xref: /freebsd/sys/kern/kern_jail.c (revision 30b72b6871140f0b29c64d41fc85c4c1d4d4b3f4)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif /* DDB */
72 
73 #include <security/mac/mac_framework.h>
74 
75 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
76 
77 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
78 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
79 
80 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
81 #ifdef INET
82 #ifdef INET6
83 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
84 #else
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
86 #endif
87 #else /* !INET */
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	0
92 #endif
93 #endif
94 
95 /* prison0 describes what is "real" about the system. */
96 struct prison prison0 = {
97 	.pr_id		= 0,
98 	.pr_name	= "0",
99 	.pr_ref		= 1,
100 	.pr_uref	= 1,
101 	.pr_path	= "/",
102 	.pr_securelevel	= -1,
103 	.pr_devfs_rsnum = 0,
104 	.pr_childmax	= JAIL_MAX,
105 	.pr_hostuuid	= DEFAULT_HOSTUUID,
106 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
107 #ifdef VIMAGE
108 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
109 #else
110 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
111 #endif
112 	.pr_allow	= PR_ALLOW_ALL,
113 };
114 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
115 
116 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
117 struct	sx allprison_lock;
118 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
119 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
120 LIST_HEAD(, prison_racct) allprison_racct;
121 int	lastprid = 0;
122 
123 static int do_jail_attach(struct thread *td, struct prison *pr);
124 static void prison_complete(void *context, int pending);
125 static void prison_deref(struct prison *pr, int flags);
126 static char *prison_path(struct prison *pr1, struct prison *pr2);
127 static void prison_remove_one(struct prison *pr);
128 #ifdef RACCT
129 static void prison_racct_attach(struct prison *pr);
130 static void prison_racct_modify(struct prison *pr);
131 static void prison_racct_detach(struct prison *pr);
132 #endif
133 #ifdef INET
134 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
135 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
136 #endif
137 #ifdef INET6
138 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
139 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
140 #endif
141 
142 /* Flags for prison_deref */
143 #define	PD_DEREF	0x01
144 #define	PD_DEUREF	0x02
145 #define	PD_LOCKED	0x04
146 #define	PD_LIST_SLOCKED	0x08
147 #define	PD_LIST_XLOCKED	0x10
148 
149 /*
150  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
151  * as we cannot figure out the size of a sparse array, or an array without a
152  * terminating entry.
153  */
154 static char *pr_flag_names[] = {
155 	[0] = "persist",
156 #ifdef INET
157 	[7] = "ip4.saddrsel",
158 #endif
159 #ifdef INET6
160 	[8] = "ip6.saddrsel",
161 #endif
162 };
163 const size_t pr_flag_names_size = sizeof(pr_flag_names);
164 
165 static char *pr_flag_nonames[] = {
166 	[0] = "nopersist",
167 #ifdef INET
168 	[7] = "ip4.nosaddrsel",
169 #endif
170 #ifdef INET6
171 	[8] = "ip6.nosaddrsel",
172 #endif
173 };
174 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
175 
176 struct jailsys_flags {
177 	const char	*name;
178 	unsigned	 disable;
179 	unsigned	 new;
180 } pr_flag_jailsys[] = {
181 	{ "host", 0, PR_HOST },
182 #ifdef VIMAGE
183 	{ "vnet", 0, PR_VNET },
184 #endif
185 #ifdef INET
186 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
187 #endif
188 #ifdef INET6
189 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
190 #endif
191 };
192 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
193 
194 static char *pr_allow_names[] = {
195 	"allow.set_hostname",
196 	"allow.sysvipc",
197 	"allow.raw_sockets",
198 	"allow.chflags",
199 	"allow.mount",
200 	"allow.quotas",
201 	"allow.socket_af",
202 	"allow.mount.devfs",
203 	"allow.mount.nullfs",
204 	"allow.mount.zfs",
205 	"allow.mount.procfs",
206 	"allow.mount.tmpfs",
207 	"allow.mount.fdescfs",
208 	"allow.mount.linprocfs",
209 	"allow.mount.linsysfs",
210 };
211 const size_t pr_allow_names_size = sizeof(pr_allow_names);
212 
213 static char *pr_allow_nonames[] = {
214 	"allow.noset_hostname",
215 	"allow.nosysvipc",
216 	"allow.noraw_sockets",
217 	"allow.nochflags",
218 	"allow.nomount",
219 	"allow.noquotas",
220 	"allow.nosocket_af",
221 	"allow.mount.nodevfs",
222 	"allow.mount.nonullfs",
223 	"allow.mount.nozfs",
224 	"allow.mount.noprocfs",
225 	"allow.mount.notmpfs",
226 	"allow.mount.nofdescfs",
227 	"allow.mount.nolinprocfs",
228 	"allow.mount.nolinsysfs",
229 };
230 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
231 
232 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
233 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
234 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
235 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
236 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
237 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
238 #if defined(INET) || defined(INET6)
239 static unsigned jail_max_af_ips = 255;
240 #endif
241 
242 /*
243  * Initialize the parts of prison0 that can't be static-initialized with
244  * constants.  This is called from proc0_init() after creating thread0 cpuset.
245  */
246 void
247 prison0_init(void)
248 {
249 
250 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
251 	prison0.pr_osreldate = osreldate;
252 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
253 }
254 
255 #ifdef INET
256 static int
257 qcmp_v4(const void *ip1, const void *ip2)
258 {
259 	in_addr_t iaa, iab;
260 
261 	/*
262 	 * We need to compare in HBO here to get the list sorted as expected
263 	 * by the result of the code.  Sorting NBO addresses gives you
264 	 * interesting results.  If you do not understand, do not try.
265 	 */
266 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
267 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
268 
269 	/*
270 	 * Do not simply return the difference of the two numbers, the int is
271 	 * not wide enough.
272 	 */
273 	if (iaa > iab)
274 		return (1);
275 	else if (iaa < iab)
276 		return (-1);
277 	else
278 		return (0);
279 }
280 #endif
281 
282 #ifdef INET6
283 static int
284 qcmp_v6(const void *ip1, const void *ip2)
285 {
286 	const struct in6_addr *ia6a, *ia6b;
287 	int i, rc;
288 
289 	ia6a = (const struct in6_addr *)ip1;
290 	ia6b = (const struct in6_addr *)ip2;
291 
292 	rc = 0;
293 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
294 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
295 			rc = 1;
296 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
297 			rc = -1;
298 	}
299 	return (rc);
300 }
301 #endif
302 
303 /*
304  * struct jail_args {
305  *	struct jail *jail;
306  * };
307  */
308 int
309 sys_jail(struct thread *td, struct jail_args *uap)
310 {
311 	uint32_t version;
312 	int error;
313 	struct jail j;
314 
315 	error = copyin(uap->jail, &version, sizeof(uint32_t));
316 	if (error)
317 		return (error);
318 
319 	switch (version) {
320 	case 0:
321 	{
322 		struct jail_v0 j0;
323 
324 		/* FreeBSD single IPv4 jails. */
325 		bzero(&j, sizeof(struct jail));
326 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
327 		if (error)
328 			return (error);
329 		j.version = j0.version;
330 		j.path = j0.path;
331 		j.hostname = j0.hostname;
332 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
333 		break;
334 	}
335 
336 	case 1:
337 		/*
338 		 * Version 1 was used by multi-IPv4 jail implementations
339 		 * that never made it into the official kernel.
340 		 */
341 		return (EINVAL);
342 
343 	case 2:	/* JAIL_API_VERSION */
344 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
345 		error = copyin(uap->jail, &j, sizeof(struct jail));
346 		if (error)
347 			return (error);
348 		break;
349 
350 	default:
351 		/* Sci-Fi jails are not supported, sorry. */
352 		return (EINVAL);
353 	}
354 	return (kern_jail(td, &j));
355 }
356 
357 int
358 kern_jail(struct thread *td, struct jail *j)
359 {
360 	struct iovec optiov[2 * (4 + nitems(pr_allow_names)
361 #ifdef INET
362 			    + 1
363 #endif
364 #ifdef INET6
365 			    + 1
366 #endif
367 			    )];
368 	struct uio opt;
369 	char *u_path, *u_hostname, *u_name;
370 #ifdef INET
371 	uint32_t ip4s;
372 	struct in_addr *u_ip4;
373 #endif
374 #ifdef INET6
375 	struct in6_addr *u_ip6;
376 #endif
377 	size_t tmplen;
378 	int error, enforce_statfs, fi;
379 
380 	bzero(&optiov, sizeof(optiov));
381 	opt.uio_iov = optiov;
382 	opt.uio_iovcnt = 0;
383 	opt.uio_offset = -1;
384 	opt.uio_resid = -1;
385 	opt.uio_segflg = UIO_SYSSPACE;
386 	opt.uio_rw = UIO_READ;
387 	opt.uio_td = td;
388 
389 	/* Set permissions for top-level jails from sysctls. */
390 	if (!jailed(td->td_ucred)) {
391 		for (fi = 0; fi < nitems(pr_allow_names); fi++) {
392 			optiov[opt.uio_iovcnt].iov_base =
393 			    (jail_default_allow & (1 << fi))
394 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
395 			optiov[opt.uio_iovcnt].iov_len =
396 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
397 			opt.uio_iovcnt += 2;
398 		}
399 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
400 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
401 		opt.uio_iovcnt++;
402 		enforce_statfs = jail_default_enforce_statfs;
403 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
404 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
405 		opt.uio_iovcnt++;
406 	}
407 
408 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
409 #ifdef INET
410 	ip4s = (j->version == 0) ? 1 : j->ip4s;
411 	if (ip4s > jail_max_af_ips)
412 		return (EINVAL);
413 	tmplen += ip4s * sizeof(struct in_addr);
414 #else
415 	if (j->ip4s > 0)
416 		return (EINVAL);
417 #endif
418 #ifdef INET6
419 	if (j->ip6s > jail_max_af_ips)
420 		return (EINVAL);
421 	tmplen += j->ip6s * sizeof(struct in6_addr);
422 #else
423 	if (j->ip6s > 0)
424 		return (EINVAL);
425 #endif
426 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
427 	u_hostname = u_path + MAXPATHLEN;
428 	u_name = u_hostname + MAXHOSTNAMELEN;
429 #ifdef INET
430 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
431 #endif
432 #ifdef INET6
433 #ifdef INET
434 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
435 #else
436 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
437 #endif
438 #endif
439 	optiov[opt.uio_iovcnt].iov_base = "path";
440 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
441 	opt.uio_iovcnt++;
442 	optiov[opt.uio_iovcnt].iov_base = u_path;
443 	error = copyinstr(j->path, u_path, MAXPATHLEN,
444 	    &optiov[opt.uio_iovcnt].iov_len);
445 	if (error) {
446 		free(u_path, M_TEMP);
447 		return (error);
448 	}
449 	opt.uio_iovcnt++;
450 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
451 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
452 	opt.uio_iovcnt++;
453 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
454 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
455 	    &optiov[opt.uio_iovcnt].iov_len);
456 	if (error) {
457 		free(u_path, M_TEMP);
458 		return (error);
459 	}
460 	opt.uio_iovcnt++;
461 	if (j->jailname != NULL) {
462 		optiov[opt.uio_iovcnt].iov_base = "name";
463 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
464 		opt.uio_iovcnt++;
465 		optiov[opt.uio_iovcnt].iov_base = u_name;
466 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
467 		    &optiov[opt.uio_iovcnt].iov_len);
468 		if (error) {
469 			free(u_path, M_TEMP);
470 			return (error);
471 		}
472 		opt.uio_iovcnt++;
473 	}
474 #ifdef INET
475 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
476 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
477 	opt.uio_iovcnt++;
478 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
479 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
480 	if (j->version == 0)
481 		u_ip4->s_addr = j->ip4s;
482 	else {
483 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
484 		if (error) {
485 			free(u_path, M_TEMP);
486 			return (error);
487 		}
488 	}
489 	opt.uio_iovcnt++;
490 #endif
491 #ifdef INET6
492 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
493 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
494 	opt.uio_iovcnt++;
495 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
496 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
497 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
498 	if (error) {
499 		free(u_path, M_TEMP);
500 		return (error);
501 	}
502 	opt.uio_iovcnt++;
503 #endif
504 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
505 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
506 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
507 	free(u_path, M_TEMP);
508 	return (error);
509 }
510 
511 
512 /*
513  * struct jail_set_args {
514  *	struct iovec *iovp;
515  *	unsigned int iovcnt;
516  *	int flags;
517  * };
518  */
519 int
520 sys_jail_set(struct thread *td, struct jail_set_args *uap)
521 {
522 	struct uio *auio;
523 	int error;
524 
525 	/* Check that we have an even number of iovecs. */
526 	if (uap->iovcnt & 1)
527 		return (EINVAL);
528 
529 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
530 	if (error)
531 		return (error);
532 	error = kern_jail_set(td, auio, uap->flags);
533 	free(auio, M_IOV);
534 	return (error);
535 }
536 
537 int
538 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
539 {
540 	struct nameidata nd;
541 #ifdef INET
542 	struct in_addr *ip4;
543 #endif
544 #ifdef INET6
545 	struct in6_addr *ip6;
546 #endif
547 	struct vfsopt *opt;
548 	struct vfsoptlist *opts;
549 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
550 	struct vnode *root;
551 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
552 	char *g_path, *osrelstr;
553 #if defined(INET) || defined(INET6)
554 	struct prison *tppr;
555 	void *op;
556 #endif
557 	unsigned long hid;
558 	size_t namelen, onamelen, pnamelen;
559 	int born, created, cuflags, descend, enforce;
560 	int error, errmsg_len, errmsg_pos;
561 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
562 	int fi, jid, jsys, len, level;
563 	int childmax, osreldt, rsnum, slevel;
564 	int fullpath_disabled;
565 #if defined(INET) || defined(INET6)
566 	int ii, ij;
567 #endif
568 #ifdef INET
569 	int ip4s, redo_ip4;
570 #endif
571 #ifdef INET6
572 	int ip6s, redo_ip6;
573 #endif
574 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
575 	unsigned tallow;
576 	char numbuf[12];
577 
578 	error = priv_check(td, PRIV_JAIL_SET);
579 	if (!error && (flags & JAIL_ATTACH))
580 		error = priv_check(td, PRIV_JAIL_ATTACH);
581 	if (error)
582 		return (error);
583 	mypr = td->td_ucred->cr_prison;
584 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
585 		return (EPERM);
586 	if (flags & ~JAIL_SET_MASK)
587 		return (EINVAL);
588 
589 	/*
590 	 * Check all the parameters before committing to anything.  Not all
591 	 * errors can be caught early, but we may as well try.  Also, this
592 	 * takes care of some expensive stuff (path lookup) before getting
593 	 * the allprison lock.
594 	 *
595 	 * XXX Jails are not filesystems, and jail parameters are not mount
596 	 *     options.  But it makes more sense to re-use the vfsopt code
597 	 *     than duplicate it under a different name.
598 	 */
599 	error = vfs_buildopts(optuio, &opts);
600 	if (error)
601 		return (error);
602 #ifdef INET
603 	ip4 = NULL;
604 #endif
605 #ifdef INET6
606 	ip6 = NULL;
607 #endif
608 	g_path = NULL;
609 
610 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
611 	if (!cuflags) {
612 		error = EINVAL;
613 		vfs_opterror(opts, "no valid operation (create or update)");
614 		goto done_errmsg;
615 	}
616 
617 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
618 	if (error == ENOENT)
619 		jid = 0;
620 	else if (error != 0)
621 		goto done_free;
622 
623 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
624 	if (error == ENOENT)
625 		gotslevel = 0;
626 	else if (error != 0)
627 		goto done_free;
628 	else
629 		gotslevel = 1;
630 
631 	error =
632 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
633 	if (error == ENOENT)
634 		gotchildmax = 0;
635 	else if (error != 0)
636 		goto done_free;
637 	else
638 		gotchildmax = 1;
639 
640 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
641 	if (error == ENOENT)
642 		gotenforce = 0;
643 	else if (error != 0)
644 		goto done_free;
645 	else if (enforce < 0 || enforce > 2) {
646 		error = EINVAL;
647 		goto done_free;
648 	} else
649 		gotenforce = 1;
650 
651 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
652 	if (error == ENOENT)
653 		gotrsnum = 0;
654 	else if (error != 0)
655 		goto done_free;
656 	else
657 		gotrsnum = 1;
658 
659 	pr_flags = ch_flags = 0;
660 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
661 		if (pr_flag_names[fi] == NULL)
662 			continue;
663 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
664 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
665 	}
666 	ch_flags |= pr_flags;
667 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
668 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
669 		    sizeof(jsys));
670 		if (error == ENOENT)
671 			continue;
672 		if (error != 0)
673 			goto done_free;
674 		switch (jsys) {
675 		case JAIL_SYS_DISABLE:
676 			if (!pr_flag_jailsys[fi].disable) {
677 				error = EINVAL;
678 				goto done_free;
679 			}
680 			pr_flags |= pr_flag_jailsys[fi].disable;
681 			break;
682 		case JAIL_SYS_NEW:
683 			pr_flags |= pr_flag_jailsys[fi].new;
684 			break;
685 		case JAIL_SYS_INHERIT:
686 			break;
687 		default:
688 			error = EINVAL;
689 			goto done_free;
690 		}
691 		ch_flags |=
692 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
693 	}
694 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
695 	    && !(pr_flags & PR_PERSIST)) {
696 		error = EINVAL;
697 		vfs_opterror(opts, "new jail must persist or attach");
698 		goto done_errmsg;
699 	}
700 #ifdef VIMAGE
701 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
702 		error = EINVAL;
703 		vfs_opterror(opts, "vnet cannot be changed after creation");
704 		goto done_errmsg;
705 	}
706 #endif
707 #ifdef INET
708 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
709 		error = EINVAL;
710 		vfs_opterror(opts, "ip4 cannot be changed after creation");
711 		goto done_errmsg;
712 	}
713 #endif
714 #ifdef INET6
715 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
716 		error = EINVAL;
717 		vfs_opterror(opts, "ip6 cannot be changed after creation");
718 		goto done_errmsg;
719 	}
720 #endif
721 
722 	pr_allow = ch_allow = 0;
723 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
724 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
725 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
726 	}
727 	ch_allow |= pr_allow;
728 
729 	error = vfs_getopt(opts, "name", (void **)&name, &len);
730 	if (error == ENOENT)
731 		name = NULL;
732 	else if (error != 0)
733 		goto done_free;
734 	else {
735 		if (len == 0 || name[len - 1] != '\0') {
736 			error = EINVAL;
737 			goto done_free;
738 		}
739 		if (len > MAXHOSTNAMELEN) {
740 			error = ENAMETOOLONG;
741 			goto done_free;
742 		}
743 	}
744 
745 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
746 	if (error == ENOENT)
747 		host = NULL;
748 	else if (error != 0)
749 		goto done_free;
750 	else {
751 		ch_flags |= PR_HOST;
752 		pr_flags |= PR_HOST;
753 		if (len == 0 || host[len - 1] != '\0') {
754 			error = EINVAL;
755 			goto done_free;
756 		}
757 		if (len > MAXHOSTNAMELEN) {
758 			error = ENAMETOOLONG;
759 			goto done_free;
760 		}
761 	}
762 
763 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
764 	if (error == ENOENT)
765 		domain = NULL;
766 	else if (error != 0)
767 		goto done_free;
768 	else {
769 		ch_flags |= PR_HOST;
770 		pr_flags |= PR_HOST;
771 		if (len == 0 || domain[len - 1] != '\0') {
772 			error = EINVAL;
773 			goto done_free;
774 		}
775 		if (len > MAXHOSTNAMELEN) {
776 			error = ENAMETOOLONG;
777 			goto done_free;
778 		}
779 	}
780 
781 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
782 	if (error == ENOENT)
783 		uuid = NULL;
784 	else if (error != 0)
785 		goto done_free;
786 	else {
787 		ch_flags |= PR_HOST;
788 		pr_flags |= PR_HOST;
789 		if (len == 0 || uuid[len - 1] != '\0') {
790 			error = EINVAL;
791 			goto done_free;
792 		}
793 		if (len > HOSTUUIDLEN) {
794 			error = ENAMETOOLONG;
795 			goto done_free;
796 		}
797 	}
798 
799 #ifdef COMPAT_FREEBSD32
800 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
801 		uint32_t hid32;
802 
803 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
804 		hid = hid32;
805 	} else
806 #endif
807 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
808 	if (error == ENOENT)
809 		gothid = 0;
810 	else if (error != 0)
811 		goto done_free;
812 	else {
813 		gothid = 1;
814 		ch_flags |= PR_HOST;
815 		pr_flags |= PR_HOST;
816 	}
817 
818 #ifdef INET
819 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
820 	if (error == ENOENT)
821 		ip4s = 0;
822 	else if (error != 0)
823 		goto done_free;
824 	else if (ip4s & (sizeof(*ip4) - 1)) {
825 		error = EINVAL;
826 		goto done_free;
827 	} else {
828 		ch_flags |= PR_IP4_USER;
829 		pr_flags |= PR_IP4_USER;
830 		if (ip4s > 0) {
831 			ip4s /= sizeof(*ip4);
832 			if (ip4s > jail_max_af_ips) {
833 				error = EINVAL;
834 				vfs_opterror(opts, "too many IPv4 addresses");
835 				goto done_errmsg;
836 			}
837 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
838 			bcopy(op, ip4, ip4s * sizeof(*ip4));
839 			/*
840 			 * IP addresses are all sorted but ip[0] to preserve
841 			 * the primary IP address as given from userland.
842 			 * This special IP is used for unbound outgoing
843 			 * connections as well for "loopback" traffic in case
844 			 * source address selection cannot find any more fitting
845 			 * address to connect from.
846 			 */
847 			if (ip4s > 1)
848 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
849 			/*
850 			 * Check for duplicate addresses and do some simple
851 			 * zero and broadcast checks. If users give other bogus
852 			 * addresses it is their problem.
853 			 *
854 			 * We do not have to care about byte order for these
855 			 * checks so we will do them in NBO.
856 			 */
857 			for (ii = 0; ii < ip4s; ii++) {
858 				if (ip4[ii].s_addr == INADDR_ANY ||
859 				    ip4[ii].s_addr == INADDR_BROADCAST) {
860 					error = EINVAL;
861 					goto done_free;
862 				}
863 				if ((ii+1) < ip4s &&
864 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
865 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
866 					error = EINVAL;
867 					goto done_free;
868 				}
869 			}
870 		}
871 	}
872 #endif
873 
874 #ifdef INET6
875 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
876 	if (error == ENOENT)
877 		ip6s = 0;
878 	else if (error != 0)
879 		goto done_free;
880 	else if (ip6s & (sizeof(*ip6) - 1)) {
881 		error = EINVAL;
882 		goto done_free;
883 	} else {
884 		ch_flags |= PR_IP6_USER;
885 		pr_flags |= PR_IP6_USER;
886 		if (ip6s > 0) {
887 			ip6s /= sizeof(*ip6);
888 			if (ip6s > jail_max_af_ips) {
889 				error = EINVAL;
890 				vfs_opterror(opts, "too many IPv6 addresses");
891 				goto done_errmsg;
892 			}
893 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
894 			bcopy(op, ip6, ip6s * sizeof(*ip6));
895 			if (ip6s > 1)
896 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
897 			for (ii = 0; ii < ip6s; ii++) {
898 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
899 					error = EINVAL;
900 					goto done_free;
901 				}
902 				if ((ii+1) < ip6s &&
903 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
904 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
905 				{
906 					error = EINVAL;
907 					goto done_free;
908 				}
909 			}
910 		}
911 	}
912 #endif
913 
914 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
915 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
916 		error = EINVAL;
917 		vfs_opterror(opts,
918 		    "vnet jails cannot have IP address restrictions");
919 		goto done_errmsg;
920 	}
921 #endif
922 
923 	fullpath_disabled = 0;
924 	root = NULL;
925 	error = vfs_getopt(opts, "path", (void **)&path, &len);
926 	if (error == ENOENT)
927 		path = NULL;
928 	else if (error != 0)
929 		goto done_free;
930 	else {
931 		if (flags & JAIL_UPDATE) {
932 			error = EINVAL;
933 			vfs_opterror(opts,
934 			    "path cannot be changed after creation");
935 			goto done_errmsg;
936 		}
937 		if (len == 0 || path[len - 1] != '\0') {
938 			error = EINVAL;
939 			goto done_free;
940 		}
941 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
942 		    path, td);
943 		error = namei(&nd);
944 		if (error)
945 			goto done_free;
946 		root = nd.ni_vp;
947 		NDFREE(&nd, NDF_ONLY_PNBUF);
948 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
949 		strlcpy(g_path, path, MAXPATHLEN);
950 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
951 		if (error == 0)
952 			path = g_path;
953 		else if (error == ENODEV) {
954 			/* proceed if sysctl debug.disablefullpath == 1 */
955 			fullpath_disabled = 1;
956 			if (len < 2 || (len == 2 && path[0] == '/'))
957 				path = NULL;
958 		} else {
959 			/* exit on other errors */
960 			goto done_free;
961 		}
962 		if (root->v_type != VDIR) {
963 			error = ENOTDIR;
964 			vput(root);
965 			goto done_free;
966 		}
967 		VOP_UNLOCK(root, 0);
968 		if (fullpath_disabled) {
969 			/* Leave room for a real-root full pathname. */
970 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
971 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
972 				error = ENAMETOOLONG;
973 				goto done_free;
974 			}
975 		}
976 	}
977 
978 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
979 	if (error == ENOENT)
980 		osrelstr = NULL;
981 	else if (error != 0)
982 		goto done_free;
983 	else {
984 		if (flags & JAIL_UPDATE) {
985 			error = EINVAL;
986 			vfs_opterror(opts,
987 			    "osrelease cannot be changed after creation");
988 			goto done_errmsg;
989 		}
990 		if (len == 0 || len >= OSRELEASELEN) {
991 			error = EINVAL;
992 			vfs_opterror(opts,
993 			    "osrelease string must be 1-%d bytes long",
994 			    OSRELEASELEN - 1);
995 			goto done_errmsg;
996 		}
997 	}
998 
999 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1000 	if (error == ENOENT)
1001 		osreldt = 0;
1002 	else if (error != 0)
1003 		goto done_free;
1004 	else {
1005 		if (flags & JAIL_UPDATE) {
1006 			error = EINVAL;
1007 			vfs_opterror(opts,
1008 			    "osreldate cannot be changed after creation");
1009 			goto done_errmsg;
1010 		}
1011 		if (osreldt == 0) {
1012 			error = EINVAL;
1013 			vfs_opterror(opts, "osreldate cannot be 0");
1014 			goto done_errmsg;
1015 		}
1016 	}
1017 
1018 	/*
1019 	 * Find the specified jail, or at least its parent.
1020 	 * This abuses the file error codes ENOENT and EEXIST.
1021 	 */
1022 	pr = NULL;
1023 	ppr = mypr;
1024 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1025 		namelc = strrchr(name, '.');
1026 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1027 		if (*p != '\0')
1028 			jid = 0;
1029 	}
1030 	sx_xlock(&allprison_lock);
1031 	if (jid != 0) {
1032 		/*
1033 		 * See if a requested jid already exists.  There is an
1034 		 * information leak here if the jid exists but is not within
1035 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1036 		 * even though they cannot see the jail, and CREATE | UPDATE
1037 		 * will return ENOENT which is not normally a valid error.
1038 		 */
1039 		if (jid < 0) {
1040 			error = EINVAL;
1041 			vfs_opterror(opts, "negative jid");
1042 			goto done_unlock_list;
1043 		}
1044 		pr = prison_find(jid);
1045 		if (pr != NULL) {
1046 			ppr = pr->pr_parent;
1047 			/* Create: jid must not exist. */
1048 			if (cuflags == JAIL_CREATE) {
1049 				mtx_unlock(&pr->pr_mtx);
1050 				error = EEXIST;
1051 				vfs_opterror(opts, "jail %d already exists",
1052 				    jid);
1053 				goto done_unlock_list;
1054 			}
1055 			if (!prison_ischild(mypr, pr)) {
1056 				mtx_unlock(&pr->pr_mtx);
1057 				pr = NULL;
1058 			} else if (pr->pr_uref == 0) {
1059 				if (!(flags & JAIL_DYING)) {
1060 					mtx_unlock(&pr->pr_mtx);
1061 					error = ENOENT;
1062 					vfs_opterror(opts, "jail %d is dying",
1063 					    jid);
1064 					goto done_unlock_list;
1065 				} else if ((flags & JAIL_ATTACH) ||
1066 				    (pr_flags & PR_PERSIST)) {
1067 					/*
1068 					 * A dying jail might be resurrected
1069 					 * (via attach or persist), but first
1070 					 * it must determine if another jail
1071 					 * has claimed its name.  Accomplish
1072 					 * this by implicitly re-setting the
1073 					 * name.
1074 					 */
1075 					if (name == NULL)
1076 						name = prison_name(mypr, pr);
1077 				}
1078 			}
1079 		}
1080 		if (pr == NULL) {
1081 			/* Update: jid must exist. */
1082 			if (cuflags == JAIL_UPDATE) {
1083 				error = ENOENT;
1084 				vfs_opterror(opts, "jail %d not found", jid);
1085 				goto done_unlock_list;
1086 			}
1087 		}
1088 	}
1089 	/*
1090 	 * If the caller provided a name, look for a jail by that name.
1091 	 * This has different semantics for creates and updates keyed by jid
1092 	 * (where the name must not already exist in a different jail),
1093 	 * and updates keyed by the name itself (where the name must exist
1094 	 * because that is the jail being updated).
1095 	 */
1096 	namelc = NULL;
1097 	if (name != NULL) {
1098 		namelc = strrchr(name, '.');
1099 		if (namelc == NULL)
1100 			namelc = name;
1101 		else {
1102 			/*
1103 			 * This is a hierarchical name.  Split it into the
1104 			 * parent and child names, and make sure the parent
1105 			 * exists or matches an already found jail.
1106 			 */
1107 			if (pr != NULL) {
1108 				if (strncmp(name, ppr->pr_name, namelc - name)
1109 				    || ppr->pr_name[namelc - name] != '\0') {
1110 					mtx_unlock(&pr->pr_mtx);
1111 					error = EINVAL;
1112 					vfs_opterror(opts,
1113 					    "cannot change jail's parent");
1114 					goto done_unlock_list;
1115 				}
1116 			} else {
1117 				*namelc = '\0';
1118 				ppr = prison_find_name(mypr, name);
1119 				if (ppr == NULL) {
1120 					error = ENOENT;
1121 					vfs_opterror(opts,
1122 					    "jail \"%s\" not found", name);
1123 					goto done_unlock_list;
1124 				}
1125 				mtx_unlock(&ppr->pr_mtx);
1126 				*namelc = '.';
1127 			}
1128 			namelc++;
1129 		}
1130 		if (namelc[0] != '\0') {
1131 			pnamelen =
1132 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1133  name_again:
1134 			deadpr = NULL;
1135 			FOREACH_PRISON_CHILD(ppr, tpr) {
1136 				if (tpr != pr && tpr->pr_ref > 0 &&
1137 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1138 					if (pr == NULL &&
1139 					    cuflags != JAIL_CREATE) {
1140 						mtx_lock(&tpr->pr_mtx);
1141 						if (tpr->pr_ref > 0) {
1142 							/*
1143 							 * Use this jail
1144 							 * for updates.
1145 							 */
1146 							if (tpr->pr_uref > 0) {
1147 								pr = tpr;
1148 								break;
1149 							}
1150 							deadpr = tpr;
1151 						}
1152 						mtx_unlock(&tpr->pr_mtx);
1153 					} else if (tpr->pr_uref > 0) {
1154 						/*
1155 						 * Create, or update(jid):
1156 						 * name must not exist in an
1157 						 * active sibling jail.
1158 						 */
1159 						error = EEXIST;
1160 						if (pr != NULL)
1161 							mtx_unlock(&pr->pr_mtx);
1162 						vfs_opterror(opts,
1163 						   "jail \"%s\" already exists",
1164 						   name);
1165 						goto done_unlock_list;
1166 					}
1167 				}
1168 			}
1169 			/* If no active jail is found, use a dying one. */
1170 			if (deadpr != NULL && pr == NULL) {
1171 				if (flags & JAIL_DYING) {
1172 					mtx_lock(&deadpr->pr_mtx);
1173 					if (deadpr->pr_ref == 0) {
1174 						mtx_unlock(&deadpr->pr_mtx);
1175 						goto name_again;
1176 					}
1177 					pr = deadpr;
1178 				} else if (cuflags == JAIL_UPDATE) {
1179 					error = ENOENT;
1180 					vfs_opterror(opts,
1181 					    "jail \"%s\" is dying", name);
1182 					goto done_unlock_list;
1183 				}
1184 			}
1185 			/* Update: name must exist if no jid. */
1186 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1187 				error = ENOENT;
1188 				vfs_opterror(opts, "jail \"%s\" not found",
1189 				    name);
1190 				goto done_unlock_list;
1191 			}
1192 		}
1193 	}
1194 	/* Update: must provide a jid or name. */
1195 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1196 		error = ENOENT;
1197 		vfs_opterror(opts, "update specified no jail");
1198 		goto done_unlock_list;
1199 	}
1200 
1201 	/* If there's no prison to update, create a new one and link it in. */
1202 	if (pr == NULL) {
1203 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1204 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1205 				error = EPERM;
1206 				vfs_opterror(opts, "prison limit exceeded");
1207 				goto done_unlock_list;
1208 			}
1209 		created = 1;
1210 		mtx_lock(&ppr->pr_mtx);
1211 		if (ppr->pr_ref == 0) {
1212 			mtx_unlock(&ppr->pr_mtx);
1213 			error = ENOENT;
1214 			vfs_opterror(opts, "jail \"%s\" not found",
1215 			    prison_name(mypr, ppr));
1216 			goto done_unlock_list;
1217 		}
1218 		ppr->pr_ref++;
1219 		ppr->pr_uref++;
1220 		mtx_unlock(&ppr->pr_mtx);
1221 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1222 		if (jid == 0) {
1223 			/* Find the next free jid. */
1224 			jid = lastprid + 1;
1225  findnext:
1226 			if (jid == JAIL_MAX)
1227 				jid = 1;
1228 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1229 				if (tpr->pr_id < jid)
1230 					continue;
1231 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1232 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1233 					break;
1234 				}
1235 				if (jid == lastprid) {
1236 					error = EAGAIN;
1237 					vfs_opterror(opts,
1238 					    "no available jail IDs");
1239 					free(pr, M_PRISON);
1240 					prison_deref(ppr, PD_DEREF |
1241 					    PD_DEUREF | PD_LIST_XLOCKED);
1242 					goto done_releroot;
1243 				}
1244 				jid++;
1245 				goto findnext;
1246 			}
1247 			lastprid = jid;
1248 		} else {
1249 			/*
1250 			 * The jail already has a jid (that did not yet exist),
1251 			 * so just find where to insert it.
1252 			 */
1253 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1254 				if (tpr->pr_id >= jid) {
1255 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1256 					break;
1257 				}
1258 		}
1259 		if (tpr == NULL)
1260 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1261 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1262 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1263 			tpr->pr_childcount++;
1264 
1265 		pr->pr_parent = ppr;
1266 		pr->pr_id = jid;
1267 
1268 		/* Set some default values, and inherit some from the parent. */
1269 		if (namelc == NULL)
1270 			namelc = "";
1271 		if (path == NULL) {
1272 			path = "/";
1273 			root = mypr->pr_root;
1274 			vref(root);
1275 		}
1276 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1277 		pr->pr_flags |= PR_HOST;
1278 #if defined(INET) || defined(INET6)
1279 #ifdef VIMAGE
1280 		if (!(pr_flags & PR_VNET))
1281 #endif
1282 		{
1283 #ifdef INET
1284 			if (!(ch_flags & PR_IP4_USER))
1285 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1286 			else if (!(pr_flags & PR_IP4_USER)) {
1287 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1288 				if (ppr->pr_ip4 != NULL) {
1289 					pr->pr_ip4s = ppr->pr_ip4s;
1290 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1291 					    sizeof(struct in_addr), M_PRISON,
1292 					    M_WAITOK);
1293 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1294 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1295 				}
1296 			}
1297 #endif
1298 #ifdef INET6
1299 			if (!(ch_flags & PR_IP6_USER))
1300 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1301 			else if (!(pr_flags & PR_IP6_USER)) {
1302 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1303 				if (ppr->pr_ip6 != NULL) {
1304 					pr->pr_ip6s = ppr->pr_ip6s;
1305 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1306 					    sizeof(struct in6_addr), M_PRISON,
1307 					    M_WAITOK);
1308 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1309 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1310 				}
1311 			}
1312 #endif
1313 		}
1314 #endif
1315 		/* Source address selection is always on by default. */
1316 		pr->pr_flags |= _PR_IP_SADDRSEL;
1317 
1318 		pr->pr_securelevel = ppr->pr_securelevel;
1319 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1320 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1321 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1322 
1323 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1324 		if (osrelstr == NULL)
1325 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1326 		else
1327 		    strcpy(pr->pr_osrelease, osrelstr);
1328 
1329 		LIST_INIT(&pr->pr_children);
1330 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1331 
1332 #ifdef VIMAGE
1333 		/* Allocate a new vnet if specified. */
1334 		pr->pr_vnet = (pr_flags & PR_VNET)
1335 		    ? vnet_alloc() : ppr->pr_vnet;
1336 #endif
1337 		/*
1338 		 * Allocate a dedicated cpuset for each jail.
1339 		 * Unlike other initial settings, this may return an erorr.
1340 		 */
1341 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1342 		if (error) {
1343 			prison_deref(pr, PD_LIST_XLOCKED);
1344 			goto done_releroot;
1345 		}
1346 
1347 		mtx_lock(&pr->pr_mtx);
1348 		/*
1349 		 * New prisons do not yet have a reference, because we do not
1350 		 * want others to see the incomplete prison once the
1351 		 * allprison_lock is downgraded.
1352 		 */
1353 	} else {
1354 		created = 0;
1355 		/*
1356 		 * Grab a reference for existing prisons, to ensure they
1357 		 * continue to exist for the duration of the call.
1358 		 */
1359 		pr->pr_ref++;
1360 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1361 		if ((pr->pr_flags & PR_VNET) &&
1362 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1363 			error = EINVAL;
1364 			vfs_opterror(opts,
1365 			    "vnet jails cannot have IP address restrictions");
1366 			goto done_deref_locked;
1367 		}
1368 #endif
1369 #ifdef INET
1370 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1371 			error = EINVAL;
1372 			vfs_opterror(opts,
1373 			    "ip4 cannot be changed after creation");
1374 			goto done_deref_locked;
1375 		}
1376 #endif
1377 #ifdef INET6
1378 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1379 			error = EINVAL;
1380 			vfs_opterror(opts,
1381 			    "ip6 cannot be changed after creation");
1382 			goto done_deref_locked;
1383 		}
1384 #endif
1385 	}
1386 
1387 	/* Do final error checking before setting anything. */
1388 	if (gotslevel) {
1389 		if (slevel < ppr->pr_securelevel) {
1390 			error = EPERM;
1391 			goto done_deref_locked;
1392 		}
1393 	}
1394 	if (gotchildmax) {
1395 		if (childmax >= ppr->pr_childmax) {
1396 			error = EPERM;
1397 			goto done_deref_locked;
1398 		}
1399 	}
1400 	if (gotenforce) {
1401 		if (enforce < ppr->pr_enforce_statfs) {
1402 			error = EPERM;
1403 			goto done_deref_locked;
1404 		}
1405 	}
1406 	if (gotrsnum) {
1407 		/*
1408 		 * devfs_rsnum is a uint16_t
1409 		 */
1410 		if (rsnum < 0 || rsnum > 65535) {
1411 			error = EINVAL;
1412 			goto done_deref_locked;
1413 		}
1414 		/*
1415 		 * Nested jails always inherit parent's devfs ruleset
1416 		 */
1417 		if (jailed(td->td_ucred)) {
1418 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1419 				error = EPERM;
1420 				goto done_deref_locked;
1421 			} else
1422 				rsnum = ppr->pr_devfs_rsnum;
1423 		}
1424 	}
1425 #ifdef INET
1426 	if (ip4s > 0) {
1427 		if (ppr->pr_flags & PR_IP4) {
1428 			/*
1429 			 * Make sure the new set of IP addresses is a
1430 			 * subset of the parent's list.  Don't worry
1431 			 * about the parent being unlocked, as any
1432 			 * setting is done with allprison_lock held.
1433 			 */
1434 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1435 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1436 					break;
1437 			if (ij == ppr->pr_ip4s) {
1438 				error = EPERM;
1439 				goto done_deref_locked;
1440 			}
1441 			if (ip4s > 1) {
1442 				for (ii = ij = 1; ii < ip4s; ii++) {
1443 					if (ip4[ii].s_addr ==
1444 					    ppr->pr_ip4[0].s_addr)
1445 						continue;
1446 					for (; ij < ppr->pr_ip4s; ij++)
1447 						if (ip4[ii].s_addr ==
1448 						    ppr->pr_ip4[ij].s_addr)
1449 							break;
1450 					if (ij == ppr->pr_ip4s)
1451 						break;
1452 				}
1453 				if (ij == ppr->pr_ip4s) {
1454 					error = EPERM;
1455 					goto done_deref_locked;
1456 				}
1457 			}
1458 		}
1459 		/*
1460 		 * Check for conflicting IP addresses.  We permit them
1461 		 * if there is no more than one IP on each jail.  If
1462 		 * there is a duplicate on a jail with more than one
1463 		 * IP stop checking and return error.
1464 		 */
1465 		tppr = ppr;
1466 #ifdef VIMAGE
1467 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1468 			if (tppr->pr_flags & PR_VNET)
1469 				break;
1470 #endif
1471 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1472 			if (tpr == pr ||
1473 #ifdef VIMAGE
1474 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1475 #endif
1476 			    tpr->pr_uref == 0) {
1477 				descend = 0;
1478 				continue;
1479 			}
1480 			if (!(tpr->pr_flags & PR_IP4_USER))
1481 				continue;
1482 			descend = 0;
1483 			if (tpr->pr_ip4 == NULL ||
1484 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1485 				continue;
1486 			for (ii = 0; ii < ip4s; ii++) {
1487 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1488 					error = EADDRINUSE;
1489 					vfs_opterror(opts,
1490 					    "IPv4 addresses clash");
1491 					goto done_deref_locked;
1492 				}
1493 			}
1494 		}
1495 	}
1496 #endif
1497 #ifdef INET6
1498 	if (ip6s > 0) {
1499 		if (ppr->pr_flags & PR_IP6) {
1500 			/*
1501 			 * Make sure the new set of IP addresses is a
1502 			 * subset of the parent's list.
1503 			 */
1504 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1505 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1506 				    &ppr->pr_ip6[ij]))
1507 					break;
1508 			if (ij == ppr->pr_ip6s) {
1509 				error = EPERM;
1510 				goto done_deref_locked;
1511 			}
1512 			if (ip6s > 1) {
1513 				for (ii = ij = 1; ii < ip6s; ii++) {
1514 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1515 					     &ppr->pr_ip6[0]))
1516 						continue;
1517 					for (; ij < ppr->pr_ip6s; ij++)
1518 						if (IN6_ARE_ADDR_EQUAL(
1519 						    &ip6[ii], &ppr->pr_ip6[ij]))
1520 							break;
1521 					if (ij == ppr->pr_ip6s)
1522 						break;
1523 				}
1524 				if (ij == ppr->pr_ip6s) {
1525 					error = EPERM;
1526 					goto done_deref_locked;
1527 				}
1528 			}
1529 		}
1530 		/* Check for conflicting IP addresses. */
1531 		tppr = ppr;
1532 #ifdef VIMAGE
1533 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1534 			if (tppr->pr_flags & PR_VNET)
1535 				break;
1536 #endif
1537 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1538 			if (tpr == pr ||
1539 #ifdef VIMAGE
1540 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1541 #endif
1542 			    tpr->pr_uref == 0) {
1543 				descend = 0;
1544 				continue;
1545 			}
1546 			if (!(tpr->pr_flags & PR_IP6_USER))
1547 				continue;
1548 			descend = 0;
1549 			if (tpr->pr_ip6 == NULL ||
1550 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1551 				continue;
1552 			for (ii = 0; ii < ip6s; ii++) {
1553 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1554 					error = EADDRINUSE;
1555 					vfs_opterror(opts,
1556 					    "IPv6 addresses clash");
1557 					goto done_deref_locked;
1558 				}
1559 			}
1560 		}
1561 	}
1562 #endif
1563 	onamelen = namelen = 0;
1564 	if (namelc != NULL) {
1565 		/* Give a default name of the jid.  Also allow the name to be
1566 		 * explicitly the jid - but not any other number, and only in
1567 		 * normal form (no leading zero/etc).
1568 		 */
1569 		if (namelc[0] == '\0')
1570 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1571 		else if ((strtoul(namelc, &p, 10) != jid ||
1572 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1573 			error = EINVAL;
1574 			vfs_opterror(opts,
1575 			    "name cannot be numeric (unless it is the jid)");
1576 			goto done_deref_locked;
1577 		}
1578 		/*
1579 		 * Make sure the name isn't too long for the prison or its
1580 		 * children.
1581 		 */
1582 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1583 		onamelen = strlen(pr->pr_name + pnamelen);
1584 		namelen = strlen(namelc);
1585 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1586 			error = ENAMETOOLONG;
1587 			goto done_deref_locked;
1588 		}
1589 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1590 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1591 			    sizeof(pr->pr_name)) {
1592 				error = ENAMETOOLONG;
1593 				goto done_deref_locked;
1594 			}
1595 		}
1596 	}
1597 	if (pr_allow & ~ppr->pr_allow) {
1598 		error = EPERM;
1599 		goto done_deref_locked;
1600 	}
1601 
1602 	/*
1603 	 * Let modules check their parameters.  This requires unlocking and
1604 	 * then re-locking the prison, but this is still a valid state as long
1605 	 * as allprison_lock remains xlocked.
1606 	 */
1607 	mtx_unlock(&pr->pr_mtx);
1608 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1609 	if (error != 0) {
1610 		prison_deref(pr, created
1611 		    ? PD_LIST_XLOCKED
1612 		    : PD_DEREF | PD_LIST_XLOCKED);
1613 		goto done_releroot;
1614 	}
1615 	mtx_lock(&pr->pr_mtx);
1616 
1617 	/* At this point, all valid parameters should have been noted. */
1618 	TAILQ_FOREACH(opt, opts, link) {
1619 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1620 			error = EINVAL;
1621 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1622 			goto done_deref_locked;
1623 		}
1624 	}
1625 
1626 	/* Set the parameters of the prison. */
1627 #ifdef INET
1628 	redo_ip4 = 0;
1629 	if (pr_flags & PR_IP4_USER) {
1630 		pr->pr_flags |= PR_IP4;
1631 		free(pr->pr_ip4, M_PRISON);
1632 		pr->pr_ip4s = ip4s;
1633 		pr->pr_ip4 = ip4;
1634 		ip4 = NULL;
1635 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1636 #ifdef VIMAGE
1637 			if (tpr->pr_flags & PR_VNET) {
1638 				descend = 0;
1639 				continue;
1640 			}
1641 #endif
1642 			if (prison_restrict_ip4(tpr, NULL)) {
1643 				redo_ip4 = 1;
1644 				descend = 0;
1645 			}
1646 		}
1647 	}
1648 #endif
1649 #ifdef INET6
1650 	redo_ip6 = 0;
1651 	if (pr_flags & PR_IP6_USER) {
1652 		pr->pr_flags |= PR_IP6;
1653 		free(pr->pr_ip6, M_PRISON);
1654 		pr->pr_ip6s = ip6s;
1655 		pr->pr_ip6 = ip6;
1656 		ip6 = NULL;
1657 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1658 #ifdef VIMAGE
1659 			if (tpr->pr_flags & PR_VNET) {
1660 				descend = 0;
1661 				continue;
1662 			}
1663 #endif
1664 			if (prison_restrict_ip6(tpr, NULL)) {
1665 				redo_ip6 = 1;
1666 				descend = 0;
1667 			}
1668 		}
1669 	}
1670 #endif
1671 	if (gotslevel) {
1672 		pr->pr_securelevel = slevel;
1673 		/* Set all child jails to be at least this level. */
1674 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1675 			if (tpr->pr_securelevel < slevel)
1676 				tpr->pr_securelevel = slevel;
1677 	}
1678 	if (gotchildmax) {
1679 		pr->pr_childmax = childmax;
1680 		/* Set all child jails to under this limit. */
1681 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1682 			if (tpr->pr_childmax > childmax - level)
1683 				tpr->pr_childmax = childmax > level
1684 				    ? childmax - level : 0;
1685 	}
1686 	if (gotenforce) {
1687 		pr->pr_enforce_statfs = enforce;
1688 		/* Pass this restriction on to the children. */
1689 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1690 			if (tpr->pr_enforce_statfs < enforce)
1691 				tpr->pr_enforce_statfs = enforce;
1692 	}
1693 	if (gotrsnum) {
1694 		pr->pr_devfs_rsnum = rsnum;
1695 		/* Pass this restriction on to the children. */
1696 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1697 			tpr->pr_devfs_rsnum = rsnum;
1698 	}
1699 	if (namelc != NULL) {
1700 		if (ppr == &prison0)
1701 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1702 		else
1703 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1704 			    ppr->pr_name, namelc);
1705 		/* Change this component of child names. */
1706 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1707 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1708 			    strlen(tpr->pr_name + onamelen) + 1);
1709 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1710 		}
1711 	}
1712 	if (path != NULL) {
1713 		/* Try to keep a real-rooted full pathname. */
1714 		if (fullpath_disabled && path[0] == '/' &&
1715 		    strcmp(mypr->pr_path, "/"))
1716 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1717 			    mypr->pr_path, path);
1718 		else
1719 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1720 		pr->pr_root = root;
1721 	}
1722 	if (PR_HOST & ch_flags & ~pr_flags) {
1723 		if (pr->pr_flags & PR_HOST) {
1724 			/*
1725 			 * Copy the parent's host info.  As with pr_ip4 above,
1726 			 * the lack of a lock on the parent is not a problem;
1727 			 * it is always set with allprison_lock at least
1728 			 * shared, and is held exclusively here.
1729 			 */
1730 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1731 			    sizeof(pr->pr_hostname));
1732 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1733 			    sizeof(pr->pr_domainname));
1734 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1735 			    sizeof(pr->pr_hostuuid));
1736 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1737 		}
1738 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1739 		/* Set this prison, and any descendants without PR_HOST. */
1740 		if (host != NULL)
1741 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1742 		if (domain != NULL)
1743 			strlcpy(pr->pr_domainname, domain,
1744 			    sizeof(pr->pr_domainname));
1745 		if (uuid != NULL)
1746 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1747 		if (gothid)
1748 			pr->pr_hostid = hid;
1749 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1750 			if (tpr->pr_flags & PR_HOST)
1751 				descend = 0;
1752 			else {
1753 				if (host != NULL)
1754 					strlcpy(tpr->pr_hostname,
1755 					    pr->pr_hostname,
1756 					    sizeof(tpr->pr_hostname));
1757 				if (domain != NULL)
1758 					strlcpy(tpr->pr_domainname,
1759 					    pr->pr_domainname,
1760 					    sizeof(tpr->pr_domainname));
1761 				if (uuid != NULL)
1762 					strlcpy(tpr->pr_hostuuid,
1763 					    pr->pr_hostuuid,
1764 					    sizeof(tpr->pr_hostuuid));
1765 				if (gothid)
1766 					tpr->pr_hostid = hid;
1767 			}
1768 		}
1769 	}
1770 	if ((tallow = ch_allow & ~pr_allow)) {
1771 		/* Clear allow bits in all children. */
1772 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1773 			tpr->pr_allow &= ~tallow;
1774 	}
1775 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1776 	/*
1777 	 * Persistent prisons get an extra reference, and prisons losing their
1778 	 * persist flag lose that reference.  Only do this for existing prisons
1779 	 * for now, so new ones will remain unseen until after the module
1780 	 * handlers have completed.
1781 	 */
1782 	born = pr->pr_uref == 0;
1783 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1784 		if (pr_flags & PR_PERSIST) {
1785 			pr->pr_ref++;
1786 			pr->pr_uref++;
1787 		} else {
1788 			pr->pr_ref--;
1789 			pr->pr_uref--;
1790 		}
1791 	}
1792 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1793 	mtx_unlock(&pr->pr_mtx);
1794 
1795 #ifdef RACCT
1796 	if (racct_enable && created)
1797 		prison_racct_attach(pr);
1798 #endif
1799 
1800 	/* Locks may have prevented a complete restriction of child IP
1801 	 * addresses.  If so, allocate some more memory and try again.
1802 	 */
1803 #ifdef INET
1804 	while (redo_ip4) {
1805 		ip4s = pr->pr_ip4s;
1806 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1807 		mtx_lock(&pr->pr_mtx);
1808 		redo_ip4 = 0;
1809 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1810 #ifdef VIMAGE
1811 			if (tpr->pr_flags & PR_VNET) {
1812 				descend = 0;
1813 				continue;
1814 			}
1815 #endif
1816 			if (prison_restrict_ip4(tpr, ip4)) {
1817 				if (ip4 != NULL)
1818 					ip4 = NULL;
1819 				else
1820 					redo_ip4 = 1;
1821 			}
1822 		}
1823 		mtx_unlock(&pr->pr_mtx);
1824 	}
1825 #endif
1826 #ifdef INET6
1827 	while (redo_ip6) {
1828 		ip6s = pr->pr_ip6s;
1829 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1830 		mtx_lock(&pr->pr_mtx);
1831 		redo_ip6 = 0;
1832 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1833 #ifdef VIMAGE
1834 			if (tpr->pr_flags & PR_VNET) {
1835 				descend = 0;
1836 				continue;
1837 			}
1838 #endif
1839 			if (prison_restrict_ip6(tpr, ip6)) {
1840 				if (ip6 != NULL)
1841 					ip6 = NULL;
1842 				else
1843 					redo_ip6 = 1;
1844 			}
1845 		}
1846 		mtx_unlock(&pr->pr_mtx);
1847 	}
1848 #endif
1849 
1850 	/* Let the modules do their work. */
1851 	sx_downgrade(&allprison_lock);
1852 	if (born) {
1853 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1854 		if (error) {
1855 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1856 			prison_deref(pr, created
1857 			    ? PD_LIST_SLOCKED
1858 			    : PD_DEREF | PD_LIST_SLOCKED);
1859 			goto done_errmsg;
1860 		}
1861 	}
1862 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1863 	if (error) {
1864 		if (born)
1865 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1866 		prison_deref(pr, created
1867 		    ? PD_LIST_SLOCKED
1868 		    : PD_DEREF | PD_LIST_SLOCKED);
1869 		goto done_errmsg;
1870 	}
1871 
1872 	/* Attach this process to the prison if requested. */
1873 	if (flags & JAIL_ATTACH) {
1874 		mtx_lock(&pr->pr_mtx);
1875 		error = do_jail_attach(td, pr);
1876 		if (error) {
1877 			vfs_opterror(opts, "attach failed");
1878 			if (!created)
1879 				prison_deref(pr, PD_DEREF);
1880 			goto done_errmsg;
1881 		}
1882 	}
1883 
1884 #ifdef RACCT
1885 	if (racct_enable && !created) {
1886 		if (!(flags & JAIL_ATTACH))
1887 			sx_sunlock(&allprison_lock);
1888 		prison_racct_modify(pr);
1889 		if (!(flags & JAIL_ATTACH))
1890 			sx_slock(&allprison_lock);
1891 	}
1892 #endif
1893 
1894 	td->td_retval[0] = pr->pr_id;
1895 
1896 	/*
1897 	 * Now that it is all there, drop the temporary reference from existing
1898 	 * prisons.  Or add a reference to newly created persistent prisons
1899 	 * (which was not done earlier so that the prison would not be publicly
1900 	 * visible).
1901 	 */
1902 	if (!created) {
1903 		prison_deref(pr, (flags & JAIL_ATTACH)
1904 		    ? PD_DEREF
1905 		    : PD_DEREF | PD_LIST_SLOCKED);
1906 	} else {
1907 		if (pr_flags & PR_PERSIST) {
1908 			mtx_lock(&pr->pr_mtx);
1909 			pr->pr_ref++;
1910 			pr->pr_uref++;
1911 			mtx_unlock(&pr->pr_mtx);
1912 		}
1913 		if (!(flags & JAIL_ATTACH))
1914 			sx_sunlock(&allprison_lock);
1915 	}
1916 
1917 	goto done_free;
1918 
1919  done_deref_locked:
1920 	prison_deref(pr, created
1921 	    ? PD_LOCKED | PD_LIST_XLOCKED
1922 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1923 	goto done_releroot;
1924  done_unlock_list:
1925 	sx_xunlock(&allprison_lock);
1926  done_releroot:
1927 	if (root != NULL)
1928 		vrele(root);
1929  done_errmsg:
1930 	if (error) {
1931 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1932 		if (errmsg_len > 0) {
1933 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1934 			if (errmsg_pos > 0) {
1935 				if (optuio->uio_segflg == UIO_SYSSPACE)
1936 					bcopy(errmsg,
1937 					   optuio->uio_iov[errmsg_pos].iov_base,
1938 					   errmsg_len);
1939 				else
1940 					copyout(errmsg,
1941 					   optuio->uio_iov[errmsg_pos].iov_base,
1942 					   errmsg_len);
1943 			}
1944 		}
1945 	}
1946  done_free:
1947 #ifdef INET
1948 	free(ip4, M_PRISON);
1949 #endif
1950 #ifdef INET6
1951 	free(ip6, M_PRISON);
1952 #endif
1953 	if (g_path != NULL)
1954 		free(g_path, M_TEMP);
1955 	vfs_freeopts(opts);
1956 	return (error);
1957 }
1958 
1959 
1960 /*
1961  * struct jail_get_args {
1962  *	struct iovec *iovp;
1963  *	unsigned int iovcnt;
1964  *	int flags;
1965  * };
1966  */
1967 int
1968 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1969 {
1970 	struct uio *auio;
1971 	int error;
1972 
1973 	/* Check that we have an even number of iovecs. */
1974 	if (uap->iovcnt & 1)
1975 		return (EINVAL);
1976 
1977 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1978 	if (error)
1979 		return (error);
1980 	error = kern_jail_get(td, auio, uap->flags);
1981 	if (error == 0)
1982 		error = copyout(auio->uio_iov, uap->iovp,
1983 		    uap->iovcnt * sizeof (struct iovec));
1984 	free(auio, M_IOV);
1985 	return (error);
1986 }
1987 
1988 int
1989 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1990 {
1991 	struct prison *pr, *mypr;
1992 	struct vfsopt *opt;
1993 	struct vfsoptlist *opts;
1994 	char *errmsg, *name;
1995 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1996 
1997 	if (flags & ~JAIL_GET_MASK)
1998 		return (EINVAL);
1999 
2000 	/* Get the parameter list. */
2001 	error = vfs_buildopts(optuio, &opts);
2002 	if (error)
2003 		return (error);
2004 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2005 	mypr = td->td_ucred->cr_prison;
2006 
2007 	/*
2008 	 * Find the prison specified by one of: lastjid, jid, name.
2009 	 */
2010 	sx_slock(&allprison_lock);
2011 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2012 	if (error == 0) {
2013 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2014 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2015 				mtx_lock(&pr->pr_mtx);
2016 				if (pr->pr_ref > 0 &&
2017 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2018 					break;
2019 				mtx_unlock(&pr->pr_mtx);
2020 			}
2021 		}
2022 		if (pr != NULL)
2023 			goto found_prison;
2024 		error = ENOENT;
2025 		vfs_opterror(opts, "no jail after %d", jid);
2026 		goto done_unlock_list;
2027 	} else if (error != ENOENT)
2028 		goto done_unlock_list;
2029 
2030 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2031 	if (error == 0) {
2032 		if (jid != 0) {
2033 			pr = prison_find_child(mypr, jid);
2034 			if (pr != NULL) {
2035 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2036 					mtx_unlock(&pr->pr_mtx);
2037 					error = ENOENT;
2038 					vfs_opterror(opts, "jail %d is dying",
2039 					    jid);
2040 					goto done_unlock_list;
2041 				}
2042 				goto found_prison;
2043 			}
2044 			error = ENOENT;
2045 			vfs_opterror(opts, "jail %d not found", jid);
2046 			goto done_unlock_list;
2047 		}
2048 	} else if (error != ENOENT)
2049 		goto done_unlock_list;
2050 
2051 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2052 	if (error == 0) {
2053 		if (len == 0 || name[len - 1] != '\0') {
2054 			error = EINVAL;
2055 			goto done_unlock_list;
2056 		}
2057 		pr = prison_find_name(mypr, name);
2058 		if (pr != NULL) {
2059 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2060 				mtx_unlock(&pr->pr_mtx);
2061 				error = ENOENT;
2062 				vfs_opterror(opts, "jail \"%s\" is dying",
2063 				    name);
2064 				goto done_unlock_list;
2065 			}
2066 			goto found_prison;
2067 		}
2068 		error = ENOENT;
2069 		vfs_opterror(opts, "jail \"%s\" not found", name);
2070 		goto done_unlock_list;
2071 	} else if (error != ENOENT)
2072 		goto done_unlock_list;
2073 
2074 	vfs_opterror(opts, "no jail specified");
2075 	error = ENOENT;
2076 	goto done_unlock_list;
2077 
2078  found_prison:
2079 	/* Get the parameters of the prison. */
2080 	pr->pr_ref++;
2081 	locked = PD_LOCKED;
2082 	td->td_retval[0] = pr->pr_id;
2083 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2084 	if (error != 0 && error != ENOENT)
2085 		goto done_deref;
2086 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2087 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2088 	if (error != 0 && error != ENOENT)
2089 		goto done_deref;
2090 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2091 	if (error != 0 && error != ENOENT)
2092 		goto done_deref;
2093 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2094 	    sizeof(pr->pr_cpuset->cs_id));
2095 	if (error != 0 && error != ENOENT)
2096 		goto done_deref;
2097 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2098 	if (error != 0 && error != ENOENT)
2099 		goto done_deref;
2100 #ifdef INET
2101 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2102 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2103 	if (error != 0 && error != ENOENT)
2104 		goto done_deref;
2105 #endif
2106 #ifdef INET6
2107 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2108 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2109 	if (error != 0 && error != ENOENT)
2110 		goto done_deref;
2111 #endif
2112 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2113 	    sizeof(pr->pr_securelevel));
2114 	if (error != 0 && error != ENOENT)
2115 		goto done_deref;
2116 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2117 	    sizeof(pr->pr_childcount));
2118 	if (error != 0 && error != ENOENT)
2119 		goto done_deref;
2120 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2121 	    sizeof(pr->pr_childmax));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2125 	if (error != 0 && error != ENOENT)
2126 		goto done_deref;
2127 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2128 	if (error != 0 && error != ENOENT)
2129 		goto done_deref;
2130 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2131 	if (error != 0 && error != ENOENT)
2132 		goto done_deref;
2133 #ifdef COMPAT_FREEBSD32
2134 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2135 		uint32_t hid32 = pr->pr_hostid;
2136 
2137 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2138 	} else
2139 #endif
2140 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2141 	    sizeof(pr->pr_hostid));
2142 	if (error != 0 && error != ENOENT)
2143 		goto done_deref;
2144 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2145 	    sizeof(pr->pr_enforce_statfs));
2146 	if (error != 0 && error != ENOENT)
2147 		goto done_deref;
2148 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2149 	    sizeof(pr->pr_devfs_rsnum));
2150 	if (error != 0 && error != ENOENT)
2151 		goto done_deref;
2152 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2153 		if (pr_flag_names[fi] == NULL)
2154 			continue;
2155 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2156 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2157 		if (error != 0 && error != ENOENT)
2158 			goto done_deref;
2159 		i = !i;
2160 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2161 		if (error != 0 && error != ENOENT)
2162 			goto done_deref;
2163 	}
2164 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
2165 		i = pr->pr_flags &
2166 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2167 		i = pr_flag_jailsys[fi].disable &&
2168 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2169 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2170 		    : JAIL_SYS_INHERIT;
2171 		error =
2172 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2173 		if (error != 0 && error != ENOENT)
2174 			goto done_deref;
2175 	}
2176 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
2177 		if (pr_allow_names[fi] == NULL)
2178 			continue;
2179 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2180 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2181 		if (error != 0 && error != ENOENT)
2182 			goto done_deref;
2183 		i = !i;
2184 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2185 		if (error != 0 && error != ENOENT)
2186 			goto done_deref;
2187 	}
2188 	i = (pr->pr_uref == 0);
2189 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2190 	if (error != 0 && error != ENOENT)
2191 		goto done_deref;
2192 	i = !i;
2193 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2194 	if (error != 0 && error != ENOENT)
2195 		goto done_deref;
2196 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2197 	    sizeof(pr->pr_osreldate));
2198 	if (error != 0 && error != ENOENT)
2199 		goto done_deref;
2200 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2201 	if (error != 0 && error != ENOENT)
2202 		goto done_deref;
2203 
2204 	/* Get the module parameters. */
2205 	mtx_unlock(&pr->pr_mtx);
2206 	locked = 0;
2207 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2208 	if (error)
2209 		goto done_deref;
2210 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2211 
2212 	/* By now, all parameters should have been noted. */
2213 	TAILQ_FOREACH(opt, opts, link) {
2214 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2215 			error = EINVAL;
2216 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2217 			goto done_errmsg;
2218 		}
2219 	}
2220 
2221 	/* Write the fetched parameters back to userspace. */
2222 	error = 0;
2223 	TAILQ_FOREACH(opt, opts, link) {
2224 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2225 			pos = 2 * opt->pos + 1;
2226 			optuio->uio_iov[pos].iov_len = opt->len;
2227 			if (opt->value != NULL) {
2228 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2229 					bcopy(opt->value,
2230 					    optuio->uio_iov[pos].iov_base,
2231 					    opt->len);
2232 				} else {
2233 					error = copyout(opt->value,
2234 					    optuio->uio_iov[pos].iov_base,
2235 					    opt->len);
2236 					if (error)
2237 						break;
2238 				}
2239 			}
2240 		}
2241 	}
2242 	goto done_errmsg;
2243 
2244  done_deref:
2245 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2246 	goto done_errmsg;
2247 
2248  done_unlock_list:
2249 	sx_sunlock(&allprison_lock);
2250  done_errmsg:
2251 	if (error && errmsg_pos >= 0) {
2252 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2253 		errmsg_pos = 2 * errmsg_pos + 1;
2254 		if (errmsg_len > 0) {
2255 			if (optuio->uio_segflg == UIO_SYSSPACE)
2256 				bcopy(errmsg,
2257 				    optuio->uio_iov[errmsg_pos].iov_base,
2258 				    errmsg_len);
2259 			else
2260 				copyout(errmsg,
2261 				    optuio->uio_iov[errmsg_pos].iov_base,
2262 				    errmsg_len);
2263 		}
2264 	}
2265 	vfs_freeopts(opts);
2266 	return (error);
2267 }
2268 
2269 
2270 /*
2271  * struct jail_remove_args {
2272  *	int jid;
2273  * };
2274  */
2275 int
2276 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2277 {
2278 	struct prison *pr, *cpr, *lpr, *tpr;
2279 	int descend, error;
2280 
2281 	error = priv_check(td, PRIV_JAIL_REMOVE);
2282 	if (error)
2283 		return (error);
2284 
2285 	sx_xlock(&allprison_lock);
2286 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2287 	if (pr == NULL) {
2288 		sx_xunlock(&allprison_lock);
2289 		return (EINVAL);
2290 	}
2291 
2292 	/* Remove all descendants of this prison, then remove this prison. */
2293 	pr->pr_ref++;
2294 	if (!LIST_EMPTY(&pr->pr_children)) {
2295 		mtx_unlock(&pr->pr_mtx);
2296 		lpr = NULL;
2297 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2298 			mtx_lock(&cpr->pr_mtx);
2299 			if (cpr->pr_ref > 0) {
2300 				tpr = cpr;
2301 				cpr->pr_ref++;
2302 			} else {
2303 				/* Already removed - do not do it again. */
2304 				tpr = NULL;
2305 			}
2306 			mtx_unlock(&cpr->pr_mtx);
2307 			if (lpr != NULL) {
2308 				mtx_lock(&lpr->pr_mtx);
2309 				prison_remove_one(lpr);
2310 				sx_xlock(&allprison_lock);
2311 			}
2312 			lpr = tpr;
2313 		}
2314 		if (lpr != NULL) {
2315 			mtx_lock(&lpr->pr_mtx);
2316 			prison_remove_one(lpr);
2317 			sx_xlock(&allprison_lock);
2318 		}
2319 		mtx_lock(&pr->pr_mtx);
2320 	}
2321 	prison_remove_one(pr);
2322 	return (0);
2323 }
2324 
2325 static void
2326 prison_remove_one(struct prison *pr)
2327 {
2328 	struct proc *p;
2329 	int deuref;
2330 
2331 	/* If the prison was persistent, it is not anymore. */
2332 	deuref = 0;
2333 	if (pr->pr_flags & PR_PERSIST) {
2334 		pr->pr_ref--;
2335 		deuref = PD_DEUREF;
2336 		pr->pr_flags &= ~PR_PERSIST;
2337 	}
2338 
2339 	/*
2340 	 * jail_remove added a reference.  If that's the only one, remove
2341 	 * the prison now.
2342 	 */
2343 	KASSERT(pr->pr_ref > 0,
2344 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2345 	if (pr->pr_ref == 1) {
2346 		prison_deref(pr,
2347 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2348 		return;
2349 	}
2350 
2351 	mtx_unlock(&pr->pr_mtx);
2352 	sx_xunlock(&allprison_lock);
2353 	/*
2354 	 * Kill all processes unfortunate enough to be attached to this prison.
2355 	 */
2356 	sx_slock(&allproc_lock);
2357 	LIST_FOREACH(p, &allproc, p_list) {
2358 		PROC_LOCK(p);
2359 		if (p->p_state != PRS_NEW && p->p_ucred &&
2360 		    p->p_ucred->cr_prison == pr)
2361 			kern_psignal(p, SIGKILL);
2362 		PROC_UNLOCK(p);
2363 	}
2364 	sx_sunlock(&allproc_lock);
2365 	/* Remove the temporary reference added by jail_remove. */
2366 	prison_deref(pr, deuref | PD_DEREF);
2367 }
2368 
2369 
2370 /*
2371  * struct jail_attach_args {
2372  *	int jid;
2373  * };
2374  */
2375 int
2376 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2377 {
2378 	struct prison *pr;
2379 	int error;
2380 
2381 	error = priv_check(td, PRIV_JAIL_ATTACH);
2382 	if (error)
2383 		return (error);
2384 
2385 	sx_slock(&allprison_lock);
2386 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2387 	if (pr == NULL) {
2388 		sx_sunlock(&allprison_lock);
2389 		return (EINVAL);
2390 	}
2391 
2392 	/*
2393 	 * Do not allow a process to attach to a prison that is not
2394 	 * considered to be "alive".
2395 	 */
2396 	if (pr->pr_uref == 0) {
2397 		mtx_unlock(&pr->pr_mtx);
2398 		sx_sunlock(&allprison_lock);
2399 		return (EINVAL);
2400 	}
2401 
2402 	return (do_jail_attach(td, pr));
2403 }
2404 
2405 static int
2406 do_jail_attach(struct thread *td, struct prison *pr)
2407 {
2408 	struct prison *ppr;
2409 	struct proc *p;
2410 	struct ucred *newcred, *oldcred;
2411 	int error;
2412 
2413 	/*
2414 	 * XXX: Note that there is a slight race here if two threads
2415 	 * in the same privileged process attempt to attach to two
2416 	 * different jails at the same time.  It is important for
2417 	 * user processes not to do this, or they might end up with
2418 	 * a process root from one prison, but attached to the jail
2419 	 * of another.
2420 	 */
2421 	pr->pr_ref++;
2422 	pr->pr_uref++;
2423 	mtx_unlock(&pr->pr_mtx);
2424 
2425 	/* Let modules do whatever they need to prepare for attaching. */
2426 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2427 	if (error) {
2428 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2429 		return (error);
2430 	}
2431 	sx_sunlock(&allprison_lock);
2432 
2433 	/*
2434 	 * Reparent the newly attached process to this jail.
2435 	 */
2436 	ppr = td->td_ucred->cr_prison;
2437 	p = td->td_proc;
2438 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2439 	if (error)
2440 		goto e_revert_osd;
2441 
2442 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2443 	if ((error = change_dir(pr->pr_root, td)) != 0)
2444 		goto e_unlock;
2445 #ifdef MAC
2446 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2447 		goto e_unlock;
2448 #endif
2449 	VOP_UNLOCK(pr->pr_root, 0);
2450 	if ((error = pwd_chroot(td, pr->pr_root)))
2451 		goto e_revert_osd;
2452 
2453 	newcred = crget();
2454 	PROC_LOCK(p);
2455 	oldcred = p->p_ucred;
2456 	setsugid(p);
2457 	crcopy(newcred, oldcred);
2458 	newcred->cr_prison = pr;
2459 	proc_set_cred(p, newcred);
2460 	PROC_UNLOCK(p);
2461 #ifdef RACCT
2462 	racct_proc_ucred_changed(p, oldcred, newcred);
2463 #endif
2464 	crfree(oldcred);
2465 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2466 	return (0);
2467  e_unlock:
2468 	VOP_UNLOCK(pr->pr_root, 0);
2469  e_revert_osd:
2470 	/* Tell modules this thread is still in its old jail after all. */
2471 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2472 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2473 	return (error);
2474 }
2475 
2476 
2477 /*
2478  * Returns a locked prison instance, or NULL on failure.
2479  */
2480 struct prison *
2481 prison_find(int prid)
2482 {
2483 	struct prison *pr;
2484 
2485 	sx_assert(&allprison_lock, SX_LOCKED);
2486 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2487 		if (pr->pr_id == prid) {
2488 			mtx_lock(&pr->pr_mtx);
2489 			if (pr->pr_ref > 0)
2490 				return (pr);
2491 			mtx_unlock(&pr->pr_mtx);
2492 		}
2493 	}
2494 	return (NULL);
2495 }
2496 
2497 /*
2498  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2499  */
2500 struct prison *
2501 prison_find_child(struct prison *mypr, int prid)
2502 {
2503 	struct prison *pr;
2504 	int descend;
2505 
2506 	sx_assert(&allprison_lock, SX_LOCKED);
2507 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2508 		if (pr->pr_id == prid) {
2509 			mtx_lock(&pr->pr_mtx);
2510 			if (pr->pr_ref > 0)
2511 				return (pr);
2512 			mtx_unlock(&pr->pr_mtx);
2513 		}
2514 	}
2515 	return (NULL);
2516 }
2517 
2518 /*
2519  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2520  */
2521 struct prison *
2522 prison_find_name(struct prison *mypr, const char *name)
2523 {
2524 	struct prison *pr, *deadpr;
2525 	size_t mylen;
2526 	int descend;
2527 
2528 	sx_assert(&allprison_lock, SX_LOCKED);
2529 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2530  again:
2531 	deadpr = NULL;
2532 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2533 		if (!strcmp(pr->pr_name + mylen, name)) {
2534 			mtx_lock(&pr->pr_mtx);
2535 			if (pr->pr_ref > 0) {
2536 				if (pr->pr_uref > 0)
2537 					return (pr);
2538 				deadpr = pr;
2539 			}
2540 			mtx_unlock(&pr->pr_mtx);
2541 		}
2542 	}
2543 	/* There was no valid prison - perhaps there was a dying one. */
2544 	if (deadpr != NULL) {
2545 		mtx_lock(&deadpr->pr_mtx);
2546 		if (deadpr->pr_ref == 0) {
2547 			mtx_unlock(&deadpr->pr_mtx);
2548 			goto again;
2549 		}
2550 	}
2551 	return (deadpr);
2552 }
2553 
2554 /*
2555  * See if a prison has the specific flag set.
2556  */
2557 int
2558 prison_flag(struct ucred *cred, unsigned flag)
2559 {
2560 
2561 	/* This is an atomic read, so no locking is necessary. */
2562 	return (cred->cr_prison->pr_flags & flag);
2563 }
2564 
2565 int
2566 prison_allow(struct ucred *cred, unsigned flag)
2567 {
2568 
2569 	/* This is an atomic read, so no locking is necessary. */
2570 	return (cred->cr_prison->pr_allow & flag);
2571 }
2572 
2573 /*
2574  * Remove a prison reference.  If that was the last reference, remove the
2575  * prison itself - but not in this context in case there are locks held.
2576  */
2577 void
2578 prison_free_locked(struct prison *pr)
2579 {
2580 
2581 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2582 	pr->pr_ref--;
2583 	if (pr->pr_ref == 0) {
2584 		mtx_unlock(&pr->pr_mtx);
2585 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2586 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2587 		return;
2588 	}
2589 	mtx_unlock(&pr->pr_mtx);
2590 }
2591 
2592 void
2593 prison_free(struct prison *pr)
2594 {
2595 
2596 	mtx_lock(&pr->pr_mtx);
2597 	prison_free_locked(pr);
2598 }
2599 
2600 static void
2601 prison_complete(void *context, int pending)
2602 {
2603 
2604 	prison_deref((struct prison *)context, 0);
2605 }
2606 
2607 /*
2608  * Remove a prison reference (usually).  This internal version assumes no
2609  * mutexes are held, except perhaps the prison itself.  If there are no more
2610  * references, release and delist the prison.  On completion, the prison lock
2611  * and the allprison lock are both unlocked.
2612  */
2613 static void
2614 prison_deref(struct prison *pr, int flags)
2615 {
2616 	struct prison *ppr, *tpr;
2617 	int ref, lasturef;
2618 
2619 	if (!(flags & PD_LOCKED))
2620 		mtx_lock(&pr->pr_mtx);
2621 	for (;;) {
2622 		if (flags & PD_DEUREF) {
2623 			pr->pr_uref--;
2624 			lasturef = pr->pr_uref == 0;
2625 			if (lasturef)
2626 				pr->pr_ref++;
2627 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2628 		} else
2629 			lasturef = 0;
2630 		if (flags & PD_DEREF)
2631 			pr->pr_ref--;
2632 		ref = pr->pr_ref;
2633 		mtx_unlock(&pr->pr_mtx);
2634 
2635 		/*
2636 		 * Tell the modules if the last user reference was removed
2637 		 * (even it sticks around in dying state).
2638 		 */
2639 		if (lasturef) {
2640 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2641 				if (ref > 1) {
2642 					sx_slock(&allprison_lock);
2643 					flags |= PD_LIST_SLOCKED;
2644 				} else {
2645 					sx_xlock(&allprison_lock);
2646 					flags |= PD_LIST_XLOCKED;
2647 				}
2648 			}
2649 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2650 			mtx_lock(&pr->pr_mtx);
2651 			ref = --pr->pr_ref;
2652 			mtx_unlock(&pr->pr_mtx);
2653 		}
2654 
2655 		/* If the prison still has references, nothing else to do. */
2656 		if (ref > 0) {
2657 			if (flags & PD_LIST_SLOCKED)
2658 				sx_sunlock(&allprison_lock);
2659 			else if (flags & PD_LIST_XLOCKED)
2660 				sx_xunlock(&allprison_lock);
2661 			return;
2662 		}
2663 
2664 		if (flags & PD_LIST_SLOCKED) {
2665 			if (!sx_try_upgrade(&allprison_lock)) {
2666 				sx_sunlock(&allprison_lock);
2667 				sx_xlock(&allprison_lock);
2668 			}
2669 		} else if (!(flags & PD_LIST_XLOCKED))
2670 			sx_xlock(&allprison_lock);
2671 
2672 		TAILQ_REMOVE(&allprison, pr, pr_list);
2673 		LIST_REMOVE(pr, pr_sibling);
2674 		ppr = pr->pr_parent;
2675 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2676 			tpr->pr_childcount--;
2677 		sx_xunlock(&allprison_lock);
2678 
2679 #ifdef VIMAGE
2680 		if (pr->pr_vnet != ppr->pr_vnet)
2681 			vnet_destroy(pr->pr_vnet);
2682 #endif
2683 		if (pr->pr_root != NULL)
2684 			vrele(pr->pr_root);
2685 		mtx_destroy(&pr->pr_mtx);
2686 #ifdef INET
2687 		free(pr->pr_ip4, M_PRISON);
2688 #endif
2689 #ifdef INET6
2690 		free(pr->pr_ip6, M_PRISON);
2691 #endif
2692 		if (pr->pr_cpuset != NULL)
2693 			cpuset_rel(pr->pr_cpuset);
2694 		osd_jail_exit(pr);
2695 #ifdef RACCT
2696 		if (racct_enable)
2697 			prison_racct_detach(pr);
2698 #endif
2699 		free(pr, M_PRISON);
2700 
2701 		/* Removing a prison frees a reference on its parent. */
2702 		pr = ppr;
2703 		mtx_lock(&pr->pr_mtx);
2704 		flags = PD_DEREF | PD_DEUREF;
2705 	}
2706 }
2707 
2708 void
2709 prison_hold_locked(struct prison *pr)
2710 {
2711 
2712 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2713 	KASSERT(pr->pr_ref > 0,
2714 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2715 	pr->pr_ref++;
2716 }
2717 
2718 void
2719 prison_hold(struct prison *pr)
2720 {
2721 
2722 	mtx_lock(&pr->pr_mtx);
2723 	prison_hold_locked(pr);
2724 	mtx_unlock(&pr->pr_mtx);
2725 }
2726 
2727 void
2728 prison_proc_hold(struct prison *pr)
2729 {
2730 
2731 	mtx_lock(&pr->pr_mtx);
2732 	KASSERT(pr->pr_uref > 0,
2733 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2734 	pr->pr_uref++;
2735 	mtx_unlock(&pr->pr_mtx);
2736 }
2737 
2738 void
2739 prison_proc_free(struct prison *pr)
2740 {
2741 
2742 	mtx_lock(&pr->pr_mtx);
2743 	KASSERT(pr->pr_uref > 0,
2744 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2745 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2746 }
2747 
2748 
2749 #ifdef INET
2750 /*
2751  * Restrict a prison's IP address list with its parent's, possibly replacing
2752  * it.  Return true if the replacement buffer was used (or would have been).
2753  */
2754 static int
2755 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2756 {
2757 	int ii, ij, used;
2758 	struct prison *ppr;
2759 
2760 	ppr = pr->pr_parent;
2761 	if (!(pr->pr_flags & PR_IP4_USER)) {
2762 		/* This has no user settings, so just copy the parent's list. */
2763 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2764 			/*
2765 			 * There's no room for the parent's list.  Use the
2766 			 * new list buffer, which is assumed to be big enough
2767 			 * (if it was passed).  If there's no buffer, try to
2768 			 * allocate one.
2769 			 */
2770 			used = 1;
2771 			if (newip4 == NULL) {
2772 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2773 				    M_PRISON, M_NOWAIT);
2774 				if (newip4 != NULL)
2775 					used = 0;
2776 			}
2777 			if (newip4 != NULL) {
2778 				bcopy(ppr->pr_ip4, newip4,
2779 				    ppr->pr_ip4s * sizeof(*newip4));
2780 				free(pr->pr_ip4, M_PRISON);
2781 				pr->pr_ip4 = newip4;
2782 				pr->pr_ip4s = ppr->pr_ip4s;
2783 			}
2784 			return (used);
2785 		}
2786 		pr->pr_ip4s = ppr->pr_ip4s;
2787 		if (pr->pr_ip4s > 0)
2788 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2789 			    pr->pr_ip4s * sizeof(*newip4));
2790 		else if (pr->pr_ip4 != NULL) {
2791 			free(pr->pr_ip4, M_PRISON);
2792 			pr->pr_ip4 = NULL;
2793 		}
2794 	} else if (pr->pr_ip4s > 0) {
2795 		/* Remove addresses that aren't in the parent. */
2796 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2797 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2798 				break;
2799 		if (ij < ppr->pr_ip4s)
2800 			ii = 1;
2801 		else {
2802 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2803 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2804 			ii = 0;
2805 		}
2806 		for (ij = 1; ii < pr->pr_ip4s; ) {
2807 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2808 				ii++;
2809 				continue;
2810 			}
2811 			switch (ij >= ppr->pr_ip4s ? -1 :
2812 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2813 			case -1:
2814 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2815 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2816 				break;
2817 			case 0:
2818 				ii++;
2819 				ij++;
2820 				break;
2821 			case 1:
2822 				ij++;
2823 				break;
2824 			}
2825 		}
2826 		if (pr->pr_ip4s == 0) {
2827 			free(pr->pr_ip4, M_PRISON);
2828 			pr->pr_ip4 = NULL;
2829 		}
2830 	}
2831 	return (0);
2832 }
2833 
2834 /*
2835  * Pass back primary IPv4 address of this jail.
2836  *
2837  * If not restricted return success but do not alter the address.  Caller has
2838  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2839  *
2840  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2841  * Address returned in NBO.
2842  */
2843 int
2844 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2845 {
2846 	struct prison *pr;
2847 
2848 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2849 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2850 
2851 	pr = cred->cr_prison;
2852 	if (!(pr->pr_flags & PR_IP4))
2853 		return (0);
2854 	mtx_lock(&pr->pr_mtx);
2855 	if (!(pr->pr_flags & PR_IP4)) {
2856 		mtx_unlock(&pr->pr_mtx);
2857 		return (0);
2858 	}
2859 	if (pr->pr_ip4 == NULL) {
2860 		mtx_unlock(&pr->pr_mtx);
2861 		return (EAFNOSUPPORT);
2862 	}
2863 
2864 	ia->s_addr = pr->pr_ip4[0].s_addr;
2865 	mtx_unlock(&pr->pr_mtx);
2866 	return (0);
2867 }
2868 
2869 /*
2870  * Return 1 if we should do proper source address selection or are not jailed.
2871  * We will return 0 if we should bypass source address selection in favour
2872  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2873  * returned in NBO.
2874  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2875  */
2876 int
2877 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2878 {
2879 	struct prison *pr;
2880 	struct in_addr lia;
2881 	int error;
2882 
2883 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2884 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2885 
2886 	if (!jailed(cred))
2887 		return (1);
2888 
2889 	pr = cred->cr_prison;
2890 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2891 		return (1);
2892 
2893 	lia.s_addr = INADDR_ANY;
2894 	error = prison_get_ip4(cred, &lia);
2895 	if (error)
2896 		return (error);
2897 	if (lia.s_addr == INADDR_ANY)
2898 		return (1);
2899 
2900 	ia->s_addr = lia.s_addr;
2901 	return (0);
2902 }
2903 
2904 /*
2905  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2906  */
2907 int
2908 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2909 {
2910 
2911 	if (pr1 == pr2)
2912 		return (1);
2913 
2914 	/*
2915 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2916 	 * existing prisons.
2917 	 */
2918 	while (pr1 != &prison0 &&
2919 #ifdef VIMAGE
2920 	       !(pr1->pr_flags & PR_VNET) &&
2921 #endif
2922 	       !(pr1->pr_flags & PR_IP4_USER))
2923 		pr1 = pr1->pr_parent;
2924 	while (pr2 != &prison0 &&
2925 #ifdef VIMAGE
2926 	       !(pr2->pr_flags & PR_VNET) &&
2927 #endif
2928 	       !(pr2->pr_flags & PR_IP4_USER))
2929 		pr2 = pr2->pr_parent;
2930 	return (pr1 == pr2);
2931 }
2932 
2933 /*
2934  * Make sure our (source) address is set to something meaningful to this
2935  * jail.
2936  *
2937  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2938  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2939  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2940  */
2941 int
2942 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2943 {
2944 	struct prison *pr;
2945 	struct in_addr ia0;
2946 	int error;
2947 
2948 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2949 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2950 
2951 	pr = cred->cr_prison;
2952 	if (!(pr->pr_flags & PR_IP4))
2953 		return (0);
2954 	mtx_lock(&pr->pr_mtx);
2955 	if (!(pr->pr_flags & PR_IP4)) {
2956 		mtx_unlock(&pr->pr_mtx);
2957 		return (0);
2958 	}
2959 	if (pr->pr_ip4 == NULL) {
2960 		mtx_unlock(&pr->pr_mtx);
2961 		return (EAFNOSUPPORT);
2962 	}
2963 
2964 	ia0.s_addr = ntohl(ia->s_addr);
2965 	if (ia0.s_addr == INADDR_LOOPBACK) {
2966 		ia->s_addr = pr->pr_ip4[0].s_addr;
2967 		mtx_unlock(&pr->pr_mtx);
2968 		return (0);
2969 	}
2970 
2971 	if (ia0.s_addr == INADDR_ANY) {
2972 		/*
2973 		 * In case there is only 1 IPv4 address, bind directly.
2974 		 */
2975 		if (pr->pr_ip4s == 1)
2976 			ia->s_addr = pr->pr_ip4[0].s_addr;
2977 		mtx_unlock(&pr->pr_mtx);
2978 		return (0);
2979 	}
2980 
2981 	error = _prison_check_ip4(pr, ia);
2982 	mtx_unlock(&pr->pr_mtx);
2983 	return (error);
2984 }
2985 
2986 /*
2987  * Rewrite destination address in case we will connect to loopback address.
2988  *
2989  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2990  * Address passed in in NBO and returned in NBO.
2991  */
2992 int
2993 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2994 {
2995 	struct prison *pr;
2996 
2997 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2998 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2999 
3000 	pr = cred->cr_prison;
3001 	if (!(pr->pr_flags & PR_IP4))
3002 		return (0);
3003 	mtx_lock(&pr->pr_mtx);
3004 	if (!(pr->pr_flags & PR_IP4)) {
3005 		mtx_unlock(&pr->pr_mtx);
3006 		return (0);
3007 	}
3008 	if (pr->pr_ip4 == NULL) {
3009 		mtx_unlock(&pr->pr_mtx);
3010 		return (EAFNOSUPPORT);
3011 	}
3012 
3013 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
3014 		ia->s_addr = pr->pr_ip4[0].s_addr;
3015 		mtx_unlock(&pr->pr_mtx);
3016 		return (0);
3017 	}
3018 
3019 	/*
3020 	 * Return success because nothing had to be changed.
3021 	 */
3022 	mtx_unlock(&pr->pr_mtx);
3023 	return (0);
3024 }
3025 
3026 /*
3027  * Check if given address belongs to the jail referenced by cred/prison.
3028  *
3029  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
3030  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3031  * doesn't allow IPv4.  Address passed in in NBO.
3032  */
3033 static int
3034 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
3035 {
3036 	int i, a, z, d;
3037 
3038 	/*
3039 	 * Check the primary IP.
3040 	 */
3041 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
3042 		return (0);
3043 
3044 	/*
3045 	 * All the other IPs are sorted so we can do a binary search.
3046 	 */
3047 	a = 0;
3048 	z = pr->pr_ip4s - 2;
3049 	while (a <= z) {
3050 		i = (a + z) / 2;
3051 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3052 		if (d > 0)
3053 			z = i - 1;
3054 		else if (d < 0)
3055 			a = i + 1;
3056 		else
3057 			return (0);
3058 	}
3059 
3060 	return (EADDRNOTAVAIL);
3061 }
3062 
3063 int
3064 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
3065 {
3066 	struct prison *pr;
3067 	int error;
3068 
3069 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3070 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3071 
3072 	pr = cred->cr_prison;
3073 	if (!(pr->pr_flags & PR_IP4))
3074 		return (0);
3075 	mtx_lock(&pr->pr_mtx);
3076 	if (!(pr->pr_flags & PR_IP4)) {
3077 		mtx_unlock(&pr->pr_mtx);
3078 		return (0);
3079 	}
3080 	if (pr->pr_ip4 == NULL) {
3081 		mtx_unlock(&pr->pr_mtx);
3082 		return (EAFNOSUPPORT);
3083 	}
3084 
3085 	error = _prison_check_ip4(pr, ia);
3086 	mtx_unlock(&pr->pr_mtx);
3087 	return (error);
3088 }
3089 #endif
3090 
3091 #ifdef INET6
3092 static int
3093 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3094 {
3095 	int ii, ij, used;
3096 	struct prison *ppr;
3097 
3098 	ppr = pr->pr_parent;
3099 	if (!(pr->pr_flags & PR_IP6_USER)) {
3100 		/* This has no user settings, so just copy the parent's list. */
3101 		if (pr->pr_ip6s < ppr->pr_ip6s) {
3102 			/*
3103 			 * There's no room for the parent's list.  Use the
3104 			 * new list buffer, which is assumed to be big enough
3105 			 * (if it was passed).  If there's no buffer, try to
3106 			 * allocate one.
3107 			 */
3108 			used = 1;
3109 			if (newip6 == NULL) {
3110 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3111 				    M_PRISON, M_NOWAIT);
3112 				if (newip6 != NULL)
3113 					used = 0;
3114 			}
3115 			if (newip6 != NULL) {
3116 				bcopy(ppr->pr_ip6, newip6,
3117 				    ppr->pr_ip6s * sizeof(*newip6));
3118 				free(pr->pr_ip6, M_PRISON);
3119 				pr->pr_ip6 = newip6;
3120 				pr->pr_ip6s = ppr->pr_ip6s;
3121 			}
3122 			return (used);
3123 		}
3124 		pr->pr_ip6s = ppr->pr_ip6s;
3125 		if (pr->pr_ip6s > 0)
3126 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3127 			    pr->pr_ip6s * sizeof(*newip6));
3128 		else if (pr->pr_ip6 != NULL) {
3129 			free(pr->pr_ip6, M_PRISON);
3130 			pr->pr_ip6 = NULL;
3131 		}
3132 	} else if (pr->pr_ip6s > 0) {
3133 		/* Remove addresses that aren't in the parent. */
3134 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3135 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3136 			    &ppr->pr_ip6[ij]))
3137 				break;
3138 		if (ij < ppr->pr_ip6s)
3139 			ii = 1;
3140 		else {
3141 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3142 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3143 			ii = 0;
3144 		}
3145 		for (ij = 1; ii < pr->pr_ip6s; ) {
3146 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3147 			    &ppr->pr_ip6[0])) {
3148 				ii++;
3149 				continue;
3150 			}
3151 			switch (ij >= ppr->pr_ip6s ? -1 :
3152 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3153 			case -1:
3154 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3155 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3156 				break;
3157 			case 0:
3158 				ii++;
3159 				ij++;
3160 				break;
3161 			case 1:
3162 				ij++;
3163 				break;
3164 			}
3165 		}
3166 		if (pr->pr_ip6s == 0) {
3167 			free(pr->pr_ip6, M_PRISON);
3168 			pr->pr_ip6 = NULL;
3169 		}
3170 	}
3171 	return 0;
3172 }
3173 
3174 /*
3175  * Pass back primary IPv6 address for this jail.
3176  *
3177  * If not restricted return success but do not alter the address.  Caller has
3178  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3179  *
3180  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3181  */
3182 int
3183 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3184 {
3185 	struct prison *pr;
3186 
3187 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3188 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3189 
3190 	pr = cred->cr_prison;
3191 	if (!(pr->pr_flags & PR_IP6))
3192 		return (0);
3193 	mtx_lock(&pr->pr_mtx);
3194 	if (!(pr->pr_flags & PR_IP6)) {
3195 		mtx_unlock(&pr->pr_mtx);
3196 		return (0);
3197 	}
3198 	if (pr->pr_ip6 == NULL) {
3199 		mtx_unlock(&pr->pr_mtx);
3200 		return (EAFNOSUPPORT);
3201 	}
3202 
3203 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3204 	mtx_unlock(&pr->pr_mtx);
3205 	return (0);
3206 }
3207 
3208 /*
3209  * Return 1 if we should do proper source address selection or are not jailed.
3210  * We will return 0 if we should bypass source address selection in favour
3211  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3212  * returned in NBO.
3213  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3214  */
3215 int
3216 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3217 {
3218 	struct prison *pr;
3219 	struct in6_addr lia6;
3220 	int error;
3221 
3222 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3223 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3224 
3225 	if (!jailed(cred))
3226 		return (1);
3227 
3228 	pr = cred->cr_prison;
3229 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3230 		return (1);
3231 
3232 	lia6 = in6addr_any;
3233 	error = prison_get_ip6(cred, &lia6);
3234 	if (error)
3235 		return (error);
3236 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3237 		return (1);
3238 
3239 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3240 	return (0);
3241 }
3242 
3243 /*
3244  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3245  */
3246 int
3247 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3248 {
3249 
3250 	if (pr1 == pr2)
3251 		return (1);
3252 
3253 	while (pr1 != &prison0 &&
3254 #ifdef VIMAGE
3255 	       !(pr1->pr_flags & PR_VNET) &&
3256 #endif
3257 	       !(pr1->pr_flags & PR_IP6_USER))
3258 		pr1 = pr1->pr_parent;
3259 	while (pr2 != &prison0 &&
3260 #ifdef VIMAGE
3261 	       !(pr2->pr_flags & PR_VNET) &&
3262 #endif
3263 	       !(pr2->pr_flags & PR_IP6_USER))
3264 		pr2 = pr2->pr_parent;
3265 	return (pr1 == pr2);
3266 }
3267 
3268 /*
3269  * Make sure our (source) address is set to something meaningful to this jail.
3270  *
3271  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3272  * when needed while binding.
3273  *
3274  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3275  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3276  * doesn't allow IPv6.
3277  */
3278 int
3279 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3280 {
3281 	struct prison *pr;
3282 	int error;
3283 
3284 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3285 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3286 
3287 	pr = cred->cr_prison;
3288 	if (!(pr->pr_flags & PR_IP6))
3289 		return (0);
3290 	mtx_lock(&pr->pr_mtx);
3291 	if (!(pr->pr_flags & PR_IP6)) {
3292 		mtx_unlock(&pr->pr_mtx);
3293 		return (0);
3294 	}
3295 	if (pr->pr_ip6 == NULL) {
3296 		mtx_unlock(&pr->pr_mtx);
3297 		return (EAFNOSUPPORT);
3298 	}
3299 
3300 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3301 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3302 		mtx_unlock(&pr->pr_mtx);
3303 		return (0);
3304 	}
3305 
3306 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3307 		/*
3308 		 * In case there is only 1 IPv6 address, and v6only is true,
3309 		 * then bind directly.
3310 		 */
3311 		if (v6only != 0 && pr->pr_ip6s == 1)
3312 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3313 		mtx_unlock(&pr->pr_mtx);
3314 		return (0);
3315 	}
3316 
3317 	error = _prison_check_ip6(pr, ia6);
3318 	mtx_unlock(&pr->pr_mtx);
3319 	return (error);
3320 }
3321 
3322 /*
3323  * Rewrite destination address in case we will connect to loopback address.
3324  *
3325  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3326  */
3327 int
3328 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3329 {
3330 	struct prison *pr;
3331 
3332 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3333 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3334 
3335 	pr = cred->cr_prison;
3336 	if (!(pr->pr_flags & PR_IP6))
3337 		return (0);
3338 	mtx_lock(&pr->pr_mtx);
3339 	if (!(pr->pr_flags & PR_IP6)) {
3340 		mtx_unlock(&pr->pr_mtx);
3341 		return (0);
3342 	}
3343 	if (pr->pr_ip6 == NULL) {
3344 		mtx_unlock(&pr->pr_mtx);
3345 		return (EAFNOSUPPORT);
3346 	}
3347 
3348 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3349 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3350 		mtx_unlock(&pr->pr_mtx);
3351 		return (0);
3352 	}
3353 
3354 	/*
3355 	 * Return success because nothing had to be changed.
3356 	 */
3357 	mtx_unlock(&pr->pr_mtx);
3358 	return (0);
3359 }
3360 
3361 /*
3362  * Check if given address belongs to the jail referenced by cred/prison.
3363  *
3364  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3365  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3366  * doesn't allow IPv6.
3367  */
3368 static int
3369 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3370 {
3371 	int i, a, z, d;
3372 
3373 	/*
3374 	 * Check the primary IP.
3375 	 */
3376 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3377 		return (0);
3378 
3379 	/*
3380 	 * All the other IPs are sorted so we can do a binary search.
3381 	 */
3382 	a = 0;
3383 	z = pr->pr_ip6s - 2;
3384 	while (a <= z) {
3385 		i = (a + z) / 2;
3386 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3387 		if (d > 0)
3388 			z = i - 1;
3389 		else if (d < 0)
3390 			a = i + 1;
3391 		else
3392 			return (0);
3393 	}
3394 
3395 	return (EADDRNOTAVAIL);
3396 }
3397 
3398 int
3399 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3400 {
3401 	struct prison *pr;
3402 	int error;
3403 
3404 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3405 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3406 
3407 	pr = cred->cr_prison;
3408 	if (!(pr->pr_flags & PR_IP6))
3409 		return (0);
3410 	mtx_lock(&pr->pr_mtx);
3411 	if (!(pr->pr_flags & PR_IP6)) {
3412 		mtx_unlock(&pr->pr_mtx);
3413 		return (0);
3414 	}
3415 	if (pr->pr_ip6 == NULL) {
3416 		mtx_unlock(&pr->pr_mtx);
3417 		return (EAFNOSUPPORT);
3418 	}
3419 
3420 	error = _prison_check_ip6(pr, ia6);
3421 	mtx_unlock(&pr->pr_mtx);
3422 	return (error);
3423 }
3424 #endif
3425 
3426 /*
3427  * Check if a jail supports the given address family.
3428  *
3429  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3430  * if not.
3431  */
3432 int
3433 prison_check_af(struct ucred *cred, int af)
3434 {
3435 	struct prison *pr;
3436 	int error;
3437 
3438 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3439 
3440 	pr = cred->cr_prison;
3441 #ifdef VIMAGE
3442 	/* Prisons with their own network stack are not limited. */
3443 	if (prison_owns_vnet(cred))
3444 		return (0);
3445 #endif
3446 
3447 	error = 0;
3448 	switch (af)
3449 	{
3450 #ifdef INET
3451 	case AF_INET:
3452 		if (pr->pr_flags & PR_IP4)
3453 		{
3454 			mtx_lock(&pr->pr_mtx);
3455 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3456 				error = EAFNOSUPPORT;
3457 			mtx_unlock(&pr->pr_mtx);
3458 		}
3459 		break;
3460 #endif
3461 #ifdef INET6
3462 	case AF_INET6:
3463 		if (pr->pr_flags & PR_IP6)
3464 		{
3465 			mtx_lock(&pr->pr_mtx);
3466 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3467 				error = EAFNOSUPPORT;
3468 			mtx_unlock(&pr->pr_mtx);
3469 		}
3470 		break;
3471 #endif
3472 	case AF_LOCAL:
3473 	case AF_ROUTE:
3474 		break;
3475 	default:
3476 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3477 			error = EAFNOSUPPORT;
3478 	}
3479 	return (error);
3480 }
3481 
3482 /*
3483  * Check if given address belongs to the jail referenced by cred (wrapper to
3484  * prison_check_ip[46]).
3485  *
3486  * Returns 0 if jail doesn't restrict the address family or if address belongs
3487  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3488  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3489  */
3490 int
3491 prison_if(struct ucred *cred, struct sockaddr *sa)
3492 {
3493 #ifdef INET
3494 	struct sockaddr_in *sai;
3495 #endif
3496 #ifdef INET6
3497 	struct sockaddr_in6 *sai6;
3498 #endif
3499 	int error;
3500 
3501 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3502 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3503 
3504 #ifdef VIMAGE
3505 	if (prison_owns_vnet(cred))
3506 		return (0);
3507 #endif
3508 
3509 	error = 0;
3510 	switch (sa->sa_family)
3511 	{
3512 #ifdef INET
3513 	case AF_INET:
3514 		sai = (struct sockaddr_in *)sa;
3515 		error = prison_check_ip4(cred, &sai->sin_addr);
3516 		break;
3517 #endif
3518 #ifdef INET6
3519 	case AF_INET6:
3520 		sai6 = (struct sockaddr_in6 *)sa;
3521 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3522 		break;
3523 #endif
3524 	default:
3525 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3526 			error = EAFNOSUPPORT;
3527 	}
3528 	return (error);
3529 }
3530 
3531 /*
3532  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3533  */
3534 int
3535 prison_check(struct ucred *cred1, struct ucred *cred2)
3536 {
3537 
3538 	return ((cred1->cr_prison == cred2->cr_prison ||
3539 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3540 }
3541 
3542 /*
3543  * Return 1 if p2 is a child of p1, otherwise 0.
3544  */
3545 int
3546 prison_ischild(struct prison *pr1, struct prison *pr2)
3547 {
3548 
3549 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3550 		if (pr1 == pr2)
3551 			return (1);
3552 	return (0);
3553 }
3554 
3555 /*
3556  * Return 1 if the passed credential is in a jail, otherwise 0.
3557  */
3558 int
3559 jailed(struct ucred *cred)
3560 {
3561 
3562 	return (cred->cr_prison != &prison0);
3563 }
3564 
3565 /*
3566  * Return 1 if the passed credential is in a jail and that jail does not
3567  * have its own virtual network stack, otherwise 0.
3568  */
3569 int
3570 jailed_without_vnet(struct ucred *cred)
3571 {
3572 
3573 	if (!jailed(cred))
3574 		return (0);
3575 #ifdef VIMAGE
3576 	if (prison_owns_vnet(cred))
3577 		return (0);
3578 #endif
3579 
3580 	return (1);
3581 }
3582 
3583 /*
3584  * Return the correct hostname (domainname, et al) for the passed credential.
3585  */
3586 void
3587 getcredhostname(struct ucred *cred, char *buf, size_t size)
3588 {
3589 	struct prison *pr;
3590 
3591 	/*
3592 	 * A NULL credential can be used to shortcut to the physical
3593 	 * system's hostname.
3594 	 */
3595 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3596 	mtx_lock(&pr->pr_mtx);
3597 	strlcpy(buf, pr->pr_hostname, size);
3598 	mtx_unlock(&pr->pr_mtx);
3599 }
3600 
3601 void
3602 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3603 {
3604 
3605 	mtx_lock(&cred->cr_prison->pr_mtx);
3606 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3607 	mtx_unlock(&cred->cr_prison->pr_mtx);
3608 }
3609 
3610 void
3611 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3612 {
3613 
3614 	mtx_lock(&cred->cr_prison->pr_mtx);
3615 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3616 	mtx_unlock(&cred->cr_prison->pr_mtx);
3617 }
3618 
3619 void
3620 getcredhostid(struct ucred *cred, unsigned long *hostid)
3621 {
3622 
3623 	mtx_lock(&cred->cr_prison->pr_mtx);
3624 	*hostid = cred->cr_prison->pr_hostid;
3625 	mtx_unlock(&cred->cr_prison->pr_mtx);
3626 }
3627 
3628 #ifdef VIMAGE
3629 /*
3630  * Determine whether the prison represented by cred owns
3631  * its vnet rather than having it inherited.
3632  *
3633  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3634  */
3635 int
3636 prison_owns_vnet(struct ucred *cred)
3637 {
3638 
3639 	/*
3640 	 * vnets cannot be added/removed after jail creation,
3641 	 * so no need to lock here.
3642 	 */
3643 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3644 }
3645 #endif
3646 
3647 /*
3648  * Determine whether the subject represented by cred can "see"
3649  * status of a mount point.
3650  * Returns: 0 for permitted, ENOENT otherwise.
3651  * XXX: This function should be called cr_canseemount() and should be
3652  *      placed in kern_prot.c.
3653  */
3654 int
3655 prison_canseemount(struct ucred *cred, struct mount *mp)
3656 {
3657 	struct prison *pr;
3658 	struct statfs *sp;
3659 	size_t len;
3660 
3661 	pr = cred->cr_prison;
3662 	if (pr->pr_enforce_statfs == 0)
3663 		return (0);
3664 	if (pr->pr_root->v_mount == mp)
3665 		return (0);
3666 	if (pr->pr_enforce_statfs == 2)
3667 		return (ENOENT);
3668 	/*
3669 	 * If jail's chroot directory is set to "/" we should be able to see
3670 	 * all mount-points from inside a jail.
3671 	 * This is ugly check, but this is the only situation when jail's
3672 	 * directory ends with '/'.
3673 	 */
3674 	if (strcmp(pr->pr_path, "/") == 0)
3675 		return (0);
3676 	len = strlen(pr->pr_path);
3677 	sp = &mp->mnt_stat;
3678 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3679 		return (ENOENT);
3680 	/*
3681 	 * Be sure that we don't have situation where jail's root directory
3682 	 * is "/some/path" and mount point is "/some/pathpath".
3683 	 */
3684 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3685 		return (ENOENT);
3686 	return (0);
3687 }
3688 
3689 void
3690 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3691 {
3692 	char jpath[MAXPATHLEN];
3693 	struct prison *pr;
3694 	size_t len;
3695 
3696 	pr = cred->cr_prison;
3697 	if (pr->pr_enforce_statfs == 0)
3698 		return;
3699 	if (prison_canseemount(cred, mp) != 0) {
3700 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3701 		strlcpy(sp->f_mntonname, "[restricted]",
3702 		    sizeof(sp->f_mntonname));
3703 		return;
3704 	}
3705 	if (pr->pr_root->v_mount == mp) {
3706 		/*
3707 		 * Clear current buffer data, so we are sure nothing from
3708 		 * the valid path left there.
3709 		 */
3710 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3711 		*sp->f_mntonname = '/';
3712 		return;
3713 	}
3714 	/*
3715 	 * If jail's chroot directory is set to "/" we should be able to see
3716 	 * all mount-points from inside a jail.
3717 	 */
3718 	if (strcmp(pr->pr_path, "/") == 0)
3719 		return;
3720 	len = strlen(pr->pr_path);
3721 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3722 	/*
3723 	 * Clear current buffer data, so we are sure nothing from
3724 	 * the valid path left there.
3725 	 */
3726 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3727 	if (*jpath == '\0') {
3728 		/* Should never happen. */
3729 		*sp->f_mntonname = '/';
3730 	} else {
3731 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3732 	}
3733 }
3734 
3735 /*
3736  * Check with permission for a specific privilege is granted within jail.  We
3737  * have a specific list of accepted privileges; the rest are denied.
3738  */
3739 int
3740 prison_priv_check(struct ucred *cred, int priv)
3741 {
3742 
3743 	if (!jailed(cred))
3744 		return (0);
3745 
3746 #ifdef VIMAGE
3747 	/*
3748 	 * Privileges specific to prisons with a virtual network stack.
3749 	 * There might be a duplicate entry here in case the privilege
3750 	 * is only granted conditionally in the legacy jail case.
3751 	 */
3752 	switch (priv) {
3753 #ifdef notyet
3754 		/*
3755 		 * NFS-specific privileges.
3756 		 */
3757 	case PRIV_NFS_DAEMON:
3758 	case PRIV_NFS_LOCKD:
3759 #endif
3760 		/*
3761 		 * Network stack privileges.
3762 		 */
3763 	case PRIV_NET_BRIDGE:
3764 	case PRIV_NET_GRE:
3765 	case PRIV_NET_BPF:
3766 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3767 	case PRIV_NET_ROUTE:
3768 	case PRIV_NET_TAP:
3769 	case PRIV_NET_SETIFMTU:
3770 	case PRIV_NET_SETIFFLAGS:
3771 	case PRIV_NET_SETIFCAP:
3772 	case PRIV_NET_SETIFDESCR:
3773 	case PRIV_NET_SETIFNAME	:
3774 	case PRIV_NET_SETIFMETRIC:
3775 	case PRIV_NET_SETIFPHYS:
3776 	case PRIV_NET_SETIFMAC:
3777 	case PRIV_NET_ADDMULTI:
3778 	case PRIV_NET_DELMULTI:
3779 	case PRIV_NET_HWIOCTL:
3780 	case PRIV_NET_SETLLADDR:
3781 	case PRIV_NET_ADDIFGROUP:
3782 	case PRIV_NET_DELIFGROUP:
3783 	case PRIV_NET_IFCREATE:
3784 	case PRIV_NET_IFDESTROY:
3785 	case PRIV_NET_ADDIFADDR:
3786 	case PRIV_NET_DELIFADDR:
3787 	case PRIV_NET_LAGG:
3788 	case PRIV_NET_GIF:
3789 	case PRIV_NET_SETIFVNET:
3790 	case PRIV_NET_SETIFFIB:
3791 
3792 		/*
3793 		 * 802.11-related privileges.
3794 		 */
3795 	case PRIV_NET80211_GETKEY:
3796 #ifdef notyet
3797 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3798 #endif
3799 
3800 #ifdef notyet
3801 		/*
3802 		 * ATM privileges.
3803 		 */
3804 	case PRIV_NETATM_CFG:
3805 	case PRIV_NETATM_ADD:
3806 	case PRIV_NETATM_DEL:
3807 	case PRIV_NETATM_SET:
3808 
3809 		/*
3810 		 * Bluetooth privileges.
3811 		 */
3812 	case PRIV_NETBLUETOOTH_RAW:
3813 #endif
3814 
3815 		/*
3816 		 * Netgraph and netgraph module privileges.
3817 		 */
3818 	case PRIV_NETGRAPH_CONTROL:
3819 #ifdef notyet
3820 	case PRIV_NETGRAPH_TTY:
3821 #endif
3822 
3823 		/*
3824 		 * IPv4 and IPv6 privileges.
3825 		 */
3826 	case PRIV_NETINET_IPFW:
3827 	case PRIV_NETINET_DIVERT:
3828 	case PRIV_NETINET_PF:
3829 	case PRIV_NETINET_DUMMYNET:
3830 	case PRIV_NETINET_CARP:
3831 	case PRIV_NETINET_MROUTE:
3832 	case PRIV_NETINET_RAW:
3833 	case PRIV_NETINET_ADDRCTRL6:
3834 	case PRIV_NETINET_ND6:
3835 	case PRIV_NETINET_SCOPE6:
3836 	case PRIV_NETINET_ALIFETIME6:
3837 	case PRIV_NETINET_IPSEC:
3838 	case PRIV_NETINET_BINDANY:
3839 
3840 #ifdef notyet
3841 		/*
3842 		 * NCP privileges.
3843 		 */
3844 	case PRIV_NETNCP:
3845 
3846 		/*
3847 		 * SMB privileges.
3848 		 */
3849 	case PRIV_NETSMB:
3850 #endif
3851 
3852 	/*
3853 	 * No default: or deny here.
3854 	 * In case of no permit fall through to next switch().
3855 	 */
3856 		if (cred->cr_prison->pr_flags & PR_VNET)
3857 			return (0);
3858 	}
3859 #endif /* VIMAGE */
3860 
3861 	switch (priv) {
3862 
3863 		/*
3864 		 * Allow ktrace privileges for root in jail.
3865 		 */
3866 	case PRIV_KTRACE:
3867 
3868 #if 0
3869 		/*
3870 		 * Allow jailed processes to configure audit identity and
3871 		 * submit audit records (login, etc).  In the future we may
3872 		 * want to further refine the relationship between audit and
3873 		 * jail.
3874 		 */
3875 	case PRIV_AUDIT_GETAUDIT:
3876 	case PRIV_AUDIT_SETAUDIT:
3877 	case PRIV_AUDIT_SUBMIT:
3878 #endif
3879 
3880 		/*
3881 		 * Allow jailed processes to manipulate process UNIX
3882 		 * credentials in any way they see fit.
3883 		 */
3884 	case PRIV_CRED_SETUID:
3885 	case PRIV_CRED_SETEUID:
3886 	case PRIV_CRED_SETGID:
3887 	case PRIV_CRED_SETEGID:
3888 	case PRIV_CRED_SETGROUPS:
3889 	case PRIV_CRED_SETREUID:
3890 	case PRIV_CRED_SETREGID:
3891 	case PRIV_CRED_SETRESUID:
3892 	case PRIV_CRED_SETRESGID:
3893 
3894 		/*
3895 		 * Jail implements visibility constraints already, so allow
3896 		 * jailed root to override uid/gid-based constraints.
3897 		 */
3898 	case PRIV_SEEOTHERGIDS:
3899 	case PRIV_SEEOTHERUIDS:
3900 
3901 		/*
3902 		 * Jail implements inter-process debugging limits already, so
3903 		 * allow jailed root various debugging privileges.
3904 		 */
3905 	case PRIV_DEBUG_DIFFCRED:
3906 	case PRIV_DEBUG_SUGID:
3907 	case PRIV_DEBUG_UNPRIV:
3908 
3909 		/*
3910 		 * Allow jail to set various resource limits and login
3911 		 * properties, and for now, exceed process resource limits.
3912 		 */
3913 	case PRIV_PROC_LIMIT:
3914 	case PRIV_PROC_SETLOGIN:
3915 	case PRIV_PROC_SETRLIMIT:
3916 
3917 		/*
3918 		 * System V and POSIX IPC privileges are granted in jail.
3919 		 */
3920 	case PRIV_IPC_READ:
3921 	case PRIV_IPC_WRITE:
3922 	case PRIV_IPC_ADMIN:
3923 	case PRIV_IPC_MSGSIZE:
3924 	case PRIV_MQ_ADMIN:
3925 
3926 		/*
3927 		 * Jail operations within a jail work on child jails.
3928 		 */
3929 	case PRIV_JAIL_ATTACH:
3930 	case PRIV_JAIL_SET:
3931 	case PRIV_JAIL_REMOVE:
3932 
3933 		/*
3934 		 * Jail implements its own inter-process limits, so allow
3935 		 * root processes in jail to change scheduling on other
3936 		 * processes in the same jail.  Likewise for signalling.
3937 		 */
3938 	case PRIV_SCHED_DIFFCRED:
3939 	case PRIV_SCHED_CPUSET:
3940 	case PRIV_SIGNAL_DIFFCRED:
3941 	case PRIV_SIGNAL_SUGID:
3942 
3943 		/*
3944 		 * Allow jailed processes to write to sysctls marked as jail
3945 		 * writable.
3946 		 */
3947 	case PRIV_SYSCTL_WRITEJAIL:
3948 
3949 		/*
3950 		 * Allow root in jail to manage a variety of quota
3951 		 * properties.  These should likely be conditional on a
3952 		 * configuration option.
3953 		 */
3954 	case PRIV_VFS_GETQUOTA:
3955 	case PRIV_VFS_SETQUOTA:
3956 
3957 		/*
3958 		 * Since Jail relies on chroot() to implement file system
3959 		 * protections, grant many VFS privileges to root in jail.
3960 		 * Be careful to exclude mount-related and NFS-related
3961 		 * privileges.
3962 		 */
3963 	case PRIV_VFS_READ:
3964 	case PRIV_VFS_WRITE:
3965 	case PRIV_VFS_ADMIN:
3966 	case PRIV_VFS_EXEC:
3967 	case PRIV_VFS_LOOKUP:
3968 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3969 	case PRIV_VFS_CHFLAGS_DEV:
3970 	case PRIV_VFS_CHOWN:
3971 	case PRIV_VFS_CHROOT:
3972 	case PRIV_VFS_RETAINSUGID:
3973 	case PRIV_VFS_FCHROOT:
3974 	case PRIV_VFS_LINK:
3975 	case PRIV_VFS_SETGID:
3976 	case PRIV_VFS_STAT:
3977 	case PRIV_VFS_STICKYFILE:
3978 
3979 		/*
3980 		 * As in the non-jail case, non-root users are expected to be
3981 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3982 		 * exists in the jail and they have permission to access it).
3983 		 */
3984 	case PRIV_KMEM_READ:
3985 		return (0);
3986 
3987 		/*
3988 		 * Depending on the global setting, allow privilege of
3989 		 * setting system flags.
3990 		 */
3991 	case PRIV_VFS_SYSFLAGS:
3992 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3993 			return (0);
3994 		else
3995 			return (EPERM);
3996 
3997 		/*
3998 		 * Depending on the global setting, allow privilege of
3999 		 * mounting/unmounting file systems.
4000 		 */
4001 	case PRIV_VFS_MOUNT:
4002 	case PRIV_VFS_UNMOUNT:
4003 	case PRIV_VFS_MOUNT_NONUSER:
4004 	case PRIV_VFS_MOUNT_OWNER:
4005 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
4006 		    cred->cr_prison->pr_enforce_statfs < 2)
4007 			return (0);
4008 		else
4009 			return (EPERM);
4010 
4011 		/*
4012 		 * Allow jailed root to bind reserved ports and reuse in-use
4013 		 * ports.
4014 		 */
4015 	case PRIV_NETINET_RESERVEDPORT:
4016 	case PRIV_NETINET_REUSEPORT:
4017 		return (0);
4018 
4019 		/*
4020 		 * Allow jailed root to set certian IPv4/6 (option) headers.
4021 		 */
4022 	case PRIV_NETINET_SETHDROPTS:
4023 		return (0);
4024 
4025 		/*
4026 		 * Conditionally allow creating raw sockets in jail.
4027 		 */
4028 	case PRIV_NETINET_RAW:
4029 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4030 			return (0);
4031 		else
4032 			return (EPERM);
4033 
4034 		/*
4035 		 * Since jail implements its own visibility limits on netstat
4036 		 * sysctls, allow getcred.  This allows identd to work in
4037 		 * jail.
4038 		 */
4039 	case PRIV_NETINET_GETCRED:
4040 		return (0);
4041 
4042 		/*
4043 		 * Allow jailed root to set loginclass.
4044 		 */
4045 	case PRIV_PROC_SETLOGINCLASS:
4046 		return (0);
4047 
4048 	default:
4049 		/*
4050 		 * In all remaining cases, deny the privilege request.  This
4051 		 * includes almost all network privileges, many system
4052 		 * configuration privileges.
4053 		 */
4054 		return (EPERM);
4055 	}
4056 }
4057 
4058 /*
4059  * Return the part of pr2's name that is relative to pr1, or the whole name
4060  * if it does not directly follow.
4061  */
4062 
4063 char *
4064 prison_name(struct prison *pr1, struct prison *pr2)
4065 {
4066 	char *name;
4067 
4068 	/* Jails see themselves as "0" (if they see themselves at all). */
4069 	if (pr1 == pr2)
4070 		return "0";
4071 	name = pr2->pr_name;
4072 	if (prison_ischild(pr1, pr2)) {
4073 		/*
4074 		 * pr1 isn't locked (and allprison_lock may not be either)
4075 		 * so its length can't be counted on.  But the number of dots
4076 		 * can be counted on - and counted.
4077 		 */
4078 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4079 			name = strchr(name, '.') + 1;
4080 	}
4081 	return (name);
4082 }
4083 
4084 /*
4085  * Return the part of pr2's path that is relative to pr1, or the whole path
4086  * if it does not directly follow.
4087  */
4088 static char *
4089 prison_path(struct prison *pr1, struct prison *pr2)
4090 {
4091 	char *path1, *path2;
4092 	int len1;
4093 
4094 	path1 = pr1->pr_path;
4095 	path2 = pr2->pr_path;
4096 	if (!strcmp(path1, "/"))
4097 		return (path2);
4098 	len1 = strlen(path1);
4099 	if (strncmp(path1, path2, len1))
4100 		return (path2);
4101 	if (path2[len1] == '\0')
4102 		return "/";
4103 	if (path2[len1] == '/')
4104 		return (path2 + len1);
4105 	return (path2);
4106 }
4107 
4108 
4109 /*
4110  * Jail-related sysctls.
4111  */
4112 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4113     "Jails");
4114 
4115 static int
4116 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4117 {
4118 	struct xprison *xp;
4119 	struct prison *pr, *cpr;
4120 #ifdef INET
4121 	struct in_addr *ip4 = NULL;
4122 	int ip4s = 0;
4123 #endif
4124 #ifdef INET6
4125 	struct in6_addr *ip6 = NULL;
4126 	int ip6s = 0;
4127 #endif
4128 	int descend, error;
4129 
4130 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4131 	pr = req->td->td_ucred->cr_prison;
4132 	error = 0;
4133 	sx_slock(&allprison_lock);
4134 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4135 #if defined(INET) || defined(INET6)
4136  again:
4137 #endif
4138 		mtx_lock(&cpr->pr_mtx);
4139 #ifdef INET
4140 		if (cpr->pr_ip4s > 0) {
4141 			if (ip4s < cpr->pr_ip4s) {
4142 				ip4s = cpr->pr_ip4s;
4143 				mtx_unlock(&cpr->pr_mtx);
4144 				ip4 = realloc(ip4, ip4s *
4145 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4146 				goto again;
4147 			}
4148 			bcopy(cpr->pr_ip4, ip4,
4149 			    cpr->pr_ip4s * sizeof(struct in_addr));
4150 		}
4151 #endif
4152 #ifdef INET6
4153 		if (cpr->pr_ip6s > 0) {
4154 			if (ip6s < cpr->pr_ip6s) {
4155 				ip6s = cpr->pr_ip6s;
4156 				mtx_unlock(&cpr->pr_mtx);
4157 				ip6 = realloc(ip6, ip6s *
4158 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4159 				goto again;
4160 			}
4161 			bcopy(cpr->pr_ip6, ip6,
4162 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4163 		}
4164 #endif
4165 		if (cpr->pr_ref == 0) {
4166 			mtx_unlock(&cpr->pr_mtx);
4167 			continue;
4168 		}
4169 		bzero(xp, sizeof(*xp));
4170 		xp->pr_version = XPRISON_VERSION;
4171 		xp->pr_id = cpr->pr_id;
4172 		xp->pr_state = cpr->pr_uref > 0
4173 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4174 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4175 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4176 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4177 #ifdef INET
4178 		xp->pr_ip4s = cpr->pr_ip4s;
4179 #endif
4180 #ifdef INET6
4181 		xp->pr_ip6s = cpr->pr_ip6s;
4182 #endif
4183 		mtx_unlock(&cpr->pr_mtx);
4184 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4185 		if (error)
4186 			break;
4187 #ifdef INET
4188 		if (xp->pr_ip4s > 0) {
4189 			error = SYSCTL_OUT(req, ip4,
4190 			    xp->pr_ip4s * sizeof(struct in_addr));
4191 			if (error)
4192 				break;
4193 		}
4194 #endif
4195 #ifdef INET6
4196 		if (xp->pr_ip6s > 0) {
4197 			error = SYSCTL_OUT(req, ip6,
4198 			    xp->pr_ip6s * sizeof(struct in6_addr));
4199 			if (error)
4200 				break;
4201 		}
4202 #endif
4203 	}
4204 	sx_sunlock(&allprison_lock);
4205 	free(xp, M_TEMP);
4206 #ifdef INET
4207 	free(ip4, M_TEMP);
4208 #endif
4209 #ifdef INET6
4210 	free(ip6, M_TEMP);
4211 #endif
4212 	return (error);
4213 }
4214 
4215 SYSCTL_OID(_security_jail, OID_AUTO, list,
4216     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4217     sysctl_jail_list, "S", "List of active jails");
4218 
4219 static int
4220 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4221 {
4222 	int error, injail;
4223 
4224 	injail = jailed(req->td->td_ucred);
4225 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4226 
4227 	return (error);
4228 }
4229 
4230 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4231     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4232     sysctl_jail_jailed, "I", "Process in jail?");
4233 
4234 static int
4235 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4236 {
4237 	int error, havevnet;
4238 #ifdef VIMAGE
4239 	struct ucred *cred = req->td->td_ucred;
4240 
4241 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4242 #else
4243 	havevnet = 0;
4244 #endif
4245 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4246 
4247 	return (error);
4248 }
4249 
4250 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4251     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4252     sysctl_jail_vnet, "I", "Jail owns VNET?");
4253 
4254 #if defined(INET) || defined(INET6)
4255 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4256     &jail_max_af_ips, 0,
4257     "Number of IP addresses a jail may have at most per address family");
4258 #endif
4259 
4260 /*
4261  * Default parameters for jail(2) compatability.  For historical reasons,
4262  * the sysctl names have varying similarity to the parameter names.  Prisons
4263  * just see their own parameters, and can't change them.
4264  */
4265 static int
4266 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4267 {
4268 	struct prison *pr;
4269 	int allow, error, i;
4270 
4271 	pr = req->td->td_ucred->cr_prison;
4272 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4273 
4274 	/* Get the current flag value, and convert it to a boolean. */
4275 	i = (allow & arg2) ? 1 : 0;
4276 	if (arg1 != NULL)
4277 		i = !i;
4278 	error = sysctl_handle_int(oidp, &i, 0, req);
4279 	if (error || !req->newptr)
4280 		return (error);
4281 	i = i ? arg2 : 0;
4282 	if (arg1 != NULL)
4283 		i ^= arg2;
4284 	/*
4285 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4286 	 * for writing.
4287 	 */
4288 	mtx_lock(&prison0.pr_mtx);
4289 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4290 	mtx_unlock(&prison0.pr_mtx);
4291 	return (0);
4292 }
4293 
4294 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4295     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4296     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4297     "Processes in jail can set their hostnames");
4298 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4299     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4300     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4301     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4302 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4303     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4304     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4305     "Processes in jail can use System V IPC primitives");
4306 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4307     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4308     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4309     "Prison root can create raw sockets");
4310 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4311     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4312     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4313     "Processes in jail can alter system file flags");
4314 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4315     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4316     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4317     "Processes in jail can mount/unmount jail-friendly file systems");
4318 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4319     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4320     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4321     "Processes in jail can mount the devfs file system");
4322 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4323     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4324     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4325     "Processes in jail can mount the fdescfs file system");
4326 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4327     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4328     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4329     "Processes in jail can mount the nullfs file system");
4330 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4331     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4332     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4333     "Processes in jail can mount the procfs file system");
4334 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
4335     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4336     NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4337     "Processes in jail can mount the linprocfs file system");
4338 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4339     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4340     NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4341     "Processes in jail can mount the linsysfs file system");
4342 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4343     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4344     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4345     "Processes in jail can mount the tmpfs file system");
4346 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4347     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4348     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4349     "Processes in jail can mount the zfs file system");
4350 
4351 static int
4352 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4353 {
4354 	struct prison *pr;
4355 	int level, error;
4356 
4357 	pr = req->td->td_ucred->cr_prison;
4358 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4359 	error = sysctl_handle_int(oidp, &level, 0, req);
4360 	if (error || !req->newptr)
4361 		return (error);
4362 	*(int *)arg1 = level;
4363 	return (0);
4364 }
4365 
4366 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4367     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4368     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4369     sysctl_jail_default_level, "I",
4370     "Processes in jail cannot see all mounted file systems");
4371 
4372 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4373     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4374     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4375     sysctl_jail_default_level, "I",
4376     "Ruleset for the devfs filesystem in jail");
4377 
4378 /*
4379  * Nodes to describe jail parameters.  Maximum length of string parameters
4380  * is returned in the string itself, and the other parameters exist merely
4381  * to make themselves and their types known.
4382  */
4383 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4384     "Jail parameters");
4385 
4386 int
4387 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4388 {
4389 	int i;
4390 	long l;
4391 	size_t s;
4392 	char numbuf[12];
4393 
4394 	switch (oidp->oid_kind & CTLTYPE)
4395 	{
4396 	case CTLTYPE_LONG:
4397 	case CTLTYPE_ULONG:
4398 		l = 0;
4399 #ifdef SCTL_MASK32
4400 		if (!(req->flags & SCTL_MASK32))
4401 #endif
4402 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4403 	case CTLTYPE_INT:
4404 	case CTLTYPE_UINT:
4405 		i = 0;
4406 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4407 	case CTLTYPE_STRING:
4408 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4409 		return
4410 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4411 	case CTLTYPE_STRUCT:
4412 		s = (size_t)arg2;
4413 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4414 	}
4415 	return (0);
4416 }
4417 
4418 /*
4419  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4420  * jail creation time but cannot be changed in an existing jail.
4421  */
4422 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4423 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4424 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4425 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4426 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4427     "I", "Jail secure level");
4428 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4429     "Jail value for kern.osreldate and uname -K");
4430 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4431     "Jail value for kern.osrelease and uname -r");
4432 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4433     "I", "Jail cannot see all mounted file systems");
4434 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4435     "I", "Ruleset for in-jail devfs mounts");
4436 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4437     "B", "Jail persistence");
4438 #ifdef VIMAGE
4439 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4440     "E,jailsys", "Virtual network stack");
4441 #endif
4442 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4443     "B", "Jail is in the process of shutting down");
4444 
4445 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4446 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4447     "I", "Current number of child jails");
4448 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4449     "I", "Maximum number of child jails");
4450 
4451 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4452 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4453     "Jail hostname");
4454 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4455     "Jail NIS domainname");
4456 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4457     "Jail host UUID");
4458 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4459     "LU", "Jail host ID");
4460 
4461 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4462 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4463 
4464 #ifdef INET
4465 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4466     "Jail IPv4 address virtualization");
4467 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4468     "S,in_addr,a", "Jail IPv4 addresses");
4469 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4470     "B", "Do (not) use IPv4 source address selection rather than the "
4471     "primary jail IPv4 address.");
4472 #endif
4473 #ifdef INET6
4474 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4475     "Jail IPv6 address virtualization");
4476 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4477     "S,in6_addr,a", "Jail IPv6 addresses");
4478 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4479     "B", "Do (not) use IPv6 source address selection rather than the "
4480     "primary jail IPv6 address.");
4481 #endif
4482 
4483 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4484 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4485     "B", "Jail may set hostname");
4486 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4487     "B", "Jail may use SYSV IPC");
4488 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4489     "B", "Jail may create raw sockets");
4490 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4491     "B", "Jail may alter system file flags");
4492 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4493     "B", "Jail may set file quotas");
4494 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4495     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4496 
4497 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4498 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4499     "B", "Jail may mount/unmount jail-friendly file systems in general");
4500 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4501     "B", "Jail may mount the devfs file system");
4502 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4503     "B", "Jail may mount the fdescfs file system");
4504 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4505     "B", "Jail may mount the nullfs file system");
4506 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4507     "B", "Jail may mount the procfs file system");
4508 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4509     "B", "Jail may mount the linprocfs file system");
4510 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4511     "B", "Jail may mount the linsysfs file system");
4512 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4513     "B", "Jail may mount the tmpfs file system");
4514 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4515     "B", "Jail may mount the zfs file system");
4516 
4517 #ifdef RACCT
4518 void
4519 prison_racct_foreach(void (*callback)(struct racct *racct,
4520     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4521     void *arg2, void *arg3)
4522 {
4523 	struct prison_racct *prr;
4524 
4525 	ASSERT_RACCT_ENABLED();
4526 
4527 	sx_slock(&allprison_lock);
4528 	if (pre != NULL)
4529 		(pre)();
4530 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4531 		(callback)(prr->prr_racct, arg2, arg3);
4532 	if (post != NULL)
4533 		(post)();
4534 	sx_sunlock(&allprison_lock);
4535 }
4536 
4537 static struct prison_racct *
4538 prison_racct_find_locked(const char *name)
4539 {
4540 	struct prison_racct *prr;
4541 
4542 	ASSERT_RACCT_ENABLED();
4543 	sx_assert(&allprison_lock, SA_XLOCKED);
4544 
4545 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4546 		return (NULL);
4547 
4548 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4549 		if (strcmp(name, prr->prr_name) != 0)
4550 			continue;
4551 
4552 		/* Found prison_racct with a matching name? */
4553 		prison_racct_hold(prr);
4554 		return (prr);
4555 	}
4556 
4557 	/* Add new prison_racct. */
4558 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4559 	racct_create(&prr->prr_racct);
4560 
4561 	strcpy(prr->prr_name, name);
4562 	refcount_init(&prr->prr_refcount, 1);
4563 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4564 
4565 	return (prr);
4566 }
4567 
4568 struct prison_racct *
4569 prison_racct_find(const char *name)
4570 {
4571 	struct prison_racct *prr;
4572 
4573 	ASSERT_RACCT_ENABLED();
4574 
4575 	sx_xlock(&allprison_lock);
4576 	prr = prison_racct_find_locked(name);
4577 	sx_xunlock(&allprison_lock);
4578 	return (prr);
4579 }
4580 
4581 void
4582 prison_racct_hold(struct prison_racct *prr)
4583 {
4584 
4585 	ASSERT_RACCT_ENABLED();
4586 
4587 	refcount_acquire(&prr->prr_refcount);
4588 }
4589 
4590 static void
4591 prison_racct_free_locked(struct prison_racct *prr)
4592 {
4593 
4594 	ASSERT_RACCT_ENABLED();
4595 	sx_assert(&allprison_lock, SA_XLOCKED);
4596 
4597 	if (refcount_release(&prr->prr_refcount)) {
4598 		racct_destroy(&prr->prr_racct);
4599 		LIST_REMOVE(prr, prr_next);
4600 		free(prr, M_PRISON_RACCT);
4601 	}
4602 }
4603 
4604 void
4605 prison_racct_free(struct prison_racct *prr)
4606 {
4607 	int old;
4608 
4609 	ASSERT_RACCT_ENABLED();
4610 	sx_assert(&allprison_lock, SA_UNLOCKED);
4611 
4612 	old = prr->prr_refcount;
4613 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4614 		return;
4615 
4616 	sx_xlock(&allprison_lock);
4617 	prison_racct_free_locked(prr);
4618 	sx_xunlock(&allprison_lock);
4619 }
4620 
4621 static void
4622 prison_racct_attach(struct prison *pr)
4623 {
4624 	struct prison_racct *prr;
4625 
4626 	ASSERT_RACCT_ENABLED();
4627 	sx_assert(&allprison_lock, SA_XLOCKED);
4628 
4629 	prr = prison_racct_find_locked(pr->pr_name);
4630 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4631 
4632 	pr->pr_prison_racct = prr;
4633 }
4634 
4635 /*
4636  * Handle jail renaming.  From the racct point of view, renaming means
4637  * moving from one prison_racct to another.
4638  */
4639 static void
4640 prison_racct_modify(struct prison *pr)
4641 {
4642 	struct proc *p;
4643 	struct ucred *cred;
4644 	struct prison_racct *oldprr;
4645 
4646 	ASSERT_RACCT_ENABLED();
4647 
4648 	sx_slock(&allproc_lock);
4649 	sx_xlock(&allprison_lock);
4650 
4651 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4652 		sx_xunlock(&allprison_lock);
4653 		sx_sunlock(&allproc_lock);
4654 		return;
4655 	}
4656 
4657 	oldprr = pr->pr_prison_racct;
4658 	pr->pr_prison_racct = NULL;
4659 
4660 	prison_racct_attach(pr);
4661 
4662 	/*
4663 	 * Move resource utilisation records.
4664 	 */
4665 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4666 
4667 	/*
4668 	 * Force rctl to reattach rules to processes.
4669 	 */
4670 	FOREACH_PROC_IN_SYSTEM(p) {
4671 		PROC_LOCK(p);
4672 		cred = crhold(p->p_ucred);
4673 		PROC_UNLOCK(p);
4674 		racct_proc_ucred_changed(p, cred, cred);
4675 		crfree(cred);
4676 	}
4677 
4678 	sx_sunlock(&allproc_lock);
4679 	prison_racct_free_locked(oldprr);
4680 	sx_xunlock(&allprison_lock);
4681 }
4682 
4683 static void
4684 prison_racct_detach(struct prison *pr)
4685 {
4686 
4687 	ASSERT_RACCT_ENABLED();
4688 	sx_assert(&allprison_lock, SA_UNLOCKED);
4689 
4690 	if (pr->pr_prison_racct == NULL)
4691 		return;
4692 	prison_racct_free(pr->pr_prison_racct);
4693 	pr->pr_prison_racct = NULL;
4694 }
4695 #endif /* RACCT */
4696 
4697 #ifdef DDB
4698 
4699 static void
4700 db_show_prison(struct prison *pr)
4701 {
4702 	int fi;
4703 #if defined(INET) || defined(INET6)
4704 	int ii;
4705 #endif
4706 	unsigned jsf;
4707 #ifdef INET6
4708 	char ip6buf[INET6_ADDRSTRLEN];
4709 #endif
4710 
4711 	db_printf("prison %p:\n", pr);
4712 	db_printf(" jid             = %d\n", pr->pr_id);
4713 	db_printf(" name            = %s\n", pr->pr_name);
4714 	db_printf(" parent          = %p\n", pr->pr_parent);
4715 	db_printf(" ref             = %d\n", pr->pr_ref);
4716 	db_printf(" uref            = %d\n", pr->pr_uref);
4717 	db_printf(" path            = %s\n", pr->pr_path);
4718 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4719 	    ? pr->pr_cpuset->cs_id : -1);
4720 #ifdef VIMAGE
4721 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4722 #endif
4723 	db_printf(" root            = %p\n", pr->pr_root);
4724 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4725 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4726 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4727 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4728 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4729 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4730 	db_printf(" flags           = 0x%x", pr->pr_flags);
4731 	for (fi = 0; fi < nitems(pr_flag_names); fi++)
4732 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4733 			db_printf(" %s", pr_flag_names[fi]);
4734 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4735 		jsf = pr->pr_flags &
4736 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4737 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4738 		    pr_flag_jailsys[fi].disable &&
4739 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4740 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4741 		    : "inherit");
4742 	}
4743 	db_printf(" allow           = 0x%x", pr->pr_allow);
4744 	for (fi = 0; fi < nitems(pr_allow_names); fi++)
4745 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4746 			db_printf(" %s", pr_allow_names[fi]);
4747 	db_printf("\n");
4748 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4749 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4750 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4751 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4752 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4753 #ifdef INET
4754 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4755 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4756 		db_printf(" %s %s\n",
4757 		    ii == 0 ? "ip4.addr        =" : "                 ",
4758 		    inet_ntoa(pr->pr_ip4[ii]));
4759 #endif
4760 #ifdef INET6
4761 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4762 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4763 		db_printf(" %s %s\n",
4764 		    ii == 0 ? "ip6.addr        =" : "                 ",
4765 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4766 #endif
4767 }
4768 
4769 DB_SHOW_COMMAND(prison, db_show_prison_command)
4770 {
4771 	struct prison *pr;
4772 
4773 	if (!have_addr) {
4774 		/*
4775 		 * Show all prisons in the list, and prison0 which is not
4776 		 * listed.
4777 		 */
4778 		db_show_prison(&prison0);
4779 		if (!db_pager_quit) {
4780 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4781 				db_show_prison(pr);
4782 				if (db_pager_quit)
4783 					break;
4784 			}
4785 		}
4786 		return;
4787 	}
4788 
4789 	if (addr == 0)
4790 		pr = &prison0;
4791 	else {
4792 		/* Look for a prison with the ID and with references. */
4793 		TAILQ_FOREACH(pr, &allprison, pr_list)
4794 			if (pr->pr_id == addr && pr->pr_ref > 0)
4795 				break;
4796 		if (pr == NULL)
4797 			/* Look again, without requiring a reference. */
4798 			TAILQ_FOREACH(pr, &allprison, pr_list)
4799 				if (pr->pr_id == addr)
4800 					break;
4801 		if (pr == NULL)
4802 			/* Assume address points to a valid prison. */
4803 			pr = (struct prison *)addr;
4804 	}
4805 	db_show_prison(pr);
4806 }
4807 
4808 #endif /* DDB */
4809