xref: /freebsd/sys/kern/kern_jail.c (revision 5861f9665471e98e544f6fa3ce73c4912229ff82)
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/sx.h>
53 #include <sys/sysent.h>
54 #include <sys/namei.h>
55 #include <sys/mount.h>
56 #include <sys/queue.h>
57 #include <sys/socket.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysctl.h>
60 #include <sys/vnode.h>
61 #include <sys/vimage.h>
62 #include <net/if.h>
63 #include <netinet/in.h>
64 #ifdef DDB
65 #include <ddb/ddb.h>
66 #ifdef INET6
67 #include <netinet6/in6_var.h>
68 #endif /* INET6 */
69 #endif /* DDB */
70 
71 #include <security/mac/mac_framework.h>
72 
73 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
74 
75 /* prison0 describes what is "real" about the system. */
76 struct prison prison0 = {
77 	.pr_id		= 0,
78 	.pr_name	= "0",
79 	.pr_ref		= 1,
80 	.pr_uref	= 1,
81 	.pr_path	= "/",
82 	.pr_securelevel	= -1,
83 	.pr_childmax	= JAIL_MAX,
84 	.pr_hostuuid	= "00000000-0000-0000-0000-000000000000",
85 	.pr_children	= LIST_HEAD_INITIALIZER(&prison0.pr_children),
86 	.pr_flags	= PR_HOST,
87 	.pr_allow	= PR_ALLOW_ALL,
88 };
89 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
90 
91 /* allprison and lastprid are protected by allprison_lock. */
92 struct	sx allprison_lock;
93 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
94 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
95 int	lastprid = 0;
96 
97 static int do_jail_attach(struct thread *td, struct prison *pr);
98 static void prison_complete(void *context, int pending);
99 static void prison_deref(struct prison *pr, int flags);
100 static char *prison_path(struct prison *pr1, struct prison *pr2);
101 static void prison_remove_one(struct prison *pr);
102 #ifdef INET
103 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
104 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
105 #endif
106 #ifdef INET6
107 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
108 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
109 #endif
110 
111 /* Flags for prison_deref */
112 #define	PD_DEREF	0x01
113 #define	PD_DEUREF	0x02
114 #define	PD_LOCKED	0x04
115 #define	PD_LIST_SLOCKED	0x08
116 #define	PD_LIST_XLOCKED	0x10
117 
118 /*
119  * Parameter names corresponding to PR_* flag values
120  */
121 static char *pr_flag_names[] = {
122 	[0] = "persist",
123 	"host",
124 #ifdef INET
125 	"ip4",
126 #endif
127 #ifdef INET6
128 	[3] = "ip6",
129 #endif
130 #ifdef VIMAGE
131 	[4] = "vnet",
132 #endif
133 };
134 
135 static char *pr_flag_nonames[] = {
136 	[0] = "nopersist",
137 	"nohost",
138 #ifdef INET
139 	"noip4",
140 #endif
141 #ifdef INET6
142 	[3] = "noip6",
143 #endif
144 #ifdef VIMAGE
145 	[4] = "novnet",
146 #endif
147 };
148 
149 static char *pr_allow_names[] = {
150 	"allow.set_hostname",
151 	"allow.sysvipc",
152 	"allow.raw_sockets",
153 	"allow.chflags",
154 	"allow.mount",
155 	"allow.quotas",
156 	"allow.socket_af",
157 };
158 
159 static char *pr_allow_nonames[] = {
160 	"allow.noset_hostname",
161 	"allow.nosysvipc",
162 	"allow.noraw_sockets",
163 	"allow.nochflags",
164 	"allow.nomount",
165 	"allow.noquotas",
166 	"allow.nosocket_af",
167 };
168 
169 #define	JAIL_DEFAULT_ALLOW	PR_ALLOW_SET_HOSTNAME
170 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
171 static int jail_default_enforce_statfs = 2;
172 #if defined(INET) || defined(INET6)
173 static unsigned jail_max_af_ips = 255;
174 #endif
175 
176 #ifdef INET
177 static int
178 qcmp_v4(const void *ip1, const void *ip2)
179 {
180 	in_addr_t iaa, iab;
181 
182 	/*
183 	 * We need to compare in HBO here to get the list sorted as expected
184 	 * by the result of the code.  Sorting NBO addresses gives you
185 	 * interesting results.  If you do not understand, do not try.
186 	 */
187 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
188 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
189 
190 	/*
191 	 * Do not simply return the difference of the two numbers, the int is
192 	 * not wide enough.
193 	 */
194 	if (iaa > iab)
195 		return (1);
196 	else if (iaa < iab)
197 		return (-1);
198 	else
199 		return (0);
200 }
201 #endif
202 
203 #ifdef INET6
204 static int
205 qcmp_v6(const void *ip1, const void *ip2)
206 {
207 	const struct in6_addr *ia6a, *ia6b;
208 	int i, rc;
209 
210 	ia6a = (const struct in6_addr *)ip1;
211 	ia6b = (const struct in6_addr *)ip2;
212 
213 	rc = 0;
214 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
215 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
216 			rc = 1;
217 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
218 			rc = -1;
219 	}
220 	return (rc);
221 }
222 #endif
223 
224 /*
225  * struct jail_args {
226  *	struct jail *jail;
227  * };
228  */
229 int
230 jail(struct thread *td, struct jail_args *uap)
231 {
232 	uint32_t version;
233 	int error;
234 	struct jail j;
235 
236 	error = copyin(uap->jail, &version, sizeof(uint32_t));
237 	if (error)
238 		return (error);
239 
240 	switch (version) {
241 	case 0:
242 	{
243 		struct jail_v0 j0;
244 
245 		/* FreeBSD single IPv4 jails. */
246 		bzero(&j, sizeof(struct jail));
247 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
248 		if (error)
249 			return (error);
250 		j.version = j0.version;
251 		j.path = j0.path;
252 		j.hostname = j0.hostname;
253 		j.ip4s = j0.ip_number;
254 		break;
255 	}
256 
257 	case 1:
258 		/*
259 		 * Version 1 was used by multi-IPv4 jail implementations
260 		 * that never made it into the official kernel.
261 		 */
262 		return (EINVAL);
263 
264 	case 2:	/* JAIL_API_VERSION */
265 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
266 		error = copyin(uap->jail, &j, sizeof(struct jail));
267 		if (error)
268 			return (error);
269 		break;
270 
271 	default:
272 		/* Sci-Fi jails are not supported, sorry. */
273 		return (EINVAL);
274 	}
275 	return (kern_jail(td, &j));
276 }
277 
278 int
279 kern_jail(struct thread *td, struct jail *j)
280 {
281 	struct iovec optiov[2 * (4
282 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
283 #ifdef INET
284 			    + 1
285 #endif
286 #ifdef INET6
287 			    + 1
288 #endif
289 			    )];
290 	struct uio opt;
291 	char *u_path, *u_hostname, *u_name;
292 #ifdef INET
293 	uint32_t ip4s;
294 	struct in_addr *u_ip4;
295 #endif
296 #ifdef INET6
297 	struct in6_addr *u_ip6;
298 #endif
299 	size_t tmplen;
300 	int error, enforce_statfs, fi;
301 
302 	bzero(&optiov, sizeof(optiov));
303 	opt.uio_iov = optiov;
304 	opt.uio_iovcnt = 0;
305 	opt.uio_offset = -1;
306 	opt.uio_resid = -1;
307 	opt.uio_segflg = UIO_SYSSPACE;
308 	opt.uio_rw = UIO_READ;
309 	opt.uio_td = td;
310 
311 	/* Set permissions for top-level jails from sysctls. */
312 	if (!jailed(td->td_ucred)) {
313 		for (fi = 0; fi < sizeof(pr_allow_names) /
314 		     sizeof(pr_allow_names[0]); fi++) {
315 			optiov[opt.uio_iovcnt].iov_base =
316 			    (jail_default_allow & (1 << fi))
317 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
318 			optiov[opt.uio_iovcnt].iov_len =
319 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
320 			opt.uio_iovcnt += 2;
321 		}
322 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
323 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
324 		opt.uio_iovcnt++;
325 		enforce_statfs = jail_default_enforce_statfs;
326 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
327 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
328 		opt.uio_iovcnt++;
329 	}
330 
331 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
332 #ifdef INET
333 	ip4s = (j->version == 0) ? 1 : j->ip4s;
334 	if (ip4s > jail_max_af_ips)
335 		return (EINVAL);
336 	tmplen += ip4s * sizeof(struct in_addr);
337 #else
338 	if (j->ip4s > 0)
339 		return (EINVAL);
340 #endif
341 #ifdef INET6
342 	if (j->ip6s > jail_max_af_ips)
343 		return (EINVAL);
344 	tmplen += j->ip6s * sizeof(struct in6_addr);
345 #else
346 	if (j->ip6s > 0)
347 		return (EINVAL);
348 #endif
349 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
350 	u_hostname = u_path + MAXPATHLEN;
351 	u_name = u_hostname + MAXHOSTNAMELEN;
352 #ifdef INET
353 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
354 #endif
355 #ifdef INET6
356 #ifdef INET
357 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
358 #else
359 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
360 #endif
361 #endif
362 	optiov[opt.uio_iovcnt].iov_base = "path";
363 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
364 	opt.uio_iovcnt++;
365 	optiov[opt.uio_iovcnt].iov_base = u_path;
366 	error = copyinstr(j->path, u_path, MAXPATHLEN,
367 	    &optiov[opt.uio_iovcnt].iov_len);
368 	if (error) {
369 		free(u_path, M_TEMP);
370 		return (error);
371 	}
372 	opt.uio_iovcnt++;
373 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
374 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
375 	opt.uio_iovcnt++;
376 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
377 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
378 	    &optiov[opt.uio_iovcnt].iov_len);
379 	if (error) {
380 		free(u_path, M_TEMP);
381 		return (error);
382 	}
383 	opt.uio_iovcnt++;
384 	if (j->jailname != NULL) {
385 		optiov[opt.uio_iovcnt].iov_base = "name";
386 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
387 		opt.uio_iovcnt++;
388 		optiov[opt.uio_iovcnt].iov_base = u_name;
389 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
390 		    &optiov[opt.uio_iovcnt].iov_len);
391 		if (error) {
392 			free(u_path, M_TEMP);
393 			return (error);
394 		}
395 		opt.uio_iovcnt++;
396 	}
397 #ifdef INET
398 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
399 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
400 	opt.uio_iovcnt++;
401 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
402 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
403 	if (j->version == 0)
404 		u_ip4->s_addr = j->ip4s;
405 	else {
406 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
407 		if (error) {
408 			free(u_path, M_TEMP);
409 			return (error);
410 		}
411 	}
412 	opt.uio_iovcnt++;
413 #endif
414 #ifdef INET6
415 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
416 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
417 	opt.uio_iovcnt++;
418 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
419 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
420 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
421 	if (error) {
422 		free(u_path, M_TEMP);
423 		return (error);
424 	}
425 	opt.uio_iovcnt++;
426 #endif
427 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
428 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
429 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
430 	free(u_path, M_TEMP);
431 	return (error);
432 }
433 
434 
435 /*
436  * struct jail_set_args {
437  *	struct iovec *iovp;
438  *	unsigned int iovcnt;
439  *	int flags;
440  * };
441  */
442 int
443 jail_set(struct thread *td, struct jail_set_args *uap)
444 {
445 	struct uio *auio;
446 	int error;
447 
448 	/* Check that we have an even number of iovecs. */
449 	if (uap->iovcnt & 1)
450 		return (EINVAL);
451 
452 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
453 	if (error)
454 		return (error);
455 	error = kern_jail_set(td, auio, uap->flags);
456 	free(auio, M_IOV);
457 	return (error);
458 }
459 
460 int
461 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
462 {
463 	struct nameidata nd;
464 #ifdef INET
465 	struct in_addr *ip4;
466 #endif
467 #ifdef INET6
468 	struct in6_addr *ip6;
469 #endif
470 	struct vfsopt *opt;
471 	struct vfsoptlist *opts;
472 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
473 	struct vnode *root;
474 	char *domain, *errmsg, *host, *name, *p, *path, *uuid;
475 #if defined(INET) || defined(INET6)
476 	void *op;
477 #endif
478 	unsigned long hid;
479 	size_t namelen, onamelen;
480 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
481 	int gotchildmax, gotenforce, gothid, gotslevel, fi, jid, len, level;
482 	int childmax, slevel, vfslocked;
483 #if defined(INET) || defined(INET6)
484 	int ii, ij;
485 #endif
486 #ifdef INET
487 	int ip4s, ip4a, redo_ip4;
488 #endif
489 #ifdef INET6
490 	int ip6s, ip6a, redo_ip6;
491 #endif
492 	unsigned pr_flags, ch_flags;
493 	unsigned pr_allow, ch_allow, tallow;
494 	char numbuf[12];
495 
496 	error = priv_check(td, PRIV_JAIL_SET);
497 	if (!error && (flags & JAIL_ATTACH))
498 		error = priv_check(td, PRIV_JAIL_ATTACH);
499 	if (error)
500 		return (error);
501 	mypr = ppr = td->td_ucred->cr_prison;
502 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
503 		return (EPERM);
504 	if (flags & ~JAIL_SET_MASK)
505 		return (EINVAL);
506 
507 	/*
508 	 * Check all the parameters before committing to anything.  Not all
509 	 * errors can be caught early, but we may as well try.  Also, this
510 	 * takes care of some expensive stuff (path lookup) before getting
511 	 * the allprison lock.
512 	 *
513 	 * XXX Jails are not filesystems, and jail parameters are not mount
514 	 *     options.  But it makes more sense to re-use the vfsopt code
515 	 *     than duplicate it under a different name.
516 	 */
517 	error = vfs_buildopts(optuio, &opts);
518 	if (error)
519 		return (error);
520 #ifdef INET
521 	ip4a = 0;
522 	ip4 = NULL;
523 #endif
524 #ifdef INET6
525 	ip6a = 0;
526 	ip6 = NULL;
527 #endif
528 
529 #if defined(INET) || defined(INET6)
530  again:
531 #endif
532 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
533 	if (error == ENOENT)
534 		jid = 0;
535 	else if (error != 0)
536 		goto done_free;
537 
538 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
539 	if (error == ENOENT)
540 		gotslevel = 0;
541 	else if (error != 0)
542 		goto done_free;
543 	else
544 		gotslevel = 1;
545 
546 	error =
547 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
548 	if (error == ENOENT)
549 		gotchildmax = 0;
550 	else if (error != 0)
551 		goto done_free;
552 	else
553 		gotchildmax = 1;
554 
555 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
556 	gotenforce = (error == 0);
557 	if (gotenforce) {
558 		if (enforce < 0 || enforce > 2)
559 			return (EINVAL);
560 	} else if (error != ENOENT)
561 		goto done_free;
562 
563 	pr_flags = ch_flags = 0;
564 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
565 	    fi++) {
566 		if (pr_flag_names[fi] == NULL)
567 			continue;
568 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
569 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
570 	}
571 	ch_flags |= pr_flags;
572 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
573 	    && !(pr_flags & PR_PERSIST)) {
574 		error = EINVAL;
575 		vfs_opterror(opts, "new jail must persist or attach");
576 		goto done_errmsg;
577 	}
578 #ifdef VIMAGE
579 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
580 		error = EINVAL;
581 		vfs_opterror(opts, "vnet cannot be changed after creation");
582 		goto done_errmsg;
583 	}
584 #endif
585 
586 	pr_allow = ch_allow = 0;
587 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
588 	    fi++) {
589 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
590 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
591 	}
592 	ch_allow |= pr_allow;
593 
594 	error = vfs_getopt(opts, "name", (void **)&name, &len);
595 	if (error == ENOENT)
596 		name = NULL;
597 	else if (error != 0)
598 		goto done_free;
599 	else {
600 		if (len == 0 || name[len - 1] != '\0') {
601 			error = EINVAL;
602 			goto done_free;
603 		}
604 		if (len > MAXHOSTNAMELEN) {
605 			error = ENAMETOOLONG;
606 			goto done_free;
607 		}
608 	}
609 
610 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
611 	if (error == ENOENT)
612 		host = NULL;
613 	else if (error != 0)
614 		goto done_free;
615 	else {
616 		ch_flags |= PR_HOST;
617 		pr_flags |= PR_HOST;
618 		if (len == 0 || host[len - 1] != '\0') {
619 			error = EINVAL;
620 			goto done_free;
621 		}
622 		if (len > MAXHOSTNAMELEN) {
623 			error = ENAMETOOLONG;
624 			goto done_free;
625 		}
626 	}
627 
628 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
629 	if (error == ENOENT)
630 		domain = NULL;
631 	else if (error != 0)
632 		goto done_free;
633 	else {
634 		ch_flags |= PR_HOST;
635 		pr_flags |= PR_HOST;
636 		if (len == 0 || domain[len - 1] != '\0') {
637 			error = EINVAL;
638 			goto done_free;
639 		}
640 		if (len > MAXHOSTNAMELEN) {
641 			error = ENAMETOOLONG;
642 			goto done_free;
643 		}
644 	}
645 
646 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
647 	if (error == ENOENT)
648 		uuid = NULL;
649 	else if (error != 0)
650 		goto done_free;
651 	else {
652 		ch_flags |= PR_HOST;
653 		pr_flags |= PR_HOST;
654 		if (len == 0 || uuid[len - 1] != '\0') {
655 			error = EINVAL;
656 			goto done_free;
657 		}
658 		if (len > HOSTUUIDLEN) {
659 			error = ENAMETOOLONG;
660 			goto done_free;
661 		}
662 	}
663 
664 #ifdef COMPAT_IA32
665 	if (td->td_proc->p_sysent->sv_flags & SV_IA32) {
666 		uint32_t hid32;
667 
668 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
669 		hid = hid32;
670 	} else
671 #endif
672 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
673 	if (error == ENOENT)
674 		gothid = 0;
675 	else if (error != 0)
676 		goto done_free;
677 	else {
678 		gothid = 1;
679 		ch_flags |= PR_HOST;
680 		pr_flags |= PR_HOST;
681 	}
682 
683 	/* This might be the second time around for this option. */
684 #ifdef INET
685 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
686 	if (error == ENOENT)
687 		ip4s = -1;
688 	else if (error != 0)
689 		goto done_free;
690 	else if (ip4s & (sizeof(*ip4) - 1)) {
691 		error = EINVAL;
692 		goto done_free;
693 	} else {
694 		ch_flags |= PR_IP4_USER;
695 		pr_flags |= PR_IP4_USER;
696 		if (ip4s > 0) {
697 			ip4s /= sizeof(*ip4);
698 			if (ip4s > jail_max_af_ips) {
699 				error = EINVAL;
700 				vfs_opterror(opts, "too many IPv4 addresses");
701 				goto done_errmsg;
702 			}
703 			if (ip4a < ip4s) {
704 				ip4a = ip4s;
705 				free(ip4, M_PRISON);
706 				ip4 = NULL;
707 			}
708 			if (ip4 == NULL)
709 				ip4 = malloc(ip4a * sizeof(*ip4), M_PRISON,
710 				    M_WAITOK);
711 			bcopy(op, ip4, ip4s * sizeof(*ip4));
712 			/*
713 			 * IP addresses are all sorted but ip[0] to preserve
714 			 * the primary IP address as given from userland.
715 			 * This special IP is used for unbound outgoing
716 			 * connections as well for "loopback" traffic.
717 			 */
718 			if (ip4s > 1)
719 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
720 			/*
721 			 * Check for duplicate addresses and do some simple
722 			 * zero and broadcast checks. If users give other bogus
723 			 * addresses it is their problem.
724 			 *
725 			 * We do not have to care about byte order for these
726 			 * checks so we will do them in NBO.
727 			 */
728 			for (ii = 0; ii < ip4s; ii++) {
729 				if (ip4[ii].s_addr == INADDR_ANY ||
730 				    ip4[ii].s_addr == INADDR_BROADCAST) {
731 					error = EINVAL;
732 					goto done_free;
733 				}
734 				if ((ii+1) < ip4s &&
735 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
736 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
737 					error = EINVAL;
738 					goto done_free;
739 				}
740 			}
741 		}
742 	}
743 #endif
744 
745 #ifdef INET6
746 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
747 	if (error == ENOENT)
748 		ip6s = -1;
749 	else if (error != 0)
750 		goto done_free;
751 	else if (ip6s & (sizeof(*ip6) - 1)) {
752 		error = EINVAL;
753 		goto done_free;
754 	} else {
755 		ch_flags |= PR_IP6_USER;
756 		pr_flags |= PR_IP6_USER;
757 		if (ip6s > 0) {
758 			ip6s /= sizeof(*ip6);
759 			if (ip6s > jail_max_af_ips) {
760 				error = EINVAL;
761 				vfs_opterror(opts, "too many IPv6 addresses");
762 				goto done_errmsg;
763 			}
764 			if (ip6a < ip6s) {
765 				ip6a = ip6s;
766 				free(ip6, M_PRISON);
767 				ip6 = NULL;
768 			}
769 			if (ip6 == NULL)
770 				ip6 = malloc(ip6a * sizeof(*ip6), M_PRISON,
771 				    M_WAITOK);
772 			bcopy(op, ip6, ip6s * sizeof(*ip6));
773 			if (ip6s > 1)
774 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
775 			for (ii = 0; ii < ip6s; ii++) {
776 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
777 					error = EINVAL;
778 					goto done_free;
779 				}
780 				if ((ii+1) < ip6s &&
781 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
782 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
783 				{
784 					error = EINVAL;
785 					goto done_free;
786 				}
787 			}
788 		}
789 	}
790 #endif
791 
792 	root = NULL;
793 	error = vfs_getopt(opts, "path", (void **)&path, &len);
794 	if (error == ENOENT)
795 		path = NULL;
796 	else if (error != 0)
797 		goto done_free;
798 	else {
799 		if (flags & JAIL_UPDATE) {
800 			error = EINVAL;
801 			vfs_opterror(opts,
802 			    "path cannot be changed after creation");
803 			goto done_errmsg;
804 		}
805 		if (len == 0 || path[len - 1] != '\0') {
806 			error = EINVAL;
807 			goto done_free;
808 		}
809 		if (len < 2 || (len == 2 && path[0] == '/'))
810 			path = NULL;
811 		else {
812 			/* Leave room for a real-root full pathname. */
813 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
814 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
815 				error = ENAMETOOLONG;
816 				goto done_free;
817 			}
818 			NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW, UIO_SYSSPACE,
819 			    path, td);
820 			error = namei(&nd);
821 			if (error)
822 				goto done_free;
823 			vfslocked = NDHASGIANT(&nd);
824 			root = nd.ni_vp;
825 			NDFREE(&nd, NDF_ONLY_PNBUF);
826 			if (root->v_type != VDIR) {
827 				error = ENOTDIR;
828 				vrele(root);
829 				VFS_UNLOCK_GIANT(vfslocked);
830 				goto done_free;
831 			}
832 			VFS_UNLOCK_GIANT(vfslocked);
833 		}
834 	}
835 
836 	/*
837 	 * Grab the allprison lock before letting modules check their
838 	 * parameters.  Once we have it, do not let go so we'll have a
839 	 * consistent view of the OSD list.
840 	 */
841 	sx_xlock(&allprison_lock);
842 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
843 	if (error)
844 		goto done_unlock_list;
845 
846 	/* By now, all parameters should have been noted. */
847 	TAILQ_FOREACH(opt, opts, link) {
848 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
849 			error = EINVAL;
850 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
851 			goto done_unlock_list;
852 		}
853 	}
854 
855 	/*
856 	 * See if we are creating a new record or updating an existing one.
857 	 * This abuses the file error codes ENOENT and EEXIST.
858 	 */
859 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
860 	if (!cuflags) {
861 		error = EINVAL;
862 		vfs_opterror(opts, "no valid operation (create or update)");
863 		goto done_unlock_list;
864 	}
865 	pr = NULL;
866 	if (jid != 0) {
867 		/*
868 		 * See if a requested jid already exists.  There is an
869 		 * information leak here if the jid exists but is not within
870 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
871 		 * even though they cannot see the jail, and CREATE | UPDATE
872 		 * will return ENOENT which is not normally a valid error.
873 		 */
874 		if (jid < 0) {
875 			error = EINVAL;
876 			vfs_opterror(opts, "negative jid");
877 			goto done_unlock_list;
878 		}
879 		pr = prison_find(jid);
880 		if (pr != NULL) {
881 			ppr = pr->pr_parent;
882 			/* Create: jid must not exist. */
883 			if (cuflags == JAIL_CREATE) {
884 				mtx_unlock(&pr->pr_mtx);
885 				error = EEXIST;
886 				vfs_opterror(opts, "jail %d already exists",
887 				    jid);
888 				goto done_unlock_list;
889 			}
890 			if (!prison_ischild(mypr, pr)) {
891 				mtx_unlock(&pr->pr_mtx);
892 				pr = NULL;
893 			} else if (pr->pr_uref == 0) {
894 				if (!(flags & JAIL_DYING)) {
895 					mtx_unlock(&pr->pr_mtx);
896 					error = ENOENT;
897 					vfs_opterror(opts, "jail %d is dying",
898 					    jid);
899 					goto done_unlock_list;
900 				} else if ((flags & JAIL_ATTACH) ||
901 				    (pr_flags & PR_PERSIST)) {
902 					/*
903 					 * A dying jail might be resurrected
904 					 * (via attach or persist), but first
905 					 * it must determine if another jail
906 					 * has claimed its name.  Accomplish
907 					 * this by implicitly re-setting the
908 					 * name.
909 					 */
910 					if (name == NULL)
911 						name = prison_name(mypr, pr);
912 				}
913 			}
914 		}
915 		if (pr == NULL) {
916 			/* Update: jid must exist. */
917 			if (cuflags == JAIL_UPDATE) {
918 				error = ENOENT;
919 				vfs_opterror(opts, "jail %d not found", jid);
920 				goto done_unlock_list;
921 			}
922 		}
923 	}
924 	/*
925 	 * If the caller provided a name, look for a jail by that name.
926 	 * This has different semantics for creates and updates keyed by jid
927 	 * (where the name must not already exist in a different jail),
928 	 * and updates keyed by the name itself (where the name must exist
929 	 * because that is the jail being updated).
930 	 */
931 	if (name != NULL) {
932 		p = strrchr(name, '.');
933 		if (p != NULL) {
934 			/*
935 			 * This is a hierarchical name.  Split it into the
936 			 * parent and child names, and make sure the parent
937 			 * exists or matches an already found jail.
938 			 */
939 			*p = '\0';
940 			if (pr != NULL) {
941 				if (strncmp(name, ppr->pr_name, p - name) ||
942 				    ppr->pr_name[p - name] != '\0') {
943 					mtx_unlock(&pr->pr_mtx);
944 					error = EINVAL;
945 					vfs_opterror(opts,
946 					    "cannot change jail's parent");
947 					goto done_unlock_list;
948 				}
949 			} else {
950 				ppr = prison_find_name(mypr, name);
951 				if (ppr == NULL) {
952 					error = ENOENT;
953 					vfs_opterror(opts,
954 					    "jail \"%s\" not found", name);
955 					goto done_unlock_list;
956 				}
957 				mtx_unlock(&ppr->pr_mtx);
958 			}
959 			name = p + 1;
960 		}
961 		if (name[0] != '\0') {
962 			namelen =
963 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
964  name_again:
965 			deadpr = NULL;
966 			FOREACH_PRISON_CHILD(ppr, tpr) {
967 				if (tpr != pr && tpr->pr_ref > 0 &&
968 				    !strcmp(tpr->pr_name + namelen, name)) {
969 					if (pr == NULL &&
970 					    cuflags != JAIL_CREATE) {
971 						mtx_lock(&tpr->pr_mtx);
972 						if (tpr->pr_ref > 0) {
973 							/*
974 							 * Use this jail
975 							 * for updates.
976 							 */
977 							if (tpr->pr_uref > 0) {
978 								pr = tpr;
979 								break;
980 							}
981 							deadpr = tpr;
982 						}
983 						mtx_unlock(&tpr->pr_mtx);
984 					} else if (tpr->pr_uref > 0) {
985 						/*
986 						 * Create, or update(jid):
987 						 * name must not exist in an
988 						 * active sibling jail.
989 						 */
990 						error = EEXIST;
991 						if (pr != NULL)
992 							mtx_unlock(&pr->pr_mtx);
993 						vfs_opterror(opts,
994 						   "jail \"%s\" already exists",
995 						   name);
996 						goto done_unlock_list;
997 					}
998 				}
999 			}
1000 			/* If no active jail is found, use a dying one. */
1001 			if (deadpr != NULL && pr == NULL) {
1002 				if (flags & JAIL_DYING) {
1003 					mtx_lock(&deadpr->pr_mtx);
1004 					if (deadpr->pr_ref == 0) {
1005 						mtx_unlock(&deadpr->pr_mtx);
1006 						goto name_again;
1007 					}
1008 					pr = deadpr;
1009 				} else if (cuflags == JAIL_UPDATE) {
1010 					error = ENOENT;
1011 					vfs_opterror(opts,
1012 					    "jail \"%s\" is dying", name);
1013 					goto done_unlock_list;
1014 				}
1015 			}
1016 			/* Update: name must exist if no jid. */
1017 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1018 				error = ENOENT;
1019 				vfs_opterror(opts, "jail \"%s\" not found",
1020 				    name);
1021 				goto done_unlock_list;
1022 			}
1023 		}
1024 	}
1025 	/* Update: must provide a jid or name. */
1026 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1027 		error = ENOENT;
1028 		vfs_opterror(opts, "update specified no jail");
1029 		goto done_unlock_list;
1030 	}
1031 
1032 	/* If there's no prison to update, create a new one and link it in. */
1033 	if (pr == NULL) {
1034 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1035 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1036 				error = EPERM;
1037 				vfs_opterror(opts, "prison limit exceeded");
1038 				goto done_unlock_list;
1039 			}
1040 		created = 1;
1041 		mtx_lock(&ppr->pr_mtx);
1042 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1043 			mtx_unlock(&ppr->pr_mtx);
1044 			error = ENOENT;
1045 			vfs_opterror(opts, "parent jail went away!");
1046 			goto done_unlock_list;
1047 		}
1048 		ppr->pr_ref++;
1049 		ppr->pr_uref++;
1050 		mtx_unlock(&ppr->pr_mtx);
1051 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1052 		if (jid == 0) {
1053 			/* Find the next free jid. */
1054 			jid = lastprid + 1;
1055  findnext:
1056 			if (jid == JAIL_MAX)
1057 				jid = 1;
1058 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1059 				if (tpr->pr_id < jid)
1060 					continue;
1061 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1062 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1063 					break;
1064 				}
1065 				if (jid == lastprid) {
1066 					error = EAGAIN;
1067 					vfs_opterror(opts,
1068 					    "no available jail IDs");
1069 					free(pr, M_PRISON);
1070 					prison_deref(ppr, PD_DEREF |
1071 					    PD_DEUREF | PD_LIST_XLOCKED);
1072 					goto done_releroot;
1073 				}
1074 				jid++;
1075 				goto findnext;
1076 			}
1077 			lastprid = jid;
1078 		} else {
1079 			/*
1080 			 * The jail already has a jid (that did not yet exist),
1081 			 * so just find where to insert it.
1082 			 */
1083 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1084 				if (tpr->pr_id >= jid) {
1085 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1086 					break;
1087 				}
1088 		}
1089 		if (tpr == NULL)
1090 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1091 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1092 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1093 			tpr->pr_childcount++;
1094 
1095 		pr->pr_parent = ppr;
1096 		pr->pr_id = jid;
1097 
1098 		/* Set some default values, and inherit some from the parent. */
1099 		if (name == NULL)
1100 			name = "";
1101 		if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1102 			if (host == NULL)
1103 				host = ppr->pr_hostname;
1104 			if (domain == NULL)
1105 				domain = ppr->pr_domainname;
1106 			if (uuid == NULL)
1107 				uuid = ppr->pr_hostuuid;
1108 			if (!gothid)
1109 				hid = ppr->pr_hostid;
1110 		}
1111 		if (path == NULL) {
1112 			path = "/";
1113 			root = mypr->pr_root;
1114 			vref(root);
1115 		}
1116 #ifdef INET
1117 		pr->pr_flags |= ppr->pr_flags & PR_IP4;
1118 		pr->pr_ip4s = ppr->pr_ip4s;
1119 		if (ppr->pr_ip4 != NULL) {
1120 			pr->pr_ip4 = malloc(pr->pr_ip4s *
1121 			    sizeof(struct in_addr), M_PRISON, M_WAITOK);
1122 			bcopy(ppr->pr_ip4, pr->pr_ip4,
1123 			    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1124 		}
1125 #endif
1126 #ifdef INET6
1127 		pr->pr_flags |= ppr->pr_flags & PR_IP6;
1128 		pr->pr_ip6s = ppr->pr_ip6s;
1129 		if (ppr->pr_ip6 != NULL) {
1130 			pr->pr_ip6 = malloc(pr->pr_ip6s *
1131 			    sizeof(struct in6_addr), M_PRISON, M_WAITOK);
1132 			bcopy(ppr->pr_ip6, pr->pr_ip6,
1133 			    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1134 		}
1135 #endif
1136 		pr->pr_securelevel = ppr->pr_securelevel;
1137 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1138 		pr->pr_enforce_statfs = ppr->pr_enforce_statfs;
1139 
1140 		LIST_INIT(&pr->pr_children);
1141 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1142 
1143 #ifdef VIMAGE
1144 		/* Allocate a new vnet if specified. */
1145 		pr->pr_vnet = (pr_flags & PR_VNET)
1146 		    ? vnet_alloc() : ppr->pr_vnet;
1147 #endif
1148 		/*
1149 		 * Allocate a dedicated cpuset for each jail.
1150 		 * Unlike other initial settings, this may return an erorr.
1151 		 */
1152 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1153 		if (error) {
1154 			prison_deref(pr, PD_LIST_XLOCKED);
1155 			goto done_releroot;
1156 		}
1157 
1158 		mtx_lock(&pr->pr_mtx);
1159 		/*
1160 		 * New prisons do not yet have a reference, because we do not
1161 		 * want other to see the incomplete prison once the
1162 		 * allprison_lock is downgraded.
1163 		 */
1164 	} else {
1165 		created = 0;
1166 		/*
1167 		 * Grab a reference for existing prisons, to ensure they
1168 		 * continue to exist for the duration of the call.
1169 		 */
1170 		pr->pr_ref++;
1171 	}
1172 
1173 	/* Do final error checking before setting anything. */
1174 	if (gotslevel) {
1175 		if (slevel < ppr->pr_securelevel) {
1176 			error = EPERM;
1177 			goto done_deref_locked;
1178 		}
1179 	}
1180 	if (gotchildmax) {
1181 		if (childmax >= ppr->pr_childmax) {
1182 			error = EPERM;
1183 			goto done_deref_locked;
1184 		}
1185 	}
1186 	if (gotenforce) {
1187 		if (enforce < ppr->pr_enforce_statfs) {
1188 			error = EPERM;
1189 			goto done_deref_locked;
1190 		}
1191 	}
1192 #ifdef INET
1193 	if (ch_flags & PR_IP4_USER) {
1194 		if (ppr->pr_flags & PR_IP4) {
1195 			if (!(pr_flags & PR_IP4_USER)) {
1196 				/*
1197 				 * Silently ignore attempts to make the IP
1198 				 * addresses unrestricted when the parent is
1199 				 * restricted; in other words, interpret
1200 				 * "unrestricted" as "as unrestricted as
1201 				 * possible".
1202 				 */
1203 				ip4s = ppr->pr_ip4s;
1204 				if (ip4s == 0) {
1205 					free(ip4, M_PRISON);
1206 					ip4 = NULL;
1207 				} else if (ip4s <= ip4a) {
1208 					/* Inherit the parent's address(es). */
1209 					bcopy(ppr->pr_ip4, ip4,
1210 					    ip4s * sizeof(*ip4));
1211 				} else {
1212 					/*
1213 					 * There's no room for the parent's
1214 					 * address list.  Allocate some more.
1215 					 */
1216 					ip4a = ip4s;
1217 					free(ip4, M_PRISON);
1218 					ip4 = malloc(ip4a * sizeof(*ip4),
1219 					    M_PRISON, M_NOWAIT);
1220 					if (ip4 != NULL)
1221 						bcopy(ppr->pr_ip4, ip4,
1222 						    ip4s * sizeof(*ip4));
1223 					else {
1224 						/* Allocation failed without
1225 						 * sleeping.  Unlocking the
1226 						 * prison now will invalidate
1227 						 * some checks and prematurely
1228 						 * show an unfinished new jail.
1229 						 * So let go of everything and
1230 						 * start over.
1231 						 */
1232 						prison_deref(pr, created
1233 						    ? PD_LOCKED |
1234 						      PD_LIST_XLOCKED
1235 						    : PD_DEREF | PD_LOCKED |
1236 						      PD_LIST_XLOCKED);
1237 						if (root != NULL) {
1238 							vfslocked =
1239 							    VFS_LOCK_GIANT(
1240 							    root->v_mount);
1241 							vrele(root);
1242 							VFS_UNLOCK_GIANT(
1243 							    vfslocked);
1244 						}
1245 						ip4 = malloc(ip4a *
1246 						    sizeof(*ip4), M_PRISON,
1247 						    M_WAITOK);
1248 						goto again;
1249 					}
1250 				}
1251 			} else if (ip4s > 0) {
1252 				/*
1253 				 * Make sure the new set of IP addresses is a
1254 				 * subset of the parent's list.  Don't worry
1255 				 * about the parent being unlocked, as any
1256 				 * setting is done with allprison_lock held.
1257 				 */
1258 				for (ij = 0; ij < ppr->pr_ip4s; ij++)
1259 					if (ip4[0].s_addr ==
1260 					    ppr->pr_ip4[ij].s_addr)
1261 						break;
1262 				if (ij == ppr->pr_ip4s) {
1263 					error = EPERM;
1264 					goto done_deref_locked;
1265 				}
1266 				if (ip4s > 1) {
1267 					for (ii = ij = 1; ii < ip4s; ii++) {
1268 						if (ip4[ii].s_addr ==
1269 						    ppr->pr_ip4[0].s_addr)
1270 							continue;
1271 						for (; ij < ppr->pr_ip4s; ij++)
1272 						    if (ip4[ii].s_addr ==
1273 							ppr->pr_ip4[ij].s_addr)
1274 							    break;
1275 						if (ij == ppr->pr_ip4s)
1276 							break;
1277 					}
1278 					if (ij == ppr->pr_ip4s) {
1279 						error = EPERM;
1280 						goto done_deref_locked;
1281 					}
1282 				}
1283 			}
1284 		}
1285 		if (ip4s > 0) {
1286 			/*
1287 			 * Check for conflicting IP addresses.  We permit them
1288 			 * if there is no more than one IP on each jail.  If
1289 			 * there is a duplicate on a jail with more than one
1290 			 * IP stop checking and return error.
1291 			 */
1292 			FOREACH_PRISON_DESCENDANT(&prison0, tpr, descend) {
1293 				if (tpr == pr || tpr->pr_uref == 0) {
1294 					descend = 0;
1295 					continue;
1296 				}
1297 				if (!(tpr->pr_flags & PR_IP4_USER))
1298 					continue;
1299 				descend = 0;
1300 				if (tpr->pr_ip4 == NULL ||
1301 				    (ip4s == 1 && tpr->pr_ip4s == 1))
1302 					continue;
1303 				for (ii = 0; ii < ip4s; ii++) {
1304 					if (_prison_check_ip4(tpr,
1305 					    &ip4[ii]) == 0) {
1306 						error = EADDRINUSE;
1307 						vfs_opterror(opts,
1308 						    "IPv4 addresses clash");
1309 						goto done_deref_locked;
1310 					}
1311 				}
1312 			}
1313 		}
1314 	}
1315 #endif
1316 #ifdef INET6
1317 	if (ch_flags & PR_IP6_USER) {
1318 		if (ppr->pr_flags & PR_IP6) {
1319 			if (!(pr_flags & PR_IP6_USER)) {
1320 				/*
1321 				 * Silently ignore attempts to make the IP
1322 				 * addresses unrestricted when the parent is
1323 				 * restricted.
1324 				 */
1325 				ip6s = ppr->pr_ip6s;
1326 				if (ip6s == 0) {
1327 					free(ip6, M_PRISON);
1328 					ip6 = NULL;
1329 				} else if (ip6s <= ip6a) {
1330 					/* Inherit the parent's address(es). */
1331 					bcopy(ppr->pr_ip6, ip6,
1332 					    ip6s * sizeof(*ip6));
1333 				} else {
1334 					/*
1335 					 * There's no room for the parent's
1336 					 * address list.
1337 					 */
1338 					ip6a = ip6s;
1339 					free(ip6, M_PRISON);
1340 					ip6 = malloc(ip6a * sizeof(*ip6),
1341 					    M_PRISON, M_NOWAIT);
1342 					if (ip6 != NULL)
1343 						bcopy(ppr->pr_ip6, ip6,
1344 						    ip6s * sizeof(*ip6));
1345 					else {
1346 						prison_deref(pr, created
1347 						    ? PD_LOCKED |
1348 						      PD_LIST_XLOCKED
1349 						    : PD_DEREF | PD_LOCKED |
1350 						      PD_LIST_XLOCKED);
1351 						if (root != NULL) {
1352 							vfslocked =
1353 							    VFS_LOCK_GIANT(
1354 							    root->v_mount);
1355 							vrele(root);
1356 							VFS_UNLOCK_GIANT(
1357 							    vfslocked);
1358 						}
1359 						ip6 = malloc(ip6a *
1360 						    sizeof(*ip6), M_PRISON,
1361 						    M_WAITOK);
1362 						goto again;
1363 					}
1364 				}
1365 			} else if (ip6s > 0) {
1366 				/*
1367 				 * Make sure the new set of IP addresses is a
1368 				 * subset of the parent's list.
1369 				 */
1370 				for (ij = 0; ij < ppr->pr_ip6s; ij++)
1371 					if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1372 					    &ppr->pr_ip6[ij]))
1373 						break;
1374 				if (ij == ppr->pr_ip6s) {
1375 					error = EPERM;
1376 					goto done_deref_locked;
1377 				}
1378 				if (ip6s > 1) {
1379 					for (ii = ij = 1; ii < ip6s; ii++) {
1380 						if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1381 						    &ppr->pr_ip6[0]))
1382 							continue;
1383 						for (; ij < ppr->pr_ip6s; ij++)
1384 							if (IN6_ARE_ADDR_EQUAL(
1385 							    &ip6[ii],
1386 							    &ppr->pr_ip6[ij]))
1387 								break;
1388 						if (ij == ppr->pr_ip6s)
1389 							break;
1390 					}
1391 					if (ij == ppr->pr_ip6s) {
1392 						error = EPERM;
1393 						goto done_deref_locked;
1394 					}
1395 				}
1396 			}
1397 		}
1398 		if (ip6s > 0) {
1399 			/* Check for conflicting IP addresses. */
1400 			FOREACH_PRISON_DESCENDANT(&prison0, tpr, descend) {
1401 				if (tpr == pr || tpr->pr_uref == 0) {
1402 					descend = 0;
1403 					continue;
1404 				}
1405 				if (!(tpr->pr_flags & PR_IP6_USER))
1406 					continue;
1407 				descend = 0;
1408 				if (tpr->pr_ip6 == NULL ||
1409 				    (ip6s == 1 && tpr->pr_ip6s == 1))
1410 					continue;
1411 				for (ii = 0; ii < ip6s; ii++) {
1412 					if (_prison_check_ip6(tpr,
1413 					    &ip6[ii]) == 0) {
1414 						error = EADDRINUSE;
1415 						vfs_opterror(opts,
1416 						    "IPv6 addresses clash");
1417 						goto done_deref_locked;
1418 					}
1419 				}
1420 			}
1421 		}
1422 	}
1423 #endif
1424 	onamelen = namelen = 0;
1425 	if (name != NULL) {
1426 		/* Give a default name of the jid. */
1427 		if (name[0] == '\0')
1428 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1429 		else if (strtoul(name, &p, 10) != jid && *p == '\0') {
1430 			error = EINVAL;
1431 			vfs_opterror(opts, "name cannot be numeric");
1432 			goto done_deref_locked;
1433 		}
1434 		/*
1435 		 * Make sure the name isn't too long for the prison or its
1436 		 * children.
1437 		 */
1438 		onamelen = strlen(pr->pr_name);
1439 		namelen = strlen(name);
1440 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1441 			error = ENAMETOOLONG;
1442 			goto done_deref_locked;
1443 		}
1444 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1445 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1446 			    sizeof(pr->pr_name)) {
1447 				error = ENAMETOOLONG;
1448 				goto done_deref_locked;
1449 			}
1450 		}
1451 	}
1452 	if (pr_allow & ~ppr->pr_allow) {
1453 		error = EPERM;
1454 		goto done_deref_locked;
1455 	}
1456 
1457 	/* Set the parameters of the prison. */
1458 #ifdef INET
1459 	redo_ip4 = 0;
1460 	if (ch_flags & PR_IP4_USER) {
1461 		if (pr_flags & PR_IP4_USER) {
1462 			/* Some restriction set. */
1463 			pr->pr_flags |= PR_IP4;
1464 			if (ip4s >= 0) {
1465 				free(pr->pr_ip4, M_PRISON);
1466 				pr->pr_ip4s = ip4s;
1467 				pr->pr_ip4 = ip4;
1468 				ip4 = NULL;
1469 			}
1470 		} else if (ppr->pr_flags & PR_IP4) {
1471 			/* This restriction cleared, but keep inherited. */
1472 			free(pr->pr_ip4, M_PRISON);
1473 			pr->pr_ip4s = ip4s;
1474 			pr->pr_ip4 = ip4;
1475 			ip4 = NULL;
1476 		} else {
1477 			/* Restriction cleared, now unrestricted. */
1478 			pr->pr_flags &= ~PR_IP4;
1479 			free(pr->pr_ip4, M_PRISON);
1480 			pr->pr_ip4s = 0;
1481 		}
1482 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1483 			if (prison_restrict_ip4(tpr, NULL)) {
1484 				redo_ip4 = 1;
1485 				descend = 0;
1486 			}
1487 		}
1488 	}
1489 #endif
1490 #ifdef INET6
1491 	redo_ip6 = 0;
1492 	if (ch_flags & PR_IP6_USER) {
1493 		if (pr_flags & PR_IP6_USER) {
1494 			/* Some restriction set. */
1495 			pr->pr_flags |= PR_IP6;
1496 			if (ip6s >= 0) {
1497 				free(pr->pr_ip6, M_PRISON);
1498 				pr->pr_ip6s = ip6s;
1499 				pr->pr_ip6 = ip6;
1500 				ip6 = NULL;
1501 			}
1502 		} else if (ppr->pr_flags & PR_IP6) {
1503 			/* This restriction cleared, but keep inherited. */
1504 			free(pr->pr_ip6, M_PRISON);
1505 			pr->pr_ip6s = ip6s;
1506 			pr->pr_ip6 = ip6;
1507 			ip6 = NULL;
1508 		} else {
1509 			/* Restriction cleared, now unrestricted. */
1510 			pr->pr_flags &= ~PR_IP6;
1511 			free(pr->pr_ip6, M_PRISON);
1512 			pr->pr_ip6s = 0;
1513 		}
1514 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1515 			if (prison_restrict_ip6(tpr, NULL)) {
1516 				redo_ip6 = 1;
1517 				descend = 0;
1518 			}
1519 		}
1520 	}
1521 #endif
1522 	if (gotslevel) {
1523 		pr->pr_securelevel = slevel;
1524 		/* Set all child jails to be at least this level. */
1525 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1526 			if (tpr->pr_securelevel < slevel)
1527 				tpr->pr_securelevel = slevel;
1528 	}
1529 	if (gotchildmax) {
1530 		pr->pr_childmax = childmax;
1531 		/* Set all child jails to under this limit. */
1532 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1533 			if (tpr->pr_childmax > childmax - level)
1534 				tpr->pr_childmax = childmax > level
1535 				    ? childmax - level : 0;
1536 	}
1537 	if (gotenforce) {
1538 		pr->pr_enforce_statfs = enforce;
1539 		/* Pass this restriction on to the children. */
1540 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1541 			if (tpr->pr_enforce_statfs < enforce)
1542 				tpr->pr_enforce_statfs = enforce;
1543 	}
1544 	if (name != NULL) {
1545 		if (ppr == &prison0)
1546 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1547 		else
1548 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1549 			    ppr->pr_name, name);
1550 		/* Change this component of child names. */
1551 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1552 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1553 			    strlen(tpr->pr_name + onamelen) + 1);
1554 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1555 		}
1556 	}
1557 	if (path != NULL) {
1558 		/* Try to keep a real-rooted full pathname. */
1559 		if (path[0] == '/' && strcmp(mypr->pr_path, "/"))
1560 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1561 			    mypr->pr_path, path);
1562 		else
1563 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1564 		pr->pr_root = root;
1565 	}
1566 	if (PR_HOST & ch_flags & ~pr_flags) {
1567 		if (pr->pr_flags & PR_HOST) {
1568 			/*
1569 			 * Copy the parent's host info.  As with pr_ip4 above,
1570 			 * the lack of a lock on the parent is not a problem;
1571 			 * it is always set with allprison_lock at least
1572 			 * shared, and is held exclusively here.
1573 			 */
1574 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1575 			    sizeof(pr->pr_hostname));
1576 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1577 			    sizeof(pr->pr_domainname));
1578 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1579 			    sizeof(pr->pr_hostuuid));
1580 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1581 		}
1582 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1583 		/* Set this prison, and any descendants without PR_HOST. */
1584 		if (host != NULL)
1585 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1586 		if (domain != NULL)
1587 			strlcpy(pr->pr_domainname, domain,
1588 			    sizeof(pr->pr_domainname));
1589 		if (uuid != NULL)
1590 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1591 		if (gothid)
1592 			pr->pr_hostid = hid;
1593 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1594 			if (tpr->pr_flags & PR_HOST)
1595 				descend = 0;
1596 			else {
1597 				if (host != NULL)
1598 					strlcpy(tpr->pr_hostname,
1599 					    pr->pr_hostname,
1600 					    sizeof(tpr->pr_hostname));
1601 				if (domain != NULL)
1602 					strlcpy(tpr->pr_domainname,
1603 					    pr->pr_domainname,
1604 					    sizeof(tpr->pr_domainname));
1605 				if (uuid != NULL)
1606 					strlcpy(tpr->pr_hostuuid,
1607 					    pr->pr_hostuuid,
1608 					    sizeof(tpr->pr_hostuuid));
1609 				if (gothid)
1610 					tpr->pr_hostid = hid;
1611 			}
1612 		}
1613 	}
1614 	if ((tallow = ch_allow & ~pr_allow)) {
1615 		/* Clear allow bits in all children. */
1616 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1617 			tpr->pr_allow &= ~tallow;
1618 	}
1619 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1620 	/*
1621 	 * Persistent prisons get an extra reference, and prisons losing their
1622 	 * persist flag lose that reference.  Only do this for existing prisons
1623 	 * for now, so new ones will remain unseen until after the module
1624 	 * handlers have completed.
1625 	 */
1626 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1627 		if (pr_flags & PR_PERSIST) {
1628 			pr->pr_ref++;
1629 			pr->pr_uref++;
1630 		} else {
1631 			pr->pr_ref--;
1632 			pr->pr_uref--;
1633 		}
1634 	}
1635 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1636 	mtx_unlock(&pr->pr_mtx);
1637 
1638 	/* Locks may have prevented a complete restriction of child IP
1639 	 * addresses.  If so, allocate some more memory and try again.
1640 	 */
1641 #ifdef INET
1642 	while (redo_ip4) {
1643 		ip4s = pr->pr_ip4s;
1644 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1645 		mtx_lock(&pr->pr_mtx);
1646 		redo_ip4 = 0;
1647 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1648 			if (prison_restrict_ip4(tpr, ip4)) {
1649 				if (ip4 != NULL)
1650 					ip4 = NULL;
1651 				else
1652 					redo_ip4 = 1;
1653 			}
1654 		}
1655 		mtx_unlock(&pr->pr_mtx);
1656 	}
1657 #endif
1658 #ifdef INET6
1659 	while (redo_ip6) {
1660 		ip6s = pr->pr_ip6s;
1661 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1662 		mtx_lock(&pr->pr_mtx);
1663 		redo_ip6 = 0;
1664 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1665 			if (prison_restrict_ip6(tpr, ip6)) {
1666 				if (ip6 != NULL)
1667 					ip6 = NULL;
1668 				else
1669 					redo_ip6 = 1;
1670 			}
1671 		}
1672 		mtx_unlock(&pr->pr_mtx);
1673 	}
1674 #endif
1675 
1676 	/* Let the modules do their work. */
1677 	sx_downgrade(&allprison_lock);
1678 	if (created) {
1679 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1680 		if (error) {
1681 			prison_deref(pr, PD_LIST_SLOCKED);
1682 			goto done_errmsg;
1683 		}
1684 	}
1685 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1686 	if (error) {
1687 		prison_deref(pr, created
1688 		    ? PD_LIST_SLOCKED
1689 		    : PD_DEREF | PD_LIST_SLOCKED);
1690 		goto done_errmsg;
1691 	}
1692 
1693 	/* Attach this process to the prison if requested. */
1694 	if (flags & JAIL_ATTACH) {
1695 		mtx_lock(&pr->pr_mtx);
1696 		error = do_jail_attach(td, pr);
1697 		if (error) {
1698 			vfs_opterror(opts, "attach failed");
1699 			if (!created)
1700 				prison_deref(pr, PD_DEREF);
1701 			goto done_errmsg;
1702 		}
1703 	}
1704 
1705 	/*
1706 	 * Now that it is all there, drop the temporary reference from existing
1707 	 * prisons.  Or add a reference to newly created persistent prisons
1708 	 * (which was not done earlier so that the prison would not be publicly
1709 	 * visible).
1710 	 */
1711 	if (!created) {
1712 		prison_deref(pr, (flags & JAIL_ATTACH)
1713 		    ? PD_DEREF
1714 		    : PD_DEREF | PD_LIST_SLOCKED);
1715 	} else {
1716 		if (pr_flags & PR_PERSIST) {
1717 			mtx_lock(&pr->pr_mtx);
1718 			pr->pr_ref++;
1719 			pr->pr_uref++;
1720 			mtx_unlock(&pr->pr_mtx);
1721 		}
1722 		if (!(flags & JAIL_ATTACH))
1723 			sx_sunlock(&allprison_lock);
1724 	}
1725 	td->td_retval[0] = pr->pr_id;
1726 	goto done_errmsg;
1727 
1728  done_deref_locked:
1729 	prison_deref(pr, created
1730 	    ? PD_LOCKED | PD_LIST_XLOCKED
1731 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1732 	goto done_releroot;
1733  done_unlock_list:
1734 	sx_xunlock(&allprison_lock);
1735  done_releroot:
1736 	if (root != NULL) {
1737 		vfslocked = VFS_LOCK_GIANT(root->v_mount);
1738 		vrele(root);
1739 		VFS_UNLOCK_GIANT(vfslocked);
1740 	}
1741  done_errmsg:
1742 	if (error) {
1743 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1744 		if (errmsg_len > 0) {
1745 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1746 			if (errmsg_pos > 0) {
1747 				if (optuio->uio_segflg == UIO_SYSSPACE)
1748 					bcopy(errmsg,
1749 					   optuio->uio_iov[errmsg_pos].iov_base,
1750 					   errmsg_len);
1751 				else
1752 					copyout(errmsg,
1753 					   optuio->uio_iov[errmsg_pos].iov_base,
1754 					   errmsg_len);
1755 			}
1756 		}
1757 	}
1758  done_free:
1759 #ifdef INET
1760 	free(ip4, M_PRISON);
1761 #endif
1762 #ifdef INET6
1763 	free(ip6, M_PRISON);
1764 #endif
1765 	vfs_freeopts(opts);
1766 	return (error);
1767 }
1768 
1769 
1770 /*
1771  * struct jail_get_args {
1772  *	struct iovec *iovp;
1773  *	unsigned int iovcnt;
1774  *	int flags;
1775  * };
1776  */
1777 int
1778 jail_get(struct thread *td, struct jail_get_args *uap)
1779 {
1780 	struct uio *auio;
1781 	int error;
1782 
1783 	/* Check that we have an even number of iovecs. */
1784 	if (uap->iovcnt & 1)
1785 		return (EINVAL);
1786 
1787 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1788 	if (error)
1789 		return (error);
1790 	error = kern_jail_get(td, auio, uap->flags);
1791 	if (error == 0)
1792 		error = copyout(auio->uio_iov, uap->iovp,
1793 		    uap->iovcnt * sizeof (struct iovec));
1794 	free(auio, M_IOV);
1795 	return (error);
1796 }
1797 
1798 int
1799 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1800 {
1801 	struct prison *pr, *mypr;
1802 	struct vfsopt *opt;
1803 	struct vfsoptlist *opts;
1804 	char *errmsg, *name;
1805 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1806 
1807 	if (flags & ~JAIL_GET_MASK)
1808 		return (EINVAL);
1809 
1810 	/* Get the parameter list. */
1811 	error = vfs_buildopts(optuio, &opts);
1812 	if (error)
1813 		return (error);
1814 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1815 	mypr = td->td_ucred->cr_prison;
1816 
1817 	/*
1818 	 * Find the prison specified by one of: lastjid, jid, name.
1819 	 */
1820 	sx_slock(&allprison_lock);
1821 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1822 	if (error == 0) {
1823 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1824 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1825 				mtx_lock(&pr->pr_mtx);
1826 				if (pr->pr_ref > 0 &&
1827 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1828 					break;
1829 				mtx_unlock(&pr->pr_mtx);
1830 			}
1831 		}
1832 		if (pr != NULL)
1833 			goto found_prison;
1834 		error = ENOENT;
1835 		vfs_opterror(opts, "no jail after %d", jid);
1836 		goto done_unlock_list;
1837 	} else if (error != ENOENT)
1838 		goto done_unlock_list;
1839 
1840 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1841 	if (error == 0) {
1842 		if (jid != 0) {
1843 			pr = prison_find_child(mypr, jid);
1844 			if (pr != NULL) {
1845 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1846 					mtx_unlock(&pr->pr_mtx);
1847 					error = ENOENT;
1848 					vfs_opterror(opts, "jail %d is dying",
1849 					    jid);
1850 					goto done_unlock_list;
1851 				}
1852 				goto found_prison;
1853 			}
1854 			error = ENOENT;
1855 			vfs_opterror(opts, "jail %d not found", jid);
1856 			goto done_unlock_list;
1857 		}
1858 	} else if (error != ENOENT)
1859 		goto done_unlock_list;
1860 
1861 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1862 	if (error == 0) {
1863 		if (len == 0 || name[len - 1] != '\0') {
1864 			error = EINVAL;
1865 			goto done_unlock_list;
1866 		}
1867 		pr = prison_find_name(mypr, name);
1868 		if (pr != NULL) {
1869 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1870 				mtx_unlock(&pr->pr_mtx);
1871 				error = ENOENT;
1872 				vfs_opterror(opts, "jail \"%s\" is dying",
1873 				    name);
1874 				goto done_unlock_list;
1875 			}
1876 			goto found_prison;
1877 		}
1878 		error = ENOENT;
1879 		vfs_opterror(opts, "jail \"%s\" not found", name);
1880 		goto done_unlock_list;
1881 	} else if (error != ENOENT)
1882 		goto done_unlock_list;
1883 
1884 	vfs_opterror(opts, "no jail specified");
1885 	error = ENOENT;
1886 	goto done_unlock_list;
1887 
1888  found_prison:
1889 	/* Get the parameters of the prison. */
1890 	pr->pr_ref++;
1891 	locked = PD_LOCKED;
1892 	td->td_retval[0] = pr->pr_id;
1893 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
1894 	if (error != 0 && error != ENOENT)
1895 		goto done_deref;
1896 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
1897 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
1898 	if (error != 0 && error != ENOENT)
1899 		goto done_deref;
1900 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
1901 	if (error != 0 && error != ENOENT)
1902 		goto done_deref;
1903 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
1904 	    sizeof(pr->pr_cpuset->cs_id));
1905 	if (error != 0 && error != ENOENT)
1906 		goto done_deref;
1907 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
1908 	if (error != 0 && error != ENOENT)
1909 		goto done_deref;
1910 #ifdef INET
1911 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
1912 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1913 	if (error != 0 && error != ENOENT)
1914 		goto done_deref;
1915 #endif
1916 #ifdef INET6
1917 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
1918 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1919 	if (error != 0 && error != ENOENT)
1920 		goto done_deref;
1921 #endif
1922 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
1923 	    sizeof(pr->pr_securelevel));
1924 	if (error != 0 && error != ENOENT)
1925 		goto done_deref;
1926 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
1927 	    sizeof(pr->pr_childcount));
1928 	if (error != 0 && error != ENOENT)
1929 		goto done_deref;
1930 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
1931 	    sizeof(pr->pr_childmax));
1932 	if (error != 0 && error != ENOENT)
1933 		goto done_deref;
1934 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
1935 	if (error != 0 && error != ENOENT)
1936 		goto done_deref;
1937 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
1938 	if (error != 0 && error != ENOENT)
1939 		goto done_deref;
1940 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
1941 	if (error != 0 && error != ENOENT)
1942 		goto done_deref;
1943 #ifdef COMPAT_IA32
1944 	if (td->td_proc->p_sysent->sv_flags & SV_IA32) {
1945 		uint32_t hid32 = pr->pr_hostid;
1946 
1947 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
1948 	} else
1949 #endif
1950 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
1951 	    sizeof(pr->pr_hostid));
1952 	if (error != 0 && error != ENOENT)
1953 		goto done_deref;
1954 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
1955 	    sizeof(pr->pr_enforce_statfs));
1956 	if (error != 0 && error != ENOENT)
1957 		goto done_deref;
1958 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
1959 	    fi++) {
1960 		if (pr_flag_names[fi] == NULL)
1961 			continue;
1962 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
1963 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
1964 		if (error != 0 && error != ENOENT)
1965 			goto done_deref;
1966 		i = !i;
1967 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
1968 		if (error != 0 && error != ENOENT)
1969 			goto done_deref;
1970 	}
1971 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
1972 	    fi++) {
1973 		if (pr_allow_names[fi] == NULL)
1974 			continue;
1975 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
1976 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
1977 		if (error != 0 && error != ENOENT)
1978 			goto done_deref;
1979 		i = !i;
1980 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
1981 		if (error != 0 && error != ENOENT)
1982 			goto done_deref;
1983 	}
1984 	i = (pr->pr_uref == 0);
1985 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
1986 	if (error != 0 && error != ENOENT)
1987 		goto done_deref;
1988 	i = !i;
1989 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
1990 	if (error != 0 && error != ENOENT)
1991 		goto done_deref;
1992 
1993 	/* Get the module parameters. */
1994 	mtx_unlock(&pr->pr_mtx);
1995 	locked = 0;
1996 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
1997 	if (error)
1998 		goto done_deref;
1999 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2000 
2001 	/* By now, all parameters should have been noted. */
2002 	TAILQ_FOREACH(opt, opts, link) {
2003 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2004 			error = EINVAL;
2005 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2006 			goto done_errmsg;
2007 		}
2008 	}
2009 
2010 	/* Write the fetched parameters back to userspace. */
2011 	error = 0;
2012 	TAILQ_FOREACH(opt, opts, link) {
2013 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2014 			pos = 2 * opt->pos + 1;
2015 			optuio->uio_iov[pos].iov_len = opt->len;
2016 			if (opt->value != NULL) {
2017 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2018 					bcopy(opt->value,
2019 					    optuio->uio_iov[pos].iov_base,
2020 					    opt->len);
2021 				} else {
2022 					error = copyout(opt->value,
2023 					    optuio->uio_iov[pos].iov_base,
2024 					    opt->len);
2025 					if (error)
2026 						break;
2027 				}
2028 			}
2029 		}
2030 	}
2031 	goto done_errmsg;
2032 
2033  done_deref:
2034 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2035 	goto done_errmsg;
2036 
2037  done_unlock_list:
2038 	sx_sunlock(&allprison_lock);
2039  done_errmsg:
2040 	if (error && errmsg_pos >= 0) {
2041 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2042 		errmsg_pos = 2 * errmsg_pos + 1;
2043 		if (errmsg_len > 0) {
2044 			if (optuio->uio_segflg == UIO_SYSSPACE)
2045 				bcopy(errmsg,
2046 				    optuio->uio_iov[errmsg_pos].iov_base,
2047 				    errmsg_len);
2048 			else
2049 				copyout(errmsg,
2050 				    optuio->uio_iov[errmsg_pos].iov_base,
2051 				    errmsg_len);
2052 		}
2053 	}
2054 	vfs_freeopts(opts);
2055 	return (error);
2056 }
2057 
2058 
2059 /*
2060  * struct jail_remove_args {
2061  *	int jid;
2062  * };
2063  */
2064 int
2065 jail_remove(struct thread *td, struct jail_remove_args *uap)
2066 {
2067 	struct prison *pr, *cpr, *lpr, *tpr;
2068 	int descend, error;
2069 
2070 	error = priv_check(td, PRIV_JAIL_REMOVE);
2071 	if (error)
2072 		return (error);
2073 
2074 	sx_xlock(&allprison_lock);
2075 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2076 	if (pr == NULL) {
2077 		sx_xunlock(&allprison_lock);
2078 		return (EINVAL);
2079 	}
2080 
2081 	/* Remove all descendants of this prison, then remove this prison. */
2082 	pr->pr_ref++;
2083 	pr->pr_flags |= PR_REMOVE;
2084 	if (!LIST_EMPTY(&pr->pr_children)) {
2085 		mtx_unlock(&pr->pr_mtx);
2086 		lpr = NULL;
2087 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2088 			mtx_lock(&cpr->pr_mtx);
2089 			if (cpr->pr_ref > 0) {
2090 				tpr = cpr;
2091 				cpr->pr_ref++;
2092 				cpr->pr_flags |= PR_REMOVE;
2093 			} else {
2094 				/* Already removed - do not do it again. */
2095 				tpr = NULL;
2096 			}
2097 			mtx_unlock(&cpr->pr_mtx);
2098 			if (lpr != NULL) {
2099 				mtx_lock(&lpr->pr_mtx);
2100 				prison_remove_one(lpr);
2101 				sx_xlock(&allprison_lock);
2102 			}
2103 			lpr = tpr;
2104 		}
2105 		if (lpr != NULL) {
2106 			mtx_lock(&lpr->pr_mtx);
2107 			prison_remove_one(lpr);
2108 			sx_xlock(&allprison_lock);
2109 		}
2110 		mtx_lock(&pr->pr_mtx);
2111 	}
2112 	prison_remove_one(pr);
2113 	return (0);
2114 }
2115 
2116 static void
2117 prison_remove_one(struct prison *pr)
2118 {
2119 	struct proc *p;
2120 	int deuref;
2121 
2122 	/* If the prison was persistent, it is not anymore. */
2123 	deuref = 0;
2124 	if (pr->pr_flags & PR_PERSIST) {
2125 		pr->pr_ref--;
2126 		deuref = PD_DEUREF;
2127 		pr->pr_flags &= ~PR_PERSIST;
2128 	}
2129 
2130 	/*
2131 	 * jail_remove added a reference.  If that's the only one, remove
2132 	 * the prison now.
2133 	 */
2134 	KASSERT(pr->pr_ref > 0,
2135 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2136 	if (pr->pr_ref == 1) {
2137 		prison_deref(pr,
2138 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2139 		return;
2140 	}
2141 
2142 	mtx_unlock(&pr->pr_mtx);
2143 	sx_xunlock(&allprison_lock);
2144 	/*
2145 	 * Kill all processes unfortunate enough to be attached to this prison.
2146 	 */
2147 	sx_slock(&allproc_lock);
2148 	LIST_FOREACH(p, &allproc, p_list) {
2149 		PROC_LOCK(p);
2150 		if (p->p_state != PRS_NEW && p->p_ucred &&
2151 		    p->p_ucred->cr_prison == pr)
2152 			psignal(p, SIGKILL);
2153 		PROC_UNLOCK(p);
2154 	}
2155 	sx_sunlock(&allproc_lock);
2156 	/* Remove the temporary reference added by jail_remove. */
2157 	prison_deref(pr, deuref | PD_DEREF);
2158 }
2159 
2160 
2161 /*
2162  * struct jail_attach_args {
2163  *	int jid;
2164  * };
2165  */
2166 int
2167 jail_attach(struct thread *td, struct jail_attach_args *uap)
2168 {
2169 	struct prison *pr;
2170 	int error;
2171 
2172 	error = priv_check(td, PRIV_JAIL_ATTACH);
2173 	if (error)
2174 		return (error);
2175 
2176 	sx_slock(&allprison_lock);
2177 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2178 	if (pr == NULL) {
2179 		sx_sunlock(&allprison_lock);
2180 		return (EINVAL);
2181 	}
2182 
2183 	/*
2184 	 * Do not allow a process to attach to a prison that is not
2185 	 * considered to be "alive".
2186 	 */
2187 	if (pr->pr_uref == 0) {
2188 		mtx_unlock(&pr->pr_mtx);
2189 		sx_sunlock(&allprison_lock);
2190 		return (EINVAL);
2191 	}
2192 
2193 	return (do_jail_attach(td, pr));
2194 }
2195 
2196 static int
2197 do_jail_attach(struct thread *td, struct prison *pr)
2198 {
2199 	struct prison *ppr;
2200 	struct proc *p;
2201 	struct ucred *newcred, *oldcred;
2202 	int vfslocked, error;
2203 
2204 	/*
2205 	 * XXX: Note that there is a slight race here if two threads
2206 	 * in the same privileged process attempt to attach to two
2207 	 * different jails at the same time.  It is important for
2208 	 * user processes not to do this, or they might end up with
2209 	 * a process root from one prison, but attached to the jail
2210 	 * of another.
2211 	 */
2212 	pr->pr_ref++;
2213 	pr->pr_uref++;
2214 	mtx_unlock(&pr->pr_mtx);
2215 
2216 	/* Let modules do whatever they need to prepare for attaching. */
2217 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2218 	if (error) {
2219 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2220 		return (error);
2221 	}
2222 	sx_sunlock(&allprison_lock);
2223 
2224 	/*
2225 	 * Reparent the newly attached process to this jail.
2226 	 */
2227 	ppr = td->td_ucred->cr_prison;
2228 	p = td->td_proc;
2229 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2230 	if (error)
2231 		goto e_revert_osd;
2232 
2233 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2234 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2235 	if ((error = change_dir(pr->pr_root, td)) != 0)
2236 		goto e_unlock;
2237 #ifdef MAC
2238 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2239 		goto e_unlock;
2240 #endif
2241 	VOP_UNLOCK(pr->pr_root, 0);
2242 	if ((error = change_root(pr->pr_root, td)))
2243 		goto e_unlock_giant;
2244 	VFS_UNLOCK_GIANT(vfslocked);
2245 
2246 	newcred = crget();
2247 	PROC_LOCK(p);
2248 	oldcred = p->p_ucred;
2249 	setsugid(p);
2250 	crcopy(newcred, oldcred);
2251 	newcred->cr_prison = pr;
2252 	p->p_ucred = newcred;
2253 	PROC_UNLOCK(p);
2254 	crfree(oldcred);
2255 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2256 	return (0);
2257  e_unlock:
2258 	VOP_UNLOCK(pr->pr_root, 0);
2259  e_unlock_giant:
2260 	VFS_UNLOCK_GIANT(vfslocked);
2261  e_revert_osd:
2262 	/* Tell modules this thread is still in its old jail after all. */
2263 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2264 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2265 	return (error);
2266 }
2267 
2268 
2269 /*
2270  * Returns a locked prison instance, or NULL on failure.
2271  */
2272 struct prison *
2273 prison_find(int prid)
2274 {
2275 	struct prison *pr;
2276 
2277 	sx_assert(&allprison_lock, SX_LOCKED);
2278 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2279 		if (pr->pr_id == prid) {
2280 			mtx_lock(&pr->pr_mtx);
2281 			if (pr->pr_ref > 0)
2282 				return (pr);
2283 			mtx_unlock(&pr->pr_mtx);
2284 		}
2285 	}
2286 	return (NULL);
2287 }
2288 
2289 /*
2290  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2291  */
2292 struct prison *
2293 prison_find_child(struct prison *mypr, int prid)
2294 {
2295 	struct prison *pr;
2296 	int descend;
2297 
2298 	sx_assert(&allprison_lock, SX_LOCKED);
2299 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2300 		if (pr->pr_id == prid) {
2301 			mtx_lock(&pr->pr_mtx);
2302 			if (pr->pr_ref > 0)
2303 				return (pr);
2304 			mtx_unlock(&pr->pr_mtx);
2305 		}
2306 	}
2307 	return (NULL);
2308 }
2309 
2310 /*
2311  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2312  */
2313 struct prison *
2314 prison_find_name(struct prison *mypr, const char *name)
2315 {
2316 	struct prison *pr, *deadpr;
2317 	size_t mylen;
2318 	int descend;
2319 
2320 	sx_assert(&allprison_lock, SX_LOCKED);
2321 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2322  again:
2323 	deadpr = NULL;
2324 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2325 		if (!strcmp(pr->pr_name + mylen, name)) {
2326 			mtx_lock(&pr->pr_mtx);
2327 			if (pr->pr_ref > 0) {
2328 				if (pr->pr_uref > 0)
2329 					return (pr);
2330 				deadpr = pr;
2331 			}
2332 			mtx_unlock(&pr->pr_mtx);
2333 		}
2334 	}
2335 	/* There was no valid prison - perhaps there was a dying one. */
2336 	if (deadpr != NULL) {
2337 		mtx_lock(&deadpr->pr_mtx);
2338 		if (deadpr->pr_ref == 0) {
2339 			mtx_unlock(&deadpr->pr_mtx);
2340 			goto again;
2341 		}
2342 	}
2343 	return (deadpr);
2344 }
2345 
2346 /*
2347  * See if a prison has the specific flag set.
2348  */
2349 int
2350 prison_flag(struct ucred *cred, unsigned flag)
2351 {
2352 
2353 	/* This is an atomic read, so no locking is necessary. */
2354 	return (cred->cr_prison->pr_flags & flag);
2355 }
2356 
2357 int
2358 prison_allow(struct ucred *cred, unsigned flag)
2359 {
2360 
2361 	/* This is an atomic read, so no locking is necessary. */
2362 	return (cred->cr_prison->pr_allow & flag);
2363 }
2364 
2365 /*
2366  * Remove a prison reference.  If that was the last reference, remove the
2367  * prison itself - but not in this context in case there are locks held.
2368  */
2369 void
2370 prison_free_locked(struct prison *pr)
2371 {
2372 
2373 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2374 	pr->pr_ref--;
2375 	if (pr->pr_ref == 0) {
2376 		mtx_unlock(&pr->pr_mtx);
2377 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2378 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2379 		return;
2380 	}
2381 	mtx_unlock(&pr->pr_mtx);
2382 }
2383 
2384 void
2385 prison_free(struct prison *pr)
2386 {
2387 
2388 	mtx_lock(&pr->pr_mtx);
2389 	prison_free_locked(pr);
2390 }
2391 
2392 static void
2393 prison_complete(void *context, int pending)
2394 {
2395 
2396 	prison_deref((struct prison *)context, 0);
2397 }
2398 
2399 /*
2400  * Remove a prison reference (usually).  This internal version assumes no
2401  * mutexes are held, except perhaps the prison itself.  If there are no more
2402  * references, release and delist the prison.  On completion, the prison lock
2403  * and the allprison lock are both unlocked.
2404  */
2405 static void
2406 prison_deref(struct prison *pr, int flags)
2407 {
2408 	struct prison *ppr, *tpr;
2409 	int vfslocked;
2410 
2411 	if (!(flags & PD_LOCKED))
2412 		mtx_lock(&pr->pr_mtx);
2413 	/* Decrement the user references in a separate loop. */
2414 	if (flags & PD_DEUREF) {
2415 		for (tpr = pr;; tpr = tpr->pr_parent) {
2416 			if (tpr != pr)
2417 				mtx_lock(&tpr->pr_mtx);
2418 			if (--tpr->pr_uref > 0)
2419 				break;
2420 			KASSERT(tpr != &prison0, ("prison0 pr_uref=0"));
2421 			mtx_unlock(&tpr->pr_mtx);
2422 		}
2423 		/* Done if there were only user references to remove. */
2424 		if (!(flags & PD_DEREF)) {
2425 			mtx_unlock(&tpr->pr_mtx);
2426 			if (flags & PD_LIST_SLOCKED)
2427 				sx_sunlock(&allprison_lock);
2428 			else if (flags & PD_LIST_XLOCKED)
2429 				sx_xunlock(&allprison_lock);
2430 			return;
2431 		}
2432 		if (tpr != pr) {
2433 			mtx_unlock(&tpr->pr_mtx);
2434 			mtx_lock(&pr->pr_mtx);
2435 		}
2436 	}
2437 
2438 	for (;;) {
2439 		if (flags & PD_DEREF)
2440 			pr->pr_ref--;
2441 		/* If the prison still has references, nothing else to do. */
2442 		if (pr->pr_ref > 0) {
2443 			mtx_unlock(&pr->pr_mtx);
2444 			if (flags & PD_LIST_SLOCKED)
2445 				sx_sunlock(&allprison_lock);
2446 			else if (flags & PD_LIST_XLOCKED)
2447 				sx_xunlock(&allprison_lock);
2448 			return;
2449 		}
2450 
2451 		mtx_unlock(&pr->pr_mtx);
2452 		if (flags & PD_LIST_SLOCKED) {
2453 			if (!sx_try_upgrade(&allprison_lock)) {
2454 				sx_sunlock(&allprison_lock);
2455 				sx_xlock(&allprison_lock);
2456 			}
2457 		} else if (!(flags & PD_LIST_XLOCKED))
2458 			sx_xlock(&allprison_lock);
2459 
2460 		TAILQ_REMOVE(&allprison, pr, pr_list);
2461 		LIST_REMOVE(pr, pr_sibling);
2462 		ppr = pr->pr_parent;
2463 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2464 			tpr->pr_childcount--;
2465 		sx_downgrade(&allprison_lock);
2466 
2467 #ifdef VIMAGE
2468 		if (pr->pr_flags & PR_VNET)
2469 			vnet_destroy(pr->pr_vnet);
2470 #endif
2471 		if (pr->pr_root != NULL) {
2472 			vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2473 			vrele(pr->pr_root);
2474 			VFS_UNLOCK_GIANT(vfslocked);
2475 		}
2476 		mtx_destroy(&pr->pr_mtx);
2477 #ifdef INET
2478 		free(pr->pr_ip4, M_PRISON);
2479 #endif
2480 #ifdef INET6
2481 		free(pr->pr_ip6, M_PRISON);
2482 #endif
2483 		if (pr->pr_cpuset != NULL)
2484 			cpuset_rel(pr->pr_cpuset);
2485 		osd_jail_exit(pr);
2486 		free(pr, M_PRISON);
2487 
2488 		/* Removing a prison frees a reference on its parent. */
2489 		pr = ppr;
2490 		mtx_lock(&pr->pr_mtx);
2491 		flags = PD_DEREF | PD_LIST_SLOCKED;
2492 	}
2493 }
2494 
2495 void
2496 prison_hold_locked(struct prison *pr)
2497 {
2498 
2499 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2500 	KASSERT(pr->pr_ref > 0,
2501 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2502 	pr->pr_ref++;
2503 }
2504 
2505 void
2506 prison_hold(struct prison *pr)
2507 {
2508 
2509 	mtx_lock(&pr->pr_mtx);
2510 	prison_hold_locked(pr);
2511 	mtx_unlock(&pr->pr_mtx);
2512 }
2513 
2514 void
2515 prison_proc_hold(struct prison *pr)
2516 {
2517 
2518 	mtx_lock(&pr->pr_mtx);
2519 	KASSERT(pr->pr_uref > 0,
2520 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2521 	pr->pr_uref++;
2522 	mtx_unlock(&pr->pr_mtx);
2523 }
2524 
2525 void
2526 prison_proc_free(struct prison *pr)
2527 {
2528 
2529 	mtx_lock(&pr->pr_mtx);
2530 	KASSERT(pr->pr_uref > 0,
2531 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2532 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2533 }
2534 
2535 
2536 #ifdef INET
2537 /*
2538  * Restrict a prison's IP address list with its parent's, possibly replacing
2539  * it.  Return true if the replacement buffer was used (or would have been).
2540  */
2541 static int
2542 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2543 {
2544 	int ii, ij, used;
2545 	struct prison *ppr;
2546 
2547 	ppr = pr->pr_parent;
2548 	if (!(pr->pr_flags & PR_IP4_USER)) {
2549 		/* This has no user settings, so just copy the parent's list. */
2550 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2551 			/*
2552 			 * There's no room for the parent's list.  Use the
2553 			 * new list buffer, which is assumed to be big enough
2554 			 * (if it was passed).  If there's no buffer, try to
2555 			 * allocate one.
2556 			 */
2557 			used = 1;
2558 			if (newip4 == NULL) {
2559 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2560 				    M_PRISON, M_NOWAIT);
2561 				if (newip4 != NULL)
2562 					used = 0;
2563 			}
2564 			if (newip4 != NULL) {
2565 				bcopy(ppr->pr_ip4, newip4,
2566 				    ppr->pr_ip4s * sizeof(*newip4));
2567 				free(pr->pr_ip4, M_PRISON);
2568 				pr->pr_ip4 = newip4;
2569 				pr->pr_ip4s = ppr->pr_ip4s;
2570 				pr->pr_flags |= PR_IP4;
2571 			}
2572 			return (used);
2573 		}
2574 		pr->pr_ip4s = ppr->pr_ip4s;
2575 		if (pr->pr_ip4s > 0)
2576 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2577 			    pr->pr_ip4s * sizeof(*newip4));
2578 		else if (pr->pr_ip4 != NULL) {
2579 			free(pr->pr_ip4, M_PRISON);
2580 			pr->pr_ip4 = NULL;
2581 		}
2582 		pr->pr_flags =
2583 			(pr->pr_flags & ~PR_IP4) | (ppr->pr_flags & PR_IP4);
2584 	} else if (pr->pr_ip4s > 0 && (ppr->pr_flags & PR_IP4)) {
2585 		/* Remove addresses that aren't in the parent. */
2586 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2587 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2588 				break;
2589 		if (ij < ppr->pr_ip4s)
2590 			ii = 1;
2591 		else {
2592 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2593 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2594 			ii = 0;
2595 		}
2596 		for (ij = 1; ii < pr->pr_ip4s; ) {
2597 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2598 				ii++;
2599 				continue;
2600 			}
2601 			switch (ij >= ppr->pr_ip4s ? -1 :
2602 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2603 			case -1:
2604 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2605 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2606 				break;
2607 			case 0:
2608 				ii++;
2609 				ij++;
2610 				break;
2611 			case 1:
2612 				ij++;
2613 				break;
2614 			}
2615 		}
2616 		if (pr->pr_ip4s == 0) {
2617 			free(pr->pr_ip4, M_PRISON);
2618 			pr->pr_ip4 = NULL;
2619 		}
2620 	}
2621 	return (0);
2622 }
2623 
2624 /*
2625  * Pass back primary IPv4 address of this jail.
2626  *
2627  * If not restricted return success but do not alter the address.  Caller has
2628  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2629  *
2630  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2631  * Address returned in NBO.
2632  */
2633 int
2634 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2635 {
2636 	struct prison *pr;
2637 
2638 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2639 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2640 
2641 	pr = cred->cr_prison;
2642 	if (!(pr->pr_flags & PR_IP4))
2643 		return (0);
2644 	mtx_lock(&pr->pr_mtx);
2645 	if (!(pr->pr_flags & PR_IP4)) {
2646 		mtx_unlock(&pr->pr_mtx);
2647 		return (0);
2648 	}
2649 	if (pr->pr_ip4 == NULL) {
2650 		mtx_unlock(&pr->pr_mtx);
2651 		return (EAFNOSUPPORT);
2652 	}
2653 
2654 	ia->s_addr = pr->pr_ip4[0].s_addr;
2655 	mtx_unlock(&pr->pr_mtx);
2656 	return (0);
2657 }
2658 
2659 /*
2660  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2661  */
2662 int
2663 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2664 {
2665 
2666 	if (pr1 == pr2)
2667 		return (1);
2668 
2669 	/*
2670 	 * jail_set maintains an exclusive hold on allprison_lock while it
2671 	 * changes the IP addresses, so only a shared hold is needed.  This is
2672 	 * easier than locking the two prisons which would require finding the
2673 	 * proper locking order and end up needing allprison_lock anyway.
2674 	 */
2675 	sx_slock(&allprison_lock);
2676 	while (pr1 != &prison0 && !(pr1->pr_flags & PR_IP4_USER))
2677 		pr1 = pr1->pr_parent;
2678 	while (pr2 != &prison0 && !(pr2->pr_flags & PR_IP4_USER))
2679 		pr2 = pr2->pr_parent;
2680 	sx_sunlock(&allprison_lock);
2681 	return (pr1 == pr2);
2682 }
2683 
2684 /*
2685  * Make sure our (source) address is set to something meaningful to this
2686  * jail.
2687  *
2688  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2689  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2690  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2691  */
2692 int
2693 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2694 {
2695 	struct prison *pr;
2696 	struct in_addr ia0;
2697 	int error;
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 	ia0.s_addr = ntohl(ia->s_addr);
2716 	if (ia0.s_addr == INADDR_LOOPBACK) {
2717 		ia->s_addr = pr->pr_ip4[0].s_addr;
2718 		mtx_unlock(&pr->pr_mtx);
2719 		return (0);
2720 	}
2721 
2722 	if (ia0.s_addr == INADDR_ANY) {
2723 		/*
2724 		 * In case there is only 1 IPv4 address, bind directly.
2725 		 */
2726 		if (pr->pr_ip4s == 1)
2727 			ia->s_addr = pr->pr_ip4[0].s_addr;
2728 		mtx_unlock(&pr->pr_mtx);
2729 		return (0);
2730 	}
2731 
2732 	error = _prison_check_ip4(pr, ia);
2733 	mtx_unlock(&pr->pr_mtx);
2734 	return (error);
2735 }
2736 
2737 /*
2738  * Rewrite destination address in case we will connect to loopback address.
2739  *
2740  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2741  * Address passed in in NBO and returned in NBO.
2742  */
2743 int
2744 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2745 {
2746 	struct prison *pr;
2747 
2748 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2749 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2750 
2751 	pr = cred->cr_prison;
2752 	if (!(pr->pr_flags & PR_IP4))
2753 		return (0);
2754 	mtx_lock(&pr->pr_mtx);
2755 	if (!(pr->pr_flags & PR_IP4)) {
2756 		mtx_unlock(&pr->pr_mtx);
2757 		return (0);
2758 	}
2759 	if (pr->pr_ip4 == NULL) {
2760 		mtx_unlock(&pr->pr_mtx);
2761 		return (EAFNOSUPPORT);
2762 	}
2763 
2764 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2765 		ia->s_addr = pr->pr_ip4[0].s_addr;
2766 		mtx_unlock(&pr->pr_mtx);
2767 		return (0);
2768 	}
2769 
2770 	/*
2771 	 * Return success because nothing had to be changed.
2772 	 */
2773 	mtx_unlock(&pr->pr_mtx);
2774 	return (0);
2775 }
2776 
2777 /*
2778  * Check if given address belongs to the jail referenced by cred/prison.
2779  *
2780  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2781  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2782  * doesn't allow IPv4.  Address passed in in NBO.
2783  */
2784 static int
2785 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
2786 {
2787 	int i, a, z, d;
2788 
2789 	/*
2790 	 * Check the primary IP.
2791 	 */
2792 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2793 		return (0);
2794 
2795 	/*
2796 	 * All the other IPs are sorted so we can do a binary search.
2797 	 */
2798 	a = 0;
2799 	z = pr->pr_ip4s - 2;
2800 	while (a <= z) {
2801 		i = (a + z) / 2;
2802 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2803 		if (d > 0)
2804 			z = i - 1;
2805 		else if (d < 0)
2806 			a = i + 1;
2807 		else
2808 			return (0);
2809 	}
2810 
2811 	return (EADDRNOTAVAIL);
2812 }
2813 
2814 int
2815 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
2816 {
2817 	struct prison *pr;
2818 	int error;
2819 
2820 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2821 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2822 
2823 	pr = cred->cr_prison;
2824 	if (!(pr->pr_flags & PR_IP4))
2825 		return (0);
2826 	mtx_lock(&pr->pr_mtx);
2827 	if (!(pr->pr_flags & PR_IP4)) {
2828 		mtx_unlock(&pr->pr_mtx);
2829 		return (0);
2830 	}
2831 	if (pr->pr_ip4 == NULL) {
2832 		mtx_unlock(&pr->pr_mtx);
2833 		return (EAFNOSUPPORT);
2834 	}
2835 
2836 	error = _prison_check_ip4(pr, ia);
2837 	mtx_unlock(&pr->pr_mtx);
2838 	return (error);
2839 }
2840 #endif
2841 
2842 #ifdef INET6
2843 static int
2844 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2845 {
2846 	int ii, ij, used;
2847 	struct prison *ppr;
2848 
2849 	ppr = pr->pr_parent;
2850 	if (!(pr->pr_flags & PR_IP6_USER)) {
2851 		/* This has no user settings, so just copy the parent's list. */
2852 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2853 			/*
2854 			 * There's no room for the parent's list.  Use the
2855 			 * new list buffer, which is assumed to be big enough
2856 			 * (if it was passed).  If there's no buffer, try to
2857 			 * allocate one.
2858 			 */
2859 			used = 1;
2860 			if (newip6 == NULL) {
2861 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
2862 				    M_PRISON, M_NOWAIT);
2863 				if (newip6 != NULL)
2864 					used = 0;
2865 			}
2866 			if (newip6 != NULL) {
2867 				bcopy(ppr->pr_ip6, newip6,
2868 				    ppr->pr_ip6s * sizeof(*newip6));
2869 				free(pr->pr_ip6, M_PRISON);
2870 				pr->pr_ip6 = newip6;
2871 				pr->pr_ip6s = ppr->pr_ip6s;
2872 				pr->pr_flags |= PR_IP6;
2873 			}
2874 			return (used);
2875 		}
2876 		pr->pr_ip6s = ppr->pr_ip6s;
2877 		if (pr->pr_ip6s > 0)
2878 			bcopy(ppr->pr_ip6, pr->pr_ip6,
2879 			    pr->pr_ip6s * sizeof(*newip6));
2880 		else if (pr->pr_ip6 != NULL) {
2881 			free(pr->pr_ip6, M_PRISON);
2882 			pr->pr_ip6 = NULL;
2883 		}
2884 		pr->pr_flags =
2885 			(pr->pr_flags & ~PR_IP6) | (ppr->pr_flags & PR_IP6);
2886 	} else if (pr->pr_ip6s > 0 && (ppr->pr_flags & PR_IP6)) {
2887 		/* Remove addresses that aren't in the parent. */
2888 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
2889 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
2890 			    &ppr->pr_ip6[ij]))
2891 				break;
2892 		if (ij < ppr->pr_ip6s)
2893 			ii = 1;
2894 		else {
2895 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
2896 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
2897 			ii = 0;
2898 		}
2899 		for (ij = 1; ii < pr->pr_ip6s; ) {
2900 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
2901 			    &ppr->pr_ip6[0])) {
2902 				ii++;
2903 				continue;
2904 			}
2905 			switch (ij >= ppr->pr_ip4s ? -1 :
2906 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
2907 			case -1:
2908 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
2909 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
2910 				break;
2911 			case 0:
2912 				ii++;
2913 				ij++;
2914 				break;
2915 			case 1:
2916 				ij++;
2917 				break;
2918 			}
2919 		}
2920 		if (pr->pr_ip6s == 0) {
2921 			free(pr->pr_ip6, M_PRISON);
2922 			pr->pr_ip6 = NULL;
2923 		}
2924 	}
2925 	return 0;
2926 }
2927 
2928 /*
2929  * Pass back primary IPv6 address for this jail.
2930  *
2931  * If not restricted return success but do not alter the address.  Caller has
2932  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
2933  *
2934  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
2935  */
2936 int
2937 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
2938 {
2939 	struct prison *pr;
2940 
2941 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2942 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
2943 
2944 	pr = cred->cr_prison;
2945 	if (!(pr->pr_flags & PR_IP6))
2946 		return (0);
2947 	mtx_lock(&pr->pr_mtx);
2948 	if (!(pr->pr_flags & PR_IP6)) {
2949 		mtx_unlock(&pr->pr_mtx);
2950 		return (0);
2951 	}
2952 	if (pr->pr_ip6 == NULL) {
2953 		mtx_unlock(&pr->pr_mtx);
2954 		return (EAFNOSUPPORT);
2955 	}
2956 
2957 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
2958 	mtx_unlock(&pr->pr_mtx);
2959 	return (0);
2960 }
2961 
2962 /*
2963  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
2964  */
2965 int
2966 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
2967 {
2968 
2969 	if (pr1 == pr2)
2970 		return (1);
2971 
2972 	sx_slock(&allprison_lock);
2973 	while (pr1 != &prison0 && !(pr1->pr_flags & PR_IP6_USER))
2974 		pr1 = pr1->pr_parent;
2975 	while (pr2 != &prison0 && !(pr2->pr_flags & PR_IP6_USER))
2976 		pr2 = pr2->pr_parent;
2977 	sx_sunlock(&allprison_lock);
2978 	return (pr1 == pr2);
2979 }
2980 
2981 /*
2982  * Make sure our (source) address is set to something meaningful to this jail.
2983  *
2984  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
2985  * when needed while binding.
2986  *
2987  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
2988  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2989  * doesn't allow IPv6.
2990  */
2991 int
2992 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
2993 {
2994 	struct prison *pr;
2995 	int error;
2996 
2997 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2998 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
2999 
3000 	pr = cred->cr_prison;
3001 	if (!(pr->pr_flags & PR_IP6))
3002 		return (0);
3003 	mtx_lock(&pr->pr_mtx);
3004 	if (!(pr->pr_flags & PR_IP6)) {
3005 		mtx_unlock(&pr->pr_mtx);
3006 		return (0);
3007 	}
3008 	if (pr->pr_ip6 == NULL) {
3009 		mtx_unlock(&pr->pr_mtx);
3010 		return (EAFNOSUPPORT);
3011 	}
3012 
3013 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3014 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3015 		mtx_unlock(&pr->pr_mtx);
3016 		return (0);
3017 	}
3018 
3019 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3020 		/*
3021 		 * In case there is only 1 IPv6 address, and v6only is true,
3022 		 * then bind directly.
3023 		 */
3024 		if (v6only != 0 && pr->pr_ip6s == 1)
3025 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3026 		mtx_unlock(&pr->pr_mtx);
3027 		return (0);
3028 	}
3029 
3030 	error = _prison_check_ip6(pr, ia6);
3031 	mtx_unlock(&pr->pr_mtx);
3032 	return (error);
3033 }
3034 
3035 /*
3036  * Rewrite destination address in case we will connect to loopback address.
3037  *
3038  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3039  */
3040 int
3041 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3042 {
3043 	struct prison *pr;
3044 
3045 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3046 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3047 
3048 	pr = cred->cr_prison;
3049 	if (!(pr->pr_flags & PR_IP6))
3050 		return (0);
3051 	mtx_lock(&pr->pr_mtx);
3052 	if (!(pr->pr_flags & PR_IP6)) {
3053 		mtx_unlock(&pr->pr_mtx);
3054 		return (0);
3055 	}
3056 	if (pr->pr_ip6 == NULL) {
3057 		mtx_unlock(&pr->pr_mtx);
3058 		return (EAFNOSUPPORT);
3059 	}
3060 
3061 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3062 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3063 		mtx_unlock(&pr->pr_mtx);
3064 		return (0);
3065 	}
3066 
3067 	/*
3068 	 * Return success because nothing had to be changed.
3069 	 */
3070 	mtx_unlock(&pr->pr_mtx);
3071 	return (0);
3072 }
3073 
3074 /*
3075  * Check if given address belongs to the jail referenced by cred/prison.
3076  *
3077  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3078  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3079  * doesn't allow IPv6.
3080  */
3081 static int
3082 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3083 {
3084 	int i, a, z, d;
3085 
3086 	/*
3087 	 * Check the primary IP.
3088 	 */
3089 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3090 		return (0);
3091 
3092 	/*
3093 	 * All the other IPs are sorted so we can do a binary search.
3094 	 */
3095 	a = 0;
3096 	z = pr->pr_ip6s - 2;
3097 	while (a <= z) {
3098 		i = (a + z) / 2;
3099 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3100 		if (d > 0)
3101 			z = i - 1;
3102 		else if (d < 0)
3103 			a = i + 1;
3104 		else
3105 			return (0);
3106 	}
3107 
3108 	return (EADDRNOTAVAIL);
3109 }
3110 
3111 int
3112 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3113 {
3114 	struct prison *pr;
3115 	int error;
3116 
3117 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3118 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3119 
3120 	pr = cred->cr_prison;
3121 	if (!(pr->pr_flags & PR_IP6))
3122 		return (0);
3123 	mtx_lock(&pr->pr_mtx);
3124 	if (!(pr->pr_flags & PR_IP6)) {
3125 		mtx_unlock(&pr->pr_mtx);
3126 		return (0);
3127 	}
3128 	if (pr->pr_ip6 == NULL) {
3129 		mtx_unlock(&pr->pr_mtx);
3130 		return (EAFNOSUPPORT);
3131 	}
3132 
3133 	error = _prison_check_ip6(pr, ia6);
3134 	mtx_unlock(&pr->pr_mtx);
3135 	return (error);
3136 }
3137 #endif
3138 
3139 /*
3140  * Check if a jail supports the given address family.
3141  *
3142  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3143  * if not.
3144  */
3145 int
3146 prison_check_af(struct ucred *cred, int af)
3147 {
3148 	struct prison *pr;
3149 	int error;
3150 
3151 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3152 
3153 	pr = cred->cr_prison;
3154 #ifdef VIMAGE
3155 	/* Prisons with their own network stack are not limited. */
3156 	if (pr->pr_flags & PR_VNET)
3157 		return (0);
3158 #endif
3159 
3160 	error = 0;
3161 	switch (af)
3162 	{
3163 #ifdef INET
3164 	case AF_INET:
3165 		if (pr->pr_flags & PR_IP4)
3166 		{
3167 			mtx_lock(&pr->pr_mtx);
3168 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3169 				error = EAFNOSUPPORT;
3170 			mtx_unlock(&pr->pr_mtx);
3171 		}
3172 		break;
3173 #endif
3174 #ifdef INET6
3175 	case AF_INET6:
3176 		if (pr->pr_flags & PR_IP6)
3177 		{
3178 			mtx_lock(&pr->pr_mtx);
3179 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3180 				error = EAFNOSUPPORT;
3181 			mtx_unlock(&pr->pr_mtx);
3182 		}
3183 		break;
3184 #endif
3185 	case AF_LOCAL:
3186 	case AF_ROUTE:
3187 		break;
3188 	default:
3189 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3190 			error = EAFNOSUPPORT;
3191 	}
3192 	return (error);
3193 }
3194 
3195 /*
3196  * Check if given address belongs to the jail referenced by cred (wrapper to
3197  * prison_check_ip[46]).
3198  *
3199  * Returns 0 if jail doesn't restrict the address family or if address belongs
3200  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3201  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3202  */
3203 int
3204 prison_if(struct ucred *cred, struct sockaddr *sa)
3205 {
3206 #ifdef INET
3207 	struct sockaddr_in *sai;
3208 #endif
3209 #ifdef INET6
3210 	struct sockaddr_in6 *sai6;
3211 #endif
3212 	int error;
3213 
3214 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3215 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3216 
3217 	error = 0;
3218 	switch (sa->sa_family)
3219 	{
3220 #ifdef INET
3221 	case AF_INET:
3222 		sai = (struct sockaddr_in *)sa;
3223 		error = prison_check_ip4(cred, &sai->sin_addr);
3224 		break;
3225 #endif
3226 #ifdef INET6
3227 	case AF_INET6:
3228 		sai6 = (struct sockaddr_in6 *)sa;
3229 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3230 		break;
3231 #endif
3232 	default:
3233 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3234 			error = EAFNOSUPPORT;
3235 	}
3236 	return (error);
3237 }
3238 
3239 /*
3240  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3241  */
3242 int
3243 prison_check(struct ucred *cred1, struct ucred *cred2)
3244 {
3245 
3246 #ifdef VIMAGE
3247 	if (cred2->cr_vimage->v_procg != cred1->cr_vimage->v_procg)
3248 		return (ESRCH);
3249 #endif
3250 	return ((cred1->cr_prison == cred2->cr_prison ||
3251 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3252 }
3253 
3254 /*
3255  * Return 1 if p2 is a child of p1, otherwise 0.
3256  */
3257 int
3258 prison_ischild(struct prison *pr1, struct prison *pr2)
3259 {
3260 
3261 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3262 		if (pr1 == pr2)
3263 			return (1);
3264 	return (0);
3265 }
3266 
3267 /*
3268  * Return 1 if the passed credential is in a jail, otherwise 0.
3269  */
3270 int
3271 jailed(struct ucred *cred)
3272 {
3273 
3274 	return (cred->cr_prison != &prison0);
3275 }
3276 
3277 /*
3278  * Return the correct hostname (domainname, et al) for the passed credential.
3279  */
3280 void
3281 getcredhostname(struct ucred *cred, char *buf, size_t size)
3282 {
3283 	struct prison *pr;
3284 
3285 	/*
3286 	 * A NULL credential can be used to shortcut to the physical
3287 	 * system's hostname.
3288 	 */
3289 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3290 	mtx_lock(&pr->pr_mtx);
3291 	strlcpy(buf, pr->pr_hostname, size);
3292 	mtx_unlock(&pr->pr_mtx);
3293 }
3294 
3295 void
3296 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3297 {
3298 
3299 	mtx_lock(&cred->cr_prison->pr_mtx);
3300 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3301 	mtx_unlock(&cred->cr_prison->pr_mtx);
3302 }
3303 
3304 void
3305 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3306 {
3307 
3308 	mtx_lock(&cred->cr_prison->pr_mtx);
3309 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3310 	mtx_unlock(&cred->cr_prison->pr_mtx);
3311 }
3312 
3313 void
3314 getcredhostid(struct ucred *cred, unsigned long *hostid)
3315 {
3316 
3317 	mtx_lock(&cred->cr_prison->pr_mtx);
3318 	*hostid = cred->cr_prison->pr_hostid;
3319 	mtx_unlock(&cred->cr_prison->pr_mtx);
3320 }
3321 
3322 /*
3323  * Determine whether the subject represented by cred can "see"
3324  * status of a mount point.
3325  * Returns: 0 for permitted, ENOENT otherwise.
3326  * XXX: This function should be called cr_canseemount() and should be
3327  *      placed in kern_prot.c.
3328  */
3329 int
3330 prison_canseemount(struct ucred *cred, struct mount *mp)
3331 {
3332 	struct prison *pr;
3333 	struct statfs *sp;
3334 	size_t len;
3335 
3336 	pr = cred->cr_prison;
3337 	if (pr->pr_enforce_statfs == 0)
3338 		return (0);
3339 	if (pr->pr_root->v_mount == mp)
3340 		return (0);
3341 	if (pr->pr_enforce_statfs == 2)
3342 		return (ENOENT);
3343 	/*
3344 	 * If jail's chroot directory is set to "/" we should be able to see
3345 	 * all mount-points from inside a jail.
3346 	 * This is ugly check, but this is the only situation when jail's
3347 	 * directory ends with '/'.
3348 	 */
3349 	if (strcmp(pr->pr_path, "/") == 0)
3350 		return (0);
3351 	len = strlen(pr->pr_path);
3352 	sp = &mp->mnt_stat;
3353 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3354 		return (ENOENT);
3355 	/*
3356 	 * Be sure that we don't have situation where jail's root directory
3357 	 * is "/some/path" and mount point is "/some/pathpath".
3358 	 */
3359 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3360 		return (ENOENT);
3361 	return (0);
3362 }
3363 
3364 void
3365 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3366 {
3367 	char jpath[MAXPATHLEN];
3368 	struct prison *pr;
3369 	size_t len;
3370 
3371 	pr = cred->cr_prison;
3372 	if (pr->pr_enforce_statfs == 0)
3373 		return;
3374 	if (prison_canseemount(cred, mp) != 0) {
3375 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3376 		strlcpy(sp->f_mntonname, "[restricted]",
3377 		    sizeof(sp->f_mntonname));
3378 		return;
3379 	}
3380 	if (pr->pr_root->v_mount == mp) {
3381 		/*
3382 		 * Clear current buffer data, so we are sure nothing from
3383 		 * the valid path left there.
3384 		 */
3385 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3386 		*sp->f_mntonname = '/';
3387 		return;
3388 	}
3389 	/*
3390 	 * If jail's chroot directory is set to "/" we should be able to see
3391 	 * all mount-points from inside a jail.
3392 	 */
3393 	if (strcmp(pr->pr_path, "/") == 0)
3394 		return;
3395 	len = strlen(pr->pr_path);
3396 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3397 	/*
3398 	 * Clear current buffer data, so we are sure nothing from
3399 	 * the valid path left there.
3400 	 */
3401 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3402 	if (*jpath == '\0') {
3403 		/* Should never happen. */
3404 		*sp->f_mntonname = '/';
3405 	} else {
3406 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3407 	}
3408 }
3409 
3410 /*
3411  * Check with permission for a specific privilege is granted within jail.  We
3412  * have a specific list of accepted privileges; the rest are denied.
3413  */
3414 int
3415 prison_priv_check(struct ucred *cred, int priv)
3416 {
3417 
3418 	if (!jailed(cred))
3419 		return (0);
3420 
3421 #ifdef VIMAGE
3422 	/*
3423 	 * Privileges specific to prisons with a virtual network stack.
3424 	 * There might be a duplicate entry here in case the privilege
3425 	 * is only granted conditionally in the legacy jail case.
3426 	 */
3427 	switch (priv) {
3428 #ifdef notyet
3429 		/*
3430 		 * NFS-specific privileges.
3431 		 */
3432 	case PRIV_NFS_DAEMON:
3433 	case PRIV_NFS_LOCKD:
3434 #endif
3435 		/*
3436 		 * Network stack privileges.
3437 		 */
3438 	case PRIV_NET_BRIDGE:
3439 	case PRIV_NET_GRE:
3440 	case PRIV_NET_BPF:
3441 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3442 	case PRIV_NET_ROUTE:
3443 	case PRIV_NET_TAP:
3444 	case PRIV_NET_SETIFMTU:
3445 	case PRIV_NET_SETIFFLAGS:
3446 	case PRIV_NET_SETIFCAP:
3447 	case PRIV_NET_SETIFNAME	:
3448 	case PRIV_NET_SETIFMETRIC:
3449 	case PRIV_NET_SETIFPHYS:
3450 	case PRIV_NET_SETIFMAC:
3451 	case PRIV_NET_ADDMULTI:
3452 	case PRIV_NET_DELMULTI:
3453 	case PRIV_NET_HWIOCTL:
3454 	case PRIV_NET_SETLLADDR:
3455 	case PRIV_NET_ADDIFGROUP:
3456 	case PRIV_NET_DELIFGROUP:
3457 	case PRIV_NET_IFCREATE:
3458 	case PRIV_NET_IFDESTROY:
3459 	case PRIV_NET_ADDIFADDR:
3460 	case PRIV_NET_DELIFADDR:
3461 	case PRIV_NET_LAGG:
3462 	case PRIV_NET_GIF:
3463 	case PRIV_NET_SETIFVNET:
3464 
3465 		/*
3466 		 * 802.11-related privileges.
3467 		 */
3468 	case PRIV_NET80211_GETKEY:
3469 #ifdef notyet
3470 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3471 #endif
3472 
3473 #ifdef notyet
3474 		/*
3475 		 * AppleTalk privileges.
3476 		 */
3477 	case PRIV_NETATALK_RESERVEDPORT:
3478 
3479 		/*
3480 		 * ATM privileges.
3481 		 */
3482 	case PRIV_NETATM_CFG:
3483 	case PRIV_NETATM_ADD:
3484 	case PRIV_NETATM_DEL:
3485 	case PRIV_NETATM_SET:
3486 
3487 		/*
3488 		 * Bluetooth privileges.
3489 		 */
3490 	case PRIV_NETBLUETOOTH_RAW:
3491 #endif
3492 
3493 		/*
3494 		 * Netgraph and netgraph module privileges.
3495 		 */
3496 	case PRIV_NETGRAPH_CONTROL:
3497 #ifdef notyet
3498 	case PRIV_NETGRAPH_TTY:
3499 #endif
3500 
3501 		/*
3502 		 * IPv4 and IPv6 privileges.
3503 		 */
3504 	case PRIV_NETINET_IPFW:
3505 	case PRIV_NETINET_DIVERT:
3506 	case PRIV_NETINET_PF:
3507 	case PRIV_NETINET_DUMMYNET:
3508 	case PRIV_NETINET_CARP:
3509 	case PRIV_NETINET_MROUTE:
3510 	case PRIV_NETINET_RAW:
3511 	case PRIV_NETINET_ADDRCTRL6:
3512 	case PRIV_NETINET_ND6:
3513 	case PRIV_NETINET_SCOPE6:
3514 	case PRIV_NETINET_ALIFETIME6:
3515 	case PRIV_NETINET_IPSEC:
3516 	case PRIV_NETINET_BINDANY:
3517 
3518 #ifdef notyet
3519 		/*
3520 		 * IPX/SPX privileges.
3521 		 */
3522 	case PRIV_NETIPX_RESERVEDPORT:
3523 	case PRIV_NETIPX_RAW:
3524 
3525 		/*
3526 		 * NCP privileges.
3527 		 */
3528 	case PRIV_NETNCP:
3529 
3530 		/*
3531 		 * SMB privileges.
3532 		 */
3533 	case PRIV_NETSMB:
3534 #endif
3535 
3536 	/*
3537 	 * No default: or deny here.
3538 	 * In case of no permit fall through to next switch().
3539 	 */
3540 		if (cred->cr_prison->pr_flags & PR_VNET)
3541 			return (0);
3542 	}
3543 #endif /* VIMAGE */
3544 
3545 	switch (priv) {
3546 
3547 		/*
3548 		 * Allow ktrace privileges for root in jail.
3549 		 */
3550 	case PRIV_KTRACE:
3551 
3552 #if 0
3553 		/*
3554 		 * Allow jailed processes to configure audit identity and
3555 		 * submit audit records (login, etc).  In the future we may
3556 		 * want to further refine the relationship between audit and
3557 		 * jail.
3558 		 */
3559 	case PRIV_AUDIT_GETAUDIT:
3560 	case PRIV_AUDIT_SETAUDIT:
3561 	case PRIV_AUDIT_SUBMIT:
3562 #endif
3563 
3564 		/*
3565 		 * Allow jailed processes to manipulate process UNIX
3566 		 * credentials in any way they see fit.
3567 		 */
3568 	case PRIV_CRED_SETUID:
3569 	case PRIV_CRED_SETEUID:
3570 	case PRIV_CRED_SETGID:
3571 	case PRIV_CRED_SETEGID:
3572 	case PRIV_CRED_SETGROUPS:
3573 	case PRIV_CRED_SETREUID:
3574 	case PRIV_CRED_SETREGID:
3575 	case PRIV_CRED_SETRESUID:
3576 	case PRIV_CRED_SETRESGID:
3577 
3578 		/*
3579 		 * Jail implements visibility constraints already, so allow
3580 		 * jailed root to override uid/gid-based constraints.
3581 		 */
3582 	case PRIV_SEEOTHERGIDS:
3583 	case PRIV_SEEOTHERUIDS:
3584 
3585 		/*
3586 		 * Jail implements inter-process debugging limits already, so
3587 		 * allow jailed root various debugging privileges.
3588 		 */
3589 	case PRIV_DEBUG_DIFFCRED:
3590 	case PRIV_DEBUG_SUGID:
3591 	case PRIV_DEBUG_UNPRIV:
3592 
3593 		/*
3594 		 * Allow jail to set various resource limits and login
3595 		 * properties, and for now, exceed process resource limits.
3596 		 */
3597 	case PRIV_PROC_LIMIT:
3598 	case PRIV_PROC_SETLOGIN:
3599 	case PRIV_PROC_SETRLIMIT:
3600 
3601 		/*
3602 		 * System V and POSIX IPC privileges are granted in jail.
3603 		 */
3604 	case PRIV_IPC_READ:
3605 	case PRIV_IPC_WRITE:
3606 	case PRIV_IPC_ADMIN:
3607 	case PRIV_IPC_MSGSIZE:
3608 	case PRIV_MQ_ADMIN:
3609 
3610 		/*
3611 		 * Jail operations within a jail work on child jails.
3612 		 */
3613 	case PRIV_JAIL_ATTACH:
3614 	case PRIV_JAIL_SET:
3615 	case PRIV_JAIL_REMOVE:
3616 
3617 		/*
3618 		 * Jail implements its own inter-process limits, so allow
3619 		 * root processes in jail to change scheduling on other
3620 		 * processes in the same jail.  Likewise for signalling.
3621 		 */
3622 	case PRIV_SCHED_DIFFCRED:
3623 	case PRIV_SCHED_CPUSET:
3624 	case PRIV_SIGNAL_DIFFCRED:
3625 	case PRIV_SIGNAL_SUGID:
3626 
3627 		/*
3628 		 * Allow jailed processes to write to sysctls marked as jail
3629 		 * writable.
3630 		 */
3631 	case PRIV_SYSCTL_WRITEJAIL:
3632 
3633 		/*
3634 		 * Allow root in jail to manage a variety of quota
3635 		 * properties.  These should likely be conditional on a
3636 		 * configuration option.
3637 		 */
3638 	case PRIV_VFS_GETQUOTA:
3639 	case PRIV_VFS_SETQUOTA:
3640 
3641 		/*
3642 		 * Since Jail relies on chroot() to implement file system
3643 		 * protections, grant many VFS privileges to root in jail.
3644 		 * Be careful to exclude mount-related and NFS-related
3645 		 * privileges.
3646 		 */
3647 	case PRIV_VFS_READ:
3648 	case PRIV_VFS_WRITE:
3649 	case PRIV_VFS_ADMIN:
3650 	case PRIV_VFS_EXEC:
3651 	case PRIV_VFS_LOOKUP:
3652 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3653 	case PRIV_VFS_CHFLAGS_DEV:
3654 	case PRIV_VFS_CHOWN:
3655 	case PRIV_VFS_CHROOT:
3656 	case PRIV_VFS_RETAINSUGID:
3657 	case PRIV_VFS_FCHROOT:
3658 	case PRIV_VFS_LINK:
3659 	case PRIV_VFS_SETGID:
3660 	case PRIV_VFS_STAT:
3661 	case PRIV_VFS_STICKYFILE:
3662 		return (0);
3663 
3664 		/*
3665 		 * Depending on the global setting, allow privilege of
3666 		 * setting system flags.
3667 		 */
3668 	case PRIV_VFS_SYSFLAGS:
3669 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3670 			return (0);
3671 		else
3672 			return (EPERM);
3673 
3674 		/*
3675 		 * Depending on the global setting, allow privilege of
3676 		 * mounting/unmounting file systems.
3677 		 */
3678 	case PRIV_VFS_MOUNT:
3679 	case PRIV_VFS_UNMOUNT:
3680 	case PRIV_VFS_MOUNT_NONUSER:
3681 	case PRIV_VFS_MOUNT_OWNER:
3682 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT)
3683 			return (0);
3684 		else
3685 			return (EPERM);
3686 
3687 		/*
3688 		 * Allow jailed root to bind reserved ports and reuse in-use
3689 		 * ports.
3690 		 */
3691 	case PRIV_NETINET_RESERVEDPORT:
3692 	case PRIV_NETINET_REUSEPORT:
3693 		return (0);
3694 
3695 		/*
3696 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3697 		 */
3698 	case PRIV_NETINET_SETHDROPTS:
3699 		return (0);
3700 
3701 		/*
3702 		 * Conditionally allow creating raw sockets in jail.
3703 		 */
3704 	case PRIV_NETINET_RAW:
3705 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3706 			return (0);
3707 		else
3708 			return (EPERM);
3709 
3710 		/*
3711 		 * Since jail implements its own visibility limits on netstat
3712 		 * sysctls, allow getcred.  This allows identd to work in
3713 		 * jail.
3714 		 */
3715 	case PRIV_NETINET_GETCRED:
3716 		return (0);
3717 
3718 	default:
3719 		/*
3720 		 * In all remaining cases, deny the privilege request.  This
3721 		 * includes almost all network privileges, many system
3722 		 * configuration privileges.
3723 		 */
3724 		return (EPERM);
3725 	}
3726 }
3727 
3728 /*
3729  * Return the part of pr2's name that is relative to pr1, or the whole name
3730  * if it does not directly follow.
3731  */
3732 
3733 char *
3734 prison_name(struct prison *pr1, struct prison *pr2)
3735 {
3736 	char *name;
3737 
3738 	/* Jails see themselves as "0" (if they see themselves at all). */
3739 	if (pr1 == pr2)
3740 		return "0";
3741 	name = pr2->pr_name;
3742 	if (prison_ischild(pr1, pr2)) {
3743 		/*
3744 		 * pr1 isn't locked (and allprison_lock may not be either)
3745 		 * so its length can't be counted on.  But the number of dots
3746 		 * can be counted on - and counted.
3747 		 */
3748 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3749 			name = strchr(name, '.') + 1;
3750 	}
3751 	return (name);
3752 }
3753 
3754 /*
3755  * Return the part of pr2's path that is relative to pr1, or the whole path
3756  * if it does not directly follow.
3757  */
3758 static char *
3759 prison_path(struct prison *pr1, struct prison *pr2)
3760 {
3761 	char *path1, *path2;
3762 	int len1;
3763 
3764 	path1 = pr1->pr_path;
3765 	path2 = pr2->pr_path;
3766 	if (!strcmp(path1, "/"))
3767 		return (path2);
3768 	len1 = strlen(path1);
3769 	if (strncmp(path1, path2, len1))
3770 		return (path2);
3771 	if (path2[len1] == '\0')
3772 		return "/";
3773 	if (path2[len1] == '/')
3774 		return (path2 + len1);
3775 	return (path2);
3776 }
3777 
3778 
3779 /*
3780  * Jail-related sysctls.
3781  */
3782 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3783     "Jails");
3784 
3785 static int
3786 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3787 {
3788 	struct xprison *xp;
3789 	struct prison *pr, *cpr;
3790 #ifdef INET
3791 	struct in_addr *ip4 = NULL;
3792 	int ip4s = 0;
3793 #endif
3794 #ifdef INET6
3795 	struct in_addr *ip6 = NULL;
3796 	int ip6s = 0;
3797 #endif
3798 	int descend, error;
3799 
3800 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3801 	pr = req->td->td_ucred->cr_prison;
3802 	error = 0;
3803 	sx_slock(&allprison_lock);
3804 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3805 #if defined(INET) || defined(INET6)
3806  again:
3807 #endif
3808 		mtx_lock(&cpr->pr_mtx);
3809 #ifdef INET
3810 		if (cpr->pr_ip4s > 0) {
3811 			if (ip4s < cpr->pr_ip4s) {
3812 				ip4s = cpr->pr_ip4s;
3813 				mtx_unlock(&cpr->pr_mtx);
3814 				ip4 = realloc(ip4, ip4s *
3815 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3816 				goto again;
3817 			}
3818 			bcopy(cpr->pr_ip4, ip4,
3819 			    cpr->pr_ip4s * sizeof(struct in_addr));
3820 		}
3821 #endif
3822 #ifdef INET6
3823 		if (cpr->pr_ip6s > 0) {
3824 			if (ip6s < cpr->pr_ip6s) {
3825 				ip6s = cpr->pr_ip6s;
3826 				mtx_unlock(&cpr->pr_mtx);
3827 				ip6 = realloc(ip6, ip6s *
3828 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3829 				goto again;
3830 			}
3831 			bcopy(cpr->pr_ip6, ip6,
3832 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3833 		}
3834 #endif
3835 		if (cpr->pr_ref == 0) {
3836 			mtx_unlock(&cpr->pr_mtx);
3837 			continue;
3838 		}
3839 		bzero(xp, sizeof(*xp));
3840 		xp->pr_version = XPRISON_VERSION;
3841 		xp->pr_id = cpr->pr_id;
3842 		xp->pr_state = cpr->pr_uref > 0
3843 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3844 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3845 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3846 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3847 #ifdef INET
3848 		xp->pr_ip4s = cpr->pr_ip4s;
3849 #endif
3850 #ifdef INET6
3851 		xp->pr_ip6s = cpr->pr_ip6s;
3852 #endif
3853 		mtx_unlock(&cpr->pr_mtx);
3854 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3855 		if (error)
3856 			break;
3857 #ifdef INET
3858 		if (xp->pr_ip4s > 0) {
3859 			error = SYSCTL_OUT(req, ip4,
3860 			    xp->pr_ip4s * sizeof(struct in_addr));
3861 			if (error)
3862 				break;
3863 		}
3864 #endif
3865 #ifdef INET6
3866 		if (xp->pr_ip6s > 0) {
3867 			error = SYSCTL_OUT(req, ip6,
3868 			    xp->pr_ip6s * sizeof(struct in6_addr));
3869 			if (error)
3870 				break;
3871 		}
3872 #endif
3873 	}
3874 	sx_sunlock(&allprison_lock);
3875 	free(xp, M_TEMP);
3876 #ifdef INET
3877 	free(ip4, M_TEMP);
3878 #endif
3879 #ifdef INET6
3880 	free(ip6, M_TEMP);
3881 #endif
3882 	return (error);
3883 }
3884 
3885 SYSCTL_OID(_security_jail, OID_AUTO, list,
3886     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3887     sysctl_jail_list, "S", "List of active jails");
3888 
3889 static int
3890 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3891 {
3892 	int error, injail;
3893 
3894 	injail = jailed(req->td->td_ucred);
3895 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3896 
3897 	return (error);
3898 }
3899 
3900 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3901     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3902     sysctl_jail_jailed, "I", "Process in jail?");
3903 
3904 #if defined(INET) || defined(INET6)
3905 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3906     &jail_max_af_ips, 0,
3907     "Number of IP addresses a jail may have at most per address family");
3908 #endif
3909 
3910 /*
3911  * Default parameters for jail(2) compatability.  For historical reasons,
3912  * the sysctl names have varying similarity to the parameter names.  Prisons
3913  * just see their own parameters, and can't change them.
3914  */
3915 static int
3916 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3917 {
3918 	struct prison *pr;
3919 	int allow, error, i;
3920 
3921 	pr = req->td->td_ucred->cr_prison;
3922 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3923 
3924 	/* Get the current flag value, and convert it to a boolean. */
3925 	i = (allow & arg2) ? 1 : 0;
3926 	if (arg1 != NULL)
3927 		i = !i;
3928 	error = sysctl_handle_int(oidp, &i, 0, req);
3929 	if (error || !req->newptr)
3930 		return (error);
3931 	i = i ? arg2 : 0;
3932 	if (arg1 != NULL)
3933 		i ^= arg2;
3934 	/*
3935 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3936 	 * for writing.
3937 	 */
3938 	mtx_lock(&prison0.pr_mtx);
3939 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3940 	mtx_unlock(&prison0.pr_mtx);
3941 	return (0);
3942 }
3943 
3944 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3945     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3946     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3947     "Processes in jail can set their hostnames");
3948 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3949     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3950     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3951     "Processes in jail are limited to creating UNIX/IP/route sockets only");
3952 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3953     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3954     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3955     "Processes in jail can use System V IPC primitives");
3956 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3957     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3958     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3959     "Prison root can create raw sockets");
3960 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3961     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3962     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3963     "Processes in jail can alter system file flags");
3964 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3965     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3966     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3967     "Processes in jail can mount/unmount jail-friendly file systems");
3968 
3969 static int
3970 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3971 {
3972 	struct prison *pr;
3973 	int level, error;
3974 
3975 	pr = req->td->td_ucred->cr_prison;
3976 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3977 	error = sysctl_handle_int(oidp, &level, 0, req);
3978 	if (error || !req->newptr)
3979 		return (error);
3980 	*(int *)arg1 = level;
3981 	return (0);
3982 }
3983 
3984 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3985     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3986     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3987     sysctl_jail_default_level, "I",
3988     "Processes in jail cannot see all mounted file systems");
3989 
3990 /*
3991  * Nodes to describe jail parameters.  Maximum length of string parameters
3992  * is returned in the string itself, and the other parameters exist merely
3993  * to make themselves and their types known.
3994  */
3995 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3996     "Jail parameters");
3997 
3998 int
3999 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4000 {
4001 	int i;
4002 	long l;
4003 	size_t s;
4004 	char numbuf[12];
4005 
4006 	switch (oidp->oid_kind & CTLTYPE)
4007 	{
4008 	case CTLTYPE_LONG:
4009 	case CTLTYPE_ULONG:
4010 		l = 0;
4011 #ifdef SCTL_MASK32
4012 		if (!(req->flags & SCTL_MASK32))
4013 #endif
4014 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4015 	case CTLTYPE_INT:
4016 	case CTLTYPE_UINT:
4017 		i = 0;
4018 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4019 	case CTLTYPE_STRING:
4020 		snprintf(numbuf, sizeof(numbuf), "%d", arg2);
4021 		return
4022 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4023 	case CTLTYPE_STRUCT:
4024 		s = (size_t)arg2;
4025 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4026 	}
4027 	return (0);
4028 }
4029 
4030 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4031 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4032 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4033 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4034 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4035     "I", "Jail secure level");
4036 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4037     "I", "Jail cannot see all mounted file systems");
4038 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4039     "B", "Jail persistence");
4040 #ifdef VIMAGE
4041 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4042     "B", "Virtual network stack");
4043 #endif
4044 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4045     "B", "Jail is in the process of shutting down");
4046 
4047 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4048 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4049     "I", "Current number of child jails");
4050 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4051     "I", "Maximum number of child jails");
4052 
4053 SYSCTL_JAIL_PARAM_NODE(host, "Jail host info");
4054 SYSCTL_JAIL_PARAM(, nohost, CTLTYPE_INT | CTLFLAG_RW,
4055     "BN", "Jail w/ no host info");
4056 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4057     "Jail hostname");
4058 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4059     "Jail NIS domainname");
4060 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4061     "Jail host UUID");
4062 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4063     "LU", "Jail host ID");
4064 
4065 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4066 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4067 
4068 #ifdef INET
4069 SYSCTL_JAIL_PARAM_NODE(ip4, "Jail IPv4 address virtualization");
4070 SYSCTL_JAIL_PARAM(, noip4, CTLTYPE_INT | CTLFLAG_RW,
4071     "BN", "Jail w/ no IP address virtualization");
4072 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4073     "S,in_addr,a", "Jail IPv4 addresses");
4074 #endif
4075 #ifdef INET6
4076 SYSCTL_JAIL_PARAM_NODE(ip6, "Jail IPv6 address virtualization");
4077 SYSCTL_JAIL_PARAM(, noip6, CTLTYPE_INT | CTLFLAG_RW,
4078     "BN", "Jail w/ no IP address virtualization");
4079 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4080     "S,in6_addr,a", "Jail IPv6 addresses");
4081 #endif
4082 
4083 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4084 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4085     "B", "Jail may set hostname");
4086 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4087     "B", "Jail may use SYSV IPC");
4088 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4089     "B", "Jail may create raw sockets");
4090 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4091     "B", "Jail may alter system file flags");
4092 SYSCTL_JAIL_PARAM(_allow, mount, CTLTYPE_INT | CTLFLAG_RW,
4093     "B", "Jail may mount/unmount jail-friendly file systems");
4094 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4095     "B", "Jail may set file quotas");
4096 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4097     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4098 
4099 
4100 #ifdef DDB
4101 
4102 static void
4103 db_show_prison(struct prison *pr)
4104 {
4105 	int fi;
4106 #if defined(INET) || defined(INET6)
4107 	int ii;
4108 #endif
4109 #ifdef INET6
4110 	char ip6buf[INET6_ADDRSTRLEN];
4111 #endif
4112 
4113 	db_printf("prison %p:\n", pr);
4114 	db_printf(" jid             = %d\n", pr->pr_id);
4115 	db_printf(" name            = %s\n", pr->pr_name);
4116 	db_printf(" parent          = %p\n", pr->pr_parent);
4117 	db_printf(" ref             = %d\n", pr->pr_ref);
4118 	db_printf(" uref            = %d\n", pr->pr_uref);
4119 	db_printf(" path            = %s\n", pr->pr_path);
4120 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4121 	    ? pr->pr_cpuset->cs_id : -1);
4122 #ifdef VIMAGE
4123 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4124 #endif
4125 	db_printf(" root            = %p\n", pr->pr_root);
4126 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4127 	db_printf(" childcount      = %d\n", pr->pr_childcount);
4128 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4129 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4130 	db_printf(" flags           = %x", pr->pr_flags);
4131 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4132 	    fi++)
4133 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4134 			db_printf(" %s", pr_flag_names[fi]);
4135 	db_printf(" allow           = %x", pr->pr_allow);
4136 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4137 	    fi++)
4138 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4139 			db_printf(" %s", pr_allow_names[fi]);
4140 	db_printf("\n");
4141 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4142 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4143 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4144 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4145 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4146 #ifdef INET
4147 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4148 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4149 		db_printf(" %s %s\n",
4150 		    ii == 0 ? "ip4             =" : "                 ",
4151 		    inet_ntoa(pr->pr_ip4[ii]));
4152 #endif
4153 #ifdef INET6
4154 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4155 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4156 		db_printf(" %s %s\n",
4157 		    ii == 0 ? "ip6             =" : "                 ",
4158 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4159 #endif
4160 }
4161 
4162 DB_SHOW_COMMAND(prison, db_show_prison_command)
4163 {
4164 	struct prison *pr;
4165 
4166 	if (!have_addr) {
4167 		/*
4168 		 * Show all prisons in the list, and prison0 which is not
4169 		 * listed.
4170 		 */
4171 		db_show_prison(&prison0);
4172 		if (!db_pager_quit) {
4173 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4174 				db_show_prison(pr);
4175 				if (db_pager_quit)
4176 					break;
4177 			}
4178 		}
4179 		return;
4180 	}
4181 
4182 	if (addr == 0)
4183 		pr = &prison0;
4184 	else {
4185 		/* Look for a prison with the ID and with references. */
4186 		TAILQ_FOREACH(pr, &allprison, pr_list)
4187 			if (pr->pr_id == addr && pr->pr_ref > 0)
4188 				break;
4189 		if (pr == NULL)
4190 			/* Look again, without requiring a reference. */
4191 			TAILQ_FOREACH(pr, &allprison, pr_list)
4192 				if (pr->pr_id == addr)
4193 					break;
4194 		if (pr == NULL)
4195 			/* Assume address points to a valid prison. */
4196 			pr = (struct prison *)addr;
4197 	}
4198 	db_show_prison(pr);
4199 }
4200 
4201 #endif /* DDB */
4202