xref: /freebsd/sys/kern/kern_jail.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #ifdef INET6
72 #include <netinet6/in6_var.h>
73 #endif /* INET6 */
74 #endif /* DDB */
75 
76 #include <security/mac/mac_framework.h>
77 
78 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
79 
80 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
81 MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
82 
83 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
84 #ifdef INET
85 #ifdef INET6
86 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
87 #else
88 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
89 #endif
90 #else /* !INET */
91 #ifdef INET6
92 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
93 #else
94 #define	_PR_IP_SADDRSEL	0
95 #endif
96 #endif
97 
98 /* prison0 describes what is "real" about the system. */
99 struct prison prison0 = {
100 	.pr_id		= 0,
101 	.pr_name	= "0",
102 	.pr_ref		= 1,
103 	.pr_uref	= 1,
104 	.pr_path	= "/",
105 	.pr_securelevel	= -1,
106 	.pr_childmax	= JAIL_MAX,
107 	.pr_hostuuid	= DEFAULT_HOSTUUID,
108 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
109 #ifdef VIMAGE
110 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
111 #else
112 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
113 #endif
114 	.pr_allow	= PR_ALLOW_ALL,
115 };
116 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
117 
118 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
119 struct	sx allprison_lock;
120 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
121 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
122 LIST_HEAD(, prison_racct) allprison_racct;
123 int	lastprid = 0;
124 
125 static int do_jail_attach(struct thread *td, struct prison *pr);
126 static void prison_complete(void *context, int pending);
127 static void prison_deref(struct prison *pr, int flags);
128 static char *prison_path(struct prison *pr1, struct prison *pr2);
129 static void prison_remove_one(struct prison *pr);
130 #ifdef RACCT
131 static void prison_racct_attach(struct prison *pr);
132 static void prison_racct_detach(struct prison *pr);
133 #endif
134 #ifdef INET
135 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
136 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
137 #endif
138 #ifdef INET6
139 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
140 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
141 #endif
142 
143 /* Flags for prison_deref */
144 #define	PD_DEREF	0x01
145 #define	PD_DEUREF	0x02
146 #define	PD_LOCKED	0x04
147 #define	PD_LIST_SLOCKED	0x08
148 #define	PD_LIST_XLOCKED	0x10
149 
150 /*
151  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
152  * as we cannot figure out the size of a sparse array, or an array without a
153  * terminating entry.
154  */
155 static char *pr_flag_names[] = {
156 	[0] = "persist",
157 #ifdef INET
158 	[7] = "ip4.saddrsel",
159 #endif
160 #ifdef INET6
161 	[8] = "ip6.saddrsel",
162 #endif
163 };
164 const size_t pr_flag_names_size = sizeof(pr_flag_names);
165 
166 static char *pr_flag_nonames[] = {
167 	[0] = "nopersist",
168 #ifdef INET
169 	[7] = "ip4.nosaddrsel",
170 #endif
171 #ifdef INET6
172 	[8] = "ip6.nosaddrsel",
173 #endif
174 };
175 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
176 
177 struct jailsys_flags {
178 	const char	*name;
179 	unsigned	 disable;
180 	unsigned	 new;
181 } pr_flag_jailsys[] = {
182 	{ "host", 0, PR_HOST },
183 #ifdef VIMAGE
184 	{ "vnet", 0, PR_VNET },
185 #endif
186 #ifdef INET
187 	{ "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
188 #endif
189 #ifdef INET6
190 	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
191 #endif
192 };
193 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
194 
195 static char *pr_allow_names[] = {
196 	"allow.set_hostname",
197 	"allow.sysvipc",
198 	"allow.raw_sockets",
199 	"allow.chflags",
200 	"allow.mount",
201 	"allow.quotas",
202 	"allow.socket_af",
203 };
204 const size_t pr_allow_names_size = sizeof(pr_allow_names);
205 
206 static char *pr_allow_nonames[] = {
207 	"allow.noset_hostname",
208 	"allow.nosysvipc",
209 	"allow.noraw_sockets",
210 	"allow.nochflags",
211 	"allow.nomount",
212 	"allow.noquotas",
213 	"allow.nosocket_af",
214 };
215 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
216 
217 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
218 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
219 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
220 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
221 #if defined(INET) || defined(INET6)
222 static unsigned jail_max_af_ips = 255;
223 #endif
224 
225 #ifdef INET
226 static int
227 qcmp_v4(const void *ip1, const void *ip2)
228 {
229 	in_addr_t iaa, iab;
230 
231 	/*
232 	 * We need to compare in HBO here to get the list sorted as expected
233 	 * by the result of the code.  Sorting NBO addresses gives you
234 	 * interesting results.  If you do not understand, do not try.
235 	 */
236 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
237 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
238 
239 	/*
240 	 * Do not simply return the difference of the two numbers, the int is
241 	 * not wide enough.
242 	 */
243 	if (iaa > iab)
244 		return (1);
245 	else if (iaa < iab)
246 		return (-1);
247 	else
248 		return (0);
249 }
250 #endif
251 
252 #ifdef INET6
253 static int
254 qcmp_v6(const void *ip1, const void *ip2)
255 {
256 	const struct in6_addr *ia6a, *ia6b;
257 	int i, rc;
258 
259 	ia6a = (const struct in6_addr *)ip1;
260 	ia6b = (const struct in6_addr *)ip2;
261 
262 	rc = 0;
263 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
264 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
265 			rc = 1;
266 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
267 			rc = -1;
268 	}
269 	return (rc);
270 }
271 #endif
272 
273 /*
274  * struct jail_args {
275  *	struct jail *jail;
276  * };
277  */
278 int
279 jail(struct thread *td, struct jail_args *uap)
280 {
281 	uint32_t version;
282 	int error;
283 	struct jail j;
284 
285 	error = copyin(uap->jail, &version, sizeof(uint32_t));
286 	if (error)
287 		return (error);
288 
289 	switch (version) {
290 	case 0:
291 	{
292 		struct jail_v0 j0;
293 
294 		/* FreeBSD single IPv4 jails. */
295 		bzero(&j, sizeof(struct jail));
296 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
297 		if (error)
298 			return (error);
299 		j.version = j0.version;
300 		j.path = j0.path;
301 		j.hostname = j0.hostname;
302 		j.ip4s = j0.ip_number;
303 		break;
304 	}
305 
306 	case 1:
307 		/*
308 		 * Version 1 was used by multi-IPv4 jail implementations
309 		 * that never made it into the official kernel.
310 		 */
311 		return (EINVAL);
312 
313 	case 2:	/* JAIL_API_VERSION */
314 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
315 		error = copyin(uap->jail, &j, sizeof(struct jail));
316 		if (error)
317 			return (error);
318 		break;
319 
320 	default:
321 		/* Sci-Fi jails are not supported, sorry. */
322 		return (EINVAL);
323 	}
324 	return (kern_jail(td, &j));
325 }
326 
327 int
328 kern_jail(struct thread *td, struct jail *j)
329 {
330 	struct iovec optiov[2 * (4
331 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
332 #ifdef INET
333 			    + 1
334 #endif
335 #ifdef INET6
336 			    + 1
337 #endif
338 			    )];
339 	struct uio opt;
340 	char *u_path, *u_hostname, *u_name;
341 #ifdef INET
342 	uint32_t ip4s;
343 	struct in_addr *u_ip4;
344 #endif
345 #ifdef INET6
346 	struct in6_addr *u_ip6;
347 #endif
348 	size_t tmplen;
349 	int error, enforce_statfs, fi;
350 
351 	bzero(&optiov, sizeof(optiov));
352 	opt.uio_iov = optiov;
353 	opt.uio_iovcnt = 0;
354 	opt.uio_offset = -1;
355 	opt.uio_resid = -1;
356 	opt.uio_segflg = UIO_SYSSPACE;
357 	opt.uio_rw = UIO_READ;
358 	opt.uio_td = td;
359 
360 	/* Set permissions for top-level jails from sysctls. */
361 	if (!jailed(td->td_ucred)) {
362 		for (fi = 0; fi < sizeof(pr_allow_names) /
363 		     sizeof(pr_allow_names[0]); fi++) {
364 			optiov[opt.uio_iovcnt].iov_base =
365 			    (jail_default_allow & (1 << fi))
366 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
367 			optiov[opt.uio_iovcnt].iov_len =
368 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
369 			opt.uio_iovcnt += 2;
370 		}
371 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
372 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
373 		opt.uio_iovcnt++;
374 		enforce_statfs = jail_default_enforce_statfs;
375 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
376 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
377 		opt.uio_iovcnt++;
378 	}
379 
380 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
381 #ifdef INET
382 	ip4s = (j->version == 0) ? 1 : j->ip4s;
383 	if (ip4s > jail_max_af_ips)
384 		return (EINVAL);
385 	tmplen += ip4s * sizeof(struct in_addr);
386 #else
387 	if (j->ip4s > 0)
388 		return (EINVAL);
389 #endif
390 #ifdef INET6
391 	if (j->ip6s > jail_max_af_ips)
392 		return (EINVAL);
393 	tmplen += j->ip6s * sizeof(struct in6_addr);
394 #else
395 	if (j->ip6s > 0)
396 		return (EINVAL);
397 #endif
398 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
399 	u_hostname = u_path + MAXPATHLEN;
400 	u_name = u_hostname + MAXHOSTNAMELEN;
401 #ifdef INET
402 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
403 #endif
404 #ifdef INET6
405 #ifdef INET
406 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
407 #else
408 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
409 #endif
410 #endif
411 	optiov[opt.uio_iovcnt].iov_base = "path";
412 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
413 	opt.uio_iovcnt++;
414 	optiov[opt.uio_iovcnt].iov_base = u_path;
415 	error = copyinstr(j->path, u_path, MAXPATHLEN,
416 	    &optiov[opt.uio_iovcnt].iov_len);
417 	if (error) {
418 		free(u_path, M_TEMP);
419 		return (error);
420 	}
421 	opt.uio_iovcnt++;
422 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
423 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
424 	opt.uio_iovcnt++;
425 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
426 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
427 	    &optiov[opt.uio_iovcnt].iov_len);
428 	if (error) {
429 		free(u_path, M_TEMP);
430 		return (error);
431 	}
432 	opt.uio_iovcnt++;
433 	if (j->jailname != NULL) {
434 		optiov[opt.uio_iovcnt].iov_base = "name";
435 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
436 		opt.uio_iovcnt++;
437 		optiov[opt.uio_iovcnt].iov_base = u_name;
438 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
439 		    &optiov[opt.uio_iovcnt].iov_len);
440 		if (error) {
441 			free(u_path, M_TEMP);
442 			return (error);
443 		}
444 		opt.uio_iovcnt++;
445 	}
446 #ifdef INET
447 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
448 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
449 	opt.uio_iovcnt++;
450 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
451 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
452 	if (j->version == 0)
453 		u_ip4->s_addr = j->ip4s;
454 	else {
455 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
456 		if (error) {
457 			free(u_path, M_TEMP);
458 			return (error);
459 		}
460 	}
461 	opt.uio_iovcnt++;
462 #endif
463 #ifdef INET6
464 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
465 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
466 	opt.uio_iovcnt++;
467 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
468 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
469 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
470 	if (error) {
471 		free(u_path, M_TEMP);
472 		return (error);
473 	}
474 	opt.uio_iovcnt++;
475 #endif
476 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
477 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
478 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
479 	free(u_path, M_TEMP);
480 	return (error);
481 }
482 
483 
484 /*
485  * struct jail_set_args {
486  *	struct iovec *iovp;
487  *	unsigned int iovcnt;
488  *	int flags;
489  * };
490  */
491 int
492 jail_set(struct thread *td, struct jail_set_args *uap)
493 {
494 	struct uio *auio;
495 	int error;
496 
497 	/* Check that we have an even number of iovecs. */
498 	if (uap->iovcnt & 1)
499 		return (EINVAL);
500 
501 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
502 	if (error)
503 		return (error);
504 	error = kern_jail_set(td, auio, uap->flags);
505 	free(auio, M_IOV);
506 	return (error);
507 }
508 
509 int
510 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
511 {
512 	struct nameidata nd;
513 #ifdef INET
514 	struct in_addr *ip4;
515 #endif
516 #ifdef INET6
517 	struct in6_addr *ip6;
518 #endif
519 	struct vfsopt *opt;
520 	struct vfsoptlist *opts;
521 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
522 	struct vnode *root;
523 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
524 #if defined(INET) || defined(INET6)
525 	struct prison *tppr;
526 	void *op;
527 #endif
528 	unsigned long hid;
529 	size_t namelen, onamelen;
530 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
531 	int gotchildmax, gotenforce, gothid, gotslevel;
532 	int fi, jid, jsys, len, level;
533 	int childmax, slevel, vfslocked;
534 #if defined(INET) || defined(INET6)
535 	int ii, ij;
536 #endif
537 #ifdef INET
538 	int ip4s, redo_ip4;
539 #endif
540 #ifdef INET6
541 	int ip6s, redo_ip6;
542 #endif
543 	unsigned pr_flags, ch_flags;
544 	unsigned pr_allow, ch_allow, tallow;
545 	char numbuf[12];
546 
547 	error = priv_check(td, PRIV_JAIL_SET);
548 	if (!error && (flags & JAIL_ATTACH))
549 		error = priv_check(td, PRIV_JAIL_ATTACH);
550 	if (error)
551 		return (error);
552 	mypr = ppr = td->td_ucred->cr_prison;
553 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
554 		return (EPERM);
555 	if (flags & ~JAIL_SET_MASK)
556 		return (EINVAL);
557 
558 	/*
559 	 * Check all the parameters before committing to anything.  Not all
560 	 * errors can be caught early, but we may as well try.  Also, this
561 	 * takes care of some expensive stuff (path lookup) before getting
562 	 * the allprison lock.
563 	 *
564 	 * XXX Jails are not filesystems, and jail parameters are not mount
565 	 *     options.  But it makes more sense to re-use the vfsopt code
566 	 *     than duplicate it under a different name.
567 	 */
568 	error = vfs_buildopts(optuio, &opts);
569 	if (error)
570 		return (error);
571 #ifdef INET
572 	ip4 = NULL;
573 #endif
574 #ifdef INET6
575 	ip6 = NULL;
576 #endif
577 
578 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
579 	if (error == ENOENT)
580 		jid = 0;
581 	else if (error != 0)
582 		goto done_free;
583 
584 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
585 	if (error == ENOENT)
586 		gotslevel = 0;
587 	else if (error != 0)
588 		goto done_free;
589 	else
590 		gotslevel = 1;
591 
592 	error =
593 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
594 	if (error == ENOENT)
595 		gotchildmax = 0;
596 	else if (error != 0)
597 		goto done_free;
598 	else
599 		gotchildmax = 1;
600 
601 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
602 	if (error == ENOENT)
603 		gotenforce = 0;
604 	else if (error != 0)
605 		goto done_free;
606 	else if (enforce < 0 || enforce > 2) {
607 		error = EINVAL;
608 		goto done_free;
609 	} else
610 		gotenforce = 1;
611 
612 	pr_flags = ch_flags = 0;
613 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
614 	    fi++) {
615 		if (pr_flag_names[fi] == NULL)
616 			continue;
617 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
618 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
619 	}
620 	ch_flags |= pr_flags;
621 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
622 	    fi++) {
623 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
624 		    sizeof(jsys));
625 		if (error == ENOENT)
626 			continue;
627 		if (error != 0)
628 			goto done_free;
629 		switch (jsys) {
630 		case JAIL_SYS_DISABLE:
631 			if (!pr_flag_jailsys[fi].disable) {
632 				error = EINVAL;
633 				goto done_free;
634 			}
635 			pr_flags |= pr_flag_jailsys[fi].disable;
636 			break;
637 		case JAIL_SYS_NEW:
638 			pr_flags |= pr_flag_jailsys[fi].new;
639 			break;
640 		case JAIL_SYS_INHERIT:
641 			break;
642 		default:
643 			error = EINVAL;
644 			goto done_free;
645 		}
646 		ch_flags |=
647 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
648 	}
649 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
650 	    && !(pr_flags & PR_PERSIST)) {
651 		error = EINVAL;
652 		vfs_opterror(opts, "new jail must persist or attach");
653 		goto done_errmsg;
654 	}
655 #ifdef VIMAGE
656 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
657 		error = EINVAL;
658 		vfs_opterror(opts, "vnet cannot be changed after creation");
659 		goto done_errmsg;
660 	}
661 #endif
662 #ifdef INET
663 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
664 		error = EINVAL;
665 		vfs_opterror(opts, "ip4 cannot be changed after creation");
666 		goto done_errmsg;
667 	}
668 #endif
669 #ifdef INET6
670 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
671 		error = EINVAL;
672 		vfs_opterror(opts, "ip6 cannot be changed after creation");
673 		goto done_errmsg;
674 	}
675 #endif
676 
677 	pr_allow = ch_allow = 0;
678 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
679 	    fi++) {
680 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
681 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
682 	}
683 	ch_allow |= pr_allow;
684 
685 	error = vfs_getopt(opts, "name", (void **)&name, &len);
686 	if (error == ENOENT)
687 		name = NULL;
688 	else if (error != 0)
689 		goto done_free;
690 	else {
691 		if (len == 0 || name[len - 1] != '\0') {
692 			error = EINVAL;
693 			goto done_free;
694 		}
695 		if (len > MAXHOSTNAMELEN) {
696 			error = ENAMETOOLONG;
697 			goto done_free;
698 		}
699 	}
700 
701 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
702 	if (error == ENOENT)
703 		host = NULL;
704 	else if (error != 0)
705 		goto done_free;
706 	else {
707 		ch_flags |= PR_HOST;
708 		pr_flags |= PR_HOST;
709 		if (len == 0 || host[len - 1] != '\0') {
710 			error = EINVAL;
711 			goto done_free;
712 		}
713 		if (len > MAXHOSTNAMELEN) {
714 			error = ENAMETOOLONG;
715 			goto done_free;
716 		}
717 	}
718 
719 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
720 	if (error == ENOENT)
721 		domain = NULL;
722 	else if (error != 0)
723 		goto done_free;
724 	else {
725 		ch_flags |= PR_HOST;
726 		pr_flags |= PR_HOST;
727 		if (len == 0 || domain[len - 1] != '\0') {
728 			error = EINVAL;
729 			goto done_free;
730 		}
731 		if (len > MAXHOSTNAMELEN) {
732 			error = ENAMETOOLONG;
733 			goto done_free;
734 		}
735 	}
736 
737 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
738 	if (error == ENOENT)
739 		uuid = NULL;
740 	else if (error != 0)
741 		goto done_free;
742 	else {
743 		ch_flags |= PR_HOST;
744 		pr_flags |= PR_HOST;
745 		if (len == 0 || uuid[len - 1] != '\0') {
746 			error = EINVAL;
747 			goto done_free;
748 		}
749 		if (len > HOSTUUIDLEN) {
750 			error = ENAMETOOLONG;
751 			goto done_free;
752 		}
753 	}
754 
755 #ifdef COMPAT_FREEBSD32
756 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
757 		uint32_t hid32;
758 
759 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
760 		hid = hid32;
761 	} else
762 #endif
763 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
764 	if (error == ENOENT)
765 		gothid = 0;
766 	else if (error != 0)
767 		goto done_free;
768 	else {
769 		gothid = 1;
770 		ch_flags |= PR_HOST;
771 		pr_flags |= PR_HOST;
772 	}
773 
774 #ifdef INET
775 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
776 	if (error == ENOENT)
777 		ip4s = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
778 	else if (error != 0)
779 		goto done_free;
780 	else if (ip4s & (sizeof(*ip4) - 1)) {
781 		error = EINVAL;
782 		goto done_free;
783 	} else {
784 		ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
785 		if (ip4s == 0)
786 			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
787 		else {
788 			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
789 			ip4s /= sizeof(*ip4);
790 			if (ip4s > jail_max_af_ips) {
791 				error = EINVAL;
792 				vfs_opterror(opts, "too many IPv4 addresses");
793 				goto done_errmsg;
794 			}
795 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
796 			bcopy(op, ip4, ip4s * sizeof(*ip4));
797 			/*
798 			 * IP addresses are all sorted but ip[0] to preserve
799 			 * the primary IP address as given from userland.
800 			 * This special IP is used for unbound outgoing
801 			 * connections as well for "loopback" traffic in case
802 			 * source address selection cannot find any more fitting
803 			 * address to connect from.
804 			 */
805 			if (ip4s > 1)
806 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
807 			/*
808 			 * Check for duplicate addresses and do some simple
809 			 * zero and broadcast checks. If users give other bogus
810 			 * addresses it is their problem.
811 			 *
812 			 * We do not have to care about byte order for these
813 			 * checks so we will do them in NBO.
814 			 */
815 			for (ii = 0; ii < ip4s; ii++) {
816 				if (ip4[ii].s_addr == INADDR_ANY ||
817 				    ip4[ii].s_addr == INADDR_BROADCAST) {
818 					error = EINVAL;
819 					goto done_free;
820 				}
821 				if ((ii+1) < ip4s &&
822 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
823 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
824 					error = EINVAL;
825 					goto done_free;
826 				}
827 			}
828 		}
829 	}
830 #endif
831 
832 #ifdef INET6
833 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
834 	if (error == ENOENT)
835 		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
836 	else if (error != 0)
837 		goto done_free;
838 	else if (ip6s & (sizeof(*ip6) - 1)) {
839 		error = EINVAL;
840 		goto done_free;
841 	} else {
842 		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
843 		if (ip6s == 0)
844 			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
845 		else {
846 			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
847 			ip6s /= sizeof(*ip6);
848 			if (ip6s > jail_max_af_ips) {
849 				error = EINVAL;
850 				vfs_opterror(opts, "too many IPv6 addresses");
851 				goto done_errmsg;
852 			}
853 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
854 			bcopy(op, ip6, ip6s * sizeof(*ip6));
855 			if (ip6s > 1)
856 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
857 			for (ii = 0; ii < ip6s; ii++) {
858 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
859 					error = EINVAL;
860 					goto done_free;
861 				}
862 				if ((ii+1) < ip6s &&
863 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
864 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
865 				{
866 					error = EINVAL;
867 					goto done_free;
868 				}
869 			}
870 		}
871 	}
872 #endif
873 
874 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
875 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
876 		error = EINVAL;
877 		vfs_opterror(opts,
878 		    "vnet jails cannot have IP address restrictions");
879 		goto done_errmsg;
880 	}
881 #endif
882 
883 	root = NULL;
884 	error = vfs_getopt(opts, "path", (void **)&path, &len);
885 	if (error == ENOENT)
886 		path = NULL;
887 	else if (error != 0)
888 		goto done_free;
889 	else {
890 		if (flags & JAIL_UPDATE) {
891 			error = EINVAL;
892 			vfs_opterror(opts,
893 			    "path cannot be changed after creation");
894 			goto done_errmsg;
895 		}
896 		if (len == 0 || path[len - 1] != '\0') {
897 			error = EINVAL;
898 			goto done_free;
899 		}
900 		if (len < 2 || (len == 2 && path[0] == '/'))
901 			path = NULL;
902 		else {
903 			/* Leave room for a real-root full pathname. */
904 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
905 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
906 				error = ENAMETOOLONG;
907 				goto done_free;
908 			}
909 			NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW, UIO_SYSSPACE,
910 			    path, td);
911 			error = namei(&nd);
912 			if (error)
913 				goto done_free;
914 			vfslocked = NDHASGIANT(&nd);
915 			root = nd.ni_vp;
916 			NDFREE(&nd, NDF_ONLY_PNBUF);
917 			if (root->v_type != VDIR) {
918 				error = ENOTDIR;
919 				vrele(root);
920 				VFS_UNLOCK_GIANT(vfslocked);
921 				goto done_free;
922 			}
923 			VFS_UNLOCK_GIANT(vfslocked);
924 		}
925 	}
926 
927 	/*
928 	 * Grab the allprison lock before letting modules check their
929 	 * parameters.  Once we have it, do not let go so we'll have a
930 	 * consistent view of the OSD list.
931 	 */
932 	sx_xlock(&allprison_lock);
933 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
934 	if (error)
935 		goto done_unlock_list;
936 
937 	/* By now, all parameters should have been noted. */
938 	TAILQ_FOREACH(opt, opts, link) {
939 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
940 			error = EINVAL;
941 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
942 			goto done_unlock_list;
943 		}
944 	}
945 
946 	/*
947 	 * See if we are creating a new record or updating an existing one.
948 	 * This abuses the file error codes ENOENT and EEXIST.
949 	 */
950 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
951 	if (!cuflags) {
952 		error = EINVAL;
953 		vfs_opterror(opts, "no valid operation (create or update)");
954 		goto done_unlock_list;
955 	}
956 	pr = NULL;
957 	namelc = NULL;
958 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
959 		namelc = strrchr(name, '.');
960 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
961 		if (*p != '\0')
962 			jid = 0;
963 	}
964 	if (jid != 0) {
965 		/*
966 		 * See if a requested jid already exists.  There is an
967 		 * information leak here if the jid exists but is not within
968 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
969 		 * even though they cannot see the jail, and CREATE | UPDATE
970 		 * will return ENOENT which is not normally a valid error.
971 		 */
972 		if (jid < 0) {
973 			error = EINVAL;
974 			vfs_opterror(opts, "negative jid");
975 			goto done_unlock_list;
976 		}
977 		pr = prison_find(jid);
978 		if (pr != NULL) {
979 			ppr = pr->pr_parent;
980 			/* Create: jid must not exist. */
981 			if (cuflags == JAIL_CREATE) {
982 				mtx_unlock(&pr->pr_mtx);
983 				error = EEXIST;
984 				vfs_opterror(opts, "jail %d already exists",
985 				    jid);
986 				goto done_unlock_list;
987 			}
988 			if (!prison_ischild(mypr, pr)) {
989 				mtx_unlock(&pr->pr_mtx);
990 				pr = NULL;
991 			} else if (pr->pr_uref == 0) {
992 				if (!(flags & JAIL_DYING)) {
993 					mtx_unlock(&pr->pr_mtx);
994 					error = ENOENT;
995 					vfs_opterror(opts, "jail %d is dying",
996 					    jid);
997 					goto done_unlock_list;
998 				} else if ((flags & JAIL_ATTACH) ||
999 				    (pr_flags & PR_PERSIST)) {
1000 					/*
1001 					 * A dying jail might be resurrected
1002 					 * (via attach or persist), but first
1003 					 * it must determine if another jail
1004 					 * has claimed its name.  Accomplish
1005 					 * this by implicitly re-setting the
1006 					 * name.
1007 					 */
1008 					if (name == NULL)
1009 						name = prison_name(mypr, pr);
1010 				}
1011 			}
1012 		}
1013 		if (pr == NULL) {
1014 			/* Update: jid must exist. */
1015 			if (cuflags == JAIL_UPDATE) {
1016 				error = ENOENT;
1017 				vfs_opterror(opts, "jail %d not found", jid);
1018 				goto done_unlock_list;
1019 			}
1020 		}
1021 	}
1022 	/*
1023 	 * If the caller provided a name, look for a jail by that name.
1024 	 * This has different semantics for creates and updates keyed by jid
1025 	 * (where the name must not already exist in a different jail),
1026 	 * and updates keyed by the name itself (where the name must exist
1027 	 * because that is the jail being updated).
1028 	 */
1029 	if (name != NULL) {
1030 		namelc = strrchr(name, '.');
1031 		if (namelc == NULL)
1032 			namelc = name;
1033 		else {
1034 			/*
1035 			 * This is a hierarchical name.  Split it into the
1036 			 * parent and child names, and make sure the parent
1037 			 * exists or matches an already found jail.
1038 			 */
1039 			*namelc = '\0';
1040 			if (pr != NULL) {
1041 				if (strncmp(name, ppr->pr_name, namelc - name)
1042 				    || ppr->pr_name[namelc - name] != '\0') {
1043 					mtx_unlock(&pr->pr_mtx);
1044 					error = EINVAL;
1045 					vfs_opterror(opts,
1046 					    "cannot change jail's parent");
1047 					goto done_unlock_list;
1048 				}
1049 			} else {
1050 				ppr = prison_find_name(mypr, name);
1051 				if (ppr == NULL) {
1052 					error = ENOENT;
1053 					vfs_opterror(opts,
1054 					    "jail \"%s\" not found", name);
1055 					goto done_unlock_list;
1056 				}
1057 				mtx_unlock(&ppr->pr_mtx);
1058 			}
1059 			name = ++namelc;
1060 		}
1061 		if (name[0] != '\0') {
1062 			namelen =
1063 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1064  name_again:
1065 			deadpr = NULL;
1066 			FOREACH_PRISON_CHILD(ppr, tpr) {
1067 				if (tpr != pr && tpr->pr_ref > 0 &&
1068 				    !strcmp(tpr->pr_name + namelen, name)) {
1069 					if (pr == NULL &&
1070 					    cuflags != JAIL_CREATE) {
1071 						mtx_lock(&tpr->pr_mtx);
1072 						if (tpr->pr_ref > 0) {
1073 							/*
1074 							 * Use this jail
1075 							 * for updates.
1076 							 */
1077 							if (tpr->pr_uref > 0) {
1078 								pr = tpr;
1079 								break;
1080 							}
1081 							deadpr = tpr;
1082 						}
1083 						mtx_unlock(&tpr->pr_mtx);
1084 					} else if (tpr->pr_uref > 0) {
1085 						/*
1086 						 * Create, or update(jid):
1087 						 * name must not exist in an
1088 						 * active sibling jail.
1089 						 */
1090 						error = EEXIST;
1091 						if (pr != NULL)
1092 							mtx_unlock(&pr->pr_mtx);
1093 						vfs_opterror(opts,
1094 						   "jail \"%s\" already exists",
1095 						   name);
1096 						goto done_unlock_list;
1097 					}
1098 				}
1099 			}
1100 			/* If no active jail is found, use a dying one. */
1101 			if (deadpr != NULL && pr == NULL) {
1102 				if (flags & JAIL_DYING) {
1103 					mtx_lock(&deadpr->pr_mtx);
1104 					if (deadpr->pr_ref == 0) {
1105 						mtx_unlock(&deadpr->pr_mtx);
1106 						goto name_again;
1107 					}
1108 					pr = deadpr;
1109 				} else if (cuflags == JAIL_UPDATE) {
1110 					error = ENOENT;
1111 					vfs_opterror(opts,
1112 					    "jail \"%s\" is dying", name);
1113 					goto done_unlock_list;
1114 				}
1115 			}
1116 			/* Update: name must exist if no jid. */
1117 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1118 				error = ENOENT;
1119 				vfs_opterror(opts, "jail \"%s\" not found",
1120 				    name);
1121 				goto done_unlock_list;
1122 			}
1123 		}
1124 	}
1125 	/* Update: must provide a jid or name. */
1126 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1127 		error = ENOENT;
1128 		vfs_opterror(opts, "update specified no jail");
1129 		goto done_unlock_list;
1130 	}
1131 
1132 	/* If there's no prison to update, create a new one and link it in. */
1133 	if (pr == NULL) {
1134 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1135 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1136 				error = EPERM;
1137 				vfs_opterror(opts, "prison limit exceeded");
1138 				goto done_unlock_list;
1139 			}
1140 		created = 1;
1141 		mtx_lock(&ppr->pr_mtx);
1142 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1143 			mtx_unlock(&ppr->pr_mtx);
1144 			error = ENOENT;
1145 			vfs_opterror(opts, "parent jail went away!");
1146 			goto done_unlock_list;
1147 		}
1148 		ppr->pr_ref++;
1149 		ppr->pr_uref++;
1150 		mtx_unlock(&ppr->pr_mtx);
1151 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1152 		if (jid == 0) {
1153 			/* Find the next free jid. */
1154 			jid = lastprid + 1;
1155  findnext:
1156 			if (jid == JAIL_MAX)
1157 				jid = 1;
1158 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1159 				if (tpr->pr_id < jid)
1160 					continue;
1161 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1162 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1163 					break;
1164 				}
1165 				if (jid == lastprid) {
1166 					error = EAGAIN;
1167 					vfs_opterror(opts,
1168 					    "no available jail IDs");
1169 					free(pr, M_PRISON);
1170 					prison_deref(ppr, PD_DEREF |
1171 					    PD_DEUREF | PD_LIST_XLOCKED);
1172 					goto done_releroot;
1173 				}
1174 				jid++;
1175 				goto findnext;
1176 			}
1177 			lastprid = jid;
1178 		} else {
1179 			/*
1180 			 * The jail already has a jid (that did not yet exist),
1181 			 * so just find where to insert it.
1182 			 */
1183 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1184 				if (tpr->pr_id >= jid) {
1185 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1186 					break;
1187 				}
1188 		}
1189 		if (tpr == NULL)
1190 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1191 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1192 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1193 			tpr->pr_childcount++;
1194 
1195 		pr->pr_parent = ppr;
1196 		pr->pr_id = jid;
1197 
1198 		/* Set some default values, and inherit some from the parent. */
1199 		if (name == NULL)
1200 			name = "";
1201 		if (path == NULL) {
1202 			path = "/";
1203 			root = mypr->pr_root;
1204 			vref(root);
1205 		}
1206 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1207 		pr->pr_flags |= PR_HOST;
1208 #if defined(INET) || defined(INET6)
1209 #ifdef VIMAGE
1210 		if (!(pr_flags & PR_VNET))
1211 #endif
1212 		{
1213 #ifdef INET
1214 			if (!(ch_flags & PR_IP4_USER))
1215 				pr->pr_flags |=
1216 				    PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1217 			else if (!(pr_flags & PR_IP4_USER)) {
1218 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1219 				if (ppr->pr_ip4 != NULL) {
1220 					pr->pr_ip4s = ppr->pr_ip4s;
1221 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1222 					    sizeof(struct in_addr), M_PRISON,
1223 					    M_WAITOK);
1224 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1225 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1226 				}
1227 			}
1228 #endif
1229 #ifdef INET6
1230 			if (!(ch_flags & PR_IP6_USER))
1231 				pr->pr_flags |=
1232 				    PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1233 			else if (!(pr_flags & PR_IP6_USER)) {
1234 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1235 				if (ppr->pr_ip6 != NULL) {
1236 					pr->pr_ip6s = ppr->pr_ip6s;
1237 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1238 					    sizeof(struct in6_addr), M_PRISON,
1239 					    M_WAITOK);
1240 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1241 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1242 				}
1243 			}
1244 #endif
1245 		}
1246 #endif
1247 		/* Source address selection is always on by default. */
1248 		pr->pr_flags |= _PR_IP_SADDRSEL;
1249 
1250 		pr->pr_securelevel = ppr->pr_securelevel;
1251 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1252 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1253 
1254 		LIST_INIT(&pr->pr_children);
1255 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1256 
1257 #ifdef VIMAGE
1258 		/* Allocate a new vnet if specified. */
1259 		pr->pr_vnet = (pr_flags & PR_VNET)
1260 		    ? vnet_alloc() : ppr->pr_vnet;
1261 #endif
1262 		/*
1263 		 * Allocate a dedicated cpuset for each jail.
1264 		 * Unlike other initial settings, this may return an erorr.
1265 		 */
1266 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1267 		if (error) {
1268 			prison_deref(pr, PD_LIST_XLOCKED);
1269 			goto done_releroot;
1270 		}
1271 
1272 		mtx_lock(&pr->pr_mtx);
1273 		/*
1274 		 * New prisons do not yet have a reference, because we do not
1275 		 * want other to see the incomplete prison once the
1276 		 * allprison_lock is downgraded.
1277 		 */
1278 	} else {
1279 		created = 0;
1280 		/*
1281 		 * Grab a reference for existing prisons, to ensure they
1282 		 * continue to exist for the duration of the call.
1283 		 */
1284 		pr->pr_ref++;
1285 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1286 		if ((pr->pr_flags & PR_VNET) &&
1287 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1288 			error = EINVAL;
1289 			vfs_opterror(opts,
1290 			    "vnet jails cannot have IP address restrictions");
1291 			goto done_deref_locked;
1292 		}
1293 #endif
1294 #ifdef INET
1295 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1296 			error = EINVAL;
1297 			vfs_opterror(opts,
1298 			    "ip4 cannot be changed after creation");
1299 			goto done_deref_locked;
1300 		}
1301 #endif
1302 #ifdef INET6
1303 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1304 			error = EINVAL;
1305 			vfs_opterror(opts,
1306 			    "ip6 cannot be changed after creation");
1307 			goto done_deref_locked;
1308 		}
1309 #endif
1310 	}
1311 
1312 	/* Do final error checking before setting anything. */
1313 	if (gotslevel) {
1314 		if (slevel < ppr->pr_securelevel) {
1315 			error = EPERM;
1316 			goto done_deref_locked;
1317 		}
1318 	}
1319 	if (gotchildmax) {
1320 		if (childmax >= ppr->pr_childmax) {
1321 			error = EPERM;
1322 			goto done_deref_locked;
1323 		}
1324 	}
1325 	if (gotenforce) {
1326 		if (enforce < ppr->pr_enforce_statfs) {
1327 			error = EPERM;
1328 			goto done_deref_locked;
1329 		}
1330 	}
1331 #ifdef INET
1332 	if (ip4s > 0) {
1333 		if (ppr->pr_flags & PR_IP4) {
1334 			/*
1335 			 * Make sure the new set of IP addresses is a
1336 			 * subset of the parent's list.  Don't worry
1337 			 * about the parent being unlocked, as any
1338 			 * setting is done with allprison_lock held.
1339 			 */
1340 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1341 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1342 					break;
1343 			if (ij == ppr->pr_ip4s) {
1344 				error = EPERM;
1345 				goto done_deref_locked;
1346 			}
1347 			if (ip4s > 1) {
1348 				for (ii = ij = 1; ii < ip4s; ii++) {
1349 					if (ip4[ii].s_addr ==
1350 					    ppr->pr_ip4[0].s_addr)
1351 						continue;
1352 					for (; ij < ppr->pr_ip4s; ij++)
1353 						if (ip4[ii].s_addr ==
1354 						    ppr->pr_ip4[ij].s_addr)
1355 							break;
1356 					if (ij == ppr->pr_ip4s)
1357 						break;
1358 				}
1359 				if (ij == ppr->pr_ip4s) {
1360 					error = EPERM;
1361 					goto done_deref_locked;
1362 				}
1363 			}
1364 		}
1365 		/*
1366 		 * Check for conflicting IP addresses.  We permit them
1367 		 * if there is no more than one IP on each jail.  If
1368 		 * there is a duplicate on a jail with more than one
1369 		 * IP stop checking and return error.
1370 		 */
1371 		tppr = ppr;
1372 #ifdef VIMAGE
1373 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1374 			if (tppr->pr_flags & PR_VNET)
1375 				break;
1376 #endif
1377 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1378 			if (tpr == pr ||
1379 #ifdef VIMAGE
1380 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1381 #endif
1382 			    tpr->pr_uref == 0) {
1383 				descend = 0;
1384 				continue;
1385 			}
1386 			if (!(tpr->pr_flags & PR_IP4_USER))
1387 				continue;
1388 			descend = 0;
1389 			if (tpr->pr_ip4 == NULL ||
1390 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1391 				continue;
1392 			for (ii = 0; ii < ip4s; ii++) {
1393 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1394 					error = EADDRINUSE;
1395 					vfs_opterror(opts,
1396 					    "IPv4 addresses clash");
1397 					goto done_deref_locked;
1398 				}
1399 			}
1400 		}
1401 	}
1402 #endif
1403 #ifdef INET6
1404 	if (ip6s > 0) {
1405 		if (ppr->pr_flags & PR_IP6) {
1406 			/*
1407 			 * Make sure the new set of IP addresses is a
1408 			 * subset of the parent's list.
1409 			 */
1410 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1411 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1412 				    &ppr->pr_ip6[ij]))
1413 					break;
1414 			if (ij == ppr->pr_ip6s) {
1415 				error = EPERM;
1416 				goto done_deref_locked;
1417 			}
1418 			if (ip6s > 1) {
1419 				for (ii = ij = 1; ii < ip6s; ii++) {
1420 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1421 					     &ppr->pr_ip6[0]))
1422 						continue;
1423 					for (; ij < ppr->pr_ip6s; ij++)
1424 						if (IN6_ARE_ADDR_EQUAL(
1425 						    &ip6[ii], &ppr->pr_ip6[ij]))
1426 							break;
1427 					if (ij == ppr->pr_ip6s)
1428 						break;
1429 				}
1430 				if (ij == ppr->pr_ip6s) {
1431 					error = EPERM;
1432 					goto done_deref_locked;
1433 				}
1434 			}
1435 		}
1436 		/* Check for conflicting IP addresses. */
1437 		tppr = ppr;
1438 #ifdef VIMAGE
1439 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1440 			if (tppr->pr_flags & PR_VNET)
1441 				break;
1442 #endif
1443 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1444 			if (tpr == pr ||
1445 #ifdef VIMAGE
1446 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1447 #endif
1448 			    tpr->pr_uref == 0) {
1449 				descend = 0;
1450 				continue;
1451 			}
1452 			if (!(tpr->pr_flags & PR_IP6_USER))
1453 				continue;
1454 			descend = 0;
1455 			if (tpr->pr_ip6 == NULL ||
1456 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1457 				continue;
1458 			for (ii = 0; ii < ip6s; ii++) {
1459 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1460 					error = EADDRINUSE;
1461 					vfs_opterror(opts,
1462 					    "IPv6 addresses clash");
1463 					goto done_deref_locked;
1464 				}
1465 			}
1466 		}
1467 	}
1468 #endif
1469 	onamelen = namelen = 0;
1470 	if (name != NULL) {
1471 		/* Give a default name of the jid. */
1472 		if (name[0] == '\0')
1473 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1474 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1475 		    *p == '\0')) {
1476 			error = EINVAL;
1477 			vfs_opterror(opts,
1478 			    "name cannot be numeric (unless it is the jid)");
1479 			goto done_deref_locked;
1480 		}
1481 		/*
1482 		 * Make sure the name isn't too long for the prison or its
1483 		 * children.
1484 		 */
1485 		onamelen = strlen(pr->pr_name);
1486 		namelen = strlen(name);
1487 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1488 			error = ENAMETOOLONG;
1489 			goto done_deref_locked;
1490 		}
1491 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1492 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1493 			    sizeof(pr->pr_name)) {
1494 				error = ENAMETOOLONG;
1495 				goto done_deref_locked;
1496 			}
1497 		}
1498 	}
1499 	if (pr_allow & ~ppr->pr_allow) {
1500 		error = EPERM;
1501 		goto done_deref_locked;
1502 	}
1503 
1504 	/* Set the parameters of the prison. */
1505 #ifdef INET
1506 	redo_ip4 = 0;
1507 	if (pr_flags & PR_IP4_USER) {
1508 		pr->pr_flags |= PR_IP4;
1509 		free(pr->pr_ip4, M_PRISON);
1510 		pr->pr_ip4s = ip4s;
1511 		pr->pr_ip4 = ip4;
1512 		ip4 = NULL;
1513 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1514 #ifdef VIMAGE
1515 			if (tpr->pr_flags & PR_VNET) {
1516 				descend = 0;
1517 				continue;
1518 			}
1519 #endif
1520 			if (prison_restrict_ip4(tpr, NULL)) {
1521 				redo_ip4 = 1;
1522 				descend = 0;
1523 			}
1524 		}
1525 	}
1526 #endif
1527 #ifdef INET6
1528 	redo_ip6 = 0;
1529 	if (pr_flags & PR_IP6_USER) {
1530 		pr->pr_flags |= PR_IP6;
1531 		free(pr->pr_ip6, M_PRISON);
1532 		pr->pr_ip6s = ip6s;
1533 		pr->pr_ip6 = ip6;
1534 		ip6 = NULL;
1535 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1536 #ifdef VIMAGE
1537 			if (tpr->pr_flags & PR_VNET) {
1538 				descend = 0;
1539 				continue;
1540 			}
1541 #endif
1542 			if (prison_restrict_ip6(tpr, NULL)) {
1543 				redo_ip6 = 1;
1544 				descend = 0;
1545 			}
1546 		}
1547 	}
1548 #endif
1549 	if (gotslevel) {
1550 		pr->pr_securelevel = slevel;
1551 		/* Set all child jails to be at least this level. */
1552 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1553 			if (tpr->pr_securelevel < slevel)
1554 				tpr->pr_securelevel = slevel;
1555 	}
1556 	if (gotchildmax) {
1557 		pr->pr_childmax = childmax;
1558 		/* Set all child jails to under this limit. */
1559 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1560 			if (tpr->pr_childmax > childmax - level)
1561 				tpr->pr_childmax = childmax > level
1562 				    ? childmax - level : 0;
1563 	}
1564 	if (gotenforce) {
1565 		pr->pr_enforce_statfs = enforce;
1566 		/* Pass this restriction on to the children. */
1567 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1568 			if (tpr->pr_enforce_statfs < enforce)
1569 				tpr->pr_enforce_statfs = enforce;
1570 	}
1571 	if (name != NULL) {
1572 		if (ppr == &prison0)
1573 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1574 		else
1575 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1576 			    ppr->pr_name, name);
1577 		/* Change this component of child names. */
1578 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1579 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1580 			    strlen(tpr->pr_name + onamelen) + 1);
1581 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1582 		}
1583 	}
1584 	if (path != NULL) {
1585 		/* Try to keep a real-rooted full pathname. */
1586 		if (path[0] == '/' && strcmp(mypr->pr_path, "/"))
1587 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1588 			    mypr->pr_path, path);
1589 		else
1590 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1591 		pr->pr_root = root;
1592 	}
1593 	if (PR_HOST & ch_flags & ~pr_flags) {
1594 		if (pr->pr_flags & PR_HOST) {
1595 			/*
1596 			 * Copy the parent's host info.  As with pr_ip4 above,
1597 			 * the lack of a lock on the parent is not a problem;
1598 			 * it is always set with allprison_lock at least
1599 			 * shared, and is held exclusively here.
1600 			 */
1601 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1602 			    sizeof(pr->pr_hostname));
1603 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1604 			    sizeof(pr->pr_domainname));
1605 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1606 			    sizeof(pr->pr_hostuuid));
1607 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1608 		}
1609 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1610 		/* Set this prison, and any descendants without PR_HOST. */
1611 		if (host != NULL)
1612 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1613 		if (domain != NULL)
1614 			strlcpy(pr->pr_domainname, domain,
1615 			    sizeof(pr->pr_domainname));
1616 		if (uuid != NULL)
1617 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1618 		if (gothid)
1619 			pr->pr_hostid = hid;
1620 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1621 			if (tpr->pr_flags & PR_HOST)
1622 				descend = 0;
1623 			else {
1624 				if (host != NULL)
1625 					strlcpy(tpr->pr_hostname,
1626 					    pr->pr_hostname,
1627 					    sizeof(tpr->pr_hostname));
1628 				if (domain != NULL)
1629 					strlcpy(tpr->pr_domainname,
1630 					    pr->pr_domainname,
1631 					    sizeof(tpr->pr_domainname));
1632 				if (uuid != NULL)
1633 					strlcpy(tpr->pr_hostuuid,
1634 					    pr->pr_hostuuid,
1635 					    sizeof(tpr->pr_hostuuid));
1636 				if (gothid)
1637 					tpr->pr_hostid = hid;
1638 			}
1639 		}
1640 	}
1641 	if ((tallow = ch_allow & ~pr_allow)) {
1642 		/* Clear allow bits in all children. */
1643 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1644 			tpr->pr_allow &= ~tallow;
1645 	}
1646 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1647 	/*
1648 	 * Persistent prisons get an extra reference, and prisons losing their
1649 	 * persist flag lose that reference.  Only do this for existing prisons
1650 	 * for now, so new ones will remain unseen until after the module
1651 	 * handlers have completed.
1652 	 */
1653 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1654 		if (pr_flags & PR_PERSIST) {
1655 			pr->pr_ref++;
1656 			pr->pr_uref++;
1657 		} else {
1658 			pr->pr_ref--;
1659 			pr->pr_uref--;
1660 		}
1661 	}
1662 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1663 	mtx_unlock(&pr->pr_mtx);
1664 
1665 #ifdef RACCT
1666 	if (created)
1667 		prison_racct_attach(pr);
1668 #endif
1669 
1670 	/* Locks may have prevented a complete restriction of child IP
1671 	 * addresses.  If so, allocate some more memory and try again.
1672 	 */
1673 #ifdef INET
1674 	while (redo_ip4) {
1675 		ip4s = pr->pr_ip4s;
1676 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1677 		mtx_lock(&pr->pr_mtx);
1678 		redo_ip4 = 0;
1679 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1680 #ifdef VIMAGE
1681 			if (tpr->pr_flags & PR_VNET) {
1682 				descend = 0;
1683 				continue;
1684 			}
1685 #endif
1686 			if (prison_restrict_ip4(tpr, ip4)) {
1687 				if (ip4 != NULL)
1688 					ip4 = NULL;
1689 				else
1690 					redo_ip4 = 1;
1691 			}
1692 		}
1693 		mtx_unlock(&pr->pr_mtx);
1694 	}
1695 #endif
1696 #ifdef INET6
1697 	while (redo_ip6) {
1698 		ip6s = pr->pr_ip6s;
1699 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1700 		mtx_lock(&pr->pr_mtx);
1701 		redo_ip6 = 0;
1702 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1703 #ifdef VIMAGE
1704 			if (tpr->pr_flags & PR_VNET) {
1705 				descend = 0;
1706 				continue;
1707 			}
1708 #endif
1709 			if (prison_restrict_ip6(tpr, ip6)) {
1710 				if (ip6 != NULL)
1711 					ip6 = NULL;
1712 				else
1713 					redo_ip6 = 1;
1714 			}
1715 		}
1716 		mtx_unlock(&pr->pr_mtx);
1717 	}
1718 #endif
1719 
1720 	/* Let the modules do their work. */
1721 	sx_downgrade(&allprison_lock);
1722 	if (created) {
1723 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1724 		if (error) {
1725 			prison_deref(pr, PD_LIST_SLOCKED);
1726 			goto done_errmsg;
1727 		}
1728 	}
1729 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1730 	if (error) {
1731 		prison_deref(pr, created
1732 		    ? PD_LIST_SLOCKED
1733 		    : PD_DEREF | PD_LIST_SLOCKED);
1734 		goto done_errmsg;
1735 	}
1736 
1737 	/* Attach this process to the prison if requested. */
1738 	if (flags & JAIL_ATTACH) {
1739 		mtx_lock(&pr->pr_mtx);
1740 		error = do_jail_attach(td, pr);
1741 		if (error) {
1742 			vfs_opterror(opts, "attach failed");
1743 			if (!created)
1744 				prison_deref(pr, PD_DEREF);
1745 			goto done_errmsg;
1746 		}
1747 	}
1748 
1749 	/*
1750 	 * Now that it is all there, drop the temporary reference from existing
1751 	 * prisons.  Or add a reference to newly created persistent prisons
1752 	 * (which was not done earlier so that the prison would not be publicly
1753 	 * visible).
1754 	 */
1755 	if (!created) {
1756 		prison_deref(pr, (flags & JAIL_ATTACH)
1757 		    ? PD_DEREF
1758 		    : PD_DEREF | PD_LIST_SLOCKED);
1759 	} else {
1760 		if (pr_flags & PR_PERSIST) {
1761 			mtx_lock(&pr->pr_mtx);
1762 			pr->pr_ref++;
1763 			pr->pr_uref++;
1764 			mtx_unlock(&pr->pr_mtx);
1765 		}
1766 		if (!(flags & JAIL_ATTACH))
1767 			sx_sunlock(&allprison_lock);
1768 	}
1769 	td->td_retval[0] = pr->pr_id;
1770 	goto done_errmsg;
1771 
1772  done_deref_locked:
1773 	prison_deref(pr, created
1774 	    ? PD_LOCKED | PD_LIST_XLOCKED
1775 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1776 	goto done_releroot;
1777  done_unlock_list:
1778 	sx_xunlock(&allprison_lock);
1779  done_releroot:
1780 	if (root != NULL) {
1781 		vfslocked = VFS_LOCK_GIANT(root->v_mount);
1782 		vrele(root);
1783 		VFS_UNLOCK_GIANT(vfslocked);
1784 	}
1785  done_errmsg:
1786 	if (error) {
1787 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1788 		if (errmsg_len > 0) {
1789 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1790 			if (errmsg_pos > 0) {
1791 				if (optuio->uio_segflg == UIO_SYSSPACE)
1792 					bcopy(errmsg,
1793 					   optuio->uio_iov[errmsg_pos].iov_base,
1794 					   errmsg_len);
1795 				else
1796 					copyout(errmsg,
1797 					   optuio->uio_iov[errmsg_pos].iov_base,
1798 					   errmsg_len);
1799 			}
1800 		}
1801 	}
1802  done_free:
1803 #ifdef INET
1804 	free(ip4, M_PRISON);
1805 #endif
1806 #ifdef INET6
1807 	free(ip6, M_PRISON);
1808 #endif
1809 	vfs_freeopts(opts);
1810 	return (error);
1811 }
1812 
1813 
1814 /*
1815  * struct jail_get_args {
1816  *	struct iovec *iovp;
1817  *	unsigned int iovcnt;
1818  *	int flags;
1819  * };
1820  */
1821 int
1822 jail_get(struct thread *td, struct jail_get_args *uap)
1823 {
1824 	struct uio *auio;
1825 	int error;
1826 
1827 	/* Check that we have an even number of iovecs. */
1828 	if (uap->iovcnt & 1)
1829 		return (EINVAL);
1830 
1831 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1832 	if (error)
1833 		return (error);
1834 	error = kern_jail_get(td, auio, uap->flags);
1835 	if (error == 0)
1836 		error = copyout(auio->uio_iov, uap->iovp,
1837 		    uap->iovcnt * sizeof (struct iovec));
1838 	free(auio, M_IOV);
1839 	return (error);
1840 }
1841 
1842 int
1843 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1844 {
1845 	struct prison *pr, *mypr;
1846 	struct vfsopt *opt;
1847 	struct vfsoptlist *opts;
1848 	char *errmsg, *name;
1849 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1850 
1851 	if (flags & ~JAIL_GET_MASK)
1852 		return (EINVAL);
1853 
1854 	/* Get the parameter list. */
1855 	error = vfs_buildopts(optuio, &opts);
1856 	if (error)
1857 		return (error);
1858 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1859 	mypr = td->td_ucred->cr_prison;
1860 
1861 	/*
1862 	 * Find the prison specified by one of: lastjid, jid, name.
1863 	 */
1864 	sx_slock(&allprison_lock);
1865 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1866 	if (error == 0) {
1867 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1868 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1869 				mtx_lock(&pr->pr_mtx);
1870 				if (pr->pr_ref > 0 &&
1871 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1872 					break;
1873 				mtx_unlock(&pr->pr_mtx);
1874 			}
1875 		}
1876 		if (pr != NULL)
1877 			goto found_prison;
1878 		error = ENOENT;
1879 		vfs_opterror(opts, "no jail after %d", jid);
1880 		goto done_unlock_list;
1881 	} else if (error != ENOENT)
1882 		goto done_unlock_list;
1883 
1884 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1885 	if (error == 0) {
1886 		if (jid != 0) {
1887 			pr = prison_find_child(mypr, jid);
1888 			if (pr != NULL) {
1889 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1890 					mtx_unlock(&pr->pr_mtx);
1891 					error = ENOENT;
1892 					vfs_opterror(opts, "jail %d is dying",
1893 					    jid);
1894 					goto done_unlock_list;
1895 				}
1896 				goto found_prison;
1897 			}
1898 			error = ENOENT;
1899 			vfs_opterror(opts, "jail %d not found", jid);
1900 			goto done_unlock_list;
1901 		}
1902 	} else if (error != ENOENT)
1903 		goto done_unlock_list;
1904 
1905 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1906 	if (error == 0) {
1907 		if (len == 0 || name[len - 1] != '\0') {
1908 			error = EINVAL;
1909 			goto done_unlock_list;
1910 		}
1911 		pr = prison_find_name(mypr, name);
1912 		if (pr != NULL) {
1913 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1914 				mtx_unlock(&pr->pr_mtx);
1915 				error = ENOENT;
1916 				vfs_opterror(opts, "jail \"%s\" is dying",
1917 				    name);
1918 				goto done_unlock_list;
1919 			}
1920 			goto found_prison;
1921 		}
1922 		error = ENOENT;
1923 		vfs_opterror(opts, "jail \"%s\" not found", name);
1924 		goto done_unlock_list;
1925 	} else if (error != ENOENT)
1926 		goto done_unlock_list;
1927 
1928 	vfs_opterror(opts, "no jail specified");
1929 	error = ENOENT;
1930 	goto done_unlock_list;
1931 
1932  found_prison:
1933 	/* Get the parameters of the prison. */
1934 	pr->pr_ref++;
1935 	locked = PD_LOCKED;
1936 	td->td_retval[0] = pr->pr_id;
1937 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
1938 	if (error != 0 && error != ENOENT)
1939 		goto done_deref;
1940 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
1941 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
1942 	if (error != 0 && error != ENOENT)
1943 		goto done_deref;
1944 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
1945 	if (error != 0 && error != ENOENT)
1946 		goto done_deref;
1947 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
1948 	    sizeof(pr->pr_cpuset->cs_id));
1949 	if (error != 0 && error != ENOENT)
1950 		goto done_deref;
1951 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
1952 	if (error != 0 && error != ENOENT)
1953 		goto done_deref;
1954 #ifdef INET
1955 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
1956 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1957 	if (error != 0 && error != ENOENT)
1958 		goto done_deref;
1959 #endif
1960 #ifdef INET6
1961 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
1962 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1963 	if (error != 0 && error != ENOENT)
1964 		goto done_deref;
1965 #endif
1966 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
1967 	    sizeof(pr->pr_securelevel));
1968 	if (error != 0 && error != ENOENT)
1969 		goto done_deref;
1970 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
1971 	    sizeof(pr->pr_childcount));
1972 	if (error != 0 && error != ENOENT)
1973 		goto done_deref;
1974 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
1975 	    sizeof(pr->pr_childmax));
1976 	if (error != 0 && error != ENOENT)
1977 		goto done_deref;
1978 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
1979 	if (error != 0 && error != ENOENT)
1980 		goto done_deref;
1981 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
1982 	if (error != 0 && error != ENOENT)
1983 		goto done_deref;
1984 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
1985 	if (error != 0 && error != ENOENT)
1986 		goto done_deref;
1987 #ifdef COMPAT_FREEBSD32
1988 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1989 		uint32_t hid32 = pr->pr_hostid;
1990 
1991 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
1992 	} else
1993 #endif
1994 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
1995 	    sizeof(pr->pr_hostid));
1996 	if (error != 0 && error != ENOENT)
1997 		goto done_deref;
1998 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
1999 	    sizeof(pr->pr_enforce_statfs));
2000 	if (error != 0 && error != ENOENT)
2001 		goto done_deref;
2002 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2003 	    fi++) {
2004 		if (pr_flag_names[fi] == NULL)
2005 			continue;
2006 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2007 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2008 		if (error != 0 && error != ENOENT)
2009 			goto done_deref;
2010 		i = !i;
2011 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2012 		if (error != 0 && error != ENOENT)
2013 			goto done_deref;
2014 	}
2015 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2016 	    fi++) {
2017 		i = pr->pr_flags &
2018 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2019 		i = pr_flag_jailsys[fi].disable &&
2020 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2021 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2022 		    : JAIL_SYS_INHERIT;
2023 		error =
2024 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2025 		if (error != 0 && error != ENOENT)
2026 			goto done_deref;
2027 	}
2028 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2029 	    fi++) {
2030 		if (pr_allow_names[fi] == NULL)
2031 			continue;
2032 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2033 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2034 		if (error != 0 && error != ENOENT)
2035 			goto done_deref;
2036 		i = !i;
2037 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2038 		if (error != 0 && error != ENOENT)
2039 			goto done_deref;
2040 	}
2041 	i = (pr->pr_uref == 0);
2042 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2043 	if (error != 0 && error != ENOENT)
2044 		goto done_deref;
2045 	i = !i;
2046 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2047 	if (error != 0 && error != ENOENT)
2048 		goto done_deref;
2049 
2050 	/* Get the module parameters. */
2051 	mtx_unlock(&pr->pr_mtx);
2052 	locked = 0;
2053 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2054 	if (error)
2055 		goto done_deref;
2056 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2057 
2058 	/* By now, all parameters should have been noted. */
2059 	TAILQ_FOREACH(opt, opts, link) {
2060 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2061 			error = EINVAL;
2062 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2063 			goto done_errmsg;
2064 		}
2065 	}
2066 
2067 	/* Write the fetched parameters back to userspace. */
2068 	error = 0;
2069 	TAILQ_FOREACH(opt, opts, link) {
2070 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2071 			pos = 2 * opt->pos + 1;
2072 			optuio->uio_iov[pos].iov_len = opt->len;
2073 			if (opt->value != NULL) {
2074 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2075 					bcopy(opt->value,
2076 					    optuio->uio_iov[pos].iov_base,
2077 					    opt->len);
2078 				} else {
2079 					error = copyout(opt->value,
2080 					    optuio->uio_iov[pos].iov_base,
2081 					    opt->len);
2082 					if (error)
2083 						break;
2084 				}
2085 			}
2086 		}
2087 	}
2088 	goto done_errmsg;
2089 
2090  done_deref:
2091 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2092 	goto done_errmsg;
2093 
2094  done_unlock_list:
2095 	sx_sunlock(&allprison_lock);
2096  done_errmsg:
2097 	if (error && errmsg_pos >= 0) {
2098 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2099 		errmsg_pos = 2 * errmsg_pos + 1;
2100 		if (errmsg_len > 0) {
2101 			if (optuio->uio_segflg == UIO_SYSSPACE)
2102 				bcopy(errmsg,
2103 				    optuio->uio_iov[errmsg_pos].iov_base,
2104 				    errmsg_len);
2105 			else
2106 				copyout(errmsg,
2107 				    optuio->uio_iov[errmsg_pos].iov_base,
2108 				    errmsg_len);
2109 		}
2110 	}
2111 	vfs_freeopts(opts);
2112 	return (error);
2113 }
2114 
2115 
2116 /*
2117  * struct jail_remove_args {
2118  *	int jid;
2119  * };
2120  */
2121 int
2122 jail_remove(struct thread *td, struct jail_remove_args *uap)
2123 {
2124 	struct prison *pr, *cpr, *lpr, *tpr;
2125 	int descend, error;
2126 
2127 	error = priv_check(td, PRIV_JAIL_REMOVE);
2128 	if (error)
2129 		return (error);
2130 
2131 	sx_xlock(&allprison_lock);
2132 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2133 	if (pr == NULL) {
2134 		sx_xunlock(&allprison_lock);
2135 		return (EINVAL);
2136 	}
2137 
2138 	/* Remove all descendants of this prison, then remove this prison. */
2139 	pr->pr_ref++;
2140 	pr->pr_flags |= PR_REMOVE;
2141 	if (!LIST_EMPTY(&pr->pr_children)) {
2142 		mtx_unlock(&pr->pr_mtx);
2143 		lpr = NULL;
2144 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2145 			mtx_lock(&cpr->pr_mtx);
2146 			if (cpr->pr_ref > 0) {
2147 				tpr = cpr;
2148 				cpr->pr_ref++;
2149 				cpr->pr_flags |= PR_REMOVE;
2150 			} else {
2151 				/* Already removed - do not do it again. */
2152 				tpr = NULL;
2153 			}
2154 			mtx_unlock(&cpr->pr_mtx);
2155 			if (lpr != NULL) {
2156 				mtx_lock(&lpr->pr_mtx);
2157 				prison_remove_one(lpr);
2158 				sx_xlock(&allprison_lock);
2159 			}
2160 			lpr = tpr;
2161 		}
2162 		if (lpr != NULL) {
2163 			mtx_lock(&lpr->pr_mtx);
2164 			prison_remove_one(lpr);
2165 			sx_xlock(&allprison_lock);
2166 		}
2167 		mtx_lock(&pr->pr_mtx);
2168 	}
2169 	prison_remove_one(pr);
2170 	return (0);
2171 }
2172 
2173 static void
2174 prison_remove_one(struct prison *pr)
2175 {
2176 	struct proc *p;
2177 	int deuref;
2178 
2179 	/* If the prison was persistent, it is not anymore. */
2180 	deuref = 0;
2181 	if (pr->pr_flags & PR_PERSIST) {
2182 		pr->pr_ref--;
2183 		deuref = PD_DEUREF;
2184 		pr->pr_flags &= ~PR_PERSIST;
2185 	}
2186 
2187 	/*
2188 	 * jail_remove added a reference.  If that's the only one, remove
2189 	 * the prison now.
2190 	 */
2191 	KASSERT(pr->pr_ref > 0,
2192 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2193 	if (pr->pr_ref == 1) {
2194 		prison_deref(pr,
2195 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2196 		return;
2197 	}
2198 
2199 	mtx_unlock(&pr->pr_mtx);
2200 	sx_xunlock(&allprison_lock);
2201 	/*
2202 	 * Kill all processes unfortunate enough to be attached to this prison.
2203 	 */
2204 	sx_slock(&allproc_lock);
2205 	LIST_FOREACH(p, &allproc, p_list) {
2206 		PROC_LOCK(p);
2207 		if (p->p_state != PRS_NEW && p->p_ucred &&
2208 		    p->p_ucred->cr_prison == pr)
2209 			psignal(p, SIGKILL);
2210 		PROC_UNLOCK(p);
2211 	}
2212 	sx_sunlock(&allproc_lock);
2213 	/* Remove the temporary reference added by jail_remove. */
2214 	prison_deref(pr, deuref | PD_DEREF);
2215 }
2216 
2217 
2218 /*
2219  * struct jail_attach_args {
2220  *	int jid;
2221  * };
2222  */
2223 int
2224 jail_attach(struct thread *td, struct jail_attach_args *uap)
2225 {
2226 	struct prison *pr;
2227 	int error;
2228 
2229 	error = priv_check(td, PRIV_JAIL_ATTACH);
2230 	if (error)
2231 		return (error);
2232 
2233 	sx_slock(&allprison_lock);
2234 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2235 	if (pr == NULL) {
2236 		sx_sunlock(&allprison_lock);
2237 		return (EINVAL);
2238 	}
2239 
2240 	/*
2241 	 * Do not allow a process to attach to a prison that is not
2242 	 * considered to be "alive".
2243 	 */
2244 	if (pr->pr_uref == 0) {
2245 		mtx_unlock(&pr->pr_mtx);
2246 		sx_sunlock(&allprison_lock);
2247 		return (EINVAL);
2248 	}
2249 
2250 	return (do_jail_attach(td, pr));
2251 }
2252 
2253 static int
2254 do_jail_attach(struct thread *td, struct prison *pr)
2255 {
2256 	struct prison *ppr;
2257 	struct proc *p;
2258 	struct ucred *newcred, *oldcred;
2259 	int vfslocked, error;
2260 
2261 	/*
2262 	 * XXX: Note that there is a slight race here if two threads
2263 	 * in the same privileged process attempt to attach to two
2264 	 * different jails at the same time.  It is important for
2265 	 * user processes not to do this, or they might end up with
2266 	 * a process root from one prison, but attached to the jail
2267 	 * of another.
2268 	 */
2269 	pr->pr_ref++;
2270 	pr->pr_uref++;
2271 	mtx_unlock(&pr->pr_mtx);
2272 
2273 	/* Let modules do whatever they need to prepare for attaching. */
2274 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2275 	if (error) {
2276 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2277 		return (error);
2278 	}
2279 	sx_sunlock(&allprison_lock);
2280 
2281 	/*
2282 	 * Reparent the newly attached process to this jail.
2283 	 */
2284 	ppr = td->td_ucred->cr_prison;
2285 	p = td->td_proc;
2286 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2287 	if (error)
2288 		goto e_revert_osd;
2289 
2290 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2291 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2292 	if ((error = change_dir(pr->pr_root, td)) != 0)
2293 		goto e_unlock;
2294 #ifdef MAC
2295 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2296 		goto e_unlock;
2297 #endif
2298 	VOP_UNLOCK(pr->pr_root, 0);
2299 	if ((error = change_root(pr->pr_root, td)))
2300 		goto e_unlock_giant;
2301 	VFS_UNLOCK_GIANT(vfslocked);
2302 
2303 	newcred = crget();
2304 	PROC_LOCK(p);
2305 	oldcred = p->p_ucred;
2306 	setsugid(p);
2307 	crcopy(newcred, oldcred);
2308 	newcred->cr_prison = pr;
2309 	p->p_ucred = newcred;
2310 	PROC_UNLOCK(p);
2311 #ifdef RACCT
2312 	racct_proc_ucred_changed(p, oldcred, newcred);
2313 #endif
2314 	crfree(oldcred);
2315 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2316 	return (0);
2317  e_unlock:
2318 	VOP_UNLOCK(pr->pr_root, 0);
2319  e_unlock_giant:
2320 	VFS_UNLOCK_GIANT(vfslocked);
2321  e_revert_osd:
2322 	/* Tell modules this thread is still in its old jail after all. */
2323 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2324 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2325 	return (error);
2326 }
2327 
2328 
2329 /*
2330  * Returns a locked prison instance, or NULL on failure.
2331  */
2332 struct prison *
2333 prison_find(int prid)
2334 {
2335 	struct prison *pr;
2336 
2337 	sx_assert(&allprison_lock, SX_LOCKED);
2338 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2339 		if (pr->pr_id == prid) {
2340 			mtx_lock(&pr->pr_mtx);
2341 			if (pr->pr_ref > 0)
2342 				return (pr);
2343 			mtx_unlock(&pr->pr_mtx);
2344 		}
2345 	}
2346 	return (NULL);
2347 }
2348 
2349 /*
2350  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2351  */
2352 struct prison *
2353 prison_find_child(struct prison *mypr, int prid)
2354 {
2355 	struct prison *pr;
2356 	int descend;
2357 
2358 	sx_assert(&allprison_lock, SX_LOCKED);
2359 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2360 		if (pr->pr_id == prid) {
2361 			mtx_lock(&pr->pr_mtx);
2362 			if (pr->pr_ref > 0)
2363 				return (pr);
2364 			mtx_unlock(&pr->pr_mtx);
2365 		}
2366 	}
2367 	return (NULL);
2368 }
2369 
2370 /*
2371  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2372  */
2373 struct prison *
2374 prison_find_name(struct prison *mypr, const char *name)
2375 {
2376 	struct prison *pr, *deadpr;
2377 	size_t mylen;
2378 	int descend;
2379 
2380 	sx_assert(&allprison_lock, SX_LOCKED);
2381 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2382  again:
2383 	deadpr = NULL;
2384 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2385 		if (!strcmp(pr->pr_name + mylen, name)) {
2386 			mtx_lock(&pr->pr_mtx);
2387 			if (pr->pr_ref > 0) {
2388 				if (pr->pr_uref > 0)
2389 					return (pr);
2390 				deadpr = pr;
2391 			}
2392 			mtx_unlock(&pr->pr_mtx);
2393 		}
2394 	}
2395 	/* There was no valid prison - perhaps there was a dying one. */
2396 	if (deadpr != NULL) {
2397 		mtx_lock(&deadpr->pr_mtx);
2398 		if (deadpr->pr_ref == 0) {
2399 			mtx_unlock(&deadpr->pr_mtx);
2400 			goto again;
2401 		}
2402 	}
2403 	return (deadpr);
2404 }
2405 
2406 /*
2407  * See if a prison has the specific flag set.
2408  */
2409 int
2410 prison_flag(struct ucred *cred, unsigned flag)
2411 {
2412 
2413 	/* This is an atomic read, so no locking is necessary. */
2414 	return (cred->cr_prison->pr_flags & flag);
2415 }
2416 
2417 int
2418 prison_allow(struct ucred *cred, unsigned flag)
2419 {
2420 
2421 	/* This is an atomic read, so no locking is necessary. */
2422 	return (cred->cr_prison->pr_allow & flag);
2423 }
2424 
2425 /*
2426  * Remove a prison reference.  If that was the last reference, remove the
2427  * prison itself - but not in this context in case there are locks held.
2428  */
2429 void
2430 prison_free_locked(struct prison *pr)
2431 {
2432 
2433 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2434 	pr->pr_ref--;
2435 	if (pr->pr_ref == 0) {
2436 		mtx_unlock(&pr->pr_mtx);
2437 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2438 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2439 		return;
2440 	}
2441 	mtx_unlock(&pr->pr_mtx);
2442 }
2443 
2444 void
2445 prison_free(struct prison *pr)
2446 {
2447 
2448 	mtx_lock(&pr->pr_mtx);
2449 	prison_free_locked(pr);
2450 }
2451 
2452 static void
2453 prison_complete(void *context, int pending)
2454 {
2455 
2456 	prison_deref((struct prison *)context, 0);
2457 }
2458 
2459 /*
2460  * Remove a prison reference (usually).  This internal version assumes no
2461  * mutexes are held, except perhaps the prison itself.  If there are no more
2462  * references, release and delist the prison.  On completion, the prison lock
2463  * and the allprison lock are both unlocked.
2464  */
2465 static void
2466 prison_deref(struct prison *pr, int flags)
2467 {
2468 	struct prison *ppr, *tpr;
2469 	int vfslocked;
2470 
2471 	if (!(flags & PD_LOCKED))
2472 		mtx_lock(&pr->pr_mtx);
2473 	/* Decrement the user references in a separate loop. */
2474 	if (flags & PD_DEUREF) {
2475 		for (tpr = pr;; tpr = tpr->pr_parent) {
2476 			if (tpr != pr)
2477 				mtx_lock(&tpr->pr_mtx);
2478 			if (--tpr->pr_uref > 0)
2479 				break;
2480 			KASSERT(tpr != &prison0, ("prison0 pr_uref=0"));
2481 			mtx_unlock(&tpr->pr_mtx);
2482 		}
2483 		/* Done if there were only user references to remove. */
2484 		if (!(flags & PD_DEREF)) {
2485 			mtx_unlock(&tpr->pr_mtx);
2486 			if (flags & PD_LIST_SLOCKED)
2487 				sx_sunlock(&allprison_lock);
2488 			else if (flags & PD_LIST_XLOCKED)
2489 				sx_xunlock(&allprison_lock);
2490 			return;
2491 		}
2492 		if (tpr != pr) {
2493 			mtx_unlock(&tpr->pr_mtx);
2494 			mtx_lock(&pr->pr_mtx);
2495 		}
2496 	}
2497 
2498 	for (;;) {
2499 		if (flags & PD_DEREF)
2500 			pr->pr_ref--;
2501 		/* If the prison still has references, nothing else to do. */
2502 		if (pr->pr_ref > 0) {
2503 			mtx_unlock(&pr->pr_mtx);
2504 			if (flags & PD_LIST_SLOCKED)
2505 				sx_sunlock(&allprison_lock);
2506 			else if (flags & PD_LIST_XLOCKED)
2507 				sx_xunlock(&allprison_lock);
2508 			return;
2509 		}
2510 
2511 		mtx_unlock(&pr->pr_mtx);
2512 		if (flags & PD_LIST_SLOCKED) {
2513 			if (!sx_try_upgrade(&allprison_lock)) {
2514 				sx_sunlock(&allprison_lock);
2515 				sx_xlock(&allprison_lock);
2516 			}
2517 		} else if (!(flags & PD_LIST_XLOCKED))
2518 			sx_xlock(&allprison_lock);
2519 
2520 		TAILQ_REMOVE(&allprison, pr, pr_list);
2521 		LIST_REMOVE(pr, pr_sibling);
2522 		ppr = pr->pr_parent;
2523 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2524 			tpr->pr_childcount--;
2525 		sx_xunlock(&allprison_lock);
2526 
2527 #ifdef VIMAGE
2528 		if (pr->pr_vnet != ppr->pr_vnet)
2529 			vnet_destroy(pr->pr_vnet);
2530 #endif
2531 		if (pr->pr_root != NULL) {
2532 			vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2533 			vrele(pr->pr_root);
2534 			VFS_UNLOCK_GIANT(vfslocked);
2535 		}
2536 		mtx_destroy(&pr->pr_mtx);
2537 #ifdef INET
2538 		free(pr->pr_ip4, M_PRISON);
2539 #endif
2540 #ifdef INET6
2541 		free(pr->pr_ip6, M_PRISON);
2542 #endif
2543 		if (pr->pr_cpuset != NULL)
2544 			cpuset_rel(pr->pr_cpuset);
2545 		osd_jail_exit(pr);
2546 #ifdef RACCT
2547 		prison_racct_detach(pr);
2548 #endif
2549 		free(pr, M_PRISON);
2550 
2551 		/* Removing a prison frees a reference on its parent. */
2552 		pr = ppr;
2553 		mtx_lock(&pr->pr_mtx);
2554 		flags = PD_DEREF;
2555 	}
2556 }
2557 
2558 void
2559 prison_hold_locked(struct prison *pr)
2560 {
2561 
2562 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2563 	KASSERT(pr->pr_ref > 0,
2564 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2565 	pr->pr_ref++;
2566 }
2567 
2568 void
2569 prison_hold(struct prison *pr)
2570 {
2571 
2572 	mtx_lock(&pr->pr_mtx);
2573 	prison_hold_locked(pr);
2574 	mtx_unlock(&pr->pr_mtx);
2575 }
2576 
2577 void
2578 prison_proc_hold(struct prison *pr)
2579 {
2580 
2581 	mtx_lock(&pr->pr_mtx);
2582 	KASSERT(pr->pr_uref > 0,
2583 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2584 	pr->pr_uref++;
2585 	mtx_unlock(&pr->pr_mtx);
2586 }
2587 
2588 void
2589 prison_proc_free(struct prison *pr)
2590 {
2591 
2592 	mtx_lock(&pr->pr_mtx);
2593 	KASSERT(pr->pr_uref > 0,
2594 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2595 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2596 }
2597 
2598 
2599 #ifdef INET
2600 /*
2601  * Restrict a prison's IP address list with its parent's, possibly replacing
2602  * it.  Return true if the replacement buffer was used (or would have been).
2603  */
2604 static int
2605 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2606 {
2607 	int ii, ij, used;
2608 	struct prison *ppr;
2609 
2610 	ppr = pr->pr_parent;
2611 	if (!(pr->pr_flags & PR_IP4_USER)) {
2612 		/* This has no user settings, so just copy the parent's list. */
2613 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2614 			/*
2615 			 * There's no room for the parent's list.  Use the
2616 			 * new list buffer, which is assumed to be big enough
2617 			 * (if it was passed).  If there's no buffer, try to
2618 			 * allocate one.
2619 			 */
2620 			used = 1;
2621 			if (newip4 == NULL) {
2622 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2623 				    M_PRISON, M_NOWAIT);
2624 				if (newip4 != NULL)
2625 					used = 0;
2626 			}
2627 			if (newip4 != NULL) {
2628 				bcopy(ppr->pr_ip4, newip4,
2629 				    ppr->pr_ip4s * sizeof(*newip4));
2630 				free(pr->pr_ip4, M_PRISON);
2631 				pr->pr_ip4 = newip4;
2632 				pr->pr_ip4s = ppr->pr_ip4s;
2633 			}
2634 			return (used);
2635 		}
2636 		pr->pr_ip4s = ppr->pr_ip4s;
2637 		if (pr->pr_ip4s > 0)
2638 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2639 			    pr->pr_ip4s * sizeof(*newip4));
2640 		else if (pr->pr_ip4 != NULL) {
2641 			free(pr->pr_ip4, M_PRISON);
2642 			pr->pr_ip4 = NULL;
2643 		}
2644 	} else if (pr->pr_ip4s > 0) {
2645 		/* Remove addresses that aren't in the parent. */
2646 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2647 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2648 				break;
2649 		if (ij < ppr->pr_ip4s)
2650 			ii = 1;
2651 		else {
2652 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2653 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2654 			ii = 0;
2655 		}
2656 		for (ij = 1; ii < pr->pr_ip4s; ) {
2657 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2658 				ii++;
2659 				continue;
2660 			}
2661 			switch (ij >= ppr->pr_ip4s ? -1 :
2662 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2663 			case -1:
2664 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2665 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2666 				break;
2667 			case 0:
2668 				ii++;
2669 				ij++;
2670 				break;
2671 			case 1:
2672 				ij++;
2673 				break;
2674 			}
2675 		}
2676 		if (pr->pr_ip4s == 0) {
2677 			pr->pr_flags |= PR_IP4_DISABLE;
2678 			free(pr->pr_ip4, M_PRISON);
2679 			pr->pr_ip4 = NULL;
2680 		}
2681 	}
2682 	return (0);
2683 }
2684 
2685 /*
2686  * Pass back primary IPv4 address of this jail.
2687  *
2688  * If not restricted return success but do not alter the address.  Caller has
2689  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2690  *
2691  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2692  * Address returned in NBO.
2693  */
2694 int
2695 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2696 {
2697 	struct prison *pr;
2698 
2699 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2700 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2701 
2702 	pr = cred->cr_prison;
2703 	if (!(pr->pr_flags & PR_IP4))
2704 		return (0);
2705 	mtx_lock(&pr->pr_mtx);
2706 	if (!(pr->pr_flags & PR_IP4)) {
2707 		mtx_unlock(&pr->pr_mtx);
2708 		return (0);
2709 	}
2710 	if (pr->pr_ip4 == NULL) {
2711 		mtx_unlock(&pr->pr_mtx);
2712 		return (EAFNOSUPPORT);
2713 	}
2714 
2715 	ia->s_addr = pr->pr_ip4[0].s_addr;
2716 	mtx_unlock(&pr->pr_mtx);
2717 	return (0);
2718 }
2719 
2720 /*
2721  * Return 1 if we should do proper source address selection or are not jailed.
2722  * We will return 0 if we should bypass source address selection in favour
2723  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2724  * returned in NBO.
2725  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2726  */
2727 int
2728 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2729 {
2730 	struct prison *pr;
2731 	struct in_addr lia;
2732 	int error;
2733 
2734 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2735 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2736 
2737 	if (!jailed(cred))
2738 		return (1);
2739 
2740 	pr = cred->cr_prison;
2741 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2742 		return (1);
2743 
2744 	lia.s_addr = INADDR_ANY;
2745 	error = prison_get_ip4(cred, &lia);
2746 	if (error)
2747 		return (error);
2748 	if (lia.s_addr == INADDR_ANY)
2749 		return (1);
2750 
2751 	ia->s_addr = lia.s_addr;
2752 	return (0);
2753 }
2754 
2755 /*
2756  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2757  */
2758 int
2759 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2760 {
2761 
2762 	if (pr1 == pr2)
2763 		return (1);
2764 
2765 	/*
2766 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2767 	 * existing prisons.
2768 	 */
2769 	while (pr1 != &prison0 &&
2770 #ifdef VIMAGE
2771 	       !(pr1->pr_flags & PR_VNET) &&
2772 #endif
2773 	       !(pr1->pr_flags & PR_IP4_USER))
2774 		pr1 = pr1->pr_parent;
2775 	while (pr2 != &prison0 &&
2776 #ifdef VIMAGE
2777 	       !(pr2->pr_flags & PR_VNET) &&
2778 #endif
2779 	       !(pr2->pr_flags & PR_IP4_USER))
2780 		pr2 = pr2->pr_parent;
2781 	return (pr1 == pr2);
2782 }
2783 
2784 /*
2785  * Make sure our (source) address is set to something meaningful to this
2786  * jail.
2787  *
2788  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2789  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2790  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2791  */
2792 int
2793 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2794 {
2795 	struct prison *pr;
2796 	struct in_addr ia0;
2797 	int error;
2798 
2799 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2800 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2801 
2802 	pr = cred->cr_prison;
2803 	if (!(pr->pr_flags & PR_IP4))
2804 		return (0);
2805 	mtx_lock(&pr->pr_mtx);
2806 	if (!(pr->pr_flags & PR_IP4)) {
2807 		mtx_unlock(&pr->pr_mtx);
2808 		return (0);
2809 	}
2810 	if (pr->pr_ip4 == NULL) {
2811 		mtx_unlock(&pr->pr_mtx);
2812 		return (EAFNOSUPPORT);
2813 	}
2814 
2815 	ia0.s_addr = ntohl(ia->s_addr);
2816 	if (ia0.s_addr == INADDR_LOOPBACK) {
2817 		ia->s_addr = pr->pr_ip4[0].s_addr;
2818 		mtx_unlock(&pr->pr_mtx);
2819 		return (0);
2820 	}
2821 
2822 	if (ia0.s_addr == INADDR_ANY) {
2823 		/*
2824 		 * In case there is only 1 IPv4 address, bind directly.
2825 		 */
2826 		if (pr->pr_ip4s == 1)
2827 			ia->s_addr = pr->pr_ip4[0].s_addr;
2828 		mtx_unlock(&pr->pr_mtx);
2829 		return (0);
2830 	}
2831 
2832 	error = _prison_check_ip4(pr, ia);
2833 	mtx_unlock(&pr->pr_mtx);
2834 	return (error);
2835 }
2836 
2837 /*
2838  * Rewrite destination address in case we will connect to loopback address.
2839  *
2840  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2841  * Address passed in in NBO and returned in NBO.
2842  */
2843 int
2844 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2845 {
2846 	struct prison *pr;
2847 
2848 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2849 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2850 
2851 	pr = cred->cr_prison;
2852 	if (!(pr->pr_flags & PR_IP4))
2853 		return (0);
2854 	mtx_lock(&pr->pr_mtx);
2855 	if (!(pr->pr_flags & PR_IP4)) {
2856 		mtx_unlock(&pr->pr_mtx);
2857 		return (0);
2858 	}
2859 	if (pr->pr_ip4 == NULL) {
2860 		mtx_unlock(&pr->pr_mtx);
2861 		return (EAFNOSUPPORT);
2862 	}
2863 
2864 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2865 		ia->s_addr = pr->pr_ip4[0].s_addr;
2866 		mtx_unlock(&pr->pr_mtx);
2867 		return (0);
2868 	}
2869 
2870 	/*
2871 	 * Return success because nothing had to be changed.
2872 	 */
2873 	mtx_unlock(&pr->pr_mtx);
2874 	return (0);
2875 }
2876 
2877 /*
2878  * Check if given address belongs to the jail referenced by cred/prison.
2879  *
2880  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2881  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2882  * doesn't allow IPv4.  Address passed in in NBO.
2883  */
2884 static int
2885 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
2886 {
2887 	int i, a, z, d;
2888 
2889 	/*
2890 	 * Check the primary IP.
2891 	 */
2892 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2893 		return (0);
2894 
2895 	/*
2896 	 * All the other IPs are sorted so we can do a binary search.
2897 	 */
2898 	a = 0;
2899 	z = pr->pr_ip4s - 2;
2900 	while (a <= z) {
2901 		i = (a + z) / 2;
2902 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2903 		if (d > 0)
2904 			z = i - 1;
2905 		else if (d < 0)
2906 			a = i + 1;
2907 		else
2908 			return (0);
2909 	}
2910 
2911 	return (EADDRNOTAVAIL);
2912 }
2913 
2914 int
2915 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
2916 {
2917 	struct prison *pr;
2918 	int error;
2919 
2920 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2921 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2922 
2923 	pr = cred->cr_prison;
2924 	if (!(pr->pr_flags & PR_IP4))
2925 		return (0);
2926 	mtx_lock(&pr->pr_mtx);
2927 	if (!(pr->pr_flags & PR_IP4)) {
2928 		mtx_unlock(&pr->pr_mtx);
2929 		return (0);
2930 	}
2931 	if (pr->pr_ip4 == NULL) {
2932 		mtx_unlock(&pr->pr_mtx);
2933 		return (EAFNOSUPPORT);
2934 	}
2935 
2936 	error = _prison_check_ip4(pr, ia);
2937 	mtx_unlock(&pr->pr_mtx);
2938 	return (error);
2939 }
2940 #endif
2941 
2942 #ifdef INET6
2943 static int
2944 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2945 {
2946 	int ii, ij, used;
2947 	struct prison *ppr;
2948 
2949 	ppr = pr->pr_parent;
2950 	if (!(pr->pr_flags & PR_IP6_USER)) {
2951 		/* This has no user settings, so just copy the parent's list. */
2952 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2953 			/*
2954 			 * There's no room for the parent's list.  Use the
2955 			 * new list buffer, which is assumed to be big enough
2956 			 * (if it was passed).  If there's no buffer, try to
2957 			 * allocate one.
2958 			 */
2959 			used = 1;
2960 			if (newip6 == NULL) {
2961 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
2962 				    M_PRISON, M_NOWAIT);
2963 				if (newip6 != NULL)
2964 					used = 0;
2965 			}
2966 			if (newip6 != NULL) {
2967 				bcopy(ppr->pr_ip6, newip6,
2968 				    ppr->pr_ip6s * sizeof(*newip6));
2969 				free(pr->pr_ip6, M_PRISON);
2970 				pr->pr_ip6 = newip6;
2971 				pr->pr_ip6s = ppr->pr_ip6s;
2972 			}
2973 			return (used);
2974 		}
2975 		pr->pr_ip6s = ppr->pr_ip6s;
2976 		if (pr->pr_ip6s > 0)
2977 			bcopy(ppr->pr_ip6, pr->pr_ip6,
2978 			    pr->pr_ip6s * sizeof(*newip6));
2979 		else if (pr->pr_ip6 != NULL) {
2980 			free(pr->pr_ip6, M_PRISON);
2981 			pr->pr_ip6 = NULL;
2982 		}
2983 	} else if (pr->pr_ip6s > 0) {
2984 		/* Remove addresses that aren't in the parent. */
2985 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
2986 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
2987 			    &ppr->pr_ip6[ij]))
2988 				break;
2989 		if (ij < ppr->pr_ip6s)
2990 			ii = 1;
2991 		else {
2992 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
2993 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
2994 			ii = 0;
2995 		}
2996 		for (ij = 1; ii < pr->pr_ip6s; ) {
2997 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
2998 			    &ppr->pr_ip6[0])) {
2999 				ii++;
3000 				continue;
3001 			}
3002 			switch (ij >= ppr->pr_ip4s ? -1 :
3003 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3004 			case -1:
3005 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3006 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3007 				break;
3008 			case 0:
3009 				ii++;
3010 				ij++;
3011 				break;
3012 			case 1:
3013 				ij++;
3014 				break;
3015 			}
3016 		}
3017 		if (pr->pr_ip6s == 0) {
3018 			pr->pr_flags |= PR_IP6_DISABLE;
3019 			free(pr->pr_ip6, M_PRISON);
3020 			pr->pr_ip6 = NULL;
3021 		}
3022 	}
3023 	return 0;
3024 }
3025 
3026 /*
3027  * Pass back primary IPv6 address for this jail.
3028  *
3029  * If not restricted return success but do not alter the address.  Caller has
3030  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3031  *
3032  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3033  */
3034 int
3035 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3036 {
3037 	struct prison *pr;
3038 
3039 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3040 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3041 
3042 	pr = cred->cr_prison;
3043 	if (!(pr->pr_flags & PR_IP6))
3044 		return (0);
3045 	mtx_lock(&pr->pr_mtx);
3046 	if (!(pr->pr_flags & PR_IP6)) {
3047 		mtx_unlock(&pr->pr_mtx);
3048 		return (0);
3049 	}
3050 	if (pr->pr_ip6 == NULL) {
3051 		mtx_unlock(&pr->pr_mtx);
3052 		return (EAFNOSUPPORT);
3053 	}
3054 
3055 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3056 	mtx_unlock(&pr->pr_mtx);
3057 	return (0);
3058 }
3059 
3060 /*
3061  * Return 1 if we should do proper source address selection or are not jailed.
3062  * We will return 0 if we should bypass source address selection in favour
3063  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3064  * returned in NBO.
3065  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3066  */
3067 int
3068 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3069 {
3070 	struct prison *pr;
3071 	struct in6_addr lia6;
3072 	int error;
3073 
3074 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3075 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3076 
3077 	if (!jailed(cred))
3078 		return (1);
3079 
3080 	pr = cred->cr_prison;
3081 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3082 		return (1);
3083 
3084 	lia6 = in6addr_any;
3085 	error = prison_get_ip6(cred, &lia6);
3086 	if (error)
3087 		return (error);
3088 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3089 		return (1);
3090 
3091 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3092 	return (0);
3093 }
3094 
3095 /*
3096  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3097  */
3098 int
3099 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3100 {
3101 
3102 	if (pr1 == pr2)
3103 		return (1);
3104 
3105 	while (pr1 != &prison0 &&
3106 #ifdef VIMAGE
3107 	       !(pr1->pr_flags & PR_VNET) &&
3108 #endif
3109 	       !(pr1->pr_flags & PR_IP6_USER))
3110 		pr1 = pr1->pr_parent;
3111 	while (pr2 != &prison0 &&
3112 #ifdef VIMAGE
3113 	       !(pr2->pr_flags & PR_VNET) &&
3114 #endif
3115 	       !(pr2->pr_flags & PR_IP6_USER))
3116 		pr2 = pr2->pr_parent;
3117 	return (pr1 == pr2);
3118 }
3119 
3120 /*
3121  * Make sure our (source) address is set to something meaningful to this jail.
3122  *
3123  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3124  * when needed while binding.
3125  *
3126  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3127  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3128  * doesn't allow IPv6.
3129  */
3130 int
3131 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3132 {
3133 	struct prison *pr;
3134 	int error;
3135 
3136 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3137 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3138 
3139 	pr = cred->cr_prison;
3140 	if (!(pr->pr_flags & PR_IP6))
3141 		return (0);
3142 	mtx_lock(&pr->pr_mtx);
3143 	if (!(pr->pr_flags & PR_IP6)) {
3144 		mtx_unlock(&pr->pr_mtx);
3145 		return (0);
3146 	}
3147 	if (pr->pr_ip6 == NULL) {
3148 		mtx_unlock(&pr->pr_mtx);
3149 		return (EAFNOSUPPORT);
3150 	}
3151 
3152 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3153 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3154 		mtx_unlock(&pr->pr_mtx);
3155 		return (0);
3156 	}
3157 
3158 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3159 		/*
3160 		 * In case there is only 1 IPv6 address, and v6only is true,
3161 		 * then bind directly.
3162 		 */
3163 		if (v6only != 0 && pr->pr_ip6s == 1)
3164 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3165 		mtx_unlock(&pr->pr_mtx);
3166 		return (0);
3167 	}
3168 
3169 	error = _prison_check_ip6(pr, ia6);
3170 	mtx_unlock(&pr->pr_mtx);
3171 	return (error);
3172 }
3173 
3174 /*
3175  * Rewrite destination address in case we will connect to loopback address.
3176  *
3177  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3178  */
3179 int
3180 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3181 {
3182 	struct prison *pr;
3183 
3184 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3185 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3186 
3187 	pr = cred->cr_prison;
3188 	if (!(pr->pr_flags & PR_IP6))
3189 		return (0);
3190 	mtx_lock(&pr->pr_mtx);
3191 	if (!(pr->pr_flags & PR_IP6)) {
3192 		mtx_unlock(&pr->pr_mtx);
3193 		return (0);
3194 	}
3195 	if (pr->pr_ip6 == NULL) {
3196 		mtx_unlock(&pr->pr_mtx);
3197 		return (EAFNOSUPPORT);
3198 	}
3199 
3200 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3201 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3202 		mtx_unlock(&pr->pr_mtx);
3203 		return (0);
3204 	}
3205 
3206 	/*
3207 	 * Return success because nothing had to be changed.
3208 	 */
3209 	mtx_unlock(&pr->pr_mtx);
3210 	return (0);
3211 }
3212 
3213 /*
3214  * Check if given address belongs to the jail referenced by cred/prison.
3215  *
3216  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3217  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3218  * doesn't allow IPv6.
3219  */
3220 static int
3221 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3222 {
3223 	int i, a, z, d;
3224 
3225 	/*
3226 	 * Check the primary IP.
3227 	 */
3228 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3229 		return (0);
3230 
3231 	/*
3232 	 * All the other IPs are sorted so we can do a binary search.
3233 	 */
3234 	a = 0;
3235 	z = pr->pr_ip6s - 2;
3236 	while (a <= z) {
3237 		i = (a + z) / 2;
3238 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3239 		if (d > 0)
3240 			z = i - 1;
3241 		else if (d < 0)
3242 			a = i + 1;
3243 		else
3244 			return (0);
3245 	}
3246 
3247 	return (EADDRNOTAVAIL);
3248 }
3249 
3250 int
3251 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3252 {
3253 	struct prison *pr;
3254 	int error;
3255 
3256 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3257 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3258 
3259 	pr = cred->cr_prison;
3260 	if (!(pr->pr_flags & PR_IP6))
3261 		return (0);
3262 	mtx_lock(&pr->pr_mtx);
3263 	if (!(pr->pr_flags & PR_IP6)) {
3264 		mtx_unlock(&pr->pr_mtx);
3265 		return (0);
3266 	}
3267 	if (pr->pr_ip6 == NULL) {
3268 		mtx_unlock(&pr->pr_mtx);
3269 		return (EAFNOSUPPORT);
3270 	}
3271 
3272 	error = _prison_check_ip6(pr, ia6);
3273 	mtx_unlock(&pr->pr_mtx);
3274 	return (error);
3275 }
3276 #endif
3277 
3278 /*
3279  * Check if a jail supports the given address family.
3280  *
3281  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3282  * if not.
3283  */
3284 int
3285 prison_check_af(struct ucred *cred, int af)
3286 {
3287 	struct prison *pr;
3288 	int error;
3289 
3290 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3291 
3292 	pr = cred->cr_prison;
3293 #ifdef VIMAGE
3294 	/* Prisons with their own network stack are not limited. */
3295 	if (prison_owns_vnet(cred))
3296 		return (0);
3297 #endif
3298 
3299 	error = 0;
3300 	switch (af)
3301 	{
3302 #ifdef INET
3303 	case AF_INET:
3304 		if (pr->pr_flags & PR_IP4)
3305 		{
3306 			mtx_lock(&pr->pr_mtx);
3307 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3308 				error = EAFNOSUPPORT;
3309 			mtx_unlock(&pr->pr_mtx);
3310 		}
3311 		break;
3312 #endif
3313 #ifdef INET6
3314 	case AF_INET6:
3315 		if (pr->pr_flags & PR_IP6)
3316 		{
3317 			mtx_lock(&pr->pr_mtx);
3318 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3319 				error = EAFNOSUPPORT;
3320 			mtx_unlock(&pr->pr_mtx);
3321 		}
3322 		break;
3323 #endif
3324 	case AF_LOCAL:
3325 	case AF_ROUTE:
3326 		break;
3327 	default:
3328 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3329 			error = EAFNOSUPPORT;
3330 	}
3331 	return (error);
3332 }
3333 
3334 /*
3335  * Check if given address belongs to the jail referenced by cred (wrapper to
3336  * prison_check_ip[46]).
3337  *
3338  * Returns 0 if jail doesn't restrict the address family or if address belongs
3339  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3340  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3341  */
3342 int
3343 prison_if(struct ucred *cred, struct sockaddr *sa)
3344 {
3345 #ifdef INET
3346 	struct sockaddr_in *sai;
3347 #endif
3348 #ifdef INET6
3349 	struct sockaddr_in6 *sai6;
3350 #endif
3351 	int error;
3352 
3353 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3354 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3355 
3356 #ifdef VIMAGE
3357 	if (prison_owns_vnet(cred))
3358 		return (0);
3359 #endif
3360 
3361 	error = 0;
3362 	switch (sa->sa_family)
3363 	{
3364 #ifdef INET
3365 	case AF_INET:
3366 		sai = (struct sockaddr_in *)sa;
3367 		error = prison_check_ip4(cred, &sai->sin_addr);
3368 		break;
3369 #endif
3370 #ifdef INET6
3371 	case AF_INET6:
3372 		sai6 = (struct sockaddr_in6 *)sa;
3373 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3374 		break;
3375 #endif
3376 	default:
3377 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3378 			error = EAFNOSUPPORT;
3379 	}
3380 	return (error);
3381 }
3382 
3383 /*
3384  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3385  */
3386 int
3387 prison_check(struct ucred *cred1, struct ucred *cred2)
3388 {
3389 
3390 	return ((cred1->cr_prison == cred2->cr_prison ||
3391 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3392 }
3393 
3394 /*
3395  * Return 1 if p2 is a child of p1, otherwise 0.
3396  */
3397 int
3398 prison_ischild(struct prison *pr1, struct prison *pr2)
3399 {
3400 
3401 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3402 		if (pr1 == pr2)
3403 			return (1);
3404 	return (0);
3405 }
3406 
3407 /*
3408  * Return 1 if the passed credential is in a jail, otherwise 0.
3409  */
3410 int
3411 jailed(struct ucred *cred)
3412 {
3413 
3414 	return (cred->cr_prison != &prison0);
3415 }
3416 
3417 /*
3418  * Return 1 if the passed credential is in a jail and that jail does not
3419  * have its own virtual network stack, otherwise 0.
3420  */
3421 int
3422 jailed_without_vnet(struct ucred *cred)
3423 {
3424 
3425 	if (!jailed(cred))
3426 		return (0);
3427 #ifdef VIMAGE
3428 	if (prison_owns_vnet(cred))
3429 		return (0);
3430 #endif
3431 
3432 	return (1);
3433 }
3434 
3435 /*
3436  * Return the correct hostname (domainname, et al) for the passed credential.
3437  */
3438 void
3439 getcredhostname(struct ucred *cred, char *buf, size_t size)
3440 {
3441 	struct prison *pr;
3442 
3443 	/*
3444 	 * A NULL credential can be used to shortcut to the physical
3445 	 * system's hostname.
3446 	 */
3447 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3448 	mtx_lock(&pr->pr_mtx);
3449 	strlcpy(buf, pr->pr_hostname, size);
3450 	mtx_unlock(&pr->pr_mtx);
3451 }
3452 
3453 void
3454 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3455 {
3456 
3457 	mtx_lock(&cred->cr_prison->pr_mtx);
3458 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3459 	mtx_unlock(&cred->cr_prison->pr_mtx);
3460 }
3461 
3462 void
3463 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3464 {
3465 
3466 	mtx_lock(&cred->cr_prison->pr_mtx);
3467 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3468 	mtx_unlock(&cred->cr_prison->pr_mtx);
3469 }
3470 
3471 void
3472 getcredhostid(struct ucred *cred, unsigned long *hostid)
3473 {
3474 
3475 	mtx_lock(&cred->cr_prison->pr_mtx);
3476 	*hostid = cred->cr_prison->pr_hostid;
3477 	mtx_unlock(&cred->cr_prison->pr_mtx);
3478 }
3479 
3480 #ifdef VIMAGE
3481 /*
3482  * Determine whether the prison represented by cred owns
3483  * its vnet rather than having it inherited.
3484  *
3485  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3486  */
3487 int
3488 prison_owns_vnet(struct ucred *cred)
3489 {
3490 
3491 	/*
3492 	 * vnets cannot be added/removed after jail creation,
3493 	 * so no need to lock here.
3494 	 */
3495 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3496 }
3497 #endif
3498 
3499 /*
3500  * Determine whether the subject represented by cred can "see"
3501  * status of a mount point.
3502  * Returns: 0 for permitted, ENOENT otherwise.
3503  * XXX: This function should be called cr_canseemount() and should be
3504  *      placed in kern_prot.c.
3505  */
3506 int
3507 prison_canseemount(struct ucred *cred, struct mount *mp)
3508 {
3509 	struct prison *pr;
3510 	struct statfs *sp;
3511 	size_t len;
3512 
3513 	pr = cred->cr_prison;
3514 	if (pr->pr_enforce_statfs == 0)
3515 		return (0);
3516 	if (pr->pr_root->v_mount == mp)
3517 		return (0);
3518 	if (pr->pr_enforce_statfs == 2)
3519 		return (ENOENT);
3520 	/*
3521 	 * If jail's chroot directory is set to "/" we should be able to see
3522 	 * all mount-points from inside a jail.
3523 	 * This is ugly check, but this is the only situation when jail's
3524 	 * directory ends with '/'.
3525 	 */
3526 	if (strcmp(pr->pr_path, "/") == 0)
3527 		return (0);
3528 	len = strlen(pr->pr_path);
3529 	sp = &mp->mnt_stat;
3530 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3531 		return (ENOENT);
3532 	/*
3533 	 * Be sure that we don't have situation where jail's root directory
3534 	 * is "/some/path" and mount point is "/some/pathpath".
3535 	 */
3536 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3537 		return (ENOENT);
3538 	return (0);
3539 }
3540 
3541 void
3542 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3543 {
3544 	char jpath[MAXPATHLEN];
3545 	struct prison *pr;
3546 	size_t len;
3547 
3548 	pr = cred->cr_prison;
3549 	if (pr->pr_enforce_statfs == 0)
3550 		return;
3551 	if (prison_canseemount(cred, mp) != 0) {
3552 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3553 		strlcpy(sp->f_mntonname, "[restricted]",
3554 		    sizeof(sp->f_mntonname));
3555 		return;
3556 	}
3557 	if (pr->pr_root->v_mount == mp) {
3558 		/*
3559 		 * Clear current buffer data, so we are sure nothing from
3560 		 * the valid path left there.
3561 		 */
3562 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3563 		*sp->f_mntonname = '/';
3564 		return;
3565 	}
3566 	/*
3567 	 * If jail's chroot directory is set to "/" we should be able to see
3568 	 * all mount-points from inside a jail.
3569 	 */
3570 	if (strcmp(pr->pr_path, "/") == 0)
3571 		return;
3572 	len = strlen(pr->pr_path);
3573 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3574 	/*
3575 	 * Clear current buffer data, so we are sure nothing from
3576 	 * the valid path left there.
3577 	 */
3578 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3579 	if (*jpath == '\0') {
3580 		/* Should never happen. */
3581 		*sp->f_mntonname = '/';
3582 	} else {
3583 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3584 	}
3585 }
3586 
3587 /*
3588  * Check with permission for a specific privilege is granted within jail.  We
3589  * have a specific list of accepted privileges; the rest are denied.
3590  */
3591 int
3592 prison_priv_check(struct ucred *cred, int priv)
3593 {
3594 
3595 	if (!jailed(cred))
3596 		return (0);
3597 
3598 #ifdef VIMAGE
3599 	/*
3600 	 * Privileges specific to prisons with a virtual network stack.
3601 	 * There might be a duplicate entry here in case the privilege
3602 	 * is only granted conditionally in the legacy jail case.
3603 	 */
3604 	switch (priv) {
3605 #ifdef notyet
3606 		/*
3607 		 * NFS-specific privileges.
3608 		 */
3609 	case PRIV_NFS_DAEMON:
3610 	case PRIV_NFS_LOCKD:
3611 #endif
3612 		/*
3613 		 * Network stack privileges.
3614 		 */
3615 	case PRIV_NET_BRIDGE:
3616 	case PRIV_NET_GRE:
3617 	case PRIV_NET_BPF:
3618 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3619 	case PRIV_NET_ROUTE:
3620 	case PRIV_NET_TAP:
3621 	case PRIV_NET_SETIFMTU:
3622 	case PRIV_NET_SETIFFLAGS:
3623 	case PRIV_NET_SETIFCAP:
3624 	case PRIV_NET_SETIFDESCR:
3625 	case PRIV_NET_SETIFNAME	:
3626 	case PRIV_NET_SETIFMETRIC:
3627 	case PRIV_NET_SETIFPHYS:
3628 	case PRIV_NET_SETIFMAC:
3629 	case PRIV_NET_ADDMULTI:
3630 	case PRIV_NET_DELMULTI:
3631 	case PRIV_NET_HWIOCTL:
3632 	case PRIV_NET_SETLLADDR:
3633 	case PRIV_NET_ADDIFGROUP:
3634 	case PRIV_NET_DELIFGROUP:
3635 	case PRIV_NET_IFCREATE:
3636 	case PRIV_NET_IFDESTROY:
3637 	case PRIV_NET_ADDIFADDR:
3638 	case PRIV_NET_DELIFADDR:
3639 	case PRIV_NET_LAGG:
3640 	case PRIV_NET_GIF:
3641 	case PRIV_NET_SETIFVNET:
3642 
3643 		/*
3644 		 * 802.11-related privileges.
3645 		 */
3646 	case PRIV_NET80211_GETKEY:
3647 #ifdef notyet
3648 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3649 #endif
3650 
3651 #ifdef notyet
3652 		/*
3653 		 * AppleTalk privileges.
3654 		 */
3655 	case PRIV_NETATALK_RESERVEDPORT:
3656 
3657 		/*
3658 		 * ATM privileges.
3659 		 */
3660 	case PRIV_NETATM_CFG:
3661 	case PRIV_NETATM_ADD:
3662 	case PRIV_NETATM_DEL:
3663 	case PRIV_NETATM_SET:
3664 
3665 		/*
3666 		 * Bluetooth privileges.
3667 		 */
3668 	case PRIV_NETBLUETOOTH_RAW:
3669 #endif
3670 
3671 		/*
3672 		 * Netgraph and netgraph module privileges.
3673 		 */
3674 	case PRIV_NETGRAPH_CONTROL:
3675 #ifdef notyet
3676 	case PRIV_NETGRAPH_TTY:
3677 #endif
3678 
3679 		/*
3680 		 * IPv4 and IPv6 privileges.
3681 		 */
3682 	case PRIV_NETINET_IPFW:
3683 	case PRIV_NETINET_DIVERT:
3684 	case PRIV_NETINET_PF:
3685 	case PRIV_NETINET_DUMMYNET:
3686 	case PRIV_NETINET_CARP:
3687 	case PRIV_NETINET_MROUTE:
3688 	case PRIV_NETINET_RAW:
3689 	case PRIV_NETINET_ADDRCTRL6:
3690 	case PRIV_NETINET_ND6:
3691 	case PRIV_NETINET_SCOPE6:
3692 	case PRIV_NETINET_ALIFETIME6:
3693 	case PRIV_NETINET_IPSEC:
3694 	case PRIV_NETINET_BINDANY:
3695 
3696 #ifdef notyet
3697 		/*
3698 		 * IPX/SPX privileges.
3699 		 */
3700 	case PRIV_NETIPX_RESERVEDPORT:
3701 	case PRIV_NETIPX_RAW:
3702 
3703 		/*
3704 		 * NCP privileges.
3705 		 */
3706 	case PRIV_NETNCP:
3707 
3708 		/*
3709 		 * SMB privileges.
3710 		 */
3711 	case PRIV_NETSMB:
3712 #endif
3713 
3714 	/*
3715 	 * No default: or deny here.
3716 	 * In case of no permit fall through to next switch().
3717 	 */
3718 		if (cred->cr_prison->pr_flags & PR_VNET)
3719 			return (0);
3720 	}
3721 #endif /* VIMAGE */
3722 
3723 	switch (priv) {
3724 
3725 		/*
3726 		 * Allow ktrace privileges for root in jail.
3727 		 */
3728 	case PRIV_KTRACE:
3729 
3730 #if 0
3731 		/*
3732 		 * Allow jailed processes to configure audit identity and
3733 		 * submit audit records (login, etc).  In the future we may
3734 		 * want to further refine the relationship between audit and
3735 		 * jail.
3736 		 */
3737 	case PRIV_AUDIT_GETAUDIT:
3738 	case PRIV_AUDIT_SETAUDIT:
3739 	case PRIV_AUDIT_SUBMIT:
3740 #endif
3741 
3742 		/*
3743 		 * Allow jailed processes to manipulate process UNIX
3744 		 * credentials in any way they see fit.
3745 		 */
3746 	case PRIV_CRED_SETUID:
3747 	case PRIV_CRED_SETEUID:
3748 	case PRIV_CRED_SETGID:
3749 	case PRIV_CRED_SETEGID:
3750 	case PRIV_CRED_SETGROUPS:
3751 	case PRIV_CRED_SETREUID:
3752 	case PRIV_CRED_SETREGID:
3753 	case PRIV_CRED_SETRESUID:
3754 	case PRIV_CRED_SETRESGID:
3755 
3756 		/*
3757 		 * Jail implements visibility constraints already, so allow
3758 		 * jailed root to override uid/gid-based constraints.
3759 		 */
3760 	case PRIV_SEEOTHERGIDS:
3761 	case PRIV_SEEOTHERUIDS:
3762 
3763 		/*
3764 		 * Jail implements inter-process debugging limits already, so
3765 		 * allow jailed root various debugging privileges.
3766 		 */
3767 	case PRIV_DEBUG_DIFFCRED:
3768 	case PRIV_DEBUG_SUGID:
3769 	case PRIV_DEBUG_UNPRIV:
3770 
3771 		/*
3772 		 * Allow jail to set various resource limits and login
3773 		 * properties, and for now, exceed process resource limits.
3774 		 */
3775 	case PRIV_PROC_LIMIT:
3776 	case PRIV_PROC_SETLOGIN:
3777 	case PRIV_PROC_SETRLIMIT:
3778 
3779 		/*
3780 		 * System V and POSIX IPC privileges are granted in jail.
3781 		 */
3782 	case PRIV_IPC_READ:
3783 	case PRIV_IPC_WRITE:
3784 	case PRIV_IPC_ADMIN:
3785 	case PRIV_IPC_MSGSIZE:
3786 	case PRIV_MQ_ADMIN:
3787 
3788 		/*
3789 		 * Jail operations within a jail work on child jails.
3790 		 */
3791 	case PRIV_JAIL_ATTACH:
3792 	case PRIV_JAIL_SET:
3793 	case PRIV_JAIL_REMOVE:
3794 
3795 		/*
3796 		 * Jail implements its own inter-process limits, so allow
3797 		 * root processes in jail to change scheduling on other
3798 		 * processes in the same jail.  Likewise for signalling.
3799 		 */
3800 	case PRIV_SCHED_DIFFCRED:
3801 	case PRIV_SCHED_CPUSET:
3802 	case PRIV_SIGNAL_DIFFCRED:
3803 	case PRIV_SIGNAL_SUGID:
3804 
3805 		/*
3806 		 * Allow jailed processes to write to sysctls marked as jail
3807 		 * writable.
3808 		 */
3809 	case PRIV_SYSCTL_WRITEJAIL:
3810 
3811 		/*
3812 		 * Allow root in jail to manage a variety of quota
3813 		 * properties.  These should likely be conditional on a
3814 		 * configuration option.
3815 		 */
3816 	case PRIV_VFS_GETQUOTA:
3817 	case PRIV_VFS_SETQUOTA:
3818 
3819 		/*
3820 		 * Since Jail relies on chroot() to implement file system
3821 		 * protections, grant many VFS privileges to root in jail.
3822 		 * Be careful to exclude mount-related and NFS-related
3823 		 * privileges.
3824 		 */
3825 	case PRIV_VFS_READ:
3826 	case PRIV_VFS_WRITE:
3827 	case PRIV_VFS_ADMIN:
3828 	case PRIV_VFS_EXEC:
3829 	case PRIV_VFS_LOOKUP:
3830 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3831 	case PRIV_VFS_CHFLAGS_DEV:
3832 	case PRIV_VFS_CHOWN:
3833 	case PRIV_VFS_CHROOT:
3834 	case PRIV_VFS_RETAINSUGID:
3835 	case PRIV_VFS_FCHROOT:
3836 	case PRIV_VFS_LINK:
3837 	case PRIV_VFS_SETGID:
3838 	case PRIV_VFS_STAT:
3839 	case PRIV_VFS_STICKYFILE:
3840 		return (0);
3841 
3842 		/*
3843 		 * Depending on the global setting, allow privilege of
3844 		 * setting system flags.
3845 		 */
3846 	case PRIV_VFS_SYSFLAGS:
3847 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3848 			return (0);
3849 		else
3850 			return (EPERM);
3851 
3852 		/*
3853 		 * Depending on the global setting, allow privilege of
3854 		 * mounting/unmounting file systems.
3855 		 */
3856 	case PRIV_VFS_MOUNT:
3857 	case PRIV_VFS_UNMOUNT:
3858 	case PRIV_VFS_MOUNT_NONUSER:
3859 	case PRIV_VFS_MOUNT_OWNER:
3860 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT)
3861 			return (0);
3862 		else
3863 			return (EPERM);
3864 
3865 		/*
3866 		 * Allow jailed root to bind reserved ports and reuse in-use
3867 		 * ports.
3868 		 */
3869 	case PRIV_NETINET_RESERVEDPORT:
3870 	case PRIV_NETINET_REUSEPORT:
3871 		return (0);
3872 
3873 		/*
3874 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3875 		 */
3876 	case PRIV_NETINET_SETHDROPTS:
3877 		return (0);
3878 
3879 		/*
3880 		 * Conditionally allow creating raw sockets in jail.
3881 		 */
3882 	case PRIV_NETINET_RAW:
3883 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3884 			return (0);
3885 		else
3886 			return (EPERM);
3887 
3888 		/*
3889 		 * Since jail implements its own visibility limits on netstat
3890 		 * sysctls, allow getcred.  This allows identd to work in
3891 		 * jail.
3892 		 */
3893 	case PRIV_NETINET_GETCRED:
3894 		return (0);
3895 
3896 		/*
3897 		 * Allow jailed root to set loginclass.
3898 		 */
3899 	case PRIV_PROC_SETLOGINCLASS:
3900 		return (0);
3901 
3902 	default:
3903 		/*
3904 		 * In all remaining cases, deny the privilege request.  This
3905 		 * includes almost all network privileges, many system
3906 		 * configuration privileges.
3907 		 */
3908 		return (EPERM);
3909 	}
3910 }
3911 
3912 /*
3913  * Return the part of pr2's name that is relative to pr1, or the whole name
3914  * if it does not directly follow.
3915  */
3916 
3917 char *
3918 prison_name(struct prison *pr1, struct prison *pr2)
3919 {
3920 	char *name;
3921 
3922 	/* Jails see themselves as "0" (if they see themselves at all). */
3923 	if (pr1 == pr2)
3924 		return "0";
3925 	name = pr2->pr_name;
3926 	if (prison_ischild(pr1, pr2)) {
3927 		/*
3928 		 * pr1 isn't locked (and allprison_lock may not be either)
3929 		 * so its length can't be counted on.  But the number of dots
3930 		 * can be counted on - and counted.
3931 		 */
3932 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3933 			name = strchr(name, '.') + 1;
3934 	}
3935 	return (name);
3936 }
3937 
3938 /*
3939  * Return the part of pr2's path that is relative to pr1, or the whole path
3940  * if it does not directly follow.
3941  */
3942 static char *
3943 prison_path(struct prison *pr1, struct prison *pr2)
3944 {
3945 	char *path1, *path2;
3946 	int len1;
3947 
3948 	path1 = pr1->pr_path;
3949 	path2 = pr2->pr_path;
3950 	if (!strcmp(path1, "/"))
3951 		return (path2);
3952 	len1 = strlen(path1);
3953 	if (strncmp(path1, path2, len1))
3954 		return (path2);
3955 	if (path2[len1] == '\0')
3956 		return "/";
3957 	if (path2[len1] == '/')
3958 		return (path2 + len1);
3959 	return (path2);
3960 }
3961 
3962 
3963 /*
3964  * Jail-related sysctls.
3965  */
3966 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3967     "Jails");
3968 
3969 static int
3970 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3971 {
3972 	struct xprison *xp;
3973 	struct prison *pr, *cpr;
3974 #ifdef INET
3975 	struct in_addr *ip4 = NULL;
3976 	int ip4s = 0;
3977 #endif
3978 #ifdef INET6
3979 	struct in6_addr *ip6 = NULL;
3980 	int ip6s = 0;
3981 #endif
3982 	int descend, error;
3983 
3984 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3985 	pr = req->td->td_ucred->cr_prison;
3986 	error = 0;
3987 	sx_slock(&allprison_lock);
3988 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3989 #if defined(INET) || defined(INET6)
3990  again:
3991 #endif
3992 		mtx_lock(&cpr->pr_mtx);
3993 #ifdef INET
3994 		if (cpr->pr_ip4s > 0) {
3995 			if (ip4s < cpr->pr_ip4s) {
3996 				ip4s = cpr->pr_ip4s;
3997 				mtx_unlock(&cpr->pr_mtx);
3998 				ip4 = realloc(ip4, ip4s *
3999 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4000 				goto again;
4001 			}
4002 			bcopy(cpr->pr_ip4, ip4,
4003 			    cpr->pr_ip4s * sizeof(struct in_addr));
4004 		}
4005 #endif
4006 #ifdef INET6
4007 		if (cpr->pr_ip6s > 0) {
4008 			if (ip6s < cpr->pr_ip6s) {
4009 				ip6s = cpr->pr_ip6s;
4010 				mtx_unlock(&cpr->pr_mtx);
4011 				ip6 = realloc(ip6, ip6s *
4012 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4013 				goto again;
4014 			}
4015 			bcopy(cpr->pr_ip6, ip6,
4016 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4017 		}
4018 #endif
4019 		if (cpr->pr_ref == 0) {
4020 			mtx_unlock(&cpr->pr_mtx);
4021 			continue;
4022 		}
4023 		bzero(xp, sizeof(*xp));
4024 		xp->pr_version = XPRISON_VERSION;
4025 		xp->pr_id = cpr->pr_id;
4026 		xp->pr_state = cpr->pr_uref > 0
4027 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4028 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4029 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4030 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4031 #ifdef INET
4032 		xp->pr_ip4s = cpr->pr_ip4s;
4033 #endif
4034 #ifdef INET6
4035 		xp->pr_ip6s = cpr->pr_ip6s;
4036 #endif
4037 		mtx_unlock(&cpr->pr_mtx);
4038 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4039 		if (error)
4040 			break;
4041 #ifdef INET
4042 		if (xp->pr_ip4s > 0) {
4043 			error = SYSCTL_OUT(req, ip4,
4044 			    xp->pr_ip4s * sizeof(struct in_addr));
4045 			if (error)
4046 				break;
4047 		}
4048 #endif
4049 #ifdef INET6
4050 		if (xp->pr_ip6s > 0) {
4051 			error = SYSCTL_OUT(req, ip6,
4052 			    xp->pr_ip6s * sizeof(struct in6_addr));
4053 			if (error)
4054 				break;
4055 		}
4056 #endif
4057 	}
4058 	sx_sunlock(&allprison_lock);
4059 	free(xp, M_TEMP);
4060 #ifdef INET
4061 	free(ip4, M_TEMP);
4062 #endif
4063 #ifdef INET6
4064 	free(ip6, M_TEMP);
4065 #endif
4066 	return (error);
4067 }
4068 
4069 SYSCTL_OID(_security_jail, OID_AUTO, list,
4070     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4071     sysctl_jail_list, "S", "List of active jails");
4072 
4073 static int
4074 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4075 {
4076 	int error, injail;
4077 
4078 	injail = jailed(req->td->td_ucred);
4079 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4080 
4081 	return (error);
4082 }
4083 
4084 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4085     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4086     sysctl_jail_jailed, "I", "Process in jail?");
4087 
4088 #if defined(INET) || defined(INET6)
4089 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4090     &jail_max_af_ips, 0,
4091     "Number of IP addresses a jail may have at most per address family");
4092 #endif
4093 
4094 /*
4095  * Default parameters for jail(2) compatability.  For historical reasons,
4096  * the sysctl names have varying similarity to the parameter names.  Prisons
4097  * just see their own parameters, and can't change them.
4098  */
4099 static int
4100 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4101 {
4102 	struct prison *pr;
4103 	int allow, error, i;
4104 
4105 	pr = req->td->td_ucred->cr_prison;
4106 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4107 
4108 	/* Get the current flag value, and convert it to a boolean. */
4109 	i = (allow & arg2) ? 1 : 0;
4110 	if (arg1 != NULL)
4111 		i = !i;
4112 	error = sysctl_handle_int(oidp, &i, 0, req);
4113 	if (error || !req->newptr)
4114 		return (error);
4115 	i = i ? arg2 : 0;
4116 	if (arg1 != NULL)
4117 		i ^= arg2;
4118 	/*
4119 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4120 	 * for writing.
4121 	 */
4122 	mtx_lock(&prison0.pr_mtx);
4123 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4124 	mtx_unlock(&prison0.pr_mtx);
4125 	return (0);
4126 }
4127 
4128 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4129     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4130     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4131     "Processes in jail can set their hostnames");
4132 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4133     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4134     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4135     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4136 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4137     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4138     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4139     "Processes in jail can use System V IPC primitives");
4140 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4141     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4142     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4143     "Prison root can create raw sockets");
4144 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4145     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4146     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4147     "Processes in jail can alter system file flags");
4148 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4149     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4150     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4151     "Processes in jail can mount/unmount jail-friendly file systems");
4152 
4153 static int
4154 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4155 {
4156 	struct prison *pr;
4157 	int level, error;
4158 
4159 	pr = req->td->td_ucred->cr_prison;
4160 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4161 	error = sysctl_handle_int(oidp, &level, 0, req);
4162 	if (error || !req->newptr)
4163 		return (error);
4164 	*(int *)arg1 = level;
4165 	return (0);
4166 }
4167 
4168 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4169     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4170     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4171     sysctl_jail_default_level, "I",
4172     "Processes in jail cannot see all mounted file systems");
4173 
4174 /*
4175  * Nodes to describe jail parameters.  Maximum length of string parameters
4176  * is returned in the string itself, and the other parameters exist merely
4177  * to make themselves and their types known.
4178  */
4179 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4180     "Jail parameters");
4181 
4182 int
4183 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4184 {
4185 	int i;
4186 	long l;
4187 	size_t s;
4188 	char numbuf[12];
4189 
4190 	switch (oidp->oid_kind & CTLTYPE)
4191 	{
4192 	case CTLTYPE_LONG:
4193 	case CTLTYPE_ULONG:
4194 		l = 0;
4195 #ifdef SCTL_MASK32
4196 		if (!(req->flags & SCTL_MASK32))
4197 #endif
4198 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4199 	case CTLTYPE_INT:
4200 	case CTLTYPE_UINT:
4201 		i = 0;
4202 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4203 	case CTLTYPE_STRING:
4204 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4205 		return
4206 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4207 	case CTLTYPE_STRUCT:
4208 		s = (size_t)arg2;
4209 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4210 	}
4211 	return (0);
4212 }
4213 
4214 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4215 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4216 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4217 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4218 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4219     "I", "Jail secure level");
4220 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4221     "I", "Jail cannot see all mounted file systems");
4222 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4223     "B", "Jail persistence");
4224 #ifdef VIMAGE
4225 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4226     "E,jailsys", "Virtual network stack");
4227 #endif
4228 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4229     "B", "Jail is in the process of shutting down");
4230 
4231 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4232 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4233     "I", "Current number of child jails");
4234 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4235     "I", "Maximum number of child jails");
4236 
4237 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4238 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4239     "Jail hostname");
4240 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4241     "Jail NIS domainname");
4242 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4243     "Jail host UUID");
4244 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4245     "LU", "Jail host ID");
4246 
4247 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4248 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4249 
4250 #ifdef INET
4251 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4252     "Jail IPv4 address virtualization");
4253 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4254     "S,in_addr,a", "Jail IPv4 addresses");
4255 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4256     "B", "Do (not) use IPv4 source address selection rather than the "
4257     "primary jail IPv4 address.");
4258 #endif
4259 #ifdef INET6
4260 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4261     "Jail IPv6 address virtualization");
4262 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4263     "S,in6_addr,a", "Jail IPv6 addresses");
4264 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4265     "B", "Do (not) use IPv6 source address selection rather than the "
4266     "primary jail IPv6 address.");
4267 #endif
4268 
4269 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4270 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4271     "B", "Jail may set hostname");
4272 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4273     "B", "Jail may use SYSV IPC");
4274 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4275     "B", "Jail may create raw sockets");
4276 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4277     "B", "Jail may alter system file flags");
4278 SYSCTL_JAIL_PARAM(_allow, mount, CTLTYPE_INT | CTLFLAG_RW,
4279     "B", "Jail may mount/unmount jail-friendly file systems");
4280 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4281     "B", "Jail may set file quotas");
4282 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4283     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4284 
4285 void
4286 prison_racct_foreach(void (*callback)(struct racct *racct,
4287     void *arg2, void *arg3), void *arg2, void *arg3)
4288 {
4289 	struct prison_racct *prr;
4290 
4291 	sx_slock(&allprison_lock);
4292 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4293 		(callback)(prr->prr_racct, arg2, arg3);
4294 	sx_sunlock(&allprison_lock);
4295 }
4296 
4297 static struct prison_racct *
4298 prison_racct_find_locked(const char *name)
4299 {
4300 	struct prison_racct *prr;
4301 
4302 	sx_assert(&allprison_lock, SA_XLOCKED);
4303 
4304 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4305 		return (NULL);
4306 
4307 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4308 		if (strcmp(name, prr->prr_name) != 0)
4309 			continue;
4310 
4311 		/* Found prison_racct with a matching name? */
4312 		prison_racct_hold(prr);
4313 		return (prr);
4314 	}
4315 
4316 	/* Add new prison_racct. */
4317 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4318 	racct_create(&prr->prr_racct);
4319 
4320 	strcpy(prr->prr_name, name);
4321 	refcount_init(&prr->prr_refcount, 1);
4322 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4323 
4324 	return (prr);
4325 }
4326 
4327 struct prison_racct *
4328 prison_racct_find(const char *name)
4329 {
4330 	struct prison_racct *prr;
4331 
4332 	sx_xlock(&allprison_lock);
4333 	prr = prison_racct_find_locked(name);
4334 	sx_xunlock(&allprison_lock);
4335 	return (prr);
4336 }
4337 
4338 void
4339 prison_racct_hold(struct prison_racct *prr)
4340 {
4341 
4342 	refcount_acquire(&prr->prr_refcount);
4343 }
4344 
4345 void
4346 prison_racct_free(struct prison_racct *prr)
4347 {
4348 	int old;
4349 
4350 	old = prr->prr_refcount;
4351 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4352 		return;
4353 
4354 	sx_xlock(&allprison_lock);
4355 	if (refcount_release(&prr->prr_refcount)) {
4356 		racct_destroy(&prr->prr_racct);
4357 		LIST_REMOVE(prr, prr_next);
4358 		sx_xunlock(&allprison_lock);
4359 		free(prr, M_PRISON_RACCT);
4360 
4361 		return;
4362 	}
4363 	sx_xunlock(&allprison_lock);
4364 }
4365 
4366 #ifdef RACCT
4367 static void
4368 prison_racct_attach(struct prison *pr)
4369 {
4370 	struct prison_racct *prr;
4371 
4372 	prr = prison_racct_find_locked(pr->pr_name);
4373 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4374 
4375 	pr->pr_prison_racct = prr;
4376 }
4377 
4378 static void
4379 prison_racct_detach(struct prison *pr)
4380 {
4381 	prison_racct_free(pr->pr_prison_racct);
4382 	pr->pr_prison_racct = NULL;
4383 }
4384 #endif /* RACCT */
4385 
4386 #ifdef DDB
4387 
4388 static void
4389 db_show_prison(struct prison *pr)
4390 {
4391 	int fi;
4392 #if defined(INET) || defined(INET6)
4393 	int ii;
4394 #endif
4395 	unsigned jsf;
4396 #ifdef INET6
4397 	char ip6buf[INET6_ADDRSTRLEN];
4398 #endif
4399 
4400 	db_printf("prison %p:\n", pr);
4401 	db_printf(" jid             = %d\n", pr->pr_id);
4402 	db_printf(" name            = %s\n", pr->pr_name);
4403 	db_printf(" parent          = %p\n", pr->pr_parent);
4404 	db_printf(" ref             = %d\n", pr->pr_ref);
4405 	db_printf(" uref            = %d\n", pr->pr_uref);
4406 	db_printf(" path            = %s\n", pr->pr_path);
4407 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4408 	    ? pr->pr_cpuset->cs_id : -1);
4409 #ifdef VIMAGE
4410 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4411 #endif
4412 	db_printf(" root            = %p\n", pr->pr_root);
4413 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4414 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4415 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4416 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4417 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4418 	db_printf(" flags           = 0x%x", pr->pr_flags);
4419 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4420 	    fi++)
4421 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4422 			db_printf(" %s", pr_flag_names[fi]);
4423 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4424 	    fi++) {
4425 		jsf = pr->pr_flags &
4426 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4427 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4428 		    pr_flag_jailsys[fi].disable &&
4429 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4430 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4431 		    : "inherit");
4432 	}
4433 	db_printf(" allow           = 0x%x", pr->pr_allow);
4434 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4435 	    fi++)
4436 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4437 			db_printf(" %s", pr_allow_names[fi]);
4438 	db_printf("\n");
4439 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4440 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4441 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4442 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4443 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4444 #ifdef INET
4445 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4446 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4447 		db_printf(" %s %s\n",
4448 		    ii == 0 ? "ip4.addr        =" : "                 ",
4449 		    inet_ntoa(pr->pr_ip4[ii]));
4450 #endif
4451 #ifdef INET6
4452 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4453 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4454 		db_printf(" %s %s\n",
4455 		    ii == 0 ? "ip6.addr        =" : "                 ",
4456 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4457 #endif
4458 }
4459 
4460 DB_SHOW_COMMAND(prison, db_show_prison_command)
4461 {
4462 	struct prison *pr;
4463 
4464 	if (!have_addr) {
4465 		/*
4466 		 * Show all prisons in the list, and prison0 which is not
4467 		 * listed.
4468 		 */
4469 		db_show_prison(&prison0);
4470 		if (!db_pager_quit) {
4471 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4472 				db_show_prison(pr);
4473 				if (db_pager_quit)
4474 					break;
4475 			}
4476 		}
4477 		return;
4478 	}
4479 
4480 	if (addr == 0)
4481 		pr = &prison0;
4482 	else {
4483 		/* Look for a prison with the ID and with references. */
4484 		TAILQ_FOREACH(pr, &allprison, pr_list)
4485 			if (pr->pr_id == addr && pr->pr_ref > 0)
4486 				break;
4487 		if (pr == NULL)
4488 			/* Look again, without requiring a reference. */
4489 			TAILQ_FOREACH(pr, &allprison, pr_list)
4490 				if (pr->pr_id == addr)
4491 					break;
4492 		if (pr == NULL)
4493 			/* Assume address points to a valid prison. */
4494 			pr = (struct prison *)addr;
4495 	}
4496 	db_show_prison(pr);
4497 }
4498 
4499 #endif /* DDB */
4500