xref: /freebsd/sys/kern/kern_jail.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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("$FreeBSD$");
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/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #ifdef INET6
72 #include <netinet6/in6_var.h>
73 #endif /* INET6 */
74 #endif /* DDB */
75 
76 #include <security/mac/mac_framework.h>
77 
78 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
79 
80 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
81 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
82 
83 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
84 #ifdef INET
85 #ifdef INET6
86 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
87 #else
88 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
89 #endif
90 #else /* !INET */
91 #ifdef INET6
92 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
93 #else
94 #define	_PR_IP_SADDRSEL	0
95 #endif
96 #endif
97 
98 /* prison0 describes what is "real" about the system. */
99 struct prison prison0 = {
100 	.pr_id		= 0,
101 	.pr_name	= "0",
102 	.pr_ref		= 1,
103 	.pr_uref	= 1,
104 	.pr_path	= "/",
105 	.pr_securelevel	= -1,
106 	.pr_devfs_rsnum = 0,
107 	.pr_childmax	= JAIL_MAX,
108 	.pr_hostuuid	= DEFAULT_HOSTUUID,
109 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
110 #ifdef VIMAGE
111 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
112 #else
113 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
114 #endif
115 	.pr_allow	= PR_ALLOW_ALL,
116 };
117 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
118 
119 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
120 struct	sx allprison_lock;
121 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
122 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
123 LIST_HEAD(, prison_racct) allprison_racct;
124 int	lastprid = 0;
125 
126 static int do_jail_attach(struct thread *td, struct prison *pr);
127 static void prison_complete(void *context, int pending);
128 static void prison_deref(struct prison *pr, int flags);
129 static char *prison_path(struct prison *pr1, struct prison *pr2);
130 static void prison_remove_one(struct prison *pr);
131 #ifdef RACCT
132 static void prison_racct_attach(struct prison *pr);
133 static void prison_racct_modify(struct prison *pr);
134 static void prison_racct_detach(struct prison *pr);
135 #endif
136 #ifdef INET
137 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
138 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
139 #endif
140 #ifdef INET6
141 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
142 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
143 #endif
144 
145 /* Flags for prison_deref */
146 #define	PD_DEREF	0x01
147 #define	PD_DEUREF	0x02
148 #define	PD_LOCKED	0x04
149 #define	PD_LIST_SLOCKED	0x08
150 #define	PD_LIST_XLOCKED	0x10
151 
152 /*
153  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
154  * as we cannot figure out the size of a sparse array, or an array without a
155  * terminating entry.
156  */
157 static char *pr_flag_names[] = {
158 	[0] = "persist",
159 #ifdef INET
160 	[7] = "ip4.saddrsel",
161 #endif
162 #ifdef INET6
163 	[8] = "ip6.saddrsel",
164 #endif
165 };
166 const size_t pr_flag_names_size = sizeof(pr_flag_names);
167 
168 static char *pr_flag_nonames[] = {
169 	[0] = "nopersist",
170 #ifdef INET
171 	[7] = "ip4.nosaddrsel",
172 #endif
173 #ifdef INET6
174 	[8] = "ip6.nosaddrsel",
175 #endif
176 };
177 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
178 
179 struct jailsys_flags {
180 	const char	*name;
181 	unsigned	 disable;
182 	unsigned	 new;
183 } pr_flag_jailsys[] = {
184 	{ "host", 0, PR_HOST },
185 #ifdef VIMAGE
186 	{ "vnet", 0, PR_VNET },
187 #endif
188 #ifdef INET
189 	{ "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
190 #endif
191 #ifdef INET6
192 	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
193 #endif
194 };
195 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
196 
197 static char *pr_allow_names[] = {
198 	"allow.set_hostname",
199 	"allow.sysvipc",
200 	"allow.raw_sockets",
201 	"allow.chflags",
202 	"allow.mount",
203 	"allow.quotas",
204 	"allow.socket_af",
205 	"allow.mount.devfs",
206 	"allow.mount.nullfs",
207 	"allow.mount.zfs",
208 	"allow.mount.procfs",
209 };
210 const size_t pr_allow_names_size = sizeof(pr_allow_names);
211 
212 static char *pr_allow_nonames[] = {
213 	"allow.noset_hostname",
214 	"allow.nosysvipc",
215 	"allow.noraw_sockets",
216 	"allow.nochflags",
217 	"allow.nomount",
218 	"allow.noquotas",
219 	"allow.nosocket_af",
220 	"allow.mount.nodevfs",
221 	"allow.mount.nonullfs",
222 	"allow.mount.nozfs",
223 	"allow.mount.noprocfs",
224 };
225 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
226 
227 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
228 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
229 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
230 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
231 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
232 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
233 #if defined(INET) || defined(INET6)
234 static unsigned jail_max_af_ips = 255;
235 #endif
236 
237 #ifdef INET
238 static int
239 qcmp_v4(const void *ip1, const void *ip2)
240 {
241 	in_addr_t iaa, iab;
242 
243 	/*
244 	 * We need to compare in HBO here to get the list sorted as expected
245 	 * by the result of the code.  Sorting NBO addresses gives you
246 	 * interesting results.  If you do not understand, do not try.
247 	 */
248 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
249 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
250 
251 	/*
252 	 * Do not simply return the difference of the two numbers, the int is
253 	 * not wide enough.
254 	 */
255 	if (iaa > iab)
256 		return (1);
257 	else if (iaa < iab)
258 		return (-1);
259 	else
260 		return (0);
261 }
262 #endif
263 
264 #ifdef INET6
265 static int
266 qcmp_v6(const void *ip1, const void *ip2)
267 {
268 	const struct in6_addr *ia6a, *ia6b;
269 	int i, rc;
270 
271 	ia6a = (const struct in6_addr *)ip1;
272 	ia6b = (const struct in6_addr *)ip2;
273 
274 	rc = 0;
275 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
276 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
277 			rc = 1;
278 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
279 			rc = -1;
280 	}
281 	return (rc);
282 }
283 #endif
284 
285 /*
286  * struct jail_args {
287  *	struct jail *jail;
288  * };
289  */
290 int
291 sys_jail(struct thread *td, struct jail_args *uap)
292 {
293 	uint32_t version;
294 	int error;
295 	struct jail j;
296 
297 	error = copyin(uap->jail, &version, sizeof(uint32_t));
298 	if (error)
299 		return (error);
300 
301 	switch (version) {
302 	case 0:
303 	{
304 		struct jail_v0 j0;
305 
306 		/* FreeBSD single IPv4 jails. */
307 		bzero(&j, sizeof(struct jail));
308 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
309 		if (error)
310 			return (error);
311 		j.version = j0.version;
312 		j.path = j0.path;
313 		j.hostname = j0.hostname;
314 		j.ip4s = j0.ip_number;
315 		break;
316 	}
317 
318 	case 1:
319 		/*
320 		 * Version 1 was used by multi-IPv4 jail implementations
321 		 * that never made it into the official kernel.
322 		 */
323 		return (EINVAL);
324 
325 	case 2:	/* JAIL_API_VERSION */
326 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
327 		error = copyin(uap->jail, &j, sizeof(struct jail));
328 		if (error)
329 			return (error);
330 		break;
331 
332 	default:
333 		/* Sci-Fi jails are not supported, sorry. */
334 		return (EINVAL);
335 	}
336 	return (kern_jail(td, &j));
337 }
338 
339 int
340 kern_jail(struct thread *td, struct jail *j)
341 {
342 	struct iovec optiov[2 * (4
343 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
344 #ifdef INET
345 			    + 1
346 #endif
347 #ifdef INET6
348 			    + 1
349 #endif
350 			    )];
351 	struct uio opt;
352 	char *u_path, *u_hostname, *u_name;
353 #ifdef INET
354 	uint32_t ip4s;
355 	struct in_addr *u_ip4;
356 #endif
357 #ifdef INET6
358 	struct in6_addr *u_ip6;
359 #endif
360 	size_t tmplen;
361 	int error, enforce_statfs, fi;
362 
363 	bzero(&optiov, sizeof(optiov));
364 	opt.uio_iov = optiov;
365 	opt.uio_iovcnt = 0;
366 	opt.uio_offset = -1;
367 	opt.uio_resid = -1;
368 	opt.uio_segflg = UIO_SYSSPACE;
369 	opt.uio_rw = UIO_READ;
370 	opt.uio_td = td;
371 
372 	/* Set permissions for top-level jails from sysctls. */
373 	if (!jailed(td->td_ucred)) {
374 		for (fi = 0; fi < sizeof(pr_allow_names) /
375 		     sizeof(pr_allow_names[0]); fi++) {
376 			optiov[opt.uio_iovcnt].iov_base =
377 			    (jail_default_allow & (1 << fi))
378 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
379 			optiov[opt.uio_iovcnt].iov_len =
380 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
381 			opt.uio_iovcnt += 2;
382 		}
383 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
384 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
385 		opt.uio_iovcnt++;
386 		enforce_statfs = jail_default_enforce_statfs;
387 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
388 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
389 		opt.uio_iovcnt++;
390 	}
391 
392 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
393 #ifdef INET
394 	ip4s = (j->version == 0) ? 1 : j->ip4s;
395 	if (ip4s > jail_max_af_ips)
396 		return (EINVAL);
397 	tmplen += ip4s * sizeof(struct in_addr);
398 #else
399 	if (j->ip4s > 0)
400 		return (EINVAL);
401 #endif
402 #ifdef INET6
403 	if (j->ip6s > jail_max_af_ips)
404 		return (EINVAL);
405 	tmplen += j->ip6s * sizeof(struct in6_addr);
406 #else
407 	if (j->ip6s > 0)
408 		return (EINVAL);
409 #endif
410 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
411 	u_hostname = u_path + MAXPATHLEN;
412 	u_name = u_hostname + MAXHOSTNAMELEN;
413 #ifdef INET
414 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
415 #endif
416 #ifdef INET6
417 #ifdef INET
418 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
419 #else
420 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
421 #endif
422 #endif
423 	optiov[opt.uio_iovcnt].iov_base = "path";
424 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
425 	opt.uio_iovcnt++;
426 	optiov[opt.uio_iovcnt].iov_base = u_path;
427 	error = copyinstr(j->path, u_path, MAXPATHLEN,
428 	    &optiov[opt.uio_iovcnt].iov_len);
429 	if (error) {
430 		free(u_path, M_TEMP);
431 		return (error);
432 	}
433 	opt.uio_iovcnt++;
434 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
435 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
436 	opt.uio_iovcnt++;
437 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
438 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
439 	    &optiov[opt.uio_iovcnt].iov_len);
440 	if (error) {
441 		free(u_path, M_TEMP);
442 		return (error);
443 	}
444 	opt.uio_iovcnt++;
445 	if (j->jailname != NULL) {
446 		optiov[opt.uio_iovcnt].iov_base = "name";
447 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
448 		opt.uio_iovcnt++;
449 		optiov[opt.uio_iovcnt].iov_base = u_name;
450 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
451 		    &optiov[opt.uio_iovcnt].iov_len);
452 		if (error) {
453 			free(u_path, M_TEMP);
454 			return (error);
455 		}
456 		opt.uio_iovcnt++;
457 	}
458 #ifdef INET
459 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
460 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
461 	opt.uio_iovcnt++;
462 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
463 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
464 	if (j->version == 0)
465 		u_ip4->s_addr = j->ip4s;
466 	else {
467 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
468 		if (error) {
469 			free(u_path, M_TEMP);
470 			return (error);
471 		}
472 	}
473 	opt.uio_iovcnt++;
474 #endif
475 #ifdef INET6
476 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
477 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
478 	opt.uio_iovcnt++;
479 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
480 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
481 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
482 	if (error) {
483 		free(u_path, M_TEMP);
484 		return (error);
485 	}
486 	opt.uio_iovcnt++;
487 #endif
488 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
489 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
490 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
491 	free(u_path, M_TEMP);
492 	return (error);
493 }
494 
495 
496 /*
497  * struct jail_set_args {
498  *	struct iovec *iovp;
499  *	unsigned int iovcnt;
500  *	int flags;
501  * };
502  */
503 int
504 sys_jail_set(struct thread *td, struct jail_set_args *uap)
505 {
506 	struct uio *auio;
507 	int error;
508 
509 	/* Check that we have an even number of iovecs. */
510 	if (uap->iovcnt & 1)
511 		return (EINVAL);
512 
513 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
514 	if (error)
515 		return (error);
516 	error = kern_jail_set(td, auio, uap->flags);
517 	free(auio, M_IOV);
518 	return (error);
519 }
520 
521 int
522 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
523 {
524 	struct nameidata nd;
525 #ifdef INET
526 	struct in_addr *ip4;
527 #endif
528 #ifdef INET6
529 	struct in6_addr *ip6;
530 #endif
531 	struct vfsopt *opt;
532 	struct vfsoptlist *opts;
533 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
534 	struct vnode *root;
535 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
536 	char *g_path;
537 #if defined(INET) || defined(INET6)
538 	struct prison *tppr;
539 	void *op;
540 #endif
541 	unsigned long hid;
542 	size_t namelen, onamelen;
543 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
544 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
545 	int fi, jid, jsys, len, level;
546 	int childmax, rsnum, slevel;
547 	int fullpath_disabled;
548 #if defined(INET) || defined(INET6)
549 	int ii, ij;
550 #endif
551 #ifdef INET
552 	int ip4s, redo_ip4;
553 #endif
554 #ifdef INET6
555 	int ip6s, redo_ip6;
556 #endif
557 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
558 	unsigned tallow;
559 	char numbuf[12];
560 
561 	error = priv_check(td, PRIV_JAIL_SET);
562 	if (!error && (flags & JAIL_ATTACH))
563 		error = priv_check(td, PRIV_JAIL_ATTACH);
564 	if (error)
565 		return (error);
566 	mypr = ppr = td->td_ucred->cr_prison;
567 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
568 		return (EPERM);
569 	if (flags & ~JAIL_SET_MASK)
570 		return (EINVAL);
571 
572 	/*
573 	 * Check all the parameters before committing to anything.  Not all
574 	 * errors can be caught early, but we may as well try.  Also, this
575 	 * takes care of some expensive stuff (path lookup) before getting
576 	 * the allprison lock.
577 	 *
578 	 * XXX Jails are not filesystems, and jail parameters are not mount
579 	 *     options.  But it makes more sense to re-use the vfsopt code
580 	 *     than duplicate it under a different name.
581 	 */
582 	error = vfs_buildopts(optuio, &opts);
583 	if (error)
584 		return (error);
585 #ifdef INET
586 	ip4 = NULL;
587 #endif
588 #ifdef INET6
589 	ip6 = NULL;
590 #endif
591 	g_path = NULL;
592 
593 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
594 	if (error == ENOENT)
595 		jid = 0;
596 	else if (error != 0)
597 		goto done_free;
598 
599 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
600 	if (error == ENOENT)
601 		gotslevel = 0;
602 	else if (error != 0)
603 		goto done_free;
604 	else
605 		gotslevel = 1;
606 
607 	error =
608 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
609 	if (error == ENOENT)
610 		gotchildmax = 0;
611 	else if (error != 0)
612 		goto done_free;
613 	else
614 		gotchildmax = 1;
615 
616 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
617 	if (error == ENOENT)
618 		gotenforce = 0;
619 	else if (error != 0)
620 		goto done_free;
621 	else if (enforce < 0 || enforce > 2) {
622 		error = EINVAL;
623 		goto done_free;
624 	} else
625 		gotenforce = 1;
626 
627 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
628 	if (error == ENOENT)
629 		gotrsnum = 0;
630 	else if (error != 0)
631 		goto done_free;
632 	else
633 		gotrsnum = 1;
634 
635 	pr_flags = ch_flags = 0;
636 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
637 	    fi++) {
638 		if (pr_flag_names[fi] == NULL)
639 			continue;
640 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
641 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
642 	}
643 	ch_flags |= pr_flags;
644 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
645 	    fi++) {
646 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
647 		    sizeof(jsys));
648 		if (error == ENOENT)
649 			continue;
650 		if (error != 0)
651 			goto done_free;
652 		switch (jsys) {
653 		case JAIL_SYS_DISABLE:
654 			if (!pr_flag_jailsys[fi].disable) {
655 				error = EINVAL;
656 				goto done_free;
657 			}
658 			pr_flags |= pr_flag_jailsys[fi].disable;
659 			break;
660 		case JAIL_SYS_NEW:
661 			pr_flags |= pr_flag_jailsys[fi].new;
662 			break;
663 		case JAIL_SYS_INHERIT:
664 			break;
665 		default:
666 			error = EINVAL;
667 			goto done_free;
668 		}
669 		ch_flags |=
670 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
671 	}
672 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
673 	    && !(pr_flags & PR_PERSIST)) {
674 		error = EINVAL;
675 		vfs_opterror(opts, "new jail must persist or attach");
676 		goto done_errmsg;
677 	}
678 #ifdef VIMAGE
679 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
680 		error = EINVAL;
681 		vfs_opterror(opts, "vnet cannot be changed after creation");
682 		goto done_errmsg;
683 	}
684 #endif
685 #ifdef INET
686 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
687 		error = EINVAL;
688 		vfs_opterror(opts, "ip4 cannot be changed after creation");
689 		goto done_errmsg;
690 	}
691 #endif
692 #ifdef INET6
693 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
694 		error = EINVAL;
695 		vfs_opterror(opts, "ip6 cannot be changed after creation");
696 		goto done_errmsg;
697 	}
698 #endif
699 
700 	pr_allow = ch_allow = 0;
701 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
702 	    fi++) {
703 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
704 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
705 	}
706 	ch_allow |= pr_allow;
707 
708 	error = vfs_getopt(opts, "name", (void **)&name, &len);
709 	if (error == ENOENT)
710 		name = NULL;
711 	else if (error != 0)
712 		goto done_free;
713 	else {
714 		if (len == 0 || name[len - 1] != '\0') {
715 			error = EINVAL;
716 			goto done_free;
717 		}
718 		if (len > MAXHOSTNAMELEN) {
719 			error = ENAMETOOLONG;
720 			goto done_free;
721 		}
722 	}
723 
724 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
725 	if (error == ENOENT)
726 		host = NULL;
727 	else if (error != 0)
728 		goto done_free;
729 	else {
730 		ch_flags |= PR_HOST;
731 		pr_flags |= PR_HOST;
732 		if (len == 0 || host[len - 1] != '\0') {
733 			error = EINVAL;
734 			goto done_free;
735 		}
736 		if (len > MAXHOSTNAMELEN) {
737 			error = ENAMETOOLONG;
738 			goto done_free;
739 		}
740 	}
741 
742 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
743 	if (error == ENOENT)
744 		domain = NULL;
745 	else if (error != 0)
746 		goto done_free;
747 	else {
748 		ch_flags |= PR_HOST;
749 		pr_flags |= PR_HOST;
750 		if (len == 0 || domain[len - 1] != '\0') {
751 			error = EINVAL;
752 			goto done_free;
753 		}
754 		if (len > MAXHOSTNAMELEN) {
755 			error = ENAMETOOLONG;
756 			goto done_free;
757 		}
758 	}
759 
760 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
761 	if (error == ENOENT)
762 		uuid = NULL;
763 	else if (error != 0)
764 		goto done_free;
765 	else {
766 		ch_flags |= PR_HOST;
767 		pr_flags |= PR_HOST;
768 		if (len == 0 || uuid[len - 1] != '\0') {
769 			error = EINVAL;
770 			goto done_free;
771 		}
772 		if (len > HOSTUUIDLEN) {
773 			error = ENAMETOOLONG;
774 			goto done_free;
775 		}
776 	}
777 
778 #ifdef COMPAT_FREEBSD32
779 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
780 		uint32_t hid32;
781 
782 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
783 		hid = hid32;
784 	} else
785 #endif
786 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
787 	if (error == ENOENT)
788 		gothid = 0;
789 	else if (error != 0)
790 		goto done_free;
791 	else {
792 		gothid = 1;
793 		ch_flags |= PR_HOST;
794 		pr_flags |= PR_HOST;
795 	}
796 
797 #ifdef INET
798 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
799 	if (error == ENOENT)
800 		ip4s = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
801 	else if (error != 0)
802 		goto done_free;
803 	else if (ip4s & (sizeof(*ip4) - 1)) {
804 		error = EINVAL;
805 		goto done_free;
806 	} else {
807 		ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
808 		if (ip4s == 0)
809 			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
810 		else {
811 			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
812 			ip4s /= sizeof(*ip4);
813 			if (ip4s > jail_max_af_ips) {
814 				error = EINVAL;
815 				vfs_opterror(opts, "too many IPv4 addresses");
816 				goto done_errmsg;
817 			}
818 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
819 			bcopy(op, ip4, ip4s * sizeof(*ip4));
820 			/*
821 			 * IP addresses are all sorted but ip[0] to preserve
822 			 * the primary IP address as given from userland.
823 			 * This special IP is used for unbound outgoing
824 			 * connections as well for "loopback" traffic in case
825 			 * source address selection cannot find any more fitting
826 			 * address to connect from.
827 			 */
828 			if (ip4s > 1)
829 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
830 			/*
831 			 * Check for duplicate addresses and do some simple
832 			 * zero and broadcast checks. If users give other bogus
833 			 * addresses it is their problem.
834 			 *
835 			 * We do not have to care about byte order for these
836 			 * checks so we will do them in NBO.
837 			 */
838 			for (ii = 0; ii < ip4s; ii++) {
839 				if (ip4[ii].s_addr == INADDR_ANY ||
840 				    ip4[ii].s_addr == INADDR_BROADCAST) {
841 					error = EINVAL;
842 					goto done_free;
843 				}
844 				if ((ii+1) < ip4s &&
845 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
846 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
847 					error = EINVAL;
848 					goto done_free;
849 				}
850 			}
851 		}
852 	}
853 #endif
854 
855 #ifdef INET6
856 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
857 	if (error == ENOENT)
858 		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
859 	else if (error != 0)
860 		goto done_free;
861 	else if (ip6s & (sizeof(*ip6) - 1)) {
862 		error = EINVAL;
863 		goto done_free;
864 	} else {
865 		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
866 		if (ip6s == 0)
867 			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
868 		else {
869 			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
870 			ip6s /= sizeof(*ip6);
871 			if (ip6s > jail_max_af_ips) {
872 				error = EINVAL;
873 				vfs_opterror(opts, "too many IPv6 addresses");
874 				goto done_errmsg;
875 			}
876 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
877 			bcopy(op, ip6, ip6s * sizeof(*ip6));
878 			if (ip6s > 1)
879 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
880 			for (ii = 0; ii < ip6s; ii++) {
881 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
882 					error = EINVAL;
883 					goto done_free;
884 				}
885 				if ((ii+1) < ip6s &&
886 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
887 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
888 				{
889 					error = EINVAL;
890 					goto done_free;
891 				}
892 			}
893 		}
894 	}
895 #endif
896 
897 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
898 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
899 		error = EINVAL;
900 		vfs_opterror(opts,
901 		    "vnet jails cannot have IP address restrictions");
902 		goto done_errmsg;
903 	}
904 #endif
905 
906 	fullpath_disabled = 0;
907 	root = NULL;
908 	error = vfs_getopt(opts, "path", (void **)&path, &len);
909 	if (error == ENOENT)
910 		path = NULL;
911 	else if (error != 0)
912 		goto done_free;
913 	else {
914 		if (flags & JAIL_UPDATE) {
915 			error = EINVAL;
916 			vfs_opterror(opts,
917 			    "path cannot be changed after creation");
918 			goto done_errmsg;
919 		}
920 		if (len == 0 || path[len - 1] != '\0') {
921 			error = EINVAL;
922 			goto done_free;
923 		}
924 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
925 		    path, td);
926 		error = namei(&nd);
927 		if (error)
928 			goto done_free;
929 		root = nd.ni_vp;
930 		NDFREE(&nd, NDF_ONLY_PNBUF);
931 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
932 		strlcpy(g_path, path, MAXPATHLEN);
933 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
934 		if (error == 0)
935 			path = g_path;
936 		else if (error == ENODEV) {
937 			/* proceed if sysctl debug.disablefullpath == 1 */
938 			fullpath_disabled = 1;
939 			if (len < 2 || (len == 2 && path[0] == '/'))
940 				path = NULL;
941 		} else {
942 			/* exit on other errors */
943 			goto done_free;
944 		}
945 		if (root->v_type != VDIR) {
946 			error = ENOTDIR;
947 			vput(root);
948 			goto done_free;
949 		}
950 		VOP_UNLOCK(root, 0);
951 		if (fullpath_disabled) {
952 			/* Leave room for a real-root full pathname. */
953 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
954 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
955 				error = ENAMETOOLONG;
956 				goto done_free;
957 			}
958 		}
959 	}
960 
961 	/*
962 	 * Grab the allprison lock before letting modules check their
963 	 * parameters.  Once we have it, do not let go so we'll have a
964 	 * consistent view of the OSD list.
965 	 */
966 	sx_xlock(&allprison_lock);
967 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
968 	if (error)
969 		goto done_unlock_list;
970 
971 	/* By now, all parameters should have been noted. */
972 	TAILQ_FOREACH(opt, opts, link) {
973 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
974 			error = EINVAL;
975 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
976 			goto done_unlock_list;
977 		}
978 	}
979 
980 	/*
981 	 * See if we are creating a new record or updating an existing one.
982 	 * This abuses the file error codes ENOENT and EEXIST.
983 	 */
984 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
985 	if (!cuflags) {
986 		error = EINVAL;
987 		vfs_opterror(opts, "no valid operation (create or update)");
988 		goto done_unlock_list;
989 	}
990 	pr = NULL;
991 	namelc = NULL;
992 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
993 		namelc = strrchr(name, '.');
994 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
995 		if (*p != '\0')
996 			jid = 0;
997 	}
998 	if (jid != 0) {
999 		/*
1000 		 * See if a requested jid already exists.  There is an
1001 		 * information leak here if the jid exists but is not within
1002 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1003 		 * even though they cannot see the jail, and CREATE | UPDATE
1004 		 * will return ENOENT which is not normally a valid error.
1005 		 */
1006 		if (jid < 0) {
1007 			error = EINVAL;
1008 			vfs_opterror(opts, "negative jid");
1009 			goto done_unlock_list;
1010 		}
1011 		pr = prison_find(jid);
1012 		if (pr != NULL) {
1013 			ppr = pr->pr_parent;
1014 			/* Create: jid must not exist. */
1015 			if (cuflags == JAIL_CREATE) {
1016 				mtx_unlock(&pr->pr_mtx);
1017 				error = EEXIST;
1018 				vfs_opterror(opts, "jail %d already exists",
1019 				    jid);
1020 				goto done_unlock_list;
1021 			}
1022 			if (!prison_ischild(mypr, pr)) {
1023 				mtx_unlock(&pr->pr_mtx);
1024 				pr = NULL;
1025 			} else if (pr->pr_uref == 0) {
1026 				if (!(flags & JAIL_DYING)) {
1027 					mtx_unlock(&pr->pr_mtx);
1028 					error = ENOENT;
1029 					vfs_opterror(opts, "jail %d is dying",
1030 					    jid);
1031 					goto done_unlock_list;
1032 				} else if ((flags & JAIL_ATTACH) ||
1033 				    (pr_flags & PR_PERSIST)) {
1034 					/*
1035 					 * A dying jail might be resurrected
1036 					 * (via attach or persist), but first
1037 					 * it must determine if another jail
1038 					 * has claimed its name.  Accomplish
1039 					 * this by implicitly re-setting the
1040 					 * name.
1041 					 */
1042 					if (name == NULL)
1043 						name = prison_name(mypr, pr);
1044 				}
1045 			}
1046 		}
1047 		if (pr == NULL) {
1048 			/* Update: jid must exist. */
1049 			if (cuflags == JAIL_UPDATE) {
1050 				error = ENOENT;
1051 				vfs_opterror(opts, "jail %d not found", jid);
1052 				goto done_unlock_list;
1053 			}
1054 		}
1055 	}
1056 	/*
1057 	 * If the caller provided a name, look for a jail by that name.
1058 	 * This has different semantics for creates and updates keyed by jid
1059 	 * (where the name must not already exist in a different jail),
1060 	 * and updates keyed by the name itself (where the name must exist
1061 	 * because that is the jail being updated).
1062 	 */
1063 	if (name != NULL) {
1064 		namelc = strrchr(name, '.');
1065 		if (namelc == NULL)
1066 			namelc = name;
1067 		else {
1068 			/*
1069 			 * This is a hierarchical name.  Split it into the
1070 			 * parent and child names, and make sure the parent
1071 			 * exists or matches an already found jail.
1072 			 */
1073 			*namelc = '\0';
1074 			if (pr != NULL) {
1075 				if (strncmp(name, ppr->pr_name, namelc - name)
1076 				    || ppr->pr_name[namelc - name] != '\0') {
1077 					mtx_unlock(&pr->pr_mtx);
1078 					error = EINVAL;
1079 					vfs_opterror(opts,
1080 					    "cannot change jail's parent");
1081 					goto done_unlock_list;
1082 				}
1083 			} else {
1084 				ppr = prison_find_name(mypr, name);
1085 				if (ppr == NULL) {
1086 					error = ENOENT;
1087 					vfs_opterror(opts,
1088 					    "jail \"%s\" not found", name);
1089 					goto done_unlock_list;
1090 				}
1091 				mtx_unlock(&ppr->pr_mtx);
1092 			}
1093 			name = ++namelc;
1094 		}
1095 		if (name[0] != '\0') {
1096 			namelen =
1097 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1098  name_again:
1099 			deadpr = NULL;
1100 			FOREACH_PRISON_CHILD(ppr, tpr) {
1101 				if (tpr != pr && tpr->pr_ref > 0 &&
1102 				    !strcmp(tpr->pr_name + namelen, name)) {
1103 					if (pr == NULL &&
1104 					    cuflags != JAIL_CREATE) {
1105 						mtx_lock(&tpr->pr_mtx);
1106 						if (tpr->pr_ref > 0) {
1107 							/*
1108 							 * Use this jail
1109 							 * for updates.
1110 							 */
1111 							if (tpr->pr_uref > 0) {
1112 								pr = tpr;
1113 								break;
1114 							}
1115 							deadpr = tpr;
1116 						}
1117 						mtx_unlock(&tpr->pr_mtx);
1118 					} else if (tpr->pr_uref > 0) {
1119 						/*
1120 						 * Create, or update(jid):
1121 						 * name must not exist in an
1122 						 * active sibling jail.
1123 						 */
1124 						error = EEXIST;
1125 						if (pr != NULL)
1126 							mtx_unlock(&pr->pr_mtx);
1127 						vfs_opterror(opts,
1128 						   "jail \"%s\" already exists",
1129 						   name);
1130 						goto done_unlock_list;
1131 					}
1132 				}
1133 			}
1134 			/* If no active jail is found, use a dying one. */
1135 			if (deadpr != NULL && pr == NULL) {
1136 				if (flags & JAIL_DYING) {
1137 					mtx_lock(&deadpr->pr_mtx);
1138 					if (deadpr->pr_ref == 0) {
1139 						mtx_unlock(&deadpr->pr_mtx);
1140 						goto name_again;
1141 					}
1142 					pr = deadpr;
1143 				} else if (cuflags == JAIL_UPDATE) {
1144 					error = ENOENT;
1145 					vfs_opterror(opts,
1146 					    "jail \"%s\" is dying", name);
1147 					goto done_unlock_list;
1148 				}
1149 			}
1150 			/* Update: name must exist if no jid. */
1151 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1152 				error = ENOENT;
1153 				vfs_opterror(opts, "jail \"%s\" not found",
1154 				    name);
1155 				goto done_unlock_list;
1156 			}
1157 		}
1158 	}
1159 	/* Update: must provide a jid or name. */
1160 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1161 		error = ENOENT;
1162 		vfs_opterror(opts, "update specified no jail");
1163 		goto done_unlock_list;
1164 	}
1165 
1166 	/* If there's no prison to update, create a new one and link it in. */
1167 	if (pr == NULL) {
1168 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1169 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1170 				error = EPERM;
1171 				vfs_opterror(opts, "prison limit exceeded");
1172 				goto done_unlock_list;
1173 			}
1174 		created = 1;
1175 		mtx_lock(&ppr->pr_mtx);
1176 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1177 			mtx_unlock(&ppr->pr_mtx);
1178 			error = ENOENT;
1179 			vfs_opterror(opts, "parent jail went away!");
1180 			goto done_unlock_list;
1181 		}
1182 		ppr->pr_ref++;
1183 		ppr->pr_uref++;
1184 		mtx_unlock(&ppr->pr_mtx);
1185 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1186 		if (jid == 0) {
1187 			/* Find the next free jid. */
1188 			jid = lastprid + 1;
1189  findnext:
1190 			if (jid == JAIL_MAX)
1191 				jid = 1;
1192 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1193 				if (tpr->pr_id < jid)
1194 					continue;
1195 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1196 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1197 					break;
1198 				}
1199 				if (jid == lastprid) {
1200 					error = EAGAIN;
1201 					vfs_opterror(opts,
1202 					    "no available jail IDs");
1203 					free(pr, M_PRISON);
1204 					prison_deref(ppr, PD_DEREF |
1205 					    PD_DEUREF | PD_LIST_XLOCKED);
1206 					goto done_releroot;
1207 				}
1208 				jid++;
1209 				goto findnext;
1210 			}
1211 			lastprid = jid;
1212 		} else {
1213 			/*
1214 			 * The jail already has a jid (that did not yet exist),
1215 			 * so just find where to insert it.
1216 			 */
1217 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1218 				if (tpr->pr_id >= jid) {
1219 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1220 					break;
1221 				}
1222 		}
1223 		if (tpr == NULL)
1224 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1225 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1226 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1227 			tpr->pr_childcount++;
1228 
1229 		pr->pr_parent = ppr;
1230 		pr->pr_id = jid;
1231 
1232 		/* Set some default values, and inherit some from the parent. */
1233 		if (name == NULL)
1234 			name = "";
1235 		if (path == NULL) {
1236 			path = "/";
1237 			root = mypr->pr_root;
1238 			vref(root);
1239 		}
1240 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1241 		pr->pr_flags |= PR_HOST;
1242 #if defined(INET) || defined(INET6)
1243 #ifdef VIMAGE
1244 		if (!(pr_flags & PR_VNET))
1245 #endif
1246 		{
1247 #ifdef INET
1248 			if (!(ch_flags & PR_IP4_USER))
1249 				pr->pr_flags |=
1250 				    PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1251 			else if (!(pr_flags & PR_IP4_USER)) {
1252 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1253 				if (ppr->pr_ip4 != NULL) {
1254 					pr->pr_ip4s = ppr->pr_ip4s;
1255 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1256 					    sizeof(struct in_addr), M_PRISON,
1257 					    M_WAITOK);
1258 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1259 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1260 				}
1261 			}
1262 #endif
1263 #ifdef INET6
1264 			if (!(ch_flags & PR_IP6_USER))
1265 				pr->pr_flags |=
1266 				    PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1267 			else if (!(pr_flags & PR_IP6_USER)) {
1268 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1269 				if (ppr->pr_ip6 != NULL) {
1270 					pr->pr_ip6s = ppr->pr_ip6s;
1271 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1272 					    sizeof(struct in6_addr), M_PRISON,
1273 					    M_WAITOK);
1274 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1275 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1276 				}
1277 			}
1278 #endif
1279 		}
1280 #endif
1281 		/* Source address selection is always on by default. */
1282 		pr->pr_flags |= _PR_IP_SADDRSEL;
1283 
1284 		pr->pr_securelevel = ppr->pr_securelevel;
1285 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1286 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1287 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1288 
1289 		LIST_INIT(&pr->pr_children);
1290 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1291 
1292 #ifdef VIMAGE
1293 		/* Allocate a new vnet if specified. */
1294 		pr->pr_vnet = (pr_flags & PR_VNET)
1295 		    ? vnet_alloc() : ppr->pr_vnet;
1296 #endif
1297 		/*
1298 		 * Allocate a dedicated cpuset for each jail.
1299 		 * Unlike other initial settings, this may return an erorr.
1300 		 */
1301 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1302 		if (error) {
1303 			prison_deref(pr, PD_LIST_XLOCKED);
1304 			goto done_releroot;
1305 		}
1306 
1307 		mtx_lock(&pr->pr_mtx);
1308 		/*
1309 		 * New prisons do not yet have a reference, because we do not
1310 		 * want other to see the incomplete prison once the
1311 		 * allprison_lock is downgraded.
1312 		 */
1313 	} else {
1314 		created = 0;
1315 		/*
1316 		 * Grab a reference for existing prisons, to ensure they
1317 		 * continue to exist for the duration of the call.
1318 		 */
1319 		pr->pr_ref++;
1320 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1321 		if ((pr->pr_flags & PR_VNET) &&
1322 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1323 			error = EINVAL;
1324 			vfs_opterror(opts,
1325 			    "vnet jails cannot have IP address restrictions");
1326 			goto done_deref_locked;
1327 		}
1328 #endif
1329 #ifdef INET
1330 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1331 			error = EINVAL;
1332 			vfs_opterror(opts,
1333 			    "ip4 cannot be changed after creation");
1334 			goto done_deref_locked;
1335 		}
1336 #endif
1337 #ifdef INET6
1338 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1339 			error = EINVAL;
1340 			vfs_opterror(opts,
1341 			    "ip6 cannot be changed after creation");
1342 			goto done_deref_locked;
1343 		}
1344 #endif
1345 	}
1346 
1347 	/* Do final error checking before setting anything. */
1348 	if (gotslevel) {
1349 		if (slevel < ppr->pr_securelevel) {
1350 			error = EPERM;
1351 			goto done_deref_locked;
1352 		}
1353 	}
1354 	if (gotchildmax) {
1355 		if (childmax >= ppr->pr_childmax) {
1356 			error = EPERM;
1357 			goto done_deref_locked;
1358 		}
1359 	}
1360 	if (gotenforce) {
1361 		if (enforce < ppr->pr_enforce_statfs) {
1362 			error = EPERM;
1363 			goto done_deref_locked;
1364 		}
1365 	}
1366 	if (gotrsnum) {
1367 		/*
1368 		 * devfs_rsnum is a uint16_t
1369 		 */
1370 		if (rsnum < 0 || rsnum > 65535) {
1371 			error = EINVAL;
1372 			goto done_deref_locked;
1373 		}
1374 		/*
1375 		 * Nested jails always inherit parent's devfs ruleset
1376 		 */
1377 		if (jailed(td->td_ucred)) {
1378 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1379 				error = EPERM;
1380 				goto done_deref_locked;
1381 			} else
1382 				rsnum = ppr->pr_devfs_rsnum;
1383 		}
1384 	}
1385 #ifdef INET
1386 	if (ip4s > 0) {
1387 		if (ppr->pr_flags & PR_IP4) {
1388 			/*
1389 			 * Make sure the new set of IP addresses is a
1390 			 * subset of the parent's list.  Don't worry
1391 			 * about the parent being unlocked, as any
1392 			 * setting is done with allprison_lock held.
1393 			 */
1394 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1395 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1396 					break;
1397 			if (ij == ppr->pr_ip4s) {
1398 				error = EPERM;
1399 				goto done_deref_locked;
1400 			}
1401 			if (ip4s > 1) {
1402 				for (ii = ij = 1; ii < ip4s; ii++) {
1403 					if (ip4[ii].s_addr ==
1404 					    ppr->pr_ip4[0].s_addr)
1405 						continue;
1406 					for (; ij < ppr->pr_ip4s; ij++)
1407 						if (ip4[ii].s_addr ==
1408 						    ppr->pr_ip4[ij].s_addr)
1409 							break;
1410 					if (ij == ppr->pr_ip4s)
1411 						break;
1412 				}
1413 				if (ij == ppr->pr_ip4s) {
1414 					error = EPERM;
1415 					goto done_deref_locked;
1416 				}
1417 			}
1418 		}
1419 		/*
1420 		 * Check for conflicting IP addresses.  We permit them
1421 		 * if there is no more than one IP on each jail.  If
1422 		 * there is a duplicate on a jail with more than one
1423 		 * IP stop checking and return error.
1424 		 */
1425 		tppr = ppr;
1426 #ifdef VIMAGE
1427 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1428 			if (tppr->pr_flags & PR_VNET)
1429 				break;
1430 #endif
1431 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1432 			if (tpr == pr ||
1433 #ifdef VIMAGE
1434 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1435 #endif
1436 			    tpr->pr_uref == 0) {
1437 				descend = 0;
1438 				continue;
1439 			}
1440 			if (!(tpr->pr_flags & PR_IP4_USER))
1441 				continue;
1442 			descend = 0;
1443 			if (tpr->pr_ip4 == NULL ||
1444 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1445 				continue;
1446 			for (ii = 0; ii < ip4s; ii++) {
1447 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1448 					error = EADDRINUSE;
1449 					vfs_opterror(opts,
1450 					    "IPv4 addresses clash");
1451 					goto done_deref_locked;
1452 				}
1453 			}
1454 		}
1455 	}
1456 #endif
1457 #ifdef INET6
1458 	if (ip6s > 0) {
1459 		if (ppr->pr_flags & PR_IP6) {
1460 			/*
1461 			 * Make sure the new set of IP addresses is a
1462 			 * subset of the parent's list.
1463 			 */
1464 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1465 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1466 				    &ppr->pr_ip6[ij]))
1467 					break;
1468 			if (ij == ppr->pr_ip6s) {
1469 				error = EPERM;
1470 				goto done_deref_locked;
1471 			}
1472 			if (ip6s > 1) {
1473 				for (ii = ij = 1; ii < ip6s; ii++) {
1474 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1475 					     &ppr->pr_ip6[0]))
1476 						continue;
1477 					for (; ij < ppr->pr_ip6s; ij++)
1478 						if (IN6_ARE_ADDR_EQUAL(
1479 						    &ip6[ii], &ppr->pr_ip6[ij]))
1480 							break;
1481 					if (ij == ppr->pr_ip6s)
1482 						break;
1483 				}
1484 				if (ij == ppr->pr_ip6s) {
1485 					error = EPERM;
1486 					goto done_deref_locked;
1487 				}
1488 			}
1489 		}
1490 		/* Check for conflicting IP addresses. */
1491 		tppr = ppr;
1492 #ifdef VIMAGE
1493 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1494 			if (tppr->pr_flags & PR_VNET)
1495 				break;
1496 #endif
1497 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1498 			if (tpr == pr ||
1499 #ifdef VIMAGE
1500 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1501 #endif
1502 			    tpr->pr_uref == 0) {
1503 				descend = 0;
1504 				continue;
1505 			}
1506 			if (!(tpr->pr_flags & PR_IP6_USER))
1507 				continue;
1508 			descend = 0;
1509 			if (tpr->pr_ip6 == NULL ||
1510 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1511 				continue;
1512 			for (ii = 0; ii < ip6s; ii++) {
1513 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1514 					error = EADDRINUSE;
1515 					vfs_opterror(opts,
1516 					    "IPv6 addresses clash");
1517 					goto done_deref_locked;
1518 				}
1519 			}
1520 		}
1521 	}
1522 #endif
1523 	onamelen = namelen = 0;
1524 	if (name != NULL) {
1525 		/* Give a default name of the jid. */
1526 		if (name[0] == '\0')
1527 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1528 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1529 		    *p == '\0')) {
1530 			error = EINVAL;
1531 			vfs_opterror(opts,
1532 			    "name cannot be numeric (unless it is the jid)");
1533 			goto done_deref_locked;
1534 		}
1535 		/*
1536 		 * Make sure the name isn't too long for the prison or its
1537 		 * children.
1538 		 */
1539 		onamelen = strlen(pr->pr_name);
1540 		namelen = strlen(name);
1541 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1542 			error = ENAMETOOLONG;
1543 			goto done_deref_locked;
1544 		}
1545 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1546 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1547 			    sizeof(pr->pr_name)) {
1548 				error = ENAMETOOLONG;
1549 				goto done_deref_locked;
1550 			}
1551 		}
1552 	}
1553 	if (pr_allow & ~ppr->pr_allow) {
1554 		error = EPERM;
1555 		goto done_deref_locked;
1556 	}
1557 
1558 	/* Set the parameters of the prison. */
1559 #ifdef INET
1560 	redo_ip4 = 0;
1561 	if (pr_flags & PR_IP4_USER) {
1562 		pr->pr_flags |= PR_IP4;
1563 		free(pr->pr_ip4, M_PRISON);
1564 		pr->pr_ip4s = ip4s;
1565 		pr->pr_ip4 = ip4;
1566 		ip4 = NULL;
1567 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1568 #ifdef VIMAGE
1569 			if (tpr->pr_flags & PR_VNET) {
1570 				descend = 0;
1571 				continue;
1572 			}
1573 #endif
1574 			if (prison_restrict_ip4(tpr, NULL)) {
1575 				redo_ip4 = 1;
1576 				descend = 0;
1577 			}
1578 		}
1579 	}
1580 #endif
1581 #ifdef INET6
1582 	redo_ip6 = 0;
1583 	if (pr_flags & PR_IP6_USER) {
1584 		pr->pr_flags |= PR_IP6;
1585 		free(pr->pr_ip6, M_PRISON);
1586 		pr->pr_ip6s = ip6s;
1587 		pr->pr_ip6 = ip6;
1588 		ip6 = NULL;
1589 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1590 #ifdef VIMAGE
1591 			if (tpr->pr_flags & PR_VNET) {
1592 				descend = 0;
1593 				continue;
1594 			}
1595 #endif
1596 			if (prison_restrict_ip6(tpr, NULL)) {
1597 				redo_ip6 = 1;
1598 				descend = 0;
1599 			}
1600 		}
1601 	}
1602 #endif
1603 	if (gotslevel) {
1604 		pr->pr_securelevel = slevel;
1605 		/* Set all child jails to be at least this level. */
1606 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1607 			if (tpr->pr_securelevel < slevel)
1608 				tpr->pr_securelevel = slevel;
1609 	}
1610 	if (gotchildmax) {
1611 		pr->pr_childmax = childmax;
1612 		/* Set all child jails to under this limit. */
1613 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1614 			if (tpr->pr_childmax > childmax - level)
1615 				tpr->pr_childmax = childmax > level
1616 				    ? childmax - level : 0;
1617 	}
1618 	if (gotenforce) {
1619 		pr->pr_enforce_statfs = enforce;
1620 		/* Pass this restriction on to the children. */
1621 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1622 			if (tpr->pr_enforce_statfs < enforce)
1623 				tpr->pr_enforce_statfs = enforce;
1624 	}
1625 	if (gotrsnum) {
1626 		pr->pr_devfs_rsnum = rsnum;
1627 		/* Pass this restriction on to the children. */
1628 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1629 			tpr->pr_devfs_rsnum = rsnum;
1630 	}
1631 	if (name != NULL) {
1632 		if (ppr == &prison0)
1633 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1634 		else
1635 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1636 			    ppr->pr_name, name);
1637 		/* Change this component of child names. */
1638 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1639 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1640 			    strlen(tpr->pr_name + onamelen) + 1);
1641 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1642 		}
1643 	}
1644 	if (path != NULL) {
1645 		/* Try to keep a real-rooted full pathname. */
1646 		if (fullpath_disabled && path[0] == '/' &&
1647 		    strcmp(mypr->pr_path, "/"))
1648 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1649 			    mypr->pr_path, path);
1650 		else
1651 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1652 		pr->pr_root = root;
1653 	}
1654 	if (PR_HOST & ch_flags & ~pr_flags) {
1655 		if (pr->pr_flags & PR_HOST) {
1656 			/*
1657 			 * Copy the parent's host info.  As with pr_ip4 above,
1658 			 * the lack of a lock on the parent is not a problem;
1659 			 * it is always set with allprison_lock at least
1660 			 * shared, and is held exclusively here.
1661 			 */
1662 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1663 			    sizeof(pr->pr_hostname));
1664 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1665 			    sizeof(pr->pr_domainname));
1666 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1667 			    sizeof(pr->pr_hostuuid));
1668 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1669 		}
1670 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1671 		/* Set this prison, and any descendants without PR_HOST. */
1672 		if (host != NULL)
1673 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1674 		if (domain != NULL)
1675 			strlcpy(pr->pr_domainname, domain,
1676 			    sizeof(pr->pr_domainname));
1677 		if (uuid != NULL)
1678 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1679 		if (gothid)
1680 			pr->pr_hostid = hid;
1681 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1682 			if (tpr->pr_flags & PR_HOST)
1683 				descend = 0;
1684 			else {
1685 				if (host != NULL)
1686 					strlcpy(tpr->pr_hostname,
1687 					    pr->pr_hostname,
1688 					    sizeof(tpr->pr_hostname));
1689 				if (domain != NULL)
1690 					strlcpy(tpr->pr_domainname,
1691 					    pr->pr_domainname,
1692 					    sizeof(tpr->pr_domainname));
1693 				if (uuid != NULL)
1694 					strlcpy(tpr->pr_hostuuid,
1695 					    pr->pr_hostuuid,
1696 					    sizeof(tpr->pr_hostuuid));
1697 				if (gothid)
1698 					tpr->pr_hostid = hid;
1699 			}
1700 		}
1701 	}
1702 	if ((tallow = ch_allow & ~pr_allow)) {
1703 		/* Clear allow bits in all children. */
1704 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1705 			tpr->pr_allow &= ~tallow;
1706 	}
1707 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1708 	/*
1709 	 * Persistent prisons get an extra reference, and prisons losing their
1710 	 * persist flag lose that reference.  Only do this for existing prisons
1711 	 * for now, so new ones will remain unseen until after the module
1712 	 * handlers have completed.
1713 	 */
1714 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1715 		if (pr_flags & PR_PERSIST) {
1716 			pr->pr_ref++;
1717 			pr->pr_uref++;
1718 		} else {
1719 			pr->pr_ref--;
1720 			pr->pr_uref--;
1721 		}
1722 	}
1723 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1724 	mtx_unlock(&pr->pr_mtx);
1725 
1726 #ifdef RACCT
1727 	if (created)
1728 		prison_racct_attach(pr);
1729 #endif
1730 
1731 	/* Locks may have prevented a complete restriction of child IP
1732 	 * addresses.  If so, allocate some more memory and try again.
1733 	 */
1734 #ifdef INET
1735 	while (redo_ip4) {
1736 		ip4s = pr->pr_ip4s;
1737 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1738 		mtx_lock(&pr->pr_mtx);
1739 		redo_ip4 = 0;
1740 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1741 #ifdef VIMAGE
1742 			if (tpr->pr_flags & PR_VNET) {
1743 				descend = 0;
1744 				continue;
1745 			}
1746 #endif
1747 			if (prison_restrict_ip4(tpr, ip4)) {
1748 				if (ip4 != NULL)
1749 					ip4 = NULL;
1750 				else
1751 					redo_ip4 = 1;
1752 			}
1753 		}
1754 		mtx_unlock(&pr->pr_mtx);
1755 	}
1756 #endif
1757 #ifdef INET6
1758 	while (redo_ip6) {
1759 		ip6s = pr->pr_ip6s;
1760 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1761 		mtx_lock(&pr->pr_mtx);
1762 		redo_ip6 = 0;
1763 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1764 #ifdef VIMAGE
1765 			if (tpr->pr_flags & PR_VNET) {
1766 				descend = 0;
1767 				continue;
1768 			}
1769 #endif
1770 			if (prison_restrict_ip6(tpr, ip6)) {
1771 				if (ip6 != NULL)
1772 					ip6 = NULL;
1773 				else
1774 					redo_ip6 = 1;
1775 			}
1776 		}
1777 		mtx_unlock(&pr->pr_mtx);
1778 	}
1779 #endif
1780 
1781 	/* Let the modules do their work. */
1782 	sx_downgrade(&allprison_lock);
1783 	if (created) {
1784 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1785 		if (error) {
1786 			prison_deref(pr, PD_LIST_SLOCKED);
1787 			goto done_errmsg;
1788 		}
1789 	}
1790 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1791 	if (error) {
1792 		prison_deref(pr, created
1793 		    ? PD_LIST_SLOCKED
1794 		    : PD_DEREF | PD_LIST_SLOCKED);
1795 		goto done_errmsg;
1796 	}
1797 
1798 	/* Attach this process to the prison if requested. */
1799 	if (flags & JAIL_ATTACH) {
1800 		mtx_lock(&pr->pr_mtx);
1801 		error = do_jail_attach(td, pr);
1802 		if (error) {
1803 			vfs_opterror(opts, "attach failed");
1804 			if (!created)
1805 				prison_deref(pr, PD_DEREF);
1806 			goto done_errmsg;
1807 		}
1808 	}
1809 
1810 #ifdef RACCT
1811 	if (!created) {
1812 		sx_sunlock(&allprison_lock);
1813 		prison_racct_modify(pr);
1814 		sx_slock(&allprison_lock);
1815 	}
1816 #endif
1817 
1818 	td->td_retval[0] = pr->pr_id;
1819 
1820 	/*
1821 	 * Now that it is all there, drop the temporary reference from existing
1822 	 * prisons.  Or add a reference to newly created persistent prisons
1823 	 * (which was not done earlier so that the prison would not be publicly
1824 	 * visible).
1825 	 */
1826 	if (!created) {
1827 		prison_deref(pr, (flags & JAIL_ATTACH)
1828 		    ? PD_DEREF
1829 		    : PD_DEREF | PD_LIST_SLOCKED);
1830 	} else {
1831 		if (pr_flags & PR_PERSIST) {
1832 			mtx_lock(&pr->pr_mtx);
1833 			pr->pr_ref++;
1834 			pr->pr_uref++;
1835 			mtx_unlock(&pr->pr_mtx);
1836 		}
1837 		if (!(flags & JAIL_ATTACH))
1838 			sx_sunlock(&allprison_lock);
1839 	}
1840 
1841 	goto done_errmsg;
1842 
1843  done_deref_locked:
1844 	prison_deref(pr, created
1845 	    ? PD_LOCKED | PD_LIST_XLOCKED
1846 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1847 	goto done_releroot;
1848  done_unlock_list:
1849 	sx_xunlock(&allprison_lock);
1850  done_releroot:
1851 	if (root != NULL)
1852 		vrele(root);
1853  done_errmsg:
1854 	if (error) {
1855 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1856 		if (errmsg_len > 0) {
1857 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1858 			if (errmsg_pos > 0) {
1859 				if (optuio->uio_segflg == UIO_SYSSPACE)
1860 					bcopy(errmsg,
1861 					   optuio->uio_iov[errmsg_pos].iov_base,
1862 					   errmsg_len);
1863 				else
1864 					copyout(errmsg,
1865 					   optuio->uio_iov[errmsg_pos].iov_base,
1866 					   errmsg_len);
1867 			}
1868 		}
1869 	}
1870  done_free:
1871 #ifdef INET
1872 	free(ip4, M_PRISON);
1873 #endif
1874 #ifdef INET6
1875 	free(ip6, M_PRISON);
1876 #endif
1877 	if (g_path != NULL)
1878 		free(g_path, M_TEMP);
1879 	vfs_freeopts(opts);
1880 	return (error);
1881 }
1882 
1883 
1884 /*
1885  * struct jail_get_args {
1886  *	struct iovec *iovp;
1887  *	unsigned int iovcnt;
1888  *	int flags;
1889  * };
1890  */
1891 int
1892 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1893 {
1894 	struct uio *auio;
1895 	int error;
1896 
1897 	/* Check that we have an even number of iovecs. */
1898 	if (uap->iovcnt & 1)
1899 		return (EINVAL);
1900 
1901 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1902 	if (error)
1903 		return (error);
1904 	error = kern_jail_get(td, auio, uap->flags);
1905 	if (error == 0)
1906 		error = copyout(auio->uio_iov, uap->iovp,
1907 		    uap->iovcnt * sizeof (struct iovec));
1908 	free(auio, M_IOV);
1909 	return (error);
1910 }
1911 
1912 int
1913 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1914 {
1915 	struct prison *pr, *mypr;
1916 	struct vfsopt *opt;
1917 	struct vfsoptlist *opts;
1918 	char *errmsg, *name;
1919 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1920 
1921 	if (flags & ~JAIL_GET_MASK)
1922 		return (EINVAL);
1923 
1924 	/* Get the parameter list. */
1925 	error = vfs_buildopts(optuio, &opts);
1926 	if (error)
1927 		return (error);
1928 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1929 	mypr = td->td_ucred->cr_prison;
1930 
1931 	/*
1932 	 * Find the prison specified by one of: lastjid, jid, name.
1933 	 */
1934 	sx_slock(&allprison_lock);
1935 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1936 	if (error == 0) {
1937 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1938 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1939 				mtx_lock(&pr->pr_mtx);
1940 				if (pr->pr_ref > 0 &&
1941 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1942 					break;
1943 				mtx_unlock(&pr->pr_mtx);
1944 			}
1945 		}
1946 		if (pr != NULL)
1947 			goto found_prison;
1948 		error = ENOENT;
1949 		vfs_opterror(opts, "no jail after %d", jid);
1950 		goto done_unlock_list;
1951 	} else if (error != ENOENT)
1952 		goto done_unlock_list;
1953 
1954 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1955 	if (error == 0) {
1956 		if (jid != 0) {
1957 			pr = prison_find_child(mypr, jid);
1958 			if (pr != NULL) {
1959 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1960 					mtx_unlock(&pr->pr_mtx);
1961 					error = ENOENT;
1962 					vfs_opterror(opts, "jail %d is dying",
1963 					    jid);
1964 					goto done_unlock_list;
1965 				}
1966 				goto found_prison;
1967 			}
1968 			error = ENOENT;
1969 			vfs_opterror(opts, "jail %d not found", jid);
1970 			goto done_unlock_list;
1971 		}
1972 	} else if (error != ENOENT)
1973 		goto done_unlock_list;
1974 
1975 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1976 	if (error == 0) {
1977 		if (len == 0 || name[len - 1] != '\0') {
1978 			error = EINVAL;
1979 			goto done_unlock_list;
1980 		}
1981 		pr = prison_find_name(mypr, name);
1982 		if (pr != NULL) {
1983 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1984 				mtx_unlock(&pr->pr_mtx);
1985 				error = ENOENT;
1986 				vfs_opterror(opts, "jail \"%s\" is dying",
1987 				    name);
1988 				goto done_unlock_list;
1989 			}
1990 			goto found_prison;
1991 		}
1992 		error = ENOENT;
1993 		vfs_opterror(opts, "jail \"%s\" not found", name);
1994 		goto done_unlock_list;
1995 	} else if (error != ENOENT)
1996 		goto done_unlock_list;
1997 
1998 	vfs_opterror(opts, "no jail specified");
1999 	error = ENOENT;
2000 	goto done_unlock_list;
2001 
2002  found_prison:
2003 	/* Get the parameters of the prison. */
2004 	pr->pr_ref++;
2005 	locked = PD_LOCKED;
2006 	td->td_retval[0] = pr->pr_id;
2007 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2008 	if (error != 0 && error != ENOENT)
2009 		goto done_deref;
2010 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2011 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2012 	if (error != 0 && error != ENOENT)
2013 		goto done_deref;
2014 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2015 	if (error != 0 && error != ENOENT)
2016 		goto done_deref;
2017 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2018 	    sizeof(pr->pr_cpuset->cs_id));
2019 	if (error != 0 && error != ENOENT)
2020 		goto done_deref;
2021 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2022 	if (error != 0 && error != ENOENT)
2023 		goto done_deref;
2024 #ifdef INET
2025 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2026 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2027 	if (error != 0 && error != ENOENT)
2028 		goto done_deref;
2029 #endif
2030 #ifdef INET6
2031 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2032 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2033 	if (error != 0 && error != ENOENT)
2034 		goto done_deref;
2035 #endif
2036 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2037 	    sizeof(pr->pr_securelevel));
2038 	if (error != 0 && error != ENOENT)
2039 		goto done_deref;
2040 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2041 	    sizeof(pr->pr_childcount));
2042 	if (error != 0 && error != ENOENT)
2043 		goto done_deref;
2044 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2045 	    sizeof(pr->pr_childmax));
2046 	if (error != 0 && error != ENOENT)
2047 		goto done_deref;
2048 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2049 	if (error != 0 && error != ENOENT)
2050 		goto done_deref;
2051 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2052 	if (error != 0 && error != ENOENT)
2053 		goto done_deref;
2054 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2055 	if (error != 0 && error != ENOENT)
2056 		goto done_deref;
2057 #ifdef COMPAT_FREEBSD32
2058 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2059 		uint32_t hid32 = pr->pr_hostid;
2060 
2061 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2062 	} else
2063 #endif
2064 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2065 	    sizeof(pr->pr_hostid));
2066 	if (error != 0 && error != ENOENT)
2067 		goto done_deref;
2068 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2069 	    sizeof(pr->pr_enforce_statfs));
2070 	if (error != 0 && error != ENOENT)
2071 		goto done_deref;
2072 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2073 	    sizeof(pr->pr_devfs_rsnum));
2074 	if (error != 0 && error != ENOENT)
2075 		goto done_deref;
2076 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2077 	    fi++) {
2078 		if (pr_flag_names[fi] == NULL)
2079 			continue;
2080 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2081 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2082 		if (error != 0 && error != ENOENT)
2083 			goto done_deref;
2084 		i = !i;
2085 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2086 		if (error != 0 && error != ENOENT)
2087 			goto done_deref;
2088 	}
2089 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2090 	    fi++) {
2091 		i = pr->pr_flags &
2092 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2093 		i = pr_flag_jailsys[fi].disable &&
2094 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2095 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2096 		    : JAIL_SYS_INHERIT;
2097 		error =
2098 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2099 		if (error != 0 && error != ENOENT)
2100 			goto done_deref;
2101 	}
2102 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2103 	    fi++) {
2104 		if (pr_allow_names[fi] == NULL)
2105 			continue;
2106 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2107 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2108 		if (error != 0 && error != ENOENT)
2109 			goto done_deref;
2110 		i = !i;
2111 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2112 		if (error != 0 && error != ENOENT)
2113 			goto done_deref;
2114 	}
2115 	i = (pr->pr_uref == 0);
2116 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2117 	if (error != 0 && error != ENOENT)
2118 		goto done_deref;
2119 	i = !i;
2120 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2121 	if (error != 0 && error != ENOENT)
2122 		goto done_deref;
2123 
2124 	/* Get the module parameters. */
2125 	mtx_unlock(&pr->pr_mtx);
2126 	locked = 0;
2127 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2128 	if (error)
2129 		goto done_deref;
2130 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2131 
2132 	/* By now, all parameters should have been noted. */
2133 	TAILQ_FOREACH(opt, opts, link) {
2134 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2135 			error = EINVAL;
2136 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2137 			goto done_errmsg;
2138 		}
2139 	}
2140 
2141 	/* Write the fetched parameters back to userspace. */
2142 	error = 0;
2143 	TAILQ_FOREACH(opt, opts, link) {
2144 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2145 			pos = 2 * opt->pos + 1;
2146 			optuio->uio_iov[pos].iov_len = opt->len;
2147 			if (opt->value != NULL) {
2148 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2149 					bcopy(opt->value,
2150 					    optuio->uio_iov[pos].iov_base,
2151 					    opt->len);
2152 				} else {
2153 					error = copyout(opt->value,
2154 					    optuio->uio_iov[pos].iov_base,
2155 					    opt->len);
2156 					if (error)
2157 						break;
2158 				}
2159 			}
2160 		}
2161 	}
2162 	goto done_errmsg;
2163 
2164  done_deref:
2165 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2166 	goto done_errmsg;
2167 
2168  done_unlock_list:
2169 	sx_sunlock(&allprison_lock);
2170  done_errmsg:
2171 	if (error && errmsg_pos >= 0) {
2172 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2173 		errmsg_pos = 2 * errmsg_pos + 1;
2174 		if (errmsg_len > 0) {
2175 			if (optuio->uio_segflg == UIO_SYSSPACE)
2176 				bcopy(errmsg,
2177 				    optuio->uio_iov[errmsg_pos].iov_base,
2178 				    errmsg_len);
2179 			else
2180 				copyout(errmsg,
2181 				    optuio->uio_iov[errmsg_pos].iov_base,
2182 				    errmsg_len);
2183 		}
2184 	}
2185 	vfs_freeopts(opts);
2186 	return (error);
2187 }
2188 
2189 
2190 /*
2191  * struct jail_remove_args {
2192  *	int jid;
2193  * };
2194  */
2195 int
2196 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2197 {
2198 	struct prison *pr, *cpr, *lpr, *tpr;
2199 	int descend, error;
2200 
2201 	error = priv_check(td, PRIV_JAIL_REMOVE);
2202 	if (error)
2203 		return (error);
2204 
2205 	sx_xlock(&allprison_lock);
2206 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2207 	if (pr == NULL) {
2208 		sx_xunlock(&allprison_lock);
2209 		return (EINVAL);
2210 	}
2211 
2212 	/* Remove all descendants of this prison, then remove this prison. */
2213 	pr->pr_ref++;
2214 	pr->pr_flags |= PR_REMOVE;
2215 	if (!LIST_EMPTY(&pr->pr_children)) {
2216 		mtx_unlock(&pr->pr_mtx);
2217 		lpr = NULL;
2218 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2219 			mtx_lock(&cpr->pr_mtx);
2220 			if (cpr->pr_ref > 0) {
2221 				tpr = cpr;
2222 				cpr->pr_ref++;
2223 				cpr->pr_flags |= PR_REMOVE;
2224 			} else {
2225 				/* Already removed - do not do it again. */
2226 				tpr = NULL;
2227 			}
2228 			mtx_unlock(&cpr->pr_mtx);
2229 			if (lpr != NULL) {
2230 				mtx_lock(&lpr->pr_mtx);
2231 				prison_remove_one(lpr);
2232 				sx_xlock(&allprison_lock);
2233 			}
2234 			lpr = tpr;
2235 		}
2236 		if (lpr != NULL) {
2237 			mtx_lock(&lpr->pr_mtx);
2238 			prison_remove_one(lpr);
2239 			sx_xlock(&allprison_lock);
2240 		}
2241 		mtx_lock(&pr->pr_mtx);
2242 	}
2243 	prison_remove_one(pr);
2244 	return (0);
2245 }
2246 
2247 static void
2248 prison_remove_one(struct prison *pr)
2249 {
2250 	struct proc *p;
2251 	int deuref;
2252 
2253 	/* If the prison was persistent, it is not anymore. */
2254 	deuref = 0;
2255 	if (pr->pr_flags & PR_PERSIST) {
2256 		pr->pr_ref--;
2257 		deuref = PD_DEUREF;
2258 		pr->pr_flags &= ~PR_PERSIST;
2259 	}
2260 
2261 	/*
2262 	 * jail_remove added a reference.  If that's the only one, remove
2263 	 * the prison now.
2264 	 */
2265 	KASSERT(pr->pr_ref > 0,
2266 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2267 	if (pr->pr_ref == 1) {
2268 		prison_deref(pr,
2269 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2270 		return;
2271 	}
2272 
2273 	mtx_unlock(&pr->pr_mtx);
2274 	sx_xunlock(&allprison_lock);
2275 	/*
2276 	 * Kill all processes unfortunate enough to be attached to this prison.
2277 	 */
2278 	sx_slock(&allproc_lock);
2279 	LIST_FOREACH(p, &allproc, p_list) {
2280 		PROC_LOCK(p);
2281 		if (p->p_state != PRS_NEW && p->p_ucred &&
2282 		    p->p_ucred->cr_prison == pr)
2283 			kern_psignal(p, SIGKILL);
2284 		PROC_UNLOCK(p);
2285 	}
2286 	sx_sunlock(&allproc_lock);
2287 	/* Remove the temporary reference added by jail_remove. */
2288 	prison_deref(pr, deuref | PD_DEREF);
2289 }
2290 
2291 
2292 /*
2293  * struct jail_attach_args {
2294  *	int jid;
2295  * };
2296  */
2297 int
2298 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2299 {
2300 	struct prison *pr;
2301 	int error;
2302 
2303 	error = priv_check(td, PRIV_JAIL_ATTACH);
2304 	if (error)
2305 		return (error);
2306 
2307 	sx_slock(&allprison_lock);
2308 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2309 	if (pr == NULL) {
2310 		sx_sunlock(&allprison_lock);
2311 		return (EINVAL);
2312 	}
2313 
2314 	/*
2315 	 * Do not allow a process to attach to a prison that is not
2316 	 * considered to be "alive".
2317 	 */
2318 	if (pr->pr_uref == 0) {
2319 		mtx_unlock(&pr->pr_mtx);
2320 		sx_sunlock(&allprison_lock);
2321 		return (EINVAL);
2322 	}
2323 
2324 	return (do_jail_attach(td, pr));
2325 }
2326 
2327 static int
2328 do_jail_attach(struct thread *td, struct prison *pr)
2329 {
2330 	struct prison *ppr;
2331 	struct proc *p;
2332 	struct ucred *newcred, *oldcred;
2333 	int error;
2334 
2335 	/*
2336 	 * XXX: Note that there is a slight race here if two threads
2337 	 * in the same privileged process attempt to attach to two
2338 	 * different jails at the same time.  It is important for
2339 	 * user processes not to do this, or they might end up with
2340 	 * a process root from one prison, but attached to the jail
2341 	 * of another.
2342 	 */
2343 	pr->pr_ref++;
2344 	pr->pr_uref++;
2345 	mtx_unlock(&pr->pr_mtx);
2346 
2347 	/* Let modules do whatever they need to prepare for attaching. */
2348 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2349 	if (error) {
2350 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2351 		return (error);
2352 	}
2353 	sx_sunlock(&allprison_lock);
2354 
2355 	/*
2356 	 * Reparent the newly attached process to this jail.
2357 	 */
2358 	ppr = td->td_ucred->cr_prison;
2359 	p = td->td_proc;
2360 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2361 	if (error)
2362 		goto e_revert_osd;
2363 
2364 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2365 	if ((error = change_dir(pr->pr_root, td)) != 0)
2366 		goto e_unlock;
2367 #ifdef MAC
2368 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2369 		goto e_unlock;
2370 #endif
2371 	VOP_UNLOCK(pr->pr_root, 0);
2372 	if ((error = change_root(pr->pr_root, td)))
2373 		goto e_revert_osd;
2374 
2375 	newcred = crget();
2376 	PROC_LOCK(p);
2377 	oldcred = p->p_ucred;
2378 	setsugid(p);
2379 	crcopy(newcred, oldcred);
2380 	newcred->cr_prison = pr;
2381 	p->p_ucred = newcred;
2382 	PROC_UNLOCK(p);
2383 #ifdef RACCT
2384 	racct_proc_ucred_changed(p, oldcred, newcred);
2385 #endif
2386 	crfree(oldcred);
2387 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2388 	return (0);
2389  e_unlock:
2390 	VOP_UNLOCK(pr->pr_root, 0);
2391  e_revert_osd:
2392 	/* Tell modules this thread is still in its old jail after all. */
2393 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2394 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2395 	return (error);
2396 }
2397 
2398 
2399 /*
2400  * Returns a locked prison instance, or NULL on failure.
2401  */
2402 struct prison *
2403 prison_find(int prid)
2404 {
2405 	struct prison *pr;
2406 
2407 	sx_assert(&allprison_lock, SX_LOCKED);
2408 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2409 		if (pr->pr_id == prid) {
2410 			mtx_lock(&pr->pr_mtx);
2411 			if (pr->pr_ref > 0)
2412 				return (pr);
2413 			mtx_unlock(&pr->pr_mtx);
2414 		}
2415 	}
2416 	return (NULL);
2417 }
2418 
2419 /*
2420  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2421  */
2422 struct prison *
2423 prison_find_child(struct prison *mypr, int prid)
2424 {
2425 	struct prison *pr;
2426 	int descend;
2427 
2428 	sx_assert(&allprison_lock, SX_LOCKED);
2429 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2430 		if (pr->pr_id == prid) {
2431 			mtx_lock(&pr->pr_mtx);
2432 			if (pr->pr_ref > 0)
2433 				return (pr);
2434 			mtx_unlock(&pr->pr_mtx);
2435 		}
2436 	}
2437 	return (NULL);
2438 }
2439 
2440 /*
2441  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2442  */
2443 struct prison *
2444 prison_find_name(struct prison *mypr, const char *name)
2445 {
2446 	struct prison *pr, *deadpr;
2447 	size_t mylen;
2448 	int descend;
2449 
2450 	sx_assert(&allprison_lock, SX_LOCKED);
2451 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2452  again:
2453 	deadpr = NULL;
2454 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2455 		if (!strcmp(pr->pr_name + mylen, name)) {
2456 			mtx_lock(&pr->pr_mtx);
2457 			if (pr->pr_ref > 0) {
2458 				if (pr->pr_uref > 0)
2459 					return (pr);
2460 				deadpr = pr;
2461 			}
2462 			mtx_unlock(&pr->pr_mtx);
2463 		}
2464 	}
2465 	/* There was no valid prison - perhaps there was a dying one. */
2466 	if (deadpr != NULL) {
2467 		mtx_lock(&deadpr->pr_mtx);
2468 		if (deadpr->pr_ref == 0) {
2469 			mtx_unlock(&deadpr->pr_mtx);
2470 			goto again;
2471 		}
2472 	}
2473 	return (deadpr);
2474 }
2475 
2476 /*
2477  * See if a prison has the specific flag set.
2478  */
2479 int
2480 prison_flag(struct ucred *cred, unsigned flag)
2481 {
2482 
2483 	/* This is an atomic read, so no locking is necessary. */
2484 	return (cred->cr_prison->pr_flags & flag);
2485 }
2486 
2487 int
2488 prison_allow(struct ucred *cred, unsigned flag)
2489 {
2490 
2491 	/* This is an atomic read, so no locking is necessary. */
2492 	return (cred->cr_prison->pr_allow & flag);
2493 }
2494 
2495 /*
2496  * Remove a prison reference.  If that was the last reference, remove the
2497  * prison itself - but not in this context in case there are locks held.
2498  */
2499 void
2500 prison_free_locked(struct prison *pr)
2501 {
2502 
2503 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2504 	pr->pr_ref--;
2505 	if (pr->pr_ref == 0) {
2506 		mtx_unlock(&pr->pr_mtx);
2507 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2508 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2509 		return;
2510 	}
2511 	mtx_unlock(&pr->pr_mtx);
2512 }
2513 
2514 void
2515 prison_free(struct prison *pr)
2516 {
2517 
2518 	mtx_lock(&pr->pr_mtx);
2519 	prison_free_locked(pr);
2520 }
2521 
2522 static void
2523 prison_complete(void *context, int pending)
2524 {
2525 
2526 	prison_deref((struct prison *)context, 0);
2527 }
2528 
2529 /*
2530  * Remove a prison reference (usually).  This internal version assumes no
2531  * mutexes are held, except perhaps the prison itself.  If there are no more
2532  * references, release and delist the prison.  On completion, the prison lock
2533  * and the allprison lock are both unlocked.
2534  */
2535 static void
2536 prison_deref(struct prison *pr, int flags)
2537 {
2538 	struct prison *ppr, *tpr;
2539 
2540 	if (!(flags & PD_LOCKED))
2541 		mtx_lock(&pr->pr_mtx);
2542 	for (;;) {
2543 		if (flags & PD_DEUREF) {
2544 			pr->pr_uref--;
2545 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2546 		}
2547 		if (flags & PD_DEREF)
2548 			pr->pr_ref--;
2549 		/* If the prison still has references, nothing else to do. */
2550 		if (pr->pr_ref > 0) {
2551 			mtx_unlock(&pr->pr_mtx);
2552 			if (flags & PD_LIST_SLOCKED)
2553 				sx_sunlock(&allprison_lock);
2554 			else if (flags & PD_LIST_XLOCKED)
2555 				sx_xunlock(&allprison_lock);
2556 			return;
2557 		}
2558 
2559 		mtx_unlock(&pr->pr_mtx);
2560 		if (flags & PD_LIST_SLOCKED) {
2561 			if (!sx_try_upgrade(&allprison_lock)) {
2562 				sx_sunlock(&allprison_lock);
2563 				sx_xlock(&allprison_lock);
2564 			}
2565 		} else if (!(flags & PD_LIST_XLOCKED))
2566 			sx_xlock(&allprison_lock);
2567 
2568 		TAILQ_REMOVE(&allprison, pr, pr_list);
2569 		LIST_REMOVE(pr, pr_sibling);
2570 		ppr = pr->pr_parent;
2571 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2572 			tpr->pr_childcount--;
2573 		sx_xunlock(&allprison_lock);
2574 
2575 #ifdef VIMAGE
2576 		if (pr->pr_vnet != ppr->pr_vnet)
2577 			vnet_destroy(pr->pr_vnet);
2578 #endif
2579 		if (pr->pr_root != NULL)
2580 			vrele(pr->pr_root);
2581 		mtx_destroy(&pr->pr_mtx);
2582 #ifdef INET
2583 		free(pr->pr_ip4, M_PRISON);
2584 #endif
2585 #ifdef INET6
2586 		free(pr->pr_ip6, M_PRISON);
2587 #endif
2588 		if (pr->pr_cpuset != NULL)
2589 			cpuset_rel(pr->pr_cpuset);
2590 		osd_jail_exit(pr);
2591 #ifdef RACCT
2592 		prison_racct_detach(pr);
2593 #endif
2594 		free(pr, M_PRISON);
2595 
2596 		/* Removing a prison frees a reference on its parent. */
2597 		pr = ppr;
2598 		mtx_lock(&pr->pr_mtx);
2599 		flags = PD_DEREF | PD_DEUREF;
2600 	}
2601 }
2602 
2603 void
2604 prison_hold_locked(struct prison *pr)
2605 {
2606 
2607 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2608 	KASSERT(pr->pr_ref > 0,
2609 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2610 	pr->pr_ref++;
2611 }
2612 
2613 void
2614 prison_hold(struct prison *pr)
2615 {
2616 
2617 	mtx_lock(&pr->pr_mtx);
2618 	prison_hold_locked(pr);
2619 	mtx_unlock(&pr->pr_mtx);
2620 }
2621 
2622 void
2623 prison_proc_hold(struct prison *pr)
2624 {
2625 
2626 	mtx_lock(&pr->pr_mtx);
2627 	KASSERT(pr->pr_uref > 0,
2628 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2629 	pr->pr_uref++;
2630 	mtx_unlock(&pr->pr_mtx);
2631 }
2632 
2633 void
2634 prison_proc_free(struct prison *pr)
2635 {
2636 
2637 	mtx_lock(&pr->pr_mtx);
2638 	KASSERT(pr->pr_uref > 0,
2639 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2640 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2641 }
2642 
2643 
2644 #ifdef INET
2645 /*
2646  * Restrict a prison's IP address list with its parent's, possibly replacing
2647  * it.  Return true if the replacement buffer was used (or would have been).
2648  */
2649 static int
2650 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2651 {
2652 	int ii, ij, used;
2653 	struct prison *ppr;
2654 
2655 	ppr = pr->pr_parent;
2656 	if (!(pr->pr_flags & PR_IP4_USER)) {
2657 		/* This has no user settings, so just copy the parent's list. */
2658 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2659 			/*
2660 			 * There's no room for the parent's list.  Use the
2661 			 * new list buffer, which is assumed to be big enough
2662 			 * (if it was passed).  If there's no buffer, try to
2663 			 * allocate one.
2664 			 */
2665 			used = 1;
2666 			if (newip4 == NULL) {
2667 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2668 				    M_PRISON, M_NOWAIT);
2669 				if (newip4 != NULL)
2670 					used = 0;
2671 			}
2672 			if (newip4 != NULL) {
2673 				bcopy(ppr->pr_ip4, newip4,
2674 				    ppr->pr_ip4s * sizeof(*newip4));
2675 				free(pr->pr_ip4, M_PRISON);
2676 				pr->pr_ip4 = newip4;
2677 				pr->pr_ip4s = ppr->pr_ip4s;
2678 			}
2679 			return (used);
2680 		}
2681 		pr->pr_ip4s = ppr->pr_ip4s;
2682 		if (pr->pr_ip4s > 0)
2683 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2684 			    pr->pr_ip4s * sizeof(*newip4));
2685 		else if (pr->pr_ip4 != NULL) {
2686 			free(pr->pr_ip4, M_PRISON);
2687 			pr->pr_ip4 = NULL;
2688 		}
2689 	} else if (pr->pr_ip4s > 0) {
2690 		/* Remove addresses that aren't in the parent. */
2691 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2692 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2693 				break;
2694 		if (ij < ppr->pr_ip4s)
2695 			ii = 1;
2696 		else {
2697 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2698 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2699 			ii = 0;
2700 		}
2701 		for (ij = 1; ii < pr->pr_ip4s; ) {
2702 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2703 				ii++;
2704 				continue;
2705 			}
2706 			switch (ij >= ppr->pr_ip4s ? -1 :
2707 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2708 			case -1:
2709 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2710 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2711 				break;
2712 			case 0:
2713 				ii++;
2714 				ij++;
2715 				break;
2716 			case 1:
2717 				ij++;
2718 				break;
2719 			}
2720 		}
2721 		if (pr->pr_ip4s == 0) {
2722 			pr->pr_flags |= PR_IP4_DISABLE;
2723 			free(pr->pr_ip4, M_PRISON);
2724 			pr->pr_ip4 = NULL;
2725 		}
2726 	}
2727 	return (0);
2728 }
2729 
2730 /*
2731  * Pass back primary IPv4 address of this jail.
2732  *
2733  * If not restricted return success but do not alter the address.  Caller has
2734  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2735  *
2736  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2737  * Address returned in NBO.
2738  */
2739 int
2740 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2741 {
2742 	struct prison *pr;
2743 
2744 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2745 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2746 
2747 	pr = cred->cr_prison;
2748 	if (!(pr->pr_flags & PR_IP4))
2749 		return (0);
2750 	mtx_lock(&pr->pr_mtx);
2751 	if (!(pr->pr_flags & PR_IP4)) {
2752 		mtx_unlock(&pr->pr_mtx);
2753 		return (0);
2754 	}
2755 	if (pr->pr_ip4 == NULL) {
2756 		mtx_unlock(&pr->pr_mtx);
2757 		return (EAFNOSUPPORT);
2758 	}
2759 
2760 	ia->s_addr = pr->pr_ip4[0].s_addr;
2761 	mtx_unlock(&pr->pr_mtx);
2762 	return (0);
2763 }
2764 
2765 /*
2766  * Return 1 if we should do proper source address selection or are not jailed.
2767  * We will return 0 if we should bypass source address selection in favour
2768  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2769  * returned in NBO.
2770  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2771  */
2772 int
2773 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2774 {
2775 	struct prison *pr;
2776 	struct in_addr lia;
2777 	int error;
2778 
2779 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2780 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2781 
2782 	if (!jailed(cred))
2783 		return (1);
2784 
2785 	pr = cred->cr_prison;
2786 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2787 		return (1);
2788 
2789 	lia.s_addr = INADDR_ANY;
2790 	error = prison_get_ip4(cred, &lia);
2791 	if (error)
2792 		return (error);
2793 	if (lia.s_addr == INADDR_ANY)
2794 		return (1);
2795 
2796 	ia->s_addr = lia.s_addr;
2797 	return (0);
2798 }
2799 
2800 /*
2801  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2802  */
2803 int
2804 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2805 {
2806 
2807 	if (pr1 == pr2)
2808 		return (1);
2809 
2810 	/*
2811 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2812 	 * existing prisons.
2813 	 */
2814 	while (pr1 != &prison0 &&
2815 #ifdef VIMAGE
2816 	       !(pr1->pr_flags & PR_VNET) &&
2817 #endif
2818 	       !(pr1->pr_flags & PR_IP4_USER))
2819 		pr1 = pr1->pr_parent;
2820 	while (pr2 != &prison0 &&
2821 #ifdef VIMAGE
2822 	       !(pr2->pr_flags & PR_VNET) &&
2823 #endif
2824 	       !(pr2->pr_flags & PR_IP4_USER))
2825 		pr2 = pr2->pr_parent;
2826 	return (pr1 == pr2);
2827 }
2828 
2829 /*
2830  * Make sure our (source) address is set to something meaningful to this
2831  * jail.
2832  *
2833  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2834  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2835  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2836  */
2837 int
2838 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2839 {
2840 	struct prison *pr;
2841 	struct in_addr ia0;
2842 	int error;
2843 
2844 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2845 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2846 
2847 	pr = cred->cr_prison;
2848 	if (!(pr->pr_flags & PR_IP4))
2849 		return (0);
2850 	mtx_lock(&pr->pr_mtx);
2851 	if (!(pr->pr_flags & PR_IP4)) {
2852 		mtx_unlock(&pr->pr_mtx);
2853 		return (0);
2854 	}
2855 	if (pr->pr_ip4 == NULL) {
2856 		mtx_unlock(&pr->pr_mtx);
2857 		return (EAFNOSUPPORT);
2858 	}
2859 
2860 	ia0.s_addr = ntohl(ia->s_addr);
2861 	if (ia0.s_addr == INADDR_LOOPBACK) {
2862 		ia->s_addr = pr->pr_ip4[0].s_addr;
2863 		mtx_unlock(&pr->pr_mtx);
2864 		return (0);
2865 	}
2866 
2867 	if (ia0.s_addr == INADDR_ANY) {
2868 		/*
2869 		 * In case there is only 1 IPv4 address, bind directly.
2870 		 */
2871 		if (pr->pr_ip4s == 1)
2872 			ia->s_addr = pr->pr_ip4[0].s_addr;
2873 		mtx_unlock(&pr->pr_mtx);
2874 		return (0);
2875 	}
2876 
2877 	error = _prison_check_ip4(pr, ia);
2878 	mtx_unlock(&pr->pr_mtx);
2879 	return (error);
2880 }
2881 
2882 /*
2883  * Rewrite destination address in case we will connect to loopback address.
2884  *
2885  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2886  * Address passed in in NBO and returned in NBO.
2887  */
2888 int
2889 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2890 {
2891 	struct prison *pr;
2892 
2893 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2894 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2895 
2896 	pr = cred->cr_prison;
2897 	if (!(pr->pr_flags & PR_IP4))
2898 		return (0);
2899 	mtx_lock(&pr->pr_mtx);
2900 	if (!(pr->pr_flags & PR_IP4)) {
2901 		mtx_unlock(&pr->pr_mtx);
2902 		return (0);
2903 	}
2904 	if (pr->pr_ip4 == NULL) {
2905 		mtx_unlock(&pr->pr_mtx);
2906 		return (EAFNOSUPPORT);
2907 	}
2908 
2909 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2910 		ia->s_addr = pr->pr_ip4[0].s_addr;
2911 		mtx_unlock(&pr->pr_mtx);
2912 		return (0);
2913 	}
2914 
2915 	/*
2916 	 * Return success because nothing had to be changed.
2917 	 */
2918 	mtx_unlock(&pr->pr_mtx);
2919 	return (0);
2920 }
2921 
2922 /*
2923  * Check if given address belongs to the jail referenced by cred/prison.
2924  *
2925  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2926  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2927  * doesn't allow IPv4.  Address passed in in NBO.
2928  */
2929 static int
2930 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
2931 {
2932 	int i, a, z, d;
2933 
2934 	/*
2935 	 * Check the primary IP.
2936 	 */
2937 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2938 		return (0);
2939 
2940 	/*
2941 	 * All the other IPs are sorted so we can do a binary search.
2942 	 */
2943 	a = 0;
2944 	z = pr->pr_ip4s - 2;
2945 	while (a <= z) {
2946 		i = (a + z) / 2;
2947 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2948 		if (d > 0)
2949 			z = i - 1;
2950 		else if (d < 0)
2951 			a = i + 1;
2952 		else
2953 			return (0);
2954 	}
2955 
2956 	return (EADDRNOTAVAIL);
2957 }
2958 
2959 int
2960 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
2961 {
2962 	struct prison *pr;
2963 	int error;
2964 
2965 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2966 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2967 
2968 	pr = cred->cr_prison;
2969 	if (!(pr->pr_flags & PR_IP4))
2970 		return (0);
2971 	mtx_lock(&pr->pr_mtx);
2972 	if (!(pr->pr_flags & PR_IP4)) {
2973 		mtx_unlock(&pr->pr_mtx);
2974 		return (0);
2975 	}
2976 	if (pr->pr_ip4 == NULL) {
2977 		mtx_unlock(&pr->pr_mtx);
2978 		return (EAFNOSUPPORT);
2979 	}
2980 
2981 	error = _prison_check_ip4(pr, ia);
2982 	mtx_unlock(&pr->pr_mtx);
2983 	return (error);
2984 }
2985 #endif
2986 
2987 #ifdef INET6
2988 static int
2989 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2990 {
2991 	int ii, ij, used;
2992 	struct prison *ppr;
2993 
2994 	ppr = pr->pr_parent;
2995 	if (!(pr->pr_flags & PR_IP6_USER)) {
2996 		/* This has no user settings, so just copy the parent's list. */
2997 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2998 			/*
2999 			 * There's no room for the parent's list.  Use the
3000 			 * new list buffer, which is assumed to be big enough
3001 			 * (if it was passed).  If there's no buffer, try to
3002 			 * allocate one.
3003 			 */
3004 			used = 1;
3005 			if (newip6 == NULL) {
3006 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3007 				    M_PRISON, M_NOWAIT);
3008 				if (newip6 != NULL)
3009 					used = 0;
3010 			}
3011 			if (newip6 != NULL) {
3012 				bcopy(ppr->pr_ip6, newip6,
3013 				    ppr->pr_ip6s * sizeof(*newip6));
3014 				free(pr->pr_ip6, M_PRISON);
3015 				pr->pr_ip6 = newip6;
3016 				pr->pr_ip6s = ppr->pr_ip6s;
3017 			}
3018 			return (used);
3019 		}
3020 		pr->pr_ip6s = ppr->pr_ip6s;
3021 		if (pr->pr_ip6s > 0)
3022 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3023 			    pr->pr_ip6s * sizeof(*newip6));
3024 		else if (pr->pr_ip6 != NULL) {
3025 			free(pr->pr_ip6, M_PRISON);
3026 			pr->pr_ip6 = NULL;
3027 		}
3028 	} else if (pr->pr_ip6s > 0) {
3029 		/* Remove addresses that aren't in the parent. */
3030 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3031 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3032 			    &ppr->pr_ip6[ij]))
3033 				break;
3034 		if (ij < ppr->pr_ip6s)
3035 			ii = 1;
3036 		else {
3037 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3038 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3039 			ii = 0;
3040 		}
3041 		for (ij = 1; ii < pr->pr_ip6s; ) {
3042 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3043 			    &ppr->pr_ip6[0])) {
3044 				ii++;
3045 				continue;
3046 			}
3047 			switch (ij >= ppr->pr_ip4s ? -1 :
3048 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3049 			case -1:
3050 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3051 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3052 				break;
3053 			case 0:
3054 				ii++;
3055 				ij++;
3056 				break;
3057 			case 1:
3058 				ij++;
3059 				break;
3060 			}
3061 		}
3062 		if (pr->pr_ip6s == 0) {
3063 			pr->pr_flags |= PR_IP6_DISABLE;
3064 			free(pr->pr_ip6, M_PRISON);
3065 			pr->pr_ip6 = NULL;
3066 		}
3067 	}
3068 	return 0;
3069 }
3070 
3071 /*
3072  * Pass back primary IPv6 address for this jail.
3073  *
3074  * If not restricted return success but do not alter the address.  Caller has
3075  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3076  *
3077  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3078  */
3079 int
3080 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3081 {
3082 	struct prison *pr;
3083 
3084 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3085 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3086 
3087 	pr = cred->cr_prison;
3088 	if (!(pr->pr_flags & PR_IP6))
3089 		return (0);
3090 	mtx_lock(&pr->pr_mtx);
3091 	if (!(pr->pr_flags & PR_IP6)) {
3092 		mtx_unlock(&pr->pr_mtx);
3093 		return (0);
3094 	}
3095 	if (pr->pr_ip6 == NULL) {
3096 		mtx_unlock(&pr->pr_mtx);
3097 		return (EAFNOSUPPORT);
3098 	}
3099 
3100 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3101 	mtx_unlock(&pr->pr_mtx);
3102 	return (0);
3103 }
3104 
3105 /*
3106  * Return 1 if we should do proper source address selection or are not jailed.
3107  * We will return 0 if we should bypass source address selection in favour
3108  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3109  * returned in NBO.
3110  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3111  */
3112 int
3113 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3114 {
3115 	struct prison *pr;
3116 	struct in6_addr lia6;
3117 	int error;
3118 
3119 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3120 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3121 
3122 	if (!jailed(cred))
3123 		return (1);
3124 
3125 	pr = cred->cr_prison;
3126 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3127 		return (1);
3128 
3129 	lia6 = in6addr_any;
3130 	error = prison_get_ip6(cred, &lia6);
3131 	if (error)
3132 		return (error);
3133 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3134 		return (1);
3135 
3136 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3137 	return (0);
3138 }
3139 
3140 /*
3141  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3142  */
3143 int
3144 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3145 {
3146 
3147 	if (pr1 == pr2)
3148 		return (1);
3149 
3150 	while (pr1 != &prison0 &&
3151 #ifdef VIMAGE
3152 	       !(pr1->pr_flags & PR_VNET) &&
3153 #endif
3154 	       !(pr1->pr_flags & PR_IP6_USER))
3155 		pr1 = pr1->pr_parent;
3156 	while (pr2 != &prison0 &&
3157 #ifdef VIMAGE
3158 	       !(pr2->pr_flags & PR_VNET) &&
3159 #endif
3160 	       !(pr2->pr_flags & PR_IP6_USER))
3161 		pr2 = pr2->pr_parent;
3162 	return (pr1 == pr2);
3163 }
3164 
3165 /*
3166  * Make sure our (source) address is set to something meaningful to this jail.
3167  *
3168  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3169  * when needed while binding.
3170  *
3171  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3172  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3173  * doesn't allow IPv6.
3174  */
3175 int
3176 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3177 {
3178 	struct prison *pr;
3179 	int error;
3180 
3181 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3182 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3183 
3184 	pr = cred->cr_prison;
3185 	if (!(pr->pr_flags & PR_IP6))
3186 		return (0);
3187 	mtx_lock(&pr->pr_mtx);
3188 	if (!(pr->pr_flags & PR_IP6)) {
3189 		mtx_unlock(&pr->pr_mtx);
3190 		return (0);
3191 	}
3192 	if (pr->pr_ip6 == NULL) {
3193 		mtx_unlock(&pr->pr_mtx);
3194 		return (EAFNOSUPPORT);
3195 	}
3196 
3197 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3198 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3199 		mtx_unlock(&pr->pr_mtx);
3200 		return (0);
3201 	}
3202 
3203 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3204 		/*
3205 		 * In case there is only 1 IPv6 address, and v6only is true,
3206 		 * then bind directly.
3207 		 */
3208 		if (v6only != 0 && pr->pr_ip6s == 1)
3209 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3210 		mtx_unlock(&pr->pr_mtx);
3211 		return (0);
3212 	}
3213 
3214 	error = _prison_check_ip6(pr, ia6);
3215 	mtx_unlock(&pr->pr_mtx);
3216 	return (error);
3217 }
3218 
3219 /*
3220  * Rewrite destination address in case we will connect to loopback address.
3221  *
3222  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3223  */
3224 int
3225 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3226 {
3227 	struct prison *pr;
3228 
3229 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3230 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3231 
3232 	pr = cred->cr_prison;
3233 	if (!(pr->pr_flags & PR_IP6))
3234 		return (0);
3235 	mtx_lock(&pr->pr_mtx);
3236 	if (!(pr->pr_flags & PR_IP6)) {
3237 		mtx_unlock(&pr->pr_mtx);
3238 		return (0);
3239 	}
3240 	if (pr->pr_ip6 == NULL) {
3241 		mtx_unlock(&pr->pr_mtx);
3242 		return (EAFNOSUPPORT);
3243 	}
3244 
3245 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3246 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3247 		mtx_unlock(&pr->pr_mtx);
3248 		return (0);
3249 	}
3250 
3251 	/*
3252 	 * Return success because nothing had to be changed.
3253 	 */
3254 	mtx_unlock(&pr->pr_mtx);
3255 	return (0);
3256 }
3257 
3258 /*
3259  * Check if given address belongs to the jail referenced by cred/prison.
3260  *
3261  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3262  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3263  * doesn't allow IPv6.
3264  */
3265 static int
3266 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3267 {
3268 	int i, a, z, d;
3269 
3270 	/*
3271 	 * Check the primary IP.
3272 	 */
3273 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3274 		return (0);
3275 
3276 	/*
3277 	 * All the other IPs are sorted so we can do a binary search.
3278 	 */
3279 	a = 0;
3280 	z = pr->pr_ip6s - 2;
3281 	while (a <= z) {
3282 		i = (a + z) / 2;
3283 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3284 		if (d > 0)
3285 			z = i - 1;
3286 		else if (d < 0)
3287 			a = i + 1;
3288 		else
3289 			return (0);
3290 	}
3291 
3292 	return (EADDRNOTAVAIL);
3293 }
3294 
3295 int
3296 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3297 {
3298 	struct prison *pr;
3299 	int error;
3300 
3301 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3302 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3303 
3304 	pr = cred->cr_prison;
3305 	if (!(pr->pr_flags & PR_IP6))
3306 		return (0);
3307 	mtx_lock(&pr->pr_mtx);
3308 	if (!(pr->pr_flags & PR_IP6)) {
3309 		mtx_unlock(&pr->pr_mtx);
3310 		return (0);
3311 	}
3312 	if (pr->pr_ip6 == NULL) {
3313 		mtx_unlock(&pr->pr_mtx);
3314 		return (EAFNOSUPPORT);
3315 	}
3316 
3317 	error = _prison_check_ip6(pr, ia6);
3318 	mtx_unlock(&pr->pr_mtx);
3319 	return (error);
3320 }
3321 #endif
3322 
3323 /*
3324  * Check if a jail supports the given address family.
3325  *
3326  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3327  * if not.
3328  */
3329 int
3330 prison_check_af(struct ucred *cred, int af)
3331 {
3332 	struct prison *pr;
3333 	int error;
3334 
3335 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3336 
3337 	pr = cred->cr_prison;
3338 #ifdef VIMAGE
3339 	/* Prisons with their own network stack are not limited. */
3340 	if (prison_owns_vnet(cred))
3341 		return (0);
3342 #endif
3343 
3344 	error = 0;
3345 	switch (af)
3346 	{
3347 #ifdef INET
3348 	case AF_INET:
3349 		if (pr->pr_flags & PR_IP4)
3350 		{
3351 			mtx_lock(&pr->pr_mtx);
3352 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3353 				error = EAFNOSUPPORT;
3354 			mtx_unlock(&pr->pr_mtx);
3355 		}
3356 		break;
3357 #endif
3358 #ifdef INET6
3359 	case AF_INET6:
3360 		if (pr->pr_flags & PR_IP6)
3361 		{
3362 			mtx_lock(&pr->pr_mtx);
3363 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3364 				error = EAFNOSUPPORT;
3365 			mtx_unlock(&pr->pr_mtx);
3366 		}
3367 		break;
3368 #endif
3369 	case AF_LOCAL:
3370 	case AF_ROUTE:
3371 		break;
3372 	default:
3373 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3374 			error = EAFNOSUPPORT;
3375 	}
3376 	return (error);
3377 }
3378 
3379 /*
3380  * Check if given address belongs to the jail referenced by cred (wrapper to
3381  * prison_check_ip[46]).
3382  *
3383  * Returns 0 if jail doesn't restrict the address family or if address belongs
3384  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3385  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3386  */
3387 int
3388 prison_if(struct ucred *cred, struct sockaddr *sa)
3389 {
3390 #ifdef INET
3391 	struct sockaddr_in *sai;
3392 #endif
3393 #ifdef INET6
3394 	struct sockaddr_in6 *sai6;
3395 #endif
3396 	int error;
3397 
3398 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3399 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3400 
3401 #ifdef VIMAGE
3402 	if (prison_owns_vnet(cred))
3403 		return (0);
3404 #endif
3405 
3406 	error = 0;
3407 	switch (sa->sa_family)
3408 	{
3409 #ifdef INET
3410 	case AF_INET:
3411 		sai = (struct sockaddr_in *)sa;
3412 		error = prison_check_ip4(cred, &sai->sin_addr);
3413 		break;
3414 #endif
3415 #ifdef INET6
3416 	case AF_INET6:
3417 		sai6 = (struct sockaddr_in6 *)sa;
3418 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3419 		break;
3420 #endif
3421 	default:
3422 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3423 			error = EAFNOSUPPORT;
3424 	}
3425 	return (error);
3426 }
3427 
3428 /*
3429  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3430  */
3431 int
3432 prison_check(struct ucred *cred1, struct ucred *cred2)
3433 {
3434 
3435 	return ((cred1->cr_prison == cred2->cr_prison ||
3436 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3437 }
3438 
3439 /*
3440  * Return 1 if p2 is a child of p1, otherwise 0.
3441  */
3442 int
3443 prison_ischild(struct prison *pr1, struct prison *pr2)
3444 {
3445 
3446 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3447 		if (pr1 == pr2)
3448 			return (1);
3449 	return (0);
3450 }
3451 
3452 /*
3453  * Return 1 if the passed credential is in a jail, otherwise 0.
3454  */
3455 int
3456 jailed(struct ucred *cred)
3457 {
3458 
3459 	return (cred->cr_prison != &prison0);
3460 }
3461 
3462 /*
3463  * Return 1 if the passed credential is in a jail and that jail does not
3464  * have its own virtual network stack, otherwise 0.
3465  */
3466 int
3467 jailed_without_vnet(struct ucred *cred)
3468 {
3469 
3470 	if (!jailed(cred))
3471 		return (0);
3472 #ifdef VIMAGE
3473 	if (prison_owns_vnet(cred))
3474 		return (0);
3475 #endif
3476 
3477 	return (1);
3478 }
3479 
3480 /*
3481  * Return the correct hostname (domainname, et al) for the passed credential.
3482  */
3483 void
3484 getcredhostname(struct ucred *cred, char *buf, size_t size)
3485 {
3486 	struct prison *pr;
3487 
3488 	/*
3489 	 * A NULL credential can be used to shortcut to the physical
3490 	 * system's hostname.
3491 	 */
3492 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3493 	mtx_lock(&pr->pr_mtx);
3494 	strlcpy(buf, pr->pr_hostname, size);
3495 	mtx_unlock(&pr->pr_mtx);
3496 }
3497 
3498 void
3499 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3500 {
3501 
3502 	mtx_lock(&cred->cr_prison->pr_mtx);
3503 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3504 	mtx_unlock(&cred->cr_prison->pr_mtx);
3505 }
3506 
3507 void
3508 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3509 {
3510 
3511 	mtx_lock(&cred->cr_prison->pr_mtx);
3512 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3513 	mtx_unlock(&cred->cr_prison->pr_mtx);
3514 }
3515 
3516 void
3517 getcredhostid(struct ucred *cred, unsigned long *hostid)
3518 {
3519 
3520 	mtx_lock(&cred->cr_prison->pr_mtx);
3521 	*hostid = cred->cr_prison->pr_hostid;
3522 	mtx_unlock(&cred->cr_prison->pr_mtx);
3523 }
3524 
3525 #ifdef VIMAGE
3526 /*
3527  * Determine whether the prison represented by cred owns
3528  * its vnet rather than having it inherited.
3529  *
3530  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3531  */
3532 int
3533 prison_owns_vnet(struct ucred *cred)
3534 {
3535 
3536 	/*
3537 	 * vnets cannot be added/removed after jail creation,
3538 	 * so no need to lock here.
3539 	 */
3540 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3541 }
3542 #endif
3543 
3544 /*
3545  * Determine whether the subject represented by cred can "see"
3546  * status of a mount point.
3547  * Returns: 0 for permitted, ENOENT otherwise.
3548  * XXX: This function should be called cr_canseemount() and should be
3549  *      placed in kern_prot.c.
3550  */
3551 int
3552 prison_canseemount(struct ucred *cred, struct mount *mp)
3553 {
3554 	struct prison *pr;
3555 	struct statfs *sp;
3556 	size_t len;
3557 
3558 	pr = cred->cr_prison;
3559 	if (pr->pr_enforce_statfs == 0)
3560 		return (0);
3561 	if (pr->pr_root->v_mount == mp)
3562 		return (0);
3563 	if (pr->pr_enforce_statfs == 2)
3564 		return (ENOENT);
3565 	/*
3566 	 * If jail's chroot directory is set to "/" we should be able to see
3567 	 * all mount-points from inside a jail.
3568 	 * This is ugly check, but this is the only situation when jail's
3569 	 * directory ends with '/'.
3570 	 */
3571 	if (strcmp(pr->pr_path, "/") == 0)
3572 		return (0);
3573 	len = strlen(pr->pr_path);
3574 	sp = &mp->mnt_stat;
3575 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3576 		return (ENOENT);
3577 	/*
3578 	 * Be sure that we don't have situation where jail's root directory
3579 	 * is "/some/path" and mount point is "/some/pathpath".
3580 	 */
3581 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3582 		return (ENOENT);
3583 	return (0);
3584 }
3585 
3586 void
3587 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3588 {
3589 	char jpath[MAXPATHLEN];
3590 	struct prison *pr;
3591 	size_t len;
3592 
3593 	pr = cred->cr_prison;
3594 	if (pr->pr_enforce_statfs == 0)
3595 		return;
3596 	if (prison_canseemount(cred, mp) != 0) {
3597 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3598 		strlcpy(sp->f_mntonname, "[restricted]",
3599 		    sizeof(sp->f_mntonname));
3600 		return;
3601 	}
3602 	if (pr->pr_root->v_mount == mp) {
3603 		/*
3604 		 * Clear current buffer data, so we are sure nothing from
3605 		 * the valid path left there.
3606 		 */
3607 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3608 		*sp->f_mntonname = '/';
3609 		return;
3610 	}
3611 	/*
3612 	 * If jail's chroot directory is set to "/" we should be able to see
3613 	 * all mount-points from inside a jail.
3614 	 */
3615 	if (strcmp(pr->pr_path, "/") == 0)
3616 		return;
3617 	len = strlen(pr->pr_path);
3618 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3619 	/*
3620 	 * Clear current buffer data, so we are sure nothing from
3621 	 * the valid path left there.
3622 	 */
3623 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3624 	if (*jpath == '\0') {
3625 		/* Should never happen. */
3626 		*sp->f_mntonname = '/';
3627 	} else {
3628 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3629 	}
3630 }
3631 
3632 /*
3633  * Check with permission for a specific privilege is granted within jail.  We
3634  * have a specific list of accepted privileges; the rest are denied.
3635  */
3636 int
3637 prison_priv_check(struct ucred *cred, int priv)
3638 {
3639 
3640 	if (!jailed(cred))
3641 		return (0);
3642 
3643 #ifdef VIMAGE
3644 	/*
3645 	 * Privileges specific to prisons with a virtual network stack.
3646 	 * There might be a duplicate entry here in case the privilege
3647 	 * is only granted conditionally in the legacy jail case.
3648 	 */
3649 	switch (priv) {
3650 #ifdef notyet
3651 		/*
3652 		 * NFS-specific privileges.
3653 		 */
3654 	case PRIV_NFS_DAEMON:
3655 	case PRIV_NFS_LOCKD:
3656 #endif
3657 		/*
3658 		 * Network stack privileges.
3659 		 */
3660 	case PRIV_NET_BRIDGE:
3661 	case PRIV_NET_GRE:
3662 	case PRIV_NET_BPF:
3663 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3664 	case PRIV_NET_ROUTE:
3665 	case PRIV_NET_TAP:
3666 	case PRIV_NET_SETIFMTU:
3667 	case PRIV_NET_SETIFFLAGS:
3668 	case PRIV_NET_SETIFCAP:
3669 	case PRIV_NET_SETIFDESCR:
3670 	case PRIV_NET_SETIFNAME	:
3671 	case PRIV_NET_SETIFMETRIC:
3672 	case PRIV_NET_SETIFPHYS:
3673 	case PRIV_NET_SETIFMAC:
3674 	case PRIV_NET_ADDMULTI:
3675 	case PRIV_NET_DELMULTI:
3676 	case PRIV_NET_HWIOCTL:
3677 	case PRIV_NET_SETLLADDR:
3678 	case PRIV_NET_ADDIFGROUP:
3679 	case PRIV_NET_DELIFGROUP:
3680 	case PRIV_NET_IFCREATE:
3681 	case PRIV_NET_IFDESTROY:
3682 	case PRIV_NET_ADDIFADDR:
3683 	case PRIV_NET_DELIFADDR:
3684 	case PRIV_NET_LAGG:
3685 	case PRIV_NET_GIF:
3686 	case PRIV_NET_SETIFVNET:
3687 	case PRIV_NET_SETIFFIB:
3688 
3689 		/*
3690 		 * 802.11-related privileges.
3691 		 */
3692 	case PRIV_NET80211_GETKEY:
3693 #ifdef notyet
3694 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3695 #endif
3696 
3697 #ifdef notyet
3698 		/*
3699 		 * AppleTalk privileges.
3700 		 */
3701 	case PRIV_NETATALK_RESERVEDPORT:
3702 
3703 		/*
3704 		 * ATM privileges.
3705 		 */
3706 	case PRIV_NETATM_CFG:
3707 	case PRIV_NETATM_ADD:
3708 	case PRIV_NETATM_DEL:
3709 	case PRIV_NETATM_SET:
3710 
3711 		/*
3712 		 * Bluetooth privileges.
3713 		 */
3714 	case PRIV_NETBLUETOOTH_RAW:
3715 #endif
3716 
3717 		/*
3718 		 * Netgraph and netgraph module privileges.
3719 		 */
3720 	case PRIV_NETGRAPH_CONTROL:
3721 #ifdef notyet
3722 	case PRIV_NETGRAPH_TTY:
3723 #endif
3724 
3725 		/*
3726 		 * IPv4 and IPv6 privileges.
3727 		 */
3728 	case PRIV_NETINET_IPFW:
3729 	case PRIV_NETINET_DIVERT:
3730 	case PRIV_NETINET_PF:
3731 	case PRIV_NETINET_DUMMYNET:
3732 	case PRIV_NETINET_CARP:
3733 	case PRIV_NETINET_MROUTE:
3734 	case PRIV_NETINET_RAW:
3735 	case PRIV_NETINET_ADDRCTRL6:
3736 	case PRIV_NETINET_ND6:
3737 	case PRIV_NETINET_SCOPE6:
3738 	case PRIV_NETINET_ALIFETIME6:
3739 	case PRIV_NETINET_IPSEC:
3740 	case PRIV_NETINET_BINDANY:
3741 
3742 #ifdef notyet
3743 		/*
3744 		 * IPX/SPX privileges.
3745 		 */
3746 	case PRIV_NETIPX_RESERVEDPORT:
3747 	case PRIV_NETIPX_RAW:
3748 
3749 		/*
3750 		 * NCP privileges.
3751 		 */
3752 	case PRIV_NETNCP:
3753 
3754 		/*
3755 		 * SMB privileges.
3756 		 */
3757 	case PRIV_NETSMB:
3758 #endif
3759 
3760 	/*
3761 	 * No default: or deny here.
3762 	 * In case of no permit fall through to next switch().
3763 	 */
3764 		if (cred->cr_prison->pr_flags & PR_VNET)
3765 			return (0);
3766 	}
3767 #endif /* VIMAGE */
3768 
3769 	switch (priv) {
3770 
3771 		/*
3772 		 * Allow ktrace privileges for root in jail.
3773 		 */
3774 	case PRIV_KTRACE:
3775 
3776 #if 0
3777 		/*
3778 		 * Allow jailed processes to configure audit identity and
3779 		 * submit audit records (login, etc).  In the future we may
3780 		 * want to further refine the relationship between audit and
3781 		 * jail.
3782 		 */
3783 	case PRIV_AUDIT_GETAUDIT:
3784 	case PRIV_AUDIT_SETAUDIT:
3785 	case PRIV_AUDIT_SUBMIT:
3786 #endif
3787 
3788 		/*
3789 		 * Allow jailed processes to manipulate process UNIX
3790 		 * credentials in any way they see fit.
3791 		 */
3792 	case PRIV_CRED_SETUID:
3793 	case PRIV_CRED_SETEUID:
3794 	case PRIV_CRED_SETGID:
3795 	case PRIV_CRED_SETEGID:
3796 	case PRIV_CRED_SETGROUPS:
3797 	case PRIV_CRED_SETREUID:
3798 	case PRIV_CRED_SETREGID:
3799 	case PRIV_CRED_SETRESUID:
3800 	case PRIV_CRED_SETRESGID:
3801 
3802 		/*
3803 		 * Jail implements visibility constraints already, so allow
3804 		 * jailed root to override uid/gid-based constraints.
3805 		 */
3806 	case PRIV_SEEOTHERGIDS:
3807 	case PRIV_SEEOTHERUIDS:
3808 
3809 		/*
3810 		 * Jail implements inter-process debugging limits already, so
3811 		 * allow jailed root various debugging privileges.
3812 		 */
3813 	case PRIV_DEBUG_DIFFCRED:
3814 	case PRIV_DEBUG_SUGID:
3815 	case PRIV_DEBUG_UNPRIV:
3816 
3817 		/*
3818 		 * Allow jail to set various resource limits and login
3819 		 * properties, and for now, exceed process resource limits.
3820 		 */
3821 	case PRIV_PROC_LIMIT:
3822 	case PRIV_PROC_SETLOGIN:
3823 	case PRIV_PROC_SETRLIMIT:
3824 
3825 		/*
3826 		 * System V and POSIX IPC privileges are granted in jail.
3827 		 */
3828 	case PRIV_IPC_READ:
3829 	case PRIV_IPC_WRITE:
3830 	case PRIV_IPC_ADMIN:
3831 	case PRIV_IPC_MSGSIZE:
3832 	case PRIV_MQ_ADMIN:
3833 
3834 		/*
3835 		 * Jail operations within a jail work on child jails.
3836 		 */
3837 	case PRIV_JAIL_ATTACH:
3838 	case PRIV_JAIL_SET:
3839 	case PRIV_JAIL_REMOVE:
3840 
3841 		/*
3842 		 * Jail implements its own inter-process limits, so allow
3843 		 * root processes in jail to change scheduling on other
3844 		 * processes in the same jail.  Likewise for signalling.
3845 		 */
3846 	case PRIV_SCHED_DIFFCRED:
3847 	case PRIV_SCHED_CPUSET:
3848 	case PRIV_SIGNAL_DIFFCRED:
3849 	case PRIV_SIGNAL_SUGID:
3850 
3851 		/*
3852 		 * Allow jailed processes to write to sysctls marked as jail
3853 		 * writable.
3854 		 */
3855 	case PRIV_SYSCTL_WRITEJAIL:
3856 
3857 		/*
3858 		 * Allow root in jail to manage a variety of quota
3859 		 * properties.  These should likely be conditional on a
3860 		 * configuration option.
3861 		 */
3862 	case PRIV_VFS_GETQUOTA:
3863 	case PRIV_VFS_SETQUOTA:
3864 
3865 		/*
3866 		 * Since Jail relies on chroot() to implement file system
3867 		 * protections, grant many VFS privileges to root in jail.
3868 		 * Be careful to exclude mount-related and NFS-related
3869 		 * privileges.
3870 		 */
3871 	case PRIV_VFS_READ:
3872 	case PRIV_VFS_WRITE:
3873 	case PRIV_VFS_ADMIN:
3874 	case PRIV_VFS_EXEC:
3875 	case PRIV_VFS_LOOKUP:
3876 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3877 	case PRIV_VFS_CHFLAGS_DEV:
3878 	case PRIV_VFS_CHOWN:
3879 	case PRIV_VFS_CHROOT:
3880 	case PRIV_VFS_RETAINSUGID:
3881 	case PRIV_VFS_FCHROOT:
3882 	case PRIV_VFS_LINK:
3883 	case PRIV_VFS_SETGID:
3884 	case PRIV_VFS_STAT:
3885 	case PRIV_VFS_STICKYFILE:
3886 		return (0);
3887 
3888 		/*
3889 		 * Depending on the global setting, allow privilege of
3890 		 * setting system flags.
3891 		 */
3892 	case PRIV_VFS_SYSFLAGS:
3893 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3894 			return (0);
3895 		else
3896 			return (EPERM);
3897 
3898 		/*
3899 		 * Depending on the global setting, allow privilege of
3900 		 * mounting/unmounting file systems.
3901 		 */
3902 	case PRIV_VFS_MOUNT:
3903 	case PRIV_VFS_UNMOUNT:
3904 	case PRIV_VFS_MOUNT_NONUSER:
3905 	case PRIV_VFS_MOUNT_OWNER:
3906 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3907 		    cred->cr_prison->pr_enforce_statfs < 2)
3908 			return (0);
3909 		else
3910 			return (EPERM);
3911 
3912 		/*
3913 		 * Allow jailed root to bind reserved ports and reuse in-use
3914 		 * ports.
3915 		 */
3916 	case PRIV_NETINET_RESERVEDPORT:
3917 	case PRIV_NETINET_REUSEPORT:
3918 		return (0);
3919 
3920 		/*
3921 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3922 		 */
3923 	case PRIV_NETINET_SETHDROPTS:
3924 		return (0);
3925 
3926 		/*
3927 		 * Conditionally allow creating raw sockets in jail.
3928 		 */
3929 	case PRIV_NETINET_RAW:
3930 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3931 			return (0);
3932 		else
3933 			return (EPERM);
3934 
3935 		/*
3936 		 * Since jail implements its own visibility limits on netstat
3937 		 * sysctls, allow getcred.  This allows identd to work in
3938 		 * jail.
3939 		 */
3940 	case PRIV_NETINET_GETCRED:
3941 		return (0);
3942 
3943 		/*
3944 		 * Allow jailed root to set loginclass.
3945 		 */
3946 	case PRIV_PROC_SETLOGINCLASS:
3947 		return (0);
3948 
3949 	default:
3950 		/*
3951 		 * In all remaining cases, deny the privilege request.  This
3952 		 * includes almost all network privileges, many system
3953 		 * configuration privileges.
3954 		 */
3955 		return (EPERM);
3956 	}
3957 }
3958 
3959 /*
3960  * Return the part of pr2's name that is relative to pr1, or the whole name
3961  * if it does not directly follow.
3962  */
3963 
3964 char *
3965 prison_name(struct prison *pr1, struct prison *pr2)
3966 {
3967 	char *name;
3968 
3969 	/* Jails see themselves as "0" (if they see themselves at all). */
3970 	if (pr1 == pr2)
3971 		return "0";
3972 	name = pr2->pr_name;
3973 	if (prison_ischild(pr1, pr2)) {
3974 		/*
3975 		 * pr1 isn't locked (and allprison_lock may not be either)
3976 		 * so its length can't be counted on.  But the number of dots
3977 		 * can be counted on - and counted.
3978 		 */
3979 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3980 			name = strchr(name, '.') + 1;
3981 	}
3982 	return (name);
3983 }
3984 
3985 /*
3986  * Return the part of pr2's path that is relative to pr1, or the whole path
3987  * if it does not directly follow.
3988  */
3989 static char *
3990 prison_path(struct prison *pr1, struct prison *pr2)
3991 {
3992 	char *path1, *path2;
3993 	int len1;
3994 
3995 	path1 = pr1->pr_path;
3996 	path2 = pr2->pr_path;
3997 	if (!strcmp(path1, "/"))
3998 		return (path2);
3999 	len1 = strlen(path1);
4000 	if (strncmp(path1, path2, len1))
4001 		return (path2);
4002 	if (path2[len1] == '\0')
4003 		return "/";
4004 	if (path2[len1] == '/')
4005 		return (path2 + len1);
4006 	return (path2);
4007 }
4008 
4009 
4010 /*
4011  * Jail-related sysctls.
4012  */
4013 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4014     "Jails");
4015 
4016 static int
4017 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4018 {
4019 	struct xprison *xp;
4020 	struct prison *pr, *cpr;
4021 #ifdef INET
4022 	struct in_addr *ip4 = NULL;
4023 	int ip4s = 0;
4024 #endif
4025 #ifdef INET6
4026 	struct in6_addr *ip6 = NULL;
4027 	int ip6s = 0;
4028 #endif
4029 	int descend, error;
4030 
4031 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4032 	pr = req->td->td_ucred->cr_prison;
4033 	error = 0;
4034 	sx_slock(&allprison_lock);
4035 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4036 #if defined(INET) || defined(INET6)
4037  again:
4038 #endif
4039 		mtx_lock(&cpr->pr_mtx);
4040 #ifdef INET
4041 		if (cpr->pr_ip4s > 0) {
4042 			if (ip4s < cpr->pr_ip4s) {
4043 				ip4s = cpr->pr_ip4s;
4044 				mtx_unlock(&cpr->pr_mtx);
4045 				ip4 = realloc(ip4, ip4s *
4046 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4047 				goto again;
4048 			}
4049 			bcopy(cpr->pr_ip4, ip4,
4050 			    cpr->pr_ip4s * sizeof(struct in_addr));
4051 		}
4052 #endif
4053 #ifdef INET6
4054 		if (cpr->pr_ip6s > 0) {
4055 			if (ip6s < cpr->pr_ip6s) {
4056 				ip6s = cpr->pr_ip6s;
4057 				mtx_unlock(&cpr->pr_mtx);
4058 				ip6 = realloc(ip6, ip6s *
4059 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4060 				goto again;
4061 			}
4062 			bcopy(cpr->pr_ip6, ip6,
4063 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4064 		}
4065 #endif
4066 		if (cpr->pr_ref == 0) {
4067 			mtx_unlock(&cpr->pr_mtx);
4068 			continue;
4069 		}
4070 		bzero(xp, sizeof(*xp));
4071 		xp->pr_version = XPRISON_VERSION;
4072 		xp->pr_id = cpr->pr_id;
4073 		xp->pr_state = cpr->pr_uref > 0
4074 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4075 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4076 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4077 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4078 #ifdef INET
4079 		xp->pr_ip4s = cpr->pr_ip4s;
4080 #endif
4081 #ifdef INET6
4082 		xp->pr_ip6s = cpr->pr_ip6s;
4083 #endif
4084 		mtx_unlock(&cpr->pr_mtx);
4085 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4086 		if (error)
4087 			break;
4088 #ifdef INET
4089 		if (xp->pr_ip4s > 0) {
4090 			error = SYSCTL_OUT(req, ip4,
4091 			    xp->pr_ip4s * sizeof(struct in_addr));
4092 			if (error)
4093 				break;
4094 		}
4095 #endif
4096 #ifdef INET6
4097 		if (xp->pr_ip6s > 0) {
4098 			error = SYSCTL_OUT(req, ip6,
4099 			    xp->pr_ip6s * sizeof(struct in6_addr));
4100 			if (error)
4101 				break;
4102 		}
4103 #endif
4104 	}
4105 	sx_sunlock(&allprison_lock);
4106 	free(xp, M_TEMP);
4107 #ifdef INET
4108 	free(ip4, M_TEMP);
4109 #endif
4110 #ifdef INET6
4111 	free(ip6, M_TEMP);
4112 #endif
4113 	return (error);
4114 }
4115 
4116 SYSCTL_OID(_security_jail, OID_AUTO, list,
4117     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4118     sysctl_jail_list, "S", "List of active jails");
4119 
4120 static int
4121 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4122 {
4123 	int error, injail;
4124 
4125 	injail = jailed(req->td->td_ucred);
4126 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4127 
4128 	return (error);
4129 }
4130 
4131 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4132     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4133     sysctl_jail_jailed, "I", "Process in jail?");
4134 
4135 static int
4136 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4137 {
4138 	int error, havevnet;
4139 #ifdef VIMAGE
4140 	struct ucred *cred = req->td->td_ucred;
4141 
4142 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4143 #else
4144 	havevnet = 0;
4145 #endif
4146 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4147 
4148 	return (error);
4149 }
4150 
4151 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4152     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4153     sysctl_jail_vnet, "I", "Jail owns VNET?");
4154 
4155 #if defined(INET) || defined(INET6)
4156 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4157     &jail_max_af_ips, 0,
4158     "Number of IP addresses a jail may have at most per address family");
4159 #endif
4160 
4161 /*
4162  * Default parameters for jail(2) compatability.  For historical reasons,
4163  * the sysctl names have varying similarity to the parameter names.  Prisons
4164  * just see their own parameters, and can't change them.
4165  */
4166 static int
4167 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4168 {
4169 	struct prison *pr;
4170 	int allow, error, i;
4171 
4172 	pr = req->td->td_ucred->cr_prison;
4173 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4174 
4175 	/* Get the current flag value, and convert it to a boolean. */
4176 	i = (allow & arg2) ? 1 : 0;
4177 	if (arg1 != NULL)
4178 		i = !i;
4179 	error = sysctl_handle_int(oidp, &i, 0, req);
4180 	if (error || !req->newptr)
4181 		return (error);
4182 	i = i ? arg2 : 0;
4183 	if (arg1 != NULL)
4184 		i ^= arg2;
4185 	/*
4186 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4187 	 * for writing.
4188 	 */
4189 	mtx_lock(&prison0.pr_mtx);
4190 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4191 	mtx_unlock(&prison0.pr_mtx);
4192 	return (0);
4193 }
4194 
4195 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4196     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4197     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4198     "Processes in jail can set their hostnames");
4199 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4200     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4201     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4202     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4203 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4204     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4205     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4206     "Processes in jail can use System V IPC primitives");
4207 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4208     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4209     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4210     "Prison root can create raw sockets");
4211 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4212     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4213     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4214     "Processes in jail can alter system file flags");
4215 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4216     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4217     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4218     "Processes in jail can mount/unmount jail-friendly file systems");
4219 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4220     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4221     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4222     "Processes in jail can mount the devfs file system");
4223 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4224     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4225     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4226     "Processes in jail can mount the nullfs file system");
4227 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4228     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4229     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4230     "Processes in jail can mount the procfs file system");
4231 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4232     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4233     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4234     "Processes in jail can mount the zfs file system");
4235 
4236 static int
4237 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4238 {
4239 	struct prison *pr;
4240 	int level, error;
4241 
4242 	pr = req->td->td_ucred->cr_prison;
4243 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4244 	error = sysctl_handle_int(oidp, &level, 0, req);
4245 	if (error || !req->newptr)
4246 		return (error);
4247 	*(int *)arg1 = level;
4248 	return (0);
4249 }
4250 
4251 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4252     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4253     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4254     sysctl_jail_default_level, "I",
4255     "Processes in jail cannot see all mounted file systems");
4256 
4257 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4258     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4259     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4260     sysctl_jail_default_level, "I",
4261     "Ruleset for the devfs filesystem in jail");
4262 
4263 /*
4264  * Nodes to describe jail parameters.  Maximum length of string parameters
4265  * is returned in the string itself, and the other parameters exist merely
4266  * to make themselves and their types known.
4267  */
4268 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4269     "Jail parameters");
4270 
4271 int
4272 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4273 {
4274 	int i;
4275 	long l;
4276 	size_t s;
4277 	char numbuf[12];
4278 
4279 	switch (oidp->oid_kind & CTLTYPE)
4280 	{
4281 	case CTLTYPE_LONG:
4282 	case CTLTYPE_ULONG:
4283 		l = 0;
4284 #ifdef SCTL_MASK32
4285 		if (!(req->flags & SCTL_MASK32))
4286 #endif
4287 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4288 	case CTLTYPE_INT:
4289 	case CTLTYPE_UINT:
4290 		i = 0;
4291 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4292 	case CTLTYPE_STRING:
4293 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4294 		return
4295 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4296 	case CTLTYPE_STRUCT:
4297 		s = (size_t)arg2;
4298 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4299 	}
4300 	return (0);
4301 }
4302 
4303 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4304 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4305 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4306 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4307 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4308     "I", "Jail secure level");
4309 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4310     "I", "Jail cannot see all mounted file systems");
4311 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4312     "I", "Ruleset for in-jail devfs mounts");
4313 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4314     "B", "Jail persistence");
4315 #ifdef VIMAGE
4316 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4317     "E,jailsys", "Virtual network stack");
4318 #endif
4319 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4320     "B", "Jail is in the process of shutting down");
4321 
4322 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4323 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4324     "I", "Current number of child jails");
4325 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4326     "I", "Maximum number of child jails");
4327 
4328 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4329 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4330     "Jail hostname");
4331 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4332     "Jail NIS domainname");
4333 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4334     "Jail host UUID");
4335 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4336     "LU", "Jail host ID");
4337 
4338 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4339 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4340 
4341 #ifdef INET
4342 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4343     "Jail IPv4 address virtualization");
4344 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4345     "S,in_addr,a", "Jail IPv4 addresses");
4346 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4347     "B", "Do (not) use IPv4 source address selection rather than the "
4348     "primary jail IPv4 address.");
4349 #endif
4350 #ifdef INET6
4351 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4352     "Jail IPv6 address virtualization");
4353 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4354     "S,in6_addr,a", "Jail IPv6 addresses");
4355 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4356     "B", "Do (not) use IPv6 source address selection rather than the "
4357     "primary jail IPv6 address.");
4358 #endif
4359 
4360 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4361 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4362     "B", "Jail may set hostname");
4363 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4364     "B", "Jail may use SYSV IPC");
4365 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4366     "B", "Jail may create raw sockets");
4367 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4368     "B", "Jail may alter system file flags");
4369 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4370     "B", "Jail may set file quotas");
4371 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4372     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4373 
4374 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4375 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4376     "B", "Jail may mount/unmount jail-friendly file systems in general");
4377 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4378     "B", "Jail may mount the devfs file system");
4379 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4380     "B", "Jail may mount the nullfs file system");
4381 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4382     "B", "Jail may mount the procfs file system");
4383 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4384     "B", "Jail may mount the zfs file system");
4385 
4386 void
4387 prison_racct_foreach(void (*callback)(struct racct *racct,
4388     void *arg2, void *arg3), void *arg2, void *arg3)
4389 {
4390 	struct prison_racct *prr;
4391 
4392 	sx_slock(&allprison_lock);
4393 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4394 		(callback)(prr->prr_racct, arg2, arg3);
4395 	sx_sunlock(&allprison_lock);
4396 }
4397 
4398 static struct prison_racct *
4399 prison_racct_find_locked(const char *name)
4400 {
4401 	struct prison_racct *prr;
4402 
4403 	sx_assert(&allprison_lock, SA_XLOCKED);
4404 
4405 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4406 		return (NULL);
4407 
4408 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4409 		if (strcmp(name, prr->prr_name) != 0)
4410 			continue;
4411 
4412 		/* Found prison_racct with a matching name? */
4413 		prison_racct_hold(prr);
4414 		return (prr);
4415 	}
4416 
4417 	/* Add new prison_racct. */
4418 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4419 	racct_create(&prr->prr_racct);
4420 
4421 	strcpy(prr->prr_name, name);
4422 	refcount_init(&prr->prr_refcount, 1);
4423 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4424 
4425 	return (prr);
4426 }
4427 
4428 struct prison_racct *
4429 prison_racct_find(const char *name)
4430 {
4431 	struct prison_racct *prr;
4432 
4433 	sx_xlock(&allprison_lock);
4434 	prr = prison_racct_find_locked(name);
4435 	sx_xunlock(&allprison_lock);
4436 	return (prr);
4437 }
4438 
4439 void
4440 prison_racct_hold(struct prison_racct *prr)
4441 {
4442 
4443 	refcount_acquire(&prr->prr_refcount);
4444 }
4445 
4446 static void
4447 prison_racct_free_locked(struct prison_racct *prr)
4448 {
4449 
4450 	sx_assert(&allprison_lock, SA_XLOCKED);
4451 
4452 	if (refcount_release(&prr->prr_refcount)) {
4453 		racct_destroy(&prr->prr_racct);
4454 		LIST_REMOVE(prr, prr_next);
4455 		free(prr, M_PRISON_RACCT);
4456 	}
4457 }
4458 
4459 void
4460 prison_racct_free(struct prison_racct *prr)
4461 {
4462 	int old;
4463 
4464 	sx_assert(&allprison_lock, SA_UNLOCKED);
4465 
4466 	old = prr->prr_refcount;
4467 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4468 		return;
4469 
4470 	sx_xlock(&allprison_lock);
4471 	prison_racct_free_locked(prr);
4472 	sx_xunlock(&allprison_lock);
4473 }
4474 
4475 #ifdef RACCT
4476 static void
4477 prison_racct_attach(struct prison *pr)
4478 {
4479 	struct prison_racct *prr;
4480 
4481 	sx_assert(&allprison_lock, SA_XLOCKED);
4482 
4483 	prr = prison_racct_find_locked(pr->pr_name);
4484 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4485 
4486 	pr->pr_prison_racct = prr;
4487 }
4488 
4489 /*
4490  * Handle jail renaming.  From the racct point of view, renaming means
4491  * moving from one prison_racct to another.
4492  */
4493 static void
4494 prison_racct_modify(struct prison *pr)
4495 {
4496 	struct proc *p;
4497 	struct ucred *cred;
4498 	struct prison_racct *oldprr;
4499 
4500 	sx_slock(&allproc_lock);
4501 	sx_xlock(&allprison_lock);
4502 
4503 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4504 		sx_xunlock(&allprison_lock);
4505 		sx_sunlock(&allproc_lock);
4506 		return;
4507 	}
4508 
4509 	oldprr = pr->pr_prison_racct;
4510 	pr->pr_prison_racct = NULL;
4511 
4512 	prison_racct_attach(pr);
4513 
4514 	/*
4515 	 * Move resource utilisation records.
4516 	 */
4517 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4518 
4519 	/*
4520 	 * Force rctl to reattach rules to processes.
4521 	 */
4522 	FOREACH_PROC_IN_SYSTEM(p) {
4523 		PROC_LOCK(p);
4524 		cred = crhold(p->p_ucred);
4525 		PROC_UNLOCK(p);
4526 		racct_proc_ucred_changed(p, cred, cred);
4527 		crfree(cred);
4528 	}
4529 
4530 	sx_sunlock(&allproc_lock);
4531 	prison_racct_free_locked(oldprr);
4532 	sx_xunlock(&allprison_lock);
4533 }
4534 
4535 static void
4536 prison_racct_detach(struct prison *pr)
4537 {
4538 
4539 	sx_assert(&allprison_lock, SA_UNLOCKED);
4540 
4541 	if (pr->pr_prison_racct == NULL)
4542 		return;
4543 	prison_racct_free(pr->pr_prison_racct);
4544 	pr->pr_prison_racct = NULL;
4545 }
4546 #endif /* RACCT */
4547 
4548 #ifdef DDB
4549 
4550 static void
4551 db_show_prison(struct prison *pr)
4552 {
4553 	int fi;
4554 #if defined(INET) || defined(INET6)
4555 	int ii;
4556 #endif
4557 	unsigned jsf;
4558 #ifdef INET6
4559 	char ip6buf[INET6_ADDRSTRLEN];
4560 #endif
4561 
4562 	db_printf("prison %p:\n", pr);
4563 	db_printf(" jid             = %d\n", pr->pr_id);
4564 	db_printf(" name            = %s\n", pr->pr_name);
4565 	db_printf(" parent          = %p\n", pr->pr_parent);
4566 	db_printf(" ref             = %d\n", pr->pr_ref);
4567 	db_printf(" uref            = %d\n", pr->pr_uref);
4568 	db_printf(" path            = %s\n", pr->pr_path);
4569 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4570 	    ? pr->pr_cpuset->cs_id : -1);
4571 #ifdef VIMAGE
4572 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4573 #endif
4574 	db_printf(" root            = %p\n", pr->pr_root);
4575 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4576 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4577 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4578 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4579 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4580 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4581 	db_printf(" flags           = 0x%x", pr->pr_flags);
4582 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4583 	    fi++)
4584 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4585 			db_printf(" %s", pr_flag_names[fi]);
4586 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4587 	    fi++) {
4588 		jsf = pr->pr_flags &
4589 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4590 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4591 		    pr_flag_jailsys[fi].disable &&
4592 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4593 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4594 		    : "inherit");
4595 	}
4596 	db_printf(" allow           = 0x%x", pr->pr_allow);
4597 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4598 	    fi++)
4599 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4600 			db_printf(" %s", pr_allow_names[fi]);
4601 	db_printf("\n");
4602 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4603 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4604 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4605 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4606 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4607 #ifdef INET
4608 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4609 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4610 		db_printf(" %s %s\n",
4611 		    ii == 0 ? "ip4.addr        =" : "                 ",
4612 		    inet_ntoa(pr->pr_ip4[ii]));
4613 #endif
4614 #ifdef INET6
4615 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4616 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4617 		db_printf(" %s %s\n",
4618 		    ii == 0 ? "ip6.addr        =" : "                 ",
4619 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4620 #endif
4621 }
4622 
4623 DB_SHOW_COMMAND(prison, db_show_prison_command)
4624 {
4625 	struct prison *pr;
4626 
4627 	if (!have_addr) {
4628 		/*
4629 		 * Show all prisons in the list, and prison0 which is not
4630 		 * listed.
4631 		 */
4632 		db_show_prison(&prison0);
4633 		if (!db_pager_quit) {
4634 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4635 				db_show_prison(pr);
4636 				if (db_pager_quit)
4637 					break;
4638 			}
4639 		}
4640 		return;
4641 	}
4642 
4643 	if (addr == 0)
4644 		pr = &prison0;
4645 	else {
4646 		/* Look for a prison with the ID and with references. */
4647 		TAILQ_FOREACH(pr, &allprison, pr_list)
4648 			if (pr->pr_id == addr && pr->pr_ref > 0)
4649 				break;
4650 		if (pr == NULL)
4651 			/* Look again, without requiring a reference. */
4652 			TAILQ_FOREACH(pr, &allprison, pr_list)
4653 				if (pr->pr_id == addr)
4654 					break;
4655 		if (pr == NULL)
4656 			/* Assume address points to a valid prison. */
4657 			pr = (struct prison *)addr;
4658 	}
4659 	db_show_prison(pr);
4660 }
4661 
4662 #endif /* DDB */
4663