xref: /freebsd/sys/kern/kern_jail.c (revision 058ac3e8063366dafa634d9107642e12b038bf09)
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 #include "opt_nfs.h"
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/kernel.h>
42 #include <sys/systm.h>
43 #include <sys/errno.h>
44 #include <sys/sysproto.h>
45 #include <sys/malloc.h>
46 #include <sys/osd.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/epoch.h>
50 #include <sys/taskqueue.h>
51 #include <sys/fcntl.h>
52 #include <sys/jail.h>
53 #include <sys/linker.h>
54 #include <sys/lock.h>
55 #include <sys/mman.h>
56 #include <sys/mutex.h>
57 #include <sys/racct.h>
58 #include <sys/rctl.h>
59 #include <sys/refcount.h>
60 #include <sys/sx.h>
61 #include <sys/sysent.h>
62 #include <sys/namei.h>
63 #include <sys/mount.h>
64 #include <sys/queue.h>
65 #include <sys/socket.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/uuid.h>
69 #include <sys/vnode.h>
70 
71 #include <net/if.h>
72 #include <net/vnet.h>
73 
74 #include <netinet/in.h>
75 
76 #ifdef DDB
77 #include <ddb/ddb.h>
78 #endif /* DDB */
79 
80 #include <security/mac/mac_framework.h>
81 
82 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
83 
84 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
85 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
86 
87 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
88 #ifdef INET
89 #ifdef INET6
90 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
91 #else
92 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
93 #endif
94 #else /* !INET */
95 #ifdef INET6
96 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
97 #else
98 #define	_PR_IP_SADDRSEL	0
99 #endif
100 #endif
101 
102 /* prison0 describes what is "real" about the system. */
103 struct prison prison0 = {
104 	.pr_id		= 0,
105 	.pr_name	= "0",
106 	.pr_ref		= 1,
107 	.pr_uref	= 1,
108 	.pr_path	= "/",
109 	.pr_securelevel	= -1,
110 	.pr_devfs_rsnum = 0,
111 	.pr_state	= PRISON_STATE_ALIVE,
112 	.pr_childmax	= JAIL_MAX,
113 	.pr_hostuuid	= DEFAULT_HOSTUUID,
114 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
115 #ifdef VIMAGE
116 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
117 #else
118 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
119 #endif
120 	.pr_allow	= PR_ALLOW_ALL_STATIC,
121 };
122 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
123 
124 struct bool_flags {
125 	const char	*name;
126 	const char	*noname;
127 	volatile u_int	 flag;
128 };
129 struct jailsys_flags {
130 	const char	*name;
131 	unsigned	 disable;
132 	unsigned	 new;
133 };
134 
135 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
136 struct	sx allprison_lock;
137 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
138 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
139 LIST_HEAD(, prison_racct) allprison_racct;
140 int	lastprid = 0;
141 
142 static int get_next_prid(struct prison **insprp);
143 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
144 static void prison_complete(void *context, int pending);
145 static void prison_deref(struct prison *pr, int flags);
146 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
147 static int prison_lock_xlock(struct prison *pr, int flags);
148 static void prison_cleanup(struct prison *pr);
149 static void prison_free_not_last(struct prison *pr);
150 static void prison_proc_free_not_last(struct prison *pr);
151 static void prison_proc_relink(struct prison *opr, struct prison *npr,
152     struct proc *p);
153 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
154     int enable);
155 static char *prison_path(struct prison *pr1, struct prison *pr2);
156 #ifdef RACCT
157 static void prison_racct_attach(struct prison *pr);
158 static void prison_racct_modify(struct prison *pr);
159 static void prison_racct_detach(struct prison *pr);
160 #endif
161 
162 /* Flags for prison_deref */
163 #define	PD_DEREF	0x01	/* Decrement pr_ref */
164 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
165 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
166 #define	PD_LOCKED	0x10	/* pr_mtx is held */
167 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
168 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
169 #define PD_OP_FLAGS	0x07	/* Operation flags */
170 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
171 
172 /*
173  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
174  * as we cannot figure out the size of a sparse array, or an array without a
175  * terminating entry.
176  */
177 static struct bool_flags pr_flag_bool[] = {
178 	{"persist", "nopersist", PR_PERSIST},
179 #ifdef INET
180 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
181 #endif
182 #ifdef INET6
183 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
184 #endif
185 };
186 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
187 
188 static struct jailsys_flags pr_flag_jailsys[] = {
189 	{"host", 0, PR_HOST},
190 #ifdef VIMAGE
191 	{"vnet", 0, PR_VNET},
192 #endif
193 #ifdef INET
194 	{"ip4", PR_IP4_USER, PR_IP4_USER},
195 #endif
196 #ifdef INET6
197 	{"ip6", PR_IP6_USER, PR_IP6_USER},
198 #endif
199 };
200 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
201 
202 /*
203  * Make this array full-size so dynamic parameters can be added.
204  * It is protected by prison0.mtx, but lockless reading is allowed
205  * with an atomic check of the flag values.
206  */
207 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
208 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
209 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
210 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
211 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
212 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
213 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
214 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
215 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
216 	{"allow.reserved_ports", "allow.noreserved_ports",
217 	 PR_ALLOW_RESERVED_PORTS},
218 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
219 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
220 	 PR_ALLOW_UNPRIV_DEBUG},
221 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
222 #if defined(VNET_NFSD) && defined(VIMAGE) && defined(NFSD)
223 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
224 #endif
225 };
226 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
227 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
228 
229 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
230 					 PR_ALLOW_RESERVED_PORTS | \
231 					 PR_ALLOW_UNPRIV_DEBUG | \
232 					 PR_ALLOW_SUSER)
233 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
234 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
235 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
236 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
237 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
238 #if defined(INET) || defined(INET6)
239 static unsigned jail_max_af_ips = 255;
240 #endif
241 
242 /*
243  * Initialize the parts of prison0 that can't be static-initialized with
244  * constants.  This is called from proc0_init() after creating thread0 cpuset.
245  */
246 void
247 prison0_init(void)
248 {
249 	uint8_t *file, *data;
250 	size_t size;
251 	char buf[sizeof(prison0.pr_hostuuid)];
252 	bool valid;
253 
254 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
255 	prison0.pr_osreldate = osreldate;
256 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
257 
258 	/* If we have a preloaded hostuuid, use it. */
259 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
260 	if (file != NULL) {
261 		data = preload_fetch_addr(file);
262 		size = preload_fetch_size(file);
263 		if (data != NULL) {
264 			/*
265 			 * The preloaded data may include trailing whitespace, almost
266 			 * certainly a newline; skip over any whitespace or
267 			 * non-printable characters to be safe.
268 			 */
269 			while (size > 0 && data[size - 1] <= 0x20) {
270 				size--;
271 			}
272 
273 			valid = false;
274 
275 			/*
276 			 * Not NUL-terminated when passed from loader, but
277 			 * validate_uuid requires that due to using sscanf (as
278 			 * does the subsequent strlcpy, since it still reads
279 			 * past the given size to return the true length);
280 			 * bounce to a temporary buffer to fix.
281 			 */
282 			if (size >= sizeof(buf))
283 				goto done;
284 
285 			memcpy(buf, data, size);
286 			buf[size] = '\0';
287 
288 			if (validate_uuid(buf, size, NULL, 0) != 0)
289 				goto done;
290 
291 			valid = true;
292 			(void)strlcpy(prison0.pr_hostuuid, buf,
293 			    sizeof(prison0.pr_hostuuid));
294 
295 done:
296 			if (bootverbose && !valid) {
297 				printf("hostuuid: preload data malformed: '%.*s'\n",
298 				    (int)size, data);
299 			}
300 		}
301 	}
302 	if (bootverbose)
303 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
304 }
305 
306 /*
307  * struct jail_args {
308  *	struct jail *jail;
309  * };
310  */
311 int
312 sys_jail(struct thread *td, struct jail_args *uap)
313 {
314 	uint32_t version;
315 	int error;
316 	struct jail j;
317 
318 	error = copyin(uap->jail, &version, sizeof(uint32_t));
319 	if (error)
320 		return (error);
321 
322 	switch (version) {
323 	case 0:
324 	{
325 		struct jail_v0 j0;
326 
327 		/* FreeBSD single IPv4 jails. */
328 		bzero(&j, sizeof(struct jail));
329 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
330 		if (error)
331 			return (error);
332 		j.version = j0.version;
333 		j.path = j0.path;
334 		j.hostname = j0.hostname;
335 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
336 		break;
337 	}
338 
339 	case 1:
340 		/*
341 		 * Version 1 was used by multi-IPv4 jail implementations
342 		 * that never made it into the official kernel.
343 		 */
344 		return (EINVAL);
345 
346 	case 2:	/* JAIL_API_VERSION */
347 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
348 		error = copyin(uap->jail, &j, sizeof(struct jail));
349 		if (error)
350 			return (error);
351 		break;
352 
353 	default:
354 		/* Sci-Fi jails are not supported, sorry. */
355 		return (EINVAL);
356 	}
357 	return (kern_jail(td, &j));
358 }
359 
360 int
361 kern_jail(struct thread *td, struct jail *j)
362 {
363 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
364 #ifdef INET
365 			    + 1
366 #endif
367 #ifdef INET6
368 			    + 1
369 #endif
370 			    )];
371 	struct uio opt;
372 	char *u_path, *u_hostname, *u_name;
373 	struct bool_flags *bf;
374 #ifdef INET
375 	uint32_t ip4s;
376 	struct in_addr *u_ip4;
377 #endif
378 #ifdef INET6
379 	struct in6_addr *u_ip6;
380 #endif
381 	size_t tmplen;
382 	int error, enforce_statfs;
383 
384 	bzero(&optiov, sizeof(optiov));
385 	opt.uio_iov = optiov;
386 	opt.uio_iovcnt = 0;
387 	opt.uio_offset = -1;
388 	opt.uio_resid = -1;
389 	opt.uio_segflg = UIO_SYSSPACE;
390 	opt.uio_rw = UIO_READ;
391 	opt.uio_td = td;
392 
393 	/* Set permissions for top-level jails from sysctls. */
394 	if (!jailed(td->td_ucred)) {
395 		for (bf = pr_flag_allow;
396 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
397 			atomic_load_int(&bf->flag) != 0;
398 		     bf++) {
399 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
400 			    (jail_default_allow & bf->flag)
401 			    ? bf->name : bf->noname);
402 			optiov[opt.uio_iovcnt].iov_len =
403 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
404 			opt.uio_iovcnt += 2;
405 		}
406 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
407 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
408 		opt.uio_iovcnt++;
409 		enforce_statfs = jail_default_enforce_statfs;
410 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
411 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
412 		opt.uio_iovcnt++;
413 	}
414 
415 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
416 #ifdef INET
417 	ip4s = (j->version == 0) ? 1 : j->ip4s;
418 	if (ip4s > jail_max_af_ips)
419 		return (EINVAL);
420 	tmplen += ip4s * sizeof(struct in_addr);
421 #else
422 	if (j->ip4s > 0)
423 		return (EINVAL);
424 #endif
425 #ifdef INET6
426 	if (j->ip6s > jail_max_af_ips)
427 		return (EINVAL);
428 	tmplen += j->ip6s * sizeof(struct in6_addr);
429 #else
430 	if (j->ip6s > 0)
431 		return (EINVAL);
432 #endif
433 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
434 	u_hostname = u_path + MAXPATHLEN;
435 	u_name = u_hostname + MAXHOSTNAMELEN;
436 #ifdef INET
437 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
438 #endif
439 #ifdef INET6
440 #ifdef INET
441 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
442 #else
443 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
444 #endif
445 #endif
446 	optiov[opt.uio_iovcnt].iov_base = "path";
447 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
448 	opt.uio_iovcnt++;
449 	optiov[opt.uio_iovcnt].iov_base = u_path;
450 	error = copyinstr(j->path, u_path, MAXPATHLEN,
451 	    &optiov[opt.uio_iovcnt].iov_len);
452 	if (error) {
453 		free(u_path, M_TEMP);
454 		return (error);
455 	}
456 	opt.uio_iovcnt++;
457 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
458 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
459 	opt.uio_iovcnt++;
460 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
461 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
462 	    &optiov[opt.uio_iovcnt].iov_len);
463 	if (error) {
464 		free(u_path, M_TEMP);
465 		return (error);
466 	}
467 	opt.uio_iovcnt++;
468 	if (j->jailname != NULL) {
469 		optiov[opt.uio_iovcnt].iov_base = "name";
470 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
471 		opt.uio_iovcnt++;
472 		optiov[opt.uio_iovcnt].iov_base = u_name;
473 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
474 		    &optiov[opt.uio_iovcnt].iov_len);
475 		if (error) {
476 			free(u_path, M_TEMP);
477 			return (error);
478 		}
479 		opt.uio_iovcnt++;
480 	}
481 #ifdef INET
482 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
483 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
484 	opt.uio_iovcnt++;
485 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
486 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
487 	if (j->version == 0)
488 		u_ip4->s_addr = j->ip4s;
489 	else {
490 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
491 		if (error) {
492 			free(u_path, M_TEMP);
493 			return (error);
494 		}
495 	}
496 	opt.uio_iovcnt++;
497 #endif
498 #ifdef INET6
499 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
500 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
501 	opt.uio_iovcnt++;
502 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
503 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
504 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
505 	if (error) {
506 		free(u_path, M_TEMP);
507 		return (error);
508 	}
509 	opt.uio_iovcnt++;
510 #endif
511 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
512 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
513 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
514 	free(u_path, M_TEMP);
515 	return (error);
516 }
517 
518 /*
519  * struct jail_set_args {
520  *	struct iovec *iovp;
521  *	unsigned int iovcnt;
522  *	int flags;
523  * };
524  */
525 int
526 sys_jail_set(struct thread *td, struct jail_set_args *uap)
527 {
528 	struct uio *auio;
529 	int error;
530 
531 	/* Check that we have an even number of iovecs. */
532 	if (uap->iovcnt & 1)
533 		return (EINVAL);
534 
535 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
536 	if (error)
537 		return (error);
538 	error = kern_jail_set(td, auio, uap->flags);
539 	free(auio, M_IOV);
540 	return (error);
541 }
542 
543 #if defined(INET) || defined(INET6)
544 typedef int prison_addr_cmp_t(const void *, const void *);
545 typedef bool prison_addr_valid_t(const void *);
546 static const struct pr_family {
547 	size_t			size;
548 	prison_addr_cmp_t	*cmp;
549 	prison_addr_valid_t	*valid;
550 	int			ip_flag;
551 } pr_families[PR_FAMILY_MAX] = {
552 #ifdef INET
553 	[PR_INET] = {
554 		.size = sizeof(struct in_addr),
555 		.cmp = prison_qcmp_v4,
556 		.valid = prison_valid_v4,
557 		.ip_flag = PR_IP4_USER,
558 	 },
559 #endif
560 #ifdef INET6
561 	[PR_INET6] = {
562 		.size = sizeof(struct in6_addr),
563 		.cmp = prison_qcmp_v6,
564 		.valid = prison_valid_v6,
565 		.ip_flag = PR_IP6_USER,
566 	},
567 #endif
568 };
569 
570 /*
571  * Network address lists (pr_addrs) allocation for jails.  The addresses
572  * are accessed locklessly by the network stack, thus need to be protected by
573  * the network epoch.
574  */
575 struct prison_ip {
576 	struct epoch_context ctx;
577 	uint32_t	ips;
578 #ifdef FUTURE_C
579 	union {
580 		struct in_addr pr_ip4[];
581 		struct in6_addr pr_ip6[];
582 	};
583 #else /* No future C :( */
584 #define	PR_IP(pip, i)	((const char *)((pip) + 1) + pr_families[af].size * (i))
585 #define	PR_IPD(pip, i)	((char *)((pip) + 1) + pr_families[af].size * (i))
586 #endif
587 };
588 
589 static struct prison_ip *
590 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
591 {
592 	struct prison_ip *pip;
593 
594 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
595 	    M_PRISON, flags);
596 	if (pip != NULL)
597 		pip->ips = cnt;
598 	return (pip);
599 }
600 
601 /*
602  * Allocate and copyin user supplied address list, sorting and validating.
603  * kern_jail_set() helper.
604  */
605 static struct prison_ip *
606 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
607 {
608 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
609 	const size_t size = pr_families[af].size;
610 	struct prison_ip *pip;
611 
612 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
613 	bcopy(op, pip + 1, cnt * size);
614 	/*
615 	 * IP addresses are all sorted but ip[0] to preserve
616 	 * the primary IP address as given from userland.
617 	 * This special IP is used for unbound outgoing
618 	 * connections as well for "loopback" traffic in case
619 	 * source address selection cannot find any more fitting
620 	 * address to connect from.
621 	 */
622 	if (cnt > 1)
623 		qsort((char *)(pip + 1) + size, cnt - 1, size,
624 		    pr_families[af].cmp);
625 	/*
626 	 * Check for duplicate addresses and do some simple
627 	 * zero and broadcast checks. If users give other bogus
628 	 * addresses it is their problem.
629 	 */
630 	for (int i = 0; i < cnt; i++) {
631 		if (!pr_families[af].valid(PR_IP(pip, i))) {
632 			free(pip, M_PRISON);
633 			return (NULL);
634 		}
635 		if (i + 1 < cnt &&
636 		    (cmp(PR_IP(pip, 0), PR_IP(pip, i + 1)) == 0 ||
637 		     cmp(PR_IP(pip, i), PR_IP(pip, i + 1)) == 0)) {
638 			free(pip, M_PRISON);
639 			return (NULL);
640 		}
641 	}
642 
643 	return (pip);
644 }
645 
646 /*
647  * Allocate and dup parent prison address list.
648  * kern_jail_set() helper.
649  */
650 static void
651 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
652 {
653 
654 	if (ppr->pr_addrs[af] != NULL) {
655 		pr->pr_addrs[af] = prison_ip_alloc(af,
656 		    ppr->pr_addrs[af]->ips, M_WAITOK);
657 		bcopy(ppr->pr_addrs[af], pr->pr_addrs[af],
658 		    pr->pr_addrs[af]->ips * pr_families[af].size);
659 	}
660 }
661 
662 /*
663  * Make sure the new set of IP addresses is a subset of the parent's list.
664  * Don't worry about the parent being unlocked, as any setting is done with
665  * allprison_lock held.
666  * kern_jail_set() helper.
667  */
668 static bool
669 prison_ip_parent_match(const struct prison_ip *ppip,
670     const struct prison_ip *pip, const pr_family_t af)
671 {
672 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
673 	int i, j;
674 
675 	if (ppip == NULL)
676 		return (false);
677 
678 	for (i = 0; i < ppip->ips; i++)
679 		if (cmp(PR_IP(pip, 0), PR_IP(ppip, i)) == 0)
680 			break;
681 
682 	if (i == ppip->ips)
683 		/* Main address not present in parent. */
684 		return (false);
685 
686 	if (pip->ips > 1) {
687 		for (i = j = 1; i < pip->ips; i++) {
688 			if (cmp(PR_IP(pip, i), PR_IP(ppip, 0)) == 0)
689 				/* Equals to parent primary address. */
690 				continue;
691 			for (; j < ppip->ips; j++)
692 				if (cmp(PR_IP(pip, i), PR_IP(ppip, j)) == 0)
693 					break;
694 			if (j == ppip->ips)
695 				break;
696 		}
697 		if (j == ppip->ips)
698 			/* Address not present in parent. */
699 			return (false);
700 	}
701 	return (true);
702 }
703 
704 /*
705  * Check for conflicting IP addresses.  We permit them if there is no more
706  * than one IP on each jail.  If there is a duplicate on a jail with more
707  * than one IP stop checking and return error.
708  * kern_jail_set() helper.
709  */
710 static bool
711 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
712     const struct prison_ip *pip, pr_family_t af)
713 {
714 	const struct prison *tppr, *tpr;
715 	int descend;
716 
717 #ifdef VIMAGE
718 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
719 		if (tppr->pr_flags & PR_VNET)
720 			break;
721 #else
722 	tppr = &prison0;
723 #endif
724 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
725 		if (tpr == pr ||
726 #ifdef VIMAGE
727 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
728 #endif
729 		    !prison_isalive(tpr)) {
730 			descend = 0;
731 			continue;
732 		}
733 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
734 			continue;
735 		descend = 0;
736 		if (tpr->pr_addrs[af] == NULL ||
737 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
738 			continue;
739 		for (int i = 0; i < pip->ips; i++)
740 			if (prison_ip_check(tpr, af, PR_IP(pip, i)) == 0)
741 				return (false);
742 	}
743 
744 	return (true);
745 }
746 
747 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
748     "prison must start with epoch context");
749 static void
750 prison_ip_free_deferred(epoch_context_t ctx)
751 {
752 
753 	free(ctx, M_PRISON);
754 }
755 
756 static void
757 prison_ip_free(struct prison_ip *pip)
758 {
759 
760 	if (pip != NULL)
761 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
762 }
763 
764 static void
765 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
766 {
767 	struct prison_ip **mem, *old;
768 
769 	mtx_assert(&pr->pr_mtx, MA_OWNED);
770 
771 	mem = &pr->pr_addrs[af];
772 
773 	old = *mem;
774 	ck_pr_store_ptr(mem, new);
775 	prison_ip_free(old);
776 }
777 
778 /*
779  * Restrict a prison's IP address list with its parent's, possibly replacing
780  * it.  Return true if the replacement buffer was used (or would have been).
781  * kern_jail_set() helper.
782  */
783 static bool
784 prison_ip_restrict(struct prison *pr, const pr_family_t af,
785     struct prison_ip *new)
786 {
787 	const struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
788 	const struct prison_ip *pip = pr->pr_addrs[af];
789 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
790 	const size_t size = pr_families[af].size;
791 	uint32_t ips;
792 	bool alloced;
793 
794 	mtx_assert(&pr->pr_mtx, MA_OWNED);
795 
796 	/*
797 	 * Due to epoch-synchronized access to the IP address lists we always
798 	 * allocate a new list even if the old one has enough space.  We could
799 	 * atomically update an IPv4 address inside a list, but that would
800 	 * screw up sorting, and in case of IPv6 we can't even atomically write
801 	 * one.
802 	 */
803 	ips = (pr->pr_flags & pr_families[af].ip_flag) ? pip->ips : ppip->ips;
804 	if (ips == 0) {
805 		prison_ip_set(pr, af, NULL);
806 		return (false);
807 	}
808 	if (new == NULL) {
809 		new = prison_ip_alloc(af, ips, M_NOWAIT);
810 		if (new == NULL)
811 			return (true);
812 		alloced = true;
813 	} else
814 		alloced = false;
815 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
816 		/* This has no user settings, so just copy the parent's list. */
817 		bcopy(ppip, new, ips * size);
818 	} else {
819 		/* Remove addresses that aren't in the parent. */
820 		int i;
821 
822 		i = 0; /* index in pip */
823 		ips = 0; /* index in new */
824 
825 		for (int pi = 0; pi < ppip->ips; pi++)
826 			if (cmp(PR_IP(pip, 0), PR_IP(ppip, pi)) == 0) {
827 				/* Found our primary address in parent. */
828 				bcopy(PR_IP(pip, i), PR_IPD(new, ips), size);
829 				i++;
830 				ips++;
831 				break;
832 			}
833 		for (int pi = 1; i < pip->ips; ) {
834 			/* Check against primary, which is unsorted. */
835 			if (cmp(PR_IP(pip, i), PR_IP(ppip, 0)) == 0) {
836 				/* Matches parent's primary address. */
837 				bcopy(PR_IP(pip, i), PR_IPD(new, ips), size);
838 				i++;
839 				ips++;
840 				continue;
841 			}
842 			/* The rest are sorted. */
843 			switch (pi >= ppip->ips ? -1 :
844 				cmp(PR_IP(pip, i), PR_IP(ppip, pi))) {
845 			case -1:
846 				i++;
847 				break;
848 			case 0:
849 				bcopy(PR_IP(pr, i), PR_IPD(new, ips), size);
850 				i++;
851 				pi++;
852 				ips++;
853 				break;
854 			case 1:
855 				pi++;
856 				break;
857 			}
858 		}
859 		if (ips == 0) {
860 			if (alloced)
861 				prison_ip_free(new);
862 			new = NULL;
863 		}
864 	}
865 	prison_ip_set(pr, af, new);
866 	return (new != NULL ? true : false);
867 }
868 
869 /*
870  * Fast-path check if an address belongs to a prison.
871  */
872 int
873 prison_ip_check(const struct prison *pr, const pr_family_t af,
874     const void *addr)
875 {
876 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
877 	const struct prison_ip *pip;
878 	int i, a, z, d;
879 
880 	MPASS(mtx_owned(&pr->pr_mtx) ||
881 	    in_epoch(net_epoch_preempt) ||
882 	    sx_xlocked(&allprison_lock));
883 
884 	pip = ck_pr_load_ptr(&pr->pr_addrs[af]);
885 	if (__predict_false(pip == NULL))
886 		return (EAFNOSUPPORT);
887 
888 	/* Check the primary IP. */
889 	if (cmp(PR_IP(pip, 0), addr) == 0)
890 		return (0);
891 
892 	/*
893 	 * All the other IPs are sorted so we can do a binary search.
894 	 */
895 	a = 0;
896 	z = pip->ips - 2;
897 	while (a <= z) {
898 		i = (a + z) / 2;
899 		d = cmp(PR_IP(pip, i + 1), addr);
900 		if (d > 0)
901 			z = i - 1;
902 		else if (d < 0)
903 			a = i + 1;
904 		else
905 			return (0);
906 	}
907 
908 	return (EADDRNOTAVAIL);
909 }
910 
911 /*
912  * Grab primary IP.  Historically required mutex, but nothing prevents
913  * us to support epoch-protected access.  Is it used in fast path?
914  * in{6}_jail.c helper
915  */
916 const void *
917 prison_ip_get0(const struct prison *pr, const pr_family_t af)
918 {
919 	const struct prison_ip *pip = pr->pr_addrs[af];
920 
921 	mtx_assert(&pr->pr_mtx, MA_OWNED);
922 	MPASS(pip);
923 
924 	return (pip + 1);
925 }
926 
927 u_int
928 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
929 {
930 
931 	return (pr->pr_addrs[af]->ips);
932 }
933 #endif	/* defined(INET) || defined(INET6) */
934 
935 int
936 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
937 {
938 	struct nameidata nd;
939 #ifdef INET
940 	struct prison_ip *ip4;
941 #endif
942 #ifdef INET6
943 	struct prison_ip *ip6;
944 #endif
945 	struct vfsopt *opt;
946 	struct vfsoptlist *opts;
947 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
948 	struct vnode *root;
949 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
950 	char *g_path, *osrelstr;
951 	struct bool_flags *bf;
952 	struct jailsys_flags *jsf;
953 #if defined(INET) || defined(INET6)
954 	void *op;
955 #endif
956 	unsigned long hid;
957 	size_t namelen, onamelen, pnamelen;
958 	int born, created, cuflags, descend, drflags, enforce;
959 	int error, errmsg_len, errmsg_pos;
960 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
961 	int jid, jsys, len, level;
962 	int childmax, osreldt, rsnum, slevel;
963 #ifdef INET
964 	int ip4s, redo_ip4;
965 #endif
966 #ifdef INET6
967 	int ip6s, redo_ip6;
968 #endif
969 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
970 	uint64_t pr_allow_diff;
971 	unsigned tallow;
972 	char numbuf[12];
973 
974 	error = priv_check(td, PRIV_JAIL_SET);
975 	if (!error && (flags & JAIL_ATTACH))
976 		error = priv_check(td, PRIV_JAIL_ATTACH);
977 	if (error)
978 		return (error);
979 	mypr = td->td_ucred->cr_prison;
980 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
981 		return (EPERM);
982 	if (flags & ~JAIL_SET_MASK)
983 		return (EINVAL);
984 
985 	/*
986 	 * Check all the parameters before committing to anything.  Not all
987 	 * errors can be caught early, but we may as well try.  Also, this
988 	 * takes care of some expensive stuff (path lookup) before getting
989 	 * the allprison lock.
990 	 *
991 	 * XXX Jails are not filesystems, and jail parameters are not mount
992 	 *     options.  But it makes more sense to re-use the vfsopt code
993 	 *     than duplicate it under a different name.
994 	 */
995 	error = vfs_buildopts(optuio, &opts);
996 	if (error)
997 		return (error);
998 #ifdef INET
999 	ip4 = NULL;
1000 #endif
1001 #ifdef INET6
1002 	ip6 = NULL;
1003 #endif
1004 	g_path = NULL;
1005 
1006 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1007 	if (!cuflags) {
1008 		error = EINVAL;
1009 		vfs_opterror(opts, "no valid operation (create or update)");
1010 		goto done_errmsg;
1011 	}
1012 
1013 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1014 	if (error == ENOENT)
1015 		jid = 0;
1016 	else if (error != 0)
1017 		goto done_free;
1018 
1019 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1020 	if (error == ENOENT)
1021 		gotslevel = 0;
1022 	else if (error != 0)
1023 		goto done_free;
1024 	else
1025 		gotslevel = 1;
1026 
1027 	error =
1028 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1029 	if (error == ENOENT)
1030 		gotchildmax = 0;
1031 	else if (error != 0)
1032 		goto done_free;
1033 	else
1034 		gotchildmax = 1;
1035 
1036 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1037 	if (error == ENOENT)
1038 		gotenforce = 0;
1039 	else if (error != 0)
1040 		goto done_free;
1041 	else if (enforce < 0 || enforce > 2) {
1042 		error = EINVAL;
1043 		goto done_free;
1044 	} else
1045 		gotenforce = 1;
1046 
1047 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1048 	if (error == ENOENT)
1049 		gotrsnum = 0;
1050 	else if (error != 0)
1051 		goto done_free;
1052 	else
1053 		gotrsnum = 1;
1054 
1055 	pr_flags = ch_flags = 0;
1056 	for (bf = pr_flag_bool;
1057 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1058 	     bf++) {
1059 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1060 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1061 	}
1062 	ch_flags |= pr_flags;
1063 	for (jsf = pr_flag_jailsys;
1064 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1065 	     jsf++) {
1066 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1067 		if (error == ENOENT)
1068 			continue;
1069 		if (error != 0)
1070 			goto done_free;
1071 		switch (jsys) {
1072 		case JAIL_SYS_DISABLE:
1073 			if (!jsf->disable) {
1074 				error = EINVAL;
1075 				goto done_free;
1076 			}
1077 			pr_flags |= jsf->disable;
1078 			break;
1079 		case JAIL_SYS_NEW:
1080 			pr_flags |= jsf->new;
1081 			break;
1082 		case JAIL_SYS_INHERIT:
1083 			break;
1084 		default:
1085 			error = EINVAL;
1086 			goto done_free;
1087 		}
1088 		ch_flags |= jsf->new | jsf->disable;
1089 	}
1090 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1091 	    && !(pr_flags & PR_PERSIST)) {
1092 		error = EINVAL;
1093 		vfs_opterror(opts, "new jail must persist or attach");
1094 		goto done_errmsg;
1095 	}
1096 #ifdef VIMAGE
1097 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1098 		error = EINVAL;
1099 		vfs_opterror(opts, "vnet cannot be changed after creation");
1100 		goto done_errmsg;
1101 	}
1102 #endif
1103 #ifdef INET
1104 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1105 		error = EINVAL;
1106 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1107 		goto done_errmsg;
1108 	}
1109 #endif
1110 #ifdef INET6
1111 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1112 		error = EINVAL;
1113 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1114 		goto done_errmsg;
1115 	}
1116 #endif
1117 
1118 	pr_allow = ch_allow = 0;
1119 	for (bf = pr_flag_allow;
1120 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1121 		atomic_load_int(&bf->flag) != 0;
1122 	     bf++) {
1123 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1124 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1125 	}
1126 	ch_allow |= pr_allow;
1127 
1128 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1129 	if (error == ENOENT)
1130 		name = NULL;
1131 	else if (error != 0)
1132 		goto done_free;
1133 	else {
1134 		if (len == 0 || name[len - 1] != '\0') {
1135 			error = EINVAL;
1136 			goto done_free;
1137 		}
1138 		if (len > MAXHOSTNAMELEN) {
1139 			error = ENAMETOOLONG;
1140 			goto done_free;
1141 		}
1142 	}
1143 
1144 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1145 	if (error == ENOENT)
1146 		host = NULL;
1147 	else if (error != 0)
1148 		goto done_free;
1149 	else {
1150 		ch_flags |= PR_HOST;
1151 		pr_flags |= PR_HOST;
1152 		if (len == 0 || host[len - 1] != '\0') {
1153 			error = EINVAL;
1154 			goto done_free;
1155 		}
1156 		if (len > MAXHOSTNAMELEN) {
1157 			error = ENAMETOOLONG;
1158 			goto done_free;
1159 		}
1160 	}
1161 
1162 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1163 	if (error == ENOENT)
1164 		domain = NULL;
1165 	else if (error != 0)
1166 		goto done_free;
1167 	else {
1168 		ch_flags |= PR_HOST;
1169 		pr_flags |= PR_HOST;
1170 		if (len == 0 || domain[len - 1] != '\0') {
1171 			error = EINVAL;
1172 			goto done_free;
1173 		}
1174 		if (len > MAXHOSTNAMELEN) {
1175 			error = ENAMETOOLONG;
1176 			goto done_free;
1177 		}
1178 	}
1179 
1180 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1181 	if (error == ENOENT)
1182 		uuid = NULL;
1183 	else if (error != 0)
1184 		goto done_free;
1185 	else {
1186 		ch_flags |= PR_HOST;
1187 		pr_flags |= PR_HOST;
1188 		if (len == 0 || uuid[len - 1] != '\0') {
1189 			error = EINVAL;
1190 			goto done_free;
1191 		}
1192 		if (len > HOSTUUIDLEN) {
1193 			error = ENAMETOOLONG;
1194 			goto done_free;
1195 		}
1196 	}
1197 
1198 #ifdef COMPAT_FREEBSD32
1199 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1200 		uint32_t hid32;
1201 
1202 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1203 		hid = hid32;
1204 	} else
1205 #endif
1206 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1207 	if (error == ENOENT)
1208 		gothid = 0;
1209 	else if (error != 0)
1210 		goto done_free;
1211 	else {
1212 		gothid = 1;
1213 		ch_flags |= PR_HOST;
1214 		pr_flags |= PR_HOST;
1215 	}
1216 
1217 #ifdef INET
1218 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1219 	if (error == ENOENT)
1220 		ip4s = 0;
1221 	else if (error != 0)
1222 		goto done_free;
1223 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1224 		error = EINVAL;
1225 		goto done_free;
1226 	} else {
1227 		ch_flags |= PR_IP4_USER;
1228 		pr_flags |= PR_IP4_USER;
1229 		if (ip4s > 0) {
1230 			ip4s /= sizeof(struct in_addr);
1231 			if (ip4s > jail_max_af_ips) {
1232 				error = EINVAL;
1233 				vfs_opterror(opts, "too many IPv4 addresses");
1234 				goto done_errmsg;
1235 			}
1236 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1237 			if (ip4 == NULL) {
1238 				error = EINVAL;
1239 				goto done_free;
1240 			}
1241 		}
1242 	}
1243 #endif
1244 
1245 #ifdef INET6
1246 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1247 	if (error == ENOENT)
1248 		ip6s = 0;
1249 	else if (error != 0)
1250 		goto done_free;
1251 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1252 		error = EINVAL;
1253 		goto done_free;
1254 	} else {
1255 		ch_flags |= PR_IP6_USER;
1256 		pr_flags |= PR_IP6_USER;
1257 		if (ip6s > 0) {
1258 			ip6s /= sizeof(struct in6_addr);
1259 			if (ip6s > jail_max_af_ips) {
1260 				error = EINVAL;
1261 				vfs_opterror(opts, "too many IPv6 addresses");
1262 				goto done_errmsg;
1263 			}
1264 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1265 			if (ip6 == NULL) {
1266 				error = EINVAL;
1267 				goto done_free;
1268 			}
1269 		}
1270 	}
1271 #endif
1272 
1273 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1274 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1275 		error = EINVAL;
1276 		vfs_opterror(opts,
1277 		    "vnet jails cannot have IP address restrictions");
1278 		goto done_errmsg;
1279 	}
1280 #endif
1281 
1282 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1283 	if (error == ENOENT)
1284 		osrelstr = NULL;
1285 	else if (error != 0)
1286 		goto done_free;
1287 	else {
1288 		if (flags & JAIL_UPDATE) {
1289 			error = EINVAL;
1290 			vfs_opterror(opts,
1291 			    "osrelease cannot be changed after creation");
1292 			goto done_errmsg;
1293 		}
1294 		if (len == 0 || osrelstr[len - 1] != '\0') {
1295 			error = EINVAL;
1296 			goto done_free;
1297 		}
1298 		if (len >= OSRELEASELEN) {
1299 			error = ENAMETOOLONG;
1300 			vfs_opterror(opts,
1301 			    "osrelease string must be 1-%d bytes long",
1302 			    OSRELEASELEN - 1);
1303 			goto done_errmsg;
1304 		}
1305 	}
1306 
1307 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1308 	if (error == ENOENT)
1309 		osreldt = 0;
1310 	else if (error != 0)
1311 		goto done_free;
1312 	else {
1313 		if (flags & JAIL_UPDATE) {
1314 			error = EINVAL;
1315 			vfs_opterror(opts,
1316 			    "osreldate cannot be changed after creation");
1317 			goto done_errmsg;
1318 		}
1319 		if (osreldt == 0) {
1320 			error = EINVAL;
1321 			vfs_opterror(opts, "osreldate cannot be 0");
1322 			goto done_errmsg;
1323 		}
1324 	}
1325 
1326 	root = NULL;
1327 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1328 	if (error == ENOENT)
1329 		path = NULL;
1330 	else if (error != 0)
1331 		goto done_free;
1332 	else {
1333 		if (flags & JAIL_UPDATE) {
1334 			error = EINVAL;
1335 			vfs_opterror(opts,
1336 			    "path cannot be changed after creation");
1337 			goto done_errmsg;
1338 		}
1339 		if (len == 0 || path[len - 1] != '\0') {
1340 			error = EINVAL;
1341 			goto done_free;
1342 		}
1343 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1344 		error = namei(&nd);
1345 		if (error)
1346 			goto done_free;
1347 		root = nd.ni_vp;
1348 		NDFREE_PNBUF(&nd);
1349 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1350 		strlcpy(g_path, path, MAXPATHLEN);
1351 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1352 		if (error == 0) {
1353 			path = g_path;
1354 		} else {
1355 			/* exit on other errors */
1356 			goto done_free;
1357 		}
1358 		if (root->v_type != VDIR) {
1359 			error = ENOTDIR;
1360 			vput(root);
1361 			goto done_free;
1362 		}
1363 		VOP_UNLOCK(root);
1364 	}
1365 
1366 	/*
1367 	 * Find the specified jail, or at least its parent.
1368 	 * This abuses the file error codes ENOENT and EEXIST.
1369 	 */
1370 	pr = NULL;
1371 	inspr = NULL;
1372 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1373 		namelc = strrchr(name, '.');
1374 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1375 		if (*p != '\0')
1376 			jid = 0;
1377 	}
1378 	sx_xlock(&allprison_lock);
1379 	drflags = PD_LIST_XLOCKED;
1380 	ppr = mypr;
1381 	if (!prison_isalive(ppr)) {
1382 		/* This jail is dying.  This process will surely follow. */
1383 		error = EAGAIN;
1384 		goto done_deref;
1385 	}
1386 	if (jid != 0) {
1387 		if (jid < 0) {
1388 			error = EINVAL;
1389 			vfs_opterror(opts, "negative jid");
1390 			goto done_deref;
1391 		}
1392 		/*
1393 		 * See if a requested jid already exists.  Keep track of
1394 		 * where it can be inserted later.
1395 		 */
1396 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1397 			if (inspr->pr_id < jid)
1398 				continue;
1399 			if (inspr->pr_id > jid)
1400 				break;
1401 			pr = inspr;
1402 			mtx_lock(&pr->pr_mtx);
1403 			drflags |= PD_LOCKED;
1404 			inspr = NULL;
1405 			break;
1406 		}
1407 		if (pr != NULL) {
1408 			/* Create: jid must not exist. */
1409 			if (cuflags == JAIL_CREATE) {
1410 				/*
1411 				 * Even creators that cannot see the jail will
1412 				 * get EEXIST.
1413 				 */
1414 				error = EEXIST;
1415 				vfs_opterror(opts, "jail %d already exists",
1416 				    jid);
1417 				goto done_deref;
1418 			}
1419 			if (!prison_ischild(mypr, pr)) {
1420 				/*
1421 				 * Updaters get ENOENT if they cannot see the
1422 				 * jail.  This is true even for CREATE | UPDATE,
1423 				 * which normally cannot give this error.
1424 				 */
1425 				error = ENOENT;
1426 				vfs_opterror(opts, "jail %d not found", jid);
1427 				goto done_deref;
1428 			}
1429 			ppr = pr->pr_parent;
1430 			if (!prison_isalive(ppr)) {
1431 				error = ENOENT;
1432 				vfs_opterror(opts, "jail %d is dying",
1433 				    ppr->pr_id);
1434 				goto done_deref;
1435 			}
1436 			if (!prison_isalive(pr)) {
1437 				if (!(flags & JAIL_DYING)) {
1438 					error = ENOENT;
1439 					vfs_opterror(opts, "jail %d is dying",
1440 					    jid);
1441 					goto done_deref;
1442 				}
1443 				if ((flags & JAIL_ATTACH) ||
1444 				    (pr_flags & PR_PERSIST)) {
1445 					/*
1446 					 * A dying jail might be resurrected
1447 					 * (via attach or persist), but first
1448 					 * it must determine if another jail
1449 					 * has claimed its name.  Accomplish
1450 					 * this by implicitly re-setting the
1451 					 * name.
1452 					 */
1453 					if (name == NULL)
1454 						name = prison_name(mypr, pr);
1455 				}
1456 			}
1457 		} else {
1458 			/* Update: jid must exist. */
1459 			if (cuflags == JAIL_UPDATE) {
1460 				error = ENOENT;
1461 				vfs_opterror(opts, "jail %d not found", jid);
1462 				goto done_deref;
1463 			}
1464 		}
1465 	}
1466 	/*
1467 	 * If the caller provided a name, look for a jail by that name.
1468 	 * This has different semantics for creates and updates keyed by jid
1469 	 * (where the name must not already exist in a different jail),
1470 	 * and updates keyed by the name itself (where the name must exist
1471 	 * because that is the jail being updated).
1472 	 */
1473 	namelc = NULL;
1474 	if (name != NULL) {
1475 		namelc = strrchr(name, '.');
1476 		if (namelc == NULL)
1477 			namelc = name;
1478 		else {
1479 			/*
1480 			 * This is a hierarchical name.  Split it into the
1481 			 * parent and child names, and make sure the parent
1482 			 * exists or matches an already found jail.
1483 			 */
1484 			if (pr != NULL) {
1485 				if (strncmp(name, ppr->pr_name, namelc - name)
1486 				    || ppr->pr_name[namelc - name] != '\0') {
1487 					error = EINVAL;
1488 					vfs_opterror(opts,
1489 					    "cannot change jail's parent");
1490 					goto done_deref;
1491 				}
1492 			} else {
1493 				*namelc = '\0';
1494 				ppr = prison_find_name(mypr, name);
1495 				if (ppr == NULL) {
1496 					error = ENOENT;
1497 					vfs_opterror(opts,
1498 					    "jail \"%s\" not found", name);
1499 					goto done_deref;
1500 				}
1501 				mtx_unlock(&ppr->pr_mtx);
1502 				if (!prison_isalive(ppr)) {
1503 					error = ENOENT;
1504 					vfs_opterror(opts,
1505 					    "jail \"%s\" is dying", name);
1506 					goto done_deref;
1507 				}
1508 				*namelc = '.';
1509 			}
1510 			namelc++;
1511 		}
1512 		if (namelc[0] != '\0') {
1513 			pnamelen =
1514 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1515 			deadpr = NULL;
1516 			FOREACH_PRISON_CHILD(ppr, tpr) {
1517 				if (tpr != pr &&
1518 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1519 					if (prison_isalive(tpr)) {
1520 						if (pr == NULL &&
1521 						    cuflags != JAIL_CREATE) {
1522 							/*
1523 							 * Use this jail
1524 							 * for updates.
1525 							 */
1526 							pr = tpr;
1527 							mtx_lock(&pr->pr_mtx);
1528 							drflags |= PD_LOCKED;
1529 							break;
1530 						}
1531 						/*
1532 						 * Create, or update(jid):
1533 						 * name must not exist in an
1534 						 * active sibling jail.
1535 						 */
1536 						error = EEXIST;
1537 						vfs_opterror(opts,
1538 						   "jail \"%s\" already exists",
1539 						   name);
1540 						goto done_deref;
1541 					}
1542 					if (pr == NULL &&
1543 					    cuflags != JAIL_CREATE) {
1544 						deadpr = tpr;
1545 					}
1546 				}
1547 			}
1548 			/* If no active jail is found, use a dying one. */
1549 			if (deadpr != NULL && pr == NULL) {
1550 				if (flags & JAIL_DYING) {
1551 					pr = deadpr;
1552 					mtx_lock(&pr->pr_mtx);
1553 					drflags |= PD_LOCKED;
1554 				} else if (cuflags == JAIL_UPDATE) {
1555 					error = ENOENT;
1556 					vfs_opterror(opts,
1557 					    "jail \"%s\" is dying", name);
1558 					goto done_deref;
1559 				}
1560 			}
1561 			/* Update: name must exist if no jid. */
1562 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1563 				error = ENOENT;
1564 				vfs_opterror(opts, "jail \"%s\" not found",
1565 				    name);
1566 				goto done_deref;
1567 			}
1568 		}
1569 	}
1570 	/* Update: must provide a jid or name. */
1571 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1572 		error = ENOENT;
1573 		vfs_opterror(opts, "update specified no jail");
1574 		goto done_deref;
1575 	}
1576 
1577 	/* If there's no prison to update, create a new one and link it in. */
1578 	created = pr == NULL;
1579 	if (created) {
1580 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1581 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1582 				error = EPERM;
1583 				vfs_opterror(opts, "prison limit exceeded");
1584 				goto done_deref;
1585 			}
1586 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1587 			error = EAGAIN;
1588 			vfs_opterror(opts, "no available jail IDs");
1589 			goto done_deref;
1590 		}
1591 
1592 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1593 		pr->pr_state = PRISON_STATE_INVALID;
1594 		refcount_init(&pr->pr_ref, 1);
1595 		refcount_init(&pr->pr_uref, 0);
1596 		drflags |= PD_DEREF;
1597 		LIST_INIT(&pr->pr_children);
1598 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1599 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1600 
1601 		pr->pr_id = jid;
1602 		if (inspr != NULL)
1603 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1604 		else
1605 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1606 
1607 		pr->pr_parent = ppr;
1608 		prison_hold(ppr);
1609 		prison_proc_hold(ppr);
1610 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1611 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1612 			tpr->pr_childcount++;
1613 
1614 		/* Set some default values, and inherit some from the parent. */
1615 		if (namelc == NULL)
1616 			namelc = "";
1617 		if (path == NULL) {
1618 			path = "/";
1619 			root = mypr->pr_root;
1620 			vref(root);
1621 		}
1622 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1623 		pr->pr_flags |= PR_HOST;
1624 #if defined(INET) || defined(INET6)
1625 #ifdef VIMAGE
1626 		if (!(pr_flags & PR_VNET))
1627 #endif
1628 		{
1629 #ifdef INET
1630 			if (!(ch_flags & PR_IP4_USER))
1631 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1632 			else if (!(pr_flags & PR_IP4_USER)) {
1633 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1634 				prison_ip_dup(ppr, pr, PR_INET);
1635 			}
1636 #endif
1637 #ifdef INET6
1638 			if (!(ch_flags & PR_IP6_USER))
1639 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1640 			else if (!(pr_flags & PR_IP6_USER)) {
1641 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1642 				prison_ip_dup(ppr, pr, PR_INET6);
1643 			}
1644 #endif
1645 		}
1646 #endif
1647 		/* Source address selection is always on by default. */
1648 		pr->pr_flags |= _PR_IP_SADDRSEL;
1649 
1650 		pr->pr_securelevel = ppr->pr_securelevel;
1651 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1652 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1653 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1654 
1655 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1656 		if (osrelstr == NULL)
1657 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1658 			    sizeof(pr->pr_osrelease));
1659 		else
1660 			strlcpy(pr->pr_osrelease, osrelstr,
1661 			    sizeof(pr->pr_osrelease));
1662 
1663 #ifdef VIMAGE
1664 		/* Allocate a new vnet if specified. */
1665 		pr->pr_vnet = (pr_flags & PR_VNET)
1666 		    ? vnet_alloc() : ppr->pr_vnet;
1667 #endif
1668 		/*
1669 		 * Allocate a dedicated cpuset for each jail.
1670 		 * Unlike other initial settings, this may return an error.
1671 		 */
1672 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1673 		if (error)
1674 			goto done_deref;
1675 
1676 		mtx_lock(&pr->pr_mtx);
1677 		drflags |= PD_LOCKED;
1678 	} else {
1679 		/*
1680 		 * Grab a reference for existing prisons, to ensure they
1681 		 * continue to exist for the duration of the call.
1682 		 */
1683 		prison_hold(pr);
1684 		drflags |= PD_DEREF;
1685 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1686 		if ((pr->pr_flags & PR_VNET) &&
1687 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1688 			error = EINVAL;
1689 			vfs_opterror(opts,
1690 			    "vnet jails cannot have IP address restrictions");
1691 			goto done_deref;
1692 		}
1693 #endif
1694 #ifdef INET
1695 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1696 			error = EINVAL;
1697 			vfs_opterror(opts,
1698 			    "ip4 cannot be changed after creation");
1699 			goto done_deref;
1700 		}
1701 #endif
1702 #ifdef INET6
1703 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1704 			error = EINVAL;
1705 			vfs_opterror(opts,
1706 			    "ip6 cannot be changed after creation");
1707 			goto done_deref;
1708 		}
1709 #endif
1710 	}
1711 
1712 	/* Do final error checking before setting anything. */
1713 	if (gotslevel) {
1714 		if (slevel < ppr->pr_securelevel) {
1715 			error = EPERM;
1716 			goto done_deref;
1717 		}
1718 	}
1719 	if (gotchildmax) {
1720 		if (childmax >= ppr->pr_childmax) {
1721 			error = EPERM;
1722 			goto done_deref;
1723 		}
1724 	}
1725 	if (gotenforce) {
1726 		if (enforce < ppr->pr_enforce_statfs) {
1727 			error = EPERM;
1728 			goto done_deref;
1729 		}
1730 	}
1731 	if (gotrsnum) {
1732 		/*
1733 		 * devfs_rsnum is a uint16_t
1734 		 */
1735 		if (rsnum < 0 || rsnum > 65535) {
1736 			error = EINVAL;
1737 			goto done_deref;
1738 		}
1739 		/*
1740 		 * Nested jails always inherit parent's devfs ruleset
1741 		 */
1742 		if (jailed(td->td_ucred)) {
1743 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1744 				error = EPERM;
1745 				goto done_deref;
1746 			} else
1747 				rsnum = ppr->pr_devfs_rsnum;
1748 		}
1749 	}
1750 #ifdef INET
1751 	if (ip4s > 0) {
1752 		if ((ppr->pr_flags & PR_IP4) &&
1753 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1754 		    PR_INET)) {
1755 			error = EPERM;
1756 			goto done_deref;
1757 		}
1758 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1759 			error = EADDRINUSE;
1760 			vfs_opterror(opts, "IPv4 addresses clash");
1761 			goto done_deref;
1762 		}
1763 	}
1764 #endif
1765 #ifdef INET6
1766 	if (ip6s > 0) {
1767 		if ((ppr->pr_flags & PR_IP6) &&
1768 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1769 		    PR_INET6)) {
1770 			error = EPERM;
1771 			goto done_deref;
1772 		}
1773 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1774 			error = EADDRINUSE;
1775 			vfs_opterror(opts, "IPv6 addresses clash");
1776 			goto done_deref;
1777 		}
1778 	}
1779 #endif
1780 	onamelen = namelen = 0;
1781 	if (namelc != NULL) {
1782 		/* Give a default name of the jid.  Also allow the name to be
1783 		 * explicitly the jid - but not any other number, and only in
1784 		 * normal form (no leading zero/etc).
1785 		 */
1786 		if (namelc[0] == '\0')
1787 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1788 		else if ((strtoul(namelc, &p, 10) != jid ||
1789 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1790 			error = EINVAL;
1791 			vfs_opterror(opts,
1792 			    "name cannot be numeric (unless it is the jid)");
1793 			goto done_deref;
1794 		}
1795 		/*
1796 		 * Make sure the name isn't too long for the prison or its
1797 		 * children.
1798 		 */
1799 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1800 		onamelen = strlen(pr->pr_name + pnamelen);
1801 		namelen = strlen(namelc);
1802 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1803 			error = ENAMETOOLONG;
1804 			goto done_deref;
1805 		}
1806 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1807 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1808 			    sizeof(pr->pr_name)) {
1809 				error = ENAMETOOLONG;
1810 				goto done_deref;
1811 			}
1812 		}
1813 	}
1814 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1815 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1816 		error = EPERM;
1817 		goto done_deref;
1818 	}
1819 
1820 	/*
1821 	 * Let modules check their parameters.  This requires unlocking and
1822 	 * then re-locking the prison, but this is still a valid state as long
1823 	 * as allprison_lock remains xlocked.
1824 	 */
1825 	mtx_unlock(&pr->pr_mtx);
1826 	drflags &= ~PD_LOCKED;
1827 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1828 	if (error != 0)
1829 		goto done_deref;
1830 	mtx_lock(&pr->pr_mtx);
1831 	drflags |= PD_LOCKED;
1832 
1833 	/* At this point, all valid parameters should have been noted. */
1834 	TAILQ_FOREACH(opt, opts, link) {
1835 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1836 			error = EINVAL;
1837 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1838 			goto done_deref;
1839 		}
1840 	}
1841 
1842 	/* Set the parameters of the prison. */
1843 #ifdef INET
1844 	redo_ip4 = 0;
1845 	if (pr_flags & PR_IP4_USER) {
1846 		pr->pr_flags |= PR_IP4;
1847 		prison_ip_set(pr, PR_INET, ip4);
1848 		ip4 = NULL;
1849 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1850 #ifdef VIMAGE
1851 			if (tpr->pr_flags & PR_VNET) {
1852 				descend = 0;
1853 				continue;
1854 			}
1855 #endif
1856 			if (prison_ip_restrict(tpr, PR_INET, NULL)) {
1857 				redo_ip4 = 1;
1858 				descend = 0;
1859 			}
1860 		}
1861 	}
1862 #endif
1863 #ifdef INET6
1864 	redo_ip6 = 0;
1865 	if (pr_flags & PR_IP6_USER) {
1866 		pr->pr_flags |= PR_IP6;
1867 		prison_ip_set(pr, PR_INET6, ip6);
1868 		ip6 = NULL;
1869 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1870 #ifdef VIMAGE
1871 			if (tpr->pr_flags & PR_VNET) {
1872 				descend = 0;
1873 				continue;
1874 			}
1875 #endif
1876 			if (prison_ip_restrict(tpr, PR_INET6, NULL)) {
1877 				redo_ip6 = 1;
1878 				descend = 0;
1879 			}
1880 		}
1881 	}
1882 #endif
1883 	if (gotslevel) {
1884 		pr->pr_securelevel = slevel;
1885 		/* Set all child jails to be at least this level. */
1886 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1887 			if (tpr->pr_securelevel < slevel)
1888 				tpr->pr_securelevel = slevel;
1889 	}
1890 	if (gotchildmax) {
1891 		pr->pr_childmax = childmax;
1892 		/* Set all child jails to under this limit. */
1893 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1894 			if (tpr->pr_childmax > childmax - level)
1895 				tpr->pr_childmax = childmax > level
1896 				    ? childmax - level : 0;
1897 	}
1898 	if (gotenforce) {
1899 		pr->pr_enforce_statfs = enforce;
1900 		/* Pass this restriction on to the children. */
1901 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1902 			if (tpr->pr_enforce_statfs < enforce)
1903 				tpr->pr_enforce_statfs = enforce;
1904 	}
1905 	if (gotrsnum) {
1906 		pr->pr_devfs_rsnum = rsnum;
1907 		/* Pass this restriction on to the children. */
1908 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1909 			tpr->pr_devfs_rsnum = rsnum;
1910 	}
1911 	if (namelc != NULL) {
1912 		if (ppr == &prison0)
1913 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1914 		else
1915 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1916 			    ppr->pr_name, namelc);
1917 		/* Change this component of child names. */
1918 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1919 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1920 			    strlen(tpr->pr_name + onamelen) + 1);
1921 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1922 		}
1923 	}
1924 	if (path != NULL) {
1925 		/* Try to keep a real-rooted full pathname. */
1926 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1927 		pr->pr_root = root;
1928 		root = NULL;
1929 	}
1930 	if (PR_HOST & ch_flags & ~pr_flags) {
1931 		if (pr->pr_flags & PR_HOST) {
1932 			/*
1933 			 * Copy the parent's host info.  As with pr_ip4 above,
1934 			 * the lack of a lock on the parent is not a problem;
1935 			 * it is always set with allprison_lock at least
1936 			 * shared, and is held exclusively here.
1937 			 */
1938 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1939 			    sizeof(pr->pr_hostname));
1940 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1941 			    sizeof(pr->pr_domainname));
1942 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1943 			    sizeof(pr->pr_hostuuid));
1944 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1945 		}
1946 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1947 		/* Set this prison, and any descendants without PR_HOST. */
1948 		if (host != NULL)
1949 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1950 		if (domain != NULL)
1951 			strlcpy(pr->pr_domainname, domain,
1952 			    sizeof(pr->pr_domainname));
1953 		if (uuid != NULL)
1954 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1955 		if (gothid)
1956 			pr->pr_hostid = hid;
1957 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1958 			if (tpr->pr_flags & PR_HOST)
1959 				descend = 0;
1960 			else {
1961 				if (host != NULL)
1962 					strlcpy(tpr->pr_hostname,
1963 					    pr->pr_hostname,
1964 					    sizeof(tpr->pr_hostname));
1965 				if (domain != NULL)
1966 					strlcpy(tpr->pr_domainname,
1967 					    pr->pr_domainname,
1968 					    sizeof(tpr->pr_domainname));
1969 				if (uuid != NULL)
1970 					strlcpy(tpr->pr_hostuuid,
1971 					    pr->pr_hostuuid,
1972 					    sizeof(tpr->pr_hostuuid));
1973 				if (gothid)
1974 					tpr->pr_hostid = hid;
1975 			}
1976 		}
1977 	}
1978 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1979 	if ((tallow = ch_allow & ~pr_allow))
1980 		prison_set_allow_locked(pr, tallow, 0);
1981 	/*
1982 	 * Persistent prisons get an extra reference, and prisons losing their
1983 	 * persist flag lose that reference.
1984 	 */
1985 	born = !prison_isalive(pr);
1986 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
1987 		if (pr_flags & PR_PERSIST) {
1988 			prison_hold(pr);
1989 			/*
1990 			 * This may make a dead prison alive again, but wait
1991 			 * to label it as such until after OSD calls have had
1992 			 * a chance to run (and perhaps to fail).
1993 			 */
1994 			refcount_acquire(&pr->pr_uref);
1995 		} else {
1996 			drflags |= PD_DEUREF;
1997 			prison_free_not_last(pr);
1998 		}
1999 	}
2000 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2001 	mtx_unlock(&pr->pr_mtx);
2002 	drflags &= ~PD_LOCKED;
2003 	/*
2004 	 * Any errors past this point will need to de-persist newly created
2005 	 * prisons, as well as call remove methods.
2006 	 */
2007 	if (born)
2008 		drflags |= PD_KILL;
2009 
2010 #ifdef RACCT
2011 	if (racct_enable && created)
2012 		prison_racct_attach(pr);
2013 #endif
2014 
2015 	/* Locks may have prevented a complete restriction of child IP
2016 	 * addresses.  If so, allocate some more memory and try again.
2017 	 */
2018 #ifdef INET
2019 	while (redo_ip4) {
2020 		ip4s = pr->pr_addrs[PR_INET]->ips;
2021 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2022 		mtx_lock(&pr->pr_mtx);
2023 		redo_ip4 = 0;
2024 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2025 #ifdef VIMAGE
2026 			if (tpr->pr_flags & PR_VNET) {
2027 				descend = 0;
2028 				continue;
2029 			}
2030 #endif
2031 			if (prison_ip_restrict(tpr, PR_INET, ip4)) {
2032 				if (ip4 != NULL)
2033 					ip4 = NULL;
2034 				else
2035 					redo_ip4 = 1;
2036 			}
2037 		}
2038 		mtx_unlock(&pr->pr_mtx);
2039 	}
2040 #endif
2041 #ifdef INET6
2042 	while (redo_ip6) {
2043 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2044 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2045 		mtx_lock(&pr->pr_mtx);
2046 		redo_ip6 = 0;
2047 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2048 #ifdef VIMAGE
2049 			if (tpr->pr_flags & PR_VNET) {
2050 				descend = 0;
2051 				continue;
2052 			}
2053 #endif
2054 			if (prison_ip_restrict(tpr, PR_INET6, ip6)) {
2055 				if (ip6 != NULL)
2056 					ip6 = NULL;
2057 				else
2058 					redo_ip6 = 1;
2059 			}
2060 		}
2061 		mtx_unlock(&pr->pr_mtx);
2062 	}
2063 #endif
2064 
2065 	/* Let the modules do their work. */
2066 	if (born) {
2067 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2068 		if (error)
2069 			goto done_deref;
2070 	}
2071 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2072 	if (error)
2073 		goto done_deref;
2074 
2075 	/*
2076 	 * A new prison is now ready to be seen; either it has gained a user
2077 	 * reference via persistence, or is about to gain one via attachment.
2078 	 */
2079 	if (born) {
2080 		drflags = prison_lock_xlock(pr, drflags);
2081 		pr->pr_state = PRISON_STATE_ALIVE;
2082 	}
2083 
2084 	/* Attach this process to the prison if requested. */
2085 	if (flags & JAIL_ATTACH) {
2086 		error = do_jail_attach(td, pr,
2087 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2088 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2089 		if (error) {
2090 			vfs_opterror(opts, "attach failed");
2091 			goto done_deref;
2092 		}
2093 	}
2094 
2095 #ifdef RACCT
2096 	if (racct_enable && !created) {
2097 		if (drflags & PD_LOCKED) {
2098 			mtx_unlock(&pr->pr_mtx);
2099 			drflags &= ~PD_LOCKED;
2100 		}
2101 		if (drflags & PD_LIST_XLOCKED) {
2102 			sx_xunlock(&allprison_lock);
2103 			drflags &= ~PD_LIST_XLOCKED;
2104 		}
2105 		prison_racct_modify(pr);
2106 	}
2107 #endif
2108 
2109 #ifdef VNET_NFSD
2110 	if (born && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2111 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2112 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2113 		   " file system\n", pr->pr_id);
2114 #endif
2115 
2116 	drflags &= ~PD_KILL;
2117 	td->td_retval[0] = pr->pr_id;
2118 
2119  done_deref:
2120 	/* Release any temporary prison holds and/or locks. */
2121 	if (pr != NULL)
2122 		prison_deref(pr, drflags);
2123 	else if (drflags & PD_LIST_SLOCKED)
2124 		sx_sunlock(&allprison_lock);
2125 	else if (drflags & PD_LIST_XLOCKED)
2126 		sx_xunlock(&allprison_lock);
2127 	if (root != NULL)
2128 		vrele(root);
2129  done_errmsg:
2130 	if (error) {
2131 		/* Write the error message back to userspace. */
2132 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2133 		    &errmsg_len) == 0 && errmsg_len > 0) {
2134 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2135 			if (optuio->uio_segflg == UIO_SYSSPACE)
2136 				bcopy(errmsg,
2137 				    optuio->uio_iov[errmsg_pos].iov_base,
2138 				    errmsg_len);
2139 			else
2140 				copyout(errmsg,
2141 				    optuio->uio_iov[errmsg_pos].iov_base,
2142 				    errmsg_len);
2143 		}
2144 	}
2145  done_free:
2146 #ifdef INET
2147 	prison_ip_free(ip4);
2148 #endif
2149 #ifdef INET6
2150 	prison_ip_free(ip6);
2151 #endif
2152 	if (g_path != NULL)
2153 		free(g_path, M_TEMP);
2154 	vfs_freeopts(opts);
2155 	return (error);
2156 }
2157 
2158 /*
2159  * Find the next available prison ID.  Return the ID on success, or zero
2160  * on failure.  Also set a pointer to the allprison list entry the prison
2161  * should be inserted before.
2162  */
2163 static int
2164 get_next_prid(struct prison **insprp)
2165 {
2166 	struct prison *inspr;
2167 	int jid, maxid;
2168 
2169 	jid = lastprid % JAIL_MAX + 1;
2170 	if (TAILQ_EMPTY(&allprison) ||
2171 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2172 		/*
2173 		 * A common case is for all jails to be implicitly numbered,
2174 		 * which means they'll go on the end of the list, at least
2175 		 * for the first JAIL_MAX times.
2176 		 */
2177 		inspr = NULL;
2178 	} else {
2179 		/*
2180 		 * Take two passes through the allprison list: first starting
2181 		 * with the proposed jid, then ending with it.
2182 		 */
2183 		for (maxid = JAIL_MAX; maxid != 0; ) {
2184 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2185 				if (inspr->pr_id < jid)
2186 					continue;
2187 				if (inspr->pr_id > jid) {
2188 					/* Found an opening. */
2189 					maxid = 0;
2190 					break;
2191 				}
2192 				if (++jid > maxid) {
2193 					if (lastprid == maxid || lastprid == 0)
2194 					{
2195 						/*
2196 						 * The entire legal range
2197 						 * has been traversed
2198 						 */
2199 						return 0;
2200 					}
2201 					/* Try again from the start. */
2202 					jid = 1;
2203 					maxid = lastprid;
2204 					break;
2205 				}
2206 			}
2207 			if (inspr == NULL) {
2208 				/* Found room at the end of the list. */
2209 				break;
2210 			}
2211 		}
2212 	}
2213 	*insprp = inspr;
2214 	lastprid = jid;
2215 	return (jid);
2216 }
2217 
2218 /*
2219  * struct jail_get_args {
2220  *	struct iovec *iovp;
2221  *	unsigned int iovcnt;
2222  *	int flags;
2223  * };
2224  */
2225 int
2226 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2227 {
2228 	struct uio *auio;
2229 	int error;
2230 
2231 	/* Check that we have an even number of iovecs. */
2232 	if (uap->iovcnt & 1)
2233 		return (EINVAL);
2234 
2235 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2236 	if (error)
2237 		return (error);
2238 	error = kern_jail_get(td, auio, uap->flags);
2239 	if (error == 0)
2240 		error = copyout(auio->uio_iov, uap->iovp,
2241 		    uap->iovcnt * sizeof (struct iovec));
2242 	free(auio, M_IOV);
2243 	return (error);
2244 }
2245 
2246 int
2247 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2248 {
2249 	struct bool_flags *bf;
2250 	struct jailsys_flags *jsf;
2251 	struct prison *pr, *mypr;
2252 	struct vfsopt *opt;
2253 	struct vfsoptlist *opts;
2254 	char *errmsg, *name;
2255 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2256 	unsigned f;
2257 
2258 	if (flags & ~JAIL_GET_MASK)
2259 		return (EINVAL);
2260 
2261 	/* Get the parameter list. */
2262 	error = vfs_buildopts(optuio, &opts);
2263 	if (error)
2264 		return (error);
2265 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2266 	mypr = td->td_ucred->cr_prison;
2267 	pr = NULL;
2268 
2269 	/*
2270 	 * Find the prison specified by one of: lastjid, jid, name.
2271 	 */
2272 	sx_slock(&allprison_lock);
2273 	drflags = PD_LIST_SLOCKED;
2274 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2275 	if (error == 0) {
2276 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2277 			if (pr->pr_id > jid &&
2278 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2279 			    prison_ischild(mypr, pr)) {
2280 				mtx_lock(&pr->pr_mtx);
2281 				drflags |= PD_LOCKED;
2282 				goto found_prison;
2283 			}
2284 		}
2285 		error = ENOENT;
2286 		vfs_opterror(opts, "no jail after %d", jid);
2287 		goto done;
2288 	} else if (error != ENOENT)
2289 		goto done;
2290 
2291 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2292 	if (error == 0) {
2293 		if (jid != 0) {
2294 			pr = prison_find_child(mypr, jid);
2295 			if (pr != NULL) {
2296 				drflags |= PD_LOCKED;
2297 				if (!(prison_isalive(pr) ||
2298 				    (flags & JAIL_DYING))) {
2299 					error = ENOENT;
2300 					vfs_opterror(opts, "jail %d is dying",
2301 					    jid);
2302 					goto done;
2303 				}
2304 				goto found_prison;
2305 			}
2306 			error = ENOENT;
2307 			vfs_opterror(opts, "jail %d not found", jid);
2308 			goto done;
2309 		}
2310 	} else if (error != ENOENT)
2311 		goto done;
2312 
2313 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2314 	if (error == 0) {
2315 		if (len == 0 || name[len - 1] != '\0') {
2316 			error = EINVAL;
2317 			goto done;
2318 		}
2319 		pr = prison_find_name(mypr, name);
2320 		if (pr != NULL) {
2321 			drflags |= PD_LOCKED;
2322 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2323 				error = ENOENT;
2324 				vfs_opterror(opts, "jail \"%s\" is dying",
2325 				    name);
2326 				goto done;
2327 			}
2328 			goto found_prison;
2329 		}
2330 		error = ENOENT;
2331 		vfs_opterror(opts, "jail \"%s\" not found", name);
2332 		goto done;
2333 	} else if (error != ENOENT)
2334 		goto done;
2335 
2336 	vfs_opterror(opts, "no jail specified");
2337 	error = ENOENT;
2338 	goto done;
2339 
2340  found_prison:
2341 	/* Get the parameters of the prison. */
2342 	prison_hold(pr);
2343 	drflags |= PD_DEREF;
2344 	td->td_retval[0] = pr->pr_id;
2345 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2346 	if (error != 0 && error != ENOENT)
2347 		goto done;
2348 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2349 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2350 	if (error != 0 && error != ENOENT)
2351 		goto done;
2352 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2353 	if (error != 0 && error != ENOENT)
2354 		goto done;
2355 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2356 	    sizeof(pr->pr_cpuset->cs_id));
2357 	if (error != 0 && error != ENOENT)
2358 		goto done;
2359 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2360 	if (error != 0 && error != ENOENT)
2361 		goto done;
2362 #ifdef INET
2363 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET] + 1,
2364 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2365 	    pr_families[PR_INET].size : 0 );
2366 	if (error != 0 && error != ENOENT)
2367 		goto done;
2368 #endif
2369 #ifdef INET6
2370 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6] + 1,
2371 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2372 	    pr_families[PR_INET6].size : 0 );
2373 	if (error != 0 && error != ENOENT)
2374 		goto done;
2375 #endif
2376 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2377 	    sizeof(pr->pr_securelevel));
2378 	if (error != 0 && error != ENOENT)
2379 		goto done;
2380 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2381 	    sizeof(pr->pr_childcount));
2382 	if (error != 0 && error != ENOENT)
2383 		goto done;
2384 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2385 	    sizeof(pr->pr_childmax));
2386 	if (error != 0 && error != ENOENT)
2387 		goto done;
2388 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2389 	if (error != 0 && error != ENOENT)
2390 		goto done;
2391 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2392 	if (error != 0 && error != ENOENT)
2393 		goto done;
2394 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2395 	if (error != 0 && error != ENOENT)
2396 		goto done;
2397 #ifdef COMPAT_FREEBSD32
2398 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2399 		uint32_t hid32 = pr->pr_hostid;
2400 
2401 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2402 	} else
2403 #endif
2404 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2405 	    sizeof(pr->pr_hostid));
2406 	if (error != 0 && error != ENOENT)
2407 		goto done;
2408 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2409 	    sizeof(pr->pr_enforce_statfs));
2410 	if (error != 0 && error != ENOENT)
2411 		goto done;
2412 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2413 	    sizeof(pr->pr_devfs_rsnum));
2414 	if (error != 0 && error != ENOENT)
2415 		goto done;
2416 	for (bf = pr_flag_bool;
2417 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2418 	     bf++) {
2419 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2420 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2421 		if (error != 0 && error != ENOENT)
2422 			goto done;
2423 		i = !i;
2424 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2425 		if (error != 0 && error != ENOENT)
2426 			goto done;
2427 	}
2428 	for (jsf = pr_flag_jailsys;
2429 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2430 	     jsf++) {
2431 		f = pr->pr_flags & (jsf->disable | jsf->new);
2432 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2433 		    : (f == jsf->new) ? JAIL_SYS_NEW
2434 		    : JAIL_SYS_INHERIT;
2435 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2436 		if (error != 0 && error != ENOENT)
2437 			goto done;
2438 	}
2439 	for (bf = pr_flag_allow;
2440 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2441 		atomic_load_int(&bf->flag) != 0;
2442 	     bf++) {
2443 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2444 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2445 		if (error != 0 && error != ENOENT)
2446 			goto done;
2447 		i = !i;
2448 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2449 		if (error != 0 && error != ENOENT)
2450 			goto done;
2451 	}
2452 	i = !prison_isalive(pr);
2453 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2454 	if (error != 0 && error != ENOENT)
2455 		goto done;
2456 	i = !i;
2457 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2458 	if (error != 0 && error != ENOENT)
2459 		goto done;
2460 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2461 	    sizeof(pr->pr_osreldate));
2462 	if (error != 0 && error != ENOENT)
2463 		goto done;
2464 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2465 	if (error != 0 && error != ENOENT)
2466 		goto done;
2467 
2468 	/* Get the module parameters. */
2469 	mtx_unlock(&pr->pr_mtx);
2470 	drflags &= ~PD_LOCKED;
2471 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2472 	if (error)
2473 		goto done;
2474 	prison_deref(pr, drflags);
2475 	pr = NULL;
2476 	drflags = 0;
2477 
2478 	/* By now, all parameters should have been noted. */
2479 	TAILQ_FOREACH(opt, opts, link) {
2480 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2481 			error = EINVAL;
2482 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2483 			goto done;
2484 		}
2485 	}
2486 
2487 	/* Write the fetched parameters back to userspace. */
2488 	error = 0;
2489 	TAILQ_FOREACH(opt, opts, link) {
2490 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2491 			pos = 2 * opt->pos + 1;
2492 			optuio->uio_iov[pos].iov_len = opt->len;
2493 			if (opt->value != NULL) {
2494 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2495 					bcopy(opt->value,
2496 					    optuio->uio_iov[pos].iov_base,
2497 					    opt->len);
2498 				} else {
2499 					error = copyout(opt->value,
2500 					    optuio->uio_iov[pos].iov_base,
2501 					    opt->len);
2502 					if (error)
2503 						break;
2504 				}
2505 			}
2506 		}
2507 	}
2508 
2509  done:
2510 	/* Release any temporary prison holds and/or locks. */
2511 	if (pr != NULL)
2512 		prison_deref(pr, drflags);
2513 	else if (drflags & PD_LIST_SLOCKED)
2514 		sx_sunlock(&allprison_lock);
2515 	if (error && errmsg_pos >= 0) {
2516 		/* Write the error message back to userspace. */
2517 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2518 		errmsg_pos = 2 * errmsg_pos + 1;
2519 		if (errmsg_len > 0) {
2520 			if (optuio->uio_segflg == UIO_SYSSPACE)
2521 				bcopy(errmsg,
2522 				    optuio->uio_iov[errmsg_pos].iov_base,
2523 				    errmsg_len);
2524 			else
2525 				copyout(errmsg,
2526 				    optuio->uio_iov[errmsg_pos].iov_base,
2527 				    errmsg_len);
2528 		}
2529 	}
2530 	vfs_freeopts(opts);
2531 	return (error);
2532 }
2533 
2534 /*
2535  * struct jail_remove_args {
2536  *	int jid;
2537  * };
2538  */
2539 int
2540 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2541 {
2542 	struct prison *pr;
2543 	int error;
2544 
2545 	error = priv_check(td, PRIV_JAIL_REMOVE);
2546 	if (error)
2547 		return (error);
2548 
2549 	sx_xlock(&allprison_lock);
2550 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2551 	if (pr == NULL) {
2552 		sx_xunlock(&allprison_lock);
2553 		return (EINVAL);
2554 	}
2555 	if (!prison_isalive(pr)) {
2556 		/* Silently ignore already-dying prisons. */
2557 		mtx_unlock(&pr->pr_mtx);
2558 		sx_xunlock(&allprison_lock);
2559 		return (0);
2560 	}
2561 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2562 	return (0);
2563 }
2564 
2565 /*
2566  * struct jail_attach_args {
2567  *	int jid;
2568  * };
2569  */
2570 int
2571 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2572 {
2573 	struct prison *pr;
2574 	int error;
2575 
2576 	error = priv_check(td, PRIV_JAIL_ATTACH);
2577 	if (error)
2578 		return (error);
2579 
2580 	sx_slock(&allprison_lock);
2581 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2582 	if (pr == NULL) {
2583 		sx_sunlock(&allprison_lock);
2584 		return (EINVAL);
2585 	}
2586 
2587 	/* Do not allow a process to attach to a prison that is not alive. */
2588 	if (!prison_isalive(pr)) {
2589 		mtx_unlock(&pr->pr_mtx);
2590 		sx_sunlock(&allprison_lock);
2591 		return (EINVAL);
2592 	}
2593 
2594 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2595 }
2596 
2597 static int
2598 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2599 {
2600 	struct proc *p;
2601 	struct ucred *newcred, *oldcred;
2602 	int error;
2603 
2604 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2605 	sx_assert(&allprison_lock, SX_LOCKED);
2606 	drflags &= PD_LOCK_FLAGS;
2607 	/*
2608 	 * XXX: Note that there is a slight race here if two threads
2609 	 * in the same privileged process attempt to attach to two
2610 	 * different jails at the same time.  It is important for
2611 	 * user processes not to do this, or they might end up with
2612 	 * a process root from one prison, but attached to the jail
2613 	 * of another.
2614 	 */
2615 	prison_hold(pr);
2616 	refcount_acquire(&pr->pr_uref);
2617 	drflags |= PD_DEREF | PD_DEUREF;
2618 	mtx_unlock(&pr->pr_mtx);
2619 	drflags &= ~PD_LOCKED;
2620 
2621 	/* Let modules do whatever they need to prepare for attaching. */
2622 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2623 	if (error) {
2624 		prison_deref(pr, drflags);
2625 		return (error);
2626 	}
2627 	sx_unlock(&allprison_lock);
2628 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2629 
2630 	/*
2631 	 * Reparent the newly attached process to this jail.
2632 	 */
2633 	p = td->td_proc;
2634 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2635 	if (error)
2636 		goto e_revert_osd;
2637 
2638 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2639 	if ((error = change_dir(pr->pr_root, td)) != 0)
2640 		goto e_unlock;
2641 #ifdef MAC
2642 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2643 		goto e_unlock;
2644 #endif
2645 	VOP_UNLOCK(pr->pr_root);
2646 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2647 		goto e_revert_osd;
2648 
2649 	newcred = crget();
2650 	PROC_LOCK(p);
2651 	oldcred = crcopysafe(p, newcred);
2652 	newcred->cr_prison = pr;
2653 	proc_set_cred(p, newcred);
2654 	setsugid(p);
2655 #ifdef RACCT
2656 	racct_proc_ucred_changed(p, oldcred, newcred);
2657 	crhold(newcred);
2658 #endif
2659 	PROC_UNLOCK(p);
2660 #ifdef RCTL
2661 	rctl_proc_ucred_changed(p, newcred);
2662 	crfree(newcred);
2663 #endif
2664 	prison_proc_relink(oldcred->cr_prison, pr, p);
2665 	prison_deref(oldcred->cr_prison, drflags);
2666 	crfree(oldcred);
2667 
2668 	/*
2669 	 * If the prison was killed while changing credentials, die along
2670 	 * with it.
2671 	 */
2672 	if (!prison_isalive(pr)) {
2673 		PROC_LOCK(p);
2674 		kern_psignal(p, SIGKILL);
2675 		PROC_UNLOCK(p);
2676 	}
2677 
2678 	return (0);
2679 
2680  e_unlock:
2681 	VOP_UNLOCK(pr->pr_root);
2682  e_revert_osd:
2683 	/* Tell modules this thread is still in its old jail after all. */
2684 	sx_slock(&allprison_lock);
2685 	drflags |= PD_LIST_SLOCKED;
2686 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2687 	prison_deref(pr, drflags);
2688 	return (error);
2689 }
2690 
2691 /*
2692  * Returns a locked prison instance, or NULL on failure.
2693  */
2694 struct prison *
2695 prison_find(int prid)
2696 {
2697 	struct prison *pr;
2698 
2699 	sx_assert(&allprison_lock, SX_LOCKED);
2700 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2701 		if (pr->pr_id < prid)
2702 			continue;
2703 		if (pr->pr_id > prid)
2704 			break;
2705 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2706 		mtx_lock(&pr->pr_mtx);
2707 		return (pr);
2708 	}
2709 	return (NULL);
2710 }
2711 
2712 /*
2713  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2714  */
2715 struct prison *
2716 prison_find_child(struct prison *mypr, int prid)
2717 {
2718 	struct prison *pr;
2719 	int descend;
2720 
2721 	sx_assert(&allprison_lock, SX_LOCKED);
2722 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2723 		if (pr->pr_id == prid) {
2724 			KASSERT(prison_isvalid(pr),
2725 			    ("Found invalid prison %p", pr));
2726 			mtx_lock(&pr->pr_mtx);
2727 			return (pr);
2728 		}
2729 	}
2730 	return (NULL);
2731 }
2732 
2733 /*
2734  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2735  */
2736 struct prison *
2737 prison_find_name(struct prison *mypr, const char *name)
2738 {
2739 	struct prison *pr, *deadpr;
2740 	size_t mylen;
2741 	int descend;
2742 
2743 	sx_assert(&allprison_lock, SX_LOCKED);
2744 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2745 	deadpr = NULL;
2746 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2747 		if (!strcmp(pr->pr_name + mylen, name)) {
2748 			KASSERT(prison_isvalid(pr),
2749 			    ("Found invalid prison %p", pr));
2750 			if (prison_isalive(pr)) {
2751 				mtx_lock(&pr->pr_mtx);
2752 				return (pr);
2753 			}
2754 			deadpr = pr;
2755 		}
2756 	}
2757 	/* There was no valid prison - perhaps there was a dying one. */
2758 	if (deadpr != NULL)
2759 		mtx_lock(&deadpr->pr_mtx);
2760 	return (deadpr);
2761 }
2762 
2763 /*
2764  * See if a prison has the specific flag set.  The prison should be locked,
2765  * unless checking for flags that are only set at jail creation (such as
2766  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2767  * to any other prison data.
2768  */
2769 int
2770 prison_flag(struct ucred *cred, unsigned flag)
2771 {
2772 
2773 	return (cred->cr_prison->pr_flags & flag);
2774 }
2775 
2776 int
2777 prison_allow(struct ucred *cred, unsigned flag)
2778 {
2779 
2780 	return ((cred->cr_prison->pr_allow & flag) != 0);
2781 }
2782 
2783 /*
2784  * Hold a prison reference, by incrementing pr_ref.  It is generally
2785  * an error to hold a prison that does not already have a reference.
2786  * A prison record will remain valid as long as it has at least one
2787  * reference, and will not be removed as long as either the prison
2788  * mutex or the allprison lock is held (allprison_lock may be shared).
2789  */
2790 void
2791 prison_hold_locked(struct prison *pr)
2792 {
2793 
2794 	/* Locking is no longer required. */
2795 	prison_hold(pr);
2796 }
2797 
2798 void
2799 prison_hold(struct prison *pr)
2800 {
2801 #ifdef INVARIANTS
2802 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2803 
2804 	KASSERT(was_valid,
2805 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2806 #else
2807 	refcount_acquire(&pr->pr_ref);
2808 #endif
2809 }
2810 
2811 /*
2812  * Remove a prison reference.  If that was the last reference, the
2813  * prison will be removed (at a later time).
2814  */
2815 void
2816 prison_free_locked(struct prison *pr)
2817 {
2818 
2819 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2820 	/*
2821 	 * Locking is no longer required, but unlock because the caller
2822 	 * expects it.
2823 	 */
2824 	mtx_unlock(&pr->pr_mtx);
2825 	prison_free(pr);
2826 }
2827 
2828 void
2829 prison_free(struct prison *pr)
2830 {
2831 
2832 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2833 	    ("Trying to free dead prison %p (jid=%d).",
2834 	     pr, pr->pr_id));
2835 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2836 		/*
2837 		 * Don't remove the last reference in this context,
2838 		 * in case there are locks held.
2839 		 */
2840 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2841 	}
2842 }
2843 
2844 static void
2845 prison_free_not_last(struct prison *pr)
2846 {
2847 #ifdef INVARIANTS
2848 	int lastref;
2849 
2850 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2851 	    ("Trying to free dead prison %p (jid=%d).",
2852 	     pr, pr->pr_id));
2853 	lastref = refcount_release(&pr->pr_ref);
2854 	KASSERT(!lastref,
2855 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2856 	     pr, pr->pr_id));
2857 #else
2858 	refcount_release(&pr->pr_ref);
2859 #endif
2860 }
2861 
2862 /*
2863  * Hold a prison for user visibility, by incrementing pr_uref.
2864  * It is generally an error to hold a prison that isn't already
2865  * user-visible, except through the jail system calls.  It is also
2866  * an error to hold an invalid prison.  A prison record will remain
2867  * alive as long as it has at least one user reference, and will not
2868  * be set to the dying state until the prison mutex and allprison_lock
2869  * are both freed.
2870  */
2871 void
2872 prison_proc_hold(struct prison *pr)
2873 {
2874 #ifdef INVARIANTS
2875 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2876 
2877 	KASSERT(was_alive,
2878 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2879 #else
2880 	refcount_acquire(&pr->pr_uref);
2881 #endif
2882 }
2883 
2884 /*
2885  * Remove a prison user reference.  If it was the last reference, the
2886  * prison will be considered "dying", and may be removed once all of
2887  * its references are dropped.
2888  */
2889 void
2890 prison_proc_free(struct prison *pr)
2891 {
2892 
2893 	/*
2894 	 * Locking is only required when releasing the last reference.
2895 	 * This allows assurance that a locked prison will remain alive
2896 	 * until it is unlocked.
2897 	 */
2898 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2899 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2900 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2901 		/*
2902 		 * Don't remove the last user reference in this context,
2903 		 * which is expected to be a process that is not only locked,
2904 		 * but also half dead.  Add a reference so any calls to
2905 		 * prison_free() won't re-submit the task.
2906 		 */
2907 		prison_hold(pr);
2908 		mtx_lock(&pr->pr_mtx);
2909 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2910 		    ("Redundant last reference in prison_proc_free (jid=%d)",
2911 		     pr->pr_id));
2912 		pr->pr_flags |= PR_COMPLETE_PROC;
2913 		mtx_unlock(&pr->pr_mtx);
2914 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2915 	}
2916 }
2917 
2918 static void
2919 prison_proc_free_not_last(struct prison *pr)
2920 {
2921 #ifdef INVARIANTS
2922 	int lastref;
2923 
2924 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2925 	    ("Trying to free dead prison %p (jid=%d).",
2926 	     pr, pr->pr_id));
2927 	lastref = refcount_release(&pr->pr_uref);
2928 	KASSERT(!lastref,
2929 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2930 	     pr, pr->pr_id));
2931 #else
2932 	refcount_release(&pr->pr_uref);
2933 #endif
2934 }
2935 
2936 void
2937 prison_proc_link(struct prison *pr, struct proc *p)
2938 {
2939 
2940 	sx_assert(&allproc_lock, SA_XLOCKED);
2941 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
2942 }
2943 
2944 void
2945 prison_proc_unlink(struct prison *pr, struct proc *p)
2946 {
2947 
2948 	sx_assert(&allproc_lock, SA_XLOCKED);
2949 	LIST_REMOVE(p, p_jaillist);
2950 }
2951 
2952 static void
2953 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
2954 {
2955 
2956 	sx_xlock(&allproc_lock);
2957 	prison_proc_unlink(opr, p);
2958 	prison_proc_link(npr, p);
2959 	sx_xunlock(&allproc_lock);
2960 }
2961 
2962 /*
2963  * Complete a call to either prison_free or prison_proc_free.
2964  */
2965 static void
2966 prison_complete(void *context, int pending)
2967 {
2968 	struct prison *pr = context;
2969 	int drflags;
2970 
2971 	/*
2972 	 * This could be called to release the last reference, or the last
2973 	 * user reference (plus the reference held in prison_proc_free).
2974 	 */
2975 	drflags = prison_lock_xlock(pr, PD_DEREF);
2976 	if (pr->pr_flags & PR_COMPLETE_PROC) {
2977 		pr->pr_flags &= ~PR_COMPLETE_PROC;
2978 		drflags |= PD_DEUREF;
2979 	}
2980 	prison_deref(pr, drflags);
2981 }
2982 
2983 static void
2984 prison_kill_processes_cb(struct proc *p, void *arg __unused)
2985 {
2986 
2987 	kern_psignal(p, SIGKILL);
2988 }
2989 
2990 /*
2991  * Note the iteration does not guarantee acting on all processes.
2992  * Most notably there may be fork or jail_attach in progress.
2993  */
2994 void
2995 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
2996     void *cbarg)
2997 {
2998 	struct prison *ppr;
2999 	struct proc *p;
3000 
3001 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3002 		sx_slock(&allproc_lock);
3003 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3004 			if (p->p_state == PRS_NEW)
3005 				continue;
3006 			PROC_LOCK(p);
3007 			cb(p, cbarg);
3008 			PROC_UNLOCK(p);
3009 		}
3010 		sx_sunlock(&allproc_lock);
3011 		if (atomic_load_int(&pr->pr_childcount) == 0)
3012 			return;
3013 		/*
3014 		 * Some jails popped up during the iteration, fall through to a
3015 		 * system-wide search.
3016 		 */
3017 	}
3018 
3019 	sx_slock(&allproc_lock);
3020 	FOREACH_PROC_IN_SYSTEM(p) {
3021 		PROC_LOCK(p);
3022 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3023 			for (ppr = p->p_ucred->cr_prison;
3024 			    ppr != &prison0;
3025 			    ppr = ppr->pr_parent) {
3026 				if (ppr == pr) {
3027 					cb(p, cbarg);
3028 					break;
3029 				}
3030 			}
3031 		}
3032 		PROC_UNLOCK(p);
3033 	}
3034 	sx_sunlock(&allproc_lock);
3035 }
3036 
3037 /*
3038  * Remove a prison reference and/or user reference (usually).
3039  * This assumes context that allows sleeping (for allprison_lock),
3040  * with no non-sleeping locks held, except perhaps the prison itself.
3041  * If there are no more references, release and delist the prison.
3042  * On completion, the prison lock and the allprison lock are both
3043  * unlocked.
3044  */
3045 static void
3046 prison_deref(struct prison *pr, int flags)
3047 {
3048 	struct prisonlist freeprison;
3049 	struct prison *killpr, *rpr, *ppr, *tpr;
3050 
3051 	killpr = NULL;
3052 	TAILQ_INIT(&freeprison);
3053 	/*
3054 	 * Release this prison as requested, which may cause its parent
3055 	 * to be released, and then maybe its grandparent, etc.
3056 	 */
3057 	for (;;) {
3058 		if (flags & PD_KILL) {
3059 			/* Kill the prison and its descendents. */
3060 			KASSERT(pr != &prison0,
3061 			    ("prison_deref trying to kill prison0"));
3062 			if (!(flags & PD_DEREF)) {
3063 				prison_hold(pr);
3064 				flags |= PD_DEREF;
3065 			}
3066 			flags = prison_lock_xlock(pr, flags);
3067 			prison_deref_kill(pr, &freeprison);
3068 		}
3069 		if (flags & PD_DEUREF) {
3070 			/* Drop a user reference. */
3071 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3072 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3073 			     pr->pr_id));
3074 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3075 				if (!(flags & PD_DEREF)) {
3076 					prison_hold(pr);
3077 					flags |= PD_DEREF;
3078 				}
3079 				flags = prison_lock_xlock(pr, flags);
3080 				if (refcount_release(&pr->pr_uref) &&
3081 				    pr->pr_state == PRISON_STATE_ALIVE) {
3082 					/*
3083 					 * When the last user references goes,
3084 					 * this becomes a dying prison.
3085 					 */
3086 					KASSERT(
3087 					    refcount_load(&prison0.pr_uref) > 0,
3088 					    ("prison0 pr_uref=0"));
3089 					pr->pr_state = PRISON_STATE_DYING;
3090 					mtx_unlock(&pr->pr_mtx);
3091 					flags &= ~PD_LOCKED;
3092 					prison_cleanup(pr);
3093 				}
3094 			}
3095 		}
3096 		if (flags & PD_KILL) {
3097 			/*
3098 			 * Any remaining user references are probably processes
3099 			 * that need to be killed, either in this prison or its
3100 			 * descendants.
3101 			 */
3102 			if (refcount_load(&pr->pr_uref) > 0)
3103 				killpr = pr;
3104 			/* Make sure the parent prison doesn't get killed. */
3105 			flags &= ~PD_KILL;
3106 		}
3107 		if (flags & PD_DEREF) {
3108 			/* Drop a reference. */
3109 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3110 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3111 			     pr->pr_id));
3112 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3113 				flags = prison_lock_xlock(pr, flags);
3114 				if (refcount_release(&pr->pr_ref)) {
3115 					/*
3116 					 * When the last reference goes,
3117 					 * unlink the prison and set it aside.
3118 					 */
3119 					KASSERT(
3120 					    refcount_load(&pr->pr_uref) == 0,
3121 					    ("prison_deref: last ref, "
3122 					     "but still has %d urefs (jid=%d)",
3123 					     pr->pr_uref, pr->pr_id));
3124 					KASSERT(
3125 					    refcount_load(&prison0.pr_ref) != 0,
3126 					    ("prison0 pr_ref=0"));
3127 					pr->pr_state = PRISON_STATE_INVALID;
3128 					TAILQ_REMOVE(&allprison, pr, pr_list);
3129 					LIST_REMOVE(pr, pr_sibling);
3130 					TAILQ_INSERT_TAIL(&freeprison, pr,
3131 					    pr_list);
3132 					for (ppr = pr->pr_parent;
3133 					     ppr != NULL;
3134 					     ppr = ppr->pr_parent)
3135 						ppr->pr_childcount--;
3136 					/*
3137 					 * Removing a prison frees references
3138 					 * from its parent.
3139 					 */
3140 					mtx_unlock(&pr->pr_mtx);
3141 					flags &= ~PD_LOCKED;
3142 					pr = pr->pr_parent;
3143 					flags |= PD_DEREF | PD_DEUREF;
3144 					continue;
3145 				}
3146 			}
3147 		}
3148 		break;
3149 	}
3150 
3151 	/* Release all the prison locks. */
3152 	if (flags & PD_LOCKED)
3153 		mtx_unlock(&pr->pr_mtx);
3154 	if (flags & PD_LIST_SLOCKED)
3155 		sx_sunlock(&allprison_lock);
3156 	else if (flags & PD_LIST_XLOCKED)
3157 		sx_xunlock(&allprison_lock);
3158 
3159 	/* Kill any processes attached to a killed prison. */
3160 	if (killpr != NULL)
3161 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3162 
3163 	/*
3164 	 * Finish removing any unreferenced prisons, which couldn't happen
3165 	 * while allprison_lock was held (to avoid a LOR on vrele).
3166 	 */
3167 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3168 #ifdef VIMAGE
3169 		if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3170 			vnet_destroy(rpr->pr_vnet);
3171 #endif
3172 		if (rpr->pr_root != NULL)
3173 			vrele(rpr->pr_root);
3174 		mtx_destroy(&rpr->pr_mtx);
3175 #ifdef INET
3176 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3177 #endif
3178 #ifdef INET6
3179 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3180 #endif
3181 		if (rpr->pr_cpuset != NULL)
3182 			cpuset_rel(rpr->pr_cpuset);
3183 		osd_jail_exit(rpr);
3184 #ifdef RACCT
3185 		if (racct_enable)
3186 			prison_racct_detach(rpr);
3187 #endif
3188 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3189 		free(rpr, M_PRISON);
3190 	}
3191 }
3192 
3193 /*
3194  * Kill the prison and its descendants.  Mark them as dying, clear the
3195  * persist flag, and call module remove methods.
3196  */
3197 static void
3198 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3199 {
3200 	struct prison *cpr, *ppr, *rpr;
3201 	bool descend;
3202 
3203 	/*
3204 	 * Unlike the descendants, the target prison can be killed
3205 	 * even if it is currently dying.  This is useful for failed
3206 	 * creation in jail_set(2).
3207 	 */
3208 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3209 	    ("Trying to kill dead prison %p (jid=%d).",
3210 	     pr, pr->pr_id));
3211 	refcount_acquire(&pr->pr_uref);
3212 	pr->pr_state = PRISON_STATE_DYING;
3213 	mtx_unlock(&pr->pr_mtx);
3214 
3215 	rpr = NULL;
3216 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3217 		if (descend) {
3218 			if (!prison_isalive(cpr)) {
3219 				descend = false;
3220 				continue;
3221 			}
3222 			prison_hold(cpr);
3223 			prison_proc_hold(cpr);
3224 			mtx_lock(&cpr->pr_mtx);
3225 			cpr->pr_state = PRISON_STATE_DYING;
3226 			cpr->pr_flags |= PR_REMOVE;
3227 			mtx_unlock(&cpr->pr_mtx);
3228 			continue;
3229 		}
3230 		if (!(cpr->pr_flags & PR_REMOVE))
3231 			continue;
3232 		prison_cleanup(cpr);
3233 		mtx_lock(&cpr->pr_mtx);
3234 		cpr->pr_flags &= ~PR_REMOVE;
3235 		if (cpr->pr_flags & PR_PERSIST) {
3236 			cpr->pr_flags &= ~PR_PERSIST;
3237 			prison_proc_free_not_last(cpr);
3238 			prison_free_not_last(cpr);
3239 		}
3240 		(void)refcount_release(&cpr->pr_uref);
3241 		if (refcount_release(&cpr->pr_ref)) {
3242 			/*
3243 			 * When the last reference goes, unlink the prison
3244 			 * and set it aside for prison_deref() to handle.
3245 			 * Delay unlinking the sibling list to keep the loop
3246 			 * safe.
3247 			 */
3248 			if (rpr != NULL)
3249 				LIST_REMOVE(rpr, pr_sibling);
3250 			rpr = cpr;
3251 			rpr->pr_state = PRISON_STATE_INVALID;
3252 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3253 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3254 			/*
3255 			 * Removing a prison frees references from its parent.
3256 			 */
3257 			ppr = rpr->pr_parent;
3258 			prison_proc_free_not_last(ppr);
3259 			prison_free_not_last(ppr);
3260 			for (; ppr != NULL; ppr = ppr->pr_parent)
3261 				ppr->pr_childcount--;
3262 		}
3263 		mtx_unlock(&cpr->pr_mtx);
3264 	}
3265 	if (rpr != NULL)
3266 		LIST_REMOVE(rpr, pr_sibling);
3267 
3268 	prison_cleanup(pr);
3269 	mtx_lock(&pr->pr_mtx);
3270 	if (pr->pr_flags & PR_PERSIST) {
3271 		pr->pr_flags &= ~PR_PERSIST;
3272 		prison_proc_free_not_last(pr);
3273 		prison_free_not_last(pr);
3274 	}
3275 	(void)refcount_release(&pr->pr_uref);
3276 }
3277 
3278 /*
3279  * Given the current locking state in the flags, make sure allprison_lock
3280  * is held exclusive, and the prison is locked.  Return flags indicating
3281  * the new state.
3282  */
3283 static int
3284 prison_lock_xlock(struct prison *pr, int flags)
3285 {
3286 
3287 	if (!(flags & PD_LIST_XLOCKED)) {
3288 		/*
3289 		 * Get allprison_lock, which may be an upgrade,
3290 		 * and may require unlocking the prison.
3291 		 */
3292 		if (flags & PD_LOCKED) {
3293 			mtx_unlock(&pr->pr_mtx);
3294 			flags &= ~PD_LOCKED;
3295 		}
3296 		if (flags & PD_LIST_SLOCKED) {
3297 			if (!sx_try_upgrade(&allprison_lock)) {
3298 				sx_sunlock(&allprison_lock);
3299 				sx_xlock(&allprison_lock);
3300 			}
3301 			flags &= ~PD_LIST_SLOCKED;
3302 		} else
3303 			sx_xlock(&allprison_lock);
3304 		flags |= PD_LIST_XLOCKED;
3305 	}
3306 	if (!(flags & PD_LOCKED)) {
3307 		/* Lock the prison mutex. */
3308 		mtx_lock(&pr->pr_mtx);
3309 		flags |= PD_LOCKED;
3310 	}
3311 	return flags;
3312 }
3313 
3314 /*
3315  * Release a prison's resources when it starts dying (when the last user
3316  * reference is dropped, or when it is killed).
3317  */
3318 static void
3319 prison_cleanup(struct prison *pr)
3320 {
3321 	sx_assert(&allprison_lock, SA_XLOCKED);
3322 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3323 	shm_remove_prison(pr);
3324 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3325 }
3326 
3327 /*
3328  * Set or clear a permission bit in the pr_allow field, passing restrictions
3329  * (cleared permission) down to child jails.
3330  */
3331 void
3332 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3333 {
3334 	struct prison *pr;
3335 
3336 	pr = cred->cr_prison;
3337 	sx_slock(&allprison_lock);
3338 	mtx_lock(&pr->pr_mtx);
3339 	prison_set_allow_locked(pr, flag, enable);
3340 	mtx_unlock(&pr->pr_mtx);
3341 	sx_sunlock(&allprison_lock);
3342 }
3343 
3344 static void
3345 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3346 {
3347 	struct prison *cpr;
3348 	int descend;
3349 
3350 	if (enable != 0)
3351 		pr->pr_allow |= flag;
3352 	else {
3353 		pr->pr_allow &= ~flag;
3354 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3355 			cpr->pr_allow &= ~flag;
3356 	}
3357 }
3358 
3359 /*
3360  * Check if a jail supports the given address family.
3361  *
3362  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3363  * if not.
3364  */
3365 int
3366 prison_check_af(struct ucred *cred, int af)
3367 {
3368 	struct prison *pr;
3369 	int error;
3370 
3371 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3372 
3373 	pr = cred->cr_prison;
3374 #ifdef VIMAGE
3375 	/* Prisons with their own network stack are not limited. */
3376 	if (prison_owns_vnet(cred))
3377 		return (0);
3378 #endif
3379 
3380 	error = 0;
3381 	switch (af)
3382 	{
3383 #ifdef INET
3384 	case AF_INET:
3385 		if (pr->pr_flags & PR_IP4)
3386 		{
3387 			mtx_lock(&pr->pr_mtx);
3388 			if ((pr->pr_flags & PR_IP4) &&
3389 			    pr->pr_addrs[PR_INET] == NULL)
3390 				error = EAFNOSUPPORT;
3391 			mtx_unlock(&pr->pr_mtx);
3392 		}
3393 		break;
3394 #endif
3395 #ifdef INET6
3396 	case AF_INET6:
3397 		if (pr->pr_flags & PR_IP6)
3398 		{
3399 			mtx_lock(&pr->pr_mtx);
3400 			if ((pr->pr_flags & PR_IP6) &&
3401 			    pr->pr_addrs[PR_INET6] == NULL)
3402 				error = EAFNOSUPPORT;
3403 			mtx_unlock(&pr->pr_mtx);
3404 		}
3405 		break;
3406 #endif
3407 	case AF_LOCAL:
3408 	case AF_ROUTE:
3409 		break;
3410 	default:
3411 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3412 			error = EAFNOSUPPORT;
3413 	}
3414 	return (error);
3415 }
3416 
3417 /*
3418  * Check if given address belongs to the jail referenced by cred (wrapper to
3419  * prison_check_ip[46]).
3420  *
3421  * Returns 0 if jail doesn't restrict the address family or if address belongs
3422  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3423  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3424  */
3425 int
3426 prison_if(struct ucred *cred, const struct sockaddr *sa)
3427 {
3428 #ifdef INET
3429 	const struct sockaddr_in *sai;
3430 #endif
3431 #ifdef INET6
3432 	const struct sockaddr_in6 *sai6;
3433 #endif
3434 	int error;
3435 
3436 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3437 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3438 
3439 #ifdef VIMAGE
3440 	if (prison_owns_vnet(cred))
3441 		return (0);
3442 #endif
3443 
3444 	error = 0;
3445 	switch (sa->sa_family)
3446 	{
3447 #ifdef INET
3448 	case AF_INET:
3449 		sai = (const struct sockaddr_in *)sa;
3450 		error = prison_check_ip4(cred, &sai->sin_addr);
3451 		break;
3452 #endif
3453 #ifdef INET6
3454 	case AF_INET6:
3455 		sai6 = (const struct sockaddr_in6 *)sa;
3456 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3457 		break;
3458 #endif
3459 	default:
3460 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3461 			error = EAFNOSUPPORT;
3462 	}
3463 	return (error);
3464 }
3465 
3466 /*
3467  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3468  */
3469 int
3470 prison_check(struct ucred *cred1, struct ucred *cred2)
3471 {
3472 
3473 	return ((cred1->cr_prison == cred2->cr_prison ||
3474 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3475 }
3476 
3477 /*
3478  * For mountd/nfsd to run within a prison, it must be:
3479  * - A vnet prison.
3480  * - PR_ALLOW_NFSD must be set on it.
3481  * - The root directory (pr_root) of the prison must be
3482  *   a file system mount point, so the mountd can hang
3483  *   export information on it.
3484  */
3485 bool
3486 prison_check_nfsd(struct ucred *cred)
3487 {
3488 
3489 	if (jailed_without_vnet(cred))
3490 		return (false);
3491 	if (!prison_allow(cred, PR_ALLOW_NFSD))
3492 		return (false);
3493 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3494 		return (false);
3495 	return (true);
3496 }
3497 
3498 /*
3499  * Return 1 if p2 is a child of p1, otherwise 0.
3500  */
3501 int
3502 prison_ischild(struct prison *pr1, struct prison *pr2)
3503 {
3504 
3505 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3506 		if (pr1 == pr2)
3507 			return (1);
3508 	return (0);
3509 }
3510 
3511 /*
3512  * Return true if the prison is currently alive.  A prison is alive if it
3513  * holds user references and it isn't being removed.
3514  */
3515 bool
3516 prison_isalive(const struct prison *pr)
3517 {
3518 
3519 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3520 		return (false);
3521 	return (true);
3522 }
3523 
3524 /*
3525  * Return true if the prison is currently valid.  A prison is valid if it has
3526  * been fully created, and is not being destroyed.  Note that dying prisons
3527  * are still considered valid.  Invalid prisons won't be found under normal
3528  * circumstances, as they're only put in that state by functions that have
3529  * an exclusive hold on allprison_lock.
3530  */
3531 bool
3532 prison_isvalid(struct prison *pr)
3533 {
3534 
3535 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3536 		return (false);
3537 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3538 		return (false);
3539 	return (true);
3540 }
3541 
3542 /*
3543  * Return 1 if the passed credential is in a jail and that jail does not
3544  * have its own virtual network stack, otherwise 0.
3545  */
3546 int
3547 jailed_without_vnet(struct ucred *cred)
3548 {
3549 
3550 	if (!jailed(cred))
3551 		return (0);
3552 #ifdef VIMAGE
3553 	if (prison_owns_vnet(cred))
3554 		return (0);
3555 #endif
3556 
3557 	return (1);
3558 }
3559 
3560 /*
3561  * Return the correct hostname (domainname, et al) for the passed credential.
3562  */
3563 void
3564 getcredhostname(struct ucred *cred, char *buf, size_t size)
3565 {
3566 	struct prison *pr;
3567 
3568 	/*
3569 	 * A NULL credential can be used to shortcut to the physical
3570 	 * system's hostname.
3571 	 */
3572 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3573 	mtx_lock(&pr->pr_mtx);
3574 	strlcpy(buf, pr->pr_hostname, size);
3575 	mtx_unlock(&pr->pr_mtx);
3576 }
3577 
3578 void
3579 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3580 {
3581 
3582 	mtx_lock(&cred->cr_prison->pr_mtx);
3583 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3584 	mtx_unlock(&cred->cr_prison->pr_mtx);
3585 }
3586 
3587 void
3588 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3589 {
3590 
3591 	mtx_lock(&cred->cr_prison->pr_mtx);
3592 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3593 	mtx_unlock(&cred->cr_prison->pr_mtx);
3594 }
3595 
3596 void
3597 getcredhostid(struct ucred *cred, unsigned long *hostid)
3598 {
3599 
3600 	mtx_lock(&cred->cr_prison->pr_mtx);
3601 	*hostid = cred->cr_prison->pr_hostid;
3602 	mtx_unlock(&cred->cr_prison->pr_mtx);
3603 }
3604 
3605 void
3606 getjailname(struct ucred *cred, char *name, size_t len)
3607 {
3608 
3609 	mtx_lock(&cred->cr_prison->pr_mtx);
3610 	strlcpy(name, cred->cr_prison->pr_name, len);
3611 	mtx_unlock(&cred->cr_prison->pr_mtx);
3612 }
3613 
3614 #ifdef VIMAGE
3615 /*
3616  * Determine whether the prison represented by cred owns
3617  * its vnet rather than having it inherited.
3618  *
3619  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3620  */
3621 int
3622 prison_owns_vnet(struct ucred *cred)
3623 {
3624 
3625 	/*
3626 	 * vnets cannot be added/removed after jail creation,
3627 	 * so no need to lock here.
3628 	 */
3629 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3630 }
3631 #endif
3632 
3633 /*
3634  * Determine whether the subject represented by cred can "see"
3635  * status of a mount point.
3636  * Returns: 0 for permitted, ENOENT otherwise.
3637  * XXX: This function should be called cr_canseemount() and should be
3638  *      placed in kern_prot.c.
3639  */
3640 int
3641 prison_canseemount(struct ucred *cred, struct mount *mp)
3642 {
3643 	struct prison *pr;
3644 	struct statfs *sp;
3645 	size_t len;
3646 
3647 	pr = cred->cr_prison;
3648 	if (pr->pr_enforce_statfs == 0)
3649 		return (0);
3650 	if (pr->pr_root->v_mount == mp)
3651 		return (0);
3652 	if (pr->pr_enforce_statfs == 2)
3653 		return (ENOENT);
3654 	/*
3655 	 * If jail's chroot directory is set to "/" we should be able to see
3656 	 * all mount-points from inside a jail.
3657 	 * This is ugly check, but this is the only situation when jail's
3658 	 * directory ends with '/'.
3659 	 */
3660 	if (strcmp(pr->pr_path, "/") == 0)
3661 		return (0);
3662 	len = strlen(pr->pr_path);
3663 	sp = &mp->mnt_stat;
3664 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3665 		return (ENOENT);
3666 	/*
3667 	 * Be sure that we don't have situation where jail's root directory
3668 	 * is "/some/path" and mount point is "/some/pathpath".
3669 	 */
3670 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3671 		return (ENOENT);
3672 	return (0);
3673 }
3674 
3675 void
3676 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3677 {
3678 	char jpath[MAXPATHLEN];
3679 	struct prison *pr;
3680 	size_t len;
3681 
3682 	pr = cred->cr_prison;
3683 	if (pr->pr_enforce_statfs == 0)
3684 		return;
3685 	if (prison_canseemount(cred, mp) != 0) {
3686 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3687 		strlcpy(sp->f_mntonname, "[restricted]",
3688 		    sizeof(sp->f_mntonname));
3689 		return;
3690 	}
3691 	if (pr->pr_root->v_mount == mp) {
3692 		/*
3693 		 * Clear current buffer data, so we are sure nothing from
3694 		 * the valid path left there.
3695 		 */
3696 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3697 		*sp->f_mntonname = '/';
3698 		return;
3699 	}
3700 	/*
3701 	 * If jail's chroot directory is set to "/" we should be able to see
3702 	 * all mount-points from inside a jail.
3703 	 */
3704 	if (strcmp(pr->pr_path, "/") == 0)
3705 		return;
3706 	len = strlen(pr->pr_path);
3707 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3708 	/*
3709 	 * Clear current buffer data, so we are sure nothing from
3710 	 * the valid path left there.
3711 	 */
3712 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3713 	if (*jpath == '\0') {
3714 		/* Should never happen. */
3715 		*sp->f_mntonname = '/';
3716 	} else {
3717 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3718 	}
3719 }
3720 
3721 /*
3722  * Check with permission for a specific privilege is granted within jail.  We
3723  * have a specific list of accepted privileges; the rest are denied.
3724  */
3725 int
3726 prison_priv_check(struct ucred *cred, int priv)
3727 {
3728 	struct prison *pr;
3729 	int error;
3730 
3731 	/*
3732 	 * Some policies have custom handlers. This routine should not be
3733 	 * called for them. See priv_check_cred().
3734 	 */
3735 	switch (priv) {
3736 	case PRIV_VFS_LOOKUP:
3737 	case PRIV_VFS_GENERATION:
3738 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3739 		    "called for %d\n", priv));
3740 	}
3741 
3742 	if (!jailed(cred))
3743 		return (0);
3744 
3745 #ifdef VIMAGE
3746 	/*
3747 	 * Privileges specific to prisons with a virtual network stack.
3748 	 * There might be a duplicate entry here in case the privilege
3749 	 * is only granted conditionally in the legacy jail case.
3750 	 */
3751 	switch (priv) {
3752 		/*
3753 		 * NFS-specific privileges.
3754 		 */
3755 	case PRIV_NFS_DAEMON:
3756 	case PRIV_VFS_GETFH:
3757 	case PRIV_VFS_MOUNT_EXPORTED:
3758 #ifdef VNET_NFSD
3759 		if (!prison_check_nfsd(cred))
3760 #else
3761 		printf("running nfsd in a prison requires a kernel "
3762 		    "built with ''options VNET_NFSD''\n");
3763 #endif
3764 			return (EPERM);
3765 #ifdef notyet
3766 	case PRIV_NFS_LOCKD:
3767 #endif
3768 		/*
3769 		 * Network stack privileges.
3770 		 */
3771 	case PRIV_NET_BRIDGE:
3772 	case PRIV_NET_GRE:
3773 	case PRIV_NET_BPF:
3774 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3775 	case PRIV_NET_ROUTE:
3776 	case PRIV_NET_TAP:
3777 	case PRIV_NET_SETIFMTU:
3778 	case PRIV_NET_SETIFFLAGS:
3779 	case PRIV_NET_SETIFCAP:
3780 	case PRIV_NET_SETIFDESCR:
3781 	case PRIV_NET_SETIFNAME	:
3782 	case PRIV_NET_SETIFMETRIC:
3783 	case PRIV_NET_SETIFPHYS:
3784 	case PRIV_NET_SETIFMAC:
3785 	case PRIV_NET_SETLANPCP:
3786 	case PRIV_NET_ADDMULTI:
3787 	case PRIV_NET_DELMULTI:
3788 	case PRIV_NET_HWIOCTL:
3789 	case PRIV_NET_SETLLADDR:
3790 	case PRIV_NET_ADDIFGROUP:
3791 	case PRIV_NET_DELIFGROUP:
3792 	case PRIV_NET_IFCREATE:
3793 	case PRIV_NET_IFDESTROY:
3794 	case PRIV_NET_ADDIFADDR:
3795 	case PRIV_NET_DELIFADDR:
3796 	case PRIV_NET_LAGG:
3797 	case PRIV_NET_GIF:
3798 	case PRIV_NET_SETIFVNET:
3799 	case PRIV_NET_SETIFFIB:
3800 	case PRIV_NET_OVPN:
3801 	case PRIV_NET_ME:
3802 	case PRIV_NET_WG:
3803 
3804 		/*
3805 		 * 802.11-related privileges.
3806 		 */
3807 	case PRIV_NET80211_VAP_GETKEY:
3808 	case PRIV_NET80211_VAP_MANAGE:
3809 
3810 #ifdef notyet
3811 		/*
3812 		 * ATM privileges.
3813 		 */
3814 	case PRIV_NETATM_CFG:
3815 	case PRIV_NETATM_ADD:
3816 	case PRIV_NETATM_DEL:
3817 	case PRIV_NETATM_SET:
3818 
3819 		/*
3820 		 * Bluetooth privileges.
3821 		 */
3822 	case PRIV_NETBLUETOOTH_RAW:
3823 #endif
3824 
3825 		/*
3826 		 * Netgraph and netgraph module privileges.
3827 		 */
3828 	case PRIV_NETGRAPH_CONTROL:
3829 #ifdef notyet
3830 	case PRIV_NETGRAPH_TTY:
3831 #endif
3832 
3833 		/*
3834 		 * IPv4 and IPv6 privileges.
3835 		 */
3836 	case PRIV_NETINET_IPFW:
3837 	case PRIV_NETINET_DIVERT:
3838 	case PRIV_NETINET_PF:
3839 	case PRIV_NETINET_DUMMYNET:
3840 	case PRIV_NETINET_CARP:
3841 	case PRIV_NETINET_MROUTE:
3842 	case PRIV_NETINET_RAW:
3843 	case PRIV_NETINET_ADDRCTRL6:
3844 	case PRIV_NETINET_ND6:
3845 	case PRIV_NETINET_SCOPE6:
3846 	case PRIV_NETINET_ALIFETIME6:
3847 	case PRIV_NETINET_IPSEC:
3848 	case PRIV_NETINET_BINDANY:
3849 
3850 #ifdef notyet
3851 		/*
3852 		 * NCP privileges.
3853 		 */
3854 	case PRIV_NETNCP:
3855 
3856 		/*
3857 		 * SMB privileges.
3858 		 */
3859 	case PRIV_NETSMB:
3860 #endif
3861 
3862 	/*
3863 	 * No default: or deny here.
3864 	 * In case of no permit fall through to next switch().
3865 	 */
3866 		if (cred->cr_prison->pr_flags & PR_VNET)
3867 			return (0);
3868 	}
3869 #endif /* VIMAGE */
3870 
3871 	switch (priv) {
3872 		/*
3873 		 * Allow ktrace privileges for root in jail.
3874 		 */
3875 	case PRIV_KTRACE:
3876 
3877 #if 0
3878 		/*
3879 		 * Allow jailed processes to configure audit identity and
3880 		 * submit audit records (login, etc).  In the future we may
3881 		 * want to further refine the relationship between audit and
3882 		 * jail.
3883 		 */
3884 	case PRIV_AUDIT_GETAUDIT:
3885 	case PRIV_AUDIT_SETAUDIT:
3886 	case PRIV_AUDIT_SUBMIT:
3887 #endif
3888 
3889 		/*
3890 		 * Allow jailed processes to manipulate process UNIX
3891 		 * credentials in any way they see fit.
3892 		 */
3893 	case PRIV_CRED_SETUID:
3894 	case PRIV_CRED_SETEUID:
3895 	case PRIV_CRED_SETGID:
3896 	case PRIV_CRED_SETEGID:
3897 	case PRIV_CRED_SETGROUPS:
3898 	case PRIV_CRED_SETREUID:
3899 	case PRIV_CRED_SETREGID:
3900 	case PRIV_CRED_SETRESUID:
3901 	case PRIV_CRED_SETRESGID:
3902 
3903 		/*
3904 		 * Jail implements visibility constraints already, so allow
3905 		 * jailed root to override uid/gid-based constraints.
3906 		 */
3907 	case PRIV_SEEOTHERGIDS:
3908 	case PRIV_SEEOTHERUIDS:
3909 
3910 		/*
3911 		 * Jail implements inter-process debugging limits already, so
3912 		 * allow jailed root various debugging privileges.
3913 		 */
3914 	case PRIV_DEBUG_DIFFCRED:
3915 	case PRIV_DEBUG_SUGID:
3916 	case PRIV_DEBUG_UNPRIV:
3917 
3918 		/*
3919 		 * Allow jail to set various resource limits and login
3920 		 * properties, and for now, exceed process resource limits.
3921 		 */
3922 	case PRIV_PROC_LIMIT:
3923 	case PRIV_PROC_SETLOGIN:
3924 	case PRIV_PROC_SETRLIMIT:
3925 
3926 		/*
3927 		 * System V and POSIX IPC privileges are granted in jail.
3928 		 */
3929 	case PRIV_IPC_READ:
3930 	case PRIV_IPC_WRITE:
3931 	case PRIV_IPC_ADMIN:
3932 	case PRIV_IPC_MSGSIZE:
3933 	case PRIV_MQ_ADMIN:
3934 
3935 		/*
3936 		 * Jail operations within a jail work on child jails.
3937 		 */
3938 	case PRIV_JAIL_ATTACH:
3939 	case PRIV_JAIL_SET:
3940 	case PRIV_JAIL_REMOVE:
3941 
3942 		/*
3943 		 * Jail implements its own inter-process limits, so allow
3944 		 * root processes in jail to change scheduling on other
3945 		 * processes in the same jail.  Likewise for signalling.
3946 		 */
3947 	case PRIV_SCHED_DIFFCRED:
3948 	case PRIV_SCHED_CPUSET:
3949 	case PRIV_SIGNAL_DIFFCRED:
3950 	case PRIV_SIGNAL_SUGID:
3951 
3952 		/*
3953 		 * Allow jailed processes to write to sysctls marked as jail
3954 		 * writable.
3955 		 */
3956 	case PRIV_SYSCTL_WRITEJAIL:
3957 
3958 		/*
3959 		 * Allow root in jail to manage a variety of quota
3960 		 * properties.  These should likely be conditional on a
3961 		 * configuration option.
3962 		 */
3963 	case PRIV_VFS_GETQUOTA:
3964 	case PRIV_VFS_SETQUOTA:
3965 
3966 		/*
3967 		 * Since Jail relies on chroot() to implement file system
3968 		 * protections, grant many VFS privileges to root in jail.
3969 		 * Be careful to exclude mount-related and NFS-related
3970 		 * privileges.
3971 		 */
3972 	case PRIV_VFS_READ:
3973 	case PRIV_VFS_WRITE:
3974 	case PRIV_VFS_ADMIN:
3975 	case PRIV_VFS_EXEC:
3976 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3977 	case PRIV_VFS_CHFLAGS_DEV:
3978 	case PRIV_VFS_CHOWN:
3979 	case PRIV_VFS_CHROOT:
3980 	case PRIV_VFS_RETAINSUGID:
3981 	case PRIV_VFS_FCHROOT:
3982 	case PRIV_VFS_LINK:
3983 	case PRIV_VFS_SETGID:
3984 	case PRIV_VFS_STAT:
3985 	case PRIV_VFS_STICKYFILE:
3986 
3987 		/*
3988 		 * As in the non-jail case, non-root users are expected to be
3989 		 * able to read kernel/physical memory (provided /dev/[k]mem
3990 		 * exists in the jail and they have permission to access it).
3991 		 */
3992 	case PRIV_KMEM_READ:
3993 		return (0);
3994 
3995 		/*
3996 		 * Depending on the global setting, allow privilege of
3997 		 * setting system flags.
3998 		 */
3999 	case PRIV_VFS_SYSFLAGS:
4000 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4001 			return (0);
4002 		else
4003 			return (EPERM);
4004 
4005 		/*
4006 		 * Depending on the global setting, allow privilege of
4007 		 * mounting/unmounting file systems.
4008 		 */
4009 	case PRIV_VFS_MOUNT:
4010 	case PRIV_VFS_UNMOUNT:
4011 	case PRIV_VFS_MOUNT_NONUSER:
4012 	case PRIV_VFS_MOUNT_OWNER:
4013 		pr = cred->cr_prison;
4014 		prison_lock(pr);
4015 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4016 			error = 0;
4017 		else
4018 			error = EPERM;
4019 		prison_unlock(pr);
4020 		return (error);
4021 
4022 		/*
4023 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4024 		 * policy.  priv_check_cred will not specifically allow it, and
4025 		 * we may want a MAC policy to allow it.
4026 		 */
4027 	case PRIV_VFS_READ_DIR:
4028 		return (0);
4029 
4030 		/*
4031 		 * Conditionnaly allow locking (unlocking) physical pages
4032 		 * in memory.
4033 		 */
4034 	case PRIV_VM_MLOCK:
4035 	case PRIV_VM_MUNLOCK:
4036 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4037 			return (0);
4038 		else
4039 			return (EPERM);
4040 
4041 		/*
4042 		 * Conditionally allow jailed root to bind reserved ports.
4043 		 */
4044 	case PRIV_NETINET_RESERVEDPORT:
4045 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4046 			return (0);
4047 		else
4048 			return (EPERM);
4049 
4050 		/*
4051 		 * Allow jailed root to reuse in-use ports.
4052 		 */
4053 	case PRIV_NETINET_REUSEPORT:
4054 		return (0);
4055 
4056 		/*
4057 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4058 		 */
4059 	case PRIV_NETINET_SETHDROPTS:
4060 		return (0);
4061 
4062 		/*
4063 		 * Conditionally allow creating raw sockets in jail.
4064 		 */
4065 	case PRIV_NETINET_RAW:
4066 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4067 			return (0);
4068 		else
4069 			return (EPERM);
4070 
4071 		/*
4072 		 * Since jail implements its own visibility limits on netstat
4073 		 * sysctls, allow getcred.  This allows identd to work in
4074 		 * jail.
4075 		 */
4076 	case PRIV_NETINET_GETCRED:
4077 		return (0);
4078 
4079 		/*
4080 		 * Allow jailed root to set loginclass.
4081 		 */
4082 	case PRIV_PROC_SETLOGINCLASS:
4083 		return (0);
4084 
4085 		/*
4086 		 * Do not allow a process inside a jail to read the kernel
4087 		 * message buffer unless explicitly permitted.
4088 		 */
4089 	case PRIV_MSGBUF:
4090 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4091 			return (0);
4092 		return (EPERM);
4093 
4094 	default:
4095 		/*
4096 		 * In all remaining cases, deny the privilege request.  This
4097 		 * includes almost all network privileges, many system
4098 		 * configuration privileges.
4099 		 */
4100 		return (EPERM);
4101 	}
4102 }
4103 
4104 /*
4105  * Return the part of pr2's name that is relative to pr1, or the whole name
4106  * if it does not directly follow.
4107  */
4108 
4109 char *
4110 prison_name(struct prison *pr1, struct prison *pr2)
4111 {
4112 	char *name;
4113 
4114 	/* Jails see themselves as "0" (if they see themselves at all). */
4115 	if (pr1 == pr2)
4116 		return "0";
4117 	name = pr2->pr_name;
4118 	if (prison_ischild(pr1, pr2)) {
4119 		/*
4120 		 * pr1 isn't locked (and allprison_lock may not be either)
4121 		 * so its length can't be counted on.  But the number of dots
4122 		 * can be counted on - and counted.
4123 		 */
4124 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4125 			name = strchr(name, '.') + 1;
4126 	}
4127 	return (name);
4128 }
4129 
4130 /*
4131  * Return the part of pr2's path that is relative to pr1, or the whole path
4132  * if it does not directly follow.
4133  */
4134 static char *
4135 prison_path(struct prison *pr1, struct prison *pr2)
4136 {
4137 	char *path1, *path2;
4138 	int len1;
4139 
4140 	path1 = pr1->pr_path;
4141 	path2 = pr2->pr_path;
4142 	if (!strcmp(path1, "/"))
4143 		return (path2);
4144 	len1 = strlen(path1);
4145 	if (strncmp(path1, path2, len1))
4146 		return (path2);
4147 	if (path2[len1] == '\0')
4148 		return "/";
4149 	if (path2[len1] == '/')
4150 		return (path2 + len1);
4151 	return (path2);
4152 }
4153 
4154 /*
4155  * Jail-related sysctls.
4156  */
4157 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4158     "Jails");
4159 
4160 #if defined(INET) || defined(INET6)
4161 /*
4162  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4163  * sysctl_jail_list() helper.
4164  */
4165 static void
4166 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4167 {
4168 	const size_t size = pr_families[af].size;
4169 
4170  again:
4171 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4172 	if (pr->pr_addrs[af] != NULL) {
4173 		if (*len < pr->pr_addrs[af]->ips) {
4174 			*len = pr->pr_addrs[af]->ips;
4175 			mtx_unlock(&pr->pr_mtx);
4176 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4177 			mtx_lock(&pr->pr_mtx);
4178 			goto again;
4179 		}
4180 		bcopy(pr->pr_addrs[af] + 1, *out, pr->pr_addrs[af]->ips * size);
4181 	}
4182 }
4183 #endif
4184 
4185 static int
4186 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4187 {
4188 	struct xprison *xp;
4189 	struct prison *pr, *cpr;
4190 #ifdef INET
4191 	struct in_addr *ip4 = NULL;
4192 	int ip4s = 0;
4193 #endif
4194 #ifdef INET6
4195 	struct in6_addr *ip6 = NULL;
4196 	int ip6s = 0;
4197 #endif
4198 	int descend, error;
4199 
4200 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4201 	pr = req->td->td_ucred->cr_prison;
4202 	error = 0;
4203 	sx_slock(&allprison_lock);
4204 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4205 		mtx_lock(&cpr->pr_mtx);
4206 #ifdef INET
4207 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4208 #endif
4209 #ifdef INET6
4210 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4211 #endif
4212 		bzero(xp, sizeof(*xp));
4213 		xp->pr_version = XPRISON_VERSION;
4214 		xp->pr_id = cpr->pr_id;
4215 		xp->pr_state = cpr->pr_state;
4216 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4217 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4218 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4219 #ifdef INET
4220 		xp->pr_ip4s = ip4s;
4221 #endif
4222 #ifdef INET6
4223 		xp->pr_ip6s = ip6s;
4224 #endif
4225 		mtx_unlock(&cpr->pr_mtx);
4226 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4227 		if (error)
4228 			break;
4229 #ifdef INET
4230 		if (xp->pr_ip4s > 0) {
4231 			error = SYSCTL_OUT(req, ip4,
4232 			    xp->pr_ip4s * sizeof(struct in_addr));
4233 			if (error)
4234 				break;
4235 		}
4236 #endif
4237 #ifdef INET6
4238 		if (xp->pr_ip6s > 0) {
4239 			error = SYSCTL_OUT(req, ip6,
4240 			    xp->pr_ip6s * sizeof(struct in6_addr));
4241 			if (error)
4242 				break;
4243 		}
4244 #endif
4245 	}
4246 	sx_sunlock(&allprison_lock);
4247 	free(xp, M_TEMP);
4248 #ifdef INET
4249 	free(ip4, M_TEMP);
4250 #endif
4251 #ifdef INET6
4252 	free(ip6, M_TEMP);
4253 #endif
4254 	return (error);
4255 }
4256 
4257 SYSCTL_OID(_security_jail, OID_AUTO, list,
4258     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4259     sysctl_jail_list, "S", "List of active jails");
4260 
4261 static int
4262 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4263 {
4264 	int error, injail;
4265 
4266 	injail = jailed(req->td->td_ucred);
4267 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4268 
4269 	return (error);
4270 }
4271 
4272 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4273     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4274     sysctl_jail_jailed, "I", "Process in jail?");
4275 
4276 static int
4277 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4278 {
4279 	int error, havevnet;
4280 #ifdef VIMAGE
4281 	struct ucred *cred = req->td->td_ucred;
4282 
4283 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4284 #else
4285 	havevnet = 0;
4286 #endif
4287 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4288 
4289 	return (error);
4290 }
4291 
4292 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4293     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4294     sysctl_jail_vnet, "I", "Jail owns vnet?");
4295 
4296 #if defined(INET) || defined(INET6)
4297 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4298     &jail_max_af_ips, 0,
4299     "Number of IP addresses a jail may have at most per address family (deprecated)");
4300 #endif
4301 
4302 /*
4303  * Default parameters for jail(2) compatibility.  For historical reasons,
4304  * the sysctl names have varying similarity to the parameter names.  Prisons
4305  * just see their own parameters, and can't change them.
4306  */
4307 static int
4308 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4309 {
4310 	int error, i;
4311 
4312 	/* Get the current flag value, and convert it to a boolean. */
4313 	if (req->td->td_ucred->cr_prison == &prison0) {
4314 		mtx_lock(&prison0.pr_mtx);
4315 		i = (jail_default_allow & arg2) != 0;
4316 		mtx_unlock(&prison0.pr_mtx);
4317 	} else
4318 		i = prison_allow(req->td->td_ucred, arg2);
4319 
4320 	if (arg1 != NULL)
4321 		i = !i;
4322 	error = sysctl_handle_int(oidp, &i, 0, req);
4323 	if (error || !req->newptr)
4324 		return (error);
4325 	i = i ? arg2 : 0;
4326 	if (arg1 != NULL)
4327 		i ^= arg2;
4328 	/*
4329 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4330 	 * for writing.
4331 	 */
4332 	mtx_lock(&prison0.pr_mtx);
4333 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4334 	mtx_unlock(&prison0.pr_mtx);
4335 	return (0);
4336 }
4337 
4338 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4339     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4340     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4341     "Processes in jail can set their hostnames (deprecated)");
4342 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4343     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4344     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4345     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4346 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4347     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4348     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4349     "Processes in jail can use System V IPC primitives (deprecated)");
4350 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4351     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4352     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4353     "Prison root can create raw sockets (deprecated)");
4354 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4355     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4356     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4357     "Processes in jail can alter system file flags (deprecated)");
4358 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4359     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4360     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4361     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4362 
4363 static int
4364 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4365 {
4366 	struct prison *pr;
4367 	int level, error;
4368 
4369 	pr = req->td->td_ucred->cr_prison;
4370 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4371 	error = sysctl_handle_int(oidp, &level, 0, req);
4372 	if (error || !req->newptr)
4373 		return (error);
4374 	*(int *)arg1 = level;
4375 	return (0);
4376 }
4377 
4378 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4379     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4380     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4381     sysctl_jail_default_level, "I",
4382     "Processes in jail cannot see all mounted file systems (deprecated)");
4383 
4384 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4385     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4386     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4387     sysctl_jail_default_level, "I",
4388     "Ruleset for the devfs filesystem in jail (deprecated)");
4389 
4390 /*
4391  * Nodes to describe jail parameters.  Maximum length of string parameters
4392  * is returned in the string itself, and the other parameters exist merely
4393  * to make themselves and their types known.
4394  */
4395 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4396     "Jail parameters");
4397 
4398 int
4399 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4400 {
4401 	int i;
4402 	long l;
4403 	size_t s;
4404 	char numbuf[12];
4405 
4406 	switch (oidp->oid_kind & CTLTYPE)
4407 	{
4408 	case CTLTYPE_LONG:
4409 	case CTLTYPE_ULONG:
4410 		l = 0;
4411 #ifdef SCTL_MASK32
4412 		if (!(req->flags & SCTL_MASK32))
4413 #endif
4414 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4415 	case CTLTYPE_INT:
4416 	case CTLTYPE_UINT:
4417 		i = 0;
4418 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4419 	case CTLTYPE_STRING:
4420 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4421 		return
4422 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4423 	case CTLTYPE_STRUCT:
4424 		s = (size_t)arg2;
4425 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4426 	}
4427 	return (0);
4428 }
4429 
4430 /*
4431  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4432  * jail creation time but cannot be changed in an existing jail.
4433  */
4434 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4435 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4436 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4437 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4438 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4439     "I", "Jail secure level");
4440 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4441     "Jail value for kern.osreldate and uname -K");
4442 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4443     "Jail value for kern.osrelease and uname -r");
4444 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4445     "I", "Jail cannot see all mounted file systems");
4446 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4447     "I", "Ruleset for in-jail devfs mounts");
4448 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4449     "B", "Jail persistence");
4450 #ifdef VIMAGE
4451 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4452     "E,jailsys", "Virtual network stack");
4453 #endif
4454 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4455     "B", "Jail is in the process of shutting down");
4456 
4457 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4458 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4459     "I", "Current number of child jails");
4460 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4461     "I", "Maximum number of child jails");
4462 
4463 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4464 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4465     "Jail hostname");
4466 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4467     "Jail NIS domainname");
4468 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4469     "Jail host UUID");
4470 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4471     "LU", "Jail host ID");
4472 
4473 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4474 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4475 
4476 #ifdef INET
4477 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4478     "Jail IPv4 address virtualization");
4479 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4480     "S,in_addr,a", "Jail IPv4 addresses");
4481 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4482     "B", "Do (not) use IPv4 source address selection rather than the "
4483     "primary jail IPv4 address.");
4484 #endif
4485 #ifdef INET6
4486 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4487     "Jail IPv6 address virtualization");
4488 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4489     "S,in6_addr,a", "Jail IPv6 addresses");
4490 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4491     "B", "Do (not) use IPv6 source address selection rather than the "
4492     "primary jail IPv6 address.");
4493 #endif
4494 
4495 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4496 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4497     "B", "Jail may set hostname");
4498 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4499     "B", "Jail may use SYSV IPC");
4500 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4501     "B", "Jail may create raw sockets");
4502 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4503     "B", "Jail may alter system file flags");
4504 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4505     "B", "Jail may set file quotas");
4506 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4507     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4508 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4509     "B", "Jail may lock (unlock) physical pages in memory");
4510 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4511     "B", "Jail may bind sockets to reserved ports");
4512 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4513     "B", "Jail may read the kernel message buffer");
4514 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4515     "B", "Unprivileged processes may use process debugging facilities");
4516 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4517     "B", "Processes in jail with uid 0 have privilege");
4518 #if defined(VNET_NFSD) && defined(VIMAGE) && defined(NFSD)
4519 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4520     "B", "Mountd/nfsd may run in the jail");
4521 #endif
4522 
4523 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4524 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4525     "B", "Jail may mount/unmount jail-friendly file systems in general");
4526 
4527 /*
4528  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4529  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4530  * not created.
4531  */
4532 unsigned
4533 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4534     const char *descr)
4535 {
4536 	struct bool_flags *bf;
4537 	struct sysctl_oid *parent;
4538 	char *allow_name, *allow_noname, *allowed;
4539 #ifndef NO_SYSCTL_DESCR
4540 	char *descr_deprecated;
4541 #endif
4542 	u_int allow_flag;
4543 
4544 	if (prefix
4545 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4546 		< 0 ||
4547 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4548 		< 0
4549 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4550 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4551 		free(allow_name, M_PRISON);
4552 		return 0;
4553 	}
4554 
4555 	/*
4556 	 * See if this parameter has already beed added, i.e. a module was
4557 	 * previously loaded/unloaded.
4558 	 */
4559 	mtx_lock(&prison0.pr_mtx);
4560 	for (bf = pr_flag_allow;
4561 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4562 		atomic_load_int(&bf->flag) != 0;
4563 	     bf++) {
4564 		if (strcmp(bf->name, allow_name) == 0) {
4565 			allow_flag = bf->flag;
4566 			goto no_add;
4567 		}
4568 	}
4569 
4570 	/*
4571 	 * Find a free bit in pr_allow_all, failing if there are none
4572 	 * (which shouldn't happen as long as we keep track of how many
4573 	 * potential dynamic flags exist).
4574 	 */
4575 	for (allow_flag = 1;; allow_flag <<= 1) {
4576 		if (allow_flag == 0)
4577 			goto no_add;
4578 		if ((pr_allow_all & allow_flag) == 0)
4579 			break;
4580 	}
4581 
4582 	/* Note the parameter in the next open slot in pr_flag_allow. */
4583 	for (bf = pr_flag_allow; ; bf++) {
4584 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4585 			/* This should never happen, but is not fatal. */
4586 			allow_flag = 0;
4587 			goto no_add;
4588 		}
4589 		if (atomic_load_int(&bf->flag) == 0)
4590 			break;
4591 	}
4592 	bf->name = allow_name;
4593 	bf->noname = allow_noname;
4594 	pr_allow_all |= allow_flag;
4595 	/*
4596 	 * prison0 always has permission for the new parameter.
4597 	 * Other jails must have it granted to them.
4598 	 */
4599 	prison0.pr_allow |= allow_flag;
4600 	/* The flag indicates a valid entry, so make sure it is set last. */
4601 	atomic_store_rel_int(&bf->flag, allow_flag);
4602 	mtx_unlock(&prison0.pr_mtx);
4603 
4604 	/*
4605 	 * Create sysctls for the parameter, and the back-compat global
4606 	 * permission.
4607 	 */
4608 	parent = prefix
4609 	    ? SYSCTL_ADD_NODE(NULL,
4610 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4611 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4612 	    : &sysctl___security_jail_param_allow;
4613 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4614 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4615 	    NULL, 0, sysctl_jail_param, "B", descr);
4616 	if ((prefix
4617 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4618 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4619 #ifndef NO_SYSCTL_DESCR
4620 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4621 		    descr);
4622 #endif
4623 		(void)SYSCTL_ADD_PROC(NULL,
4624 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4625 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4626 		    sysctl_jail_default_allow, "I", descr_deprecated);
4627 #ifndef NO_SYSCTL_DESCR
4628 		free(descr_deprecated, M_TEMP);
4629 #endif
4630 		free(allowed, M_TEMP);
4631 	}
4632 	return allow_flag;
4633 
4634  no_add:
4635 	mtx_unlock(&prison0.pr_mtx);
4636 	free(allow_name, M_PRISON);
4637 	free(allow_noname, M_PRISON);
4638 	return allow_flag;
4639 }
4640 
4641 /*
4642  * The VFS system will register jail-aware filesystems here.  They each get
4643  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4644  * attempts to mount.
4645  */
4646 void
4647 prison_add_vfs(struct vfsconf *vfsp)
4648 {
4649 #ifdef NO_SYSCTL_DESCR
4650 
4651 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4652 	    NULL, NULL);
4653 #else
4654 	char *descr;
4655 
4656 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4657 	    vfsp->vfc_name);
4658 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4659 	    NULL, descr);
4660 	free(descr, M_TEMP);
4661 #endif
4662 }
4663 
4664 #ifdef RACCT
4665 void
4666 prison_racct_foreach(void (*callback)(struct racct *racct,
4667     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4668     void *arg2, void *arg3)
4669 {
4670 	struct prison_racct *prr;
4671 
4672 	ASSERT_RACCT_ENABLED();
4673 
4674 	sx_slock(&allprison_lock);
4675 	if (pre != NULL)
4676 		(pre)();
4677 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4678 		(callback)(prr->prr_racct, arg2, arg3);
4679 	if (post != NULL)
4680 		(post)();
4681 	sx_sunlock(&allprison_lock);
4682 }
4683 
4684 static struct prison_racct *
4685 prison_racct_find_locked(const char *name)
4686 {
4687 	struct prison_racct *prr;
4688 
4689 	ASSERT_RACCT_ENABLED();
4690 	sx_assert(&allprison_lock, SA_XLOCKED);
4691 
4692 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4693 		return (NULL);
4694 
4695 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4696 		if (strcmp(name, prr->prr_name) != 0)
4697 			continue;
4698 
4699 		/* Found prison_racct with a matching name? */
4700 		prison_racct_hold(prr);
4701 		return (prr);
4702 	}
4703 
4704 	/* Add new prison_racct. */
4705 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4706 	racct_create(&prr->prr_racct);
4707 
4708 	strcpy(prr->prr_name, name);
4709 	refcount_init(&prr->prr_refcount, 1);
4710 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4711 
4712 	return (prr);
4713 }
4714 
4715 struct prison_racct *
4716 prison_racct_find(const char *name)
4717 {
4718 	struct prison_racct *prr;
4719 
4720 	ASSERT_RACCT_ENABLED();
4721 
4722 	sx_xlock(&allprison_lock);
4723 	prr = prison_racct_find_locked(name);
4724 	sx_xunlock(&allprison_lock);
4725 	return (prr);
4726 }
4727 
4728 void
4729 prison_racct_hold(struct prison_racct *prr)
4730 {
4731 
4732 	ASSERT_RACCT_ENABLED();
4733 
4734 	refcount_acquire(&prr->prr_refcount);
4735 }
4736 
4737 static void
4738 prison_racct_free_locked(struct prison_racct *prr)
4739 {
4740 
4741 	ASSERT_RACCT_ENABLED();
4742 	sx_assert(&allprison_lock, SA_XLOCKED);
4743 
4744 	if (refcount_release(&prr->prr_refcount)) {
4745 		racct_destroy(&prr->prr_racct);
4746 		LIST_REMOVE(prr, prr_next);
4747 		free(prr, M_PRISON_RACCT);
4748 	}
4749 }
4750 
4751 void
4752 prison_racct_free(struct prison_racct *prr)
4753 {
4754 
4755 	ASSERT_RACCT_ENABLED();
4756 	sx_assert(&allprison_lock, SA_UNLOCKED);
4757 
4758 	if (refcount_release_if_not_last(&prr->prr_refcount))
4759 		return;
4760 
4761 	sx_xlock(&allprison_lock);
4762 	prison_racct_free_locked(prr);
4763 	sx_xunlock(&allprison_lock);
4764 }
4765 
4766 static void
4767 prison_racct_attach(struct prison *pr)
4768 {
4769 	struct prison_racct *prr;
4770 
4771 	ASSERT_RACCT_ENABLED();
4772 	sx_assert(&allprison_lock, SA_XLOCKED);
4773 
4774 	prr = prison_racct_find_locked(pr->pr_name);
4775 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4776 
4777 	pr->pr_prison_racct = prr;
4778 }
4779 
4780 /*
4781  * Handle jail renaming.  From the racct point of view, renaming means
4782  * moving from one prison_racct to another.
4783  */
4784 static void
4785 prison_racct_modify(struct prison *pr)
4786 {
4787 #ifdef RCTL
4788 	struct proc *p;
4789 	struct ucred *cred;
4790 #endif
4791 	struct prison_racct *oldprr;
4792 
4793 	ASSERT_RACCT_ENABLED();
4794 
4795 	sx_slock(&allproc_lock);
4796 	sx_xlock(&allprison_lock);
4797 
4798 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4799 		sx_xunlock(&allprison_lock);
4800 		sx_sunlock(&allproc_lock);
4801 		return;
4802 	}
4803 
4804 	oldprr = pr->pr_prison_racct;
4805 	pr->pr_prison_racct = NULL;
4806 
4807 	prison_racct_attach(pr);
4808 
4809 	/*
4810 	 * Move resource utilisation records.
4811 	 */
4812 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4813 
4814 #ifdef RCTL
4815 	/*
4816 	 * Force rctl to reattach rules to processes.
4817 	 */
4818 	FOREACH_PROC_IN_SYSTEM(p) {
4819 		PROC_LOCK(p);
4820 		cred = crhold(p->p_ucred);
4821 		PROC_UNLOCK(p);
4822 		rctl_proc_ucred_changed(p, cred);
4823 		crfree(cred);
4824 	}
4825 #endif
4826 
4827 	sx_sunlock(&allproc_lock);
4828 	prison_racct_free_locked(oldprr);
4829 	sx_xunlock(&allprison_lock);
4830 }
4831 
4832 static void
4833 prison_racct_detach(struct prison *pr)
4834 {
4835 
4836 	ASSERT_RACCT_ENABLED();
4837 	sx_assert(&allprison_lock, SA_UNLOCKED);
4838 
4839 	if (pr->pr_prison_racct == NULL)
4840 		return;
4841 	prison_racct_free(pr->pr_prison_racct);
4842 	pr->pr_prison_racct = NULL;
4843 }
4844 #endif /* RACCT */
4845 
4846 #ifdef DDB
4847 
4848 static void
4849 db_show_prison(struct prison *pr)
4850 {
4851 	struct bool_flags *bf;
4852 	struct jailsys_flags *jsf;
4853 #if defined(INET) || defined(INET6)
4854 	int ii;
4855 #endif
4856 	unsigned f;
4857 #ifdef INET
4858 	char ip4buf[INET_ADDRSTRLEN];
4859 #endif
4860 #ifdef INET6
4861 	char ip6buf[INET6_ADDRSTRLEN];
4862 #endif
4863 
4864 	db_printf("prison %p:\n", pr);
4865 	db_printf(" jid             = %d\n", pr->pr_id);
4866 	db_printf(" name            = %s\n", pr->pr_name);
4867 	db_printf(" parent          = %p\n", pr->pr_parent);
4868 	db_printf(" ref             = %d\n", pr->pr_ref);
4869 	db_printf(" uref            = %d\n", pr->pr_uref);
4870 	db_printf(" state           = %s\n",
4871 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4872 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
4873 	    "invalid");
4874 	db_printf(" path            = %s\n", pr->pr_path);
4875 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4876 	    ? pr->pr_cpuset->cs_id : -1);
4877 #ifdef VIMAGE
4878 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4879 #endif
4880 	db_printf(" root            = %p\n", pr->pr_root);
4881 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4882 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4883 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4884 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4885 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4886 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4887 	db_printf(" flags           = 0x%x", pr->pr_flags);
4888 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4889 		if (pr->pr_flags & bf->flag)
4890 			db_printf(" %s", bf->name);
4891 	for (jsf = pr_flag_jailsys;
4892 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4893 	     jsf++) {
4894 		f = pr->pr_flags & (jsf->disable | jsf->new);
4895 		db_printf(" %-16s= %s\n", jsf->name,
4896 		    (f != 0 && f == jsf->disable) ? "disable"
4897 		    : (f == jsf->new) ? "new"
4898 		    : "inherit");
4899 	}
4900 	db_printf(" allow           = 0x%x", pr->pr_allow);
4901 	for (bf = pr_flag_allow;
4902 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4903 		atomic_load_int(&bf->flag) != 0;
4904 	     bf++)
4905 		if (pr->pr_allow & bf->flag)
4906 			db_printf(" %s", bf->name);
4907 	db_printf("\n");
4908 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4909 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4910 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4911 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4912 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4913 #ifdef INET
4914 	if (pr->pr_addrs[PR_INET] != NULL) {
4915 		pr_family_t af = PR_INET;
4916 
4917 		db_printf(" ip4s            = %d\n", pr->pr_addrs[af]->ips);
4918 		for (ii = 0; ii < pr->pr_addrs[af]->ips; ii++)
4919 			db_printf(" %s %s\n",
4920 			    ii == 0 ? "ip4.addr        =" : "                 ",
4921 			    inet_ntoa_r(
4922 			    *(const struct in_addr *)PR_IP(pr->pr_addrs[af], ii),
4923 			    ip4buf));
4924 	}
4925 #endif
4926 #ifdef INET6
4927 	if (pr->pr_addrs[PR_INET6] != NULL) {
4928 		pr_family_t af = PR_INET6;
4929 
4930 		db_printf(" ip6s            = %d\n", pr->pr_addrs[af]->ips);
4931 		for (ii = 0; ii < pr->pr_addrs[af]->ips; ii++)
4932 			db_printf(" %s %s\n",
4933 			    ii == 0 ? "ip6.addr        =" : "                 ",
4934 			    ip6_sprintf(ip6buf,
4935 			    (const struct in6_addr *)PR_IP(pr->pr_addrs[af], ii)));
4936 	}
4937 #endif
4938 }
4939 
4940 DB_SHOW_COMMAND(prison, db_show_prison_command)
4941 {
4942 	struct prison *pr;
4943 
4944 	if (!have_addr) {
4945 		/*
4946 		 * Show all prisons in the list, and prison0 which is not
4947 		 * listed.
4948 		 */
4949 		db_show_prison(&prison0);
4950 		if (!db_pager_quit) {
4951 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4952 				db_show_prison(pr);
4953 				if (db_pager_quit)
4954 					break;
4955 			}
4956 		}
4957 		return;
4958 	}
4959 
4960 	if (addr == 0)
4961 		pr = &prison0;
4962 	else {
4963 		/* Look for a prison with the ID and with references. */
4964 		TAILQ_FOREACH(pr, &allprison, pr_list)
4965 			if (pr->pr_id == addr && pr->pr_ref > 0)
4966 				break;
4967 		if (pr == NULL)
4968 			/* Look again, without requiring a reference. */
4969 			TAILQ_FOREACH(pr, &allprison, pr_list)
4970 				if (pr->pr_id == addr)
4971 					break;
4972 		if (pr == NULL)
4973 			/* Assume address points to a valid prison. */
4974 			pr = (struct prison *)addr;
4975 	}
4976 	db_show_prison(pr);
4977 }
4978 
4979 #endif /* DDB */
4980