xref: /freebsd/sys/kern/kern_jail.c (revision 43f8c763cdeea29f95b6f0eebce3ad80dd210c7a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/epoch.h>
49 #include <sys/taskqueue.h>
50 #include <sys/fcntl.h>
51 #include <sys/jail.h>
52 #include <sys/linker.h>
53 #include <sys/lock.h>
54 #include <sys/mman.h>
55 #include <sys/mutex.h>
56 #include <sys/racct.h>
57 #include <sys/rctl.h>
58 #include <sys/refcount.h>
59 #include <sys/sx.h>
60 #include <sys/sysent.h>
61 #include <sys/namei.h>
62 #include <sys/mount.h>
63 #include <sys/queue.h>
64 #include <sys/socket.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysctl.h>
67 #include <sys/uuid.h>
68 #include <sys/vnode.h>
69 
70 #include <net/if.h>
71 #include <net/vnet.h>
72 
73 #include <netinet/in.h>
74 
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #endif /* DDB */
78 
79 #include <security/mac/mac_framework.h>
80 
81 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
82 
83 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
84 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
85 
86 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
87 #ifdef INET
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
92 #endif
93 #else /* !INET */
94 #ifdef INET6
95 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
96 #else
97 #define	_PR_IP_SADDRSEL	0
98 #endif
99 #endif
100 
101 /* prison0 describes what is "real" about the system. */
102 struct prison prison0 = {
103 	.pr_id		= 0,
104 	.pr_name	= "0",
105 	.pr_ref		= 1,
106 	.pr_uref	= 1,
107 	.pr_path	= "/",
108 	.pr_securelevel	= -1,
109 	.pr_devfs_rsnum = 0,
110 	.pr_state	= PRISON_STATE_ALIVE,
111 	.pr_childmax	= JAIL_MAX,
112 	.pr_hostuuid	= DEFAULT_HOSTUUID,
113 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
114 #ifdef VIMAGE
115 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
116 #else
117 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
118 #endif
119 	.pr_allow	= PR_ALLOW_ALL_STATIC,
120 };
121 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
122 
123 struct bool_flags {
124 	const char	*name;
125 	const char	*noname;
126 	volatile u_int	 flag;
127 };
128 struct jailsys_flags {
129 	const char	*name;
130 	unsigned	 disable;
131 	unsigned	 new;
132 };
133 
134 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
135 struct	sx allprison_lock;
136 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
137 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
138 LIST_HEAD(, prison_racct) allprison_racct;
139 int	lastprid = 0;
140 
141 static int get_next_prid(struct prison **insprp);
142 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
143 static void prison_complete(void *context, int pending);
144 static void prison_deref(struct prison *pr, int flags);
145 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
146 static int prison_lock_xlock(struct prison *pr, int flags);
147 static void prison_cleanup(struct prison *pr);
148 static void prison_free_not_last(struct prison *pr);
149 static void prison_proc_free_not_last(struct prison *pr);
150 static void prison_proc_relink(struct prison *opr, struct prison *npr,
151     struct proc *p);
152 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
153     int enable);
154 static char *prison_path(struct prison *pr1, struct prison *pr2);
155 #ifdef RACCT
156 static void prison_racct_attach(struct prison *pr);
157 static void prison_racct_modify(struct prison *pr);
158 static void prison_racct_detach(struct prison *pr);
159 #endif
160 
161 /* Flags for prison_deref */
162 #define	PD_DEREF	0x01	/* Decrement pr_ref */
163 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
164 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
165 #define	PD_LOCKED	0x10	/* pr_mtx is held */
166 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
167 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
168 #define PD_OP_FLAGS	0x07	/* Operation flags */
169 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
170 
171 /*
172  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
173  * as we cannot figure out the size of a sparse array, or an array without a
174  * terminating entry.
175  */
176 static struct bool_flags pr_flag_bool[] = {
177 	{"persist", "nopersist", PR_PERSIST},
178 #ifdef INET
179 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
180 #endif
181 #ifdef INET6
182 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
183 #endif
184 };
185 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
186 
187 static struct jailsys_flags pr_flag_jailsys[] = {
188 	{"host", 0, PR_HOST},
189 #ifdef VIMAGE
190 	{"vnet", 0, PR_VNET},
191 #endif
192 #ifdef INET
193 	{"ip4", PR_IP4_USER, PR_IP4_USER},
194 #endif
195 #ifdef INET6
196 	{"ip6", PR_IP6_USER, PR_IP6_USER},
197 #endif
198 };
199 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
200 
201 /*
202  * Make this array full-size so dynamic parameters can be added.
203  * It is protected by prison0.mtx, but lockless reading is allowed
204  * with an atomic check of the flag values.
205  */
206 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
207 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
208 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
209 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
210 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
211 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
212 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
213 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
214 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
215 	{"allow.reserved_ports", "allow.noreserved_ports",
216 	 PR_ALLOW_RESERVED_PORTS},
217 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
218 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
219 	 PR_ALLOW_UNPRIV_DEBUG},
220 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
221 };
222 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
223 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
224 
225 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
226 					 PR_ALLOW_RESERVED_PORTS | \
227 					 PR_ALLOW_UNPRIV_DEBUG | \
228 					 PR_ALLOW_SUSER)
229 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
230 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
231 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
232 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
233 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
234 #if defined(INET) || defined(INET6)
235 static unsigned jail_max_af_ips = 255;
236 #endif
237 
238 /*
239  * Initialize the parts of prison0 that can't be static-initialized with
240  * constants.  This is called from proc0_init() after creating thread0 cpuset.
241  */
242 void
243 prison0_init(void)
244 {
245 	uint8_t *file, *data;
246 	size_t size;
247 	char buf[sizeof(prison0.pr_hostuuid)];
248 	bool valid;
249 
250 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
251 	prison0.pr_osreldate = osreldate;
252 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
253 
254 	/* If we have a preloaded hostuuid, use it. */
255 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
256 	if (file != NULL) {
257 		data = preload_fetch_addr(file);
258 		size = preload_fetch_size(file);
259 		if (data != NULL) {
260 			/*
261 			 * The preloaded data may include trailing whitespace, almost
262 			 * certainly a newline; skip over any whitespace or
263 			 * non-printable characters to be safe.
264 			 */
265 			while (size > 0 && data[size - 1] <= 0x20) {
266 				size--;
267 			}
268 
269 			valid = false;
270 
271 			/*
272 			 * Not NUL-terminated when passed from loader, but
273 			 * validate_uuid requires that due to using sscanf (as
274 			 * does the subsequent strlcpy, since it still reads
275 			 * past the given size to return the true length);
276 			 * bounce to a temporary buffer to fix.
277 			 */
278 			if (size >= sizeof(buf))
279 				goto done;
280 
281 			memcpy(buf, data, size);
282 			buf[size] = '\0';
283 
284 			if (validate_uuid(buf, size, NULL, 0) != 0)
285 				goto done;
286 
287 			valid = true;
288 			(void)strlcpy(prison0.pr_hostuuid, buf,
289 			    sizeof(prison0.pr_hostuuid));
290 
291 done:
292 			if (bootverbose && !valid) {
293 				printf("hostuuid: preload data malformed: '%.*s'\n",
294 				    (int)size, data);
295 			}
296 		}
297 	}
298 	if (bootverbose)
299 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
300 }
301 
302 /*
303  * struct jail_args {
304  *	struct jail *jail;
305  * };
306  */
307 int
308 sys_jail(struct thread *td, struct jail_args *uap)
309 {
310 	uint32_t version;
311 	int error;
312 	struct jail j;
313 
314 	error = copyin(uap->jail, &version, sizeof(uint32_t));
315 	if (error)
316 		return (error);
317 
318 	switch (version) {
319 	case 0:
320 	{
321 		struct jail_v0 j0;
322 
323 		/* FreeBSD single IPv4 jails. */
324 		bzero(&j, sizeof(struct jail));
325 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
326 		if (error)
327 			return (error);
328 		j.version = j0.version;
329 		j.path = j0.path;
330 		j.hostname = j0.hostname;
331 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
332 		break;
333 	}
334 
335 	case 1:
336 		/*
337 		 * Version 1 was used by multi-IPv4 jail implementations
338 		 * that never made it into the official kernel.
339 		 */
340 		return (EINVAL);
341 
342 	case 2:	/* JAIL_API_VERSION */
343 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
344 		error = copyin(uap->jail, &j, sizeof(struct jail));
345 		if (error)
346 			return (error);
347 		break;
348 
349 	default:
350 		/* Sci-Fi jails are not supported, sorry. */
351 		return (EINVAL);
352 	}
353 	return (kern_jail(td, &j));
354 }
355 
356 int
357 kern_jail(struct thread *td, struct jail *j)
358 {
359 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
360 #ifdef INET
361 			    + 1
362 #endif
363 #ifdef INET6
364 			    + 1
365 #endif
366 			    )];
367 	struct uio opt;
368 	char *u_path, *u_hostname, *u_name;
369 	struct bool_flags *bf;
370 #ifdef INET
371 	uint32_t ip4s;
372 	struct in_addr *u_ip4;
373 #endif
374 #ifdef INET6
375 	struct in6_addr *u_ip6;
376 #endif
377 	size_t tmplen;
378 	int error, enforce_statfs;
379 
380 	bzero(&optiov, sizeof(optiov));
381 	opt.uio_iov = optiov;
382 	opt.uio_iovcnt = 0;
383 	opt.uio_offset = -1;
384 	opt.uio_resid = -1;
385 	opt.uio_segflg = UIO_SYSSPACE;
386 	opt.uio_rw = UIO_READ;
387 	opt.uio_td = td;
388 
389 	/* Set permissions for top-level jails from sysctls. */
390 	if (!jailed(td->td_ucred)) {
391 		for (bf = pr_flag_allow;
392 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
393 			atomic_load_int(&bf->flag) != 0;
394 		     bf++) {
395 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
396 			    (jail_default_allow & bf->flag)
397 			    ? bf->name : bf->noname);
398 			optiov[opt.uio_iovcnt].iov_len =
399 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
400 			opt.uio_iovcnt += 2;
401 		}
402 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
403 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
404 		opt.uio_iovcnt++;
405 		enforce_statfs = jail_default_enforce_statfs;
406 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
407 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
408 		opt.uio_iovcnt++;
409 	}
410 
411 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
412 #ifdef INET
413 	ip4s = (j->version == 0) ? 1 : j->ip4s;
414 	if (ip4s > jail_max_af_ips)
415 		return (EINVAL);
416 	tmplen += ip4s * sizeof(struct in_addr);
417 #else
418 	if (j->ip4s > 0)
419 		return (EINVAL);
420 #endif
421 #ifdef INET6
422 	if (j->ip6s > jail_max_af_ips)
423 		return (EINVAL);
424 	tmplen += j->ip6s * sizeof(struct in6_addr);
425 #else
426 	if (j->ip6s > 0)
427 		return (EINVAL);
428 #endif
429 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
430 	u_hostname = u_path + MAXPATHLEN;
431 	u_name = u_hostname + MAXHOSTNAMELEN;
432 #ifdef INET
433 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
434 #endif
435 #ifdef INET6
436 #ifdef INET
437 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
438 #else
439 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
440 #endif
441 #endif
442 	optiov[opt.uio_iovcnt].iov_base = "path";
443 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
444 	opt.uio_iovcnt++;
445 	optiov[opt.uio_iovcnt].iov_base = u_path;
446 	error = copyinstr(j->path, u_path, MAXPATHLEN,
447 	    &optiov[opt.uio_iovcnt].iov_len);
448 	if (error) {
449 		free(u_path, M_TEMP);
450 		return (error);
451 	}
452 	opt.uio_iovcnt++;
453 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
454 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
455 	opt.uio_iovcnt++;
456 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
457 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
458 	    &optiov[opt.uio_iovcnt].iov_len);
459 	if (error) {
460 		free(u_path, M_TEMP);
461 		return (error);
462 	}
463 	opt.uio_iovcnt++;
464 	if (j->jailname != NULL) {
465 		optiov[opt.uio_iovcnt].iov_base = "name";
466 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
467 		opt.uio_iovcnt++;
468 		optiov[opt.uio_iovcnt].iov_base = u_name;
469 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
470 		    &optiov[opt.uio_iovcnt].iov_len);
471 		if (error) {
472 			free(u_path, M_TEMP);
473 			return (error);
474 		}
475 		opt.uio_iovcnt++;
476 	}
477 #ifdef INET
478 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
479 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
480 	opt.uio_iovcnt++;
481 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
482 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
483 	if (j->version == 0)
484 		u_ip4->s_addr = j->ip4s;
485 	else {
486 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
487 		if (error) {
488 			free(u_path, M_TEMP);
489 			return (error);
490 		}
491 	}
492 	opt.uio_iovcnt++;
493 #endif
494 #ifdef INET6
495 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
496 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
497 	opt.uio_iovcnt++;
498 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
499 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
500 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
501 	if (error) {
502 		free(u_path, M_TEMP);
503 		return (error);
504 	}
505 	opt.uio_iovcnt++;
506 #endif
507 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
508 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
509 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
510 	free(u_path, M_TEMP);
511 	return (error);
512 }
513 
514 /*
515  * struct jail_set_args {
516  *	struct iovec *iovp;
517  *	unsigned int iovcnt;
518  *	int flags;
519  * };
520  */
521 int
522 sys_jail_set(struct thread *td, struct jail_set_args *uap)
523 {
524 	struct uio *auio;
525 	int error;
526 
527 	/* Check that we have an even number of iovecs. */
528 	if (uap->iovcnt & 1)
529 		return (EINVAL);
530 
531 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
532 	if (error)
533 		return (error);
534 	error = kern_jail_set(td, auio, uap->flags);
535 	free(auio, M_IOV);
536 	return (error);
537 }
538 
539 #if defined(INET) || defined(INET6)
540 typedef int prison_addr_cmp_t(const void *, const void *);
541 typedef bool prison_addr_valid_t(const void *);
542 static const struct pr_family {
543 	size_t			size;
544 	prison_addr_cmp_t	*cmp;
545 	prison_addr_valid_t	*valid;
546 	int			ip_flag;
547 } pr_families[PR_FAMILY_MAX] = {
548 #ifdef INET
549 	[PR_INET] = {
550 		.size = sizeof(struct in_addr),
551 		.cmp = prison_qcmp_v4,
552 		.valid = prison_valid_v4,
553 		.ip_flag = PR_IP4_USER,
554 	 },
555 #endif
556 #ifdef INET6
557 	[PR_INET6] = {
558 		.size = sizeof(struct in6_addr),
559 		.cmp = prison_qcmp_v6,
560 		.valid = prison_valid_v6,
561 		.ip_flag = PR_IP6_USER,
562 	},
563 #endif
564 };
565 
566 /*
567  * Network address lists (pr_addrs) allocation for jails.  The addresses
568  * are accessed locklessly by the network stack, thus need to be protected by
569  * the network epoch.
570  */
571 struct prison_ip {
572 	struct epoch_context ctx;
573 	uint32_t	ips;
574 #ifdef FUTURE_C
575 	union {
576 		struct in_addr pr_ip4[];
577 		struct in6_addr pr_ip6[];
578 	};
579 #else /* No future C :( */
580 #define	PR_IP(pip, i)	((const char *)((pip) + 1) + pr_families[af].size * (i))
581 #define	PR_IPD(pip, i)	((char *)((pip) + 1) + pr_families[af].size * (i))
582 #endif
583 };
584 
585 static struct prison_ip *
586 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
587 {
588 	struct prison_ip *pip;
589 
590 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
591 	    M_PRISON, flags);
592 	if (pip != NULL)
593 		pip->ips = cnt;
594 	return (pip);
595 }
596 
597 /*
598  * Allocate and copyin user supplied address list, sorting and validating.
599  * kern_jail_set() helper.
600  */
601 static struct prison_ip *
602 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
603 {
604 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
605 	const size_t size = pr_families[af].size;
606 	struct prison_ip *pip;
607 
608 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
609 	bcopy(op, pip + 1, cnt * size);
610 	/*
611 	 * IP addresses are all sorted but ip[0] to preserve
612 	 * the primary IP address as given from userland.
613 	 * This special IP is used for unbound outgoing
614 	 * connections as well for "loopback" traffic in case
615 	 * source address selection cannot find any more fitting
616 	 * address to connect from.
617 	 */
618 	if (cnt > 1)
619 		qsort((char *)(pip + 1) + size, cnt - 1, size,
620 		    pr_families[af].cmp);
621 	/*
622 	 * Check for duplicate addresses and do some simple
623 	 * zero and broadcast checks. If users give other bogus
624 	 * addresses it is their problem.
625 	 */
626 	for (int i = 0; i < cnt; i++) {
627 		if (!pr_families[af].valid(PR_IP(pip, i))) {
628 			free(pip, M_PRISON);
629 			return (NULL);
630 		}
631 		if (i + 1 < cnt &&
632 		    (cmp(PR_IP(pip, 0), PR_IP(pip, i + 1)) == 0 ||
633 		     cmp(PR_IP(pip, i), PR_IP(pip, i + 1)) == 0)) {
634 			free(pip, M_PRISON);
635 			return (NULL);
636 		}
637 	}
638 
639 	return (pip);
640 }
641 
642 /*
643  * Allocate and dup parent prison address list.
644  * kern_jail_set() helper.
645  */
646 static void
647 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
648 {
649 
650 	if (ppr->pr_addrs[af] != NULL) {
651 		pr->pr_addrs[af] = prison_ip_alloc(af,
652 		    ppr->pr_addrs[af]->ips, M_WAITOK);
653 		bcopy(ppr->pr_addrs[af], pr->pr_addrs[af],
654 		    pr->pr_addrs[af]->ips * pr_families[af].size);
655 	}
656 }
657 
658 /*
659  * Make sure the new set of IP addresses is a subset of the parent's list.
660  * Don't worry about the parent being unlocked, as any setting is done with
661  * allprison_lock held.
662  * kern_jail_set() helper.
663  */
664 static bool
665 prison_ip_parent_match(const struct prison_ip *ppip,
666     const struct prison_ip *pip, const pr_family_t af)
667 {
668 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
669 	int i, j;
670 
671 	if (ppip == NULL)
672 		return (false);
673 
674 	for (i = 0; i < ppip->ips; i++)
675 		if (cmp(PR_IP(pip, 0), PR_IP(ppip, i)) == 0)
676 			break;
677 
678 	if (i == ppip->ips)
679 		/* Main address not present in parent. */
680 		return (false);
681 
682 	if (pip->ips > 1) {
683 		for (i = j = 1; i < pip->ips; i++) {
684 			if (cmp(PR_IP(pip, i), PR_IP(ppip, 0)) == 0)
685 				/* Equals to parent primary address. */
686 				continue;
687 			for (; j < ppip->ips; j++)
688 				if (cmp(PR_IP(pip, i), PR_IP(ppip, j)) == 0)
689 					break;
690 			if (j == ppip->ips)
691 				break;
692 		}
693 		if (j == ppip->ips)
694 			/* Address not present in parent. */
695 			return (false);
696 	}
697 	return (true);
698 }
699 
700 /*
701  * Check for conflicting IP addresses.  We permit them if there is no more
702  * than one IP on each jail.  If there is a duplicate on a jail with more
703  * than one IP stop checking and return error.
704  * kern_jail_set() helper.
705  */
706 static bool
707 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
708     const struct prison_ip *pip, pr_family_t af)
709 {
710 	const struct prison *tppr, *tpr;
711 	int descend;
712 
713 #ifdef VIMAGE
714 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
715 		if (tppr->pr_flags & PR_VNET)
716 			break;
717 #else
718 	tppr = &prison0;
719 #endif
720 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
721 		if (tpr == pr ||
722 #ifdef VIMAGE
723 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
724 #endif
725 		    !prison_isalive(tpr)) {
726 			descend = 0;
727 			continue;
728 		}
729 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
730 			continue;
731 		descend = 0;
732 		if (tpr->pr_addrs[af] == NULL ||
733 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
734 			continue;
735 		for (int i = 0; i < pip->ips; i++)
736 			if (prison_ip_check(tpr, af, PR_IP(pip, i)) == 0)
737 				return (false);
738 	}
739 
740 	return (true);
741 }
742 
743 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
744     "prison must start with epoch context");
745 static void
746 prison_ip_free_deferred(epoch_context_t ctx)
747 {
748 
749 	free(ctx, M_PRISON);
750 }
751 
752 static void
753 prison_ip_free(struct prison_ip *pip)
754 {
755 
756 	if (pip != NULL)
757 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
758 }
759 
760 static void
761 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
762 {
763 	struct prison_ip **mem, *old;
764 
765 	mtx_assert(&pr->pr_mtx, MA_OWNED);
766 
767 	mem = &pr->pr_addrs[af];
768 
769 	old = *mem;
770 	ck_pr_store_ptr(mem, new);
771 	prison_ip_free(old);
772 }
773 
774 /*
775  * Restrict a prison's IP address list with its parent's, possibly replacing
776  * it.  Return true if the replacement buffer was used (or would have been).
777  * kern_jail_set() helper.
778  */
779 static bool
780 prison_ip_restrict(struct prison *pr, const pr_family_t af,
781     struct prison_ip *new)
782 {
783 	const struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
784 	const struct prison_ip *pip = pr->pr_addrs[af];
785 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
786 	const size_t size = pr_families[af].size;
787 	uint32_t ips;
788 	bool alloced;
789 
790 	mtx_assert(&pr->pr_mtx, MA_OWNED);
791 
792 	/*
793 	 * Due to epoch-synchronized access to the IP address lists we always
794 	 * allocate a new list even if the old one has enough space.  We could
795 	 * atomically update an IPv4 address inside a list, but that would
796 	 * screw up sorting, and in case of IPv6 we can't even atomically write
797 	 * one.
798 	 */
799 	ips = (pr->pr_flags & pr_families[af].ip_flag) ? pip->ips : ppip->ips;
800 	if (ips == 0) {
801 		prison_ip_set(pr, af, NULL);
802 		return (false);
803 	}
804 	if (new == NULL) {
805 		new = prison_ip_alloc(af, ips, M_NOWAIT);
806 		if (new == NULL)
807 			return (true);
808 		alloced = true;
809 	} else
810 		alloced = false;
811 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
812 		/* This has no user settings, so just copy the parent's list. */
813 		bcopy(ppip, new, ips * size);
814 	} else {
815 		/* Remove addresses that aren't in the parent. */
816 		int i;
817 
818 		i = 0; /* index in pip */
819 		ips = 0; /* index in new */
820 
821 		for (int pi = 0; pi < ppip->ips; pi++)
822 			if (cmp(PR_IP(pip, 0), PR_IP(ppip, pi)) == 0) {
823 				/* Found our primary address in parent. */
824 				bcopy(PR_IP(pip, i), PR_IPD(new, ips), size);
825 				i++;
826 				ips++;
827 				break;
828 			}
829 		for (int pi = 1; i < pip->ips; ) {
830 			/* Check against primary, which is unsorted. */
831 			if (cmp(PR_IP(pip, i), PR_IP(ppip, 0)) == 0) {
832 				/* Matches parent's primary address. */
833 				bcopy(PR_IP(pip, i), PR_IPD(new, ips), size);
834 				i++;
835 				ips++;
836 				continue;
837 			}
838 			/* The rest are sorted. */
839 			switch (pi >= ppip->ips ? -1 :
840 				cmp(PR_IP(pip, i), PR_IP(ppip, pi))) {
841 			case -1:
842 				i++;
843 				break;
844 			case 0:
845 				bcopy(PR_IP(pr, i), PR_IPD(new, ips), size);
846 				i++;
847 				pi++;
848 				ips++;
849 				break;
850 			case 1:
851 				pi++;
852 				break;
853 			}
854 		}
855 		if (ips == 0) {
856 			if (alloced)
857 				prison_ip_free(new);
858 			new = NULL;
859 		}
860 	}
861 	prison_ip_set(pr, af, new);
862 	return (new != NULL ? true : false);
863 }
864 
865 /*
866  * Fast-path check if an address belongs to a prison.
867  */
868 int
869 prison_ip_check(const struct prison *pr, const pr_family_t af,
870     const void *addr)
871 {
872 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
873 	const struct prison_ip *pip;
874 	int i, a, z, d;
875 
876 	MPASS(mtx_owned(&pr->pr_mtx) ||
877 	    in_epoch(net_epoch_preempt) ||
878 	    sx_xlocked(&allprison_lock));
879 
880 	pip = ck_pr_load_ptr(&pr->pr_addrs[af]);
881 	if (__predict_false(pip == NULL))
882 		return (EAFNOSUPPORT);
883 
884 	/* Check the primary IP. */
885 	if (cmp(PR_IP(pip, 0), addr) == 0)
886 		return (0);
887 
888 	/*
889 	 * All the other IPs are sorted so we can do a binary search.
890 	 */
891 	a = 0;
892 	z = pip->ips - 2;
893 	while (a <= z) {
894 		i = (a + z) / 2;
895 		d = cmp(PR_IP(pip, i + 1), addr);
896 		if (d > 0)
897 			z = i - 1;
898 		else if (d < 0)
899 			a = i + 1;
900 		else
901 			return (0);
902 	}
903 
904 	return (EADDRNOTAVAIL);
905 }
906 
907 /*
908  * Grab primary IP.  Historically required mutex, but nothing prevents
909  * us to support epoch-protected access.  Is it used in fast path?
910  * in{6}_jail.c helper
911  */
912 const void *
913 prison_ip_get0(const struct prison *pr, const pr_family_t af)
914 {
915 	const struct prison_ip *pip = pr->pr_addrs[af];
916 
917 	mtx_assert(&pr->pr_mtx, MA_OWNED);
918 	MPASS(pip);
919 
920 	return (pip + 1);
921 }
922 
923 u_int
924 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
925 {
926 
927 	return (pr->pr_addrs[af]->ips);
928 }
929 #endif	/* defined(INET) || defined(INET6) */
930 
931 int
932 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
933 {
934 	struct nameidata nd;
935 #ifdef INET
936 	struct prison_ip *ip4;
937 #endif
938 #ifdef INET6
939 	struct prison_ip *ip6;
940 #endif
941 	struct vfsopt *opt;
942 	struct vfsoptlist *opts;
943 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
944 	struct vnode *root;
945 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
946 	char *g_path, *osrelstr;
947 	struct bool_flags *bf;
948 	struct jailsys_flags *jsf;
949 #if defined(INET) || defined(INET6)
950 	void *op;
951 #endif
952 	unsigned long hid;
953 	size_t namelen, onamelen, pnamelen;
954 	int born, created, cuflags, descend, drflags, enforce;
955 	int error, errmsg_len, errmsg_pos;
956 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
957 	int jid, jsys, len, level;
958 	int childmax, osreldt, rsnum, slevel;
959 #ifdef INET
960 	int ip4s, redo_ip4;
961 #endif
962 #ifdef INET6
963 	int ip6s, redo_ip6;
964 #endif
965 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
966 	uint64_t pr_allow_diff;
967 	unsigned tallow;
968 	char numbuf[12];
969 
970 	error = priv_check(td, PRIV_JAIL_SET);
971 	if (!error && (flags & JAIL_ATTACH))
972 		error = priv_check(td, PRIV_JAIL_ATTACH);
973 	if (error)
974 		return (error);
975 	mypr = td->td_ucred->cr_prison;
976 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
977 		return (EPERM);
978 	if (flags & ~JAIL_SET_MASK)
979 		return (EINVAL);
980 
981 	/*
982 	 * Check all the parameters before committing to anything.  Not all
983 	 * errors can be caught early, but we may as well try.  Also, this
984 	 * takes care of some expensive stuff (path lookup) before getting
985 	 * the allprison lock.
986 	 *
987 	 * XXX Jails are not filesystems, and jail parameters are not mount
988 	 *     options.  But it makes more sense to re-use the vfsopt code
989 	 *     than duplicate it under a different name.
990 	 */
991 	error = vfs_buildopts(optuio, &opts);
992 	if (error)
993 		return (error);
994 #ifdef INET
995 	ip4 = NULL;
996 #endif
997 #ifdef INET6
998 	ip6 = NULL;
999 #endif
1000 	g_path = NULL;
1001 
1002 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1003 	if (!cuflags) {
1004 		error = EINVAL;
1005 		vfs_opterror(opts, "no valid operation (create or update)");
1006 		goto done_errmsg;
1007 	}
1008 
1009 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1010 	if (error == ENOENT)
1011 		jid = 0;
1012 	else if (error != 0)
1013 		goto done_free;
1014 
1015 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1016 	if (error == ENOENT)
1017 		gotslevel = 0;
1018 	else if (error != 0)
1019 		goto done_free;
1020 	else
1021 		gotslevel = 1;
1022 
1023 	error =
1024 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1025 	if (error == ENOENT)
1026 		gotchildmax = 0;
1027 	else if (error != 0)
1028 		goto done_free;
1029 	else
1030 		gotchildmax = 1;
1031 
1032 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1033 	if (error == ENOENT)
1034 		gotenforce = 0;
1035 	else if (error != 0)
1036 		goto done_free;
1037 	else if (enforce < 0 || enforce > 2) {
1038 		error = EINVAL;
1039 		goto done_free;
1040 	} else
1041 		gotenforce = 1;
1042 
1043 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1044 	if (error == ENOENT)
1045 		gotrsnum = 0;
1046 	else if (error != 0)
1047 		goto done_free;
1048 	else
1049 		gotrsnum = 1;
1050 
1051 	pr_flags = ch_flags = 0;
1052 	for (bf = pr_flag_bool;
1053 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1054 	     bf++) {
1055 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1056 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1057 	}
1058 	ch_flags |= pr_flags;
1059 	for (jsf = pr_flag_jailsys;
1060 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1061 	     jsf++) {
1062 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1063 		if (error == ENOENT)
1064 			continue;
1065 		if (error != 0)
1066 			goto done_free;
1067 		switch (jsys) {
1068 		case JAIL_SYS_DISABLE:
1069 			if (!jsf->disable) {
1070 				error = EINVAL;
1071 				goto done_free;
1072 			}
1073 			pr_flags |= jsf->disable;
1074 			break;
1075 		case JAIL_SYS_NEW:
1076 			pr_flags |= jsf->new;
1077 			break;
1078 		case JAIL_SYS_INHERIT:
1079 			break;
1080 		default:
1081 			error = EINVAL;
1082 			goto done_free;
1083 		}
1084 		ch_flags |= jsf->new | jsf->disable;
1085 	}
1086 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1087 	    && !(pr_flags & PR_PERSIST)) {
1088 		error = EINVAL;
1089 		vfs_opterror(opts, "new jail must persist or attach");
1090 		goto done_errmsg;
1091 	}
1092 #ifdef VIMAGE
1093 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1094 		error = EINVAL;
1095 		vfs_opterror(opts, "vnet cannot be changed after creation");
1096 		goto done_errmsg;
1097 	}
1098 #endif
1099 #ifdef INET
1100 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1101 		error = EINVAL;
1102 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1103 		goto done_errmsg;
1104 	}
1105 #endif
1106 #ifdef INET6
1107 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1108 		error = EINVAL;
1109 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1110 		goto done_errmsg;
1111 	}
1112 #endif
1113 
1114 	pr_allow = ch_allow = 0;
1115 	for (bf = pr_flag_allow;
1116 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1117 		atomic_load_int(&bf->flag) != 0;
1118 	     bf++) {
1119 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1120 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1121 	}
1122 	ch_allow |= pr_allow;
1123 
1124 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1125 	if (error == ENOENT)
1126 		name = NULL;
1127 	else if (error != 0)
1128 		goto done_free;
1129 	else {
1130 		if (len == 0 || name[len - 1] != '\0') {
1131 			error = EINVAL;
1132 			goto done_free;
1133 		}
1134 		if (len > MAXHOSTNAMELEN) {
1135 			error = ENAMETOOLONG;
1136 			goto done_free;
1137 		}
1138 	}
1139 
1140 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1141 	if (error == ENOENT)
1142 		host = NULL;
1143 	else if (error != 0)
1144 		goto done_free;
1145 	else {
1146 		ch_flags |= PR_HOST;
1147 		pr_flags |= PR_HOST;
1148 		if (len == 0 || host[len - 1] != '\0') {
1149 			error = EINVAL;
1150 			goto done_free;
1151 		}
1152 		if (len > MAXHOSTNAMELEN) {
1153 			error = ENAMETOOLONG;
1154 			goto done_free;
1155 		}
1156 	}
1157 
1158 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1159 	if (error == ENOENT)
1160 		domain = NULL;
1161 	else if (error != 0)
1162 		goto done_free;
1163 	else {
1164 		ch_flags |= PR_HOST;
1165 		pr_flags |= PR_HOST;
1166 		if (len == 0 || domain[len - 1] != '\0') {
1167 			error = EINVAL;
1168 			goto done_free;
1169 		}
1170 		if (len > MAXHOSTNAMELEN) {
1171 			error = ENAMETOOLONG;
1172 			goto done_free;
1173 		}
1174 	}
1175 
1176 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1177 	if (error == ENOENT)
1178 		uuid = NULL;
1179 	else if (error != 0)
1180 		goto done_free;
1181 	else {
1182 		ch_flags |= PR_HOST;
1183 		pr_flags |= PR_HOST;
1184 		if (len == 0 || uuid[len - 1] != '\0') {
1185 			error = EINVAL;
1186 			goto done_free;
1187 		}
1188 		if (len > HOSTUUIDLEN) {
1189 			error = ENAMETOOLONG;
1190 			goto done_free;
1191 		}
1192 	}
1193 
1194 #ifdef COMPAT_FREEBSD32
1195 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1196 		uint32_t hid32;
1197 
1198 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1199 		hid = hid32;
1200 	} else
1201 #endif
1202 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1203 	if (error == ENOENT)
1204 		gothid = 0;
1205 	else if (error != 0)
1206 		goto done_free;
1207 	else {
1208 		gothid = 1;
1209 		ch_flags |= PR_HOST;
1210 		pr_flags |= PR_HOST;
1211 	}
1212 
1213 #ifdef INET
1214 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1215 	if (error == ENOENT)
1216 		ip4s = 0;
1217 	else if (error != 0)
1218 		goto done_free;
1219 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1220 		error = EINVAL;
1221 		goto done_free;
1222 	} else {
1223 		ch_flags |= PR_IP4_USER;
1224 		pr_flags |= PR_IP4_USER;
1225 		if (ip4s > 0) {
1226 			ip4s /= sizeof(struct in_addr);
1227 			if (ip4s > jail_max_af_ips) {
1228 				error = EINVAL;
1229 				vfs_opterror(opts, "too many IPv4 addresses");
1230 				goto done_errmsg;
1231 			}
1232 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1233 			if (ip4 == NULL) {
1234 				error = EINVAL;
1235 				goto done_free;
1236 			}
1237 		}
1238 	}
1239 #endif
1240 
1241 #ifdef INET6
1242 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1243 	if (error == ENOENT)
1244 		ip6s = 0;
1245 	else if (error != 0)
1246 		goto done_free;
1247 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1248 		error = EINVAL;
1249 		goto done_free;
1250 	} else {
1251 		ch_flags |= PR_IP6_USER;
1252 		pr_flags |= PR_IP6_USER;
1253 		if (ip6s > 0) {
1254 			ip6s /= sizeof(struct in6_addr);
1255 			if (ip6s > jail_max_af_ips) {
1256 				error = EINVAL;
1257 				vfs_opterror(opts, "too many IPv6 addresses");
1258 				goto done_errmsg;
1259 			}
1260 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1261 			if (ip6 == NULL) {
1262 				error = EINVAL;
1263 				goto done_free;
1264 			}
1265 		}
1266 	}
1267 #endif
1268 
1269 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1270 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1271 		error = EINVAL;
1272 		vfs_opterror(opts,
1273 		    "vnet jails cannot have IP address restrictions");
1274 		goto done_errmsg;
1275 	}
1276 #endif
1277 
1278 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1279 	if (error == ENOENT)
1280 		osrelstr = NULL;
1281 	else if (error != 0)
1282 		goto done_free;
1283 	else {
1284 		if (flags & JAIL_UPDATE) {
1285 			error = EINVAL;
1286 			vfs_opterror(opts,
1287 			    "osrelease cannot be changed after creation");
1288 			goto done_errmsg;
1289 		}
1290 		if (len == 0 || osrelstr[len - 1] != '\0') {
1291 			error = EINVAL;
1292 			goto done_free;
1293 		}
1294 		if (len >= OSRELEASELEN) {
1295 			error = ENAMETOOLONG;
1296 			vfs_opterror(opts,
1297 			    "osrelease string must be 1-%d bytes long",
1298 			    OSRELEASELEN - 1);
1299 			goto done_errmsg;
1300 		}
1301 	}
1302 
1303 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1304 	if (error == ENOENT)
1305 		osreldt = 0;
1306 	else if (error != 0)
1307 		goto done_free;
1308 	else {
1309 		if (flags & JAIL_UPDATE) {
1310 			error = EINVAL;
1311 			vfs_opterror(opts,
1312 			    "osreldate cannot be changed after creation");
1313 			goto done_errmsg;
1314 		}
1315 		if (osreldt == 0) {
1316 			error = EINVAL;
1317 			vfs_opterror(opts, "osreldate cannot be 0");
1318 			goto done_errmsg;
1319 		}
1320 	}
1321 
1322 	root = NULL;
1323 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1324 	if (error == ENOENT)
1325 		path = NULL;
1326 	else if (error != 0)
1327 		goto done_free;
1328 	else {
1329 		if (flags & JAIL_UPDATE) {
1330 			error = EINVAL;
1331 			vfs_opterror(opts,
1332 			    "path cannot be changed after creation");
1333 			goto done_errmsg;
1334 		}
1335 		if (len == 0 || path[len - 1] != '\0') {
1336 			error = EINVAL;
1337 			goto done_free;
1338 		}
1339 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1340 		error = namei(&nd);
1341 		if (error)
1342 			goto done_free;
1343 		root = nd.ni_vp;
1344 		NDFREE_PNBUF(&nd);
1345 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1346 		strlcpy(g_path, path, MAXPATHLEN);
1347 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1348 		if (error == 0) {
1349 			path = g_path;
1350 		} else {
1351 			/* exit on other errors */
1352 			goto done_free;
1353 		}
1354 		if (root->v_type != VDIR) {
1355 			error = ENOTDIR;
1356 			vput(root);
1357 			goto done_free;
1358 		}
1359 		VOP_UNLOCK(root);
1360 	}
1361 
1362 	/*
1363 	 * Find the specified jail, or at least its parent.
1364 	 * This abuses the file error codes ENOENT and EEXIST.
1365 	 */
1366 	pr = NULL;
1367 	inspr = NULL;
1368 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1369 		namelc = strrchr(name, '.');
1370 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1371 		if (*p != '\0')
1372 			jid = 0;
1373 	}
1374 	sx_xlock(&allprison_lock);
1375 	drflags = PD_LIST_XLOCKED;
1376 	ppr = mypr;
1377 	if (!prison_isalive(ppr)) {
1378 		/* This jail is dying.  This process will surely follow. */
1379 		error = EAGAIN;
1380 		goto done_deref;
1381 	}
1382 	if (jid != 0) {
1383 		if (jid < 0) {
1384 			error = EINVAL;
1385 			vfs_opterror(opts, "negative jid");
1386 			goto done_deref;
1387 		}
1388 		/*
1389 		 * See if a requested jid already exists.  Keep track of
1390 		 * where it can be inserted later.
1391 		 */
1392 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1393 			if (inspr->pr_id < jid)
1394 				continue;
1395 			if (inspr->pr_id > jid)
1396 				break;
1397 			pr = inspr;
1398 			mtx_lock(&pr->pr_mtx);
1399 			drflags |= PD_LOCKED;
1400 			inspr = NULL;
1401 			break;
1402 		}
1403 		if (pr != NULL) {
1404 			/* Create: jid must not exist. */
1405 			if (cuflags == JAIL_CREATE) {
1406 				/*
1407 				 * Even creators that cannot see the jail will
1408 				 * get EEXIST.
1409 				 */
1410 				error = EEXIST;
1411 				vfs_opterror(opts, "jail %d already exists",
1412 				    jid);
1413 				goto done_deref;
1414 			}
1415 			if (!prison_ischild(mypr, pr)) {
1416 				/*
1417 				 * Updaters get ENOENT if they cannot see the
1418 				 * jail.  This is true even for CREATE | UPDATE,
1419 				 * which normally cannot give this error.
1420 				 */
1421 				error = ENOENT;
1422 				vfs_opterror(opts, "jail %d not found", jid);
1423 				goto done_deref;
1424 			}
1425 			ppr = pr->pr_parent;
1426 			if (!prison_isalive(ppr)) {
1427 				error = ENOENT;
1428 				vfs_opterror(opts, "jail %d is dying",
1429 				    ppr->pr_id);
1430 				goto done_deref;
1431 			}
1432 			if (!prison_isalive(pr)) {
1433 				if (!(flags & JAIL_DYING)) {
1434 					error = ENOENT;
1435 					vfs_opterror(opts, "jail %d is dying",
1436 					    jid);
1437 					goto done_deref;
1438 				}
1439 				if ((flags & JAIL_ATTACH) ||
1440 				    (pr_flags & PR_PERSIST)) {
1441 					/*
1442 					 * A dying jail might be resurrected
1443 					 * (via attach or persist), but first
1444 					 * it must determine if another jail
1445 					 * has claimed its name.  Accomplish
1446 					 * this by implicitly re-setting the
1447 					 * name.
1448 					 */
1449 					if (name == NULL)
1450 						name = prison_name(mypr, pr);
1451 				}
1452 			}
1453 		} else {
1454 			/* Update: jid must exist. */
1455 			if (cuflags == JAIL_UPDATE) {
1456 				error = ENOENT;
1457 				vfs_opterror(opts, "jail %d not found", jid);
1458 				goto done_deref;
1459 			}
1460 		}
1461 	}
1462 	/*
1463 	 * If the caller provided a name, look for a jail by that name.
1464 	 * This has different semantics for creates and updates keyed by jid
1465 	 * (where the name must not already exist in a different jail),
1466 	 * and updates keyed by the name itself (where the name must exist
1467 	 * because that is the jail being updated).
1468 	 */
1469 	namelc = NULL;
1470 	if (name != NULL) {
1471 		namelc = strrchr(name, '.');
1472 		if (namelc == NULL)
1473 			namelc = name;
1474 		else {
1475 			/*
1476 			 * This is a hierarchical name.  Split it into the
1477 			 * parent and child names, and make sure the parent
1478 			 * exists or matches an already found jail.
1479 			 */
1480 			if (pr != NULL) {
1481 				if (strncmp(name, ppr->pr_name, namelc - name)
1482 				    || ppr->pr_name[namelc - name] != '\0') {
1483 					error = EINVAL;
1484 					vfs_opterror(opts,
1485 					    "cannot change jail's parent");
1486 					goto done_deref;
1487 				}
1488 			} else {
1489 				*namelc = '\0';
1490 				ppr = prison_find_name(mypr, name);
1491 				if (ppr == NULL) {
1492 					error = ENOENT;
1493 					vfs_opterror(opts,
1494 					    "jail \"%s\" not found", name);
1495 					goto done_deref;
1496 				}
1497 				mtx_unlock(&ppr->pr_mtx);
1498 				if (!prison_isalive(ppr)) {
1499 					error = ENOENT;
1500 					vfs_opterror(opts,
1501 					    "jail \"%s\" is dying", name);
1502 					goto done_deref;
1503 				}
1504 				*namelc = '.';
1505 			}
1506 			namelc++;
1507 		}
1508 		if (namelc[0] != '\0') {
1509 			pnamelen =
1510 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1511 			deadpr = NULL;
1512 			FOREACH_PRISON_CHILD(ppr, tpr) {
1513 				if (tpr != pr &&
1514 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1515 					if (prison_isalive(tpr)) {
1516 						if (pr == NULL &&
1517 						    cuflags != JAIL_CREATE) {
1518 							/*
1519 							 * Use this jail
1520 							 * for updates.
1521 							 */
1522 							pr = tpr;
1523 							mtx_lock(&pr->pr_mtx);
1524 							drflags |= PD_LOCKED;
1525 							break;
1526 						}
1527 						/*
1528 						 * Create, or update(jid):
1529 						 * name must not exist in an
1530 						 * active sibling jail.
1531 						 */
1532 						error = EEXIST;
1533 						vfs_opterror(opts,
1534 						   "jail \"%s\" already exists",
1535 						   name);
1536 						goto done_deref;
1537 					}
1538 					if (pr == NULL &&
1539 					    cuflags != JAIL_CREATE) {
1540 						deadpr = tpr;
1541 					}
1542 				}
1543 			}
1544 			/* If no active jail is found, use a dying one. */
1545 			if (deadpr != NULL && pr == NULL) {
1546 				if (flags & JAIL_DYING) {
1547 					pr = deadpr;
1548 					mtx_lock(&pr->pr_mtx);
1549 					drflags |= PD_LOCKED;
1550 				} else if (cuflags == JAIL_UPDATE) {
1551 					error = ENOENT;
1552 					vfs_opterror(opts,
1553 					    "jail \"%s\" is dying", name);
1554 					goto done_deref;
1555 				}
1556 			}
1557 			/* Update: name must exist if no jid. */
1558 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1559 				error = ENOENT;
1560 				vfs_opterror(opts, "jail \"%s\" not found",
1561 				    name);
1562 				goto done_deref;
1563 			}
1564 		}
1565 	}
1566 	/* Update: must provide a jid or name. */
1567 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1568 		error = ENOENT;
1569 		vfs_opterror(opts, "update specified no jail");
1570 		goto done_deref;
1571 	}
1572 
1573 	/* If there's no prison to update, create a new one and link it in. */
1574 	created = pr == NULL;
1575 	if (created) {
1576 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1577 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1578 				error = EPERM;
1579 				vfs_opterror(opts, "prison limit exceeded");
1580 				goto done_deref;
1581 			}
1582 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1583 			error = EAGAIN;
1584 			vfs_opterror(opts, "no available jail IDs");
1585 			goto done_deref;
1586 		}
1587 
1588 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1589 		pr->pr_state = PRISON_STATE_INVALID;
1590 		refcount_init(&pr->pr_ref, 1);
1591 		refcount_init(&pr->pr_uref, 0);
1592 		drflags |= PD_DEREF;
1593 		LIST_INIT(&pr->pr_children);
1594 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1595 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1596 
1597 		pr->pr_id = jid;
1598 		if (inspr != NULL)
1599 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1600 		else
1601 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1602 
1603 		pr->pr_parent = ppr;
1604 		prison_hold(ppr);
1605 		prison_proc_hold(ppr);
1606 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1607 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1608 			tpr->pr_childcount++;
1609 
1610 		/* Set some default values, and inherit some from the parent. */
1611 		if (namelc == NULL)
1612 			namelc = "";
1613 		if (path == NULL) {
1614 			path = "/";
1615 			root = mypr->pr_root;
1616 			vref(root);
1617 		}
1618 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1619 		pr->pr_flags |= PR_HOST;
1620 #if defined(INET) || defined(INET6)
1621 #ifdef VIMAGE
1622 		if (!(pr_flags & PR_VNET))
1623 #endif
1624 		{
1625 #ifdef INET
1626 			if (!(ch_flags & PR_IP4_USER))
1627 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1628 			else if (!(pr_flags & PR_IP4_USER)) {
1629 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1630 				prison_ip_dup(ppr, pr, PR_INET);
1631 			}
1632 #endif
1633 #ifdef INET6
1634 			if (!(ch_flags & PR_IP6_USER))
1635 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1636 			else if (!(pr_flags & PR_IP6_USER)) {
1637 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1638 				prison_ip_dup(ppr, pr, PR_INET6);
1639 			}
1640 #endif
1641 		}
1642 #endif
1643 		/* Source address selection is always on by default. */
1644 		pr->pr_flags |= _PR_IP_SADDRSEL;
1645 
1646 		pr->pr_securelevel = ppr->pr_securelevel;
1647 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1648 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1649 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1650 
1651 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1652 		if (osrelstr == NULL)
1653 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1654 			    sizeof(pr->pr_osrelease));
1655 		else
1656 			strlcpy(pr->pr_osrelease, osrelstr,
1657 			    sizeof(pr->pr_osrelease));
1658 
1659 #ifdef VIMAGE
1660 		/* Allocate a new vnet if specified. */
1661 		pr->pr_vnet = (pr_flags & PR_VNET)
1662 		    ? vnet_alloc() : ppr->pr_vnet;
1663 #endif
1664 		/*
1665 		 * Allocate a dedicated cpuset for each jail.
1666 		 * Unlike other initial settings, this may return an error.
1667 		 */
1668 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1669 		if (error)
1670 			goto done_deref;
1671 
1672 		mtx_lock(&pr->pr_mtx);
1673 		drflags |= PD_LOCKED;
1674 	} else {
1675 		/*
1676 		 * Grab a reference for existing prisons, to ensure they
1677 		 * continue to exist for the duration of the call.
1678 		 */
1679 		prison_hold(pr);
1680 		drflags |= PD_DEREF;
1681 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1682 		if ((pr->pr_flags & PR_VNET) &&
1683 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1684 			error = EINVAL;
1685 			vfs_opterror(opts,
1686 			    "vnet jails cannot have IP address restrictions");
1687 			goto done_deref;
1688 		}
1689 #endif
1690 #ifdef INET
1691 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1692 			error = EINVAL;
1693 			vfs_opterror(opts,
1694 			    "ip4 cannot be changed after creation");
1695 			goto done_deref;
1696 		}
1697 #endif
1698 #ifdef INET6
1699 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1700 			error = EINVAL;
1701 			vfs_opterror(opts,
1702 			    "ip6 cannot be changed after creation");
1703 			goto done_deref;
1704 		}
1705 #endif
1706 	}
1707 
1708 	/* Do final error checking before setting anything. */
1709 	if (gotslevel) {
1710 		if (slevel < ppr->pr_securelevel) {
1711 			error = EPERM;
1712 			goto done_deref;
1713 		}
1714 	}
1715 	if (gotchildmax) {
1716 		if (childmax >= ppr->pr_childmax) {
1717 			error = EPERM;
1718 			goto done_deref;
1719 		}
1720 	}
1721 	if (gotenforce) {
1722 		if (enforce < ppr->pr_enforce_statfs) {
1723 			error = EPERM;
1724 			goto done_deref;
1725 		}
1726 	}
1727 	if (gotrsnum) {
1728 		/*
1729 		 * devfs_rsnum is a uint16_t
1730 		 */
1731 		if (rsnum < 0 || rsnum > 65535) {
1732 			error = EINVAL;
1733 			goto done_deref;
1734 		}
1735 		/*
1736 		 * Nested jails always inherit parent's devfs ruleset
1737 		 */
1738 		if (jailed(td->td_ucred)) {
1739 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1740 				error = EPERM;
1741 				goto done_deref;
1742 			} else
1743 				rsnum = ppr->pr_devfs_rsnum;
1744 		}
1745 	}
1746 #ifdef INET
1747 	if (ip4s > 0) {
1748 		if ((ppr->pr_flags & PR_IP4) &&
1749 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1750 		    PR_INET)) {
1751 			error = EPERM;
1752 			goto done_deref;
1753 		}
1754 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1755 			error = EADDRINUSE;
1756 			vfs_opterror(opts, "IPv4 addresses clash");
1757 			goto done_deref;
1758 		}
1759 	}
1760 #endif
1761 #ifdef INET6
1762 	if (ip6s > 0) {
1763 		if ((ppr->pr_flags & PR_IP6) &&
1764 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1765 		    PR_INET6)) {
1766 			error = EPERM;
1767 			goto done_deref;
1768 		}
1769 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1770 			error = EADDRINUSE;
1771 			vfs_opterror(opts, "IPv6 addresses clash");
1772 			goto done_deref;
1773 		}
1774 	}
1775 #endif
1776 	onamelen = namelen = 0;
1777 	if (namelc != NULL) {
1778 		/* Give a default name of the jid.  Also allow the name to be
1779 		 * explicitly the jid - but not any other number, and only in
1780 		 * normal form (no leading zero/etc).
1781 		 */
1782 		if (namelc[0] == '\0')
1783 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1784 		else if ((strtoul(namelc, &p, 10) != jid ||
1785 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1786 			error = EINVAL;
1787 			vfs_opterror(opts,
1788 			    "name cannot be numeric (unless it is the jid)");
1789 			goto done_deref;
1790 		}
1791 		/*
1792 		 * Make sure the name isn't too long for the prison or its
1793 		 * children.
1794 		 */
1795 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1796 		onamelen = strlen(pr->pr_name + pnamelen);
1797 		namelen = strlen(namelc);
1798 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1799 			error = ENAMETOOLONG;
1800 			goto done_deref;
1801 		}
1802 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1803 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1804 			    sizeof(pr->pr_name)) {
1805 				error = ENAMETOOLONG;
1806 				goto done_deref;
1807 			}
1808 		}
1809 	}
1810 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1811 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1812 		error = EPERM;
1813 		goto done_deref;
1814 	}
1815 
1816 	/*
1817 	 * Let modules check their parameters.  This requires unlocking and
1818 	 * then re-locking the prison, but this is still a valid state as long
1819 	 * as allprison_lock remains xlocked.
1820 	 */
1821 	mtx_unlock(&pr->pr_mtx);
1822 	drflags &= ~PD_LOCKED;
1823 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1824 	if (error != 0)
1825 		goto done_deref;
1826 	mtx_lock(&pr->pr_mtx);
1827 	drflags |= PD_LOCKED;
1828 
1829 	/* At this point, all valid parameters should have been noted. */
1830 	TAILQ_FOREACH(opt, opts, link) {
1831 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1832 			error = EINVAL;
1833 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1834 			goto done_deref;
1835 		}
1836 	}
1837 
1838 	/* Set the parameters of the prison. */
1839 #ifdef INET
1840 	redo_ip4 = 0;
1841 	if (pr_flags & PR_IP4_USER) {
1842 		pr->pr_flags |= PR_IP4;
1843 		prison_ip_set(pr, PR_INET, ip4);
1844 		ip4 = NULL;
1845 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1846 #ifdef VIMAGE
1847 			if (tpr->pr_flags & PR_VNET) {
1848 				descend = 0;
1849 				continue;
1850 			}
1851 #endif
1852 			if (prison_ip_restrict(tpr, PR_INET, NULL)) {
1853 				redo_ip4 = 1;
1854 				descend = 0;
1855 			}
1856 		}
1857 	}
1858 #endif
1859 #ifdef INET6
1860 	redo_ip6 = 0;
1861 	if (pr_flags & PR_IP6_USER) {
1862 		pr->pr_flags |= PR_IP6;
1863 		prison_ip_set(pr, PR_INET6, ip6);
1864 		ip6 = NULL;
1865 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1866 #ifdef VIMAGE
1867 			if (tpr->pr_flags & PR_VNET) {
1868 				descend = 0;
1869 				continue;
1870 			}
1871 #endif
1872 			if (prison_ip_restrict(tpr, PR_INET6, NULL)) {
1873 				redo_ip6 = 1;
1874 				descend = 0;
1875 			}
1876 		}
1877 	}
1878 #endif
1879 	if (gotslevel) {
1880 		pr->pr_securelevel = slevel;
1881 		/* Set all child jails to be at least this level. */
1882 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1883 			if (tpr->pr_securelevel < slevel)
1884 				tpr->pr_securelevel = slevel;
1885 	}
1886 	if (gotchildmax) {
1887 		pr->pr_childmax = childmax;
1888 		/* Set all child jails to under this limit. */
1889 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1890 			if (tpr->pr_childmax > childmax - level)
1891 				tpr->pr_childmax = childmax > level
1892 				    ? childmax - level : 0;
1893 	}
1894 	if (gotenforce) {
1895 		pr->pr_enforce_statfs = enforce;
1896 		/* Pass this restriction on to the children. */
1897 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1898 			if (tpr->pr_enforce_statfs < enforce)
1899 				tpr->pr_enforce_statfs = enforce;
1900 	}
1901 	if (gotrsnum) {
1902 		pr->pr_devfs_rsnum = rsnum;
1903 		/* Pass this restriction on to the children. */
1904 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1905 			tpr->pr_devfs_rsnum = rsnum;
1906 	}
1907 	if (namelc != NULL) {
1908 		if (ppr == &prison0)
1909 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1910 		else
1911 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1912 			    ppr->pr_name, namelc);
1913 		/* Change this component of child names. */
1914 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1915 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1916 			    strlen(tpr->pr_name + onamelen) + 1);
1917 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1918 		}
1919 	}
1920 	if (path != NULL) {
1921 		/* Try to keep a real-rooted full pathname. */
1922 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1923 		pr->pr_root = root;
1924 		root = NULL;
1925 	}
1926 	if (PR_HOST & ch_flags & ~pr_flags) {
1927 		if (pr->pr_flags & PR_HOST) {
1928 			/*
1929 			 * Copy the parent's host info.  As with pr_ip4 above,
1930 			 * the lack of a lock on the parent is not a problem;
1931 			 * it is always set with allprison_lock at least
1932 			 * shared, and is held exclusively here.
1933 			 */
1934 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1935 			    sizeof(pr->pr_hostname));
1936 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1937 			    sizeof(pr->pr_domainname));
1938 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1939 			    sizeof(pr->pr_hostuuid));
1940 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1941 		}
1942 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1943 		/* Set this prison, and any descendants without PR_HOST. */
1944 		if (host != NULL)
1945 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1946 		if (domain != NULL)
1947 			strlcpy(pr->pr_domainname, domain,
1948 			    sizeof(pr->pr_domainname));
1949 		if (uuid != NULL)
1950 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1951 		if (gothid)
1952 			pr->pr_hostid = hid;
1953 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1954 			if (tpr->pr_flags & PR_HOST)
1955 				descend = 0;
1956 			else {
1957 				if (host != NULL)
1958 					strlcpy(tpr->pr_hostname,
1959 					    pr->pr_hostname,
1960 					    sizeof(tpr->pr_hostname));
1961 				if (domain != NULL)
1962 					strlcpy(tpr->pr_domainname,
1963 					    pr->pr_domainname,
1964 					    sizeof(tpr->pr_domainname));
1965 				if (uuid != NULL)
1966 					strlcpy(tpr->pr_hostuuid,
1967 					    pr->pr_hostuuid,
1968 					    sizeof(tpr->pr_hostuuid));
1969 				if (gothid)
1970 					tpr->pr_hostid = hid;
1971 			}
1972 		}
1973 	}
1974 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1975 	if ((tallow = ch_allow & ~pr_allow))
1976 		prison_set_allow_locked(pr, tallow, 0);
1977 	/*
1978 	 * Persistent prisons get an extra reference, and prisons losing their
1979 	 * persist flag lose that reference.
1980 	 */
1981 	born = !prison_isalive(pr);
1982 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
1983 		if (pr_flags & PR_PERSIST) {
1984 			prison_hold(pr);
1985 			/*
1986 			 * This may make a dead prison alive again, but wait
1987 			 * to label it as such until after OSD calls have had
1988 			 * a chance to run (and perhaps to fail).
1989 			 */
1990 			refcount_acquire(&pr->pr_uref);
1991 		} else {
1992 			drflags |= PD_DEUREF;
1993 			prison_free_not_last(pr);
1994 		}
1995 	}
1996 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1997 	mtx_unlock(&pr->pr_mtx);
1998 	drflags &= ~PD_LOCKED;
1999 	/*
2000 	 * Any errors past this point will need to de-persist newly created
2001 	 * prisons, as well as call remove methods.
2002 	 */
2003 	if (born)
2004 		drflags |= PD_KILL;
2005 
2006 #ifdef RACCT
2007 	if (racct_enable && created)
2008 		prison_racct_attach(pr);
2009 #endif
2010 
2011 	/* Locks may have prevented a complete restriction of child IP
2012 	 * addresses.  If so, allocate some more memory and try again.
2013 	 */
2014 #ifdef INET
2015 	while (redo_ip4) {
2016 		ip4s = pr->pr_addrs[PR_INET]->ips;
2017 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2018 		mtx_lock(&pr->pr_mtx);
2019 		redo_ip4 = 0;
2020 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2021 #ifdef VIMAGE
2022 			if (tpr->pr_flags & PR_VNET) {
2023 				descend = 0;
2024 				continue;
2025 			}
2026 #endif
2027 			if (prison_ip_restrict(tpr, PR_INET, ip4)) {
2028 				if (ip4 != NULL)
2029 					ip4 = NULL;
2030 				else
2031 					redo_ip4 = 1;
2032 			}
2033 		}
2034 		mtx_unlock(&pr->pr_mtx);
2035 	}
2036 #endif
2037 #ifdef INET6
2038 	while (redo_ip6) {
2039 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2040 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2041 		mtx_lock(&pr->pr_mtx);
2042 		redo_ip6 = 0;
2043 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2044 #ifdef VIMAGE
2045 			if (tpr->pr_flags & PR_VNET) {
2046 				descend = 0;
2047 				continue;
2048 			}
2049 #endif
2050 			if (prison_ip_restrict(tpr, PR_INET6, ip6)) {
2051 				if (ip6 != NULL)
2052 					ip6 = NULL;
2053 				else
2054 					redo_ip6 = 1;
2055 			}
2056 		}
2057 		mtx_unlock(&pr->pr_mtx);
2058 	}
2059 #endif
2060 
2061 	/* Let the modules do their work. */
2062 	if (born) {
2063 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2064 		if (error)
2065 			goto done_deref;
2066 	}
2067 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2068 	if (error)
2069 		goto done_deref;
2070 
2071 	/*
2072 	 * A new prison is now ready to be seen; either it has gained a user
2073 	 * reference via persistence, or is about to gain one via attachment.
2074 	 */
2075 	if (born) {
2076 		drflags = prison_lock_xlock(pr, drflags);
2077 		pr->pr_state = PRISON_STATE_ALIVE;
2078 	}
2079 
2080 	/* Attach this process to the prison if requested. */
2081 	if (flags & JAIL_ATTACH) {
2082 		error = do_jail_attach(td, pr,
2083 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2084 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2085 		if (error) {
2086 			vfs_opterror(opts, "attach failed");
2087 			goto done_deref;
2088 		}
2089 	}
2090 
2091 #ifdef RACCT
2092 	if (racct_enable && !created) {
2093 		if (drflags & PD_LOCKED) {
2094 			mtx_unlock(&pr->pr_mtx);
2095 			drflags &= ~PD_LOCKED;
2096 		}
2097 		if (drflags & PD_LIST_XLOCKED) {
2098 			sx_xunlock(&allprison_lock);
2099 			drflags &= ~PD_LIST_XLOCKED;
2100 		}
2101 		prison_racct_modify(pr);
2102 	}
2103 #endif
2104 
2105 	drflags &= ~PD_KILL;
2106 	td->td_retval[0] = pr->pr_id;
2107 
2108  done_deref:
2109 	/* Release any temporary prison holds and/or locks. */
2110 	if (pr != NULL)
2111 		prison_deref(pr, drflags);
2112 	else if (drflags & PD_LIST_SLOCKED)
2113 		sx_sunlock(&allprison_lock);
2114 	else if (drflags & PD_LIST_XLOCKED)
2115 		sx_xunlock(&allprison_lock);
2116 	if (root != NULL)
2117 		vrele(root);
2118  done_errmsg:
2119 	if (error) {
2120 		/* Write the error message back to userspace. */
2121 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2122 		    &errmsg_len) == 0 && errmsg_len > 0) {
2123 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2124 			if (optuio->uio_segflg == UIO_SYSSPACE)
2125 				bcopy(errmsg,
2126 				    optuio->uio_iov[errmsg_pos].iov_base,
2127 				    errmsg_len);
2128 			else
2129 				copyout(errmsg,
2130 				    optuio->uio_iov[errmsg_pos].iov_base,
2131 				    errmsg_len);
2132 		}
2133 	}
2134  done_free:
2135 #ifdef INET
2136 	prison_ip_free(ip4);
2137 #endif
2138 #ifdef INET6
2139 	prison_ip_free(ip6);
2140 #endif
2141 	if (g_path != NULL)
2142 		free(g_path, M_TEMP);
2143 	vfs_freeopts(opts);
2144 	return (error);
2145 }
2146 
2147 /*
2148  * Find the next available prison ID.  Return the ID on success, or zero
2149  * on failure.  Also set a pointer to the allprison list entry the prison
2150  * should be inserted before.
2151  */
2152 static int
2153 get_next_prid(struct prison **insprp)
2154 {
2155 	struct prison *inspr;
2156 	int jid, maxid;
2157 
2158 	jid = lastprid % JAIL_MAX + 1;
2159 	if (TAILQ_EMPTY(&allprison) ||
2160 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2161 		/*
2162 		 * A common case is for all jails to be implicitly numbered,
2163 		 * which means they'll go on the end of the list, at least
2164 		 * for the first JAIL_MAX times.
2165 		 */
2166 		inspr = NULL;
2167 	} else {
2168 		/*
2169 		 * Take two passes through the allprison list: first starting
2170 		 * with the proposed jid, then ending with it.
2171 		 */
2172 		for (maxid = JAIL_MAX; maxid != 0; ) {
2173 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2174 				if (inspr->pr_id < jid)
2175 					continue;
2176 				if (inspr->pr_id > jid) {
2177 					/* Found an opening. */
2178 					maxid = 0;
2179 					break;
2180 				}
2181 				if (++jid > maxid) {
2182 					if (lastprid == maxid || lastprid == 0)
2183 					{
2184 						/*
2185 						 * The entire legal range
2186 						 * has been traversed
2187 						 */
2188 						return 0;
2189 					}
2190 					/* Try again from the start. */
2191 					jid = 1;
2192 					maxid = lastprid;
2193 					break;
2194 				}
2195 			}
2196 			if (inspr == NULL) {
2197 				/* Found room at the end of the list. */
2198 				break;
2199 			}
2200 		}
2201 	}
2202 	*insprp = inspr;
2203 	lastprid = jid;
2204 	return (jid);
2205 }
2206 
2207 /*
2208  * struct jail_get_args {
2209  *	struct iovec *iovp;
2210  *	unsigned int iovcnt;
2211  *	int flags;
2212  * };
2213  */
2214 int
2215 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2216 {
2217 	struct uio *auio;
2218 	int error;
2219 
2220 	/* Check that we have an even number of iovecs. */
2221 	if (uap->iovcnt & 1)
2222 		return (EINVAL);
2223 
2224 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2225 	if (error)
2226 		return (error);
2227 	error = kern_jail_get(td, auio, uap->flags);
2228 	if (error == 0)
2229 		error = copyout(auio->uio_iov, uap->iovp,
2230 		    uap->iovcnt * sizeof (struct iovec));
2231 	free(auio, M_IOV);
2232 	return (error);
2233 }
2234 
2235 int
2236 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2237 {
2238 	struct bool_flags *bf;
2239 	struct jailsys_flags *jsf;
2240 	struct prison *pr, *mypr;
2241 	struct vfsopt *opt;
2242 	struct vfsoptlist *opts;
2243 	char *errmsg, *name;
2244 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2245 	unsigned f;
2246 
2247 	if (flags & ~JAIL_GET_MASK)
2248 		return (EINVAL);
2249 
2250 	/* Get the parameter list. */
2251 	error = vfs_buildopts(optuio, &opts);
2252 	if (error)
2253 		return (error);
2254 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2255 	mypr = td->td_ucred->cr_prison;
2256 	pr = NULL;
2257 
2258 	/*
2259 	 * Find the prison specified by one of: lastjid, jid, name.
2260 	 */
2261 	sx_slock(&allprison_lock);
2262 	drflags = PD_LIST_SLOCKED;
2263 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2264 	if (error == 0) {
2265 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2266 			if (pr->pr_id > jid &&
2267 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2268 			    prison_ischild(mypr, pr)) {
2269 				mtx_lock(&pr->pr_mtx);
2270 				drflags |= PD_LOCKED;
2271 				goto found_prison;
2272 			}
2273 		}
2274 		error = ENOENT;
2275 		vfs_opterror(opts, "no jail after %d", jid);
2276 		goto done;
2277 	} else if (error != ENOENT)
2278 		goto done;
2279 
2280 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2281 	if (error == 0) {
2282 		if (jid != 0) {
2283 			pr = prison_find_child(mypr, jid);
2284 			if (pr != NULL) {
2285 				drflags |= PD_LOCKED;
2286 				if (!(prison_isalive(pr) ||
2287 				    (flags & JAIL_DYING))) {
2288 					error = ENOENT;
2289 					vfs_opterror(opts, "jail %d is dying",
2290 					    jid);
2291 					goto done;
2292 				}
2293 				goto found_prison;
2294 			}
2295 			error = ENOENT;
2296 			vfs_opterror(opts, "jail %d not found", jid);
2297 			goto done;
2298 		}
2299 	} else if (error != ENOENT)
2300 		goto done;
2301 
2302 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2303 	if (error == 0) {
2304 		if (len == 0 || name[len - 1] != '\0') {
2305 			error = EINVAL;
2306 			goto done;
2307 		}
2308 		pr = prison_find_name(mypr, name);
2309 		if (pr != NULL) {
2310 			drflags |= PD_LOCKED;
2311 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2312 				error = ENOENT;
2313 				vfs_opterror(opts, "jail \"%s\" is dying",
2314 				    name);
2315 				goto done;
2316 			}
2317 			goto found_prison;
2318 		}
2319 		error = ENOENT;
2320 		vfs_opterror(opts, "jail \"%s\" not found", name);
2321 		goto done;
2322 	} else if (error != ENOENT)
2323 		goto done;
2324 
2325 	vfs_opterror(opts, "no jail specified");
2326 	error = ENOENT;
2327 	goto done;
2328 
2329  found_prison:
2330 	/* Get the parameters of the prison. */
2331 	prison_hold(pr);
2332 	drflags |= PD_DEREF;
2333 	td->td_retval[0] = pr->pr_id;
2334 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2335 	if (error != 0 && error != ENOENT)
2336 		goto done;
2337 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2338 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2339 	if (error != 0 && error != ENOENT)
2340 		goto done;
2341 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2342 	if (error != 0 && error != ENOENT)
2343 		goto done;
2344 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2345 	    sizeof(pr->pr_cpuset->cs_id));
2346 	if (error != 0 && error != ENOENT)
2347 		goto done;
2348 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2349 	if (error != 0 && error != ENOENT)
2350 		goto done;
2351 #ifdef INET
2352 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET] + 1,
2353 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2354 	    pr_families[PR_INET].size : 0 );
2355 	if (error != 0 && error != ENOENT)
2356 		goto done;
2357 #endif
2358 #ifdef INET6
2359 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6] + 1,
2360 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2361 	    pr_families[PR_INET6].size : 0 );
2362 	if (error != 0 && error != ENOENT)
2363 		goto done;
2364 #endif
2365 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2366 	    sizeof(pr->pr_securelevel));
2367 	if (error != 0 && error != ENOENT)
2368 		goto done;
2369 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2370 	    sizeof(pr->pr_childcount));
2371 	if (error != 0 && error != ENOENT)
2372 		goto done;
2373 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2374 	    sizeof(pr->pr_childmax));
2375 	if (error != 0 && error != ENOENT)
2376 		goto done;
2377 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2378 	if (error != 0 && error != ENOENT)
2379 		goto done;
2380 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2381 	if (error != 0 && error != ENOENT)
2382 		goto done;
2383 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2384 	if (error != 0 && error != ENOENT)
2385 		goto done;
2386 #ifdef COMPAT_FREEBSD32
2387 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2388 		uint32_t hid32 = pr->pr_hostid;
2389 
2390 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2391 	} else
2392 #endif
2393 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2394 	    sizeof(pr->pr_hostid));
2395 	if (error != 0 && error != ENOENT)
2396 		goto done;
2397 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2398 	    sizeof(pr->pr_enforce_statfs));
2399 	if (error != 0 && error != ENOENT)
2400 		goto done;
2401 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2402 	    sizeof(pr->pr_devfs_rsnum));
2403 	if (error != 0 && error != ENOENT)
2404 		goto done;
2405 	for (bf = pr_flag_bool;
2406 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2407 	     bf++) {
2408 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2409 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2410 		if (error != 0 && error != ENOENT)
2411 			goto done;
2412 		i = !i;
2413 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2414 		if (error != 0 && error != ENOENT)
2415 			goto done;
2416 	}
2417 	for (jsf = pr_flag_jailsys;
2418 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2419 	     jsf++) {
2420 		f = pr->pr_flags & (jsf->disable | jsf->new);
2421 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2422 		    : (f == jsf->new) ? JAIL_SYS_NEW
2423 		    : JAIL_SYS_INHERIT;
2424 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2425 		if (error != 0 && error != ENOENT)
2426 			goto done;
2427 	}
2428 	for (bf = pr_flag_allow;
2429 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2430 		atomic_load_int(&bf->flag) != 0;
2431 	     bf++) {
2432 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2433 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2434 		if (error != 0 && error != ENOENT)
2435 			goto done;
2436 		i = !i;
2437 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2438 		if (error != 0 && error != ENOENT)
2439 			goto done;
2440 	}
2441 	i = !prison_isalive(pr);
2442 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2443 	if (error != 0 && error != ENOENT)
2444 		goto done;
2445 	i = !i;
2446 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2447 	if (error != 0 && error != ENOENT)
2448 		goto done;
2449 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2450 	    sizeof(pr->pr_osreldate));
2451 	if (error != 0 && error != ENOENT)
2452 		goto done;
2453 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2454 	if (error != 0 && error != ENOENT)
2455 		goto done;
2456 
2457 	/* Get the module parameters. */
2458 	mtx_unlock(&pr->pr_mtx);
2459 	drflags &= ~PD_LOCKED;
2460 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2461 	if (error)
2462 		goto done;
2463 	prison_deref(pr, drflags);
2464 	pr = NULL;
2465 	drflags = 0;
2466 
2467 	/* By now, all parameters should have been noted. */
2468 	TAILQ_FOREACH(opt, opts, link) {
2469 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2470 			error = EINVAL;
2471 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2472 			goto done;
2473 		}
2474 	}
2475 
2476 	/* Write the fetched parameters back to userspace. */
2477 	error = 0;
2478 	TAILQ_FOREACH(opt, opts, link) {
2479 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2480 			pos = 2 * opt->pos + 1;
2481 			optuio->uio_iov[pos].iov_len = opt->len;
2482 			if (opt->value != NULL) {
2483 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2484 					bcopy(opt->value,
2485 					    optuio->uio_iov[pos].iov_base,
2486 					    opt->len);
2487 				} else {
2488 					error = copyout(opt->value,
2489 					    optuio->uio_iov[pos].iov_base,
2490 					    opt->len);
2491 					if (error)
2492 						break;
2493 				}
2494 			}
2495 		}
2496 	}
2497 
2498  done:
2499 	/* Release any temporary prison holds and/or locks. */
2500 	if (pr != NULL)
2501 		prison_deref(pr, drflags);
2502 	else if (drflags & PD_LIST_SLOCKED)
2503 		sx_sunlock(&allprison_lock);
2504 	if (error && errmsg_pos >= 0) {
2505 		/* Write the error message back to userspace. */
2506 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2507 		errmsg_pos = 2 * errmsg_pos + 1;
2508 		if (errmsg_len > 0) {
2509 			if (optuio->uio_segflg == UIO_SYSSPACE)
2510 				bcopy(errmsg,
2511 				    optuio->uio_iov[errmsg_pos].iov_base,
2512 				    errmsg_len);
2513 			else
2514 				copyout(errmsg,
2515 				    optuio->uio_iov[errmsg_pos].iov_base,
2516 				    errmsg_len);
2517 		}
2518 	}
2519 	vfs_freeopts(opts);
2520 	return (error);
2521 }
2522 
2523 /*
2524  * struct jail_remove_args {
2525  *	int jid;
2526  * };
2527  */
2528 int
2529 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2530 {
2531 	struct prison *pr;
2532 	int error;
2533 
2534 	error = priv_check(td, PRIV_JAIL_REMOVE);
2535 	if (error)
2536 		return (error);
2537 
2538 	sx_xlock(&allprison_lock);
2539 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2540 	if (pr == NULL) {
2541 		sx_xunlock(&allprison_lock);
2542 		return (EINVAL);
2543 	}
2544 	if (!prison_isalive(pr)) {
2545 		/* Silently ignore already-dying prisons. */
2546 		mtx_unlock(&pr->pr_mtx);
2547 		sx_xunlock(&allprison_lock);
2548 		return (0);
2549 	}
2550 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2551 	return (0);
2552 }
2553 
2554 /*
2555  * struct jail_attach_args {
2556  *	int jid;
2557  * };
2558  */
2559 int
2560 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2561 {
2562 	struct prison *pr;
2563 	int error;
2564 
2565 	error = priv_check(td, PRIV_JAIL_ATTACH);
2566 	if (error)
2567 		return (error);
2568 
2569 	sx_slock(&allprison_lock);
2570 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2571 	if (pr == NULL) {
2572 		sx_sunlock(&allprison_lock);
2573 		return (EINVAL);
2574 	}
2575 
2576 	/* Do not allow a process to attach to a prison that is not alive. */
2577 	if (!prison_isalive(pr)) {
2578 		mtx_unlock(&pr->pr_mtx);
2579 		sx_sunlock(&allprison_lock);
2580 		return (EINVAL);
2581 	}
2582 
2583 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2584 }
2585 
2586 static int
2587 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2588 {
2589 	struct proc *p;
2590 	struct ucred *newcred, *oldcred;
2591 	int error;
2592 
2593 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2594 	sx_assert(&allprison_lock, SX_LOCKED);
2595 	drflags &= PD_LOCK_FLAGS;
2596 	/*
2597 	 * XXX: Note that there is a slight race here if two threads
2598 	 * in the same privileged process attempt to attach to two
2599 	 * different jails at the same time.  It is important for
2600 	 * user processes not to do this, or they might end up with
2601 	 * a process root from one prison, but attached to the jail
2602 	 * of another.
2603 	 */
2604 	prison_hold(pr);
2605 	refcount_acquire(&pr->pr_uref);
2606 	drflags |= PD_DEREF | PD_DEUREF;
2607 	mtx_unlock(&pr->pr_mtx);
2608 	drflags &= ~PD_LOCKED;
2609 
2610 	/* Let modules do whatever they need to prepare for attaching. */
2611 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2612 	if (error) {
2613 		prison_deref(pr, drflags);
2614 		return (error);
2615 	}
2616 	sx_unlock(&allprison_lock);
2617 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2618 
2619 	/*
2620 	 * Reparent the newly attached process to this jail.
2621 	 */
2622 	p = td->td_proc;
2623 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2624 	if (error)
2625 		goto e_revert_osd;
2626 
2627 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2628 	if ((error = change_dir(pr->pr_root, td)) != 0)
2629 		goto e_unlock;
2630 #ifdef MAC
2631 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2632 		goto e_unlock;
2633 #endif
2634 	VOP_UNLOCK(pr->pr_root);
2635 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2636 		goto e_revert_osd;
2637 
2638 	newcred = crget();
2639 	PROC_LOCK(p);
2640 	oldcred = crcopysafe(p, newcred);
2641 	newcred->cr_prison = pr;
2642 	proc_set_cred(p, newcred);
2643 	setsugid(p);
2644 #ifdef RACCT
2645 	racct_proc_ucred_changed(p, oldcred, newcred);
2646 	crhold(newcred);
2647 #endif
2648 	PROC_UNLOCK(p);
2649 #ifdef RCTL
2650 	rctl_proc_ucred_changed(p, newcred);
2651 	crfree(newcred);
2652 #endif
2653 	prison_proc_relink(oldcred->cr_prison, pr, p);
2654 	prison_deref(oldcred->cr_prison, drflags);
2655 	crfree(oldcred);
2656 
2657 	/*
2658 	 * If the prison was killed while changing credentials, die along
2659 	 * with it.
2660 	 */
2661 	if (!prison_isalive(pr)) {
2662 		PROC_LOCK(p);
2663 		kern_psignal(p, SIGKILL);
2664 		PROC_UNLOCK(p);
2665 	}
2666 
2667 	return (0);
2668 
2669  e_unlock:
2670 	VOP_UNLOCK(pr->pr_root);
2671  e_revert_osd:
2672 	/* Tell modules this thread is still in its old jail after all. */
2673 	sx_slock(&allprison_lock);
2674 	drflags |= PD_LIST_SLOCKED;
2675 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2676 	prison_deref(pr, drflags);
2677 	return (error);
2678 }
2679 
2680 /*
2681  * Returns a locked prison instance, or NULL on failure.
2682  */
2683 struct prison *
2684 prison_find(int prid)
2685 {
2686 	struct prison *pr;
2687 
2688 	sx_assert(&allprison_lock, SX_LOCKED);
2689 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2690 		if (pr->pr_id < prid)
2691 			continue;
2692 		if (pr->pr_id > prid)
2693 			break;
2694 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2695 		mtx_lock(&pr->pr_mtx);
2696 		return (pr);
2697 	}
2698 	return (NULL);
2699 }
2700 
2701 /*
2702  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2703  */
2704 struct prison *
2705 prison_find_child(struct prison *mypr, int prid)
2706 {
2707 	struct prison *pr;
2708 	int descend;
2709 
2710 	sx_assert(&allprison_lock, SX_LOCKED);
2711 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2712 		if (pr->pr_id == prid) {
2713 			KASSERT(prison_isvalid(pr),
2714 			    ("Found invalid prison %p", pr));
2715 			mtx_lock(&pr->pr_mtx);
2716 			return (pr);
2717 		}
2718 	}
2719 	return (NULL);
2720 }
2721 
2722 /*
2723  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2724  */
2725 struct prison *
2726 prison_find_name(struct prison *mypr, const char *name)
2727 {
2728 	struct prison *pr, *deadpr;
2729 	size_t mylen;
2730 	int descend;
2731 
2732 	sx_assert(&allprison_lock, SX_LOCKED);
2733 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2734 	deadpr = NULL;
2735 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2736 		if (!strcmp(pr->pr_name + mylen, name)) {
2737 			KASSERT(prison_isvalid(pr),
2738 			    ("Found invalid prison %p", pr));
2739 			if (prison_isalive(pr)) {
2740 				mtx_lock(&pr->pr_mtx);
2741 				return (pr);
2742 			}
2743 			deadpr = pr;
2744 		}
2745 	}
2746 	/* There was no valid prison - perhaps there was a dying one. */
2747 	if (deadpr != NULL)
2748 		mtx_lock(&deadpr->pr_mtx);
2749 	return (deadpr);
2750 }
2751 
2752 /*
2753  * See if a prison has the specific flag set.  The prison should be locked,
2754  * unless checking for flags that are only set at jail creation (such as
2755  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2756  * to any other prison data.
2757  */
2758 int
2759 prison_flag(struct ucred *cred, unsigned flag)
2760 {
2761 
2762 	return (cred->cr_prison->pr_flags & flag);
2763 }
2764 
2765 int
2766 prison_allow(struct ucred *cred, unsigned flag)
2767 {
2768 
2769 	return ((cred->cr_prison->pr_allow & flag) != 0);
2770 }
2771 
2772 /*
2773  * Hold a prison reference, by incrementing pr_ref.  It is generally
2774  * an error to hold a prison that does not already have a reference.
2775  * A prison record will remain valid as long as it has at least one
2776  * reference, and will not be removed as long as either the prison
2777  * mutex or the allprison lock is held (allprison_lock may be shared).
2778  */
2779 void
2780 prison_hold_locked(struct prison *pr)
2781 {
2782 
2783 	/* Locking is no longer required. */
2784 	prison_hold(pr);
2785 }
2786 
2787 void
2788 prison_hold(struct prison *pr)
2789 {
2790 #ifdef INVARIANTS
2791 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2792 
2793 	KASSERT(was_valid,
2794 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2795 #else
2796 	refcount_acquire(&pr->pr_ref);
2797 #endif
2798 }
2799 
2800 /*
2801  * Remove a prison reference.  If that was the last reference, the
2802  * prison will be removed (at a later time).
2803  */
2804 void
2805 prison_free_locked(struct prison *pr)
2806 {
2807 
2808 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2809 	/*
2810 	 * Locking is no longer required, but unlock because the caller
2811 	 * expects it.
2812 	 */
2813 	mtx_unlock(&pr->pr_mtx);
2814 	prison_free(pr);
2815 }
2816 
2817 void
2818 prison_free(struct prison *pr)
2819 {
2820 
2821 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2822 	    ("Trying to free dead prison %p (jid=%d).",
2823 	     pr, pr->pr_id));
2824 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2825 		/*
2826 		 * Don't remove the last reference in this context,
2827 		 * in case there are locks held.
2828 		 */
2829 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2830 	}
2831 }
2832 
2833 static void
2834 prison_free_not_last(struct prison *pr)
2835 {
2836 #ifdef INVARIANTS
2837 	int lastref;
2838 
2839 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2840 	    ("Trying to free dead prison %p (jid=%d).",
2841 	     pr, pr->pr_id));
2842 	lastref = refcount_release(&pr->pr_ref);
2843 	KASSERT(!lastref,
2844 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2845 	     pr, pr->pr_id));
2846 #else
2847 	refcount_release(&pr->pr_ref);
2848 #endif
2849 }
2850 
2851 /*
2852  * Hold a prison for user visibility, by incrementing pr_uref.
2853  * It is generally an error to hold a prison that isn't already
2854  * user-visible, except through the jail system calls.  It is also
2855  * an error to hold an invalid prison.  A prison record will remain
2856  * alive as long as it has at least one user reference, and will not
2857  * be set to the dying state until the prison mutex and allprison_lock
2858  * are both freed.
2859  */
2860 void
2861 prison_proc_hold(struct prison *pr)
2862 {
2863 #ifdef INVARIANTS
2864 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2865 
2866 	KASSERT(was_alive,
2867 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2868 #else
2869 	refcount_acquire(&pr->pr_uref);
2870 #endif
2871 }
2872 
2873 /*
2874  * Remove a prison user reference.  If it was the last reference, the
2875  * prison will be considered "dying", and may be removed once all of
2876  * its references are dropped.
2877  */
2878 void
2879 prison_proc_free(struct prison *pr)
2880 {
2881 
2882 	/*
2883 	 * Locking is only required when releasing the last reference.
2884 	 * This allows assurance that a locked prison will remain alive
2885 	 * until it is unlocked.
2886 	 */
2887 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2888 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2889 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2890 		/*
2891 		 * Don't remove the last user reference in this context,
2892 		 * which is expected to be a process that is not only locked,
2893 		 * but also half dead.  Add a reference so any calls to
2894 		 * prison_free() won't re-submit the task.
2895 		 */
2896 		prison_hold(pr);
2897 		mtx_lock(&pr->pr_mtx);
2898 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2899 		    ("Redundant last reference in prison_proc_free (jid=%d)",
2900 		     pr->pr_id));
2901 		pr->pr_flags |= PR_COMPLETE_PROC;
2902 		mtx_unlock(&pr->pr_mtx);
2903 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2904 	}
2905 }
2906 
2907 static void
2908 prison_proc_free_not_last(struct prison *pr)
2909 {
2910 #ifdef INVARIANTS
2911 	int lastref;
2912 
2913 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2914 	    ("Trying to free dead prison %p (jid=%d).",
2915 	     pr, pr->pr_id));
2916 	lastref = refcount_release(&pr->pr_uref);
2917 	KASSERT(!lastref,
2918 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2919 	     pr, pr->pr_id));
2920 #else
2921 	refcount_release(&pr->pr_uref);
2922 #endif
2923 }
2924 
2925 void
2926 prison_proc_link(struct prison *pr, struct proc *p)
2927 {
2928 
2929 	sx_assert(&allproc_lock, SA_XLOCKED);
2930 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
2931 }
2932 
2933 void
2934 prison_proc_unlink(struct prison *pr, struct proc *p)
2935 {
2936 
2937 	sx_assert(&allproc_lock, SA_XLOCKED);
2938 	LIST_REMOVE(p, p_jaillist);
2939 }
2940 
2941 static void
2942 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
2943 {
2944 
2945 	sx_xlock(&allproc_lock);
2946 	prison_proc_unlink(opr, p);
2947 	prison_proc_link(npr, p);
2948 	sx_xunlock(&allproc_lock);
2949 }
2950 
2951 /*
2952  * Complete a call to either prison_free or prison_proc_free.
2953  */
2954 static void
2955 prison_complete(void *context, int pending)
2956 {
2957 	struct prison *pr = context;
2958 	int drflags;
2959 
2960 	/*
2961 	 * This could be called to release the last reference, or the last
2962 	 * user reference (plus the reference held in prison_proc_free).
2963 	 */
2964 	drflags = prison_lock_xlock(pr, PD_DEREF);
2965 	if (pr->pr_flags & PR_COMPLETE_PROC) {
2966 		pr->pr_flags &= ~PR_COMPLETE_PROC;
2967 		drflags |= PD_DEUREF;
2968 	}
2969 	prison_deref(pr, drflags);
2970 }
2971 
2972 static void
2973 prison_kill_processes_cb(struct proc *p, void *arg __unused)
2974 {
2975 
2976 	kern_psignal(p, SIGKILL);
2977 }
2978 
2979 /*
2980  * Note the iteration does not guarantee acting on all processes.
2981  * Most notably there may be fork or jail_attach in progress.
2982  */
2983 void
2984 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
2985     void *cbarg)
2986 {
2987 	struct prison *ppr;
2988 	struct proc *p;
2989 
2990 	if (atomic_load_int(&pr->pr_childcount) == 0) {
2991 		sx_slock(&allproc_lock);
2992 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
2993 			if (p->p_state == PRS_NEW)
2994 				continue;
2995 			PROC_LOCK(p);
2996 			cb(p, cbarg);
2997 			PROC_UNLOCK(p);
2998 		}
2999 		sx_sunlock(&allproc_lock);
3000 		if (atomic_load_int(&pr->pr_childcount) == 0)
3001 			return;
3002 		/*
3003 		 * Some jails popped up during the iteration, fall through to a
3004 		 * system-wide search.
3005 		 */
3006 	}
3007 
3008 	sx_slock(&allproc_lock);
3009 	FOREACH_PROC_IN_SYSTEM(p) {
3010 		PROC_LOCK(p);
3011 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3012 			for (ppr = p->p_ucred->cr_prison;
3013 			    ppr != &prison0;
3014 			    ppr = ppr->pr_parent) {
3015 				if (ppr == pr) {
3016 					cb(p, cbarg);
3017 					break;
3018 				}
3019 			}
3020 		}
3021 		PROC_UNLOCK(p);
3022 	}
3023 	sx_sunlock(&allproc_lock);
3024 }
3025 
3026 /*
3027  * Remove a prison reference and/or user reference (usually).
3028  * This assumes context that allows sleeping (for allprison_lock),
3029  * with no non-sleeping locks held, except perhaps the prison itself.
3030  * If there are no more references, release and delist the prison.
3031  * On completion, the prison lock and the allprison lock are both
3032  * unlocked.
3033  */
3034 static void
3035 prison_deref(struct prison *pr, int flags)
3036 {
3037 	struct prisonlist freeprison;
3038 	struct prison *killpr, *rpr, *ppr, *tpr;
3039 
3040 	killpr = NULL;
3041 	TAILQ_INIT(&freeprison);
3042 	/*
3043 	 * Release this prison as requested, which may cause its parent
3044 	 * to be released, and then maybe its grandparent, etc.
3045 	 */
3046 	for (;;) {
3047 		if (flags & PD_KILL) {
3048 			/* Kill the prison and its descendents. */
3049 			KASSERT(pr != &prison0,
3050 			    ("prison_deref trying to kill prison0"));
3051 			if (!(flags & PD_DEREF)) {
3052 				prison_hold(pr);
3053 				flags |= PD_DEREF;
3054 			}
3055 			flags = prison_lock_xlock(pr, flags);
3056 			prison_deref_kill(pr, &freeprison);
3057 		}
3058 		if (flags & PD_DEUREF) {
3059 			/* Drop a user reference. */
3060 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3061 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3062 			     pr->pr_id));
3063 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3064 				if (!(flags & PD_DEREF)) {
3065 					prison_hold(pr);
3066 					flags |= PD_DEREF;
3067 				}
3068 				flags = prison_lock_xlock(pr, flags);
3069 				if (refcount_release(&pr->pr_uref) &&
3070 				    pr->pr_state == PRISON_STATE_ALIVE) {
3071 					/*
3072 					 * When the last user references goes,
3073 					 * this becomes a dying prison.
3074 					 */
3075 					KASSERT(
3076 					    refcount_load(&prison0.pr_uref) > 0,
3077 					    ("prison0 pr_uref=0"));
3078 					pr->pr_state = PRISON_STATE_DYING;
3079 					mtx_unlock(&pr->pr_mtx);
3080 					flags &= ~PD_LOCKED;
3081 					prison_cleanup(pr);
3082 				}
3083 			}
3084 		}
3085 		if (flags & PD_KILL) {
3086 			/*
3087 			 * Any remaining user references are probably processes
3088 			 * that need to be killed, either in this prison or its
3089 			 * descendants.
3090 			 */
3091 			if (refcount_load(&pr->pr_uref) > 0)
3092 				killpr = pr;
3093 			/* Make sure the parent prison doesn't get killed. */
3094 			flags &= ~PD_KILL;
3095 		}
3096 		if (flags & PD_DEREF) {
3097 			/* Drop a reference. */
3098 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3099 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3100 			     pr->pr_id));
3101 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3102 				flags = prison_lock_xlock(pr, flags);
3103 				if (refcount_release(&pr->pr_ref)) {
3104 					/*
3105 					 * When the last reference goes,
3106 					 * unlink the prison and set it aside.
3107 					 */
3108 					KASSERT(
3109 					    refcount_load(&pr->pr_uref) == 0,
3110 					    ("prison_deref: last ref, "
3111 					     "but still has %d urefs (jid=%d)",
3112 					     pr->pr_uref, pr->pr_id));
3113 					KASSERT(
3114 					    refcount_load(&prison0.pr_ref) != 0,
3115 					    ("prison0 pr_ref=0"));
3116 					pr->pr_state = PRISON_STATE_INVALID;
3117 					TAILQ_REMOVE(&allprison, pr, pr_list);
3118 					LIST_REMOVE(pr, pr_sibling);
3119 					TAILQ_INSERT_TAIL(&freeprison, pr,
3120 					    pr_list);
3121 					for (ppr = pr->pr_parent;
3122 					     ppr != NULL;
3123 					     ppr = ppr->pr_parent)
3124 						ppr->pr_childcount--;
3125 					/*
3126 					 * Removing a prison frees references
3127 					 * from its parent.
3128 					 */
3129 					mtx_unlock(&pr->pr_mtx);
3130 					flags &= ~PD_LOCKED;
3131 					pr = pr->pr_parent;
3132 					flags |= PD_DEREF | PD_DEUREF;
3133 					continue;
3134 				}
3135 			}
3136 		}
3137 		break;
3138 	}
3139 
3140 	/* Release all the prison locks. */
3141 	if (flags & PD_LOCKED)
3142 		mtx_unlock(&pr->pr_mtx);
3143 	if (flags & PD_LIST_SLOCKED)
3144 		sx_sunlock(&allprison_lock);
3145 	else if (flags & PD_LIST_XLOCKED)
3146 		sx_xunlock(&allprison_lock);
3147 
3148 	/* Kill any processes attached to a killed prison. */
3149 	if (killpr != NULL)
3150 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3151 
3152 	/*
3153 	 * Finish removing any unreferenced prisons, which couldn't happen
3154 	 * while allprison_lock was held (to avoid a LOR on vrele).
3155 	 */
3156 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3157 #ifdef VIMAGE
3158 		if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3159 			vnet_destroy(rpr->pr_vnet);
3160 #endif
3161 		if (rpr->pr_root != NULL)
3162 			vrele(rpr->pr_root);
3163 		mtx_destroy(&rpr->pr_mtx);
3164 #ifdef INET
3165 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3166 #endif
3167 #ifdef INET6
3168 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3169 #endif
3170 		if (rpr->pr_cpuset != NULL)
3171 			cpuset_rel(rpr->pr_cpuset);
3172 		osd_jail_exit(rpr);
3173 #ifdef RACCT
3174 		if (racct_enable)
3175 			prison_racct_detach(rpr);
3176 #endif
3177 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3178 		free(rpr, M_PRISON);
3179 	}
3180 }
3181 
3182 /*
3183  * Kill the prison and its descendants.  Mark them as dying, clear the
3184  * persist flag, and call module remove methods.
3185  */
3186 static void
3187 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3188 {
3189 	struct prison *cpr, *ppr, *rpr;
3190 	bool descend;
3191 
3192 	/*
3193 	 * Unlike the descendants, the target prison can be killed
3194 	 * even if it is currently dying.  This is useful for failed
3195 	 * creation in jail_set(2).
3196 	 */
3197 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3198 	    ("Trying to kill dead prison %p (jid=%d).",
3199 	     pr, pr->pr_id));
3200 	refcount_acquire(&pr->pr_uref);
3201 	pr->pr_state = PRISON_STATE_DYING;
3202 	mtx_unlock(&pr->pr_mtx);
3203 
3204 	rpr = NULL;
3205 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3206 		if (descend) {
3207 			if (!prison_isalive(cpr)) {
3208 				descend = false;
3209 				continue;
3210 			}
3211 			prison_hold(cpr);
3212 			prison_proc_hold(cpr);
3213 			mtx_lock(&cpr->pr_mtx);
3214 			cpr->pr_state = PRISON_STATE_DYING;
3215 			cpr->pr_flags |= PR_REMOVE;
3216 			mtx_unlock(&cpr->pr_mtx);
3217 			continue;
3218 		}
3219 		if (!(cpr->pr_flags & PR_REMOVE))
3220 			continue;
3221 		prison_cleanup(cpr);
3222 		mtx_lock(&cpr->pr_mtx);
3223 		cpr->pr_flags &= ~PR_REMOVE;
3224 		if (cpr->pr_flags & PR_PERSIST) {
3225 			cpr->pr_flags &= ~PR_PERSIST;
3226 			prison_proc_free_not_last(cpr);
3227 			prison_free_not_last(cpr);
3228 		}
3229 		(void)refcount_release(&cpr->pr_uref);
3230 		if (refcount_release(&cpr->pr_ref)) {
3231 			/*
3232 			 * When the last reference goes, unlink the prison
3233 			 * and set it aside for prison_deref() to handle.
3234 			 * Delay unlinking the sibling list to keep the loop
3235 			 * safe.
3236 			 */
3237 			if (rpr != NULL)
3238 				LIST_REMOVE(rpr, pr_sibling);
3239 			rpr = cpr;
3240 			rpr->pr_state = PRISON_STATE_INVALID;
3241 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3242 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3243 			/*
3244 			 * Removing a prison frees references from its parent.
3245 			 */
3246 			ppr = rpr->pr_parent;
3247 			prison_proc_free_not_last(ppr);
3248 			prison_free_not_last(ppr);
3249 			for (; ppr != NULL; ppr = ppr->pr_parent)
3250 				ppr->pr_childcount--;
3251 		}
3252 		mtx_unlock(&cpr->pr_mtx);
3253 	}
3254 	if (rpr != NULL)
3255 		LIST_REMOVE(rpr, pr_sibling);
3256 
3257 	prison_cleanup(pr);
3258 	mtx_lock(&pr->pr_mtx);
3259 	if (pr->pr_flags & PR_PERSIST) {
3260 		pr->pr_flags &= ~PR_PERSIST;
3261 		prison_proc_free_not_last(pr);
3262 		prison_free_not_last(pr);
3263 	}
3264 	(void)refcount_release(&pr->pr_uref);
3265 }
3266 
3267 /*
3268  * Given the current locking state in the flags, make sure allprison_lock
3269  * is held exclusive, and the prison is locked.  Return flags indicating
3270  * the new state.
3271  */
3272 static int
3273 prison_lock_xlock(struct prison *pr, int flags)
3274 {
3275 
3276 	if (!(flags & PD_LIST_XLOCKED)) {
3277 		/*
3278 		 * Get allprison_lock, which may be an upgrade,
3279 		 * and may require unlocking the prison.
3280 		 */
3281 		if (flags & PD_LOCKED) {
3282 			mtx_unlock(&pr->pr_mtx);
3283 			flags &= ~PD_LOCKED;
3284 		}
3285 		if (flags & PD_LIST_SLOCKED) {
3286 			if (!sx_try_upgrade(&allprison_lock)) {
3287 				sx_sunlock(&allprison_lock);
3288 				sx_xlock(&allprison_lock);
3289 			}
3290 			flags &= ~PD_LIST_SLOCKED;
3291 		} else
3292 			sx_xlock(&allprison_lock);
3293 		flags |= PD_LIST_XLOCKED;
3294 	}
3295 	if (!(flags & PD_LOCKED)) {
3296 		/* Lock the prison mutex. */
3297 		mtx_lock(&pr->pr_mtx);
3298 		flags |= PD_LOCKED;
3299 	}
3300 	return flags;
3301 }
3302 
3303 /*
3304  * Release a prison's resources when it starts dying (when the last user
3305  * reference is dropped, or when it is killed).
3306  */
3307 static void
3308 prison_cleanup(struct prison *pr)
3309 {
3310 	sx_assert(&allprison_lock, SA_XLOCKED);
3311 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3312 	shm_remove_prison(pr);
3313 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3314 }
3315 
3316 /*
3317  * Set or clear a permission bit in the pr_allow field, passing restrictions
3318  * (cleared permission) down to child jails.
3319  */
3320 void
3321 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3322 {
3323 	struct prison *pr;
3324 
3325 	pr = cred->cr_prison;
3326 	sx_slock(&allprison_lock);
3327 	mtx_lock(&pr->pr_mtx);
3328 	prison_set_allow_locked(pr, flag, enable);
3329 	mtx_unlock(&pr->pr_mtx);
3330 	sx_sunlock(&allprison_lock);
3331 }
3332 
3333 static void
3334 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3335 {
3336 	struct prison *cpr;
3337 	int descend;
3338 
3339 	if (enable != 0)
3340 		pr->pr_allow |= flag;
3341 	else {
3342 		pr->pr_allow &= ~flag;
3343 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3344 			cpr->pr_allow &= ~flag;
3345 	}
3346 }
3347 
3348 /*
3349  * Check if a jail supports the given address family.
3350  *
3351  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3352  * if not.
3353  */
3354 int
3355 prison_check_af(struct ucred *cred, int af)
3356 {
3357 	struct prison *pr;
3358 	int error;
3359 
3360 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3361 
3362 	pr = cred->cr_prison;
3363 #ifdef VIMAGE
3364 	/* Prisons with their own network stack are not limited. */
3365 	if (prison_owns_vnet(cred))
3366 		return (0);
3367 #endif
3368 
3369 	error = 0;
3370 	switch (af)
3371 	{
3372 #ifdef INET
3373 	case AF_INET:
3374 		if (pr->pr_flags & PR_IP4)
3375 		{
3376 			mtx_lock(&pr->pr_mtx);
3377 			if ((pr->pr_flags & PR_IP4) &&
3378 			    pr->pr_addrs[PR_INET] == NULL)
3379 				error = EAFNOSUPPORT;
3380 			mtx_unlock(&pr->pr_mtx);
3381 		}
3382 		break;
3383 #endif
3384 #ifdef INET6
3385 	case AF_INET6:
3386 		if (pr->pr_flags & PR_IP6)
3387 		{
3388 			mtx_lock(&pr->pr_mtx);
3389 			if ((pr->pr_flags & PR_IP6) &&
3390 			    pr->pr_addrs[PR_INET6] == NULL)
3391 				error = EAFNOSUPPORT;
3392 			mtx_unlock(&pr->pr_mtx);
3393 		}
3394 		break;
3395 #endif
3396 	case AF_LOCAL:
3397 	case AF_ROUTE:
3398 		break;
3399 	default:
3400 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3401 			error = EAFNOSUPPORT;
3402 	}
3403 	return (error);
3404 }
3405 
3406 /*
3407  * Check if given address belongs to the jail referenced by cred (wrapper to
3408  * prison_check_ip[46]).
3409  *
3410  * Returns 0 if jail doesn't restrict the address family or if address belongs
3411  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3412  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3413  */
3414 int
3415 prison_if(struct ucred *cred, const struct sockaddr *sa)
3416 {
3417 #ifdef INET
3418 	const struct sockaddr_in *sai;
3419 #endif
3420 #ifdef INET6
3421 	const struct sockaddr_in6 *sai6;
3422 #endif
3423 	int error;
3424 
3425 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3426 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3427 
3428 #ifdef VIMAGE
3429 	if (prison_owns_vnet(cred))
3430 		return (0);
3431 #endif
3432 
3433 	error = 0;
3434 	switch (sa->sa_family)
3435 	{
3436 #ifdef INET
3437 	case AF_INET:
3438 		sai = (const struct sockaddr_in *)sa;
3439 		error = prison_check_ip4(cred, &sai->sin_addr);
3440 		break;
3441 #endif
3442 #ifdef INET6
3443 	case AF_INET6:
3444 		sai6 = (const struct sockaddr_in6 *)sa;
3445 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3446 		break;
3447 #endif
3448 	default:
3449 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3450 			error = EAFNOSUPPORT;
3451 	}
3452 	return (error);
3453 }
3454 
3455 /*
3456  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3457  */
3458 int
3459 prison_check(struct ucred *cred1, struct ucred *cred2)
3460 {
3461 
3462 	return ((cred1->cr_prison == cred2->cr_prison ||
3463 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3464 }
3465 
3466 /*
3467  * Return 1 if p2 is a child of p1, otherwise 0.
3468  */
3469 int
3470 prison_ischild(struct prison *pr1, struct prison *pr2)
3471 {
3472 
3473 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3474 		if (pr1 == pr2)
3475 			return (1);
3476 	return (0);
3477 }
3478 
3479 /*
3480  * Return true if the prison is currently alive.  A prison is alive if it
3481  * holds user references and it isn't being removed.
3482  */
3483 bool
3484 prison_isalive(const struct prison *pr)
3485 {
3486 
3487 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3488 		return (false);
3489 	return (true);
3490 }
3491 
3492 /*
3493  * Return true if the prison is currently valid.  A prison is valid if it has
3494  * been fully created, and is not being destroyed.  Note that dying prisons
3495  * are still considered valid.  Invalid prisons won't be found under normal
3496  * circumstances, as they're only put in that state by functions that have
3497  * an exclusive hold on allprison_lock.
3498  */
3499 bool
3500 prison_isvalid(struct prison *pr)
3501 {
3502 
3503 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3504 		return (false);
3505 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3506 		return (false);
3507 	return (true);
3508 }
3509 
3510 /*
3511  * Return 1 if the passed credential is in a jail and that jail does not
3512  * have its own virtual network stack, otherwise 0.
3513  */
3514 int
3515 jailed_without_vnet(struct ucred *cred)
3516 {
3517 
3518 	if (!jailed(cred))
3519 		return (0);
3520 #ifdef VIMAGE
3521 	if (prison_owns_vnet(cred))
3522 		return (0);
3523 #endif
3524 
3525 	return (1);
3526 }
3527 
3528 /*
3529  * Return the correct hostname (domainname, et al) for the passed credential.
3530  */
3531 void
3532 getcredhostname(struct ucred *cred, char *buf, size_t size)
3533 {
3534 	struct prison *pr;
3535 
3536 	/*
3537 	 * A NULL credential can be used to shortcut to the physical
3538 	 * system's hostname.
3539 	 */
3540 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3541 	mtx_lock(&pr->pr_mtx);
3542 	strlcpy(buf, pr->pr_hostname, size);
3543 	mtx_unlock(&pr->pr_mtx);
3544 }
3545 
3546 void
3547 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3548 {
3549 
3550 	mtx_lock(&cred->cr_prison->pr_mtx);
3551 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3552 	mtx_unlock(&cred->cr_prison->pr_mtx);
3553 }
3554 
3555 void
3556 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3557 {
3558 
3559 	mtx_lock(&cred->cr_prison->pr_mtx);
3560 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3561 	mtx_unlock(&cred->cr_prison->pr_mtx);
3562 }
3563 
3564 void
3565 getcredhostid(struct ucred *cred, unsigned long *hostid)
3566 {
3567 
3568 	mtx_lock(&cred->cr_prison->pr_mtx);
3569 	*hostid = cred->cr_prison->pr_hostid;
3570 	mtx_unlock(&cred->cr_prison->pr_mtx);
3571 }
3572 
3573 void
3574 getjailname(struct ucred *cred, char *name, size_t len)
3575 {
3576 
3577 	mtx_lock(&cred->cr_prison->pr_mtx);
3578 	strlcpy(name, cred->cr_prison->pr_name, len);
3579 	mtx_unlock(&cred->cr_prison->pr_mtx);
3580 }
3581 
3582 #ifdef VIMAGE
3583 /*
3584  * Determine whether the prison represented by cred owns
3585  * its vnet rather than having it inherited.
3586  *
3587  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3588  */
3589 int
3590 prison_owns_vnet(struct ucred *cred)
3591 {
3592 
3593 	/*
3594 	 * vnets cannot be added/removed after jail creation,
3595 	 * so no need to lock here.
3596 	 */
3597 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3598 }
3599 #endif
3600 
3601 /*
3602  * Determine whether the subject represented by cred can "see"
3603  * status of a mount point.
3604  * Returns: 0 for permitted, ENOENT otherwise.
3605  * XXX: This function should be called cr_canseemount() and should be
3606  *      placed in kern_prot.c.
3607  */
3608 int
3609 prison_canseemount(struct ucred *cred, struct mount *mp)
3610 {
3611 	struct prison *pr;
3612 	struct statfs *sp;
3613 	size_t len;
3614 
3615 	pr = cred->cr_prison;
3616 	if (pr->pr_enforce_statfs == 0)
3617 		return (0);
3618 	if (pr->pr_root->v_mount == mp)
3619 		return (0);
3620 	if (pr->pr_enforce_statfs == 2)
3621 		return (ENOENT);
3622 	/*
3623 	 * If jail's chroot directory is set to "/" we should be able to see
3624 	 * all mount-points from inside a jail.
3625 	 * This is ugly check, but this is the only situation when jail's
3626 	 * directory ends with '/'.
3627 	 */
3628 	if (strcmp(pr->pr_path, "/") == 0)
3629 		return (0);
3630 	len = strlen(pr->pr_path);
3631 	sp = &mp->mnt_stat;
3632 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3633 		return (ENOENT);
3634 	/*
3635 	 * Be sure that we don't have situation where jail's root directory
3636 	 * is "/some/path" and mount point is "/some/pathpath".
3637 	 */
3638 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3639 		return (ENOENT);
3640 	return (0);
3641 }
3642 
3643 void
3644 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3645 {
3646 	char jpath[MAXPATHLEN];
3647 	struct prison *pr;
3648 	size_t len;
3649 
3650 	pr = cred->cr_prison;
3651 	if (pr->pr_enforce_statfs == 0)
3652 		return;
3653 	if (prison_canseemount(cred, mp) != 0) {
3654 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3655 		strlcpy(sp->f_mntonname, "[restricted]",
3656 		    sizeof(sp->f_mntonname));
3657 		return;
3658 	}
3659 	if (pr->pr_root->v_mount == mp) {
3660 		/*
3661 		 * Clear current buffer data, so we are sure nothing from
3662 		 * the valid path left there.
3663 		 */
3664 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3665 		*sp->f_mntonname = '/';
3666 		return;
3667 	}
3668 	/*
3669 	 * If jail's chroot directory is set to "/" we should be able to see
3670 	 * all mount-points from inside a jail.
3671 	 */
3672 	if (strcmp(pr->pr_path, "/") == 0)
3673 		return;
3674 	len = strlen(pr->pr_path);
3675 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3676 	/*
3677 	 * Clear current buffer data, so we are sure nothing from
3678 	 * the valid path left there.
3679 	 */
3680 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3681 	if (*jpath == '\0') {
3682 		/* Should never happen. */
3683 		*sp->f_mntonname = '/';
3684 	} else {
3685 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3686 	}
3687 }
3688 
3689 /*
3690  * Check with permission for a specific privilege is granted within jail.  We
3691  * have a specific list of accepted privileges; the rest are denied.
3692  */
3693 int
3694 prison_priv_check(struct ucred *cred, int priv)
3695 {
3696 	struct prison *pr;
3697 	int error;
3698 
3699 	/*
3700 	 * Some policies have custom handlers. This routine should not be
3701 	 * called for them. See priv_check_cred().
3702 	 */
3703 	switch (priv) {
3704 	case PRIV_VFS_LOOKUP:
3705 	case PRIV_VFS_GENERATION:
3706 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3707 		    "called for %d\n", priv));
3708 	}
3709 
3710 	if (!jailed(cred))
3711 		return (0);
3712 
3713 #ifdef VIMAGE
3714 	/*
3715 	 * Privileges specific to prisons with a virtual network stack.
3716 	 * There might be a duplicate entry here in case the privilege
3717 	 * is only granted conditionally in the legacy jail case.
3718 	 */
3719 	switch (priv) {
3720 #ifdef notyet
3721 		/*
3722 		 * NFS-specific privileges.
3723 		 */
3724 	case PRIV_NFS_DAEMON:
3725 	case PRIV_NFS_LOCKD:
3726 #endif
3727 		/*
3728 		 * Network stack privileges.
3729 		 */
3730 	case PRIV_NET_BRIDGE:
3731 	case PRIV_NET_GRE:
3732 	case PRIV_NET_BPF:
3733 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3734 	case PRIV_NET_ROUTE:
3735 	case PRIV_NET_TAP:
3736 	case PRIV_NET_SETIFMTU:
3737 	case PRIV_NET_SETIFFLAGS:
3738 	case PRIV_NET_SETIFCAP:
3739 	case PRIV_NET_SETIFDESCR:
3740 	case PRIV_NET_SETIFNAME	:
3741 	case PRIV_NET_SETIFMETRIC:
3742 	case PRIV_NET_SETIFPHYS:
3743 	case PRIV_NET_SETIFMAC:
3744 	case PRIV_NET_SETLANPCP:
3745 	case PRIV_NET_ADDMULTI:
3746 	case PRIV_NET_DELMULTI:
3747 	case PRIV_NET_HWIOCTL:
3748 	case PRIV_NET_SETLLADDR:
3749 	case PRIV_NET_ADDIFGROUP:
3750 	case PRIV_NET_DELIFGROUP:
3751 	case PRIV_NET_IFCREATE:
3752 	case PRIV_NET_IFDESTROY:
3753 	case PRIV_NET_ADDIFADDR:
3754 	case PRIV_NET_DELIFADDR:
3755 	case PRIV_NET_LAGG:
3756 	case PRIV_NET_GIF:
3757 	case PRIV_NET_SETIFVNET:
3758 	case PRIV_NET_SETIFFIB:
3759 	case PRIV_NET_OVPN:
3760 	case PRIV_NET_ME:
3761 
3762 		/*
3763 		 * 802.11-related privileges.
3764 		 */
3765 	case PRIV_NET80211_VAP_GETKEY:
3766 	case PRIV_NET80211_VAP_MANAGE:
3767 
3768 #ifdef notyet
3769 		/*
3770 		 * ATM privileges.
3771 		 */
3772 	case PRIV_NETATM_CFG:
3773 	case PRIV_NETATM_ADD:
3774 	case PRIV_NETATM_DEL:
3775 	case PRIV_NETATM_SET:
3776 
3777 		/*
3778 		 * Bluetooth privileges.
3779 		 */
3780 	case PRIV_NETBLUETOOTH_RAW:
3781 #endif
3782 
3783 		/*
3784 		 * Netgraph and netgraph module privileges.
3785 		 */
3786 	case PRIV_NETGRAPH_CONTROL:
3787 #ifdef notyet
3788 	case PRIV_NETGRAPH_TTY:
3789 #endif
3790 
3791 		/*
3792 		 * IPv4 and IPv6 privileges.
3793 		 */
3794 	case PRIV_NETINET_IPFW:
3795 	case PRIV_NETINET_DIVERT:
3796 	case PRIV_NETINET_PF:
3797 	case PRIV_NETINET_DUMMYNET:
3798 	case PRIV_NETINET_CARP:
3799 	case PRIV_NETINET_MROUTE:
3800 	case PRIV_NETINET_RAW:
3801 	case PRIV_NETINET_ADDRCTRL6:
3802 	case PRIV_NETINET_ND6:
3803 	case PRIV_NETINET_SCOPE6:
3804 	case PRIV_NETINET_ALIFETIME6:
3805 	case PRIV_NETINET_IPSEC:
3806 	case PRIV_NETINET_BINDANY:
3807 
3808 #ifdef notyet
3809 		/*
3810 		 * NCP privileges.
3811 		 */
3812 	case PRIV_NETNCP:
3813 
3814 		/*
3815 		 * SMB privileges.
3816 		 */
3817 	case PRIV_NETSMB:
3818 #endif
3819 
3820 	/*
3821 	 * No default: or deny here.
3822 	 * In case of no permit fall through to next switch().
3823 	 */
3824 		if (cred->cr_prison->pr_flags & PR_VNET)
3825 			return (0);
3826 	}
3827 #endif /* VIMAGE */
3828 
3829 	switch (priv) {
3830 		/*
3831 		 * Allow ktrace privileges for root in jail.
3832 		 */
3833 	case PRIV_KTRACE:
3834 
3835 #if 0
3836 		/*
3837 		 * Allow jailed processes to configure audit identity and
3838 		 * submit audit records (login, etc).  In the future we may
3839 		 * want to further refine the relationship between audit and
3840 		 * jail.
3841 		 */
3842 	case PRIV_AUDIT_GETAUDIT:
3843 	case PRIV_AUDIT_SETAUDIT:
3844 	case PRIV_AUDIT_SUBMIT:
3845 #endif
3846 
3847 		/*
3848 		 * Allow jailed processes to manipulate process UNIX
3849 		 * credentials in any way they see fit.
3850 		 */
3851 	case PRIV_CRED_SETUID:
3852 	case PRIV_CRED_SETEUID:
3853 	case PRIV_CRED_SETGID:
3854 	case PRIV_CRED_SETEGID:
3855 	case PRIV_CRED_SETGROUPS:
3856 	case PRIV_CRED_SETREUID:
3857 	case PRIV_CRED_SETREGID:
3858 	case PRIV_CRED_SETRESUID:
3859 	case PRIV_CRED_SETRESGID:
3860 
3861 		/*
3862 		 * Jail implements visibility constraints already, so allow
3863 		 * jailed root to override uid/gid-based constraints.
3864 		 */
3865 	case PRIV_SEEOTHERGIDS:
3866 	case PRIV_SEEOTHERUIDS:
3867 
3868 		/*
3869 		 * Jail implements inter-process debugging limits already, so
3870 		 * allow jailed root various debugging privileges.
3871 		 */
3872 	case PRIV_DEBUG_DIFFCRED:
3873 	case PRIV_DEBUG_SUGID:
3874 	case PRIV_DEBUG_UNPRIV:
3875 
3876 		/*
3877 		 * Allow jail to set various resource limits and login
3878 		 * properties, and for now, exceed process resource limits.
3879 		 */
3880 	case PRIV_PROC_LIMIT:
3881 	case PRIV_PROC_SETLOGIN:
3882 	case PRIV_PROC_SETRLIMIT:
3883 
3884 		/*
3885 		 * System V and POSIX IPC privileges are granted in jail.
3886 		 */
3887 	case PRIV_IPC_READ:
3888 	case PRIV_IPC_WRITE:
3889 	case PRIV_IPC_ADMIN:
3890 	case PRIV_IPC_MSGSIZE:
3891 	case PRIV_MQ_ADMIN:
3892 
3893 		/*
3894 		 * Jail operations within a jail work on child jails.
3895 		 */
3896 	case PRIV_JAIL_ATTACH:
3897 	case PRIV_JAIL_SET:
3898 	case PRIV_JAIL_REMOVE:
3899 
3900 		/*
3901 		 * Jail implements its own inter-process limits, so allow
3902 		 * root processes in jail to change scheduling on other
3903 		 * processes in the same jail.  Likewise for signalling.
3904 		 */
3905 	case PRIV_SCHED_DIFFCRED:
3906 	case PRIV_SCHED_CPUSET:
3907 	case PRIV_SIGNAL_DIFFCRED:
3908 	case PRIV_SIGNAL_SUGID:
3909 
3910 		/*
3911 		 * Allow jailed processes to write to sysctls marked as jail
3912 		 * writable.
3913 		 */
3914 	case PRIV_SYSCTL_WRITEJAIL:
3915 
3916 		/*
3917 		 * Allow root in jail to manage a variety of quota
3918 		 * properties.  These should likely be conditional on a
3919 		 * configuration option.
3920 		 */
3921 	case PRIV_VFS_GETQUOTA:
3922 	case PRIV_VFS_SETQUOTA:
3923 
3924 		/*
3925 		 * Since Jail relies on chroot() to implement file system
3926 		 * protections, grant many VFS privileges to root in jail.
3927 		 * Be careful to exclude mount-related and NFS-related
3928 		 * privileges.
3929 		 */
3930 	case PRIV_VFS_READ:
3931 	case PRIV_VFS_WRITE:
3932 	case PRIV_VFS_ADMIN:
3933 	case PRIV_VFS_EXEC:
3934 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3935 	case PRIV_VFS_CHFLAGS_DEV:
3936 	case PRIV_VFS_CHOWN:
3937 	case PRIV_VFS_CHROOT:
3938 	case PRIV_VFS_RETAINSUGID:
3939 	case PRIV_VFS_FCHROOT:
3940 	case PRIV_VFS_LINK:
3941 	case PRIV_VFS_SETGID:
3942 	case PRIV_VFS_STAT:
3943 	case PRIV_VFS_STICKYFILE:
3944 
3945 		/*
3946 		 * As in the non-jail case, non-root users are expected to be
3947 		 * able to read kernel/physical memory (provided /dev/[k]mem
3948 		 * exists in the jail and they have permission to access it).
3949 		 */
3950 	case PRIV_KMEM_READ:
3951 		return (0);
3952 
3953 		/*
3954 		 * Depending on the global setting, allow privilege of
3955 		 * setting system flags.
3956 		 */
3957 	case PRIV_VFS_SYSFLAGS:
3958 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3959 			return (0);
3960 		else
3961 			return (EPERM);
3962 
3963 		/*
3964 		 * Depending on the global setting, allow privilege of
3965 		 * mounting/unmounting file systems.
3966 		 */
3967 	case PRIV_VFS_MOUNT:
3968 	case PRIV_VFS_UNMOUNT:
3969 	case PRIV_VFS_MOUNT_NONUSER:
3970 	case PRIV_VFS_MOUNT_OWNER:
3971 		pr = cred->cr_prison;
3972 		prison_lock(pr);
3973 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3974 			error = 0;
3975 		else
3976 			error = EPERM;
3977 		prison_unlock(pr);
3978 		return (error);
3979 
3980 		/*
3981 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3982 		 * policy.  priv_check_cred will not specifically allow it, and
3983 		 * we may want a MAC policy to allow it.
3984 		 */
3985 	case PRIV_VFS_READ_DIR:
3986 		return (0);
3987 
3988 		/*
3989 		 * Conditionnaly allow locking (unlocking) physical pages
3990 		 * in memory.
3991 		 */
3992 	case PRIV_VM_MLOCK:
3993 	case PRIV_VM_MUNLOCK:
3994 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3995 			return (0);
3996 		else
3997 			return (EPERM);
3998 
3999 		/*
4000 		 * Conditionally allow jailed root to bind reserved ports.
4001 		 */
4002 	case PRIV_NETINET_RESERVEDPORT:
4003 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4004 			return (0);
4005 		else
4006 			return (EPERM);
4007 
4008 		/*
4009 		 * Allow jailed root to reuse in-use ports.
4010 		 */
4011 	case PRIV_NETINET_REUSEPORT:
4012 		return (0);
4013 
4014 		/*
4015 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4016 		 */
4017 	case PRIV_NETINET_SETHDROPTS:
4018 		return (0);
4019 
4020 		/*
4021 		 * Conditionally allow creating raw sockets in jail.
4022 		 */
4023 	case PRIV_NETINET_RAW:
4024 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4025 			return (0);
4026 		else
4027 			return (EPERM);
4028 
4029 		/*
4030 		 * Since jail implements its own visibility limits on netstat
4031 		 * sysctls, allow getcred.  This allows identd to work in
4032 		 * jail.
4033 		 */
4034 	case PRIV_NETINET_GETCRED:
4035 		return (0);
4036 
4037 		/*
4038 		 * Allow jailed root to set loginclass.
4039 		 */
4040 	case PRIV_PROC_SETLOGINCLASS:
4041 		return (0);
4042 
4043 		/*
4044 		 * Do not allow a process inside a jail to read the kernel
4045 		 * message buffer unless explicitly permitted.
4046 		 */
4047 	case PRIV_MSGBUF:
4048 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4049 			return (0);
4050 		return (EPERM);
4051 
4052 	default:
4053 		/*
4054 		 * In all remaining cases, deny the privilege request.  This
4055 		 * includes almost all network privileges, many system
4056 		 * configuration privileges.
4057 		 */
4058 		return (EPERM);
4059 	}
4060 }
4061 
4062 /*
4063  * Return the part of pr2's name that is relative to pr1, or the whole name
4064  * if it does not directly follow.
4065  */
4066 
4067 char *
4068 prison_name(struct prison *pr1, struct prison *pr2)
4069 {
4070 	char *name;
4071 
4072 	/* Jails see themselves as "0" (if they see themselves at all). */
4073 	if (pr1 == pr2)
4074 		return "0";
4075 	name = pr2->pr_name;
4076 	if (prison_ischild(pr1, pr2)) {
4077 		/*
4078 		 * pr1 isn't locked (and allprison_lock may not be either)
4079 		 * so its length can't be counted on.  But the number of dots
4080 		 * can be counted on - and counted.
4081 		 */
4082 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4083 			name = strchr(name, '.') + 1;
4084 	}
4085 	return (name);
4086 }
4087 
4088 /*
4089  * Return the part of pr2's path that is relative to pr1, or the whole path
4090  * if it does not directly follow.
4091  */
4092 static char *
4093 prison_path(struct prison *pr1, struct prison *pr2)
4094 {
4095 	char *path1, *path2;
4096 	int len1;
4097 
4098 	path1 = pr1->pr_path;
4099 	path2 = pr2->pr_path;
4100 	if (!strcmp(path1, "/"))
4101 		return (path2);
4102 	len1 = strlen(path1);
4103 	if (strncmp(path1, path2, len1))
4104 		return (path2);
4105 	if (path2[len1] == '\0')
4106 		return "/";
4107 	if (path2[len1] == '/')
4108 		return (path2 + len1);
4109 	return (path2);
4110 }
4111 
4112 /*
4113  * Jail-related sysctls.
4114  */
4115 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4116     "Jails");
4117 
4118 #if defined(INET) || defined(INET6)
4119 /*
4120  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4121  * sysctl_jail_list() helper.
4122  */
4123 static void
4124 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4125 {
4126 	const size_t size = pr_families[af].size;
4127 
4128  again:
4129 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4130 	if (pr->pr_addrs[af] != NULL) {
4131 		if (*len < pr->pr_addrs[af]->ips) {
4132 			*len = pr->pr_addrs[af]->ips;
4133 			mtx_unlock(&pr->pr_mtx);
4134 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4135 			mtx_lock(&pr->pr_mtx);
4136 			goto again;
4137 		}
4138 		bcopy(pr->pr_addrs[af] + 1, *out, pr->pr_addrs[af]->ips * size);
4139 	}
4140 }
4141 #endif
4142 
4143 static int
4144 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4145 {
4146 	struct xprison *xp;
4147 	struct prison *pr, *cpr;
4148 #ifdef INET
4149 	struct in_addr *ip4 = NULL;
4150 	int ip4s = 0;
4151 #endif
4152 #ifdef INET6
4153 	struct in6_addr *ip6 = NULL;
4154 	int ip6s = 0;
4155 #endif
4156 	int descend, error;
4157 
4158 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4159 	pr = req->td->td_ucred->cr_prison;
4160 	error = 0;
4161 	sx_slock(&allprison_lock);
4162 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4163 		mtx_lock(&cpr->pr_mtx);
4164 #ifdef INET
4165 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4166 #endif
4167 #ifdef INET6
4168 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4169 #endif
4170 		bzero(xp, sizeof(*xp));
4171 		xp->pr_version = XPRISON_VERSION;
4172 		xp->pr_id = cpr->pr_id;
4173 		xp->pr_state = cpr->pr_state;
4174 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4175 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4176 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4177 #ifdef INET
4178 		xp->pr_ip4s = ip4s;
4179 #endif
4180 #ifdef INET6
4181 		xp->pr_ip6s = ip6s;
4182 #endif
4183 		mtx_unlock(&cpr->pr_mtx);
4184 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4185 		if (error)
4186 			break;
4187 #ifdef INET
4188 		if (xp->pr_ip4s > 0) {
4189 			error = SYSCTL_OUT(req, ip4,
4190 			    xp->pr_ip4s * sizeof(struct in_addr));
4191 			if (error)
4192 				break;
4193 		}
4194 #endif
4195 #ifdef INET6
4196 		if (xp->pr_ip6s > 0) {
4197 			error = SYSCTL_OUT(req, ip6,
4198 			    xp->pr_ip6s * sizeof(struct in6_addr));
4199 			if (error)
4200 				break;
4201 		}
4202 #endif
4203 	}
4204 	sx_sunlock(&allprison_lock);
4205 	free(xp, M_TEMP);
4206 #ifdef INET
4207 	free(ip4, M_TEMP);
4208 #endif
4209 #ifdef INET6
4210 	free(ip6, M_TEMP);
4211 #endif
4212 	return (error);
4213 }
4214 
4215 SYSCTL_OID(_security_jail, OID_AUTO, list,
4216     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4217     sysctl_jail_list, "S", "List of active jails");
4218 
4219 static int
4220 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4221 {
4222 	int error, injail;
4223 
4224 	injail = jailed(req->td->td_ucred);
4225 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4226 
4227 	return (error);
4228 }
4229 
4230 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4231     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4232     sysctl_jail_jailed, "I", "Process in jail?");
4233 
4234 static int
4235 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4236 {
4237 	int error, havevnet;
4238 #ifdef VIMAGE
4239 	struct ucred *cred = req->td->td_ucred;
4240 
4241 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4242 #else
4243 	havevnet = 0;
4244 #endif
4245 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4246 
4247 	return (error);
4248 }
4249 
4250 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4251     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4252     sysctl_jail_vnet, "I", "Jail owns vnet?");
4253 
4254 #if defined(INET) || defined(INET6)
4255 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4256     &jail_max_af_ips, 0,
4257     "Number of IP addresses a jail may have at most per address family (deprecated)");
4258 #endif
4259 
4260 /*
4261  * Default parameters for jail(2) compatibility.  For historical reasons,
4262  * the sysctl names have varying similarity to the parameter names.  Prisons
4263  * just see their own parameters, and can't change them.
4264  */
4265 static int
4266 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4267 {
4268 	int error, i;
4269 
4270 	/* Get the current flag value, and convert it to a boolean. */
4271 	if (req->td->td_ucred->cr_prison == &prison0) {
4272 		mtx_lock(&prison0.pr_mtx);
4273 		i = (jail_default_allow & arg2) != 0;
4274 		mtx_unlock(&prison0.pr_mtx);
4275 	} else
4276 		i = prison_allow(req->td->td_ucred, arg2);
4277 
4278 	if (arg1 != NULL)
4279 		i = !i;
4280 	error = sysctl_handle_int(oidp, &i, 0, req);
4281 	if (error || !req->newptr)
4282 		return (error);
4283 	i = i ? arg2 : 0;
4284 	if (arg1 != NULL)
4285 		i ^= arg2;
4286 	/*
4287 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4288 	 * for writing.
4289 	 */
4290 	mtx_lock(&prison0.pr_mtx);
4291 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4292 	mtx_unlock(&prison0.pr_mtx);
4293 	return (0);
4294 }
4295 
4296 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4297     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4298     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4299     "Processes in jail can set their hostnames (deprecated)");
4300 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4301     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4302     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4303     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4304 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4305     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4306     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4307     "Processes in jail can use System V IPC primitives (deprecated)");
4308 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4309     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4310     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4311     "Prison root can create raw sockets (deprecated)");
4312 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4313     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4314     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4315     "Processes in jail can alter system file flags (deprecated)");
4316 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4317     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4318     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4319     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4320 
4321 static int
4322 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4323 {
4324 	struct prison *pr;
4325 	int level, error;
4326 
4327 	pr = req->td->td_ucred->cr_prison;
4328 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4329 	error = sysctl_handle_int(oidp, &level, 0, req);
4330 	if (error || !req->newptr)
4331 		return (error);
4332 	*(int *)arg1 = level;
4333 	return (0);
4334 }
4335 
4336 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4337     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4338     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4339     sysctl_jail_default_level, "I",
4340     "Processes in jail cannot see all mounted file systems (deprecated)");
4341 
4342 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4343     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4344     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4345     sysctl_jail_default_level, "I",
4346     "Ruleset for the devfs filesystem in jail (deprecated)");
4347 
4348 /*
4349  * Nodes to describe jail parameters.  Maximum length of string parameters
4350  * is returned in the string itself, and the other parameters exist merely
4351  * to make themselves and their types known.
4352  */
4353 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4354     "Jail parameters");
4355 
4356 int
4357 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4358 {
4359 	int i;
4360 	long l;
4361 	size_t s;
4362 	char numbuf[12];
4363 
4364 	switch (oidp->oid_kind & CTLTYPE)
4365 	{
4366 	case CTLTYPE_LONG:
4367 	case CTLTYPE_ULONG:
4368 		l = 0;
4369 #ifdef SCTL_MASK32
4370 		if (!(req->flags & SCTL_MASK32))
4371 #endif
4372 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4373 	case CTLTYPE_INT:
4374 	case CTLTYPE_UINT:
4375 		i = 0;
4376 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4377 	case CTLTYPE_STRING:
4378 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4379 		return
4380 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4381 	case CTLTYPE_STRUCT:
4382 		s = (size_t)arg2;
4383 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4384 	}
4385 	return (0);
4386 }
4387 
4388 /*
4389  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4390  * jail creation time but cannot be changed in an existing jail.
4391  */
4392 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4393 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4394 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4395 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4396 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4397     "I", "Jail secure level");
4398 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4399     "Jail value for kern.osreldate and uname -K");
4400 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4401     "Jail value for kern.osrelease and uname -r");
4402 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4403     "I", "Jail cannot see all mounted file systems");
4404 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4405     "I", "Ruleset for in-jail devfs mounts");
4406 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4407     "B", "Jail persistence");
4408 #ifdef VIMAGE
4409 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4410     "E,jailsys", "Virtual network stack");
4411 #endif
4412 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4413     "B", "Jail is in the process of shutting down");
4414 
4415 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4416 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4417     "I", "Current number of child jails");
4418 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4419     "I", "Maximum number of child jails");
4420 
4421 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4422 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4423     "Jail hostname");
4424 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4425     "Jail NIS domainname");
4426 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4427     "Jail host UUID");
4428 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4429     "LU", "Jail host ID");
4430 
4431 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4432 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4433 
4434 #ifdef INET
4435 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4436     "Jail IPv4 address virtualization");
4437 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4438     "S,in_addr,a", "Jail IPv4 addresses");
4439 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4440     "B", "Do (not) use IPv4 source address selection rather than the "
4441     "primary jail IPv4 address.");
4442 #endif
4443 #ifdef INET6
4444 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4445     "Jail IPv6 address virtualization");
4446 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4447     "S,in6_addr,a", "Jail IPv6 addresses");
4448 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4449     "B", "Do (not) use IPv6 source address selection rather than the "
4450     "primary jail IPv6 address.");
4451 #endif
4452 
4453 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4454 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4455     "B", "Jail may set hostname");
4456 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4457     "B", "Jail may use SYSV IPC");
4458 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4459     "B", "Jail may create raw sockets");
4460 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4461     "B", "Jail may alter system file flags");
4462 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4463     "B", "Jail may set file quotas");
4464 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4465     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4466 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4467     "B", "Jail may lock (unlock) physical pages in memory");
4468 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4469     "B", "Jail may bind sockets to reserved ports");
4470 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4471     "B", "Jail may read the kernel message buffer");
4472 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4473     "B", "Unprivileged processes may use process debugging facilities");
4474 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4475     "B", "Processes in jail with uid 0 have privilege");
4476 
4477 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4478 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4479     "B", "Jail may mount/unmount jail-friendly file systems in general");
4480 
4481 /*
4482  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4483  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4484  * not created.
4485  */
4486 unsigned
4487 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4488     const char *descr)
4489 {
4490 	struct bool_flags *bf;
4491 	struct sysctl_oid *parent;
4492 	char *allow_name, *allow_noname, *allowed;
4493 #ifndef NO_SYSCTL_DESCR
4494 	char *descr_deprecated;
4495 #endif
4496 	u_int allow_flag;
4497 
4498 	if (prefix
4499 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4500 		< 0 ||
4501 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4502 		< 0
4503 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4504 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4505 		free(allow_name, M_PRISON);
4506 		return 0;
4507 	}
4508 
4509 	/*
4510 	 * See if this parameter has already beed added, i.e. a module was
4511 	 * previously loaded/unloaded.
4512 	 */
4513 	mtx_lock(&prison0.pr_mtx);
4514 	for (bf = pr_flag_allow;
4515 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4516 		atomic_load_int(&bf->flag) != 0;
4517 	     bf++) {
4518 		if (strcmp(bf->name, allow_name) == 0) {
4519 			allow_flag = bf->flag;
4520 			goto no_add;
4521 		}
4522 	}
4523 
4524 	/*
4525 	 * Find a free bit in pr_allow_all, failing if there are none
4526 	 * (which shouldn't happen as long as we keep track of how many
4527 	 * potential dynamic flags exist).
4528 	 */
4529 	for (allow_flag = 1;; allow_flag <<= 1) {
4530 		if (allow_flag == 0)
4531 			goto no_add;
4532 		if ((pr_allow_all & allow_flag) == 0)
4533 			break;
4534 	}
4535 
4536 	/* Note the parameter in the next open slot in pr_flag_allow. */
4537 	for (bf = pr_flag_allow; ; bf++) {
4538 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4539 			/* This should never happen, but is not fatal. */
4540 			allow_flag = 0;
4541 			goto no_add;
4542 		}
4543 		if (atomic_load_int(&bf->flag) == 0)
4544 			break;
4545 	}
4546 	bf->name = allow_name;
4547 	bf->noname = allow_noname;
4548 	pr_allow_all |= allow_flag;
4549 	/*
4550 	 * prison0 always has permission for the new parameter.
4551 	 * Other jails must have it granted to them.
4552 	 */
4553 	prison0.pr_allow |= allow_flag;
4554 	/* The flag indicates a valid entry, so make sure it is set last. */
4555 	atomic_store_rel_int(&bf->flag, allow_flag);
4556 	mtx_unlock(&prison0.pr_mtx);
4557 
4558 	/*
4559 	 * Create sysctls for the parameter, and the back-compat global
4560 	 * permission.
4561 	 */
4562 	parent = prefix
4563 	    ? SYSCTL_ADD_NODE(NULL,
4564 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4565 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4566 	    : &sysctl___security_jail_param_allow;
4567 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4568 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4569 	    NULL, 0, sysctl_jail_param, "B", descr);
4570 	if ((prefix
4571 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4572 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4573 #ifndef NO_SYSCTL_DESCR
4574 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4575 		    descr);
4576 #endif
4577 		(void)SYSCTL_ADD_PROC(NULL,
4578 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4579 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4580 		    sysctl_jail_default_allow, "I", descr_deprecated);
4581 #ifndef NO_SYSCTL_DESCR
4582 		free(descr_deprecated, M_TEMP);
4583 #endif
4584 		free(allowed, M_TEMP);
4585 	}
4586 	return allow_flag;
4587 
4588  no_add:
4589 	mtx_unlock(&prison0.pr_mtx);
4590 	free(allow_name, M_PRISON);
4591 	free(allow_noname, M_PRISON);
4592 	return allow_flag;
4593 }
4594 
4595 /*
4596  * The VFS system will register jail-aware filesystems here.  They each get
4597  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4598  * attempts to mount.
4599  */
4600 void
4601 prison_add_vfs(struct vfsconf *vfsp)
4602 {
4603 #ifdef NO_SYSCTL_DESCR
4604 
4605 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4606 	    NULL, NULL);
4607 #else
4608 	char *descr;
4609 
4610 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4611 	    vfsp->vfc_name);
4612 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4613 	    NULL, descr);
4614 	free(descr, M_TEMP);
4615 #endif
4616 }
4617 
4618 #ifdef RACCT
4619 void
4620 prison_racct_foreach(void (*callback)(struct racct *racct,
4621     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4622     void *arg2, void *arg3)
4623 {
4624 	struct prison_racct *prr;
4625 
4626 	ASSERT_RACCT_ENABLED();
4627 
4628 	sx_slock(&allprison_lock);
4629 	if (pre != NULL)
4630 		(pre)();
4631 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4632 		(callback)(prr->prr_racct, arg2, arg3);
4633 	if (post != NULL)
4634 		(post)();
4635 	sx_sunlock(&allprison_lock);
4636 }
4637 
4638 static struct prison_racct *
4639 prison_racct_find_locked(const char *name)
4640 {
4641 	struct prison_racct *prr;
4642 
4643 	ASSERT_RACCT_ENABLED();
4644 	sx_assert(&allprison_lock, SA_XLOCKED);
4645 
4646 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4647 		return (NULL);
4648 
4649 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4650 		if (strcmp(name, prr->prr_name) != 0)
4651 			continue;
4652 
4653 		/* Found prison_racct with a matching name? */
4654 		prison_racct_hold(prr);
4655 		return (prr);
4656 	}
4657 
4658 	/* Add new prison_racct. */
4659 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4660 	racct_create(&prr->prr_racct);
4661 
4662 	strcpy(prr->prr_name, name);
4663 	refcount_init(&prr->prr_refcount, 1);
4664 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4665 
4666 	return (prr);
4667 }
4668 
4669 struct prison_racct *
4670 prison_racct_find(const char *name)
4671 {
4672 	struct prison_racct *prr;
4673 
4674 	ASSERT_RACCT_ENABLED();
4675 
4676 	sx_xlock(&allprison_lock);
4677 	prr = prison_racct_find_locked(name);
4678 	sx_xunlock(&allprison_lock);
4679 	return (prr);
4680 }
4681 
4682 void
4683 prison_racct_hold(struct prison_racct *prr)
4684 {
4685 
4686 	ASSERT_RACCT_ENABLED();
4687 
4688 	refcount_acquire(&prr->prr_refcount);
4689 }
4690 
4691 static void
4692 prison_racct_free_locked(struct prison_racct *prr)
4693 {
4694 
4695 	ASSERT_RACCT_ENABLED();
4696 	sx_assert(&allprison_lock, SA_XLOCKED);
4697 
4698 	if (refcount_release(&prr->prr_refcount)) {
4699 		racct_destroy(&prr->prr_racct);
4700 		LIST_REMOVE(prr, prr_next);
4701 		free(prr, M_PRISON_RACCT);
4702 	}
4703 }
4704 
4705 void
4706 prison_racct_free(struct prison_racct *prr)
4707 {
4708 
4709 	ASSERT_RACCT_ENABLED();
4710 	sx_assert(&allprison_lock, SA_UNLOCKED);
4711 
4712 	if (refcount_release_if_not_last(&prr->prr_refcount))
4713 		return;
4714 
4715 	sx_xlock(&allprison_lock);
4716 	prison_racct_free_locked(prr);
4717 	sx_xunlock(&allprison_lock);
4718 }
4719 
4720 static void
4721 prison_racct_attach(struct prison *pr)
4722 {
4723 	struct prison_racct *prr;
4724 
4725 	ASSERT_RACCT_ENABLED();
4726 	sx_assert(&allprison_lock, SA_XLOCKED);
4727 
4728 	prr = prison_racct_find_locked(pr->pr_name);
4729 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4730 
4731 	pr->pr_prison_racct = prr;
4732 }
4733 
4734 /*
4735  * Handle jail renaming.  From the racct point of view, renaming means
4736  * moving from one prison_racct to another.
4737  */
4738 static void
4739 prison_racct_modify(struct prison *pr)
4740 {
4741 #ifdef RCTL
4742 	struct proc *p;
4743 	struct ucred *cred;
4744 #endif
4745 	struct prison_racct *oldprr;
4746 
4747 	ASSERT_RACCT_ENABLED();
4748 
4749 	sx_slock(&allproc_lock);
4750 	sx_xlock(&allprison_lock);
4751 
4752 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4753 		sx_xunlock(&allprison_lock);
4754 		sx_sunlock(&allproc_lock);
4755 		return;
4756 	}
4757 
4758 	oldprr = pr->pr_prison_racct;
4759 	pr->pr_prison_racct = NULL;
4760 
4761 	prison_racct_attach(pr);
4762 
4763 	/*
4764 	 * Move resource utilisation records.
4765 	 */
4766 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4767 
4768 #ifdef RCTL
4769 	/*
4770 	 * Force rctl to reattach rules to processes.
4771 	 */
4772 	FOREACH_PROC_IN_SYSTEM(p) {
4773 		PROC_LOCK(p);
4774 		cred = crhold(p->p_ucred);
4775 		PROC_UNLOCK(p);
4776 		rctl_proc_ucred_changed(p, cred);
4777 		crfree(cred);
4778 	}
4779 #endif
4780 
4781 	sx_sunlock(&allproc_lock);
4782 	prison_racct_free_locked(oldprr);
4783 	sx_xunlock(&allprison_lock);
4784 }
4785 
4786 static void
4787 prison_racct_detach(struct prison *pr)
4788 {
4789 
4790 	ASSERT_RACCT_ENABLED();
4791 	sx_assert(&allprison_lock, SA_UNLOCKED);
4792 
4793 	if (pr->pr_prison_racct == NULL)
4794 		return;
4795 	prison_racct_free(pr->pr_prison_racct);
4796 	pr->pr_prison_racct = NULL;
4797 }
4798 #endif /* RACCT */
4799 
4800 #ifdef DDB
4801 
4802 static void
4803 db_show_prison(struct prison *pr)
4804 {
4805 	struct bool_flags *bf;
4806 	struct jailsys_flags *jsf;
4807 #if defined(INET) || defined(INET6)
4808 	int ii;
4809 #endif
4810 	unsigned f;
4811 #ifdef INET
4812 	char ip4buf[INET_ADDRSTRLEN];
4813 #endif
4814 #ifdef INET6
4815 	char ip6buf[INET6_ADDRSTRLEN];
4816 #endif
4817 
4818 	db_printf("prison %p:\n", pr);
4819 	db_printf(" jid             = %d\n", pr->pr_id);
4820 	db_printf(" name            = %s\n", pr->pr_name);
4821 	db_printf(" parent          = %p\n", pr->pr_parent);
4822 	db_printf(" ref             = %d\n", pr->pr_ref);
4823 	db_printf(" uref            = %d\n", pr->pr_uref);
4824 	db_printf(" state           = %s\n",
4825 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4826 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
4827 	    "invalid");
4828 	db_printf(" path            = %s\n", pr->pr_path);
4829 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4830 	    ? pr->pr_cpuset->cs_id : -1);
4831 #ifdef VIMAGE
4832 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4833 #endif
4834 	db_printf(" root            = %p\n", pr->pr_root);
4835 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4836 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4837 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4838 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4839 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4840 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4841 	db_printf(" flags           = 0x%x", pr->pr_flags);
4842 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4843 		if (pr->pr_flags & bf->flag)
4844 			db_printf(" %s", bf->name);
4845 	for (jsf = pr_flag_jailsys;
4846 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4847 	     jsf++) {
4848 		f = pr->pr_flags & (jsf->disable | jsf->new);
4849 		db_printf(" %-16s= %s\n", jsf->name,
4850 		    (f != 0 && f == jsf->disable) ? "disable"
4851 		    : (f == jsf->new) ? "new"
4852 		    : "inherit");
4853 	}
4854 	db_printf(" allow           = 0x%x", pr->pr_allow);
4855 	for (bf = pr_flag_allow;
4856 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4857 		atomic_load_int(&bf->flag) != 0;
4858 	     bf++)
4859 		if (pr->pr_allow & bf->flag)
4860 			db_printf(" %s", bf->name);
4861 	db_printf("\n");
4862 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4863 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4864 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4865 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4866 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4867 #ifdef INET
4868 	if (pr->pr_addrs[PR_INET] != NULL) {
4869 		pr_family_t af = PR_INET;
4870 
4871 		db_printf(" ip4s            = %d\n", pr->pr_addrs[af]->ips);
4872 		for (ii = 0; ii < pr->pr_addrs[af]->ips; ii++)
4873 			db_printf(" %s %s\n",
4874 			    ii == 0 ? "ip4.addr        =" : "                 ",
4875 			    inet_ntoa_r(
4876 			    *(const struct in_addr *)PR_IP(pr, ii),
4877 			    ip4buf));
4878 	}
4879 #endif
4880 #ifdef INET6
4881 	if (pr->pr_addrs[PR_INET6] != NULL) {
4882 		pr_family_t af = PR_INET6;
4883 
4884 		db_printf(" ip6s            = %d\n", pr->pr_addrs[af]->ips);
4885 		for (ii = 0; ii < pr->pr_addrs[af]->ips; ii++)
4886 			db_printf(" %s %s\n",
4887 			    ii == 0 ? "ip6.addr        =" : "                 ",
4888 			    ip6_sprintf(ip6buf,
4889 			    (const struct in6_addr *)PR_IP(pr, ii)));
4890 	}
4891 #endif
4892 }
4893 
4894 DB_SHOW_COMMAND(prison, db_show_prison_command)
4895 {
4896 	struct prison *pr;
4897 
4898 	if (!have_addr) {
4899 		/*
4900 		 * Show all prisons in the list, and prison0 which is not
4901 		 * listed.
4902 		 */
4903 		db_show_prison(&prison0);
4904 		if (!db_pager_quit) {
4905 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4906 				db_show_prison(pr);
4907 				if (db_pager_quit)
4908 					break;
4909 			}
4910 		}
4911 		return;
4912 	}
4913 
4914 	if (addr == 0)
4915 		pr = &prison0;
4916 	else {
4917 		/* Look for a prison with the ID and with references. */
4918 		TAILQ_FOREACH(pr, &allprison, pr_list)
4919 			if (pr->pr_id == addr && pr->pr_ref > 0)
4920 				break;
4921 		if (pr == NULL)
4922 			/* Look again, without requiring a reference. */
4923 			TAILQ_FOREACH(pr, &allprison, pr_list)
4924 				if (pr->pr_id == addr)
4925 					break;
4926 		if (pr == NULL)
4927 			/* Assume address points to a valid prison. */
4928 			pr = (struct prison *)addr;
4929 	}
4930 	db_show_prison(pr);
4931 }
4932 
4933 #endif /* DDB */
4934