xref: /freebsd/sys/kern/kern_jail.c (revision 3823d5e198425b4f5e5a80267d195769d1063773)
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_DISABLE, PR_IP4_USER },
191 #endif
192 #ifdef INET6
193 	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, 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 = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
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 | PR_IP4_DISABLE;
811 		if (ip4s == 0)
812 			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
813 		else {
814 			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
815 			ip4s /= sizeof(*ip4);
816 			if (ip4s > jail_max_af_ips) {
817 				error = EINVAL;
818 				vfs_opterror(opts, "too many IPv4 addresses");
819 				goto done_errmsg;
820 			}
821 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
822 			bcopy(op, ip4, ip4s * sizeof(*ip4));
823 			/*
824 			 * IP addresses are all sorted but ip[0] to preserve
825 			 * the primary IP address as given from userland.
826 			 * This special IP is used for unbound outgoing
827 			 * connections as well for "loopback" traffic in case
828 			 * source address selection cannot find any more fitting
829 			 * address to connect from.
830 			 */
831 			if (ip4s > 1)
832 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
833 			/*
834 			 * Check for duplicate addresses and do some simple
835 			 * zero and broadcast checks. If users give other bogus
836 			 * addresses it is their problem.
837 			 *
838 			 * We do not have to care about byte order for these
839 			 * checks so we will do them in NBO.
840 			 */
841 			for (ii = 0; ii < ip4s; ii++) {
842 				if (ip4[ii].s_addr == INADDR_ANY ||
843 				    ip4[ii].s_addr == INADDR_BROADCAST) {
844 					error = EINVAL;
845 					goto done_free;
846 				}
847 				if ((ii+1) < ip4s &&
848 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
849 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
850 					error = EINVAL;
851 					goto done_free;
852 				}
853 			}
854 		}
855 	}
856 #endif
857 
858 #ifdef INET6
859 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
860 	if (error == ENOENT)
861 		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
862 	else if (error != 0)
863 		goto done_free;
864 	else if (ip6s & (sizeof(*ip6) - 1)) {
865 		error = EINVAL;
866 		goto done_free;
867 	} else {
868 		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
869 		if (ip6s == 0)
870 			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
871 		else {
872 			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
873 			ip6s /= sizeof(*ip6);
874 			if (ip6s > jail_max_af_ips) {
875 				error = EINVAL;
876 				vfs_opterror(opts, "too many IPv6 addresses");
877 				goto done_errmsg;
878 			}
879 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
880 			bcopy(op, ip6, ip6s * sizeof(*ip6));
881 			if (ip6s > 1)
882 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
883 			for (ii = 0; ii < ip6s; ii++) {
884 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
885 					error = EINVAL;
886 					goto done_free;
887 				}
888 				if ((ii+1) < ip6s &&
889 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
890 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
891 				{
892 					error = EINVAL;
893 					goto done_free;
894 				}
895 			}
896 		}
897 	}
898 #endif
899 
900 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
901 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
902 		error = EINVAL;
903 		vfs_opterror(opts,
904 		    "vnet jails cannot have IP address restrictions");
905 		goto done_errmsg;
906 	}
907 #endif
908 
909 	fullpath_disabled = 0;
910 	root = NULL;
911 	error = vfs_getopt(opts, "path", (void **)&path, &len);
912 	if (error == ENOENT)
913 		path = NULL;
914 	else if (error != 0)
915 		goto done_free;
916 	else {
917 		if (flags & JAIL_UPDATE) {
918 			error = EINVAL;
919 			vfs_opterror(opts,
920 			    "path cannot be changed after creation");
921 			goto done_errmsg;
922 		}
923 		if (len == 0 || path[len - 1] != '\0') {
924 			error = EINVAL;
925 			goto done_free;
926 		}
927 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
928 		    path, td);
929 		error = namei(&nd);
930 		if (error)
931 			goto done_free;
932 		root = nd.ni_vp;
933 		NDFREE(&nd, NDF_ONLY_PNBUF);
934 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
935 		strlcpy(g_path, path, MAXPATHLEN);
936 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
937 		if (error == 0)
938 			path = g_path;
939 		else if (error == ENODEV) {
940 			/* proceed if sysctl debug.disablefullpath == 1 */
941 			fullpath_disabled = 1;
942 			if (len < 2 || (len == 2 && path[0] == '/'))
943 				path = NULL;
944 		} else {
945 			/* exit on other errors */
946 			goto done_free;
947 		}
948 		if (root->v_type != VDIR) {
949 			error = ENOTDIR;
950 			vput(root);
951 			goto done_free;
952 		}
953 		VOP_UNLOCK(root, 0);
954 		if (fullpath_disabled) {
955 			/* Leave room for a real-root full pathname. */
956 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
957 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
958 				error = ENAMETOOLONG;
959 				goto done_free;
960 			}
961 		}
962 	}
963 
964 	/*
965 	 * Grab the allprison lock before letting modules check their
966 	 * parameters.  Once we have it, do not let go so we'll have a
967 	 * consistent view of the OSD list.
968 	 */
969 	sx_xlock(&allprison_lock);
970 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
971 	if (error)
972 		goto done_unlock_list;
973 
974 	/* By now, all parameters should have been noted. */
975 	TAILQ_FOREACH(opt, opts, link) {
976 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
977 			error = EINVAL;
978 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
979 			goto done_unlock_list;
980 		}
981 	}
982 
983 	/*
984 	 * See if we are creating a new record or updating an existing one.
985 	 * This abuses the file error codes ENOENT and EEXIST.
986 	 */
987 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
988 	if (!cuflags) {
989 		error = EINVAL;
990 		vfs_opterror(opts, "no valid operation (create or update)");
991 		goto done_unlock_list;
992 	}
993 	pr = NULL;
994 	namelc = NULL;
995 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
996 		namelc = strrchr(name, '.');
997 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
998 		if (*p != '\0')
999 			jid = 0;
1000 	}
1001 	if (jid != 0) {
1002 		/*
1003 		 * See if a requested jid already exists.  There is an
1004 		 * information leak here if the jid exists but is not within
1005 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1006 		 * even though they cannot see the jail, and CREATE | UPDATE
1007 		 * will return ENOENT which is not normally a valid error.
1008 		 */
1009 		if (jid < 0) {
1010 			error = EINVAL;
1011 			vfs_opterror(opts, "negative jid");
1012 			goto done_unlock_list;
1013 		}
1014 		pr = prison_find(jid);
1015 		if (pr != NULL) {
1016 			ppr = pr->pr_parent;
1017 			/* Create: jid must not exist. */
1018 			if (cuflags == JAIL_CREATE) {
1019 				mtx_unlock(&pr->pr_mtx);
1020 				error = EEXIST;
1021 				vfs_opterror(opts, "jail %d already exists",
1022 				    jid);
1023 				goto done_unlock_list;
1024 			}
1025 			if (!prison_ischild(mypr, pr)) {
1026 				mtx_unlock(&pr->pr_mtx);
1027 				pr = NULL;
1028 			} else if (pr->pr_uref == 0) {
1029 				if (!(flags & JAIL_DYING)) {
1030 					mtx_unlock(&pr->pr_mtx);
1031 					error = ENOENT;
1032 					vfs_opterror(opts, "jail %d is dying",
1033 					    jid);
1034 					goto done_unlock_list;
1035 				} else if ((flags & JAIL_ATTACH) ||
1036 				    (pr_flags & PR_PERSIST)) {
1037 					/*
1038 					 * A dying jail might be resurrected
1039 					 * (via attach or persist), but first
1040 					 * it must determine if another jail
1041 					 * has claimed its name.  Accomplish
1042 					 * this by implicitly re-setting the
1043 					 * name.
1044 					 */
1045 					if (name == NULL)
1046 						name = prison_name(mypr, pr);
1047 				}
1048 			}
1049 		}
1050 		if (pr == NULL) {
1051 			/* Update: jid must exist. */
1052 			if (cuflags == JAIL_UPDATE) {
1053 				error = ENOENT;
1054 				vfs_opterror(opts, "jail %d not found", jid);
1055 				goto done_unlock_list;
1056 			}
1057 		}
1058 	}
1059 	/*
1060 	 * If the caller provided a name, look for a jail by that name.
1061 	 * This has different semantics for creates and updates keyed by jid
1062 	 * (where the name must not already exist in a different jail),
1063 	 * and updates keyed by the name itself (where the name must exist
1064 	 * because that is the jail being updated).
1065 	 */
1066 	if (name != NULL) {
1067 		namelc = strrchr(name, '.');
1068 		if (namelc == NULL)
1069 			namelc = name;
1070 		else {
1071 			/*
1072 			 * This is a hierarchical name.  Split it into the
1073 			 * parent and child names, and make sure the parent
1074 			 * exists or matches an already found jail.
1075 			 */
1076 			*namelc = '\0';
1077 			if (pr != NULL) {
1078 				if (strncmp(name, ppr->pr_name, namelc - name)
1079 				    || ppr->pr_name[namelc - name] != '\0') {
1080 					mtx_unlock(&pr->pr_mtx);
1081 					error = EINVAL;
1082 					vfs_opterror(opts,
1083 					    "cannot change jail's parent");
1084 					goto done_unlock_list;
1085 				}
1086 			} else {
1087 				ppr = prison_find_name(mypr, name);
1088 				if (ppr == NULL) {
1089 					error = ENOENT;
1090 					vfs_opterror(opts,
1091 					    "jail \"%s\" not found", name);
1092 					goto done_unlock_list;
1093 				}
1094 				mtx_unlock(&ppr->pr_mtx);
1095 			}
1096 			name = ++namelc;
1097 		}
1098 		if (name[0] != '\0') {
1099 			namelen =
1100 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1101  name_again:
1102 			deadpr = NULL;
1103 			FOREACH_PRISON_CHILD(ppr, tpr) {
1104 				if (tpr != pr && tpr->pr_ref > 0 &&
1105 				    !strcmp(tpr->pr_name + namelen, name)) {
1106 					if (pr == NULL &&
1107 					    cuflags != JAIL_CREATE) {
1108 						mtx_lock(&tpr->pr_mtx);
1109 						if (tpr->pr_ref > 0) {
1110 							/*
1111 							 * Use this jail
1112 							 * for updates.
1113 							 */
1114 							if (tpr->pr_uref > 0) {
1115 								pr = tpr;
1116 								break;
1117 							}
1118 							deadpr = tpr;
1119 						}
1120 						mtx_unlock(&tpr->pr_mtx);
1121 					} else if (tpr->pr_uref > 0) {
1122 						/*
1123 						 * Create, or update(jid):
1124 						 * name must not exist in an
1125 						 * active sibling jail.
1126 						 */
1127 						error = EEXIST;
1128 						if (pr != NULL)
1129 							mtx_unlock(&pr->pr_mtx);
1130 						vfs_opterror(opts,
1131 						   "jail \"%s\" already exists",
1132 						   name);
1133 						goto done_unlock_list;
1134 					}
1135 				}
1136 			}
1137 			/* If no active jail is found, use a dying one. */
1138 			if (deadpr != NULL && pr == NULL) {
1139 				if (flags & JAIL_DYING) {
1140 					mtx_lock(&deadpr->pr_mtx);
1141 					if (deadpr->pr_ref == 0) {
1142 						mtx_unlock(&deadpr->pr_mtx);
1143 						goto name_again;
1144 					}
1145 					pr = deadpr;
1146 				} else if (cuflags == JAIL_UPDATE) {
1147 					error = ENOENT;
1148 					vfs_opterror(opts,
1149 					    "jail \"%s\" is dying", name);
1150 					goto done_unlock_list;
1151 				}
1152 			}
1153 			/* Update: name must exist if no jid. */
1154 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1155 				error = ENOENT;
1156 				vfs_opterror(opts, "jail \"%s\" not found",
1157 				    name);
1158 				goto done_unlock_list;
1159 			}
1160 		}
1161 	}
1162 	/* Update: must provide a jid or name. */
1163 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1164 		error = ENOENT;
1165 		vfs_opterror(opts, "update specified no jail");
1166 		goto done_unlock_list;
1167 	}
1168 
1169 	/* If there's no prison to update, create a new one and link it in. */
1170 	if (pr == NULL) {
1171 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1172 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1173 				error = EPERM;
1174 				vfs_opterror(opts, "prison limit exceeded");
1175 				goto done_unlock_list;
1176 			}
1177 		created = 1;
1178 		mtx_lock(&ppr->pr_mtx);
1179 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1180 			mtx_unlock(&ppr->pr_mtx);
1181 			error = ENOENT;
1182 			vfs_opterror(opts, "parent jail went away!");
1183 			goto done_unlock_list;
1184 		}
1185 		ppr->pr_ref++;
1186 		ppr->pr_uref++;
1187 		mtx_unlock(&ppr->pr_mtx);
1188 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1189 		if (jid == 0) {
1190 			/* Find the next free jid. */
1191 			jid = lastprid + 1;
1192  findnext:
1193 			if (jid == JAIL_MAX)
1194 				jid = 1;
1195 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1196 				if (tpr->pr_id < jid)
1197 					continue;
1198 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1199 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1200 					break;
1201 				}
1202 				if (jid == lastprid) {
1203 					error = EAGAIN;
1204 					vfs_opterror(opts,
1205 					    "no available jail IDs");
1206 					free(pr, M_PRISON);
1207 					prison_deref(ppr, PD_DEREF |
1208 					    PD_DEUREF | PD_LIST_XLOCKED);
1209 					goto done_releroot;
1210 				}
1211 				jid++;
1212 				goto findnext;
1213 			}
1214 			lastprid = jid;
1215 		} else {
1216 			/*
1217 			 * The jail already has a jid (that did not yet exist),
1218 			 * so just find where to insert it.
1219 			 */
1220 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1221 				if (tpr->pr_id >= jid) {
1222 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1223 					break;
1224 				}
1225 		}
1226 		if (tpr == NULL)
1227 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1228 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1229 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1230 			tpr->pr_childcount++;
1231 
1232 		pr->pr_parent = ppr;
1233 		pr->pr_id = jid;
1234 
1235 		/* Set some default values, and inherit some from the parent. */
1236 		if (name == NULL)
1237 			name = "";
1238 		if (path == NULL) {
1239 			path = "/";
1240 			root = mypr->pr_root;
1241 			vref(root);
1242 		}
1243 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1244 		pr->pr_flags |= PR_HOST;
1245 #if defined(INET) || defined(INET6)
1246 #ifdef VIMAGE
1247 		if (!(pr_flags & PR_VNET))
1248 #endif
1249 		{
1250 #ifdef INET
1251 			if (!(ch_flags & PR_IP4_USER))
1252 				pr->pr_flags |=
1253 				    PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1254 			else if (!(pr_flags & PR_IP4_USER)) {
1255 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1256 				if (ppr->pr_ip4 != NULL) {
1257 					pr->pr_ip4s = ppr->pr_ip4s;
1258 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1259 					    sizeof(struct in_addr), M_PRISON,
1260 					    M_WAITOK);
1261 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1262 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1263 				}
1264 			}
1265 #endif
1266 #ifdef INET6
1267 			if (!(ch_flags & PR_IP6_USER))
1268 				pr->pr_flags |=
1269 				    PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1270 			else if (!(pr_flags & PR_IP6_USER)) {
1271 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1272 				if (ppr->pr_ip6 != NULL) {
1273 					pr->pr_ip6s = ppr->pr_ip6s;
1274 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1275 					    sizeof(struct in6_addr), M_PRISON,
1276 					    M_WAITOK);
1277 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1278 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1279 				}
1280 			}
1281 #endif
1282 		}
1283 #endif
1284 		/* Source address selection is always on by default. */
1285 		pr->pr_flags |= _PR_IP_SADDRSEL;
1286 
1287 		pr->pr_securelevel = ppr->pr_securelevel;
1288 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1289 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1290 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1291 
1292 		LIST_INIT(&pr->pr_children);
1293 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1294 
1295 #ifdef VIMAGE
1296 		/* Allocate a new vnet if specified. */
1297 		pr->pr_vnet = (pr_flags & PR_VNET)
1298 		    ? vnet_alloc() : ppr->pr_vnet;
1299 #endif
1300 		/*
1301 		 * Allocate a dedicated cpuset for each jail.
1302 		 * Unlike other initial settings, this may return an erorr.
1303 		 */
1304 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1305 		if (error) {
1306 			prison_deref(pr, PD_LIST_XLOCKED);
1307 			goto done_releroot;
1308 		}
1309 
1310 		mtx_lock(&pr->pr_mtx);
1311 		/*
1312 		 * New prisons do not yet have a reference, because we do not
1313 		 * want other to see the incomplete prison once the
1314 		 * allprison_lock is downgraded.
1315 		 */
1316 	} else {
1317 		created = 0;
1318 		/*
1319 		 * Grab a reference for existing prisons, to ensure they
1320 		 * continue to exist for the duration of the call.
1321 		 */
1322 		pr->pr_ref++;
1323 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1324 		if ((pr->pr_flags & PR_VNET) &&
1325 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1326 			error = EINVAL;
1327 			vfs_opterror(opts,
1328 			    "vnet jails cannot have IP address restrictions");
1329 			goto done_deref_locked;
1330 		}
1331 #endif
1332 #ifdef INET
1333 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1334 			error = EINVAL;
1335 			vfs_opterror(opts,
1336 			    "ip4 cannot be changed after creation");
1337 			goto done_deref_locked;
1338 		}
1339 #endif
1340 #ifdef INET6
1341 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1342 			error = EINVAL;
1343 			vfs_opterror(opts,
1344 			    "ip6 cannot be changed after creation");
1345 			goto done_deref_locked;
1346 		}
1347 #endif
1348 	}
1349 
1350 	/* Do final error checking before setting anything. */
1351 	if (gotslevel) {
1352 		if (slevel < ppr->pr_securelevel) {
1353 			error = EPERM;
1354 			goto done_deref_locked;
1355 		}
1356 	}
1357 	if (gotchildmax) {
1358 		if (childmax >= ppr->pr_childmax) {
1359 			error = EPERM;
1360 			goto done_deref_locked;
1361 		}
1362 	}
1363 	if (gotenforce) {
1364 		if (enforce < ppr->pr_enforce_statfs) {
1365 			error = EPERM;
1366 			goto done_deref_locked;
1367 		}
1368 	}
1369 	if (gotrsnum) {
1370 		/*
1371 		 * devfs_rsnum is a uint16_t
1372 		 */
1373 		if (rsnum < 0 || rsnum > 65535) {
1374 			error = EINVAL;
1375 			goto done_deref_locked;
1376 		}
1377 		/*
1378 		 * Nested jails always inherit parent's devfs ruleset
1379 		 */
1380 		if (jailed(td->td_ucred)) {
1381 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1382 				error = EPERM;
1383 				goto done_deref_locked;
1384 			} else
1385 				rsnum = ppr->pr_devfs_rsnum;
1386 		}
1387 	}
1388 #ifdef INET
1389 	if (ip4s > 0) {
1390 		if (ppr->pr_flags & PR_IP4) {
1391 			/*
1392 			 * Make sure the new set of IP addresses is a
1393 			 * subset of the parent's list.  Don't worry
1394 			 * about the parent being unlocked, as any
1395 			 * setting is done with allprison_lock held.
1396 			 */
1397 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1398 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1399 					break;
1400 			if (ij == ppr->pr_ip4s) {
1401 				error = EPERM;
1402 				goto done_deref_locked;
1403 			}
1404 			if (ip4s > 1) {
1405 				for (ii = ij = 1; ii < ip4s; ii++) {
1406 					if (ip4[ii].s_addr ==
1407 					    ppr->pr_ip4[0].s_addr)
1408 						continue;
1409 					for (; ij < ppr->pr_ip4s; ij++)
1410 						if (ip4[ii].s_addr ==
1411 						    ppr->pr_ip4[ij].s_addr)
1412 							break;
1413 					if (ij == ppr->pr_ip4s)
1414 						break;
1415 				}
1416 				if (ij == ppr->pr_ip4s) {
1417 					error = EPERM;
1418 					goto done_deref_locked;
1419 				}
1420 			}
1421 		}
1422 		/*
1423 		 * Check for conflicting IP addresses.  We permit them
1424 		 * if there is no more than one IP on each jail.  If
1425 		 * there is a duplicate on a jail with more than one
1426 		 * IP stop checking and return error.
1427 		 */
1428 		tppr = ppr;
1429 #ifdef VIMAGE
1430 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1431 			if (tppr->pr_flags & PR_VNET)
1432 				break;
1433 #endif
1434 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1435 			if (tpr == pr ||
1436 #ifdef VIMAGE
1437 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1438 #endif
1439 			    tpr->pr_uref == 0) {
1440 				descend = 0;
1441 				continue;
1442 			}
1443 			if (!(tpr->pr_flags & PR_IP4_USER))
1444 				continue;
1445 			descend = 0;
1446 			if (tpr->pr_ip4 == NULL ||
1447 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1448 				continue;
1449 			for (ii = 0; ii < ip4s; ii++) {
1450 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1451 					error = EADDRINUSE;
1452 					vfs_opterror(opts,
1453 					    "IPv4 addresses clash");
1454 					goto done_deref_locked;
1455 				}
1456 			}
1457 		}
1458 	}
1459 #endif
1460 #ifdef INET6
1461 	if (ip6s > 0) {
1462 		if (ppr->pr_flags & PR_IP6) {
1463 			/*
1464 			 * Make sure the new set of IP addresses is a
1465 			 * subset of the parent's list.
1466 			 */
1467 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1468 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1469 				    &ppr->pr_ip6[ij]))
1470 					break;
1471 			if (ij == ppr->pr_ip6s) {
1472 				error = EPERM;
1473 				goto done_deref_locked;
1474 			}
1475 			if (ip6s > 1) {
1476 				for (ii = ij = 1; ii < ip6s; ii++) {
1477 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1478 					     &ppr->pr_ip6[0]))
1479 						continue;
1480 					for (; ij < ppr->pr_ip6s; ij++)
1481 						if (IN6_ARE_ADDR_EQUAL(
1482 						    &ip6[ii], &ppr->pr_ip6[ij]))
1483 							break;
1484 					if (ij == ppr->pr_ip6s)
1485 						break;
1486 				}
1487 				if (ij == ppr->pr_ip6s) {
1488 					error = EPERM;
1489 					goto done_deref_locked;
1490 				}
1491 			}
1492 		}
1493 		/* Check for conflicting IP addresses. */
1494 		tppr = ppr;
1495 #ifdef VIMAGE
1496 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1497 			if (tppr->pr_flags & PR_VNET)
1498 				break;
1499 #endif
1500 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1501 			if (tpr == pr ||
1502 #ifdef VIMAGE
1503 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1504 #endif
1505 			    tpr->pr_uref == 0) {
1506 				descend = 0;
1507 				continue;
1508 			}
1509 			if (!(tpr->pr_flags & PR_IP6_USER))
1510 				continue;
1511 			descend = 0;
1512 			if (tpr->pr_ip6 == NULL ||
1513 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1514 				continue;
1515 			for (ii = 0; ii < ip6s; ii++) {
1516 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1517 					error = EADDRINUSE;
1518 					vfs_opterror(opts,
1519 					    "IPv6 addresses clash");
1520 					goto done_deref_locked;
1521 				}
1522 			}
1523 		}
1524 	}
1525 #endif
1526 	onamelen = namelen = 0;
1527 	if (name != NULL) {
1528 		/* Give a default name of the jid. */
1529 		if (name[0] == '\0')
1530 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1531 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1532 		    *p == '\0')) {
1533 			error = EINVAL;
1534 			vfs_opterror(opts,
1535 			    "name cannot be numeric (unless it is the jid)");
1536 			goto done_deref_locked;
1537 		}
1538 		/*
1539 		 * Make sure the name isn't too long for the prison or its
1540 		 * children.
1541 		 */
1542 		onamelen = strlen(pr->pr_name);
1543 		namelen = strlen(name);
1544 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1545 			error = ENAMETOOLONG;
1546 			goto done_deref_locked;
1547 		}
1548 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1549 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1550 			    sizeof(pr->pr_name)) {
1551 				error = ENAMETOOLONG;
1552 				goto done_deref_locked;
1553 			}
1554 		}
1555 	}
1556 	if (pr_allow & ~ppr->pr_allow) {
1557 		error = EPERM;
1558 		goto done_deref_locked;
1559 	}
1560 
1561 	/* Set the parameters of the prison. */
1562 #ifdef INET
1563 	redo_ip4 = 0;
1564 	if (pr_flags & PR_IP4_USER) {
1565 		pr->pr_flags |= PR_IP4;
1566 		free(pr->pr_ip4, M_PRISON);
1567 		pr->pr_ip4s = ip4s;
1568 		pr->pr_ip4 = ip4;
1569 		ip4 = NULL;
1570 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1571 #ifdef VIMAGE
1572 			if (tpr->pr_flags & PR_VNET) {
1573 				descend = 0;
1574 				continue;
1575 			}
1576 #endif
1577 			if (prison_restrict_ip4(tpr, NULL)) {
1578 				redo_ip4 = 1;
1579 				descend = 0;
1580 			}
1581 		}
1582 	}
1583 #endif
1584 #ifdef INET6
1585 	redo_ip6 = 0;
1586 	if (pr_flags & PR_IP6_USER) {
1587 		pr->pr_flags |= PR_IP6;
1588 		free(pr->pr_ip6, M_PRISON);
1589 		pr->pr_ip6s = ip6s;
1590 		pr->pr_ip6 = ip6;
1591 		ip6 = NULL;
1592 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1593 #ifdef VIMAGE
1594 			if (tpr->pr_flags & PR_VNET) {
1595 				descend = 0;
1596 				continue;
1597 			}
1598 #endif
1599 			if (prison_restrict_ip6(tpr, NULL)) {
1600 				redo_ip6 = 1;
1601 				descend = 0;
1602 			}
1603 		}
1604 	}
1605 #endif
1606 	if (gotslevel) {
1607 		pr->pr_securelevel = slevel;
1608 		/* Set all child jails to be at least this level. */
1609 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1610 			if (tpr->pr_securelevel < slevel)
1611 				tpr->pr_securelevel = slevel;
1612 	}
1613 	if (gotchildmax) {
1614 		pr->pr_childmax = childmax;
1615 		/* Set all child jails to under this limit. */
1616 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1617 			if (tpr->pr_childmax > childmax - level)
1618 				tpr->pr_childmax = childmax > level
1619 				    ? childmax - level : 0;
1620 	}
1621 	if (gotenforce) {
1622 		pr->pr_enforce_statfs = enforce;
1623 		/* Pass this restriction on to the children. */
1624 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1625 			if (tpr->pr_enforce_statfs < enforce)
1626 				tpr->pr_enforce_statfs = enforce;
1627 	}
1628 	if (gotrsnum) {
1629 		pr->pr_devfs_rsnum = rsnum;
1630 		/* Pass this restriction on to the children. */
1631 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1632 			tpr->pr_devfs_rsnum = rsnum;
1633 	}
1634 	if (name != NULL) {
1635 		if (ppr == &prison0)
1636 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1637 		else
1638 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1639 			    ppr->pr_name, name);
1640 		/* Change this component of child names. */
1641 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1642 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1643 			    strlen(tpr->pr_name + onamelen) + 1);
1644 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1645 		}
1646 	}
1647 	if (path != NULL) {
1648 		/* Try to keep a real-rooted full pathname. */
1649 		if (fullpath_disabled && path[0] == '/' &&
1650 		    strcmp(mypr->pr_path, "/"))
1651 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1652 			    mypr->pr_path, path);
1653 		else
1654 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1655 		pr->pr_root = root;
1656 	}
1657 	if (PR_HOST & ch_flags & ~pr_flags) {
1658 		if (pr->pr_flags & PR_HOST) {
1659 			/*
1660 			 * Copy the parent's host info.  As with pr_ip4 above,
1661 			 * the lack of a lock on the parent is not a problem;
1662 			 * it is always set with allprison_lock at least
1663 			 * shared, and is held exclusively here.
1664 			 */
1665 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1666 			    sizeof(pr->pr_hostname));
1667 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1668 			    sizeof(pr->pr_domainname));
1669 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1670 			    sizeof(pr->pr_hostuuid));
1671 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1672 		}
1673 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1674 		/* Set this prison, and any descendants without PR_HOST. */
1675 		if (host != NULL)
1676 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1677 		if (domain != NULL)
1678 			strlcpy(pr->pr_domainname, domain,
1679 			    sizeof(pr->pr_domainname));
1680 		if (uuid != NULL)
1681 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1682 		if (gothid)
1683 			pr->pr_hostid = hid;
1684 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1685 			if (tpr->pr_flags & PR_HOST)
1686 				descend = 0;
1687 			else {
1688 				if (host != NULL)
1689 					strlcpy(tpr->pr_hostname,
1690 					    pr->pr_hostname,
1691 					    sizeof(tpr->pr_hostname));
1692 				if (domain != NULL)
1693 					strlcpy(tpr->pr_domainname,
1694 					    pr->pr_domainname,
1695 					    sizeof(tpr->pr_domainname));
1696 				if (uuid != NULL)
1697 					strlcpy(tpr->pr_hostuuid,
1698 					    pr->pr_hostuuid,
1699 					    sizeof(tpr->pr_hostuuid));
1700 				if (gothid)
1701 					tpr->pr_hostid = hid;
1702 			}
1703 		}
1704 	}
1705 	if ((tallow = ch_allow & ~pr_allow)) {
1706 		/* Clear allow bits in all children. */
1707 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1708 			tpr->pr_allow &= ~tallow;
1709 	}
1710 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1711 	/*
1712 	 * Persistent prisons get an extra reference, and prisons losing their
1713 	 * persist flag lose that reference.  Only do this for existing prisons
1714 	 * for now, so new ones will remain unseen until after the module
1715 	 * handlers have completed.
1716 	 */
1717 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1718 		if (pr_flags & PR_PERSIST) {
1719 			pr->pr_ref++;
1720 			pr->pr_uref++;
1721 		} else {
1722 			pr->pr_ref--;
1723 			pr->pr_uref--;
1724 		}
1725 	}
1726 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1727 	mtx_unlock(&pr->pr_mtx);
1728 
1729 #ifdef RACCT
1730 	if (created)
1731 		prison_racct_attach(pr);
1732 #endif
1733 
1734 	/* Locks may have prevented a complete restriction of child IP
1735 	 * addresses.  If so, allocate some more memory and try again.
1736 	 */
1737 #ifdef INET
1738 	while (redo_ip4) {
1739 		ip4s = pr->pr_ip4s;
1740 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1741 		mtx_lock(&pr->pr_mtx);
1742 		redo_ip4 = 0;
1743 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1744 #ifdef VIMAGE
1745 			if (tpr->pr_flags & PR_VNET) {
1746 				descend = 0;
1747 				continue;
1748 			}
1749 #endif
1750 			if (prison_restrict_ip4(tpr, ip4)) {
1751 				if (ip4 != NULL)
1752 					ip4 = NULL;
1753 				else
1754 					redo_ip4 = 1;
1755 			}
1756 		}
1757 		mtx_unlock(&pr->pr_mtx);
1758 	}
1759 #endif
1760 #ifdef INET6
1761 	while (redo_ip6) {
1762 		ip6s = pr->pr_ip6s;
1763 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1764 		mtx_lock(&pr->pr_mtx);
1765 		redo_ip6 = 0;
1766 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1767 #ifdef VIMAGE
1768 			if (tpr->pr_flags & PR_VNET) {
1769 				descend = 0;
1770 				continue;
1771 			}
1772 #endif
1773 			if (prison_restrict_ip6(tpr, ip6)) {
1774 				if (ip6 != NULL)
1775 					ip6 = NULL;
1776 				else
1777 					redo_ip6 = 1;
1778 			}
1779 		}
1780 		mtx_unlock(&pr->pr_mtx);
1781 	}
1782 #endif
1783 
1784 	/* Let the modules do their work. */
1785 	sx_downgrade(&allprison_lock);
1786 	if (created) {
1787 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1788 		if (error) {
1789 			prison_deref(pr, PD_LIST_SLOCKED);
1790 			goto done_errmsg;
1791 		}
1792 	}
1793 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1794 	if (error) {
1795 		prison_deref(pr, created
1796 		    ? PD_LIST_SLOCKED
1797 		    : PD_DEREF | PD_LIST_SLOCKED);
1798 		goto done_errmsg;
1799 	}
1800 
1801 	/* Attach this process to the prison if requested. */
1802 	if (flags & JAIL_ATTACH) {
1803 		mtx_lock(&pr->pr_mtx);
1804 		error = do_jail_attach(td, pr);
1805 		if (error) {
1806 			vfs_opterror(opts, "attach failed");
1807 			if (!created)
1808 				prison_deref(pr, PD_DEREF);
1809 			goto done_errmsg;
1810 		}
1811 	}
1812 
1813 #ifdef RACCT
1814 	if (!created) {
1815 		if (!(flags & JAIL_ATTACH))
1816 			sx_sunlock(&allprison_lock);
1817 		prison_racct_modify(pr);
1818 		if (!(flags & JAIL_ATTACH))
1819 			sx_slock(&allprison_lock);
1820 	}
1821 #endif
1822 
1823 	td->td_retval[0] = pr->pr_id;
1824 
1825 	/*
1826 	 * Now that it is all there, drop the temporary reference from existing
1827 	 * prisons.  Or add a reference to newly created persistent prisons
1828 	 * (which was not done earlier so that the prison would not be publicly
1829 	 * visible).
1830 	 */
1831 	if (!created) {
1832 		prison_deref(pr, (flags & JAIL_ATTACH)
1833 		    ? PD_DEREF
1834 		    : PD_DEREF | PD_LIST_SLOCKED);
1835 	} else {
1836 		if (pr_flags & PR_PERSIST) {
1837 			mtx_lock(&pr->pr_mtx);
1838 			pr->pr_ref++;
1839 			pr->pr_uref++;
1840 			mtx_unlock(&pr->pr_mtx);
1841 		}
1842 		if (!(flags & JAIL_ATTACH))
1843 			sx_sunlock(&allprison_lock);
1844 	}
1845 
1846 	goto done_errmsg;
1847 
1848  done_deref_locked:
1849 	prison_deref(pr, created
1850 	    ? PD_LOCKED | PD_LIST_XLOCKED
1851 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1852 	goto done_releroot;
1853  done_unlock_list:
1854 	sx_xunlock(&allprison_lock);
1855  done_releroot:
1856 	if (root != NULL)
1857 		vrele(root);
1858  done_errmsg:
1859 	if (error) {
1860 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1861 		if (errmsg_len > 0) {
1862 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1863 			if (errmsg_pos > 0) {
1864 				if (optuio->uio_segflg == UIO_SYSSPACE)
1865 					bcopy(errmsg,
1866 					   optuio->uio_iov[errmsg_pos].iov_base,
1867 					   errmsg_len);
1868 				else
1869 					copyout(errmsg,
1870 					   optuio->uio_iov[errmsg_pos].iov_base,
1871 					   errmsg_len);
1872 			}
1873 		}
1874 	}
1875  done_free:
1876 #ifdef INET
1877 	free(ip4, M_PRISON);
1878 #endif
1879 #ifdef INET6
1880 	free(ip6, M_PRISON);
1881 #endif
1882 	if (g_path != NULL)
1883 		free(g_path, M_TEMP);
1884 	vfs_freeopts(opts);
1885 	return (error);
1886 }
1887 
1888 
1889 /*
1890  * struct jail_get_args {
1891  *	struct iovec *iovp;
1892  *	unsigned int iovcnt;
1893  *	int flags;
1894  * };
1895  */
1896 int
1897 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1898 {
1899 	struct uio *auio;
1900 	int error;
1901 
1902 	/* Check that we have an even number of iovecs. */
1903 	if (uap->iovcnt & 1)
1904 		return (EINVAL);
1905 
1906 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1907 	if (error)
1908 		return (error);
1909 	error = kern_jail_get(td, auio, uap->flags);
1910 	if (error == 0)
1911 		error = copyout(auio->uio_iov, uap->iovp,
1912 		    uap->iovcnt * sizeof (struct iovec));
1913 	free(auio, M_IOV);
1914 	return (error);
1915 }
1916 
1917 int
1918 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1919 {
1920 	struct prison *pr, *mypr;
1921 	struct vfsopt *opt;
1922 	struct vfsoptlist *opts;
1923 	char *errmsg, *name;
1924 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1925 
1926 	if (flags & ~JAIL_GET_MASK)
1927 		return (EINVAL);
1928 
1929 	/* Get the parameter list. */
1930 	error = vfs_buildopts(optuio, &opts);
1931 	if (error)
1932 		return (error);
1933 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1934 	mypr = td->td_ucred->cr_prison;
1935 
1936 	/*
1937 	 * Find the prison specified by one of: lastjid, jid, name.
1938 	 */
1939 	sx_slock(&allprison_lock);
1940 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1941 	if (error == 0) {
1942 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1943 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1944 				mtx_lock(&pr->pr_mtx);
1945 				if (pr->pr_ref > 0 &&
1946 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1947 					break;
1948 				mtx_unlock(&pr->pr_mtx);
1949 			}
1950 		}
1951 		if (pr != NULL)
1952 			goto found_prison;
1953 		error = ENOENT;
1954 		vfs_opterror(opts, "no jail after %d", jid);
1955 		goto done_unlock_list;
1956 	} else if (error != ENOENT)
1957 		goto done_unlock_list;
1958 
1959 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1960 	if (error == 0) {
1961 		if (jid != 0) {
1962 			pr = prison_find_child(mypr, jid);
1963 			if (pr != NULL) {
1964 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1965 					mtx_unlock(&pr->pr_mtx);
1966 					error = ENOENT;
1967 					vfs_opterror(opts, "jail %d is dying",
1968 					    jid);
1969 					goto done_unlock_list;
1970 				}
1971 				goto found_prison;
1972 			}
1973 			error = ENOENT;
1974 			vfs_opterror(opts, "jail %d not found", jid);
1975 			goto done_unlock_list;
1976 		}
1977 	} else if (error != ENOENT)
1978 		goto done_unlock_list;
1979 
1980 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1981 	if (error == 0) {
1982 		if (len == 0 || name[len - 1] != '\0') {
1983 			error = EINVAL;
1984 			goto done_unlock_list;
1985 		}
1986 		pr = prison_find_name(mypr, name);
1987 		if (pr != NULL) {
1988 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1989 				mtx_unlock(&pr->pr_mtx);
1990 				error = ENOENT;
1991 				vfs_opterror(opts, "jail \"%s\" is dying",
1992 				    name);
1993 				goto done_unlock_list;
1994 			}
1995 			goto found_prison;
1996 		}
1997 		error = ENOENT;
1998 		vfs_opterror(opts, "jail \"%s\" not found", name);
1999 		goto done_unlock_list;
2000 	} else if (error != ENOENT)
2001 		goto done_unlock_list;
2002 
2003 	vfs_opterror(opts, "no jail specified");
2004 	error = ENOENT;
2005 	goto done_unlock_list;
2006 
2007  found_prison:
2008 	/* Get the parameters of the prison. */
2009 	pr->pr_ref++;
2010 	locked = PD_LOCKED;
2011 	td->td_retval[0] = pr->pr_id;
2012 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2013 	if (error != 0 && error != ENOENT)
2014 		goto done_deref;
2015 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2016 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2017 	if (error != 0 && error != ENOENT)
2018 		goto done_deref;
2019 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2020 	if (error != 0 && error != ENOENT)
2021 		goto done_deref;
2022 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2023 	    sizeof(pr->pr_cpuset->cs_id));
2024 	if (error != 0 && error != ENOENT)
2025 		goto done_deref;
2026 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2027 	if (error != 0 && error != ENOENT)
2028 		goto done_deref;
2029 #ifdef INET
2030 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2031 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2032 	if (error != 0 && error != ENOENT)
2033 		goto done_deref;
2034 #endif
2035 #ifdef INET6
2036 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2037 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2038 	if (error != 0 && error != ENOENT)
2039 		goto done_deref;
2040 #endif
2041 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2042 	    sizeof(pr->pr_securelevel));
2043 	if (error != 0 && error != ENOENT)
2044 		goto done_deref;
2045 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2046 	    sizeof(pr->pr_childcount));
2047 	if (error != 0 && error != ENOENT)
2048 		goto done_deref;
2049 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2050 	    sizeof(pr->pr_childmax));
2051 	if (error != 0 && error != ENOENT)
2052 		goto done_deref;
2053 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2054 	if (error != 0 && error != ENOENT)
2055 		goto done_deref;
2056 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2057 	if (error != 0 && error != ENOENT)
2058 		goto done_deref;
2059 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2060 	if (error != 0 && error != ENOENT)
2061 		goto done_deref;
2062 #ifdef COMPAT_FREEBSD32
2063 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2064 		uint32_t hid32 = pr->pr_hostid;
2065 
2066 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2067 	} else
2068 #endif
2069 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2070 	    sizeof(pr->pr_hostid));
2071 	if (error != 0 && error != ENOENT)
2072 		goto done_deref;
2073 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2074 	    sizeof(pr->pr_enforce_statfs));
2075 	if (error != 0 && error != ENOENT)
2076 		goto done_deref;
2077 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2078 	    sizeof(pr->pr_devfs_rsnum));
2079 	if (error != 0 && error != ENOENT)
2080 		goto done_deref;
2081 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2082 	    fi++) {
2083 		if (pr_flag_names[fi] == NULL)
2084 			continue;
2085 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2086 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2087 		if (error != 0 && error != ENOENT)
2088 			goto done_deref;
2089 		i = !i;
2090 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2091 		if (error != 0 && error != ENOENT)
2092 			goto done_deref;
2093 	}
2094 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2095 	    fi++) {
2096 		i = pr->pr_flags &
2097 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2098 		i = pr_flag_jailsys[fi].disable &&
2099 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2100 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2101 		    : JAIL_SYS_INHERIT;
2102 		error =
2103 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2104 		if (error != 0 && error != ENOENT)
2105 			goto done_deref;
2106 	}
2107 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2108 	    fi++) {
2109 		if (pr_allow_names[fi] == NULL)
2110 			continue;
2111 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2112 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2113 		if (error != 0 && error != ENOENT)
2114 			goto done_deref;
2115 		i = !i;
2116 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2117 		if (error != 0 && error != ENOENT)
2118 			goto done_deref;
2119 	}
2120 	i = (pr->pr_uref == 0);
2121 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 	i = !i;
2125 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2126 	if (error != 0 && error != ENOENT)
2127 		goto done_deref;
2128 
2129 	/* Get the module parameters. */
2130 	mtx_unlock(&pr->pr_mtx);
2131 	locked = 0;
2132 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2133 	if (error)
2134 		goto done_deref;
2135 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2136 
2137 	/* By now, all parameters should have been noted. */
2138 	TAILQ_FOREACH(opt, opts, link) {
2139 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2140 			error = EINVAL;
2141 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2142 			goto done_errmsg;
2143 		}
2144 	}
2145 
2146 	/* Write the fetched parameters back to userspace. */
2147 	error = 0;
2148 	TAILQ_FOREACH(opt, opts, link) {
2149 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2150 			pos = 2 * opt->pos + 1;
2151 			optuio->uio_iov[pos].iov_len = opt->len;
2152 			if (opt->value != NULL) {
2153 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2154 					bcopy(opt->value,
2155 					    optuio->uio_iov[pos].iov_base,
2156 					    opt->len);
2157 				} else {
2158 					error = copyout(opt->value,
2159 					    optuio->uio_iov[pos].iov_base,
2160 					    opt->len);
2161 					if (error)
2162 						break;
2163 				}
2164 			}
2165 		}
2166 	}
2167 	goto done_errmsg;
2168 
2169  done_deref:
2170 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2171 	goto done_errmsg;
2172 
2173  done_unlock_list:
2174 	sx_sunlock(&allprison_lock);
2175  done_errmsg:
2176 	if (error && errmsg_pos >= 0) {
2177 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2178 		errmsg_pos = 2 * errmsg_pos + 1;
2179 		if (errmsg_len > 0) {
2180 			if (optuio->uio_segflg == UIO_SYSSPACE)
2181 				bcopy(errmsg,
2182 				    optuio->uio_iov[errmsg_pos].iov_base,
2183 				    errmsg_len);
2184 			else
2185 				copyout(errmsg,
2186 				    optuio->uio_iov[errmsg_pos].iov_base,
2187 				    errmsg_len);
2188 		}
2189 	}
2190 	vfs_freeopts(opts);
2191 	return (error);
2192 }
2193 
2194 
2195 /*
2196  * struct jail_remove_args {
2197  *	int jid;
2198  * };
2199  */
2200 int
2201 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2202 {
2203 	struct prison *pr, *cpr, *lpr, *tpr;
2204 	int descend, error;
2205 
2206 	error = priv_check(td, PRIV_JAIL_REMOVE);
2207 	if (error)
2208 		return (error);
2209 
2210 	sx_xlock(&allprison_lock);
2211 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2212 	if (pr == NULL) {
2213 		sx_xunlock(&allprison_lock);
2214 		return (EINVAL);
2215 	}
2216 
2217 	/* Remove all descendants of this prison, then remove this prison. */
2218 	pr->pr_ref++;
2219 	pr->pr_flags |= PR_REMOVE;
2220 	if (!LIST_EMPTY(&pr->pr_children)) {
2221 		mtx_unlock(&pr->pr_mtx);
2222 		lpr = NULL;
2223 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2224 			mtx_lock(&cpr->pr_mtx);
2225 			if (cpr->pr_ref > 0) {
2226 				tpr = cpr;
2227 				cpr->pr_ref++;
2228 				cpr->pr_flags |= PR_REMOVE;
2229 			} else {
2230 				/* Already removed - do not do it again. */
2231 				tpr = NULL;
2232 			}
2233 			mtx_unlock(&cpr->pr_mtx);
2234 			if (lpr != NULL) {
2235 				mtx_lock(&lpr->pr_mtx);
2236 				prison_remove_one(lpr);
2237 				sx_xlock(&allprison_lock);
2238 			}
2239 			lpr = tpr;
2240 		}
2241 		if (lpr != NULL) {
2242 			mtx_lock(&lpr->pr_mtx);
2243 			prison_remove_one(lpr);
2244 			sx_xlock(&allprison_lock);
2245 		}
2246 		mtx_lock(&pr->pr_mtx);
2247 	}
2248 	prison_remove_one(pr);
2249 	return (0);
2250 }
2251 
2252 static void
2253 prison_remove_one(struct prison *pr)
2254 {
2255 	struct proc *p;
2256 	int deuref;
2257 
2258 	/* If the prison was persistent, it is not anymore. */
2259 	deuref = 0;
2260 	if (pr->pr_flags & PR_PERSIST) {
2261 		pr->pr_ref--;
2262 		deuref = PD_DEUREF;
2263 		pr->pr_flags &= ~PR_PERSIST;
2264 	}
2265 
2266 	/*
2267 	 * jail_remove added a reference.  If that's the only one, remove
2268 	 * the prison now.
2269 	 */
2270 	KASSERT(pr->pr_ref > 0,
2271 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2272 	if (pr->pr_ref == 1) {
2273 		prison_deref(pr,
2274 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2275 		return;
2276 	}
2277 
2278 	mtx_unlock(&pr->pr_mtx);
2279 	sx_xunlock(&allprison_lock);
2280 	/*
2281 	 * Kill all processes unfortunate enough to be attached to this prison.
2282 	 */
2283 	sx_slock(&allproc_lock);
2284 	LIST_FOREACH(p, &allproc, p_list) {
2285 		PROC_LOCK(p);
2286 		if (p->p_state != PRS_NEW && p->p_ucred &&
2287 		    p->p_ucred->cr_prison == pr)
2288 			kern_psignal(p, SIGKILL);
2289 		PROC_UNLOCK(p);
2290 	}
2291 	sx_sunlock(&allproc_lock);
2292 	/* Remove the temporary reference added by jail_remove. */
2293 	prison_deref(pr, deuref | PD_DEREF);
2294 }
2295 
2296 
2297 /*
2298  * struct jail_attach_args {
2299  *	int jid;
2300  * };
2301  */
2302 int
2303 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2304 {
2305 	struct prison *pr;
2306 	int error;
2307 
2308 	error = priv_check(td, PRIV_JAIL_ATTACH);
2309 	if (error)
2310 		return (error);
2311 
2312 	sx_slock(&allprison_lock);
2313 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2314 	if (pr == NULL) {
2315 		sx_sunlock(&allprison_lock);
2316 		return (EINVAL);
2317 	}
2318 
2319 	/*
2320 	 * Do not allow a process to attach to a prison that is not
2321 	 * considered to be "alive".
2322 	 */
2323 	if (pr->pr_uref == 0) {
2324 		mtx_unlock(&pr->pr_mtx);
2325 		sx_sunlock(&allprison_lock);
2326 		return (EINVAL);
2327 	}
2328 
2329 	return (do_jail_attach(td, pr));
2330 }
2331 
2332 static int
2333 do_jail_attach(struct thread *td, struct prison *pr)
2334 {
2335 	struct prison *ppr;
2336 	struct proc *p;
2337 	struct ucred *newcred, *oldcred;
2338 	int error;
2339 
2340 	/*
2341 	 * XXX: Note that there is a slight race here if two threads
2342 	 * in the same privileged process attempt to attach to two
2343 	 * different jails at the same time.  It is important for
2344 	 * user processes not to do this, or they might end up with
2345 	 * a process root from one prison, but attached to the jail
2346 	 * of another.
2347 	 */
2348 	pr->pr_ref++;
2349 	pr->pr_uref++;
2350 	mtx_unlock(&pr->pr_mtx);
2351 
2352 	/* Let modules do whatever they need to prepare for attaching. */
2353 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2354 	if (error) {
2355 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2356 		return (error);
2357 	}
2358 	sx_sunlock(&allprison_lock);
2359 
2360 	/*
2361 	 * Reparent the newly attached process to this jail.
2362 	 */
2363 	ppr = td->td_ucred->cr_prison;
2364 	p = td->td_proc;
2365 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2366 	if (error)
2367 		goto e_revert_osd;
2368 
2369 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2370 	if ((error = change_dir(pr->pr_root, td)) != 0)
2371 		goto e_unlock;
2372 #ifdef MAC
2373 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2374 		goto e_unlock;
2375 #endif
2376 	VOP_UNLOCK(pr->pr_root, 0);
2377 	if ((error = change_root(pr->pr_root, td)))
2378 		goto e_revert_osd;
2379 
2380 	newcred = crget();
2381 	PROC_LOCK(p);
2382 	oldcred = p->p_ucred;
2383 	setsugid(p);
2384 	crcopy(newcred, oldcred);
2385 	newcred->cr_prison = pr;
2386 	p->p_ucred = newcred;
2387 	PROC_UNLOCK(p);
2388 #ifdef RACCT
2389 	racct_proc_ucred_changed(p, oldcred, newcred);
2390 #endif
2391 	crfree(oldcred);
2392 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2393 	return (0);
2394  e_unlock:
2395 	VOP_UNLOCK(pr->pr_root, 0);
2396  e_revert_osd:
2397 	/* Tell modules this thread is still in its old jail after all. */
2398 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2399 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2400 	return (error);
2401 }
2402 
2403 
2404 /*
2405  * Returns a locked prison instance, or NULL on failure.
2406  */
2407 struct prison *
2408 prison_find(int prid)
2409 {
2410 	struct prison *pr;
2411 
2412 	sx_assert(&allprison_lock, SX_LOCKED);
2413 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2414 		if (pr->pr_id == prid) {
2415 			mtx_lock(&pr->pr_mtx);
2416 			if (pr->pr_ref > 0)
2417 				return (pr);
2418 			mtx_unlock(&pr->pr_mtx);
2419 		}
2420 	}
2421 	return (NULL);
2422 }
2423 
2424 /*
2425  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2426  */
2427 struct prison *
2428 prison_find_child(struct prison *mypr, int prid)
2429 {
2430 	struct prison *pr;
2431 	int descend;
2432 
2433 	sx_assert(&allprison_lock, SX_LOCKED);
2434 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2435 		if (pr->pr_id == prid) {
2436 			mtx_lock(&pr->pr_mtx);
2437 			if (pr->pr_ref > 0)
2438 				return (pr);
2439 			mtx_unlock(&pr->pr_mtx);
2440 		}
2441 	}
2442 	return (NULL);
2443 }
2444 
2445 /*
2446  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2447  */
2448 struct prison *
2449 prison_find_name(struct prison *mypr, const char *name)
2450 {
2451 	struct prison *pr, *deadpr;
2452 	size_t mylen;
2453 	int descend;
2454 
2455 	sx_assert(&allprison_lock, SX_LOCKED);
2456 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2457  again:
2458 	deadpr = NULL;
2459 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2460 		if (!strcmp(pr->pr_name + mylen, name)) {
2461 			mtx_lock(&pr->pr_mtx);
2462 			if (pr->pr_ref > 0) {
2463 				if (pr->pr_uref > 0)
2464 					return (pr);
2465 				deadpr = pr;
2466 			}
2467 			mtx_unlock(&pr->pr_mtx);
2468 		}
2469 	}
2470 	/* There was no valid prison - perhaps there was a dying one. */
2471 	if (deadpr != NULL) {
2472 		mtx_lock(&deadpr->pr_mtx);
2473 		if (deadpr->pr_ref == 0) {
2474 			mtx_unlock(&deadpr->pr_mtx);
2475 			goto again;
2476 		}
2477 	}
2478 	return (deadpr);
2479 }
2480 
2481 /*
2482  * See if a prison has the specific flag set.
2483  */
2484 int
2485 prison_flag(struct ucred *cred, unsigned flag)
2486 {
2487 
2488 	/* This is an atomic read, so no locking is necessary. */
2489 	return (cred->cr_prison->pr_flags & flag);
2490 }
2491 
2492 int
2493 prison_allow(struct ucred *cred, unsigned flag)
2494 {
2495 
2496 	/* This is an atomic read, so no locking is necessary. */
2497 	return (cred->cr_prison->pr_allow & flag);
2498 }
2499 
2500 /*
2501  * Remove a prison reference.  If that was the last reference, remove the
2502  * prison itself - but not in this context in case there are locks held.
2503  */
2504 void
2505 prison_free_locked(struct prison *pr)
2506 {
2507 
2508 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2509 	pr->pr_ref--;
2510 	if (pr->pr_ref == 0) {
2511 		mtx_unlock(&pr->pr_mtx);
2512 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2513 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2514 		return;
2515 	}
2516 	mtx_unlock(&pr->pr_mtx);
2517 }
2518 
2519 void
2520 prison_free(struct prison *pr)
2521 {
2522 
2523 	mtx_lock(&pr->pr_mtx);
2524 	prison_free_locked(pr);
2525 }
2526 
2527 static void
2528 prison_complete(void *context, int pending)
2529 {
2530 
2531 	prison_deref((struct prison *)context, 0);
2532 }
2533 
2534 /*
2535  * Remove a prison reference (usually).  This internal version assumes no
2536  * mutexes are held, except perhaps the prison itself.  If there are no more
2537  * references, release and delist the prison.  On completion, the prison lock
2538  * and the allprison lock are both unlocked.
2539  */
2540 static void
2541 prison_deref(struct prison *pr, int flags)
2542 {
2543 	struct prison *ppr, *tpr;
2544 
2545 	if (!(flags & PD_LOCKED))
2546 		mtx_lock(&pr->pr_mtx);
2547 	for (;;) {
2548 		if (flags & PD_DEUREF) {
2549 			pr->pr_uref--;
2550 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2551 		}
2552 		if (flags & PD_DEREF)
2553 			pr->pr_ref--;
2554 		/* If the prison still has references, nothing else to do. */
2555 		if (pr->pr_ref > 0) {
2556 			mtx_unlock(&pr->pr_mtx);
2557 			if (flags & PD_LIST_SLOCKED)
2558 				sx_sunlock(&allprison_lock);
2559 			else if (flags & PD_LIST_XLOCKED)
2560 				sx_xunlock(&allprison_lock);
2561 			return;
2562 		}
2563 
2564 		mtx_unlock(&pr->pr_mtx);
2565 		if (flags & PD_LIST_SLOCKED) {
2566 			if (!sx_try_upgrade(&allprison_lock)) {
2567 				sx_sunlock(&allprison_lock);
2568 				sx_xlock(&allprison_lock);
2569 			}
2570 		} else if (!(flags & PD_LIST_XLOCKED))
2571 			sx_xlock(&allprison_lock);
2572 
2573 		TAILQ_REMOVE(&allprison, pr, pr_list);
2574 		LIST_REMOVE(pr, pr_sibling);
2575 		ppr = pr->pr_parent;
2576 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2577 			tpr->pr_childcount--;
2578 		sx_xunlock(&allprison_lock);
2579 
2580 #ifdef VIMAGE
2581 		if (pr->pr_vnet != ppr->pr_vnet)
2582 			vnet_destroy(pr->pr_vnet);
2583 #endif
2584 		if (pr->pr_root != NULL)
2585 			vrele(pr->pr_root);
2586 		mtx_destroy(&pr->pr_mtx);
2587 #ifdef INET
2588 		free(pr->pr_ip4, M_PRISON);
2589 #endif
2590 #ifdef INET6
2591 		free(pr->pr_ip6, M_PRISON);
2592 #endif
2593 		if (pr->pr_cpuset != NULL)
2594 			cpuset_rel(pr->pr_cpuset);
2595 		osd_jail_exit(pr);
2596 #ifdef RACCT
2597 		prison_racct_detach(pr);
2598 #endif
2599 		free(pr, M_PRISON);
2600 
2601 		/* Removing a prison frees a reference on its parent. */
2602 		pr = ppr;
2603 		mtx_lock(&pr->pr_mtx);
2604 		flags = PD_DEREF | PD_DEUREF;
2605 	}
2606 }
2607 
2608 void
2609 prison_hold_locked(struct prison *pr)
2610 {
2611 
2612 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2613 	KASSERT(pr->pr_ref > 0,
2614 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2615 	pr->pr_ref++;
2616 }
2617 
2618 void
2619 prison_hold(struct prison *pr)
2620 {
2621 
2622 	mtx_lock(&pr->pr_mtx);
2623 	prison_hold_locked(pr);
2624 	mtx_unlock(&pr->pr_mtx);
2625 }
2626 
2627 void
2628 prison_proc_hold(struct prison *pr)
2629 {
2630 
2631 	mtx_lock(&pr->pr_mtx);
2632 	KASSERT(pr->pr_uref > 0,
2633 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2634 	pr->pr_uref++;
2635 	mtx_unlock(&pr->pr_mtx);
2636 }
2637 
2638 void
2639 prison_proc_free(struct prison *pr)
2640 {
2641 
2642 	mtx_lock(&pr->pr_mtx);
2643 	KASSERT(pr->pr_uref > 0,
2644 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2645 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2646 }
2647 
2648 
2649 #ifdef INET
2650 /*
2651  * Restrict a prison's IP address list with its parent's, possibly replacing
2652  * it.  Return true if the replacement buffer was used (or would have been).
2653  */
2654 static int
2655 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2656 {
2657 	int ii, ij, used;
2658 	struct prison *ppr;
2659 
2660 	ppr = pr->pr_parent;
2661 	if (!(pr->pr_flags & PR_IP4_USER)) {
2662 		/* This has no user settings, so just copy the parent's list. */
2663 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2664 			/*
2665 			 * There's no room for the parent's list.  Use the
2666 			 * new list buffer, which is assumed to be big enough
2667 			 * (if it was passed).  If there's no buffer, try to
2668 			 * allocate one.
2669 			 */
2670 			used = 1;
2671 			if (newip4 == NULL) {
2672 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2673 				    M_PRISON, M_NOWAIT);
2674 				if (newip4 != NULL)
2675 					used = 0;
2676 			}
2677 			if (newip4 != NULL) {
2678 				bcopy(ppr->pr_ip4, newip4,
2679 				    ppr->pr_ip4s * sizeof(*newip4));
2680 				free(pr->pr_ip4, M_PRISON);
2681 				pr->pr_ip4 = newip4;
2682 				pr->pr_ip4s = ppr->pr_ip4s;
2683 			}
2684 			return (used);
2685 		}
2686 		pr->pr_ip4s = ppr->pr_ip4s;
2687 		if (pr->pr_ip4s > 0)
2688 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2689 			    pr->pr_ip4s * sizeof(*newip4));
2690 		else if (pr->pr_ip4 != NULL) {
2691 			free(pr->pr_ip4, M_PRISON);
2692 			pr->pr_ip4 = NULL;
2693 		}
2694 	} else if (pr->pr_ip4s > 0) {
2695 		/* Remove addresses that aren't in the parent. */
2696 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2697 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2698 				break;
2699 		if (ij < ppr->pr_ip4s)
2700 			ii = 1;
2701 		else {
2702 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2703 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2704 			ii = 0;
2705 		}
2706 		for (ij = 1; ii < pr->pr_ip4s; ) {
2707 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2708 				ii++;
2709 				continue;
2710 			}
2711 			switch (ij >= ppr->pr_ip4s ? -1 :
2712 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2713 			case -1:
2714 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2715 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2716 				break;
2717 			case 0:
2718 				ii++;
2719 				ij++;
2720 				break;
2721 			case 1:
2722 				ij++;
2723 				break;
2724 			}
2725 		}
2726 		if (pr->pr_ip4s == 0) {
2727 			pr->pr_flags |= PR_IP4_DISABLE;
2728 			free(pr->pr_ip4, M_PRISON);
2729 			pr->pr_ip4 = NULL;
2730 		}
2731 	}
2732 	return (0);
2733 }
2734 
2735 /*
2736  * Pass back primary IPv4 address of this jail.
2737  *
2738  * If not restricted return success but do not alter the address.  Caller has
2739  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2740  *
2741  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2742  * Address returned in NBO.
2743  */
2744 int
2745 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2746 {
2747 	struct prison *pr;
2748 
2749 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2750 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2751 
2752 	pr = cred->cr_prison;
2753 	if (!(pr->pr_flags & PR_IP4))
2754 		return (0);
2755 	mtx_lock(&pr->pr_mtx);
2756 	if (!(pr->pr_flags & PR_IP4)) {
2757 		mtx_unlock(&pr->pr_mtx);
2758 		return (0);
2759 	}
2760 	if (pr->pr_ip4 == NULL) {
2761 		mtx_unlock(&pr->pr_mtx);
2762 		return (EAFNOSUPPORT);
2763 	}
2764 
2765 	ia->s_addr = pr->pr_ip4[0].s_addr;
2766 	mtx_unlock(&pr->pr_mtx);
2767 	return (0);
2768 }
2769 
2770 /*
2771  * Return 1 if we should do proper source address selection or are not jailed.
2772  * We will return 0 if we should bypass source address selection in favour
2773  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2774  * returned in NBO.
2775  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2776  */
2777 int
2778 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2779 {
2780 	struct prison *pr;
2781 	struct in_addr lia;
2782 	int error;
2783 
2784 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2785 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2786 
2787 	if (!jailed(cred))
2788 		return (1);
2789 
2790 	pr = cred->cr_prison;
2791 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2792 		return (1);
2793 
2794 	lia.s_addr = INADDR_ANY;
2795 	error = prison_get_ip4(cred, &lia);
2796 	if (error)
2797 		return (error);
2798 	if (lia.s_addr == INADDR_ANY)
2799 		return (1);
2800 
2801 	ia->s_addr = lia.s_addr;
2802 	return (0);
2803 }
2804 
2805 /*
2806  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2807  */
2808 int
2809 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2810 {
2811 
2812 	if (pr1 == pr2)
2813 		return (1);
2814 
2815 	/*
2816 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2817 	 * existing prisons.
2818 	 */
2819 	while (pr1 != &prison0 &&
2820 #ifdef VIMAGE
2821 	       !(pr1->pr_flags & PR_VNET) &&
2822 #endif
2823 	       !(pr1->pr_flags & PR_IP4_USER))
2824 		pr1 = pr1->pr_parent;
2825 	while (pr2 != &prison0 &&
2826 #ifdef VIMAGE
2827 	       !(pr2->pr_flags & PR_VNET) &&
2828 #endif
2829 	       !(pr2->pr_flags & PR_IP4_USER))
2830 		pr2 = pr2->pr_parent;
2831 	return (pr1 == pr2);
2832 }
2833 
2834 /*
2835  * Make sure our (source) address is set to something meaningful to this
2836  * jail.
2837  *
2838  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2839  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2840  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2841  */
2842 int
2843 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2844 {
2845 	struct prison *pr;
2846 	struct in_addr ia0;
2847 	int error;
2848 
2849 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2850 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2851 
2852 	pr = cred->cr_prison;
2853 	if (!(pr->pr_flags & PR_IP4))
2854 		return (0);
2855 	mtx_lock(&pr->pr_mtx);
2856 	if (!(pr->pr_flags & PR_IP4)) {
2857 		mtx_unlock(&pr->pr_mtx);
2858 		return (0);
2859 	}
2860 	if (pr->pr_ip4 == NULL) {
2861 		mtx_unlock(&pr->pr_mtx);
2862 		return (EAFNOSUPPORT);
2863 	}
2864 
2865 	ia0.s_addr = ntohl(ia->s_addr);
2866 	if (ia0.s_addr == INADDR_LOOPBACK) {
2867 		ia->s_addr = pr->pr_ip4[0].s_addr;
2868 		mtx_unlock(&pr->pr_mtx);
2869 		return (0);
2870 	}
2871 
2872 	if (ia0.s_addr == INADDR_ANY) {
2873 		/*
2874 		 * In case there is only 1 IPv4 address, bind directly.
2875 		 */
2876 		if (pr->pr_ip4s == 1)
2877 			ia->s_addr = pr->pr_ip4[0].s_addr;
2878 		mtx_unlock(&pr->pr_mtx);
2879 		return (0);
2880 	}
2881 
2882 	error = _prison_check_ip4(pr, ia);
2883 	mtx_unlock(&pr->pr_mtx);
2884 	return (error);
2885 }
2886 
2887 /*
2888  * Rewrite destination address in case we will connect to loopback address.
2889  *
2890  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2891  * Address passed in in NBO and returned in NBO.
2892  */
2893 int
2894 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2895 {
2896 	struct prison *pr;
2897 
2898 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2899 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2900 
2901 	pr = cred->cr_prison;
2902 	if (!(pr->pr_flags & PR_IP4))
2903 		return (0);
2904 	mtx_lock(&pr->pr_mtx);
2905 	if (!(pr->pr_flags & PR_IP4)) {
2906 		mtx_unlock(&pr->pr_mtx);
2907 		return (0);
2908 	}
2909 	if (pr->pr_ip4 == NULL) {
2910 		mtx_unlock(&pr->pr_mtx);
2911 		return (EAFNOSUPPORT);
2912 	}
2913 
2914 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2915 		ia->s_addr = pr->pr_ip4[0].s_addr;
2916 		mtx_unlock(&pr->pr_mtx);
2917 		return (0);
2918 	}
2919 
2920 	/*
2921 	 * Return success because nothing had to be changed.
2922 	 */
2923 	mtx_unlock(&pr->pr_mtx);
2924 	return (0);
2925 }
2926 
2927 /*
2928  * Check if given address belongs to the jail referenced by cred/prison.
2929  *
2930  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2931  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2932  * doesn't allow IPv4.  Address passed in in NBO.
2933  */
2934 static int
2935 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
2936 {
2937 	int i, a, z, d;
2938 
2939 	/*
2940 	 * Check the primary IP.
2941 	 */
2942 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2943 		return (0);
2944 
2945 	/*
2946 	 * All the other IPs are sorted so we can do a binary search.
2947 	 */
2948 	a = 0;
2949 	z = pr->pr_ip4s - 2;
2950 	while (a <= z) {
2951 		i = (a + z) / 2;
2952 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2953 		if (d > 0)
2954 			z = i - 1;
2955 		else if (d < 0)
2956 			a = i + 1;
2957 		else
2958 			return (0);
2959 	}
2960 
2961 	return (EADDRNOTAVAIL);
2962 }
2963 
2964 int
2965 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
2966 {
2967 	struct prison *pr;
2968 	int error;
2969 
2970 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2971 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2972 
2973 	pr = cred->cr_prison;
2974 	if (!(pr->pr_flags & PR_IP4))
2975 		return (0);
2976 	mtx_lock(&pr->pr_mtx);
2977 	if (!(pr->pr_flags & PR_IP4)) {
2978 		mtx_unlock(&pr->pr_mtx);
2979 		return (0);
2980 	}
2981 	if (pr->pr_ip4 == NULL) {
2982 		mtx_unlock(&pr->pr_mtx);
2983 		return (EAFNOSUPPORT);
2984 	}
2985 
2986 	error = _prison_check_ip4(pr, ia);
2987 	mtx_unlock(&pr->pr_mtx);
2988 	return (error);
2989 }
2990 #endif
2991 
2992 #ifdef INET6
2993 static int
2994 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2995 {
2996 	int ii, ij, used;
2997 	struct prison *ppr;
2998 
2999 	ppr = pr->pr_parent;
3000 	if (!(pr->pr_flags & PR_IP6_USER)) {
3001 		/* This has no user settings, so just copy the parent's list. */
3002 		if (pr->pr_ip6s < ppr->pr_ip6s) {
3003 			/*
3004 			 * There's no room for the parent's list.  Use the
3005 			 * new list buffer, which is assumed to be big enough
3006 			 * (if it was passed).  If there's no buffer, try to
3007 			 * allocate one.
3008 			 */
3009 			used = 1;
3010 			if (newip6 == NULL) {
3011 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3012 				    M_PRISON, M_NOWAIT);
3013 				if (newip6 != NULL)
3014 					used = 0;
3015 			}
3016 			if (newip6 != NULL) {
3017 				bcopy(ppr->pr_ip6, newip6,
3018 				    ppr->pr_ip6s * sizeof(*newip6));
3019 				free(pr->pr_ip6, M_PRISON);
3020 				pr->pr_ip6 = newip6;
3021 				pr->pr_ip6s = ppr->pr_ip6s;
3022 			}
3023 			return (used);
3024 		}
3025 		pr->pr_ip6s = ppr->pr_ip6s;
3026 		if (pr->pr_ip6s > 0)
3027 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3028 			    pr->pr_ip6s * sizeof(*newip6));
3029 		else if (pr->pr_ip6 != NULL) {
3030 			free(pr->pr_ip6, M_PRISON);
3031 			pr->pr_ip6 = NULL;
3032 		}
3033 	} else if (pr->pr_ip6s > 0) {
3034 		/* Remove addresses that aren't in the parent. */
3035 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3036 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3037 			    &ppr->pr_ip6[ij]))
3038 				break;
3039 		if (ij < ppr->pr_ip6s)
3040 			ii = 1;
3041 		else {
3042 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3043 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3044 			ii = 0;
3045 		}
3046 		for (ij = 1; ii < pr->pr_ip6s; ) {
3047 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3048 			    &ppr->pr_ip6[0])) {
3049 				ii++;
3050 				continue;
3051 			}
3052 			switch (ij >= ppr->pr_ip6s ? -1 :
3053 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3054 			case -1:
3055 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3056 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3057 				break;
3058 			case 0:
3059 				ii++;
3060 				ij++;
3061 				break;
3062 			case 1:
3063 				ij++;
3064 				break;
3065 			}
3066 		}
3067 		if (pr->pr_ip6s == 0) {
3068 			pr->pr_flags |= PR_IP6_DISABLE;
3069 			free(pr->pr_ip6, M_PRISON);
3070 			pr->pr_ip6 = NULL;
3071 		}
3072 	}
3073 	return 0;
3074 }
3075 
3076 /*
3077  * Pass back primary IPv6 address for this jail.
3078  *
3079  * If not restricted return success but do not alter the address.  Caller has
3080  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3081  *
3082  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3083  */
3084 int
3085 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3086 {
3087 	struct prison *pr;
3088 
3089 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3090 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3091 
3092 	pr = cred->cr_prison;
3093 	if (!(pr->pr_flags & PR_IP6))
3094 		return (0);
3095 	mtx_lock(&pr->pr_mtx);
3096 	if (!(pr->pr_flags & PR_IP6)) {
3097 		mtx_unlock(&pr->pr_mtx);
3098 		return (0);
3099 	}
3100 	if (pr->pr_ip6 == NULL) {
3101 		mtx_unlock(&pr->pr_mtx);
3102 		return (EAFNOSUPPORT);
3103 	}
3104 
3105 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3106 	mtx_unlock(&pr->pr_mtx);
3107 	return (0);
3108 }
3109 
3110 /*
3111  * Return 1 if we should do proper source address selection or are not jailed.
3112  * We will return 0 if we should bypass source address selection in favour
3113  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3114  * returned in NBO.
3115  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3116  */
3117 int
3118 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3119 {
3120 	struct prison *pr;
3121 	struct in6_addr lia6;
3122 	int error;
3123 
3124 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3125 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3126 
3127 	if (!jailed(cred))
3128 		return (1);
3129 
3130 	pr = cred->cr_prison;
3131 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3132 		return (1);
3133 
3134 	lia6 = in6addr_any;
3135 	error = prison_get_ip6(cred, &lia6);
3136 	if (error)
3137 		return (error);
3138 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3139 		return (1);
3140 
3141 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3142 	return (0);
3143 }
3144 
3145 /*
3146  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3147  */
3148 int
3149 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3150 {
3151 
3152 	if (pr1 == pr2)
3153 		return (1);
3154 
3155 	while (pr1 != &prison0 &&
3156 #ifdef VIMAGE
3157 	       !(pr1->pr_flags & PR_VNET) &&
3158 #endif
3159 	       !(pr1->pr_flags & PR_IP6_USER))
3160 		pr1 = pr1->pr_parent;
3161 	while (pr2 != &prison0 &&
3162 #ifdef VIMAGE
3163 	       !(pr2->pr_flags & PR_VNET) &&
3164 #endif
3165 	       !(pr2->pr_flags & PR_IP6_USER))
3166 		pr2 = pr2->pr_parent;
3167 	return (pr1 == pr2);
3168 }
3169 
3170 /*
3171  * Make sure our (source) address is set to something meaningful to this jail.
3172  *
3173  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3174  * when needed while binding.
3175  *
3176  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3177  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3178  * doesn't allow IPv6.
3179  */
3180 int
3181 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3182 {
3183 	struct prison *pr;
3184 	int error;
3185 
3186 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3187 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3188 
3189 	pr = cred->cr_prison;
3190 	if (!(pr->pr_flags & PR_IP6))
3191 		return (0);
3192 	mtx_lock(&pr->pr_mtx);
3193 	if (!(pr->pr_flags & PR_IP6)) {
3194 		mtx_unlock(&pr->pr_mtx);
3195 		return (0);
3196 	}
3197 	if (pr->pr_ip6 == NULL) {
3198 		mtx_unlock(&pr->pr_mtx);
3199 		return (EAFNOSUPPORT);
3200 	}
3201 
3202 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3203 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3204 		mtx_unlock(&pr->pr_mtx);
3205 		return (0);
3206 	}
3207 
3208 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3209 		/*
3210 		 * In case there is only 1 IPv6 address, and v6only is true,
3211 		 * then bind directly.
3212 		 */
3213 		if (v6only != 0 && pr->pr_ip6s == 1)
3214 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3215 		mtx_unlock(&pr->pr_mtx);
3216 		return (0);
3217 	}
3218 
3219 	error = _prison_check_ip6(pr, ia6);
3220 	mtx_unlock(&pr->pr_mtx);
3221 	return (error);
3222 }
3223 
3224 /*
3225  * Rewrite destination address in case we will connect to loopback address.
3226  *
3227  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3228  */
3229 int
3230 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3231 {
3232 	struct prison *pr;
3233 
3234 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3235 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3236 
3237 	pr = cred->cr_prison;
3238 	if (!(pr->pr_flags & PR_IP6))
3239 		return (0);
3240 	mtx_lock(&pr->pr_mtx);
3241 	if (!(pr->pr_flags & PR_IP6)) {
3242 		mtx_unlock(&pr->pr_mtx);
3243 		return (0);
3244 	}
3245 	if (pr->pr_ip6 == NULL) {
3246 		mtx_unlock(&pr->pr_mtx);
3247 		return (EAFNOSUPPORT);
3248 	}
3249 
3250 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3251 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3252 		mtx_unlock(&pr->pr_mtx);
3253 		return (0);
3254 	}
3255 
3256 	/*
3257 	 * Return success because nothing had to be changed.
3258 	 */
3259 	mtx_unlock(&pr->pr_mtx);
3260 	return (0);
3261 }
3262 
3263 /*
3264  * Check if given address belongs to the jail referenced by cred/prison.
3265  *
3266  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3267  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3268  * doesn't allow IPv6.
3269  */
3270 static int
3271 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3272 {
3273 	int i, a, z, d;
3274 
3275 	/*
3276 	 * Check the primary IP.
3277 	 */
3278 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3279 		return (0);
3280 
3281 	/*
3282 	 * All the other IPs are sorted so we can do a binary search.
3283 	 */
3284 	a = 0;
3285 	z = pr->pr_ip6s - 2;
3286 	while (a <= z) {
3287 		i = (a + z) / 2;
3288 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3289 		if (d > 0)
3290 			z = i - 1;
3291 		else if (d < 0)
3292 			a = i + 1;
3293 		else
3294 			return (0);
3295 	}
3296 
3297 	return (EADDRNOTAVAIL);
3298 }
3299 
3300 int
3301 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3302 {
3303 	struct prison *pr;
3304 	int error;
3305 
3306 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3307 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3308 
3309 	pr = cred->cr_prison;
3310 	if (!(pr->pr_flags & PR_IP6))
3311 		return (0);
3312 	mtx_lock(&pr->pr_mtx);
3313 	if (!(pr->pr_flags & PR_IP6)) {
3314 		mtx_unlock(&pr->pr_mtx);
3315 		return (0);
3316 	}
3317 	if (pr->pr_ip6 == NULL) {
3318 		mtx_unlock(&pr->pr_mtx);
3319 		return (EAFNOSUPPORT);
3320 	}
3321 
3322 	error = _prison_check_ip6(pr, ia6);
3323 	mtx_unlock(&pr->pr_mtx);
3324 	return (error);
3325 }
3326 #endif
3327 
3328 /*
3329  * Check if a jail supports the given address family.
3330  *
3331  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3332  * if not.
3333  */
3334 int
3335 prison_check_af(struct ucred *cred, int af)
3336 {
3337 	struct prison *pr;
3338 	int error;
3339 
3340 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3341 
3342 	pr = cred->cr_prison;
3343 #ifdef VIMAGE
3344 	/* Prisons with their own network stack are not limited. */
3345 	if (prison_owns_vnet(cred))
3346 		return (0);
3347 #endif
3348 
3349 	error = 0;
3350 	switch (af)
3351 	{
3352 #ifdef INET
3353 	case AF_INET:
3354 		if (pr->pr_flags & PR_IP4)
3355 		{
3356 			mtx_lock(&pr->pr_mtx);
3357 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3358 				error = EAFNOSUPPORT;
3359 			mtx_unlock(&pr->pr_mtx);
3360 		}
3361 		break;
3362 #endif
3363 #ifdef INET6
3364 	case AF_INET6:
3365 		if (pr->pr_flags & PR_IP6)
3366 		{
3367 			mtx_lock(&pr->pr_mtx);
3368 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3369 				error = EAFNOSUPPORT;
3370 			mtx_unlock(&pr->pr_mtx);
3371 		}
3372 		break;
3373 #endif
3374 	case AF_LOCAL:
3375 	case AF_ROUTE:
3376 		break;
3377 	default:
3378 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3379 			error = EAFNOSUPPORT;
3380 	}
3381 	return (error);
3382 }
3383 
3384 /*
3385  * Check if given address belongs to the jail referenced by cred (wrapper to
3386  * prison_check_ip[46]).
3387  *
3388  * Returns 0 if jail doesn't restrict the address family or if address belongs
3389  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3390  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3391  */
3392 int
3393 prison_if(struct ucred *cred, struct sockaddr *sa)
3394 {
3395 #ifdef INET
3396 	struct sockaddr_in *sai;
3397 #endif
3398 #ifdef INET6
3399 	struct sockaddr_in6 *sai6;
3400 #endif
3401 	int error;
3402 
3403 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3404 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3405 
3406 #ifdef VIMAGE
3407 	if (prison_owns_vnet(cred))
3408 		return (0);
3409 #endif
3410 
3411 	error = 0;
3412 	switch (sa->sa_family)
3413 	{
3414 #ifdef INET
3415 	case AF_INET:
3416 		sai = (struct sockaddr_in *)sa;
3417 		error = prison_check_ip4(cred, &sai->sin_addr);
3418 		break;
3419 #endif
3420 #ifdef INET6
3421 	case AF_INET6:
3422 		sai6 = (struct sockaddr_in6 *)sa;
3423 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3424 		break;
3425 #endif
3426 	default:
3427 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3428 			error = EAFNOSUPPORT;
3429 	}
3430 	return (error);
3431 }
3432 
3433 /*
3434  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3435  */
3436 int
3437 prison_check(struct ucred *cred1, struct ucred *cred2)
3438 {
3439 
3440 	return ((cred1->cr_prison == cred2->cr_prison ||
3441 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3442 }
3443 
3444 /*
3445  * Return 1 if p2 is a child of p1, otherwise 0.
3446  */
3447 int
3448 prison_ischild(struct prison *pr1, struct prison *pr2)
3449 {
3450 
3451 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3452 		if (pr1 == pr2)
3453 			return (1);
3454 	return (0);
3455 }
3456 
3457 /*
3458  * Return 1 if the passed credential is in a jail, otherwise 0.
3459  */
3460 int
3461 jailed(struct ucred *cred)
3462 {
3463 
3464 	return (cred->cr_prison != &prison0);
3465 }
3466 
3467 /*
3468  * Return 1 if the passed credential is in a jail and that jail does not
3469  * have its own virtual network stack, otherwise 0.
3470  */
3471 int
3472 jailed_without_vnet(struct ucred *cred)
3473 {
3474 
3475 	if (!jailed(cred))
3476 		return (0);
3477 #ifdef VIMAGE
3478 	if (prison_owns_vnet(cred))
3479 		return (0);
3480 #endif
3481 
3482 	return (1);
3483 }
3484 
3485 /*
3486  * Return the correct hostname (domainname, et al) for the passed credential.
3487  */
3488 void
3489 getcredhostname(struct ucred *cred, char *buf, size_t size)
3490 {
3491 	struct prison *pr;
3492 
3493 	/*
3494 	 * A NULL credential can be used to shortcut to the physical
3495 	 * system's hostname.
3496 	 */
3497 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3498 	mtx_lock(&pr->pr_mtx);
3499 	strlcpy(buf, pr->pr_hostname, size);
3500 	mtx_unlock(&pr->pr_mtx);
3501 }
3502 
3503 void
3504 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3505 {
3506 
3507 	mtx_lock(&cred->cr_prison->pr_mtx);
3508 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3509 	mtx_unlock(&cred->cr_prison->pr_mtx);
3510 }
3511 
3512 void
3513 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3514 {
3515 
3516 	mtx_lock(&cred->cr_prison->pr_mtx);
3517 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3518 	mtx_unlock(&cred->cr_prison->pr_mtx);
3519 }
3520 
3521 void
3522 getcredhostid(struct ucred *cred, unsigned long *hostid)
3523 {
3524 
3525 	mtx_lock(&cred->cr_prison->pr_mtx);
3526 	*hostid = cred->cr_prison->pr_hostid;
3527 	mtx_unlock(&cred->cr_prison->pr_mtx);
3528 }
3529 
3530 #ifdef VIMAGE
3531 /*
3532  * Determine whether the prison represented by cred owns
3533  * its vnet rather than having it inherited.
3534  *
3535  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3536  */
3537 int
3538 prison_owns_vnet(struct ucred *cred)
3539 {
3540 
3541 	/*
3542 	 * vnets cannot be added/removed after jail creation,
3543 	 * so no need to lock here.
3544 	 */
3545 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3546 }
3547 #endif
3548 
3549 /*
3550  * Determine whether the subject represented by cred can "see"
3551  * status of a mount point.
3552  * Returns: 0 for permitted, ENOENT otherwise.
3553  * XXX: This function should be called cr_canseemount() and should be
3554  *      placed in kern_prot.c.
3555  */
3556 int
3557 prison_canseemount(struct ucred *cred, struct mount *mp)
3558 {
3559 	struct prison *pr;
3560 	struct statfs *sp;
3561 	size_t len;
3562 
3563 	pr = cred->cr_prison;
3564 	if (pr->pr_enforce_statfs == 0)
3565 		return (0);
3566 	if (pr->pr_root->v_mount == mp)
3567 		return (0);
3568 	if (pr->pr_enforce_statfs == 2)
3569 		return (ENOENT);
3570 	/*
3571 	 * If jail's chroot directory is set to "/" we should be able to see
3572 	 * all mount-points from inside a jail.
3573 	 * This is ugly check, but this is the only situation when jail's
3574 	 * directory ends with '/'.
3575 	 */
3576 	if (strcmp(pr->pr_path, "/") == 0)
3577 		return (0);
3578 	len = strlen(pr->pr_path);
3579 	sp = &mp->mnt_stat;
3580 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3581 		return (ENOENT);
3582 	/*
3583 	 * Be sure that we don't have situation where jail's root directory
3584 	 * is "/some/path" and mount point is "/some/pathpath".
3585 	 */
3586 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3587 		return (ENOENT);
3588 	return (0);
3589 }
3590 
3591 void
3592 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3593 {
3594 	char jpath[MAXPATHLEN];
3595 	struct prison *pr;
3596 	size_t len;
3597 
3598 	pr = cred->cr_prison;
3599 	if (pr->pr_enforce_statfs == 0)
3600 		return;
3601 	if (prison_canseemount(cred, mp) != 0) {
3602 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3603 		strlcpy(sp->f_mntonname, "[restricted]",
3604 		    sizeof(sp->f_mntonname));
3605 		return;
3606 	}
3607 	if (pr->pr_root->v_mount == mp) {
3608 		/*
3609 		 * Clear current buffer data, so we are sure nothing from
3610 		 * the valid path left there.
3611 		 */
3612 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3613 		*sp->f_mntonname = '/';
3614 		return;
3615 	}
3616 	/*
3617 	 * If jail's chroot directory is set to "/" we should be able to see
3618 	 * all mount-points from inside a jail.
3619 	 */
3620 	if (strcmp(pr->pr_path, "/") == 0)
3621 		return;
3622 	len = strlen(pr->pr_path);
3623 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3624 	/*
3625 	 * Clear current buffer data, so we are sure nothing from
3626 	 * the valid path left there.
3627 	 */
3628 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3629 	if (*jpath == '\0') {
3630 		/* Should never happen. */
3631 		*sp->f_mntonname = '/';
3632 	} else {
3633 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3634 	}
3635 }
3636 
3637 /*
3638  * Check with permission for a specific privilege is granted within jail.  We
3639  * have a specific list of accepted privileges; the rest are denied.
3640  */
3641 int
3642 prison_priv_check(struct ucred *cred, int priv)
3643 {
3644 
3645 	if (!jailed(cred))
3646 		return (0);
3647 
3648 #ifdef VIMAGE
3649 	/*
3650 	 * Privileges specific to prisons with a virtual network stack.
3651 	 * There might be a duplicate entry here in case the privilege
3652 	 * is only granted conditionally in the legacy jail case.
3653 	 */
3654 	switch (priv) {
3655 #ifdef notyet
3656 		/*
3657 		 * NFS-specific privileges.
3658 		 */
3659 	case PRIV_NFS_DAEMON:
3660 	case PRIV_NFS_LOCKD:
3661 #endif
3662 		/*
3663 		 * Network stack privileges.
3664 		 */
3665 	case PRIV_NET_BRIDGE:
3666 	case PRIV_NET_GRE:
3667 	case PRIV_NET_BPF:
3668 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3669 	case PRIV_NET_ROUTE:
3670 	case PRIV_NET_TAP:
3671 	case PRIV_NET_SETIFMTU:
3672 	case PRIV_NET_SETIFFLAGS:
3673 	case PRIV_NET_SETIFCAP:
3674 	case PRIV_NET_SETIFDESCR:
3675 	case PRIV_NET_SETIFNAME	:
3676 	case PRIV_NET_SETIFMETRIC:
3677 	case PRIV_NET_SETIFPHYS:
3678 	case PRIV_NET_SETIFMAC:
3679 	case PRIV_NET_ADDMULTI:
3680 	case PRIV_NET_DELMULTI:
3681 	case PRIV_NET_HWIOCTL:
3682 	case PRIV_NET_SETLLADDR:
3683 	case PRIV_NET_ADDIFGROUP:
3684 	case PRIV_NET_DELIFGROUP:
3685 	case PRIV_NET_IFCREATE:
3686 	case PRIV_NET_IFDESTROY:
3687 	case PRIV_NET_ADDIFADDR:
3688 	case PRIV_NET_DELIFADDR:
3689 	case PRIV_NET_LAGG:
3690 	case PRIV_NET_GIF:
3691 	case PRIV_NET_SETIFVNET:
3692 	case PRIV_NET_SETIFFIB:
3693 
3694 		/*
3695 		 * 802.11-related privileges.
3696 		 */
3697 	case PRIV_NET80211_GETKEY:
3698 #ifdef notyet
3699 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3700 #endif
3701 
3702 #ifdef notyet
3703 		/*
3704 		 * ATM privileges.
3705 		 */
3706 	case PRIV_NETATM_CFG:
3707 	case PRIV_NETATM_ADD:
3708 	case PRIV_NETATM_DEL:
3709 	case PRIV_NETATM_SET:
3710 
3711 		/*
3712 		 * Bluetooth privileges.
3713 		 */
3714 	case PRIV_NETBLUETOOTH_RAW:
3715 #endif
3716 
3717 		/*
3718 		 * Netgraph and netgraph module privileges.
3719 		 */
3720 	case PRIV_NETGRAPH_CONTROL:
3721 #ifdef notyet
3722 	case PRIV_NETGRAPH_TTY:
3723 #endif
3724 
3725 		/*
3726 		 * IPv4 and IPv6 privileges.
3727 		 */
3728 	case PRIV_NETINET_IPFW:
3729 	case PRIV_NETINET_DIVERT:
3730 	case PRIV_NETINET_PF:
3731 	case PRIV_NETINET_DUMMYNET:
3732 	case PRIV_NETINET_CARP:
3733 	case PRIV_NETINET_MROUTE:
3734 	case PRIV_NETINET_RAW:
3735 	case PRIV_NETINET_ADDRCTRL6:
3736 	case PRIV_NETINET_ND6:
3737 	case PRIV_NETINET_SCOPE6:
3738 	case PRIV_NETINET_ALIFETIME6:
3739 	case PRIV_NETINET_IPSEC:
3740 	case PRIV_NETINET_BINDANY:
3741 
3742 #ifdef notyet
3743 		/*
3744 		 * NCP privileges.
3745 		 */
3746 	case PRIV_NETNCP:
3747 
3748 		/*
3749 		 * SMB privileges.
3750 		 */
3751 	case PRIV_NETSMB:
3752 #endif
3753 
3754 	/*
3755 	 * No default: or deny here.
3756 	 * In case of no permit fall through to next switch().
3757 	 */
3758 		if (cred->cr_prison->pr_flags & PR_VNET)
3759 			return (0);
3760 	}
3761 #endif /* VIMAGE */
3762 
3763 	switch (priv) {
3764 
3765 		/*
3766 		 * Allow ktrace privileges for root in jail.
3767 		 */
3768 	case PRIV_KTRACE:
3769 
3770 #if 0
3771 		/*
3772 		 * Allow jailed processes to configure audit identity and
3773 		 * submit audit records (login, etc).  In the future we may
3774 		 * want to further refine the relationship between audit and
3775 		 * jail.
3776 		 */
3777 	case PRIV_AUDIT_GETAUDIT:
3778 	case PRIV_AUDIT_SETAUDIT:
3779 	case PRIV_AUDIT_SUBMIT:
3780 #endif
3781 
3782 		/*
3783 		 * Allow jailed processes to manipulate process UNIX
3784 		 * credentials in any way they see fit.
3785 		 */
3786 	case PRIV_CRED_SETUID:
3787 	case PRIV_CRED_SETEUID:
3788 	case PRIV_CRED_SETGID:
3789 	case PRIV_CRED_SETEGID:
3790 	case PRIV_CRED_SETGROUPS:
3791 	case PRIV_CRED_SETREUID:
3792 	case PRIV_CRED_SETREGID:
3793 	case PRIV_CRED_SETRESUID:
3794 	case PRIV_CRED_SETRESGID:
3795 
3796 		/*
3797 		 * Jail implements visibility constraints already, so allow
3798 		 * jailed root to override uid/gid-based constraints.
3799 		 */
3800 	case PRIV_SEEOTHERGIDS:
3801 	case PRIV_SEEOTHERUIDS:
3802 
3803 		/*
3804 		 * Jail implements inter-process debugging limits already, so
3805 		 * allow jailed root various debugging privileges.
3806 		 */
3807 	case PRIV_DEBUG_DIFFCRED:
3808 	case PRIV_DEBUG_SUGID:
3809 	case PRIV_DEBUG_UNPRIV:
3810 
3811 		/*
3812 		 * Allow jail to set various resource limits and login
3813 		 * properties, and for now, exceed process resource limits.
3814 		 */
3815 	case PRIV_PROC_LIMIT:
3816 	case PRIV_PROC_SETLOGIN:
3817 	case PRIV_PROC_SETRLIMIT:
3818 
3819 		/*
3820 		 * System V and POSIX IPC privileges are granted in jail.
3821 		 */
3822 	case PRIV_IPC_READ:
3823 	case PRIV_IPC_WRITE:
3824 	case PRIV_IPC_ADMIN:
3825 	case PRIV_IPC_MSGSIZE:
3826 	case PRIV_MQ_ADMIN:
3827 
3828 		/*
3829 		 * Jail operations within a jail work on child jails.
3830 		 */
3831 	case PRIV_JAIL_ATTACH:
3832 	case PRIV_JAIL_SET:
3833 	case PRIV_JAIL_REMOVE:
3834 
3835 		/*
3836 		 * Jail implements its own inter-process limits, so allow
3837 		 * root processes in jail to change scheduling on other
3838 		 * processes in the same jail.  Likewise for signalling.
3839 		 */
3840 	case PRIV_SCHED_DIFFCRED:
3841 	case PRIV_SCHED_CPUSET:
3842 	case PRIV_SIGNAL_DIFFCRED:
3843 	case PRIV_SIGNAL_SUGID:
3844 
3845 		/*
3846 		 * Allow jailed processes to write to sysctls marked as jail
3847 		 * writable.
3848 		 */
3849 	case PRIV_SYSCTL_WRITEJAIL:
3850 
3851 		/*
3852 		 * Allow root in jail to manage a variety of quota
3853 		 * properties.  These should likely be conditional on a
3854 		 * configuration option.
3855 		 */
3856 	case PRIV_VFS_GETQUOTA:
3857 	case PRIV_VFS_SETQUOTA:
3858 
3859 		/*
3860 		 * Since Jail relies on chroot() to implement file system
3861 		 * protections, grant many VFS privileges to root in jail.
3862 		 * Be careful to exclude mount-related and NFS-related
3863 		 * privileges.
3864 		 */
3865 	case PRIV_VFS_READ:
3866 	case PRIV_VFS_WRITE:
3867 	case PRIV_VFS_ADMIN:
3868 	case PRIV_VFS_EXEC:
3869 	case PRIV_VFS_LOOKUP:
3870 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3871 	case PRIV_VFS_CHFLAGS_DEV:
3872 	case PRIV_VFS_CHOWN:
3873 	case PRIV_VFS_CHROOT:
3874 	case PRIV_VFS_RETAINSUGID:
3875 	case PRIV_VFS_FCHROOT:
3876 	case PRIV_VFS_LINK:
3877 	case PRIV_VFS_SETGID:
3878 	case PRIV_VFS_STAT:
3879 	case PRIV_VFS_STICKYFILE:
3880 
3881 		/*
3882 		 * As in the non-jail case, non-root users are expected to be
3883 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3884 		 * exists in the jail and they have permission to access it).
3885 		 */
3886 	case PRIV_KMEM_READ:
3887 		return (0);
3888 
3889 		/*
3890 		 * Depending on the global setting, allow privilege of
3891 		 * setting system flags.
3892 		 */
3893 	case PRIV_VFS_SYSFLAGS:
3894 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3895 			return (0);
3896 		else
3897 			return (EPERM);
3898 
3899 		/*
3900 		 * Depending on the global setting, allow privilege of
3901 		 * mounting/unmounting file systems.
3902 		 */
3903 	case PRIV_VFS_MOUNT:
3904 	case PRIV_VFS_UNMOUNT:
3905 	case PRIV_VFS_MOUNT_NONUSER:
3906 	case PRIV_VFS_MOUNT_OWNER:
3907 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3908 		    cred->cr_prison->pr_enforce_statfs < 2)
3909 			return (0);
3910 		else
3911 			return (EPERM);
3912 
3913 		/*
3914 		 * Allow jailed root to bind reserved ports and reuse in-use
3915 		 * ports.
3916 		 */
3917 	case PRIV_NETINET_RESERVEDPORT:
3918 	case PRIV_NETINET_REUSEPORT:
3919 		return (0);
3920 
3921 		/*
3922 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3923 		 */
3924 	case PRIV_NETINET_SETHDROPTS:
3925 		return (0);
3926 
3927 		/*
3928 		 * Conditionally allow creating raw sockets in jail.
3929 		 */
3930 	case PRIV_NETINET_RAW:
3931 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3932 			return (0);
3933 		else
3934 			return (EPERM);
3935 
3936 		/*
3937 		 * Since jail implements its own visibility limits on netstat
3938 		 * sysctls, allow getcred.  This allows identd to work in
3939 		 * jail.
3940 		 */
3941 	case PRIV_NETINET_GETCRED:
3942 		return (0);
3943 
3944 		/*
3945 		 * Allow jailed root to set loginclass.
3946 		 */
3947 	case PRIV_PROC_SETLOGINCLASS:
3948 		return (0);
3949 
3950 	default:
3951 		/*
3952 		 * In all remaining cases, deny the privilege request.  This
3953 		 * includes almost all network privileges, many system
3954 		 * configuration privileges.
3955 		 */
3956 		return (EPERM);
3957 	}
3958 }
3959 
3960 /*
3961  * Return the part of pr2's name that is relative to pr1, or the whole name
3962  * if it does not directly follow.
3963  */
3964 
3965 char *
3966 prison_name(struct prison *pr1, struct prison *pr2)
3967 {
3968 	char *name;
3969 
3970 	/* Jails see themselves as "0" (if they see themselves at all). */
3971 	if (pr1 == pr2)
3972 		return "0";
3973 	name = pr2->pr_name;
3974 	if (prison_ischild(pr1, pr2)) {
3975 		/*
3976 		 * pr1 isn't locked (and allprison_lock may not be either)
3977 		 * so its length can't be counted on.  But the number of dots
3978 		 * can be counted on - and counted.
3979 		 */
3980 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3981 			name = strchr(name, '.') + 1;
3982 	}
3983 	return (name);
3984 }
3985 
3986 /*
3987  * Return the part of pr2's path that is relative to pr1, or the whole path
3988  * if it does not directly follow.
3989  */
3990 static char *
3991 prison_path(struct prison *pr1, struct prison *pr2)
3992 {
3993 	char *path1, *path2;
3994 	int len1;
3995 
3996 	path1 = pr1->pr_path;
3997 	path2 = pr2->pr_path;
3998 	if (!strcmp(path1, "/"))
3999 		return (path2);
4000 	len1 = strlen(path1);
4001 	if (strncmp(path1, path2, len1))
4002 		return (path2);
4003 	if (path2[len1] == '\0')
4004 		return "/";
4005 	if (path2[len1] == '/')
4006 		return (path2 + len1);
4007 	return (path2);
4008 }
4009 
4010 
4011 /*
4012  * Jail-related sysctls.
4013  */
4014 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4015     "Jails");
4016 
4017 static int
4018 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4019 {
4020 	struct xprison *xp;
4021 	struct prison *pr, *cpr;
4022 #ifdef INET
4023 	struct in_addr *ip4 = NULL;
4024 	int ip4s = 0;
4025 #endif
4026 #ifdef INET6
4027 	struct in6_addr *ip6 = NULL;
4028 	int ip6s = 0;
4029 #endif
4030 	int descend, error;
4031 
4032 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4033 	pr = req->td->td_ucred->cr_prison;
4034 	error = 0;
4035 	sx_slock(&allprison_lock);
4036 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4037 #if defined(INET) || defined(INET6)
4038  again:
4039 #endif
4040 		mtx_lock(&cpr->pr_mtx);
4041 #ifdef INET
4042 		if (cpr->pr_ip4s > 0) {
4043 			if (ip4s < cpr->pr_ip4s) {
4044 				ip4s = cpr->pr_ip4s;
4045 				mtx_unlock(&cpr->pr_mtx);
4046 				ip4 = realloc(ip4, ip4s *
4047 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4048 				goto again;
4049 			}
4050 			bcopy(cpr->pr_ip4, ip4,
4051 			    cpr->pr_ip4s * sizeof(struct in_addr));
4052 		}
4053 #endif
4054 #ifdef INET6
4055 		if (cpr->pr_ip6s > 0) {
4056 			if (ip6s < cpr->pr_ip6s) {
4057 				ip6s = cpr->pr_ip6s;
4058 				mtx_unlock(&cpr->pr_mtx);
4059 				ip6 = realloc(ip6, ip6s *
4060 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4061 				goto again;
4062 			}
4063 			bcopy(cpr->pr_ip6, ip6,
4064 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4065 		}
4066 #endif
4067 		if (cpr->pr_ref == 0) {
4068 			mtx_unlock(&cpr->pr_mtx);
4069 			continue;
4070 		}
4071 		bzero(xp, sizeof(*xp));
4072 		xp->pr_version = XPRISON_VERSION;
4073 		xp->pr_id = cpr->pr_id;
4074 		xp->pr_state = cpr->pr_uref > 0
4075 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4076 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4077 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4078 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4079 #ifdef INET
4080 		xp->pr_ip4s = cpr->pr_ip4s;
4081 #endif
4082 #ifdef INET6
4083 		xp->pr_ip6s = cpr->pr_ip6s;
4084 #endif
4085 		mtx_unlock(&cpr->pr_mtx);
4086 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4087 		if (error)
4088 			break;
4089 #ifdef INET
4090 		if (xp->pr_ip4s > 0) {
4091 			error = SYSCTL_OUT(req, ip4,
4092 			    xp->pr_ip4s * sizeof(struct in_addr));
4093 			if (error)
4094 				break;
4095 		}
4096 #endif
4097 #ifdef INET6
4098 		if (xp->pr_ip6s > 0) {
4099 			error = SYSCTL_OUT(req, ip6,
4100 			    xp->pr_ip6s * sizeof(struct in6_addr));
4101 			if (error)
4102 				break;
4103 		}
4104 #endif
4105 	}
4106 	sx_sunlock(&allprison_lock);
4107 	free(xp, M_TEMP);
4108 #ifdef INET
4109 	free(ip4, M_TEMP);
4110 #endif
4111 #ifdef INET6
4112 	free(ip6, M_TEMP);
4113 #endif
4114 	return (error);
4115 }
4116 
4117 SYSCTL_OID(_security_jail, OID_AUTO, list,
4118     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4119     sysctl_jail_list, "S", "List of active jails");
4120 
4121 static int
4122 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4123 {
4124 	int error, injail;
4125 
4126 	injail = jailed(req->td->td_ucred);
4127 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4128 
4129 	return (error);
4130 }
4131 
4132 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4133     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4134     sysctl_jail_jailed, "I", "Process in jail?");
4135 
4136 static int
4137 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4138 {
4139 	int error, havevnet;
4140 #ifdef VIMAGE
4141 	struct ucred *cred = req->td->td_ucred;
4142 
4143 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4144 #else
4145 	havevnet = 0;
4146 #endif
4147 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4148 
4149 	return (error);
4150 }
4151 
4152 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4153     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4154     sysctl_jail_vnet, "I", "Jail owns VNET?");
4155 
4156 #if defined(INET) || defined(INET6)
4157 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4158     &jail_max_af_ips, 0,
4159     "Number of IP addresses a jail may have at most per address family");
4160 #endif
4161 
4162 /*
4163  * Default parameters for jail(2) compatability.  For historical reasons,
4164  * the sysctl names have varying similarity to the parameter names.  Prisons
4165  * just see their own parameters, and can't change them.
4166  */
4167 static int
4168 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4169 {
4170 	struct prison *pr;
4171 	int allow, error, i;
4172 
4173 	pr = req->td->td_ucred->cr_prison;
4174 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4175 
4176 	/* Get the current flag value, and convert it to a boolean. */
4177 	i = (allow & arg2) ? 1 : 0;
4178 	if (arg1 != NULL)
4179 		i = !i;
4180 	error = sysctl_handle_int(oidp, &i, 0, req);
4181 	if (error || !req->newptr)
4182 		return (error);
4183 	i = i ? arg2 : 0;
4184 	if (arg1 != NULL)
4185 		i ^= arg2;
4186 	/*
4187 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4188 	 * for writing.
4189 	 */
4190 	mtx_lock(&prison0.pr_mtx);
4191 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4192 	mtx_unlock(&prison0.pr_mtx);
4193 	return (0);
4194 }
4195 
4196 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4197     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4198     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4199     "Processes in jail can set their hostnames");
4200 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4201     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4202     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4203     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4204 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4205     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4206     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4207     "Processes in jail can use System V IPC primitives");
4208 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4209     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4210     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4211     "Prison root can create raw sockets");
4212 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4213     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4214     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4215     "Processes in jail can alter system file flags");
4216 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4217     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4218     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4219     "Processes in jail can mount/unmount jail-friendly file systems");
4220 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4221     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4222     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4223     "Processes in jail can mount the devfs file system");
4224 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4225     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4226     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4227     "Processes in jail can mount the nullfs file system");
4228 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4229     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4230     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4231     "Processes in jail can mount the procfs file system");
4232 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4233     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4234     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4235     "Processes in jail can mount the tmpfs file system");
4236 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4237     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4238     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4239     "Processes in jail can mount the zfs file system");
4240 
4241 static int
4242 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4243 {
4244 	struct prison *pr;
4245 	int level, error;
4246 
4247 	pr = req->td->td_ucred->cr_prison;
4248 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4249 	error = sysctl_handle_int(oidp, &level, 0, req);
4250 	if (error || !req->newptr)
4251 		return (error);
4252 	*(int *)arg1 = level;
4253 	return (0);
4254 }
4255 
4256 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4257     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4258     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4259     sysctl_jail_default_level, "I",
4260     "Processes in jail cannot see all mounted file systems");
4261 
4262 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4263     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4264     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4265     sysctl_jail_default_level, "I",
4266     "Ruleset for the devfs filesystem in jail");
4267 
4268 /*
4269  * Nodes to describe jail parameters.  Maximum length of string parameters
4270  * is returned in the string itself, and the other parameters exist merely
4271  * to make themselves and their types known.
4272  */
4273 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4274     "Jail parameters");
4275 
4276 int
4277 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4278 {
4279 	int i;
4280 	long l;
4281 	size_t s;
4282 	char numbuf[12];
4283 
4284 	switch (oidp->oid_kind & CTLTYPE)
4285 	{
4286 	case CTLTYPE_LONG:
4287 	case CTLTYPE_ULONG:
4288 		l = 0;
4289 #ifdef SCTL_MASK32
4290 		if (!(req->flags & SCTL_MASK32))
4291 #endif
4292 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4293 	case CTLTYPE_INT:
4294 	case CTLTYPE_UINT:
4295 		i = 0;
4296 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4297 	case CTLTYPE_STRING:
4298 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4299 		return
4300 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4301 	case CTLTYPE_STRUCT:
4302 		s = (size_t)arg2;
4303 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4304 	}
4305 	return (0);
4306 }
4307 
4308 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4309 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4310 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4311 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4312 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4313     "I", "Jail secure level");
4314 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4315     "I", "Jail cannot see all mounted file systems");
4316 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4317     "I", "Ruleset for in-jail devfs mounts");
4318 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4319     "B", "Jail persistence");
4320 #ifdef VIMAGE
4321 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4322     "E,jailsys", "Virtual network stack");
4323 #endif
4324 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4325     "B", "Jail is in the process of shutting down");
4326 
4327 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4328 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4329     "I", "Current number of child jails");
4330 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4331     "I", "Maximum number of child jails");
4332 
4333 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4334 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4335     "Jail hostname");
4336 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4337     "Jail NIS domainname");
4338 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4339     "Jail host UUID");
4340 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4341     "LU", "Jail host ID");
4342 
4343 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4344 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4345 
4346 #ifdef INET
4347 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4348     "Jail IPv4 address virtualization");
4349 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4350     "S,in_addr,a", "Jail IPv4 addresses");
4351 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4352     "B", "Do (not) use IPv4 source address selection rather than the "
4353     "primary jail IPv4 address.");
4354 #endif
4355 #ifdef INET6
4356 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4357     "Jail IPv6 address virtualization");
4358 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4359     "S,in6_addr,a", "Jail IPv6 addresses");
4360 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4361     "B", "Do (not) use IPv6 source address selection rather than the "
4362     "primary jail IPv6 address.");
4363 #endif
4364 
4365 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4366 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4367     "B", "Jail may set hostname");
4368 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4369     "B", "Jail may use SYSV IPC");
4370 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4371     "B", "Jail may create raw sockets");
4372 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4373     "B", "Jail may alter system file flags");
4374 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4375     "B", "Jail may set file quotas");
4376 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4377     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4378 
4379 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4380 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4381     "B", "Jail may mount/unmount jail-friendly file systems in general");
4382 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4383     "B", "Jail may mount the devfs file system");
4384 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4385     "B", "Jail may mount the nullfs file system");
4386 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4387     "B", "Jail may mount the procfs file system");
4388 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4389     "B", "Jail may mount the tmpfs file system");
4390 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4391     "B", "Jail may mount the zfs file system");
4392 
4393 void
4394 prison_racct_foreach(void (*callback)(struct racct *racct,
4395     void *arg2, void *arg3), void *arg2, void *arg3)
4396 {
4397 	struct prison_racct *prr;
4398 
4399 	sx_slock(&allprison_lock);
4400 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4401 		(callback)(prr->prr_racct, arg2, arg3);
4402 	sx_sunlock(&allprison_lock);
4403 }
4404 
4405 static struct prison_racct *
4406 prison_racct_find_locked(const char *name)
4407 {
4408 	struct prison_racct *prr;
4409 
4410 	sx_assert(&allprison_lock, SA_XLOCKED);
4411 
4412 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4413 		return (NULL);
4414 
4415 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4416 		if (strcmp(name, prr->prr_name) != 0)
4417 			continue;
4418 
4419 		/* Found prison_racct with a matching name? */
4420 		prison_racct_hold(prr);
4421 		return (prr);
4422 	}
4423 
4424 	/* Add new prison_racct. */
4425 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4426 	racct_create(&prr->prr_racct);
4427 
4428 	strcpy(prr->prr_name, name);
4429 	refcount_init(&prr->prr_refcount, 1);
4430 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4431 
4432 	return (prr);
4433 }
4434 
4435 struct prison_racct *
4436 prison_racct_find(const char *name)
4437 {
4438 	struct prison_racct *prr;
4439 
4440 	sx_xlock(&allprison_lock);
4441 	prr = prison_racct_find_locked(name);
4442 	sx_xunlock(&allprison_lock);
4443 	return (prr);
4444 }
4445 
4446 void
4447 prison_racct_hold(struct prison_racct *prr)
4448 {
4449 
4450 	refcount_acquire(&prr->prr_refcount);
4451 }
4452 
4453 static void
4454 prison_racct_free_locked(struct prison_racct *prr)
4455 {
4456 
4457 	sx_assert(&allprison_lock, SA_XLOCKED);
4458 
4459 	if (refcount_release(&prr->prr_refcount)) {
4460 		racct_destroy(&prr->prr_racct);
4461 		LIST_REMOVE(prr, prr_next);
4462 		free(prr, M_PRISON_RACCT);
4463 	}
4464 }
4465 
4466 void
4467 prison_racct_free(struct prison_racct *prr)
4468 {
4469 	int old;
4470 
4471 	sx_assert(&allprison_lock, SA_UNLOCKED);
4472 
4473 	old = prr->prr_refcount;
4474 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4475 		return;
4476 
4477 	sx_xlock(&allprison_lock);
4478 	prison_racct_free_locked(prr);
4479 	sx_xunlock(&allprison_lock);
4480 }
4481 
4482 #ifdef RACCT
4483 static void
4484 prison_racct_attach(struct prison *pr)
4485 {
4486 	struct prison_racct *prr;
4487 
4488 	sx_assert(&allprison_lock, SA_XLOCKED);
4489 
4490 	prr = prison_racct_find_locked(pr->pr_name);
4491 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4492 
4493 	pr->pr_prison_racct = prr;
4494 }
4495 
4496 /*
4497  * Handle jail renaming.  From the racct point of view, renaming means
4498  * moving from one prison_racct to another.
4499  */
4500 static void
4501 prison_racct_modify(struct prison *pr)
4502 {
4503 	struct proc *p;
4504 	struct ucred *cred;
4505 	struct prison_racct *oldprr;
4506 
4507 	sx_slock(&allproc_lock);
4508 	sx_xlock(&allprison_lock);
4509 
4510 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4511 		sx_xunlock(&allprison_lock);
4512 		sx_sunlock(&allproc_lock);
4513 		return;
4514 	}
4515 
4516 	oldprr = pr->pr_prison_racct;
4517 	pr->pr_prison_racct = NULL;
4518 
4519 	prison_racct_attach(pr);
4520 
4521 	/*
4522 	 * Move resource utilisation records.
4523 	 */
4524 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4525 
4526 	/*
4527 	 * Force rctl to reattach rules to processes.
4528 	 */
4529 	FOREACH_PROC_IN_SYSTEM(p) {
4530 		PROC_LOCK(p);
4531 		cred = crhold(p->p_ucred);
4532 		PROC_UNLOCK(p);
4533 		racct_proc_ucred_changed(p, cred, cred);
4534 		crfree(cred);
4535 	}
4536 
4537 	sx_sunlock(&allproc_lock);
4538 	prison_racct_free_locked(oldprr);
4539 	sx_xunlock(&allprison_lock);
4540 }
4541 
4542 static void
4543 prison_racct_detach(struct prison *pr)
4544 {
4545 
4546 	sx_assert(&allprison_lock, SA_UNLOCKED);
4547 
4548 	if (pr->pr_prison_racct == NULL)
4549 		return;
4550 	prison_racct_free(pr->pr_prison_racct);
4551 	pr->pr_prison_racct = NULL;
4552 }
4553 #endif /* RACCT */
4554 
4555 #ifdef DDB
4556 
4557 static void
4558 db_show_prison(struct prison *pr)
4559 {
4560 	int fi;
4561 #if defined(INET) || defined(INET6)
4562 	int ii;
4563 #endif
4564 	unsigned jsf;
4565 #ifdef INET6
4566 	char ip6buf[INET6_ADDRSTRLEN];
4567 #endif
4568 
4569 	db_printf("prison %p:\n", pr);
4570 	db_printf(" jid             = %d\n", pr->pr_id);
4571 	db_printf(" name            = %s\n", pr->pr_name);
4572 	db_printf(" parent          = %p\n", pr->pr_parent);
4573 	db_printf(" ref             = %d\n", pr->pr_ref);
4574 	db_printf(" uref            = %d\n", pr->pr_uref);
4575 	db_printf(" path            = %s\n", pr->pr_path);
4576 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4577 	    ? pr->pr_cpuset->cs_id : -1);
4578 #ifdef VIMAGE
4579 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4580 #endif
4581 	db_printf(" root            = %p\n", pr->pr_root);
4582 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4583 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4584 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4585 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4586 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4587 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4588 	db_printf(" flags           = 0x%x", pr->pr_flags);
4589 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4590 	    fi++)
4591 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4592 			db_printf(" %s", pr_flag_names[fi]);
4593 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4594 	    fi++) {
4595 		jsf = pr->pr_flags &
4596 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4597 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4598 		    pr_flag_jailsys[fi].disable &&
4599 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4600 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4601 		    : "inherit");
4602 	}
4603 	db_printf(" allow           = 0x%x", pr->pr_allow);
4604 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4605 	    fi++)
4606 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4607 			db_printf(" %s", pr_allow_names[fi]);
4608 	db_printf("\n");
4609 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4610 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4611 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4612 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4613 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4614 #ifdef INET
4615 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4616 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4617 		db_printf(" %s %s\n",
4618 		    ii == 0 ? "ip4.addr        =" : "                 ",
4619 		    inet_ntoa(pr->pr_ip4[ii]));
4620 #endif
4621 #ifdef INET6
4622 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4623 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4624 		db_printf(" %s %s\n",
4625 		    ii == 0 ? "ip6.addr        =" : "                 ",
4626 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4627 #endif
4628 }
4629 
4630 DB_SHOW_COMMAND(prison, db_show_prison_command)
4631 {
4632 	struct prison *pr;
4633 
4634 	if (!have_addr) {
4635 		/*
4636 		 * Show all prisons in the list, and prison0 which is not
4637 		 * listed.
4638 		 */
4639 		db_show_prison(&prison0);
4640 		if (!db_pager_quit) {
4641 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4642 				db_show_prison(pr);
4643 				if (db_pager_quit)
4644 					break;
4645 			}
4646 		}
4647 		return;
4648 	}
4649 
4650 	if (addr == 0)
4651 		pr = &prison0;
4652 	else {
4653 		/* Look for a prison with the ID and with references. */
4654 		TAILQ_FOREACH(pr, &allprison, pr_list)
4655 			if (pr->pr_id == addr && pr->pr_ref > 0)
4656 				break;
4657 		if (pr == NULL)
4658 			/* Look again, without requiring a reference. */
4659 			TAILQ_FOREACH(pr, &allprison, pr_list)
4660 				if (pr->pr_id == addr)
4661 					break;
4662 		if (pr == NULL)
4663 			/* Assume address points to a valid prison. */
4664 			pr = (struct prison *)addr;
4665 	}
4666 	db_show_prison(pr);
4667 }
4668 
4669 #endif /* DDB */
4670