xref: /freebsd/sys/kern/kern_jail.c (revision 7447ca0eb235974642312b9555caec00b57d8fc1)
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 #endif /* DDB */
72 
73 #include <security/mac/mac_framework.h>
74 
75 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
76 
77 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
78 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
79 
80 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
81 #ifdef INET
82 #ifdef INET6
83 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
84 #else
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
86 #endif
87 #else /* !INET */
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	0
92 #endif
93 #endif
94 
95 /* prison0 describes what is "real" about the system. */
96 struct prison prison0 = {
97 	.pr_id		= 0,
98 	.pr_name	= "0",
99 	.pr_ref		= 1,
100 	.pr_uref	= 1,
101 	.pr_path	= "/",
102 	.pr_securelevel	= -1,
103 	.pr_devfs_rsnum = 0,
104 	.pr_childmax	= JAIL_MAX,
105 	.pr_hostuuid	= DEFAULT_HOSTUUID,
106 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
107 #ifdef VIMAGE
108 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
109 #else
110 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
111 #endif
112 	.pr_allow	= PR_ALLOW_ALL,
113 };
114 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
115 
116 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
117 struct	sx allprison_lock;
118 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
119 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
120 LIST_HEAD(, prison_racct) allprison_racct;
121 int	lastprid = 0;
122 
123 static int do_jail_attach(struct thread *td, struct prison *pr);
124 static void prison_complete(void *context, int pending);
125 static void prison_deref(struct prison *pr, int flags);
126 static char *prison_path(struct prison *pr1, struct prison *pr2);
127 static void prison_remove_one(struct prison *pr);
128 #ifdef RACCT
129 static void prison_racct_attach(struct prison *pr);
130 static void prison_racct_modify(struct prison *pr);
131 static void prison_racct_detach(struct prison *pr);
132 #endif
133 
134 /* Flags for prison_deref */
135 #define	PD_DEREF	0x01
136 #define	PD_DEUREF	0x02
137 #define	PD_LOCKED	0x04
138 #define	PD_LIST_SLOCKED	0x08
139 #define	PD_LIST_XLOCKED	0x10
140 
141 /*
142  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
143  * as we cannot figure out the size of a sparse array, or an array without a
144  * terminating entry.
145  */
146 static char *pr_flag_names[] = {
147 	[0] = "persist",
148 #ifdef INET
149 	[7] = "ip4.saddrsel",
150 #endif
151 #ifdef INET6
152 	[8] = "ip6.saddrsel",
153 #endif
154 };
155 const size_t pr_flag_names_size = sizeof(pr_flag_names);
156 
157 static char *pr_flag_nonames[] = {
158 	[0] = "nopersist",
159 #ifdef INET
160 	[7] = "ip4.nosaddrsel",
161 #endif
162 #ifdef INET6
163 	[8] = "ip6.nosaddrsel",
164 #endif
165 };
166 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
167 
168 struct jailsys_flags {
169 	const char	*name;
170 	unsigned	 disable;
171 	unsigned	 new;
172 } pr_flag_jailsys[] = {
173 	{ "host", 0, PR_HOST },
174 #ifdef VIMAGE
175 	{ "vnet", 0, PR_VNET },
176 #endif
177 #ifdef INET
178 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
179 #endif
180 #ifdef INET6
181 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
182 #endif
183 };
184 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
185 
186 static char *pr_allow_names[] = {
187 	"allow.set_hostname",
188 	"allow.sysvipc",
189 	"allow.raw_sockets",
190 	"allow.chflags",
191 	"allow.mount",
192 	"allow.quotas",
193 	"allow.socket_af",
194 	"allow.mount.devfs",
195 	"allow.mount.nullfs",
196 	"allow.mount.zfs",
197 	"allow.mount.procfs",
198 	"allow.mount.tmpfs",
199 	"allow.mount.fdescfs",
200 	"allow.mount.linprocfs",
201 	"allow.mount.linsysfs",
202 };
203 const size_t pr_allow_names_size = sizeof(pr_allow_names);
204 
205 static char *pr_allow_nonames[] = {
206 	"allow.noset_hostname",
207 	"allow.nosysvipc",
208 	"allow.noraw_sockets",
209 	"allow.nochflags",
210 	"allow.nomount",
211 	"allow.noquotas",
212 	"allow.nosocket_af",
213 	"allow.mount.nodevfs",
214 	"allow.mount.nonullfs",
215 	"allow.mount.nozfs",
216 	"allow.mount.noprocfs",
217 	"allow.mount.notmpfs",
218 	"allow.mount.nofdescfs",
219 	"allow.mount.nolinprocfs",
220 	"allow.mount.nolinsysfs",
221 };
222 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
223 
224 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
225 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
226 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
227 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
228 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
229 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
230 #if defined(INET) || defined(INET6)
231 static unsigned jail_max_af_ips = 255;
232 #endif
233 
234 /*
235  * Initialize the parts of prison0 that can't be static-initialized with
236  * constants.  This is called from proc0_init() after creating thread0 cpuset.
237  */
238 void
239 prison0_init(void)
240 {
241 
242 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
243 	prison0.pr_osreldate = osreldate;
244 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
245 }
246 
247 /*
248  * struct jail_args {
249  *	struct jail *jail;
250  * };
251  */
252 int
253 sys_jail(struct thread *td, struct jail_args *uap)
254 {
255 	uint32_t version;
256 	int error;
257 	struct jail j;
258 
259 	error = copyin(uap->jail, &version, sizeof(uint32_t));
260 	if (error)
261 		return (error);
262 
263 	switch (version) {
264 	case 0:
265 	{
266 		struct jail_v0 j0;
267 
268 		/* FreeBSD single IPv4 jails. */
269 		bzero(&j, sizeof(struct jail));
270 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
271 		if (error)
272 			return (error);
273 		j.version = j0.version;
274 		j.path = j0.path;
275 		j.hostname = j0.hostname;
276 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
277 		break;
278 	}
279 
280 	case 1:
281 		/*
282 		 * Version 1 was used by multi-IPv4 jail implementations
283 		 * that never made it into the official kernel.
284 		 */
285 		return (EINVAL);
286 
287 	case 2:	/* JAIL_API_VERSION */
288 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
289 		error = copyin(uap->jail, &j, sizeof(struct jail));
290 		if (error)
291 			return (error);
292 		break;
293 
294 	default:
295 		/* Sci-Fi jails are not supported, sorry. */
296 		return (EINVAL);
297 	}
298 	return (kern_jail(td, &j));
299 }
300 
301 int
302 kern_jail(struct thread *td, struct jail *j)
303 {
304 	struct iovec optiov[2 * (4 + nitems(pr_allow_names)
305 #ifdef INET
306 			    + 1
307 #endif
308 #ifdef INET6
309 			    + 1
310 #endif
311 			    )];
312 	struct uio opt;
313 	char *u_path, *u_hostname, *u_name;
314 #ifdef INET
315 	uint32_t ip4s;
316 	struct in_addr *u_ip4;
317 #endif
318 #ifdef INET6
319 	struct in6_addr *u_ip6;
320 #endif
321 	size_t tmplen;
322 	int error, enforce_statfs, fi;
323 
324 	bzero(&optiov, sizeof(optiov));
325 	opt.uio_iov = optiov;
326 	opt.uio_iovcnt = 0;
327 	opt.uio_offset = -1;
328 	opt.uio_resid = -1;
329 	opt.uio_segflg = UIO_SYSSPACE;
330 	opt.uio_rw = UIO_READ;
331 	opt.uio_td = td;
332 
333 	/* Set permissions for top-level jails from sysctls. */
334 	if (!jailed(td->td_ucred)) {
335 		for (fi = 0; fi < nitems(pr_allow_names); fi++) {
336 			optiov[opt.uio_iovcnt].iov_base =
337 			    (jail_default_allow & (1 << fi))
338 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
339 			optiov[opt.uio_iovcnt].iov_len =
340 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
341 			opt.uio_iovcnt += 2;
342 		}
343 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
344 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
345 		opt.uio_iovcnt++;
346 		enforce_statfs = jail_default_enforce_statfs;
347 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
348 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
349 		opt.uio_iovcnt++;
350 	}
351 
352 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
353 #ifdef INET
354 	ip4s = (j->version == 0) ? 1 : j->ip4s;
355 	if (ip4s > jail_max_af_ips)
356 		return (EINVAL);
357 	tmplen += ip4s * sizeof(struct in_addr);
358 #else
359 	if (j->ip4s > 0)
360 		return (EINVAL);
361 #endif
362 #ifdef INET6
363 	if (j->ip6s > jail_max_af_ips)
364 		return (EINVAL);
365 	tmplen += j->ip6s * sizeof(struct in6_addr);
366 #else
367 	if (j->ip6s > 0)
368 		return (EINVAL);
369 #endif
370 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
371 	u_hostname = u_path + MAXPATHLEN;
372 	u_name = u_hostname + MAXHOSTNAMELEN;
373 #ifdef INET
374 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
375 #endif
376 #ifdef INET6
377 #ifdef INET
378 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
379 #else
380 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
381 #endif
382 #endif
383 	optiov[opt.uio_iovcnt].iov_base = "path";
384 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
385 	opt.uio_iovcnt++;
386 	optiov[opt.uio_iovcnt].iov_base = u_path;
387 	error = copyinstr(j->path, u_path, MAXPATHLEN,
388 	    &optiov[opt.uio_iovcnt].iov_len);
389 	if (error) {
390 		free(u_path, M_TEMP);
391 		return (error);
392 	}
393 	opt.uio_iovcnt++;
394 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
395 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
396 	opt.uio_iovcnt++;
397 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
398 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
399 	    &optiov[opt.uio_iovcnt].iov_len);
400 	if (error) {
401 		free(u_path, M_TEMP);
402 		return (error);
403 	}
404 	opt.uio_iovcnt++;
405 	if (j->jailname != NULL) {
406 		optiov[opt.uio_iovcnt].iov_base = "name";
407 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
408 		opt.uio_iovcnt++;
409 		optiov[opt.uio_iovcnt].iov_base = u_name;
410 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
411 		    &optiov[opt.uio_iovcnt].iov_len);
412 		if (error) {
413 			free(u_path, M_TEMP);
414 			return (error);
415 		}
416 		opt.uio_iovcnt++;
417 	}
418 #ifdef INET
419 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
420 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
421 	opt.uio_iovcnt++;
422 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
423 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
424 	if (j->version == 0)
425 		u_ip4->s_addr = j->ip4s;
426 	else {
427 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
428 		if (error) {
429 			free(u_path, M_TEMP);
430 			return (error);
431 		}
432 	}
433 	opt.uio_iovcnt++;
434 #endif
435 #ifdef INET6
436 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
437 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
438 	opt.uio_iovcnt++;
439 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
440 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
441 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
442 	if (error) {
443 		free(u_path, M_TEMP);
444 		return (error);
445 	}
446 	opt.uio_iovcnt++;
447 #endif
448 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
449 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
450 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
451 	free(u_path, M_TEMP);
452 	return (error);
453 }
454 
455 
456 /*
457  * struct jail_set_args {
458  *	struct iovec *iovp;
459  *	unsigned int iovcnt;
460  *	int flags;
461  * };
462  */
463 int
464 sys_jail_set(struct thread *td, struct jail_set_args *uap)
465 {
466 	struct uio *auio;
467 	int error;
468 
469 	/* Check that we have an even number of iovecs. */
470 	if (uap->iovcnt & 1)
471 		return (EINVAL);
472 
473 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
474 	if (error)
475 		return (error);
476 	error = kern_jail_set(td, auio, uap->flags);
477 	free(auio, M_IOV);
478 	return (error);
479 }
480 
481 int
482 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
483 {
484 	struct nameidata nd;
485 #ifdef INET
486 	struct in_addr *ip4;
487 #endif
488 #ifdef INET6
489 	struct in6_addr *ip6;
490 #endif
491 	struct vfsopt *opt;
492 	struct vfsoptlist *opts;
493 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
494 	struct vnode *root;
495 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
496 	char *g_path, *osrelstr;
497 #if defined(INET) || defined(INET6)
498 	struct prison *tppr;
499 	void *op;
500 #endif
501 	unsigned long hid;
502 	size_t namelen, onamelen, pnamelen;
503 	int born, created, cuflags, descend, enforce;
504 	int error, errmsg_len, errmsg_pos;
505 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
506 	int fi, jid, jsys, len, level;
507 	int childmax, osreldt, rsnum, slevel;
508 	int fullpath_disabled;
509 #if defined(INET) || defined(INET6)
510 	int ii, ij;
511 #endif
512 #ifdef INET
513 	int ip4s, redo_ip4;
514 #endif
515 #ifdef INET6
516 	int ip6s, redo_ip6;
517 #endif
518 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
519 	unsigned tallow;
520 	char numbuf[12];
521 
522 	error = priv_check(td, PRIV_JAIL_SET);
523 	if (!error && (flags & JAIL_ATTACH))
524 		error = priv_check(td, PRIV_JAIL_ATTACH);
525 	if (error)
526 		return (error);
527 	mypr = td->td_ucred->cr_prison;
528 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
529 		return (EPERM);
530 	if (flags & ~JAIL_SET_MASK)
531 		return (EINVAL);
532 
533 	/*
534 	 * Check all the parameters before committing to anything.  Not all
535 	 * errors can be caught early, but we may as well try.  Also, this
536 	 * takes care of some expensive stuff (path lookup) before getting
537 	 * the allprison lock.
538 	 *
539 	 * XXX Jails are not filesystems, and jail parameters are not mount
540 	 *     options.  But it makes more sense to re-use the vfsopt code
541 	 *     than duplicate it under a different name.
542 	 */
543 	error = vfs_buildopts(optuio, &opts);
544 	if (error)
545 		return (error);
546 #ifdef INET
547 	ip4 = NULL;
548 #endif
549 #ifdef INET6
550 	ip6 = NULL;
551 #endif
552 	g_path = NULL;
553 
554 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
555 	if (!cuflags) {
556 		error = EINVAL;
557 		vfs_opterror(opts, "no valid operation (create or update)");
558 		goto done_errmsg;
559 	}
560 
561 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
562 	if (error == ENOENT)
563 		jid = 0;
564 	else if (error != 0)
565 		goto done_free;
566 
567 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
568 	if (error == ENOENT)
569 		gotslevel = 0;
570 	else if (error != 0)
571 		goto done_free;
572 	else
573 		gotslevel = 1;
574 
575 	error =
576 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
577 	if (error == ENOENT)
578 		gotchildmax = 0;
579 	else if (error != 0)
580 		goto done_free;
581 	else
582 		gotchildmax = 1;
583 
584 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
585 	if (error == ENOENT)
586 		gotenforce = 0;
587 	else if (error != 0)
588 		goto done_free;
589 	else if (enforce < 0 || enforce > 2) {
590 		error = EINVAL;
591 		goto done_free;
592 	} else
593 		gotenforce = 1;
594 
595 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
596 	if (error == ENOENT)
597 		gotrsnum = 0;
598 	else if (error != 0)
599 		goto done_free;
600 	else
601 		gotrsnum = 1;
602 
603 	pr_flags = ch_flags = 0;
604 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
605 		if (pr_flag_names[fi] == NULL)
606 			continue;
607 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
608 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
609 	}
610 	ch_flags |= pr_flags;
611 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
612 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
613 		    sizeof(jsys));
614 		if (error == ENOENT)
615 			continue;
616 		if (error != 0)
617 			goto done_free;
618 		switch (jsys) {
619 		case JAIL_SYS_DISABLE:
620 			if (!pr_flag_jailsys[fi].disable) {
621 				error = EINVAL;
622 				goto done_free;
623 			}
624 			pr_flags |= pr_flag_jailsys[fi].disable;
625 			break;
626 		case JAIL_SYS_NEW:
627 			pr_flags |= pr_flag_jailsys[fi].new;
628 			break;
629 		case JAIL_SYS_INHERIT:
630 			break;
631 		default:
632 			error = EINVAL;
633 			goto done_free;
634 		}
635 		ch_flags |=
636 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
637 	}
638 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
639 	    && !(pr_flags & PR_PERSIST)) {
640 		error = EINVAL;
641 		vfs_opterror(opts, "new jail must persist or attach");
642 		goto done_errmsg;
643 	}
644 #ifdef VIMAGE
645 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
646 		error = EINVAL;
647 		vfs_opterror(opts, "vnet cannot be changed after creation");
648 		goto done_errmsg;
649 	}
650 #endif
651 #ifdef INET
652 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
653 		error = EINVAL;
654 		vfs_opterror(opts, "ip4 cannot be changed after creation");
655 		goto done_errmsg;
656 	}
657 #endif
658 #ifdef INET6
659 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
660 		error = EINVAL;
661 		vfs_opterror(opts, "ip6 cannot be changed after creation");
662 		goto done_errmsg;
663 	}
664 #endif
665 
666 	pr_allow = ch_allow = 0;
667 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
668 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
669 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
670 	}
671 	ch_allow |= pr_allow;
672 
673 	error = vfs_getopt(opts, "name", (void **)&name, &len);
674 	if (error == ENOENT)
675 		name = NULL;
676 	else if (error != 0)
677 		goto done_free;
678 	else {
679 		if (len == 0 || name[len - 1] != '\0') {
680 			error = EINVAL;
681 			goto done_free;
682 		}
683 		if (len > MAXHOSTNAMELEN) {
684 			error = ENAMETOOLONG;
685 			goto done_free;
686 		}
687 	}
688 
689 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
690 	if (error == ENOENT)
691 		host = NULL;
692 	else if (error != 0)
693 		goto done_free;
694 	else {
695 		ch_flags |= PR_HOST;
696 		pr_flags |= PR_HOST;
697 		if (len == 0 || host[len - 1] != '\0') {
698 			error = EINVAL;
699 			goto done_free;
700 		}
701 		if (len > MAXHOSTNAMELEN) {
702 			error = ENAMETOOLONG;
703 			goto done_free;
704 		}
705 	}
706 
707 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
708 	if (error == ENOENT)
709 		domain = NULL;
710 	else if (error != 0)
711 		goto done_free;
712 	else {
713 		ch_flags |= PR_HOST;
714 		pr_flags |= PR_HOST;
715 		if (len == 0 || domain[len - 1] != '\0') {
716 			error = EINVAL;
717 			goto done_free;
718 		}
719 		if (len > MAXHOSTNAMELEN) {
720 			error = ENAMETOOLONG;
721 			goto done_free;
722 		}
723 	}
724 
725 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
726 	if (error == ENOENT)
727 		uuid = NULL;
728 	else if (error != 0)
729 		goto done_free;
730 	else {
731 		ch_flags |= PR_HOST;
732 		pr_flags |= PR_HOST;
733 		if (len == 0 || uuid[len - 1] != '\0') {
734 			error = EINVAL;
735 			goto done_free;
736 		}
737 		if (len > HOSTUUIDLEN) {
738 			error = ENAMETOOLONG;
739 			goto done_free;
740 		}
741 	}
742 
743 #ifdef COMPAT_FREEBSD32
744 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
745 		uint32_t hid32;
746 
747 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
748 		hid = hid32;
749 	} else
750 #endif
751 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
752 	if (error == ENOENT)
753 		gothid = 0;
754 	else if (error != 0)
755 		goto done_free;
756 	else {
757 		gothid = 1;
758 		ch_flags |= PR_HOST;
759 		pr_flags |= PR_HOST;
760 	}
761 
762 #ifdef INET
763 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
764 	if (error == ENOENT)
765 		ip4s = 0;
766 	else if (error != 0)
767 		goto done_free;
768 	else if (ip4s & (sizeof(*ip4) - 1)) {
769 		error = EINVAL;
770 		goto done_free;
771 	} else {
772 		ch_flags |= PR_IP4_USER;
773 		pr_flags |= PR_IP4_USER;
774 		if (ip4s > 0) {
775 			ip4s /= sizeof(*ip4);
776 			if (ip4s > jail_max_af_ips) {
777 				error = EINVAL;
778 				vfs_opterror(opts, "too many IPv4 addresses");
779 				goto done_errmsg;
780 			}
781 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
782 			bcopy(op, ip4, ip4s * sizeof(*ip4));
783 			/*
784 			 * IP addresses are all sorted but ip[0] to preserve
785 			 * the primary IP address as given from userland.
786 			 * This special IP is used for unbound outgoing
787 			 * connections as well for "loopback" traffic in case
788 			 * source address selection cannot find any more fitting
789 			 * address to connect from.
790 			 */
791 			if (ip4s > 1)
792 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
793 				    prison_qcmp_v4);
794 			/*
795 			 * Check for duplicate addresses and do some simple
796 			 * zero and broadcast checks. If users give other bogus
797 			 * addresses it is their problem.
798 			 *
799 			 * We do not have to care about byte order for these
800 			 * checks so we will do them in NBO.
801 			 */
802 			for (ii = 0; ii < ip4s; ii++) {
803 				if (ip4[ii].s_addr == INADDR_ANY ||
804 				    ip4[ii].s_addr == INADDR_BROADCAST) {
805 					error = EINVAL;
806 					goto done_free;
807 				}
808 				if ((ii+1) < ip4s &&
809 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
810 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
811 					error = EINVAL;
812 					goto done_free;
813 				}
814 			}
815 		}
816 	}
817 #endif
818 
819 #ifdef INET6
820 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
821 	if (error == ENOENT)
822 		ip6s = 0;
823 	else if (error != 0)
824 		goto done_free;
825 	else if (ip6s & (sizeof(*ip6) - 1)) {
826 		error = EINVAL;
827 		goto done_free;
828 	} else {
829 		ch_flags |= PR_IP6_USER;
830 		pr_flags |= PR_IP6_USER;
831 		if (ip6s > 0) {
832 			ip6s /= sizeof(*ip6);
833 			if (ip6s > jail_max_af_ips) {
834 				error = EINVAL;
835 				vfs_opterror(opts, "too many IPv6 addresses");
836 				goto done_errmsg;
837 			}
838 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
839 			bcopy(op, ip6, ip6s * sizeof(*ip6));
840 			if (ip6s > 1)
841 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
842 				    prison_qcmp_v6);
843 			for (ii = 0; ii < ip6s; ii++) {
844 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
845 					error = EINVAL;
846 					goto done_free;
847 				}
848 				if ((ii+1) < ip6s &&
849 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
850 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
851 				{
852 					error = EINVAL;
853 					goto done_free;
854 				}
855 			}
856 		}
857 	}
858 #endif
859 
860 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
861 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
862 		error = EINVAL;
863 		vfs_opterror(opts,
864 		    "vnet jails cannot have IP address restrictions");
865 		goto done_errmsg;
866 	}
867 #endif
868 
869 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
870 	if (error == ENOENT)
871 		osrelstr = NULL;
872 	else if (error != 0)
873 		goto done_free;
874 	else {
875 		if (flags & JAIL_UPDATE) {
876 			error = EINVAL;
877 			vfs_opterror(opts,
878 			    "osrelease cannot be changed after creation");
879 			goto done_errmsg;
880 		}
881 		if (len == 0 || len >= OSRELEASELEN) {
882 			error = EINVAL;
883 			vfs_opterror(opts,
884 			    "osrelease string must be 1-%d bytes long",
885 			    OSRELEASELEN - 1);
886 			goto done_errmsg;
887 		}
888 	}
889 
890 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
891 	if (error == ENOENT)
892 		osreldt = 0;
893 	else if (error != 0)
894 		goto done_free;
895 	else {
896 		if (flags & JAIL_UPDATE) {
897 			error = EINVAL;
898 			vfs_opterror(opts,
899 			    "osreldate cannot be changed after creation");
900 			goto done_errmsg;
901 		}
902 		if (osreldt == 0) {
903 			error = EINVAL;
904 			vfs_opterror(opts, "osreldate cannot be 0");
905 			goto done_errmsg;
906 		}
907 	}
908 
909 	fullpath_disabled = 0;
910 	root = NULL;
911 	error = vfs_getopt(opts, "path", (void **)&path, &len);
912 	if (error == ENOENT)
913 		path = NULL;
914 	else if (error != 0)
915 		goto done_free;
916 	else {
917 		if (flags & JAIL_UPDATE) {
918 			error = EINVAL;
919 			vfs_opterror(opts,
920 			    "path cannot be changed after creation");
921 			goto done_errmsg;
922 		}
923 		if (len == 0 || path[len - 1] != '\0') {
924 			error = EINVAL;
925 			goto done_free;
926 		}
927 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
928 		    path, td);
929 		error = namei(&nd);
930 		if (error)
931 			goto done_free;
932 		root = nd.ni_vp;
933 		NDFREE(&nd, NDF_ONLY_PNBUF);
934 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
935 		strlcpy(g_path, path, MAXPATHLEN);
936 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
937 		if (error == 0)
938 			path = g_path;
939 		else if (error == ENODEV) {
940 			/* proceed if sysctl debug.disablefullpath == 1 */
941 			fullpath_disabled = 1;
942 			if (len < 2 || (len == 2 && path[0] == '/'))
943 				path = NULL;
944 		} else {
945 			/* exit on other errors */
946 			goto done_free;
947 		}
948 		if (root->v_type != VDIR) {
949 			error = ENOTDIR;
950 			vput(root);
951 			goto done_free;
952 		}
953 		VOP_UNLOCK(root, 0);
954 		if (fullpath_disabled) {
955 			/* Leave room for a real-root full pathname. */
956 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
957 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
958 				error = ENAMETOOLONG;
959 				vrele(root);
960 				goto done_free;
961 			}
962 		}
963 	}
964 
965 	/*
966 	 * Find the specified jail, or at least its parent.
967 	 * This abuses the file error codes ENOENT and EEXIST.
968 	 */
969 	pr = NULL;
970 	ppr = mypr;
971 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
972 		namelc = strrchr(name, '.');
973 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
974 		if (*p != '\0')
975 			jid = 0;
976 	}
977 	sx_xlock(&allprison_lock);
978 	if (jid != 0) {
979 		/*
980 		 * See if a requested jid already exists.  There is an
981 		 * information leak here if the jid exists but is not within
982 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
983 		 * even though they cannot see the jail, and CREATE | UPDATE
984 		 * will return ENOENT which is not normally a valid error.
985 		 */
986 		if (jid < 0) {
987 			error = EINVAL;
988 			vfs_opterror(opts, "negative jid");
989 			goto done_unlock_list;
990 		}
991 		pr = prison_find(jid);
992 		if (pr != NULL) {
993 			ppr = pr->pr_parent;
994 			/* Create: jid must not exist. */
995 			if (cuflags == JAIL_CREATE) {
996 				mtx_unlock(&pr->pr_mtx);
997 				error = EEXIST;
998 				vfs_opterror(opts, "jail %d already exists",
999 				    jid);
1000 				goto done_unlock_list;
1001 			}
1002 			if (!prison_ischild(mypr, pr)) {
1003 				mtx_unlock(&pr->pr_mtx);
1004 				pr = NULL;
1005 			} else if (pr->pr_uref == 0) {
1006 				if (!(flags & JAIL_DYING)) {
1007 					mtx_unlock(&pr->pr_mtx);
1008 					error = ENOENT;
1009 					vfs_opterror(opts, "jail %d is dying",
1010 					    jid);
1011 					goto done_unlock_list;
1012 				} else if ((flags & JAIL_ATTACH) ||
1013 				    (pr_flags & PR_PERSIST)) {
1014 					/*
1015 					 * A dying jail might be resurrected
1016 					 * (via attach or persist), but first
1017 					 * it must determine if another jail
1018 					 * has claimed its name.  Accomplish
1019 					 * this by implicitly re-setting the
1020 					 * name.
1021 					 */
1022 					if (name == NULL)
1023 						name = prison_name(mypr, pr);
1024 				}
1025 			}
1026 		}
1027 		if (pr == NULL) {
1028 			/* Update: jid must exist. */
1029 			if (cuflags == JAIL_UPDATE) {
1030 				error = ENOENT;
1031 				vfs_opterror(opts, "jail %d not found", jid);
1032 				goto done_unlock_list;
1033 			}
1034 		}
1035 	}
1036 	/*
1037 	 * If the caller provided a name, look for a jail by that name.
1038 	 * This has different semantics for creates and updates keyed by jid
1039 	 * (where the name must not already exist in a different jail),
1040 	 * and updates keyed by the name itself (where the name must exist
1041 	 * because that is the jail being updated).
1042 	 */
1043 	namelc = NULL;
1044 	if (name != NULL) {
1045 		namelc = strrchr(name, '.');
1046 		if (namelc == NULL)
1047 			namelc = name;
1048 		else {
1049 			/*
1050 			 * This is a hierarchical name.  Split it into the
1051 			 * parent and child names, and make sure the parent
1052 			 * exists or matches an already found jail.
1053 			 */
1054 			if (pr != NULL) {
1055 				if (strncmp(name, ppr->pr_name, namelc - name)
1056 				    || ppr->pr_name[namelc - name] != '\0') {
1057 					mtx_unlock(&pr->pr_mtx);
1058 					error = EINVAL;
1059 					vfs_opterror(opts,
1060 					    "cannot change jail's parent");
1061 					goto done_unlock_list;
1062 				}
1063 			} else {
1064 				*namelc = '\0';
1065 				ppr = prison_find_name(mypr, name);
1066 				if (ppr == NULL) {
1067 					error = ENOENT;
1068 					vfs_opterror(opts,
1069 					    "jail \"%s\" not found", name);
1070 					goto done_unlock_list;
1071 				}
1072 				mtx_unlock(&ppr->pr_mtx);
1073 				*namelc = '.';
1074 			}
1075 			namelc++;
1076 		}
1077 		if (namelc[0] != '\0') {
1078 			pnamelen =
1079 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1080  name_again:
1081 			deadpr = NULL;
1082 			FOREACH_PRISON_CHILD(ppr, tpr) {
1083 				if (tpr != pr && tpr->pr_ref > 0 &&
1084 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1085 					if (pr == NULL &&
1086 					    cuflags != JAIL_CREATE) {
1087 						mtx_lock(&tpr->pr_mtx);
1088 						if (tpr->pr_ref > 0) {
1089 							/*
1090 							 * Use this jail
1091 							 * for updates.
1092 							 */
1093 							if (tpr->pr_uref > 0) {
1094 								pr = tpr;
1095 								break;
1096 							}
1097 							deadpr = tpr;
1098 						}
1099 						mtx_unlock(&tpr->pr_mtx);
1100 					} else if (tpr->pr_uref > 0) {
1101 						/*
1102 						 * Create, or update(jid):
1103 						 * name must not exist in an
1104 						 * active sibling jail.
1105 						 */
1106 						error = EEXIST;
1107 						if (pr != NULL)
1108 							mtx_unlock(&pr->pr_mtx);
1109 						vfs_opterror(opts,
1110 						   "jail \"%s\" already exists",
1111 						   name);
1112 						goto done_unlock_list;
1113 					}
1114 				}
1115 			}
1116 			/* If no active jail is found, use a dying one. */
1117 			if (deadpr != NULL && pr == NULL) {
1118 				if (flags & JAIL_DYING) {
1119 					mtx_lock(&deadpr->pr_mtx);
1120 					if (deadpr->pr_ref == 0) {
1121 						mtx_unlock(&deadpr->pr_mtx);
1122 						goto name_again;
1123 					}
1124 					pr = deadpr;
1125 				} else if (cuflags == JAIL_UPDATE) {
1126 					error = ENOENT;
1127 					vfs_opterror(opts,
1128 					    "jail \"%s\" is dying", name);
1129 					goto done_unlock_list;
1130 				}
1131 			}
1132 			/* Update: name must exist if no jid. */
1133 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1134 				error = ENOENT;
1135 				vfs_opterror(opts, "jail \"%s\" not found",
1136 				    name);
1137 				goto done_unlock_list;
1138 			}
1139 		}
1140 	}
1141 	/* Update: must provide a jid or name. */
1142 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1143 		error = ENOENT;
1144 		vfs_opterror(opts, "update specified no jail");
1145 		goto done_unlock_list;
1146 	}
1147 
1148 	/* If there's no prison to update, create a new one and link it in. */
1149 	if (pr == NULL) {
1150 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1151 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1152 				error = EPERM;
1153 				vfs_opterror(opts, "prison limit exceeded");
1154 				goto done_unlock_list;
1155 			}
1156 		created = 1;
1157 		mtx_lock(&ppr->pr_mtx);
1158 		if (ppr->pr_ref == 0) {
1159 			mtx_unlock(&ppr->pr_mtx);
1160 			error = ENOENT;
1161 			vfs_opterror(opts, "jail \"%s\" not found",
1162 			    prison_name(mypr, ppr));
1163 			goto done_unlock_list;
1164 		}
1165 		ppr->pr_ref++;
1166 		ppr->pr_uref++;
1167 		mtx_unlock(&ppr->pr_mtx);
1168 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1169 		if (jid == 0) {
1170 			/* Find the next free jid. */
1171 			jid = lastprid + 1;
1172  findnext:
1173 			if (jid == JAIL_MAX)
1174 				jid = 1;
1175 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1176 				if (tpr->pr_id < jid)
1177 					continue;
1178 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1179 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1180 					break;
1181 				}
1182 				if (jid == lastprid) {
1183 					error = EAGAIN;
1184 					vfs_opterror(opts,
1185 					    "no available jail IDs");
1186 					free(pr, M_PRISON);
1187 					prison_deref(ppr, PD_DEREF |
1188 					    PD_DEUREF | PD_LIST_XLOCKED);
1189 					goto done_releroot;
1190 				}
1191 				jid++;
1192 				goto findnext;
1193 			}
1194 			lastprid = jid;
1195 		} else {
1196 			/*
1197 			 * The jail already has a jid (that did not yet exist),
1198 			 * so just find where to insert it.
1199 			 */
1200 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1201 				if (tpr->pr_id >= jid) {
1202 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1203 					break;
1204 				}
1205 		}
1206 		if (tpr == NULL)
1207 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1208 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1209 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1210 			tpr->pr_childcount++;
1211 
1212 		pr->pr_parent = ppr;
1213 		pr->pr_id = jid;
1214 
1215 		/* Set some default values, and inherit some from the parent. */
1216 		if (namelc == NULL)
1217 			namelc = "";
1218 		if (path == NULL) {
1219 			path = "/";
1220 			root = mypr->pr_root;
1221 			vref(root);
1222 		}
1223 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1224 		pr->pr_flags |= PR_HOST;
1225 #if defined(INET) || defined(INET6)
1226 #ifdef VIMAGE
1227 		if (!(pr_flags & PR_VNET))
1228 #endif
1229 		{
1230 #ifdef INET
1231 			if (!(ch_flags & PR_IP4_USER))
1232 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1233 			else if (!(pr_flags & PR_IP4_USER)) {
1234 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1235 				if (ppr->pr_ip4 != NULL) {
1236 					pr->pr_ip4s = ppr->pr_ip4s;
1237 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1238 					    sizeof(struct in_addr), M_PRISON,
1239 					    M_WAITOK);
1240 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1241 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1242 				}
1243 			}
1244 #endif
1245 #ifdef INET6
1246 			if (!(ch_flags & PR_IP6_USER))
1247 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1248 			else if (!(pr_flags & PR_IP6_USER)) {
1249 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1250 				if (ppr->pr_ip6 != NULL) {
1251 					pr->pr_ip6s = ppr->pr_ip6s;
1252 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1253 					    sizeof(struct in6_addr), M_PRISON,
1254 					    M_WAITOK);
1255 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1256 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1257 				}
1258 			}
1259 #endif
1260 		}
1261 #endif
1262 		/* Source address selection is always on by default. */
1263 		pr->pr_flags |= _PR_IP_SADDRSEL;
1264 
1265 		pr->pr_securelevel = ppr->pr_securelevel;
1266 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1267 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1268 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1269 
1270 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1271 		if (osrelstr == NULL)
1272 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1273 		else
1274 		    strcpy(pr->pr_osrelease, osrelstr);
1275 
1276 		LIST_INIT(&pr->pr_children);
1277 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1278 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1279 
1280 #ifdef VIMAGE
1281 		/* Allocate a new vnet if specified. */
1282 		pr->pr_vnet = (pr_flags & PR_VNET)
1283 		    ? vnet_alloc() : ppr->pr_vnet;
1284 #endif
1285 		/*
1286 		 * Allocate a dedicated cpuset for each jail.
1287 		 * Unlike other initial settings, this may return an erorr.
1288 		 */
1289 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1290 		if (error) {
1291 			prison_deref(pr, PD_LIST_XLOCKED);
1292 			goto done_releroot;
1293 		}
1294 
1295 		mtx_lock(&pr->pr_mtx);
1296 		/*
1297 		 * New prisons do not yet have a reference, because we do not
1298 		 * want others to see the incomplete prison once the
1299 		 * allprison_lock is downgraded.
1300 		 */
1301 	} else {
1302 		created = 0;
1303 		/*
1304 		 * Grab a reference for existing prisons, to ensure they
1305 		 * continue to exist for the duration of the call.
1306 		 */
1307 		pr->pr_ref++;
1308 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1309 		if ((pr->pr_flags & PR_VNET) &&
1310 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1311 			error = EINVAL;
1312 			vfs_opterror(opts,
1313 			    "vnet jails cannot have IP address restrictions");
1314 			goto done_deref_locked;
1315 		}
1316 #endif
1317 #ifdef INET
1318 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1319 			error = EINVAL;
1320 			vfs_opterror(opts,
1321 			    "ip4 cannot be changed after creation");
1322 			goto done_deref_locked;
1323 		}
1324 #endif
1325 #ifdef INET6
1326 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1327 			error = EINVAL;
1328 			vfs_opterror(opts,
1329 			    "ip6 cannot be changed after creation");
1330 			goto done_deref_locked;
1331 		}
1332 #endif
1333 	}
1334 
1335 	/* Do final error checking before setting anything. */
1336 	if (gotslevel) {
1337 		if (slevel < ppr->pr_securelevel) {
1338 			error = EPERM;
1339 			goto done_deref_locked;
1340 		}
1341 	}
1342 	if (gotchildmax) {
1343 		if (childmax >= ppr->pr_childmax) {
1344 			error = EPERM;
1345 			goto done_deref_locked;
1346 		}
1347 	}
1348 	if (gotenforce) {
1349 		if (enforce < ppr->pr_enforce_statfs) {
1350 			error = EPERM;
1351 			goto done_deref_locked;
1352 		}
1353 	}
1354 	if (gotrsnum) {
1355 		/*
1356 		 * devfs_rsnum is a uint16_t
1357 		 */
1358 		if (rsnum < 0 || rsnum > 65535) {
1359 			error = EINVAL;
1360 			goto done_deref_locked;
1361 		}
1362 		/*
1363 		 * Nested jails always inherit parent's devfs ruleset
1364 		 */
1365 		if (jailed(td->td_ucred)) {
1366 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1367 				error = EPERM;
1368 				goto done_deref_locked;
1369 			} else
1370 				rsnum = ppr->pr_devfs_rsnum;
1371 		}
1372 	}
1373 #ifdef INET
1374 	if (ip4s > 0) {
1375 		if (ppr->pr_flags & PR_IP4) {
1376 			/*
1377 			 * Make sure the new set of IP addresses is a
1378 			 * subset of the parent's list.  Don't worry
1379 			 * about the parent being unlocked, as any
1380 			 * setting is done with allprison_lock held.
1381 			 */
1382 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1383 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1384 					break;
1385 			if (ij == ppr->pr_ip4s) {
1386 				error = EPERM;
1387 				goto done_deref_locked;
1388 			}
1389 			if (ip4s > 1) {
1390 				for (ii = ij = 1; ii < ip4s; ii++) {
1391 					if (ip4[ii].s_addr ==
1392 					    ppr->pr_ip4[0].s_addr)
1393 						continue;
1394 					for (; ij < ppr->pr_ip4s; ij++)
1395 						if (ip4[ii].s_addr ==
1396 						    ppr->pr_ip4[ij].s_addr)
1397 							break;
1398 					if (ij == ppr->pr_ip4s)
1399 						break;
1400 				}
1401 				if (ij == ppr->pr_ip4s) {
1402 					error = EPERM;
1403 					goto done_deref_locked;
1404 				}
1405 			}
1406 		}
1407 		/*
1408 		 * Check for conflicting IP addresses.  We permit them
1409 		 * if there is no more than one IP on each jail.  If
1410 		 * there is a duplicate on a jail with more than one
1411 		 * IP stop checking and return error.
1412 		 */
1413 		tppr = ppr;
1414 #ifdef VIMAGE
1415 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1416 			if (tppr->pr_flags & PR_VNET)
1417 				break;
1418 #endif
1419 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1420 			if (tpr == pr ||
1421 #ifdef VIMAGE
1422 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1423 #endif
1424 			    tpr->pr_uref == 0) {
1425 				descend = 0;
1426 				continue;
1427 			}
1428 			if (!(tpr->pr_flags & PR_IP4_USER))
1429 				continue;
1430 			descend = 0;
1431 			if (tpr->pr_ip4 == NULL ||
1432 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1433 				continue;
1434 			for (ii = 0; ii < ip4s; ii++) {
1435 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1436 				    0) {
1437 					error = EADDRINUSE;
1438 					vfs_opterror(opts,
1439 					    "IPv4 addresses clash");
1440 					goto done_deref_locked;
1441 				}
1442 			}
1443 		}
1444 	}
1445 #endif
1446 #ifdef INET6
1447 	if (ip6s > 0) {
1448 		if (ppr->pr_flags & PR_IP6) {
1449 			/*
1450 			 * Make sure the new set of IP addresses is a
1451 			 * subset of the parent's list.
1452 			 */
1453 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1454 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1455 				    &ppr->pr_ip6[ij]))
1456 					break;
1457 			if (ij == ppr->pr_ip6s) {
1458 				error = EPERM;
1459 				goto done_deref_locked;
1460 			}
1461 			if (ip6s > 1) {
1462 				for (ii = ij = 1; ii < ip6s; ii++) {
1463 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1464 					     &ppr->pr_ip6[0]))
1465 						continue;
1466 					for (; ij < ppr->pr_ip6s; ij++)
1467 						if (IN6_ARE_ADDR_EQUAL(
1468 						    &ip6[ii], &ppr->pr_ip6[ij]))
1469 							break;
1470 					if (ij == ppr->pr_ip6s)
1471 						break;
1472 				}
1473 				if (ij == ppr->pr_ip6s) {
1474 					error = EPERM;
1475 					goto done_deref_locked;
1476 				}
1477 			}
1478 		}
1479 		/* Check for conflicting IP addresses. */
1480 		tppr = ppr;
1481 #ifdef VIMAGE
1482 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1483 			if (tppr->pr_flags & PR_VNET)
1484 				break;
1485 #endif
1486 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1487 			if (tpr == pr ||
1488 #ifdef VIMAGE
1489 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1490 #endif
1491 			    tpr->pr_uref == 0) {
1492 				descend = 0;
1493 				continue;
1494 			}
1495 			if (!(tpr->pr_flags & PR_IP6_USER))
1496 				continue;
1497 			descend = 0;
1498 			if (tpr->pr_ip6 == NULL ||
1499 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1500 				continue;
1501 			for (ii = 0; ii < ip6s; ii++) {
1502 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1503 				    0) {
1504 					error = EADDRINUSE;
1505 					vfs_opterror(opts,
1506 					    "IPv6 addresses clash");
1507 					goto done_deref_locked;
1508 				}
1509 			}
1510 		}
1511 	}
1512 #endif
1513 	onamelen = namelen = 0;
1514 	if (namelc != NULL) {
1515 		/* Give a default name of the jid.  Also allow the name to be
1516 		 * explicitly the jid - but not any other number, and only in
1517 		 * normal form (no leading zero/etc).
1518 		 */
1519 		if (namelc[0] == '\0')
1520 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1521 		else if ((strtoul(namelc, &p, 10) != jid ||
1522 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1523 			error = EINVAL;
1524 			vfs_opterror(opts,
1525 			    "name cannot be numeric (unless it is the jid)");
1526 			goto done_deref_locked;
1527 		}
1528 		/*
1529 		 * Make sure the name isn't too long for the prison or its
1530 		 * children.
1531 		 */
1532 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1533 		onamelen = strlen(pr->pr_name + pnamelen);
1534 		namelen = strlen(namelc);
1535 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1536 			error = ENAMETOOLONG;
1537 			goto done_deref_locked;
1538 		}
1539 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1540 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1541 			    sizeof(pr->pr_name)) {
1542 				error = ENAMETOOLONG;
1543 				goto done_deref_locked;
1544 			}
1545 		}
1546 	}
1547 	if (pr_allow & ~ppr->pr_allow) {
1548 		error = EPERM;
1549 		goto done_deref_locked;
1550 	}
1551 
1552 	/*
1553 	 * Let modules check their parameters.  This requires unlocking and
1554 	 * then re-locking the prison, but this is still a valid state as long
1555 	 * as allprison_lock remains xlocked.
1556 	 */
1557 	mtx_unlock(&pr->pr_mtx);
1558 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1559 	if (error != 0) {
1560 		prison_deref(pr, created
1561 		    ? PD_LIST_XLOCKED
1562 		    : PD_DEREF | PD_LIST_XLOCKED);
1563 		goto done_releroot;
1564 	}
1565 	mtx_lock(&pr->pr_mtx);
1566 
1567 	/* At this point, all valid parameters should have been noted. */
1568 	TAILQ_FOREACH(opt, opts, link) {
1569 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1570 			error = EINVAL;
1571 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1572 			goto done_deref_locked;
1573 		}
1574 	}
1575 
1576 	/* Set the parameters of the prison. */
1577 #ifdef INET
1578 	redo_ip4 = 0;
1579 	if (pr_flags & PR_IP4_USER) {
1580 		pr->pr_flags |= PR_IP4;
1581 		free(pr->pr_ip4, M_PRISON);
1582 		pr->pr_ip4s = ip4s;
1583 		pr->pr_ip4 = ip4;
1584 		ip4 = NULL;
1585 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1586 #ifdef VIMAGE
1587 			if (tpr->pr_flags & PR_VNET) {
1588 				descend = 0;
1589 				continue;
1590 			}
1591 #endif
1592 			if (prison_restrict_ip4(tpr, NULL)) {
1593 				redo_ip4 = 1;
1594 				descend = 0;
1595 			}
1596 		}
1597 	}
1598 #endif
1599 #ifdef INET6
1600 	redo_ip6 = 0;
1601 	if (pr_flags & PR_IP6_USER) {
1602 		pr->pr_flags |= PR_IP6;
1603 		free(pr->pr_ip6, M_PRISON);
1604 		pr->pr_ip6s = ip6s;
1605 		pr->pr_ip6 = ip6;
1606 		ip6 = NULL;
1607 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1608 #ifdef VIMAGE
1609 			if (tpr->pr_flags & PR_VNET) {
1610 				descend = 0;
1611 				continue;
1612 			}
1613 #endif
1614 			if (prison_restrict_ip6(tpr, NULL)) {
1615 				redo_ip6 = 1;
1616 				descend = 0;
1617 			}
1618 		}
1619 	}
1620 #endif
1621 	if (gotslevel) {
1622 		pr->pr_securelevel = slevel;
1623 		/* Set all child jails to be at least this level. */
1624 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1625 			if (tpr->pr_securelevel < slevel)
1626 				tpr->pr_securelevel = slevel;
1627 	}
1628 	if (gotchildmax) {
1629 		pr->pr_childmax = childmax;
1630 		/* Set all child jails to under this limit. */
1631 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1632 			if (tpr->pr_childmax > childmax - level)
1633 				tpr->pr_childmax = childmax > level
1634 				    ? childmax - level : 0;
1635 	}
1636 	if (gotenforce) {
1637 		pr->pr_enforce_statfs = enforce;
1638 		/* Pass this restriction on to the children. */
1639 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1640 			if (tpr->pr_enforce_statfs < enforce)
1641 				tpr->pr_enforce_statfs = enforce;
1642 	}
1643 	if (gotrsnum) {
1644 		pr->pr_devfs_rsnum = rsnum;
1645 		/* Pass this restriction on to the children. */
1646 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1647 			tpr->pr_devfs_rsnum = rsnum;
1648 	}
1649 	if (namelc != NULL) {
1650 		if (ppr == &prison0)
1651 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1652 		else
1653 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1654 			    ppr->pr_name, namelc);
1655 		/* Change this component of child names. */
1656 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1657 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1658 			    strlen(tpr->pr_name + onamelen) + 1);
1659 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1660 		}
1661 	}
1662 	if (path != NULL) {
1663 		/* Try to keep a real-rooted full pathname. */
1664 		if (fullpath_disabled && path[0] == '/' &&
1665 		    strcmp(mypr->pr_path, "/"))
1666 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1667 			    mypr->pr_path, path);
1668 		else
1669 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1670 		pr->pr_root = root;
1671 	}
1672 	if (PR_HOST & ch_flags & ~pr_flags) {
1673 		if (pr->pr_flags & PR_HOST) {
1674 			/*
1675 			 * Copy the parent's host info.  As with pr_ip4 above,
1676 			 * the lack of a lock on the parent is not a problem;
1677 			 * it is always set with allprison_lock at least
1678 			 * shared, and is held exclusively here.
1679 			 */
1680 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1681 			    sizeof(pr->pr_hostname));
1682 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1683 			    sizeof(pr->pr_domainname));
1684 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1685 			    sizeof(pr->pr_hostuuid));
1686 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1687 		}
1688 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1689 		/* Set this prison, and any descendants without PR_HOST. */
1690 		if (host != NULL)
1691 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1692 		if (domain != NULL)
1693 			strlcpy(pr->pr_domainname, domain,
1694 			    sizeof(pr->pr_domainname));
1695 		if (uuid != NULL)
1696 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1697 		if (gothid)
1698 			pr->pr_hostid = hid;
1699 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1700 			if (tpr->pr_flags & PR_HOST)
1701 				descend = 0;
1702 			else {
1703 				if (host != NULL)
1704 					strlcpy(tpr->pr_hostname,
1705 					    pr->pr_hostname,
1706 					    sizeof(tpr->pr_hostname));
1707 				if (domain != NULL)
1708 					strlcpy(tpr->pr_domainname,
1709 					    pr->pr_domainname,
1710 					    sizeof(tpr->pr_domainname));
1711 				if (uuid != NULL)
1712 					strlcpy(tpr->pr_hostuuid,
1713 					    pr->pr_hostuuid,
1714 					    sizeof(tpr->pr_hostuuid));
1715 				if (gothid)
1716 					tpr->pr_hostid = hid;
1717 			}
1718 		}
1719 	}
1720 	if ((tallow = ch_allow & ~pr_allow)) {
1721 		/* Clear allow bits in all children. */
1722 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1723 			tpr->pr_allow &= ~tallow;
1724 	}
1725 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1726 	/*
1727 	 * Persistent prisons get an extra reference, and prisons losing their
1728 	 * persist flag lose that reference.  Only do this for existing prisons
1729 	 * for now, so new ones will remain unseen until after the module
1730 	 * handlers have completed.
1731 	 */
1732 	born = pr->pr_uref == 0;
1733 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1734 		if (pr_flags & PR_PERSIST) {
1735 			pr->pr_ref++;
1736 			pr->pr_uref++;
1737 		} else {
1738 			pr->pr_ref--;
1739 			pr->pr_uref--;
1740 		}
1741 	}
1742 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1743 	mtx_unlock(&pr->pr_mtx);
1744 
1745 #ifdef RACCT
1746 	if (racct_enable && created)
1747 		prison_racct_attach(pr);
1748 #endif
1749 
1750 	/* Locks may have prevented a complete restriction of child IP
1751 	 * addresses.  If so, allocate some more memory and try again.
1752 	 */
1753 #ifdef INET
1754 	while (redo_ip4) {
1755 		ip4s = pr->pr_ip4s;
1756 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1757 		mtx_lock(&pr->pr_mtx);
1758 		redo_ip4 = 0;
1759 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1760 #ifdef VIMAGE
1761 			if (tpr->pr_flags & PR_VNET) {
1762 				descend = 0;
1763 				continue;
1764 			}
1765 #endif
1766 			if (prison_restrict_ip4(tpr, ip4)) {
1767 				if (ip4 != NULL)
1768 					ip4 = NULL;
1769 				else
1770 					redo_ip4 = 1;
1771 			}
1772 		}
1773 		mtx_unlock(&pr->pr_mtx);
1774 	}
1775 #endif
1776 #ifdef INET6
1777 	while (redo_ip6) {
1778 		ip6s = pr->pr_ip6s;
1779 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1780 		mtx_lock(&pr->pr_mtx);
1781 		redo_ip6 = 0;
1782 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1783 #ifdef VIMAGE
1784 			if (tpr->pr_flags & PR_VNET) {
1785 				descend = 0;
1786 				continue;
1787 			}
1788 #endif
1789 			if (prison_restrict_ip6(tpr, ip6)) {
1790 				if (ip6 != NULL)
1791 					ip6 = NULL;
1792 				else
1793 					redo_ip6 = 1;
1794 			}
1795 		}
1796 		mtx_unlock(&pr->pr_mtx);
1797 	}
1798 #endif
1799 
1800 	/* Let the modules do their work. */
1801 	sx_downgrade(&allprison_lock);
1802 	if (born) {
1803 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1804 		if (error) {
1805 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1806 			prison_deref(pr, created
1807 			    ? PD_LIST_SLOCKED
1808 			    : PD_DEREF | PD_LIST_SLOCKED);
1809 			goto done_errmsg;
1810 		}
1811 	}
1812 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1813 	if (error) {
1814 		if (born)
1815 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1816 		prison_deref(pr, created
1817 		    ? PD_LIST_SLOCKED
1818 		    : PD_DEREF | PD_LIST_SLOCKED);
1819 		goto done_errmsg;
1820 	}
1821 
1822 	/* Attach this process to the prison if requested. */
1823 	if (flags & JAIL_ATTACH) {
1824 		mtx_lock(&pr->pr_mtx);
1825 		error = do_jail_attach(td, pr);
1826 		if (error) {
1827 			vfs_opterror(opts, "attach failed");
1828 			if (!created)
1829 				prison_deref(pr, PD_DEREF);
1830 			goto done_errmsg;
1831 		}
1832 	}
1833 
1834 #ifdef RACCT
1835 	if (racct_enable && !created) {
1836 		if (!(flags & JAIL_ATTACH))
1837 			sx_sunlock(&allprison_lock);
1838 		prison_racct_modify(pr);
1839 		if (!(flags & JAIL_ATTACH))
1840 			sx_slock(&allprison_lock);
1841 	}
1842 #endif
1843 
1844 	td->td_retval[0] = pr->pr_id;
1845 
1846 	/*
1847 	 * Now that it is all there, drop the temporary reference from existing
1848 	 * prisons.  Or add a reference to newly created persistent prisons
1849 	 * (which was not done earlier so that the prison would not be publicly
1850 	 * visible).
1851 	 */
1852 	if (!created) {
1853 		prison_deref(pr, (flags & JAIL_ATTACH)
1854 		    ? PD_DEREF
1855 		    : PD_DEREF | PD_LIST_SLOCKED);
1856 	} else {
1857 		if (pr_flags & PR_PERSIST) {
1858 			mtx_lock(&pr->pr_mtx);
1859 			pr->pr_ref++;
1860 			pr->pr_uref++;
1861 			mtx_unlock(&pr->pr_mtx);
1862 		}
1863 		if (!(flags & JAIL_ATTACH))
1864 			sx_sunlock(&allprison_lock);
1865 	}
1866 
1867 	goto done_free;
1868 
1869  done_deref_locked:
1870 	prison_deref(pr, created
1871 	    ? PD_LOCKED | PD_LIST_XLOCKED
1872 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1873 	goto done_releroot;
1874  done_unlock_list:
1875 	sx_xunlock(&allprison_lock);
1876  done_releroot:
1877 	if (root != NULL)
1878 		vrele(root);
1879  done_errmsg:
1880 	if (error) {
1881 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1882 		    &errmsg_len) == 0 && errmsg_len > 0) {
1883 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1884 			if (optuio->uio_segflg == UIO_SYSSPACE)
1885 				bcopy(errmsg,
1886 				    optuio->uio_iov[errmsg_pos].iov_base,
1887 				    errmsg_len);
1888 			else
1889 				copyout(errmsg,
1890 				    optuio->uio_iov[errmsg_pos].iov_base,
1891 				    errmsg_len);
1892 		}
1893 	}
1894  done_free:
1895 #ifdef INET
1896 	free(ip4, M_PRISON);
1897 #endif
1898 #ifdef INET6
1899 	free(ip6, M_PRISON);
1900 #endif
1901 	if (g_path != NULL)
1902 		free(g_path, M_TEMP);
1903 	vfs_freeopts(opts);
1904 	return (error);
1905 }
1906 
1907 
1908 /*
1909  * struct jail_get_args {
1910  *	struct iovec *iovp;
1911  *	unsigned int iovcnt;
1912  *	int flags;
1913  * };
1914  */
1915 int
1916 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1917 {
1918 	struct uio *auio;
1919 	int error;
1920 
1921 	/* Check that we have an even number of iovecs. */
1922 	if (uap->iovcnt & 1)
1923 		return (EINVAL);
1924 
1925 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1926 	if (error)
1927 		return (error);
1928 	error = kern_jail_get(td, auio, uap->flags);
1929 	if (error == 0)
1930 		error = copyout(auio->uio_iov, uap->iovp,
1931 		    uap->iovcnt * sizeof (struct iovec));
1932 	free(auio, M_IOV);
1933 	return (error);
1934 }
1935 
1936 int
1937 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1938 {
1939 	struct prison *pr, *mypr;
1940 	struct vfsopt *opt;
1941 	struct vfsoptlist *opts;
1942 	char *errmsg, *name;
1943 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1944 
1945 	if (flags & ~JAIL_GET_MASK)
1946 		return (EINVAL);
1947 
1948 	/* Get the parameter list. */
1949 	error = vfs_buildopts(optuio, &opts);
1950 	if (error)
1951 		return (error);
1952 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1953 	mypr = td->td_ucred->cr_prison;
1954 
1955 	/*
1956 	 * Find the prison specified by one of: lastjid, jid, name.
1957 	 */
1958 	sx_slock(&allprison_lock);
1959 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1960 	if (error == 0) {
1961 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1962 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1963 				mtx_lock(&pr->pr_mtx);
1964 				if (pr->pr_ref > 0 &&
1965 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1966 					break;
1967 				mtx_unlock(&pr->pr_mtx);
1968 			}
1969 		}
1970 		if (pr != NULL)
1971 			goto found_prison;
1972 		error = ENOENT;
1973 		vfs_opterror(opts, "no jail after %d", jid);
1974 		goto done_unlock_list;
1975 	} else if (error != ENOENT)
1976 		goto done_unlock_list;
1977 
1978 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1979 	if (error == 0) {
1980 		if (jid != 0) {
1981 			pr = prison_find_child(mypr, jid);
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 %d is dying",
1987 					    jid);
1988 					goto done_unlock_list;
1989 				}
1990 				goto found_prison;
1991 			}
1992 			error = ENOENT;
1993 			vfs_opterror(opts, "jail %d not found", jid);
1994 			goto done_unlock_list;
1995 		}
1996 	} else if (error != ENOENT)
1997 		goto done_unlock_list;
1998 
1999 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2000 	if (error == 0) {
2001 		if (len == 0 || name[len - 1] != '\0') {
2002 			error = EINVAL;
2003 			goto done_unlock_list;
2004 		}
2005 		pr = prison_find_name(mypr, name);
2006 		if (pr != NULL) {
2007 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2008 				mtx_unlock(&pr->pr_mtx);
2009 				error = ENOENT;
2010 				vfs_opterror(opts, "jail \"%s\" is dying",
2011 				    name);
2012 				goto done_unlock_list;
2013 			}
2014 			goto found_prison;
2015 		}
2016 		error = ENOENT;
2017 		vfs_opterror(opts, "jail \"%s\" not found", name);
2018 		goto done_unlock_list;
2019 	} else if (error != ENOENT)
2020 		goto done_unlock_list;
2021 
2022 	vfs_opterror(opts, "no jail specified");
2023 	error = ENOENT;
2024 	goto done_unlock_list;
2025 
2026  found_prison:
2027 	/* Get the parameters of the prison. */
2028 	pr->pr_ref++;
2029 	locked = PD_LOCKED;
2030 	td->td_retval[0] = pr->pr_id;
2031 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2032 	if (error != 0 && error != ENOENT)
2033 		goto done_deref;
2034 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2035 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2036 	if (error != 0 && error != ENOENT)
2037 		goto done_deref;
2038 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2039 	if (error != 0 && error != ENOENT)
2040 		goto done_deref;
2041 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2042 	    sizeof(pr->pr_cpuset->cs_id));
2043 	if (error != 0 && error != ENOENT)
2044 		goto done_deref;
2045 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2046 	if (error != 0 && error != ENOENT)
2047 		goto done_deref;
2048 #ifdef INET
2049 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2050 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2051 	if (error != 0 && error != ENOENT)
2052 		goto done_deref;
2053 #endif
2054 #ifdef INET6
2055 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2056 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2057 	if (error != 0 && error != ENOENT)
2058 		goto done_deref;
2059 #endif
2060 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2061 	    sizeof(pr->pr_securelevel));
2062 	if (error != 0 && error != ENOENT)
2063 		goto done_deref;
2064 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2065 	    sizeof(pr->pr_childcount));
2066 	if (error != 0 && error != ENOENT)
2067 		goto done_deref;
2068 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2069 	    sizeof(pr->pr_childmax));
2070 	if (error != 0 && error != ENOENT)
2071 		goto done_deref;
2072 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2073 	if (error != 0 && error != ENOENT)
2074 		goto done_deref;
2075 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2076 	if (error != 0 && error != ENOENT)
2077 		goto done_deref;
2078 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2079 	if (error != 0 && error != ENOENT)
2080 		goto done_deref;
2081 #ifdef COMPAT_FREEBSD32
2082 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2083 		uint32_t hid32 = pr->pr_hostid;
2084 
2085 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2086 	} else
2087 #endif
2088 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2089 	    sizeof(pr->pr_hostid));
2090 	if (error != 0 && error != ENOENT)
2091 		goto done_deref;
2092 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2093 	    sizeof(pr->pr_enforce_statfs));
2094 	if (error != 0 && error != ENOENT)
2095 		goto done_deref;
2096 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2097 	    sizeof(pr->pr_devfs_rsnum));
2098 	if (error != 0 && error != ENOENT)
2099 		goto done_deref;
2100 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2101 		if (pr_flag_names[fi] == NULL)
2102 			continue;
2103 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2104 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2105 		if (error != 0 && error != ENOENT)
2106 			goto done_deref;
2107 		i = !i;
2108 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2109 		if (error != 0 && error != ENOENT)
2110 			goto done_deref;
2111 	}
2112 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
2113 		i = pr->pr_flags &
2114 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2115 		i = pr_flag_jailsys[fi].disable &&
2116 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2117 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2118 		    : JAIL_SYS_INHERIT;
2119 		error =
2120 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2121 		if (error != 0 && error != ENOENT)
2122 			goto done_deref;
2123 	}
2124 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
2125 		if (pr_allow_names[fi] == NULL)
2126 			continue;
2127 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2128 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2129 		if (error != 0 && error != ENOENT)
2130 			goto done_deref;
2131 		i = !i;
2132 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2133 		if (error != 0 && error != ENOENT)
2134 			goto done_deref;
2135 	}
2136 	i = (pr->pr_uref == 0);
2137 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2138 	if (error != 0 && error != ENOENT)
2139 		goto done_deref;
2140 	i = !i;
2141 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2142 	if (error != 0 && error != ENOENT)
2143 		goto done_deref;
2144 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2145 	    sizeof(pr->pr_osreldate));
2146 	if (error != 0 && error != ENOENT)
2147 		goto done_deref;
2148 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2149 	if (error != 0 && error != ENOENT)
2150 		goto done_deref;
2151 
2152 	/* Get the module parameters. */
2153 	mtx_unlock(&pr->pr_mtx);
2154 	locked = 0;
2155 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2156 	if (error)
2157 		goto done_deref;
2158 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2159 
2160 	/* By now, all parameters should have been noted. */
2161 	TAILQ_FOREACH(opt, opts, link) {
2162 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2163 			error = EINVAL;
2164 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2165 			goto done_errmsg;
2166 		}
2167 	}
2168 
2169 	/* Write the fetched parameters back to userspace. */
2170 	error = 0;
2171 	TAILQ_FOREACH(opt, opts, link) {
2172 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2173 			pos = 2 * opt->pos + 1;
2174 			optuio->uio_iov[pos].iov_len = opt->len;
2175 			if (opt->value != NULL) {
2176 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2177 					bcopy(opt->value,
2178 					    optuio->uio_iov[pos].iov_base,
2179 					    opt->len);
2180 				} else {
2181 					error = copyout(opt->value,
2182 					    optuio->uio_iov[pos].iov_base,
2183 					    opt->len);
2184 					if (error)
2185 						break;
2186 				}
2187 			}
2188 		}
2189 	}
2190 	goto done_errmsg;
2191 
2192  done_deref:
2193 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2194 	goto done_errmsg;
2195 
2196  done_unlock_list:
2197 	sx_sunlock(&allprison_lock);
2198  done_errmsg:
2199 	if (error && errmsg_pos >= 0) {
2200 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2201 		errmsg_pos = 2 * errmsg_pos + 1;
2202 		if (errmsg_len > 0) {
2203 			if (optuio->uio_segflg == UIO_SYSSPACE)
2204 				bcopy(errmsg,
2205 				    optuio->uio_iov[errmsg_pos].iov_base,
2206 				    errmsg_len);
2207 			else
2208 				copyout(errmsg,
2209 				    optuio->uio_iov[errmsg_pos].iov_base,
2210 				    errmsg_len);
2211 		}
2212 	}
2213 	vfs_freeopts(opts);
2214 	return (error);
2215 }
2216 
2217 
2218 /*
2219  * struct jail_remove_args {
2220  *	int jid;
2221  * };
2222  */
2223 int
2224 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2225 {
2226 	struct prison *pr, *cpr, *lpr, *tpr;
2227 	int descend, error;
2228 
2229 	error = priv_check(td, PRIV_JAIL_REMOVE);
2230 	if (error)
2231 		return (error);
2232 
2233 	sx_xlock(&allprison_lock);
2234 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2235 	if (pr == NULL) {
2236 		sx_xunlock(&allprison_lock);
2237 		return (EINVAL);
2238 	}
2239 
2240 	/* Remove all descendants of this prison, then remove this prison. */
2241 	pr->pr_ref++;
2242 	if (!LIST_EMPTY(&pr->pr_children)) {
2243 		mtx_unlock(&pr->pr_mtx);
2244 		lpr = NULL;
2245 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2246 			mtx_lock(&cpr->pr_mtx);
2247 			if (cpr->pr_ref > 0) {
2248 				tpr = cpr;
2249 				cpr->pr_ref++;
2250 			} else {
2251 				/* Already removed - do not do it again. */
2252 				tpr = NULL;
2253 			}
2254 			mtx_unlock(&cpr->pr_mtx);
2255 			if (lpr != NULL) {
2256 				mtx_lock(&lpr->pr_mtx);
2257 				prison_remove_one(lpr);
2258 				sx_xlock(&allprison_lock);
2259 			}
2260 			lpr = tpr;
2261 		}
2262 		if (lpr != NULL) {
2263 			mtx_lock(&lpr->pr_mtx);
2264 			prison_remove_one(lpr);
2265 			sx_xlock(&allprison_lock);
2266 		}
2267 		mtx_lock(&pr->pr_mtx);
2268 	}
2269 	prison_remove_one(pr);
2270 	return (0);
2271 }
2272 
2273 static void
2274 prison_remove_one(struct prison *pr)
2275 {
2276 	struct proc *p;
2277 	int deuref;
2278 
2279 	/* If the prison was persistent, it is not anymore. */
2280 	deuref = 0;
2281 	if (pr->pr_flags & PR_PERSIST) {
2282 		pr->pr_ref--;
2283 		deuref = PD_DEUREF;
2284 		pr->pr_flags &= ~PR_PERSIST;
2285 	}
2286 
2287 	/*
2288 	 * jail_remove added a reference.  If that's the only one, remove
2289 	 * the prison now.
2290 	 */
2291 	KASSERT(pr->pr_ref > 0,
2292 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2293 	if (pr->pr_ref == 1) {
2294 		prison_deref(pr,
2295 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2296 		return;
2297 	}
2298 
2299 	mtx_unlock(&pr->pr_mtx);
2300 	sx_xunlock(&allprison_lock);
2301 	/*
2302 	 * Kill all processes unfortunate enough to be attached to this prison.
2303 	 */
2304 	sx_slock(&allproc_lock);
2305 	LIST_FOREACH(p, &allproc, p_list) {
2306 		PROC_LOCK(p);
2307 		if (p->p_state != PRS_NEW && p->p_ucred &&
2308 		    p->p_ucred->cr_prison == pr)
2309 			kern_psignal(p, SIGKILL);
2310 		PROC_UNLOCK(p);
2311 	}
2312 	sx_sunlock(&allproc_lock);
2313 	/* Remove the temporary reference added by jail_remove. */
2314 	prison_deref(pr, deuref | PD_DEREF);
2315 }
2316 
2317 
2318 /*
2319  * struct jail_attach_args {
2320  *	int jid;
2321  * };
2322  */
2323 int
2324 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2325 {
2326 	struct prison *pr;
2327 	int error;
2328 
2329 	error = priv_check(td, PRIV_JAIL_ATTACH);
2330 	if (error)
2331 		return (error);
2332 
2333 	/*
2334 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2335 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2336 	 * But then immediately downgrade it since we don't need to stop
2337 	 * readers.
2338 	 */
2339 	sx_xlock(&allprison_lock);
2340 	sx_downgrade(&allprison_lock);
2341 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2342 	if (pr == NULL) {
2343 		sx_sunlock(&allprison_lock);
2344 		return (EINVAL);
2345 	}
2346 
2347 	/*
2348 	 * Do not allow a process to attach to a prison that is not
2349 	 * considered to be "alive".
2350 	 */
2351 	if (pr->pr_uref == 0) {
2352 		mtx_unlock(&pr->pr_mtx);
2353 		sx_sunlock(&allprison_lock);
2354 		return (EINVAL);
2355 	}
2356 
2357 	return (do_jail_attach(td, pr));
2358 }
2359 
2360 static int
2361 do_jail_attach(struct thread *td, struct prison *pr)
2362 {
2363 	struct proc *p;
2364 	struct ucred *newcred, *oldcred;
2365 	int error;
2366 
2367 	/*
2368 	 * XXX: Note that there is a slight race here if two threads
2369 	 * in the same privileged process attempt to attach to two
2370 	 * different jails at the same time.  It is important for
2371 	 * user processes not to do this, or they might end up with
2372 	 * a process root from one prison, but attached to the jail
2373 	 * of another.
2374 	 */
2375 	pr->pr_ref++;
2376 	pr->pr_uref++;
2377 	mtx_unlock(&pr->pr_mtx);
2378 
2379 	/* Let modules do whatever they need to prepare for attaching. */
2380 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2381 	if (error) {
2382 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2383 		return (error);
2384 	}
2385 	sx_sunlock(&allprison_lock);
2386 
2387 	/*
2388 	 * Reparent the newly attached process to this jail.
2389 	 */
2390 	p = td->td_proc;
2391 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2392 	if (error)
2393 		goto e_revert_osd;
2394 
2395 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2396 	if ((error = change_dir(pr->pr_root, td)) != 0)
2397 		goto e_unlock;
2398 #ifdef MAC
2399 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2400 		goto e_unlock;
2401 #endif
2402 	VOP_UNLOCK(pr->pr_root, 0);
2403 	if ((error = pwd_chroot(td, pr->pr_root)))
2404 		goto e_revert_osd;
2405 
2406 	newcred = crget();
2407 	PROC_LOCK(p);
2408 	oldcred = crcopysafe(p, newcred);
2409 	newcred->cr_prison = pr;
2410 	proc_set_cred(p, newcred);
2411 	setsugid(p);
2412 	PROC_UNLOCK(p);
2413 #ifdef RACCT
2414 	racct_proc_ucred_changed(p, oldcred, newcred);
2415 #endif
2416 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2417 	crfree(oldcred);
2418 	return (0);
2419 
2420  e_unlock:
2421 	VOP_UNLOCK(pr->pr_root, 0);
2422  e_revert_osd:
2423 	/* Tell modules this thread is still in its old jail after all. */
2424 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2425 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2426 	return (error);
2427 }
2428 
2429 
2430 /*
2431  * Returns a locked prison instance, or NULL on failure.
2432  */
2433 struct prison *
2434 prison_find(int prid)
2435 {
2436 	struct prison *pr;
2437 
2438 	sx_assert(&allprison_lock, SX_LOCKED);
2439 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2440 		if (pr->pr_id == prid) {
2441 			mtx_lock(&pr->pr_mtx);
2442 			if (pr->pr_ref > 0)
2443 				return (pr);
2444 			mtx_unlock(&pr->pr_mtx);
2445 		}
2446 	}
2447 	return (NULL);
2448 }
2449 
2450 /*
2451  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2452  */
2453 struct prison *
2454 prison_find_child(struct prison *mypr, int prid)
2455 {
2456 	struct prison *pr;
2457 	int descend;
2458 
2459 	sx_assert(&allprison_lock, SX_LOCKED);
2460 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2461 		if (pr->pr_id == prid) {
2462 			mtx_lock(&pr->pr_mtx);
2463 			if (pr->pr_ref > 0)
2464 				return (pr);
2465 			mtx_unlock(&pr->pr_mtx);
2466 		}
2467 	}
2468 	return (NULL);
2469 }
2470 
2471 /*
2472  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2473  */
2474 struct prison *
2475 prison_find_name(struct prison *mypr, const char *name)
2476 {
2477 	struct prison *pr, *deadpr;
2478 	size_t mylen;
2479 	int descend;
2480 
2481 	sx_assert(&allprison_lock, SX_LOCKED);
2482 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2483  again:
2484 	deadpr = NULL;
2485 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2486 		if (!strcmp(pr->pr_name + mylen, name)) {
2487 			mtx_lock(&pr->pr_mtx);
2488 			if (pr->pr_ref > 0) {
2489 				if (pr->pr_uref > 0)
2490 					return (pr);
2491 				deadpr = pr;
2492 			}
2493 			mtx_unlock(&pr->pr_mtx);
2494 		}
2495 	}
2496 	/* There was no valid prison - perhaps there was a dying one. */
2497 	if (deadpr != NULL) {
2498 		mtx_lock(&deadpr->pr_mtx);
2499 		if (deadpr->pr_ref == 0) {
2500 			mtx_unlock(&deadpr->pr_mtx);
2501 			goto again;
2502 		}
2503 	}
2504 	return (deadpr);
2505 }
2506 
2507 /*
2508  * See if a prison has the specific flag set.
2509  */
2510 int
2511 prison_flag(struct ucred *cred, unsigned flag)
2512 {
2513 
2514 	/* This is an atomic read, so no locking is necessary. */
2515 	return (cred->cr_prison->pr_flags & flag);
2516 }
2517 
2518 int
2519 prison_allow(struct ucred *cred, unsigned flag)
2520 {
2521 
2522 	/* This is an atomic read, so no locking is necessary. */
2523 	return (cred->cr_prison->pr_allow & flag);
2524 }
2525 
2526 /*
2527  * Remove a prison reference.  If that was the last reference, remove the
2528  * prison itself - but not in this context in case there are locks held.
2529  */
2530 void
2531 prison_free_locked(struct prison *pr)
2532 {
2533 	int ref;
2534 
2535 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2536 	ref = --pr->pr_ref;
2537 	mtx_unlock(&pr->pr_mtx);
2538 	if (ref == 0)
2539 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2540 }
2541 
2542 void
2543 prison_free(struct prison *pr)
2544 {
2545 
2546 	mtx_lock(&pr->pr_mtx);
2547 	prison_free_locked(pr);
2548 }
2549 
2550 /*
2551  * Complete a call to either prison_free or prison_proc_free.
2552  */
2553 static void
2554 prison_complete(void *context, int pending)
2555 {
2556 	struct prison *pr = context;
2557 
2558 	sx_xlock(&allprison_lock);
2559 	mtx_lock(&pr->pr_mtx);
2560 	prison_deref(pr, pr->pr_uref
2561 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2562 	    : PD_LOCKED | PD_LIST_XLOCKED);
2563 }
2564 
2565 /*
2566  * Remove a prison reference (usually).  This internal version assumes no
2567  * mutexes are held, except perhaps the prison itself.  If there are no more
2568  * references, release and delist the prison.  On completion, the prison lock
2569  * and the allprison lock are both unlocked.
2570  */
2571 static void
2572 prison_deref(struct prison *pr, int flags)
2573 {
2574 	struct prison *ppr, *tpr;
2575 	int ref, lasturef;
2576 
2577 	if (!(flags & PD_LOCKED))
2578 		mtx_lock(&pr->pr_mtx);
2579 	for (;;) {
2580 		if (flags & PD_DEUREF) {
2581 			KASSERT(pr->pr_uref > 0,
2582 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2583 			     pr->pr_id));
2584 			pr->pr_uref--;
2585 			lasturef = pr->pr_uref == 0;
2586 			if (lasturef)
2587 				pr->pr_ref++;
2588 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2589 		} else
2590 			lasturef = 0;
2591 		if (flags & PD_DEREF) {
2592 			KASSERT(pr->pr_ref > 0,
2593 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2594 			     pr->pr_id));
2595 			pr->pr_ref--;
2596 		}
2597 		ref = pr->pr_ref;
2598 		mtx_unlock(&pr->pr_mtx);
2599 
2600 		/*
2601 		 * Tell the modules if the last user reference was removed
2602 		 * (even it sticks around in dying state).
2603 		 */
2604 		if (lasturef) {
2605 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2606 				sx_xlock(&allprison_lock);
2607 				flags |= PD_LIST_XLOCKED;
2608 			}
2609 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2610 			mtx_lock(&pr->pr_mtx);
2611 			ref = --pr->pr_ref;
2612 			mtx_unlock(&pr->pr_mtx);
2613 		}
2614 
2615 		/* If the prison still has references, nothing else to do. */
2616 		if (ref > 0) {
2617 			if (flags & PD_LIST_SLOCKED)
2618 				sx_sunlock(&allprison_lock);
2619 			else if (flags & PD_LIST_XLOCKED)
2620 				sx_xunlock(&allprison_lock);
2621 			return;
2622 		}
2623 
2624 		if (flags & PD_LIST_SLOCKED) {
2625 			if (!sx_try_upgrade(&allprison_lock)) {
2626 				sx_sunlock(&allprison_lock);
2627 				sx_xlock(&allprison_lock);
2628 			}
2629 		} else if (!(flags & PD_LIST_XLOCKED))
2630 			sx_xlock(&allprison_lock);
2631 
2632 		TAILQ_REMOVE(&allprison, pr, pr_list);
2633 		LIST_REMOVE(pr, pr_sibling);
2634 		ppr = pr->pr_parent;
2635 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2636 			tpr->pr_childcount--;
2637 		sx_xunlock(&allprison_lock);
2638 
2639 #ifdef VIMAGE
2640 		if (pr->pr_vnet != ppr->pr_vnet)
2641 			vnet_destroy(pr->pr_vnet);
2642 #endif
2643 		if (pr->pr_root != NULL)
2644 			vrele(pr->pr_root);
2645 		mtx_destroy(&pr->pr_mtx);
2646 #ifdef INET
2647 		free(pr->pr_ip4, M_PRISON);
2648 #endif
2649 #ifdef INET6
2650 		free(pr->pr_ip6, M_PRISON);
2651 #endif
2652 		if (pr->pr_cpuset != NULL)
2653 			cpuset_rel(pr->pr_cpuset);
2654 		osd_jail_exit(pr);
2655 #ifdef RACCT
2656 		if (racct_enable)
2657 			prison_racct_detach(pr);
2658 #endif
2659 		free(pr, M_PRISON);
2660 
2661 		/* Removing a prison frees a reference on its parent. */
2662 		pr = ppr;
2663 		mtx_lock(&pr->pr_mtx);
2664 		flags = PD_DEREF | PD_DEUREF;
2665 	}
2666 }
2667 
2668 void
2669 prison_hold_locked(struct prison *pr)
2670 {
2671 
2672 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2673 	KASSERT(pr->pr_ref > 0,
2674 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2675 	pr->pr_ref++;
2676 }
2677 
2678 void
2679 prison_hold(struct prison *pr)
2680 {
2681 
2682 	mtx_lock(&pr->pr_mtx);
2683 	prison_hold_locked(pr);
2684 	mtx_unlock(&pr->pr_mtx);
2685 }
2686 
2687 void
2688 prison_proc_hold(struct prison *pr)
2689 {
2690 
2691 	mtx_lock(&pr->pr_mtx);
2692 	KASSERT(pr->pr_uref > 0,
2693 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2694 	pr->pr_uref++;
2695 	mtx_unlock(&pr->pr_mtx);
2696 }
2697 
2698 void
2699 prison_proc_free(struct prison *pr)
2700 {
2701 
2702 	mtx_lock(&pr->pr_mtx);
2703 	KASSERT(pr->pr_uref > 0,
2704 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2705 	if (pr->pr_uref > 1)
2706 		pr->pr_uref--;
2707 	else {
2708 		/*
2709 		 * Don't remove the last user reference in this context, which
2710 		 * is expected to be a process that is not only locked, but
2711 		 * also half dead.
2712 		 */
2713 		pr->pr_ref++;
2714 		mtx_unlock(&pr->pr_mtx);
2715 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2716 		return;
2717 	}
2718 	mtx_unlock(&pr->pr_mtx);
2719 }
2720 
2721 /*
2722  * Check if a jail supports the given address family.
2723  *
2724  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2725  * if not.
2726  */
2727 int
2728 prison_check_af(struct ucred *cred, int af)
2729 {
2730 	struct prison *pr;
2731 	int error;
2732 
2733 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2734 
2735 	pr = cred->cr_prison;
2736 #ifdef VIMAGE
2737 	/* Prisons with their own network stack are not limited. */
2738 	if (prison_owns_vnet(cred))
2739 		return (0);
2740 #endif
2741 
2742 	error = 0;
2743 	switch (af)
2744 	{
2745 #ifdef INET
2746 	case AF_INET:
2747 		if (pr->pr_flags & PR_IP4)
2748 		{
2749 			mtx_lock(&pr->pr_mtx);
2750 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2751 				error = EAFNOSUPPORT;
2752 			mtx_unlock(&pr->pr_mtx);
2753 		}
2754 		break;
2755 #endif
2756 #ifdef INET6
2757 	case AF_INET6:
2758 		if (pr->pr_flags & PR_IP6)
2759 		{
2760 			mtx_lock(&pr->pr_mtx);
2761 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2762 				error = EAFNOSUPPORT;
2763 			mtx_unlock(&pr->pr_mtx);
2764 		}
2765 		break;
2766 #endif
2767 	case AF_LOCAL:
2768 	case AF_ROUTE:
2769 		break;
2770 	default:
2771 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2772 			error = EAFNOSUPPORT;
2773 	}
2774 	return (error);
2775 }
2776 
2777 /*
2778  * Check if given address belongs to the jail referenced by cred (wrapper to
2779  * prison_check_ip[46]).
2780  *
2781  * Returns 0 if jail doesn't restrict the address family or if address belongs
2782  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2783  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2784  */
2785 int
2786 prison_if(struct ucred *cred, struct sockaddr *sa)
2787 {
2788 #ifdef INET
2789 	struct sockaddr_in *sai;
2790 #endif
2791 #ifdef INET6
2792 	struct sockaddr_in6 *sai6;
2793 #endif
2794 	int error;
2795 
2796 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2797 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2798 
2799 #ifdef VIMAGE
2800 	if (prison_owns_vnet(cred))
2801 		return (0);
2802 #endif
2803 
2804 	error = 0;
2805 	switch (sa->sa_family)
2806 	{
2807 #ifdef INET
2808 	case AF_INET:
2809 		sai = (struct sockaddr_in *)sa;
2810 		error = prison_check_ip4(cred, &sai->sin_addr);
2811 		break;
2812 #endif
2813 #ifdef INET6
2814 	case AF_INET6:
2815 		sai6 = (struct sockaddr_in6 *)sa;
2816 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2817 		break;
2818 #endif
2819 	default:
2820 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2821 			error = EAFNOSUPPORT;
2822 	}
2823 	return (error);
2824 }
2825 
2826 /*
2827  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2828  */
2829 int
2830 prison_check(struct ucred *cred1, struct ucred *cred2)
2831 {
2832 
2833 	return ((cred1->cr_prison == cred2->cr_prison ||
2834 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2835 }
2836 
2837 /*
2838  * Return 1 if p2 is a child of p1, otherwise 0.
2839  */
2840 int
2841 prison_ischild(struct prison *pr1, struct prison *pr2)
2842 {
2843 
2844 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2845 		if (pr1 == pr2)
2846 			return (1);
2847 	return (0);
2848 }
2849 
2850 /*
2851  * Return 1 if the passed credential is in a jail, otherwise 0.
2852  */
2853 int
2854 jailed(struct ucred *cred)
2855 {
2856 
2857 	return (cred->cr_prison != &prison0);
2858 }
2859 
2860 /*
2861  * Return 1 if the passed credential is in a jail and that jail does not
2862  * have its own virtual network stack, otherwise 0.
2863  */
2864 int
2865 jailed_without_vnet(struct ucred *cred)
2866 {
2867 
2868 	if (!jailed(cred))
2869 		return (0);
2870 #ifdef VIMAGE
2871 	if (prison_owns_vnet(cred))
2872 		return (0);
2873 #endif
2874 
2875 	return (1);
2876 }
2877 
2878 /*
2879  * Return the correct hostname (domainname, et al) for the passed credential.
2880  */
2881 void
2882 getcredhostname(struct ucred *cred, char *buf, size_t size)
2883 {
2884 	struct prison *pr;
2885 
2886 	/*
2887 	 * A NULL credential can be used to shortcut to the physical
2888 	 * system's hostname.
2889 	 */
2890 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2891 	mtx_lock(&pr->pr_mtx);
2892 	strlcpy(buf, pr->pr_hostname, size);
2893 	mtx_unlock(&pr->pr_mtx);
2894 }
2895 
2896 void
2897 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2898 {
2899 
2900 	mtx_lock(&cred->cr_prison->pr_mtx);
2901 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2902 	mtx_unlock(&cred->cr_prison->pr_mtx);
2903 }
2904 
2905 void
2906 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2907 {
2908 
2909 	mtx_lock(&cred->cr_prison->pr_mtx);
2910 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2911 	mtx_unlock(&cred->cr_prison->pr_mtx);
2912 }
2913 
2914 void
2915 getcredhostid(struct ucred *cred, unsigned long *hostid)
2916 {
2917 
2918 	mtx_lock(&cred->cr_prison->pr_mtx);
2919 	*hostid = cred->cr_prison->pr_hostid;
2920 	mtx_unlock(&cred->cr_prison->pr_mtx);
2921 }
2922 
2923 #ifdef VIMAGE
2924 /*
2925  * Determine whether the prison represented by cred owns
2926  * its vnet rather than having it inherited.
2927  *
2928  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2929  */
2930 int
2931 prison_owns_vnet(struct ucred *cred)
2932 {
2933 
2934 	/*
2935 	 * vnets cannot be added/removed after jail creation,
2936 	 * so no need to lock here.
2937 	 */
2938 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2939 }
2940 #endif
2941 
2942 /*
2943  * Determine whether the subject represented by cred can "see"
2944  * status of a mount point.
2945  * Returns: 0 for permitted, ENOENT otherwise.
2946  * XXX: This function should be called cr_canseemount() and should be
2947  *      placed in kern_prot.c.
2948  */
2949 int
2950 prison_canseemount(struct ucred *cred, struct mount *mp)
2951 {
2952 	struct prison *pr;
2953 	struct statfs *sp;
2954 	size_t len;
2955 
2956 	pr = cred->cr_prison;
2957 	if (pr->pr_enforce_statfs == 0)
2958 		return (0);
2959 	if (pr->pr_root->v_mount == mp)
2960 		return (0);
2961 	if (pr->pr_enforce_statfs == 2)
2962 		return (ENOENT);
2963 	/*
2964 	 * If jail's chroot directory is set to "/" we should be able to see
2965 	 * all mount-points from inside a jail.
2966 	 * This is ugly check, but this is the only situation when jail's
2967 	 * directory ends with '/'.
2968 	 */
2969 	if (strcmp(pr->pr_path, "/") == 0)
2970 		return (0);
2971 	len = strlen(pr->pr_path);
2972 	sp = &mp->mnt_stat;
2973 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2974 		return (ENOENT);
2975 	/*
2976 	 * Be sure that we don't have situation where jail's root directory
2977 	 * is "/some/path" and mount point is "/some/pathpath".
2978 	 */
2979 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2980 		return (ENOENT);
2981 	return (0);
2982 }
2983 
2984 void
2985 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2986 {
2987 	char jpath[MAXPATHLEN];
2988 	struct prison *pr;
2989 	size_t len;
2990 
2991 	pr = cred->cr_prison;
2992 	if (pr->pr_enforce_statfs == 0)
2993 		return;
2994 	if (prison_canseemount(cred, mp) != 0) {
2995 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2996 		strlcpy(sp->f_mntonname, "[restricted]",
2997 		    sizeof(sp->f_mntonname));
2998 		return;
2999 	}
3000 	if (pr->pr_root->v_mount == mp) {
3001 		/*
3002 		 * Clear current buffer data, so we are sure nothing from
3003 		 * the valid path left there.
3004 		 */
3005 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3006 		*sp->f_mntonname = '/';
3007 		return;
3008 	}
3009 	/*
3010 	 * If jail's chroot directory is set to "/" we should be able to see
3011 	 * all mount-points from inside a jail.
3012 	 */
3013 	if (strcmp(pr->pr_path, "/") == 0)
3014 		return;
3015 	len = strlen(pr->pr_path);
3016 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3017 	/*
3018 	 * Clear current buffer data, so we are sure nothing from
3019 	 * the valid path left there.
3020 	 */
3021 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3022 	if (*jpath == '\0') {
3023 		/* Should never happen. */
3024 		*sp->f_mntonname = '/';
3025 	} else {
3026 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3027 	}
3028 }
3029 
3030 /*
3031  * Check with permission for a specific privilege is granted within jail.  We
3032  * have a specific list of accepted privileges; the rest are denied.
3033  */
3034 int
3035 prison_priv_check(struct ucred *cred, int priv)
3036 {
3037 
3038 	if (!jailed(cred))
3039 		return (0);
3040 
3041 #ifdef VIMAGE
3042 	/*
3043 	 * Privileges specific to prisons with a virtual network stack.
3044 	 * There might be a duplicate entry here in case the privilege
3045 	 * is only granted conditionally in the legacy jail case.
3046 	 */
3047 	switch (priv) {
3048 #ifdef notyet
3049 		/*
3050 		 * NFS-specific privileges.
3051 		 */
3052 	case PRIV_NFS_DAEMON:
3053 	case PRIV_NFS_LOCKD:
3054 #endif
3055 		/*
3056 		 * Network stack privileges.
3057 		 */
3058 	case PRIV_NET_BRIDGE:
3059 	case PRIV_NET_GRE:
3060 	case PRIV_NET_BPF:
3061 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3062 	case PRIV_NET_ROUTE:
3063 	case PRIV_NET_TAP:
3064 	case PRIV_NET_SETIFMTU:
3065 	case PRIV_NET_SETIFFLAGS:
3066 	case PRIV_NET_SETIFCAP:
3067 	case PRIV_NET_SETIFDESCR:
3068 	case PRIV_NET_SETIFNAME	:
3069 	case PRIV_NET_SETIFMETRIC:
3070 	case PRIV_NET_SETIFPHYS:
3071 	case PRIV_NET_SETIFMAC:
3072 	case PRIV_NET_ADDMULTI:
3073 	case PRIV_NET_DELMULTI:
3074 	case PRIV_NET_HWIOCTL:
3075 	case PRIV_NET_SETLLADDR:
3076 	case PRIV_NET_ADDIFGROUP:
3077 	case PRIV_NET_DELIFGROUP:
3078 	case PRIV_NET_IFCREATE:
3079 	case PRIV_NET_IFDESTROY:
3080 	case PRIV_NET_ADDIFADDR:
3081 	case PRIV_NET_DELIFADDR:
3082 	case PRIV_NET_LAGG:
3083 	case PRIV_NET_GIF:
3084 	case PRIV_NET_SETIFVNET:
3085 	case PRIV_NET_SETIFFIB:
3086 
3087 		/*
3088 		 * 802.11-related privileges.
3089 		 */
3090 	case PRIV_NET80211_GETKEY:
3091 #ifdef notyet
3092 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3093 #endif
3094 
3095 #ifdef notyet
3096 		/*
3097 		 * ATM privileges.
3098 		 */
3099 	case PRIV_NETATM_CFG:
3100 	case PRIV_NETATM_ADD:
3101 	case PRIV_NETATM_DEL:
3102 	case PRIV_NETATM_SET:
3103 
3104 		/*
3105 		 * Bluetooth privileges.
3106 		 */
3107 	case PRIV_NETBLUETOOTH_RAW:
3108 #endif
3109 
3110 		/*
3111 		 * Netgraph and netgraph module privileges.
3112 		 */
3113 	case PRIV_NETGRAPH_CONTROL:
3114 #ifdef notyet
3115 	case PRIV_NETGRAPH_TTY:
3116 #endif
3117 
3118 		/*
3119 		 * IPv4 and IPv6 privileges.
3120 		 */
3121 	case PRIV_NETINET_IPFW:
3122 	case PRIV_NETINET_DIVERT:
3123 	case PRIV_NETINET_PF:
3124 	case PRIV_NETINET_DUMMYNET:
3125 	case PRIV_NETINET_CARP:
3126 	case PRIV_NETINET_MROUTE:
3127 	case PRIV_NETINET_RAW:
3128 	case PRIV_NETINET_ADDRCTRL6:
3129 	case PRIV_NETINET_ND6:
3130 	case PRIV_NETINET_SCOPE6:
3131 	case PRIV_NETINET_ALIFETIME6:
3132 	case PRIV_NETINET_IPSEC:
3133 	case PRIV_NETINET_BINDANY:
3134 
3135 #ifdef notyet
3136 		/*
3137 		 * NCP privileges.
3138 		 */
3139 	case PRIV_NETNCP:
3140 
3141 		/*
3142 		 * SMB privileges.
3143 		 */
3144 	case PRIV_NETSMB:
3145 #endif
3146 
3147 	/*
3148 	 * No default: or deny here.
3149 	 * In case of no permit fall through to next switch().
3150 	 */
3151 		if (cred->cr_prison->pr_flags & PR_VNET)
3152 			return (0);
3153 	}
3154 #endif /* VIMAGE */
3155 
3156 	switch (priv) {
3157 
3158 		/*
3159 		 * Allow ktrace privileges for root in jail.
3160 		 */
3161 	case PRIV_KTRACE:
3162 
3163 #if 0
3164 		/*
3165 		 * Allow jailed processes to configure audit identity and
3166 		 * submit audit records (login, etc).  In the future we may
3167 		 * want to further refine the relationship between audit and
3168 		 * jail.
3169 		 */
3170 	case PRIV_AUDIT_GETAUDIT:
3171 	case PRIV_AUDIT_SETAUDIT:
3172 	case PRIV_AUDIT_SUBMIT:
3173 #endif
3174 
3175 		/*
3176 		 * Allow jailed processes to manipulate process UNIX
3177 		 * credentials in any way they see fit.
3178 		 */
3179 	case PRIV_CRED_SETUID:
3180 	case PRIV_CRED_SETEUID:
3181 	case PRIV_CRED_SETGID:
3182 	case PRIV_CRED_SETEGID:
3183 	case PRIV_CRED_SETGROUPS:
3184 	case PRIV_CRED_SETREUID:
3185 	case PRIV_CRED_SETREGID:
3186 	case PRIV_CRED_SETRESUID:
3187 	case PRIV_CRED_SETRESGID:
3188 
3189 		/*
3190 		 * Jail implements visibility constraints already, so allow
3191 		 * jailed root to override uid/gid-based constraints.
3192 		 */
3193 	case PRIV_SEEOTHERGIDS:
3194 	case PRIV_SEEOTHERUIDS:
3195 
3196 		/*
3197 		 * Jail implements inter-process debugging limits already, so
3198 		 * allow jailed root various debugging privileges.
3199 		 */
3200 	case PRIV_DEBUG_DIFFCRED:
3201 	case PRIV_DEBUG_SUGID:
3202 	case PRIV_DEBUG_UNPRIV:
3203 
3204 		/*
3205 		 * Allow jail to set various resource limits and login
3206 		 * properties, and for now, exceed process resource limits.
3207 		 */
3208 	case PRIV_PROC_LIMIT:
3209 	case PRIV_PROC_SETLOGIN:
3210 	case PRIV_PROC_SETRLIMIT:
3211 
3212 		/*
3213 		 * System V and POSIX IPC privileges are granted in jail.
3214 		 */
3215 	case PRIV_IPC_READ:
3216 	case PRIV_IPC_WRITE:
3217 	case PRIV_IPC_ADMIN:
3218 	case PRIV_IPC_MSGSIZE:
3219 	case PRIV_MQ_ADMIN:
3220 
3221 		/*
3222 		 * Jail operations within a jail work on child jails.
3223 		 */
3224 	case PRIV_JAIL_ATTACH:
3225 	case PRIV_JAIL_SET:
3226 	case PRIV_JAIL_REMOVE:
3227 
3228 		/*
3229 		 * Jail implements its own inter-process limits, so allow
3230 		 * root processes in jail to change scheduling on other
3231 		 * processes in the same jail.  Likewise for signalling.
3232 		 */
3233 	case PRIV_SCHED_DIFFCRED:
3234 	case PRIV_SCHED_CPUSET:
3235 	case PRIV_SIGNAL_DIFFCRED:
3236 	case PRIV_SIGNAL_SUGID:
3237 
3238 		/*
3239 		 * Allow jailed processes to write to sysctls marked as jail
3240 		 * writable.
3241 		 */
3242 	case PRIV_SYSCTL_WRITEJAIL:
3243 
3244 		/*
3245 		 * Allow root in jail to manage a variety of quota
3246 		 * properties.  These should likely be conditional on a
3247 		 * configuration option.
3248 		 */
3249 	case PRIV_VFS_GETQUOTA:
3250 	case PRIV_VFS_SETQUOTA:
3251 
3252 		/*
3253 		 * Since Jail relies on chroot() to implement file system
3254 		 * protections, grant many VFS privileges to root in jail.
3255 		 * Be careful to exclude mount-related and NFS-related
3256 		 * privileges.
3257 		 */
3258 	case PRIV_VFS_READ:
3259 	case PRIV_VFS_WRITE:
3260 	case PRIV_VFS_ADMIN:
3261 	case PRIV_VFS_EXEC:
3262 	case PRIV_VFS_LOOKUP:
3263 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3264 	case PRIV_VFS_CHFLAGS_DEV:
3265 	case PRIV_VFS_CHOWN:
3266 	case PRIV_VFS_CHROOT:
3267 	case PRIV_VFS_RETAINSUGID:
3268 	case PRIV_VFS_FCHROOT:
3269 	case PRIV_VFS_LINK:
3270 	case PRIV_VFS_SETGID:
3271 	case PRIV_VFS_STAT:
3272 	case PRIV_VFS_STICKYFILE:
3273 
3274 		/*
3275 		 * As in the non-jail case, non-root users are expected to be
3276 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3277 		 * exists in the jail and they have permission to access it).
3278 		 */
3279 	case PRIV_KMEM_READ:
3280 		return (0);
3281 
3282 		/*
3283 		 * Depending on the global setting, allow privilege of
3284 		 * setting system flags.
3285 		 */
3286 	case PRIV_VFS_SYSFLAGS:
3287 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3288 			return (0);
3289 		else
3290 			return (EPERM);
3291 
3292 		/*
3293 		 * Depending on the global setting, allow privilege of
3294 		 * mounting/unmounting file systems.
3295 		 */
3296 	case PRIV_VFS_MOUNT:
3297 	case PRIV_VFS_UNMOUNT:
3298 	case PRIV_VFS_MOUNT_NONUSER:
3299 	case PRIV_VFS_MOUNT_OWNER:
3300 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3301 		    cred->cr_prison->pr_enforce_statfs < 2)
3302 			return (0);
3303 		else
3304 			return (EPERM);
3305 
3306 		/*
3307 		 * Allow jailed root to bind reserved ports and reuse in-use
3308 		 * ports.
3309 		 */
3310 	case PRIV_NETINET_RESERVEDPORT:
3311 	case PRIV_NETINET_REUSEPORT:
3312 		return (0);
3313 
3314 		/*
3315 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3316 		 */
3317 	case PRIV_NETINET_SETHDROPTS:
3318 		return (0);
3319 
3320 		/*
3321 		 * Conditionally allow creating raw sockets in jail.
3322 		 */
3323 	case PRIV_NETINET_RAW:
3324 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3325 			return (0);
3326 		else
3327 			return (EPERM);
3328 
3329 		/*
3330 		 * Since jail implements its own visibility limits on netstat
3331 		 * sysctls, allow getcred.  This allows identd to work in
3332 		 * jail.
3333 		 */
3334 	case PRIV_NETINET_GETCRED:
3335 		return (0);
3336 
3337 		/*
3338 		 * Allow jailed root to set loginclass.
3339 		 */
3340 	case PRIV_PROC_SETLOGINCLASS:
3341 		return (0);
3342 
3343 	default:
3344 		/*
3345 		 * In all remaining cases, deny the privilege request.  This
3346 		 * includes almost all network privileges, many system
3347 		 * configuration privileges.
3348 		 */
3349 		return (EPERM);
3350 	}
3351 }
3352 
3353 /*
3354  * Return the part of pr2's name that is relative to pr1, or the whole name
3355  * if it does not directly follow.
3356  */
3357 
3358 char *
3359 prison_name(struct prison *pr1, struct prison *pr2)
3360 {
3361 	char *name;
3362 
3363 	/* Jails see themselves as "0" (if they see themselves at all). */
3364 	if (pr1 == pr2)
3365 		return "0";
3366 	name = pr2->pr_name;
3367 	if (prison_ischild(pr1, pr2)) {
3368 		/*
3369 		 * pr1 isn't locked (and allprison_lock may not be either)
3370 		 * so its length can't be counted on.  But the number of dots
3371 		 * can be counted on - and counted.
3372 		 */
3373 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3374 			name = strchr(name, '.') + 1;
3375 	}
3376 	return (name);
3377 }
3378 
3379 /*
3380  * Return the part of pr2's path that is relative to pr1, or the whole path
3381  * if it does not directly follow.
3382  */
3383 static char *
3384 prison_path(struct prison *pr1, struct prison *pr2)
3385 {
3386 	char *path1, *path2;
3387 	int len1;
3388 
3389 	path1 = pr1->pr_path;
3390 	path2 = pr2->pr_path;
3391 	if (!strcmp(path1, "/"))
3392 		return (path2);
3393 	len1 = strlen(path1);
3394 	if (strncmp(path1, path2, len1))
3395 		return (path2);
3396 	if (path2[len1] == '\0')
3397 		return "/";
3398 	if (path2[len1] == '/')
3399 		return (path2 + len1);
3400 	return (path2);
3401 }
3402 
3403 
3404 /*
3405  * Jail-related sysctls.
3406  */
3407 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3408     "Jails");
3409 
3410 static int
3411 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3412 {
3413 	struct xprison *xp;
3414 	struct prison *pr, *cpr;
3415 #ifdef INET
3416 	struct in_addr *ip4 = NULL;
3417 	int ip4s = 0;
3418 #endif
3419 #ifdef INET6
3420 	struct in6_addr *ip6 = NULL;
3421 	int ip6s = 0;
3422 #endif
3423 	int descend, error;
3424 
3425 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3426 	pr = req->td->td_ucred->cr_prison;
3427 	error = 0;
3428 	sx_slock(&allprison_lock);
3429 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3430 #if defined(INET) || defined(INET6)
3431  again:
3432 #endif
3433 		mtx_lock(&cpr->pr_mtx);
3434 #ifdef INET
3435 		if (cpr->pr_ip4s > 0) {
3436 			if (ip4s < cpr->pr_ip4s) {
3437 				ip4s = cpr->pr_ip4s;
3438 				mtx_unlock(&cpr->pr_mtx);
3439 				ip4 = realloc(ip4, ip4s *
3440 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3441 				goto again;
3442 			}
3443 			bcopy(cpr->pr_ip4, ip4,
3444 			    cpr->pr_ip4s * sizeof(struct in_addr));
3445 		}
3446 #endif
3447 #ifdef INET6
3448 		if (cpr->pr_ip6s > 0) {
3449 			if (ip6s < cpr->pr_ip6s) {
3450 				ip6s = cpr->pr_ip6s;
3451 				mtx_unlock(&cpr->pr_mtx);
3452 				ip6 = realloc(ip6, ip6s *
3453 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3454 				goto again;
3455 			}
3456 			bcopy(cpr->pr_ip6, ip6,
3457 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3458 		}
3459 #endif
3460 		if (cpr->pr_ref == 0) {
3461 			mtx_unlock(&cpr->pr_mtx);
3462 			continue;
3463 		}
3464 		bzero(xp, sizeof(*xp));
3465 		xp->pr_version = XPRISON_VERSION;
3466 		xp->pr_id = cpr->pr_id;
3467 		xp->pr_state = cpr->pr_uref > 0
3468 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3469 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3470 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3471 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3472 #ifdef INET
3473 		xp->pr_ip4s = cpr->pr_ip4s;
3474 #endif
3475 #ifdef INET6
3476 		xp->pr_ip6s = cpr->pr_ip6s;
3477 #endif
3478 		mtx_unlock(&cpr->pr_mtx);
3479 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3480 		if (error)
3481 			break;
3482 #ifdef INET
3483 		if (xp->pr_ip4s > 0) {
3484 			error = SYSCTL_OUT(req, ip4,
3485 			    xp->pr_ip4s * sizeof(struct in_addr));
3486 			if (error)
3487 				break;
3488 		}
3489 #endif
3490 #ifdef INET6
3491 		if (xp->pr_ip6s > 0) {
3492 			error = SYSCTL_OUT(req, ip6,
3493 			    xp->pr_ip6s * sizeof(struct in6_addr));
3494 			if (error)
3495 				break;
3496 		}
3497 #endif
3498 	}
3499 	sx_sunlock(&allprison_lock);
3500 	free(xp, M_TEMP);
3501 #ifdef INET
3502 	free(ip4, M_TEMP);
3503 #endif
3504 #ifdef INET6
3505 	free(ip6, M_TEMP);
3506 #endif
3507 	return (error);
3508 }
3509 
3510 SYSCTL_OID(_security_jail, OID_AUTO, list,
3511     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3512     sysctl_jail_list, "S", "List of active jails");
3513 
3514 static int
3515 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3516 {
3517 	int error, injail;
3518 
3519 	injail = jailed(req->td->td_ucred);
3520 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3521 
3522 	return (error);
3523 }
3524 
3525 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3526     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3527     sysctl_jail_jailed, "I", "Process in jail?");
3528 
3529 static int
3530 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3531 {
3532 	int error, havevnet;
3533 #ifdef VIMAGE
3534 	struct ucred *cred = req->td->td_ucred;
3535 
3536 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3537 #else
3538 	havevnet = 0;
3539 #endif
3540 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3541 
3542 	return (error);
3543 }
3544 
3545 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3546     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3547     sysctl_jail_vnet, "I", "Jail owns VNET?");
3548 
3549 #if defined(INET) || defined(INET6)
3550 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3551     &jail_max_af_ips, 0,
3552     "Number of IP addresses a jail may have at most per address family (deprecated)");
3553 #endif
3554 
3555 /*
3556  * Default parameters for jail(2) compatibility.  For historical reasons,
3557  * the sysctl names have varying similarity to the parameter names.  Prisons
3558  * just see their own parameters, and can't change them.
3559  */
3560 static int
3561 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3562 {
3563 	struct prison *pr;
3564 	int allow, error, i;
3565 
3566 	pr = req->td->td_ucred->cr_prison;
3567 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3568 
3569 	/* Get the current flag value, and convert it to a boolean. */
3570 	i = (allow & arg2) ? 1 : 0;
3571 	if (arg1 != NULL)
3572 		i = !i;
3573 	error = sysctl_handle_int(oidp, &i, 0, req);
3574 	if (error || !req->newptr)
3575 		return (error);
3576 	i = i ? arg2 : 0;
3577 	if (arg1 != NULL)
3578 		i ^= arg2;
3579 	/*
3580 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3581 	 * for writing.
3582 	 */
3583 	mtx_lock(&prison0.pr_mtx);
3584 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3585 	mtx_unlock(&prison0.pr_mtx);
3586 	return (0);
3587 }
3588 
3589 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3590     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3591     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3592     "Processes in jail can set their hostnames (deprecated)");
3593 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3594     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3595     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3596     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3597 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3598     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3599     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3600     "Processes in jail can use System V IPC primitives (deprecated)");
3601 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3602     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3603     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3604     "Prison root can create raw sockets (deprecated)");
3605 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3606     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3607     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3608     "Processes in jail can alter system file flags (deprecated)");
3609 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3610     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3611     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3612     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3613 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
3614     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3615     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
3616     "Processes in jail can mount the devfs file system (deprecated)");
3617 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
3618     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3619     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
3620     "Processes in jail can mount the fdescfs file system (deprecated)");
3621 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
3622     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3623     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
3624     "Processes in jail can mount the nullfs file system (deprecated)");
3625 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
3626     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3627     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
3628     "Processes in jail can mount the procfs file system (deprecated)");
3629 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
3630     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3631     NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
3632     "Processes in jail can mount the linprocfs file system (deprecated)");
3633 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
3634     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3635     NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
3636     "Processes in jail can mount the linsysfs file system (deprecated)");
3637 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
3638     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3639     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
3640     "Processes in jail can mount the tmpfs file system (deprecated)");
3641 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
3642     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3643     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
3644     "Processes in jail can mount the zfs file system (deprecated)");
3645 
3646 static int
3647 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3648 {
3649 	struct prison *pr;
3650 	int level, error;
3651 
3652 	pr = req->td->td_ucred->cr_prison;
3653 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3654 	error = sysctl_handle_int(oidp, &level, 0, req);
3655 	if (error || !req->newptr)
3656 		return (error);
3657 	*(int *)arg1 = level;
3658 	return (0);
3659 }
3660 
3661 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3662     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3663     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3664     sysctl_jail_default_level, "I",
3665     "Processes in jail cannot see all mounted file systems (deprecated)");
3666 
3667 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3668     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3669     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3670     sysctl_jail_default_level, "I",
3671     "Ruleset for the devfs filesystem in jail (deprecated)");
3672 
3673 /*
3674  * Nodes to describe jail parameters.  Maximum length of string parameters
3675  * is returned in the string itself, and the other parameters exist merely
3676  * to make themselves and their types known.
3677  */
3678 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3679     "Jail parameters");
3680 
3681 int
3682 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3683 {
3684 	int i;
3685 	long l;
3686 	size_t s;
3687 	char numbuf[12];
3688 
3689 	switch (oidp->oid_kind & CTLTYPE)
3690 	{
3691 	case CTLTYPE_LONG:
3692 	case CTLTYPE_ULONG:
3693 		l = 0;
3694 #ifdef SCTL_MASK32
3695 		if (!(req->flags & SCTL_MASK32))
3696 #endif
3697 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3698 	case CTLTYPE_INT:
3699 	case CTLTYPE_UINT:
3700 		i = 0;
3701 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3702 	case CTLTYPE_STRING:
3703 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3704 		return
3705 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3706 	case CTLTYPE_STRUCT:
3707 		s = (size_t)arg2;
3708 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3709 	}
3710 	return (0);
3711 }
3712 
3713 /*
3714  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3715  * jail creation time but cannot be changed in an existing jail.
3716  */
3717 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3718 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3719 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3720 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3721 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3722     "I", "Jail secure level");
3723 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3724     "Jail value for kern.osreldate and uname -K");
3725 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3726     "Jail value for kern.osrelease and uname -r");
3727 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3728     "I", "Jail cannot see all mounted file systems");
3729 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3730     "I", "Ruleset for in-jail devfs mounts");
3731 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3732     "B", "Jail persistence");
3733 #ifdef VIMAGE
3734 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3735     "E,jailsys", "Virtual network stack");
3736 #endif
3737 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3738     "B", "Jail is in the process of shutting down");
3739 
3740 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3741 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3742     "I", "Current number of child jails");
3743 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3744     "I", "Maximum number of child jails");
3745 
3746 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3747 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3748     "Jail hostname");
3749 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3750     "Jail NIS domainname");
3751 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3752     "Jail host UUID");
3753 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3754     "LU", "Jail host ID");
3755 
3756 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3757 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3758 
3759 #ifdef INET
3760 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3761     "Jail IPv4 address virtualization");
3762 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3763     "S,in_addr,a", "Jail IPv4 addresses");
3764 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3765     "B", "Do (not) use IPv4 source address selection rather than the "
3766     "primary jail IPv4 address.");
3767 #endif
3768 #ifdef INET6
3769 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3770     "Jail IPv6 address virtualization");
3771 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3772     "S,in6_addr,a", "Jail IPv6 addresses");
3773 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3774     "B", "Do (not) use IPv6 source address selection rather than the "
3775     "primary jail IPv6 address.");
3776 #endif
3777 
3778 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3779 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3780     "B", "Jail may set hostname");
3781 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3782     "B", "Jail may use SYSV IPC");
3783 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3784     "B", "Jail may create raw sockets");
3785 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3786     "B", "Jail may alter system file flags");
3787 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3788     "B", "Jail may set file quotas");
3789 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3790     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3791 
3792 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3793 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3794     "B", "Jail may mount/unmount jail-friendly file systems in general");
3795 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
3796     "B", "Jail may mount the devfs file system");
3797 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
3798     "B", "Jail may mount the fdescfs file system");
3799 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
3800     "B", "Jail may mount the nullfs file system");
3801 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
3802     "B", "Jail may mount the procfs file system");
3803 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
3804     "B", "Jail may mount the linprocfs file system");
3805 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
3806     "B", "Jail may mount the linsysfs file system");
3807 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
3808     "B", "Jail may mount the tmpfs file system");
3809 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
3810     "B", "Jail may mount the zfs file system");
3811 
3812 #ifdef RACCT
3813 void
3814 prison_racct_foreach(void (*callback)(struct racct *racct,
3815     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3816     void *arg2, void *arg3)
3817 {
3818 	struct prison_racct *prr;
3819 
3820 	ASSERT_RACCT_ENABLED();
3821 
3822 	sx_slock(&allprison_lock);
3823 	if (pre != NULL)
3824 		(pre)();
3825 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3826 		(callback)(prr->prr_racct, arg2, arg3);
3827 	if (post != NULL)
3828 		(post)();
3829 	sx_sunlock(&allprison_lock);
3830 }
3831 
3832 static struct prison_racct *
3833 prison_racct_find_locked(const char *name)
3834 {
3835 	struct prison_racct *prr;
3836 
3837 	ASSERT_RACCT_ENABLED();
3838 	sx_assert(&allprison_lock, SA_XLOCKED);
3839 
3840 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3841 		return (NULL);
3842 
3843 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3844 		if (strcmp(name, prr->prr_name) != 0)
3845 			continue;
3846 
3847 		/* Found prison_racct with a matching name? */
3848 		prison_racct_hold(prr);
3849 		return (prr);
3850 	}
3851 
3852 	/* Add new prison_racct. */
3853 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3854 	racct_create(&prr->prr_racct);
3855 
3856 	strcpy(prr->prr_name, name);
3857 	refcount_init(&prr->prr_refcount, 1);
3858 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3859 
3860 	return (prr);
3861 }
3862 
3863 struct prison_racct *
3864 prison_racct_find(const char *name)
3865 {
3866 	struct prison_racct *prr;
3867 
3868 	ASSERT_RACCT_ENABLED();
3869 
3870 	sx_xlock(&allprison_lock);
3871 	prr = prison_racct_find_locked(name);
3872 	sx_xunlock(&allprison_lock);
3873 	return (prr);
3874 }
3875 
3876 void
3877 prison_racct_hold(struct prison_racct *prr)
3878 {
3879 
3880 	ASSERT_RACCT_ENABLED();
3881 
3882 	refcount_acquire(&prr->prr_refcount);
3883 }
3884 
3885 static void
3886 prison_racct_free_locked(struct prison_racct *prr)
3887 {
3888 
3889 	ASSERT_RACCT_ENABLED();
3890 	sx_assert(&allprison_lock, SA_XLOCKED);
3891 
3892 	if (refcount_release(&prr->prr_refcount)) {
3893 		racct_destroy(&prr->prr_racct);
3894 		LIST_REMOVE(prr, prr_next);
3895 		free(prr, M_PRISON_RACCT);
3896 	}
3897 }
3898 
3899 void
3900 prison_racct_free(struct prison_racct *prr)
3901 {
3902 	int old;
3903 
3904 	ASSERT_RACCT_ENABLED();
3905 	sx_assert(&allprison_lock, SA_UNLOCKED);
3906 
3907 	old = prr->prr_refcount;
3908 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
3909 		return;
3910 
3911 	sx_xlock(&allprison_lock);
3912 	prison_racct_free_locked(prr);
3913 	sx_xunlock(&allprison_lock);
3914 }
3915 
3916 static void
3917 prison_racct_attach(struct prison *pr)
3918 {
3919 	struct prison_racct *prr;
3920 
3921 	ASSERT_RACCT_ENABLED();
3922 	sx_assert(&allprison_lock, SA_XLOCKED);
3923 
3924 	prr = prison_racct_find_locked(pr->pr_name);
3925 	KASSERT(prr != NULL, ("cannot find prison_racct"));
3926 
3927 	pr->pr_prison_racct = prr;
3928 }
3929 
3930 /*
3931  * Handle jail renaming.  From the racct point of view, renaming means
3932  * moving from one prison_racct to another.
3933  */
3934 static void
3935 prison_racct_modify(struct prison *pr)
3936 {
3937 	struct proc *p;
3938 	struct ucred *cred;
3939 	struct prison_racct *oldprr;
3940 
3941 	ASSERT_RACCT_ENABLED();
3942 
3943 	sx_slock(&allproc_lock);
3944 	sx_xlock(&allprison_lock);
3945 
3946 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
3947 		sx_xunlock(&allprison_lock);
3948 		sx_sunlock(&allproc_lock);
3949 		return;
3950 	}
3951 
3952 	oldprr = pr->pr_prison_racct;
3953 	pr->pr_prison_racct = NULL;
3954 
3955 	prison_racct_attach(pr);
3956 
3957 	/*
3958 	 * Move resource utilisation records.
3959 	 */
3960 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
3961 
3962 	/*
3963 	 * Force rctl to reattach rules to processes.
3964 	 */
3965 	FOREACH_PROC_IN_SYSTEM(p) {
3966 		PROC_LOCK(p);
3967 		cred = crhold(p->p_ucred);
3968 		PROC_UNLOCK(p);
3969 		racct_proc_ucred_changed(p, cred, cred);
3970 		crfree(cred);
3971 	}
3972 
3973 	sx_sunlock(&allproc_lock);
3974 	prison_racct_free_locked(oldprr);
3975 	sx_xunlock(&allprison_lock);
3976 }
3977 
3978 static void
3979 prison_racct_detach(struct prison *pr)
3980 {
3981 
3982 	ASSERT_RACCT_ENABLED();
3983 	sx_assert(&allprison_lock, SA_UNLOCKED);
3984 
3985 	if (pr->pr_prison_racct == NULL)
3986 		return;
3987 	prison_racct_free(pr->pr_prison_racct);
3988 	pr->pr_prison_racct = NULL;
3989 }
3990 #endif /* RACCT */
3991 
3992 #ifdef DDB
3993 
3994 static void
3995 db_show_prison(struct prison *pr)
3996 {
3997 	int fi;
3998 #if defined(INET) || defined(INET6)
3999 	int ii;
4000 #endif
4001 	unsigned jsf;
4002 #ifdef INET6
4003 	char ip6buf[INET6_ADDRSTRLEN];
4004 #endif
4005 
4006 	db_printf("prison %p:\n", pr);
4007 	db_printf(" jid             = %d\n", pr->pr_id);
4008 	db_printf(" name            = %s\n", pr->pr_name);
4009 	db_printf(" parent          = %p\n", pr->pr_parent);
4010 	db_printf(" ref             = %d\n", pr->pr_ref);
4011 	db_printf(" uref            = %d\n", pr->pr_uref);
4012 	db_printf(" path            = %s\n", pr->pr_path);
4013 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4014 	    ? pr->pr_cpuset->cs_id : -1);
4015 #ifdef VIMAGE
4016 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4017 #endif
4018 	db_printf(" root            = %p\n", pr->pr_root);
4019 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4020 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4021 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4022 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4023 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4024 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4025 	db_printf(" flags           = 0x%x", pr->pr_flags);
4026 	for (fi = 0; fi < nitems(pr_flag_names); fi++)
4027 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4028 			db_printf(" %s", pr_flag_names[fi]);
4029 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4030 		jsf = pr->pr_flags &
4031 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4032 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4033 		    pr_flag_jailsys[fi].disable &&
4034 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4035 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4036 		    : "inherit");
4037 	}
4038 	db_printf(" allow           = 0x%x", pr->pr_allow);
4039 	for (fi = 0; fi < nitems(pr_allow_names); fi++)
4040 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4041 			db_printf(" %s", pr_allow_names[fi]);
4042 	db_printf("\n");
4043 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4044 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4045 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4046 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4047 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4048 #ifdef INET
4049 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4050 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4051 		db_printf(" %s %s\n",
4052 		    ii == 0 ? "ip4.addr        =" : "                 ",
4053 		    inet_ntoa(pr->pr_ip4[ii]));
4054 #endif
4055 #ifdef INET6
4056 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4057 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4058 		db_printf(" %s %s\n",
4059 		    ii == 0 ? "ip6.addr        =" : "                 ",
4060 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4061 #endif
4062 }
4063 
4064 DB_SHOW_COMMAND(prison, db_show_prison_command)
4065 {
4066 	struct prison *pr;
4067 
4068 	if (!have_addr) {
4069 		/*
4070 		 * Show all prisons in the list, and prison0 which is not
4071 		 * listed.
4072 		 */
4073 		db_show_prison(&prison0);
4074 		if (!db_pager_quit) {
4075 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4076 				db_show_prison(pr);
4077 				if (db_pager_quit)
4078 					break;
4079 			}
4080 		}
4081 		return;
4082 	}
4083 
4084 	if (addr == 0)
4085 		pr = &prison0;
4086 	else {
4087 		/* Look for a prison with the ID and with references. */
4088 		TAILQ_FOREACH(pr, &allprison, pr_list)
4089 			if (pr->pr_id == addr && pr->pr_ref > 0)
4090 				break;
4091 		if (pr == NULL)
4092 			/* Look again, without requiring a reference. */
4093 			TAILQ_FOREACH(pr, &allprison, pr_list)
4094 				if (pr->pr_id == addr)
4095 					break;
4096 		if (pr == NULL)
4097 			/* Assume address points to a valid prison. */
4098 			pr = (struct prison *)addr;
4099 	}
4100 	db_show_prison(pr);
4101 }
4102 
4103 #endif /* DDB */
4104