xref: /freebsd/sys/kern/kern_jail.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 
70 #ifdef DDB
71 #include <ddb/ddb.h>
72 #ifdef INET6
73 #include <netinet6/in6_var.h>
74 #endif /* INET6 */
75 #endif /* DDB */
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
80 
81 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
82 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
83 
84 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
85 #ifdef INET
86 #ifdef INET6
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
88 #else
89 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
90 #endif
91 #else /* !INET */
92 #ifdef INET6
93 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
94 #else
95 #define	_PR_IP_SADDRSEL	0
96 #endif
97 #endif
98 
99 /* prison0 describes what is "real" about the system. */
100 struct prison prison0 = {
101 	.pr_id		= 0,
102 	.pr_name	= "0",
103 	.pr_ref		= 1,
104 	.pr_uref	= 1,
105 	.pr_path	= "/",
106 	.pr_securelevel	= -1,
107 	.pr_devfs_rsnum = 0,
108 	.pr_childmax	= JAIL_MAX,
109 	.pr_hostuuid	= DEFAULT_HOSTUUID,
110 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
111 #ifdef VIMAGE
112 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
113 #else
114 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
115 #endif
116 	.pr_allow	= PR_ALLOW_ALL,
117 };
118 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
119 
120 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
121 struct	sx allprison_lock;
122 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
123 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
124 LIST_HEAD(, prison_racct) allprison_racct;
125 int	lastprid = 0;
126 
127 static int do_jail_attach(struct thread *td, struct prison *pr);
128 static void prison_complete(void *context, int pending);
129 static void prison_deref(struct prison *pr, int flags);
130 static char *prison_path(struct prison *pr1, struct prison *pr2);
131 static void prison_remove_one(struct prison *pr);
132 #ifdef RACCT
133 static void prison_racct_attach(struct prison *pr);
134 static void prison_racct_modify(struct prison *pr);
135 static void prison_racct_detach(struct prison *pr);
136 #endif
137 #ifdef INET
138 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
139 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
140 #endif
141 #ifdef INET6
142 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
143 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
144 #endif
145 
146 /* Flags for prison_deref */
147 #define	PD_DEREF	0x01
148 #define	PD_DEUREF	0x02
149 #define	PD_LOCKED	0x04
150 #define	PD_LIST_SLOCKED	0x08
151 #define	PD_LIST_XLOCKED	0x10
152 
153 /*
154  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
155  * as we cannot figure out the size of a sparse array, or an array without a
156  * terminating entry.
157  */
158 static char *pr_flag_names[] = {
159 	[0] = "persist",
160 #ifdef INET
161 	[7] = "ip4.saddrsel",
162 #endif
163 #ifdef INET6
164 	[8] = "ip6.saddrsel",
165 #endif
166 };
167 const size_t pr_flag_names_size = sizeof(pr_flag_names);
168 
169 static char *pr_flag_nonames[] = {
170 	[0] = "nopersist",
171 #ifdef INET
172 	[7] = "ip4.nosaddrsel",
173 #endif
174 #ifdef INET6
175 	[8] = "ip6.nosaddrsel",
176 #endif
177 };
178 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
179 
180 struct jailsys_flags {
181 	const char	*name;
182 	unsigned	 disable;
183 	unsigned	 new;
184 } pr_flag_jailsys[] = {
185 	{ "host", 0, PR_HOST },
186 #ifdef VIMAGE
187 	{ "vnet", 0, PR_VNET },
188 #endif
189 #ifdef INET
190 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
191 #endif
192 #ifdef INET6
193 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
194 #endif
195 };
196 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
197 
198 static char *pr_allow_names[] = {
199 	"allow.set_hostname",
200 	"allow.sysvipc",
201 	"allow.raw_sockets",
202 	"allow.chflags",
203 	"allow.mount",
204 	"allow.quotas",
205 	"allow.socket_af",
206 	"allow.mount.devfs",
207 	"allow.mount.nullfs",
208 	"allow.mount.zfs",
209 	"allow.mount.procfs",
210 	"allow.mount.tmpfs",
211 	"allow.mount.fdescfs",
212 };
213 const size_t pr_allow_names_size = sizeof(pr_allow_names);
214 
215 static char *pr_allow_nonames[] = {
216 	"allow.noset_hostname",
217 	"allow.nosysvipc",
218 	"allow.noraw_sockets",
219 	"allow.nochflags",
220 	"allow.nomount",
221 	"allow.noquotas",
222 	"allow.nosocket_af",
223 	"allow.mount.nodevfs",
224 	"allow.mount.nonullfs",
225 	"allow.mount.nozfs",
226 	"allow.mount.noprocfs",
227 	"allow.mount.notmpfs",
228 	"allow.mount.nofdescfs",
229 };
230 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
231 
232 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
233 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
234 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
235 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
236 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
237 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
238 #if defined(INET) || defined(INET6)
239 static unsigned jail_max_af_ips = 255;
240 #endif
241 
242 #ifdef INET
243 static int
244 qcmp_v4(const void *ip1, const void *ip2)
245 {
246 	in_addr_t iaa, iab;
247 
248 	/*
249 	 * We need to compare in HBO here to get the list sorted as expected
250 	 * by the result of the code.  Sorting NBO addresses gives you
251 	 * interesting results.  If you do not understand, do not try.
252 	 */
253 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
254 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
255 
256 	/*
257 	 * Do not simply return the difference of the two numbers, the int is
258 	 * not wide enough.
259 	 */
260 	if (iaa > iab)
261 		return (1);
262 	else if (iaa < iab)
263 		return (-1);
264 	else
265 		return (0);
266 }
267 #endif
268 
269 #ifdef INET6
270 static int
271 qcmp_v6(const void *ip1, const void *ip2)
272 {
273 	const struct in6_addr *ia6a, *ia6b;
274 	int i, rc;
275 
276 	ia6a = (const struct in6_addr *)ip1;
277 	ia6b = (const struct in6_addr *)ip2;
278 
279 	rc = 0;
280 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
281 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
282 			rc = 1;
283 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
284 			rc = -1;
285 	}
286 	return (rc);
287 }
288 #endif
289 
290 /*
291  * struct jail_args {
292  *	struct jail *jail;
293  * };
294  */
295 int
296 sys_jail(struct thread *td, struct jail_args *uap)
297 {
298 	uint32_t version;
299 	int error;
300 	struct jail j;
301 
302 	error = copyin(uap->jail, &version, sizeof(uint32_t));
303 	if (error)
304 		return (error);
305 
306 	switch (version) {
307 	case 0:
308 	{
309 		struct jail_v0 j0;
310 
311 		/* FreeBSD single IPv4 jails. */
312 		bzero(&j, sizeof(struct jail));
313 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
314 		if (error)
315 			return (error);
316 		j.version = j0.version;
317 		j.path = j0.path;
318 		j.hostname = j0.hostname;
319 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
320 		break;
321 	}
322 
323 	case 1:
324 		/*
325 		 * Version 1 was used by multi-IPv4 jail implementations
326 		 * that never made it into the official kernel.
327 		 */
328 		return (EINVAL);
329 
330 	case 2:	/* JAIL_API_VERSION */
331 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
332 		error = copyin(uap->jail, &j, sizeof(struct jail));
333 		if (error)
334 			return (error);
335 		break;
336 
337 	default:
338 		/* Sci-Fi jails are not supported, sorry. */
339 		return (EINVAL);
340 	}
341 	return (kern_jail(td, &j));
342 }
343 
344 int
345 kern_jail(struct thread *td, struct jail *j)
346 {
347 	struct iovec optiov[2 * (4
348 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
349 #ifdef INET
350 			    + 1
351 #endif
352 #ifdef INET6
353 			    + 1
354 #endif
355 			    )];
356 	struct uio opt;
357 	char *u_path, *u_hostname, *u_name;
358 #ifdef INET
359 	uint32_t ip4s;
360 	struct in_addr *u_ip4;
361 #endif
362 #ifdef INET6
363 	struct in6_addr *u_ip6;
364 #endif
365 	size_t tmplen;
366 	int error, enforce_statfs, fi;
367 
368 	bzero(&optiov, sizeof(optiov));
369 	opt.uio_iov = optiov;
370 	opt.uio_iovcnt = 0;
371 	opt.uio_offset = -1;
372 	opt.uio_resid = -1;
373 	opt.uio_segflg = UIO_SYSSPACE;
374 	opt.uio_rw = UIO_READ;
375 	opt.uio_td = td;
376 
377 	/* Set permissions for top-level jails from sysctls. */
378 	if (!jailed(td->td_ucred)) {
379 		for (fi = 0; fi < sizeof(pr_allow_names) /
380 		     sizeof(pr_allow_names[0]); fi++) {
381 			optiov[opt.uio_iovcnt].iov_base =
382 			    (jail_default_allow & (1 << fi))
383 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
384 			optiov[opt.uio_iovcnt].iov_len =
385 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
386 			opt.uio_iovcnt += 2;
387 		}
388 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
389 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
390 		opt.uio_iovcnt++;
391 		enforce_statfs = jail_default_enforce_statfs;
392 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
393 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
394 		opt.uio_iovcnt++;
395 	}
396 
397 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
398 #ifdef INET
399 	ip4s = (j->version == 0) ? 1 : j->ip4s;
400 	if (ip4s > jail_max_af_ips)
401 		return (EINVAL);
402 	tmplen += ip4s * sizeof(struct in_addr);
403 #else
404 	if (j->ip4s > 0)
405 		return (EINVAL);
406 #endif
407 #ifdef INET6
408 	if (j->ip6s > jail_max_af_ips)
409 		return (EINVAL);
410 	tmplen += j->ip6s * sizeof(struct in6_addr);
411 #else
412 	if (j->ip6s > 0)
413 		return (EINVAL);
414 #endif
415 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
416 	u_hostname = u_path + MAXPATHLEN;
417 	u_name = u_hostname + MAXHOSTNAMELEN;
418 #ifdef INET
419 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
420 #endif
421 #ifdef INET6
422 #ifdef INET
423 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
424 #else
425 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
426 #endif
427 #endif
428 	optiov[opt.uio_iovcnt].iov_base = "path";
429 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
430 	opt.uio_iovcnt++;
431 	optiov[opt.uio_iovcnt].iov_base = u_path;
432 	error = copyinstr(j->path, u_path, MAXPATHLEN,
433 	    &optiov[opt.uio_iovcnt].iov_len);
434 	if (error) {
435 		free(u_path, M_TEMP);
436 		return (error);
437 	}
438 	opt.uio_iovcnt++;
439 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
440 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
441 	opt.uio_iovcnt++;
442 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
443 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
444 	    &optiov[opt.uio_iovcnt].iov_len);
445 	if (error) {
446 		free(u_path, M_TEMP);
447 		return (error);
448 	}
449 	opt.uio_iovcnt++;
450 	if (j->jailname != NULL) {
451 		optiov[opt.uio_iovcnt].iov_base = "name";
452 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
453 		opt.uio_iovcnt++;
454 		optiov[opt.uio_iovcnt].iov_base = u_name;
455 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
456 		    &optiov[opt.uio_iovcnt].iov_len);
457 		if (error) {
458 			free(u_path, M_TEMP);
459 			return (error);
460 		}
461 		opt.uio_iovcnt++;
462 	}
463 #ifdef INET
464 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
465 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
466 	opt.uio_iovcnt++;
467 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
468 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
469 	if (j->version == 0)
470 		u_ip4->s_addr = j->ip4s;
471 	else {
472 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
473 		if (error) {
474 			free(u_path, M_TEMP);
475 			return (error);
476 		}
477 	}
478 	opt.uio_iovcnt++;
479 #endif
480 #ifdef INET6
481 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
482 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
483 	opt.uio_iovcnt++;
484 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
485 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
486 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
487 	if (error) {
488 		free(u_path, M_TEMP);
489 		return (error);
490 	}
491 	opt.uio_iovcnt++;
492 #endif
493 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
494 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
495 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
496 	free(u_path, M_TEMP);
497 	return (error);
498 }
499 
500 
501 /*
502  * struct jail_set_args {
503  *	struct iovec *iovp;
504  *	unsigned int iovcnt;
505  *	int flags;
506  * };
507  */
508 int
509 sys_jail_set(struct thread *td, struct jail_set_args *uap)
510 {
511 	struct uio *auio;
512 	int error;
513 
514 	/* Check that we have an even number of iovecs. */
515 	if (uap->iovcnt & 1)
516 		return (EINVAL);
517 
518 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
519 	if (error)
520 		return (error);
521 	error = kern_jail_set(td, auio, uap->flags);
522 	free(auio, M_IOV);
523 	return (error);
524 }
525 
526 int
527 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
528 {
529 	struct nameidata nd;
530 #ifdef INET
531 	struct in_addr *ip4;
532 #endif
533 #ifdef INET6
534 	struct in6_addr *ip6;
535 #endif
536 	struct vfsopt *opt;
537 	struct vfsoptlist *opts;
538 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
539 	struct vnode *root;
540 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
541 	char *g_path;
542 #if defined(INET) || defined(INET6)
543 	struct prison *tppr;
544 	void *op;
545 #endif
546 	unsigned long hid;
547 	size_t namelen, onamelen;
548 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
549 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
550 	int fi, jid, jsys, len, level;
551 	int childmax, rsnum, slevel;
552 	int fullpath_disabled;
553 #if defined(INET) || defined(INET6)
554 	int ii, ij;
555 #endif
556 #ifdef INET
557 	int ip4s, redo_ip4;
558 #endif
559 #ifdef INET6
560 	int ip6s, redo_ip6;
561 #endif
562 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
563 	unsigned tallow;
564 	char numbuf[12];
565 
566 	error = priv_check(td, PRIV_JAIL_SET);
567 	if (!error && (flags & JAIL_ATTACH))
568 		error = priv_check(td, PRIV_JAIL_ATTACH);
569 	if (error)
570 		return (error);
571 	mypr = ppr = td->td_ucred->cr_prison;
572 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
573 		return (EPERM);
574 	if (flags & ~JAIL_SET_MASK)
575 		return (EINVAL);
576 
577 	/*
578 	 * Check all the parameters before committing to anything.  Not all
579 	 * errors can be caught early, but we may as well try.  Also, this
580 	 * takes care of some expensive stuff (path lookup) before getting
581 	 * the allprison lock.
582 	 *
583 	 * XXX Jails are not filesystems, and jail parameters are not mount
584 	 *     options.  But it makes more sense to re-use the vfsopt code
585 	 *     than duplicate it under a different name.
586 	 */
587 	error = vfs_buildopts(optuio, &opts);
588 	if (error)
589 		return (error);
590 #ifdef INET
591 	ip4 = NULL;
592 #endif
593 #ifdef INET6
594 	ip6 = NULL;
595 #endif
596 	g_path = NULL;
597 
598 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
599 	if (error == ENOENT)
600 		jid = 0;
601 	else if (error != 0)
602 		goto done_free;
603 
604 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
605 	if (error == ENOENT)
606 		gotslevel = 0;
607 	else if (error != 0)
608 		goto done_free;
609 	else
610 		gotslevel = 1;
611 
612 	error =
613 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
614 	if (error == ENOENT)
615 		gotchildmax = 0;
616 	else if (error != 0)
617 		goto done_free;
618 	else
619 		gotchildmax = 1;
620 
621 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
622 	if (error == ENOENT)
623 		gotenforce = 0;
624 	else if (error != 0)
625 		goto done_free;
626 	else if (enforce < 0 || enforce > 2) {
627 		error = EINVAL;
628 		goto done_free;
629 	} else
630 		gotenforce = 1;
631 
632 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
633 	if (error == ENOENT)
634 		gotrsnum = 0;
635 	else if (error != 0)
636 		goto done_free;
637 	else
638 		gotrsnum = 1;
639 
640 	pr_flags = ch_flags = 0;
641 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
642 	    fi++) {
643 		if (pr_flag_names[fi] == NULL)
644 			continue;
645 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
646 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
647 	}
648 	ch_flags |= pr_flags;
649 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
650 	    fi++) {
651 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
652 		    sizeof(jsys));
653 		if (error == ENOENT)
654 			continue;
655 		if (error != 0)
656 			goto done_free;
657 		switch (jsys) {
658 		case JAIL_SYS_DISABLE:
659 			if (!pr_flag_jailsys[fi].disable) {
660 				error = EINVAL;
661 				goto done_free;
662 			}
663 			pr_flags |= pr_flag_jailsys[fi].disable;
664 			break;
665 		case JAIL_SYS_NEW:
666 			pr_flags |= pr_flag_jailsys[fi].new;
667 			break;
668 		case JAIL_SYS_INHERIT:
669 			break;
670 		default:
671 			error = EINVAL;
672 			goto done_free;
673 		}
674 		ch_flags |=
675 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
676 	}
677 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
678 	    && !(pr_flags & PR_PERSIST)) {
679 		error = EINVAL;
680 		vfs_opterror(opts, "new jail must persist or attach");
681 		goto done_errmsg;
682 	}
683 #ifdef VIMAGE
684 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
685 		error = EINVAL;
686 		vfs_opterror(opts, "vnet cannot be changed after creation");
687 		goto done_errmsg;
688 	}
689 #endif
690 #ifdef INET
691 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
692 		error = EINVAL;
693 		vfs_opterror(opts, "ip4 cannot be changed after creation");
694 		goto done_errmsg;
695 	}
696 #endif
697 #ifdef INET6
698 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
699 		error = EINVAL;
700 		vfs_opterror(opts, "ip6 cannot be changed after creation");
701 		goto done_errmsg;
702 	}
703 #endif
704 
705 	pr_allow = ch_allow = 0;
706 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
707 	    fi++) {
708 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
709 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
710 	}
711 	ch_allow |= pr_allow;
712 
713 	error = vfs_getopt(opts, "name", (void **)&name, &len);
714 	if (error == ENOENT)
715 		name = NULL;
716 	else if (error != 0)
717 		goto done_free;
718 	else {
719 		if (len == 0 || name[len - 1] != '\0') {
720 			error = EINVAL;
721 			goto done_free;
722 		}
723 		if (len > MAXHOSTNAMELEN) {
724 			error = ENAMETOOLONG;
725 			goto done_free;
726 		}
727 	}
728 
729 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
730 	if (error == ENOENT)
731 		host = NULL;
732 	else if (error != 0)
733 		goto done_free;
734 	else {
735 		ch_flags |= PR_HOST;
736 		pr_flags |= PR_HOST;
737 		if (len == 0 || host[len - 1] != '\0') {
738 			error = EINVAL;
739 			goto done_free;
740 		}
741 		if (len > MAXHOSTNAMELEN) {
742 			error = ENAMETOOLONG;
743 			goto done_free;
744 		}
745 	}
746 
747 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
748 	if (error == ENOENT)
749 		domain = NULL;
750 	else if (error != 0)
751 		goto done_free;
752 	else {
753 		ch_flags |= PR_HOST;
754 		pr_flags |= PR_HOST;
755 		if (len == 0 || domain[len - 1] != '\0') {
756 			error = EINVAL;
757 			goto done_free;
758 		}
759 		if (len > MAXHOSTNAMELEN) {
760 			error = ENAMETOOLONG;
761 			goto done_free;
762 		}
763 	}
764 
765 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
766 	if (error == ENOENT)
767 		uuid = NULL;
768 	else if (error != 0)
769 		goto done_free;
770 	else {
771 		ch_flags |= PR_HOST;
772 		pr_flags |= PR_HOST;
773 		if (len == 0 || uuid[len - 1] != '\0') {
774 			error = EINVAL;
775 			goto done_free;
776 		}
777 		if (len > HOSTUUIDLEN) {
778 			error = ENAMETOOLONG;
779 			goto done_free;
780 		}
781 	}
782 
783 #ifdef COMPAT_FREEBSD32
784 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
785 		uint32_t hid32;
786 
787 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
788 		hid = hid32;
789 	} else
790 #endif
791 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
792 	if (error == ENOENT)
793 		gothid = 0;
794 	else if (error != 0)
795 		goto done_free;
796 	else {
797 		gothid = 1;
798 		ch_flags |= PR_HOST;
799 		pr_flags |= PR_HOST;
800 	}
801 
802 #ifdef INET
803 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
804 	if (error == ENOENT)
805 		ip4s = 0;
806 	else if (error != 0)
807 		goto done_free;
808 	else if (ip4s & (sizeof(*ip4) - 1)) {
809 		error = EINVAL;
810 		goto done_free;
811 	} else {
812 		ch_flags |= PR_IP4_USER;
813 		pr_flags |= PR_IP4_USER;
814 		if (ip4s > 0) {
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 = 0;
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;
869 		pr_flags |= PR_IP6_USER;
870 		if (ip6s > 0) {
871 			ip6s /= sizeof(*ip6);
872 			if (ip6s > jail_max_af_ips) {
873 				error = EINVAL;
874 				vfs_opterror(opts, "too many IPv6 addresses");
875 				goto done_errmsg;
876 			}
877 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
878 			bcopy(op, ip6, ip6s * sizeof(*ip6));
879 			if (ip6s > 1)
880 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
881 			for (ii = 0; ii < ip6s; ii++) {
882 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
883 					error = EINVAL;
884 					goto done_free;
885 				}
886 				if ((ii+1) < ip6s &&
887 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
888 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
889 				{
890 					error = EINVAL;
891 					goto done_free;
892 				}
893 			}
894 		}
895 	}
896 #endif
897 
898 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
899 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
900 		error = EINVAL;
901 		vfs_opterror(opts,
902 		    "vnet jails cannot have IP address restrictions");
903 		goto done_errmsg;
904 	}
905 #endif
906 
907 	fullpath_disabled = 0;
908 	root = NULL;
909 	error = vfs_getopt(opts, "path", (void **)&path, &len);
910 	if (error == ENOENT)
911 		path = NULL;
912 	else if (error != 0)
913 		goto done_free;
914 	else {
915 		if (flags & JAIL_UPDATE) {
916 			error = EINVAL;
917 			vfs_opterror(opts,
918 			    "path cannot be changed after creation");
919 			goto done_errmsg;
920 		}
921 		if (len == 0 || path[len - 1] != '\0') {
922 			error = EINVAL;
923 			goto done_free;
924 		}
925 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
926 		    path, td);
927 		error = namei(&nd);
928 		if (error)
929 			goto done_free;
930 		root = nd.ni_vp;
931 		NDFREE(&nd, NDF_ONLY_PNBUF);
932 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
933 		strlcpy(g_path, path, MAXPATHLEN);
934 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
935 		if (error == 0)
936 			path = g_path;
937 		else if (error == ENODEV) {
938 			/* proceed if sysctl debug.disablefullpath == 1 */
939 			fullpath_disabled = 1;
940 			if (len < 2 || (len == 2 && path[0] == '/'))
941 				path = NULL;
942 		} else {
943 			/* exit on other errors */
944 			goto done_free;
945 		}
946 		if (root->v_type != VDIR) {
947 			error = ENOTDIR;
948 			vput(root);
949 			goto done_free;
950 		}
951 		VOP_UNLOCK(root, 0);
952 		if (fullpath_disabled) {
953 			/* Leave room for a real-root full pathname. */
954 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
955 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
956 				error = ENAMETOOLONG;
957 				goto done_free;
958 			}
959 		}
960 	}
961 
962 	/*
963 	 * Grab the allprison lock before letting modules check their
964 	 * parameters.  Once we have it, do not let go so we'll have a
965 	 * consistent view of the OSD list.
966 	 */
967 	sx_xlock(&allprison_lock);
968 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
969 	if (error)
970 		goto done_unlock_list;
971 
972 	/* By now, all parameters should have been noted. */
973 	TAILQ_FOREACH(opt, opts, link) {
974 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
975 			error = EINVAL;
976 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
977 			goto done_unlock_list;
978 		}
979 	}
980 
981 	/*
982 	 * See if we are creating a new record or updating an existing one.
983 	 * This abuses the file error codes ENOENT and EEXIST.
984 	 */
985 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
986 	if (!cuflags) {
987 		error = EINVAL;
988 		vfs_opterror(opts, "no valid operation (create or update)");
989 		goto done_unlock_list;
990 	}
991 	pr = NULL;
992 	namelc = NULL;
993 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
994 		namelc = strrchr(name, '.');
995 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
996 		if (*p != '\0')
997 			jid = 0;
998 	}
999 	if (jid != 0) {
1000 		/*
1001 		 * See if a requested jid already exists.  There is an
1002 		 * information leak here if the jid exists but is not within
1003 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1004 		 * even though they cannot see the jail, and CREATE | UPDATE
1005 		 * will return ENOENT which is not normally a valid error.
1006 		 */
1007 		if (jid < 0) {
1008 			error = EINVAL;
1009 			vfs_opterror(opts, "negative jid");
1010 			goto done_unlock_list;
1011 		}
1012 		pr = prison_find(jid);
1013 		if (pr != NULL) {
1014 			ppr = pr->pr_parent;
1015 			/* Create: jid must not exist. */
1016 			if (cuflags == JAIL_CREATE) {
1017 				mtx_unlock(&pr->pr_mtx);
1018 				error = EEXIST;
1019 				vfs_opterror(opts, "jail %d already exists",
1020 				    jid);
1021 				goto done_unlock_list;
1022 			}
1023 			if (!prison_ischild(mypr, pr)) {
1024 				mtx_unlock(&pr->pr_mtx);
1025 				pr = NULL;
1026 			} else if (pr->pr_uref == 0) {
1027 				if (!(flags & JAIL_DYING)) {
1028 					mtx_unlock(&pr->pr_mtx);
1029 					error = ENOENT;
1030 					vfs_opterror(opts, "jail %d is dying",
1031 					    jid);
1032 					goto done_unlock_list;
1033 				} else if ((flags & JAIL_ATTACH) ||
1034 				    (pr_flags & PR_PERSIST)) {
1035 					/*
1036 					 * A dying jail might be resurrected
1037 					 * (via attach or persist), but first
1038 					 * it must determine if another jail
1039 					 * has claimed its name.  Accomplish
1040 					 * this by implicitly re-setting the
1041 					 * name.
1042 					 */
1043 					if (name == NULL)
1044 						name = prison_name(mypr, pr);
1045 				}
1046 			}
1047 		}
1048 		if (pr == NULL) {
1049 			/* Update: jid must exist. */
1050 			if (cuflags == JAIL_UPDATE) {
1051 				error = ENOENT;
1052 				vfs_opterror(opts, "jail %d not found", jid);
1053 				goto done_unlock_list;
1054 			}
1055 		}
1056 	}
1057 	/*
1058 	 * If the caller provided a name, look for a jail by that name.
1059 	 * This has different semantics for creates and updates keyed by jid
1060 	 * (where the name must not already exist in a different jail),
1061 	 * and updates keyed by the name itself (where the name must exist
1062 	 * because that is the jail being updated).
1063 	 */
1064 	if (name != NULL) {
1065 		namelc = strrchr(name, '.');
1066 		if (namelc == NULL)
1067 			namelc = name;
1068 		else {
1069 			/*
1070 			 * This is a hierarchical name.  Split it into the
1071 			 * parent and child names, and make sure the parent
1072 			 * exists or matches an already found jail.
1073 			 */
1074 			*namelc = '\0';
1075 			if (pr != NULL) {
1076 				if (strncmp(name, ppr->pr_name, namelc - name)
1077 				    || ppr->pr_name[namelc - name] != '\0') {
1078 					mtx_unlock(&pr->pr_mtx);
1079 					error = EINVAL;
1080 					vfs_opterror(opts,
1081 					    "cannot change jail's parent");
1082 					goto done_unlock_list;
1083 				}
1084 			} else {
1085 				ppr = prison_find_name(mypr, name);
1086 				if (ppr == NULL) {
1087 					error = ENOENT;
1088 					vfs_opterror(opts,
1089 					    "jail \"%s\" not found", name);
1090 					goto done_unlock_list;
1091 				}
1092 				mtx_unlock(&ppr->pr_mtx);
1093 			}
1094 			name = ++namelc;
1095 		}
1096 		if (name[0] != '\0') {
1097 			namelen =
1098 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1099  name_again:
1100 			deadpr = NULL;
1101 			FOREACH_PRISON_CHILD(ppr, tpr) {
1102 				if (tpr != pr && tpr->pr_ref > 0 &&
1103 				    !strcmp(tpr->pr_name + namelen, name)) {
1104 					if (pr == NULL &&
1105 					    cuflags != JAIL_CREATE) {
1106 						mtx_lock(&tpr->pr_mtx);
1107 						if (tpr->pr_ref > 0) {
1108 							/*
1109 							 * Use this jail
1110 							 * for updates.
1111 							 */
1112 							if (tpr->pr_uref > 0) {
1113 								pr = tpr;
1114 								break;
1115 							}
1116 							deadpr = tpr;
1117 						}
1118 						mtx_unlock(&tpr->pr_mtx);
1119 					} else if (tpr->pr_uref > 0) {
1120 						/*
1121 						 * Create, or update(jid):
1122 						 * name must not exist in an
1123 						 * active sibling jail.
1124 						 */
1125 						error = EEXIST;
1126 						if (pr != NULL)
1127 							mtx_unlock(&pr->pr_mtx);
1128 						vfs_opterror(opts,
1129 						   "jail \"%s\" already exists",
1130 						   name);
1131 						goto done_unlock_list;
1132 					}
1133 				}
1134 			}
1135 			/* If no active jail is found, use a dying one. */
1136 			if (deadpr != NULL && pr == NULL) {
1137 				if (flags & JAIL_DYING) {
1138 					mtx_lock(&deadpr->pr_mtx);
1139 					if (deadpr->pr_ref == 0) {
1140 						mtx_unlock(&deadpr->pr_mtx);
1141 						goto name_again;
1142 					}
1143 					pr = deadpr;
1144 				} else if (cuflags == JAIL_UPDATE) {
1145 					error = ENOENT;
1146 					vfs_opterror(opts,
1147 					    "jail \"%s\" is dying", name);
1148 					goto done_unlock_list;
1149 				}
1150 			}
1151 			/* Update: name must exist if no jid. */
1152 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1153 				error = ENOENT;
1154 				vfs_opterror(opts, "jail \"%s\" not found",
1155 				    name);
1156 				goto done_unlock_list;
1157 			}
1158 		}
1159 	}
1160 	/* Update: must provide a jid or name. */
1161 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1162 		error = ENOENT;
1163 		vfs_opterror(opts, "update specified no jail");
1164 		goto done_unlock_list;
1165 	}
1166 
1167 	/* If there's no prison to update, create a new one and link it in. */
1168 	if (pr == NULL) {
1169 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1170 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1171 				error = EPERM;
1172 				vfs_opterror(opts, "prison limit exceeded");
1173 				goto done_unlock_list;
1174 			}
1175 		created = 1;
1176 		mtx_lock(&ppr->pr_mtx);
1177 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1178 			mtx_unlock(&ppr->pr_mtx);
1179 			error = ENOENT;
1180 			vfs_opterror(opts, "parent jail went away!");
1181 			goto done_unlock_list;
1182 		}
1183 		ppr->pr_ref++;
1184 		ppr->pr_uref++;
1185 		mtx_unlock(&ppr->pr_mtx);
1186 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1187 		if (jid == 0) {
1188 			/* Find the next free jid. */
1189 			jid = lastprid + 1;
1190  findnext:
1191 			if (jid == JAIL_MAX)
1192 				jid = 1;
1193 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1194 				if (tpr->pr_id < jid)
1195 					continue;
1196 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1197 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1198 					break;
1199 				}
1200 				if (jid == lastprid) {
1201 					error = EAGAIN;
1202 					vfs_opterror(opts,
1203 					    "no available jail IDs");
1204 					free(pr, M_PRISON);
1205 					prison_deref(ppr, PD_DEREF |
1206 					    PD_DEUREF | PD_LIST_XLOCKED);
1207 					goto done_releroot;
1208 				}
1209 				jid++;
1210 				goto findnext;
1211 			}
1212 			lastprid = jid;
1213 		} else {
1214 			/*
1215 			 * The jail already has a jid (that did not yet exist),
1216 			 * so just find where to insert it.
1217 			 */
1218 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1219 				if (tpr->pr_id >= jid) {
1220 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1221 					break;
1222 				}
1223 		}
1224 		if (tpr == NULL)
1225 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1226 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1227 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1228 			tpr->pr_childcount++;
1229 
1230 		pr->pr_parent = ppr;
1231 		pr->pr_id = jid;
1232 
1233 		/* Set some default values, and inherit some from the parent. */
1234 		if (name == NULL)
1235 			name = "";
1236 		if (path == NULL) {
1237 			path = "/";
1238 			root = mypr->pr_root;
1239 			vref(root);
1240 		}
1241 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1242 		pr->pr_flags |= PR_HOST;
1243 #if defined(INET) || defined(INET6)
1244 #ifdef VIMAGE
1245 		if (!(pr_flags & PR_VNET))
1246 #endif
1247 		{
1248 #ifdef INET
1249 			if (!(ch_flags & PR_IP4_USER))
1250 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1251 			else if (!(pr_flags & PR_IP4_USER)) {
1252 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1253 				if (ppr->pr_ip4 != NULL) {
1254 					pr->pr_ip4s = ppr->pr_ip4s;
1255 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1256 					    sizeof(struct in_addr), M_PRISON,
1257 					    M_WAITOK);
1258 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1259 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1260 				}
1261 			}
1262 #endif
1263 #ifdef INET6
1264 			if (!(ch_flags & PR_IP6_USER))
1265 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1266 			else if (!(pr_flags & PR_IP6_USER)) {
1267 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1268 				if (ppr->pr_ip6 != NULL) {
1269 					pr->pr_ip6s = ppr->pr_ip6s;
1270 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1271 					    sizeof(struct in6_addr), M_PRISON,
1272 					    M_WAITOK);
1273 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1274 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1275 				}
1276 			}
1277 #endif
1278 		}
1279 #endif
1280 		/* Source address selection is always on by default. */
1281 		pr->pr_flags |= _PR_IP_SADDRSEL;
1282 
1283 		pr->pr_securelevel = ppr->pr_securelevel;
1284 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1285 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1286 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1287 
1288 		LIST_INIT(&pr->pr_children);
1289 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1290 
1291 #ifdef VIMAGE
1292 		/* Allocate a new vnet if specified. */
1293 		pr->pr_vnet = (pr_flags & PR_VNET)
1294 		    ? vnet_alloc() : ppr->pr_vnet;
1295 #endif
1296 		/*
1297 		 * Allocate a dedicated cpuset for each jail.
1298 		 * Unlike other initial settings, this may return an erorr.
1299 		 */
1300 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1301 		if (error) {
1302 			prison_deref(pr, PD_LIST_XLOCKED);
1303 			goto done_releroot;
1304 		}
1305 
1306 		mtx_lock(&pr->pr_mtx);
1307 		/*
1308 		 * New prisons do not yet have a reference, because we do not
1309 		 * want other to see the incomplete prison once the
1310 		 * allprison_lock is downgraded.
1311 		 */
1312 	} else {
1313 		created = 0;
1314 		/*
1315 		 * Grab a reference for existing prisons, to ensure they
1316 		 * continue to exist for the duration of the call.
1317 		 */
1318 		pr->pr_ref++;
1319 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1320 		if ((pr->pr_flags & PR_VNET) &&
1321 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1322 			error = EINVAL;
1323 			vfs_opterror(opts,
1324 			    "vnet jails cannot have IP address restrictions");
1325 			goto done_deref_locked;
1326 		}
1327 #endif
1328 #ifdef INET
1329 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1330 			error = EINVAL;
1331 			vfs_opterror(opts,
1332 			    "ip4 cannot be changed after creation");
1333 			goto done_deref_locked;
1334 		}
1335 #endif
1336 #ifdef INET6
1337 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1338 			error = EINVAL;
1339 			vfs_opterror(opts,
1340 			    "ip6 cannot be changed after creation");
1341 			goto done_deref_locked;
1342 		}
1343 #endif
1344 	}
1345 
1346 	/* Do final error checking before setting anything. */
1347 	if (gotslevel) {
1348 		if (slevel < ppr->pr_securelevel) {
1349 			error = EPERM;
1350 			goto done_deref_locked;
1351 		}
1352 	}
1353 	if (gotchildmax) {
1354 		if (childmax >= ppr->pr_childmax) {
1355 			error = EPERM;
1356 			goto done_deref_locked;
1357 		}
1358 	}
1359 	if (gotenforce) {
1360 		if (enforce < ppr->pr_enforce_statfs) {
1361 			error = EPERM;
1362 			goto done_deref_locked;
1363 		}
1364 	}
1365 	if (gotrsnum) {
1366 		/*
1367 		 * devfs_rsnum is a uint16_t
1368 		 */
1369 		if (rsnum < 0 || rsnum > 65535) {
1370 			error = EINVAL;
1371 			goto done_deref_locked;
1372 		}
1373 		/*
1374 		 * Nested jails always inherit parent's devfs ruleset
1375 		 */
1376 		if (jailed(td->td_ucred)) {
1377 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1378 				error = EPERM;
1379 				goto done_deref_locked;
1380 			} else
1381 				rsnum = ppr->pr_devfs_rsnum;
1382 		}
1383 	}
1384 #ifdef INET
1385 	if (ip4s > 0) {
1386 		if (ppr->pr_flags & PR_IP4) {
1387 			/*
1388 			 * Make sure the new set of IP addresses is a
1389 			 * subset of the parent's list.  Don't worry
1390 			 * about the parent being unlocked, as any
1391 			 * setting is done with allprison_lock held.
1392 			 */
1393 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1394 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1395 					break;
1396 			if (ij == ppr->pr_ip4s) {
1397 				error = EPERM;
1398 				goto done_deref_locked;
1399 			}
1400 			if (ip4s > 1) {
1401 				for (ii = ij = 1; ii < ip4s; ii++) {
1402 					if (ip4[ii].s_addr ==
1403 					    ppr->pr_ip4[0].s_addr)
1404 						continue;
1405 					for (; ij < ppr->pr_ip4s; ij++)
1406 						if (ip4[ii].s_addr ==
1407 						    ppr->pr_ip4[ij].s_addr)
1408 							break;
1409 					if (ij == ppr->pr_ip4s)
1410 						break;
1411 				}
1412 				if (ij == ppr->pr_ip4s) {
1413 					error = EPERM;
1414 					goto done_deref_locked;
1415 				}
1416 			}
1417 		}
1418 		/*
1419 		 * Check for conflicting IP addresses.  We permit them
1420 		 * if there is no more than one IP on each jail.  If
1421 		 * there is a duplicate on a jail with more than one
1422 		 * IP stop checking and return error.
1423 		 */
1424 		tppr = ppr;
1425 #ifdef VIMAGE
1426 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1427 			if (tppr->pr_flags & PR_VNET)
1428 				break;
1429 #endif
1430 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1431 			if (tpr == pr ||
1432 #ifdef VIMAGE
1433 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1434 #endif
1435 			    tpr->pr_uref == 0) {
1436 				descend = 0;
1437 				continue;
1438 			}
1439 			if (!(tpr->pr_flags & PR_IP4_USER))
1440 				continue;
1441 			descend = 0;
1442 			if (tpr->pr_ip4 == NULL ||
1443 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1444 				continue;
1445 			for (ii = 0; ii < ip4s; ii++) {
1446 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1447 					error = EADDRINUSE;
1448 					vfs_opterror(opts,
1449 					    "IPv4 addresses clash");
1450 					goto done_deref_locked;
1451 				}
1452 			}
1453 		}
1454 	}
1455 #endif
1456 #ifdef INET6
1457 	if (ip6s > 0) {
1458 		if (ppr->pr_flags & PR_IP6) {
1459 			/*
1460 			 * Make sure the new set of IP addresses is a
1461 			 * subset of the parent's list.
1462 			 */
1463 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1464 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1465 				    &ppr->pr_ip6[ij]))
1466 					break;
1467 			if (ij == ppr->pr_ip6s) {
1468 				error = EPERM;
1469 				goto done_deref_locked;
1470 			}
1471 			if (ip6s > 1) {
1472 				for (ii = ij = 1; ii < ip6s; ii++) {
1473 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1474 					     &ppr->pr_ip6[0]))
1475 						continue;
1476 					for (; ij < ppr->pr_ip6s; ij++)
1477 						if (IN6_ARE_ADDR_EQUAL(
1478 						    &ip6[ii], &ppr->pr_ip6[ij]))
1479 							break;
1480 					if (ij == ppr->pr_ip6s)
1481 						break;
1482 				}
1483 				if (ij == ppr->pr_ip6s) {
1484 					error = EPERM;
1485 					goto done_deref_locked;
1486 				}
1487 			}
1488 		}
1489 		/* Check for conflicting IP addresses. */
1490 		tppr = ppr;
1491 #ifdef VIMAGE
1492 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1493 			if (tppr->pr_flags & PR_VNET)
1494 				break;
1495 #endif
1496 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1497 			if (tpr == pr ||
1498 #ifdef VIMAGE
1499 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1500 #endif
1501 			    tpr->pr_uref == 0) {
1502 				descend = 0;
1503 				continue;
1504 			}
1505 			if (!(tpr->pr_flags & PR_IP6_USER))
1506 				continue;
1507 			descend = 0;
1508 			if (tpr->pr_ip6 == NULL ||
1509 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1510 				continue;
1511 			for (ii = 0; ii < ip6s; ii++) {
1512 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1513 					error = EADDRINUSE;
1514 					vfs_opterror(opts,
1515 					    "IPv6 addresses clash");
1516 					goto done_deref_locked;
1517 				}
1518 			}
1519 		}
1520 	}
1521 #endif
1522 	onamelen = namelen = 0;
1523 	if (name != NULL) {
1524 		/* Give a default name of the jid. */
1525 		if (name[0] == '\0')
1526 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1527 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1528 		    *p == '\0')) {
1529 			error = EINVAL;
1530 			vfs_opterror(opts,
1531 			    "name cannot be numeric (unless it is the jid)");
1532 			goto done_deref_locked;
1533 		}
1534 		/*
1535 		 * Make sure the name isn't too long for the prison or its
1536 		 * children.
1537 		 */
1538 		onamelen = strlen(pr->pr_name);
1539 		namelen = strlen(name);
1540 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1541 			error = ENAMETOOLONG;
1542 			goto done_deref_locked;
1543 		}
1544 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1545 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1546 			    sizeof(pr->pr_name)) {
1547 				error = ENAMETOOLONG;
1548 				goto done_deref_locked;
1549 			}
1550 		}
1551 	}
1552 	if (pr_allow & ~ppr->pr_allow) {
1553 		error = EPERM;
1554 		goto done_deref_locked;
1555 	}
1556 
1557 	/* Set the parameters of the prison. */
1558 #ifdef INET
1559 	redo_ip4 = 0;
1560 	if (pr_flags & PR_IP4_USER) {
1561 		pr->pr_flags |= PR_IP4;
1562 		free(pr->pr_ip4, M_PRISON);
1563 		pr->pr_ip4s = ip4s;
1564 		pr->pr_ip4 = ip4;
1565 		ip4 = NULL;
1566 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1567 #ifdef VIMAGE
1568 			if (tpr->pr_flags & PR_VNET) {
1569 				descend = 0;
1570 				continue;
1571 			}
1572 #endif
1573 			if (prison_restrict_ip4(tpr, NULL)) {
1574 				redo_ip4 = 1;
1575 				descend = 0;
1576 			}
1577 		}
1578 	}
1579 #endif
1580 #ifdef INET6
1581 	redo_ip6 = 0;
1582 	if (pr_flags & PR_IP6_USER) {
1583 		pr->pr_flags |= PR_IP6;
1584 		free(pr->pr_ip6, M_PRISON);
1585 		pr->pr_ip6s = ip6s;
1586 		pr->pr_ip6 = ip6;
1587 		ip6 = NULL;
1588 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1589 #ifdef VIMAGE
1590 			if (tpr->pr_flags & PR_VNET) {
1591 				descend = 0;
1592 				continue;
1593 			}
1594 #endif
1595 			if (prison_restrict_ip6(tpr, NULL)) {
1596 				redo_ip6 = 1;
1597 				descend = 0;
1598 			}
1599 		}
1600 	}
1601 #endif
1602 	if (gotslevel) {
1603 		pr->pr_securelevel = slevel;
1604 		/* Set all child jails to be at least this level. */
1605 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1606 			if (tpr->pr_securelevel < slevel)
1607 				tpr->pr_securelevel = slevel;
1608 	}
1609 	if (gotchildmax) {
1610 		pr->pr_childmax = childmax;
1611 		/* Set all child jails to under this limit. */
1612 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1613 			if (tpr->pr_childmax > childmax - level)
1614 				tpr->pr_childmax = childmax > level
1615 				    ? childmax - level : 0;
1616 	}
1617 	if (gotenforce) {
1618 		pr->pr_enforce_statfs = enforce;
1619 		/* Pass this restriction on to the children. */
1620 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1621 			if (tpr->pr_enforce_statfs < enforce)
1622 				tpr->pr_enforce_statfs = enforce;
1623 	}
1624 	if (gotrsnum) {
1625 		pr->pr_devfs_rsnum = rsnum;
1626 		/* Pass this restriction on to the children. */
1627 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1628 			tpr->pr_devfs_rsnum = rsnum;
1629 	}
1630 	if (name != NULL) {
1631 		if (ppr == &prison0)
1632 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1633 		else
1634 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1635 			    ppr->pr_name, name);
1636 		/* Change this component of child names. */
1637 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1638 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1639 			    strlen(tpr->pr_name + onamelen) + 1);
1640 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1641 		}
1642 	}
1643 	if (path != NULL) {
1644 		/* Try to keep a real-rooted full pathname. */
1645 		if (fullpath_disabled && path[0] == '/' &&
1646 		    strcmp(mypr->pr_path, "/"))
1647 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1648 			    mypr->pr_path, path);
1649 		else
1650 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1651 		pr->pr_root = root;
1652 	}
1653 	if (PR_HOST & ch_flags & ~pr_flags) {
1654 		if (pr->pr_flags & PR_HOST) {
1655 			/*
1656 			 * Copy the parent's host info.  As with pr_ip4 above,
1657 			 * the lack of a lock on the parent is not a problem;
1658 			 * it is always set with allprison_lock at least
1659 			 * shared, and is held exclusively here.
1660 			 */
1661 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1662 			    sizeof(pr->pr_hostname));
1663 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1664 			    sizeof(pr->pr_domainname));
1665 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1666 			    sizeof(pr->pr_hostuuid));
1667 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1668 		}
1669 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1670 		/* Set this prison, and any descendants without PR_HOST. */
1671 		if (host != NULL)
1672 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1673 		if (domain != NULL)
1674 			strlcpy(pr->pr_domainname, domain,
1675 			    sizeof(pr->pr_domainname));
1676 		if (uuid != NULL)
1677 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1678 		if (gothid)
1679 			pr->pr_hostid = hid;
1680 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1681 			if (tpr->pr_flags & PR_HOST)
1682 				descend = 0;
1683 			else {
1684 				if (host != NULL)
1685 					strlcpy(tpr->pr_hostname,
1686 					    pr->pr_hostname,
1687 					    sizeof(tpr->pr_hostname));
1688 				if (domain != NULL)
1689 					strlcpy(tpr->pr_domainname,
1690 					    pr->pr_domainname,
1691 					    sizeof(tpr->pr_domainname));
1692 				if (uuid != NULL)
1693 					strlcpy(tpr->pr_hostuuid,
1694 					    pr->pr_hostuuid,
1695 					    sizeof(tpr->pr_hostuuid));
1696 				if (gothid)
1697 					tpr->pr_hostid = hid;
1698 			}
1699 		}
1700 	}
1701 	if ((tallow = ch_allow & ~pr_allow)) {
1702 		/* Clear allow bits in all children. */
1703 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1704 			tpr->pr_allow &= ~tallow;
1705 	}
1706 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1707 	/*
1708 	 * Persistent prisons get an extra reference, and prisons losing their
1709 	 * persist flag lose that reference.  Only do this for existing prisons
1710 	 * for now, so new ones will remain unseen until after the module
1711 	 * handlers have completed.
1712 	 */
1713 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1714 		if (pr_flags & PR_PERSIST) {
1715 			pr->pr_ref++;
1716 			pr->pr_uref++;
1717 		} else {
1718 			pr->pr_ref--;
1719 			pr->pr_uref--;
1720 		}
1721 	}
1722 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1723 	mtx_unlock(&pr->pr_mtx);
1724 
1725 #ifdef RACCT
1726 	if (created)
1727 		prison_racct_attach(pr);
1728 #endif
1729 
1730 	/* Locks may have prevented a complete restriction of child IP
1731 	 * addresses.  If so, allocate some more memory and try again.
1732 	 */
1733 #ifdef INET
1734 	while (redo_ip4) {
1735 		ip4s = pr->pr_ip4s;
1736 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1737 		mtx_lock(&pr->pr_mtx);
1738 		redo_ip4 = 0;
1739 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1740 #ifdef VIMAGE
1741 			if (tpr->pr_flags & PR_VNET) {
1742 				descend = 0;
1743 				continue;
1744 			}
1745 #endif
1746 			if (prison_restrict_ip4(tpr, ip4)) {
1747 				if (ip4 != NULL)
1748 					ip4 = NULL;
1749 				else
1750 					redo_ip4 = 1;
1751 			}
1752 		}
1753 		mtx_unlock(&pr->pr_mtx);
1754 	}
1755 #endif
1756 #ifdef INET6
1757 	while (redo_ip6) {
1758 		ip6s = pr->pr_ip6s;
1759 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1760 		mtx_lock(&pr->pr_mtx);
1761 		redo_ip6 = 0;
1762 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1763 #ifdef VIMAGE
1764 			if (tpr->pr_flags & PR_VNET) {
1765 				descend = 0;
1766 				continue;
1767 			}
1768 #endif
1769 			if (prison_restrict_ip6(tpr, ip6)) {
1770 				if (ip6 != NULL)
1771 					ip6 = NULL;
1772 				else
1773 					redo_ip6 = 1;
1774 			}
1775 		}
1776 		mtx_unlock(&pr->pr_mtx);
1777 	}
1778 #endif
1779 
1780 	/* Let the modules do their work. */
1781 	sx_downgrade(&allprison_lock);
1782 	if (created) {
1783 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1784 		if (error) {
1785 			prison_deref(pr, PD_LIST_SLOCKED);
1786 			goto done_errmsg;
1787 		}
1788 	}
1789 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1790 	if (error) {
1791 		prison_deref(pr, created
1792 		    ? PD_LIST_SLOCKED
1793 		    : PD_DEREF | PD_LIST_SLOCKED);
1794 		goto done_errmsg;
1795 	}
1796 
1797 	/* Attach this process to the prison if requested. */
1798 	if (flags & JAIL_ATTACH) {
1799 		mtx_lock(&pr->pr_mtx);
1800 		error = do_jail_attach(td, pr);
1801 		if (error) {
1802 			vfs_opterror(opts, "attach failed");
1803 			if (!created)
1804 				prison_deref(pr, PD_DEREF);
1805 			goto done_errmsg;
1806 		}
1807 	}
1808 
1809 #ifdef RACCT
1810 	if (!created) {
1811 		if (!(flags & JAIL_ATTACH))
1812 			sx_sunlock(&allprison_lock);
1813 		prison_racct_modify(pr);
1814 		if (!(flags & JAIL_ATTACH))
1815 			sx_slock(&allprison_lock);
1816 	}
1817 #endif
1818 
1819 	td->td_retval[0] = pr->pr_id;
1820 
1821 	/*
1822 	 * Now that it is all there, drop the temporary reference from existing
1823 	 * prisons.  Or add a reference to newly created persistent prisons
1824 	 * (which was not done earlier so that the prison would not be publicly
1825 	 * visible).
1826 	 */
1827 	if (!created) {
1828 		prison_deref(pr, (flags & JAIL_ATTACH)
1829 		    ? PD_DEREF
1830 		    : PD_DEREF | PD_LIST_SLOCKED);
1831 	} else {
1832 		if (pr_flags & PR_PERSIST) {
1833 			mtx_lock(&pr->pr_mtx);
1834 			pr->pr_ref++;
1835 			pr->pr_uref++;
1836 			mtx_unlock(&pr->pr_mtx);
1837 		}
1838 		if (!(flags & JAIL_ATTACH))
1839 			sx_sunlock(&allprison_lock);
1840 	}
1841 
1842 	goto done_errmsg;
1843 
1844  done_deref_locked:
1845 	prison_deref(pr, created
1846 	    ? PD_LOCKED | PD_LIST_XLOCKED
1847 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1848 	goto done_releroot;
1849  done_unlock_list:
1850 	sx_xunlock(&allprison_lock);
1851  done_releroot:
1852 	if (root != NULL)
1853 		vrele(root);
1854  done_errmsg:
1855 	if (error) {
1856 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1857 		if (errmsg_len > 0) {
1858 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1859 			if (errmsg_pos > 0) {
1860 				if (optuio->uio_segflg == UIO_SYSSPACE)
1861 					bcopy(errmsg,
1862 					   optuio->uio_iov[errmsg_pos].iov_base,
1863 					   errmsg_len);
1864 				else
1865 					copyout(errmsg,
1866 					   optuio->uio_iov[errmsg_pos].iov_base,
1867 					   errmsg_len);
1868 			}
1869 		}
1870 	}
1871  done_free:
1872 #ifdef INET
1873 	free(ip4, M_PRISON);
1874 #endif
1875 #ifdef INET6
1876 	free(ip6, M_PRISON);
1877 #endif
1878 	if (g_path != NULL)
1879 		free(g_path, M_TEMP);
1880 	vfs_freeopts(opts);
1881 	return (error);
1882 }
1883 
1884 
1885 /*
1886  * struct jail_get_args {
1887  *	struct iovec *iovp;
1888  *	unsigned int iovcnt;
1889  *	int flags;
1890  * };
1891  */
1892 int
1893 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1894 {
1895 	struct uio *auio;
1896 	int error;
1897 
1898 	/* Check that we have an even number of iovecs. */
1899 	if (uap->iovcnt & 1)
1900 		return (EINVAL);
1901 
1902 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1903 	if (error)
1904 		return (error);
1905 	error = kern_jail_get(td, auio, uap->flags);
1906 	if (error == 0)
1907 		error = copyout(auio->uio_iov, uap->iovp,
1908 		    uap->iovcnt * sizeof (struct iovec));
1909 	free(auio, M_IOV);
1910 	return (error);
1911 }
1912 
1913 int
1914 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1915 {
1916 	struct prison *pr, *mypr;
1917 	struct vfsopt *opt;
1918 	struct vfsoptlist *opts;
1919 	char *errmsg, *name;
1920 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1921 
1922 	if (flags & ~JAIL_GET_MASK)
1923 		return (EINVAL);
1924 
1925 	/* Get the parameter list. */
1926 	error = vfs_buildopts(optuio, &opts);
1927 	if (error)
1928 		return (error);
1929 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1930 	mypr = td->td_ucred->cr_prison;
1931 
1932 	/*
1933 	 * Find the prison specified by one of: lastjid, jid, name.
1934 	 */
1935 	sx_slock(&allprison_lock);
1936 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1937 	if (error == 0) {
1938 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1939 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1940 				mtx_lock(&pr->pr_mtx);
1941 				if (pr->pr_ref > 0 &&
1942 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1943 					break;
1944 				mtx_unlock(&pr->pr_mtx);
1945 			}
1946 		}
1947 		if (pr != NULL)
1948 			goto found_prison;
1949 		error = ENOENT;
1950 		vfs_opterror(opts, "no jail after %d", jid);
1951 		goto done_unlock_list;
1952 	} else if (error != ENOENT)
1953 		goto done_unlock_list;
1954 
1955 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1956 	if (error == 0) {
1957 		if (jid != 0) {
1958 			pr = prison_find_child(mypr, jid);
1959 			if (pr != NULL) {
1960 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1961 					mtx_unlock(&pr->pr_mtx);
1962 					error = ENOENT;
1963 					vfs_opterror(opts, "jail %d is dying",
1964 					    jid);
1965 					goto done_unlock_list;
1966 				}
1967 				goto found_prison;
1968 			}
1969 			error = ENOENT;
1970 			vfs_opterror(opts, "jail %d not found", jid);
1971 			goto done_unlock_list;
1972 		}
1973 	} else if (error != ENOENT)
1974 		goto done_unlock_list;
1975 
1976 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1977 	if (error == 0) {
1978 		if (len == 0 || name[len - 1] != '\0') {
1979 			error = EINVAL;
1980 			goto done_unlock_list;
1981 		}
1982 		pr = prison_find_name(mypr, name);
1983 		if (pr != NULL) {
1984 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1985 				mtx_unlock(&pr->pr_mtx);
1986 				error = ENOENT;
1987 				vfs_opterror(opts, "jail \"%s\" is dying",
1988 				    name);
1989 				goto done_unlock_list;
1990 			}
1991 			goto found_prison;
1992 		}
1993 		error = ENOENT;
1994 		vfs_opterror(opts, "jail \"%s\" not found", name);
1995 		goto done_unlock_list;
1996 	} else if (error != ENOENT)
1997 		goto done_unlock_list;
1998 
1999 	vfs_opterror(opts, "no jail specified");
2000 	error = ENOENT;
2001 	goto done_unlock_list;
2002 
2003  found_prison:
2004 	/* Get the parameters of the prison. */
2005 	pr->pr_ref++;
2006 	locked = PD_LOCKED;
2007 	td->td_retval[0] = pr->pr_id;
2008 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2009 	if (error != 0 && error != ENOENT)
2010 		goto done_deref;
2011 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2012 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2013 	if (error != 0 && error != ENOENT)
2014 		goto done_deref;
2015 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2016 	if (error != 0 && error != ENOENT)
2017 		goto done_deref;
2018 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2019 	    sizeof(pr->pr_cpuset->cs_id));
2020 	if (error != 0 && error != ENOENT)
2021 		goto done_deref;
2022 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2023 	if (error != 0 && error != ENOENT)
2024 		goto done_deref;
2025 #ifdef INET
2026 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2027 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2028 	if (error != 0 && error != ENOENT)
2029 		goto done_deref;
2030 #endif
2031 #ifdef INET6
2032 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2033 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2034 	if (error != 0 && error != ENOENT)
2035 		goto done_deref;
2036 #endif
2037 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2038 	    sizeof(pr->pr_securelevel));
2039 	if (error != 0 && error != ENOENT)
2040 		goto done_deref;
2041 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2042 	    sizeof(pr->pr_childcount));
2043 	if (error != 0 && error != ENOENT)
2044 		goto done_deref;
2045 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2046 	    sizeof(pr->pr_childmax));
2047 	if (error != 0 && error != ENOENT)
2048 		goto done_deref;
2049 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2050 	if (error != 0 && error != ENOENT)
2051 		goto done_deref;
2052 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2053 	if (error != 0 && error != ENOENT)
2054 		goto done_deref;
2055 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2056 	if (error != 0 && error != ENOENT)
2057 		goto done_deref;
2058 #ifdef COMPAT_FREEBSD32
2059 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2060 		uint32_t hid32 = pr->pr_hostid;
2061 
2062 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2063 	} else
2064 #endif
2065 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2066 	    sizeof(pr->pr_hostid));
2067 	if (error != 0 && error != ENOENT)
2068 		goto done_deref;
2069 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2070 	    sizeof(pr->pr_enforce_statfs));
2071 	if (error != 0 && error != ENOENT)
2072 		goto done_deref;
2073 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2074 	    sizeof(pr->pr_devfs_rsnum));
2075 	if (error != 0 && error != ENOENT)
2076 		goto done_deref;
2077 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2078 	    fi++) {
2079 		if (pr_flag_names[fi] == NULL)
2080 			continue;
2081 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2082 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2083 		if (error != 0 && error != ENOENT)
2084 			goto done_deref;
2085 		i = !i;
2086 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2087 		if (error != 0 && error != ENOENT)
2088 			goto done_deref;
2089 	}
2090 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2091 	    fi++) {
2092 		i = pr->pr_flags &
2093 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2094 		i = pr_flag_jailsys[fi].disable &&
2095 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2096 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2097 		    : JAIL_SYS_INHERIT;
2098 		error =
2099 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2100 		if (error != 0 && error != ENOENT)
2101 			goto done_deref;
2102 	}
2103 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2104 	    fi++) {
2105 		if (pr_allow_names[fi] == NULL)
2106 			continue;
2107 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2108 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2109 		if (error != 0 && error != ENOENT)
2110 			goto done_deref;
2111 		i = !i;
2112 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2113 		if (error != 0 && error != ENOENT)
2114 			goto done_deref;
2115 	}
2116 	i = (pr->pr_uref == 0);
2117 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2118 	if (error != 0 && error != ENOENT)
2119 		goto done_deref;
2120 	i = !i;
2121 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 
2125 	/* Get the module parameters. */
2126 	mtx_unlock(&pr->pr_mtx);
2127 	locked = 0;
2128 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2129 	if (error)
2130 		goto done_deref;
2131 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2132 
2133 	/* By now, all parameters should have been noted. */
2134 	TAILQ_FOREACH(opt, opts, link) {
2135 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2136 			error = EINVAL;
2137 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2138 			goto done_errmsg;
2139 		}
2140 	}
2141 
2142 	/* Write the fetched parameters back to userspace. */
2143 	error = 0;
2144 	TAILQ_FOREACH(opt, opts, link) {
2145 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2146 			pos = 2 * opt->pos + 1;
2147 			optuio->uio_iov[pos].iov_len = opt->len;
2148 			if (opt->value != NULL) {
2149 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2150 					bcopy(opt->value,
2151 					    optuio->uio_iov[pos].iov_base,
2152 					    opt->len);
2153 				} else {
2154 					error = copyout(opt->value,
2155 					    optuio->uio_iov[pos].iov_base,
2156 					    opt->len);
2157 					if (error)
2158 						break;
2159 				}
2160 			}
2161 		}
2162 	}
2163 	goto done_errmsg;
2164 
2165  done_deref:
2166 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2167 	goto done_errmsg;
2168 
2169  done_unlock_list:
2170 	sx_sunlock(&allprison_lock);
2171  done_errmsg:
2172 	if (error && errmsg_pos >= 0) {
2173 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2174 		errmsg_pos = 2 * errmsg_pos + 1;
2175 		if (errmsg_len > 0) {
2176 			if (optuio->uio_segflg == UIO_SYSSPACE)
2177 				bcopy(errmsg,
2178 				    optuio->uio_iov[errmsg_pos].iov_base,
2179 				    errmsg_len);
2180 			else
2181 				copyout(errmsg,
2182 				    optuio->uio_iov[errmsg_pos].iov_base,
2183 				    errmsg_len);
2184 		}
2185 	}
2186 	vfs_freeopts(opts);
2187 	return (error);
2188 }
2189 
2190 
2191 /*
2192  * struct jail_remove_args {
2193  *	int jid;
2194  * };
2195  */
2196 int
2197 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2198 {
2199 	struct prison *pr, *cpr, *lpr, *tpr;
2200 	int descend, error;
2201 
2202 	error = priv_check(td, PRIV_JAIL_REMOVE);
2203 	if (error)
2204 		return (error);
2205 
2206 	sx_xlock(&allprison_lock);
2207 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2208 	if (pr == NULL) {
2209 		sx_xunlock(&allprison_lock);
2210 		return (EINVAL);
2211 	}
2212 
2213 	/* Remove all descendants of this prison, then remove this prison. */
2214 	pr->pr_ref++;
2215 	pr->pr_flags |= PR_REMOVE;
2216 	if (!LIST_EMPTY(&pr->pr_children)) {
2217 		mtx_unlock(&pr->pr_mtx);
2218 		lpr = NULL;
2219 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2220 			mtx_lock(&cpr->pr_mtx);
2221 			if (cpr->pr_ref > 0) {
2222 				tpr = cpr;
2223 				cpr->pr_ref++;
2224 				cpr->pr_flags |= PR_REMOVE;
2225 			} else {
2226 				/* Already removed - do not do it again. */
2227 				tpr = NULL;
2228 			}
2229 			mtx_unlock(&cpr->pr_mtx);
2230 			if (lpr != NULL) {
2231 				mtx_lock(&lpr->pr_mtx);
2232 				prison_remove_one(lpr);
2233 				sx_xlock(&allprison_lock);
2234 			}
2235 			lpr = tpr;
2236 		}
2237 		if (lpr != NULL) {
2238 			mtx_lock(&lpr->pr_mtx);
2239 			prison_remove_one(lpr);
2240 			sx_xlock(&allprison_lock);
2241 		}
2242 		mtx_lock(&pr->pr_mtx);
2243 	}
2244 	prison_remove_one(pr);
2245 	return (0);
2246 }
2247 
2248 static void
2249 prison_remove_one(struct prison *pr)
2250 {
2251 	struct proc *p;
2252 	int deuref;
2253 
2254 	/* If the prison was persistent, it is not anymore. */
2255 	deuref = 0;
2256 	if (pr->pr_flags & PR_PERSIST) {
2257 		pr->pr_ref--;
2258 		deuref = PD_DEUREF;
2259 		pr->pr_flags &= ~PR_PERSIST;
2260 	}
2261 
2262 	/*
2263 	 * jail_remove added a reference.  If that's the only one, remove
2264 	 * the prison now.
2265 	 */
2266 	KASSERT(pr->pr_ref > 0,
2267 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2268 	if (pr->pr_ref == 1) {
2269 		prison_deref(pr,
2270 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2271 		return;
2272 	}
2273 
2274 	mtx_unlock(&pr->pr_mtx);
2275 	sx_xunlock(&allprison_lock);
2276 	/*
2277 	 * Kill all processes unfortunate enough to be attached to this prison.
2278 	 */
2279 	sx_slock(&allproc_lock);
2280 	LIST_FOREACH(p, &allproc, p_list) {
2281 		PROC_LOCK(p);
2282 		if (p->p_state != PRS_NEW && p->p_ucred &&
2283 		    p->p_ucred->cr_prison == pr)
2284 			kern_psignal(p, SIGKILL);
2285 		PROC_UNLOCK(p);
2286 	}
2287 	sx_sunlock(&allproc_lock);
2288 	/* Remove the temporary reference added by jail_remove. */
2289 	prison_deref(pr, deuref | PD_DEREF);
2290 }
2291 
2292 
2293 /*
2294  * struct jail_attach_args {
2295  *	int jid;
2296  * };
2297  */
2298 int
2299 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2300 {
2301 	struct prison *pr;
2302 	int error;
2303 
2304 	error = priv_check(td, PRIV_JAIL_ATTACH);
2305 	if (error)
2306 		return (error);
2307 
2308 	sx_slock(&allprison_lock);
2309 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2310 	if (pr == NULL) {
2311 		sx_sunlock(&allprison_lock);
2312 		return (EINVAL);
2313 	}
2314 
2315 	/*
2316 	 * Do not allow a process to attach to a prison that is not
2317 	 * considered to be "alive".
2318 	 */
2319 	if (pr->pr_uref == 0) {
2320 		mtx_unlock(&pr->pr_mtx);
2321 		sx_sunlock(&allprison_lock);
2322 		return (EINVAL);
2323 	}
2324 
2325 	return (do_jail_attach(td, pr));
2326 }
2327 
2328 static int
2329 do_jail_attach(struct thread *td, struct prison *pr)
2330 {
2331 	struct prison *ppr;
2332 	struct proc *p;
2333 	struct ucred *newcred, *oldcred;
2334 	int error;
2335 
2336 	/*
2337 	 * XXX: Note that there is a slight race here if two threads
2338 	 * in the same privileged process attempt to attach to two
2339 	 * different jails at the same time.  It is important for
2340 	 * user processes not to do this, or they might end up with
2341 	 * a process root from one prison, but attached to the jail
2342 	 * of another.
2343 	 */
2344 	pr->pr_ref++;
2345 	pr->pr_uref++;
2346 	mtx_unlock(&pr->pr_mtx);
2347 
2348 	/* Let modules do whatever they need to prepare for attaching. */
2349 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2350 	if (error) {
2351 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2352 		return (error);
2353 	}
2354 	sx_sunlock(&allprison_lock);
2355 
2356 	/*
2357 	 * Reparent the newly attached process to this jail.
2358 	 */
2359 	ppr = td->td_ucred->cr_prison;
2360 	p = td->td_proc;
2361 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2362 	if (error)
2363 		goto e_revert_osd;
2364 
2365 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2366 	if ((error = change_dir(pr->pr_root, td)) != 0)
2367 		goto e_unlock;
2368 #ifdef MAC
2369 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2370 		goto e_unlock;
2371 #endif
2372 	VOP_UNLOCK(pr->pr_root, 0);
2373 	if ((error = change_root(pr->pr_root, td)))
2374 		goto e_revert_osd;
2375 
2376 	newcred = crget();
2377 	PROC_LOCK(p);
2378 	oldcred = p->p_ucred;
2379 	setsugid(p);
2380 	crcopy(newcred, oldcred);
2381 	newcred->cr_prison = pr;
2382 	p->p_ucred = newcred;
2383 	PROC_UNLOCK(p);
2384 #ifdef RACCT
2385 	racct_proc_ucred_changed(p, oldcred, newcred);
2386 #endif
2387 	crfree(oldcred);
2388 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2389 	return (0);
2390  e_unlock:
2391 	VOP_UNLOCK(pr->pr_root, 0);
2392  e_revert_osd:
2393 	/* Tell modules this thread is still in its old jail after all. */
2394 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2395 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2396 	return (error);
2397 }
2398 
2399 
2400 /*
2401  * Returns a locked prison instance, or NULL on failure.
2402  */
2403 struct prison *
2404 prison_find(int prid)
2405 {
2406 	struct prison *pr;
2407 
2408 	sx_assert(&allprison_lock, SX_LOCKED);
2409 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2410 		if (pr->pr_id == prid) {
2411 			mtx_lock(&pr->pr_mtx);
2412 			if (pr->pr_ref > 0)
2413 				return (pr);
2414 			mtx_unlock(&pr->pr_mtx);
2415 		}
2416 	}
2417 	return (NULL);
2418 }
2419 
2420 /*
2421  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2422  */
2423 struct prison *
2424 prison_find_child(struct prison *mypr, int prid)
2425 {
2426 	struct prison *pr;
2427 	int descend;
2428 
2429 	sx_assert(&allprison_lock, SX_LOCKED);
2430 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2431 		if (pr->pr_id == prid) {
2432 			mtx_lock(&pr->pr_mtx);
2433 			if (pr->pr_ref > 0)
2434 				return (pr);
2435 			mtx_unlock(&pr->pr_mtx);
2436 		}
2437 	}
2438 	return (NULL);
2439 }
2440 
2441 /*
2442  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2443  */
2444 struct prison *
2445 prison_find_name(struct prison *mypr, const char *name)
2446 {
2447 	struct prison *pr, *deadpr;
2448 	size_t mylen;
2449 	int descend;
2450 
2451 	sx_assert(&allprison_lock, SX_LOCKED);
2452 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2453  again:
2454 	deadpr = NULL;
2455 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2456 		if (!strcmp(pr->pr_name + mylen, name)) {
2457 			mtx_lock(&pr->pr_mtx);
2458 			if (pr->pr_ref > 0) {
2459 				if (pr->pr_uref > 0)
2460 					return (pr);
2461 				deadpr = pr;
2462 			}
2463 			mtx_unlock(&pr->pr_mtx);
2464 		}
2465 	}
2466 	/* There was no valid prison - perhaps there was a dying one. */
2467 	if (deadpr != NULL) {
2468 		mtx_lock(&deadpr->pr_mtx);
2469 		if (deadpr->pr_ref == 0) {
2470 			mtx_unlock(&deadpr->pr_mtx);
2471 			goto again;
2472 		}
2473 	}
2474 	return (deadpr);
2475 }
2476 
2477 /*
2478  * See if a prison has the specific flag set.
2479  */
2480 int
2481 prison_flag(struct ucred *cred, unsigned flag)
2482 {
2483 
2484 	/* This is an atomic read, so no locking is necessary. */
2485 	return (cred->cr_prison->pr_flags & flag);
2486 }
2487 
2488 int
2489 prison_allow(struct ucred *cred, unsigned flag)
2490 {
2491 
2492 	/* This is an atomic read, so no locking is necessary. */
2493 	return (cred->cr_prison->pr_allow & flag);
2494 }
2495 
2496 /*
2497  * Remove a prison reference.  If that was the last reference, remove the
2498  * prison itself - but not in this context in case there are locks held.
2499  */
2500 void
2501 prison_free_locked(struct prison *pr)
2502 {
2503 
2504 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2505 	pr->pr_ref--;
2506 	if (pr->pr_ref == 0) {
2507 		mtx_unlock(&pr->pr_mtx);
2508 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2509 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2510 		return;
2511 	}
2512 	mtx_unlock(&pr->pr_mtx);
2513 }
2514 
2515 void
2516 prison_free(struct prison *pr)
2517 {
2518 
2519 	mtx_lock(&pr->pr_mtx);
2520 	prison_free_locked(pr);
2521 }
2522 
2523 static void
2524 prison_complete(void *context, int pending)
2525 {
2526 
2527 	prison_deref((struct prison *)context, 0);
2528 }
2529 
2530 /*
2531  * Remove a prison reference (usually).  This internal version assumes no
2532  * mutexes are held, except perhaps the prison itself.  If there are no more
2533  * references, release and delist the prison.  On completion, the prison lock
2534  * and the allprison lock are both unlocked.
2535  */
2536 static void
2537 prison_deref(struct prison *pr, int flags)
2538 {
2539 	struct prison *ppr, *tpr;
2540 
2541 	if (!(flags & PD_LOCKED))
2542 		mtx_lock(&pr->pr_mtx);
2543 	for (;;) {
2544 		if (flags & PD_DEUREF) {
2545 			pr->pr_uref--;
2546 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2547 		}
2548 		if (flags & PD_DEREF)
2549 			pr->pr_ref--;
2550 		/* If the prison still has references, nothing else to do. */
2551 		if (pr->pr_ref > 0) {
2552 			mtx_unlock(&pr->pr_mtx);
2553 			if (flags & PD_LIST_SLOCKED)
2554 				sx_sunlock(&allprison_lock);
2555 			else if (flags & PD_LIST_XLOCKED)
2556 				sx_xunlock(&allprison_lock);
2557 			return;
2558 		}
2559 
2560 		mtx_unlock(&pr->pr_mtx);
2561 		if (flags & PD_LIST_SLOCKED) {
2562 			if (!sx_try_upgrade(&allprison_lock)) {
2563 				sx_sunlock(&allprison_lock);
2564 				sx_xlock(&allprison_lock);
2565 			}
2566 		} else if (!(flags & PD_LIST_XLOCKED))
2567 			sx_xlock(&allprison_lock);
2568 
2569 		TAILQ_REMOVE(&allprison, pr, pr_list);
2570 		LIST_REMOVE(pr, pr_sibling);
2571 		ppr = pr->pr_parent;
2572 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2573 			tpr->pr_childcount--;
2574 		sx_xunlock(&allprison_lock);
2575 
2576 #ifdef VIMAGE
2577 		if (pr->pr_vnet != ppr->pr_vnet)
2578 			vnet_destroy(pr->pr_vnet);
2579 #endif
2580 		if (pr->pr_root != NULL)
2581 			vrele(pr->pr_root);
2582 		mtx_destroy(&pr->pr_mtx);
2583 #ifdef INET
2584 		free(pr->pr_ip4, M_PRISON);
2585 #endif
2586 #ifdef INET6
2587 		free(pr->pr_ip6, M_PRISON);
2588 #endif
2589 		if (pr->pr_cpuset != NULL)
2590 			cpuset_rel(pr->pr_cpuset);
2591 		osd_jail_exit(pr);
2592 #ifdef RACCT
2593 		prison_racct_detach(pr);
2594 #endif
2595 		free(pr, M_PRISON);
2596 
2597 		/* Removing a prison frees a reference on its parent. */
2598 		pr = ppr;
2599 		mtx_lock(&pr->pr_mtx);
2600 		flags = PD_DEREF | PD_DEUREF;
2601 	}
2602 }
2603 
2604 void
2605 prison_hold_locked(struct prison *pr)
2606 {
2607 
2608 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2609 	KASSERT(pr->pr_ref > 0,
2610 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2611 	pr->pr_ref++;
2612 }
2613 
2614 void
2615 prison_hold(struct prison *pr)
2616 {
2617 
2618 	mtx_lock(&pr->pr_mtx);
2619 	prison_hold_locked(pr);
2620 	mtx_unlock(&pr->pr_mtx);
2621 }
2622 
2623 void
2624 prison_proc_hold(struct prison *pr)
2625 {
2626 
2627 	mtx_lock(&pr->pr_mtx);
2628 	KASSERT(pr->pr_uref > 0,
2629 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2630 	pr->pr_uref++;
2631 	mtx_unlock(&pr->pr_mtx);
2632 }
2633 
2634 void
2635 prison_proc_free(struct prison *pr)
2636 {
2637 
2638 	mtx_lock(&pr->pr_mtx);
2639 	KASSERT(pr->pr_uref > 0,
2640 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2641 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2642 }
2643 
2644 
2645 #ifdef INET
2646 /*
2647  * Restrict a prison's IP address list with its parent's, possibly replacing
2648  * it.  Return true if the replacement buffer was used (or would have been).
2649  */
2650 static int
2651 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2652 {
2653 	int ii, ij, used;
2654 	struct prison *ppr;
2655 
2656 	ppr = pr->pr_parent;
2657 	if (!(pr->pr_flags & PR_IP4_USER)) {
2658 		/* This has no user settings, so just copy the parent's list. */
2659 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2660 			/*
2661 			 * There's no room for the parent's list.  Use the
2662 			 * new list buffer, which is assumed to be big enough
2663 			 * (if it was passed).  If there's no buffer, try to
2664 			 * allocate one.
2665 			 */
2666 			used = 1;
2667 			if (newip4 == NULL) {
2668 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2669 				    M_PRISON, M_NOWAIT);
2670 				if (newip4 != NULL)
2671 					used = 0;
2672 			}
2673 			if (newip4 != NULL) {
2674 				bcopy(ppr->pr_ip4, newip4,
2675 				    ppr->pr_ip4s * sizeof(*newip4));
2676 				free(pr->pr_ip4, M_PRISON);
2677 				pr->pr_ip4 = newip4;
2678 				pr->pr_ip4s = ppr->pr_ip4s;
2679 			}
2680 			return (used);
2681 		}
2682 		pr->pr_ip4s = ppr->pr_ip4s;
2683 		if (pr->pr_ip4s > 0)
2684 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2685 			    pr->pr_ip4s * sizeof(*newip4));
2686 		else if (pr->pr_ip4 != NULL) {
2687 			free(pr->pr_ip4, M_PRISON);
2688 			pr->pr_ip4 = NULL;
2689 		}
2690 	} else if (pr->pr_ip4s > 0) {
2691 		/* Remove addresses that aren't in the parent. */
2692 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2693 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2694 				break;
2695 		if (ij < ppr->pr_ip4s)
2696 			ii = 1;
2697 		else {
2698 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2699 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2700 			ii = 0;
2701 		}
2702 		for (ij = 1; ii < pr->pr_ip4s; ) {
2703 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2704 				ii++;
2705 				continue;
2706 			}
2707 			switch (ij >= ppr->pr_ip4s ? -1 :
2708 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2709 			case -1:
2710 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2711 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2712 				break;
2713 			case 0:
2714 				ii++;
2715 				ij++;
2716 				break;
2717 			case 1:
2718 				ij++;
2719 				break;
2720 			}
2721 		}
2722 		if (pr->pr_ip4s == 0) {
2723 			free(pr->pr_ip4, M_PRISON);
2724 			pr->pr_ip4 = NULL;
2725 		}
2726 	}
2727 	return (0);
2728 }
2729 
2730 /*
2731  * Pass back primary IPv4 address of this jail.
2732  *
2733  * If not restricted return success but do not alter the address.  Caller has
2734  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2735  *
2736  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2737  * Address returned in NBO.
2738  */
2739 int
2740 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2741 {
2742 	struct prison *pr;
2743 
2744 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2745 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2746 
2747 	pr = cred->cr_prison;
2748 	if (!(pr->pr_flags & PR_IP4))
2749 		return (0);
2750 	mtx_lock(&pr->pr_mtx);
2751 	if (!(pr->pr_flags & PR_IP4)) {
2752 		mtx_unlock(&pr->pr_mtx);
2753 		return (0);
2754 	}
2755 	if (pr->pr_ip4 == NULL) {
2756 		mtx_unlock(&pr->pr_mtx);
2757 		return (EAFNOSUPPORT);
2758 	}
2759 
2760 	ia->s_addr = pr->pr_ip4[0].s_addr;
2761 	mtx_unlock(&pr->pr_mtx);
2762 	return (0);
2763 }
2764 
2765 /*
2766  * Return 1 if we should do proper source address selection or are not jailed.
2767  * We will return 0 if we should bypass source address selection in favour
2768  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2769  * returned in NBO.
2770  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2771  */
2772 int
2773 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2774 {
2775 	struct prison *pr;
2776 	struct in_addr lia;
2777 	int error;
2778 
2779 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2780 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2781 
2782 	if (!jailed(cred))
2783 		return (1);
2784 
2785 	pr = cred->cr_prison;
2786 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2787 		return (1);
2788 
2789 	lia.s_addr = INADDR_ANY;
2790 	error = prison_get_ip4(cred, &lia);
2791 	if (error)
2792 		return (error);
2793 	if (lia.s_addr == INADDR_ANY)
2794 		return (1);
2795 
2796 	ia->s_addr = lia.s_addr;
2797 	return (0);
2798 }
2799 
2800 /*
2801  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2802  */
2803 int
2804 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2805 {
2806 
2807 	if (pr1 == pr2)
2808 		return (1);
2809 
2810 	/*
2811 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2812 	 * existing prisons.
2813 	 */
2814 	while (pr1 != &prison0 &&
2815 #ifdef VIMAGE
2816 	       !(pr1->pr_flags & PR_VNET) &&
2817 #endif
2818 	       !(pr1->pr_flags & PR_IP4_USER))
2819 		pr1 = pr1->pr_parent;
2820 	while (pr2 != &prison0 &&
2821 #ifdef VIMAGE
2822 	       !(pr2->pr_flags & PR_VNET) &&
2823 #endif
2824 	       !(pr2->pr_flags & PR_IP4_USER))
2825 		pr2 = pr2->pr_parent;
2826 	return (pr1 == pr2);
2827 }
2828 
2829 /*
2830  * Make sure our (source) address is set to something meaningful to this
2831  * jail.
2832  *
2833  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2834  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2835  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2836  */
2837 int
2838 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2839 {
2840 	struct prison *pr;
2841 	struct in_addr ia0;
2842 	int error;
2843 
2844 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2845 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2846 
2847 	pr = cred->cr_prison;
2848 	if (!(pr->pr_flags & PR_IP4))
2849 		return (0);
2850 	mtx_lock(&pr->pr_mtx);
2851 	if (!(pr->pr_flags & PR_IP4)) {
2852 		mtx_unlock(&pr->pr_mtx);
2853 		return (0);
2854 	}
2855 	if (pr->pr_ip4 == NULL) {
2856 		mtx_unlock(&pr->pr_mtx);
2857 		return (EAFNOSUPPORT);
2858 	}
2859 
2860 	ia0.s_addr = ntohl(ia->s_addr);
2861 	if (ia0.s_addr == INADDR_LOOPBACK) {
2862 		ia->s_addr = pr->pr_ip4[0].s_addr;
2863 		mtx_unlock(&pr->pr_mtx);
2864 		return (0);
2865 	}
2866 
2867 	if (ia0.s_addr == INADDR_ANY) {
2868 		/*
2869 		 * In case there is only 1 IPv4 address, bind directly.
2870 		 */
2871 		if (pr->pr_ip4s == 1)
2872 			ia->s_addr = pr->pr_ip4[0].s_addr;
2873 		mtx_unlock(&pr->pr_mtx);
2874 		return (0);
2875 	}
2876 
2877 	error = _prison_check_ip4(pr, ia);
2878 	mtx_unlock(&pr->pr_mtx);
2879 	return (error);
2880 }
2881 
2882 /*
2883  * Rewrite destination address in case we will connect to loopback address.
2884  *
2885  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2886  * Address passed in in NBO and returned in NBO.
2887  */
2888 int
2889 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2890 {
2891 	struct prison *pr;
2892 
2893 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2894 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2895 
2896 	pr = cred->cr_prison;
2897 	if (!(pr->pr_flags & PR_IP4))
2898 		return (0);
2899 	mtx_lock(&pr->pr_mtx);
2900 	if (!(pr->pr_flags & PR_IP4)) {
2901 		mtx_unlock(&pr->pr_mtx);
2902 		return (0);
2903 	}
2904 	if (pr->pr_ip4 == NULL) {
2905 		mtx_unlock(&pr->pr_mtx);
2906 		return (EAFNOSUPPORT);
2907 	}
2908 
2909 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2910 		ia->s_addr = pr->pr_ip4[0].s_addr;
2911 		mtx_unlock(&pr->pr_mtx);
2912 		return (0);
2913 	}
2914 
2915 	/*
2916 	 * Return success because nothing had to be changed.
2917 	 */
2918 	mtx_unlock(&pr->pr_mtx);
2919 	return (0);
2920 }
2921 
2922 /*
2923  * Check if given address belongs to the jail referenced by cred/prison.
2924  *
2925  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2926  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2927  * doesn't allow IPv4.  Address passed in in NBO.
2928  */
2929 static int
2930 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
2931 {
2932 	int i, a, z, d;
2933 
2934 	/*
2935 	 * Check the primary IP.
2936 	 */
2937 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2938 		return (0);
2939 
2940 	/*
2941 	 * All the other IPs are sorted so we can do a binary search.
2942 	 */
2943 	a = 0;
2944 	z = pr->pr_ip4s - 2;
2945 	while (a <= z) {
2946 		i = (a + z) / 2;
2947 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2948 		if (d > 0)
2949 			z = i - 1;
2950 		else if (d < 0)
2951 			a = i + 1;
2952 		else
2953 			return (0);
2954 	}
2955 
2956 	return (EADDRNOTAVAIL);
2957 }
2958 
2959 int
2960 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
2961 {
2962 	struct prison *pr;
2963 	int error;
2964 
2965 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2966 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2967 
2968 	pr = cred->cr_prison;
2969 	if (!(pr->pr_flags & PR_IP4))
2970 		return (0);
2971 	mtx_lock(&pr->pr_mtx);
2972 	if (!(pr->pr_flags & PR_IP4)) {
2973 		mtx_unlock(&pr->pr_mtx);
2974 		return (0);
2975 	}
2976 	if (pr->pr_ip4 == NULL) {
2977 		mtx_unlock(&pr->pr_mtx);
2978 		return (EAFNOSUPPORT);
2979 	}
2980 
2981 	error = _prison_check_ip4(pr, ia);
2982 	mtx_unlock(&pr->pr_mtx);
2983 	return (error);
2984 }
2985 #endif
2986 
2987 #ifdef INET6
2988 static int
2989 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2990 {
2991 	int ii, ij, used;
2992 	struct prison *ppr;
2993 
2994 	ppr = pr->pr_parent;
2995 	if (!(pr->pr_flags & PR_IP6_USER)) {
2996 		/* This has no user settings, so just copy the parent's list. */
2997 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2998 			/*
2999 			 * There's no room for the parent's list.  Use the
3000 			 * new list buffer, which is assumed to be big enough
3001 			 * (if it was passed).  If there's no buffer, try to
3002 			 * allocate one.
3003 			 */
3004 			used = 1;
3005 			if (newip6 == NULL) {
3006 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3007 				    M_PRISON, M_NOWAIT);
3008 				if (newip6 != NULL)
3009 					used = 0;
3010 			}
3011 			if (newip6 != NULL) {
3012 				bcopy(ppr->pr_ip6, newip6,
3013 				    ppr->pr_ip6s * sizeof(*newip6));
3014 				free(pr->pr_ip6, M_PRISON);
3015 				pr->pr_ip6 = newip6;
3016 				pr->pr_ip6s = ppr->pr_ip6s;
3017 			}
3018 			return (used);
3019 		}
3020 		pr->pr_ip6s = ppr->pr_ip6s;
3021 		if (pr->pr_ip6s > 0)
3022 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3023 			    pr->pr_ip6s * sizeof(*newip6));
3024 		else if (pr->pr_ip6 != NULL) {
3025 			free(pr->pr_ip6, M_PRISON);
3026 			pr->pr_ip6 = NULL;
3027 		}
3028 	} else if (pr->pr_ip6s > 0) {
3029 		/* Remove addresses that aren't in the parent. */
3030 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3031 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3032 			    &ppr->pr_ip6[ij]))
3033 				break;
3034 		if (ij < ppr->pr_ip6s)
3035 			ii = 1;
3036 		else {
3037 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3038 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3039 			ii = 0;
3040 		}
3041 		for (ij = 1; ii < pr->pr_ip6s; ) {
3042 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3043 			    &ppr->pr_ip6[0])) {
3044 				ii++;
3045 				continue;
3046 			}
3047 			switch (ij >= ppr->pr_ip6s ? -1 :
3048 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3049 			case -1:
3050 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3051 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3052 				break;
3053 			case 0:
3054 				ii++;
3055 				ij++;
3056 				break;
3057 			case 1:
3058 				ij++;
3059 				break;
3060 			}
3061 		}
3062 		if (pr->pr_ip6s == 0) {
3063 			free(pr->pr_ip6, M_PRISON);
3064 			pr->pr_ip6 = NULL;
3065 		}
3066 	}
3067 	return 0;
3068 }
3069 
3070 /*
3071  * Pass back primary IPv6 address for this jail.
3072  *
3073  * If not restricted return success but do not alter the address.  Caller has
3074  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3075  *
3076  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3077  */
3078 int
3079 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3080 {
3081 	struct prison *pr;
3082 
3083 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3084 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3085 
3086 	pr = cred->cr_prison;
3087 	if (!(pr->pr_flags & PR_IP6))
3088 		return (0);
3089 	mtx_lock(&pr->pr_mtx);
3090 	if (!(pr->pr_flags & PR_IP6)) {
3091 		mtx_unlock(&pr->pr_mtx);
3092 		return (0);
3093 	}
3094 	if (pr->pr_ip6 == NULL) {
3095 		mtx_unlock(&pr->pr_mtx);
3096 		return (EAFNOSUPPORT);
3097 	}
3098 
3099 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3100 	mtx_unlock(&pr->pr_mtx);
3101 	return (0);
3102 }
3103 
3104 /*
3105  * Return 1 if we should do proper source address selection or are not jailed.
3106  * We will return 0 if we should bypass source address selection in favour
3107  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3108  * returned in NBO.
3109  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3110  */
3111 int
3112 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3113 {
3114 	struct prison *pr;
3115 	struct in6_addr lia6;
3116 	int error;
3117 
3118 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3119 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3120 
3121 	if (!jailed(cred))
3122 		return (1);
3123 
3124 	pr = cred->cr_prison;
3125 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3126 		return (1);
3127 
3128 	lia6 = in6addr_any;
3129 	error = prison_get_ip6(cred, &lia6);
3130 	if (error)
3131 		return (error);
3132 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3133 		return (1);
3134 
3135 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3136 	return (0);
3137 }
3138 
3139 /*
3140  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3141  */
3142 int
3143 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3144 {
3145 
3146 	if (pr1 == pr2)
3147 		return (1);
3148 
3149 	while (pr1 != &prison0 &&
3150 #ifdef VIMAGE
3151 	       !(pr1->pr_flags & PR_VNET) &&
3152 #endif
3153 	       !(pr1->pr_flags & PR_IP6_USER))
3154 		pr1 = pr1->pr_parent;
3155 	while (pr2 != &prison0 &&
3156 #ifdef VIMAGE
3157 	       !(pr2->pr_flags & PR_VNET) &&
3158 #endif
3159 	       !(pr2->pr_flags & PR_IP6_USER))
3160 		pr2 = pr2->pr_parent;
3161 	return (pr1 == pr2);
3162 }
3163 
3164 /*
3165  * Make sure our (source) address is set to something meaningful to this jail.
3166  *
3167  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3168  * when needed while binding.
3169  *
3170  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3171  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3172  * doesn't allow IPv6.
3173  */
3174 int
3175 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3176 {
3177 	struct prison *pr;
3178 	int error;
3179 
3180 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3181 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3182 
3183 	pr = cred->cr_prison;
3184 	if (!(pr->pr_flags & PR_IP6))
3185 		return (0);
3186 	mtx_lock(&pr->pr_mtx);
3187 	if (!(pr->pr_flags & PR_IP6)) {
3188 		mtx_unlock(&pr->pr_mtx);
3189 		return (0);
3190 	}
3191 	if (pr->pr_ip6 == NULL) {
3192 		mtx_unlock(&pr->pr_mtx);
3193 		return (EAFNOSUPPORT);
3194 	}
3195 
3196 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3197 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3198 		mtx_unlock(&pr->pr_mtx);
3199 		return (0);
3200 	}
3201 
3202 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3203 		/*
3204 		 * In case there is only 1 IPv6 address, and v6only is true,
3205 		 * then bind directly.
3206 		 */
3207 		if (v6only != 0 && pr->pr_ip6s == 1)
3208 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3209 		mtx_unlock(&pr->pr_mtx);
3210 		return (0);
3211 	}
3212 
3213 	error = _prison_check_ip6(pr, ia6);
3214 	mtx_unlock(&pr->pr_mtx);
3215 	return (error);
3216 }
3217 
3218 /*
3219  * Rewrite destination address in case we will connect to loopback address.
3220  *
3221  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3222  */
3223 int
3224 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3225 {
3226 	struct prison *pr;
3227 
3228 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3229 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3230 
3231 	pr = cred->cr_prison;
3232 	if (!(pr->pr_flags & PR_IP6))
3233 		return (0);
3234 	mtx_lock(&pr->pr_mtx);
3235 	if (!(pr->pr_flags & PR_IP6)) {
3236 		mtx_unlock(&pr->pr_mtx);
3237 		return (0);
3238 	}
3239 	if (pr->pr_ip6 == NULL) {
3240 		mtx_unlock(&pr->pr_mtx);
3241 		return (EAFNOSUPPORT);
3242 	}
3243 
3244 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3245 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3246 		mtx_unlock(&pr->pr_mtx);
3247 		return (0);
3248 	}
3249 
3250 	/*
3251 	 * Return success because nothing had to be changed.
3252 	 */
3253 	mtx_unlock(&pr->pr_mtx);
3254 	return (0);
3255 }
3256 
3257 /*
3258  * Check if given address belongs to the jail referenced by cred/prison.
3259  *
3260  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3261  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3262  * doesn't allow IPv6.
3263  */
3264 static int
3265 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3266 {
3267 	int i, a, z, d;
3268 
3269 	/*
3270 	 * Check the primary IP.
3271 	 */
3272 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3273 		return (0);
3274 
3275 	/*
3276 	 * All the other IPs are sorted so we can do a binary search.
3277 	 */
3278 	a = 0;
3279 	z = pr->pr_ip6s - 2;
3280 	while (a <= z) {
3281 		i = (a + z) / 2;
3282 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3283 		if (d > 0)
3284 			z = i - 1;
3285 		else if (d < 0)
3286 			a = i + 1;
3287 		else
3288 			return (0);
3289 	}
3290 
3291 	return (EADDRNOTAVAIL);
3292 }
3293 
3294 int
3295 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3296 {
3297 	struct prison *pr;
3298 	int error;
3299 
3300 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3301 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3302 
3303 	pr = cred->cr_prison;
3304 	if (!(pr->pr_flags & PR_IP6))
3305 		return (0);
3306 	mtx_lock(&pr->pr_mtx);
3307 	if (!(pr->pr_flags & PR_IP6)) {
3308 		mtx_unlock(&pr->pr_mtx);
3309 		return (0);
3310 	}
3311 	if (pr->pr_ip6 == NULL) {
3312 		mtx_unlock(&pr->pr_mtx);
3313 		return (EAFNOSUPPORT);
3314 	}
3315 
3316 	error = _prison_check_ip6(pr, ia6);
3317 	mtx_unlock(&pr->pr_mtx);
3318 	return (error);
3319 }
3320 #endif
3321 
3322 /*
3323  * Check if a jail supports the given address family.
3324  *
3325  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3326  * if not.
3327  */
3328 int
3329 prison_check_af(struct ucred *cred, int af)
3330 {
3331 	struct prison *pr;
3332 	int error;
3333 
3334 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3335 
3336 	pr = cred->cr_prison;
3337 #ifdef VIMAGE
3338 	/* Prisons with their own network stack are not limited. */
3339 	if (prison_owns_vnet(cred))
3340 		return (0);
3341 #endif
3342 
3343 	error = 0;
3344 	switch (af)
3345 	{
3346 #ifdef INET
3347 	case AF_INET:
3348 		if (pr->pr_flags & PR_IP4)
3349 		{
3350 			mtx_lock(&pr->pr_mtx);
3351 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3352 				error = EAFNOSUPPORT;
3353 			mtx_unlock(&pr->pr_mtx);
3354 		}
3355 		break;
3356 #endif
3357 #ifdef INET6
3358 	case AF_INET6:
3359 		if (pr->pr_flags & PR_IP6)
3360 		{
3361 			mtx_lock(&pr->pr_mtx);
3362 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3363 				error = EAFNOSUPPORT;
3364 			mtx_unlock(&pr->pr_mtx);
3365 		}
3366 		break;
3367 #endif
3368 	case AF_LOCAL:
3369 	case AF_ROUTE:
3370 		break;
3371 	default:
3372 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3373 			error = EAFNOSUPPORT;
3374 	}
3375 	return (error);
3376 }
3377 
3378 /*
3379  * Check if given address belongs to the jail referenced by cred (wrapper to
3380  * prison_check_ip[46]).
3381  *
3382  * Returns 0 if jail doesn't restrict the address family or if address belongs
3383  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3384  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3385  */
3386 int
3387 prison_if(struct ucred *cred, struct sockaddr *sa)
3388 {
3389 #ifdef INET
3390 	struct sockaddr_in *sai;
3391 #endif
3392 #ifdef INET6
3393 	struct sockaddr_in6 *sai6;
3394 #endif
3395 	int error;
3396 
3397 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3398 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3399 
3400 #ifdef VIMAGE
3401 	if (prison_owns_vnet(cred))
3402 		return (0);
3403 #endif
3404 
3405 	error = 0;
3406 	switch (sa->sa_family)
3407 	{
3408 #ifdef INET
3409 	case AF_INET:
3410 		sai = (struct sockaddr_in *)sa;
3411 		error = prison_check_ip4(cred, &sai->sin_addr);
3412 		break;
3413 #endif
3414 #ifdef INET6
3415 	case AF_INET6:
3416 		sai6 = (struct sockaddr_in6 *)sa;
3417 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3418 		break;
3419 #endif
3420 	default:
3421 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3422 			error = EAFNOSUPPORT;
3423 	}
3424 	return (error);
3425 }
3426 
3427 /*
3428  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3429  */
3430 int
3431 prison_check(struct ucred *cred1, struct ucred *cred2)
3432 {
3433 
3434 	return ((cred1->cr_prison == cred2->cr_prison ||
3435 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3436 }
3437 
3438 /*
3439  * Return 1 if p2 is a child of p1, otherwise 0.
3440  */
3441 int
3442 prison_ischild(struct prison *pr1, struct prison *pr2)
3443 {
3444 
3445 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3446 		if (pr1 == pr2)
3447 			return (1);
3448 	return (0);
3449 }
3450 
3451 /*
3452  * Return 1 if the passed credential is in a jail, otherwise 0.
3453  */
3454 int
3455 jailed(struct ucred *cred)
3456 {
3457 
3458 	return (cred->cr_prison != &prison0);
3459 }
3460 
3461 /*
3462  * Return 1 if the passed credential is in a jail and that jail does not
3463  * have its own virtual network stack, otherwise 0.
3464  */
3465 int
3466 jailed_without_vnet(struct ucred *cred)
3467 {
3468 
3469 	if (!jailed(cred))
3470 		return (0);
3471 #ifdef VIMAGE
3472 	if (prison_owns_vnet(cred))
3473 		return (0);
3474 #endif
3475 
3476 	return (1);
3477 }
3478 
3479 /*
3480  * Return the correct hostname (domainname, et al) for the passed credential.
3481  */
3482 void
3483 getcredhostname(struct ucred *cred, char *buf, size_t size)
3484 {
3485 	struct prison *pr;
3486 
3487 	/*
3488 	 * A NULL credential can be used to shortcut to the physical
3489 	 * system's hostname.
3490 	 */
3491 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3492 	mtx_lock(&pr->pr_mtx);
3493 	strlcpy(buf, pr->pr_hostname, size);
3494 	mtx_unlock(&pr->pr_mtx);
3495 }
3496 
3497 void
3498 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3499 {
3500 
3501 	mtx_lock(&cred->cr_prison->pr_mtx);
3502 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3503 	mtx_unlock(&cred->cr_prison->pr_mtx);
3504 }
3505 
3506 void
3507 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3508 {
3509 
3510 	mtx_lock(&cred->cr_prison->pr_mtx);
3511 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3512 	mtx_unlock(&cred->cr_prison->pr_mtx);
3513 }
3514 
3515 void
3516 getcredhostid(struct ucred *cred, unsigned long *hostid)
3517 {
3518 
3519 	mtx_lock(&cred->cr_prison->pr_mtx);
3520 	*hostid = cred->cr_prison->pr_hostid;
3521 	mtx_unlock(&cred->cr_prison->pr_mtx);
3522 }
3523 
3524 #ifdef VIMAGE
3525 /*
3526  * Determine whether the prison represented by cred owns
3527  * its vnet rather than having it inherited.
3528  *
3529  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3530  */
3531 int
3532 prison_owns_vnet(struct ucred *cred)
3533 {
3534 
3535 	/*
3536 	 * vnets cannot be added/removed after jail creation,
3537 	 * so no need to lock here.
3538 	 */
3539 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3540 }
3541 #endif
3542 
3543 /*
3544  * Determine whether the subject represented by cred can "see"
3545  * status of a mount point.
3546  * Returns: 0 for permitted, ENOENT otherwise.
3547  * XXX: This function should be called cr_canseemount() and should be
3548  *      placed in kern_prot.c.
3549  */
3550 int
3551 prison_canseemount(struct ucred *cred, struct mount *mp)
3552 {
3553 	struct prison *pr;
3554 	struct statfs *sp;
3555 	size_t len;
3556 
3557 	pr = cred->cr_prison;
3558 	if (pr->pr_enforce_statfs == 0)
3559 		return (0);
3560 	if (pr->pr_root->v_mount == mp)
3561 		return (0);
3562 	if (pr->pr_enforce_statfs == 2)
3563 		return (ENOENT);
3564 	/*
3565 	 * If jail's chroot directory is set to "/" we should be able to see
3566 	 * all mount-points from inside a jail.
3567 	 * This is ugly check, but this is the only situation when jail's
3568 	 * directory ends with '/'.
3569 	 */
3570 	if (strcmp(pr->pr_path, "/") == 0)
3571 		return (0);
3572 	len = strlen(pr->pr_path);
3573 	sp = &mp->mnt_stat;
3574 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3575 		return (ENOENT);
3576 	/*
3577 	 * Be sure that we don't have situation where jail's root directory
3578 	 * is "/some/path" and mount point is "/some/pathpath".
3579 	 */
3580 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3581 		return (ENOENT);
3582 	return (0);
3583 }
3584 
3585 void
3586 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3587 {
3588 	char jpath[MAXPATHLEN];
3589 	struct prison *pr;
3590 	size_t len;
3591 
3592 	pr = cred->cr_prison;
3593 	if (pr->pr_enforce_statfs == 0)
3594 		return;
3595 	if (prison_canseemount(cred, mp) != 0) {
3596 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3597 		strlcpy(sp->f_mntonname, "[restricted]",
3598 		    sizeof(sp->f_mntonname));
3599 		return;
3600 	}
3601 	if (pr->pr_root->v_mount == mp) {
3602 		/*
3603 		 * Clear current buffer data, so we are sure nothing from
3604 		 * the valid path left there.
3605 		 */
3606 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3607 		*sp->f_mntonname = '/';
3608 		return;
3609 	}
3610 	/*
3611 	 * If jail's chroot directory is set to "/" we should be able to see
3612 	 * all mount-points from inside a jail.
3613 	 */
3614 	if (strcmp(pr->pr_path, "/") == 0)
3615 		return;
3616 	len = strlen(pr->pr_path);
3617 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3618 	/*
3619 	 * Clear current buffer data, so we are sure nothing from
3620 	 * the valid path left there.
3621 	 */
3622 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3623 	if (*jpath == '\0') {
3624 		/* Should never happen. */
3625 		*sp->f_mntonname = '/';
3626 	} else {
3627 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3628 	}
3629 }
3630 
3631 /*
3632  * Check with permission for a specific privilege is granted within jail.  We
3633  * have a specific list of accepted privileges; the rest are denied.
3634  */
3635 int
3636 prison_priv_check(struct ucred *cred, int priv)
3637 {
3638 
3639 	if (!jailed(cred))
3640 		return (0);
3641 
3642 #ifdef VIMAGE
3643 	/*
3644 	 * Privileges specific to prisons with a virtual network stack.
3645 	 * There might be a duplicate entry here in case the privilege
3646 	 * is only granted conditionally in the legacy jail case.
3647 	 */
3648 	switch (priv) {
3649 #ifdef notyet
3650 		/*
3651 		 * NFS-specific privileges.
3652 		 */
3653 	case PRIV_NFS_DAEMON:
3654 	case PRIV_NFS_LOCKD:
3655 #endif
3656 		/*
3657 		 * Network stack privileges.
3658 		 */
3659 	case PRIV_NET_BRIDGE:
3660 	case PRIV_NET_GRE:
3661 	case PRIV_NET_BPF:
3662 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3663 	case PRIV_NET_ROUTE:
3664 	case PRIV_NET_TAP:
3665 	case PRIV_NET_SETIFMTU:
3666 	case PRIV_NET_SETIFFLAGS:
3667 	case PRIV_NET_SETIFCAP:
3668 	case PRIV_NET_SETIFDESCR:
3669 	case PRIV_NET_SETIFNAME	:
3670 	case PRIV_NET_SETIFMETRIC:
3671 	case PRIV_NET_SETIFPHYS:
3672 	case PRIV_NET_SETIFMAC:
3673 	case PRIV_NET_ADDMULTI:
3674 	case PRIV_NET_DELMULTI:
3675 	case PRIV_NET_HWIOCTL:
3676 	case PRIV_NET_SETLLADDR:
3677 	case PRIV_NET_ADDIFGROUP:
3678 	case PRIV_NET_DELIFGROUP:
3679 	case PRIV_NET_IFCREATE:
3680 	case PRIV_NET_IFDESTROY:
3681 	case PRIV_NET_ADDIFADDR:
3682 	case PRIV_NET_DELIFADDR:
3683 	case PRIV_NET_LAGG:
3684 	case PRIV_NET_GIF:
3685 	case PRIV_NET_SETIFVNET:
3686 	case PRIV_NET_SETIFFIB:
3687 
3688 		/*
3689 		 * 802.11-related privileges.
3690 		 */
3691 	case PRIV_NET80211_GETKEY:
3692 #ifdef notyet
3693 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3694 #endif
3695 
3696 #ifdef notyet
3697 		/*
3698 		 * ATM privileges.
3699 		 */
3700 	case PRIV_NETATM_CFG:
3701 	case PRIV_NETATM_ADD:
3702 	case PRIV_NETATM_DEL:
3703 	case PRIV_NETATM_SET:
3704 
3705 		/*
3706 		 * Bluetooth privileges.
3707 		 */
3708 	case PRIV_NETBLUETOOTH_RAW:
3709 #endif
3710 
3711 		/*
3712 		 * Netgraph and netgraph module privileges.
3713 		 */
3714 	case PRIV_NETGRAPH_CONTROL:
3715 #ifdef notyet
3716 	case PRIV_NETGRAPH_TTY:
3717 #endif
3718 
3719 		/*
3720 		 * IPv4 and IPv6 privileges.
3721 		 */
3722 	case PRIV_NETINET_IPFW:
3723 	case PRIV_NETINET_DIVERT:
3724 	case PRIV_NETINET_PF:
3725 	case PRIV_NETINET_DUMMYNET:
3726 	case PRIV_NETINET_CARP:
3727 	case PRIV_NETINET_MROUTE:
3728 	case PRIV_NETINET_RAW:
3729 	case PRIV_NETINET_ADDRCTRL6:
3730 	case PRIV_NETINET_ND6:
3731 	case PRIV_NETINET_SCOPE6:
3732 	case PRIV_NETINET_ALIFETIME6:
3733 	case PRIV_NETINET_IPSEC:
3734 	case PRIV_NETINET_BINDANY:
3735 
3736 #ifdef notyet
3737 		/*
3738 		 * NCP privileges.
3739 		 */
3740 	case PRIV_NETNCP:
3741 
3742 		/*
3743 		 * SMB privileges.
3744 		 */
3745 	case PRIV_NETSMB:
3746 #endif
3747 
3748 	/*
3749 	 * No default: or deny here.
3750 	 * In case of no permit fall through to next switch().
3751 	 */
3752 		if (cred->cr_prison->pr_flags & PR_VNET)
3753 			return (0);
3754 	}
3755 #endif /* VIMAGE */
3756 
3757 	switch (priv) {
3758 
3759 		/*
3760 		 * Allow ktrace privileges for root in jail.
3761 		 */
3762 	case PRIV_KTRACE:
3763 
3764 #if 0
3765 		/*
3766 		 * Allow jailed processes to configure audit identity and
3767 		 * submit audit records (login, etc).  In the future we may
3768 		 * want to further refine the relationship between audit and
3769 		 * jail.
3770 		 */
3771 	case PRIV_AUDIT_GETAUDIT:
3772 	case PRIV_AUDIT_SETAUDIT:
3773 	case PRIV_AUDIT_SUBMIT:
3774 #endif
3775 
3776 		/*
3777 		 * Allow jailed processes to manipulate process UNIX
3778 		 * credentials in any way they see fit.
3779 		 */
3780 	case PRIV_CRED_SETUID:
3781 	case PRIV_CRED_SETEUID:
3782 	case PRIV_CRED_SETGID:
3783 	case PRIV_CRED_SETEGID:
3784 	case PRIV_CRED_SETGROUPS:
3785 	case PRIV_CRED_SETREUID:
3786 	case PRIV_CRED_SETREGID:
3787 	case PRIV_CRED_SETRESUID:
3788 	case PRIV_CRED_SETRESGID:
3789 
3790 		/*
3791 		 * Jail implements visibility constraints already, so allow
3792 		 * jailed root to override uid/gid-based constraints.
3793 		 */
3794 	case PRIV_SEEOTHERGIDS:
3795 	case PRIV_SEEOTHERUIDS:
3796 
3797 		/*
3798 		 * Jail implements inter-process debugging limits already, so
3799 		 * allow jailed root various debugging privileges.
3800 		 */
3801 	case PRIV_DEBUG_DIFFCRED:
3802 	case PRIV_DEBUG_SUGID:
3803 	case PRIV_DEBUG_UNPRIV:
3804 
3805 		/*
3806 		 * Allow jail to set various resource limits and login
3807 		 * properties, and for now, exceed process resource limits.
3808 		 */
3809 	case PRIV_PROC_LIMIT:
3810 	case PRIV_PROC_SETLOGIN:
3811 	case PRIV_PROC_SETRLIMIT:
3812 
3813 		/*
3814 		 * System V and POSIX IPC privileges are granted in jail.
3815 		 */
3816 	case PRIV_IPC_READ:
3817 	case PRIV_IPC_WRITE:
3818 	case PRIV_IPC_ADMIN:
3819 	case PRIV_IPC_MSGSIZE:
3820 	case PRIV_MQ_ADMIN:
3821 
3822 		/*
3823 		 * Jail operations within a jail work on child jails.
3824 		 */
3825 	case PRIV_JAIL_ATTACH:
3826 	case PRIV_JAIL_SET:
3827 	case PRIV_JAIL_REMOVE:
3828 
3829 		/*
3830 		 * Jail implements its own inter-process limits, so allow
3831 		 * root processes in jail to change scheduling on other
3832 		 * processes in the same jail.  Likewise for signalling.
3833 		 */
3834 	case PRIV_SCHED_DIFFCRED:
3835 	case PRIV_SCHED_CPUSET:
3836 	case PRIV_SIGNAL_DIFFCRED:
3837 	case PRIV_SIGNAL_SUGID:
3838 
3839 		/*
3840 		 * Allow jailed processes to write to sysctls marked as jail
3841 		 * writable.
3842 		 */
3843 	case PRIV_SYSCTL_WRITEJAIL:
3844 
3845 		/*
3846 		 * Allow root in jail to manage a variety of quota
3847 		 * properties.  These should likely be conditional on a
3848 		 * configuration option.
3849 		 */
3850 	case PRIV_VFS_GETQUOTA:
3851 	case PRIV_VFS_SETQUOTA:
3852 
3853 		/*
3854 		 * Since Jail relies on chroot() to implement file system
3855 		 * protections, grant many VFS privileges to root in jail.
3856 		 * Be careful to exclude mount-related and NFS-related
3857 		 * privileges.
3858 		 */
3859 	case PRIV_VFS_READ:
3860 	case PRIV_VFS_WRITE:
3861 	case PRIV_VFS_ADMIN:
3862 	case PRIV_VFS_EXEC:
3863 	case PRIV_VFS_LOOKUP:
3864 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3865 	case PRIV_VFS_CHFLAGS_DEV:
3866 	case PRIV_VFS_CHOWN:
3867 	case PRIV_VFS_CHROOT:
3868 	case PRIV_VFS_RETAINSUGID:
3869 	case PRIV_VFS_FCHROOT:
3870 	case PRIV_VFS_LINK:
3871 	case PRIV_VFS_SETGID:
3872 	case PRIV_VFS_STAT:
3873 	case PRIV_VFS_STICKYFILE:
3874 
3875 		/*
3876 		 * As in the non-jail case, non-root users are expected to be
3877 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3878 		 * exists in the jail and they have permission to access it).
3879 		 */
3880 	case PRIV_KMEM_READ:
3881 		return (0);
3882 
3883 		/*
3884 		 * Depending on the global setting, allow privilege of
3885 		 * setting system flags.
3886 		 */
3887 	case PRIV_VFS_SYSFLAGS:
3888 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3889 			return (0);
3890 		else
3891 			return (EPERM);
3892 
3893 		/*
3894 		 * Depending on the global setting, allow privilege of
3895 		 * mounting/unmounting file systems.
3896 		 */
3897 	case PRIV_VFS_MOUNT:
3898 	case PRIV_VFS_UNMOUNT:
3899 	case PRIV_VFS_MOUNT_NONUSER:
3900 	case PRIV_VFS_MOUNT_OWNER:
3901 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3902 		    cred->cr_prison->pr_enforce_statfs < 2)
3903 			return (0);
3904 		else
3905 			return (EPERM);
3906 
3907 		/*
3908 		 * Allow jailed root to bind reserved ports and reuse in-use
3909 		 * ports.
3910 		 */
3911 	case PRIV_NETINET_RESERVEDPORT:
3912 	case PRIV_NETINET_REUSEPORT:
3913 		return (0);
3914 
3915 		/*
3916 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3917 		 */
3918 	case PRIV_NETINET_SETHDROPTS:
3919 		return (0);
3920 
3921 		/*
3922 		 * Conditionally allow creating raw sockets in jail.
3923 		 */
3924 	case PRIV_NETINET_RAW:
3925 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3926 			return (0);
3927 		else
3928 			return (EPERM);
3929 
3930 		/*
3931 		 * Since jail implements its own visibility limits on netstat
3932 		 * sysctls, allow getcred.  This allows identd to work in
3933 		 * jail.
3934 		 */
3935 	case PRIV_NETINET_GETCRED:
3936 		return (0);
3937 
3938 		/*
3939 		 * Allow jailed root to set loginclass.
3940 		 */
3941 	case PRIV_PROC_SETLOGINCLASS:
3942 		return (0);
3943 
3944 	default:
3945 		/*
3946 		 * In all remaining cases, deny the privilege request.  This
3947 		 * includes almost all network privileges, many system
3948 		 * configuration privileges.
3949 		 */
3950 		return (EPERM);
3951 	}
3952 }
3953 
3954 /*
3955  * Return the part of pr2's name that is relative to pr1, or the whole name
3956  * if it does not directly follow.
3957  */
3958 
3959 char *
3960 prison_name(struct prison *pr1, struct prison *pr2)
3961 {
3962 	char *name;
3963 
3964 	/* Jails see themselves as "0" (if they see themselves at all). */
3965 	if (pr1 == pr2)
3966 		return "0";
3967 	name = pr2->pr_name;
3968 	if (prison_ischild(pr1, pr2)) {
3969 		/*
3970 		 * pr1 isn't locked (and allprison_lock may not be either)
3971 		 * so its length can't be counted on.  But the number of dots
3972 		 * can be counted on - and counted.
3973 		 */
3974 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3975 			name = strchr(name, '.') + 1;
3976 	}
3977 	return (name);
3978 }
3979 
3980 /*
3981  * Return the part of pr2's path that is relative to pr1, or the whole path
3982  * if it does not directly follow.
3983  */
3984 static char *
3985 prison_path(struct prison *pr1, struct prison *pr2)
3986 {
3987 	char *path1, *path2;
3988 	int len1;
3989 
3990 	path1 = pr1->pr_path;
3991 	path2 = pr2->pr_path;
3992 	if (!strcmp(path1, "/"))
3993 		return (path2);
3994 	len1 = strlen(path1);
3995 	if (strncmp(path1, path2, len1))
3996 		return (path2);
3997 	if (path2[len1] == '\0')
3998 		return "/";
3999 	if (path2[len1] == '/')
4000 		return (path2 + len1);
4001 	return (path2);
4002 }
4003 
4004 
4005 /*
4006  * Jail-related sysctls.
4007  */
4008 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4009     "Jails");
4010 
4011 static int
4012 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4013 {
4014 	struct xprison *xp;
4015 	struct prison *pr, *cpr;
4016 #ifdef INET
4017 	struct in_addr *ip4 = NULL;
4018 	int ip4s = 0;
4019 #endif
4020 #ifdef INET6
4021 	struct in6_addr *ip6 = NULL;
4022 	int ip6s = 0;
4023 #endif
4024 	int descend, error;
4025 
4026 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4027 	pr = req->td->td_ucred->cr_prison;
4028 	error = 0;
4029 	sx_slock(&allprison_lock);
4030 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4031 #if defined(INET) || defined(INET6)
4032  again:
4033 #endif
4034 		mtx_lock(&cpr->pr_mtx);
4035 #ifdef INET
4036 		if (cpr->pr_ip4s > 0) {
4037 			if (ip4s < cpr->pr_ip4s) {
4038 				ip4s = cpr->pr_ip4s;
4039 				mtx_unlock(&cpr->pr_mtx);
4040 				ip4 = realloc(ip4, ip4s *
4041 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4042 				goto again;
4043 			}
4044 			bcopy(cpr->pr_ip4, ip4,
4045 			    cpr->pr_ip4s * sizeof(struct in_addr));
4046 		}
4047 #endif
4048 #ifdef INET6
4049 		if (cpr->pr_ip6s > 0) {
4050 			if (ip6s < cpr->pr_ip6s) {
4051 				ip6s = cpr->pr_ip6s;
4052 				mtx_unlock(&cpr->pr_mtx);
4053 				ip6 = realloc(ip6, ip6s *
4054 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4055 				goto again;
4056 			}
4057 			bcopy(cpr->pr_ip6, ip6,
4058 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4059 		}
4060 #endif
4061 		if (cpr->pr_ref == 0) {
4062 			mtx_unlock(&cpr->pr_mtx);
4063 			continue;
4064 		}
4065 		bzero(xp, sizeof(*xp));
4066 		xp->pr_version = XPRISON_VERSION;
4067 		xp->pr_id = cpr->pr_id;
4068 		xp->pr_state = cpr->pr_uref > 0
4069 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4070 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4071 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4072 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4073 #ifdef INET
4074 		xp->pr_ip4s = cpr->pr_ip4s;
4075 #endif
4076 #ifdef INET6
4077 		xp->pr_ip6s = cpr->pr_ip6s;
4078 #endif
4079 		mtx_unlock(&cpr->pr_mtx);
4080 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4081 		if (error)
4082 			break;
4083 #ifdef INET
4084 		if (xp->pr_ip4s > 0) {
4085 			error = SYSCTL_OUT(req, ip4,
4086 			    xp->pr_ip4s * sizeof(struct in_addr));
4087 			if (error)
4088 				break;
4089 		}
4090 #endif
4091 #ifdef INET6
4092 		if (xp->pr_ip6s > 0) {
4093 			error = SYSCTL_OUT(req, ip6,
4094 			    xp->pr_ip6s * sizeof(struct in6_addr));
4095 			if (error)
4096 				break;
4097 		}
4098 #endif
4099 	}
4100 	sx_sunlock(&allprison_lock);
4101 	free(xp, M_TEMP);
4102 #ifdef INET
4103 	free(ip4, M_TEMP);
4104 #endif
4105 #ifdef INET6
4106 	free(ip6, M_TEMP);
4107 #endif
4108 	return (error);
4109 }
4110 
4111 SYSCTL_OID(_security_jail, OID_AUTO, list,
4112     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4113     sysctl_jail_list, "S", "List of active jails");
4114 
4115 static int
4116 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4117 {
4118 	int error, injail;
4119 
4120 	injail = jailed(req->td->td_ucred);
4121 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4122 
4123 	return (error);
4124 }
4125 
4126 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4127     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4128     sysctl_jail_jailed, "I", "Process in jail?");
4129 
4130 static int
4131 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4132 {
4133 	int error, havevnet;
4134 #ifdef VIMAGE
4135 	struct ucred *cred = req->td->td_ucred;
4136 
4137 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4138 #else
4139 	havevnet = 0;
4140 #endif
4141 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4142 
4143 	return (error);
4144 }
4145 
4146 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4147     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4148     sysctl_jail_vnet, "I", "Jail owns VNET?");
4149 
4150 #if defined(INET) || defined(INET6)
4151 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4152     &jail_max_af_ips, 0,
4153     "Number of IP addresses a jail may have at most per address family");
4154 #endif
4155 
4156 /*
4157  * Default parameters for jail(2) compatability.  For historical reasons,
4158  * the sysctl names have varying similarity to the parameter names.  Prisons
4159  * just see their own parameters, and can't change them.
4160  */
4161 static int
4162 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4163 {
4164 	struct prison *pr;
4165 	int allow, error, i;
4166 
4167 	pr = req->td->td_ucred->cr_prison;
4168 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4169 
4170 	/* Get the current flag value, and convert it to a boolean. */
4171 	i = (allow & arg2) ? 1 : 0;
4172 	if (arg1 != NULL)
4173 		i = !i;
4174 	error = sysctl_handle_int(oidp, &i, 0, req);
4175 	if (error || !req->newptr)
4176 		return (error);
4177 	i = i ? arg2 : 0;
4178 	if (arg1 != NULL)
4179 		i ^= arg2;
4180 	/*
4181 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4182 	 * for writing.
4183 	 */
4184 	mtx_lock(&prison0.pr_mtx);
4185 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4186 	mtx_unlock(&prison0.pr_mtx);
4187 	return (0);
4188 }
4189 
4190 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4191     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4192     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4193     "Processes in jail can set their hostnames");
4194 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4195     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4196     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4197     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4198 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4199     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4200     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4201     "Processes in jail can use System V IPC primitives");
4202 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4203     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4204     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4205     "Prison root can create raw sockets");
4206 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4207     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4208     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4209     "Processes in jail can alter system file flags");
4210 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4211     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4212     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4213     "Processes in jail can mount/unmount jail-friendly file systems");
4214 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4215     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4216     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4217     "Processes in jail can mount the devfs file system");
4218 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4219     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4220     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4221     "Processes in jail can mount the fdescfs file system");
4222 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4223     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4224     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4225     "Processes in jail can mount the nullfs file system");
4226 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4227     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4228     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4229     "Processes in jail can mount the procfs file system");
4230 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4231     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4232     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4233     "Processes in jail can mount the tmpfs file system");
4234 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4235     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4236     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4237     "Processes in jail can mount the zfs file system");
4238 
4239 static int
4240 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4241 {
4242 	struct prison *pr;
4243 	int level, error;
4244 
4245 	pr = req->td->td_ucred->cr_prison;
4246 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4247 	error = sysctl_handle_int(oidp, &level, 0, req);
4248 	if (error || !req->newptr)
4249 		return (error);
4250 	*(int *)arg1 = level;
4251 	return (0);
4252 }
4253 
4254 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4255     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4256     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4257     sysctl_jail_default_level, "I",
4258     "Processes in jail cannot see all mounted file systems");
4259 
4260 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4261     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4262     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4263     sysctl_jail_default_level, "I",
4264     "Ruleset for the devfs filesystem in jail");
4265 
4266 /*
4267  * Nodes to describe jail parameters.  Maximum length of string parameters
4268  * is returned in the string itself, and the other parameters exist merely
4269  * to make themselves and their types known.
4270  */
4271 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4272     "Jail parameters");
4273 
4274 int
4275 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4276 {
4277 	int i;
4278 	long l;
4279 	size_t s;
4280 	char numbuf[12];
4281 
4282 	switch (oidp->oid_kind & CTLTYPE)
4283 	{
4284 	case CTLTYPE_LONG:
4285 	case CTLTYPE_ULONG:
4286 		l = 0;
4287 #ifdef SCTL_MASK32
4288 		if (!(req->flags & SCTL_MASK32))
4289 #endif
4290 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4291 	case CTLTYPE_INT:
4292 	case CTLTYPE_UINT:
4293 		i = 0;
4294 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4295 	case CTLTYPE_STRING:
4296 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4297 		return
4298 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4299 	case CTLTYPE_STRUCT:
4300 		s = (size_t)arg2;
4301 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4302 	}
4303 	return (0);
4304 }
4305 
4306 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4307 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4308 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4309 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4310 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4311     "I", "Jail secure level");
4312 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4313     "I", "Jail cannot see all mounted file systems");
4314 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4315     "I", "Ruleset for in-jail devfs mounts");
4316 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4317     "B", "Jail persistence");
4318 #ifdef VIMAGE
4319 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4320     "E,jailsys", "Virtual network stack");
4321 #endif
4322 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4323     "B", "Jail is in the process of shutting down");
4324 
4325 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4326 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4327     "I", "Current number of child jails");
4328 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4329     "I", "Maximum number of child jails");
4330 
4331 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4332 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4333     "Jail hostname");
4334 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4335     "Jail NIS domainname");
4336 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4337     "Jail host UUID");
4338 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4339     "LU", "Jail host ID");
4340 
4341 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4342 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4343 
4344 #ifdef INET
4345 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4346     "Jail IPv4 address virtualization");
4347 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4348     "S,in_addr,a", "Jail IPv4 addresses");
4349 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4350     "B", "Do (not) use IPv4 source address selection rather than the "
4351     "primary jail IPv4 address.");
4352 #endif
4353 #ifdef INET6
4354 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4355     "Jail IPv6 address virtualization");
4356 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4357     "S,in6_addr,a", "Jail IPv6 addresses");
4358 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4359     "B", "Do (not) use IPv6 source address selection rather than the "
4360     "primary jail IPv6 address.");
4361 #endif
4362 
4363 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4364 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4365     "B", "Jail may set hostname");
4366 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4367     "B", "Jail may use SYSV IPC");
4368 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4369     "B", "Jail may create raw sockets");
4370 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4371     "B", "Jail may alter system file flags");
4372 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4373     "B", "Jail may set file quotas");
4374 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4375     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4376 
4377 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4378 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4379     "B", "Jail may mount/unmount jail-friendly file systems in general");
4380 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4381     "B", "Jail may mount the devfs file system");
4382 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4383     "B", "Jail may mount the fdescfs 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