xref: /freebsd/sys/kern/kern_jail.c (revision 69c5fa5cd1ec9b09ed88a086607a8a0993818db9)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/linker.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/racct.h>
55 #include <sys/rctl.h>
56 #include <sys/refcount.h>
57 #include <sys/sx.h>
58 #include <sys/sysent.h>
59 #include <sys/namei.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysctl.h>
65 #include <sys/uuid.h>
66 #include <sys/vnode.h>
67 
68 #include <net/if.h>
69 #include <net/vnet.h>
70 
71 #include <netinet/in.h>
72 
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif /* DDB */
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
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_childmax	= JAIL_MAX,
110 	.pr_hostuuid	= DEFAULT_HOSTUUID,
111 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
112 #ifdef VIMAGE
113 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
114 #else
115 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
116 #endif
117 	.pr_allow	= PR_ALLOW_ALL_STATIC,
118 };
119 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
120 
121 struct bool_flags {
122 	const char	*name;
123 	const char	*noname;
124 	volatile u_int	 flag;
125 };
126 struct jailsys_flags {
127 	const char	*name;
128 	unsigned	 disable;
129 	unsigned	 new;
130 };
131 
132 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
133 struct	sx allprison_lock;
134 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
135 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
136 LIST_HEAD(, prison_racct) allprison_racct;
137 int	lastprid = 0;
138 
139 static int get_next_prid(struct prison **insprp);
140 static int do_jail_attach(struct thread *td, struct prison *pr);
141 static void prison_complete(void *context, int pending);
142 static void prison_deref(struct prison *pr, int flags);
143 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
144     int enable);
145 static char *prison_path(struct prison *pr1, struct prison *pr2);
146 static void prison_remove_one(struct prison *pr);
147 #ifdef RACCT
148 static void prison_racct_attach(struct prison *pr);
149 static void prison_racct_modify(struct prison *pr);
150 static void prison_racct_detach(struct prison *pr);
151 #endif
152 
153 /* Flags for prison_deref */
154 #define	PD_DEREF	0x01	/* Decrement pr_ref */
155 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
156 #define	PD_LOCKED	0x04	/* pr_mtx is held */
157 #define	PD_LIST_SLOCKED	0x08	/* allprison_lock is held shared */
158 #define	PD_LIST_XLOCKED	0x10	/* allprison_lock is held exclusive */
159 
160 /*
161  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
162  * as we cannot figure out the size of a sparse array, or an array without a
163  * terminating entry.
164  */
165 static struct bool_flags pr_flag_bool[] = {
166 	{"persist", "nopersist", PR_PERSIST},
167 #ifdef INET
168 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
169 #endif
170 #ifdef INET6
171 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
172 #endif
173 };
174 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
175 
176 static struct jailsys_flags pr_flag_jailsys[] = {
177 	{"host", 0, PR_HOST},
178 #ifdef VIMAGE
179 	{"vnet", 0, PR_VNET},
180 #endif
181 #ifdef INET
182 	{"ip4", PR_IP4_USER, PR_IP4_USER},
183 #endif
184 #ifdef INET6
185 	{"ip6", PR_IP6_USER, PR_IP6_USER},
186 #endif
187 };
188 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
189 
190 /*
191  * Make this array full-size so dynamic parameters can be added.
192  * It is protected by prison0.mtx, but lockless reading is allowed
193  * with an atomic check of the flag values.
194  */
195 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
196 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
197 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
198 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
199 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
200 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
201 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
202 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
203 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
204 	{"allow.reserved_ports", "allow.noreserved_ports",
205 	 PR_ALLOW_RESERVED_PORTS},
206 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
207 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
208 	 PR_ALLOW_UNPRIV_DEBUG},
209 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
210 };
211 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
212 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
213 
214 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
215 					 PR_ALLOW_RESERVED_PORTS | \
216 					 PR_ALLOW_UNPRIV_DEBUG | \
217 					 PR_ALLOW_SUSER)
218 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
219 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
220 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
221 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
222 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
223 #if defined(INET) || defined(INET6)
224 static unsigned jail_max_af_ips = 255;
225 #endif
226 
227 /*
228  * Initialize the parts of prison0 that can't be static-initialized with
229  * constants.  This is called from proc0_init() after creating thread0 cpuset.
230  */
231 void
232 prison0_init(void)
233 {
234 	uint8_t *file, *data;
235 	size_t size;
236 
237 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
238 	prison0.pr_osreldate = osreldate;
239 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
240 
241 	/* If we have a preloaded hostuuid, use it. */
242 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
243 	if (file != NULL) {
244 		data = preload_fetch_addr(file);
245 		size = preload_fetch_size(file);
246 		if (data != NULL) {
247 			/*
248 			 * The preloaded data may include trailing whitespace, almost
249 			 * certainly a newline; skip over any whitespace or
250 			 * non-printable characters to be safe.
251 			 */
252 			while (size > 0 && data[size - 1] <= 0x20) {
253 				data[size--] = '\0';
254 			}
255 			if (validate_uuid(data, size, NULL, 0) == 0) {
256 				(void)strlcpy(prison0.pr_hostuuid, data,
257 				    size + 1);
258 			} else if (bootverbose) {
259 				printf("hostuuid: preload data malformed: '%s'",
260 				    data);
261 			}
262 		}
263 	}
264 	if (bootverbose)
265 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
266 }
267 
268 /*
269  * struct jail_args {
270  *	struct jail *jail;
271  * };
272  */
273 int
274 sys_jail(struct thread *td, struct jail_args *uap)
275 {
276 	uint32_t version;
277 	int error;
278 	struct jail j;
279 
280 	error = copyin(uap->jail, &version, sizeof(uint32_t));
281 	if (error)
282 		return (error);
283 
284 	switch (version) {
285 	case 0:
286 	{
287 		struct jail_v0 j0;
288 
289 		/* FreeBSD single IPv4 jails. */
290 		bzero(&j, sizeof(struct jail));
291 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
292 		if (error)
293 			return (error);
294 		j.version = j0.version;
295 		j.path = j0.path;
296 		j.hostname = j0.hostname;
297 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
298 		break;
299 	}
300 
301 	case 1:
302 		/*
303 		 * Version 1 was used by multi-IPv4 jail implementations
304 		 * that never made it into the official kernel.
305 		 */
306 		return (EINVAL);
307 
308 	case 2:	/* JAIL_API_VERSION */
309 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
310 		error = copyin(uap->jail, &j, sizeof(struct jail));
311 		if (error)
312 			return (error);
313 		break;
314 
315 	default:
316 		/* Sci-Fi jails are not supported, sorry. */
317 		return (EINVAL);
318 	}
319 	return (kern_jail(td, &j));
320 }
321 
322 int
323 kern_jail(struct thread *td, struct jail *j)
324 {
325 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
326 #ifdef INET
327 			    + 1
328 #endif
329 #ifdef INET6
330 			    + 1
331 #endif
332 			    )];
333 	struct uio opt;
334 	char *u_path, *u_hostname, *u_name;
335 	struct bool_flags *bf;
336 #ifdef INET
337 	uint32_t ip4s;
338 	struct in_addr *u_ip4;
339 #endif
340 #ifdef INET6
341 	struct in6_addr *u_ip6;
342 #endif
343 	size_t tmplen;
344 	int error, enforce_statfs;
345 
346 	bzero(&optiov, sizeof(optiov));
347 	opt.uio_iov = optiov;
348 	opt.uio_iovcnt = 0;
349 	opt.uio_offset = -1;
350 	opt.uio_resid = -1;
351 	opt.uio_segflg = UIO_SYSSPACE;
352 	opt.uio_rw = UIO_READ;
353 	opt.uio_td = td;
354 
355 	/* Set permissions for top-level jails from sysctls. */
356 	if (!jailed(td->td_ucred)) {
357 		for (bf = pr_flag_allow;
358 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
359 			atomic_load_int(&bf->flag) != 0;
360 		     bf++) {
361 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
362 			    (jail_default_allow & bf->flag)
363 			    ? bf->name : bf->noname);
364 			optiov[opt.uio_iovcnt].iov_len =
365 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
366 			opt.uio_iovcnt += 2;
367 		}
368 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
369 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
370 		opt.uio_iovcnt++;
371 		enforce_statfs = jail_default_enforce_statfs;
372 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
373 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
374 		opt.uio_iovcnt++;
375 	}
376 
377 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
378 #ifdef INET
379 	ip4s = (j->version == 0) ? 1 : j->ip4s;
380 	if (ip4s > jail_max_af_ips)
381 		return (EINVAL);
382 	tmplen += ip4s * sizeof(struct in_addr);
383 #else
384 	if (j->ip4s > 0)
385 		return (EINVAL);
386 #endif
387 #ifdef INET6
388 	if (j->ip6s > jail_max_af_ips)
389 		return (EINVAL);
390 	tmplen += j->ip6s * sizeof(struct in6_addr);
391 #else
392 	if (j->ip6s > 0)
393 		return (EINVAL);
394 #endif
395 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
396 	u_hostname = u_path + MAXPATHLEN;
397 	u_name = u_hostname + MAXHOSTNAMELEN;
398 #ifdef INET
399 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
400 #endif
401 #ifdef INET6
402 #ifdef INET
403 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
404 #else
405 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
406 #endif
407 #endif
408 	optiov[opt.uio_iovcnt].iov_base = "path";
409 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
410 	opt.uio_iovcnt++;
411 	optiov[opt.uio_iovcnt].iov_base = u_path;
412 	error = copyinstr(j->path, u_path, MAXPATHLEN,
413 	    &optiov[opt.uio_iovcnt].iov_len);
414 	if (error) {
415 		free(u_path, M_TEMP);
416 		return (error);
417 	}
418 	opt.uio_iovcnt++;
419 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
420 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
421 	opt.uio_iovcnt++;
422 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
423 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
424 	    &optiov[opt.uio_iovcnt].iov_len);
425 	if (error) {
426 		free(u_path, M_TEMP);
427 		return (error);
428 	}
429 	opt.uio_iovcnt++;
430 	if (j->jailname != NULL) {
431 		optiov[opt.uio_iovcnt].iov_base = "name";
432 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
433 		opt.uio_iovcnt++;
434 		optiov[opt.uio_iovcnt].iov_base = u_name;
435 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
436 		    &optiov[opt.uio_iovcnt].iov_len);
437 		if (error) {
438 			free(u_path, M_TEMP);
439 			return (error);
440 		}
441 		opt.uio_iovcnt++;
442 	}
443 #ifdef INET
444 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
445 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
446 	opt.uio_iovcnt++;
447 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
448 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
449 	if (j->version == 0)
450 		u_ip4->s_addr = j->ip4s;
451 	else {
452 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
453 		if (error) {
454 			free(u_path, M_TEMP);
455 			return (error);
456 		}
457 	}
458 	opt.uio_iovcnt++;
459 #endif
460 #ifdef INET6
461 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
462 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
463 	opt.uio_iovcnt++;
464 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
465 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
466 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
467 	if (error) {
468 		free(u_path, M_TEMP);
469 		return (error);
470 	}
471 	opt.uio_iovcnt++;
472 #endif
473 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
474 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
475 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
476 	free(u_path, M_TEMP);
477 	return (error);
478 }
479 
480 /*
481  * struct jail_set_args {
482  *	struct iovec *iovp;
483  *	unsigned int iovcnt;
484  *	int flags;
485  * };
486  */
487 int
488 sys_jail_set(struct thread *td, struct jail_set_args *uap)
489 {
490 	struct uio *auio;
491 	int error;
492 
493 	/* Check that we have an even number of iovecs. */
494 	if (uap->iovcnt & 1)
495 		return (EINVAL);
496 
497 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
498 	if (error)
499 		return (error);
500 	error = kern_jail_set(td, auio, uap->flags);
501 	free(auio, M_IOV);
502 	return (error);
503 }
504 
505 int
506 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
507 {
508 	struct nameidata nd;
509 #ifdef INET
510 	struct in_addr *ip4;
511 #endif
512 #ifdef INET6
513 	struct in6_addr *ip6;
514 #endif
515 	struct vfsopt *opt;
516 	struct vfsoptlist *opts;
517 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
518 	struct vnode *root;
519 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
520 	char *g_path, *osrelstr;
521 	struct bool_flags *bf;
522 	struct jailsys_flags *jsf;
523 #if defined(INET) || defined(INET6)
524 	struct prison *tppr;
525 	void *op;
526 #endif
527 	unsigned long hid;
528 	size_t namelen, onamelen, pnamelen;
529 	int born, created, cuflags, descend, drflags, enforce;
530 	int error, errmsg_len, errmsg_pos;
531 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
532 	int jid, jsys, len, level;
533 	int childmax, osreldt, rsnum, slevel;
534 #if defined(INET) || defined(INET6)
535 	int ii, ij;
536 #endif
537 #ifdef INET
538 	int ip4s, redo_ip4;
539 #endif
540 #ifdef INET6
541 	int ip6s, redo_ip6;
542 #endif
543 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
544 	uint64_t pr_allow_diff;
545 	unsigned tallow;
546 	char numbuf[12];
547 
548 	error = priv_check(td, PRIV_JAIL_SET);
549 	if (!error && (flags & JAIL_ATTACH))
550 		error = priv_check(td, PRIV_JAIL_ATTACH);
551 	if (error)
552 		return (error);
553 	mypr = td->td_ucred->cr_prison;
554 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
555 		return (EPERM);
556 	if (flags & ~JAIL_SET_MASK)
557 		return (EINVAL);
558 
559 	/*
560 	 * Check all the parameters before committing to anything.  Not all
561 	 * errors can be caught early, but we may as well try.  Also, this
562 	 * takes care of some expensive stuff (path lookup) before getting
563 	 * the allprison lock.
564 	 *
565 	 * XXX Jails are not filesystems, and jail parameters are not mount
566 	 *     options.  But it makes more sense to re-use the vfsopt code
567 	 *     than duplicate it under a different name.
568 	 */
569 	error = vfs_buildopts(optuio, &opts);
570 	if (error)
571 		return (error);
572 #ifdef INET
573 	ip4 = NULL;
574 #endif
575 #ifdef INET6
576 	ip6 = NULL;
577 #endif
578 	g_path = NULL;
579 
580 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
581 	if (!cuflags) {
582 		error = EINVAL;
583 		vfs_opterror(opts, "no valid operation (create or update)");
584 		goto done_errmsg;
585 	}
586 
587 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
588 	if (error == ENOENT)
589 		jid = 0;
590 	else if (error != 0)
591 		goto done_free;
592 
593 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
594 	if (error == ENOENT)
595 		gotslevel = 0;
596 	else if (error != 0)
597 		goto done_free;
598 	else
599 		gotslevel = 1;
600 
601 	error =
602 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
603 	if (error == ENOENT)
604 		gotchildmax = 0;
605 	else if (error != 0)
606 		goto done_free;
607 	else
608 		gotchildmax = 1;
609 
610 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
611 	if (error == ENOENT)
612 		gotenforce = 0;
613 	else if (error != 0)
614 		goto done_free;
615 	else if (enforce < 0 || enforce > 2) {
616 		error = EINVAL;
617 		goto done_free;
618 	} else
619 		gotenforce = 1;
620 
621 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
622 	if (error == ENOENT)
623 		gotrsnum = 0;
624 	else if (error != 0)
625 		goto done_free;
626 	else
627 		gotrsnum = 1;
628 
629 	pr_flags = ch_flags = 0;
630 	for (bf = pr_flag_bool;
631 	     bf < pr_flag_bool + nitems(pr_flag_bool);
632 	     bf++) {
633 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
634 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
635 	}
636 	ch_flags |= pr_flags;
637 	for (jsf = pr_flag_jailsys;
638 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
639 	     jsf++) {
640 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
641 		if (error == ENOENT)
642 			continue;
643 		if (error != 0)
644 			goto done_free;
645 		switch (jsys) {
646 		case JAIL_SYS_DISABLE:
647 			if (!jsf->disable) {
648 				error = EINVAL;
649 				goto done_free;
650 			}
651 			pr_flags |= jsf->disable;
652 			break;
653 		case JAIL_SYS_NEW:
654 			pr_flags |= jsf->new;
655 			break;
656 		case JAIL_SYS_INHERIT:
657 			break;
658 		default:
659 			error = EINVAL;
660 			goto done_free;
661 		}
662 		ch_flags |= jsf->new | jsf->disable;
663 	}
664 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
665 	    && !(pr_flags & PR_PERSIST)) {
666 		error = EINVAL;
667 		vfs_opterror(opts, "new jail must persist or attach");
668 		goto done_errmsg;
669 	}
670 #ifdef VIMAGE
671 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
672 		error = EINVAL;
673 		vfs_opterror(opts, "vnet cannot be changed after creation");
674 		goto done_errmsg;
675 	}
676 #endif
677 #ifdef INET
678 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
679 		error = EINVAL;
680 		vfs_opterror(opts, "ip4 cannot be changed after creation");
681 		goto done_errmsg;
682 	}
683 #endif
684 #ifdef INET6
685 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
686 		error = EINVAL;
687 		vfs_opterror(opts, "ip6 cannot be changed after creation");
688 		goto done_errmsg;
689 	}
690 #endif
691 
692 	pr_allow = ch_allow = 0;
693 	for (bf = pr_flag_allow;
694 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
695 		atomic_load_int(&bf->flag) != 0;
696 	     bf++) {
697 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
698 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
699 	}
700 	ch_allow |= pr_allow;
701 
702 	error = vfs_getopt(opts, "name", (void **)&name, &len);
703 	if (error == ENOENT)
704 		name = NULL;
705 	else if (error != 0)
706 		goto done_free;
707 	else {
708 		if (len == 0 || name[len - 1] != '\0') {
709 			error = EINVAL;
710 			goto done_free;
711 		}
712 		if (len > MAXHOSTNAMELEN) {
713 			error = ENAMETOOLONG;
714 			goto done_free;
715 		}
716 	}
717 
718 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
719 	if (error == ENOENT)
720 		host = NULL;
721 	else if (error != 0)
722 		goto done_free;
723 	else {
724 		ch_flags |= PR_HOST;
725 		pr_flags |= PR_HOST;
726 		if (len == 0 || host[len - 1] != '\0') {
727 			error = EINVAL;
728 			goto done_free;
729 		}
730 		if (len > MAXHOSTNAMELEN) {
731 			error = ENAMETOOLONG;
732 			goto done_free;
733 		}
734 	}
735 
736 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
737 	if (error == ENOENT)
738 		domain = NULL;
739 	else if (error != 0)
740 		goto done_free;
741 	else {
742 		ch_flags |= PR_HOST;
743 		pr_flags |= PR_HOST;
744 		if (len == 0 || domain[len - 1] != '\0') {
745 			error = EINVAL;
746 			goto done_free;
747 		}
748 		if (len > MAXHOSTNAMELEN) {
749 			error = ENAMETOOLONG;
750 			goto done_free;
751 		}
752 	}
753 
754 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
755 	if (error == ENOENT)
756 		uuid = NULL;
757 	else if (error != 0)
758 		goto done_free;
759 	else {
760 		ch_flags |= PR_HOST;
761 		pr_flags |= PR_HOST;
762 		if (len == 0 || uuid[len - 1] != '\0') {
763 			error = EINVAL;
764 			goto done_free;
765 		}
766 		if (len > HOSTUUIDLEN) {
767 			error = ENAMETOOLONG;
768 			goto done_free;
769 		}
770 	}
771 
772 #ifdef COMPAT_FREEBSD32
773 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
774 		uint32_t hid32;
775 
776 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
777 		hid = hid32;
778 	} else
779 #endif
780 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
781 	if (error == ENOENT)
782 		gothid = 0;
783 	else if (error != 0)
784 		goto done_free;
785 	else {
786 		gothid = 1;
787 		ch_flags |= PR_HOST;
788 		pr_flags |= PR_HOST;
789 	}
790 
791 #ifdef INET
792 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
793 	if (error == ENOENT)
794 		ip4s = 0;
795 	else if (error != 0)
796 		goto done_free;
797 	else if (ip4s & (sizeof(*ip4) - 1)) {
798 		error = EINVAL;
799 		goto done_free;
800 	} else {
801 		ch_flags |= PR_IP4_USER;
802 		pr_flags |= PR_IP4_USER;
803 		if (ip4s > 0) {
804 			ip4s /= sizeof(*ip4);
805 			if (ip4s > jail_max_af_ips) {
806 				error = EINVAL;
807 				vfs_opterror(opts, "too many IPv4 addresses");
808 				goto done_errmsg;
809 			}
810 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
811 			bcopy(op, ip4, ip4s * sizeof(*ip4));
812 			/*
813 			 * IP addresses are all sorted but ip[0] to preserve
814 			 * the primary IP address as given from userland.
815 			 * This special IP is used for unbound outgoing
816 			 * connections as well for "loopback" traffic in case
817 			 * source address selection cannot find any more fitting
818 			 * address to connect from.
819 			 */
820 			if (ip4s > 1)
821 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
822 				    prison_qcmp_v4);
823 			/*
824 			 * Check for duplicate addresses and do some simple
825 			 * zero and broadcast checks. If users give other bogus
826 			 * addresses it is their problem.
827 			 *
828 			 * We do not have to care about byte order for these
829 			 * checks so we will do them in NBO.
830 			 */
831 			for (ii = 0; ii < ip4s; ii++) {
832 				if (ip4[ii].s_addr == INADDR_ANY ||
833 				    ip4[ii].s_addr == INADDR_BROADCAST) {
834 					error = EINVAL;
835 					goto done_free;
836 				}
837 				if ((ii+1) < ip4s &&
838 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
839 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
840 					error = EINVAL;
841 					goto done_free;
842 				}
843 			}
844 		}
845 	}
846 #endif
847 
848 #ifdef INET6
849 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
850 	if (error == ENOENT)
851 		ip6s = 0;
852 	else if (error != 0)
853 		goto done_free;
854 	else if (ip6s & (sizeof(*ip6) - 1)) {
855 		error = EINVAL;
856 		goto done_free;
857 	} else {
858 		ch_flags |= PR_IP6_USER;
859 		pr_flags |= PR_IP6_USER;
860 		if (ip6s > 0) {
861 			ip6s /= sizeof(*ip6);
862 			if (ip6s > jail_max_af_ips) {
863 				error = EINVAL;
864 				vfs_opterror(opts, "too many IPv6 addresses");
865 				goto done_errmsg;
866 			}
867 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
868 			bcopy(op, ip6, ip6s * sizeof(*ip6));
869 			if (ip6s > 1)
870 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
871 				    prison_qcmp_v6);
872 			for (ii = 0; ii < ip6s; ii++) {
873 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
874 					error = EINVAL;
875 					goto done_free;
876 				}
877 				if ((ii+1) < ip6s &&
878 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
879 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
880 				{
881 					error = EINVAL;
882 					goto done_free;
883 				}
884 			}
885 		}
886 	}
887 #endif
888 
889 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
890 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
891 		error = EINVAL;
892 		vfs_opterror(opts,
893 		    "vnet jails cannot have IP address restrictions");
894 		goto done_errmsg;
895 	}
896 #endif
897 
898 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
899 	if (error == ENOENT)
900 		osrelstr = NULL;
901 	else if (error != 0)
902 		goto done_free;
903 	else {
904 		if (flags & JAIL_UPDATE) {
905 			error = EINVAL;
906 			vfs_opterror(opts,
907 			    "osrelease cannot be changed after creation");
908 			goto done_errmsg;
909 		}
910 		if (len == 0 || osrelstr[len - 1] != '\0') {
911 			error = EINVAL;
912 			goto done_free;
913 		}
914 		if (len >= OSRELEASELEN) {
915 			error = ENAMETOOLONG;
916 			vfs_opterror(opts,
917 			    "osrelease string must be 1-%d bytes long",
918 			    OSRELEASELEN - 1);
919 			goto done_errmsg;
920 		}
921 	}
922 
923 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
924 	if (error == ENOENT)
925 		osreldt = 0;
926 	else if (error != 0)
927 		goto done_free;
928 	else {
929 		if (flags & JAIL_UPDATE) {
930 			error = EINVAL;
931 			vfs_opterror(opts,
932 			    "osreldate cannot be changed after creation");
933 			goto done_errmsg;
934 		}
935 		if (osreldt == 0) {
936 			error = EINVAL;
937 			vfs_opterror(opts, "osreldate cannot be 0");
938 			goto done_errmsg;
939 		}
940 	}
941 
942 	root = NULL;
943 	error = vfs_getopt(opts, "path", (void **)&path, &len);
944 	if (error == ENOENT)
945 		path = NULL;
946 	else if (error != 0)
947 		goto done_free;
948 	else {
949 		if (flags & JAIL_UPDATE) {
950 			error = EINVAL;
951 			vfs_opterror(opts,
952 			    "path cannot be changed after creation");
953 			goto done_errmsg;
954 		}
955 		if (len == 0 || path[len - 1] != '\0') {
956 			error = EINVAL;
957 			goto done_free;
958 		}
959 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
960 		    path, td);
961 		error = namei(&nd);
962 		if (error)
963 			goto done_free;
964 		root = nd.ni_vp;
965 		NDFREE(&nd, NDF_ONLY_PNBUF);
966 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
967 		strlcpy(g_path, path, MAXPATHLEN);
968 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
969 		if (error == 0) {
970 			path = g_path;
971 		} else {
972 			/* exit on other errors */
973 			goto done_free;
974 		}
975 		if (root->v_type != VDIR) {
976 			error = ENOTDIR;
977 			vput(root);
978 			goto done_free;
979 		}
980 		VOP_UNLOCK(root);
981 	}
982 
983 	/*
984 	 * Find the specified jail, or at least its parent.
985 	 * This abuses the file error codes ENOENT and EEXIST.
986 	 */
987 	pr = NULL;
988 	ppr = mypr;
989 	inspr = NULL;
990 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
991 		namelc = strrchr(name, '.');
992 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
993 		if (*p != '\0')
994 			jid = 0;
995 	}
996 	sx_xlock(&allprison_lock);
997 	drflags = PD_LIST_XLOCKED;
998 	if (jid != 0) {
999 		if (jid < 0) {
1000 			error = EINVAL;
1001 			vfs_opterror(opts, "negative jid");
1002 			goto done_deref;
1003 		}
1004 		/*
1005 		 * See if a requested jid already exists.  Keep track of
1006 		 * where it can be inserted later.
1007 		 */
1008 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1009 			if (inspr->pr_id == jid) {
1010 				mtx_lock(&inspr->pr_mtx);
1011 				if (prison_isvalid(inspr)) {
1012 					pr = inspr;
1013 					drflags |= PD_LOCKED;
1014 					inspr = NULL;
1015 				} else
1016 					mtx_unlock(&inspr->pr_mtx);
1017 				break;
1018 			}
1019 			if (inspr->pr_id > jid)
1020 				break;
1021 		}
1022 		if (pr != NULL) {
1023 			ppr = pr->pr_parent;
1024 			/* Create: jid must not exist. */
1025 			if (cuflags == JAIL_CREATE) {
1026 				/*
1027 				 * Even creators that cannot see the jail will
1028 				 * get EEXIST.
1029 				 */
1030 				error = EEXIST;
1031 				vfs_opterror(opts, "jail %d already exists",
1032 				    jid);
1033 				goto done_deref;
1034 			}
1035 			if (!prison_ischild(mypr, pr)) {
1036 				/*
1037 				 * Updaters get ENOENT if they cannot see the
1038 				 * jail.  This is true even for CREATE | UPDATE,
1039 				 * which normally cannot give this error.
1040 				 */
1041 				error = ENOENT;
1042 				vfs_opterror(opts, "jail %d not found", jid);
1043 				goto done_deref;
1044 			} else if (!prison_isalive(pr)) {
1045 				if (!(flags & JAIL_DYING)) {
1046 					error = ENOENT;
1047 					vfs_opterror(opts, "jail %d is dying",
1048 					    jid);
1049 					goto done_deref;
1050 				} else if ((flags & JAIL_ATTACH) ||
1051 				    (pr_flags & PR_PERSIST)) {
1052 					/*
1053 					 * A dying jail might be resurrected
1054 					 * (via attach or persist), but first
1055 					 * it must determine if another jail
1056 					 * has claimed its name.  Accomplish
1057 					 * this by implicitly re-setting the
1058 					 * name.
1059 					 */
1060 					if (name == NULL)
1061 						name = prison_name(mypr, pr);
1062 				}
1063 			}
1064 		} else {
1065 			/* Update: jid must exist. */
1066 			if (cuflags == JAIL_UPDATE) {
1067 				error = ENOENT;
1068 				vfs_opterror(opts, "jail %d not found", jid);
1069 				goto done_deref;
1070 			}
1071 		}
1072 	}
1073 	/*
1074 	 * If the caller provided a name, look for a jail by that name.
1075 	 * This has different semantics for creates and updates keyed by jid
1076 	 * (where the name must not already exist in a different jail),
1077 	 * and updates keyed by the name itself (where the name must exist
1078 	 * because that is the jail being updated).
1079 	 */
1080 	namelc = NULL;
1081 	if (name != NULL) {
1082 		namelc = strrchr(name, '.');
1083 		if (namelc == NULL)
1084 			namelc = name;
1085 		else {
1086 			/*
1087 			 * This is a hierarchical name.  Split it into the
1088 			 * parent and child names, and make sure the parent
1089 			 * exists or matches an already found jail.
1090 			 */
1091 			if (pr != NULL) {
1092 				if (strncmp(name, ppr->pr_name, namelc - name)
1093 				    || ppr->pr_name[namelc - name] != '\0') {
1094 					error = EINVAL;
1095 					vfs_opterror(opts,
1096 					    "cannot change jail's parent");
1097 					goto done_deref;
1098 				}
1099 			} else {
1100 				*namelc = '\0';
1101 				ppr = prison_find_name(mypr, name);
1102 				if (ppr == NULL) {
1103 					error = ENOENT;
1104 					vfs_opterror(opts,
1105 					    "jail \"%s\" not found", name);
1106 					goto done_deref;
1107 				}
1108 				if (!(flags & JAIL_DYING) &&
1109 				    !prison_isalive(ppr)) {
1110 					mtx_unlock(&ppr->pr_mtx);
1111 					error = ENOENT;
1112 					vfs_opterror(opts,
1113 					    "jail \"%s\" is dying", name);
1114 					goto done_deref;
1115 				}
1116 				mtx_unlock(&ppr->pr_mtx);
1117 				*namelc = '.';
1118 			}
1119 			namelc++;
1120 		}
1121 		if (namelc[0] != '\0') {
1122 			pnamelen =
1123 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1124  name_again:
1125 			deadpr = NULL;
1126 			FOREACH_PRISON_CHILD(ppr, tpr) {
1127 				if (tpr != pr &&
1128 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1129 					mtx_lock(&tpr->pr_mtx);
1130 					if (prison_isalive(tpr)) {
1131 						if (pr == NULL &&
1132 						    cuflags != JAIL_CREATE) {
1133 							/*
1134 							 * Use this jail
1135 							 * for updates.
1136 							 */
1137 							pr = tpr;
1138 							drflags |= PD_LOCKED;
1139 							break;
1140 						}
1141 						/*
1142 						 * Create, or update(jid):
1143 						 * name must not exist in an
1144 						 * active sibling jail.
1145 						 */
1146 						error = EEXIST;
1147 						mtx_unlock(&tpr->pr_mtx);
1148 						vfs_opterror(opts,
1149 						   "jail \"%s\" already exists",
1150 						   name);
1151 						goto done_deref;
1152 					}
1153 					if (pr == NULL &&
1154 					    cuflags != JAIL_CREATE &&
1155 					    prison_isvalid(tpr))
1156 						deadpr = tpr;
1157 					mtx_unlock(&tpr->pr_mtx);
1158 				}
1159 			}
1160 			/* If no active jail is found, use a dying one. */
1161 			if (deadpr != NULL && pr == NULL) {
1162 				if (flags & JAIL_DYING) {
1163 					mtx_lock(&deadpr->pr_mtx);
1164 					if (!prison_isvalid(deadpr)) {
1165 						mtx_unlock(&deadpr->pr_mtx);
1166 						goto name_again;
1167 					}
1168 					pr = deadpr;
1169 					drflags |= PD_LOCKED;
1170 				} else if (cuflags == JAIL_UPDATE) {
1171 					error = ENOENT;
1172 					vfs_opterror(opts,
1173 					    "jail \"%s\" is dying", name);
1174 					goto done_deref;
1175 				}
1176 			}
1177 			/* Update: name must exist if no jid. */
1178 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1179 				error = ENOENT;
1180 				vfs_opterror(opts, "jail \"%s\" not found",
1181 				    name);
1182 				goto done_deref;
1183 			}
1184 		}
1185 	}
1186 	/* Update: must provide a jid or name. */
1187 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1188 		error = ENOENT;
1189 		vfs_opterror(opts, "update specified no jail");
1190 		goto done_deref;
1191 	}
1192 
1193 	/* If there's no prison to update, create a new one and link it in. */
1194 	created = pr == NULL;
1195 	if (created) {
1196 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1197 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1198 				error = EPERM;
1199 				vfs_opterror(opts, "prison limit exceeded");
1200 				goto done_deref;
1201 			}
1202 		mtx_lock(&ppr->pr_mtx);
1203 		if (!prison_isvalid(ppr)) {
1204 			mtx_unlock(&ppr->pr_mtx);
1205 			error = ENOENT;
1206 			vfs_opterror(opts, "jail \"%s\" not found",
1207 			    prison_name(mypr, ppr));
1208 			goto done_deref;
1209 		}
1210 		prison_hold(ppr);
1211 		if (refcount_acquire(&ppr->pr_uref))
1212 			mtx_unlock(&ppr->pr_mtx);
1213 		else {
1214 			/* This brings the parent back to life. */
1215 			mtx_unlock(&ppr->pr_mtx);
1216 			error = osd_jail_call(ppr, PR_METHOD_CREATE, opts);
1217 			if (error) {
1218 				pr = ppr;
1219 				drflags |= PD_DEREF | PD_DEUREF;
1220 				goto done_deref;
1221 			}
1222                 }
1223 
1224 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1225 			error = EAGAIN;
1226 			vfs_opterror(opts, "no available jail IDs");
1227 			pr = ppr;
1228 			drflags |= PD_DEREF | PD_DEUREF;
1229 			goto done_deref;
1230 		}
1231 
1232 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1233 		LIST_INIT(&pr->pr_children);
1234 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1235 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1236 
1237 		pr->pr_id = jid;
1238 		if (inspr != NULL)
1239 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1240 		else
1241 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1242 
1243 		pr->pr_parent = ppr;
1244 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1245 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1246 			tpr->pr_childcount++;
1247 
1248 		/* Set some default values, and inherit some from the parent. */
1249 		if (namelc == NULL)
1250 			namelc = "";
1251 		if (path == NULL) {
1252 			path = "/";
1253 			root = mypr->pr_root;
1254 			vref(root);
1255 		}
1256 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1257 		pr->pr_flags |= PR_HOST;
1258 #if defined(INET) || defined(INET6)
1259 #ifdef VIMAGE
1260 		if (!(pr_flags & PR_VNET))
1261 #endif
1262 		{
1263 #ifdef INET
1264 			if (!(ch_flags & PR_IP4_USER))
1265 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1266 			else if (!(pr_flags & PR_IP4_USER)) {
1267 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1268 				if (ppr->pr_ip4 != NULL) {
1269 					pr->pr_ip4s = ppr->pr_ip4s;
1270 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1271 					    sizeof(struct in_addr), M_PRISON,
1272 					    M_WAITOK);
1273 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1274 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1275 				}
1276 			}
1277 #endif
1278 #ifdef INET6
1279 			if (!(ch_flags & PR_IP6_USER))
1280 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1281 			else if (!(pr_flags & PR_IP6_USER)) {
1282 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1283 				if (ppr->pr_ip6 != NULL) {
1284 					pr->pr_ip6s = ppr->pr_ip6s;
1285 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1286 					    sizeof(struct in6_addr), M_PRISON,
1287 					    M_WAITOK);
1288 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1289 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1290 				}
1291 			}
1292 #endif
1293 		}
1294 #endif
1295 		/* Source address selection is always on by default. */
1296 		pr->pr_flags |= _PR_IP_SADDRSEL;
1297 
1298 		pr->pr_securelevel = ppr->pr_securelevel;
1299 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1300 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1301 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1302 
1303 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1304 		if (osrelstr == NULL)
1305 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1306 			    sizeof(pr->pr_osrelease));
1307 		else
1308 			strlcpy(pr->pr_osrelease, osrelstr,
1309 			    sizeof(pr->pr_osrelease));
1310 
1311 #ifdef VIMAGE
1312 		/* Allocate a new vnet if specified. */
1313 		pr->pr_vnet = (pr_flags & PR_VNET)
1314 		    ? vnet_alloc() : ppr->pr_vnet;
1315 #endif
1316 		/*
1317 		 * Allocate a dedicated cpuset for each jail.
1318 		 * Unlike other initial settings, this may return an erorr.
1319 		 */
1320 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1321 		if (error)
1322 			goto done_deref;
1323 
1324 		mtx_lock(&pr->pr_mtx);
1325 		drflags |= PD_LOCKED;
1326 		/*
1327 		 * New prisons do not yet have a reference, because we do not
1328 		 * want others to see the incomplete prison once the
1329 		 * allprison_lock is downgraded.
1330 		 */
1331 	} else {
1332 		/*
1333 		 * Grab a reference for existing prisons, to ensure they
1334 		 * continue to exist for the duration of the call.
1335 		 */
1336 		prison_hold(pr);
1337 		drflags |= PD_DEREF;
1338 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1339 		if ((pr->pr_flags & PR_VNET) &&
1340 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1341 			error = EINVAL;
1342 			vfs_opterror(opts,
1343 			    "vnet jails cannot have IP address restrictions");
1344 			goto done_deref;
1345 		}
1346 #endif
1347 #ifdef INET
1348 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1349 			error = EINVAL;
1350 			vfs_opterror(opts,
1351 			    "ip4 cannot be changed after creation");
1352 			goto done_deref;
1353 		}
1354 #endif
1355 #ifdef INET6
1356 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1357 			error = EINVAL;
1358 			vfs_opterror(opts,
1359 			    "ip6 cannot be changed after creation");
1360 			goto done_deref;
1361 		}
1362 #endif
1363 	}
1364 
1365 	/* Do final error checking before setting anything. */
1366 	if (gotslevel) {
1367 		if (slevel < ppr->pr_securelevel) {
1368 			error = EPERM;
1369 			goto done_deref;
1370 		}
1371 	}
1372 	if (gotchildmax) {
1373 		if (childmax >= ppr->pr_childmax) {
1374 			error = EPERM;
1375 			goto done_deref;
1376 		}
1377 	}
1378 	if (gotenforce) {
1379 		if (enforce < ppr->pr_enforce_statfs) {
1380 			error = EPERM;
1381 			goto done_deref;
1382 		}
1383 	}
1384 	if (gotrsnum) {
1385 		/*
1386 		 * devfs_rsnum is a uint16_t
1387 		 */
1388 		if (rsnum < 0 || rsnum > 65535) {
1389 			error = EINVAL;
1390 			goto done_deref;
1391 		}
1392 		/*
1393 		 * Nested jails always inherit parent's devfs ruleset
1394 		 */
1395 		if (jailed(td->td_ucred)) {
1396 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1397 				error = EPERM;
1398 				goto done_deref;
1399 			} else
1400 				rsnum = ppr->pr_devfs_rsnum;
1401 		}
1402 	}
1403 #ifdef INET
1404 	if (ip4s > 0) {
1405 		if (ppr->pr_flags & PR_IP4) {
1406 			/*
1407 			 * Make sure the new set of IP addresses is a
1408 			 * subset of the parent's list.  Don't worry
1409 			 * about the parent being unlocked, as any
1410 			 * setting is done with allprison_lock held.
1411 			 */
1412 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1413 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1414 					break;
1415 			if (ij == ppr->pr_ip4s) {
1416 				error = EPERM;
1417 				goto done_deref;
1418 			}
1419 			if (ip4s > 1) {
1420 				for (ii = ij = 1; ii < ip4s; ii++) {
1421 					if (ip4[ii].s_addr ==
1422 					    ppr->pr_ip4[0].s_addr)
1423 						continue;
1424 					for (; ij < ppr->pr_ip4s; ij++)
1425 						if (ip4[ii].s_addr ==
1426 						    ppr->pr_ip4[ij].s_addr)
1427 							break;
1428 					if (ij == ppr->pr_ip4s)
1429 						break;
1430 				}
1431 				if (ij == ppr->pr_ip4s) {
1432 					error = EPERM;
1433 					goto done_deref;
1434 				}
1435 			}
1436 		}
1437 		/*
1438 		 * Check for conflicting IP addresses.  We permit them
1439 		 * if there is no more than one IP on each jail.  If
1440 		 * there is a duplicate on a jail with more than one
1441 		 * IP stop checking and return error.
1442 		 */
1443 #ifdef VIMAGE
1444 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1445 			if (tppr->pr_flags & PR_VNET)
1446 				break;
1447 #else
1448 		tppr = &prison0;
1449 #endif
1450 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1451 			if (tpr == pr ||
1452 #ifdef VIMAGE
1453 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1454 #endif
1455 			    refcount_load(&tpr->pr_uref) == 0) {
1456 				descend = 0;
1457 				continue;
1458 			}
1459 			if (!(tpr->pr_flags & PR_IP4_USER))
1460 				continue;
1461 			descend = 0;
1462 			if (tpr->pr_ip4 == NULL ||
1463 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1464 				continue;
1465 			for (ii = 0; ii < ip4s; ii++) {
1466 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1467 				    0) {
1468 					error = EADDRINUSE;
1469 					vfs_opterror(opts,
1470 					    "IPv4 addresses clash");
1471 					goto done_deref;
1472 				}
1473 			}
1474 		}
1475 	}
1476 #endif
1477 #ifdef INET6
1478 	if (ip6s > 0) {
1479 		if (ppr->pr_flags & PR_IP6) {
1480 			/*
1481 			 * Make sure the new set of IP addresses is a
1482 			 * subset of the parent's list.
1483 			 */
1484 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1485 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1486 				    &ppr->pr_ip6[ij]))
1487 					break;
1488 			if (ij == ppr->pr_ip6s) {
1489 				error = EPERM;
1490 				goto done_deref;
1491 			}
1492 			if (ip6s > 1) {
1493 				for (ii = ij = 1; ii < ip6s; ii++) {
1494 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1495 					     &ppr->pr_ip6[0]))
1496 						continue;
1497 					for (; ij < ppr->pr_ip6s; ij++)
1498 						if (IN6_ARE_ADDR_EQUAL(
1499 						    &ip6[ii], &ppr->pr_ip6[ij]))
1500 							break;
1501 					if (ij == ppr->pr_ip6s)
1502 						break;
1503 				}
1504 				if (ij == ppr->pr_ip6s) {
1505 					error = EPERM;
1506 					goto done_deref;
1507 				}
1508 			}
1509 		}
1510 		/* Check for conflicting IP addresses. */
1511 #ifdef VIMAGE
1512 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1513 			if (tppr->pr_flags & PR_VNET)
1514 				break;
1515 #else
1516 		tppr = &prison0;
1517 #endif
1518 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1519 			if (tpr == pr ||
1520 #ifdef VIMAGE
1521 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1522 #endif
1523 			    refcount_load(&tpr->pr_uref) == 0) {
1524 				descend = 0;
1525 				continue;
1526 			}
1527 			if (!(tpr->pr_flags & PR_IP6_USER))
1528 				continue;
1529 			descend = 0;
1530 			if (tpr->pr_ip6 == NULL ||
1531 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1532 				continue;
1533 			for (ii = 0; ii < ip6s; ii++) {
1534 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1535 				    0) {
1536 					error = EADDRINUSE;
1537 					vfs_opterror(opts,
1538 					    "IPv6 addresses clash");
1539 					goto done_deref;
1540 				}
1541 			}
1542 		}
1543 	}
1544 #endif
1545 	onamelen = namelen = 0;
1546 	if (namelc != NULL) {
1547 		/* Give a default name of the jid.  Also allow the name to be
1548 		 * explicitly the jid - but not any other number, and only in
1549 		 * normal form (no leading zero/etc).
1550 		 */
1551 		if (namelc[0] == '\0')
1552 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1553 		else if ((strtoul(namelc, &p, 10) != jid ||
1554 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1555 			error = EINVAL;
1556 			vfs_opterror(opts,
1557 			    "name cannot be numeric (unless it is the jid)");
1558 			goto done_deref;
1559 		}
1560 		/*
1561 		 * Make sure the name isn't too long for the prison or its
1562 		 * children.
1563 		 */
1564 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1565 		onamelen = strlen(pr->pr_name + pnamelen);
1566 		namelen = strlen(namelc);
1567 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1568 			error = ENAMETOOLONG;
1569 			goto done_deref;
1570 		}
1571 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1572 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1573 			    sizeof(pr->pr_name)) {
1574 				error = ENAMETOOLONG;
1575 				goto done_deref;
1576 			}
1577 		}
1578 	}
1579 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1580 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1581 		error = EPERM;
1582 		goto done_deref;
1583 	}
1584 
1585 	/*
1586 	 * Let modules check their parameters.  This requires unlocking and
1587 	 * then re-locking the prison, but this is still a valid state as long
1588 	 * as allprison_lock remains xlocked.
1589 	 */
1590 	mtx_unlock(&pr->pr_mtx);
1591 	drflags &= ~PD_LOCKED;
1592 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1593 	if (error != 0)
1594 		goto done_deref;
1595 	mtx_lock(&pr->pr_mtx);
1596 	drflags |= PD_LOCKED;
1597 
1598 	/* At this point, all valid parameters should have been noted. */
1599 	TAILQ_FOREACH(opt, opts, link) {
1600 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1601 			error = EINVAL;
1602 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1603 			goto done_deref;
1604 		}
1605 	}
1606 
1607 	/* Set the parameters of the prison. */
1608 #ifdef INET
1609 	redo_ip4 = 0;
1610 	if (pr_flags & PR_IP4_USER) {
1611 		pr->pr_flags |= PR_IP4;
1612 		free(pr->pr_ip4, M_PRISON);
1613 		pr->pr_ip4s = ip4s;
1614 		pr->pr_ip4 = ip4;
1615 		ip4 = NULL;
1616 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1617 #ifdef VIMAGE
1618 			if (tpr->pr_flags & PR_VNET) {
1619 				descend = 0;
1620 				continue;
1621 			}
1622 #endif
1623 			if (prison_restrict_ip4(tpr, NULL)) {
1624 				redo_ip4 = 1;
1625 				descend = 0;
1626 			}
1627 		}
1628 	}
1629 #endif
1630 #ifdef INET6
1631 	redo_ip6 = 0;
1632 	if (pr_flags & PR_IP6_USER) {
1633 		pr->pr_flags |= PR_IP6;
1634 		free(pr->pr_ip6, M_PRISON);
1635 		pr->pr_ip6s = ip6s;
1636 		pr->pr_ip6 = ip6;
1637 		ip6 = NULL;
1638 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1639 #ifdef VIMAGE
1640 			if (tpr->pr_flags & PR_VNET) {
1641 				descend = 0;
1642 				continue;
1643 			}
1644 #endif
1645 			if (prison_restrict_ip6(tpr, NULL)) {
1646 				redo_ip6 = 1;
1647 				descend = 0;
1648 			}
1649 		}
1650 	}
1651 #endif
1652 	if (gotslevel) {
1653 		pr->pr_securelevel = slevel;
1654 		/* Set all child jails to be at least this level. */
1655 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1656 			if (tpr->pr_securelevel < slevel)
1657 				tpr->pr_securelevel = slevel;
1658 	}
1659 	if (gotchildmax) {
1660 		pr->pr_childmax = childmax;
1661 		/* Set all child jails to under this limit. */
1662 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1663 			if (tpr->pr_childmax > childmax - level)
1664 				tpr->pr_childmax = childmax > level
1665 				    ? childmax - level : 0;
1666 	}
1667 	if (gotenforce) {
1668 		pr->pr_enforce_statfs = enforce;
1669 		/* Pass this restriction on to the children. */
1670 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1671 			if (tpr->pr_enforce_statfs < enforce)
1672 				tpr->pr_enforce_statfs = enforce;
1673 	}
1674 	if (gotrsnum) {
1675 		pr->pr_devfs_rsnum = rsnum;
1676 		/* Pass this restriction on to the children. */
1677 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1678 			tpr->pr_devfs_rsnum = rsnum;
1679 	}
1680 	if (namelc != NULL) {
1681 		if (ppr == &prison0)
1682 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1683 		else
1684 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1685 			    ppr->pr_name, namelc);
1686 		/* Change this component of child names. */
1687 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1688 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1689 			    strlen(tpr->pr_name + onamelen) + 1);
1690 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1691 		}
1692 	}
1693 	if (path != NULL) {
1694 		/* Try to keep a real-rooted full pathname. */
1695 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1696 		pr->pr_root = root;
1697 		root = NULL;
1698 	}
1699 	if (PR_HOST & ch_flags & ~pr_flags) {
1700 		if (pr->pr_flags & PR_HOST) {
1701 			/*
1702 			 * Copy the parent's host info.  As with pr_ip4 above,
1703 			 * the lack of a lock on the parent is not a problem;
1704 			 * it is always set with allprison_lock at least
1705 			 * shared, and is held exclusively here.
1706 			 */
1707 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1708 			    sizeof(pr->pr_hostname));
1709 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1710 			    sizeof(pr->pr_domainname));
1711 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1712 			    sizeof(pr->pr_hostuuid));
1713 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1714 		}
1715 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1716 		/* Set this prison, and any descendants without PR_HOST. */
1717 		if (host != NULL)
1718 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1719 		if (domain != NULL)
1720 			strlcpy(pr->pr_domainname, domain,
1721 			    sizeof(pr->pr_domainname));
1722 		if (uuid != NULL)
1723 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1724 		if (gothid)
1725 			pr->pr_hostid = hid;
1726 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1727 			if (tpr->pr_flags & PR_HOST)
1728 				descend = 0;
1729 			else {
1730 				if (host != NULL)
1731 					strlcpy(tpr->pr_hostname,
1732 					    pr->pr_hostname,
1733 					    sizeof(tpr->pr_hostname));
1734 				if (domain != NULL)
1735 					strlcpy(tpr->pr_domainname,
1736 					    pr->pr_domainname,
1737 					    sizeof(tpr->pr_domainname));
1738 				if (uuid != NULL)
1739 					strlcpy(tpr->pr_hostuuid,
1740 					    pr->pr_hostuuid,
1741 					    sizeof(tpr->pr_hostuuid));
1742 				if (gothid)
1743 					tpr->pr_hostid = hid;
1744 			}
1745 		}
1746 	}
1747 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1748 	if ((tallow = ch_allow & ~pr_allow))
1749 		prison_set_allow_locked(pr, tallow, 0);
1750 	/*
1751 	 * Persistent prisons get an extra reference, and prisons losing their
1752 	 * persist flag lose that reference.  Only do this for existing prisons
1753 	 * for now, so new ones will remain unseen until after the module
1754 	 * handlers have completed.
1755 	 */
1756 	born = !prison_isalive(pr);
1757 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1758 		if (pr_flags & PR_PERSIST) {
1759 			prison_hold(pr);
1760 			refcount_acquire(&pr->pr_uref);
1761 		} else {
1762 			refcount_release(&pr->pr_ref);
1763 			drflags |= PD_DEUREF;
1764 		}
1765 	}
1766 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1767 	mtx_unlock(&pr->pr_mtx);
1768 	drflags &= ~PD_LOCKED;
1769 
1770 #ifdef RACCT
1771 	if (racct_enable && created)
1772 		prison_racct_attach(pr);
1773 #endif
1774 
1775 	/* Locks may have prevented a complete restriction of child IP
1776 	 * addresses.  If so, allocate some more memory and try again.
1777 	 */
1778 #ifdef INET
1779 	while (redo_ip4) {
1780 		ip4s = pr->pr_ip4s;
1781 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1782 		mtx_lock(&pr->pr_mtx);
1783 		redo_ip4 = 0;
1784 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1785 #ifdef VIMAGE
1786 			if (tpr->pr_flags & PR_VNET) {
1787 				descend = 0;
1788 				continue;
1789 			}
1790 #endif
1791 			if (prison_restrict_ip4(tpr, ip4)) {
1792 				if (ip4 != NULL)
1793 					ip4 = NULL;
1794 				else
1795 					redo_ip4 = 1;
1796 			}
1797 		}
1798 		mtx_unlock(&pr->pr_mtx);
1799 	}
1800 #endif
1801 #ifdef INET6
1802 	while (redo_ip6) {
1803 		ip6s = pr->pr_ip6s;
1804 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1805 		mtx_lock(&pr->pr_mtx);
1806 		redo_ip6 = 0;
1807 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1808 #ifdef VIMAGE
1809 			if (tpr->pr_flags & PR_VNET) {
1810 				descend = 0;
1811 				continue;
1812 			}
1813 #endif
1814 			if (prison_restrict_ip6(tpr, ip6)) {
1815 				if (ip6 != NULL)
1816 					ip6 = NULL;
1817 				else
1818 					redo_ip6 = 1;
1819 			}
1820 		}
1821 		mtx_unlock(&pr->pr_mtx);
1822 	}
1823 #endif
1824 
1825 	/* Let the modules do their work. */
1826 	sx_downgrade(&allprison_lock);
1827 	drflags = (drflags & ~PD_LIST_XLOCKED) | PD_LIST_SLOCKED;
1828 	if (born) {
1829 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1830 		if (error) {
1831 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1832 			goto done_deref;
1833 		}
1834 	}
1835 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1836 	if (error) {
1837 		if (born)
1838 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1839 		goto done_deref;
1840 	}
1841 
1842 	/* Attach this process to the prison if requested. */
1843 	if (flags & JAIL_ATTACH) {
1844 		mtx_lock(&pr->pr_mtx);
1845 		error = do_jail_attach(td, pr);
1846 		drflags &= ~PD_LIST_SLOCKED;
1847 		if (error) {
1848 			if (created) {
1849 				/* do_jail_attach has removed the prison. */
1850 				pr = NULL;
1851 			}
1852 			vfs_opterror(opts, "attach failed");
1853 			goto done_deref;
1854 		}
1855 	}
1856 
1857 #ifdef RACCT
1858 	if (racct_enable && !created) {
1859 		if (drflags & PD_LIST_SLOCKED) {
1860 			sx_sunlock(&allprison_lock);
1861 			drflags &= ~PD_LIST_SLOCKED;
1862 		}
1863 		prison_racct_modify(pr);
1864 	}
1865 #endif
1866 
1867 	td->td_retval[0] = pr->pr_id;
1868 
1869 	if (created) {
1870 		/*
1871 		 * Add a reference to newly created persistent prisons
1872 		 * (which was not done earlier so that the prison would
1873 		 * not be publicly visible).
1874 		 */
1875 		if (pr_flags & PR_PERSIST) {
1876 			mtx_lock(&pr->pr_mtx);
1877 			drflags |= PD_LOCKED;
1878 			refcount_acquire(&pr->pr_ref);
1879 			refcount_acquire(&pr->pr_uref);
1880 		} else {
1881 			/* Non-persistent jails need no further changes. */
1882 			pr = NULL;
1883 		}
1884 	}
1885 
1886  done_deref:
1887 	/* Release any temporary prison holds and/or locks. */
1888 	if (pr != NULL)
1889 		prison_deref(pr, drflags);
1890 	else if (drflags & PD_LIST_SLOCKED)
1891 		sx_sunlock(&allprison_lock);
1892 	else if (drflags & PD_LIST_XLOCKED)
1893 		sx_xunlock(&allprison_lock);
1894 	if (root != NULL)
1895 		vrele(root);
1896  done_errmsg:
1897 	if (error) {
1898 		/* Write the error message back to userspace. */
1899 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1900 		    &errmsg_len) == 0 && errmsg_len > 0) {
1901 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1902 			if (optuio->uio_segflg == UIO_SYSSPACE)
1903 				bcopy(errmsg,
1904 				    optuio->uio_iov[errmsg_pos].iov_base,
1905 				    errmsg_len);
1906 			else
1907 				copyout(errmsg,
1908 				    optuio->uio_iov[errmsg_pos].iov_base,
1909 				    errmsg_len);
1910 		}
1911 	}
1912  done_free:
1913 #ifdef INET
1914 	free(ip4, M_PRISON);
1915 #endif
1916 #ifdef INET6
1917 	free(ip6, M_PRISON);
1918 #endif
1919 	if (g_path != NULL)
1920 		free(g_path, M_TEMP);
1921 	vfs_freeopts(opts);
1922 	return (error);
1923 }
1924 
1925 /*
1926  * Find the next available prison ID.  Return the ID on success, or zero
1927  * on failure.  Also set a pointer to the allprison list entry the prison
1928  * should be inserted before.
1929  */
1930 static int
1931 get_next_prid(struct prison **insprp)
1932 {
1933 	struct prison *inspr;
1934 	int jid, maxid;
1935 
1936 	jid = lastprid % JAIL_MAX + 1;
1937 	if (TAILQ_EMPTY(&allprison) ||
1938 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
1939 		/*
1940 		 * A common case is for all jails to be implicitly numbered,
1941 		 * which means they'll go on the end of the list, at least
1942 		 * for the first JAIL_MAX times.
1943 		 */
1944 		inspr = NULL;
1945 	} else {
1946 		/*
1947 		 * Take two passes through the allprison list: first starting
1948 		 * with the proposed jid, then ending with it.
1949 		 */
1950 		for (maxid = JAIL_MAX; maxid != 0; ) {
1951 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
1952 				if (inspr->pr_id < jid)
1953 					continue;
1954 				if (inspr->pr_id > jid ||
1955 				    refcount_load(&inspr->pr_ref) == 0) {
1956 					/*
1957 					 * Found an opening.  This may be a gap
1958 					 * in the list, or a dead jail with the
1959 					 * same ID.
1960 					 */
1961 					maxid = 0;
1962 					break;
1963 				}
1964 				if (++jid > maxid) {
1965 					if (lastprid == maxid || lastprid == 0)
1966 					{
1967 						/*
1968 						 * The entire legal range
1969 						 * has been traversed
1970 						 */
1971 						return 0;
1972 					}
1973 					/* Try again from the start. */
1974 					jid = 1;
1975 					maxid = lastprid;
1976 					break;
1977 				}
1978 			}
1979 			if (inspr == NULL) {
1980 				/* Found room at the end of the list. */
1981 				break;
1982 			}
1983 		}
1984 	}
1985 	*insprp = inspr;
1986 	lastprid = jid;
1987 	return (jid);
1988 }
1989 
1990 /*
1991  * struct jail_get_args {
1992  *	struct iovec *iovp;
1993  *	unsigned int iovcnt;
1994  *	int flags;
1995  * };
1996  */
1997 int
1998 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1999 {
2000 	struct uio *auio;
2001 	int error;
2002 
2003 	/* Check that we have an even number of iovecs. */
2004 	if (uap->iovcnt & 1)
2005 		return (EINVAL);
2006 
2007 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2008 	if (error)
2009 		return (error);
2010 	error = kern_jail_get(td, auio, uap->flags);
2011 	if (error == 0)
2012 		error = copyout(auio->uio_iov, uap->iovp,
2013 		    uap->iovcnt * sizeof (struct iovec));
2014 	free(auio, M_IOV);
2015 	return (error);
2016 }
2017 
2018 int
2019 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2020 {
2021 	struct bool_flags *bf;
2022 	struct jailsys_flags *jsf;
2023 	struct prison *pr, *mypr;
2024 	struct vfsopt *opt;
2025 	struct vfsoptlist *opts;
2026 	char *errmsg, *name;
2027 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2028 	unsigned f;
2029 
2030 	if (flags & ~JAIL_GET_MASK)
2031 		return (EINVAL);
2032 
2033 	/* Get the parameter list. */
2034 	error = vfs_buildopts(optuio, &opts);
2035 	if (error)
2036 		return (error);
2037 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2038 	mypr = td->td_ucred->cr_prison;
2039 	pr = NULL;
2040 
2041 	/*
2042 	 * Find the prison specified by one of: lastjid, jid, name.
2043 	 */
2044 	sx_slock(&allprison_lock);
2045 	drflags = PD_LIST_SLOCKED;
2046 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2047 	if (error == 0) {
2048 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2049 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2050 				mtx_lock(&pr->pr_mtx);
2051 				if ((flags & JAIL_DYING)
2052 				    ? prison_isvalid(pr) : prison_isalive(pr))
2053 					break;
2054 				mtx_unlock(&pr->pr_mtx);
2055 			}
2056 		}
2057 		if (pr != NULL) {
2058 			drflags |= PD_LOCKED;
2059 			goto found_prison;
2060 		}
2061 		error = ENOENT;
2062 		vfs_opterror(opts, "no jail after %d", jid);
2063 		goto done;
2064 	} else if (error != ENOENT)
2065 		goto done;
2066 
2067 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2068 	if (error == 0) {
2069 		if (jid != 0) {
2070 			pr = prison_find_child(mypr, jid);
2071 			if (pr != NULL) {
2072 				drflags |= PD_LOCKED;
2073 				if (!(prison_isalive(pr) ||
2074 				    (flags & JAIL_DYING))) {
2075 					error = ENOENT;
2076 					vfs_opterror(opts, "jail %d is dying",
2077 					    jid);
2078 					goto done;
2079 				}
2080 				goto found_prison;
2081 			}
2082 			error = ENOENT;
2083 			vfs_opterror(opts, "jail %d not found", jid);
2084 			goto done;
2085 		}
2086 	} else if (error != ENOENT)
2087 		goto done;
2088 
2089 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2090 	if (error == 0) {
2091 		if (len == 0 || name[len - 1] != '\0') {
2092 			error = EINVAL;
2093 			goto done;
2094 		}
2095 		pr = prison_find_name(mypr, name);
2096 		if (pr != NULL) {
2097 			drflags |= PD_LOCKED;
2098 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2099 				error = ENOENT;
2100 				vfs_opterror(opts, "jail \"%s\" is dying",
2101 				    name);
2102 				goto done;
2103 			}
2104 			goto found_prison;
2105 		}
2106 		error = ENOENT;
2107 		vfs_opterror(opts, "jail \"%s\" not found", name);
2108 		goto done;
2109 	} else if (error != ENOENT)
2110 		goto done;
2111 
2112 	vfs_opterror(opts, "no jail specified");
2113 	error = ENOENT;
2114 	goto done;
2115 
2116  found_prison:
2117 	/* Get the parameters of the prison. */
2118 	prison_hold(pr);
2119 	drflags |= PD_DEREF;
2120 	td->td_retval[0] = pr->pr_id;
2121 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done;
2124 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2125 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2126 	if (error != 0 && error != ENOENT)
2127 		goto done;
2128 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2129 	if (error != 0 && error != ENOENT)
2130 		goto done;
2131 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2132 	    sizeof(pr->pr_cpuset->cs_id));
2133 	if (error != 0 && error != ENOENT)
2134 		goto done;
2135 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2136 	if (error != 0 && error != ENOENT)
2137 		goto done;
2138 #ifdef INET
2139 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2140 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2141 	if (error != 0 && error != ENOENT)
2142 		goto done;
2143 #endif
2144 #ifdef INET6
2145 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2146 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2147 	if (error != 0 && error != ENOENT)
2148 		goto done;
2149 #endif
2150 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2151 	    sizeof(pr->pr_securelevel));
2152 	if (error != 0 && error != ENOENT)
2153 		goto done;
2154 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2155 	    sizeof(pr->pr_childcount));
2156 	if (error != 0 && error != ENOENT)
2157 		goto done;
2158 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2159 	    sizeof(pr->pr_childmax));
2160 	if (error != 0 && error != ENOENT)
2161 		goto done;
2162 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2163 	if (error != 0 && error != ENOENT)
2164 		goto done;
2165 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2166 	if (error != 0 && error != ENOENT)
2167 		goto done;
2168 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2169 	if (error != 0 && error != ENOENT)
2170 		goto done;
2171 #ifdef COMPAT_FREEBSD32
2172 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2173 		uint32_t hid32 = pr->pr_hostid;
2174 
2175 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2176 	} else
2177 #endif
2178 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2179 	    sizeof(pr->pr_hostid));
2180 	if (error != 0 && error != ENOENT)
2181 		goto done;
2182 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2183 	    sizeof(pr->pr_enforce_statfs));
2184 	if (error != 0 && error != ENOENT)
2185 		goto done;
2186 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2187 	    sizeof(pr->pr_devfs_rsnum));
2188 	if (error != 0 && error != ENOENT)
2189 		goto done;
2190 	for (bf = pr_flag_bool;
2191 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2192 	     bf++) {
2193 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2194 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2195 		if (error != 0 && error != ENOENT)
2196 			goto done;
2197 		i = !i;
2198 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2199 		if (error != 0 && error != ENOENT)
2200 			goto done;
2201 	}
2202 	for (jsf = pr_flag_jailsys;
2203 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2204 	     jsf++) {
2205 		f = pr->pr_flags & (jsf->disable | jsf->new);
2206 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2207 		    : (f == jsf->new) ? JAIL_SYS_NEW
2208 		    : JAIL_SYS_INHERIT;
2209 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2210 		if (error != 0 && error != ENOENT)
2211 			goto done;
2212 	}
2213 	for (bf = pr_flag_allow;
2214 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2215 		atomic_load_int(&bf->flag) != 0;
2216 	     bf++) {
2217 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2218 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2219 		if (error != 0 && error != ENOENT)
2220 			goto done;
2221 		i = !i;
2222 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2223 		if (error != 0 && error != ENOENT)
2224 			goto done;
2225 	}
2226 	i = !prison_isalive(pr);
2227 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2228 	if (error != 0 && error != ENOENT)
2229 		goto done;
2230 	i = !i;
2231 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2232 	if (error != 0 && error != ENOENT)
2233 		goto done;
2234 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2235 	    sizeof(pr->pr_osreldate));
2236 	if (error != 0 && error != ENOENT)
2237 		goto done;
2238 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2239 	if (error != 0 && error != ENOENT)
2240 		goto done;
2241 
2242 	/* Get the module parameters. */
2243 	mtx_unlock(&pr->pr_mtx);
2244 	drflags &= ~PD_LOCKED;
2245 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2246 	if (error)
2247 		goto done;
2248 	prison_deref(pr, drflags);
2249 	pr = NULL;
2250 	drflags = 0;
2251 
2252 	/* By now, all parameters should have been noted. */
2253 	TAILQ_FOREACH(opt, opts, link) {
2254 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2255 			error = EINVAL;
2256 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2257 			goto done;
2258 		}
2259 	}
2260 
2261 	/* Write the fetched parameters back to userspace. */
2262 	error = 0;
2263 	TAILQ_FOREACH(opt, opts, link) {
2264 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2265 			pos = 2 * opt->pos + 1;
2266 			optuio->uio_iov[pos].iov_len = opt->len;
2267 			if (opt->value != NULL) {
2268 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2269 					bcopy(opt->value,
2270 					    optuio->uio_iov[pos].iov_base,
2271 					    opt->len);
2272 				} else {
2273 					error = copyout(opt->value,
2274 					    optuio->uio_iov[pos].iov_base,
2275 					    opt->len);
2276 					if (error)
2277 						break;
2278 				}
2279 			}
2280 		}
2281 	}
2282 
2283  done:
2284 	/* Release any temporary prison holds and/or locks. */
2285 	if (pr != NULL)
2286 		prison_deref(pr, drflags);
2287 	else if (drflags & PD_LIST_SLOCKED)
2288 		sx_sunlock(&allprison_lock);
2289 	if (error && errmsg_pos >= 0) {
2290 		/* Write the error message back to userspace. */
2291 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2292 		errmsg_pos = 2 * errmsg_pos + 1;
2293 		if (errmsg_len > 0) {
2294 			if (optuio->uio_segflg == UIO_SYSSPACE)
2295 				bcopy(errmsg,
2296 				    optuio->uio_iov[errmsg_pos].iov_base,
2297 				    errmsg_len);
2298 			else
2299 				copyout(errmsg,
2300 				    optuio->uio_iov[errmsg_pos].iov_base,
2301 				    errmsg_len);
2302 		}
2303 	}
2304 	vfs_freeopts(opts);
2305 	return (error);
2306 }
2307 
2308 /*
2309  * struct jail_remove_args {
2310  *	int jid;
2311  * };
2312  */
2313 int
2314 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2315 {
2316 	struct prison *pr, *cpr, *lpr, *tpr;
2317 	int descend, error;
2318 
2319 	error = priv_check(td, PRIV_JAIL_REMOVE);
2320 	if (error)
2321 		return (error);
2322 
2323 	sx_xlock(&allprison_lock);
2324 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2325 	if (pr == NULL) {
2326 		sx_xunlock(&allprison_lock);
2327 		return (EINVAL);
2328 	}
2329 
2330 	/* Remove all descendants of this prison, then remove this prison. */
2331 	prison_hold(pr);
2332 	if (!LIST_EMPTY(&pr->pr_children)) {
2333 		mtx_unlock(&pr->pr_mtx);
2334 		lpr = NULL;
2335 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2336 			mtx_lock(&cpr->pr_mtx);
2337 			if (prison_isvalid(cpr)) {
2338 				tpr = cpr;
2339 				prison_hold(cpr);
2340 			} else {
2341 				/* Already removed - do not do it again. */
2342 				tpr = NULL;
2343 			}
2344 			mtx_unlock(&cpr->pr_mtx);
2345 			if (lpr != NULL) {
2346 				mtx_lock(&lpr->pr_mtx);
2347 				prison_remove_one(lpr);
2348 				sx_xlock(&allprison_lock);
2349 			}
2350 			lpr = tpr;
2351 		}
2352 		if (lpr != NULL) {
2353 			mtx_lock(&lpr->pr_mtx);
2354 			prison_remove_one(lpr);
2355 			sx_xlock(&allprison_lock);
2356 		}
2357 		mtx_lock(&pr->pr_mtx);
2358 	}
2359 	prison_remove_one(pr);
2360 	return (0);
2361 }
2362 
2363 static void
2364 prison_remove_one(struct prison *pr)
2365 {
2366 	struct proc *p;
2367 	int drflags;
2368 
2369 	drflags = PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED;
2370 
2371 	/* If the prison was persistent, it is not anymore. */
2372 	if (pr->pr_flags & PR_PERSIST) {
2373 		refcount_release(&pr->pr_ref);
2374 		drflags |= PD_DEUREF;
2375 		pr->pr_flags &= ~PR_PERSIST;
2376 	}
2377 
2378 	/*
2379 	 * jail_remove added a reference.  If that's the only one, remove
2380 	 * the prison now.  refcount(9) doesn't guarantee the cache coherence
2381 	 * of non-zero counters, so force it here.
2382 	 */
2383 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2384 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2385 	if (atomic_load_acq_int(&pr->pr_ref) == 1) {
2386 		prison_deref(pr, drflags);
2387 		return;
2388 	}
2389 
2390 	mtx_unlock(&pr->pr_mtx);
2391 	sx_xunlock(&allprison_lock);
2392 	drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2393 	/*
2394 	 * Kill all processes unfortunate enough to be attached to this prison.
2395 	 */
2396 	sx_slock(&allproc_lock);
2397 	FOREACH_PROC_IN_SYSTEM(p) {
2398 		PROC_LOCK(p);
2399 		if (p->p_state != PRS_NEW && p->p_ucred &&
2400 		    p->p_ucred->cr_prison == pr)
2401 			kern_psignal(p, SIGKILL);
2402 		PROC_UNLOCK(p);
2403 	}
2404 	sx_sunlock(&allproc_lock);
2405 	/* Remove the temporary reference added by jail_remove. */
2406 	prison_deref(pr, drflags);
2407 }
2408 
2409 /*
2410  * struct jail_attach_args {
2411  *	int jid;
2412  * };
2413  */
2414 int
2415 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2416 {
2417 	struct prison *pr;
2418 	int error;
2419 
2420 	error = priv_check(td, PRIV_JAIL_ATTACH);
2421 	if (error)
2422 		return (error);
2423 
2424 	/*
2425 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2426 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2427 	 * But then immediately downgrade it since we don't need to stop
2428 	 * readers.
2429 	 */
2430 	sx_xlock(&allprison_lock);
2431 	sx_downgrade(&allprison_lock);
2432 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2433 	if (pr == NULL) {
2434 		sx_sunlock(&allprison_lock);
2435 		return (EINVAL);
2436 	}
2437 
2438 	/* Do not allow a process to attach to a prison that is not alive. */
2439 	if (!prison_isalive(pr)) {
2440 		mtx_unlock(&pr->pr_mtx);
2441 		sx_sunlock(&allprison_lock);
2442 		return (EINVAL);
2443 	}
2444 
2445 	return (do_jail_attach(td, pr));
2446 }
2447 
2448 static int
2449 do_jail_attach(struct thread *td, struct prison *pr)
2450 {
2451 	struct proc *p;
2452 	struct ucred *newcred, *oldcred;
2453 	int error;
2454 
2455 	/*
2456 	 * XXX: Note that there is a slight race here if two threads
2457 	 * in the same privileged process attempt to attach to two
2458 	 * different jails at the same time.  It is important for
2459 	 * user processes not to do this, or they might end up with
2460 	 * a process root from one prison, but attached to the jail
2461 	 * of another.
2462 	 */
2463 	refcount_acquire(&pr->pr_ref);
2464 	refcount_acquire(&pr->pr_uref);
2465 	mtx_unlock(&pr->pr_mtx);
2466 
2467 	/* Let modules do whatever they need to prepare for attaching. */
2468 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2469 	if (error) {
2470 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2471 		return (error);
2472 	}
2473 	sx_sunlock(&allprison_lock);
2474 
2475 	/*
2476 	 * Reparent the newly attached process to this jail.
2477 	 */
2478 	p = td->td_proc;
2479 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2480 	if (error)
2481 		goto e_revert_osd;
2482 
2483 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2484 	if ((error = change_dir(pr->pr_root, td)) != 0)
2485 		goto e_unlock;
2486 #ifdef MAC
2487 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2488 		goto e_unlock;
2489 #endif
2490 	VOP_UNLOCK(pr->pr_root);
2491 	if ((error = pwd_chroot(td, pr->pr_root)))
2492 		goto e_revert_osd;
2493 
2494 	newcred = crget();
2495 	PROC_LOCK(p);
2496 	oldcred = crcopysafe(p, newcred);
2497 	newcred->cr_prison = pr;
2498 	proc_set_cred(p, newcred);
2499 	setsugid(p);
2500 #ifdef RACCT
2501 	racct_proc_ucred_changed(p, oldcred, newcred);
2502 	crhold(newcred);
2503 #endif
2504 	PROC_UNLOCK(p);
2505 #ifdef RCTL
2506 	rctl_proc_ucred_changed(p, newcred);
2507 	crfree(newcred);
2508 #endif
2509 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2510 	crfree(oldcred);
2511 	return (0);
2512 
2513  e_unlock:
2514 	VOP_UNLOCK(pr->pr_root);
2515  e_revert_osd:
2516 	/* Tell modules this thread is still in its old jail after all. */
2517 	sx_slock(&allprison_lock);
2518 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2519 	prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2520 	return (error);
2521 }
2522 
2523 /*
2524  * Returns a locked prison instance, or NULL on failure.
2525  */
2526 struct prison *
2527 prison_find(int prid)
2528 {
2529 	struct prison *pr;
2530 
2531 	sx_assert(&allprison_lock, SX_LOCKED);
2532 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2533 		if (pr->pr_id == prid) {
2534 			mtx_lock(&pr->pr_mtx);
2535 			if (prison_isvalid(pr))
2536 				return (pr);
2537 			/*
2538 			 * Any active prison with the same ID would have
2539 			 * been inserted before a dead one.
2540 			 */
2541 			mtx_unlock(&pr->pr_mtx);
2542 			break;
2543 		}
2544 		if (pr->pr_id > prid)
2545 			break;
2546 	}
2547 	return (NULL);
2548 }
2549 
2550 /*
2551  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2552  */
2553 struct prison *
2554 prison_find_child(struct prison *mypr, int prid)
2555 {
2556 	struct prison *pr;
2557 	int descend;
2558 
2559 	sx_assert(&allprison_lock, SX_LOCKED);
2560 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2561 		if (pr->pr_id == prid) {
2562 			mtx_lock(&pr->pr_mtx);
2563 			if (prison_isvalid(pr))
2564 				return (pr);
2565 			mtx_unlock(&pr->pr_mtx);
2566 		}
2567 	}
2568 	return (NULL);
2569 }
2570 
2571 /*
2572  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2573  */
2574 struct prison *
2575 prison_find_name(struct prison *mypr, const char *name)
2576 {
2577 	struct prison *pr, *deadpr;
2578 	size_t mylen;
2579 	int descend;
2580 
2581 	sx_assert(&allprison_lock, SX_LOCKED);
2582 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2583  again:
2584 	deadpr = NULL;
2585 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2586 		if (!strcmp(pr->pr_name + mylen, name)) {
2587 			mtx_lock(&pr->pr_mtx);
2588 			if (prison_isalive(pr))
2589 				return (pr);
2590 			if (prison_isvalid(pr))
2591 				deadpr = pr;
2592 			mtx_unlock(&pr->pr_mtx);
2593 		}
2594 	}
2595 	/* There was no valid prison - perhaps there was a dying one. */
2596 	if (deadpr != NULL) {
2597 		mtx_lock(&deadpr->pr_mtx);
2598 		if (!prison_isvalid(deadpr)) {
2599 			mtx_unlock(&deadpr->pr_mtx);
2600 			goto again;
2601 		}
2602 	}
2603 	return (deadpr);
2604 }
2605 
2606 /*
2607  * See if a prison has the specific flag set.  The prison should be locked,
2608  * unless checking for flags that are only set at jail creation (such as
2609  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2610  * to any other prison data.
2611  */
2612 int
2613 prison_flag(struct ucred *cred, unsigned flag)
2614 {
2615 
2616 	return (cred->cr_prison->pr_flags & flag);
2617 }
2618 
2619 int
2620 prison_allow(struct ucred *cred, unsigned flag)
2621 {
2622 
2623 	return ((cred->cr_prison->pr_allow & flag) != 0);
2624 }
2625 
2626 /*
2627  * Hold a prison reference, by incrementing pr_ref.  It is generally
2628  * an error to hold a prison that does not already have a reference.
2629  * A prison record will remain valid as long as it has at least one
2630  * reference, and will not be removed as long as either the prison
2631  * mutex or the allprison lock is held (allprison_lock may be shared).
2632  */
2633 void
2634 prison_hold_locked(struct prison *pr)
2635 {
2636 
2637 	/* Locking is no longer required. */
2638 	prison_hold(pr);
2639 }
2640 
2641 void
2642 prison_hold(struct prison *pr)
2643 {
2644 #ifdef INVARIANTS
2645 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2646 
2647 	KASSERT(was_valid,
2648 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2649 #else
2650 	refcount_acquire(&pr->pr_ref);
2651 #endif
2652 }
2653 
2654 /*
2655  * Remove a prison reference.  If that was the last reference, the
2656  * prison will be removed (at a later time).  Return with the prison
2657  * unlocked.
2658  */
2659 void
2660 prison_free_locked(struct prison *pr)
2661 {
2662 	int lastref;
2663 
2664 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2665 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2666 	    ("Trying to free dead prison %p (jid=%d).",
2667 	     pr, pr->pr_id));
2668 	lastref = refcount_release(&pr->pr_ref);
2669 	mtx_unlock(&pr->pr_mtx);
2670 	if (lastref) {
2671 		/*
2672 		 * Don't remove the prison itself in this context,
2673 		 * in case there are locks held.
2674 		 */
2675 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2676 	}
2677 }
2678 
2679 void
2680 prison_free(struct prison *pr)
2681 {
2682 
2683 	/*
2684 	 * Locking is only required when releasing the last reference.
2685 	 * This allows assurance that a locked prison will remain valid
2686 	 * until it is unlocked.
2687 	 */
2688 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2689 	    ("Trying to free dead prison %p (jid=%d).",
2690 	     pr, pr->pr_id));
2691 	if (refcount_release_if_not_last(&pr->pr_ref))
2692 		return;
2693 	mtx_lock(&pr->pr_mtx);
2694 	prison_free_locked(pr);
2695 }
2696 
2697 /*
2698  * Hold a a prison for user visibility, by incrementing pr_uref.
2699  * It is generally an error to hold a prison that isn't already
2700  * user-visible, except through the the jail system calls.  It is also
2701  * an error to hold an invalid prison.  A prison record will remain
2702  * alive as long as it has at least one user reference, and will not
2703  * be set to the dying state was long as the prison mutex is held.
2704  */
2705 void
2706 prison_proc_hold(struct prison *pr)
2707 {
2708 #ifdef INVARIANTS
2709 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2710 
2711 	KASSERT(was_alive,
2712 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2713 #else
2714 	refcount_acquire(&pr->pr_uref);
2715 #endif
2716 }
2717 
2718 /*
2719  * Remove a prison user reference.  If it was the last reference, the
2720  * prison will be considered "dying", and may be removed once all of
2721  * its references are dropped.
2722  */
2723 void
2724 prison_proc_free(struct prison *pr)
2725 {
2726 
2727 	/*
2728 	 * Locking is only required when releasing the last reference.
2729 	 * This allows assurance that a locked prison will remain alive
2730 	 * until it is unlocked.
2731 	 */
2732 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2733 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2734 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2735 		/*
2736 		 * Don't remove the last user reference in this context,
2737 		 * which is expected to be a process that is not only locked,
2738 		 * but also half dead.  Add a reference so any calls to
2739 		 * prison_free() won't re-submit the task.
2740 		 */
2741 		refcount_acquire(&pr->pr_ref);
2742 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2743 	}
2744 }
2745 
2746 /*
2747  * Complete a call to either prison_free or prison_proc_free.
2748  */
2749 static void
2750 prison_complete(void *context, int pending)
2751 {
2752 	struct prison *pr = context;
2753 
2754 	sx_xlock(&allprison_lock);
2755 	mtx_lock(&pr->pr_mtx);
2756 	/*
2757 	 * If this is completing a call to prison_proc_free, there will still
2758 	 * be a user reference held; clear that as well as the reference that
2759 	 * was added.  No references are expected if this is completing a call
2760 	 * to prison_free, but prison_deref is still called for the cleanup.
2761 	 */
2762 	prison_deref(pr, refcount_load(&pr->pr_uref) > 0
2763 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2764 	    : PD_LOCKED | PD_LIST_XLOCKED);
2765 }
2766 
2767 /*
2768  * Remove a prison reference and/or user reference (usually).
2769  * This assumes context that allows sleeping (for allprison_lock),
2770  * with no non-sleeping locks held, except perhaps the prison itself.
2771  * If there are no more references, release and delist the prison.
2772  * On completion, the prison lock and the allprison lock are both
2773  * unlocked.
2774  */
2775 static void
2776 prison_deref(struct prison *pr, int flags)
2777 {
2778 	struct prison *ppr, *tpr;
2779 	int lastref, lasturef;
2780 
2781 	if (!(flags & PD_LOCKED))
2782 		mtx_lock(&pr->pr_mtx);
2783 	for (;;) {
2784 		if (flags & PD_DEUREF) {
2785 			KASSERT(refcount_load(&pr->pr_uref) > 0,
2786 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2787 			     pr->pr_id));
2788 			lasturef = refcount_release(&pr->pr_uref);
2789 			if (lasturef)
2790 				refcount_acquire(&pr->pr_ref);
2791 			KASSERT(refcount_load(&prison0.pr_uref) > 0,
2792 			    ("prison0 pr_uref=0"));
2793 		} else
2794 			lasturef = 0;
2795 		if (flags & PD_DEREF) {
2796 			KASSERT(refcount_load(&pr->pr_ref) > 0,
2797 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2798 			     pr->pr_id));
2799 			lastref = refcount_release(&pr->pr_ref);
2800 		}
2801 		else
2802 			lastref = refcount_load(&pr->pr_ref) == 0;
2803 		mtx_unlock(&pr->pr_mtx);
2804 
2805 		/*
2806 		 * Tell the modules if the last user reference was removed
2807 		 * (even it sticks around in dying state).
2808 		 */
2809 		if (lasturef) {
2810 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2811 				if (atomic_load_acq_int(&pr->pr_ref) > 1) {
2812 					sx_slock(&allprison_lock);
2813 					flags |= PD_LIST_SLOCKED;
2814 				} else {
2815 					sx_xlock(&allprison_lock);
2816 					flags |= PD_LIST_XLOCKED;
2817 				}
2818 			}
2819 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2820 			mtx_lock(&pr->pr_mtx);
2821 			lastref = refcount_release(&pr->pr_ref);
2822 			mtx_unlock(&pr->pr_mtx);
2823 		}
2824 
2825 		/* If the prison still has references, nothing else to do. */
2826 		if (!lastref) {
2827 			if (flags & PD_LIST_SLOCKED)
2828 				sx_sunlock(&allprison_lock);
2829 			else if (flags & PD_LIST_XLOCKED)
2830 				sx_xunlock(&allprison_lock);
2831 			return;
2832 		}
2833 
2834 		if (flags & PD_LIST_SLOCKED) {
2835 			if (!sx_try_upgrade(&allprison_lock)) {
2836 				sx_sunlock(&allprison_lock);
2837 				sx_xlock(&allprison_lock);
2838 			}
2839 		} else if (!(flags & PD_LIST_XLOCKED))
2840 			sx_xlock(&allprison_lock);
2841 
2842 		TAILQ_REMOVE(&allprison, pr, pr_list);
2843 		LIST_REMOVE(pr, pr_sibling);
2844 		ppr = pr->pr_parent;
2845 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2846 			tpr->pr_childcount--;
2847 		sx_xunlock(&allprison_lock);
2848 
2849 #ifdef VIMAGE
2850 		if (pr->pr_vnet != ppr->pr_vnet)
2851 			vnet_destroy(pr->pr_vnet);
2852 #endif
2853 		if (pr->pr_root != NULL)
2854 			vrele(pr->pr_root);
2855 		mtx_destroy(&pr->pr_mtx);
2856 #ifdef INET
2857 		free(pr->pr_ip4, M_PRISON);
2858 #endif
2859 #ifdef INET6
2860 		free(pr->pr_ip6, M_PRISON);
2861 #endif
2862 		if (pr->pr_cpuset != NULL)
2863 			cpuset_rel(pr->pr_cpuset);
2864 		osd_jail_exit(pr);
2865 #ifdef RACCT
2866 		if (racct_enable)
2867 			prison_racct_detach(pr);
2868 #endif
2869 		free(pr, M_PRISON);
2870 
2871 		/* Removing a prison frees a reference on its parent. */
2872 		pr = ppr;
2873 		mtx_lock(&pr->pr_mtx);
2874 		flags = PD_DEREF | PD_DEUREF;
2875 	}
2876 }
2877 
2878 /*
2879  * Set or clear a permission bit in the pr_allow field, passing restrictions
2880  * (cleared permission) down to child jails.
2881  */
2882 void
2883 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
2884 {
2885 	struct prison *pr;
2886 
2887 	pr = cred->cr_prison;
2888 	sx_slock(&allprison_lock);
2889 	mtx_lock(&pr->pr_mtx);
2890 	prison_set_allow_locked(pr, flag, enable);
2891 	mtx_unlock(&pr->pr_mtx);
2892 	sx_sunlock(&allprison_lock);
2893 }
2894 
2895 static void
2896 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
2897 {
2898 	struct prison *cpr;
2899 	int descend;
2900 
2901 	if (enable != 0)
2902 		pr->pr_allow |= flag;
2903 	else {
2904 		pr->pr_allow &= ~flag;
2905 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
2906 			cpr->pr_allow &= ~flag;
2907 	}
2908 }
2909 
2910 /*
2911  * Check if a jail supports the given address family.
2912  *
2913  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2914  * if not.
2915  */
2916 int
2917 prison_check_af(struct ucred *cred, int af)
2918 {
2919 	struct prison *pr;
2920 	int error;
2921 
2922 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2923 
2924 	pr = cred->cr_prison;
2925 #ifdef VIMAGE
2926 	/* Prisons with their own network stack are not limited. */
2927 	if (prison_owns_vnet(cred))
2928 		return (0);
2929 #endif
2930 
2931 	error = 0;
2932 	switch (af)
2933 	{
2934 #ifdef INET
2935 	case AF_INET:
2936 		if (pr->pr_flags & PR_IP4)
2937 		{
2938 			mtx_lock(&pr->pr_mtx);
2939 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2940 				error = EAFNOSUPPORT;
2941 			mtx_unlock(&pr->pr_mtx);
2942 		}
2943 		break;
2944 #endif
2945 #ifdef INET6
2946 	case AF_INET6:
2947 		if (pr->pr_flags & PR_IP6)
2948 		{
2949 			mtx_lock(&pr->pr_mtx);
2950 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2951 				error = EAFNOSUPPORT;
2952 			mtx_unlock(&pr->pr_mtx);
2953 		}
2954 		break;
2955 #endif
2956 	case AF_LOCAL:
2957 	case AF_ROUTE:
2958 		break;
2959 	default:
2960 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2961 			error = EAFNOSUPPORT;
2962 	}
2963 	return (error);
2964 }
2965 
2966 /*
2967  * Check if given address belongs to the jail referenced by cred (wrapper to
2968  * prison_check_ip[46]).
2969  *
2970  * Returns 0 if jail doesn't restrict the address family or if address belongs
2971  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2972  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2973  */
2974 int
2975 prison_if(struct ucred *cred, const struct sockaddr *sa)
2976 {
2977 #ifdef INET
2978 	const struct sockaddr_in *sai;
2979 #endif
2980 #ifdef INET6
2981 	const struct sockaddr_in6 *sai6;
2982 #endif
2983 	int error;
2984 
2985 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2986 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2987 
2988 #ifdef VIMAGE
2989 	if (prison_owns_vnet(cred))
2990 		return (0);
2991 #endif
2992 
2993 	error = 0;
2994 	switch (sa->sa_family)
2995 	{
2996 #ifdef INET
2997 	case AF_INET:
2998 		sai = (const struct sockaddr_in *)sa;
2999 		error = prison_check_ip4(cred, &sai->sin_addr);
3000 		break;
3001 #endif
3002 #ifdef INET6
3003 	case AF_INET6:
3004 		sai6 = (const struct sockaddr_in6 *)sa;
3005 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3006 		break;
3007 #endif
3008 	default:
3009 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3010 			error = EAFNOSUPPORT;
3011 	}
3012 	return (error);
3013 }
3014 
3015 /*
3016  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3017  */
3018 int
3019 prison_check(struct ucred *cred1, struct ucred *cred2)
3020 {
3021 
3022 	return ((cred1->cr_prison == cred2->cr_prison ||
3023 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3024 }
3025 
3026 /*
3027  * Return 1 if p2 is a child of p1, otherwise 0.
3028  */
3029 int
3030 prison_ischild(struct prison *pr1, struct prison *pr2)
3031 {
3032 
3033 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3034 		if (pr1 == pr2)
3035 			return (1);
3036 	return (0);
3037 }
3038 
3039 /*
3040  * Return true if the prison is currently alive.  A prison is alive if it is
3041  * valid and it holds user references.
3042  */
3043 bool
3044 prison_isalive(struct prison *pr)
3045 {
3046 
3047 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3048 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3049 		return (false);
3050 	if (__predict_false(refcount_load(&pr->pr_uref) == 0))
3051 		return (false);
3052 	return (true);
3053 }
3054 
3055 /*
3056  * Return true if the prison is currently valid.  A prison is valid if it has
3057  * been fully created, and is not being destroyed.  Note that dying prisons
3058  * are still considered valid.
3059  */
3060 bool
3061 prison_isvalid(struct prison *pr)
3062 {
3063 
3064 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3065 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3066 		return (false);
3067 	return (true);
3068 }
3069 
3070 /*
3071  * Return 1 if the passed credential is in a jail and that jail does not
3072  * have its own virtual network stack, otherwise 0.
3073  */
3074 int
3075 jailed_without_vnet(struct ucred *cred)
3076 {
3077 
3078 	if (!jailed(cred))
3079 		return (0);
3080 #ifdef VIMAGE
3081 	if (prison_owns_vnet(cred))
3082 		return (0);
3083 #endif
3084 
3085 	return (1);
3086 }
3087 
3088 /*
3089  * Return the correct hostname (domainname, et al) for the passed credential.
3090  */
3091 void
3092 getcredhostname(struct ucred *cred, char *buf, size_t size)
3093 {
3094 	struct prison *pr;
3095 
3096 	/*
3097 	 * A NULL credential can be used to shortcut to the physical
3098 	 * system's hostname.
3099 	 */
3100 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3101 	mtx_lock(&pr->pr_mtx);
3102 	strlcpy(buf, pr->pr_hostname, size);
3103 	mtx_unlock(&pr->pr_mtx);
3104 }
3105 
3106 void
3107 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3108 {
3109 
3110 	mtx_lock(&cred->cr_prison->pr_mtx);
3111 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3112 	mtx_unlock(&cred->cr_prison->pr_mtx);
3113 }
3114 
3115 void
3116 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3117 {
3118 
3119 	mtx_lock(&cred->cr_prison->pr_mtx);
3120 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3121 	mtx_unlock(&cred->cr_prison->pr_mtx);
3122 }
3123 
3124 void
3125 getcredhostid(struct ucred *cred, unsigned long *hostid)
3126 {
3127 
3128 	mtx_lock(&cred->cr_prison->pr_mtx);
3129 	*hostid = cred->cr_prison->pr_hostid;
3130 	mtx_unlock(&cred->cr_prison->pr_mtx);
3131 }
3132 
3133 void
3134 getjailname(struct ucred *cred, char *name, size_t len)
3135 {
3136 
3137 	mtx_lock(&cred->cr_prison->pr_mtx);
3138 	strlcpy(name, cred->cr_prison->pr_name, len);
3139 	mtx_unlock(&cred->cr_prison->pr_mtx);
3140 }
3141 
3142 #ifdef VIMAGE
3143 /*
3144  * Determine whether the prison represented by cred owns
3145  * its vnet rather than having it inherited.
3146  *
3147  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3148  */
3149 int
3150 prison_owns_vnet(struct ucred *cred)
3151 {
3152 
3153 	/*
3154 	 * vnets cannot be added/removed after jail creation,
3155 	 * so no need to lock here.
3156 	 */
3157 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3158 }
3159 #endif
3160 
3161 /*
3162  * Determine whether the subject represented by cred can "see"
3163  * status of a mount point.
3164  * Returns: 0 for permitted, ENOENT otherwise.
3165  * XXX: This function should be called cr_canseemount() and should be
3166  *      placed in kern_prot.c.
3167  */
3168 int
3169 prison_canseemount(struct ucred *cred, struct mount *mp)
3170 {
3171 	struct prison *pr;
3172 	struct statfs *sp;
3173 	size_t len;
3174 
3175 	pr = cred->cr_prison;
3176 	if (pr->pr_enforce_statfs == 0)
3177 		return (0);
3178 	if (pr->pr_root->v_mount == mp)
3179 		return (0);
3180 	if (pr->pr_enforce_statfs == 2)
3181 		return (ENOENT);
3182 	/*
3183 	 * If jail's chroot directory is set to "/" we should be able to see
3184 	 * all mount-points from inside a jail.
3185 	 * This is ugly check, but this is the only situation when jail's
3186 	 * directory ends with '/'.
3187 	 */
3188 	if (strcmp(pr->pr_path, "/") == 0)
3189 		return (0);
3190 	len = strlen(pr->pr_path);
3191 	sp = &mp->mnt_stat;
3192 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3193 		return (ENOENT);
3194 	/*
3195 	 * Be sure that we don't have situation where jail's root directory
3196 	 * is "/some/path" and mount point is "/some/pathpath".
3197 	 */
3198 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3199 		return (ENOENT);
3200 	return (0);
3201 }
3202 
3203 void
3204 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3205 {
3206 	char jpath[MAXPATHLEN];
3207 	struct prison *pr;
3208 	size_t len;
3209 
3210 	pr = cred->cr_prison;
3211 	if (pr->pr_enforce_statfs == 0)
3212 		return;
3213 	if (prison_canseemount(cred, mp) != 0) {
3214 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3215 		strlcpy(sp->f_mntonname, "[restricted]",
3216 		    sizeof(sp->f_mntonname));
3217 		return;
3218 	}
3219 	if (pr->pr_root->v_mount == mp) {
3220 		/*
3221 		 * Clear current buffer data, so we are sure nothing from
3222 		 * the valid path left there.
3223 		 */
3224 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3225 		*sp->f_mntonname = '/';
3226 		return;
3227 	}
3228 	/*
3229 	 * If jail's chroot directory is set to "/" we should be able to see
3230 	 * all mount-points from inside a jail.
3231 	 */
3232 	if (strcmp(pr->pr_path, "/") == 0)
3233 		return;
3234 	len = strlen(pr->pr_path);
3235 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3236 	/*
3237 	 * Clear current buffer data, so we are sure nothing from
3238 	 * the valid path left there.
3239 	 */
3240 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3241 	if (*jpath == '\0') {
3242 		/* Should never happen. */
3243 		*sp->f_mntonname = '/';
3244 	} else {
3245 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3246 	}
3247 }
3248 
3249 /*
3250  * Check with permission for a specific privilege is granted within jail.  We
3251  * have a specific list of accepted privileges; the rest are denied.
3252  */
3253 int
3254 prison_priv_check(struct ucred *cred, int priv)
3255 {
3256 	struct prison *pr;
3257 	int error;
3258 
3259 	/*
3260 	 * Some policies have custom handlers. This routine should not be
3261 	 * called for them. See priv_check_cred().
3262 	 */
3263 	switch (priv) {
3264 	case PRIV_VFS_LOOKUP:
3265 	case PRIV_VFS_GENERATION:
3266 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3267 		    "called for %d\n", priv));
3268 	}
3269 
3270 	if (!jailed(cred))
3271 		return (0);
3272 
3273 #ifdef VIMAGE
3274 	/*
3275 	 * Privileges specific to prisons with a virtual network stack.
3276 	 * There might be a duplicate entry here in case the privilege
3277 	 * is only granted conditionally in the legacy jail case.
3278 	 */
3279 	switch (priv) {
3280 #ifdef notyet
3281 		/*
3282 		 * NFS-specific privileges.
3283 		 */
3284 	case PRIV_NFS_DAEMON:
3285 	case PRIV_NFS_LOCKD:
3286 #endif
3287 		/*
3288 		 * Network stack privileges.
3289 		 */
3290 	case PRIV_NET_BRIDGE:
3291 	case PRIV_NET_GRE:
3292 	case PRIV_NET_BPF:
3293 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3294 	case PRIV_NET_ROUTE:
3295 	case PRIV_NET_TAP:
3296 	case PRIV_NET_SETIFMTU:
3297 	case PRIV_NET_SETIFFLAGS:
3298 	case PRIV_NET_SETIFCAP:
3299 	case PRIV_NET_SETIFDESCR:
3300 	case PRIV_NET_SETIFNAME	:
3301 	case PRIV_NET_SETIFMETRIC:
3302 	case PRIV_NET_SETIFPHYS:
3303 	case PRIV_NET_SETIFMAC:
3304 	case PRIV_NET_SETLANPCP:
3305 	case PRIV_NET_ADDMULTI:
3306 	case PRIV_NET_DELMULTI:
3307 	case PRIV_NET_HWIOCTL:
3308 	case PRIV_NET_SETLLADDR:
3309 	case PRIV_NET_ADDIFGROUP:
3310 	case PRIV_NET_DELIFGROUP:
3311 	case PRIV_NET_IFCREATE:
3312 	case PRIV_NET_IFDESTROY:
3313 	case PRIV_NET_ADDIFADDR:
3314 	case PRIV_NET_DELIFADDR:
3315 	case PRIV_NET_LAGG:
3316 	case PRIV_NET_GIF:
3317 	case PRIV_NET_SETIFVNET:
3318 	case PRIV_NET_SETIFFIB:
3319 
3320 		/*
3321 		 * 802.11-related privileges.
3322 		 */
3323 	case PRIV_NET80211_VAP_GETKEY:
3324 	case PRIV_NET80211_VAP_MANAGE:
3325 
3326 #ifdef notyet
3327 		/*
3328 		 * ATM privileges.
3329 		 */
3330 	case PRIV_NETATM_CFG:
3331 	case PRIV_NETATM_ADD:
3332 	case PRIV_NETATM_DEL:
3333 	case PRIV_NETATM_SET:
3334 
3335 		/*
3336 		 * Bluetooth privileges.
3337 		 */
3338 	case PRIV_NETBLUETOOTH_RAW:
3339 #endif
3340 
3341 		/*
3342 		 * Netgraph and netgraph module privileges.
3343 		 */
3344 	case PRIV_NETGRAPH_CONTROL:
3345 #ifdef notyet
3346 	case PRIV_NETGRAPH_TTY:
3347 #endif
3348 
3349 		/*
3350 		 * IPv4 and IPv6 privileges.
3351 		 */
3352 	case PRIV_NETINET_IPFW:
3353 	case PRIV_NETINET_DIVERT:
3354 	case PRIV_NETINET_PF:
3355 	case PRIV_NETINET_DUMMYNET:
3356 	case PRIV_NETINET_CARP:
3357 	case PRIV_NETINET_MROUTE:
3358 	case PRIV_NETINET_RAW:
3359 	case PRIV_NETINET_ADDRCTRL6:
3360 	case PRIV_NETINET_ND6:
3361 	case PRIV_NETINET_SCOPE6:
3362 	case PRIV_NETINET_ALIFETIME6:
3363 	case PRIV_NETINET_IPSEC:
3364 	case PRIV_NETINET_BINDANY:
3365 
3366 #ifdef notyet
3367 		/*
3368 		 * NCP privileges.
3369 		 */
3370 	case PRIV_NETNCP:
3371 
3372 		/*
3373 		 * SMB privileges.
3374 		 */
3375 	case PRIV_NETSMB:
3376 #endif
3377 
3378 	/*
3379 	 * No default: or deny here.
3380 	 * In case of no permit fall through to next switch().
3381 	 */
3382 		if (cred->cr_prison->pr_flags & PR_VNET)
3383 			return (0);
3384 	}
3385 #endif /* VIMAGE */
3386 
3387 	switch (priv) {
3388 		/*
3389 		 * Allow ktrace privileges for root in jail.
3390 		 */
3391 	case PRIV_KTRACE:
3392 
3393 #if 0
3394 		/*
3395 		 * Allow jailed processes to configure audit identity and
3396 		 * submit audit records (login, etc).  In the future we may
3397 		 * want to further refine the relationship between audit and
3398 		 * jail.
3399 		 */
3400 	case PRIV_AUDIT_GETAUDIT:
3401 	case PRIV_AUDIT_SETAUDIT:
3402 	case PRIV_AUDIT_SUBMIT:
3403 #endif
3404 
3405 		/*
3406 		 * Allow jailed processes to manipulate process UNIX
3407 		 * credentials in any way they see fit.
3408 		 */
3409 	case PRIV_CRED_SETUID:
3410 	case PRIV_CRED_SETEUID:
3411 	case PRIV_CRED_SETGID:
3412 	case PRIV_CRED_SETEGID:
3413 	case PRIV_CRED_SETGROUPS:
3414 	case PRIV_CRED_SETREUID:
3415 	case PRIV_CRED_SETREGID:
3416 	case PRIV_CRED_SETRESUID:
3417 	case PRIV_CRED_SETRESGID:
3418 
3419 		/*
3420 		 * Jail implements visibility constraints already, so allow
3421 		 * jailed root to override uid/gid-based constraints.
3422 		 */
3423 	case PRIV_SEEOTHERGIDS:
3424 	case PRIV_SEEOTHERUIDS:
3425 
3426 		/*
3427 		 * Jail implements inter-process debugging limits already, so
3428 		 * allow jailed root various debugging privileges.
3429 		 */
3430 	case PRIV_DEBUG_DIFFCRED:
3431 	case PRIV_DEBUG_SUGID:
3432 	case PRIV_DEBUG_UNPRIV:
3433 
3434 		/*
3435 		 * Allow jail to set various resource limits and login
3436 		 * properties, and for now, exceed process resource limits.
3437 		 */
3438 	case PRIV_PROC_LIMIT:
3439 	case PRIV_PROC_SETLOGIN:
3440 	case PRIV_PROC_SETRLIMIT:
3441 
3442 		/*
3443 		 * System V and POSIX IPC privileges are granted in jail.
3444 		 */
3445 	case PRIV_IPC_READ:
3446 	case PRIV_IPC_WRITE:
3447 	case PRIV_IPC_ADMIN:
3448 	case PRIV_IPC_MSGSIZE:
3449 	case PRIV_MQ_ADMIN:
3450 
3451 		/*
3452 		 * Jail operations within a jail work on child jails.
3453 		 */
3454 	case PRIV_JAIL_ATTACH:
3455 	case PRIV_JAIL_SET:
3456 	case PRIV_JAIL_REMOVE:
3457 
3458 		/*
3459 		 * Jail implements its own inter-process limits, so allow
3460 		 * root processes in jail to change scheduling on other
3461 		 * processes in the same jail.  Likewise for signalling.
3462 		 */
3463 	case PRIV_SCHED_DIFFCRED:
3464 	case PRIV_SCHED_CPUSET:
3465 	case PRIV_SIGNAL_DIFFCRED:
3466 	case PRIV_SIGNAL_SUGID:
3467 
3468 		/*
3469 		 * Allow jailed processes to write to sysctls marked as jail
3470 		 * writable.
3471 		 */
3472 	case PRIV_SYSCTL_WRITEJAIL:
3473 
3474 		/*
3475 		 * Allow root in jail to manage a variety of quota
3476 		 * properties.  These should likely be conditional on a
3477 		 * configuration option.
3478 		 */
3479 	case PRIV_VFS_GETQUOTA:
3480 	case PRIV_VFS_SETQUOTA:
3481 
3482 		/*
3483 		 * Since Jail relies on chroot() to implement file system
3484 		 * protections, grant many VFS privileges to root in jail.
3485 		 * Be careful to exclude mount-related and NFS-related
3486 		 * privileges.
3487 		 */
3488 	case PRIV_VFS_READ:
3489 	case PRIV_VFS_WRITE:
3490 	case PRIV_VFS_ADMIN:
3491 	case PRIV_VFS_EXEC:
3492 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3493 	case PRIV_VFS_CHFLAGS_DEV:
3494 	case PRIV_VFS_CHOWN:
3495 	case PRIV_VFS_CHROOT:
3496 	case PRIV_VFS_RETAINSUGID:
3497 	case PRIV_VFS_FCHROOT:
3498 	case PRIV_VFS_LINK:
3499 	case PRIV_VFS_SETGID:
3500 	case PRIV_VFS_STAT:
3501 	case PRIV_VFS_STICKYFILE:
3502 
3503 		/*
3504 		 * As in the non-jail case, non-root users are expected to be
3505 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3506 		 * exists in the jail and they have permission to access it).
3507 		 */
3508 	case PRIV_KMEM_READ:
3509 		return (0);
3510 
3511 		/*
3512 		 * Depending on the global setting, allow privilege of
3513 		 * setting system flags.
3514 		 */
3515 	case PRIV_VFS_SYSFLAGS:
3516 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3517 			return (0);
3518 		else
3519 			return (EPERM);
3520 
3521 		/*
3522 		 * Depending on the global setting, allow privilege of
3523 		 * mounting/unmounting file systems.
3524 		 */
3525 	case PRIV_VFS_MOUNT:
3526 	case PRIV_VFS_UNMOUNT:
3527 	case PRIV_VFS_MOUNT_NONUSER:
3528 	case PRIV_VFS_MOUNT_OWNER:
3529 		pr = cred->cr_prison;
3530 		prison_lock(pr);
3531 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3532 			error = 0;
3533 		else
3534 			error = EPERM;
3535 		prison_unlock(pr);
3536 		return (error);
3537 
3538 		/*
3539 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3540 		 * policy.  priv_check_cred will not specifically allow it, and
3541 		 * we may want a MAC policy to allow it.
3542 		 */
3543 	case PRIV_VFS_READ_DIR:
3544 		return (0);
3545 
3546 		/*
3547 		 * Conditionnaly allow locking (unlocking) physical pages
3548 		 * in memory.
3549 		 */
3550 	case PRIV_VM_MLOCK:
3551 	case PRIV_VM_MUNLOCK:
3552 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3553 			return (0);
3554 		else
3555 			return (EPERM);
3556 
3557 		/*
3558 		 * Conditionally allow jailed root to bind reserved ports.
3559 		 */
3560 	case PRIV_NETINET_RESERVEDPORT:
3561 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3562 			return (0);
3563 		else
3564 			return (EPERM);
3565 
3566 		/*
3567 		 * Allow jailed root to reuse in-use ports.
3568 		 */
3569 	case PRIV_NETINET_REUSEPORT:
3570 		return (0);
3571 
3572 		/*
3573 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3574 		 */
3575 	case PRIV_NETINET_SETHDROPTS:
3576 		return (0);
3577 
3578 		/*
3579 		 * Conditionally allow creating raw sockets in jail.
3580 		 */
3581 	case PRIV_NETINET_RAW:
3582 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3583 			return (0);
3584 		else
3585 			return (EPERM);
3586 
3587 		/*
3588 		 * Since jail implements its own visibility limits on netstat
3589 		 * sysctls, allow getcred.  This allows identd to work in
3590 		 * jail.
3591 		 */
3592 	case PRIV_NETINET_GETCRED:
3593 		return (0);
3594 
3595 		/*
3596 		 * Allow jailed root to set loginclass.
3597 		 */
3598 	case PRIV_PROC_SETLOGINCLASS:
3599 		return (0);
3600 
3601 		/*
3602 		 * Do not allow a process inside a jail to read the kernel
3603 		 * message buffer unless explicitly permitted.
3604 		 */
3605 	case PRIV_MSGBUF:
3606 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3607 			return (0);
3608 		return (EPERM);
3609 
3610 	default:
3611 		/*
3612 		 * In all remaining cases, deny the privilege request.  This
3613 		 * includes almost all network privileges, many system
3614 		 * configuration privileges.
3615 		 */
3616 		return (EPERM);
3617 	}
3618 }
3619 
3620 /*
3621  * Return the part of pr2's name that is relative to pr1, or the whole name
3622  * if it does not directly follow.
3623  */
3624 
3625 char *
3626 prison_name(struct prison *pr1, struct prison *pr2)
3627 {
3628 	char *name;
3629 
3630 	/* Jails see themselves as "0" (if they see themselves at all). */
3631 	if (pr1 == pr2)
3632 		return "0";
3633 	name = pr2->pr_name;
3634 	if (prison_ischild(pr1, pr2)) {
3635 		/*
3636 		 * pr1 isn't locked (and allprison_lock may not be either)
3637 		 * so its length can't be counted on.  But the number of dots
3638 		 * can be counted on - and counted.
3639 		 */
3640 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3641 			name = strchr(name, '.') + 1;
3642 	}
3643 	return (name);
3644 }
3645 
3646 /*
3647  * Return the part of pr2's path that is relative to pr1, or the whole path
3648  * if it does not directly follow.
3649  */
3650 static char *
3651 prison_path(struct prison *pr1, struct prison *pr2)
3652 {
3653 	char *path1, *path2;
3654 	int len1;
3655 
3656 	path1 = pr1->pr_path;
3657 	path2 = pr2->pr_path;
3658 	if (!strcmp(path1, "/"))
3659 		return (path2);
3660 	len1 = strlen(path1);
3661 	if (strncmp(path1, path2, len1))
3662 		return (path2);
3663 	if (path2[len1] == '\0')
3664 		return "/";
3665 	if (path2[len1] == '/')
3666 		return (path2 + len1);
3667 	return (path2);
3668 }
3669 
3670 /*
3671  * Jail-related sysctls.
3672  */
3673 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3674     "Jails");
3675 
3676 static int
3677 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3678 {
3679 	struct xprison *xp;
3680 	struct prison *pr, *cpr;
3681 #ifdef INET
3682 	struct in_addr *ip4 = NULL;
3683 	int ip4s = 0;
3684 #endif
3685 #ifdef INET6
3686 	struct in6_addr *ip6 = NULL;
3687 	int ip6s = 0;
3688 #endif
3689 	int descend, error;
3690 
3691 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3692 	pr = req->td->td_ucred->cr_prison;
3693 	error = 0;
3694 	sx_slock(&allprison_lock);
3695 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3696 #if defined(INET) || defined(INET6)
3697  again:
3698 #endif
3699 		mtx_lock(&cpr->pr_mtx);
3700 #ifdef INET
3701 		if (cpr->pr_ip4s > 0) {
3702 			if (ip4s < cpr->pr_ip4s) {
3703 				ip4s = cpr->pr_ip4s;
3704 				mtx_unlock(&cpr->pr_mtx);
3705 				ip4 = realloc(ip4, ip4s *
3706 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3707 				goto again;
3708 			}
3709 			bcopy(cpr->pr_ip4, ip4,
3710 			    cpr->pr_ip4s * sizeof(struct in_addr));
3711 		}
3712 #endif
3713 #ifdef INET6
3714 		if (cpr->pr_ip6s > 0) {
3715 			if (ip6s < cpr->pr_ip6s) {
3716 				ip6s = cpr->pr_ip6s;
3717 				mtx_unlock(&cpr->pr_mtx);
3718 				ip6 = realloc(ip6, ip6s *
3719 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3720 				goto again;
3721 			}
3722 			bcopy(cpr->pr_ip6, ip6,
3723 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3724 		}
3725 #endif
3726 		if (!prison_isvalid(cpr)) {
3727 			mtx_unlock(&cpr->pr_mtx);
3728 			continue;
3729 		}
3730 		bzero(xp, sizeof(*xp));
3731 		xp->pr_version = XPRISON_VERSION;
3732 		xp->pr_id = cpr->pr_id;
3733 		xp->pr_state = prison_isalive(cpr)
3734 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3735 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3736 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3737 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3738 #ifdef INET
3739 		xp->pr_ip4s = cpr->pr_ip4s;
3740 #endif
3741 #ifdef INET6
3742 		xp->pr_ip6s = cpr->pr_ip6s;
3743 #endif
3744 		mtx_unlock(&cpr->pr_mtx);
3745 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3746 		if (error)
3747 			break;
3748 #ifdef INET
3749 		if (xp->pr_ip4s > 0) {
3750 			error = SYSCTL_OUT(req, ip4,
3751 			    xp->pr_ip4s * sizeof(struct in_addr));
3752 			if (error)
3753 				break;
3754 		}
3755 #endif
3756 #ifdef INET6
3757 		if (xp->pr_ip6s > 0) {
3758 			error = SYSCTL_OUT(req, ip6,
3759 			    xp->pr_ip6s * sizeof(struct in6_addr));
3760 			if (error)
3761 				break;
3762 		}
3763 #endif
3764 	}
3765 	sx_sunlock(&allprison_lock);
3766 	free(xp, M_TEMP);
3767 #ifdef INET
3768 	free(ip4, M_TEMP);
3769 #endif
3770 #ifdef INET6
3771 	free(ip6, M_TEMP);
3772 #endif
3773 	return (error);
3774 }
3775 
3776 SYSCTL_OID(_security_jail, OID_AUTO, list,
3777     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3778     sysctl_jail_list, "S", "List of active jails");
3779 
3780 static int
3781 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3782 {
3783 	int error, injail;
3784 
3785 	injail = jailed(req->td->td_ucred);
3786 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3787 
3788 	return (error);
3789 }
3790 
3791 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3792     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3793     sysctl_jail_jailed, "I", "Process in jail?");
3794 
3795 static int
3796 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3797 {
3798 	int error, havevnet;
3799 #ifdef VIMAGE
3800 	struct ucred *cred = req->td->td_ucred;
3801 
3802 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3803 #else
3804 	havevnet = 0;
3805 #endif
3806 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3807 
3808 	return (error);
3809 }
3810 
3811 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3812     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3813     sysctl_jail_vnet, "I", "Jail owns vnet?");
3814 
3815 #if defined(INET) || defined(INET6)
3816 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3817     &jail_max_af_ips, 0,
3818     "Number of IP addresses a jail may have at most per address family (deprecated)");
3819 #endif
3820 
3821 /*
3822  * Default parameters for jail(2) compatibility.  For historical reasons,
3823  * the sysctl names have varying similarity to the parameter names.  Prisons
3824  * just see their own parameters, and can't change them.
3825  */
3826 static int
3827 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3828 {
3829 	int error, i;
3830 
3831 	/* Get the current flag value, and convert it to a boolean. */
3832 	if (req->td->td_ucred->cr_prison == &prison0) {
3833 		mtx_lock(&prison0.pr_mtx);
3834 		i = (jail_default_allow & arg2) != 0;
3835 		mtx_unlock(&prison0.pr_mtx);
3836 	} else
3837 		i = prison_allow(req->td->td_ucred, arg2);
3838 
3839 	if (arg1 != NULL)
3840 		i = !i;
3841 	error = sysctl_handle_int(oidp, &i, 0, req);
3842 	if (error || !req->newptr)
3843 		return (error);
3844 	i = i ? arg2 : 0;
3845 	if (arg1 != NULL)
3846 		i ^= arg2;
3847 	/*
3848 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3849 	 * for writing.
3850 	 */
3851 	mtx_lock(&prison0.pr_mtx);
3852 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3853 	mtx_unlock(&prison0.pr_mtx);
3854 	return (0);
3855 }
3856 
3857 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3858     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3859     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3860     "Processes in jail can set their hostnames (deprecated)");
3861 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3862     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3863     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3864     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3865 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3866     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3867     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3868     "Processes in jail can use System V IPC primitives (deprecated)");
3869 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3870     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3871     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3872     "Prison root can create raw sockets (deprecated)");
3873 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3874     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3875     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3876     "Processes in jail can alter system file flags (deprecated)");
3877 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3878     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3879     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3880     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3881 
3882 static int
3883 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3884 {
3885 	struct prison *pr;
3886 	int level, error;
3887 
3888 	pr = req->td->td_ucred->cr_prison;
3889 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3890 	error = sysctl_handle_int(oidp, &level, 0, req);
3891 	if (error || !req->newptr)
3892 		return (error);
3893 	*(int *)arg1 = level;
3894 	return (0);
3895 }
3896 
3897 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3898     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3899     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3900     sysctl_jail_default_level, "I",
3901     "Processes in jail cannot see all mounted file systems (deprecated)");
3902 
3903 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3904     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3905     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3906     sysctl_jail_default_level, "I",
3907     "Ruleset for the devfs filesystem in jail (deprecated)");
3908 
3909 /*
3910  * Nodes to describe jail parameters.  Maximum length of string parameters
3911  * is returned in the string itself, and the other parameters exist merely
3912  * to make themselves and their types known.
3913  */
3914 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3915     "Jail parameters");
3916 
3917 int
3918 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3919 {
3920 	int i;
3921 	long l;
3922 	size_t s;
3923 	char numbuf[12];
3924 
3925 	switch (oidp->oid_kind & CTLTYPE)
3926 	{
3927 	case CTLTYPE_LONG:
3928 	case CTLTYPE_ULONG:
3929 		l = 0;
3930 #ifdef SCTL_MASK32
3931 		if (!(req->flags & SCTL_MASK32))
3932 #endif
3933 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3934 	case CTLTYPE_INT:
3935 	case CTLTYPE_UINT:
3936 		i = 0;
3937 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3938 	case CTLTYPE_STRING:
3939 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3940 		return
3941 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3942 	case CTLTYPE_STRUCT:
3943 		s = (size_t)arg2;
3944 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3945 	}
3946 	return (0);
3947 }
3948 
3949 /*
3950  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3951  * jail creation time but cannot be changed in an existing jail.
3952  */
3953 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3954 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3955 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3956 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3957 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3958     "I", "Jail secure level");
3959 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3960     "Jail value for kern.osreldate and uname -K");
3961 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3962     "Jail value for kern.osrelease and uname -r");
3963 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3964     "I", "Jail cannot see all mounted file systems");
3965 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3966     "I", "Ruleset for in-jail devfs mounts");
3967 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3968     "B", "Jail persistence");
3969 #ifdef VIMAGE
3970 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3971     "E,jailsys", "Virtual network stack");
3972 #endif
3973 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3974     "B", "Jail is in the process of shutting down");
3975 
3976 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3977 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3978     "I", "Current number of child jails");
3979 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3980     "I", "Maximum number of child jails");
3981 
3982 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3983 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3984     "Jail hostname");
3985 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3986     "Jail NIS domainname");
3987 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3988     "Jail host UUID");
3989 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3990     "LU", "Jail host ID");
3991 
3992 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3993 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3994 
3995 #ifdef INET
3996 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3997     "Jail IPv4 address virtualization");
3998 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3999     "S,in_addr,a", "Jail IPv4 addresses");
4000 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4001     "B", "Do (not) use IPv4 source address selection rather than the "
4002     "primary jail IPv4 address.");
4003 #endif
4004 #ifdef INET6
4005 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4006     "Jail IPv6 address virtualization");
4007 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4008     "S,in6_addr,a", "Jail IPv6 addresses");
4009 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4010     "B", "Do (not) use IPv6 source address selection rather than the "
4011     "primary jail IPv6 address.");
4012 #endif
4013 
4014 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4015 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4016     "B", "Jail may set hostname");
4017 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4018     "B", "Jail may use SYSV IPC");
4019 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4020     "B", "Jail may create raw sockets");
4021 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4022     "B", "Jail may alter system file flags");
4023 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4024     "B", "Jail may set file quotas");
4025 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4026     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4027 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4028     "B", "Jail may lock (unlock) physical pages in memory");
4029 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4030     "B", "Jail may bind sockets to reserved ports");
4031 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4032     "B", "Jail may read the kernel message buffer");
4033 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4034     "B", "Unprivileged processes may use process debugging facilities");
4035 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4036     "B", "Processes in jail with uid 0 have privilege");
4037 
4038 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4039 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4040     "B", "Jail may mount/unmount jail-friendly file systems in general");
4041 
4042 /*
4043  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4044  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4045  * not created.
4046  */
4047 unsigned
4048 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4049     const char *descr)
4050 {
4051 	struct bool_flags *bf;
4052 	struct sysctl_oid *parent;
4053 	char *allow_name, *allow_noname, *allowed;
4054 #ifndef NO_SYSCTL_DESCR
4055 	char *descr_deprecated;
4056 #endif
4057 	u_int allow_flag;
4058 
4059 	if (prefix
4060 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4061 		< 0 ||
4062 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4063 		< 0
4064 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4065 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4066 		free(allow_name, M_PRISON);
4067 		return 0;
4068 	}
4069 
4070 	/*
4071 	 * See if this parameter has already beed added, i.e. a module was
4072 	 * previously loaded/unloaded.
4073 	 */
4074 	mtx_lock(&prison0.pr_mtx);
4075 	for (bf = pr_flag_allow;
4076 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4077 		atomic_load_int(&bf->flag) != 0;
4078 	     bf++) {
4079 		if (strcmp(bf->name, allow_name) == 0) {
4080 			allow_flag = bf->flag;
4081 			goto no_add;
4082 		}
4083 	}
4084 
4085 	/*
4086 	 * Find a free bit in pr_allow_all, failing if there are none
4087 	 * (which shouldn't happen as long as we keep track of how many
4088 	 * potential dynamic flags exist).
4089 	 */
4090 	for (allow_flag = 1;; allow_flag <<= 1) {
4091 		if (allow_flag == 0)
4092 			goto no_add;
4093 		if ((pr_allow_all & allow_flag) == 0)
4094 			break;
4095 	}
4096 
4097 	/* Note the parameter in the next open slot in pr_flag_allow. */
4098 	for (bf = pr_flag_allow; ; bf++) {
4099 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4100 			/* This should never happen, but is not fatal. */
4101 			allow_flag = 0;
4102 			goto no_add;
4103 		}
4104 		if (atomic_load_int(&bf->flag) == 0)
4105 			break;
4106 	}
4107 	bf->name = allow_name;
4108 	bf->noname = allow_noname;
4109 	pr_allow_all |= allow_flag;
4110 	/*
4111 	 * prison0 always has permission for the new parameter.
4112 	 * Other jails must have it granted to them.
4113 	 */
4114 	prison0.pr_allow |= allow_flag;
4115 	/* The flag indicates a valid entry, so make sure it is set last. */
4116 	atomic_store_rel_int(&bf->flag, allow_flag);
4117 	mtx_unlock(&prison0.pr_mtx);
4118 
4119 	/*
4120 	 * Create sysctls for the paramter, and the back-compat global
4121 	 * permission.
4122 	 */
4123 	parent = prefix
4124 	    ? SYSCTL_ADD_NODE(NULL,
4125 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4126 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4127 	    : &sysctl___security_jail_param_allow;
4128 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4129 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4130 	    NULL, 0, sysctl_jail_param, "B", descr);
4131 	if ((prefix
4132 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4133 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4134 #ifndef NO_SYSCTL_DESCR
4135 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4136 		    descr);
4137 #endif
4138 		(void)SYSCTL_ADD_PROC(NULL,
4139 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4140 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4141 		    sysctl_jail_default_allow, "I", descr_deprecated);
4142 #ifndef NO_SYSCTL_DESCR
4143 		free(descr_deprecated, M_TEMP);
4144 #endif
4145 		free(allowed, M_TEMP);
4146 	}
4147 	return allow_flag;
4148 
4149  no_add:
4150 	mtx_unlock(&prison0.pr_mtx);
4151 	free(allow_name, M_PRISON);
4152 	free(allow_noname, M_PRISON);
4153 	return allow_flag;
4154 }
4155 
4156 /*
4157  * The VFS system will register jail-aware filesystems here.  They each get
4158  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4159  * attempts to mount.
4160  */
4161 void
4162 prison_add_vfs(struct vfsconf *vfsp)
4163 {
4164 #ifdef NO_SYSCTL_DESCR
4165 
4166 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4167 	    NULL, NULL);
4168 #else
4169 	char *descr;
4170 
4171 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4172 	    vfsp->vfc_name);
4173 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4174 	    NULL, descr);
4175 	free(descr, M_TEMP);
4176 #endif
4177 }
4178 
4179 #ifdef RACCT
4180 void
4181 prison_racct_foreach(void (*callback)(struct racct *racct,
4182     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4183     void *arg2, void *arg3)
4184 {
4185 	struct prison_racct *prr;
4186 
4187 	ASSERT_RACCT_ENABLED();
4188 
4189 	sx_slock(&allprison_lock);
4190 	if (pre != NULL)
4191 		(pre)();
4192 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4193 		(callback)(prr->prr_racct, arg2, arg3);
4194 	if (post != NULL)
4195 		(post)();
4196 	sx_sunlock(&allprison_lock);
4197 }
4198 
4199 static struct prison_racct *
4200 prison_racct_find_locked(const char *name)
4201 {
4202 	struct prison_racct *prr;
4203 
4204 	ASSERT_RACCT_ENABLED();
4205 	sx_assert(&allprison_lock, SA_XLOCKED);
4206 
4207 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4208 		return (NULL);
4209 
4210 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4211 		if (strcmp(name, prr->prr_name) != 0)
4212 			continue;
4213 
4214 		/* Found prison_racct with a matching name? */
4215 		prison_racct_hold(prr);
4216 		return (prr);
4217 	}
4218 
4219 	/* Add new prison_racct. */
4220 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4221 	racct_create(&prr->prr_racct);
4222 
4223 	strcpy(prr->prr_name, name);
4224 	refcount_init(&prr->prr_refcount, 1);
4225 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4226 
4227 	return (prr);
4228 }
4229 
4230 struct prison_racct *
4231 prison_racct_find(const char *name)
4232 {
4233 	struct prison_racct *prr;
4234 
4235 	ASSERT_RACCT_ENABLED();
4236 
4237 	sx_xlock(&allprison_lock);
4238 	prr = prison_racct_find_locked(name);
4239 	sx_xunlock(&allprison_lock);
4240 	return (prr);
4241 }
4242 
4243 void
4244 prison_racct_hold(struct prison_racct *prr)
4245 {
4246 
4247 	ASSERT_RACCT_ENABLED();
4248 
4249 	refcount_acquire(&prr->prr_refcount);
4250 }
4251 
4252 static void
4253 prison_racct_free_locked(struct prison_racct *prr)
4254 {
4255 
4256 	ASSERT_RACCT_ENABLED();
4257 	sx_assert(&allprison_lock, SA_XLOCKED);
4258 
4259 	if (refcount_release(&prr->prr_refcount)) {
4260 		racct_destroy(&prr->prr_racct);
4261 		LIST_REMOVE(prr, prr_next);
4262 		free(prr, M_PRISON_RACCT);
4263 	}
4264 }
4265 
4266 void
4267 prison_racct_free(struct prison_racct *prr)
4268 {
4269 
4270 	ASSERT_RACCT_ENABLED();
4271 	sx_assert(&allprison_lock, SA_UNLOCKED);
4272 
4273 	if (refcount_release_if_not_last(&prr->prr_refcount))
4274 		return;
4275 
4276 	sx_xlock(&allprison_lock);
4277 	prison_racct_free_locked(prr);
4278 	sx_xunlock(&allprison_lock);
4279 }
4280 
4281 static void
4282 prison_racct_attach(struct prison *pr)
4283 {
4284 	struct prison_racct *prr;
4285 
4286 	ASSERT_RACCT_ENABLED();
4287 	sx_assert(&allprison_lock, SA_XLOCKED);
4288 
4289 	prr = prison_racct_find_locked(pr->pr_name);
4290 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4291 
4292 	pr->pr_prison_racct = prr;
4293 }
4294 
4295 /*
4296  * Handle jail renaming.  From the racct point of view, renaming means
4297  * moving from one prison_racct to another.
4298  */
4299 static void
4300 prison_racct_modify(struct prison *pr)
4301 {
4302 #ifdef RCTL
4303 	struct proc *p;
4304 	struct ucred *cred;
4305 #endif
4306 	struct prison_racct *oldprr;
4307 
4308 	ASSERT_RACCT_ENABLED();
4309 
4310 	sx_slock(&allproc_lock);
4311 	sx_xlock(&allprison_lock);
4312 
4313 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4314 		sx_xunlock(&allprison_lock);
4315 		sx_sunlock(&allproc_lock);
4316 		return;
4317 	}
4318 
4319 	oldprr = pr->pr_prison_racct;
4320 	pr->pr_prison_racct = NULL;
4321 
4322 	prison_racct_attach(pr);
4323 
4324 	/*
4325 	 * Move resource utilisation records.
4326 	 */
4327 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4328 
4329 #ifdef RCTL
4330 	/*
4331 	 * Force rctl to reattach rules to processes.
4332 	 */
4333 	FOREACH_PROC_IN_SYSTEM(p) {
4334 		PROC_LOCK(p);
4335 		cred = crhold(p->p_ucred);
4336 		PROC_UNLOCK(p);
4337 		rctl_proc_ucred_changed(p, cred);
4338 		crfree(cred);
4339 	}
4340 #endif
4341 
4342 	sx_sunlock(&allproc_lock);
4343 	prison_racct_free_locked(oldprr);
4344 	sx_xunlock(&allprison_lock);
4345 }
4346 
4347 static void
4348 prison_racct_detach(struct prison *pr)
4349 {
4350 
4351 	ASSERT_RACCT_ENABLED();
4352 	sx_assert(&allprison_lock, SA_UNLOCKED);
4353 
4354 	if (pr->pr_prison_racct == NULL)
4355 		return;
4356 	prison_racct_free(pr->pr_prison_racct);
4357 	pr->pr_prison_racct = NULL;
4358 }
4359 #endif /* RACCT */
4360 
4361 #ifdef DDB
4362 
4363 static void
4364 db_show_prison(struct prison *pr)
4365 {
4366 	struct bool_flags *bf;
4367 	struct jailsys_flags *jsf;
4368 #if defined(INET) || defined(INET6)
4369 	int ii;
4370 #endif
4371 	unsigned f;
4372 #ifdef INET
4373 	char ip4buf[INET_ADDRSTRLEN];
4374 #endif
4375 #ifdef INET6
4376 	char ip6buf[INET6_ADDRSTRLEN];
4377 #endif
4378 
4379 	db_printf("prison %p:\n", pr);
4380 	db_printf(" jid             = %d\n", pr->pr_id);
4381 	db_printf(" name            = %s\n", pr->pr_name);
4382 	db_printf(" parent          = %p\n", pr->pr_parent);
4383 	db_printf(" ref             = %d\n", pr->pr_ref);
4384 	db_printf(" uref            = %d\n", pr->pr_uref);
4385 	db_printf(" path            = %s\n", pr->pr_path);
4386 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4387 	    ? pr->pr_cpuset->cs_id : -1);
4388 #ifdef VIMAGE
4389 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4390 #endif
4391 	db_printf(" root            = %p\n", pr->pr_root);
4392 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4393 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4394 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4395 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4396 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4397 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4398 	db_printf(" flags           = 0x%x", pr->pr_flags);
4399 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4400 		if (pr->pr_flags & bf->flag)
4401 			db_printf(" %s", bf->name);
4402 	for (jsf = pr_flag_jailsys;
4403 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4404 	     jsf++) {
4405 		f = pr->pr_flags & (jsf->disable | jsf->new);
4406 		db_printf(" %-16s= %s\n", jsf->name,
4407 		    (f != 0 && f == jsf->disable) ? "disable"
4408 		    : (f == jsf->new) ? "new"
4409 		    : "inherit");
4410 	}
4411 	db_printf(" allow           = 0x%x", pr->pr_allow);
4412 	for (bf = pr_flag_allow;
4413 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4414 		atomic_load_int(&bf->flag) != 0;
4415 	     bf++)
4416 		if (pr->pr_allow & bf->flag)
4417 			db_printf(" %s", bf->name);
4418 	db_printf("\n");
4419 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4420 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4421 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4422 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4423 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4424 #ifdef INET
4425 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4426 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4427 		db_printf(" %s %s\n",
4428 		    ii == 0 ? "ip4.addr        =" : "                 ",
4429 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4430 #endif
4431 #ifdef INET6
4432 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4433 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4434 		db_printf(" %s %s\n",
4435 		    ii == 0 ? "ip6.addr        =" : "                 ",
4436 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4437 #endif
4438 }
4439 
4440 DB_SHOW_COMMAND(prison, db_show_prison_command)
4441 {
4442 	struct prison *pr;
4443 
4444 	if (!have_addr) {
4445 		/*
4446 		 * Show all prisons in the list, and prison0 which is not
4447 		 * listed.
4448 		 */
4449 		db_show_prison(&prison0);
4450 		if (!db_pager_quit) {
4451 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4452 				db_show_prison(pr);
4453 				if (db_pager_quit)
4454 					break;
4455 			}
4456 		}
4457 		return;
4458 	}
4459 
4460 	if (addr == 0)
4461 		pr = &prison0;
4462 	else {
4463 		/* Look for a prison with the ID and with references. */
4464 		TAILQ_FOREACH(pr, &allprison, pr_list)
4465 			if (pr->pr_id == addr && pr->pr_ref > 0)
4466 				break;
4467 		if (pr == NULL)
4468 			/* Look again, without requiring a reference. */
4469 			TAILQ_FOREACH(pr, &allprison, pr_list)
4470 				if (pr->pr_id == addr)
4471 					break;
4472 		if (pr == NULL)
4473 			/* Assume address points to a valid prison. */
4474 			pr = (struct prison *)addr;
4475 	}
4476 	db_show_prison(pr);
4477 }
4478 
4479 #endif /* DDB */
4480