xref: /freebsd/sys/kern/kern_jail.c (revision c9504239e8e891292b751f9c498dc3b03234004b)
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/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/racct.h>
54 #include <sys/rctl.h>
55 #include <sys/refcount.h>
56 #include <sys/sx.h>
57 #include <sys/sysent.h>
58 #include <sys/namei.h>
59 #include <sys/mount.h>
60 #include <sys/queue.h>
61 #include <sys/socket.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 
66 #include <net/if.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif /* DDB */
74 
75 #include <security/mac/mac_framework.h>
76 
77 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
78 
79 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
80 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
81 
82 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
83 #ifdef INET
84 #ifdef INET6
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
86 #else
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
88 #endif
89 #else /* !INET */
90 #ifdef INET6
91 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
92 #else
93 #define	_PR_IP_SADDRSEL	0
94 #endif
95 #endif
96 
97 /* prison0 describes what is "real" about the system. */
98 struct prison prison0 = {
99 	.pr_id		= 0,
100 	.pr_name	= "0",
101 	.pr_ref		= 1,
102 	.pr_uref	= 1,
103 	.pr_path	= "/",
104 	.pr_securelevel	= -1,
105 	.pr_devfs_rsnum = 0,
106 	.pr_childmax	= JAIL_MAX,
107 	.pr_hostuuid	= DEFAULT_HOSTUUID,
108 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
109 #ifdef VIMAGE
110 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
111 #else
112 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
113 #endif
114 	.pr_allow	= PR_ALLOW_ALL_STATIC,
115 };
116 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
117 
118 struct bool_flags {
119 	const char	*name;
120 	const char	*noname;
121 	unsigned	 flag;
122 };
123 struct jailsys_flags {
124 	const char	*name;
125 	unsigned	 disable;
126 	unsigned	 new;
127 };
128 
129 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
130 struct	sx allprison_lock;
131 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
132 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
133 LIST_HEAD(, prison_racct) allprison_racct;
134 int	lastprid = 0;
135 
136 static int do_jail_attach(struct thread *td, struct prison *pr);
137 static void prison_complete(void *context, int pending);
138 static void prison_deref(struct prison *pr, int flags);
139 static char *prison_path(struct prison *pr1, struct prison *pr2);
140 static void prison_remove_one(struct prison *pr);
141 #ifdef RACCT
142 static void prison_racct_attach(struct prison *pr);
143 static void prison_racct_modify(struct prison *pr);
144 static void prison_racct_detach(struct prison *pr);
145 #endif
146 
147 /* Flags for prison_deref */
148 #define	PD_DEREF	0x01
149 #define	PD_DEUREF	0x02
150 #define	PD_LOCKED	0x04
151 #define	PD_LIST_SLOCKED	0x08
152 #define	PD_LIST_XLOCKED	0x10
153 
154 /*
155  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
156  * as we cannot figure out the size of a sparse array, or an array without a
157  * terminating entry.
158  */
159 static struct bool_flags pr_flag_bool[] = {
160 	{"persist", "nopersist", PR_PERSIST},
161 #ifdef INET
162 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
163 #endif
164 #ifdef INET6
165 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
166 #endif
167 };
168 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
169 
170 static struct jailsys_flags pr_flag_jailsys[] = {
171 	{"host", 0, PR_HOST},
172 #ifdef VIMAGE
173 	{"vnet", 0, PR_VNET},
174 #endif
175 #ifdef INET
176 	{"ip4", PR_IP4_USER, PR_IP4_USER},
177 #endif
178 #ifdef INET6
179 	{"ip6", PR_IP6_USER, PR_IP6_USER},
180 #endif
181 };
182 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
183 
184 /* Make this array full-size so dynamic parameters can be added. */
185 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
186 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
187 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
188 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
189 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
190 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
191 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
192 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
193 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
194 	{"allow.reserved_ports", "allow.noreserved_ports",
195 	 PR_ALLOW_RESERVED_PORTS},
196 };
197 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
198 
199 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | PR_ALLOW_RESERVED_PORTS)
200 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
201 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
202 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
203 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
204 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
205 #if defined(INET) || defined(INET6)
206 static unsigned jail_max_af_ips = 255;
207 #endif
208 
209 /*
210  * Initialize the parts of prison0 that can't be static-initialized with
211  * constants.  This is called from proc0_init() after creating thread0 cpuset.
212  */
213 void
214 prison0_init(void)
215 {
216 
217 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
218 	prison0.pr_osreldate = osreldate;
219 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
220 }
221 
222 /*
223  * struct jail_args {
224  *	struct jail *jail;
225  * };
226  */
227 int
228 sys_jail(struct thread *td, struct jail_args *uap)
229 {
230 	uint32_t version;
231 	int error;
232 	struct jail j;
233 
234 	error = copyin(uap->jail, &version, sizeof(uint32_t));
235 	if (error)
236 		return (error);
237 
238 	switch (version) {
239 	case 0:
240 	{
241 		struct jail_v0 j0;
242 
243 		/* FreeBSD single IPv4 jails. */
244 		bzero(&j, sizeof(struct jail));
245 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
246 		if (error)
247 			return (error);
248 		j.version = j0.version;
249 		j.path = j0.path;
250 		j.hostname = j0.hostname;
251 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
252 		break;
253 	}
254 
255 	case 1:
256 		/*
257 		 * Version 1 was used by multi-IPv4 jail implementations
258 		 * that never made it into the official kernel.
259 		 */
260 		return (EINVAL);
261 
262 	case 2:	/* JAIL_API_VERSION */
263 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
264 		error = copyin(uap->jail, &j, sizeof(struct jail));
265 		if (error)
266 			return (error);
267 		break;
268 
269 	default:
270 		/* Sci-Fi jails are not supported, sorry. */
271 		return (EINVAL);
272 	}
273 	return (kern_jail(td, &j));
274 }
275 
276 int
277 kern_jail(struct thread *td, struct jail *j)
278 {
279 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
280 #ifdef INET
281 			    + 1
282 #endif
283 #ifdef INET6
284 			    + 1
285 #endif
286 			    )];
287 	struct uio opt;
288 	char *u_path, *u_hostname, *u_name;
289 	struct bool_flags *bf;
290 #ifdef INET
291 	uint32_t ip4s;
292 	struct in_addr *u_ip4;
293 #endif
294 #ifdef INET6
295 	struct in6_addr *u_ip6;
296 #endif
297 	size_t tmplen;
298 	int error, enforce_statfs;
299 
300 	bzero(&optiov, sizeof(optiov));
301 	opt.uio_iov = optiov;
302 	opt.uio_iovcnt = 0;
303 	opt.uio_offset = -1;
304 	opt.uio_resid = -1;
305 	opt.uio_segflg = UIO_SYSSPACE;
306 	opt.uio_rw = UIO_READ;
307 	opt.uio_td = td;
308 
309 	/* Set permissions for top-level jails from sysctls. */
310 	if (!jailed(td->td_ucred)) {
311 		for (bf = pr_flag_allow;
312 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
313 			bf->flag != 0;
314 		     bf++) {
315 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
316 			    (jail_default_allow & bf->flag)
317 			    ? bf->name : bf->noname);
318 			optiov[opt.uio_iovcnt].iov_len =
319 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
320 			opt.uio_iovcnt += 2;
321 		}
322 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
323 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
324 		opt.uio_iovcnt++;
325 		enforce_statfs = jail_default_enforce_statfs;
326 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
327 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
328 		opt.uio_iovcnt++;
329 	}
330 
331 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
332 #ifdef INET
333 	ip4s = (j->version == 0) ? 1 : j->ip4s;
334 	if (ip4s > jail_max_af_ips)
335 		return (EINVAL);
336 	tmplen += ip4s * sizeof(struct in_addr);
337 #else
338 	if (j->ip4s > 0)
339 		return (EINVAL);
340 #endif
341 #ifdef INET6
342 	if (j->ip6s > jail_max_af_ips)
343 		return (EINVAL);
344 	tmplen += j->ip6s * sizeof(struct in6_addr);
345 #else
346 	if (j->ip6s > 0)
347 		return (EINVAL);
348 #endif
349 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
350 	u_hostname = u_path + MAXPATHLEN;
351 	u_name = u_hostname + MAXHOSTNAMELEN;
352 #ifdef INET
353 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
354 #endif
355 #ifdef INET6
356 #ifdef INET
357 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
358 #else
359 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
360 #endif
361 #endif
362 	optiov[opt.uio_iovcnt].iov_base = "path";
363 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
364 	opt.uio_iovcnt++;
365 	optiov[opt.uio_iovcnt].iov_base = u_path;
366 	error = copyinstr(j->path, u_path, MAXPATHLEN,
367 	    &optiov[opt.uio_iovcnt].iov_len);
368 	if (error) {
369 		free(u_path, M_TEMP);
370 		return (error);
371 	}
372 	opt.uio_iovcnt++;
373 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
374 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
375 	opt.uio_iovcnt++;
376 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
377 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
378 	    &optiov[opt.uio_iovcnt].iov_len);
379 	if (error) {
380 		free(u_path, M_TEMP);
381 		return (error);
382 	}
383 	opt.uio_iovcnt++;
384 	if (j->jailname != NULL) {
385 		optiov[opt.uio_iovcnt].iov_base = "name";
386 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
387 		opt.uio_iovcnt++;
388 		optiov[opt.uio_iovcnt].iov_base = u_name;
389 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
390 		    &optiov[opt.uio_iovcnt].iov_len);
391 		if (error) {
392 			free(u_path, M_TEMP);
393 			return (error);
394 		}
395 		opt.uio_iovcnt++;
396 	}
397 #ifdef INET
398 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
399 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
400 	opt.uio_iovcnt++;
401 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
402 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
403 	if (j->version == 0)
404 		u_ip4->s_addr = j->ip4s;
405 	else {
406 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
407 		if (error) {
408 			free(u_path, M_TEMP);
409 			return (error);
410 		}
411 	}
412 	opt.uio_iovcnt++;
413 #endif
414 #ifdef INET6
415 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
416 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
417 	opt.uio_iovcnt++;
418 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
419 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
420 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
421 	if (error) {
422 		free(u_path, M_TEMP);
423 		return (error);
424 	}
425 	opt.uio_iovcnt++;
426 #endif
427 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
428 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
429 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
430 	free(u_path, M_TEMP);
431 	return (error);
432 }
433 
434 
435 /*
436  * struct jail_set_args {
437  *	struct iovec *iovp;
438  *	unsigned int iovcnt;
439  *	int flags;
440  * };
441  */
442 int
443 sys_jail_set(struct thread *td, struct jail_set_args *uap)
444 {
445 	struct uio *auio;
446 	int error;
447 
448 	/* Check that we have an even number of iovecs. */
449 	if (uap->iovcnt & 1)
450 		return (EINVAL);
451 
452 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
453 	if (error)
454 		return (error);
455 	error = kern_jail_set(td, auio, uap->flags);
456 	free(auio, M_IOV);
457 	return (error);
458 }
459 
460 int
461 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
462 {
463 	struct nameidata nd;
464 #ifdef INET
465 	struct in_addr *ip4;
466 #endif
467 #ifdef INET6
468 	struct in6_addr *ip6;
469 #endif
470 	struct vfsopt *opt;
471 	struct vfsoptlist *opts;
472 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
473 	struct vnode *root;
474 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
475 	char *g_path, *osrelstr;
476 	struct bool_flags *bf;
477 	struct jailsys_flags *jsf;
478 #if defined(INET) || defined(INET6)
479 	struct prison *tppr;
480 	void *op;
481 #endif
482 	unsigned long hid;
483 	size_t namelen, onamelen, pnamelen;
484 	int born, created, cuflags, descend, enforce;
485 	int error, errmsg_len, errmsg_pos;
486 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
487 	int jid, jsys, len, level;
488 	int childmax, osreldt, rsnum, slevel;
489 	int fullpath_disabled;
490 #if defined(INET) || defined(INET6)
491 	int ii, ij;
492 #endif
493 #ifdef INET
494 	int ip4s, redo_ip4;
495 #endif
496 #ifdef INET6
497 	int ip6s, redo_ip6;
498 #endif
499 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
500 	unsigned tallow;
501 	char numbuf[12];
502 
503 	error = priv_check(td, PRIV_JAIL_SET);
504 	if (!error && (flags & JAIL_ATTACH))
505 		error = priv_check(td, PRIV_JAIL_ATTACH);
506 	if (error)
507 		return (error);
508 	mypr = td->td_ucred->cr_prison;
509 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
510 		return (EPERM);
511 	if (flags & ~JAIL_SET_MASK)
512 		return (EINVAL);
513 
514 	/*
515 	 * Check all the parameters before committing to anything.  Not all
516 	 * errors can be caught early, but we may as well try.  Also, this
517 	 * takes care of some expensive stuff (path lookup) before getting
518 	 * the allprison lock.
519 	 *
520 	 * XXX Jails are not filesystems, and jail parameters are not mount
521 	 *     options.  But it makes more sense to re-use the vfsopt code
522 	 *     than duplicate it under a different name.
523 	 */
524 	error = vfs_buildopts(optuio, &opts);
525 	if (error)
526 		return (error);
527 #ifdef INET
528 	ip4 = NULL;
529 #endif
530 #ifdef INET6
531 	ip6 = NULL;
532 #endif
533 	g_path = NULL;
534 
535 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
536 	if (!cuflags) {
537 		error = EINVAL;
538 		vfs_opterror(opts, "no valid operation (create or update)");
539 		goto done_errmsg;
540 	}
541 
542 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
543 	if (error == ENOENT)
544 		jid = 0;
545 	else if (error != 0)
546 		goto done_free;
547 
548 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
549 	if (error == ENOENT)
550 		gotslevel = 0;
551 	else if (error != 0)
552 		goto done_free;
553 	else
554 		gotslevel = 1;
555 
556 	error =
557 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
558 	if (error == ENOENT)
559 		gotchildmax = 0;
560 	else if (error != 0)
561 		goto done_free;
562 	else
563 		gotchildmax = 1;
564 
565 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
566 	if (error == ENOENT)
567 		gotenforce = 0;
568 	else if (error != 0)
569 		goto done_free;
570 	else if (enforce < 0 || enforce > 2) {
571 		error = EINVAL;
572 		goto done_free;
573 	} else
574 		gotenforce = 1;
575 
576 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
577 	if (error == ENOENT)
578 		gotrsnum = 0;
579 	else if (error != 0)
580 		goto done_free;
581 	else
582 		gotrsnum = 1;
583 
584 	pr_flags = ch_flags = 0;
585 	for (bf = pr_flag_bool;
586 	     bf < pr_flag_bool + nitems(pr_flag_bool);
587 	     bf++) {
588 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
589 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
590 	}
591 	ch_flags |= pr_flags;
592 	for (jsf = pr_flag_jailsys;
593 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
594 	     jsf++) {
595 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
596 		if (error == ENOENT)
597 			continue;
598 		if (error != 0)
599 			goto done_free;
600 		switch (jsys) {
601 		case JAIL_SYS_DISABLE:
602 			if (!jsf->disable) {
603 				error = EINVAL;
604 				goto done_free;
605 			}
606 			pr_flags |= jsf->disable;
607 			break;
608 		case JAIL_SYS_NEW:
609 			pr_flags |= jsf->new;
610 			break;
611 		case JAIL_SYS_INHERIT:
612 			break;
613 		default:
614 			error = EINVAL;
615 			goto done_free;
616 		}
617 		ch_flags |= jsf->new | jsf->disable;
618 	}
619 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
620 	    && !(pr_flags & PR_PERSIST)) {
621 		error = EINVAL;
622 		vfs_opterror(opts, "new jail must persist or attach");
623 		goto done_errmsg;
624 	}
625 #ifdef VIMAGE
626 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
627 		error = EINVAL;
628 		vfs_opterror(opts, "vnet cannot be changed after creation");
629 		goto done_errmsg;
630 	}
631 #endif
632 #ifdef INET
633 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
634 		error = EINVAL;
635 		vfs_opterror(opts, "ip4 cannot be changed after creation");
636 		goto done_errmsg;
637 	}
638 #endif
639 #ifdef INET6
640 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
641 		error = EINVAL;
642 		vfs_opterror(opts, "ip6 cannot be changed after creation");
643 		goto done_errmsg;
644 	}
645 #endif
646 
647 	pr_allow = ch_allow = 0;
648 	for (bf = pr_flag_allow;
649 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
650 	     bf++) {
651 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
652 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
653 	}
654 	ch_allow |= pr_allow;
655 
656 	error = vfs_getopt(opts, "name", (void **)&name, &len);
657 	if (error == ENOENT)
658 		name = NULL;
659 	else if (error != 0)
660 		goto done_free;
661 	else {
662 		if (len == 0 || name[len - 1] != '\0') {
663 			error = EINVAL;
664 			goto done_free;
665 		}
666 		if (len > MAXHOSTNAMELEN) {
667 			error = ENAMETOOLONG;
668 			goto done_free;
669 		}
670 	}
671 
672 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
673 	if (error == ENOENT)
674 		host = NULL;
675 	else if (error != 0)
676 		goto done_free;
677 	else {
678 		ch_flags |= PR_HOST;
679 		pr_flags |= PR_HOST;
680 		if (len == 0 || host[len - 1] != '\0') {
681 			error = EINVAL;
682 			goto done_free;
683 		}
684 		if (len > MAXHOSTNAMELEN) {
685 			error = ENAMETOOLONG;
686 			goto done_free;
687 		}
688 	}
689 
690 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
691 	if (error == ENOENT)
692 		domain = NULL;
693 	else if (error != 0)
694 		goto done_free;
695 	else {
696 		ch_flags |= PR_HOST;
697 		pr_flags |= PR_HOST;
698 		if (len == 0 || domain[len - 1] != '\0') {
699 			error = EINVAL;
700 			goto done_free;
701 		}
702 		if (len > MAXHOSTNAMELEN) {
703 			error = ENAMETOOLONG;
704 			goto done_free;
705 		}
706 	}
707 
708 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
709 	if (error == ENOENT)
710 		uuid = NULL;
711 	else if (error != 0)
712 		goto done_free;
713 	else {
714 		ch_flags |= PR_HOST;
715 		pr_flags |= PR_HOST;
716 		if (len == 0 || uuid[len - 1] != '\0') {
717 			error = EINVAL;
718 			goto done_free;
719 		}
720 		if (len > HOSTUUIDLEN) {
721 			error = ENAMETOOLONG;
722 			goto done_free;
723 		}
724 	}
725 
726 #ifdef COMPAT_FREEBSD32
727 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
728 		uint32_t hid32;
729 
730 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
731 		hid = hid32;
732 	} else
733 #endif
734 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
735 	if (error == ENOENT)
736 		gothid = 0;
737 	else if (error != 0)
738 		goto done_free;
739 	else {
740 		gothid = 1;
741 		ch_flags |= PR_HOST;
742 		pr_flags |= PR_HOST;
743 	}
744 
745 #ifdef INET
746 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
747 	if (error == ENOENT)
748 		ip4s = 0;
749 	else if (error != 0)
750 		goto done_free;
751 	else if (ip4s & (sizeof(*ip4) - 1)) {
752 		error = EINVAL;
753 		goto done_free;
754 	} else {
755 		ch_flags |= PR_IP4_USER;
756 		pr_flags |= PR_IP4_USER;
757 		if (ip4s > 0) {
758 			ip4s /= sizeof(*ip4);
759 			if (ip4s > jail_max_af_ips) {
760 				error = EINVAL;
761 				vfs_opterror(opts, "too many IPv4 addresses");
762 				goto done_errmsg;
763 			}
764 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
765 			bcopy(op, ip4, ip4s * sizeof(*ip4));
766 			/*
767 			 * IP addresses are all sorted but ip[0] to preserve
768 			 * the primary IP address as given from userland.
769 			 * This special IP is used for unbound outgoing
770 			 * connections as well for "loopback" traffic in case
771 			 * source address selection cannot find any more fitting
772 			 * address to connect from.
773 			 */
774 			if (ip4s > 1)
775 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
776 				    prison_qcmp_v4);
777 			/*
778 			 * Check for duplicate addresses and do some simple
779 			 * zero and broadcast checks. If users give other bogus
780 			 * addresses it is their problem.
781 			 *
782 			 * We do not have to care about byte order for these
783 			 * checks so we will do them in NBO.
784 			 */
785 			for (ii = 0; ii < ip4s; ii++) {
786 				if (ip4[ii].s_addr == INADDR_ANY ||
787 				    ip4[ii].s_addr == INADDR_BROADCAST) {
788 					error = EINVAL;
789 					goto done_free;
790 				}
791 				if ((ii+1) < ip4s &&
792 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
793 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
794 					error = EINVAL;
795 					goto done_free;
796 				}
797 			}
798 		}
799 	}
800 #endif
801 
802 #ifdef INET6
803 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
804 	if (error == ENOENT)
805 		ip6s = 0;
806 	else if (error != 0)
807 		goto done_free;
808 	else if (ip6s & (sizeof(*ip6) - 1)) {
809 		error = EINVAL;
810 		goto done_free;
811 	} else {
812 		ch_flags |= PR_IP6_USER;
813 		pr_flags |= PR_IP6_USER;
814 		if (ip6s > 0) {
815 			ip6s /= sizeof(*ip6);
816 			if (ip6s > jail_max_af_ips) {
817 				error = EINVAL;
818 				vfs_opterror(opts, "too many IPv6 addresses");
819 				goto done_errmsg;
820 			}
821 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
822 			bcopy(op, ip6, ip6s * sizeof(*ip6));
823 			if (ip6s > 1)
824 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
825 				    prison_qcmp_v6);
826 			for (ii = 0; ii < ip6s; ii++) {
827 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
828 					error = EINVAL;
829 					goto done_free;
830 				}
831 				if ((ii+1) < ip6s &&
832 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
833 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
834 				{
835 					error = EINVAL;
836 					goto done_free;
837 				}
838 			}
839 		}
840 	}
841 #endif
842 
843 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
844 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
845 		error = EINVAL;
846 		vfs_opterror(opts,
847 		    "vnet jails cannot have IP address restrictions");
848 		goto done_errmsg;
849 	}
850 #endif
851 
852 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
853 	if (error == ENOENT)
854 		osrelstr = NULL;
855 	else if (error != 0)
856 		goto done_free;
857 	else {
858 		if (flags & JAIL_UPDATE) {
859 			error = EINVAL;
860 			vfs_opterror(opts,
861 			    "osrelease cannot be changed after creation");
862 			goto done_errmsg;
863 		}
864 		if (len == 0 || len >= OSRELEASELEN) {
865 			error = EINVAL;
866 			vfs_opterror(opts,
867 			    "osrelease string must be 1-%d bytes long",
868 			    OSRELEASELEN - 1);
869 			goto done_errmsg;
870 		}
871 	}
872 
873 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
874 	if (error == ENOENT)
875 		osreldt = 0;
876 	else if (error != 0)
877 		goto done_free;
878 	else {
879 		if (flags & JAIL_UPDATE) {
880 			error = EINVAL;
881 			vfs_opterror(opts,
882 			    "osreldate cannot be changed after creation");
883 			goto done_errmsg;
884 		}
885 		if (osreldt == 0) {
886 			error = EINVAL;
887 			vfs_opterror(opts, "osreldate cannot be 0");
888 			goto done_errmsg;
889 		}
890 	}
891 
892 	fullpath_disabled = 0;
893 	root = NULL;
894 	error = vfs_getopt(opts, "path", (void **)&path, &len);
895 	if (error == ENOENT)
896 		path = NULL;
897 	else if (error != 0)
898 		goto done_free;
899 	else {
900 		if (flags & JAIL_UPDATE) {
901 			error = EINVAL;
902 			vfs_opterror(opts,
903 			    "path cannot be changed after creation");
904 			goto done_errmsg;
905 		}
906 		if (len == 0 || path[len - 1] != '\0') {
907 			error = EINVAL;
908 			goto done_free;
909 		}
910 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
911 		    path, td);
912 		error = namei(&nd);
913 		if (error)
914 			goto done_free;
915 		root = nd.ni_vp;
916 		NDFREE(&nd, NDF_ONLY_PNBUF);
917 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
918 		strlcpy(g_path, path, MAXPATHLEN);
919 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
920 		if (error == 0)
921 			path = g_path;
922 		else if (error == ENODEV) {
923 			/* proceed if sysctl debug.disablefullpath == 1 */
924 			fullpath_disabled = 1;
925 			if (len < 2 || (len == 2 && path[0] == '/'))
926 				path = NULL;
927 		} else {
928 			/* exit on other errors */
929 			goto done_free;
930 		}
931 		if (root->v_type != VDIR) {
932 			error = ENOTDIR;
933 			vput(root);
934 			goto done_free;
935 		}
936 		VOP_UNLOCK(root, 0);
937 		if (fullpath_disabled) {
938 			/* Leave room for a real-root full pathname. */
939 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
940 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
941 				error = ENAMETOOLONG;
942 				vrele(root);
943 				goto done_free;
944 			}
945 		}
946 	}
947 
948 	/*
949 	 * Find the specified jail, or at least its parent.
950 	 * This abuses the file error codes ENOENT and EEXIST.
951 	 */
952 	pr = NULL;
953 	ppr = mypr;
954 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
955 		namelc = strrchr(name, '.');
956 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
957 		if (*p != '\0')
958 			jid = 0;
959 	}
960 	sx_xlock(&allprison_lock);
961 	if (jid != 0) {
962 		/*
963 		 * See if a requested jid already exists.  There is an
964 		 * information leak here if the jid exists but is not within
965 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
966 		 * even though they cannot see the jail, and CREATE | UPDATE
967 		 * will return ENOENT which is not normally a valid error.
968 		 */
969 		if (jid < 0) {
970 			error = EINVAL;
971 			vfs_opterror(opts, "negative jid");
972 			goto done_unlock_list;
973 		}
974 		pr = prison_find(jid);
975 		if (pr != NULL) {
976 			ppr = pr->pr_parent;
977 			/* Create: jid must not exist. */
978 			if (cuflags == JAIL_CREATE) {
979 				mtx_unlock(&pr->pr_mtx);
980 				error = EEXIST;
981 				vfs_opterror(opts, "jail %d already exists",
982 				    jid);
983 				goto done_unlock_list;
984 			}
985 			if (!prison_ischild(mypr, pr)) {
986 				mtx_unlock(&pr->pr_mtx);
987 				pr = NULL;
988 			} else if (pr->pr_uref == 0) {
989 				if (!(flags & JAIL_DYING)) {
990 					mtx_unlock(&pr->pr_mtx);
991 					error = ENOENT;
992 					vfs_opterror(opts, "jail %d is dying",
993 					    jid);
994 					goto done_unlock_list;
995 				} else if ((flags & JAIL_ATTACH) ||
996 				    (pr_flags & PR_PERSIST)) {
997 					/*
998 					 * A dying jail might be resurrected
999 					 * (via attach or persist), but first
1000 					 * it must determine if another jail
1001 					 * has claimed its name.  Accomplish
1002 					 * this by implicitly re-setting the
1003 					 * name.
1004 					 */
1005 					if (name == NULL)
1006 						name = prison_name(mypr, pr);
1007 				}
1008 			}
1009 		}
1010 		if (pr == NULL) {
1011 			/* Update: jid must exist. */
1012 			if (cuflags == JAIL_UPDATE) {
1013 				error = ENOENT;
1014 				vfs_opterror(opts, "jail %d not found", jid);
1015 				goto done_unlock_list;
1016 			}
1017 		}
1018 	}
1019 	/*
1020 	 * If the caller provided a name, look for a jail by that name.
1021 	 * This has different semantics for creates and updates keyed by jid
1022 	 * (where the name must not already exist in a different jail),
1023 	 * and updates keyed by the name itself (where the name must exist
1024 	 * because that is the jail being updated).
1025 	 */
1026 	namelc = NULL;
1027 	if (name != NULL) {
1028 		namelc = strrchr(name, '.');
1029 		if (namelc == NULL)
1030 			namelc = name;
1031 		else {
1032 			/*
1033 			 * This is a hierarchical name.  Split it into the
1034 			 * parent and child names, and make sure the parent
1035 			 * exists or matches an already found jail.
1036 			 */
1037 			if (pr != NULL) {
1038 				if (strncmp(name, ppr->pr_name, namelc - name)
1039 				    || ppr->pr_name[namelc - name] != '\0') {
1040 					mtx_unlock(&pr->pr_mtx);
1041 					error = EINVAL;
1042 					vfs_opterror(opts,
1043 					    "cannot change jail's parent");
1044 					goto done_unlock_list;
1045 				}
1046 			} else {
1047 				*namelc = '\0';
1048 				ppr = prison_find_name(mypr, name);
1049 				if (ppr == NULL) {
1050 					error = ENOENT;
1051 					vfs_opterror(opts,
1052 					    "jail \"%s\" not found", name);
1053 					goto done_unlock_list;
1054 				}
1055 				mtx_unlock(&ppr->pr_mtx);
1056 				*namelc = '.';
1057 			}
1058 			namelc++;
1059 		}
1060 		if (namelc[0] != '\0') {
1061 			pnamelen =
1062 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1063  name_again:
1064 			deadpr = NULL;
1065 			FOREACH_PRISON_CHILD(ppr, tpr) {
1066 				if (tpr != pr && tpr->pr_ref > 0 &&
1067 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1068 					if (pr == NULL &&
1069 					    cuflags != JAIL_CREATE) {
1070 						mtx_lock(&tpr->pr_mtx);
1071 						if (tpr->pr_ref > 0) {
1072 							/*
1073 							 * Use this jail
1074 							 * for updates.
1075 							 */
1076 							if (tpr->pr_uref > 0) {
1077 								pr = tpr;
1078 								break;
1079 							}
1080 							deadpr = tpr;
1081 						}
1082 						mtx_unlock(&tpr->pr_mtx);
1083 					} else if (tpr->pr_uref > 0) {
1084 						/*
1085 						 * Create, or update(jid):
1086 						 * name must not exist in an
1087 						 * active sibling jail.
1088 						 */
1089 						error = EEXIST;
1090 						if (pr != NULL)
1091 							mtx_unlock(&pr->pr_mtx);
1092 						vfs_opterror(opts,
1093 						   "jail \"%s\" already exists",
1094 						   name);
1095 						goto done_unlock_list;
1096 					}
1097 				}
1098 			}
1099 			/* If no active jail is found, use a dying one. */
1100 			if (deadpr != NULL && pr == NULL) {
1101 				if (flags & JAIL_DYING) {
1102 					mtx_lock(&deadpr->pr_mtx);
1103 					if (deadpr->pr_ref == 0) {
1104 						mtx_unlock(&deadpr->pr_mtx);
1105 						goto name_again;
1106 					}
1107 					pr = deadpr;
1108 				} else if (cuflags == JAIL_UPDATE) {
1109 					error = ENOENT;
1110 					vfs_opterror(opts,
1111 					    "jail \"%s\" is dying", name);
1112 					goto done_unlock_list;
1113 				}
1114 			}
1115 			/* Update: name must exist if no jid. */
1116 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1117 				error = ENOENT;
1118 				vfs_opterror(opts, "jail \"%s\" not found",
1119 				    name);
1120 				goto done_unlock_list;
1121 			}
1122 		}
1123 	}
1124 	/* Update: must provide a jid or name. */
1125 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1126 		error = ENOENT;
1127 		vfs_opterror(opts, "update specified no jail");
1128 		goto done_unlock_list;
1129 	}
1130 
1131 	/* If there's no prison to update, create a new one and link it in. */
1132 	if (pr == NULL) {
1133 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1134 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1135 				error = EPERM;
1136 				vfs_opterror(opts, "prison limit exceeded");
1137 				goto done_unlock_list;
1138 			}
1139 		created = 1;
1140 		mtx_lock(&ppr->pr_mtx);
1141 		if (ppr->pr_ref == 0) {
1142 			mtx_unlock(&ppr->pr_mtx);
1143 			error = ENOENT;
1144 			vfs_opterror(opts, "jail \"%s\" not found",
1145 			    prison_name(mypr, ppr));
1146 			goto done_unlock_list;
1147 		}
1148 		ppr->pr_ref++;
1149 		ppr->pr_uref++;
1150 		mtx_unlock(&ppr->pr_mtx);
1151 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1152 		if (jid == 0) {
1153 			/* Find the next free jid. */
1154 			jid = lastprid + 1;
1155  findnext:
1156 			if (jid == JAIL_MAX)
1157 				jid = 1;
1158 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1159 				if (tpr->pr_id < jid)
1160 					continue;
1161 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1162 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1163 					break;
1164 				}
1165 				if (jid == lastprid) {
1166 					error = EAGAIN;
1167 					vfs_opterror(opts,
1168 					    "no available jail IDs");
1169 					free(pr, M_PRISON);
1170 					prison_deref(ppr, PD_DEREF |
1171 					    PD_DEUREF | PD_LIST_XLOCKED);
1172 					goto done_releroot;
1173 				}
1174 				jid++;
1175 				goto findnext;
1176 			}
1177 			lastprid = jid;
1178 		} else {
1179 			/*
1180 			 * The jail already has a jid (that did not yet exist),
1181 			 * so just find where to insert it.
1182 			 */
1183 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1184 				if (tpr->pr_id >= jid) {
1185 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1186 					break;
1187 				}
1188 		}
1189 		if (tpr == NULL)
1190 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1191 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1192 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1193 			tpr->pr_childcount++;
1194 
1195 		pr->pr_parent = ppr;
1196 		pr->pr_id = jid;
1197 
1198 		/* Set some default values, and inherit some from the parent. */
1199 		if (namelc == NULL)
1200 			namelc = "";
1201 		if (path == NULL) {
1202 			path = "/";
1203 			root = mypr->pr_root;
1204 			vref(root);
1205 		}
1206 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1207 		pr->pr_flags |= PR_HOST;
1208 #if defined(INET) || defined(INET6)
1209 #ifdef VIMAGE
1210 		if (!(pr_flags & PR_VNET))
1211 #endif
1212 		{
1213 #ifdef INET
1214 			if (!(ch_flags & PR_IP4_USER))
1215 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1216 			else if (!(pr_flags & PR_IP4_USER)) {
1217 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1218 				if (ppr->pr_ip4 != NULL) {
1219 					pr->pr_ip4s = ppr->pr_ip4s;
1220 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1221 					    sizeof(struct in_addr), M_PRISON,
1222 					    M_WAITOK);
1223 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1224 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1225 				}
1226 			}
1227 #endif
1228 #ifdef INET6
1229 			if (!(ch_flags & PR_IP6_USER))
1230 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1231 			else if (!(pr_flags & PR_IP6_USER)) {
1232 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1233 				if (ppr->pr_ip6 != NULL) {
1234 					pr->pr_ip6s = ppr->pr_ip6s;
1235 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1236 					    sizeof(struct in6_addr), M_PRISON,
1237 					    M_WAITOK);
1238 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1239 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1240 				}
1241 			}
1242 #endif
1243 		}
1244 #endif
1245 		/* Source address selection is always on by default. */
1246 		pr->pr_flags |= _PR_IP_SADDRSEL;
1247 
1248 		pr->pr_securelevel = ppr->pr_securelevel;
1249 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1250 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1251 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1252 
1253 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1254 		if (osrelstr == NULL)
1255 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1256 		else
1257 		    strcpy(pr->pr_osrelease, osrelstr);
1258 
1259 		LIST_INIT(&pr->pr_children);
1260 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1261 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1262 
1263 #ifdef VIMAGE
1264 		/* Allocate a new vnet if specified. */
1265 		pr->pr_vnet = (pr_flags & PR_VNET)
1266 		    ? vnet_alloc() : ppr->pr_vnet;
1267 #endif
1268 		/*
1269 		 * Allocate a dedicated cpuset for each jail.
1270 		 * Unlike other initial settings, this may return an erorr.
1271 		 */
1272 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1273 		if (error) {
1274 			prison_deref(pr, PD_LIST_XLOCKED);
1275 			goto done_releroot;
1276 		}
1277 
1278 		mtx_lock(&pr->pr_mtx);
1279 		/*
1280 		 * New prisons do not yet have a reference, because we do not
1281 		 * want others to see the incomplete prison once the
1282 		 * allprison_lock is downgraded.
1283 		 */
1284 	} else {
1285 		created = 0;
1286 		/*
1287 		 * Grab a reference for existing prisons, to ensure they
1288 		 * continue to exist for the duration of the call.
1289 		 */
1290 		pr->pr_ref++;
1291 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1292 		if ((pr->pr_flags & PR_VNET) &&
1293 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1294 			error = EINVAL;
1295 			vfs_opterror(opts,
1296 			    "vnet jails cannot have IP address restrictions");
1297 			goto done_deref_locked;
1298 		}
1299 #endif
1300 #ifdef INET
1301 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1302 			error = EINVAL;
1303 			vfs_opterror(opts,
1304 			    "ip4 cannot be changed after creation");
1305 			goto done_deref_locked;
1306 		}
1307 #endif
1308 #ifdef INET6
1309 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1310 			error = EINVAL;
1311 			vfs_opterror(opts,
1312 			    "ip6 cannot be changed after creation");
1313 			goto done_deref_locked;
1314 		}
1315 #endif
1316 	}
1317 
1318 	/* Do final error checking before setting anything. */
1319 	if (gotslevel) {
1320 		if (slevel < ppr->pr_securelevel) {
1321 			error = EPERM;
1322 			goto done_deref_locked;
1323 		}
1324 	}
1325 	if (gotchildmax) {
1326 		if (childmax >= ppr->pr_childmax) {
1327 			error = EPERM;
1328 			goto done_deref_locked;
1329 		}
1330 	}
1331 	if (gotenforce) {
1332 		if (enforce < ppr->pr_enforce_statfs) {
1333 			error = EPERM;
1334 			goto done_deref_locked;
1335 		}
1336 	}
1337 	if (gotrsnum) {
1338 		/*
1339 		 * devfs_rsnum is a uint16_t
1340 		 */
1341 		if (rsnum < 0 || rsnum > 65535) {
1342 			error = EINVAL;
1343 			goto done_deref_locked;
1344 		}
1345 		/*
1346 		 * Nested jails always inherit parent's devfs ruleset
1347 		 */
1348 		if (jailed(td->td_ucred)) {
1349 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1350 				error = EPERM;
1351 				goto done_deref_locked;
1352 			} else
1353 				rsnum = ppr->pr_devfs_rsnum;
1354 		}
1355 	}
1356 #ifdef INET
1357 	if (ip4s > 0) {
1358 		if (ppr->pr_flags & PR_IP4) {
1359 			/*
1360 			 * Make sure the new set of IP addresses is a
1361 			 * subset of the parent's list.  Don't worry
1362 			 * about the parent being unlocked, as any
1363 			 * setting is done with allprison_lock held.
1364 			 */
1365 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1366 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1367 					break;
1368 			if (ij == ppr->pr_ip4s) {
1369 				error = EPERM;
1370 				goto done_deref_locked;
1371 			}
1372 			if (ip4s > 1) {
1373 				for (ii = ij = 1; ii < ip4s; ii++) {
1374 					if (ip4[ii].s_addr ==
1375 					    ppr->pr_ip4[0].s_addr)
1376 						continue;
1377 					for (; ij < ppr->pr_ip4s; ij++)
1378 						if (ip4[ii].s_addr ==
1379 						    ppr->pr_ip4[ij].s_addr)
1380 							break;
1381 					if (ij == ppr->pr_ip4s)
1382 						break;
1383 				}
1384 				if (ij == ppr->pr_ip4s) {
1385 					error = EPERM;
1386 					goto done_deref_locked;
1387 				}
1388 			}
1389 		}
1390 		/*
1391 		 * Check for conflicting IP addresses.  We permit them
1392 		 * if there is no more than one IP on each jail.  If
1393 		 * there is a duplicate on a jail with more than one
1394 		 * IP stop checking and return error.
1395 		 */
1396 #ifdef VIMAGE
1397 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1398 			if (tppr->pr_flags & PR_VNET)
1399 				break;
1400 #else
1401 		tppr = &prison0;
1402 #endif
1403 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1404 			if (tpr == pr ||
1405 #ifdef VIMAGE
1406 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1407 #endif
1408 			    tpr->pr_uref == 0) {
1409 				descend = 0;
1410 				continue;
1411 			}
1412 			if (!(tpr->pr_flags & PR_IP4_USER))
1413 				continue;
1414 			descend = 0;
1415 			if (tpr->pr_ip4 == NULL ||
1416 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1417 				continue;
1418 			for (ii = 0; ii < ip4s; ii++) {
1419 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1420 				    0) {
1421 					error = EADDRINUSE;
1422 					vfs_opterror(opts,
1423 					    "IPv4 addresses clash");
1424 					goto done_deref_locked;
1425 				}
1426 			}
1427 		}
1428 	}
1429 #endif
1430 #ifdef INET6
1431 	if (ip6s > 0) {
1432 		if (ppr->pr_flags & PR_IP6) {
1433 			/*
1434 			 * Make sure the new set of IP addresses is a
1435 			 * subset of the parent's list.
1436 			 */
1437 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1438 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1439 				    &ppr->pr_ip6[ij]))
1440 					break;
1441 			if (ij == ppr->pr_ip6s) {
1442 				error = EPERM;
1443 				goto done_deref_locked;
1444 			}
1445 			if (ip6s > 1) {
1446 				for (ii = ij = 1; ii < ip6s; ii++) {
1447 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1448 					     &ppr->pr_ip6[0]))
1449 						continue;
1450 					for (; ij < ppr->pr_ip6s; ij++)
1451 						if (IN6_ARE_ADDR_EQUAL(
1452 						    &ip6[ii], &ppr->pr_ip6[ij]))
1453 							break;
1454 					if (ij == ppr->pr_ip6s)
1455 						break;
1456 				}
1457 				if (ij == ppr->pr_ip6s) {
1458 					error = EPERM;
1459 					goto done_deref_locked;
1460 				}
1461 			}
1462 		}
1463 		/* Check for conflicting IP addresses. */
1464 #ifdef VIMAGE
1465 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1466 			if (tppr->pr_flags & PR_VNET)
1467 				break;
1468 #else
1469 		tppr = &prison0;
1470 #endif
1471 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1472 			if (tpr == pr ||
1473 #ifdef VIMAGE
1474 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1475 #endif
1476 			    tpr->pr_uref == 0) {
1477 				descend = 0;
1478 				continue;
1479 			}
1480 			if (!(tpr->pr_flags & PR_IP6_USER))
1481 				continue;
1482 			descend = 0;
1483 			if (tpr->pr_ip6 == NULL ||
1484 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1485 				continue;
1486 			for (ii = 0; ii < ip6s; ii++) {
1487 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1488 				    0) {
1489 					error = EADDRINUSE;
1490 					vfs_opterror(opts,
1491 					    "IPv6 addresses clash");
1492 					goto done_deref_locked;
1493 				}
1494 			}
1495 		}
1496 	}
1497 #endif
1498 	onamelen = namelen = 0;
1499 	if (namelc != NULL) {
1500 		/* Give a default name of the jid.  Also allow the name to be
1501 		 * explicitly the jid - but not any other number, and only in
1502 		 * normal form (no leading zero/etc).
1503 		 */
1504 		if (namelc[0] == '\0')
1505 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1506 		else if ((strtoul(namelc, &p, 10) != jid ||
1507 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1508 			error = EINVAL;
1509 			vfs_opterror(opts,
1510 			    "name cannot be numeric (unless it is the jid)");
1511 			goto done_deref_locked;
1512 		}
1513 		/*
1514 		 * Make sure the name isn't too long for the prison or its
1515 		 * children.
1516 		 */
1517 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1518 		onamelen = strlen(pr->pr_name + pnamelen);
1519 		namelen = strlen(namelc);
1520 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1521 			error = ENAMETOOLONG;
1522 			goto done_deref_locked;
1523 		}
1524 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1525 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1526 			    sizeof(pr->pr_name)) {
1527 				error = ENAMETOOLONG;
1528 				goto done_deref_locked;
1529 			}
1530 		}
1531 	}
1532 	if (pr_allow & ~ppr->pr_allow) {
1533 		error = EPERM;
1534 		goto done_deref_locked;
1535 	}
1536 
1537 	/*
1538 	 * Let modules check their parameters.  This requires unlocking and
1539 	 * then re-locking the prison, but this is still a valid state as long
1540 	 * as allprison_lock remains xlocked.
1541 	 */
1542 	mtx_unlock(&pr->pr_mtx);
1543 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1544 	if (error != 0) {
1545 		prison_deref(pr, created
1546 		    ? PD_LIST_XLOCKED
1547 		    : PD_DEREF | PD_LIST_XLOCKED);
1548 		goto done_releroot;
1549 	}
1550 	mtx_lock(&pr->pr_mtx);
1551 
1552 	/* At this point, all valid parameters should have been noted. */
1553 	TAILQ_FOREACH(opt, opts, link) {
1554 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1555 			error = EINVAL;
1556 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1557 			goto done_deref_locked;
1558 		}
1559 	}
1560 
1561 	/* Set the parameters of the prison. */
1562 #ifdef INET
1563 	redo_ip4 = 0;
1564 	if (pr_flags & PR_IP4_USER) {
1565 		pr->pr_flags |= PR_IP4;
1566 		free(pr->pr_ip4, M_PRISON);
1567 		pr->pr_ip4s = ip4s;
1568 		pr->pr_ip4 = ip4;
1569 		ip4 = NULL;
1570 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1571 #ifdef VIMAGE
1572 			if (tpr->pr_flags & PR_VNET) {
1573 				descend = 0;
1574 				continue;
1575 			}
1576 #endif
1577 			if (prison_restrict_ip4(tpr, NULL)) {
1578 				redo_ip4 = 1;
1579 				descend = 0;
1580 			}
1581 		}
1582 	}
1583 #endif
1584 #ifdef INET6
1585 	redo_ip6 = 0;
1586 	if (pr_flags & PR_IP6_USER) {
1587 		pr->pr_flags |= PR_IP6;
1588 		free(pr->pr_ip6, M_PRISON);
1589 		pr->pr_ip6s = ip6s;
1590 		pr->pr_ip6 = ip6;
1591 		ip6 = NULL;
1592 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1593 #ifdef VIMAGE
1594 			if (tpr->pr_flags & PR_VNET) {
1595 				descend = 0;
1596 				continue;
1597 			}
1598 #endif
1599 			if (prison_restrict_ip6(tpr, NULL)) {
1600 				redo_ip6 = 1;
1601 				descend = 0;
1602 			}
1603 		}
1604 	}
1605 #endif
1606 	if (gotslevel) {
1607 		pr->pr_securelevel = slevel;
1608 		/* Set all child jails to be at least this level. */
1609 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1610 			if (tpr->pr_securelevel < slevel)
1611 				tpr->pr_securelevel = slevel;
1612 	}
1613 	if (gotchildmax) {
1614 		pr->pr_childmax = childmax;
1615 		/* Set all child jails to under this limit. */
1616 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1617 			if (tpr->pr_childmax > childmax - level)
1618 				tpr->pr_childmax = childmax > level
1619 				    ? childmax - level : 0;
1620 	}
1621 	if (gotenforce) {
1622 		pr->pr_enforce_statfs = enforce;
1623 		/* Pass this restriction on to the children. */
1624 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1625 			if (tpr->pr_enforce_statfs < enforce)
1626 				tpr->pr_enforce_statfs = enforce;
1627 	}
1628 	if (gotrsnum) {
1629 		pr->pr_devfs_rsnum = rsnum;
1630 		/* Pass this restriction on to the children. */
1631 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1632 			tpr->pr_devfs_rsnum = rsnum;
1633 	}
1634 	if (namelc != NULL) {
1635 		if (ppr == &prison0)
1636 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1637 		else
1638 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1639 			    ppr->pr_name, namelc);
1640 		/* Change this component of child names. */
1641 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1642 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1643 			    strlen(tpr->pr_name + onamelen) + 1);
1644 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1645 		}
1646 	}
1647 	if (path != NULL) {
1648 		/* Try to keep a real-rooted full pathname. */
1649 		if (fullpath_disabled && path[0] == '/' &&
1650 		    strcmp(mypr->pr_path, "/"))
1651 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1652 			    mypr->pr_path, path);
1653 		else
1654 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1655 		pr->pr_root = root;
1656 	}
1657 	if (PR_HOST & ch_flags & ~pr_flags) {
1658 		if (pr->pr_flags & PR_HOST) {
1659 			/*
1660 			 * Copy the parent's host info.  As with pr_ip4 above,
1661 			 * the lack of a lock on the parent is not a problem;
1662 			 * it is always set with allprison_lock at least
1663 			 * shared, and is held exclusively here.
1664 			 */
1665 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1666 			    sizeof(pr->pr_hostname));
1667 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1668 			    sizeof(pr->pr_domainname));
1669 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1670 			    sizeof(pr->pr_hostuuid));
1671 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1672 		}
1673 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1674 		/* Set this prison, and any descendants without PR_HOST. */
1675 		if (host != NULL)
1676 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1677 		if (domain != NULL)
1678 			strlcpy(pr->pr_domainname, domain,
1679 			    sizeof(pr->pr_domainname));
1680 		if (uuid != NULL)
1681 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1682 		if (gothid)
1683 			pr->pr_hostid = hid;
1684 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1685 			if (tpr->pr_flags & PR_HOST)
1686 				descend = 0;
1687 			else {
1688 				if (host != NULL)
1689 					strlcpy(tpr->pr_hostname,
1690 					    pr->pr_hostname,
1691 					    sizeof(tpr->pr_hostname));
1692 				if (domain != NULL)
1693 					strlcpy(tpr->pr_domainname,
1694 					    pr->pr_domainname,
1695 					    sizeof(tpr->pr_domainname));
1696 				if (uuid != NULL)
1697 					strlcpy(tpr->pr_hostuuid,
1698 					    pr->pr_hostuuid,
1699 					    sizeof(tpr->pr_hostuuid));
1700 				if (gothid)
1701 					tpr->pr_hostid = hid;
1702 			}
1703 		}
1704 	}
1705 	if ((tallow = ch_allow & ~pr_allow)) {
1706 		/* Clear allow bits in all children. */
1707 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1708 			tpr->pr_allow &= ~tallow;
1709 	}
1710 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1711 	/*
1712 	 * Persistent prisons get an extra reference, and prisons losing their
1713 	 * persist flag lose that reference.  Only do this for existing prisons
1714 	 * for now, so new ones will remain unseen until after the module
1715 	 * handlers have completed.
1716 	 */
1717 	born = pr->pr_uref == 0;
1718 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1719 		if (pr_flags & PR_PERSIST) {
1720 			pr->pr_ref++;
1721 			pr->pr_uref++;
1722 		} else {
1723 			pr->pr_ref--;
1724 			pr->pr_uref--;
1725 		}
1726 	}
1727 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1728 	mtx_unlock(&pr->pr_mtx);
1729 
1730 #ifdef RACCT
1731 	if (racct_enable && created)
1732 		prison_racct_attach(pr);
1733 #endif
1734 
1735 	/* Locks may have prevented a complete restriction of child IP
1736 	 * addresses.  If so, allocate some more memory and try again.
1737 	 */
1738 #ifdef INET
1739 	while (redo_ip4) {
1740 		ip4s = pr->pr_ip4s;
1741 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1742 		mtx_lock(&pr->pr_mtx);
1743 		redo_ip4 = 0;
1744 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1745 #ifdef VIMAGE
1746 			if (tpr->pr_flags & PR_VNET) {
1747 				descend = 0;
1748 				continue;
1749 			}
1750 #endif
1751 			if (prison_restrict_ip4(tpr, ip4)) {
1752 				if (ip4 != NULL)
1753 					ip4 = NULL;
1754 				else
1755 					redo_ip4 = 1;
1756 			}
1757 		}
1758 		mtx_unlock(&pr->pr_mtx);
1759 	}
1760 #endif
1761 #ifdef INET6
1762 	while (redo_ip6) {
1763 		ip6s = pr->pr_ip6s;
1764 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1765 		mtx_lock(&pr->pr_mtx);
1766 		redo_ip6 = 0;
1767 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1768 #ifdef VIMAGE
1769 			if (tpr->pr_flags & PR_VNET) {
1770 				descend = 0;
1771 				continue;
1772 			}
1773 #endif
1774 			if (prison_restrict_ip6(tpr, ip6)) {
1775 				if (ip6 != NULL)
1776 					ip6 = NULL;
1777 				else
1778 					redo_ip6 = 1;
1779 			}
1780 		}
1781 		mtx_unlock(&pr->pr_mtx);
1782 	}
1783 #endif
1784 
1785 	/* Let the modules do their work. */
1786 	sx_downgrade(&allprison_lock);
1787 	if (born) {
1788 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1789 		if (error) {
1790 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1791 			prison_deref(pr, created
1792 			    ? PD_LIST_SLOCKED
1793 			    : PD_DEREF | PD_LIST_SLOCKED);
1794 			goto done_errmsg;
1795 		}
1796 	}
1797 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1798 	if (error) {
1799 		if (born)
1800 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1801 		prison_deref(pr, created
1802 		    ? PD_LIST_SLOCKED
1803 		    : PD_DEREF | PD_LIST_SLOCKED);
1804 		goto done_errmsg;
1805 	}
1806 
1807 	/* Attach this process to the prison if requested. */
1808 	if (flags & JAIL_ATTACH) {
1809 		mtx_lock(&pr->pr_mtx);
1810 		error = do_jail_attach(td, pr);
1811 		if (error) {
1812 			vfs_opterror(opts, "attach failed");
1813 			if (!created)
1814 				prison_deref(pr, PD_DEREF);
1815 			goto done_errmsg;
1816 		}
1817 	}
1818 
1819 #ifdef RACCT
1820 	if (racct_enable && !created) {
1821 		if (!(flags & JAIL_ATTACH))
1822 			sx_sunlock(&allprison_lock);
1823 		prison_racct_modify(pr);
1824 		if (!(flags & JAIL_ATTACH))
1825 			sx_slock(&allprison_lock);
1826 	}
1827 #endif
1828 
1829 	td->td_retval[0] = pr->pr_id;
1830 
1831 	/*
1832 	 * Now that it is all there, drop the temporary reference from existing
1833 	 * prisons.  Or add a reference to newly created persistent prisons
1834 	 * (which was not done earlier so that the prison would not be publicly
1835 	 * visible).
1836 	 */
1837 	if (!created) {
1838 		prison_deref(pr, (flags & JAIL_ATTACH)
1839 		    ? PD_DEREF
1840 		    : PD_DEREF | PD_LIST_SLOCKED);
1841 	} else {
1842 		if (pr_flags & PR_PERSIST) {
1843 			mtx_lock(&pr->pr_mtx);
1844 			pr->pr_ref++;
1845 			pr->pr_uref++;
1846 			mtx_unlock(&pr->pr_mtx);
1847 		}
1848 		if (!(flags & JAIL_ATTACH))
1849 			sx_sunlock(&allprison_lock);
1850 	}
1851 
1852 	goto done_free;
1853 
1854  done_deref_locked:
1855 	prison_deref(pr, created
1856 	    ? PD_LOCKED | PD_LIST_XLOCKED
1857 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1858 	goto done_releroot;
1859  done_unlock_list:
1860 	sx_xunlock(&allprison_lock);
1861  done_releroot:
1862 	if (root != NULL)
1863 		vrele(root);
1864  done_errmsg:
1865 	if (error) {
1866 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1867 		    &errmsg_len) == 0 && errmsg_len > 0) {
1868 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1869 			if (optuio->uio_segflg == UIO_SYSSPACE)
1870 				bcopy(errmsg,
1871 				    optuio->uio_iov[errmsg_pos].iov_base,
1872 				    errmsg_len);
1873 			else
1874 				copyout(errmsg,
1875 				    optuio->uio_iov[errmsg_pos].iov_base,
1876 				    errmsg_len);
1877 		}
1878 	}
1879  done_free:
1880 #ifdef INET
1881 	free(ip4, M_PRISON);
1882 #endif
1883 #ifdef INET6
1884 	free(ip6, M_PRISON);
1885 #endif
1886 	if (g_path != NULL)
1887 		free(g_path, M_TEMP);
1888 	vfs_freeopts(opts);
1889 	return (error);
1890 }
1891 
1892 
1893 /*
1894  * struct jail_get_args {
1895  *	struct iovec *iovp;
1896  *	unsigned int iovcnt;
1897  *	int flags;
1898  * };
1899  */
1900 int
1901 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1902 {
1903 	struct uio *auio;
1904 	int error;
1905 
1906 	/* Check that we have an even number of iovecs. */
1907 	if (uap->iovcnt & 1)
1908 		return (EINVAL);
1909 
1910 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1911 	if (error)
1912 		return (error);
1913 	error = kern_jail_get(td, auio, uap->flags);
1914 	if (error == 0)
1915 		error = copyout(auio->uio_iov, uap->iovp,
1916 		    uap->iovcnt * sizeof (struct iovec));
1917 	free(auio, M_IOV);
1918 	return (error);
1919 }
1920 
1921 int
1922 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1923 {
1924 	struct bool_flags *bf;
1925 	struct jailsys_flags *jsf;
1926 	struct prison *pr, *mypr;
1927 	struct vfsopt *opt;
1928 	struct vfsoptlist *opts;
1929 	char *errmsg, *name;
1930 	int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1931 	unsigned f;
1932 
1933 	if (flags & ~JAIL_GET_MASK)
1934 		return (EINVAL);
1935 
1936 	/* Get the parameter list. */
1937 	error = vfs_buildopts(optuio, &opts);
1938 	if (error)
1939 		return (error);
1940 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1941 	mypr = td->td_ucred->cr_prison;
1942 
1943 	/*
1944 	 * Find the prison specified by one of: lastjid, jid, name.
1945 	 */
1946 	sx_slock(&allprison_lock);
1947 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1948 	if (error == 0) {
1949 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1950 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1951 				mtx_lock(&pr->pr_mtx);
1952 				if (pr->pr_ref > 0 &&
1953 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1954 					break;
1955 				mtx_unlock(&pr->pr_mtx);
1956 			}
1957 		}
1958 		if (pr != NULL)
1959 			goto found_prison;
1960 		error = ENOENT;
1961 		vfs_opterror(opts, "no jail after %d", jid);
1962 		goto done_unlock_list;
1963 	} else if (error != ENOENT)
1964 		goto done_unlock_list;
1965 
1966 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1967 	if (error == 0) {
1968 		if (jid != 0) {
1969 			pr = prison_find_child(mypr, jid);
1970 			if (pr != NULL) {
1971 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1972 					mtx_unlock(&pr->pr_mtx);
1973 					error = ENOENT;
1974 					vfs_opterror(opts, "jail %d is dying",
1975 					    jid);
1976 					goto done_unlock_list;
1977 				}
1978 				goto found_prison;
1979 			}
1980 			error = ENOENT;
1981 			vfs_opterror(opts, "jail %d not found", jid);
1982 			goto done_unlock_list;
1983 		}
1984 	} else if (error != ENOENT)
1985 		goto done_unlock_list;
1986 
1987 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1988 	if (error == 0) {
1989 		if (len == 0 || name[len - 1] != '\0') {
1990 			error = EINVAL;
1991 			goto done_unlock_list;
1992 		}
1993 		pr = prison_find_name(mypr, name);
1994 		if (pr != NULL) {
1995 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1996 				mtx_unlock(&pr->pr_mtx);
1997 				error = ENOENT;
1998 				vfs_opterror(opts, "jail \"%s\" is dying",
1999 				    name);
2000 				goto done_unlock_list;
2001 			}
2002 			goto found_prison;
2003 		}
2004 		error = ENOENT;
2005 		vfs_opterror(opts, "jail \"%s\" not found", name);
2006 		goto done_unlock_list;
2007 	} else if (error != ENOENT)
2008 		goto done_unlock_list;
2009 
2010 	vfs_opterror(opts, "no jail specified");
2011 	error = ENOENT;
2012 	goto done_unlock_list;
2013 
2014  found_prison:
2015 	/* Get the parameters of the prison. */
2016 	pr->pr_ref++;
2017 	locked = PD_LOCKED;
2018 	td->td_retval[0] = pr->pr_id;
2019 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2020 	if (error != 0 && error != ENOENT)
2021 		goto done_deref;
2022 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2023 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2024 	if (error != 0 && error != ENOENT)
2025 		goto done_deref;
2026 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2027 	if (error != 0 && error != ENOENT)
2028 		goto done_deref;
2029 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2030 	    sizeof(pr->pr_cpuset->cs_id));
2031 	if (error != 0 && error != ENOENT)
2032 		goto done_deref;
2033 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2034 	if (error != 0 && error != ENOENT)
2035 		goto done_deref;
2036 #ifdef INET
2037 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2038 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2039 	if (error != 0 && error != ENOENT)
2040 		goto done_deref;
2041 #endif
2042 #ifdef INET6
2043 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2044 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2045 	if (error != 0 && error != ENOENT)
2046 		goto done_deref;
2047 #endif
2048 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2049 	    sizeof(pr->pr_securelevel));
2050 	if (error != 0 && error != ENOENT)
2051 		goto done_deref;
2052 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2053 	    sizeof(pr->pr_childcount));
2054 	if (error != 0 && error != ENOENT)
2055 		goto done_deref;
2056 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2057 	    sizeof(pr->pr_childmax));
2058 	if (error != 0 && error != ENOENT)
2059 		goto done_deref;
2060 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2061 	if (error != 0 && error != ENOENT)
2062 		goto done_deref;
2063 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2064 	if (error != 0 && error != ENOENT)
2065 		goto done_deref;
2066 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2067 	if (error != 0 && error != ENOENT)
2068 		goto done_deref;
2069 #ifdef COMPAT_FREEBSD32
2070 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2071 		uint32_t hid32 = pr->pr_hostid;
2072 
2073 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2074 	} else
2075 #endif
2076 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2077 	    sizeof(pr->pr_hostid));
2078 	if (error != 0 && error != ENOENT)
2079 		goto done_deref;
2080 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2081 	    sizeof(pr->pr_enforce_statfs));
2082 	if (error != 0 && error != ENOENT)
2083 		goto done_deref;
2084 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2085 	    sizeof(pr->pr_devfs_rsnum));
2086 	if (error != 0 && error != ENOENT)
2087 		goto done_deref;
2088 	for (bf = pr_flag_bool;
2089 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2090 	     bf++) {
2091 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2092 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2093 		if (error != 0 && error != ENOENT)
2094 			goto done_deref;
2095 		i = !i;
2096 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2097 		if (error != 0 && error != ENOENT)
2098 			goto done_deref;
2099 	}
2100 	for (jsf = pr_flag_jailsys;
2101 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2102 	     jsf++) {
2103 		f = pr->pr_flags & (jsf->disable | jsf->new);
2104 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2105 		    : (f == jsf->new) ? JAIL_SYS_NEW
2106 		    : JAIL_SYS_INHERIT;
2107 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2108 		if (error != 0 && error != ENOENT)
2109 			goto done_deref;
2110 	}
2111 	for (bf = pr_flag_allow;
2112 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2113 	     bf++) {
2114 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2115 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2116 		if (error != 0 && error != ENOENT)
2117 			goto done_deref;
2118 		i = !i;
2119 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2120 		if (error != 0 && error != ENOENT)
2121 			goto done_deref;
2122 	}
2123 	i = (pr->pr_uref == 0);
2124 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2125 	if (error != 0 && error != ENOENT)
2126 		goto done_deref;
2127 	i = !i;
2128 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2129 	if (error != 0 && error != ENOENT)
2130 		goto done_deref;
2131 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2132 	    sizeof(pr->pr_osreldate));
2133 	if (error != 0 && error != ENOENT)
2134 		goto done_deref;
2135 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2136 	if (error != 0 && error != ENOENT)
2137 		goto done_deref;
2138 
2139 	/* Get the module parameters. */
2140 	mtx_unlock(&pr->pr_mtx);
2141 	locked = 0;
2142 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2143 	if (error)
2144 		goto done_deref;
2145 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2146 
2147 	/* By now, all parameters should have been noted. */
2148 	TAILQ_FOREACH(opt, opts, link) {
2149 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2150 			error = EINVAL;
2151 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2152 			goto done_errmsg;
2153 		}
2154 	}
2155 
2156 	/* Write the fetched parameters back to userspace. */
2157 	error = 0;
2158 	TAILQ_FOREACH(opt, opts, link) {
2159 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2160 			pos = 2 * opt->pos + 1;
2161 			optuio->uio_iov[pos].iov_len = opt->len;
2162 			if (opt->value != NULL) {
2163 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2164 					bcopy(opt->value,
2165 					    optuio->uio_iov[pos].iov_base,
2166 					    opt->len);
2167 				} else {
2168 					error = copyout(opt->value,
2169 					    optuio->uio_iov[pos].iov_base,
2170 					    opt->len);
2171 					if (error)
2172 						break;
2173 				}
2174 			}
2175 		}
2176 	}
2177 	goto done_errmsg;
2178 
2179  done_deref:
2180 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2181 	goto done_errmsg;
2182 
2183  done_unlock_list:
2184 	sx_sunlock(&allprison_lock);
2185  done_errmsg:
2186 	if (error && errmsg_pos >= 0) {
2187 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2188 		errmsg_pos = 2 * errmsg_pos + 1;
2189 		if (errmsg_len > 0) {
2190 			if (optuio->uio_segflg == UIO_SYSSPACE)
2191 				bcopy(errmsg,
2192 				    optuio->uio_iov[errmsg_pos].iov_base,
2193 				    errmsg_len);
2194 			else
2195 				copyout(errmsg,
2196 				    optuio->uio_iov[errmsg_pos].iov_base,
2197 				    errmsg_len);
2198 		}
2199 	}
2200 	vfs_freeopts(opts);
2201 	return (error);
2202 }
2203 
2204 
2205 /*
2206  * struct jail_remove_args {
2207  *	int jid;
2208  * };
2209  */
2210 int
2211 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2212 {
2213 	struct prison *pr, *cpr, *lpr, *tpr;
2214 	int descend, error;
2215 
2216 	error = priv_check(td, PRIV_JAIL_REMOVE);
2217 	if (error)
2218 		return (error);
2219 
2220 	sx_xlock(&allprison_lock);
2221 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2222 	if (pr == NULL) {
2223 		sx_xunlock(&allprison_lock);
2224 		return (EINVAL);
2225 	}
2226 
2227 	/* Remove all descendants of this prison, then remove this prison. */
2228 	pr->pr_ref++;
2229 	if (!LIST_EMPTY(&pr->pr_children)) {
2230 		mtx_unlock(&pr->pr_mtx);
2231 		lpr = NULL;
2232 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2233 			mtx_lock(&cpr->pr_mtx);
2234 			if (cpr->pr_ref > 0) {
2235 				tpr = cpr;
2236 				cpr->pr_ref++;
2237 			} else {
2238 				/* Already removed - do not do it again. */
2239 				tpr = NULL;
2240 			}
2241 			mtx_unlock(&cpr->pr_mtx);
2242 			if (lpr != NULL) {
2243 				mtx_lock(&lpr->pr_mtx);
2244 				prison_remove_one(lpr);
2245 				sx_xlock(&allprison_lock);
2246 			}
2247 			lpr = tpr;
2248 		}
2249 		if (lpr != NULL) {
2250 			mtx_lock(&lpr->pr_mtx);
2251 			prison_remove_one(lpr);
2252 			sx_xlock(&allprison_lock);
2253 		}
2254 		mtx_lock(&pr->pr_mtx);
2255 	}
2256 	prison_remove_one(pr);
2257 	return (0);
2258 }
2259 
2260 static void
2261 prison_remove_one(struct prison *pr)
2262 {
2263 	struct proc *p;
2264 	int deuref;
2265 
2266 	/* If the prison was persistent, it is not anymore. */
2267 	deuref = 0;
2268 	if (pr->pr_flags & PR_PERSIST) {
2269 		pr->pr_ref--;
2270 		deuref = PD_DEUREF;
2271 		pr->pr_flags &= ~PR_PERSIST;
2272 	}
2273 
2274 	/*
2275 	 * jail_remove added a reference.  If that's the only one, remove
2276 	 * the prison now.
2277 	 */
2278 	KASSERT(pr->pr_ref > 0,
2279 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2280 	if (pr->pr_ref == 1) {
2281 		prison_deref(pr,
2282 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2283 		return;
2284 	}
2285 
2286 	mtx_unlock(&pr->pr_mtx);
2287 	sx_xunlock(&allprison_lock);
2288 	/*
2289 	 * Kill all processes unfortunate enough to be attached to this prison.
2290 	 */
2291 	sx_slock(&allproc_lock);
2292 	FOREACH_PROC_IN_SYSTEM(p) {
2293 		PROC_LOCK(p);
2294 		if (p->p_state != PRS_NEW && p->p_ucred &&
2295 		    p->p_ucred->cr_prison == pr)
2296 			kern_psignal(p, SIGKILL);
2297 		PROC_UNLOCK(p);
2298 	}
2299 	sx_sunlock(&allproc_lock);
2300 	/* Remove the temporary reference added by jail_remove. */
2301 	prison_deref(pr, deuref | PD_DEREF);
2302 }
2303 
2304 
2305 /*
2306  * struct jail_attach_args {
2307  *	int jid;
2308  * };
2309  */
2310 int
2311 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2312 {
2313 	struct prison *pr;
2314 	int error;
2315 
2316 	error = priv_check(td, PRIV_JAIL_ATTACH);
2317 	if (error)
2318 		return (error);
2319 
2320 	/*
2321 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2322 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2323 	 * But then immediately downgrade it since we don't need to stop
2324 	 * readers.
2325 	 */
2326 	sx_xlock(&allprison_lock);
2327 	sx_downgrade(&allprison_lock);
2328 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2329 	if (pr == NULL) {
2330 		sx_sunlock(&allprison_lock);
2331 		return (EINVAL);
2332 	}
2333 
2334 	/*
2335 	 * Do not allow a process to attach to a prison that is not
2336 	 * considered to be "alive".
2337 	 */
2338 	if (pr->pr_uref == 0) {
2339 		mtx_unlock(&pr->pr_mtx);
2340 		sx_sunlock(&allprison_lock);
2341 		return (EINVAL);
2342 	}
2343 
2344 	return (do_jail_attach(td, pr));
2345 }
2346 
2347 static int
2348 do_jail_attach(struct thread *td, struct prison *pr)
2349 {
2350 	struct proc *p;
2351 	struct ucred *newcred, *oldcred;
2352 	int error;
2353 
2354 	/*
2355 	 * XXX: Note that there is a slight race here if two threads
2356 	 * in the same privileged process attempt to attach to two
2357 	 * different jails at the same time.  It is important for
2358 	 * user processes not to do this, or they might end up with
2359 	 * a process root from one prison, but attached to the jail
2360 	 * of another.
2361 	 */
2362 	pr->pr_ref++;
2363 	pr->pr_uref++;
2364 	mtx_unlock(&pr->pr_mtx);
2365 
2366 	/* Let modules do whatever they need to prepare for attaching. */
2367 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2368 	if (error) {
2369 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2370 		return (error);
2371 	}
2372 	sx_sunlock(&allprison_lock);
2373 
2374 	/*
2375 	 * Reparent the newly attached process to this jail.
2376 	 */
2377 	p = td->td_proc;
2378 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2379 	if (error)
2380 		goto e_revert_osd;
2381 
2382 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2383 	if ((error = change_dir(pr->pr_root, td)) != 0)
2384 		goto e_unlock;
2385 #ifdef MAC
2386 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2387 		goto e_unlock;
2388 #endif
2389 	VOP_UNLOCK(pr->pr_root, 0);
2390 	if ((error = pwd_chroot(td, pr->pr_root)))
2391 		goto e_revert_osd;
2392 
2393 	newcred = crget();
2394 	PROC_LOCK(p);
2395 	oldcred = crcopysafe(p, newcred);
2396 	newcred->cr_prison = pr;
2397 	proc_set_cred(p, newcred);
2398 	setsugid(p);
2399 #ifdef RACCT
2400 	racct_proc_ucred_changed(p, oldcred, newcred);
2401 	crhold(newcred);
2402 #endif
2403 	PROC_UNLOCK(p);
2404 #ifdef RCTL
2405 	rctl_proc_ucred_changed(p, newcred);
2406 	crfree(newcred);
2407 #endif
2408 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2409 	crfree(oldcred);
2410 	return (0);
2411 
2412  e_unlock:
2413 	VOP_UNLOCK(pr->pr_root, 0);
2414  e_revert_osd:
2415 	/* Tell modules this thread is still in its old jail after all. */
2416 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2417 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2418 	return (error);
2419 }
2420 
2421 
2422 /*
2423  * Returns a locked prison instance, or NULL on failure.
2424  */
2425 struct prison *
2426 prison_find(int prid)
2427 {
2428 	struct prison *pr;
2429 
2430 	sx_assert(&allprison_lock, SX_LOCKED);
2431 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2432 		if (pr->pr_id == prid) {
2433 			mtx_lock(&pr->pr_mtx);
2434 			if (pr->pr_ref > 0)
2435 				return (pr);
2436 			mtx_unlock(&pr->pr_mtx);
2437 		}
2438 	}
2439 	return (NULL);
2440 }
2441 
2442 /*
2443  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2444  */
2445 struct prison *
2446 prison_find_child(struct prison *mypr, int prid)
2447 {
2448 	struct prison *pr;
2449 	int descend;
2450 
2451 	sx_assert(&allprison_lock, SX_LOCKED);
2452 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2453 		if (pr->pr_id == prid) {
2454 			mtx_lock(&pr->pr_mtx);
2455 			if (pr->pr_ref > 0)
2456 				return (pr);
2457 			mtx_unlock(&pr->pr_mtx);
2458 		}
2459 	}
2460 	return (NULL);
2461 }
2462 
2463 /*
2464  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2465  */
2466 struct prison *
2467 prison_find_name(struct prison *mypr, const char *name)
2468 {
2469 	struct prison *pr, *deadpr;
2470 	size_t mylen;
2471 	int descend;
2472 
2473 	sx_assert(&allprison_lock, SX_LOCKED);
2474 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2475  again:
2476 	deadpr = NULL;
2477 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2478 		if (!strcmp(pr->pr_name + mylen, name)) {
2479 			mtx_lock(&pr->pr_mtx);
2480 			if (pr->pr_ref > 0) {
2481 				if (pr->pr_uref > 0)
2482 					return (pr);
2483 				deadpr = pr;
2484 			}
2485 			mtx_unlock(&pr->pr_mtx);
2486 		}
2487 	}
2488 	/* There was no valid prison - perhaps there was a dying one. */
2489 	if (deadpr != NULL) {
2490 		mtx_lock(&deadpr->pr_mtx);
2491 		if (deadpr->pr_ref == 0) {
2492 			mtx_unlock(&deadpr->pr_mtx);
2493 			goto again;
2494 		}
2495 	}
2496 	return (deadpr);
2497 }
2498 
2499 /*
2500  * See if a prison has the specific flag set.
2501  */
2502 int
2503 prison_flag(struct ucred *cred, unsigned flag)
2504 {
2505 
2506 	/* This is an atomic read, so no locking is necessary. */
2507 	return (cred->cr_prison->pr_flags & flag);
2508 }
2509 
2510 int
2511 prison_allow(struct ucred *cred, unsigned flag)
2512 {
2513 
2514 	/* This is an atomic read, so no locking is necessary. */
2515 	return (cred->cr_prison->pr_allow & flag);
2516 }
2517 
2518 /*
2519  * Remove a prison reference.  If that was the last reference, remove the
2520  * prison itself - but not in this context in case there are locks held.
2521  */
2522 void
2523 prison_free_locked(struct prison *pr)
2524 {
2525 	int ref;
2526 
2527 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2528 	ref = --pr->pr_ref;
2529 	mtx_unlock(&pr->pr_mtx);
2530 	if (ref == 0)
2531 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2532 }
2533 
2534 void
2535 prison_free(struct prison *pr)
2536 {
2537 
2538 	mtx_lock(&pr->pr_mtx);
2539 	prison_free_locked(pr);
2540 }
2541 
2542 /*
2543  * Complete a call to either prison_free or prison_proc_free.
2544  */
2545 static void
2546 prison_complete(void *context, int pending)
2547 {
2548 	struct prison *pr = context;
2549 
2550 	sx_xlock(&allprison_lock);
2551 	mtx_lock(&pr->pr_mtx);
2552 	prison_deref(pr, pr->pr_uref
2553 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2554 	    : PD_LOCKED | PD_LIST_XLOCKED);
2555 }
2556 
2557 /*
2558  * Remove a prison reference (usually).  This internal version assumes no
2559  * mutexes are held, except perhaps the prison itself.  If there are no more
2560  * references, release and delist the prison.  On completion, the prison lock
2561  * and the allprison lock are both unlocked.
2562  */
2563 static void
2564 prison_deref(struct prison *pr, int flags)
2565 {
2566 	struct prison *ppr, *tpr;
2567 	int ref, lasturef;
2568 
2569 	if (!(flags & PD_LOCKED))
2570 		mtx_lock(&pr->pr_mtx);
2571 	for (;;) {
2572 		if (flags & PD_DEUREF) {
2573 			KASSERT(pr->pr_uref > 0,
2574 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2575 			     pr->pr_id));
2576 			pr->pr_uref--;
2577 			lasturef = pr->pr_uref == 0;
2578 			if (lasturef)
2579 				pr->pr_ref++;
2580 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2581 		} else
2582 			lasturef = 0;
2583 		if (flags & PD_DEREF) {
2584 			KASSERT(pr->pr_ref > 0,
2585 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2586 			     pr->pr_id));
2587 			pr->pr_ref--;
2588 		}
2589 		ref = pr->pr_ref;
2590 		mtx_unlock(&pr->pr_mtx);
2591 
2592 		/*
2593 		 * Tell the modules if the last user reference was removed
2594 		 * (even it sticks around in dying state).
2595 		 */
2596 		if (lasturef) {
2597 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2598 				sx_xlock(&allprison_lock);
2599 				flags |= PD_LIST_XLOCKED;
2600 			}
2601 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2602 			mtx_lock(&pr->pr_mtx);
2603 			ref = --pr->pr_ref;
2604 			mtx_unlock(&pr->pr_mtx);
2605 		}
2606 
2607 		/* If the prison still has references, nothing else to do. */
2608 		if (ref > 0) {
2609 			if (flags & PD_LIST_SLOCKED)
2610 				sx_sunlock(&allprison_lock);
2611 			else if (flags & PD_LIST_XLOCKED)
2612 				sx_xunlock(&allprison_lock);
2613 			return;
2614 		}
2615 
2616 		if (flags & PD_LIST_SLOCKED) {
2617 			if (!sx_try_upgrade(&allprison_lock)) {
2618 				sx_sunlock(&allprison_lock);
2619 				sx_xlock(&allprison_lock);
2620 			}
2621 		} else if (!(flags & PD_LIST_XLOCKED))
2622 			sx_xlock(&allprison_lock);
2623 
2624 		TAILQ_REMOVE(&allprison, pr, pr_list);
2625 		LIST_REMOVE(pr, pr_sibling);
2626 		ppr = pr->pr_parent;
2627 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2628 			tpr->pr_childcount--;
2629 		sx_xunlock(&allprison_lock);
2630 
2631 #ifdef VIMAGE
2632 		if (pr->pr_vnet != ppr->pr_vnet)
2633 			vnet_destroy(pr->pr_vnet);
2634 #endif
2635 		if (pr->pr_root != NULL)
2636 			vrele(pr->pr_root);
2637 		mtx_destroy(&pr->pr_mtx);
2638 #ifdef INET
2639 		free(pr->pr_ip4, M_PRISON);
2640 #endif
2641 #ifdef INET6
2642 		free(pr->pr_ip6, M_PRISON);
2643 #endif
2644 		if (pr->pr_cpuset != NULL)
2645 			cpuset_rel(pr->pr_cpuset);
2646 		osd_jail_exit(pr);
2647 #ifdef RACCT
2648 		if (racct_enable)
2649 			prison_racct_detach(pr);
2650 #endif
2651 		free(pr, M_PRISON);
2652 
2653 		/* Removing a prison frees a reference on its parent. */
2654 		pr = ppr;
2655 		mtx_lock(&pr->pr_mtx);
2656 		flags = PD_DEREF | PD_DEUREF;
2657 	}
2658 }
2659 
2660 void
2661 prison_hold_locked(struct prison *pr)
2662 {
2663 
2664 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2665 	KASSERT(pr->pr_ref > 0,
2666 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2667 	pr->pr_ref++;
2668 }
2669 
2670 void
2671 prison_hold(struct prison *pr)
2672 {
2673 
2674 	mtx_lock(&pr->pr_mtx);
2675 	prison_hold_locked(pr);
2676 	mtx_unlock(&pr->pr_mtx);
2677 }
2678 
2679 void
2680 prison_proc_hold(struct prison *pr)
2681 {
2682 
2683 	mtx_lock(&pr->pr_mtx);
2684 	KASSERT(pr->pr_uref > 0,
2685 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2686 	pr->pr_uref++;
2687 	mtx_unlock(&pr->pr_mtx);
2688 }
2689 
2690 void
2691 prison_proc_free(struct prison *pr)
2692 {
2693 
2694 	mtx_lock(&pr->pr_mtx);
2695 	KASSERT(pr->pr_uref > 0,
2696 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2697 	if (pr->pr_uref > 1)
2698 		pr->pr_uref--;
2699 	else {
2700 		/*
2701 		 * Don't remove the last user reference in this context, which
2702 		 * is expected to be a process that is not only locked, but
2703 		 * also half dead.
2704 		 */
2705 		pr->pr_ref++;
2706 		mtx_unlock(&pr->pr_mtx);
2707 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2708 		return;
2709 	}
2710 	mtx_unlock(&pr->pr_mtx);
2711 }
2712 
2713 /*
2714  * Check if a jail supports the given address family.
2715  *
2716  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2717  * if not.
2718  */
2719 int
2720 prison_check_af(struct ucred *cred, int af)
2721 {
2722 	struct prison *pr;
2723 	int error;
2724 
2725 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2726 
2727 	pr = cred->cr_prison;
2728 #ifdef VIMAGE
2729 	/* Prisons with their own network stack are not limited. */
2730 	if (prison_owns_vnet(cred))
2731 		return (0);
2732 #endif
2733 
2734 	error = 0;
2735 	switch (af)
2736 	{
2737 #ifdef INET
2738 	case AF_INET:
2739 		if (pr->pr_flags & PR_IP4)
2740 		{
2741 			mtx_lock(&pr->pr_mtx);
2742 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2743 				error = EAFNOSUPPORT;
2744 			mtx_unlock(&pr->pr_mtx);
2745 		}
2746 		break;
2747 #endif
2748 #ifdef INET6
2749 	case AF_INET6:
2750 		if (pr->pr_flags & PR_IP6)
2751 		{
2752 			mtx_lock(&pr->pr_mtx);
2753 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2754 				error = EAFNOSUPPORT;
2755 			mtx_unlock(&pr->pr_mtx);
2756 		}
2757 		break;
2758 #endif
2759 	case AF_LOCAL:
2760 	case AF_ROUTE:
2761 		break;
2762 	default:
2763 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2764 			error = EAFNOSUPPORT;
2765 	}
2766 	return (error);
2767 }
2768 
2769 /*
2770  * Check if given address belongs to the jail referenced by cred (wrapper to
2771  * prison_check_ip[46]).
2772  *
2773  * Returns 0 if jail doesn't restrict the address family or if address belongs
2774  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2775  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2776  */
2777 int
2778 prison_if(struct ucred *cred, struct sockaddr *sa)
2779 {
2780 #ifdef INET
2781 	struct sockaddr_in *sai;
2782 #endif
2783 #ifdef INET6
2784 	struct sockaddr_in6 *sai6;
2785 #endif
2786 	int error;
2787 
2788 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2789 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2790 
2791 #ifdef VIMAGE
2792 	if (prison_owns_vnet(cred))
2793 		return (0);
2794 #endif
2795 
2796 	error = 0;
2797 	switch (sa->sa_family)
2798 	{
2799 #ifdef INET
2800 	case AF_INET:
2801 		sai = (struct sockaddr_in *)sa;
2802 		error = prison_check_ip4(cred, &sai->sin_addr);
2803 		break;
2804 #endif
2805 #ifdef INET6
2806 	case AF_INET6:
2807 		sai6 = (struct sockaddr_in6 *)sa;
2808 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2809 		break;
2810 #endif
2811 	default:
2812 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2813 			error = EAFNOSUPPORT;
2814 	}
2815 	return (error);
2816 }
2817 
2818 /*
2819  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2820  */
2821 int
2822 prison_check(struct ucred *cred1, struct ucred *cred2)
2823 {
2824 
2825 	return ((cred1->cr_prison == cred2->cr_prison ||
2826 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2827 }
2828 
2829 /*
2830  * Return 1 if p2 is a child of p1, otherwise 0.
2831  */
2832 int
2833 prison_ischild(struct prison *pr1, struct prison *pr2)
2834 {
2835 
2836 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2837 		if (pr1 == pr2)
2838 			return (1);
2839 	return (0);
2840 }
2841 
2842 /*
2843  * Return 1 if the passed credential is in a jail, otherwise 0.
2844  */
2845 int
2846 jailed(struct ucred *cred)
2847 {
2848 
2849 	return (cred->cr_prison != &prison0);
2850 }
2851 
2852 /*
2853  * Return 1 if the passed credential is in a jail and that jail does not
2854  * have its own virtual network stack, otherwise 0.
2855  */
2856 int
2857 jailed_without_vnet(struct ucred *cred)
2858 {
2859 
2860 	if (!jailed(cred))
2861 		return (0);
2862 #ifdef VIMAGE
2863 	if (prison_owns_vnet(cred))
2864 		return (0);
2865 #endif
2866 
2867 	return (1);
2868 }
2869 
2870 /*
2871  * Return the correct hostname (domainname, et al) for the passed credential.
2872  */
2873 void
2874 getcredhostname(struct ucred *cred, char *buf, size_t size)
2875 {
2876 	struct prison *pr;
2877 
2878 	/*
2879 	 * A NULL credential can be used to shortcut to the physical
2880 	 * system's hostname.
2881 	 */
2882 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2883 	mtx_lock(&pr->pr_mtx);
2884 	strlcpy(buf, pr->pr_hostname, size);
2885 	mtx_unlock(&pr->pr_mtx);
2886 }
2887 
2888 void
2889 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2890 {
2891 
2892 	mtx_lock(&cred->cr_prison->pr_mtx);
2893 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2894 	mtx_unlock(&cred->cr_prison->pr_mtx);
2895 }
2896 
2897 void
2898 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2899 {
2900 
2901 	mtx_lock(&cred->cr_prison->pr_mtx);
2902 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2903 	mtx_unlock(&cred->cr_prison->pr_mtx);
2904 }
2905 
2906 void
2907 getcredhostid(struct ucred *cred, unsigned long *hostid)
2908 {
2909 
2910 	mtx_lock(&cred->cr_prison->pr_mtx);
2911 	*hostid = cred->cr_prison->pr_hostid;
2912 	mtx_unlock(&cred->cr_prison->pr_mtx);
2913 }
2914 
2915 #ifdef VIMAGE
2916 /*
2917  * Determine whether the prison represented by cred owns
2918  * its vnet rather than having it inherited.
2919  *
2920  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2921  */
2922 int
2923 prison_owns_vnet(struct ucred *cred)
2924 {
2925 
2926 	/*
2927 	 * vnets cannot be added/removed after jail creation,
2928 	 * so no need to lock here.
2929 	 */
2930 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2931 }
2932 #endif
2933 
2934 /*
2935  * Determine whether the subject represented by cred can "see"
2936  * status of a mount point.
2937  * Returns: 0 for permitted, ENOENT otherwise.
2938  * XXX: This function should be called cr_canseemount() and should be
2939  *      placed in kern_prot.c.
2940  */
2941 int
2942 prison_canseemount(struct ucred *cred, struct mount *mp)
2943 {
2944 	struct prison *pr;
2945 	struct statfs *sp;
2946 	size_t len;
2947 
2948 	pr = cred->cr_prison;
2949 	if (pr->pr_enforce_statfs == 0)
2950 		return (0);
2951 	if (pr->pr_root->v_mount == mp)
2952 		return (0);
2953 	if (pr->pr_enforce_statfs == 2)
2954 		return (ENOENT);
2955 	/*
2956 	 * If jail's chroot directory is set to "/" we should be able to see
2957 	 * all mount-points from inside a jail.
2958 	 * This is ugly check, but this is the only situation when jail's
2959 	 * directory ends with '/'.
2960 	 */
2961 	if (strcmp(pr->pr_path, "/") == 0)
2962 		return (0);
2963 	len = strlen(pr->pr_path);
2964 	sp = &mp->mnt_stat;
2965 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2966 		return (ENOENT);
2967 	/*
2968 	 * Be sure that we don't have situation where jail's root directory
2969 	 * is "/some/path" and mount point is "/some/pathpath".
2970 	 */
2971 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2972 		return (ENOENT);
2973 	return (0);
2974 }
2975 
2976 void
2977 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2978 {
2979 	char jpath[MAXPATHLEN];
2980 	struct prison *pr;
2981 	size_t len;
2982 
2983 	pr = cred->cr_prison;
2984 	if (pr->pr_enforce_statfs == 0)
2985 		return;
2986 	if (prison_canseemount(cred, mp) != 0) {
2987 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2988 		strlcpy(sp->f_mntonname, "[restricted]",
2989 		    sizeof(sp->f_mntonname));
2990 		return;
2991 	}
2992 	if (pr->pr_root->v_mount == mp) {
2993 		/*
2994 		 * Clear current buffer data, so we are sure nothing from
2995 		 * the valid path left there.
2996 		 */
2997 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2998 		*sp->f_mntonname = '/';
2999 		return;
3000 	}
3001 	/*
3002 	 * If jail's chroot directory is set to "/" we should be able to see
3003 	 * all mount-points from inside a jail.
3004 	 */
3005 	if (strcmp(pr->pr_path, "/") == 0)
3006 		return;
3007 	len = strlen(pr->pr_path);
3008 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3009 	/*
3010 	 * Clear current buffer data, so we are sure nothing from
3011 	 * the valid path left there.
3012 	 */
3013 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3014 	if (*jpath == '\0') {
3015 		/* Should never happen. */
3016 		*sp->f_mntonname = '/';
3017 	} else {
3018 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3019 	}
3020 }
3021 
3022 /*
3023  * Check with permission for a specific privilege is granted within jail.  We
3024  * have a specific list of accepted privileges; the rest are denied.
3025  */
3026 int
3027 prison_priv_check(struct ucred *cred, int priv)
3028 {
3029 
3030 	if (!jailed(cred))
3031 		return (0);
3032 
3033 #ifdef VIMAGE
3034 	/*
3035 	 * Privileges specific to prisons with a virtual network stack.
3036 	 * There might be a duplicate entry here in case the privilege
3037 	 * is only granted conditionally in the legacy jail case.
3038 	 */
3039 	switch (priv) {
3040 #ifdef notyet
3041 		/*
3042 		 * NFS-specific privileges.
3043 		 */
3044 	case PRIV_NFS_DAEMON:
3045 	case PRIV_NFS_LOCKD:
3046 #endif
3047 		/*
3048 		 * Network stack privileges.
3049 		 */
3050 	case PRIV_NET_BRIDGE:
3051 	case PRIV_NET_GRE:
3052 	case PRIV_NET_BPF:
3053 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3054 	case PRIV_NET_ROUTE:
3055 	case PRIV_NET_TAP:
3056 	case PRIV_NET_SETIFMTU:
3057 	case PRIV_NET_SETIFFLAGS:
3058 	case PRIV_NET_SETIFCAP:
3059 	case PRIV_NET_SETIFDESCR:
3060 	case PRIV_NET_SETIFNAME	:
3061 	case PRIV_NET_SETIFMETRIC:
3062 	case PRIV_NET_SETIFPHYS:
3063 	case PRIV_NET_SETIFMAC:
3064 	case PRIV_NET_ADDMULTI:
3065 	case PRIV_NET_DELMULTI:
3066 	case PRIV_NET_HWIOCTL:
3067 	case PRIV_NET_SETLLADDR:
3068 	case PRIV_NET_ADDIFGROUP:
3069 	case PRIV_NET_DELIFGROUP:
3070 	case PRIV_NET_IFCREATE:
3071 	case PRIV_NET_IFDESTROY:
3072 	case PRIV_NET_ADDIFADDR:
3073 	case PRIV_NET_DELIFADDR:
3074 	case PRIV_NET_LAGG:
3075 	case PRIV_NET_GIF:
3076 	case PRIV_NET_SETIFVNET:
3077 	case PRIV_NET_SETIFFIB:
3078 
3079 		/*
3080 		 * 802.11-related privileges.
3081 		 */
3082 	case PRIV_NET80211_GETKEY:
3083 #ifdef notyet
3084 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3085 #endif
3086 
3087 #ifdef notyet
3088 		/*
3089 		 * ATM privileges.
3090 		 */
3091 	case PRIV_NETATM_CFG:
3092 	case PRIV_NETATM_ADD:
3093 	case PRIV_NETATM_DEL:
3094 	case PRIV_NETATM_SET:
3095 
3096 		/*
3097 		 * Bluetooth privileges.
3098 		 */
3099 	case PRIV_NETBLUETOOTH_RAW:
3100 #endif
3101 
3102 		/*
3103 		 * Netgraph and netgraph module privileges.
3104 		 */
3105 	case PRIV_NETGRAPH_CONTROL:
3106 #ifdef notyet
3107 	case PRIV_NETGRAPH_TTY:
3108 #endif
3109 
3110 		/*
3111 		 * IPv4 and IPv6 privileges.
3112 		 */
3113 	case PRIV_NETINET_IPFW:
3114 	case PRIV_NETINET_DIVERT:
3115 	case PRIV_NETINET_PF:
3116 	case PRIV_NETINET_DUMMYNET:
3117 	case PRIV_NETINET_CARP:
3118 	case PRIV_NETINET_MROUTE:
3119 	case PRIV_NETINET_RAW:
3120 	case PRIV_NETINET_ADDRCTRL6:
3121 	case PRIV_NETINET_ND6:
3122 	case PRIV_NETINET_SCOPE6:
3123 	case PRIV_NETINET_ALIFETIME6:
3124 	case PRIV_NETINET_IPSEC:
3125 	case PRIV_NETINET_BINDANY:
3126 
3127 #ifdef notyet
3128 		/*
3129 		 * NCP privileges.
3130 		 */
3131 	case PRIV_NETNCP:
3132 
3133 		/*
3134 		 * SMB privileges.
3135 		 */
3136 	case PRIV_NETSMB:
3137 #endif
3138 
3139 	/*
3140 	 * No default: or deny here.
3141 	 * In case of no permit fall through to next switch().
3142 	 */
3143 		if (cred->cr_prison->pr_flags & PR_VNET)
3144 			return (0);
3145 	}
3146 #endif /* VIMAGE */
3147 
3148 	switch (priv) {
3149 
3150 		/*
3151 		 * Allow ktrace privileges for root in jail.
3152 		 */
3153 	case PRIV_KTRACE:
3154 
3155 #if 0
3156 		/*
3157 		 * Allow jailed processes to configure audit identity and
3158 		 * submit audit records (login, etc).  In the future we may
3159 		 * want to further refine the relationship between audit and
3160 		 * jail.
3161 		 */
3162 	case PRIV_AUDIT_GETAUDIT:
3163 	case PRIV_AUDIT_SETAUDIT:
3164 	case PRIV_AUDIT_SUBMIT:
3165 #endif
3166 
3167 		/*
3168 		 * Allow jailed processes to manipulate process UNIX
3169 		 * credentials in any way they see fit.
3170 		 */
3171 	case PRIV_CRED_SETUID:
3172 	case PRIV_CRED_SETEUID:
3173 	case PRIV_CRED_SETGID:
3174 	case PRIV_CRED_SETEGID:
3175 	case PRIV_CRED_SETGROUPS:
3176 	case PRIV_CRED_SETREUID:
3177 	case PRIV_CRED_SETREGID:
3178 	case PRIV_CRED_SETRESUID:
3179 	case PRIV_CRED_SETRESGID:
3180 
3181 		/*
3182 		 * Jail implements visibility constraints already, so allow
3183 		 * jailed root to override uid/gid-based constraints.
3184 		 */
3185 	case PRIV_SEEOTHERGIDS:
3186 	case PRIV_SEEOTHERUIDS:
3187 
3188 		/*
3189 		 * Jail implements inter-process debugging limits already, so
3190 		 * allow jailed root various debugging privileges.
3191 		 */
3192 	case PRIV_DEBUG_DIFFCRED:
3193 	case PRIV_DEBUG_SUGID:
3194 	case PRIV_DEBUG_UNPRIV:
3195 
3196 		/*
3197 		 * Allow jail to set various resource limits and login
3198 		 * properties, and for now, exceed process resource limits.
3199 		 */
3200 	case PRIV_PROC_LIMIT:
3201 	case PRIV_PROC_SETLOGIN:
3202 	case PRIV_PROC_SETRLIMIT:
3203 
3204 		/*
3205 		 * System V and POSIX IPC privileges are granted in jail.
3206 		 */
3207 	case PRIV_IPC_READ:
3208 	case PRIV_IPC_WRITE:
3209 	case PRIV_IPC_ADMIN:
3210 	case PRIV_IPC_MSGSIZE:
3211 	case PRIV_MQ_ADMIN:
3212 
3213 		/*
3214 		 * Jail operations within a jail work on child jails.
3215 		 */
3216 	case PRIV_JAIL_ATTACH:
3217 	case PRIV_JAIL_SET:
3218 	case PRIV_JAIL_REMOVE:
3219 
3220 		/*
3221 		 * Jail implements its own inter-process limits, so allow
3222 		 * root processes in jail to change scheduling on other
3223 		 * processes in the same jail.  Likewise for signalling.
3224 		 */
3225 	case PRIV_SCHED_DIFFCRED:
3226 	case PRIV_SCHED_CPUSET:
3227 	case PRIV_SIGNAL_DIFFCRED:
3228 	case PRIV_SIGNAL_SUGID:
3229 
3230 		/*
3231 		 * Allow jailed processes to write to sysctls marked as jail
3232 		 * writable.
3233 		 */
3234 	case PRIV_SYSCTL_WRITEJAIL:
3235 
3236 		/*
3237 		 * Allow root in jail to manage a variety of quota
3238 		 * properties.  These should likely be conditional on a
3239 		 * configuration option.
3240 		 */
3241 	case PRIV_VFS_GETQUOTA:
3242 	case PRIV_VFS_SETQUOTA:
3243 
3244 		/*
3245 		 * Since Jail relies on chroot() to implement file system
3246 		 * protections, grant many VFS privileges to root in jail.
3247 		 * Be careful to exclude mount-related and NFS-related
3248 		 * privileges.
3249 		 */
3250 	case PRIV_VFS_READ:
3251 	case PRIV_VFS_WRITE:
3252 	case PRIV_VFS_ADMIN:
3253 	case PRIV_VFS_EXEC:
3254 	case PRIV_VFS_LOOKUP:
3255 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3256 	case PRIV_VFS_CHFLAGS_DEV:
3257 	case PRIV_VFS_CHOWN:
3258 	case PRIV_VFS_CHROOT:
3259 	case PRIV_VFS_RETAINSUGID:
3260 	case PRIV_VFS_FCHROOT:
3261 	case PRIV_VFS_LINK:
3262 	case PRIV_VFS_SETGID:
3263 	case PRIV_VFS_STAT:
3264 	case PRIV_VFS_STICKYFILE:
3265 
3266 		/*
3267 		 * As in the non-jail case, non-root users are expected to be
3268 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3269 		 * exists in the jail and they have permission to access it).
3270 		 */
3271 	case PRIV_KMEM_READ:
3272 		return (0);
3273 
3274 		/*
3275 		 * Depending on the global setting, allow privilege of
3276 		 * setting system flags.
3277 		 */
3278 	case PRIV_VFS_SYSFLAGS:
3279 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3280 			return (0);
3281 		else
3282 			return (EPERM);
3283 
3284 		/*
3285 		 * Depending on the global setting, allow privilege of
3286 		 * mounting/unmounting file systems.
3287 		 */
3288 	case PRIV_VFS_MOUNT:
3289 	case PRIV_VFS_UNMOUNT:
3290 	case PRIV_VFS_MOUNT_NONUSER:
3291 	case PRIV_VFS_MOUNT_OWNER:
3292 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3293 		    cred->cr_prison->pr_enforce_statfs < 2)
3294 			return (0);
3295 		else
3296 			return (EPERM);
3297 
3298 		/*
3299 		 * Conditionnaly allow locking (unlocking) physical pages
3300 		 * in memory.
3301 		 */
3302 	case PRIV_VM_MLOCK:
3303 	case PRIV_VM_MUNLOCK:
3304 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3305 			return (0);
3306 		else
3307 			return (EPERM);
3308 
3309 		/*
3310 		 * Conditionally allow jailed root to bind reserved ports.
3311 		 */
3312 	case PRIV_NETINET_RESERVEDPORT:
3313 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3314 			return (0);
3315 		else
3316 			return (EPERM);
3317 
3318 		/*
3319 		 * Allow jailed root to reuse in-use ports.
3320 		 */
3321 	case PRIV_NETINET_REUSEPORT:
3322 		return (0);
3323 
3324 		/*
3325 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3326 		 */
3327 	case PRIV_NETINET_SETHDROPTS:
3328 		return (0);
3329 
3330 		/*
3331 		 * Conditionally allow creating raw sockets in jail.
3332 		 */
3333 	case PRIV_NETINET_RAW:
3334 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3335 			return (0);
3336 		else
3337 			return (EPERM);
3338 
3339 		/*
3340 		 * Since jail implements its own visibility limits on netstat
3341 		 * sysctls, allow getcred.  This allows identd to work in
3342 		 * jail.
3343 		 */
3344 	case PRIV_NETINET_GETCRED:
3345 		return (0);
3346 
3347 		/*
3348 		 * Allow jailed root to set loginclass.
3349 		 */
3350 	case PRIV_PROC_SETLOGINCLASS:
3351 		return (0);
3352 
3353 	default:
3354 		/*
3355 		 * In all remaining cases, deny the privilege request.  This
3356 		 * includes almost all network privileges, many system
3357 		 * configuration privileges.
3358 		 */
3359 		return (EPERM);
3360 	}
3361 }
3362 
3363 /*
3364  * Return the part of pr2's name that is relative to pr1, or the whole name
3365  * if it does not directly follow.
3366  */
3367 
3368 char *
3369 prison_name(struct prison *pr1, struct prison *pr2)
3370 {
3371 	char *name;
3372 
3373 	/* Jails see themselves as "0" (if they see themselves at all). */
3374 	if (pr1 == pr2)
3375 		return "0";
3376 	name = pr2->pr_name;
3377 	if (prison_ischild(pr1, pr2)) {
3378 		/*
3379 		 * pr1 isn't locked (and allprison_lock may not be either)
3380 		 * so its length can't be counted on.  But the number of dots
3381 		 * can be counted on - and counted.
3382 		 */
3383 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3384 			name = strchr(name, '.') + 1;
3385 	}
3386 	return (name);
3387 }
3388 
3389 /*
3390  * Return the part of pr2's path that is relative to pr1, or the whole path
3391  * if it does not directly follow.
3392  */
3393 static char *
3394 prison_path(struct prison *pr1, struct prison *pr2)
3395 {
3396 	char *path1, *path2;
3397 	int len1;
3398 
3399 	path1 = pr1->pr_path;
3400 	path2 = pr2->pr_path;
3401 	if (!strcmp(path1, "/"))
3402 		return (path2);
3403 	len1 = strlen(path1);
3404 	if (strncmp(path1, path2, len1))
3405 		return (path2);
3406 	if (path2[len1] == '\0')
3407 		return "/";
3408 	if (path2[len1] == '/')
3409 		return (path2 + len1);
3410 	return (path2);
3411 }
3412 
3413 
3414 /*
3415  * Jail-related sysctls.
3416  */
3417 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3418     "Jails");
3419 
3420 static int
3421 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3422 {
3423 	struct xprison *xp;
3424 	struct prison *pr, *cpr;
3425 #ifdef INET
3426 	struct in_addr *ip4 = NULL;
3427 	int ip4s = 0;
3428 #endif
3429 #ifdef INET6
3430 	struct in6_addr *ip6 = NULL;
3431 	int ip6s = 0;
3432 #endif
3433 	int descend, error;
3434 
3435 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3436 	pr = req->td->td_ucred->cr_prison;
3437 	error = 0;
3438 	sx_slock(&allprison_lock);
3439 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3440 #if defined(INET) || defined(INET6)
3441  again:
3442 #endif
3443 		mtx_lock(&cpr->pr_mtx);
3444 #ifdef INET
3445 		if (cpr->pr_ip4s > 0) {
3446 			if (ip4s < cpr->pr_ip4s) {
3447 				ip4s = cpr->pr_ip4s;
3448 				mtx_unlock(&cpr->pr_mtx);
3449 				ip4 = realloc(ip4, ip4s *
3450 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3451 				goto again;
3452 			}
3453 			bcopy(cpr->pr_ip4, ip4,
3454 			    cpr->pr_ip4s * sizeof(struct in_addr));
3455 		}
3456 #endif
3457 #ifdef INET6
3458 		if (cpr->pr_ip6s > 0) {
3459 			if (ip6s < cpr->pr_ip6s) {
3460 				ip6s = cpr->pr_ip6s;
3461 				mtx_unlock(&cpr->pr_mtx);
3462 				ip6 = realloc(ip6, ip6s *
3463 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3464 				goto again;
3465 			}
3466 			bcopy(cpr->pr_ip6, ip6,
3467 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3468 		}
3469 #endif
3470 		if (cpr->pr_ref == 0) {
3471 			mtx_unlock(&cpr->pr_mtx);
3472 			continue;
3473 		}
3474 		bzero(xp, sizeof(*xp));
3475 		xp->pr_version = XPRISON_VERSION;
3476 		xp->pr_id = cpr->pr_id;
3477 		xp->pr_state = cpr->pr_uref > 0
3478 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3479 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3480 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3481 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3482 #ifdef INET
3483 		xp->pr_ip4s = cpr->pr_ip4s;
3484 #endif
3485 #ifdef INET6
3486 		xp->pr_ip6s = cpr->pr_ip6s;
3487 #endif
3488 		mtx_unlock(&cpr->pr_mtx);
3489 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3490 		if (error)
3491 			break;
3492 #ifdef INET
3493 		if (xp->pr_ip4s > 0) {
3494 			error = SYSCTL_OUT(req, ip4,
3495 			    xp->pr_ip4s * sizeof(struct in_addr));
3496 			if (error)
3497 				break;
3498 		}
3499 #endif
3500 #ifdef INET6
3501 		if (xp->pr_ip6s > 0) {
3502 			error = SYSCTL_OUT(req, ip6,
3503 			    xp->pr_ip6s * sizeof(struct in6_addr));
3504 			if (error)
3505 				break;
3506 		}
3507 #endif
3508 	}
3509 	sx_sunlock(&allprison_lock);
3510 	free(xp, M_TEMP);
3511 #ifdef INET
3512 	free(ip4, M_TEMP);
3513 #endif
3514 #ifdef INET6
3515 	free(ip6, M_TEMP);
3516 #endif
3517 	return (error);
3518 }
3519 
3520 SYSCTL_OID(_security_jail, OID_AUTO, list,
3521     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3522     sysctl_jail_list, "S", "List of active jails");
3523 
3524 static int
3525 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3526 {
3527 	int error, injail;
3528 
3529 	injail = jailed(req->td->td_ucred);
3530 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3531 
3532 	return (error);
3533 }
3534 
3535 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3536     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3537     sysctl_jail_jailed, "I", "Process in jail?");
3538 
3539 static int
3540 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3541 {
3542 	int error, havevnet;
3543 #ifdef VIMAGE
3544 	struct ucred *cred = req->td->td_ucred;
3545 
3546 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3547 #else
3548 	havevnet = 0;
3549 #endif
3550 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3551 
3552 	return (error);
3553 }
3554 
3555 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3556     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3557     sysctl_jail_vnet, "I", "Jail owns vnet?");
3558 
3559 #if defined(INET) || defined(INET6)
3560 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3561     &jail_max_af_ips, 0,
3562     "Number of IP addresses a jail may have at most per address family (deprecated)");
3563 #endif
3564 
3565 /*
3566  * Default parameters for jail(2) compatibility.  For historical reasons,
3567  * the sysctl names have varying similarity to the parameter names.  Prisons
3568  * just see their own parameters, and can't change them.
3569  */
3570 static int
3571 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3572 {
3573 	struct prison *pr;
3574 	int allow, error, i;
3575 
3576 	pr = req->td->td_ucred->cr_prison;
3577 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3578 
3579 	/* Get the current flag value, and convert it to a boolean. */
3580 	i = (allow & arg2) ? 1 : 0;
3581 	if (arg1 != NULL)
3582 		i = !i;
3583 	error = sysctl_handle_int(oidp, &i, 0, req);
3584 	if (error || !req->newptr)
3585 		return (error);
3586 	i = i ? arg2 : 0;
3587 	if (arg1 != NULL)
3588 		i ^= arg2;
3589 	/*
3590 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3591 	 * for writing.
3592 	 */
3593 	mtx_lock(&prison0.pr_mtx);
3594 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3595 	mtx_unlock(&prison0.pr_mtx);
3596 	return (0);
3597 }
3598 
3599 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3600     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3601     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3602     "Processes in jail can set their hostnames (deprecated)");
3603 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3604     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3605     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3606     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3607 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3608     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3609     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3610     "Processes in jail can use System V IPC primitives (deprecated)");
3611 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3612     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3613     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3614     "Prison root can create raw sockets (deprecated)");
3615 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3616     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3617     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3618     "Processes in jail can alter system file flags (deprecated)");
3619 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3620     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3621     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3622     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3623 
3624 static int
3625 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3626 {
3627 	struct prison *pr;
3628 	int level, error;
3629 
3630 	pr = req->td->td_ucred->cr_prison;
3631 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3632 	error = sysctl_handle_int(oidp, &level, 0, req);
3633 	if (error || !req->newptr)
3634 		return (error);
3635 	*(int *)arg1 = level;
3636 	return (0);
3637 }
3638 
3639 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3640     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3641     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3642     sysctl_jail_default_level, "I",
3643     "Processes in jail cannot see all mounted file systems (deprecated)");
3644 
3645 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3646     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3647     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3648     sysctl_jail_default_level, "I",
3649     "Ruleset for the devfs filesystem in jail (deprecated)");
3650 
3651 /*
3652  * Nodes to describe jail parameters.  Maximum length of string parameters
3653  * is returned in the string itself, and the other parameters exist merely
3654  * to make themselves and their types known.
3655  */
3656 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3657     "Jail parameters");
3658 
3659 int
3660 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3661 {
3662 	int i;
3663 	long l;
3664 	size_t s;
3665 	char numbuf[12];
3666 
3667 	switch (oidp->oid_kind & CTLTYPE)
3668 	{
3669 	case CTLTYPE_LONG:
3670 	case CTLTYPE_ULONG:
3671 		l = 0;
3672 #ifdef SCTL_MASK32
3673 		if (!(req->flags & SCTL_MASK32))
3674 #endif
3675 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3676 	case CTLTYPE_INT:
3677 	case CTLTYPE_UINT:
3678 		i = 0;
3679 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3680 	case CTLTYPE_STRING:
3681 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3682 		return
3683 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3684 	case CTLTYPE_STRUCT:
3685 		s = (size_t)arg2;
3686 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3687 	}
3688 	return (0);
3689 }
3690 
3691 /*
3692  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3693  * jail creation time but cannot be changed in an existing jail.
3694  */
3695 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3696 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3697 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3698 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3699 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3700     "I", "Jail secure level");
3701 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3702     "Jail value for kern.osreldate and uname -K");
3703 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3704     "Jail value for kern.osrelease and uname -r");
3705 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3706     "I", "Jail cannot see all mounted file systems");
3707 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3708     "I", "Ruleset for in-jail devfs mounts");
3709 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3710     "B", "Jail persistence");
3711 #ifdef VIMAGE
3712 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3713     "E,jailsys", "Virtual network stack");
3714 #endif
3715 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3716     "B", "Jail is in the process of shutting down");
3717 
3718 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3719 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3720     "I", "Current number of child jails");
3721 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3722     "I", "Maximum number of child jails");
3723 
3724 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3725 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3726     "Jail hostname");
3727 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3728     "Jail NIS domainname");
3729 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3730     "Jail host UUID");
3731 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3732     "LU", "Jail host ID");
3733 
3734 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3735 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3736 
3737 #ifdef INET
3738 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3739     "Jail IPv4 address virtualization");
3740 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3741     "S,in_addr,a", "Jail IPv4 addresses");
3742 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3743     "B", "Do (not) use IPv4 source address selection rather than the "
3744     "primary jail IPv4 address.");
3745 #endif
3746 #ifdef INET6
3747 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3748     "Jail IPv6 address virtualization");
3749 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3750     "S,in6_addr,a", "Jail IPv6 addresses");
3751 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3752     "B", "Do (not) use IPv6 source address selection rather than the "
3753     "primary jail IPv6 address.");
3754 #endif
3755 
3756 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3757 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3758     "B", "Jail may set hostname");
3759 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3760     "B", "Jail may use SYSV IPC");
3761 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3762     "B", "Jail may create raw sockets");
3763 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3764     "B", "Jail may alter system file flags");
3765 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3766     "B", "Jail may set file quotas");
3767 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3768     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3769 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3770     "B", "Jail may lock (unlock) physical pages in memory");
3771 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3772     "B", "Jail may bind sockets to reserved ports");
3773 
3774 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3775 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3776     "B", "Jail may mount/unmount jail-friendly file systems in general");
3777 
3778 /*
3779  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
3780  * its associated bit in the pr_allow bitmask, or zero if the parameter was
3781  * not created.
3782  */
3783 unsigned
3784 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3785     const char *descr)
3786 {
3787 	struct bool_flags *bf;
3788 	struct sysctl_oid *parent;
3789 	char *allow_name, *allow_noname, *allowed;
3790 #ifndef NO_SYSCTL_DESCR
3791 	char *descr_deprecated;
3792 #endif
3793 	unsigned allow_flag;
3794 
3795 	if (prefix
3796 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3797 		< 0 ||
3798 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3799 		< 0
3800 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3801 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3802 		free(allow_name, M_PRISON);
3803 		return 0;
3804 	}
3805 
3806 	/*
3807 	 * See if this parameter has already beed added, i.e. a module was
3808 	 * previously loaded/unloaded.
3809 	 */
3810 	mtx_lock(&prison0.pr_mtx);
3811 	for (bf = pr_flag_allow;
3812 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3813 	     bf++) {
3814 		if (strcmp(bf->name, allow_name) == 0) {
3815 			allow_flag = bf->flag;
3816 			goto no_add;
3817 		}
3818 	}
3819 
3820 	/*
3821 	 * Find a free bit in prison0's pr_allow, failing if there are none
3822 	 * (which shouldn't happen as long as we keep track of how many
3823 	 * potential dynamic flags exist).
3824 	 */
3825 	for (allow_flag = 1;; allow_flag <<= 1) {
3826 		if (allow_flag == 0)
3827 			goto no_add;
3828 		if ((prison0.pr_allow & allow_flag) == 0)
3829 			break;
3830 	}
3831 
3832 	/*
3833 	 * Note the parameter in the next open slot in pr_flag_allow.
3834 	 * Set the flag last so code that checks pr_flag_allow can do so
3835 	 * without locking.
3836 	 */
3837 	for (bf = pr_flag_allow; bf->flag != 0; bf++)
3838 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3839 			/* This should never happen, but is not fatal. */
3840 			allow_flag = 0;
3841 			goto no_add;
3842 		}
3843 	prison0.pr_allow |= allow_flag;
3844 	bf->name = allow_name;
3845 	bf->noname = allow_noname;
3846 	bf->flag = allow_flag;
3847 	mtx_unlock(&prison0.pr_mtx);
3848 
3849 	/*
3850 	 * Create sysctls for the paramter, and the back-compat global
3851 	 * permission.
3852 	 */
3853 	parent = prefix
3854 	    ? SYSCTL_ADD_NODE(NULL,
3855 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3856 		  OID_AUTO, prefix, 0, 0, prefix_descr)
3857 	    : &sysctl___security_jail_param_allow;
3858 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3859 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3860 	    NULL, 0, sysctl_jail_param, "B", descr);
3861 	if ((prefix
3862 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3863 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3864 #ifndef NO_SYSCTL_DESCR
3865 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3866 		    descr);
3867 #endif
3868 		(void)SYSCTL_ADD_PROC(NULL,
3869 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3870 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3871 		    sysctl_jail_default_allow, "I", descr_deprecated);
3872 #ifndef NO_SYSCTL_DESCR
3873 		free(descr_deprecated, M_TEMP);
3874 #endif
3875 		free(allowed, M_TEMP);
3876 	}
3877 	return allow_flag;
3878 
3879  no_add:
3880 	mtx_unlock(&prison0.pr_mtx);
3881 	free(allow_name, M_PRISON);
3882 	free(allow_noname, M_PRISON);
3883 	return allow_flag;
3884 }
3885 
3886 /*
3887  * The VFS system will register jail-aware filesystems here.  They each get
3888  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3889  * attempts to mount.
3890  */
3891 void
3892 prison_add_vfs(struct vfsconf *vfsp)
3893 {
3894 #ifdef NO_SYSCTL_DESCR
3895 
3896 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3897 	    NULL, NULL);
3898 #else
3899 	char *descr;
3900 
3901 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3902 	    vfsp->vfc_name);
3903 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3904 	    NULL, descr);
3905 	free(descr, M_TEMP);
3906 #endif
3907 }
3908 
3909 #ifdef RACCT
3910 void
3911 prison_racct_foreach(void (*callback)(struct racct *racct,
3912     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3913     void *arg2, void *arg3)
3914 {
3915 	struct prison_racct *prr;
3916 
3917 	ASSERT_RACCT_ENABLED();
3918 
3919 	sx_slock(&allprison_lock);
3920 	if (pre != NULL)
3921 		(pre)();
3922 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3923 		(callback)(prr->prr_racct, arg2, arg3);
3924 	if (post != NULL)
3925 		(post)();
3926 	sx_sunlock(&allprison_lock);
3927 }
3928 
3929 static struct prison_racct *
3930 prison_racct_find_locked(const char *name)
3931 {
3932 	struct prison_racct *prr;
3933 
3934 	ASSERT_RACCT_ENABLED();
3935 	sx_assert(&allprison_lock, SA_XLOCKED);
3936 
3937 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3938 		return (NULL);
3939 
3940 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3941 		if (strcmp(name, prr->prr_name) != 0)
3942 			continue;
3943 
3944 		/* Found prison_racct with a matching name? */
3945 		prison_racct_hold(prr);
3946 		return (prr);
3947 	}
3948 
3949 	/* Add new prison_racct. */
3950 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3951 	racct_create(&prr->prr_racct);
3952 
3953 	strcpy(prr->prr_name, name);
3954 	refcount_init(&prr->prr_refcount, 1);
3955 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3956 
3957 	return (prr);
3958 }
3959 
3960 struct prison_racct *
3961 prison_racct_find(const char *name)
3962 {
3963 	struct prison_racct *prr;
3964 
3965 	ASSERT_RACCT_ENABLED();
3966 
3967 	sx_xlock(&allprison_lock);
3968 	prr = prison_racct_find_locked(name);
3969 	sx_xunlock(&allprison_lock);
3970 	return (prr);
3971 }
3972 
3973 void
3974 prison_racct_hold(struct prison_racct *prr)
3975 {
3976 
3977 	ASSERT_RACCT_ENABLED();
3978 
3979 	refcount_acquire(&prr->prr_refcount);
3980 }
3981 
3982 static void
3983 prison_racct_free_locked(struct prison_racct *prr)
3984 {
3985 
3986 	ASSERT_RACCT_ENABLED();
3987 	sx_assert(&allprison_lock, SA_XLOCKED);
3988 
3989 	if (refcount_release(&prr->prr_refcount)) {
3990 		racct_destroy(&prr->prr_racct);
3991 		LIST_REMOVE(prr, prr_next);
3992 		free(prr, M_PRISON_RACCT);
3993 	}
3994 }
3995 
3996 void
3997 prison_racct_free(struct prison_racct *prr)
3998 {
3999 	int old;
4000 
4001 	ASSERT_RACCT_ENABLED();
4002 	sx_assert(&allprison_lock, SA_UNLOCKED);
4003 
4004 	old = prr->prr_refcount;
4005 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4006 		return;
4007 
4008 	sx_xlock(&allprison_lock);
4009 	prison_racct_free_locked(prr);
4010 	sx_xunlock(&allprison_lock);
4011 }
4012 
4013 static void
4014 prison_racct_attach(struct prison *pr)
4015 {
4016 	struct prison_racct *prr;
4017 
4018 	ASSERT_RACCT_ENABLED();
4019 	sx_assert(&allprison_lock, SA_XLOCKED);
4020 
4021 	prr = prison_racct_find_locked(pr->pr_name);
4022 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4023 
4024 	pr->pr_prison_racct = prr;
4025 }
4026 
4027 /*
4028  * Handle jail renaming.  From the racct point of view, renaming means
4029  * moving from one prison_racct to another.
4030  */
4031 static void
4032 prison_racct_modify(struct prison *pr)
4033 {
4034 #ifdef RCTL
4035 	struct proc *p;
4036 	struct ucred *cred;
4037 #endif
4038 	struct prison_racct *oldprr;
4039 
4040 	ASSERT_RACCT_ENABLED();
4041 
4042 	sx_slock(&allproc_lock);
4043 	sx_xlock(&allprison_lock);
4044 
4045 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4046 		sx_xunlock(&allprison_lock);
4047 		sx_sunlock(&allproc_lock);
4048 		return;
4049 	}
4050 
4051 	oldprr = pr->pr_prison_racct;
4052 	pr->pr_prison_racct = NULL;
4053 
4054 	prison_racct_attach(pr);
4055 
4056 	/*
4057 	 * Move resource utilisation records.
4058 	 */
4059 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4060 
4061 #ifdef RCTL
4062 	/*
4063 	 * Force rctl to reattach rules to processes.
4064 	 */
4065 	FOREACH_PROC_IN_SYSTEM(p) {
4066 		PROC_LOCK(p);
4067 		cred = crhold(p->p_ucred);
4068 		PROC_UNLOCK(p);
4069 		rctl_proc_ucred_changed(p, cred);
4070 		crfree(cred);
4071 	}
4072 #endif
4073 
4074 	sx_sunlock(&allproc_lock);
4075 	prison_racct_free_locked(oldprr);
4076 	sx_xunlock(&allprison_lock);
4077 }
4078 
4079 static void
4080 prison_racct_detach(struct prison *pr)
4081 {
4082 
4083 	ASSERT_RACCT_ENABLED();
4084 	sx_assert(&allprison_lock, SA_UNLOCKED);
4085 
4086 	if (pr->pr_prison_racct == NULL)
4087 		return;
4088 	prison_racct_free(pr->pr_prison_racct);
4089 	pr->pr_prison_racct = NULL;
4090 }
4091 #endif /* RACCT */
4092 
4093 #ifdef DDB
4094 
4095 static void
4096 db_show_prison(struct prison *pr)
4097 {
4098 	struct bool_flags *bf;
4099 	struct jailsys_flags *jsf;
4100 #if defined(INET) || defined(INET6)
4101 	int ii;
4102 #endif
4103 	unsigned f;
4104 #ifdef INET
4105 	char ip4buf[INET_ADDRSTRLEN];
4106 #endif
4107 #ifdef INET6
4108 	char ip6buf[INET6_ADDRSTRLEN];
4109 #endif
4110 
4111 	db_printf("prison %p:\n", pr);
4112 	db_printf(" jid             = %d\n", pr->pr_id);
4113 	db_printf(" name            = %s\n", pr->pr_name);
4114 	db_printf(" parent          = %p\n", pr->pr_parent);
4115 	db_printf(" ref             = %d\n", pr->pr_ref);
4116 	db_printf(" uref            = %d\n", pr->pr_uref);
4117 	db_printf(" path            = %s\n", pr->pr_path);
4118 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4119 	    ? pr->pr_cpuset->cs_id : -1);
4120 #ifdef VIMAGE
4121 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4122 #endif
4123 	db_printf(" root            = %p\n", pr->pr_root);
4124 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4125 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4126 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4127 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4128 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4129 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4130 	db_printf(" flags           = 0x%x", pr->pr_flags);
4131 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4132 		if (pr->pr_flags & bf->flag)
4133 			db_printf(" %s", bf->name);
4134 	for (jsf = pr_flag_jailsys;
4135 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4136 	     jsf++) {
4137 		f = pr->pr_flags & (jsf->disable | jsf->new);
4138 		db_printf(" %-16s= %s\n", jsf->name,
4139 		    (f != 0 && f == jsf->disable) ? "disable"
4140 		    : (f == jsf->new) ? "new"
4141 		    : "inherit");
4142 	}
4143 	db_printf(" allow           = 0x%x", pr->pr_allow);
4144 	for (bf = pr_flag_allow;
4145 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4146 	     bf++)
4147 		if (pr->pr_allow & bf->flag)
4148 			db_printf(" %s", bf->name);
4149 	db_printf("\n");
4150 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4151 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4152 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4153 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4154 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4155 #ifdef INET
4156 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4157 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4158 		db_printf(" %s %s\n",
4159 		    ii == 0 ? "ip4.addr        =" : "                 ",
4160 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4161 #endif
4162 #ifdef INET6
4163 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4164 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4165 		db_printf(" %s %s\n",
4166 		    ii == 0 ? "ip6.addr        =" : "                 ",
4167 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4168 #endif
4169 }
4170 
4171 DB_SHOW_COMMAND(prison, db_show_prison_command)
4172 {
4173 	struct prison *pr;
4174 
4175 	if (!have_addr) {
4176 		/*
4177 		 * Show all prisons in the list, and prison0 which is not
4178 		 * listed.
4179 		 */
4180 		db_show_prison(&prison0);
4181 		if (!db_pager_quit) {
4182 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4183 				db_show_prison(pr);
4184 				if (db_pager_quit)
4185 					break;
4186 			}
4187 		}
4188 		return;
4189 	}
4190 
4191 	if (addr == 0)
4192 		pr = &prison0;
4193 	else {
4194 		/* Look for a prison with the ID and with references. */
4195 		TAILQ_FOREACH(pr, &allprison, pr_list)
4196 			if (pr->pr_id == addr && pr->pr_ref > 0)
4197 				break;
4198 		if (pr == NULL)
4199 			/* Look again, without requiring a reference. */
4200 			TAILQ_FOREACH(pr, &allprison, pr_list)
4201 				if (pr->pr_id == addr)
4202 					break;
4203 		if (pr == NULL)
4204 			/* Assume address points to a valid prison. */
4205 			pr = (struct prison *)addr;
4206 	}
4207 	db_show_prison(pr);
4208 }
4209 
4210 #endif /* DDB */
4211