xref: /freebsd/sys/kern/kern_jail.c (revision a5ff72cb0e51a7675d4e2b5810a2b6dad5b91960)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif /* DDB */
72 
73 #include <security/mac/mac_framework.h>
74 
75 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
76 
77 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
78 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
79 
80 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
81 #ifdef INET
82 #ifdef INET6
83 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
84 #else
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
86 #endif
87 #else /* !INET */
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	0
92 #endif
93 #endif
94 
95 /* prison0 describes what is "real" about the system. */
96 struct prison prison0 = {
97 	.pr_id		= 0,
98 	.pr_name	= "0",
99 	.pr_ref		= 1,
100 	.pr_uref	= 1,
101 	.pr_path	= "/",
102 	.pr_securelevel	= -1,
103 	.pr_devfs_rsnum = 0,
104 	.pr_childmax	= JAIL_MAX,
105 	.pr_hostuuid	= DEFAULT_HOSTUUID,
106 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
107 #ifdef VIMAGE
108 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
109 #else
110 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
111 #endif
112 	.pr_allow	= PR_ALLOW_ALL,
113 };
114 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
115 
116 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
117 struct	sx allprison_lock;
118 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
119 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
120 LIST_HEAD(, prison_racct) allprison_racct;
121 int	lastprid = 0;
122 
123 static int do_jail_attach(struct thread *td, struct prison *pr);
124 static void prison_complete(void *context, int pending);
125 static void prison_deref(struct prison *pr, int flags);
126 static char *prison_path(struct prison *pr1, struct prison *pr2);
127 static void prison_remove_one(struct prison *pr);
128 #ifdef RACCT
129 static void prison_racct_attach(struct prison *pr);
130 static void prison_racct_modify(struct prison *pr);
131 static void prison_racct_detach(struct prison *pr);
132 #endif
133 #ifdef INET
134 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
135 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
136 #endif
137 #ifdef INET6
138 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
139 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
140 #endif
141 
142 /* Flags for prison_deref */
143 #define	PD_DEREF	0x01
144 #define	PD_DEUREF	0x02
145 #define	PD_LOCKED	0x04
146 #define	PD_LIST_SLOCKED	0x08
147 #define	PD_LIST_XLOCKED	0x10
148 
149 /*
150  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
151  * as we cannot figure out the size of a sparse array, or an array without a
152  * terminating entry.
153  */
154 static char *pr_flag_names[] = {
155 	[0] = "persist",
156 #ifdef INET
157 	[7] = "ip4.saddrsel",
158 #endif
159 #ifdef INET6
160 	[8] = "ip6.saddrsel",
161 #endif
162 };
163 const size_t pr_flag_names_size = sizeof(pr_flag_names);
164 
165 static char *pr_flag_nonames[] = {
166 	[0] = "nopersist",
167 #ifdef INET
168 	[7] = "ip4.nosaddrsel",
169 #endif
170 #ifdef INET6
171 	[8] = "ip6.nosaddrsel",
172 #endif
173 };
174 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
175 
176 struct jailsys_flags {
177 	const char	*name;
178 	unsigned	 disable;
179 	unsigned	 new;
180 } pr_flag_jailsys[] = {
181 	{ "host", 0, PR_HOST },
182 #ifdef VIMAGE
183 	{ "vnet", 0, PR_VNET },
184 #endif
185 #ifdef INET
186 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
187 #endif
188 #ifdef INET6
189 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
190 #endif
191 };
192 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
193 
194 static char *pr_allow_names[] = {
195 	"allow.set_hostname",
196 	"allow.sysvipc",
197 	"allow.raw_sockets",
198 	"allow.chflags",
199 	"allow.mount",
200 	"allow.quotas",
201 	"allow.socket_af",
202 	"allow.mount.devfs",
203 	"allow.mount.nullfs",
204 	"allow.mount.zfs",
205 	"allow.mount.procfs",
206 	"allow.mount.tmpfs",
207 	"allow.mount.fdescfs",
208 	"allow.mount.linprocfs",
209 	"allow.mount.linsysfs",
210 };
211 const size_t pr_allow_names_size = sizeof(pr_allow_names);
212 
213 static char *pr_allow_nonames[] = {
214 	"allow.noset_hostname",
215 	"allow.nosysvipc",
216 	"allow.noraw_sockets",
217 	"allow.nochflags",
218 	"allow.nomount",
219 	"allow.noquotas",
220 	"allow.nosocket_af",
221 	"allow.mount.nodevfs",
222 	"allow.mount.nonullfs",
223 	"allow.mount.nozfs",
224 	"allow.mount.noprocfs",
225 	"allow.mount.notmpfs",
226 	"allow.mount.nofdescfs",
227 	"allow.mount.nolinprocfs",
228 	"allow.mount.nolinsysfs",
229 };
230 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
231 
232 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
233 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
234 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
235 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
236 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
237 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
238 #if defined(INET) || defined(INET6)
239 static unsigned jail_max_af_ips = 255;
240 #endif
241 
242 /*
243  * Initialize the parts of prison0 that can't be static-initialized with
244  * constants.  This is called from proc0_init() after creating thread0 cpuset.
245  */
246 void
247 prison0_init(void)
248 {
249 
250 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
251 	prison0.pr_osreldate = osreldate;
252 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
253 }
254 
255 #ifdef INET
256 static int
257 qcmp_v4(const void *ip1, const void *ip2)
258 {
259 	in_addr_t iaa, iab;
260 
261 	/*
262 	 * We need to compare in HBO here to get the list sorted as expected
263 	 * by the result of the code.  Sorting NBO addresses gives you
264 	 * interesting results.  If you do not understand, do not try.
265 	 */
266 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
267 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
268 
269 	/*
270 	 * Do not simply return the difference of the two numbers, the int is
271 	 * not wide enough.
272 	 */
273 	if (iaa > iab)
274 		return (1);
275 	else if (iaa < iab)
276 		return (-1);
277 	else
278 		return (0);
279 }
280 #endif
281 
282 #ifdef INET6
283 static int
284 qcmp_v6(const void *ip1, const void *ip2)
285 {
286 	const struct in6_addr *ia6a, *ia6b;
287 	int i, rc;
288 
289 	ia6a = (const struct in6_addr *)ip1;
290 	ia6b = (const struct in6_addr *)ip2;
291 
292 	rc = 0;
293 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
294 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
295 			rc = 1;
296 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
297 			rc = -1;
298 	}
299 	return (rc);
300 }
301 #endif
302 
303 /*
304  * struct jail_args {
305  *	struct jail *jail;
306  * };
307  */
308 int
309 sys_jail(struct thread *td, struct jail_args *uap)
310 {
311 	uint32_t version;
312 	int error;
313 	struct jail j;
314 
315 	error = copyin(uap->jail, &version, sizeof(uint32_t));
316 	if (error)
317 		return (error);
318 
319 	switch (version) {
320 	case 0:
321 	{
322 		struct jail_v0 j0;
323 
324 		/* FreeBSD single IPv4 jails. */
325 		bzero(&j, sizeof(struct jail));
326 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
327 		if (error)
328 			return (error);
329 		j.version = j0.version;
330 		j.path = j0.path;
331 		j.hostname = j0.hostname;
332 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
333 		break;
334 	}
335 
336 	case 1:
337 		/*
338 		 * Version 1 was used by multi-IPv4 jail implementations
339 		 * that never made it into the official kernel.
340 		 */
341 		return (EINVAL);
342 
343 	case 2:	/* JAIL_API_VERSION */
344 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
345 		error = copyin(uap->jail, &j, sizeof(struct jail));
346 		if (error)
347 			return (error);
348 		break;
349 
350 	default:
351 		/* Sci-Fi jails are not supported, sorry. */
352 		return (EINVAL);
353 	}
354 	return (kern_jail(td, &j));
355 }
356 
357 int
358 kern_jail(struct thread *td, struct jail *j)
359 {
360 	struct iovec optiov[2 * (4 + nitems(pr_allow_names)
361 #ifdef INET
362 			    + 1
363 #endif
364 #ifdef INET6
365 			    + 1
366 #endif
367 			    )];
368 	struct uio opt;
369 	char *u_path, *u_hostname, *u_name;
370 #ifdef INET
371 	uint32_t ip4s;
372 	struct in_addr *u_ip4;
373 #endif
374 #ifdef INET6
375 	struct in6_addr *u_ip6;
376 #endif
377 	size_t tmplen;
378 	int error, enforce_statfs, fi;
379 
380 	bzero(&optiov, sizeof(optiov));
381 	opt.uio_iov = optiov;
382 	opt.uio_iovcnt = 0;
383 	opt.uio_offset = -1;
384 	opt.uio_resid = -1;
385 	opt.uio_segflg = UIO_SYSSPACE;
386 	opt.uio_rw = UIO_READ;
387 	opt.uio_td = td;
388 
389 	/* Set permissions for top-level jails from sysctls. */
390 	if (!jailed(td->td_ucred)) {
391 		for (fi = 0; fi < nitems(pr_allow_names); fi++) {
392 			optiov[opt.uio_iovcnt].iov_base =
393 			    (jail_default_allow & (1 << fi))
394 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
395 			optiov[opt.uio_iovcnt].iov_len =
396 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
397 			opt.uio_iovcnt += 2;
398 		}
399 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
400 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
401 		opt.uio_iovcnt++;
402 		enforce_statfs = jail_default_enforce_statfs;
403 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
404 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
405 		opt.uio_iovcnt++;
406 	}
407 
408 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
409 #ifdef INET
410 	ip4s = (j->version == 0) ? 1 : j->ip4s;
411 	if (ip4s > jail_max_af_ips)
412 		return (EINVAL);
413 	tmplen += ip4s * sizeof(struct in_addr);
414 #else
415 	if (j->ip4s > 0)
416 		return (EINVAL);
417 #endif
418 #ifdef INET6
419 	if (j->ip6s > jail_max_af_ips)
420 		return (EINVAL);
421 	tmplen += j->ip6s * sizeof(struct in6_addr);
422 #else
423 	if (j->ip6s > 0)
424 		return (EINVAL);
425 #endif
426 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
427 	u_hostname = u_path + MAXPATHLEN;
428 	u_name = u_hostname + MAXHOSTNAMELEN;
429 #ifdef INET
430 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
431 #endif
432 #ifdef INET6
433 #ifdef INET
434 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
435 #else
436 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
437 #endif
438 #endif
439 	optiov[opt.uio_iovcnt].iov_base = "path";
440 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
441 	opt.uio_iovcnt++;
442 	optiov[opt.uio_iovcnt].iov_base = u_path;
443 	error = copyinstr(j->path, u_path, MAXPATHLEN,
444 	    &optiov[opt.uio_iovcnt].iov_len);
445 	if (error) {
446 		free(u_path, M_TEMP);
447 		return (error);
448 	}
449 	opt.uio_iovcnt++;
450 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
451 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
452 	opt.uio_iovcnt++;
453 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
454 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
455 	    &optiov[opt.uio_iovcnt].iov_len);
456 	if (error) {
457 		free(u_path, M_TEMP);
458 		return (error);
459 	}
460 	opt.uio_iovcnt++;
461 	if (j->jailname != NULL) {
462 		optiov[opt.uio_iovcnt].iov_base = "name";
463 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
464 		opt.uio_iovcnt++;
465 		optiov[opt.uio_iovcnt].iov_base = u_name;
466 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
467 		    &optiov[opt.uio_iovcnt].iov_len);
468 		if (error) {
469 			free(u_path, M_TEMP);
470 			return (error);
471 		}
472 		opt.uio_iovcnt++;
473 	}
474 #ifdef INET
475 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
476 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
477 	opt.uio_iovcnt++;
478 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
479 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
480 	if (j->version == 0)
481 		u_ip4->s_addr = j->ip4s;
482 	else {
483 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
484 		if (error) {
485 			free(u_path, M_TEMP);
486 			return (error);
487 		}
488 	}
489 	opt.uio_iovcnt++;
490 #endif
491 #ifdef INET6
492 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
493 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
494 	opt.uio_iovcnt++;
495 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
496 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
497 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
498 	if (error) {
499 		free(u_path, M_TEMP);
500 		return (error);
501 	}
502 	opt.uio_iovcnt++;
503 #endif
504 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
505 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
506 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
507 	free(u_path, M_TEMP);
508 	return (error);
509 }
510 
511 
512 /*
513  * struct jail_set_args {
514  *	struct iovec *iovp;
515  *	unsigned int iovcnt;
516  *	int flags;
517  * };
518  */
519 int
520 sys_jail_set(struct thread *td, struct jail_set_args *uap)
521 {
522 	struct uio *auio;
523 	int error;
524 
525 	/* Check that we have an even number of iovecs. */
526 	if (uap->iovcnt & 1)
527 		return (EINVAL);
528 
529 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
530 	if (error)
531 		return (error);
532 	error = kern_jail_set(td, auio, uap->flags);
533 	free(auio, M_IOV);
534 	return (error);
535 }
536 
537 int
538 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
539 {
540 	struct nameidata nd;
541 #ifdef INET
542 	struct in_addr *ip4;
543 #endif
544 #ifdef INET6
545 	struct in6_addr *ip6;
546 #endif
547 	struct vfsopt *opt;
548 	struct vfsoptlist *opts;
549 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
550 	struct vnode *root;
551 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
552 	char *g_path, *osrelstr;
553 #if defined(INET) || defined(INET6)
554 	struct prison *tppr;
555 	void *op;
556 #endif
557 	unsigned long hid;
558 	size_t namelen, onamelen;
559 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
560 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
561 	int fi, jid, jsys, len, level;
562 	int childmax, osreldt, rsnum, slevel;
563 	int fullpath_disabled;
564 #if defined(INET) || defined(INET6)
565 	int ii, ij;
566 #endif
567 #ifdef INET
568 	int ip4s, redo_ip4;
569 #endif
570 #ifdef INET6
571 	int ip6s, redo_ip6;
572 #endif
573 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
574 	unsigned tallow;
575 	char numbuf[12];
576 
577 	error = priv_check(td, PRIV_JAIL_SET);
578 	if (!error && (flags & JAIL_ATTACH))
579 		error = priv_check(td, PRIV_JAIL_ATTACH);
580 	if (error)
581 		return (error);
582 	mypr = ppr = td->td_ucred->cr_prison;
583 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
584 		return (EPERM);
585 	if (flags & ~JAIL_SET_MASK)
586 		return (EINVAL);
587 
588 	/*
589 	 * Check all the parameters before committing to anything.  Not all
590 	 * errors can be caught early, but we may as well try.  Also, this
591 	 * takes care of some expensive stuff (path lookup) before getting
592 	 * the allprison lock.
593 	 *
594 	 * XXX Jails are not filesystems, and jail parameters are not mount
595 	 *     options.  But it makes more sense to re-use the vfsopt code
596 	 *     than duplicate it under a different name.
597 	 */
598 	error = vfs_buildopts(optuio, &opts);
599 	if (error)
600 		return (error);
601 #ifdef INET
602 	ip4 = NULL;
603 #endif
604 #ifdef INET6
605 	ip6 = NULL;
606 #endif
607 	g_path = NULL;
608 
609 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
610 	if (error == ENOENT)
611 		jid = 0;
612 	else if (error != 0)
613 		goto done_free;
614 
615 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
616 	if (error == ENOENT)
617 		gotslevel = 0;
618 	else if (error != 0)
619 		goto done_free;
620 	else
621 		gotslevel = 1;
622 
623 	error =
624 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
625 	if (error == ENOENT)
626 		gotchildmax = 0;
627 	else if (error != 0)
628 		goto done_free;
629 	else
630 		gotchildmax = 1;
631 
632 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
633 	if (error == ENOENT)
634 		gotenforce = 0;
635 	else if (error != 0)
636 		goto done_free;
637 	else if (enforce < 0 || enforce > 2) {
638 		error = EINVAL;
639 		goto done_free;
640 	} else
641 		gotenforce = 1;
642 
643 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
644 	if (error == ENOENT)
645 		gotrsnum = 0;
646 	else if (error != 0)
647 		goto done_free;
648 	else
649 		gotrsnum = 1;
650 
651 	pr_flags = ch_flags = 0;
652 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
653 		if (pr_flag_names[fi] == NULL)
654 			continue;
655 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
656 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
657 	}
658 	ch_flags |= pr_flags;
659 	for (fi = 0; fi < nitems(pr_flag_jailsys); 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 < nitems(pr_allow_names); fi++) {
716 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
717 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
718 	}
719 	ch_allow |= pr_allow;
720 
721 	error = vfs_getopt(opts, "name", (void **)&name, &len);
722 	if (error == ENOENT)
723 		name = NULL;
724 	else if (error != 0)
725 		goto done_free;
726 	else {
727 		if (len == 0 || name[len - 1] != '\0') {
728 			error = EINVAL;
729 			goto done_free;
730 		}
731 		if (len > MAXHOSTNAMELEN) {
732 			error = ENAMETOOLONG;
733 			goto done_free;
734 		}
735 	}
736 
737 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
738 	if (error == ENOENT)
739 		host = NULL;
740 	else if (error != 0)
741 		goto done_free;
742 	else {
743 		ch_flags |= PR_HOST;
744 		pr_flags |= PR_HOST;
745 		if (len == 0 || host[len - 1] != '\0') {
746 			error = EINVAL;
747 			goto done_free;
748 		}
749 		if (len > MAXHOSTNAMELEN) {
750 			error = ENAMETOOLONG;
751 			goto done_free;
752 		}
753 	}
754 
755 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
756 	if (error == ENOENT)
757 		domain = NULL;
758 	else if (error != 0)
759 		goto done_free;
760 	else {
761 		ch_flags |= PR_HOST;
762 		pr_flags |= PR_HOST;
763 		if (len == 0 || domain[len - 1] != '\0') {
764 			error = EINVAL;
765 			goto done_free;
766 		}
767 		if (len > MAXHOSTNAMELEN) {
768 			error = ENAMETOOLONG;
769 			goto done_free;
770 		}
771 	}
772 
773 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
774 	if (error == ENOENT)
775 		uuid = NULL;
776 	else if (error != 0)
777 		goto done_free;
778 	else {
779 		ch_flags |= PR_HOST;
780 		pr_flags |= PR_HOST;
781 		if (len == 0 || uuid[len - 1] != '\0') {
782 			error = EINVAL;
783 			goto done_free;
784 		}
785 		if (len > HOSTUUIDLEN) {
786 			error = ENAMETOOLONG;
787 			goto done_free;
788 		}
789 	}
790 
791 #ifdef COMPAT_FREEBSD32
792 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
793 		uint32_t hid32;
794 
795 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
796 		hid = hid32;
797 	} else
798 #endif
799 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
800 	if (error == ENOENT)
801 		gothid = 0;
802 	else if (error != 0)
803 		goto done_free;
804 	else {
805 		gothid = 1;
806 		ch_flags |= PR_HOST;
807 		pr_flags |= PR_HOST;
808 	}
809 
810 #ifdef INET
811 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
812 	if (error == ENOENT)
813 		ip4s = 0;
814 	else if (error != 0)
815 		goto done_free;
816 	else if (ip4s & (sizeof(*ip4) - 1)) {
817 		error = EINVAL;
818 		goto done_free;
819 	} else {
820 		ch_flags |= PR_IP4_USER;
821 		pr_flags |= PR_IP4_USER;
822 		if (ip4s > 0) {
823 			ip4s /= sizeof(*ip4);
824 			if (ip4s > jail_max_af_ips) {
825 				error = EINVAL;
826 				vfs_opterror(opts, "too many IPv4 addresses");
827 				goto done_errmsg;
828 			}
829 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
830 			bcopy(op, ip4, ip4s * sizeof(*ip4));
831 			/*
832 			 * IP addresses are all sorted but ip[0] to preserve
833 			 * the primary IP address as given from userland.
834 			 * This special IP is used for unbound outgoing
835 			 * connections as well for "loopback" traffic in case
836 			 * source address selection cannot find any more fitting
837 			 * address to connect from.
838 			 */
839 			if (ip4s > 1)
840 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
841 			/*
842 			 * Check for duplicate addresses and do some simple
843 			 * zero and broadcast checks. If users give other bogus
844 			 * addresses it is their problem.
845 			 *
846 			 * We do not have to care about byte order for these
847 			 * checks so we will do them in NBO.
848 			 */
849 			for (ii = 0; ii < ip4s; ii++) {
850 				if (ip4[ii].s_addr == INADDR_ANY ||
851 				    ip4[ii].s_addr == INADDR_BROADCAST) {
852 					error = EINVAL;
853 					goto done_free;
854 				}
855 				if ((ii+1) < ip4s &&
856 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
857 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
858 					error = EINVAL;
859 					goto done_free;
860 				}
861 			}
862 		}
863 	}
864 #endif
865 
866 #ifdef INET6
867 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
868 	if (error == ENOENT)
869 		ip6s = 0;
870 	else if (error != 0)
871 		goto done_free;
872 	else if (ip6s & (sizeof(*ip6) - 1)) {
873 		error = EINVAL;
874 		goto done_free;
875 	} else {
876 		ch_flags |= PR_IP6_USER;
877 		pr_flags |= PR_IP6_USER;
878 		if (ip6s > 0) {
879 			ip6s /= sizeof(*ip6);
880 			if (ip6s > jail_max_af_ips) {
881 				error = EINVAL;
882 				vfs_opterror(opts, "too many IPv6 addresses");
883 				goto done_errmsg;
884 			}
885 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
886 			bcopy(op, ip6, ip6s * sizeof(*ip6));
887 			if (ip6s > 1)
888 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
889 			for (ii = 0; ii < ip6s; ii++) {
890 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
891 					error = EINVAL;
892 					goto done_free;
893 				}
894 				if ((ii+1) < ip6s &&
895 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
896 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
897 				{
898 					error = EINVAL;
899 					goto done_free;
900 				}
901 			}
902 		}
903 	}
904 #endif
905 
906 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
907 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
908 		error = EINVAL;
909 		vfs_opterror(opts,
910 		    "vnet jails cannot have IP address restrictions");
911 		goto done_errmsg;
912 	}
913 #endif
914 
915 	fullpath_disabled = 0;
916 	root = NULL;
917 	error = vfs_getopt(opts, "path", (void **)&path, &len);
918 	if (error == ENOENT)
919 		path = NULL;
920 	else if (error != 0)
921 		goto done_free;
922 	else {
923 		if (flags & JAIL_UPDATE) {
924 			error = EINVAL;
925 			vfs_opterror(opts,
926 			    "path cannot be changed after creation");
927 			goto done_errmsg;
928 		}
929 		if (len == 0 || path[len - 1] != '\0') {
930 			error = EINVAL;
931 			goto done_free;
932 		}
933 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
934 		    path, td);
935 		error = namei(&nd);
936 		if (error)
937 			goto done_free;
938 		root = nd.ni_vp;
939 		NDFREE(&nd, NDF_ONLY_PNBUF);
940 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
941 		strlcpy(g_path, path, MAXPATHLEN);
942 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
943 		if (error == 0)
944 			path = g_path;
945 		else if (error == ENODEV) {
946 			/* proceed if sysctl debug.disablefullpath == 1 */
947 			fullpath_disabled = 1;
948 			if (len < 2 || (len == 2 && path[0] == '/'))
949 				path = NULL;
950 		} else {
951 			/* exit on other errors */
952 			goto done_free;
953 		}
954 		if (root->v_type != VDIR) {
955 			error = ENOTDIR;
956 			vput(root);
957 			goto done_free;
958 		}
959 		VOP_UNLOCK(root, 0);
960 		if (fullpath_disabled) {
961 			/* Leave room for a real-root full pathname. */
962 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
963 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
964 				error = ENAMETOOLONG;
965 				goto done_free;
966 			}
967 		}
968 	}
969 
970 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
971 	if (error == ENOENT)
972 		osrelstr = NULL;
973 	else if (error != 0)
974 		goto done_free;
975 	else {
976 		if (flags & JAIL_UPDATE) {
977 			error = EINVAL;
978 			vfs_opterror(opts,
979 			    "osrelease cannot be changed after creation");
980 			goto done_errmsg;
981 		}
982 		if (len == 0 || len >= OSRELEASELEN) {
983 			error = EINVAL;
984 			vfs_opterror(opts,
985 			    "osrelease string must be 1-%d bytes long",
986 			    OSRELEASELEN - 1);
987 			goto done_errmsg;
988 		}
989 	}
990 
991 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
992 	if (error == ENOENT)
993 		osreldt = 0;
994 	else if (error != 0)
995 		goto done_free;
996 	else {
997 		if (flags & JAIL_UPDATE) {
998 			error = EINVAL;
999 			vfs_opterror(opts,
1000 			    "osreldate cannot be changed after creation");
1001 			goto done_errmsg;
1002 		}
1003 		if (osreldt == 0) {
1004 			error = EINVAL;
1005 			vfs_opterror(opts, "osreldate cannot be 0");
1006 			goto done_errmsg;
1007 		}
1008 	}
1009 
1010 	/*
1011 	 * Grab the allprison lock before letting modules check their
1012 	 * parameters.  Once we have it, do not let go so we'll have a
1013 	 * consistent view of the OSD list.
1014 	 */
1015 	sx_xlock(&allprison_lock);
1016 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
1017 	if (error)
1018 		goto done_unlock_list;
1019 
1020 	/* By now, all parameters should have been noted. */
1021 	TAILQ_FOREACH(opt, opts, link) {
1022 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1023 			error = EINVAL;
1024 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1025 			goto done_unlock_list;
1026 		}
1027 	}
1028 
1029 	/*
1030 	 * See if we are creating a new record or updating an existing one.
1031 	 * This abuses the file error codes ENOENT and EEXIST.
1032 	 */
1033 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1034 	if (!cuflags) {
1035 		error = EINVAL;
1036 		vfs_opterror(opts, "no valid operation (create or update)");
1037 		goto done_unlock_list;
1038 	}
1039 	pr = NULL;
1040 	namelc = NULL;
1041 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1042 		namelc = strrchr(name, '.');
1043 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1044 		if (*p != '\0')
1045 			jid = 0;
1046 	}
1047 	if (jid != 0) {
1048 		/*
1049 		 * See if a requested jid already exists.  There is an
1050 		 * information leak here if the jid exists but is not within
1051 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1052 		 * even though they cannot see the jail, and CREATE | UPDATE
1053 		 * will return ENOENT which is not normally a valid error.
1054 		 */
1055 		if (jid < 0) {
1056 			error = EINVAL;
1057 			vfs_opterror(opts, "negative jid");
1058 			goto done_unlock_list;
1059 		}
1060 		pr = prison_find(jid);
1061 		if (pr != NULL) {
1062 			ppr = pr->pr_parent;
1063 			/* Create: jid must not exist. */
1064 			if (cuflags == JAIL_CREATE) {
1065 				mtx_unlock(&pr->pr_mtx);
1066 				error = EEXIST;
1067 				vfs_opterror(opts, "jail %d already exists",
1068 				    jid);
1069 				goto done_unlock_list;
1070 			}
1071 			if (!prison_ischild(mypr, pr)) {
1072 				mtx_unlock(&pr->pr_mtx);
1073 				pr = NULL;
1074 			} else if (pr->pr_uref == 0) {
1075 				if (!(flags & JAIL_DYING)) {
1076 					mtx_unlock(&pr->pr_mtx);
1077 					error = ENOENT;
1078 					vfs_opterror(opts, "jail %d is dying",
1079 					    jid);
1080 					goto done_unlock_list;
1081 				} else if ((flags & JAIL_ATTACH) ||
1082 				    (pr_flags & PR_PERSIST)) {
1083 					/*
1084 					 * A dying jail might be resurrected
1085 					 * (via attach or persist), but first
1086 					 * it must determine if another jail
1087 					 * has claimed its name.  Accomplish
1088 					 * this by implicitly re-setting the
1089 					 * name.
1090 					 */
1091 					if (name == NULL)
1092 						name = prison_name(mypr, pr);
1093 				}
1094 			}
1095 		}
1096 		if (pr == NULL) {
1097 			/* Update: jid must exist. */
1098 			if (cuflags == JAIL_UPDATE) {
1099 				error = ENOENT;
1100 				vfs_opterror(opts, "jail %d not found", jid);
1101 				goto done_unlock_list;
1102 			}
1103 		}
1104 	}
1105 	/*
1106 	 * If the caller provided a name, look for a jail by that name.
1107 	 * This has different semantics for creates and updates keyed by jid
1108 	 * (where the name must not already exist in a different jail),
1109 	 * and updates keyed by the name itself (where the name must exist
1110 	 * because that is the jail being updated).
1111 	 */
1112 	if (name != NULL) {
1113 		namelc = strrchr(name, '.');
1114 		if (namelc == NULL)
1115 			namelc = name;
1116 		else {
1117 			/*
1118 			 * This is a hierarchical name.  Split it into the
1119 			 * parent and child names, and make sure the parent
1120 			 * exists or matches an already found jail.
1121 			 */
1122 			*namelc = '\0';
1123 			if (pr != NULL) {
1124 				if (strncmp(name, ppr->pr_name, namelc - name)
1125 				    || ppr->pr_name[namelc - name] != '\0') {
1126 					mtx_unlock(&pr->pr_mtx);
1127 					error = EINVAL;
1128 					vfs_opterror(opts,
1129 					    "cannot change jail's parent");
1130 					goto done_unlock_list;
1131 				}
1132 			} else {
1133 				ppr = prison_find_name(mypr, name);
1134 				if (ppr == NULL) {
1135 					error = ENOENT;
1136 					vfs_opterror(opts,
1137 					    "jail \"%s\" not found", name);
1138 					goto done_unlock_list;
1139 				}
1140 				mtx_unlock(&ppr->pr_mtx);
1141 			}
1142 			name = ++namelc;
1143 		}
1144 		if (name[0] != '\0') {
1145 			namelen =
1146 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1147  name_again:
1148 			deadpr = NULL;
1149 			FOREACH_PRISON_CHILD(ppr, tpr) {
1150 				if (tpr != pr && tpr->pr_ref > 0 &&
1151 				    !strcmp(tpr->pr_name + namelen, name)) {
1152 					if (pr == NULL &&
1153 					    cuflags != JAIL_CREATE) {
1154 						mtx_lock(&tpr->pr_mtx);
1155 						if (tpr->pr_ref > 0) {
1156 							/*
1157 							 * Use this jail
1158 							 * for updates.
1159 							 */
1160 							if (tpr->pr_uref > 0) {
1161 								pr = tpr;
1162 								break;
1163 							}
1164 							deadpr = tpr;
1165 						}
1166 						mtx_unlock(&tpr->pr_mtx);
1167 					} else if (tpr->pr_uref > 0) {
1168 						/*
1169 						 * Create, or update(jid):
1170 						 * name must not exist in an
1171 						 * active sibling jail.
1172 						 */
1173 						error = EEXIST;
1174 						if (pr != NULL)
1175 							mtx_unlock(&pr->pr_mtx);
1176 						vfs_opterror(opts,
1177 						   "jail \"%s\" already exists",
1178 						   name);
1179 						goto done_unlock_list;
1180 					}
1181 				}
1182 			}
1183 			/* If no active jail is found, use a dying one. */
1184 			if (deadpr != NULL && pr == NULL) {
1185 				if (flags & JAIL_DYING) {
1186 					mtx_lock(&deadpr->pr_mtx);
1187 					if (deadpr->pr_ref == 0) {
1188 						mtx_unlock(&deadpr->pr_mtx);
1189 						goto name_again;
1190 					}
1191 					pr = deadpr;
1192 				} else if (cuflags == JAIL_UPDATE) {
1193 					error = ENOENT;
1194 					vfs_opterror(opts,
1195 					    "jail \"%s\" is dying", name);
1196 					goto done_unlock_list;
1197 				}
1198 			}
1199 			/* Update: name must exist if no jid. */
1200 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1201 				error = ENOENT;
1202 				vfs_opterror(opts, "jail \"%s\" not found",
1203 				    name);
1204 				goto done_unlock_list;
1205 			}
1206 		}
1207 	}
1208 	/* Update: must provide a jid or name. */
1209 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1210 		error = ENOENT;
1211 		vfs_opterror(opts, "update specified no jail");
1212 		goto done_unlock_list;
1213 	}
1214 
1215 	/* If there's no prison to update, create a new one and link it in. */
1216 	if (pr == NULL) {
1217 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1218 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1219 				error = EPERM;
1220 				vfs_opterror(opts, "prison limit exceeded");
1221 				goto done_unlock_list;
1222 			}
1223 		created = 1;
1224 		mtx_lock(&ppr->pr_mtx);
1225 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1226 			mtx_unlock(&ppr->pr_mtx);
1227 			error = ENOENT;
1228 			vfs_opterror(opts, "parent jail went away!");
1229 			goto done_unlock_list;
1230 		}
1231 		ppr->pr_ref++;
1232 		ppr->pr_uref++;
1233 		mtx_unlock(&ppr->pr_mtx);
1234 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1235 		if (jid == 0) {
1236 			/* Find the next free jid. */
1237 			jid = lastprid + 1;
1238  findnext:
1239 			if (jid == JAIL_MAX)
1240 				jid = 1;
1241 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1242 				if (tpr->pr_id < jid)
1243 					continue;
1244 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1245 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1246 					break;
1247 				}
1248 				if (jid == lastprid) {
1249 					error = EAGAIN;
1250 					vfs_opterror(opts,
1251 					    "no available jail IDs");
1252 					free(pr, M_PRISON);
1253 					prison_deref(ppr, PD_DEREF |
1254 					    PD_DEUREF | PD_LIST_XLOCKED);
1255 					goto done_releroot;
1256 				}
1257 				jid++;
1258 				goto findnext;
1259 			}
1260 			lastprid = jid;
1261 		} else {
1262 			/*
1263 			 * The jail already has a jid (that did not yet exist),
1264 			 * so just find where to insert it.
1265 			 */
1266 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1267 				if (tpr->pr_id >= jid) {
1268 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1269 					break;
1270 				}
1271 		}
1272 		if (tpr == NULL)
1273 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1274 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1275 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1276 			tpr->pr_childcount++;
1277 
1278 		pr->pr_parent = ppr;
1279 		pr->pr_id = jid;
1280 
1281 		/* Set some default values, and inherit some from the parent. */
1282 		if (name == NULL)
1283 			name = "";
1284 		if (path == NULL) {
1285 			path = "/";
1286 			root = mypr->pr_root;
1287 			vref(root);
1288 		}
1289 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1290 		pr->pr_flags |= PR_HOST;
1291 #if defined(INET) || defined(INET6)
1292 #ifdef VIMAGE
1293 		if (!(pr_flags & PR_VNET))
1294 #endif
1295 		{
1296 #ifdef INET
1297 			if (!(ch_flags & PR_IP4_USER))
1298 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1299 			else if (!(pr_flags & PR_IP4_USER)) {
1300 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1301 				if (ppr->pr_ip4 != NULL) {
1302 					pr->pr_ip4s = ppr->pr_ip4s;
1303 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1304 					    sizeof(struct in_addr), M_PRISON,
1305 					    M_WAITOK);
1306 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1307 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1308 				}
1309 			}
1310 #endif
1311 #ifdef INET6
1312 			if (!(ch_flags & PR_IP6_USER))
1313 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1314 			else if (!(pr_flags & PR_IP6_USER)) {
1315 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1316 				if (ppr->pr_ip6 != NULL) {
1317 					pr->pr_ip6s = ppr->pr_ip6s;
1318 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1319 					    sizeof(struct in6_addr), M_PRISON,
1320 					    M_WAITOK);
1321 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1322 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1323 				}
1324 			}
1325 #endif
1326 		}
1327 #endif
1328 		/* Source address selection is always on by default. */
1329 		pr->pr_flags |= _PR_IP_SADDRSEL;
1330 
1331 		pr->pr_securelevel = ppr->pr_securelevel;
1332 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1333 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1334 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1335 
1336 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1337 		if (osrelstr == NULL)
1338 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1339 		else
1340 		    strcpy(pr->pr_osrelease, osrelstr);
1341 
1342 		LIST_INIT(&pr->pr_children);
1343 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1344 
1345 #ifdef VIMAGE
1346 		/* Allocate a new vnet if specified. */
1347 		pr->pr_vnet = (pr_flags & PR_VNET)
1348 		    ? vnet_alloc() : ppr->pr_vnet;
1349 #endif
1350 		/*
1351 		 * Allocate a dedicated cpuset for each jail.
1352 		 * Unlike other initial settings, this may return an erorr.
1353 		 */
1354 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1355 		if (error) {
1356 			prison_deref(pr, PD_LIST_XLOCKED);
1357 			goto done_releroot;
1358 		}
1359 
1360 		mtx_lock(&pr->pr_mtx);
1361 		/*
1362 		 * New prisons do not yet have a reference, because we do not
1363 		 * want other to see the incomplete prison once the
1364 		 * allprison_lock is downgraded.
1365 		 */
1366 	} else {
1367 		created = 0;
1368 		/*
1369 		 * Grab a reference for existing prisons, to ensure they
1370 		 * continue to exist for the duration of the call.
1371 		 */
1372 		pr->pr_ref++;
1373 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1374 		if ((pr->pr_flags & PR_VNET) &&
1375 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1376 			error = EINVAL;
1377 			vfs_opterror(opts,
1378 			    "vnet jails cannot have IP address restrictions");
1379 			goto done_deref_locked;
1380 		}
1381 #endif
1382 #ifdef INET
1383 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1384 			error = EINVAL;
1385 			vfs_opterror(opts,
1386 			    "ip4 cannot be changed after creation");
1387 			goto done_deref_locked;
1388 		}
1389 #endif
1390 #ifdef INET6
1391 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1392 			error = EINVAL;
1393 			vfs_opterror(opts,
1394 			    "ip6 cannot be changed after creation");
1395 			goto done_deref_locked;
1396 		}
1397 #endif
1398 	}
1399 
1400 	/* Do final error checking before setting anything. */
1401 	if (gotslevel) {
1402 		if (slevel < ppr->pr_securelevel) {
1403 			error = EPERM;
1404 			goto done_deref_locked;
1405 		}
1406 	}
1407 	if (gotchildmax) {
1408 		if (childmax >= ppr->pr_childmax) {
1409 			error = EPERM;
1410 			goto done_deref_locked;
1411 		}
1412 	}
1413 	if (gotenforce) {
1414 		if (enforce < ppr->pr_enforce_statfs) {
1415 			error = EPERM;
1416 			goto done_deref_locked;
1417 		}
1418 	}
1419 	if (gotrsnum) {
1420 		/*
1421 		 * devfs_rsnum is a uint16_t
1422 		 */
1423 		if (rsnum < 0 || rsnum > 65535) {
1424 			error = EINVAL;
1425 			goto done_deref_locked;
1426 		}
1427 		/*
1428 		 * Nested jails always inherit parent's devfs ruleset
1429 		 */
1430 		if (jailed(td->td_ucred)) {
1431 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1432 				error = EPERM;
1433 				goto done_deref_locked;
1434 			} else
1435 				rsnum = ppr->pr_devfs_rsnum;
1436 		}
1437 	}
1438 #ifdef INET
1439 	if (ip4s > 0) {
1440 		if (ppr->pr_flags & PR_IP4) {
1441 			/*
1442 			 * Make sure the new set of IP addresses is a
1443 			 * subset of the parent's list.  Don't worry
1444 			 * about the parent being unlocked, as any
1445 			 * setting is done with allprison_lock held.
1446 			 */
1447 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1448 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1449 					break;
1450 			if (ij == ppr->pr_ip4s) {
1451 				error = EPERM;
1452 				goto done_deref_locked;
1453 			}
1454 			if (ip4s > 1) {
1455 				for (ii = ij = 1; ii < ip4s; ii++) {
1456 					if (ip4[ii].s_addr ==
1457 					    ppr->pr_ip4[0].s_addr)
1458 						continue;
1459 					for (; ij < ppr->pr_ip4s; ij++)
1460 						if (ip4[ii].s_addr ==
1461 						    ppr->pr_ip4[ij].s_addr)
1462 							break;
1463 					if (ij == ppr->pr_ip4s)
1464 						break;
1465 				}
1466 				if (ij == ppr->pr_ip4s) {
1467 					error = EPERM;
1468 					goto done_deref_locked;
1469 				}
1470 			}
1471 		}
1472 		/*
1473 		 * Check for conflicting IP addresses.  We permit them
1474 		 * if there is no more than one IP on each jail.  If
1475 		 * there is a duplicate on a jail with more than one
1476 		 * IP stop checking and return error.
1477 		 */
1478 		tppr = ppr;
1479 #ifdef VIMAGE
1480 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1481 			if (tppr->pr_flags & PR_VNET)
1482 				break;
1483 #endif
1484 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1485 			if (tpr == pr ||
1486 #ifdef VIMAGE
1487 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1488 #endif
1489 			    tpr->pr_uref == 0) {
1490 				descend = 0;
1491 				continue;
1492 			}
1493 			if (!(tpr->pr_flags & PR_IP4_USER))
1494 				continue;
1495 			descend = 0;
1496 			if (tpr->pr_ip4 == NULL ||
1497 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1498 				continue;
1499 			for (ii = 0; ii < ip4s; ii++) {
1500 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1501 					error = EADDRINUSE;
1502 					vfs_opterror(opts,
1503 					    "IPv4 addresses clash");
1504 					goto done_deref_locked;
1505 				}
1506 			}
1507 		}
1508 	}
1509 #endif
1510 #ifdef INET6
1511 	if (ip6s > 0) {
1512 		if (ppr->pr_flags & PR_IP6) {
1513 			/*
1514 			 * Make sure the new set of IP addresses is a
1515 			 * subset of the parent's list.
1516 			 */
1517 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1518 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1519 				    &ppr->pr_ip6[ij]))
1520 					break;
1521 			if (ij == ppr->pr_ip6s) {
1522 				error = EPERM;
1523 				goto done_deref_locked;
1524 			}
1525 			if (ip6s > 1) {
1526 				for (ii = ij = 1; ii < ip6s; ii++) {
1527 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1528 					     &ppr->pr_ip6[0]))
1529 						continue;
1530 					for (; ij < ppr->pr_ip6s; ij++)
1531 						if (IN6_ARE_ADDR_EQUAL(
1532 						    &ip6[ii], &ppr->pr_ip6[ij]))
1533 							break;
1534 					if (ij == ppr->pr_ip6s)
1535 						break;
1536 				}
1537 				if (ij == ppr->pr_ip6s) {
1538 					error = EPERM;
1539 					goto done_deref_locked;
1540 				}
1541 			}
1542 		}
1543 		/* Check for conflicting IP addresses. */
1544 		tppr = ppr;
1545 #ifdef VIMAGE
1546 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1547 			if (tppr->pr_flags & PR_VNET)
1548 				break;
1549 #endif
1550 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1551 			if (tpr == pr ||
1552 #ifdef VIMAGE
1553 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1554 #endif
1555 			    tpr->pr_uref == 0) {
1556 				descend = 0;
1557 				continue;
1558 			}
1559 			if (!(tpr->pr_flags & PR_IP6_USER))
1560 				continue;
1561 			descend = 0;
1562 			if (tpr->pr_ip6 == NULL ||
1563 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1564 				continue;
1565 			for (ii = 0; ii < ip6s; ii++) {
1566 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1567 					error = EADDRINUSE;
1568 					vfs_opterror(opts,
1569 					    "IPv6 addresses clash");
1570 					goto done_deref_locked;
1571 				}
1572 			}
1573 		}
1574 	}
1575 #endif
1576 	onamelen = namelen = 0;
1577 	if (name != NULL) {
1578 		/* Give a default name of the jid.  Also allow the name to be
1579 		 * explicitly the jid - but not any other number, and only in
1580 		 * normal form (no leading zero/etc).
1581 		 */
1582 		if (name[0] == '\0')
1583 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1584 		else if ((strtoul(namelc, &p, 10) != jid ||
1585 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1586 			error = EINVAL;
1587 			vfs_opterror(opts,
1588 			    "name cannot be numeric (unless it is the jid)");
1589 			goto done_deref_locked;
1590 		}
1591 		/*
1592 		 * Make sure the name isn't too long for the prison or its
1593 		 * children.
1594 		 */
1595 		onamelen = strlen(pr->pr_name);
1596 		namelen = strlen(name);
1597 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1598 			error = ENAMETOOLONG;
1599 			goto done_deref_locked;
1600 		}
1601 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1602 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1603 			    sizeof(pr->pr_name)) {
1604 				error = ENAMETOOLONG;
1605 				goto done_deref_locked;
1606 			}
1607 		}
1608 	}
1609 	if (pr_allow & ~ppr->pr_allow) {
1610 		error = EPERM;
1611 		goto done_deref_locked;
1612 	}
1613 
1614 	/* Set the parameters of the prison. */
1615 #ifdef INET
1616 	redo_ip4 = 0;
1617 	if (pr_flags & PR_IP4_USER) {
1618 		pr->pr_flags |= PR_IP4;
1619 		free(pr->pr_ip4, M_PRISON);
1620 		pr->pr_ip4s = ip4s;
1621 		pr->pr_ip4 = ip4;
1622 		ip4 = NULL;
1623 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1624 #ifdef VIMAGE
1625 			if (tpr->pr_flags & PR_VNET) {
1626 				descend = 0;
1627 				continue;
1628 			}
1629 #endif
1630 			if (prison_restrict_ip4(tpr, NULL)) {
1631 				redo_ip4 = 1;
1632 				descend = 0;
1633 			}
1634 		}
1635 	}
1636 #endif
1637 #ifdef INET6
1638 	redo_ip6 = 0;
1639 	if (pr_flags & PR_IP6_USER) {
1640 		pr->pr_flags |= PR_IP6;
1641 		free(pr->pr_ip6, M_PRISON);
1642 		pr->pr_ip6s = ip6s;
1643 		pr->pr_ip6 = ip6;
1644 		ip6 = NULL;
1645 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1646 #ifdef VIMAGE
1647 			if (tpr->pr_flags & PR_VNET) {
1648 				descend = 0;
1649 				continue;
1650 			}
1651 #endif
1652 			if (prison_restrict_ip6(tpr, NULL)) {
1653 				redo_ip6 = 1;
1654 				descend = 0;
1655 			}
1656 		}
1657 	}
1658 #endif
1659 	if (gotslevel) {
1660 		pr->pr_securelevel = slevel;
1661 		/* Set all child jails to be at least this level. */
1662 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1663 			if (tpr->pr_securelevel < slevel)
1664 				tpr->pr_securelevel = slevel;
1665 	}
1666 	if (gotchildmax) {
1667 		pr->pr_childmax = childmax;
1668 		/* Set all child jails to under this limit. */
1669 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1670 			if (tpr->pr_childmax > childmax - level)
1671 				tpr->pr_childmax = childmax > level
1672 				    ? childmax - level : 0;
1673 	}
1674 	if (gotenforce) {
1675 		pr->pr_enforce_statfs = enforce;
1676 		/* Pass this restriction on to the children. */
1677 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1678 			if (tpr->pr_enforce_statfs < enforce)
1679 				tpr->pr_enforce_statfs = enforce;
1680 	}
1681 	if (gotrsnum) {
1682 		pr->pr_devfs_rsnum = rsnum;
1683 		/* Pass this restriction on to the children. */
1684 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1685 			tpr->pr_devfs_rsnum = rsnum;
1686 	}
1687 	if (name != NULL) {
1688 		if (ppr == &prison0)
1689 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1690 		else
1691 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1692 			    ppr->pr_name, name);
1693 		/* Change this component of child names. */
1694 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1695 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1696 			    strlen(tpr->pr_name + onamelen) + 1);
1697 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1698 		}
1699 	}
1700 	if (path != NULL) {
1701 		/* Try to keep a real-rooted full pathname. */
1702 		if (fullpath_disabled && path[0] == '/' &&
1703 		    strcmp(mypr->pr_path, "/"))
1704 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1705 			    mypr->pr_path, path);
1706 		else
1707 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1708 		pr->pr_root = root;
1709 	}
1710 	if (PR_HOST & ch_flags & ~pr_flags) {
1711 		if (pr->pr_flags & PR_HOST) {
1712 			/*
1713 			 * Copy the parent's host info.  As with pr_ip4 above,
1714 			 * the lack of a lock on the parent is not a problem;
1715 			 * it is always set with allprison_lock at least
1716 			 * shared, and is held exclusively here.
1717 			 */
1718 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1719 			    sizeof(pr->pr_hostname));
1720 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1721 			    sizeof(pr->pr_domainname));
1722 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1723 			    sizeof(pr->pr_hostuuid));
1724 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1725 		}
1726 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1727 		/* Set this prison, and any descendants without PR_HOST. */
1728 		if (host != NULL)
1729 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1730 		if (domain != NULL)
1731 			strlcpy(pr->pr_domainname, domain,
1732 			    sizeof(pr->pr_domainname));
1733 		if (uuid != NULL)
1734 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1735 		if (gothid)
1736 			pr->pr_hostid = hid;
1737 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1738 			if (tpr->pr_flags & PR_HOST)
1739 				descend = 0;
1740 			else {
1741 				if (host != NULL)
1742 					strlcpy(tpr->pr_hostname,
1743 					    pr->pr_hostname,
1744 					    sizeof(tpr->pr_hostname));
1745 				if (domain != NULL)
1746 					strlcpy(tpr->pr_domainname,
1747 					    pr->pr_domainname,
1748 					    sizeof(tpr->pr_domainname));
1749 				if (uuid != NULL)
1750 					strlcpy(tpr->pr_hostuuid,
1751 					    pr->pr_hostuuid,
1752 					    sizeof(tpr->pr_hostuuid));
1753 				if (gothid)
1754 					tpr->pr_hostid = hid;
1755 			}
1756 		}
1757 	}
1758 	if ((tallow = ch_allow & ~pr_allow)) {
1759 		/* Clear allow bits in all children. */
1760 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1761 			tpr->pr_allow &= ~tallow;
1762 	}
1763 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1764 	/*
1765 	 * Persistent prisons get an extra reference, and prisons losing their
1766 	 * persist flag lose that reference.  Only do this for existing prisons
1767 	 * for now, so new ones will remain unseen until after the module
1768 	 * handlers have completed.
1769 	 */
1770 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1771 		if (pr_flags & PR_PERSIST) {
1772 			pr->pr_ref++;
1773 			pr->pr_uref++;
1774 		} else {
1775 			pr->pr_ref--;
1776 			pr->pr_uref--;
1777 		}
1778 	}
1779 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1780 	mtx_unlock(&pr->pr_mtx);
1781 
1782 #ifdef RACCT
1783 	if (racct_enable && created)
1784 		prison_racct_attach(pr);
1785 #endif
1786 
1787 	/* Locks may have prevented a complete restriction of child IP
1788 	 * addresses.  If so, allocate some more memory and try again.
1789 	 */
1790 #ifdef INET
1791 	while (redo_ip4) {
1792 		ip4s = pr->pr_ip4s;
1793 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1794 		mtx_lock(&pr->pr_mtx);
1795 		redo_ip4 = 0;
1796 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1797 #ifdef VIMAGE
1798 			if (tpr->pr_flags & PR_VNET) {
1799 				descend = 0;
1800 				continue;
1801 			}
1802 #endif
1803 			if (prison_restrict_ip4(tpr, ip4)) {
1804 				if (ip4 != NULL)
1805 					ip4 = NULL;
1806 				else
1807 					redo_ip4 = 1;
1808 			}
1809 		}
1810 		mtx_unlock(&pr->pr_mtx);
1811 	}
1812 #endif
1813 #ifdef INET6
1814 	while (redo_ip6) {
1815 		ip6s = pr->pr_ip6s;
1816 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1817 		mtx_lock(&pr->pr_mtx);
1818 		redo_ip6 = 0;
1819 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1820 #ifdef VIMAGE
1821 			if (tpr->pr_flags & PR_VNET) {
1822 				descend = 0;
1823 				continue;
1824 			}
1825 #endif
1826 			if (prison_restrict_ip6(tpr, ip6)) {
1827 				if (ip6 != NULL)
1828 					ip6 = NULL;
1829 				else
1830 					redo_ip6 = 1;
1831 			}
1832 		}
1833 		mtx_unlock(&pr->pr_mtx);
1834 	}
1835 #endif
1836 
1837 	/* Let the modules do their work. */
1838 	sx_downgrade(&allprison_lock);
1839 	if (created) {
1840 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1841 		if (error) {
1842 			prison_deref(pr, PD_LIST_SLOCKED);
1843 			goto done_errmsg;
1844 		}
1845 	}
1846 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1847 	if (error) {
1848 		prison_deref(pr, created
1849 		    ? PD_LIST_SLOCKED
1850 		    : PD_DEREF | PD_LIST_SLOCKED);
1851 		goto done_errmsg;
1852 	}
1853 
1854 	/* Attach this process to the prison if requested. */
1855 	if (flags & JAIL_ATTACH) {
1856 		mtx_lock(&pr->pr_mtx);
1857 		error = do_jail_attach(td, pr);
1858 		if (error) {
1859 			vfs_opterror(opts, "attach failed");
1860 			if (!created)
1861 				prison_deref(pr, PD_DEREF);
1862 			goto done_errmsg;
1863 		}
1864 	}
1865 
1866 #ifdef RACCT
1867 	if (racct_enable && !created) {
1868 		if (!(flags & JAIL_ATTACH))
1869 			sx_sunlock(&allprison_lock);
1870 		prison_racct_modify(pr);
1871 		if (!(flags & JAIL_ATTACH))
1872 			sx_slock(&allprison_lock);
1873 	}
1874 #endif
1875 
1876 	td->td_retval[0] = pr->pr_id;
1877 
1878 	/*
1879 	 * Now that it is all there, drop the temporary reference from existing
1880 	 * prisons.  Or add a reference to newly created persistent prisons
1881 	 * (which was not done earlier so that the prison would not be publicly
1882 	 * visible).
1883 	 */
1884 	if (!created) {
1885 		prison_deref(pr, (flags & JAIL_ATTACH)
1886 		    ? PD_DEREF
1887 		    : PD_DEREF | PD_LIST_SLOCKED);
1888 	} else {
1889 		if (pr_flags & PR_PERSIST) {
1890 			mtx_lock(&pr->pr_mtx);
1891 			pr->pr_ref++;
1892 			pr->pr_uref++;
1893 			mtx_unlock(&pr->pr_mtx);
1894 		}
1895 		if (!(flags & JAIL_ATTACH))
1896 			sx_sunlock(&allprison_lock);
1897 	}
1898 
1899 	goto done_errmsg;
1900 
1901  done_deref_locked:
1902 	prison_deref(pr, created
1903 	    ? PD_LOCKED | PD_LIST_XLOCKED
1904 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1905 	goto done_releroot;
1906  done_unlock_list:
1907 	sx_xunlock(&allprison_lock);
1908  done_releroot:
1909 	if (root != NULL)
1910 		vrele(root);
1911  done_errmsg:
1912 	if (error) {
1913 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1914 		if (errmsg_len > 0) {
1915 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1916 			if (errmsg_pos > 0) {
1917 				if (optuio->uio_segflg == UIO_SYSSPACE)
1918 					bcopy(errmsg,
1919 					   optuio->uio_iov[errmsg_pos].iov_base,
1920 					   errmsg_len);
1921 				else
1922 					copyout(errmsg,
1923 					   optuio->uio_iov[errmsg_pos].iov_base,
1924 					   errmsg_len);
1925 			}
1926 		}
1927 	}
1928  done_free:
1929 #ifdef INET
1930 	free(ip4, M_PRISON);
1931 #endif
1932 #ifdef INET6
1933 	free(ip6, M_PRISON);
1934 #endif
1935 	if (g_path != NULL)
1936 		free(g_path, M_TEMP);
1937 	vfs_freeopts(opts);
1938 	return (error);
1939 }
1940 
1941 
1942 /*
1943  * struct jail_get_args {
1944  *	struct iovec *iovp;
1945  *	unsigned int iovcnt;
1946  *	int flags;
1947  * };
1948  */
1949 int
1950 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1951 {
1952 	struct uio *auio;
1953 	int error;
1954 
1955 	/* Check that we have an even number of iovecs. */
1956 	if (uap->iovcnt & 1)
1957 		return (EINVAL);
1958 
1959 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1960 	if (error)
1961 		return (error);
1962 	error = kern_jail_get(td, auio, uap->flags);
1963 	if (error == 0)
1964 		error = copyout(auio->uio_iov, uap->iovp,
1965 		    uap->iovcnt * sizeof (struct iovec));
1966 	free(auio, M_IOV);
1967 	return (error);
1968 }
1969 
1970 int
1971 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1972 {
1973 	struct prison *pr, *mypr;
1974 	struct vfsopt *opt;
1975 	struct vfsoptlist *opts;
1976 	char *errmsg, *name;
1977 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1978 
1979 	if (flags & ~JAIL_GET_MASK)
1980 		return (EINVAL);
1981 
1982 	/* Get the parameter list. */
1983 	error = vfs_buildopts(optuio, &opts);
1984 	if (error)
1985 		return (error);
1986 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1987 	mypr = td->td_ucred->cr_prison;
1988 
1989 	/*
1990 	 * Find the prison specified by one of: lastjid, jid, name.
1991 	 */
1992 	sx_slock(&allprison_lock);
1993 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1994 	if (error == 0) {
1995 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1996 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1997 				mtx_lock(&pr->pr_mtx);
1998 				if (pr->pr_ref > 0 &&
1999 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2000 					break;
2001 				mtx_unlock(&pr->pr_mtx);
2002 			}
2003 		}
2004 		if (pr != NULL)
2005 			goto found_prison;
2006 		error = ENOENT;
2007 		vfs_opterror(opts, "no jail after %d", jid);
2008 		goto done_unlock_list;
2009 	} else if (error != ENOENT)
2010 		goto done_unlock_list;
2011 
2012 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2013 	if (error == 0) {
2014 		if (jid != 0) {
2015 			pr = prison_find_child(mypr, jid);
2016 			if (pr != NULL) {
2017 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2018 					mtx_unlock(&pr->pr_mtx);
2019 					error = ENOENT;
2020 					vfs_opterror(opts, "jail %d is dying",
2021 					    jid);
2022 					goto done_unlock_list;
2023 				}
2024 				goto found_prison;
2025 			}
2026 			error = ENOENT;
2027 			vfs_opterror(opts, "jail %d not found", jid);
2028 			goto done_unlock_list;
2029 		}
2030 	} else if (error != ENOENT)
2031 		goto done_unlock_list;
2032 
2033 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2034 	if (error == 0) {
2035 		if (len == 0 || name[len - 1] != '\0') {
2036 			error = EINVAL;
2037 			goto done_unlock_list;
2038 		}
2039 		pr = prison_find_name(mypr, name);
2040 		if (pr != NULL) {
2041 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2042 				mtx_unlock(&pr->pr_mtx);
2043 				error = ENOENT;
2044 				vfs_opterror(opts, "jail \"%s\" is dying",
2045 				    name);
2046 				goto done_unlock_list;
2047 			}
2048 			goto found_prison;
2049 		}
2050 		error = ENOENT;
2051 		vfs_opterror(opts, "jail \"%s\" not found", name);
2052 		goto done_unlock_list;
2053 	} else if (error != ENOENT)
2054 		goto done_unlock_list;
2055 
2056 	vfs_opterror(opts, "no jail specified");
2057 	error = ENOENT;
2058 	goto done_unlock_list;
2059 
2060  found_prison:
2061 	/* Get the parameters of the prison. */
2062 	pr->pr_ref++;
2063 	locked = PD_LOCKED;
2064 	td->td_retval[0] = pr->pr_id;
2065 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2066 	if (error != 0 && error != ENOENT)
2067 		goto done_deref;
2068 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2069 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2070 	if (error != 0 && error != ENOENT)
2071 		goto done_deref;
2072 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2073 	if (error != 0 && error != ENOENT)
2074 		goto done_deref;
2075 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2076 	    sizeof(pr->pr_cpuset->cs_id));
2077 	if (error != 0 && error != ENOENT)
2078 		goto done_deref;
2079 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2080 	if (error != 0 && error != ENOENT)
2081 		goto done_deref;
2082 #ifdef INET
2083 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2084 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2085 	if (error != 0 && error != ENOENT)
2086 		goto done_deref;
2087 #endif
2088 #ifdef INET6
2089 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2090 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2091 	if (error != 0 && error != ENOENT)
2092 		goto done_deref;
2093 #endif
2094 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2095 	    sizeof(pr->pr_securelevel));
2096 	if (error != 0 && error != ENOENT)
2097 		goto done_deref;
2098 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2099 	    sizeof(pr->pr_childcount));
2100 	if (error != 0 && error != ENOENT)
2101 		goto done_deref;
2102 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2103 	    sizeof(pr->pr_childmax));
2104 	if (error != 0 && error != ENOENT)
2105 		goto done_deref;
2106 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2107 	if (error != 0 && error != ENOENT)
2108 		goto done_deref;
2109 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2110 	if (error != 0 && error != ENOENT)
2111 		goto done_deref;
2112 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2113 	if (error != 0 && error != ENOENT)
2114 		goto done_deref;
2115 #ifdef COMPAT_FREEBSD32
2116 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2117 		uint32_t hid32 = pr->pr_hostid;
2118 
2119 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2120 	} else
2121 #endif
2122 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2123 	    sizeof(pr->pr_hostid));
2124 	if (error != 0 && error != ENOENT)
2125 		goto done_deref;
2126 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2127 	    sizeof(pr->pr_enforce_statfs));
2128 	if (error != 0 && error != ENOENT)
2129 		goto done_deref;
2130 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2131 	    sizeof(pr->pr_devfs_rsnum));
2132 	if (error != 0 && error != ENOENT)
2133 		goto done_deref;
2134 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2135 		if (pr_flag_names[fi] == NULL)
2136 			continue;
2137 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2138 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2139 		if (error != 0 && error != ENOENT)
2140 			goto done_deref;
2141 		i = !i;
2142 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2143 		if (error != 0 && error != ENOENT)
2144 			goto done_deref;
2145 	}
2146 	for (fi = 0; fi < nitems(pr_flag_jailsys); 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 < nitems(pr_allow_names); fi++) {
2159 		if (pr_allow_names[fi] == NULL)
2160 			continue;
2161 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2162 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2163 		if (error != 0 && error != ENOENT)
2164 			goto done_deref;
2165 		i = !i;
2166 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2167 		if (error != 0 && error != ENOENT)
2168 			goto done_deref;
2169 	}
2170 	i = (pr->pr_uref == 0);
2171 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2172 	if (error != 0 && error != ENOENT)
2173 		goto done_deref;
2174 	i = !i;
2175 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2176 	if (error != 0 && error != ENOENT)
2177 		goto done_deref;
2178 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2179 	    sizeof(pr->pr_osreldate));
2180 	if (error != 0 && error != ENOENT)
2181 		goto done_deref;
2182 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2183 	if (error != 0 && error != ENOENT)
2184 		goto done_deref;
2185 
2186 	/* Get the module parameters. */
2187 	mtx_unlock(&pr->pr_mtx);
2188 	locked = 0;
2189 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2190 	if (error)
2191 		goto done_deref;
2192 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2193 
2194 	/* By now, all parameters should have been noted. */
2195 	TAILQ_FOREACH(opt, opts, link) {
2196 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2197 			error = EINVAL;
2198 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2199 			goto done_errmsg;
2200 		}
2201 	}
2202 
2203 	/* Write the fetched parameters back to userspace. */
2204 	error = 0;
2205 	TAILQ_FOREACH(opt, opts, link) {
2206 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2207 			pos = 2 * opt->pos + 1;
2208 			optuio->uio_iov[pos].iov_len = opt->len;
2209 			if (opt->value != NULL) {
2210 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2211 					bcopy(opt->value,
2212 					    optuio->uio_iov[pos].iov_base,
2213 					    opt->len);
2214 				} else {
2215 					error = copyout(opt->value,
2216 					    optuio->uio_iov[pos].iov_base,
2217 					    opt->len);
2218 					if (error)
2219 						break;
2220 				}
2221 			}
2222 		}
2223 	}
2224 	goto done_errmsg;
2225 
2226  done_deref:
2227 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2228 	goto done_errmsg;
2229 
2230  done_unlock_list:
2231 	sx_sunlock(&allprison_lock);
2232  done_errmsg:
2233 	if (error && errmsg_pos >= 0) {
2234 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2235 		errmsg_pos = 2 * errmsg_pos + 1;
2236 		if (errmsg_len > 0) {
2237 			if (optuio->uio_segflg == UIO_SYSSPACE)
2238 				bcopy(errmsg,
2239 				    optuio->uio_iov[errmsg_pos].iov_base,
2240 				    errmsg_len);
2241 			else
2242 				copyout(errmsg,
2243 				    optuio->uio_iov[errmsg_pos].iov_base,
2244 				    errmsg_len);
2245 		}
2246 	}
2247 	vfs_freeopts(opts);
2248 	return (error);
2249 }
2250 
2251 
2252 /*
2253  * struct jail_remove_args {
2254  *	int jid;
2255  * };
2256  */
2257 int
2258 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2259 {
2260 	struct prison *pr, *cpr, *lpr, *tpr;
2261 	int descend, error;
2262 
2263 	error = priv_check(td, PRIV_JAIL_REMOVE);
2264 	if (error)
2265 		return (error);
2266 
2267 	sx_xlock(&allprison_lock);
2268 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2269 	if (pr == NULL) {
2270 		sx_xunlock(&allprison_lock);
2271 		return (EINVAL);
2272 	}
2273 
2274 	/* Remove all descendants of this prison, then remove this prison. */
2275 	pr->pr_ref++;
2276 	pr->pr_flags |= PR_REMOVE;
2277 	if (!LIST_EMPTY(&pr->pr_children)) {
2278 		mtx_unlock(&pr->pr_mtx);
2279 		lpr = NULL;
2280 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2281 			mtx_lock(&cpr->pr_mtx);
2282 			if (cpr->pr_ref > 0) {
2283 				tpr = cpr;
2284 				cpr->pr_ref++;
2285 				cpr->pr_flags |= PR_REMOVE;
2286 			} else {
2287 				/* Already removed - do not do it again. */
2288 				tpr = NULL;
2289 			}
2290 			mtx_unlock(&cpr->pr_mtx);
2291 			if (lpr != NULL) {
2292 				mtx_lock(&lpr->pr_mtx);
2293 				prison_remove_one(lpr);
2294 				sx_xlock(&allprison_lock);
2295 			}
2296 			lpr = tpr;
2297 		}
2298 		if (lpr != NULL) {
2299 			mtx_lock(&lpr->pr_mtx);
2300 			prison_remove_one(lpr);
2301 			sx_xlock(&allprison_lock);
2302 		}
2303 		mtx_lock(&pr->pr_mtx);
2304 	}
2305 	prison_remove_one(pr);
2306 	return (0);
2307 }
2308 
2309 static void
2310 prison_remove_one(struct prison *pr)
2311 {
2312 	struct proc *p;
2313 	int deuref;
2314 
2315 	/* If the prison was persistent, it is not anymore. */
2316 	deuref = 0;
2317 	if (pr->pr_flags & PR_PERSIST) {
2318 		pr->pr_ref--;
2319 		deuref = PD_DEUREF;
2320 		pr->pr_flags &= ~PR_PERSIST;
2321 	}
2322 
2323 	/*
2324 	 * jail_remove added a reference.  If that's the only one, remove
2325 	 * the prison now.
2326 	 */
2327 	KASSERT(pr->pr_ref > 0,
2328 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2329 	if (pr->pr_ref == 1) {
2330 		prison_deref(pr,
2331 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2332 		return;
2333 	}
2334 
2335 	mtx_unlock(&pr->pr_mtx);
2336 	sx_xunlock(&allprison_lock);
2337 	/*
2338 	 * Kill all processes unfortunate enough to be attached to this prison.
2339 	 */
2340 	sx_slock(&allproc_lock);
2341 	LIST_FOREACH(p, &allproc, p_list) {
2342 		PROC_LOCK(p);
2343 		if (p->p_state != PRS_NEW && p->p_ucred &&
2344 		    p->p_ucred->cr_prison == pr)
2345 			kern_psignal(p, SIGKILL);
2346 		PROC_UNLOCK(p);
2347 	}
2348 	sx_sunlock(&allproc_lock);
2349 	/* Remove the temporary reference added by jail_remove. */
2350 	prison_deref(pr, deuref | PD_DEREF);
2351 }
2352 
2353 
2354 /*
2355  * struct jail_attach_args {
2356  *	int jid;
2357  * };
2358  */
2359 int
2360 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2361 {
2362 	struct prison *pr;
2363 	int error;
2364 
2365 	error = priv_check(td, PRIV_JAIL_ATTACH);
2366 	if (error)
2367 		return (error);
2368 
2369 	sx_slock(&allprison_lock);
2370 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2371 	if (pr == NULL) {
2372 		sx_sunlock(&allprison_lock);
2373 		return (EINVAL);
2374 	}
2375 
2376 	/*
2377 	 * Do not allow a process to attach to a prison that is not
2378 	 * considered to be "alive".
2379 	 */
2380 	if (pr->pr_uref == 0) {
2381 		mtx_unlock(&pr->pr_mtx);
2382 		sx_sunlock(&allprison_lock);
2383 		return (EINVAL);
2384 	}
2385 
2386 	return (do_jail_attach(td, pr));
2387 }
2388 
2389 static int
2390 do_jail_attach(struct thread *td, struct prison *pr)
2391 {
2392 	struct prison *ppr;
2393 	struct proc *p;
2394 	struct ucred *newcred, *oldcred;
2395 	int error;
2396 
2397 	/*
2398 	 * XXX: Note that there is a slight race here if two threads
2399 	 * in the same privileged process attempt to attach to two
2400 	 * different jails at the same time.  It is important for
2401 	 * user processes not to do this, or they might end up with
2402 	 * a process root from one prison, but attached to the jail
2403 	 * of another.
2404 	 */
2405 	pr->pr_ref++;
2406 	pr->pr_uref++;
2407 	mtx_unlock(&pr->pr_mtx);
2408 
2409 	/* Let modules do whatever they need to prepare for attaching. */
2410 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2411 	if (error) {
2412 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2413 		return (error);
2414 	}
2415 	sx_sunlock(&allprison_lock);
2416 
2417 	/*
2418 	 * Reparent the newly attached process to this jail.
2419 	 */
2420 	ppr = td->td_ucred->cr_prison;
2421 	p = td->td_proc;
2422 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2423 	if (error)
2424 		goto e_revert_osd;
2425 
2426 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2427 	if ((error = change_dir(pr->pr_root, td)) != 0)
2428 		goto e_unlock;
2429 #ifdef MAC
2430 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2431 		goto e_unlock;
2432 #endif
2433 	VOP_UNLOCK(pr->pr_root, 0);
2434 	if ((error = pwd_chroot(td, pr->pr_root)))
2435 		goto e_revert_osd;
2436 
2437 	newcred = crget();
2438 	PROC_LOCK(p);
2439 	oldcred = p->p_ucred;
2440 	setsugid(p);
2441 	crcopy(newcred, oldcred);
2442 	newcred->cr_prison = pr;
2443 	proc_set_cred(p, newcred);
2444 	PROC_UNLOCK(p);
2445 #ifdef RACCT
2446 	racct_proc_ucred_changed(p, oldcred, newcred);
2447 #endif
2448 	crfree(oldcred);
2449 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2450 	return (0);
2451  e_unlock:
2452 	VOP_UNLOCK(pr->pr_root, 0);
2453  e_revert_osd:
2454 	/* Tell modules this thread is still in its old jail after all. */
2455 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2456 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2457 	return (error);
2458 }
2459 
2460 
2461 /*
2462  * Returns a locked prison instance, or NULL on failure.
2463  */
2464 struct prison *
2465 prison_find(int prid)
2466 {
2467 	struct prison *pr;
2468 
2469 	sx_assert(&allprison_lock, SX_LOCKED);
2470 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2471 		if (pr->pr_id == prid) {
2472 			mtx_lock(&pr->pr_mtx);
2473 			if (pr->pr_ref > 0)
2474 				return (pr);
2475 			mtx_unlock(&pr->pr_mtx);
2476 		}
2477 	}
2478 	return (NULL);
2479 }
2480 
2481 /*
2482  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2483  */
2484 struct prison *
2485 prison_find_child(struct prison *mypr, int prid)
2486 {
2487 	struct prison *pr;
2488 	int descend;
2489 
2490 	sx_assert(&allprison_lock, SX_LOCKED);
2491 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2492 		if (pr->pr_id == prid) {
2493 			mtx_lock(&pr->pr_mtx);
2494 			if (pr->pr_ref > 0)
2495 				return (pr);
2496 			mtx_unlock(&pr->pr_mtx);
2497 		}
2498 	}
2499 	return (NULL);
2500 }
2501 
2502 /*
2503  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2504  */
2505 struct prison *
2506 prison_find_name(struct prison *mypr, const char *name)
2507 {
2508 	struct prison *pr, *deadpr;
2509 	size_t mylen;
2510 	int descend;
2511 
2512 	sx_assert(&allprison_lock, SX_LOCKED);
2513 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2514  again:
2515 	deadpr = NULL;
2516 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2517 		if (!strcmp(pr->pr_name + mylen, name)) {
2518 			mtx_lock(&pr->pr_mtx);
2519 			if (pr->pr_ref > 0) {
2520 				if (pr->pr_uref > 0)
2521 					return (pr);
2522 				deadpr = pr;
2523 			}
2524 			mtx_unlock(&pr->pr_mtx);
2525 		}
2526 	}
2527 	/* There was no valid prison - perhaps there was a dying one. */
2528 	if (deadpr != NULL) {
2529 		mtx_lock(&deadpr->pr_mtx);
2530 		if (deadpr->pr_ref == 0) {
2531 			mtx_unlock(&deadpr->pr_mtx);
2532 			goto again;
2533 		}
2534 	}
2535 	return (deadpr);
2536 }
2537 
2538 /*
2539  * See if a prison has the specific flag set.
2540  */
2541 int
2542 prison_flag(struct ucred *cred, unsigned flag)
2543 {
2544 
2545 	/* This is an atomic read, so no locking is necessary. */
2546 	return (cred->cr_prison->pr_flags & flag);
2547 }
2548 
2549 int
2550 prison_allow(struct ucred *cred, unsigned flag)
2551 {
2552 
2553 	/* This is an atomic read, so no locking is necessary. */
2554 	return (cred->cr_prison->pr_allow & flag);
2555 }
2556 
2557 /*
2558  * Remove a prison reference.  If that was the last reference, remove the
2559  * prison itself - but not in this context in case there are locks held.
2560  */
2561 void
2562 prison_free_locked(struct prison *pr)
2563 {
2564 
2565 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2566 	pr->pr_ref--;
2567 	if (pr->pr_ref == 0) {
2568 		mtx_unlock(&pr->pr_mtx);
2569 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2570 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2571 		return;
2572 	}
2573 	mtx_unlock(&pr->pr_mtx);
2574 }
2575 
2576 void
2577 prison_free(struct prison *pr)
2578 {
2579 
2580 	mtx_lock(&pr->pr_mtx);
2581 	prison_free_locked(pr);
2582 }
2583 
2584 static void
2585 prison_complete(void *context, int pending)
2586 {
2587 
2588 	prison_deref((struct prison *)context, 0);
2589 }
2590 
2591 /*
2592  * Remove a prison reference (usually).  This internal version assumes no
2593  * mutexes are held, except perhaps the prison itself.  If there are no more
2594  * references, release and delist the prison.  On completion, the prison lock
2595  * and the allprison lock are both unlocked.
2596  */
2597 static void
2598 prison_deref(struct prison *pr, int flags)
2599 {
2600 	struct prison *ppr, *tpr;
2601 
2602 	if (!(flags & PD_LOCKED))
2603 		mtx_lock(&pr->pr_mtx);
2604 	for (;;) {
2605 		if (flags & PD_DEUREF) {
2606 			pr->pr_uref--;
2607 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2608 		}
2609 		if (flags & PD_DEREF)
2610 			pr->pr_ref--;
2611 		/* If the prison still has references, nothing else to do. */
2612 		if (pr->pr_ref > 0) {
2613 			mtx_unlock(&pr->pr_mtx);
2614 			if (flags & PD_LIST_SLOCKED)
2615 				sx_sunlock(&allprison_lock);
2616 			else if (flags & PD_LIST_XLOCKED)
2617 				sx_xunlock(&allprison_lock);
2618 			return;
2619 		}
2620 
2621 		mtx_unlock(&pr->pr_mtx);
2622 		if (flags & PD_LIST_SLOCKED) {
2623 			if (!sx_try_upgrade(&allprison_lock)) {
2624 				sx_sunlock(&allprison_lock);
2625 				sx_xlock(&allprison_lock);
2626 			}
2627 		} else if (!(flags & PD_LIST_XLOCKED))
2628 			sx_xlock(&allprison_lock);
2629 
2630 		TAILQ_REMOVE(&allprison, pr, pr_list);
2631 		LIST_REMOVE(pr, pr_sibling);
2632 		ppr = pr->pr_parent;
2633 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2634 			tpr->pr_childcount--;
2635 		sx_xunlock(&allprison_lock);
2636 
2637 #ifdef VIMAGE
2638 		if (pr->pr_vnet != ppr->pr_vnet)
2639 			vnet_destroy(pr->pr_vnet);
2640 #endif
2641 		if (pr->pr_root != NULL)
2642 			vrele(pr->pr_root);
2643 		mtx_destroy(&pr->pr_mtx);
2644 #ifdef INET
2645 		free(pr->pr_ip4, M_PRISON);
2646 #endif
2647 #ifdef INET6
2648 		free(pr->pr_ip6, M_PRISON);
2649 #endif
2650 		if (pr->pr_cpuset != NULL)
2651 			cpuset_rel(pr->pr_cpuset);
2652 		osd_jail_exit(pr);
2653 #ifdef RACCT
2654 		if (racct_enable)
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_linprocfs_allowed,
4293     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4294     NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4295     "Processes in jail can mount the linprocfs file system");
4296 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4297     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4298     NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4299     "Processes in jail can mount the linsysfs file system");
4300 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4301     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4302     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4303     "Processes in jail can mount the tmpfs file system");
4304 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4305     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4306     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4307     "Processes in jail can mount the zfs file system");
4308 
4309 static int
4310 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4311 {
4312 	struct prison *pr;
4313 	int level, error;
4314 
4315 	pr = req->td->td_ucred->cr_prison;
4316 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4317 	error = sysctl_handle_int(oidp, &level, 0, req);
4318 	if (error || !req->newptr)
4319 		return (error);
4320 	*(int *)arg1 = level;
4321 	return (0);
4322 }
4323 
4324 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4325     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4326     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4327     sysctl_jail_default_level, "I",
4328     "Processes in jail cannot see all mounted file systems");
4329 
4330 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4331     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4332     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4333     sysctl_jail_default_level, "I",
4334     "Ruleset for the devfs filesystem in jail");
4335 
4336 /*
4337  * Nodes to describe jail parameters.  Maximum length of string parameters
4338  * is returned in the string itself, and the other parameters exist merely
4339  * to make themselves and their types known.
4340  */
4341 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4342     "Jail parameters");
4343 
4344 int
4345 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4346 {
4347 	int i;
4348 	long l;
4349 	size_t s;
4350 	char numbuf[12];
4351 
4352 	switch (oidp->oid_kind & CTLTYPE)
4353 	{
4354 	case CTLTYPE_LONG:
4355 	case CTLTYPE_ULONG:
4356 		l = 0;
4357 #ifdef SCTL_MASK32
4358 		if (!(req->flags & SCTL_MASK32))
4359 #endif
4360 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4361 	case CTLTYPE_INT:
4362 	case CTLTYPE_UINT:
4363 		i = 0;
4364 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4365 	case CTLTYPE_STRING:
4366 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4367 		return
4368 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4369 	case CTLTYPE_STRUCT:
4370 		s = (size_t)arg2;
4371 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4372 	}
4373 	return (0);
4374 }
4375 
4376 /*
4377  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4378  * jail creation time but cannot be changed in an existing jail.
4379  */
4380 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4381 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4382 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4383 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4384 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4385     "I", "Jail secure level");
4386 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4387     "Jail value for kern.osreldate and uname -K");
4388 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4389     "Jail value for kern.osrelease and uname -r");
4390 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4391     "I", "Jail cannot see all mounted file systems");
4392 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4393     "I", "Ruleset for in-jail devfs mounts");
4394 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4395     "B", "Jail persistence");
4396 #ifdef VIMAGE
4397 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4398     "E,jailsys", "Virtual network stack");
4399 #endif
4400 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4401     "B", "Jail is in the process of shutting down");
4402 
4403 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4404 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4405     "I", "Current number of child jails");
4406 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4407     "I", "Maximum number of child jails");
4408 
4409 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4410 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4411     "Jail hostname");
4412 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4413     "Jail NIS domainname");
4414 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4415     "Jail host UUID");
4416 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4417     "LU", "Jail host ID");
4418 
4419 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4420 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4421 
4422 #ifdef INET
4423 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4424     "Jail IPv4 address virtualization");
4425 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4426     "S,in_addr,a", "Jail IPv4 addresses");
4427 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4428     "B", "Do (not) use IPv4 source address selection rather than the "
4429     "primary jail IPv4 address.");
4430 #endif
4431 #ifdef INET6
4432 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4433     "Jail IPv6 address virtualization");
4434 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4435     "S,in6_addr,a", "Jail IPv6 addresses");
4436 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4437     "B", "Do (not) use IPv6 source address selection rather than the "
4438     "primary jail IPv6 address.");
4439 #endif
4440 
4441 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4442 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4443     "B", "Jail may set hostname");
4444 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4445     "B", "Jail may use SYSV IPC");
4446 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4447     "B", "Jail may create raw sockets");
4448 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4449     "B", "Jail may alter system file flags");
4450 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4451     "B", "Jail may set file quotas");
4452 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4453     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4454 
4455 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4456 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4457     "B", "Jail may mount/unmount jail-friendly file systems in general");
4458 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4459     "B", "Jail may mount the devfs file system");
4460 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4461     "B", "Jail may mount the fdescfs file system");
4462 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4463     "B", "Jail may mount the nullfs file system");
4464 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4465     "B", "Jail may mount the procfs file system");
4466 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4467     "B", "Jail may mount the linprocfs file system");
4468 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4469     "B", "Jail may mount the linsysfs file system");
4470 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4471     "B", "Jail may mount the tmpfs file system");
4472 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4473     "B", "Jail may mount the zfs file system");
4474 
4475 #ifdef RACCT
4476 void
4477 prison_racct_foreach(void (*callback)(struct racct *racct,
4478     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4479     void *arg2, void *arg3)
4480 {
4481 	struct prison_racct *prr;
4482 
4483 	ASSERT_RACCT_ENABLED();
4484 
4485 	sx_slock(&allprison_lock);
4486 	if (pre != NULL)
4487 		(pre)();
4488 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4489 		(callback)(prr->prr_racct, arg2, arg3);
4490 	if (post != NULL)
4491 		(post)();
4492 	sx_sunlock(&allprison_lock);
4493 }
4494 
4495 static struct prison_racct *
4496 prison_racct_find_locked(const char *name)
4497 {
4498 	struct prison_racct *prr;
4499 
4500 	ASSERT_RACCT_ENABLED();
4501 	sx_assert(&allprison_lock, SA_XLOCKED);
4502 
4503 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4504 		return (NULL);
4505 
4506 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4507 		if (strcmp(name, prr->prr_name) != 0)
4508 			continue;
4509 
4510 		/* Found prison_racct with a matching name? */
4511 		prison_racct_hold(prr);
4512 		return (prr);
4513 	}
4514 
4515 	/* Add new prison_racct. */
4516 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4517 	racct_create(&prr->prr_racct);
4518 
4519 	strcpy(prr->prr_name, name);
4520 	refcount_init(&prr->prr_refcount, 1);
4521 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4522 
4523 	return (prr);
4524 }
4525 
4526 struct prison_racct *
4527 prison_racct_find(const char *name)
4528 {
4529 	struct prison_racct *prr;
4530 
4531 	ASSERT_RACCT_ENABLED();
4532 
4533 	sx_xlock(&allprison_lock);
4534 	prr = prison_racct_find_locked(name);
4535 	sx_xunlock(&allprison_lock);
4536 	return (prr);
4537 }
4538 
4539 void
4540 prison_racct_hold(struct prison_racct *prr)
4541 {
4542 
4543 	ASSERT_RACCT_ENABLED();
4544 
4545 	refcount_acquire(&prr->prr_refcount);
4546 }
4547 
4548 static void
4549 prison_racct_free_locked(struct prison_racct *prr)
4550 {
4551 
4552 	ASSERT_RACCT_ENABLED();
4553 	sx_assert(&allprison_lock, SA_XLOCKED);
4554 
4555 	if (refcount_release(&prr->prr_refcount)) {
4556 		racct_destroy(&prr->prr_racct);
4557 		LIST_REMOVE(prr, prr_next);
4558 		free(prr, M_PRISON_RACCT);
4559 	}
4560 }
4561 
4562 void
4563 prison_racct_free(struct prison_racct *prr)
4564 {
4565 	int old;
4566 
4567 	ASSERT_RACCT_ENABLED();
4568 	sx_assert(&allprison_lock, SA_UNLOCKED);
4569 
4570 	old = prr->prr_refcount;
4571 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4572 		return;
4573 
4574 	sx_xlock(&allprison_lock);
4575 	prison_racct_free_locked(prr);
4576 	sx_xunlock(&allprison_lock);
4577 }
4578 
4579 static void
4580 prison_racct_attach(struct prison *pr)
4581 {
4582 	struct prison_racct *prr;
4583 
4584 	ASSERT_RACCT_ENABLED();
4585 	sx_assert(&allprison_lock, SA_XLOCKED);
4586 
4587 	prr = prison_racct_find_locked(pr->pr_name);
4588 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4589 
4590 	pr->pr_prison_racct = prr;
4591 }
4592 
4593 /*
4594  * Handle jail renaming.  From the racct point of view, renaming means
4595  * moving from one prison_racct to another.
4596  */
4597 static void
4598 prison_racct_modify(struct prison *pr)
4599 {
4600 	struct proc *p;
4601 	struct ucred *cred;
4602 	struct prison_racct *oldprr;
4603 
4604 	ASSERT_RACCT_ENABLED();
4605 
4606 	sx_slock(&allproc_lock);
4607 	sx_xlock(&allprison_lock);
4608 
4609 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4610 		sx_xunlock(&allprison_lock);
4611 		sx_sunlock(&allproc_lock);
4612 		return;
4613 	}
4614 
4615 	oldprr = pr->pr_prison_racct;
4616 	pr->pr_prison_racct = NULL;
4617 
4618 	prison_racct_attach(pr);
4619 
4620 	/*
4621 	 * Move resource utilisation records.
4622 	 */
4623 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4624 
4625 	/*
4626 	 * Force rctl to reattach rules to processes.
4627 	 */
4628 	FOREACH_PROC_IN_SYSTEM(p) {
4629 		PROC_LOCK(p);
4630 		cred = crhold(p->p_ucred);
4631 		PROC_UNLOCK(p);
4632 		racct_proc_ucred_changed(p, cred, cred);
4633 		crfree(cred);
4634 	}
4635 
4636 	sx_sunlock(&allproc_lock);
4637 	prison_racct_free_locked(oldprr);
4638 	sx_xunlock(&allprison_lock);
4639 }
4640 
4641 static void
4642 prison_racct_detach(struct prison *pr)
4643 {
4644 
4645 	ASSERT_RACCT_ENABLED();
4646 	sx_assert(&allprison_lock, SA_UNLOCKED);
4647 
4648 	if (pr->pr_prison_racct == NULL)
4649 		return;
4650 	prison_racct_free(pr->pr_prison_racct);
4651 	pr->pr_prison_racct = NULL;
4652 }
4653 #endif /* RACCT */
4654 
4655 #ifdef DDB
4656 
4657 static void
4658 db_show_prison(struct prison *pr)
4659 {
4660 	int fi;
4661 #if defined(INET) || defined(INET6)
4662 	int ii;
4663 #endif
4664 	unsigned jsf;
4665 #ifdef INET6
4666 	char ip6buf[INET6_ADDRSTRLEN];
4667 #endif
4668 
4669 	db_printf("prison %p:\n", pr);
4670 	db_printf(" jid             = %d\n", pr->pr_id);
4671 	db_printf(" name            = %s\n", pr->pr_name);
4672 	db_printf(" parent          = %p\n", pr->pr_parent);
4673 	db_printf(" ref             = %d\n", pr->pr_ref);
4674 	db_printf(" uref            = %d\n", pr->pr_uref);
4675 	db_printf(" path            = %s\n", pr->pr_path);
4676 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4677 	    ? pr->pr_cpuset->cs_id : -1);
4678 #ifdef VIMAGE
4679 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4680 #endif
4681 	db_printf(" root            = %p\n", pr->pr_root);
4682 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4683 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4684 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4685 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4686 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4687 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4688 	db_printf(" flags           = 0x%x", pr->pr_flags);
4689 	for (fi = 0; fi < nitems(pr_flag_names); fi++)
4690 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4691 			db_printf(" %s", pr_flag_names[fi]);
4692 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4693 		jsf = pr->pr_flags &
4694 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4695 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4696 		    pr_flag_jailsys[fi].disable &&
4697 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4698 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4699 		    : "inherit");
4700 	}
4701 	db_printf(" allow           = 0x%x", pr->pr_allow);
4702 	for (fi = 0; fi < nitems(pr_allow_names); fi++)
4703 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4704 			db_printf(" %s", pr_allow_names[fi]);
4705 	db_printf("\n");
4706 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4707 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4708 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4709 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4710 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4711 #ifdef INET
4712 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4713 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4714 		db_printf(" %s %s\n",
4715 		    ii == 0 ? "ip4.addr        =" : "                 ",
4716 		    inet_ntoa(pr->pr_ip4[ii]));
4717 #endif
4718 #ifdef INET6
4719 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4720 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4721 		db_printf(" %s %s\n",
4722 		    ii == 0 ? "ip6.addr        =" : "                 ",
4723 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4724 #endif
4725 }
4726 
4727 DB_SHOW_COMMAND(prison, db_show_prison_command)
4728 {
4729 	struct prison *pr;
4730 
4731 	if (!have_addr) {
4732 		/*
4733 		 * Show all prisons in the list, and prison0 which is not
4734 		 * listed.
4735 		 */
4736 		db_show_prison(&prison0);
4737 		if (!db_pager_quit) {
4738 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4739 				db_show_prison(pr);
4740 				if (db_pager_quit)
4741 					break;
4742 			}
4743 		}
4744 		return;
4745 	}
4746 
4747 	if (addr == 0)
4748 		pr = &prison0;
4749 	else {
4750 		/* Look for a prison with the ID and with references. */
4751 		TAILQ_FOREACH(pr, &allprison, pr_list)
4752 			if (pr->pr_id == addr && pr->pr_ref > 0)
4753 				break;
4754 		if (pr == NULL)
4755 			/* Look again, without requiring a reference. */
4756 			TAILQ_FOREACH(pr, &allprison, pr_list)
4757 				if (pr->pr_id == addr)
4758 					break;
4759 		if (pr == NULL)
4760 			/* Assume address points to a valid prison. */
4761 			pr = (struct prison *)addr;
4762 	}
4763 	db_show_prison(pr);
4764 }
4765 
4766 #endif /* DDB */
4767