xref: /freebsd/sys/kern/kern_jail.c (revision f7167e0ea0bf5aaabff9490453b3b71b3f1b4d51)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 
70 #ifdef DDB
71 #include <ddb/ddb.h>
72 #ifdef INET6
73 #include <netinet6/in6_var.h>
74 #endif /* INET6 */
75 #endif /* DDB */
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
80 
81 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
82 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
83 
84 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
85 #ifdef INET
86 #ifdef INET6
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
88 #else
89 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
90 #endif
91 #else /* !INET */
92 #ifdef INET6
93 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
94 #else
95 #define	_PR_IP_SADDRSEL	0
96 #endif
97 #endif
98 
99 /* prison0 describes what is "real" about the system. */
100 struct prison prison0 = {
101 	.pr_id		= 0,
102 	.pr_name	= "0",
103 	.pr_ref		= 1,
104 	.pr_uref	= 1,
105 	.pr_path	= "/",
106 	.pr_securelevel	= -1,
107 	.pr_devfs_rsnum = 0,
108 	.pr_childmax	= JAIL_MAX,
109 	.pr_hostuuid	= DEFAULT_HOSTUUID,
110 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
111 #ifdef VIMAGE
112 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
113 #else
114 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
115 #endif
116 	.pr_allow	= PR_ALLOW_ALL,
117 };
118 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
119 
120 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
121 struct	sx allprison_lock;
122 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
123 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
124 LIST_HEAD(, prison_racct) allprison_racct;
125 int	lastprid = 0;
126 
127 static int do_jail_attach(struct thread *td, struct prison *pr);
128 static void prison_complete(void *context, int pending);
129 static void prison_deref(struct prison *pr, int flags);
130 static char *prison_path(struct prison *pr1, struct prison *pr2);
131 static void prison_remove_one(struct prison *pr);
132 #ifdef RACCT
133 static void prison_racct_attach(struct prison *pr);
134 static void prison_racct_modify(struct prison *pr);
135 static void prison_racct_detach(struct prison *pr);
136 #endif
137 #ifdef INET
138 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
139 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
140 #endif
141 #ifdef INET6
142 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
143 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
144 #endif
145 
146 /* Flags for prison_deref */
147 #define	PD_DEREF	0x01
148 #define	PD_DEUREF	0x02
149 #define	PD_LOCKED	0x04
150 #define	PD_LIST_SLOCKED	0x08
151 #define	PD_LIST_XLOCKED	0x10
152 
153 /*
154  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
155  * as we cannot figure out the size of a sparse array, or an array without a
156  * terminating entry.
157  */
158 static char *pr_flag_names[] = {
159 	[0] = "persist",
160 #ifdef INET
161 	[7] = "ip4.saddrsel",
162 #endif
163 #ifdef INET6
164 	[8] = "ip6.saddrsel",
165 #endif
166 };
167 const size_t pr_flag_names_size = sizeof(pr_flag_names);
168 
169 static char *pr_flag_nonames[] = {
170 	[0] = "nopersist",
171 #ifdef INET
172 	[7] = "ip4.nosaddrsel",
173 #endif
174 #ifdef INET6
175 	[8] = "ip6.nosaddrsel",
176 #endif
177 };
178 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
179 
180 struct jailsys_flags {
181 	const char	*name;
182 	unsigned	 disable;
183 	unsigned	 new;
184 } pr_flag_jailsys[] = {
185 	{ "host", 0, PR_HOST },
186 #ifdef VIMAGE
187 	{ "vnet", 0, PR_VNET },
188 #endif
189 #ifdef INET
190 	{ "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
191 #endif
192 #ifdef INET6
193 	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
194 #endif
195 };
196 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
197 
198 static char *pr_allow_names[] = {
199 	"allow.set_hostname",
200 	"allow.sysvipc",
201 	"allow.raw_sockets",
202 	"allow.chflags",
203 	"allow.mount",
204 	"allow.quotas",
205 	"allow.socket_af",
206 	"allow.mount.devfs",
207 	"allow.mount.nullfs",
208 	"allow.mount.zfs",
209 	"allow.mount.procfs",
210 	"allow.mount.tmpfs",
211 	"allow.kmem",
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.nokmem",
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 = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
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 | PR_IP4_DISABLE;
813 		if (ip4s == 0)
814 			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
815 		else {
816 			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
817 			ip4s /= sizeof(*ip4);
818 			if (ip4s > jail_max_af_ips) {
819 				error = EINVAL;
820 				vfs_opterror(opts, "too many IPv4 addresses");
821 				goto done_errmsg;
822 			}
823 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
824 			bcopy(op, ip4, ip4s * sizeof(*ip4));
825 			/*
826 			 * IP addresses are all sorted but ip[0] to preserve
827 			 * the primary IP address as given from userland.
828 			 * This special IP is used for unbound outgoing
829 			 * connections as well for "loopback" traffic in case
830 			 * source address selection cannot find any more fitting
831 			 * address to connect from.
832 			 */
833 			if (ip4s > 1)
834 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
835 			/*
836 			 * Check for duplicate addresses and do some simple
837 			 * zero and broadcast checks. If users give other bogus
838 			 * addresses it is their problem.
839 			 *
840 			 * We do not have to care about byte order for these
841 			 * checks so we will do them in NBO.
842 			 */
843 			for (ii = 0; ii < ip4s; ii++) {
844 				if (ip4[ii].s_addr == INADDR_ANY ||
845 				    ip4[ii].s_addr == INADDR_BROADCAST) {
846 					error = EINVAL;
847 					goto done_free;
848 				}
849 				if ((ii+1) < ip4s &&
850 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
851 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
852 					error = EINVAL;
853 					goto done_free;
854 				}
855 			}
856 		}
857 	}
858 #endif
859 
860 #ifdef INET6
861 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
862 	if (error == ENOENT)
863 		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
864 	else if (error != 0)
865 		goto done_free;
866 	else if (ip6s & (sizeof(*ip6) - 1)) {
867 		error = EINVAL;
868 		goto done_free;
869 	} else {
870 		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
871 		if (ip6s == 0)
872 			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
873 		else {
874 			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
875 			ip6s /= sizeof(*ip6);
876 			if (ip6s > jail_max_af_ips) {
877 				error = EINVAL;
878 				vfs_opterror(opts, "too many IPv6 addresses");
879 				goto done_errmsg;
880 			}
881 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
882 			bcopy(op, ip6, ip6s * sizeof(*ip6));
883 			if (ip6s > 1)
884 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
885 			for (ii = 0; ii < ip6s; ii++) {
886 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
887 					error = EINVAL;
888 					goto done_free;
889 				}
890 				if ((ii+1) < ip6s &&
891 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
892 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
893 				{
894 					error = EINVAL;
895 					goto done_free;
896 				}
897 			}
898 		}
899 	}
900 #endif
901 
902 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
903 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
904 		error = EINVAL;
905 		vfs_opterror(opts,
906 		    "vnet jails cannot have IP address restrictions");
907 		goto done_errmsg;
908 	}
909 #endif
910 
911 	fullpath_disabled = 0;
912 	root = NULL;
913 	error = vfs_getopt(opts, "path", (void **)&path, &len);
914 	if (error == ENOENT)
915 		path = NULL;
916 	else if (error != 0)
917 		goto done_free;
918 	else {
919 		if (flags & JAIL_UPDATE) {
920 			error = EINVAL;
921 			vfs_opterror(opts,
922 			    "path cannot be changed after creation");
923 			goto done_errmsg;
924 		}
925 		if (len == 0 || path[len - 1] != '\0') {
926 			error = EINVAL;
927 			goto done_free;
928 		}
929 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
930 		    path, td);
931 		error = namei(&nd);
932 		if (error)
933 			goto done_free;
934 		root = nd.ni_vp;
935 		NDFREE(&nd, NDF_ONLY_PNBUF);
936 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
937 		strlcpy(g_path, path, MAXPATHLEN);
938 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
939 		if (error == 0)
940 			path = g_path;
941 		else if (error == ENODEV) {
942 			/* proceed if sysctl debug.disablefullpath == 1 */
943 			fullpath_disabled = 1;
944 			if (len < 2 || (len == 2 && path[0] == '/'))
945 				path = NULL;
946 		} else {
947 			/* exit on other errors */
948 			goto done_free;
949 		}
950 		if (root->v_type != VDIR) {
951 			error = ENOTDIR;
952 			vput(root);
953 			goto done_free;
954 		}
955 		VOP_UNLOCK(root, 0);
956 		if (fullpath_disabled) {
957 			/* Leave room for a real-root full pathname. */
958 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
959 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
960 				error = ENAMETOOLONG;
961 				goto done_free;
962 			}
963 		}
964 	}
965 
966 	/*
967 	 * Grab the allprison lock before letting modules check their
968 	 * parameters.  Once we have it, do not let go so we'll have a
969 	 * consistent view of the OSD list.
970 	 */
971 	sx_xlock(&allprison_lock);
972 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
973 	if (error)
974 		goto done_unlock_list;
975 
976 	/* By now, all parameters should have been noted. */
977 	TAILQ_FOREACH(opt, opts, link) {
978 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
979 			error = EINVAL;
980 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
981 			goto done_unlock_list;
982 		}
983 	}
984 
985 	/*
986 	 * See if we are creating a new record or updating an existing one.
987 	 * This abuses the file error codes ENOENT and EEXIST.
988 	 */
989 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
990 	if (!cuflags) {
991 		error = EINVAL;
992 		vfs_opterror(opts, "no valid operation (create or update)");
993 		goto done_unlock_list;
994 	}
995 	pr = NULL;
996 	namelc = NULL;
997 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
998 		namelc = strrchr(name, '.');
999 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1000 		if (*p != '\0')
1001 			jid = 0;
1002 	}
1003 	if (jid != 0) {
1004 		/*
1005 		 * See if a requested jid already exists.  There is an
1006 		 * information leak here if the jid exists but is not within
1007 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1008 		 * even though they cannot see the jail, and CREATE | UPDATE
1009 		 * will return ENOENT which is not normally a valid error.
1010 		 */
1011 		if (jid < 0) {
1012 			error = EINVAL;
1013 			vfs_opterror(opts, "negative jid");
1014 			goto done_unlock_list;
1015 		}
1016 		pr = prison_find(jid);
1017 		if (pr != NULL) {
1018 			ppr = pr->pr_parent;
1019 			/* Create: jid must not exist. */
1020 			if (cuflags == JAIL_CREATE) {
1021 				mtx_unlock(&pr->pr_mtx);
1022 				error = EEXIST;
1023 				vfs_opterror(opts, "jail %d already exists",
1024 				    jid);
1025 				goto done_unlock_list;
1026 			}
1027 			if (!prison_ischild(mypr, pr)) {
1028 				mtx_unlock(&pr->pr_mtx);
1029 				pr = NULL;
1030 			} else if (pr->pr_uref == 0) {
1031 				if (!(flags & JAIL_DYING)) {
1032 					mtx_unlock(&pr->pr_mtx);
1033 					error = ENOENT;
1034 					vfs_opterror(opts, "jail %d is dying",
1035 					    jid);
1036 					goto done_unlock_list;
1037 				} else if ((flags & JAIL_ATTACH) ||
1038 				    (pr_flags & PR_PERSIST)) {
1039 					/*
1040 					 * A dying jail might be resurrected
1041 					 * (via attach or persist), but first
1042 					 * it must determine if another jail
1043 					 * has claimed its name.  Accomplish
1044 					 * this by implicitly re-setting the
1045 					 * name.
1046 					 */
1047 					if (name == NULL)
1048 						name = prison_name(mypr, pr);
1049 				}
1050 			}
1051 		}
1052 		if (pr == NULL) {
1053 			/* Update: jid must exist. */
1054 			if (cuflags == JAIL_UPDATE) {
1055 				error = ENOENT;
1056 				vfs_opterror(opts, "jail %d not found", jid);
1057 				goto done_unlock_list;
1058 			}
1059 		}
1060 	}
1061 	/*
1062 	 * If the caller provided a name, look for a jail by that name.
1063 	 * This has different semantics for creates and updates keyed by jid
1064 	 * (where the name must not already exist in a different jail),
1065 	 * and updates keyed by the name itself (where the name must exist
1066 	 * because that is the jail being updated).
1067 	 */
1068 	if (name != NULL) {
1069 		namelc = strrchr(name, '.');
1070 		if (namelc == NULL)
1071 			namelc = name;
1072 		else {
1073 			/*
1074 			 * This is a hierarchical name.  Split it into the
1075 			 * parent and child names, and make sure the parent
1076 			 * exists or matches an already found jail.
1077 			 */
1078 			*namelc = '\0';
1079 			if (pr != NULL) {
1080 				if (strncmp(name, ppr->pr_name, namelc - name)
1081 				    || ppr->pr_name[namelc - name] != '\0') {
1082 					mtx_unlock(&pr->pr_mtx);
1083 					error = EINVAL;
1084 					vfs_opterror(opts,
1085 					    "cannot change jail's parent");
1086 					goto done_unlock_list;
1087 				}
1088 			} else {
1089 				ppr = prison_find_name(mypr, name);
1090 				if (ppr == NULL) {
1091 					error = ENOENT;
1092 					vfs_opterror(opts,
1093 					    "jail \"%s\" not found", name);
1094 					goto done_unlock_list;
1095 				}
1096 				mtx_unlock(&ppr->pr_mtx);
1097 			}
1098 			name = ++namelc;
1099 		}
1100 		if (name[0] != '\0') {
1101 			namelen =
1102 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1103  name_again:
1104 			deadpr = NULL;
1105 			FOREACH_PRISON_CHILD(ppr, tpr) {
1106 				if (tpr != pr && tpr->pr_ref > 0 &&
1107 				    !strcmp(tpr->pr_name + namelen, name)) {
1108 					if (pr == NULL &&
1109 					    cuflags != JAIL_CREATE) {
1110 						mtx_lock(&tpr->pr_mtx);
1111 						if (tpr->pr_ref > 0) {
1112 							/*
1113 							 * Use this jail
1114 							 * for updates.
1115 							 */
1116 							if (tpr->pr_uref > 0) {
1117 								pr = tpr;
1118 								break;
1119 							}
1120 							deadpr = tpr;
1121 						}
1122 						mtx_unlock(&tpr->pr_mtx);
1123 					} else if (tpr->pr_uref > 0) {
1124 						/*
1125 						 * Create, or update(jid):
1126 						 * name must not exist in an
1127 						 * active sibling jail.
1128 						 */
1129 						error = EEXIST;
1130 						if (pr != NULL)
1131 							mtx_unlock(&pr->pr_mtx);
1132 						vfs_opterror(opts,
1133 						   "jail \"%s\" already exists",
1134 						   name);
1135 						goto done_unlock_list;
1136 					}
1137 				}
1138 			}
1139 			/* If no active jail is found, use a dying one. */
1140 			if (deadpr != NULL && pr == NULL) {
1141 				if (flags & JAIL_DYING) {
1142 					mtx_lock(&deadpr->pr_mtx);
1143 					if (deadpr->pr_ref == 0) {
1144 						mtx_unlock(&deadpr->pr_mtx);
1145 						goto name_again;
1146 					}
1147 					pr = deadpr;
1148 				} else if (cuflags == JAIL_UPDATE) {
1149 					error = ENOENT;
1150 					vfs_opterror(opts,
1151 					    "jail \"%s\" is dying", name);
1152 					goto done_unlock_list;
1153 				}
1154 			}
1155 			/* Update: name must exist if no jid. */
1156 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1157 				error = ENOENT;
1158 				vfs_opterror(opts, "jail \"%s\" not found",
1159 				    name);
1160 				goto done_unlock_list;
1161 			}
1162 		}
1163 	}
1164 	/* Update: must provide a jid or name. */
1165 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1166 		error = ENOENT;
1167 		vfs_opterror(opts, "update specified no jail");
1168 		goto done_unlock_list;
1169 	}
1170 
1171 	/* If there's no prison to update, create a new one and link it in. */
1172 	if (pr == NULL) {
1173 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1174 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1175 				error = EPERM;
1176 				vfs_opterror(opts, "prison limit exceeded");
1177 				goto done_unlock_list;
1178 			}
1179 		created = 1;
1180 		mtx_lock(&ppr->pr_mtx);
1181 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1182 			mtx_unlock(&ppr->pr_mtx);
1183 			error = ENOENT;
1184 			vfs_opterror(opts, "parent jail went away!");
1185 			goto done_unlock_list;
1186 		}
1187 		ppr->pr_ref++;
1188 		ppr->pr_uref++;
1189 		mtx_unlock(&ppr->pr_mtx);
1190 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1191 		if (jid == 0) {
1192 			/* Find the next free jid. */
1193 			jid = lastprid + 1;
1194  findnext:
1195 			if (jid == JAIL_MAX)
1196 				jid = 1;
1197 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1198 				if (tpr->pr_id < jid)
1199 					continue;
1200 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1201 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1202 					break;
1203 				}
1204 				if (jid == lastprid) {
1205 					error = EAGAIN;
1206 					vfs_opterror(opts,
1207 					    "no available jail IDs");
1208 					free(pr, M_PRISON);
1209 					prison_deref(ppr, PD_DEREF |
1210 					    PD_DEUREF | PD_LIST_XLOCKED);
1211 					goto done_releroot;
1212 				}
1213 				jid++;
1214 				goto findnext;
1215 			}
1216 			lastprid = jid;
1217 		} else {
1218 			/*
1219 			 * The jail already has a jid (that did not yet exist),
1220 			 * so just find where to insert it.
1221 			 */
1222 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1223 				if (tpr->pr_id >= jid) {
1224 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1225 					break;
1226 				}
1227 		}
1228 		if (tpr == NULL)
1229 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1230 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1231 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1232 			tpr->pr_childcount++;
1233 
1234 		pr->pr_parent = ppr;
1235 		pr->pr_id = jid;
1236 
1237 		/* Set some default values, and inherit some from the parent. */
1238 		if (name == NULL)
1239 			name = "";
1240 		if (path == NULL) {
1241 			path = "/";
1242 			root = mypr->pr_root;
1243 			vref(root);
1244 		}
1245 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1246 		pr->pr_flags |= PR_HOST;
1247 #if defined(INET) || defined(INET6)
1248 #ifdef VIMAGE
1249 		if (!(pr_flags & PR_VNET))
1250 #endif
1251 		{
1252 #ifdef INET
1253 			if (!(ch_flags & PR_IP4_USER))
1254 				pr->pr_flags |=
1255 				    PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1256 			else if (!(pr_flags & PR_IP4_USER)) {
1257 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1258 				if (ppr->pr_ip4 != NULL) {
1259 					pr->pr_ip4s = ppr->pr_ip4s;
1260 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1261 					    sizeof(struct in_addr), M_PRISON,
1262 					    M_WAITOK);
1263 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1264 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1265 				}
1266 			}
1267 #endif
1268 #ifdef INET6
1269 			if (!(ch_flags & PR_IP6_USER))
1270 				pr->pr_flags |=
1271 				    PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1272 			else if (!(pr_flags & PR_IP6_USER)) {
1273 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1274 				if (ppr->pr_ip6 != NULL) {
1275 					pr->pr_ip6s = ppr->pr_ip6s;
1276 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1277 					    sizeof(struct in6_addr), M_PRISON,
1278 					    M_WAITOK);
1279 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1280 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1281 				}
1282 			}
1283 #endif
1284 		}
1285 #endif
1286 		/* Source address selection is always on by default. */
1287 		pr->pr_flags |= _PR_IP_SADDRSEL;
1288 
1289 		pr->pr_securelevel = ppr->pr_securelevel;
1290 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1291 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1292 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1293 
1294 		LIST_INIT(&pr->pr_children);
1295 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1296 
1297 #ifdef VIMAGE
1298 		/* Allocate a new vnet if specified. */
1299 		pr->pr_vnet = (pr_flags & PR_VNET)
1300 		    ? vnet_alloc() : ppr->pr_vnet;
1301 #endif
1302 		/*
1303 		 * Allocate a dedicated cpuset for each jail.
1304 		 * Unlike other initial settings, this may return an erorr.
1305 		 */
1306 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1307 		if (error) {
1308 			prison_deref(pr, PD_LIST_XLOCKED);
1309 			goto done_releroot;
1310 		}
1311 
1312 		mtx_lock(&pr->pr_mtx);
1313 		/*
1314 		 * New prisons do not yet have a reference, because we do not
1315 		 * want other to see the incomplete prison once the
1316 		 * allprison_lock is downgraded.
1317 		 */
1318 	} else {
1319 		created = 0;
1320 		/*
1321 		 * Grab a reference for existing prisons, to ensure they
1322 		 * continue to exist for the duration of the call.
1323 		 */
1324 		pr->pr_ref++;
1325 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1326 		if ((pr->pr_flags & PR_VNET) &&
1327 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1328 			error = EINVAL;
1329 			vfs_opterror(opts,
1330 			    "vnet jails cannot have IP address restrictions");
1331 			goto done_deref_locked;
1332 		}
1333 #endif
1334 #ifdef INET
1335 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1336 			error = EINVAL;
1337 			vfs_opterror(opts,
1338 			    "ip4 cannot be changed after creation");
1339 			goto done_deref_locked;
1340 		}
1341 #endif
1342 #ifdef INET6
1343 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1344 			error = EINVAL;
1345 			vfs_opterror(opts,
1346 			    "ip6 cannot be changed after creation");
1347 			goto done_deref_locked;
1348 		}
1349 #endif
1350 	}
1351 
1352 	/* Do final error checking before setting anything. */
1353 	if (gotslevel) {
1354 		if (slevel < ppr->pr_securelevel) {
1355 			error = EPERM;
1356 			goto done_deref_locked;
1357 		}
1358 	}
1359 	if (gotchildmax) {
1360 		if (childmax >= ppr->pr_childmax) {
1361 			error = EPERM;
1362 			goto done_deref_locked;
1363 		}
1364 	}
1365 	if (gotenforce) {
1366 		if (enforce < ppr->pr_enforce_statfs) {
1367 			error = EPERM;
1368 			goto done_deref_locked;
1369 		}
1370 	}
1371 	if (gotrsnum) {
1372 		/*
1373 		 * devfs_rsnum is a uint16_t
1374 		 */
1375 		if (rsnum < 0 || rsnum > 65535) {
1376 			error = EINVAL;
1377 			goto done_deref_locked;
1378 		}
1379 		/*
1380 		 * Nested jails always inherit parent's devfs ruleset
1381 		 */
1382 		if (jailed(td->td_ucred)) {
1383 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1384 				error = EPERM;
1385 				goto done_deref_locked;
1386 			} else
1387 				rsnum = ppr->pr_devfs_rsnum;
1388 		}
1389 	}
1390 #ifdef INET
1391 	if (ip4s > 0) {
1392 		if (ppr->pr_flags & PR_IP4) {
1393 			/*
1394 			 * Make sure the new set of IP addresses is a
1395 			 * subset of the parent's list.  Don't worry
1396 			 * about the parent being unlocked, as any
1397 			 * setting is done with allprison_lock held.
1398 			 */
1399 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1400 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1401 					break;
1402 			if (ij == ppr->pr_ip4s) {
1403 				error = EPERM;
1404 				goto done_deref_locked;
1405 			}
1406 			if (ip4s > 1) {
1407 				for (ii = ij = 1; ii < ip4s; ii++) {
1408 					if (ip4[ii].s_addr ==
1409 					    ppr->pr_ip4[0].s_addr)
1410 						continue;
1411 					for (; ij < ppr->pr_ip4s; ij++)
1412 						if (ip4[ii].s_addr ==
1413 						    ppr->pr_ip4[ij].s_addr)
1414 							break;
1415 					if (ij == ppr->pr_ip4s)
1416 						break;
1417 				}
1418 				if (ij == ppr->pr_ip4s) {
1419 					error = EPERM;
1420 					goto done_deref_locked;
1421 				}
1422 			}
1423 		}
1424 		/*
1425 		 * Check for conflicting IP addresses.  We permit them
1426 		 * if there is no more than one IP on each jail.  If
1427 		 * there is a duplicate on a jail with more than one
1428 		 * IP stop checking and return error.
1429 		 */
1430 		tppr = ppr;
1431 #ifdef VIMAGE
1432 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1433 			if (tppr->pr_flags & PR_VNET)
1434 				break;
1435 #endif
1436 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1437 			if (tpr == pr ||
1438 #ifdef VIMAGE
1439 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1440 #endif
1441 			    tpr->pr_uref == 0) {
1442 				descend = 0;
1443 				continue;
1444 			}
1445 			if (!(tpr->pr_flags & PR_IP4_USER))
1446 				continue;
1447 			descend = 0;
1448 			if (tpr->pr_ip4 == NULL ||
1449 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1450 				continue;
1451 			for (ii = 0; ii < ip4s; ii++) {
1452 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1453 					error = EADDRINUSE;
1454 					vfs_opterror(opts,
1455 					    "IPv4 addresses clash");
1456 					goto done_deref_locked;
1457 				}
1458 			}
1459 		}
1460 	}
1461 #endif
1462 #ifdef INET6
1463 	if (ip6s > 0) {
1464 		if (ppr->pr_flags & PR_IP6) {
1465 			/*
1466 			 * Make sure the new set of IP addresses is a
1467 			 * subset of the parent's list.
1468 			 */
1469 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1470 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1471 				    &ppr->pr_ip6[ij]))
1472 					break;
1473 			if (ij == ppr->pr_ip6s) {
1474 				error = EPERM;
1475 				goto done_deref_locked;
1476 			}
1477 			if (ip6s > 1) {
1478 				for (ii = ij = 1; ii < ip6s; ii++) {
1479 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1480 					     &ppr->pr_ip6[0]))
1481 						continue;
1482 					for (; ij < ppr->pr_ip6s; ij++)
1483 						if (IN6_ARE_ADDR_EQUAL(
1484 						    &ip6[ii], &ppr->pr_ip6[ij]))
1485 							break;
1486 					if (ij == ppr->pr_ip6s)
1487 						break;
1488 				}
1489 				if (ij == ppr->pr_ip6s) {
1490 					error = EPERM;
1491 					goto done_deref_locked;
1492 				}
1493 			}
1494 		}
1495 		/* Check for conflicting IP addresses. */
1496 		tppr = ppr;
1497 #ifdef VIMAGE
1498 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1499 			if (tppr->pr_flags & PR_VNET)
1500 				break;
1501 #endif
1502 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1503 			if (tpr == pr ||
1504 #ifdef VIMAGE
1505 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1506 #endif
1507 			    tpr->pr_uref == 0) {
1508 				descend = 0;
1509 				continue;
1510 			}
1511 			if (!(tpr->pr_flags & PR_IP6_USER))
1512 				continue;
1513 			descend = 0;
1514 			if (tpr->pr_ip6 == NULL ||
1515 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1516 				continue;
1517 			for (ii = 0; ii < ip6s; ii++) {
1518 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1519 					error = EADDRINUSE;
1520 					vfs_opterror(opts,
1521 					    "IPv6 addresses clash");
1522 					goto done_deref_locked;
1523 				}
1524 			}
1525 		}
1526 	}
1527 #endif
1528 	onamelen = namelen = 0;
1529 	if (name != NULL) {
1530 		/* Give a default name of the jid. */
1531 		if (name[0] == '\0')
1532 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1533 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1534 		    *p == '\0')) {
1535 			error = EINVAL;
1536 			vfs_opterror(opts,
1537 			    "name cannot be numeric (unless it is the jid)");
1538 			goto done_deref_locked;
1539 		}
1540 		/*
1541 		 * Make sure the name isn't too long for the prison or its
1542 		 * children.
1543 		 */
1544 		onamelen = strlen(pr->pr_name);
1545 		namelen = strlen(name);
1546 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1547 			error = ENAMETOOLONG;
1548 			goto done_deref_locked;
1549 		}
1550 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1551 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1552 			    sizeof(pr->pr_name)) {
1553 				error = ENAMETOOLONG;
1554 				goto done_deref_locked;
1555 			}
1556 		}
1557 	}
1558 	if (pr_allow & ~ppr->pr_allow) {
1559 		error = EPERM;
1560 		goto done_deref_locked;
1561 	}
1562 
1563 	/* Set the parameters of the prison. */
1564 #ifdef INET
1565 	redo_ip4 = 0;
1566 	if (pr_flags & PR_IP4_USER) {
1567 		pr->pr_flags |= PR_IP4;
1568 		free(pr->pr_ip4, M_PRISON);
1569 		pr->pr_ip4s = ip4s;
1570 		pr->pr_ip4 = ip4;
1571 		ip4 = NULL;
1572 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1573 #ifdef VIMAGE
1574 			if (tpr->pr_flags & PR_VNET) {
1575 				descend = 0;
1576 				continue;
1577 			}
1578 #endif
1579 			if (prison_restrict_ip4(tpr, NULL)) {
1580 				redo_ip4 = 1;
1581 				descend = 0;
1582 			}
1583 		}
1584 	}
1585 #endif
1586 #ifdef INET6
1587 	redo_ip6 = 0;
1588 	if (pr_flags & PR_IP6_USER) {
1589 		pr->pr_flags |= PR_IP6;
1590 		free(pr->pr_ip6, M_PRISON);
1591 		pr->pr_ip6s = ip6s;
1592 		pr->pr_ip6 = ip6;
1593 		ip6 = NULL;
1594 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1595 #ifdef VIMAGE
1596 			if (tpr->pr_flags & PR_VNET) {
1597 				descend = 0;
1598 				continue;
1599 			}
1600 #endif
1601 			if (prison_restrict_ip6(tpr, NULL)) {
1602 				redo_ip6 = 1;
1603 				descend = 0;
1604 			}
1605 		}
1606 	}
1607 #endif
1608 	if (gotslevel) {
1609 		pr->pr_securelevel = slevel;
1610 		/* Set all child jails to be at least this level. */
1611 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1612 			if (tpr->pr_securelevel < slevel)
1613 				tpr->pr_securelevel = slevel;
1614 	}
1615 	if (gotchildmax) {
1616 		pr->pr_childmax = childmax;
1617 		/* Set all child jails to under this limit. */
1618 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1619 			if (tpr->pr_childmax > childmax - level)
1620 				tpr->pr_childmax = childmax > level
1621 				    ? childmax - level : 0;
1622 	}
1623 	if (gotenforce) {
1624 		pr->pr_enforce_statfs = enforce;
1625 		/* Pass this restriction on to the children. */
1626 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1627 			if (tpr->pr_enforce_statfs < enforce)
1628 				tpr->pr_enforce_statfs = enforce;
1629 	}
1630 	if (gotrsnum) {
1631 		pr->pr_devfs_rsnum = rsnum;
1632 		/* Pass this restriction on to the children. */
1633 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1634 			tpr->pr_devfs_rsnum = rsnum;
1635 	}
1636 	if (name != NULL) {
1637 		if (ppr == &prison0)
1638 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1639 		else
1640 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1641 			    ppr->pr_name, name);
1642 		/* Change this component of child names. */
1643 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1644 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1645 			    strlen(tpr->pr_name + onamelen) + 1);
1646 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1647 		}
1648 	}
1649 	if (path != NULL) {
1650 		/* Try to keep a real-rooted full pathname. */
1651 		if (fullpath_disabled && path[0] == '/' &&
1652 		    strcmp(mypr->pr_path, "/"))
1653 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1654 			    mypr->pr_path, path);
1655 		else
1656 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1657 		pr->pr_root = root;
1658 	}
1659 	if (PR_HOST & ch_flags & ~pr_flags) {
1660 		if (pr->pr_flags & PR_HOST) {
1661 			/*
1662 			 * Copy the parent's host info.  As with pr_ip4 above,
1663 			 * the lack of a lock on the parent is not a problem;
1664 			 * it is always set with allprison_lock at least
1665 			 * shared, and is held exclusively here.
1666 			 */
1667 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1668 			    sizeof(pr->pr_hostname));
1669 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1670 			    sizeof(pr->pr_domainname));
1671 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1672 			    sizeof(pr->pr_hostuuid));
1673 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1674 		}
1675 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1676 		/* Set this prison, and any descendants without PR_HOST. */
1677 		if (host != NULL)
1678 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1679 		if (domain != NULL)
1680 			strlcpy(pr->pr_domainname, domain,
1681 			    sizeof(pr->pr_domainname));
1682 		if (uuid != NULL)
1683 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1684 		if (gothid)
1685 			pr->pr_hostid = hid;
1686 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1687 			if (tpr->pr_flags & PR_HOST)
1688 				descend = 0;
1689 			else {
1690 				if (host != NULL)
1691 					strlcpy(tpr->pr_hostname,
1692 					    pr->pr_hostname,
1693 					    sizeof(tpr->pr_hostname));
1694 				if (domain != NULL)
1695 					strlcpy(tpr->pr_domainname,
1696 					    pr->pr_domainname,
1697 					    sizeof(tpr->pr_domainname));
1698 				if (uuid != NULL)
1699 					strlcpy(tpr->pr_hostuuid,
1700 					    pr->pr_hostuuid,
1701 					    sizeof(tpr->pr_hostuuid));
1702 				if (gothid)
1703 					tpr->pr_hostid = hid;
1704 			}
1705 		}
1706 	}
1707 	if ((tallow = ch_allow & ~pr_allow)) {
1708 		/* Clear allow bits in all children. */
1709 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1710 			tpr->pr_allow &= ~tallow;
1711 	}
1712 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1713 	/*
1714 	 * Persistent prisons get an extra reference, and prisons losing their
1715 	 * persist flag lose that reference.  Only do this for existing prisons
1716 	 * for now, so new ones will remain unseen until after the module
1717 	 * handlers have completed.
1718 	 */
1719 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1720 		if (pr_flags & PR_PERSIST) {
1721 			pr->pr_ref++;
1722 			pr->pr_uref++;
1723 		} else {
1724 			pr->pr_ref--;
1725 			pr->pr_uref--;
1726 		}
1727 	}
1728 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1729 	mtx_unlock(&pr->pr_mtx);
1730 
1731 #ifdef RACCT
1732 	if (created)
1733 		prison_racct_attach(pr);
1734 #endif
1735 
1736 	/* Locks may have prevented a complete restriction of child IP
1737 	 * addresses.  If so, allocate some more memory and try again.
1738 	 */
1739 #ifdef INET
1740 	while (redo_ip4) {
1741 		ip4s = pr->pr_ip4s;
1742 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1743 		mtx_lock(&pr->pr_mtx);
1744 		redo_ip4 = 0;
1745 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1746 #ifdef VIMAGE
1747 			if (tpr->pr_flags & PR_VNET) {
1748 				descend = 0;
1749 				continue;
1750 			}
1751 #endif
1752 			if (prison_restrict_ip4(tpr, ip4)) {
1753 				if (ip4 != NULL)
1754 					ip4 = NULL;
1755 				else
1756 					redo_ip4 = 1;
1757 			}
1758 		}
1759 		mtx_unlock(&pr->pr_mtx);
1760 	}
1761 #endif
1762 #ifdef INET6
1763 	while (redo_ip6) {
1764 		ip6s = pr->pr_ip6s;
1765 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1766 		mtx_lock(&pr->pr_mtx);
1767 		redo_ip6 = 0;
1768 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1769 #ifdef VIMAGE
1770 			if (tpr->pr_flags & PR_VNET) {
1771 				descend = 0;
1772 				continue;
1773 			}
1774 #endif
1775 			if (prison_restrict_ip6(tpr, ip6)) {
1776 				if (ip6 != NULL)
1777 					ip6 = NULL;
1778 				else
1779 					redo_ip6 = 1;
1780 			}
1781 		}
1782 		mtx_unlock(&pr->pr_mtx);
1783 	}
1784 #endif
1785 
1786 	/* Let the modules do their work. */
1787 	sx_downgrade(&allprison_lock);
1788 	if (created) {
1789 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1790 		if (error) {
1791 			prison_deref(pr, PD_LIST_SLOCKED);
1792 			goto done_errmsg;
1793 		}
1794 	}
1795 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1796 	if (error) {
1797 		prison_deref(pr, created
1798 		    ? PD_LIST_SLOCKED
1799 		    : PD_DEREF | PD_LIST_SLOCKED);
1800 		goto done_errmsg;
1801 	}
1802 
1803 	/* Attach this process to the prison if requested. */
1804 	if (flags & JAIL_ATTACH) {
1805 		mtx_lock(&pr->pr_mtx);
1806 		error = do_jail_attach(td, pr);
1807 		if (error) {
1808 			vfs_opterror(opts, "attach failed");
1809 			if (!created)
1810 				prison_deref(pr, PD_DEREF);
1811 			goto done_errmsg;
1812 		}
1813 	}
1814 
1815 #ifdef RACCT
1816 	if (!created) {
1817 		sx_sunlock(&allprison_lock);
1818 		prison_racct_modify(pr);
1819 		sx_slock(&allprison_lock);
1820 	}
1821 #endif
1822 
1823 	td->td_retval[0] = pr->pr_id;
1824 
1825 	/*
1826 	 * Now that it is all there, drop the temporary reference from existing
1827 	 * prisons.  Or add a reference to newly created persistent prisons
1828 	 * (which was not done earlier so that the prison would not be publicly
1829 	 * visible).
1830 	 */
1831 	if (!created) {
1832 		prison_deref(pr, (flags & JAIL_ATTACH)
1833 		    ? PD_DEREF
1834 		    : PD_DEREF | PD_LIST_SLOCKED);
1835 	} else {
1836 		if (pr_flags & PR_PERSIST) {
1837 			mtx_lock(&pr->pr_mtx);
1838 			pr->pr_ref++;
1839 			pr->pr_uref++;
1840 			mtx_unlock(&pr->pr_mtx);
1841 		}
1842 		if (!(flags & JAIL_ATTACH))
1843 			sx_sunlock(&allprison_lock);
1844 	}
1845 
1846 	goto done_errmsg;
1847 
1848  done_deref_locked:
1849 	prison_deref(pr, created
1850 	    ? PD_LOCKED | PD_LIST_XLOCKED
1851 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1852 	goto done_releroot;
1853  done_unlock_list:
1854 	sx_xunlock(&allprison_lock);
1855  done_releroot:
1856 	if (root != NULL)
1857 		vrele(root);
1858  done_errmsg:
1859 	if (error) {
1860 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1861 		if (errmsg_len > 0) {
1862 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1863 			if (errmsg_pos > 0) {
1864 				if (optuio->uio_segflg == UIO_SYSSPACE)
1865 					bcopy(errmsg,
1866 					   optuio->uio_iov[errmsg_pos].iov_base,
1867 					   errmsg_len);
1868 				else
1869 					copyout(errmsg,
1870 					   optuio->uio_iov[errmsg_pos].iov_base,
1871 					   errmsg_len);
1872 			}
1873 		}
1874 	}
1875  done_free:
1876 #ifdef INET
1877 	free(ip4, M_PRISON);
1878 #endif
1879 #ifdef INET6
1880 	free(ip6, M_PRISON);
1881 #endif
1882 	if (g_path != NULL)
1883 		free(g_path, M_TEMP);
1884 	vfs_freeopts(opts);
1885 	return (error);
1886 }
1887 
1888 
1889 /*
1890  * struct jail_get_args {
1891  *	struct iovec *iovp;
1892  *	unsigned int iovcnt;
1893  *	int flags;
1894  * };
1895  */
1896 int
1897 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1898 {
1899 	struct uio *auio;
1900 	int error;
1901 
1902 	/* Check that we have an even number of iovecs. */
1903 	if (uap->iovcnt & 1)
1904 		return (EINVAL);
1905 
1906 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1907 	if (error)
1908 		return (error);
1909 	error = kern_jail_get(td, auio, uap->flags);
1910 	if (error == 0)
1911 		error = copyout(auio->uio_iov, uap->iovp,
1912 		    uap->iovcnt * sizeof (struct iovec));
1913 	free(auio, M_IOV);
1914 	return (error);
1915 }
1916 
1917 int
1918 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1919 {
1920 	struct prison *pr, *mypr;
1921 	struct vfsopt *opt;
1922 	struct vfsoptlist *opts;
1923 	char *errmsg, *name;
1924 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1925 
1926 	if (flags & ~JAIL_GET_MASK)
1927 		return (EINVAL);
1928 
1929 	/* Get the parameter list. */
1930 	error = vfs_buildopts(optuio, &opts);
1931 	if (error)
1932 		return (error);
1933 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1934 	mypr = td->td_ucred->cr_prison;
1935 
1936 	/*
1937 	 * Find the prison specified by one of: lastjid, jid, name.
1938 	 */
1939 	sx_slock(&allprison_lock);
1940 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1941 	if (error == 0) {
1942 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1943 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1944 				mtx_lock(&pr->pr_mtx);
1945 				if (pr->pr_ref > 0 &&
1946 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1947 					break;
1948 				mtx_unlock(&pr->pr_mtx);
1949 			}
1950 		}
1951 		if (pr != NULL)
1952 			goto found_prison;
1953 		error = ENOENT;
1954 		vfs_opterror(opts, "no jail after %d", jid);
1955 		goto done_unlock_list;
1956 	} else if (error != ENOENT)
1957 		goto done_unlock_list;
1958 
1959 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1960 	if (error == 0) {
1961 		if (jid != 0) {
1962 			pr = prison_find_child(mypr, jid);
1963 			if (pr != NULL) {
1964 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1965 					mtx_unlock(&pr->pr_mtx);
1966 					error = ENOENT;
1967 					vfs_opterror(opts, "jail %d is dying",
1968 					    jid);
1969 					goto done_unlock_list;
1970 				}
1971 				goto found_prison;
1972 			}
1973 			error = ENOENT;
1974 			vfs_opterror(opts, "jail %d not found", jid);
1975 			goto done_unlock_list;
1976 		}
1977 	} else if (error != ENOENT)
1978 		goto done_unlock_list;
1979 
1980 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1981 	if (error == 0) {
1982 		if (len == 0 || name[len - 1] != '\0') {
1983 			error = EINVAL;
1984 			goto done_unlock_list;
1985 		}
1986 		pr = prison_find_name(mypr, name);
1987 		if (pr != NULL) {
1988 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1989 				mtx_unlock(&pr->pr_mtx);
1990 				error = ENOENT;
1991 				vfs_opterror(opts, "jail \"%s\" is dying",
1992 				    name);
1993 				goto done_unlock_list;
1994 			}
1995 			goto found_prison;
1996 		}
1997 		error = ENOENT;
1998 		vfs_opterror(opts, "jail \"%s\" not found", name);
1999 		goto done_unlock_list;
2000 	} else if (error != ENOENT)
2001 		goto done_unlock_list;
2002 
2003 	vfs_opterror(opts, "no jail specified");
2004 	error = ENOENT;
2005 	goto done_unlock_list;
2006 
2007  found_prison:
2008 	/* Get the parameters of the prison. */
2009 	pr->pr_ref++;
2010 	locked = PD_LOCKED;
2011 	td->td_retval[0] = pr->pr_id;
2012 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2013 	if (error != 0 && error != ENOENT)
2014 		goto done_deref;
2015 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2016 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2017 	if (error != 0 && error != ENOENT)
2018 		goto done_deref;
2019 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2020 	if (error != 0 && error != ENOENT)
2021 		goto done_deref;
2022 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2023 	    sizeof(pr->pr_cpuset->cs_id));
2024 	if (error != 0 && error != ENOENT)
2025 		goto done_deref;
2026 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2027 	if (error != 0 && error != ENOENT)
2028 		goto done_deref;
2029 #ifdef INET
2030 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2031 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2032 	if (error != 0 && error != ENOENT)
2033 		goto done_deref;
2034 #endif
2035 #ifdef INET6
2036 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2037 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2038 	if (error != 0 && error != ENOENT)
2039 		goto done_deref;
2040 #endif
2041 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2042 	    sizeof(pr->pr_securelevel));
2043 	if (error != 0 && error != ENOENT)
2044 		goto done_deref;
2045 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2046 	    sizeof(pr->pr_childcount));
2047 	if (error != 0 && error != ENOENT)
2048 		goto done_deref;
2049 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2050 	    sizeof(pr->pr_childmax));
2051 	if (error != 0 && error != ENOENT)
2052 		goto done_deref;
2053 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2054 	if (error != 0 && error != ENOENT)
2055 		goto done_deref;
2056 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2057 	if (error != 0 && error != ENOENT)
2058 		goto done_deref;
2059 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2060 	if (error != 0 && error != ENOENT)
2061 		goto done_deref;
2062 #ifdef COMPAT_FREEBSD32
2063 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2064 		uint32_t hid32 = pr->pr_hostid;
2065 
2066 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2067 	} else
2068 #endif
2069 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2070 	    sizeof(pr->pr_hostid));
2071 	if (error != 0 && error != ENOENT)
2072 		goto done_deref;
2073 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2074 	    sizeof(pr->pr_enforce_statfs));
2075 	if (error != 0 && error != ENOENT)
2076 		goto done_deref;
2077 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2078 	    sizeof(pr->pr_devfs_rsnum));
2079 	if (error != 0 && error != ENOENT)
2080 		goto done_deref;
2081 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2082 	    fi++) {
2083 		if (pr_flag_names[fi] == NULL)
2084 			continue;
2085 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2086 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2087 		if (error != 0 && error != ENOENT)
2088 			goto done_deref;
2089 		i = !i;
2090 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2091 		if (error != 0 && error != ENOENT)
2092 			goto done_deref;
2093 	}
2094 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2095 	    fi++) {
2096 		i = pr->pr_flags &
2097 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2098 		i = pr_flag_jailsys[fi].disable &&
2099 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2100 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2101 		    : JAIL_SYS_INHERIT;
2102 		error =
2103 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2104 		if (error != 0 && error != ENOENT)
2105 			goto done_deref;
2106 	}
2107 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2108 	    fi++) {
2109 		if (pr_allow_names[fi] == NULL)
2110 			continue;
2111 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2112 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2113 		if (error != 0 && error != ENOENT)
2114 			goto done_deref;
2115 		i = !i;
2116 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2117 		if (error != 0 && error != ENOENT)
2118 			goto done_deref;
2119 	}
2120 	i = (pr->pr_uref == 0);
2121 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 	i = !i;
2125 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2126 	if (error != 0 && error != ENOENT)
2127 		goto done_deref;
2128 
2129 	/* Get the module parameters. */
2130 	mtx_unlock(&pr->pr_mtx);
2131 	locked = 0;
2132 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2133 	if (error)
2134 		goto done_deref;
2135 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2136 
2137 	/* By now, all parameters should have been noted. */
2138 	TAILQ_FOREACH(opt, opts, link) {
2139 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2140 			error = EINVAL;
2141 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2142 			goto done_errmsg;
2143 		}
2144 	}
2145 
2146 	/* Write the fetched parameters back to userspace. */
2147 	error = 0;
2148 	TAILQ_FOREACH(opt, opts, link) {
2149 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2150 			pos = 2 * opt->pos + 1;
2151 			optuio->uio_iov[pos].iov_len = opt->len;
2152 			if (opt->value != NULL) {
2153 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2154 					bcopy(opt->value,
2155 					    optuio->uio_iov[pos].iov_base,
2156 					    opt->len);
2157 				} else {
2158 					error = copyout(opt->value,
2159 					    optuio->uio_iov[pos].iov_base,
2160 					    opt->len);
2161 					if (error)
2162 						break;
2163 				}
2164 			}
2165 		}
2166 	}
2167 	goto done_errmsg;
2168 
2169  done_deref:
2170 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2171 	goto done_errmsg;
2172 
2173  done_unlock_list:
2174 	sx_sunlock(&allprison_lock);
2175  done_errmsg:
2176 	if (error && errmsg_pos >= 0) {
2177 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2178 		errmsg_pos = 2 * errmsg_pos + 1;
2179 		if (errmsg_len > 0) {
2180 			if (optuio->uio_segflg == UIO_SYSSPACE)
2181 				bcopy(errmsg,
2182 				    optuio->uio_iov[errmsg_pos].iov_base,
2183 				    errmsg_len);
2184 			else
2185 				copyout(errmsg,
2186 				    optuio->uio_iov[errmsg_pos].iov_base,
2187 				    errmsg_len);
2188 		}
2189 	}
2190 	vfs_freeopts(opts);
2191 	return (error);
2192 }
2193 
2194 
2195 /*
2196  * struct jail_remove_args {
2197  *	int jid;
2198  * };
2199  */
2200 int
2201 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2202 {
2203 	struct prison *pr, *cpr, *lpr, *tpr;
2204 	int descend, error;
2205 
2206 	error = priv_check(td, PRIV_JAIL_REMOVE);
2207 	if (error)
2208 		return (error);
2209 
2210 	sx_xlock(&allprison_lock);
2211 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2212 	if (pr == NULL) {
2213 		sx_xunlock(&allprison_lock);
2214 		return (EINVAL);
2215 	}
2216 
2217 	/* Remove all descendants of this prison, then remove this prison. */
2218 	pr->pr_ref++;
2219 	pr->pr_flags |= PR_REMOVE;
2220 	if (!LIST_EMPTY(&pr->pr_children)) {
2221 		mtx_unlock(&pr->pr_mtx);
2222 		lpr = NULL;
2223 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2224 			mtx_lock(&cpr->pr_mtx);
2225 			if (cpr->pr_ref > 0) {
2226 				tpr = cpr;
2227 				cpr->pr_ref++;
2228 				cpr->pr_flags |= PR_REMOVE;
2229 			} else {
2230 				/* Already removed - do not do it again. */
2231 				tpr = NULL;
2232 			}
2233 			mtx_unlock(&cpr->pr_mtx);
2234 			if (lpr != NULL) {
2235 				mtx_lock(&lpr->pr_mtx);
2236 				prison_remove_one(lpr);
2237 				sx_xlock(&allprison_lock);
2238 			}
2239 			lpr = tpr;
2240 		}
2241 		if (lpr != NULL) {
2242 			mtx_lock(&lpr->pr_mtx);
2243 			prison_remove_one(lpr);
2244 			sx_xlock(&allprison_lock);
2245 		}
2246 		mtx_lock(&pr->pr_mtx);
2247 	}
2248 	prison_remove_one(pr);
2249 	return (0);
2250 }
2251 
2252 static void
2253 prison_remove_one(struct prison *pr)
2254 {
2255 	struct proc *p;
2256 	int deuref;
2257 
2258 	/* If the prison was persistent, it is not anymore. */
2259 	deuref = 0;
2260 	if (pr->pr_flags & PR_PERSIST) {
2261 		pr->pr_ref--;
2262 		deuref = PD_DEUREF;
2263 		pr->pr_flags &= ~PR_PERSIST;
2264 	}
2265 
2266 	/*
2267 	 * jail_remove added a reference.  If that's the only one, remove
2268 	 * the prison now.
2269 	 */
2270 	KASSERT(pr->pr_ref > 0,
2271 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2272 	if (pr->pr_ref == 1) {
2273 		prison_deref(pr,
2274 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2275 		return;
2276 	}
2277 
2278 	mtx_unlock(&pr->pr_mtx);
2279 	sx_xunlock(&allprison_lock);
2280 	/*
2281 	 * Kill all processes unfortunate enough to be attached to this prison.
2282 	 */
2283 	sx_slock(&allproc_lock);
2284 	LIST_FOREACH(p, &allproc, p_list) {
2285 		PROC_LOCK(p);
2286 		if (p->p_state != PRS_NEW && p->p_ucred &&
2287 		    p->p_ucred->cr_prison == pr)
2288 			kern_psignal(p, SIGKILL);
2289 		PROC_UNLOCK(p);
2290 	}
2291 	sx_sunlock(&allproc_lock);
2292 	/* Remove the temporary reference added by jail_remove. */
2293 	prison_deref(pr, deuref | PD_DEREF);
2294 }
2295 
2296 
2297 /*
2298  * struct jail_attach_args {
2299  *	int jid;
2300  * };
2301  */
2302 int
2303 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2304 {
2305 	struct prison *pr;
2306 	int error;
2307 
2308 	error = priv_check(td, PRIV_JAIL_ATTACH);
2309 	if (error)
2310 		return (error);
2311 
2312 	sx_slock(&allprison_lock);
2313 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2314 	if (pr == NULL) {
2315 		sx_sunlock(&allprison_lock);
2316 		return (EINVAL);
2317 	}
2318 
2319 	/*
2320 	 * Do not allow a process to attach to a prison that is not
2321 	 * considered to be "alive".
2322 	 */
2323 	if (pr->pr_uref == 0) {
2324 		mtx_unlock(&pr->pr_mtx);
2325 		sx_sunlock(&allprison_lock);
2326 		return (EINVAL);
2327 	}
2328 
2329 	return (do_jail_attach(td, pr));
2330 }
2331 
2332 static int
2333 do_jail_attach(struct thread *td, struct prison *pr)
2334 {
2335 	struct prison *ppr;
2336 	struct proc *p;
2337 	struct ucred *newcred, *oldcred;
2338 	int error;
2339 
2340 	/*
2341 	 * XXX: Note that there is a slight race here if two threads
2342 	 * in the same privileged process attempt to attach to two
2343 	 * different jails at the same time.  It is important for
2344 	 * user processes not to do this, or they might end up with
2345 	 * a process root from one prison, but attached to the jail
2346 	 * of another.
2347 	 */
2348 	pr->pr_ref++;
2349 	pr->pr_uref++;
2350 	mtx_unlock(&pr->pr_mtx);
2351 
2352 	/* Let modules do whatever they need to prepare for attaching. */
2353 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2354 	if (error) {
2355 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2356 		return (error);
2357 	}
2358 	sx_sunlock(&allprison_lock);
2359 
2360 	/*
2361 	 * Reparent the newly attached process to this jail.
2362 	 */
2363 	ppr = td->td_ucred->cr_prison;
2364 	p = td->td_proc;
2365 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2366 	if (error)
2367 		goto e_revert_osd;
2368 
2369 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2370 	if ((error = change_dir(pr->pr_root, td)) != 0)
2371 		goto e_unlock;
2372 #ifdef MAC
2373 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2374 		goto e_unlock;
2375 #endif
2376 	VOP_UNLOCK(pr->pr_root, 0);
2377 	if ((error = change_root(pr->pr_root, td)))
2378 		goto e_revert_osd;
2379 
2380 	newcred = crget();
2381 	PROC_LOCK(p);
2382 	oldcred = p->p_ucred;
2383 	setsugid(p);
2384 	crcopy(newcred, oldcred);
2385 	newcred->cr_prison = pr;
2386 	p->p_ucred = newcred;
2387 	PROC_UNLOCK(p);
2388 #ifdef RACCT
2389 	racct_proc_ucred_changed(p, oldcred, newcred);
2390 #endif
2391 	crfree(oldcred);
2392 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2393 	return (0);
2394  e_unlock:
2395 	VOP_UNLOCK(pr->pr_root, 0);
2396  e_revert_osd:
2397 	/* Tell modules this thread is still in its old jail after all. */
2398 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2399 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2400 	return (error);
2401 }
2402 
2403 
2404 /*
2405  * Returns a locked prison instance, or NULL on failure.
2406  */
2407 struct prison *
2408 prison_find(int prid)
2409 {
2410 	struct prison *pr;
2411 
2412 	sx_assert(&allprison_lock, SX_LOCKED);
2413 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2414 		if (pr->pr_id == prid) {
2415 			mtx_lock(&pr->pr_mtx);
2416 			if (pr->pr_ref > 0)
2417 				return (pr);
2418 			mtx_unlock(&pr->pr_mtx);
2419 		}
2420 	}
2421 	return (NULL);
2422 }
2423 
2424 /*
2425  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2426  */
2427 struct prison *
2428 prison_find_child(struct prison *mypr, int prid)
2429 {
2430 	struct prison *pr;
2431 	int descend;
2432 
2433 	sx_assert(&allprison_lock, SX_LOCKED);
2434 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2435 		if (pr->pr_id == prid) {
2436 			mtx_lock(&pr->pr_mtx);
2437 			if (pr->pr_ref > 0)
2438 				return (pr);
2439 			mtx_unlock(&pr->pr_mtx);
2440 		}
2441 	}
2442 	return (NULL);
2443 }
2444 
2445 /*
2446  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2447  */
2448 struct prison *
2449 prison_find_name(struct prison *mypr, const char *name)
2450 {
2451 	struct prison *pr, *deadpr;
2452 	size_t mylen;
2453 	int descend;
2454 
2455 	sx_assert(&allprison_lock, SX_LOCKED);
2456 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2457  again:
2458 	deadpr = NULL;
2459 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2460 		if (!strcmp(pr->pr_name + mylen, name)) {
2461 			mtx_lock(&pr->pr_mtx);
2462 			if (pr->pr_ref > 0) {
2463 				if (pr->pr_uref > 0)
2464 					return (pr);
2465 				deadpr = pr;
2466 			}
2467 			mtx_unlock(&pr->pr_mtx);
2468 		}
2469 	}
2470 	/* There was no valid prison - perhaps there was a dying one. */
2471 	if (deadpr != NULL) {
2472 		mtx_lock(&deadpr->pr_mtx);
2473 		if (deadpr->pr_ref == 0) {
2474 			mtx_unlock(&deadpr->pr_mtx);
2475 			goto again;
2476 		}
2477 	}
2478 	return (deadpr);
2479 }
2480 
2481 /*
2482  * See if a prison has the specific flag set.
2483  */
2484 int
2485 prison_flag(struct ucred *cred, unsigned flag)
2486 {
2487 
2488 	/* This is an atomic read, so no locking is necessary. */
2489 	return (cred->cr_prison->pr_flags & flag);
2490 }
2491 
2492 int
2493 prison_allow(struct ucred *cred, unsigned flag)
2494 {
2495 
2496 	/* This is an atomic read, so no locking is necessary. */
2497 	return (cred->cr_prison->pr_allow & flag);
2498 }
2499 
2500 /*
2501  * Remove a prison reference.  If that was the last reference, remove the
2502  * prison itself - but not in this context in case there are locks held.
2503  */
2504 void
2505 prison_free_locked(struct prison *pr)
2506 {
2507 
2508 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2509 	pr->pr_ref--;
2510 	if (pr->pr_ref == 0) {
2511 		mtx_unlock(&pr->pr_mtx);
2512 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2513 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2514 		return;
2515 	}
2516 	mtx_unlock(&pr->pr_mtx);
2517 }
2518 
2519 void
2520 prison_free(struct prison *pr)
2521 {
2522 
2523 	mtx_lock(&pr->pr_mtx);
2524 	prison_free_locked(pr);
2525 }
2526 
2527 static void
2528 prison_complete(void *context, int pending)
2529 {
2530 
2531 	prison_deref((struct prison *)context, 0);
2532 }
2533 
2534 /*
2535  * Remove a prison reference (usually).  This internal version assumes no
2536  * mutexes are held, except perhaps the prison itself.  If there are no more
2537  * references, release and delist the prison.  On completion, the prison lock
2538  * and the allprison lock are both unlocked.
2539  */
2540 static void
2541 prison_deref(struct prison *pr, int flags)
2542 {
2543 	struct prison *ppr, *tpr;
2544 
2545 	if (!(flags & PD_LOCKED))
2546 		mtx_lock(&pr->pr_mtx);
2547 	for (;;) {
2548 		if (flags & PD_DEUREF) {
2549 			pr->pr_uref--;
2550 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2551 		}
2552 		if (flags & PD_DEREF)
2553 			pr->pr_ref--;
2554 		/* If the prison still has references, nothing else to do. */
2555 		if (pr->pr_ref > 0) {
2556 			mtx_unlock(&pr->pr_mtx);
2557 			if (flags & PD_LIST_SLOCKED)
2558 				sx_sunlock(&allprison_lock);
2559 			else if (flags & PD_LIST_XLOCKED)
2560 				sx_xunlock(&allprison_lock);
2561 			return;
2562 		}
2563 
2564 		mtx_unlock(&pr->pr_mtx);
2565 		if (flags & PD_LIST_SLOCKED) {
2566 			if (!sx_try_upgrade(&allprison_lock)) {
2567 				sx_sunlock(&allprison_lock);
2568 				sx_xlock(&allprison_lock);
2569 			}
2570 		} else if (!(flags & PD_LIST_XLOCKED))
2571 			sx_xlock(&allprison_lock);
2572 
2573 		TAILQ_REMOVE(&allprison, pr, pr_list);
2574 		LIST_REMOVE(pr, pr_sibling);
2575 		ppr = pr->pr_parent;
2576 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2577 			tpr->pr_childcount--;
2578 		sx_xunlock(&allprison_lock);
2579 
2580 #ifdef VIMAGE
2581 		if (pr->pr_vnet != ppr->pr_vnet)
2582 			vnet_destroy(pr->pr_vnet);
2583 #endif
2584 		if (pr->pr_root != NULL)
2585 			vrele(pr->pr_root);
2586 		mtx_destroy(&pr->pr_mtx);
2587 #ifdef INET
2588 		free(pr->pr_ip4, M_PRISON);
2589 #endif
2590 #ifdef INET6
2591 		free(pr->pr_ip6, M_PRISON);
2592 #endif
2593 		if (pr->pr_cpuset != NULL)
2594 			cpuset_rel(pr->pr_cpuset);
2595 		osd_jail_exit(pr);
2596 #ifdef RACCT
2597 		prison_racct_detach(pr);
2598 #endif
2599 		free(pr, M_PRISON);
2600 
2601 		/* Removing a prison frees a reference on its parent. */
2602 		pr = ppr;
2603 		mtx_lock(&pr->pr_mtx);
2604 		flags = PD_DEREF | PD_DEUREF;
2605 	}
2606 }
2607 
2608 void
2609 prison_hold_locked(struct prison *pr)
2610 {
2611 
2612 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2613 	KASSERT(pr->pr_ref > 0,
2614 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2615 	pr->pr_ref++;
2616 }
2617 
2618 void
2619 prison_hold(struct prison *pr)
2620 {
2621 
2622 	mtx_lock(&pr->pr_mtx);
2623 	prison_hold_locked(pr);
2624 	mtx_unlock(&pr->pr_mtx);
2625 }
2626 
2627 void
2628 prison_proc_hold(struct prison *pr)
2629 {
2630 
2631 	mtx_lock(&pr->pr_mtx);
2632 	KASSERT(pr->pr_uref > 0,
2633 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2634 	pr->pr_uref++;
2635 	mtx_unlock(&pr->pr_mtx);
2636 }
2637 
2638 void
2639 prison_proc_free(struct prison *pr)
2640 {
2641 
2642 	mtx_lock(&pr->pr_mtx);
2643 	KASSERT(pr->pr_uref > 0,
2644 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2645 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2646 }
2647 
2648 
2649 #ifdef INET
2650 /*
2651  * Restrict a prison's IP address list with its parent's, possibly replacing
2652  * it.  Return true if the replacement buffer was used (or would have been).
2653  */
2654 static int
2655 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2656 {
2657 	int ii, ij, used;
2658 	struct prison *ppr;
2659 
2660 	ppr = pr->pr_parent;
2661 	if (!(pr->pr_flags & PR_IP4_USER)) {
2662 		/* This has no user settings, so just copy the parent's list. */
2663 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2664 			/*
2665 			 * There's no room for the parent's list.  Use the
2666 			 * new list buffer, which is assumed to be big enough
2667 			 * (if it was passed).  If there's no buffer, try to
2668 			 * allocate one.
2669 			 */
2670 			used = 1;
2671 			if (newip4 == NULL) {
2672 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2673 				    M_PRISON, M_NOWAIT);
2674 				if (newip4 != NULL)
2675 					used = 0;
2676 			}
2677 			if (newip4 != NULL) {
2678 				bcopy(ppr->pr_ip4, newip4,
2679 				    ppr->pr_ip4s * sizeof(*newip4));
2680 				free(pr->pr_ip4, M_PRISON);
2681 				pr->pr_ip4 = newip4;
2682 				pr->pr_ip4s = ppr->pr_ip4s;
2683 			}
2684 			return (used);
2685 		}
2686 		pr->pr_ip4s = ppr->pr_ip4s;
2687 		if (pr->pr_ip4s > 0)
2688 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2689 			    pr->pr_ip4s * sizeof(*newip4));
2690 		else if (pr->pr_ip4 != NULL) {
2691 			free(pr->pr_ip4, M_PRISON);
2692 			pr->pr_ip4 = NULL;
2693 		}
2694 	} else if (pr->pr_ip4s > 0) {
2695 		/* Remove addresses that aren't in the parent. */
2696 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2697 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2698 				break;
2699 		if (ij < ppr->pr_ip4s)
2700 			ii = 1;
2701 		else {
2702 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2703 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2704 			ii = 0;
2705 		}
2706 		for (ij = 1; ii < pr->pr_ip4s; ) {
2707 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2708 				ii++;
2709 				continue;
2710 			}
2711 			switch (ij >= ppr->pr_ip4s ? -1 :
2712 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2713 			case -1:
2714 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2715 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2716 				break;
2717 			case 0:
2718 				ii++;
2719 				ij++;
2720 				break;
2721 			case 1:
2722 				ij++;
2723 				break;
2724 			}
2725 		}
2726 		if (pr->pr_ip4s == 0) {
2727 			pr->pr_flags |= PR_IP4_DISABLE;
2728 			free(pr->pr_ip4, M_PRISON);
2729 			pr->pr_ip4 = NULL;
2730 		}
2731 	}
2732 	return (0);
2733 }
2734 
2735 /*
2736  * Pass back primary IPv4 address of this jail.
2737  *
2738  * If not restricted return success but do not alter the address.  Caller has
2739  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2740  *
2741  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2742  * Address returned in NBO.
2743  */
2744 int
2745 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2746 {
2747 	struct prison *pr;
2748 
2749 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2750 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2751 
2752 	pr = cred->cr_prison;
2753 	if (!(pr->pr_flags & PR_IP4))
2754 		return (0);
2755 	mtx_lock(&pr->pr_mtx);
2756 	if (!(pr->pr_flags & PR_IP4)) {
2757 		mtx_unlock(&pr->pr_mtx);
2758 		return (0);
2759 	}
2760 	if (pr->pr_ip4 == NULL) {
2761 		mtx_unlock(&pr->pr_mtx);
2762 		return (EAFNOSUPPORT);
2763 	}
2764 
2765 	ia->s_addr = pr->pr_ip4[0].s_addr;
2766 	mtx_unlock(&pr->pr_mtx);
2767 	return (0);
2768 }
2769 
2770 /*
2771  * Return 1 if we should do proper source address selection or are not jailed.
2772  * We will return 0 if we should bypass source address selection in favour
2773  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2774  * returned in NBO.
2775  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2776  */
2777 int
2778 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2779 {
2780 	struct prison *pr;
2781 	struct in_addr lia;
2782 	int error;
2783 
2784 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2785 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2786 
2787 	if (!jailed(cred))
2788 		return (1);
2789 
2790 	pr = cred->cr_prison;
2791 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2792 		return (1);
2793 
2794 	lia.s_addr = INADDR_ANY;
2795 	error = prison_get_ip4(cred, &lia);
2796 	if (error)
2797 		return (error);
2798 	if (lia.s_addr == INADDR_ANY)
2799 		return (1);
2800 
2801 	ia->s_addr = lia.s_addr;
2802 	return (0);
2803 }
2804 
2805 /*
2806  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2807  */
2808 int
2809 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2810 {
2811 
2812 	if (pr1 == pr2)
2813 		return (1);
2814 
2815 	/*
2816 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2817 	 * existing prisons.
2818 	 */
2819 	while (pr1 != &prison0 &&
2820 #ifdef VIMAGE
2821 	       !(pr1->pr_flags & PR_VNET) &&
2822 #endif
2823 	       !(pr1->pr_flags & PR_IP4_USER))
2824 		pr1 = pr1->pr_parent;
2825 	while (pr2 != &prison0 &&
2826 #ifdef VIMAGE
2827 	       !(pr2->pr_flags & PR_VNET) &&
2828 #endif
2829 	       !(pr2->pr_flags & PR_IP4_USER))
2830 		pr2 = pr2->pr_parent;
2831 	return (pr1 == pr2);
2832 }
2833 
2834 /*
2835  * Make sure our (source) address is set to something meaningful to this
2836  * jail.
2837  *
2838  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2839  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2840  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2841  */
2842 int
2843 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2844 {
2845 	struct prison *pr;
2846 	struct in_addr ia0;
2847 	int error;
2848 
2849 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2850 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2851 
2852 	pr = cred->cr_prison;
2853 	if (!(pr->pr_flags & PR_IP4))
2854 		return (0);
2855 	mtx_lock(&pr->pr_mtx);
2856 	if (!(pr->pr_flags & PR_IP4)) {
2857 		mtx_unlock(&pr->pr_mtx);
2858 		return (0);
2859 	}
2860 	if (pr->pr_ip4 == NULL) {
2861 		mtx_unlock(&pr->pr_mtx);
2862 		return (EAFNOSUPPORT);
2863 	}
2864 
2865 	ia0.s_addr = ntohl(ia->s_addr);
2866 	if (ia0.s_addr == INADDR_LOOPBACK) {
2867 		ia->s_addr = pr->pr_ip4[0].s_addr;
2868 		mtx_unlock(&pr->pr_mtx);
2869 		return (0);
2870 	}
2871 
2872 	if (ia0.s_addr == INADDR_ANY) {
2873 		/*
2874 		 * In case there is only 1 IPv4 address, bind directly.
2875 		 */
2876 		if (pr->pr_ip4s == 1)
2877 			ia->s_addr = pr->pr_ip4[0].s_addr;
2878 		mtx_unlock(&pr->pr_mtx);
2879 		return (0);
2880 	}
2881 
2882 	error = _prison_check_ip4(pr, ia);
2883 	mtx_unlock(&pr->pr_mtx);
2884 	return (error);
2885 }
2886 
2887 /*
2888  * Rewrite destination address in case we will connect to loopback address.
2889  *
2890  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2891  * Address passed in in NBO and returned in NBO.
2892  */
2893 int
2894 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2895 {
2896 	struct prison *pr;
2897 
2898 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2899 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2900 
2901 	pr = cred->cr_prison;
2902 	if (!(pr->pr_flags & PR_IP4))
2903 		return (0);
2904 	mtx_lock(&pr->pr_mtx);
2905 	if (!(pr->pr_flags & PR_IP4)) {
2906 		mtx_unlock(&pr->pr_mtx);
2907 		return (0);
2908 	}
2909 	if (pr->pr_ip4 == NULL) {
2910 		mtx_unlock(&pr->pr_mtx);
2911 		return (EAFNOSUPPORT);
2912 	}
2913 
2914 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2915 		ia->s_addr = pr->pr_ip4[0].s_addr;
2916 		mtx_unlock(&pr->pr_mtx);
2917 		return (0);
2918 	}
2919 
2920 	/*
2921 	 * Return success because nothing had to be changed.
2922 	 */
2923 	mtx_unlock(&pr->pr_mtx);
2924 	return (0);
2925 }
2926 
2927 /*
2928  * Check if given address belongs to the jail referenced by cred/prison.
2929  *
2930  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2931  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2932  * doesn't allow IPv4.  Address passed in in NBO.
2933  */
2934 static int
2935 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
2936 {
2937 	int i, a, z, d;
2938 
2939 	/*
2940 	 * Check the primary IP.
2941 	 */
2942 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2943 		return (0);
2944 
2945 	/*
2946 	 * All the other IPs are sorted so we can do a binary search.
2947 	 */
2948 	a = 0;
2949 	z = pr->pr_ip4s - 2;
2950 	while (a <= z) {
2951 		i = (a + z) / 2;
2952 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2953 		if (d > 0)
2954 			z = i - 1;
2955 		else if (d < 0)
2956 			a = i + 1;
2957 		else
2958 			return (0);
2959 	}
2960 
2961 	return (EADDRNOTAVAIL);
2962 }
2963 
2964 int
2965 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
2966 {
2967 	struct prison *pr;
2968 	int error;
2969 
2970 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2971 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2972 
2973 	pr = cred->cr_prison;
2974 	if (!(pr->pr_flags & PR_IP4))
2975 		return (0);
2976 	mtx_lock(&pr->pr_mtx);
2977 	if (!(pr->pr_flags & PR_IP4)) {
2978 		mtx_unlock(&pr->pr_mtx);
2979 		return (0);
2980 	}
2981 	if (pr->pr_ip4 == NULL) {
2982 		mtx_unlock(&pr->pr_mtx);
2983 		return (EAFNOSUPPORT);
2984 	}
2985 
2986 	error = _prison_check_ip4(pr, ia);
2987 	mtx_unlock(&pr->pr_mtx);
2988 	return (error);
2989 }
2990 #endif
2991 
2992 #ifdef INET6
2993 static int
2994 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2995 {
2996 	int ii, ij, used;
2997 	struct prison *ppr;
2998 
2999 	ppr = pr->pr_parent;
3000 	if (!(pr->pr_flags & PR_IP6_USER)) {
3001 		/* This has no user settings, so just copy the parent's list. */
3002 		if (pr->pr_ip6s < ppr->pr_ip6s) {
3003 			/*
3004 			 * There's no room for the parent's list.  Use the
3005 			 * new list buffer, which is assumed to be big enough
3006 			 * (if it was passed).  If there's no buffer, try to
3007 			 * allocate one.
3008 			 */
3009 			used = 1;
3010 			if (newip6 == NULL) {
3011 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3012 				    M_PRISON, M_NOWAIT);
3013 				if (newip6 != NULL)
3014 					used = 0;
3015 			}
3016 			if (newip6 != NULL) {
3017 				bcopy(ppr->pr_ip6, newip6,
3018 				    ppr->pr_ip6s * sizeof(*newip6));
3019 				free(pr->pr_ip6, M_PRISON);
3020 				pr->pr_ip6 = newip6;
3021 				pr->pr_ip6s = ppr->pr_ip6s;
3022 			}
3023 			return (used);
3024 		}
3025 		pr->pr_ip6s = ppr->pr_ip6s;
3026 		if (pr->pr_ip6s > 0)
3027 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3028 			    pr->pr_ip6s * sizeof(*newip6));
3029 		else if (pr->pr_ip6 != NULL) {
3030 			free(pr->pr_ip6, M_PRISON);
3031 			pr->pr_ip6 = NULL;
3032 		}
3033 	} else if (pr->pr_ip6s > 0) {
3034 		/* Remove addresses that aren't in the parent. */
3035 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3036 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3037 			    &ppr->pr_ip6[ij]))
3038 				break;
3039 		if (ij < ppr->pr_ip6s)
3040 			ii = 1;
3041 		else {
3042 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3043 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3044 			ii = 0;
3045 		}
3046 		for (ij = 1; ii < pr->pr_ip6s; ) {
3047 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3048 			    &ppr->pr_ip6[0])) {
3049 				ii++;
3050 				continue;
3051 			}
3052 			switch (ij >= ppr->pr_ip6s ? -1 :
3053 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3054 			case -1:
3055 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3056 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3057 				break;
3058 			case 0:
3059 				ii++;
3060 				ij++;
3061 				break;
3062 			case 1:
3063 				ij++;
3064 				break;
3065 			}
3066 		}
3067 		if (pr->pr_ip6s == 0) {
3068 			pr->pr_flags |= PR_IP6_DISABLE;
3069 			free(pr->pr_ip6, M_PRISON);
3070 			pr->pr_ip6 = NULL;
3071 		}
3072 	}
3073 	return 0;
3074 }
3075 
3076 /*
3077  * Pass back primary IPv6 address for this jail.
3078  *
3079  * If not restricted return success but do not alter the address.  Caller has
3080  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3081  *
3082  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3083  */
3084 int
3085 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3086 {
3087 	struct prison *pr;
3088 
3089 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3090 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3091 
3092 	pr = cred->cr_prison;
3093 	if (!(pr->pr_flags & PR_IP6))
3094 		return (0);
3095 	mtx_lock(&pr->pr_mtx);
3096 	if (!(pr->pr_flags & PR_IP6)) {
3097 		mtx_unlock(&pr->pr_mtx);
3098 		return (0);
3099 	}
3100 	if (pr->pr_ip6 == NULL) {
3101 		mtx_unlock(&pr->pr_mtx);
3102 		return (EAFNOSUPPORT);
3103 	}
3104 
3105 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3106 	mtx_unlock(&pr->pr_mtx);
3107 	return (0);
3108 }
3109 
3110 /*
3111  * Return 1 if we should do proper source address selection or are not jailed.
3112  * We will return 0 if we should bypass source address selection in favour
3113  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3114  * returned in NBO.
3115  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3116  */
3117 int
3118 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3119 {
3120 	struct prison *pr;
3121 	struct in6_addr lia6;
3122 	int error;
3123 
3124 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3125 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3126 
3127 	if (!jailed(cred))
3128 		return (1);
3129 
3130 	pr = cred->cr_prison;
3131 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3132 		return (1);
3133 
3134 	lia6 = in6addr_any;
3135 	error = prison_get_ip6(cred, &lia6);
3136 	if (error)
3137 		return (error);
3138 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3139 		return (1);
3140 
3141 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3142 	return (0);
3143 }
3144 
3145 /*
3146  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3147  */
3148 int
3149 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3150 {
3151 
3152 	if (pr1 == pr2)
3153 		return (1);
3154 
3155 	while (pr1 != &prison0 &&
3156 #ifdef VIMAGE
3157 	       !(pr1->pr_flags & PR_VNET) &&
3158 #endif
3159 	       !(pr1->pr_flags & PR_IP6_USER))
3160 		pr1 = pr1->pr_parent;
3161 	while (pr2 != &prison0 &&
3162 #ifdef VIMAGE
3163 	       !(pr2->pr_flags & PR_VNET) &&
3164 #endif
3165 	       !(pr2->pr_flags & PR_IP6_USER))
3166 		pr2 = pr2->pr_parent;
3167 	return (pr1 == pr2);
3168 }
3169 
3170 /*
3171  * Make sure our (source) address is set to something meaningful to this jail.
3172  *
3173  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3174  * when needed while binding.
3175  *
3176  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3177  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3178  * doesn't allow IPv6.
3179  */
3180 int
3181 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3182 {
3183 	struct prison *pr;
3184 	int error;
3185 
3186 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3187 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3188 
3189 	pr = cred->cr_prison;
3190 	if (!(pr->pr_flags & PR_IP6))
3191 		return (0);
3192 	mtx_lock(&pr->pr_mtx);
3193 	if (!(pr->pr_flags & PR_IP6)) {
3194 		mtx_unlock(&pr->pr_mtx);
3195 		return (0);
3196 	}
3197 	if (pr->pr_ip6 == NULL) {
3198 		mtx_unlock(&pr->pr_mtx);
3199 		return (EAFNOSUPPORT);
3200 	}
3201 
3202 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3203 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3204 		mtx_unlock(&pr->pr_mtx);
3205 		return (0);
3206 	}
3207 
3208 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3209 		/*
3210 		 * In case there is only 1 IPv6 address, and v6only is true,
3211 		 * then bind directly.
3212 		 */
3213 		if (v6only != 0 && pr->pr_ip6s == 1)
3214 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3215 		mtx_unlock(&pr->pr_mtx);
3216 		return (0);
3217 	}
3218 
3219 	error = _prison_check_ip6(pr, ia6);
3220 	mtx_unlock(&pr->pr_mtx);
3221 	return (error);
3222 }
3223 
3224 /*
3225  * Rewrite destination address in case we will connect to loopback address.
3226  *
3227  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3228  */
3229 int
3230 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3231 {
3232 	struct prison *pr;
3233 
3234 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3235 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3236 
3237 	pr = cred->cr_prison;
3238 	if (!(pr->pr_flags & PR_IP6))
3239 		return (0);
3240 	mtx_lock(&pr->pr_mtx);
3241 	if (!(pr->pr_flags & PR_IP6)) {
3242 		mtx_unlock(&pr->pr_mtx);
3243 		return (0);
3244 	}
3245 	if (pr->pr_ip6 == NULL) {
3246 		mtx_unlock(&pr->pr_mtx);
3247 		return (EAFNOSUPPORT);
3248 	}
3249 
3250 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3251 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3252 		mtx_unlock(&pr->pr_mtx);
3253 		return (0);
3254 	}
3255 
3256 	/*
3257 	 * Return success because nothing had to be changed.
3258 	 */
3259 	mtx_unlock(&pr->pr_mtx);
3260 	return (0);
3261 }
3262 
3263 /*
3264  * Check if given address belongs to the jail referenced by cred/prison.
3265  *
3266  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3267  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3268  * doesn't allow IPv6.
3269  */
3270 static int
3271 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3272 {
3273 	int i, a, z, d;
3274 
3275 	/*
3276 	 * Check the primary IP.
3277 	 */
3278 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3279 		return (0);
3280 
3281 	/*
3282 	 * All the other IPs are sorted so we can do a binary search.
3283 	 */
3284 	a = 0;
3285 	z = pr->pr_ip6s - 2;
3286 	while (a <= z) {
3287 		i = (a + z) / 2;
3288 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3289 		if (d > 0)
3290 			z = i - 1;
3291 		else if (d < 0)
3292 			a = i + 1;
3293 		else
3294 			return (0);
3295 	}
3296 
3297 	return (EADDRNOTAVAIL);
3298 }
3299 
3300 int
3301 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3302 {
3303 	struct prison *pr;
3304 	int error;
3305 
3306 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3307 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3308 
3309 	pr = cred->cr_prison;
3310 	if (!(pr->pr_flags & PR_IP6))
3311 		return (0);
3312 	mtx_lock(&pr->pr_mtx);
3313 	if (!(pr->pr_flags & PR_IP6)) {
3314 		mtx_unlock(&pr->pr_mtx);
3315 		return (0);
3316 	}
3317 	if (pr->pr_ip6 == NULL) {
3318 		mtx_unlock(&pr->pr_mtx);
3319 		return (EAFNOSUPPORT);
3320 	}
3321 
3322 	error = _prison_check_ip6(pr, ia6);
3323 	mtx_unlock(&pr->pr_mtx);
3324 	return (error);
3325 }
3326 #endif
3327 
3328 /*
3329  * Check if a jail supports the given address family.
3330  *
3331  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3332  * if not.
3333  */
3334 int
3335 prison_check_af(struct ucred *cred, int af)
3336 {
3337 	struct prison *pr;
3338 	int error;
3339 
3340 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3341 
3342 	pr = cred->cr_prison;
3343 #ifdef VIMAGE
3344 	/* Prisons with their own network stack are not limited. */
3345 	if (prison_owns_vnet(cred))
3346 		return (0);
3347 #endif
3348 
3349 	error = 0;
3350 	switch (af)
3351 	{
3352 #ifdef INET
3353 	case AF_INET:
3354 		if (pr->pr_flags & PR_IP4)
3355 		{
3356 			mtx_lock(&pr->pr_mtx);
3357 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3358 				error = EAFNOSUPPORT;
3359 			mtx_unlock(&pr->pr_mtx);
3360 		}
3361 		break;
3362 #endif
3363 #ifdef INET6
3364 	case AF_INET6:
3365 		if (pr->pr_flags & PR_IP6)
3366 		{
3367 			mtx_lock(&pr->pr_mtx);
3368 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3369 				error = EAFNOSUPPORT;
3370 			mtx_unlock(&pr->pr_mtx);
3371 		}
3372 		break;
3373 #endif
3374 	case AF_LOCAL:
3375 	case AF_ROUTE:
3376 		break;
3377 	default:
3378 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3379 			error = EAFNOSUPPORT;
3380 	}
3381 	return (error);
3382 }
3383 
3384 /*
3385  * Check if given address belongs to the jail referenced by cred (wrapper to
3386  * prison_check_ip[46]).
3387  *
3388  * Returns 0 if jail doesn't restrict the address family or if address belongs
3389  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3390  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3391  */
3392 int
3393 prison_if(struct ucred *cred, struct sockaddr *sa)
3394 {
3395 #ifdef INET
3396 	struct sockaddr_in *sai;
3397 #endif
3398 #ifdef INET6
3399 	struct sockaddr_in6 *sai6;
3400 #endif
3401 	int error;
3402 
3403 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3404 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3405 
3406 #ifdef VIMAGE
3407 	if (prison_owns_vnet(cred))
3408 		return (0);
3409 #endif
3410 
3411 	error = 0;
3412 	switch (sa->sa_family)
3413 	{
3414 #ifdef INET
3415 	case AF_INET:
3416 		sai = (struct sockaddr_in *)sa;
3417 		error = prison_check_ip4(cred, &sai->sin_addr);
3418 		break;
3419 #endif
3420 #ifdef INET6
3421 	case AF_INET6:
3422 		sai6 = (struct sockaddr_in6 *)sa;
3423 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3424 		break;
3425 #endif
3426 	default:
3427 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3428 			error = EAFNOSUPPORT;
3429 	}
3430 	return (error);
3431 }
3432 
3433 /*
3434  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3435  */
3436 int
3437 prison_check(struct ucred *cred1, struct ucred *cred2)
3438 {
3439 
3440 	return ((cred1->cr_prison == cred2->cr_prison ||
3441 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3442 }
3443 
3444 /*
3445  * Return 1 if p2 is a child of p1, otherwise 0.
3446  */
3447 int
3448 prison_ischild(struct prison *pr1, struct prison *pr2)
3449 {
3450 
3451 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3452 		if (pr1 == pr2)
3453 			return (1);
3454 	return (0);
3455 }
3456 
3457 /*
3458  * Return 1 if the passed credential is in a jail, otherwise 0.
3459  */
3460 int
3461 jailed(struct ucred *cred)
3462 {
3463 
3464 	return (cred->cr_prison != &prison0);
3465 }
3466 
3467 /*
3468  * Return 1 if the passed credential is in a jail and that jail does not
3469  * have its own virtual network stack, otherwise 0.
3470  */
3471 int
3472 jailed_without_vnet(struct ucred *cred)
3473 {
3474 
3475 	if (!jailed(cred))
3476 		return (0);
3477 #ifdef VIMAGE
3478 	if (prison_owns_vnet(cred))
3479 		return (0);
3480 #endif
3481 
3482 	return (1);
3483 }
3484 
3485 /*
3486  * Return the correct hostname (domainname, et al) for the passed credential.
3487  */
3488 void
3489 getcredhostname(struct ucred *cred, char *buf, size_t size)
3490 {
3491 	struct prison *pr;
3492 
3493 	/*
3494 	 * A NULL credential can be used to shortcut to the physical
3495 	 * system's hostname.
3496 	 */
3497 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3498 	mtx_lock(&pr->pr_mtx);
3499 	strlcpy(buf, pr->pr_hostname, size);
3500 	mtx_unlock(&pr->pr_mtx);
3501 }
3502 
3503 void
3504 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3505 {
3506 
3507 	mtx_lock(&cred->cr_prison->pr_mtx);
3508 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3509 	mtx_unlock(&cred->cr_prison->pr_mtx);
3510 }
3511 
3512 void
3513 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3514 {
3515 
3516 	mtx_lock(&cred->cr_prison->pr_mtx);
3517 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3518 	mtx_unlock(&cred->cr_prison->pr_mtx);
3519 }
3520 
3521 void
3522 getcredhostid(struct ucred *cred, unsigned long *hostid)
3523 {
3524 
3525 	mtx_lock(&cred->cr_prison->pr_mtx);
3526 	*hostid = cred->cr_prison->pr_hostid;
3527 	mtx_unlock(&cred->cr_prison->pr_mtx);
3528 }
3529 
3530 #ifdef VIMAGE
3531 /*
3532  * Determine whether the prison represented by cred owns
3533  * its vnet rather than having it inherited.
3534  *
3535  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3536  */
3537 int
3538 prison_owns_vnet(struct ucred *cred)
3539 {
3540 
3541 	/*
3542 	 * vnets cannot be added/removed after jail creation,
3543 	 * so no need to lock here.
3544 	 */
3545 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3546 }
3547 #endif
3548 
3549 /*
3550  * Determine whether the subject represented by cred can "see"
3551  * status of a mount point.
3552  * Returns: 0 for permitted, ENOENT otherwise.
3553  * XXX: This function should be called cr_canseemount() and should be
3554  *      placed in kern_prot.c.
3555  */
3556 int
3557 prison_canseemount(struct ucred *cred, struct mount *mp)
3558 {
3559 	struct prison *pr;
3560 	struct statfs *sp;
3561 	size_t len;
3562 
3563 	pr = cred->cr_prison;
3564 	if (pr->pr_enforce_statfs == 0)
3565 		return (0);
3566 	if (pr->pr_root->v_mount == mp)
3567 		return (0);
3568 	if (pr->pr_enforce_statfs == 2)
3569 		return (ENOENT);
3570 	/*
3571 	 * If jail's chroot directory is set to "/" we should be able to see
3572 	 * all mount-points from inside a jail.
3573 	 * This is ugly check, but this is the only situation when jail's
3574 	 * directory ends with '/'.
3575 	 */
3576 	if (strcmp(pr->pr_path, "/") == 0)
3577 		return (0);
3578 	len = strlen(pr->pr_path);
3579 	sp = &mp->mnt_stat;
3580 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3581 		return (ENOENT);
3582 	/*
3583 	 * Be sure that we don't have situation where jail's root directory
3584 	 * is "/some/path" and mount point is "/some/pathpath".
3585 	 */
3586 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3587 		return (ENOENT);
3588 	return (0);
3589 }
3590 
3591 void
3592 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3593 {
3594 	char jpath[MAXPATHLEN];
3595 	struct prison *pr;
3596 	size_t len;
3597 
3598 	pr = cred->cr_prison;
3599 	if (pr->pr_enforce_statfs == 0)
3600 		return;
3601 	if (prison_canseemount(cred, mp) != 0) {
3602 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3603 		strlcpy(sp->f_mntonname, "[restricted]",
3604 		    sizeof(sp->f_mntonname));
3605 		return;
3606 	}
3607 	if (pr->pr_root->v_mount == mp) {
3608 		/*
3609 		 * Clear current buffer data, so we are sure nothing from
3610 		 * the valid path left there.
3611 		 */
3612 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3613 		*sp->f_mntonname = '/';
3614 		return;
3615 	}
3616 	/*
3617 	 * If jail's chroot directory is set to "/" we should be able to see
3618 	 * all mount-points from inside a jail.
3619 	 */
3620 	if (strcmp(pr->pr_path, "/") == 0)
3621 		return;
3622 	len = strlen(pr->pr_path);
3623 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3624 	/*
3625 	 * Clear current buffer data, so we are sure nothing from
3626 	 * the valid path left there.
3627 	 */
3628 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3629 	if (*jpath == '\0') {
3630 		/* Should never happen. */
3631 		*sp->f_mntonname = '/';
3632 	} else {
3633 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3634 	}
3635 }
3636 
3637 /*
3638  * Check with permission for a specific privilege is granted within jail.  We
3639  * have a specific list of accepted privileges; the rest are denied.
3640  */
3641 int
3642 prison_priv_check(struct ucred *cred, int priv)
3643 {
3644 
3645 	if (!jailed(cred))
3646 		return (0);
3647 
3648 #ifdef VIMAGE
3649 	/*
3650 	 * Privileges specific to prisons with a virtual network stack.
3651 	 * There might be a duplicate entry here in case the privilege
3652 	 * is only granted conditionally in the legacy jail case.
3653 	 */
3654 	switch (priv) {
3655 #ifdef notyet
3656 		/*
3657 		 * NFS-specific privileges.
3658 		 */
3659 	case PRIV_NFS_DAEMON:
3660 	case PRIV_NFS_LOCKD:
3661 #endif
3662 		/*
3663 		 * Network stack privileges.
3664 		 */
3665 	case PRIV_NET_BRIDGE:
3666 	case PRIV_NET_GRE:
3667 	case PRIV_NET_BPF:
3668 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3669 	case PRIV_NET_ROUTE:
3670 	case PRIV_NET_TAP:
3671 	case PRIV_NET_SETIFMTU:
3672 	case PRIV_NET_SETIFFLAGS:
3673 	case PRIV_NET_SETIFCAP:
3674 	case PRIV_NET_SETIFDESCR:
3675 	case PRIV_NET_SETIFNAME	:
3676 	case PRIV_NET_SETIFMETRIC:
3677 	case PRIV_NET_SETIFPHYS:
3678 	case PRIV_NET_SETIFMAC:
3679 	case PRIV_NET_ADDMULTI:
3680 	case PRIV_NET_DELMULTI:
3681 	case PRIV_NET_HWIOCTL:
3682 	case PRIV_NET_SETLLADDR:
3683 	case PRIV_NET_ADDIFGROUP:
3684 	case PRIV_NET_DELIFGROUP:
3685 	case PRIV_NET_IFCREATE:
3686 	case PRIV_NET_IFDESTROY:
3687 	case PRIV_NET_ADDIFADDR:
3688 	case PRIV_NET_DELIFADDR:
3689 	case PRIV_NET_LAGG:
3690 	case PRIV_NET_GIF:
3691 	case PRIV_NET_SETIFVNET:
3692 	case PRIV_NET_SETIFFIB:
3693 
3694 		/*
3695 		 * 802.11-related privileges.
3696 		 */
3697 	case PRIV_NET80211_GETKEY:
3698 #ifdef notyet
3699 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3700 #endif
3701 
3702 #ifdef notyet
3703 		/*
3704 		 * AppleTalk privileges.
3705 		 */
3706 	case PRIV_NETATALK_RESERVEDPORT:
3707 
3708 		/*
3709 		 * ATM privileges.
3710 		 */
3711 	case PRIV_NETATM_CFG:
3712 	case PRIV_NETATM_ADD:
3713 	case PRIV_NETATM_DEL:
3714 	case PRIV_NETATM_SET:
3715 
3716 		/*
3717 		 * Bluetooth privileges.
3718 		 */
3719 	case PRIV_NETBLUETOOTH_RAW:
3720 #endif
3721 
3722 		/*
3723 		 * Netgraph and netgraph module privileges.
3724 		 */
3725 	case PRIV_NETGRAPH_CONTROL:
3726 #ifdef notyet
3727 	case PRIV_NETGRAPH_TTY:
3728 #endif
3729 
3730 		/*
3731 		 * IPv4 and IPv6 privileges.
3732 		 */
3733 	case PRIV_NETINET_IPFW:
3734 	case PRIV_NETINET_DIVERT:
3735 	case PRIV_NETINET_PF:
3736 	case PRIV_NETINET_DUMMYNET:
3737 	case PRIV_NETINET_CARP:
3738 	case PRIV_NETINET_MROUTE:
3739 	case PRIV_NETINET_RAW:
3740 	case PRIV_NETINET_ADDRCTRL6:
3741 	case PRIV_NETINET_ND6:
3742 	case PRIV_NETINET_SCOPE6:
3743 	case PRIV_NETINET_ALIFETIME6:
3744 	case PRIV_NETINET_IPSEC:
3745 	case PRIV_NETINET_BINDANY:
3746 
3747 #ifdef notyet
3748 		/*
3749 		 * IPX/SPX privileges.
3750 		 */
3751 	case PRIV_NETIPX_RESERVEDPORT:
3752 	case PRIV_NETIPX_RAW:
3753 
3754 		/*
3755 		 * NCP privileges.
3756 		 */
3757 	case PRIV_NETNCP:
3758 
3759 		/*
3760 		 * SMB privileges.
3761 		 */
3762 	case PRIV_NETSMB:
3763 #endif
3764 
3765 	/*
3766 	 * No default: or deny here.
3767 	 * In case of no permit fall through to next switch().
3768 	 */
3769 		if (cred->cr_prison->pr_flags & PR_VNET)
3770 			return (0);
3771 	}
3772 #endif /* VIMAGE */
3773 
3774 	switch (priv) {
3775 
3776 		/*
3777 		 * Allow ktrace privileges for root in jail.
3778 		 */
3779 	case PRIV_KTRACE:
3780 
3781 #if 0
3782 		/*
3783 		 * Allow jailed processes to configure audit identity and
3784 		 * submit audit records (login, etc).  In the future we may
3785 		 * want to further refine the relationship between audit and
3786 		 * jail.
3787 		 */
3788 	case PRIV_AUDIT_GETAUDIT:
3789 	case PRIV_AUDIT_SETAUDIT:
3790 	case PRIV_AUDIT_SUBMIT:
3791 #endif
3792 
3793 		/*
3794 		 * Allow jailed processes to manipulate process UNIX
3795 		 * credentials in any way they see fit.
3796 		 */
3797 	case PRIV_CRED_SETUID:
3798 	case PRIV_CRED_SETEUID:
3799 	case PRIV_CRED_SETGID:
3800 	case PRIV_CRED_SETEGID:
3801 	case PRIV_CRED_SETGROUPS:
3802 	case PRIV_CRED_SETREUID:
3803 	case PRIV_CRED_SETREGID:
3804 	case PRIV_CRED_SETRESUID:
3805 	case PRIV_CRED_SETRESGID:
3806 
3807 		/*
3808 		 * Jail implements visibility constraints already, so allow
3809 		 * jailed root to override uid/gid-based constraints.
3810 		 */
3811 	case PRIV_SEEOTHERGIDS:
3812 	case PRIV_SEEOTHERUIDS:
3813 
3814 		/*
3815 		 * Jail implements inter-process debugging limits already, so
3816 		 * allow jailed root various debugging privileges.
3817 		 */
3818 	case PRIV_DEBUG_DIFFCRED:
3819 	case PRIV_DEBUG_SUGID:
3820 	case PRIV_DEBUG_UNPRIV:
3821 
3822 		/*
3823 		 * Allow jail to set various resource limits and login
3824 		 * properties, and for now, exceed process resource limits.
3825 		 */
3826 	case PRIV_PROC_LIMIT:
3827 	case PRIV_PROC_SETLOGIN:
3828 	case PRIV_PROC_SETRLIMIT:
3829 
3830 		/*
3831 		 * System V and POSIX IPC privileges are granted in jail.
3832 		 */
3833 	case PRIV_IPC_READ:
3834 	case PRIV_IPC_WRITE:
3835 	case PRIV_IPC_ADMIN:
3836 	case PRIV_IPC_MSGSIZE:
3837 	case PRIV_MQ_ADMIN:
3838 
3839 		/*
3840 		 * Jail operations within a jail work on child jails.
3841 		 */
3842 	case PRIV_JAIL_ATTACH:
3843 	case PRIV_JAIL_SET:
3844 	case PRIV_JAIL_REMOVE:
3845 
3846 		/*
3847 		 * Jail implements its own inter-process limits, so allow
3848 		 * root processes in jail to change scheduling on other
3849 		 * processes in the same jail.  Likewise for signalling.
3850 		 */
3851 	case PRIV_SCHED_DIFFCRED:
3852 	case PRIV_SCHED_CPUSET:
3853 	case PRIV_SIGNAL_DIFFCRED:
3854 	case PRIV_SIGNAL_SUGID:
3855 
3856 		/*
3857 		 * Allow jailed processes to write to sysctls marked as jail
3858 		 * writable.
3859 		 */
3860 	case PRIV_SYSCTL_WRITEJAIL:
3861 
3862 		/*
3863 		 * Allow root in jail to manage a variety of quota
3864 		 * properties.  These should likely be conditional on a
3865 		 * configuration option.
3866 		 */
3867 	case PRIV_VFS_GETQUOTA:
3868 	case PRIV_VFS_SETQUOTA:
3869 
3870 		/*
3871 		 * Since Jail relies on chroot() to implement file system
3872 		 * protections, grant many VFS privileges to root in jail.
3873 		 * Be careful to exclude mount-related and NFS-related
3874 		 * privileges.
3875 		 */
3876 	case PRIV_VFS_READ:
3877 	case PRIV_VFS_WRITE:
3878 	case PRIV_VFS_ADMIN:
3879 	case PRIV_VFS_EXEC:
3880 	case PRIV_VFS_LOOKUP:
3881 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3882 	case PRIV_VFS_CHFLAGS_DEV:
3883 	case PRIV_VFS_CHOWN:
3884 	case PRIV_VFS_CHROOT:
3885 	case PRIV_VFS_RETAINSUGID:
3886 	case PRIV_VFS_FCHROOT:
3887 	case PRIV_VFS_LINK:
3888 	case PRIV_VFS_SETGID:
3889 	case PRIV_VFS_STAT:
3890 	case PRIV_VFS_STICKYFILE:
3891 
3892 		/*
3893 		 * As in the non-jail case, non-root users are expected to be
3894 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3895 		 * exists in the jail and they have permission to access it).
3896 		 */
3897 	case PRIV_KMEM_READ:
3898 		return (0);
3899 
3900 		/*
3901 		 * Depending on the global setting, allow privilege of
3902 		 * setting system flags.
3903 		 */
3904 	case PRIV_VFS_SYSFLAGS:
3905 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3906 			return (0);
3907 		else
3908 			return (EPERM);
3909 
3910 		/*
3911 		 * Depending on the global setting, allow privilege of
3912 		 * mounting/unmounting file systems.
3913 		 */
3914 	case PRIV_VFS_MOUNT:
3915 	case PRIV_VFS_UNMOUNT:
3916 	case PRIV_VFS_MOUNT_NONUSER:
3917 	case PRIV_VFS_MOUNT_OWNER:
3918 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3919 		    cred->cr_prison->pr_enforce_statfs < 2)
3920 			return (0);
3921 		else
3922 			return (EPERM);
3923 
3924 		/*
3925 		 * Allow jailed root to bind reserved ports and reuse in-use
3926 		 * ports.
3927 		 */
3928 	case PRIV_NETINET_RESERVEDPORT:
3929 	case PRIV_NETINET_REUSEPORT:
3930 		return (0);
3931 
3932 		/*
3933 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3934 		 */
3935 	case PRIV_NETINET_SETHDROPTS:
3936 		return (0);
3937 
3938 		/*
3939 		 * Conditionally allow creating raw sockets in jail.
3940 		 */
3941 	case PRIV_NETINET_RAW:
3942 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3943 			return (0);
3944 		else
3945 			return (EPERM);
3946 
3947 		/*
3948 		 * Since jail implements its own visibility limits on netstat
3949 		 * sysctls, allow getcred.  This allows identd to work in
3950 		 * jail.
3951 		 */
3952 	case PRIV_NETINET_GETCRED:
3953 		return (0);
3954 
3955 		/*
3956 		 * Allow access to /dev/io in a jail if the non-jailed admin
3957 		 * requests this and if /dev/io exists in the jail. This
3958 		 * allows Xorg to probe a card.
3959 		 */
3960 	case PRIV_IO:
3961 		if (cred->cr_prison->pr_allow & PR_ALLOW_KMEM)
3962 			return (0);
3963 		else
3964 			return (EPERM);
3965 
3966 		/*
3967 		 * Allow low level access to KMEM-like devices (e.g. to
3968 		 * allow Xorg to use DRI).
3969 		 */
3970 	case PRIV_KMEM_WRITE:
3971 		if (cred->cr_prison->pr_allow & PR_ALLOW_KMEM)
3972 			return (0);
3973 		else
3974 			return (EPERM);
3975 
3976 		/*
3977 		 * Allow jailed root to set loginclass.
3978 		 */
3979 	case PRIV_PROC_SETLOGINCLASS:
3980 		return (0);
3981 
3982 	default:
3983 		/*
3984 		 * In all remaining cases, deny the privilege request.  This
3985 		 * includes almost all network privileges, many system
3986 		 * configuration privileges.
3987 		 */
3988 		return (EPERM);
3989 	}
3990 }
3991 
3992 /*
3993  * Return the part of pr2's name that is relative to pr1, or the whole name
3994  * if it does not directly follow.
3995  */
3996 
3997 char *
3998 prison_name(struct prison *pr1, struct prison *pr2)
3999 {
4000 	char *name;
4001 
4002 	/* Jails see themselves as "0" (if they see themselves at all). */
4003 	if (pr1 == pr2)
4004 		return "0";
4005 	name = pr2->pr_name;
4006 	if (prison_ischild(pr1, pr2)) {
4007 		/*
4008 		 * pr1 isn't locked (and allprison_lock may not be either)
4009 		 * so its length can't be counted on.  But the number of dots
4010 		 * can be counted on - and counted.
4011 		 */
4012 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4013 			name = strchr(name, '.') + 1;
4014 	}
4015 	return (name);
4016 }
4017 
4018 /*
4019  * Return the part of pr2's path that is relative to pr1, or the whole path
4020  * if it does not directly follow.
4021  */
4022 static char *
4023 prison_path(struct prison *pr1, struct prison *pr2)
4024 {
4025 	char *path1, *path2;
4026 	int len1;
4027 
4028 	path1 = pr1->pr_path;
4029 	path2 = pr2->pr_path;
4030 	if (!strcmp(path1, "/"))
4031 		return (path2);
4032 	len1 = strlen(path1);
4033 	if (strncmp(path1, path2, len1))
4034 		return (path2);
4035 	if (path2[len1] == '\0')
4036 		return "/";
4037 	if (path2[len1] == '/')
4038 		return (path2 + len1);
4039 	return (path2);
4040 }
4041 
4042 
4043 /*
4044  * Jail-related sysctls.
4045  */
4046 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4047     "Jails");
4048 
4049 static int
4050 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4051 {
4052 	struct xprison *xp;
4053 	struct prison *pr, *cpr;
4054 #ifdef INET
4055 	struct in_addr *ip4 = NULL;
4056 	int ip4s = 0;
4057 #endif
4058 #ifdef INET6
4059 	struct in6_addr *ip6 = NULL;
4060 	int ip6s = 0;
4061 #endif
4062 	int descend, error;
4063 
4064 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4065 	pr = req->td->td_ucred->cr_prison;
4066 	error = 0;
4067 	sx_slock(&allprison_lock);
4068 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4069 #if defined(INET) || defined(INET6)
4070  again:
4071 #endif
4072 		mtx_lock(&cpr->pr_mtx);
4073 #ifdef INET
4074 		if (cpr->pr_ip4s > 0) {
4075 			if (ip4s < cpr->pr_ip4s) {
4076 				ip4s = cpr->pr_ip4s;
4077 				mtx_unlock(&cpr->pr_mtx);
4078 				ip4 = realloc(ip4, ip4s *
4079 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4080 				goto again;
4081 			}
4082 			bcopy(cpr->pr_ip4, ip4,
4083 			    cpr->pr_ip4s * sizeof(struct in_addr));
4084 		}
4085 #endif
4086 #ifdef INET6
4087 		if (cpr->pr_ip6s > 0) {
4088 			if (ip6s < cpr->pr_ip6s) {
4089 				ip6s = cpr->pr_ip6s;
4090 				mtx_unlock(&cpr->pr_mtx);
4091 				ip6 = realloc(ip6, ip6s *
4092 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4093 				goto again;
4094 			}
4095 			bcopy(cpr->pr_ip6, ip6,
4096 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4097 		}
4098 #endif
4099 		if (cpr->pr_ref == 0) {
4100 			mtx_unlock(&cpr->pr_mtx);
4101 			continue;
4102 		}
4103 		bzero(xp, sizeof(*xp));
4104 		xp->pr_version = XPRISON_VERSION;
4105 		xp->pr_id = cpr->pr_id;
4106 		xp->pr_state = cpr->pr_uref > 0
4107 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4108 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4109 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4110 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4111 #ifdef INET
4112 		xp->pr_ip4s = cpr->pr_ip4s;
4113 #endif
4114 #ifdef INET6
4115 		xp->pr_ip6s = cpr->pr_ip6s;
4116 #endif
4117 		mtx_unlock(&cpr->pr_mtx);
4118 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4119 		if (error)
4120 			break;
4121 #ifdef INET
4122 		if (xp->pr_ip4s > 0) {
4123 			error = SYSCTL_OUT(req, ip4,
4124 			    xp->pr_ip4s * sizeof(struct in_addr));
4125 			if (error)
4126 				break;
4127 		}
4128 #endif
4129 #ifdef INET6
4130 		if (xp->pr_ip6s > 0) {
4131 			error = SYSCTL_OUT(req, ip6,
4132 			    xp->pr_ip6s * sizeof(struct in6_addr));
4133 			if (error)
4134 				break;
4135 		}
4136 #endif
4137 	}
4138 	sx_sunlock(&allprison_lock);
4139 	free(xp, M_TEMP);
4140 #ifdef INET
4141 	free(ip4, M_TEMP);
4142 #endif
4143 #ifdef INET6
4144 	free(ip6, M_TEMP);
4145 #endif
4146 	return (error);
4147 }
4148 
4149 SYSCTL_OID(_security_jail, OID_AUTO, list,
4150     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4151     sysctl_jail_list, "S", "List of active jails");
4152 
4153 static int
4154 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4155 {
4156 	int error, injail;
4157 
4158 	injail = jailed(req->td->td_ucred);
4159 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4160 
4161 	return (error);
4162 }
4163 
4164 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4165     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4166     sysctl_jail_jailed, "I", "Process in jail?");
4167 
4168 static int
4169 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4170 {
4171 	int error, havevnet;
4172 #ifdef VIMAGE
4173 	struct ucred *cred = req->td->td_ucred;
4174 
4175 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4176 #else
4177 	havevnet = 0;
4178 #endif
4179 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4180 
4181 	return (error);
4182 }
4183 
4184 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4185     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4186     sysctl_jail_vnet, "I", "Jail owns VNET?");
4187 
4188 #if defined(INET) || defined(INET6)
4189 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4190     &jail_max_af_ips, 0,
4191     "Number of IP addresses a jail may have at most per address family");
4192 #endif
4193 
4194 /*
4195  * Default parameters for jail(2) compatability.  For historical reasons,
4196  * the sysctl names have varying similarity to the parameter names.  Prisons
4197  * just see their own parameters, and can't change them.
4198  */
4199 static int
4200 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4201 {
4202 	struct prison *pr;
4203 	int allow, error, i;
4204 
4205 	pr = req->td->td_ucred->cr_prison;
4206 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4207 
4208 	/* Get the current flag value, and convert it to a boolean. */
4209 	i = (allow & arg2) ? 1 : 0;
4210 	if (arg1 != NULL)
4211 		i = !i;
4212 	error = sysctl_handle_int(oidp, &i, 0, req);
4213 	if (error || !req->newptr)
4214 		return (error);
4215 	i = i ? arg2 : 0;
4216 	if (arg1 != NULL)
4217 		i ^= arg2;
4218 	/*
4219 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4220 	 * for writing.
4221 	 */
4222 	mtx_lock(&prison0.pr_mtx);
4223 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4224 	mtx_unlock(&prison0.pr_mtx);
4225 	return (0);
4226 }
4227 
4228 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4229     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4230     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4231     "Processes in jail can set their hostnames");
4232 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4233     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4234     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4235     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4236 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4237     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4238     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4239     "Processes in jail can use System V IPC primitives");
4240 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4241     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4242     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4243     "Prison root can create raw sockets");
4244 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4245     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4246     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4247     "Processes in jail can alter system file flags");
4248 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4249     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4250     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4251     "Processes in jail can mount/unmount jail-friendly file systems");
4252 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4253     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4254     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4255     "Processes in jail can mount the devfs file system");
4256 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4257     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4258     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4259     "Processes in jail can mount the nullfs file system");
4260 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4261     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4262     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4263     "Processes in jail can mount the procfs file system");
4264 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4265     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4266     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4267     "Processes in jail can mount the tmpfs file system");
4268 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4269     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4270     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4271     "Processes in jail can mount the zfs file system");
4272 
4273 static int
4274 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4275 {
4276 	struct prison *pr;
4277 	int level, error;
4278 
4279 	pr = req->td->td_ucred->cr_prison;
4280 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4281 	error = sysctl_handle_int(oidp, &level, 0, req);
4282 	if (error || !req->newptr)
4283 		return (error);
4284 	*(int *)arg1 = level;
4285 	return (0);
4286 }
4287 
4288 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4289     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4290     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4291     sysctl_jail_default_level, "I",
4292     "Processes in jail cannot see all mounted file systems");
4293 
4294 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4295     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4296     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4297     sysctl_jail_default_level, "I",
4298     "Ruleset for the devfs filesystem in jail");
4299 
4300 /*
4301  * Nodes to describe jail parameters.  Maximum length of string parameters
4302  * is returned in the string itself, and the other parameters exist merely
4303  * to make themselves and their types known.
4304  */
4305 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4306     "Jail parameters");
4307 
4308 int
4309 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4310 {
4311 	int i;
4312 	long l;
4313 	size_t s;
4314 	char numbuf[12];
4315 
4316 	switch (oidp->oid_kind & CTLTYPE)
4317 	{
4318 	case CTLTYPE_LONG:
4319 	case CTLTYPE_ULONG:
4320 		l = 0;
4321 #ifdef SCTL_MASK32
4322 		if (!(req->flags & SCTL_MASK32))
4323 #endif
4324 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4325 	case CTLTYPE_INT:
4326 	case CTLTYPE_UINT:
4327 		i = 0;
4328 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4329 	case CTLTYPE_STRING:
4330 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4331 		return
4332 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4333 	case CTLTYPE_STRUCT:
4334 		s = (size_t)arg2;
4335 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4336 	}
4337 	return (0);
4338 }
4339 
4340 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4341 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4342 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4343 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4344 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4345     "I", "Jail secure level");
4346 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4347     "I", "Jail cannot see all mounted file systems");
4348 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4349     "I", "Ruleset for in-jail devfs mounts");
4350 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4351     "B", "Jail persistence");
4352 #ifdef VIMAGE
4353 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4354     "E,jailsys", "Virtual network stack");
4355 #endif
4356 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4357     "B", "Jail is in the process of shutting down");
4358 
4359 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4360 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4361     "I", "Current number of child jails");
4362 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4363     "I", "Maximum number of child jails");
4364 
4365 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4366 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4367     "Jail hostname");
4368 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4369     "Jail NIS domainname");
4370 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4371     "Jail host UUID");
4372 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4373     "LU", "Jail host ID");
4374 
4375 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4376 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4377 
4378 #ifdef INET
4379 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4380     "Jail IPv4 address virtualization");
4381 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4382     "S,in_addr,a", "Jail IPv4 addresses");
4383 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4384     "B", "Do (not) use IPv4 source address selection rather than the "
4385     "primary jail IPv4 address.");
4386 #endif
4387 #ifdef INET6
4388 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4389     "Jail IPv6 address virtualization");
4390 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4391     "S,in6_addr,a", "Jail IPv6 addresses");
4392 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4393     "B", "Do (not) use IPv6 source address selection rather than the "
4394     "primary jail IPv6 address.");
4395 #endif
4396 
4397 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4398 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4399     "B", "Jail may set hostname");
4400 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4401     "B", "Jail may use SYSV IPC");
4402 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4403     "B", "Jail may create raw sockets");
4404 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4405     "B", "Jail may alter system file flags");
4406 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4407     "B", "Jail may set file quotas");
4408 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4409     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4410 SYSCTL_JAIL_PARAM(_allow, kmem, CTLTYPE_INT | CTLFLAG_RW,
4411     "B", "Jail may access kmem-like devices (io, dri) if they exist");
4412 
4413 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4414 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4415     "B", "Jail may mount/unmount jail-friendly file systems in general");
4416 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4417     "B", "Jail may mount the devfs file system");
4418 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4419     "B", "Jail may mount the nullfs file system");
4420 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4421     "B", "Jail may mount the procfs file system");
4422 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4423     "B", "Jail may mount the tmpfs file system");
4424 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4425     "B", "Jail may mount the zfs file system");
4426 
4427 void
4428 prison_racct_foreach(void (*callback)(struct racct *racct,
4429     void *arg2, void *arg3), void *arg2, void *arg3)
4430 {
4431 	struct prison_racct *prr;
4432 
4433 	sx_slock(&allprison_lock);
4434 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4435 		(callback)(prr->prr_racct, arg2, arg3);
4436 	sx_sunlock(&allprison_lock);
4437 }
4438 
4439 static struct prison_racct *
4440 prison_racct_find_locked(const char *name)
4441 {
4442 	struct prison_racct *prr;
4443 
4444 	sx_assert(&allprison_lock, SA_XLOCKED);
4445 
4446 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4447 		return (NULL);
4448 
4449 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4450 		if (strcmp(name, prr->prr_name) != 0)
4451 			continue;
4452 
4453 		/* Found prison_racct with a matching name? */
4454 		prison_racct_hold(prr);
4455 		return (prr);
4456 	}
4457 
4458 	/* Add new prison_racct. */
4459 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4460 	racct_create(&prr->prr_racct);
4461 
4462 	strcpy(prr->prr_name, name);
4463 	refcount_init(&prr->prr_refcount, 1);
4464 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4465 
4466 	return (prr);
4467 }
4468 
4469 struct prison_racct *
4470 prison_racct_find(const char *name)
4471 {
4472 	struct prison_racct *prr;
4473 
4474 	sx_xlock(&allprison_lock);
4475 	prr = prison_racct_find_locked(name);
4476 	sx_xunlock(&allprison_lock);
4477 	return (prr);
4478 }
4479 
4480 void
4481 prison_racct_hold(struct prison_racct *prr)
4482 {
4483 
4484 	refcount_acquire(&prr->prr_refcount);
4485 }
4486 
4487 static void
4488 prison_racct_free_locked(struct prison_racct *prr)
4489 {
4490 
4491 	sx_assert(&allprison_lock, SA_XLOCKED);
4492 
4493 	if (refcount_release(&prr->prr_refcount)) {
4494 		racct_destroy(&prr->prr_racct);
4495 		LIST_REMOVE(prr, prr_next);
4496 		free(prr, M_PRISON_RACCT);
4497 	}
4498 }
4499 
4500 void
4501 prison_racct_free(struct prison_racct *prr)
4502 {
4503 	int old;
4504 
4505 	sx_assert(&allprison_lock, SA_UNLOCKED);
4506 
4507 	old = prr->prr_refcount;
4508 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4509 		return;
4510 
4511 	sx_xlock(&allprison_lock);
4512 	prison_racct_free_locked(prr);
4513 	sx_xunlock(&allprison_lock);
4514 }
4515 
4516 #ifdef RACCT
4517 static void
4518 prison_racct_attach(struct prison *pr)
4519 {
4520 	struct prison_racct *prr;
4521 
4522 	sx_assert(&allprison_lock, SA_XLOCKED);
4523 
4524 	prr = prison_racct_find_locked(pr->pr_name);
4525 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4526 
4527 	pr->pr_prison_racct = prr;
4528 }
4529 
4530 /*
4531  * Handle jail renaming.  From the racct point of view, renaming means
4532  * moving from one prison_racct to another.
4533  */
4534 static void
4535 prison_racct_modify(struct prison *pr)
4536 {
4537 	struct proc *p;
4538 	struct ucred *cred;
4539 	struct prison_racct *oldprr;
4540 
4541 	sx_slock(&allproc_lock);
4542 	sx_xlock(&allprison_lock);
4543 
4544 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4545 		sx_xunlock(&allprison_lock);
4546 		sx_sunlock(&allproc_lock);
4547 		return;
4548 	}
4549 
4550 	oldprr = pr->pr_prison_racct;
4551 	pr->pr_prison_racct = NULL;
4552 
4553 	prison_racct_attach(pr);
4554 
4555 	/*
4556 	 * Move resource utilisation records.
4557 	 */
4558 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4559 
4560 	/*
4561 	 * Force rctl to reattach rules to processes.
4562 	 */
4563 	FOREACH_PROC_IN_SYSTEM(p) {
4564 		PROC_LOCK(p);
4565 		cred = crhold(p->p_ucred);
4566 		PROC_UNLOCK(p);
4567 		racct_proc_ucred_changed(p, cred, cred);
4568 		crfree(cred);
4569 	}
4570 
4571 	sx_sunlock(&allproc_lock);
4572 	prison_racct_free_locked(oldprr);
4573 	sx_xunlock(&allprison_lock);
4574 }
4575 
4576 static void
4577 prison_racct_detach(struct prison *pr)
4578 {
4579 
4580 	sx_assert(&allprison_lock, SA_UNLOCKED);
4581 
4582 	if (pr->pr_prison_racct == NULL)
4583 		return;
4584 	prison_racct_free(pr->pr_prison_racct);
4585 	pr->pr_prison_racct = NULL;
4586 }
4587 #endif /* RACCT */
4588 
4589 #ifdef DDB
4590 
4591 static void
4592 db_show_prison(struct prison *pr)
4593 {
4594 	int fi;
4595 #if defined(INET) || defined(INET6)
4596 	int ii;
4597 #endif
4598 	unsigned jsf;
4599 #ifdef INET6
4600 	char ip6buf[INET6_ADDRSTRLEN];
4601 #endif
4602 
4603 	db_printf("prison %p:\n", pr);
4604 	db_printf(" jid             = %d\n", pr->pr_id);
4605 	db_printf(" name            = %s\n", pr->pr_name);
4606 	db_printf(" parent          = %p\n", pr->pr_parent);
4607 	db_printf(" ref             = %d\n", pr->pr_ref);
4608 	db_printf(" uref            = %d\n", pr->pr_uref);
4609 	db_printf(" path            = %s\n", pr->pr_path);
4610 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4611 	    ? pr->pr_cpuset->cs_id : -1);
4612 #ifdef VIMAGE
4613 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4614 #endif
4615 	db_printf(" root            = %p\n", pr->pr_root);
4616 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4617 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4618 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4619 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4620 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4621 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4622 	db_printf(" flags           = 0x%x", pr->pr_flags);
4623 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4624 	    fi++)
4625 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4626 			db_printf(" %s", pr_flag_names[fi]);
4627 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4628 	    fi++) {
4629 		jsf = pr->pr_flags &
4630 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4631 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4632 		    pr_flag_jailsys[fi].disable &&
4633 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4634 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4635 		    : "inherit");
4636 	}
4637 	db_printf(" allow           = 0x%x", pr->pr_allow);
4638 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4639 	    fi++)
4640 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4641 			db_printf(" %s", pr_allow_names[fi]);
4642 	db_printf("\n");
4643 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4644 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4645 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4646 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4647 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4648 #ifdef INET
4649 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4650 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4651 		db_printf(" %s %s\n",
4652 		    ii == 0 ? "ip4.addr        =" : "                 ",
4653 		    inet_ntoa(pr->pr_ip4[ii]));
4654 #endif
4655 #ifdef INET6
4656 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4657 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4658 		db_printf(" %s %s\n",
4659 		    ii == 0 ? "ip6.addr        =" : "                 ",
4660 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4661 #endif
4662 }
4663 
4664 DB_SHOW_COMMAND(prison, db_show_prison_command)
4665 {
4666 	struct prison *pr;
4667 
4668 	if (!have_addr) {
4669 		/*
4670 		 * Show all prisons in the list, and prison0 which is not
4671 		 * listed.
4672 		 */
4673 		db_show_prison(&prison0);
4674 		if (!db_pager_quit) {
4675 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4676 				db_show_prison(pr);
4677 				if (db_pager_quit)
4678 					break;
4679 			}
4680 		}
4681 		return;
4682 	}
4683 
4684 	if (addr == 0)
4685 		pr = &prison0;
4686 	else {
4687 		/* Look for a prison with the ID and with references. */
4688 		TAILQ_FOREACH(pr, &allprison, pr_list)
4689 			if (pr->pr_id == addr && pr->pr_ref > 0)
4690 				break;
4691 		if (pr == NULL)
4692 			/* Look again, without requiring a reference. */
4693 			TAILQ_FOREACH(pr, &allprison, pr_list)
4694 				if (pr->pr_id == addr)
4695 					break;
4696 		if (pr == NULL)
4697 			/* Assume address points to a valid prison. */
4698 			pr = (struct prison *)addr;
4699 	}
4700 	db_show_prison(pr);
4701 }
4702 
4703 #endif /* DDB */
4704