xref: /freebsd/sys/kern/kern_jail.c (revision 545ddfbe7d4fe8adfb862903b24eac1d5896c1ef)
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/if_var.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 
70 #ifdef DDB
71 #include <ddb/ddb.h>
72 #ifdef INET6
73 #include <netinet6/in6_var.h>
74 #endif /* INET6 */
75 #endif /* DDB */
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
80 
81 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
82 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
83 
84 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
85 #ifdef INET
86 #ifdef INET6
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
88 #else
89 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
90 #endif
91 #else /* !INET */
92 #ifdef INET6
93 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
94 #else
95 #define	_PR_IP_SADDRSEL	0
96 #endif
97 #endif
98 
99 /* prison0 describes what is "real" about the system. */
100 struct prison prison0 = {
101 	.pr_id		= 0,
102 	.pr_name	= "0",
103 	.pr_ref		= 1,
104 	.pr_uref	= 1,
105 	.pr_path	= "/",
106 	.pr_securelevel	= -1,
107 	.pr_devfs_rsnum = 0,
108 	.pr_childmax	= JAIL_MAX,
109 	.pr_hostuuid	= DEFAULT_HOSTUUID,
110 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
111 #ifdef VIMAGE
112 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
113 #else
114 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
115 #endif
116 	.pr_allow	= PR_ALLOW_ALL,
117 };
118 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
119 
120 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
121 struct	sx allprison_lock;
122 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
123 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
124 LIST_HEAD(, prison_racct) allprison_racct;
125 int	lastprid = 0;
126 
127 static int do_jail_attach(struct thread *td, struct prison *pr);
128 static void prison_complete(void *context, int pending);
129 static void prison_deref(struct prison *pr, int flags);
130 static char *prison_path(struct prison *pr1, struct prison *pr2);
131 static void prison_remove_one(struct prison *pr);
132 #ifdef RACCT
133 static void prison_racct_attach(struct prison *pr);
134 static void prison_racct_modify(struct prison *pr);
135 static void prison_racct_detach(struct prison *pr);
136 #endif
137 #ifdef INET
138 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
139 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
140 #endif
141 #ifdef INET6
142 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
143 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
144 #endif
145 
146 /* Flags for prison_deref */
147 #define	PD_DEREF	0x01
148 #define	PD_DEUREF	0x02
149 #define	PD_LOCKED	0x04
150 #define	PD_LIST_SLOCKED	0x08
151 #define	PD_LIST_XLOCKED	0x10
152 
153 /*
154  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
155  * as we cannot figure out the size of a sparse array, or an array without a
156  * terminating entry.
157  */
158 static char *pr_flag_names[] = {
159 	[0] = "persist",
160 #ifdef INET
161 	[7] = "ip4.saddrsel",
162 #endif
163 #ifdef INET6
164 	[8] = "ip6.saddrsel",
165 #endif
166 };
167 const size_t pr_flag_names_size = sizeof(pr_flag_names);
168 
169 static char *pr_flag_nonames[] = {
170 	[0] = "nopersist",
171 #ifdef INET
172 	[7] = "ip4.nosaddrsel",
173 #endif
174 #ifdef INET6
175 	[8] = "ip6.nosaddrsel",
176 #endif
177 };
178 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
179 
180 struct jailsys_flags {
181 	const char	*name;
182 	unsigned	 disable;
183 	unsigned	 new;
184 } pr_flag_jailsys[] = {
185 	{ "host", 0, PR_HOST },
186 #ifdef VIMAGE
187 	{ "vnet", 0, PR_VNET },
188 #endif
189 #ifdef INET
190 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
191 #endif
192 #ifdef INET6
193 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
194 #endif
195 };
196 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
197 
198 static char *pr_allow_names[] = {
199 	"allow.set_hostname",
200 	"allow.sysvipc",
201 	"allow.raw_sockets",
202 	"allow.chflags",
203 	"allow.mount",
204 	"allow.quotas",
205 	"allow.socket_af",
206 	"allow.mount.devfs",
207 	"allow.mount.nullfs",
208 	"allow.mount.zfs",
209 	"allow.mount.procfs",
210 	"allow.mount.tmpfs",
211 };
212 const size_t pr_allow_names_size = sizeof(pr_allow_names);
213 
214 static char *pr_allow_nonames[] = {
215 	"allow.noset_hostname",
216 	"allow.nosysvipc",
217 	"allow.noraw_sockets",
218 	"allow.nochflags",
219 	"allow.nomount",
220 	"allow.noquotas",
221 	"allow.nosocket_af",
222 	"allow.mount.nodevfs",
223 	"allow.mount.nonullfs",
224 	"allow.mount.nozfs",
225 	"allow.mount.noprocfs",
226 	"allow.mount.notmpfs",
227 };
228 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
229 
230 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
231 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
232 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
233 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
234 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
235 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
236 #if defined(INET) || defined(INET6)
237 static unsigned jail_max_af_ips = 255;
238 #endif
239 
240 #ifdef INET
241 static int
242 qcmp_v4(const void *ip1, const void *ip2)
243 {
244 	in_addr_t iaa, iab;
245 
246 	/*
247 	 * We need to compare in HBO here to get the list sorted as expected
248 	 * by the result of the code.  Sorting NBO addresses gives you
249 	 * interesting results.  If you do not understand, do not try.
250 	 */
251 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
252 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
253 
254 	/*
255 	 * Do not simply return the difference of the two numbers, the int is
256 	 * not wide enough.
257 	 */
258 	if (iaa > iab)
259 		return (1);
260 	else if (iaa < iab)
261 		return (-1);
262 	else
263 		return (0);
264 }
265 #endif
266 
267 #ifdef INET6
268 static int
269 qcmp_v6(const void *ip1, const void *ip2)
270 {
271 	const struct in6_addr *ia6a, *ia6b;
272 	int i, rc;
273 
274 	ia6a = (const struct in6_addr *)ip1;
275 	ia6b = (const struct in6_addr *)ip2;
276 
277 	rc = 0;
278 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
279 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
280 			rc = 1;
281 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
282 			rc = -1;
283 	}
284 	return (rc);
285 }
286 #endif
287 
288 /*
289  * struct jail_args {
290  *	struct jail *jail;
291  * };
292  */
293 int
294 sys_jail(struct thread *td, struct jail_args *uap)
295 {
296 	uint32_t version;
297 	int error;
298 	struct jail j;
299 
300 	error = copyin(uap->jail, &version, sizeof(uint32_t));
301 	if (error)
302 		return (error);
303 
304 	switch (version) {
305 	case 0:
306 	{
307 		struct jail_v0 j0;
308 
309 		/* FreeBSD single IPv4 jails. */
310 		bzero(&j, sizeof(struct jail));
311 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
312 		if (error)
313 			return (error);
314 		j.version = j0.version;
315 		j.path = j0.path;
316 		j.hostname = j0.hostname;
317 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
318 		break;
319 	}
320 
321 	case 1:
322 		/*
323 		 * Version 1 was used by multi-IPv4 jail implementations
324 		 * that never made it into the official kernel.
325 		 */
326 		return (EINVAL);
327 
328 	case 2:	/* JAIL_API_VERSION */
329 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
330 		error = copyin(uap->jail, &j, sizeof(struct jail));
331 		if (error)
332 			return (error);
333 		break;
334 
335 	default:
336 		/* Sci-Fi jails are not supported, sorry. */
337 		return (EINVAL);
338 	}
339 	return (kern_jail(td, &j));
340 }
341 
342 int
343 kern_jail(struct thread *td, struct jail *j)
344 {
345 	struct iovec optiov[2 * (4
346 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
347 #ifdef INET
348 			    + 1
349 #endif
350 #ifdef INET6
351 			    + 1
352 #endif
353 			    )];
354 	struct uio opt;
355 	char *u_path, *u_hostname, *u_name;
356 #ifdef INET
357 	uint32_t ip4s;
358 	struct in_addr *u_ip4;
359 #endif
360 #ifdef INET6
361 	struct in6_addr *u_ip6;
362 #endif
363 	size_t tmplen;
364 	int error, enforce_statfs, fi;
365 
366 	bzero(&optiov, sizeof(optiov));
367 	opt.uio_iov = optiov;
368 	opt.uio_iovcnt = 0;
369 	opt.uio_offset = -1;
370 	opt.uio_resid = -1;
371 	opt.uio_segflg = UIO_SYSSPACE;
372 	opt.uio_rw = UIO_READ;
373 	opt.uio_td = td;
374 
375 	/* Set permissions for top-level jails from sysctls. */
376 	if (!jailed(td->td_ucred)) {
377 		for (fi = 0; fi < sizeof(pr_allow_names) /
378 		     sizeof(pr_allow_names[0]); fi++) {
379 			optiov[opt.uio_iovcnt].iov_base =
380 			    (jail_default_allow & (1 << fi))
381 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
382 			optiov[opt.uio_iovcnt].iov_len =
383 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
384 			opt.uio_iovcnt += 2;
385 		}
386 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
387 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
388 		opt.uio_iovcnt++;
389 		enforce_statfs = jail_default_enforce_statfs;
390 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
391 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
392 		opt.uio_iovcnt++;
393 	}
394 
395 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
396 #ifdef INET
397 	ip4s = (j->version == 0) ? 1 : j->ip4s;
398 	if (ip4s > jail_max_af_ips)
399 		return (EINVAL);
400 	tmplen += ip4s * sizeof(struct in_addr);
401 #else
402 	if (j->ip4s > 0)
403 		return (EINVAL);
404 #endif
405 #ifdef INET6
406 	if (j->ip6s > jail_max_af_ips)
407 		return (EINVAL);
408 	tmplen += j->ip6s * sizeof(struct in6_addr);
409 #else
410 	if (j->ip6s > 0)
411 		return (EINVAL);
412 #endif
413 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
414 	u_hostname = u_path + MAXPATHLEN;
415 	u_name = u_hostname + MAXHOSTNAMELEN;
416 #ifdef INET
417 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
418 #endif
419 #ifdef INET6
420 #ifdef INET
421 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
422 #else
423 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
424 #endif
425 #endif
426 	optiov[opt.uio_iovcnt].iov_base = "path";
427 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
428 	opt.uio_iovcnt++;
429 	optiov[opt.uio_iovcnt].iov_base = u_path;
430 	error = copyinstr(j->path, u_path, MAXPATHLEN,
431 	    &optiov[opt.uio_iovcnt].iov_len);
432 	if (error) {
433 		free(u_path, M_TEMP);
434 		return (error);
435 	}
436 	opt.uio_iovcnt++;
437 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
438 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
439 	opt.uio_iovcnt++;
440 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
441 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
442 	    &optiov[opt.uio_iovcnt].iov_len);
443 	if (error) {
444 		free(u_path, M_TEMP);
445 		return (error);
446 	}
447 	opt.uio_iovcnt++;
448 	if (j->jailname != NULL) {
449 		optiov[opt.uio_iovcnt].iov_base = "name";
450 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
451 		opt.uio_iovcnt++;
452 		optiov[opt.uio_iovcnt].iov_base = u_name;
453 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
454 		    &optiov[opt.uio_iovcnt].iov_len);
455 		if (error) {
456 			free(u_path, M_TEMP);
457 			return (error);
458 		}
459 		opt.uio_iovcnt++;
460 	}
461 #ifdef INET
462 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
463 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
464 	opt.uio_iovcnt++;
465 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
466 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
467 	if (j->version == 0)
468 		u_ip4->s_addr = j->ip4s;
469 	else {
470 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
471 		if (error) {
472 			free(u_path, M_TEMP);
473 			return (error);
474 		}
475 	}
476 	opt.uio_iovcnt++;
477 #endif
478 #ifdef INET6
479 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
480 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
481 	opt.uio_iovcnt++;
482 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
483 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
484 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
485 	if (error) {
486 		free(u_path, M_TEMP);
487 		return (error);
488 	}
489 	opt.uio_iovcnt++;
490 #endif
491 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
492 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
493 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
494 	free(u_path, M_TEMP);
495 	return (error);
496 }
497 
498 
499 /*
500  * struct jail_set_args {
501  *	struct iovec *iovp;
502  *	unsigned int iovcnt;
503  *	int flags;
504  * };
505  */
506 int
507 sys_jail_set(struct thread *td, struct jail_set_args *uap)
508 {
509 	struct uio *auio;
510 	int error;
511 
512 	/* Check that we have an even number of iovecs. */
513 	if (uap->iovcnt & 1)
514 		return (EINVAL);
515 
516 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
517 	if (error)
518 		return (error);
519 	error = kern_jail_set(td, auio, uap->flags);
520 	free(auio, M_IOV);
521 	return (error);
522 }
523 
524 int
525 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
526 {
527 	struct nameidata nd;
528 #ifdef INET
529 	struct in_addr *ip4;
530 #endif
531 #ifdef INET6
532 	struct in6_addr *ip6;
533 #endif
534 	struct vfsopt *opt;
535 	struct vfsoptlist *opts;
536 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
537 	struct vnode *root;
538 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
539 	char *g_path;
540 #if defined(INET) || defined(INET6)
541 	struct prison *tppr;
542 	void *op;
543 #endif
544 	unsigned long hid;
545 	size_t namelen, onamelen;
546 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
547 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
548 	int fi, jid, jsys, len, level;
549 	int childmax, rsnum, slevel;
550 	int fullpath_disabled;
551 #if defined(INET) || defined(INET6)
552 	int ii, ij;
553 #endif
554 #ifdef INET
555 	int ip4s, redo_ip4;
556 #endif
557 #ifdef INET6
558 	int ip6s, redo_ip6;
559 #endif
560 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
561 	unsigned tallow;
562 	char numbuf[12];
563 
564 	error = priv_check(td, PRIV_JAIL_SET);
565 	if (!error && (flags & JAIL_ATTACH))
566 		error = priv_check(td, PRIV_JAIL_ATTACH);
567 	if (error)
568 		return (error);
569 	mypr = ppr = td->td_ucred->cr_prison;
570 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
571 		return (EPERM);
572 	if (flags & ~JAIL_SET_MASK)
573 		return (EINVAL);
574 
575 	/*
576 	 * Check all the parameters before committing to anything.  Not all
577 	 * errors can be caught early, but we may as well try.  Also, this
578 	 * takes care of some expensive stuff (path lookup) before getting
579 	 * the allprison lock.
580 	 *
581 	 * XXX Jails are not filesystems, and jail parameters are not mount
582 	 *     options.  But it makes more sense to re-use the vfsopt code
583 	 *     than duplicate it under a different name.
584 	 */
585 	error = vfs_buildopts(optuio, &opts);
586 	if (error)
587 		return (error);
588 #ifdef INET
589 	ip4 = NULL;
590 #endif
591 #ifdef INET6
592 	ip6 = NULL;
593 #endif
594 	g_path = NULL;
595 
596 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
597 	if (error == ENOENT)
598 		jid = 0;
599 	else if (error != 0)
600 		goto done_free;
601 
602 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
603 	if (error == ENOENT)
604 		gotslevel = 0;
605 	else if (error != 0)
606 		goto done_free;
607 	else
608 		gotslevel = 1;
609 
610 	error =
611 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
612 	if (error == ENOENT)
613 		gotchildmax = 0;
614 	else if (error != 0)
615 		goto done_free;
616 	else
617 		gotchildmax = 1;
618 
619 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
620 	if (error == ENOENT)
621 		gotenforce = 0;
622 	else if (error != 0)
623 		goto done_free;
624 	else if (enforce < 0 || enforce > 2) {
625 		error = EINVAL;
626 		goto done_free;
627 	} else
628 		gotenforce = 1;
629 
630 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
631 	if (error == ENOENT)
632 		gotrsnum = 0;
633 	else if (error != 0)
634 		goto done_free;
635 	else
636 		gotrsnum = 1;
637 
638 	pr_flags = ch_flags = 0;
639 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
640 	    fi++) {
641 		if (pr_flag_names[fi] == NULL)
642 			continue;
643 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
644 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
645 	}
646 	ch_flags |= pr_flags;
647 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
648 	    fi++) {
649 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
650 		    sizeof(jsys));
651 		if (error == ENOENT)
652 			continue;
653 		if (error != 0)
654 			goto done_free;
655 		switch (jsys) {
656 		case JAIL_SYS_DISABLE:
657 			if (!pr_flag_jailsys[fi].disable) {
658 				error = EINVAL;
659 				goto done_free;
660 			}
661 			pr_flags |= pr_flag_jailsys[fi].disable;
662 			break;
663 		case JAIL_SYS_NEW:
664 			pr_flags |= pr_flag_jailsys[fi].new;
665 			break;
666 		case JAIL_SYS_INHERIT:
667 			break;
668 		default:
669 			error = EINVAL;
670 			goto done_free;
671 		}
672 		ch_flags |=
673 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
674 	}
675 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
676 	    && !(pr_flags & PR_PERSIST)) {
677 		error = EINVAL;
678 		vfs_opterror(opts, "new jail must persist or attach");
679 		goto done_errmsg;
680 	}
681 #ifdef VIMAGE
682 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
683 		error = EINVAL;
684 		vfs_opterror(opts, "vnet cannot be changed after creation");
685 		goto done_errmsg;
686 	}
687 #endif
688 #ifdef INET
689 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
690 		error = EINVAL;
691 		vfs_opterror(opts, "ip4 cannot be changed after creation");
692 		goto done_errmsg;
693 	}
694 #endif
695 #ifdef INET6
696 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
697 		error = EINVAL;
698 		vfs_opterror(opts, "ip6 cannot be changed after creation");
699 		goto done_errmsg;
700 	}
701 #endif
702 
703 	pr_allow = ch_allow = 0;
704 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
705 	    fi++) {
706 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
707 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
708 	}
709 	ch_allow |= pr_allow;
710 
711 	error = vfs_getopt(opts, "name", (void **)&name, &len);
712 	if (error == ENOENT)
713 		name = NULL;
714 	else if (error != 0)
715 		goto done_free;
716 	else {
717 		if (len == 0 || name[len - 1] != '\0') {
718 			error = EINVAL;
719 			goto done_free;
720 		}
721 		if (len > MAXHOSTNAMELEN) {
722 			error = ENAMETOOLONG;
723 			goto done_free;
724 		}
725 	}
726 
727 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
728 	if (error == ENOENT)
729 		host = NULL;
730 	else if (error != 0)
731 		goto done_free;
732 	else {
733 		ch_flags |= PR_HOST;
734 		pr_flags |= PR_HOST;
735 		if (len == 0 || host[len - 1] != '\0') {
736 			error = EINVAL;
737 			goto done_free;
738 		}
739 		if (len > MAXHOSTNAMELEN) {
740 			error = ENAMETOOLONG;
741 			goto done_free;
742 		}
743 	}
744 
745 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
746 	if (error == ENOENT)
747 		domain = NULL;
748 	else if (error != 0)
749 		goto done_free;
750 	else {
751 		ch_flags |= PR_HOST;
752 		pr_flags |= PR_HOST;
753 		if (len == 0 || domain[len - 1] != '\0') {
754 			error = EINVAL;
755 			goto done_free;
756 		}
757 		if (len > MAXHOSTNAMELEN) {
758 			error = ENAMETOOLONG;
759 			goto done_free;
760 		}
761 	}
762 
763 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
764 	if (error == ENOENT)
765 		uuid = NULL;
766 	else if (error != 0)
767 		goto done_free;
768 	else {
769 		ch_flags |= PR_HOST;
770 		pr_flags |= PR_HOST;
771 		if (len == 0 || uuid[len - 1] != '\0') {
772 			error = EINVAL;
773 			goto done_free;
774 		}
775 		if (len > HOSTUUIDLEN) {
776 			error = ENAMETOOLONG;
777 			goto done_free;
778 		}
779 	}
780 
781 #ifdef COMPAT_FREEBSD32
782 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
783 		uint32_t hid32;
784 
785 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
786 		hid = hid32;
787 	} else
788 #endif
789 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
790 	if (error == ENOENT)
791 		gothid = 0;
792 	else if (error != 0)
793 		goto done_free;
794 	else {
795 		gothid = 1;
796 		ch_flags |= PR_HOST;
797 		pr_flags |= PR_HOST;
798 	}
799 
800 #ifdef INET
801 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
802 	if (error == ENOENT)
803 		ip4s = 0;
804 	else if (error != 0)
805 		goto done_free;
806 	else if (ip4s & (sizeof(*ip4) - 1)) {
807 		error = EINVAL;
808 		goto done_free;
809 	} else {
810 		ch_flags |= PR_IP4_USER;
811 		pr_flags |= PR_IP4_USER;
812 		if (ip4s > 0) {
813 			ip4s /= sizeof(*ip4);
814 			if (ip4s > jail_max_af_ips) {
815 				error = EINVAL;
816 				vfs_opterror(opts, "too many IPv4 addresses");
817 				goto done_errmsg;
818 			}
819 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
820 			bcopy(op, ip4, ip4s * sizeof(*ip4));
821 			/*
822 			 * IP addresses are all sorted but ip[0] to preserve
823 			 * the primary IP address as given from userland.
824 			 * This special IP is used for unbound outgoing
825 			 * connections as well for "loopback" traffic in case
826 			 * source address selection cannot find any more fitting
827 			 * address to connect from.
828 			 */
829 			if (ip4s > 1)
830 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
831 			/*
832 			 * Check for duplicate addresses and do some simple
833 			 * zero and broadcast checks. If users give other bogus
834 			 * addresses it is their problem.
835 			 *
836 			 * We do not have to care about byte order for these
837 			 * checks so we will do them in NBO.
838 			 */
839 			for (ii = 0; ii < ip4s; ii++) {
840 				if (ip4[ii].s_addr == INADDR_ANY ||
841 				    ip4[ii].s_addr == INADDR_BROADCAST) {
842 					error = EINVAL;
843 					goto done_free;
844 				}
845 				if ((ii+1) < ip4s &&
846 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
847 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
848 					error = EINVAL;
849 					goto done_free;
850 				}
851 			}
852 		}
853 	}
854 #endif
855 
856 #ifdef INET6
857 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
858 	if (error == ENOENT)
859 		ip6s = 0;
860 	else if (error != 0)
861 		goto done_free;
862 	else if (ip6s & (sizeof(*ip6) - 1)) {
863 		error = EINVAL;
864 		goto done_free;
865 	} else {
866 		ch_flags |= PR_IP6_USER;
867 		pr_flags |= PR_IP6_USER;
868 		if (ip6s > 0) {
869 			ip6s /= sizeof(*ip6);
870 			if (ip6s > jail_max_af_ips) {
871 				error = EINVAL;
872 				vfs_opterror(opts, "too many IPv6 addresses");
873 				goto done_errmsg;
874 			}
875 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
876 			bcopy(op, ip6, ip6s * sizeof(*ip6));
877 			if (ip6s > 1)
878 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
879 			for (ii = 0; ii < ip6s; ii++) {
880 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
881 					error = EINVAL;
882 					goto done_free;
883 				}
884 				if ((ii+1) < ip6s &&
885 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
886 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
887 				{
888 					error = EINVAL;
889 					goto done_free;
890 				}
891 			}
892 		}
893 	}
894 #endif
895 
896 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
897 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
898 		error = EINVAL;
899 		vfs_opterror(opts,
900 		    "vnet jails cannot have IP address restrictions");
901 		goto done_errmsg;
902 	}
903 #endif
904 
905 	fullpath_disabled = 0;
906 	root = NULL;
907 	error = vfs_getopt(opts, "path", (void **)&path, &len);
908 	if (error == ENOENT)
909 		path = NULL;
910 	else if (error != 0)
911 		goto done_free;
912 	else {
913 		if (flags & JAIL_UPDATE) {
914 			error = EINVAL;
915 			vfs_opterror(opts,
916 			    "path cannot be changed after creation");
917 			goto done_errmsg;
918 		}
919 		if (len == 0 || path[len - 1] != '\0') {
920 			error = EINVAL;
921 			goto done_free;
922 		}
923 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
924 		    path, td);
925 		error = namei(&nd);
926 		if (error)
927 			goto done_free;
928 		root = nd.ni_vp;
929 		NDFREE(&nd, NDF_ONLY_PNBUF);
930 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
931 		strlcpy(g_path, path, MAXPATHLEN);
932 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
933 		if (error == 0)
934 			path = g_path;
935 		else if (error == ENODEV) {
936 			/* proceed if sysctl debug.disablefullpath == 1 */
937 			fullpath_disabled = 1;
938 			if (len < 2 || (len == 2 && path[0] == '/'))
939 				path = NULL;
940 		} else {
941 			/* exit on other errors */
942 			goto done_free;
943 		}
944 		if (root->v_type != VDIR) {
945 			error = ENOTDIR;
946 			vput(root);
947 			goto done_free;
948 		}
949 		VOP_UNLOCK(root, 0);
950 		if (fullpath_disabled) {
951 			/* Leave room for a real-root full pathname. */
952 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
953 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
954 				error = ENAMETOOLONG;
955 				goto done_free;
956 			}
957 		}
958 	}
959 
960 	/*
961 	 * Grab the allprison lock before letting modules check their
962 	 * parameters.  Once we have it, do not let go so we'll have a
963 	 * consistent view of the OSD list.
964 	 */
965 	sx_xlock(&allprison_lock);
966 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
967 	if (error)
968 		goto done_unlock_list;
969 
970 	/* By now, all parameters should have been noted. */
971 	TAILQ_FOREACH(opt, opts, link) {
972 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
973 			error = EINVAL;
974 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
975 			goto done_unlock_list;
976 		}
977 	}
978 
979 	/*
980 	 * See if we are creating a new record or updating an existing one.
981 	 * This abuses the file error codes ENOENT and EEXIST.
982 	 */
983 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
984 	if (!cuflags) {
985 		error = EINVAL;
986 		vfs_opterror(opts, "no valid operation (create or update)");
987 		goto done_unlock_list;
988 	}
989 	pr = NULL;
990 	namelc = NULL;
991 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
992 		namelc = strrchr(name, '.');
993 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
994 		if (*p != '\0')
995 			jid = 0;
996 	}
997 	if (jid != 0) {
998 		/*
999 		 * See if a requested jid already exists.  There is an
1000 		 * information leak here if the jid exists but is not within
1001 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1002 		 * even though they cannot see the jail, and CREATE | UPDATE
1003 		 * will return ENOENT which is not normally a valid error.
1004 		 */
1005 		if (jid < 0) {
1006 			error = EINVAL;
1007 			vfs_opterror(opts, "negative jid");
1008 			goto done_unlock_list;
1009 		}
1010 		pr = prison_find(jid);
1011 		if (pr != NULL) {
1012 			ppr = pr->pr_parent;
1013 			/* Create: jid must not exist. */
1014 			if (cuflags == JAIL_CREATE) {
1015 				mtx_unlock(&pr->pr_mtx);
1016 				error = EEXIST;
1017 				vfs_opterror(opts, "jail %d already exists",
1018 				    jid);
1019 				goto done_unlock_list;
1020 			}
1021 			if (!prison_ischild(mypr, pr)) {
1022 				mtx_unlock(&pr->pr_mtx);
1023 				pr = NULL;
1024 			} else if (pr->pr_uref == 0) {
1025 				if (!(flags & JAIL_DYING)) {
1026 					mtx_unlock(&pr->pr_mtx);
1027 					error = ENOENT;
1028 					vfs_opterror(opts, "jail %d is dying",
1029 					    jid);
1030 					goto done_unlock_list;
1031 				} else if ((flags & JAIL_ATTACH) ||
1032 				    (pr_flags & PR_PERSIST)) {
1033 					/*
1034 					 * A dying jail might be resurrected
1035 					 * (via attach or persist), but first
1036 					 * it must determine if another jail
1037 					 * has claimed its name.  Accomplish
1038 					 * this by implicitly re-setting the
1039 					 * name.
1040 					 */
1041 					if (name == NULL)
1042 						name = prison_name(mypr, pr);
1043 				}
1044 			}
1045 		}
1046 		if (pr == NULL) {
1047 			/* Update: jid must exist. */
1048 			if (cuflags == JAIL_UPDATE) {
1049 				error = ENOENT;
1050 				vfs_opterror(opts, "jail %d not found", jid);
1051 				goto done_unlock_list;
1052 			}
1053 		}
1054 	}
1055 	/*
1056 	 * If the caller provided a name, look for a jail by that name.
1057 	 * This has different semantics for creates and updates keyed by jid
1058 	 * (where the name must not already exist in a different jail),
1059 	 * and updates keyed by the name itself (where the name must exist
1060 	 * because that is the jail being updated).
1061 	 */
1062 	if (name != NULL) {
1063 		namelc = strrchr(name, '.');
1064 		if (namelc == NULL)
1065 			namelc = name;
1066 		else {
1067 			/*
1068 			 * This is a hierarchical name.  Split it into the
1069 			 * parent and child names, and make sure the parent
1070 			 * exists or matches an already found jail.
1071 			 */
1072 			*namelc = '\0';
1073 			if (pr != NULL) {
1074 				if (strncmp(name, ppr->pr_name, namelc - name)
1075 				    || ppr->pr_name[namelc - name] != '\0') {
1076 					mtx_unlock(&pr->pr_mtx);
1077 					error = EINVAL;
1078 					vfs_opterror(opts,
1079 					    "cannot change jail's parent");
1080 					goto done_unlock_list;
1081 				}
1082 			} else {
1083 				ppr = prison_find_name(mypr, name);
1084 				if (ppr == NULL) {
1085 					error = ENOENT;
1086 					vfs_opterror(opts,
1087 					    "jail \"%s\" not found", name);
1088 					goto done_unlock_list;
1089 				}
1090 				mtx_unlock(&ppr->pr_mtx);
1091 			}
1092 			name = ++namelc;
1093 		}
1094 		if (name[0] != '\0') {
1095 			namelen =
1096 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1097  name_again:
1098 			deadpr = NULL;
1099 			FOREACH_PRISON_CHILD(ppr, tpr) {
1100 				if (tpr != pr && tpr->pr_ref > 0 &&
1101 				    !strcmp(tpr->pr_name + namelen, name)) {
1102 					if (pr == NULL &&
1103 					    cuflags != JAIL_CREATE) {
1104 						mtx_lock(&tpr->pr_mtx);
1105 						if (tpr->pr_ref > 0) {
1106 							/*
1107 							 * Use this jail
1108 							 * for updates.
1109 							 */
1110 							if (tpr->pr_uref > 0) {
1111 								pr = tpr;
1112 								break;
1113 							}
1114 							deadpr = tpr;
1115 						}
1116 						mtx_unlock(&tpr->pr_mtx);
1117 					} else if (tpr->pr_uref > 0) {
1118 						/*
1119 						 * Create, or update(jid):
1120 						 * name must not exist in an
1121 						 * active sibling jail.
1122 						 */
1123 						error = EEXIST;
1124 						if (pr != NULL)
1125 							mtx_unlock(&pr->pr_mtx);
1126 						vfs_opterror(opts,
1127 						   "jail \"%s\" already exists",
1128 						   name);
1129 						goto done_unlock_list;
1130 					}
1131 				}
1132 			}
1133 			/* If no active jail is found, use a dying one. */
1134 			if (deadpr != NULL && pr == NULL) {
1135 				if (flags & JAIL_DYING) {
1136 					mtx_lock(&deadpr->pr_mtx);
1137 					if (deadpr->pr_ref == 0) {
1138 						mtx_unlock(&deadpr->pr_mtx);
1139 						goto name_again;
1140 					}
1141 					pr = deadpr;
1142 				} else if (cuflags == JAIL_UPDATE) {
1143 					error = ENOENT;
1144 					vfs_opterror(opts,
1145 					    "jail \"%s\" is dying", name);
1146 					goto done_unlock_list;
1147 				}
1148 			}
1149 			/* Update: name must exist if no jid. */
1150 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1151 				error = ENOENT;
1152 				vfs_opterror(opts, "jail \"%s\" not found",
1153 				    name);
1154 				goto done_unlock_list;
1155 			}
1156 		}
1157 	}
1158 	/* Update: must provide a jid or name. */
1159 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1160 		error = ENOENT;
1161 		vfs_opterror(opts, "update specified no jail");
1162 		goto done_unlock_list;
1163 	}
1164 
1165 	/* If there's no prison to update, create a new one and link it in. */
1166 	if (pr == NULL) {
1167 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1168 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1169 				error = EPERM;
1170 				vfs_opterror(opts, "prison limit exceeded");
1171 				goto done_unlock_list;
1172 			}
1173 		created = 1;
1174 		mtx_lock(&ppr->pr_mtx);
1175 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1176 			mtx_unlock(&ppr->pr_mtx);
1177 			error = ENOENT;
1178 			vfs_opterror(opts, "parent jail went away!");
1179 			goto done_unlock_list;
1180 		}
1181 		ppr->pr_ref++;
1182 		ppr->pr_uref++;
1183 		mtx_unlock(&ppr->pr_mtx);
1184 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1185 		if (jid == 0) {
1186 			/* Find the next free jid. */
1187 			jid = lastprid + 1;
1188  findnext:
1189 			if (jid == JAIL_MAX)
1190 				jid = 1;
1191 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1192 				if (tpr->pr_id < jid)
1193 					continue;
1194 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1195 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1196 					break;
1197 				}
1198 				if (jid == lastprid) {
1199 					error = EAGAIN;
1200 					vfs_opterror(opts,
1201 					    "no available jail IDs");
1202 					free(pr, M_PRISON);
1203 					prison_deref(ppr, PD_DEREF |
1204 					    PD_DEUREF | PD_LIST_XLOCKED);
1205 					goto done_releroot;
1206 				}
1207 				jid++;
1208 				goto findnext;
1209 			}
1210 			lastprid = jid;
1211 		} else {
1212 			/*
1213 			 * The jail already has a jid (that did not yet exist),
1214 			 * so just find where to insert it.
1215 			 */
1216 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1217 				if (tpr->pr_id >= jid) {
1218 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1219 					break;
1220 				}
1221 		}
1222 		if (tpr == NULL)
1223 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1224 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1225 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1226 			tpr->pr_childcount++;
1227 
1228 		pr->pr_parent = ppr;
1229 		pr->pr_id = jid;
1230 
1231 		/* Set some default values, and inherit some from the parent. */
1232 		if (name == NULL)
1233 			name = "";
1234 		if (path == NULL) {
1235 			path = "/";
1236 			root = mypr->pr_root;
1237 			vref(root);
1238 		}
1239 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1240 		pr->pr_flags |= PR_HOST;
1241 #if defined(INET) || defined(INET6)
1242 #ifdef VIMAGE
1243 		if (!(pr_flags & PR_VNET))
1244 #endif
1245 		{
1246 #ifdef INET
1247 			if (!(ch_flags & PR_IP4_USER))
1248 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1249 			else if (!(pr_flags & PR_IP4_USER)) {
1250 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1251 				if (ppr->pr_ip4 != NULL) {
1252 					pr->pr_ip4s = ppr->pr_ip4s;
1253 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1254 					    sizeof(struct in_addr), M_PRISON,
1255 					    M_WAITOK);
1256 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1257 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1258 				}
1259 			}
1260 #endif
1261 #ifdef INET6
1262 			if (!(ch_flags & PR_IP6_USER))
1263 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1264 			else if (!(pr_flags & PR_IP6_USER)) {
1265 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1266 				if (ppr->pr_ip6 != NULL) {
1267 					pr->pr_ip6s = ppr->pr_ip6s;
1268 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1269 					    sizeof(struct in6_addr), M_PRISON,
1270 					    M_WAITOK);
1271 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1272 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1273 				}
1274 			}
1275 #endif
1276 		}
1277 #endif
1278 		/* Source address selection is always on by default. */
1279 		pr->pr_flags |= _PR_IP_SADDRSEL;
1280 
1281 		pr->pr_securelevel = ppr->pr_securelevel;
1282 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1283 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1284 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1285 
1286 		LIST_INIT(&pr->pr_children);
1287 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1288 
1289 #ifdef VIMAGE
1290 		/* Allocate a new vnet if specified. */
1291 		pr->pr_vnet = (pr_flags & PR_VNET)
1292 		    ? vnet_alloc() : ppr->pr_vnet;
1293 #endif
1294 		/*
1295 		 * Allocate a dedicated cpuset for each jail.
1296 		 * Unlike other initial settings, this may return an erorr.
1297 		 */
1298 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1299 		if (error) {
1300 			prison_deref(pr, PD_LIST_XLOCKED);
1301 			goto done_releroot;
1302 		}
1303 
1304 		mtx_lock(&pr->pr_mtx);
1305 		/*
1306 		 * New prisons do not yet have a reference, because we do not
1307 		 * want other to see the incomplete prison once the
1308 		 * allprison_lock is downgraded.
1309 		 */
1310 	} else {
1311 		created = 0;
1312 		/*
1313 		 * Grab a reference for existing prisons, to ensure they
1314 		 * continue to exist for the duration of the call.
1315 		 */
1316 		pr->pr_ref++;
1317 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1318 		if ((pr->pr_flags & PR_VNET) &&
1319 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1320 			error = EINVAL;
1321 			vfs_opterror(opts,
1322 			    "vnet jails cannot have IP address restrictions");
1323 			goto done_deref_locked;
1324 		}
1325 #endif
1326 #ifdef INET
1327 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1328 			error = EINVAL;
1329 			vfs_opterror(opts,
1330 			    "ip4 cannot be changed after creation");
1331 			goto done_deref_locked;
1332 		}
1333 #endif
1334 #ifdef INET6
1335 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1336 			error = EINVAL;
1337 			vfs_opterror(opts,
1338 			    "ip6 cannot be changed after creation");
1339 			goto done_deref_locked;
1340 		}
1341 #endif
1342 	}
1343 
1344 	/* Do final error checking before setting anything. */
1345 	if (gotslevel) {
1346 		if (slevel < ppr->pr_securelevel) {
1347 			error = EPERM;
1348 			goto done_deref_locked;
1349 		}
1350 	}
1351 	if (gotchildmax) {
1352 		if (childmax >= ppr->pr_childmax) {
1353 			error = EPERM;
1354 			goto done_deref_locked;
1355 		}
1356 	}
1357 	if (gotenforce) {
1358 		if (enforce < ppr->pr_enforce_statfs) {
1359 			error = EPERM;
1360 			goto done_deref_locked;
1361 		}
1362 	}
1363 	if (gotrsnum) {
1364 		/*
1365 		 * devfs_rsnum is a uint16_t
1366 		 */
1367 		if (rsnum < 0 || rsnum > 65535) {
1368 			error = EINVAL;
1369 			goto done_deref_locked;
1370 		}
1371 		/*
1372 		 * Nested jails always inherit parent's devfs ruleset
1373 		 */
1374 		if (jailed(td->td_ucred)) {
1375 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1376 				error = EPERM;
1377 				goto done_deref_locked;
1378 			} else
1379 				rsnum = ppr->pr_devfs_rsnum;
1380 		}
1381 	}
1382 #ifdef INET
1383 	if (ip4s > 0) {
1384 		if (ppr->pr_flags & PR_IP4) {
1385 			/*
1386 			 * Make sure the new set of IP addresses is a
1387 			 * subset of the parent's list.  Don't worry
1388 			 * about the parent being unlocked, as any
1389 			 * setting is done with allprison_lock held.
1390 			 */
1391 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1392 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1393 					break;
1394 			if (ij == ppr->pr_ip4s) {
1395 				error = EPERM;
1396 				goto done_deref_locked;
1397 			}
1398 			if (ip4s > 1) {
1399 				for (ii = ij = 1; ii < ip4s; ii++) {
1400 					if (ip4[ii].s_addr ==
1401 					    ppr->pr_ip4[0].s_addr)
1402 						continue;
1403 					for (; ij < ppr->pr_ip4s; ij++)
1404 						if (ip4[ii].s_addr ==
1405 						    ppr->pr_ip4[ij].s_addr)
1406 							break;
1407 					if (ij == ppr->pr_ip4s)
1408 						break;
1409 				}
1410 				if (ij == ppr->pr_ip4s) {
1411 					error = EPERM;
1412 					goto done_deref_locked;
1413 				}
1414 			}
1415 		}
1416 		/*
1417 		 * Check for conflicting IP addresses.  We permit them
1418 		 * if there is no more than one IP on each jail.  If
1419 		 * there is a duplicate on a jail with more than one
1420 		 * IP stop checking and return error.
1421 		 */
1422 		tppr = ppr;
1423 #ifdef VIMAGE
1424 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1425 			if (tppr->pr_flags & PR_VNET)
1426 				break;
1427 #endif
1428 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1429 			if (tpr == pr ||
1430 #ifdef VIMAGE
1431 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1432 #endif
1433 			    tpr->pr_uref == 0) {
1434 				descend = 0;
1435 				continue;
1436 			}
1437 			if (!(tpr->pr_flags & PR_IP4_USER))
1438 				continue;
1439 			descend = 0;
1440 			if (tpr->pr_ip4 == NULL ||
1441 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1442 				continue;
1443 			for (ii = 0; ii < ip4s; ii++) {
1444 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1445 					error = EADDRINUSE;
1446 					vfs_opterror(opts,
1447 					    "IPv4 addresses clash");
1448 					goto done_deref_locked;
1449 				}
1450 			}
1451 		}
1452 	}
1453 #endif
1454 #ifdef INET6
1455 	if (ip6s > 0) {
1456 		if (ppr->pr_flags & PR_IP6) {
1457 			/*
1458 			 * Make sure the new set of IP addresses is a
1459 			 * subset of the parent's list.
1460 			 */
1461 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1462 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1463 				    &ppr->pr_ip6[ij]))
1464 					break;
1465 			if (ij == ppr->pr_ip6s) {
1466 				error = EPERM;
1467 				goto done_deref_locked;
1468 			}
1469 			if (ip6s > 1) {
1470 				for (ii = ij = 1; ii < ip6s; ii++) {
1471 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1472 					     &ppr->pr_ip6[0]))
1473 						continue;
1474 					for (; ij < ppr->pr_ip6s; ij++)
1475 						if (IN6_ARE_ADDR_EQUAL(
1476 						    &ip6[ii], &ppr->pr_ip6[ij]))
1477 							break;
1478 					if (ij == ppr->pr_ip6s)
1479 						break;
1480 				}
1481 				if (ij == ppr->pr_ip6s) {
1482 					error = EPERM;
1483 					goto done_deref_locked;
1484 				}
1485 			}
1486 		}
1487 		/* Check for conflicting IP addresses. */
1488 		tppr = ppr;
1489 #ifdef VIMAGE
1490 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1491 			if (tppr->pr_flags & PR_VNET)
1492 				break;
1493 #endif
1494 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1495 			if (tpr == pr ||
1496 #ifdef VIMAGE
1497 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1498 #endif
1499 			    tpr->pr_uref == 0) {
1500 				descend = 0;
1501 				continue;
1502 			}
1503 			if (!(tpr->pr_flags & PR_IP6_USER))
1504 				continue;
1505 			descend = 0;
1506 			if (tpr->pr_ip6 == NULL ||
1507 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1508 				continue;
1509 			for (ii = 0; ii < ip6s; ii++) {
1510 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1511 					error = EADDRINUSE;
1512 					vfs_opterror(opts,
1513 					    "IPv6 addresses clash");
1514 					goto done_deref_locked;
1515 				}
1516 			}
1517 		}
1518 	}
1519 #endif
1520 	onamelen = namelen = 0;
1521 	if (name != NULL) {
1522 		/* Give a default name of the jid. */
1523 		if (name[0] == '\0')
1524 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1525 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1526 		    *p == '\0')) {
1527 			error = EINVAL;
1528 			vfs_opterror(opts,
1529 			    "name cannot be numeric (unless it is the jid)");
1530 			goto done_deref_locked;
1531 		}
1532 		/*
1533 		 * Make sure the name isn't too long for the prison or its
1534 		 * children.
1535 		 */
1536 		onamelen = strlen(pr->pr_name);
1537 		namelen = strlen(name);
1538 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1539 			error = ENAMETOOLONG;
1540 			goto done_deref_locked;
1541 		}
1542 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1543 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1544 			    sizeof(pr->pr_name)) {
1545 				error = ENAMETOOLONG;
1546 				goto done_deref_locked;
1547 			}
1548 		}
1549 	}
1550 	if (pr_allow & ~ppr->pr_allow) {
1551 		error = EPERM;
1552 		goto done_deref_locked;
1553 	}
1554 
1555 	/* Set the parameters of the prison. */
1556 #ifdef INET
1557 	redo_ip4 = 0;
1558 	if (pr_flags & PR_IP4_USER) {
1559 		pr->pr_flags |= PR_IP4;
1560 		free(pr->pr_ip4, M_PRISON);
1561 		pr->pr_ip4s = ip4s;
1562 		pr->pr_ip4 = ip4;
1563 		ip4 = NULL;
1564 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1565 #ifdef VIMAGE
1566 			if (tpr->pr_flags & PR_VNET) {
1567 				descend = 0;
1568 				continue;
1569 			}
1570 #endif
1571 			if (prison_restrict_ip4(tpr, NULL)) {
1572 				redo_ip4 = 1;
1573 				descend = 0;
1574 			}
1575 		}
1576 	}
1577 #endif
1578 #ifdef INET6
1579 	redo_ip6 = 0;
1580 	if (pr_flags & PR_IP6_USER) {
1581 		pr->pr_flags |= PR_IP6;
1582 		free(pr->pr_ip6, M_PRISON);
1583 		pr->pr_ip6s = ip6s;
1584 		pr->pr_ip6 = ip6;
1585 		ip6 = NULL;
1586 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1587 #ifdef VIMAGE
1588 			if (tpr->pr_flags & PR_VNET) {
1589 				descend = 0;
1590 				continue;
1591 			}
1592 #endif
1593 			if (prison_restrict_ip6(tpr, NULL)) {
1594 				redo_ip6 = 1;
1595 				descend = 0;
1596 			}
1597 		}
1598 	}
1599 #endif
1600 	if (gotslevel) {
1601 		pr->pr_securelevel = slevel;
1602 		/* Set all child jails to be at least this level. */
1603 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1604 			if (tpr->pr_securelevel < slevel)
1605 				tpr->pr_securelevel = slevel;
1606 	}
1607 	if (gotchildmax) {
1608 		pr->pr_childmax = childmax;
1609 		/* Set all child jails to under this limit. */
1610 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1611 			if (tpr->pr_childmax > childmax - level)
1612 				tpr->pr_childmax = childmax > level
1613 				    ? childmax - level : 0;
1614 	}
1615 	if (gotenforce) {
1616 		pr->pr_enforce_statfs = enforce;
1617 		/* Pass this restriction on to the children. */
1618 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1619 			if (tpr->pr_enforce_statfs < enforce)
1620 				tpr->pr_enforce_statfs = enforce;
1621 	}
1622 	if (gotrsnum) {
1623 		pr->pr_devfs_rsnum = rsnum;
1624 		/* Pass this restriction on to the children. */
1625 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1626 			tpr->pr_devfs_rsnum = rsnum;
1627 	}
1628 	if (name != NULL) {
1629 		if (ppr == &prison0)
1630 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1631 		else
1632 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1633 			    ppr->pr_name, name);
1634 		/* Change this component of child names. */
1635 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1636 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1637 			    strlen(tpr->pr_name + onamelen) + 1);
1638 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1639 		}
1640 	}
1641 	if (path != NULL) {
1642 		/* Try to keep a real-rooted full pathname. */
1643 		if (fullpath_disabled && path[0] == '/' &&
1644 		    strcmp(mypr->pr_path, "/"))
1645 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1646 			    mypr->pr_path, path);
1647 		else
1648 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1649 		pr->pr_root = root;
1650 	}
1651 	if (PR_HOST & ch_flags & ~pr_flags) {
1652 		if (pr->pr_flags & PR_HOST) {
1653 			/*
1654 			 * Copy the parent's host info.  As with pr_ip4 above,
1655 			 * the lack of a lock on the parent is not a problem;
1656 			 * it is always set with allprison_lock at least
1657 			 * shared, and is held exclusively here.
1658 			 */
1659 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1660 			    sizeof(pr->pr_hostname));
1661 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1662 			    sizeof(pr->pr_domainname));
1663 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1664 			    sizeof(pr->pr_hostuuid));
1665 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1666 		}
1667 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1668 		/* Set this prison, and any descendants without PR_HOST. */
1669 		if (host != NULL)
1670 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1671 		if (domain != NULL)
1672 			strlcpy(pr->pr_domainname, domain,
1673 			    sizeof(pr->pr_domainname));
1674 		if (uuid != NULL)
1675 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1676 		if (gothid)
1677 			pr->pr_hostid = hid;
1678 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1679 			if (tpr->pr_flags & PR_HOST)
1680 				descend = 0;
1681 			else {
1682 				if (host != NULL)
1683 					strlcpy(tpr->pr_hostname,
1684 					    pr->pr_hostname,
1685 					    sizeof(tpr->pr_hostname));
1686 				if (domain != NULL)
1687 					strlcpy(tpr->pr_domainname,
1688 					    pr->pr_domainname,
1689 					    sizeof(tpr->pr_domainname));
1690 				if (uuid != NULL)
1691 					strlcpy(tpr->pr_hostuuid,
1692 					    pr->pr_hostuuid,
1693 					    sizeof(tpr->pr_hostuuid));
1694 				if (gothid)
1695 					tpr->pr_hostid = hid;
1696 			}
1697 		}
1698 	}
1699 	if ((tallow = ch_allow & ~pr_allow)) {
1700 		/* Clear allow bits in all children. */
1701 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1702 			tpr->pr_allow &= ~tallow;
1703 	}
1704 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1705 	/*
1706 	 * Persistent prisons get an extra reference, and prisons losing their
1707 	 * persist flag lose that reference.  Only do this for existing prisons
1708 	 * for now, so new ones will remain unseen until after the module
1709 	 * handlers have completed.
1710 	 */
1711 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1712 		if (pr_flags & PR_PERSIST) {
1713 			pr->pr_ref++;
1714 			pr->pr_uref++;
1715 		} else {
1716 			pr->pr_ref--;
1717 			pr->pr_uref--;
1718 		}
1719 	}
1720 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1721 	mtx_unlock(&pr->pr_mtx);
1722 
1723 #ifdef RACCT
1724 	if (created)
1725 		prison_racct_attach(pr);
1726 #endif
1727 
1728 	/* Locks may have prevented a complete restriction of child IP
1729 	 * addresses.  If so, allocate some more memory and try again.
1730 	 */
1731 #ifdef INET
1732 	while (redo_ip4) {
1733 		ip4s = pr->pr_ip4s;
1734 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1735 		mtx_lock(&pr->pr_mtx);
1736 		redo_ip4 = 0;
1737 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1738 #ifdef VIMAGE
1739 			if (tpr->pr_flags & PR_VNET) {
1740 				descend = 0;
1741 				continue;
1742 			}
1743 #endif
1744 			if (prison_restrict_ip4(tpr, ip4)) {
1745 				if (ip4 != NULL)
1746 					ip4 = NULL;
1747 				else
1748 					redo_ip4 = 1;
1749 			}
1750 		}
1751 		mtx_unlock(&pr->pr_mtx);
1752 	}
1753 #endif
1754 #ifdef INET6
1755 	while (redo_ip6) {
1756 		ip6s = pr->pr_ip6s;
1757 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1758 		mtx_lock(&pr->pr_mtx);
1759 		redo_ip6 = 0;
1760 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1761 #ifdef VIMAGE
1762 			if (tpr->pr_flags & PR_VNET) {
1763 				descend = 0;
1764 				continue;
1765 			}
1766 #endif
1767 			if (prison_restrict_ip6(tpr, ip6)) {
1768 				if (ip6 != NULL)
1769 					ip6 = NULL;
1770 				else
1771 					redo_ip6 = 1;
1772 			}
1773 		}
1774 		mtx_unlock(&pr->pr_mtx);
1775 	}
1776 #endif
1777 
1778 	/* Let the modules do their work. */
1779 	sx_downgrade(&allprison_lock);
1780 	if (created) {
1781 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1782 		if (error) {
1783 			prison_deref(pr, PD_LIST_SLOCKED);
1784 			goto done_errmsg;
1785 		}
1786 	}
1787 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1788 	if (error) {
1789 		prison_deref(pr, created
1790 		    ? PD_LIST_SLOCKED
1791 		    : PD_DEREF | PD_LIST_SLOCKED);
1792 		goto done_errmsg;
1793 	}
1794 
1795 	/* Attach this process to the prison if requested. */
1796 	if (flags & JAIL_ATTACH) {
1797 		mtx_lock(&pr->pr_mtx);
1798 		error = do_jail_attach(td, pr);
1799 		if (error) {
1800 			vfs_opterror(opts, "attach failed");
1801 			if (!created)
1802 				prison_deref(pr, PD_DEREF);
1803 			goto done_errmsg;
1804 		}
1805 	}
1806 
1807 #ifdef RACCT
1808 	if (!created) {
1809 		if (!(flags & JAIL_ATTACH))
1810 			sx_sunlock(&allprison_lock);
1811 		prison_racct_modify(pr);
1812 		if (!(flags & JAIL_ATTACH))
1813 			sx_slock(&allprison_lock);
1814 	}
1815 #endif
1816 
1817 	td->td_retval[0] = pr->pr_id;
1818 
1819 	/*
1820 	 * Now that it is all there, drop the temporary reference from existing
1821 	 * prisons.  Or add a reference to newly created persistent prisons
1822 	 * (which was not done earlier so that the prison would not be publicly
1823 	 * visible).
1824 	 */
1825 	if (!created) {
1826 		prison_deref(pr, (flags & JAIL_ATTACH)
1827 		    ? PD_DEREF
1828 		    : PD_DEREF | PD_LIST_SLOCKED);
1829 	} else {
1830 		if (pr_flags & PR_PERSIST) {
1831 			mtx_lock(&pr->pr_mtx);
1832 			pr->pr_ref++;
1833 			pr->pr_uref++;
1834 			mtx_unlock(&pr->pr_mtx);
1835 		}
1836 		if (!(flags & JAIL_ATTACH))
1837 			sx_sunlock(&allprison_lock);
1838 	}
1839 
1840 	goto done_errmsg;
1841 
1842  done_deref_locked:
1843 	prison_deref(pr, created
1844 	    ? PD_LOCKED | PD_LIST_XLOCKED
1845 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1846 	goto done_releroot;
1847  done_unlock_list:
1848 	sx_xunlock(&allprison_lock);
1849  done_releroot:
1850 	if (root != NULL)
1851 		vrele(root);
1852  done_errmsg:
1853 	if (error) {
1854 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1855 		if (errmsg_len > 0) {
1856 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1857 			if (errmsg_pos > 0) {
1858 				if (optuio->uio_segflg == UIO_SYSSPACE)
1859 					bcopy(errmsg,
1860 					   optuio->uio_iov[errmsg_pos].iov_base,
1861 					   errmsg_len);
1862 				else
1863 					copyout(errmsg,
1864 					   optuio->uio_iov[errmsg_pos].iov_base,
1865 					   errmsg_len);
1866 			}
1867 		}
1868 	}
1869  done_free:
1870 #ifdef INET
1871 	free(ip4, M_PRISON);
1872 #endif
1873 #ifdef INET6
1874 	free(ip6, M_PRISON);
1875 #endif
1876 	if (g_path != NULL)
1877 		free(g_path, M_TEMP);
1878 	vfs_freeopts(opts);
1879 	return (error);
1880 }
1881 
1882 
1883 /*
1884  * struct jail_get_args {
1885  *	struct iovec *iovp;
1886  *	unsigned int iovcnt;
1887  *	int flags;
1888  * };
1889  */
1890 int
1891 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1892 {
1893 	struct uio *auio;
1894 	int error;
1895 
1896 	/* Check that we have an even number of iovecs. */
1897 	if (uap->iovcnt & 1)
1898 		return (EINVAL);
1899 
1900 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1901 	if (error)
1902 		return (error);
1903 	error = kern_jail_get(td, auio, uap->flags);
1904 	if (error == 0)
1905 		error = copyout(auio->uio_iov, uap->iovp,
1906 		    uap->iovcnt * sizeof (struct iovec));
1907 	free(auio, M_IOV);
1908 	return (error);
1909 }
1910 
1911 int
1912 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1913 {
1914 	struct prison *pr, *mypr;
1915 	struct vfsopt *opt;
1916 	struct vfsoptlist *opts;
1917 	char *errmsg, *name;
1918 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1919 
1920 	if (flags & ~JAIL_GET_MASK)
1921 		return (EINVAL);
1922 
1923 	/* Get the parameter list. */
1924 	error = vfs_buildopts(optuio, &opts);
1925 	if (error)
1926 		return (error);
1927 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1928 	mypr = td->td_ucred->cr_prison;
1929 
1930 	/*
1931 	 * Find the prison specified by one of: lastjid, jid, name.
1932 	 */
1933 	sx_slock(&allprison_lock);
1934 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1935 	if (error == 0) {
1936 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1937 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1938 				mtx_lock(&pr->pr_mtx);
1939 				if (pr->pr_ref > 0 &&
1940 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1941 					break;
1942 				mtx_unlock(&pr->pr_mtx);
1943 			}
1944 		}
1945 		if (pr != NULL)
1946 			goto found_prison;
1947 		error = ENOENT;
1948 		vfs_opterror(opts, "no jail after %d", jid);
1949 		goto done_unlock_list;
1950 	} else if (error != ENOENT)
1951 		goto done_unlock_list;
1952 
1953 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1954 	if (error == 0) {
1955 		if (jid != 0) {
1956 			pr = prison_find_child(mypr, jid);
1957 			if (pr != NULL) {
1958 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1959 					mtx_unlock(&pr->pr_mtx);
1960 					error = ENOENT;
1961 					vfs_opterror(opts, "jail %d is dying",
1962 					    jid);
1963 					goto done_unlock_list;
1964 				}
1965 				goto found_prison;
1966 			}
1967 			error = ENOENT;
1968 			vfs_opterror(opts, "jail %d not found", jid);
1969 			goto done_unlock_list;
1970 		}
1971 	} else if (error != ENOENT)
1972 		goto done_unlock_list;
1973 
1974 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1975 	if (error == 0) {
1976 		if (len == 0 || name[len - 1] != '\0') {
1977 			error = EINVAL;
1978 			goto done_unlock_list;
1979 		}
1980 		pr = prison_find_name(mypr, name);
1981 		if (pr != NULL) {
1982 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1983 				mtx_unlock(&pr->pr_mtx);
1984 				error = ENOENT;
1985 				vfs_opterror(opts, "jail \"%s\" is dying",
1986 				    name);
1987 				goto done_unlock_list;
1988 			}
1989 			goto found_prison;
1990 		}
1991 		error = ENOENT;
1992 		vfs_opterror(opts, "jail \"%s\" not found", name);
1993 		goto done_unlock_list;
1994 	} else if (error != ENOENT)
1995 		goto done_unlock_list;
1996 
1997 	vfs_opterror(opts, "no jail specified");
1998 	error = ENOENT;
1999 	goto done_unlock_list;
2000 
2001  found_prison:
2002 	/* Get the parameters of the prison. */
2003 	pr->pr_ref++;
2004 	locked = PD_LOCKED;
2005 	td->td_retval[0] = pr->pr_id;
2006 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2007 	if (error != 0 && error != ENOENT)
2008 		goto done_deref;
2009 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2010 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2011 	if (error != 0 && error != ENOENT)
2012 		goto done_deref;
2013 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2014 	if (error != 0 && error != ENOENT)
2015 		goto done_deref;
2016 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2017 	    sizeof(pr->pr_cpuset->cs_id));
2018 	if (error != 0 && error != ENOENT)
2019 		goto done_deref;
2020 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2021 	if (error != 0 && error != ENOENT)
2022 		goto done_deref;
2023 #ifdef INET
2024 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2025 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2026 	if (error != 0 && error != ENOENT)
2027 		goto done_deref;
2028 #endif
2029 #ifdef INET6
2030 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2031 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2032 	if (error != 0 && error != ENOENT)
2033 		goto done_deref;
2034 #endif
2035 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2036 	    sizeof(pr->pr_securelevel));
2037 	if (error != 0 && error != ENOENT)
2038 		goto done_deref;
2039 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2040 	    sizeof(pr->pr_childcount));
2041 	if (error != 0 && error != ENOENT)
2042 		goto done_deref;
2043 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2044 	    sizeof(pr->pr_childmax));
2045 	if (error != 0 && error != ENOENT)
2046 		goto done_deref;
2047 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2048 	if (error != 0 && error != ENOENT)
2049 		goto done_deref;
2050 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2051 	if (error != 0 && error != ENOENT)
2052 		goto done_deref;
2053 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2054 	if (error != 0 && error != ENOENT)
2055 		goto done_deref;
2056 #ifdef COMPAT_FREEBSD32
2057 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2058 		uint32_t hid32 = pr->pr_hostid;
2059 
2060 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2061 	} else
2062 #endif
2063 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2064 	    sizeof(pr->pr_hostid));
2065 	if (error != 0 && error != ENOENT)
2066 		goto done_deref;
2067 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2068 	    sizeof(pr->pr_enforce_statfs));
2069 	if (error != 0 && error != ENOENT)
2070 		goto done_deref;
2071 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2072 	    sizeof(pr->pr_devfs_rsnum));
2073 	if (error != 0 && error != ENOENT)
2074 		goto done_deref;
2075 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2076 	    fi++) {
2077 		if (pr_flag_names[fi] == NULL)
2078 			continue;
2079 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2080 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2081 		if (error != 0 && error != ENOENT)
2082 			goto done_deref;
2083 		i = !i;
2084 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2085 		if (error != 0 && error != ENOENT)
2086 			goto done_deref;
2087 	}
2088 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2089 	    fi++) {
2090 		i = pr->pr_flags &
2091 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2092 		i = pr_flag_jailsys[fi].disable &&
2093 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2094 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2095 		    : JAIL_SYS_INHERIT;
2096 		error =
2097 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2098 		if (error != 0 && error != ENOENT)
2099 			goto done_deref;
2100 	}
2101 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2102 	    fi++) {
2103 		if (pr_allow_names[fi] == NULL)
2104 			continue;
2105 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2106 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2107 		if (error != 0 && error != ENOENT)
2108 			goto done_deref;
2109 		i = !i;
2110 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2111 		if (error != 0 && error != ENOENT)
2112 			goto done_deref;
2113 	}
2114 	i = (pr->pr_uref == 0);
2115 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2116 	if (error != 0 && error != ENOENT)
2117 		goto done_deref;
2118 	i = !i;
2119 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2120 	if (error != 0 && error != ENOENT)
2121 		goto done_deref;
2122 
2123 	/* Get the module parameters. */
2124 	mtx_unlock(&pr->pr_mtx);
2125 	locked = 0;
2126 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2127 	if (error)
2128 		goto done_deref;
2129 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2130 
2131 	/* By now, all parameters should have been noted. */
2132 	TAILQ_FOREACH(opt, opts, link) {
2133 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2134 			error = EINVAL;
2135 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2136 			goto done_errmsg;
2137 		}
2138 	}
2139 
2140 	/* Write the fetched parameters back to userspace. */
2141 	error = 0;
2142 	TAILQ_FOREACH(opt, opts, link) {
2143 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2144 			pos = 2 * opt->pos + 1;
2145 			optuio->uio_iov[pos].iov_len = opt->len;
2146 			if (opt->value != NULL) {
2147 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2148 					bcopy(opt->value,
2149 					    optuio->uio_iov[pos].iov_base,
2150 					    opt->len);
2151 				} else {
2152 					error = copyout(opt->value,
2153 					    optuio->uio_iov[pos].iov_base,
2154 					    opt->len);
2155 					if (error)
2156 						break;
2157 				}
2158 			}
2159 		}
2160 	}
2161 	goto done_errmsg;
2162 
2163  done_deref:
2164 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2165 	goto done_errmsg;
2166 
2167  done_unlock_list:
2168 	sx_sunlock(&allprison_lock);
2169  done_errmsg:
2170 	if (error && errmsg_pos >= 0) {
2171 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2172 		errmsg_pos = 2 * errmsg_pos + 1;
2173 		if (errmsg_len > 0) {
2174 			if (optuio->uio_segflg == UIO_SYSSPACE)
2175 				bcopy(errmsg,
2176 				    optuio->uio_iov[errmsg_pos].iov_base,
2177 				    errmsg_len);
2178 			else
2179 				copyout(errmsg,
2180 				    optuio->uio_iov[errmsg_pos].iov_base,
2181 				    errmsg_len);
2182 		}
2183 	}
2184 	vfs_freeopts(opts);
2185 	return (error);
2186 }
2187 
2188 
2189 /*
2190  * struct jail_remove_args {
2191  *	int jid;
2192  * };
2193  */
2194 int
2195 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2196 {
2197 	struct prison *pr, *cpr, *lpr, *tpr;
2198 	int descend, error;
2199 
2200 	error = priv_check(td, PRIV_JAIL_REMOVE);
2201 	if (error)
2202 		return (error);
2203 
2204 	sx_xlock(&allprison_lock);
2205 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2206 	if (pr == NULL) {
2207 		sx_xunlock(&allprison_lock);
2208 		return (EINVAL);
2209 	}
2210 
2211 	/* Remove all descendants of this prison, then remove this prison. */
2212 	pr->pr_ref++;
2213 	pr->pr_flags |= PR_REMOVE;
2214 	if (!LIST_EMPTY(&pr->pr_children)) {
2215 		mtx_unlock(&pr->pr_mtx);
2216 		lpr = NULL;
2217 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2218 			mtx_lock(&cpr->pr_mtx);
2219 			if (cpr->pr_ref > 0) {
2220 				tpr = cpr;
2221 				cpr->pr_ref++;
2222 				cpr->pr_flags |= PR_REMOVE;
2223 			} else {
2224 				/* Already removed - do not do it again. */
2225 				tpr = NULL;
2226 			}
2227 			mtx_unlock(&cpr->pr_mtx);
2228 			if (lpr != NULL) {
2229 				mtx_lock(&lpr->pr_mtx);
2230 				prison_remove_one(lpr);
2231 				sx_xlock(&allprison_lock);
2232 			}
2233 			lpr = tpr;
2234 		}
2235 		if (lpr != NULL) {
2236 			mtx_lock(&lpr->pr_mtx);
2237 			prison_remove_one(lpr);
2238 			sx_xlock(&allprison_lock);
2239 		}
2240 		mtx_lock(&pr->pr_mtx);
2241 	}
2242 	prison_remove_one(pr);
2243 	return (0);
2244 }
2245 
2246 static void
2247 prison_remove_one(struct prison *pr)
2248 {
2249 	struct proc *p;
2250 	int deuref;
2251 
2252 	/* If the prison was persistent, it is not anymore. */
2253 	deuref = 0;
2254 	if (pr->pr_flags & PR_PERSIST) {
2255 		pr->pr_ref--;
2256 		deuref = PD_DEUREF;
2257 		pr->pr_flags &= ~PR_PERSIST;
2258 	}
2259 
2260 	/*
2261 	 * jail_remove added a reference.  If that's the only one, remove
2262 	 * the prison now.
2263 	 */
2264 	KASSERT(pr->pr_ref > 0,
2265 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2266 	if (pr->pr_ref == 1) {
2267 		prison_deref(pr,
2268 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2269 		return;
2270 	}
2271 
2272 	mtx_unlock(&pr->pr_mtx);
2273 	sx_xunlock(&allprison_lock);
2274 	/*
2275 	 * Kill all processes unfortunate enough to be attached to this prison.
2276 	 */
2277 	sx_slock(&allproc_lock);
2278 	LIST_FOREACH(p, &allproc, p_list) {
2279 		PROC_LOCK(p);
2280 		if (p->p_state != PRS_NEW && p->p_ucred &&
2281 		    p->p_ucred->cr_prison == pr)
2282 			kern_psignal(p, SIGKILL);
2283 		PROC_UNLOCK(p);
2284 	}
2285 	sx_sunlock(&allproc_lock);
2286 	/* Remove the temporary reference added by jail_remove. */
2287 	prison_deref(pr, deuref | PD_DEREF);
2288 }
2289 
2290 
2291 /*
2292  * struct jail_attach_args {
2293  *	int jid;
2294  * };
2295  */
2296 int
2297 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2298 {
2299 	struct prison *pr;
2300 	int error;
2301 
2302 	error = priv_check(td, PRIV_JAIL_ATTACH);
2303 	if (error)
2304 		return (error);
2305 
2306 	sx_slock(&allprison_lock);
2307 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2308 	if (pr == NULL) {
2309 		sx_sunlock(&allprison_lock);
2310 		return (EINVAL);
2311 	}
2312 
2313 	/*
2314 	 * Do not allow a process to attach to a prison that is not
2315 	 * considered to be "alive".
2316 	 */
2317 	if (pr->pr_uref == 0) {
2318 		mtx_unlock(&pr->pr_mtx);
2319 		sx_sunlock(&allprison_lock);
2320 		return (EINVAL);
2321 	}
2322 
2323 	return (do_jail_attach(td, pr));
2324 }
2325 
2326 static int
2327 do_jail_attach(struct thread *td, struct prison *pr)
2328 {
2329 	struct prison *ppr;
2330 	struct proc *p;
2331 	struct ucred *newcred, *oldcred;
2332 	int error;
2333 
2334 	/*
2335 	 * XXX: Note that there is a slight race here if two threads
2336 	 * in the same privileged process attempt to attach to two
2337 	 * different jails at the same time.  It is important for
2338 	 * user processes not to do this, or they might end up with
2339 	 * a process root from one prison, but attached to the jail
2340 	 * of another.
2341 	 */
2342 	pr->pr_ref++;
2343 	pr->pr_uref++;
2344 	mtx_unlock(&pr->pr_mtx);
2345 
2346 	/* Let modules do whatever they need to prepare for attaching. */
2347 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2348 	if (error) {
2349 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2350 		return (error);
2351 	}
2352 	sx_sunlock(&allprison_lock);
2353 
2354 	/*
2355 	 * Reparent the newly attached process to this jail.
2356 	 */
2357 	ppr = td->td_ucred->cr_prison;
2358 	p = td->td_proc;
2359 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2360 	if (error)
2361 		goto e_revert_osd;
2362 
2363 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2364 	if ((error = change_dir(pr->pr_root, td)) != 0)
2365 		goto e_unlock;
2366 #ifdef MAC
2367 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2368 		goto e_unlock;
2369 #endif
2370 	VOP_UNLOCK(pr->pr_root, 0);
2371 	if ((error = change_root(pr->pr_root, td)))
2372 		goto e_revert_osd;
2373 
2374 	newcred = crget();
2375 	PROC_LOCK(p);
2376 	oldcred = p->p_ucred;
2377 	setsugid(p);
2378 	crcopy(newcred, oldcred);
2379 	newcred->cr_prison = pr;
2380 	p->p_ucred = newcred;
2381 	PROC_UNLOCK(p);
2382 #ifdef RACCT
2383 	racct_proc_ucred_changed(p, oldcred, newcred);
2384 #endif
2385 	crfree(oldcred);
2386 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2387 	return (0);
2388  e_unlock:
2389 	VOP_UNLOCK(pr->pr_root, 0);
2390  e_revert_osd:
2391 	/* Tell modules this thread is still in its old jail after all. */
2392 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2393 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2394 	return (error);
2395 }
2396 
2397 
2398 /*
2399  * Returns a locked prison instance, or NULL on failure.
2400  */
2401 struct prison *
2402 prison_find(int prid)
2403 {
2404 	struct prison *pr;
2405 
2406 	sx_assert(&allprison_lock, SX_LOCKED);
2407 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2408 		if (pr->pr_id == prid) {
2409 			mtx_lock(&pr->pr_mtx);
2410 			if (pr->pr_ref > 0)
2411 				return (pr);
2412 			mtx_unlock(&pr->pr_mtx);
2413 		}
2414 	}
2415 	return (NULL);
2416 }
2417 
2418 /*
2419  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2420  */
2421 struct prison *
2422 prison_find_child(struct prison *mypr, int prid)
2423 {
2424 	struct prison *pr;
2425 	int descend;
2426 
2427 	sx_assert(&allprison_lock, SX_LOCKED);
2428 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2429 		if (pr->pr_id == prid) {
2430 			mtx_lock(&pr->pr_mtx);
2431 			if (pr->pr_ref > 0)
2432 				return (pr);
2433 			mtx_unlock(&pr->pr_mtx);
2434 		}
2435 	}
2436 	return (NULL);
2437 }
2438 
2439 /*
2440  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2441  */
2442 struct prison *
2443 prison_find_name(struct prison *mypr, const char *name)
2444 {
2445 	struct prison *pr, *deadpr;
2446 	size_t mylen;
2447 	int descend;
2448 
2449 	sx_assert(&allprison_lock, SX_LOCKED);
2450 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2451  again:
2452 	deadpr = NULL;
2453 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2454 		if (!strcmp(pr->pr_name + mylen, name)) {
2455 			mtx_lock(&pr->pr_mtx);
2456 			if (pr->pr_ref > 0) {
2457 				if (pr->pr_uref > 0)
2458 					return (pr);
2459 				deadpr = pr;
2460 			}
2461 			mtx_unlock(&pr->pr_mtx);
2462 		}
2463 	}
2464 	/* There was no valid prison - perhaps there was a dying one. */
2465 	if (deadpr != NULL) {
2466 		mtx_lock(&deadpr->pr_mtx);
2467 		if (deadpr->pr_ref == 0) {
2468 			mtx_unlock(&deadpr->pr_mtx);
2469 			goto again;
2470 		}
2471 	}
2472 	return (deadpr);
2473 }
2474 
2475 /*
2476  * See if a prison has the specific flag set.
2477  */
2478 int
2479 prison_flag(struct ucred *cred, unsigned flag)
2480 {
2481 
2482 	/* This is an atomic read, so no locking is necessary. */
2483 	return (cred->cr_prison->pr_flags & flag);
2484 }
2485 
2486 int
2487 prison_allow(struct ucred *cred, unsigned flag)
2488 {
2489 
2490 	/* This is an atomic read, so no locking is necessary. */
2491 	return (cred->cr_prison->pr_allow & flag);
2492 }
2493 
2494 /*
2495  * Remove a prison reference.  If that was the last reference, remove the
2496  * prison itself - but not in this context in case there are locks held.
2497  */
2498 void
2499 prison_free_locked(struct prison *pr)
2500 {
2501 
2502 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2503 	pr->pr_ref--;
2504 	if (pr->pr_ref == 0) {
2505 		mtx_unlock(&pr->pr_mtx);
2506 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2507 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2508 		return;
2509 	}
2510 	mtx_unlock(&pr->pr_mtx);
2511 }
2512 
2513 void
2514 prison_free(struct prison *pr)
2515 {
2516 
2517 	mtx_lock(&pr->pr_mtx);
2518 	prison_free_locked(pr);
2519 }
2520 
2521 static void
2522 prison_complete(void *context, int pending)
2523 {
2524 
2525 	prison_deref((struct prison *)context, 0);
2526 }
2527 
2528 /*
2529  * Remove a prison reference (usually).  This internal version assumes no
2530  * mutexes are held, except perhaps the prison itself.  If there are no more
2531  * references, release and delist the prison.  On completion, the prison lock
2532  * and the allprison lock are both unlocked.
2533  */
2534 static void
2535 prison_deref(struct prison *pr, int flags)
2536 {
2537 	struct prison *ppr, *tpr;
2538 
2539 	if (!(flags & PD_LOCKED))
2540 		mtx_lock(&pr->pr_mtx);
2541 	for (;;) {
2542 		if (flags & PD_DEUREF) {
2543 			pr->pr_uref--;
2544 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2545 		}
2546 		if (flags & PD_DEREF)
2547 			pr->pr_ref--;
2548 		/* If the prison still has references, nothing else to do. */
2549 		if (pr->pr_ref > 0) {
2550 			mtx_unlock(&pr->pr_mtx);
2551 			if (flags & PD_LIST_SLOCKED)
2552 				sx_sunlock(&allprison_lock);
2553 			else if (flags & PD_LIST_XLOCKED)
2554 				sx_xunlock(&allprison_lock);
2555 			return;
2556 		}
2557 
2558 		mtx_unlock(&pr->pr_mtx);
2559 		if (flags & PD_LIST_SLOCKED) {
2560 			if (!sx_try_upgrade(&allprison_lock)) {
2561 				sx_sunlock(&allprison_lock);
2562 				sx_xlock(&allprison_lock);
2563 			}
2564 		} else if (!(flags & PD_LIST_XLOCKED))
2565 			sx_xlock(&allprison_lock);
2566 
2567 		TAILQ_REMOVE(&allprison, pr, pr_list);
2568 		LIST_REMOVE(pr, pr_sibling);
2569 		ppr = pr->pr_parent;
2570 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2571 			tpr->pr_childcount--;
2572 		sx_xunlock(&allprison_lock);
2573 
2574 #ifdef VIMAGE
2575 		if (pr->pr_vnet != ppr->pr_vnet)
2576 			vnet_destroy(pr->pr_vnet);
2577 #endif
2578 		if (pr->pr_root != NULL)
2579 			vrele(pr->pr_root);
2580 		mtx_destroy(&pr->pr_mtx);
2581 #ifdef INET
2582 		free(pr->pr_ip4, M_PRISON);
2583 #endif
2584 #ifdef INET6
2585 		free(pr->pr_ip6, M_PRISON);
2586 #endif
2587 		if (pr->pr_cpuset != NULL)
2588 			cpuset_rel(pr->pr_cpuset);
2589 		osd_jail_exit(pr);
2590 #ifdef RACCT
2591 		prison_racct_detach(pr);
2592 #endif
2593 		free(pr, M_PRISON);
2594 
2595 		/* Removing a prison frees a reference on its parent. */
2596 		pr = ppr;
2597 		mtx_lock(&pr->pr_mtx);
2598 		flags = PD_DEREF | PD_DEUREF;
2599 	}
2600 }
2601 
2602 void
2603 prison_hold_locked(struct prison *pr)
2604 {
2605 
2606 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2607 	KASSERT(pr->pr_ref > 0,
2608 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2609 	pr->pr_ref++;
2610 }
2611 
2612 void
2613 prison_hold(struct prison *pr)
2614 {
2615 
2616 	mtx_lock(&pr->pr_mtx);
2617 	prison_hold_locked(pr);
2618 	mtx_unlock(&pr->pr_mtx);
2619 }
2620 
2621 void
2622 prison_proc_hold(struct prison *pr)
2623 {
2624 
2625 	mtx_lock(&pr->pr_mtx);
2626 	KASSERT(pr->pr_uref > 0,
2627 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2628 	pr->pr_uref++;
2629 	mtx_unlock(&pr->pr_mtx);
2630 }
2631 
2632 void
2633 prison_proc_free(struct prison *pr)
2634 {
2635 
2636 	mtx_lock(&pr->pr_mtx);
2637 	KASSERT(pr->pr_uref > 0,
2638 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2639 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2640 }
2641 
2642 
2643 #ifdef INET
2644 /*
2645  * Restrict a prison's IP address list with its parent's, possibly replacing
2646  * it.  Return true if the replacement buffer was used (or would have been).
2647  */
2648 static int
2649 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2650 {
2651 	int ii, ij, used;
2652 	struct prison *ppr;
2653 
2654 	ppr = pr->pr_parent;
2655 	if (!(pr->pr_flags & PR_IP4_USER)) {
2656 		/* This has no user settings, so just copy the parent's list. */
2657 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2658 			/*
2659 			 * There's no room for the parent's list.  Use the
2660 			 * new list buffer, which is assumed to be big enough
2661 			 * (if it was passed).  If there's no buffer, try to
2662 			 * allocate one.
2663 			 */
2664 			used = 1;
2665 			if (newip4 == NULL) {
2666 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2667 				    M_PRISON, M_NOWAIT);
2668 				if (newip4 != NULL)
2669 					used = 0;
2670 			}
2671 			if (newip4 != NULL) {
2672 				bcopy(ppr->pr_ip4, newip4,
2673 				    ppr->pr_ip4s * sizeof(*newip4));
2674 				free(pr->pr_ip4, M_PRISON);
2675 				pr->pr_ip4 = newip4;
2676 				pr->pr_ip4s = ppr->pr_ip4s;
2677 			}
2678 			return (used);
2679 		}
2680 		pr->pr_ip4s = ppr->pr_ip4s;
2681 		if (pr->pr_ip4s > 0)
2682 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2683 			    pr->pr_ip4s * sizeof(*newip4));
2684 		else if (pr->pr_ip4 != NULL) {
2685 			free(pr->pr_ip4, M_PRISON);
2686 			pr->pr_ip4 = NULL;
2687 		}
2688 	} else if (pr->pr_ip4s > 0) {
2689 		/* Remove addresses that aren't in the parent. */
2690 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2691 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2692 				break;
2693 		if (ij < ppr->pr_ip4s)
2694 			ii = 1;
2695 		else {
2696 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2697 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2698 			ii = 0;
2699 		}
2700 		for (ij = 1; ii < pr->pr_ip4s; ) {
2701 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2702 				ii++;
2703 				continue;
2704 			}
2705 			switch (ij >= ppr->pr_ip4s ? -1 :
2706 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2707 			case -1:
2708 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2709 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2710 				break;
2711 			case 0:
2712 				ii++;
2713 				ij++;
2714 				break;
2715 			case 1:
2716 				ij++;
2717 				break;
2718 			}
2719 		}
2720 		if (pr->pr_ip4s == 0) {
2721 			free(pr->pr_ip4, M_PRISON);
2722 			pr->pr_ip4 = NULL;
2723 		}
2724 	}
2725 	return (0);
2726 }
2727 
2728 /*
2729  * Pass back primary IPv4 address of this jail.
2730  *
2731  * If not restricted return success but do not alter the address.  Caller has
2732  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2733  *
2734  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2735  * Address returned in NBO.
2736  */
2737 int
2738 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2739 {
2740 	struct prison *pr;
2741 
2742 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2743 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2744 
2745 	pr = cred->cr_prison;
2746 	if (!(pr->pr_flags & PR_IP4))
2747 		return (0);
2748 	mtx_lock(&pr->pr_mtx);
2749 	if (!(pr->pr_flags & PR_IP4)) {
2750 		mtx_unlock(&pr->pr_mtx);
2751 		return (0);
2752 	}
2753 	if (pr->pr_ip4 == NULL) {
2754 		mtx_unlock(&pr->pr_mtx);
2755 		return (EAFNOSUPPORT);
2756 	}
2757 
2758 	ia->s_addr = pr->pr_ip4[0].s_addr;
2759 	mtx_unlock(&pr->pr_mtx);
2760 	return (0);
2761 }
2762 
2763 /*
2764  * Return 1 if we should do proper source address selection or are not jailed.
2765  * We will return 0 if we should bypass source address selection in favour
2766  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2767  * returned in NBO.
2768  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2769  */
2770 int
2771 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2772 {
2773 	struct prison *pr;
2774 	struct in_addr lia;
2775 	int error;
2776 
2777 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2778 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2779 
2780 	if (!jailed(cred))
2781 		return (1);
2782 
2783 	pr = cred->cr_prison;
2784 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2785 		return (1);
2786 
2787 	lia.s_addr = INADDR_ANY;
2788 	error = prison_get_ip4(cred, &lia);
2789 	if (error)
2790 		return (error);
2791 	if (lia.s_addr == INADDR_ANY)
2792 		return (1);
2793 
2794 	ia->s_addr = lia.s_addr;
2795 	return (0);
2796 }
2797 
2798 /*
2799  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2800  */
2801 int
2802 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2803 {
2804 
2805 	if (pr1 == pr2)
2806 		return (1);
2807 
2808 	/*
2809 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2810 	 * existing prisons.
2811 	 */
2812 	while (pr1 != &prison0 &&
2813 #ifdef VIMAGE
2814 	       !(pr1->pr_flags & PR_VNET) &&
2815 #endif
2816 	       !(pr1->pr_flags & PR_IP4_USER))
2817 		pr1 = pr1->pr_parent;
2818 	while (pr2 != &prison0 &&
2819 #ifdef VIMAGE
2820 	       !(pr2->pr_flags & PR_VNET) &&
2821 #endif
2822 	       !(pr2->pr_flags & PR_IP4_USER))
2823 		pr2 = pr2->pr_parent;
2824 	return (pr1 == pr2);
2825 }
2826 
2827 /*
2828  * Make sure our (source) address is set to something meaningful to this
2829  * jail.
2830  *
2831  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2832  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2833  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2834  */
2835 int
2836 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2837 {
2838 	struct prison *pr;
2839 	struct in_addr ia0;
2840 	int error;
2841 
2842 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2843 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2844 
2845 	pr = cred->cr_prison;
2846 	if (!(pr->pr_flags & PR_IP4))
2847 		return (0);
2848 	mtx_lock(&pr->pr_mtx);
2849 	if (!(pr->pr_flags & PR_IP4)) {
2850 		mtx_unlock(&pr->pr_mtx);
2851 		return (0);
2852 	}
2853 	if (pr->pr_ip4 == NULL) {
2854 		mtx_unlock(&pr->pr_mtx);
2855 		return (EAFNOSUPPORT);
2856 	}
2857 
2858 	ia0.s_addr = ntohl(ia->s_addr);
2859 	if (ia0.s_addr == INADDR_LOOPBACK) {
2860 		ia->s_addr = pr->pr_ip4[0].s_addr;
2861 		mtx_unlock(&pr->pr_mtx);
2862 		return (0);
2863 	}
2864 
2865 	if (ia0.s_addr == INADDR_ANY) {
2866 		/*
2867 		 * In case there is only 1 IPv4 address, bind directly.
2868 		 */
2869 		if (pr->pr_ip4s == 1)
2870 			ia->s_addr = pr->pr_ip4[0].s_addr;
2871 		mtx_unlock(&pr->pr_mtx);
2872 		return (0);
2873 	}
2874 
2875 	error = _prison_check_ip4(pr, ia);
2876 	mtx_unlock(&pr->pr_mtx);
2877 	return (error);
2878 }
2879 
2880 /*
2881  * Rewrite destination address in case we will connect to loopback address.
2882  *
2883  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2884  * Address passed in in NBO and returned in NBO.
2885  */
2886 int
2887 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2888 {
2889 	struct prison *pr;
2890 
2891 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2892 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2893 
2894 	pr = cred->cr_prison;
2895 	if (!(pr->pr_flags & PR_IP4))
2896 		return (0);
2897 	mtx_lock(&pr->pr_mtx);
2898 	if (!(pr->pr_flags & PR_IP4)) {
2899 		mtx_unlock(&pr->pr_mtx);
2900 		return (0);
2901 	}
2902 	if (pr->pr_ip4 == NULL) {
2903 		mtx_unlock(&pr->pr_mtx);
2904 		return (EAFNOSUPPORT);
2905 	}
2906 
2907 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2908 		ia->s_addr = pr->pr_ip4[0].s_addr;
2909 		mtx_unlock(&pr->pr_mtx);
2910 		return (0);
2911 	}
2912 
2913 	/*
2914 	 * Return success because nothing had to be changed.
2915 	 */
2916 	mtx_unlock(&pr->pr_mtx);
2917 	return (0);
2918 }
2919 
2920 /*
2921  * Check if given address belongs to the jail referenced by cred/prison.
2922  *
2923  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2924  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2925  * doesn't allow IPv4.  Address passed in in NBO.
2926  */
2927 static int
2928 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
2929 {
2930 	int i, a, z, d;
2931 
2932 	/*
2933 	 * Check the primary IP.
2934 	 */
2935 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2936 		return (0);
2937 
2938 	/*
2939 	 * All the other IPs are sorted so we can do a binary search.
2940 	 */
2941 	a = 0;
2942 	z = pr->pr_ip4s - 2;
2943 	while (a <= z) {
2944 		i = (a + z) / 2;
2945 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2946 		if (d > 0)
2947 			z = i - 1;
2948 		else if (d < 0)
2949 			a = i + 1;
2950 		else
2951 			return (0);
2952 	}
2953 
2954 	return (EADDRNOTAVAIL);
2955 }
2956 
2957 int
2958 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
2959 {
2960 	struct prison *pr;
2961 	int error;
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 	error = _prison_check_ip4(pr, ia);
2980 	mtx_unlock(&pr->pr_mtx);
2981 	return (error);
2982 }
2983 #endif
2984 
2985 #ifdef INET6
2986 static int
2987 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2988 {
2989 	int ii, ij, used;
2990 	struct prison *ppr;
2991 
2992 	ppr = pr->pr_parent;
2993 	if (!(pr->pr_flags & PR_IP6_USER)) {
2994 		/* This has no user settings, so just copy the parent's list. */
2995 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2996 			/*
2997 			 * There's no room for the parent's list.  Use the
2998 			 * new list buffer, which is assumed to be big enough
2999 			 * (if it was passed).  If there's no buffer, try to
3000 			 * allocate one.
3001 			 */
3002 			used = 1;
3003 			if (newip6 == NULL) {
3004 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3005 				    M_PRISON, M_NOWAIT);
3006 				if (newip6 != NULL)
3007 					used = 0;
3008 			}
3009 			if (newip6 != NULL) {
3010 				bcopy(ppr->pr_ip6, newip6,
3011 				    ppr->pr_ip6s * sizeof(*newip6));
3012 				free(pr->pr_ip6, M_PRISON);
3013 				pr->pr_ip6 = newip6;
3014 				pr->pr_ip6s = ppr->pr_ip6s;
3015 			}
3016 			return (used);
3017 		}
3018 		pr->pr_ip6s = ppr->pr_ip6s;
3019 		if (pr->pr_ip6s > 0)
3020 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3021 			    pr->pr_ip6s * sizeof(*newip6));
3022 		else if (pr->pr_ip6 != NULL) {
3023 			free(pr->pr_ip6, M_PRISON);
3024 			pr->pr_ip6 = NULL;
3025 		}
3026 	} else if (pr->pr_ip6s > 0) {
3027 		/* Remove addresses that aren't in the parent. */
3028 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3029 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3030 			    &ppr->pr_ip6[ij]))
3031 				break;
3032 		if (ij < ppr->pr_ip6s)
3033 			ii = 1;
3034 		else {
3035 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3036 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3037 			ii = 0;
3038 		}
3039 		for (ij = 1; ii < pr->pr_ip6s; ) {
3040 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3041 			    &ppr->pr_ip6[0])) {
3042 				ii++;
3043 				continue;
3044 			}
3045 			switch (ij >= ppr->pr_ip6s ? -1 :
3046 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3047 			case -1:
3048 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3049 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3050 				break;
3051 			case 0:
3052 				ii++;
3053 				ij++;
3054 				break;
3055 			case 1:
3056 				ij++;
3057 				break;
3058 			}
3059 		}
3060 		if (pr->pr_ip6s == 0) {
3061 			free(pr->pr_ip6, M_PRISON);
3062 			pr->pr_ip6 = NULL;
3063 		}
3064 	}
3065 	return 0;
3066 }
3067 
3068 /*
3069  * Pass back primary IPv6 address for this jail.
3070  *
3071  * If not restricted return success but do not alter the address.  Caller has
3072  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3073  *
3074  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3075  */
3076 int
3077 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3078 {
3079 	struct prison *pr;
3080 
3081 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3082 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3083 
3084 	pr = cred->cr_prison;
3085 	if (!(pr->pr_flags & PR_IP6))
3086 		return (0);
3087 	mtx_lock(&pr->pr_mtx);
3088 	if (!(pr->pr_flags & PR_IP6)) {
3089 		mtx_unlock(&pr->pr_mtx);
3090 		return (0);
3091 	}
3092 	if (pr->pr_ip6 == NULL) {
3093 		mtx_unlock(&pr->pr_mtx);
3094 		return (EAFNOSUPPORT);
3095 	}
3096 
3097 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3098 	mtx_unlock(&pr->pr_mtx);
3099 	return (0);
3100 }
3101 
3102 /*
3103  * Return 1 if we should do proper source address selection or are not jailed.
3104  * We will return 0 if we should bypass source address selection in favour
3105  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3106  * returned in NBO.
3107  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3108  */
3109 int
3110 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3111 {
3112 	struct prison *pr;
3113 	struct in6_addr lia6;
3114 	int error;
3115 
3116 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3117 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3118 
3119 	if (!jailed(cred))
3120 		return (1);
3121 
3122 	pr = cred->cr_prison;
3123 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3124 		return (1);
3125 
3126 	lia6 = in6addr_any;
3127 	error = prison_get_ip6(cred, &lia6);
3128 	if (error)
3129 		return (error);
3130 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3131 		return (1);
3132 
3133 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3134 	return (0);
3135 }
3136 
3137 /*
3138  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3139  */
3140 int
3141 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3142 {
3143 
3144 	if (pr1 == pr2)
3145 		return (1);
3146 
3147 	while (pr1 != &prison0 &&
3148 #ifdef VIMAGE
3149 	       !(pr1->pr_flags & PR_VNET) &&
3150 #endif
3151 	       !(pr1->pr_flags & PR_IP6_USER))
3152 		pr1 = pr1->pr_parent;
3153 	while (pr2 != &prison0 &&
3154 #ifdef VIMAGE
3155 	       !(pr2->pr_flags & PR_VNET) &&
3156 #endif
3157 	       !(pr2->pr_flags & PR_IP6_USER))
3158 		pr2 = pr2->pr_parent;
3159 	return (pr1 == pr2);
3160 }
3161 
3162 /*
3163  * Make sure our (source) address is set to something meaningful to this jail.
3164  *
3165  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3166  * when needed while binding.
3167  *
3168  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3169  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3170  * doesn't allow IPv6.
3171  */
3172 int
3173 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3174 {
3175 	struct prison *pr;
3176 	int error;
3177 
3178 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3179 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3180 
3181 	pr = cred->cr_prison;
3182 	if (!(pr->pr_flags & PR_IP6))
3183 		return (0);
3184 	mtx_lock(&pr->pr_mtx);
3185 	if (!(pr->pr_flags & PR_IP6)) {
3186 		mtx_unlock(&pr->pr_mtx);
3187 		return (0);
3188 	}
3189 	if (pr->pr_ip6 == NULL) {
3190 		mtx_unlock(&pr->pr_mtx);
3191 		return (EAFNOSUPPORT);
3192 	}
3193 
3194 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3195 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3196 		mtx_unlock(&pr->pr_mtx);
3197 		return (0);
3198 	}
3199 
3200 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3201 		/*
3202 		 * In case there is only 1 IPv6 address, and v6only is true,
3203 		 * then bind directly.
3204 		 */
3205 		if (v6only != 0 && pr->pr_ip6s == 1)
3206 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3207 		mtx_unlock(&pr->pr_mtx);
3208 		return (0);
3209 	}
3210 
3211 	error = _prison_check_ip6(pr, ia6);
3212 	mtx_unlock(&pr->pr_mtx);
3213 	return (error);
3214 }
3215 
3216 /*
3217  * Rewrite destination address in case we will connect to loopback address.
3218  *
3219  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3220  */
3221 int
3222 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3223 {
3224 	struct prison *pr;
3225 
3226 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3227 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3228 
3229 	pr = cred->cr_prison;
3230 	if (!(pr->pr_flags & PR_IP6))
3231 		return (0);
3232 	mtx_lock(&pr->pr_mtx);
3233 	if (!(pr->pr_flags & PR_IP6)) {
3234 		mtx_unlock(&pr->pr_mtx);
3235 		return (0);
3236 	}
3237 	if (pr->pr_ip6 == NULL) {
3238 		mtx_unlock(&pr->pr_mtx);
3239 		return (EAFNOSUPPORT);
3240 	}
3241 
3242 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3243 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3244 		mtx_unlock(&pr->pr_mtx);
3245 		return (0);
3246 	}
3247 
3248 	/*
3249 	 * Return success because nothing had to be changed.
3250 	 */
3251 	mtx_unlock(&pr->pr_mtx);
3252 	return (0);
3253 }
3254 
3255 /*
3256  * Check if given address belongs to the jail referenced by cred/prison.
3257  *
3258  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3259  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3260  * doesn't allow IPv6.
3261  */
3262 static int
3263 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3264 {
3265 	int i, a, z, d;
3266 
3267 	/*
3268 	 * Check the primary IP.
3269 	 */
3270 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3271 		return (0);
3272 
3273 	/*
3274 	 * All the other IPs are sorted so we can do a binary search.
3275 	 */
3276 	a = 0;
3277 	z = pr->pr_ip6s - 2;
3278 	while (a <= z) {
3279 		i = (a + z) / 2;
3280 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3281 		if (d > 0)
3282 			z = i - 1;
3283 		else if (d < 0)
3284 			a = i + 1;
3285 		else
3286 			return (0);
3287 	}
3288 
3289 	return (EADDRNOTAVAIL);
3290 }
3291 
3292 int
3293 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3294 {
3295 	struct prison *pr;
3296 	int error;
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 	error = _prison_check_ip6(pr, ia6);
3315 	mtx_unlock(&pr->pr_mtx);
3316 	return (error);
3317 }
3318 #endif
3319 
3320 /*
3321  * Check if a jail supports the given address family.
3322  *
3323  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3324  * if not.
3325  */
3326 int
3327 prison_check_af(struct ucred *cred, int af)
3328 {
3329 	struct prison *pr;
3330 	int error;
3331 
3332 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3333 
3334 	pr = cred->cr_prison;
3335 #ifdef VIMAGE
3336 	/* Prisons with their own network stack are not limited. */
3337 	if (prison_owns_vnet(cred))
3338 		return (0);
3339 #endif
3340 
3341 	error = 0;
3342 	switch (af)
3343 	{
3344 #ifdef INET
3345 	case AF_INET:
3346 		if (pr->pr_flags & PR_IP4)
3347 		{
3348 			mtx_lock(&pr->pr_mtx);
3349 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3350 				error = EAFNOSUPPORT;
3351 			mtx_unlock(&pr->pr_mtx);
3352 		}
3353 		break;
3354 #endif
3355 #ifdef INET6
3356 	case AF_INET6:
3357 		if (pr->pr_flags & PR_IP6)
3358 		{
3359 			mtx_lock(&pr->pr_mtx);
3360 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3361 				error = EAFNOSUPPORT;
3362 			mtx_unlock(&pr->pr_mtx);
3363 		}
3364 		break;
3365 #endif
3366 	case AF_LOCAL:
3367 	case AF_ROUTE:
3368 		break;
3369 	default:
3370 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3371 			error = EAFNOSUPPORT;
3372 	}
3373 	return (error);
3374 }
3375 
3376 /*
3377  * Check if given address belongs to the jail referenced by cred (wrapper to
3378  * prison_check_ip[46]).
3379  *
3380  * Returns 0 if jail doesn't restrict the address family or if address belongs
3381  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3382  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3383  */
3384 int
3385 prison_if(struct ucred *cred, struct sockaddr *sa)
3386 {
3387 #ifdef INET
3388 	struct sockaddr_in *sai;
3389 #endif
3390 #ifdef INET6
3391 	struct sockaddr_in6 *sai6;
3392 #endif
3393 	int error;
3394 
3395 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3396 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3397 
3398 #ifdef VIMAGE
3399 	if (prison_owns_vnet(cred))
3400 		return (0);
3401 #endif
3402 
3403 	error = 0;
3404 	switch (sa->sa_family)
3405 	{
3406 #ifdef INET
3407 	case AF_INET:
3408 		sai = (struct sockaddr_in *)sa;
3409 		error = prison_check_ip4(cred, &sai->sin_addr);
3410 		break;
3411 #endif
3412 #ifdef INET6
3413 	case AF_INET6:
3414 		sai6 = (struct sockaddr_in6 *)sa;
3415 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3416 		break;
3417 #endif
3418 	default:
3419 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3420 			error = EAFNOSUPPORT;
3421 	}
3422 	return (error);
3423 }
3424 
3425 /*
3426  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3427  */
3428 int
3429 prison_check(struct ucred *cred1, struct ucred *cred2)
3430 {
3431 
3432 	return ((cred1->cr_prison == cred2->cr_prison ||
3433 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3434 }
3435 
3436 /*
3437  * Return 1 if p2 is a child of p1, otherwise 0.
3438  */
3439 int
3440 prison_ischild(struct prison *pr1, struct prison *pr2)
3441 {
3442 
3443 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3444 		if (pr1 == pr2)
3445 			return (1);
3446 	return (0);
3447 }
3448 
3449 /*
3450  * Return 1 if the passed credential is in a jail, otherwise 0.
3451  */
3452 int
3453 jailed(struct ucred *cred)
3454 {
3455 
3456 	return (cred->cr_prison != &prison0);
3457 }
3458 
3459 /*
3460  * Return 1 if the passed credential is in a jail and that jail does not
3461  * have its own virtual network stack, otherwise 0.
3462  */
3463 int
3464 jailed_without_vnet(struct ucred *cred)
3465 {
3466 
3467 	if (!jailed(cred))
3468 		return (0);
3469 #ifdef VIMAGE
3470 	if (prison_owns_vnet(cred))
3471 		return (0);
3472 #endif
3473 
3474 	return (1);
3475 }
3476 
3477 /*
3478  * Return the correct hostname (domainname, et al) for the passed credential.
3479  */
3480 void
3481 getcredhostname(struct ucred *cred, char *buf, size_t size)
3482 {
3483 	struct prison *pr;
3484 
3485 	/*
3486 	 * A NULL credential can be used to shortcut to the physical
3487 	 * system's hostname.
3488 	 */
3489 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3490 	mtx_lock(&pr->pr_mtx);
3491 	strlcpy(buf, pr->pr_hostname, size);
3492 	mtx_unlock(&pr->pr_mtx);
3493 }
3494 
3495 void
3496 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3497 {
3498 
3499 	mtx_lock(&cred->cr_prison->pr_mtx);
3500 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3501 	mtx_unlock(&cred->cr_prison->pr_mtx);
3502 }
3503 
3504 void
3505 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3506 {
3507 
3508 	mtx_lock(&cred->cr_prison->pr_mtx);
3509 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3510 	mtx_unlock(&cred->cr_prison->pr_mtx);
3511 }
3512 
3513 void
3514 getcredhostid(struct ucred *cred, unsigned long *hostid)
3515 {
3516 
3517 	mtx_lock(&cred->cr_prison->pr_mtx);
3518 	*hostid = cred->cr_prison->pr_hostid;
3519 	mtx_unlock(&cred->cr_prison->pr_mtx);
3520 }
3521 
3522 #ifdef VIMAGE
3523 /*
3524  * Determine whether the prison represented by cred owns
3525  * its vnet rather than having it inherited.
3526  *
3527  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3528  */
3529 int
3530 prison_owns_vnet(struct ucred *cred)
3531 {
3532 
3533 	/*
3534 	 * vnets cannot be added/removed after jail creation,
3535 	 * so no need to lock here.
3536 	 */
3537 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3538 }
3539 #endif
3540 
3541 /*
3542  * Determine whether the subject represented by cred can "see"
3543  * status of a mount point.
3544  * Returns: 0 for permitted, ENOENT otherwise.
3545  * XXX: This function should be called cr_canseemount() and should be
3546  *      placed in kern_prot.c.
3547  */
3548 int
3549 prison_canseemount(struct ucred *cred, struct mount *mp)
3550 {
3551 	struct prison *pr;
3552 	struct statfs *sp;
3553 	size_t len;
3554 
3555 	pr = cred->cr_prison;
3556 	if (pr->pr_enforce_statfs == 0)
3557 		return (0);
3558 	if (pr->pr_root->v_mount == mp)
3559 		return (0);
3560 	if (pr->pr_enforce_statfs == 2)
3561 		return (ENOENT);
3562 	/*
3563 	 * If jail's chroot directory is set to "/" we should be able to see
3564 	 * all mount-points from inside a jail.
3565 	 * This is ugly check, but this is the only situation when jail's
3566 	 * directory ends with '/'.
3567 	 */
3568 	if (strcmp(pr->pr_path, "/") == 0)
3569 		return (0);
3570 	len = strlen(pr->pr_path);
3571 	sp = &mp->mnt_stat;
3572 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3573 		return (ENOENT);
3574 	/*
3575 	 * Be sure that we don't have situation where jail's root directory
3576 	 * is "/some/path" and mount point is "/some/pathpath".
3577 	 */
3578 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3579 		return (ENOENT);
3580 	return (0);
3581 }
3582 
3583 void
3584 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3585 {
3586 	char jpath[MAXPATHLEN];
3587 	struct prison *pr;
3588 	size_t len;
3589 
3590 	pr = cred->cr_prison;
3591 	if (pr->pr_enforce_statfs == 0)
3592 		return;
3593 	if (prison_canseemount(cred, mp) != 0) {
3594 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3595 		strlcpy(sp->f_mntonname, "[restricted]",
3596 		    sizeof(sp->f_mntonname));
3597 		return;
3598 	}
3599 	if (pr->pr_root->v_mount == mp) {
3600 		/*
3601 		 * Clear current buffer data, so we are sure nothing from
3602 		 * the valid path left there.
3603 		 */
3604 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3605 		*sp->f_mntonname = '/';
3606 		return;
3607 	}
3608 	/*
3609 	 * If jail's chroot directory is set to "/" we should be able to see
3610 	 * all mount-points from inside a jail.
3611 	 */
3612 	if (strcmp(pr->pr_path, "/") == 0)
3613 		return;
3614 	len = strlen(pr->pr_path);
3615 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3616 	/*
3617 	 * Clear current buffer data, so we are sure nothing from
3618 	 * the valid path left there.
3619 	 */
3620 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3621 	if (*jpath == '\0') {
3622 		/* Should never happen. */
3623 		*sp->f_mntonname = '/';
3624 	} else {
3625 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3626 	}
3627 }
3628 
3629 /*
3630  * Check with permission for a specific privilege is granted within jail.  We
3631  * have a specific list of accepted privileges; the rest are denied.
3632  */
3633 int
3634 prison_priv_check(struct ucred *cred, int priv)
3635 {
3636 
3637 	if (!jailed(cred))
3638 		return (0);
3639 
3640 #ifdef VIMAGE
3641 	/*
3642 	 * Privileges specific to prisons with a virtual network stack.
3643 	 * There might be a duplicate entry here in case the privilege
3644 	 * is only granted conditionally in the legacy jail case.
3645 	 */
3646 	switch (priv) {
3647 #ifdef notyet
3648 		/*
3649 		 * NFS-specific privileges.
3650 		 */
3651 	case PRIV_NFS_DAEMON:
3652 	case PRIV_NFS_LOCKD:
3653 #endif
3654 		/*
3655 		 * Network stack privileges.
3656 		 */
3657 	case PRIV_NET_BRIDGE:
3658 	case PRIV_NET_GRE:
3659 	case PRIV_NET_BPF:
3660 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3661 	case PRIV_NET_ROUTE:
3662 	case PRIV_NET_TAP:
3663 	case PRIV_NET_SETIFMTU:
3664 	case PRIV_NET_SETIFFLAGS:
3665 	case PRIV_NET_SETIFCAP:
3666 	case PRIV_NET_SETIFDESCR:
3667 	case PRIV_NET_SETIFNAME	:
3668 	case PRIV_NET_SETIFMETRIC:
3669 	case PRIV_NET_SETIFPHYS:
3670 	case PRIV_NET_SETIFMAC:
3671 	case PRIV_NET_ADDMULTI:
3672 	case PRIV_NET_DELMULTI:
3673 	case PRIV_NET_HWIOCTL:
3674 	case PRIV_NET_SETLLADDR:
3675 	case PRIV_NET_ADDIFGROUP:
3676 	case PRIV_NET_DELIFGROUP:
3677 	case PRIV_NET_IFCREATE:
3678 	case PRIV_NET_IFDESTROY:
3679 	case PRIV_NET_ADDIFADDR:
3680 	case PRIV_NET_DELIFADDR:
3681 	case PRIV_NET_LAGG:
3682 	case PRIV_NET_GIF:
3683 	case PRIV_NET_SETIFVNET:
3684 	case PRIV_NET_SETIFFIB:
3685 
3686 		/*
3687 		 * 802.11-related privileges.
3688 		 */
3689 	case PRIV_NET80211_GETKEY:
3690 #ifdef notyet
3691 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3692 #endif
3693 
3694 #ifdef notyet
3695 		/*
3696 		 * ATM privileges.
3697 		 */
3698 	case PRIV_NETATM_CFG:
3699 	case PRIV_NETATM_ADD:
3700 	case PRIV_NETATM_DEL:
3701 	case PRIV_NETATM_SET:
3702 
3703 		/*
3704 		 * Bluetooth privileges.
3705 		 */
3706 	case PRIV_NETBLUETOOTH_RAW:
3707 #endif
3708 
3709 		/*
3710 		 * Netgraph and netgraph module privileges.
3711 		 */
3712 	case PRIV_NETGRAPH_CONTROL:
3713 #ifdef notyet
3714 	case PRIV_NETGRAPH_TTY:
3715 #endif
3716 
3717 		/*
3718 		 * IPv4 and IPv6 privileges.
3719 		 */
3720 	case PRIV_NETINET_IPFW:
3721 	case PRIV_NETINET_DIVERT:
3722 	case PRIV_NETINET_PF:
3723 	case PRIV_NETINET_DUMMYNET:
3724 	case PRIV_NETINET_CARP:
3725 	case PRIV_NETINET_MROUTE:
3726 	case PRIV_NETINET_RAW:
3727 	case PRIV_NETINET_ADDRCTRL6:
3728 	case PRIV_NETINET_ND6:
3729 	case PRIV_NETINET_SCOPE6:
3730 	case PRIV_NETINET_ALIFETIME6:
3731 	case PRIV_NETINET_IPSEC:
3732 	case PRIV_NETINET_BINDANY:
3733 
3734 #ifdef notyet
3735 		/*
3736 		 * NCP privileges.
3737 		 */
3738 	case PRIV_NETNCP:
3739 
3740 		/*
3741 		 * SMB privileges.
3742 		 */
3743 	case PRIV_NETSMB:
3744 #endif
3745 
3746 	/*
3747 	 * No default: or deny here.
3748 	 * In case of no permit fall through to next switch().
3749 	 */
3750 		if (cred->cr_prison->pr_flags & PR_VNET)
3751 			return (0);
3752 	}
3753 #endif /* VIMAGE */
3754 
3755 	switch (priv) {
3756 
3757 		/*
3758 		 * Allow ktrace privileges for root in jail.
3759 		 */
3760 	case PRIV_KTRACE:
3761 
3762 #if 0
3763 		/*
3764 		 * Allow jailed processes to configure audit identity and
3765 		 * submit audit records (login, etc).  In the future we may
3766 		 * want to further refine the relationship between audit and
3767 		 * jail.
3768 		 */
3769 	case PRIV_AUDIT_GETAUDIT:
3770 	case PRIV_AUDIT_SETAUDIT:
3771 	case PRIV_AUDIT_SUBMIT:
3772 #endif
3773 
3774 		/*
3775 		 * Allow jailed processes to manipulate process UNIX
3776 		 * credentials in any way they see fit.
3777 		 */
3778 	case PRIV_CRED_SETUID:
3779 	case PRIV_CRED_SETEUID:
3780 	case PRIV_CRED_SETGID:
3781 	case PRIV_CRED_SETEGID:
3782 	case PRIV_CRED_SETGROUPS:
3783 	case PRIV_CRED_SETREUID:
3784 	case PRIV_CRED_SETREGID:
3785 	case PRIV_CRED_SETRESUID:
3786 	case PRIV_CRED_SETRESGID:
3787 
3788 		/*
3789 		 * Jail implements visibility constraints already, so allow
3790 		 * jailed root to override uid/gid-based constraints.
3791 		 */
3792 	case PRIV_SEEOTHERGIDS:
3793 	case PRIV_SEEOTHERUIDS:
3794 
3795 		/*
3796 		 * Jail implements inter-process debugging limits already, so
3797 		 * allow jailed root various debugging privileges.
3798 		 */
3799 	case PRIV_DEBUG_DIFFCRED:
3800 	case PRIV_DEBUG_SUGID:
3801 	case PRIV_DEBUG_UNPRIV:
3802 
3803 		/*
3804 		 * Allow jail to set various resource limits and login
3805 		 * properties, and for now, exceed process resource limits.
3806 		 */
3807 	case PRIV_PROC_LIMIT:
3808 	case PRIV_PROC_SETLOGIN:
3809 	case PRIV_PROC_SETRLIMIT:
3810 
3811 		/*
3812 		 * System V and POSIX IPC privileges are granted in jail.
3813 		 */
3814 	case PRIV_IPC_READ:
3815 	case PRIV_IPC_WRITE:
3816 	case PRIV_IPC_ADMIN:
3817 	case PRIV_IPC_MSGSIZE:
3818 	case PRIV_MQ_ADMIN:
3819 
3820 		/*
3821 		 * Jail operations within a jail work on child jails.
3822 		 */
3823 	case PRIV_JAIL_ATTACH:
3824 	case PRIV_JAIL_SET:
3825 	case PRIV_JAIL_REMOVE:
3826 
3827 		/*
3828 		 * Jail implements its own inter-process limits, so allow
3829 		 * root processes in jail to change scheduling on other
3830 		 * processes in the same jail.  Likewise for signalling.
3831 		 */
3832 	case PRIV_SCHED_DIFFCRED:
3833 	case PRIV_SCHED_CPUSET:
3834 	case PRIV_SIGNAL_DIFFCRED:
3835 	case PRIV_SIGNAL_SUGID:
3836 
3837 		/*
3838 		 * Allow jailed processes to write to sysctls marked as jail
3839 		 * writable.
3840 		 */
3841 	case PRIV_SYSCTL_WRITEJAIL:
3842 
3843 		/*
3844 		 * Allow root in jail to manage a variety of quota
3845 		 * properties.  These should likely be conditional on a
3846 		 * configuration option.
3847 		 */
3848 	case PRIV_VFS_GETQUOTA:
3849 	case PRIV_VFS_SETQUOTA:
3850 
3851 		/*
3852 		 * Since Jail relies on chroot() to implement file system
3853 		 * protections, grant many VFS privileges to root in jail.
3854 		 * Be careful to exclude mount-related and NFS-related
3855 		 * privileges.
3856 		 */
3857 	case PRIV_VFS_READ:
3858 	case PRIV_VFS_WRITE:
3859 	case PRIV_VFS_ADMIN:
3860 	case PRIV_VFS_EXEC:
3861 	case PRIV_VFS_LOOKUP:
3862 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3863 	case PRIV_VFS_CHFLAGS_DEV:
3864 	case PRIV_VFS_CHOWN:
3865 	case PRIV_VFS_CHROOT:
3866 	case PRIV_VFS_RETAINSUGID:
3867 	case PRIV_VFS_FCHROOT:
3868 	case PRIV_VFS_LINK:
3869 	case PRIV_VFS_SETGID:
3870 	case PRIV_VFS_STAT:
3871 	case PRIV_VFS_STICKYFILE:
3872 
3873 		/*
3874 		 * As in the non-jail case, non-root users are expected to be
3875 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3876 		 * exists in the jail and they have permission to access it).
3877 		 */
3878 	case PRIV_KMEM_READ:
3879 		return (0);
3880 
3881 		/*
3882 		 * Depending on the global setting, allow privilege of
3883 		 * setting system flags.
3884 		 */
3885 	case PRIV_VFS_SYSFLAGS:
3886 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3887 			return (0);
3888 		else
3889 			return (EPERM);
3890 
3891 		/*
3892 		 * Depending on the global setting, allow privilege of
3893 		 * mounting/unmounting file systems.
3894 		 */
3895 	case PRIV_VFS_MOUNT:
3896 	case PRIV_VFS_UNMOUNT:
3897 	case PRIV_VFS_MOUNT_NONUSER:
3898 	case PRIV_VFS_MOUNT_OWNER:
3899 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3900 		    cred->cr_prison->pr_enforce_statfs < 2)
3901 			return (0);
3902 		else
3903 			return (EPERM);
3904 
3905 		/*
3906 		 * Allow jailed root to bind reserved ports and reuse in-use
3907 		 * ports.
3908 		 */
3909 	case PRIV_NETINET_RESERVEDPORT:
3910 	case PRIV_NETINET_REUSEPORT:
3911 		return (0);
3912 
3913 		/*
3914 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3915 		 */
3916 	case PRIV_NETINET_SETHDROPTS:
3917 		return (0);
3918 
3919 		/*
3920 		 * Conditionally allow creating raw sockets in jail.
3921 		 */
3922 	case PRIV_NETINET_RAW:
3923 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3924 			return (0);
3925 		else
3926 			return (EPERM);
3927 
3928 		/*
3929 		 * Since jail implements its own visibility limits on netstat
3930 		 * sysctls, allow getcred.  This allows identd to work in
3931 		 * jail.
3932 		 */
3933 	case PRIV_NETINET_GETCRED:
3934 		return (0);
3935 
3936 		/*
3937 		 * Allow jailed root to set loginclass.
3938 		 */
3939 	case PRIV_PROC_SETLOGINCLASS:
3940 		return (0);
3941 
3942 	default:
3943 		/*
3944 		 * In all remaining cases, deny the privilege request.  This
3945 		 * includes almost all network privileges, many system
3946 		 * configuration privileges.
3947 		 */
3948 		return (EPERM);
3949 	}
3950 }
3951 
3952 /*
3953  * Return the part of pr2's name that is relative to pr1, or the whole name
3954  * if it does not directly follow.
3955  */
3956 
3957 char *
3958 prison_name(struct prison *pr1, struct prison *pr2)
3959 {
3960 	char *name;
3961 
3962 	/* Jails see themselves as "0" (if they see themselves at all). */
3963 	if (pr1 == pr2)
3964 		return "0";
3965 	name = pr2->pr_name;
3966 	if (prison_ischild(pr1, pr2)) {
3967 		/*
3968 		 * pr1 isn't locked (and allprison_lock may not be either)
3969 		 * so its length can't be counted on.  But the number of dots
3970 		 * can be counted on - and counted.
3971 		 */
3972 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3973 			name = strchr(name, '.') + 1;
3974 	}
3975 	return (name);
3976 }
3977 
3978 /*
3979  * Return the part of pr2's path that is relative to pr1, or the whole path
3980  * if it does not directly follow.
3981  */
3982 static char *
3983 prison_path(struct prison *pr1, struct prison *pr2)
3984 {
3985 	char *path1, *path2;
3986 	int len1;
3987 
3988 	path1 = pr1->pr_path;
3989 	path2 = pr2->pr_path;
3990 	if (!strcmp(path1, "/"))
3991 		return (path2);
3992 	len1 = strlen(path1);
3993 	if (strncmp(path1, path2, len1))
3994 		return (path2);
3995 	if (path2[len1] == '\0')
3996 		return "/";
3997 	if (path2[len1] == '/')
3998 		return (path2 + len1);
3999 	return (path2);
4000 }
4001 
4002 
4003 /*
4004  * Jail-related sysctls.
4005  */
4006 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4007     "Jails");
4008 
4009 static int
4010 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4011 {
4012 	struct xprison *xp;
4013 	struct prison *pr, *cpr;
4014 #ifdef INET
4015 	struct in_addr *ip4 = NULL;
4016 	int ip4s = 0;
4017 #endif
4018 #ifdef INET6
4019 	struct in6_addr *ip6 = NULL;
4020 	int ip6s = 0;
4021 #endif
4022 	int descend, error;
4023 
4024 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4025 	pr = req->td->td_ucred->cr_prison;
4026 	error = 0;
4027 	sx_slock(&allprison_lock);
4028 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4029 #if defined(INET) || defined(INET6)
4030  again:
4031 #endif
4032 		mtx_lock(&cpr->pr_mtx);
4033 #ifdef INET
4034 		if (cpr->pr_ip4s > 0) {
4035 			if (ip4s < cpr->pr_ip4s) {
4036 				ip4s = cpr->pr_ip4s;
4037 				mtx_unlock(&cpr->pr_mtx);
4038 				ip4 = realloc(ip4, ip4s *
4039 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4040 				goto again;
4041 			}
4042 			bcopy(cpr->pr_ip4, ip4,
4043 			    cpr->pr_ip4s * sizeof(struct in_addr));
4044 		}
4045 #endif
4046 #ifdef INET6
4047 		if (cpr->pr_ip6s > 0) {
4048 			if (ip6s < cpr->pr_ip6s) {
4049 				ip6s = cpr->pr_ip6s;
4050 				mtx_unlock(&cpr->pr_mtx);
4051 				ip6 = realloc(ip6, ip6s *
4052 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4053 				goto again;
4054 			}
4055 			bcopy(cpr->pr_ip6, ip6,
4056 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4057 		}
4058 #endif
4059 		if (cpr->pr_ref == 0) {
4060 			mtx_unlock(&cpr->pr_mtx);
4061 			continue;
4062 		}
4063 		bzero(xp, sizeof(*xp));
4064 		xp->pr_version = XPRISON_VERSION;
4065 		xp->pr_id = cpr->pr_id;
4066 		xp->pr_state = cpr->pr_uref > 0
4067 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4068 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4069 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4070 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4071 #ifdef INET
4072 		xp->pr_ip4s = cpr->pr_ip4s;
4073 #endif
4074 #ifdef INET6
4075 		xp->pr_ip6s = cpr->pr_ip6s;
4076 #endif
4077 		mtx_unlock(&cpr->pr_mtx);
4078 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4079 		if (error)
4080 			break;
4081 #ifdef INET
4082 		if (xp->pr_ip4s > 0) {
4083 			error = SYSCTL_OUT(req, ip4,
4084 			    xp->pr_ip4s * sizeof(struct in_addr));
4085 			if (error)
4086 				break;
4087 		}
4088 #endif
4089 #ifdef INET6
4090 		if (xp->pr_ip6s > 0) {
4091 			error = SYSCTL_OUT(req, ip6,
4092 			    xp->pr_ip6s * sizeof(struct in6_addr));
4093 			if (error)
4094 				break;
4095 		}
4096 #endif
4097 	}
4098 	sx_sunlock(&allprison_lock);
4099 	free(xp, M_TEMP);
4100 #ifdef INET
4101 	free(ip4, M_TEMP);
4102 #endif
4103 #ifdef INET6
4104 	free(ip6, M_TEMP);
4105 #endif
4106 	return (error);
4107 }
4108 
4109 SYSCTL_OID(_security_jail, OID_AUTO, list,
4110     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4111     sysctl_jail_list, "S", "List of active jails");
4112 
4113 static int
4114 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4115 {
4116 	int error, injail;
4117 
4118 	injail = jailed(req->td->td_ucred);
4119 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4120 
4121 	return (error);
4122 }
4123 
4124 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4125     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4126     sysctl_jail_jailed, "I", "Process in jail?");
4127 
4128 static int
4129 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4130 {
4131 	int error, havevnet;
4132 #ifdef VIMAGE
4133 	struct ucred *cred = req->td->td_ucred;
4134 
4135 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4136 #else
4137 	havevnet = 0;
4138 #endif
4139 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4140 
4141 	return (error);
4142 }
4143 
4144 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4145     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4146     sysctl_jail_vnet, "I", "Jail owns VNET?");
4147 
4148 #if defined(INET) || defined(INET6)
4149 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4150     &jail_max_af_ips, 0,
4151     "Number of IP addresses a jail may have at most per address family");
4152 #endif
4153 
4154 /*
4155  * Default parameters for jail(2) compatability.  For historical reasons,
4156  * the sysctl names have varying similarity to the parameter names.  Prisons
4157  * just see their own parameters, and can't change them.
4158  */
4159 static int
4160 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4161 {
4162 	struct prison *pr;
4163 	int allow, error, i;
4164 
4165 	pr = req->td->td_ucred->cr_prison;
4166 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4167 
4168 	/* Get the current flag value, and convert it to a boolean. */
4169 	i = (allow & arg2) ? 1 : 0;
4170 	if (arg1 != NULL)
4171 		i = !i;
4172 	error = sysctl_handle_int(oidp, &i, 0, req);
4173 	if (error || !req->newptr)
4174 		return (error);
4175 	i = i ? arg2 : 0;
4176 	if (arg1 != NULL)
4177 		i ^= arg2;
4178 	/*
4179 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4180 	 * for writing.
4181 	 */
4182 	mtx_lock(&prison0.pr_mtx);
4183 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4184 	mtx_unlock(&prison0.pr_mtx);
4185 	return (0);
4186 }
4187 
4188 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4189     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4190     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4191     "Processes in jail can set their hostnames");
4192 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4193     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4194     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4195     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4196 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4197     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4198     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4199     "Processes in jail can use System V IPC primitives");
4200 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4201     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4202     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4203     "Prison root can create raw sockets");
4204 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4205     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4206     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4207     "Processes in jail can alter system file flags");
4208 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4209     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4210     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4211     "Processes in jail can mount/unmount jail-friendly file systems");
4212 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4213     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4214     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4215     "Processes in jail can mount the devfs file system");
4216 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4217     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4218     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4219     "Processes in jail can mount the nullfs file system");
4220 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4221     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4222     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4223     "Processes in jail can mount the procfs file system");
4224 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4225     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4226     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4227     "Processes in jail can mount the tmpfs file system");
4228 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4229     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4230     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4231     "Processes in jail can mount the zfs file system");
4232 
4233 static int
4234 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4235 {
4236 	struct prison *pr;
4237 	int level, error;
4238 
4239 	pr = req->td->td_ucred->cr_prison;
4240 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4241 	error = sysctl_handle_int(oidp, &level, 0, req);
4242 	if (error || !req->newptr)
4243 		return (error);
4244 	*(int *)arg1 = level;
4245 	return (0);
4246 }
4247 
4248 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4249     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4250     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4251     sysctl_jail_default_level, "I",
4252     "Processes in jail cannot see all mounted file systems");
4253 
4254 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4255     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4256     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4257     sysctl_jail_default_level, "I",
4258     "Ruleset for the devfs filesystem in jail");
4259 
4260 /*
4261  * Nodes to describe jail parameters.  Maximum length of string parameters
4262  * is returned in the string itself, and the other parameters exist merely
4263  * to make themselves and their types known.
4264  */
4265 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4266     "Jail parameters");
4267 
4268 int
4269 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4270 {
4271 	int i;
4272 	long l;
4273 	size_t s;
4274 	char numbuf[12];
4275 
4276 	switch (oidp->oid_kind & CTLTYPE)
4277 	{
4278 	case CTLTYPE_LONG:
4279 	case CTLTYPE_ULONG:
4280 		l = 0;
4281 #ifdef SCTL_MASK32
4282 		if (!(req->flags & SCTL_MASK32))
4283 #endif
4284 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4285 	case CTLTYPE_INT:
4286 	case CTLTYPE_UINT:
4287 		i = 0;
4288 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4289 	case CTLTYPE_STRING:
4290 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4291 		return
4292 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4293 	case CTLTYPE_STRUCT:
4294 		s = (size_t)arg2;
4295 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4296 	}
4297 	return (0);
4298 }
4299 
4300 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4301 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4302 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4303 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4304 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4305     "I", "Jail secure level");
4306 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4307     "I", "Jail cannot see all mounted file systems");
4308 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4309     "I", "Ruleset for in-jail devfs mounts");
4310 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4311     "B", "Jail persistence");
4312 #ifdef VIMAGE
4313 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4314     "E,jailsys", "Virtual network stack");
4315 #endif
4316 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4317     "B", "Jail is in the process of shutting down");
4318 
4319 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4320 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4321     "I", "Current number of child jails");
4322 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4323     "I", "Maximum number of child jails");
4324 
4325 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4326 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4327     "Jail hostname");
4328 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4329     "Jail NIS domainname");
4330 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4331     "Jail host UUID");
4332 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4333     "LU", "Jail host ID");
4334 
4335 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4336 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4337 
4338 #ifdef INET
4339 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4340     "Jail IPv4 address virtualization");
4341 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4342     "S,in_addr,a", "Jail IPv4 addresses");
4343 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4344     "B", "Do (not) use IPv4 source address selection rather than the "
4345     "primary jail IPv4 address.");
4346 #endif
4347 #ifdef INET6
4348 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4349     "Jail IPv6 address virtualization");
4350 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4351     "S,in6_addr,a", "Jail IPv6 addresses");
4352 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4353     "B", "Do (not) use IPv6 source address selection rather than the "
4354     "primary jail IPv6 address.");
4355 #endif
4356 
4357 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4358 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4359     "B", "Jail may set hostname");
4360 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4361     "B", "Jail may use SYSV IPC");
4362 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4363     "B", "Jail may create raw sockets");
4364 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4365     "B", "Jail may alter system file flags");
4366 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4367     "B", "Jail may set file quotas");
4368 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4369     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4370 
4371 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4372 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4373     "B", "Jail may mount/unmount jail-friendly file systems in general");
4374 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4375     "B", "Jail may mount the devfs file system");
4376 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4377     "B", "Jail may mount the nullfs file system");
4378 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4379     "B", "Jail may mount the procfs file system");
4380 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4381     "B", "Jail may mount the tmpfs file system");
4382 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4383     "B", "Jail may mount the zfs file system");
4384 
4385 void
4386 prison_racct_foreach(void (*callback)(struct racct *racct,
4387     void *arg2, void *arg3), void *arg2, void *arg3)
4388 {
4389 	struct prison_racct *prr;
4390 
4391 	sx_slock(&allprison_lock);
4392 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4393 		(callback)(prr->prr_racct, arg2, arg3);
4394 	sx_sunlock(&allprison_lock);
4395 }
4396 
4397 static struct prison_racct *
4398 prison_racct_find_locked(const char *name)
4399 {
4400 	struct prison_racct *prr;
4401 
4402 	sx_assert(&allprison_lock, SA_XLOCKED);
4403 
4404 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4405 		return (NULL);
4406 
4407 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4408 		if (strcmp(name, prr->prr_name) != 0)
4409 			continue;
4410 
4411 		/* Found prison_racct with a matching name? */
4412 		prison_racct_hold(prr);
4413 		return (prr);
4414 	}
4415 
4416 	/* Add new prison_racct. */
4417 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4418 	racct_create(&prr->prr_racct);
4419 
4420 	strcpy(prr->prr_name, name);
4421 	refcount_init(&prr->prr_refcount, 1);
4422 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4423 
4424 	return (prr);
4425 }
4426 
4427 struct prison_racct *
4428 prison_racct_find(const char *name)
4429 {
4430 	struct prison_racct *prr;
4431 
4432 	sx_xlock(&allprison_lock);
4433 	prr = prison_racct_find_locked(name);
4434 	sx_xunlock(&allprison_lock);
4435 	return (prr);
4436 }
4437 
4438 void
4439 prison_racct_hold(struct prison_racct *prr)
4440 {
4441 
4442 	refcount_acquire(&prr->prr_refcount);
4443 }
4444 
4445 static void
4446 prison_racct_free_locked(struct prison_racct *prr)
4447 {
4448 
4449 	sx_assert(&allprison_lock, SA_XLOCKED);
4450 
4451 	if (refcount_release(&prr->prr_refcount)) {
4452 		racct_destroy(&prr->prr_racct);
4453 		LIST_REMOVE(prr, prr_next);
4454 		free(prr, M_PRISON_RACCT);
4455 	}
4456 }
4457 
4458 void
4459 prison_racct_free(struct prison_racct *prr)
4460 {
4461 	int old;
4462 
4463 	sx_assert(&allprison_lock, SA_UNLOCKED);
4464 
4465 	old = prr->prr_refcount;
4466 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4467 		return;
4468 
4469 	sx_xlock(&allprison_lock);
4470 	prison_racct_free_locked(prr);
4471 	sx_xunlock(&allprison_lock);
4472 }
4473 
4474 #ifdef RACCT
4475 static void
4476 prison_racct_attach(struct prison *pr)
4477 {
4478 	struct prison_racct *prr;
4479 
4480 	sx_assert(&allprison_lock, SA_XLOCKED);
4481 
4482 	prr = prison_racct_find_locked(pr->pr_name);
4483 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4484 
4485 	pr->pr_prison_racct = prr;
4486 }
4487 
4488 /*
4489  * Handle jail renaming.  From the racct point of view, renaming means
4490  * moving from one prison_racct to another.
4491  */
4492 static void
4493 prison_racct_modify(struct prison *pr)
4494 {
4495 	struct proc *p;
4496 	struct ucred *cred;
4497 	struct prison_racct *oldprr;
4498 
4499 	sx_slock(&allproc_lock);
4500 	sx_xlock(&allprison_lock);
4501 
4502 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4503 		sx_xunlock(&allprison_lock);
4504 		sx_sunlock(&allproc_lock);
4505 		return;
4506 	}
4507 
4508 	oldprr = pr->pr_prison_racct;
4509 	pr->pr_prison_racct = NULL;
4510 
4511 	prison_racct_attach(pr);
4512 
4513 	/*
4514 	 * Move resource utilisation records.
4515 	 */
4516 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4517 
4518 	/*
4519 	 * Force rctl to reattach rules to processes.
4520 	 */
4521 	FOREACH_PROC_IN_SYSTEM(p) {
4522 		PROC_LOCK(p);
4523 		cred = crhold(p->p_ucred);
4524 		PROC_UNLOCK(p);
4525 		racct_proc_ucred_changed(p, cred, cred);
4526 		crfree(cred);
4527 	}
4528 
4529 	sx_sunlock(&allproc_lock);
4530 	prison_racct_free_locked(oldprr);
4531 	sx_xunlock(&allprison_lock);
4532 }
4533 
4534 static void
4535 prison_racct_detach(struct prison *pr)
4536 {
4537 
4538 	sx_assert(&allprison_lock, SA_UNLOCKED);
4539 
4540 	if (pr->pr_prison_racct == NULL)
4541 		return;
4542 	prison_racct_free(pr->pr_prison_racct);
4543 	pr->pr_prison_racct = NULL;
4544 }
4545 #endif /* RACCT */
4546 
4547 #ifdef DDB
4548 
4549 static void
4550 db_show_prison(struct prison *pr)
4551 {
4552 	int fi;
4553 #if defined(INET) || defined(INET6)
4554 	int ii;
4555 #endif
4556 	unsigned jsf;
4557 #ifdef INET6
4558 	char ip6buf[INET6_ADDRSTRLEN];
4559 #endif
4560 
4561 	db_printf("prison %p:\n", pr);
4562 	db_printf(" jid             = %d\n", pr->pr_id);
4563 	db_printf(" name            = %s\n", pr->pr_name);
4564 	db_printf(" parent          = %p\n", pr->pr_parent);
4565 	db_printf(" ref             = %d\n", pr->pr_ref);
4566 	db_printf(" uref            = %d\n", pr->pr_uref);
4567 	db_printf(" path            = %s\n", pr->pr_path);
4568 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4569 	    ? pr->pr_cpuset->cs_id : -1);
4570 #ifdef VIMAGE
4571 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4572 #endif
4573 	db_printf(" root            = %p\n", pr->pr_root);
4574 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4575 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4576 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4577 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4578 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4579 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4580 	db_printf(" flags           = 0x%x", pr->pr_flags);
4581 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4582 	    fi++)
4583 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4584 			db_printf(" %s", pr_flag_names[fi]);
4585 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4586 	    fi++) {
4587 		jsf = pr->pr_flags &
4588 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4589 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4590 		    pr_flag_jailsys[fi].disable &&
4591 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4592 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4593 		    : "inherit");
4594 	}
4595 	db_printf(" allow           = 0x%x", pr->pr_allow);
4596 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4597 	    fi++)
4598 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4599 			db_printf(" %s", pr_allow_names[fi]);
4600 	db_printf("\n");
4601 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4602 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4603 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4604 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4605 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4606 #ifdef INET
4607 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4608 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4609 		db_printf(" %s %s\n",
4610 		    ii == 0 ? "ip4.addr        =" : "                 ",
4611 		    inet_ntoa(pr->pr_ip4[ii]));
4612 #endif
4613 #ifdef INET6
4614 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4615 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4616 		db_printf(" %s %s\n",
4617 		    ii == 0 ? "ip6.addr        =" : "                 ",
4618 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4619 #endif
4620 }
4621 
4622 DB_SHOW_COMMAND(prison, db_show_prison_command)
4623 {
4624 	struct prison *pr;
4625 
4626 	if (!have_addr) {
4627 		/*
4628 		 * Show all prisons in the list, and prison0 which is not
4629 		 * listed.
4630 		 */
4631 		db_show_prison(&prison0);
4632 		if (!db_pager_quit) {
4633 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4634 				db_show_prison(pr);
4635 				if (db_pager_quit)
4636 					break;
4637 			}
4638 		}
4639 		return;
4640 	}
4641 
4642 	if (addr == 0)
4643 		pr = &prison0;
4644 	else {
4645 		/* Look for a prison with the ID and with references. */
4646 		TAILQ_FOREACH(pr, &allprison, pr_list)
4647 			if (pr->pr_id == addr && pr->pr_ref > 0)
4648 				break;
4649 		if (pr == NULL)
4650 			/* Look again, without requiring a reference. */
4651 			TAILQ_FOREACH(pr, &allprison, pr_list)
4652 				if (pr->pr_id == addr)
4653 					break;
4654 		if (pr == NULL)
4655 			/* Assume address points to a valid prison. */
4656 			pr = (struct prison *)addr;
4657 	}
4658 	db_show_prison(pr);
4659 }
4660 
4661 #endif /* DDB */
4662