xref: /freebsd/sys/kern/kern_jail.c (revision f95c930e6701aa675377bc5871ea490dd565eeba)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif /* DDB */
72 
73 #include <security/mac/mac_framework.h>
74 
75 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
76 
77 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
78 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
79 
80 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
81 #ifdef INET
82 #ifdef INET6
83 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
84 #else
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
86 #endif
87 #else /* !INET */
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	0
92 #endif
93 #endif
94 
95 /* prison0 describes what is "real" about the system. */
96 struct prison prison0 = {
97 	.pr_id		= 0,
98 	.pr_name	= "0",
99 	.pr_ref		= 1,
100 	.pr_uref	= 1,
101 	.pr_path	= "/",
102 	.pr_securelevel	= -1,
103 	.pr_devfs_rsnum = 0,
104 	.pr_childmax	= JAIL_MAX,
105 	.pr_hostuuid	= DEFAULT_HOSTUUID,
106 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
107 #ifdef VIMAGE
108 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
109 #else
110 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
111 #endif
112 	.pr_allow	= PR_ALLOW_ALL,
113 };
114 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
115 
116 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
117 struct	sx allprison_lock;
118 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
119 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
120 LIST_HEAD(, prison_racct) allprison_racct;
121 int	lastprid = 0;
122 
123 static int do_jail_attach(struct thread *td, struct prison *pr);
124 static void prison_complete(void *context, int pending);
125 static void prison_deref(struct prison *pr, int flags);
126 static char *prison_path(struct prison *pr1, struct prison *pr2);
127 static void prison_remove_one(struct prison *pr);
128 #ifdef RACCT
129 static void prison_racct_attach(struct prison *pr);
130 static void prison_racct_modify(struct prison *pr);
131 static void prison_racct_detach(struct prison *pr);
132 #endif
133 #ifdef INET
134 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
135 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
136 #endif
137 #ifdef INET6
138 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
139 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
140 #endif
141 
142 /* Flags for prison_deref */
143 #define	PD_DEREF	0x01
144 #define	PD_DEUREF	0x02
145 #define	PD_LOCKED	0x04
146 #define	PD_LIST_SLOCKED	0x08
147 #define	PD_LIST_XLOCKED	0x10
148 
149 /*
150  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
151  * as we cannot figure out the size of a sparse array, or an array without a
152  * terminating entry.
153  */
154 static char *pr_flag_names[] = {
155 	[0] = "persist",
156 #ifdef INET
157 	[7] = "ip4.saddrsel",
158 #endif
159 #ifdef INET6
160 	[8] = "ip6.saddrsel",
161 #endif
162 };
163 const size_t pr_flag_names_size = sizeof(pr_flag_names);
164 
165 static char *pr_flag_nonames[] = {
166 	[0] = "nopersist",
167 #ifdef INET
168 	[7] = "ip4.nosaddrsel",
169 #endif
170 #ifdef INET6
171 	[8] = "ip6.nosaddrsel",
172 #endif
173 };
174 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
175 
176 struct jailsys_flags {
177 	const char	*name;
178 	unsigned	 disable;
179 	unsigned	 new;
180 } pr_flag_jailsys[] = {
181 	{ "host", 0, PR_HOST },
182 #ifdef VIMAGE
183 	{ "vnet", 0, PR_VNET },
184 #endif
185 #ifdef INET
186 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
187 #endif
188 #ifdef INET6
189 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
190 #endif
191 };
192 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
193 
194 static char *pr_allow_names[] = {
195 	"allow.set_hostname",
196 	"allow.sysvipc",
197 	"allow.raw_sockets",
198 	"allow.chflags",
199 	"allow.mount",
200 	"allow.quotas",
201 	"allow.socket_af",
202 	"allow.mount.devfs",
203 	"allow.mount.nullfs",
204 	"allow.mount.zfs",
205 	"allow.mount.procfs",
206 	"allow.mount.tmpfs",
207 	"allow.mount.fdescfs",
208 };
209 const size_t pr_allow_names_size = sizeof(pr_allow_names);
210 
211 static char *pr_allow_nonames[] = {
212 	"allow.noset_hostname",
213 	"allow.nosysvipc",
214 	"allow.noraw_sockets",
215 	"allow.nochflags",
216 	"allow.nomount",
217 	"allow.noquotas",
218 	"allow.nosocket_af",
219 	"allow.mount.nodevfs",
220 	"allow.mount.nonullfs",
221 	"allow.mount.nozfs",
222 	"allow.mount.noprocfs",
223 	"allow.mount.notmpfs",
224 	"allow.mount.nofdescfs",
225 };
226 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
227 
228 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
229 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
230 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
231 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
232 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
233 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
234 #if defined(INET) || defined(INET6)
235 static unsigned jail_max_af_ips = 255;
236 #endif
237 
238 /*
239  * Initialize the parts of prison0 that can't be static-initialized with
240  * constants.  This is called from proc0_init() after creating thread0 cpuset.
241  */
242 void
243 prison0_init(void)
244 {
245 
246 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
247 	prison0.pr_osreldate = osreldate;
248 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
249 }
250 
251 #ifdef INET
252 static int
253 qcmp_v4(const void *ip1, const void *ip2)
254 {
255 	in_addr_t iaa, iab;
256 
257 	/*
258 	 * We need to compare in HBO here to get the list sorted as expected
259 	 * by the result of the code.  Sorting NBO addresses gives you
260 	 * interesting results.  If you do not understand, do not try.
261 	 */
262 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
263 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
264 
265 	/*
266 	 * Do not simply return the difference of the two numbers, the int is
267 	 * not wide enough.
268 	 */
269 	if (iaa > iab)
270 		return (1);
271 	else if (iaa < iab)
272 		return (-1);
273 	else
274 		return (0);
275 }
276 #endif
277 
278 #ifdef INET6
279 static int
280 qcmp_v6(const void *ip1, const void *ip2)
281 {
282 	const struct in6_addr *ia6a, *ia6b;
283 	int i, rc;
284 
285 	ia6a = (const struct in6_addr *)ip1;
286 	ia6b = (const struct in6_addr *)ip2;
287 
288 	rc = 0;
289 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
290 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
291 			rc = 1;
292 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
293 			rc = -1;
294 	}
295 	return (rc);
296 }
297 #endif
298 
299 /*
300  * struct jail_args {
301  *	struct jail *jail;
302  * };
303  */
304 int
305 sys_jail(struct thread *td, struct jail_args *uap)
306 {
307 	uint32_t version;
308 	int error;
309 	struct jail j;
310 
311 	error = copyin(uap->jail, &version, sizeof(uint32_t));
312 	if (error)
313 		return (error);
314 
315 	switch (version) {
316 	case 0:
317 	{
318 		struct jail_v0 j0;
319 
320 		/* FreeBSD single IPv4 jails. */
321 		bzero(&j, sizeof(struct jail));
322 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
323 		if (error)
324 			return (error);
325 		j.version = j0.version;
326 		j.path = j0.path;
327 		j.hostname = j0.hostname;
328 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
329 		break;
330 	}
331 
332 	case 1:
333 		/*
334 		 * Version 1 was used by multi-IPv4 jail implementations
335 		 * that never made it into the official kernel.
336 		 */
337 		return (EINVAL);
338 
339 	case 2:	/* JAIL_API_VERSION */
340 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
341 		error = copyin(uap->jail, &j, sizeof(struct jail));
342 		if (error)
343 			return (error);
344 		break;
345 
346 	default:
347 		/* Sci-Fi jails are not supported, sorry. */
348 		return (EINVAL);
349 	}
350 	return (kern_jail(td, &j));
351 }
352 
353 int
354 kern_jail(struct thread *td, struct jail *j)
355 {
356 	struct iovec optiov[2 * (4
357 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
358 #ifdef INET
359 			    + 1
360 #endif
361 #ifdef INET6
362 			    + 1
363 #endif
364 			    )];
365 	struct uio opt;
366 	char *u_path, *u_hostname, *u_name;
367 #ifdef INET
368 	uint32_t ip4s;
369 	struct in_addr *u_ip4;
370 #endif
371 #ifdef INET6
372 	struct in6_addr *u_ip6;
373 #endif
374 	size_t tmplen;
375 	int error, enforce_statfs, fi;
376 
377 	bzero(&optiov, sizeof(optiov));
378 	opt.uio_iov = optiov;
379 	opt.uio_iovcnt = 0;
380 	opt.uio_offset = -1;
381 	opt.uio_resid = -1;
382 	opt.uio_segflg = UIO_SYSSPACE;
383 	opt.uio_rw = UIO_READ;
384 	opt.uio_td = td;
385 
386 	/* Set permissions for top-level jails from sysctls. */
387 	if (!jailed(td->td_ucred)) {
388 		for (fi = 0; fi < sizeof(pr_allow_names) /
389 		     sizeof(pr_allow_names[0]); fi++) {
390 			optiov[opt.uio_iovcnt].iov_base =
391 			    (jail_default_allow & (1 << fi))
392 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
393 			optiov[opt.uio_iovcnt].iov_len =
394 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
395 			opt.uio_iovcnt += 2;
396 		}
397 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
398 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
399 		opt.uio_iovcnt++;
400 		enforce_statfs = jail_default_enforce_statfs;
401 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
402 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
403 		opt.uio_iovcnt++;
404 	}
405 
406 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
407 #ifdef INET
408 	ip4s = (j->version == 0) ? 1 : j->ip4s;
409 	if (ip4s > jail_max_af_ips)
410 		return (EINVAL);
411 	tmplen += ip4s * sizeof(struct in_addr);
412 #else
413 	if (j->ip4s > 0)
414 		return (EINVAL);
415 #endif
416 #ifdef INET6
417 	if (j->ip6s > jail_max_af_ips)
418 		return (EINVAL);
419 	tmplen += j->ip6s * sizeof(struct in6_addr);
420 #else
421 	if (j->ip6s > 0)
422 		return (EINVAL);
423 #endif
424 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
425 	u_hostname = u_path + MAXPATHLEN;
426 	u_name = u_hostname + MAXHOSTNAMELEN;
427 #ifdef INET
428 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
429 #endif
430 #ifdef INET6
431 #ifdef INET
432 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
433 #else
434 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
435 #endif
436 #endif
437 	optiov[opt.uio_iovcnt].iov_base = "path";
438 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
439 	opt.uio_iovcnt++;
440 	optiov[opt.uio_iovcnt].iov_base = u_path;
441 	error = copyinstr(j->path, u_path, MAXPATHLEN,
442 	    &optiov[opt.uio_iovcnt].iov_len);
443 	if (error) {
444 		free(u_path, M_TEMP);
445 		return (error);
446 	}
447 	opt.uio_iovcnt++;
448 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
449 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
450 	opt.uio_iovcnt++;
451 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
452 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
453 	    &optiov[opt.uio_iovcnt].iov_len);
454 	if (error) {
455 		free(u_path, M_TEMP);
456 		return (error);
457 	}
458 	opt.uio_iovcnt++;
459 	if (j->jailname != NULL) {
460 		optiov[opt.uio_iovcnt].iov_base = "name";
461 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
462 		opt.uio_iovcnt++;
463 		optiov[opt.uio_iovcnt].iov_base = u_name;
464 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
465 		    &optiov[opt.uio_iovcnt].iov_len);
466 		if (error) {
467 			free(u_path, M_TEMP);
468 			return (error);
469 		}
470 		opt.uio_iovcnt++;
471 	}
472 #ifdef INET
473 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
474 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
475 	opt.uio_iovcnt++;
476 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
477 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
478 	if (j->version == 0)
479 		u_ip4->s_addr = j->ip4s;
480 	else {
481 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
482 		if (error) {
483 			free(u_path, M_TEMP);
484 			return (error);
485 		}
486 	}
487 	opt.uio_iovcnt++;
488 #endif
489 #ifdef INET6
490 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
491 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
492 	opt.uio_iovcnt++;
493 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
494 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
495 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
496 	if (error) {
497 		free(u_path, M_TEMP);
498 		return (error);
499 	}
500 	opt.uio_iovcnt++;
501 #endif
502 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
503 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
504 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
505 	free(u_path, M_TEMP);
506 	return (error);
507 }
508 
509 
510 /*
511  * struct jail_set_args {
512  *	struct iovec *iovp;
513  *	unsigned int iovcnt;
514  *	int flags;
515  * };
516  */
517 int
518 sys_jail_set(struct thread *td, struct jail_set_args *uap)
519 {
520 	struct uio *auio;
521 	int error;
522 
523 	/* Check that we have an even number of iovecs. */
524 	if (uap->iovcnt & 1)
525 		return (EINVAL);
526 
527 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
528 	if (error)
529 		return (error);
530 	error = kern_jail_set(td, auio, uap->flags);
531 	free(auio, M_IOV);
532 	return (error);
533 }
534 
535 int
536 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
537 {
538 	struct nameidata nd;
539 #ifdef INET
540 	struct in_addr *ip4;
541 #endif
542 #ifdef INET6
543 	struct in6_addr *ip6;
544 #endif
545 	struct vfsopt *opt;
546 	struct vfsoptlist *opts;
547 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
548 	struct vnode *root;
549 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
550 	char *g_path, *osrelstr;
551 #if defined(INET) || defined(INET6)
552 	struct prison *tppr;
553 	void *op;
554 #endif
555 	unsigned long hid;
556 	size_t namelen, onamelen;
557 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
558 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
559 	int fi, jid, jsys, len, level;
560 	int childmax, osreldt, rsnum, slevel;
561 	int fullpath_disabled;
562 #if defined(INET) || defined(INET6)
563 	int ii, ij;
564 #endif
565 #ifdef INET
566 	int ip4s, redo_ip4;
567 #endif
568 #ifdef INET6
569 	int ip6s, redo_ip6;
570 #endif
571 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
572 	unsigned tallow;
573 	char numbuf[12];
574 
575 	error = priv_check(td, PRIV_JAIL_SET);
576 	if (!error && (flags & JAIL_ATTACH))
577 		error = priv_check(td, PRIV_JAIL_ATTACH);
578 	if (error)
579 		return (error);
580 	mypr = ppr = td->td_ucred->cr_prison;
581 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
582 		return (EPERM);
583 	if (flags & ~JAIL_SET_MASK)
584 		return (EINVAL);
585 
586 	/*
587 	 * Check all the parameters before committing to anything.  Not all
588 	 * errors can be caught early, but we may as well try.  Also, this
589 	 * takes care of some expensive stuff (path lookup) before getting
590 	 * the allprison lock.
591 	 *
592 	 * XXX Jails are not filesystems, and jail parameters are not mount
593 	 *     options.  But it makes more sense to re-use the vfsopt code
594 	 *     than duplicate it under a different name.
595 	 */
596 	error = vfs_buildopts(optuio, &opts);
597 	if (error)
598 		return (error);
599 #ifdef INET
600 	ip4 = NULL;
601 #endif
602 #ifdef INET6
603 	ip6 = NULL;
604 #endif
605 	g_path = NULL;
606 
607 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
608 	if (error == ENOENT)
609 		jid = 0;
610 	else if (error != 0)
611 		goto done_free;
612 
613 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
614 	if (error == ENOENT)
615 		gotslevel = 0;
616 	else if (error != 0)
617 		goto done_free;
618 	else
619 		gotslevel = 1;
620 
621 	error =
622 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
623 	if (error == ENOENT)
624 		gotchildmax = 0;
625 	else if (error != 0)
626 		goto done_free;
627 	else
628 		gotchildmax = 1;
629 
630 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
631 	if (error == ENOENT)
632 		gotenforce = 0;
633 	else if (error != 0)
634 		goto done_free;
635 	else if (enforce < 0 || enforce > 2) {
636 		error = EINVAL;
637 		goto done_free;
638 	} else
639 		gotenforce = 1;
640 
641 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
642 	if (error == ENOENT)
643 		gotrsnum = 0;
644 	else if (error != 0)
645 		goto done_free;
646 	else
647 		gotrsnum = 1;
648 
649 	pr_flags = ch_flags = 0;
650 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
651 	    fi++) {
652 		if (pr_flag_names[fi] == NULL)
653 			continue;
654 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
655 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
656 	}
657 	ch_flags |= pr_flags;
658 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
659 	    fi++) {
660 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
661 		    sizeof(jsys));
662 		if (error == ENOENT)
663 			continue;
664 		if (error != 0)
665 			goto done_free;
666 		switch (jsys) {
667 		case JAIL_SYS_DISABLE:
668 			if (!pr_flag_jailsys[fi].disable) {
669 				error = EINVAL;
670 				goto done_free;
671 			}
672 			pr_flags |= pr_flag_jailsys[fi].disable;
673 			break;
674 		case JAIL_SYS_NEW:
675 			pr_flags |= pr_flag_jailsys[fi].new;
676 			break;
677 		case JAIL_SYS_INHERIT:
678 			break;
679 		default:
680 			error = EINVAL;
681 			goto done_free;
682 		}
683 		ch_flags |=
684 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
685 	}
686 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
687 	    && !(pr_flags & PR_PERSIST)) {
688 		error = EINVAL;
689 		vfs_opterror(opts, "new jail must persist or attach");
690 		goto done_errmsg;
691 	}
692 #ifdef VIMAGE
693 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
694 		error = EINVAL;
695 		vfs_opterror(opts, "vnet cannot be changed after creation");
696 		goto done_errmsg;
697 	}
698 #endif
699 #ifdef INET
700 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
701 		error = EINVAL;
702 		vfs_opterror(opts, "ip4 cannot be changed after creation");
703 		goto done_errmsg;
704 	}
705 #endif
706 #ifdef INET6
707 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
708 		error = EINVAL;
709 		vfs_opterror(opts, "ip6 cannot be changed after creation");
710 		goto done_errmsg;
711 	}
712 #endif
713 
714 	pr_allow = ch_allow = 0;
715 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
716 	    fi++) {
717 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
718 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
719 	}
720 	ch_allow |= pr_allow;
721 
722 	error = vfs_getopt(opts, "name", (void **)&name, &len);
723 	if (error == ENOENT)
724 		name = NULL;
725 	else if (error != 0)
726 		goto done_free;
727 	else {
728 		if (len == 0 || name[len - 1] != '\0') {
729 			error = EINVAL;
730 			goto done_free;
731 		}
732 		if (len > MAXHOSTNAMELEN) {
733 			error = ENAMETOOLONG;
734 			goto done_free;
735 		}
736 	}
737 
738 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
739 	if (error == ENOENT)
740 		host = NULL;
741 	else if (error != 0)
742 		goto done_free;
743 	else {
744 		ch_flags |= PR_HOST;
745 		pr_flags |= PR_HOST;
746 		if (len == 0 || host[len - 1] != '\0') {
747 			error = EINVAL;
748 			goto done_free;
749 		}
750 		if (len > MAXHOSTNAMELEN) {
751 			error = ENAMETOOLONG;
752 			goto done_free;
753 		}
754 	}
755 
756 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
757 	if (error == ENOENT)
758 		domain = NULL;
759 	else if (error != 0)
760 		goto done_free;
761 	else {
762 		ch_flags |= PR_HOST;
763 		pr_flags |= PR_HOST;
764 		if (len == 0 || domain[len - 1] != '\0') {
765 			error = EINVAL;
766 			goto done_free;
767 		}
768 		if (len > MAXHOSTNAMELEN) {
769 			error = ENAMETOOLONG;
770 			goto done_free;
771 		}
772 	}
773 
774 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
775 	if (error == ENOENT)
776 		uuid = NULL;
777 	else if (error != 0)
778 		goto done_free;
779 	else {
780 		ch_flags |= PR_HOST;
781 		pr_flags |= PR_HOST;
782 		if (len == 0 || uuid[len - 1] != '\0') {
783 			error = EINVAL;
784 			goto done_free;
785 		}
786 		if (len > HOSTUUIDLEN) {
787 			error = ENAMETOOLONG;
788 			goto done_free;
789 		}
790 	}
791 
792 #ifdef COMPAT_FREEBSD32
793 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
794 		uint32_t hid32;
795 
796 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
797 		hid = hid32;
798 	} else
799 #endif
800 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
801 	if (error == ENOENT)
802 		gothid = 0;
803 	else if (error != 0)
804 		goto done_free;
805 	else {
806 		gothid = 1;
807 		ch_flags |= PR_HOST;
808 		pr_flags |= PR_HOST;
809 	}
810 
811 #ifdef INET
812 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
813 	if (error == ENOENT)
814 		ip4s = 0;
815 	else if (error != 0)
816 		goto done_free;
817 	else if (ip4s & (sizeof(*ip4) - 1)) {
818 		error = EINVAL;
819 		goto done_free;
820 	} else {
821 		ch_flags |= PR_IP4_USER;
822 		pr_flags |= PR_IP4_USER;
823 		if (ip4s > 0) {
824 			ip4s /= sizeof(*ip4);
825 			if (ip4s > jail_max_af_ips) {
826 				error = EINVAL;
827 				vfs_opterror(opts, "too many IPv4 addresses");
828 				goto done_errmsg;
829 			}
830 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
831 			bcopy(op, ip4, ip4s * sizeof(*ip4));
832 			/*
833 			 * IP addresses are all sorted but ip[0] to preserve
834 			 * the primary IP address as given from userland.
835 			 * This special IP is used for unbound outgoing
836 			 * connections as well for "loopback" traffic in case
837 			 * source address selection cannot find any more fitting
838 			 * address to connect from.
839 			 */
840 			if (ip4s > 1)
841 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
842 			/*
843 			 * Check for duplicate addresses and do some simple
844 			 * zero and broadcast checks. If users give other bogus
845 			 * addresses it is their problem.
846 			 *
847 			 * We do not have to care about byte order for these
848 			 * checks so we will do them in NBO.
849 			 */
850 			for (ii = 0; ii < ip4s; ii++) {
851 				if (ip4[ii].s_addr == INADDR_ANY ||
852 				    ip4[ii].s_addr == INADDR_BROADCAST) {
853 					error = EINVAL;
854 					goto done_free;
855 				}
856 				if ((ii+1) < ip4s &&
857 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
858 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
859 					error = EINVAL;
860 					goto done_free;
861 				}
862 			}
863 		}
864 	}
865 #endif
866 
867 #ifdef INET6
868 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
869 	if (error == ENOENT)
870 		ip6s = 0;
871 	else if (error != 0)
872 		goto done_free;
873 	else if (ip6s & (sizeof(*ip6) - 1)) {
874 		error = EINVAL;
875 		goto done_free;
876 	} else {
877 		ch_flags |= PR_IP6_USER;
878 		pr_flags |= PR_IP6_USER;
879 		if (ip6s > 0) {
880 			ip6s /= sizeof(*ip6);
881 			if (ip6s > jail_max_af_ips) {
882 				error = EINVAL;
883 				vfs_opterror(opts, "too many IPv6 addresses");
884 				goto done_errmsg;
885 			}
886 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
887 			bcopy(op, ip6, ip6s * sizeof(*ip6));
888 			if (ip6s > 1)
889 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
890 			for (ii = 0; ii < ip6s; ii++) {
891 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
892 					error = EINVAL;
893 					goto done_free;
894 				}
895 				if ((ii+1) < ip6s &&
896 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
897 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
898 				{
899 					error = EINVAL;
900 					goto done_free;
901 				}
902 			}
903 		}
904 	}
905 #endif
906 
907 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
908 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
909 		error = EINVAL;
910 		vfs_opterror(opts,
911 		    "vnet jails cannot have IP address restrictions");
912 		goto done_errmsg;
913 	}
914 #endif
915 
916 	fullpath_disabled = 0;
917 	root = NULL;
918 	error = vfs_getopt(opts, "path", (void **)&path, &len);
919 	if (error == ENOENT)
920 		path = NULL;
921 	else if (error != 0)
922 		goto done_free;
923 	else {
924 		if (flags & JAIL_UPDATE) {
925 			error = EINVAL;
926 			vfs_opterror(opts,
927 			    "path cannot be changed after creation");
928 			goto done_errmsg;
929 		}
930 		if (len == 0 || path[len - 1] != '\0') {
931 			error = EINVAL;
932 			goto done_free;
933 		}
934 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
935 		    path, td);
936 		error = namei(&nd);
937 		if (error)
938 			goto done_free;
939 		root = nd.ni_vp;
940 		NDFREE(&nd, NDF_ONLY_PNBUF);
941 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
942 		strlcpy(g_path, path, MAXPATHLEN);
943 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
944 		if (error == 0)
945 			path = g_path;
946 		else if (error == ENODEV) {
947 			/* proceed if sysctl debug.disablefullpath == 1 */
948 			fullpath_disabled = 1;
949 			if (len < 2 || (len == 2 && path[0] == '/'))
950 				path = NULL;
951 		} else {
952 			/* exit on other errors */
953 			goto done_free;
954 		}
955 		if (root->v_type != VDIR) {
956 			error = ENOTDIR;
957 			vput(root);
958 			goto done_free;
959 		}
960 		VOP_UNLOCK(root, 0);
961 		if (fullpath_disabled) {
962 			/* Leave room for a real-root full pathname. */
963 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
964 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
965 				error = ENAMETOOLONG;
966 				goto done_free;
967 			}
968 		}
969 	}
970 
971 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
972 	if (error == ENOENT)
973 		osrelstr = NULL;
974 	else if (error != 0)
975 		goto done_free;
976 	else {
977 		if (flags & JAIL_UPDATE) {
978 			error = EINVAL;
979 			vfs_opterror(opts,
980 			    "osrelease cannot be changed after creation");
981 			goto done_errmsg;
982 		}
983 		if (len == 0 || len >= OSRELEASELEN) {
984 			error = EINVAL;
985 			vfs_opterror(opts,
986 			    "osrelease string must be 1-%d bytes long",
987 			    OSRELEASELEN - 1);
988 			goto done_errmsg;
989 		}
990 	}
991 
992 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
993 	if (error == ENOENT)
994 		osreldt = 0;
995 	else if (error != 0)
996 		goto done_free;
997 	else {
998 		if (flags & JAIL_UPDATE) {
999 			error = EINVAL;
1000 			vfs_opterror(opts,
1001 			    "osreldate cannot be changed after creation");
1002 			goto done_errmsg;
1003 		}
1004 		if (osreldt == 0) {
1005 			error = EINVAL;
1006 			vfs_opterror(opts, "osreldate cannot be 0");
1007 			goto done_errmsg;
1008 		}
1009 	}
1010 
1011 	/*
1012 	 * Grab the allprison lock before letting modules check their
1013 	 * parameters.  Once we have it, do not let go so we'll have a
1014 	 * consistent view of the OSD list.
1015 	 */
1016 	sx_xlock(&allprison_lock);
1017 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
1018 	if (error)
1019 		goto done_unlock_list;
1020 
1021 	/* By now, all parameters should have been noted. */
1022 	TAILQ_FOREACH(opt, opts, link) {
1023 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1024 			error = EINVAL;
1025 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1026 			goto done_unlock_list;
1027 		}
1028 	}
1029 
1030 	/*
1031 	 * See if we are creating a new record or updating an existing one.
1032 	 * This abuses the file error codes ENOENT and EEXIST.
1033 	 */
1034 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1035 	if (!cuflags) {
1036 		error = EINVAL;
1037 		vfs_opterror(opts, "no valid operation (create or update)");
1038 		goto done_unlock_list;
1039 	}
1040 	pr = NULL;
1041 	namelc = NULL;
1042 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1043 		namelc = strrchr(name, '.');
1044 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1045 		if (*p != '\0')
1046 			jid = 0;
1047 	}
1048 	if (jid != 0) {
1049 		/*
1050 		 * See if a requested jid already exists.  There is an
1051 		 * information leak here if the jid exists but is not within
1052 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1053 		 * even though they cannot see the jail, and CREATE | UPDATE
1054 		 * will return ENOENT which is not normally a valid error.
1055 		 */
1056 		if (jid < 0) {
1057 			error = EINVAL;
1058 			vfs_opterror(opts, "negative jid");
1059 			goto done_unlock_list;
1060 		}
1061 		pr = prison_find(jid);
1062 		if (pr != NULL) {
1063 			ppr = pr->pr_parent;
1064 			/* Create: jid must not exist. */
1065 			if (cuflags == JAIL_CREATE) {
1066 				mtx_unlock(&pr->pr_mtx);
1067 				error = EEXIST;
1068 				vfs_opterror(opts, "jail %d already exists",
1069 				    jid);
1070 				goto done_unlock_list;
1071 			}
1072 			if (!prison_ischild(mypr, pr)) {
1073 				mtx_unlock(&pr->pr_mtx);
1074 				pr = NULL;
1075 			} else if (pr->pr_uref == 0) {
1076 				if (!(flags & JAIL_DYING)) {
1077 					mtx_unlock(&pr->pr_mtx);
1078 					error = ENOENT;
1079 					vfs_opterror(opts, "jail %d is dying",
1080 					    jid);
1081 					goto done_unlock_list;
1082 				} else if ((flags & JAIL_ATTACH) ||
1083 				    (pr_flags & PR_PERSIST)) {
1084 					/*
1085 					 * A dying jail might be resurrected
1086 					 * (via attach or persist), but first
1087 					 * it must determine if another jail
1088 					 * has claimed its name.  Accomplish
1089 					 * this by implicitly re-setting the
1090 					 * name.
1091 					 */
1092 					if (name == NULL)
1093 						name = prison_name(mypr, pr);
1094 				}
1095 			}
1096 		}
1097 		if (pr == NULL) {
1098 			/* Update: jid must exist. */
1099 			if (cuflags == JAIL_UPDATE) {
1100 				error = ENOENT;
1101 				vfs_opterror(opts, "jail %d not found", jid);
1102 				goto done_unlock_list;
1103 			}
1104 		}
1105 	}
1106 	/*
1107 	 * If the caller provided a name, look for a jail by that name.
1108 	 * This has different semantics for creates and updates keyed by jid
1109 	 * (where the name must not already exist in a different jail),
1110 	 * and updates keyed by the name itself (where the name must exist
1111 	 * because that is the jail being updated).
1112 	 */
1113 	if (name != NULL) {
1114 		namelc = strrchr(name, '.');
1115 		if (namelc == NULL)
1116 			namelc = name;
1117 		else {
1118 			/*
1119 			 * This is a hierarchical name.  Split it into the
1120 			 * parent and child names, and make sure the parent
1121 			 * exists or matches an already found jail.
1122 			 */
1123 			*namelc = '\0';
1124 			if (pr != NULL) {
1125 				if (strncmp(name, ppr->pr_name, namelc - name)
1126 				    || ppr->pr_name[namelc - name] != '\0') {
1127 					mtx_unlock(&pr->pr_mtx);
1128 					error = EINVAL;
1129 					vfs_opterror(opts,
1130 					    "cannot change jail's parent");
1131 					goto done_unlock_list;
1132 				}
1133 			} else {
1134 				ppr = prison_find_name(mypr, name);
1135 				if (ppr == NULL) {
1136 					error = ENOENT;
1137 					vfs_opterror(opts,
1138 					    "jail \"%s\" not found", name);
1139 					goto done_unlock_list;
1140 				}
1141 				mtx_unlock(&ppr->pr_mtx);
1142 			}
1143 			name = ++namelc;
1144 		}
1145 		if (name[0] != '\0') {
1146 			namelen =
1147 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1148  name_again:
1149 			deadpr = NULL;
1150 			FOREACH_PRISON_CHILD(ppr, tpr) {
1151 				if (tpr != pr && tpr->pr_ref > 0 &&
1152 				    !strcmp(tpr->pr_name + namelen, name)) {
1153 					if (pr == NULL &&
1154 					    cuflags != JAIL_CREATE) {
1155 						mtx_lock(&tpr->pr_mtx);
1156 						if (tpr->pr_ref > 0) {
1157 							/*
1158 							 * Use this jail
1159 							 * for updates.
1160 							 */
1161 							if (tpr->pr_uref > 0) {
1162 								pr = tpr;
1163 								break;
1164 							}
1165 							deadpr = tpr;
1166 						}
1167 						mtx_unlock(&tpr->pr_mtx);
1168 					} else if (tpr->pr_uref > 0) {
1169 						/*
1170 						 * Create, or update(jid):
1171 						 * name must not exist in an
1172 						 * active sibling jail.
1173 						 */
1174 						error = EEXIST;
1175 						if (pr != NULL)
1176 							mtx_unlock(&pr->pr_mtx);
1177 						vfs_opterror(opts,
1178 						   "jail \"%s\" already exists",
1179 						   name);
1180 						goto done_unlock_list;
1181 					}
1182 				}
1183 			}
1184 			/* If no active jail is found, use a dying one. */
1185 			if (deadpr != NULL && pr == NULL) {
1186 				if (flags & JAIL_DYING) {
1187 					mtx_lock(&deadpr->pr_mtx);
1188 					if (deadpr->pr_ref == 0) {
1189 						mtx_unlock(&deadpr->pr_mtx);
1190 						goto name_again;
1191 					}
1192 					pr = deadpr;
1193 				} else if (cuflags == JAIL_UPDATE) {
1194 					error = ENOENT;
1195 					vfs_opterror(opts,
1196 					    "jail \"%s\" is dying", name);
1197 					goto done_unlock_list;
1198 				}
1199 			}
1200 			/* Update: name must exist if no jid. */
1201 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1202 				error = ENOENT;
1203 				vfs_opterror(opts, "jail \"%s\" not found",
1204 				    name);
1205 				goto done_unlock_list;
1206 			}
1207 		}
1208 	}
1209 	/* Update: must provide a jid or name. */
1210 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1211 		error = ENOENT;
1212 		vfs_opterror(opts, "update specified no jail");
1213 		goto done_unlock_list;
1214 	}
1215 
1216 	/* If there's no prison to update, create a new one and link it in. */
1217 	if (pr == NULL) {
1218 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1219 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1220 				error = EPERM;
1221 				vfs_opterror(opts, "prison limit exceeded");
1222 				goto done_unlock_list;
1223 			}
1224 		created = 1;
1225 		mtx_lock(&ppr->pr_mtx);
1226 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1227 			mtx_unlock(&ppr->pr_mtx);
1228 			error = ENOENT;
1229 			vfs_opterror(opts, "parent jail went away!");
1230 			goto done_unlock_list;
1231 		}
1232 		ppr->pr_ref++;
1233 		ppr->pr_uref++;
1234 		mtx_unlock(&ppr->pr_mtx);
1235 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1236 		if (jid == 0) {
1237 			/* Find the next free jid. */
1238 			jid = lastprid + 1;
1239  findnext:
1240 			if (jid == JAIL_MAX)
1241 				jid = 1;
1242 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1243 				if (tpr->pr_id < jid)
1244 					continue;
1245 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1246 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1247 					break;
1248 				}
1249 				if (jid == lastprid) {
1250 					error = EAGAIN;
1251 					vfs_opterror(opts,
1252 					    "no available jail IDs");
1253 					free(pr, M_PRISON);
1254 					prison_deref(ppr, PD_DEREF |
1255 					    PD_DEUREF | PD_LIST_XLOCKED);
1256 					goto done_releroot;
1257 				}
1258 				jid++;
1259 				goto findnext;
1260 			}
1261 			lastprid = jid;
1262 		} else {
1263 			/*
1264 			 * The jail already has a jid (that did not yet exist),
1265 			 * so just find where to insert it.
1266 			 */
1267 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1268 				if (tpr->pr_id >= jid) {
1269 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1270 					break;
1271 				}
1272 		}
1273 		if (tpr == NULL)
1274 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1275 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1276 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1277 			tpr->pr_childcount++;
1278 
1279 		pr->pr_parent = ppr;
1280 		pr->pr_id = jid;
1281 
1282 		/* Set some default values, and inherit some from the parent. */
1283 		if (name == NULL)
1284 			name = "";
1285 		if (path == NULL) {
1286 			path = "/";
1287 			root = mypr->pr_root;
1288 			vref(root);
1289 		}
1290 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1291 		pr->pr_flags |= PR_HOST;
1292 #if defined(INET) || defined(INET6)
1293 #ifdef VIMAGE
1294 		if (!(pr_flags & PR_VNET))
1295 #endif
1296 		{
1297 #ifdef INET
1298 			if (!(ch_flags & PR_IP4_USER))
1299 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1300 			else if (!(pr_flags & PR_IP4_USER)) {
1301 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1302 				if (ppr->pr_ip4 != NULL) {
1303 					pr->pr_ip4s = ppr->pr_ip4s;
1304 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1305 					    sizeof(struct in_addr), M_PRISON,
1306 					    M_WAITOK);
1307 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1308 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1309 				}
1310 			}
1311 #endif
1312 #ifdef INET6
1313 			if (!(ch_flags & PR_IP6_USER))
1314 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1315 			else if (!(pr_flags & PR_IP6_USER)) {
1316 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1317 				if (ppr->pr_ip6 != NULL) {
1318 					pr->pr_ip6s = ppr->pr_ip6s;
1319 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1320 					    sizeof(struct in6_addr), M_PRISON,
1321 					    M_WAITOK);
1322 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1323 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1324 				}
1325 			}
1326 #endif
1327 		}
1328 #endif
1329 		/* Source address selection is always on by default. */
1330 		pr->pr_flags |= _PR_IP_SADDRSEL;
1331 
1332 		pr->pr_securelevel = ppr->pr_securelevel;
1333 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1334 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1335 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1336 
1337 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1338 		if (osrelstr == NULL)
1339 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1340 		else
1341 		    strcpy(pr->pr_osrelease, osrelstr);
1342 
1343 		LIST_INIT(&pr->pr_children);
1344 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1345 
1346 #ifdef VIMAGE
1347 		/* Allocate a new vnet if specified. */
1348 		pr->pr_vnet = (pr_flags & PR_VNET)
1349 		    ? vnet_alloc() : ppr->pr_vnet;
1350 #endif
1351 		/*
1352 		 * Allocate a dedicated cpuset for each jail.
1353 		 * Unlike other initial settings, this may return an erorr.
1354 		 */
1355 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1356 		if (error) {
1357 			prison_deref(pr, PD_LIST_XLOCKED);
1358 			goto done_releroot;
1359 		}
1360 
1361 		mtx_lock(&pr->pr_mtx);
1362 		/*
1363 		 * New prisons do not yet have a reference, because we do not
1364 		 * want other to see the incomplete prison once the
1365 		 * allprison_lock is downgraded.
1366 		 */
1367 	} else {
1368 		created = 0;
1369 		/*
1370 		 * Grab a reference for existing prisons, to ensure they
1371 		 * continue to exist for the duration of the call.
1372 		 */
1373 		pr->pr_ref++;
1374 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1375 		if ((pr->pr_flags & PR_VNET) &&
1376 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1377 			error = EINVAL;
1378 			vfs_opterror(opts,
1379 			    "vnet jails cannot have IP address restrictions");
1380 			goto done_deref_locked;
1381 		}
1382 #endif
1383 #ifdef INET
1384 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1385 			error = EINVAL;
1386 			vfs_opterror(opts,
1387 			    "ip4 cannot be changed after creation");
1388 			goto done_deref_locked;
1389 		}
1390 #endif
1391 #ifdef INET6
1392 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1393 			error = EINVAL;
1394 			vfs_opterror(opts,
1395 			    "ip6 cannot be changed after creation");
1396 			goto done_deref_locked;
1397 		}
1398 #endif
1399 	}
1400 
1401 	/* Do final error checking before setting anything. */
1402 	if (gotslevel) {
1403 		if (slevel < ppr->pr_securelevel) {
1404 			error = EPERM;
1405 			goto done_deref_locked;
1406 		}
1407 	}
1408 	if (gotchildmax) {
1409 		if (childmax >= ppr->pr_childmax) {
1410 			error = EPERM;
1411 			goto done_deref_locked;
1412 		}
1413 	}
1414 	if (gotenforce) {
1415 		if (enforce < ppr->pr_enforce_statfs) {
1416 			error = EPERM;
1417 			goto done_deref_locked;
1418 		}
1419 	}
1420 	if (gotrsnum) {
1421 		/*
1422 		 * devfs_rsnum is a uint16_t
1423 		 */
1424 		if (rsnum < 0 || rsnum > 65535) {
1425 			error = EINVAL;
1426 			goto done_deref_locked;
1427 		}
1428 		/*
1429 		 * Nested jails always inherit parent's devfs ruleset
1430 		 */
1431 		if (jailed(td->td_ucred)) {
1432 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1433 				error = EPERM;
1434 				goto done_deref_locked;
1435 			} else
1436 				rsnum = ppr->pr_devfs_rsnum;
1437 		}
1438 	}
1439 #ifdef INET
1440 	if (ip4s > 0) {
1441 		if (ppr->pr_flags & PR_IP4) {
1442 			/*
1443 			 * Make sure the new set of IP addresses is a
1444 			 * subset of the parent's list.  Don't worry
1445 			 * about the parent being unlocked, as any
1446 			 * setting is done with allprison_lock held.
1447 			 */
1448 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1449 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1450 					break;
1451 			if (ij == ppr->pr_ip4s) {
1452 				error = EPERM;
1453 				goto done_deref_locked;
1454 			}
1455 			if (ip4s > 1) {
1456 				for (ii = ij = 1; ii < ip4s; ii++) {
1457 					if (ip4[ii].s_addr ==
1458 					    ppr->pr_ip4[0].s_addr)
1459 						continue;
1460 					for (; ij < ppr->pr_ip4s; ij++)
1461 						if (ip4[ii].s_addr ==
1462 						    ppr->pr_ip4[ij].s_addr)
1463 							break;
1464 					if (ij == ppr->pr_ip4s)
1465 						break;
1466 				}
1467 				if (ij == ppr->pr_ip4s) {
1468 					error = EPERM;
1469 					goto done_deref_locked;
1470 				}
1471 			}
1472 		}
1473 		/*
1474 		 * Check for conflicting IP addresses.  We permit them
1475 		 * if there is no more than one IP on each jail.  If
1476 		 * there is a duplicate on a jail with more than one
1477 		 * IP stop checking and return error.
1478 		 */
1479 		tppr = ppr;
1480 #ifdef VIMAGE
1481 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1482 			if (tppr->pr_flags & PR_VNET)
1483 				break;
1484 #endif
1485 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1486 			if (tpr == pr ||
1487 #ifdef VIMAGE
1488 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1489 #endif
1490 			    tpr->pr_uref == 0) {
1491 				descend = 0;
1492 				continue;
1493 			}
1494 			if (!(tpr->pr_flags & PR_IP4_USER))
1495 				continue;
1496 			descend = 0;
1497 			if (tpr->pr_ip4 == NULL ||
1498 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1499 				continue;
1500 			for (ii = 0; ii < ip4s; ii++) {
1501 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1502 					error = EADDRINUSE;
1503 					vfs_opterror(opts,
1504 					    "IPv4 addresses clash");
1505 					goto done_deref_locked;
1506 				}
1507 			}
1508 		}
1509 	}
1510 #endif
1511 #ifdef INET6
1512 	if (ip6s > 0) {
1513 		if (ppr->pr_flags & PR_IP6) {
1514 			/*
1515 			 * Make sure the new set of IP addresses is a
1516 			 * subset of the parent's list.
1517 			 */
1518 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1519 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1520 				    &ppr->pr_ip6[ij]))
1521 					break;
1522 			if (ij == ppr->pr_ip6s) {
1523 				error = EPERM;
1524 				goto done_deref_locked;
1525 			}
1526 			if (ip6s > 1) {
1527 				for (ii = ij = 1; ii < ip6s; ii++) {
1528 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1529 					     &ppr->pr_ip6[0]))
1530 						continue;
1531 					for (; ij < ppr->pr_ip6s; ij++)
1532 						if (IN6_ARE_ADDR_EQUAL(
1533 						    &ip6[ii], &ppr->pr_ip6[ij]))
1534 							break;
1535 					if (ij == ppr->pr_ip6s)
1536 						break;
1537 				}
1538 				if (ij == ppr->pr_ip6s) {
1539 					error = EPERM;
1540 					goto done_deref_locked;
1541 				}
1542 			}
1543 		}
1544 		/* Check for conflicting IP addresses. */
1545 		tppr = ppr;
1546 #ifdef VIMAGE
1547 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1548 			if (tppr->pr_flags & PR_VNET)
1549 				break;
1550 #endif
1551 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1552 			if (tpr == pr ||
1553 #ifdef VIMAGE
1554 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1555 #endif
1556 			    tpr->pr_uref == 0) {
1557 				descend = 0;
1558 				continue;
1559 			}
1560 			if (!(tpr->pr_flags & PR_IP6_USER))
1561 				continue;
1562 			descend = 0;
1563 			if (tpr->pr_ip6 == NULL ||
1564 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1565 				continue;
1566 			for (ii = 0; ii < ip6s; ii++) {
1567 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1568 					error = EADDRINUSE;
1569 					vfs_opterror(opts,
1570 					    "IPv6 addresses clash");
1571 					goto done_deref_locked;
1572 				}
1573 			}
1574 		}
1575 	}
1576 #endif
1577 	onamelen = namelen = 0;
1578 	if (name != NULL) {
1579 		/* Give a default name of the jid. */
1580 		if (name[0] == '\0')
1581 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1582 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1583 		    *p == '\0')) {
1584 			error = EINVAL;
1585 			vfs_opterror(opts,
1586 			    "name cannot be numeric (unless it is the jid)");
1587 			goto done_deref_locked;
1588 		}
1589 		/*
1590 		 * Make sure the name isn't too long for the prison or its
1591 		 * children.
1592 		 */
1593 		onamelen = strlen(pr->pr_name);
1594 		namelen = strlen(name);
1595 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1596 			error = ENAMETOOLONG;
1597 			goto done_deref_locked;
1598 		}
1599 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1600 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1601 			    sizeof(pr->pr_name)) {
1602 				error = ENAMETOOLONG;
1603 				goto done_deref_locked;
1604 			}
1605 		}
1606 	}
1607 	if (pr_allow & ~ppr->pr_allow) {
1608 		error = EPERM;
1609 		goto done_deref_locked;
1610 	}
1611 
1612 	/* Set the parameters of the prison. */
1613 #ifdef INET
1614 	redo_ip4 = 0;
1615 	if (pr_flags & PR_IP4_USER) {
1616 		pr->pr_flags |= PR_IP4;
1617 		free(pr->pr_ip4, M_PRISON);
1618 		pr->pr_ip4s = ip4s;
1619 		pr->pr_ip4 = ip4;
1620 		ip4 = NULL;
1621 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1622 #ifdef VIMAGE
1623 			if (tpr->pr_flags & PR_VNET) {
1624 				descend = 0;
1625 				continue;
1626 			}
1627 #endif
1628 			if (prison_restrict_ip4(tpr, NULL)) {
1629 				redo_ip4 = 1;
1630 				descend = 0;
1631 			}
1632 		}
1633 	}
1634 #endif
1635 #ifdef INET6
1636 	redo_ip6 = 0;
1637 	if (pr_flags & PR_IP6_USER) {
1638 		pr->pr_flags |= PR_IP6;
1639 		free(pr->pr_ip6, M_PRISON);
1640 		pr->pr_ip6s = ip6s;
1641 		pr->pr_ip6 = ip6;
1642 		ip6 = NULL;
1643 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1644 #ifdef VIMAGE
1645 			if (tpr->pr_flags & PR_VNET) {
1646 				descend = 0;
1647 				continue;
1648 			}
1649 #endif
1650 			if (prison_restrict_ip6(tpr, NULL)) {
1651 				redo_ip6 = 1;
1652 				descend = 0;
1653 			}
1654 		}
1655 	}
1656 #endif
1657 	if (gotslevel) {
1658 		pr->pr_securelevel = slevel;
1659 		/* Set all child jails to be at least this level. */
1660 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1661 			if (tpr->pr_securelevel < slevel)
1662 				tpr->pr_securelevel = slevel;
1663 	}
1664 	if (gotchildmax) {
1665 		pr->pr_childmax = childmax;
1666 		/* Set all child jails to under this limit. */
1667 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1668 			if (tpr->pr_childmax > childmax - level)
1669 				tpr->pr_childmax = childmax > level
1670 				    ? childmax - level : 0;
1671 	}
1672 	if (gotenforce) {
1673 		pr->pr_enforce_statfs = enforce;
1674 		/* Pass this restriction on to the children. */
1675 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1676 			if (tpr->pr_enforce_statfs < enforce)
1677 				tpr->pr_enforce_statfs = enforce;
1678 	}
1679 	if (gotrsnum) {
1680 		pr->pr_devfs_rsnum = rsnum;
1681 		/* Pass this restriction on to the children. */
1682 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1683 			tpr->pr_devfs_rsnum = rsnum;
1684 	}
1685 	if (name != NULL) {
1686 		if (ppr == &prison0)
1687 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1688 		else
1689 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1690 			    ppr->pr_name, name);
1691 		/* Change this component of child names. */
1692 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1693 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1694 			    strlen(tpr->pr_name + onamelen) + 1);
1695 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1696 		}
1697 	}
1698 	if (path != NULL) {
1699 		/* Try to keep a real-rooted full pathname. */
1700 		if (fullpath_disabled && path[0] == '/' &&
1701 		    strcmp(mypr->pr_path, "/"))
1702 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1703 			    mypr->pr_path, path);
1704 		else
1705 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1706 		pr->pr_root = root;
1707 	}
1708 	if (PR_HOST & ch_flags & ~pr_flags) {
1709 		if (pr->pr_flags & PR_HOST) {
1710 			/*
1711 			 * Copy the parent's host info.  As with pr_ip4 above,
1712 			 * the lack of a lock on the parent is not a problem;
1713 			 * it is always set with allprison_lock at least
1714 			 * shared, and is held exclusively here.
1715 			 */
1716 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1717 			    sizeof(pr->pr_hostname));
1718 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1719 			    sizeof(pr->pr_domainname));
1720 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1721 			    sizeof(pr->pr_hostuuid));
1722 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1723 		}
1724 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1725 		/* Set this prison, and any descendants without PR_HOST. */
1726 		if (host != NULL)
1727 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1728 		if (domain != NULL)
1729 			strlcpy(pr->pr_domainname, domain,
1730 			    sizeof(pr->pr_domainname));
1731 		if (uuid != NULL)
1732 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1733 		if (gothid)
1734 			pr->pr_hostid = hid;
1735 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1736 			if (tpr->pr_flags & PR_HOST)
1737 				descend = 0;
1738 			else {
1739 				if (host != NULL)
1740 					strlcpy(tpr->pr_hostname,
1741 					    pr->pr_hostname,
1742 					    sizeof(tpr->pr_hostname));
1743 				if (domain != NULL)
1744 					strlcpy(tpr->pr_domainname,
1745 					    pr->pr_domainname,
1746 					    sizeof(tpr->pr_domainname));
1747 				if (uuid != NULL)
1748 					strlcpy(tpr->pr_hostuuid,
1749 					    pr->pr_hostuuid,
1750 					    sizeof(tpr->pr_hostuuid));
1751 				if (gothid)
1752 					tpr->pr_hostid = hid;
1753 			}
1754 		}
1755 	}
1756 	if ((tallow = ch_allow & ~pr_allow)) {
1757 		/* Clear allow bits in all children. */
1758 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1759 			tpr->pr_allow &= ~tallow;
1760 	}
1761 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1762 	/*
1763 	 * Persistent prisons get an extra reference, and prisons losing their
1764 	 * persist flag lose that reference.  Only do this for existing prisons
1765 	 * for now, so new ones will remain unseen until after the module
1766 	 * handlers have completed.
1767 	 */
1768 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1769 		if (pr_flags & PR_PERSIST) {
1770 			pr->pr_ref++;
1771 			pr->pr_uref++;
1772 		} else {
1773 			pr->pr_ref--;
1774 			pr->pr_uref--;
1775 		}
1776 	}
1777 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1778 	mtx_unlock(&pr->pr_mtx);
1779 
1780 #ifdef RACCT
1781 	if (created)
1782 		prison_racct_attach(pr);
1783 #endif
1784 
1785 	/* Locks may have prevented a complete restriction of child IP
1786 	 * addresses.  If so, allocate some more memory and try again.
1787 	 */
1788 #ifdef INET
1789 	while (redo_ip4) {
1790 		ip4s = pr->pr_ip4s;
1791 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1792 		mtx_lock(&pr->pr_mtx);
1793 		redo_ip4 = 0;
1794 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1795 #ifdef VIMAGE
1796 			if (tpr->pr_flags & PR_VNET) {
1797 				descend = 0;
1798 				continue;
1799 			}
1800 #endif
1801 			if (prison_restrict_ip4(tpr, ip4)) {
1802 				if (ip4 != NULL)
1803 					ip4 = NULL;
1804 				else
1805 					redo_ip4 = 1;
1806 			}
1807 		}
1808 		mtx_unlock(&pr->pr_mtx);
1809 	}
1810 #endif
1811 #ifdef INET6
1812 	while (redo_ip6) {
1813 		ip6s = pr->pr_ip6s;
1814 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1815 		mtx_lock(&pr->pr_mtx);
1816 		redo_ip6 = 0;
1817 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1818 #ifdef VIMAGE
1819 			if (tpr->pr_flags & PR_VNET) {
1820 				descend = 0;
1821 				continue;
1822 			}
1823 #endif
1824 			if (prison_restrict_ip6(tpr, ip6)) {
1825 				if (ip6 != NULL)
1826 					ip6 = NULL;
1827 				else
1828 					redo_ip6 = 1;
1829 			}
1830 		}
1831 		mtx_unlock(&pr->pr_mtx);
1832 	}
1833 #endif
1834 
1835 	/* Let the modules do their work. */
1836 	sx_downgrade(&allprison_lock);
1837 	if (created) {
1838 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1839 		if (error) {
1840 			prison_deref(pr, PD_LIST_SLOCKED);
1841 			goto done_errmsg;
1842 		}
1843 	}
1844 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1845 	if (error) {
1846 		prison_deref(pr, created
1847 		    ? PD_LIST_SLOCKED
1848 		    : PD_DEREF | PD_LIST_SLOCKED);
1849 		goto done_errmsg;
1850 	}
1851 
1852 	/* Attach this process to the prison if requested. */
1853 	if (flags & JAIL_ATTACH) {
1854 		mtx_lock(&pr->pr_mtx);
1855 		error = do_jail_attach(td, pr);
1856 		if (error) {
1857 			vfs_opterror(opts, "attach failed");
1858 			if (!created)
1859 				prison_deref(pr, PD_DEREF);
1860 			goto done_errmsg;
1861 		}
1862 	}
1863 
1864 #ifdef RACCT
1865 	if (!created) {
1866 		if (!(flags & JAIL_ATTACH))
1867 			sx_sunlock(&allprison_lock);
1868 		prison_racct_modify(pr);
1869 		if (!(flags & JAIL_ATTACH))
1870 			sx_slock(&allprison_lock);
1871 	}
1872 #endif
1873 
1874 	td->td_retval[0] = pr->pr_id;
1875 
1876 	/*
1877 	 * Now that it is all there, drop the temporary reference from existing
1878 	 * prisons.  Or add a reference to newly created persistent prisons
1879 	 * (which was not done earlier so that the prison would not be publicly
1880 	 * visible).
1881 	 */
1882 	if (!created) {
1883 		prison_deref(pr, (flags & JAIL_ATTACH)
1884 		    ? PD_DEREF
1885 		    : PD_DEREF | PD_LIST_SLOCKED);
1886 	} else {
1887 		if (pr_flags & PR_PERSIST) {
1888 			mtx_lock(&pr->pr_mtx);
1889 			pr->pr_ref++;
1890 			pr->pr_uref++;
1891 			mtx_unlock(&pr->pr_mtx);
1892 		}
1893 		if (!(flags & JAIL_ATTACH))
1894 			sx_sunlock(&allprison_lock);
1895 	}
1896 
1897 	goto done_errmsg;
1898 
1899  done_deref_locked:
1900 	prison_deref(pr, created
1901 	    ? PD_LOCKED | PD_LIST_XLOCKED
1902 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1903 	goto done_releroot;
1904  done_unlock_list:
1905 	sx_xunlock(&allprison_lock);
1906  done_releroot:
1907 	if (root != NULL)
1908 		vrele(root);
1909  done_errmsg:
1910 	if (error) {
1911 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1912 		if (errmsg_len > 0) {
1913 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1914 			if (errmsg_pos > 0) {
1915 				if (optuio->uio_segflg == UIO_SYSSPACE)
1916 					bcopy(errmsg,
1917 					   optuio->uio_iov[errmsg_pos].iov_base,
1918 					   errmsg_len);
1919 				else
1920 					copyout(errmsg,
1921 					   optuio->uio_iov[errmsg_pos].iov_base,
1922 					   errmsg_len);
1923 			}
1924 		}
1925 	}
1926  done_free:
1927 #ifdef INET
1928 	free(ip4, M_PRISON);
1929 #endif
1930 #ifdef INET6
1931 	free(ip6, M_PRISON);
1932 #endif
1933 	if (g_path != NULL)
1934 		free(g_path, M_TEMP);
1935 	vfs_freeopts(opts);
1936 	return (error);
1937 }
1938 
1939 
1940 /*
1941  * struct jail_get_args {
1942  *	struct iovec *iovp;
1943  *	unsigned int iovcnt;
1944  *	int flags;
1945  * };
1946  */
1947 int
1948 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1949 {
1950 	struct uio *auio;
1951 	int error;
1952 
1953 	/* Check that we have an even number of iovecs. */
1954 	if (uap->iovcnt & 1)
1955 		return (EINVAL);
1956 
1957 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1958 	if (error)
1959 		return (error);
1960 	error = kern_jail_get(td, auio, uap->flags);
1961 	if (error == 0)
1962 		error = copyout(auio->uio_iov, uap->iovp,
1963 		    uap->iovcnt * sizeof (struct iovec));
1964 	free(auio, M_IOV);
1965 	return (error);
1966 }
1967 
1968 int
1969 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1970 {
1971 	struct prison *pr, *mypr;
1972 	struct vfsopt *opt;
1973 	struct vfsoptlist *opts;
1974 	char *errmsg, *name;
1975 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1976 
1977 	if (flags & ~JAIL_GET_MASK)
1978 		return (EINVAL);
1979 
1980 	/* Get the parameter list. */
1981 	error = vfs_buildopts(optuio, &opts);
1982 	if (error)
1983 		return (error);
1984 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1985 	mypr = td->td_ucred->cr_prison;
1986 
1987 	/*
1988 	 * Find the prison specified by one of: lastjid, jid, name.
1989 	 */
1990 	sx_slock(&allprison_lock);
1991 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1992 	if (error == 0) {
1993 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1994 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1995 				mtx_lock(&pr->pr_mtx);
1996 				if (pr->pr_ref > 0 &&
1997 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1998 					break;
1999 				mtx_unlock(&pr->pr_mtx);
2000 			}
2001 		}
2002 		if (pr != NULL)
2003 			goto found_prison;
2004 		error = ENOENT;
2005 		vfs_opterror(opts, "no jail after %d", jid);
2006 		goto done_unlock_list;
2007 	} else if (error != ENOENT)
2008 		goto done_unlock_list;
2009 
2010 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2011 	if (error == 0) {
2012 		if (jid != 0) {
2013 			pr = prison_find_child(mypr, jid);
2014 			if (pr != NULL) {
2015 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2016 					mtx_unlock(&pr->pr_mtx);
2017 					error = ENOENT;
2018 					vfs_opterror(opts, "jail %d is dying",
2019 					    jid);
2020 					goto done_unlock_list;
2021 				}
2022 				goto found_prison;
2023 			}
2024 			error = ENOENT;
2025 			vfs_opterror(opts, "jail %d not found", jid);
2026 			goto done_unlock_list;
2027 		}
2028 	} else if (error != ENOENT)
2029 		goto done_unlock_list;
2030 
2031 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2032 	if (error == 0) {
2033 		if (len == 0 || name[len - 1] != '\0') {
2034 			error = EINVAL;
2035 			goto done_unlock_list;
2036 		}
2037 		pr = prison_find_name(mypr, name);
2038 		if (pr != NULL) {
2039 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2040 				mtx_unlock(&pr->pr_mtx);
2041 				error = ENOENT;
2042 				vfs_opterror(opts, "jail \"%s\" is dying",
2043 				    name);
2044 				goto done_unlock_list;
2045 			}
2046 			goto found_prison;
2047 		}
2048 		error = ENOENT;
2049 		vfs_opterror(opts, "jail \"%s\" not found", name);
2050 		goto done_unlock_list;
2051 	} else if (error != ENOENT)
2052 		goto done_unlock_list;
2053 
2054 	vfs_opterror(opts, "no jail specified");
2055 	error = ENOENT;
2056 	goto done_unlock_list;
2057 
2058  found_prison:
2059 	/* Get the parameters of the prison. */
2060 	pr->pr_ref++;
2061 	locked = PD_LOCKED;
2062 	td->td_retval[0] = pr->pr_id;
2063 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2064 	if (error != 0 && error != ENOENT)
2065 		goto done_deref;
2066 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2067 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2068 	if (error != 0 && error != ENOENT)
2069 		goto done_deref;
2070 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2071 	if (error != 0 && error != ENOENT)
2072 		goto done_deref;
2073 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2074 	    sizeof(pr->pr_cpuset->cs_id));
2075 	if (error != 0 && error != ENOENT)
2076 		goto done_deref;
2077 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2078 	if (error != 0 && error != ENOENT)
2079 		goto done_deref;
2080 #ifdef INET
2081 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2082 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2083 	if (error != 0 && error != ENOENT)
2084 		goto done_deref;
2085 #endif
2086 #ifdef INET6
2087 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2088 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2089 	if (error != 0 && error != ENOENT)
2090 		goto done_deref;
2091 #endif
2092 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2093 	    sizeof(pr->pr_securelevel));
2094 	if (error != 0 && error != ENOENT)
2095 		goto done_deref;
2096 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2097 	    sizeof(pr->pr_childcount));
2098 	if (error != 0 && error != ENOENT)
2099 		goto done_deref;
2100 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2101 	    sizeof(pr->pr_childmax));
2102 	if (error != 0 && error != ENOENT)
2103 		goto done_deref;
2104 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2105 	if (error != 0 && error != ENOENT)
2106 		goto done_deref;
2107 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2108 	if (error != 0 && error != ENOENT)
2109 		goto done_deref;
2110 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2111 	if (error != 0 && error != ENOENT)
2112 		goto done_deref;
2113 #ifdef COMPAT_FREEBSD32
2114 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2115 		uint32_t hid32 = pr->pr_hostid;
2116 
2117 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2118 	} else
2119 #endif
2120 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2121 	    sizeof(pr->pr_hostid));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2125 	    sizeof(pr->pr_enforce_statfs));
2126 	if (error != 0 && error != ENOENT)
2127 		goto done_deref;
2128 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2129 	    sizeof(pr->pr_devfs_rsnum));
2130 	if (error != 0 && error != ENOENT)
2131 		goto done_deref;
2132 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2133 	    fi++) {
2134 		if (pr_flag_names[fi] == NULL)
2135 			continue;
2136 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2137 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2138 		if (error != 0 && error != ENOENT)
2139 			goto done_deref;
2140 		i = !i;
2141 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2142 		if (error != 0 && error != ENOENT)
2143 			goto done_deref;
2144 	}
2145 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2146 	    fi++) {
2147 		i = pr->pr_flags &
2148 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2149 		i = pr_flag_jailsys[fi].disable &&
2150 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2151 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2152 		    : JAIL_SYS_INHERIT;
2153 		error =
2154 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2155 		if (error != 0 && error != ENOENT)
2156 			goto done_deref;
2157 	}
2158 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2159 	    fi++) {
2160 		if (pr_allow_names[fi] == NULL)
2161 			continue;
2162 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2163 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2164 		if (error != 0 && error != ENOENT)
2165 			goto done_deref;
2166 		i = !i;
2167 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2168 		if (error != 0 && error != ENOENT)
2169 			goto done_deref;
2170 	}
2171 	i = (pr->pr_uref == 0);
2172 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2173 	if (error != 0 && error != ENOENT)
2174 		goto done_deref;
2175 	i = !i;
2176 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2177 	if (error != 0 && error != ENOENT)
2178 		goto done_deref;
2179 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2180 	    sizeof(pr->pr_osreldate));
2181 	if (error != 0 && error != ENOENT)
2182 		goto done_deref;
2183 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2184 	if (error != 0 && error != ENOENT)
2185 		goto done_deref;
2186 
2187 	/* Get the module parameters. */
2188 	mtx_unlock(&pr->pr_mtx);
2189 	locked = 0;
2190 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2191 	if (error)
2192 		goto done_deref;
2193 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2194 
2195 	/* By now, all parameters should have been noted. */
2196 	TAILQ_FOREACH(opt, opts, link) {
2197 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2198 			error = EINVAL;
2199 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2200 			goto done_errmsg;
2201 		}
2202 	}
2203 
2204 	/* Write the fetched parameters back to userspace. */
2205 	error = 0;
2206 	TAILQ_FOREACH(opt, opts, link) {
2207 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2208 			pos = 2 * opt->pos + 1;
2209 			optuio->uio_iov[pos].iov_len = opt->len;
2210 			if (opt->value != NULL) {
2211 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2212 					bcopy(opt->value,
2213 					    optuio->uio_iov[pos].iov_base,
2214 					    opt->len);
2215 				} else {
2216 					error = copyout(opt->value,
2217 					    optuio->uio_iov[pos].iov_base,
2218 					    opt->len);
2219 					if (error)
2220 						break;
2221 				}
2222 			}
2223 		}
2224 	}
2225 	goto done_errmsg;
2226 
2227  done_deref:
2228 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2229 	goto done_errmsg;
2230 
2231  done_unlock_list:
2232 	sx_sunlock(&allprison_lock);
2233  done_errmsg:
2234 	if (error && errmsg_pos >= 0) {
2235 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2236 		errmsg_pos = 2 * errmsg_pos + 1;
2237 		if (errmsg_len > 0) {
2238 			if (optuio->uio_segflg == UIO_SYSSPACE)
2239 				bcopy(errmsg,
2240 				    optuio->uio_iov[errmsg_pos].iov_base,
2241 				    errmsg_len);
2242 			else
2243 				copyout(errmsg,
2244 				    optuio->uio_iov[errmsg_pos].iov_base,
2245 				    errmsg_len);
2246 		}
2247 	}
2248 	vfs_freeopts(opts);
2249 	return (error);
2250 }
2251 
2252 
2253 /*
2254  * struct jail_remove_args {
2255  *	int jid;
2256  * };
2257  */
2258 int
2259 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2260 {
2261 	struct prison *pr, *cpr, *lpr, *tpr;
2262 	int descend, error;
2263 
2264 	error = priv_check(td, PRIV_JAIL_REMOVE);
2265 	if (error)
2266 		return (error);
2267 
2268 	sx_xlock(&allprison_lock);
2269 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2270 	if (pr == NULL) {
2271 		sx_xunlock(&allprison_lock);
2272 		return (EINVAL);
2273 	}
2274 
2275 	/* Remove all descendants of this prison, then remove this prison. */
2276 	pr->pr_ref++;
2277 	pr->pr_flags |= PR_REMOVE;
2278 	if (!LIST_EMPTY(&pr->pr_children)) {
2279 		mtx_unlock(&pr->pr_mtx);
2280 		lpr = NULL;
2281 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2282 			mtx_lock(&cpr->pr_mtx);
2283 			if (cpr->pr_ref > 0) {
2284 				tpr = cpr;
2285 				cpr->pr_ref++;
2286 				cpr->pr_flags |= PR_REMOVE;
2287 			} else {
2288 				/* Already removed - do not do it again. */
2289 				tpr = NULL;
2290 			}
2291 			mtx_unlock(&cpr->pr_mtx);
2292 			if (lpr != NULL) {
2293 				mtx_lock(&lpr->pr_mtx);
2294 				prison_remove_one(lpr);
2295 				sx_xlock(&allprison_lock);
2296 			}
2297 			lpr = tpr;
2298 		}
2299 		if (lpr != NULL) {
2300 			mtx_lock(&lpr->pr_mtx);
2301 			prison_remove_one(lpr);
2302 			sx_xlock(&allprison_lock);
2303 		}
2304 		mtx_lock(&pr->pr_mtx);
2305 	}
2306 	prison_remove_one(pr);
2307 	return (0);
2308 }
2309 
2310 static void
2311 prison_remove_one(struct prison *pr)
2312 {
2313 	struct proc *p;
2314 	int deuref;
2315 
2316 	/* If the prison was persistent, it is not anymore. */
2317 	deuref = 0;
2318 	if (pr->pr_flags & PR_PERSIST) {
2319 		pr->pr_ref--;
2320 		deuref = PD_DEUREF;
2321 		pr->pr_flags &= ~PR_PERSIST;
2322 	}
2323 
2324 	/*
2325 	 * jail_remove added a reference.  If that's the only one, remove
2326 	 * the prison now.
2327 	 */
2328 	KASSERT(pr->pr_ref > 0,
2329 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2330 	if (pr->pr_ref == 1) {
2331 		prison_deref(pr,
2332 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2333 		return;
2334 	}
2335 
2336 	mtx_unlock(&pr->pr_mtx);
2337 	sx_xunlock(&allprison_lock);
2338 	/*
2339 	 * Kill all processes unfortunate enough to be attached to this prison.
2340 	 */
2341 	sx_slock(&allproc_lock);
2342 	LIST_FOREACH(p, &allproc, p_list) {
2343 		PROC_LOCK(p);
2344 		if (p->p_state != PRS_NEW && p->p_ucred &&
2345 		    p->p_ucred->cr_prison == pr)
2346 			kern_psignal(p, SIGKILL);
2347 		PROC_UNLOCK(p);
2348 	}
2349 	sx_sunlock(&allproc_lock);
2350 	/* Remove the temporary reference added by jail_remove. */
2351 	prison_deref(pr, deuref | PD_DEREF);
2352 }
2353 
2354 
2355 /*
2356  * struct jail_attach_args {
2357  *	int jid;
2358  * };
2359  */
2360 int
2361 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2362 {
2363 	struct prison *pr;
2364 	int error;
2365 
2366 	error = priv_check(td, PRIV_JAIL_ATTACH);
2367 	if (error)
2368 		return (error);
2369 
2370 	sx_slock(&allprison_lock);
2371 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2372 	if (pr == NULL) {
2373 		sx_sunlock(&allprison_lock);
2374 		return (EINVAL);
2375 	}
2376 
2377 	/*
2378 	 * Do not allow a process to attach to a prison that is not
2379 	 * considered to be "alive".
2380 	 */
2381 	if (pr->pr_uref == 0) {
2382 		mtx_unlock(&pr->pr_mtx);
2383 		sx_sunlock(&allprison_lock);
2384 		return (EINVAL);
2385 	}
2386 
2387 	return (do_jail_attach(td, pr));
2388 }
2389 
2390 static int
2391 do_jail_attach(struct thread *td, struct prison *pr)
2392 {
2393 	struct prison *ppr;
2394 	struct proc *p;
2395 	struct ucred *newcred, *oldcred;
2396 	int error;
2397 
2398 	/*
2399 	 * XXX: Note that there is a slight race here if two threads
2400 	 * in the same privileged process attempt to attach to two
2401 	 * different jails at the same time.  It is important for
2402 	 * user processes not to do this, or they might end up with
2403 	 * a process root from one prison, but attached to the jail
2404 	 * of another.
2405 	 */
2406 	pr->pr_ref++;
2407 	pr->pr_uref++;
2408 	mtx_unlock(&pr->pr_mtx);
2409 
2410 	/* Let modules do whatever they need to prepare for attaching. */
2411 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2412 	if (error) {
2413 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2414 		return (error);
2415 	}
2416 	sx_sunlock(&allprison_lock);
2417 
2418 	/*
2419 	 * Reparent the newly attached process to this jail.
2420 	 */
2421 	ppr = td->td_ucred->cr_prison;
2422 	p = td->td_proc;
2423 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2424 	if (error)
2425 		goto e_revert_osd;
2426 
2427 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2428 	if ((error = change_dir(pr->pr_root, td)) != 0)
2429 		goto e_unlock;
2430 #ifdef MAC
2431 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2432 		goto e_unlock;
2433 #endif
2434 	VOP_UNLOCK(pr->pr_root, 0);
2435 	if ((error = change_root(pr->pr_root, td)))
2436 		goto e_revert_osd;
2437 
2438 	newcred = crget();
2439 	PROC_LOCK(p);
2440 	oldcred = p->p_ucred;
2441 	setsugid(p);
2442 	crcopy(newcred, oldcred);
2443 	newcred->cr_prison = pr;
2444 	proc_set_cred(p, newcred);
2445 	PROC_UNLOCK(p);
2446 #ifdef RACCT
2447 	racct_proc_ucred_changed(p, oldcred, newcred);
2448 #endif
2449 	crfree(oldcred);
2450 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2451 	return (0);
2452  e_unlock:
2453 	VOP_UNLOCK(pr->pr_root, 0);
2454  e_revert_osd:
2455 	/* Tell modules this thread is still in its old jail after all. */
2456 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2457 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2458 	return (error);
2459 }
2460 
2461 
2462 /*
2463  * Returns a locked prison instance, or NULL on failure.
2464  */
2465 struct prison *
2466 prison_find(int prid)
2467 {
2468 	struct prison *pr;
2469 
2470 	sx_assert(&allprison_lock, SX_LOCKED);
2471 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2472 		if (pr->pr_id == prid) {
2473 			mtx_lock(&pr->pr_mtx);
2474 			if (pr->pr_ref > 0)
2475 				return (pr);
2476 			mtx_unlock(&pr->pr_mtx);
2477 		}
2478 	}
2479 	return (NULL);
2480 }
2481 
2482 /*
2483  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2484  */
2485 struct prison *
2486 prison_find_child(struct prison *mypr, int prid)
2487 {
2488 	struct prison *pr;
2489 	int descend;
2490 
2491 	sx_assert(&allprison_lock, SX_LOCKED);
2492 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2493 		if (pr->pr_id == prid) {
2494 			mtx_lock(&pr->pr_mtx);
2495 			if (pr->pr_ref > 0)
2496 				return (pr);
2497 			mtx_unlock(&pr->pr_mtx);
2498 		}
2499 	}
2500 	return (NULL);
2501 }
2502 
2503 /*
2504  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2505  */
2506 struct prison *
2507 prison_find_name(struct prison *mypr, const char *name)
2508 {
2509 	struct prison *pr, *deadpr;
2510 	size_t mylen;
2511 	int descend;
2512 
2513 	sx_assert(&allprison_lock, SX_LOCKED);
2514 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2515  again:
2516 	deadpr = NULL;
2517 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2518 		if (!strcmp(pr->pr_name + mylen, name)) {
2519 			mtx_lock(&pr->pr_mtx);
2520 			if (pr->pr_ref > 0) {
2521 				if (pr->pr_uref > 0)
2522 					return (pr);
2523 				deadpr = pr;
2524 			}
2525 			mtx_unlock(&pr->pr_mtx);
2526 		}
2527 	}
2528 	/* There was no valid prison - perhaps there was a dying one. */
2529 	if (deadpr != NULL) {
2530 		mtx_lock(&deadpr->pr_mtx);
2531 		if (deadpr->pr_ref == 0) {
2532 			mtx_unlock(&deadpr->pr_mtx);
2533 			goto again;
2534 		}
2535 	}
2536 	return (deadpr);
2537 }
2538 
2539 /*
2540  * See if a prison has the specific flag set.
2541  */
2542 int
2543 prison_flag(struct ucred *cred, unsigned flag)
2544 {
2545 
2546 	/* This is an atomic read, so no locking is necessary. */
2547 	return (cred->cr_prison->pr_flags & flag);
2548 }
2549 
2550 int
2551 prison_allow(struct ucred *cred, unsigned flag)
2552 {
2553 
2554 	/* This is an atomic read, so no locking is necessary. */
2555 	return (cred->cr_prison->pr_allow & flag);
2556 }
2557 
2558 /*
2559  * Remove a prison reference.  If that was the last reference, remove the
2560  * prison itself - but not in this context in case there are locks held.
2561  */
2562 void
2563 prison_free_locked(struct prison *pr)
2564 {
2565 
2566 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2567 	pr->pr_ref--;
2568 	if (pr->pr_ref == 0) {
2569 		mtx_unlock(&pr->pr_mtx);
2570 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2571 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2572 		return;
2573 	}
2574 	mtx_unlock(&pr->pr_mtx);
2575 }
2576 
2577 void
2578 prison_free(struct prison *pr)
2579 {
2580 
2581 	mtx_lock(&pr->pr_mtx);
2582 	prison_free_locked(pr);
2583 }
2584 
2585 static void
2586 prison_complete(void *context, int pending)
2587 {
2588 
2589 	prison_deref((struct prison *)context, 0);
2590 }
2591 
2592 /*
2593  * Remove a prison reference (usually).  This internal version assumes no
2594  * mutexes are held, except perhaps the prison itself.  If there are no more
2595  * references, release and delist the prison.  On completion, the prison lock
2596  * and the allprison lock are both unlocked.
2597  */
2598 static void
2599 prison_deref(struct prison *pr, int flags)
2600 {
2601 	struct prison *ppr, *tpr;
2602 
2603 	if (!(flags & PD_LOCKED))
2604 		mtx_lock(&pr->pr_mtx);
2605 	for (;;) {
2606 		if (flags & PD_DEUREF) {
2607 			pr->pr_uref--;
2608 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2609 		}
2610 		if (flags & PD_DEREF)
2611 			pr->pr_ref--;
2612 		/* If the prison still has references, nothing else to do. */
2613 		if (pr->pr_ref > 0) {
2614 			mtx_unlock(&pr->pr_mtx);
2615 			if (flags & PD_LIST_SLOCKED)
2616 				sx_sunlock(&allprison_lock);
2617 			else if (flags & PD_LIST_XLOCKED)
2618 				sx_xunlock(&allprison_lock);
2619 			return;
2620 		}
2621 
2622 		mtx_unlock(&pr->pr_mtx);
2623 		if (flags & PD_LIST_SLOCKED) {
2624 			if (!sx_try_upgrade(&allprison_lock)) {
2625 				sx_sunlock(&allprison_lock);
2626 				sx_xlock(&allprison_lock);
2627 			}
2628 		} else if (!(flags & PD_LIST_XLOCKED))
2629 			sx_xlock(&allprison_lock);
2630 
2631 		TAILQ_REMOVE(&allprison, pr, pr_list);
2632 		LIST_REMOVE(pr, pr_sibling);
2633 		ppr = pr->pr_parent;
2634 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2635 			tpr->pr_childcount--;
2636 		sx_xunlock(&allprison_lock);
2637 
2638 #ifdef VIMAGE
2639 		if (pr->pr_vnet != ppr->pr_vnet)
2640 			vnet_destroy(pr->pr_vnet);
2641 #endif
2642 		if (pr->pr_root != NULL)
2643 			vrele(pr->pr_root);
2644 		mtx_destroy(&pr->pr_mtx);
2645 #ifdef INET
2646 		free(pr->pr_ip4, M_PRISON);
2647 #endif
2648 #ifdef INET6
2649 		free(pr->pr_ip6, M_PRISON);
2650 #endif
2651 		if (pr->pr_cpuset != NULL)
2652 			cpuset_rel(pr->pr_cpuset);
2653 		osd_jail_exit(pr);
2654 #ifdef RACCT
2655 		prison_racct_detach(pr);
2656 #endif
2657 		free(pr, M_PRISON);
2658 
2659 		/* Removing a prison frees a reference on its parent. */
2660 		pr = ppr;
2661 		mtx_lock(&pr->pr_mtx);
2662 		flags = PD_DEREF | PD_DEUREF;
2663 	}
2664 }
2665 
2666 void
2667 prison_hold_locked(struct prison *pr)
2668 {
2669 
2670 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2671 	KASSERT(pr->pr_ref > 0,
2672 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2673 	pr->pr_ref++;
2674 }
2675 
2676 void
2677 prison_hold(struct prison *pr)
2678 {
2679 
2680 	mtx_lock(&pr->pr_mtx);
2681 	prison_hold_locked(pr);
2682 	mtx_unlock(&pr->pr_mtx);
2683 }
2684 
2685 void
2686 prison_proc_hold(struct prison *pr)
2687 {
2688 
2689 	mtx_lock(&pr->pr_mtx);
2690 	KASSERT(pr->pr_uref > 0,
2691 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2692 	pr->pr_uref++;
2693 	mtx_unlock(&pr->pr_mtx);
2694 }
2695 
2696 void
2697 prison_proc_free(struct prison *pr)
2698 {
2699 
2700 	mtx_lock(&pr->pr_mtx);
2701 	KASSERT(pr->pr_uref > 0,
2702 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2703 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2704 }
2705 
2706 
2707 #ifdef INET
2708 /*
2709  * Restrict a prison's IP address list with its parent's, possibly replacing
2710  * it.  Return true if the replacement buffer was used (or would have been).
2711  */
2712 static int
2713 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2714 {
2715 	int ii, ij, used;
2716 	struct prison *ppr;
2717 
2718 	ppr = pr->pr_parent;
2719 	if (!(pr->pr_flags & PR_IP4_USER)) {
2720 		/* This has no user settings, so just copy the parent's list. */
2721 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2722 			/*
2723 			 * There's no room for the parent's list.  Use the
2724 			 * new list buffer, which is assumed to be big enough
2725 			 * (if it was passed).  If there's no buffer, try to
2726 			 * allocate one.
2727 			 */
2728 			used = 1;
2729 			if (newip4 == NULL) {
2730 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2731 				    M_PRISON, M_NOWAIT);
2732 				if (newip4 != NULL)
2733 					used = 0;
2734 			}
2735 			if (newip4 != NULL) {
2736 				bcopy(ppr->pr_ip4, newip4,
2737 				    ppr->pr_ip4s * sizeof(*newip4));
2738 				free(pr->pr_ip4, M_PRISON);
2739 				pr->pr_ip4 = newip4;
2740 				pr->pr_ip4s = ppr->pr_ip4s;
2741 			}
2742 			return (used);
2743 		}
2744 		pr->pr_ip4s = ppr->pr_ip4s;
2745 		if (pr->pr_ip4s > 0)
2746 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2747 			    pr->pr_ip4s * sizeof(*newip4));
2748 		else if (pr->pr_ip4 != NULL) {
2749 			free(pr->pr_ip4, M_PRISON);
2750 			pr->pr_ip4 = NULL;
2751 		}
2752 	} else if (pr->pr_ip4s > 0) {
2753 		/* Remove addresses that aren't in the parent. */
2754 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2755 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2756 				break;
2757 		if (ij < ppr->pr_ip4s)
2758 			ii = 1;
2759 		else {
2760 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2761 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2762 			ii = 0;
2763 		}
2764 		for (ij = 1; ii < pr->pr_ip4s; ) {
2765 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2766 				ii++;
2767 				continue;
2768 			}
2769 			switch (ij >= ppr->pr_ip4s ? -1 :
2770 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2771 			case -1:
2772 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2773 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2774 				break;
2775 			case 0:
2776 				ii++;
2777 				ij++;
2778 				break;
2779 			case 1:
2780 				ij++;
2781 				break;
2782 			}
2783 		}
2784 		if (pr->pr_ip4s == 0) {
2785 			free(pr->pr_ip4, M_PRISON);
2786 			pr->pr_ip4 = NULL;
2787 		}
2788 	}
2789 	return (0);
2790 }
2791 
2792 /*
2793  * Pass back primary IPv4 address of this jail.
2794  *
2795  * If not restricted return success but do not alter the address.  Caller has
2796  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2797  *
2798  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2799  * Address returned in NBO.
2800  */
2801 int
2802 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2803 {
2804 	struct prison *pr;
2805 
2806 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2807 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2808 
2809 	pr = cred->cr_prison;
2810 	if (!(pr->pr_flags & PR_IP4))
2811 		return (0);
2812 	mtx_lock(&pr->pr_mtx);
2813 	if (!(pr->pr_flags & PR_IP4)) {
2814 		mtx_unlock(&pr->pr_mtx);
2815 		return (0);
2816 	}
2817 	if (pr->pr_ip4 == NULL) {
2818 		mtx_unlock(&pr->pr_mtx);
2819 		return (EAFNOSUPPORT);
2820 	}
2821 
2822 	ia->s_addr = pr->pr_ip4[0].s_addr;
2823 	mtx_unlock(&pr->pr_mtx);
2824 	return (0);
2825 }
2826 
2827 /*
2828  * Return 1 if we should do proper source address selection or are not jailed.
2829  * We will return 0 if we should bypass source address selection in favour
2830  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2831  * returned in NBO.
2832  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2833  */
2834 int
2835 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2836 {
2837 	struct prison *pr;
2838 	struct in_addr lia;
2839 	int error;
2840 
2841 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2842 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2843 
2844 	if (!jailed(cred))
2845 		return (1);
2846 
2847 	pr = cred->cr_prison;
2848 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2849 		return (1);
2850 
2851 	lia.s_addr = INADDR_ANY;
2852 	error = prison_get_ip4(cred, &lia);
2853 	if (error)
2854 		return (error);
2855 	if (lia.s_addr == INADDR_ANY)
2856 		return (1);
2857 
2858 	ia->s_addr = lia.s_addr;
2859 	return (0);
2860 }
2861 
2862 /*
2863  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2864  */
2865 int
2866 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2867 {
2868 
2869 	if (pr1 == pr2)
2870 		return (1);
2871 
2872 	/*
2873 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2874 	 * existing prisons.
2875 	 */
2876 	while (pr1 != &prison0 &&
2877 #ifdef VIMAGE
2878 	       !(pr1->pr_flags & PR_VNET) &&
2879 #endif
2880 	       !(pr1->pr_flags & PR_IP4_USER))
2881 		pr1 = pr1->pr_parent;
2882 	while (pr2 != &prison0 &&
2883 #ifdef VIMAGE
2884 	       !(pr2->pr_flags & PR_VNET) &&
2885 #endif
2886 	       !(pr2->pr_flags & PR_IP4_USER))
2887 		pr2 = pr2->pr_parent;
2888 	return (pr1 == pr2);
2889 }
2890 
2891 /*
2892  * Make sure our (source) address is set to something meaningful to this
2893  * jail.
2894  *
2895  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2896  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2897  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2898  */
2899 int
2900 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2901 {
2902 	struct prison *pr;
2903 	struct in_addr ia0;
2904 	int error;
2905 
2906 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2907 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2908 
2909 	pr = cred->cr_prison;
2910 	if (!(pr->pr_flags & PR_IP4))
2911 		return (0);
2912 	mtx_lock(&pr->pr_mtx);
2913 	if (!(pr->pr_flags & PR_IP4)) {
2914 		mtx_unlock(&pr->pr_mtx);
2915 		return (0);
2916 	}
2917 	if (pr->pr_ip4 == NULL) {
2918 		mtx_unlock(&pr->pr_mtx);
2919 		return (EAFNOSUPPORT);
2920 	}
2921 
2922 	ia0.s_addr = ntohl(ia->s_addr);
2923 	if (ia0.s_addr == INADDR_LOOPBACK) {
2924 		ia->s_addr = pr->pr_ip4[0].s_addr;
2925 		mtx_unlock(&pr->pr_mtx);
2926 		return (0);
2927 	}
2928 
2929 	if (ia0.s_addr == INADDR_ANY) {
2930 		/*
2931 		 * In case there is only 1 IPv4 address, bind directly.
2932 		 */
2933 		if (pr->pr_ip4s == 1)
2934 			ia->s_addr = pr->pr_ip4[0].s_addr;
2935 		mtx_unlock(&pr->pr_mtx);
2936 		return (0);
2937 	}
2938 
2939 	error = _prison_check_ip4(pr, ia);
2940 	mtx_unlock(&pr->pr_mtx);
2941 	return (error);
2942 }
2943 
2944 /*
2945  * Rewrite destination address in case we will connect to loopback address.
2946  *
2947  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2948  * Address passed in in NBO and returned in NBO.
2949  */
2950 int
2951 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2952 {
2953 	struct prison *pr;
2954 
2955 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2956 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2957 
2958 	pr = cred->cr_prison;
2959 	if (!(pr->pr_flags & PR_IP4))
2960 		return (0);
2961 	mtx_lock(&pr->pr_mtx);
2962 	if (!(pr->pr_flags & PR_IP4)) {
2963 		mtx_unlock(&pr->pr_mtx);
2964 		return (0);
2965 	}
2966 	if (pr->pr_ip4 == NULL) {
2967 		mtx_unlock(&pr->pr_mtx);
2968 		return (EAFNOSUPPORT);
2969 	}
2970 
2971 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2972 		ia->s_addr = pr->pr_ip4[0].s_addr;
2973 		mtx_unlock(&pr->pr_mtx);
2974 		return (0);
2975 	}
2976 
2977 	/*
2978 	 * Return success because nothing had to be changed.
2979 	 */
2980 	mtx_unlock(&pr->pr_mtx);
2981 	return (0);
2982 }
2983 
2984 /*
2985  * Check if given address belongs to the jail referenced by cred/prison.
2986  *
2987  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2988  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2989  * doesn't allow IPv4.  Address passed in in NBO.
2990  */
2991 static int
2992 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
2993 {
2994 	int i, a, z, d;
2995 
2996 	/*
2997 	 * Check the primary IP.
2998 	 */
2999 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
3000 		return (0);
3001 
3002 	/*
3003 	 * All the other IPs are sorted so we can do a binary search.
3004 	 */
3005 	a = 0;
3006 	z = pr->pr_ip4s - 2;
3007 	while (a <= z) {
3008 		i = (a + z) / 2;
3009 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3010 		if (d > 0)
3011 			z = i - 1;
3012 		else if (d < 0)
3013 			a = i + 1;
3014 		else
3015 			return (0);
3016 	}
3017 
3018 	return (EADDRNOTAVAIL);
3019 }
3020 
3021 int
3022 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
3023 {
3024 	struct prison *pr;
3025 	int error;
3026 
3027 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3028 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3029 
3030 	pr = cred->cr_prison;
3031 	if (!(pr->pr_flags & PR_IP4))
3032 		return (0);
3033 	mtx_lock(&pr->pr_mtx);
3034 	if (!(pr->pr_flags & PR_IP4)) {
3035 		mtx_unlock(&pr->pr_mtx);
3036 		return (0);
3037 	}
3038 	if (pr->pr_ip4 == NULL) {
3039 		mtx_unlock(&pr->pr_mtx);
3040 		return (EAFNOSUPPORT);
3041 	}
3042 
3043 	error = _prison_check_ip4(pr, ia);
3044 	mtx_unlock(&pr->pr_mtx);
3045 	return (error);
3046 }
3047 #endif
3048 
3049 #ifdef INET6
3050 static int
3051 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3052 {
3053 	int ii, ij, used;
3054 	struct prison *ppr;
3055 
3056 	ppr = pr->pr_parent;
3057 	if (!(pr->pr_flags & PR_IP6_USER)) {
3058 		/* This has no user settings, so just copy the parent's list. */
3059 		if (pr->pr_ip6s < ppr->pr_ip6s) {
3060 			/*
3061 			 * There's no room for the parent's list.  Use the
3062 			 * new list buffer, which is assumed to be big enough
3063 			 * (if it was passed).  If there's no buffer, try to
3064 			 * allocate one.
3065 			 */
3066 			used = 1;
3067 			if (newip6 == NULL) {
3068 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3069 				    M_PRISON, M_NOWAIT);
3070 				if (newip6 != NULL)
3071 					used = 0;
3072 			}
3073 			if (newip6 != NULL) {
3074 				bcopy(ppr->pr_ip6, newip6,
3075 				    ppr->pr_ip6s * sizeof(*newip6));
3076 				free(pr->pr_ip6, M_PRISON);
3077 				pr->pr_ip6 = newip6;
3078 				pr->pr_ip6s = ppr->pr_ip6s;
3079 			}
3080 			return (used);
3081 		}
3082 		pr->pr_ip6s = ppr->pr_ip6s;
3083 		if (pr->pr_ip6s > 0)
3084 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3085 			    pr->pr_ip6s * sizeof(*newip6));
3086 		else if (pr->pr_ip6 != NULL) {
3087 			free(pr->pr_ip6, M_PRISON);
3088 			pr->pr_ip6 = NULL;
3089 		}
3090 	} else if (pr->pr_ip6s > 0) {
3091 		/* Remove addresses that aren't in the parent. */
3092 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3093 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3094 			    &ppr->pr_ip6[ij]))
3095 				break;
3096 		if (ij < ppr->pr_ip6s)
3097 			ii = 1;
3098 		else {
3099 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3100 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3101 			ii = 0;
3102 		}
3103 		for (ij = 1; ii < pr->pr_ip6s; ) {
3104 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3105 			    &ppr->pr_ip6[0])) {
3106 				ii++;
3107 				continue;
3108 			}
3109 			switch (ij >= ppr->pr_ip6s ? -1 :
3110 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3111 			case -1:
3112 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3113 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3114 				break;
3115 			case 0:
3116 				ii++;
3117 				ij++;
3118 				break;
3119 			case 1:
3120 				ij++;
3121 				break;
3122 			}
3123 		}
3124 		if (pr->pr_ip6s == 0) {
3125 			free(pr->pr_ip6, M_PRISON);
3126 			pr->pr_ip6 = NULL;
3127 		}
3128 	}
3129 	return 0;
3130 }
3131 
3132 /*
3133  * Pass back primary IPv6 address for this jail.
3134  *
3135  * If not restricted return success but do not alter the address.  Caller has
3136  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3137  *
3138  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3139  */
3140 int
3141 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3142 {
3143 	struct prison *pr;
3144 
3145 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3146 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3147 
3148 	pr = cred->cr_prison;
3149 	if (!(pr->pr_flags & PR_IP6))
3150 		return (0);
3151 	mtx_lock(&pr->pr_mtx);
3152 	if (!(pr->pr_flags & PR_IP6)) {
3153 		mtx_unlock(&pr->pr_mtx);
3154 		return (0);
3155 	}
3156 	if (pr->pr_ip6 == NULL) {
3157 		mtx_unlock(&pr->pr_mtx);
3158 		return (EAFNOSUPPORT);
3159 	}
3160 
3161 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3162 	mtx_unlock(&pr->pr_mtx);
3163 	return (0);
3164 }
3165 
3166 /*
3167  * Return 1 if we should do proper source address selection or are not jailed.
3168  * We will return 0 if we should bypass source address selection in favour
3169  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3170  * returned in NBO.
3171  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3172  */
3173 int
3174 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3175 {
3176 	struct prison *pr;
3177 	struct in6_addr lia6;
3178 	int error;
3179 
3180 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3181 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3182 
3183 	if (!jailed(cred))
3184 		return (1);
3185 
3186 	pr = cred->cr_prison;
3187 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3188 		return (1);
3189 
3190 	lia6 = in6addr_any;
3191 	error = prison_get_ip6(cred, &lia6);
3192 	if (error)
3193 		return (error);
3194 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3195 		return (1);
3196 
3197 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3198 	return (0);
3199 }
3200 
3201 /*
3202  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3203  */
3204 int
3205 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3206 {
3207 
3208 	if (pr1 == pr2)
3209 		return (1);
3210 
3211 	while (pr1 != &prison0 &&
3212 #ifdef VIMAGE
3213 	       !(pr1->pr_flags & PR_VNET) &&
3214 #endif
3215 	       !(pr1->pr_flags & PR_IP6_USER))
3216 		pr1 = pr1->pr_parent;
3217 	while (pr2 != &prison0 &&
3218 #ifdef VIMAGE
3219 	       !(pr2->pr_flags & PR_VNET) &&
3220 #endif
3221 	       !(pr2->pr_flags & PR_IP6_USER))
3222 		pr2 = pr2->pr_parent;
3223 	return (pr1 == pr2);
3224 }
3225 
3226 /*
3227  * Make sure our (source) address is set to something meaningful to this jail.
3228  *
3229  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3230  * when needed while binding.
3231  *
3232  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3233  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3234  * doesn't allow IPv6.
3235  */
3236 int
3237 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3238 {
3239 	struct prison *pr;
3240 	int error;
3241 
3242 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3243 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3244 
3245 	pr = cred->cr_prison;
3246 	if (!(pr->pr_flags & PR_IP6))
3247 		return (0);
3248 	mtx_lock(&pr->pr_mtx);
3249 	if (!(pr->pr_flags & PR_IP6)) {
3250 		mtx_unlock(&pr->pr_mtx);
3251 		return (0);
3252 	}
3253 	if (pr->pr_ip6 == NULL) {
3254 		mtx_unlock(&pr->pr_mtx);
3255 		return (EAFNOSUPPORT);
3256 	}
3257 
3258 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3259 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3260 		mtx_unlock(&pr->pr_mtx);
3261 		return (0);
3262 	}
3263 
3264 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3265 		/*
3266 		 * In case there is only 1 IPv6 address, and v6only is true,
3267 		 * then bind directly.
3268 		 */
3269 		if (v6only != 0 && pr->pr_ip6s == 1)
3270 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3271 		mtx_unlock(&pr->pr_mtx);
3272 		return (0);
3273 	}
3274 
3275 	error = _prison_check_ip6(pr, ia6);
3276 	mtx_unlock(&pr->pr_mtx);
3277 	return (error);
3278 }
3279 
3280 /*
3281  * Rewrite destination address in case we will connect to loopback address.
3282  *
3283  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3284  */
3285 int
3286 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3287 {
3288 	struct prison *pr;
3289 
3290 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3291 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3292 
3293 	pr = cred->cr_prison;
3294 	if (!(pr->pr_flags & PR_IP6))
3295 		return (0);
3296 	mtx_lock(&pr->pr_mtx);
3297 	if (!(pr->pr_flags & PR_IP6)) {
3298 		mtx_unlock(&pr->pr_mtx);
3299 		return (0);
3300 	}
3301 	if (pr->pr_ip6 == NULL) {
3302 		mtx_unlock(&pr->pr_mtx);
3303 		return (EAFNOSUPPORT);
3304 	}
3305 
3306 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3307 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3308 		mtx_unlock(&pr->pr_mtx);
3309 		return (0);
3310 	}
3311 
3312 	/*
3313 	 * Return success because nothing had to be changed.
3314 	 */
3315 	mtx_unlock(&pr->pr_mtx);
3316 	return (0);
3317 }
3318 
3319 /*
3320  * Check if given address belongs to the jail referenced by cred/prison.
3321  *
3322  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3323  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3324  * doesn't allow IPv6.
3325  */
3326 static int
3327 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3328 {
3329 	int i, a, z, d;
3330 
3331 	/*
3332 	 * Check the primary IP.
3333 	 */
3334 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3335 		return (0);
3336 
3337 	/*
3338 	 * All the other IPs are sorted so we can do a binary search.
3339 	 */
3340 	a = 0;
3341 	z = pr->pr_ip6s - 2;
3342 	while (a <= z) {
3343 		i = (a + z) / 2;
3344 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3345 		if (d > 0)
3346 			z = i - 1;
3347 		else if (d < 0)
3348 			a = i + 1;
3349 		else
3350 			return (0);
3351 	}
3352 
3353 	return (EADDRNOTAVAIL);
3354 }
3355 
3356 int
3357 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3358 {
3359 	struct prison *pr;
3360 	int error;
3361 
3362 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3363 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3364 
3365 	pr = cred->cr_prison;
3366 	if (!(pr->pr_flags & PR_IP6))
3367 		return (0);
3368 	mtx_lock(&pr->pr_mtx);
3369 	if (!(pr->pr_flags & PR_IP6)) {
3370 		mtx_unlock(&pr->pr_mtx);
3371 		return (0);
3372 	}
3373 	if (pr->pr_ip6 == NULL) {
3374 		mtx_unlock(&pr->pr_mtx);
3375 		return (EAFNOSUPPORT);
3376 	}
3377 
3378 	error = _prison_check_ip6(pr, ia6);
3379 	mtx_unlock(&pr->pr_mtx);
3380 	return (error);
3381 }
3382 #endif
3383 
3384 /*
3385  * Check if a jail supports the given address family.
3386  *
3387  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3388  * if not.
3389  */
3390 int
3391 prison_check_af(struct ucred *cred, int af)
3392 {
3393 	struct prison *pr;
3394 	int error;
3395 
3396 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3397 
3398 	pr = cred->cr_prison;
3399 #ifdef VIMAGE
3400 	/* Prisons with their own network stack are not limited. */
3401 	if (prison_owns_vnet(cred))
3402 		return (0);
3403 #endif
3404 
3405 	error = 0;
3406 	switch (af)
3407 	{
3408 #ifdef INET
3409 	case AF_INET:
3410 		if (pr->pr_flags & PR_IP4)
3411 		{
3412 			mtx_lock(&pr->pr_mtx);
3413 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3414 				error = EAFNOSUPPORT;
3415 			mtx_unlock(&pr->pr_mtx);
3416 		}
3417 		break;
3418 #endif
3419 #ifdef INET6
3420 	case AF_INET6:
3421 		if (pr->pr_flags & PR_IP6)
3422 		{
3423 			mtx_lock(&pr->pr_mtx);
3424 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3425 				error = EAFNOSUPPORT;
3426 			mtx_unlock(&pr->pr_mtx);
3427 		}
3428 		break;
3429 #endif
3430 	case AF_LOCAL:
3431 	case AF_ROUTE:
3432 		break;
3433 	default:
3434 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3435 			error = EAFNOSUPPORT;
3436 	}
3437 	return (error);
3438 }
3439 
3440 /*
3441  * Check if given address belongs to the jail referenced by cred (wrapper to
3442  * prison_check_ip[46]).
3443  *
3444  * Returns 0 if jail doesn't restrict the address family or if address belongs
3445  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3446  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3447  */
3448 int
3449 prison_if(struct ucred *cred, struct sockaddr *sa)
3450 {
3451 #ifdef INET
3452 	struct sockaddr_in *sai;
3453 #endif
3454 #ifdef INET6
3455 	struct sockaddr_in6 *sai6;
3456 #endif
3457 	int error;
3458 
3459 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3460 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3461 
3462 #ifdef VIMAGE
3463 	if (prison_owns_vnet(cred))
3464 		return (0);
3465 #endif
3466 
3467 	error = 0;
3468 	switch (sa->sa_family)
3469 	{
3470 #ifdef INET
3471 	case AF_INET:
3472 		sai = (struct sockaddr_in *)sa;
3473 		error = prison_check_ip4(cred, &sai->sin_addr);
3474 		break;
3475 #endif
3476 #ifdef INET6
3477 	case AF_INET6:
3478 		sai6 = (struct sockaddr_in6 *)sa;
3479 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3480 		break;
3481 #endif
3482 	default:
3483 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3484 			error = EAFNOSUPPORT;
3485 	}
3486 	return (error);
3487 }
3488 
3489 /*
3490  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3491  */
3492 int
3493 prison_check(struct ucred *cred1, struct ucred *cred2)
3494 {
3495 
3496 	return ((cred1->cr_prison == cred2->cr_prison ||
3497 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3498 }
3499 
3500 /*
3501  * Return 1 if p2 is a child of p1, otherwise 0.
3502  */
3503 int
3504 prison_ischild(struct prison *pr1, struct prison *pr2)
3505 {
3506 
3507 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3508 		if (pr1 == pr2)
3509 			return (1);
3510 	return (0);
3511 }
3512 
3513 /*
3514  * Return 1 if the passed credential is in a jail, otherwise 0.
3515  */
3516 int
3517 jailed(struct ucred *cred)
3518 {
3519 
3520 	return (cred->cr_prison != &prison0);
3521 }
3522 
3523 /*
3524  * Return 1 if the passed credential is in a jail and that jail does not
3525  * have its own virtual network stack, otherwise 0.
3526  */
3527 int
3528 jailed_without_vnet(struct ucred *cred)
3529 {
3530 
3531 	if (!jailed(cred))
3532 		return (0);
3533 #ifdef VIMAGE
3534 	if (prison_owns_vnet(cred))
3535 		return (0);
3536 #endif
3537 
3538 	return (1);
3539 }
3540 
3541 /*
3542  * Return the correct hostname (domainname, et al) for the passed credential.
3543  */
3544 void
3545 getcredhostname(struct ucred *cred, char *buf, size_t size)
3546 {
3547 	struct prison *pr;
3548 
3549 	/*
3550 	 * A NULL credential can be used to shortcut to the physical
3551 	 * system's hostname.
3552 	 */
3553 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3554 	mtx_lock(&pr->pr_mtx);
3555 	strlcpy(buf, pr->pr_hostname, size);
3556 	mtx_unlock(&pr->pr_mtx);
3557 }
3558 
3559 void
3560 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3561 {
3562 
3563 	mtx_lock(&cred->cr_prison->pr_mtx);
3564 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3565 	mtx_unlock(&cred->cr_prison->pr_mtx);
3566 }
3567 
3568 void
3569 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3570 {
3571 
3572 	mtx_lock(&cred->cr_prison->pr_mtx);
3573 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3574 	mtx_unlock(&cred->cr_prison->pr_mtx);
3575 }
3576 
3577 void
3578 getcredhostid(struct ucred *cred, unsigned long *hostid)
3579 {
3580 
3581 	mtx_lock(&cred->cr_prison->pr_mtx);
3582 	*hostid = cred->cr_prison->pr_hostid;
3583 	mtx_unlock(&cred->cr_prison->pr_mtx);
3584 }
3585 
3586 #ifdef VIMAGE
3587 /*
3588  * Determine whether the prison represented by cred owns
3589  * its vnet rather than having it inherited.
3590  *
3591  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3592  */
3593 int
3594 prison_owns_vnet(struct ucred *cred)
3595 {
3596 
3597 	/*
3598 	 * vnets cannot be added/removed after jail creation,
3599 	 * so no need to lock here.
3600 	 */
3601 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3602 }
3603 #endif
3604 
3605 /*
3606  * Determine whether the subject represented by cred can "see"
3607  * status of a mount point.
3608  * Returns: 0 for permitted, ENOENT otherwise.
3609  * XXX: This function should be called cr_canseemount() and should be
3610  *      placed in kern_prot.c.
3611  */
3612 int
3613 prison_canseemount(struct ucred *cred, struct mount *mp)
3614 {
3615 	struct prison *pr;
3616 	struct statfs *sp;
3617 	size_t len;
3618 
3619 	pr = cred->cr_prison;
3620 	if (pr->pr_enforce_statfs == 0)
3621 		return (0);
3622 	if (pr->pr_root->v_mount == mp)
3623 		return (0);
3624 	if (pr->pr_enforce_statfs == 2)
3625 		return (ENOENT);
3626 	/*
3627 	 * If jail's chroot directory is set to "/" we should be able to see
3628 	 * all mount-points from inside a jail.
3629 	 * This is ugly check, but this is the only situation when jail's
3630 	 * directory ends with '/'.
3631 	 */
3632 	if (strcmp(pr->pr_path, "/") == 0)
3633 		return (0);
3634 	len = strlen(pr->pr_path);
3635 	sp = &mp->mnt_stat;
3636 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3637 		return (ENOENT);
3638 	/*
3639 	 * Be sure that we don't have situation where jail's root directory
3640 	 * is "/some/path" and mount point is "/some/pathpath".
3641 	 */
3642 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3643 		return (ENOENT);
3644 	return (0);
3645 }
3646 
3647 void
3648 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3649 {
3650 	char jpath[MAXPATHLEN];
3651 	struct prison *pr;
3652 	size_t len;
3653 
3654 	pr = cred->cr_prison;
3655 	if (pr->pr_enforce_statfs == 0)
3656 		return;
3657 	if (prison_canseemount(cred, mp) != 0) {
3658 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3659 		strlcpy(sp->f_mntonname, "[restricted]",
3660 		    sizeof(sp->f_mntonname));
3661 		return;
3662 	}
3663 	if (pr->pr_root->v_mount == mp) {
3664 		/*
3665 		 * Clear current buffer data, so we are sure nothing from
3666 		 * the valid path left there.
3667 		 */
3668 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3669 		*sp->f_mntonname = '/';
3670 		return;
3671 	}
3672 	/*
3673 	 * If jail's chroot directory is set to "/" we should be able to see
3674 	 * all mount-points from inside a jail.
3675 	 */
3676 	if (strcmp(pr->pr_path, "/") == 0)
3677 		return;
3678 	len = strlen(pr->pr_path);
3679 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3680 	/*
3681 	 * Clear current buffer data, so we are sure nothing from
3682 	 * the valid path left there.
3683 	 */
3684 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3685 	if (*jpath == '\0') {
3686 		/* Should never happen. */
3687 		*sp->f_mntonname = '/';
3688 	} else {
3689 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3690 	}
3691 }
3692 
3693 /*
3694  * Check with permission for a specific privilege is granted within jail.  We
3695  * have a specific list of accepted privileges; the rest are denied.
3696  */
3697 int
3698 prison_priv_check(struct ucred *cred, int priv)
3699 {
3700 
3701 	if (!jailed(cred))
3702 		return (0);
3703 
3704 #ifdef VIMAGE
3705 	/*
3706 	 * Privileges specific to prisons with a virtual network stack.
3707 	 * There might be a duplicate entry here in case the privilege
3708 	 * is only granted conditionally in the legacy jail case.
3709 	 */
3710 	switch (priv) {
3711 #ifdef notyet
3712 		/*
3713 		 * NFS-specific privileges.
3714 		 */
3715 	case PRIV_NFS_DAEMON:
3716 	case PRIV_NFS_LOCKD:
3717 #endif
3718 		/*
3719 		 * Network stack privileges.
3720 		 */
3721 	case PRIV_NET_BRIDGE:
3722 	case PRIV_NET_GRE:
3723 	case PRIV_NET_BPF:
3724 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3725 	case PRIV_NET_ROUTE:
3726 	case PRIV_NET_TAP:
3727 	case PRIV_NET_SETIFMTU:
3728 	case PRIV_NET_SETIFFLAGS:
3729 	case PRIV_NET_SETIFCAP:
3730 	case PRIV_NET_SETIFDESCR:
3731 	case PRIV_NET_SETIFNAME	:
3732 	case PRIV_NET_SETIFMETRIC:
3733 	case PRIV_NET_SETIFPHYS:
3734 	case PRIV_NET_SETIFMAC:
3735 	case PRIV_NET_ADDMULTI:
3736 	case PRIV_NET_DELMULTI:
3737 	case PRIV_NET_HWIOCTL:
3738 	case PRIV_NET_SETLLADDR:
3739 	case PRIV_NET_ADDIFGROUP:
3740 	case PRIV_NET_DELIFGROUP:
3741 	case PRIV_NET_IFCREATE:
3742 	case PRIV_NET_IFDESTROY:
3743 	case PRIV_NET_ADDIFADDR:
3744 	case PRIV_NET_DELIFADDR:
3745 	case PRIV_NET_LAGG:
3746 	case PRIV_NET_GIF:
3747 	case PRIV_NET_SETIFVNET:
3748 	case PRIV_NET_SETIFFIB:
3749 
3750 		/*
3751 		 * 802.11-related privileges.
3752 		 */
3753 	case PRIV_NET80211_GETKEY:
3754 #ifdef notyet
3755 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3756 #endif
3757 
3758 #ifdef notyet
3759 		/*
3760 		 * ATM privileges.
3761 		 */
3762 	case PRIV_NETATM_CFG:
3763 	case PRIV_NETATM_ADD:
3764 	case PRIV_NETATM_DEL:
3765 	case PRIV_NETATM_SET:
3766 
3767 		/*
3768 		 * Bluetooth privileges.
3769 		 */
3770 	case PRIV_NETBLUETOOTH_RAW:
3771 #endif
3772 
3773 		/*
3774 		 * Netgraph and netgraph module privileges.
3775 		 */
3776 	case PRIV_NETGRAPH_CONTROL:
3777 #ifdef notyet
3778 	case PRIV_NETGRAPH_TTY:
3779 #endif
3780 
3781 		/*
3782 		 * IPv4 and IPv6 privileges.
3783 		 */
3784 	case PRIV_NETINET_IPFW:
3785 	case PRIV_NETINET_DIVERT:
3786 	case PRIV_NETINET_PF:
3787 	case PRIV_NETINET_DUMMYNET:
3788 	case PRIV_NETINET_CARP:
3789 	case PRIV_NETINET_MROUTE:
3790 	case PRIV_NETINET_RAW:
3791 	case PRIV_NETINET_ADDRCTRL6:
3792 	case PRIV_NETINET_ND6:
3793 	case PRIV_NETINET_SCOPE6:
3794 	case PRIV_NETINET_ALIFETIME6:
3795 	case PRIV_NETINET_IPSEC:
3796 	case PRIV_NETINET_BINDANY:
3797 
3798 #ifdef notyet
3799 		/*
3800 		 * NCP privileges.
3801 		 */
3802 	case PRIV_NETNCP:
3803 
3804 		/*
3805 		 * SMB privileges.
3806 		 */
3807 	case PRIV_NETSMB:
3808 #endif
3809 
3810 	/*
3811 	 * No default: or deny here.
3812 	 * In case of no permit fall through to next switch().
3813 	 */
3814 		if (cred->cr_prison->pr_flags & PR_VNET)
3815 			return (0);
3816 	}
3817 #endif /* VIMAGE */
3818 
3819 	switch (priv) {
3820 
3821 		/*
3822 		 * Allow ktrace privileges for root in jail.
3823 		 */
3824 	case PRIV_KTRACE:
3825 
3826 #if 0
3827 		/*
3828 		 * Allow jailed processes to configure audit identity and
3829 		 * submit audit records (login, etc).  In the future we may
3830 		 * want to further refine the relationship between audit and
3831 		 * jail.
3832 		 */
3833 	case PRIV_AUDIT_GETAUDIT:
3834 	case PRIV_AUDIT_SETAUDIT:
3835 	case PRIV_AUDIT_SUBMIT:
3836 #endif
3837 
3838 		/*
3839 		 * Allow jailed processes to manipulate process UNIX
3840 		 * credentials in any way they see fit.
3841 		 */
3842 	case PRIV_CRED_SETUID:
3843 	case PRIV_CRED_SETEUID:
3844 	case PRIV_CRED_SETGID:
3845 	case PRIV_CRED_SETEGID:
3846 	case PRIV_CRED_SETGROUPS:
3847 	case PRIV_CRED_SETREUID:
3848 	case PRIV_CRED_SETREGID:
3849 	case PRIV_CRED_SETRESUID:
3850 	case PRIV_CRED_SETRESGID:
3851 
3852 		/*
3853 		 * Jail implements visibility constraints already, so allow
3854 		 * jailed root to override uid/gid-based constraints.
3855 		 */
3856 	case PRIV_SEEOTHERGIDS:
3857 	case PRIV_SEEOTHERUIDS:
3858 
3859 		/*
3860 		 * Jail implements inter-process debugging limits already, so
3861 		 * allow jailed root various debugging privileges.
3862 		 */
3863 	case PRIV_DEBUG_DIFFCRED:
3864 	case PRIV_DEBUG_SUGID:
3865 	case PRIV_DEBUG_UNPRIV:
3866 
3867 		/*
3868 		 * Allow jail to set various resource limits and login
3869 		 * properties, and for now, exceed process resource limits.
3870 		 */
3871 	case PRIV_PROC_LIMIT:
3872 	case PRIV_PROC_SETLOGIN:
3873 	case PRIV_PROC_SETRLIMIT:
3874 
3875 		/*
3876 		 * System V and POSIX IPC privileges are granted in jail.
3877 		 */
3878 	case PRIV_IPC_READ:
3879 	case PRIV_IPC_WRITE:
3880 	case PRIV_IPC_ADMIN:
3881 	case PRIV_IPC_MSGSIZE:
3882 	case PRIV_MQ_ADMIN:
3883 
3884 		/*
3885 		 * Jail operations within a jail work on child jails.
3886 		 */
3887 	case PRIV_JAIL_ATTACH:
3888 	case PRIV_JAIL_SET:
3889 	case PRIV_JAIL_REMOVE:
3890 
3891 		/*
3892 		 * Jail implements its own inter-process limits, so allow
3893 		 * root processes in jail to change scheduling on other
3894 		 * processes in the same jail.  Likewise for signalling.
3895 		 */
3896 	case PRIV_SCHED_DIFFCRED:
3897 	case PRIV_SCHED_CPUSET:
3898 	case PRIV_SIGNAL_DIFFCRED:
3899 	case PRIV_SIGNAL_SUGID:
3900 
3901 		/*
3902 		 * Allow jailed processes to write to sysctls marked as jail
3903 		 * writable.
3904 		 */
3905 	case PRIV_SYSCTL_WRITEJAIL:
3906 
3907 		/*
3908 		 * Allow root in jail to manage a variety of quota
3909 		 * properties.  These should likely be conditional on a
3910 		 * configuration option.
3911 		 */
3912 	case PRIV_VFS_GETQUOTA:
3913 	case PRIV_VFS_SETQUOTA:
3914 
3915 		/*
3916 		 * Since Jail relies on chroot() to implement file system
3917 		 * protections, grant many VFS privileges to root in jail.
3918 		 * Be careful to exclude mount-related and NFS-related
3919 		 * privileges.
3920 		 */
3921 	case PRIV_VFS_READ:
3922 	case PRIV_VFS_WRITE:
3923 	case PRIV_VFS_ADMIN:
3924 	case PRIV_VFS_EXEC:
3925 	case PRIV_VFS_LOOKUP:
3926 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3927 	case PRIV_VFS_CHFLAGS_DEV:
3928 	case PRIV_VFS_CHOWN:
3929 	case PRIV_VFS_CHROOT:
3930 	case PRIV_VFS_RETAINSUGID:
3931 	case PRIV_VFS_FCHROOT:
3932 	case PRIV_VFS_LINK:
3933 	case PRIV_VFS_SETGID:
3934 	case PRIV_VFS_STAT:
3935 	case PRIV_VFS_STICKYFILE:
3936 
3937 		/*
3938 		 * As in the non-jail case, non-root users are expected to be
3939 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3940 		 * exists in the jail and they have permission to access it).
3941 		 */
3942 	case PRIV_KMEM_READ:
3943 		return (0);
3944 
3945 		/*
3946 		 * Depending on the global setting, allow privilege of
3947 		 * setting system flags.
3948 		 */
3949 	case PRIV_VFS_SYSFLAGS:
3950 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3951 			return (0);
3952 		else
3953 			return (EPERM);
3954 
3955 		/*
3956 		 * Depending on the global setting, allow privilege of
3957 		 * mounting/unmounting file systems.
3958 		 */
3959 	case PRIV_VFS_MOUNT:
3960 	case PRIV_VFS_UNMOUNT:
3961 	case PRIV_VFS_MOUNT_NONUSER:
3962 	case PRIV_VFS_MOUNT_OWNER:
3963 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3964 		    cred->cr_prison->pr_enforce_statfs < 2)
3965 			return (0);
3966 		else
3967 			return (EPERM);
3968 
3969 		/*
3970 		 * Allow jailed root to bind reserved ports and reuse in-use
3971 		 * ports.
3972 		 */
3973 	case PRIV_NETINET_RESERVEDPORT:
3974 	case PRIV_NETINET_REUSEPORT:
3975 		return (0);
3976 
3977 		/*
3978 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3979 		 */
3980 	case PRIV_NETINET_SETHDROPTS:
3981 		return (0);
3982 
3983 		/*
3984 		 * Conditionally allow creating raw sockets in jail.
3985 		 */
3986 	case PRIV_NETINET_RAW:
3987 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3988 			return (0);
3989 		else
3990 			return (EPERM);
3991 
3992 		/*
3993 		 * Since jail implements its own visibility limits on netstat
3994 		 * sysctls, allow getcred.  This allows identd to work in
3995 		 * jail.
3996 		 */
3997 	case PRIV_NETINET_GETCRED:
3998 		return (0);
3999 
4000 		/*
4001 		 * Allow jailed root to set loginclass.
4002 		 */
4003 	case PRIV_PROC_SETLOGINCLASS:
4004 		return (0);
4005 
4006 	default:
4007 		/*
4008 		 * In all remaining cases, deny the privilege request.  This
4009 		 * includes almost all network privileges, many system
4010 		 * configuration privileges.
4011 		 */
4012 		return (EPERM);
4013 	}
4014 }
4015 
4016 /*
4017  * Return the part of pr2's name that is relative to pr1, or the whole name
4018  * if it does not directly follow.
4019  */
4020 
4021 char *
4022 prison_name(struct prison *pr1, struct prison *pr2)
4023 {
4024 	char *name;
4025 
4026 	/* Jails see themselves as "0" (if they see themselves at all). */
4027 	if (pr1 == pr2)
4028 		return "0";
4029 	name = pr2->pr_name;
4030 	if (prison_ischild(pr1, pr2)) {
4031 		/*
4032 		 * pr1 isn't locked (and allprison_lock may not be either)
4033 		 * so its length can't be counted on.  But the number of dots
4034 		 * can be counted on - and counted.
4035 		 */
4036 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4037 			name = strchr(name, '.') + 1;
4038 	}
4039 	return (name);
4040 }
4041 
4042 /*
4043  * Return the part of pr2's path that is relative to pr1, or the whole path
4044  * if it does not directly follow.
4045  */
4046 static char *
4047 prison_path(struct prison *pr1, struct prison *pr2)
4048 {
4049 	char *path1, *path2;
4050 	int len1;
4051 
4052 	path1 = pr1->pr_path;
4053 	path2 = pr2->pr_path;
4054 	if (!strcmp(path1, "/"))
4055 		return (path2);
4056 	len1 = strlen(path1);
4057 	if (strncmp(path1, path2, len1))
4058 		return (path2);
4059 	if (path2[len1] == '\0')
4060 		return "/";
4061 	if (path2[len1] == '/')
4062 		return (path2 + len1);
4063 	return (path2);
4064 }
4065 
4066 
4067 /*
4068  * Jail-related sysctls.
4069  */
4070 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4071     "Jails");
4072 
4073 static int
4074 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4075 {
4076 	struct xprison *xp;
4077 	struct prison *pr, *cpr;
4078 #ifdef INET
4079 	struct in_addr *ip4 = NULL;
4080 	int ip4s = 0;
4081 #endif
4082 #ifdef INET6
4083 	struct in6_addr *ip6 = NULL;
4084 	int ip6s = 0;
4085 #endif
4086 	int descend, error;
4087 
4088 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4089 	pr = req->td->td_ucred->cr_prison;
4090 	error = 0;
4091 	sx_slock(&allprison_lock);
4092 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4093 #if defined(INET) || defined(INET6)
4094  again:
4095 #endif
4096 		mtx_lock(&cpr->pr_mtx);
4097 #ifdef INET
4098 		if (cpr->pr_ip4s > 0) {
4099 			if (ip4s < cpr->pr_ip4s) {
4100 				ip4s = cpr->pr_ip4s;
4101 				mtx_unlock(&cpr->pr_mtx);
4102 				ip4 = realloc(ip4, ip4s *
4103 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4104 				goto again;
4105 			}
4106 			bcopy(cpr->pr_ip4, ip4,
4107 			    cpr->pr_ip4s * sizeof(struct in_addr));
4108 		}
4109 #endif
4110 #ifdef INET6
4111 		if (cpr->pr_ip6s > 0) {
4112 			if (ip6s < cpr->pr_ip6s) {
4113 				ip6s = cpr->pr_ip6s;
4114 				mtx_unlock(&cpr->pr_mtx);
4115 				ip6 = realloc(ip6, ip6s *
4116 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4117 				goto again;
4118 			}
4119 			bcopy(cpr->pr_ip6, ip6,
4120 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4121 		}
4122 #endif
4123 		if (cpr->pr_ref == 0) {
4124 			mtx_unlock(&cpr->pr_mtx);
4125 			continue;
4126 		}
4127 		bzero(xp, sizeof(*xp));
4128 		xp->pr_version = XPRISON_VERSION;
4129 		xp->pr_id = cpr->pr_id;
4130 		xp->pr_state = cpr->pr_uref > 0
4131 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4132 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4133 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4134 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4135 #ifdef INET
4136 		xp->pr_ip4s = cpr->pr_ip4s;
4137 #endif
4138 #ifdef INET6
4139 		xp->pr_ip6s = cpr->pr_ip6s;
4140 #endif
4141 		mtx_unlock(&cpr->pr_mtx);
4142 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4143 		if (error)
4144 			break;
4145 #ifdef INET
4146 		if (xp->pr_ip4s > 0) {
4147 			error = SYSCTL_OUT(req, ip4,
4148 			    xp->pr_ip4s * sizeof(struct in_addr));
4149 			if (error)
4150 				break;
4151 		}
4152 #endif
4153 #ifdef INET6
4154 		if (xp->pr_ip6s > 0) {
4155 			error = SYSCTL_OUT(req, ip6,
4156 			    xp->pr_ip6s * sizeof(struct in6_addr));
4157 			if (error)
4158 				break;
4159 		}
4160 #endif
4161 	}
4162 	sx_sunlock(&allprison_lock);
4163 	free(xp, M_TEMP);
4164 #ifdef INET
4165 	free(ip4, M_TEMP);
4166 #endif
4167 #ifdef INET6
4168 	free(ip6, M_TEMP);
4169 #endif
4170 	return (error);
4171 }
4172 
4173 SYSCTL_OID(_security_jail, OID_AUTO, list,
4174     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4175     sysctl_jail_list, "S", "List of active jails");
4176 
4177 static int
4178 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4179 {
4180 	int error, injail;
4181 
4182 	injail = jailed(req->td->td_ucred);
4183 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4184 
4185 	return (error);
4186 }
4187 
4188 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4189     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4190     sysctl_jail_jailed, "I", "Process in jail?");
4191 
4192 static int
4193 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4194 {
4195 	int error, havevnet;
4196 #ifdef VIMAGE
4197 	struct ucred *cred = req->td->td_ucred;
4198 
4199 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4200 #else
4201 	havevnet = 0;
4202 #endif
4203 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4204 
4205 	return (error);
4206 }
4207 
4208 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4209     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4210     sysctl_jail_vnet, "I", "Jail owns VNET?");
4211 
4212 #if defined(INET) || defined(INET6)
4213 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4214     &jail_max_af_ips, 0,
4215     "Number of IP addresses a jail may have at most per address family");
4216 #endif
4217 
4218 /*
4219  * Default parameters for jail(2) compatability.  For historical reasons,
4220  * the sysctl names have varying similarity to the parameter names.  Prisons
4221  * just see their own parameters, and can't change them.
4222  */
4223 static int
4224 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4225 {
4226 	struct prison *pr;
4227 	int allow, error, i;
4228 
4229 	pr = req->td->td_ucred->cr_prison;
4230 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4231 
4232 	/* Get the current flag value, and convert it to a boolean. */
4233 	i = (allow & arg2) ? 1 : 0;
4234 	if (arg1 != NULL)
4235 		i = !i;
4236 	error = sysctl_handle_int(oidp, &i, 0, req);
4237 	if (error || !req->newptr)
4238 		return (error);
4239 	i = i ? arg2 : 0;
4240 	if (arg1 != NULL)
4241 		i ^= arg2;
4242 	/*
4243 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4244 	 * for writing.
4245 	 */
4246 	mtx_lock(&prison0.pr_mtx);
4247 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4248 	mtx_unlock(&prison0.pr_mtx);
4249 	return (0);
4250 }
4251 
4252 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4253     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4254     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4255     "Processes in jail can set their hostnames");
4256 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4257     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4258     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4259     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4260 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4261     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4262     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4263     "Processes in jail can use System V IPC primitives");
4264 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4265     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4266     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4267     "Prison root can create raw sockets");
4268 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4269     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4270     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4271     "Processes in jail can alter system file flags");
4272 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4273     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4274     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4275     "Processes in jail can mount/unmount jail-friendly file systems");
4276 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4277     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4278     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4279     "Processes in jail can mount the devfs file system");
4280 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4281     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4282     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4283     "Processes in jail can mount the fdescfs file system");
4284 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4285     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4286     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4287     "Processes in jail can mount the nullfs file system");
4288 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4289     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4290     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4291     "Processes in jail can mount the procfs file system");
4292 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4293     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4294     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4295     "Processes in jail can mount the tmpfs file system");
4296 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4297     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4298     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4299     "Processes in jail can mount the zfs file system");
4300 
4301 static int
4302 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4303 {
4304 	struct prison *pr;
4305 	int level, error;
4306 
4307 	pr = req->td->td_ucred->cr_prison;
4308 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4309 	error = sysctl_handle_int(oidp, &level, 0, req);
4310 	if (error || !req->newptr)
4311 		return (error);
4312 	*(int *)arg1 = level;
4313 	return (0);
4314 }
4315 
4316 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4317     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4318     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4319     sysctl_jail_default_level, "I",
4320     "Processes in jail cannot see all mounted file systems");
4321 
4322 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4323     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4324     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4325     sysctl_jail_default_level, "I",
4326     "Ruleset for the devfs filesystem in jail");
4327 
4328 /*
4329  * Nodes to describe jail parameters.  Maximum length of string parameters
4330  * is returned in the string itself, and the other parameters exist merely
4331  * to make themselves and their types known.
4332  */
4333 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4334     "Jail parameters");
4335 
4336 int
4337 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4338 {
4339 	int i;
4340 	long l;
4341 	size_t s;
4342 	char numbuf[12];
4343 
4344 	switch (oidp->oid_kind & CTLTYPE)
4345 	{
4346 	case CTLTYPE_LONG:
4347 	case CTLTYPE_ULONG:
4348 		l = 0;
4349 #ifdef SCTL_MASK32
4350 		if (!(req->flags & SCTL_MASK32))
4351 #endif
4352 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4353 	case CTLTYPE_INT:
4354 	case CTLTYPE_UINT:
4355 		i = 0;
4356 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4357 	case CTLTYPE_STRING:
4358 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4359 		return
4360 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4361 	case CTLTYPE_STRUCT:
4362 		s = (size_t)arg2;
4363 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4364 	}
4365 	return (0);
4366 }
4367 
4368 /*
4369  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4370  * jail creation time but cannot be changed in an existing jail.
4371  */
4372 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4373 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4374 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4375 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4376 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4377     "I", "Jail secure level");
4378 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4379     "Jail value for kern.osreldate and uname -K");
4380 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4381     "Jail value for kern.osrelease and uname -r");
4382 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4383     "I", "Jail cannot see all mounted file systems");
4384 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4385     "I", "Ruleset for in-jail devfs mounts");
4386 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4387     "B", "Jail persistence");
4388 #ifdef VIMAGE
4389 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4390     "E,jailsys", "Virtual network stack");
4391 #endif
4392 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4393     "B", "Jail is in the process of shutting down");
4394 
4395 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4396 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4397     "I", "Current number of child jails");
4398 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4399     "I", "Maximum number of child jails");
4400 
4401 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4402 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4403     "Jail hostname");
4404 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4405     "Jail NIS domainname");
4406 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4407     "Jail host UUID");
4408 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4409     "LU", "Jail host ID");
4410 
4411 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4412 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4413 
4414 #ifdef INET
4415 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4416     "Jail IPv4 address virtualization");
4417 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4418     "S,in_addr,a", "Jail IPv4 addresses");
4419 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4420     "B", "Do (not) use IPv4 source address selection rather than the "
4421     "primary jail IPv4 address.");
4422 #endif
4423 #ifdef INET6
4424 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4425     "Jail IPv6 address virtualization");
4426 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4427     "S,in6_addr,a", "Jail IPv6 addresses");
4428 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4429     "B", "Do (not) use IPv6 source address selection rather than the "
4430     "primary jail IPv6 address.");
4431 #endif
4432 
4433 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4434 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4435     "B", "Jail may set hostname");
4436 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4437     "B", "Jail may use SYSV IPC");
4438 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4439     "B", "Jail may create raw sockets");
4440 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4441     "B", "Jail may alter system file flags");
4442 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4443     "B", "Jail may set file quotas");
4444 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4445     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4446 
4447 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4448 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4449     "B", "Jail may mount/unmount jail-friendly file systems in general");
4450 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4451     "B", "Jail may mount the devfs file system");
4452 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4453     "B", "Jail may mount the fdescfs file system");
4454 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4455     "B", "Jail may mount the nullfs file system");
4456 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4457     "B", "Jail may mount the procfs file system");
4458 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4459     "B", "Jail may mount the tmpfs file system");
4460 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4461     "B", "Jail may mount the zfs file system");
4462 
4463 void
4464 prison_racct_foreach(void (*callback)(struct racct *racct,
4465     void *arg2, void *arg3), void *arg2, void *arg3)
4466 {
4467 	struct prison_racct *prr;
4468 
4469 	sx_slock(&allprison_lock);
4470 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4471 		(callback)(prr->prr_racct, arg2, arg3);
4472 	sx_sunlock(&allprison_lock);
4473 }
4474 
4475 static struct prison_racct *
4476 prison_racct_find_locked(const char *name)
4477 {
4478 	struct prison_racct *prr;
4479 
4480 	sx_assert(&allprison_lock, SA_XLOCKED);
4481 
4482 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4483 		return (NULL);
4484 
4485 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4486 		if (strcmp(name, prr->prr_name) != 0)
4487 			continue;
4488 
4489 		/* Found prison_racct with a matching name? */
4490 		prison_racct_hold(prr);
4491 		return (prr);
4492 	}
4493 
4494 	/* Add new prison_racct. */
4495 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4496 	racct_create(&prr->prr_racct);
4497 
4498 	strcpy(prr->prr_name, name);
4499 	refcount_init(&prr->prr_refcount, 1);
4500 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4501 
4502 	return (prr);
4503 }
4504 
4505 struct prison_racct *
4506 prison_racct_find(const char *name)
4507 {
4508 	struct prison_racct *prr;
4509 
4510 	sx_xlock(&allprison_lock);
4511 	prr = prison_racct_find_locked(name);
4512 	sx_xunlock(&allprison_lock);
4513 	return (prr);
4514 }
4515 
4516 void
4517 prison_racct_hold(struct prison_racct *prr)
4518 {
4519 
4520 	refcount_acquire(&prr->prr_refcount);
4521 }
4522 
4523 static void
4524 prison_racct_free_locked(struct prison_racct *prr)
4525 {
4526 
4527 	sx_assert(&allprison_lock, SA_XLOCKED);
4528 
4529 	if (refcount_release(&prr->prr_refcount)) {
4530 		racct_destroy(&prr->prr_racct);
4531 		LIST_REMOVE(prr, prr_next);
4532 		free(prr, M_PRISON_RACCT);
4533 	}
4534 }
4535 
4536 void
4537 prison_racct_free(struct prison_racct *prr)
4538 {
4539 	int old;
4540 
4541 	sx_assert(&allprison_lock, SA_UNLOCKED);
4542 
4543 	old = prr->prr_refcount;
4544 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4545 		return;
4546 
4547 	sx_xlock(&allprison_lock);
4548 	prison_racct_free_locked(prr);
4549 	sx_xunlock(&allprison_lock);
4550 }
4551 
4552 #ifdef RACCT
4553 static void
4554 prison_racct_attach(struct prison *pr)
4555 {
4556 	struct prison_racct *prr;
4557 
4558 	sx_assert(&allprison_lock, SA_XLOCKED);
4559 
4560 	prr = prison_racct_find_locked(pr->pr_name);
4561 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4562 
4563 	pr->pr_prison_racct = prr;
4564 }
4565 
4566 /*
4567  * Handle jail renaming.  From the racct point of view, renaming means
4568  * moving from one prison_racct to another.
4569  */
4570 static void
4571 prison_racct_modify(struct prison *pr)
4572 {
4573 	struct proc *p;
4574 	struct ucred *cred;
4575 	struct prison_racct *oldprr;
4576 
4577 	sx_slock(&allproc_lock);
4578 	sx_xlock(&allprison_lock);
4579 
4580 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4581 		sx_xunlock(&allprison_lock);
4582 		sx_sunlock(&allproc_lock);
4583 		return;
4584 	}
4585 
4586 	oldprr = pr->pr_prison_racct;
4587 	pr->pr_prison_racct = NULL;
4588 
4589 	prison_racct_attach(pr);
4590 
4591 	/*
4592 	 * Move resource utilisation records.
4593 	 */
4594 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4595 
4596 	/*
4597 	 * Force rctl to reattach rules to processes.
4598 	 */
4599 	FOREACH_PROC_IN_SYSTEM(p) {
4600 		PROC_LOCK(p);
4601 		cred = crhold(p->p_ucred);
4602 		PROC_UNLOCK(p);
4603 		racct_proc_ucred_changed(p, cred, cred);
4604 		crfree(cred);
4605 	}
4606 
4607 	sx_sunlock(&allproc_lock);
4608 	prison_racct_free_locked(oldprr);
4609 	sx_xunlock(&allprison_lock);
4610 }
4611 
4612 static void
4613 prison_racct_detach(struct prison *pr)
4614 {
4615 
4616 	sx_assert(&allprison_lock, SA_UNLOCKED);
4617 
4618 	if (pr->pr_prison_racct == NULL)
4619 		return;
4620 	prison_racct_free(pr->pr_prison_racct);
4621 	pr->pr_prison_racct = NULL;
4622 }
4623 #endif /* RACCT */
4624 
4625 #ifdef DDB
4626 
4627 static void
4628 db_show_prison(struct prison *pr)
4629 {
4630 	int fi;
4631 #if defined(INET) || defined(INET6)
4632 	int ii;
4633 #endif
4634 	unsigned jsf;
4635 #ifdef INET6
4636 	char ip6buf[INET6_ADDRSTRLEN];
4637 #endif
4638 
4639 	db_printf("prison %p:\n", pr);
4640 	db_printf(" jid             = %d\n", pr->pr_id);
4641 	db_printf(" name            = %s\n", pr->pr_name);
4642 	db_printf(" parent          = %p\n", pr->pr_parent);
4643 	db_printf(" ref             = %d\n", pr->pr_ref);
4644 	db_printf(" uref            = %d\n", pr->pr_uref);
4645 	db_printf(" path            = %s\n", pr->pr_path);
4646 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4647 	    ? pr->pr_cpuset->cs_id : -1);
4648 #ifdef VIMAGE
4649 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4650 #endif
4651 	db_printf(" root            = %p\n", pr->pr_root);
4652 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4653 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4654 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4655 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4656 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4657 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4658 	db_printf(" flags           = 0x%x", pr->pr_flags);
4659 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4660 	    fi++)
4661 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4662 			db_printf(" %s", pr_flag_names[fi]);
4663 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4664 	    fi++) {
4665 		jsf = pr->pr_flags &
4666 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4667 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4668 		    pr_flag_jailsys[fi].disable &&
4669 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4670 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4671 		    : "inherit");
4672 	}
4673 	db_printf(" allow           = 0x%x", pr->pr_allow);
4674 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4675 	    fi++)
4676 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4677 			db_printf(" %s", pr_allow_names[fi]);
4678 	db_printf("\n");
4679 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4680 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4681 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4682 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4683 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4684 #ifdef INET
4685 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4686 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4687 		db_printf(" %s %s\n",
4688 		    ii == 0 ? "ip4.addr        =" : "                 ",
4689 		    inet_ntoa(pr->pr_ip4[ii]));
4690 #endif
4691 #ifdef INET6
4692 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4693 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4694 		db_printf(" %s %s\n",
4695 		    ii == 0 ? "ip6.addr        =" : "                 ",
4696 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4697 #endif
4698 }
4699 
4700 DB_SHOW_COMMAND(prison, db_show_prison_command)
4701 {
4702 	struct prison *pr;
4703 
4704 	if (!have_addr) {
4705 		/*
4706 		 * Show all prisons in the list, and prison0 which is not
4707 		 * listed.
4708 		 */
4709 		db_show_prison(&prison0);
4710 		if (!db_pager_quit) {
4711 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4712 				db_show_prison(pr);
4713 				if (db_pager_quit)
4714 					break;
4715 			}
4716 		}
4717 		return;
4718 	}
4719 
4720 	if (addr == 0)
4721 		pr = &prison0;
4722 	else {
4723 		/* Look for a prison with the ID and with references. */
4724 		TAILQ_FOREACH(pr, &allprison, pr_list)
4725 			if (pr->pr_id == addr && pr->pr_ref > 0)
4726 				break;
4727 		if (pr == NULL)
4728 			/* Look again, without requiring a reference. */
4729 			TAILQ_FOREACH(pr, &allprison, pr_list)
4730 				if (pr->pr_id == addr)
4731 					break;
4732 		if (pr == NULL)
4733 			/* Assume address points to a valid prison. */
4734 			pr = (struct prison *)addr;
4735 	}
4736 	db_show_prison(pr);
4737 }
4738 
4739 #endif /* DDB */
4740