xref: /freebsd/sys/kern/kern_jail.c (revision 35c0a8c449fd2b7f75029ebed5e10852240f0865)
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 /*
134  * Handle jail teardown in a dedicated thread to avoid deadlocks from
135  * vnet_destroy().
136  */
137 TASKQUEUE_DEFINE_THREAD(jail_remove);
138 
139 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
140 struct	sx allprison_lock;
141 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
142 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
143 LIST_HEAD(, prison_racct) allprison_racct;
144 int	lastprid = 0;
145 int	lastdeadid = 0;
146 
147 static int get_next_prid(struct prison **insprp);
148 static int get_next_deadid(struct prison **insprp);
149 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
150 static void prison_complete(void *context, int pending);
151 static void prison_deref(struct prison *pr, int flags);
152 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
153 static int prison_lock_xlock(struct prison *pr, int flags);
154 static void prison_cleanup(struct prison *pr);
155 static void prison_free_not_last(struct prison *pr);
156 static void prison_proc_free_not_last(struct prison *pr);
157 static void prison_proc_relink(struct prison *opr, struct prison *npr,
158     struct proc *p);
159 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
160     int enable);
161 static char *prison_path(struct prison *pr1, struct prison *pr2);
162 #ifdef RACCT
163 static void prison_racct_attach(struct prison *pr);
164 static void prison_racct_modify(struct prison *pr);
165 static void prison_racct_detach(struct prison *pr);
166 #endif
167 
168 /* Flags for prison_deref */
169 #define	PD_DEREF	0x01	/* Decrement pr_ref */
170 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
171 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
172 #define	PD_LOCKED	0x10	/* pr_mtx is held */
173 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
174 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
175 #define PD_OP_FLAGS	0x07	/* Operation flags */
176 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
177 
178 /*
179  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
180  * as we cannot figure out the size of a sparse array, or an array without a
181  * terminating entry.
182  */
183 static struct bool_flags pr_flag_bool[] = {
184 	{"persist", "nopersist", PR_PERSIST},
185 #ifdef INET
186 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
187 #endif
188 #ifdef INET6
189 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
190 #endif
191 };
192 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
193 
194 static struct jailsys_flags pr_flag_jailsys[] = {
195 	{"host", 0, PR_HOST},
196 #ifdef VIMAGE
197 	{"vnet", 0, PR_VNET},
198 #endif
199 #ifdef INET
200 	{"ip4", PR_IP4_USER, PR_IP4_USER},
201 #endif
202 #ifdef INET6
203 	{"ip6", PR_IP6_USER, PR_IP6_USER},
204 #endif
205 };
206 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
207 
208 /*
209  * Make this array full-size so dynamic parameters can be added.
210  * It is protected by prison0.mtx, but lockless reading is allowed
211  * with an atomic check of the flag values.
212  */
213 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
214 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
215 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
216 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
217 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
218 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
219 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
220 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
221 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
222 	{"allow.reserved_ports", "allow.noreserved_ports",
223 	 PR_ALLOW_RESERVED_PORTS},
224 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
225 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
226 	 PR_ALLOW_UNPRIV_DEBUG},
227 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
228 #ifdef VIMAGE
229 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
230 #endif
231 	{"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
232 	{"allow.adjtime", "allow.noadjtime", PR_ALLOW_ADJTIME},
233 	{"allow.settime", "allow.nosettime", PR_ALLOW_SETTIME},
234 };
235 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
236 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
237 
238 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
239 					 PR_ALLOW_RESERVED_PORTS | \
240 					 PR_ALLOW_UNPRIV_DEBUG | \
241 					 PR_ALLOW_SUSER)
242 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
243 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
244 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
245 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
246 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
247 #if defined(INET) || defined(INET6)
248 static unsigned jail_max_af_ips = 255;
249 #endif
250 
251 /*
252  * Initialize the parts of prison0 that can't be static-initialized with
253  * constants.  This is called from proc0_init() after creating thread0 cpuset.
254  */
255 void
256 prison0_init(void)
257 {
258 	uint8_t *file, *data;
259 	size_t size;
260 	char buf[sizeof(prison0.pr_hostuuid)];
261 	bool valid;
262 
263 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
264 	prison0.pr_osreldate = osreldate;
265 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
266 
267 	/* If we have a preloaded hostuuid, use it. */
268 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
269 	if (file != NULL) {
270 		data = preload_fetch_addr(file);
271 		size = preload_fetch_size(file);
272 		if (data != NULL) {
273 			/*
274 			 * The preloaded data may include trailing whitespace, almost
275 			 * certainly a newline; skip over any whitespace or
276 			 * non-printable characters to be safe.
277 			 */
278 			while (size > 0 && data[size - 1] <= 0x20) {
279 				size--;
280 			}
281 
282 			valid = false;
283 
284 			/*
285 			 * Not NUL-terminated when passed from loader, but
286 			 * validate_uuid requires that due to using sscanf (as
287 			 * does the subsequent strlcpy, since it still reads
288 			 * past the given size to return the true length);
289 			 * bounce to a temporary buffer to fix.
290 			 */
291 			if (size >= sizeof(buf))
292 				goto done;
293 
294 			memcpy(buf, data, size);
295 			buf[size] = '\0';
296 
297 			if (validate_uuid(buf, size, NULL, 0) != 0)
298 				goto done;
299 
300 			valid = true;
301 			(void)strlcpy(prison0.pr_hostuuid, buf,
302 			    sizeof(prison0.pr_hostuuid));
303 
304 done:
305 			if (bootverbose && !valid) {
306 				printf("hostuuid: preload data malformed: '%.*s'\n",
307 				    (int)size, data);
308 			}
309 		}
310 	}
311 	if (bootverbose)
312 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
313 }
314 
315 /*
316  * struct jail_args {
317  *	struct jail *jail;
318  * };
319  */
320 int
321 sys_jail(struct thread *td, struct jail_args *uap)
322 {
323 	uint32_t version;
324 	int error;
325 	struct jail j;
326 
327 	error = copyin(uap->jail, &version, sizeof(uint32_t));
328 	if (error)
329 		return (error);
330 
331 	switch (version) {
332 	case 0:
333 	{
334 		struct jail_v0 j0;
335 
336 		/* FreeBSD single IPv4 jails. */
337 		bzero(&j, sizeof(struct jail));
338 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
339 		if (error)
340 			return (error);
341 		j.version = j0.version;
342 		j.path = j0.path;
343 		j.hostname = j0.hostname;
344 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
345 		break;
346 	}
347 
348 	case 1:
349 		/*
350 		 * Version 1 was used by multi-IPv4 jail implementations
351 		 * that never made it into the official kernel.
352 		 */
353 		return (EINVAL);
354 
355 	case 2:	/* JAIL_API_VERSION */
356 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
357 		error = copyin(uap->jail, &j, sizeof(struct jail));
358 		if (error)
359 			return (error);
360 		break;
361 
362 	default:
363 		/* Sci-Fi jails are not supported, sorry. */
364 		return (EINVAL);
365 	}
366 	return (kern_jail(td, &j));
367 }
368 
369 int
370 kern_jail(struct thread *td, struct jail *j)
371 {
372 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
373 #ifdef INET
374 			    + 1
375 #endif
376 #ifdef INET6
377 			    + 1
378 #endif
379 			    )];
380 	struct uio opt;
381 	char *u_path, *u_hostname, *u_name;
382 	struct bool_flags *bf;
383 #ifdef INET
384 	uint32_t ip4s;
385 	struct in_addr *u_ip4;
386 #endif
387 #ifdef INET6
388 	struct in6_addr *u_ip6;
389 #endif
390 	size_t tmplen;
391 	int error, enforce_statfs;
392 
393 	bzero(&optiov, sizeof(optiov));
394 	opt.uio_iov = optiov;
395 	opt.uio_iovcnt = 0;
396 	opt.uio_offset = -1;
397 	opt.uio_resid = -1;
398 	opt.uio_segflg = UIO_SYSSPACE;
399 	opt.uio_rw = UIO_READ;
400 	opt.uio_td = td;
401 
402 	/* Set permissions for top-level jails from sysctls. */
403 	if (!jailed(td->td_ucred)) {
404 		for (bf = pr_flag_allow;
405 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
406 			atomic_load_int(&bf->flag) != 0;
407 		     bf++) {
408 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
409 			    (jail_default_allow & bf->flag)
410 			    ? bf->name : bf->noname);
411 			optiov[opt.uio_iovcnt].iov_len =
412 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
413 			opt.uio_iovcnt += 2;
414 		}
415 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
416 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
417 		opt.uio_iovcnt++;
418 		enforce_statfs = jail_default_enforce_statfs;
419 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
420 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
421 		opt.uio_iovcnt++;
422 	}
423 
424 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
425 #ifdef INET
426 	ip4s = (j->version == 0) ? 1 : j->ip4s;
427 	if (ip4s > jail_max_af_ips)
428 		return (EINVAL);
429 	tmplen += ip4s * sizeof(struct in_addr);
430 #else
431 	if (j->ip4s > 0)
432 		return (EINVAL);
433 #endif
434 #ifdef INET6
435 	if (j->ip6s > jail_max_af_ips)
436 		return (EINVAL);
437 	tmplen += j->ip6s * sizeof(struct in6_addr);
438 #else
439 	if (j->ip6s > 0)
440 		return (EINVAL);
441 #endif
442 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
443 	u_hostname = u_path + MAXPATHLEN;
444 	u_name = u_hostname + MAXHOSTNAMELEN;
445 #ifdef INET
446 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
447 #endif
448 #ifdef INET6
449 #ifdef INET
450 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
451 #else
452 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
453 #endif
454 #endif
455 	optiov[opt.uio_iovcnt].iov_base = "path";
456 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
457 	opt.uio_iovcnt++;
458 	optiov[opt.uio_iovcnt].iov_base = u_path;
459 	error = copyinstr(j->path, u_path, MAXPATHLEN,
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 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
467 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
468 	opt.uio_iovcnt++;
469 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
470 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
471 	    &optiov[opt.uio_iovcnt].iov_len);
472 	if (error) {
473 		free(u_path, M_TEMP);
474 		return (error);
475 	}
476 	opt.uio_iovcnt++;
477 	if (j->jailname != NULL) {
478 		optiov[opt.uio_iovcnt].iov_base = "name";
479 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
480 		opt.uio_iovcnt++;
481 		optiov[opt.uio_iovcnt].iov_base = u_name;
482 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
483 		    &optiov[opt.uio_iovcnt].iov_len);
484 		if (error) {
485 			free(u_path, M_TEMP);
486 			return (error);
487 		}
488 		opt.uio_iovcnt++;
489 	}
490 #ifdef INET
491 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
492 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
493 	opt.uio_iovcnt++;
494 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
495 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
496 	if (j->version == 0)
497 		u_ip4->s_addr = j->ip4s;
498 	else {
499 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
500 		if (error) {
501 			free(u_path, M_TEMP);
502 			return (error);
503 		}
504 	}
505 	opt.uio_iovcnt++;
506 #endif
507 #ifdef INET6
508 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
509 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
510 	opt.uio_iovcnt++;
511 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
512 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
513 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
514 	if (error) {
515 		free(u_path, M_TEMP);
516 		return (error);
517 	}
518 	opt.uio_iovcnt++;
519 #endif
520 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
521 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
522 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
523 	free(u_path, M_TEMP);
524 	return (error);
525 }
526 
527 /*
528  * struct jail_set_args {
529  *	struct iovec *iovp;
530  *	unsigned int iovcnt;
531  *	int flags;
532  * };
533  */
534 int
535 sys_jail_set(struct thread *td, struct jail_set_args *uap)
536 {
537 	struct uio *auio;
538 	int error;
539 
540 	/* Check that we have an even number of iovecs. */
541 	if (uap->iovcnt & 1)
542 		return (EINVAL);
543 
544 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
545 	if (error)
546 		return (error);
547 	error = kern_jail_set(td, auio, uap->flags);
548 	freeuio(auio);
549 	return (error);
550 }
551 
552 #if defined(INET) || defined(INET6)
553 typedef int prison_addr_cmp_t(const void *, const void *);
554 typedef bool prison_addr_valid_t(const void *);
555 static const struct pr_family {
556 	size_t			size;
557 	prison_addr_cmp_t	*cmp;
558 	prison_addr_valid_t	*valid;
559 	int			ip_flag;
560 } pr_families[PR_FAMILY_MAX] = {
561 #ifdef INET
562 	[PR_INET] = {
563 		.size = sizeof(struct in_addr),
564 		.cmp = prison_qcmp_v4,
565 		.valid = prison_valid_v4,
566 		.ip_flag = PR_IP4_USER,
567 	 },
568 #endif
569 #ifdef INET6
570 	[PR_INET6] = {
571 		.size = sizeof(struct in6_addr),
572 		.cmp = prison_qcmp_v6,
573 		.valid = prison_valid_v6,
574 		.ip_flag = PR_IP6_USER,
575 	},
576 #endif
577 };
578 
579 /*
580  * Network address lists (pr_addrs) allocation for jails.  The addresses
581  * are accessed locklessly by the network stack, thus need to be protected by
582  * the network epoch.
583  */
584 struct prison_ip {
585 	struct epoch_context ctx;
586 	uint32_t	ips;
587 #ifdef FUTURE_C
588 	/*
589 	 * XXX Variable-length automatic arrays in union may be
590 	 * supported in future C.
591 	 */
592 	union {
593 		char pr_ip[];
594 		struct in_addr pr_ip4[];
595 		struct in6_addr pr_ip6[];
596 	};
597 #else /* No future C :( */
598 	char pr_ip[];
599 #endif
600 };
601 
602 static char *
603 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
604 {
605 	MPASS(pip);
606 	MPASS(af < PR_FAMILY_MAX);
607 	MPASS(idx >= 0 && idx < pip->ips);
608 
609 	return (pip->pr_ip + pr_families[af].size * idx);
610 }
611 
612 static struct prison_ip *
613 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
614 {
615 	struct prison_ip *pip;
616 
617 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
618 	    M_PRISON, flags);
619 	if (pip != NULL)
620 		pip->ips = cnt;
621 	return (pip);
622 }
623 
624 /*
625  * Allocate and copyin user supplied address list, sorting and validating.
626  * kern_jail_set() helper.
627  */
628 static struct prison_ip *
629 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
630 {
631 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
632 	const size_t size = pr_families[af].size;
633 	struct prison_ip *pip;
634 
635 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
636 	bcopy(op, pip->pr_ip, cnt * size);
637 	/*
638 	 * IP addresses are all sorted but ip[0] to preserve
639 	 * the primary IP address as given from userland.
640 	 * This special IP is used for unbound outgoing
641 	 * connections as well for "loopback" traffic in case
642 	 * source address selection cannot find any more fitting
643 	 * address to connect from.
644 	 */
645 	if (cnt > 1)
646 		qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
647 	/*
648 	 * Check for duplicate addresses and do some simple
649 	 * zero and broadcast checks. If users give other bogus
650 	 * addresses it is their problem.
651 	 */
652 	for (int i = 0; i < cnt; i++) {
653 		if (!pr_families[af].valid(PR_IP(pip, af, i))) {
654 			free(pip, M_PRISON);
655 			return (NULL);
656 		}
657 		if (i + 1 < cnt &&
658 		    (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
659 		     cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
660 			free(pip, M_PRISON);
661 			return (NULL);
662 		}
663 	}
664 
665 	return (pip);
666 }
667 
668 /*
669  * Allocate and dup parent prison address list.
670  * kern_jail_set() helper.
671  */
672 static void
673 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
674 {
675 	const struct prison_ip *ppip = ppr->pr_addrs[af];
676 	struct prison_ip *pip;
677 
678 	if (ppip != NULL) {
679 		pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
680 		bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
681 		pr->pr_addrs[af] = pip;
682 	}
683 }
684 
685 /*
686  * Make sure the new set of IP addresses is a subset of the parent's list.
687  * Don't worry about the parent being unlocked, as any setting is done with
688  * allprison_lock held.
689  * kern_jail_set() helper.
690  */
691 static bool
692 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
693     const pr_family_t af)
694 {
695 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
696 	int i, j;
697 
698 	if (ppip == NULL)
699 		return (false);
700 
701 	for (i = 0; i < ppip->ips; i++)
702 		if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
703 			break;
704 
705 	if (i == ppip->ips)
706 		/* Main address not present in parent. */
707 		return (false);
708 
709 	if (pip->ips > 1) {
710 		for (i = j = 1; i < pip->ips; i++) {
711 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
712 				/* Equals to parent primary address. */
713 				continue;
714 			for (; j < ppip->ips; j++)
715 				if (cmp(PR_IP(pip, af, i),
716 				    PR_IP(ppip, af, j)) == 0)
717 					break;
718 			if (j == ppip->ips)
719 				break;
720 		}
721 		if (j == ppip->ips)
722 			/* Address not present in parent. */
723 			return (false);
724 	}
725 	return (true);
726 }
727 
728 /*
729  * Check for conflicting IP addresses.  We permit them if there is no more
730  * than one IP on each jail.  If there is a duplicate on a jail with more
731  * than one IP stop checking and return error.
732  * kern_jail_set() helper.
733  */
734 static bool
735 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
736     struct prison_ip *pip, pr_family_t af)
737 {
738 	const struct prison *tppr, *tpr;
739 	int descend;
740 
741 #ifdef VIMAGE
742 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
743 		if (tppr->pr_flags & PR_VNET)
744 			break;
745 #else
746 	tppr = &prison0;
747 #endif
748 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
749 		if (tpr == pr ||
750 #ifdef VIMAGE
751 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
752 #endif
753 		    !prison_isalive(tpr)) {
754 			descend = 0;
755 			continue;
756 		}
757 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
758 			continue;
759 		descend = 0;
760 		if (tpr->pr_addrs[af] == NULL ||
761 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
762 			continue;
763 		for (int i = 0; i < pip->ips; i++)
764 			if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
765 				return (false);
766 	}
767 
768 	return (true);
769 }
770 
771 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
772     "prison must start with epoch context");
773 static void
774 prison_ip_free_deferred(epoch_context_t ctx)
775 {
776 
777 	free(ctx, M_PRISON);
778 }
779 
780 static void
781 prison_ip_free(struct prison_ip *pip)
782 {
783 
784 	if (pip != NULL)
785 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
786 }
787 
788 static void
789 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
790 {
791 	struct prison_ip **mem, *old;
792 
793 	mtx_assert(&pr->pr_mtx, MA_OWNED);
794 
795 	mem = &pr->pr_addrs[af];
796 
797 	old = *mem;
798 	atomic_store_ptr(mem, new);
799 	prison_ip_free(old);
800 }
801 
802 /*
803  * Restrict a prison's IP address list with its parent's, possibly replacing
804  * it.  Return true if succeed, otherwise should redo.
805  * kern_jail_set() helper.
806  */
807 static bool
808 prison_ip_restrict(struct prison *pr, const pr_family_t af,
809     struct prison_ip **newp)
810 {
811 	struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
812 	struct prison_ip *pip = pr->pr_addrs[af];
813 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
814 	const size_t size = pr_families[af].size;
815 	struct prison_ip *new = newp != NULL ? *newp : NULL;
816 	uint32_t ips;
817 
818 	mtx_assert(&pr->pr_mtx, MA_OWNED);
819 
820 	/*
821 	 * Due to epoch-synchronized access to the IP address lists we always
822 	 * allocate a new list even if the old one has enough space.  We could
823 	 * atomically update an IPv4 address inside a list, but that would
824 	 * screw up sorting, and in case of IPv6 we can't even atomically write
825 	 * one.
826 	 */
827 	if (ppip == NULL) {
828 		if (pip != NULL)
829 			prison_ip_set(pr, af, NULL);
830 		return (true);
831 	}
832 
833 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
834 		if (new == NULL) {
835 			new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
836 			if (new == NULL)
837 				return (false); /* Redo */
838 		}
839 		/* This has no user settings, so just copy the parent's list. */
840 		MPASS(new->ips == ppip->ips);
841 		bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
842 		prison_ip_set(pr, af, new);
843 		if (newp != NULL)
844 			*newp = NULL; /* Used */
845 	} else if (pip != NULL) {
846 		/* Remove addresses that aren't in the parent. */
847 		int i;
848 
849 		i = 0; /* index in pip */
850 		ips = 0; /* index in new */
851 
852 		if (new == NULL) {
853 			new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
854 			if (new == NULL)
855 				return (false); /* Redo */
856 		}
857 
858 		for (int pi = 0; pi < ppip->ips; pi++)
859 			if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
860 				/* Found our primary address in parent. */
861 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
862 				    size);
863 				i++;
864 				ips++;
865 				break;
866 			}
867 		for (int pi = 1; i < pip->ips; ) {
868 			/* Check against primary, which is unsorted. */
869 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
870 				/* Matches parent's primary address. */
871 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
872 				    size);
873 				i++;
874 				ips++;
875 				continue;
876 			}
877 			/* The rest are sorted. */
878 			switch (pi >= ppip->ips ? -1 :
879 				cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
880 			case -1:
881 				i++;
882 				break;
883 			case 0:
884 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
885 				    size);
886 				i++;
887 				pi++;
888 				ips++;
889 				break;
890 			case 1:
891 				pi++;
892 				break;
893 			}
894 		}
895 		if (ips == 0) {
896 			if (newp == NULL || *newp == NULL)
897 				prison_ip_free(new);
898 			new = NULL;
899 		} else {
900 			/* Shrink to real size */
901 			KASSERT((new->ips >= ips),
902 			    ("Out-of-bounds write to prison_ip %p", new));
903 			new->ips = ips;
904 		}
905 		prison_ip_set(pr, af, new);
906 		if (newp != NULL)
907 			*newp = NULL; /* Used */
908 	}
909 	return (true);
910 }
911 
912 /*
913  * Fast-path check if an address belongs to a prison.
914  */
915 int
916 prison_ip_check(const struct prison *pr, const pr_family_t af,
917     const void *addr)
918 {
919 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
920 	struct prison_ip *pip;
921 	int i, a, z, d;
922 
923 	MPASS(mtx_owned(&pr->pr_mtx) ||
924 	    in_epoch(net_epoch_preempt) ||
925 	    sx_xlocked(&allprison_lock));
926 
927 	pip = atomic_load_ptr(&pr->pr_addrs[af]);
928 	if (__predict_false(pip == NULL))
929 		return (EAFNOSUPPORT);
930 
931 	/* Check the primary IP. */
932 	if (cmp(PR_IP(pip, af, 0), addr) == 0)
933 		return (0);
934 
935 	/*
936 	 * All the other IPs are sorted so we can do a binary search.
937 	 */
938 	a = 0;
939 	z = pip->ips - 2;
940 	while (a <= z) {
941 		i = (a + z) / 2;
942 		d = cmp(PR_IP(pip, af, i + 1), addr);
943 		if (d > 0)
944 			z = i - 1;
945 		else if (d < 0)
946 			a = i + 1;
947 		else
948 			return (0);
949 	}
950 
951 	return (EADDRNOTAVAIL);
952 }
953 
954 /*
955  * Grab primary IP.  Historically required mutex, but nothing prevents
956  * us to support epoch-protected access.  Is it used in fast path?
957  * in{6}_jail.c helper
958  */
959 const void *
960 prison_ip_get0(const struct prison *pr, const pr_family_t af)
961 {
962 	const struct prison_ip *pip = pr->pr_addrs[af];
963 
964 	mtx_assert(&pr->pr_mtx, MA_OWNED);
965 	MPASS(pip);
966 
967 	return (pip->pr_ip);
968 }
969 
970 u_int
971 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
972 {
973 
974 	return (pr->pr_addrs[af]->ips);
975 }
976 #endif	/* defined(INET) || defined(INET6) */
977 
978 int
979 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
980 {
981 	struct nameidata nd;
982 #ifdef INET
983 	struct prison_ip *ip4;
984 #endif
985 #ifdef INET6
986 	struct prison_ip *ip6;
987 #endif
988 	struct vfsopt *opt;
989 	struct vfsoptlist *opts;
990 	struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
991 	struct vnode *root;
992 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
993 	char *g_path, *osrelstr;
994 	struct bool_flags *bf;
995 	struct jailsys_flags *jsf;
996 #if defined(INET) || defined(INET6)
997 	void *op;
998 #endif
999 	unsigned long hid;
1000 	size_t namelen, onamelen, pnamelen;
1001 	int created, cuflags, descend, drflags, enforce;
1002 	int error, errmsg_len, errmsg_pos;
1003 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1004 	int deadid, jid, jsys, len, level;
1005 	int childmax, osreldt, rsnum, slevel;
1006 #ifdef INET
1007 	int ip4s;
1008 	bool redo_ip4;
1009 #endif
1010 #ifdef INET6
1011 	int ip6s;
1012 	bool redo_ip6;
1013 #endif
1014 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1015 	uint64_t pr_allow_diff;
1016 	unsigned tallow;
1017 	char numbuf[12];
1018 
1019 	error = priv_check(td, PRIV_JAIL_SET);
1020 	if (!error && (flags & JAIL_ATTACH))
1021 		error = priv_check(td, PRIV_JAIL_ATTACH);
1022 	if (error)
1023 		return (error);
1024 	mypr = td->td_ucred->cr_prison;
1025 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
1026 		return (EPERM);
1027 	if (flags & ~JAIL_SET_MASK)
1028 		return (EINVAL);
1029 
1030 	/*
1031 	 * Check all the parameters before committing to anything.  Not all
1032 	 * errors can be caught early, but we may as well try.  Also, this
1033 	 * takes care of some expensive stuff (path lookup) before getting
1034 	 * the allprison lock.
1035 	 *
1036 	 * XXX Jails are not filesystems, and jail parameters are not mount
1037 	 *     options.  But it makes more sense to re-use the vfsopt code
1038 	 *     than duplicate it under a different name.
1039 	 */
1040 	error = vfs_buildopts(optuio, &opts);
1041 	if (error)
1042 		return (error);
1043 #ifdef INET
1044 	ip4 = NULL;
1045 #endif
1046 #ifdef INET6
1047 	ip6 = NULL;
1048 #endif
1049 	g_path = NULL;
1050 
1051 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1052 	if (!cuflags) {
1053 		error = EINVAL;
1054 		vfs_opterror(opts, "no valid operation (create or update)");
1055 		goto done_errmsg;
1056 	}
1057 
1058 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1059 	if (error == ENOENT)
1060 		jid = 0;
1061 	else if (error != 0)
1062 		goto done_free;
1063 
1064 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1065 	if (error == ENOENT)
1066 		gotslevel = 0;
1067 	else if (error != 0)
1068 		goto done_free;
1069 	else
1070 		gotslevel = 1;
1071 
1072 	error =
1073 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1074 	if (error == ENOENT)
1075 		gotchildmax = 0;
1076 	else if (error != 0)
1077 		goto done_free;
1078 	else
1079 		gotchildmax = 1;
1080 
1081 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1082 	if (error == ENOENT)
1083 		gotenforce = 0;
1084 	else if (error != 0)
1085 		goto done_free;
1086 	else if (enforce < 0 || enforce > 2) {
1087 		error = EINVAL;
1088 		goto done_free;
1089 	} else
1090 		gotenforce = 1;
1091 
1092 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1093 	if (error == ENOENT)
1094 		gotrsnum = 0;
1095 	else if (error != 0)
1096 		goto done_free;
1097 	else
1098 		gotrsnum = 1;
1099 
1100 	pr_flags = ch_flags = 0;
1101 	for (bf = pr_flag_bool;
1102 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1103 	     bf++) {
1104 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1105 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1106 	}
1107 	ch_flags |= pr_flags;
1108 	for (jsf = pr_flag_jailsys;
1109 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1110 	     jsf++) {
1111 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1112 		if (error == ENOENT)
1113 			continue;
1114 		if (error != 0)
1115 			goto done_free;
1116 		switch (jsys) {
1117 		case JAIL_SYS_DISABLE:
1118 			if (!jsf->disable) {
1119 				error = EINVAL;
1120 				goto done_free;
1121 			}
1122 			pr_flags |= jsf->disable;
1123 			break;
1124 		case JAIL_SYS_NEW:
1125 			pr_flags |= jsf->new;
1126 			break;
1127 		case JAIL_SYS_INHERIT:
1128 			break;
1129 		default:
1130 			error = EINVAL;
1131 			goto done_free;
1132 		}
1133 		ch_flags |= jsf->new | jsf->disable;
1134 	}
1135 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1136 	    && !(pr_flags & PR_PERSIST)) {
1137 		error = EINVAL;
1138 		vfs_opterror(opts, "new jail must persist or attach");
1139 		goto done_errmsg;
1140 	}
1141 #ifdef VIMAGE
1142 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1143 		error = EINVAL;
1144 		vfs_opterror(opts, "vnet cannot be changed after creation");
1145 		goto done_errmsg;
1146 	}
1147 #endif
1148 #ifdef INET
1149 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1150 		error = EINVAL;
1151 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1152 		goto done_errmsg;
1153 	}
1154 #endif
1155 #ifdef INET6
1156 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1157 		error = EINVAL;
1158 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1159 		goto done_errmsg;
1160 	}
1161 #endif
1162 
1163 	pr_allow = ch_allow = 0;
1164 	for (bf = pr_flag_allow;
1165 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1166 		atomic_load_int(&bf->flag) != 0;
1167 	     bf++) {
1168 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1169 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1170 	}
1171 	ch_allow |= pr_allow;
1172 
1173 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1174 	if (error == ENOENT)
1175 		name = NULL;
1176 	else if (error != 0)
1177 		goto done_free;
1178 	else {
1179 		if (len == 0 || name[len - 1] != '\0') {
1180 			error = EINVAL;
1181 			goto done_free;
1182 		}
1183 		if (len > MAXHOSTNAMELEN) {
1184 			error = ENAMETOOLONG;
1185 			goto done_free;
1186 		}
1187 	}
1188 
1189 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1190 	if (error == ENOENT)
1191 		host = NULL;
1192 	else if (error != 0)
1193 		goto done_free;
1194 	else {
1195 		ch_flags |= PR_HOST;
1196 		pr_flags |= PR_HOST;
1197 		if (len == 0 || host[len - 1] != '\0') {
1198 			error = EINVAL;
1199 			goto done_free;
1200 		}
1201 		if (len > MAXHOSTNAMELEN) {
1202 			error = ENAMETOOLONG;
1203 			goto done_free;
1204 		}
1205 	}
1206 
1207 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1208 	if (error == ENOENT)
1209 		domain = NULL;
1210 	else if (error != 0)
1211 		goto done_free;
1212 	else {
1213 		ch_flags |= PR_HOST;
1214 		pr_flags |= PR_HOST;
1215 		if (len == 0 || domain[len - 1] != '\0') {
1216 			error = EINVAL;
1217 			goto done_free;
1218 		}
1219 		if (len > MAXHOSTNAMELEN) {
1220 			error = ENAMETOOLONG;
1221 			goto done_free;
1222 		}
1223 	}
1224 
1225 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1226 	if (error == ENOENT)
1227 		uuid = NULL;
1228 	else if (error != 0)
1229 		goto done_free;
1230 	else {
1231 		ch_flags |= PR_HOST;
1232 		pr_flags |= PR_HOST;
1233 		if (len == 0 || uuid[len - 1] != '\0') {
1234 			error = EINVAL;
1235 			goto done_free;
1236 		}
1237 		if (len > HOSTUUIDLEN) {
1238 			error = ENAMETOOLONG;
1239 			goto done_free;
1240 		}
1241 	}
1242 
1243 #ifdef COMPAT_FREEBSD32
1244 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1245 		uint32_t hid32;
1246 
1247 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1248 		hid = hid32;
1249 	} else
1250 #endif
1251 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1252 	if (error == ENOENT)
1253 		gothid = 0;
1254 	else if (error != 0)
1255 		goto done_free;
1256 	else {
1257 		gothid = 1;
1258 		ch_flags |= PR_HOST;
1259 		pr_flags |= PR_HOST;
1260 	}
1261 
1262 #ifdef INET
1263 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1264 	if (error == ENOENT)
1265 		ip4s = 0;
1266 	else if (error != 0)
1267 		goto done_free;
1268 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1269 		error = EINVAL;
1270 		goto done_free;
1271 	} else {
1272 		ch_flags |= PR_IP4_USER;
1273 		pr_flags |= PR_IP4_USER;
1274 		if (ip4s > 0) {
1275 			ip4s /= sizeof(struct in_addr);
1276 			if (ip4s > jail_max_af_ips) {
1277 				error = EINVAL;
1278 				vfs_opterror(opts, "too many IPv4 addresses");
1279 				goto done_errmsg;
1280 			}
1281 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1282 			if (ip4 == NULL) {
1283 				error = EINVAL;
1284 				goto done_free;
1285 			}
1286 		}
1287 	}
1288 #endif
1289 
1290 #ifdef INET6
1291 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1292 	if (error == ENOENT)
1293 		ip6s = 0;
1294 	else if (error != 0)
1295 		goto done_free;
1296 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1297 		error = EINVAL;
1298 		goto done_free;
1299 	} else {
1300 		ch_flags |= PR_IP6_USER;
1301 		pr_flags |= PR_IP6_USER;
1302 		if (ip6s > 0) {
1303 			ip6s /= sizeof(struct in6_addr);
1304 			if (ip6s > jail_max_af_ips) {
1305 				error = EINVAL;
1306 				vfs_opterror(opts, "too many IPv6 addresses");
1307 				goto done_errmsg;
1308 			}
1309 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1310 			if (ip6 == NULL) {
1311 				error = EINVAL;
1312 				goto done_free;
1313 			}
1314 		}
1315 	}
1316 #endif
1317 
1318 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1319 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1320 		error = EINVAL;
1321 		vfs_opterror(opts,
1322 		    "vnet jails cannot have IP address restrictions");
1323 		goto done_errmsg;
1324 	}
1325 #endif
1326 
1327 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1328 	if (error == ENOENT)
1329 		osrelstr = NULL;
1330 	else if (error != 0)
1331 		goto done_free;
1332 	else {
1333 		if (flags & JAIL_UPDATE) {
1334 			error = EINVAL;
1335 			vfs_opterror(opts,
1336 			    "osrelease cannot be changed after creation");
1337 			goto done_errmsg;
1338 		}
1339 		if (len == 0 || osrelstr[len - 1] != '\0') {
1340 			error = EINVAL;
1341 			goto done_free;
1342 		}
1343 		if (len >= OSRELEASELEN) {
1344 			error = ENAMETOOLONG;
1345 			vfs_opterror(opts,
1346 			    "osrelease string must be 1-%d bytes long",
1347 			    OSRELEASELEN - 1);
1348 			goto done_errmsg;
1349 		}
1350 	}
1351 
1352 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1353 	if (error == ENOENT)
1354 		osreldt = 0;
1355 	else if (error != 0)
1356 		goto done_free;
1357 	else {
1358 		if (flags & JAIL_UPDATE) {
1359 			error = EINVAL;
1360 			vfs_opterror(opts,
1361 			    "osreldate cannot be changed after creation");
1362 			goto done_errmsg;
1363 		}
1364 		if (osreldt == 0) {
1365 			error = EINVAL;
1366 			vfs_opterror(opts, "osreldate cannot be 0");
1367 			goto done_errmsg;
1368 		}
1369 	}
1370 
1371 	root = NULL;
1372 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1373 	if (error == ENOENT)
1374 		path = NULL;
1375 	else if (error != 0)
1376 		goto done_free;
1377 	else {
1378 		if (flags & JAIL_UPDATE) {
1379 			error = EINVAL;
1380 			vfs_opterror(opts,
1381 			    "path cannot be changed after creation");
1382 			goto done_errmsg;
1383 		}
1384 		if (len == 0 || path[len - 1] != '\0') {
1385 			error = EINVAL;
1386 			goto done_free;
1387 		}
1388 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1389 		error = namei(&nd);
1390 		if (error)
1391 			goto done_free;
1392 		root = nd.ni_vp;
1393 		NDFREE_PNBUF(&nd);
1394 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1395 		strlcpy(g_path, path, MAXPATHLEN);
1396 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1397 		if (error == 0) {
1398 			path = g_path;
1399 		} else {
1400 			/* exit on other errors */
1401 			goto done_free;
1402 		}
1403 		if (root->v_type != VDIR) {
1404 			error = ENOTDIR;
1405 			vput(root);
1406 			goto done_free;
1407 		}
1408 		VOP_UNLOCK(root);
1409 	}
1410 
1411 	/*
1412 	 * Find the specified jail, or at least its parent.
1413 	 * This abuses the file error codes ENOENT and EEXIST.
1414 	 */
1415 	pr = NULL;
1416 	inspr = NULL;
1417 	deadpr = NULL;
1418 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1419 		namelc = strrchr(name, '.');
1420 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1421 		if (*p != '\0')
1422 			jid = 0;
1423 	}
1424 	sx_xlock(&allprison_lock);
1425 	drflags = PD_LIST_XLOCKED;
1426 	ppr = mypr;
1427 	if (!prison_isalive(ppr)) {
1428 		/* This jail is dying.  This process will surely follow. */
1429 		error = EAGAIN;
1430 		goto done_deref;
1431 	}
1432 	if (jid != 0) {
1433 		if (jid < 0) {
1434 			error = EINVAL;
1435 			vfs_opterror(opts, "negative jid");
1436 			goto done_deref;
1437 		}
1438 		/*
1439 		 * See if a requested jid already exists.  Keep track of
1440 		 * where it can be inserted later.
1441 		 */
1442 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1443 			if (inspr->pr_id < jid)
1444 				continue;
1445 			if (inspr->pr_id > jid)
1446 				break;
1447 			if (prison_isalive(inspr)) {
1448 				pr = inspr;
1449 				mtx_lock(&pr->pr_mtx);
1450 				drflags |= PD_LOCKED;
1451 			} else {
1452 				/* Note a dying jail to handle later. */
1453 				deadpr = inspr;
1454 			}
1455 			inspr = NULL;
1456 			break;
1457 		}
1458 		if (cuflags == JAIL_CREATE && pr != NULL) {
1459 			/*
1460 			 * Even creators that cannot see the jail will
1461 			 * get EEXIST.
1462 			 */
1463 			error = EEXIST;
1464 			vfs_opterror(opts, "jail %d already exists", jid);
1465 			goto done_deref;
1466 		}
1467 		if ((pr == NULL)
1468 		    ? cuflags == JAIL_UPDATE
1469 		    : !prison_ischild(mypr, pr)) {
1470 			/*
1471 			 * Updaters get ENOENT for nonexistent jails,
1472 			 * or for jails they cannot see.  The latter
1473 			 * case is true even for CREATE | UPDATE,
1474 			 * which normally cannot give this error.
1475 			 */
1476 			error = ENOENT;
1477 			vfs_opterror(opts, "jail %d not found", jid);
1478 			goto done_deref;
1479 		}
1480 	}
1481 	/*
1482 	 * If the caller provided a name, look for a jail by that name.
1483 	 * This has different semantics for creates and updates keyed by jid
1484 	 * (where the name must not already exist in a different jail),
1485 	 * and updates keyed by the name itself (where the name must exist
1486 	 * because that is the jail being updated).
1487 	 */
1488 	namelc = NULL;
1489 	if (name != NULL) {
1490 		namelc = strrchr(name, '.');
1491 		if (namelc == NULL)
1492 			namelc = name;
1493 		else {
1494 			/*
1495 			 * This is a hierarchical name.  Split it into the
1496 			 * parent and child names, and make sure the parent
1497 			 * exists or matches an already found jail.
1498 			 */
1499 			if (pr != NULL) {
1500 				if (strncmp(name, ppr->pr_name, namelc - name)
1501 				    || ppr->pr_name[namelc - name] != '\0') {
1502 					error = EINVAL;
1503 					vfs_opterror(opts,
1504 					    "cannot change jail's parent");
1505 					goto done_deref;
1506 				}
1507 			} else {
1508 				*namelc = '\0';
1509 				ppr = prison_find_name(mypr, name);
1510 				if (ppr == NULL) {
1511 					error = ENOENT;
1512 					vfs_opterror(opts,
1513 					    "jail \"%s\" not found", name);
1514 					goto done_deref;
1515 				}
1516 				mtx_unlock(&ppr->pr_mtx);
1517 				if (!prison_isalive(ppr)) {
1518 					error = ENOENT;
1519 					vfs_opterror(opts,
1520 					    "jail \"%s\" is dying", name);
1521 					goto done_deref;
1522 				}
1523 				*namelc = '.';
1524 			}
1525 			namelc++;
1526 		}
1527 		if (namelc[0] != '\0') {
1528 			pnamelen =
1529 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1530 			FOREACH_PRISON_CHILD(ppr, tpr) {
1531 				if (tpr == pr || !prison_isalive(tpr) ||
1532 				    strcmp(tpr->pr_name + pnamelen, namelc))
1533 					continue;
1534 				if (cuflags == JAIL_CREATE || pr != NULL) {
1535 					/*
1536 					 * Create, or update(jid): name must
1537 					 * not exist in an active sibling jail.
1538 					 */
1539 					error = EEXIST;
1540 					vfs_opterror(opts,
1541 					    "jail \"%s\" already exists", name);
1542 					goto done_deref;
1543 				}
1544 				/* Use this jail for updates. */
1545 				pr = tpr;
1546 				mtx_lock(&pr->pr_mtx);
1547 				drflags |= PD_LOCKED;
1548 				break;
1549 			}
1550 			/*
1551 			 * Update: name must exist if no jid is specified.
1552 			 * As with the jid case, the jail must be currently
1553 			 * visible, or else even CREATE | UPDATE will get
1554 			 * an error.
1555 			 */
1556 			if ((pr == NULL)
1557 			    ? cuflags == JAIL_UPDATE
1558 			    : !prison_isalive(pr)) {
1559 				error = ENOENT;
1560 				vfs_opterror(opts, "jail \"%s\" not found",
1561 				    name);
1562 				goto done_deref;
1563 			}
1564 		}
1565 	}
1566 	/* Update: must provide a jid or name. */
1567 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1568 		error = ENOENT;
1569 		vfs_opterror(opts, "update specified no jail");
1570 		goto done_deref;
1571 	}
1572 
1573 	/* If there's no prison to update, create a new one and link it in. */
1574 	created = pr == NULL;
1575 	if (created) {
1576 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1577 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1578 				error = EPERM;
1579 				vfs_opterror(opts, "prison limit exceeded");
1580 				goto done_deref;
1581 			}
1582 
1583 		if (deadpr != NULL) {
1584 			/*
1585 			 * The prison being created has the same ID as a dying
1586 			 * one.  Handle this by giving the dying jail a new ID.
1587 			 * This may cause some confusion to user space, but
1588 			 * only to those listing dying jails.
1589 			 */
1590 			deadid = get_next_deadid(&dinspr);
1591 			if (deadid == 0) {
1592 				error = EAGAIN;
1593 				vfs_opterror(opts, "no available jail IDs");
1594 				goto done_deref;
1595 			}
1596 			mtx_lock(&deadpr->pr_mtx);
1597 			deadpr->pr_id = deadid;
1598 			mtx_unlock(&deadpr->pr_mtx);
1599 			if (dinspr == deadpr)
1600 				inspr = deadpr;
1601 			else {
1602 				inspr = TAILQ_NEXT(deadpr, pr_list);
1603 				TAILQ_REMOVE(&allprison, deadpr, pr_list);
1604 				if (dinspr != NULL)
1605 					TAILQ_INSERT_AFTER(&allprison, dinspr,
1606 					    deadpr, pr_list);
1607 				else
1608 					TAILQ_INSERT_HEAD(&allprison, deadpr,
1609 					    pr_list);
1610 			}
1611 		}
1612 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1613 			error = EAGAIN;
1614 			vfs_opterror(opts, "no available jail IDs");
1615 			goto done_deref;
1616 		}
1617 
1618 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1619 		pr->pr_state = PRISON_STATE_INVALID;
1620 		refcount_init(&pr->pr_ref, 1);
1621 		refcount_init(&pr->pr_uref, 0);
1622 		drflags |= PD_DEREF;
1623 		LIST_INIT(&pr->pr_children);
1624 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1625 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1626 
1627 		pr->pr_id = jid;
1628 		if (inspr != NULL)
1629 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1630 		else
1631 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1632 
1633 		pr->pr_parent = ppr;
1634 		prison_hold(ppr);
1635 		prison_proc_hold(ppr);
1636 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1637 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1638 			tpr->pr_childcount++;
1639 
1640 		/* Set some default values, and inherit some from the parent. */
1641 		if (namelc == NULL)
1642 			namelc = "";
1643 		if (path == NULL) {
1644 			path = "/";
1645 			root = mypr->pr_root;
1646 			vref(root);
1647 		}
1648 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1649 		pr->pr_flags |= PR_HOST;
1650 #if defined(INET) || defined(INET6)
1651 #ifdef VIMAGE
1652 		if (!(pr_flags & PR_VNET))
1653 #endif
1654 		{
1655 #ifdef INET
1656 			if (!(ch_flags & PR_IP4_USER))
1657 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1658 			else if (!(pr_flags & PR_IP4_USER)) {
1659 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1660 				prison_ip_dup(ppr, pr, PR_INET);
1661 			}
1662 #endif
1663 #ifdef INET6
1664 			if (!(ch_flags & PR_IP6_USER))
1665 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1666 			else if (!(pr_flags & PR_IP6_USER)) {
1667 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1668 				prison_ip_dup(ppr, pr, PR_INET6);
1669 			}
1670 #endif
1671 		}
1672 #endif
1673 		/* Source address selection is always on by default. */
1674 		pr->pr_flags |= _PR_IP_SADDRSEL;
1675 
1676 		pr->pr_securelevel = ppr->pr_securelevel;
1677 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1678 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1679 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1680 
1681 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1682 		if (osrelstr == NULL)
1683 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1684 			    sizeof(pr->pr_osrelease));
1685 		else
1686 			strlcpy(pr->pr_osrelease, osrelstr,
1687 			    sizeof(pr->pr_osrelease));
1688 
1689 #ifdef VIMAGE
1690 		/* Allocate a new vnet if specified. */
1691 		pr->pr_vnet = (pr_flags & PR_VNET)
1692 		    ? vnet_alloc() : ppr->pr_vnet;
1693 #endif
1694 		/*
1695 		 * Allocate a dedicated cpuset for each jail.
1696 		 * Unlike other initial settings, this may return an error.
1697 		 */
1698 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1699 		if (error)
1700 			goto done_deref;
1701 
1702 		mtx_lock(&pr->pr_mtx);
1703 		drflags |= PD_LOCKED;
1704 	} else {
1705 		/*
1706 		 * Grab a reference for existing prisons, to ensure they
1707 		 * continue to exist for the duration of the call.
1708 		 */
1709 		prison_hold(pr);
1710 		drflags |= PD_DEREF;
1711 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1712 		if ((pr->pr_flags & PR_VNET) &&
1713 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1714 			error = EINVAL;
1715 			vfs_opterror(opts,
1716 			    "vnet jails cannot have IP address restrictions");
1717 			goto done_deref;
1718 		}
1719 #endif
1720 #ifdef INET
1721 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1722 			error = EINVAL;
1723 			vfs_opterror(opts,
1724 			    "ip4 cannot be changed after creation");
1725 			goto done_deref;
1726 		}
1727 #endif
1728 #ifdef INET6
1729 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1730 			error = EINVAL;
1731 			vfs_opterror(opts,
1732 			    "ip6 cannot be changed after creation");
1733 			goto done_deref;
1734 		}
1735 #endif
1736 	}
1737 
1738 	/* Do final error checking before setting anything. */
1739 	if (gotslevel) {
1740 		if (slevel < ppr->pr_securelevel) {
1741 			error = EPERM;
1742 			goto done_deref;
1743 		}
1744 	}
1745 	if (gotchildmax) {
1746 		if (childmax >= ppr->pr_childmax) {
1747 			error = EPERM;
1748 			goto done_deref;
1749 		}
1750 	}
1751 	if (gotenforce) {
1752 		if (enforce < ppr->pr_enforce_statfs) {
1753 			error = EPERM;
1754 			goto done_deref;
1755 		}
1756 	}
1757 	if (gotrsnum) {
1758 		/*
1759 		 * devfs_rsnum is a uint16_t
1760 		 */
1761 		if (rsnum < 0 || rsnum > 65535) {
1762 			error = EINVAL;
1763 			goto done_deref;
1764 		}
1765 		/*
1766 		 * Nested jails always inherit parent's devfs ruleset
1767 		 */
1768 		if (jailed(td->td_ucred)) {
1769 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1770 				error = EPERM;
1771 				goto done_deref;
1772 			} else
1773 				rsnum = ppr->pr_devfs_rsnum;
1774 		}
1775 	}
1776 #ifdef INET
1777 	if (ip4s > 0) {
1778 		if ((ppr->pr_flags & PR_IP4) &&
1779 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1780 		    PR_INET)) {
1781 			error = EPERM;
1782 			goto done_deref;
1783 		}
1784 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1785 			error = EADDRINUSE;
1786 			vfs_opterror(opts, "IPv4 addresses clash");
1787 			goto done_deref;
1788 		}
1789 	}
1790 #endif
1791 #ifdef INET6
1792 	if (ip6s > 0) {
1793 		if ((ppr->pr_flags & PR_IP6) &&
1794 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1795 		    PR_INET6)) {
1796 			error = EPERM;
1797 			goto done_deref;
1798 		}
1799 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1800 			error = EADDRINUSE;
1801 			vfs_opterror(opts, "IPv6 addresses clash");
1802 			goto done_deref;
1803 		}
1804 	}
1805 #endif
1806 	onamelen = namelen = 0;
1807 	if (namelc != NULL) {
1808 		/* Give a default name of the jid.  Also allow the name to be
1809 		 * explicitly the jid - but not any other number, and only in
1810 		 * normal form (no leading zero/etc).
1811 		 */
1812 		if (namelc[0] == '\0')
1813 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1814 		else if ((strtoul(namelc, &p, 10) != jid ||
1815 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1816 			error = EINVAL;
1817 			vfs_opterror(opts,
1818 			    "name cannot be numeric (unless it is the jid)");
1819 			goto done_deref;
1820 		}
1821 		/*
1822 		 * Make sure the name isn't too long for the prison or its
1823 		 * children.
1824 		 */
1825 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1826 		onamelen = strlen(pr->pr_name + pnamelen);
1827 		namelen = strlen(namelc);
1828 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1829 			error = ENAMETOOLONG;
1830 			goto done_deref;
1831 		}
1832 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1833 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1834 			    sizeof(pr->pr_name)) {
1835 				error = ENAMETOOLONG;
1836 				goto done_deref;
1837 			}
1838 		}
1839 	}
1840 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1841 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1842 		error = EPERM;
1843 		goto done_deref;
1844 	}
1845 
1846 	/*
1847 	 * Let modules check their parameters.  This requires unlocking and
1848 	 * then re-locking the prison, but this is still a valid state as long
1849 	 * as allprison_lock remains xlocked.
1850 	 */
1851 	mtx_unlock(&pr->pr_mtx);
1852 	drflags &= ~PD_LOCKED;
1853 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1854 	if (error != 0)
1855 		goto done_deref;
1856 	mtx_lock(&pr->pr_mtx);
1857 	drflags |= PD_LOCKED;
1858 
1859 	/* At this point, all valid parameters should have been noted. */
1860 	TAILQ_FOREACH(opt, opts, link) {
1861 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1862 			error = EINVAL;
1863 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1864 			goto done_deref;
1865 		}
1866 	}
1867 
1868 	/* Set the parameters of the prison. */
1869 #ifdef INET
1870 	redo_ip4 = false;
1871 	if (pr_flags & PR_IP4_USER) {
1872 		pr->pr_flags |= PR_IP4;
1873 		prison_ip_set(pr, PR_INET, ip4);
1874 		ip4 = NULL;
1875 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1876 #ifdef VIMAGE
1877 			if (tpr->pr_flags & PR_VNET) {
1878 				descend = 0;
1879 				continue;
1880 			}
1881 #endif
1882 			if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
1883 				redo_ip4 = true;
1884 				descend = 0;
1885 			}
1886 		}
1887 	}
1888 #endif
1889 #ifdef INET6
1890 	redo_ip6 = false;
1891 	if (pr_flags & PR_IP6_USER) {
1892 		pr->pr_flags |= PR_IP6;
1893 		prison_ip_set(pr, PR_INET6, ip6);
1894 		ip6 = NULL;
1895 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1896 #ifdef VIMAGE
1897 			if (tpr->pr_flags & PR_VNET) {
1898 				descend = 0;
1899 				continue;
1900 			}
1901 #endif
1902 			if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
1903 				redo_ip6 = true;
1904 				descend = 0;
1905 			}
1906 		}
1907 	}
1908 #endif
1909 	if (gotslevel) {
1910 		pr->pr_securelevel = slevel;
1911 		/* Set all child jails to be at least this level. */
1912 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1913 			if (tpr->pr_securelevel < slevel)
1914 				tpr->pr_securelevel = slevel;
1915 	}
1916 	if (gotchildmax) {
1917 		pr->pr_childmax = childmax;
1918 		/* Set all child jails to under this limit. */
1919 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1920 			if (tpr->pr_childmax > childmax - level)
1921 				tpr->pr_childmax = childmax > level
1922 				    ? childmax - level : 0;
1923 	}
1924 	if (gotenforce) {
1925 		pr->pr_enforce_statfs = enforce;
1926 		/* Pass this restriction on to the children. */
1927 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1928 			if (tpr->pr_enforce_statfs < enforce)
1929 				tpr->pr_enforce_statfs = enforce;
1930 	}
1931 	if (gotrsnum) {
1932 		pr->pr_devfs_rsnum = rsnum;
1933 		/* Pass this restriction on to the children. */
1934 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1935 			tpr->pr_devfs_rsnum = rsnum;
1936 	}
1937 	if (namelc != NULL) {
1938 		if (ppr == &prison0)
1939 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1940 		else
1941 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1942 			    ppr->pr_name, namelc);
1943 		/* Change this component of child names. */
1944 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1945 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1946 			    strlen(tpr->pr_name + onamelen) + 1);
1947 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1948 		}
1949 	}
1950 	if (path != NULL) {
1951 		/* Try to keep a real-rooted full pathname. */
1952 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1953 		pr->pr_root = root;
1954 		root = NULL;
1955 	}
1956 	if (PR_HOST & ch_flags & ~pr_flags) {
1957 		if (pr->pr_flags & PR_HOST) {
1958 			/*
1959 			 * Copy the parent's host info.  As with pr_ip4 above,
1960 			 * the lack of a lock on the parent is not a problem;
1961 			 * it is always set with allprison_lock at least
1962 			 * shared, and is held exclusively here.
1963 			 */
1964 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1965 			    sizeof(pr->pr_hostname));
1966 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1967 			    sizeof(pr->pr_domainname));
1968 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1969 			    sizeof(pr->pr_hostuuid));
1970 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1971 		}
1972 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1973 		/* Set this prison, and any descendants without PR_HOST. */
1974 		if (host != NULL)
1975 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1976 		if (domain != NULL)
1977 			strlcpy(pr->pr_domainname, domain,
1978 			    sizeof(pr->pr_domainname));
1979 		if (uuid != NULL)
1980 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1981 		if (gothid)
1982 			pr->pr_hostid = hid;
1983 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1984 			if (tpr->pr_flags & PR_HOST)
1985 				descend = 0;
1986 			else {
1987 				if (host != NULL)
1988 					strlcpy(tpr->pr_hostname,
1989 					    pr->pr_hostname,
1990 					    sizeof(tpr->pr_hostname));
1991 				if (domain != NULL)
1992 					strlcpy(tpr->pr_domainname,
1993 					    pr->pr_domainname,
1994 					    sizeof(tpr->pr_domainname));
1995 				if (uuid != NULL)
1996 					strlcpy(tpr->pr_hostuuid,
1997 					    pr->pr_hostuuid,
1998 					    sizeof(tpr->pr_hostuuid));
1999 				if (gothid)
2000 					tpr->pr_hostid = hid;
2001 			}
2002 		}
2003 	}
2004 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2005 	if ((tallow = ch_allow & ~pr_allow))
2006 		prison_set_allow_locked(pr, tallow, 0);
2007 	/*
2008 	 * Persistent prisons get an extra reference, and prisons losing their
2009 	 * persist flag lose that reference.
2010 	 */
2011 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2012 		if (pr_flags & PR_PERSIST) {
2013 			prison_hold(pr);
2014 			/*
2015 			 * This may be a new prison's first user reference,
2016 			 * but wait to call it alive until after OSD calls
2017 			 * have had a chance to run (and perhaps to fail).
2018 			 */
2019 			refcount_acquire(&pr->pr_uref);
2020 		} else {
2021 			drflags |= PD_DEUREF;
2022 			prison_free_not_last(pr);
2023 		}
2024 	}
2025 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2026 	mtx_unlock(&pr->pr_mtx);
2027 	drflags &= ~PD_LOCKED;
2028 	/*
2029 	 * Any errors past this point will need to de-persist newly created
2030 	 * prisons, as well as call remove methods.
2031 	 */
2032 	if (created)
2033 		drflags |= PD_KILL;
2034 
2035 #ifdef RACCT
2036 	if (racct_enable && created)
2037 		prison_racct_attach(pr);
2038 #endif
2039 
2040 	/* Locks may have prevented a complete restriction of child IP
2041 	 * addresses.  If so, allocate some more memory and try again.
2042 	 */
2043 #ifdef INET
2044 	while (redo_ip4) {
2045 		ip4s = pr->pr_addrs[PR_INET]->ips;
2046 		MPASS(ip4 == NULL);
2047 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2048 		mtx_lock(&pr->pr_mtx);
2049 		redo_ip4 = false;
2050 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2051 #ifdef VIMAGE
2052 			if (tpr->pr_flags & PR_VNET) {
2053 				descend = 0;
2054 				continue;
2055 			}
2056 #endif
2057 			if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2058 				redo_ip4 = true;
2059 		}
2060 		mtx_unlock(&pr->pr_mtx);
2061 	}
2062 #endif
2063 #ifdef INET6
2064 	while (redo_ip6) {
2065 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2066 		MPASS(ip6 == NULL);
2067 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2068 		mtx_lock(&pr->pr_mtx);
2069 		redo_ip6 = false;
2070 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2071 #ifdef VIMAGE
2072 			if (tpr->pr_flags & PR_VNET) {
2073 				descend = 0;
2074 				continue;
2075 			}
2076 #endif
2077 			if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2078 				redo_ip6 = true;
2079 		}
2080 		mtx_unlock(&pr->pr_mtx);
2081 	}
2082 #endif
2083 
2084 	/* Let the modules do their work. */
2085 	if (created) {
2086 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2087 		if (error)
2088 			goto done_deref;
2089 	}
2090 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2091 	if (error)
2092 		goto done_deref;
2093 
2094 	/*
2095 	 * A new prison is now ready to be seen; either it has gained a user
2096 	 * reference via persistence, or is about to gain one via attachment.
2097 	 */
2098 	if (created) {
2099 		drflags = prison_lock_xlock(pr, drflags);
2100 		pr->pr_state = PRISON_STATE_ALIVE;
2101 	}
2102 
2103 	/* Attach this process to the prison if requested. */
2104 	if (flags & JAIL_ATTACH) {
2105 		error = do_jail_attach(td, pr,
2106 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2107 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2108 		if (error) {
2109 			vfs_opterror(opts, "attach failed");
2110 			goto done_deref;
2111 		}
2112 	}
2113 
2114 #ifdef RACCT
2115 	if (racct_enable && !created) {
2116 		if (drflags & PD_LOCKED) {
2117 			mtx_unlock(&pr->pr_mtx);
2118 			drflags &= ~PD_LOCKED;
2119 		}
2120 		if (drflags & PD_LIST_XLOCKED) {
2121 			sx_xunlock(&allprison_lock);
2122 			drflags &= ~PD_LIST_XLOCKED;
2123 		}
2124 		prison_racct_modify(pr);
2125 	}
2126 #endif
2127 
2128 	if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2129 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2130 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2131 		   " file system\n", pr->pr_id);
2132 
2133 	drflags &= ~PD_KILL;
2134 	td->td_retval[0] = pr->pr_id;
2135 
2136  done_deref:
2137 	/* Release any temporary prison holds and/or locks. */
2138 	if (pr != NULL)
2139 		prison_deref(pr, drflags);
2140 	else if (drflags & PD_LIST_SLOCKED)
2141 		sx_sunlock(&allprison_lock);
2142 	else if (drflags & PD_LIST_XLOCKED)
2143 		sx_xunlock(&allprison_lock);
2144 	if (root != NULL)
2145 		vrele(root);
2146  done_errmsg:
2147 	if (error) {
2148 		/* Write the error message back to userspace. */
2149 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2150 		    &errmsg_len) == 0 && errmsg_len > 0) {
2151 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2152 			if (optuio->uio_segflg == UIO_SYSSPACE)
2153 				bcopy(errmsg,
2154 				    optuio->uio_iov[errmsg_pos].iov_base,
2155 				    errmsg_len);
2156 			else
2157 				(void)copyout(errmsg,
2158 				    optuio->uio_iov[errmsg_pos].iov_base,
2159 				    errmsg_len);
2160 		}
2161 	}
2162  done_free:
2163 #ifdef INET
2164 	prison_ip_free(ip4);
2165 #endif
2166 #ifdef INET6
2167 	prison_ip_free(ip6);
2168 #endif
2169 	if (g_path != NULL)
2170 		free(g_path, M_TEMP);
2171 	vfs_freeopts(opts);
2172 	return (error);
2173 }
2174 
2175 /*
2176  * Find the next available prison ID.  Return the ID on success, or zero
2177  * on failure.  Also set a pointer to the allprison list entry the prison
2178  * should be inserted before.
2179  */
2180 static int
2181 get_next_prid(struct prison **insprp)
2182 {
2183 	struct prison *inspr;
2184 	int jid, maxid;
2185 
2186 	jid = lastprid % JAIL_MAX + 1;
2187 	if (TAILQ_EMPTY(&allprison) ||
2188 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2189 		/*
2190 		 * A common case is for all jails to be implicitly numbered,
2191 		 * which means they'll go on the end of the list, at least
2192 		 * for the first JAIL_MAX times.
2193 		 */
2194 		inspr = NULL;
2195 	} else {
2196 		/*
2197 		 * Take two passes through the allprison list: first starting
2198 		 * with the proposed jid, then ending with it.
2199 		 */
2200 		for (maxid = JAIL_MAX; maxid != 0; ) {
2201 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2202 				if (inspr->pr_id < jid)
2203 					continue;
2204 				if (inspr->pr_id > jid) {
2205 					/* Found an opening. */
2206 					maxid = 0;
2207 					break;
2208 				}
2209 				if (++jid > maxid) {
2210 					if (lastprid == maxid || lastprid == 0)
2211 					{
2212 						/*
2213 						 * The entire legal range
2214 						 * has been traversed
2215 						 */
2216 						return 0;
2217 					}
2218 					/* Try again from the start. */
2219 					jid = 1;
2220 					maxid = lastprid;
2221 					break;
2222 				}
2223 			}
2224 			if (inspr == NULL) {
2225 				/* Found room at the end of the list. */
2226 				break;
2227 			}
2228 		}
2229 	}
2230 	*insprp = inspr;
2231 	lastprid = jid;
2232 	return (jid);
2233 }
2234 
2235 /*
2236  * Find the next available ID for a renumbered dead prison.  This is the same
2237  * as get_next_prid, but counting backward from the end of the range.
2238  */
2239 static int
2240 get_next_deadid(struct prison **dinsprp)
2241 {
2242 	struct prison *dinspr;
2243 	int deadid, minid;
2244 
2245 	deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2246 	/*
2247 	 * Take two reverse passes through the allprison list: first
2248 	 * starting with the proposed deadid, then ending with it.
2249 	 */
2250 	for (minid = 1; minid != 0; ) {
2251 		TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2252 			if (dinspr->pr_id > deadid)
2253 				continue;
2254 			if (dinspr->pr_id < deadid) {
2255 				/* Found an opening. */
2256 				minid = 0;
2257 				break;
2258 			}
2259 			if (--deadid < minid) {
2260 				if (lastdeadid == minid || lastdeadid == 0)
2261 				{
2262 					/*
2263 					 * The entire legal range
2264 					 * has been traversed
2265 					 */
2266 					return 0;
2267 				}
2268 				/* Try again from the end. */
2269 				deadid = JAIL_MAX;
2270 				minid = lastdeadid;
2271 				break;
2272 			}
2273 		}
2274 		if (dinspr == NULL) {
2275 			/* Found room at the beginning of the list. */
2276 			break;
2277 		}
2278 	}
2279 	*dinsprp = dinspr;
2280 	lastdeadid = deadid;
2281 	return (deadid);
2282 }
2283 
2284 /*
2285  * struct jail_get_args {
2286  *	struct iovec *iovp;
2287  *	unsigned int iovcnt;
2288  *	int flags;
2289  * };
2290  */
2291 int
2292 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2293 {
2294 	struct uio *auio;
2295 	int error;
2296 
2297 	/* Check that we have an even number of iovecs. */
2298 	if (uap->iovcnt & 1)
2299 		return (EINVAL);
2300 
2301 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2302 	if (error)
2303 		return (error);
2304 	error = kern_jail_get(td, auio, uap->flags);
2305 	if (error == 0)
2306 		error = copyout(auio->uio_iov, uap->iovp,
2307 		    uap->iovcnt * sizeof(struct iovec));
2308 	freeuio(auio);
2309 	return (error);
2310 }
2311 
2312 int
2313 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2314 {
2315 	struct bool_flags *bf;
2316 	struct jailsys_flags *jsf;
2317 	struct prison *pr, *mypr;
2318 	struct vfsopt *opt;
2319 	struct vfsoptlist *opts;
2320 	char *errmsg, *name;
2321 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2322 	unsigned f;
2323 
2324 	if (flags & ~JAIL_GET_MASK)
2325 		return (EINVAL);
2326 
2327 	/* Get the parameter list. */
2328 	error = vfs_buildopts(optuio, &opts);
2329 	if (error)
2330 		return (error);
2331 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2332 	mypr = td->td_ucred->cr_prison;
2333 	pr = NULL;
2334 
2335 	/*
2336 	 * Find the prison specified by one of: lastjid, jid, name.
2337 	 */
2338 	sx_slock(&allprison_lock);
2339 	drflags = PD_LIST_SLOCKED;
2340 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2341 	if (error == 0) {
2342 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2343 			if (pr->pr_id > jid &&
2344 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2345 			    prison_ischild(mypr, pr)) {
2346 				mtx_lock(&pr->pr_mtx);
2347 				drflags |= PD_LOCKED;
2348 				goto found_prison;
2349 			}
2350 		}
2351 		error = ENOENT;
2352 		vfs_opterror(opts, "no jail after %d", jid);
2353 		goto done;
2354 	} else if (error != ENOENT)
2355 		goto done;
2356 
2357 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2358 	if (error == 0) {
2359 		if (jid != 0) {
2360 			pr = prison_find_child(mypr, jid);
2361 			if (pr != NULL) {
2362 				drflags |= PD_LOCKED;
2363 				if (!(prison_isalive(pr) ||
2364 				    (flags & JAIL_DYING))) {
2365 					error = ENOENT;
2366 					vfs_opterror(opts, "jail %d is dying",
2367 					    jid);
2368 					goto done;
2369 				}
2370 				goto found_prison;
2371 			}
2372 			error = ENOENT;
2373 			vfs_opterror(opts, "jail %d not found", jid);
2374 			goto done;
2375 		}
2376 	} else if (error != ENOENT)
2377 		goto done;
2378 
2379 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2380 	if (error == 0) {
2381 		if (len == 0 || name[len - 1] != '\0') {
2382 			error = EINVAL;
2383 			goto done;
2384 		}
2385 		pr = prison_find_name(mypr, name);
2386 		if (pr != NULL) {
2387 			drflags |= PD_LOCKED;
2388 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2389 				error = ENOENT;
2390 				vfs_opterror(opts, "jail \"%s\" is dying",
2391 				    name);
2392 				goto done;
2393 			}
2394 			goto found_prison;
2395 		}
2396 		error = ENOENT;
2397 		vfs_opterror(opts, "jail \"%s\" not found", name);
2398 		goto done;
2399 	} else if (error != ENOENT)
2400 		goto done;
2401 
2402 	vfs_opterror(opts, "no jail specified");
2403 	error = ENOENT;
2404 	goto done;
2405 
2406  found_prison:
2407 	/* Get the parameters of the prison. */
2408 	prison_hold(pr);
2409 	drflags |= PD_DEREF;
2410 	td->td_retval[0] = pr->pr_id;
2411 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2412 	if (error != 0 && error != ENOENT)
2413 		goto done;
2414 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2415 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2416 	if (error != 0 && error != ENOENT)
2417 		goto done;
2418 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2419 	if (error != 0 && error != ENOENT)
2420 		goto done;
2421 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2422 	    sizeof(pr->pr_cpuset->cs_id));
2423 	if (error != 0 && error != ENOENT)
2424 		goto done;
2425 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2426 	if (error != 0 && error != ENOENT)
2427 		goto done;
2428 #ifdef INET
2429 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2430 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2431 	    pr_families[PR_INET].size : 0 );
2432 	if (error != 0 && error != ENOENT)
2433 		goto done;
2434 #endif
2435 #ifdef INET6
2436 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2437 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2438 	    pr_families[PR_INET6].size : 0 );
2439 	if (error != 0 && error != ENOENT)
2440 		goto done;
2441 #endif
2442 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2443 	    sizeof(pr->pr_securelevel));
2444 	if (error != 0 && error != ENOENT)
2445 		goto done;
2446 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2447 	    sizeof(pr->pr_childcount));
2448 	if (error != 0 && error != ENOENT)
2449 		goto done;
2450 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2451 	    sizeof(pr->pr_childmax));
2452 	if (error != 0 && error != ENOENT)
2453 		goto done;
2454 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2455 	if (error != 0 && error != ENOENT)
2456 		goto done;
2457 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2458 	if (error != 0 && error != ENOENT)
2459 		goto done;
2460 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2461 	if (error != 0 && error != ENOENT)
2462 		goto done;
2463 #ifdef COMPAT_FREEBSD32
2464 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2465 		uint32_t hid32 = pr->pr_hostid;
2466 
2467 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2468 	} else
2469 #endif
2470 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2471 	    sizeof(pr->pr_hostid));
2472 	if (error != 0 && error != ENOENT)
2473 		goto done;
2474 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2475 	    sizeof(pr->pr_enforce_statfs));
2476 	if (error != 0 && error != ENOENT)
2477 		goto done;
2478 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2479 	    sizeof(pr->pr_devfs_rsnum));
2480 	if (error != 0 && error != ENOENT)
2481 		goto done;
2482 	for (bf = pr_flag_bool;
2483 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2484 	     bf++) {
2485 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2486 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2487 		if (error != 0 && error != ENOENT)
2488 			goto done;
2489 		i = !i;
2490 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2491 		if (error != 0 && error != ENOENT)
2492 			goto done;
2493 	}
2494 	for (jsf = pr_flag_jailsys;
2495 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2496 	     jsf++) {
2497 		f = pr->pr_flags & (jsf->disable | jsf->new);
2498 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2499 		    : (f == jsf->new) ? JAIL_SYS_NEW
2500 		    : JAIL_SYS_INHERIT;
2501 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2502 		if (error != 0 && error != ENOENT)
2503 			goto done;
2504 	}
2505 	for (bf = pr_flag_allow;
2506 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2507 		atomic_load_int(&bf->flag) != 0;
2508 	     bf++) {
2509 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2510 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2511 		if (error != 0 && error != ENOENT)
2512 			goto done;
2513 		i = !i;
2514 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2515 		if (error != 0 && error != ENOENT)
2516 			goto done;
2517 	}
2518 	i = !prison_isalive(pr);
2519 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2520 	if (error != 0 && error != ENOENT)
2521 		goto done;
2522 	i = !i;
2523 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2524 	if (error != 0 && error != ENOENT)
2525 		goto done;
2526 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2527 	    sizeof(pr->pr_osreldate));
2528 	if (error != 0 && error != ENOENT)
2529 		goto done;
2530 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2531 	if (error != 0 && error != ENOENT)
2532 		goto done;
2533 
2534 	/* Get the module parameters. */
2535 	mtx_unlock(&pr->pr_mtx);
2536 	drflags &= ~PD_LOCKED;
2537 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2538 	if (error)
2539 		goto done;
2540 	prison_deref(pr, drflags);
2541 	pr = NULL;
2542 	drflags = 0;
2543 
2544 	/* By now, all parameters should have been noted. */
2545 	TAILQ_FOREACH(opt, opts, link) {
2546 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2547 			error = EINVAL;
2548 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2549 			goto done;
2550 		}
2551 	}
2552 
2553 	/* Write the fetched parameters back to userspace. */
2554 	error = 0;
2555 	TAILQ_FOREACH(opt, opts, link) {
2556 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2557 			pos = 2 * opt->pos + 1;
2558 			optuio->uio_iov[pos].iov_len = opt->len;
2559 			if (opt->value != NULL) {
2560 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2561 					bcopy(opt->value,
2562 					    optuio->uio_iov[pos].iov_base,
2563 					    opt->len);
2564 				} else {
2565 					error = copyout(opt->value,
2566 					    optuio->uio_iov[pos].iov_base,
2567 					    opt->len);
2568 					if (error)
2569 						break;
2570 				}
2571 			}
2572 		}
2573 	}
2574 
2575  done:
2576 	/* Release any temporary prison holds and/or locks. */
2577 	if (pr != NULL)
2578 		prison_deref(pr, drflags);
2579 	else if (drflags & PD_LIST_SLOCKED)
2580 		sx_sunlock(&allprison_lock);
2581 	if (error && errmsg_pos >= 0) {
2582 		/* Write the error message back to userspace. */
2583 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2584 		errmsg_pos = 2 * errmsg_pos + 1;
2585 		if (errmsg_len > 0) {
2586 			if (optuio->uio_segflg == UIO_SYSSPACE)
2587 				bcopy(errmsg,
2588 				    optuio->uio_iov[errmsg_pos].iov_base,
2589 				    errmsg_len);
2590 			else
2591 				(void)copyout(errmsg,
2592 				    optuio->uio_iov[errmsg_pos].iov_base,
2593 				    errmsg_len);
2594 		}
2595 	}
2596 	vfs_freeopts(opts);
2597 	return (error);
2598 }
2599 
2600 /*
2601  * struct jail_remove_args {
2602  *	int jid;
2603  * };
2604  */
2605 int
2606 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2607 {
2608 	struct prison *pr;
2609 	int error;
2610 
2611 	error = priv_check(td, PRIV_JAIL_REMOVE);
2612 	if (error)
2613 		return (error);
2614 
2615 	sx_xlock(&allprison_lock);
2616 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2617 	if (pr == NULL) {
2618 		sx_xunlock(&allprison_lock);
2619 		return (EINVAL);
2620 	}
2621 	if (!prison_isalive(pr)) {
2622 		/* Silently ignore already-dying prisons. */
2623 		mtx_unlock(&pr->pr_mtx);
2624 		sx_xunlock(&allprison_lock);
2625 		return (0);
2626 	}
2627 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2628 	return (0);
2629 }
2630 
2631 /*
2632  * struct jail_attach_args {
2633  *	int jid;
2634  * };
2635  */
2636 int
2637 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2638 {
2639 	struct prison *pr;
2640 	int error;
2641 
2642 	error = priv_check(td, PRIV_JAIL_ATTACH);
2643 	if (error)
2644 		return (error);
2645 
2646 	sx_slock(&allprison_lock);
2647 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2648 	if (pr == NULL) {
2649 		sx_sunlock(&allprison_lock);
2650 		return (EINVAL);
2651 	}
2652 
2653 	/* Do not allow a process to attach to a prison that is not alive. */
2654 	if (!prison_isalive(pr)) {
2655 		mtx_unlock(&pr->pr_mtx);
2656 		sx_sunlock(&allprison_lock);
2657 		return (EINVAL);
2658 	}
2659 
2660 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2661 }
2662 
2663 static int
2664 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2665 {
2666 	struct proc *p;
2667 	struct ucred *newcred, *oldcred;
2668 	int error;
2669 
2670 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2671 	sx_assert(&allprison_lock, SX_LOCKED);
2672 	drflags &= PD_LOCK_FLAGS;
2673 	/*
2674 	 * XXX: Note that there is a slight race here if two threads
2675 	 * in the same privileged process attempt to attach to two
2676 	 * different jails at the same time.  It is important for
2677 	 * user processes not to do this, or they might end up with
2678 	 * a process root from one prison, but attached to the jail
2679 	 * of another.
2680 	 */
2681 	prison_hold(pr);
2682 	refcount_acquire(&pr->pr_uref);
2683 	drflags |= PD_DEREF | PD_DEUREF;
2684 	mtx_unlock(&pr->pr_mtx);
2685 	drflags &= ~PD_LOCKED;
2686 
2687 	/* Let modules do whatever they need to prepare for attaching. */
2688 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2689 	if (error) {
2690 		prison_deref(pr, drflags);
2691 		return (error);
2692 	}
2693 	sx_unlock(&allprison_lock);
2694 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2695 
2696 	/*
2697 	 * Reparent the newly attached process to this jail.
2698 	 */
2699 	p = td->td_proc;
2700 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2701 	if (error)
2702 		goto e_revert_osd;
2703 
2704 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2705 	if ((error = change_dir(pr->pr_root, td)) != 0)
2706 		goto e_unlock;
2707 #ifdef MAC
2708 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2709 		goto e_unlock;
2710 #endif
2711 	VOP_UNLOCK(pr->pr_root);
2712 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2713 		goto e_revert_osd;
2714 
2715 	newcred = crget();
2716 	PROC_LOCK(p);
2717 	oldcred = crcopysafe(p, newcred);
2718 	newcred->cr_prison = pr;
2719 	proc_set_cred(p, newcred);
2720 	setsugid(p);
2721 #ifdef RACCT
2722 	racct_proc_ucred_changed(p, oldcred, newcred);
2723 	crhold(newcred);
2724 #endif
2725 	PROC_UNLOCK(p);
2726 #ifdef RCTL
2727 	rctl_proc_ucred_changed(p, newcred);
2728 	crfree(newcred);
2729 #endif
2730 	prison_proc_relink(oldcred->cr_prison, pr, p);
2731 	prison_deref(oldcred->cr_prison, drflags);
2732 	crfree(oldcred);
2733 
2734 	/*
2735 	 * If the prison was killed while changing credentials, die along
2736 	 * with it.
2737 	 */
2738 	if (!prison_isalive(pr)) {
2739 		PROC_LOCK(p);
2740 		kern_psignal(p, SIGKILL);
2741 		PROC_UNLOCK(p);
2742 	}
2743 
2744 	return (0);
2745 
2746  e_unlock:
2747 	VOP_UNLOCK(pr->pr_root);
2748  e_revert_osd:
2749 	/* Tell modules this thread is still in its old jail after all. */
2750 	sx_slock(&allprison_lock);
2751 	drflags |= PD_LIST_SLOCKED;
2752 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2753 	prison_deref(pr, drflags);
2754 	return (error);
2755 }
2756 
2757 /*
2758  * Returns a locked prison instance, or NULL on failure.
2759  */
2760 struct prison *
2761 prison_find(int prid)
2762 {
2763 	struct prison *pr;
2764 
2765 	sx_assert(&allprison_lock, SX_LOCKED);
2766 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2767 		if (pr->pr_id < prid)
2768 			continue;
2769 		if (pr->pr_id > prid)
2770 			break;
2771 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2772 		mtx_lock(&pr->pr_mtx);
2773 		return (pr);
2774 	}
2775 	return (NULL);
2776 }
2777 
2778 /*
2779  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2780  */
2781 struct prison *
2782 prison_find_child(struct prison *mypr, int prid)
2783 {
2784 	struct prison *pr;
2785 	int descend;
2786 
2787 	sx_assert(&allprison_lock, SX_LOCKED);
2788 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2789 		if (pr->pr_id == prid) {
2790 			KASSERT(prison_isvalid(pr),
2791 			    ("Found invalid prison %p", pr));
2792 			mtx_lock(&pr->pr_mtx);
2793 			return (pr);
2794 		}
2795 	}
2796 	return (NULL);
2797 }
2798 
2799 /*
2800  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2801  */
2802 struct prison *
2803 prison_find_name(struct prison *mypr, const char *name)
2804 {
2805 	struct prison *pr, *deadpr;
2806 	size_t mylen;
2807 	int descend;
2808 
2809 	sx_assert(&allprison_lock, SX_LOCKED);
2810 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2811 	deadpr = NULL;
2812 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2813 		if (!strcmp(pr->pr_name + mylen, name)) {
2814 			KASSERT(prison_isvalid(pr),
2815 			    ("Found invalid prison %p", pr));
2816 			if (prison_isalive(pr)) {
2817 				mtx_lock(&pr->pr_mtx);
2818 				return (pr);
2819 			}
2820 			deadpr = pr;
2821 		}
2822 	}
2823 	/* There was no valid prison - perhaps there was a dying one. */
2824 	if (deadpr != NULL)
2825 		mtx_lock(&deadpr->pr_mtx);
2826 	return (deadpr);
2827 }
2828 
2829 /*
2830  * See if a prison has the specific flag set.  The prison should be locked,
2831  * unless checking for flags that are only set at jail creation (such as
2832  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2833  * to any other prison data.
2834  */
2835 bool
2836 prison_flag(struct ucred *cred, unsigned flag)
2837 {
2838 
2839 	return ((cred->cr_prison->pr_flags & flag) != 0);
2840 }
2841 
2842 /*
2843  * See if a prison has the specific allow flag set.
2844  * The prison *should* be locked, or only a single bit is examined, without
2845  * regard to any other prison data.
2846  */
2847 bool
2848 prison_allow(struct ucred *cred, unsigned flag)
2849 {
2850 
2851 	return ((cred->cr_prison->pr_allow & flag) != 0);
2852 }
2853 
2854 /*
2855  * Hold a prison reference, by incrementing pr_ref.  It is generally
2856  * an error to hold a prison that does not already have a reference.
2857  * A prison record will remain valid as long as it has at least one
2858  * reference, and will not be removed as long as either the prison
2859  * mutex or the allprison lock is held (allprison_lock may be shared).
2860  */
2861 void
2862 prison_hold_locked(struct prison *pr)
2863 {
2864 
2865 	/* Locking is no longer required. */
2866 	prison_hold(pr);
2867 }
2868 
2869 void
2870 prison_hold(struct prison *pr)
2871 {
2872 #ifdef INVARIANTS
2873 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2874 
2875 	KASSERT(was_valid,
2876 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2877 #else
2878 	refcount_acquire(&pr->pr_ref);
2879 #endif
2880 }
2881 
2882 /*
2883  * Remove a prison reference.  If that was the last reference, the
2884  * prison will be removed (at a later time).
2885  */
2886 void
2887 prison_free_locked(struct prison *pr)
2888 {
2889 
2890 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2891 	/*
2892 	 * Locking is no longer required, but unlock because the caller
2893 	 * expects it.
2894 	 */
2895 	mtx_unlock(&pr->pr_mtx);
2896 	prison_free(pr);
2897 }
2898 
2899 void
2900 prison_free(struct prison *pr)
2901 {
2902 
2903 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2904 	    ("Trying to free dead prison %p (jid=%d).",
2905 	     pr, pr->pr_id));
2906 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2907 		/*
2908 		 * Don't remove the last reference in this context,
2909 		 * in case there are locks held.
2910 		 */
2911 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
2912 	}
2913 }
2914 
2915 static void
2916 prison_free_not_last(struct prison *pr)
2917 {
2918 #ifdef INVARIANTS
2919 	int lastref;
2920 
2921 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2922 	    ("Trying to free dead prison %p (jid=%d).",
2923 	     pr, pr->pr_id));
2924 	lastref = refcount_release(&pr->pr_ref);
2925 	KASSERT(!lastref,
2926 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2927 	     pr, pr->pr_id));
2928 #else
2929 	refcount_release(&pr->pr_ref);
2930 #endif
2931 }
2932 
2933 /*
2934  * Hold a prison for user visibility, by incrementing pr_uref.
2935  * It is generally an error to hold a prison that isn't already
2936  * user-visible, except through the jail system calls.  It is also
2937  * an error to hold an invalid prison.  A prison record will remain
2938  * alive as long as it has at least one user reference, and will not
2939  * be set to the dying state until the prison mutex and allprison_lock
2940  * are both freed.
2941  */
2942 void
2943 prison_proc_hold(struct prison *pr)
2944 {
2945 #ifdef INVARIANTS
2946 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2947 
2948 	KASSERT(was_alive,
2949 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2950 #else
2951 	refcount_acquire(&pr->pr_uref);
2952 #endif
2953 }
2954 
2955 /*
2956  * Remove a prison user reference.  If it was the last reference, the
2957  * prison will be considered "dying", and may be removed once all of
2958  * its references are dropped.
2959  */
2960 void
2961 prison_proc_free(struct prison *pr)
2962 {
2963 
2964 	/*
2965 	 * Locking is only required when releasing the last reference.
2966 	 * This allows assurance that a locked prison will remain alive
2967 	 * until it is unlocked.
2968 	 */
2969 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2970 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2971 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2972 		/*
2973 		 * Don't remove the last user reference in this context,
2974 		 * which is expected to be a process that is not only locked,
2975 		 * but also half dead.  Add a reference so any calls to
2976 		 * prison_free() won't re-submit the task.
2977 		 */
2978 		prison_hold(pr);
2979 		mtx_lock(&pr->pr_mtx);
2980 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2981 		    ("Redundant last reference in prison_proc_free (jid=%d)",
2982 		     pr->pr_id));
2983 		pr->pr_flags |= PR_COMPLETE_PROC;
2984 		mtx_unlock(&pr->pr_mtx);
2985 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
2986 	}
2987 }
2988 
2989 static void
2990 prison_proc_free_not_last(struct prison *pr)
2991 {
2992 #ifdef INVARIANTS
2993 	int lastref;
2994 
2995 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2996 	    ("Trying to free dead prison %p (jid=%d).",
2997 	     pr, pr->pr_id));
2998 	lastref = refcount_release(&pr->pr_uref);
2999 	KASSERT(!lastref,
3000 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3001 	     pr, pr->pr_id));
3002 #else
3003 	refcount_release(&pr->pr_uref);
3004 #endif
3005 }
3006 
3007 void
3008 prison_proc_link(struct prison *pr, struct proc *p)
3009 {
3010 
3011 	sx_assert(&allproc_lock, SA_XLOCKED);
3012 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3013 }
3014 
3015 void
3016 prison_proc_unlink(struct prison *pr, struct proc *p)
3017 {
3018 
3019 	sx_assert(&allproc_lock, SA_XLOCKED);
3020 	LIST_REMOVE(p, p_jaillist);
3021 }
3022 
3023 static void
3024 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3025 {
3026 
3027 	sx_xlock(&allproc_lock);
3028 	prison_proc_unlink(opr, p);
3029 	prison_proc_link(npr, p);
3030 	sx_xunlock(&allproc_lock);
3031 }
3032 
3033 /*
3034  * Complete a call to either prison_free or prison_proc_free.
3035  */
3036 static void
3037 prison_complete(void *context, int pending)
3038 {
3039 	struct prison *pr = context;
3040 	int drflags;
3041 
3042 	/*
3043 	 * This could be called to release the last reference, or the last
3044 	 * user reference (plus the reference held in prison_proc_free).
3045 	 */
3046 	drflags = prison_lock_xlock(pr, PD_DEREF);
3047 	if (pr->pr_flags & PR_COMPLETE_PROC) {
3048 		pr->pr_flags &= ~PR_COMPLETE_PROC;
3049 		drflags |= PD_DEUREF;
3050 	}
3051 	prison_deref(pr, drflags);
3052 }
3053 
3054 static void
3055 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3056 {
3057 
3058 	kern_psignal(p, SIGKILL);
3059 }
3060 
3061 /*
3062  * Note the iteration does not guarantee acting on all processes.
3063  * Most notably there may be fork or jail_attach in progress.
3064  */
3065 void
3066 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3067     void *cbarg)
3068 {
3069 	struct prison *ppr;
3070 	struct proc *p;
3071 
3072 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3073 		sx_slock(&allproc_lock);
3074 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3075 			if (p->p_state == PRS_NEW)
3076 				continue;
3077 			PROC_LOCK(p);
3078 			cb(p, cbarg);
3079 			PROC_UNLOCK(p);
3080 		}
3081 		sx_sunlock(&allproc_lock);
3082 		if (atomic_load_int(&pr->pr_childcount) == 0)
3083 			return;
3084 		/*
3085 		 * Some jails popped up during the iteration, fall through to a
3086 		 * system-wide search.
3087 		 */
3088 	}
3089 
3090 	sx_slock(&allproc_lock);
3091 	FOREACH_PROC_IN_SYSTEM(p) {
3092 		PROC_LOCK(p);
3093 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3094 			for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3095 			    ppr = ppr->pr_parent) {
3096 				if (ppr == pr) {
3097 					cb(p, cbarg);
3098 					break;
3099 				}
3100 			}
3101 		}
3102 		PROC_UNLOCK(p);
3103 	}
3104 	sx_sunlock(&allproc_lock);
3105 }
3106 
3107 /*
3108  * Remove a prison reference and/or user reference (usually).
3109  * This assumes context that allows sleeping (for allprison_lock),
3110  * with no non-sleeping locks held, except perhaps the prison itself.
3111  * If there are no more references, release and delist the prison.
3112  * On completion, the prison lock and the allprison lock are both
3113  * unlocked.
3114  */
3115 static void
3116 prison_deref(struct prison *pr, int flags)
3117 {
3118 	struct prisonlist freeprison;
3119 	struct prison *killpr, *rpr, *ppr, *tpr;
3120 
3121 	killpr = NULL;
3122 	TAILQ_INIT(&freeprison);
3123 	/*
3124 	 * Release this prison as requested, which may cause its parent
3125 	 * to be released, and then maybe its grandparent, etc.
3126 	 */
3127 	for (;;) {
3128 		if (flags & PD_KILL) {
3129 			/* Kill the prison and its descendents. */
3130 			KASSERT(pr != &prison0,
3131 			    ("prison_deref trying to kill prison0"));
3132 			if (!(flags & PD_DEREF)) {
3133 				prison_hold(pr);
3134 				flags |= PD_DEREF;
3135 			}
3136 			flags = prison_lock_xlock(pr, flags);
3137 			prison_deref_kill(pr, &freeprison);
3138 		}
3139 		if (flags & PD_DEUREF) {
3140 			/* Drop a user reference. */
3141 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3142 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3143 			     pr->pr_id));
3144 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3145 				if (!(flags & PD_DEREF)) {
3146 					prison_hold(pr);
3147 					flags |= PD_DEREF;
3148 				}
3149 				flags = prison_lock_xlock(pr, flags);
3150 				if (refcount_release(&pr->pr_uref) &&
3151 				    pr->pr_state == PRISON_STATE_ALIVE) {
3152 					/*
3153 					 * When the last user references goes,
3154 					 * this becomes a dying prison.
3155 					 */
3156 					KASSERT(
3157 					    refcount_load(&prison0.pr_uref) > 0,
3158 					    ("prison0 pr_uref=0"));
3159 					pr->pr_state = PRISON_STATE_DYING;
3160 					mtx_unlock(&pr->pr_mtx);
3161 					flags &= ~PD_LOCKED;
3162 					prison_cleanup(pr);
3163 				}
3164 			}
3165 		}
3166 		if (flags & PD_KILL) {
3167 			/*
3168 			 * Any remaining user references are probably processes
3169 			 * that need to be killed, either in this prison or its
3170 			 * descendants.
3171 			 */
3172 			if (refcount_load(&pr->pr_uref) > 0)
3173 				killpr = pr;
3174 			/* Make sure the parent prison doesn't get killed. */
3175 			flags &= ~PD_KILL;
3176 		}
3177 		if (flags & PD_DEREF) {
3178 			/* Drop a reference. */
3179 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3180 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3181 			     pr->pr_id));
3182 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3183 				flags = prison_lock_xlock(pr, flags);
3184 				if (refcount_release(&pr->pr_ref)) {
3185 					/*
3186 					 * When the last reference goes,
3187 					 * unlink the prison and set it aside.
3188 					 */
3189 					KASSERT(
3190 					    refcount_load(&pr->pr_uref) == 0,
3191 					    ("prison_deref: last ref, "
3192 					     "but still has %d urefs (jid=%d)",
3193 					     pr->pr_uref, pr->pr_id));
3194 					KASSERT(
3195 					    refcount_load(&prison0.pr_ref) != 0,
3196 					    ("prison0 pr_ref=0"));
3197 					pr->pr_state = PRISON_STATE_INVALID;
3198 					TAILQ_REMOVE(&allprison, pr, pr_list);
3199 					LIST_REMOVE(pr, pr_sibling);
3200 					TAILQ_INSERT_TAIL(&freeprison, pr,
3201 					    pr_list);
3202 					for (ppr = pr->pr_parent;
3203 					     ppr != NULL;
3204 					     ppr = ppr->pr_parent)
3205 						ppr->pr_childcount--;
3206 					/*
3207 					 * Removing a prison frees references
3208 					 * from its parent.
3209 					 */
3210 					mtx_unlock(&pr->pr_mtx);
3211 					flags &= ~PD_LOCKED;
3212 					pr = pr->pr_parent;
3213 					flags |= PD_DEREF | PD_DEUREF;
3214 					continue;
3215 				}
3216 			}
3217 		}
3218 		break;
3219 	}
3220 
3221 	/* Release all the prison locks. */
3222 	if (flags & PD_LOCKED)
3223 		mtx_unlock(&pr->pr_mtx);
3224 	if (flags & PD_LIST_SLOCKED)
3225 		sx_sunlock(&allprison_lock);
3226 	else if (flags & PD_LIST_XLOCKED)
3227 		sx_xunlock(&allprison_lock);
3228 
3229 	/* Kill any processes attached to a killed prison. */
3230 	if (killpr != NULL)
3231 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3232 
3233 	/*
3234 	 * Finish removing any unreferenced prisons, which couldn't happen
3235 	 * while allprison_lock was held (to avoid a LOR on vrele).
3236 	 */
3237 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3238 #ifdef VIMAGE
3239 		if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3240 			vnet_destroy(rpr->pr_vnet);
3241 #endif
3242 		if (rpr->pr_root != NULL)
3243 			vrele(rpr->pr_root);
3244 		mtx_destroy(&rpr->pr_mtx);
3245 #ifdef INET
3246 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3247 #endif
3248 #ifdef INET6
3249 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3250 #endif
3251 		if (rpr->pr_cpuset != NULL)
3252 			cpuset_rel(rpr->pr_cpuset);
3253 		osd_jail_exit(rpr);
3254 #ifdef RACCT
3255 		if (racct_enable)
3256 			prison_racct_detach(rpr);
3257 #endif
3258 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3259 		free(rpr, M_PRISON);
3260 	}
3261 }
3262 
3263 /*
3264  * Kill the prison and its descendants.  Mark them as dying, clear the
3265  * persist flag, and call module remove methods.
3266  */
3267 static void
3268 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3269 {
3270 	struct prison *cpr, *ppr, *rpr;
3271 	bool descend;
3272 
3273 	/*
3274 	 * Unlike the descendants, the target prison can be killed
3275 	 * even if it is currently dying.  This is useful for failed
3276 	 * creation in jail_set(2).
3277 	 */
3278 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3279 	    ("Trying to kill dead prison %p (jid=%d).",
3280 	     pr, pr->pr_id));
3281 	refcount_acquire(&pr->pr_uref);
3282 	pr->pr_state = PRISON_STATE_DYING;
3283 	mtx_unlock(&pr->pr_mtx);
3284 
3285 	rpr = NULL;
3286 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3287 		if (descend) {
3288 			if (!prison_isalive(cpr)) {
3289 				descend = false;
3290 				continue;
3291 			}
3292 			prison_hold(cpr);
3293 			prison_proc_hold(cpr);
3294 			mtx_lock(&cpr->pr_mtx);
3295 			cpr->pr_state = PRISON_STATE_DYING;
3296 			cpr->pr_flags |= PR_REMOVE;
3297 			mtx_unlock(&cpr->pr_mtx);
3298 			continue;
3299 		}
3300 		if (!(cpr->pr_flags & PR_REMOVE))
3301 			continue;
3302 		prison_cleanup(cpr);
3303 		mtx_lock(&cpr->pr_mtx);
3304 		cpr->pr_flags &= ~PR_REMOVE;
3305 		if (cpr->pr_flags & PR_PERSIST) {
3306 			cpr->pr_flags &= ~PR_PERSIST;
3307 			prison_proc_free_not_last(cpr);
3308 			prison_free_not_last(cpr);
3309 		}
3310 		(void)refcount_release(&cpr->pr_uref);
3311 		if (refcount_release(&cpr->pr_ref)) {
3312 			/*
3313 			 * When the last reference goes, unlink the prison
3314 			 * and set it aside for prison_deref() to handle.
3315 			 * Delay unlinking the sibling list to keep the loop
3316 			 * safe.
3317 			 */
3318 			if (rpr != NULL)
3319 				LIST_REMOVE(rpr, pr_sibling);
3320 			rpr = cpr;
3321 			rpr->pr_state = PRISON_STATE_INVALID;
3322 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3323 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3324 			/*
3325 			 * Removing a prison frees references from its parent.
3326 			 */
3327 			ppr = rpr->pr_parent;
3328 			prison_proc_free_not_last(ppr);
3329 			prison_free_not_last(ppr);
3330 			for (; ppr != NULL; ppr = ppr->pr_parent)
3331 				ppr->pr_childcount--;
3332 		}
3333 		mtx_unlock(&cpr->pr_mtx);
3334 	}
3335 	if (rpr != NULL)
3336 		LIST_REMOVE(rpr, pr_sibling);
3337 
3338 	prison_cleanup(pr);
3339 	mtx_lock(&pr->pr_mtx);
3340 	if (pr->pr_flags & PR_PERSIST) {
3341 		pr->pr_flags &= ~PR_PERSIST;
3342 		prison_proc_free_not_last(pr);
3343 		prison_free_not_last(pr);
3344 	}
3345 	(void)refcount_release(&pr->pr_uref);
3346 }
3347 
3348 /*
3349  * Given the current locking state in the flags, make sure allprison_lock
3350  * is held exclusive, and the prison is locked.  Return flags indicating
3351  * the new state.
3352  */
3353 static int
3354 prison_lock_xlock(struct prison *pr, int flags)
3355 {
3356 
3357 	if (!(flags & PD_LIST_XLOCKED)) {
3358 		/*
3359 		 * Get allprison_lock, which may be an upgrade,
3360 		 * and may require unlocking the prison.
3361 		 */
3362 		if (flags & PD_LOCKED) {
3363 			mtx_unlock(&pr->pr_mtx);
3364 			flags &= ~PD_LOCKED;
3365 		}
3366 		if (flags & PD_LIST_SLOCKED) {
3367 			if (!sx_try_upgrade(&allprison_lock)) {
3368 				sx_sunlock(&allprison_lock);
3369 				sx_xlock(&allprison_lock);
3370 			}
3371 			flags &= ~PD_LIST_SLOCKED;
3372 		} else
3373 			sx_xlock(&allprison_lock);
3374 		flags |= PD_LIST_XLOCKED;
3375 	}
3376 	if (!(flags & PD_LOCKED)) {
3377 		/* Lock the prison mutex. */
3378 		mtx_lock(&pr->pr_mtx);
3379 		flags |= PD_LOCKED;
3380 	}
3381 	return flags;
3382 }
3383 
3384 /*
3385  * Release a prison's resources when it starts dying (when the last user
3386  * reference is dropped, or when it is killed).
3387  */
3388 static void
3389 prison_cleanup(struct prison *pr)
3390 {
3391 	sx_assert(&allprison_lock, SA_XLOCKED);
3392 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3393 	vfs_exjail_delete(pr);
3394 	shm_remove_prison(pr);
3395 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3396 }
3397 
3398 /*
3399  * Set or clear a permission bit in the pr_allow field, passing restrictions
3400  * (cleared permission) down to child jails.
3401  */
3402 void
3403 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3404 {
3405 	struct prison *pr;
3406 
3407 	pr = cred->cr_prison;
3408 	sx_slock(&allprison_lock);
3409 	mtx_lock(&pr->pr_mtx);
3410 	prison_set_allow_locked(pr, flag, enable);
3411 	mtx_unlock(&pr->pr_mtx);
3412 	sx_sunlock(&allprison_lock);
3413 }
3414 
3415 static void
3416 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3417 {
3418 	struct prison *cpr;
3419 	int descend;
3420 
3421 	if (enable != 0)
3422 		pr->pr_allow |= flag;
3423 	else {
3424 		pr->pr_allow &= ~flag;
3425 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3426 			cpr->pr_allow &= ~flag;
3427 	}
3428 }
3429 
3430 /*
3431  * Check if a jail supports the given address family.
3432  *
3433  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3434  * if not.
3435  */
3436 int
3437 prison_check_af(struct ucred *cred, int af)
3438 {
3439 	struct prison *pr;
3440 	int error;
3441 
3442 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3443 
3444 	pr = cred->cr_prison;
3445 #ifdef VIMAGE
3446 	/* Prisons with their own network stack are not limited. */
3447 	if (prison_owns_vnet(cred))
3448 		return (0);
3449 #endif
3450 
3451 	error = 0;
3452 	switch (af)
3453 	{
3454 #ifdef INET
3455 	case AF_INET:
3456 		if (pr->pr_flags & PR_IP4)
3457 		{
3458 			mtx_lock(&pr->pr_mtx);
3459 			if ((pr->pr_flags & PR_IP4) &&
3460 			    pr->pr_addrs[PR_INET] == NULL)
3461 				error = EAFNOSUPPORT;
3462 			mtx_unlock(&pr->pr_mtx);
3463 		}
3464 		break;
3465 #endif
3466 #ifdef INET6
3467 	case AF_INET6:
3468 		if (pr->pr_flags & PR_IP6)
3469 		{
3470 			mtx_lock(&pr->pr_mtx);
3471 			if ((pr->pr_flags & PR_IP6) &&
3472 			    pr->pr_addrs[PR_INET6] == NULL)
3473 				error = EAFNOSUPPORT;
3474 			mtx_unlock(&pr->pr_mtx);
3475 		}
3476 		break;
3477 #endif
3478 	case AF_LOCAL:
3479 	case AF_ROUTE:
3480 	case AF_NETLINK:
3481 		break;
3482 	default:
3483 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3484 			error = EAFNOSUPPORT;
3485 	}
3486 	return (error);
3487 }
3488 
3489 /*
3490  * Check if given address belongs to the jail referenced by cred (wrapper to
3491  * prison_check_ip[46]).
3492  *
3493  * Returns 0 if jail doesn't restrict the address family or if address belongs
3494  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3495  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3496  */
3497 int
3498 prison_if(struct ucred *cred, const struct sockaddr *sa)
3499 {
3500 #ifdef INET
3501 	const struct sockaddr_in *sai;
3502 #endif
3503 #ifdef INET6
3504 	const struct sockaddr_in6 *sai6;
3505 #endif
3506 	int error;
3507 
3508 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3509 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3510 
3511 #ifdef VIMAGE
3512 	if (prison_owns_vnet(cred))
3513 		return (0);
3514 #endif
3515 
3516 	error = 0;
3517 	switch (sa->sa_family)
3518 	{
3519 #ifdef INET
3520 	case AF_INET:
3521 		sai = (const struct sockaddr_in *)sa;
3522 		error = prison_check_ip4(cred, &sai->sin_addr);
3523 		break;
3524 #endif
3525 #ifdef INET6
3526 	case AF_INET6:
3527 		sai6 = (const struct sockaddr_in6 *)sa;
3528 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3529 		break;
3530 #endif
3531 	default:
3532 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3533 			error = EAFNOSUPPORT;
3534 	}
3535 	return (error);
3536 }
3537 
3538 /*
3539  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3540  */
3541 int
3542 prison_check(struct ucred *cred1, struct ucred *cred2)
3543 {
3544 
3545 	return ((cred1->cr_prison == cred2->cr_prison ||
3546 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3547 }
3548 
3549 /*
3550  * For mountd/nfsd to run within a prison, it must be:
3551  * - A vnet prison.
3552  * - PR_ALLOW_NFSD must be set on it.
3553  * - The root directory (pr_root) of the prison must be
3554  *   a file system mount point, so the mountd can hang
3555  *   export information on it.
3556  * - The prison's enforce_statfs cannot be 0, so that
3557  *   mountd(8) can do exports.
3558  */
3559 bool
3560 prison_check_nfsd(struct ucred *cred)
3561 {
3562 
3563 	if (jailed_without_vnet(cred))
3564 		return (false);
3565 	if (!prison_allow(cred, PR_ALLOW_NFSD))
3566 		return (false);
3567 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3568 		return (false);
3569 	if (cred->cr_prison->pr_enforce_statfs == 0)
3570 		return (false);
3571 	return (true);
3572 }
3573 
3574 /*
3575  * Return true if p2 is a child of p1, otherwise false.
3576  */
3577 bool
3578 prison_ischild(struct prison *pr1, struct prison *pr2)
3579 {
3580 
3581 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3582 		if (pr1 == pr2)
3583 			return (true);
3584 	return (false);
3585 }
3586 
3587 /*
3588  * Return true if the prison is currently alive.  A prison is alive if it
3589  * holds user references and it isn't being removed.
3590  */
3591 bool
3592 prison_isalive(const struct prison *pr)
3593 {
3594 
3595 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3596 		return (false);
3597 	return (true);
3598 }
3599 
3600 /*
3601  * Return true if the prison is currently valid.  A prison is valid if it has
3602  * been fully created, and is not being destroyed.  Note that dying prisons
3603  * are still considered valid.  Invalid prisons won't be found under normal
3604  * circumstances, as they're only put in that state by functions that have
3605  * an exclusive hold on allprison_lock.
3606  */
3607 bool
3608 prison_isvalid(struct prison *pr)
3609 {
3610 
3611 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3612 		return (false);
3613 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3614 		return (false);
3615 	return (true);
3616 }
3617 
3618 /*
3619  * Return true if the passed credential is in a jail and that jail does not
3620  * have its own virtual network stack, otherwise false.
3621  */
3622 bool
3623 jailed_without_vnet(struct ucred *cred)
3624 {
3625 
3626 	if (!jailed(cred))
3627 		return (false);
3628 #ifdef VIMAGE
3629 	if (prison_owns_vnet(cred))
3630 		return (false);
3631 #endif
3632 
3633 	return (true);
3634 }
3635 
3636 /*
3637  * Return the correct hostname (domainname, et al) for the passed credential.
3638  */
3639 void
3640 getcredhostname(struct ucred *cred, char *buf, size_t size)
3641 {
3642 	struct prison *pr;
3643 
3644 	/*
3645 	 * A NULL credential can be used to shortcut to the physical
3646 	 * system's hostname.
3647 	 */
3648 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3649 	mtx_lock(&pr->pr_mtx);
3650 	strlcpy(buf, pr->pr_hostname, size);
3651 	mtx_unlock(&pr->pr_mtx);
3652 }
3653 
3654 void
3655 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3656 {
3657 
3658 	mtx_lock(&cred->cr_prison->pr_mtx);
3659 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3660 	mtx_unlock(&cred->cr_prison->pr_mtx);
3661 }
3662 
3663 void
3664 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3665 {
3666 
3667 	mtx_lock(&cred->cr_prison->pr_mtx);
3668 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3669 	mtx_unlock(&cred->cr_prison->pr_mtx);
3670 }
3671 
3672 void
3673 getcredhostid(struct ucred *cred, unsigned long *hostid)
3674 {
3675 
3676 	mtx_lock(&cred->cr_prison->pr_mtx);
3677 	*hostid = cred->cr_prison->pr_hostid;
3678 	mtx_unlock(&cred->cr_prison->pr_mtx);
3679 }
3680 
3681 void
3682 getjailname(struct ucred *cred, char *name, size_t len)
3683 {
3684 
3685 	mtx_lock(&cred->cr_prison->pr_mtx);
3686 	strlcpy(name, cred->cr_prison->pr_name, len);
3687 	mtx_unlock(&cred->cr_prison->pr_mtx);
3688 }
3689 
3690 #ifdef VIMAGE
3691 /*
3692  * Determine whether the prison represented by cred owns
3693  * its vnet rather than having it inherited.
3694  *
3695  * Returns true in case the prison owns the vnet, false otherwise.
3696  */
3697 bool
3698 prison_owns_vnet(struct ucred *cred)
3699 {
3700 
3701 	/*
3702 	 * vnets cannot be added/removed after jail creation,
3703 	 * so no need to lock here.
3704 	 */
3705 	return ((cred->cr_prison->pr_flags & PR_VNET) != 0);
3706 }
3707 #endif
3708 
3709 /*
3710  * Determine whether the subject represented by cred can "see"
3711  * status of a mount point.
3712  * Returns: 0 for permitted, ENOENT otherwise.
3713  * XXX: This function should be called cr_canseemount() and should be
3714  *      placed in kern_prot.c.
3715  */
3716 int
3717 prison_canseemount(struct ucred *cred, struct mount *mp)
3718 {
3719 	struct prison *pr;
3720 	struct statfs *sp;
3721 	size_t len;
3722 
3723 	pr = cred->cr_prison;
3724 	if (pr->pr_enforce_statfs == 0)
3725 		return (0);
3726 	if (pr->pr_root->v_mount == mp)
3727 		return (0);
3728 	if (pr->pr_enforce_statfs == 2)
3729 		return (ENOENT);
3730 	/*
3731 	 * If jail's chroot directory is set to "/" we should be able to see
3732 	 * all mount-points from inside a jail.
3733 	 * This is ugly check, but this is the only situation when jail's
3734 	 * directory ends with '/'.
3735 	 */
3736 	if (strcmp(pr->pr_path, "/") == 0)
3737 		return (0);
3738 	len = strlen(pr->pr_path);
3739 	sp = &mp->mnt_stat;
3740 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3741 		return (ENOENT);
3742 	/*
3743 	 * Be sure that we don't have situation where jail's root directory
3744 	 * is "/some/path" and mount point is "/some/pathpath".
3745 	 */
3746 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3747 		return (ENOENT);
3748 	return (0);
3749 }
3750 
3751 void
3752 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3753 {
3754 	char jpath[MAXPATHLEN];
3755 	struct prison *pr;
3756 	size_t len;
3757 
3758 	pr = cred->cr_prison;
3759 	if (pr->pr_enforce_statfs == 0)
3760 		return;
3761 	if (prison_canseemount(cred, mp) != 0) {
3762 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3763 		strlcpy(sp->f_mntonname, "[restricted]",
3764 		    sizeof(sp->f_mntonname));
3765 		return;
3766 	}
3767 	if (pr->pr_root->v_mount == mp) {
3768 		/*
3769 		 * Clear current buffer data, so we are sure nothing from
3770 		 * the valid path left there.
3771 		 */
3772 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3773 		*sp->f_mntonname = '/';
3774 		return;
3775 	}
3776 	/*
3777 	 * If jail's chroot directory is set to "/" we should be able to see
3778 	 * all mount-points from inside a jail.
3779 	 */
3780 	if (strcmp(pr->pr_path, "/") == 0)
3781 		return;
3782 	len = strlen(pr->pr_path);
3783 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3784 	/*
3785 	 * Clear current buffer data, so we are sure nothing from
3786 	 * the valid path left there.
3787 	 */
3788 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3789 	if (*jpath == '\0') {
3790 		/* Should never happen. */
3791 		*sp->f_mntonname = '/';
3792 	} else {
3793 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3794 	}
3795 }
3796 
3797 /*
3798  * Check with permission for a specific privilege is granted within jail.  We
3799  * have a specific list of accepted privileges; the rest are denied.
3800  */
3801 int
3802 prison_priv_check(struct ucred *cred, int priv)
3803 {
3804 	struct prison *pr;
3805 	int error;
3806 
3807 	/*
3808 	 * Some policies have custom handlers. This routine should not be
3809 	 * called for them. See priv_check_cred().
3810 	 */
3811 	switch (priv) {
3812 	case PRIV_VFS_LOOKUP:
3813 	case PRIV_VFS_GENERATION:
3814 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3815 		    "called for %d\n", priv));
3816 	}
3817 
3818 	if (!jailed(cred))
3819 		return (0);
3820 
3821 #ifdef VIMAGE
3822 	/*
3823 	 * Privileges specific to prisons with a virtual network stack.
3824 	 * There might be a duplicate entry here in case the privilege
3825 	 * is only granted conditionally in the legacy jail case.
3826 	 */
3827 	switch (priv) {
3828 		/*
3829 		 * NFS-specific privileges.
3830 		 */
3831 	case PRIV_NFS_DAEMON:
3832 	case PRIV_VFS_GETFH:
3833 	case PRIV_VFS_MOUNT_EXPORTED:
3834 		if (!prison_check_nfsd(cred))
3835 			return (EPERM);
3836 #ifdef notyet
3837 	case PRIV_NFS_LOCKD:
3838 #endif
3839 		/*
3840 		 * Network stack privileges.
3841 		 */
3842 	case PRIV_NET_BRIDGE:
3843 	case PRIV_NET_GRE:
3844 	case PRIV_NET_BPF:
3845 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3846 	case PRIV_NET_ROUTE:
3847 	case PRIV_NET_TAP:
3848 	case PRIV_NET_SETIFMTU:
3849 	case PRIV_NET_SETIFFLAGS:
3850 	case PRIV_NET_SETIFCAP:
3851 	case PRIV_NET_SETIFDESCR:
3852 	case PRIV_NET_SETIFNAME	:
3853 	case PRIV_NET_SETIFMETRIC:
3854 	case PRIV_NET_SETIFPHYS:
3855 	case PRIV_NET_SETIFMAC:
3856 	case PRIV_NET_SETLANPCP:
3857 	case PRIV_NET_ADDMULTI:
3858 	case PRIV_NET_DELMULTI:
3859 	case PRIV_NET_HWIOCTL:
3860 	case PRIV_NET_SETLLADDR:
3861 	case PRIV_NET_ADDIFGROUP:
3862 	case PRIV_NET_DELIFGROUP:
3863 	case PRIV_NET_IFCREATE:
3864 	case PRIV_NET_IFDESTROY:
3865 	case PRIV_NET_ADDIFADDR:
3866 	case PRIV_NET_DELIFADDR:
3867 	case PRIV_NET_LAGG:
3868 	case PRIV_NET_GIF:
3869 	case PRIV_NET_SETIFVNET:
3870 	case PRIV_NET_SETIFFIB:
3871 	case PRIV_NET_OVPN:
3872 	case PRIV_NET_ME:
3873 	case PRIV_NET_WG:
3874 
3875 		/*
3876 		 * 802.11-related privileges.
3877 		 */
3878 	case PRIV_NET80211_VAP_GETKEY:
3879 	case PRIV_NET80211_VAP_MANAGE:
3880 
3881 #ifdef notyet
3882 		/*
3883 		 * ATM privileges.
3884 		 */
3885 	case PRIV_NETATM_CFG:
3886 	case PRIV_NETATM_ADD:
3887 	case PRIV_NETATM_DEL:
3888 	case PRIV_NETATM_SET:
3889 
3890 		/*
3891 		 * Bluetooth privileges.
3892 		 */
3893 	case PRIV_NETBLUETOOTH_RAW:
3894 #endif
3895 
3896 		/*
3897 		 * Netgraph and netgraph module privileges.
3898 		 */
3899 	case PRIV_NETGRAPH_CONTROL:
3900 #ifdef notyet
3901 	case PRIV_NETGRAPH_TTY:
3902 #endif
3903 
3904 		/*
3905 		 * IPv4 and IPv6 privileges.
3906 		 */
3907 	case PRIV_NETINET_IPFW:
3908 	case PRIV_NETINET_DIVERT:
3909 	case PRIV_NETINET_PF:
3910 	case PRIV_NETINET_DUMMYNET:
3911 	case PRIV_NETINET_CARP:
3912 	case PRIV_NETINET_MROUTE:
3913 	case PRIV_NETINET_RAW:
3914 	case PRIV_NETINET_ADDRCTRL6:
3915 	case PRIV_NETINET_ND6:
3916 	case PRIV_NETINET_SCOPE6:
3917 	case PRIV_NETINET_ALIFETIME6:
3918 	case PRIV_NETINET_IPSEC:
3919 	case PRIV_NETINET_BINDANY:
3920 
3921 #ifdef notyet
3922 		/*
3923 		 * NCP privileges.
3924 		 */
3925 	case PRIV_NETNCP:
3926 
3927 		/*
3928 		 * SMB privileges.
3929 		 */
3930 	case PRIV_NETSMB:
3931 #endif
3932 
3933 	/*
3934 	 * No default: or deny here.
3935 	 * In case of no permit fall through to next switch().
3936 	 */
3937 		if (cred->cr_prison->pr_flags & PR_VNET)
3938 			return (0);
3939 	}
3940 #endif /* VIMAGE */
3941 
3942 	switch (priv) {
3943 		/*
3944 		 * Allow ktrace privileges for root in jail.
3945 		 */
3946 	case PRIV_KTRACE:
3947 
3948 #if 0
3949 		/*
3950 		 * Allow jailed processes to configure audit identity and
3951 		 * submit audit records (login, etc).  In the future we may
3952 		 * want to further refine the relationship between audit and
3953 		 * jail.
3954 		 */
3955 	case PRIV_AUDIT_GETAUDIT:
3956 	case PRIV_AUDIT_SETAUDIT:
3957 	case PRIV_AUDIT_SUBMIT:
3958 #endif
3959 
3960 		/*
3961 		 * Allow jailed processes to manipulate process UNIX
3962 		 * credentials in any way they see fit.
3963 		 */
3964 	case PRIV_CRED_SETCRED:
3965 	case PRIV_CRED_SETUID:
3966 	case PRIV_CRED_SETEUID:
3967 	case PRIV_CRED_SETGID:
3968 	case PRIV_CRED_SETEGID:
3969 	case PRIV_CRED_SETGROUPS:
3970 	case PRIV_CRED_SETREUID:
3971 	case PRIV_CRED_SETREGID:
3972 	case PRIV_CRED_SETRESUID:
3973 	case PRIV_CRED_SETRESGID:
3974 
3975 		/*
3976 		 * Jail implements visibility constraints already, so allow
3977 		 * jailed root to override uid/gid-based constraints.
3978 		 */
3979 	case PRIV_SEEOTHERGIDS:
3980 	case PRIV_SEEOTHERUIDS:
3981 	case PRIV_SEEJAILPROC:
3982 
3983 		/*
3984 		 * Jail implements inter-process debugging limits already, so
3985 		 * allow jailed root various debugging privileges.
3986 		 */
3987 	case PRIV_DEBUG_DIFFCRED:
3988 	case PRIV_DEBUG_SUGID:
3989 	case PRIV_DEBUG_UNPRIV:
3990 
3991 		/*
3992 		 * Allow jail to set various resource limits and login
3993 		 * properties, and for now, exceed process resource limits.
3994 		 */
3995 	case PRIV_PROC_LIMIT:
3996 	case PRIV_PROC_SETLOGIN:
3997 	case PRIV_PROC_SETRLIMIT:
3998 
3999 		/*
4000 		 * System V and POSIX IPC privileges are granted in jail.
4001 		 */
4002 	case PRIV_IPC_READ:
4003 	case PRIV_IPC_WRITE:
4004 	case PRIV_IPC_ADMIN:
4005 	case PRIV_IPC_MSGSIZE:
4006 	case PRIV_MQ_ADMIN:
4007 
4008 		/*
4009 		 * Jail operations within a jail work on child jails.
4010 		 */
4011 	case PRIV_JAIL_ATTACH:
4012 	case PRIV_JAIL_SET:
4013 	case PRIV_JAIL_REMOVE:
4014 
4015 		/*
4016 		 * Jail implements its own inter-process limits, so allow
4017 		 * root processes in jail to change scheduling on other
4018 		 * processes in the same jail.  Likewise for signalling.
4019 		 */
4020 	case PRIV_SCHED_DIFFCRED:
4021 	case PRIV_SCHED_CPUSET:
4022 	case PRIV_SIGNAL_DIFFCRED:
4023 	case PRIV_SIGNAL_SUGID:
4024 
4025 		/*
4026 		 * Allow jailed processes to write to sysctls marked as jail
4027 		 * writable.
4028 		 */
4029 	case PRIV_SYSCTL_WRITEJAIL:
4030 
4031 		/*
4032 		 * Allow root in jail to manage a variety of quota
4033 		 * properties.  These should likely be conditional on a
4034 		 * configuration option.
4035 		 */
4036 	case PRIV_VFS_GETQUOTA:
4037 	case PRIV_VFS_SETQUOTA:
4038 
4039 		/*
4040 		 * Since Jail relies on chroot() to implement file system
4041 		 * protections, grant many VFS privileges to root in jail.
4042 		 * Be careful to exclude mount-related and NFS-related
4043 		 * privileges.
4044 		 */
4045 	case PRIV_VFS_READ:
4046 	case PRIV_VFS_WRITE:
4047 	case PRIV_VFS_ADMIN:
4048 	case PRIV_VFS_EXEC:
4049 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
4050 	case PRIV_VFS_CHFLAGS_DEV:
4051 	case PRIV_VFS_CHOWN:
4052 	case PRIV_VFS_CHROOT:
4053 	case PRIV_VFS_RETAINSUGID:
4054 	case PRIV_VFS_FCHROOT:
4055 	case PRIV_VFS_LINK:
4056 	case PRIV_VFS_SETGID:
4057 	case PRIV_VFS_STAT:
4058 	case PRIV_VFS_STICKYFILE:
4059 
4060 		/*
4061 		 * As in the non-jail case, non-root users are expected to be
4062 		 * able to read kernel/physical memory (provided /dev/[k]mem
4063 		 * exists in the jail and they have permission to access it).
4064 		 */
4065 	case PRIV_KMEM_READ:
4066 		return (0);
4067 
4068 		/*
4069 		 * Depending on the global setting, allow privilege of
4070 		 * setting system flags.
4071 		 */
4072 	case PRIV_VFS_SYSFLAGS:
4073 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4074 			return (0);
4075 		else
4076 			return (EPERM);
4077 
4078 		/*
4079 		 * Depending on the global setting, allow privilege of
4080 		 * mounting/unmounting file systems.
4081 		 */
4082 	case PRIV_VFS_MOUNT:
4083 	case PRIV_VFS_UNMOUNT:
4084 	case PRIV_VFS_MOUNT_NONUSER:
4085 	case PRIV_VFS_MOUNT_OWNER:
4086 		pr = cred->cr_prison;
4087 		prison_lock(pr);
4088 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4089 			error = 0;
4090 		else
4091 			error = EPERM;
4092 		prison_unlock(pr);
4093 		return (error);
4094 
4095 		/*
4096 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4097 		 * policy.  priv_check_cred will not specifically allow it, and
4098 		 * we may want a MAC policy to allow it.
4099 		 */
4100 	case PRIV_VFS_READ_DIR:
4101 		return (0);
4102 
4103 		/*
4104 		 * Conditionally allow privileged process in the jail to
4105 		 * manipulate filesystem extended attributes in the system
4106 		 * namespace.
4107 		 */
4108 	case PRIV_VFS_EXTATTR_SYSTEM:
4109 		if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4110 			return (0);
4111 		else
4112 			return (EPERM);
4113 
4114 		/*
4115 		 * Conditionnaly allow locking (unlocking) physical pages
4116 		 * in memory.
4117 		 */
4118 	case PRIV_VM_MLOCK:
4119 	case PRIV_VM_MUNLOCK:
4120 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4121 			return (0);
4122 		else
4123 			return (EPERM);
4124 
4125 		/*
4126 		 * Conditionally allow jailed root to bind reserved ports.
4127 		 */
4128 	case PRIV_NETINET_RESERVEDPORT:
4129 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4130 			return (0);
4131 		else
4132 			return (EPERM);
4133 
4134 		/*
4135 		 * Allow jailed root to reuse in-use ports.
4136 		 */
4137 	case PRIV_NETINET_REUSEPORT:
4138 		return (0);
4139 
4140 		/*
4141 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4142 		 */
4143 	case PRIV_NETINET_SETHDROPTS:
4144 		return (0);
4145 
4146 		/*
4147 		 * Conditionally allow creating raw sockets in jail.
4148 		 */
4149 	case PRIV_NETINET_RAW:
4150 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4151 			return (0);
4152 		else
4153 			return (EPERM);
4154 
4155 		/*
4156 		 * Since jail implements its own visibility limits on netstat
4157 		 * sysctls, allow getcred.  This allows identd to work in
4158 		 * jail.
4159 		 */
4160 	case PRIV_NETINET_GETCRED:
4161 		return (0);
4162 
4163 		/*
4164 		 * Allow jailed root to set loginclass.
4165 		 */
4166 	case PRIV_PROC_SETLOGINCLASS:
4167 		return (0);
4168 
4169 		/*
4170 		 * Do not allow a process inside a jail to read the kernel
4171 		 * message buffer unless explicitly permitted.
4172 		 */
4173 	case PRIV_MSGBUF:
4174 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4175 			return (0);
4176 		return (EPERM);
4177 
4178 		/*
4179 		 * Conditionally allow privileged process in the jail adjust
4180 		 * machine time.
4181 		 */
4182 	case PRIV_ADJTIME:
4183 	case PRIV_NTP_ADJTIME:
4184 		if (cred->cr_prison->pr_allow &
4185 		    (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4186 			return (0);
4187 		}
4188 		return (EPERM);
4189 
4190 		/*
4191 		 * Conditionally allow privileged process in the jail set
4192 		 * machine time.
4193 		 */
4194 	case PRIV_CLOCK_SETTIME:
4195 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4196 			return (0);
4197 		else
4198 			return (EPERM);
4199 
4200 	default:
4201 		/*
4202 		 * In all remaining cases, deny the privilege request.  This
4203 		 * includes almost all network privileges, many system
4204 		 * configuration privileges.
4205 		 */
4206 		return (EPERM);
4207 	}
4208 }
4209 
4210 /*
4211  * Return the part of pr2's name that is relative to pr1, or the whole name
4212  * if it does not directly follow.
4213  */
4214 
4215 char *
4216 prison_name(struct prison *pr1, struct prison *pr2)
4217 {
4218 	char *name;
4219 
4220 	/* Jails see themselves as "0" (if they see themselves at all). */
4221 	if (pr1 == pr2)
4222 		return "0";
4223 	name = pr2->pr_name;
4224 	if (prison_ischild(pr1, pr2)) {
4225 		/*
4226 		 * pr1 isn't locked (and allprison_lock may not be either)
4227 		 * so its length can't be counted on.  But the number of dots
4228 		 * can be counted on - and counted.
4229 		 */
4230 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4231 			name = strchr(name, '.') + 1;
4232 	}
4233 	return (name);
4234 }
4235 
4236 /*
4237  * Return the part of pr2's path that is relative to pr1, or the whole path
4238  * if it does not directly follow.
4239  */
4240 static char *
4241 prison_path(struct prison *pr1, struct prison *pr2)
4242 {
4243 	char *path1, *path2;
4244 	int len1;
4245 
4246 	path1 = pr1->pr_path;
4247 	path2 = pr2->pr_path;
4248 	if (!strcmp(path1, "/"))
4249 		return (path2);
4250 	len1 = strlen(path1);
4251 	if (strncmp(path1, path2, len1))
4252 		return (path2);
4253 	if (path2[len1] == '\0')
4254 		return "/";
4255 	if (path2[len1] == '/')
4256 		return (path2 + len1);
4257 	return (path2);
4258 }
4259 
4260 /*
4261  * Jail-related sysctls.
4262  */
4263 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4264     "Jails");
4265 
4266 #if defined(INET) || defined(INET6)
4267 /*
4268  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4269  * sysctl_jail_list() helper.
4270  */
4271 static void
4272 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4273 {
4274 	const struct prison_ip *pip;
4275 	const size_t size = pr_families[af].size;
4276 
4277  again:
4278 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4279 	if ((pip = pr->pr_addrs[af]) != NULL) {
4280 		if (*len < pip->ips) {
4281 			*len = pip->ips;
4282 			mtx_unlock(&pr->pr_mtx);
4283 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4284 			mtx_lock(&pr->pr_mtx);
4285 			goto again;
4286 		}
4287 		bcopy(pip->pr_ip, *out, pip->ips * size);
4288 	}
4289 }
4290 #endif
4291 
4292 static int
4293 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4294 {
4295 	struct xprison *xp;
4296 	struct prison *pr, *cpr;
4297 #ifdef INET
4298 	struct in_addr *ip4 = NULL;
4299 	int ip4s = 0;
4300 #endif
4301 #ifdef INET6
4302 	struct in6_addr *ip6 = NULL;
4303 	int ip6s = 0;
4304 #endif
4305 	int descend, error;
4306 
4307 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4308 	pr = req->td->td_ucred->cr_prison;
4309 	error = 0;
4310 	sx_slock(&allprison_lock);
4311 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4312 		mtx_lock(&cpr->pr_mtx);
4313 #ifdef INET
4314 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4315 #endif
4316 #ifdef INET6
4317 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4318 #endif
4319 		bzero(xp, sizeof(*xp));
4320 		xp->pr_version = XPRISON_VERSION;
4321 		xp->pr_id = cpr->pr_id;
4322 		xp->pr_state = cpr->pr_state;
4323 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4324 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4325 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4326 #ifdef INET
4327 		xp->pr_ip4s = ip4s;
4328 #endif
4329 #ifdef INET6
4330 		xp->pr_ip6s = ip6s;
4331 #endif
4332 		mtx_unlock(&cpr->pr_mtx);
4333 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4334 		if (error)
4335 			break;
4336 #ifdef INET
4337 		if (xp->pr_ip4s > 0) {
4338 			error = SYSCTL_OUT(req, ip4,
4339 			    xp->pr_ip4s * sizeof(struct in_addr));
4340 			if (error)
4341 				break;
4342 		}
4343 #endif
4344 #ifdef INET6
4345 		if (xp->pr_ip6s > 0) {
4346 			error = SYSCTL_OUT(req, ip6,
4347 			    xp->pr_ip6s * sizeof(struct in6_addr));
4348 			if (error)
4349 				break;
4350 		}
4351 #endif
4352 	}
4353 	sx_sunlock(&allprison_lock);
4354 	free(xp, M_TEMP);
4355 #ifdef INET
4356 	free(ip4, M_TEMP);
4357 #endif
4358 #ifdef INET6
4359 	free(ip6, M_TEMP);
4360 #endif
4361 	return (error);
4362 }
4363 
4364 SYSCTL_OID(_security_jail, OID_AUTO, list,
4365     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4366     sysctl_jail_list, "S", "List of active jails");
4367 
4368 static int
4369 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4370 {
4371 	int error, injail;
4372 
4373 	injail = jailed(req->td->td_ucred);
4374 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4375 
4376 	return (error);
4377 }
4378 
4379 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4380     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4381     sysctl_jail_jailed, "I", "Process in jail?");
4382 
4383 static int
4384 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4385 {
4386 	int error, havevnet;
4387 #ifdef VIMAGE
4388 	struct ucred *cred = req->td->td_ucred;
4389 
4390 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4391 #else
4392 	havevnet = 0;
4393 #endif
4394 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4395 
4396 	return (error);
4397 }
4398 
4399 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4400     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4401     sysctl_jail_vnet, "I", "Jail owns vnet?");
4402 
4403 #if defined(INET) || defined(INET6)
4404 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4405     &jail_max_af_ips, 0,
4406     "Number of IP addresses a jail may have at most per address family (deprecated)");
4407 #endif
4408 
4409 /*
4410  * Default parameters for jail(2) compatibility.  For historical reasons,
4411  * the sysctl names have varying similarity to the parameter names.  Prisons
4412  * just see their own parameters, and can't change them.
4413  */
4414 static int
4415 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4416 {
4417 	int error, i;
4418 
4419 	/* Get the current flag value, and convert it to a boolean. */
4420 	if (req->td->td_ucred->cr_prison == &prison0) {
4421 		mtx_lock(&prison0.pr_mtx);
4422 		i = (jail_default_allow & arg2) != 0;
4423 		mtx_unlock(&prison0.pr_mtx);
4424 	} else
4425 		i = prison_allow(req->td->td_ucred, arg2);
4426 
4427 	if (arg1 != NULL)
4428 		i = !i;
4429 	error = sysctl_handle_int(oidp, &i, 0, req);
4430 	if (error || !req->newptr)
4431 		return (error);
4432 	i = i ? arg2 : 0;
4433 	if (arg1 != NULL)
4434 		i ^= arg2;
4435 	/*
4436 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4437 	 * for writing.
4438 	 */
4439 	mtx_lock(&prison0.pr_mtx);
4440 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4441 	mtx_unlock(&prison0.pr_mtx);
4442 	return (0);
4443 }
4444 
4445 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4446     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4447     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4448     "Processes in jail can set their hostnames (deprecated)");
4449 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4450     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4451     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4452     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4453 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4454     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4455     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4456     "Processes in jail can use System V IPC primitives (deprecated)");
4457 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4458     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4459     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4460     "Prison root can create raw sockets (deprecated)");
4461 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4462     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4463     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4464     "Processes in jail can alter system file flags (deprecated)");
4465 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4466     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4467     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4468     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4469 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
4470     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4471     NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
4472     "Processes in jail can lock/unlock physical pages in memory");
4473 
4474 static int
4475 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4476 {
4477 	struct prison *pr;
4478 	int level, error;
4479 
4480 	pr = req->td->td_ucred->cr_prison;
4481 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4482 	error = sysctl_handle_int(oidp, &level, 0, req);
4483 	if (error || !req->newptr)
4484 		return (error);
4485 	*(int *)arg1 = level;
4486 	return (0);
4487 }
4488 
4489 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4490     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4491     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4492     sysctl_jail_default_level, "I",
4493     "Processes in jail cannot see all mounted file systems (deprecated)");
4494 
4495 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4496     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4497     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4498     sysctl_jail_default_level, "I",
4499     "Ruleset for the devfs filesystem in jail (deprecated)");
4500 
4501 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4502     "Limits and stats of child jails");
4503 
4504 static int
4505 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
4506 {
4507 	struct prison *pr;
4508 	int i;
4509 
4510 	pr = req->td->td_ucred->cr_prison;
4511 
4512 	switch (oidp->oid_kind & CTLTYPE) {
4513 	case CTLTYPE_INT:
4514 		i = *(int *)((char *)pr + arg2);
4515 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4516 	}
4517 
4518 	return (0);
4519 }
4520 
4521 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
4522     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4523     NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
4524     "I", "Maximum number of child jails");
4525 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
4526     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4527     NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
4528     "I", "Current number of child jails");
4529 
4530 /*
4531  * Nodes to describe jail parameters.  Maximum length of string parameters
4532  * is returned in the string itself, and the other parameters exist merely
4533  * to make themselves and their types known.
4534  */
4535 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4536     "Jail parameters");
4537 
4538 int
4539 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4540 {
4541 	int i;
4542 	long l;
4543 	size_t s;
4544 	char numbuf[12];
4545 
4546 	switch (oidp->oid_kind & CTLTYPE)
4547 	{
4548 	case CTLTYPE_LONG:
4549 	case CTLTYPE_ULONG:
4550 		l = 0;
4551 #ifdef SCTL_MASK32
4552 		if (!(req->flags & SCTL_MASK32))
4553 #endif
4554 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4555 	case CTLTYPE_INT:
4556 	case CTLTYPE_UINT:
4557 		i = 0;
4558 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4559 	case CTLTYPE_STRING:
4560 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4561 		return
4562 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4563 	case CTLTYPE_STRUCT:
4564 		s = (size_t)arg2;
4565 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4566 	}
4567 	return (0);
4568 }
4569 
4570 /*
4571  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4572  * jail creation time but cannot be changed in an existing jail.
4573  */
4574 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4575 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4576 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4577 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4578 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4579     "I", "Jail secure level");
4580 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4581     "Jail value for kern.osreldate and uname -K");
4582 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4583     "Jail value for kern.osrelease and uname -r");
4584 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4585     "I", "Jail cannot see all mounted file systems");
4586 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4587     "I", "Ruleset for in-jail devfs mounts");
4588 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4589     "B", "Jail persistence");
4590 #ifdef VIMAGE
4591 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4592     "E,jailsys", "Virtual network stack");
4593 #endif
4594 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4595     "B", "Jail is in the process of shutting down");
4596 
4597 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4598 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4599     "I", "Current number of child jails");
4600 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4601     "I", "Maximum number of child jails");
4602 
4603 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4604 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4605     "Jail hostname");
4606 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4607     "Jail NIS domainname");
4608 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4609     "Jail host UUID");
4610 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4611     "LU", "Jail host ID");
4612 
4613 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4614 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4615 
4616 #ifdef INET
4617 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4618     "Jail IPv4 address virtualization");
4619 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4620     "S,in_addr,a", "Jail IPv4 addresses");
4621 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4622     "B", "Do (not) use IPv4 source address selection rather than the "
4623     "primary jail IPv4 address.");
4624 #endif
4625 #ifdef INET6
4626 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4627     "Jail IPv6 address virtualization");
4628 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4629     "S,in6_addr,a", "Jail IPv6 addresses");
4630 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4631     "B", "Do (not) use IPv6 source address selection rather than the "
4632     "primary jail IPv6 address.");
4633 #endif
4634 
4635 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4636 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4637     "B", "Jail may set hostname");
4638 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4639     "B", "Jail may use SYSV IPC");
4640 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4641     "B", "Jail may create raw sockets");
4642 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4643     "B", "Jail may alter system file flags");
4644 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4645     "B", "Jail may set file quotas");
4646 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4647     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4648 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4649     "B", "Jail may lock (unlock) physical pages in memory");
4650 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4651     "B", "Jail may bind sockets to reserved ports");
4652 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4653     "B", "Jail may read the kernel message buffer");
4654 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4655     "B", "Unprivileged processes may use process debugging facilities");
4656 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4657     "B", "Processes in jail with uid 0 have privilege");
4658 #ifdef VIMAGE
4659 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4660     "B", "Mountd/nfsd may run in the jail");
4661 #endif
4662 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
4663     "B", "Jail may set system-level filesystem extended attributes");
4664 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
4665     "B", "Jail may adjust system time");
4666 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
4667     "B", "Jail may set system time");
4668 
4669 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4670 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4671     "B", "Jail may mount/unmount jail-friendly file systems in general");
4672 
4673 /*
4674  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4675  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4676  * not created.
4677  */
4678 unsigned
4679 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4680     const char *descr)
4681 {
4682 	struct bool_flags *bf;
4683 	struct sysctl_oid *parent;
4684 	char *allow_name, *allow_noname, *allowed;
4685 #ifndef NO_SYSCTL_DESCR
4686 	char *descr_deprecated;
4687 #endif
4688 	u_int allow_flag;
4689 
4690 	if (prefix
4691 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4692 		< 0 ||
4693 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4694 		< 0
4695 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4696 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4697 		free(allow_name, M_PRISON);
4698 		return 0;
4699 	}
4700 
4701 	/*
4702 	 * See if this parameter has already beed added, i.e. a module was
4703 	 * previously loaded/unloaded.
4704 	 */
4705 	mtx_lock(&prison0.pr_mtx);
4706 	for (bf = pr_flag_allow;
4707 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4708 		atomic_load_int(&bf->flag) != 0;
4709 	     bf++) {
4710 		if (strcmp(bf->name, allow_name) == 0) {
4711 			allow_flag = bf->flag;
4712 			goto no_add;
4713 		}
4714 	}
4715 
4716 	/*
4717 	 * Find a free bit in pr_allow_all, failing if there are none
4718 	 * (which shouldn't happen as long as we keep track of how many
4719 	 * potential dynamic flags exist).
4720 	 */
4721 	for (allow_flag = 1;; allow_flag <<= 1) {
4722 		if (allow_flag == 0)
4723 			goto no_add;
4724 		if ((pr_allow_all & allow_flag) == 0)
4725 			break;
4726 	}
4727 
4728 	/* Note the parameter in the next open slot in pr_flag_allow. */
4729 	for (bf = pr_flag_allow; ; bf++) {
4730 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4731 			/* This should never happen, but is not fatal. */
4732 			allow_flag = 0;
4733 			goto no_add;
4734 		}
4735 		if (atomic_load_int(&bf->flag) == 0)
4736 			break;
4737 	}
4738 	bf->name = allow_name;
4739 	bf->noname = allow_noname;
4740 	pr_allow_all |= allow_flag;
4741 	/*
4742 	 * prison0 always has permission for the new parameter.
4743 	 * Other jails must have it granted to them.
4744 	 */
4745 	prison0.pr_allow |= allow_flag;
4746 	/* The flag indicates a valid entry, so make sure it is set last. */
4747 	atomic_store_rel_int(&bf->flag, allow_flag);
4748 	mtx_unlock(&prison0.pr_mtx);
4749 
4750 	/*
4751 	 * Create sysctls for the parameter, and the back-compat global
4752 	 * permission.
4753 	 */
4754 	parent = prefix
4755 	    ? SYSCTL_ADD_NODE(NULL,
4756 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4757 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4758 	    : &sysctl___security_jail_param_allow;
4759 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4760 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4761 	    NULL, 0, sysctl_jail_param, "B", descr);
4762 	if ((prefix
4763 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4764 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4765 #ifndef NO_SYSCTL_DESCR
4766 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4767 		    descr);
4768 #endif
4769 		(void)SYSCTL_ADD_PROC(NULL,
4770 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4771 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4772 		    sysctl_jail_default_allow, "I", descr_deprecated);
4773 #ifndef NO_SYSCTL_DESCR
4774 		free(descr_deprecated, M_TEMP);
4775 #endif
4776 		free(allowed, M_TEMP);
4777 	}
4778 	return allow_flag;
4779 
4780  no_add:
4781 	mtx_unlock(&prison0.pr_mtx);
4782 	free(allow_name, M_PRISON);
4783 	free(allow_noname, M_PRISON);
4784 	return allow_flag;
4785 }
4786 
4787 /*
4788  * The VFS system will register jail-aware filesystems here.  They each get
4789  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4790  * attempts to mount.
4791  */
4792 void
4793 prison_add_vfs(struct vfsconf *vfsp)
4794 {
4795 #ifdef NO_SYSCTL_DESCR
4796 
4797 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4798 	    NULL, NULL);
4799 #else
4800 	char *descr;
4801 
4802 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4803 	    vfsp->vfc_name);
4804 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4805 	    NULL, descr);
4806 	free(descr, M_TEMP);
4807 #endif
4808 }
4809 
4810 #ifdef RACCT
4811 void
4812 prison_racct_foreach(void (*callback)(struct racct *racct,
4813     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4814     void *arg2, void *arg3)
4815 {
4816 	struct prison_racct *prr;
4817 
4818 	ASSERT_RACCT_ENABLED();
4819 
4820 	sx_slock(&allprison_lock);
4821 	if (pre != NULL)
4822 		(pre)();
4823 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4824 		(callback)(prr->prr_racct, arg2, arg3);
4825 	if (post != NULL)
4826 		(post)();
4827 	sx_sunlock(&allprison_lock);
4828 }
4829 
4830 static struct prison_racct *
4831 prison_racct_find_locked(const char *name)
4832 {
4833 	struct prison_racct *prr;
4834 
4835 	ASSERT_RACCT_ENABLED();
4836 	sx_assert(&allprison_lock, SA_XLOCKED);
4837 
4838 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4839 		return (NULL);
4840 
4841 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4842 		if (strcmp(name, prr->prr_name) != 0)
4843 			continue;
4844 
4845 		/* Found prison_racct with a matching name? */
4846 		prison_racct_hold(prr);
4847 		return (prr);
4848 	}
4849 
4850 	/* Add new prison_racct. */
4851 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4852 	racct_create(&prr->prr_racct);
4853 
4854 	strcpy(prr->prr_name, name);
4855 	refcount_init(&prr->prr_refcount, 1);
4856 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4857 
4858 	return (prr);
4859 }
4860 
4861 struct prison_racct *
4862 prison_racct_find(const char *name)
4863 {
4864 	struct prison_racct *prr;
4865 
4866 	ASSERT_RACCT_ENABLED();
4867 
4868 	sx_xlock(&allprison_lock);
4869 	prr = prison_racct_find_locked(name);
4870 	sx_xunlock(&allprison_lock);
4871 	return (prr);
4872 }
4873 
4874 void
4875 prison_racct_hold(struct prison_racct *prr)
4876 {
4877 
4878 	ASSERT_RACCT_ENABLED();
4879 
4880 	refcount_acquire(&prr->prr_refcount);
4881 }
4882 
4883 static void
4884 prison_racct_free_locked(struct prison_racct *prr)
4885 {
4886 
4887 	ASSERT_RACCT_ENABLED();
4888 	sx_assert(&allprison_lock, SA_XLOCKED);
4889 
4890 	if (refcount_release(&prr->prr_refcount)) {
4891 		racct_destroy(&prr->prr_racct);
4892 		LIST_REMOVE(prr, prr_next);
4893 		free(prr, M_PRISON_RACCT);
4894 	}
4895 }
4896 
4897 void
4898 prison_racct_free(struct prison_racct *prr)
4899 {
4900 
4901 	ASSERT_RACCT_ENABLED();
4902 	sx_assert(&allprison_lock, SA_UNLOCKED);
4903 
4904 	if (refcount_release_if_not_last(&prr->prr_refcount))
4905 		return;
4906 
4907 	sx_xlock(&allprison_lock);
4908 	prison_racct_free_locked(prr);
4909 	sx_xunlock(&allprison_lock);
4910 }
4911 
4912 static void
4913 prison_racct_attach(struct prison *pr)
4914 {
4915 	struct prison_racct *prr;
4916 
4917 	ASSERT_RACCT_ENABLED();
4918 	sx_assert(&allprison_lock, SA_XLOCKED);
4919 
4920 	prr = prison_racct_find_locked(pr->pr_name);
4921 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4922 
4923 	pr->pr_prison_racct = prr;
4924 }
4925 
4926 /*
4927  * Handle jail renaming.  From the racct point of view, renaming means
4928  * moving from one prison_racct to another.
4929  */
4930 static void
4931 prison_racct_modify(struct prison *pr)
4932 {
4933 #ifdef RCTL
4934 	struct proc *p;
4935 	struct ucred *cred;
4936 #endif
4937 	struct prison_racct *oldprr;
4938 
4939 	ASSERT_RACCT_ENABLED();
4940 
4941 	sx_slock(&allproc_lock);
4942 	sx_xlock(&allprison_lock);
4943 
4944 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4945 		sx_xunlock(&allprison_lock);
4946 		sx_sunlock(&allproc_lock);
4947 		return;
4948 	}
4949 
4950 	oldprr = pr->pr_prison_racct;
4951 	pr->pr_prison_racct = NULL;
4952 
4953 	prison_racct_attach(pr);
4954 
4955 	/*
4956 	 * Move resource utilisation records.
4957 	 */
4958 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4959 
4960 #ifdef RCTL
4961 	/*
4962 	 * Force rctl to reattach rules to processes.
4963 	 */
4964 	FOREACH_PROC_IN_SYSTEM(p) {
4965 		PROC_LOCK(p);
4966 		cred = crhold(p->p_ucred);
4967 		PROC_UNLOCK(p);
4968 		rctl_proc_ucred_changed(p, cred);
4969 		crfree(cred);
4970 	}
4971 #endif
4972 
4973 	sx_sunlock(&allproc_lock);
4974 	prison_racct_free_locked(oldprr);
4975 	sx_xunlock(&allprison_lock);
4976 }
4977 
4978 static void
4979 prison_racct_detach(struct prison *pr)
4980 {
4981 
4982 	ASSERT_RACCT_ENABLED();
4983 	sx_assert(&allprison_lock, SA_UNLOCKED);
4984 
4985 	if (pr->pr_prison_racct == NULL)
4986 		return;
4987 	prison_racct_free(pr->pr_prison_racct);
4988 	pr->pr_prison_racct = NULL;
4989 }
4990 #endif /* RACCT */
4991 
4992 #ifdef DDB
4993 
4994 static void
4995 db_show_prison(struct prison *pr)
4996 {
4997 	struct bool_flags *bf;
4998 	struct jailsys_flags *jsf;
4999 #if defined(INET) || defined(INET6)
5000 	int ii;
5001 	struct prison_ip *pip;
5002 #endif
5003 	unsigned f;
5004 #ifdef INET
5005 	char ip4buf[INET_ADDRSTRLEN];
5006 #endif
5007 #ifdef INET6
5008 	char ip6buf[INET6_ADDRSTRLEN];
5009 #endif
5010 
5011 	db_printf("prison %p:\n", pr);
5012 	db_printf(" jid             = %d\n", pr->pr_id);
5013 	db_printf(" name            = %s\n", pr->pr_name);
5014 	db_printf(" parent          = %p\n", pr->pr_parent);
5015 	db_printf(" ref             = %d\n", pr->pr_ref);
5016 	db_printf(" uref            = %d\n", pr->pr_uref);
5017 	db_printf(" state           = %s\n",
5018 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5019 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
5020 	    "invalid");
5021 	db_printf(" path            = %s\n", pr->pr_path);
5022 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
5023 	    ? pr->pr_cpuset->cs_id : -1);
5024 #ifdef VIMAGE
5025 	db_printf(" vnet            = %p\n", pr->pr_vnet);
5026 #endif
5027 	db_printf(" root            = %p\n", pr->pr_root);
5028 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
5029 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
5030 	db_printf(" children.max    = %d\n", pr->pr_childmax);
5031 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
5032 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
5033 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
5034 	db_printf(" flags           = 0x%x", pr->pr_flags);
5035 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5036 		if (pr->pr_flags & bf->flag)
5037 			db_printf(" %s", bf->name);
5038 	for (jsf = pr_flag_jailsys;
5039 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5040 	     jsf++) {
5041 		f = pr->pr_flags & (jsf->disable | jsf->new);
5042 		db_printf(" %-16s= %s\n", jsf->name,
5043 		    (f != 0 && f == jsf->disable) ? "disable"
5044 		    : (f == jsf->new) ? "new"
5045 		    : "inherit");
5046 	}
5047 	db_printf(" allow           = 0x%x", pr->pr_allow);
5048 	for (bf = pr_flag_allow;
5049 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5050 		atomic_load_int(&bf->flag) != 0;
5051 	     bf++)
5052 		if (pr->pr_allow & bf->flag)
5053 			db_printf(" %s", bf->name);
5054 	db_printf("\n");
5055 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
5056 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
5057 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
5058 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
5059 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
5060 #ifdef INET
5061 	if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5062 		db_printf(" ip4s            = %d\n", pip->ips);
5063 		for (ii = 0; ii < pip->ips; ii++)
5064 			db_printf(" %s %s\n",
5065 			    ii == 0 ? "ip4.addr        =" : "                 ",
5066 			    inet_ntoa_r(
5067 			    *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5068 			    ip4buf));
5069 	}
5070 #endif
5071 #ifdef INET6
5072 	if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5073 		db_printf(" ip6s            = %d\n", pip->ips);
5074 		for (ii = 0; ii < pip->ips; ii++)
5075 			db_printf(" %s %s\n",
5076 			    ii == 0 ? "ip6.addr        =" : "                 ",
5077 			    ip6_sprintf(ip6buf,
5078 			    (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5079 	}
5080 #endif
5081 }
5082 
5083 DB_SHOW_COMMAND(prison, db_show_prison_command)
5084 {
5085 	struct prison *pr;
5086 
5087 	if (!have_addr) {
5088 		/*
5089 		 * Show all prisons in the list, and prison0 which is not
5090 		 * listed.
5091 		 */
5092 		db_show_prison(&prison0);
5093 		if (!db_pager_quit) {
5094 			TAILQ_FOREACH(pr, &allprison, pr_list) {
5095 				db_show_prison(pr);
5096 				if (db_pager_quit)
5097 					break;
5098 			}
5099 		}
5100 		return;
5101 	}
5102 
5103 	if (addr == 0)
5104 		pr = &prison0;
5105 	else {
5106 		/* Look for a prison with the ID and with references. */
5107 		TAILQ_FOREACH(pr, &allprison, pr_list)
5108 			if (pr->pr_id == addr && pr->pr_ref > 0)
5109 				break;
5110 		if (pr == NULL)
5111 			/* Look again, without requiring a reference. */
5112 			TAILQ_FOREACH(pr, &allprison, pr_list)
5113 				if (pr->pr_id == addr)
5114 					break;
5115 		if (pr == NULL)
5116 			/* Assume address points to a valid prison. */
5117 			pr = (struct prison *)addr;
5118 	}
5119 	db_show_prison(pr);
5120 }
5121 
5122 #endif /* DDB */
5123