xref: /freebsd/sys/kern/kern_jail.c (revision 56e53cb8ef000c3ef72337a4095987a932cdedef)
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_compat.h"
35 #include "opt_ddb.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/kernel.h>
42 #include <sys/systm.h>
43 #include <sys/errno.h>
44 #include <sys/sysproto.h>
45 #include <sys/malloc.h>
46 #include <sys/osd.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/taskqueue.h>
50 #include <sys/fcntl.h>
51 #include <sys/jail.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/racct.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,
115 };
116 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
117 
118 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
119 struct	sx allprison_lock;
120 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
121 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
122 LIST_HEAD(, prison_racct) allprison_racct;
123 int	lastprid = 0;
124 
125 static int do_jail_attach(struct thread *td, struct prison *pr);
126 static void prison_complete(void *context, int pending);
127 static void prison_deref(struct prison *pr, int flags);
128 static char *prison_path(struct prison *pr1, struct prison *pr2);
129 static void prison_remove_one(struct prison *pr);
130 #ifdef RACCT
131 static void prison_racct_attach(struct prison *pr);
132 static void prison_racct_modify(struct prison *pr);
133 static void prison_racct_detach(struct prison *pr);
134 #endif
135 
136 /* Flags for prison_deref */
137 #define	PD_DEREF	0x01
138 #define	PD_DEUREF	0x02
139 #define	PD_LOCKED	0x04
140 #define	PD_LIST_SLOCKED	0x08
141 #define	PD_LIST_XLOCKED	0x10
142 
143 /*
144  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
145  * as we cannot figure out the size of a sparse array, or an array without a
146  * terminating entry.
147  */
148 static char *pr_flag_names[] = {
149 	[0] = "persist",
150 #ifdef INET
151 	[7] = "ip4.saddrsel",
152 #endif
153 #ifdef INET6
154 	[8] = "ip6.saddrsel",
155 #endif
156 };
157 const size_t pr_flag_names_size = sizeof(pr_flag_names);
158 
159 static char *pr_flag_nonames[] = {
160 	[0] = "nopersist",
161 #ifdef INET
162 	[7] = "ip4.nosaddrsel",
163 #endif
164 #ifdef INET6
165 	[8] = "ip6.nosaddrsel",
166 #endif
167 };
168 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
169 
170 struct jailsys_flags {
171 	const char	*name;
172 	unsigned	 disable;
173 	unsigned	 new;
174 } pr_flag_jailsys[] = {
175 	{ "host", 0, PR_HOST },
176 #ifdef VIMAGE
177 	{ "vnet", 0, PR_VNET },
178 #endif
179 #ifdef INET
180 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
181 #endif
182 #ifdef INET6
183 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
184 #endif
185 };
186 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
187 
188 static char *pr_allow_names[] = {
189 	"allow.set_hostname",
190 	"allow.sysvipc",
191 	"allow.raw_sockets",
192 	"allow.chflags",
193 	"allow.mount",
194 	"allow.quotas",
195 	"allow.socket_af",
196 	"allow.mount.devfs",
197 	"allow.mount.nullfs",
198 	"allow.mount.zfs",
199 	"allow.mount.procfs",
200 	"allow.mount.tmpfs",
201 	"allow.mount.fdescfs",
202 	"allow.mount.linprocfs",
203 	"allow.mount.linsysfs",
204 	"allow.reserved_ports",
205 };
206 const size_t pr_allow_names_size = sizeof(pr_allow_names);
207 
208 static char *pr_allow_nonames[] = {
209 	"allow.noset_hostname",
210 	"allow.nosysvipc",
211 	"allow.noraw_sockets",
212 	"allow.nochflags",
213 	"allow.nomount",
214 	"allow.noquotas",
215 	"allow.nosocket_af",
216 	"allow.mount.nodevfs",
217 	"allow.mount.nonullfs",
218 	"allow.mount.nozfs",
219 	"allow.mount.noprocfs",
220 	"allow.mount.notmpfs",
221 	"allow.mount.nofdescfs",
222 	"allow.mount.nolinprocfs",
223 	"allow.mount.nolinsysfs",
224 	"allow.noreserved_ports",
225 };
226 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
227 
228 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | PR_ALLOW_RESERVED_PORTS)
229 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
230 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
231 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
232 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
233 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
234 #if defined(INET) || defined(INET6)
235 static unsigned jail_max_af_ips = 255;
236 #endif
237 
238 /*
239  * Initialize the parts of prison0 that can't be static-initialized with
240  * constants.  This is called from proc0_init() after creating thread0 cpuset.
241  */
242 void
243 prison0_init(void)
244 {
245 
246 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
247 	prison0.pr_osreldate = osreldate;
248 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
249 }
250 
251 /*
252  * struct jail_args {
253  *	struct jail *jail;
254  * };
255  */
256 int
257 sys_jail(struct thread *td, struct jail_args *uap)
258 {
259 	uint32_t version;
260 	int error;
261 	struct jail j;
262 
263 	error = copyin(uap->jail, &version, sizeof(uint32_t));
264 	if (error)
265 		return (error);
266 
267 	switch (version) {
268 	case 0:
269 	{
270 		struct jail_v0 j0;
271 
272 		/* FreeBSD single IPv4 jails. */
273 		bzero(&j, sizeof(struct jail));
274 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
275 		if (error)
276 			return (error);
277 		j.version = j0.version;
278 		j.path = j0.path;
279 		j.hostname = j0.hostname;
280 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
281 		break;
282 	}
283 
284 	case 1:
285 		/*
286 		 * Version 1 was used by multi-IPv4 jail implementations
287 		 * that never made it into the official kernel.
288 		 */
289 		return (EINVAL);
290 
291 	case 2:	/* JAIL_API_VERSION */
292 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
293 		error = copyin(uap->jail, &j, sizeof(struct jail));
294 		if (error)
295 			return (error);
296 		break;
297 
298 	default:
299 		/* Sci-Fi jails are not supported, sorry. */
300 		return (EINVAL);
301 	}
302 	return (kern_jail(td, &j));
303 }
304 
305 int
306 kern_jail(struct thread *td, struct jail *j)
307 {
308 	struct iovec optiov[2 * (4 + nitems(pr_allow_names)
309 #ifdef INET
310 			    + 1
311 #endif
312 #ifdef INET6
313 			    + 1
314 #endif
315 			    )];
316 	struct uio opt;
317 	char *u_path, *u_hostname, *u_name;
318 #ifdef INET
319 	uint32_t ip4s;
320 	struct in_addr *u_ip4;
321 #endif
322 #ifdef INET6
323 	struct in6_addr *u_ip6;
324 #endif
325 	size_t tmplen;
326 	int error, enforce_statfs, fi;
327 
328 	bzero(&optiov, sizeof(optiov));
329 	opt.uio_iov = optiov;
330 	opt.uio_iovcnt = 0;
331 	opt.uio_offset = -1;
332 	opt.uio_resid = -1;
333 	opt.uio_segflg = UIO_SYSSPACE;
334 	opt.uio_rw = UIO_READ;
335 	opt.uio_td = td;
336 
337 	/* Set permissions for top-level jails from sysctls. */
338 	if (!jailed(td->td_ucred)) {
339 		for (fi = 0; fi < nitems(pr_allow_names); fi++) {
340 			optiov[opt.uio_iovcnt].iov_base =
341 			    (jail_default_allow & (1 << fi))
342 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
343 			optiov[opt.uio_iovcnt].iov_len =
344 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
345 			opt.uio_iovcnt += 2;
346 		}
347 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
348 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
349 		opt.uio_iovcnt++;
350 		enforce_statfs = jail_default_enforce_statfs;
351 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
352 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
353 		opt.uio_iovcnt++;
354 	}
355 
356 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
357 #ifdef INET
358 	ip4s = (j->version == 0) ? 1 : j->ip4s;
359 	if (ip4s > jail_max_af_ips)
360 		return (EINVAL);
361 	tmplen += ip4s * sizeof(struct in_addr);
362 #else
363 	if (j->ip4s > 0)
364 		return (EINVAL);
365 #endif
366 #ifdef INET6
367 	if (j->ip6s > jail_max_af_ips)
368 		return (EINVAL);
369 	tmplen += j->ip6s * sizeof(struct in6_addr);
370 #else
371 	if (j->ip6s > 0)
372 		return (EINVAL);
373 #endif
374 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
375 	u_hostname = u_path + MAXPATHLEN;
376 	u_name = u_hostname + MAXHOSTNAMELEN;
377 #ifdef INET
378 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
379 #endif
380 #ifdef INET6
381 #ifdef INET
382 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
383 #else
384 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
385 #endif
386 #endif
387 	optiov[opt.uio_iovcnt].iov_base = "path";
388 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
389 	opt.uio_iovcnt++;
390 	optiov[opt.uio_iovcnt].iov_base = u_path;
391 	error = copyinstr(j->path, u_path, MAXPATHLEN,
392 	    &optiov[opt.uio_iovcnt].iov_len);
393 	if (error) {
394 		free(u_path, M_TEMP);
395 		return (error);
396 	}
397 	opt.uio_iovcnt++;
398 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
399 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
400 	opt.uio_iovcnt++;
401 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
402 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
403 	    &optiov[opt.uio_iovcnt].iov_len);
404 	if (error) {
405 		free(u_path, M_TEMP);
406 		return (error);
407 	}
408 	opt.uio_iovcnt++;
409 	if (j->jailname != NULL) {
410 		optiov[opt.uio_iovcnt].iov_base = "name";
411 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
412 		opt.uio_iovcnt++;
413 		optiov[opt.uio_iovcnt].iov_base = u_name;
414 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
415 		    &optiov[opt.uio_iovcnt].iov_len);
416 		if (error) {
417 			free(u_path, M_TEMP);
418 			return (error);
419 		}
420 		opt.uio_iovcnt++;
421 	}
422 #ifdef INET
423 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
424 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
425 	opt.uio_iovcnt++;
426 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
427 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
428 	if (j->version == 0)
429 		u_ip4->s_addr = j->ip4s;
430 	else {
431 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
432 		if (error) {
433 			free(u_path, M_TEMP);
434 			return (error);
435 		}
436 	}
437 	opt.uio_iovcnt++;
438 #endif
439 #ifdef INET6
440 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
441 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
442 	opt.uio_iovcnt++;
443 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
444 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
445 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
446 	if (error) {
447 		free(u_path, M_TEMP);
448 		return (error);
449 	}
450 	opt.uio_iovcnt++;
451 #endif
452 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
453 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
454 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
455 	free(u_path, M_TEMP);
456 	return (error);
457 }
458 
459 
460 /*
461  * struct jail_set_args {
462  *	struct iovec *iovp;
463  *	unsigned int iovcnt;
464  *	int flags;
465  * };
466  */
467 int
468 sys_jail_set(struct thread *td, struct jail_set_args *uap)
469 {
470 	struct uio *auio;
471 	int error;
472 
473 	/* Check that we have an even number of iovecs. */
474 	if (uap->iovcnt & 1)
475 		return (EINVAL);
476 
477 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
478 	if (error)
479 		return (error);
480 	error = kern_jail_set(td, auio, uap->flags);
481 	free(auio, M_IOV);
482 	return (error);
483 }
484 
485 int
486 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
487 {
488 	struct nameidata nd;
489 #ifdef INET
490 	struct in_addr *ip4;
491 #endif
492 #ifdef INET6
493 	struct in6_addr *ip6;
494 #endif
495 	struct vfsopt *opt;
496 	struct vfsoptlist *opts;
497 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
498 	struct vnode *root;
499 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
500 	char *g_path, *osrelstr;
501 #if defined(INET) || defined(INET6)
502 	struct prison *tppr;
503 	void *op;
504 #endif
505 	unsigned long hid;
506 	size_t namelen, onamelen, pnamelen;
507 	int born, created, cuflags, descend, enforce;
508 	int error, errmsg_len, errmsg_pos;
509 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
510 	int fi, jid, jsys, len, level;
511 	int childmax, osreldt, rsnum, slevel;
512 	int fullpath_disabled;
513 #if defined(INET) || defined(INET6)
514 	int ii, ij;
515 #endif
516 #ifdef INET
517 	int ip4s, redo_ip4;
518 #endif
519 #ifdef INET6
520 	int ip6s, redo_ip6;
521 #endif
522 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
523 	unsigned tallow;
524 	char numbuf[12];
525 
526 	error = priv_check(td, PRIV_JAIL_SET);
527 	if (!error && (flags & JAIL_ATTACH))
528 		error = priv_check(td, PRIV_JAIL_ATTACH);
529 	if (error)
530 		return (error);
531 	mypr = td->td_ucred->cr_prison;
532 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
533 		return (EPERM);
534 	if (flags & ~JAIL_SET_MASK)
535 		return (EINVAL);
536 
537 	/*
538 	 * Check all the parameters before committing to anything.  Not all
539 	 * errors can be caught early, but we may as well try.  Also, this
540 	 * takes care of some expensive stuff (path lookup) before getting
541 	 * the allprison lock.
542 	 *
543 	 * XXX Jails are not filesystems, and jail parameters are not mount
544 	 *     options.  But it makes more sense to re-use the vfsopt code
545 	 *     than duplicate it under a different name.
546 	 */
547 	error = vfs_buildopts(optuio, &opts);
548 	if (error)
549 		return (error);
550 #ifdef INET
551 	ip4 = NULL;
552 #endif
553 #ifdef INET6
554 	ip6 = NULL;
555 #endif
556 	g_path = NULL;
557 
558 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
559 	if (!cuflags) {
560 		error = EINVAL;
561 		vfs_opterror(opts, "no valid operation (create or update)");
562 		goto done_errmsg;
563 	}
564 
565 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
566 	if (error == ENOENT)
567 		jid = 0;
568 	else if (error != 0)
569 		goto done_free;
570 
571 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
572 	if (error == ENOENT)
573 		gotslevel = 0;
574 	else if (error != 0)
575 		goto done_free;
576 	else
577 		gotslevel = 1;
578 
579 	error =
580 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
581 	if (error == ENOENT)
582 		gotchildmax = 0;
583 	else if (error != 0)
584 		goto done_free;
585 	else
586 		gotchildmax = 1;
587 
588 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
589 	if (error == ENOENT)
590 		gotenforce = 0;
591 	else if (error != 0)
592 		goto done_free;
593 	else if (enforce < 0 || enforce > 2) {
594 		error = EINVAL;
595 		goto done_free;
596 	} else
597 		gotenforce = 1;
598 
599 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
600 	if (error == ENOENT)
601 		gotrsnum = 0;
602 	else if (error != 0)
603 		goto done_free;
604 	else
605 		gotrsnum = 1;
606 
607 	pr_flags = ch_flags = 0;
608 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
609 		if (pr_flag_names[fi] == NULL)
610 			continue;
611 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
612 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
613 	}
614 	ch_flags |= pr_flags;
615 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
616 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
617 		    sizeof(jsys));
618 		if (error == ENOENT)
619 			continue;
620 		if (error != 0)
621 			goto done_free;
622 		switch (jsys) {
623 		case JAIL_SYS_DISABLE:
624 			if (!pr_flag_jailsys[fi].disable) {
625 				error = EINVAL;
626 				goto done_free;
627 			}
628 			pr_flags |= pr_flag_jailsys[fi].disable;
629 			break;
630 		case JAIL_SYS_NEW:
631 			pr_flags |= pr_flag_jailsys[fi].new;
632 			break;
633 		case JAIL_SYS_INHERIT:
634 			break;
635 		default:
636 			error = EINVAL;
637 			goto done_free;
638 		}
639 		ch_flags |=
640 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
641 	}
642 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
643 	    && !(pr_flags & PR_PERSIST)) {
644 		error = EINVAL;
645 		vfs_opterror(opts, "new jail must persist or attach");
646 		goto done_errmsg;
647 	}
648 #ifdef VIMAGE
649 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
650 		error = EINVAL;
651 		vfs_opterror(opts, "vnet cannot be changed after creation");
652 		goto done_errmsg;
653 	}
654 #endif
655 #ifdef INET
656 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
657 		error = EINVAL;
658 		vfs_opterror(opts, "ip4 cannot be changed after creation");
659 		goto done_errmsg;
660 	}
661 #endif
662 #ifdef INET6
663 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
664 		error = EINVAL;
665 		vfs_opterror(opts, "ip6 cannot be changed after creation");
666 		goto done_errmsg;
667 	}
668 #endif
669 
670 	pr_allow = ch_allow = 0;
671 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
672 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
673 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
674 	}
675 	ch_allow |= pr_allow;
676 
677 	error = vfs_getopt(opts, "name", (void **)&name, &len);
678 	if (error == ENOENT)
679 		name = NULL;
680 	else if (error != 0)
681 		goto done_free;
682 	else {
683 		if (len == 0 || name[len - 1] != '\0') {
684 			error = EINVAL;
685 			goto done_free;
686 		}
687 		if (len > MAXHOSTNAMELEN) {
688 			error = ENAMETOOLONG;
689 			goto done_free;
690 		}
691 	}
692 
693 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
694 	if (error == ENOENT)
695 		host = NULL;
696 	else if (error != 0)
697 		goto done_free;
698 	else {
699 		ch_flags |= PR_HOST;
700 		pr_flags |= PR_HOST;
701 		if (len == 0 || host[len - 1] != '\0') {
702 			error = EINVAL;
703 			goto done_free;
704 		}
705 		if (len > MAXHOSTNAMELEN) {
706 			error = ENAMETOOLONG;
707 			goto done_free;
708 		}
709 	}
710 
711 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
712 	if (error == ENOENT)
713 		domain = NULL;
714 	else if (error != 0)
715 		goto done_free;
716 	else {
717 		ch_flags |= PR_HOST;
718 		pr_flags |= PR_HOST;
719 		if (len == 0 || domain[len - 1] != '\0') {
720 			error = EINVAL;
721 			goto done_free;
722 		}
723 		if (len > MAXHOSTNAMELEN) {
724 			error = ENAMETOOLONG;
725 			goto done_free;
726 		}
727 	}
728 
729 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
730 	if (error == ENOENT)
731 		uuid = NULL;
732 	else if (error != 0)
733 		goto done_free;
734 	else {
735 		ch_flags |= PR_HOST;
736 		pr_flags |= PR_HOST;
737 		if (len == 0 || uuid[len - 1] != '\0') {
738 			error = EINVAL;
739 			goto done_free;
740 		}
741 		if (len > HOSTUUIDLEN) {
742 			error = ENAMETOOLONG;
743 			goto done_free;
744 		}
745 	}
746 
747 #ifdef COMPAT_FREEBSD32
748 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
749 		uint32_t hid32;
750 
751 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
752 		hid = hid32;
753 	} else
754 #endif
755 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
756 	if (error == ENOENT)
757 		gothid = 0;
758 	else if (error != 0)
759 		goto done_free;
760 	else {
761 		gothid = 1;
762 		ch_flags |= PR_HOST;
763 		pr_flags |= PR_HOST;
764 	}
765 
766 #ifdef INET
767 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
768 	if (error == ENOENT)
769 		ip4s = 0;
770 	else if (error != 0)
771 		goto done_free;
772 	else if (ip4s & (sizeof(*ip4) - 1)) {
773 		error = EINVAL;
774 		goto done_free;
775 	} else {
776 		ch_flags |= PR_IP4_USER;
777 		pr_flags |= PR_IP4_USER;
778 		if (ip4s > 0) {
779 			ip4s /= sizeof(*ip4);
780 			if (ip4s > jail_max_af_ips) {
781 				error = EINVAL;
782 				vfs_opterror(opts, "too many IPv4 addresses");
783 				goto done_errmsg;
784 			}
785 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
786 			bcopy(op, ip4, ip4s * sizeof(*ip4));
787 			/*
788 			 * IP addresses are all sorted but ip[0] to preserve
789 			 * the primary IP address as given from userland.
790 			 * This special IP is used for unbound outgoing
791 			 * connections as well for "loopback" traffic in case
792 			 * source address selection cannot find any more fitting
793 			 * address to connect from.
794 			 */
795 			if (ip4s > 1)
796 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
797 				    prison_qcmp_v4);
798 			/*
799 			 * Check for duplicate addresses and do some simple
800 			 * zero and broadcast checks. If users give other bogus
801 			 * addresses it is their problem.
802 			 *
803 			 * We do not have to care about byte order for these
804 			 * checks so we will do them in NBO.
805 			 */
806 			for (ii = 0; ii < ip4s; ii++) {
807 				if (ip4[ii].s_addr == INADDR_ANY ||
808 				    ip4[ii].s_addr == INADDR_BROADCAST) {
809 					error = EINVAL;
810 					goto done_free;
811 				}
812 				if ((ii+1) < ip4s &&
813 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
814 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
815 					error = EINVAL;
816 					goto done_free;
817 				}
818 			}
819 		}
820 	}
821 #endif
822 
823 #ifdef INET6
824 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
825 	if (error == ENOENT)
826 		ip6s = 0;
827 	else if (error != 0)
828 		goto done_free;
829 	else if (ip6s & (sizeof(*ip6) - 1)) {
830 		error = EINVAL;
831 		goto done_free;
832 	} else {
833 		ch_flags |= PR_IP6_USER;
834 		pr_flags |= PR_IP6_USER;
835 		if (ip6s > 0) {
836 			ip6s /= sizeof(*ip6);
837 			if (ip6s > jail_max_af_ips) {
838 				error = EINVAL;
839 				vfs_opterror(opts, "too many IPv6 addresses");
840 				goto done_errmsg;
841 			}
842 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
843 			bcopy(op, ip6, ip6s * sizeof(*ip6));
844 			if (ip6s > 1)
845 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
846 				    prison_qcmp_v6);
847 			for (ii = 0; ii < ip6s; ii++) {
848 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
849 					error = EINVAL;
850 					goto done_free;
851 				}
852 				if ((ii+1) < ip6s &&
853 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
854 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
855 				{
856 					error = EINVAL;
857 					goto done_free;
858 				}
859 			}
860 		}
861 	}
862 #endif
863 
864 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
865 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
866 		error = EINVAL;
867 		vfs_opterror(opts,
868 		    "vnet jails cannot have IP address restrictions");
869 		goto done_errmsg;
870 	}
871 #endif
872 
873 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
874 	if (error == ENOENT)
875 		osrelstr = NULL;
876 	else if (error != 0)
877 		goto done_free;
878 	else {
879 		if (flags & JAIL_UPDATE) {
880 			error = EINVAL;
881 			vfs_opterror(opts,
882 			    "osrelease cannot be changed after creation");
883 			goto done_errmsg;
884 		}
885 		if (len == 0 || len >= OSRELEASELEN) {
886 			error = EINVAL;
887 			vfs_opterror(opts,
888 			    "osrelease string must be 1-%d bytes long",
889 			    OSRELEASELEN - 1);
890 			goto done_errmsg;
891 		}
892 	}
893 
894 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
895 	if (error == ENOENT)
896 		osreldt = 0;
897 	else if (error != 0)
898 		goto done_free;
899 	else {
900 		if (flags & JAIL_UPDATE) {
901 			error = EINVAL;
902 			vfs_opterror(opts,
903 			    "osreldate cannot be changed after creation");
904 			goto done_errmsg;
905 		}
906 		if (osreldt == 0) {
907 			error = EINVAL;
908 			vfs_opterror(opts, "osreldate cannot be 0");
909 			goto done_errmsg;
910 		}
911 	}
912 
913 	fullpath_disabled = 0;
914 	root = NULL;
915 	error = vfs_getopt(opts, "path", (void **)&path, &len);
916 	if (error == ENOENT)
917 		path = NULL;
918 	else if (error != 0)
919 		goto done_free;
920 	else {
921 		if (flags & JAIL_UPDATE) {
922 			error = EINVAL;
923 			vfs_opterror(opts,
924 			    "path cannot be changed after creation");
925 			goto done_errmsg;
926 		}
927 		if (len == 0 || path[len - 1] != '\0') {
928 			error = EINVAL;
929 			goto done_free;
930 		}
931 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
932 		    path, td);
933 		error = namei(&nd);
934 		if (error)
935 			goto done_free;
936 		root = nd.ni_vp;
937 		NDFREE(&nd, NDF_ONLY_PNBUF);
938 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
939 		strlcpy(g_path, path, MAXPATHLEN);
940 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
941 		if (error == 0)
942 			path = g_path;
943 		else if (error == ENODEV) {
944 			/* proceed if sysctl debug.disablefullpath == 1 */
945 			fullpath_disabled = 1;
946 			if (len < 2 || (len == 2 && path[0] == '/'))
947 				path = NULL;
948 		} else {
949 			/* exit on other errors */
950 			goto done_free;
951 		}
952 		if (root->v_type != VDIR) {
953 			error = ENOTDIR;
954 			vput(root);
955 			goto done_free;
956 		}
957 		VOP_UNLOCK(root, 0);
958 		if (fullpath_disabled) {
959 			/* Leave room for a real-root full pathname. */
960 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
961 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
962 				error = ENAMETOOLONG;
963 				vrele(root);
964 				goto done_free;
965 			}
966 		}
967 	}
968 
969 	/*
970 	 * Find the specified jail, or at least its parent.
971 	 * This abuses the file error codes ENOENT and EEXIST.
972 	 */
973 	pr = NULL;
974 	ppr = mypr;
975 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
976 		namelc = strrchr(name, '.');
977 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
978 		if (*p != '\0')
979 			jid = 0;
980 	}
981 	sx_xlock(&allprison_lock);
982 	if (jid != 0) {
983 		/*
984 		 * See if a requested jid already exists.  There is an
985 		 * information leak here if the jid exists but is not within
986 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
987 		 * even though they cannot see the jail, and CREATE | UPDATE
988 		 * will return ENOENT which is not normally a valid error.
989 		 */
990 		if (jid < 0) {
991 			error = EINVAL;
992 			vfs_opterror(opts, "negative jid");
993 			goto done_unlock_list;
994 		}
995 		pr = prison_find(jid);
996 		if (pr != NULL) {
997 			ppr = pr->pr_parent;
998 			/* Create: jid must not exist. */
999 			if (cuflags == JAIL_CREATE) {
1000 				mtx_unlock(&pr->pr_mtx);
1001 				error = EEXIST;
1002 				vfs_opterror(opts, "jail %d already exists",
1003 				    jid);
1004 				goto done_unlock_list;
1005 			}
1006 			if (!prison_ischild(mypr, pr)) {
1007 				mtx_unlock(&pr->pr_mtx);
1008 				pr = NULL;
1009 			} else if (pr->pr_uref == 0) {
1010 				if (!(flags & JAIL_DYING)) {
1011 					mtx_unlock(&pr->pr_mtx);
1012 					error = ENOENT;
1013 					vfs_opterror(opts, "jail %d is dying",
1014 					    jid);
1015 					goto done_unlock_list;
1016 				} else if ((flags & JAIL_ATTACH) ||
1017 				    (pr_flags & PR_PERSIST)) {
1018 					/*
1019 					 * A dying jail might be resurrected
1020 					 * (via attach or persist), but first
1021 					 * it must determine if another jail
1022 					 * has claimed its name.  Accomplish
1023 					 * this by implicitly re-setting the
1024 					 * name.
1025 					 */
1026 					if (name == NULL)
1027 						name = prison_name(mypr, pr);
1028 				}
1029 			}
1030 		}
1031 		if (pr == NULL) {
1032 			/* Update: jid must exist. */
1033 			if (cuflags == JAIL_UPDATE) {
1034 				error = ENOENT;
1035 				vfs_opterror(opts, "jail %d not found", jid);
1036 				goto done_unlock_list;
1037 			}
1038 		}
1039 	}
1040 	/*
1041 	 * If the caller provided a name, look for a jail by that name.
1042 	 * This has different semantics for creates and updates keyed by jid
1043 	 * (where the name must not already exist in a different jail),
1044 	 * and updates keyed by the name itself (where the name must exist
1045 	 * because that is the jail being updated).
1046 	 */
1047 	namelc = NULL;
1048 	if (name != NULL) {
1049 		namelc = strrchr(name, '.');
1050 		if (namelc == NULL)
1051 			namelc = name;
1052 		else {
1053 			/*
1054 			 * This is a hierarchical name.  Split it into the
1055 			 * parent and child names, and make sure the parent
1056 			 * exists or matches an already found jail.
1057 			 */
1058 			if (pr != NULL) {
1059 				if (strncmp(name, ppr->pr_name, namelc - name)
1060 				    || ppr->pr_name[namelc - name] != '\0') {
1061 					mtx_unlock(&pr->pr_mtx);
1062 					error = EINVAL;
1063 					vfs_opterror(opts,
1064 					    "cannot change jail's parent");
1065 					goto done_unlock_list;
1066 				}
1067 			} else {
1068 				*namelc = '\0';
1069 				ppr = prison_find_name(mypr, name);
1070 				if (ppr == NULL) {
1071 					error = ENOENT;
1072 					vfs_opterror(opts,
1073 					    "jail \"%s\" not found", name);
1074 					goto done_unlock_list;
1075 				}
1076 				mtx_unlock(&ppr->pr_mtx);
1077 				*namelc = '.';
1078 			}
1079 			namelc++;
1080 		}
1081 		if (namelc[0] != '\0') {
1082 			pnamelen =
1083 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1084  name_again:
1085 			deadpr = NULL;
1086 			FOREACH_PRISON_CHILD(ppr, tpr) {
1087 				if (tpr != pr && tpr->pr_ref > 0 &&
1088 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1089 					if (pr == NULL &&
1090 					    cuflags != JAIL_CREATE) {
1091 						mtx_lock(&tpr->pr_mtx);
1092 						if (tpr->pr_ref > 0) {
1093 							/*
1094 							 * Use this jail
1095 							 * for updates.
1096 							 */
1097 							if (tpr->pr_uref > 0) {
1098 								pr = tpr;
1099 								break;
1100 							}
1101 							deadpr = tpr;
1102 						}
1103 						mtx_unlock(&tpr->pr_mtx);
1104 					} else if (tpr->pr_uref > 0) {
1105 						/*
1106 						 * Create, or update(jid):
1107 						 * name must not exist in an
1108 						 * active sibling jail.
1109 						 */
1110 						error = EEXIST;
1111 						if (pr != NULL)
1112 							mtx_unlock(&pr->pr_mtx);
1113 						vfs_opterror(opts,
1114 						   "jail \"%s\" already exists",
1115 						   name);
1116 						goto done_unlock_list;
1117 					}
1118 				}
1119 			}
1120 			/* If no active jail is found, use a dying one. */
1121 			if (deadpr != NULL && pr == NULL) {
1122 				if (flags & JAIL_DYING) {
1123 					mtx_lock(&deadpr->pr_mtx);
1124 					if (deadpr->pr_ref == 0) {
1125 						mtx_unlock(&deadpr->pr_mtx);
1126 						goto name_again;
1127 					}
1128 					pr = deadpr;
1129 				} else if (cuflags == JAIL_UPDATE) {
1130 					error = ENOENT;
1131 					vfs_opterror(opts,
1132 					    "jail \"%s\" is dying", name);
1133 					goto done_unlock_list;
1134 				}
1135 			}
1136 			/* Update: name must exist if no jid. */
1137 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1138 				error = ENOENT;
1139 				vfs_opterror(opts, "jail \"%s\" not found",
1140 				    name);
1141 				goto done_unlock_list;
1142 			}
1143 		}
1144 	}
1145 	/* Update: must provide a jid or name. */
1146 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1147 		error = ENOENT;
1148 		vfs_opterror(opts, "update specified no jail");
1149 		goto done_unlock_list;
1150 	}
1151 
1152 	/* If there's no prison to update, create a new one and link it in. */
1153 	if (pr == NULL) {
1154 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1155 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1156 				error = EPERM;
1157 				vfs_opterror(opts, "prison limit exceeded");
1158 				goto done_unlock_list;
1159 			}
1160 		created = 1;
1161 		mtx_lock(&ppr->pr_mtx);
1162 		if (ppr->pr_ref == 0) {
1163 			mtx_unlock(&ppr->pr_mtx);
1164 			error = ENOENT;
1165 			vfs_opterror(opts, "jail \"%s\" not found",
1166 			    prison_name(mypr, ppr));
1167 			goto done_unlock_list;
1168 		}
1169 		ppr->pr_ref++;
1170 		ppr->pr_uref++;
1171 		mtx_unlock(&ppr->pr_mtx);
1172 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1173 		if (jid == 0) {
1174 			/* Find the next free jid. */
1175 			jid = lastprid + 1;
1176  findnext:
1177 			if (jid == JAIL_MAX)
1178 				jid = 1;
1179 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1180 				if (tpr->pr_id < jid)
1181 					continue;
1182 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1183 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1184 					break;
1185 				}
1186 				if (jid == lastprid) {
1187 					error = EAGAIN;
1188 					vfs_opterror(opts,
1189 					    "no available jail IDs");
1190 					free(pr, M_PRISON);
1191 					prison_deref(ppr, PD_DEREF |
1192 					    PD_DEUREF | PD_LIST_XLOCKED);
1193 					goto done_releroot;
1194 				}
1195 				jid++;
1196 				goto findnext;
1197 			}
1198 			lastprid = jid;
1199 		} else {
1200 			/*
1201 			 * The jail already has a jid (that did not yet exist),
1202 			 * so just find where to insert it.
1203 			 */
1204 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1205 				if (tpr->pr_id >= jid) {
1206 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1207 					break;
1208 				}
1209 		}
1210 		if (tpr == NULL)
1211 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1212 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1213 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1214 			tpr->pr_childcount++;
1215 
1216 		pr->pr_parent = ppr;
1217 		pr->pr_id = jid;
1218 
1219 		/* Set some default values, and inherit some from the parent. */
1220 		if (namelc == NULL)
1221 			namelc = "";
1222 		if (path == NULL) {
1223 			path = "/";
1224 			root = mypr->pr_root;
1225 			vref(root);
1226 		}
1227 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1228 		pr->pr_flags |= PR_HOST;
1229 #if defined(INET) || defined(INET6)
1230 #ifdef VIMAGE
1231 		if (!(pr_flags & PR_VNET))
1232 #endif
1233 		{
1234 #ifdef INET
1235 			if (!(ch_flags & PR_IP4_USER))
1236 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1237 			else if (!(pr_flags & PR_IP4_USER)) {
1238 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1239 				if (ppr->pr_ip4 != NULL) {
1240 					pr->pr_ip4s = ppr->pr_ip4s;
1241 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1242 					    sizeof(struct in_addr), M_PRISON,
1243 					    M_WAITOK);
1244 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1245 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1246 				}
1247 			}
1248 #endif
1249 #ifdef INET6
1250 			if (!(ch_flags & PR_IP6_USER))
1251 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1252 			else if (!(pr_flags & PR_IP6_USER)) {
1253 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1254 				if (ppr->pr_ip6 != NULL) {
1255 					pr->pr_ip6s = ppr->pr_ip6s;
1256 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1257 					    sizeof(struct in6_addr), M_PRISON,
1258 					    M_WAITOK);
1259 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1260 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1261 				}
1262 			}
1263 #endif
1264 		}
1265 #endif
1266 		/* Source address selection is always on by default. */
1267 		pr->pr_flags |= _PR_IP_SADDRSEL;
1268 
1269 		pr->pr_securelevel = ppr->pr_securelevel;
1270 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1271 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1272 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1273 
1274 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1275 		if (osrelstr == NULL)
1276 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1277 		else
1278 		    strcpy(pr->pr_osrelease, osrelstr);
1279 
1280 		LIST_INIT(&pr->pr_children);
1281 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1282 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1283 
1284 #ifdef VIMAGE
1285 		/* Allocate a new vnet if specified. */
1286 		pr->pr_vnet = (pr_flags & PR_VNET)
1287 		    ? vnet_alloc() : ppr->pr_vnet;
1288 #endif
1289 		/*
1290 		 * Allocate a dedicated cpuset for each jail.
1291 		 * Unlike other initial settings, this may return an erorr.
1292 		 */
1293 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1294 		if (error) {
1295 			prison_deref(pr, PD_LIST_XLOCKED);
1296 			goto done_releroot;
1297 		}
1298 
1299 		mtx_lock(&pr->pr_mtx);
1300 		/*
1301 		 * New prisons do not yet have a reference, because we do not
1302 		 * want others to see the incomplete prison once the
1303 		 * allprison_lock is downgraded.
1304 		 */
1305 	} else {
1306 		created = 0;
1307 		/*
1308 		 * Grab a reference for existing prisons, to ensure they
1309 		 * continue to exist for the duration of the call.
1310 		 */
1311 		pr->pr_ref++;
1312 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1313 		if ((pr->pr_flags & PR_VNET) &&
1314 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1315 			error = EINVAL;
1316 			vfs_opterror(opts,
1317 			    "vnet jails cannot have IP address restrictions");
1318 			goto done_deref_locked;
1319 		}
1320 #endif
1321 #ifdef INET
1322 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1323 			error = EINVAL;
1324 			vfs_opterror(opts,
1325 			    "ip4 cannot be changed after creation");
1326 			goto done_deref_locked;
1327 		}
1328 #endif
1329 #ifdef INET6
1330 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1331 			error = EINVAL;
1332 			vfs_opterror(opts,
1333 			    "ip6 cannot be changed after creation");
1334 			goto done_deref_locked;
1335 		}
1336 #endif
1337 	}
1338 
1339 	/* Do final error checking before setting anything. */
1340 	if (gotslevel) {
1341 		if (slevel < ppr->pr_securelevel) {
1342 			error = EPERM;
1343 			goto done_deref_locked;
1344 		}
1345 	}
1346 	if (gotchildmax) {
1347 		if (childmax >= ppr->pr_childmax) {
1348 			error = EPERM;
1349 			goto done_deref_locked;
1350 		}
1351 	}
1352 	if (gotenforce) {
1353 		if (enforce < ppr->pr_enforce_statfs) {
1354 			error = EPERM;
1355 			goto done_deref_locked;
1356 		}
1357 	}
1358 	if (gotrsnum) {
1359 		/*
1360 		 * devfs_rsnum is a uint16_t
1361 		 */
1362 		if (rsnum < 0 || rsnum > 65535) {
1363 			error = EINVAL;
1364 			goto done_deref_locked;
1365 		}
1366 		/*
1367 		 * Nested jails always inherit parent's devfs ruleset
1368 		 */
1369 		if (jailed(td->td_ucred)) {
1370 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1371 				error = EPERM;
1372 				goto done_deref_locked;
1373 			} else
1374 				rsnum = ppr->pr_devfs_rsnum;
1375 		}
1376 	}
1377 #ifdef INET
1378 	if (ip4s > 0) {
1379 		if (ppr->pr_flags & PR_IP4) {
1380 			/*
1381 			 * Make sure the new set of IP addresses is a
1382 			 * subset of the parent's list.  Don't worry
1383 			 * about the parent being unlocked, as any
1384 			 * setting is done with allprison_lock held.
1385 			 */
1386 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1387 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1388 					break;
1389 			if (ij == ppr->pr_ip4s) {
1390 				error = EPERM;
1391 				goto done_deref_locked;
1392 			}
1393 			if (ip4s > 1) {
1394 				for (ii = ij = 1; ii < ip4s; ii++) {
1395 					if (ip4[ii].s_addr ==
1396 					    ppr->pr_ip4[0].s_addr)
1397 						continue;
1398 					for (; ij < ppr->pr_ip4s; ij++)
1399 						if (ip4[ii].s_addr ==
1400 						    ppr->pr_ip4[ij].s_addr)
1401 							break;
1402 					if (ij == ppr->pr_ip4s)
1403 						break;
1404 				}
1405 				if (ij == ppr->pr_ip4s) {
1406 					error = EPERM;
1407 					goto done_deref_locked;
1408 				}
1409 			}
1410 		}
1411 		/*
1412 		 * Check for conflicting IP addresses.  We permit them
1413 		 * if there is no more than one IP on each jail.  If
1414 		 * there is a duplicate on a jail with more than one
1415 		 * IP stop checking and return error.
1416 		 */
1417 		tppr = ppr;
1418 #ifdef VIMAGE
1419 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1420 			if (tppr->pr_flags & PR_VNET)
1421 				break;
1422 #endif
1423 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1424 			if (tpr == pr ||
1425 #ifdef VIMAGE
1426 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1427 #endif
1428 			    tpr->pr_uref == 0) {
1429 				descend = 0;
1430 				continue;
1431 			}
1432 			if (!(tpr->pr_flags & PR_IP4_USER))
1433 				continue;
1434 			descend = 0;
1435 			if (tpr->pr_ip4 == NULL ||
1436 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1437 				continue;
1438 			for (ii = 0; ii < ip4s; ii++) {
1439 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1440 				    0) {
1441 					error = EADDRINUSE;
1442 					vfs_opterror(opts,
1443 					    "IPv4 addresses clash");
1444 					goto done_deref_locked;
1445 				}
1446 			}
1447 		}
1448 	}
1449 #endif
1450 #ifdef INET6
1451 	if (ip6s > 0) {
1452 		if (ppr->pr_flags & PR_IP6) {
1453 			/*
1454 			 * Make sure the new set of IP addresses is a
1455 			 * subset of the parent's list.
1456 			 */
1457 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1458 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1459 				    &ppr->pr_ip6[ij]))
1460 					break;
1461 			if (ij == ppr->pr_ip6s) {
1462 				error = EPERM;
1463 				goto done_deref_locked;
1464 			}
1465 			if (ip6s > 1) {
1466 				for (ii = ij = 1; ii < ip6s; ii++) {
1467 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1468 					     &ppr->pr_ip6[0]))
1469 						continue;
1470 					for (; ij < ppr->pr_ip6s; ij++)
1471 						if (IN6_ARE_ADDR_EQUAL(
1472 						    &ip6[ii], &ppr->pr_ip6[ij]))
1473 							break;
1474 					if (ij == ppr->pr_ip6s)
1475 						break;
1476 				}
1477 				if (ij == ppr->pr_ip6s) {
1478 					error = EPERM;
1479 					goto done_deref_locked;
1480 				}
1481 			}
1482 		}
1483 		/* Check for conflicting IP addresses. */
1484 		tppr = ppr;
1485 #ifdef VIMAGE
1486 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1487 			if (tppr->pr_flags & PR_VNET)
1488 				break;
1489 #endif
1490 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1491 			if (tpr == pr ||
1492 #ifdef VIMAGE
1493 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1494 #endif
1495 			    tpr->pr_uref == 0) {
1496 				descend = 0;
1497 				continue;
1498 			}
1499 			if (!(tpr->pr_flags & PR_IP6_USER))
1500 				continue;
1501 			descend = 0;
1502 			if (tpr->pr_ip6 == NULL ||
1503 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1504 				continue;
1505 			for (ii = 0; ii < ip6s; ii++) {
1506 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1507 				    0) {
1508 					error = EADDRINUSE;
1509 					vfs_opterror(opts,
1510 					    "IPv6 addresses clash");
1511 					goto done_deref_locked;
1512 				}
1513 			}
1514 		}
1515 	}
1516 #endif
1517 	onamelen = namelen = 0;
1518 	if (namelc != NULL) {
1519 		/* Give a default name of the jid.  Also allow the name to be
1520 		 * explicitly the jid - but not any other number, and only in
1521 		 * normal form (no leading zero/etc).
1522 		 */
1523 		if (namelc[0] == '\0')
1524 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1525 		else if ((strtoul(namelc, &p, 10) != jid ||
1526 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1527 			error = EINVAL;
1528 			vfs_opterror(opts,
1529 			    "name cannot be numeric (unless it is the jid)");
1530 			goto done_deref_locked;
1531 		}
1532 		/*
1533 		 * Make sure the name isn't too long for the prison or its
1534 		 * children.
1535 		 */
1536 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1537 		onamelen = strlen(pr->pr_name + pnamelen);
1538 		namelen = strlen(namelc);
1539 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1540 			error = ENAMETOOLONG;
1541 			goto done_deref_locked;
1542 		}
1543 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1544 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1545 			    sizeof(pr->pr_name)) {
1546 				error = ENAMETOOLONG;
1547 				goto done_deref_locked;
1548 			}
1549 		}
1550 	}
1551 	if (pr_allow & ~ppr->pr_allow) {
1552 		error = EPERM;
1553 		goto done_deref_locked;
1554 	}
1555 
1556 	/*
1557 	 * Let modules check their parameters.  This requires unlocking and
1558 	 * then re-locking the prison, but this is still a valid state as long
1559 	 * as allprison_lock remains xlocked.
1560 	 */
1561 	mtx_unlock(&pr->pr_mtx);
1562 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1563 	if (error != 0) {
1564 		prison_deref(pr, created
1565 		    ? PD_LIST_XLOCKED
1566 		    : PD_DEREF | PD_LIST_XLOCKED);
1567 		goto done_releroot;
1568 	}
1569 	mtx_lock(&pr->pr_mtx);
1570 
1571 	/* At this point, all valid parameters should have been noted. */
1572 	TAILQ_FOREACH(opt, opts, link) {
1573 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1574 			error = EINVAL;
1575 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1576 			goto done_deref_locked;
1577 		}
1578 	}
1579 
1580 	/* Set the parameters of the prison. */
1581 #ifdef INET
1582 	redo_ip4 = 0;
1583 	if (pr_flags & PR_IP4_USER) {
1584 		pr->pr_flags |= PR_IP4;
1585 		free(pr->pr_ip4, M_PRISON);
1586 		pr->pr_ip4s = ip4s;
1587 		pr->pr_ip4 = ip4;
1588 		ip4 = NULL;
1589 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1590 #ifdef VIMAGE
1591 			if (tpr->pr_flags & PR_VNET) {
1592 				descend = 0;
1593 				continue;
1594 			}
1595 #endif
1596 			if (prison_restrict_ip4(tpr, NULL)) {
1597 				redo_ip4 = 1;
1598 				descend = 0;
1599 			}
1600 		}
1601 	}
1602 #endif
1603 #ifdef INET6
1604 	redo_ip6 = 0;
1605 	if (pr_flags & PR_IP6_USER) {
1606 		pr->pr_flags |= PR_IP6;
1607 		free(pr->pr_ip6, M_PRISON);
1608 		pr->pr_ip6s = ip6s;
1609 		pr->pr_ip6 = ip6;
1610 		ip6 = NULL;
1611 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1612 #ifdef VIMAGE
1613 			if (tpr->pr_flags & PR_VNET) {
1614 				descend = 0;
1615 				continue;
1616 			}
1617 #endif
1618 			if (prison_restrict_ip6(tpr, NULL)) {
1619 				redo_ip6 = 1;
1620 				descend = 0;
1621 			}
1622 		}
1623 	}
1624 #endif
1625 	if (gotslevel) {
1626 		pr->pr_securelevel = slevel;
1627 		/* Set all child jails to be at least this level. */
1628 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1629 			if (tpr->pr_securelevel < slevel)
1630 				tpr->pr_securelevel = slevel;
1631 	}
1632 	if (gotchildmax) {
1633 		pr->pr_childmax = childmax;
1634 		/* Set all child jails to under this limit. */
1635 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1636 			if (tpr->pr_childmax > childmax - level)
1637 				tpr->pr_childmax = childmax > level
1638 				    ? childmax - level : 0;
1639 	}
1640 	if (gotenforce) {
1641 		pr->pr_enforce_statfs = enforce;
1642 		/* Pass this restriction on to the children. */
1643 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1644 			if (tpr->pr_enforce_statfs < enforce)
1645 				tpr->pr_enforce_statfs = enforce;
1646 	}
1647 	if (gotrsnum) {
1648 		pr->pr_devfs_rsnum = rsnum;
1649 		/* Pass this restriction on to the children. */
1650 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1651 			tpr->pr_devfs_rsnum = rsnum;
1652 	}
1653 	if (namelc != NULL) {
1654 		if (ppr == &prison0)
1655 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1656 		else
1657 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1658 			    ppr->pr_name, namelc);
1659 		/* Change this component of child names. */
1660 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1661 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1662 			    strlen(tpr->pr_name + onamelen) + 1);
1663 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1664 		}
1665 	}
1666 	if (path != NULL) {
1667 		/* Try to keep a real-rooted full pathname. */
1668 		if (fullpath_disabled && path[0] == '/' &&
1669 		    strcmp(mypr->pr_path, "/"))
1670 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1671 			    mypr->pr_path, path);
1672 		else
1673 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1674 		pr->pr_root = root;
1675 	}
1676 	if (PR_HOST & ch_flags & ~pr_flags) {
1677 		if (pr->pr_flags & PR_HOST) {
1678 			/*
1679 			 * Copy the parent's host info.  As with pr_ip4 above,
1680 			 * the lack of a lock on the parent is not a problem;
1681 			 * it is always set with allprison_lock at least
1682 			 * shared, and is held exclusively here.
1683 			 */
1684 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1685 			    sizeof(pr->pr_hostname));
1686 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1687 			    sizeof(pr->pr_domainname));
1688 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1689 			    sizeof(pr->pr_hostuuid));
1690 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1691 		}
1692 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1693 		/* Set this prison, and any descendants without PR_HOST. */
1694 		if (host != NULL)
1695 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1696 		if (domain != NULL)
1697 			strlcpy(pr->pr_domainname, domain,
1698 			    sizeof(pr->pr_domainname));
1699 		if (uuid != NULL)
1700 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1701 		if (gothid)
1702 			pr->pr_hostid = hid;
1703 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1704 			if (tpr->pr_flags & PR_HOST)
1705 				descend = 0;
1706 			else {
1707 				if (host != NULL)
1708 					strlcpy(tpr->pr_hostname,
1709 					    pr->pr_hostname,
1710 					    sizeof(tpr->pr_hostname));
1711 				if (domain != NULL)
1712 					strlcpy(tpr->pr_domainname,
1713 					    pr->pr_domainname,
1714 					    sizeof(tpr->pr_domainname));
1715 				if (uuid != NULL)
1716 					strlcpy(tpr->pr_hostuuid,
1717 					    pr->pr_hostuuid,
1718 					    sizeof(tpr->pr_hostuuid));
1719 				if (gothid)
1720 					tpr->pr_hostid = hid;
1721 			}
1722 		}
1723 	}
1724 	if ((tallow = ch_allow & ~pr_allow)) {
1725 		/* Clear allow bits in all children. */
1726 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1727 			tpr->pr_allow &= ~tallow;
1728 	}
1729 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1730 	/*
1731 	 * Persistent prisons get an extra reference, and prisons losing their
1732 	 * persist flag lose that reference.  Only do this for existing prisons
1733 	 * for now, so new ones will remain unseen until after the module
1734 	 * handlers have completed.
1735 	 */
1736 	born = pr->pr_uref == 0;
1737 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1738 		if (pr_flags & PR_PERSIST) {
1739 			pr->pr_ref++;
1740 			pr->pr_uref++;
1741 		} else {
1742 			pr->pr_ref--;
1743 			pr->pr_uref--;
1744 		}
1745 	}
1746 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1747 	mtx_unlock(&pr->pr_mtx);
1748 
1749 #ifdef RACCT
1750 	if (racct_enable && created)
1751 		prison_racct_attach(pr);
1752 #endif
1753 
1754 	/* Locks may have prevented a complete restriction of child IP
1755 	 * addresses.  If so, allocate some more memory and try again.
1756 	 */
1757 #ifdef INET
1758 	while (redo_ip4) {
1759 		ip4s = pr->pr_ip4s;
1760 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1761 		mtx_lock(&pr->pr_mtx);
1762 		redo_ip4 = 0;
1763 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1764 #ifdef VIMAGE
1765 			if (tpr->pr_flags & PR_VNET) {
1766 				descend = 0;
1767 				continue;
1768 			}
1769 #endif
1770 			if (prison_restrict_ip4(tpr, ip4)) {
1771 				if (ip4 != NULL)
1772 					ip4 = NULL;
1773 				else
1774 					redo_ip4 = 1;
1775 			}
1776 		}
1777 		mtx_unlock(&pr->pr_mtx);
1778 	}
1779 #endif
1780 #ifdef INET6
1781 	while (redo_ip6) {
1782 		ip6s = pr->pr_ip6s;
1783 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1784 		mtx_lock(&pr->pr_mtx);
1785 		redo_ip6 = 0;
1786 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1787 #ifdef VIMAGE
1788 			if (tpr->pr_flags & PR_VNET) {
1789 				descend = 0;
1790 				continue;
1791 			}
1792 #endif
1793 			if (prison_restrict_ip6(tpr, ip6)) {
1794 				if (ip6 != NULL)
1795 					ip6 = NULL;
1796 				else
1797 					redo_ip6 = 1;
1798 			}
1799 		}
1800 		mtx_unlock(&pr->pr_mtx);
1801 	}
1802 #endif
1803 
1804 	/* Let the modules do their work. */
1805 	sx_downgrade(&allprison_lock);
1806 	if (born) {
1807 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1808 		if (error) {
1809 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1810 			prison_deref(pr, created
1811 			    ? PD_LIST_SLOCKED
1812 			    : PD_DEREF | PD_LIST_SLOCKED);
1813 			goto done_errmsg;
1814 		}
1815 	}
1816 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1817 	if (error) {
1818 		if (born)
1819 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1820 		prison_deref(pr, created
1821 		    ? PD_LIST_SLOCKED
1822 		    : PD_DEREF | PD_LIST_SLOCKED);
1823 		goto done_errmsg;
1824 	}
1825 
1826 	/* Attach this process to the prison if requested. */
1827 	if (flags & JAIL_ATTACH) {
1828 		mtx_lock(&pr->pr_mtx);
1829 		error = do_jail_attach(td, pr);
1830 		if (error) {
1831 			vfs_opterror(opts, "attach failed");
1832 			if (!created)
1833 				prison_deref(pr, PD_DEREF);
1834 			goto done_errmsg;
1835 		}
1836 	}
1837 
1838 #ifdef RACCT
1839 	if (racct_enable && !created) {
1840 		if (!(flags & JAIL_ATTACH))
1841 			sx_sunlock(&allprison_lock);
1842 		prison_racct_modify(pr);
1843 		if (!(flags & JAIL_ATTACH))
1844 			sx_slock(&allprison_lock);
1845 	}
1846 #endif
1847 
1848 	td->td_retval[0] = pr->pr_id;
1849 
1850 	/*
1851 	 * Now that it is all there, drop the temporary reference from existing
1852 	 * prisons.  Or add a reference to newly created persistent prisons
1853 	 * (which was not done earlier so that the prison would not be publicly
1854 	 * visible).
1855 	 */
1856 	if (!created) {
1857 		prison_deref(pr, (flags & JAIL_ATTACH)
1858 		    ? PD_DEREF
1859 		    : PD_DEREF | PD_LIST_SLOCKED);
1860 	} else {
1861 		if (pr_flags & PR_PERSIST) {
1862 			mtx_lock(&pr->pr_mtx);
1863 			pr->pr_ref++;
1864 			pr->pr_uref++;
1865 			mtx_unlock(&pr->pr_mtx);
1866 		}
1867 		if (!(flags & JAIL_ATTACH))
1868 			sx_sunlock(&allprison_lock);
1869 	}
1870 
1871 	goto done_free;
1872 
1873  done_deref_locked:
1874 	prison_deref(pr, created
1875 	    ? PD_LOCKED | PD_LIST_XLOCKED
1876 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1877 	goto done_releroot;
1878  done_unlock_list:
1879 	sx_xunlock(&allprison_lock);
1880  done_releroot:
1881 	if (root != NULL)
1882 		vrele(root);
1883  done_errmsg:
1884 	if (error) {
1885 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1886 		    &errmsg_len) == 0 && errmsg_len > 0) {
1887 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1888 			if (optuio->uio_segflg == UIO_SYSSPACE)
1889 				bcopy(errmsg,
1890 				    optuio->uio_iov[errmsg_pos].iov_base,
1891 				    errmsg_len);
1892 			else
1893 				copyout(errmsg,
1894 				    optuio->uio_iov[errmsg_pos].iov_base,
1895 				    errmsg_len);
1896 		}
1897 	}
1898  done_free:
1899 #ifdef INET
1900 	free(ip4, M_PRISON);
1901 #endif
1902 #ifdef INET6
1903 	free(ip6, M_PRISON);
1904 #endif
1905 	if (g_path != NULL)
1906 		free(g_path, M_TEMP);
1907 	vfs_freeopts(opts);
1908 	return (error);
1909 }
1910 
1911 
1912 /*
1913  * struct jail_get_args {
1914  *	struct iovec *iovp;
1915  *	unsigned int iovcnt;
1916  *	int flags;
1917  * };
1918  */
1919 int
1920 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1921 {
1922 	struct uio *auio;
1923 	int error;
1924 
1925 	/* Check that we have an even number of iovecs. */
1926 	if (uap->iovcnt & 1)
1927 		return (EINVAL);
1928 
1929 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1930 	if (error)
1931 		return (error);
1932 	error = kern_jail_get(td, auio, uap->flags);
1933 	if (error == 0)
1934 		error = copyout(auio->uio_iov, uap->iovp,
1935 		    uap->iovcnt * sizeof (struct iovec));
1936 	free(auio, M_IOV);
1937 	return (error);
1938 }
1939 
1940 int
1941 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1942 {
1943 	struct prison *pr, *mypr;
1944 	struct vfsopt *opt;
1945 	struct vfsoptlist *opts;
1946 	char *errmsg, *name;
1947 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1948 
1949 	if (flags & ~JAIL_GET_MASK)
1950 		return (EINVAL);
1951 
1952 	/* Get the parameter list. */
1953 	error = vfs_buildopts(optuio, &opts);
1954 	if (error)
1955 		return (error);
1956 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1957 	mypr = td->td_ucred->cr_prison;
1958 
1959 	/*
1960 	 * Find the prison specified by one of: lastjid, jid, name.
1961 	 */
1962 	sx_slock(&allprison_lock);
1963 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1964 	if (error == 0) {
1965 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1966 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1967 				mtx_lock(&pr->pr_mtx);
1968 				if (pr->pr_ref > 0 &&
1969 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1970 					break;
1971 				mtx_unlock(&pr->pr_mtx);
1972 			}
1973 		}
1974 		if (pr != NULL)
1975 			goto found_prison;
1976 		error = ENOENT;
1977 		vfs_opterror(opts, "no jail after %d", jid);
1978 		goto done_unlock_list;
1979 	} else if (error != ENOENT)
1980 		goto done_unlock_list;
1981 
1982 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1983 	if (error == 0) {
1984 		if (jid != 0) {
1985 			pr = prison_find_child(mypr, jid);
1986 			if (pr != NULL) {
1987 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1988 					mtx_unlock(&pr->pr_mtx);
1989 					error = ENOENT;
1990 					vfs_opterror(opts, "jail %d is dying",
1991 					    jid);
1992 					goto done_unlock_list;
1993 				}
1994 				goto found_prison;
1995 			}
1996 			error = ENOENT;
1997 			vfs_opterror(opts, "jail %d not found", jid);
1998 			goto done_unlock_list;
1999 		}
2000 	} else if (error != ENOENT)
2001 		goto done_unlock_list;
2002 
2003 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2004 	if (error == 0) {
2005 		if (len == 0 || name[len - 1] != '\0') {
2006 			error = EINVAL;
2007 			goto done_unlock_list;
2008 		}
2009 		pr = prison_find_name(mypr, name);
2010 		if (pr != NULL) {
2011 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2012 				mtx_unlock(&pr->pr_mtx);
2013 				error = ENOENT;
2014 				vfs_opterror(opts, "jail \"%s\" is dying",
2015 				    name);
2016 				goto done_unlock_list;
2017 			}
2018 			goto found_prison;
2019 		}
2020 		error = ENOENT;
2021 		vfs_opterror(opts, "jail \"%s\" not found", name);
2022 		goto done_unlock_list;
2023 	} else if (error != ENOENT)
2024 		goto done_unlock_list;
2025 
2026 	vfs_opterror(opts, "no jail specified");
2027 	error = ENOENT;
2028 	goto done_unlock_list;
2029 
2030  found_prison:
2031 	/* Get the parameters of the prison. */
2032 	pr->pr_ref++;
2033 	locked = PD_LOCKED;
2034 	td->td_retval[0] = pr->pr_id;
2035 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2036 	if (error != 0 && error != ENOENT)
2037 		goto done_deref;
2038 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2039 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2040 	if (error != 0 && error != ENOENT)
2041 		goto done_deref;
2042 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2043 	if (error != 0 && error != ENOENT)
2044 		goto done_deref;
2045 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2046 	    sizeof(pr->pr_cpuset->cs_id));
2047 	if (error != 0 && error != ENOENT)
2048 		goto done_deref;
2049 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2050 	if (error != 0 && error != ENOENT)
2051 		goto done_deref;
2052 #ifdef INET
2053 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2054 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2055 	if (error != 0 && error != ENOENT)
2056 		goto done_deref;
2057 #endif
2058 #ifdef INET6
2059 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2060 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2061 	if (error != 0 && error != ENOENT)
2062 		goto done_deref;
2063 #endif
2064 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2065 	    sizeof(pr->pr_securelevel));
2066 	if (error != 0 && error != ENOENT)
2067 		goto done_deref;
2068 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2069 	    sizeof(pr->pr_childcount));
2070 	if (error != 0 && error != ENOENT)
2071 		goto done_deref;
2072 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2073 	    sizeof(pr->pr_childmax));
2074 	if (error != 0 && error != ENOENT)
2075 		goto done_deref;
2076 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2077 	if (error != 0 && error != ENOENT)
2078 		goto done_deref;
2079 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2080 	if (error != 0 && error != ENOENT)
2081 		goto done_deref;
2082 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2083 	if (error != 0 && error != ENOENT)
2084 		goto done_deref;
2085 #ifdef COMPAT_FREEBSD32
2086 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2087 		uint32_t hid32 = pr->pr_hostid;
2088 
2089 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2090 	} else
2091 #endif
2092 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2093 	    sizeof(pr->pr_hostid));
2094 	if (error != 0 && error != ENOENT)
2095 		goto done_deref;
2096 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2097 	    sizeof(pr->pr_enforce_statfs));
2098 	if (error != 0 && error != ENOENT)
2099 		goto done_deref;
2100 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2101 	    sizeof(pr->pr_devfs_rsnum));
2102 	if (error != 0 && error != ENOENT)
2103 		goto done_deref;
2104 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2105 		if (pr_flag_names[fi] == NULL)
2106 			continue;
2107 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2108 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2109 		if (error != 0 && error != ENOENT)
2110 			goto done_deref;
2111 		i = !i;
2112 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2113 		if (error != 0 && error != ENOENT)
2114 			goto done_deref;
2115 	}
2116 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
2117 		i = pr->pr_flags &
2118 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2119 		i = pr_flag_jailsys[fi].disable &&
2120 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2121 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2122 		    : JAIL_SYS_INHERIT;
2123 		error =
2124 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2125 		if (error != 0 && error != ENOENT)
2126 			goto done_deref;
2127 	}
2128 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
2129 		if (pr_allow_names[fi] == NULL)
2130 			continue;
2131 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2132 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2133 		if (error != 0 && error != ENOENT)
2134 			goto done_deref;
2135 		i = !i;
2136 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2137 		if (error != 0 && error != ENOENT)
2138 			goto done_deref;
2139 	}
2140 	i = (pr->pr_uref == 0);
2141 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2142 	if (error != 0 && error != ENOENT)
2143 		goto done_deref;
2144 	i = !i;
2145 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2146 	if (error != 0 && error != ENOENT)
2147 		goto done_deref;
2148 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2149 	    sizeof(pr->pr_osreldate));
2150 	if (error != 0 && error != ENOENT)
2151 		goto done_deref;
2152 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2153 	if (error != 0 && error != ENOENT)
2154 		goto done_deref;
2155 
2156 	/* Get the module parameters. */
2157 	mtx_unlock(&pr->pr_mtx);
2158 	locked = 0;
2159 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2160 	if (error)
2161 		goto done_deref;
2162 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2163 
2164 	/* By now, all parameters should have been noted. */
2165 	TAILQ_FOREACH(opt, opts, link) {
2166 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2167 			error = EINVAL;
2168 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2169 			goto done_errmsg;
2170 		}
2171 	}
2172 
2173 	/* Write the fetched parameters back to userspace. */
2174 	error = 0;
2175 	TAILQ_FOREACH(opt, opts, link) {
2176 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2177 			pos = 2 * opt->pos + 1;
2178 			optuio->uio_iov[pos].iov_len = opt->len;
2179 			if (opt->value != NULL) {
2180 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2181 					bcopy(opt->value,
2182 					    optuio->uio_iov[pos].iov_base,
2183 					    opt->len);
2184 				} else {
2185 					error = copyout(opt->value,
2186 					    optuio->uio_iov[pos].iov_base,
2187 					    opt->len);
2188 					if (error)
2189 						break;
2190 				}
2191 			}
2192 		}
2193 	}
2194 	goto done_errmsg;
2195 
2196  done_deref:
2197 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2198 	goto done_errmsg;
2199 
2200  done_unlock_list:
2201 	sx_sunlock(&allprison_lock);
2202  done_errmsg:
2203 	if (error && errmsg_pos >= 0) {
2204 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2205 		errmsg_pos = 2 * errmsg_pos + 1;
2206 		if (errmsg_len > 0) {
2207 			if (optuio->uio_segflg == UIO_SYSSPACE)
2208 				bcopy(errmsg,
2209 				    optuio->uio_iov[errmsg_pos].iov_base,
2210 				    errmsg_len);
2211 			else
2212 				copyout(errmsg,
2213 				    optuio->uio_iov[errmsg_pos].iov_base,
2214 				    errmsg_len);
2215 		}
2216 	}
2217 	vfs_freeopts(opts);
2218 	return (error);
2219 }
2220 
2221 
2222 /*
2223  * struct jail_remove_args {
2224  *	int jid;
2225  * };
2226  */
2227 int
2228 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2229 {
2230 	struct prison *pr, *cpr, *lpr, *tpr;
2231 	int descend, error;
2232 
2233 	error = priv_check(td, PRIV_JAIL_REMOVE);
2234 	if (error)
2235 		return (error);
2236 
2237 	sx_xlock(&allprison_lock);
2238 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2239 	if (pr == NULL) {
2240 		sx_xunlock(&allprison_lock);
2241 		return (EINVAL);
2242 	}
2243 
2244 	/* Remove all descendants of this prison, then remove this prison. */
2245 	pr->pr_ref++;
2246 	if (!LIST_EMPTY(&pr->pr_children)) {
2247 		mtx_unlock(&pr->pr_mtx);
2248 		lpr = NULL;
2249 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2250 			mtx_lock(&cpr->pr_mtx);
2251 			if (cpr->pr_ref > 0) {
2252 				tpr = cpr;
2253 				cpr->pr_ref++;
2254 			} else {
2255 				/* Already removed - do not do it again. */
2256 				tpr = NULL;
2257 			}
2258 			mtx_unlock(&cpr->pr_mtx);
2259 			if (lpr != NULL) {
2260 				mtx_lock(&lpr->pr_mtx);
2261 				prison_remove_one(lpr);
2262 				sx_xlock(&allprison_lock);
2263 			}
2264 			lpr = tpr;
2265 		}
2266 		if (lpr != NULL) {
2267 			mtx_lock(&lpr->pr_mtx);
2268 			prison_remove_one(lpr);
2269 			sx_xlock(&allprison_lock);
2270 		}
2271 		mtx_lock(&pr->pr_mtx);
2272 	}
2273 	prison_remove_one(pr);
2274 	return (0);
2275 }
2276 
2277 static void
2278 prison_remove_one(struct prison *pr)
2279 {
2280 	struct proc *p;
2281 	int deuref;
2282 
2283 	/* If the prison was persistent, it is not anymore. */
2284 	deuref = 0;
2285 	if (pr->pr_flags & PR_PERSIST) {
2286 		pr->pr_ref--;
2287 		deuref = PD_DEUREF;
2288 		pr->pr_flags &= ~PR_PERSIST;
2289 	}
2290 
2291 	/*
2292 	 * jail_remove added a reference.  If that's the only one, remove
2293 	 * the prison now.
2294 	 */
2295 	KASSERT(pr->pr_ref > 0,
2296 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2297 	if (pr->pr_ref == 1) {
2298 		prison_deref(pr,
2299 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2300 		return;
2301 	}
2302 
2303 	mtx_unlock(&pr->pr_mtx);
2304 	sx_xunlock(&allprison_lock);
2305 	/*
2306 	 * Kill all processes unfortunate enough to be attached to this prison.
2307 	 */
2308 	sx_slock(&allproc_lock);
2309 	LIST_FOREACH(p, &allproc, p_list) {
2310 		PROC_LOCK(p);
2311 		if (p->p_state != PRS_NEW && p->p_ucred &&
2312 		    p->p_ucred->cr_prison == pr)
2313 			kern_psignal(p, SIGKILL);
2314 		PROC_UNLOCK(p);
2315 	}
2316 	sx_sunlock(&allproc_lock);
2317 	/* Remove the temporary reference added by jail_remove. */
2318 	prison_deref(pr, deuref | PD_DEREF);
2319 }
2320 
2321 
2322 /*
2323  * struct jail_attach_args {
2324  *	int jid;
2325  * };
2326  */
2327 int
2328 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2329 {
2330 	struct prison *pr;
2331 	int error;
2332 
2333 	error = priv_check(td, PRIV_JAIL_ATTACH);
2334 	if (error)
2335 		return (error);
2336 
2337 	/*
2338 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2339 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2340 	 * But then immediately downgrade it since we don't need to stop
2341 	 * readers.
2342 	 */
2343 	sx_xlock(&allprison_lock);
2344 	sx_downgrade(&allprison_lock);
2345 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2346 	if (pr == NULL) {
2347 		sx_sunlock(&allprison_lock);
2348 		return (EINVAL);
2349 	}
2350 
2351 	/*
2352 	 * Do not allow a process to attach to a prison that is not
2353 	 * considered to be "alive".
2354 	 */
2355 	if (pr->pr_uref == 0) {
2356 		mtx_unlock(&pr->pr_mtx);
2357 		sx_sunlock(&allprison_lock);
2358 		return (EINVAL);
2359 	}
2360 
2361 	return (do_jail_attach(td, pr));
2362 }
2363 
2364 static int
2365 do_jail_attach(struct thread *td, struct prison *pr)
2366 {
2367 	struct proc *p;
2368 	struct ucred *newcred, *oldcred;
2369 	int error;
2370 
2371 	/*
2372 	 * XXX: Note that there is a slight race here if two threads
2373 	 * in the same privileged process attempt to attach to two
2374 	 * different jails at the same time.  It is important for
2375 	 * user processes not to do this, or they might end up with
2376 	 * a process root from one prison, but attached to the jail
2377 	 * of another.
2378 	 */
2379 	pr->pr_ref++;
2380 	pr->pr_uref++;
2381 	mtx_unlock(&pr->pr_mtx);
2382 
2383 	/* Let modules do whatever they need to prepare for attaching. */
2384 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2385 	if (error) {
2386 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2387 		return (error);
2388 	}
2389 	sx_sunlock(&allprison_lock);
2390 
2391 	/*
2392 	 * Reparent the newly attached process to this jail.
2393 	 */
2394 	p = td->td_proc;
2395 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2396 	if (error)
2397 		goto e_revert_osd;
2398 
2399 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2400 	if ((error = change_dir(pr->pr_root, td)) != 0)
2401 		goto e_unlock;
2402 #ifdef MAC
2403 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2404 		goto e_unlock;
2405 #endif
2406 	VOP_UNLOCK(pr->pr_root, 0);
2407 	if ((error = pwd_chroot(td, pr->pr_root)))
2408 		goto e_revert_osd;
2409 
2410 	newcred = crget();
2411 	PROC_LOCK(p);
2412 	oldcred = crcopysafe(p, newcred);
2413 	newcred->cr_prison = pr;
2414 	proc_set_cred(p, newcred);
2415 	setsugid(p);
2416 	PROC_UNLOCK(p);
2417 #ifdef RACCT
2418 	racct_proc_ucred_changed(p, oldcred, newcred);
2419 #endif
2420 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2421 	crfree(oldcred);
2422 	return (0);
2423 
2424  e_unlock:
2425 	VOP_UNLOCK(pr->pr_root, 0);
2426  e_revert_osd:
2427 	/* Tell modules this thread is still in its old jail after all. */
2428 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2429 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2430 	return (error);
2431 }
2432 
2433 
2434 /*
2435  * Returns a locked prison instance, or NULL on failure.
2436  */
2437 struct prison *
2438 prison_find(int prid)
2439 {
2440 	struct prison *pr;
2441 
2442 	sx_assert(&allprison_lock, SX_LOCKED);
2443 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2444 		if (pr->pr_id == prid) {
2445 			mtx_lock(&pr->pr_mtx);
2446 			if (pr->pr_ref > 0)
2447 				return (pr);
2448 			mtx_unlock(&pr->pr_mtx);
2449 		}
2450 	}
2451 	return (NULL);
2452 }
2453 
2454 /*
2455  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2456  */
2457 struct prison *
2458 prison_find_child(struct prison *mypr, int prid)
2459 {
2460 	struct prison *pr;
2461 	int descend;
2462 
2463 	sx_assert(&allprison_lock, SX_LOCKED);
2464 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2465 		if (pr->pr_id == prid) {
2466 			mtx_lock(&pr->pr_mtx);
2467 			if (pr->pr_ref > 0)
2468 				return (pr);
2469 			mtx_unlock(&pr->pr_mtx);
2470 		}
2471 	}
2472 	return (NULL);
2473 }
2474 
2475 /*
2476  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2477  */
2478 struct prison *
2479 prison_find_name(struct prison *mypr, const char *name)
2480 {
2481 	struct prison *pr, *deadpr;
2482 	size_t mylen;
2483 	int descend;
2484 
2485 	sx_assert(&allprison_lock, SX_LOCKED);
2486 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2487  again:
2488 	deadpr = NULL;
2489 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2490 		if (!strcmp(pr->pr_name + mylen, name)) {
2491 			mtx_lock(&pr->pr_mtx);
2492 			if (pr->pr_ref > 0) {
2493 				if (pr->pr_uref > 0)
2494 					return (pr);
2495 				deadpr = pr;
2496 			}
2497 			mtx_unlock(&pr->pr_mtx);
2498 		}
2499 	}
2500 	/* There was no valid prison - perhaps there was a dying one. */
2501 	if (deadpr != NULL) {
2502 		mtx_lock(&deadpr->pr_mtx);
2503 		if (deadpr->pr_ref == 0) {
2504 			mtx_unlock(&deadpr->pr_mtx);
2505 			goto again;
2506 		}
2507 	}
2508 	return (deadpr);
2509 }
2510 
2511 /*
2512  * See if a prison has the specific flag set.
2513  */
2514 int
2515 prison_flag(struct ucred *cred, unsigned flag)
2516 {
2517 
2518 	/* This is an atomic read, so no locking is necessary. */
2519 	return (cred->cr_prison->pr_flags & flag);
2520 }
2521 
2522 int
2523 prison_allow(struct ucred *cred, unsigned flag)
2524 {
2525 
2526 	/* This is an atomic read, so no locking is necessary. */
2527 	return (cred->cr_prison->pr_allow & flag);
2528 }
2529 
2530 /*
2531  * Remove a prison reference.  If that was the last reference, remove the
2532  * prison itself - but not in this context in case there are locks held.
2533  */
2534 void
2535 prison_free_locked(struct prison *pr)
2536 {
2537 	int ref;
2538 
2539 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2540 	ref = --pr->pr_ref;
2541 	mtx_unlock(&pr->pr_mtx);
2542 	if (ref == 0)
2543 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2544 }
2545 
2546 void
2547 prison_free(struct prison *pr)
2548 {
2549 
2550 	mtx_lock(&pr->pr_mtx);
2551 	prison_free_locked(pr);
2552 }
2553 
2554 /*
2555  * Complete a call to either prison_free or prison_proc_free.
2556  */
2557 static void
2558 prison_complete(void *context, int pending)
2559 {
2560 	struct prison *pr = context;
2561 
2562 	sx_xlock(&allprison_lock);
2563 	mtx_lock(&pr->pr_mtx);
2564 	prison_deref(pr, pr->pr_uref
2565 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2566 	    : PD_LOCKED | PD_LIST_XLOCKED);
2567 }
2568 
2569 /*
2570  * Remove a prison reference (usually).  This internal version assumes no
2571  * mutexes are held, except perhaps the prison itself.  If there are no more
2572  * references, release and delist the prison.  On completion, the prison lock
2573  * and the allprison lock are both unlocked.
2574  */
2575 static void
2576 prison_deref(struct prison *pr, int flags)
2577 {
2578 	struct prison *ppr, *tpr;
2579 	int ref, lasturef;
2580 
2581 	if (!(flags & PD_LOCKED))
2582 		mtx_lock(&pr->pr_mtx);
2583 	for (;;) {
2584 		if (flags & PD_DEUREF) {
2585 			KASSERT(pr->pr_uref > 0,
2586 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2587 			     pr->pr_id));
2588 			pr->pr_uref--;
2589 			lasturef = pr->pr_uref == 0;
2590 			if (lasturef)
2591 				pr->pr_ref++;
2592 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2593 		} else
2594 			lasturef = 0;
2595 		if (flags & PD_DEREF) {
2596 			KASSERT(pr->pr_ref > 0,
2597 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2598 			     pr->pr_id));
2599 			pr->pr_ref--;
2600 		}
2601 		ref = pr->pr_ref;
2602 		mtx_unlock(&pr->pr_mtx);
2603 
2604 		/*
2605 		 * Tell the modules if the last user reference was removed
2606 		 * (even it sticks around in dying state).
2607 		 */
2608 		if (lasturef) {
2609 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2610 				sx_xlock(&allprison_lock);
2611 				flags |= PD_LIST_XLOCKED;
2612 			}
2613 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2614 			mtx_lock(&pr->pr_mtx);
2615 			ref = --pr->pr_ref;
2616 			mtx_unlock(&pr->pr_mtx);
2617 		}
2618 
2619 		/* If the prison still has references, nothing else to do. */
2620 		if (ref > 0) {
2621 			if (flags & PD_LIST_SLOCKED)
2622 				sx_sunlock(&allprison_lock);
2623 			else if (flags & PD_LIST_XLOCKED)
2624 				sx_xunlock(&allprison_lock);
2625 			return;
2626 		}
2627 
2628 		if (flags & PD_LIST_SLOCKED) {
2629 			if (!sx_try_upgrade(&allprison_lock)) {
2630 				sx_sunlock(&allprison_lock);
2631 				sx_xlock(&allprison_lock);
2632 			}
2633 		} else if (!(flags & PD_LIST_XLOCKED))
2634 			sx_xlock(&allprison_lock);
2635 
2636 		TAILQ_REMOVE(&allprison, pr, pr_list);
2637 		LIST_REMOVE(pr, pr_sibling);
2638 		ppr = pr->pr_parent;
2639 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2640 			tpr->pr_childcount--;
2641 		sx_xunlock(&allprison_lock);
2642 
2643 #ifdef VIMAGE
2644 		if (pr->pr_vnet != ppr->pr_vnet)
2645 			vnet_destroy(pr->pr_vnet);
2646 #endif
2647 		if (pr->pr_root != NULL)
2648 			vrele(pr->pr_root);
2649 		mtx_destroy(&pr->pr_mtx);
2650 #ifdef INET
2651 		free(pr->pr_ip4, M_PRISON);
2652 #endif
2653 #ifdef INET6
2654 		free(pr->pr_ip6, M_PRISON);
2655 #endif
2656 		if (pr->pr_cpuset != NULL)
2657 			cpuset_rel(pr->pr_cpuset);
2658 		osd_jail_exit(pr);
2659 #ifdef RACCT
2660 		if (racct_enable)
2661 			prison_racct_detach(pr);
2662 #endif
2663 		free(pr, M_PRISON);
2664 
2665 		/* Removing a prison frees a reference on its parent. */
2666 		pr = ppr;
2667 		mtx_lock(&pr->pr_mtx);
2668 		flags = PD_DEREF | PD_DEUREF;
2669 	}
2670 }
2671 
2672 void
2673 prison_hold_locked(struct prison *pr)
2674 {
2675 
2676 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2677 	KASSERT(pr->pr_ref > 0,
2678 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2679 	pr->pr_ref++;
2680 }
2681 
2682 void
2683 prison_hold(struct prison *pr)
2684 {
2685 
2686 	mtx_lock(&pr->pr_mtx);
2687 	prison_hold_locked(pr);
2688 	mtx_unlock(&pr->pr_mtx);
2689 }
2690 
2691 void
2692 prison_proc_hold(struct prison *pr)
2693 {
2694 
2695 	mtx_lock(&pr->pr_mtx);
2696 	KASSERT(pr->pr_uref > 0,
2697 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2698 	pr->pr_uref++;
2699 	mtx_unlock(&pr->pr_mtx);
2700 }
2701 
2702 void
2703 prison_proc_free(struct prison *pr)
2704 {
2705 
2706 	mtx_lock(&pr->pr_mtx);
2707 	KASSERT(pr->pr_uref > 0,
2708 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2709 	if (pr->pr_uref > 1)
2710 		pr->pr_uref--;
2711 	else {
2712 		/*
2713 		 * Don't remove the last user reference in this context, which
2714 		 * is expected to be a process that is not only locked, but
2715 		 * also half dead.
2716 		 */
2717 		pr->pr_ref++;
2718 		mtx_unlock(&pr->pr_mtx);
2719 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2720 		return;
2721 	}
2722 	mtx_unlock(&pr->pr_mtx);
2723 }
2724 
2725 /*
2726  * Check if a jail supports the given address family.
2727  *
2728  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2729  * if not.
2730  */
2731 int
2732 prison_check_af(struct ucred *cred, int af)
2733 {
2734 	struct prison *pr;
2735 	int error;
2736 
2737 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2738 
2739 	pr = cred->cr_prison;
2740 #ifdef VIMAGE
2741 	/* Prisons with their own network stack are not limited. */
2742 	if (prison_owns_vnet(cred))
2743 		return (0);
2744 #endif
2745 
2746 	error = 0;
2747 	switch (af)
2748 	{
2749 #ifdef INET
2750 	case AF_INET:
2751 		if (pr->pr_flags & PR_IP4)
2752 		{
2753 			mtx_lock(&pr->pr_mtx);
2754 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2755 				error = EAFNOSUPPORT;
2756 			mtx_unlock(&pr->pr_mtx);
2757 		}
2758 		break;
2759 #endif
2760 #ifdef INET6
2761 	case AF_INET6:
2762 		if (pr->pr_flags & PR_IP6)
2763 		{
2764 			mtx_lock(&pr->pr_mtx);
2765 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2766 				error = EAFNOSUPPORT;
2767 			mtx_unlock(&pr->pr_mtx);
2768 		}
2769 		break;
2770 #endif
2771 	case AF_LOCAL:
2772 	case AF_ROUTE:
2773 		break;
2774 	default:
2775 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2776 			error = EAFNOSUPPORT;
2777 	}
2778 	return (error);
2779 }
2780 
2781 /*
2782  * Check if given address belongs to the jail referenced by cred (wrapper to
2783  * prison_check_ip[46]).
2784  *
2785  * Returns 0 if jail doesn't restrict the address family or if address belongs
2786  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2787  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2788  */
2789 int
2790 prison_if(struct ucred *cred, struct sockaddr *sa)
2791 {
2792 #ifdef INET
2793 	struct sockaddr_in *sai;
2794 #endif
2795 #ifdef INET6
2796 	struct sockaddr_in6 *sai6;
2797 #endif
2798 	int error;
2799 
2800 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2801 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2802 
2803 #ifdef VIMAGE
2804 	if (prison_owns_vnet(cred))
2805 		return (0);
2806 #endif
2807 
2808 	error = 0;
2809 	switch (sa->sa_family)
2810 	{
2811 #ifdef INET
2812 	case AF_INET:
2813 		sai = (struct sockaddr_in *)sa;
2814 		error = prison_check_ip4(cred, &sai->sin_addr);
2815 		break;
2816 #endif
2817 #ifdef INET6
2818 	case AF_INET6:
2819 		sai6 = (struct sockaddr_in6 *)sa;
2820 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2821 		break;
2822 #endif
2823 	default:
2824 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2825 			error = EAFNOSUPPORT;
2826 	}
2827 	return (error);
2828 }
2829 
2830 /*
2831  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2832  */
2833 int
2834 prison_check(struct ucred *cred1, struct ucred *cred2)
2835 {
2836 
2837 	return ((cred1->cr_prison == cred2->cr_prison ||
2838 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2839 }
2840 
2841 /*
2842  * Return 1 if p2 is a child of p1, otherwise 0.
2843  */
2844 int
2845 prison_ischild(struct prison *pr1, struct prison *pr2)
2846 {
2847 
2848 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2849 		if (pr1 == pr2)
2850 			return (1);
2851 	return (0);
2852 }
2853 
2854 /*
2855  * Return 1 if the passed credential is in a jail, otherwise 0.
2856  */
2857 int
2858 jailed(struct ucred *cred)
2859 {
2860 
2861 	return (cred->cr_prison != &prison0);
2862 }
2863 
2864 /*
2865  * Return 1 if the passed credential is in a jail and that jail does not
2866  * have its own virtual network stack, otherwise 0.
2867  */
2868 int
2869 jailed_without_vnet(struct ucred *cred)
2870 {
2871 
2872 	if (!jailed(cred))
2873 		return (0);
2874 #ifdef VIMAGE
2875 	if (prison_owns_vnet(cred))
2876 		return (0);
2877 #endif
2878 
2879 	return (1);
2880 }
2881 
2882 /*
2883  * Return the correct hostname (domainname, et al) for the passed credential.
2884  */
2885 void
2886 getcredhostname(struct ucred *cred, char *buf, size_t size)
2887 {
2888 	struct prison *pr;
2889 
2890 	/*
2891 	 * A NULL credential can be used to shortcut to the physical
2892 	 * system's hostname.
2893 	 */
2894 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2895 	mtx_lock(&pr->pr_mtx);
2896 	strlcpy(buf, pr->pr_hostname, size);
2897 	mtx_unlock(&pr->pr_mtx);
2898 }
2899 
2900 void
2901 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2902 {
2903 
2904 	mtx_lock(&cred->cr_prison->pr_mtx);
2905 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2906 	mtx_unlock(&cred->cr_prison->pr_mtx);
2907 }
2908 
2909 void
2910 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2911 {
2912 
2913 	mtx_lock(&cred->cr_prison->pr_mtx);
2914 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2915 	mtx_unlock(&cred->cr_prison->pr_mtx);
2916 }
2917 
2918 void
2919 getcredhostid(struct ucred *cred, unsigned long *hostid)
2920 {
2921 
2922 	mtx_lock(&cred->cr_prison->pr_mtx);
2923 	*hostid = cred->cr_prison->pr_hostid;
2924 	mtx_unlock(&cred->cr_prison->pr_mtx);
2925 }
2926 
2927 #ifdef VIMAGE
2928 /*
2929  * Determine whether the prison represented by cred owns
2930  * its vnet rather than having it inherited.
2931  *
2932  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2933  */
2934 int
2935 prison_owns_vnet(struct ucred *cred)
2936 {
2937 
2938 	/*
2939 	 * vnets cannot be added/removed after jail creation,
2940 	 * so no need to lock here.
2941 	 */
2942 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2943 }
2944 #endif
2945 
2946 /*
2947  * Determine whether the subject represented by cred can "see"
2948  * status of a mount point.
2949  * Returns: 0 for permitted, ENOENT otherwise.
2950  * XXX: This function should be called cr_canseemount() and should be
2951  *      placed in kern_prot.c.
2952  */
2953 int
2954 prison_canseemount(struct ucred *cred, struct mount *mp)
2955 {
2956 	struct prison *pr;
2957 	struct statfs *sp;
2958 	size_t len;
2959 
2960 	pr = cred->cr_prison;
2961 	if (pr->pr_enforce_statfs == 0)
2962 		return (0);
2963 	if (pr->pr_root->v_mount == mp)
2964 		return (0);
2965 	if (pr->pr_enforce_statfs == 2)
2966 		return (ENOENT);
2967 	/*
2968 	 * If jail's chroot directory is set to "/" we should be able to see
2969 	 * all mount-points from inside a jail.
2970 	 * This is ugly check, but this is the only situation when jail's
2971 	 * directory ends with '/'.
2972 	 */
2973 	if (strcmp(pr->pr_path, "/") == 0)
2974 		return (0);
2975 	len = strlen(pr->pr_path);
2976 	sp = &mp->mnt_stat;
2977 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2978 		return (ENOENT);
2979 	/*
2980 	 * Be sure that we don't have situation where jail's root directory
2981 	 * is "/some/path" and mount point is "/some/pathpath".
2982 	 */
2983 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2984 		return (ENOENT);
2985 	return (0);
2986 }
2987 
2988 void
2989 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2990 {
2991 	char jpath[MAXPATHLEN];
2992 	struct prison *pr;
2993 	size_t len;
2994 
2995 	pr = cred->cr_prison;
2996 	if (pr->pr_enforce_statfs == 0)
2997 		return;
2998 	if (prison_canseemount(cred, mp) != 0) {
2999 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3000 		strlcpy(sp->f_mntonname, "[restricted]",
3001 		    sizeof(sp->f_mntonname));
3002 		return;
3003 	}
3004 	if (pr->pr_root->v_mount == mp) {
3005 		/*
3006 		 * Clear current buffer data, so we are sure nothing from
3007 		 * the valid path left there.
3008 		 */
3009 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3010 		*sp->f_mntonname = '/';
3011 		return;
3012 	}
3013 	/*
3014 	 * If jail's chroot directory is set to "/" we should be able to see
3015 	 * all mount-points from inside a jail.
3016 	 */
3017 	if (strcmp(pr->pr_path, "/") == 0)
3018 		return;
3019 	len = strlen(pr->pr_path);
3020 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3021 	/*
3022 	 * Clear current buffer data, so we are sure nothing from
3023 	 * the valid path left there.
3024 	 */
3025 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3026 	if (*jpath == '\0') {
3027 		/* Should never happen. */
3028 		*sp->f_mntonname = '/';
3029 	} else {
3030 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3031 	}
3032 }
3033 
3034 /*
3035  * Check with permission for a specific privilege is granted within jail.  We
3036  * have a specific list of accepted privileges; the rest are denied.
3037  */
3038 int
3039 prison_priv_check(struct ucred *cred, int priv)
3040 {
3041 
3042 	if (!jailed(cred))
3043 		return (0);
3044 
3045 #ifdef VIMAGE
3046 	/*
3047 	 * Privileges specific to prisons with a virtual network stack.
3048 	 * There might be a duplicate entry here in case the privilege
3049 	 * is only granted conditionally in the legacy jail case.
3050 	 */
3051 	switch (priv) {
3052 #ifdef notyet
3053 		/*
3054 		 * NFS-specific privileges.
3055 		 */
3056 	case PRIV_NFS_DAEMON:
3057 	case PRIV_NFS_LOCKD:
3058 #endif
3059 		/*
3060 		 * Network stack privileges.
3061 		 */
3062 	case PRIV_NET_BRIDGE:
3063 	case PRIV_NET_GRE:
3064 	case PRIV_NET_BPF:
3065 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3066 	case PRIV_NET_ROUTE:
3067 	case PRIV_NET_TAP:
3068 	case PRIV_NET_SETIFMTU:
3069 	case PRIV_NET_SETIFFLAGS:
3070 	case PRIV_NET_SETIFCAP:
3071 	case PRIV_NET_SETIFDESCR:
3072 	case PRIV_NET_SETIFNAME	:
3073 	case PRIV_NET_SETIFMETRIC:
3074 	case PRIV_NET_SETIFPHYS:
3075 	case PRIV_NET_SETIFMAC:
3076 	case PRIV_NET_ADDMULTI:
3077 	case PRIV_NET_DELMULTI:
3078 	case PRIV_NET_HWIOCTL:
3079 	case PRIV_NET_SETLLADDR:
3080 	case PRIV_NET_ADDIFGROUP:
3081 	case PRIV_NET_DELIFGROUP:
3082 	case PRIV_NET_IFCREATE:
3083 	case PRIV_NET_IFDESTROY:
3084 	case PRIV_NET_ADDIFADDR:
3085 	case PRIV_NET_DELIFADDR:
3086 	case PRIV_NET_LAGG:
3087 	case PRIV_NET_GIF:
3088 	case PRIV_NET_SETIFVNET:
3089 	case PRIV_NET_SETIFFIB:
3090 
3091 		/*
3092 		 * 802.11-related privileges.
3093 		 */
3094 	case PRIV_NET80211_GETKEY:
3095 #ifdef notyet
3096 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3097 #endif
3098 
3099 #ifdef notyet
3100 		/*
3101 		 * ATM privileges.
3102 		 */
3103 	case PRIV_NETATM_CFG:
3104 	case PRIV_NETATM_ADD:
3105 	case PRIV_NETATM_DEL:
3106 	case PRIV_NETATM_SET:
3107 
3108 		/*
3109 		 * Bluetooth privileges.
3110 		 */
3111 	case PRIV_NETBLUETOOTH_RAW:
3112 #endif
3113 
3114 		/*
3115 		 * Netgraph and netgraph module privileges.
3116 		 */
3117 	case PRIV_NETGRAPH_CONTROL:
3118 #ifdef notyet
3119 	case PRIV_NETGRAPH_TTY:
3120 #endif
3121 
3122 		/*
3123 		 * IPv4 and IPv6 privileges.
3124 		 */
3125 	case PRIV_NETINET_IPFW:
3126 	case PRIV_NETINET_DIVERT:
3127 	case PRIV_NETINET_PF:
3128 	case PRIV_NETINET_DUMMYNET:
3129 	case PRIV_NETINET_CARP:
3130 	case PRIV_NETINET_MROUTE:
3131 	case PRIV_NETINET_RAW:
3132 	case PRIV_NETINET_ADDRCTRL6:
3133 	case PRIV_NETINET_ND6:
3134 	case PRIV_NETINET_SCOPE6:
3135 	case PRIV_NETINET_ALIFETIME6:
3136 	case PRIV_NETINET_IPSEC:
3137 	case PRIV_NETINET_BINDANY:
3138 
3139 #ifdef notyet
3140 		/*
3141 		 * NCP privileges.
3142 		 */
3143 	case PRIV_NETNCP:
3144 
3145 		/*
3146 		 * SMB privileges.
3147 		 */
3148 	case PRIV_NETSMB:
3149 #endif
3150 
3151 	/*
3152 	 * No default: or deny here.
3153 	 * In case of no permit fall through to next switch().
3154 	 */
3155 		if (cred->cr_prison->pr_flags & PR_VNET)
3156 			return (0);
3157 	}
3158 #endif /* VIMAGE */
3159 
3160 	switch (priv) {
3161 
3162 		/*
3163 		 * Allow ktrace privileges for root in jail.
3164 		 */
3165 	case PRIV_KTRACE:
3166 
3167 #if 0
3168 		/*
3169 		 * Allow jailed processes to configure audit identity and
3170 		 * submit audit records (login, etc).  In the future we may
3171 		 * want to further refine the relationship between audit and
3172 		 * jail.
3173 		 */
3174 	case PRIV_AUDIT_GETAUDIT:
3175 	case PRIV_AUDIT_SETAUDIT:
3176 	case PRIV_AUDIT_SUBMIT:
3177 #endif
3178 
3179 		/*
3180 		 * Allow jailed processes to manipulate process UNIX
3181 		 * credentials in any way they see fit.
3182 		 */
3183 	case PRIV_CRED_SETUID:
3184 	case PRIV_CRED_SETEUID:
3185 	case PRIV_CRED_SETGID:
3186 	case PRIV_CRED_SETEGID:
3187 	case PRIV_CRED_SETGROUPS:
3188 	case PRIV_CRED_SETREUID:
3189 	case PRIV_CRED_SETREGID:
3190 	case PRIV_CRED_SETRESUID:
3191 	case PRIV_CRED_SETRESGID:
3192 
3193 		/*
3194 		 * Jail implements visibility constraints already, so allow
3195 		 * jailed root to override uid/gid-based constraints.
3196 		 */
3197 	case PRIV_SEEOTHERGIDS:
3198 	case PRIV_SEEOTHERUIDS:
3199 
3200 		/*
3201 		 * Jail implements inter-process debugging limits already, so
3202 		 * allow jailed root various debugging privileges.
3203 		 */
3204 	case PRIV_DEBUG_DIFFCRED:
3205 	case PRIV_DEBUG_SUGID:
3206 	case PRIV_DEBUG_UNPRIV:
3207 
3208 		/*
3209 		 * Allow jail to set various resource limits and login
3210 		 * properties, and for now, exceed process resource limits.
3211 		 */
3212 	case PRIV_PROC_LIMIT:
3213 	case PRIV_PROC_SETLOGIN:
3214 	case PRIV_PROC_SETRLIMIT:
3215 
3216 		/*
3217 		 * System V and POSIX IPC privileges are granted in jail.
3218 		 */
3219 	case PRIV_IPC_READ:
3220 	case PRIV_IPC_WRITE:
3221 	case PRIV_IPC_ADMIN:
3222 	case PRIV_IPC_MSGSIZE:
3223 	case PRIV_MQ_ADMIN:
3224 
3225 		/*
3226 		 * Jail operations within a jail work on child jails.
3227 		 */
3228 	case PRIV_JAIL_ATTACH:
3229 	case PRIV_JAIL_SET:
3230 	case PRIV_JAIL_REMOVE:
3231 
3232 		/*
3233 		 * Jail implements its own inter-process limits, so allow
3234 		 * root processes in jail to change scheduling on other
3235 		 * processes in the same jail.  Likewise for signalling.
3236 		 */
3237 	case PRIV_SCHED_DIFFCRED:
3238 	case PRIV_SCHED_CPUSET:
3239 	case PRIV_SIGNAL_DIFFCRED:
3240 	case PRIV_SIGNAL_SUGID:
3241 
3242 		/*
3243 		 * Allow jailed processes to write to sysctls marked as jail
3244 		 * writable.
3245 		 */
3246 	case PRIV_SYSCTL_WRITEJAIL:
3247 
3248 		/*
3249 		 * Allow root in jail to manage a variety of quota
3250 		 * properties.  These should likely be conditional on a
3251 		 * configuration option.
3252 		 */
3253 	case PRIV_VFS_GETQUOTA:
3254 	case PRIV_VFS_SETQUOTA:
3255 
3256 		/*
3257 		 * Since Jail relies on chroot() to implement file system
3258 		 * protections, grant many VFS privileges to root in jail.
3259 		 * Be careful to exclude mount-related and NFS-related
3260 		 * privileges.
3261 		 */
3262 	case PRIV_VFS_READ:
3263 	case PRIV_VFS_WRITE:
3264 	case PRIV_VFS_ADMIN:
3265 	case PRIV_VFS_EXEC:
3266 	case PRIV_VFS_LOOKUP:
3267 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3268 	case PRIV_VFS_CHFLAGS_DEV:
3269 	case PRIV_VFS_CHOWN:
3270 	case PRIV_VFS_CHROOT:
3271 	case PRIV_VFS_RETAINSUGID:
3272 	case PRIV_VFS_FCHROOT:
3273 	case PRIV_VFS_LINK:
3274 	case PRIV_VFS_SETGID:
3275 	case PRIV_VFS_STAT:
3276 	case PRIV_VFS_STICKYFILE:
3277 
3278 		/*
3279 		 * As in the non-jail case, non-root users are expected to be
3280 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3281 		 * exists in the jail and they have permission to access it).
3282 		 */
3283 	case PRIV_KMEM_READ:
3284 		return (0);
3285 
3286 		/*
3287 		 * Depending on the global setting, allow privilege of
3288 		 * setting system flags.
3289 		 */
3290 	case PRIV_VFS_SYSFLAGS:
3291 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3292 			return (0);
3293 		else
3294 			return (EPERM);
3295 
3296 		/*
3297 		 * Depending on the global setting, allow privilege of
3298 		 * mounting/unmounting file systems.
3299 		 */
3300 	case PRIV_VFS_MOUNT:
3301 	case PRIV_VFS_UNMOUNT:
3302 	case PRIV_VFS_MOUNT_NONUSER:
3303 	case PRIV_VFS_MOUNT_OWNER:
3304 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3305 		    cred->cr_prison->pr_enforce_statfs < 2)
3306 			return (0);
3307 		else
3308 			return (EPERM);
3309 
3310 		/*
3311 		 * Conditionally allow jailed root to bind reserved ports.
3312 		 */
3313 	case PRIV_NETINET_RESERVEDPORT:
3314 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3315 			return (0);
3316 		else
3317 			return (EPERM);
3318 
3319 		/*
3320 		 * Allow jailed root to reuse in-use ports.
3321 		 */
3322 	case PRIV_NETINET_REUSEPORT:
3323 		return (0);
3324 
3325 		/*
3326 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3327 		 */
3328 	case PRIV_NETINET_SETHDROPTS:
3329 		return (0);
3330 
3331 		/*
3332 		 * Conditionally allow creating raw sockets in jail.
3333 		 */
3334 	case PRIV_NETINET_RAW:
3335 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3336 			return (0);
3337 		else
3338 			return (EPERM);
3339 
3340 		/*
3341 		 * Since jail implements its own visibility limits on netstat
3342 		 * sysctls, allow getcred.  This allows identd to work in
3343 		 * jail.
3344 		 */
3345 	case PRIV_NETINET_GETCRED:
3346 		return (0);
3347 
3348 		/*
3349 		 * Allow jailed root to set loginclass.
3350 		 */
3351 	case PRIV_PROC_SETLOGINCLASS:
3352 		return (0);
3353 
3354 	default:
3355 		/*
3356 		 * In all remaining cases, deny the privilege request.  This
3357 		 * includes almost all network privileges, many system
3358 		 * configuration privileges.
3359 		 */
3360 		return (EPERM);
3361 	}
3362 }
3363 
3364 /*
3365  * Return the part of pr2's name that is relative to pr1, or the whole name
3366  * if it does not directly follow.
3367  */
3368 
3369 char *
3370 prison_name(struct prison *pr1, struct prison *pr2)
3371 {
3372 	char *name;
3373 
3374 	/* Jails see themselves as "0" (if they see themselves at all). */
3375 	if (pr1 == pr2)
3376 		return "0";
3377 	name = pr2->pr_name;
3378 	if (prison_ischild(pr1, pr2)) {
3379 		/*
3380 		 * pr1 isn't locked (and allprison_lock may not be either)
3381 		 * so its length can't be counted on.  But the number of dots
3382 		 * can be counted on - and counted.
3383 		 */
3384 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3385 			name = strchr(name, '.') + 1;
3386 	}
3387 	return (name);
3388 }
3389 
3390 /*
3391  * Return the part of pr2's path that is relative to pr1, or the whole path
3392  * if it does not directly follow.
3393  */
3394 static char *
3395 prison_path(struct prison *pr1, struct prison *pr2)
3396 {
3397 	char *path1, *path2;
3398 	int len1;
3399 
3400 	path1 = pr1->pr_path;
3401 	path2 = pr2->pr_path;
3402 	if (!strcmp(path1, "/"))
3403 		return (path2);
3404 	len1 = strlen(path1);
3405 	if (strncmp(path1, path2, len1))
3406 		return (path2);
3407 	if (path2[len1] == '\0')
3408 		return "/";
3409 	if (path2[len1] == '/')
3410 		return (path2 + len1);
3411 	return (path2);
3412 }
3413 
3414 
3415 /*
3416  * Jail-related sysctls.
3417  */
3418 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3419     "Jails");
3420 
3421 static int
3422 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3423 {
3424 	struct xprison *xp;
3425 	struct prison *pr, *cpr;
3426 #ifdef INET
3427 	struct in_addr *ip4 = NULL;
3428 	int ip4s = 0;
3429 #endif
3430 #ifdef INET6
3431 	struct in6_addr *ip6 = NULL;
3432 	int ip6s = 0;
3433 #endif
3434 	int descend, error;
3435 
3436 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3437 	pr = req->td->td_ucred->cr_prison;
3438 	error = 0;
3439 	sx_slock(&allprison_lock);
3440 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3441 #if defined(INET) || defined(INET6)
3442  again:
3443 #endif
3444 		mtx_lock(&cpr->pr_mtx);
3445 #ifdef INET
3446 		if (cpr->pr_ip4s > 0) {
3447 			if (ip4s < cpr->pr_ip4s) {
3448 				ip4s = cpr->pr_ip4s;
3449 				mtx_unlock(&cpr->pr_mtx);
3450 				ip4 = realloc(ip4, ip4s *
3451 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3452 				goto again;
3453 			}
3454 			bcopy(cpr->pr_ip4, ip4,
3455 			    cpr->pr_ip4s * sizeof(struct in_addr));
3456 		}
3457 #endif
3458 #ifdef INET6
3459 		if (cpr->pr_ip6s > 0) {
3460 			if (ip6s < cpr->pr_ip6s) {
3461 				ip6s = cpr->pr_ip6s;
3462 				mtx_unlock(&cpr->pr_mtx);
3463 				ip6 = realloc(ip6, ip6s *
3464 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3465 				goto again;
3466 			}
3467 			bcopy(cpr->pr_ip6, ip6,
3468 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3469 		}
3470 #endif
3471 		if (cpr->pr_ref == 0) {
3472 			mtx_unlock(&cpr->pr_mtx);
3473 			continue;
3474 		}
3475 		bzero(xp, sizeof(*xp));
3476 		xp->pr_version = XPRISON_VERSION;
3477 		xp->pr_id = cpr->pr_id;
3478 		xp->pr_state = cpr->pr_uref > 0
3479 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3480 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3481 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3482 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3483 #ifdef INET
3484 		xp->pr_ip4s = cpr->pr_ip4s;
3485 #endif
3486 #ifdef INET6
3487 		xp->pr_ip6s = cpr->pr_ip6s;
3488 #endif
3489 		mtx_unlock(&cpr->pr_mtx);
3490 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3491 		if (error)
3492 			break;
3493 #ifdef INET
3494 		if (xp->pr_ip4s > 0) {
3495 			error = SYSCTL_OUT(req, ip4,
3496 			    xp->pr_ip4s * sizeof(struct in_addr));
3497 			if (error)
3498 				break;
3499 		}
3500 #endif
3501 #ifdef INET6
3502 		if (xp->pr_ip6s > 0) {
3503 			error = SYSCTL_OUT(req, ip6,
3504 			    xp->pr_ip6s * sizeof(struct in6_addr));
3505 			if (error)
3506 				break;
3507 		}
3508 #endif
3509 	}
3510 	sx_sunlock(&allprison_lock);
3511 	free(xp, M_TEMP);
3512 #ifdef INET
3513 	free(ip4, M_TEMP);
3514 #endif
3515 #ifdef INET6
3516 	free(ip6, M_TEMP);
3517 #endif
3518 	return (error);
3519 }
3520 
3521 SYSCTL_OID(_security_jail, OID_AUTO, list,
3522     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3523     sysctl_jail_list, "S", "List of active jails");
3524 
3525 static int
3526 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3527 {
3528 	int error, injail;
3529 
3530 	injail = jailed(req->td->td_ucred);
3531 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3532 
3533 	return (error);
3534 }
3535 
3536 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3537     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3538     sysctl_jail_jailed, "I", "Process in jail?");
3539 
3540 static int
3541 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3542 {
3543 	int error, havevnet;
3544 #ifdef VIMAGE
3545 	struct ucred *cred = req->td->td_ucred;
3546 
3547 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3548 #else
3549 	havevnet = 0;
3550 #endif
3551 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3552 
3553 	return (error);
3554 }
3555 
3556 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3557     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3558     sysctl_jail_vnet, "I", "Jail owns VNET?");
3559 
3560 #if defined(INET) || defined(INET6)
3561 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3562     &jail_max_af_ips, 0,
3563     "Number of IP addresses a jail may have at most per address family (deprecated)");
3564 #endif
3565 
3566 /*
3567  * Default parameters for jail(2) compatibility.  For historical reasons,
3568  * the sysctl names have varying similarity to the parameter names.  Prisons
3569  * just see their own parameters, and can't change them.
3570  */
3571 static int
3572 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3573 {
3574 	struct prison *pr;
3575 	int allow, error, i;
3576 
3577 	pr = req->td->td_ucred->cr_prison;
3578 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3579 
3580 	/* Get the current flag value, and convert it to a boolean. */
3581 	i = (allow & arg2) ? 1 : 0;
3582 	if (arg1 != NULL)
3583 		i = !i;
3584 	error = sysctl_handle_int(oidp, &i, 0, req);
3585 	if (error || !req->newptr)
3586 		return (error);
3587 	i = i ? arg2 : 0;
3588 	if (arg1 != NULL)
3589 		i ^= arg2;
3590 	/*
3591 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3592 	 * for writing.
3593 	 */
3594 	mtx_lock(&prison0.pr_mtx);
3595 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3596 	mtx_unlock(&prison0.pr_mtx);
3597 	return (0);
3598 }
3599 
3600 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3601     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3602     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3603     "Processes in jail can set their hostnames (deprecated)");
3604 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3605     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3606     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3607     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3608 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3609     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3610     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3611     "Processes in jail can use System V IPC primitives (deprecated)");
3612 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3613     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3614     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3615     "Prison root can create raw sockets (deprecated)");
3616 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3617     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3618     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3619     "Processes in jail can alter system file flags (deprecated)");
3620 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3621     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3622     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3623     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3624 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
3625     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3626     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
3627     "Processes in jail can mount the devfs file system (deprecated)");
3628 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
3629     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3630     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
3631     "Processes in jail can mount the fdescfs file system (deprecated)");
3632 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
3633     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3634     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
3635     "Processes in jail can mount the nullfs file system (deprecated)");
3636 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
3637     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3638     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
3639     "Processes in jail can mount the procfs file system (deprecated)");
3640 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
3641     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3642     NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
3643     "Processes in jail can mount the linprocfs file system (deprecated)");
3644 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
3645     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3646     NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
3647     "Processes in jail can mount the linsysfs file system (deprecated)");
3648 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
3649     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3650     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
3651     "Processes in jail can mount the tmpfs file system (deprecated)");
3652 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
3653     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3654     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
3655     "Processes in jail can mount the zfs file system (deprecated)");
3656 
3657 static int
3658 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3659 {
3660 	struct prison *pr;
3661 	int level, error;
3662 
3663 	pr = req->td->td_ucred->cr_prison;
3664 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3665 	error = sysctl_handle_int(oidp, &level, 0, req);
3666 	if (error || !req->newptr)
3667 		return (error);
3668 	*(int *)arg1 = level;
3669 	return (0);
3670 }
3671 
3672 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3673     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3674     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3675     sysctl_jail_default_level, "I",
3676     "Processes in jail cannot see all mounted file systems (deprecated)");
3677 
3678 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3679     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3680     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3681     sysctl_jail_default_level, "I",
3682     "Ruleset for the devfs filesystem in jail (deprecated)");
3683 
3684 /*
3685  * Nodes to describe jail parameters.  Maximum length of string parameters
3686  * is returned in the string itself, and the other parameters exist merely
3687  * to make themselves and their types known.
3688  */
3689 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3690     "Jail parameters");
3691 
3692 int
3693 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3694 {
3695 	int i;
3696 	long l;
3697 	size_t s;
3698 	char numbuf[12];
3699 
3700 	switch (oidp->oid_kind & CTLTYPE)
3701 	{
3702 	case CTLTYPE_LONG:
3703 	case CTLTYPE_ULONG:
3704 		l = 0;
3705 #ifdef SCTL_MASK32
3706 		if (!(req->flags & SCTL_MASK32))
3707 #endif
3708 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3709 	case CTLTYPE_INT:
3710 	case CTLTYPE_UINT:
3711 		i = 0;
3712 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3713 	case CTLTYPE_STRING:
3714 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3715 		return
3716 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3717 	case CTLTYPE_STRUCT:
3718 		s = (size_t)arg2;
3719 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3720 	}
3721 	return (0);
3722 }
3723 
3724 /*
3725  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3726  * jail creation time but cannot be changed in an existing jail.
3727  */
3728 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3729 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3730 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3731 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3732 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3733     "I", "Jail secure level");
3734 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3735     "Jail value for kern.osreldate and uname -K");
3736 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3737     "Jail value for kern.osrelease and uname -r");
3738 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3739     "I", "Jail cannot see all mounted file systems");
3740 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3741     "I", "Ruleset for in-jail devfs mounts");
3742 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3743     "B", "Jail persistence");
3744 #ifdef VIMAGE
3745 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3746     "E,jailsys", "Virtual network stack");
3747 #endif
3748 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3749     "B", "Jail is in the process of shutting down");
3750 
3751 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3752 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3753     "I", "Current number of child jails");
3754 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3755     "I", "Maximum number of child jails");
3756 
3757 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3758 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3759     "Jail hostname");
3760 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3761     "Jail NIS domainname");
3762 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3763     "Jail host UUID");
3764 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3765     "LU", "Jail host ID");
3766 
3767 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3768 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3769 
3770 #ifdef INET
3771 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3772     "Jail IPv4 address virtualization");
3773 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3774     "S,in_addr,a", "Jail IPv4 addresses");
3775 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3776     "B", "Do (not) use IPv4 source address selection rather than the "
3777     "primary jail IPv4 address.");
3778 #endif
3779 #ifdef INET6
3780 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3781     "Jail IPv6 address virtualization");
3782 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3783     "S,in6_addr,a", "Jail IPv6 addresses");
3784 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3785     "B", "Do (not) use IPv6 source address selection rather than the "
3786     "primary jail IPv6 address.");
3787 #endif
3788 
3789 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3790 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3791     "B", "Jail may set hostname");
3792 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3793     "B", "Jail may use SYSV IPC");
3794 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3795     "B", "Jail may create raw sockets");
3796 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3797     "B", "Jail may alter system file flags");
3798 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3799     "B", "Jail may set file quotas");
3800 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3801     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3802 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3803     "B", "Jail may bind sockets to reserved ports");
3804 
3805 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3806 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3807     "B", "Jail may mount/unmount jail-friendly file systems in general");
3808 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
3809     "B", "Jail may mount the devfs file system");
3810 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
3811     "B", "Jail may mount the fdescfs file system");
3812 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
3813     "B", "Jail may mount the nullfs file system");
3814 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
3815     "B", "Jail may mount the procfs file system");
3816 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
3817     "B", "Jail may mount the linprocfs file system");
3818 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
3819     "B", "Jail may mount the linsysfs file system");
3820 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
3821     "B", "Jail may mount the tmpfs file system");
3822 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
3823     "B", "Jail may mount the zfs file system");
3824 
3825 #ifdef RACCT
3826 void
3827 prison_racct_foreach(void (*callback)(struct racct *racct,
3828     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3829     void *arg2, void *arg3)
3830 {
3831 	struct prison_racct *prr;
3832 
3833 	ASSERT_RACCT_ENABLED();
3834 
3835 	sx_slock(&allprison_lock);
3836 	if (pre != NULL)
3837 		(pre)();
3838 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3839 		(callback)(prr->prr_racct, arg2, arg3);
3840 	if (post != NULL)
3841 		(post)();
3842 	sx_sunlock(&allprison_lock);
3843 }
3844 
3845 static struct prison_racct *
3846 prison_racct_find_locked(const char *name)
3847 {
3848 	struct prison_racct *prr;
3849 
3850 	ASSERT_RACCT_ENABLED();
3851 	sx_assert(&allprison_lock, SA_XLOCKED);
3852 
3853 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3854 		return (NULL);
3855 
3856 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3857 		if (strcmp(name, prr->prr_name) != 0)
3858 			continue;
3859 
3860 		/* Found prison_racct with a matching name? */
3861 		prison_racct_hold(prr);
3862 		return (prr);
3863 	}
3864 
3865 	/* Add new prison_racct. */
3866 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3867 	racct_create(&prr->prr_racct);
3868 
3869 	strcpy(prr->prr_name, name);
3870 	refcount_init(&prr->prr_refcount, 1);
3871 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3872 
3873 	return (prr);
3874 }
3875 
3876 struct prison_racct *
3877 prison_racct_find(const char *name)
3878 {
3879 	struct prison_racct *prr;
3880 
3881 	ASSERT_RACCT_ENABLED();
3882 
3883 	sx_xlock(&allprison_lock);
3884 	prr = prison_racct_find_locked(name);
3885 	sx_xunlock(&allprison_lock);
3886 	return (prr);
3887 }
3888 
3889 void
3890 prison_racct_hold(struct prison_racct *prr)
3891 {
3892 
3893 	ASSERT_RACCT_ENABLED();
3894 
3895 	refcount_acquire(&prr->prr_refcount);
3896 }
3897 
3898 static void
3899 prison_racct_free_locked(struct prison_racct *prr)
3900 {
3901 
3902 	ASSERT_RACCT_ENABLED();
3903 	sx_assert(&allprison_lock, SA_XLOCKED);
3904 
3905 	if (refcount_release(&prr->prr_refcount)) {
3906 		racct_destroy(&prr->prr_racct);
3907 		LIST_REMOVE(prr, prr_next);
3908 		free(prr, M_PRISON_RACCT);
3909 	}
3910 }
3911 
3912 void
3913 prison_racct_free(struct prison_racct *prr)
3914 {
3915 	int old;
3916 
3917 	ASSERT_RACCT_ENABLED();
3918 	sx_assert(&allprison_lock, SA_UNLOCKED);
3919 
3920 	old = prr->prr_refcount;
3921 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
3922 		return;
3923 
3924 	sx_xlock(&allprison_lock);
3925 	prison_racct_free_locked(prr);
3926 	sx_xunlock(&allprison_lock);
3927 }
3928 
3929 static void
3930 prison_racct_attach(struct prison *pr)
3931 {
3932 	struct prison_racct *prr;
3933 
3934 	ASSERT_RACCT_ENABLED();
3935 	sx_assert(&allprison_lock, SA_XLOCKED);
3936 
3937 	prr = prison_racct_find_locked(pr->pr_name);
3938 	KASSERT(prr != NULL, ("cannot find prison_racct"));
3939 
3940 	pr->pr_prison_racct = prr;
3941 }
3942 
3943 /*
3944  * Handle jail renaming.  From the racct point of view, renaming means
3945  * moving from one prison_racct to another.
3946  */
3947 static void
3948 prison_racct_modify(struct prison *pr)
3949 {
3950 	struct proc *p;
3951 	struct ucred *cred;
3952 	struct prison_racct *oldprr;
3953 
3954 	ASSERT_RACCT_ENABLED();
3955 
3956 	sx_slock(&allproc_lock);
3957 	sx_xlock(&allprison_lock);
3958 
3959 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
3960 		sx_xunlock(&allprison_lock);
3961 		sx_sunlock(&allproc_lock);
3962 		return;
3963 	}
3964 
3965 	oldprr = pr->pr_prison_racct;
3966 	pr->pr_prison_racct = NULL;
3967 
3968 	prison_racct_attach(pr);
3969 
3970 	/*
3971 	 * Move resource utilisation records.
3972 	 */
3973 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
3974 
3975 	/*
3976 	 * Force rctl to reattach rules to processes.
3977 	 */
3978 	FOREACH_PROC_IN_SYSTEM(p) {
3979 		PROC_LOCK(p);
3980 		cred = crhold(p->p_ucred);
3981 		PROC_UNLOCK(p);
3982 		racct_proc_ucred_changed(p, cred, cred);
3983 		crfree(cred);
3984 	}
3985 
3986 	sx_sunlock(&allproc_lock);
3987 	prison_racct_free_locked(oldprr);
3988 	sx_xunlock(&allprison_lock);
3989 }
3990 
3991 static void
3992 prison_racct_detach(struct prison *pr)
3993 {
3994 
3995 	ASSERT_RACCT_ENABLED();
3996 	sx_assert(&allprison_lock, SA_UNLOCKED);
3997 
3998 	if (pr->pr_prison_racct == NULL)
3999 		return;
4000 	prison_racct_free(pr->pr_prison_racct);
4001 	pr->pr_prison_racct = NULL;
4002 }
4003 #endif /* RACCT */
4004 
4005 #ifdef DDB
4006 
4007 static void
4008 db_show_prison(struct prison *pr)
4009 {
4010 	int fi;
4011 #if defined(INET) || defined(INET6)
4012 	int ii;
4013 #endif
4014 	unsigned jsf;
4015 #ifdef INET
4016 	char ip4buf[INET_ADDRSTRLEN];
4017 #endif
4018 #ifdef INET6
4019 	char ip6buf[INET6_ADDRSTRLEN];
4020 #endif
4021 
4022 	db_printf("prison %p:\n", pr);
4023 	db_printf(" jid             = %d\n", pr->pr_id);
4024 	db_printf(" name            = %s\n", pr->pr_name);
4025 	db_printf(" parent          = %p\n", pr->pr_parent);
4026 	db_printf(" ref             = %d\n", pr->pr_ref);
4027 	db_printf(" uref            = %d\n", pr->pr_uref);
4028 	db_printf(" path            = %s\n", pr->pr_path);
4029 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4030 	    ? pr->pr_cpuset->cs_id : -1);
4031 #ifdef VIMAGE
4032 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4033 #endif
4034 	db_printf(" root            = %p\n", pr->pr_root);
4035 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4036 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4037 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4038 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4039 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4040 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4041 	db_printf(" flags           = 0x%x", pr->pr_flags);
4042 	for (fi = 0; fi < nitems(pr_flag_names); fi++)
4043 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4044 			db_printf(" %s", pr_flag_names[fi]);
4045 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4046 		jsf = pr->pr_flags &
4047 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4048 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4049 		    pr_flag_jailsys[fi].disable &&
4050 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4051 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4052 		    : "inherit");
4053 	}
4054 	db_printf(" allow           = 0x%x", pr->pr_allow);
4055 	for (fi = 0; fi < nitems(pr_allow_names); fi++)
4056 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4057 			db_printf(" %s", pr_allow_names[fi]);
4058 	db_printf("\n");
4059 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4060 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4061 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4062 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4063 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4064 #ifdef INET
4065 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4066 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4067 		db_printf(" %s %s\n",
4068 		    ii == 0 ? "ip4.addr        =" : "                 ",
4069 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4070 #endif
4071 #ifdef INET6
4072 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4073 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4074 		db_printf(" %s %s\n",
4075 		    ii == 0 ? "ip6.addr        =" : "                 ",
4076 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4077 #endif
4078 }
4079 
4080 DB_SHOW_COMMAND(prison, db_show_prison_command)
4081 {
4082 	struct prison *pr;
4083 
4084 	if (!have_addr) {
4085 		/*
4086 		 * Show all prisons in the list, and prison0 which is not
4087 		 * listed.
4088 		 */
4089 		db_show_prison(&prison0);
4090 		if (!db_pager_quit) {
4091 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4092 				db_show_prison(pr);
4093 				if (db_pager_quit)
4094 					break;
4095 			}
4096 		}
4097 		return;
4098 	}
4099 
4100 	if (addr == 0)
4101 		pr = &prison0;
4102 	else {
4103 		/* Look for a prison with the ID and with references. */
4104 		TAILQ_FOREACH(pr, &allprison, pr_list)
4105 			if (pr->pr_id == addr && pr->pr_ref > 0)
4106 				break;
4107 		if (pr == NULL)
4108 			/* Look again, without requiring a reference. */
4109 			TAILQ_FOREACH(pr, &allprison, pr_list)
4110 				if (pr->pr_id == addr)
4111 					break;
4112 		if (pr == NULL)
4113 			/* Assume address points to a valid prison. */
4114 			pr = (struct prison *)addr;
4115 	}
4116 	db_show_prison(pr);
4117 }
4118 
4119 #endif /* DDB */
4120