xref: /freebsd/sys/kern/kern_jail.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
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
361 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
362 #ifdef INET
363 			    + 1
364 #endif
365 #ifdef INET6
366 			    + 1
367 #endif
368 			    )];
369 	struct uio opt;
370 	char *u_path, *u_hostname, *u_name;
371 #ifdef INET
372 	uint32_t ip4s;
373 	struct in_addr *u_ip4;
374 #endif
375 #ifdef INET6
376 	struct in6_addr *u_ip6;
377 #endif
378 	size_t tmplen;
379 	int error, enforce_statfs, fi;
380 
381 	bzero(&optiov, sizeof(optiov));
382 	opt.uio_iov = optiov;
383 	opt.uio_iovcnt = 0;
384 	opt.uio_offset = -1;
385 	opt.uio_resid = -1;
386 	opt.uio_segflg = UIO_SYSSPACE;
387 	opt.uio_rw = UIO_READ;
388 	opt.uio_td = td;
389 
390 	/* Set permissions for top-level jails from sysctls. */
391 	if (!jailed(td->td_ucred)) {
392 		for (fi = 0; fi < sizeof(pr_allow_names) /
393 		     sizeof(pr_allow_names[0]); fi++) {
394 			optiov[opt.uio_iovcnt].iov_base =
395 			    (jail_default_allow & (1 << fi))
396 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
397 			optiov[opt.uio_iovcnt].iov_len =
398 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
399 			opt.uio_iovcnt += 2;
400 		}
401 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
402 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
403 		opt.uio_iovcnt++;
404 		enforce_statfs = jail_default_enforce_statfs;
405 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
406 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
407 		opt.uio_iovcnt++;
408 	}
409 
410 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
411 #ifdef INET
412 	ip4s = (j->version == 0) ? 1 : j->ip4s;
413 	if (ip4s > jail_max_af_ips)
414 		return (EINVAL);
415 	tmplen += ip4s * sizeof(struct in_addr);
416 #else
417 	if (j->ip4s > 0)
418 		return (EINVAL);
419 #endif
420 #ifdef INET6
421 	if (j->ip6s > jail_max_af_ips)
422 		return (EINVAL);
423 	tmplen += j->ip6s * sizeof(struct in6_addr);
424 #else
425 	if (j->ip6s > 0)
426 		return (EINVAL);
427 #endif
428 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
429 	u_hostname = u_path + MAXPATHLEN;
430 	u_name = u_hostname + MAXHOSTNAMELEN;
431 #ifdef INET
432 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
433 #endif
434 #ifdef INET6
435 #ifdef INET
436 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
437 #else
438 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
439 #endif
440 #endif
441 	optiov[opt.uio_iovcnt].iov_base = "path";
442 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
443 	opt.uio_iovcnt++;
444 	optiov[opt.uio_iovcnt].iov_base = u_path;
445 	error = copyinstr(j->path, u_path, MAXPATHLEN,
446 	    &optiov[opt.uio_iovcnt].iov_len);
447 	if (error) {
448 		free(u_path, M_TEMP);
449 		return (error);
450 	}
451 	opt.uio_iovcnt++;
452 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
453 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
454 	opt.uio_iovcnt++;
455 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
456 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
457 	    &optiov[opt.uio_iovcnt].iov_len);
458 	if (error) {
459 		free(u_path, M_TEMP);
460 		return (error);
461 	}
462 	opt.uio_iovcnt++;
463 	if (j->jailname != NULL) {
464 		optiov[opt.uio_iovcnt].iov_base = "name";
465 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
466 		opt.uio_iovcnt++;
467 		optiov[opt.uio_iovcnt].iov_base = u_name;
468 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
469 		    &optiov[opt.uio_iovcnt].iov_len);
470 		if (error) {
471 			free(u_path, M_TEMP);
472 			return (error);
473 		}
474 		opt.uio_iovcnt++;
475 	}
476 #ifdef INET
477 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
478 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
479 	opt.uio_iovcnt++;
480 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
481 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
482 	if (j->version == 0)
483 		u_ip4->s_addr = j->ip4s;
484 	else {
485 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
486 		if (error) {
487 			free(u_path, M_TEMP);
488 			return (error);
489 		}
490 	}
491 	opt.uio_iovcnt++;
492 #endif
493 #ifdef INET6
494 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
495 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
496 	opt.uio_iovcnt++;
497 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
498 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
499 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
500 	if (error) {
501 		free(u_path, M_TEMP);
502 		return (error);
503 	}
504 	opt.uio_iovcnt++;
505 #endif
506 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
507 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
508 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
509 	free(u_path, M_TEMP);
510 	return (error);
511 }
512 
513 
514 /*
515  * struct jail_set_args {
516  *	struct iovec *iovp;
517  *	unsigned int iovcnt;
518  *	int flags;
519  * };
520  */
521 int
522 sys_jail_set(struct thread *td, struct jail_set_args *uap)
523 {
524 	struct uio *auio;
525 	int error;
526 
527 	/* Check that we have an even number of iovecs. */
528 	if (uap->iovcnt & 1)
529 		return (EINVAL);
530 
531 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
532 	if (error)
533 		return (error);
534 	error = kern_jail_set(td, auio, uap->flags);
535 	free(auio, M_IOV);
536 	return (error);
537 }
538 
539 int
540 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
541 {
542 	struct nameidata nd;
543 #ifdef INET
544 	struct in_addr *ip4;
545 #endif
546 #ifdef INET6
547 	struct in6_addr *ip6;
548 #endif
549 	struct vfsopt *opt;
550 	struct vfsoptlist *opts;
551 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
552 	struct vnode *root;
553 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
554 	char *g_path, *osrelstr;
555 #if defined(INET) || defined(INET6)
556 	struct prison *tppr;
557 	void *op;
558 #endif
559 	unsigned long hid;
560 	size_t namelen, onamelen;
561 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
562 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
563 	int fi, jid, jsys, len, level;
564 	int childmax, osreldt, rsnum, slevel;
565 	int fullpath_disabled;
566 #if defined(INET) || defined(INET6)
567 	int ii, ij;
568 #endif
569 #ifdef INET
570 	int ip4s, redo_ip4;
571 #endif
572 #ifdef INET6
573 	int ip6s, redo_ip6;
574 #endif
575 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
576 	unsigned tallow;
577 	char numbuf[12];
578 
579 	error = priv_check(td, PRIV_JAIL_SET);
580 	if (!error && (flags & JAIL_ATTACH))
581 		error = priv_check(td, PRIV_JAIL_ATTACH);
582 	if (error)
583 		return (error);
584 	mypr = ppr = td->td_ucred->cr_prison;
585 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
586 		return (EPERM);
587 	if (flags & ~JAIL_SET_MASK)
588 		return (EINVAL);
589 
590 	/*
591 	 * Check all the parameters before committing to anything.  Not all
592 	 * errors can be caught early, but we may as well try.  Also, this
593 	 * takes care of some expensive stuff (path lookup) before getting
594 	 * the allprison lock.
595 	 *
596 	 * XXX Jails are not filesystems, and jail parameters are not mount
597 	 *     options.  But it makes more sense to re-use the vfsopt code
598 	 *     than duplicate it under a different name.
599 	 */
600 	error = vfs_buildopts(optuio, &opts);
601 	if (error)
602 		return (error);
603 #ifdef INET
604 	ip4 = NULL;
605 #endif
606 #ifdef INET6
607 	ip6 = NULL;
608 #endif
609 	g_path = NULL;
610 
611 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
612 	if (error == ENOENT)
613 		jid = 0;
614 	else if (error != 0)
615 		goto done_free;
616 
617 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
618 	if (error == ENOENT)
619 		gotslevel = 0;
620 	else if (error != 0)
621 		goto done_free;
622 	else
623 		gotslevel = 1;
624 
625 	error =
626 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
627 	if (error == ENOENT)
628 		gotchildmax = 0;
629 	else if (error != 0)
630 		goto done_free;
631 	else
632 		gotchildmax = 1;
633 
634 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
635 	if (error == ENOENT)
636 		gotenforce = 0;
637 	else if (error != 0)
638 		goto done_free;
639 	else if (enforce < 0 || enforce > 2) {
640 		error = EINVAL;
641 		goto done_free;
642 	} else
643 		gotenforce = 1;
644 
645 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
646 	if (error == ENOENT)
647 		gotrsnum = 0;
648 	else if (error != 0)
649 		goto done_free;
650 	else
651 		gotrsnum = 1;
652 
653 	pr_flags = ch_flags = 0;
654 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
655 	    fi++) {
656 		if (pr_flag_names[fi] == NULL)
657 			continue;
658 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
659 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
660 	}
661 	ch_flags |= pr_flags;
662 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
663 	    fi++) {
664 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
665 		    sizeof(jsys));
666 		if (error == ENOENT)
667 			continue;
668 		if (error != 0)
669 			goto done_free;
670 		switch (jsys) {
671 		case JAIL_SYS_DISABLE:
672 			if (!pr_flag_jailsys[fi].disable) {
673 				error = EINVAL;
674 				goto done_free;
675 			}
676 			pr_flags |= pr_flag_jailsys[fi].disable;
677 			break;
678 		case JAIL_SYS_NEW:
679 			pr_flags |= pr_flag_jailsys[fi].new;
680 			break;
681 		case JAIL_SYS_INHERIT:
682 			break;
683 		default:
684 			error = EINVAL;
685 			goto done_free;
686 		}
687 		ch_flags |=
688 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
689 	}
690 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
691 	    && !(pr_flags & PR_PERSIST)) {
692 		error = EINVAL;
693 		vfs_opterror(opts, "new jail must persist or attach");
694 		goto done_errmsg;
695 	}
696 #ifdef VIMAGE
697 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
698 		error = EINVAL;
699 		vfs_opterror(opts, "vnet cannot be changed after creation");
700 		goto done_errmsg;
701 	}
702 #endif
703 #ifdef INET
704 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
705 		error = EINVAL;
706 		vfs_opterror(opts, "ip4 cannot be changed after creation");
707 		goto done_errmsg;
708 	}
709 #endif
710 #ifdef INET6
711 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
712 		error = EINVAL;
713 		vfs_opterror(opts, "ip6 cannot be changed after creation");
714 		goto done_errmsg;
715 	}
716 #endif
717 
718 	pr_allow = ch_allow = 0;
719 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
720 	    fi++) {
721 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
722 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
723 	}
724 	ch_allow |= pr_allow;
725 
726 	error = vfs_getopt(opts, "name", (void **)&name, &len);
727 	if (error == ENOENT)
728 		name = NULL;
729 	else if (error != 0)
730 		goto done_free;
731 	else {
732 		if (len == 0 || name[len - 1] != '\0') {
733 			error = EINVAL;
734 			goto done_free;
735 		}
736 		if (len > MAXHOSTNAMELEN) {
737 			error = ENAMETOOLONG;
738 			goto done_free;
739 		}
740 	}
741 
742 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
743 	if (error == ENOENT)
744 		host = NULL;
745 	else if (error != 0)
746 		goto done_free;
747 	else {
748 		ch_flags |= PR_HOST;
749 		pr_flags |= PR_HOST;
750 		if (len == 0 || host[len - 1] != '\0') {
751 			error = EINVAL;
752 			goto done_free;
753 		}
754 		if (len > MAXHOSTNAMELEN) {
755 			error = ENAMETOOLONG;
756 			goto done_free;
757 		}
758 	}
759 
760 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
761 	if (error == ENOENT)
762 		domain = NULL;
763 	else if (error != 0)
764 		goto done_free;
765 	else {
766 		ch_flags |= PR_HOST;
767 		pr_flags |= PR_HOST;
768 		if (len == 0 || domain[len - 1] != '\0') {
769 			error = EINVAL;
770 			goto done_free;
771 		}
772 		if (len > MAXHOSTNAMELEN) {
773 			error = ENAMETOOLONG;
774 			goto done_free;
775 		}
776 	}
777 
778 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
779 	if (error == ENOENT)
780 		uuid = NULL;
781 	else if (error != 0)
782 		goto done_free;
783 	else {
784 		ch_flags |= PR_HOST;
785 		pr_flags |= PR_HOST;
786 		if (len == 0 || uuid[len - 1] != '\0') {
787 			error = EINVAL;
788 			goto done_free;
789 		}
790 		if (len > HOSTUUIDLEN) {
791 			error = ENAMETOOLONG;
792 			goto done_free;
793 		}
794 	}
795 
796 #ifdef COMPAT_FREEBSD32
797 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
798 		uint32_t hid32;
799 
800 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
801 		hid = hid32;
802 	} else
803 #endif
804 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
805 	if (error == ENOENT)
806 		gothid = 0;
807 	else if (error != 0)
808 		goto done_free;
809 	else {
810 		gothid = 1;
811 		ch_flags |= PR_HOST;
812 		pr_flags |= PR_HOST;
813 	}
814 
815 #ifdef INET
816 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
817 	if (error == ENOENT)
818 		ip4s = 0;
819 	else if (error != 0)
820 		goto done_free;
821 	else if (ip4s & (sizeof(*ip4) - 1)) {
822 		error = EINVAL;
823 		goto done_free;
824 	} else {
825 		ch_flags |= PR_IP4_USER;
826 		pr_flags |= PR_IP4_USER;
827 		if (ip4s > 0) {
828 			ip4s /= sizeof(*ip4);
829 			if (ip4s > jail_max_af_ips) {
830 				error = EINVAL;
831 				vfs_opterror(opts, "too many IPv4 addresses");
832 				goto done_errmsg;
833 			}
834 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
835 			bcopy(op, ip4, ip4s * sizeof(*ip4));
836 			/*
837 			 * IP addresses are all sorted but ip[0] to preserve
838 			 * the primary IP address as given from userland.
839 			 * This special IP is used for unbound outgoing
840 			 * connections as well for "loopback" traffic in case
841 			 * source address selection cannot find any more fitting
842 			 * address to connect from.
843 			 */
844 			if (ip4s > 1)
845 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
846 			/*
847 			 * Check for duplicate addresses and do some simple
848 			 * zero and broadcast checks. If users give other bogus
849 			 * addresses it is their problem.
850 			 *
851 			 * We do not have to care about byte order for these
852 			 * checks so we will do them in NBO.
853 			 */
854 			for (ii = 0; ii < ip4s; ii++) {
855 				if (ip4[ii].s_addr == INADDR_ANY ||
856 				    ip4[ii].s_addr == INADDR_BROADCAST) {
857 					error = EINVAL;
858 					goto done_free;
859 				}
860 				if ((ii+1) < ip4s &&
861 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
862 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
863 					error = EINVAL;
864 					goto done_free;
865 				}
866 			}
867 		}
868 	}
869 #endif
870 
871 #ifdef INET6
872 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
873 	if (error == ENOENT)
874 		ip6s = 0;
875 	else if (error != 0)
876 		goto done_free;
877 	else if (ip6s & (sizeof(*ip6) - 1)) {
878 		error = EINVAL;
879 		goto done_free;
880 	} else {
881 		ch_flags |= PR_IP6_USER;
882 		pr_flags |= PR_IP6_USER;
883 		if (ip6s > 0) {
884 			ip6s /= sizeof(*ip6);
885 			if (ip6s > jail_max_af_ips) {
886 				error = EINVAL;
887 				vfs_opterror(opts, "too many IPv6 addresses");
888 				goto done_errmsg;
889 			}
890 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
891 			bcopy(op, ip6, ip6s * sizeof(*ip6));
892 			if (ip6s > 1)
893 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
894 			for (ii = 0; ii < ip6s; ii++) {
895 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
896 					error = EINVAL;
897 					goto done_free;
898 				}
899 				if ((ii+1) < ip6s &&
900 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
901 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
902 				{
903 					error = EINVAL;
904 					goto done_free;
905 				}
906 			}
907 		}
908 	}
909 #endif
910 
911 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
912 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
913 		error = EINVAL;
914 		vfs_opterror(opts,
915 		    "vnet jails cannot have IP address restrictions");
916 		goto done_errmsg;
917 	}
918 #endif
919 
920 	fullpath_disabled = 0;
921 	root = NULL;
922 	error = vfs_getopt(opts, "path", (void **)&path, &len);
923 	if (error == ENOENT)
924 		path = NULL;
925 	else if (error != 0)
926 		goto done_free;
927 	else {
928 		if (flags & JAIL_UPDATE) {
929 			error = EINVAL;
930 			vfs_opterror(opts,
931 			    "path cannot be changed after creation");
932 			goto done_errmsg;
933 		}
934 		if (len == 0 || path[len - 1] != '\0') {
935 			error = EINVAL;
936 			goto done_free;
937 		}
938 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
939 		    path, td);
940 		error = namei(&nd);
941 		if (error)
942 			goto done_free;
943 		root = nd.ni_vp;
944 		NDFREE(&nd, NDF_ONLY_PNBUF);
945 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
946 		strlcpy(g_path, path, MAXPATHLEN);
947 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
948 		if (error == 0)
949 			path = g_path;
950 		else if (error == ENODEV) {
951 			/* proceed if sysctl debug.disablefullpath == 1 */
952 			fullpath_disabled = 1;
953 			if (len < 2 || (len == 2 && path[0] == '/'))
954 				path = NULL;
955 		} else {
956 			/* exit on other errors */
957 			goto done_free;
958 		}
959 		if (root->v_type != VDIR) {
960 			error = ENOTDIR;
961 			vput(root);
962 			goto done_free;
963 		}
964 		VOP_UNLOCK(root, 0);
965 		if (fullpath_disabled) {
966 			/* Leave room for a real-root full pathname. */
967 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
968 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
969 				error = ENAMETOOLONG;
970 				goto done_free;
971 			}
972 		}
973 	}
974 
975 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
976 	if (error == ENOENT)
977 		osrelstr = NULL;
978 	else if (error != 0)
979 		goto done_free;
980 	else {
981 		if (flags & JAIL_UPDATE) {
982 			error = EINVAL;
983 			vfs_opterror(opts,
984 			    "osrelease cannot be changed after creation");
985 			goto done_errmsg;
986 		}
987 		if (len == 0 || len >= OSRELEASELEN) {
988 			error = EINVAL;
989 			vfs_opterror(opts,
990 			    "osrelease string must be 1-%d bytes long",
991 			    OSRELEASELEN - 1);
992 			goto done_errmsg;
993 		}
994 	}
995 
996 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
997 	if (error == ENOENT)
998 		osreldt = 0;
999 	else if (error != 0)
1000 		goto done_free;
1001 	else {
1002 		if (flags & JAIL_UPDATE) {
1003 			error = EINVAL;
1004 			vfs_opterror(opts,
1005 			    "osreldate cannot be changed after creation");
1006 			goto done_errmsg;
1007 		}
1008 		if (osreldt == 0) {
1009 			error = EINVAL;
1010 			vfs_opterror(opts, "osreldate cannot be 0");
1011 			goto done_errmsg;
1012 		}
1013 	}
1014 
1015 	/*
1016 	 * Grab the allprison lock before letting modules check their
1017 	 * parameters.  Once we have it, do not let go so we'll have a
1018 	 * consistent view of the OSD list.
1019 	 */
1020 	sx_xlock(&allprison_lock);
1021 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
1022 	if (error)
1023 		goto done_unlock_list;
1024 
1025 	/* By now, all parameters should have been noted. */
1026 	TAILQ_FOREACH(opt, opts, link) {
1027 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1028 			error = EINVAL;
1029 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1030 			goto done_unlock_list;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * See if we are creating a new record or updating an existing one.
1036 	 * This abuses the file error codes ENOENT and EEXIST.
1037 	 */
1038 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1039 	if (!cuflags) {
1040 		error = EINVAL;
1041 		vfs_opterror(opts, "no valid operation (create or update)");
1042 		goto done_unlock_list;
1043 	}
1044 	pr = NULL;
1045 	namelc = NULL;
1046 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1047 		namelc = strrchr(name, '.');
1048 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1049 		if (*p != '\0')
1050 			jid = 0;
1051 	}
1052 	if (jid != 0) {
1053 		/*
1054 		 * See if a requested jid already exists.  There is an
1055 		 * information leak here if the jid exists but is not within
1056 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1057 		 * even though they cannot see the jail, and CREATE | UPDATE
1058 		 * will return ENOENT which is not normally a valid error.
1059 		 */
1060 		if (jid < 0) {
1061 			error = EINVAL;
1062 			vfs_opterror(opts, "negative jid");
1063 			goto done_unlock_list;
1064 		}
1065 		pr = prison_find(jid);
1066 		if (pr != NULL) {
1067 			ppr = pr->pr_parent;
1068 			/* Create: jid must not exist. */
1069 			if (cuflags == JAIL_CREATE) {
1070 				mtx_unlock(&pr->pr_mtx);
1071 				error = EEXIST;
1072 				vfs_opterror(opts, "jail %d already exists",
1073 				    jid);
1074 				goto done_unlock_list;
1075 			}
1076 			if (!prison_ischild(mypr, pr)) {
1077 				mtx_unlock(&pr->pr_mtx);
1078 				pr = NULL;
1079 			} else if (pr->pr_uref == 0) {
1080 				if (!(flags & JAIL_DYING)) {
1081 					mtx_unlock(&pr->pr_mtx);
1082 					error = ENOENT;
1083 					vfs_opterror(opts, "jail %d is dying",
1084 					    jid);
1085 					goto done_unlock_list;
1086 				} else if ((flags & JAIL_ATTACH) ||
1087 				    (pr_flags & PR_PERSIST)) {
1088 					/*
1089 					 * A dying jail might be resurrected
1090 					 * (via attach or persist), but first
1091 					 * it must determine if another jail
1092 					 * has claimed its name.  Accomplish
1093 					 * this by implicitly re-setting the
1094 					 * name.
1095 					 */
1096 					if (name == NULL)
1097 						name = prison_name(mypr, pr);
1098 				}
1099 			}
1100 		}
1101 		if (pr == NULL) {
1102 			/* Update: jid must exist. */
1103 			if (cuflags == JAIL_UPDATE) {
1104 				error = ENOENT;
1105 				vfs_opterror(opts, "jail %d not found", jid);
1106 				goto done_unlock_list;
1107 			}
1108 		}
1109 	}
1110 	/*
1111 	 * If the caller provided a name, look for a jail by that name.
1112 	 * This has different semantics for creates and updates keyed by jid
1113 	 * (where the name must not already exist in a different jail),
1114 	 * and updates keyed by the name itself (where the name must exist
1115 	 * because that is the jail being updated).
1116 	 */
1117 	if (name != NULL) {
1118 		namelc = strrchr(name, '.');
1119 		if (namelc == NULL)
1120 			namelc = name;
1121 		else {
1122 			/*
1123 			 * This is a hierarchical name.  Split it into the
1124 			 * parent and child names, and make sure the parent
1125 			 * exists or matches an already found jail.
1126 			 */
1127 			*namelc = '\0';
1128 			if (pr != NULL) {
1129 				if (strncmp(name, ppr->pr_name, namelc - name)
1130 				    || ppr->pr_name[namelc - name] != '\0') {
1131 					mtx_unlock(&pr->pr_mtx);
1132 					error = EINVAL;
1133 					vfs_opterror(opts,
1134 					    "cannot change jail's parent");
1135 					goto done_unlock_list;
1136 				}
1137 			} else {
1138 				ppr = prison_find_name(mypr, name);
1139 				if (ppr == NULL) {
1140 					error = ENOENT;
1141 					vfs_opterror(opts,
1142 					    "jail \"%s\" not found", name);
1143 					goto done_unlock_list;
1144 				}
1145 				mtx_unlock(&ppr->pr_mtx);
1146 			}
1147 			name = ++namelc;
1148 		}
1149 		if (name[0] != '\0') {
1150 			namelen =
1151 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1152  name_again:
1153 			deadpr = NULL;
1154 			FOREACH_PRISON_CHILD(ppr, tpr) {
1155 				if (tpr != pr && tpr->pr_ref > 0 &&
1156 				    !strcmp(tpr->pr_name + namelen, name)) {
1157 					if (pr == NULL &&
1158 					    cuflags != JAIL_CREATE) {
1159 						mtx_lock(&tpr->pr_mtx);
1160 						if (tpr->pr_ref > 0) {
1161 							/*
1162 							 * Use this jail
1163 							 * for updates.
1164 							 */
1165 							if (tpr->pr_uref > 0) {
1166 								pr = tpr;
1167 								break;
1168 							}
1169 							deadpr = tpr;
1170 						}
1171 						mtx_unlock(&tpr->pr_mtx);
1172 					} else if (tpr->pr_uref > 0) {
1173 						/*
1174 						 * Create, or update(jid):
1175 						 * name must not exist in an
1176 						 * active sibling jail.
1177 						 */
1178 						error = EEXIST;
1179 						if (pr != NULL)
1180 							mtx_unlock(&pr->pr_mtx);
1181 						vfs_opterror(opts,
1182 						   "jail \"%s\" already exists",
1183 						   name);
1184 						goto done_unlock_list;
1185 					}
1186 				}
1187 			}
1188 			/* If no active jail is found, use a dying one. */
1189 			if (deadpr != NULL && pr == NULL) {
1190 				if (flags & JAIL_DYING) {
1191 					mtx_lock(&deadpr->pr_mtx);
1192 					if (deadpr->pr_ref == 0) {
1193 						mtx_unlock(&deadpr->pr_mtx);
1194 						goto name_again;
1195 					}
1196 					pr = deadpr;
1197 				} else if (cuflags == JAIL_UPDATE) {
1198 					error = ENOENT;
1199 					vfs_opterror(opts,
1200 					    "jail \"%s\" is dying", name);
1201 					goto done_unlock_list;
1202 				}
1203 			}
1204 			/* Update: name must exist if no jid. */
1205 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1206 				error = ENOENT;
1207 				vfs_opterror(opts, "jail \"%s\" not found",
1208 				    name);
1209 				goto done_unlock_list;
1210 			}
1211 		}
1212 	}
1213 	/* Update: must provide a jid or name. */
1214 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1215 		error = ENOENT;
1216 		vfs_opterror(opts, "update specified no jail");
1217 		goto done_unlock_list;
1218 	}
1219 
1220 	/* If there's no prison to update, create a new one and link it in. */
1221 	if (pr == NULL) {
1222 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1223 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1224 				error = EPERM;
1225 				vfs_opterror(opts, "prison limit exceeded");
1226 				goto done_unlock_list;
1227 			}
1228 		created = 1;
1229 		mtx_lock(&ppr->pr_mtx);
1230 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1231 			mtx_unlock(&ppr->pr_mtx);
1232 			error = ENOENT;
1233 			vfs_opterror(opts, "parent jail went away!");
1234 			goto done_unlock_list;
1235 		}
1236 		ppr->pr_ref++;
1237 		ppr->pr_uref++;
1238 		mtx_unlock(&ppr->pr_mtx);
1239 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1240 		if (jid == 0) {
1241 			/* Find the next free jid. */
1242 			jid = lastprid + 1;
1243  findnext:
1244 			if (jid == JAIL_MAX)
1245 				jid = 1;
1246 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1247 				if (tpr->pr_id < jid)
1248 					continue;
1249 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1250 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1251 					break;
1252 				}
1253 				if (jid == lastprid) {
1254 					error = EAGAIN;
1255 					vfs_opterror(opts,
1256 					    "no available jail IDs");
1257 					free(pr, M_PRISON);
1258 					prison_deref(ppr, PD_DEREF |
1259 					    PD_DEUREF | PD_LIST_XLOCKED);
1260 					goto done_releroot;
1261 				}
1262 				jid++;
1263 				goto findnext;
1264 			}
1265 			lastprid = jid;
1266 		} else {
1267 			/*
1268 			 * The jail already has a jid (that did not yet exist),
1269 			 * so just find where to insert it.
1270 			 */
1271 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1272 				if (tpr->pr_id >= jid) {
1273 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1274 					break;
1275 				}
1276 		}
1277 		if (tpr == NULL)
1278 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1279 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1280 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1281 			tpr->pr_childcount++;
1282 
1283 		pr->pr_parent = ppr;
1284 		pr->pr_id = jid;
1285 
1286 		/* Set some default values, and inherit some from the parent. */
1287 		if (name == NULL)
1288 			name = "";
1289 		if (path == NULL) {
1290 			path = "/";
1291 			root = mypr->pr_root;
1292 			vref(root);
1293 		}
1294 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1295 		pr->pr_flags |= PR_HOST;
1296 #if defined(INET) || defined(INET6)
1297 #ifdef VIMAGE
1298 		if (!(pr_flags & PR_VNET))
1299 #endif
1300 		{
1301 #ifdef INET
1302 			if (!(ch_flags & PR_IP4_USER))
1303 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1304 			else if (!(pr_flags & PR_IP4_USER)) {
1305 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1306 				if (ppr->pr_ip4 != NULL) {
1307 					pr->pr_ip4s = ppr->pr_ip4s;
1308 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1309 					    sizeof(struct in_addr), M_PRISON,
1310 					    M_WAITOK);
1311 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1312 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1313 				}
1314 			}
1315 #endif
1316 #ifdef INET6
1317 			if (!(ch_flags & PR_IP6_USER))
1318 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1319 			else if (!(pr_flags & PR_IP6_USER)) {
1320 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1321 				if (ppr->pr_ip6 != NULL) {
1322 					pr->pr_ip6s = ppr->pr_ip6s;
1323 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1324 					    sizeof(struct in6_addr), M_PRISON,
1325 					    M_WAITOK);
1326 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1327 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1328 				}
1329 			}
1330 #endif
1331 		}
1332 #endif
1333 		/* Source address selection is always on by default. */
1334 		pr->pr_flags |= _PR_IP_SADDRSEL;
1335 
1336 		pr->pr_securelevel = ppr->pr_securelevel;
1337 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1338 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1339 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1340 
1341 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1342 		if (osrelstr == NULL)
1343 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1344 		else
1345 		    strcpy(pr->pr_osrelease, osrelstr);
1346 
1347 		LIST_INIT(&pr->pr_children);
1348 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1349 
1350 #ifdef VIMAGE
1351 		/* Allocate a new vnet if specified. */
1352 		pr->pr_vnet = (pr_flags & PR_VNET)
1353 		    ? vnet_alloc() : ppr->pr_vnet;
1354 #endif
1355 		/*
1356 		 * Allocate a dedicated cpuset for each jail.
1357 		 * Unlike other initial settings, this may return an erorr.
1358 		 */
1359 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1360 		if (error) {
1361 			prison_deref(pr, PD_LIST_XLOCKED);
1362 			goto done_releroot;
1363 		}
1364 
1365 		mtx_lock(&pr->pr_mtx);
1366 		/*
1367 		 * New prisons do not yet have a reference, because we do not
1368 		 * want other to see the incomplete prison once the
1369 		 * allprison_lock is downgraded.
1370 		 */
1371 	} else {
1372 		created = 0;
1373 		/*
1374 		 * Grab a reference for existing prisons, to ensure they
1375 		 * continue to exist for the duration of the call.
1376 		 */
1377 		pr->pr_ref++;
1378 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1379 		if ((pr->pr_flags & PR_VNET) &&
1380 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1381 			error = EINVAL;
1382 			vfs_opterror(opts,
1383 			    "vnet jails cannot have IP address restrictions");
1384 			goto done_deref_locked;
1385 		}
1386 #endif
1387 #ifdef INET
1388 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1389 			error = EINVAL;
1390 			vfs_opterror(opts,
1391 			    "ip4 cannot be changed after creation");
1392 			goto done_deref_locked;
1393 		}
1394 #endif
1395 #ifdef INET6
1396 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1397 			error = EINVAL;
1398 			vfs_opterror(opts,
1399 			    "ip6 cannot be changed after creation");
1400 			goto done_deref_locked;
1401 		}
1402 #endif
1403 	}
1404 
1405 	/* Do final error checking before setting anything. */
1406 	if (gotslevel) {
1407 		if (slevel < ppr->pr_securelevel) {
1408 			error = EPERM;
1409 			goto done_deref_locked;
1410 		}
1411 	}
1412 	if (gotchildmax) {
1413 		if (childmax >= ppr->pr_childmax) {
1414 			error = EPERM;
1415 			goto done_deref_locked;
1416 		}
1417 	}
1418 	if (gotenforce) {
1419 		if (enforce < ppr->pr_enforce_statfs) {
1420 			error = EPERM;
1421 			goto done_deref_locked;
1422 		}
1423 	}
1424 	if (gotrsnum) {
1425 		/*
1426 		 * devfs_rsnum is a uint16_t
1427 		 */
1428 		if (rsnum < 0 || rsnum > 65535) {
1429 			error = EINVAL;
1430 			goto done_deref_locked;
1431 		}
1432 		/*
1433 		 * Nested jails always inherit parent's devfs ruleset
1434 		 */
1435 		if (jailed(td->td_ucred)) {
1436 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1437 				error = EPERM;
1438 				goto done_deref_locked;
1439 			} else
1440 				rsnum = ppr->pr_devfs_rsnum;
1441 		}
1442 	}
1443 #ifdef INET
1444 	if (ip4s > 0) {
1445 		if (ppr->pr_flags & PR_IP4) {
1446 			/*
1447 			 * Make sure the new set of IP addresses is a
1448 			 * subset of the parent's list.  Don't worry
1449 			 * about the parent being unlocked, as any
1450 			 * setting is done with allprison_lock held.
1451 			 */
1452 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1453 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1454 					break;
1455 			if (ij == ppr->pr_ip4s) {
1456 				error = EPERM;
1457 				goto done_deref_locked;
1458 			}
1459 			if (ip4s > 1) {
1460 				for (ii = ij = 1; ii < ip4s; ii++) {
1461 					if (ip4[ii].s_addr ==
1462 					    ppr->pr_ip4[0].s_addr)
1463 						continue;
1464 					for (; ij < ppr->pr_ip4s; ij++)
1465 						if (ip4[ii].s_addr ==
1466 						    ppr->pr_ip4[ij].s_addr)
1467 							break;
1468 					if (ij == ppr->pr_ip4s)
1469 						break;
1470 				}
1471 				if (ij == ppr->pr_ip4s) {
1472 					error = EPERM;
1473 					goto done_deref_locked;
1474 				}
1475 			}
1476 		}
1477 		/*
1478 		 * Check for conflicting IP addresses.  We permit them
1479 		 * if there is no more than one IP on each jail.  If
1480 		 * there is a duplicate on a jail with more than one
1481 		 * IP stop checking and return error.
1482 		 */
1483 		tppr = ppr;
1484 #ifdef VIMAGE
1485 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1486 			if (tppr->pr_flags & PR_VNET)
1487 				break;
1488 #endif
1489 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1490 			if (tpr == pr ||
1491 #ifdef VIMAGE
1492 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1493 #endif
1494 			    tpr->pr_uref == 0) {
1495 				descend = 0;
1496 				continue;
1497 			}
1498 			if (!(tpr->pr_flags & PR_IP4_USER))
1499 				continue;
1500 			descend = 0;
1501 			if (tpr->pr_ip4 == NULL ||
1502 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1503 				continue;
1504 			for (ii = 0; ii < ip4s; ii++) {
1505 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1506 					error = EADDRINUSE;
1507 					vfs_opterror(opts,
1508 					    "IPv4 addresses clash");
1509 					goto done_deref_locked;
1510 				}
1511 			}
1512 		}
1513 	}
1514 #endif
1515 #ifdef INET6
1516 	if (ip6s > 0) {
1517 		if (ppr->pr_flags & PR_IP6) {
1518 			/*
1519 			 * Make sure the new set of IP addresses is a
1520 			 * subset of the parent's list.
1521 			 */
1522 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1523 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1524 				    &ppr->pr_ip6[ij]))
1525 					break;
1526 			if (ij == ppr->pr_ip6s) {
1527 				error = EPERM;
1528 				goto done_deref_locked;
1529 			}
1530 			if (ip6s > 1) {
1531 				for (ii = ij = 1; ii < ip6s; ii++) {
1532 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1533 					     &ppr->pr_ip6[0]))
1534 						continue;
1535 					for (; ij < ppr->pr_ip6s; ij++)
1536 						if (IN6_ARE_ADDR_EQUAL(
1537 						    &ip6[ii], &ppr->pr_ip6[ij]))
1538 							break;
1539 					if (ij == ppr->pr_ip6s)
1540 						break;
1541 				}
1542 				if (ij == ppr->pr_ip6s) {
1543 					error = EPERM;
1544 					goto done_deref_locked;
1545 				}
1546 			}
1547 		}
1548 		/* Check for conflicting IP addresses. */
1549 		tppr = ppr;
1550 #ifdef VIMAGE
1551 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1552 			if (tppr->pr_flags & PR_VNET)
1553 				break;
1554 #endif
1555 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1556 			if (tpr == pr ||
1557 #ifdef VIMAGE
1558 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1559 #endif
1560 			    tpr->pr_uref == 0) {
1561 				descend = 0;
1562 				continue;
1563 			}
1564 			if (!(tpr->pr_flags & PR_IP6_USER))
1565 				continue;
1566 			descend = 0;
1567 			if (tpr->pr_ip6 == NULL ||
1568 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1569 				continue;
1570 			for (ii = 0; ii < ip6s; ii++) {
1571 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1572 					error = EADDRINUSE;
1573 					vfs_opterror(opts,
1574 					    "IPv6 addresses clash");
1575 					goto done_deref_locked;
1576 				}
1577 			}
1578 		}
1579 	}
1580 #endif
1581 	onamelen = namelen = 0;
1582 	if (name != NULL) {
1583 		/* Give a default name of the jid.  Also allow the name to be
1584 		 * explicitly the jid - but not any other number, and only in
1585 		 * normal form (no leading zero/etc).
1586 		 */
1587 		if (name[0] == '\0')
1588 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1589 		else if ((strtoul(namelc, &p, 10) != jid ||
1590 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1591 			error = EINVAL;
1592 			vfs_opterror(opts,
1593 			    "name cannot be numeric (unless it is the jid)");
1594 			goto done_deref_locked;
1595 		}
1596 		/*
1597 		 * Make sure the name isn't too long for the prison or its
1598 		 * children.
1599 		 */
1600 		onamelen = strlen(pr->pr_name);
1601 		namelen = strlen(name);
1602 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1603 			error = ENAMETOOLONG;
1604 			goto done_deref_locked;
1605 		}
1606 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1607 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1608 			    sizeof(pr->pr_name)) {
1609 				error = ENAMETOOLONG;
1610 				goto done_deref_locked;
1611 			}
1612 		}
1613 	}
1614 	if (pr_allow & ~ppr->pr_allow) {
1615 		error = EPERM;
1616 		goto done_deref_locked;
1617 	}
1618 
1619 	/* Set the parameters of the prison. */
1620 #ifdef INET
1621 	redo_ip4 = 0;
1622 	if (pr_flags & PR_IP4_USER) {
1623 		pr->pr_flags |= PR_IP4;
1624 		free(pr->pr_ip4, M_PRISON);
1625 		pr->pr_ip4s = ip4s;
1626 		pr->pr_ip4 = ip4;
1627 		ip4 = NULL;
1628 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1629 #ifdef VIMAGE
1630 			if (tpr->pr_flags & PR_VNET) {
1631 				descend = 0;
1632 				continue;
1633 			}
1634 #endif
1635 			if (prison_restrict_ip4(tpr, NULL)) {
1636 				redo_ip4 = 1;
1637 				descend = 0;
1638 			}
1639 		}
1640 	}
1641 #endif
1642 #ifdef INET6
1643 	redo_ip6 = 0;
1644 	if (pr_flags & PR_IP6_USER) {
1645 		pr->pr_flags |= PR_IP6;
1646 		free(pr->pr_ip6, M_PRISON);
1647 		pr->pr_ip6s = ip6s;
1648 		pr->pr_ip6 = ip6;
1649 		ip6 = NULL;
1650 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1651 #ifdef VIMAGE
1652 			if (tpr->pr_flags & PR_VNET) {
1653 				descend = 0;
1654 				continue;
1655 			}
1656 #endif
1657 			if (prison_restrict_ip6(tpr, NULL)) {
1658 				redo_ip6 = 1;
1659 				descend = 0;
1660 			}
1661 		}
1662 	}
1663 #endif
1664 	if (gotslevel) {
1665 		pr->pr_securelevel = slevel;
1666 		/* Set all child jails to be at least this level. */
1667 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1668 			if (tpr->pr_securelevel < slevel)
1669 				tpr->pr_securelevel = slevel;
1670 	}
1671 	if (gotchildmax) {
1672 		pr->pr_childmax = childmax;
1673 		/* Set all child jails to under this limit. */
1674 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1675 			if (tpr->pr_childmax > childmax - level)
1676 				tpr->pr_childmax = childmax > level
1677 				    ? childmax - level : 0;
1678 	}
1679 	if (gotenforce) {
1680 		pr->pr_enforce_statfs = enforce;
1681 		/* Pass this restriction on to the children. */
1682 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1683 			if (tpr->pr_enforce_statfs < enforce)
1684 				tpr->pr_enforce_statfs = enforce;
1685 	}
1686 	if (gotrsnum) {
1687 		pr->pr_devfs_rsnum = rsnum;
1688 		/* Pass this restriction on to the children. */
1689 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1690 			tpr->pr_devfs_rsnum = rsnum;
1691 	}
1692 	if (name != NULL) {
1693 		if (ppr == &prison0)
1694 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1695 		else
1696 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1697 			    ppr->pr_name, name);
1698 		/* Change this component of child names. */
1699 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1700 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1701 			    strlen(tpr->pr_name + onamelen) + 1);
1702 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1703 		}
1704 	}
1705 	if (path != NULL) {
1706 		/* Try to keep a real-rooted full pathname. */
1707 		if (fullpath_disabled && path[0] == '/' &&
1708 		    strcmp(mypr->pr_path, "/"))
1709 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1710 			    mypr->pr_path, path);
1711 		else
1712 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1713 		pr->pr_root = root;
1714 	}
1715 	if (PR_HOST & ch_flags & ~pr_flags) {
1716 		if (pr->pr_flags & PR_HOST) {
1717 			/*
1718 			 * Copy the parent's host info.  As with pr_ip4 above,
1719 			 * the lack of a lock on the parent is not a problem;
1720 			 * it is always set with allprison_lock at least
1721 			 * shared, and is held exclusively here.
1722 			 */
1723 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1724 			    sizeof(pr->pr_hostname));
1725 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1726 			    sizeof(pr->pr_domainname));
1727 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1728 			    sizeof(pr->pr_hostuuid));
1729 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1730 		}
1731 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1732 		/* Set this prison, and any descendants without PR_HOST. */
1733 		if (host != NULL)
1734 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1735 		if (domain != NULL)
1736 			strlcpy(pr->pr_domainname, domain,
1737 			    sizeof(pr->pr_domainname));
1738 		if (uuid != NULL)
1739 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1740 		if (gothid)
1741 			pr->pr_hostid = hid;
1742 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1743 			if (tpr->pr_flags & PR_HOST)
1744 				descend = 0;
1745 			else {
1746 				if (host != NULL)
1747 					strlcpy(tpr->pr_hostname,
1748 					    pr->pr_hostname,
1749 					    sizeof(tpr->pr_hostname));
1750 				if (domain != NULL)
1751 					strlcpy(tpr->pr_domainname,
1752 					    pr->pr_domainname,
1753 					    sizeof(tpr->pr_domainname));
1754 				if (uuid != NULL)
1755 					strlcpy(tpr->pr_hostuuid,
1756 					    pr->pr_hostuuid,
1757 					    sizeof(tpr->pr_hostuuid));
1758 				if (gothid)
1759 					tpr->pr_hostid = hid;
1760 			}
1761 		}
1762 	}
1763 	if ((tallow = ch_allow & ~pr_allow)) {
1764 		/* Clear allow bits in all children. */
1765 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1766 			tpr->pr_allow &= ~tallow;
1767 	}
1768 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1769 	/*
1770 	 * Persistent prisons get an extra reference, and prisons losing their
1771 	 * persist flag lose that reference.  Only do this for existing prisons
1772 	 * for now, so new ones will remain unseen until after the module
1773 	 * handlers have completed.
1774 	 */
1775 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1776 		if (pr_flags & PR_PERSIST) {
1777 			pr->pr_ref++;
1778 			pr->pr_uref++;
1779 		} else {
1780 			pr->pr_ref--;
1781 			pr->pr_uref--;
1782 		}
1783 	}
1784 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1785 	mtx_unlock(&pr->pr_mtx);
1786 
1787 #ifdef RACCT
1788 	if (racct_enable && created)
1789 		prison_racct_attach(pr);
1790 #endif
1791 
1792 	/* Locks may have prevented a complete restriction of child IP
1793 	 * addresses.  If so, allocate some more memory and try again.
1794 	 */
1795 #ifdef INET
1796 	while (redo_ip4) {
1797 		ip4s = pr->pr_ip4s;
1798 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1799 		mtx_lock(&pr->pr_mtx);
1800 		redo_ip4 = 0;
1801 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1802 #ifdef VIMAGE
1803 			if (tpr->pr_flags & PR_VNET) {
1804 				descend = 0;
1805 				continue;
1806 			}
1807 #endif
1808 			if (prison_restrict_ip4(tpr, ip4)) {
1809 				if (ip4 != NULL)
1810 					ip4 = NULL;
1811 				else
1812 					redo_ip4 = 1;
1813 			}
1814 		}
1815 		mtx_unlock(&pr->pr_mtx);
1816 	}
1817 #endif
1818 #ifdef INET6
1819 	while (redo_ip6) {
1820 		ip6s = pr->pr_ip6s;
1821 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1822 		mtx_lock(&pr->pr_mtx);
1823 		redo_ip6 = 0;
1824 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1825 #ifdef VIMAGE
1826 			if (tpr->pr_flags & PR_VNET) {
1827 				descend = 0;
1828 				continue;
1829 			}
1830 #endif
1831 			if (prison_restrict_ip6(tpr, ip6)) {
1832 				if (ip6 != NULL)
1833 					ip6 = NULL;
1834 				else
1835 					redo_ip6 = 1;
1836 			}
1837 		}
1838 		mtx_unlock(&pr->pr_mtx);
1839 	}
1840 #endif
1841 
1842 	/* Let the modules do their work. */
1843 	sx_downgrade(&allprison_lock);
1844 	if (created) {
1845 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1846 		if (error) {
1847 			prison_deref(pr, PD_LIST_SLOCKED);
1848 			goto done_errmsg;
1849 		}
1850 	}
1851 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1852 	if (error) {
1853 		prison_deref(pr, created
1854 		    ? PD_LIST_SLOCKED
1855 		    : PD_DEREF | PD_LIST_SLOCKED);
1856 		goto done_errmsg;
1857 	}
1858 
1859 	/* Attach this process to the prison if requested. */
1860 	if (flags & JAIL_ATTACH) {
1861 		mtx_lock(&pr->pr_mtx);
1862 		error = do_jail_attach(td, pr);
1863 		if (error) {
1864 			vfs_opterror(opts, "attach failed");
1865 			if (!created)
1866 				prison_deref(pr, PD_DEREF);
1867 			goto done_errmsg;
1868 		}
1869 	}
1870 
1871 #ifdef RACCT
1872 	if (racct_enable && !created) {
1873 		if (!(flags & JAIL_ATTACH))
1874 			sx_sunlock(&allprison_lock);
1875 		prison_racct_modify(pr);
1876 		if (!(flags & JAIL_ATTACH))
1877 			sx_slock(&allprison_lock);
1878 	}
1879 #endif
1880 
1881 	td->td_retval[0] = pr->pr_id;
1882 
1883 	/*
1884 	 * Now that it is all there, drop the temporary reference from existing
1885 	 * prisons.  Or add a reference to newly created persistent prisons
1886 	 * (which was not done earlier so that the prison would not be publicly
1887 	 * visible).
1888 	 */
1889 	if (!created) {
1890 		prison_deref(pr, (flags & JAIL_ATTACH)
1891 		    ? PD_DEREF
1892 		    : PD_DEREF | PD_LIST_SLOCKED);
1893 	} else {
1894 		if (pr_flags & PR_PERSIST) {
1895 			mtx_lock(&pr->pr_mtx);
1896 			pr->pr_ref++;
1897 			pr->pr_uref++;
1898 			mtx_unlock(&pr->pr_mtx);
1899 		}
1900 		if (!(flags & JAIL_ATTACH))
1901 			sx_sunlock(&allprison_lock);
1902 	}
1903 
1904 	goto done_errmsg;
1905 
1906  done_deref_locked:
1907 	prison_deref(pr, created
1908 	    ? PD_LOCKED | PD_LIST_XLOCKED
1909 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1910 	goto done_releroot;
1911  done_unlock_list:
1912 	sx_xunlock(&allprison_lock);
1913  done_releroot:
1914 	if (root != NULL)
1915 		vrele(root);
1916  done_errmsg:
1917 	if (error) {
1918 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1919 		if (errmsg_len > 0) {
1920 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1921 			if (errmsg_pos > 0) {
1922 				if (optuio->uio_segflg == UIO_SYSSPACE)
1923 					bcopy(errmsg,
1924 					   optuio->uio_iov[errmsg_pos].iov_base,
1925 					   errmsg_len);
1926 				else
1927 					copyout(errmsg,
1928 					   optuio->uio_iov[errmsg_pos].iov_base,
1929 					   errmsg_len);
1930 			}
1931 		}
1932 	}
1933  done_free:
1934 #ifdef INET
1935 	free(ip4, M_PRISON);
1936 #endif
1937 #ifdef INET6
1938 	free(ip6, M_PRISON);
1939 #endif
1940 	if (g_path != NULL)
1941 		free(g_path, M_TEMP);
1942 	vfs_freeopts(opts);
1943 	return (error);
1944 }
1945 
1946 
1947 /*
1948  * struct jail_get_args {
1949  *	struct iovec *iovp;
1950  *	unsigned int iovcnt;
1951  *	int flags;
1952  * };
1953  */
1954 int
1955 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1956 {
1957 	struct uio *auio;
1958 	int error;
1959 
1960 	/* Check that we have an even number of iovecs. */
1961 	if (uap->iovcnt & 1)
1962 		return (EINVAL);
1963 
1964 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1965 	if (error)
1966 		return (error);
1967 	error = kern_jail_get(td, auio, uap->flags);
1968 	if (error == 0)
1969 		error = copyout(auio->uio_iov, uap->iovp,
1970 		    uap->iovcnt * sizeof (struct iovec));
1971 	free(auio, M_IOV);
1972 	return (error);
1973 }
1974 
1975 int
1976 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1977 {
1978 	struct prison *pr, *mypr;
1979 	struct vfsopt *opt;
1980 	struct vfsoptlist *opts;
1981 	char *errmsg, *name;
1982 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1983 
1984 	if (flags & ~JAIL_GET_MASK)
1985 		return (EINVAL);
1986 
1987 	/* Get the parameter list. */
1988 	error = vfs_buildopts(optuio, &opts);
1989 	if (error)
1990 		return (error);
1991 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1992 	mypr = td->td_ucred->cr_prison;
1993 
1994 	/*
1995 	 * Find the prison specified by one of: lastjid, jid, name.
1996 	 */
1997 	sx_slock(&allprison_lock);
1998 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1999 	if (error == 0) {
2000 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2001 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2002 				mtx_lock(&pr->pr_mtx);
2003 				if (pr->pr_ref > 0 &&
2004 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2005 					break;
2006 				mtx_unlock(&pr->pr_mtx);
2007 			}
2008 		}
2009 		if (pr != NULL)
2010 			goto found_prison;
2011 		error = ENOENT;
2012 		vfs_opterror(opts, "no jail after %d", jid);
2013 		goto done_unlock_list;
2014 	} else if (error != ENOENT)
2015 		goto done_unlock_list;
2016 
2017 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2018 	if (error == 0) {
2019 		if (jid != 0) {
2020 			pr = prison_find_child(mypr, jid);
2021 			if (pr != NULL) {
2022 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2023 					mtx_unlock(&pr->pr_mtx);
2024 					error = ENOENT;
2025 					vfs_opterror(opts, "jail %d is dying",
2026 					    jid);
2027 					goto done_unlock_list;
2028 				}
2029 				goto found_prison;
2030 			}
2031 			error = ENOENT;
2032 			vfs_opterror(opts, "jail %d not found", jid);
2033 			goto done_unlock_list;
2034 		}
2035 	} else if (error != ENOENT)
2036 		goto done_unlock_list;
2037 
2038 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2039 	if (error == 0) {
2040 		if (len == 0 || name[len - 1] != '\0') {
2041 			error = EINVAL;
2042 			goto done_unlock_list;
2043 		}
2044 		pr = prison_find_name(mypr, name);
2045 		if (pr != NULL) {
2046 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2047 				mtx_unlock(&pr->pr_mtx);
2048 				error = ENOENT;
2049 				vfs_opterror(opts, "jail \"%s\" is dying",
2050 				    name);
2051 				goto done_unlock_list;
2052 			}
2053 			goto found_prison;
2054 		}
2055 		error = ENOENT;
2056 		vfs_opterror(opts, "jail \"%s\" not found", name);
2057 		goto done_unlock_list;
2058 	} else if (error != ENOENT)
2059 		goto done_unlock_list;
2060 
2061 	vfs_opterror(opts, "no jail specified");
2062 	error = ENOENT;
2063 	goto done_unlock_list;
2064 
2065  found_prison:
2066 	/* Get the parameters of the prison. */
2067 	pr->pr_ref++;
2068 	locked = PD_LOCKED;
2069 	td->td_retval[0] = pr->pr_id;
2070 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2071 	if (error != 0 && error != ENOENT)
2072 		goto done_deref;
2073 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2074 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2075 	if (error != 0 && error != ENOENT)
2076 		goto done_deref;
2077 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2078 	if (error != 0 && error != ENOENT)
2079 		goto done_deref;
2080 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2081 	    sizeof(pr->pr_cpuset->cs_id));
2082 	if (error != 0 && error != ENOENT)
2083 		goto done_deref;
2084 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2085 	if (error != 0 && error != ENOENT)
2086 		goto done_deref;
2087 #ifdef INET
2088 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2089 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2090 	if (error != 0 && error != ENOENT)
2091 		goto done_deref;
2092 #endif
2093 #ifdef INET6
2094 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2095 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2096 	if (error != 0 && error != ENOENT)
2097 		goto done_deref;
2098 #endif
2099 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2100 	    sizeof(pr->pr_securelevel));
2101 	if (error != 0 && error != ENOENT)
2102 		goto done_deref;
2103 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2104 	    sizeof(pr->pr_childcount));
2105 	if (error != 0 && error != ENOENT)
2106 		goto done_deref;
2107 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2108 	    sizeof(pr->pr_childmax));
2109 	if (error != 0 && error != ENOENT)
2110 		goto done_deref;
2111 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2112 	if (error != 0 && error != ENOENT)
2113 		goto done_deref;
2114 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2115 	if (error != 0 && error != ENOENT)
2116 		goto done_deref;
2117 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2118 	if (error != 0 && error != ENOENT)
2119 		goto done_deref;
2120 #ifdef COMPAT_FREEBSD32
2121 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2122 		uint32_t hid32 = pr->pr_hostid;
2123 
2124 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2125 	} else
2126 #endif
2127 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2128 	    sizeof(pr->pr_hostid));
2129 	if (error != 0 && error != ENOENT)
2130 		goto done_deref;
2131 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2132 	    sizeof(pr->pr_enforce_statfs));
2133 	if (error != 0 && error != ENOENT)
2134 		goto done_deref;
2135 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2136 	    sizeof(pr->pr_devfs_rsnum));
2137 	if (error != 0 && error != ENOENT)
2138 		goto done_deref;
2139 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2140 	    fi++) {
2141 		if (pr_flag_names[fi] == NULL)
2142 			continue;
2143 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2144 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2145 		if (error != 0 && error != ENOENT)
2146 			goto done_deref;
2147 		i = !i;
2148 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2149 		if (error != 0 && error != ENOENT)
2150 			goto done_deref;
2151 	}
2152 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2153 	    fi++) {
2154 		i = pr->pr_flags &
2155 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2156 		i = pr_flag_jailsys[fi].disable &&
2157 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2158 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2159 		    : JAIL_SYS_INHERIT;
2160 		error =
2161 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2162 		if (error != 0 && error != ENOENT)
2163 			goto done_deref;
2164 	}
2165 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2166 	    fi++) {
2167 		if (pr_allow_names[fi] == NULL)
2168 			continue;
2169 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2170 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2171 		if (error != 0 && error != ENOENT)
2172 			goto done_deref;
2173 		i = !i;
2174 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2175 		if (error != 0 && error != ENOENT)
2176 			goto done_deref;
2177 	}
2178 	i = (pr->pr_uref == 0);
2179 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2180 	if (error != 0 && error != ENOENT)
2181 		goto done_deref;
2182 	i = !i;
2183 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2184 	if (error != 0 && error != ENOENT)
2185 		goto done_deref;
2186 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2187 	    sizeof(pr->pr_osreldate));
2188 	if (error != 0 && error != ENOENT)
2189 		goto done_deref;
2190 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2191 	if (error != 0 && error != ENOENT)
2192 		goto done_deref;
2193 
2194 	/* Get the module parameters. */
2195 	mtx_unlock(&pr->pr_mtx);
2196 	locked = 0;
2197 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2198 	if (error)
2199 		goto done_deref;
2200 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2201 
2202 	/* By now, all parameters should have been noted. */
2203 	TAILQ_FOREACH(opt, opts, link) {
2204 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2205 			error = EINVAL;
2206 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2207 			goto done_errmsg;
2208 		}
2209 	}
2210 
2211 	/* Write the fetched parameters back to userspace. */
2212 	error = 0;
2213 	TAILQ_FOREACH(opt, opts, link) {
2214 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2215 			pos = 2 * opt->pos + 1;
2216 			optuio->uio_iov[pos].iov_len = opt->len;
2217 			if (opt->value != NULL) {
2218 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2219 					bcopy(opt->value,
2220 					    optuio->uio_iov[pos].iov_base,
2221 					    opt->len);
2222 				} else {
2223 					error = copyout(opt->value,
2224 					    optuio->uio_iov[pos].iov_base,
2225 					    opt->len);
2226 					if (error)
2227 						break;
2228 				}
2229 			}
2230 		}
2231 	}
2232 	goto done_errmsg;
2233 
2234  done_deref:
2235 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2236 	goto done_errmsg;
2237 
2238  done_unlock_list:
2239 	sx_sunlock(&allprison_lock);
2240  done_errmsg:
2241 	if (error && errmsg_pos >= 0) {
2242 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2243 		errmsg_pos = 2 * errmsg_pos + 1;
2244 		if (errmsg_len > 0) {
2245 			if (optuio->uio_segflg == UIO_SYSSPACE)
2246 				bcopy(errmsg,
2247 				    optuio->uio_iov[errmsg_pos].iov_base,
2248 				    errmsg_len);
2249 			else
2250 				copyout(errmsg,
2251 				    optuio->uio_iov[errmsg_pos].iov_base,
2252 				    errmsg_len);
2253 		}
2254 	}
2255 	vfs_freeopts(opts);
2256 	return (error);
2257 }
2258 
2259 
2260 /*
2261  * struct jail_remove_args {
2262  *	int jid;
2263  * };
2264  */
2265 int
2266 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2267 {
2268 	struct prison *pr, *cpr, *lpr, *tpr;
2269 	int descend, error;
2270 
2271 	error = priv_check(td, PRIV_JAIL_REMOVE);
2272 	if (error)
2273 		return (error);
2274 
2275 	sx_xlock(&allprison_lock);
2276 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2277 	if (pr == NULL) {
2278 		sx_xunlock(&allprison_lock);
2279 		return (EINVAL);
2280 	}
2281 
2282 	/* Remove all descendants of this prison, then remove this prison. */
2283 	pr->pr_ref++;
2284 	pr->pr_flags |= PR_REMOVE;
2285 	if (!LIST_EMPTY(&pr->pr_children)) {
2286 		mtx_unlock(&pr->pr_mtx);
2287 		lpr = NULL;
2288 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2289 			mtx_lock(&cpr->pr_mtx);
2290 			if (cpr->pr_ref > 0) {
2291 				tpr = cpr;
2292 				cpr->pr_ref++;
2293 				cpr->pr_flags |= PR_REMOVE;
2294 			} else {
2295 				/* Already removed - do not do it again. */
2296 				tpr = NULL;
2297 			}
2298 			mtx_unlock(&cpr->pr_mtx);
2299 			if (lpr != NULL) {
2300 				mtx_lock(&lpr->pr_mtx);
2301 				prison_remove_one(lpr);
2302 				sx_xlock(&allprison_lock);
2303 			}
2304 			lpr = tpr;
2305 		}
2306 		if (lpr != NULL) {
2307 			mtx_lock(&lpr->pr_mtx);
2308 			prison_remove_one(lpr);
2309 			sx_xlock(&allprison_lock);
2310 		}
2311 		mtx_lock(&pr->pr_mtx);
2312 	}
2313 	prison_remove_one(pr);
2314 	return (0);
2315 }
2316 
2317 static void
2318 prison_remove_one(struct prison *pr)
2319 {
2320 	struct proc *p;
2321 	int deuref;
2322 
2323 	/* If the prison was persistent, it is not anymore. */
2324 	deuref = 0;
2325 	if (pr->pr_flags & PR_PERSIST) {
2326 		pr->pr_ref--;
2327 		deuref = PD_DEUREF;
2328 		pr->pr_flags &= ~PR_PERSIST;
2329 	}
2330 
2331 	/*
2332 	 * jail_remove added a reference.  If that's the only one, remove
2333 	 * the prison now.
2334 	 */
2335 	KASSERT(pr->pr_ref > 0,
2336 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2337 	if (pr->pr_ref == 1) {
2338 		prison_deref(pr,
2339 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2340 		return;
2341 	}
2342 
2343 	mtx_unlock(&pr->pr_mtx);
2344 	sx_xunlock(&allprison_lock);
2345 	/*
2346 	 * Kill all processes unfortunate enough to be attached to this prison.
2347 	 */
2348 	sx_slock(&allproc_lock);
2349 	LIST_FOREACH(p, &allproc, p_list) {
2350 		PROC_LOCK(p);
2351 		if (p->p_state != PRS_NEW && p->p_ucred &&
2352 		    p->p_ucred->cr_prison == pr)
2353 			kern_psignal(p, SIGKILL);
2354 		PROC_UNLOCK(p);
2355 	}
2356 	sx_sunlock(&allproc_lock);
2357 	/* Remove the temporary reference added by jail_remove. */
2358 	prison_deref(pr, deuref | PD_DEREF);
2359 }
2360 
2361 
2362 /*
2363  * struct jail_attach_args {
2364  *	int jid;
2365  * };
2366  */
2367 int
2368 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2369 {
2370 	struct prison *pr;
2371 	int error;
2372 
2373 	error = priv_check(td, PRIV_JAIL_ATTACH);
2374 	if (error)
2375 		return (error);
2376 
2377 	sx_slock(&allprison_lock);
2378 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2379 	if (pr == NULL) {
2380 		sx_sunlock(&allprison_lock);
2381 		return (EINVAL);
2382 	}
2383 
2384 	/*
2385 	 * Do not allow a process to attach to a prison that is not
2386 	 * considered to be "alive".
2387 	 */
2388 	if (pr->pr_uref == 0) {
2389 		mtx_unlock(&pr->pr_mtx);
2390 		sx_sunlock(&allprison_lock);
2391 		return (EINVAL);
2392 	}
2393 
2394 	return (do_jail_attach(td, pr));
2395 }
2396 
2397 static int
2398 do_jail_attach(struct thread *td, struct prison *pr)
2399 {
2400 	struct prison *ppr;
2401 	struct proc *p;
2402 	struct ucred *newcred, *oldcred;
2403 	int error;
2404 
2405 	/*
2406 	 * XXX: Note that there is a slight race here if two threads
2407 	 * in the same privileged process attempt to attach to two
2408 	 * different jails at the same time.  It is important for
2409 	 * user processes not to do this, or they might end up with
2410 	 * a process root from one prison, but attached to the jail
2411 	 * of another.
2412 	 */
2413 	pr->pr_ref++;
2414 	pr->pr_uref++;
2415 	mtx_unlock(&pr->pr_mtx);
2416 
2417 	/* Let modules do whatever they need to prepare for attaching. */
2418 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2419 	if (error) {
2420 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2421 		return (error);
2422 	}
2423 	sx_sunlock(&allprison_lock);
2424 
2425 	/*
2426 	 * Reparent the newly attached process to this jail.
2427 	 */
2428 	ppr = td->td_ucred->cr_prison;
2429 	p = td->td_proc;
2430 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2431 	if (error)
2432 		goto e_revert_osd;
2433 
2434 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2435 	if ((error = change_dir(pr->pr_root, td)) != 0)
2436 		goto e_unlock;
2437 #ifdef MAC
2438 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2439 		goto e_unlock;
2440 #endif
2441 	VOP_UNLOCK(pr->pr_root, 0);
2442 	if ((error = pwd_chroot(td, pr->pr_root)))
2443 		goto e_revert_osd;
2444 
2445 	newcred = crget();
2446 	PROC_LOCK(p);
2447 	oldcred = p->p_ucred;
2448 	setsugid(p);
2449 	crcopy(newcred, oldcred);
2450 	newcred->cr_prison = pr;
2451 	proc_set_cred(p, newcred);
2452 	PROC_UNLOCK(p);
2453 #ifdef RACCT
2454 	racct_proc_ucred_changed(p, oldcred, newcred);
2455 #endif
2456 	crfree(oldcred);
2457 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2458 	return (0);
2459  e_unlock:
2460 	VOP_UNLOCK(pr->pr_root, 0);
2461  e_revert_osd:
2462 	/* Tell modules this thread is still in its old jail after all. */
2463 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2464 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2465 	return (error);
2466 }
2467 
2468 
2469 /*
2470  * Returns a locked prison instance, or NULL on failure.
2471  */
2472 struct prison *
2473 prison_find(int prid)
2474 {
2475 	struct prison *pr;
2476 
2477 	sx_assert(&allprison_lock, SX_LOCKED);
2478 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2479 		if (pr->pr_id == prid) {
2480 			mtx_lock(&pr->pr_mtx);
2481 			if (pr->pr_ref > 0)
2482 				return (pr);
2483 			mtx_unlock(&pr->pr_mtx);
2484 		}
2485 	}
2486 	return (NULL);
2487 }
2488 
2489 /*
2490  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2491  */
2492 struct prison *
2493 prison_find_child(struct prison *mypr, int prid)
2494 {
2495 	struct prison *pr;
2496 	int descend;
2497 
2498 	sx_assert(&allprison_lock, SX_LOCKED);
2499 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2500 		if (pr->pr_id == prid) {
2501 			mtx_lock(&pr->pr_mtx);
2502 			if (pr->pr_ref > 0)
2503 				return (pr);
2504 			mtx_unlock(&pr->pr_mtx);
2505 		}
2506 	}
2507 	return (NULL);
2508 }
2509 
2510 /*
2511  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2512  */
2513 struct prison *
2514 prison_find_name(struct prison *mypr, const char *name)
2515 {
2516 	struct prison *pr, *deadpr;
2517 	size_t mylen;
2518 	int descend;
2519 
2520 	sx_assert(&allprison_lock, SX_LOCKED);
2521 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2522  again:
2523 	deadpr = NULL;
2524 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2525 		if (!strcmp(pr->pr_name + mylen, name)) {
2526 			mtx_lock(&pr->pr_mtx);
2527 			if (pr->pr_ref > 0) {
2528 				if (pr->pr_uref > 0)
2529 					return (pr);
2530 				deadpr = pr;
2531 			}
2532 			mtx_unlock(&pr->pr_mtx);
2533 		}
2534 	}
2535 	/* There was no valid prison - perhaps there was a dying one. */
2536 	if (deadpr != NULL) {
2537 		mtx_lock(&deadpr->pr_mtx);
2538 		if (deadpr->pr_ref == 0) {
2539 			mtx_unlock(&deadpr->pr_mtx);
2540 			goto again;
2541 		}
2542 	}
2543 	return (deadpr);
2544 }
2545 
2546 /*
2547  * See if a prison has the specific flag set.
2548  */
2549 int
2550 prison_flag(struct ucred *cred, unsigned flag)
2551 {
2552 
2553 	/* This is an atomic read, so no locking is necessary. */
2554 	return (cred->cr_prison->pr_flags & flag);
2555 }
2556 
2557 int
2558 prison_allow(struct ucred *cred, unsigned flag)
2559 {
2560 
2561 	/* This is an atomic read, so no locking is necessary. */
2562 	return (cred->cr_prison->pr_allow & flag);
2563 }
2564 
2565 /*
2566  * Remove a prison reference.  If that was the last reference, remove the
2567  * prison itself - but not in this context in case there are locks held.
2568  */
2569 void
2570 prison_free_locked(struct prison *pr)
2571 {
2572 
2573 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2574 	pr->pr_ref--;
2575 	if (pr->pr_ref == 0) {
2576 		mtx_unlock(&pr->pr_mtx);
2577 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2578 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2579 		return;
2580 	}
2581 	mtx_unlock(&pr->pr_mtx);
2582 }
2583 
2584 void
2585 prison_free(struct prison *pr)
2586 {
2587 
2588 	mtx_lock(&pr->pr_mtx);
2589 	prison_free_locked(pr);
2590 }
2591 
2592 static void
2593 prison_complete(void *context, int pending)
2594 {
2595 
2596 	prison_deref((struct prison *)context, 0);
2597 }
2598 
2599 /*
2600  * Remove a prison reference (usually).  This internal version assumes no
2601  * mutexes are held, except perhaps the prison itself.  If there are no more
2602  * references, release and delist the prison.  On completion, the prison lock
2603  * and the allprison lock are both unlocked.
2604  */
2605 static void
2606 prison_deref(struct prison *pr, int flags)
2607 {
2608 	struct prison *ppr, *tpr;
2609 
2610 	if (!(flags & PD_LOCKED))
2611 		mtx_lock(&pr->pr_mtx);
2612 	for (;;) {
2613 		if (flags & PD_DEUREF) {
2614 			pr->pr_uref--;
2615 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2616 		}
2617 		if (flags & PD_DEREF)
2618 			pr->pr_ref--;
2619 		/* If the prison still has references, nothing else to do. */
2620 		if (pr->pr_ref > 0) {
2621 			mtx_unlock(&pr->pr_mtx);
2622 			if (flags & PD_LIST_SLOCKED)
2623 				sx_sunlock(&allprison_lock);
2624 			else if (flags & PD_LIST_XLOCKED)
2625 				sx_xunlock(&allprison_lock);
2626 			return;
2627 		}
2628 
2629 		mtx_unlock(&pr->pr_mtx);
2630 		if (flags & PD_LIST_SLOCKED) {
2631 			if (!sx_try_upgrade(&allprison_lock)) {
2632 				sx_sunlock(&allprison_lock);
2633 				sx_xlock(&allprison_lock);
2634 			}
2635 		} else if (!(flags & PD_LIST_XLOCKED))
2636 			sx_xlock(&allprison_lock);
2637 
2638 		TAILQ_REMOVE(&allprison, pr, pr_list);
2639 		LIST_REMOVE(pr, pr_sibling);
2640 		ppr = pr->pr_parent;
2641 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2642 			tpr->pr_childcount--;
2643 		sx_xunlock(&allprison_lock);
2644 
2645 #ifdef VIMAGE
2646 		if (pr->pr_vnet != ppr->pr_vnet)
2647 			vnet_destroy(pr->pr_vnet);
2648 #endif
2649 		if (pr->pr_root != NULL)
2650 			vrele(pr->pr_root);
2651 		mtx_destroy(&pr->pr_mtx);
2652 #ifdef INET
2653 		free(pr->pr_ip4, M_PRISON);
2654 #endif
2655 #ifdef INET6
2656 		free(pr->pr_ip6, M_PRISON);
2657 #endif
2658 		if (pr->pr_cpuset != NULL)
2659 			cpuset_rel(pr->pr_cpuset);
2660 		osd_jail_exit(pr);
2661 #ifdef RACCT
2662 		if (racct_enable)
2663 			prison_racct_detach(pr);
2664 #endif
2665 		free(pr, M_PRISON);
2666 
2667 		/* Removing a prison frees a reference on its parent. */
2668 		pr = ppr;
2669 		mtx_lock(&pr->pr_mtx);
2670 		flags = PD_DEREF | PD_DEUREF;
2671 	}
2672 }
2673 
2674 void
2675 prison_hold_locked(struct prison *pr)
2676 {
2677 
2678 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2679 	KASSERT(pr->pr_ref > 0,
2680 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2681 	pr->pr_ref++;
2682 }
2683 
2684 void
2685 prison_hold(struct prison *pr)
2686 {
2687 
2688 	mtx_lock(&pr->pr_mtx);
2689 	prison_hold_locked(pr);
2690 	mtx_unlock(&pr->pr_mtx);
2691 }
2692 
2693 void
2694 prison_proc_hold(struct prison *pr)
2695 {
2696 
2697 	mtx_lock(&pr->pr_mtx);
2698 	KASSERT(pr->pr_uref > 0,
2699 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2700 	pr->pr_uref++;
2701 	mtx_unlock(&pr->pr_mtx);
2702 }
2703 
2704 void
2705 prison_proc_free(struct prison *pr)
2706 {
2707 
2708 	mtx_lock(&pr->pr_mtx);
2709 	KASSERT(pr->pr_uref > 0,
2710 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2711 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2712 }
2713 
2714 
2715 #ifdef INET
2716 /*
2717  * Restrict a prison's IP address list with its parent's, possibly replacing
2718  * it.  Return true if the replacement buffer was used (or would have been).
2719  */
2720 static int
2721 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2722 {
2723 	int ii, ij, used;
2724 	struct prison *ppr;
2725 
2726 	ppr = pr->pr_parent;
2727 	if (!(pr->pr_flags & PR_IP4_USER)) {
2728 		/* This has no user settings, so just copy the parent's list. */
2729 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2730 			/*
2731 			 * There's no room for the parent's list.  Use the
2732 			 * new list buffer, which is assumed to be big enough
2733 			 * (if it was passed).  If there's no buffer, try to
2734 			 * allocate one.
2735 			 */
2736 			used = 1;
2737 			if (newip4 == NULL) {
2738 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2739 				    M_PRISON, M_NOWAIT);
2740 				if (newip4 != NULL)
2741 					used = 0;
2742 			}
2743 			if (newip4 != NULL) {
2744 				bcopy(ppr->pr_ip4, newip4,
2745 				    ppr->pr_ip4s * sizeof(*newip4));
2746 				free(pr->pr_ip4, M_PRISON);
2747 				pr->pr_ip4 = newip4;
2748 				pr->pr_ip4s = ppr->pr_ip4s;
2749 			}
2750 			return (used);
2751 		}
2752 		pr->pr_ip4s = ppr->pr_ip4s;
2753 		if (pr->pr_ip4s > 0)
2754 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2755 			    pr->pr_ip4s * sizeof(*newip4));
2756 		else if (pr->pr_ip4 != NULL) {
2757 			free(pr->pr_ip4, M_PRISON);
2758 			pr->pr_ip4 = NULL;
2759 		}
2760 	} else if (pr->pr_ip4s > 0) {
2761 		/* Remove addresses that aren't in the parent. */
2762 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2763 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2764 				break;
2765 		if (ij < ppr->pr_ip4s)
2766 			ii = 1;
2767 		else {
2768 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2769 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2770 			ii = 0;
2771 		}
2772 		for (ij = 1; ii < pr->pr_ip4s; ) {
2773 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2774 				ii++;
2775 				continue;
2776 			}
2777 			switch (ij >= ppr->pr_ip4s ? -1 :
2778 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2779 			case -1:
2780 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2781 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2782 				break;
2783 			case 0:
2784 				ii++;
2785 				ij++;
2786 				break;
2787 			case 1:
2788 				ij++;
2789 				break;
2790 			}
2791 		}
2792 		if (pr->pr_ip4s == 0) {
2793 			free(pr->pr_ip4, M_PRISON);
2794 			pr->pr_ip4 = NULL;
2795 		}
2796 	}
2797 	return (0);
2798 }
2799 
2800 /*
2801  * Pass back primary IPv4 address of this jail.
2802  *
2803  * If not restricted return success but do not alter the address.  Caller has
2804  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2805  *
2806  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2807  * Address returned in NBO.
2808  */
2809 int
2810 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2811 {
2812 	struct prison *pr;
2813 
2814 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2815 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2816 
2817 	pr = cred->cr_prison;
2818 	if (!(pr->pr_flags & PR_IP4))
2819 		return (0);
2820 	mtx_lock(&pr->pr_mtx);
2821 	if (!(pr->pr_flags & PR_IP4)) {
2822 		mtx_unlock(&pr->pr_mtx);
2823 		return (0);
2824 	}
2825 	if (pr->pr_ip4 == NULL) {
2826 		mtx_unlock(&pr->pr_mtx);
2827 		return (EAFNOSUPPORT);
2828 	}
2829 
2830 	ia->s_addr = pr->pr_ip4[0].s_addr;
2831 	mtx_unlock(&pr->pr_mtx);
2832 	return (0);
2833 }
2834 
2835 /*
2836  * Return 1 if we should do proper source address selection or are not jailed.
2837  * We will return 0 if we should bypass source address selection in favour
2838  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2839  * returned in NBO.
2840  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2841  */
2842 int
2843 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2844 {
2845 	struct prison *pr;
2846 	struct in_addr lia;
2847 	int error;
2848 
2849 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2850 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2851 
2852 	if (!jailed(cred))
2853 		return (1);
2854 
2855 	pr = cred->cr_prison;
2856 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2857 		return (1);
2858 
2859 	lia.s_addr = INADDR_ANY;
2860 	error = prison_get_ip4(cred, &lia);
2861 	if (error)
2862 		return (error);
2863 	if (lia.s_addr == INADDR_ANY)
2864 		return (1);
2865 
2866 	ia->s_addr = lia.s_addr;
2867 	return (0);
2868 }
2869 
2870 /*
2871  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2872  */
2873 int
2874 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2875 {
2876 
2877 	if (pr1 == pr2)
2878 		return (1);
2879 
2880 	/*
2881 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2882 	 * existing prisons.
2883 	 */
2884 	while (pr1 != &prison0 &&
2885 #ifdef VIMAGE
2886 	       !(pr1->pr_flags & PR_VNET) &&
2887 #endif
2888 	       !(pr1->pr_flags & PR_IP4_USER))
2889 		pr1 = pr1->pr_parent;
2890 	while (pr2 != &prison0 &&
2891 #ifdef VIMAGE
2892 	       !(pr2->pr_flags & PR_VNET) &&
2893 #endif
2894 	       !(pr2->pr_flags & PR_IP4_USER))
2895 		pr2 = pr2->pr_parent;
2896 	return (pr1 == pr2);
2897 }
2898 
2899 /*
2900  * Make sure our (source) address is set to something meaningful to this
2901  * jail.
2902  *
2903  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2904  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2905  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2906  */
2907 int
2908 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2909 {
2910 	struct prison *pr;
2911 	struct in_addr ia0;
2912 	int error;
2913 
2914 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2915 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2916 
2917 	pr = cred->cr_prison;
2918 	if (!(pr->pr_flags & PR_IP4))
2919 		return (0);
2920 	mtx_lock(&pr->pr_mtx);
2921 	if (!(pr->pr_flags & PR_IP4)) {
2922 		mtx_unlock(&pr->pr_mtx);
2923 		return (0);
2924 	}
2925 	if (pr->pr_ip4 == NULL) {
2926 		mtx_unlock(&pr->pr_mtx);
2927 		return (EAFNOSUPPORT);
2928 	}
2929 
2930 	ia0.s_addr = ntohl(ia->s_addr);
2931 	if (ia0.s_addr == INADDR_LOOPBACK) {
2932 		ia->s_addr = pr->pr_ip4[0].s_addr;
2933 		mtx_unlock(&pr->pr_mtx);
2934 		return (0);
2935 	}
2936 
2937 	if (ia0.s_addr == INADDR_ANY) {
2938 		/*
2939 		 * In case there is only 1 IPv4 address, bind directly.
2940 		 */
2941 		if (pr->pr_ip4s == 1)
2942 			ia->s_addr = pr->pr_ip4[0].s_addr;
2943 		mtx_unlock(&pr->pr_mtx);
2944 		return (0);
2945 	}
2946 
2947 	error = _prison_check_ip4(pr, ia);
2948 	mtx_unlock(&pr->pr_mtx);
2949 	return (error);
2950 }
2951 
2952 /*
2953  * Rewrite destination address in case we will connect to loopback address.
2954  *
2955  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2956  * Address passed in in NBO and returned in NBO.
2957  */
2958 int
2959 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2960 {
2961 	struct prison *pr;
2962 
2963 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2964 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2965 
2966 	pr = cred->cr_prison;
2967 	if (!(pr->pr_flags & PR_IP4))
2968 		return (0);
2969 	mtx_lock(&pr->pr_mtx);
2970 	if (!(pr->pr_flags & PR_IP4)) {
2971 		mtx_unlock(&pr->pr_mtx);
2972 		return (0);
2973 	}
2974 	if (pr->pr_ip4 == NULL) {
2975 		mtx_unlock(&pr->pr_mtx);
2976 		return (EAFNOSUPPORT);
2977 	}
2978 
2979 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2980 		ia->s_addr = pr->pr_ip4[0].s_addr;
2981 		mtx_unlock(&pr->pr_mtx);
2982 		return (0);
2983 	}
2984 
2985 	/*
2986 	 * Return success because nothing had to be changed.
2987 	 */
2988 	mtx_unlock(&pr->pr_mtx);
2989 	return (0);
2990 }
2991 
2992 /*
2993  * Check if given address belongs to the jail referenced by cred/prison.
2994  *
2995  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2996  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2997  * doesn't allow IPv4.  Address passed in in NBO.
2998  */
2999 static int
3000 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
3001 {
3002 	int i, a, z, d;
3003 
3004 	/*
3005 	 * Check the primary IP.
3006 	 */
3007 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
3008 		return (0);
3009 
3010 	/*
3011 	 * All the other IPs are sorted so we can do a binary search.
3012 	 */
3013 	a = 0;
3014 	z = pr->pr_ip4s - 2;
3015 	while (a <= z) {
3016 		i = (a + z) / 2;
3017 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3018 		if (d > 0)
3019 			z = i - 1;
3020 		else if (d < 0)
3021 			a = i + 1;
3022 		else
3023 			return (0);
3024 	}
3025 
3026 	return (EADDRNOTAVAIL);
3027 }
3028 
3029 int
3030 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
3031 {
3032 	struct prison *pr;
3033 	int error;
3034 
3035 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3036 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3037 
3038 	pr = cred->cr_prison;
3039 	if (!(pr->pr_flags & PR_IP4))
3040 		return (0);
3041 	mtx_lock(&pr->pr_mtx);
3042 	if (!(pr->pr_flags & PR_IP4)) {
3043 		mtx_unlock(&pr->pr_mtx);
3044 		return (0);
3045 	}
3046 	if (pr->pr_ip4 == NULL) {
3047 		mtx_unlock(&pr->pr_mtx);
3048 		return (EAFNOSUPPORT);
3049 	}
3050 
3051 	error = _prison_check_ip4(pr, ia);
3052 	mtx_unlock(&pr->pr_mtx);
3053 	return (error);
3054 }
3055 #endif
3056 
3057 #ifdef INET6
3058 static int
3059 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3060 {
3061 	int ii, ij, used;
3062 	struct prison *ppr;
3063 
3064 	ppr = pr->pr_parent;
3065 	if (!(pr->pr_flags & PR_IP6_USER)) {
3066 		/* This has no user settings, so just copy the parent's list. */
3067 		if (pr->pr_ip6s < ppr->pr_ip6s) {
3068 			/*
3069 			 * There's no room for the parent's list.  Use the
3070 			 * new list buffer, which is assumed to be big enough
3071 			 * (if it was passed).  If there's no buffer, try to
3072 			 * allocate one.
3073 			 */
3074 			used = 1;
3075 			if (newip6 == NULL) {
3076 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3077 				    M_PRISON, M_NOWAIT);
3078 				if (newip6 != NULL)
3079 					used = 0;
3080 			}
3081 			if (newip6 != NULL) {
3082 				bcopy(ppr->pr_ip6, newip6,
3083 				    ppr->pr_ip6s * sizeof(*newip6));
3084 				free(pr->pr_ip6, M_PRISON);
3085 				pr->pr_ip6 = newip6;
3086 				pr->pr_ip6s = ppr->pr_ip6s;
3087 			}
3088 			return (used);
3089 		}
3090 		pr->pr_ip6s = ppr->pr_ip6s;
3091 		if (pr->pr_ip6s > 0)
3092 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3093 			    pr->pr_ip6s * sizeof(*newip6));
3094 		else if (pr->pr_ip6 != NULL) {
3095 			free(pr->pr_ip6, M_PRISON);
3096 			pr->pr_ip6 = NULL;
3097 		}
3098 	} else if (pr->pr_ip6s > 0) {
3099 		/* Remove addresses that aren't in the parent. */
3100 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3101 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3102 			    &ppr->pr_ip6[ij]))
3103 				break;
3104 		if (ij < ppr->pr_ip6s)
3105 			ii = 1;
3106 		else {
3107 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3108 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3109 			ii = 0;
3110 		}
3111 		for (ij = 1; ii < pr->pr_ip6s; ) {
3112 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3113 			    &ppr->pr_ip6[0])) {
3114 				ii++;
3115 				continue;
3116 			}
3117 			switch (ij >= ppr->pr_ip6s ? -1 :
3118 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3119 			case -1:
3120 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3121 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3122 				break;
3123 			case 0:
3124 				ii++;
3125 				ij++;
3126 				break;
3127 			case 1:
3128 				ij++;
3129 				break;
3130 			}
3131 		}
3132 		if (pr->pr_ip6s == 0) {
3133 			free(pr->pr_ip6, M_PRISON);
3134 			pr->pr_ip6 = NULL;
3135 		}
3136 	}
3137 	return 0;
3138 }
3139 
3140 /*
3141  * Pass back primary IPv6 address for this jail.
3142  *
3143  * If not restricted return success but do not alter the address.  Caller has
3144  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3145  *
3146  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3147  */
3148 int
3149 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3150 {
3151 	struct prison *pr;
3152 
3153 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3154 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3155 
3156 	pr = cred->cr_prison;
3157 	if (!(pr->pr_flags & PR_IP6))
3158 		return (0);
3159 	mtx_lock(&pr->pr_mtx);
3160 	if (!(pr->pr_flags & PR_IP6)) {
3161 		mtx_unlock(&pr->pr_mtx);
3162 		return (0);
3163 	}
3164 	if (pr->pr_ip6 == NULL) {
3165 		mtx_unlock(&pr->pr_mtx);
3166 		return (EAFNOSUPPORT);
3167 	}
3168 
3169 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3170 	mtx_unlock(&pr->pr_mtx);
3171 	return (0);
3172 }
3173 
3174 /*
3175  * Return 1 if we should do proper source address selection or are not jailed.
3176  * We will return 0 if we should bypass source address selection in favour
3177  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3178  * returned in NBO.
3179  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3180  */
3181 int
3182 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3183 {
3184 	struct prison *pr;
3185 	struct in6_addr lia6;
3186 	int error;
3187 
3188 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3189 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3190 
3191 	if (!jailed(cred))
3192 		return (1);
3193 
3194 	pr = cred->cr_prison;
3195 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3196 		return (1);
3197 
3198 	lia6 = in6addr_any;
3199 	error = prison_get_ip6(cred, &lia6);
3200 	if (error)
3201 		return (error);
3202 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3203 		return (1);
3204 
3205 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3206 	return (0);
3207 }
3208 
3209 /*
3210  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3211  */
3212 int
3213 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3214 {
3215 
3216 	if (pr1 == pr2)
3217 		return (1);
3218 
3219 	while (pr1 != &prison0 &&
3220 #ifdef VIMAGE
3221 	       !(pr1->pr_flags & PR_VNET) &&
3222 #endif
3223 	       !(pr1->pr_flags & PR_IP6_USER))
3224 		pr1 = pr1->pr_parent;
3225 	while (pr2 != &prison0 &&
3226 #ifdef VIMAGE
3227 	       !(pr2->pr_flags & PR_VNET) &&
3228 #endif
3229 	       !(pr2->pr_flags & PR_IP6_USER))
3230 		pr2 = pr2->pr_parent;
3231 	return (pr1 == pr2);
3232 }
3233 
3234 /*
3235  * Make sure our (source) address is set to something meaningful to this jail.
3236  *
3237  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3238  * when needed while binding.
3239  *
3240  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3241  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3242  * doesn't allow IPv6.
3243  */
3244 int
3245 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3246 {
3247 	struct prison *pr;
3248 	int error;
3249 
3250 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3251 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3252 
3253 	pr = cred->cr_prison;
3254 	if (!(pr->pr_flags & PR_IP6))
3255 		return (0);
3256 	mtx_lock(&pr->pr_mtx);
3257 	if (!(pr->pr_flags & PR_IP6)) {
3258 		mtx_unlock(&pr->pr_mtx);
3259 		return (0);
3260 	}
3261 	if (pr->pr_ip6 == NULL) {
3262 		mtx_unlock(&pr->pr_mtx);
3263 		return (EAFNOSUPPORT);
3264 	}
3265 
3266 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3267 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3268 		mtx_unlock(&pr->pr_mtx);
3269 		return (0);
3270 	}
3271 
3272 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3273 		/*
3274 		 * In case there is only 1 IPv6 address, and v6only is true,
3275 		 * then bind directly.
3276 		 */
3277 		if (v6only != 0 && pr->pr_ip6s == 1)
3278 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3279 		mtx_unlock(&pr->pr_mtx);
3280 		return (0);
3281 	}
3282 
3283 	error = _prison_check_ip6(pr, ia6);
3284 	mtx_unlock(&pr->pr_mtx);
3285 	return (error);
3286 }
3287 
3288 /*
3289  * Rewrite destination address in case we will connect to loopback address.
3290  *
3291  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3292  */
3293 int
3294 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3295 {
3296 	struct prison *pr;
3297 
3298 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3299 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3300 
3301 	pr = cred->cr_prison;
3302 	if (!(pr->pr_flags & PR_IP6))
3303 		return (0);
3304 	mtx_lock(&pr->pr_mtx);
3305 	if (!(pr->pr_flags & PR_IP6)) {
3306 		mtx_unlock(&pr->pr_mtx);
3307 		return (0);
3308 	}
3309 	if (pr->pr_ip6 == NULL) {
3310 		mtx_unlock(&pr->pr_mtx);
3311 		return (EAFNOSUPPORT);
3312 	}
3313 
3314 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3315 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3316 		mtx_unlock(&pr->pr_mtx);
3317 		return (0);
3318 	}
3319 
3320 	/*
3321 	 * Return success because nothing had to be changed.
3322 	 */
3323 	mtx_unlock(&pr->pr_mtx);
3324 	return (0);
3325 }
3326 
3327 /*
3328  * Check if given address belongs to the jail referenced by cred/prison.
3329  *
3330  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3331  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3332  * doesn't allow IPv6.
3333  */
3334 static int
3335 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3336 {
3337 	int i, a, z, d;
3338 
3339 	/*
3340 	 * Check the primary IP.
3341 	 */
3342 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3343 		return (0);
3344 
3345 	/*
3346 	 * All the other IPs are sorted so we can do a binary search.
3347 	 */
3348 	a = 0;
3349 	z = pr->pr_ip6s - 2;
3350 	while (a <= z) {
3351 		i = (a + z) / 2;
3352 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3353 		if (d > 0)
3354 			z = i - 1;
3355 		else if (d < 0)
3356 			a = i + 1;
3357 		else
3358 			return (0);
3359 	}
3360 
3361 	return (EADDRNOTAVAIL);
3362 }
3363 
3364 int
3365 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3366 {
3367 	struct prison *pr;
3368 	int error;
3369 
3370 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3371 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3372 
3373 	pr = cred->cr_prison;
3374 	if (!(pr->pr_flags & PR_IP6))
3375 		return (0);
3376 	mtx_lock(&pr->pr_mtx);
3377 	if (!(pr->pr_flags & PR_IP6)) {
3378 		mtx_unlock(&pr->pr_mtx);
3379 		return (0);
3380 	}
3381 	if (pr->pr_ip6 == NULL) {
3382 		mtx_unlock(&pr->pr_mtx);
3383 		return (EAFNOSUPPORT);
3384 	}
3385 
3386 	error = _prison_check_ip6(pr, ia6);
3387 	mtx_unlock(&pr->pr_mtx);
3388 	return (error);
3389 }
3390 #endif
3391 
3392 /*
3393  * Check if a jail supports the given address family.
3394  *
3395  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3396  * if not.
3397  */
3398 int
3399 prison_check_af(struct ucred *cred, int af)
3400 {
3401 	struct prison *pr;
3402 	int error;
3403 
3404 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3405 
3406 	pr = cred->cr_prison;
3407 #ifdef VIMAGE
3408 	/* Prisons with their own network stack are not limited. */
3409 	if (prison_owns_vnet(cred))
3410 		return (0);
3411 #endif
3412 
3413 	error = 0;
3414 	switch (af)
3415 	{
3416 #ifdef INET
3417 	case AF_INET:
3418 		if (pr->pr_flags & PR_IP4)
3419 		{
3420 			mtx_lock(&pr->pr_mtx);
3421 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3422 				error = EAFNOSUPPORT;
3423 			mtx_unlock(&pr->pr_mtx);
3424 		}
3425 		break;
3426 #endif
3427 #ifdef INET6
3428 	case AF_INET6:
3429 		if (pr->pr_flags & PR_IP6)
3430 		{
3431 			mtx_lock(&pr->pr_mtx);
3432 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3433 				error = EAFNOSUPPORT;
3434 			mtx_unlock(&pr->pr_mtx);
3435 		}
3436 		break;
3437 #endif
3438 	case AF_LOCAL:
3439 	case AF_ROUTE:
3440 		break;
3441 	default:
3442 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3443 			error = EAFNOSUPPORT;
3444 	}
3445 	return (error);
3446 }
3447 
3448 /*
3449  * Check if given address belongs to the jail referenced by cred (wrapper to
3450  * prison_check_ip[46]).
3451  *
3452  * Returns 0 if jail doesn't restrict the address family or if address belongs
3453  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3454  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3455  */
3456 int
3457 prison_if(struct ucred *cred, struct sockaddr *sa)
3458 {
3459 #ifdef INET
3460 	struct sockaddr_in *sai;
3461 #endif
3462 #ifdef INET6
3463 	struct sockaddr_in6 *sai6;
3464 #endif
3465 	int error;
3466 
3467 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3468 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3469 
3470 #ifdef VIMAGE
3471 	if (prison_owns_vnet(cred))
3472 		return (0);
3473 #endif
3474 
3475 	error = 0;
3476 	switch (sa->sa_family)
3477 	{
3478 #ifdef INET
3479 	case AF_INET:
3480 		sai = (struct sockaddr_in *)sa;
3481 		error = prison_check_ip4(cred, &sai->sin_addr);
3482 		break;
3483 #endif
3484 #ifdef INET6
3485 	case AF_INET6:
3486 		sai6 = (struct sockaddr_in6 *)sa;
3487 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3488 		break;
3489 #endif
3490 	default:
3491 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3492 			error = EAFNOSUPPORT;
3493 	}
3494 	return (error);
3495 }
3496 
3497 /*
3498  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3499  */
3500 int
3501 prison_check(struct ucred *cred1, struct ucred *cred2)
3502 {
3503 
3504 	return ((cred1->cr_prison == cred2->cr_prison ||
3505 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3506 }
3507 
3508 /*
3509  * Return 1 if p2 is a child of p1, otherwise 0.
3510  */
3511 int
3512 prison_ischild(struct prison *pr1, struct prison *pr2)
3513 {
3514 
3515 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3516 		if (pr1 == pr2)
3517 			return (1);
3518 	return (0);
3519 }
3520 
3521 /*
3522  * Return 1 if the passed credential is in a jail, otherwise 0.
3523  */
3524 int
3525 jailed(struct ucred *cred)
3526 {
3527 
3528 	return (cred->cr_prison != &prison0);
3529 }
3530 
3531 /*
3532  * Return 1 if the passed credential is in a jail and that jail does not
3533  * have its own virtual network stack, otherwise 0.
3534  */
3535 int
3536 jailed_without_vnet(struct ucred *cred)
3537 {
3538 
3539 	if (!jailed(cred))
3540 		return (0);
3541 #ifdef VIMAGE
3542 	if (prison_owns_vnet(cred))
3543 		return (0);
3544 #endif
3545 
3546 	return (1);
3547 }
3548 
3549 /*
3550  * Return the correct hostname (domainname, et al) for the passed credential.
3551  */
3552 void
3553 getcredhostname(struct ucred *cred, char *buf, size_t size)
3554 {
3555 	struct prison *pr;
3556 
3557 	/*
3558 	 * A NULL credential can be used to shortcut to the physical
3559 	 * system's hostname.
3560 	 */
3561 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3562 	mtx_lock(&pr->pr_mtx);
3563 	strlcpy(buf, pr->pr_hostname, size);
3564 	mtx_unlock(&pr->pr_mtx);
3565 }
3566 
3567 void
3568 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3569 {
3570 
3571 	mtx_lock(&cred->cr_prison->pr_mtx);
3572 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3573 	mtx_unlock(&cred->cr_prison->pr_mtx);
3574 }
3575 
3576 void
3577 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3578 {
3579 
3580 	mtx_lock(&cred->cr_prison->pr_mtx);
3581 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3582 	mtx_unlock(&cred->cr_prison->pr_mtx);
3583 }
3584 
3585 void
3586 getcredhostid(struct ucred *cred, unsigned long *hostid)
3587 {
3588 
3589 	mtx_lock(&cred->cr_prison->pr_mtx);
3590 	*hostid = cred->cr_prison->pr_hostid;
3591 	mtx_unlock(&cred->cr_prison->pr_mtx);
3592 }
3593 
3594 #ifdef VIMAGE
3595 /*
3596  * Determine whether the prison represented by cred owns
3597  * its vnet rather than having it inherited.
3598  *
3599  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3600  */
3601 int
3602 prison_owns_vnet(struct ucred *cred)
3603 {
3604 
3605 	/*
3606 	 * vnets cannot be added/removed after jail creation,
3607 	 * so no need to lock here.
3608 	 */
3609 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3610 }
3611 #endif
3612 
3613 /*
3614  * Determine whether the subject represented by cred can "see"
3615  * status of a mount point.
3616  * Returns: 0 for permitted, ENOENT otherwise.
3617  * XXX: This function should be called cr_canseemount() and should be
3618  *      placed in kern_prot.c.
3619  */
3620 int
3621 prison_canseemount(struct ucred *cred, struct mount *mp)
3622 {
3623 	struct prison *pr;
3624 	struct statfs *sp;
3625 	size_t len;
3626 
3627 	pr = cred->cr_prison;
3628 	if (pr->pr_enforce_statfs == 0)
3629 		return (0);
3630 	if (pr->pr_root->v_mount == mp)
3631 		return (0);
3632 	if (pr->pr_enforce_statfs == 2)
3633 		return (ENOENT);
3634 	/*
3635 	 * If jail's chroot directory is set to "/" we should be able to see
3636 	 * all mount-points from inside a jail.
3637 	 * This is ugly check, but this is the only situation when jail's
3638 	 * directory ends with '/'.
3639 	 */
3640 	if (strcmp(pr->pr_path, "/") == 0)
3641 		return (0);
3642 	len = strlen(pr->pr_path);
3643 	sp = &mp->mnt_stat;
3644 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3645 		return (ENOENT);
3646 	/*
3647 	 * Be sure that we don't have situation where jail's root directory
3648 	 * is "/some/path" and mount point is "/some/pathpath".
3649 	 */
3650 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3651 		return (ENOENT);
3652 	return (0);
3653 }
3654 
3655 void
3656 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3657 {
3658 	char jpath[MAXPATHLEN];
3659 	struct prison *pr;
3660 	size_t len;
3661 
3662 	pr = cred->cr_prison;
3663 	if (pr->pr_enforce_statfs == 0)
3664 		return;
3665 	if (prison_canseemount(cred, mp) != 0) {
3666 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3667 		strlcpy(sp->f_mntonname, "[restricted]",
3668 		    sizeof(sp->f_mntonname));
3669 		return;
3670 	}
3671 	if (pr->pr_root->v_mount == mp) {
3672 		/*
3673 		 * Clear current buffer data, so we are sure nothing from
3674 		 * the valid path left there.
3675 		 */
3676 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3677 		*sp->f_mntonname = '/';
3678 		return;
3679 	}
3680 	/*
3681 	 * If jail's chroot directory is set to "/" we should be able to see
3682 	 * all mount-points from inside a jail.
3683 	 */
3684 	if (strcmp(pr->pr_path, "/") == 0)
3685 		return;
3686 	len = strlen(pr->pr_path);
3687 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3688 	/*
3689 	 * Clear current buffer data, so we are sure nothing from
3690 	 * the valid path left there.
3691 	 */
3692 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3693 	if (*jpath == '\0') {
3694 		/* Should never happen. */
3695 		*sp->f_mntonname = '/';
3696 	} else {
3697 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3698 	}
3699 }
3700 
3701 /*
3702  * Check with permission for a specific privilege is granted within jail.  We
3703  * have a specific list of accepted privileges; the rest are denied.
3704  */
3705 int
3706 prison_priv_check(struct ucred *cred, int priv)
3707 {
3708 
3709 	if (!jailed(cred))
3710 		return (0);
3711 
3712 #ifdef VIMAGE
3713 	/*
3714 	 * Privileges specific to prisons with a virtual network stack.
3715 	 * There might be a duplicate entry here in case the privilege
3716 	 * is only granted conditionally in the legacy jail case.
3717 	 */
3718 	switch (priv) {
3719 #ifdef notyet
3720 		/*
3721 		 * NFS-specific privileges.
3722 		 */
3723 	case PRIV_NFS_DAEMON:
3724 	case PRIV_NFS_LOCKD:
3725 #endif
3726 		/*
3727 		 * Network stack privileges.
3728 		 */
3729 	case PRIV_NET_BRIDGE:
3730 	case PRIV_NET_GRE:
3731 	case PRIV_NET_BPF:
3732 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3733 	case PRIV_NET_ROUTE:
3734 	case PRIV_NET_TAP:
3735 	case PRIV_NET_SETIFMTU:
3736 	case PRIV_NET_SETIFFLAGS:
3737 	case PRIV_NET_SETIFCAP:
3738 	case PRIV_NET_SETIFDESCR:
3739 	case PRIV_NET_SETIFNAME	:
3740 	case PRIV_NET_SETIFMETRIC:
3741 	case PRIV_NET_SETIFPHYS:
3742 	case PRIV_NET_SETIFMAC:
3743 	case PRIV_NET_ADDMULTI:
3744 	case PRIV_NET_DELMULTI:
3745 	case PRIV_NET_HWIOCTL:
3746 	case PRIV_NET_SETLLADDR:
3747 	case PRIV_NET_ADDIFGROUP:
3748 	case PRIV_NET_DELIFGROUP:
3749 	case PRIV_NET_IFCREATE:
3750 	case PRIV_NET_IFDESTROY:
3751 	case PRIV_NET_ADDIFADDR:
3752 	case PRIV_NET_DELIFADDR:
3753 	case PRIV_NET_LAGG:
3754 	case PRIV_NET_GIF:
3755 	case PRIV_NET_SETIFVNET:
3756 	case PRIV_NET_SETIFFIB:
3757 
3758 		/*
3759 		 * 802.11-related privileges.
3760 		 */
3761 	case PRIV_NET80211_GETKEY:
3762 #ifdef notyet
3763 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3764 #endif
3765 
3766 #ifdef notyet
3767 		/*
3768 		 * ATM privileges.
3769 		 */
3770 	case PRIV_NETATM_CFG:
3771 	case PRIV_NETATM_ADD:
3772 	case PRIV_NETATM_DEL:
3773 	case PRIV_NETATM_SET:
3774 
3775 		/*
3776 		 * Bluetooth privileges.
3777 		 */
3778 	case PRIV_NETBLUETOOTH_RAW:
3779 #endif
3780 
3781 		/*
3782 		 * Netgraph and netgraph module privileges.
3783 		 */
3784 	case PRIV_NETGRAPH_CONTROL:
3785 #ifdef notyet
3786 	case PRIV_NETGRAPH_TTY:
3787 #endif
3788 
3789 		/*
3790 		 * IPv4 and IPv6 privileges.
3791 		 */
3792 	case PRIV_NETINET_IPFW:
3793 	case PRIV_NETINET_DIVERT:
3794 	case PRIV_NETINET_PF:
3795 	case PRIV_NETINET_DUMMYNET:
3796 	case PRIV_NETINET_CARP:
3797 	case PRIV_NETINET_MROUTE:
3798 	case PRIV_NETINET_RAW:
3799 	case PRIV_NETINET_ADDRCTRL6:
3800 	case PRIV_NETINET_ND6:
3801 	case PRIV_NETINET_SCOPE6:
3802 	case PRIV_NETINET_ALIFETIME6:
3803 	case PRIV_NETINET_IPSEC:
3804 	case PRIV_NETINET_BINDANY:
3805 
3806 #ifdef notyet
3807 		/*
3808 		 * NCP privileges.
3809 		 */
3810 	case PRIV_NETNCP:
3811 
3812 		/*
3813 		 * SMB privileges.
3814 		 */
3815 	case PRIV_NETSMB:
3816 #endif
3817 
3818 	/*
3819 	 * No default: or deny here.
3820 	 * In case of no permit fall through to next switch().
3821 	 */
3822 		if (cred->cr_prison->pr_flags & PR_VNET)
3823 			return (0);
3824 	}
3825 #endif /* VIMAGE */
3826 
3827 	switch (priv) {
3828 
3829 		/*
3830 		 * Allow ktrace privileges for root in jail.
3831 		 */
3832 	case PRIV_KTRACE:
3833 
3834 #if 0
3835 		/*
3836 		 * Allow jailed processes to configure audit identity and
3837 		 * submit audit records (login, etc).  In the future we may
3838 		 * want to further refine the relationship between audit and
3839 		 * jail.
3840 		 */
3841 	case PRIV_AUDIT_GETAUDIT:
3842 	case PRIV_AUDIT_SETAUDIT:
3843 	case PRIV_AUDIT_SUBMIT:
3844 #endif
3845 
3846 		/*
3847 		 * Allow jailed processes to manipulate process UNIX
3848 		 * credentials in any way they see fit.
3849 		 */
3850 	case PRIV_CRED_SETUID:
3851 	case PRIV_CRED_SETEUID:
3852 	case PRIV_CRED_SETGID:
3853 	case PRIV_CRED_SETEGID:
3854 	case PRIV_CRED_SETGROUPS:
3855 	case PRIV_CRED_SETREUID:
3856 	case PRIV_CRED_SETREGID:
3857 	case PRIV_CRED_SETRESUID:
3858 	case PRIV_CRED_SETRESGID:
3859 
3860 		/*
3861 		 * Jail implements visibility constraints already, so allow
3862 		 * jailed root to override uid/gid-based constraints.
3863 		 */
3864 	case PRIV_SEEOTHERGIDS:
3865 	case PRIV_SEEOTHERUIDS:
3866 
3867 		/*
3868 		 * Jail implements inter-process debugging limits already, so
3869 		 * allow jailed root various debugging privileges.
3870 		 */
3871 	case PRIV_DEBUG_DIFFCRED:
3872 	case PRIV_DEBUG_SUGID:
3873 	case PRIV_DEBUG_UNPRIV:
3874 
3875 		/*
3876 		 * Allow jail to set various resource limits and login
3877 		 * properties, and for now, exceed process resource limits.
3878 		 */
3879 	case PRIV_PROC_LIMIT:
3880 	case PRIV_PROC_SETLOGIN:
3881 	case PRIV_PROC_SETRLIMIT:
3882 
3883 		/*
3884 		 * System V and POSIX IPC privileges are granted in jail.
3885 		 */
3886 	case PRIV_IPC_READ:
3887 	case PRIV_IPC_WRITE:
3888 	case PRIV_IPC_ADMIN:
3889 	case PRIV_IPC_MSGSIZE:
3890 	case PRIV_MQ_ADMIN:
3891 
3892 		/*
3893 		 * Jail operations within a jail work on child jails.
3894 		 */
3895 	case PRIV_JAIL_ATTACH:
3896 	case PRIV_JAIL_SET:
3897 	case PRIV_JAIL_REMOVE:
3898 
3899 		/*
3900 		 * Jail implements its own inter-process limits, so allow
3901 		 * root processes in jail to change scheduling on other
3902 		 * processes in the same jail.  Likewise for signalling.
3903 		 */
3904 	case PRIV_SCHED_DIFFCRED:
3905 	case PRIV_SCHED_CPUSET:
3906 	case PRIV_SIGNAL_DIFFCRED:
3907 	case PRIV_SIGNAL_SUGID:
3908 
3909 		/*
3910 		 * Allow jailed processes to write to sysctls marked as jail
3911 		 * writable.
3912 		 */
3913 	case PRIV_SYSCTL_WRITEJAIL:
3914 
3915 		/*
3916 		 * Allow root in jail to manage a variety of quota
3917 		 * properties.  These should likely be conditional on a
3918 		 * configuration option.
3919 		 */
3920 	case PRIV_VFS_GETQUOTA:
3921 	case PRIV_VFS_SETQUOTA:
3922 
3923 		/*
3924 		 * Since Jail relies on chroot() to implement file system
3925 		 * protections, grant many VFS privileges to root in jail.
3926 		 * Be careful to exclude mount-related and NFS-related
3927 		 * privileges.
3928 		 */
3929 	case PRIV_VFS_READ:
3930 	case PRIV_VFS_WRITE:
3931 	case PRIV_VFS_ADMIN:
3932 	case PRIV_VFS_EXEC:
3933 	case PRIV_VFS_LOOKUP:
3934 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3935 	case PRIV_VFS_CHFLAGS_DEV:
3936 	case PRIV_VFS_CHOWN:
3937 	case PRIV_VFS_CHROOT:
3938 	case PRIV_VFS_RETAINSUGID:
3939 	case PRIV_VFS_FCHROOT:
3940 	case PRIV_VFS_LINK:
3941 	case PRIV_VFS_SETGID:
3942 	case PRIV_VFS_STAT:
3943 	case PRIV_VFS_STICKYFILE:
3944 
3945 		/*
3946 		 * As in the non-jail case, non-root users are expected to be
3947 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3948 		 * exists in the jail and they have permission to access it).
3949 		 */
3950 	case PRIV_KMEM_READ:
3951 		return (0);
3952 
3953 		/*
3954 		 * Depending on the global setting, allow privilege of
3955 		 * setting system flags.
3956 		 */
3957 	case PRIV_VFS_SYSFLAGS:
3958 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3959 			return (0);
3960 		else
3961 			return (EPERM);
3962 
3963 		/*
3964 		 * Depending on the global setting, allow privilege of
3965 		 * mounting/unmounting file systems.
3966 		 */
3967 	case PRIV_VFS_MOUNT:
3968 	case PRIV_VFS_UNMOUNT:
3969 	case PRIV_VFS_MOUNT_NONUSER:
3970 	case PRIV_VFS_MOUNT_OWNER:
3971 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3972 		    cred->cr_prison->pr_enforce_statfs < 2)
3973 			return (0);
3974 		else
3975 			return (EPERM);
3976 
3977 		/*
3978 		 * Allow jailed root to bind reserved ports and reuse in-use
3979 		 * ports.
3980 		 */
3981 	case PRIV_NETINET_RESERVEDPORT:
3982 	case PRIV_NETINET_REUSEPORT:
3983 		return (0);
3984 
3985 		/*
3986 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3987 		 */
3988 	case PRIV_NETINET_SETHDROPTS:
3989 		return (0);
3990 
3991 		/*
3992 		 * Conditionally allow creating raw sockets in jail.
3993 		 */
3994 	case PRIV_NETINET_RAW:
3995 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3996 			return (0);
3997 		else
3998 			return (EPERM);
3999 
4000 		/*
4001 		 * Since jail implements its own visibility limits on netstat
4002 		 * sysctls, allow getcred.  This allows identd to work in
4003 		 * jail.
4004 		 */
4005 	case PRIV_NETINET_GETCRED:
4006 		return (0);
4007 
4008 		/*
4009 		 * Allow jailed root to set loginclass.
4010 		 */
4011 	case PRIV_PROC_SETLOGINCLASS:
4012 		return (0);
4013 
4014 	default:
4015 		/*
4016 		 * In all remaining cases, deny the privilege request.  This
4017 		 * includes almost all network privileges, many system
4018 		 * configuration privileges.
4019 		 */
4020 		return (EPERM);
4021 	}
4022 }
4023 
4024 /*
4025  * Return the part of pr2's name that is relative to pr1, or the whole name
4026  * if it does not directly follow.
4027  */
4028 
4029 char *
4030 prison_name(struct prison *pr1, struct prison *pr2)
4031 {
4032 	char *name;
4033 
4034 	/* Jails see themselves as "0" (if they see themselves at all). */
4035 	if (pr1 == pr2)
4036 		return "0";
4037 	name = pr2->pr_name;
4038 	if (prison_ischild(pr1, pr2)) {
4039 		/*
4040 		 * pr1 isn't locked (and allprison_lock may not be either)
4041 		 * so its length can't be counted on.  But the number of dots
4042 		 * can be counted on - and counted.
4043 		 */
4044 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4045 			name = strchr(name, '.') + 1;
4046 	}
4047 	return (name);
4048 }
4049 
4050 /*
4051  * Return the part of pr2's path that is relative to pr1, or the whole path
4052  * if it does not directly follow.
4053  */
4054 static char *
4055 prison_path(struct prison *pr1, struct prison *pr2)
4056 {
4057 	char *path1, *path2;
4058 	int len1;
4059 
4060 	path1 = pr1->pr_path;
4061 	path2 = pr2->pr_path;
4062 	if (!strcmp(path1, "/"))
4063 		return (path2);
4064 	len1 = strlen(path1);
4065 	if (strncmp(path1, path2, len1))
4066 		return (path2);
4067 	if (path2[len1] == '\0')
4068 		return "/";
4069 	if (path2[len1] == '/')
4070 		return (path2 + len1);
4071 	return (path2);
4072 }
4073 
4074 
4075 /*
4076  * Jail-related sysctls.
4077  */
4078 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4079     "Jails");
4080 
4081 static int
4082 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4083 {
4084 	struct xprison *xp;
4085 	struct prison *pr, *cpr;
4086 #ifdef INET
4087 	struct in_addr *ip4 = NULL;
4088 	int ip4s = 0;
4089 #endif
4090 #ifdef INET6
4091 	struct in6_addr *ip6 = NULL;
4092 	int ip6s = 0;
4093 #endif
4094 	int descend, error;
4095 
4096 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4097 	pr = req->td->td_ucred->cr_prison;
4098 	error = 0;
4099 	sx_slock(&allprison_lock);
4100 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4101 #if defined(INET) || defined(INET6)
4102  again:
4103 #endif
4104 		mtx_lock(&cpr->pr_mtx);
4105 #ifdef INET
4106 		if (cpr->pr_ip4s > 0) {
4107 			if (ip4s < cpr->pr_ip4s) {
4108 				ip4s = cpr->pr_ip4s;
4109 				mtx_unlock(&cpr->pr_mtx);
4110 				ip4 = realloc(ip4, ip4s *
4111 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4112 				goto again;
4113 			}
4114 			bcopy(cpr->pr_ip4, ip4,
4115 			    cpr->pr_ip4s * sizeof(struct in_addr));
4116 		}
4117 #endif
4118 #ifdef INET6
4119 		if (cpr->pr_ip6s > 0) {
4120 			if (ip6s < cpr->pr_ip6s) {
4121 				ip6s = cpr->pr_ip6s;
4122 				mtx_unlock(&cpr->pr_mtx);
4123 				ip6 = realloc(ip6, ip6s *
4124 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4125 				goto again;
4126 			}
4127 			bcopy(cpr->pr_ip6, ip6,
4128 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4129 		}
4130 #endif
4131 		if (cpr->pr_ref == 0) {
4132 			mtx_unlock(&cpr->pr_mtx);
4133 			continue;
4134 		}
4135 		bzero(xp, sizeof(*xp));
4136 		xp->pr_version = XPRISON_VERSION;
4137 		xp->pr_id = cpr->pr_id;
4138 		xp->pr_state = cpr->pr_uref > 0
4139 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4140 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4141 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4142 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4143 #ifdef INET
4144 		xp->pr_ip4s = cpr->pr_ip4s;
4145 #endif
4146 #ifdef INET6
4147 		xp->pr_ip6s = cpr->pr_ip6s;
4148 #endif
4149 		mtx_unlock(&cpr->pr_mtx);
4150 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4151 		if (error)
4152 			break;
4153 #ifdef INET
4154 		if (xp->pr_ip4s > 0) {
4155 			error = SYSCTL_OUT(req, ip4,
4156 			    xp->pr_ip4s * sizeof(struct in_addr));
4157 			if (error)
4158 				break;
4159 		}
4160 #endif
4161 #ifdef INET6
4162 		if (xp->pr_ip6s > 0) {
4163 			error = SYSCTL_OUT(req, ip6,
4164 			    xp->pr_ip6s * sizeof(struct in6_addr));
4165 			if (error)
4166 				break;
4167 		}
4168 #endif
4169 	}
4170 	sx_sunlock(&allprison_lock);
4171 	free(xp, M_TEMP);
4172 #ifdef INET
4173 	free(ip4, M_TEMP);
4174 #endif
4175 #ifdef INET6
4176 	free(ip6, M_TEMP);
4177 #endif
4178 	return (error);
4179 }
4180 
4181 SYSCTL_OID(_security_jail, OID_AUTO, list,
4182     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4183     sysctl_jail_list, "S", "List of active jails");
4184 
4185 static int
4186 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4187 {
4188 	int error, injail;
4189 
4190 	injail = jailed(req->td->td_ucred);
4191 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4192 
4193 	return (error);
4194 }
4195 
4196 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4197     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4198     sysctl_jail_jailed, "I", "Process in jail?");
4199 
4200 static int
4201 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4202 {
4203 	int error, havevnet;
4204 #ifdef VIMAGE
4205 	struct ucred *cred = req->td->td_ucred;
4206 
4207 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4208 #else
4209 	havevnet = 0;
4210 #endif
4211 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4212 
4213 	return (error);
4214 }
4215 
4216 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4217     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4218     sysctl_jail_vnet, "I", "Jail owns VNET?");
4219 
4220 #if defined(INET) || defined(INET6)
4221 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4222     &jail_max_af_ips, 0,
4223     "Number of IP addresses a jail may have at most per address family");
4224 #endif
4225 
4226 /*
4227  * Default parameters for jail(2) compatability.  For historical reasons,
4228  * the sysctl names have varying similarity to the parameter names.  Prisons
4229  * just see their own parameters, and can't change them.
4230  */
4231 static int
4232 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4233 {
4234 	struct prison *pr;
4235 	int allow, error, i;
4236 
4237 	pr = req->td->td_ucred->cr_prison;
4238 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4239 
4240 	/* Get the current flag value, and convert it to a boolean. */
4241 	i = (allow & arg2) ? 1 : 0;
4242 	if (arg1 != NULL)
4243 		i = !i;
4244 	error = sysctl_handle_int(oidp, &i, 0, req);
4245 	if (error || !req->newptr)
4246 		return (error);
4247 	i = i ? arg2 : 0;
4248 	if (arg1 != NULL)
4249 		i ^= arg2;
4250 	/*
4251 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4252 	 * for writing.
4253 	 */
4254 	mtx_lock(&prison0.pr_mtx);
4255 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4256 	mtx_unlock(&prison0.pr_mtx);
4257 	return (0);
4258 }
4259 
4260 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4261     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4262     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4263     "Processes in jail can set their hostnames");
4264 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4265     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4266     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4267     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4268 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4269     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4270     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4271     "Processes in jail can use System V IPC primitives");
4272 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4273     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4274     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4275     "Prison root can create raw sockets");
4276 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4277     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4278     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4279     "Processes in jail can alter system file flags");
4280 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4281     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4282     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4283     "Processes in jail can mount/unmount jail-friendly file systems");
4284 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4285     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4286     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4287     "Processes in jail can mount the devfs file system");
4288 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4289     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4290     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4291     "Processes in jail can mount the fdescfs file system");
4292 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4293     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4294     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4295     "Processes in jail can mount the nullfs file system");
4296 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4297     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4298     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4299     "Processes in jail can mount the procfs file system");
4300 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
4301     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4302     NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4303     "Processes in jail can mount the linprocfs file system");
4304 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4305     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4306     NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4307     "Processes in jail can mount the linsysfs file system");
4308 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4309     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4310     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4311     "Processes in jail can mount the tmpfs file system");
4312 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4313     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4314     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4315     "Processes in jail can mount the zfs file system");
4316 
4317 static int
4318 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4319 {
4320 	struct prison *pr;
4321 	int level, error;
4322 
4323 	pr = req->td->td_ucred->cr_prison;
4324 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4325 	error = sysctl_handle_int(oidp, &level, 0, req);
4326 	if (error || !req->newptr)
4327 		return (error);
4328 	*(int *)arg1 = level;
4329 	return (0);
4330 }
4331 
4332 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4333     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4334     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4335     sysctl_jail_default_level, "I",
4336     "Processes in jail cannot see all mounted file systems");
4337 
4338 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4339     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4340     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4341     sysctl_jail_default_level, "I",
4342     "Ruleset for the devfs filesystem in jail");
4343 
4344 /*
4345  * Nodes to describe jail parameters.  Maximum length of string parameters
4346  * is returned in the string itself, and the other parameters exist merely
4347  * to make themselves and their types known.
4348  */
4349 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4350     "Jail parameters");
4351 
4352 int
4353 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4354 {
4355 	int i;
4356 	long l;
4357 	size_t s;
4358 	char numbuf[12];
4359 
4360 	switch (oidp->oid_kind & CTLTYPE)
4361 	{
4362 	case CTLTYPE_LONG:
4363 	case CTLTYPE_ULONG:
4364 		l = 0;
4365 #ifdef SCTL_MASK32
4366 		if (!(req->flags & SCTL_MASK32))
4367 #endif
4368 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4369 	case CTLTYPE_INT:
4370 	case CTLTYPE_UINT:
4371 		i = 0;
4372 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4373 	case CTLTYPE_STRING:
4374 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4375 		return
4376 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4377 	case CTLTYPE_STRUCT:
4378 		s = (size_t)arg2;
4379 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4380 	}
4381 	return (0);
4382 }
4383 
4384 /*
4385  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4386  * jail creation time but cannot be changed in an existing jail.
4387  */
4388 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4389 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4390 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4391 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4392 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4393     "I", "Jail secure level");
4394 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4395     "Jail value for kern.osreldate and uname -K");
4396 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4397     "Jail value for kern.osrelease and uname -r");
4398 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4399     "I", "Jail cannot see all mounted file systems");
4400 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4401     "I", "Ruleset for in-jail devfs mounts");
4402 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4403     "B", "Jail persistence");
4404 #ifdef VIMAGE
4405 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4406     "E,jailsys", "Virtual network stack");
4407 #endif
4408 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4409     "B", "Jail is in the process of shutting down");
4410 
4411 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4412 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4413     "I", "Current number of child jails");
4414 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4415     "I", "Maximum number of child jails");
4416 
4417 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4418 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4419     "Jail hostname");
4420 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4421     "Jail NIS domainname");
4422 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4423     "Jail host UUID");
4424 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4425     "LU", "Jail host ID");
4426 
4427 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4428 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4429 
4430 #ifdef INET
4431 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4432     "Jail IPv4 address virtualization");
4433 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4434     "S,in_addr,a", "Jail IPv4 addresses");
4435 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4436     "B", "Do (not) use IPv4 source address selection rather than the "
4437     "primary jail IPv4 address.");
4438 #endif
4439 #ifdef INET6
4440 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4441     "Jail IPv6 address virtualization");
4442 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4443     "S,in6_addr,a", "Jail IPv6 addresses");
4444 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4445     "B", "Do (not) use IPv6 source address selection rather than the "
4446     "primary jail IPv6 address.");
4447 #endif
4448 
4449 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4450 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4451     "B", "Jail may set hostname");
4452 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4453     "B", "Jail may use SYSV IPC");
4454 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4455     "B", "Jail may create raw sockets");
4456 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4457     "B", "Jail may alter system file flags");
4458 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4459     "B", "Jail may set file quotas");
4460 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4461     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4462 
4463 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4464 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4465     "B", "Jail may mount/unmount jail-friendly file systems in general");
4466 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4467     "B", "Jail may mount the devfs file system");
4468 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4469     "B", "Jail may mount the fdescfs file system");
4470 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4471     "B", "Jail may mount the nullfs file system");
4472 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4473     "B", "Jail may mount the procfs file system");
4474 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4475     "B", "Jail may mount the linprocfs file system");
4476 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4477     "B", "Jail may mount the linsysfs file system");
4478 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4479     "B", "Jail may mount the tmpfs file system");
4480 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4481     "B", "Jail may mount the zfs file system");
4482 
4483 #ifdef RACCT
4484 void
4485 prison_racct_foreach(void (*callback)(struct racct *racct,
4486     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4487     void *arg2, void *arg3)
4488 {
4489 	struct prison_racct *prr;
4490 
4491 	ASSERT_RACCT_ENABLED();
4492 
4493 	sx_slock(&allprison_lock);
4494 	if (pre != NULL)
4495 		(pre)();
4496 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4497 		(callback)(prr->prr_racct, arg2, arg3);
4498 	if (post != NULL)
4499 		(post)();
4500 	sx_sunlock(&allprison_lock);
4501 }
4502 
4503 static struct prison_racct *
4504 prison_racct_find_locked(const char *name)
4505 {
4506 	struct prison_racct *prr;
4507 
4508 	ASSERT_RACCT_ENABLED();
4509 	sx_assert(&allprison_lock, SA_XLOCKED);
4510 
4511 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4512 		return (NULL);
4513 
4514 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4515 		if (strcmp(name, prr->prr_name) != 0)
4516 			continue;
4517 
4518 		/* Found prison_racct with a matching name? */
4519 		prison_racct_hold(prr);
4520 		return (prr);
4521 	}
4522 
4523 	/* Add new prison_racct. */
4524 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4525 	racct_create(&prr->prr_racct);
4526 
4527 	strcpy(prr->prr_name, name);
4528 	refcount_init(&prr->prr_refcount, 1);
4529 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4530 
4531 	return (prr);
4532 }
4533 
4534 struct prison_racct *
4535 prison_racct_find(const char *name)
4536 {
4537 	struct prison_racct *prr;
4538 
4539 	ASSERT_RACCT_ENABLED();
4540 
4541 	sx_xlock(&allprison_lock);
4542 	prr = prison_racct_find_locked(name);
4543 	sx_xunlock(&allprison_lock);
4544 	return (prr);
4545 }
4546 
4547 void
4548 prison_racct_hold(struct prison_racct *prr)
4549 {
4550 
4551 	ASSERT_RACCT_ENABLED();
4552 
4553 	refcount_acquire(&prr->prr_refcount);
4554 }
4555 
4556 static void
4557 prison_racct_free_locked(struct prison_racct *prr)
4558 {
4559 
4560 	ASSERT_RACCT_ENABLED();
4561 	sx_assert(&allprison_lock, SA_XLOCKED);
4562 
4563 	if (refcount_release(&prr->prr_refcount)) {
4564 		racct_destroy(&prr->prr_racct);
4565 		LIST_REMOVE(prr, prr_next);
4566 		free(prr, M_PRISON_RACCT);
4567 	}
4568 }
4569 
4570 void
4571 prison_racct_free(struct prison_racct *prr)
4572 {
4573 	int old;
4574 
4575 	ASSERT_RACCT_ENABLED();
4576 	sx_assert(&allprison_lock, SA_UNLOCKED);
4577 
4578 	old = prr->prr_refcount;
4579 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4580 		return;
4581 
4582 	sx_xlock(&allprison_lock);
4583 	prison_racct_free_locked(prr);
4584 	sx_xunlock(&allprison_lock);
4585 }
4586 
4587 static void
4588 prison_racct_attach(struct prison *pr)
4589 {
4590 	struct prison_racct *prr;
4591 
4592 	ASSERT_RACCT_ENABLED();
4593 	sx_assert(&allprison_lock, SA_XLOCKED);
4594 
4595 	prr = prison_racct_find_locked(pr->pr_name);
4596 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4597 
4598 	pr->pr_prison_racct = prr;
4599 }
4600 
4601 /*
4602  * Handle jail renaming.  From the racct point of view, renaming means
4603  * moving from one prison_racct to another.
4604  */
4605 static void
4606 prison_racct_modify(struct prison *pr)
4607 {
4608 	struct proc *p;
4609 	struct ucred *cred;
4610 	struct prison_racct *oldprr;
4611 
4612 	ASSERT_RACCT_ENABLED();
4613 
4614 	sx_slock(&allproc_lock);
4615 	sx_xlock(&allprison_lock);
4616 
4617 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4618 		sx_xunlock(&allprison_lock);
4619 		sx_sunlock(&allproc_lock);
4620 		return;
4621 	}
4622 
4623 	oldprr = pr->pr_prison_racct;
4624 	pr->pr_prison_racct = NULL;
4625 
4626 	prison_racct_attach(pr);
4627 
4628 	/*
4629 	 * Move resource utilisation records.
4630 	 */
4631 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4632 
4633 	/*
4634 	 * Force rctl to reattach rules to processes.
4635 	 */
4636 	FOREACH_PROC_IN_SYSTEM(p) {
4637 		PROC_LOCK(p);
4638 		cred = crhold(p->p_ucred);
4639 		PROC_UNLOCK(p);
4640 		racct_proc_ucred_changed(p, cred, cred);
4641 		crfree(cred);
4642 	}
4643 
4644 	sx_sunlock(&allproc_lock);
4645 	prison_racct_free_locked(oldprr);
4646 	sx_xunlock(&allprison_lock);
4647 }
4648 
4649 static void
4650 prison_racct_detach(struct prison *pr)
4651 {
4652 
4653 	ASSERT_RACCT_ENABLED();
4654 	sx_assert(&allprison_lock, SA_UNLOCKED);
4655 
4656 	if (pr->pr_prison_racct == NULL)
4657 		return;
4658 	prison_racct_free(pr->pr_prison_racct);
4659 	pr->pr_prison_racct = NULL;
4660 }
4661 #endif /* RACCT */
4662 
4663 #ifdef DDB
4664 
4665 static void
4666 db_show_prison(struct prison *pr)
4667 {
4668 	int fi;
4669 #if defined(INET) || defined(INET6)
4670 	int ii;
4671 #endif
4672 	unsigned jsf;
4673 #ifdef INET6
4674 	char ip6buf[INET6_ADDRSTRLEN];
4675 #endif
4676 
4677 	db_printf("prison %p:\n", pr);
4678 	db_printf(" jid             = %d\n", pr->pr_id);
4679 	db_printf(" name            = %s\n", pr->pr_name);
4680 	db_printf(" parent          = %p\n", pr->pr_parent);
4681 	db_printf(" ref             = %d\n", pr->pr_ref);
4682 	db_printf(" uref            = %d\n", pr->pr_uref);
4683 	db_printf(" path            = %s\n", pr->pr_path);
4684 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4685 	    ? pr->pr_cpuset->cs_id : -1);
4686 #ifdef VIMAGE
4687 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4688 #endif
4689 	db_printf(" root            = %p\n", pr->pr_root);
4690 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4691 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4692 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4693 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4694 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4695 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4696 	db_printf(" flags           = 0x%x", pr->pr_flags);
4697 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4698 	    fi++)
4699 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4700 			db_printf(" %s", pr_flag_names[fi]);
4701 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4702 	    fi++) {
4703 		jsf = pr->pr_flags &
4704 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4705 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4706 		    pr_flag_jailsys[fi].disable &&
4707 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4708 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4709 		    : "inherit");
4710 	}
4711 	db_printf(" allow           = 0x%x", pr->pr_allow);
4712 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4713 	    fi++)
4714 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4715 			db_printf(" %s", pr_allow_names[fi]);
4716 	db_printf("\n");
4717 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4718 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4719 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4720 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4721 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4722 #ifdef INET
4723 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4724 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4725 		db_printf(" %s %s\n",
4726 		    ii == 0 ? "ip4.addr        =" : "                 ",
4727 		    inet_ntoa(pr->pr_ip4[ii]));
4728 #endif
4729 #ifdef INET6
4730 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4731 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4732 		db_printf(" %s %s\n",
4733 		    ii == 0 ? "ip6.addr        =" : "                 ",
4734 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4735 #endif
4736 }
4737 
4738 DB_SHOW_COMMAND(prison, db_show_prison_command)
4739 {
4740 	struct prison *pr;
4741 
4742 	if (!have_addr) {
4743 		/*
4744 		 * Show all prisons in the list, and prison0 which is not
4745 		 * listed.
4746 		 */
4747 		db_show_prison(&prison0);
4748 		if (!db_pager_quit) {
4749 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4750 				db_show_prison(pr);
4751 				if (db_pager_quit)
4752 					break;
4753 			}
4754 		}
4755 		return;
4756 	}
4757 
4758 	if (addr == 0)
4759 		pr = &prison0;
4760 	else {
4761 		/* Look for a prison with the ID and with references. */
4762 		TAILQ_FOREACH(pr, &allprison, pr_list)
4763 			if (pr->pr_id == addr && pr->pr_ref > 0)
4764 				break;
4765 		if (pr == NULL)
4766 			/* Look again, without requiring a reference. */
4767 			TAILQ_FOREACH(pr, &allprison, pr_list)
4768 				if (pr->pr_id == addr)
4769 					break;
4770 		if (pr == NULL)
4771 			/* Assume address points to a valid prison. */
4772 			pr = (struct prison *)addr;
4773 	}
4774 	db_show_prison(pr);
4775 }
4776 
4777 #endif /* DDB */
4778