kern_jail.c revision 195870 
1/*-
2 * Copyright (c) 1999 Poul-Henning Kamp.
3 * Copyright (c) 2008 Bjoern A. Zeeb.
4 * Copyright (c) 2009 James Gritton.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: head/sys/kern/kern_jail.c 195870 2009-07-25 14:48:57Z jamie $");
31
32#include "opt_compat.h"
33#include "opt_ddb.h"
34#include "opt_inet.h"
35#include "opt_inet6.h"
36
37#include <sys/param.h>
38#include <sys/types.h>
39#include <sys/kernel.h>
40#include <sys/systm.h>
41#include <sys/errno.h>
42#include <sys/sysproto.h>
43#include <sys/malloc.h>
44#include <sys/osd.h>
45#include <sys/priv.h>
46#include <sys/proc.h>
47#include <sys/taskqueue.h>
48#include <sys/fcntl.h>
49#include <sys/jail.h>
50#include <sys/lock.h>
51#include <sys/mutex.h>
52#include <sys/sx.h>
53#include <sys/sysent.h>
54#include <sys/namei.h>
55#include <sys/mount.h>
56#include <sys/queue.h>
57#include <sys/socket.h>
58#include <sys/syscallsubr.h>
59#include <sys/sysctl.h>
60#include <sys/vnode.h>
61#include <sys/vimage.h>
62#include <net/if.h>
63#include <netinet/in.h>
64#ifdef DDB
65#include <ddb/ddb.h>
66#ifdef INET6
67#include <netinet6/in6_var.h>
68#endif /* INET6 */
69#endif /* DDB */
70
71#include <security/mac/mac_framework.h>
72
73MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
74
75/* prison0 describes what is "real" about the system. */
76struct prison prison0 = {
77	.pr_id		= 0,
78	.pr_name	= "0",
79	.pr_ref		= 1,
80	.pr_uref	= 1,
81	.pr_path	= "/",
82	.pr_securelevel	= -1,
83	.pr_childmax	= JAIL_MAX,
84	.pr_hostuuid	= "00000000-0000-0000-0000-000000000000",
85	.pr_children	= LIST_HEAD_INITIALIZER(&prison0.pr_children),
86	.pr_flags	= PR_HOST,
87	.pr_allow	= PR_ALLOW_ALL,
88};
89MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
90
91/* allprison and lastprid are protected by allprison_lock. */
92struct	sx allprison_lock;
93SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
94struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
95int	lastprid = 0;
96
97static int do_jail_attach(struct thread *td, struct prison *pr);
98static void prison_complete(void *context, int pending);
99static void prison_deref(struct prison *pr, int flags);
100static char *prison_path(struct prison *pr1, struct prison *pr2);
101static void prison_remove_one(struct prison *pr);
102#ifdef INET
103static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
104static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
105#endif
106#ifdef INET6
107static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
108static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
109#endif
110
111/* Flags for prison_deref */
112#define	PD_DEREF	0x01
113#define	PD_DEUREF	0x02
114#define	PD_LOCKED	0x04
115#define	PD_LIST_SLOCKED	0x08
116#define	PD_LIST_XLOCKED	0x10
117
118/*
119 * Parameter names corresponding to PR_* flag values
120 */
121static char *pr_flag_names[] = {
122	[0] = "persist",
123};
124
125static char *pr_flag_nonames[] = {
126	[0] = "nopersist",
127};
128
129struct jailsys_flags {
130	const char	*name;
131	unsigned	 disable;
132	unsigned	 new;
133} pr_flag_jailsys[] = {
134	{ "host", 0, PR_HOST },
135#ifdef VIMAGE
136	{ "vnet", 0, PR_VNET },
137#endif
138#ifdef INET
139	{ "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
140#endif
141#ifdef INET6
142	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
143#endif
144};
145
146static char *pr_allow_names[] = {
147	"allow.set_hostname",
148	"allow.sysvipc",
149	"allow.raw_sockets",
150	"allow.chflags",
151	"allow.mount",
152	"allow.quotas",
153	"allow.socket_af",
154};
155
156static char *pr_allow_nonames[] = {
157	"allow.noset_hostname",
158	"allow.nosysvipc",
159	"allow.noraw_sockets",
160	"allow.nochflags",
161	"allow.nomount",
162	"allow.noquotas",
163	"allow.nosocket_af",
164};
165
166#define	JAIL_DEFAULT_ALLOW	PR_ALLOW_SET_HOSTNAME
167static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
168static int jail_default_enforce_statfs = 2;
169#if defined(INET) || defined(INET6)
170static unsigned jail_max_af_ips = 255;
171#endif
172
173#ifdef INET
174static int
175qcmp_v4(const void *ip1, const void *ip2)
176{
177	in_addr_t iaa, iab;
178
179	/*
180	 * We need to compare in HBO here to get the list sorted as expected
181	 * by the result of the code.  Sorting NBO addresses gives you
182	 * interesting results.  If you do not understand, do not try.
183	 */
184	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
185	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
186
187	/*
188	 * Do not simply return the difference of the two numbers, the int is
189	 * not wide enough.
190	 */
191	if (iaa > iab)
192		return (1);
193	else if (iaa < iab)
194		return (-1);
195	else
196		return (0);
197}
198#endif
199
200#ifdef INET6
201static int
202qcmp_v6(const void *ip1, const void *ip2)
203{
204	const struct in6_addr *ia6a, *ia6b;
205	int i, rc;
206
207	ia6a = (const struct in6_addr *)ip1;
208	ia6b = (const struct in6_addr *)ip2;
209
210	rc = 0;
211	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
212		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
213			rc = 1;
214		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
215			rc = -1;
216	}
217	return (rc);
218}
219#endif
220
221/*
222 * struct jail_args {
223 *	struct jail *jail;
224 * };
225 */
226int
227jail(struct thread *td, struct jail_args *uap)
228{
229	uint32_t version;
230	int error;
231	struct jail j;
232
233	error = copyin(uap->jail, &version, sizeof(uint32_t));
234	if (error)
235		return (error);
236
237	switch (version) {
238	case 0:
239	{
240		struct jail_v0 j0;
241
242		/* FreeBSD single IPv4 jails. */
243		bzero(&j, sizeof(struct jail));
244		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
245		if (error)
246			return (error);
247		j.version = j0.version;
248		j.path = j0.path;
249		j.hostname = j0.hostname;
250		j.ip4s = j0.ip_number;
251		break;
252	}
253
254	case 1:
255		/*
256		 * Version 1 was used by multi-IPv4 jail implementations
257		 * that never made it into the official kernel.
258		 */
259		return (EINVAL);
260
261	case 2:	/* JAIL_API_VERSION */
262		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
263		error = copyin(uap->jail, &j, sizeof(struct jail));
264		if (error)
265			return (error);
266		break;
267
268	default:
269		/* Sci-Fi jails are not supported, sorry. */
270		return (EINVAL);
271	}
272	return (kern_jail(td, &j));
273}
274
275int
276kern_jail(struct thread *td, struct jail *j)
277{
278	struct iovec optiov[2 * (4
279			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
280#ifdef INET
281			    + 1
282#endif
283#ifdef INET6
284			    + 1
285#endif
286			    )];
287	struct uio opt;
288	char *u_path, *u_hostname, *u_name;
289#ifdef INET
290	uint32_t ip4s;
291	struct in_addr *u_ip4;
292#endif
293#ifdef INET6
294	struct in6_addr *u_ip6;
295#endif
296	size_t tmplen;
297	int error, enforce_statfs, fi;
298
299	bzero(&optiov, sizeof(optiov));
300	opt.uio_iov = optiov;
301	opt.uio_iovcnt = 0;
302	opt.uio_offset = -1;
303	opt.uio_resid = -1;
304	opt.uio_segflg = UIO_SYSSPACE;
305	opt.uio_rw = UIO_READ;
306	opt.uio_td = td;
307
308	/* Set permissions for top-level jails from sysctls. */
309	if (!jailed(td->td_ucred)) {
310		for (fi = 0; fi < sizeof(pr_allow_names) /
311		     sizeof(pr_allow_names[0]); fi++) {
312			optiov[opt.uio_iovcnt].iov_base =
313			    (jail_default_allow & (1 << fi))
314			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
315			optiov[opt.uio_iovcnt].iov_len =
316			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
317			opt.uio_iovcnt += 2;
318		}
319		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
320		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
321		opt.uio_iovcnt++;
322		enforce_statfs = jail_default_enforce_statfs;
323		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
324		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
325		opt.uio_iovcnt++;
326	}
327
328	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
329#ifdef INET
330	ip4s = (j->version == 0) ? 1 : j->ip4s;
331	if (ip4s > jail_max_af_ips)
332		return (EINVAL);
333	tmplen += ip4s * sizeof(struct in_addr);
334#else
335	if (j->ip4s > 0)
336		return (EINVAL);
337#endif
338#ifdef INET6
339	if (j->ip6s > jail_max_af_ips)
340		return (EINVAL);
341	tmplen += j->ip6s * sizeof(struct in6_addr);
342#else
343	if (j->ip6s > 0)
344		return (EINVAL);
345#endif
346	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
347	u_hostname = u_path + MAXPATHLEN;
348	u_name = u_hostname + MAXHOSTNAMELEN;
349#ifdef INET
350	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
351#endif
352#ifdef INET6
353#ifdef INET
354	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
355#else
356	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
357#endif
358#endif
359	optiov[opt.uio_iovcnt].iov_base = "path";
360	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
361	opt.uio_iovcnt++;
362	optiov[opt.uio_iovcnt].iov_base = u_path;
363	error = copyinstr(j->path, u_path, MAXPATHLEN,
364	    &optiov[opt.uio_iovcnt].iov_len);
365	if (error) {
366		free(u_path, M_TEMP);
367		return (error);
368	}
369	opt.uio_iovcnt++;
370	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
371	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
372	opt.uio_iovcnt++;
373	optiov[opt.uio_iovcnt].iov_base = u_hostname;
374	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
375	    &optiov[opt.uio_iovcnt].iov_len);
376	if (error) {
377		free(u_path, M_TEMP);
378		return (error);
379	}
380	opt.uio_iovcnt++;
381	if (j->jailname != NULL) {
382		optiov[opt.uio_iovcnt].iov_base = "name";
383		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
384		opt.uio_iovcnt++;
385		optiov[opt.uio_iovcnt].iov_base = u_name;
386		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
387		    &optiov[opt.uio_iovcnt].iov_len);
388		if (error) {
389			free(u_path, M_TEMP);
390			return (error);
391		}
392		opt.uio_iovcnt++;
393	}
394#ifdef INET
395	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
396	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
397	opt.uio_iovcnt++;
398	optiov[opt.uio_iovcnt].iov_base = u_ip4;
399	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
400	if (j->version == 0)
401		u_ip4->s_addr = j->ip4s;
402	else {
403		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
404		if (error) {
405			free(u_path, M_TEMP);
406			return (error);
407		}
408	}
409	opt.uio_iovcnt++;
410#endif
411#ifdef INET6
412	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
413	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
414	opt.uio_iovcnt++;
415	optiov[opt.uio_iovcnt].iov_base = u_ip6;
416	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
417	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
418	if (error) {
419		free(u_path, M_TEMP);
420		return (error);
421	}
422	opt.uio_iovcnt++;
423#endif
424	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
425	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
426	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
427	free(u_path, M_TEMP);
428	return (error);
429}
430
431
432/*
433 * struct jail_set_args {
434 *	struct iovec *iovp;
435 *	unsigned int iovcnt;
436 *	int flags;
437 * };
438 */
439int
440jail_set(struct thread *td, struct jail_set_args *uap)
441{
442	struct uio *auio;
443	int error;
444
445	/* Check that we have an even number of iovecs. */
446	if (uap->iovcnt & 1)
447		return (EINVAL);
448
449	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
450	if (error)
451		return (error);
452	error = kern_jail_set(td, auio, uap->flags);
453	free(auio, M_IOV);
454	return (error);
455}
456
457int
458kern_jail_set(struct thread *td, struct uio *optuio, int flags)
459{
460	struct nameidata nd;
461#ifdef INET
462	struct in_addr *ip4;
463#endif
464#ifdef INET6
465	struct in6_addr *ip6;
466#endif
467	struct vfsopt *opt;
468	struct vfsoptlist *opts;
469	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
470	struct vnode *root;
471	char *domain, *errmsg, *host, *name, *p, *path, *uuid;
472#if defined(INET) || defined(INET6)
473	void *op;
474#endif
475	unsigned long hid;
476	size_t namelen, onamelen;
477	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
478	int gotchildmax, gotenforce, gothid, gotslevel;
479	int fi, jid, jsys, len, level;
480	int childmax, slevel, vfslocked;
481#if defined(INET) || defined(INET6)
482	int ii, ij;
483#endif
484#ifdef INET
485	int ip4s, ip4a, redo_ip4;
486#endif
487#ifdef INET6
488	int ip6s, ip6a, redo_ip6;
489#endif
490	unsigned pr_flags, ch_flags;
491	unsigned pr_allow, ch_allow, tallow;
492	char numbuf[12];
493
494	error = priv_check(td, PRIV_JAIL_SET);
495	if (!error && (flags & JAIL_ATTACH))
496		error = priv_check(td, PRIV_JAIL_ATTACH);
497	if (error)
498		return (error);
499	mypr = ppr = td->td_ucred->cr_prison;
500	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
501		return (EPERM);
502	if (flags & ~JAIL_SET_MASK)
503		return (EINVAL);
504
505	/*
506	 * Check all the parameters before committing to anything.  Not all
507	 * errors can be caught early, but we may as well try.  Also, this
508	 * takes care of some expensive stuff (path lookup) before getting
509	 * the allprison lock.
510	 *
511	 * XXX Jails are not filesystems, and jail parameters are not mount
512	 *     options.  But it makes more sense to re-use the vfsopt code
513	 *     than duplicate it under a different name.
514	 */
515	error = vfs_buildopts(optuio, &opts);
516	if (error)
517		return (error);
518#ifdef INET
519	ip4a = 0;
520	ip4 = NULL;
521#endif
522#ifdef INET6
523	ip6a = 0;
524	ip6 = NULL;
525#endif
526
527#if defined(INET) || defined(INET6)
528 again:
529#endif
530	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
531	if (error == ENOENT)
532		jid = 0;
533	else if (error != 0)
534		goto done_free;
535
536	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
537	if (error == ENOENT)
538		gotslevel = 0;
539	else if (error != 0)
540		goto done_free;
541	else
542		gotslevel = 1;
543
544	error =
545	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
546	if (error == ENOENT)
547		gotchildmax = 0;
548	else if (error != 0)
549		goto done_free;
550	else
551		gotchildmax = 1;
552
553	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
554	gotenforce = (error == 0);
555	if (gotenforce) {
556		if (enforce < 0 || enforce > 2)
557			return (EINVAL);
558	} else if (error != ENOENT)
559		goto done_free;
560
561	pr_flags = ch_flags = 0;
562	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
563	    fi++) {
564		if (pr_flag_names[fi] == NULL)
565			continue;
566		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
567		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
568	}
569	ch_flags |= pr_flags;
570	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
571	    fi++) {
572		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
573		    sizeof(jsys));
574		if (error == ENOENT)
575			continue;
576		if (error != 0)
577			goto done_free;
578		switch (jsys) {
579		case JAIL_SYS_DISABLE:
580			if (!pr_flag_jailsys[fi].disable) {
581				error = EINVAL;
582				goto done_free;
583			}
584			pr_flags |= pr_flag_jailsys[fi].disable;
585			break;
586		case JAIL_SYS_NEW:
587			pr_flags |= pr_flag_jailsys[fi].new;
588			break;
589		case JAIL_SYS_INHERIT:
590			break;
591		default:
592			error = EINVAL;
593			goto done_free;
594		}
595		ch_flags |=
596		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
597	}
598	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
599	    && !(pr_flags & PR_PERSIST)) {
600		error = EINVAL;
601		vfs_opterror(opts, "new jail must persist or attach");
602		goto done_errmsg;
603	}
604#ifdef VIMAGE
605	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
606		error = EINVAL;
607		vfs_opterror(opts, "vnet cannot be changed after creation");
608		goto done_errmsg;
609	}
610#endif
611
612	pr_allow = ch_allow = 0;
613	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
614	    fi++) {
615		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
616		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
617	}
618	ch_allow |= pr_allow;
619
620	error = vfs_getopt(opts, "name", (void **)&name, &len);
621	if (error == ENOENT)
622		name = NULL;
623	else if (error != 0)
624		goto done_free;
625	else {
626		if (len == 0 || name[len - 1] != '\0') {
627			error = EINVAL;
628			goto done_free;
629		}
630		if (len > MAXHOSTNAMELEN) {
631			error = ENAMETOOLONG;
632			goto done_free;
633		}
634	}
635
636	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
637	if (error == ENOENT)
638		host = NULL;
639	else if (error != 0)
640		goto done_free;
641	else {
642		ch_flags |= PR_HOST;
643		pr_flags |= PR_HOST;
644		if (len == 0 || host[len - 1] != '\0') {
645			error = EINVAL;
646			goto done_free;
647		}
648		if (len > MAXHOSTNAMELEN) {
649			error = ENAMETOOLONG;
650			goto done_free;
651		}
652	}
653
654	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
655	if (error == ENOENT)
656		domain = NULL;
657	else if (error != 0)
658		goto done_free;
659	else {
660		ch_flags |= PR_HOST;
661		pr_flags |= PR_HOST;
662		if (len == 0 || domain[len - 1] != '\0') {
663			error = EINVAL;
664			goto done_free;
665		}
666		if (len > MAXHOSTNAMELEN) {
667			error = ENAMETOOLONG;
668			goto done_free;
669		}
670	}
671
672	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
673	if (error == ENOENT)
674		uuid = NULL;
675	else if (error != 0)
676		goto done_free;
677	else {
678		ch_flags |= PR_HOST;
679		pr_flags |= PR_HOST;
680		if (len == 0 || uuid[len - 1] != '\0') {
681			error = EINVAL;
682			goto done_free;
683		}
684		if (len > HOSTUUIDLEN) {
685			error = ENAMETOOLONG;
686			goto done_free;
687		}
688	}
689
690#ifdef COMPAT_IA32
691	if (td->td_proc->p_sysent->sv_flags & SV_IA32) {
692		uint32_t hid32;
693
694		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
695		hid = hid32;
696	} else
697#endif
698		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
699	if (error == ENOENT)
700		gothid = 0;
701	else if (error != 0)
702		goto done_free;
703	else {
704		gothid = 1;
705		ch_flags |= PR_HOST;
706		pr_flags |= PR_HOST;
707	}
708
709	/* This might be the second time around for this option. */
710#ifdef INET
711	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
712	if (error == ENOENT)
713		ip4s = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
714	else if (error != 0)
715		goto done_free;
716	else if (ip4s & (sizeof(*ip4) - 1)) {
717		error = EINVAL;
718		goto done_free;
719	} else {
720		ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
721		if (ip4s == 0)
722			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
723		else {
724			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
725			ip4s /= sizeof(*ip4);
726			if (ip4s > jail_max_af_ips) {
727				error = EINVAL;
728				vfs_opterror(opts, "too many IPv4 addresses");
729				goto done_errmsg;
730			}
731			if (ip4a < ip4s) {
732				ip4a = ip4s;
733				free(ip4, M_PRISON);
734				ip4 = NULL;
735			}
736			if (ip4 == NULL)
737				ip4 = malloc(ip4a * sizeof(*ip4), M_PRISON,
738				    M_WAITOK);
739			bcopy(op, ip4, ip4s * sizeof(*ip4));
740			/*
741			 * IP addresses are all sorted but ip[0] to preserve
742			 * the primary IP address as given from userland.
743			 * This special IP is used for unbound outgoing
744			 * connections as well for "loopback" traffic.
745			 */
746			if (ip4s > 1)
747				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
748			/*
749			 * Check for duplicate addresses and do some simple
750			 * zero and broadcast checks. If users give other bogus
751			 * addresses it is their problem.
752			 *
753			 * We do not have to care about byte order for these
754			 * checks so we will do them in NBO.
755			 */
756			for (ii = 0; ii < ip4s; ii++) {
757				if (ip4[ii].s_addr == INADDR_ANY ||
758				    ip4[ii].s_addr == INADDR_BROADCAST) {
759					error = EINVAL;
760					goto done_free;
761				}
762				if ((ii+1) < ip4s &&
763				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
764				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
765					error = EINVAL;
766					goto done_free;
767				}
768			}
769		}
770	}
771#endif
772
773#ifdef INET6
774	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
775	if (error == ENOENT)
776		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
777	else if (error != 0)
778		goto done_free;
779	else if (ip6s & (sizeof(*ip6) - 1)) {
780		error = EINVAL;
781		goto done_free;
782	} else {
783		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
784		if (ip6s == 0)
785			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
786		else {
787			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
788			ip6s /= sizeof(*ip6);
789			if (ip6s > jail_max_af_ips) {
790				error = EINVAL;
791				vfs_opterror(opts, "too many IPv6 addresses");
792				goto done_errmsg;
793			}
794			if (ip6a < ip6s) {
795				ip6a = ip6s;
796				free(ip6, M_PRISON);
797				ip6 = NULL;
798			}
799			if (ip6 == NULL)
800				ip6 = malloc(ip6a * sizeof(*ip6), M_PRISON,
801				    M_WAITOK);
802			bcopy(op, ip6, ip6s * sizeof(*ip6));
803			if (ip6s > 1)
804				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
805			for (ii = 0; ii < ip6s; ii++) {
806				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
807					error = EINVAL;
808					goto done_free;
809				}
810				if ((ii+1) < ip6s &&
811				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
812				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
813				{
814					error = EINVAL;
815					goto done_free;
816				}
817			}
818		}
819	}
820#endif
821
822	root = NULL;
823	error = vfs_getopt(opts, "path", (void **)&path, &len);
824	if (error == ENOENT)
825		path = NULL;
826	else if (error != 0)
827		goto done_free;
828	else {
829		if (flags & JAIL_UPDATE) {
830			error = EINVAL;
831			vfs_opterror(opts,
832			    "path cannot be changed after creation");
833			goto done_errmsg;
834		}
835		if (len == 0 || path[len - 1] != '\0') {
836			error = EINVAL;
837			goto done_free;
838		}
839		if (len < 2 || (len == 2 && path[0] == '/'))
840			path = NULL;
841		else {
842			/* Leave room for a real-root full pathname. */
843			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
844			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
845				error = ENAMETOOLONG;
846				goto done_free;
847			}
848			NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW, UIO_SYSSPACE,
849			    path, td);
850			error = namei(&nd);
851			if (error)
852				goto done_free;
853			vfslocked = NDHASGIANT(&nd);
854			root = nd.ni_vp;
855			NDFREE(&nd, NDF_ONLY_PNBUF);
856			if (root->v_type != VDIR) {
857				error = ENOTDIR;
858				vrele(root);
859				VFS_UNLOCK_GIANT(vfslocked);
860				goto done_free;
861			}
862			VFS_UNLOCK_GIANT(vfslocked);
863		}
864	}
865
866	/*
867	 * Grab the allprison lock before letting modules check their
868	 * parameters.  Once we have it, do not let go so we'll have a
869	 * consistent view of the OSD list.
870	 */
871	sx_xlock(&allprison_lock);
872	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
873	if (error)
874		goto done_unlock_list;
875
876	/* By now, all parameters should have been noted. */
877	TAILQ_FOREACH(opt, opts, link) {
878		if (!opt->seen && strcmp(opt->name, "errmsg")) {
879			error = EINVAL;
880			vfs_opterror(opts, "unknown parameter: %s", opt->name);
881			goto done_unlock_list;
882		}
883	}
884
885	/*
886	 * See if we are creating a new record or updating an existing one.
887	 * This abuses the file error codes ENOENT and EEXIST.
888	 */
889	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
890	if (!cuflags) {
891		error = EINVAL;
892		vfs_opterror(opts, "no valid operation (create or update)");
893		goto done_unlock_list;
894	}
895	pr = NULL;
896	if (jid != 0) {
897		/*
898		 * See if a requested jid already exists.  There is an
899		 * information leak here if the jid exists but is not within
900		 * the caller's jail hierarchy.  Jail creators will get EEXIST
901		 * even though they cannot see the jail, and CREATE | UPDATE
902		 * will return ENOENT which is not normally a valid error.
903		 */
904		if (jid < 0) {
905			error = EINVAL;
906			vfs_opterror(opts, "negative jid");
907			goto done_unlock_list;
908		}
909		pr = prison_find(jid);
910		if (pr != NULL) {
911			ppr = pr->pr_parent;
912			/* Create: jid must not exist. */
913			if (cuflags == JAIL_CREATE) {
914				mtx_unlock(&pr->pr_mtx);
915				error = EEXIST;
916				vfs_opterror(opts, "jail %d already exists",
917				    jid);
918				goto done_unlock_list;
919			}
920			if (!prison_ischild(mypr, pr)) {
921				mtx_unlock(&pr->pr_mtx);
922				pr = NULL;
923			} else if (pr->pr_uref == 0) {
924				if (!(flags & JAIL_DYING)) {
925					mtx_unlock(&pr->pr_mtx);
926					error = ENOENT;
927					vfs_opterror(opts, "jail %d is dying",
928					    jid);
929					goto done_unlock_list;
930				} else if ((flags & JAIL_ATTACH) ||
931				    (pr_flags & PR_PERSIST)) {
932					/*
933					 * A dying jail might be resurrected
934					 * (via attach or persist), but first
935					 * it must determine if another jail
936					 * has claimed its name.  Accomplish
937					 * this by implicitly re-setting the
938					 * name.
939					 */
940					if (name == NULL)
941						name = prison_name(mypr, pr);
942				}
943			}
944		}
945		if (pr == NULL) {
946			/* Update: jid must exist. */
947			if (cuflags == JAIL_UPDATE) {
948				error = ENOENT;
949				vfs_opterror(opts, "jail %d not found", jid);
950				goto done_unlock_list;
951			}
952		}
953	}
954	/*
955	 * If the caller provided a name, look for a jail by that name.
956	 * This has different semantics for creates and updates keyed by jid
957	 * (where the name must not already exist in a different jail),
958	 * and updates keyed by the name itself (where the name must exist
959	 * because that is the jail being updated).
960	 */
961	if (name != NULL) {
962		p = strrchr(name, '.');
963		if (p != NULL) {
964			/*
965			 * This is a hierarchical name.  Split it into the
966			 * parent and child names, and make sure the parent
967			 * exists or matches an already found jail.
968			 */
969			*p = '\0';
970			if (pr != NULL) {
971				if (strncmp(name, ppr->pr_name, p - name) ||
972				    ppr->pr_name[p - name] != '\0') {
973					mtx_unlock(&pr->pr_mtx);
974					error = EINVAL;
975					vfs_opterror(opts,
976					    "cannot change jail's parent");
977					goto done_unlock_list;
978				}
979			} else {
980				ppr = prison_find_name(mypr, name);
981				if (ppr == NULL) {
982					error = ENOENT;
983					vfs_opterror(opts,
984					    "jail \"%s\" not found", name);
985					goto done_unlock_list;
986				}
987				mtx_unlock(&ppr->pr_mtx);
988			}
989			name = p + 1;
990		}
991		if (name[0] != '\0') {
992			namelen =
993			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
994 name_again:
995			deadpr = NULL;
996			FOREACH_PRISON_CHILD(ppr, tpr) {
997				if (tpr != pr && tpr->pr_ref > 0 &&
998				    !strcmp(tpr->pr_name + namelen, name)) {
999					if (pr == NULL &&
1000					    cuflags != JAIL_CREATE) {
1001						mtx_lock(&tpr->pr_mtx);
1002						if (tpr->pr_ref > 0) {
1003							/*
1004							 * Use this jail
1005							 * for updates.
1006							 */
1007							if (tpr->pr_uref > 0) {
1008								pr = tpr;
1009								break;
1010							}
1011							deadpr = tpr;
1012						}
1013						mtx_unlock(&tpr->pr_mtx);
1014					} else if (tpr->pr_uref > 0) {
1015						/*
1016						 * Create, or update(jid):
1017						 * name must not exist in an
1018						 * active sibling jail.
1019						 */
1020						error = EEXIST;
1021						if (pr != NULL)
1022							mtx_unlock(&pr->pr_mtx);
1023						vfs_opterror(opts,
1024						   "jail \"%s\" already exists",
1025						   name);
1026						goto done_unlock_list;
1027					}
1028				}
1029			}
1030			/* If no active jail is found, use a dying one. */
1031			if (deadpr != NULL && pr == NULL) {
1032				if (flags & JAIL_DYING) {
1033					mtx_lock(&deadpr->pr_mtx);
1034					if (deadpr->pr_ref == 0) {
1035						mtx_unlock(&deadpr->pr_mtx);
1036						goto name_again;
1037					}
1038					pr = deadpr;
1039				} else if (cuflags == JAIL_UPDATE) {
1040					error = ENOENT;
1041					vfs_opterror(opts,
1042					    "jail \"%s\" is dying", name);
1043					goto done_unlock_list;
1044				}
1045			}
1046			/* Update: name must exist if no jid. */
1047			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1048				error = ENOENT;
1049				vfs_opterror(opts, "jail \"%s\" not found",
1050				    name);
1051				goto done_unlock_list;
1052			}
1053		}
1054	}
1055	/* Update: must provide a jid or name. */
1056	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1057		error = ENOENT;
1058		vfs_opterror(opts, "update specified no jail");
1059		goto done_unlock_list;
1060	}
1061
1062	/* If there's no prison to update, create a new one and link it in. */
1063	if (pr == NULL) {
1064		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1065			if (tpr->pr_childcount >= tpr->pr_childmax) {
1066				error = EPERM;
1067				vfs_opterror(opts, "prison limit exceeded");
1068				goto done_unlock_list;
1069			}
1070		created = 1;
1071		mtx_lock(&ppr->pr_mtx);
1072		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1073			mtx_unlock(&ppr->pr_mtx);
1074			error = ENOENT;
1075			vfs_opterror(opts, "parent jail went away!");
1076			goto done_unlock_list;
1077		}
1078		ppr->pr_ref++;
1079		ppr->pr_uref++;
1080		mtx_unlock(&ppr->pr_mtx);
1081		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1082		if (jid == 0) {
1083			/* Find the next free jid. */
1084			jid = lastprid + 1;
1085 findnext:
1086			if (jid == JAIL_MAX)
1087				jid = 1;
1088			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1089				if (tpr->pr_id < jid)
1090					continue;
1091				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1092					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1093					break;
1094				}
1095				if (jid == lastprid) {
1096					error = EAGAIN;
1097					vfs_opterror(opts,
1098					    "no available jail IDs");
1099					free(pr, M_PRISON);
1100					prison_deref(ppr, PD_DEREF |
1101					    PD_DEUREF | PD_LIST_XLOCKED);
1102					goto done_releroot;
1103				}
1104				jid++;
1105				goto findnext;
1106			}
1107			lastprid = jid;
1108		} else {
1109			/*
1110			 * The jail already has a jid (that did not yet exist),
1111			 * so just find where to insert it.
1112			 */
1113			TAILQ_FOREACH(tpr, &allprison, pr_list)
1114				if (tpr->pr_id >= jid) {
1115					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1116					break;
1117				}
1118		}
1119		if (tpr == NULL)
1120			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1121		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1122		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1123			tpr->pr_childcount++;
1124
1125		pr->pr_parent = ppr;
1126		pr->pr_id = jid;
1127
1128		/* Set some default values, and inherit some from the parent. */
1129		if (name == NULL)
1130			name = "";
1131		if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1132			if (host == NULL)
1133				host = ppr->pr_hostname;
1134			if (domain == NULL)
1135				domain = ppr->pr_domainname;
1136			if (uuid == NULL)
1137				uuid = ppr->pr_hostuuid;
1138			if (!gothid)
1139				hid = ppr->pr_hostid;
1140		}
1141		if (path == NULL) {
1142			path = "/";
1143			root = mypr->pr_root;
1144			vref(root);
1145		}
1146#ifdef INET
1147		pr->pr_flags |= ppr->pr_flags & PR_IP4;
1148		pr->pr_ip4s = ppr->pr_ip4s;
1149		if (ppr->pr_ip4 != NULL) {
1150			pr->pr_ip4 = malloc(pr->pr_ip4s *
1151			    sizeof(struct in_addr), M_PRISON, M_WAITOK);
1152			bcopy(ppr->pr_ip4, pr->pr_ip4,
1153			    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1154		}
1155#endif
1156#ifdef INET6
1157		pr->pr_flags |= ppr->pr_flags & PR_IP6;
1158		pr->pr_ip6s = ppr->pr_ip6s;
1159		if (ppr->pr_ip6 != NULL) {
1160			pr->pr_ip6 = malloc(pr->pr_ip6s *
1161			    sizeof(struct in6_addr), M_PRISON, M_WAITOK);
1162			bcopy(ppr->pr_ip6, pr->pr_ip6,
1163			    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1164		}
1165#endif
1166		pr->pr_securelevel = ppr->pr_securelevel;
1167		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1168		pr->pr_enforce_statfs = ppr->pr_enforce_statfs;
1169
1170		LIST_INIT(&pr->pr_children);
1171		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1172
1173#ifdef VIMAGE
1174		/* Allocate a new vnet if specified. */
1175		pr->pr_vnet = (pr_flags & PR_VNET)
1176		    ? vnet_alloc() : ppr->pr_vnet;
1177#endif
1178		/*
1179		 * Allocate a dedicated cpuset for each jail.
1180		 * Unlike other initial settings, this may return an erorr.
1181		 */
1182		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1183		if (error) {
1184			prison_deref(pr, PD_LIST_XLOCKED);
1185			goto done_releroot;
1186		}
1187
1188		mtx_lock(&pr->pr_mtx);
1189		/*
1190		 * New prisons do not yet have a reference, because we do not
1191		 * want other to see the incomplete prison once the
1192		 * allprison_lock is downgraded.
1193		 */
1194	} else {
1195		created = 0;
1196		/*
1197		 * Grab a reference for existing prisons, to ensure they
1198		 * continue to exist for the duration of the call.
1199		 */
1200		pr->pr_ref++;
1201	}
1202
1203	/* Do final error checking before setting anything. */
1204	if (gotslevel) {
1205		if (slevel < ppr->pr_securelevel) {
1206			error = EPERM;
1207			goto done_deref_locked;
1208		}
1209	}
1210	if (gotchildmax) {
1211		if (childmax >= ppr->pr_childmax) {
1212			error = EPERM;
1213			goto done_deref_locked;
1214		}
1215	}
1216	if (gotenforce) {
1217		if (enforce < ppr->pr_enforce_statfs) {
1218			error = EPERM;
1219			goto done_deref_locked;
1220		}
1221	}
1222#ifdef INET
1223	if (ch_flags & PR_IP4_USER) {
1224		if (ppr->pr_flags & PR_IP4) {
1225			if (!(pr_flags & PR_IP4_USER)) {
1226				/*
1227				 * Silently ignore attempts to make the IP
1228				 * addresses unrestricted when the parent is
1229				 * restricted; in other words, interpret
1230				 * "unrestricted" as "as unrestricted as
1231				 * possible".
1232				 */
1233				ip4s = ppr->pr_ip4s;
1234				if (ip4s == 0) {
1235					free(ip4, M_PRISON);
1236					ip4 = NULL;
1237				} else if (ip4s <= ip4a) {
1238					/* Inherit the parent's address(es). */
1239					bcopy(ppr->pr_ip4, ip4,
1240					    ip4s * sizeof(*ip4));
1241				} else {
1242					/*
1243					 * There's no room for the parent's
1244					 * address list.  Allocate some more.
1245					 */
1246					ip4a = ip4s;
1247					free(ip4, M_PRISON);
1248					ip4 = malloc(ip4a * sizeof(*ip4),
1249					    M_PRISON, M_NOWAIT);
1250					if (ip4 != NULL)
1251						bcopy(ppr->pr_ip4, ip4,
1252						    ip4s * sizeof(*ip4));
1253					else {
1254						/* Allocation failed without
1255						 * sleeping.  Unlocking the
1256						 * prison now will invalidate
1257						 * some checks and prematurely
1258						 * show an unfinished new jail.
1259						 * So let go of everything and
1260						 * start over.
1261						 */
1262						prison_deref(pr, created
1263						    ? PD_LOCKED |
1264						      PD_LIST_XLOCKED
1265						    : PD_DEREF | PD_LOCKED |
1266						      PD_LIST_XLOCKED);
1267						if (root != NULL) {
1268							vfslocked =
1269							    VFS_LOCK_GIANT(
1270							    root->v_mount);
1271							vrele(root);
1272							VFS_UNLOCK_GIANT(
1273							    vfslocked);
1274						}
1275						ip4 = malloc(ip4a *
1276						    sizeof(*ip4), M_PRISON,
1277						    M_WAITOK);
1278						goto again;
1279					}
1280				}
1281			} else if (ip4s > 0) {
1282				/*
1283				 * Make sure the new set of IP addresses is a
1284				 * subset of the parent's list.  Don't worry
1285				 * about the parent being unlocked, as any
1286				 * setting is done with allprison_lock held.
1287				 */
1288				for (ij = 0; ij < ppr->pr_ip4s; ij++)
1289					if (ip4[0].s_addr ==
1290					    ppr->pr_ip4[ij].s_addr)
1291						break;
1292				if (ij == ppr->pr_ip4s) {
1293					error = EPERM;
1294					goto done_deref_locked;
1295				}
1296				if (ip4s > 1) {
1297					for (ii = ij = 1; ii < ip4s; ii++) {
1298						if (ip4[ii].s_addr ==
1299						    ppr->pr_ip4[0].s_addr)
1300							continue;
1301						for (; ij < ppr->pr_ip4s; ij++)
1302						    if (ip4[ii].s_addr ==
1303							ppr->pr_ip4[ij].s_addr)
1304							    break;
1305						if (ij == ppr->pr_ip4s)
1306							break;
1307					}
1308					if (ij == ppr->pr_ip4s) {
1309						error = EPERM;
1310						goto done_deref_locked;
1311					}
1312				}
1313			}
1314		}
1315		if (ip4s > 0) {
1316			/*
1317			 * Check for conflicting IP addresses.  We permit them
1318			 * if there is no more than one IP on each jail.  If
1319			 * there is a duplicate on a jail with more than one
1320			 * IP stop checking and return error.
1321			 */
1322			FOREACH_PRISON_DESCENDANT(&prison0, tpr, descend) {
1323				if (tpr == pr || tpr->pr_uref == 0) {
1324					descend = 0;
1325					continue;
1326				}
1327				if (!(tpr->pr_flags & PR_IP4_USER))
1328					continue;
1329				descend = 0;
1330				if (tpr->pr_ip4 == NULL ||
1331				    (ip4s == 1 && tpr->pr_ip4s == 1))
1332					continue;
1333				for (ii = 0; ii < ip4s; ii++) {
1334					if (_prison_check_ip4(tpr,
1335					    &ip4[ii]) == 0) {
1336						error = EADDRINUSE;
1337						vfs_opterror(opts,
1338						    "IPv4 addresses clash");
1339						goto done_deref_locked;
1340					}
1341				}
1342			}
1343		}
1344	}
1345#endif
1346#ifdef INET6
1347	if (ch_flags & PR_IP6_USER) {
1348		if (ppr->pr_flags & PR_IP6) {
1349			if (!(pr_flags & PR_IP6_USER)) {
1350				/*
1351				 * Silently ignore attempts to make the IP
1352				 * addresses unrestricted when the parent is
1353				 * restricted.
1354				 */
1355				ip6s = ppr->pr_ip6s;
1356				if (ip6s == 0) {
1357					free(ip6, M_PRISON);
1358					ip6 = NULL;
1359				} else if (ip6s <= ip6a) {
1360					/* Inherit the parent's address(es). */
1361					bcopy(ppr->pr_ip6, ip6,
1362					    ip6s * sizeof(*ip6));
1363				} else {
1364					/*
1365					 * There's no room for the parent's
1366					 * address list.
1367					 */
1368					ip6a = ip6s;
1369					free(ip6, M_PRISON);
1370					ip6 = malloc(ip6a * sizeof(*ip6),
1371					    M_PRISON, M_NOWAIT);
1372					if (ip6 != NULL)
1373						bcopy(ppr->pr_ip6, ip6,
1374						    ip6s * sizeof(*ip6));
1375					else {
1376						prison_deref(pr, created
1377						    ? PD_LOCKED |
1378						      PD_LIST_XLOCKED
1379						    : PD_DEREF | PD_LOCKED |
1380						      PD_LIST_XLOCKED);
1381						if (root != NULL) {
1382							vfslocked =
1383							    VFS_LOCK_GIANT(
1384							    root->v_mount);
1385							vrele(root);
1386							VFS_UNLOCK_GIANT(
1387							    vfslocked);
1388						}
1389						ip6 = malloc(ip6a *
1390						    sizeof(*ip6), M_PRISON,
1391						    M_WAITOK);
1392						goto again;
1393					}
1394				}
1395			} else if (ip6s > 0) {
1396				/*
1397				 * Make sure the new set of IP addresses is a
1398				 * subset of the parent's list.
1399				 */
1400				for (ij = 0; ij < ppr->pr_ip6s; ij++)
1401					if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1402					    &ppr->pr_ip6[ij]))
1403						break;
1404				if (ij == ppr->pr_ip6s) {
1405					error = EPERM;
1406					goto done_deref_locked;
1407				}
1408				if (ip6s > 1) {
1409					for (ii = ij = 1; ii < ip6s; ii++) {
1410						if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1411						    &ppr->pr_ip6[0]))
1412							continue;
1413						for (; ij < ppr->pr_ip6s; ij++)
1414							if (IN6_ARE_ADDR_EQUAL(
1415							    &ip6[ii],
1416							    &ppr->pr_ip6[ij]))
1417								break;
1418						if (ij == ppr->pr_ip6s)
1419							break;
1420					}
1421					if (ij == ppr->pr_ip6s) {
1422						error = EPERM;
1423						goto done_deref_locked;
1424					}
1425				}
1426			}
1427		}
1428		if (ip6s > 0) {
1429			/* Check for conflicting IP addresses. */
1430			FOREACH_PRISON_DESCENDANT(&prison0, tpr, descend) {
1431				if (tpr == pr || tpr->pr_uref == 0) {
1432					descend = 0;
1433					continue;
1434				}
1435				if (!(tpr->pr_flags & PR_IP6_USER))
1436					continue;
1437				descend = 0;
1438				if (tpr->pr_ip6 == NULL ||
1439				    (ip6s == 1 && tpr->pr_ip6s == 1))
1440					continue;
1441				for (ii = 0; ii < ip6s; ii++) {
1442					if (_prison_check_ip6(tpr,
1443					    &ip6[ii]) == 0) {
1444						error = EADDRINUSE;
1445						vfs_opterror(opts,
1446						    "IPv6 addresses clash");
1447						goto done_deref_locked;
1448					}
1449				}
1450			}
1451		}
1452	}
1453#endif
1454	onamelen = namelen = 0;
1455	if (name != NULL) {
1456		/* Give a default name of the jid. */
1457		if (name[0] == '\0')
1458			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1459		else if (strtoul(name, &p, 10) != jid && *p == '\0') {
1460			error = EINVAL;
1461			vfs_opterror(opts, "name cannot be numeric");
1462			goto done_deref_locked;
1463		}
1464		/*
1465		 * Make sure the name isn't too long for the prison or its
1466		 * children.
1467		 */
1468		onamelen = strlen(pr->pr_name);
1469		namelen = strlen(name);
1470		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1471			error = ENAMETOOLONG;
1472			goto done_deref_locked;
1473		}
1474		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1475			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1476			    sizeof(pr->pr_name)) {
1477				error = ENAMETOOLONG;
1478				goto done_deref_locked;
1479			}
1480		}
1481	}
1482	if (pr_allow & ~ppr->pr_allow) {
1483		error = EPERM;
1484		goto done_deref_locked;
1485	}
1486
1487	/* Set the parameters of the prison. */
1488#ifdef INET
1489	redo_ip4 = 0;
1490	if (ch_flags & PR_IP4_USER) {
1491		if (pr_flags & PR_IP4_USER) {
1492			/* Some restriction set. */
1493			pr->pr_flags |= PR_IP4;
1494			if (ip4s >= 0) {
1495				free(pr->pr_ip4, M_PRISON);
1496				pr->pr_ip4s = ip4s;
1497				pr->pr_ip4 = ip4;
1498				ip4 = NULL;
1499			}
1500		} else if (ppr->pr_flags & PR_IP4) {
1501			/* This restriction cleared, but keep inherited. */
1502			free(pr->pr_ip4, M_PRISON);
1503			pr->pr_ip4s = ip4s;
1504			pr->pr_ip4 = ip4;
1505			ip4 = NULL;
1506		} else {
1507			/* Restriction cleared, now unrestricted. */
1508			pr->pr_flags &= ~PR_IP4;
1509			free(pr->pr_ip4, M_PRISON);
1510			pr->pr_ip4s = 0;
1511		}
1512		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1513			if (prison_restrict_ip4(tpr, NULL)) {
1514				redo_ip4 = 1;
1515				descend = 0;
1516			}
1517		}
1518	}
1519#endif
1520#ifdef INET6
1521	redo_ip6 = 0;
1522	if (ch_flags & PR_IP6_USER) {
1523		if (pr_flags & PR_IP6_USER) {
1524			/* Some restriction set. */
1525			pr->pr_flags |= PR_IP6;
1526			if (ip6s >= 0) {
1527				free(pr->pr_ip6, M_PRISON);
1528				pr->pr_ip6s = ip6s;
1529				pr->pr_ip6 = ip6;
1530				ip6 = NULL;
1531			}
1532		} else if (ppr->pr_flags & PR_IP6) {
1533			/* This restriction cleared, but keep inherited. */
1534			free(pr->pr_ip6, M_PRISON);
1535			pr->pr_ip6s = ip6s;
1536			pr->pr_ip6 = ip6;
1537			ip6 = NULL;
1538		} else {
1539			/* Restriction cleared, now unrestricted. */
1540			pr->pr_flags &= ~PR_IP6;
1541			free(pr->pr_ip6, M_PRISON);
1542			pr->pr_ip6s = 0;
1543		}
1544		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1545			if (prison_restrict_ip6(tpr, NULL)) {
1546				redo_ip6 = 1;
1547				descend = 0;
1548			}
1549		}
1550	}
1551#endif
1552	if (gotslevel) {
1553		pr->pr_securelevel = slevel;
1554		/* Set all child jails to be at least this level. */
1555		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1556			if (tpr->pr_securelevel < slevel)
1557				tpr->pr_securelevel = slevel;
1558	}
1559	if (gotchildmax) {
1560		pr->pr_childmax = childmax;
1561		/* Set all child jails to under this limit. */
1562		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1563			if (tpr->pr_childmax > childmax - level)
1564				tpr->pr_childmax = childmax > level
1565				    ? childmax - level : 0;
1566	}
1567	if (gotenforce) {
1568		pr->pr_enforce_statfs = enforce;
1569		/* Pass this restriction on to the children. */
1570		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1571			if (tpr->pr_enforce_statfs < enforce)
1572				tpr->pr_enforce_statfs = enforce;
1573	}
1574	if (name != NULL) {
1575		if (ppr == &prison0)
1576			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1577		else
1578			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1579			    ppr->pr_name, name);
1580		/* Change this component of child names. */
1581		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1582			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1583			    strlen(tpr->pr_name + onamelen) + 1);
1584			bcopy(pr->pr_name, tpr->pr_name, namelen);
1585		}
1586	}
1587	if (path != NULL) {
1588		/* Try to keep a real-rooted full pathname. */
1589		if (path[0] == '/' && strcmp(mypr->pr_path, "/"))
1590			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1591			    mypr->pr_path, path);
1592		else
1593			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1594		pr->pr_root = root;
1595	}
1596	if (PR_HOST & ch_flags & ~pr_flags) {
1597		if (pr->pr_flags & PR_HOST) {
1598			/*
1599			 * Copy the parent's host info.  As with pr_ip4 above,
1600			 * the lack of a lock on the parent is not a problem;
1601			 * it is always set with allprison_lock at least
1602			 * shared, and is held exclusively here.
1603			 */
1604			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1605			    sizeof(pr->pr_hostname));
1606			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1607			    sizeof(pr->pr_domainname));
1608			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1609			    sizeof(pr->pr_hostuuid));
1610			pr->pr_hostid = pr->pr_parent->pr_hostid;
1611		}
1612	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1613		/* Set this prison, and any descendants without PR_HOST. */
1614		if (host != NULL)
1615			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1616		if (domain != NULL)
1617			strlcpy(pr->pr_domainname, domain,
1618			    sizeof(pr->pr_domainname));
1619		if (uuid != NULL)
1620			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1621		if (gothid)
1622			pr->pr_hostid = hid;
1623		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1624			if (tpr->pr_flags & PR_HOST)
1625				descend = 0;
1626			else {
1627				if (host != NULL)
1628					strlcpy(tpr->pr_hostname,
1629					    pr->pr_hostname,
1630					    sizeof(tpr->pr_hostname));
1631				if (domain != NULL)
1632					strlcpy(tpr->pr_domainname,
1633					    pr->pr_domainname,
1634					    sizeof(tpr->pr_domainname));
1635				if (uuid != NULL)
1636					strlcpy(tpr->pr_hostuuid,
1637					    pr->pr_hostuuid,
1638					    sizeof(tpr->pr_hostuuid));
1639				if (gothid)
1640					tpr->pr_hostid = hid;
1641			}
1642		}
1643	}
1644	if ((tallow = ch_allow & ~pr_allow)) {
1645		/* Clear allow bits in all children. */
1646		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1647			tpr->pr_allow &= ~tallow;
1648	}
1649	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1650	/*
1651	 * Persistent prisons get an extra reference, and prisons losing their
1652	 * persist flag lose that reference.  Only do this for existing prisons
1653	 * for now, so new ones will remain unseen until after the module
1654	 * handlers have completed.
1655	 */
1656	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1657		if (pr_flags & PR_PERSIST) {
1658			pr->pr_ref++;
1659			pr->pr_uref++;
1660		} else {
1661			pr->pr_ref--;
1662			pr->pr_uref--;
1663		}
1664	}
1665	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1666	mtx_unlock(&pr->pr_mtx);
1667
1668	/* Locks may have prevented a complete restriction of child IP
1669	 * addresses.  If so, allocate some more memory and try again.
1670	 */
1671#ifdef INET
1672	while (redo_ip4) {
1673		ip4s = pr->pr_ip4s;
1674		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1675		mtx_lock(&pr->pr_mtx);
1676		redo_ip4 = 0;
1677		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1678			if (prison_restrict_ip4(tpr, ip4)) {
1679				if (ip4 != NULL)
1680					ip4 = NULL;
1681				else
1682					redo_ip4 = 1;
1683			}
1684		}
1685		mtx_unlock(&pr->pr_mtx);
1686	}
1687#endif
1688#ifdef INET6
1689	while (redo_ip6) {
1690		ip6s = pr->pr_ip6s;
1691		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1692		mtx_lock(&pr->pr_mtx);
1693		redo_ip6 = 0;
1694		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1695			if (prison_restrict_ip6(tpr, ip6)) {
1696				if (ip6 != NULL)
1697					ip6 = NULL;
1698				else
1699					redo_ip6 = 1;
1700			}
1701		}
1702		mtx_unlock(&pr->pr_mtx);
1703	}
1704#endif
1705
1706	/* Let the modules do their work. */
1707	sx_downgrade(&allprison_lock);
1708	if (created) {
1709		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1710		if (error) {
1711			prison_deref(pr, PD_LIST_SLOCKED);
1712			goto done_errmsg;
1713		}
1714	}
1715	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1716	if (error) {
1717		prison_deref(pr, created
1718		    ? PD_LIST_SLOCKED
1719		    : PD_DEREF | PD_LIST_SLOCKED);
1720		goto done_errmsg;
1721	}
1722
1723	/* Attach this process to the prison if requested. */
1724	if (flags & JAIL_ATTACH) {
1725		mtx_lock(&pr->pr_mtx);
1726		error = do_jail_attach(td, pr);
1727		if (error) {
1728			vfs_opterror(opts, "attach failed");
1729			if (!created)
1730				prison_deref(pr, PD_DEREF);
1731			goto done_errmsg;
1732		}
1733	}
1734
1735	/*
1736	 * Now that it is all there, drop the temporary reference from existing
1737	 * prisons.  Or add a reference to newly created persistent prisons
1738	 * (which was not done earlier so that the prison would not be publicly
1739	 * visible).
1740	 */
1741	if (!created) {
1742		prison_deref(pr, (flags & JAIL_ATTACH)
1743		    ? PD_DEREF
1744		    : PD_DEREF | PD_LIST_SLOCKED);
1745	} else {
1746		if (pr_flags & PR_PERSIST) {
1747			mtx_lock(&pr->pr_mtx);
1748			pr->pr_ref++;
1749			pr->pr_uref++;
1750			mtx_unlock(&pr->pr_mtx);
1751		}
1752		if (!(flags & JAIL_ATTACH))
1753			sx_sunlock(&allprison_lock);
1754	}
1755	td->td_retval[0] = pr->pr_id;
1756	goto done_errmsg;
1757
1758 done_deref_locked:
1759	prison_deref(pr, created
1760	    ? PD_LOCKED | PD_LIST_XLOCKED
1761	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1762	goto done_releroot;
1763 done_unlock_list:
1764	sx_xunlock(&allprison_lock);
1765 done_releroot:
1766	if (root != NULL) {
1767		vfslocked = VFS_LOCK_GIANT(root->v_mount);
1768		vrele(root);
1769		VFS_UNLOCK_GIANT(vfslocked);
1770	}
1771 done_errmsg:
1772	if (error) {
1773		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1774		if (errmsg_len > 0) {
1775			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1776			if (errmsg_pos > 0) {
1777				if (optuio->uio_segflg == UIO_SYSSPACE)
1778					bcopy(errmsg,
1779					   optuio->uio_iov[errmsg_pos].iov_base,
1780					   errmsg_len);
1781				else
1782					copyout(errmsg,
1783					   optuio->uio_iov[errmsg_pos].iov_base,
1784					   errmsg_len);
1785			}
1786		}
1787	}
1788 done_free:
1789#ifdef INET
1790	free(ip4, M_PRISON);
1791#endif
1792#ifdef INET6
1793	free(ip6, M_PRISON);
1794#endif
1795	vfs_freeopts(opts);
1796	return (error);
1797}
1798
1799
1800/*
1801 * struct jail_get_args {
1802 *	struct iovec *iovp;
1803 *	unsigned int iovcnt;
1804 *	int flags;
1805 * };
1806 */
1807int
1808jail_get(struct thread *td, struct jail_get_args *uap)
1809{
1810	struct uio *auio;
1811	int error;
1812
1813	/* Check that we have an even number of iovecs. */
1814	if (uap->iovcnt & 1)
1815		return (EINVAL);
1816
1817	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1818	if (error)
1819		return (error);
1820	error = kern_jail_get(td, auio, uap->flags);
1821	if (error == 0)
1822		error = copyout(auio->uio_iov, uap->iovp,
1823		    uap->iovcnt * sizeof (struct iovec));
1824	free(auio, M_IOV);
1825	return (error);
1826}
1827
1828int
1829kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1830{
1831	struct prison *pr, *mypr;
1832	struct vfsopt *opt;
1833	struct vfsoptlist *opts;
1834	char *errmsg, *name;
1835	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1836
1837	if (flags & ~JAIL_GET_MASK)
1838		return (EINVAL);
1839
1840	/* Get the parameter list. */
1841	error = vfs_buildopts(optuio, &opts);
1842	if (error)
1843		return (error);
1844	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1845	mypr = td->td_ucred->cr_prison;
1846
1847	/*
1848	 * Find the prison specified by one of: lastjid, jid, name.
1849	 */
1850	sx_slock(&allprison_lock);
1851	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1852	if (error == 0) {
1853		TAILQ_FOREACH(pr, &allprison, pr_list) {
1854			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1855				mtx_lock(&pr->pr_mtx);
1856				if (pr->pr_ref > 0 &&
1857				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1858					break;
1859				mtx_unlock(&pr->pr_mtx);
1860			}
1861		}
1862		if (pr != NULL)
1863			goto found_prison;
1864		error = ENOENT;
1865		vfs_opterror(opts, "no jail after %d", jid);
1866		goto done_unlock_list;
1867	} else if (error != ENOENT)
1868		goto done_unlock_list;
1869
1870	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1871	if (error == 0) {
1872		if (jid != 0) {
1873			pr = prison_find_child(mypr, jid);
1874			if (pr != NULL) {
1875				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1876					mtx_unlock(&pr->pr_mtx);
1877					error = ENOENT;
1878					vfs_opterror(opts, "jail %d is dying",
1879					    jid);
1880					goto done_unlock_list;
1881				}
1882				goto found_prison;
1883			}
1884			error = ENOENT;
1885			vfs_opterror(opts, "jail %d not found", jid);
1886			goto done_unlock_list;
1887		}
1888	} else if (error != ENOENT)
1889		goto done_unlock_list;
1890
1891	error = vfs_getopt(opts, "name", (void **)&name, &len);
1892	if (error == 0) {
1893		if (len == 0 || name[len - 1] != '\0') {
1894			error = EINVAL;
1895			goto done_unlock_list;
1896		}
1897		pr = prison_find_name(mypr, name);
1898		if (pr != NULL) {
1899			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1900				mtx_unlock(&pr->pr_mtx);
1901				error = ENOENT;
1902				vfs_opterror(opts, "jail \"%s\" is dying",
1903				    name);
1904				goto done_unlock_list;
1905			}
1906			goto found_prison;
1907		}
1908		error = ENOENT;
1909		vfs_opterror(opts, "jail \"%s\" not found", name);
1910		goto done_unlock_list;
1911	} else if (error != ENOENT)
1912		goto done_unlock_list;
1913
1914	vfs_opterror(opts, "no jail specified");
1915	error = ENOENT;
1916	goto done_unlock_list;
1917
1918 found_prison:
1919	/* Get the parameters of the prison. */
1920	pr->pr_ref++;
1921	locked = PD_LOCKED;
1922	td->td_retval[0] = pr->pr_id;
1923	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
1924	if (error != 0 && error != ENOENT)
1925		goto done_deref;
1926	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
1927	error = vfs_setopt(opts, "parent", &i, sizeof(i));
1928	if (error != 0 && error != ENOENT)
1929		goto done_deref;
1930	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
1931	if (error != 0 && error != ENOENT)
1932		goto done_deref;
1933	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
1934	    sizeof(pr->pr_cpuset->cs_id));
1935	if (error != 0 && error != ENOENT)
1936		goto done_deref;
1937	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
1938	if (error != 0 && error != ENOENT)
1939		goto done_deref;
1940#ifdef INET
1941	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
1942	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1943	if (error != 0 && error != ENOENT)
1944		goto done_deref;
1945#endif
1946#ifdef INET6
1947	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
1948	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1949	if (error != 0 && error != ENOENT)
1950		goto done_deref;
1951#endif
1952	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
1953	    sizeof(pr->pr_securelevel));
1954	if (error != 0 && error != ENOENT)
1955		goto done_deref;
1956	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
1957	    sizeof(pr->pr_childcount));
1958	if (error != 0 && error != ENOENT)
1959		goto done_deref;
1960	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
1961	    sizeof(pr->pr_childmax));
1962	if (error != 0 && error != ENOENT)
1963		goto done_deref;
1964	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
1965	if (error != 0 && error != ENOENT)
1966		goto done_deref;
1967	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
1968	if (error != 0 && error != ENOENT)
1969		goto done_deref;
1970	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
1971	if (error != 0 && error != ENOENT)
1972		goto done_deref;
1973#ifdef COMPAT_IA32
1974	if (td->td_proc->p_sysent->sv_flags & SV_IA32) {
1975		uint32_t hid32 = pr->pr_hostid;
1976
1977		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
1978	} else
1979#endif
1980	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
1981	    sizeof(pr->pr_hostid));
1982	if (error != 0 && error != ENOENT)
1983		goto done_deref;
1984	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
1985	    sizeof(pr->pr_enforce_statfs));
1986	if (error != 0 && error != ENOENT)
1987		goto done_deref;
1988	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
1989	    fi++) {
1990		if (pr_flag_names[fi] == NULL)
1991			continue;
1992		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
1993		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
1994		if (error != 0 && error != ENOENT)
1995			goto done_deref;
1996		i = !i;
1997		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
1998		if (error != 0 && error != ENOENT)
1999			goto done_deref;
2000	}
2001	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2002	    fi++) {
2003		i = pr->pr_flags &
2004		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2005		i = pr_flag_jailsys[fi].disable &&
2006		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2007		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2008		    : JAIL_SYS_INHERIT;
2009		error =
2010		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2011		if (error != 0 && error != ENOENT)
2012			goto done_deref;
2013	}
2014	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2015	    fi++) {
2016		if (pr_allow_names[fi] == NULL)
2017			continue;
2018		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2019		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2020		if (error != 0 && error != ENOENT)
2021			goto done_deref;
2022		i = !i;
2023		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2024		if (error != 0 && error != ENOENT)
2025			goto done_deref;
2026	}
2027	i = (pr->pr_uref == 0);
2028	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2029	if (error != 0 && error != ENOENT)
2030		goto done_deref;
2031	i = !i;
2032	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2033	if (error != 0 && error != ENOENT)
2034		goto done_deref;
2035
2036	/* Get the module parameters. */
2037	mtx_unlock(&pr->pr_mtx);
2038	locked = 0;
2039	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2040	if (error)
2041		goto done_deref;
2042	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2043
2044	/* By now, all parameters should have been noted. */
2045	TAILQ_FOREACH(opt, opts, link) {
2046		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2047			error = EINVAL;
2048			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2049			goto done_errmsg;
2050		}
2051	}
2052
2053	/* Write the fetched parameters back to userspace. */
2054	error = 0;
2055	TAILQ_FOREACH(opt, opts, link) {
2056		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2057			pos = 2 * opt->pos + 1;
2058			optuio->uio_iov[pos].iov_len = opt->len;
2059			if (opt->value != NULL) {
2060				if (optuio->uio_segflg == UIO_SYSSPACE) {
2061					bcopy(opt->value,
2062					    optuio->uio_iov[pos].iov_base,
2063					    opt->len);
2064				} else {
2065					error = copyout(opt->value,
2066					    optuio->uio_iov[pos].iov_base,
2067					    opt->len);
2068					if (error)
2069						break;
2070				}
2071			}
2072		}
2073	}
2074	goto done_errmsg;
2075
2076 done_deref:
2077	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2078	goto done_errmsg;
2079
2080 done_unlock_list:
2081	sx_sunlock(&allprison_lock);
2082 done_errmsg:
2083	if (error && errmsg_pos >= 0) {
2084		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2085		errmsg_pos = 2 * errmsg_pos + 1;
2086		if (errmsg_len > 0) {
2087			if (optuio->uio_segflg == UIO_SYSSPACE)
2088				bcopy(errmsg,
2089				    optuio->uio_iov[errmsg_pos].iov_base,
2090				    errmsg_len);
2091			else
2092				copyout(errmsg,
2093				    optuio->uio_iov[errmsg_pos].iov_base,
2094				    errmsg_len);
2095		}
2096	}
2097	vfs_freeopts(opts);
2098	return (error);
2099}
2100
2101
2102/*
2103 * struct jail_remove_args {
2104 *	int jid;
2105 * };
2106 */
2107int
2108jail_remove(struct thread *td, struct jail_remove_args *uap)
2109{
2110	struct prison *pr, *cpr, *lpr, *tpr;
2111	int descend, error;
2112
2113	error = priv_check(td, PRIV_JAIL_REMOVE);
2114	if (error)
2115		return (error);
2116
2117	sx_xlock(&allprison_lock);
2118	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2119	if (pr == NULL) {
2120		sx_xunlock(&allprison_lock);
2121		return (EINVAL);
2122	}
2123
2124	/* Remove all descendants of this prison, then remove this prison. */
2125	pr->pr_ref++;
2126	pr->pr_flags |= PR_REMOVE;
2127	if (!LIST_EMPTY(&pr->pr_children)) {
2128		mtx_unlock(&pr->pr_mtx);
2129		lpr = NULL;
2130		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2131			mtx_lock(&cpr->pr_mtx);
2132			if (cpr->pr_ref > 0) {
2133				tpr = cpr;
2134				cpr->pr_ref++;
2135				cpr->pr_flags |= PR_REMOVE;
2136			} else {
2137				/* Already removed - do not do it again. */
2138				tpr = NULL;
2139			}
2140			mtx_unlock(&cpr->pr_mtx);
2141			if (lpr != NULL) {
2142				mtx_lock(&lpr->pr_mtx);
2143				prison_remove_one(lpr);
2144				sx_xlock(&allprison_lock);
2145			}
2146			lpr = tpr;
2147		}
2148		if (lpr != NULL) {
2149			mtx_lock(&lpr->pr_mtx);
2150			prison_remove_one(lpr);
2151			sx_xlock(&allprison_lock);
2152		}
2153		mtx_lock(&pr->pr_mtx);
2154	}
2155	prison_remove_one(pr);
2156	return (0);
2157}
2158
2159static void
2160prison_remove_one(struct prison *pr)
2161{
2162	struct proc *p;
2163	int deuref;
2164
2165	/* If the prison was persistent, it is not anymore. */
2166	deuref = 0;
2167	if (pr->pr_flags & PR_PERSIST) {
2168		pr->pr_ref--;
2169		deuref = PD_DEUREF;
2170		pr->pr_flags &= ~PR_PERSIST;
2171	}
2172
2173	/*
2174	 * jail_remove added a reference.  If that's the only one, remove
2175	 * the prison now.
2176	 */
2177	KASSERT(pr->pr_ref > 0,
2178	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2179	if (pr->pr_ref == 1) {
2180		prison_deref(pr,
2181		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2182		return;
2183	}
2184
2185	mtx_unlock(&pr->pr_mtx);
2186	sx_xunlock(&allprison_lock);
2187	/*
2188	 * Kill all processes unfortunate enough to be attached to this prison.
2189	 */
2190	sx_slock(&allproc_lock);
2191	LIST_FOREACH(p, &allproc, p_list) {
2192		PROC_LOCK(p);
2193		if (p->p_state != PRS_NEW && p->p_ucred &&
2194		    p->p_ucred->cr_prison == pr)
2195			psignal(p, SIGKILL);
2196		PROC_UNLOCK(p);
2197	}
2198	sx_sunlock(&allproc_lock);
2199	/* Remove the temporary reference added by jail_remove. */
2200	prison_deref(pr, deuref | PD_DEREF);
2201}
2202
2203
2204/*
2205 * struct jail_attach_args {
2206 *	int jid;
2207 * };
2208 */
2209int
2210jail_attach(struct thread *td, struct jail_attach_args *uap)
2211{
2212	struct prison *pr;
2213	int error;
2214
2215	error = priv_check(td, PRIV_JAIL_ATTACH);
2216	if (error)
2217		return (error);
2218
2219	sx_slock(&allprison_lock);
2220	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2221	if (pr == NULL) {
2222		sx_sunlock(&allprison_lock);
2223		return (EINVAL);
2224	}
2225
2226	/*
2227	 * Do not allow a process to attach to a prison that is not
2228	 * considered to be "alive".
2229	 */
2230	if (pr->pr_uref == 0) {
2231		mtx_unlock(&pr->pr_mtx);
2232		sx_sunlock(&allprison_lock);
2233		return (EINVAL);
2234	}
2235
2236	return (do_jail_attach(td, pr));
2237}
2238
2239static int
2240do_jail_attach(struct thread *td, struct prison *pr)
2241{
2242	struct prison *ppr;
2243	struct proc *p;
2244	struct ucred *newcred, *oldcred;
2245	int vfslocked, error;
2246
2247	/*
2248	 * XXX: Note that there is a slight race here if two threads
2249	 * in the same privileged process attempt to attach to two
2250	 * different jails at the same time.  It is important for
2251	 * user processes not to do this, or they might end up with
2252	 * a process root from one prison, but attached to the jail
2253	 * of another.
2254	 */
2255	pr->pr_ref++;
2256	pr->pr_uref++;
2257	mtx_unlock(&pr->pr_mtx);
2258
2259	/* Let modules do whatever they need to prepare for attaching. */
2260	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2261	if (error) {
2262		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2263		return (error);
2264	}
2265	sx_sunlock(&allprison_lock);
2266
2267	/*
2268	 * Reparent the newly attached process to this jail.
2269	 */
2270	ppr = td->td_ucred->cr_prison;
2271	p = td->td_proc;
2272	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2273	if (error)
2274		goto e_revert_osd;
2275
2276	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2277	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2278	if ((error = change_dir(pr->pr_root, td)) != 0)
2279		goto e_unlock;
2280#ifdef MAC
2281	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2282		goto e_unlock;
2283#endif
2284	VOP_UNLOCK(pr->pr_root, 0);
2285	if ((error = change_root(pr->pr_root, td)))
2286		goto e_unlock_giant;
2287	VFS_UNLOCK_GIANT(vfslocked);
2288
2289	newcred = crget();
2290	PROC_LOCK(p);
2291	oldcred = p->p_ucred;
2292	setsugid(p);
2293	crcopy(newcred, oldcred);
2294	newcred->cr_prison = pr;
2295	p->p_ucred = newcred;
2296	PROC_UNLOCK(p);
2297	crfree(oldcred);
2298	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2299	return (0);
2300 e_unlock:
2301	VOP_UNLOCK(pr->pr_root, 0);
2302 e_unlock_giant:
2303	VFS_UNLOCK_GIANT(vfslocked);
2304 e_revert_osd:
2305	/* Tell modules this thread is still in its old jail after all. */
2306	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2307	prison_deref(pr, PD_DEREF | PD_DEUREF);
2308	return (error);
2309}
2310
2311
2312/*
2313 * Returns a locked prison instance, or NULL on failure.
2314 */
2315struct prison *
2316prison_find(int prid)
2317{
2318	struct prison *pr;
2319
2320	sx_assert(&allprison_lock, SX_LOCKED);
2321	TAILQ_FOREACH(pr, &allprison, pr_list) {
2322		if (pr->pr_id == prid) {
2323			mtx_lock(&pr->pr_mtx);
2324			if (pr->pr_ref > 0)
2325				return (pr);
2326			mtx_unlock(&pr->pr_mtx);
2327		}
2328	}
2329	return (NULL);
2330}
2331
2332/*
2333 * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2334 */
2335struct prison *
2336prison_find_child(struct prison *mypr, int prid)
2337{
2338	struct prison *pr;
2339	int descend;
2340
2341	sx_assert(&allprison_lock, SX_LOCKED);
2342	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2343		if (pr->pr_id == prid) {
2344			mtx_lock(&pr->pr_mtx);
2345			if (pr->pr_ref > 0)
2346				return (pr);
2347			mtx_unlock(&pr->pr_mtx);
2348		}
2349	}
2350	return (NULL);
2351}
2352
2353/*
2354 * Look for the name relative to mypr.  Returns a locked prison or NULL.
2355 */
2356struct prison *
2357prison_find_name(struct prison *mypr, const char *name)
2358{
2359	struct prison *pr, *deadpr;
2360	size_t mylen;
2361	int descend;
2362
2363	sx_assert(&allprison_lock, SX_LOCKED);
2364	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2365 again:
2366	deadpr = NULL;
2367	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2368		if (!strcmp(pr->pr_name + mylen, name)) {
2369			mtx_lock(&pr->pr_mtx);
2370			if (pr->pr_ref > 0) {
2371				if (pr->pr_uref > 0)
2372					return (pr);
2373				deadpr = pr;
2374			}
2375			mtx_unlock(&pr->pr_mtx);
2376		}
2377	}
2378	/* There was no valid prison - perhaps there was a dying one. */
2379	if (deadpr != NULL) {
2380		mtx_lock(&deadpr->pr_mtx);
2381		if (deadpr->pr_ref == 0) {
2382			mtx_unlock(&deadpr->pr_mtx);
2383			goto again;
2384		}
2385	}
2386	return (deadpr);
2387}
2388
2389/*
2390 * See if a prison has the specific flag set.
2391 */
2392int
2393prison_flag(struct ucred *cred, unsigned flag)
2394{
2395
2396	/* This is an atomic read, so no locking is necessary. */
2397	return (cred->cr_prison->pr_flags & flag);
2398}
2399
2400int
2401prison_allow(struct ucred *cred, unsigned flag)
2402{
2403
2404	/* This is an atomic read, so no locking is necessary. */
2405	return (cred->cr_prison->pr_allow & flag);
2406}
2407
2408/*
2409 * Remove a prison reference.  If that was the last reference, remove the
2410 * prison itself - but not in this context in case there are locks held.
2411 */
2412void
2413prison_free_locked(struct prison *pr)
2414{
2415
2416	mtx_assert(&pr->pr_mtx, MA_OWNED);
2417	pr->pr_ref--;
2418	if (pr->pr_ref == 0) {
2419		mtx_unlock(&pr->pr_mtx);
2420		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2421		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2422		return;
2423	}
2424	mtx_unlock(&pr->pr_mtx);
2425}
2426
2427void
2428prison_free(struct prison *pr)
2429{
2430
2431	mtx_lock(&pr->pr_mtx);
2432	prison_free_locked(pr);
2433}
2434
2435static void
2436prison_complete(void *context, int pending)
2437{
2438
2439	prison_deref((struct prison *)context, 0);
2440}
2441
2442/*
2443 * Remove a prison reference (usually).  This internal version assumes no
2444 * mutexes are held, except perhaps the prison itself.  If there are no more
2445 * references, release and delist the prison.  On completion, the prison lock
2446 * and the allprison lock are both unlocked.
2447 */
2448static void
2449prison_deref(struct prison *pr, int flags)
2450{
2451	struct prison *ppr, *tpr;
2452	int vfslocked;
2453
2454	if (!(flags & PD_LOCKED))
2455		mtx_lock(&pr->pr_mtx);
2456	/* Decrement the user references in a separate loop. */
2457	if (flags & PD_DEUREF) {
2458		for (tpr = pr;; tpr = tpr->pr_parent) {
2459			if (tpr != pr)
2460				mtx_lock(&tpr->pr_mtx);
2461			if (--tpr->pr_uref > 0)
2462				break;
2463			KASSERT(tpr != &prison0, ("prison0 pr_uref=0"));
2464			mtx_unlock(&tpr->pr_mtx);
2465		}
2466		/* Done if there were only user references to remove. */
2467		if (!(flags & PD_DEREF)) {
2468			mtx_unlock(&tpr->pr_mtx);
2469			if (flags & PD_LIST_SLOCKED)
2470				sx_sunlock(&allprison_lock);
2471			else if (flags & PD_LIST_XLOCKED)
2472				sx_xunlock(&allprison_lock);
2473			return;
2474		}
2475		if (tpr != pr) {
2476			mtx_unlock(&tpr->pr_mtx);
2477			mtx_lock(&pr->pr_mtx);
2478		}
2479	}
2480
2481	for (;;) {
2482		if (flags & PD_DEREF)
2483			pr->pr_ref--;
2484		/* If the prison still has references, nothing else to do. */
2485		if (pr->pr_ref > 0) {
2486			mtx_unlock(&pr->pr_mtx);
2487			if (flags & PD_LIST_SLOCKED)
2488				sx_sunlock(&allprison_lock);
2489			else if (flags & PD_LIST_XLOCKED)
2490				sx_xunlock(&allprison_lock);
2491			return;
2492		}
2493
2494		mtx_unlock(&pr->pr_mtx);
2495		if (flags & PD_LIST_SLOCKED) {
2496			if (!sx_try_upgrade(&allprison_lock)) {
2497				sx_sunlock(&allprison_lock);
2498				sx_xlock(&allprison_lock);
2499			}
2500		} else if (!(flags & PD_LIST_XLOCKED))
2501			sx_xlock(&allprison_lock);
2502
2503		TAILQ_REMOVE(&allprison, pr, pr_list);
2504		LIST_REMOVE(pr, pr_sibling);
2505		ppr = pr->pr_parent;
2506		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2507			tpr->pr_childcount--;
2508		sx_downgrade(&allprison_lock);
2509
2510#ifdef VIMAGE
2511		if (pr->pr_flags & PR_VNET)
2512			vnet_destroy(pr->pr_vnet);
2513#endif
2514		if (pr->pr_root != NULL) {
2515			vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2516			vrele(pr->pr_root);
2517			VFS_UNLOCK_GIANT(vfslocked);
2518		}
2519		mtx_destroy(&pr->pr_mtx);
2520#ifdef INET
2521		free(pr->pr_ip4, M_PRISON);
2522#endif
2523#ifdef INET6
2524		free(pr->pr_ip6, M_PRISON);
2525#endif
2526		if (pr->pr_cpuset != NULL)
2527			cpuset_rel(pr->pr_cpuset);
2528		osd_jail_exit(pr);
2529		free(pr, M_PRISON);
2530
2531		/* Removing a prison frees a reference on its parent. */
2532		pr = ppr;
2533		mtx_lock(&pr->pr_mtx);
2534		flags = PD_DEREF | PD_LIST_SLOCKED;
2535	}
2536}
2537
2538void
2539prison_hold_locked(struct prison *pr)
2540{
2541
2542	mtx_assert(&pr->pr_mtx, MA_OWNED);
2543	KASSERT(pr->pr_ref > 0,
2544	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2545	pr->pr_ref++;
2546}
2547
2548void
2549prison_hold(struct prison *pr)
2550{
2551
2552	mtx_lock(&pr->pr_mtx);
2553	prison_hold_locked(pr);
2554	mtx_unlock(&pr->pr_mtx);
2555}
2556
2557void
2558prison_proc_hold(struct prison *pr)
2559{
2560
2561	mtx_lock(&pr->pr_mtx);
2562	KASSERT(pr->pr_uref > 0,
2563	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2564	pr->pr_uref++;
2565	mtx_unlock(&pr->pr_mtx);
2566}
2567
2568void
2569prison_proc_free(struct prison *pr)
2570{
2571
2572	mtx_lock(&pr->pr_mtx);
2573	KASSERT(pr->pr_uref > 0,
2574	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2575	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2576}
2577
2578
2579#ifdef INET
2580/*
2581 * Restrict a prison's IP address list with its parent's, possibly replacing
2582 * it.  Return true if the replacement buffer was used (or would have been).
2583 */
2584static int
2585prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2586{
2587	int ii, ij, used;
2588	struct prison *ppr;
2589
2590	ppr = pr->pr_parent;
2591	if (!(pr->pr_flags & PR_IP4_USER)) {
2592		/* This has no user settings, so just copy the parent's list. */
2593		if (pr->pr_ip4s < ppr->pr_ip4s) {
2594			/*
2595			 * There's no room for the parent's list.  Use the
2596			 * new list buffer, which is assumed to be big enough
2597			 * (if it was passed).  If there's no buffer, try to
2598			 * allocate one.
2599			 */
2600			used = 1;
2601			if (newip4 == NULL) {
2602				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2603				    M_PRISON, M_NOWAIT);
2604				if (newip4 != NULL)
2605					used = 0;
2606			}
2607			if (newip4 != NULL) {
2608				bcopy(ppr->pr_ip4, newip4,
2609				    ppr->pr_ip4s * sizeof(*newip4));
2610				free(pr->pr_ip4, M_PRISON);
2611				pr->pr_ip4 = newip4;
2612				pr->pr_ip4s = ppr->pr_ip4s;
2613				pr->pr_flags |= PR_IP4;
2614			}
2615			return (used);
2616		}
2617		pr->pr_ip4s = ppr->pr_ip4s;
2618		if (pr->pr_ip4s > 0)
2619			bcopy(ppr->pr_ip4, pr->pr_ip4,
2620			    pr->pr_ip4s * sizeof(*newip4));
2621		else if (pr->pr_ip4 != NULL) {
2622			free(pr->pr_ip4, M_PRISON);
2623			pr->pr_ip4 = NULL;
2624		}
2625		pr->pr_flags =
2626			(pr->pr_flags & ~PR_IP4) | (ppr->pr_flags & PR_IP4);
2627	} else if (pr->pr_ip4s > 0 && (ppr->pr_flags & PR_IP4)) {
2628		/* Remove addresses that aren't in the parent. */
2629		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2630			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2631				break;
2632		if (ij < ppr->pr_ip4s)
2633			ii = 1;
2634		else {
2635			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2636			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2637			ii = 0;
2638		}
2639		for (ij = 1; ii < pr->pr_ip4s; ) {
2640			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2641				ii++;
2642				continue;
2643			}
2644			switch (ij >= ppr->pr_ip4s ? -1 :
2645				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2646			case -1:
2647				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2648				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2649				break;
2650			case 0:
2651				ii++;
2652				ij++;
2653				break;
2654			case 1:
2655				ij++;
2656				break;
2657			}
2658		}
2659		if (pr->pr_ip4s == 0) {
2660			pr->pr_flags |= PR_IP4_DISABLE;
2661			free(pr->pr_ip4, M_PRISON);
2662			pr->pr_ip4 = NULL;
2663		}
2664	}
2665	return (0);
2666}
2667
2668/*
2669 * Pass back primary IPv4 address of this jail.
2670 *
2671 * If not restricted return success but do not alter the address.  Caller has
2672 * to make sure to initialize it correctly (e.g. INADDR_ANY).
2673 *
2674 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2675 * Address returned in NBO.
2676 */
2677int
2678prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2679{
2680	struct prison *pr;
2681
2682	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2683	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2684
2685	pr = cred->cr_prison;
2686	if (!(pr->pr_flags & PR_IP4))
2687		return (0);
2688	mtx_lock(&pr->pr_mtx);
2689	if (!(pr->pr_flags & PR_IP4)) {
2690		mtx_unlock(&pr->pr_mtx);
2691		return (0);
2692	}
2693	if (pr->pr_ip4 == NULL) {
2694		mtx_unlock(&pr->pr_mtx);
2695		return (EAFNOSUPPORT);
2696	}
2697
2698	ia->s_addr = pr->pr_ip4[0].s_addr;
2699	mtx_unlock(&pr->pr_mtx);
2700	return (0);
2701}
2702
2703/*
2704 * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2705 */
2706int
2707prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2708{
2709
2710	if (pr1 == pr2)
2711		return (1);
2712
2713	/*
2714	 * jail_set maintains an exclusive hold on allprison_lock while it
2715	 * changes the IP addresses, so only a shared hold is needed.  This is
2716	 * easier than locking the two prisons which would require finding the
2717	 * proper locking order and end up needing allprison_lock anyway.
2718	 */
2719	sx_slock(&allprison_lock);
2720	while (pr1 != &prison0 && !(pr1->pr_flags & PR_IP4_USER))
2721		pr1 = pr1->pr_parent;
2722	while (pr2 != &prison0 && !(pr2->pr_flags & PR_IP4_USER))
2723		pr2 = pr2->pr_parent;
2724	sx_sunlock(&allprison_lock);
2725	return (pr1 == pr2);
2726}
2727
2728/*
2729 * Make sure our (source) address is set to something meaningful to this
2730 * jail.
2731 *
2732 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2733 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2734 * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2735 */
2736int
2737prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2738{
2739	struct prison *pr;
2740	struct in_addr ia0;
2741	int error;
2742
2743	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2744	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2745
2746	pr = cred->cr_prison;
2747	if (!(pr->pr_flags & PR_IP4))
2748		return (0);
2749	mtx_lock(&pr->pr_mtx);
2750	if (!(pr->pr_flags & PR_IP4)) {
2751		mtx_unlock(&pr->pr_mtx);
2752		return (0);
2753	}
2754	if (pr->pr_ip4 == NULL) {
2755		mtx_unlock(&pr->pr_mtx);
2756		return (EAFNOSUPPORT);
2757	}
2758
2759	ia0.s_addr = ntohl(ia->s_addr);
2760	if (ia0.s_addr == INADDR_LOOPBACK) {
2761		ia->s_addr = pr->pr_ip4[0].s_addr;
2762		mtx_unlock(&pr->pr_mtx);
2763		return (0);
2764	}
2765
2766	if (ia0.s_addr == INADDR_ANY) {
2767		/*
2768		 * In case there is only 1 IPv4 address, bind directly.
2769		 */
2770		if (pr->pr_ip4s == 1)
2771			ia->s_addr = pr->pr_ip4[0].s_addr;
2772		mtx_unlock(&pr->pr_mtx);
2773		return (0);
2774	}
2775
2776	error = _prison_check_ip4(pr, ia);
2777	mtx_unlock(&pr->pr_mtx);
2778	return (error);
2779}
2780
2781/*
2782 * Rewrite destination address in case we will connect to loopback address.
2783 *
2784 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2785 * Address passed in in NBO and returned in NBO.
2786 */
2787int
2788prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2789{
2790	struct prison *pr;
2791
2792	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2793	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2794
2795	pr = cred->cr_prison;
2796	if (!(pr->pr_flags & PR_IP4))
2797		return (0);
2798	mtx_lock(&pr->pr_mtx);
2799	if (!(pr->pr_flags & PR_IP4)) {
2800		mtx_unlock(&pr->pr_mtx);
2801		return (0);
2802	}
2803	if (pr->pr_ip4 == NULL) {
2804		mtx_unlock(&pr->pr_mtx);
2805		return (EAFNOSUPPORT);
2806	}
2807
2808	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2809		ia->s_addr = pr->pr_ip4[0].s_addr;
2810		mtx_unlock(&pr->pr_mtx);
2811		return (0);
2812	}
2813
2814	/*
2815	 * Return success because nothing had to be changed.
2816	 */
2817	mtx_unlock(&pr->pr_mtx);
2818	return (0);
2819}
2820
2821/*
2822 * Check if given address belongs to the jail referenced by cred/prison.
2823 *
2824 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2825 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2826 * doesn't allow IPv4.  Address passed in in NBO.
2827 */
2828static int
2829_prison_check_ip4(struct prison *pr, struct in_addr *ia)
2830{
2831	int i, a, z, d;
2832
2833	/*
2834	 * Check the primary IP.
2835	 */
2836	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2837		return (0);
2838
2839	/*
2840	 * All the other IPs are sorted so we can do a binary search.
2841	 */
2842	a = 0;
2843	z = pr->pr_ip4s - 2;
2844	while (a <= z) {
2845		i = (a + z) / 2;
2846		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2847		if (d > 0)
2848			z = i - 1;
2849		else if (d < 0)
2850			a = i + 1;
2851		else
2852			return (0);
2853	}
2854
2855	return (EADDRNOTAVAIL);
2856}
2857
2858int
2859prison_check_ip4(struct ucred *cred, struct in_addr *ia)
2860{
2861	struct prison *pr;
2862	int error;
2863
2864	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2865	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2866
2867	pr = cred->cr_prison;
2868	if (!(pr->pr_flags & PR_IP4))
2869		return (0);
2870	mtx_lock(&pr->pr_mtx);
2871	if (!(pr->pr_flags & PR_IP4)) {
2872		mtx_unlock(&pr->pr_mtx);
2873		return (0);
2874	}
2875	if (pr->pr_ip4 == NULL) {
2876		mtx_unlock(&pr->pr_mtx);
2877		return (EAFNOSUPPORT);
2878	}
2879
2880	error = _prison_check_ip4(pr, ia);
2881	mtx_unlock(&pr->pr_mtx);
2882	return (error);
2883}
2884#endif
2885
2886#ifdef INET6
2887static int
2888prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2889{
2890	int ii, ij, used;
2891	struct prison *ppr;
2892
2893	ppr = pr->pr_parent;
2894	if (!(pr->pr_flags & PR_IP6_USER)) {
2895		/* This has no user settings, so just copy the parent's list. */
2896		if (pr->pr_ip6s < ppr->pr_ip6s) {
2897			/*
2898			 * There's no room for the parent's list.  Use the
2899			 * new list buffer, which is assumed to be big enough
2900			 * (if it was passed).  If there's no buffer, try to
2901			 * allocate one.
2902			 */
2903			used = 1;
2904			if (newip6 == NULL) {
2905				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
2906				    M_PRISON, M_NOWAIT);
2907				if (newip6 != NULL)
2908					used = 0;
2909			}
2910			if (newip6 != NULL) {
2911				bcopy(ppr->pr_ip6, newip6,
2912				    ppr->pr_ip6s * sizeof(*newip6));
2913				free(pr->pr_ip6, M_PRISON);
2914				pr->pr_ip6 = newip6;
2915				pr->pr_ip6s = ppr->pr_ip6s;
2916				pr->pr_flags |= PR_IP6;
2917			}
2918			return (used);
2919		}
2920		pr->pr_ip6s = ppr->pr_ip6s;
2921		if (pr->pr_ip6s > 0)
2922			bcopy(ppr->pr_ip6, pr->pr_ip6,
2923			    pr->pr_ip6s * sizeof(*newip6));
2924		else if (pr->pr_ip6 != NULL) {
2925			free(pr->pr_ip6, M_PRISON);
2926			pr->pr_ip6 = NULL;
2927		}
2928		pr->pr_flags =
2929			(pr->pr_flags & ~PR_IP6) | (ppr->pr_flags & PR_IP6);
2930	} else if (pr->pr_ip6s > 0 && (ppr->pr_flags & PR_IP6)) {
2931		/* Remove addresses that aren't in the parent. */
2932		for (ij = 0; ij < ppr->pr_ip6s; ij++)
2933			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
2934			    &ppr->pr_ip6[ij]))
2935				break;
2936		if (ij < ppr->pr_ip6s)
2937			ii = 1;
2938		else {
2939			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
2940			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
2941			ii = 0;
2942		}
2943		for (ij = 1; ii < pr->pr_ip6s; ) {
2944			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
2945			    &ppr->pr_ip6[0])) {
2946				ii++;
2947				continue;
2948			}
2949			switch (ij >= ppr->pr_ip4s ? -1 :
2950				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
2951			case -1:
2952				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
2953				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
2954				break;
2955			case 0:
2956				ii++;
2957				ij++;
2958				break;
2959			case 1:
2960				ij++;
2961				break;
2962			}
2963		}
2964		if (pr->pr_ip6s == 0) {
2965			pr->pr_flags |= PR_IP6_DISABLE;
2966			free(pr->pr_ip6, M_PRISON);
2967			pr->pr_ip6 = NULL;
2968		}
2969	}
2970	return 0;
2971}
2972
2973/*
2974 * Pass back primary IPv6 address for this jail.
2975 *
2976 * If not restricted return success but do not alter the address.  Caller has
2977 * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
2978 *
2979 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
2980 */
2981int
2982prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
2983{
2984	struct prison *pr;
2985
2986	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2987	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
2988
2989	pr = cred->cr_prison;
2990	if (!(pr->pr_flags & PR_IP6))
2991		return (0);
2992	mtx_lock(&pr->pr_mtx);
2993	if (!(pr->pr_flags & PR_IP6)) {
2994		mtx_unlock(&pr->pr_mtx);
2995		return (0);
2996	}
2997	if (pr->pr_ip6 == NULL) {
2998		mtx_unlock(&pr->pr_mtx);
2999		return (EAFNOSUPPORT);
3000	}
3001
3002	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3003	mtx_unlock(&pr->pr_mtx);
3004	return (0);
3005}
3006
3007/*
3008 * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3009 */
3010int
3011prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3012{
3013
3014	if (pr1 == pr2)
3015		return (1);
3016
3017	sx_slock(&allprison_lock);
3018	while (pr1 != &prison0 && !(pr1->pr_flags & PR_IP6_USER))
3019		pr1 = pr1->pr_parent;
3020	while (pr2 != &prison0 && !(pr2->pr_flags & PR_IP6_USER))
3021		pr2 = pr2->pr_parent;
3022	sx_sunlock(&allprison_lock);
3023	return (pr1 == pr2);
3024}
3025
3026/*
3027 * Make sure our (source) address is set to something meaningful to this jail.
3028 *
3029 * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3030 * when needed while binding.
3031 *
3032 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3033 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3034 * doesn't allow IPv6.
3035 */
3036int
3037prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3038{
3039	struct prison *pr;
3040	int error;
3041
3042	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3043	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3044
3045	pr = cred->cr_prison;
3046	if (!(pr->pr_flags & PR_IP6))
3047		return (0);
3048	mtx_lock(&pr->pr_mtx);
3049	if (!(pr->pr_flags & PR_IP6)) {
3050		mtx_unlock(&pr->pr_mtx);
3051		return (0);
3052	}
3053	if (pr->pr_ip6 == NULL) {
3054		mtx_unlock(&pr->pr_mtx);
3055		return (EAFNOSUPPORT);
3056	}
3057
3058	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3059		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3060		mtx_unlock(&pr->pr_mtx);
3061		return (0);
3062	}
3063
3064	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3065		/*
3066		 * In case there is only 1 IPv6 address, and v6only is true,
3067		 * then bind directly.
3068		 */
3069		if (v6only != 0 && pr->pr_ip6s == 1)
3070			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3071		mtx_unlock(&pr->pr_mtx);
3072		return (0);
3073	}
3074
3075	error = _prison_check_ip6(pr, ia6);
3076	mtx_unlock(&pr->pr_mtx);
3077	return (error);
3078}
3079
3080/*
3081 * Rewrite destination address in case we will connect to loopback address.
3082 *
3083 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3084 */
3085int
3086prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3087{
3088	struct prison *pr;
3089
3090	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3091	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3092
3093	pr = cred->cr_prison;
3094	if (!(pr->pr_flags & PR_IP6))
3095		return (0);
3096	mtx_lock(&pr->pr_mtx);
3097	if (!(pr->pr_flags & PR_IP6)) {
3098		mtx_unlock(&pr->pr_mtx);
3099		return (0);
3100	}
3101	if (pr->pr_ip6 == NULL) {
3102		mtx_unlock(&pr->pr_mtx);
3103		return (EAFNOSUPPORT);
3104	}
3105
3106	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3107		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3108		mtx_unlock(&pr->pr_mtx);
3109		return (0);
3110	}
3111
3112	/*
3113	 * Return success because nothing had to be changed.
3114	 */
3115	mtx_unlock(&pr->pr_mtx);
3116	return (0);
3117}
3118
3119/*
3120 * Check if given address belongs to the jail referenced by cred/prison.
3121 *
3122 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3123 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3124 * doesn't allow IPv6.
3125 */
3126static int
3127_prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3128{
3129	int i, a, z, d;
3130
3131	/*
3132	 * Check the primary IP.
3133	 */
3134	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3135		return (0);
3136
3137	/*
3138	 * All the other IPs are sorted so we can do a binary search.
3139	 */
3140	a = 0;
3141	z = pr->pr_ip6s - 2;
3142	while (a <= z) {
3143		i = (a + z) / 2;
3144		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3145		if (d > 0)
3146			z = i - 1;
3147		else if (d < 0)
3148			a = i + 1;
3149		else
3150			return (0);
3151	}
3152
3153	return (EADDRNOTAVAIL);
3154}
3155
3156int
3157prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3158{
3159	struct prison *pr;
3160	int error;
3161
3162	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3163	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3164
3165	pr = cred->cr_prison;
3166	if (!(pr->pr_flags & PR_IP6))
3167		return (0);
3168	mtx_lock(&pr->pr_mtx);
3169	if (!(pr->pr_flags & PR_IP6)) {
3170		mtx_unlock(&pr->pr_mtx);
3171		return (0);
3172	}
3173	if (pr->pr_ip6 == NULL) {
3174		mtx_unlock(&pr->pr_mtx);
3175		return (EAFNOSUPPORT);
3176	}
3177
3178	error = _prison_check_ip6(pr, ia6);
3179	mtx_unlock(&pr->pr_mtx);
3180	return (error);
3181}
3182#endif
3183
3184/*
3185 * Check if a jail supports the given address family.
3186 *
3187 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3188 * if not.
3189 */
3190int
3191prison_check_af(struct ucred *cred, int af)
3192{
3193	struct prison *pr;
3194	int error;
3195
3196	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3197
3198	pr = cred->cr_prison;
3199#ifdef VIMAGE
3200	/* Prisons with their own network stack are not limited. */
3201	if (pr->pr_flags & PR_VNET)
3202		return (0);
3203#endif
3204
3205	error = 0;
3206	switch (af)
3207	{
3208#ifdef INET
3209	case AF_INET:
3210		if (pr->pr_flags & PR_IP4)
3211		{
3212			mtx_lock(&pr->pr_mtx);
3213			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3214				error = EAFNOSUPPORT;
3215			mtx_unlock(&pr->pr_mtx);
3216		}
3217		break;
3218#endif
3219#ifdef INET6
3220	case AF_INET6:
3221		if (pr->pr_flags & PR_IP6)
3222		{
3223			mtx_lock(&pr->pr_mtx);
3224			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3225				error = EAFNOSUPPORT;
3226			mtx_unlock(&pr->pr_mtx);
3227		}
3228		break;
3229#endif
3230	case AF_LOCAL:
3231	case AF_ROUTE:
3232		break;
3233	default:
3234		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3235			error = EAFNOSUPPORT;
3236	}
3237	return (error);
3238}
3239
3240/*
3241 * Check if given address belongs to the jail referenced by cred (wrapper to
3242 * prison_check_ip[46]).
3243 *
3244 * Returns 0 if jail doesn't restrict the address family or if address belongs
3245 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3246 * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3247 */
3248int
3249prison_if(struct ucred *cred, struct sockaddr *sa)
3250{
3251#ifdef INET
3252	struct sockaddr_in *sai;
3253#endif
3254#ifdef INET6
3255	struct sockaddr_in6 *sai6;
3256#endif
3257	int error;
3258
3259	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3260	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3261
3262	error = 0;
3263	switch (sa->sa_family)
3264	{
3265#ifdef INET
3266	case AF_INET:
3267		sai = (struct sockaddr_in *)sa;
3268		error = prison_check_ip4(cred, &sai->sin_addr);
3269		break;
3270#endif
3271#ifdef INET6
3272	case AF_INET6:
3273		sai6 = (struct sockaddr_in6 *)sa;
3274		error = prison_check_ip6(cred, &sai6->sin6_addr);
3275		break;
3276#endif
3277	default:
3278		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3279			error = EAFNOSUPPORT;
3280	}
3281	return (error);
3282}
3283
3284/*
3285 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3286 */
3287int
3288prison_check(struct ucred *cred1, struct ucred *cred2)
3289{
3290
3291	return ((cred1->cr_prison == cred2->cr_prison ||
3292	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3293}
3294
3295/*
3296 * Return 1 if p2 is a child of p1, otherwise 0.
3297 */
3298int
3299prison_ischild(struct prison *pr1, struct prison *pr2)
3300{
3301
3302	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3303		if (pr1 == pr2)
3304			return (1);
3305	return (0);
3306}
3307
3308/*
3309 * Return 1 if the passed credential is in a jail, otherwise 0.
3310 */
3311int
3312jailed(struct ucred *cred)
3313{
3314
3315	return (cred->cr_prison != &prison0);
3316}
3317
3318/*
3319 * Return the correct hostname (domainname, et al) for the passed credential.
3320 */
3321void
3322getcredhostname(struct ucred *cred, char *buf, size_t size)
3323{
3324	struct prison *pr;
3325
3326	/*
3327	 * A NULL credential can be used to shortcut to the physical
3328	 * system's hostname.
3329	 */
3330	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3331	mtx_lock(&pr->pr_mtx);
3332	strlcpy(buf, pr->pr_hostname, size);
3333	mtx_unlock(&pr->pr_mtx);
3334}
3335
3336void
3337getcreddomainname(struct ucred *cred, char *buf, size_t size)
3338{
3339
3340	mtx_lock(&cred->cr_prison->pr_mtx);
3341	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3342	mtx_unlock(&cred->cr_prison->pr_mtx);
3343}
3344
3345void
3346getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3347{
3348
3349	mtx_lock(&cred->cr_prison->pr_mtx);
3350	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3351	mtx_unlock(&cred->cr_prison->pr_mtx);
3352}
3353
3354void
3355getcredhostid(struct ucred *cred, unsigned long *hostid)
3356{
3357
3358	mtx_lock(&cred->cr_prison->pr_mtx);
3359	*hostid = cred->cr_prison->pr_hostid;
3360	mtx_unlock(&cred->cr_prison->pr_mtx);
3361}
3362
3363/*
3364 * Determine whether the subject represented by cred can "see"
3365 * status of a mount point.
3366 * Returns: 0 for permitted, ENOENT otherwise.
3367 * XXX: This function should be called cr_canseemount() and should be
3368 *      placed in kern_prot.c.
3369 */
3370int
3371prison_canseemount(struct ucred *cred, struct mount *mp)
3372{
3373	struct prison *pr;
3374	struct statfs *sp;
3375	size_t len;
3376
3377	pr = cred->cr_prison;
3378	if (pr->pr_enforce_statfs == 0)
3379		return (0);
3380	if (pr->pr_root->v_mount == mp)
3381		return (0);
3382	if (pr->pr_enforce_statfs == 2)
3383		return (ENOENT);
3384	/*
3385	 * If jail's chroot directory is set to "/" we should be able to see
3386	 * all mount-points from inside a jail.
3387	 * This is ugly check, but this is the only situation when jail's
3388	 * directory ends with '/'.
3389	 */
3390	if (strcmp(pr->pr_path, "/") == 0)
3391		return (0);
3392	len = strlen(pr->pr_path);
3393	sp = &mp->mnt_stat;
3394	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3395		return (ENOENT);
3396	/*
3397	 * Be sure that we don't have situation where jail's root directory
3398	 * is "/some/path" and mount point is "/some/pathpath".
3399	 */
3400	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3401		return (ENOENT);
3402	return (0);
3403}
3404
3405void
3406prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3407{
3408	char jpath[MAXPATHLEN];
3409	struct prison *pr;
3410	size_t len;
3411
3412	pr = cred->cr_prison;
3413	if (pr->pr_enforce_statfs == 0)
3414		return;
3415	if (prison_canseemount(cred, mp) != 0) {
3416		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3417		strlcpy(sp->f_mntonname, "[restricted]",
3418		    sizeof(sp->f_mntonname));
3419		return;
3420	}
3421	if (pr->pr_root->v_mount == mp) {
3422		/*
3423		 * Clear current buffer data, so we are sure nothing from
3424		 * the valid path left there.
3425		 */
3426		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3427		*sp->f_mntonname = '/';
3428		return;
3429	}
3430	/*
3431	 * If jail's chroot directory is set to "/" we should be able to see
3432	 * all mount-points from inside a jail.
3433	 */
3434	if (strcmp(pr->pr_path, "/") == 0)
3435		return;
3436	len = strlen(pr->pr_path);
3437	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3438	/*
3439	 * Clear current buffer data, so we are sure nothing from
3440	 * the valid path left there.
3441	 */
3442	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3443	if (*jpath == '\0') {
3444		/* Should never happen. */
3445		*sp->f_mntonname = '/';
3446	} else {
3447		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3448	}
3449}
3450
3451/*
3452 * Check with permission for a specific privilege is granted within jail.  We
3453 * have a specific list of accepted privileges; the rest are denied.
3454 */
3455int
3456prison_priv_check(struct ucred *cred, int priv)
3457{
3458
3459	if (!jailed(cred))
3460		return (0);
3461
3462#ifdef VIMAGE
3463	/*
3464	 * Privileges specific to prisons with a virtual network stack.
3465	 * There might be a duplicate entry here in case the privilege
3466	 * is only granted conditionally in the legacy jail case.
3467	 */
3468	switch (priv) {
3469#ifdef notyet
3470		/*
3471		 * NFS-specific privileges.
3472		 */
3473	case PRIV_NFS_DAEMON:
3474	case PRIV_NFS_LOCKD:
3475#endif
3476		/*
3477		 * Network stack privileges.
3478		 */
3479	case PRIV_NET_BRIDGE:
3480	case PRIV_NET_GRE:
3481	case PRIV_NET_BPF:
3482	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3483	case PRIV_NET_ROUTE:
3484	case PRIV_NET_TAP:
3485	case PRIV_NET_SETIFMTU:
3486	case PRIV_NET_SETIFFLAGS:
3487	case PRIV_NET_SETIFCAP:
3488	case PRIV_NET_SETIFNAME	:
3489	case PRIV_NET_SETIFMETRIC:
3490	case PRIV_NET_SETIFPHYS:
3491	case PRIV_NET_SETIFMAC:
3492	case PRIV_NET_ADDMULTI:
3493	case PRIV_NET_DELMULTI:
3494	case PRIV_NET_HWIOCTL:
3495	case PRIV_NET_SETLLADDR:
3496	case PRIV_NET_ADDIFGROUP:
3497	case PRIV_NET_DELIFGROUP:
3498	case PRIV_NET_IFCREATE:
3499	case PRIV_NET_IFDESTROY:
3500	case PRIV_NET_ADDIFADDR:
3501	case PRIV_NET_DELIFADDR:
3502	case PRIV_NET_LAGG:
3503	case PRIV_NET_GIF:
3504	case PRIV_NET_SETIFVNET:
3505
3506		/*
3507		 * 802.11-related privileges.
3508		 */
3509	case PRIV_NET80211_GETKEY:
3510#ifdef notyet
3511	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3512#endif
3513
3514#ifdef notyet
3515		/*
3516		 * AppleTalk privileges.
3517		 */
3518	case PRIV_NETATALK_RESERVEDPORT:
3519
3520		/*
3521		 * ATM privileges.
3522		 */
3523	case PRIV_NETATM_CFG:
3524	case PRIV_NETATM_ADD:
3525	case PRIV_NETATM_DEL:
3526	case PRIV_NETATM_SET:
3527
3528		/*
3529		 * Bluetooth privileges.
3530		 */
3531	case PRIV_NETBLUETOOTH_RAW:
3532#endif
3533
3534		/*
3535		 * Netgraph and netgraph module privileges.
3536		 */
3537	case PRIV_NETGRAPH_CONTROL:
3538#ifdef notyet
3539	case PRIV_NETGRAPH_TTY:
3540#endif
3541
3542		/*
3543		 * IPv4 and IPv6 privileges.
3544		 */
3545	case PRIV_NETINET_IPFW:
3546	case PRIV_NETINET_DIVERT:
3547	case PRIV_NETINET_PF:
3548	case PRIV_NETINET_DUMMYNET:
3549	case PRIV_NETINET_CARP:
3550	case PRIV_NETINET_MROUTE:
3551	case PRIV_NETINET_RAW:
3552	case PRIV_NETINET_ADDRCTRL6:
3553	case PRIV_NETINET_ND6:
3554	case PRIV_NETINET_SCOPE6:
3555	case PRIV_NETINET_ALIFETIME6:
3556	case PRIV_NETINET_IPSEC:
3557	case PRIV_NETINET_BINDANY:
3558
3559#ifdef notyet
3560		/*
3561		 * IPX/SPX privileges.
3562		 */
3563	case PRIV_NETIPX_RESERVEDPORT:
3564	case PRIV_NETIPX_RAW:
3565
3566		/*
3567		 * NCP privileges.
3568		 */
3569	case PRIV_NETNCP:
3570
3571		/*
3572		 * SMB privileges.
3573		 */
3574	case PRIV_NETSMB:
3575#endif
3576
3577	/*
3578	 * No default: or deny here.
3579	 * In case of no permit fall through to next switch().
3580	 */
3581		if (cred->cr_prison->pr_flags & PR_VNET)
3582			return (0);
3583	}
3584#endif /* VIMAGE */
3585
3586	switch (priv) {
3587
3588		/*
3589		 * Allow ktrace privileges for root in jail.
3590		 */
3591	case PRIV_KTRACE:
3592
3593#if 0
3594		/*
3595		 * Allow jailed processes to configure audit identity and
3596		 * submit audit records (login, etc).  In the future we may
3597		 * want to further refine the relationship between audit and
3598		 * jail.
3599		 */
3600	case PRIV_AUDIT_GETAUDIT:
3601	case PRIV_AUDIT_SETAUDIT:
3602	case PRIV_AUDIT_SUBMIT:
3603#endif
3604
3605		/*
3606		 * Allow jailed processes to manipulate process UNIX
3607		 * credentials in any way they see fit.
3608		 */
3609	case PRIV_CRED_SETUID:
3610	case PRIV_CRED_SETEUID:
3611	case PRIV_CRED_SETGID:
3612	case PRIV_CRED_SETEGID:
3613	case PRIV_CRED_SETGROUPS:
3614	case PRIV_CRED_SETREUID:
3615	case PRIV_CRED_SETREGID:
3616	case PRIV_CRED_SETRESUID:
3617	case PRIV_CRED_SETRESGID:
3618
3619		/*
3620		 * Jail implements visibility constraints already, so allow
3621		 * jailed root to override uid/gid-based constraints.
3622		 */
3623	case PRIV_SEEOTHERGIDS:
3624	case PRIV_SEEOTHERUIDS:
3625
3626		/*
3627		 * Jail implements inter-process debugging limits already, so
3628		 * allow jailed root various debugging privileges.
3629		 */
3630	case PRIV_DEBUG_DIFFCRED:
3631	case PRIV_DEBUG_SUGID:
3632	case PRIV_DEBUG_UNPRIV:
3633
3634		/*
3635		 * Allow jail to set various resource limits and login
3636		 * properties, and for now, exceed process resource limits.
3637		 */
3638	case PRIV_PROC_LIMIT:
3639	case PRIV_PROC_SETLOGIN:
3640	case PRIV_PROC_SETRLIMIT:
3641
3642		/*
3643		 * System V and POSIX IPC privileges are granted in jail.
3644		 */
3645	case PRIV_IPC_READ:
3646	case PRIV_IPC_WRITE:
3647	case PRIV_IPC_ADMIN:
3648	case PRIV_IPC_MSGSIZE:
3649	case PRIV_MQ_ADMIN:
3650
3651		/*
3652		 * Jail operations within a jail work on child jails.
3653		 */
3654	case PRIV_JAIL_ATTACH:
3655	case PRIV_JAIL_SET:
3656	case PRIV_JAIL_REMOVE:
3657
3658		/*
3659		 * Jail implements its own inter-process limits, so allow
3660		 * root processes in jail to change scheduling on other
3661		 * processes in the same jail.  Likewise for signalling.
3662		 */
3663	case PRIV_SCHED_DIFFCRED:
3664	case PRIV_SCHED_CPUSET:
3665	case PRIV_SIGNAL_DIFFCRED:
3666	case PRIV_SIGNAL_SUGID:
3667
3668		/*
3669		 * Allow jailed processes to write to sysctls marked as jail
3670		 * writable.
3671		 */
3672	case PRIV_SYSCTL_WRITEJAIL:
3673
3674		/*
3675		 * Allow root in jail to manage a variety of quota
3676		 * properties.  These should likely be conditional on a
3677		 * configuration option.
3678		 */
3679	case PRIV_VFS_GETQUOTA:
3680	case PRIV_VFS_SETQUOTA:
3681
3682		/*
3683		 * Since Jail relies on chroot() to implement file system
3684		 * protections, grant many VFS privileges to root in jail.
3685		 * Be careful to exclude mount-related and NFS-related
3686		 * privileges.
3687		 */
3688	case PRIV_VFS_READ:
3689	case PRIV_VFS_WRITE:
3690	case PRIV_VFS_ADMIN:
3691	case PRIV_VFS_EXEC:
3692	case PRIV_VFS_LOOKUP:
3693	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3694	case PRIV_VFS_CHFLAGS_DEV:
3695	case PRIV_VFS_CHOWN:
3696	case PRIV_VFS_CHROOT:
3697	case PRIV_VFS_RETAINSUGID:
3698	case PRIV_VFS_FCHROOT:
3699	case PRIV_VFS_LINK:
3700	case PRIV_VFS_SETGID:
3701	case PRIV_VFS_STAT:
3702	case PRIV_VFS_STICKYFILE:
3703		return (0);
3704
3705		/*
3706		 * Depending on the global setting, allow privilege of
3707		 * setting system flags.
3708		 */
3709	case PRIV_VFS_SYSFLAGS:
3710		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3711			return (0);
3712		else
3713			return (EPERM);
3714
3715		/*
3716		 * Depending on the global setting, allow privilege of
3717		 * mounting/unmounting file systems.
3718		 */
3719	case PRIV_VFS_MOUNT:
3720	case PRIV_VFS_UNMOUNT:
3721	case PRIV_VFS_MOUNT_NONUSER:
3722	case PRIV_VFS_MOUNT_OWNER:
3723		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT)
3724			return (0);
3725		else
3726			return (EPERM);
3727
3728		/*
3729		 * Allow jailed root to bind reserved ports and reuse in-use
3730		 * ports.
3731		 */
3732	case PRIV_NETINET_RESERVEDPORT:
3733	case PRIV_NETINET_REUSEPORT:
3734		return (0);
3735
3736		/*
3737		 * Allow jailed root to set certian IPv4/6 (option) headers.
3738		 */
3739	case PRIV_NETINET_SETHDROPTS:
3740		return (0);
3741
3742		/*
3743		 * Conditionally allow creating raw sockets in jail.
3744		 */
3745	case PRIV_NETINET_RAW:
3746		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3747			return (0);
3748		else
3749			return (EPERM);
3750
3751		/*
3752		 * Since jail implements its own visibility limits on netstat
3753		 * sysctls, allow getcred.  This allows identd to work in
3754		 * jail.
3755		 */
3756	case PRIV_NETINET_GETCRED:
3757		return (0);
3758
3759	default:
3760		/*
3761		 * In all remaining cases, deny the privilege request.  This
3762		 * includes almost all network privileges, many system
3763		 * configuration privileges.
3764		 */
3765		return (EPERM);
3766	}
3767}
3768
3769/*
3770 * Return the part of pr2's name that is relative to pr1, or the whole name
3771 * if it does not directly follow.
3772 */
3773
3774char *
3775prison_name(struct prison *pr1, struct prison *pr2)
3776{
3777	char *name;
3778
3779	/* Jails see themselves as "0" (if they see themselves at all). */
3780	if (pr1 == pr2)
3781		return "0";
3782	name = pr2->pr_name;
3783	if (prison_ischild(pr1, pr2)) {
3784		/*
3785		 * pr1 isn't locked (and allprison_lock may not be either)
3786		 * so its length can't be counted on.  But the number of dots
3787		 * can be counted on - and counted.
3788		 */
3789		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3790			name = strchr(name, '.') + 1;
3791	}
3792	return (name);
3793}
3794
3795/*
3796 * Return the part of pr2's path that is relative to pr1, or the whole path
3797 * if it does not directly follow.
3798 */
3799static char *
3800prison_path(struct prison *pr1, struct prison *pr2)
3801{
3802	char *path1, *path2;
3803	int len1;
3804
3805	path1 = pr1->pr_path;
3806	path2 = pr2->pr_path;
3807	if (!strcmp(path1, "/"))
3808		return (path2);
3809	len1 = strlen(path1);
3810	if (strncmp(path1, path2, len1))
3811		return (path2);
3812	if (path2[len1] == '\0')
3813		return "/";
3814	if (path2[len1] == '/')
3815		return (path2 + len1);
3816	return (path2);
3817}
3818
3819
3820/*
3821 * Jail-related sysctls.
3822 */
3823SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3824    "Jails");
3825
3826static int
3827sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3828{
3829	struct xprison *xp;
3830	struct prison *pr, *cpr;
3831#ifdef INET
3832	struct in_addr *ip4 = NULL;
3833	int ip4s = 0;
3834#endif
3835#ifdef INET6
3836	struct in_addr *ip6 = NULL;
3837	int ip6s = 0;
3838#endif
3839	int descend, error;
3840
3841	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3842	pr = req->td->td_ucred->cr_prison;
3843	error = 0;
3844	sx_slock(&allprison_lock);
3845	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3846#if defined(INET) || defined(INET6)
3847 again:
3848#endif
3849		mtx_lock(&cpr->pr_mtx);
3850#ifdef INET
3851		if (cpr->pr_ip4s > 0) {
3852			if (ip4s < cpr->pr_ip4s) {
3853				ip4s = cpr->pr_ip4s;
3854				mtx_unlock(&cpr->pr_mtx);
3855				ip4 = realloc(ip4, ip4s *
3856				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3857				goto again;
3858			}
3859			bcopy(cpr->pr_ip4, ip4,
3860			    cpr->pr_ip4s * sizeof(struct in_addr));
3861		}
3862#endif
3863#ifdef INET6
3864		if (cpr->pr_ip6s > 0) {
3865			if (ip6s < cpr->pr_ip6s) {
3866				ip6s = cpr->pr_ip6s;
3867				mtx_unlock(&cpr->pr_mtx);
3868				ip6 = realloc(ip6, ip6s *
3869				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3870				goto again;
3871			}
3872			bcopy(cpr->pr_ip6, ip6,
3873			    cpr->pr_ip6s * sizeof(struct in6_addr));
3874		}
3875#endif
3876		if (cpr->pr_ref == 0) {
3877			mtx_unlock(&cpr->pr_mtx);
3878			continue;
3879		}
3880		bzero(xp, sizeof(*xp));
3881		xp->pr_version = XPRISON_VERSION;
3882		xp->pr_id = cpr->pr_id;
3883		xp->pr_state = cpr->pr_uref > 0
3884		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3885		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3886		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3887		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3888#ifdef INET
3889		xp->pr_ip4s = cpr->pr_ip4s;
3890#endif
3891#ifdef INET6
3892		xp->pr_ip6s = cpr->pr_ip6s;
3893#endif
3894		mtx_unlock(&cpr->pr_mtx);
3895		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3896		if (error)
3897			break;
3898#ifdef INET
3899		if (xp->pr_ip4s > 0) {
3900			error = SYSCTL_OUT(req, ip4,
3901			    xp->pr_ip4s * sizeof(struct in_addr));
3902			if (error)
3903				break;
3904		}
3905#endif
3906#ifdef INET6
3907		if (xp->pr_ip6s > 0) {
3908			error = SYSCTL_OUT(req, ip6,
3909			    xp->pr_ip6s * sizeof(struct in6_addr));
3910			if (error)
3911				break;
3912		}
3913#endif
3914	}
3915	sx_sunlock(&allprison_lock);
3916	free(xp, M_TEMP);
3917#ifdef INET
3918	free(ip4, M_TEMP);
3919#endif
3920#ifdef INET6
3921	free(ip6, M_TEMP);
3922#endif
3923	return (error);
3924}
3925
3926SYSCTL_OID(_security_jail, OID_AUTO, list,
3927    CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3928    sysctl_jail_list, "S", "List of active jails");
3929
3930static int
3931sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3932{
3933	int error, injail;
3934
3935	injail = jailed(req->td->td_ucred);
3936	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3937
3938	return (error);
3939}
3940
3941SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3942    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3943    sysctl_jail_jailed, "I", "Process in jail?");
3944
3945#if defined(INET) || defined(INET6)
3946SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3947    &jail_max_af_ips, 0,
3948    "Number of IP addresses a jail may have at most per address family");
3949#endif
3950
3951/*
3952 * Default parameters for jail(2) compatability.  For historical reasons,
3953 * the sysctl names have varying similarity to the parameter names.  Prisons
3954 * just see their own parameters, and can't change them.
3955 */
3956static int
3957sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3958{
3959	struct prison *pr;
3960	int allow, error, i;
3961
3962	pr = req->td->td_ucred->cr_prison;
3963	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3964
3965	/* Get the current flag value, and convert it to a boolean. */
3966	i = (allow & arg2) ? 1 : 0;
3967	if (arg1 != NULL)
3968		i = !i;
3969	error = sysctl_handle_int(oidp, &i, 0, req);
3970	if (error || !req->newptr)
3971		return (error);
3972	i = i ? arg2 : 0;
3973	if (arg1 != NULL)
3974		i ^= arg2;
3975	/*
3976	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3977	 * for writing.
3978	 */
3979	mtx_lock(&prison0.pr_mtx);
3980	jail_default_allow = (jail_default_allow & ~arg2) | i;
3981	mtx_unlock(&prison0.pr_mtx);
3982	return (0);
3983}
3984
3985SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3986    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3987    NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3988    "Processes in jail can set their hostnames");
3989SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3990    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3991    (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3992    "Processes in jail are limited to creating UNIX/IP/route sockets only");
3993SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3994    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3995    NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3996    "Processes in jail can use System V IPC primitives");
3997SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3998    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3999    NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4000    "Prison root can create raw sockets");
4001SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4002    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4003    NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4004    "Processes in jail can alter system file flags");
4005SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4006    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4007    NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4008    "Processes in jail can mount/unmount jail-friendly file systems");
4009
4010static int
4011sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4012{
4013	struct prison *pr;
4014	int level, error;
4015
4016	pr = req->td->td_ucred->cr_prison;
4017	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4018	error = sysctl_handle_int(oidp, &level, 0, req);
4019	if (error || !req->newptr)
4020		return (error);
4021	*(int *)arg1 = level;
4022	return (0);
4023}
4024
4025SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4026    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4027    &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4028    sysctl_jail_default_level, "I",
4029    "Processes in jail cannot see all mounted file systems");
4030
4031/*
4032 * Nodes to describe jail parameters.  Maximum length of string parameters
4033 * is returned in the string itself, and the other parameters exist merely
4034 * to make themselves and their types known.
4035 */
4036SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4037    "Jail parameters");
4038
4039int
4040sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4041{
4042	int i;
4043	long l;
4044	size_t s;
4045	char numbuf[12];
4046
4047	switch (oidp->oid_kind & CTLTYPE)
4048	{
4049	case CTLTYPE_LONG:
4050	case CTLTYPE_ULONG:
4051		l = 0;
4052#ifdef SCTL_MASK32
4053		if (!(req->flags & SCTL_MASK32))
4054#endif
4055			return (SYSCTL_OUT(req, &l, sizeof(l)));
4056	case CTLTYPE_INT:
4057	case CTLTYPE_UINT:
4058		i = 0;
4059		return (SYSCTL_OUT(req, &i, sizeof(i)));
4060	case CTLTYPE_STRING:
4061		snprintf(numbuf, sizeof(numbuf), "%d", arg2);
4062		return
4063		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4064	case CTLTYPE_STRUCT:
4065		s = (size_t)arg2;
4066		return (SYSCTL_OUT(req, &s, sizeof(s)));
4067	}
4068	return (0);
4069}
4070
4071SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4072SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4073SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4074SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4075SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4076    "I", "Jail secure level");
4077SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4078    "I", "Jail cannot see all mounted file systems");
4079SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4080    "B", "Jail persistence");
4081#ifdef VIMAGE
4082SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4083    "E,jailsys", "Virtual network stack");
4084#endif
4085SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4086    "B", "Jail is in the process of shutting down");
4087
4088SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4089SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4090    "I", "Current number of child jails");
4091SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4092    "I", "Maximum number of child jails");
4093
4094SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4095SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4096    "Jail hostname");
4097SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4098    "Jail NIS domainname");
4099SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4100    "Jail host UUID");
4101SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4102    "LU", "Jail host ID");
4103
4104SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4105SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4106
4107#ifdef INET
4108SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RW, "Jail IPv4 address virtualization");
4109SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4110    "S,in_addr,a", "Jail IPv4 addresses");
4111#endif
4112#ifdef INET6
4113SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RW, "Jail IPv6 address virtualization");
4114SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4115    "S,in6_addr,a", "Jail IPv6 addresses");
4116#endif
4117
4118SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4119SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4120    "B", "Jail may set hostname");
4121SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4122    "B", "Jail may use SYSV IPC");
4123SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4124    "B", "Jail may create raw sockets");
4125SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4126    "B", "Jail may alter system file flags");
4127SYSCTL_JAIL_PARAM(_allow, mount, CTLTYPE_INT | CTLFLAG_RW,
4128    "B", "Jail may mount/unmount jail-friendly file systems");
4129SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4130    "B", "Jail may set file quotas");
4131SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4132    "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4133
4134
4135#ifdef DDB
4136
4137static void
4138db_show_prison(struct prison *pr)
4139{
4140	int fi;
4141#if defined(INET) || defined(INET6)
4142	int ii;
4143#endif
4144	unsigned jsf;
4145#ifdef INET6
4146	char ip6buf[INET6_ADDRSTRLEN];
4147#endif
4148
4149	db_printf("prison %p:\n", pr);
4150	db_printf(" jid             = %d\n", pr->pr_id);
4151	db_printf(" name            = %s\n", pr->pr_name);
4152	db_printf(" parent          = %p\n", pr->pr_parent);
4153	db_printf(" ref             = %d\n", pr->pr_ref);
4154	db_printf(" uref            = %d\n", pr->pr_uref);
4155	db_printf(" path            = %s\n", pr->pr_path);
4156	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4157	    ? pr->pr_cpuset->cs_id : -1);
4158#ifdef VIMAGE
4159	db_printf(" vnet            = %p\n", pr->pr_vnet);
4160#endif
4161	db_printf(" root            = %p\n", pr->pr_root);
4162	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4163	db_printf(" childcount      = %d\n", pr->pr_childcount);
4164	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4165	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4166	db_printf(" flags           = %x", pr->pr_flags);
4167	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4168	    fi++)
4169		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4170			db_printf(" %s", pr_flag_names[fi]);
4171	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4172	    fi++) {
4173		jsf = pr->pr_flags &
4174		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4175		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4176		    pr_flag_jailsys[fi].disable &&
4177		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4178		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4179		    : "inherit");
4180	}
4181	db_printf(" allow           = %x", pr->pr_allow);
4182	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4183	    fi++)
4184		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4185			db_printf(" %s", pr_allow_names[fi]);
4186	db_printf("\n");
4187	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4188	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4189	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4190	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4191	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4192#ifdef INET
4193	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4194	for (ii = 0; ii < pr->pr_ip4s; ii++)
4195		db_printf(" %s %s\n",
4196		    ii == 0 ? "ip4             =" : "                 ",
4197		    inet_ntoa(pr->pr_ip4[ii]));
4198#endif
4199#ifdef INET6
4200	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4201	for (ii = 0; ii < pr->pr_ip6s; ii++)
4202		db_printf(" %s %s\n",
4203		    ii == 0 ? "ip6             =" : "                 ",
4204		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4205#endif
4206}
4207
4208DB_SHOW_COMMAND(prison, db_show_prison_command)
4209{
4210	struct prison *pr;
4211
4212	if (!have_addr) {
4213		/*
4214		 * Show all prisons in the list, and prison0 which is not
4215		 * listed.
4216		 */
4217		db_show_prison(&prison0);
4218		if (!db_pager_quit) {
4219			TAILQ_FOREACH(pr, &allprison, pr_list) {
4220				db_show_prison(pr);
4221				if (db_pager_quit)
4222					break;
4223			}
4224		}
4225		return;
4226	}
4227
4228	if (addr == 0)
4229		pr = &prison0;
4230	else {
4231		/* Look for a prison with the ID and with references. */
4232		TAILQ_FOREACH(pr, &allprison, pr_list)
4233			if (pr->pr_id == addr && pr->pr_ref > 0)
4234				break;
4235		if (pr == NULL)
4236			/* Look again, without requiring a reference. */
4237			TAILQ_FOREACH(pr, &allprison, pr_list)
4238				if (pr->pr_id == addr)
4239					break;
4240		if (pr == NULL)
4241			/* Assume address points to a valid prison. */
4242			pr = (struct prison *)addr;
4243	}
4244	db_show_prison(pr);
4245}
4246
4247#endif /* DDB */
4248