xref: /freebsd/sys/kern/kern_jail.c (revision 47dd1d1b619cc035b82b49a91a25544309ff95ae)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/racct.h>
54 #include <sys/rctl.h>
55 #include <sys/refcount.h>
56 #include <sys/sx.h>
57 #include <sys/sysent.h>
58 #include <sys/namei.h>
59 #include <sys/mount.h>
60 #include <sys/queue.h>
61 #include <sys/socket.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 
66 #include <net/if.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif /* DDB */
74 
75 #include <security/mac/mac_framework.h>
76 
77 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
78 
79 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
80 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
81 
82 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
83 #ifdef INET
84 #ifdef INET6
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
86 #else
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
88 #endif
89 #else /* !INET */
90 #ifdef INET6
91 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
92 #else
93 #define	_PR_IP_SADDRSEL	0
94 #endif
95 #endif
96 
97 /* prison0 describes what is "real" about the system. */
98 struct prison prison0 = {
99 	.pr_id		= 0,
100 	.pr_name	= "0",
101 	.pr_ref		= 1,
102 	.pr_uref	= 1,
103 	.pr_path	= "/",
104 	.pr_securelevel	= -1,
105 	.pr_devfs_rsnum = 0,
106 	.pr_childmax	= JAIL_MAX,
107 	.pr_hostuuid	= DEFAULT_HOSTUUID,
108 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
109 #ifdef VIMAGE
110 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
111 #else
112 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
113 #endif
114 	.pr_allow	= PR_ALLOW_ALL_STATIC,
115 };
116 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
117 
118 struct bool_flags {
119 	const char	*name;
120 	const char	*noname;
121 	unsigned	 flag;
122 };
123 struct jailsys_flags {
124 	const char	*name;
125 	unsigned	 disable;
126 	unsigned	 new;
127 };
128 
129 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
130 struct	sx allprison_lock;
131 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
132 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
133 LIST_HEAD(, prison_racct) allprison_racct;
134 int	lastprid = 0;
135 
136 static int do_jail_attach(struct thread *td, struct prison *pr);
137 static void prison_complete(void *context, int pending);
138 static void prison_deref(struct prison *pr, int flags);
139 static char *prison_path(struct prison *pr1, struct prison *pr2);
140 static void prison_remove_one(struct prison *pr);
141 #ifdef RACCT
142 static void prison_racct_attach(struct prison *pr);
143 static void prison_racct_modify(struct prison *pr);
144 static void prison_racct_detach(struct prison *pr);
145 #endif
146 
147 /* Flags for prison_deref */
148 #define	PD_DEREF	0x01
149 #define	PD_DEUREF	0x02
150 #define	PD_LOCKED	0x04
151 #define	PD_LIST_SLOCKED	0x08
152 #define	PD_LIST_XLOCKED	0x10
153 
154 /*
155  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
156  * as we cannot figure out the size of a sparse array, or an array without a
157  * terminating entry.
158  */
159 static struct bool_flags pr_flag_bool[] = {
160 	{"persist", "nopersist", PR_PERSIST},
161 #ifdef INET
162 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
163 #endif
164 #ifdef INET6
165 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
166 #endif
167 };
168 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
169 
170 static struct jailsys_flags pr_flag_jailsys[] = {
171 	{"host", 0, PR_HOST},
172 #ifdef VIMAGE
173 	{"vnet", 0, PR_VNET},
174 #endif
175 #ifdef INET
176 	{"ip4", PR_IP4_USER, PR_IP4_USER},
177 #endif
178 #ifdef INET6
179 	{"ip6", PR_IP6_USER, PR_IP6_USER},
180 #endif
181 };
182 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
183 
184 /* Make this array full-size so dynamic parameters can be added. */
185 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
186 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
187 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
188 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
189 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
190 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
191 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
192 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
193 	{"allow.reserved_ports", "allow.noreserved_ports",
194 	 PR_ALLOW_RESERVED_PORTS},
195 };
196 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
197 
198 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | PR_ALLOW_RESERVED_PORTS)
199 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
200 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
201 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
202 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
203 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
204 #if defined(INET) || defined(INET6)
205 static unsigned jail_max_af_ips = 255;
206 #endif
207 
208 /*
209  * Initialize the parts of prison0 that can't be static-initialized with
210  * constants.  This is called from proc0_init() after creating thread0 cpuset.
211  */
212 void
213 prison0_init(void)
214 {
215 
216 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
217 	prison0.pr_osreldate = osreldate;
218 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
219 }
220 
221 /*
222  * struct jail_args {
223  *	struct jail *jail;
224  * };
225  */
226 int
227 sys_jail(struct thread *td, struct jail_args *uap)
228 {
229 	uint32_t version;
230 	int error;
231 	struct jail j;
232 
233 	error = copyin(uap->jail, &version, sizeof(uint32_t));
234 	if (error)
235 		return (error);
236 
237 	switch (version) {
238 	case 0:
239 	{
240 		struct jail_v0 j0;
241 
242 		/* FreeBSD single IPv4 jails. */
243 		bzero(&j, sizeof(struct jail));
244 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
245 		if (error)
246 			return (error);
247 		j.version = j0.version;
248 		j.path = j0.path;
249 		j.hostname = j0.hostname;
250 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
251 		break;
252 	}
253 
254 	case 1:
255 		/*
256 		 * Version 1 was used by multi-IPv4 jail implementations
257 		 * that never made it into the official kernel.
258 		 */
259 		return (EINVAL);
260 
261 	case 2:	/* JAIL_API_VERSION */
262 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
263 		error = copyin(uap->jail, &j, sizeof(struct jail));
264 		if (error)
265 			return (error);
266 		break;
267 
268 	default:
269 		/* Sci-Fi jails are not supported, sorry. */
270 		return (EINVAL);
271 	}
272 	return (kern_jail(td, &j));
273 }
274 
275 int
276 kern_jail(struct thread *td, struct jail *j)
277 {
278 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
279 #ifdef INET
280 			    + 1
281 #endif
282 #ifdef INET6
283 			    + 1
284 #endif
285 			    )];
286 	struct uio opt;
287 	char *u_path, *u_hostname, *u_name;
288 	struct bool_flags *bf;
289 #ifdef INET
290 	uint32_t ip4s;
291 	struct in_addr *u_ip4;
292 #endif
293 #ifdef INET6
294 	struct in6_addr *u_ip6;
295 #endif
296 	size_t tmplen;
297 	int error, enforce_statfs;
298 
299 	bzero(&optiov, sizeof(optiov));
300 	opt.uio_iov = optiov;
301 	opt.uio_iovcnt = 0;
302 	opt.uio_offset = -1;
303 	opt.uio_resid = -1;
304 	opt.uio_segflg = UIO_SYSSPACE;
305 	opt.uio_rw = UIO_READ;
306 	opt.uio_td = td;
307 
308 	/* Set permissions for top-level jails from sysctls. */
309 	if (!jailed(td->td_ucred)) {
310 		for (bf = pr_flag_allow;
311 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
312 			bf->flag != 0;
313 		     bf++) {
314 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
315 			    (jail_default_allow & bf->flag)
316 			    ? bf->name : bf->noname);
317 			optiov[opt.uio_iovcnt].iov_len =
318 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
319 			opt.uio_iovcnt += 2;
320 		}
321 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
322 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
323 		opt.uio_iovcnt++;
324 		enforce_statfs = jail_default_enforce_statfs;
325 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
326 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
327 		opt.uio_iovcnt++;
328 	}
329 
330 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
331 #ifdef INET
332 	ip4s = (j->version == 0) ? 1 : j->ip4s;
333 	if (ip4s > jail_max_af_ips)
334 		return (EINVAL);
335 	tmplen += ip4s * sizeof(struct in_addr);
336 #else
337 	if (j->ip4s > 0)
338 		return (EINVAL);
339 #endif
340 #ifdef INET6
341 	if (j->ip6s > jail_max_af_ips)
342 		return (EINVAL);
343 	tmplen += j->ip6s * sizeof(struct in6_addr);
344 #else
345 	if (j->ip6s > 0)
346 		return (EINVAL);
347 #endif
348 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
349 	u_hostname = u_path + MAXPATHLEN;
350 	u_name = u_hostname + MAXHOSTNAMELEN;
351 #ifdef INET
352 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
353 #endif
354 #ifdef INET6
355 #ifdef INET
356 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
357 #else
358 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
359 #endif
360 #endif
361 	optiov[opt.uio_iovcnt].iov_base = "path";
362 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
363 	opt.uio_iovcnt++;
364 	optiov[opt.uio_iovcnt].iov_base = u_path;
365 	error = copyinstr(j->path, u_path, MAXPATHLEN,
366 	    &optiov[opt.uio_iovcnt].iov_len);
367 	if (error) {
368 		free(u_path, M_TEMP);
369 		return (error);
370 	}
371 	opt.uio_iovcnt++;
372 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
373 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
374 	opt.uio_iovcnt++;
375 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
376 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
377 	    &optiov[opt.uio_iovcnt].iov_len);
378 	if (error) {
379 		free(u_path, M_TEMP);
380 		return (error);
381 	}
382 	opt.uio_iovcnt++;
383 	if (j->jailname != NULL) {
384 		optiov[opt.uio_iovcnt].iov_base = "name";
385 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
386 		opt.uio_iovcnt++;
387 		optiov[opt.uio_iovcnt].iov_base = u_name;
388 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
389 		    &optiov[opt.uio_iovcnt].iov_len);
390 		if (error) {
391 			free(u_path, M_TEMP);
392 			return (error);
393 		}
394 		opt.uio_iovcnt++;
395 	}
396 #ifdef INET
397 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
398 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
399 	opt.uio_iovcnt++;
400 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
401 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
402 	if (j->version == 0)
403 		u_ip4->s_addr = j->ip4s;
404 	else {
405 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
406 		if (error) {
407 			free(u_path, M_TEMP);
408 			return (error);
409 		}
410 	}
411 	opt.uio_iovcnt++;
412 #endif
413 #ifdef INET6
414 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
415 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
416 	opt.uio_iovcnt++;
417 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
418 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
419 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
420 	if (error) {
421 		free(u_path, M_TEMP);
422 		return (error);
423 	}
424 	opt.uio_iovcnt++;
425 #endif
426 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
427 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
428 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
429 	free(u_path, M_TEMP);
430 	return (error);
431 }
432 
433 
434 /*
435  * struct jail_set_args {
436  *	struct iovec *iovp;
437  *	unsigned int iovcnt;
438  *	int flags;
439  * };
440  */
441 int
442 sys_jail_set(struct thread *td, struct jail_set_args *uap)
443 {
444 	struct uio *auio;
445 	int error;
446 
447 	/* Check that we have an even number of iovecs. */
448 	if (uap->iovcnt & 1)
449 		return (EINVAL);
450 
451 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
452 	if (error)
453 		return (error);
454 	error = kern_jail_set(td, auio, uap->flags);
455 	free(auio, M_IOV);
456 	return (error);
457 }
458 
459 int
460 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
461 {
462 	struct nameidata nd;
463 #ifdef INET
464 	struct in_addr *ip4;
465 #endif
466 #ifdef INET6
467 	struct in6_addr *ip6;
468 #endif
469 	struct vfsopt *opt;
470 	struct vfsoptlist *opts;
471 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
472 	struct vnode *root;
473 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
474 	char *g_path, *osrelstr;
475 	struct bool_flags *bf;
476 	struct jailsys_flags *jsf;
477 #if defined(INET) || defined(INET6)
478 	struct prison *tppr;
479 	void *op;
480 #endif
481 	unsigned long hid;
482 	size_t namelen, onamelen, pnamelen;
483 	int born, created, cuflags, descend, enforce;
484 	int error, errmsg_len, errmsg_pos;
485 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
486 	int jid, jsys, len, level;
487 	int childmax, osreldt, rsnum, slevel;
488 	int fullpath_disabled;
489 #if defined(INET) || defined(INET6)
490 	int ii, ij;
491 #endif
492 #ifdef INET
493 	int ip4s, redo_ip4;
494 #endif
495 #ifdef INET6
496 	int ip6s, redo_ip6;
497 #endif
498 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
499 	unsigned tallow;
500 	char numbuf[12];
501 
502 	error = priv_check(td, PRIV_JAIL_SET);
503 	if (!error && (flags & JAIL_ATTACH))
504 		error = priv_check(td, PRIV_JAIL_ATTACH);
505 	if (error)
506 		return (error);
507 	mypr = td->td_ucred->cr_prison;
508 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
509 		return (EPERM);
510 	if (flags & ~JAIL_SET_MASK)
511 		return (EINVAL);
512 
513 	/*
514 	 * Check all the parameters before committing to anything.  Not all
515 	 * errors can be caught early, but we may as well try.  Also, this
516 	 * takes care of some expensive stuff (path lookup) before getting
517 	 * the allprison lock.
518 	 *
519 	 * XXX Jails are not filesystems, and jail parameters are not mount
520 	 *     options.  But it makes more sense to re-use the vfsopt code
521 	 *     than duplicate it under a different name.
522 	 */
523 	error = vfs_buildopts(optuio, &opts);
524 	if (error)
525 		return (error);
526 #ifdef INET
527 	ip4 = NULL;
528 #endif
529 #ifdef INET6
530 	ip6 = NULL;
531 #endif
532 	g_path = NULL;
533 
534 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
535 	if (!cuflags) {
536 		error = EINVAL;
537 		vfs_opterror(opts, "no valid operation (create or update)");
538 		goto done_errmsg;
539 	}
540 
541 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
542 	if (error == ENOENT)
543 		jid = 0;
544 	else if (error != 0)
545 		goto done_free;
546 
547 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
548 	if (error == ENOENT)
549 		gotslevel = 0;
550 	else if (error != 0)
551 		goto done_free;
552 	else
553 		gotslevel = 1;
554 
555 	error =
556 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
557 	if (error == ENOENT)
558 		gotchildmax = 0;
559 	else if (error != 0)
560 		goto done_free;
561 	else
562 		gotchildmax = 1;
563 
564 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
565 	if (error == ENOENT)
566 		gotenforce = 0;
567 	else if (error != 0)
568 		goto done_free;
569 	else if (enforce < 0 || enforce > 2) {
570 		error = EINVAL;
571 		goto done_free;
572 	} else
573 		gotenforce = 1;
574 
575 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
576 	if (error == ENOENT)
577 		gotrsnum = 0;
578 	else if (error != 0)
579 		goto done_free;
580 	else
581 		gotrsnum = 1;
582 
583 	pr_flags = ch_flags = 0;
584 	for (bf = pr_flag_bool;
585 	     bf < pr_flag_bool + nitems(pr_flag_bool);
586 	     bf++) {
587 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
588 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
589 	}
590 	ch_flags |= pr_flags;
591 	for (jsf = pr_flag_jailsys;
592 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
593 	     jsf++) {
594 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
595 		if (error == ENOENT)
596 			continue;
597 		if (error != 0)
598 			goto done_free;
599 		switch (jsys) {
600 		case JAIL_SYS_DISABLE:
601 			if (!jsf->disable) {
602 				error = EINVAL;
603 				goto done_free;
604 			}
605 			pr_flags |= jsf->disable;
606 			break;
607 		case JAIL_SYS_NEW:
608 			pr_flags |= jsf->new;
609 			break;
610 		case JAIL_SYS_INHERIT:
611 			break;
612 		default:
613 			error = EINVAL;
614 			goto done_free;
615 		}
616 		ch_flags |= jsf->new | jsf->disable;
617 	}
618 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
619 	    && !(pr_flags & PR_PERSIST)) {
620 		error = EINVAL;
621 		vfs_opterror(opts, "new jail must persist or attach");
622 		goto done_errmsg;
623 	}
624 #ifdef VIMAGE
625 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
626 		error = EINVAL;
627 		vfs_opterror(opts, "vnet cannot be changed after creation");
628 		goto done_errmsg;
629 	}
630 #endif
631 #ifdef INET
632 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
633 		error = EINVAL;
634 		vfs_opterror(opts, "ip4 cannot be changed after creation");
635 		goto done_errmsg;
636 	}
637 #endif
638 #ifdef INET6
639 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
640 		error = EINVAL;
641 		vfs_opterror(opts, "ip6 cannot be changed after creation");
642 		goto done_errmsg;
643 	}
644 #endif
645 
646 	pr_allow = ch_allow = 0;
647 	for (bf = pr_flag_allow;
648 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
649 	     bf++) {
650 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
651 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
652 	}
653 	ch_allow |= pr_allow;
654 
655 	error = vfs_getopt(opts, "name", (void **)&name, &len);
656 	if (error == ENOENT)
657 		name = NULL;
658 	else if (error != 0)
659 		goto done_free;
660 	else {
661 		if (len == 0 || name[len - 1] != '\0') {
662 			error = EINVAL;
663 			goto done_free;
664 		}
665 		if (len > MAXHOSTNAMELEN) {
666 			error = ENAMETOOLONG;
667 			goto done_free;
668 		}
669 	}
670 
671 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
672 	if (error == ENOENT)
673 		host = NULL;
674 	else if (error != 0)
675 		goto done_free;
676 	else {
677 		ch_flags |= PR_HOST;
678 		pr_flags |= PR_HOST;
679 		if (len == 0 || host[len - 1] != '\0') {
680 			error = EINVAL;
681 			goto done_free;
682 		}
683 		if (len > MAXHOSTNAMELEN) {
684 			error = ENAMETOOLONG;
685 			goto done_free;
686 		}
687 	}
688 
689 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
690 	if (error == ENOENT)
691 		domain = NULL;
692 	else if (error != 0)
693 		goto done_free;
694 	else {
695 		ch_flags |= PR_HOST;
696 		pr_flags |= PR_HOST;
697 		if (len == 0 || domain[len - 1] != '\0') {
698 			error = EINVAL;
699 			goto done_free;
700 		}
701 		if (len > MAXHOSTNAMELEN) {
702 			error = ENAMETOOLONG;
703 			goto done_free;
704 		}
705 	}
706 
707 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
708 	if (error == ENOENT)
709 		uuid = NULL;
710 	else if (error != 0)
711 		goto done_free;
712 	else {
713 		ch_flags |= PR_HOST;
714 		pr_flags |= PR_HOST;
715 		if (len == 0 || uuid[len - 1] != '\0') {
716 			error = EINVAL;
717 			goto done_free;
718 		}
719 		if (len > HOSTUUIDLEN) {
720 			error = ENAMETOOLONG;
721 			goto done_free;
722 		}
723 	}
724 
725 #ifdef COMPAT_FREEBSD32
726 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
727 		uint32_t hid32;
728 
729 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
730 		hid = hid32;
731 	} else
732 #endif
733 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
734 	if (error == ENOENT)
735 		gothid = 0;
736 	else if (error != 0)
737 		goto done_free;
738 	else {
739 		gothid = 1;
740 		ch_flags |= PR_HOST;
741 		pr_flags |= PR_HOST;
742 	}
743 
744 #ifdef INET
745 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
746 	if (error == ENOENT)
747 		ip4s = 0;
748 	else if (error != 0)
749 		goto done_free;
750 	else if (ip4s & (sizeof(*ip4) - 1)) {
751 		error = EINVAL;
752 		goto done_free;
753 	} else {
754 		ch_flags |= PR_IP4_USER;
755 		pr_flags |= PR_IP4_USER;
756 		if (ip4s > 0) {
757 			ip4s /= sizeof(*ip4);
758 			if (ip4s > jail_max_af_ips) {
759 				error = EINVAL;
760 				vfs_opterror(opts, "too many IPv4 addresses");
761 				goto done_errmsg;
762 			}
763 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
764 			bcopy(op, ip4, ip4s * sizeof(*ip4));
765 			/*
766 			 * IP addresses are all sorted but ip[0] to preserve
767 			 * the primary IP address as given from userland.
768 			 * This special IP is used for unbound outgoing
769 			 * connections as well for "loopback" traffic in case
770 			 * source address selection cannot find any more fitting
771 			 * address to connect from.
772 			 */
773 			if (ip4s > 1)
774 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
775 				    prison_qcmp_v4);
776 			/*
777 			 * Check for duplicate addresses and do some simple
778 			 * zero and broadcast checks. If users give other bogus
779 			 * addresses it is their problem.
780 			 *
781 			 * We do not have to care about byte order for these
782 			 * checks so we will do them in NBO.
783 			 */
784 			for (ii = 0; ii < ip4s; ii++) {
785 				if (ip4[ii].s_addr == INADDR_ANY ||
786 				    ip4[ii].s_addr == INADDR_BROADCAST) {
787 					error = EINVAL;
788 					goto done_free;
789 				}
790 				if ((ii+1) < ip4s &&
791 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
792 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
793 					error = EINVAL;
794 					goto done_free;
795 				}
796 			}
797 		}
798 	}
799 #endif
800 
801 #ifdef INET6
802 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
803 	if (error == ENOENT)
804 		ip6s = 0;
805 	else if (error != 0)
806 		goto done_free;
807 	else if (ip6s & (sizeof(*ip6) - 1)) {
808 		error = EINVAL;
809 		goto done_free;
810 	} else {
811 		ch_flags |= PR_IP6_USER;
812 		pr_flags |= PR_IP6_USER;
813 		if (ip6s > 0) {
814 			ip6s /= sizeof(*ip6);
815 			if (ip6s > jail_max_af_ips) {
816 				error = EINVAL;
817 				vfs_opterror(opts, "too many IPv6 addresses");
818 				goto done_errmsg;
819 			}
820 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
821 			bcopy(op, ip6, ip6s * sizeof(*ip6));
822 			if (ip6s > 1)
823 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
824 				    prison_qcmp_v6);
825 			for (ii = 0; ii < ip6s; ii++) {
826 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
827 					error = EINVAL;
828 					goto done_free;
829 				}
830 				if ((ii+1) < ip6s &&
831 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
832 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
833 				{
834 					error = EINVAL;
835 					goto done_free;
836 				}
837 			}
838 		}
839 	}
840 #endif
841 
842 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
843 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
844 		error = EINVAL;
845 		vfs_opterror(opts,
846 		    "vnet jails cannot have IP address restrictions");
847 		goto done_errmsg;
848 	}
849 #endif
850 
851 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
852 	if (error == ENOENT)
853 		osrelstr = NULL;
854 	else if (error != 0)
855 		goto done_free;
856 	else {
857 		if (flags & JAIL_UPDATE) {
858 			error = EINVAL;
859 			vfs_opterror(opts,
860 			    "osrelease cannot be changed after creation");
861 			goto done_errmsg;
862 		}
863 		if (len == 0 || len >= OSRELEASELEN) {
864 			error = EINVAL;
865 			vfs_opterror(opts,
866 			    "osrelease string must be 1-%d bytes long",
867 			    OSRELEASELEN - 1);
868 			goto done_errmsg;
869 		}
870 	}
871 
872 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
873 	if (error == ENOENT)
874 		osreldt = 0;
875 	else if (error != 0)
876 		goto done_free;
877 	else {
878 		if (flags & JAIL_UPDATE) {
879 			error = EINVAL;
880 			vfs_opterror(opts,
881 			    "osreldate cannot be changed after creation");
882 			goto done_errmsg;
883 		}
884 		if (osreldt == 0) {
885 			error = EINVAL;
886 			vfs_opterror(opts, "osreldate cannot be 0");
887 			goto done_errmsg;
888 		}
889 	}
890 
891 	fullpath_disabled = 0;
892 	root = NULL;
893 	error = vfs_getopt(opts, "path", (void **)&path, &len);
894 	if (error == ENOENT)
895 		path = NULL;
896 	else if (error != 0)
897 		goto done_free;
898 	else {
899 		if (flags & JAIL_UPDATE) {
900 			error = EINVAL;
901 			vfs_opterror(opts,
902 			    "path cannot be changed after creation");
903 			goto done_errmsg;
904 		}
905 		if (len == 0 || path[len - 1] != '\0') {
906 			error = EINVAL;
907 			goto done_free;
908 		}
909 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
910 		    path, td);
911 		error = namei(&nd);
912 		if (error)
913 			goto done_free;
914 		root = nd.ni_vp;
915 		NDFREE(&nd, NDF_ONLY_PNBUF);
916 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
917 		strlcpy(g_path, path, MAXPATHLEN);
918 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
919 		if (error == 0)
920 			path = g_path;
921 		else if (error == ENODEV) {
922 			/* proceed if sysctl debug.disablefullpath == 1 */
923 			fullpath_disabled = 1;
924 			if (len < 2 || (len == 2 && path[0] == '/'))
925 				path = NULL;
926 		} else {
927 			/* exit on other errors */
928 			goto done_free;
929 		}
930 		if (root->v_type != VDIR) {
931 			error = ENOTDIR;
932 			vput(root);
933 			goto done_free;
934 		}
935 		VOP_UNLOCK(root, 0);
936 		if (fullpath_disabled) {
937 			/* Leave room for a real-root full pathname. */
938 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
939 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
940 				error = ENAMETOOLONG;
941 				vrele(root);
942 				goto done_free;
943 			}
944 		}
945 	}
946 
947 	/*
948 	 * Find the specified jail, or at least its parent.
949 	 * This abuses the file error codes ENOENT and EEXIST.
950 	 */
951 	pr = NULL;
952 	ppr = mypr;
953 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
954 		namelc = strrchr(name, '.');
955 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
956 		if (*p != '\0')
957 			jid = 0;
958 	}
959 	sx_xlock(&allprison_lock);
960 	if (jid != 0) {
961 		/*
962 		 * See if a requested jid already exists.  There is an
963 		 * information leak here if the jid exists but is not within
964 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
965 		 * even though they cannot see the jail, and CREATE | UPDATE
966 		 * will return ENOENT which is not normally a valid error.
967 		 */
968 		if (jid < 0) {
969 			error = EINVAL;
970 			vfs_opterror(opts, "negative jid");
971 			goto done_unlock_list;
972 		}
973 		pr = prison_find(jid);
974 		if (pr != NULL) {
975 			ppr = pr->pr_parent;
976 			/* Create: jid must not exist. */
977 			if (cuflags == JAIL_CREATE) {
978 				mtx_unlock(&pr->pr_mtx);
979 				error = EEXIST;
980 				vfs_opterror(opts, "jail %d already exists",
981 				    jid);
982 				goto done_unlock_list;
983 			}
984 			if (!prison_ischild(mypr, pr)) {
985 				mtx_unlock(&pr->pr_mtx);
986 				pr = NULL;
987 			} else if (pr->pr_uref == 0) {
988 				if (!(flags & JAIL_DYING)) {
989 					mtx_unlock(&pr->pr_mtx);
990 					error = ENOENT;
991 					vfs_opterror(opts, "jail %d is dying",
992 					    jid);
993 					goto done_unlock_list;
994 				} else if ((flags & JAIL_ATTACH) ||
995 				    (pr_flags & PR_PERSIST)) {
996 					/*
997 					 * A dying jail might be resurrected
998 					 * (via attach or persist), but first
999 					 * it must determine if another jail
1000 					 * has claimed its name.  Accomplish
1001 					 * this by implicitly re-setting the
1002 					 * name.
1003 					 */
1004 					if (name == NULL)
1005 						name = prison_name(mypr, pr);
1006 				}
1007 			}
1008 		}
1009 		if (pr == NULL) {
1010 			/* Update: jid must exist. */
1011 			if (cuflags == JAIL_UPDATE) {
1012 				error = ENOENT;
1013 				vfs_opterror(opts, "jail %d not found", jid);
1014 				goto done_unlock_list;
1015 			}
1016 		}
1017 	}
1018 	/*
1019 	 * If the caller provided a name, look for a jail by that name.
1020 	 * This has different semantics for creates and updates keyed by jid
1021 	 * (where the name must not already exist in a different jail),
1022 	 * and updates keyed by the name itself (where the name must exist
1023 	 * because that is the jail being updated).
1024 	 */
1025 	namelc = NULL;
1026 	if (name != NULL) {
1027 		namelc = strrchr(name, '.');
1028 		if (namelc == NULL)
1029 			namelc = name;
1030 		else {
1031 			/*
1032 			 * This is a hierarchical name.  Split it into the
1033 			 * parent and child names, and make sure the parent
1034 			 * exists or matches an already found jail.
1035 			 */
1036 			if (pr != NULL) {
1037 				if (strncmp(name, ppr->pr_name, namelc - name)
1038 				    || ppr->pr_name[namelc - name] != '\0') {
1039 					mtx_unlock(&pr->pr_mtx);
1040 					error = EINVAL;
1041 					vfs_opterror(opts,
1042 					    "cannot change jail's parent");
1043 					goto done_unlock_list;
1044 				}
1045 			} else {
1046 				*namelc = '\0';
1047 				ppr = prison_find_name(mypr, name);
1048 				if (ppr == NULL) {
1049 					error = ENOENT;
1050 					vfs_opterror(opts,
1051 					    "jail \"%s\" not found", name);
1052 					goto done_unlock_list;
1053 				}
1054 				mtx_unlock(&ppr->pr_mtx);
1055 				*namelc = '.';
1056 			}
1057 			namelc++;
1058 		}
1059 		if (namelc[0] != '\0') {
1060 			pnamelen =
1061 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1062  name_again:
1063 			deadpr = NULL;
1064 			FOREACH_PRISON_CHILD(ppr, tpr) {
1065 				if (tpr != pr && tpr->pr_ref > 0 &&
1066 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1067 					if (pr == NULL &&
1068 					    cuflags != JAIL_CREATE) {
1069 						mtx_lock(&tpr->pr_mtx);
1070 						if (tpr->pr_ref > 0) {
1071 							/*
1072 							 * Use this jail
1073 							 * for updates.
1074 							 */
1075 							if (tpr->pr_uref > 0) {
1076 								pr = tpr;
1077 								break;
1078 							}
1079 							deadpr = tpr;
1080 						}
1081 						mtx_unlock(&tpr->pr_mtx);
1082 					} else if (tpr->pr_uref > 0) {
1083 						/*
1084 						 * Create, or update(jid):
1085 						 * name must not exist in an
1086 						 * active sibling jail.
1087 						 */
1088 						error = EEXIST;
1089 						if (pr != NULL)
1090 							mtx_unlock(&pr->pr_mtx);
1091 						vfs_opterror(opts,
1092 						   "jail \"%s\" already exists",
1093 						   name);
1094 						goto done_unlock_list;
1095 					}
1096 				}
1097 			}
1098 			/* If no active jail is found, use a dying one. */
1099 			if (deadpr != NULL && pr == NULL) {
1100 				if (flags & JAIL_DYING) {
1101 					mtx_lock(&deadpr->pr_mtx);
1102 					if (deadpr->pr_ref == 0) {
1103 						mtx_unlock(&deadpr->pr_mtx);
1104 						goto name_again;
1105 					}
1106 					pr = deadpr;
1107 				} else if (cuflags == JAIL_UPDATE) {
1108 					error = ENOENT;
1109 					vfs_opterror(opts,
1110 					    "jail \"%s\" is dying", name);
1111 					goto done_unlock_list;
1112 				}
1113 			}
1114 			/* Update: name must exist if no jid. */
1115 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1116 				error = ENOENT;
1117 				vfs_opterror(opts, "jail \"%s\" not found",
1118 				    name);
1119 				goto done_unlock_list;
1120 			}
1121 		}
1122 	}
1123 	/* Update: must provide a jid or name. */
1124 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1125 		error = ENOENT;
1126 		vfs_opterror(opts, "update specified no jail");
1127 		goto done_unlock_list;
1128 	}
1129 
1130 	/* If there's no prison to update, create a new one and link it in. */
1131 	if (pr == NULL) {
1132 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1133 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1134 				error = EPERM;
1135 				vfs_opterror(opts, "prison limit exceeded");
1136 				goto done_unlock_list;
1137 			}
1138 		created = 1;
1139 		mtx_lock(&ppr->pr_mtx);
1140 		if (ppr->pr_ref == 0) {
1141 			mtx_unlock(&ppr->pr_mtx);
1142 			error = ENOENT;
1143 			vfs_opterror(opts, "jail \"%s\" not found",
1144 			    prison_name(mypr, ppr));
1145 			goto done_unlock_list;
1146 		}
1147 		ppr->pr_ref++;
1148 		ppr->pr_uref++;
1149 		mtx_unlock(&ppr->pr_mtx);
1150 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1151 		if (jid == 0) {
1152 			/* Find the next free jid. */
1153 			jid = lastprid + 1;
1154  findnext:
1155 			if (jid == JAIL_MAX)
1156 				jid = 1;
1157 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1158 				if (tpr->pr_id < jid)
1159 					continue;
1160 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1161 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1162 					break;
1163 				}
1164 				if (jid == lastprid) {
1165 					error = EAGAIN;
1166 					vfs_opterror(opts,
1167 					    "no available jail IDs");
1168 					free(pr, M_PRISON);
1169 					prison_deref(ppr, PD_DEREF |
1170 					    PD_DEUREF | PD_LIST_XLOCKED);
1171 					goto done_releroot;
1172 				}
1173 				jid++;
1174 				goto findnext;
1175 			}
1176 			lastprid = jid;
1177 		} else {
1178 			/*
1179 			 * The jail already has a jid (that did not yet exist),
1180 			 * so just find where to insert it.
1181 			 */
1182 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1183 				if (tpr->pr_id >= jid) {
1184 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1185 					break;
1186 				}
1187 		}
1188 		if (tpr == NULL)
1189 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1190 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1191 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1192 			tpr->pr_childcount++;
1193 
1194 		pr->pr_parent = ppr;
1195 		pr->pr_id = jid;
1196 
1197 		/* Set some default values, and inherit some from the parent. */
1198 		if (namelc == NULL)
1199 			namelc = "";
1200 		if (path == NULL) {
1201 			path = "/";
1202 			root = mypr->pr_root;
1203 			vref(root);
1204 		}
1205 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1206 		pr->pr_flags |= PR_HOST;
1207 #if defined(INET) || defined(INET6)
1208 #ifdef VIMAGE
1209 		if (!(pr_flags & PR_VNET))
1210 #endif
1211 		{
1212 #ifdef INET
1213 			if (!(ch_flags & PR_IP4_USER))
1214 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1215 			else if (!(pr_flags & PR_IP4_USER)) {
1216 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1217 				if (ppr->pr_ip4 != NULL) {
1218 					pr->pr_ip4s = ppr->pr_ip4s;
1219 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1220 					    sizeof(struct in_addr), M_PRISON,
1221 					    M_WAITOK);
1222 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1223 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1224 				}
1225 			}
1226 #endif
1227 #ifdef INET6
1228 			if (!(ch_flags & PR_IP6_USER))
1229 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1230 			else if (!(pr_flags & PR_IP6_USER)) {
1231 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1232 				if (ppr->pr_ip6 != NULL) {
1233 					pr->pr_ip6s = ppr->pr_ip6s;
1234 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1235 					    sizeof(struct in6_addr), M_PRISON,
1236 					    M_WAITOK);
1237 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1238 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1239 				}
1240 			}
1241 #endif
1242 		}
1243 #endif
1244 		/* Source address selection is always on by default. */
1245 		pr->pr_flags |= _PR_IP_SADDRSEL;
1246 
1247 		pr->pr_securelevel = ppr->pr_securelevel;
1248 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1249 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1250 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1251 
1252 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1253 		if (osrelstr == NULL)
1254 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1255 		else
1256 		    strcpy(pr->pr_osrelease, osrelstr);
1257 
1258 		LIST_INIT(&pr->pr_children);
1259 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1260 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1261 
1262 #ifdef VIMAGE
1263 		/* Allocate a new vnet if specified. */
1264 		pr->pr_vnet = (pr_flags & PR_VNET)
1265 		    ? vnet_alloc() : ppr->pr_vnet;
1266 #endif
1267 		/*
1268 		 * Allocate a dedicated cpuset for each jail.
1269 		 * Unlike other initial settings, this may return an erorr.
1270 		 */
1271 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1272 		if (error) {
1273 			prison_deref(pr, PD_LIST_XLOCKED);
1274 			goto done_releroot;
1275 		}
1276 
1277 		mtx_lock(&pr->pr_mtx);
1278 		/*
1279 		 * New prisons do not yet have a reference, because we do not
1280 		 * want others to see the incomplete prison once the
1281 		 * allprison_lock is downgraded.
1282 		 */
1283 	} else {
1284 		created = 0;
1285 		/*
1286 		 * Grab a reference for existing prisons, to ensure they
1287 		 * continue to exist for the duration of the call.
1288 		 */
1289 		pr->pr_ref++;
1290 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1291 		if ((pr->pr_flags & PR_VNET) &&
1292 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1293 			error = EINVAL;
1294 			vfs_opterror(opts,
1295 			    "vnet jails cannot have IP address restrictions");
1296 			goto done_deref_locked;
1297 		}
1298 #endif
1299 #ifdef INET
1300 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1301 			error = EINVAL;
1302 			vfs_opterror(opts,
1303 			    "ip4 cannot be changed after creation");
1304 			goto done_deref_locked;
1305 		}
1306 #endif
1307 #ifdef INET6
1308 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1309 			error = EINVAL;
1310 			vfs_opterror(opts,
1311 			    "ip6 cannot be changed after creation");
1312 			goto done_deref_locked;
1313 		}
1314 #endif
1315 	}
1316 
1317 	/* Do final error checking before setting anything. */
1318 	if (gotslevel) {
1319 		if (slevel < ppr->pr_securelevel) {
1320 			error = EPERM;
1321 			goto done_deref_locked;
1322 		}
1323 	}
1324 	if (gotchildmax) {
1325 		if (childmax >= ppr->pr_childmax) {
1326 			error = EPERM;
1327 			goto done_deref_locked;
1328 		}
1329 	}
1330 	if (gotenforce) {
1331 		if (enforce < ppr->pr_enforce_statfs) {
1332 			error = EPERM;
1333 			goto done_deref_locked;
1334 		}
1335 	}
1336 	if (gotrsnum) {
1337 		/*
1338 		 * devfs_rsnum is a uint16_t
1339 		 */
1340 		if (rsnum < 0 || rsnum > 65535) {
1341 			error = EINVAL;
1342 			goto done_deref_locked;
1343 		}
1344 		/*
1345 		 * Nested jails always inherit parent's devfs ruleset
1346 		 */
1347 		if (jailed(td->td_ucred)) {
1348 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1349 				error = EPERM;
1350 				goto done_deref_locked;
1351 			} else
1352 				rsnum = ppr->pr_devfs_rsnum;
1353 		}
1354 	}
1355 #ifdef INET
1356 	if (ip4s > 0) {
1357 		if (ppr->pr_flags & PR_IP4) {
1358 			/*
1359 			 * Make sure the new set of IP addresses is a
1360 			 * subset of the parent's list.  Don't worry
1361 			 * about the parent being unlocked, as any
1362 			 * setting is done with allprison_lock held.
1363 			 */
1364 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1365 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1366 					break;
1367 			if (ij == ppr->pr_ip4s) {
1368 				error = EPERM;
1369 				goto done_deref_locked;
1370 			}
1371 			if (ip4s > 1) {
1372 				for (ii = ij = 1; ii < ip4s; ii++) {
1373 					if (ip4[ii].s_addr ==
1374 					    ppr->pr_ip4[0].s_addr)
1375 						continue;
1376 					for (; ij < ppr->pr_ip4s; ij++)
1377 						if (ip4[ii].s_addr ==
1378 						    ppr->pr_ip4[ij].s_addr)
1379 							break;
1380 					if (ij == ppr->pr_ip4s)
1381 						break;
1382 				}
1383 				if (ij == ppr->pr_ip4s) {
1384 					error = EPERM;
1385 					goto done_deref_locked;
1386 				}
1387 			}
1388 		}
1389 		/*
1390 		 * Check for conflicting IP addresses.  We permit them
1391 		 * if there is no more than one IP on each jail.  If
1392 		 * there is a duplicate on a jail with more than one
1393 		 * IP stop checking and return error.
1394 		 */
1395 		tppr = ppr;
1396 #ifdef VIMAGE
1397 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1398 			if (tppr->pr_flags & PR_VNET)
1399 				break;
1400 #endif
1401 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1402 			if (tpr == pr ||
1403 #ifdef VIMAGE
1404 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1405 #endif
1406 			    tpr->pr_uref == 0) {
1407 				descend = 0;
1408 				continue;
1409 			}
1410 			if (!(tpr->pr_flags & PR_IP4_USER))
1411 				continue;
1412 			descend = 0;
1413 			if (tpr->pr_ip4 == NULL ||
1414 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1415 				continue;
1416 			for (ii = 0; ii < ip4s; ii++) {
1417 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1418 				    0) {
1419 					error = EADDRINUSE;
1420 					vfs_opterror(opts,
1421 					    "IPv4 addresses clash");
1422 					goto done_deref_locked;
1423 				}
1424 			}
1425 		}
1426 	}
1427 #endif
1428 #ifdef INET6
1429 	if (ip6s > 0) {
1430 		if (ppr->pr_flags & PR_IP6) {
1431 			/*
1432 			 * Make sure the new set of IP addresses is a
1433 			 * subset of the parent's list.
1434 			 */
1435 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1436 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1437 				    &ppr->pr_ip6[ij]))
1438 					break;
1439 			if (ij == ppr->pr_ip6s) {
1440 				error = EPERM;
1441 				goto done_deref_locked;
1442 			}
1443 			if (ip6s > 1) {
1444 				for (ii = ij = 1; ii < ip6s; ii++) {
1445 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1446 					     &ppr->pr_ip6[0]))
1447 						continue;
1448 					for (; ij < ppr->pr_ip6s; ij++)
1449 						if (IN6_ARE_ADDR_EQUAL(
1450 						    &ip6[ii], &ppr->pr_ip6[ij]))
1451 							break;
1452 					if (ij == ppr->pr_ip6s)
1453 						break;
1454 				}
1455 				if (ij == ppr->pr_ip6s) {
1456 					error = EPERM;
1457 					goto done_deref_locked;
1458 				}
1459 			}
1460 		}
1461 		/* Check for conflicting IP addresses. */
1462 		tppr = ppr;
1463 #ifdef VIMAGE
1464 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1465 			if (tppr->pr_flags & PR_VNET)
1466 				break;
1467 #endif
1468 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1469 			if (tpr == pr ||
1470 #ifdef VIMAGE
1471 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1472 #endif
1473 			    tpr->pr_uref == 0) {
1474 				descend = 0;
1475 				continue;
1476 			}
1477 			if (!(tpr->pr_flags & PR_IP6_USER))
1478 				continue;
1479 			descend = 0;
1480 			if (tpr->pr_ip6 == NULL ||
1481 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1482 				continue;
1483 			for (ii = 0; ii < ip6s; ii++) {
1484 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1485 				    0) {
1486 					error = EADDRINUSE;
1487 					vfs_opterror(opts,
1488 					    "IPv6 addresses clash");
1489 					goto done_deref_locked;
1490 				}
1491 			}
1492 		}
1493 	}
1494 #endif
1495 	onamelen = namelen = 0;
1496 	if (namelc != NULL) {
1497 		/* Give a default name of the jid.  Also allow the name to be
1498 		 * explicitly the jid - but not any other number, and only in
1499 		 * normal form (no leading zero/etc).
1500 		 */
1501 		if (namelc[0] == '\0')
1502 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1503 		else if ((strtoul(namelc, &p, 10) != jid ||
1504 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1505 			error = EINVAL;
1506 			vfs_opterror(opts,
1507 			    "name cannot be numeric (unless it is the jid)");
1508 			goto done_deref_locked;
1509 		}
1510 		/*
1511 		 * Make sure the name isn't too long for the prison or its
1512 		 * children.
1513 		 */
1514 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1515 		onamelen = strlen(pr->pr_name + pnamelen);
1516 		namelen = strlen(namelc);
1517 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1518 			error = ENAMETOOLONG;
1519 			goto done_deref_locked;
1520 		}
1521 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1522 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1523 			    sizeof(pr->pr_name)) {
1524 				error = ENAMETOOLONG;
1525 				goto done_deref_locked;
1526 			}
1527 		}
1528 	}
1529 	if (pr_allow & ~ppr->pr_allow) {
1530 		error = EPERM;
1531 		goto done_deref_locked;
1532 	}
1533 
1534 	/*
1535 	 * Let modules check their parameters.  This requires unlocking and
1536 	 * then re-locking the prison, but this is still a valid state as long
1537 	 * as allprison_lock remains xlocked.
1538 	 */
1539 	mtx_unlock(&pr->pr_mtx);
1540 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1541 	if (error != 0) {
1542 		prison_deref(pr, created
1543 		    ? PD_LIST_XLOCKED
1544 		    : PD_DEREF | PD_LIST_XLOCKED);
1545 		goto done_releroot;
1546 	}
1547 	mtx_lock(&pr->pr_mtx);
1548 
1549 	/* At this point, all valid parameters should have been noted. */
1550 	TAILQ_FOREACH(opt, opts, link) {
1551 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1552 			error = EINVAL;
1553 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1554 			goto done_deref_locked;
1555 		}
1556 	}
1557 
1558 	/* Set the parameters of the prison. */
1559 #ifdef INET
1560 	redo_ip4 = 0;
1561 	if (pr_flags & PR_IP4_USER) {
1562 		pr->pr_flags |= PR_IP4;
1563 		free(pr->pr_ip4, M_PRISON);
1564 		pr->pr_ip4s = ip4s;
1565 		pr->pr_ip4 = ip4;
1566 		ip4 = NULL;
1567 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1568 #ifdef VIMAGE
1569 			if (tpr->pr_flags & PR_VNET) {
1570 				descend = 0;
1571 				continue;
1572 			}
1573 #endif
1574 			if (prison_restrict_ip4(tpr, NULL)) {
1575 				redo_ip4 = 1;
1576 				descend = 0;
1577 			}
1578 		}
1579 	}
1580 #endif
1581 #ifdef INET6
1582 	redo_ip6 = 0;
1583 	if (pr_flags & PR_IP6_USER) {
1584 		pr->pr_flags |= PR_IP6;
1585 		free(pr->pr_ip6, M_PRISON);
1586 		pr->pr_ip6s = ip6s;
1587 		pr->pr_ip6 = ip6;
1588 		ip6 = NULL;
1589 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1590 #ifdef VIMAGE
1591 			if (tpr->pr_flags & PR_VNET) {
1592 				descend = 0;
1593 				continue;
1594 			}
1595 #endif
1596 			if (prison_restrict_ip6(tpr, NULL)) {
1597 				redo_ip6 = 1;
1598 				descend = 0;
1599 			}
1600 		}
1601 	}
1602 #endif
1603 	if (gotslevel) {
1604 		pr->pr_securelevel = slevel;
1605 		/* Set all child jails to be at least this level. */
1606 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1607 			if (tpr->pr_securelevel < slevel)
1608 				tpr->pr_securelevel = slevel;
1609 	}
1610 	if (gotchildmax) {
1611 		pr->pr_childmax = childmax;
1612 		/* Set all child jails to under this limit. */
1613 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1614 			if (tpr->pr_childmax > childmax - level)
1615 				tpr->pr_childmax = childmax > level
1616 				    ? childmax - level : 0;
1617 	}
1618 	if (gotenforce) {
1619 		pr->pr_enforce_statfs = enforce;
1620 		/* Pass this restriction on to the children. */
1621 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1622 			if (tpr->pr_enforce_statfs < enforce)
1623 				tpr->pr_enforce_statfs = enforce;
1624 	}
1625 	if (gotrsnum) {
1626 		pr->pr_devfs_rsnum = rsnum;
1627 		/* Pass this restriction on to the children. */
1628 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1629 			tpr->pr_devfs_rsnum = rsnum;
1630 	}
1631 	if (namelc != NULL) {
1632 		if (ppr == &prison0)
1633 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1634 		else
1635 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1636 			    ppr->pr_name, namelc);
1637 		/* Change this component of child names. */
1638 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1639 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1640 			    strlen(tpr->pr_name + onamelen) + 1);
1641 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1642 		}
1643 	}
1644 	if (path != NULL) {
1645 		/* Try to keep a real-rooted full pathname. */
1646 		if (fullpath_disabled && path[0] == '/' &&
1647 		    strcmp(mypr->pr_path, "/"))
1648 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1649 			    mypr->pr_path, path);
1650 		else
1651 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1652 		pr->pr_root = root;
1653 	}
1654 	if (PR_HOST & ch_flags & ~pr_flags) {
1655 		if (pr->pr_flags & PR_HOST) {
1656 			/*
1657 			 * Copy the parent's host info.  As with pr_ip4 above,
1658 			 * the lack of a lock on the parent is not a problem;
1659 			 * it is always set with allprison_lock at least
1660 			 * shared, and is held exclusively here.
1661 			 */
1662 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1663 			    sizeof(pr->pr_hostname));
1664 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1665 			    sizeof(pr->pr_domainname));
1666 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1667 			    sizeof(pr->pr_hostuuid));
1668 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1669 		}
1670 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1671 		/* Set this prison, and any descendants without PR_HOST. */
1672 		if (host != NULL)
1673 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1674 		if (domain != NULL)
1675 			strlcpy(pr->pr_domainname, domain,
1676 			    sizeof(pr->pr_domainname));
1677 		if (uuid != NULL)
1678 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1679 		if (gothid)
1680 			pr->pr_hostid = hid;
1681 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1682 			if (tpr->pr_flags & PR_HOST)
1683 				descend = 0;
1684 			else {
1685 				if (host != NULL)
1686 					strlcpy(tpr->pr_hostname,
1687 					    pr->pr_hostname,
1688 					    sizeof(tpr->pr_hostname));
1689 				if (domain != NULL)
1690 					strlcpy(tpr->pr_domainname,
1691 					    pr->pr_domainname,
1692 					    sizeof(tpr->pr_domainname));
1693 				if (uuid != NULL)
1694 					strlcpy(tpr->pr_hostuuid,
1695 					    pr->pr_hostuuid,
1696 					    sizeof(tpr->pr_hostuuid));
1697 				if (gothid)
1698 					tpr->pr_hostid = hid;
1699 			}
1700 		}
1701 	}
1702 	if ((tallow = ch_allow & ~pr_allow)) {
1703 		/* Clear allow bits in all children. */
1704 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1705 			tpr->pr_allow &= ~tallow;
1706 	}
1707 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1708 	/*
1709 	 * Persistent prisons get an extra reference, and prisons losing their
1710 	 * persist flag lose that reference.  Only do this for existing prisons
1711 	 * for now, so new ones will remain unseen until after the module
1712 	 * handlers have completed.
1713 	 */
1714 	born = pr->pr_uref == 0;
1715 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1716 		if (pr_flags & PR_PERSIST) {
1717 			pr->pr_ref++;
1718 			pr->pr_uref++;
1719 		} else {
1720 			pr->pr_ref--;
1721 			pr->pr_uref--;
1722 		}
1723 	}
1724 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1725 	mtx_unlock(&pr->pr_mtx);
1726 
1727 #ifdef RACCT
1728 	if (racct_enable && created)
1729 		prison_racct_attach(pr);
1730 #endif
1731 
1732 	/* Locks may have prevented a complete restriction of child IP
1733 	 * addresses.  If so, allocate some more memory and try again.
1734 	 */
1735 #ifdef INET
1736 	while (redo_ip4) {
1737 		ip4s = pr->pr_ip4s;
1738 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1739 		mtx_lock(&pr->pr_mtx);
1740 		redo_ip4 = 0;
1741 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1742 #ifdef VIMAGE
1743 			if (tpr->pr_flags & PR_VNET) {
1744 				descend = 0;
1745 				continue;
1746 			}
1747 #endif
1748 			if (prison_restrict_ip4(tpr, ip4)) {
1749 				if (ip4 != NULL)
1750 					ip4 = NULL;
1751 				else
1752 					redo_ip4 = 1;
1753 			}
1754 		}
1755 		mtx_unlock(&pr->pr_mtx);
1756 	}
1757 #endif
1758 #ifdef INET6
1759 	while (redo_ip6) {
1760 		ip6s = pr->pr_ip6s;
1761 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1762 		mtx_lock(&pr->pr_mtx);
1763 		redo_ip6 = 0;
1764 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1765 #ifdef VIMAGE
1766 			if (tpr->pr_flags & PR_VNET) {
1767 				descend = 0;
1768 				continue;
1769 			}
1770 #endif
1771 			if (prison_restrict_ip6(tpr, ip6)) {
1772 				if (ip6 != NULL)
1773 					ip6 = NULL;
1774 				else
1775 					redo_ip6 = 1;
1776 			}
1777 		}
1778 		mtx_unlock(&pr->pr_mtx);
1779 	}
1780 #endif
1781 
1782 	/* Let the modules do their work. */
1783 	sx_downgrade(&allprison_lock);
1784 	if (born) {
1785 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1786 		if (error) {
1787 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1788 			prison_deref(pr, created
1789 			    ? PD_LIST_SLOCKED
1790 			    : PD_DEREF | PD_LIST_SLOCKED);
1791 			goto done_errmsg;
1792 		}
1793 	}
1794 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1795 	if (error) {
1796 		if (born)
1797 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1798 		prison_deref(pr, created
1799 		    ? PD_LIST_SLOCKED
1800 		    : PD_DEREF | PD_LIST_SLOCKED);
1801 		goto done_errmsg;
1802 	}
1803 
1804 	/* Attach this process to the prison if requested. */
1805 	if (flags & JAIL_ATTACH) {
1806 		mtx_lock(&pr->pr_mtx);
1807 		error = do_jail_attach(td, pr);
1808 		if (error) {
1809 			vfs_opterror(opts, "attach failed");
1810 			if (!created)
1811 				prison_deref(pr, PD_DEREF);
1812 			goto done_errmsg;
1813 		}
1814 	}
1815 
1816 #ifdef RACCT
1817 	if (racct_enable && !created) {
1818 		if (!(flags & JAIL_ATTACH))
1819 			sx_sunlock(&allprison_lock);
1820 		prison_racct_modify(pr);
1821 		if (!(flags & JAIL_ATTACH))
1822 			sx_slock(&allprison_lock);
1823 	}
1824 #endif
1825 
1826 	td->td_retval[0] = pr->pr_id;
1827 
1828 	/*
1829 	 * Now that it is all there, drop the temporary reference from existing
1830 	 * prisons.  Or add a reference to newly created persistent prisons
1831 	 * (which was not done earlier so that the prison would not be publicly
1832 	 * visible).
1833 	 */
1834 	if (!created) {
1835 		prison_deref(pr, (flags & JAIL_ATTACH)
1836 		    ? PD_DEREF
1837 		    : PD_DEREF | PD_LIST_SLOCKED);
1838 	} else {
1839 		if (pr_flags & PR_PERSIST) {
1840 			mtx_lock(&pr->pr_mtx);
1841 			pr->pr_ref++;
1842 			pr->pr_uref++;
1843 			mtx_unlock(&pr->pr_mtx);
1844 		}
1845 		if (!(flags & JAIL_ATTACH))
1846 			sx_sunlock(&allprison_lock);
1847 	}
1848 
1849 	goto done_free;
1850 
1851  done_deref_locked:
1852 	prison_deref(pr, created
1853 	    ? PD_LOCKED | PD_LIST_XLOCKED
1854 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1855 	goto done_releroot;
1856  done_unlock_list:
1857 	sx_xunlock(&allprison_lock);
1858  done_releroot:
1859 	if (root != NULL)
1860 		vrele(root);
1861  done_errmsg:
1862 	if (error) {
1863 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1864 		    &errmsg_len) == 0 && errmsg_len > 0) {
1865 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1866 			if (optuio->uio_segflg == UIO_SYSSPACE)
1867 				bcopy(errmsg,
1868 				    optuio->uio_iov[errmsg_pos].iov_base,
1869 				    errmsg_len);
1870 			else
1871 				copyout(errmsg,
1872 				    optuio->uio_iov[errmsg_pos].iov_base,
1873 				    errmsg_len);
1874 		}
1875 	}
1876  done_free:
1877 #ifdef INET
1878 	free(ip4, M_PRISON);
1879 #endif
1880 #ifdef INET6
1881 	free(ip6, M_PRISON);
1882 #endif
1883 	if (g_path != NULL)
1884 		free(g_path, M_TEMP);
1885 	vfs_freeopts(opts);
1886 	return (error);
1887 }
1888 
1889 
1890 /*
1891  * struct jail_get_args {
1892  *	struct iovec *iovp;
1893  *	unsigned int iovcnt;
1894  *	int flags;
1895  * };
1896  */
1897 int
1898 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1899 {
1900 	struct uio *auio;
1901 	int error;
1902 
1903 	/* Check that we have an even number of iovecs. */
1904 	if (uap->iovcnt & 1)
1905 		return (EINVAL);
1906 
1907 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1908 	if (error)
1909 		return (error);
1910 	error = kern_jail_get(td, auio, uap->flags);
1911 	if (error == 0)
1912 		error = copyout(auio->uio_iov, uap->iovp,
1913 		    uap->iovcnt * sizeof (struct iovec));
1914 	free(auio, M_IOV);
1915 	return (error);
1916 }
1917 
1918 int
1919 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1920 {
1921 	struct bool_flags *bf;
1922 	struct jailsys_flags *jsf;
1923 	struct prison *pr, *mypr;
1924 	struct vfsopt *opt;
1925 	struct vfsoptlist *opts;
1926 	char *errmsg, *name;
1927 	int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1928 	unsigned f;
1929 
1930 	if (flags & ~JAIL_GET_MASK)
1931 		return (EINVAL);
1932 
1933 	/* Get the parameter list. */
1934 	error = vfs_buildopts(optuio, &opts);
1935 	if (error)
1936 		return (error);
1937 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1938 	mypr = td->td_ucred->cr_prison;
1939 
1940 	/*
1941 	 * Find the prison specified by one of: lastjid, jid, name.
1942 	 */
1943 	sx_slock(&allprison_lock);
1944 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1945 	if (error == 0) {
1946 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1947 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1948 				mtx_lock(&pr->pr_mtx);
1949 				if (pr->pr_ref > 0 &&
1950 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1951 					break;
1952 				mtx_unlock(&pr->pr_mtx);
1953 			}
1954 		}
1955 		if (pr != NULL)
1956 			goto found_prison;
1957 		error = ENOENT;
1958 		vfs_opterror(opts, "no jail after %d", jid);
1959 		goto done_unlock_list;
1960 	} else if (error != ENOENT)
1961 		goto done_unlock_list;
1962 
1963 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1964 	if (error == 0) {
1965 		if (jid != 0) {
1966 			pr = prison_find_child(mypr, jid);
1967 			if (pr != NULL) {
1968 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1969 					mtx_unlock(&pr->pr_mtx);
1970 					error = ENOENT;
1971 					vfs_opterror(opts, "jail %d is dying",
1972 					    jid);
1973 					goto done_unlock_list;
1974 				}
1975 				goto found_prison;
1976 			}
1977 			error = ENOENT;
1978 			vfs_opterror(opts, "jail %d not found", jid);
1979 			goto done_unlock_list;
1980 		}
1981 	} else if (error != ENOENT)
1982 		goto done_unlock_list;
1983 
1984 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1985 	if (error == 0) {
1986 		if (len == 0 || name[len - 1] != '\0') {
1987 			error = EINVAL;
1988 			goto done_unlock_list;
1989 		}
1990 		pr = prison_find_name(mypr, name);
1991 		if (pr != NULL) {
1992 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1993 				mtx_unlock(&pr->pr_mtx);
1994 				error = ENOENT;
1995 				vfs_opterror(opts, "jail \"%s\" is dying",
1996 				    name);
1997 				goto done_unlock_list;
1998 			}
1999 			goto found_prison;
2000 		}
2001 		error = ENOENT;
2002 		vfs_opterror(opts, "jail \"%s\" not found", name);
2003 		goto done_unlock_list;
2004 	} else if (error != ENOENT)
2005 		goto done_unlock_list;
2006 
2007 	vfs_opterror(opts, "no jail specified");
2008 	error = ENOENT;
2009 	goto done_unlock_list;
2010 
2011  found_prison:
2012 	/* Get the parameters of the prison. */
2013 	pr->pr_ref++;
2014 	locked = PD_LOCKED;
2015 	td->td_retval[0] = pr->pr_id;
2016 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2017 	if (error != 0 && error != ENOENT)
2018 		goto done_deref;
2019 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2020 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2021 	if (error != 0 && error != ENOENT)
2022 		goto done_deref;
2023 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2024 	if (error != 0 && error != ENOENT)
2025 		goto done_deref;
2026 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2027 	    sizeof(pr->pr_cpuset->cs_id));
2028 	if (error != 0 && error != ENOENT)
2029 		goto done_deref;
2030 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2031 	if (error != 0 && error != ENOENT)
2032 		goto done_deref;
2033 #ifdef INET
2034 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2035 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2036 	if (error != 0 && error != ENOENT)
2037 		goto done_deref;
2038 #endif
2039 #ifdef INET6
2040 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2041 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2042 	if (error != 0 && error != ENOENT)
2043 		goto done_deref;
2044 #endif
2045 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2046 	    sizeof(pr->pr_securelevel));
2047 	if (error != 0 && error != ENOENT)
2048 		goto done_deref;
2049 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2050 	    sizeof(pr->pr_childcount));
2051 	if (error != 0 && error != ENOENT)
2052 		goto done_deref;
2053 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2054 	    sizeof(pr->pr_childmax));
2055 	if (error != 0 && error != ENOENT)
2056 		goto done_deref;
2057 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2058 	if (error != 0 && error != ENOENT)
2059 		goto done_deref;
2060 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2061 	if (error != 0 && error != ENOENT)
2062 		goto done_deref;
2063 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2064 	if (error != 0 && error != ENOENT)
2065 		goto done_deref;
2066 #ifdef COMPAT_FREEBSD32
2067 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2068 		uint32_t hid32 = pr->pr_hostid;
2069 
2070 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2071 	} else
2072 #endif
2073 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2074 	    sizeof(pr->pr_hostid));
2075 	if (error != 0 && error != ENOENT)
2076 		goto done_deref;
2077 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2078 	    sizeof(pr->pr_enforce_statfs));
2079 	if (error != 0 && error != ENOENT)
2080 		goto done_deref;
2081 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2082 	    sizeof(pr->pr_devfs_rsnum));
2083 	if (error != 0 && error != ENOENT)
2084 		goto done_deref;
2085 	for (bf = pr_flag_bool;
2086 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2087 	     bf++) {
2088 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2089 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2090 		if (error != 0 && error != ENOENT)
2091 			goto done_deref;
2092 		i = !i;
2093 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2094 		if (error != 0 && error != ENOENT)
2095 			goto done_deref;
2096 	}
2097 	for (jsf = pr_flag_jailsys;
2098 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2099 	     jsf++) {
2100 		f = pr->pr_flags & (jsf->disable | jsf->new);
2101 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2102 		    : (f == jsf->new) ? JAIL_SYS_NEW
2103 		    : JAIL_SYS_INHERIT;
2104 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2105 		if (error != 0 && error != ENOENT)
2106 			goto done_deref;
2107 	}
2108 	for (bf = pr_flag_allow;
2109 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2110 	     bf++) {
2111 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2112 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2113 		if (error != 0 && error != ENOENT)
2114 			goto done_deref;
2115 		i = !i;
2116 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2117 		if (error != 0 && error != ENOENT)
2118 			goto done_deref;
2119 	}
2120 	i = (pr->pr_uref == 0);
2121 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 	i = !i;
2125 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2126 	if (error != 0 && error != ENOENT)
2127 		goto done_deref;
2128 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2129 	    sizeof(pr->pr_osreldate));
2130 	if (error != 0 && error != ENOENT)
2131 		goto done_deref;
2132 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2133 	if (error != 0 && error != ENOENT)
2134 		goto done_deref;
2135 
2136 	/* Get the module parameters. */
2137 	mtx_unlock(&pr->pr_mtx);
2138 	locked = 0;
2139 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2140 	if (error)
2141 		goto done_deref;
2142 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2143 
2144 	/* By now, all parameters should have been noted. */
2145 	TAILQ_FOREACH(opt, opts, link) {
2146 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2147 			error = EINVAL;
2148 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2149 			goto done_errmsg;
2150 		}
2151 	}
2152 
2153 	/* Write the fetched parameters back to userspace. */
2154 	error = 0;
2155 	TAILQ_FOREACH(opt, opts, link) {
2156 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2157 			pos = 2 * opt->pos + 1;
2158 			optuio->uio_iov[pos].iov_len = opt->len;
2159 			if (opt->value != NULL) {
2160 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2161 					bcopy(opt->value,
2162 					    optuio->uio_iov[pos].iov_base,
2163 					    opt->len);
2164 				} else {
2165 					error = copyout(opt->value,
2166 					    optuio->uio_iov[pos].iov_base,
2167 					    opt->len);
2168 					if (error)
2169 						break;
2170 				}
2171 			}
2172 		}
2173 	}
2174 	goto done_errmsg;
2175 
2176  done_deref:
2177 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2178 	goto done_errmsg;
2179 
2180  done_unlock_list:
2181 	sx_sunlock(&allprison_lock);
2182  done_errmsg:
2183 	if (error && errmsg_pos >= 0) {
2184 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2185 		errmsg_pos = 2 * errmsg_pos + 1;
2186 		if (errmsg_len > 0) {
2187 			if (optuio->uio_segflg == UIO_SYSSPACE)
2188 				bcopy(errmsg,
2189 				    optuio->uio_iov[errmsg_pos].iov_base,
2190 				    errmsg_len);
2191 			else
2192 				copyout(errmsg,
2193 				    optuio->uio_iov[errmsg_pos].iov_base,
2194 				    errmsg_len);
2195 		}
2196 	}
2197 	vfs_freeopts(opts);
2198 	return (error);
2199 }
2200 
2201 
2202 /*
2203  * struct jail_remove_args {
2204  *	int jid;
2205  * };
2206  */
2207 int
2208 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2209 {
2210 	struct prison *pr, *cpr, *lpr, *tpr;
2211 	int descend, error;
2212 
2213 	error = priv_check(td, PRIV_JAIL_REMOVE);
2214 	if (error)
2215 		return (error);
2216 
2217 	sx_xlock(&allprison_lock);
2218 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2219 	if (pr == NULL) {
2220 		sx_xunlock(&allprison_lock);
2221 		return (EINVAL);
2222 	}
2223 
2224 	/* Remove all descendants of this prison, then remove this prison. */
2225 	pr->pr_ref++;
2226 	if (!LIST_EMPTY(&pr->pr_children)) {
2227 		mtx_unlock(&pr->pr_mtx);
2228 		lpr = NULL;
2229 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2230 			mtx_lock(&cpr->pr_mtx);
2231 			if (cpr->pr_ref > 0) {
2232 				tpr = cpr;
2233 				cpr->pr_ref++;
2234 			} else {
2235 				/* Already removed - do not do it again. */
2236 				tpr = NULL;
2237 			}
2238 			mtx_unlock(&cpr->pr_mtx);
2239 			if (lpr != NULL) {
2240 				mtx_lock(&lpr->pr_mtx);
2241 				prison_remove_one(lpr);
2242 				sx_xlock(&allprison_lock);
2243 			}
2244 			lpr = tpr;
2245 		}
2246 		if (lpr != NULL) {
2247 			mtx_lock(&lpr->pr_mtx);
2248 			prison_remove_one(lpr);
2249 			sx_xlock(&allprison_lock);
2250 		}
2251 		mtx_lock(&pr->pr_mtx);
2252 	}
2253 	prison_remove_one(pr);
2254 	return (0);
2255 }
2256 
2257 static void
2258 prison_remove_one(struct prison *pr)
2259 {
2260 	struct proc *p;
2261 	int deuref;
2262 
2263 	/* If the prison was persistent, it is not anymore. */
2264 	deuref = 0;
2265 	if (pr->pr_flags & PR_PERSIST) {
2266 		pr->pr_ref--;
2267 		deuref = PD_DEUREF;
2268 		pr->pr_flags &= ~PR_PERSIST;
2269 	}
2270 
2271 	/*
2272 	 * jail_remove added a reference.  If that's the only one, remove
2273 	 * the prison now.
2274 	 */
2275 	KASSERT(pr->pr_ref > 0,
2276 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2277 	if (pr->pr_ref == 1) {
2278 		prison_deref(pr,
2279 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2280 		return;
2281 	}
2282 
2283 	mtx_unlock(&pr->pr_mtx);
2284 	sx_xunlock(&allprison_lock);
2285 	/*
2286 	 * Kill all processes unfortunate enough to be attached to this prison.
2287 	 */
2288 	sx_slock(&allproc_lock);
2289 	LIST_FOREACH(p, &allproc, p_list) {
2290 		PROC_LOCK(p);
2291 		if (p->p_state != PRS_NEW && p->p_ucred &&
2292 		    p->p_ucred->cr_prison == pr)
2293 			kern_psignal(p, SIGKILL);
2294 		PROC_UNLOCK(p);
2295 	}
2296 	sx_sunlock(&allproc_lock);
2297 	/* Remove the temporary reference added by jail_remove. */
2298 	prison_deref(pr, deuref | PD_DEREF);
2299 }
2300 
2301 
2302 /*
2303  * struct jail_attach_args {
2304  *	int jid;
2305  * };
2306  */
2307 int
2308 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2309 {
2310 	struct prison *pr;
2311 	int error;
2312 
2313 	error = priv_check(td, PRIV_JAIL_ATTACH);
2314 	if (error)
2315 		return (error);
2316 
2317 	/*
2318 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2319 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2320 	 * But then immediately downgrade it since we don't need to stop
2321 	 * readers.
2322 	 */
2323 	sx_xlock(&allprison_lock);
2324 	sx_downgrade(&allprison_lock);
2325 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2326 	if (pr == NULL) {
2327 		sx_sunlock(&allprison_lock);
2328 		return (EINVAL);
2329 	}
2330 
2331 	/*
2332 	 * Do not allow a process to attach to a prison that is not
2333 	 * considered to be "alive".
2334 	 */
2335 	if (pr->pr_uref == 0) {
2336 		mtx_unlock(&pr->pr_mtx);
2337 		sx_sunlock(&allprison_lock);
2338 		return (EINVAL);
2339 	}
2340 
2341 	return (do_jail_attach(td, pr));
2342 }
2343 
2344 static int
2345 do_jail_attach(struct thread *td, struct prison *pr)
2346 {
2347 	struct proc *p;
2348 	struct ucred *newcred, *oldcred;
2349 	int error;
2350 
2351 	/*
2352 	 * XXX: Note that there is a slight race here if two threads
2353 	 * in the same privileged process attempt to attach to two
2354 	 * different jails at the same time.  It is important for
2355 	 * user processes not to do this, or they might end up with
2356 	 * a process root from one prison, but attached to the jail
2357 	 * of another.
2358 	 */
2359 	pr->pr_ref++;
2360 	pr->pr_uref++;
2361 	mtx_unlock(&pr->pr_mtx);
2362 
2363 	/* Let modules do whatever they need to prepare for attaching. */
2364 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2365 	if (error) {
2366 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2367 		return (error);
2368 	}
2369 	sx_sunlock(&allprison_lock);
2370 
2371 	/*
2372 	 * Reparent the newly attached process to this jail.
2373 	 */
2374 	p = td->td_proc;
2375 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2376 	if (error)
2377 		goto e_revert_osd;
2378 
2379 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2380 	if ((error = change_dir(pr->pr_root, td)) != 0)
2381 		goto e_unlock;
2382 #ifdef MAC
2383 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2384 		goto e_unlock;
2385 #endif
2386 	VOP_UNLOCK(pr->pr_root, 0);
2387 	if ((error = pwd_chroot(td, pr->pr_root)))
2388 		goto e_revert_osd;
2389 
2390 	newcred = crget();
2391 	PROC_LOCK(p);
2392 	oldcred = crcopysafe(p, newcred);
2393 	newcred->cr_prison = pr;
2394 	proc_set_cred(p, newcred);
2395 	setsugid(p);
2396 #ifdef RACCT
2397 	racct_proc_ucred_changed(p, oldcred, newcred);
2398 	crhold(newcred);
2399 #endif
2400 	PROC_UNLOCK(p);
2401 #ifdef RCTL
2402 	rctl_proc_ucred_changed(p, newcred);
2403 	crfree(newcred);
2404 #endif
2405 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2406 	crfree(oldcred);
2407 	return (0);
2408 
2409  e_unlock:
2410 	VOP_UNLOCK(pr->pr_root, 0);
2411  e_revert_osd:
2412 	/* Tell modules this thread is still in its old jail after all. */
2413 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2414 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2415 	return (error);
2416 }
2417 
2418 
2419 /*
2420  * Returns a locked prison instance, or NULL on failure.
2421  */
2422 struct prison *
2423 prison_find(int prid)
2424 {
2425 	struct prison *pr;
2426 
2427 	sx_assert(&allprison_lock, SX_LOCKED);
2428 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2429 		if (pr->pr_id == prid) {
2430 			mtx_lock(&pr->pr_mtx);
2431 			if (pr->pr_ref > 0)
2432 				return (pr);
2433 			mtx_unlock(&pr->pr_mtx);
2434 		}
2435 	}
2436 	return (NULL);
2437 }
2438 
2439 /*
2440  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2441  */
2442 struct prison *
2443 prison_find_child(struct prison *mypr, int prid)
2444 {
2445 	struct prison *pr;
2446 	int descend;
2447 
2448 	sx_assert(&allprison_lock, SX_LOCKED);
2449 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2450 		if (pr->pr_id == prid) {
2451 			mtx_lock(&pr->pr_mtx);
2452 			if (pr->pr_ref > 0)
2453 				return (pr);
2454 			mtx_unlock(&pr->pr_mtx);
2455 		}
2456 	}
2457 	return (NULL);
2458 }
2459 
2460 /*
2461  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2462  */
2463 struct prison *
2464 prison_find_name(struct prison *mypr, const char *name)
2465 {
2466 	struct prison *pr, *deadpr;
2467 	size_t mylen;
2468 	int descend;
2469 
2470 	sx_assert(&allprison_lock, SX_LOCKED);
2471 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2472  again:
2473 	deadpr = NULL;
2474 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2475 		if (!strcmp(pr->pr_name + mylen, name)) {
2476 			mtx_lock(&pr->pr_mtx);
2477 			if (pr->pr_ref > 0) {
2478 				if (pr->pr_uref > 0)
2479 					return (pr);
2480 				deadpr = pr;
2481 			}
2482 			mtx_unlock(&pr->pr_mtx);
2483 		}
2484 	}
2485 	/* There was no valid prison - perhaps there was a dying one. */
2486 	if (deadpr != NULL) {
2487 		mtx_lock(&deadpr->pr_mtx);
2488 		if (deadpr->pr_ref == 0) {
2489 			mtx_unlock(&deadpr->pr_mtx);
2490 			goto again;
2491 		}
2492 	}
2493 	return (deadpr);
2494 }
2495 
2496 /*
2497  * See if a prison has the specific flag set.
2498  */
2499 int
2500 prison_flag(struct ucred *cred, unsigned flag)
2501 {
2502 
2503 	/* This is an atomic read, so no locking is necessary. */
2504 	return (cred->cr_prison->pr_flags & flag);
2505 }
2506 
2507 int
2508 prison_allow(struct ucred *cred, unsigned flag)
2509 {
2510 
2511 	/* This is an atomic read, so no locking is necessary. */
2512 	return (cred->cr_prison->pr_allow & flag);
2513 }
2514 
2515 /*
2516  * Remove a prison reference.  If that was the last reference, remove the
2517  * prison itself - but not in this context in case there are locks held.
2518  */
2519 void
2520 prison_free_locked(struct prison *pr)
2521 {
2522 	int ref;
2523 
2524 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2525 	ref = --pr->pr_ref;
2526 	mtx_unlock(&pr->pr_mtx);
2527 	if (ref == 0)
2528 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2529 }
2530 
2531 void
2532 prison_free(struct prison *pr)
2533 {
2534 
2535 	mtx_lock(&pr->pr_mtx);
2536 	prison_free_locked(pr);
2537 }
2538 
2539 /*
2540  * Complete a call to either prison_free or prison_proc_free.
2541  */
2542 static void
2543 prison_complete(void *context, int pending)
2544 {
2545 	struct prison *pr = context;
2546 
2547 	sx_xlock(&allprison_lock);
2548 	mtx_lock(&pr->pr_mtx);
2549 	prison_deref(pr, pr->pr_uref
2550 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2551 	    : PD_LOCKED | PD_LIST_XLOCKED);
2552 }
2553 
2554 /*
2555  * Remove a prison reference (usually).  This internal version assumes no
2556  * mutexes are held, except perhaps the prison itself.  If there are no more
2557  * references, release and delist the prison.  On completion, the prison lock
2558  * and the allprison lock are both unlocked.
2559  */
2560 static void
2561 prison_deref(struct prison *pr, int flags)
2562 {
2563 	struct prison *ppr, *tpr;
2564 	int ref, lasturef;
2565 
2566 	if (!(flags & PD_LOCKED))
2567 		mtx_lock(&pr->pr_mtx);
2568 	for (;;) {
2569 		if (flags & PD_DEUREF) {
2570 			KASSERT(pr->pr_uref > 0,
2571 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2572 			     pr->pr_id));
2573 			pr->pr_uref--;
2574 			lasturef = pr->pr_uref == 0;
2575 			if (lasturef)
2576 				pr->pr_ref++;
2577 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2578 		} else
2579 			lasturef = 0;
2580 		if (flags & PD_DEREF) {
2581 			KASSERT(pr->pr_ref > 0,
2582 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2583 			     pr->pr_id));
2584 			pr->pr_ref--;
2585 		}
2586 		ref = pr->pr_ref;
2587 		mtx_unlock(&pr->pr_mtx);
2588 
2589 		/*
2590 		 * Tell the modules if the last user reference was removed
2591 		 * (even it sticks around in dying state).
2592 		 */
2593 		if (lasturef) {
2594 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2595 				sx_xlock(&allprison_lock);
2596 				flags |= PD_LIST_XLOCKED;
2597 			}
2598 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2599 			mtx_lock(&pr->pr_mtx);
2600 			ref = --pr->pr_ref;
2601 			mtx_unlock(&pr->pr_mtx);
2602 		}
2603 
2604 		/* If the prison still has references, nothing else to do. */
2605 		if (ref > 0) {
2606 			if (flags & PD_LIST_SLOCKED)
2607 				sx_sunlock(&allprison_lock);
2608 			else if (flags & PD_LIST_XLOCKED)
2609 				sx_xunlock(&allprison_lock);
2610 			return;
2611 		}
2612 
2613 		if (flags & PD_LIST_SLOCKED) {
2614 			if (!sx_try_upgrade(&allprison_lock)) {
2615 				sx_sunlock(&allprison_lock);
2616 				sx_xlock(&allprison_lock);
2617 			}
2618 		} else if (!(flags & PD_LIST_XLOCKED))
2619 			sx_xlock(&allprison_lock);
2620 
2621 		TAILQ_REMOVE(&allprison, pr, pr_list);
2622 		LIST_REMOVE(pr, pr_sibling);
2623 		ppr = pr->pr_parent;
2624 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2625 			tpr->pr_childcount--;
2626 		sx_xunlock(&allprison_lock);
2627 
2628 #ifdef VIMAGE
2629 		if (pr->pr_vnet != ppr->pr_vnet)
2630 			vnet_destroy(pr->pr_vnet);
2631 #endif
2632 		if (pr->pr_root != NULL)
2633 			vrele(pr->pr_root);
2634 		mtx_destroy(&pr->pr_mtx);
2635 #ifdef INET
2636 		free(pr->pr_ip4, M_PRISON);
2637 #endif
2638 #ifdef INET6
2639 		free(pr->pr_ip6, M_PRISON);
2640 #endif
2641 		if (pr->pr_cpuset != NULL)
2642 			cpuset_rel(pr->pr_cpuset);
2643 		osd_jail_exit(pr);
2644 #ifdef RACCT
2645 		if (racct_enable)
2646 			prison_racct_detach(pr);
2647 #endif
2648 		free(pr, M_PRISON);
2649 
2650 		/* Removing a prison frees a reference on its parent. */
2651 		pr = ppr;
2652 		mtx_lock(&pr->pr_mtx);
2653 		flags = PD_DEREF | PD_DEUREF;
2654 	}
2655 }
2656 
2657 void
2658 prison_hold_locked(struct prison *pr)
2659 {
2660 
2661 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2662 	KASSERT(pr->pr_ref > 0,
2663 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2664 	pr->pr_ref++;
2665 }
2666 
2667 void
2668 prison_hold(struct prison *pr)
2669 {
2670 
2671 	mtx_lock(&pr->pr_mtx);
2672 	prison_hold_locked(pr);
2673 	mtx_unlock(&pr->pr_mtx);
2674 }
2675 
2676 void
2677 prison_proc_hold(struct prison *pr)
2678 {
2679 
2680 	mtx_lock(&pr->pr_mtx);
2681 	KASSERT(pr->pr_uref > 0,
2682 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2683 	pr->pr_uref++;
2684 	mtx_unlock(&pr->pr_mtx);
2685 }
2686 
2687 void
2688 prison_proc_free(struct prison *pr)
2689 {
2690 
2691 	mtx_lock(&pr->pr_mtx);
2692 	KASSERT(pr->pr_uref > 0,
2693 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2694 	if (pr->pr_uref > 1)
2695 		pr->pr_uref--;
2696 	else {
2697 		/*
2698 		 * Don't remove the last user reference in this context, which
2699 		 * is expected to be a process that is not only locked, but
2700 		 * also half dead.
2701 		 */
2702 		pr->pr_ref++;
2703 		mtx_unlock(&pr->pr_mtx);
2704 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2705 		return;
2706 	}
2707 	mtx_unlock(&pr->pr_mtx);
2708 }
2709 
2710 /*
2711  * Check if a jail supports the given address family.
2712  *
2713  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2714  * if not.
2715  */
2716 int
2717 prison_check_af(struct ucred *cred, int af)
2718 {
2719 	struct prison *pr;
2720 	int error;
2721 
2722 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2723 
2724 	pr = cred->cr_prison;
2725 #ifdef VIMAGE
2726 	/* Prisons with their own network stack are not limited. */
2727 	if (prison_owns_vnet(cred))
2728 		return (0);
2729 #endif
2730 
2731 	error = 0;
2732 	switch (af)
2733 	{
2734 #ifdef INET
2735 	case AF_INET:
2736 		if (pr->pr_flags & PR_IP4)
2737 		{
2738 			mtx_lock(&pr->pr_mtx);
2739 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2740 				error = EAFNOSUPPORT;
2741 			mtx_unlock(&pr->pr_mtx);
2742 		}
2743 		break;
2744 #endif
2745 #ifdef INET6
2746 	case AF_INET6:
2747 		if (pr->pr_flags & PR_IP6)
2748 		{
2749 			mtx_lock(&pr->pr_mtx);
2750 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2751 				error = EAFNOSUPPORT;
2752 			mtx_unlock(&pr->pr_mtx);
2753 		}
2754 		break;
2755 #endif
2756 	case AF_LOCAL:
2757 	case AF_ROUTE:
2758 		break;
2759 	default:
2760 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2761 			error = EAFNOSUPPORT;
2762 	}
2763 	return (error);
2764 }
2765 
2766 /*
2767  * Check if given address belongs to the jail referenced by cred (wrapper to
2768  * prison_check_ip[46]).
2769  *
2770  * Returns 0 if jail doesn't restrict the address family or if address belongs
2771  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2772  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2773  */
2774 int
2775 prison_if(struct ucred *cred, struct sockaddr *sa)
2776 {
2777 #ifdef INET
2778 	struct sockaddr_in *sai;
2779 #endif
2780 #ifdef INET6
2781 	struct sockaddr_in6 *sai6;
2782 #endif
2783 	int error;
2784 
2785 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2786 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2787 
2788 #ifdef VIMAGE
2789 	if (prison_owns_vnet(cred))
2790 		return (0);
2791 #endif
2792 
2793 	error = 0;
2794 	switch (sa->sa_family)
2795 	{
2796 #ifdef INET
2797 	case AF_INET:
2798 		sai = (struct sockaddr_in *)sa;
2799 		error = prison_check_ip4(cred, &sai->sin_addr);
2800 		break;
2801 #endif
2802 #ifdef INET6
2803 	case AF_INET6:
2804 		sai6 = (struct sockaddr_in6 *)sa;
2805 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2806 		break;
2807 #endif
2808 	default:
2809 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2810 			error = EAFNOSUPPORT;
2811 	}
2812 	return (error);
2813 }
2814 
2815 /*
2816  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2817  */
2818 int
2819 prison_check(struct ucred *cred1, struct ucred *cred2)
2820 {
2821 
2822 	return ((cred1->cr_prison == cred2->cr_prison ||
2823 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2824 }
2825 
2826 /*
2827  * Return 1 if p2 is a child of p1, otherwise 0.
2828  */
2829 int
2830 prison_ischild(struct prison *pr1, struct prison *pr2)
2831 {
2832 
2833 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2834 		if (pr1 == pr2)
2835 			return (1);
2836 	return (0);
2837 }
2838 
2839 /*
2840  * Return 1 if the passed credential is in a jail, otherwise 0.
2841  */
2842 int
2843 jailed(struct ucred *cred)
2844 {
2845 
2846 	return (cred->cr_prison != &prison0);
2847 }
2848 
2849 /*
2850  * Return 1 if the passed credential is in a jail and that jail does not
2851  * have its own virtual network stack, otherwise 0.
2852  */
2853 int
2854 jailed_without_vnet(struct ucred *cred)
2855 {
2856 
2857 	if (!jailed(cred))
2858 		return (0);
2859 #ifdef VIMAGE
2860 	if (prison_owns_vnet(cred))
2861 		return (0);
2862 #endif
2863 
2864 	return (1);
2865 }
2866 
2867 /*
2868  * Return the correct hostname (domainname, et al) for the passed credential.
2869  */
2870 void
2871 getcredhostname(struct ucred *cred, char *buf, size_t size)
2872 {
2873 	struct prison *pr;
2874 
2875 	/*
2876 	 * A NULL credential can be used to shortcut to the physical
2877 	 * system's hostname.
2878 	 */
2879 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2880 	mtx_lock(&pr->pr_mtx);
2881 	strlcpy(buf, pr->pr_hostname, size);
2882 	mtx_unlock(&pr->pr_mtx);
2883 }
2884 
2885 void
2886 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2887 {
2888 
2889 	mtx_lock(&cred->cr_prison->pr_mtx);
2890 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2891 	mtx_unlock(&cred->cr_prison->pr_mtx);
2892 }
2893 
2894 void
2895 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2896 {
2897 
2898 	mtx_lock(&cred->cr_prison->pr_mtx);
2899 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2900 	mtx_unlock(&cred->cr_prison->pr_mtx);
2901 }
2902 
2903 void
2904 getcredhostid(struct ucred *cred, unsigned long *hostid)
2905 {
2906 
2907 	mtx_lock(&cred->cr_prison->pr_mtx);
2908 	*hostid = cred->cr_prison->pr_hostid;
2909 	mtx_unlock(&cred->cr_prison->pr_mtx);
2910 }
2911 
2912 #ifdef VIMAGE
2913 /*
2914  * Determine whether the prison represented by cred owns
2915  * its vnet rather than having it inherited.
2916  *
2917  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2918  */
2919 int
2920 prison_owns_vnet(struct ucred *cred)
2921 {
2922 
2923 	/*
2924 	 * vnets cannot be added/removed after jail creation,
2925 	 * so no need to lock here.
2926 	 */
2927 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2928 }
2929 #endif
2930 
2931 /*
2932  * Determine whether the subject represented by cred can "see"
2933  * status of a mount point.
2934  * Returns: 0 for permitted, ENOENT otherwise.
2935  * XXX: This function should be called cr_canseemount() and should be
2936  *      placed in kern_prot.c.
2937  */
2938 int
2939 prison_canseemount(struct ucred *cred, struct mount *mp)
2940 {
2941 	struct prison *pr;
2942 	struct statfs *sp;
2943 	size_t len;
2944 
2945 	pr = cred->cr_prison;
2946 	if (pr->pr_enforce_statfs == 0)
2947 		return (0);
2948 	if (pr->pr_root->v_mount == mp)
2949 		return (0);
2950 	if (pr->pr_enforce_statfs == 2)
2951 		return (ENOENT);
2952 	/*
2953 	 * If jail's chroot directory is set to "/" we should be able to see
2954 	 * all mount-points from inside a jail.
2955 	 * This is ugly check, but this is the only situation when jail's
2956 	 * directory ends with '/'.
2957 	 */
2958 	if (strcmp(pr->pr_path, "/") == 0)
2959 		return (0);
2960 	len = strlen(pr->pr_path);
2961 	sp = &mp->mnt_stat;
2962 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2963 		return (ENOENT);
2964 	/*
2965 	 * Be sure that we don't have situation where jail's root directory
2966 	 * is "/some/path" and mount point is "/some/pathpath".
2967 	 */
2968 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2969 		return (ENOENT);
2970 	return (0);
2971 }
2972 
2973 void
2974 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2975 {
2976 	char jpath[MAXPATHLEN];
2977 	struct prison *pr;
2978 	size_t len;
2979 
2980 	pr = cred->cr_prison;
2981 	if (pr->pr_enforce_statfs == 0)
2982 		return;
2983 	if (prison_canseemount(cred, mp) != 0) {
2984 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2985 		strlcpy(sp->f_mntonname, "[restricted]",
2986 		    sizeof(sp->f_mntonname));
2987 		return;
2988 	}
2989 	if (pr->pr_root->v_mount == mp) {
2990 		/*
2991 		 * Clear current buffer data, so we are sure nothing from
2992 		 * the valid path left there.
2993 		 */
2994 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2995 		*sp->f_mntonname = '/';
2996 		return;
2997 	}
2998 	/*
2999 	 * If jail's chroot directory is set to "/" we should be able to see
3000 	 * all mount-points from inside a jail.
3001 	 */
3002 	if (strcmp(pr->pr_path, "/") == 0)
3003 		return;
3004 	len = strlen(pr->pr_path);
3005 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3006 	/*
3007 	 * Clear current buffer data, so we are sure nothing from
3008 	 * the valid path left there.
3009 	 */
3010 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3011 	if (*jpath == '\0') {
3012 		/* Should never happen. */
3013 		*sp->f_mntonname = '/';
3014 	} else {
3015 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3016 	}
3017 }
3018 
3019 /*
3020  * Check with permission for a specific privilege is granted within jail.  We
3021  * have a specific list of accepted privileges; the rest are denied.
3022  */
3023 int
3024 prison_priv_check(struct ucred *cred, int priv)
3025 {
3026 
3027 	if (!jailed(cred))
3028 		return (0);
3029 
3030 #ifdef VIMAGE
3031 	/*
3032 	 * Privileges specific to prisons with a virtual network stack.
3033 	 * There might be a duplicate entry here in case the privilege
3034 	 * is only granted conditionally in the legacy jail case.
3035 	 */
3036 	switch (priv) {
3037 #ifdef notyet
3038 		/*
3039 		 * NFS-specific privileges.
3040 		 */
3041 	case PRIV_NFS_DAEMON:
3042 	case PRIV_NFS_LOCKD:
3043 #endif
3044 		/*
3045 		 * Network stack privileges.
3046 		 */
3047 	case PRIV_NET_BRIDGE:
3048 	case PRIV_NET_GRE:
3049 	case PRIV_NET_BPF:
3050 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3051 	case PRIV_NET_ROUTE:
3052 	case PRIV_NET_TAP:
3053 	case PRIV_NET_SETIFMTU:
3054 	case PRIV_NET_SETIFFLAGS:
3055 	case PRIV_NET_SETIFCAP:
3056 	case PRIV_NET_SETIFDESCR:
3057 	case PRIV_NET_SETIFNAME	:
3058 	case PRIV_NET_SETIFMETRIC:
3059 	case PRIV_NET_SETIFPHYS:
3060 	case PRIV_NET_SETIFMAC:
3061 	case PRIV_NET_ADDMULTI:
3062 	case PRIV_NET_DELMULTI:
3063 	case PRIV_NET_HWIOCTL:
3064 	case PRIV_NET_SETLLADDR:
3065 	case PRIV_NET_ADDIFGROUP:
3066 	case PRIV_NET_DELIFGROUP:
3067 	case PRIV_NET_IFCREATE:
3068 	case PRIV_NET_IFDESTROY:
3069 	case PRIV_NET_ADDIFADDR:
3070 	case PRIV_NET_DELIFADDR:
3071 	case PRIV_NET_LAGG:
3072 	case PRIV_NET_GIF:
3073 	case PRIV_NET_SETIFVNET:
3074 	case PRIV_NET_SETIFFIB:
3075 
3076 		/*
3077 		 * 802.11-related privileges.
3078 		 */
3079 	case PRIV_NET80211_GETKEY:
3080 #ifdef notyet
3081 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3082 #endif
3083 
3084 #ifdef notyet
3085 		/*
3086 		 * ATM privileges.
3087 		 */
3088 	case PRIV_NETATM_CFG:
3089 	case PRIV_NETATM_ADD:
3090 	case PRIV_NETATM_DEL:
3091 	case PRIV_NETATM_SET:
3092 
3093 		/*
3094 		 * Bluetooth privileges.
3095 		 */
3096 	case PRIV_NETBLUETOOTH_RAW:
3097 #endif
3098 
3099 		/*
3100 		 * Netgraph and netgraph module privileges.
3101 		 */
3102 	case PRIV_NETGRAPH_CONTROL:
3103 #ifdef notyet
3104 	case PRIV_NETGRAPH_TTY:
3105 #endif
3106 
3107 		/*
3108 		 * IPv4 and IPv6 privileges.
3109 		 */
3110 	case PRIV_NETINET_IPFW:
3111 	case PRIV_NETINET_DIVERT:
3112 	case PRIV_NETINET_PF:
3113 	case PRIV_NETINET_DUMMYNET:
3114 	case PRIV_NETINET_CARP:
3115 	case PRIV_NETINET_MROUTE:
3116 	case PRIV_NETINET_RAW:
3117 	case PRIV_NETINET_ADDRCTRL6:
3118 	case PRIV_NETINET_ND6:
3119 	case PRIV_NETINET_SCOPE6:
3120 	case PRIV_NETINET_ALIFETIME6:
3121 	case PRIV_NETINET_IPSEC:
3122 	case PRIV_NETINET_BINDANY:
3123 
3124 #ifdef notyet
3125 		/*
3126 		 * NCP privileges.
3127 		 */
3128 	case PRIV_NETNCP:
3129 
3130 		/*
3131 		 * SMB privileges.
3132 		 */
3133 	case PRIV_NETSMB:
3134 #endif
3135 
3136 	/*
3137 	 * No default: or deny here.
3138 	 * In case of no permit fall through to next switch().
3139 	 */
3140 		if (cred->cr_prison->pr_flags & PR_VNET)
3141 			return (0);
3142 	}
3143 #endif /* VIMAGE */
3144 
3145 	switch (priv) {
3146 
3147 		/*
3148 		 * Allow ktrace privileges for root in jail.
3149 		 */
3150 	case PRIV_KTRACE:
3151 
3152 #if 0
3153 		/*
3154 		 * Allow jailed processes to configure audit identity and
3155 		 * submit audit records (login, etc).  In the future we may
3156 		 * want to further refine the relationship between audit and
3157 		 * jail.
3158 		 */
3159 	case PRIV_AUDIT_GETAUDIT:
3160 	case PRIV_AUDIT_SETAUDIT:
3161 	case PRIV_AUDIT_SUBMIT:
3162 #endif
3163 
3164 		/*
3165 		 * Allow jailed processes to manipulate process UNIX
3166 		 * credentials in any way they see fit.
3167 		 */
3168 	case PRIV_CRED_SETUID:
3169 	case PRIV_CRED_SETEUID:
3170 	case PRIV_CRED_SETGID:
3171 	case PRIV_CRED_SETEGID:
3172 	case PRIV_CRED_SETGROUPS:
3173 	case PRIV_CRED_SETREUID:
3174 	case PRIV_CRED_SETREGID:
3175 	case PRIV_CRED_SETRESUID:
3176 	case PRIV_CRED_SETRESGID:
3177 
3178 		/*
3179 		 * Jail implements visibility constraints already, so allow
3180 		 * jailed root to override uid/gid-based constraints.
3181 		 */
3182 	case PRIV_SEEOTHERGIDS:
3183 	case PRIV_SEEOTHERUIDS:
3184 
3185 		/*
3186 		 * Jail implements inter-process debugging limits already, so
3187 		 * allow jailed root various debugging privileges.
3188 		 */
3189 	case PRIV_DEBUG_DIFFCRED:
3190 	case PRIV_DEBUG_SUGID:
3191 	case PRIV_DEBUG_UNPRIV:
3192 
3193 		/*
3194 		 * Allow jail to set various resource limits and login
3195 		 * properties, and for now, exceed process resource limits.
3196 		 */
3197 	case PRIV_PROC_LIMIT:
3198 	case PRIV_PROC_SETLOGIN:
3199 	case PRIV_PROC_SETRLIMIT:
3200 
3201 		/*
3202 		 * System V and POSIX IPC privileges are granted in jail.
3203 		 */
3204 	case PRIV_IPC_READ:
3205 	case PRIV_IPC_WRITE:
3206 	case PRIV_IPC_ADMIN:
3207 	case PRIV_IPC_MSGSIZE:
3208 	case PRIV_MQ_ADMIN:
3209 
3210 		/*
3211 		 * Jail operations within a jail work on child jails.
3212 		 */
3213 	case PRIV_JAIL_ATTACH:
3214 	case PRIV_JAIL_SET:
3215 	case PRIV_JAIL_REMOVE:
3216 
3217 		/*
3218 		 * Jail implements its own inter-process limits, so allow
3219 		 * root processes in jail to change scheduling on other
3220 		 * processes in the same jail.  Likewise for signalling.
3221 		 */
3222 	case PRIV_SCHED_DIFFCRED:
3223 	case PRIV_SCHED_CPUSET:
3224 	case PRIV_SIGNAL_DIFFCRED:
3225 	case PRIV_SIGNAL_SUGID:
3226 
3227 		/*
3228 		 * Allow jailed processes to write to sysctls marked as jail
3229 		 * writable.
3230 		 */
3231 	case PRIV_SYSCTL_WRITEJAIL:
3232 
3233 		/*
3234 		 * Allow root in jail to manage a variety of quota
3235 		 * properties.  These should likely be conditional on a
3236 		 * configuration option.
3237 		 */
3238 	case PRIV_VFS_GETQUOTA:
3239 	case PRIV_VFS_SETQUOTA:
3240 
3241 		/*
3242 		 * Since Jail relies on chroot() to implement file system
3243 		 * protections, grant many VFS privileges to root in jail.
3244 		 * Be careful to exclude mount-related and NFS-related
3245 		 * privileges.
3246 		 */
3247 	case PRIV_VFS_READ:
3248 	case PRIV_VFS_WRITE:
3249 	case PRIV_VFS_ADMIN:
3250 	case PRIV_VFS_EXEC:
3251 	case PRIV_VFS_LOOKUP:
3252 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3253 	case PRIV_VFS_CHFLAGS_DEV:
3254 	case PRIV_VFS_CHOWN:
3255 	case PRIV_VFS_CHROOT:
3256 	case PRIV_VFS_RETAINSUGID:
3257 	case PRIV_VFS_FCHROOT:
3258 	case PRIV_VFS_LINK:
3259 	case PRIV_VFS_SETGID:
3260 	case PRIV_VFS_STAT:
3261 	case PRIV_VFS_STICKYFILE:
3262 
3263 		/*
3264 		 * As in the non-jail case, non-root users are expected to be
3265 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3266 		 * exists in the jail and they have permission to access it).
3267 		 */
3268 	case PRIV_KMEM_READ:
3269 		return (0);
3270 
3271 		/*
3272 		 * Depending on the global setting, allow privilege of
3273 		 * setting system flags.
3274 		 */
3275 	case PRIV_VFS_SYSFLAGS:
3276 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3277 			return (0);
3278 		else
3279 			return (EPERM);
3280 
3281 		/*
3282 		 * Depending on the global setting, allow privilege of
3283 		 * mounting/unmounting file systems.
3284 		 */
3285 	case PRIV_VFS_MOUNT:
3286 	case PRIV_VFS_UNMOUNT:
3287 	case PRIV_VFS_MOUNT_NONUSER:
3288 	case PRIV_VFS_MOUNT_OWNER:
3289 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3290 		    cred->cr_prison->pr_enforce_statfs < 2)
3291 			return (0);
3292 		else
3293 			return (EPERM);
3294 
3295 		/*
3296 		 * Conditionally allow jailed root to bind reserved ports.
3297 		 */
3298 	case PRIV_NETINET_RESERVEDPORT:
3299 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3300 			return (0);
3301 		else
3302 			return (EPERM);
3303 
3304 		/*
3305 		 * Allow jailed root to reuse in-use ports.
3306 		 */
3307 	case PRIV_NETINET_REUSEPORT:
3308 		return (0);
3309 
3310 		/*
3311 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3312 		 */
3313 	case PRIV_NETINET_SETHDROPTS:
3314 		return (0);
3315 
3316 		/*
3317 		 * Conditionally allow creating raw sockets in jail.
3318 		 */
3319 	case PRIV_NETINET_RAW:
3320 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3321 			return (0);
3322 		else
3323 			return (EPERM);
3324 
3325 		/*
3326 		 * Since jail implements its own visibility limits on netstat
3327 		 * sysctls, allow getcred.  This allows identd to work in
3328 		 * jail.
3329 		 */
3330 	case PRIV_NETINET_GETCRED:
3331 		return (0);
3332 
3333 		/*
3334 		 * Allow jailed root to set loginclass.
3335 		 */
3336 	case PRIV_PROC_SETLOGINCLASS:
3337 		return (0);
3338 
3339 	default:
3340 		/*
3341 		 * In all remaining cases, deny the privilege request.  This
3342 		 * includes almost all network privileges, many system
3343 		 * configuration privileges.
3344 		 */
3345 		return (EPERM);
3346 	}
3347 }
3348 
3349 /*
3350  * Return the part of pr2's name that is relative to pr1, or the whole name
3351  * if it does not directly follow.
3352  */
3353 
3354 char *
3355 prison_name(struct prison *pr1, struct prison *pr2)
3356 {
3357 	char *name;
3358 
3359 	/* Jails see themselves as "0" (if they see themselves at all). */
3360 	if (pr1 == pr2)
3361 		return "0";
3362 	name = pr2->pr_name;
3363 	if (prison_ischild(pr1, pr2)) {
3364 		/*
3365 		 * pr1 isn't locked (and allprison_lock may not be either)
3366 		 * so its length can't be counted on.  But the number of dots
3367 		 * can be counted on - and counted.
3368 		 */
3369 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3370 			name = strchr(name, '.') + 1;
3371 	}
3372 	return (name);
3373 }
3374 
3375 /*
3376  * Return the part of pr2's path that is relative to pr1, or the whole path
3377  * if it does not directly follow.
3378  */
3379 static char *
3380 prison_path(struct prison *pr1, struct prison *pr2)
3381 {
3382 	char *path1, *path2;
3383 	int len1;
3384 
3385 	path1 = pr1->pr_path;
3386 	path2 = pr2->pr_path;
3387 	if (!strcmp(path1, "/"))
3388 		return (path2);
3389 	len1 = strlen(path1);
3390 	if (strncmp(path1, path2, len1))
3391 		return (path2);
3392 	if (path2[len1] == '\0')
3393 		return "/";
3394 	if (path2[len1] == '/')
3395 		return (path2 + len1);
3396 	return (path2);
3397 }
3398 
3399 
3400 /*
3401  * Jail-related sysctls.
3402  */
3403 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3404     "Jails");
3405 
3406 static int
3407 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3408 {
3409 	struct xprison *xp;
3410 	struct prison *pr, *cpr;
3411 #ifdef INET
3412 	struct in_addr *ip4 = NULL;
3413 	int ip4s = 0;
3414 #endif
3415 #ifdef INET6
3416 	struct in6_addr *ip6 = NULL;
3417 	int ip6s = 0;
3418 #endif
3419 	int descend, error;
3420 
3421 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3422 	pr = req->td->td_ucred->cr_prison;
3423 	error = 0;
3424 	sx_slock(&allprison_lock);
3425 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3426 #if defined(INET) || defined(INET6)
3427  again:
3428 #endif
3429 		mtx_lock(&cpr->pr_mtx);
3430 #ifdef INET
3431 		if (cpr->pr_ip4s > 0) {
3432 			if (ip4s < cpr->pr_ip4s) {
3433 				ip4s = cpr->pr_ip4s;
3434 				mtx_unlock(&cpr->pr_mtx);
3435 				ip4 = realloc(ip4, ip4s *
3436 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3437 				goto again;
3438 			}
3439 			bcopy(cpr->pr_ip4, ip4,
3440 			    cpr->pr_ip4s * sizeof(struct in_addr));
3441 		}
3442 #endif
3443 #ifdef INET6
3444 		if (cpr->pr_ip6s > 0) {
3445 			if (ip6s < cpr->pr_ip6s) {
3446 				ip6s = cpr->pr_ip6s;
3447 				mtx_unlock(&cpr->pr_mtx);
3448 				ip6 = realloc(ip6, ip6s *
3449 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3450 				goto again;
3451 			}
3452 			bcopy(cpr->pr_ip6, ip6,
3453 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3454 		}
3455 #endif
3456 		if (cpr->pr_ref == 0) {
3457 			mtx_unlock(&cpr->pr_mtx);
3458 			continue;
3459 		}
3460 		bzero(xp, sizeof(*xp));
3461 		xp->pr_version = XPRISON_VERSION;
3462 		xp->pr_id = cpr->pr_id;
3463 		xp->pr_state = cpr->pr_uref > 0
3464 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3465 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3466 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3467 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3468 #ifdef INET
3469 		xp->pr_ip4s = cpr->pr_ip4s;
3470 #endif
3471 #ifdef INET6
3472 		xp->pr_ip6s = cpr->pr_ip6s;
3473 #endif
3474 		mtx_unlock(&cpr->pr_mtx);
3475 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3476 		if (error)
3477 			break;
3478 #ifdef INET
3479 		if (xp->pr_ip4s > 0) {
3480 			error = SYSCTL_OUT(req, ip4,
3481 			    xp->pr_ip4s * sizeof(struct in_addr));
3482 			if (error)
3483 				break;
3484 		}
3485 #endif
3486 #ifdef INET6
3487 		if (xp->pr_ip6s > 0) {
3488 			error = SYSCTL_OUT(req, ip6,
3489 			    xp->pr_ip6s * sizeof(struct in6_addr));
3490 			if (error)
3491 				break;
3492 		}
3493 #endif
3494 	}
3495 	sx_sunlock(&allprison_lock);
3496 	free(xp, M_TEMP);
3497 #ifdef INET
3498 	free(ip4, M_TEMP);
3499 #endif
3500 #ifdef INET6
3501 	free(ip6, M_TEMP);
3502 #endif
3503 	return (error);
3504 }
3505 
3506 SYSCTL_OID(_security_jail, OID_AUTO, list,
3507     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3508     sysctl_jail_list, "S", "List of active jails");
3509 
3510 static int
3511 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3512 {
3513 	int error, injail;
3514 
3515 	injail = jailed(req->td->td_ucred);
3516 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3517 
3518 	return (error);
3519 }
3520 
3521 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3522     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3523     sysctl_jail_jailed, "I", "Process in jail?");
3524 
3525 static int
3526 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3527 {
3528 	int error, havevnet;
3529 #ifdef VIMAGE
3530 	struct ucred *cred = req->td->td_ucred;
3531 
3532 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3533 #else
3534 	havevnet = 0;
3535 #endif
3536 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3537 
3538 	return (error);
3539 }
3540 
3541 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3542     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3543     sysctl_jail_vnet, "I", "Jail owns VNET?");
3544 
3545 #if defined(INET) || defined(INET6)
3546 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3547     &jail_max_af_ips, 0,
3548     "Number of IP addresses a jail may have at most per address family (deprecated)");
3549 #endif
3550 
3551 /*
3552  * Default parameters for jail(2) compatibility.  For historical reasons,
3553  * the sysctl names have varying similarity to the parameter names.  Prisons
3554  * just see their own parameters, and can't change them.
3555  */
3556 static int
3557 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3558 {
3559 	struct prison *pr;
3560 	int allow, error, i;
3561 
3562 	pr = req->td->td_ucred->cr_prison;
3563 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3564 
3565 	/* Get the current flag value, and convert it to a boolean. */
3566 	i = (allow & arg2) ? 1 : 0;
3567 	if (arg1 != NULL)
3568 		i = !i;
3569 	error = sysctl_handle_int(oidp, &i, 0, req);
3570 	if (error || !req->newptr)
3571 		return (error);
3572 	i = i ? arg2 : 0;
3573 	if (arg1 != NULL)
3574 		i ^= arg2;
3575 	/*
3576 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3577 	 * for writing.
3578 	 */
3579 	mtx_lock(&prison0.pr_mtx);
3580 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3581 	mtx_unlock(&prison0.pr_mtx);
3582 	return (0);
3583 }
3584 
3585 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3586     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3587     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3588     "Processes in jail can set their hostnames (deprecated)");
3589 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3590     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3591     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3592     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3593 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3594     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3595     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3596     "Processes in jail can use System V IPC primitives (deprecated)");
3597 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3598     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3599     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3600     "Prison root can create raw sockets (deprecated)");
3601 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3602     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3603     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3604     "Processes in jail can alter system file flags (deprecated)");
3605 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3606     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3607     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3608     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3609 
3610 static int
3611 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3612 {
3613 	struct prison *pr;
3614 	int level, error;
3615 
3616 	pr = req->td->td_ucred->cr_prison;
3617 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3618 	error = sysctl_handle_int(oidp, &level, 0, req);
3619 	if (error || !req->newptr)
3620 		return (error);
3621 	*(int *)arg1 = level;
3622 	return (0);
3623 }
3624 
3625 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3626     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3627     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3628     sysctl_jail_default_level, "I",
3629     "Processes in jail cannot see all mounted file systems (deprecated)");
3630 
3631 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3632     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3633     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3634     sysctl_jail_default_level, "I",
3635     "Ruleset for the devfs filesystem in jail (deprecated)");
3636 
3637 /*
3638  * Nodes to describe jail parameters.  Maximum length of string parameters
3639  * is returned in the string itself, and the other parameters exist merely
3640  * to make themselves and their types known.
3641  */
3642 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3643     "Jail parameters");
3644 
3645 int
3646 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3647 {
3648 	int i;
3649 	long l;
3650 	size_t s;
3651 	char numbuf[12];
3652 
3653 	switch (oidp->oid_kind & CTLTYPE)
3654 	{
3655 	case CTLTYPE_LONG:
3656 	case CTLTYPE_ULONG:
3657 		l = 0;
3658 #ifdef SCTL_MASK32
3659 		if (!(req->flags & SCTL_MASK32))
3660 #endif
3661 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3662 	case CTLTYPE_INT:
3663 	case CTLTYPE_UINT:
3664 		i = 0;
3665 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3666 	case CTLTYPE_STRING:
3667 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3668 		return
3669 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3670 	case CTLTYPE_STRUCT:
3671 		s = (size_t)arg2;
3672 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3673 	}
3674 	return (0);
3675 }
3676 
3677 /*
3678  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3679  * jail creation time but cannot be changed in an existing jail.
3680  */
3681 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3682 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3683 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3684 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3685 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3686     "I", "Jail secure level");
3687 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3688     "Jail value for kern.osreldate and uname -K");
3689 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3690     "Jail value for kern.osrelease and uname -r");
3691 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3692     "I", "Jail cannot see all mounted file systems");
3693 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3694     "I", "Ruleset for in-jail devfs mounts");
3695 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3696     "B", "Jail persistence");
3697 #ifdef VIMAGE
3698 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3699     "E,jailsys", "Virtual network stack");
3700 #endif
3701 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3702     "B", "Jail is in the process of shutting down");
3703 
3704 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3705 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3706     "I", "Current number of child jails");
3707 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3708     "I", "Maximum number of child jails");
3709 
3710 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3711 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3712     "Jail hostname");
3713 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3714     "Jail NIS domainname");
3715 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3716     "Jail host UUID");
3717 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3718     "LU", "Jail host ID");
3719 
3720 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3721 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3722 
3723 #ifdef INET
3724 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3725     "Jail IPv4 address virtualization");
3726 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3727     "S,in_addr,a", "Jail IPv4 addresses");
3728 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3729     "B", "Do (not) use IPv4 source address selection rather than the "
3730     "primary jail IPv4 address.");
3731 #endif
3732 #ifdef INET6
3733 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3734     "Jail IPv6 address virtualization");
3735 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3736     "S,in6_addr,a", "Jail IPv6 addresses");
3737 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3738     "B", "Do (not) use IPv6 source address selection rather than the "
3739     "primary jail IPv6 address.");
3740 #endif
3741 
3742 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3743 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3744     "B", "Jail may set hostname");
3745 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3746     "B", "Jail may use SYSV IPC");
3747 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3748     "B", "Jail may create raw sockets");
3749 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3750     "B", "Jail may alter system file flags");
3751 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3752     "B", "Jail may set file quotas");
3753 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3754     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3755 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3756     "B", "Jail may bind sockets to reserved ports");
3757 
3758 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3759 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3760     "B", "Jail may mount/unmount jail-friendly file systems in general");
3761 
3762 /*
3763  * The VFS system will register jail-aware filesystems here.  They each get
3764  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3765  * attempts to mount.
3766  */
3767 void
3768 prison_add_vfs(struct vfsconf *vfsp)
3769 {
3770 	char *allow_name, *allow_noname, *mount_allowed;
3771 	struct bool_flags *bf;
3772 #ifndef NO_SYSCTL_DESCR
3773 	char *descr;
3774 #endif
3775 	unsigned allow_flag;
3776 
3777 	if (asprintf(&allow_name, M_PRISON, "allow.mount.%s", vfsp->vfc_name) <
3778 	    0 || asprintf(&allow_noname, M_PRISON, "allow.mount.no%s",
3779 	    vfsp->vfc_name) < 0) {
3780 		free(allow_name, M_PRISON);
3781 		return;
3782 	}
3783 
3784 	/*
3785 	 * See if this parameter has already beed added, i.e. if the filesystem
3786 	 * was previously loaded/unloaded.
3787 	 */
3788 	mtx_lock(&prison0.pr_mtx);
3789 	for (bf = pr_flag_allow;
3790 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3791 	     bf++) {
3792 		if (strcmp(bf->name, allow_name) == 0) {
3793 			vfsp->vfc_prison_flag = bf->flag;
3794 			goto no_add;
3795 		}
3796 	}
3797 
3798 	/*
3799 	 * Find a free bit in prison0's pr_allow, failing if there are none
3800 	 * (which shouldn't happen as long as we keep track of how many
3801 	 * filesystems are jail-aware).
3802 	 */
3803 	for (allow_flag = 1;; allow_flag <<= 1) {
3804 		if (allow_flag == 0)
3805 			goto no_add;
3806 		if ((prison0.pr_allow & allow_flag) == 0)
3807 			break;
3808 	}
3809 
3810 	/*
3811 	 * Note the parameter in the next open slot in pr_flag_allow.
3812 	 * Set the flag last so code that checks pr_flag_allow can do so
3813 	 * without locking.
3814 	 */
3815 	for (bf = pr_flag_allow; bf->flag != 0; bf++)
3816 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3817 			/* This should never happen, but is not fatal. */
3818 			goto no_add;
3819 		}
3820 	prison0.pr_allow |= allow_flag;
3821 	bf->name = allow_name;
3822 	bf->noname = allow_noname;
3823 	bf->flag = allow_flag;
3824 	vfsp->vfc_prison_flag = allow_flag;
3825 	mtx_unlock(&prison0.pr_mtx);
3826 
3827 	/*
3828 	 * Create sysctls for the paramter, and the back-compat global
3829 	 * permission.
3830 	 */
3831 #ifndef NO_SYSCTL_DESCR
3832 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3833 	    vfsp->vfc_name);
3834 #endif
3835 	(void)SYSCTL_ADD_PROC(NULL,
3836 	    SYSCTL_CHILDREN(&sysctl___security_jail_param_allow_mount),
3837 	    OID_AUTO, vfsp->vfc_name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3838 	    NULL, 0, sysctl_jail_param, "B", descr);
3839 #ifndef NO_SYSCTL_DESCR
3840 	free(descr, M_TEMP);
3841 #endif
3842 	if (asprintf(&mount_allowed, M_TEMP, "mount_%s_allowed",
3843 		vfsp->vfc_name) >= 0) {
3844 #ifndef NO_SYSCTL_DESCR
3845 		(void)asprintf(&descr, M_TEMP,
3846 		  "Processes in jail can mount the %s file system (deprecated)",
3847 		  vfsp->vfc_name);
3848 #endif
3849 		(void)SYSCTL_ADD_PROC(NULL,
3850 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO,
3851 		    mount_allowed, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3852 		    NULL, allow_flag, sysctl_jail_default_allow, "I", descr);
3853 #ifndef NO_SYSCTL_DESCR
3854 		free(descr, M_TEMP);
3855 #endif
3856 		free(mount_allowed, M_TEMP);
3857 	}
3858 	return;
3859 
3860  no_add:
3861 	mtx_unlock(&prison0.pr_mtx);
3862 	free(allow_name, M_PRISON);
3863 	free(allow_noname, M_PRISON);
3864 }
3865 
3866 #ifdef RACCT
3867 void
3868 prison_racct_foreach(void (*callback)(struct racct *racct,
3869     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3870     void *arg2, void *arg3)
3871 {
3872 	struct prison_racct *prr;
3873 
3874 	ASSERT_RACCT_ENABLED();
3875 
3876 	sx_slock(&allprison_lock);
3877 	if (pre != NULL)
3878 		(pre)();
3879 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3880 		(callback)(prr->prr_racct, arg2, arg3);
3881 	if (post != NULL)
3882 		(post)();
3883 	sx_sunlock(&allprison_lock);
3884 }
3885 
3886 static struct prison_racct *
3887 prison_racct_find_locked(const char *name)
3888 {
3889 	struct prison_racct *prr;
3890 
3891 	ASSERT_RACCT_ENABLED();
3892 	sx_assert(&allprison_lock, SA_XLOCKED);
3893 
3894 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3895 		return (NULL);
3896 
3897 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3898 		if (strcmp(name, prr->prr_name) != 0)
3899 			continue;
3900 
3901 		/* Found prison_racct with a matching name? */
3902 		prison_racct_hold(prr);
3903 		return (prr);
3904 	}
3905 
3906 	/* Add new prison_racct. */
3907 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3908 	racct_create(&prr->prr_racct);
3909 
3910 	strcpy(prr->prr_name, name);
3911 	refcount_init(&prr->prr_refcount, 1);
3912 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3913 
3914 	return (prr);
3915 }
3916 
3917 struct prison_racct *
3918 prison_racct_find(const char *name)
3919 {
3920 	struct prison_racct *prr;
3921 
3922 	ASSERT_RACCT_ENABLED();
3923 
3924 	sx_xlock(&allprison_lock);
3925 	prr = prison_racct_find_locked(name);
3926 	sx_xunlock(&allprison_lock);
3927 	return (prr);
3928 }
3929 
3930 void
3931 prison_racct_hold(struct prison_racct *prr)
3932 {
3933 
3934 	ASSERT_RACCT_ENABLED();
3935 
3936 	refcount_acquire(&prr->prr_refcount);
3937 }
3938 
3939 static void
3940 prison_racct_free_locked(struct prison_racct *prr)
3941 {
3942 
3943 	ASSERT_RACCT_ENABLED();
3944 	sx_assert(&allprison_lock, SA_XLOCKED);
3945 
3946 	if (refcount_release(&prr->prr_refcount)) {
3947 		racct_destroy(&prr->prr_racct);
3948 		LIST_REMOVE(prr, prr_next);
3949 		free(prr, M_PRISON_RACCT);
3950 	}
3951 }
3952 
3953 void
3954 prison_racct_free(struct prison_racct *prr)
3955 {
3956 	int old;
3957 
3958 	ASSERT_RACCT_ENABLED();
3959 	sx_assert(&allprison_lock, SA_UNLOCKED);
3960 
3961 	old = prr->prr_refcount;
3962 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
3963 		return;
3964 
3965 	sx_xlock(&allprison_lock);
3966 	prison_racct_free_locked(prr);
3967 	sx_xunlock(&allprison_lock);
3968 }
3969 
3970 static void
3971 prison_racct_attach(struct prison *pr)
3972 {
3973 	struct prison_racct *prr;
3974 
3975 	ASSERT_RACCT_ENABLED();
3976 	sx_assert(&allprison_lock, SA_XLOCKED);
3977 
3978 	prr = prison_racct_find_locked(pr->pr_name);
3979 	KASSERT(prr != NULL, ("cannot find prison_racct"));
3980 
3981 	pr->pr_prison_racct = prr;
3982 }
3983 
3984 /*
3985  * Handle jail renaming.  From the racct point of view, renaming means
3986  * moving from one prison_racct to another.
3987  */
3988 static void
3989 prison_racct_modify(struct prison *pr)
3990 {
3991 	struct proc *p;
3992 	struct ucred *cred;
3993 	struct prison_racct *oldprr;
3994 
3995 	ASSERT_RACCT_ENABLED();
3996 
3997 	sx_slock(&allproc_lock);
3998 	sx_xlock(&allprison_lock);
3999 
4000 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4001 		sx_xunlock(&allprison_lock);
4002 		sx_sunlock(&allproc_lock);
4003 		return;
4004 	}
4005 
4006 	oldprr = pr->pr_prison_racct;
4007 	pr->pr_prison_racct = NULL;
4008 
4009 	prison_racct_attach(pr);
4010 
4011 	/*
4012 	 * Move resource utilisation records.
4013 	 */
4014 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4015 
4016 #ifdef RCTL
4017 	/*
4018 	 * Force rctl to reattach rules to processes.
4019 	 */
4020 	FOREACH_PROC_IN_SYSTEM(p) {
4021 		PROC_LOCK(p);
4022 		cred = crhold(p->p_ucred);
4023 		PROC_UNLOCK(p);
4024 		rctl_proc_ucred_changed(p, cred);
4025 		crfree(cred);
4026 	}
4027 #endif
4028 
4029 	sx_sunlock(&allproc_lock);
4030 	prison_racct_free_locked(oldprr);
4031 	sx_xunlock(&allprison_lock);
4032 }
4033 
4034 static void
4035 prison_racct_detach(struct prison *pr)
4036 {
4037 
4038 	ASSERT_RACCT_ENABLED();
4039 	sx_assert(&allprison_lock, SA_UNLOCKED);
4040 
4041 	if (pr->pr_prison_racct == NULL)
4042 		return;
4043 	prison_racct_free(pr->pr_prison_racct);
4044 	pr->pr_prison_racct = NULL;
4045 }
4046 #endif /* RACCT */
4047 
4048 #ifdef DDB
4049 
4050 static void
4051 db_show_prison(struct prison *pr)
4052 {
4053 	struct bool_flags *bf;
4054 	struct jailsys_flags *jsf;
4055 #if defined(INET) || defined(INET6)
4056 	int ii;
4057 #endif
4058 	unsigned f;
4059 #ifdef INET
4060 	char ip4buf[INET_ADDRSTRLEN];
4061 #endif
4062 #ifdef INET6
4063 	char ip6buf[INET6_ADDRSTRLEN];
4064 #endif
4065 
4066 	db_printf("prison %p:\n", pr);
4067 	db_printf(" jid             = %d\n", pr->pr_id);
4068 	db_printf(" name            = %s\n", pr->pr_name);
4069 	db_printf(" parent          = %p\n", pr->pr_parent);
4070 	db_printf(" ref             = %d\n", pr->pr_ref);
4071 	db_printf(" uref            = %d\n", pr->pr_uref);
4072 	db_printf(" path            = %s\n", pr->pr_path);
4073 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4074 	    ? pr->pr_cpuset->cs_id : -1);
4075 #ifdef VIMAGE
4076 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4077 #endif
4078 	db_printf(" root            = %p\n", pr->pr_root);
4079 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4080 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4081 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4082 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4083 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4084 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4085 	db_printf(" flags           = 0x%x", pr->pr_flags);
4086 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4087 		if (pr->pr_flags & bf->flag)
4088 			db_printf(" %s", bf->name);
4089 	for (jsf = pr_flag_jailsys;
4090 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4091 	     jsf++) {
4092 		f = pr->pr_flags & (jsf->disable | jsf->new);
4093 		db_printf(" %-16s= %s\n", jsf->name,
4094 		    (f != 0 && f == jsf->disable) ? "disable"
4095 		    : (f == jsf->new) ? "new"
4096 		    : "inherit");
4097 	}
4098 	db_printf(" allow           = 0x%x", pr->pr_allow);
4099 	for (bf = pr_flag_allow;
4100 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4101 	     bf++)
4102 		if (pr->pr_allow & bf->flag)
4103 			db_printf(" %s", bf->name);
4104 	db_printf("\n");
4105 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4106 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4107 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4108 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4109 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4110 #ifdef INET
4111 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4112 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4113 		db_printf(" %s %s\n",
4114 		    ii == 0 ? "ip4.addr        =" : "                 ",
4115 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4116 #endif
4117 #ifdef INET6
4118 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4119 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4120 		db_printf(" %s %s\n",
4121 		    ii == 0 ? "ip6.addr        =" : "                 ",
4122 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4123 #endif
4124 }
4125 
4126 DB_SHOW_COMMAND(prison, db_show_prison_command)
4127 {
4128 	struct prison *pr;
4129 
4130 	if (!have_addr) {
4131 		/*
4132 		 * Show all prisons in the list, and prison0 which is not
4133 		 * listed.
4134 		 */
4135 		db_show_prison(&prison0);
4136 		if (!db_pager_quit) {
4137 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4138 				db_show_prison(pr);
4139 				if (db_pager_quit)
4140 					break;
4141 			}
4142 		}
4143 		return;
4144 	}
4145 
4146 	if (addr == 0)
4147 		pr = &prison0;
4148 	else {
4149 		/* Look for a prison with the ID and with references. */
4150 		TAILQ_FOREACH(pr, &allprison, pr_list)
4151 			if (pr->pr_id == addr && pr->pr_ref > 0)
4152 				break;
4153 		if (pr == NULL)
4154 			/* Look again, without requiring a reference. */
4155 			TAILQ_FOREACH(pr, &allprison, pr_list)
4156 				if (pr->pr_id == addr)
4157 					break;
4158 		if (pr == NULL)
4159 			/* Assume address points to a valid prison. */
4160 			pr = (struct prison *)addr;
4161 	}
4162 	db_show_prison(pr);
4163 }
4164 
4165 #endif /* DDB */
4166