xref: /freebsd/sys/kern/kern_jail.c (revision 5bb3134a8c21cb87b30e135ef168483f0333dabb)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/linker.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/racct.h>
55 #include <sys/rctl.h>
56 #include <sys/refcount.h>
57 #include <sys/sx.h>
58 #include <sys/sysent.h>
59 #include <sys/namei.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysctl.h>
65 #include <sys/uuid.h>
66 #include <sys/vnode.h>
67 
68 #include <net/if.h>
69 #include <net/vnet.h>
70 
71 #include <netinet/in.h>
72 
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif /* DDB */
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
80 
81 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
82 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
83 
84 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
85 #ifdef INET
86 #ifdef INET6
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
88 #else
89 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
90 #endif
91 #else /* !INET */
92 #ifdef INET6
93 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
94 #else
95 #define	_PR_IP_SADDRSEL	0
96 #endif
97 #endif
98 
99 /* prison0 describes what is "real" about the system. */
100 struct prison prison0 = {
101 	.pr_id		= 0,
102 	.pr_name	= "0",
103 	.pr_ref		= 1,
104 	.pr_uref	= 1,
105 	.pr_path	= "/",
106 	.pr_securelevel	= -1,
107 	.pr_devfs_rsnum = 0,
108 	.pr_state	= PRISON_STATE_ALIVE,
109 	.pr_childmax	= JAIL_MAX,
110 	.pr_hostuuid	= DEFAULT_HOSTUUID,
111 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
112 #ifdef VIMAGE
113 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
114 #else
115 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
116 #endif
117 	.pr_allow	= PR_ALLOW_ALL_STATIC,
118 };
119 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
120 
121 struct bool_flags {
122 	const char	*name;
123 	const char	*noname;
124 	volatile u_int	 flag;
125 };
126 struct jailsys_flags {
127 	const char	*name;
128 	unsigned	 disable;
129 	unsigned	 new;
130 };
131 
132 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
133 struct	sx allprison_lock;
134 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
135 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
136 LIST_HEAD(, prison_racct) allprison_racct;
137 int	lastprid = 0;
138 
139 static int get_next_prid(struct prison **insprp);
140 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
141 static void prison_complete(void *context, int pending);
142 static void prison_deref(struct prison *pr, int flags);
143 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
144 static int prison_lock_xlock(struct prison *pr, int flags);
145 static void prison_free_not_last(struct prison *pr);
146 static void prison_proc_free_not_last(struct prison *pr);
147 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
148     int enable);
149 static char *prison_path(struct prison *pr1, struct prison *pr2);
150 #ifdef RACCT
151 static void prison_racct_attach(struct prison *pr);
152 static void prison_racct_modify(struct prison *pr);
153 static void prison_racct_detach(struct prison *pr);
154 #endif
155 
156 /* Flags for prison_deref */
157 #define	PD_DEREF	0x01	/* Decrement pr_ref */
158 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
159 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
160 #define	PD_LOCKED	0x10	/* pr_mtx is held */
161 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
162 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
163 #define PD_OP_FLAGS	0x07	/* Operation flags */
164 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
165 
166 /*
167  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
168  * as we cannot figure out the size of a sparse array, or an array without a
169  * terminating entry.
170  */
171 static struct bool_flags pr_flag_bool[] = {
172 	{"persist", "nopersist", PR_PERSIST},
173 #ifdef INET
174 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
175 #endif
176 #ifdef INET6
177 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
178 #endif
179 };
180 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
181 
182 static struct jailsys_flags pr_flag_jailsys[] = {
183 	{"host", 0, PR_HOST},
184 #ifdef VIMAGE
185 	{"vnet", 0, PR_VNET},
186 #endif
187 #ifdef INET
188 	{"ip4", PR_IP4_USER, PR_IP4_USER},
189 #endif
190 #ifdef INET6
191 	{"ip6", PR_IP6_USER, PR_IP6_USER},
192 #endif
193 };
194 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
195 
196 /*
197  * Make this array full-size so dynamic parameters can be added.
198  * It is protected by prison0.mtx, but lockless reading is allowed
199  * with an atomic check of the flag values.
200  */
201 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
202 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
203 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
204 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
205 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
206 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
207 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
208 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
209 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
210 	{"allow.reserved_ports", "allow.noreserved_ports",
211 	 PR_ALLOW_RESERVED_PORTS},
212 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
213 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
214 	 PR_ALLOW_UNPRIV_DEBUG},
215 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
216 };
217 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
218 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
219 
220 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
221 					 PR_ALLOW_RESERVED_PORTS | \
222 					 PR_ALLOW_UNPRIV_DEBUG | \
223 					 PR_ALLOW_SUSER)
224 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
225 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
226 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
227 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
228 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
229 #if defined(INET) || defined(INET6)
230 static unsigned jail_max_af_ips = 255;
231 #endif
232 
233 /*
234  * Initialize the parts of prison0 that can't be static-initialized with
235  * constants.  This is called from proc0_init() after creating thread0 cpuset.
236  */
237 void
238 prison0_init(void)
239 {
240 	uint8_t *file, *data;
241 	size_t size;
242 
243 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
244 	prison0.pr_osreldate = osreldate;
245 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
246 
247 	/* If we have a preloaded hostuuid, use it. */
248 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
249 	if (file != NULL) {
250 		data = preload_fetch_addr(file);
251 		size = preload_fetch_size(file);
252 		if (data != NULL) {
253 			/*
254 			 * The preloaded data may include trailing whitespace, almost
255 			 * certainly a newline; skip over any whitespace or
256 			 * non-printable characters to be safe.
257 			 */
258 			while (size > 0 && data[size - 1] <= 0x20) {
259 				size--;
260 			}
261 			if (validate_uuid(data, size, NULL, 0) == 0) {
262 				(void)strlcpy(prison0.pr_hostuuid, data,
263 				    size + 1);
264 			} else if (bootverbose) {
265 				printf("hostuuid: preload data malformed: '%.*s'\n",
266 				    (int)size, data);
267 			}
268 		}
269 	}
270 	if (bootverbose)
271 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
272 }
273 
274 /*
275  * struct jail_args {
276  *	struct jail *jail;
277  * };
278  */
279 int
280 sys_jail(struct thread *td, struct jail_args *uap)
281 {
282 	uint32_t version;
283 	int error;
284 	struct jail j;
285 
286 	error = copyin(uap->jail, &version, sizeof(uint32_t));
287 	if (error)
288 		return (error);
289 
290 	switch (version) {
291 	case 0:
292 	{
293 		struct jail_v0 j0;
294 
295 		/* FreeBSD single IPv4 jails. */
296 		bzero(&j, sizeof(struct jail));
297 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
298 		if (error)
299 			return (error);
300 		j.version = j0.version;
301 		j.path = j0.path;
302 		j.hostname = j0.hostname;
303 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
304 		break;
305 	}
306 
307 	case 1:
308 		/*
309 		 * Version 1 was used by multi-IPv4 jail implementations
310 		 * that never made it into the official kernel.
311 		 */
312 		return (EINVAL);
313 
314 	case 2:	/* JAIL_API_VERSION */
315 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
316 		error = copyin(uap->jail, &j, sizeof(struct jail));
317 		if (error)
318 			return (error);
319 		break;
320 
321 	default:
322 		/* Sci-Fi jails are not supported, sorry. */
323 		return (EINVAL);
324 	}
325 	return (kern_jail(td, &j));
326 }
327 
328 int
329 kern_jail(struct thread *td, struct jail *j)
330 {
331 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
332 #ifdef INET
333 			    + 1
334 #endif
335 #ifdef INET6
336 			    + 1
337 #endif
338 			    )];
339 	struct uio opt;
340 	char *u_path, *u_hostname, *u_name;
341 	struct bool_flags *bf;
342 #ifdef INET
343 	uint32_t ip4s;
344 	struct in_addr *u_ip4;
345 #endif
346 #ifdef INET6
347 	struct in6_addr *u_ip6;
348 #endif
349 	size_t tmplen;
350 	int error, enforce_statfs;
351 
352 	bzero(&optiov, sizeof(optiov));
353 	opt.uio_iov = optiov;
354 	opt.uio_iovcnt = 0;
355 	opt.uio_offset = -1;
356 	opt.uio_resid = -1;
357 	opt.uio_segflg = UIO_SYSSPACE;
358 	opt.uio_rw = UIO_READ;
359 	opt.uio_td = td;
360 
361 	/* Set permissions for top-level jails from sysctls. */
362 	if (!jailed(td->td_ucred)) {
363 		for (bf = pr_flag_allow;
364 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
365 			atomic_load_int(&bf->flag) != 0;
366 		     bf++) {
367 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
368 			    (jail_default_allow & bf->flag)
369 			    ? bf->name : bf->noname);
370 			optiov[opt.uio_iovcnt].iov_len =
371 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
372 			opt.uio_iovcnt += 2;
373 		}
374 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
375 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
376 		opt.uio_iovcnt++;
377 		enforce_statfs = jail_default_enforce_statfs;
378 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
379 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
380 		opt.uio_iovcnt++;
381 	}
382 
383 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
384 #ifdef INET
385 	ip4s = (j->version == 0) ? 1 : j->ip4s;
386 	if (ip4s > jail_max_af_ips)
387 		return (EINVAL);
388 	tmplen += ip4s * sizeof(struct in_addr);
389 #else
390 	if (j->ip4s > 0)
391 		return (EINVAL);
392 #endif
393 #ifdef INET6
394 	if (j->ip6s > jail_max_af_ips)
395 		return (EINVAL);
396 	tmplen += j->ip6s * sizeof(struct in6_addr);
397 #else
398 	if (j->ip6s > 0)
399 		return (EINVAL);
400 #endif
401 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
402 	u_hostname = u_path + MAXPATHLEN;
403 	u_name = u_hostname + MAXHOSTNAMELEN;
404 #ifdef INET
405 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
406 #endif
407 #ifdef INET6
408 #ifdef INET
409 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
410 #else
411 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
412 #endif
413 #endif
414 	optiov[opt.uio_iovcnt].iov_base = "path";
415 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
416 	opt.uio_iovcnt++;
417 	optiov[opt.uio_iovcnt].iov_base = u_path;
418 	error = copyinstr(j->path, u_path, MAXPATHLEN,
419 	    &optiov[opt.uio_iovcnt].iov_len);
420 	if (error) {
421 		free(u_path, M_TEMP);
422 		return (error);
423 	}
424 	opt.uio_iovcnt++;
425 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
426 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
427 	opt.uio_iovcnt++;
428 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
429 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
430 	    &optiov[opt.uio_iovcnt].iov_len);
431 	if (error) {
432 		free(u_path, M_TEMP);
433 		return (error);
434 	}
435 	opt.uio_iovcnt++;
436 	if (j->jailname != NULL) {
437 		optiov[opt.uio_iovcnt].iov_base = "name";
438 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
439 		opt.uio_iovcnt++;
440 		optiov[opt.uio_iovcnt].iov_base = u_name;
441 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
442 		    &optiov[opt.uio_iovcnt].iov_len);
443 		if (error) {
444 			free(u_path, M_TEMP);
445 			return (error);
446 		}
447 		opt.uio_iovcnt++;
448 	}
449 #ifdef INET
450 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
451 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
452 	opt.uio_iovcnt++;
453 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
454 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
455 	if (j->version == 0)
456 		u_ip4->s_addr = j->ip4s;
457 	else {
458 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
459 		if (error) {
460 			free(u_path, M_TEMP);
461 			return (error);
462 		}
463 	}
464 	opt.uio_iovcnt++;
465 #endif
466 #ifdef INET6
467 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
468 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
469 	opt.uio_iovcnt++;
470 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
471 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
472 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
473 	if (error) {
474 		free(u_path, M_TEMP);
475 		return (error);
476 	}
477 	opt.uio_iovcnt++;
478 #endif
479 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
480 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
481 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
482 	free(u_path, M_TEMP);
483 	return (error);
484 }
485 
486 /*
487  * struct jail_set_args {
488  *	struct iovec *iovp;
489  *	unsigned int iovcnt;
490  *	int flags;
491  * };
492  */
493 int
494 sys_jail_set(struct thread *td, struct jail_set_args *uap)
495 {
496 	struct uio *auio;
497 	int error;
498 
499 	/* Check that we have an even number of iovecs. */
500 	if (uap->iovcnt & 1)
501 		return (EINVAL);
502 
503 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
504 	if (error)
505 		return (error);
506 	error = kern_jail_set(td, auio, uap->flags);
507 	free(auio, M_IOV);
508 	return (error);
509 }
510 
511 int
512 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
513 {
514 	struct nameidata nd;
515 #ifdef INET
516 	struct in_addr *ip4;
517 #endif
518 #ifdef INET6
519 	struct in6_addr *ip6;
520 #endif
521 	struct vfsopt *opt;
522 	struct vfsoptlist *opts;
523 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
524 	struct vnode *root;
525 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
526 	char *g_path, *osrelstr;
527 	struct bool_flags *bf;
528 	struct jailsys_flags *jsf;
529 #if defined(INET) || defined(INET6)
530 	struct prison *tppr;
531 	void *op;
532 #endif
533 	unsigned long hid;
534 	size_t namelen, onamelen, pnamelen;
535 	int born, created, cuflags, descend, drflags, enforce;
536 	int error, errmsg_len, errmsg_pos;
537 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
538 	int jid, jsys, len, level;
539 	int childmax, osreldt, rsnum, slevel;
540 #if defined(INET) || defined(INET6)
541 	int ii, ij;
542 #endif
543 #ifdef INET
544 	int ip4s, redo_ip4;
545 #endif
546 #ifdef INET6
547 	int ip6s, redo_ip6;
548 #endif
549 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
550 	uint64_t pr_allow_diff;
551 	unsigned tallow;
552 	char numbuf[12];
553 
554 	error = priv_check(td, PRIV_JAIL_SET);
555 	if (!error && (flags & JAIL_ATTACH))
556 		error = priv_check(td, PRIV_JAIL_ATTACH);
557 	if (error)
558 		return (error);
559 	mypr = td->td_ucred->cr_prison;
560 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
561 		return (EPERM);
562 	if (flags & ~JAIL_SET_MASK)
563 		return (EINVAL);
564 
565 	/*
566 	 * Check all the parameters before committing to anything.  Not all
567 	 * errors can be caught early, but we may as well try.  Also, this
568 	 * takes care of some expensive stuff (path lookup) before getting
569 	 * the allprison lock.
570 	 *
571 	 * XXX Jails are not filesystems, and jail parameters are not mount
572 	 *     options.  But it makes more sense to re-use the vfsopt code
573 	 *     than duplicate it under a different name.
574 	 */
575 	error = vfs_buildopts(optuio, &opts);
576 	if (error)
577 		return (error);
578 #ifdef INET
579 	ip4 = NULL;
580 #endif
581 #ifdef INET6
582 	ip6 = NULL;
583 #endif
584 	g_path = NULL;
585 
586 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
587 	if (!cuflags) {
588 		error = EINVAL;
589 		vfs_opterror(opts, "no valid operation (create or update)");
590 		goto done_errmsg;
591 	}
592 
593 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
594 	if (error == ENOENT)
595 		jid = 0;
596 	else if (error != 0)
597 		goto done_free;
598 
599 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
600 	if (error == ENOENT)
601 		gotslevel = 0;
602 	else if (error != 0)
603 		goto done_free;
604 	else
605 		gotslevel = 1;
606 
607 	error =
608 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
609 	if (error == ENOENT)
610 		gotchildmax = 0;
611 	else if (error != 0)
612 		goto done_free;
613 	else
614 		gotchildmax = 1;
615 
616 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
617 	if (error == ENOENT)
618 		gotenforce = 0;
619 	else if (error != 0)
620 		goto done_free;
621 	else if (enforce < 0 || enforce > 2) {
622 		error = EINVAL;
623 		goto done_free;
624 	} else
625 		gotenforce = 1;
626 
627 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
628 	if (error == ENOENT)
629 		gotrsnum = 0;
630 	else if (error != 0)
631 		goto done_free;
632 	else
633 		gotrsnum = 1;
634 
635 	pr_flags = ch_flags = 0;
636 	for (bf = pr_flag_bool;
637 	     bf < pr_flag_bool + nitems(pr_flag_bool);
638 	     bf++) {
639 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
640 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
641 	}
642 	ch_flags |= pr_flags;
643 	for (jsf = pr_flag_jailsys;
644 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
645 	     jsf++) {
646 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
647 		if (error == ENOENT)
648 			continue;
649 		if (error != 0)
650 			goto done_free;
651 		switch (jsys) {
652 		case JAIL_SYS_DISABLE:
653 			if (!jsf->disable) {
654 				error = EINVAL;
655 				goto done_free;
656 			}
657 			pr_flags |= jsf->disable;
658 			break;
659 		case JAIL_SYS_NEW:
660 			pr_flags |= jsf->new;
661 			break;
662 		case JAIL_SYS_INHERIT:
663 			break;
664 		default:
665 			error = EINVAL;
666 			goto done_free;
667 		}
668 		ch_flags |= jsf->new | jsf->disable;
669 	}
670 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
671 	    && !(pr_flags & PR_PERSIST)) {
672 		error = EINVAL;
673 		vfs_opterror(opts, "new jail must persist or attach");
674 		goto done_errmsg;
675 	}
676 #ifdef VIMAGE
677 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
678 		error = EINVAL;
679 		vfs_opterror(opts, "vnet cannot be changed after creation");
680 		goto done_errmsg;
681 	}
682 #endif
683 #ifdef INET
684 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
685 		error = EINVAL;
686 		vfs_opterror(opts, "ip4 cannot be changed after creation");
687 		goto done_errmsg;
688 	}
689 #endif
690 #ifdef INET6
691 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
692 		error = EINVAL;
693 		vfs_opterror(opts, "ip6 cannot be changed after creation");
694 		goto done_errmsg;
695 	}
696 #endif
697 
698 	pr_allow = ch_allow = 0;
699 	for (bf = pr_flag_allow;
700 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
701 		atomic_load_int(&bf->flag) != 0;
702 	     bf++) {
703 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
704 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
705 	}
706 	ch_allow |= pr_allow;
707 
708 	error = vfs_getopt(opts, "name", (void **)&name, &len);
709 	if (error == ENOENT)
710 		name = NULL;
711 	else if (error != 0)
712 		goto done_free;
713 	else {
714 		if (len == 0 || name[len - 1] != '\0') {
715 			error = EINVAL;
716 			goto done_free;
717 		}
718 		if (len > MAXHOSTNAMELEN) {
719 			error = ENAMETOOLONG;
720 			goto done_free;
721 		}
722 	}
723 
724 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
725 	if (error == ENOENT)
726 		host = NULL;
727 	else if (error != 0)
728 		goto done_free;
729 	else {
730 		ch_flags |= PR_HOST;
731 		pr_flags |= PR_HOST;
732 		if (len == 0 || host[len - 1] != '\0') {
733 			error = EINVAL;
734 			goto done_free;
735 		}
736 		if (len > MAXHOSTNAMELEN) {
737 			error = ENAMETOOLONG;
738 			goto done_free;
739 		}
740 	}
741 
742 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
743 	if (error == ENOENT)
744 		domain = NULL;
745 	else if (error != 0)
746 		goto done_free;
747 	else {
748 		ch_flags |= PR_HOST;
749 		pr_flags |= PR_HOST;
750 		if (len == 0 || domain[len - 1] != '\0') {
751 			error = EINVAL;
752 			goto done_free;
753 		}
754 		if (len > MAXHOSTNAMELEN) {
755 			error = ENAMETOOLONG;
756 			goto done_free;
757 		}
758 	}
759 
760 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
761 	if (error == ENOENT)
762 		uuid = NULL;
763 	else if (error != 0)
764 		goto done_free;
765 	else {
766 		ch_flags |= PR_HOST;
767 		pr_flags |= PR_HOST;
768 		if (len == 0 || uuid[len - 1] != '\0') {
769 			error = EINVAL;
770 			goto done_free;
771 		}
772 		if (len > HOSTUUIDLEN) {
773 			error = ENAMETOOLONG;
774 			goto done_free;
775 		}
776 	}
777 
778 #ifdef COMPAT_FREEBSD32
779 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
780 		uint32_t hid32;
781 
782 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
783 		hid = hid32;
784 	} else
785 #endif
786 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
787 	if (error == ENOENT)
788 		gothid = 0;
789 	else if (error != 0)
790 		goto done_free;
791 	else {
792 		gothid = 1;
793 		ch_flags |= PR_HOST;
794 		pr_flags |= PR_HOST;
795 	}
796 
797 #ifdef INET
798 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
799 	if (error == ENOENT)
800 		ip4s = 0;
801 	else if (error != 0)
802 		goto done_free;
803 	else if (ip4s & (sizeof(*ip4) - 1)) {
804 		error = EINVAL;
805 		goto done_free;
806 	} else {
807 		ch_flags |= PR_IP4_USER;
808 		pr_flags |= PR_IP4_USER;
809 		if (ip4s > 0) {
810 			ip4s /= sizeof(*ip4);
811 			if (ip4s > jail_max_af_ips) {
812 				error = EINVAL;
813 				vfs_opterror(opts, "too many IPv4 addresses");
814 				goto done_errmsg;
815 			}
816 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
817 			bcopy(op, ip4, ip4s * sizeof(*ip4));
818 			/*
819 			 * IP addresses are all sorted but ip[0] to preserve
820 			 * the primary IP address as given from userland.
821 			 * This special IP is used for unbound outgoing
822 			 * connections as well for "loopback" traffic in case
823 			 * source address selection cannot find any more fitting
824 			 * address to connect from.
825 			 */
826 			if (ip4s > 1)
827 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
828 				    prison_qcmp_v4);
829 			/*
830 			 * Check for duplicate addresses and do some simple
831 			 * zero and broadcast checks. If users give other bogus
832 			 * addresses it is their problem.
833 			 *
834 			 * We do not have to care about byte order for these
835 			 * checks so we will do them in NBO.
836 			 */
837 			for (ii = 0; ii < ip4s; ii++) {
838 				if (ip4[ii].s_addr == INADDR_ANY ||
839 				    ip4[ii].s_addr == INADDR_BROADCAST) {
840 					error = EINVAL;
841 					goto done_free;
842 				}
843 				if ((ii+1) < ip4s &&
844 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
845 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
846 					error = EINVAL;
847 					goto done_free;
848 				}
849 			}
850 		}
851 	}
852 #endif
853 
854 #ifdef INET6
855 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
856 	if (error == ENOENT)
857 		ip6s = 0;
858 	else if (error != 0)
859 		goto done_free;
860 	else if (ip6s & (sizeof(*ip6) - 1)) {
861 		error = EINVAL;
862 		goto done_free;
863 	} else {
864 		ch_flags |= PR_IP6_USER;
865 		pr_flags |= PR_IP6_USER;
866 		if (ip6s > 0) {
867 			ip6s /= sizeof(*ip6);
868 			if (ip6s > jail_max_af_ips) {
869 				error = EINVAL;
870 				vfs_opterror(opts, "too many IPv6 addresses");
871 				goto done_errmsg;
872 			}
873 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
874 			bcopy(op, ip6, ip6s * sizeof(*ip6));
875 			if (ip6s > 1)
876 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
877 				    prison_qcmp_v6);
878 			for (ii = 0; ii < ip6s; ii++) {
879 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
880 					error = EINVAL;
881 					goto done_free;
882 				}
883 				if ((ii+1) < ip6s &&
884 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
885 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
886 				{
887 					error = EINVAL;
888 					goto done_free;
889 				}
890 			}
891 		}
892 	}
893 #endif
894 
895 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
896 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
897 		error = EINVAL;
898 		vfs_opterror(opts,
899 		    "vnet jails cannot have IP address restrictions");
900 		goto done_errmsg;
901 	}
902 #endif
903 
904 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
905 	if (error == ENOENT)
906 		osrelstr = NULL;
907 	else if (error != 0)
908 		goto done_free;
909 	else {
910 		if (flags & JAIL_UPDATE) {
911 			error = EINVAL;
912 			vfs_opterror(opts,
913 			    "osrelease cannot be changed after creation");
914 			goto done_errmsg;
915 		}
916 		if (len == 0 || osrelstr[len - 1] != '\0') {
917 			error = EINVAL;
918 			goto done_free;
919 		}
920 		if (len >= OSRELEASELEN) {
921 			error = ENAMETOOLONG;
922 			vfs_opterror(opts,
923 			    "osrelease string must be 1-%d bytes long",
924 			    OSRELEASELEN - 1);
925 			goto done_errmsg;
926 		}
927 	}
928 
929 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
930 	if (error == ENOENT)
931 		osreldt = 0;
932 	else if (error != 0)
933 		goto done_free;
934 	else {
935 		if (flags & JAIL_UPDATE) {
936 			error = EINVAL;
937 			vfs_opterror(opts,
938 			    "osreldate cannot be changed after creation");
939 			goto done_errmsg;
940 		}
941 		if (osreldt == 0) {
942 			error = EINVAL;
943 			vfs_opterror(opts, "osreldate cannot be 0");
944 			goto done_errmsg;
945 		}
946 	}
947 
948 	root = NULL;
949 	error = vfs_getopt(opts, "path", (void **)&path, &len);
950 	if (error == ENOENT)
951 		path = NULL;
952 	else if (error != 0)
953 		goto done_free;
954 	else {
955 		if (flags & JAIL_UPDATE) {
956 			error = EINVAL;
957 			vfs_opterror(opts,
958 			    "path cannot be changed after creation");
959 			goto done_errmsg;
960 		}
961 		if (len == 0 || path[len - 1] != '\0') {
962 			error = EINVAL;
963 			goto done_free;
964 		}
965 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
966 		    path, td);
967 		error = namei(&nd);
968 		if (error)
969 			goto done_free;
970 		root = nd.ni_vp;
971 		NDFREE(&nd, NDF_ONLY_PNBUF);
972 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
973 		strlcpy(g_path, path, MAXPATHLEN);
974 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
975 		if (error == 0) {
976 			path = g_path;
977 		} else {
978 			/* exit on other errors */
979 			goto done_free;
980 		}
981 		if (root->v_type != VDIR) {
982 			error = ENOTDIR;
983 			vput(root);
984 			goto done_free;
985 		}
986 		VOP_UNLOCK(root);
987 	}
988 
989 	/*
990 	 * Find the specified jail, or at least its parent.
991 	 * This abuses the file error codes ENOENT and EEXIST.
992 	 */
993 	pr = NULL;
994 	inspr = NULL;
995 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
996 		namelc = strrchr(name, '.');
997 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
998 		if (*p != '\0')
999 			jid = 0;
1000 	}
1001 	sx_xlock(&allprison_lock);
1002 	drflags = PD_LIST_XLOCKED;
1003 	ppr = mypr;
1004 	if (!prison_isalive(ppr)) {
1005 		/* This jail is dying.  This process will surely follow. */
1006 		error = EAGAIN;
1007 		goto done_deref;
1008 	}
1009 	if (jid != 0) {
1010 		if (jid < 0) {
1011 			error = EINVAL;
1012 			vfs_opterror(opts, "negative jid");
1013 			goto done_deref;
1014 		}
1015 		/*
1016 		 * See if a requested jid already exists.  Keep track of
1017 		 * where it can be inserted later.
1018 		 */
1019 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1020 			if (inspr->pr_id < jid)
1021 				continue;
1022 			if (inspr->pr_id > jid)
1023 				break;
1024 			pr = inspr;
1025 			mtx_lock(&pr->pr_mtx);
1026 			drflags |= PD_LOCKED;
1027 			inspr = NULL;
1028 			break;
1029 		}
1030 		if (pr != NULL) {
1031 			/* Create: jid must not exist. */
1032 			if (cuflags == JAIL_CREATE) {
1033 				/*
1034 				 * Even creators that cannot see the jail will
1035 				 * get EEXIST.
1036 				 */
1037 				error = EEXIST;
1038 				vfs_opterror(opts, "jail %d already exists",
1039 				    jid);
1040 				goto done_deref;
1041 			}
1042 			if (!prison_ischild(mypr, pr)) {
1043 				/*
1044 				 * Updaters get ENOENT if they cannot see the
1045 				 * jail.  This is true even for CREATE | UPDATE,
1046 				 * which normally cannot give this error.
1047 				 */
1048 				error = ENOENT;
1049 				vfs_opterror(opts, "jail %d not found", jid);
1050 				goto done_deref;
1051 			}
1052 			ppr = pr->pr_parent;
1053 			if (!prison_isalive(ppr)) {
1054 				error = ENOENT;
1055 				vfs_opterror(opts, "jail %d is dying",
1056 				    ppr->pr_id);
1057 				goto done_deref;
1058 			}
1059 			if (!prison_isalive(pr)) {
1060 				if (!(flags & JAIL_DYING)) {
1061 					error = ENOENT;
1062 					vfs_opterror(opts, "jail %d is dying",
1063 					    jid);
1064 					goto done_deref;
1065 				}
1066 				if ((flags & JAIL_ATTACH) ||
1067 				    (pr_flags & PR_PERSIST)) {
1068 					/*
1069 					 * A dying jail might be resurrected
1070 					 * (via attach or persist), but first
1071 					 * it must determine if another jail
1072 					 * has claimed its name.  Accomplish
1073 					 * this by implicitly re-setting the
1074 					 * name.
1075 					 */
1076 					if (name == NULL)
1077 						name = prison_name(mypr, pr);
1078 				}
1079 			}
1080 		} else {
1081 			/* Update: jid must exist. */
1082 			if (cuflags == JAIL_UPDATE) {
1083 				error = ENOENT;
1084 				vfs_opterror(opts, "jail %d not found", jid);
1085 				goto done_deref;
1086 			}
1087 		}
1088 	}
1089 	/*
1090 	 * If the caller provided a name, look for a jail by that name.
1091 	 * This has different semantics for creates and updates keyed by jid
1092 	 * (where the name must not already exist in a different jail),
1093 	 * and updates keyed by the name itself (where the name must exist
1094 	 * because that is the jail being updated).
1095 	 */
1096 	namelc = NULL;
1097 	if (name != NULL) {
1098 		namelc = strrchr(name, '.');
1099 		if (namelc == NULL)
1100 			namelc = name;
1101 		else {
1102 			/*
1103 			 * This is a hierarchical name.  Split it into the
1104 			 * parent and child names, and make sure the parent
1105 			 * exists or matches an already found jail.
1106 			 */
1107 			if (pr != NULL) {
1108 				if (strncmp(name, ppr->pr_name, namelc - name)
1109 				    || ppr->pr_name[namelc - name] != '\0') {
1110 					error = EINVAL;
1111 					vfs_opterror(opts,
1112 					    "cannot change jail's parent");
1113 					goto done_deref;
1114 				}
1115 			} else {
1116 				*namelc = '\0';
1117 				ppr = prison_find_name(mypr, name);
1118 				if (ppr == NULL) {
1119 					error = ENOENT;
1120 					vfs_opterror(opts,
1121 					    "jail \"%s\" not found", name);
1122 					goto done_deref;
1123 				}
1124 				mtx_unlock(&ppr->pr_mtx);
1125 				if (!prison_isalive(ppr)) {
1126 					error = ENOENT;
1127 					vfs_opterror(opts,
1128 					    "jail \"%s\" is dying", name);
1129 					goto done_deref;
1130 				}
1131 				*namelc = '.';
1132 			}
1133 			namelc++;
1134 		}
1135 		if (namelc[0] != '\0') {
1136 			pnamelen =
1137 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1138 			deadpr = NULL;
1139 			FOREACH_PRISON_CHILD(ppr, tpr) {
1140 				if (tpr != pr &&
1141 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1142 					if (prison_isalive(tpr)) {
1143 						if (pr == NULL &&
1144 						    cuflags != JAIL_CREATE) {
1145 							/*
1146 							 * Use this jail
1147 							 * for updates.
1148 							 */
1149 							pr = tpr;
1150 							mtx_lock(&pr->pr_mtx);
1151 							drflags |= PD_LOCKED;
1152 							break;
1153 						}
1154 						/*
1155 						 * Create, or update(jid):
1156 						 * name must not exist in an
1157 						 * active sibling jail.
1158 						 */
1159 						error = EEXIST;
1160 						vfs_opterror(opts,
1161 						   "jail \"%s\" already exists",
1162 						   name);
1163 						goto done_deref;
1164 					}
1165 					if (pr == NULL &&
1166 					    cuflags != JAIL_CREATE) {
1167 						deadpr = tpr;
1168 					}
1169 				}
1170 			}
1171 			/* If no active jail is found, use a dying one. */
1172 			if (deadpr != NULL && pr == NULL) {
1173 				if (flags & JAIL_DYING) {
1174 					pr = deadpr;
1175 					mtx_lock(&pr->pr_mtx);
1176 					drflags |= PD_LOCKED;
1177 				} else if (cuflags == JAIL_UPDATE) {
1178 					error = ENOENT;
1179 					vfs_opterror(opts,
1180 					    "jail \"%s\" is dying", name);
1181 					goto done_deref;
1182 				}
1183 			}
1184 			/* Update: name must exist if no jid. */
1185 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1186 				error = ENOENT;
1187 				vfs_opterror(opts, "jail \"%s\" not found",
1188 				    name);
1189 				goto done_deref;
1190 			}
1191 		}
1192 	}
1193 	/* Update: must provide a jid or name. */
1194 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1195 		error = ENOENT;
1196 		vfs_opterror(opts, "update specified no jail");
1197 		goto done_deref;
1198 	}
1199 
1200 	/* If there's no prison to update, create a new one and link it in. */
1201 	created = pr == NULL;
1202 	if (created) {
1203 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1204 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1205 				error = EPERM;
1206 				vfs_opterror(opts, "prison limit exceeded");
1207 				goto done_deref;
1208 			}
1209 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1210 			error = EAGAIN;
1211 			vfs_opterror(opts, "no available jail IDs");
1212 			goto done_deref;
1213 		}
1214 
1215 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1216 		pr->pr_state = PRISON_STATE_INVALID;
1217 		refcount_init(&pr->pr_ref, 1);
1218 		refcount_init(&pr->pr_uref, 0);
1219 		drflags |= PD_DEREF;
1220 		LIST_INIT(&pr->pr_children);
1221 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1222 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1223 
1224 		pr->pr_id = jid;
1225 		if (inspr != NULL)
1226 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1227 		else
1228 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1229 
1230 		pr->pr_parent = ppr;
1231 		prison_hold(ppr);
1232 		prison_proc_hold(ppr);
1233 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1234 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1235 			tpr->pr_childcount++;
1236 
1237 		/* Set some default values, and inherit some from the parent. */
1238 		if (namelc == NULL)
1239 			namelc = "";
1240 		if (path == NULL) {
1241 			path = "/";
1242 			root = mypr->pr_root;
1243 			vref(root);
1244 		}
1245 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1246 		pr->pr_flags |= PR_HOST;
1247 #if defined(INET) || defined(INET6)
1248 #ifdef VIMAGE
1249 		if (!(pr_flags & PR_VNET))
1250 #endif
1251 		{
1252 #ifdef INET
1253 			if (!(ch_flags & PR_IP4_USER))
1254 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1255 			else if (!(pr_flags & PR_IP4_USER)) {
1256 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1257 				if (ppr->pr_ip4 != NULL) {
1258 					pr->pr_ip4s = ppr->pr_ip4s;
1259 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1260 					    sizeof(struct in_addr), M_PRISON,
1261 					    M_WAITOK);
1262 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1263 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1264 				}
1265 			}
1266 #endif
1267 #ifdef INET6
1268 			if (!(ch_flags & PR_IP6_USER))
1269 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1270 			else if (!(pr_flags & PR_IP6_USER)) {
1271 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1272 				if (ppr->pr_ip6 != NULL) {
1273 					pr->pr_ip6s = ppr->pr_ip6s;
1274 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1275 					    sizeof(struct in6_addr), M_PRISON,
1276 					    M_WAITOK);
1277 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1278 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1279 				}
1280 			}
1281 #endif
1282 		}
1283 #endif
1284 		/* Source address selection is always on by default. */
1285 		pr->pr_flags |= _PR_IP_SADDRSEL;
1286 
1287 		pr->pr_securelevel = ppr->pr_securelevel;
1288 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1289 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1290 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1291 
1292 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1293 		if (osrelstr == NULL)
1294 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1295 			    sizeof(pr->pr_osrelease));
1296 		else
1297 			strlcpy(pr->pr_osrelease, osrelstr,
1298 			    sizeof(pr->pr_osrelease));
1299 
1300 #ifdef VIMAGE
1301 		/* Allocate a new vnet if specified. */
1302 		pr->pr_vnet = (pr_flags & PR_VNET)
1303 		    ? vnet_alloc() : ppr->pr_vnet;
1304 #endif
1305 		/*
1306 		 * Allocate a dedicated cpuset for each jail.
1307 		 * Unlike other initial settings, this may return an error.
1308 		 */
1309 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1310 		if (error)
1311 			goto done_deref;
1312 
1313 		mtx_lock(&pr->pr_mtx);
1314 		drflags |= PD_LOCKED;
1315 	} else {
1316 		/*
1317 		 * Grab a reference for existing prisons, to ensure they
1318 		 * continue to exist for the duration of the call.
1319 		 */
1320 		prison_hold(pr);
1321 		drflags |= PD_DEREF;
1322 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1323 		if ((pr->pr_flags & PR_VNET) &&
1324 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1325 			error = EINVAL;
1326 			vfs_opterror(opts,
1327 			    "vnet jails cannot have IP address restrictions");
1328 			goto done_deref;
1329 		}
1330 #endif
1331 #ifdef INET
1332 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1333 			error = EINVAL;
1334 			vfs_opterror(opts,
1335 			    "ip4 cannot be changed after creation");
1336 			goto done_deref;
1337 		}
1338 #endif
1339 #ifdef INET6
1340 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1341 			error = EINVAL;
1342 			vfs_opterror(opts,
1343 			    "ip6 cannot be changed after creation");
1344 			goto done_deref;
1345 		}
1346 #endif
1347 	}
1348 
1349 	/* Do final error checking before setting anything. */
1350 	if (gotslevel) {
1351 		if (slevel < ppr->pr_securelevel) {
1352 			error = EPERM;
1353 			goto done_deref;
1354 		}
1355 	}
1356 	if (gotchildmax) {
1357 		if (childmax >= ppr->pr_childmax) {
1358 			error = EPERM;
1359 			goto done_deref;
1360 		}
1361 	}
1362 	if (gotenforce) {
1363 		if (enforce < ppr->pr_enforce_statfs) {
1364 			error = EPERM;
1365 			goto done_deref;
1366 		}
1367 	}
1368 	if (gotrsnum) {
1369 		/*
1370 		 * devfs_rsnum is a uint16_t
1371 		 */
1372 		if (rsnum < 0 || rsnum > 65535) {
1373 			error = EINVAL;
1374 			goto done_deref;
1375 		}
1376 		/*
1377 		 * Nested jails always inherit parent's devfs ruleset
1378 		 */
1379 		if (jailed(td->td_ucred)) {
1380 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1381 				error = EPERM;
1382 				goto done_deref;
1383 			} else
1384 				rsnum = ppr->pr_devfs_rsnum;
1385 		}
1386 	}
1387 #ifdef INET
1388 	if (ip4s > 0) {
1389 		if (ppr->pr_flags & PR_IP4) {
1390 			/*
1391 			 * Make sure the new set of IP addresses is a
1392 			 * subset of the parent's list.  Don't worry
1393 			 * about the parent being unlocked, as any
1394 			 * setting is done with allprison_lock held.
1395 			 */
1396 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1397 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1398 					break;
1399 			if (ij == ppr->pr_ip4s) {
1400 				error = EPERM;
1401 				goto done_deref;
1402 			}
1403 			if (ip4s > 1) {
1404 				for (ii = ij = 1; ii < ip4s; ii++) {
1405 					if (ip4[ii].s_addr ==
1406 					    ppr->pr_ip4[0].s_addr)
1407 						continue;
1408 					for (; ij < ppr->pr_ip4s; ij++)
1409 						if (ip4[ii].s_addr ==
1410 						    ppr->pr_ip4[ij].s_addr)
1411 							break;
1412 					if (ij == ppr->pr_ip4s)
1413 						break;
1414 				}
1415 				if (ij == ppr->pr_ip4s) {
1416 					error = EPERM;
1417 					goto done_deref;
1418 				}
1419 			}
1420 		}
1421 		/*
1422 		 * Check for conflicting IP addresses.  We permit them
1423 		 * if there is no more than one IP on each jail.  If
1424 		 * there is a duplicate on a jail with more than one
1425 		 * IP stop checking and return error.
1426 		 */
1427 #ifdef VIMAGE
1428 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1429 			if (tppr->pr_flags & PR_VNET)
1430 				break;
1431 #else
1432 		tppr = &prison0;
1433 #endif
1434 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1435 			if (tpr == pr ||
1436 #ifdef VIMAGE
1437 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1438 #endif
1439 			    !prison_isalive(tpr)) {
1440 				descend = 0;
1441 				continue;
1442 			}
1443 			if (!(tpr->pr_flags & PR_IP4_USER))
1444 				continue;
1445 			descend = 0;
1446 			if (tpr->pr_ip4 == NULL ||
1447 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1448 				continue;
1449 			for (ii = 0; ii < ip4s; ii++) {
1450 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1451 				    0) {
1452 					error = EADDRINUSE;
1453 					vfs_opterror(opts,
1454 					    "IPv4 addresses clash");
1455 					goto done_deref;
1456 				}
1457 			}
1458 		}
1459 	}
1460 #endif
1461 #ifdef INET6
1462 	if (ip6s > 0) {
1463 		if (ppr->pr_flags & PR_IP6) {
1464 			/*
1465 			 * Make sure the new set of IP addresses is a
1466 			 * subset of the parent's list.
1467 			 */
1468 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1469 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1470 				    &ppr->pr_ip6[ij]))
1471 					break;
1472 			if (ij == ppr->pr_ip6s) {
1473 				error = EPERM;
1474 				goto done_deref;
1475 			}
1476 			if (ip6s > 1) {
1477 				for (ii = ij = 1; ii < ip6s; ii++) {
1478 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1479 					     &ppr->pr_ip6[0]))
1480 						continue;
1481 					for (; ij < ppr->pr_ip6s; ij++)
1482 						if (IN6_ARE_ADDR_EQUAL(
1483 						    &ip6[ii], &ppr->pr_ip6[ij]))
1484 							break;
1485 					if (ij == ppr->pr_ip6s)
1486 						break;
1487 				}
1488 				if (ij == ppr->pr_ip6s) {
1489 					error = EPERM;
1490 					goto done_deref;
1491 				}
1492 			}
1493 		}
1494 		/* Check for conflicting IP addresses. */
1495 #ifdef VIMAGE
1496 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1497 			if (tppr->pr_flags & PR_VNET)
1498 				break;
1499 #else
1500 		tppr = &prison0;
1501 #endif
1502 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1503 			if (tpr == pr ||
1504 #ifdef VIMAGE
1505 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1506 #endif
1507 			    !prison_isalive(tpr)) {
1508 				descend = 0;
1509 				continue;
1510 			}
1511 			if (!(tpr->pr_flags & PR_IP6_USER))
1512 				continue;
1513 			descend = 0;
1514 			if (tpr->pr_ip6 == NULL ||
1515 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1516 				continue;
1517 			for (ii = 0; ii < ip6s; ii++) {
1518 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1519 				    0) {
1520 					error = EADDRINUSE;
1521 					vfs_opterror(opts,
1522 					    "IPv6 addresses clash");
1523 					goto done_deref;
1524 				}
1525 			}
1526 		}
1527 	}
1528 #endif
1529 	onamelen = namelen = 0;
1530 	if (namelc != NULL) {
1531 		/* Give a default name of the jid.  Also allow the name to be
1532 		 * explicitly the jid - but not any other number, and only in
1533 		 * normal form (no leading zero/etc).
1534 		 */
1535 		if (namelc[0] == '\0')
1536 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1537 		else if ((strtoul(namelc, &p, 10) != jid ||
1538 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1539 			error = EINVAL;
1540 			vfs_opterror(opts,
1541 			    "name cannot be numeric (unless it is the jid)");
1542 			goto done_deref;
1543 		}
1544 		/*
1545 		 * Make sure the name isn't too long for the prison or its
1546 		 * children.
1547 		 */
1548 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1549 		onamelen = strlen(pr->pr_name + pnamelen);
1550 		namelen = strlen(namelc);
1551 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1552 			error = ENAMETOOLONG;
1553 			goto done_deref;
1554 		}
1555 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1556 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1557 			    sizeof(pr->pr_name)) {
1558 				error = ENAMETOOLONG;
1559 				goto done_deref;
1560 			}
1561 		}
1562 	}
1563 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1564 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1565 		error = EPERM;
1566 		goto done_deref;
1567 	}
1568 
1569 	/*
1570 	 * Let modules check their parameters.  This requires unlocking and
1571 	 * then re-locking the prison, but this is still a valid state as long
1572 	 * as allprison_lock remains xlocked.
1573 	 */
1574 	mtx_unlock(&pr->pr_mtx);
1575 	drflags &= ~PD_LOCKED;
1576 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1577 	if (error != 0)
1578 		goto done_deref;
1579 	mtx_lock(&pr->pr_mtx);
1580 	drflags |= PD_LOCKED;
1581 
1582 	/* At this point, all valid parameters should have been noted. */
1583 	TAILQ_FOREACH(opt, opts, link) {
1584 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1585 			error = EINVAL;
1586 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1587 			goto done_deref;
1588 		}
1589 	}
1590 
1591 	/* Set the parameters of the prison. */
1592 #ifdef INET
1593 	redo_ip4 = 0;
1594 	if (pr_flags & PR_IP4_USER) {
1595 		pr->pr_flags |= PR_IP4;
1596 		free(pr->pr_ip4, M_PRISON);
1597 		pr->pr_ip4s = ip4s;
1598 		pr->pr_ip4 = ip4;
1599 		ip4 = NULL;
1600 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1601 #ifdef VIMAGE
1602 			if (tpr->pr_flags & PR_VNET) {
1603 				descend = 0;
1604 				continue;
1605 			}
1606 #endif
1607 			if (prison_restrict_ip4(tpr, NULL)) {
1608 				redo_ip4 = 1;
1609 				descend = 0;
1610 			}
1611 		}
1612 	}
1613 #endif
1614 #ifdef INET6
1615 	redo_ip6 = 0;
1616 	if (pr_flags & PR_IP6_USER) {
1617 		pr->pr_flags |= PR_IP6;
1618 		free(pr->pr_ip6, M_PRISON);
1619 		pr->pr_ip6s = ip6s;
1620 		pr->pr_ip6 = ip6;
1621 		ip6 = NULL;
1622 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1623 #ifdef VIMAGE
1624 			if (tpr->pr_flags & PR_VNET) {
1625 				descend = 0;
1626 				continue;
1627 			}
1628 #endif
1629 			if (prison_restrict_ip6(tpr, NULL)) {
1630 				redo_ip6 = 1;
1631 				descend = 0;
1632 			}
1633 		}
1634 	}
1635 #endif
1636 	if (gotslevel) {
1637 		pr->pr_securelevel = slevel;
1638 		/* Set all child jails to be at least this level. */
1639 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1640 			if (tpr->pr_securelevel < slevel)
1641 				tpr->pr_securelevel = slevel;
1642 	}
1643 	if (gotchildmax) {
1644 		pr->pr_childmax = childmax;
1645 		/* Set all child jails to under this limit. */
1646 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1647 			if (tpr->pr_childmax > childmax - level)
1648 				tpr->pr_childmax = childmax > level
1649 				    ? childmax - level : 0;
1650 	}
1651 	if (gotenforce) {
1652 		pr->pr_enforce_statfs = enforce;
1653 		/* Pass this restriction on to the children. */
1654 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1655 			if (tpr->pr_enforce_statfs < enforce)
1656 				tpr->pr_enforce_statfs = enforce;
1657 	}
1658 	if (gotrsnum) {
1659 		pr->pr_devfs_rsnum = rsnum;
1660 		/* Pass this restriction on to the children. */
1661 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1662 			tpr->pr_devfs_rsnum = rsnum;
1663 	}
1664 	if (namelc != NULL) {
1665 		if (ppr == &prison0)
1666 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1667 		else
1668 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1669 			    ppr->pr_name, namelc);
1670 		/* Change this component of child names. */
1671 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1672 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1673 			    strlen(tpr->pr_name + onamelen) + 1);
1674 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1675 		}
1676 	}
1677 	if (path != NULL) {
1678 		/* Try to keep a real-rooted full pathname. */
1679 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1680 		pr->pr_root = root;
1681 		root = NULL;
1682 	}
1683 	if (PR_HOST & ch_flags & ~pr_flags) {
1684 		if (pr->pr_flags & PR_HOST) {
1685 			/*
1686 			 * Copy the parent's host info.  As with pr_ip4 above,
1687 			 * the lack of a lock on the parent is not a problem;
1688 			 * it is always set with allprison_lock at least
1689 			 * shared, and is held exclusively here.
1690 			 */
1691 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1692 			    sizeof(pr->pr_hostname));
1693 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1694 			    sizeof(pr->pr_domainname));
1695 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1696 			    sizeof(pr->pr_hostuuid));
1697 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1698 		}
1699 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1700 		/* Set this prison, and any descendants without PR_HOST. */
1701 		if (host != NULL)
1702 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1703 		if (domain != NULL)
1704 			strlcpy(pr->pr_domainname, domain,
1705 			    sizeof(pr->pr_domainname));
1706 		if (uuid != NULL)
1707 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1708 		if (gothid)
1709 			pr->pr_hostid = hid;
1710 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1711 			if (tpr->pr_flags & PR_HOST)
1712 				descend = 0;
1713 			else {
1714 				if (host != NULL)
1715 					strlcpy(tpr->pr_hostname,
1716 					    pr->pr_hostname,
1717 					    sizeof(tpr->pr_hostname));
1718 				if (domain != NULL)
1719 					strlcpy(tpr->pr_domainname,
1720 					    pr->pr_domainname,
1721 					    sizeof(tpr->pr_domainname));
1722 				if (uuid != NULL)
1723 					strlcpy(tpr->pr_hostuuid,
1724 					    pr->pr_hostuuid,
1725 					    sizeof(tpr->pr_hostuuid));
1726 				if (gothid)
1727 					tpr->pr_hostid = hid;
1728 			}
1729 		}
1730 	}
1731 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1732 	if ((tallow = ch_allow & ~pr_allow))
1733 		prison_set_allow_locked(pr, tallow, 0);
1734 	/*
1735 	 * Persistent prisons get an extra reference, and prisons losing their
1736 	 * persist flag lose that reference.
1737 	 */
1738 	born = !prison_isalive(pr);
1739 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
1740 		if (pr_flags & PR_PERSIST) {
1741 			prison_hold(pr);
1742 			/*
1743 			 * This may make a dead prison alive again, but wait
1744 			 * to label it as such until after OSD calls have had
1745 			 * a chance to run (and perhaps to fail).
1746 			 */
1747 			refcount_acquire(&pr->pr_uref);
1748 		} else {
1749 			drflags |= PD_DEUREF;
1750 			prison_free_not_last(pr);
1751 		}
1752 	}
1753 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1754 	mtx_unlock(&pr->pr_mtx);
1755 	drflags &= ~PD_LOCKED;
1756 	/*
1757 	 * Any errors past this point will need to de-persist newly created
1758 	 * prisons, as well as call remove methods.
1759 	 */
1760 	if (born)
1761 		drflags |= PD_KILL;
1762 
1763 #ifdef RACCT
1764 	if (racct_enable && created)
1765 		prison_racct_attach(pr);
1766 #endif
1767 
1768 	/* Locks may have prevented a complete restriction of child IP
1769 	 * addresses.  If so, allocate some more memory and try again.
1770 	 */
1771 #ifdef INET
1772 	while (redo_ip4) {
1773 		ip4s = pr->pr_ip4s;
1774 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1775 		mtx_lock(&pr->pr_mtx);
1776 		redo_ip4 = 0;
1777 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1778 #ifdef VIMAGE
1779 			if (tpr->pr_flags & PR_VNET) {
1780 				descend = 0;
1781 				continue;
1782 			}
1783 #endif
1784 			if (prison_restrict_ip4(tpr, ip4)) {
1785 				if (ip4 != NULL)
1786 					ip4 = NULL;
1787 				else
1788 					redo_ip4 = 1;
1789 			}
1790 		}
1791 		mtx_unlock(&pr->pr_mtx);
1792 	}
1793 #endif
1794 #ifdef INET6
1795 	while (redo_ip6) {
1796 		ip6s = pr->pr_ip6s;
1797 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1798 		mtx_lock(&pr->pr_mtx);
1799 		redo_ip6 = 0;
1800 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1801 #ifdef VIMAGE
1802 			if (tpr->pr_flags & PR_VNET) {
1803 				descend = 0;
1804 				continue;
1805 			}
1806 #endif
1807 			if (prison_restrict_ip6(tpr, ip6)) {
1808 				if (ip6 != NULL)
1809 					ip6 = NULL;
1810 				else
1811 					redo_ip6 = 1;
1812 			}
1813 		}
1814 		mtx_unlock(&pr->pr_mtx);
1815 	}
1816 #endif
1817 
1818 	/* Let the modules do their work. */
1819 	if (born) {
1820 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1821 		if (error)
1822 			goto done_deref;
1823 	}
1824 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1825 	if (error)
1826 		goto done_deref;
1827 
1828 	/*
1829 	 * A new prison is now ready to be seen; either it has gained a user
1830 	 * reference via persistence, or is about to gain one via attachment.
1831 	 */
1832 	if (born) {
1833 		drflags = prison_lock_xlock(pr, drflags);
1834 		pr->pr_state = PRISON_STATE_ALIVE;
1835 	}
1836 
1837 	/* Attach this process to the prison if requested. */
1838 	if (flags & JAIL_ATTACH) {
1839 		error = do_jail_attach(td, pr,
1840 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
1841 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
1842 		if (error) {
1843 			vfs_opterror(opts, "attach failed");
1844 			goto done_deref;
1845 		}
1846 	}
1847 
1848 #ifdef RACCT
1849 	if (racct_enable && !created) {
1850 		if (drflags & PD_LOCKED) {
1851 			mtx_unlock(&pr->pr_mtx);
1852 			drflags &= ~PD_LOCKED;
1853 		}
1854 		if (drflags & PD_LIST_XLOCKED) {
1855 			sx_xunlock(&allprison_lock);
1856 			drflags &= ~PD_LIST_XLOCKED;
1857 		}
1858 		prison_racct_modify(pr);
1859 	}
1860 #endif
1861 
1862 	drflags &= ~PD_KILL;
1863 	td->td_retval[0] = pr->pr_id;
1864 
1865  done_deref:
1866 	/* Release any temporary prison holds and/or locks. */
1867 	if (pr != NULL)
1868 		prison_deref(pr, drflags);
1869 	else if (drflags & PD_LIST_SLOCKED)
1870 		sx_sunlock(&allprison_lock);
1871 	else if (drflags & PD_LIST_XLOCKED)
1872 		sx_xunlock(&allprison_lock);
1873 	if (root != NULL)
1874 		vrele(root);
1875  done_errmsg:
1876 	if (error) {
1877 		/* Write the error message back to userspace. */
1878 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1879 		    &errmsg_len) == 0 && errmsg_len > 0) {
1880 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1881 			if (optuio->uio_segflg == UIO_SYSSPACE)
1882 				bcopy(errmsg,
1883 				    optuio->uio_iov[errmsg_pos].iov_base,
1884 				    errmsg_len);
1885 			else
1886 				copyout(errmsg,
1887 				    optuio->uio_iov[errmsg_pos].iov_base,
1888 				    errmsg_len);
1889 		}
1890 	}
1891  done_free:
1892 #ifdef INET
1893 	free(ip4, M_PRISON);
1894 #endif
1895 #ifdef INET6
1896 	free(ip6, M_PRISON);
1897 #endif
1898 	if (g_path != NULL)
1899 		free(g_path, M_TEMP);
1900 	vfs_freeopts(opts);
1901 	return (error);
1902 }
1903 
1904 /*
1905  * Find the next available prison ID.  Return the ID on success, or zero
1906  * on failure.  Also set a pointer to the allprison list entry the prison
1907  * should be inserted before.
1908  */
1909 static int
1910 get_next_prid(struct prison **insprp)
1911 {
1912 	struct prison *inspr;
1913 	int jid, maxid;
1914 
1915 	jid = lastprid % JAIL_MAX + 1;
1916 	if (TAILQ_EMPTY(&allprison) ||
1917 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
1918 		/*
1919 		 * A common case is for all jails to be implicitly numbered,
1920 		 * which means they'll go on the end of the list, at least
1921 		 * for the first JAIL_MAX times.
1922 		 */
1923 		inspr = NULL;
1924 	} else {
1925 		/*
1926 		 * Take two passes through the allprison list: first starting
1927 		 * with the proposed jid, then ending with it.
1928 		 */
1929 		for (maxid = JAIL_MAX; maxid != 0; ) {
1930 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
1931 				if (inspr->pr_id < jid)
1932 					continue;
1933 				if (inspr->pr_id > jid) {
1934 					/* Found an opening. */
1935 					maxid = 0;
1936 					break;
1937 				}
1938 				if (++jid > maxid) {
1939 					if (lastprid == maxid || lastprid == 0)
1940 					{
1941 						/*
1942 						 * The entire legal range
1943 						 * has been traversed
1944 						 */
1945 						return 0;
1946 					}
1947 					/* Try again from the start. */
1948 					jid = 1;
1949 					maxid = lastprid;
1950 					break;
1951 				}
1952 			}
1953 			if (inspr == NULL) {
1954 				/* Found room at the end of the list. */
1955 				break;
1956 			}
1957 		}
1958 	}
1959 	*insprp = inspr;
1960 	lastprid = jid;
1961 	return (jid);
1962 }
1963 
1964 /*
1965  * struct jail_get_args {
1966  *	struct iovec *iovp;
1967  *	unsigned int iovcnt;
1968  *	int flags;
1969  * };
1970  */
1971 int
1972 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1973 {
1974 	struct uio *auio;
1975 	int error;
1976 
1977 	/* Check that we have an even number of iovecs. */
1978 	if (uap->iovcnt & 1)
1979 		return (EINVAL);
1980 
1981 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1982 	if (error)
1983 		return (error);
1984 	error = kern_jail_get(td, auio, uap->flags);
1985 	if (error == 0)
1986 		error = copyout(auio->uio_iov, uap->iovp,
1987 		    uap->iovcnt * sizeof (struct iovec));
1988 	free(auio, M_IOV);
1989 	return (error);
1990 }
1991 
1992 int
1993 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1994 {
1995 	struct bool_flags *bf;
1996 	struct jailsys_flags *jsf;
1997 	struct prison *pr, *mypr;
1998 	struct vfsopt *opt;
1999 	struct vfsoptlist *opts;
2000 	char *errmsg, *name;
2001 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2002 	unsigned f;
2003 
2004 	if (flags & ~JAIL_GET_MASK)
2005 		return (EINVAL);
2006 
2007 	/* Get the parameter list. */
2008 	error = vfs_buildopts(optuio, &opts);
2009 	if (error)
2010 		return (error);
2011 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2012 	mypr = td->td_ucred->cr_prison;
2013 	pr = NULL;
2014 
2015 	/*
2016 	 * Find the prison specified by one of: lastjid, jid, name.
2017 	 */
2018 	sx_slock(&allprison_lock);
2019 	drflags = PD_LIST_SLOCKED;
2020 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2021 	if (error == 0) {
2022 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2023 			if (pr->pr_id > jid &&
2024 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2025 			    prison_ischild(mypr, pr)) {
2026 				mtx_lock(&pr->pr_mtx);
2027 				drflags |= PD_LOCKED;
2028 				goto found_prison;
2029 			}
2030 		}
2031 		error = ENOENT;
2032 		vfs_opterror(opts, "no jail after %d", jid);
2033 		goto done;
2034 	} else if (error != ENOENT)
2035 		goto done;
2036 
2037 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2038 	if (error == 0) {
2039 		if (jid != 0) {
2040 			pr = prison_find_child(mypr, jid);
2041 			if (pr != NULL) {
2042 				drflags |= PD_LOCKED;
2043 				if (!(prison_isalive(pr) ||
2044 				    (flags & JAIL_DYING))) {
2045 					error = ENOENT;
2046 					vfs_opterror(opts, "jail %d is dying",
2047 					    jid);
2048 					goto done;
2049 				}
2050 				goto found_prison;
2051 			}
2052 			error = ENOENT;
2053 			vfs_opterror(opts, "jail %d not found", jid);
2054 			goto done;
2055 		}
2056 	} else if (error != ENOENT)
2057 		goto done;
2058 
2059 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2060 	if (error == 0) {
2061 		if (len == 0 || name[len - 1] != '\0') {
2062 			error = EINVAL;
2063 			goto done;
2064 		}
2065 		pr = prison_find_name(mypr, name);
2066 		if (pr != NULL) {
2067 			drflags |= PD_LOCKED;
2068 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2069 				error = ENOENT;
2070 				vfs_opterror(opts, "jail \"%s\" is dying",
2071 				    name);
2072 				goto done;
2073 			}
2074 			goto found_prison;
2075 		}
2076 		error = ENOENT;
2077 		vfs_opterror(opts, "jail \"%s\" not found", name);
2078 		goto done;
2079 	} else if (error != ENOENT)
2080 		goto done;
2081 
2082 	vfs_opterror(opts, "no jail specified");
2083 	error = ENOENT;
2084 	goto done;
2085 
2086  found_prison:
2087 	/* Get the parameters of the prison. */
2088 	prison_hold(pr);
2089 	drflags |= PD_DEREF;
2090 	td->td_retval[0] = pr->pr_id;
2091 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2092 	if (error != 0 && error != ENOENT)
2093 		goto done;
2094 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2095 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2096 	if (error != 0 && error != ENOENT)
2097 		goto done;
2098 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2099 	if (error != 0 && error != ENOENT)
2100 		goto done;
2101 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2102 	    sizeof(pr->pr_cpuset->cs_id));
2103 	if (error != 0 && error != ENOENT)
2104 		goto done;
2105 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2106 	if (error != 0 && error != ENOENT)
2107 		goto done;
2108 #ifdef INET
2109 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2110 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2111 	if (error != 0 && error != ENOENT)
2112 		goto done;
2113 #endif
2114 #ifdef INET6
2115 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2116 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2117 	if (error != 0 && error != ENOENT)
2118 		goto done;
2119 #endif
2120 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2121 	    sizeof(pr->pr_securelevel));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done;
2124 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2125 	    sizeof(pr->pr_childcount));
2126 	if (error != 0 && error != ENOENT)
2127 		goto done;
2128 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2129 	    sizeof(pr->pr_childmax));
2130 	if (error != 0 && error != ENOENT)
2131 		goto done;
2132 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2133 	if (error != 0 && error != ENOENT)
2134 		goto done;
2135 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2136 	if (error != 0 && error != ENOENT)
2137 		goto done;
2138 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2139 	if (error != 0 && error != ENOENT)
2140 		goto done;
2141 #ifdef COMPAT_FREEBSD32
2142 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2143 		uint32_t hid32 = pr->pr_hostid;
2144 
2145 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2146 	} else
2147 #endif
2148 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2149 	    sizeof(pr->pr_hostid));
2150 	if (error != 0 && error != ENOENT)
2151 		goto done;
2152 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2153 	    sizeof(pr->pr_enforce_statfs));
2154 	if (error != 0 && error != ENOENT)
2155 		goto done;
2156 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2157 	    sizeof(pr->pr_devfs_rsnum));
2158 	if (error != 0 && error != ENOENT)
2159 		goto done;
2160 	for (bf = pr_flag_bool;
2161 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2162 	     bf++) {
2163 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2164 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2165 		if (error != 0 && error != ENOENT)
2166 			goto done;
2167 		i = !i;
2168 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2169 		if (error != 0 && error != ENOENT)
2170 			goto done;
2171 	}
2172 	for (jsf = pr_flag_jailsys;
2173 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2174 	     jsf++) {
2175 		f = pr->pr_flags & (jsf->disable | jsf->new);
2176 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2177 		    : (f == jsf->new) ? JAIL_SYS_NEW
2178 		    : JAIL_SYS_INHERIT;
2179 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2180 		if (error != 0 && error != ENOENT)
2181 			goto done;
2182 	}
2183 	for (bf = pr_flag_allow;
2184 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2185 		atomic_load_int(&bf->flag) != 0;
2186 	     bf++) {
2187 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2188 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2189 		if (error != 0 && error != ENOENT)
2190 			goto done;
2191 		i = !i;
2192 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2193 		if (error != 0 && error != ENOENT)
2194 			goto done;
2195 	}
2196 	i = !prison_isalive(pr);
2197 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2198 	if (error != 0 && error != ENOENT)
2199 		goto done;
2200 	i = !i;
2201 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2202 	if (error != 0 && error != ENOENT)
2203 		goto done;
2204 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2205 	    sizeof(pr->pr_osreldate));
2206 	if (error != 0 && error != ENOENT)
2207 		goto done;
2208 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2209 	if (error != 0 && error != ENOENT)
2210 		goto done;
2211 
2212 	/* Get the module parameters. */
2213 	mtx_unlock(&pr->pr_mtx);
2214 	drflags &= ~PD_LOCKED;
2215 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2216 	if (error)
2217 		goto done;
2218 	prison_deref(pr, drflags);
2219 	pr = NULL;
2220 	drflags = 0;
2221 
2222 	/* By now, all parameters should have been noted. */
2223 	TAILQ_FOREACH(opt, opts, link) {
2224 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2225 			error = EINVAL;
2226 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2227 			goto done;
2228 		}
2229 	}
2230 
2231 	/* Write the fetched parameters back to userspace. */
2232 	error = 0;
2233 	TAILQ_FOREACH(opt, opts, link) {
2234 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2235 			pos = 2 * opt->pos + 1;
2236 			optuio->uio_iov[pos].iov_len = opt->len;
2237 			if (opt->value != NULL) {
2238 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2239 					bcopy(opt->value,
2240 					    optuio->uio_iov[pos].iov_base,
2241 					    opt->len);
2242 				} else {
2243 					error = copyout(opt->value,
2244 					    optuio->uio_iov[pos].iov_base,
2245 					    opt->len);
2246 					if (error)
2247 						break;
2248 				}
2249 			}
2250 		}
2251 	}
2252 
2253  done:
2254 	/* Release any temporary prison holds and/or locks. */
2255 	if (pr != NULL)
2256 		prison_deref(pr, drflags);
2257 	else if (drflags & PD_LIST_SLOCKED)
2258 		sx_sunlock(&allprison_lock);
2259 	if (error && errmsg_pos >= 0) {
2260 		/* Write the error message back to userspace. */
2261 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2262 		errmsg_pos = 2 * errmsg_pos + 1;
2263 		if (errmsg_len > 0) {
2264 			if (optuio->uio_segflg == UIO_SYSSPACE)
2265 				bcopy(errmsg,
2266 				    optuio->uio_iov[errmsg_pos].iov_base,
2267 				    errmsg_len);
2268 			else
2269 				copyout(errmsg,
2270 				    optuio->uio_iov[errmsg_pos].iov_base,
2271 				    errmsg_len);
2272 		}
2273 	}
2274 	vfs_freeopts(opts);
2275 	return (error);
2276 }
2277 
2278 /*
2279  * struct jail_remove_args {
2280  *	int jid;
2281  * };
2282  */
2283 int
2284 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2285 {
2286 	struct prison *pr;
2287 	int error;
2288 
2289 	error = priv_check(td, PRIV_JAIL_REMOVE);
2290 	if (error)
2291 		return (error);
2292 
2293 	sx_xlock(&allprison_lock);
2294 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2295 	if (pr == NULL) {
2296 		sx_xunlock(&allprison_lock);
2297 		return (EINVAL);
2298 	}
2299 	if (!prison_isalive(pr)) {
2300 		/* Silently ignore already-dying prisons. */
2301 		mtx_unlock(&pr->pr_mtx);
2302 		sx_xunlock(&allprison_lock);
2303 		return (0);
2304 	}
2305 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2306 	return (0);
2307 }
2308 
2309 /*
2310  * struct jail_attach_args {
2311  *	int jid;
2312  * };
2313  */
2314 int
2315 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2316 {
2317 	struct prison *pr;
2318 	int error;
2319 
2320 	error = priv_check(td, PRIV_JAIL_ATTACH);
2321 	if (error)
2322 		return (error);
2323 
2324 	sx_slock(&allprison_lock);
2325 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2326 	if (pr == NULL) {
2327 		sx_sunlock(&allprison_lock);
2328 		return (EINVAL);
2329 	}
2330 
2331 	/* Do not allow a process to attach to a prison that is not alive. */
2332 	if (!prison_isalive(pr)) {
2333 		mtx_unlock(&pr->pr_mtx);
2334 		sx_sunlock(&allprison_lock);
2335 		return (EINVAL);
2336 	}
2337 
2338 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2339 }
2340 
2341 static int
2342 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2343 {
2344 	struct proc *p;
2345 	struct ucred *newcred, *oldcred;
2346 	int error;
2347 
2348 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2349 	sx_assert(&allprison_lock, SX_LOCKED);
2350 	drflags &= PD_LOCK_FLAGS;
2351 	/*
2352 	 * XXX: Note that there is a slight race here if two threads
2353 	 * in the same privileged process attempt to attach to two
2354 	 * different jails at the same time.  It is important for
2355 	 * user processes not to do this, or they might end up with
2356 	 * a process root from one prison, but attached to the jail
2357 	 * of another.
2358 	 */
2359 	prison_hold(pr);
2360 	refcount_acquire(&pr->pr_uref);
2361 	drflags |= PD_DEREF | PD_DEUREF;
2362 	mtx_unlock(&pr->pr_mtx);
2363 	drflags &= ~PD_LOCKED;
2364 
2365 	/* Let modules do whatever they need to prepare for attaching. */
2366 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2367 	if (error) {
2368 		prison_deref(pr, drflags);
2369 		return (error);
2370 	}
2371 	sx_unlock(&allprison_lock);
2372 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2373 
2374 	/*
2375 	 * Reparent the newly attached process to this jail.
2376 	 */
2377 	p = td->td_proc;
2378 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2379 	if (error)
2380 		goto e_revert_osd;
2381 
2382 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2383 	if ((error = change_dir(pr->pr_root, td)) != 0)
2384 		goto e_unlock;
2385 #ifdef MAC
2386 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2387 		goto e_unlock;
2388 #endif
2389 	VOP_UNLOCK(pr->pr_root);
2390 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2391 		goto e_revert_osd;
2392 
2393 	newcred = crget();
2394 	PROC_LOCK(p);
2395 	oldcred = crcopysafe(p, newcred);
2396 	newcred->cr_prison = pr;
2397 	proc_set_cred(p, newcred);
2398 	setsugid(p);
2399 #ifdef RACCT
2400 	racct_proc_ucred_changed(p, oldcred, newcred);
2401 	crhold(newcred);
2402 #endif
2403 	PROC_UNLOCK(p);
2404 #ifdef RCTL
2405 	rctl_proc_ucred_changed(p, newcred);
2406 	crfree(newcred);
2407 #endif
2408 	prison_deref(oldcred->cr_prison, drflags);
2409 	crfree(oldcred);
2410 
2411 	/*
2412 	 * If the prison was killed while changing credentials, die along
2413 	 * with it.
2414 	 */
2415 	if (!prison_isalive(pr)) {
2416 		PROC_LOCK(p);
2417 		kern_psignal(p, SIGKILL);
2418 		PROC_UNLOCK(p);
2419 	}
2420 
2421 	return (0);
2422 
2423  e_unlock:
2424 	VOP_UNLOCK(pr->pr_root);
2425  e_revert_osd:
2426 	/* Tell modules this thread is still in its old jail after all. */
2427 	sx_slock(&allprison_lock);
2428 	drflags |= PD_LIST_SLOCKED;
2429 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2430 	prison_deref(pr, drflags);
2431 	return (error);
2432 }
2433 
2434 /*
2435  * Returns a locked prison instance, or NULL on failure.
2436  */
2437 struct prison *
2438 prison_find(int prid)
2439 {
2440 	struct prison *pr;
2441 
2442 	sx_assert(&allprison_lock, SX_LOCKED);
2443 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2444 		if (pr->pr_id < prid)
2445 			continue;
2446 		if (pr->pr_id > prid)
2447 			break;
2448 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2449 		mtx_lock(&pr->pr_mtx);
2450 		return (pr);
2451 	}
2452 	return (NULL);
2453 }
2454 
2455 /*
2456  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2457  */
2458 struct prison *
2459 prison_find_child(struct prison *mypr, int prid)
2460 {
2461 	struct prison *pr;
2462 	int descend;
2463 
2464 	sx_assert(&allprison_lock, SX_LOCKED);
2465 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2466 		if (pr->pr_id == prid) {
2467 			KASSERT(prison_isvalid(pr),
2468 			    ("Found invalid prison %p", pr));
2469 			mtx_lock(&pr->pr_mtx);
2470 			return (pr);
2471 		}
2472 	}
2473 	return (NULL);
2474 }
2475 
2476 /*
2477  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2478  */
2479 struct prison *
2480 prison_find_name(struct prison *mypr, const char *name)
2481 {
2482 	struct prison *pr, *deadpr;
2483 	size_t mylen;
2484 	int descend;
2485 
2486 	sx_assert(&allprison_lock, SX_LOCKED);
2487 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2488 	deadpr = NULL;
2489 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2490 		if (!strcmp(pr->pr_name + mylen, name)) {
2491 			KASSERT(prison_isvalid(pr),
2492 			    ("Found invalid prison %p", pr));
2493 			if (prison_isalive(pr)) {
2494 				mtx_lock(&pr->pr_mtx);
2495 				return (pr);
2496 			}
2497 			deadpr = pr;
2498 		}
2499 	}
2500 	/* There was no valid prison - perhaps there was a dying one. */
2501 	if (deadpr != NULL)
2502 		mtx_lock(&deadpr->pr_mtx);
2503 	return (deadpr);
2504 }
2505 
2506 /*
2507  * See if a prison has the specific flag set.  The prison should be locked,
2508  * unless checking for flags that are only set at jail creation (such as
2509  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2510  * to any other prison data.
2511  */
2512 int
2513 prison_flag(struct ucred *cred, unsigned flag)
2514 {
2515 
2516 	return (cred->cr_prison->pr_flags & flag);
2517 }
2518 
2519 int
2520 prison_allow(struct ucred *cred, unsigned flag)
2521 {
2522 
2523 	return ((cred->cr_prison->pr_allow & flag) != 0);
2524 }
2525 
2526 /*
2527  * Hold a prison reference, by incrementing pr_ref.  It is generally
2528  * an error to hold a prison that does not already have a reference.
2529  * A prison record will remain valid as long as it has at least one
2530  * reference, and will not be removed as long as either the prison
2531  * mutex or the allprison lock is held (allprison_lock may be shared).
2532  */
2533 void
2534 prison_hold_locked(struct prison *pr)
2535 {
2536 
2537 	/* Locking is no longer required. */
2538 	prison_hold(pr);
2539 }
2540 
2541 void
2542 prison_hold(struct prison *pr)
2543 {
2544 #ifdef INVARIANTS
2545 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2546 
2547 	KASSERT(was_valid,
2548 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2549 #else
2550 	refcount_acquire(&pr->pr_ref);
2551 #endif
2552 }
2553 
2554 /*
2555  * Remove a prison reference.  If that was the last reference, the
2556  * prison will be removed (at a later time).
2557  */
2558 void
2559 prison_free_locked(struct prison *pr)
2560 {
2561 
2562 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2563 	/*
2564 	 * Locking is no longer required, but unlock because the caller
2565 	 * expects it.
2566 	 */
2567 	mtx_unlock(&pr->pr_mtx);
2568 	prison_free(pr);
2569 }
2570 
2571 void
2572 prison_free(struct prison *pr)
2573 {
2574 
2575 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2576 	    ("Trying to free dead prison %p (jid=%d).",
2577 	     pr, pr->pr_id));
2578 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2579 		/*
2580 		 * Don't remove the last reference in this context,
2581 		 * in case there are locks held.
2582 		 */
2583 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2584 	}
2585 }
2586 
2587 static void
2588 prison_free_not_last(struct prison *pr)
2589 {
2590 #ifdef INVARIANTS
2591 	int lastref;
2592 
2593 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2594 	    ("Trying to free dead prison %p (jid=%d).",
2595 	     pr, pr->pr_id));
2596 	lastref = refcount_release(&pr->pr_ref);
2597 	KASSERT(!lastref,
2598 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2599 	     pr, pr->pr_id));
2600 #else
2601 	refcount_release(&pr->pr_ref);
2602 #endif
2603 }
2604 
2605 /*
2606  * Hold a a prison for user visibility, by incrementing pr_uref.
2607  * It is generally an error to hold a prison that isn't already
2608  * user-visible, except through the the jail system calls.  It is also
2609  * an error to hold an invalid prison.  A prison record will remain
2610  * alive as long as it has at least one user reference, and will not
2611  * be set to the dying state until the prison mutex and allprison_lock
2612  * are both freed.
2613  */
2614 void
2615 prison_proc_hold(struct prison *pr)
2616 {
2617 #ifdef INVARIANTS
2618 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2619 
2620 	KASSERT(was_alive,
2621 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2622 #else
2623 	refcount_acquire(&pr->pr_uref);
2624 #endif
2625 }
2626 
2627 /*
2628  * Remove a prison user reference.  If it was the last reference, the
2629  * prison will be considered "dying", and may be removed once all of
2630  * its references are dropped.
2631  */
2632 void
2633 prison_proc_free(struct prison *pr)
2634 {
2635 
2636 	/*
2637 	 * Locking is only required when releasing the last reference.
2638 	 * This allows assurance that a locked prison will remain alive
2639 	 * until it is unlocked.
2640 	 */
2641 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2642 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2643 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2644 		/*
2645 		 * Don't remove the last user reference in this context,
2646 		 * which is expected to be a process that is not only locked,
2647 		 * but also half dead.  Add a reference so any calls to
2648 		 * prison_free() won't re-submit the task.
2649 		 */
2650 		prison_hold(pr);
2651 		mtx_lock(&pr->pr_mtx);
2652 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2653 		    ("Redundant last reference in prison_proc_free (jid=%d)",
2654 		     pr->pr_id));
2655 		pr->pr_flags |= PR_COMPLETE_PROC;
2656 		mtx_unlock(&pr->pr_mtx);
2657 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2658 	}
2659 }
2660 
2661 static void
2662 prison_proc_free_not_last(struct prison *pr)
2663 {
2664 #ifdef INVARIANTS
2665 	int lastref;
2666 
2667 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2668 	    ("Trying to free dead prison %p (jid=%d).",
2669 	     pr, pr->pr_id));
2670 	lastref = refcount_release(&pr->pr_uref);
2671 	KASSERT(!lastref,
2672 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2673 	     pr, pr->pr_id));
2674 #else
2675 	refcount_release(&pr->pr_uref);
2676 #endif
2677 }
2678 
2679 /*
2680  * Complete a call to either prison_free or prison_proc_free.
2681  */
2682 static void
2683 prison_complete(void *context, int pending)
2684 {
2685 	struct prison *pr = context;
2686 	int drflags;
2687 
2688 	/*
2689 	 * This could be called to release the last reference, or the last
2690 	 * user reference (plus the reference held in prison_proc_free).
2691 	 */
2692 	drflags = prison_lock_xlock(pr, PD_DEREF);
2693 	if (pr->pr_flags & PR_COMPLETE_PROC) {
2694 		pr->pr_flags &= ~PR_COMPLETE_PROC;
2695 		drflags |= PD_DEUREF;
2696 	}
2697 	prison_deref(pr, drflags);
2698 }
2699 
2700 /*
2701  * Remove a prison reference and/or user reference (usually).
2702  * This assumes context that allows sleeping (for allprison_lock),
2703  * with no non-sleeping locks held, except perhaps the prison itself.
2704  * If there are no more references, release and delist the prison.
2705  * On completion, the prison lock and the allprison lock are both
2706  * unlocked.
2707  */
2708 static void
2709 prison_deref(struct prison *pr, int flags)
2710 {
2711 	struct prisonlist freeprison;
2712 	struct prison *killpr, *rpr, *ppr, *tpr;
2713 	struct proc *p;
2714 
2715 	killpr = NULL;
2716 	TAILQ_INIT(&freeprison);
2717 	/*
2718 	 * Release this prison as requested, which may cause its parent
2719 	 * to be released, and then maybe its grandparent, etc.
2720 	 */
2721 	for (;;) {
2722 		if (flags & PD_KILL) {
2723 			/* Kill the prison and its descendents. */
2724 			KASSERT(pr != &prison0,
2725 			    ("prison_deref trying to kill prison0"));
2726 			if (!(flags & PD_DEREF)) {
2727 				prison_hold(pr);
2728 				flags |= PD_DEREF;
2729 			}
2730 			flags = prison_lock_xlock(pr, flags);
2731 			prison_deref_kill(pr, &freeprison);
2732 		}
2733 		if (flags & PD_DEUREF) {
2734 			/* Drop a user reference. */
2735 			KASSERT(refcount_load(&pr->pr_uref) > 0,
2736 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2737 			     pr->pr_id));
2738 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
2739 				if (!(flags & PD_DEREF)) {
2740 					prison_hold(pr);
2741 					flags |= PD_DEREF;
2742 				}
2743 				flags = prison_lock_xlock(pr, flags);
2744 				if (refcount_release(&pr->pr_uref) &&
2745 				    pr->pr_state == PRISON_STATE_ALIVE) {
2746 					/*
2747 					 * When the last user references goes,
2748 					 * this becomes a dying prison.
2749 					 */
2750 					KASSERT(
2751 					    refcount_load(&prison0.pr_uref) > 0,
2752 					    ("prison0 pr_uref=0"));
2753 					pr->pr_state = PRISON_STATE_DYING;
2754 					mtx_unlock(&pr->pr_mtx);
2755 					flags &= ~PD_LOCKED;
2756 					(void)osd_jail_call(pr,
2757 					    PR_METHOD_REMOVE, NULL);
2758 				}
2759 			}
2760 		}
2761 		if (flags & PD_KILL) {
2762 			/*
2763 			 * Any remaining user references are probably processes
2764 			 * that need to be killed, either in this prison or its
2765 			 * descendants.
2766 			 */
2767 			if (refcount_load(&pr->pr_uref) > 0)
2768 				killpr = pr;
2769 			/* Make sure the parent prison doesn't get killed. */
2770 			flags &= ~PD_KILL;
2771 		}
2772 		if (flags & PD_DEREF) {
2773 			/* Drop a reference. */
2774 			KASSERT(refcount_load(&pr->pr_ref) > 0,
2775 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2776 			     pr->pr_id));
2777 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
2778 				flags = prison_lock_xlock(pr, flags);
2779 				if (refcount_release(&pr->pr_ref)) {
2780 					/*
2781 					 * When the last reference goes,
2782 					 * unlink the prison and set it aside.
2783 					 */
2784 					KASSERT(
2785 					    refcount_load(&pr->pr_uref) == 0,
2786 					    ("prison_deref: last ref, "
2787 					     "but still has %d urefs (jid=%d)",
2788 					     pr->pr_uref, pr->pr_id));
2789 					KASSERT(
2790 					    refcount_load(&prison0.pr_ref) != 0,
2791 					    ("prison0 pr_ref=0"));
2792 					pr->pr_state = PRISON_STATE_INVALID;
2793 					TAILQ_REMOVE(&allprison, pr, pr_list);
2794 					LIST_REMOVE(pr, pr_sibling);
2795 					TAILQ_INSERT_TAIL(&freeprison, pr,
2796 					    pr_list);
2797 					for (ppr = pr->pr_parent;
2798 					     ppr != NULL;
2799 					     ppr = ppr->pr_parent)
2800 						ppr->pr_childcount--;
2801 					/*
2802 					 * Removing a prison frees references
2803 					 * from its parent.
2804 					 */
2805 					mtx_unlock(&pr->pr_mtx);
2806 					flags &= ~PD_LOCKED;
2807 					pr = pr->pr_parent;
2808 					flags |= PD_DEREF | PD_DEUREF;
2809 					continue;
2810 				}
2811 			}
2812 		}
2813 		break;
2814 	}
2815 
2816 	/* Release all the prison locks. */
2817 	if (flags & PD_LOCKED)
2818 		mtx_unlock(&pr->pr_mtx);
2819 	if (flags & PD_LIST_SLOCKED)
2820 		sx_sunlock(&allprison_lock);
2821 	else if (flags & PD_LIST_XLOCKED)
2822 		sx_xunlock(&allprison_lock);
2823 
2824 	/* Kill any processes attached to a killed prison. */
2825 	if (killpr != NULL) {
2826 		sx_slock(&allproc_lock);
2827 		FOREACH_PROC_IN_SYSTEM(p) {
2828 			PROC_LOCK(p);
2829 			if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
2830 				for (ppr = p->p_ucred->cr_prison;
2831 				     ppr != &prison0;
2832 				     ppr = ppr->pr_parent)
2833 					if (ppr == killpr) {
2834 						kern_psignal(p, SIGKILL);
2835 						break;
2836 					}
2837 			}
2838 			PROC_UNLOCK(p);
2839 		}
2840 		sx_sunlock(&allproc_lock);
2841 	}
2842 
2843 	/*
2844 	 * Finish removing any unreferenced prisons, which couldn't happen
2845 	 * while allprison_lock was held (to avoid a LOR on vrele).
2846 	 */
2847 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
2848 #ifdef VIMAGE
2849 		if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
2850 			vnet_destroy(rpr->pr_vnet);
2851 #endif
2852 		if (rpr->pr_root != NULL)
2853 			vrele(rpr->pr_root);
2854 		mtx_destroy(&rpr->pr_mtx);
2855 #ifdef INET
2856 		free(rpr->pr_ip4, M_PRISON);
2857 #endif
2858 #ifdef INET6
2859 		free(rpr->pr_ip6, M_PRISON);
2860 #endif
2861 		if (rpr->pr_cpuset != NULL)
2862 			cpuset_rel(rpr->pr_cpuset);
2863 		osd_jail_exit(rpr);
2864 #ifdef RACCT
2865 		if (racct_enable)
2866 			prison_racct_detach(rpr);
2867 #endif
2868 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
2869 		free(rpr, M_PRISON);
2870 	}
2871 }
2872 
2873 /*
2874  * Kill the prison and its descendants.  Mark them as dying, clear the
2875  * persist flag, and call module remove methods.
2876  */
2877 static void
2878 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
2879 {
2880 	struct prison *cpr, *ppr, *rpr;
2881 	bool descend;
2882 
2883 	/*
2884 	 * Unlike the descendants, the target prison can be killed
2885 	 * even if it is currently dying.  This is useful for failed
2886 	 * creation in jail_set(2).
2887 	 */
2888 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2889 	    ("Trying to kill dead prison %p (jid=%d).",
2890 	     pr, pr->pr_id));
2891 	refcount_acquire(&pr->pr_uref);
2892 	pr->pr_state = PRISON_STATE_DYING;
2893 	mtx_unlock(&pr->pr_mtx);
2894 
2895 	rpr = NULL;
2896 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
2897 		if (descend) {
2898 			if (!prison_isalive(cpr)) {
2899 				descend = false;
2900 				continue;
2901 			}
2902 			prison_hold(cpr);
2903 			prison_proc_hold(cpr);
2904 			mtx_lock(&cpr->pr_mtx);
2905 			cpr->pr_state = PRISON_STATE_DYING;
2906 			cpr->pr_flags |= PR_REMOVE;
2907 			mtx_unlock(&cpr->pr_mtx);
2908 			continue;
2909 		}
2910 		if (!(cpr->pr_flags & PR_REMOVE))
2911 			continue;
2912 		(void)osd_jail_call(cpr, PR_METHOD_REMOVE, NULL);
2913 		mtx_lock(&cpr->pr_mtx);
2914 		cpr->pr_flags &= ~PR_REMOVE;
2915 		if (cpr->pr_flags & PR_PERSIST) {
2916 			cpr->pr_flags &= ~PR_PERSIST;
2917 			prison_proc_free_not_last(cpr);
2918 			prison_free_not_last(cpr);
2919 		}
2920 		(void)refcount_release(&cpr->pr_uref);
2921 		if (refcount_release(&cpr->pr_ref)) {
2922 			/*
2923 			 * When the last reference goes, unlink the prison
2924 			 * and set it aside for prison_deref() to handle.
2925 			 * Delay unlinking the sibling list to keep the loop
2926 			 * safe.
2927 			 */
2928 			if (rpr != NULL)
2929 				LIST_REMOVE(rpr, pr_sibling);
2930 			rpr = cpr;
2931 			rpr->pr_state = PRISON_STATE_INVALID;
2932 			TAILQ_REMOVE(&allprison, rpr, pr_list);
2933 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
2934 			/*
2935 			 * Removing a prison frees references from its parent.
2936 			 */
2937 			ppr = rpr->pr_parent;
2938 			prison_proc_free_not_last(ppr);
2939 			prison_free_not_last(ppr);
2940 			for (; ppr != NULL; ppr = ppr->pr_parent)
2941 				ppr->pr_childcount--;
2942 		}
2943 		mtx_unlock(&cpr->pr_mtx);
2944 	}
2945 	if (rpr != NULL)
2946 		LIST_REMOVE(rpr, pr_sibling);
2947 
2948 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2949 	mtx_lock(&pr->pr_mtx);
2950 	if (pr->pr_flags & PR_PERSIST) {
2951 		pr->pr_flags &= ~PR_PERSIST;
2952 		prison_proc_free_not_last(pr);
2953 		prison_free_not_last(pr);
2954 	}
2955 	(void)refcount_release(&pr->pr_uref);
2956 }
2957 
2958 /*
2959  * Given the current locking state in the flags, make sure allprison_lock
2960  * is held exclusive, and the prison is locked.  Return flags indicating
2961  * the new state.
2962  */
2963 static int
2964 prison_lock_xlock(struct prison *pr, int flags)
2965 {
2966 
2967 	if (!(flags & PD_LIST_XLOCKED)) {
2968 		/*
2969 		 * Get allprison_lock, which may be an upgrade,
2970 		 * and may require unlocking the prison.
2971 		 */
2972 		if (flags & PD_LOCKED) {
2973 			mtx_unlock(&pr->pr_mtx);
2974 			flags &= ~PD_LOCKED;
2975 		}
2976 		if (flags & PD_LIST_SLOCKED) {
2977 			if (!sx_try_upgrade(&allprison_lock)) {
2978 				sx_sunlock(&allprison_lock);
2979 				sx_xlock(&allprison_lock);
2980 			}
2981 			flags &= ~PD_LIST_SLOCKED;
2982 		} else
2983 			sx_xlock(&allprison_lock);
2984 		flags |= PD_LIST_XLOCKED;
2985 	}
2986 	if (!(flags & PD_LOCKED)) {
2987 		/* Lock the prison mutex. */
2988 		mtx_lock(&pr->pr_mtx);
2989 		flags |= PD_LOCKED;
2990 	}
2991 	return flags;
2992 }
2993 
2994 /*
2995  * Set or clear a permission bit in the pr_allow field, passing restrictions
2996  * (cleared permission) down to child jails.
2997  */
2998 void
2999 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3000 {
3001 	struct prison *pr;
3002 
3003 	pr = cred->cr_prison;
3004 	sx_slock(&allprison_lock);
3005 	mtx_lock(&pr->pr_mtx);
3006 	prison_set_allow_locked(pr, flag, enable);
3007 	mtx_unlock(&pr->pr_mtx);
3008 	sx_sunlock(&allprison_lock);
3009 }
3010 
3011 static void
3012 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3013 {
3014 	struct prison *cpr;
3015 	int descend;
3016 
3017 	if (enable != 0)
3018 		pr->pr_allow |= flag;
3019 	else {
3020 		pr->pr_allow &= ~flag;
3021 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3022 			cpr->pr_allow &= ~flag;
3023 	}
3024 }
3025 
3026 /*
3027  * Check if a jail supports the given address family.
3028  *
3029  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3030  * if not.
3031  */
3032 int
3033 prison_check_af(struct ucred *cred, int af)
3034 {
3035 	struct prison *pr;
3036 	int error;
3037 
3038 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3039 
3040 	pr = cred->cr_prison;
3041 #ifdef VIMAGE
3042 	/* Prisons with their own network stack are not limited. */
3043 	if (prison_owns_vnet(cred))
3044 		return (0);
3045 #endif
3046 
3047 	error = 0;
3048 	switch (af)
3049 	{
3050 #ifdef INET
3051 	case AF_INET:
3052 		if (pr->pr_flags & PR_IP4)
3053 		{
3054 			mtx_lock(&pr->pr_mtx);
3055 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3056 				error = EAFNOSUPPORT;
3057 			mtx_unlock(&pr->pr_mtx);
3058 		}
3059 		break;
3060 #endif
3061 #ifdef INET6
3062 	case AF_INET6:
3063 		if (pr->pr_flags & PR_IP6)
3064 		{
3065 			mtx_lock(&pr->pr_mtx);
3066 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3067 				error = EAFNOSUPPORT;
3068 			mtx_unlock(&pr->pr_mtx);
3069 		}
3070 		break;
3071 #endif
3072 	case AF_LOCAL:
3073 	case AF_ROUTE:
3074 		break;
3075 	default:
3076 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3077 			error = EAFNOSUPPORT;
3078 	}
3079 	return (error);
3080 }
3081 
3082 /*
3083  * Check if given address belongs to the jail referenced by cred (wrapper to
3084  * prison_check_ip[46]).
3085  *
3086  * Returns 0 if jail doesn't restrict the address family or if address belongs
3087  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3088  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3089  */
3090 int
3091 prison_if(struct ucred *cred, const struct sockaddr *sa)
3092 {
3093 #ifdef INET
3094 	const struct sockaddr_in *sai;
3095 #endif
3096 #ifdef INET6
3097 	const struct sockaddr_in6 *sai6;
3098 #endif
3099 	int error;
3100 
3101 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3102 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3103 
3104 #ifdef VIMAGE
3105 	if (prison_owns_vnet(cred))
3106 		return (0);
3107 #endif
3108 
3109 	error = 0;
3110 	switch (sa->sa_family)
3111 	{
3112 #ifdef INET
3113 	case AF_INET:
3114 		sai = (const struct sockaddr_in *)sa;
3115 		error = prison_check_ip4(cred, &sai->sin_addr);
3116 		break;
3117 #endif
3118 #ifdef INET6
3119 	case AF_INET6:
3120 		sai6 = (const struct sockaddr_in6 *)sa;
3121 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3122 		break;
3123 #endif
3124 	default:
3125 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3126 			error = EAFNOSUPPORT;
3127 	}
3128 	return (error);
3129 }
3130 
3131 /*
3132  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3133  */
3134 int
3135 prison_check(struct ucred *cred1, struct ucred *cred2)
3136 {
3137 
3138 	return ((cred1->cr_prison == cred2->cr_prison ||
3139 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3140 }
3141 
3142 /*
3143  * Return 1 if p2 is a child of p1, otherwise 0.
3144  */
3145 int
3146 prison_ischild(struct prison *pr1, struct prison *pr2)
3147 {
3148 
3149 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3150 		if (pr1 == pr2)
3151 			return (1);
3152 	return (0);
3153 }
3154 
3155 /*
3156  * Return true if the prison is currently alive.  A prison is alive if it
3157  * holds user references and it isn't being removed.
3158  */
3159 bool
3160 prison_isalive(struct prison *pr)
3161 {
3162 
3163 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3164 		return (false);
3165 	return (true);
3166 }
3167 
3168 /*
3169  * Return true if the prison is currently valid.  A prison is valid if it has
3170  * been fully created, and is not being destroyed.  Note that dying prisons
3171  * are still considered valid.  Invalid prisons won't be found under normal
3172  * circumstances, as they're only put in that state by functions that have
3173  * an exclusive hold on allprison_lock.
3174  */
3175 bool
3176 prison_isvalid(struct prison *pr)
3177 {
3178 
3179 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3180 		return (false);
3181 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3182 		return (false);
3183 	return (true);
3184 }
3185 
3186 /*
3187  * Return 1 if the passed credential is in a jail and that jail does not
3188  * have its own virtual network stack, otherwise 0.
3189  */
3190 int
3191 jailed_without_vnet(struct ucred *cred)
3192 {
3193 
3194 	if (!jailed(cred))
3195 		return (0);
3196 #ifdef VIMAGE
3197 	if (prison_owns_vnet(cred))
3198 		return (0);
3199 #endif
3200 
3201 	return (1);
3202 }
3203 
3204 /*
3205  * Return the correct hostname (domainname, et al) for the passed credential.
3206  */
3207 void
3208 getcredhostname(struct ucred *cred, char *buf, size_t size)
3209 {
3210 	struct prison *pr;
3211 
3212 	/*
3213 	 * A NULL credential can be used to shortcut to the physical
3214 	 * system's hostname.
3215 	 */
3216 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3217 	mtx_lock(&pr->pr_mtx);
3218 	strlcpy(buf, pr->pr_hostname, size);
3219 	mtx_unlock(&pr->pr_mtx);
3220 }
3221 
3222 void
3223 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3224 {
3225 
3226 	mtx_lock(&cred->cr_prison->pr_mtx);
3227 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3228 	mtx_unlock(&cred->cr_prison->pr_mtx);
3229 }
3230 
3231 void
3232 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3233 {
3234 
3235 	mtx_lock(&cred->cr_prison->pr_mtx);
3236 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3237 	mtx_unlock(&cred->cr_prison->pr_mtx);
3238 }
3239 
3240 void
3241 getcredhostid(struct ucred *cred, unsigned long *hostid)
3242 {
3243 
3244 	mtx_lock(&cred->cr_prison->pr_mtx);
3245 	*hostid = cred->cr_prison->pr_hostid;
3246 	mtx_unlock(&cred->cr_prison->pr_mtx);
3247 }
3248 
3249 void
3250 getjailname(struct ucred *cred, char *name, size_t len)
3251 {
3252 
3253 	mtx_lock(&cred->cr_prison->pr_mtx);
3254 	strlcpy(name, cred->cr_prison->pr_name, len);
3255 	mtx_unlock(&cred->cr_prison->pr_mtx);
3256 }
3257 
3258 #ifdef VIMAGE
3259 /*
3260  * Determine whether the prison represented by cred owns
3261  * its vnet rather than having it inherited.
3262  *
3263  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3264  */
3265 int
3266 prison_owns_vnet(struct ucred *cred)
3267 {
3268 
3269 	/*
3270 	 * vnets cannot be added/removed after jail creation,
3271 	 * so no need to lock here.
3272 	 */
3273 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3274 }
3275 #endif
3276 
3277 /*
3278  * Determine whether the subject represented by cred can "see"
3279  * status of a mount point.
3280  * Returns: 0 for permitted, ENOENT otherwise.
3281  * XXX: This function should be called cr_canseemount() and should be
3282  *      placed in kern_prot.c.
3283  */
3284 int
3285 prison_canseemount(struct ucred *cred, struct mount *mp)
3286 {
3287 	struct prison *pr;
3288 	struct statfs *sp;
3289 	size_t len;
3290 
3291 	pr = cred->cr_prison;
3292 	if (pr->pr_enforce_statfs == 0)
3293 		return (0);
3294 	if (pr->pr_root->v_mount == mp)
3295 		return (0);
3296 	if (pr->pr_enforce_statfs == 2)
3297 		return (ENOENT);
3298 	/*
3299 	 * If jail's chroot directory is set to "/" we should be able to see
3300 	 * all mount-points from inside a jail.
3301 	 * This is ugly check, but this is the only situation when jail's
3302 	 * directory ends with '/'.
3303 	 */
3304 	if (strcmp(pr->pr_path, "/") == 0)
3305 		return (0);
3306 	len = strlen(pr->pr_path);
3307 	sp = &mp->mnt_stat;
3308 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3309 		return (ENOENT);
3310 	/*
3311 	 * Be sure that we don't have situation where jail's root directory
3312 	 * is "/some/path" and mount point is "/some/pathpath".
3313 	 */
3314 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3315 		return (ENOENT);
3316 	return (0);
3317 }
3318 
3319 void
3320 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3321 {
3322 	char jpath[MAXPATHLEN];
3323 	struct prison *pr;
3324 	size_t len;
3325 
3326 	pr = cred->cr_prison;
3327 	if (pr->pr_enforce_statfs == 0)
3328 		return;
3329 	if (prison_canseemount(cred, mp) != 0) {
3330 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3331 		strlcpy(sp->f_mntonname, "[restricted]",
3332 		    sizeof(sp->f_mntonname));
3333 		return;
3334 	}
3335 	if (pr->pr_root->v_mount == mp) {
3336 		/*
3337 		 * Clear current buffer data, so we are sure nothing from
3338 		 * the valid path left there.
3339 		 */
3340 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3341 		*sp->f_mntonname = '/';
3342 		return;
3343 	}
3344 	/*
3345 	 * If jail's chroot directory is set to "/" we should be able to see
3346 	 * all mount-points from inside a jail.
3347 	 */
3348 	if (strcmp(pr->pr_path, "/") == 0)
3349 		return;
3350 	len = strlen(pr->pr_path);
3351 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3352 	/*
3353 	 * Clear current buffer data, so we are sure nothing from
3354 	 * the valid path left there.
3355 	 */
3356 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3357 	if (*jpath == '\0') {
3358 		/* Should never happen. */
3359 		*sp->f_mntonname = '/';
3360 	} else {
3361 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3362 	}
3363 }
3364 
3365 /*
3366  * Check with permission for a specific privilege is granted within jail.  We
3367  * have a specific list of accepted privileges; the rest are denied.
3368  */
3369 int
3370 prison_priv_check(struct ucred *cred, int priv)
3371 {
3372 	struct prison *pr;
3373 	int error;
3374 
3375 	/*
3376 	 * Some policies have custom handlers. This routine should not be
3377 	 * called for them. See priv_check_cred().
3378 	 */
3379 	switch (priv) {
3380 	case PRIV_VFS_LOOKUP:
3381 	case PRIV_VFS_GENERATION:
3382 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3383 		    "called for %d\n", priv));
3384 	}
3385 
3386 	if (!jailed(cred))
3387 		return (0);
3388 
3389 #ifdef VIMAGE
3390 	/*
3391 	 * Privileges specific to prisons with a virtual network stack.
3392 	 * There might be a duplicate entry here in case the privilege
3393 	 * is only granted conditionally in the legacy jail case.
3394 	 */
3395 	switch (priv) {
3396 #ifdef notyet
3397 		/*
3398 		 * NFS-specific privileges.
3399 		 */
3400 	case PRIV_NFS_DAEMON:
3401 	case PRIV_NFS_LOCKD:
3402 #endif
3403 		/*
3404 		 * Network stack privileges.
3405 		 */
3406 	case PRIV_NET_BRIDGE:
3407 	case PRIV_NET_GRE:
3408 	case PRIV_NET_BPF:
3409 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3410 	case PRIV_NET_ROUTE:
3411 	case PRIV_NET_TAP:
3412 	case PRIV_NET_SETIFMTU:
3413 	case PRIV_NET_SETIFFLAGS:
3414 	case PRIV_NET_SETIFCAP:
3415 	case PRIV_NET_SETIFDESCR:
3416 	case PRIV_NET_SETIFNAME	:
3417 	case PRIV_NET_SETIFMETRIC:
3418 	case PRIV_NET_SETIFPHYS:
3419 	case PRIV_NET_SETIFMAC:
3420 	case PRIV_NET_SETLANPCP:
3421 	case PRIV_NET_ADDMULTI:
3422 	case PRIV_NET_DELMULTI:
3423 	case PRIV_NET_HWIOCTL:
3424 	case PRIV_NET_SETLLADDR:
3425 	case PRIV_NET_ADDIFGROUP:
3426 	case PRIV_NET_DELIFGROUP:
3427 	case PRIV_NET_IFCREATE:
3428 	case PRIV_NET_IFDESTROY:
3429 	case PRIV_NET_ADDIFADDR:
3430 	case PRIV_NET_DELIFADDR:
3431 	case PRIV_NET_LAGG:
3432 	case PRIV_NET_GIF:
3433 	case PRIV_NET_SETIFVNET:
3434 	case PRIV_NET_SETIFFIB:
3435 
3436 		/*
3437 		 * 802.11-related privileges.
3438 		 */
3439 	case PRIV_NET80211_VAP_GETKEY:
3440 	case PRIV_NET80211_VAP_MANAGE:
3441 
3442 #ifdef notyet
3443 		/*
3444 		 * ATM privileges.
3445 		 */
3446 	case PRIV_NETATM_CFG:
3447 	case PRIV_NETATM_ADD:
3448 	case PRIV_NETATM_DEL:
3449 	case PRIV_NETATM_SET:
3450 
3451 		/*
3452 		 * Bluetooth privileges.
3453 		 */
3454 	case PRIV_NETBLUETOOTH_RAW:
3455 #endif
3456 
3457 		/*
3458 		 * Netgraph and netgraph module privileges.
3459 		 */
3460 	case PRIV_NETGRAPH_CONTROL:
3461 #ifdef notyet
3462 	case PRIV_NETGRAPH_TTY:
3463 #endif
3464 
3465 		/*
3466 		 * IPv4 and IPv6 privileges.
3467 		 */
3468 	case PRIV_NETINET_IPFW:
3469 	case PRIV_NETINET_DIVERT:
3470 	case PRIV_NETINET_PF:
3471 	case PRIV_NETINET_DUMMYNET:
3472 	case PRIV_NETINET_CARP:
3473 	case PRIV_NETINET_MROUTE:
3474 	case PRIV_NETINET_RAW:
3475 	case PRIV_NETINET_ADDRCTRL6:
3476 	case PRIV_NETINET_ND6:
3477 	case PRIV_NETINET_SCOPE6:
3478 	case PRIV_NETINET_ALIFETIME6:
3479 	case PRIV_NETINET_IPSEC:
3480 	case PRIV_NETINET_BINDANY:
3481 
3482 #ifdef notyet
3483 		/*
3484 		 * NCP privileges.
3485 		 */
3486 	case PRIV_NETNCP:
3487 
3488 		/*
3489 		 * SMB privileges.
3490 		 */
3491 	case PRIV_NETSMB:
3492 #endif
3493 
3494 	/*
3495 	 * No default: or deny here.
3496 	 * In case of no permit fall through to next switch().
3497 	 */
3498 		if (cred->cr_prison->pr_flags & PR_VNET)
3499 			return (0);
3500 	}
3501 #endif /* VIMAGE */
3502 
3503 	switch (priv) {
3504 		/*
3505 		 * Allow ktrace privileges for root in jail.
3506 		 */
3507 	case PRIV_KTRACE:
3508 
3509 #if 0
3510 		/*
3511 		 * Allow jailed processes to configure audit identity and
3512 		 * submit audit records (login, etc).  In the future we may
3513 		 * want to further refine the relationship between audit and
3514 		 * jail.
3515 		 */
3516 	case PRIV_AUDIT_GETAUDIT:
3517 	case PRIV_AUDIT_SETAUDIT:
3518 	case PRIV_AUDIT_SUBMIT:
3519 #endif
3520 
3521 		/*
3522 		 * Allow jailed processes to manipulate process UNIX
3523 		 * credentials in any way they see fit.
3524 		 */
3525 	case PRIV_CRED_SETUID:
3526 	case PRIV_CRED_SETEUID:
3527 	case PRIV_CRED_SETGID:
3528 	case PRIV_CRED_SETEGID:
3529 	case PRIV_CRED_SETGROUPS:
3530 	case PRIV_CRED_SETREUID:
3531 	case PRIV_CRED_SETREGID:
3532 	case PRIV_CRED_SETRESUID:
3533 	case PRIV_CRED_SETRESGID:
3534 
3535 		/*
3536 		 * Jail implements visibility constraints already, so allow
3537 		 * jailed root to override uid/gid-based constraints.
3538 		 */
3539 	case PRIV_SEEOTHERGIDS:
3540 	case PRIV_SEEOTHERUIDS:
3541 
3542 		/*
3543 		 * Jail implements inter-process debugging limits already, so
3544 		 * allow jailed root various debugging privileges.
3545 		 */
3546 	case PRIV_DEBUG_DIFFCRED:
3547 	case PRIV_DEBUG_SUGID:
3548 	case PRIV_DEBUG_UNPRIV:
3549 
3550 		/*
3551 		 * Allow jail to set various resource limits and login
3552 		 * properties, and for now, exceed process resource limits.
3553 		 */
3554 	case PRIV_PROC_LIMIT:
3555 	case PRIV_PROC_SETLOGIN:
3556 	case PRIV_PROC_SETRLIMIT:
3557 
3558 		/*
3559 		 * System V and POSIX IPC privileges are granted in jail.
3560 		 */
3561 	case PRIV_IPC_READ:
3562 	case PRIV_IPC_WRITE:
3563 	case PRIV_IPC_ADMIN:
3564 	case PRIV_IPC_MSGSIZE:
3565 	case PRIV_MQ_ADMIN:
3566 
3567 		/*
3568 		 * Jail operations within a jail work on child jails.
3569 		 */
3570 	case PRIV_JAIL_ATTACH:
3571 	case PRIV_JAIL_SET:
3572 	case PRIV_JAIL_REMOVE:
3573 
3574 		/*
3575 		 * Jail implements its own inter-process limits, so allow
3576 		 * root processes in jail to change scheduling on other
3577 		 * processes in the same jail.  Likewise for signalling.
3578 		 */
3579 	case PRIV_SCHED_DIFFCRED:
3580 	case PRIV_SCHED_CPUSET:
3581 	case PRIV_SIGNAL_DIFFCRED:
3582 	case PRIV_SIGNAL_SUGID:
3583 
3584 		/*
3585 		 * Allow jailed processes to write to sysctls marked as jail
3586 		 * writable.
3587 		 */
3588 	case PRIV_SYSCTL_WRITEJAIL:
3589 
3590 		/*
3591 		 * Allow root in jail to manage a variety of quota
3592 		 * properties.  These should likely be conditional on a
3593 		 * configuration option.
3594 		 */
3595 	case PRIV_VFS_GETQUOTA:
3596 	case PRIV_VFS_SETQUOTA:
3597 
3598 		/*
3599 		 * Since Jail relies on chroot() to implement file system
3600 		 * protections, grant many VFS privileges to root in jail.
3601 		 * Be careful to exclude mount-related and NFS-related
3602 		 * privileges.
3603 		 */
3604 	case PRIV_VFS_READ:
3605 	case PRIV_VFS_WRITE:
3606 	case PRIV_VFS_ADMIN:
3607 	case PRIV_VFS_EXEC:
3608 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3609 	case PRIV_VFS_CHFLAGS_DEV:
3610 	case PRIV_VFS_CHOWN:
3611 	case PRIV_VFS_CHROOT:
3612 	case PRIV_VFS_RETAINSUGID:
3613 	case PRIV_VFS_FCHROOT:
3614 	case PRIV_VFS_LINK:
3615 	case PRIV_VFS_SETGID:
3616 	case PRIV_VFS_STAT:
3617 	case PRIV_VFS_STICKYFILE:
3618 
3619 		/*
3620 		 * As in the non-jail case, non-root users are expected to be
3621 		 * able to read kernel/physical memory (provided /dev/[k]mem
3622 		 * exists in the jail and they have permission to access it).
3623 		 */
3624 	case PRIV_KMEM_READ:
3625 		return (0);
3626 
3627 		/*
3628 		 * Depending on the global setting, allow privilege of
3629 		 * setting system flags.
3630 		 */
3631 	case PRIV_VFS_SYSFLAGS:
3632 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3633 			return (0);
3634 		else
3635 			return (EPERM);
3636 
3637 		/*
3638 		 * Depending on the global setting, allow privilege of
3639 		 * mounting/unmounting file systems.
3640 		 */
3641 	case PRIV_VFS_MOUNT:
3642 	case PRIV_VFS_UNMOUNT:
3643 	case PRIV_VFS_MOUNT_NONUSER:
3644 	case PRIV_VFS_MOUNT_OWNER:
3645 		pr = cred->cr_prison;
3646 		prison_lock(pr);
3647 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3648 			error = 0;
3649 		else
3650 			error = EPERM;
3651 		prison_unlock(pr);
3652 		return (error);
3653 
3654 		/*
3655 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3656 		 * policy.  priv_check_cred will not specifically allow it, and
3657 		 * we may want a MAC policy to allow it.
3658 		 */
3659 	case PRIV_VFS_READ_DIR:
3660 		return (0);
3661 
3662 		/*
3663 		 * Conditionnaly allow locking (unlocking) physical pages
3664 		 * in memory.
3665 		 */
3666 	case PRIV_VM_MLOCK:
3667 	case PRIV_VM_MUNLOCK:
3668 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3669 			return (0);
3670 		else
3671 			return (EPERM);
3672 
3673 		/*
3674 		 * Conditionally allow jailed root to bind reserved ports.
3675 		 */
3676 	case PRIV_NETINET_RESERVEDPORT:
3677 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3678 			return (0);
3679 		else
3680 			return (EPERM);
3681 
3682 		/*
3683 		 * Allow jailed root to reuse in-use ports.
3684 		 */
3685 	case PRIV_NETINET_REUSEPORT:
3686 		return (0);
3687 
3688 		/*
3689 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3690 		 */
3691 	case PRIV_NETINET_SETHDROPTS:
3692 		return (0);
3693 
3694 		/*
3695 		 * Conditionally allow creating raw sockets in jail.
3696 		 */
3697 	case PRIV_NETINET_RAW:
3698 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3699 			return (0);
3700 		else
3701 			return (EPERM);
3702 
3703 		/*
3704 		 * Since jail implements its own visibility limits on netstat
3705 		 * sysctls, allow getcred.  This allows identd to work in
3706 		 * jail.
3707 		 */
3708 	case PRIV_NETINET_GETCRED:
3709 		return (0);
3710 
3711 		/*
3712 		 * Allow jailed root to set loginclass.
3713 		 */
3714 	case PRIV_PROC_SETLOGINCLASS:
3715 		return (0);
3716 
3717 		/*
3718 		 * Do not allow a process inside a jail to read the kernel
3719 		 * message buffer unless explicitly permitted.
3720 		 */
3721 	case PRIV_MSGBUF:
3722 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3723 			return (0);
3724 		return (EPERM);
3725 
3726 	default:
3727 		/*
3728 		 * In all remaining cases, deny the privilege request.  This
3729 		 * includes almost all network privileges, many system
3730 		 * configuration privileges.
3731 		 */
3732 		return (EPERM);
3733 	}
3734 }
3735 
3736 /*
3737  * Return the part of pr2's name that is relative to pr1, or the whole name
3738  * if it does not directly follow.
3739  */
3740 
3741 char *
3742 prison_name(struct prison *pr1, struct prison *pr2)
3743 {
3744 	char *name;
3745 
3746 	/* Jails see themselves as "0" (if they see themselves at all). */
3747 	if (pr1 == pr2)
3748 		return "0";
3749 	name = pr2->pr_name;
3750 	if (prison_ischild(pr1, pr2)) {
3751 		/*
3752 		 * pr1 isn't locked (and allprison_lock may not be either)
3753 		 * so its length can't be counted on.  But the number of dots
3754 		 * can be counted on - and counted.
3755 		 */
3756 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3757 			name = strchr(name, '.') + 1;
3758 	}
3759 	return (name);
3760 }
3761 
3762 /*
3763  * Return the part of pr2's path that is relative to pr1, or the whole path
3764  * if it does not directly follow.
3765  */
3766 static char *
3767 prison_path(struct prison *pr1, struct prison *pr2)
3768 {
3769 	char *path1, *path2;
3770 	int len1;
3771 
3772 	path1 = pr1->pr_path;
3773 	path2 = pr2->pr_path;
3774 	if (!strcmp(path1, "/"))
3775 		return (path2);
3776 	len1 = strlen(path1);
3777 	if (strncmp(path1, path2, len1))
3778 		return (path2);
3779 	if (path2[len1] == '\0')
3780 		return "/";
3781 	if (path2[len1] == '/')
3782 		return (path2 + len1);
3783 	return (path2);
3784 }
3785 
3786 /*
3787  * Jail-related sysctls.
3788  */
3789 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3790     "Jails");
3791 
3792 static int
3793 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3794 {
3795 	struct xprison *xp;
3796 	struct prison *pr, *cpr;
3797 #ifdef INET
3798 	struct in_addr *ip4 = NULL;
3799 	int ip4s = 0;
3800 #endif
3801 #ifdef INET6
3802 	struct in6_addr *ip6 = NULL;
3803 	int ip6s = 0;
3804 #endif
3805 	int descend, error;
3806 
3807 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3808 	pr = req->td->td_ucred->cr_prison;
3809 	error = 0;
3810 	sx_slock(&allprison_lock);
3811 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3812 #if defined(INET) || defined(INET6)
3813  again:
3814 #endif
3815 		mtx_lock(&cpr->pr_mtx);
3816 #ifdef INET
3817 		if (cpr->pr_ip4s > 0) {
3818 			if (ip4s < cpr->pr_ip4s) {
3819 				ip4s = cpr->pr_ip4s;
3820 				mtx_unlock(&cpr->pr_mtx);
3821 				ip4 = realloc(ip4, ip4s *
3822 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3823 				goto again;
3824 			}
3825 			bcopy(cpr->pr_ip4, ip4,
3826 			    cpr->pr_ip4s * sizeof(struct in_addr));
3827 		}
3828 #endif
3829 #ifdef INET6
3830 		if (cpr->pr_ip6s > 0) {
3831 			if (ip6s < cpr->pr_ip6s) {
3832 				ip6s = cpr->pr_ip6s;
3833 				mtx_unlock(&cpr->pr_mtx);
3834 				ip6 = realloc(ip6, ip6s *
3835 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3836 				goto again;
3837 			}
3838 			bcopy(cpr->pr_ip6, ip6,
3839 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3840 		}
3841 #endif
3842 		bzero(xp, sizeof(*xp));
3843 		xp->pr_version = XPRISON_VERSION;
3844 		xp->pr_id = cpr->pr_id;
3845 		xp->pr_state = cpr->pr_state;
3846 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3847 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3848 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3849 #ifdef INET
3850 		xp->pr_ip4s = cpr->pr_ip4s;
3851 #endif
3852 #ifdef INET6
3853 		xp->pr_ip6s = cpr->pr_ip6s;
3854 #endif
3855 		mtx_unlock(&cpr->pr_mtx);
3856 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3857 		if (error)
3858 			break;
3859 #ifdef INET
3860 		if (xp->pr_ip4s > 0) {
3861 			error = SYSCTL_OUT(req, ip4,
3862 			    xp->pr_ip4s * sizeof(struct in_addr));
3863 			if (error)
3864 				break;
3865 		}
3866 #endif
3867 #ifdef INET6
3868 		if (xp->pr_ip6s > 0) {
3869 			error = SYSCTL_OUT(req, ip6,
3870 			    xp->pr_ip6s * sizeof(struct in6_addr));
3871 			if (error)
3872 				break;
3873 		}
3874 #endif
3875 	}
3876 	sx_sunlock(&allprison_lock);
3877 	free(xp, M_TEMP);
3878 #ifdef INET
3879 	free(ip4, M_TEMP);
3880 #endif
3881 #ifdef INET6
3882 	free(ip6, M_TEMP);
3883 #endif
3884 	return (error);
3885 }
3886 
3887 SYSCTL_OID(_security_jail, OID_AUTO, list,
3888     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3889     sysctl_jail_list, "S", "List of active jails");
3890 
3891 static int
3892 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3893 {
3894 	int error, injail;
3895 
3896 	injail = jailed(req->td->td_ucred);
3897 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3898 
3899 	return (error);
3900 }
3901 
3902 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3903     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3904     sysctl_jail_jailed, "I", "Process in jail?");
3905 
3906 static int
3907 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3908 {
3909 	int error, havevnet;
3910 #ifdef VIMAGE
3911 	struct ucred *cred = req->td->td_ucred;
3912 
3913 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3914 #else
3915 	havevnet = 0;
3916 #endif
3917 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3918 
3919 	return (error);
3920 }
3921 
3922 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3923     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3924     sysctl_jail_vnet, "I", "Jail owns vnet?");
3925 
3926 #if defined(INET) || defined(INET6)
3927 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3928     &jail_max_af_ips, 0,
3929     "Number of IP addresses a jail may have at most per address family (deprecated)");
3930 #endif
3931 
3932 /*
3933  * Default parameters for jail(2) compatibility.  For historical reasons,
3934  * the sysctl names have varying similarity to the parameter names.  Prisons
3935  * just see their own parameters, and can't change them.
3936  */
3937 static int
3938 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3939 {
3940 	int error, i;
3941 
3942 	/* Get the current flag value, and convert it to a boolean. */
3943 	if (req->td->td_ucred->cr_prison == &prison0) {
3944 		mtx_lock(&prison0.pr_mtx);
3945 		i = (jail_default_allow & arg2) != 0;
3946 		mtx_unlock(&prison0.pr_mtx);
3947 	} else
3948 		i = prison_allow(req->td->td_ucred, arg2);
3949 
3950 	if (arg1 != NULL)
3951 		i = !i;
3952 	error = sysctl_handle_int(oidp, &i, 0, req);
3953 	if (error || !req->newptr)
3954 		return (error);
3955 	i = i ? arg2 : 0;
3956 	if (arg1 != NULL)
3957 		i ^= arg2;
3958 	/*
3959 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3960 	 * for writing.
3961 	 */
3962 	mtx_lock(&prison0.pr_mtx);
3963 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3964 	mtx_unlock(&prison0.pr_mtx);
3965 	return (0);
3966 }
3967 
3968 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3969     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3970     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3971     "Processes in jail can set their hostnames (deprecated)");
3972 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3973     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3974     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3975     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3976 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3977     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3978     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3979     "Processes in jail can use System V IPC primitives (deprecated)");
3980 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3981     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3982     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3983     "Prison root can create raw sockets (deprecated)");
3984 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3985     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3986     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3987     "Processes in jail can alter system file flags (deprecated)");
3988 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3989     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3990     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3991     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3992 
3993 static int
3994 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3995 {
3996 	struct prison *pr;
3997 	int level, error;
3998 
3999 	pr = req->td->td_ucred->cr_prison;
4000 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4001 	error = sysctl_handle_int(oidp, &level, 0, req);
4002 	if (error || !req->newptr)
4003 		return (error);
4004 	*(int *)arg1 = level;
4005 	return (0);
4006 }
4007 
4008 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4009     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4010     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4011     sysctl_jail_default_level, "I",
4012     "Processes in jail cannot see all mounted file systems (deprecated)");
4013 
4014 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4015     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4016     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4017     sysctl_jail_default_level, "I",
4018     "Ruleset for the devfs filesystem in jail (deprecated)");
4019 
4020 /*
4021  * Nodes to describe jail parameters.  Maximum length of string parameters
4022  * is returned in the string itself, and the other parameters exist merely
4023  * to make themselves and their types known.
4024  */
4025 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4026     "Jail parameters");
4027 
4028 int
4029 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4030 {
4031 	int i;
4032 	long l;
4033 	size_t s;
4034 	char numbuf[12];
4035 
4036 	switch (oidp->oid_kind & CTLTYPE)
4037 	{
4038 	case CTLTYPE_LONG:
4039 	case CTLTYPE_ULONG:
4040 		l = 0;
4041 #ifdef SCTL_MASK32
4042 		if (!(req->flags & SCTL_MASK32))
4043 #endif
4044 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4045 	case CTLTYPE_INT:
4046 	case CTLTYPE_UINT:
4047 		i = 0;
4048 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4049 	case CTLTYPE_STRING:
4050 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4051 		return
4052 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4053 	case CTLTYPE_STRUCT:
4054 		s = (size_t)arg2;
4055 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4056 	}
4057 	return (0);
4058 }
4059 
4060 /*
4061  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4062  * jail creation time but cannot be changed in an existing jail.
4063  */
4064 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4065 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4066 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4067 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4068 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4069     "I", "Jail secure level");
4070 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4071     "Jail value for kern.osreldate and uname -K");
4072 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4073     "Jail value for kern.osrelease and uname -r");
4074 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4075     "I", "Jail cannot see all mounted file systems");
4076 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4077     "I", "Ruleset for in-jail devfs mounts");
4078 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4079     "B", "Jail persistence");
4080 #ifdef VIMAGE
4081 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4082     "E,jailsys", "Virtual network stack");
4083 #endif
4084 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4085     "B", "Jail is in the process of shutting down");
4086 
4087 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4088 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4089     "I", "Current number of child jails");
4090 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4091     "I", "Maximum number of child jails");
4092 
4093 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4094 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4095     "Jail hostname");
4096 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4097     "Jail NIS domainname");
4098 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4099     "Jail host UUID");
4100 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4101     "LU", "Jail host ID");
4102 
4103 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4104 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4105 
4106 #ifdef INET
4107 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4108     "Jail IPv4 address virtualization");
4109 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4110     "S,in_addr,a", "Jail IPv4 addresses");
4111 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4112     "B", "Do (not) use IPv4 source address selection rather than the "
4113     "primary jail IPv4 address.");
4114 #endif
4115 #ifdef INET6
4116 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4117     "Jail IPv6 address virtualization");
4118 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4119     "S,in6_addr,a", "Jail IPv6 addresses");
4120 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4121     "B", "Do (not) use IPv6 source address selection rather than the "
4122     "primary jail IPv6 address.");
4123 #endif
4124 
4125 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4126 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4127     "B", "Jail may set hostname");
4128 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4129     "B", "Jail may use SYSV IPC");
4130 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4131     "B", "Jail may create raw sockets");
4132 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4133     "B", "Jail may alter system file flags");
4134 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4135     "B", "Jail may set file quotas");
4136 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4137     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4138 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4139     "B", "Jail may lock (unlock) physical pages in memory");
4140 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4141     "B", "Jail may bind sockets to reserved ports");
4142 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4143     "B", "Jail may read the kernel message buffer");
4144 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4145     "B", "Unprivileged processes may use process debugging facilities");
4146 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4147     "B", "Processes in jail with uid 0 have privilege");
4148 
4149 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4150 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4151     "B", "Jail may mount/unmount jail-friendly file systems in general");
4152 
4153 /*
4154  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4155  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4156  * not created.
4157  */
4158 unsigned
4159 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4160     const char *descr)
4161 {
4162 	struct bool_flags *bf;
4163 	struct sysctl_oid *parent;
4164 	char *allow_name, *allow_noname, *allowed;
4165 #ifndef NO_SYSCTL_DESCR
4166 	char *descr_deprecated;
4167 #endif
4168 	u_int allow_flag;
4169 
4170 	if (prefix
4171 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4172 		< 0 ||
4173 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4174 		< 0
4175 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4176 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4177 		free(allow_name, M_PRISON);
4178 		return 0;
4179 	}
4180 
4181 	/*
4182 	 * See if this parameter has already beed added, i.e. a module was
4183 	 * previously loaded/unloaded.
4184 	 */
4185 	mtx_lock(&prison0.pr_mtx);
4186 	for (bf = pr_flag_allow;
4187 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4188 		atomic_load_int(&bf->flag) != 0;
4189 	     bf++) {
4190 		if (strcmp(bf->name, allow_name) == 0) {
4191 			allow_flag = bf->flag;
4192 			goto no_add;
4193 		}
4194 	}
4195 
4196 	/*
4197 	 * Find a free bit in pr_allow_all, failing if there are none
4198 	 * (which shouldn't happen as long as we keep track of how many
4199 	 * potential dynamic flags exist).
4200 	 */
4201 	for (allow_flag = 1;; allow_flag <<= 1) {
4202 		if (allow_flag == 0)
4203 			goto no_add;
4204 		if ((pr_allow_all & allow_flag) == 0)
4205 			break;
4206 	}
4207 
4208 	/* Note the parameter in the next open slot in pr_flag_allow. */
4209 	for (bf = pr_flag_allow; ; bf++) {
4210 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4211 			/* This should never happen, but is not fatal. */
4212 			allow_flag = 0;
4213 			goto no_add;
4214 		}
4215 		if (atomic_load_int(&bf->flag) == 0)
4216 			break;
4217 	}
4218 	bf->name = allow_name;
4219 	bf->noname = allow_noname;
4220 	pr_allow_all |= allow_flag;
4221 	/*
4222 	 * prison0 always has permission for the new parameter.
4223 	 * Other jails must have it granted to them.
4224 	 */
4225 	prison0.pr_allow |= allow_flag;
4226 	/* The flag indicates a valid entry, so make sure it is set last. */
4227 	atomic_store_rel_int(&bf->flag, allow_flag);
4228 	mtx_unlock(&prison0.pr_mtx);
4229 
4230 	/*
4231 	 * Create sysctls for the paramter, and the back-compat global
4232 	 * permission.
4233 	 */
4234 	parent = prefix
4235 	    ? SYSCTL_ADD_NODE(NULL,
4236 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4237 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4238 	    : &sysctl___security_jail_param_allow;
4239 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4240 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4241 	    NULL, 0, sysctl_jail_param, "B", descr);
4242 	if ((prefix
4243 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4244 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4245 #ifndef NO_SYSCTL_DESCR
4246 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4247 		    descr);
4248 #endif
4249 		(void)SYSCTL_ADD_PROC(NULL,
4250 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4251 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4252 		    sysctl_jail_default_allow, "I", descr_deprecated);
4253 #ifndef NO_SYSCTL_DESCR
4254 		free(descr_deprecated, M_TEMP);
4255 #endif
4256 		free(allowed, M_TEMP);
4257 	}
4258 	return allow_flag;
4259 
4260  no_add:
4261 	mtx_unlock(&prison0.pr_mtx);
4262 	free(allow_name, M_PRISON);
4263 	free(allow_noname, M_PRISON);
4264 	return allow_flag;
4265 }
4266 
4267 /*
4268  * The VFS system will register jail-aware filesystems here.  They each get
4269  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4270  * attempts to mount.
4271  */
4272 void
4273 prison_add_vfs(struct vfsconf *vfsp)
4274 {
4275 #ifdef NO_SYSCTL_DESCR
4276 
4277 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4278 	    NULL, NULL);
4279 #else
4280 	char *descr;
4281 
4282 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4283 	    vfsp->vfc_name);
4284 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4285 	    NULL, descr);
4286 	free(descr, M_TEMP);
4287 #endif
4288 }
4289 
4290 #ifdef RACCT
4291 void
4292 prison_racct_foreach(void (*callback)(struct racct *racct,
4293     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4294     void *arg2, void *arg3)
4295 {
4296 	struct prison_racct *prr;
4297 
4298 	ASSERT_RACCT_ENABLED();
4299 
4300 	sx_slock(&allprison_lock);
4301 	if (pre != NULL)
4302 		(pre)();
4303 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4304 		(callback)(prr->prr_racct, arg2, arg3);
4305 	if (post != NULL)
4306 		(post)();
4307 	sx_sunlock(&allprison_lock);
4308 }
4309 
4310 static struct prison_racct *
4311 prison_racct_find_locked(const char *name)
4312 {
4313 	struct prison_racct *prr;
4314 
4315 	ASSERT_RACCT_ENABLED();
4316 	sx_assert(&allprison_lock, SA_XLOCKED);
4317 
4318 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4319 		return (NULL);
4320 
4321 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4322 		if (strcmp(name, prr->prr_name) != 0)
4323 			continue;
4324 
4325 		/* Found prison_racct with a matching name? */
4326 		prison_racct_hold(prr);
4327 		return (prr);
4328 	}
4329 
4330 	/* Add new prison_racct. */
4331 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4332 	racct_create(&prr->prr_racct);
4333 
4334 	strcpy(prr->prr_name, name);
4335 	refcount_init(&prr->prr_refcount, 1);
4336 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4337 
4338 	return (prr);
4339 }
4340 
4341 struct prison_racct *
4342 prison_racct_find(const char *name)
4343 {
4344 	struct prison_racct *prr;
4345 
4346 	ASSERT_RACCT_ENABLED();
4347 
4348 	sx_xlock(&allprison_lock);
4349 	prr = prison_racct_find_locked(name);
4350 	sx_xunlock(&allprison_lock);
4351 	return (prr);
4352 }
4353 
4354 void
4355 prison_racct_hold(struct prison_racct *prr)
4356 {
4357 
4358 	ASSERT_RACCT_ENABLED();
4359 
4360 	refcount_acquire(&prr->prr_refcount);
4361 }
4362 
4363 static void
4364 prison_racct_free_locked(struct prison_racct *prr)
4365 {
4366 
4367 	ASSERT_RACCT_ENABLED();
4368 	sx_assert(&allprison_lock, SA_XLOCKED);
4369 
4370 	if (refcount_release(&prr->prr_refcount)) {
4371 		racct_destroy(&prr->prr_racct);
4372 		LIST_REMOVE(prr, prr_next);
4373 		free(prr, M_PRISON_RACCT);
4374 	}
4375 }
4376 
4377 void
4378 prison_racct_free(struct prison_racct *prr)
4379 {
4380 
4381 	ASSERT_RACCT_ENABLED();
4382 	sx_assert(&allprison_lock, SA_UNLOCKED);
4383 
4384 	if (refcount_release_if_not_last(&prr->prr_refcount))
4385 		return;
4386 
4387 	sx_xlock(&allprison_lock);
4388 	prison_racct_free_locked(prr);
4389 	sx_xunlock(&allprison_lock);
4390 }
4391 
4392 static void
4393 prison_racct_attach(struct prison *pr)
4394 {
4395 	struct prison_racct *prr;
4396 
4397 	ASSERT_RACCT_ENABLED();
4398 	sx_assert(&allprison_lock, SA_XLOCKED);
4399 
4400 	prr = prison_racct_find_locked(pr->pr_name);
4401 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4402 
4403 	pr->pr_prison_racct = prr;
4404 }
4405 
4406 /*
4407  * Handle jail renaming.  From the racct point of view, renaming means
4408  * moving from one prison_racct to another.
4409  */
4410 static void
4411 prison_racct_modify(struct prison *pr)
4412 {
4413 #ifdef RCTL
4414 	struct proc *p;
4415 	struct ucred *cred;
4416 #endif
4417 	struct prison_racct *oldprr;
4418 
4419 	ASSERT_RACCT_ENABLED();
4420 
4421 	sx_slock(&allproc_lock);
4422 	sx_xlock(&allprison_lock);
4423 
4424 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4425 		sx_xunlock(&allprison_lock);
4426 		sx_sunlock(&allproc_lock);
4427 		return;
4428 	}
4429 
4430 	oldprr = pr->pr_prison_racct;
4431 	pr->pr_prison_racct = NULL;
4432 
4433 	prison_racct_attach(pr);
4434 
4435 	/*
4436 	 * Move resource utilisation records.
4437 	 */
4438 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4439 
4440 #ifdef RCTL
4441 	/*
4442 	 * Force rctl to reattach rules to processes.
4443 	 */
4444 	FOREACH_PROC_IN_SYSTEM(p) {
4445 		PROC_LOCK(p);
4446 		cred = crhold(p->p_ucred);
4447 		PROC_UNLOCK(p);
4448 		rctl_proc_ucred_changed(p, cred);
4449 		crfree(cred);
4450 	}
4451 #endif
4452 
4453 	sx_sunlock(&allproc_lock);
4454 	prison_racct_free_locked(oldprr);
4455 	sx_xunlock(&allprison_lock);
4456 }
4457 
4458 static void
4459 prison_racct_detach(struct prison *pr)
4460 {
4461 
4462 	ASSERT_RACCT_ENABLED();
4463 	sx_assert(&allprison_lock, SA_UNLOCKED);
4464 
4465 	if (pr->pr_prison_racct == NULL)
4466 		return;
4467 	prison_racct_free(pr->pr_prison_racct);
4468 	pr->pr_prison_racct = NULL;
4469 }
4470 #endif /* RACCT */
4471 
4472 #ifdef DDB
4473 
4474 static void
4475 db_show_prison(struct prison *pr)
4476 {
4477 	struct bool_flags *bf;
4478 	struct jailsys_flags *jsf;
4479 #if defined(INET) || defined(INET6)
4480 	int ii;
4481 #endif
4482 	unsigned f;
4483 #ifdef INET
4484 	char ip4buf[INET_ADDRSTRLEN];
4485 #endif
4486 #ifdef INET6
4487 	char ip6buf[INET6_ADDRSTRLEN];
4488 #endif
4489 
4490 	db_printf("prison %p:\n", pr);
4491 	db_printf(" jid             = %d\n", pr->pr_id);
4492 	db_printf(" name            = %s\n", pr->pr_name);
4493 	db_printf(" parent          = %p\n", pr->pr_parent);
4494 	db_printf(" ref             = %d\n", pr->pr_ref);
4495 	db_printf(" uref            = %d\n", pr->pr_uref);
4496 	db_printf(" state           = %s\n",
4497 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4498 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
4499 	    "invalid");
4500 	db_printf(" path            = %s\n", pr->pr_path);
4501 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4502 	    ? pr->pr_cpuset->cs_id : -1);
4503 #ifdef VIMAGE
4504 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4505 #endif
4506 	db_printf(" root            = %p\n", pr->pr_root);
4507 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4508 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4509 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4510 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4511 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4512 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4513 	db_printf(" flags           = 0x%x", pr->pr_flags);
4514 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4515 		if (pr->pr_flags & bf->flag)
4516 			db_printf(" %s", bf->name);
4517 	for (jsf = pr_flag_jailsys;
4518 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4519 	     jsf++) {
4520 		f = pr->pr_flags & (jsf->disable | jsf->new);
4521 		db_printf(" %-16s= %s\n", jsf->name,
4522 		    (f != 0 && f == jsf->disable) ? "disable"
4523 		    : (f == jsf->new) ? "new"
4524 		    : "inherit");
4525 	}
4526 	db_printf(" allow           = 0x%x", pr->pr_allow);
4527 	for (bf = pr_flag_allow;
4528 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4529 		atomic_load_int(&bf->flag) != 0;
4530 	     bf++)
4531 		if (pr->pr_allow & bf->flag)
4532 			db_printf(" %s", bf->name);
4533 	db_printf("\n");
4534 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4535 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4536 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4537 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4538 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4539 #ifdef INET
4540 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4541 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4542 		db_printf(" %s %s\n",
4543 		    ii == 0 ? "ip4.addr        =" : "                 ",
4544 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4545 #endif
4546 #ifdef INET6
4547 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4548 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4549 		db_printf(" %s %s\n",
4550 		    ii == 0 ? "ip6.addr        =" : "                 ",
4551 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4552 #endif
4553 }
4554 
4555 DB_SHOW_COMMAND(prison, db_show_prison_command)
4556 {
4557 	struct prison *pr;
4558 
4559 	if (!have_addr) {
4560 		/*
4561 		 * Show all prisons in the list, and prison0 which is not
4562 		 * listed.
4563 		 */
4564 		db_show_prison(&prison0);
4565 		if (!db_pager_quit) {
4566 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4567 				db_show_prison(pr);
4568 				if (db_pager_quit)
4569 					break;
4570 			}
4571 		}
4572 		return;
4573 	}
4574 
4575 	if (addr == 0)
4576 		pr = &prison0;
4577 	else {
4578 		/* Look for a prison with the ID and with references. */
4579 		TAILQ_FOREACH(pr, &allprison, pr_list)
4580 			if (pr->pr_id == addr && pr->pr_ref > 0)
4581 				break;
4582 		if (pr == NULL)
4583 			/* Look again, without requiring a reference. */
4584 			TAILQ_FOREACH(pr, &allprison, pr_list)
4585 				if (pr->pr_id == addr)
4586 					break;
4587 		if (pr == NULL)
4588 			/* Assume address points to a valid prison. */
4589 			pr = (struct prison *)addr;
4590 	}
4591 	db_show_prison(pr);
4592 }
4593 
4594 #endif /* DDB */
4595