xref: /freebsd/sys/kern/kern_jail.c (revision 734e82fe33aa764367791a7d603b383996c6b40b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_ddb.h"
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_nfs.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/epoch.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/linker.h>
52 #include <sys/lock.h>
53 #include <sys/mman.h>
54 #include <sys/mutex.h>
55 #include <sys/racct.h>
56 #include <sys/rctl.h>
57 #include <sys/refcount.h>
58 #include <sys/sx.h>
59 #include <sys/sysent.h>
60 #include <sys/namei.h>
61 #include <sys/mount.h>
62 #include <sys/queue.h>
63 #include <sys/socket.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/uuid.h>
67 #include <sys/vnode.h>
68 
69 #include <net/if.h>
70 #include <net/vnet.h>
71 
72 #include <netinet/in.h>
73 
74 #ifdef DDB
75 #include <ddb/ddb.h>
76 #endif /* DDB */
77 
78 #include <security/mac/mac_framework.h>
79 
80 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
81 
82 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
83 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
84 
85 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
86 #ifdef INET
87 #ifdef INET6
88 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
89 #else
90 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
91 #endif
92 #else /* !INET */
93 #ifdef INET6
94 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
95 #else
96 #define	_PR_IP_SADDRSEL	0
97 #endif
98 #endif
99 
100 /* prison0 describes what is "real" about the system. */
101 struct prison prison0 = {
102 	.pr_id		= 0,
103 	.pr_name	= "0",
104 	.pr_ref		= 1,
105 	.pr_uref	= 1,
106 	.pr_path	= "/",
107 	.pr_securelevel	= -1,
108 	.pr_devfs_rsnum = 0,
109 	.pr_state	= PRISON_STATE_ALIVE,
110 	.pr_childmax	= JAIL_MAX,
111 	.pr_hostuuid	= DEFAULT_HOSTUUID,
112 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
113 #ifdef VIMAGE
114 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
115 #else
116 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
117 #endif
118 	.pr_allow	= PR_ALLOW_ALL_STATIC,
119 };
120 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
121 
122 struct bool_flags {
123 	const char	*name;
124 	const char	*noname;
125 	volatile u_int	 flag;
126 };
127 struct jailsys_flags {
128 	const char	*name;
129 	unsigned	 disable;
130 	unsigned	 new;
131 };
132 
133 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
134 struct	sx allprison_lock;
135 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
136 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
137 LIST_HEAD(, prison_racct) allprison_racct;
138 int	lastprid = 0;
139 
140 static int get_next_prid(struct prison **insprp);
141 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
142 static void prison_complete(void *context, int pending);
143 static void prison_deref(struct prison *pr, int flags);
144 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
145 static int prison_lock_xlock(struct prison *pr, int flags);
146 static void prison_cleanup(struct prison *pr);
147 static void prison_free_not_last(struct prison *pr);
148 static void prison_proc_free_not_last(struct prison *pr);
149 static void prison_proc_relink(struct prison *opr, struct prison *npr,
150     struct proc *p);
151 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
152     int enable);
153 static char *prison_path(struct prison *pr1, struct prison *pr2);
154 #ifdef RACCT
155 static void prison_racct_attach(struct prison *pr);
156 static void prison_racct_modify(struct prison *pr);
157 static void prison_racct_detach(struct prison *pr);
158 #endif
159 
160 /* Flags for prison_deref */
161 #define	PD_DEREF	0x01	/* Decrement pr_ref */
162 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
163 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
164 #define	PD_LOCKED	0x10	/* pr_mtx is held */
165 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
166 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
167 #define PD_OP_FLAGS	0x07	/* Operation flags */
168 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
169 
170 /*
171  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
172  * as we cannot figure out the size of a sparse array, or an array without a
173  * terminating entry.
174  */
175 static struct bool_flags pr_flag_bool[] = {
176 	{"persist", "nopersist", PR_PERSIST},
177 #ifdef INET
178 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
179 #endif
180 #ifdef INET6
181 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
182 #endif
183 };
184 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
185 
186 static struct jailsys_flags pr_flag_jailsys[] = {
187 	{"host", 0, PR_HOST},
188 #ifdef VIMAGE
189 	{"vnet", 0, PR_VNET},
190 #endif
191 #ifdef INET
192 	{"ip4", PR_IP4_USER, PR_IP4_USER},
193 #endif
194 #ifdef INET6
195 	{"ip6", PR_IP6_USER, PR_IP6_USER},
196 #endif
197 };
198 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
199 
200 /*
201  * Make this array full-size so dynamic parameters can be added.
202  * It is protected by prison0.mtx, but lockless reading is allowed
203  * with an atomic check of the flag values.
204  */
205 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
206 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
207 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
208 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
209 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
210 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
211 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
212 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
213 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
214 	{"allow.reserved_ports", "allow.noreserved_ports",
215 	 PR_ALLOW_RESERVED_PORTS},
216 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
217 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
218 	 PR_ALLOW_UNPRIV_DEBUG},
219 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
220 #ifdef VIMAGE
221 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
222 #endif
223 };
224 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
225 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
226 
227 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
228 					 PR_ALLOW_RESERVED_PORTS | \
229 					 PR_ALLOW_UNPRIV_DEBUG | \
230 					 PR_ALLOW_SUSER)
231 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
232 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
233 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
234 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
235 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
236 #if defined(INET) || defined(INET6)
237 static unsigned jail_max_af_ips = 255;
238 #endif
239 
240 /*
241  * Initialize the parts of prison0 that can't be static-initialized with
242  * constants.  This is called from proc0_init() after creating thread0 cpuset.
243  */
244 void
245 prison0_init(void)
246 {
247 	uint8_t *file, *data;
248 	size_t size;
249 	char buf[sizeof(prison0.pr_hostuuid)];
250 	bool valid;
251 
252 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
253 	prison0.pr_osreldate = osreldate;
254 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
255 
256 	/* If we have a preloaded hostuuid, use it. */
257 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
258 	if (file != NULL) {
259 		data = preload_fetch_addr(file);
260 		size = preload_fetch_size(file);
261 		if (data != NULL) {
262 			/*
263 			 * The preloaded data may include trailing whitespace, almost
264 			 * certainly a newline; skip over any whitespace or
265 			 * non-printable characters to be safe.
266 			 */
267 			while (size > 0 && data[size - 1] <= 0x20) {
268 				size--;
269 			}
270 
271 			valid = false;
272 
273 			/*
274 			 * Not NUL-terminated when passed from loader, but
275 			 * validate_uuid requires that due to using sscanf (as
276 			 * does the subsequent strlcpy, since it still reads
277 			 * past the given size to return the true length);
278 			 * bounce to a temporary buffer to fix.
279 			 */
280 			if (size >= sizeof(buf))
281 				goto done;
282 
283 			memcpy(buf, data, size);
284 			buf[size] = '\0';
285 
286 			if (validate_uuid(buf, size, NULL, 0) != 0)
287 				goto done;
288 
289 			valid = true;
290 			(void)strlcpy(prison0.pr_hostuuid, buf,
291 			    sizeof(prison0.pr_hostuuid));
292 
293 done:
294 			if (bootverbose && !valid) {
295 				printf("hostuuid: preload data malformed: '%.*s'\n",
296 				    (int)size, data);
297 			}
298 		}
299 	}
300 	if (bootverbose)
301 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
302 }
303 
304 /*
305  * struct jail_args {
306  *	struct jail *jail;
307  * };
308  */
309 int
310 sys_jail(struct thread *td, struct jail_args *uap)
311 {
312 	uint32_t version;
313 	int error;
314 	struct jail j;
315 
316 	error = copyin(uap->jail, &version, sizeof(uint32_t));
317 	if (error)
318 		return (error);
319 
320 	switch (version) {
321 	case 0:
322 	{
323 		struct jail_v0 j0;
324 
325 		/* FreeBSD single IPv4 jails. */
326 		bzero(&j, sizeof(struct jail));
327 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
328 		if (error)
329 			return (error);
330 		j.version = j0.version;
331 		j.path = j0.path;
332 		j.hostname = j0.hostname;
333 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
334 		break;
335 	}
336 
337 	case 1:
338 		/*
339 		 * Version 1 was used by multi-IPv4 jail implementations
340 		 * that never made it into the official kernel.
341 		 */
342 		return (EINVAL);
343 
344 	case 2:	/* JAIL_API_VERSION */
345 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
346 		error = copyin(uap->jail, &j, sizeof(struct jail));
347 		if (error)
348 			return (error);
349 		break;
350 
351 	default:
352 		/* Sci-Fi jails are not supported, sorry. */
353 		return (EINVAL);
354 	}
355 	return (kern_jail(td, &j));
356 }
357 
358 int
359 kern_jail(struct thread *td, struct jail *j)
360 {
361 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
362 #ifdef INET
363 			    + 1
364 #endif
365 #ifdef INET6
366 			    + 1
367 #endif
368 			    )];
369 	struct uio opt;
370 	char *u_path, *u_hostname, *u_name;
371 	struct bool_flags *bf;
372 #ifdef INET
373 	uint32_t ip4s;
374 	struct in_addr *u_ip4;
375 #endif
376 #ifdef INET6
377 	struct in6_addr *u_ip6;
378 #endif
379 	size_t tmplen;
380 	int error, enforce_statfs;
381 
382 	bzero(&optiov, sizeof(optiov));
383 	opt.uio_iov = optiov;
384 	opt.uio_iovcnt = 0;
385 	opt.uio_offset = -1;
386 	opt.uio_resid = -1;
387 	opt.uio_segflg = UIO_SYSSPACE;
388 	opt.uio_rw = UIO_READ;
389 	opt.uio_td = td;
390 
391 	/* Set permissions for top-level jails from sysctls. */
392 	if (!jailed(td->td_ucred)) {
393 		for (bf = pr_flag_allow;
394 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
395 			atomic_load_int(&bf->flag) != 0;
396 		     bf++) {
397 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
398 			    (jail_default_allow & bf->flag)
399 			    ? bf->name : bf->noname);
400 			optiov[opt.uio_iovcnt].iov_len =
401 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
402 			opt.uio_iovcnt += 2;
403 		}
404 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
405 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
406 		opt.uio_iovcnt++;
407 		enforce_statfs = jail_default_enforce_statfs;
408 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
409 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
410 		opt.uio_iovcnt++;
411 	}
412 
413 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
414 #ifdef INET
415 	ip4s = (j->version == 0) ? 1 : j->ip4s;
416 	if (ip4s > jail_max_af_ips)
417 		return (EINVAL);
418 	tmplen += ip4s * sizeof(struct in_addr);
419 #else
420 	if (j->ip4s > 0)
421 		return (EINVAL);
422 #endif
423 #ifdef INET6
424 	if (j->ip6s > jail_max_af_ips)
425 		return (EINVAL);
426 	tmplen += j->ip6s * sizeof(struct in6_addr);
427 #else
428 	if (j->ip6s > 0)
429 		return (EINVAL);
430 #endif
431 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
432 	u_hostname = u_path + MAXPATHLEN;
433 	u_name = u_hostname + MAXHOSTNAMELEN;
434 #ifdef INET
435 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
436 #endif
437 #ifdef INET6
438 #ifdef INET
439 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
440 #else
441 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
442 #endif
443 #endif
444 	optiov[opt.uio_iovcnt].iov_base = "path";
445 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
446 	opt.uio_iovcnt++;
447 	optiov[opt.uio_iovcnt].iov_base = u_path;
448 	error = copyinstr(j->path, u_path, MAXPATHLEN,
449 	    &optiov[opt.uio_iovcnt].iov_len);
450 	if (error) {
451 		free(u_path, M_TEMP);
452 		return (error);
453 	}
454 	opt.uio_iovcnt++;
455 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
456 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
457 	opt.uio_iovcnt++;
458 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
459 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
460 	    &optiov[opt.uio_iovcnt].iov_len);
461 	if (error) {
462 		free(u_path, M_TEMP);
463 		return (error);
464 	}
465 	opt.uio_iovcnt++;
466 	if (j->jailname != NULL) {
467 		optiov[opt.uio_iovcnt].iov_base = "name";
468 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
469 		opt.uio_iovcnt++;
470 		optiov[opt.uio_iovcnt].iov_base = u_name;
471 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
472 		    &optiov[opt.uio_iovcnt].iov_len);
473 		if (error) {
474 			free(u_path, M_TEMP);
475 			return (error);
476 		}
477 		opt.uio_iovcnt++;
478 	}
479 #ifdef INET
480 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
481 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
482 	opt.uio_iovcnt++;
483 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
484 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
485 	if (j->version == 0)
486 		u_ip4->s_addr = j->ip4s;
487 	else {
488 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
489 		if (error) {
490 			free(u_path, M_TEMP);
491 			return (error);
492 		}
493 	}
494 	opt.uio_iovcnt++;
495 #endif
496 #ifdef INET6
497 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
498 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
499 	opt.uio_iovcnt++;
500 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
501 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
502 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
503 	if (error) {
504 		free(u_path, M_TEMP);
505 		return (error);
506 	}
507 	opt.uio_iovcnt++;
508 #endif
509 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
510 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
511 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
512 	free(u_path, M_TEMP);
513 	return (error);
514 }
515 
516 /*
517  * struct jail_set_args {
518  *	struct iovec *iovp;
519  *	unsigned int iovcnt;
520  *	int flags;
521  * };
522  */
523 int
524 sys_jail_set(struct thread *td, struct jail_set_args *uap)
525 {
526 	struct uio *auio;
527 	int error;
528 
529 	/* Check that we have an even number of iovecs. */
530 	if (uap->iovcnt & 1)
531 		return (EINVAL);
532 
533 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
534 	if (error)
535 		return (error);
536 	error = kern_jail_set(td, auio, uap->flags);
537 	free(auio, M_IOV);
538 	return (error);
539 }
540 
541 #if defined(INET) || defined(INET6)
542 typedef int prison_addr_cmp_t(const void *, const void *);
543 typedef bool prison_addr_valid_t(const void *);
544 static const struct pr_family {
545 	size_t			size;
546 	prison_addr_cmp_t	*cmp;
547 	prison_addr_valid_t	*valid;
548 	int			ip_flag;
549 } pr_families[PR_FAMILY_MAX] = {
550 #ifdef INET
551 	[PR_INET] = {
552 		.size = sizeof(struct in_addr),
553 		.cmp = prison_qcmp_v4,
554 		.valid = prison_valid_v4,
555 		.ip_flag = PR_IP4_USER,
556 	 },
557 #endif
558 #ifdef INET6
559 	[PR_INET6] = {
560 		.size = sizeof(struct in6_addr),
561 		.cmp = prison_qcmp_v6,
562 		.valid = prison_valid_v6,
563 		.ip_flag = PR_IP6_USER,
564 	},
565 #endif
566 };
567 
568 /*
569  * Network address lists (pr_addrs) allocation for jails.  The addresses
570  * are accessed locklessly by the network stack, thus need to be protected by
571  * the network epoch.
572  */
573 struct prison_ip {
574 	struct epoch_context ctx;
575 	uint32_t	ips;
576 #ifdef FUTURE_C
577 	/*
578 	 * XXX Variable-length automatic arrays in union may be
579 	 * supported in future C.
580 	 */
581 	union {
582 		char pr_ip[];
583 		struct in_addr pr_ip4[];
584 		struct in6_addr pr_ip6[];
585 	};
586 #else /* No future C :( */
587 	char pr_ip[];
588 #endif
589 };
590 
591 static char *
592 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
593 {
594 	MPASS(pip);
595 	MPASS(af < PR_FAMILY_MAX);
596 	MPASS(idx >= 0 && idx < pip->ips);
597 
598 	return (pip->pr_ip + pr_families[af].size * idx);
599 }
600 
601 static struct prison_ip *
602 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
603 {
604 	struct prison_ip *pip;
605 
606 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
607 	    M_PRISON, flags);
608 	if (pip != NULL)
609 		pip->ips = cnt;
610 	return (pip);
611 }
612 
613 /*
614  * Allocate and copyin user supplied address list, sorting and validating.
615  * kern_jail_set() helper.
616  */
617 static struct prison_ip *
618 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
619 {
620 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
621 	const size_t size = pr_families[af].size;
622 	struct prison_ip *pip;
623 
624 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
625 	bcopy(op, pip->pr_ip, cnt * size);
626 	/*
627 	 * IP addresses are all sorted but ip[0] to preserve
628 	 * the primary IP address as given from userland.
629 	 * This special IP is used for unbound outgoing
630 	 * connections as well for "loopback" traffic in case
631 	 * source address selection cannot find any more fitting
632 	 * address to connect from.
633 	 */
634 	if (cnt > 1)
635 		qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
636 	/*
637 	 * Check for duplicate addresses and do some simple
638 	 * zero and broadcast checks. If users give other bogus
639 	 * addresses it is their problem.
640 	 */
641 	for (int i = 0; i < cnt; i++) {
642 		if (!pr_families[af].valid(PR_IP(pip, af, i))) {
643 			free(pip, M_PRISON);
644 			return (NULL);
645 		}
646 		if (i + 1 < cnt &&
647 		    (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
648 		     cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
649 			free(pip, M_PRISON);
650 			return (NULL);
651 		}
652 	}
653 
654 	return (pip);
655 }
656 
657 /*
658  * Allocate and dup parent prison address list.
659  * kern_jail_set() helper.
660  */
661 static void
662 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
663 {
664 	const struct prison_ip *ppip = ppr->pr_addrs[af];
665 	struct prison_ip *pip;
666 
667 	if (ppip != NULL) {
668 		pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
669 		bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
670 		pr->pr_addrs[af] = pip;
671 	}
672 }
673 
674 /*
675  * Make sure the new set of IP addresses is a subset of the parent's list.
676  * Don't worry about the parent being unlocked, as any setting is done with
677  * allprison_lock held.
678  * kern_jail_set() helper.
679  */
680 static bool
681 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
682     const pr_family_t af)
683 {
684 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
685 	int i, j;
686 
687 	if (ppip == NULL)
688 		return (false);
689 
690 	for (i = 0; i < ppip->ips; i++)
691 		if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
692 			break;
693 
694 	if (i == ppip->ips)
695 		/* Main address not present in parent. */
696 		return (false);
697 
698 	if (pip->ips > 1) {
699 		for (i = j = 1; i < pip->ips; i++) {
700 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
701 				/* Equals to parent primary address. */
702 				continue;
703 			for (; j < ppip->ips; j++)
704 				if (cmp(PR_IP(pip, af, i),
705 				    PR_IP(ppip, af, j)) == 0)
706 					break;
707 			if (j == ppip->ips)
708 				break;
709 		}
710 		if (j == ppip->ips)
711 			/* Address not present in parent. */
712 			return (false);
713 	}
714 	return (true);
715 }
716 
717 /*
718  * Check for conflicting IP addresses.  We permit them if there is no more
719  * than one IP on each jail.  If there is a duplicate on a jail with more
720  * than one IP stop checking and return error.
721  * kern_jail_set() helper.
722  */
723 static bool
724 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
725     struct prison_ip *pip, pr_family_t af)
726 {
727 	const struct prison *tppr, *tpr;
728 	int descend;
729 
730 #ifdef VIMAGE
731 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
732 		if (tppr->pr_flags & PR_VNET)
733 			break;
734 #else
735 	tppr = &prison0;
736 #endif
737 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
738 		if (tpr == pr ||
739 #ifdef VIMAGE
740 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
741 #endif
742 		    !prison_isalive(tpr)) {
743 			descend = 0;
744 			continue;
745 		}
746 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
747 			continue;
748 		descend = 0;
749 		if (tpr->pr_addrs[af] == NULL ||
750 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
751 			continue;
752 		for (int i = 0; i < pip->ips; i++)
753 			if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
754 				return (false);
755 	}
756 
757 	return (true);
758 }
759 
760 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
761     "prison must start with epoch context");
762 static void
763 prison_ip_free_deferred(epoch_context_t ctx)
764 {
765 
766 	free(ctx, M_PRISON);
767 }
768 
769 static void
770 prison_ip_free(struct prison_ip *pip)
771 {
772 
773 	if (pip != NULL)
774 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
775 }
776 
777 static void
778 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
779 {
780 	struct prison_ip **mem, *old;
781 
782 	mtx_assert(&pr->pr_mtx, MA_OWNED);
783 
784 	mem = &pr->pr_addrs[af];
785 
786 	old = *mem;
787 	atomic_store_ptr(mem, new);
788 	prison_ip_free(old);
789 }
790 
791 /*
792  * Restrict a prison's IP address list with its parent's, possibly replacing
793  * it.  Return true if succeed, otherwise should redo.
794  * kern_jail_set() helper.
795  */
796 static bool
797 prison_ip_restrict(struct prison *pr, const pr_family_t af,
798     struct prison_ip **newp)
799 {
800 	struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
801 	struct prison_ip *pip = pr->pr_addrs[af];
802 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
803 	const size_t size = pr_families[af].size;
804 	struct prison_ip *new = newp != NULL ? *newp : NULL;
805 	uint32_t ips;
806 
807 	mtx_assert(&pr->pr_mtx, MA_OWNED);
808 
809 	/*
810 	 * Due to epoch-synchronized access to the IP address lists we always
811 	 * allocate a new list even if the old one has enough space.  We could
812 	 * atomically update an IPv4 address inside a list, but that would
813 	 * screw up sorting, and in case of IPv6 we can't even atomically write
814 	 * one.
815 	 */
816 	if (ppip == NULL) {
817 		if (pip != NULL)
818 			prison_ip_set(pr, af, NULL);
819 		return (true);
820 	}
821 
822 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
823 		if (new == NULL) {
824 			new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
825 			if (new == NULL)
826 				return (false); /* Redo */
827 		}
828 		/* This has no user settings, so just copy the parent's list. */
829 		MPASS(new->ips == ppip->ips);
830 		bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
831 		prison_ip_set(pr, af, new);
832 		if (newp != NULL)
833 			*newp = NULL; /* Used */
834 	} else if (pip != NULL) {
835 		/* Remove addresses that aren't in the parent. */
836 		int i;
837 
838 		i = 0; /* index in pip */
839 		ips = 0; /* index in new */
840 
841 		if (new == NULL) {
842 			new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
843 			if (new == NULL)
844 				return (false); /* Redo */
845 		}
846 
847 		for (int pi = 0; pi < ppip->ips; pi++)
848 			if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
849 				/* Found our primary address in parent. */
850 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
851 				    size);
852 				i++;
853 				ips++;
854 				break;
855 			}
856 		for (int pi = 1; i < pip->ips; ) {
857 			/* Check against primary, which is unsorted. */
858 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
859 				/* Matches parent's primary address. */
860 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
861 				    size);
862 				i++;
863 				ips++;
864 				continue;
865 			}
866 			/* The rest are sorted. */
867 			switch (pi >= ppip->ips ? -1 :
868 				cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
869 			case -1:
870 				i++;
871 				break;
872 			case 0:
873 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
874 				    size);
875 				i++;
876 				pi++;
877 				ips++;
878 				break;
879 			case 1:
880 				pi++;
881 				break;
882 			}
883 		}
884 		if (ips == 0) {
885 			if (newp == NULL || *newp == NULL)
886 				prison_ip_free(new);
887 			new = NULL;
888 		} else {
889 			/* Shrink to real size */
890 			KASSERT((new->ips >= ips),
891 			    ("Out-of-bounds write to prison_ip %p", new));
892 			new->ips = ips;
893 		}
894 		prison_ip_set(pr, af, new);
895 		if (newp != NULL)
896 			*newp = NULL; /* Used */
897 	}
898 	return (true);
899 }
900 
901 /*
902  * Fast-path check if an address belongs to a prison.
903  */
904 int
905 prison_ip_check(const struct prison *pr, const pr_family_t af,
906     const void *addr)
907 {
908 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
909 	struct prison_ip *pip;
910 	int i, a, z, d;
911 
912 	MPASS(mtx_owned(&pr->pr_mtx) ||
913 	    in_epoch(net_epoch_preempt) ||
914 	    sx_xlocked(&allprison_lock));
915 
916 	pip = atomic_load_ptr(&pr->pr_addrs[af]);
917 	if (__predict_false(pip == NULL))
918 		return (EAFNOSUPPORT);
919 
920 	/* Check the primary IP. */
921 	if (cmp(PR_IP(pip, af, 0), addr) == 0)
922 		return (0);
923 
924 	/*
925 	 * All the other IPs are sorted so we can do a binary search.
926 	 */
927 	a = 0;
928 	z = pip->ips - 2;
929 	while (a <= z) {
930 		i = (a + z) / 2;
931 		d = cmp(PR_IP(pip, af, i + 1), addr);
932 		if (d > 0)
933 			z = i - 1;
934 		else if (d < 0)
935 			a = i + 1;
936 		else
937 			return (0);
938 	}
939 
940 	return (EADDRNOTAVAIL);
941 }
942 
943 /*
944  * Grab primary IP.  Historically required mutex, but nothing prevents
945  * us to support epoch-protected access.  Is it used in fast path?
946  * in{6}_jail.c helper
947  */
948 const void *
949 prison_ip_get0(const struct prison *pr, const pr_family_t af)
950 {
951 	const struct prison_ip *pip = pr->pr_addrs[af];
952 
953 	mtx_assert(&pr->pr_mtx, MA_OWNED);
954 	MPASS(pip);
955 
956 	return (pip->pr_ip);
957 }
958 
959 u_int
960 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
961 {
962 
963 	return (pr->pr_addrs[af]->ips);
964 }
965 #endif	/* defined(INET) || defined(INET6) */
966 
967 int
968 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
969 {
970 	struct nameidata nd;
971 #ifdef INET
972 	struct prison_ip *ip4;
973 #endif
974 #ifdef INET6
975 	struct prison_ip *ip6;
976 #endif
977 	struct vfsopt *opt;
978 	struct vfsoptlist *opts;
979 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
980 	struct vnode *root;
981 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
982 	char *g_path, *osrelstr;
983 	struct bool_flags *bf;
984 	struct jailsys_flags *jsf;
985 #if defined(INET) || defined(INET6)
986 	void *op;
987 #endif
988 	unsigned long hid;
989 	size_t namelen, onamelen, pnamelen;
990 	int born, created, cuflags, descend, drflags, enforce;
991 	int error, errmsg_len, errmsg_pos;
992 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
993 	int jid, jsys, len, level;
994 	int childmax, osreldt, rsnum, slevel;
995 #ifdef INET
996 	int ip4s;
997 	bool redo_ip4;
998 #endif
999 #ifdef INET6
1000 	int ip6s;
1001 	bool redo_ip6;
1002 #endif
1003 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1004 	uint64_t pr_allow_diff;
1005 	unsigned tallow;
1006 	char numbuf[12];
1007 
1008 	error = priv_check(td, PRIV_JAIL_SET);
1009 	if (!error && (flags & JAIL_ATTACH))
1010 		error = priv_check(td, PRIV_JAIL_ATTACH);
1011 	if (error)
1012 		return (error);
1013 	mypr = td->td_ucred->cr_prison;
1014 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
1015 		return (EPERM);
1016 	if (flags & ~JAIL_SET_MASK)
1017 		return (EINVAL);
1018 
1019 	/*
1020 	 * Check all the parameters before committing to anything.  Not all
1021 	 * errors can be caught early, but we may as well try.  Also, this
1022 	 * takes care of some expensive stuff (path lookup) before getting
1023 	 * the allprison lock.
1024 	 *
1025 	 * XXX Jails are not filesystems, and jail parameters are not mount
1026 	 *     options.  But it makes more sense to re-use the vfsopt code
1027 	 *     than duplicate it under a different name.
1028 	 */
1029 	error = vfs_buildopts(optuio, &opts);
1030 	if (error)
1031 		return (error);
1032 #ifdef INET
1033 	ip4 = NULL;
1034 #endif
1035 #ifdef INET6
1036 	ip6 = NULL;
1037 #endif
1038 	g_path = NULL;
1039 
1040 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1041 	if (!cuflags) {
1042 		error = EINVAL;
1043 		vfs_opterror(opts, "no valid operation (create or update)");
1044 		goto done_errmsg;
1045 	}
1046 
1047 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1048 	if (error == ENOENT)
1049 		jid = 0;
1050 	else if (error != 0)
1051 		goto done_free;
1052 
1053 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1054 	if (error == ENOENT)
1055 		gotslevel = 0;
1056 	else if (error != 0)
1057 		goto done_free;
1058 	else
1059 		gotslevel = 1;
1060 
1061 	error =
1062 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1063 	if (error == ENOENT)
1064 		gotchildmax = 0;
1065 	else if (error != 0)
1066 		goto done_free;
1067 	else
1068 		gotchildmax = 1;
1069 
1070 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1071 	if (error == ENOENT)
1072 		gotenforce = 0;
1073 	else if (error != 0)
1074 		goto done_free;
1075 	else if (enforce < 0 || enforce > 2) {
1076 		error = EINVAL;
1077 		goto done_free;
1078 	} else
1079 		gotenforce = 1;
1080 
1081 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1082 	if (error == ENOENT)
1083 		gotrsnum = 0;
1084 	else if (error != 0)
1085 		goto done_free;
1086 	else
1087 		gotrsnum = 1;
1088 
1089 	pr_flags = ch_flags = 0;
1090 	for (bf = pr_flag_bool;
1091 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1092 	     bf++) {
1093 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1094 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1095 	}
1096 	ch_flags |= pr_flags;
1097 	for (jsf = pr_flag_jailsys;
1098 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1099 	     jsf++) {
1100 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1101 		if (error == ENOENT)
1102 			continue;
1103 		if (error != 0)
1104 			goto done_free;
1105 		switch (jsys) {
1106 		case JAIL_SYS_DISABLE:
1107 			if (!jsf->disable) {
1108 				error = EINVAL;
1109 				goto done_free;
1110 			}
1111 			pr_flags |= jsf->disable;
1112 			break;
1113 		case JAIL_SYS_NEW:
1114 			pr_flags |= jsf->new;
1115 			break;
1116 		case JAIL_SYS_INHERIT:
1117 			break;
1118 		default:
1119 			error = EINVAL;
1120 			goto done_free;
1121 		}
1122 		ch_flags |= jsf->new | jsf->disable;
1123 	}
1124 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1125 	    && !(pr_flags & PR_PERSIST)) {
1126 		error = EINVAL;
1127 		vfs_opterror(opts, "new jail must persist or attach");
1128 		goto done_errmsg;
1129 	}
1130 #ifdef VIMAGE
1131 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1132 		error = EINVAL;
1133 		vfs_opterror(opts, "vnet cannot be changed after creation");
1134 		goto done_errmsg;
1135 	}
1136 #endif
1137 #ifdef INET
1138 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1139 		error = EINVAL;
1140 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1141 		goto done_errmsg;
1142 	}
1143 #endif
1144 #ifdef INET6
1145 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1146 		error = EINVAL;
1147 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1148 		goto done_errmsg;
1149 	}
1150 #endif
1151 
1152 	pr_allow = ch_allow = 0;
1153 	for (bf = pr_flag_allow;
1154 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1155 		atomic_load_int(&bf->flag) != 0;
1156 	     bf++) {
1157 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1158 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1159 	}
1160 	ch_allow |= pr_allow;
1161 
1162 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1163 	if (error == ENOENT)
1164 		name = NULL;
1165 	else if (error != 0)
1166 		goto done_free;
1167 	else {
1168 		if (len == 0 || name[len - 1] != '\0') {
1169 			error = EINVAL;
1170 			goto done_free;
1171 		}
1172 		if (len > MAXHOSTNAMELEN) {
1173 			error = ENAMETOOLONG;
1174 			goto done_free;
1175 		}
1176 	}
1177 
1178 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1179 	if (error == ENOENT)
1180 		host = NULL;
1181 	else if (error != 0)
1182 		goto done_free;
1183 	else {
1184 		ch_flags |= PR_HOST;
1185 		pr_flags |= PR_HOST;
1186 		if (len == 0 || host[len - 1] != '\0') {
1187 			error = EINVAL;
1188 			goto done_free;
1189 		}
1190 		if (len > MAXHOSTNAMELEN) {
1191 			error = ENAMETOOLONG;
1192 			goto done_free;
1193 		}
1194 	}
1195 
1196 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1197 	if (error == ENOENT)
1198 		domain = NULL;
1199 	else if (error != 0)
1200 		goto done_free;
1201 	else {
1202 		ch_flags |= PR_HOST;
1203 		pr_flags |= PR_HOST;
1204 		if (len == 0 || domain[len - 1] != '\0') {
1205 			error = EINVAL;
1206 			goto done_free;
1207 		}
1208 		if (len > MAXHOSTNAMELEN) {
1209 			error = ENAMETOOLONG;
1210 			goto done_free;
1211 		}
1212 	}
1213 
1214 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1215 	if (error == ENOENT)
1216 		uuid = NULL;
1217 	else if (error != 0)
1218 		goto done_free;
1219 	else {
1220 		ch_flags |= PR_HOST;
1221 		pr_flags |= PR_HOST;
1222 		if (len == 0 || uuid[len - 1] != '\0') {
1223 			error = EINVAL;
1224 			goto done_free;
1225 		}
1226 		if (len > HOSTUUIDLEN) {
1227 			error = ENAMETOOLONG;
1228 			goto done_free;
1229 		}
1230 	}
1231 
1232 #ifdef COMPAT_FREEBSD32
1233 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1234 		uint32_t hid32;
1235 
1236 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1237 		hid = hid32;
1238 	} else
1239 #endif
1240 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1241 	if (error == ENOENT)
1242 		gothid = 0;
1243 	else if (error != 0)
1244 		goto done_free;
1245 	else {
1246 		gothid = 1;
1247 		ch_flags |= PR_HOST;
1248 		pr_flags |= PR_HOST;
1249 	}
1250 
1251 #ifdef INET
1252 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1253 	if (error == ENOENT)
1254 		ip4s = 0;
1255 	else if (error != 0)
1256 		goto done_free;
1257 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1258 		error = EINVAL;
1259 		goto done_free;
1260 	} else {
1261 		ch_flags |= PR_IP4_USER;
1262 		pr_flags |= PR_IP4_USER;
1263 		if (ip4s > 0) {
1264 			ip4s /= sizeof(struct in_addr);
1265 			if (ip4s > jail_max_af_ips) {
1266 				error = EINVAL;
1267 				vfs_opterror(opts, "too many IPv4 addresses");
1268 				goto done_errmsg;
1269 			}
1270 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1271 			if (ip4 == NULL) {
1272 				error = EINVAL;
1273 				goto done_free;
1274 			}
1275 		}
1276 	}
1277 #endif
1278 
1279 #ifdef INET6
1280 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1281 	if (error == ENOENT)
1282 		ip6s = 0;
1283 	else if (error != 0)
1284 		goto done_free;
1285 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1286 		error = EINVAL;
1287 		goto done_free;
1288 	} else {
1289 		ch_flags |= PR_IP6_USER;
1290 		pr_flags |= PR_IP6_USER;
1291 		if (ip6s > 0) {
1292 			ip6s /= sizeof(struct in6_addr);
1293 			if (ip6s > jail_max_af_ips) {
1294 				error = EINVAL;
1295 				vfs_opterror(opts, "too many IPv6 addresses");
1296 				goto done_errmsg;
1297 			}
1298 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1299 			if (ip6 == NULL) {
1300 				error = EINVAL;
1301 				goto done_free;
1302 			}
1303 		}
1304 	}
1305 #endif
1306 
1307 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1308 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1309 		error = EINVAL;
1310 		vfs_opterror(opts,
1311 		    "vnet jails cannot have IP address restrictions");
1312 		goto done_errmsg;
1313 	}
1314 #endif
1315 
1316 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1317 	if (error == ENOENT)
1318 		osrelstr = NULL;
1319 	else if (error != 0)
1320 		goto done_free;
1321 	else {
1322 		if (flags & JAIL_UPDATE) {
1323 			error = EINVAL;
1324 			vfs_opterror(opts,
1325 			    "osrelease cannot be changed after creation");
1326 			goto done_errmsg;
1327 		}
1328 		if (len == 0 || osrelstr[len - 1] != '\0') {
1329 			error = EINVAL;
1330 			goto done_free;
1331 		}
1332 		if (len >= OSRELEASELEN) {
1333 			error = ENAMETOOLONG;
1334 			vfs_opterror(opts,
1335 			    "osrelease string must be 1-%d bytes long",
1336 			    OSRELEASELEN - 1);
1337 			goto done_errmsg;
1338 		}
1339 	}
1340 
1341 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1342 	if (error == ENOENT)
1343 		osreldt = 0;
1344 	else if (error != 0)
1345 		goto done_free;
1346 	else {
1347 		if (flags & JAIL_UPDATE) {
1348 			error = EINVAL;
1349 			vfs_opterror(opts,
1350 			    "osreldate cannot be changed after creation");
1351 			goto done_errmsg;
1352 		}
1353 		if (osreldt == 0) {
1354 			error = EINVAL;
1355 			vfs_opterror(opts, "osreldate cannot be 0");
1356 			goto done_errmsg;
1357 		}
1358 	}
1359 
1360 	root = NULL;
1361 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1362 	if (error == ENOENT)
1363 		path = NULL;
1364 	else if (error != 0)
1365 		goto done_free;
1366 	else {
1367 		if (flags & JAIL_UPDATE) {
1368 			error = EINVAL;
1369 			vfs_opterror(opts,
1370 			    "path cannot be changed after creation");
1371 			goto done_errmsg;
1372 		}
1373 		if (len == 0 || path[len - 1] != '\0') {
1374 			error = EINVAL;
1375 			goto done_free;
1376 		}
1377 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1378 		error = namei(&nd);
1379 		if (error)
1380 			goto done_free;
1381 		root = nd.ni_vp;
1382 		NDFREE_PNBUF(&nd);
1383 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1384 		strlcpy(g_path, path, MAXPATHLEN);
1385 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1386 		if (error == 0) {
1387 			path = g_path;
1388 		} else {
1389 			/* exit on other errors */
1390 			goto done_free;
1391 		}
1392 		if (root->v_type != VDIR) {
1393 			error = ENOTDIR;
1394 			vput(root);
1395 			goto done_free;
1396 		}
1397 		VOP_UNLOCK(root);
1398 	}
1399 
1400 	/*
1401 	 * Find the specified jail, or at least its parent.
1402 	 * This abuses the file error codes ENOENT and EEXIST.
1403 	 */
1404 	pr = NULL;
1405 	inspr = NULL;
1406 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1407 		namelc = strrchr(name, '.');
1408 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1409 		if (*p != '\0')
1410 			jid = 0;
1411 	}
1412 	sx_xlock(&allprison_lock);
1413 	drflags = PD_LIST_XLOCKED;
1414 	ppr = mypr;
1415 	if (!prison_isalive(ppr)) {
1416 		/* This jail is dying.  This process will surely follow. */
1417 		error = EAGAIN;
1418 		goto done_deref;
1419 	}
1420 	if (jid != 0) {
1421 		if (jid < 0) {
1422 			error = EINVAL;
1423 			vfs_opterror(opts, "negative jid");
1424 			goto done_deref;
1425 		}
1426 		/*
1427 		 * See if a requested jid already exists.  Keep track of
1428 		 * where it can be inserted later.
1429 		 */
1430 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1431 			if (inspr->pr_id < jid)
1432 				continue;
1433 			if (inspr->pr_id > jid)
1434 				break;
1435 			pr = inspr;
1436 			mtx_lock(&pr->pr_mtx);
1437 			drflags |= PD_LOCKED;
1438 			inspr = NULL;
1439 			break;
1440 		}
1441 		if (pr != NULL) {
1442 			/* Create: jid must not exist. */
1443 			if (cuflags == JAIL_CREATE) {
1444 				/*
1445 				 * Even creators that cannot see the jail will
1446 				 * get EEXIST.
1447 				 */
1448 				error = EEXIST;
1449 				vfs_opterror(opts, "jail %d already exists",
1450 				    jid);
1451 				goto done_deref;
1452 			}
1453 			if (!prison_ischild(mypr, pr)) {
1454 				/*
1455 				 * Updaters get ENOENT if they cannot see the
1456 				 * jail.  This is true even for CREATE | UPDATE,
1457 				 * which normally cannot give this error.
1458 				 */
1459 				error = ENOENT;
1460 				vfs_opterror(opts, "jail %d not found", jid);
1461 				goto done_deref;
1462 			}
1463 			ppr = pr->pr_parent;
1464 			if (!prison_isalive(ppr)) {
1465 				error = ENOENT;
1466 				vfs_opterror(opts, "jail %d is dying",
1467 				    ppr->pr_id);
1468 				goto done_deref;
1469 			}
1470 			if (!prison_isalive(pr)) {
1471 				if (!(flags & JAIL_DYING)) {
1472 					error = ENOENT;
1473 					vfs_opterror(opts, "jail %d is dying",
1474 					    jid);
1475 					goto done_deref;
1476 				}
1477 				if ((flags & JAIL_ATTACH) ||
1478 				    (pr_flags & PR_PERSIST)) {
1479 					/*
1480 					 * A dying jail might be resurrected
1481 					 * (via attach or persist), but first
1482 					 * it must determine if another jail
1483 					 * has claimed its name.  Accomplish
1484 					 * this by implicitly re-setting the
1485 					 * name.
1486 					 */
1487 					if (name == NULL)
1488 						name = prison_name(mypr, pr);
1489 				}
1490 			}
1491 		} else {
1492 			/* Update: jid must exist. */
1493 			if (cuflags == JAIL_UPDATE) {
1494 				error = ENOENT;
1495 				vfs_opterror(opts, "jail %d not found", jid);
1496 				goto done_deref;
1497 			}
1498 		}
1499 	}
1500 	/*
1501 	 * If the caller provided a name, look for a jail by that name.
1502 	 * This has different semantics for creates and updates keyed by jid
1503 	 * (where the name must not already exist in a different jail),
1504 	 * and updates keyed by the name itself (where the name must exist
1505 	 * because that is the jail being updated).
1506 	 */
1507 	namelc = NULL;
1508 	if (name != NULL) {
1509 		namelc = strrchr(name, '.');
1510 		if (namelc == NULL)
1511 			namelc = name;
1512 		else {
1513 			/*
1514 			 * This is a hierarchical name.  Split it into the
1515 			 * parent and child names, and make sure the parent
1516 			 * exists or matches an already found jail.
1517 			 */
1518 			if (pr != NULL) {
1519 				if (strncmp(name, ppr->pr_name, namelc - name)
1520 				    || ppr->pr_name[namelc - name] != '\0') {
1521 					error = EINVAL;
1522 					vfs_opterror(opts,
1523 					    "cannot change jail's parent");
1524 					goto done_deref;
1525 				}
1526 			} else {
1527 				*namelc = '\0';
1528 				ppr = prison_find_name(mypr, name);
1529 				if (ppr == NULL) {
1530 					error = ENOENT;
1531 					vfs_opterror(opts,
1532 					    "jail \"%s\" not found", name);
1533 					goto done_deref;
1534 				}
1535 				mtx_unlock(&ppr->pr_mtx);
1536 				if (!prison_isalive(ppr)) {
1537 					error = ENOENT;
1538 					vfs_opterror(opts,
1539 					    "jail \"%s\" is dying", name);
1540 					goto done_deref;
1541 				}
1542 				*namelc = '.';
1543 			}
1544 			namelc++;
1545 		}
1546 		if (namelc[0] != '\0') {
1547 			pnamelen =
1548 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1549 			deadpr = NULL;
1550 			FOREACH_PRISON_CHILD(ppr, tpr) {
1551 				if (tpr != pr &&
1552 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1553 					if (prison_isalive(tpr)) {
1554 						if (pr == NULL &&
1555 						    cuflags != JAIL_CREATE) {
1556 							/*
1557 							 * Use this jail
1558 							 * for updates.
1559 							 */
1560 							pr = tpr;
1561 							mtx_lock(&pr->pr_mtx);
1562 							drflags |= PD_LOCKED;
1563 							break;
1564 						}
1565 						/*
1566 						 * Create, or update(jid):
1567 						 * name must not exist in an
1568 						 * active sibling jail.
1569 						 */
1570 						error = EEXIST;
1571 						vfs_opterror(opts,
1572 						   "jail \"%s\" already exists",
1573 						   name);
1574 						goto done_deref;
1575 					}
1576 					if (pr == NULL &&
1577 					    cuflags != JAIL_CREATE) {
1578 						deadpr = tpr;
1579 					}
1580 				}
1581 			}
1582 			/* If no active jail is found, use a dying one. */
1583 			if (deadpr != NULL && pr == NULL) {
1584 				if (flags & JAIL_DYING) {
1585 					pr = deadpr;
1586 					mtx_lock(&pr->pr_mtx);
1587 					drflags |= PD_LOCKED;
1588 				} else if (cuflags == JAIL_UPDATE) {
1589 					error = ENOENT;
1590 					vfs_opterror(opts,
1591 					    "jail \"%s\" is dying", name);
1592 					goto done_deref;
1593 				}
1594 			}
1595 			/* Update: name must exist if no jid. */
1596 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1597 				error = ENOENT;
1598 				vfs_opterror(opts, "jail \"%s\" not found",
1599 				    name);
1600 				goto done_deref;
1601 			}
1602 		}
1603 	}
1604 	/* Update: must provide a jid or name. */
1605 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1606 		error = ENOENT;
1607 		vfs_opterror(opts, "update specified no jail");
1608 		goto done_deref;
1609 	}
1610 
1611 	/* If there's no prison to update, create a new one and link it in. */
1612 	created = pr == NULL;
1613 	if (created) {
1614 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1615 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1616 				error = EPERM;
1617 				vfs_opterror(opts, "prison limit exceeded");
1618 				goto done_deref;
1619 			}
1620 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1621 			error = EAGAIN;
1622 			vfs_opterror(opts, "no available jail IDs");
1623 			goto done_deref;
1624 		}
1625 
1626 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1627 		pr->pr_state = PRISON_STATE_INVALID;
1628 		refcount_init(&pr->pr_ref, 1);
1629 		refcount_init(&pr->pr_uref, 0);
1630 		drflags |= PD_DEREF;
1631 		LIST_INIT(&pr->pr_children);
1632 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1633 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1634 
1635 		pr->pr_id = jid;
1636 		if (inspr != NULL)
1637 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1638 		else
1639 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1640 
1641 		pr->pr_parent = ppr;
1642 		prison_hold(ppr);
1643 		prison_proc_hold(ppr);
1644 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1645 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1646 			tpr->pr_childcount++;
1647 
1648 		/* Set some default values, and inherit some from the parent. */
1649 		if (namelc == NULL)
1650 			namelc = "";
1651 		if (path == NULL) {
1652 			path = "/";
1653 			root = mypr->pr_root;
1654 			vref(root);
1655 		}
1656 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1657 		pr->pr_flags |= PR_HOST;
1658 #if defined(INET) || defined(INET6)
1659 #ifdef VIMAGE
1660 		if (!(pr_flags & PR_VNET))
1661 #endif
1662 		{
1663 #ifdef INET
1664 			if (!(ch_flags & PR_IP4_USER))
1665 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1666 			else if (!(pr_flags & PR_IP4_USER)) {
1667 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1668 				prison_ip_dup(ppr, pr, PR_INET);
1669 			}
1670 #endif
1671 #ifdef INET6
1672 			if (!(ch_flags & PR_IP6_USER))
1673 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1674 			else if (!(pr_flags & PR_IP6_USER)) {
1675 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1676 				prison_ip_dup(ppr, pr, PR_INET6);
1677 			}
1678 #endif
1679 		}
1680 #endif
1681 		/* Source address selection is always on by default. */
1682 		pr->pr_flags |= _PR_IP_SADDRSEL;
1683 
1684 		pr->pr_securelevel = ppr->pr_securelevel;
1685 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1686 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1687 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1688 
1689 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1690 		if (osrelstr == NULL)
1691 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1692 			    sizeof(pr->pr_osrelease));
1693 		else
1694 			strlcpy(pr->pr_osrelease, osrelstr,
1695 			    sizeof(pr->pr_osrelease));
1696 
1697 #ifdef VIMAGE
1698 		/* Allocate a new vnet if specified. */
1699 		pr->pr_vnet = (pr_flags & PR_VNET)
1700 		    ? vnet_alloc() : ppr->pr_vnet;
1701 #endif
1702 		/*
1703 		 * Allocate a dedicated cpuset for each jail.
1704 		 * Unlike other initial settings, this may return an error.
1705 		 */
1706 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1707 		if (error)
1708 			goto done_deref;
1709 
1710 		mtx_lock(&pr->pr_mtx);
1711 		drflags |= PD_LOCKED;
1712 	} else {
1713 		/*
1714 		 * Grab a reference for existing prisons, to ensure they
1715 		 * continue to exist for the duration of the call.
1716 		 */
1717 		prison_hold(pr);
1718 		drflags |= PD_DEREF;
1719 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1720 		if ((pr->pr_flags & PR_VNET) &&
1721 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1722 			error = EINVAL;
1723 			vfs_opterror(opts,
1724 			    "vnet jails cannot have IP address restrictions");
1725 			goto done_deref;
1726 		}
1727 #endif
1728 #ifdef INET
1729 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1730 			error = EINVAL;
1731 			vfs_opterror(opts,
1732 			    "ip4 cannot be changed after creation");
1733 			goto done_deref;
1734 		}
1735 #endif
1736 #ifdef INET6
1737 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1738 			error = EINVAL;
1739 			vfs_opterror(opts,
1740 			    "ip6 cannot be changed after creation");
1741 			goto done_deref;
1742 		}
1743 #endif
1744 	}
1745 
1746 	/* Do final error checking before setting anything. */
1747 	if (gotslevel) {
1748 		if (slevel < ppr->pr_securelevel) {
1749 			error = EPERM;
1750 			goto done_deref;
1751 		}
1752 	}
1753 	if (gotchildmax) {
1754 		if (childmax >= ppr->pr_childmax) {
1755 			error = EPERM;
1756 			goto done_deref;
1757 		}
1758 	}
1759 	if (gotenforce) {
1760 		if (enforce < ppr->pr_enforce_statfs) {
1761 			error = EPERM;
1762 			goto done_deref;
1763 		}
1764 	}
1765 	if (gotrsnum) {
1766 		/*
1767 		 * devfs_rsnum is a uint16_t
1768 		 */
1769 		if (rsnum < 0 || rsnum > 65535) {
1770 			error = EINVAL;
1771 			goto done_deref;
1772 		}
1773 		/*
1774 		 * Nested jails always inherit parent's devfs ruleset
1775 		 */
1776 		if (jailed(td->td_ucred)) {
1777 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1778 				error = EPERM;
1779 				goto done_deref;
1780 			} else
1781 				rsnum = ppr->pr_devfs_rsnum;
1782 		}
1783 	}
1784 #ifdef INET
1785 	if (ip4s > 0) {
1786 		if ((ppr->pr_flags & PR_IP4) &&
1787 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1788 		    PR_INET)) {
1789 			error = EPERM;
1790 			goto done_deref;
1791 		}
1792 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1793 			error = EADDRINUSE;
1794 			vfs_opterror(opts, "IPv4 addresses clash");
1795 			goto done_deref;
1796 		}
1797 	}
1798 #endif
1799 #ifdef INET6
1800 	if (ip6s > 0) {
1801 		if ((ppr->pr_flags & PR_IP6) &&
1802 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1803 		    PR_INET6)) {
1804 			error = EPERM;
1805 			goto done_deref;
1806 		}
1807 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1808 			error = EADDRINUSE;
1809 			vfs_opterror(opts, "IPv6 addresses clash");
1810 			goto done_deref;
1811 		}
1812 	}
1813 #endif
1814 	onamelen = namelen = 0;
1815 	if (namelc != NULL) {
1816 		/* Give a default name of the jid.  Also allow the name to be
1817 		 * explicitly the jid - but not any other number, and only in
1818 		 * normal form (no leading zero/etc).
1819 		 */
1820 		if (namelc[0] == '\0')
1821 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1822 		else if ((strtoul(namelc, &p, 10) != jid ||
1823 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1824 			error = EINVAL;
1825 			vfs_opterror(opts,
1826 			    "name cannot be numeric (unless it is the jid)");
1827 			goto done_deref;
1828 		}
1829 		/*
1830 		 * Make sure the name isn't too long for the prison or its
1831 		 * children.
1832 		 */
1833 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1834 		onamelen = strlen(pr->pr_name + pnamelen);
1835 		namelen = strlen(namelc);
1836 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1837 			error = ENAMETOOLONG;
1838 			goto done_deref;
1839 		}
1840 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1841 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1842 			    sizeof(pr->pr_name)) {
1843 				error = ENAMETOOLONG;
1844 				goto done_deref;
1845 			}
1846 		}
1847 	}
1848 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1849 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1850 		error = EPERM;
1851 		goto done_deref;
1852 	}
1853 
1854 	/*
1855 	 * Let modules check their parameters.  This requires unlocking and
1856 	 * then re-locking the prison, but this is still a valid state as long
1857 	 * as allprison_lock remains xlocked.
1858 	 */
1859 	mtx_unlock(&pr->pr_mtx);
1860 	drflags &= ~PD_LOCKED;
1861 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1862 	if (error != 0)
1863 		goto done_deref;
1864 	mtx_lock(&pr->pr_mtx);
1865 	drflags |= PD_LOCKED;
1866 
1867 	/* At this point, all valid parameters should have been noted. */
1868 	TAILQ_FOREACH(opt, opts, link) {
1869 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1870 			error = EINVAL;
1871 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1872 			goto done_deref;
1873 		}
1874 	}
1875 
1876 	/* Set the parameters of the prison. */
1877 #ifdef INET
1878 	redo_ip4 = false;
1879 	if (pr_flags & PR_IP4_USER) {
1880 		pr->pr_flags |= PR_IP4;
1881 		prison_ip_set(pr, PR_INET, ip4);
1882 		ip4 = NULL;
1883 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1884 #ifdef VIMAGE
1885 			if (tpr->pr_flags & PR_VNET) {
1886 				descend = 0;
1887 				continue;
1888 			}
1889 #endif
1890 			if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
1891 				redo_ip4 = true;
1892 				descend = 0;
1893 			}
1894 		}
1895 	}
1896 #endif
1897 #ifdef INET6
1898 	redo_ip6 = false;
1899 	if (pr_flags & PR_IP6_USER) {
1900 		pr->pr_flags |= PR_IP6;
1901 		prison_ip_set(pr, PR_INET6, ip6);
1902 		ip6 = NULL;
1903 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1904 #ifdef VIMAGE
1905 			if (tpr->pr_flags & PR_VNET) {
1906 				descend = 0;
1907 				continue;
1908 			}
1909 #endif
1910 			if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
1911 				redo_ip6 = true;
1912 				descend = 0;
1913 			}
1914 		}
1915 	}
1916 #endif
1917 	if (gotslevel) {
1918 		pr->pr_securelevel = slevel;
1919 		/* Set all child jails to be at least this level. */
1920 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1921 			if (tpr->pr_securelevel < slevel)
1922 				tpr->pr_securelevel = slevel;
1923 	}
1924 	if (gotchildmax) {
1925 		pr->pr_childmax = childmax;
1926 		/* Set all child jails to under this limit. */
1927 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1928 			if (tpr->pr_childmax > childmax - level)
1929 				tpr->pr_childmax = childmax > level
1930 				    ? childmax - level : 0;
1931 	}
1932 	if (gotenforce) {
1933 		pr->pr_enforce_statfs = enforce;
1934 		/* Pass this restriction on to the children. */
1935 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1936 			if (tpr->pr_enforce_statfs < enforce)
1937 				tpr->pr_enforce_statfs = enforce;
1938 	}
1939 	if (gotrsnum) {
1940 		pr->pr_devfs_rsnum = rsnum;
1941 		/* Pass this restriction on to the children. */
1942 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1943 			tpr->pr_devfs_rsnum = rsnum;
1944 	}
1945 	if (namelc != NULL) {
1946 		if (ppr == &prison0)
1947 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1948 		else
1949 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1950 			    ppr->pr_name, namelc);
1951 		/* Change this component of child names. */
1952 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1953 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1954 			    strlen(tpr->pr_name + onamelen) + 1);
1955 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1956 		}
1957 	}
1958 	if (path != NULL) {
1959 		/* Try to keep a real-rooted full pathname. */
1960 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1961 		pr->pr_root = root;
1962 		root = NULL;
1963 	}
1964 	if (PR_HOST & ch_flags & ~pr_flags) {
1965 		if (pr->pr_flags & PR_HOST) {
1966 			/*
1967 			 * Copy the parent's host info.  As with pr_ip4 above,
1968 			 * the lack of a lock on the parent is not a problem;
1969 			 * it is always set with allprison_lock at least
1970 			 * shared, and is held exclusively here.
1971 			 */
1972 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1973 			    sizeof(pr->pr_hostname));
1974 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1975 			    sizeof(pr->pr_domainname));
1976 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1977 			    sizeof(pr->pr_hostuuid));
1978 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1979 		}
1980 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1981 		/* Set this prison, and any descendants without PR_HOST. */
1982 		if (host != NULL)
1983 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1984 		if (domain != NULL)
1985 			strlcpy(pr->pr_domainname, domain,
1986 			    sizeof(pr->pr_domainname));
1987 		if (uuid != NULL)
1988 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1989 		if (gothid)
1990 			pr->pr_hostid = hid;
1991 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1992 			if (tpr->pr_flags & PR_HOST)
1993 				descend = 0;
1994 			else {
1995 				if (host != NULL)
1996 					strlcpy(tpr->pr_hostname,
1997 					    pr->pr_hostname,
1998 					    sizeof(tpr->pr_hostname));
1999 				if (domain != NULL)
2000 					strlcpy(tpr->pr_domainname,
2001 					    pr->pr_domainname,
2002 					    sizeof(tpr->pr_domainname));
2003 				if (uuid != NULL)
2004 					strlcpy(tpr->pr_hostuuid,
2005 					    pr->pr_hostuuid,
2006 					    sizeof(tpr->pr_hostuuid));
2007 				if (gothid)
2008 					tpr->pr_hostid = hid;
2009 			}
2010 		}
2011 	}
2012 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2013 	if ((tallow = ch_allow & ~pr_allow))
2014 		prison_set_allow_locked(pr, tallow, 0);
2015 	/*
2016 	 * Persistent prisons get an extra reference, and prisons losing their
2017 	 * persist flag lose that reference.
2018 	 */
2019 	born = !prison_isalive(pr);
2020 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2021 		if (pr_flags & PR_PERSIST) {
2022 			prison_hold(pr);
2023 			/*
2024 			 * This may make a dead prison alive again, but wait
2025 			 * to label it as such until after OSD calls have had
2026 			 * a chance to run (and perhaps to fail).
2027 			 */
2028 			refcount_acquire(&pr->pr_uref);
2029 		} else {
2030 			drflags |= PD_DEUREF;
2031 			prison_free_not_last(pr);
2032 		}
2033 	}
2034 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2035 	mtx_unlock(&pr->pr_mtx);
2036 	drflags &= ~PD_LOCKED;
2037 	/*
2038 	 * Any errors past this point will need to de-persist newly created
2039 	 * prisons, as well as call remove methods.
2040 	 */
2041 	if (born)
2042 		drflags |= PD_KILL;
2043 
2044 #ifdef RACCT
2045 	if (racct_enable && created)
2046 		prison_racct_attach(pr);
2047 #endif
2048 
2049 	/* Locks may have prevented a complete restriction of child IP
2050 	 * addresses.  If so, allocate some more memory and try again.
2051 	 */
2052 #ifdef INET
2053 	while (redo_ip4) {
2054 		ip4s = pr->pr_addrs[PR_INET]->ips;
2055 		MPASS(ip4 == NULL);
2056 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2057 		mtx_lock(&pr->pr_mtx);
2058 		redo_ip4 = false;
2059 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2060 #ifdef VIMAGE
2061 			if (tpr->pr_flags & PR_VNET) {
2062 				descend = 0;
2063 				continue;
2064 			}
2065 #endif
2066 			if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2067 				redo_ip4 = true;
2068 		}
2069 		mtx_unlock(&pr->pr_mtx);
2070 	}
2071 #endif
2072 #ifdef INET6
2073 	while (redo_ip6) {
2074 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2075 		MPASS(ip6 == NULL);
2076 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2077 		mtx_lock(&pr->pr_mtx);
2078 		redo_ip6 = false;
2079 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2080 #ifdef VIMAGE
2081 			if (tpr->pr_flags & PR_VNET) {
2082 				descend = 0;
2083 				continue;
2084 			}
2085 #endif
2086 			if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2087 				redo_ip6 = true;
2088 		}
2089 		mtx_unlock(&pr->pr_mtx);
2090 	}
2091 #endif
2092 
2093 	/* Let the modules do their work. */
2094 	if (born) {
2095 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2096 		if (error)
2097 			goto done_deref;
2098 	}
2099 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2100 	if (error)
2101 		goto done_deref;
2102 
2103 	/*
2104 	 * A new prison is now ready to be seen; either it has gained a user
2105 	 * reference via persistence, or is about to gain one via attachment.
2106 	 */
2107 	if (born) {
2108 		drflags = prison_lock_xlock(pr, drflags);
2109 		pr->pr_state = PRISON_STATE_ALIVE;
2110 	}
2111 
2112 	/* Attach this process to the prison if requested. */
2113 	if (flags & JAIL_ATTACH) {
2114 		error = do_jail_attach(td, pr,
2115 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2116 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2117 		if (error) {
2118 			vfs_opterror(opts, "attach failed");
2119 			goto done_deref;
2120 		}
2121 	}
2122 
2123 #ifdef RACCT
2124 	if (racct_enable && !created) {
2125 		if (drflags & PD_LOCKED) {
2126 			mtx_unlock(&pr->pr_mtx);
2127 			drflags &= ~PD_LOCKED;
2128 		}
2129 		if (drflags & PD_LIST_XLOCKED) {
2130 			sx_xunlock(&allprison_lock);
2131 			drflags &= ~PD_LIST_XLOCKED;
2132 		}
2133 		prison_racct_modify(pr);
2134 	}
2135 #endif
2136 
2137 	if (born && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2138 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2139 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2140 		   " file system\n", pr->pr_id);
2141 
2142 	drflags &= ~PD_KILL;
2143 	td->td_retval[0] = pr->pr_id;
2144 
2145  done_deref:
2146 	/* Release any temporary prison holds and/or locks. */
2147 	if (pr != NULL)
2148 		prison_deref(pr, drflags);
2149 	else if (drflags & PD_LIST_SLOCKED)
2150 		sx_sunlock(&allprison_lock);
2151 	else if (drflags & PD_LIST_XLOCKED)
2152 		sx_xunlock(&allprison_lock);
2153 	if (root != NULL)
2154 		vrele(root);
2155  done_errmsg:
2156 	if (error) {
2157 		/* Write the error message back to userspace. */
2158 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2159 		    &errmsg_len) == 0 && errmsg_len > 0) {
2160 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2161 			if (optuio->uio_segflg == UIO_SYSSPACE)
2162 				bcopy(errmsg,
2163 				    optuio->uio_iov[errmsg_pos].iov_base,
2164 				    errmsg_len);
2165 			else
2166 				copyout(errmsg,
2167 				    optuio->uio_iov[errmsg_pos].iov_base,
2168 				    errmsg_len);
2169 		}
2170 	}
2171  done_free:
2172 #ifdef INET
2173 	prison_ip_free(ip4);
2174 #endif
2175 #ifdef INET6
2176 	prison_ip_free(ip6);
2177 #endif
2178 	if (g_path != NULL)
2179 		free(g_path, M_TEMP);
2180 	vfs_freeopts(opts);
2181 	return (error);
2182 }
2183 
2184 /*
2185  * Find the next available prison ID.  Return the ID on success, or zero
2186  * on failure.  Also set a pointer to the allprison list entry the prison
2187  * should be inserted before.
2188  */
2189 static int
2190 get_next_prid(struct prison **insprp)
2191 {
2192 	struct prison *inspr;
2193 	int jid, maxid;
2194 
2195 	jid = lastprid % JAIL_MAX + 1;
2196 	if (TAILQ_EMPTY(&allprison) ||
2197 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2198 		/*
2199 		 * A common case is for all jails to be implicitly numbered,
2200 		 * which means they'll go on the end of the list, at least
2201 		 * for the first JAIL_MAX times.
2202 		 */
2203 		inspr = NULL;
2204 	} else {
2205 		/*
2206 		 * Take two passes through the allprison list: first starting
2207 		 * with the proposed jid, then ending with it.
2208 		 */
2209 		for (maxid = JAIL_MAX; maxid != 0; ) {
2210 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2211 				if (inspr->pr_id < jid)
2212 					continue;
2213 				if (inspr->pr_id > jid) {
2214 					/* Found an opening. */
2215 					maxid = 0;
2216 					break;
2217 				}
2218 				if (++jid > maxid) {
2219 					if (lastprid == maxid || lastprid == 0)
2220 					{
2221 						/*
2222 						 * The entire legal range
2223 						 * has been traversed
2224 						 */
2225 						return 0;
2226 					}
2227 					/* Try again from the start. */
2228 					jid = 1;
2229 					maxid = lastprid;
2230 					break;
2231 				}
2232 			}
2233 			if (inspr == NULL) {
2234 				/* Found room at the end of the list. */
2235 				break;
2236 			}
2237 		}
2238 	}
2239 	*insprp = inspr;
2240 	lastprid = jid;
2241 	return (jid);
2242 }
2243 
2244 /*
2245  * struct jail_get_args {
2246  *	struct iovec *iovp;
2247  *	unsigned int iovcnt;
2248  *	int flags;
2249  * };
2250  */
2251 int
2252 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2253 {
2254 	struct uio *auio;
2255 	int error;
2256 
2257 	/* Check that we have an even number of iovecs. */
2258 	if (uap->iovcnt & 1)
2259 		return (EINVAL);
2260 
2261 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2262 	if (error)
2263 		return (error);
2264 	error = kern_jail_get(td, auio, uap->flags);
2265 	if (error == 0)
2266 		error = copyout(auio->uio_iov, uap->iovp,
2267 		    uap->iovcnt * sizeof (struct iovec));
2268 	free(auio, M_IOV);
2269 	return (error);
2270 }
2271 
2272 int
2273 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2274 {
2275 	struct bool_flags *bf;
2276 	struct jailsys_flags *jsf;
2277 	struct prison *pr, *mypr;
2278 	struct vfsopt *opt;
2279 	struct vfsoptlist *opts;
2280 	char *errmsg, *name;
2281 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2282 	unsigned f;
2283 
2284 	if (flags & ~JAIL_GET_MASK)
2285 		return (EINVAL);
2286 
2287 	/* Get the parameter list. */
2288 	error = vfs_buildopts(optuio, &opts);
2289 	if (error)
2290 		return (error);
2291 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2292 	mypr = td->td_ucred->cr_prison;
2293 	pr = NULL;
2294 
2295 	/*
2296 	 * Find the prison specified by one of: lastjid, jid, name.
2297 	 */
2298 	sx_slock(&allprison_lock);
2299 	drflags = PD_LIST_SLOCKED;
2300 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2301 	if (error == 0) {
2302 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2303 			if (pr->pr_id > jid &&
2304 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2305 			    prison_ischild(mypr, pr)) {
2306 				mtx_lock(&pr->pr_mtx);
2307 				drflags |= PD_LOCKED;
2308 				goto found_prison;
2309 			}
2310 		}
2311 		error = ENOENT;
2312 		vfs_opterror(opts, "no jail after %d", jid);
2313 		goto done;
2314 	} else if (error != ENOENT)
2315 		goto done;
2316 
2317 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2318 	if (error == 0) {
2319 		if (jid != 0) {
2320 			pr = prison_find_child(mypr, jid);
2321 			if (pr != NULL) {
2322 				drflags |= PD_LOCKED;
2323 				if (!(prison_isalive(pr) ||
2324 				    (flags & JAIL_DYING))) {
2325 					error = ENOENT;
2326 					vfs_opterror(opts, "jail %d is dying",
2327 					    jid);
2328 					goto done;
2329 				}
2330 				goto found_prison;
2331 			}
2332 			error = ENOENT;
2333 			vfs_opterror(opts, "jail %d not found", jid);
2334 			goto done;
2335 		}
2336 	} else if (error != ENOENT)
2337 		goto done;
2338 
2339 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2340 	if (error == 0) {
2341 		if (len == 0 || name[len - 1] != '\0') {
2342 			error = EINVAL;
2343 			goto done;
2344 		}
2345 		pr = prison_find_name(mypr, name);
2346 		if (pr != NULL) {
2347 			drflags |= PD_LOCKED;
2348 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2349 				error = ENOENT;
2350 				vfs_opterror(opts, "jail \"%s\" is dying",
2351 				    name);
2352 				goto done;
2353 			}
2354 			goto found_prison;
2355 		}
2356 		error = ENOENT;
2357 		vfs_opterror(opts, "jail \"%s\" not found", name);
2358 		goto done;
2359 	} else if (error != ENOENT)
2360 		goto done;
2361 
2362 	vfs_opterror(opts, "no jail specified");
2363 	error = ENOENT;
2364 	goto done;
2365 
2366  found_prison:
2367 	/* Get the parameters of the prison. */
2368 	prison_hold(pr);
2369 	drflags |= PD_DEREF;
2370 	td->td_retval[0] = pr->pr_id;
2371 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2372 	if (error != 0 && error != ENOENT)
2373 		goto done;
2374 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2375 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2376 	if (error != 0 && error != ENOENT)
2377 		goto done;
2378 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2379 	if (error != 0 && error != ENOENT)
2380 		goto done;
2381 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2382 	    sizeof(pr->pr_cpuset->cs_id));
2383 	if (error != 0 && error != ENOENT)
2384 		goto done;
2385 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2386 	if (error != 0 && error != ENOENT)
2387 		goto done;
2388 #ifdef INET
2389 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2390 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2391 	    pr_families[PR_INET].size : 0 );
2392 	if (error != 0 && error != ENOENT)
2393 		goto done;
2394 #endif
2395 #ifdef INET6
2396 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2397 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2398 	    pr_families[PR_INET6].size : 0 );
2399 	if (error != 0 && error != ENOENT)
2400 		goto done;
2401 #endif
2402 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2403 	    sizeof(pr->pr_securelevel));
2404 	if (error != 0 && error != ENOENT)
2405 		goto done;
2406 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2407 	    sizeof(pr->pr_childcount));
2408 	if (error != 0 && error != ENOENT)
2409 		goto done;
2410 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2411 	    sizeof(pr->pr_childmax));
2412 	if (error != 0 && error != ENOENT)
2413 		goto done;
2414 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2415 	if (error != 0 && error != ENOENT)
2416 		goto done;
2417 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2418 	if (error != 0 && error != ENOENT)
2419 		goto done;
2420 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2421 	if (error != 0 && error != ENOENT)
2422 		goto done;
2423 #ifdef COMPAT_FREEBSD32
2424 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2425 		uint32_t hid32 = pr->pr_hostid;
2426 
2427 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2428 	} else
2429 #endif
2430 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2431 	    sizeof(pr->pr_hostid));
2432 	if (error != 0 && error != ENOENT)
2433 		goto done;
2434 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2435 	    sizeof(pr->pr_enforce_statfs));
2436 	if (error != 0 && error != ENOENT)
2437 		goto done;
2438 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2439 	    sizeof(pr->pr_devfs_rsnum));
2440 	if (error != 0 && error != ENOENT)
2441 		goto done;
2442 	for (bf = pr_flag_bool;
2443 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2444 	     bf++) {
2445 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2446 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2447 		if (error != 0 && error != ENOENT)
2448 			goto done;
2449 		i = !i;
2450 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2451 		if (error != 0 && error != ENOENT)
2452 			goto done;
2453 	}
2454 	for (jsf = pr_flag_jailsys;
2455 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2456 	     jsf++) {
2457 		f = pr->pr_flags & (jsf->disable | jsf->new);
2458 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2459 		    : (f == jsf->new) ? JAIL_SYS_NEW
2460 		    : JAIL_SYS_INHERIT;
2461 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2462 		if (error != 0 && error != ENOENT)
2463 			goto done;
2464 	}
2465 	for (bf = pr_flag_allow;
2466 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2467 		atomic_load_int(&bf->flag) != 0;
2468 	     bf++) {
2469 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2470 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2471 		if (error != 0 && error != ENOENT)
2472 			goto done;
2473 		i = !i;
2474 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2475 		if (error != 0 && error != ENOENT)
2476 			goto done;
2477 	}
2478 	i = !prison_isalive(pr);
2479 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2480 	if (error != 0 && error != ENOENT)
2481 		goto done;
2482 	i = !i;
2483 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2484 	if (error != 0 && error != ENOENT)
2485 		goto done;
2486 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2487 	    sizeof(pr->pr_osreldate));
2488 	if (error != 0 && error != ENOENT)
2489 		goto done;
2490 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2491 	if (error != 0 && error != ENOENT)
2492 		goto done;
2493 
2494 	/* Get the module parameters. */
2495 	mtx_unlock(&pr->pr_mtx);
2496 	drflags &= ~PD_LOCKED;
2497 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2498 	if (error)
2499 		goto done;
2500 	prison_deref(pr, drflags);
2501 	pr = NULL;
2502 	drflags = 0;
2503 
2504 	/* By now, all parameters should have been noted. */
2505 	TAILQ_FOREACH(opt, opts, link) {
2506 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2507 			error = EINVAL;
2508 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2509 			goto done;
2510 		}
2511 	}
2512 
2513 	/* Write the fetched parameters back to userspace. */
2514 	error = 0;
2515 	TAILQ_FOREACH(opt, opts, link) {
2516 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2517 			pos = 2 * opt->pos + 1;
2518 			optuio->uio_iov[pos].iov_len = opt->len;
2519 			if (opt->value != NULL) {
2520 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2521 					bcopy(opt->value,
2522 					    optuio->uio_iov[pos].iov_base,
2523 					    opt->len);
2524 				} else {
2525 					error = copyout(opt->value,
2526 					    optuio->uio_iov[pos].iov_base,
2527 					    opt->len);
2528 					if (error)
2529 						break;
2530 				}
2531 			}
2532 		}
2533 	}
2534 
2535  done:
2536 	/* Release any temporary prison holds and/or locks. */
2537 	if (pr != NULL)
2538 		prison_deref(pr, drflags);
2539 	else if (drflags & PD_LIST_SLOCKED)
2540 		sx_sunlock(&allprison_lock);
2541 	if (error && errmsg_pos >= 0) {
2542 		/* Write the error message back to userspace. */
2543 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2544 		errmsg_pos = 2 * errmsg_pos + 1;
2545 		if (errmsg_len > 0) {
2546 			if (optuio->uio_segflg == UIO_SYSSPACE)
2547 				bcopy(errmsg,
2548 				    optuio->uio_iov[errmsg_pos].iov_base,
2549 				    errmsg_len);
2550 			else
2551 				copyout(errmsg,
2552 				    optuio->uio_iov[errmsg_pos].iov_base,
2553 				    errmsg_len);
2554 		}
2555 	}
2556 	vfs_freeopts(opts);
2557 	return (error);
2558 }
2559 
2560 /*
2561  * struct jail_remove_args {
2562  *	int jid;
2563  * };
2564  */
2565 int
2566 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2567 {
2568 	struct prison *pr;
2569 	int error;
2570 
2571 	error = priv_check(td, PRIV_JAIL_REMOVE);
2572 	if (error)
2573 		return (error);
2574 
2575 	sx_xlock(&allprison_lock);
2576 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2577 	if (pr == NULL) {
2578 		sx_xunlock(&allprison_lock);
2579 		return (EINVAL);
2580 	}
2581 	if (!prison_isalive(pr)) {
2582 		/* Silently ignore already-dying prisons. */
2583 		mtx_unlock(&pr->pr_mtx);
2584 		sx_xunlock(&allprison_lock);
2585 		return (0);
2586 	}
2587 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2588 	return (0);
2589 }
2590 
2591 /*
2592  * struct jail_attach_args {
2593  *	int jid;
2594  * };
2595  */
2596 int
2597 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2598 {
2599 	struct prison *pr;
2600 	int error;
2601 
2602 	error = priv_check(td, PRIV_JAIL_ATTACH);
2603 	if (error)
2604 		return (error);
2605 
2606 	sx_slock(&allprison_lock);
2607 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2608 	if (pr == NULL) {
2609 		sx_sunlock(&allprison_lock);
2610 		return (EINVAL);
2611 	}
2612 
2613 	/* Do not allow a process to attach to a prison that is not alive. */
2614 	if (!prison_isalive(pr)) {
2615 		mtx_unlock(&pr->pr_mtx);
2616 		sx_sunlock(&allprison_lock);
2617 		return (EINVAL);
2618 	}
2619 
2620 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2621 }
2622 
2623 static int
2624 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2625 {
2626 	struct proc *p;
2627 	struct ucred *newcred, *oldcred;
2628 	int error;
2629 
2630 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2631 	sx_assert(&allprison_lock, SX_LOCKED);
2632 	drflags &= PD_LOCK_FLAGS;
2633 	/*
2634 	 * XXX: Note that there is a slight race here if two threads
2635 	 * in the same privileged process attempt to attach to two
2636 	 * different jails at the same time.  It is important for
2637 	 * user processes not to do this, or they might end up with
2638 	 * a process root from one prison, but attached to the jail
2639 	 * of another.
2640 	 */
2641 	prison_hold(pr);
2642 	refcount_acquire(&pr->pr_uref);
2643 	drflags |= PD_DEREF | PD_DEUREF;
2644 	mtx_unlock(&pr->pr_mtx);
2645 	drflags &= ~PD_LOCKED;
2646 
2647 	/* Let modules do whatever they need to prepare for attaching. */
2648 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2649 	if (error) {
2650 		prison_deref(pr, drflags);
2651 		return (error);
2652 	}
2653 	sx_unlock(&allprison_lock);
2654 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2655 
2656 	/*
2657 	 * Reparent the newly attached process to this jail.
2658 	 */
2659 	p = td->td_proc;
2660 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2661 	if (error)
2662 		goto e_revert_osd;
2663 
2664 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2665 	if ((error = change_dir(pr->pr_root, td)) != 0)
2666 		goto e_unlock;
2667 #ifdef MAC
2668 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2669 		goto e_unlock;
2670 #endif
2671 	VOP_UNLOCK(pr->pr_root);
2672 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2673 		goto e_revert_osd;
2674 
2675 	newcred = crget();
2676 	PROC_LOCK(p);
2677 	oldcred = crcopysafe(p, newcred);
2678 	newcred->cr_prison = pr;
2679 	proc_set_cred(p, newcred);
2680 	setsugid(p);
2681 #ifdef RACCT
2682 	racct_proc_ucred_changed(p, oldcred, newcred);
2683 	crhold(newcred);
2684 #endif
2685 	PROC_UNLOCK(p);
2686 #ifdef RCTL
2687 	rctl_proc_ucred_changed(p, newcred);
2688 	crfree(newcred);
2689 #endif
2690 	prison_proc_relink(oldcred->cr_prison, pr, p);
2691 	prison_deref(oldcred->cr_prison, drflags);
2692 	crfree(oldcred);
2693 
2694 	/*
2695 	 * If the prison was killed while changing credentials, die along
2696 	 * with it.
2697 	 */
2698 	if (!prison_isalive(pr)) {
2699 		PROC_LOCK(p);
2700 		kern_psignal(p, SIGKILL);
2701 		PROC_UNLOCK(p);
2702 	}
2703 
2704 	return (0);
2705 
2706  e_unlock:
2707 	VOP_UNLOCK(pr->pr_root);
2708  e_revert_osd:
2709 	/* Tell modules this thread is still in its old jail after all. */
2710 	sx_slock(&allprison_lock);
2711 	drflags |= PD_LIST_SLOCKED;
2712 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2713 	prison_deref(pr, drflags);
2714 	return (error);
2715 }
2716 
2717 /*
2718  * Returns a locked prison instance, or NULL on failure.
2719  */
2720 struct prison *
2721 prison_find(int prid)
2722 {
2723 	struct prison *pr;
2724 
2725 	sx_assert(&allprison_lock, SX_LOCKED);
2726 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2727 		if (pr->pr_id < prid)
2728 			continue;
2729 		if (pr->pr_id > prid)
2730 			break;
2731 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2732 		mtx_lock(&pr->pr_mtx);
2733 		return (pr);
2734 	}
2735 	return (NULL);
2736 }
2737 
2738 /*
2739  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2740  */
2741 struct prison *
2742 prison_find_child(struct prison *mypr, int prid)
2743 {
2744 	struct prison *pr;
2745 	int descend;
2746 
2747 	sx_assert(&allprison_lock, SX_LOCKED);
2748 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2749 		if (pr->pr_id == prid) {
2750 			KASSERT(prison_isvalid(pr),
2751 			    ("Found invalid prison %p", pr));
2752 			mtx_lock(&pr->pr_mtx);
2753 			return (pr);
2754 		}
2755 	}
2756 	return (NULL);
2757 }
2758 
2759 /*
2760  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2761  */
2762 struct prison *
2763 prison_find_name(struct prison *mypr, const char *name)
2764 {
2765 	struct prison *pr, *deadpr;
2766 	size_t mylen;
2767 	int descend;
2768 
2769 	sx_assert(&allprison_lock, SX_LOCKED);
2770 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2771 	deadpr = NULL;
2772 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2773 		if (!strcmp(pr->pr_name + mylen, name)) {
2774 			KASSERT(prison_isvalid(pr),
2775 			    ("Found invalid prison %p", pr));
2776 			if (prison_isalive(pr)) {
2777 				mtx_lock(&pr->pr_mtx);
2778 				return (pr);
2779 			}
2780 			deadpr = pr;
2781 		}
2782 	}
2783 	/* There was no valid prison - perhaps there was a dying one. */
2784 	if (deadpr != NULL)
2785 		mtx_lock(&deadpr->pr_mtx);
2786 	return (deadpr);
2787 }
2788 
2789 /*
2790  * See if a prison has the specific flag set.  The prison should be locked,
2791  * unless checking for flags that are only set at jail creation (such as
2792  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2793  * to any other prison data.
2794  */
2795 bool
2796 prison_flag(struct ucred *cred, unsigned flag)
2797 {
2798 
2799 	return ((cred->cr_prison->pr_flags & flag) != 0);
2800 }
2801 
2802 /*
2803  * See if a prison has the specific allow flag set.
2804  * The prison *should* be locked, or only a single bit is examined, without
2805  * regard to any other prison data.
2806  */
2807 bool
2808 prison_allow(struct ucred *cred, unsigned flag)
2809 {
2810 
2811 	return ((cred->cr_prison->pr_allow & flag) != 0);
2812 }
2813 
2814 /*
2815  * Hold a prison reference, by incrementing pr_ref.  It is generally
2816  * an error to hold a prison that does not already have a reference.
2817  * A prison record will remain valid as long as it has at least one
2818  * reference, and will not be removed as long as either the prison
2819  * mutex or the allprison lock is held (allprison_lock may be shared).
2820  */
2821 void
2822 prison_hold_locked(struct prison *pr)
2823 {
2824 
2825 	/* Locking is no longer required. */
2826 	prison_hold(pr);
2827 }
2828 
2829 void
2830 prison_hold(struct prison *pr)
2831 {
2832 #ifdef INVARIANTS
2833 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2834 
2835 	KASSERT(was_valid,
2836 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2837 #else
2838 	refcount_acquire(&pr->pr_ref);
2839 #endif
2840 }
2841 
2842 /*
2843  * Remove a prison reference.  If that was the last reference, the
2844  * prison will be removed (at a later time).
2845  */
2846 void
2847 prison_free_locked(struct prison *pr)
2848 {
2849 
2850 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2851 	/*
2852 	 * Locking is no longer required, but unlock because the caller
2853 	 * expects it.
2854 	 */
2855 	mtx_unlock(&pr->pr_mtx);
2856 	prison_free(pr);
2857 }
2858 
2859 void
2860 prison_free(struct prison *pr)
2861 {
2862 
2863 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2864 	    ("Trying to free dead prison %p (jid=%d).",
2865 	     pr, pr->pr_id));
2866 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2867 		/*
2868 		 * Don't remove the last reference in this context,
2869 		 * in case there are locks held.
2870 		 */
2871 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2872 	}
2873 }
2874 
2875 static void
2876 prison_free_not_last(struct prison *pr)
2877 {
2878 #ifdef INVARIANTS
2879 	int lastref;
2880 
2881 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2882 	    ("Trying to free dead prison %p (jid=%d).",
2883 	     pr, pr->pr_id));
2884 	lastref = refcount_release(&pr->pr_ref);
2885 	KASSERT(!lastref,
2886 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2887 	     pr, pr->pr_id));
2888 #else
2889 	refcount_release(&pr->pr_ref);
2890 #endif
2891 }
2892 
2893 /*
2894  * Hold a prison for user visibility, by incrementing pr_uref.
2895  * It is generally an error to hold a prison that isn't already
2896  * user-visible, except through the jail system calls.  It is also
2897  * an error to hold an invalid prison.  A prison record will remain
2898  * alive as long as it has at least one user reference, and will not
2899  * be set to the dying state until the prison mutex and allprison_lock
2900  * are both freed.
2901  */
2902 void
2903 prison_proc_hold(struct prison *pr)
2904 {
2905 #ifdef INVARIANTS
2906 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2907 
2908 	KASSERT(was_alive,
2909 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2910 #else
2911 	refcount_acquire(&pr->pr_uref);
2912 #endif
2913 }
2914 
2915 /*
2916  * Remove a prison user reference.  If it was the last reference, the
2917  * prison will be considered "dying", and may be removed once all of
2918  * its references are dropped.
2919  */
2920 void
2921 prison_proc_free(struct prison *pr)
2922 {
2923 
2924 	/*
2925 	 * Locking is only required when releasing the last reference.
2926 	 * This allows assurance that a locked prison will remain alive
2927 	 * until it is unlocked.
2928 	 */
2929 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2930 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2931 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2932 		/*
2933 		 * Don't remove the last user reference in this context,
2934 		 * which is expected to be a process that is not only locked,
2935 		 * but also half dead.  Add a reference so any calls to
2936 		 * prison_free() won't re-submit the task.
2937 		 */
2938 		prison_hold(pr);
2939 		mtx_lock(&pr->pr_mtx);
2940 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2941 		    ("Redundant last reference in prison_proc_free (jid=%d)",
2942 		     pr->pr_id));
2943 		pr->pr_flags |= PR_COMPLETE_PROC;
2944 		mtx_unlock(&pr->pr_mtx);
2945 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2946 	}
2947 }
2948 
2949 static void
2950 prison_proc_free_not_last(struct prison *pr)
2951 {
2952 #ifdef INVARIANTS
2953 	int lastref;
2954 
2955 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2956 	    ("Trying to free dead prison %p (jid=%d).",
2957 	     pr, pr->pr_id));
2958 	lastref = refcount_release(&pr->pr_uref);
2959 	KASSERT(!lastref,
2960 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2961 	     pr, pr->pr_id));
2962 #else
2963 	refcount_release(&pr->pr_uref);
2964 #endif
2965 }
2966 
2967 void
2968 prison_proc_link(struct prison *pr, struct proc *p)
2969 {
2970 
2971 	sx_assert(&allproc_lock, SA_XLOCKED);
2972 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
2973 }
2974 
2975 void
2976 prison_proc_unlink(struct prison *pr, struct proc *p)
2977 {
2978 
2979 	sx_assert(&allproc_lock, SA_XLOCKED);
2980 	LIST_REMOVE(p, p_jaillist);
2981 }
2982 
2983 static void
2984 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
2985 {
2986 
2987 	sx_xlock(&allproc_lock);
2988 	prison_proc_unlink(opr, p);
2989 	prison_proc_link(npr, p);
2990 	sx_xunlock(&allproc_lock);
2991 }
2992 
2993 /*
2994  * Complete a call to either prison_free or prison_proc_free.
2995  */
2996 static void
2997 prison_complete(void *context, int pending)
2998 {
2999 	struct prison *pr = context;
3000 	int drflags;
3001 
3002 	/*
3003 	 * This could be called to release the last reference, or the last
3004 	 * user reference (plus the reference held in prison_proc_free).
3005 	 */
3006 	drflags = prison_lock_xlock(pr, PD_DEREF);
3007 	if (pr->pr_flags & PR_COMPLETE_PROC) {
3008 		pr->pr_flags &= ~PR_COMPLETE_PROC;
3009 		drflags |= PD_DEUREF;
3010 	}
3011 	prison_deref(pr, drflags);
3012 }
3013 
3014 static void
3015 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3016 {
3017 
3018 	kern_psignal(p, SIGKILL);
3019 }
3020 
3021 /*
3022  * Note the iteration does not guarantee acting on all processes.
3023  * Most notably there may be fork or jail_attach in progress.
3024  */
3025 void
3026 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3027     void *cbarg)
3028 {
3029 	struct prison *ppr;
3030 	struct proc *p;
3031 
3032 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3033 		sx_slock(&allproc_lock);
3034 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3035 			if (p->p_state == PRS_NEW)
3036 				continue;
3037 			PROC_LOCK(p);
3038 			cb(p, cbarg);
3039 			PROC_UNLOCK(p);
3040 		}
3041 		sx_sunlock(&allproc_lock);
3042 		if (atomic_load_int(&pr->pr_childcount) == 0)
3043 			return;
3044 		/*
3045 		 * Some jails popped up during the iteration, fall through to a
3046 		 * system-wide search.
3047 		 */
3048 	}
3049 
3050 	sx_slock(&allproc_lock);
3051 	FOREACH_PROC_IN_SYSTEM(p) {
3052 		PROC_LOCK(p);
3053 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3054 			for (ppr = p->p_ucred->cr_prison;
3055 			    ppr != &prison0;
3056 			    ppr = ppr->pr_parent) {
3057 				if (ppr == pr) {
3058 					cb(p, cbarg);
3059 					break;
3060 				}
3061 			}
3062 		}
3063 		PROC_UNLOCK(p);
3064 	}
3065 	sx_sunlock(&allproc_lock);
3066 }
3067 
3068 /*
3069  * Remove a prison reference and/or user reference (usually).
3070  * This assumes context that allows sleeping (for allprison_lock),
3071  * with no non-sleeping locks held, except perhaps the prison itself.
3072  * If there are no more references, release and delist the prison.
3073  * On completion, the prison lock and the allprison lock are both
3074  * unlocked.
3075  */
3076 static void
3077 prison_deref(struct prison *pr, int flags)
3078 {
3079 	struct prisonlist freeprison;
3080 	struct prison *killpr, *rpr, *ppr, *tpr;
3081 
3082 	killpr = NULL;
3083 	TAILQ_INIT(&freeprison);
3084 	/*
3085 	 * Release this prison as requested, which may cause its parent
3086 	 * to be released, and then maybe its grandparent, etc.
3087 	 */
3088 	for (;;) {
3089 		if (flags & PD_KILL) {
3090 			/* Kill the prison and its descendents. */
3091 			KASSERT(pr != &prison0,
3092 			    ("prison_deref trying to kill prison0"));
3093 			if (!(flags & PD_DEREF)) {
3094 				prison_hold(pr);
3095 				flags |= PD_DEREF;
3096 			}
3097 			flags = prison_lock_xlock(pr, flags);
3098 			prison_deref_kill(pr, &freeprison);
3099 		}
3100 		if (flags & PD_DEUREF) {
3101 			/* Drop a user reference. */
3102 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3103 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3104 			     pr->pr_id));
3105 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3106 				if (!(flags & PD_DEREF)) {
3107 					prison_hold(pr);
3108 					flags |= PD_DEREF;
3109 				}
3110 				flags = prison_lock_xlock(pr, flags);
3111 				if (refcount_release(&pr->pr_uref) &&
3112 				    pr->pr_state == PRISON_STATE_ALIVE) {
3113 					/*
3114 					 * When the last user references goes,
3115 					 * this becomes a dying prison.
3116 					 */
3117 					KASSERT(
3118 					    refcount_load(&prison0.pr_uref) > 0,
3119 					    ("prison0 pr_uref=0"));
3120 					pr->pr_state = PRISON_STATE_DYING;
3121 					mtx_unlock(&pr->pr_mtx);
3122 					flags &= ~PD_LOCKED;
3123 					prison_cleanup(pr);
3124 				}
3125 			}
3126 		}
3127 		if (flags & PD_KILL) {
3128 			/*
3129 			 * Any remaining user references are probably processes
3130 			 * that need to be killed, either in this prison or its
3131 			 * descendants.
3132 			 */
3133 			if (refcount_load(&pr->pr_uref) > 0)
3134 				killpr = pr;
3135 			/* Make sure the parent prison doesn't get killed. */
3136 			flags &= ~PD_KILL;
3137 		}
3138 		if (flags & PD_DEREF) {
3139 			/* Drop a reference. */
3140 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3141 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3142 			     pr->pr_id));
3143 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3144 				flags = prison_lock_xlock(pr, flags);
3145 				if (refcount_release(&pr->pr_ref)) {
3146 					/*
3147 					 * When the last reference goes,
3148 					 * unlink the prison and set it aside.
3149 					 */
3150 					KASSERT(
3151 					    refcount_load(&pr->pr_uref) == 0,
3152 					    ("prison_deref: last ref, "
3153 					     "but still has %d urefs (jid=%d)",
3154 					     pr->pr_uref, pr->pr_id));
3155 					KASSERT(
3156 					    refcount_load(&prison0.pr_ref) != 0,
3157 					    ("prison0 pr_ref=0"));
3158 					pr->pr_state = PRISON_STATE_INVALID;
3159 					TAILQ_REMOVE(&allprison, pr, pr_list);
3160 					LIST_REMOVE(pr, pr_sibling);
3161 					TAILQ_INSERT_TAIL(&freeprison, pr,
3162 					    pr_list);
3163 					for (ppr = pr->pr_parent;
3164 					     ppr != NULL;
3165 					     ppr = ppr->pr_parent)
3166 						ppr->pr_childcount--;
3167 					/*
3168 					 * Removing a prison frees references
3169 					 * from its parent.
3170 					 */
3171 					mtx_unlock(&pr->pr_mtx);
3172 					flags &= ~PD_LOCKED;
3173 					pr = pr->pr_parent;
3174 					flags |= PD_DEREF | PD_DEUREF;
3175 					continue;
3176 				}
3177 			}
3178 		}
3179 		break;
3180 	}
3181 
3182 	/* Release all the prison locks. */
3183 	if (flags & PD_LOCKED)
3184 		mtx_unlock(&pr->pr_mtx);
3185 	if (flags & PD_LIST_SLOCKED)
3186 		sx_sunlock(&allprison_lock);
3187 	else if (flags & PD_LIST_XLOCKED)
3188 		sx_xunlock(&allprison_lock);
3189 
3190 	/* Kill any processes attached to a killed prison. */
3191 	if (killpr != NULL)
3192 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3193 
3194 	/*
3195 	 * Finish removing any unreferenced prisons, which couldn't happen
3196 	 * while allprison_lock was held (to avoid a LOR on vrele).
3197 	 */
3198 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3199 #ifdef VIMAGE
3200 		if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3201 			vnet_destroy(rpr->pr_vnet);
3202 #endif
3203 		if (rpr->pr_root != NULL)
3204 			vrele(rpr->pr_root);
3205 		mtx_destroy(&rpr->pr_mtx);
3206 #ifdef INET
3207 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3208 #endif
3209 #ifdef INET6
3210 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3211 #endif
3212 		if (rpr->pr_cpuset != NULL)
3213 			cpuset_rel(rpr->pr_cpuset);
3214 		osd_jail_exit(rpr);
3215 #ifdef RACCT
3216 		if (racct_enable)
3217 			prison_racct_detach(rpr);
3218 #endif
3219 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3220 		free(rpr, M_PRISON);
3221 	}
3222 }
3223 
3224 /*
3225  * Kill the prison and its descendants.  Mark them as dying, clear the
3226  * persist flag, and call module remove methods.
3227  */
3228 static void
3229 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3230 {
3231 	struct prison *cpr, *ppr, *rpr;
3232 	bool descend;
3233 
3234 	/*
3235 	 * Unlike the descendants, the target prison can be killed
3236 	 * even if it is currently dying.  This is useful for failed
3237 	 * creation in jail_set(2).
3238 	 */
3239 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3240 	    ("Trying to kill dead prison %p (jid=%d).",
3241 	     pr, pr->pr_id));
3242 	refcount_acquire(&pr->pr_uref);
3243 	pr->pr_state = PRISON_STATE_DYING;
3244 	mtx_unlock(&pr->pr_mtx);
3245 
3246 	rpr = NULL;
3247 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3248 		if (descend) {
3249 			if (!prison_isalive(cpr)) {
3250 				descend = false;
3251 				continue;
3252 			}
3253 			prison_hold(cpr);
3254 			prison_proc_hold(cpr);
3255 			mtx_lock(&cpr->pr_mtx);
3256 			cpr->pr_state = PRISON_STATE_DYING;
3257 			cpr->pr_flags |= PR_REMOVE;
3258 			mtx_unlock(&cpr->pr_mtx);
3259 			continue;
3260 		}
3261 		if (!(cpr->pr_flags & PR_REMOVE))
3262 			continue;
3263 		prison_cleanup(cpr);
3264 		mtx_lock(&cpr->pr_mtx);
3265 		cpr->pr_flags &= ~PR_REMOVE;
3266 		if (cpr->pr_flags & PR_PERSIST) {
3267 			cpr->pr_flags &= ~PR_PERSIST;
3268 			prison_proc_free_not_last(cpr);
3269 			prison_free_not_last(cpr);
3270 		}
3271 		(void)refcount_release(&cpr->pr_uref);
3272 		if (refcount_release(&cpr->pr_ref)) {
3273 			/*
3274 			 * When the last reference goes, unlink the prison
3275 			 * and set it aside for prison_deref() to handle.
3276 			 * Delay unlinking the sibling list to keep the loop
3277 			 * safe.
3278 			 */
3279 			if (rpr != NULL)
3280 				LIST_REMOVE(rpr, pr_sibling);
3281 			rpr = cpr;
3282 			rpr->pr_state = PRISON_STATE_INVALID;
3283 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3284 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3285 			/*
3286 			 * Removing a prison frees references from its parent.
3287 			 */
3288 			ppr = rpr->pr_parent;
3289 			prison_proc_free_not_last(ppr);
3290 			prison_free_not_last(ppr);
3291 			for (; ppr != NULL; ppr = ppr->pr_parent)
3292 				ppr->pr_childcount--;
3293 		}
3294 		mtx_unlock(&cpr->pr_mtx);
3295 	}
3296 	if (rpr != NULL)
3297 		LIST_REMOVE(rpr, pr_sibling);
3298 
3299 	prison_cleanup(pr);
3300 	mtx_lock(&pr->pr_mtx);
3301 	if (pr->pr_flags & PR_PERSIST) {
3302 		pr->pr_flags &= ~PR_PERSIST;
3303 		prison_proc_free_not_last(pr);
3304 		prison_free_not_last(pr);
3305 	}
3306 	(void)refcount_release(&pr->pr_uref);
3307 }
3308 
3309 /*
3310  * Given the current locking state in the flags, make sure allprison_lock
3311  * is held exclusive, and the prison is locked.  Return flags indicating
3312  * the new state.
3313  */
3314 static int
3315 prison_lock_xlock(struct prison *pr, int flags)
3316 {
3317 
3318 	if (!(flags & PD_LIST_XLOCKED)) {
3319 		/*
3320 		 * Get allprison_lock, which may be an upgrade,
3321 		 * and may require unlocking the prison.
3322 		 */
3323 		if (flags & PD_LOCKED) {
3324 			mtx_unlock(&pr->pr_mtx);
3325 			flags &= ~PD_LOCKED;
3326 		}
3327 		if (flags & PD_LIST_SLOCKED) {
3328 			if (!sx_try_upgrade(&allprison_lock)) {
3329 				sx_sunlock(&allprison_lock);
3330 				sx_xlock(&allprison_lock);
3331 			}
3332 			flags &= ~PD_LIST_SLOCKED;
3333 		} else
3334 			sx_xlock(&allprison_lock);
3335 		flags |= PD_LIST_XLOCKED;
3336 	}
3337 	if (!(flags & PD_LOCKED)) {
3338 		/* Lock the prison mutex. */
3339 		mtx_lock(&pr->pr_mtx);
3340 		flags |= PD_LOCKED;
3341 	}
3342 	return flags;
3343 }
3344 
3345 /*
3346  * Release a prison's resources when it starts dying (when the last user
3347  * reference is dropped, or when it is killed).
3348  */
3349 static void
3350 prison_cleanup(struct prison *pr)
3351 {
3352 	sx_assert(&allprison_lock, SA_XLOCKED);
3353 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3354 	vfs_exjail_delete(pr);
3355 	shm_remove_prison(pr);
3356 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3357 }
3358 
3359 /*
3360  * Set or clear a permission bit in the pr_allow field, passing restrictions
3361  * (cleared permission) down to child jails.
3362  */
3363 void
3364 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3365 {
3366 	struct prison *pr;
3367 
3368 	pr = cred->cr_prison;
3369 	sx_slock(&allprison_lock);
3370 	mtx_lock(&pr->pr_mtx);
3371 	prison_set_allow_locked(pr, flag, enable);
3372 	mtx_unlock(&pr->pr_mtx);
3373 	sx_sunlock(&allprison_lock);
3374 }
3375 
3376 static void
3377 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3378 {
3379 	struct prison *cpr;
3380 	int descend;
3381 
3382 	if (enable != 0)
3383 		pr->pr_allow |= flag;
3384 	else {
3385 		pr->pr_allow &= ~flag;
3386 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3387 			cpr->pr_allow &= ~flag;
3388 	}
3389 }
3390 
3391 /*
3392  * Check if a jail supports the given address family.
3393  *
3394  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3395  * if not.
3396  */
3397 int
3398 prison_check_af(struct ucred *cred, int af)
3399 {
3400 	struct prison *pr;
3401 	int error;
3402 
3403 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3404 
3405 	pr = cred->cr_prison;
3406 #ifdef VIMAGE
3407 	/* Prisons with their own network stack are not limited. */
3408 	if (prison_owns_vnet(cred))
3409 		return (0);
3410 #endif
3411 
3412 	error = 0;
3413 	switch (af)
3414 	{
3415 #ifdef INET
3416 	case AF_INET:
3417 		if (pr->pr_flags & PR_IP4)
3418 		{
3419 			mtx_lock(&pr->pr_mtx);
3420 			if ((pr->pr_flags & PR_IP4) &&
3421 			    pr->pr_addrs[PR_INET] == NULL)
3422 				error = EAFNOSUPPORT;
3423 			mtx_unlock(&pr->pr_mtx);
3424 		}
3425 		break;
3426 #endif
3427 #ifdef INET6
3428 	case AF_INET6:
3429 		if (pr->pr_flags & PR_IP6)
3430 		{
3431 			mtx_lock(&pr->pr_mtx);
3432 			if ((pr->pr_flags & PR_IP6) &&
3433 			    pr->pr_addrs[PR_INET6] == NULL)
3434 				error = EAFNOSUPPORT;
3435 			mtx_unlock(&pr->pr_mtx);
3436 		}
3437 		break;
3438 #endif
3439 	case AF_LOCAL:
3440 	case AF_ROUTE:
3441 	case AF_NETLINK:
3442 		break;
3443 	default:
3444 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3445 			error = EAFNOSUPPORT;
3446 	}
3447 	return (error);
3448 }
3449 
3450 /*
3451  * Check if given address belongs to the jail referenced by cred (wrapper to
3452  * prison_check_ip[46]).
3453  *
3454  * Returns 0 if jail doesn't restrict the address family or if address belongs
3455  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3456  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3457  */
3458 int
3459 prison_if(struct ucred *cred, const struct sockaddr *sa)
3460 {
3461 #ifdef INET
3462 	const struct sockaddr_in *sai;
3463 #endif
3464 #ifdef INET6
3465 	const struct sockaddr_in6 *sai6;
3466 #endif
3467 	int error;
3468 
3469 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3470 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3471 
3472 #ifdef VIMAGE
3473 	if (prison_owns_vnet(cred))
3474 		return (0);
3475 #endif
3476 
3477 	error = 0;
3478 	switch (sa->sa_family)
3479 	{
3480 #ifdef INET
3481 	case AF_INET:
3482 		sai = (const struct sockaddr_in *)sa;
3483 		error = prison_check_ip4(cred, &sai->sin_addr);
3484 		break;
3485 #endif
3486 #ifdef INET6
3487 	case AF_INET6:
3488 		sai6 = (const struct sockaddr_in6 *)sa;
3489 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3490 		break;
3491 #endif
3492 	default:
3493 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3494 			error = EAFNOSUPPORT;
3495 	}
3496 	return (error);
3497 }
3498 
3499 /*
3500  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3501  */
3502 int
3503 prison_check(struct ucred *cred1, struct ucred *cred2)
3504 {
3505 
3506 	return ((cred1->cr_prison == cred2->cr_prison ||
3507 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3508 }
3509 
3510 /*
3511  * For mountd/nfsd to run within a prison, it must be:
3512  * - A vnet prison.
3513  * - PR_ALLOW_NFSD must be set on it.
3514  * - The root directory (pr_root) of the prison must be
3515  *   a file system mount point, so the mountd can hang
3516  *   export information on it.
3517  * - The prison's enforce_statfs cannot be 0, so that
3518  *   mountd(8) can do exports.
3519  */
3520 bool
3521 prison_check_nfsd(struct ucred *cred)
3522 {
3523 
3524 	if (jailed_without_vnet(cred))
3525 		return (false);
3526 	if (!prison_allow(cred, PR_ALLOW_NFSD))
3527 		return (false);
3528 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3529 		return (false);
3530 	if (cred->cr_prison->pr_enforce_statfs == 0)
3531 		return (false);
3532 	return (true);
3533 }
3534 
3535 /*
3536  * Return true if p2 is a child of p1, otherwise false.
3537  */
3538 bool
3539 prison_ischild(struct prison *pr1, struct prison *pr2)
3540 {
3541 
3542 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3543 		if (pr1 == pr2)
3544 			return (true);
3545 	return (false);
3546 }
3547 
3548 /*
3549  * Return true if the prison is currently alive.  A prison is alive if it
3550  * holds user references and it isn't being removed.
3551  */
3552 bool
3553 prison_isalive(const struct prison *pr)
3554 {
3555 
3556 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3557 		return (false);
3558 	return (true);
3559 }
3560 
3561 /*
3562  * Return true if the prison is currently valid.  A prison is valid if it has
3563  * been fully created, and is not being destroyed.  Note that dying prisons
3564  * are still considered valid.  Invalid prisons won't be found under normal
3565  * circumstances, as they're only put in that state by functions that have
3566  * an exclusive hold on allprison_lock.
3567  */
3568 bool
3569 prison_isvalid(struct prison *pr)
3570 {
3571 
3572 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3573 		return (false);
3574 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3575 		return (false);
3576 	return (true);
3577 }
3578 
3579 /*
3580  * Return true if the passed credential is in a jail and that jail does not
3581  * have its own virtual network stack, otherwise false.
3582  */
3583 bool
3584 jailed_without_vnet(struct ucred *cred)
3585 {
3586 
3587 	if (!jailed(cred))
3588 		return (false);
3589 #ifdef VIMAGE
3590 	if (prison_owns_vnet(cred))
3591 		return (false);
3592 #endif
3593 
3594 	return (true);
3595 }
3596 
3597 /*
3598  * Return the correct hostname (domainname, et al) for the passed credential.
3599  */
3600 void
3601 getcredhostname(struct ucred *cred, char *buf, size_t size)
3602 {
3603 	struct prison *pr;
3604 
3605 	/*
3606 	 * A NULL credential can be used to shortcut to the physical
3607 	 * system's hostname.
3608 	 */
3609 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3610 	mtx_lock(&pr->pr_mtx);
3611 	strlcpy(buf, pr->pr_hostname, size);
3612 	mtx_unlock(&pr->pr_mtx);
3613 }
3614 
3615 void
3616 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3617 {
3618 
3619 	mtx_lock(&cred->cr_prison->pr_mtx);
3620 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3621 	mtx_unlock(&cred->cr_prison->pr_mtx);
3622 }
3623 
3624 void
3625 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3626 {
3627 
3628 	mtx_lock(&cred->cr_prison->pr_mtx);
3629 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3630 	mtx_unlock(&cred->cr_prison->pr_mtx);
3631 }
3632 
3633 void
3634 getcredhostid(struct ucred *cred, unsigned long *hostid)
3635 {
3636 
3637 	mtx_lock(&cred->cr_prison->pr_mtx);
3638 	*hostid = cred->cr_prison->pr_hostid;
3639 	mtx_unlock(&cred->cr_prison->pr_mtx);
3640 }
3641 
3642 void
3643 getjailname(struct ucred *cred, char *name, size_t len)
3644 {
3645 
3646 	mtx_lock(&cred->cr_prison->pr_mtx);
3647 	strlcpy(name, cred->cr_prison->pr_name, len);
3648 	mtx_unlock(&cred->cr_prison->pr_mtx);
3649 }
3650 
3651 #ifdef VIMAGE
3652 /*
3653  * Determine whether the prison represented by cred owns
3654  * its vnet rather than having it inherited.
3655  *
3656  * Returns true in case the prison owns the vnet, false otherwise.
3657  */
3658 bool
3659 prison_owns_vnet(struct ucred *cred)
3660 {
3661 
3662 	/*
3663 	 * vnets cannot be added/removed after jail creation,
3664 	 * so no need to lock here.
3665 	 */
3666 	return ((cred->cr_prison->pr_flags & PR_VNET) != 0);
3667 }
3668 #endif
3669 
3670 /*
3671  * Determine whether the subject represented by cred can "see"
3672  * status of a mount point.
3673  * Returns: 0 for permitted, ENOENT otherwise.
3674  * XXX: This function should be called cr_canseemount() and should be
3675  *      placed in kern_prot.c.
3676  */
3677 int
3678 prison_canseemount(struct ucred *cred, struct mount *mp)
3679 {
3680 	struct prison *pr;
3681 	struct statfs *sp;
3682 	size_t len;
3683 
3684 	pr = cred->cr_prison;
3685 	if (pr->pr_enforce_statfs == 0)
3686 		return (0);
3687 	if (pr->pr_root->v_mount == mp)
3688 		return (0);
3689 	if (pr->pr_enforce_statfs == 2)
3690 		return (ENOENT);
3691 	/*
3692 	 * If jail's chroot directory is set to "/" we should be able to see
3693 	 * all mount-points from inside a jail.
3694 	 * This is ugly check, but this is the only situation when jail's
3695 	 * directory ends with '/'.
3696 	 */
3697 	if (strcmp(pr->pr_path, "/") == 0)
3698 		return (0);
3699 	len = strlen(pr->pr_path);
3700 	sp = &mp->mnt_stat;
3701 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3702 		return (ENOENT);
3703 	/*
3704 	 * Be sure that we don't have situation where jail's root directory
3705 	 * is "/some/path" and mount point is "/some/pathpath".
3706 	 */
3707 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3708 		return (ENOENT);
3709 	return (0);
3710 }
3711 
3712 void
3713 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3714 {
3715 	char jpath[MAXPATHLEN];
3716 	struct prison *pr;
3717 	size_t len;
3718 
3719 	pr = cred->cr_prison;
3720 	if (pr->pr_enforce_statfs == 0)
3721 		return;
3722 	if (prison_canseemount(cred, mp) != 0) {
3723 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3724 		strlcpy(sp->f_mntonname, "[restricted]",
3725 		    sizeof(sp->f_mntonname));
3726 		return;
3727 	}
3728 	if (pr->pr_root->v_mount == mp) {
3729 		/*
3730 		 * Clear current buffer data, so we are sure nothing from
3731 		 * the valid path left there.
3732 		 */
3733 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3734 		*sp->f_mntonname = '/';
3735 		return;
3736 	}
3737 	/*
3738 	 * If jail's chroot directory is set to "/" we should be able to see
3739 	 * all mount-points from inside a jail.
3740 	 */
3741 	if (strcmp(pr->pr_path, "/") == 0)
3742 		return;
3743 	len = strlen(pr->pr_path);
3744 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3745 	/*
3746 	 * Clear current buffer data, so we are sure nothing from
3747 	 * the valid path left there.
3748 	 */
3749 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3750 	if (*jpath == '\0') {
3751 		/* Should never happen. */
3752 		*sp->f_mntonname = '/';
3753 	} else {
3754 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3755 	}
3756 }
3757 
3758 /*
3759  * Check with permission for a specific privilege is granted within jail.  We
3760  * have a specific list of accepted privileges; the rest are denied.
3761  */
3762 int
3763 prison_priv_check(struct ucred *cred, int priv)
3764 {
3765 	struct prison *pr;
3766 	int error;
3767 
3768 	/*
3769 	 * Some policies have custom handlers. This routine should not be
3770 	 * called for them. See priv_check_cred().
3771 	 */
3772 	switch (priv) {
3773 	case PRIV_VFS_LOOKUP:
3774 	case PRIV_VFS_GENERATION:
3775 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3776 		    "called for %d\n", priv));
3777 	}
3778 
3779 	if (!jailed(cred))
3780 		return (0);
3781 
3782 #ifdef VIMAGE
3783 	/*
3784 	 * Privileges specific to prisons with a virtual network stack.
3785 	 * There might be a duplicate entry here in case the privilege
3786 	 * is only granted conditionally in the legacy jail case.
3787 	 */
3788 	switch (priv) {
3789 		/*
3790 		 * NFS-specific privileges.
3791 		 */
3792 	case PRIV_NFS_DAEMON:
3793 	case PRIV_VFS_GETFH:
3794 	case PRIV_VFS_MOUNT_EXPORTED:
3795 		if (!prison_check_nfsd(cred))
3796 			return (EPERM);
3797 #ifdef notyet
3798 	case PRIV_NFS_LOCKD:
3799 #endif
3800 		/*
3801 		 * Network stack privileges.
3802 		 */
3803 	case PRIV_NET_BRIDGE:
3804 	case PRIV_NET_GRE:
3805 	case PRIV_NET_BPF:
3806 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3807 	case PRIV_NET_ROUTE:
3808 	case PRIV_NET_TAP:
3809 	case PRIV_NET_SETIFMTU:
3810 	case PRIV_NET_SETIFFLAGS:
3811 	case PRIV_NET_SETIFCAP:
3812 	case PRIV_NET_SETIFDESCR:
3813 	case PRIV_NET_SETIFNAME	:
3814 	case PRIV_NET_SETIFMETRIC:
3815 	case PRIV_NET_SETIFPHYS:
3816 	case PRIV_NET_SETIFMAC:
3817 	case PRIV_NET_SETLANPCP:
3818 	case PRIV_NET_ADDMULTI:
3819 	case PRIV_NET_DELMULTI:
3820 	case PRIV_NET_HWIOCTL:
3821 	case PRIV_NET_SETLLADDR:
3822 	case PRIV_NET_ADDIFGROUP:
3823 	case PRIV_NET_DELIFGROUP:
3824 	case PRIV_NET_IFCREATE:
3825 	case PRIV_NET_IFDESTROY:
3826 	case PRIV_NET_ADDIFADDR:
3827 	case PRIV_NET_DELIFADDR:
3828 	case PRIV_NET_LAGG:
3829 	case PRIV_NET_GIF:
3830 	case PRIV_NET_SETIFVNET:
3831 	case PRIV_NET_SETIFFIB:
3832 	case PRIV_NET_OVPN:
3833 	case PRIV_NET_ME:
3834 	case PRIV_NET_WG:
3835 
3836 		/*
3837 		 * 802.11-related privileges.
3838 		 */
3839 	case PRIV_NET80211_VAP_GETKEY:
3840 	case PRIV_NET80211_VAP_MANAGE:
3841 
3842 #ifdef notyet
3843 		/*
3844 		 * ATM privileges.
3845 		 */
3846 	case PRIV_NETATM_CFG:
3847 	case PRIV_NETATM_ADD:
3848 	case PRIV_NETATM_DEL:
3849 	case PRIV_NETATM_SET:
3850 
3851 		/*
3852 		 * Bluetooth privileges.
3853 		 */
3854 	case PRIV_NETBLUETOOTH_RAW:
3855 #endif
3856 
3857 		/*
3858 		 * Netgraph and netgraph module privileges.
3859 		 */
3860 	case PRIV_NETGRAPH_CONTROL:
3861 #ifdef notyet
3862 	case PRIV_NETGRAPH_TTY:
3863 #endif
3864 
3865 		/*
3866 		 * IPv4 and IPv6 privileges.
3867 		 */
3868 	case PRIV_NETINET_IPFW:
3869 	case PRIV_NETINET_DIVERT:
3870 	case PRIV_NETINET_PF:
3871 	case PRIV_NETINET_DUMMYNET:
3872 	case PRIV_NETINET_CARP:
3873 	case PRIV_NETINET_MROUTE:
3874 	case PRIV_NETINET_RAW:
3875 	case PRIV_NETINET_ADDRCTRL6:
3876 	case PRIV_NETINET_ND6:
3877 	case PRIV_NETINET_SCOPE6:
3878 	case PRIV_NETINET_ALIFETIME6:
3879 	case PRIV_NETINET_IPSEC:
3880 	case PRIV_NETINET_BINDANY:
3881 
3882 #ifdef notyet
3883 		/*
3884 		 * NCP privileges.
3885 		 */
3886 	case PRIV_NETNCP:
3887 
3888 		/*
3889 		 * SMB privileges.
3890 		 */
3891 	case PRIV_NETSMB:
3892 #endif
3893 
3894 	/*
3895 	 * No default: or deny here.
3896 	 * In case of no permit fall through to next switch().
3897 	 */
3898 		if (cred->cr_prison->pr_flags & PR_VNET)
3899 			return (0);
3900 	}
3901 #endif /* VIMAGE */
3902 
3903 	switch (priv) {
3904 		/*
3905 		 * Allow ktrace privileges for root in jail.
3906 		 */
3907 	case PRIV_KTRACE:
3908 
3909 #if 0
3910 		/*
3911 		 * Allow jailed processes to configure audit identity and
3912 		 * submit audit records (login, etc).  In the future we may
3913 		 * want to further refine the relationship between audit and
3914 		 * jail.
3915 		 */
3916 	case PRIV_AUDIT_GETAUDIT:
3917 	case PRIV_AUDIT_SETAUDIT:
3918 	case PRIV_AUDIT_SUBMIT:
3919 #endif
3920 
3921 		/*
3922 		 * Allow jailed processes to manipulate process UNIX
3923 		 * credentials in any way they see fit.
3924 		 */
3925 	case PRIV_CRED_SETUID:
3926 	case PRIV_CRED_SETEUID:
3927 	case PRIV_CRED_SETGID:
3928 	case PRIV_CRED_SETEGID:
3929 	case PRIV_CRED_SETGROUPS:
3930 	case PRIV_CRED_SETREUID:
3931 	case PRIV_CRED_SETREGID:
3932 	case PRIV_CRED_SETRESUID:
3933 	case PRIV_CRED_SETRESGID:
3934 
3935 		/*
3936 		 * Jail implements visibility constraints already, so allow
3937 		 * jailed root to override uid/gid-based constraints.
3938 		 */
3939 	case PRIV_SEEOTHERGIDS:
3940 	case PRIV_SEEOTHERUIDS:
3941 
3942 		/*
3943 		 * Jail implements inter-process debugging limits already, so
3944 		 * allow jailed root various debugging privileges.
3945 		 */
3946 	case PRIV_DEBUG_DIFFCRED:
3947 	case PRIV_DEBUG_SUGID:
3948 	case PRIV_DEBUG_UNPRIV:
3949 
3950 		/*
3951 		 * Allow jail to set various resource limits and login
3952 		 * properties, and for now, exceed process resource limits.
3953 		 */
3954 	case PRIV_PROC_LIMIT:
3955 	case PRIV_PROC_SETLOGIN:
3956 	case PRIV_PROC_SETRLIMIT:
3957 
3958 		/*
3959 		 * System V and POSIX IPC privileges are granted in jail.
3960 		 */
3961 	case PRIV_IPC_READ:
3962 	case PRIV_IPC_WRITE:
3963 	case PRIV_IPC_ADMIN:
3964 	case PRIV_IPC_MSGSIZE:
3965 	case PRIV_MQ_ADMIN:
3966 
3967 		/*
3968 		 * Jail operations within a jail work on child jails.
3969 		 */
3970 	case PRIV_JAIL_ATTACH:
3971 	case PRIV_JAIL_SET:
3972 	case PRIV_JAIL_REMOVE:
3973 
3974 		/*
3975 		 * Jail implements its own inter-process limits, so allow
3976 		 * root processes in jail to change scheduling on other
3977 		 * processes in the same jail.  Likewise for signalling.
3978 		 */
3979 	case PRIV_SCHED_DIFFCRED:
3980 	case PRIV_SCHED_CPUSET:
3981 	case PRIV_SIGNAL_DIFFCRED:
3982 	case PRIV_SIGNAL_SUGID:
3983 
3984 		/*
3985 		 * Allow jailed processes to write to sysctls marked as jail
3986 		 * writable.
3987 		 */
3988 	case PRIV_SYSCTL_WRITEJAIL:
3989 
3990 		/*
3991 		 * Allow root in jail to manage a variety of quota
3992 		 * properties.  These should likely be conditional on a
3993 		 * configuration option.
3994 		 */
3995 	case PRIV_VFS_GETQUOTA:
3996 	case PRIV_VFS_SETQUOTA:
3997 
3998 		/*
3999 		 * Since Jail relies on chroot() to implement file system
4000 		 * protections, grant many VFS privileges to root in jail.
4001 		 * Be careful to exclude mount-related and NFS-related
4002 		 * privileges.
4003 		 */
4004 	case PRIV_VFS_READ:
4005 	case PRIV_VFS_WRITE:
4006 	case PRIV_VFS_ADMIN:
4007 	case PRIV_VFS_EXEC:
4008 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
4009 	case PRIV_VFS_CHFLAGS_DEV:
4010 	case PRIV_VFS_CHOWN:
4011 	case PRIV_VFS_CHROOT:
4012 	case PRIV_VFS_RETAINSUGID:
4013 	case PRIV_VFS_FCHROOT:
4014 	case PRIV_VFS_LINK:
4015 	case PRIV_VFS_SETGID:
4016 	case PRIV_VFS_STAT:
4017 	case PRIV_VFS_STICKYFILE:
4018 
4019 		/*
4020 		 * As in the non-jail case, non-root users are expected to be
4021 		 * able to read kernel/physical memory (provided /dev/[k]mem
4022 		 * exists in the jail and they have permission to access it).
4023 		 */
4024 	case PRIV_KMEM_READ:
4025 		return (0);
4026 
4027 		/*
4028 		 * Depending on the global setting, allow privilege of
4029 		 * setting system flags.
4030 		 */
4031 	case PRIV_VFS_SYSFLAGS:
4032 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4033 			return (0);
4034 		else
4035 			return (EPERM);
4036 
4037 		/*
4038 		 * Depending on the global setting, allow privilege of
4039 		 * mounting/unmounting file systems.
4040 		 */
4041 	case PRIV_VFS_MOUNT:
4042 	case PRIV_VFS_UNMOUNT:
4043 	case PRIV_VFS_MOUNT_NONUSER:
4044 	case PRIV_VFS_MOUNT_OWNER:
4045 		pr = cred->cr_prison;
4046 		prison_lock(pr);
4047 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4048 			error = 0;
4049 		else
4050 			error = EPERM;
4051 		prison_unlock(pr);
4052 		return (error);
4053 
4054 		/*
4055 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4056 		 * policy.  priv_check_cred will not specifically allow it, and
4057 		 * we may want a MAC policy to allow it.
4058 		 */
4059 	case PRIV_VFS_READ_DIR:
4060 		return (0);
4061 
4062 		/*
4063 		 * Conditionnaly allow locking (unlocking) physical pages
4064 		 * in memory.
4065 		 */
4066 	case PRIV_VM_MLOCK:
4067 	case PRIV_VM_MUNLOCK:
4068 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4069 			return (0);
4070 		else
4071 			return (EPERM);
4072 
4073 		/*
4074 		 * Conditionally allow jailed root to bind reserved ports.
4075 		 */
4076 	case PRIV_NETINET_RESERVEDPORT:
4077 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4078 			return (0);
4079 		else
4080 			return (EPERM);
4081 
4082 		/*
4083 		 * Allow jailed root to reuse in-use ports.
4084 		 */
4085 	case PRIV_NETINET_REUSEPORT:
4086 		return (0);
4087 
4088 		/*
4089 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4090 		 */
4091 	case PRIV_NETINET_SETHDROPTS:
4092 		return (0);
4093 
4094 		/*
4095 		 * Conditionally allow creating raw sockets in jail.
4096 		 */
4097 	case PRIV_NETINET_RAW:
4098 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4099 			return (0);
4100 		else
4101 			return (EPERM);
4102 
4103 		/*
4104 		 * Since jail implements its own visibility limits on netstat
4105 		 * sysctls, allow getcred.  This allows identd to work in
4106 		 * jail.
4107 		 */
4108 	case PRIV_NETINET_GETCRED:
4109 		return (0);
4110 
4111 		/*
4112 		 * Allow jailed root to set loginclass.
4113 		 */
4114 	case PRIV_PROC_SETLOGINCLASS:
4115 		return (0);
4116 
4117 		/*
4118 		 * Do not allow a process inside a jail to read the kernel
4119 		 * message buffer unless explicitly permitted.
4120 		 */
4121 	case PRIV_MSGBUF:
4122 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4123 			return (0);
4124 		return (EPERM);
4125 
4126 	default:
4127 		/*
4128 		 * In all remaining cases, deny the privilege request.  This
4129 		 * includes almost all network privileges, many system
4130 		 * configuration privileges.
4131 		 */
4132 		return (EPERM);
4133 	}
4134 }
4135 
4136 /*
4137  * Return the part of pr2's name that is relative to pr1, or the whole name
4138  * if it does not directly follow.
4139  */
4140 
4141 char *
4142 prison_name(struct prison *pr1, struct prison *pr2)
4143 {
4144 	char *name;
4145 
4146 	/* Jails see themselves as "0" (if they see themselves at all). */
4147 	if (pr1 == pr2)
4148 		return "0";
4149 	name = pr2->pr_name;
4150 	if (prison_ischild(pr1, pr2)) {
4151 		/*
4152 		 * pr1 isn't locked (and allprison_lock may not be either)
4153 		 * so its length can't be counted on.  But the number of dots
4154 		 * can be counted on - and counted.
4155 		 */
4156 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4157 			name = strchr(name, '.') + 1;
4158 	}
4159 	return (name);
4160 }
4161 
4162 /*
4163  * Return the part of pr2's path that is relative to pr1, or the whole path
4164  * if it does not directly follow.
4165  */
4166 static char *
4167 prison_path(struct prison *pr1, struct prison *pr2)
4168 {
4169 	char *path1, *path2;
4170 	int len1;
4171 
4172 	path1 = pr1->pr_path;
4173 	path2 = pr2->pr_path;
4174 	if (!strcmp(path1, "/"))
4175 		return (path2);
4176 	len1 = strlen(path1);
4177 	if (strncmp(path1, path2, len1))
4178 		return (path2);
4179 	if (path2[len1] == '\0')
4180 		return "/";
4181 	if (path2[len1] == '/')
4182 		return (path2 + len1);
4183 	return (path2);
4184 }
4185 
4186 /*
4187  * Jail-related sysctls.
4188  */
4189 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4190     "Jails");
4191 
4192 #if defined(INET) || defined(INET6)
4193 /*
4194  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4195  * sysctl_jail_list() helper.
4196  */
4197 static void
4198 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4199 {
4200 	const struct prison_ip *pip;
4201 	const size_t size = pr_families[af].size;
4202 
4203  again:
4204 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4205 	if ((pip = pr->pr_addrs[af]) != NULL) {
4206 		if (*len < pip->ips) {
4207 			*len = pip->ips;
4208 			mtx_unlock(&pr->pr_mtx);
4209 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4210 			mtx_lock(&pr->pr_mtx);
4211 			goto again;
4212 		}
4213 		bcopy(pip->pr_ip, *out, pip->ips * size);
4214 	}
4215 }
4216 #endif
4217 
4218 static int
4219 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4220 {
4221 	struct xprison *xp;
4222 	struct prison *pr, *cpr;
4223 #ifdef INET
4224 	struct in_addr *ip4 = NULL;
4225 	int ip4s = 0;
4226 #endif
4227 #ifdef INET6
4228 	struct in6_addr *ip6 = NULL;
4229 	int ip6s = 0;
4230 #endif
4231 	int descend, error;
4232 
4233 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4234 	pr = req->td->td_ucred->cr_prison;
4235 	error = 0;
4236 	sx_slock(&allprison_lock);
4237 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4238 		mtx_lock(&cpr->pr_mtx);
4239 #ifdef INET
4240 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4241 #endif
4242 #ifdef INET6
4243 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4244 #endif
4245 		bzero(xp, sizeof(*xp));
4246 		xp->pr_version = XPRISON_VERSION;
4247 		xp->pr_id = cpr->pr_id;
4248 		xp->pr_state = cpr->pr_state;
4249 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4250 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4251 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4252 #ifdef INET
4253 		xp->pr_ip4s = ip4s;
4254 #endif
4255 #ifdef INET6
4256 		xp->pr_ip6s = ip6s;
4257 #endif
4258 		mtx_unlock(&cpr->pr_mtx);
4259 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4260 		if (error)
4261 			break;
4262 #ifdef INET
4263 		if (xp->pr_ip4s > 0) {
4264 			error = SYSCTL_OUT(req, ip4,
4265 			    xp->pr_ip4s * sizeof(struct in_addr));
4266 			if (error)
4267 				break;
4268 		}
4269 #endif
4270 #ifdef INET6
4271 		if (xp->pr_ip6s > 0) {
4272 			error = SYSCTL_OUT(req, ip6,
4273 			    xp->pr_ip6s * sizeof(struct in6_addr));
4274 			if (error)
4275 				break;
4276 		}
4277 #endif
4278 	}
4279 	sx_sunlock(&allprison_lock);
4280 	free(xp, M_TEMP);
4281 #ifdef INET
4282 	free(ip4, M_TEMP);
4283 #endif
4284 #ifdef INET6
4285 	free(ip6, M_TEMP);
4286 #endif
4287 	return (error);
4288 }
4289 
4290 SYSCTL_OID(_security_jail, OID_AUTO, list,
4291     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4292     sysctl_jail_list, "S", "List of active jails");
4293 
4294 static int
4295 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4296 {
4297 	int error, injail;
4298 
4299 	injail = jailed(req->td->td_ucred);
4300 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4301 
4302 	return (error);
4303 }
4304 
4305 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4306     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4307     sysctl_jail_jailed, "I", "Process in jail?");
4308 
4309 static int
4310 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4311 {
4312 	int error, havevnet;
4313 #ifdef VIMAGE
4314 	struct ucred *cred = req->td->td_ucred;
4315 
4316 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4317 #else
4318 	havevnet = 0;
4319 #endif
4320 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4321 
4322 	return (error);
4323 }
4324 
4325 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4326     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4327     sysctl_jail_vnet, "I", "Jail owns vnet?");
4328 
4329 #if defined(INET) || defined(INET6)
4330 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4331     &jail_max_af_ips, 0,
4332     "Number of IP addresses a jail may have at most per address family (deprecated)");
4333 #endif
4334 
4335 /*
4336  * Default parameters for jail(2) compatibility.  For historical reasons,
4337  * the sysctl names have varying similarity to the parameter names.  Prisons
4338  * just see their own parameters, and can't change them.
4339  */
4340 static int
4341 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4342 {
4343 	int error, i;
4344 
4345 	/* Get the current flag value, and convert it to a boolean. */
4346 	if (req->td->td_ucred->cr_prison == &prison0) {
4347 		mtx_lock(&prison0.pr_mtx);
4348 		i = (jail_default_allow & arg2) != 0;
4349 		mtx_unlock(&prison0.pr_mtx);
4350 	} else
4351 		i = prison_allow(req->td->td_ucred, arg2);
4352 
4353 	if (arg1 != NULL)
4354 		i = !i;
4355 	error = sysctl_handle_int(oidp, &i, 0, req);
4356 	if (error || !req->newptr)
4357 		return (error);
4358 	i = i ? arg2 : 0;
4359 	if (arg1 != NULL)
4360 		i ^= arg2;
4361 	/*
4362 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4363 	 * for writing.
4364 	 */
4365 	mtx_lock(&prison0.pr_mtx);
4366 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4367 	mtx_unlock(&prison0.pr_mtx);
4368 	return (0);
4369 }
4370 
4371 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4372     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4373     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4374     "Processes in jail can set their hostnames (deprecated)");
4375 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4376     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4377     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4378     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4379 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4380     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4381     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4382     "Processes in jail can use System V IPC primitives (deprecated)");
4383 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4384     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4385     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4386     "Prison root can create raw sockets (deprecated)");
4387 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4388     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4389     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4390     "Processes in jail can alter system file flags (deprecated)");
4391 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4392     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4393     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4394     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4395 
4396 static int
4397 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4398 {
4399 	struct prison *pr;
4400 	int level, error;
4401 
4402 	pr = req->td->td_ucred->cr_prison;
4403 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4404 	error = sysctl_handle_int(oidp, &level, 0, req);
4405 	if (error || !req->newptr)
4406 		return (error);
4407 	*(int *)arg1 = level;
4408 	return (0);
4409 }
4410 
4411 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4412     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4413     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4414     sysctl_jail_default_level, "I",
4415     "Processes in jail cannot see all mounted file systems (deprecated)");
4416 
4417 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4418     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4419     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4420     sysctl_jail_default_level, "I",
4421     "Ruleset for the devfs filesystem in jail (deprecated)");
4422 
4423 /*
4424  * Nodes to describe jail parameters.  Maximum length of string parameters
4425  * is returned in the string itself, and the other parameters exist merely
4426  * to make themselves and their types known.
4427  */
4428 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4429     "Jail parameters");
4430 
4431 int
4432 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4433 {
4434 	int i;
4435 	long l;
4436 	size_t s;
4437 	char numbuf[12];
4438 
4439 	switch (oidp->oid_kind & CTLTYPE)
4440 	{
4441 	case CTLTYPE_LONG:
4442 	case CTLTYPE_ULONG:
4443 		l = 0;
4444 #ifdef SCTL_MASK32
4445 		if (!(req->flags & SCTL_MASK32))
4446 #endif
4447 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4448 	case CTLTYPE_INT:
4449 	case CTLTYPE_UINT:
4450 		i = 0;
4451 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4452 	case CTLTYPE_STRING:
4453 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4454 		return
4455 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4456 	case CTLTYPE_STRUCT:
4457 		s = (size_t)arg2;
4458 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4459 	}
4460 	return (0);
4461 }
4462 
4463 /*
4464  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4465  * jail creation time but cannot be changed in an existing jail.
4466  */
4467 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4468 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4469 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4470 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4471 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4472     "I", "Jail secure level");
4473 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4474     "Jail value for kern.osreldate and uname -K");
4475 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4476     "Jail value for kern.osrelease and uname -r");
4477 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4478     "I", "Jail cannot see all mounted file systems");
4479 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4480     "I", "Ruleset for in-jail devfs mounts");
4481 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4482     "B", "Jail persistence");
4483 #ifdef VIMAGE
4484 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4485     "E,jailsys", "Virtual network stack");
4486 #endif
4487 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4488     "B", "Jail is in the process of shutting down");
4489 
4490 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4491 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4492     "I", "Current number of child jails");
4493 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4494     "I", "Maximum number of child jails");
4495 
4496 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4497 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4498     "Jail hostname");
4499 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4500     "Jail NIS domainname");
4501 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4502     "Jail host UUID");
4503 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4504     "LU", "Jail host ID");
4505 
4506 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4507 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4508 
4509 #ifdef INET
4510 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4511     "Jail IPv4 address virtualization");
4512 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4513     "S,in_addr,a", "Jail IPv4 addresses");
4514 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4515     "B", "Do (not) use IPv4 source address selection rather than the "
4516     "primary jail IPv4 address.");
4517 #endif
4518 #ifdef INET6
4519 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4520     "Jail IPv6 address virtualization");
4521 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4522     "S,in6_addr,a", "Jail IPv6 addresses");
4523 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4524     "B", "Do (not) use IPv6 source address selection rather than the "
4525     "primary jail IPv6 address.");
4526 #endif
4527 
4528 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4529 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4530     "B", "Jail may set hostname");
4531 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4532     "B", "Jail may use SYSV IPC");
4533 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4534     "B", "Jail may create raw sockets");
4535 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4536     "B", "Jail may alter system file flags");
4537 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4538     "B", "Jail may set file quotas");
4539 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4540     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4541 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4542     "B", "Jail may lock (unlock) physical pages in memory");
4543 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4544     "B", "Jail may bind sockets to reserved ports");
4545 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4546     "B", "Jail may read the kernel message buffer");
4547 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4548     "B", "Unprivileged processes may use process debugging facilities");
4549 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4550     "B", "Processes in jail with uid 0 have privilege");
4551 #ifdef VIMAGE
4552 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4553     "B", "Mountd/nfsd may run in the jail");
4554 #endif
4555 
4556 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4557 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4558     "B", "Jail may mount/unmount jail-friendly file systems in general");
4559 
4560 /*
4561  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4562  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4563  * not created.
4564  */
4565 unsigned
4566 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4567     const char *descr)
4568 {
4569 	struct bool_flags *bf;
4570 	struct sysctl_oid *parent;
4571 	char *allow_name, *allow_noname, *allowed;
4572 #ifndef NO_SYSCTL_DESCR
4573 	char *descr_deprecated;
4574 #endif
4575 	u_int allow_flag;
4576 
4577 	if (prefix
4578 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4579 		< 0 ||
4580 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4581 		< 0
4582 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4583 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4584 		free(allow_name, M_PRISON);
4585 		return 0;
4586 	}
4587 
4588 	/*
4589 	 * See if this parameter has already beed added, i.e. a module was
4590 	 * previously loaded/unloaded.
4591 	 */
4592 	mtx_lock(&prison0.pr_mtx);
4593 	for (bf = pr_flag_allow;
4594 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4595 		atomic_load_int(&bf->flag) != 0;
4596 	     bf++) {
4597 		if (strcmp(bf->name, allow_name) == 0) {
4598 			allow_flag = bf->flag;
4599 			goto no_add;
4600 		}
4601 	}
4602 
4603 	/*
4604 	 * Find a free bit in pr_allow_all, failing if there are none
4605 	 * (which shouldn't happen as long as we keep track of how many
4606 	 * potential dynamic flags exist).
4607 	 */
4608 	for (allow_flag = 1;; allow_flag <<= 1) {
4609 		if (allow_flag == 0)
4610 			goto no_add;
4611 		if ((pr_allow_all & allow_flag) == 0)
4612 			break;
4613 	}
4614 
4615 	/* Note the parameter in the next open slot in pr_flag_allow. */
4616 	for (bf = pr_flag_allow; ; bf++) {
4617 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4618 			/* This should never happen, but is not fatal. */
4619 			allow_flag = 0;
4620 			goto no_add;
4621 		}
4622 		if (atomic_load_int(&bf->flag) == 0)
4623 			break;
4624 	}
4625 	bf->name = allow_name;
4626 	bf->noname = allow_noname;
4627 	pr_allow_all |= allow_flag;
4628 	/*
4629 	 * prison0 always has permission for the new parameter.
4630 	 * Other jails must have it granted to them.
4631 	 */
4632 	prison0.pr_allow |= allow_flag;
4633 	/* The flag indicates a valid entry, so make sure it is set last. */
4634 	atomic_store_rel_int(&bf->flag, allow_flag);
4635 	mtx_unlock(&prison0.pr_mtx);
4636 
4637 	/*
4638 	 * Create sysctls for the parameter, and the back-compat global
4639 	 * permission.
4640 	 */
4641 	parent = prefix
4642 	    ? SYSCTL_ADD_NODE(NULL,
4643 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4644 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4645 	    : &sysctl___security_jail_param_allow;
4646 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4647 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4648 	    NULL, 0, sysctl_jail_param, "B", descr);
4649 	if ((prefix
4650 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4651 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4652 #ifndef NO_SYSCTL_DESCR
4653 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4654 		    descr);
4655 #endif
4656 		(void)SYSCTL_ADD_PROC(NULL,
4657 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4658 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4659 		    sysctl_jail_default_allow, "I", descr_deprecated);
4660 #ifndef NO_SYSCTL_DESCR
4661 		free(descr_deprecated, M_TEMP);
4662 #endif
4663 		free(allowed, M_TEMP);
4664 	}
4665 	return allow_flag;
4666 
4667  no_add:
4668 	mtx_unlock(&prison0.pr_mtx);
4669 	free(allow_name, M_PRISON);
4670 	free(allow_noname, M_PRISON);
4671 	return allow_flag;
4672 }
4673 
4674 /*
4675  * The VFS system will register jail-aware filesystems here.  They each get
4676  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4677  * attempts to mount.
4678  */
4679 void
4680 prison_add_vfs(struct vfsconf *vfsp)
4681 {
4682 #ifdef NO_SYSCTL_DESCR
4683 
4684 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4685 	    NULL, NULL);
4686 #else
4687 	char *descr;
4688 
4689 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4690 	    vfsp->vfc_name);
4691 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4692 	    NULL, descr);
4693 	free(descr, M_TEMP);
4694 #endif
4695 }
4696 
4697 #ifdef RACCT
4698 void
4699 prison_racct_foreach(void (*callback)(struct racct *racct,
4700     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4701     void *arg2, void *arg3)
4702 {
4703 	struct prison_racct *prr;
4704 
4705 	ASSERT_RACCT_ENABLED();
4706 
4707 	sx_slock(&allprison_lock);
4708 	if (pre != NULL)
4709 		(pre)();
4710 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4711 		(callback)(prr->prr_racct, arg2, arg3);
4712 	if (post != NULL)
4713 		(post)();
4714 	sx_sunlock(&allprison_lock);
4715 }
4716 
4717 static struct prison_racct *
4718 prison_racct_find_locked(const char *name)
4719 {
4720 	struct prison_racct *prr;
4721 
4722 	ASSERT_RACCT_ENABLED();
4723 	sx_assert(&allprison_lock, SA_XLOCKED);
4724 
4725 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4726 		return (NULL);
4727 
4728 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4729 		if (strcmp(name, prr->prr_name) != 0)
4730 			continue;
4731 
4732 		/* Found prison_racct with a matching name? */
4733 		prison_racct_hold(prr);
4734 		return (prr);
4735 	}
4736 
4737 	/* Add new prison_racct. */
4738 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4739 	racct_create(&prr->prr_racct);
4740 
4741 	strcpy(prr->prr_name, name);
4742 	refcount_init(&prr->prr_refcount, 1);
4743 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4744 
4745 	return (prr);
4746 }
4747 
4748 struct prison_racct *
4749 prison_racct_find(const char *name)
4750 {
4751 	struct prison_racct *prr;
4752 
4753 	ASSERT_RACCT_ENABLED();
4754 
4755 	sx_xlock(&allprison_lock);
4756 	prr = prison_racct_find_locked(name);
4757 	sx_xunlock(&allprison_lock);
4758 	return (prr);
4759 }
4760 
4761 void
4762 prison_racct_hold(struct prison_racct *prr)
4763 {
4764 
4765 	ASSERT_RACCT_ENABLED();
4766 
4767 	refcount_acquire(&prr->prr_refcount);
4768 }
4769 
4770 static void
4771 prison_racct_free_locked(struct prison_racct *prr)
4772 {
4773 
4774 	ASSERT_RACCT_ENABLED();
4775 	sx_assert(&allprison_lock, SA_XLOCKED);
4776 
4777 	if (refcount_release(&prr->prr_refcount)) {
4778 		racct_destroy(&prr->prr_racct);
4779 		LIST_REMOVE(prr, prr_next);
4780 		free(prr, M_PRISON_RACCT);
4781 	}
4782 }
4783 
4784 void
4785 prison_racct_free(struct prison_racct *prr)
4786 {
4787 
4788 	ASSERT_RACCT_ENABLED();
4789 	sx_assert(&allprison_lock, SA_UNLOCKED);
4790 
4791 	if (refcount_release_if_not_last(&prr->prr_refcount))
4792 		return;
4793 
4794 	sx_xlock(&allprison_lock);
4795 	prison_racct_free_locked(prr);
4796 	sx_xunlock(&allprison_lock);
4797 }
4798 
4799 static void
4800 prison_racct_attach(struct prison *pr)
4801 {
4802 	struct prison_racct *prr;
4803 
4804 	ASSERT_RACCT_ENABLED();
4805 	sx_assert(&allprison_lock, SA_XLOCKED);
4806 
4807 	prr = prison_racct_find_locked(pr->pr_name);
4808 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4809 
4810 	pr->pr_prison_racct = prr;
4811 }
4812 
4813 /*
4814  * Handle jail renaming.  From the racct point of view, renaming means
4815  * moving from one prison_racct to another.
4816  */
4817 static void
4818 prison_racct_modify(struct prison *pr)
4819 {
4820 #ifdef RCTL
4821 	struct proc *p;
4822 	struct ucred *cred;
4823 #endif
4824 	struct prison_racct *oldprr;
4825 
4826 	ASSERT_RACCT_ENABLED();
4827 
4828 	sx_slock(&allproc_lock);
4829 	sx_xlock(&allprison_lock);
4830 
4831 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4832 		sx_xunlock(&allprison_lock);
4833 		sx_sunlock(&allproc_lock);
4834 		return;
4835 	}
4836 
4837 	oldprr = pr->pr_prison_racct;
4838 	pr->pr_prison_racct = NULL;
4839 
4840 	prison_racct_attach(pr);
4841 
4842 	/*
4843 	 * Move resource utilisation records.
4844 	 */
4845 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4846 
4847 #ifdef RCTL
4848 	/*
4849 	 * Force rctl to reattach rules to processes.
4850 	 */
4851 	FOREACH_PROC_IN_SYSTEM(p) {
4852 		PROC_LOCK(p);
4853 		cred = crhold(p->p_ucred);
4854 		PROC_UNLOCK(p);
4855 		rctl_proc_ucred_changed(p, cred);
4856 		crfree(cred);
4857 	}
4858 #endif
4859 
4860 	sx_sunlock(&allproc_lock);
4861 	prison_racct_free_locked(oldprr);
4862 	sx_xunlock(&allprison_lock);
4863 }
4864 
4865 static void
4866 prison_racct_detach(struct prison *pr)
4867 {
4868 
4869 	ASSERT_RACCT_ENABLED();
4870 	sx_assert(&allprison_lock, SA_UNLOCKED);
4871 
4872 	if (pr->pr_prison_racct == NULL)
4873 		return;
4874 	prison_racct_free(pr->pr_prison_racct);
4875 	pr->pr_prison_racct = NULL;
4876 }
4877 #endif /* RACCT */
4878 
4879 #ifdef DDB
4880 
4881 static void
4882 db_show_prison(struct prison *pr)
4883 {
4884 	struct bool_flags *bf;
4885 	struct jailsys_flags *jsf;
4886 #if defined(INET) || defined(INET6)
4887 	int ii;
4888 	struct prison_ip *pip;
4889 #endif
4890 	unsigned f;
4891 #ifdef INET
4892 	char ip4buf[INET_ADDRSTRLEN];
4893 #endif
4894 #ifdef INET6
4895 	char ip6buf[INET6_ADDRSTRLEN];
4896 #endif
4897 
4898 	db_printf("prison %p:\n", pr);
4899 	db_printf(" jid             = %d\n", pr->pr_id);
4900 	db_printf(" name            = %s\n", pr->pr_name);
4901 	db_printf(" parent          = %p\n", pr->pr_parent);
4902 	db_printf(" ref             = %d\n", pr->pr_ref);
4903 	db_printf(" uref            = %d\n", pr->pr_uref);
4904 	db_printf(" state           = %s\n",
4905 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4906 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
4907 	    "invalid");
4908 	db_printf(" path            = %s\n", pr->pr_path);
4909 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4910 	    ? pr->pr_cpuset->cs_id : -1);
4911 #ifdef VIMAGE
4912 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4913 #endif
4914 	db_printf(" root            = %p\n", pr->pr_root);
4915 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4916 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4917 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4918 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4919 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4920 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4921 	db_printf(" flags           = 0x%x", pr->pr_flags);
4922 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4923 		if (pr->pr_flags & bf->flag)
4924 			db_printf(" %s", bf->name);
4925 	for (jsf = pr_flag_jailsys;
4926 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4927 	     jsf++) {
4928 		f = pr->pr_flags & (jsf->disable | jsf->new);
4929 		db_printf(" %-16s= %s\n", jsf->name,
4930 		    (f != 0 && f == jsf->disable) ? "disable"
4931 		    : (f == jsf->new) ? "new"
4932 		    : "inherit");
4933 	}
4934 	db_printf(" allow           = 0x%x", pr->pr_allow);
4935 	for (bf = pr_flag_allow;
4936 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4937 		atomic_load_int(&bf->flag) != 0;
4938 	     bf++)
4939 		if (pr->pr_allow & bf->flag)
4940 			db_printf(" %s", bf->name);
4941 	db_printf("\n");
4942 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4943 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4944 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4945 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4946 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4947 #ifdef INET
4948 	if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
4949 		db_printf(" ip4s            = %d\n", pip->ips);
4950 		for (ii = 0; ii < pip->ips; ii++)
4951 			db_printf(" %s %s\n",
4952 			    ii == 0 ? "ip4.addr        =" : "                 ",
4953 			    inet_ntoa_r(
4954 			    *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
4955 			    ip4buf));
4956 	}
4957 #endif
4958 #ifdef INET6
4959 	if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
4960 		db_printf(" ip6s            = %d\n", pip->ips);
4961 		for (ii = 0; ii < pip->ips; ii++)
4962 			db_printf(" %s %s\n",
4963 			    ii == 0 ? "ip6.addr        =" : "                 ",
4964 			    ip6_sprintf(ip6buf,
4965 			    (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
4966 	}
4967 #endif
4968 }
4969 
4970 DB_SHOW_COMMAND(prison, db_show_prison_command)
4971 {
4972 	struct prison *pr;
4973 
4974 	if (!have_addr) {
4975 		/*
4976 		 * Show all prisons in the list, and prison0 which is not
4977 		 * listed.
4978 		 */
4979 		db_show_prison(&prison0);
4980 		if (!db_pager_quit) {
4981 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4982 				db_show_prison(pr);
4983 				if (db_pager_quit)
4984 					break;
4985 			}
4986 		}
4987 		return;
4988 	}
4989 
4990 	if (addr == 0)
4991 		pr = &prison0;
4992 	else {
4993 		/* Look for a prison with the ID and with references. */
4994 		TAILQ_FOREACH(pr, &allprison, pr_list)
4995 			if (pr->pr_id == addr && pr->pr_ref > 0)
4996 				break;
4997 		if (pr == NULL)
4998 			/* Look again, without requiring a reference. */
4999 			TAILQ_FOREACH(pr, &allprison, pr_list)
5000 				if (pr->pr_id == addr)
5001 					break;
5002 		if (pr == NULL)
5003 			/* Assume address points to a valid prison. */
5004 			pr = (struct prison *)addr;
5005 	}
5006 	db_show_prison(pr);
5007 }
5008 
5009 #endif /* DDB */
5010