xref: /freebsd/sys/kern/kern_jail.c (revision 9d54812421274e490dc5f0fe4722ab8d35d9b258)
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/epoch.h>
49 #include <sys/taskqueue.h>
50 #include <sys/fcntl.h>
51 #include <sys/jail.h>
52 #include <sys/linker.h>
53 #include <sys/lock.h>
54 #include <sys/mman.h>
55 #include <sys/mutex.h>
56 #include <sys/racct.h>
57 #include <sys/rctl.h>
58 #include <sys/refcount.h>
59 #include <sys/sx.h>
60 #include <sys/sysent.h>
61 #include <sys/namei.h>
62 #include <sys/mount.h>
63 #include <sys/queue.h>
64 #include <sys/socket.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysctl.h>
67 #include <sys/uuid.h>
68 #include <sys/vnode.h>
69 
70 #include <net/if.h>
71 #include <net/vnet.h>
72 
73 #include <netinet/in.h>
74 
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #endif /* DDB */
78 
79 #include <security/mac/mac_framework.h>
80 
81 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
82 
83 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
84 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
85 
86 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
87 #ifdef INET
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
92 #endif
93 #else /* !INET */
94 #ifdef INET6
95 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
96 #else
97 #define	_PR_IP_SADDRSEL	0
98 #endif
99 #endif
100 
101 /* prison0 describes what is "real" about the system. */
102 struct prison prison0 = {
103 	.pr_id		= 0,
104 	.pr_name	= "0",
105 	.pr_ref		= 1,
106 	.pr_uref	= 1,
107 	.pr_path	= "/",
108 	.pr_securelevel	= -1,
109 	.pr_devfs_rsnum = 0,
110 	.pr_state	= PRISON_STATE_ALIVE,
111 	.pr_childmax	= JAIL_MAX,
112 	.pr_hostuuid	= DEFAULT_HOSTUUID,
113 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
114 #ifdef VIMAGE
115 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
116 #else
117 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
118 #endif
119 	.pr_allow	= PR_ALLOW_ALL_STATIC,
120 };
121 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
122 
123 struct bool_flags {
124 	const char	*name;
125 	const char	*noname;
126 	volatile u_int	 flag;
127 };
128 struct jailsys_flags {
129 	const char	*name;
130 	unsigned	 disable;
131 	unsigned	 new;
132 };
133 
134 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
135 struct	sx allprison_lock;
136 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
137 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
138 LIST_HEAD(, prison_racct) allprison_racct;
139 int	lastprid = 0;
140 
141 static int get_next_prid(struct prison **insprp);
142 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
143 static void prison_complete(void *context, int pending);
144 static void prison_deref(struct prison *pr, int flags);
145 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
146 static int prison_lock_xlock(struct prison *pr, int flags);
147 static void prison_cleanup(struct prison *pr);
148 static void prison_free_not_last(struct prison *pr);
149 static void prison_proc_free_not_last(struct prison *pr);
150 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
151     int enable);
152 static char *prison_path(struct prison *pr1, struct prison *pr2);
153 #ifdef RACCT
154 static void prison_racct_attach(struct prison *pr);
155 static void prison_racct_modify(struct prison *pr);
156 static void prison_racct_detach(struct prison *pr);
157 #endif
158 
159 /* Flags for prison_deref */
160 #define	PD_DEREF	0x01	/* Decrement pr_ref */
161 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
162 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
163 #define	PD_LOCKED	0x10	/* pr_mtx is held */
164 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
165 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
166 #define PD_OP_FLAGS	0x07	/* Operation flags */
167 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
168 
169 /*
170  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
171  * as we cannot figure out the size of a sparse array, or an array without a
172  * terminating entry.
173  */
174 static struct bool_flags pr_flag_bool[] = {
175 	{"persist", "nopersist", PR_PERSIST},
176 #ifdef INET
177 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
178 #endif
179 #ifdef INET6
180 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
181 #endif
182 };
183 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
184 
185 static struct jailsys_flags pr_flag_jailsys[] = {
186 	{"host", 0, PR_HOST},
187 #ifdef VIMAGE
188 	{"vnet", 0, PR_VNET},
189 #endif
190 #ifdef INET
191 	{"ip4", PR_IP4_USER, PR_IP4_USER},
192 #endif
193 #ifdef INET6
194 	{"ip6", PR_IP6_USER, PR_IP6_USER},
195 #endif
196 };
197 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
198 
199 /*
200  * Make this array full-size so dynamic parameters can be added.
201  * It is protected by prison0.mtx, but lockless reading is allowed
202  * with an atomic check of the flag values.
203  */
204 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
205 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
206 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
207 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
208 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
209 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
210 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
211 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
212 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
213 	{"allow.reserved_ports", "allow.noreserved_ports",
214 	 PR_ALLOW_RESERVED_PORTS},
215 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
216 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
217 	 PR_ALLOW_UNPRIV_DEBUG},
218 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
219 };
220 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
221 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
222 
223 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
224 					 PR_ALLOW_RESERVED_PORTS | \
225 					 PR_ALLOW_UNPRIV_DEBUG | \
226 					 PR_ALLOW_SUSER)
227 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
228 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
229 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
230 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
231 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
232 #if defined(INET) || defined(INET6)
233 static unsigned jail_max_af_ips = 255;
234 #endif
235 
236 /*
237  * Initialize the parts of prison0 that can't be static-initialized with
238  * constants.  This is called from proc0_init() after creating thread0 cpuset.
239  */
240 void
241 prison0_init(void)
242 {
243 	uint8_t *file, *data;
244 	size_t size;
245 	char buf[sizeof(prison0.pr_hostuuid)];
246 	bool valid;
247 
248 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
249 	prison0.pr_osreldate = osreldate;
250 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
251 
252 	/* If we have a preloaded hostuuid, use it. */
253 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
254 	if (file != NULL) {
255 		data = preload_fetch_addr(file);
256 		size = preload_fetch_size(file);
257 		if (data != NULL) {
258 			/*
259 			 * The preloaded data may include trailing whitespace, almost
260 			 * certainly a newline; skip over any whitespace or
261 			 * non-printable characters to be safe.
262 			 */
263 			while (size > 0 && data[size - 1] <= 0x20) {
264 				size--;
265 			}
266 
267 			valid = false;
268 
269 			/*
270 			 * Not NUL-terminated when passed from loader, but
271 			 * validate_uuid requires that due to using sscanf (as
272 			 * does the subsequent strlcpy, since it still reads
273 			 * past the given size to return the true length);
274 			 * bounce to a temporary buffer to fix.
275 			 */
276 			if (size >= sizeof(buf))
277 				goto done;
278 
279 			memcpy(buf, data, size);
280 			buf[size] = '\0';
281 
282 			if (validate_uuid(buf, size, NULL, 0) != 0)
283 				goto done;
284 
285 			valid = true;
286 			(void)strlcpy(prison0.pr_hostuuid, buf,
287 			    sizeof(prison0.pr_hostuuid));
288 
289 done:
290 			if (bootverbose && !valid) {
291 				printf("hostuuid: preload data malformed: '%.*s'\n",
292 				    (int)size, data);
293 			}
294 		}
295 	}
296 	if (bootverbose)
297 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
298 }
299 
300 /*
301  * struct jail_args {
302  *	struct jail *jail;
303  * };
304  */
305 int
306 sys_jail(struct thread *td, struct jail_args *uap)
307 {
308 	uint32_t version;
309 	int error;
310 	struct jail j;
311 
312 	error = copyin(uap->jail, &version, sizeof(uint32_t));
313 	if (error)
314 		return (error);
315 
316 	switch (version) {
317 	case 0:
318 	{
319 		struct jail_v0 j0;
320 
321 		/* FreeBSD single IPv4 jails. */
322 		bzero(&j, sizeof(struct jail));
323 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
324 		if (error)
325 			return (error);
326 		j.version = j0.version;
327 		j.path = j0.path;
328 		j.hostname = j0.hostname;
329 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
330 		break;
331 	}
332 
333 	case 1:
334 		/*
335 		 * Version 1 was used by multi-IPv4 jail implementations
336 		 * that never made it into the official kernel.
337 		 */
338 		return (EINVAL);
339 
340 	case 2:	/* JAIL_API_VERSION */
341 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
342 		error = copyin(uap->jail, &j, sizeof(struct jail));
343 		if (error)
344 			return (error);
345 		break;
346 
347 	default:
348 		/* Sci-Fi jails are not supported, sorry. */
349 		return (EINVAL);
350 	}
351 	return (kern_jail(td, &j));
352 }
353 
354 int
355 kern_jail(struct thread *td, struct jail *j)
356 {
357 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
358 #ifdef INET
359 			    + 1
360 #endif
361 #ifdef INET6
362 			    + 1
363 #endif
364 			    )];
365 	struct uio opt;
366 	char *u_path, *u_hostname, *u_name;
367 	struct bool_flags *bf;
368 #ifdef INET
369 	uint32_t ip4s;
370 	struct in_addr *u_ip4;
371 #endif
372 #ifdef INET6
373 	struct in6_addr *u_ip6;
374 #endif
375 	size_t tmplen;
376 	int error, enforce_statfs;
377 
378 	bzero(&optiov, sizeof(optiov));
379 	opt.uio_iov = optiov;
380 	opt.uio_iovcnt = 0;
381 	opt.uio_offset = -1;
382 	opt.uio_resid = -1;
383 	opt.uio_segflg = UIO_SYSSPACE;
384 	opt.uio_rw = UIO_READ;
385 	opt.uio_td = td;
386 
387 	/* Set permissions for top-level jails from sysctls. */
388 	if (!jailed(td->td_ucred)) {
389 		for (bf = pr_flag_allow;
390 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
391 			atomic_load_int(&bf->flag) != 0;
392 		     bf++) {
393 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
394 			    (jail_default_allow & bf->flag)
395 			    ? bf->name : bf->noname);
396 			optiov[opt.uio_iovcnt].iov_len =
397 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
398 			opt.uio_iovcnt += 2;
399 		}
400 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
401 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
402 		opt.uio_iovcnt++;
403 		enforce_statfs = jail_default_enforce_statfs;
404 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
405 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
406 		opt.uio_iovcnt++;
407 	}
408 
409 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
410 #ifdef INET
411 	ip4s = (j->version == 0) ? 1 : j->ip4s;
412 	if (ip4s > jail_max_af_ips)
413 		return (EINVAL);
414 	tmplen += ip4s * sizeof(struct in_addr);
415 #else
416 	if (j->ip4s > 0)
417 		return (EINVAL);
418 #endif
419 #ifdef INET6
420 	if (j->ip6s > jail_max_af_ips)
421 		return (EINVAL);
422 	tmplen += j->ip6s * sizeof(struct in6_addr);
423 #else
424 	if (j->ip6s > 0)
425 		return (EINVAL);
426 #endif
427 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
428 	u_hostname = u_path + MAXPATHLEN;
429 	u_name = u_hostname + MAXHOSTNAMELEN;
430 #ifdef INET
431 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
432 #endif
433 #ifdef INET6
434 #ifdef INET
435 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
436 #else
437 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
438 #endif
439 #endif
440 	optiov[opt.uio_iovcnt].iov_base = "path";
441 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
442 	opt.uio_iovcnt++;
443 	optiov[opt.uio_iovcnt].iov_base = u_path;
444 	error = copyinstr(j->path, u_path, MAXPATHLEN,
445 	    &optiov[opt.uio_iovcnt].iov_len);
446 	if (error) {
447 		free(u_path, M_TEMP);
448 		return (error);
449 	}
450 	opt.uio_iovcnt++;
451 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
452 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
453 	opt.uio_iovcnt++;
454 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
455 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
456 	    &optiov[opt.uio_iovcnt].iov_len);
457 	if (error) {
458 		free(u_path, M_TEMP);
459 		return (error);
460 	}
461 	opt.uio_iovcnt++;
462 	if (j->jailname != NULL) {
463 		optiov[opt.uio_iovcnt].iov_base = "name";
464 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
465 		opt.uio_iovcnt++;
466 		optiov[opt.uio_iovcnt].iov_base = u_name;
467 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
468 		    &optiov[opt.uio_iovcnt].iov_len);
469 		if (error) {
470 			free(u_path, M_TEMP);
471 			return (error);
472 		}
473 		opt.uio_iovcnt++;
474 	}
475 #ifdef INET
476 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
477 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
478 	opt.uio_iovcnt++;
479 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
480 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
481 	if (j->version == 0)
482 		u_ip4->s_addr = j->ip4s;
483 	else {
484 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
485 		if (error) {
486 			free(u_path, M_TEMP);
487 			return (error);
488 		}
489 	}
490 	opt.uio_iovcnt++;
491 #endif
492 #ifdef INET6
493 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
494 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
495 	opt.uio_iovcnt++;
496 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
497 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
498 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
499 	if (error) {
500 		free(u_path, M_TEMP);
501 		return (error);
502 	}
503 	opt.uio_iovcnt++;
504 #endif
505 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
506 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
507 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
508 	free(u_path, M_TEMP);
509 	return (error);
510 }
511 
512 /*
513  * struct jail_set_args {
514  *	struct iovec *iovp;
515  *	unsigned int iovcnt;
516  *	int flags;
517  * };
518  */
519 int
520 sys_jail_set(struct thread *td, struct jail_set_args *uap)
521 {
522 	struct uio *auio;
523 	int error;
524 
525 	/* Check that we have an even number of iovecs. */
526 	if (uap->iovcnt & 1)
527 		return (EINVAL);
528 
529 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
530 	if (error)
531 		return (error);
532 	error = kern_jail_set(td, auio, uap->flags);
533 	free(auio, M_IOV);
534 	return (error);
535 }
536 
537 #if defined(INET) || defined(INET6)
538 typedef int prison_addr_cmp_t(const void *, const void *);
539 typedef bool prison_addr_valid_t(const void *);
540 static const struct pr_family {
541 	size_t			size;
542 	prison_addr_cmp_t	*cmp;
543 	prison_addr_valid_t	*valid;
544 	int			ip_flag;
545 } pr_families[PR_FAMILY_MAX] = {
546 #ifdef INET
547 	[PR_INET] = {
548 		.size = sizeof(struct in_addr),
549 		.cmp = prison_qcmp_v4,
550 		.valid = prison_valid_v4,
551 		.ip_flag = PR_IP4_USER,
552 	 },
553 #endif
554 #ifdef INET6
555 	[PR_INET6] = {
556 		.size = sizeof(struct in6_addr),
557 		.cmp = prison_qcmp_v6,
558 		.valid = prison_valid_v6,
559 		.ip_flag = PR_IP6_USER,
560 	},
561 #endif
562 };
563 
564 /*
565  * Network address lists (pr_addrs) allocation for jails.  The addresses
566  * are accessed locklessly by the network stack, thus need to be protected by
567  * the network epoch.
568  */
569 struct prison_ip {
570 	struct epoch_context ctx;
571 	uint32_t	ips;
572 #ifdef FUTURE_C
573 	union {
574 		struct in_addr pr_ip4[];
575 		struct in6_addr pr_ip6[];
576 	};
577 #else /* No future C :( */
578 #define	PR_IP(pip, i)	((const char *)((pip) + 1) + pr_families[af].size * (i))
579 #define	PR_IPD(pip, i)	((char *)((pip) + 1) + pr_families[af].size * (i))
580 #endif
581 };
582 
583 static struct prison_ip *
584 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
585 {
586 	struct prison_ip *pip;
587 
588 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
589 	    M_PRISON, flags);
590 	if (pip != NULL)
591 		pip->ips = cnt;
592 	return (pip);
593 }
594 
595 /*
596  * Allocate and copyin user supplied address list, sorting and validating.
597  * kern_jail_set() helper.
598  */
599 static struct prison_ip *
600 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
601 {
602 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
603 	const size_t size = pr_families[af].size;
604 	struct prison_ip *pip;
605 
606 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
607 	bcopy(op, pip + 1, cnt * size);
608 	/*
609 	 * IP addresses are all sorted but ip[0] to preserve
610 	 * the primary IP address as given from userland.
611 	 * This special IP is used for unbound outgoing
612 	 * connections as well for "loopback" traffic in case
613 	 * source address selection cannot find any more fitting
614 	 * address to connect from.
615 	 */
616 	if (cnt > 1)
617 		qsort((char *)(pip + 1) + size, cnt - 1, size,
618 		    pr_families[af].cmp);
619 	/*
620 	 * Check for duplicate addresses and do some simple
621 	 * zero and broadcast checks. If users give other bogus
622 	 * addresses it is their problem.
623 	 */
624 	for (int i = 0; i < cnt; i++) {
625 		if (!pr_families[af].valid(PR_IP(pip, i))) {
626 			free(pip, M_PRISON);
627 			return (NULL);
628 		}
629 		if (i + 1 < cnt &&
630 		    (cmp(PR_IP(pip, 0), PR_IP(pip, i + 1)) == 0 ||
631 		     cmp(PR_IP(pip, i), PR_IP(pip, i + 1)) == 0)) {
632 			free(pip, M_PRISON);
633 			return (NULL);
634 		}
635 	}
636 
637 	return (pip);
638 }
639 
640 /*
641  * Allocate and dup parent prison address list.
642  * kern_jail_set() helper.
643  */
644 static void
645 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
646 {
647 
648 	if (ppr->pr_addrs[af] != NULL) {
649 		pr->pr_addrs[af] = prison_ip_alloc(af,
650 		    ppr->pr_addrs[af]->ips, M_WAITOK);
651 		bcopy(ppr->pr_addrs[af], pr->pr_addrs[af],
652 		    pr->pr_addrs[af]->ips * pr_families[af].size);
653 	}
654 }
655 
656 /*
657  * Make sure the new set of IP addresses is a subset of the parent's list.
658  * Don't worry about the parent being unlocked, as any setting is done with
659  * allprison_lock held.
660  * kern_jail_set() helper.
661  */
662 static bool
663 prison_ip_parent_match(const struct prison_ip *ppip,
664     const struct prison_ip *pip, const pr_family_t af)
665 {
666 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
667 	int i, j;
668 
669 	if (ppip == NULL)
670 		return (false);
671 
672 	for (i = 0; i < ppip->ips; i++)
673 		if (cmp(PR_IP(pip, 0), PR_IP(ppip, i)) == 0)
674 			break;
675 
676 	if (i == ppip->ips)
677 		/* Main address not present in parent. */
678 		return (false);
679 
680 	if (pip->ips > 1) {
681 		for (i = j = 1; i < pip->ips; i++) {
682 			if (cmp(PR_IP(pip, i), PR_IP(ppip, 0)) == 0)
683 				/* Equals to parent primary address. */
684 				continue;
685 			for (; j < ppip->ips; j++)
686 				if (cmp(PR_IP(pip, i), PR_IP(ppip, j)) == 0)
687 					break;
688 			if (j == ppip->ips)
689 				break;
690 		}
691 		if (j == ppip->ips)
692 			/* Address not present in parent. */
693 			return (false);
694 	}
695 	return (true);
696 }
697 
698 /*
699  * Check for conflicting IP addresses.  We permit them if there is no more
700  * than one IP on each jail.  If there is a duplicate on a jail with more
701  * than one IP stop checking and return error.
702  * kern_jail_set() helper.
703  */
704 static bool
705 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
706     const struct prison_ip *pip, pr_family_t af)
707 {
708 	const struct prison *tppr, *tpr;
709 	int descend;
710 
711 #ifdef VIMAGE
712 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
713 		if (tppr->pr_flags & PR_VNET)
714 			break;
715 #else
716 	tppr = &prison0;
717 #endif
718 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
719 		if (tpr == pr ||
720 #ifdef VIMAGE
721 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
722 #endif
723 		    !prison_isalive(tpr)) {
724 			descend = 0;
725 			continue;
726 		}
727 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
728 			continue;
729 		descend = 0;
730 		if (tpr->pr_addrs[af] == NULL ||
731 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
732 			continue;
733 		for (int i = 0; i < pip->ips; i++)
734 			if (prison_ip_check(tpr, af, PR_IP(pip, i)) == 0)
735 				return (false);
736 	}
737 
738 	return (true);
739 }
740 
741 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
742     "prison must start with epoch context");
743 static void
744 prison_ip_free_deferred(epoch_context_t ctx)
745 {
746 
747 	free(ctx, M_PRISON);
748 }
749 
750 static void
751 prison_ip_free(struct prison_ip *pip)
752 {
753 
754 	if (pip != NULL)
755 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
756 }
757 
758 static void
759 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
760 {
761 	struct prison_ip **mem, *old;
762 
763 	mtx_assert(&pr->pr_mtx, MA_OWNED);
764 
765 	mem = &pr->pr_addrs[af];
766 
767 	old = *mem;
768 	ck_pr_store_ptr(mem, new);
769 	prison_ip_free(old);
770 }
771 
772 /*
773  * Restrict a prison's IP address list with its parent's, possibly replacing
774  * it.  Return true if the replacement buffer was used (or would have been).
775  * kern_jail_set() helper.
776  */
777 static bool
778 prison_ip_restrict(struct prison *pr, const pr_family_t af,
779     struct prison_ip *new)
780 {
781 	const struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
782 	const struct prison_ip *pip = pr->pr_addrs[af];
783 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
784 	const size_t size = pr_families[af].size;
785 	uint32_t ips;
786 	bool alloced;
787 
788 	mtx_assert(&pr->pr_mtx, MA_OWNED);
789 
790 	/*
791 	 * Due to epoch-synchronized access to the IP address lists we always
792 	 * allocate a new list even if the old one has enough space.  We could
793 	 * atomically update an IPv4 address inside a list, but that would
794 	 * screw up sorting, and in case of IPv6 we can't even atomically write
795 	 * one.
796 	 */
797 	ips = (pr->pr_flags & pr_families[af].ip_flag) ? pip->ips : ppip->ips;
798 	if (ips == 0) {
799 		prison_ip_set(pr, af, NULL);
800 		return (false);
801 	}
802 	if (new == NULL) {
803 		new = prison_ip_alloc(af, ips, M_NOWAIT);
804 		if (new == NULL)
805 			return (true);
806 		alloced = true;
807 	} else
808 		alloced = false;
809 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
810 		/* This has no user settings, so just copy the parent's list. */
811 		bcopy(ppip, new, ips * size);
812 	} else {
813 		/* Remove addresses that aren't in the parent. */
814 		int i;
815 
816 		i = 0; /* index in pip */
817 		ips = 0; /* index in new */
818 
819 		for (int pi = 0; pi < ppip->ips; pi++)
820 			if (cmp(PR_IP(pip, 0), PR_IP(ppip, pi)) == 0) {
821 				/* Found our primary address in parent. */
822 				bcopy(PR_IP(pip, i), PR_IPD(new, ips), size);
823 				i++;
824 				ips++;
825 				break;
826 			}
827 		for (int pi = 1; i < pip->ips; ) {
828 			/* Check against primary, which is unsorted. */
829 			if (cmp(PR_IP(pip, i), PR_IP(ppip, 0)) == 0) {
830 				/* Matches parent's primary address. */
831 				bcopy(PR_IP(pip, i), PR_IPD(new, ips), size);
832 				i++;
833 				ips++;
834 				continue;
835 			}
836 			/* The rest are sorted. */
837 			switch (pi >= ppip->ips ? -1 :
838 				cmp(PR_IP(pip, i), PR_IP(ppip, pi))) {
839 			case -1:
840 				i++;
841 				break;
842 			case 0:
843 				bcopy(PR_IP(pr, i), PR_IPD(new, ips), size);
844 				i++;
845 				pi++;
846 				ips++;
847 				break;
848 			case 1:
849 				pi++;
850 				break;
851 			}
852 		}
853 		if (ips == 0) {
854 			if (alloced)
855 				prison_ip_free(new);
856 			new = NULL;
857 		}
858 	}
859 	prison_ip_set(pr, af, new);
860 	return (new != NULL ? true : false);
861 }
862 
863 /*
864  * Fast-path check if an address belongs to a prison.
865  */
866 int
867 prison_ip_check(const struct prison *pr, const pr_family_t af,
868     const void *addr)
869 {
870 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
871 	const struct prison_ip *pip;
872 	int i, a, z, d;
873 
874 	MPASS(mtx_owned(&pr->pr_mtx) ||
875 	    in_epoch(net_epoch_preempt) ||
876 	    sx_xlocked(&allprison_lock));
877 
878 	pip = ck_pr_load_ptr(&pr->pr_addrs[af]);
879 	if (__predict_false(pip == NULL))
880 		return (EAFNOSUPPORT);
881 
882 	/* Check the primary IP. */
883 	if (cmp(PR_IP(pip, 0), addr) == 0)
884 		return (0);
885 
886 	/*
887 	 * All the other IPs are sorted so we can do a binary search.
888 	 */
889 	a = 0;
890 	z = pip->ips - 2;
891 	while (a <= z) {
892 		i = (a + z) / 2;
893 		d = cmp(PR_IP(pip, i + 1), addr);
894 		if (d > 0)
895 			z = i - 1;
896 		else if (d < 0)
897 			a = i + 1;
898 		else
899 			return (0);
900 	}
901 
902 	return (EADDRNOTAVAIL);
903 }
904 
905 /*
906  * Grab primary IP.  Historically required mutex, but nothing prevents
907  * us to support epoch-protected access.  Is it used in fast path?
908  * in{6}_jail.c helper
909  */
910 const void *
911 prison_ip_get0(const struct prison *pr, const pr_family_t af)
912 {
913 	const struct prison_ip *pip = pr->pr_addrs[af];
914 
915 	mtx_assert(&pr->pr_mtx, MA_OWNED);
916 	MPASS(pip);
917 
918 	return (pip + 1);
919 }
920 
921 u_int
922 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
923 {
924 
925 	return (pr->pr_addrs[af]->ips);
926 }
927 #endif	/* defined(INET) || defined(INET6) */
928 
929 int
930 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
931 {
932 	struct nameidata nd;
933 #ifdef INET
934 	struct prison_ip *ip4;
935 #endif
936 #ifdef INET6
937 	struct prison_ip *ip6;
938 #endif
939 	struct vfsopt *opt;
940 	struct vfsoptlist *opts;
941 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
942 	struct vnode *root;
943 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
944 	char *g_path, *osrelstr;
945 	struct bool_flags *bf;
946 	struct jailsys_flags *jsf;
947 #if defined(INET) || defined(INET6)
948 	void *op;
949 #endif
950 	unsigned long hid;
951 	size_t namelen, onamelen, pnamelen;
952 	int born, created, cuflags, descend, drflags, enforce;
953 	int error, errmsg_len, errmsg_pos;
954 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
955 	int jid, jsys, len, level;
956 	int childmax, osreldt, rsnum, slevel;
957 #ifdef INET
958 	int ip4s, redo_ip4;
959 #endif
960 #ifdef INET6
961 	int ip6s, redo_ip6;
962 #endif
963 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
964 	uint64_t pr_allow_diff;
965 	unsigned tallow;
966 	char numbuf[12];
967 
968 	error = priv_check(td, PRIV_JAIL_SET);
969 	if (!error && (flags & JAIL_ATTACH))
970 		error = priv_check(td, PRIV_JAIL_ATTACH);
971 	if (error)
972 		return (error);
973 	mypr = td->td_ucred->cr_prison;
974 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
975 		return (EPERM);
976 	if (flags & ~JAIL_SET_MASK)
977 		return (EINVAL);
978 
979 	/*
980 	 * Check all the parameters before committing to anything.  Not all
981 	 * errors can be caught early, but we may as well try.  Also, this
982 	 * takes care of some expensive stuff (path lookup) before getting
983 	 * the allprison lock.
984 	 *
985 	 * XXX Jails are not filesystems, and jail parameters are not mount
986 	 *     options.  But it makes more sense to re-use the vfsopt code
987 	 *     than duplicate it under a different name.
988 	 */
989 	error = vfs_buildopts(optuio, &opts);
990 	if (error)
991 		return (error);
992 #ifdef INET
993 	ip4 = NULL;
994 #endif
995 #ifdef INET6
996 	ip6 = NULL;
997 #endif
998 	g_path = NULL;
999 
1000 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1001 	if (!cuflags) {
1002 		error = EINVAL;
1003 		vfs_opterror(opts, "no valid operation (create or update)");
1004 		goto done_errmsg;
1005 	}
1006 
1007 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1008 	if (error == ENOENT)
1009 		jid = 0;
1010 	else if (error != 0)
1011 		goto done_free;
1012 
1013 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1014 	if (error == ENOENT)
1015 		gotslevel = 0;
1016 	else if (error != 0)
1017 		goto done_free;
1018 	else
1019 		gotslevel = 1;
1020 
1021 	error =
1022 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1023 	if (error == ENOENT)
1024 		gotchildmax = 0;
1025 	else if (error != 0)
1026 		goto done_free;
1027 	else
1028 		gotchildmax = 1;
1029 
1030 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1031 	if (error == ENOENT)
1032 		gotenforce = 0;
1033 	else if (error != 0)
1034 		goto done_free;
1035 	else if (enforce < 0 || enforce > 2) {
1036 		error = EINVAL;
1037 		goto done_free;
1038 	} else
1039 		gotenforce = 1;
1040 
1041 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1042 	if (error == ENOENT)
1043 		gotrsnum = 0;
1044 	else if (error != 0)
1045 		goto done_free;
1046 	else
1047 		gotrsnum = 1;
1048 
1049 	pr_flags = ch_flags = 0;
1050 	for (bf = pr_flag_bool;
1051 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1052 	     bf++) {
1053 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1054 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1055 	}
1056 	ch_flags |= pr_flags;
1057 	for (jsf = pr_flag_jailsys;
1058 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1059 	     jsf++) {
1060 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1061 		if (error == ENOENT)
1062 			continue;
1063 		if (error != 0)
1064 			goto done_free;
1065 		switch (jsys) {
1066 		case JAIL_SYS_DISABLE:
1067 			if (!jsf->disable) {
1068 				error = EINVAL;
1069 				goto done_free;
1070 			}
1071 			pr_flags |= jsf->disable;
1072 			break;
1073 		case JAIL_SYS_NEW:
1074 			pr_flags |= jsf->new;
1075 			break;
1076 		case JAIL_SYS_INHERIT:
1077 			break;
1078 		default:
1079 			error = EINVAL;
1080 			goto done_free;
1081 		}
1082 		ch_flags |= jsf->new | jsf->disable;
1083 	}
1084 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1085 	    && !(pr_flags & PR_PERSIST)) {
1086 		error = EINVAL;
1087 		vfs_opterror(opts, "new jail must persist or attach");
1088 		goto done_errmsg;
1089 	}
1090 #ifdef VIMAGE
1091 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1092 		error = EINVAL;
1093 		vfs_opterror(opts, "vnet cannot be changed after creation");
1094 		goto done_errmsg;
1095 	}
1096 #endif
1097 #ifdef INET
1098 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1099 		error = EINVAL;
1100 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1101 		goto done_errmsg;
1102 	}
1103 #endif
1104 #ifdef INET6
1105 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1106 		error = EINVAL;
1107 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1108 		goto done_errmsg;
1109 	}
1110 #endif
1111 
1112 	pr_allow = ch_allow = 0;
1113 	for (bf = pr_flag_allow;
1114 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1115 		atomic_load_int(&bf->flag) != 0;
1116 	     bf++) {
1117 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1118 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1119 	}
1120 	ch_allow |= pr_allow;
1121 
1122 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1123 	if (error == ENOENT)
1124 		name = NULL;
1125 	else if (error != 0)
1126 		goto done_free;
1127 	else {
1128 		if (len == 0 || name[len - 1] != '\0') {
1129 			error = EINVAL;
1130 			goto done_free;
1131 		}
1132 		if (len > MAXHOSTNAMELEN) {
1133 			error = ENAMETOOLONG;
1134 			goto done_free;
1135 		}
1136 	}
1137 
1138 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1139 	if (error == ENOENT)
1140 		host = NULL;
1141 	else if (error != 0)
1142 		goto done_free;
1143 	else {
1144 		ch_flags |= PR_HOST;
1145 		pr_flags |= PR_HOST;
1146 		if (len == 0 || host[len - 1] != '\0') {
1147 			error = EINVAL;
1148 			goto done_free;
1149 		}
1150 		if (len > MAXHOSTNAMELEN) {
1151 			error = ENAMETOOLONG;
1152 			goto done_free;
1153 		}
1154 	}
1155 
1156 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1157 	if (error == ENOENT)
1158 		domain = NULL;
1159 	else if (error != 0)
1160 		goto done_free;
1161 	else {
1162 		ch_flags |= PR_HOST;
1163 		pr_flags |= PR_HOST;
1164 		if (len == 0 || domain[len - 1] != '\0') {
1165 			error = EINVAL;
1166 			goto done_free;
1167 		}
1168 		if (len > MAXHOSTNAMELEN) {
1169 			error = ENAMETOOLONG;
1170 			goto done_free;
1171 		}
1172 	}
1173 
1174 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1175 	if (error == ENOENT)
1176 		uuid = NULL;
1177 	else if (error != 0)
1178 		goto done_free;
1179 	else {
1180 		ch_flags |= PR_HOST;
1181 		pr_flags |= PR_HOST;
1182 		if (len == 0 || uuid[len - 1] != '\0') {
1183 			error = EINVAL;
1184 			goto done_free;
1185 		}
1186 		if (len > HOSTUUIDLEN) {
1187 			error = ENAMETOOLONG;
1188 			goto done_free;
1189 		}
1190 	}
1191 
1192 #ifdef COMPAT_FREEBSD32
1193 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1194 		uint32_t hid32;
1195 
1196 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1197 		hid = hid32;
1198 	} else
1199 #endif
1200 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1201 	if (error == ENOENT)
1202 		gothid = 0;
1203 	else if (error != 0)
1204 		goto done_free;
1205 	else {
1206 		gothid = 1;
1207 		ch_flags |= PR_HOST;
1208 		pr_flags |= PR_HOST;
1209 	}
1210 
1211 #ifdef INET
1212 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1213 	if (error == ENOENT)
1214 		ip4s = 0;
1215 	else if (error != 0)
1216 		goto done_free;
1217 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1218 		error = EINVAL;
1219 		goto done_free;
1220 	} else {
1221 		ch_flags |= PR_IP4_USER;
1222 		pr_flags |= PR_IP4_USER;
1223 		if (ip4s > 0) {
1224 			ip4s /= sizeof(struct in_addr);
1225 			if (ip4s > jail_max_af_ips) {
1226 				error = EINVAL;
1227 				vfs_opterror(opts, "too many IPv4 addresses");
1228 				goto done_errmsg;
1229 			}
1230 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1231 			if (ip4 == NULL) {
1232 				error = EINVAL;
1233 				goto done_free;
1234 			}
1235 		}
1236 	}
1237 #endif
1238 
1239 #ifdef INET6
1240 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1241 	if (error == ENOENT)
1242 		ip6s = 0;
1243 	else if (error != 0)
1244 		goto done_free;
1245 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1246 		error = EINVAL;
1247 		goto done_free;
1248 	} else {
1249 		ch_flags |= PR_IP6_USER;
1250 		pr_flags |= PR_IP6_USER;
1251 		if (ip6s > 0) {
1252 			ip6s /= sizeof(struct in6_addr);
1253 			if (ip6s > jail_max_af_ips) {
1254 				error = EINVAL;
1255 				vfs_opterror(opts, "too many IPv6 addresses");
1256 				goto done_errmsg;
1257 			}
1258 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1259 			if (ip6 == NULL) {
1260 				error = EINVAL;
1261 				goto done_free;
1262 			}
1263 		}
1264 	}
1265 #endif
1266 
1267 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1268 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1269 		error = EINVAL;
1270 		vfs_opterror(opts,
1271 		    "vnet jails cannot have IP address restrictions");
1272 		goto done_errmsg;
1273 	}
1274 #endif
1275 
1276 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1277 	if (error == ENOENT)
1278 		osrelstr = NULL;
1279 	else if (error != 0)
1280 		goto done_free;
1281 	else {
1282 		if (flags & JAIL_UPDATE) {
1283 			error = EINVAL;
1284 			vfs_opterror(opts,
1285 			    "osrelease cannot be changed after creation");
1286 			goto done_errmsg;
1287 		}
1288 		if (len == 0 || osrelstr[len - 1] != '\0') {
1289 			error = EINVAL;
1290 			goto done_free;
1291 		}
1292 		if (len >= OSRELEASELEN) {
1293 			error = ENAMETOOLONG;
1294 			vfs_opterror(opts,
1295 			    "osrelease string must be 1-%d bytes long",
1296 			    OSRELEASELEN - 1);
1297 			goto done_errmsg;
1298 		}
1299 	}
1300 
1301 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1302 	if (error == ENOENT)
1303 		osreldt = 0;
1304 	else if (error != 0)
1305 		goto done_free;
1306 	else {
1307 		if (flags & JAIL_UPDATE) {
1308 			error = EINVAL;
1309 			vfs_opterror(opts,
1310 			    "osreldate cannot be changed after creation");
1311 			goto done_errmsg;
1312 		}
1313 		if (osreldt == 0) {
1314 			error = EINVAL;
1315 			vfs_opterror(opts, "osreldate cannot be 0");
1316 			goto done_errmsg;
1317 		}
1318 	}
1319 
1320 	root = NULL;
1321 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1322 	if (error == ENOENT)
1323 		path = NULL;
1324 	else if (error != 0)
1325 		goto done_free;
1326 	else {
1327 		if (flags & JAIL_UPDATE) {
1328 			error = EINVAL;
1329 			vfs_opterror(opts,
1330 			    "path cannot be changed after creation");
1331 			goto done_errmsg;
1332 		}
1333 		if (len == 0 || path[len - 1] != '\0') {
1334 			error = EINVAL;
1335 			goto done_free;
1336 		}
1337 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1338 		error = namei(&nd);
1339 		if (error)
1340 			goto done_free;
1341 		root = nd.ni_vp;
1342 		NDFREE_PNBUF(&nd);
1343 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1344 		strlcpy(g_path, path, MAXPATHLEN);
1345 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1346 		if (error == 0) {
1347 			path = g_path;
1348 		} else {
1349 			/* exit on other errors */
1350 			goto done_free;
1351 		}
1352 		if (root->v_type != VDIR) {
1353 			error = ENOTDIR;
1354 			vput(root);
1355 			goto done_free;
1356 		}
1357 		VOP_UNLOCK(root);
1358 	}
1359 
1360 	/*
1361 	 * Find the specified jail, or at least its parent.
1362 	 * This abuses the file error codes ENOENT and EEXIST.
1363 	 */
1364 	pr = NULL;
1365 	inspr = NULL;
1366 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1367 		namelc = strrchr(name, '.');
1368 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1369 		if (*p != '\0')
1370 			jid = 0;
1371 	}
1372 	sx_xlock(&allprison_lock);
1373 	drflags = PD_LIST_XLOCKED;
1374 	ppr = mypr;
1375 	if (!prison_isalive(ppr)) {
1376 		/* This jail is dying.  This process will surely follow. */
1377 		error = EAGAIN;
1378 		goto done_deref;
1379 	}
1380 	if (jid != 0) {
1381 		if (jid < 0) {
1382 			error = EINVAL;
1383 			vfs_opterror(opts, "negative jid");
1384 			goto done_deref;
1385 		}
1386 		/*
1387 		 * See if a requested jid already exists.  Keep track of
1388 		 * where it can be inserted later.
1389 		 */
1390 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1391 			if (inspr->pr_id < jid)
1392 				continue;
1393 			if (inspr->pr_id > jid)
1394 				break;
1395 			pr = inspr;
1396 			mtx_lock(&pr->pr_mtx);
1397 			drflags |= PD_LOCKED;
1398 			inspr = NULL;
1399 			break;
1400 		}
1401 		if (pr != NULL) {
1402 			/* Create: jid must not exist. */
1403 			if (cuflags == JAIL_CREATE) {
1404 				/*
1405 				 * Even creators that cannot see the jail will
1406 				 * get EEXIST.
1407 				 */
1408 				error = EEXIST;
1409 				vfs_opterror(opts, "jail %d already exists",
1410 				    jid);
1411 				goto done_deref;
1412 			}
1413 			if (!prison_ischild(mypr, pr)) {
1414 				/*
1415 				 * Updaters get ENOENT if they cannot see the
1416 				 * jail.  This is true even for CREATE | UPDATE,
1417 				 * which normally cannot give this error.
1418 				 */
1419 				error = ENOENT;
1420 				vfs_opterror(opts, "jail %d not found", jid);
1421 				goto done_deref;
1422 			}
1423 			ppr = pr->pr_parent;
1424 			if (!prison_isalive(ppr)) {
1425 				error = ENOENT;
1426 				vfs_opterror(opts, "jail %d is dying",
1427 				    ppr->pr_id);
1428 				goto done_deref;
1429 			}
1430 			if (!prison_isalive(pr)) {
1431 				if (!(flags & JAIL_DYING)) {
1432 					error = ENOENT;
1433 					vfs_opterror(opts, "jail %d is dying",
1434 					    jid);
1435 					goto done_deref;
1436 				}
1437 				if ((flags & JAIL_ATTACH) ||
1438 				    (pr_flags & PR_PERSIST)) {
1439 					/*
1440 					 * A dying jail might be resurrected
1441 					 * (via attach or persist), but first
1442 					 * it must determine if another jail
1443 					 * has claimed its name.  Accomplish
1444 					 * this by implicitly re-setting the
1445 					 * name.
1446 					 */
1447 					if (name == NULL)
1448 						name = prison_name(mypr, pr);
1449 				}
1450 			}
1451 		} else {
1452 			/* Update: jid must exist. */
1453 			if (cuflags == JAIL_UPDATE) {
1454 				error = ENOENT;
1455 				vfs_opterror(opts, "jail %d not found", jid);
1456 				goto done_deref;
1457 			}
1458 		}
1459 	}
1460 	/*
1461 	 * If the caller provided a name, look for a jail by that name.
1462 	 * This has different semantics for creates and updates keyed by jid
1463 	 * (where the name must not already exist in a different jail),
1464 	 * and updates keyed by the name itself (where the name must exist
1465 	 * because that is the jail being updated).
1466 	 */
1467 	namelc = NULL;
1468 	if (name != NULL) {
1469 		namelc = strrchr(name, '.');
1470 		if (namelc == NULL)
1471 			namelc = name;
1472 		else {
1473 			/*
1474 			 * This is a hierarchical name.  Split it into the
1475 			 * parent and child names, and make sure the parent
1476 			 * exists or matches an already found jail.
1477 			 */
1478 			if (pr != NULL) {
1479 				if (strncmp(name, ppr->pr_name, namelc - name)
1480 				    || ppr->pr_name[namelc - name] != '\0') {
1481 					error = EINVAL;
1482 					vfs_opterror(opts,
1483 					    "cannot change jail's parent");
1484 					goto done_deref;
1485 				}
1486 			} else {
1487 				*namelc = '\0';
1488 				ppr = prison_find_name(mypr, name);
1489 				if (ppr == NULL) {
1490 					error = ENOENT;
1491 					vfs_opterror(opts,
1492 					    "jail \"%s\" not found", name);
1493 					goto done_deref;
1494 				}
1495 				mtx_unlock(&ppr->pr_mtx);
1496 				if (!prison_isalive(ppr)) {
1497 					error = ENOENT;
1498 					vfs_opterror(opts,
1499 					    "jail \"%s\" is dying", name);
1500 					goto done_deref;
1501 				}
1502 				*namelc = '.';
1503 			}
1504 			namelc++;
1505 		}
1506 		if (namelc[0] != '\0') {
1507 			pnamelen =
1508 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1509 			deadpr = NULL;
1510 			FOREACH_PRISON_CHILD(ppr, tpr) {
1511 				if (tpr != pr &&
1512 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1513 					if (prison_isalive(tpr)) {
1514 						if (pr == NULL &&
1515 						    cuflags != JAIL_CREATE) {
1516 							/*
1517 							 * Use this jail
1518 							 * for updates.
1519 							 */
1520 							pr = tpr;
1521 							mtx_lock(&pr->pr_mtx);
1522 							drflags |= PD_LOCKED;
1523 							break;
1524 						}
1525 						/*
1526 						 * Create, or update(jid):
1527 						 * name must not exist in an
1528 						 * active sibling jail.
1529 						 */
1530 						error = EEXIST;
1531 						vfs_opterror(opts,
1532 						   "jail \"%s\" already exists",
1533 						   name);
1534 						goto done_deref;
1535 					}
1536 					if (pr == NULL &&
1537 					    cuflags != JAIL_CREATE) {
1538 						deadpr = tpr;
1539 					}
1540 				}
1541 			}
1542 			/* If no active jail is found, use a dying one. */
1543 			if (deadpr != NULL && pr == NULL) {
1544 				if (flags & JAIL_DYING) {
1545 					pr = deadpr;
1546 					mtx_lock(&pr->pr_mtx);
1547 					drflags |= PD_LOCKED;
1548 				} else if (cuflags == JAIL_UPDATE) {
1549 					error = ENOENT;
1550 					vfs_opterror(opts,
1551 					    "jail \"%s\" is dying", name);
1552 					goto done_deref;
1553 				}
1554 			}
1555 			/* Update: name must exist if no jid. */
1556 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1557 				error = ENOENT;
1558 				vfs_opterror(opts, "jail \"%s\" not found",
1559 				    name);
1560 				goto done_deref;
1561 			}
1562 		}
1563 	}
1564 	/* Update: must provide a jid or name. */
1565 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1566 		error = ENOENT;
1567 		vfs_opterror(opts, "update specified no jail");
1568 		goto done_deref;
1569 	}
1570 
1571 	/* If there's no prison to update, create a new one and link it in. */
1572 	created = pr == NULL;
1573 	if (created) {
1574 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1575 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1576 				error = EPERM;
1577 				vfs_opterror(opts, "prison limit exceeded");
1578 				goto done_deref;
1579 			}
1580 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1581 			error = EAGAIN;
1582 			vfs_opterror(opts, "no available jail IDs");
1583 			goto done_deref;
1584 		}
1585 
1586 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1587 		pr->pr_state = PRISON_STATE_INVALID;
1588 		refcount_init(&pr->pr_ref, 1);
1589 		refcount_init(&pr->pr_uref, 0);
1590 		drflags |= PD_DEREF;
1591 		LIST_INIT(&pr->pr_children);
1592 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1593 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1594 
1595 		pr->pr_id = jid;
1596 		if (inspr != NULL)
1597 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1598 		else
1599 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1600 
1601 		pr->pr_parent = ppr;
1602 		prison_hold(ppr);
1603 		prison_proc_hold(ppr);
1604 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1605 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1606 			tpr->pr_childcount++;
1607 
1608 		/* Set some default values, and inherit some from the parent. */
1609 		if (namelc == NULL)
1610 			namelc = "";
1611 		if (path == NULL) {
1612 			path = "/";
1613 			root = mypr->pr_root;
1614 			vref(root);
1615 		}
1616 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1617 		pr->pr_flags |= PR_HOST;
1618 #if defined(INET) || defined(INET6)
1619 #ifdef VIMAGE
1620 		if (!(pr_flags & PR_VNET))
1621 #endif
1622 		{
1623 #ifdef INET
1624 			if (!(ch_flags & PR_IP4_USER))
1625 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1626 			else if (!(pr_flags & PR_IP4_USER)) {
1627 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1628 				prison_ip_dup(ppr, pr, PR_INET);
1629 			}
1630 #endif
1631 #ifdef INET6
1632 			if (!(ch_flags & PR_IP6_USER))
1633 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1634 			else if (!(pr_flags & PR_IP6_USER)) {
1635 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1636 				prison_ip_dup(ppr, pr, PR_INET6);
1637 			}
1638 #endif
1639 		}
1640 #endif
1641 		/* Source address selection is always on by default. */
1642 		pr->pr_flags |= _PR_IP_SADDRSEL;
1643 
1644 		pr->pr_securelevel = ppr->pr_securelevel;
1645 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1646 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1647 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1648 
1649 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1650 		if (osrelstr == NULL)
1651 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1652 			    sizeof(pr->pr_osrelease));
1653 		else
1654 			strlcpy(pr->pr_osrelease, osrelstr,
1655 			    sizeof(pr->pr_osrelease));
1656 
1657 #ifdef VIMAGE
1658 		/* Allocate a new vnet if specified. */
1659 		pr->pr_vnet = (pr_flags & PR_VNET)
1660 		    ? vnet_alloc() : ppr->pr_vnet;
1661 #endif
1662 		/*
1663 		 * Allocate a dedicated cpuset for each jail.
1664 		 * Unlike other initial settings, this may return an error.
1665 		 */
1666 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1667 		if (error)
1668 			goto done_deref;
1669 
1670 		mtx_lock(&pr->pr_mtx);
1671 		drflags |= PD_LOCKED;
1672 	} else {
1673 		/*
1674 		 * Grab a reference for existing prisons, to ensure they
1675 		 * continue to exist for the duration of the call.
1676 		 */
1677 		prison_hold(pr);
1678 		drflags |= PD_DEREF;
1679 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1680 		if ((pr->pr_flags & PR_VNET) &&
1681 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1682 			error = EINVAL;
1683 			vfs_opterror(opts,
1684 			    "vnet jails cannot have IP address restrictions");
1685 			goto done_deref;
1686 		}
1687 #endif
1688 #ifdef INET
1689 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1690 			error = EINVAL;
1691 			vfs_opterror(opts,
1692 			    "ip4 cannot be changed after creation");
1693 			goto done_deref;
1694 		}
1695 #endif
1696 #ifdef INET6
1697 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1698 			error = EINVAL;
1699 			vfs_opterror(opts,
1700 			    "ip6 cannot be changed after creation");
1701 			goto done_deref;
1702 		}
1703 #endif
1704 	}
1705 
1706 	/* Do final error checking before setting anything. */
1707 	if (gotslevel) {
1708 		if (slevel < ppr->pr_securelevel) {
1709 			error = EPERM;
1710 			goto done_deref;
1711 		}
1712 	}
1713 	if (gotchildmax) {
1714 		if (childmax >= ppr->pr_childmax) {
1715 			error = EPERM;
1716 			goto done_deref;
1717 		}
1718 	}
1719 	if (gotenforce) {
1720 		if (enforce < ppr->pr_enforce_statfs) {
1721 			error = EPERM;
1722 			goto done_deref;
1723 		}
1724 	}
1725 	if (gotrsnum) {
1726 		/*
1727 		 * devfs_rsnum is a uint16_t
1728 		 */
1729 		if (rsnum < 0 || rsnum > 65535) {
1730 			error = EINVAL;
1731 			goto done_deref;
1732 		}
1733 		/*
1734 		 * Nested jails always inherit parent's devfs ruleset
1735 		 */
1736 		if (jailed(td->td_ucred)) {
1737 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1738 				error = EPERM;
1739 				goto done_deref;
1740 			} else
1741 				rsnum = ppr->pr_devfs_rsnum;
1742 		}
1743 	}
1744 #ifdef INET
1745 	if (ip4s > 0) {
1746 		if ((ppr->pr_flags & PR_IP4) &&
1747 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1748 		    PR_INET)) {
1749 			error = EPERM;
1750 			goto done_deref;
1751 		}
1752 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1753 			error = EADDRINUSE;
1754 			vfs_opterror(opts, "IPv4 addresses clash");
1755 			goto done_deref;
1756 		}
1757 	}
1758 #endif
1759 #ifdef INET6
1760 	if (ip6s > 0) {
1761 		if ((ppr->pr_flags & PR_IP6) &&
1762 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1763 		    PR_INET6)) {
1764 			error = EPERM;
1765 			goto done_deref;
1766 		}
1767 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1768 			error = EADDRINUSE;
1769 			vfs_opterror(opts, "IPv6 addresses clash");
1770 			goto done_deref;
1771 		}
1772 	}
1773 #endif
1774 	onamelen = namelen = 0;
1775 	if (namelc != NULL) {
1776 		/* Give a default name of the jid.  Also allow the name to be
1777 		 * explicitly the jid - but not any other number, and only in
1778 		 * normal form (no leading zero/etc).
1779 		 */
1780 		if (namelc[0] == '\0')
1781 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1782 		else if ((strtoul(namelc, &p, 10) != jid ||
1783 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1784 			error = EINVAL;
1785 			vfs_opterror(opts,
1786 			    "name cannot be numeric (unless it is the jid)");
1787 			goto done_deref;
1788 		}
1789 		/*
1790 		 * Make sure the name isn't too long for the prison or its
1791 		 * children.
1792 		 */
1793 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1794 		onamelen = strlen(pr->pr_name + pnamelen);
1795 		namelen = strlen(namelc);
1796 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1797 			error = ENAMETOOLONG;
1798 			goto done_deref;
1799 		}
1800 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1801 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1802 			    sizeof(pr->pr_name)) {
1803 				error = ENAMETOOLONG;
1804 				goto done_deref;
1805 			}
1806 		}
1807 	}
1808 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1809 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1810 		error = EPERM;
1811 		goto done_deref;
1812 	}
1813 
1814 	/*
1815 	 * Let modules check their parameters.  This requires unlocking and
1816 	 * then re-locking the prison, but this is still a valid state as long
1817 	 * as allprison_lock remains xlocked.
1818 	 */
1819 	mtx_unlock(&pr->pr_mtx);
1820 	drflags &= ~PD_LOCKED;
1821 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1822 	if (error != 0)
1823 		goto done_deref;
1824 	mtx_lock(&pr->pr_mtx);
1825 	drflags |= PD_LOCKED;
1826 
1827 	/* At this point, all valid parameters should have been noted. */
1828 	TAILQ_FOREACH(opt, opts, link) {
1829 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1830 			error = EINVAL;
1831 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1832 			goto done_deref;
1833 		}
1834 	}
1835 
1836 	/* Set the parameters of the prison. */
1837 #ifdef INET
1838 	redo_ip4 = 0;
1839 	if (pr_flags & PR_IP4_USER) {
1840 		pr->pr_flags |= PR_IP4;
1841 		prison_ip_set(pr, PR_INET, ip4);
1842 		ip4 = NULL;
1843 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1844 #ifdef VIMAGE
1845 			if (tpr->pr_flags & PR_VNET) {
1846 				descend = 0;
1847 				continue;
1848 			}
1849 #endif
1850 			if (prison_ip_restrict(tpr, PR_INET, NULL)) {
1851 				redo_ip4 = 1;
1852 				descend = 0;
1853 			}
1854 		}
1855 	}
1856 #endif
1857 #ifdef INET6
1858 	redo_ip6 = 0;
1859 	if (pr_flags & PR_IP6_USER) {
1860 		pr->pr_flags |= PR_IP6;
1861 		prison_ip_set(pr, PR_INET6, ip6);
1862 		ip6 = NULL;
1863 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1864 #ifdef VIMAGE
1865 			if (tpr->pr_flags & PR_VNET) {
1866 				descend = 0;
1867 				continue;
1868 			}
1869 #endif
1870 			if (prison_ip_restrict(tpr, PR_INET6, NULL)) {
1871 				redo_ip6 = 1;
1872 				descend = 0;
1873 			}
1874 		}
1875 	}
1876 #endif
1877 	if (gotslevel) {
1878 		pr->pr_securelevel = slevel;
1879 		/* Set all child jails to be at least this level. */
1880 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1881 			if (tpr->pr_securelevel < slevel)
1882 				tpr->pr_securelevel = slevel;
1883 	}
1884 	if (gotchildmax) {
1885 		pr->pr_childmax = childmax;
1886 		/* Set all child jails to under this limit. */
1887 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1888 			if (tpr->pr_childmax > childmax - level)
1889 				tpr->pr_childmax = childmax > level
1890 				    ? childmax - level : 0;
1891 	}
1892 	if (gotenforce) {
1893 		pr->pr_enforce_statfs = enforce;
1894 		/* Pass this restriction on to the children. */
1895 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1896 			if (tpr->pr_enforce_statfs < enforce)
1897 				tpr->pr_enforce_statfs = enforce;
1898 	}
1899 	if (gotrsnum) {
1900 		pr->pr_devfs_rsnum = rsnum;
1901 		/* Pass this restriction on to the children. */
1902 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1903 			tpr->pr_devfs_rsnum = rsnum;
1904 	}
1905 	if (namelc != NULL) {
1906 		if (ppr == &prison0)
1907 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1908 		else
1909 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1910 			    ppr->pr_name, namelc);
1911 		/* Change this component of child names. */
1912 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1913 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1914 			    strlen(tpr->pr_name + onamelen) + 1);
1915 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1916 		}
1917 	}
1918 	if (path != NULL) {
1919 		/* Try to keep a real-rooted full pathname. */
1920 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1921 		pr->pr_root = root;
1922 		root = NULL;
1923 	}
1924 	if (PR_HOST & ch_flags & ~pr_flags) {
1925 		if (pr->pr_flags & PR_HOST) {
1926 			/*
1927 			 * Copy the parent's host info.  As with pr_ip4 above,
1928 			 * the lack of a lock on the parent is not a problem;
1929 			 * it is always set with allprison_lock at least
1930 			 * shared, and is held exclusively here.
1931 			 */
1932 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1933 			    sizeof(pr->pr_hostname));
1934 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1935 			    sizeof(pr->pr_domainname));
1936 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1937 			    sizeof(pr->pr_hostuuid));
1938 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1939 		}
1940 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1941 		/* Set this prison, and any descendants without PR_HOST. */
1942 		if (host != NULL)
1943 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1944 		if (domain != NULL)
1945 			strlcpy(pr->pr_domainname, domain,
1946 			    sizeof(pr->pr_domainname));
1947 		if (uuid != NULL)
1948 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1949 		if (gothid)
1950 			pr->pr_hostid = hid;
1951 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1952 			if (tpr->pr_flags & PR_HOST)
1953 				descend = 0;
1954 			else {
1955 				if (host != NULL)
1956 					strlcpy(tpr->pr_hostname,
1957 					    pr->pr_hostname,
1958 					    sizeof(tpr->pr_hostname));
1959 				if (domain != NULL)
1960 					strlcpy(tpr->pr_domainname,
1961 					    pr->pr_domainname,
1962 					    sizeof(tpr->pr_domainname));
1963 				if (uuid != NULL)
1964 					strlcpy(tpr->pr_hostuuid,
1965 					    pr->pr_hostuuid,
1966 					    sizeof(tpr->pr_hostuuid));
1967 				if (gothid)
1968 					tpr->pr_hostid = hid;
1969 			}
1970 		}
1971 	}
1972 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1973 	if ((tallow = ch_allow & ~pr_allow))
1974 		prison_set_allow_locked(pr, tallow, 0);
1975 	/*
1976 	 * Persistent prisons get an extra reference, and prisons losing their
1977 	 * persist flag lose that reference.
1978 	 */
1979 	born = !prison_isalive(pr);
1980 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
1981 		if (pr_flags & PR_PERSIST) {
1982 			prison_hold(pr);
1983 			/*
1984 			 * This may make a dead prison alive again, but wait
1985 			 * to label it as such until after OSD calls have had
1986 			 * a chance to run (and perhaps to fail).
1987 			 */
1988 			refcount_acquire(&pr->pr_uref);
1989 		} else {
1990 			drflags |= PD_DEUREF;
1991 			prison_free_not_last(pr);
1992 		}
1993 	}
1994 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1995 	mtx_unlock(&pr->pr_mtx);
1996 	drflags &= ~PD_LOCKED;
1997 	/*
1998 	 * Any errors past this point will need to de-persist newly created
1999 	 * prisons, as well as call remove methods.
2000 	 */
2001 	if (born)
2002 		drflags |= PD_KILL;
2003 
2004 #ifdef RACCT
2005 	if (racct_enable && created)
2006 		prison_racct_attach(pr);
2007 #endif
2008 
2009 	/* Locks may have prevented a complete restriction of child IP
2010 	 * addresses.  If so, allocate some more memory and try again.
2011 	 */
2012 #ifdef INET
2013 	while (redo_ip4) {
2014 		ip4s = pr->pr_addrs[PR_INET]->ips;
2015 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2016 		mtx_lock(&pr->pr_mtx);
2017 		redo_ip4 = 0;
2018 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2019 #ifdef VIMAGE
2020 			if (tpr->pr_flags & PR_VNET) {
2021 				descend = 0;
2022 				continue;
2023 			}
2024 #endif
2025 			if (prison_ip_restrict(tpr, PR_INET, ip4)) {
2026 				if (ip4 != NULL)
2027 					ip4 = NULL;
2028 				else
2029 					redo_ip4 = 1;
2030 			}
2031 		}
2032 		mtx_unlock(&pr->pr_mtx);
2033 	}
2034 #endif
2035 #ifdef INET6
2036 	while (redo_ip6) {
2037 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2038 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2039 		mtx_lock(&pr->pr_mtx);
2040 		redo_ip6 = 0;
2041 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2042 #ifdef VIMAGE
2043 			if (tpr->pr_flags & PR_VNET) {
2044 				descend = 0;
2045 				continue;
2046 			}
2047 #endif
2048 			if (prison_ip_restrict(tpr, PR_INET6, ip6)) {
2049 				if (ip6 != NULL)
2050 					ip6 = NULL;
2051 				else
2052 					redo_ip6 = 1;
2053 			}
2054 		}
2055 		mtx_unlock(&pr->pr_mtx);
2056 	}
2057 #endif
2058 
2059 	/* Let the modules do their work. */
2060 	if (born) {
2061 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2062 		if (error)
2063 			goto done_deref;
2064 	}
2065 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2066 	if (error)
2067 		goto done_deref;
2068 
2069 	/*
2070 	 * A new prison is now ready to be seen; either it has gained a user
2071 	 * reference via persistence, or is about to gain one via attachment.
2072 	 */
2073 	if (born) {
2074 		drflags = prison_lock_xlock(pr, drflags);
2075 		pr->pr_state = PRISON_STATE_ALIVE;
2076 	}
2077 
2078 	/* Attach this process to the prison if requested. */
2079 	if (flags & JAIL_ATTACH) {
2080 		error = do_jail_attach(td, pr,
2081 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2082 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2083 		if (error) {
2084 			vfs_opterror(opts, "attach failed");
2085 			goto done_deref;
2086 		}
2087 	}
2088 
2089 #ifdef RACCT
2090 	if (racct_enable && !created) {
2091 		if (drflags & PD_LOCKED) {
2092 			mtx_unlock(&pr->pr_mtx);
2093 			drflags &= ~PD_LOCKED;
2094 		}
2095 		if (drflags & PD_LIST_XLOCKED) {
2096 			sx_xunlock(&allprison_lock);
2097 			drflags &= ~PD_LIST_XLOCKED;
2098 		}
2099 		prison_racct_modify(pr);
2100 	}
2101 #endif
2102 
2103 	drflags &= ~PD_KILL;
2104 	td->td_retval[0] = pr->pr_id;
2105 
2106  done_deref:
2107 	/* Release any temporary prison holds and/or locks. */
2108 	if (pr != NULL)
2109 		prison_deref(pr, drflags);
2110 	else if (drflags & PD_LIST_SLOCKED)
2111 		sx_sunlock(&allprison_lock);
2112 	else if (drflags & PD_LIST_XLOCKED)
2113 		sx_xunlock(&allprison_lock);
2114 	if (root != NULL)
2115 		vrele(root);
2116  done_errmsg:
2117 	if (error) {
2118 		/* Write the error message back to userspace. */
2119 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2120 		    &errmsg_len) == 0 && errmsg_len > 0) {
2121 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2122 			if (optuio->uio_segflg == UIO_SYSSPACE)
2123 				bcopy(errmsg,
2124 				    optuio->uio_iov[errmsg_pos].iov_base,
2125 				    errmsg_len);
2126 			else
2127 				copyout(errmsg,
2128 				    optuio->uio_iov[errmsg_pos].iov_base,
2129 				    errmsg_len);
2130 		}
2131 	}
2132  done_free:
2133 #ifdef INET
2134 	prison_ip_free(ip4);
2135 #endif
2136 #ifdef INET6
2137 	prison_ip_free(ip6);
2138 #endif
2139 	if (g_path != NULL)
2140 		free(g_path, M_TEMP);
2141 	vfs_freeopts(opts);
2142 	return (error);
2143 }
2144 
2145 /*
2146  * Find the next available prison ID.  Return the ID on success, or zero
2147  * on failure.  Also set a pointer to the allprison list entry the prison
2148  * should be inserted before.
2149  */
2150 static int
2151 get_next_prid(struct prison **insprp)
2152 {
2153 	struct prison *inspr;
2154 	int jid, maxid;
2155 
2156 	jid = lastprid % JAIL_MAX + 1;
2157 	if (TAILQ_EMPTY(&allprison) ||
2158 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2159 		/*
2160 		 * A common case is for all jails to be implicitly numbered,
2161 		 * which means they'll go on the end of the list, at least
2162 		 * for the first JAIL_MAX times.
2163 		 */
2164 		inspr = NULL;
2165 	} else {
2166 		/*
2167 		 * Take two passes through the allprison list: first starting
2168 		 * with the proposed jid, then ending with it.
2169 		 */
2170 		for (maxid = JAIL_MAX; maxid != 0; ) {
2171 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2172 				if (inspr->pr_id < jid)
2173 					continue;
2174 				if (inspr->pr_id > jid) {
2175 					/* Found an opening. */
2176 					maxid = 0;
2177 					break;
2178 				}
2179 				if (++jid > maxid) {
2180 					if (lastprid == maxid || lastprid == 0)
2181 					{
2182 						/*
2183 						 * The entire legal range
2184 						 * has been traversed
2185 						 */
2186 						return 0;
2187 					}
2188 					/* Try again from the start. */
2189 					jid = 1;
2190 					maxid = lastprid;
2191 					break;
2192 				}
2193 			}
2194 			if (inspr == NULL) {
2195 				/* Found room at the end of the list. */
2196 				break;
2197 			}
2198 		}
2199 	}
2200 	*insprp = inspr;
2201 	lastprid = jid;
2202 	return (jid);
2203 }
2204 
2205 /*
2206  * struct jail_get_args {
2207  *	struct iovec *iovp;
2208  *	unsigned int iovcnt;
2209  *	int flags;
2210  * };
2211  */
2212 int
2213 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2214 {
2215 	struct uio *auio;
2216 	int error;
2217 
2218 	/* Check that we have an even number of iovecs. */
2219 	if (uap->iovcnt & 1)
2220 		return (EINVAL);
2221 
2222 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2223 	if (error)
2224 		return (error);
2225 	error = kern_jail_get(td, auio, uap->flags);
2226 	if (error == 0)
2227 		error = copyout(auio->uio_iov, uap->iovp,
2228 		    uap->iovcnt * sizeof (struct iovec));
2229 	free(auio, M_IOV);
2230 	return (error);
2231 }
2232 
2233 int
2234 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2235 {
2236 	struct bool_flags *bf;
2237 	struct jailsys_flags *jsf;
2238 	struct prison *pr, *mypr;
2239 	struct vfsopt *opt;
2240 	struct vfsoptlist *opts;
2241 	char *errmsg, *name;
2242 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2243 	unsigned f;
2244 
2245 	if (flags & ~JAIL_GET_MASK)
2246 		return (EINVAL);
2247 
2248 	/* Get the parameter list. */
2249 	error = vfs_buildopts(optuio, &opts);
2250 	if (error)
2251 		return (error);
2252 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2253 	mypr = td->td_ucred->cr_prison;
2254 	pr = NULL;
2255 
2256 	/*
2257 	 * Find the prison specified by one of: lastjid, jid, name.
2258 	 */
2259 	sx_slock(&allprison_lock);
2260 	drflags = PD_LIST_SLOCKED;
2261 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2262 	if (error == 0) {
2263 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2264 			if (pr->pr_id > jid &&
2265 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2266 			    prison_ischild(mypr, pr)) {
2267 				mtx_lock(&pr->pr_mtx);
2268 				drflags |= PD_LOCKED;
2269 				goto found_prison;
2270 			}
2271 		}
2272 		error = ENOENT;
2273 		vfs_opterror(opts, "no jail after %d", jid);
2274 		goto done;
2275 	} else if (error != ENOENT)
2276 		goto done;
2277 
2278 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2279 	if (error == 0) {
2280 		if (jid != 0) {
2281 			pr = prison_find_child(mypr, jid);
2282 			if (pr != NULL) {
2283 				drflags |= PD_LOCKED;
2284 				if (!(prison_isalive(pr) ||
2285 				    (flags & JAIL_DYING))) {
2286 					error = ENOENT;
2287 					vfs_opterror(opts, "jail %d is dying",
2288 					    jid);
2289 					goto done;
2290 				}
2291 				goto found_prison;
2292 			}
2293 			error = ENOENT;
2294 			vfs_opterror(opts, "jail %d not found", jid);
2295 			goto done;
2296 		}
2297 	} else if (error != ENOENT)
2298 		goto done;
2299 
2300 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2301 	if (error == 0) {
2302 		if (len == 0 || name[len - 1] != '\0') {
2303 			error = EINVAL;
2304 			goto done;
2305 		}
2306 		pr = prison_find_name(mypr, name);
2307 		if (pr != NULL) {
2308 			drflags |= PD_LOCKED;
2309 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2310 				error = ENOENT;
2311 				vfs_opterror(opts, "jail \"%s\" is dying",
2312 				    name);
2313 				goto done;
2314 			}
2315 			goto found_prison;
2316 		}
2317 		error = ENOENT;
2318 		vfs_opterror(opts, "jail \"%s\" not found", name);
2319 		goto done;
2320 	} else if (error != ENOENT)
2321 		goto done;
2322 
2323 	vfs_opterror(opts, "no jail specified");
2324 	error = ENOENT;
2325 	goto done;
2326 
2327  found_prison:
2328 	/* Get the parameters of the prison. */
2329 	prison_hold(pr);
2330 	drflags |= PD_DEREF;
2331 	td->td_retval[0] = pr->pr_id;
2332 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2333 	if (error != 0 && error != ENOENT)
2334 		goto done;
2335 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2336 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2337 	if (error != 0 && error != ENOENT)
2338 		goto done;
2339 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2340 	if (error != 0 && error != ENOENT)
2341 		goto done;
2342 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2343 	    sizeof(pr->pr_cpuset->cs_id));
2344 	if (error != 0 && error != ENOENT)
2345 		goto done;
2346 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2347 	if (error != 0 && error != ENOENT)
2348 		goto done;
2349 #ifdef INET
2350 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET] + 1,
2351 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2352 	    pr_families[PR_INET].size : 0 );
2353 	if (error != 0 && error != ENOENT)
2354 		goto done;
2355 #endif
2356 #ifdef INET6
2357 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6] + 1,
2358 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2359 	    pr_families[PR_INET6].size : 0 );
2360 	if (error != 0 && error != ENOENT)
2361 		goto done;
2362 #endif
2363 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2364 	    sizeof(pr->pr_securelevel));
2365 	if (error != 0 && error != ENOENT)
2366 		goto done;
2367 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2368 	    sizeof(pr->pr_childcount));
2369 	if (error != 0 && error != ENOENT)
2370 		goto done;
2371 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2372 	    sizeof(pr->pr_childmax));
2373 	if (error != 0 && error != ENOENT)
2374 		goto done;
2375 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2376 	if (error != 0 && error != ENOENT)
2377 		goto done;
2378 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2379 	if (error != 0 && error != ENOENT)
2380 		goto done;
2381 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2382 	if (error != 0 && error != ENOENT)
2383 		goto done;
2384 #ifdef COMPAT_FREEBSD32
2385 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2386 		uint32_t hid32 = pr->pr_hostid;
2387 
2388 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2389 	} else
2390 #endif
2391 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2392 	    sizeof(pr->pr_hostid));
2393 	if (error != 0 && error != ENOENT)
2394 		goto done;
2395 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2396 	    sizeof(pr->pr_enforce_statfs));
2397 	if (error != 0 && error != ENOENT)
2398 		goto done;
2399 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2400 	    sizeof(pr->pr_devfs_rsnum));
2401 	if (error != 0 && error != ENOENT)
2402 		goto done;
2403 	for (bf = pr_flag_bool;
2404 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2405 	     bf++) {
2406 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2407 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2408 		if (error != 0 && error != ENOENT)
2409 			goto done;
2410 		i = !i;
2411 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2412 		if (error != 0 && error != ENOENT)
2413 			goto done;
2414 	}
2415 	for (jsf = pr_flag_jailsys;
2416 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2417 	     jsf++) {
2418 		f = pr->pr_flags & (jsf->disable | jsf->new);
2419 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2420 		    : (f == jsf->new) ? JAIL_SYS_NEW
2421 		    : JAIL_SYS_INHERIT;
2422 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2423 		if (error != 0 && error != ENOENT)
2424 			goto done;
2425 	}
2426 	for (bf = pr_flag_allow;
2427 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2428 		atomic_load_int(&bf->flag) != 0;
2429 	     bf++) {
2430 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2431 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2432 		if (error != 0 && error != ENOENT)
2433 			goto done;
2434 		i = !i;
2435 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2436 		if (error != 0 && error != ENOENT)
2437 			goto done;
2438 	}
2439 	i = !prison_isalive(pr);
2440 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2441 	if (error != 0 && error != ENOENT)
2442 		goto done;
2443 	i = !i;
2444 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2445 	if (error != 0 && error != ENOENT)
2446 		goto done;
2447 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2448 	    sizeof(pr->pr_osreldate));
2449 	if (error != 0 && error != ENOENT)
2450 		goto done;
2451 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2452 	if (error != 0 && error != ENOENT)
2453 		goto done;
2454 
2455 	/* Get the module parameters. */
2456 	mtx_unlock(&pr->pr_mtx);
2457 	drflags &= ~PD_LOCKED;
2458 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2459 	if (error)
2460 		goto done;
2461 	prison_deref(pr, drflags);
2462 	pr = NULL;
2463 	drflags = 0;
2464 
2465 	/* By now, all parameters should have been noted. */
2466 	TAILQ_FOREACH(opt, opts, link) {
2467 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2468 			error = EINVAL;
2469 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2470 			goto done;
2471 		}
2472 	}
2473 
2474 	/* Write the fetched parameters back to userspace. */
2475 	error = 0;
2476 	TAILQ_FOREACH(opt, opts, link) {
2477 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2478 			pos = 2 * opt->pos + 1;
2479 			optuio->uio_iov[pos].iov_len = opt->len;
2480 			if (opt->value != NULL) {
2481 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2482 					bcopy(opt->value,
2483 					    optuio->uio_iov[pos].iov_base,
2484 					    opt->len);
2485 				} else {
2486 					error = copyout(opt->value,
2487 					    optuio->uio_iov[pos].iov_base,
2488 					    opt->len);
2489 					if (error)
2490 						break;
2491 				}
2492 			}
2493 		}
2494 	}
2495 
2496  done:
2497 	/* Release any temporary prison holds and/or locks. */
2498 	if (pr != NULL)
2499 		prison_deref(pr, drflags);
2500 	else if (drflags & PD_LIST_SLOCKED)
2501 		sx_sunlock(&allprison_lock);
2502 	if (error && errmsg_pos >= 0) {
2503 		/* Write the error message back to userspace. */
2504 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2505 		errmsg_pos = 2 * errmsg_pos + 1;
2506 		if (errmsg_len > 0) {
2507 			if (optuio->uio_segflg == UIO_SYSSPACE)
2508 				bcopy(errmsg,
2509 				    optuio->uio_iov[errmsg_pos].iov_base,
2510 				    errmsg_len);
2511 			else
2512 				copyout(errmsg,
2513 				    optuio->uio_iov[errmsg_pos].iov_base,
2514 				    errmsg_len);
2515 		}
2516 	}
2517 	vfs_freeopts(opts);
2518 	return (error);
2519 }
2520 
2521 /*
2522  * struct jail_remove_args {
2523  *	int jid;
2524  * };
2525  */
2526 int
2527 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2528 {
2529 	struct prison *pr;
2530 	int error;
2531 
2532 	error = priv_check(td, PRIV_JAIL_REMOVE);
2533 	if (error)
2534 		return (error);
2535 
2536 	sx_xlock(&allprison_lock);
2537 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2538 	if (pr == NULL) {
2539 		sx_xunlock(&allprison_lock);
2540 		return (EINVAL);
2541 	}
2542 	if (!prison_isalive(pr)) {
2543 		/* Silently ignore already-dying prisons. */
2544 		mtx_unlock(&pr->pr_mtx);
2545 		sx_xunlock(&allprison_lock);
2546 		return (0);
2547 	}
2548 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2549 	return (0);
2550 }
2551 
2552 /*
2553  * struct jail_attach_args {
2554  *	int jid;
2555  * };
2556  */
2557 int
2558 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2559 {
2560 	struct prison *pr;
2561 	int error;
2562 
2563 	error = priv_check(td, PRIV_JAIL_ATTACH);
2564 	if (error)
2565 		return (error);
2566 
2567 	sx_slock(&allprison_lock);
2568 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2569 	if (pr == NULL) {
2570 		sx_sunlock(&allprison_lock);
2571 		return (EINVAL);
2572 	}
2573 
2574 	/* Do not allow a process to attach to a prison that is not alive. */
2575 	if (!prison_isalive(pr)) {
2576 		mtx_unlock(&pr->pr_mtx);
2577 		sx_sunlock(&allprison_lock);
2578 		return (EINVAL);
2579 	}
2580 
2581 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2582 }
2583 
2584 static int
2585 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2586 {
2587 	struct proc *p;
2588 	struct ucred *newcred, *oldcred;
2589 	int error;
2590 
2591 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2592 	sx_assert(&allprison_lock, SX_LOCKED);
2593 	drflags &= PD_LOCK_FLAGS;
2594 	/*
2595 	 * XXX: Note that there is a slight race here if two threads
2596 	 * in the same privileged process attempt to attach to two
2597 	 * different jails at the same time.  It is important for
2598 	 * user processes not to do this, or they might end up with
2599 	 * a process root from one prison, but attached to the jail
2600 	 * of another.
2601 	 */
2602 	prison_hold(pr);
2603 	refcount_acquire(&pr->pr_uref);
2604 	drflags |= PD_DEREF | PD_DEUREF;
2605 	mtx_unlock(&pr->pr_mtx);
2606 	drflags &= ~PD_LOCKED;
2607 
2608 	/* Let modules do whatever they need to prepare for attaching. */
2609 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2610 	if (error) {
2611 		prison_deref(pr, drflags);
2612 		return (error);
2613 	}
2614 	sx_unlock(&allprison_lock);
2615 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2616 
2617 	/*
2618 	 * Reparent the newly attached process to this jail.
2619 	 */
2620 	p = td->td_proc;
2621 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2622 	if (error)
2623 		goto e_revert_osd;
2624 
2625 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2626 	if ((error = change_dir(pr->pr_root, td)) != 0)
2627 		goto e_unlock;
2628 #ifdef MAC
2629 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2630 		goto e_unlock;
2631 #endif
2632 	VOP_UNLOCK(pr->pr_root);
2633 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2634 		goto e_revert_osd;
2635 
2636 	newcred = crget();
2637 	PROC_LOCK(p);
2638 	oldcred = crcopysafe(p, newcred);
2639 	newcred->cr_prison = pr;
2640 	proc_set_cred(p, newcred);
2641 	setsugid(p);
2642 #ifdef RACCT
2643 	racct_proc_ucred_changed(p, oldcred, newcred);
2644 	crhold(newcred);
2645 #endif
2646 	PROC_UNLOCK(p);
2647 #ifdef RCTL
2648 	rctl_proc_ucred_changed(p, newcred);
2649 	crfree(newcred);
2650 #endif
2651 	prison_deref(oldcred->cr_prison, drflags);
2652 	crfree(oldcred);
2653 
2654 	/*
2655 	 * If the prison was killed while changing credentials, die along
2656 	 * with it.
2657 	 */
2658 	if (!prison_isalive(pr)) {
2659 		PROC_LOCK(p);
2660 		kern_psignal(p, SIGKILL);
2661 		PROC_UNLOCK(p);
2662 	}
2663 
2664 	return (0);
2665 
2666  e_unlock:
2667 	VOP_UNLOCK(pr->pr_root);
2668  e_revert_osd:
2669 	/* Tell modules this thread is still in its old jail after all. */
2670 	sx_slock(&allprison_lock);
2671 	drflags |= PD_LIST_SLOCKED;
2672 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2673 	prison_deref(pr, drflags);
2674 	return (error);
2675 }
2676 
2677 /*
2678  * Returns a locked prison instance, or NULL on failure.
2679  */
2680 struct prison *
2681 prison_find(int prid)
2682 {
2683 	struct prison *pr;
2684 
2685 	sx_assert(&allprison_lock, SX_LOCKED);
2686 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2687 		if (pr->pr_id < prid)
2688 			continue;
2689 		if (pr->pr_id > prid)
2690 			break;
2691 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2692 		mtx_lock(&pr->pr_mtx);
2693 		return (pr);
2694 	}
2695 	return (NULL);
2696 }
2697 
2698 /*
2699  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2700  */
2701 struct prison *
2702 prison_find_child(struct prison *mypr, int prid)
2703 {
2704 	struct prison *pr;
2705 	int descend;
2706 
2707 	sx_assert(&allprison_lock, SX_LOCKED);
2708 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2709 		if (pr->pr_id == prid) {
2710 			KASSERT(prison_isvalid(pr),
2711 			    ("Found invalid prison %p", pr));
2712 			mtx_lock(&pr->pr_mtx);
2713 			return (pr);
2714 		}
2715 	}
2716 	return (NULL);
2717 }
2718 
2719 /*
2720  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2721  */
2722 struct prison *
2723 prison_find_name(struct prison *mypr, const char *name)
2724 {
2725 	struct prison *pr, *deadpr;
2726 	size_t mylen;
2727 	int descend;
2728 
2729 	sx_assert(&allprison_lock, SX_LOCKED);
2730 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2731 	deadpr = NULL;
2732 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2733 		if (!strcmp(pr->pr_name + mylen, name)) {
2734 			KASSERT(prison_isvalid(pr),
2735 			    ("Found invalid prison %p", pr));
2736 			if (prison_isalive(pr)) {
2737 				mtx_lock(&pr->pr_mtx);
2738 				return (pr);
2739 			}
2740 			deadpr = pr;
2741 		}
2742 	}
2743 	/* There was no valid prison - perhaps there was a dying one. */
2744 	if (deadpr != NULL)
2745 		mtx_lock(&deadpr->pr_mtx);
2746 	return (deadpr);
2747 }
2748 
2749 /*
2750  * See if a prison has the specific flag set.  The prison should be locked,
2751  * unless checking for flags that are only set at jail creation (such as
2752  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2753  * to any other prison data.
2754  */
2755 int
2756 prison_flag(struct ucred *cred, unsigned flag)
2757 {
2758 
2759 	return (cred->cr_prison->pr_flags & flag);
2760 }
2761 
2762 int
2763 prison_allow(struct ucred *cred, unsigned flag)
2764 {
2765 
2766 	return ((cred->cr_prison->pr_allow & flag) != 0);
2767 }
2768 
2769 /*
2770  * Hold a prison reference, by incrementing pr_ref.  It is generally
2771  * an error to hold a prison that does not already have a reference.
2772  * A prison record will remain valid as long as it has at least one
2773  * reference, and will not be removed as long as either the prison
2774  * mutex or the allprison lock is held (allprison_lock may be shared).
2775  */
2776 void
2777 prison_hold_locked(struct prison *pr)
2778 {
2779 
2780 	/* Locking is no longer required. */
2781 	prison_hold(pr);
2782 }
2783 
2784 void
2785 prison_hold(struct prison *pr)
2786 {
2787 #ifdef INVARIANTS
2788 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2789 
2790 	KASSERT(was_valid,
2791 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2792 #else
2793 	refcount_acquire(&pr->pr_ref);
2794 #endif
2795 }
2796 
2797 /*
2798  * Remove a prison reference.  If that was the last reference, the
2799  * prison will be removed (at a later time).
2800  */
2801 void
2802 prison_free_locked(struct prison *pr)
2803 {
2804 
2805 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2806 	/*
2807 	 * Locking is no longer required, but unlock because the caller
2808 	 * expects it.
2809 	 */
2810 	mtx_unlock(&pr->pr_mtx);
2811 	prison_free(pr);
2812 }
2813 
2814 void
2815 prison_free(struct prison *pr)
2816 {
2817 
2818 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2819 	    ("Trying to free dead prison %p (jid=%d).",
2820 	     pr, pr->pr_id));
2821 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2822 		/*
2823 		 * Don't remove the last reference in this context,
2824 		 * in case there are locks held.
2825 		 */
2826 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2827 	}
2828 }
2829 
2830 static void
2831 prison_free_not_last(struct prison *pr)
2832 {
2833 #ifdef INVARIANTS
2834 	int lastref;
2835 
2836 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2837 	    ("Trying to free dead prison %p (jid=%d).",
2838 	     pr, pr->pr_id));
2839 	lastref = refcount_release(&pr->pr_ref);
2840 	KASSERT(!lastref,
2841 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2842 	     pr, pr->pr_id));
2843 #else
2844 	refcount_release(&pr->pr_ref);
2845 #endif
2846 }
2847 
2848 /*
2849  * Hold a prison for user visibility, by incrementing pr_uref.
2850  * It is generally an error to hold a prison that isn't already
2851  * user-visible, except through the the jail system calls.  It is also
2852  * an error to hold an invalid prison.  A prison record will remain
2853  * alive as long as it has at least one user reference, and will not
2854  * be set to the dying state until the prison mutex and allprison_lock
2855  * are both freed.
2856  */
2857 void
2858 prison_proc_hold(struct prison *pr)
2859 {
2860 #ifdef INVARIANTS
2861 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2862 
2863 	KASSERT(was_alive,
2864 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2865 #else
2866 	refcount_acquire(&pr->pr_uref);
2867 #endif
2868 }
2869 
2870 /*
2871  * Remove a prison user reference.  If it was the last reference, the
2872  * prison will be considered "dying", and may be removed once all of
2873  * its references are dropped.
2874  */
2875 void
2876 prison_proc_free(struct prison *pr)
2877 {
2878 
2879 	/*
2880 	 * Locking is only required when releasing the last reference.
2881 	 * This allows assurance that a locked prison will remain alive
2882 	 * until it is unlocked.
2883 	 */
2884 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2885 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2886 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2887 		/*
2888 		 * Don't remove the last user reference in this context,
2889 		 * which is expected to be a process that is not only locked,
2890 		 * but also half dead.  Add a reference so any calls to
2891 		 * prison_free() won't re-submit the task.
2892 		 */
2893 		prison_hold(pr);
2894 		mtx_lock(&pr->pr_mtx);
2895 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2896 		    ("Redundant last reference in prison_proc_free (jid=%d)",
2897 		     pr->pr_id));
2898 		pr->pr_flags |= PR_COMPLETE_PROC;
2899 		mtx_unlock(&pr->pr_mtx);
2900 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2901 	}
2902 }
2903 
2904 static void
2905 prison_proc_free_not_last(struct prison *pr)
2906 {
2907 #ifdef INVARIANTS
2908 	int lastref;
2909 
2910 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2911 	    ("Trying to free dead prison %p (jid=%d).",
2912 	     pr, pr->pr_id));
2913 	lastref = refcount_release(&pr->pr_uref);
2914 	KASSERT(!lastref,
2915 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2916 	     pr, pr->pr_id));
2917 #else
2918 	refcount_release(&pr->pr_uref);
2919 #endif
2920 }
2921 
2922 /*
2923  * Complete a call to either prison_free or prison_proc_free.
2924  */
2925 static void
2926 prison_complete(void *context, int pending)
2927 {
2928 	struct prison *pr = context;
2929 	int drflags;
2930 
2931 	/*
2932 	 * This could be called to release the last reference, or the last
2933 	 * user reference (plus the reference held in prison_proc_free).
2934 	 */
2935 	drflags = prison_lock_xlock(pr, PD_DEREF);
2936 	if (pr->pr_flags & PR_COMPLETE_PROC) {
2937 		pr->pr_flags &= ~PR_COMPLETE_PROC;
2938 		drflags |= PD_DEUREF;
2939 	}
2940 	prison_deref(pr, drflags);
2941 }
2942 
2943 /*
2944  * Remove a prison reference and/or user reference (usually).
2945  * This assumes context that allows sleeping (for allprison_lock),
2946  * with no non-sleeping locks held, except perhaps the prison itself.
2947  * If there are no more references, release and delist the prison.
2948  * On completion, the prison lock and the allprison lock are both
2949  * unlocked.
2950  */
2951 static void
2952 prison_deref(struct prison *pr, int flags)
2953 {
2954 	struct prisonlist freeprison;
2955 	struct prison *killpr, *rpr, *ppr, *tpr;
2956 	struct proc *p;
2957 
2958 	killpr = NULL;
2959 	TAILQ_INIT(&freeprison);
2960 	/*
2961 	 * Release this prison as requested, which may cause its parent
2962 	 * to be released, and then maybe its grandparent, etc.
2963 	 */
2964 	for (;;) {
2965 		if (flags & PD_KILL) {
2966 			/* Kill the prison and its descendents. */
2967 			KASSERT(pr != &prison0,
2968 			    ("prison_deref trying to kill prison0"));
2969 			if (!(flags & PD_DEREF)) {
2970 				prison_hold(pr);
2971 				flags |= PD_DEREF;
2972 			}
2973 			flags = prison_lock_xlock(pr, flags);
2974 			prison_deref_kill(pr, &freeprison);
2975 		}
2976 		if (flags & PD_DEUREF) {
2977 			/* Drop a user reference. */
2978 			KASSERT(refcount_load(&pr->pr_uref) > 0,
2979 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2980 			     pr->pr_id));
2981 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
2982 				if (!(flags & PD_DEREF)) {
2983 					prison_hold(pr);
2984 					flags |= PD_DEREF;
2985 				}
2986 				flags = prison_lock_xlock(pr, flags);
2987 				if (refcount_release(&pr->pr_uref) &&
2988 				    pr->pr_state == PRISON_STATE_ALIVE) {
2989 					/*
2990 					 * When the last user references goes,
2991 					 * this becomes a dying prison.
2992 					 */
2993 					KASSERT(
2994 					    refcount_load(&prison0.pr_uref) > 0,
2995 					    ("prison0 pr_uref=0"));
2996 					pr->pr_state = PRISON_STATE_DYING;
2997 					mtx_unlock(&pr->pr_mtx);
2998 					flags &= ~PD_LOCKED;
2999 					prison_cleanup(pr);
3000 				}
3001 			}
3002 		}
3003 		if (flags & PD_KILL) {
3004 			/*
3005 			 * Any remaining user references are probably processes
3006 			 * that need to be killed, either in this prison or its
3007 			 * descendants.
3008 			 */
3009 			if (refcount_load(&pr->pr_uref) > 0)
3010 				killpr = pr;
3011 			/* Make sure the parent prison doesn't get killed. */
3012 			flags &= ~PD_KILL;
3013 		}
3014 		if (flags & PD_DEREF) {
3015 			/* Drop a reference. */
3016 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3017 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3018 			     pr->pr_id));
3019 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3020 				flags = prison_lock_xlock(pr, flags);
3021 				if (refcount_release(&pr->pr_ref)) {
3022 					/*
3023 					 * When the last reference goes,
3024 					 * unlink the prison and set it aside.
3025 					 */
3026 					KASSERT(
3027 					    refcount_load(&pr->pr_uref) == 0,
3028 					    ("prison_deref: last ref, "
3029 					     "but still has %d urefs (jid=%d)",
3030 					     pr->pr_uref, pr->pr_id));
3031 					KASSERT(
3032 					    refcount_load(&prison0.pr_ref) != 0,
3033 					    ("prison0 pr_ref=0"));
3034 					pr->pr_state = PRISON_STATE_INVALID;
3035 					TAILQ_REMOVE(&allprison, pr, pr_list);
3036 					LIST_REMOVE(pr, pr_sibling);
3037 					TAILQ_INSERT_TAIL(&freeprison, pr,
3038 					    pr_list);
3039 					for (ppr = pr->pr_parent;
3040 					     ppr != NULL;
3041 					     ppr = ppr->pr_parent)
3042 						ppr->pr_childcount--;
3043 					/*
3044 					 * Removing a prison frees references
3045 					 * from its parent.
3046 					 */
3047 					mtx_unlock(&pr->pr_mtx);
3048 					flags &= ~PD_LOCKED;
3049 					pr = pr->pr_parent;
3050 					flags |= PD_DEREF | PD_DEUREF;
3051 					continue;
3052 				}
3053 			}
3054 		}
3055 		break;
3056 	}
3057 
3058 	/* Release all the prison locks. */
3059 	if (flags & PD_LOCKED)
3060 		mtx_unlock(&pr->pr_mtx);
3061 	if (flags & PD_LIST_SLOCKED)
3062 		sx_sunlock(&allprison_lock);
3063 	else if (flags & PD_LIST_XLOCKED)
3064 		sx_xunlock(&allprison_lock);
3065 
3066 	/* Kill any processes attached to a killed prison. */
3067 	if (killpr != NULL) {
3068 		sx_slock(&allproc_lock);
3069 		FOREACH_PROC_IN_SYSTEM(p) {
3070 			PROC_LOCK(p);
3071 			if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3072 				for (ppr = p->p_ucred->cr_prison;
3073 				     ppr != &prison0;
3074 				     ppr = ppr->pr_parent)
3075 					if (ppr == killpr) {
3076 						kern_psignal(p, SIGKILL);
3077 						break;
3078 					}
3079 			}
3080 			PROC_UNLOCK(p);
3081 		}
3082 		sx_sunlock(&allproc_lock);
3083 	}
3084 
3085 	/*
3086 	 * Finish removing any unreferenced prisons, which couldn't happen
3087 	 * while allprison_lock was held (to avoid a LOR on vrele).
3088 	 */
3089 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3090 #ifdef VIMAGE
3091 		if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
3092 			vnet_destroy(rpr->pr_vnet);
3093 #endif
3094 		if (rpr->pr_root != NULL)
3095 			vrele(rpr->pr_root);
3096 		mtx_destroy(&rpr->pr_mtx);
3097 #ifdef INET
3098 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3099 #endif
3100 #ifdef INET6
3101 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3102 #endif
3103 		if (rpr->pr_cpuset != NULL)
3104 			cpuset_rel(rpr->pr_cpuset);
3105 		osd_jail_exit(rpr);
3106 #ifdef RACCT
3107 		if (racct_enable)
3108 			prison_racct_detach(rpr);
3109 #endif
3110 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3111 		free(rpr, M_PRISON);
3112 	}
3113 }
3114 
3115 /*
3116  * Kill the prison and its descendants.  Mark them as dying, clear the
3117  * persist flag, and call module remove methods.
3118  */
3119 static void
3120 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3121 {
3122 	struct prison *cpr, *ppr, *rpr;
3123 	bool descend;
3124 
3125 	/*
3126 	 * Unlike the descendants, the target prison can be killed
3127 	 * even if it is currently dying.  This is useful for failed
3128 	 * creation in jail_set(2).
3129 	 */
3130 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3131 	    ("Trying to kill dead prison %p (jid=%d).",
3132 	     pr, pr->pr_id));
3133 	refcount_acquire(&pr->pr_uref);
3134 	pr->pr_state = PRISON_STATE_DYING;
3135 	mtx_unlock(&pr->pr_mtx);
3136 
3137 	rpr = NULL;
3138 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3139 		if (descend) {
3140 			if (!prison_isalive(cpr)) {
3141 				descend = false;
3142 				continue;
3143 			}
3144 			prison_hold(cpr);
3145 			prison_proc_hold(cpr);
3146 			mtx_lock(&cpr->pr_mtx);
3147 			cpr->pr_state = PRISON_STATE_DYING;
3148 			cpr->pr_flags |= PR_REMOVE;
3149 			mtx_unlock(&cpr->pr_mtx);
3150 			continue;
3151 		}
3152 		if (!(cpr->pr_flags & PR_REMOVE))
3153 			continue;
3154 		prison_cleanup(cpr);
3155 		mtx_lock(&cpr->pr_mtx);
3156 		cpr->pr_flags &= ~PR_REMOVE;
3157 		if (cpr->pr_flags & PR_PERSIST) {
3158 			cpr->pr_flags &= ~PR_PERSIST;
3159 			prison_proc_free_not_last(cpr);
3160 			prison_free_not_last(cpr);
3161 		}
3162 		(void)refcount_release(&cpr->pr_uref);
3163 		if (refcount_release(&cpr->pr_ref)) {
3164 			/*
3165 			 * When the last reference goes, unlink the prison
3166 			 * and set it aside for prison_deref() to handle.
3167 			 * Delay unlinking the sibling list to keep the loop
3168 			 * safe.
3169 			 */
3170 			if (rpr != NULL)
3171 				LIST_REMOVE(rpr, pr_sibling);
3172 			rpr = cpr;
3173 			rpr->pr_state = PRISON_STATE_INVALID;
3174 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3175 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3176 			/*
3177 			 * Removing a prison frees references from its parent.
3178 			 */
3179 			ppr = rpr->pr_parent;
3180 			prison_proc_free_not_last(ppr);
3181 			prison_free_not_last(ppr);
3182 			for (; ppr != NULL; ppr = ppr->pr_parent)
3183 				ppr->pr_childcount--;
3184 		}
3185 		mtx_unlock(&cpr->pr_mtx);
3186 	}
3187 	if (rpr != NULL)
3188 		LIST_REMOVE(rpr, pr_sibling);
3189 
3190 	prison_cleanup(pr);
3191 	mtx_lock(&pr->pr_mtx);
3192 	if (pr->pr_flags & PR_PERSIST) {
3193 		pr->pr_flags &= ~PR_PERSIST;
3194 		prison_proc_free_not_last(pr);
3195 		prison_free_not_last(pr);
3196 	}
3197 	(void)refcount_release(&pr->pr_uref);
3198 }
3199 
3200 /*
3201  * Given the current locking state in the flags, make sure allprison_lock
3202  * is held exclusive, and the prison is locked.  Return flags indicating
3203  * the new state.
3204  */
3205 static int
3206 prison_lock_xlock(struct prison *pr, int flags)
3207 {
3208 
3209 	if (!(flags & PD_LIST_XLOCKED)) {
3210 		/*
3211 		 * Get allprison_lock, which may be an upgrade,
3212 		 * and may require unlocking the prison.
3213 		 */
3214 		if (flags & PD_LOCKED) {
3215 			mtx_unlock(&pr->pr_mtx);
3216 			flags &= ~PD_LOCKED;
3217 		}
3218 		if (flags & PD_LIST_SLOCKED) {
3219 			if (!sx_try_upgrade(&allprison_lock)) {
3220 				sx_sunlock(&allprison_lock);
3221 				sx_xlock(&allprison_lock);
3222 			}
3223 			flags &= ~PD_LIST_SLOCKED;
3224 		} else
3225 			sx_xlock(&allprison_lock);
3226 		flags |= PD_LIST_XLOCKED;
3227 	}
3228 	if (!(flags & PD_LOCKED)) {
3229 		/* Lock the prison mutex. */
3230 		mtx_lock(&pr->pr_mtx);
3231 		flags |= PD_LOCKED;
3232 	}
3233 	return flags;
3234 }
3235 
3236 /*
3237  * Release a prison's resources when it starts dying (when the last user
3238  * reference is dropped, or when it is killed).
3239  */
3240 static void
3241 prison_cleanup(struct prison *pr)
3242 {
3243 	sx_assert(&allprison_lock, SA_XLOCKED);
3244 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3245 	shm_remove_prison(pr);
3246 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3247 }
3248 
3249 /*
3250  * Set or clear a permission bit in the pr_allow field, passing restrictions
3251  * (cleared permission) down to child jails.
3252  */
3253 void
3254 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3255 {
3256 	struct prison *pr;
3257 
3258 	pr = cred->cr_prison;
3259 	sx_slock(&allprison_lock);
3260 	mtx_lock(&pr->pr_mtx);
3261 	prison_set_allow_locked(pr, flag, enable);
3262 	mtx_unlock(&pr->pr_mtx);
3263 	sx_sunlock(&allprison_lock);
3264 }
3265 
3266 static void
3267 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3268 {
3269 	struct prison *cpr;
3270 	int descend;
3271 
3272 	if (enable != 0)
3273 		pr->pr_allow |= flag;
3274 	else {
3275 		pr->pr_allow &= ~flag;
3276 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3277 			cpr->pr_allow &= ~flag;
3278 	}
3279 }
3280 
3281 /*
3282  * Check if a jail supports the given address family.
3283  *
3284  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3285  * if not.
3286  */
3287 int
3288 prison_check_af(struct ucred *cred, int af)
3289 {
3290 	struct prison *pr;
3291 	int error;
3292 
3293 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3294 
3295 	pr = cred->cr_prison;
3296 #ifdef VIMAGE
3297 	/* Prisons with their own network stack are not limited. */
3298 	if (prison_owns_vnet(cred))
3299 		return (0);
3300 #endif
3301 
3302 	error = 0;
3303 	switch (af)
3304 	{
3305 #ifdef INET
3306 	case AF_INET:
3307 		if (pr->pr_flags & PR_IP4)
3308 		{
3309 			mtx_lock(&pr->pr_mtx);
3310 			if ((pr->pr_flags & PR_IP4) &&
3311 			    pr->pr_addrs[PR_INET] == NULL)
3312 				error = EAFNOSUPPORT;
3313 			mtx_unlock(&pr->pr_mtx);
3314 		}
3315 		break;
3316 #endif
3317 #ifdef INET6
3318 	case AF_INET6:
3319 		if (pr->pr_flags & PR_IP6)
3320 		{
3321 			mtx_lock(&pr->pr_mtx);
3322 			if ((pr->pr_flags & PR_IP6) &&
3323 			    pr->pr_addrs[PR_INET6] == NULL)
3324 				error = EAFNOSUPPORT;
3325 			mtx_unlock(&pr->pr_mtx);
3326 		}
3327 		break;
3328 #endif
3329 	case AF_LOCAL:
3330 	case AF_ROUTE:
3331 		break;
3332 	default:
3333 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3334 			error = EAFNOSUPPORT;
3335 	}
3336 	return (error);
3337 }
3338 
3339 /*
3340  * Check if given address belongs to the jail referenced by cred (wrapper to
3341  * prison_check_ip[46]).
3342  *
3343  * Returns 0 if jail doesn't restrict the address family or if address belongs
3344  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3345  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3346  */
3347 int
3348 prison_if(struct ucred *cred, const struct sockaddr *sa)
3349 {
3350 #ifdef INET
3351 	const struct sockaddr_in *sai;
3352 #endif
3353 #ifdef INET6
3354 	const struct sockaddr_in6 *sai6;
3355 #endif
3356 	int error;
3357 
3358 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3359 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3360 
3361 #ifdef VIMAGE
3362 	if (prison_owns_vnet(cred))
3363 		return (0);
3364 #endif
3365 
3366 	error = 0;
3367 	switch (sa->sa_family)
3368 	{
3369 #ifdef INET
3370 	case AF_INET:
3371 		sai = (const struct sockaddr_in *)sa;
3372 		error = prison_check_ip4(cred, &sai->sin_addr);
3373 		break;
3374 #endif
3375 #ifdef INET6
3376 	case AF_INET6:
3377 		sai6 = (const struct sockaddr_in6 *)sa;
3378 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3379 		break;
3380 #endif
3381 	default:
3382 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3383 			error = EAFNOSUPPORT;
3384 	}
3385 	return (error);
3386 }
3387 
3388 /*
3389  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3390  */
3391 int
3392 prison_check(struct ucred *cred1, struct ucred *cred2)
3393 {
3394 
3395 	return ((cred1->cr_prison == cred2->cr_prison ||
3396 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3397 }
3398 
3399 /*
3400  * Return 1 if p2 is a child of p1, otherwise 0.
3401  */
3402 int
3403 prison_ischild(struct prison *pr1, struct prison *pr2)
3404 {
3405 
3406 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3407 		if (pr1 == pr2)
3408 			return (1);
3409 	return (0);
3410 }
3411 
3412 /*
3413  * Return true if the prison is currently alive.  A prison is alive if it
3414  * holds user references and it isn't being removed.
3415  */
3416 bool
3417 prison_isalive(const struct prison *pr)
3418 {
3419 
3420 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3421 		return (false);
3422 	return (true);
3423 }
3424 
3425 /*
3426  * Return true if the prison is currently valid.  A prison is valid if it has
3427  * been fully created, and is not being destroyed.  Note that dying prisons
3428  * are still considered valid.  Invalid prisons won't be found under normal
3429  * circumstances, as they're only put in that state by functions that have
3430  * an exclusive hold on allprison_lock.
3431  */
3432 bool
3433 prison_isvalid(struct prison *pr)
3434 {
3435 
3436 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3437 		return (false);
3438 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3439 		return (false);
3440 	return (true);
3441 }
3442 
3443 /*
3444  * Return 1 if the passed credential is in a jail and that jail does not
3445  * have its own virtual network stack, otherwise 0.
3446  */
3447 int
3448 jailed_without_vnet(struct ucred *cred)
3449 {
3450 
3451 	if (!jailed(cred))
3452 		return (0);
3453 #ifdef VIMAGE
3454 	if (prison_owns_vnet(cred))
3455 		return (0);
3456 #endif
3457 
3458 	return (1);
3459 }
3460 
3461 /*
3462  * Return the correct hostname (domainname, et al) for the passed credential.
3463  */
3464 void
3465 getcredhostname(struct ucred *cred, char *buf, size_t size)
3466 {
3467 	struct prison *pr;
3468 
3469 	/*
3470 	 * A NULL credential can be used to shortcut to the physical
3471 	 * system's hostname.
3472 	 */
3473 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3474 	mtx_lock(&pr->pr_mtx);
3475 	strlcpy(buf, pr->pr_hostname, size);
3476 	mtx_unlock(&pr->pr_mtx);
3477 }
3478 
3479 void
3480 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3481 {
3482 
3483 	mtx_lock(&cred->cr_prison->pr_mtx);
3484 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3485 	mtx_unlock(&cred->cr_prison->pr_mtx);
3486 }
3487 
3488 void
3489 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3490 {
3491 
3492 	mtx_lock(&cred->cr_prison->pr_mtx);
3493 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3494 	mtx_unlock(&cred->cr_prison->pr_mtx);
3495 }
3496 
3497 void
3498 getcredhostid(struct ucred *cred, unsigned long *hostid)
3499 {
3500 
3501 	mtx_lock(&cred->cr_prison->pr_mtx);
3502 	*hostid = cred->cr_prison->pr_hostid;
3503 	mtx_unlock(&cred->cr_prison->pr_mtx);
3504 }
3505 
3506 void
3507 getjailname(struct ucred *cred, char *name, size_t len)
3508 {
3509 
3510 	mtx_lock(&cred->cr_prison->pr_mtx);
3511 	strlcpy(name, cred->cr_prison->pr_name, len);
3512 	mtx_unlock(&cred->cr_prison->pr_mtx);
3513 }
3514 
3515 #ifdef VIMAGE
3516 /*
3517  * Determine whether the prison represented by cred owns
3518  * its vnet rather than having it inherited.
3519  *
3520  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3521  */
3522 int
3523 prison_owns_vnet(struct ucred *cred)
3524 {
3525 
3526 	/*
3527 	 * vnets cannot be added/removed after jail creation,
3528 	 * so no need to lock here.
3529 	 */
3530 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3531 }
3532 #endif
3533 
3534 /*
3535  * Determine whether the subject represented by cred can "see"
3536  * status of a mount point.
3537  * Returns: 0 for permitted, ENOENT otherwise.
3538  * XXX: This function should be called cr_canseemount() and should be
3539  *      placed in kern_prot.c.
3540  */
3541 int
3542 prison_canseemount(struct ucred *cred, struct mount *mp)
3543 {
3544 	struct prison *pr;
3545 	struct statfs *sp;
3546 	size_t len;
3547 
3548 	pr = cred->cr_prison;
3549 	if (pr->pr_enforce_statfs == 0)
3550 		return (0);
3551 	if (pr->pr_root->v_mount == mp)
3552 		return (0);
3553 	if (pr->pr_enforce_statfs == 2)
3554 		return (ENOENT);
3555 	/*
3556 	 * If jail's chroot directory is set to "/" we should be able to see
3557 	 * all mount-points from inside a jail.
3558 	 * This is ugly check, but this is the only situation when jail's
3559 	 * directory ends with '/'.
3560 	 */
3561 	if (strcmp(pr->pr_path, "/") == 0)
3562 		return (0);
3563 	len = strlen(pr->pr_path);
3564 	sp = &mp->mnt_stat;
3565 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3566 		return (ENOENT);
3567 	/*
3568 	 * Be sure that we don't have situation where jail's root directory
3569 	 * is "/some/path" and mount point is "/some/pathpath".
3570 	 */
3571 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3572 		return (ENOENT);
3573 	return (0);
3574 }
3575 
3576 void
3577 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3578 {
3579 	char jpath[MAXPATHLEN];
3580 	struct prison *pr;
3581 	size_t len;
3582 
3583 	pr = cred->cr_prison;
3584 	if (pr->pr_enforce_statfs == 0)
3585 		return;
3586 	if (prison_canseemount(cred, mp) != 0) {
3587 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3588 		strlcpy(sp->f_mntonname, "[restricted]",
3589 		    sizeof(sp->f_mntonname));
3590 		return;
3591 	}
3592 	if (pr->pr_root->v_mount == mp) {
3593 		/*
3594 		 * Clear current buffer data, so we are sure nothing from
3595 		 * the valid path left there.
3596 		 */
3597 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3598 		*sp->f_mntonname = '/';
3599 		return;
3600 	}
3601 	/*
3602 	 * If jail's chroot directory is set to "/" we should be able to see
3603 	 * all mount-points from inside a jail.
3604 	 */
3605 	if (strcmp(pr->pr_path, "/") == 0)
3606 		return;
3607 	len = strlen(pr->pr_path);
3608 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3609 	/*
3610 	 * Clear current buffer data, so we are sure nothing from
3611 	 * the valid path left there.
3612 	 */
3613 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3614 	if (*jpath == '\0') {
3615 		/* Should never happen. */
3616 		*sp->f_mntonname = '/';
3617 	} else {
3618 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3619 	}
3620 }
3621 
3622 /*
3623  * Check with permission for a specific privilege is granted within jail.  We
3624  * have a specific list of accepted privileges; the rest are denied.
3625  */
3626 int
3627 prison_priv_check(struct ucred *cred, int priv)
3628 {
3629 	struct prison *pr;
3630 	int error;
3631 
3632 	/*
3633 	 * Some policies have custom handlers. This routine should not be
3634 	 * called for them. See priv_check_cred().
3635 	 */
3636 	switch (priv) {
3637 	case PRIV_VFS_LOOKUP:
3638 	case PRIV_VFS_GENERATION:
3639 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3640 		    "called for %d\n", priv));
3641 	}
3642 
3643 	if (!jailed(cred))
3644 		return (0);
3645 
3646 #ifdef VIMAGE
3647 	/*
3648 	 * Privileges specific to prisons with a virtual network stack.
3649 	 * There might be a duplicate entry here in case the privilege
3650 	 * is only granted conditionally in the legacy jail case.
3651 	 */
3652 	switch (priv) {
3653 #ifdef notyet
3654 		/*
3655 		 * NFS-specific privileges.
3656 		 */
3657 	case PRIV_NFS_DAEMON:
3658 	case PRIV_NFS_LOCKD:
3659 #endif
3660 		/*
3661 		 * Network stack privileges.
3662 		 */
3663 	case PRIV_NET_BRIDGE:
3664 	case PRIV_NET_GRE:
3665 	case PRIV_NET_BPF:
3666 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3667 	case PRIV_NET_ROUTE:
3668 	case PRIV_NET_TAP:
3669 	case PRIV_NET_SETIFMTU:
3670 	case PRIV_NET_SETIFFLAGS:
3671 	case PRIV_NET_SETIFCAP:
3672 	case PRIV_NET_SETIFDESCR:
3673 	case PRIV_NET_SETIFNAME	:
3674 	case PRIV_NET_SETIFMETRIC:
3675 	case PRIV_NET_SETIFPHYS:
3676 	case PRIV_NET_SETIFMAC:
3677 	case PRIV_NET_SETLANPCP:
3678 	case PRIV_NET_ADDMULTI:
3679 	case PRIV_NET_DELMULTI:
3680 	case PRIV_NET_HWIOCTL:
3681 	case PRIV_NET_SETLLADDR:
3682 	case PRIV_NET_ADDIFGROUP:
3683 	case PRIV_NET_DELIFGROUP:
3684 	case PRIV_NET_IFCREATE:
3685 	case PRIV_NET_IFDESTROY:
3686 	case PRIV_NET_ADDIFADDR:
3687 	case PRIV_NET_DELIFADDR:
3688 	case PRIV_NET_LAGG:
3689 	case PRIV_NET_GIF:
3690 	case PRIV_NET_SETIFVNET:
3691 	case PRIV_NET_SETIFFIB:
3692 	case PRIV_NET_OVPN:
3693 
3694 		/*
3695 		 * 802.11-related privileges.
3696 		 */
3697 	case PRIV_NET80211_VAP_GETKEY:
3698 	case PRIV_NET80211_VAP_MANAGE:
3699 
3700 #ifdef notyet
3701 		/*
3702 		 * ATM privileges.
3703 		 */
3704 	case PRIV_NETATM_CFG:
3705 	case PRIV_NETATM_ADD:
3706 	case PRIV_NETATM_DEL:
3707 	case PRIV_NETATM_SET:
3708 
3709 		/*
3710 		 * Bluetooth privileges.
3711 		 */
3712 	case PRIV_NETBLUETOOTH_RAW:
3713 #endif
3714 
3715 		/*
3716 		 * Netgraph and netgraph module privileges.
3717 		 */
3718 	case PRIV_NETGRAPH_CONTROL:
3719 #ifdef notyet
3720 	case PRIV_NETGRAPH_TTY:
3721 #endif
3722 
3723 		/*
3724 		 * IPv4 and IPv6 privileges.
3725 		 */
3726 	case PRIV_NETINET_IPFW:
3727 	case PRIV_NETINET_DIVERT:
3728 	case PRIV_NETINET_PF:
3729 	case PRIV_NETINET_DUMMYNET:
3730 	case PRIV_NETINET_CARP:
3731 	case PRIV_NETINET_MROUTE:
3732 	case PRIV_NETINET_RAW:
3733 	case PRIV_NETINET_ADDRCTRL6:
3734 	case PRIV_NETINET_ND6:
3735 	case PRIV_NETINET_SCOPE6:
3736 	case PRIV_NETINET_ALIFETIME6:
3737 	case PRIV_NETINET_IPSEC:
3738 	case PRIV_NETINET_BINDANY:
3739 
3740 #ifdef notyet
3741 		/*
3742 		 * NCP privileges.
3743 		 */
3744 	case PRIV_NETNCP:
3745 
3746 		/*
3747 		 * SMB privileges.
3748 		 */
3749 	case PRIV_NETSMB:
3750 #endif
3751 
3752 	/*
3753 	 * No default: or deny here.
3754 	 * In case of no permit fall through to next switch().
3755 	 */
3756 		if (cred->cr_prison->pr_flags & PR_VNET)
3757 			return (0);
3758 	}
3759 #endif /* VIMAGE */
3760 
3761 	switch (priv) {
3762 		/*
3763 		 * Allow ktrace privileges for root in jail.
3764 		 */
3765 	case PRIV_KTRACE:
3766 
3767 #if 0
3768 		/*
3769 		 * Allow jailed processes to configure audit identity and
3770 		 * submit audit records (login, etc).  In the future we may
3771 		 * want to further refine the relationship between audit and
3772 		 * jail.
3773 		 */
3774 	case PRIV_AUDIT_GETAUDIT:
3775 	case PRIV_AUDIT_SETAUDIT:
3776 	case PRIV_AUDIT_SUBMIT:
3777 #endif
3778 
3779 		/*
3780 		 * Allow jailed processes to manipulate process UNIX
3781 		 * credentials in any way they see fit.
3782 		 */
3783 	case PRIV_CRED_SETUID:
3784 	case PRIV_CRED_SETEUID:
3785 	case PRIV_CRED_SETGID:
3786 	case PRIV_CRED_SETEGID:
3787 	case PRIV_CRED_SETGROUPS:
3788 	case PRIV_CRED_SETREUID:
3789 	case PRIV_CRED_SETREGID:
3790 	case PRIV_CRED_SETRESUID:
3791 	case PRIV_CRED_SETRESGID:
3792 
3793 		/*
3794 		 * Jail implements visibility constraints already, so allow
3795 		 * jailed root to override uid/gid-based constraints.
3796 		 */
3797 	case PRIV_SEEOTHERGIDS:
3798 	case PRIV_SEEOTHERUIDS:
3799 
3800 		/*
3801 		 * Jail implements inter-process debugging limits already, so
3802 		 * allow jailed root various debugging privileges.
3803 		 */
3804 	case PRIV_DEBUG_DIFFCRED:
3805 	case PRIV_DEBUG_SUGID:
3806 	case PRIV_DEBUG_UNPRIV:
3807 
3808 		/*
3809 		 * Allow jail to set various resource limits and login
3810 		 * properties, and for now, exceed process resource limits.
3811 		 */
3812 	case PRIV_PROC_LIMIT:
3813 	case PRIV_PROC_SETLOGIN:
3814 	case PRIV_PROC_SETRLIMIT:
3815 
3816 		/*
3817 		 * System V and POSIX IPC privileges are granted in jail.
3818 		 */
3819 	case PRIV_IPC_READ:
3820 	case PRIV_IPC_WRITE:
3821 	case PRIV_IPC_ADMIN:
3822 	case PRIV_IPC_MSGSIZE:
3823 	case PRIV_MQ_ADMIN:
3824 
3825 		/*
3826 		 * Jail operations within a jail work on child jails.
3827 		 */
3828 	case PRIV_JAIL_ATTACH:
3829 	case PRIV_JAIL_SET:
3830 	case PRIV_JAIL_REMOVE:
3831 
3832 		/*
3833 		 * Jail implements its own inter-process limits, so allow
3834 		 * root processes in jail to change scheduling on other
3835 		 * processes in the same jail.  Likewise for signalling.
3836 		 */
3837 	case PRIV_SCHED_DIFFCRED:
3838 	case PRIV_SCHED_CPUSET:
3839 	case PRIV_SIGNAL_DIFFCRED:
3840 	case PRIV_SIGNAL_SUGID:
3841 
3842 		/*
3843 		 * Allow jailed processes to write to sysctls marked as jail
3844 		 * writable.
3845 		 */
3846 	case PRIV_SYSCTL_WRITEJAIL:
3847 
3848 		/*
3849 		 * Allow root in jail to manage a variety of quota
3850 		 * properties.  These should likely be conditional on a
3851 		 * configuration option.
3852 		 */
3853 	case PRIV_VFS_GETQUOTA:
3854 	case PRIV_VFS_SETQUOTA:
3855 
3856 		/*
3857 		 * Since Jail relies on chroot() to implement file system
3858 		 * protections, grant many VFS privileges to root in jail.
3859 		 * Be careful to exclude mount-related and NFS-related
3860 		 * privileges.
3861 		 */
3862 	case PRIV_VFS_READ:
3863 	case PRIV_VFS_WRITE:
3864 	case PRIV_VFS_ADMIN:
3865 	case PRIV_VFS_EXEC:
3866 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3867 	case PRIV_VFS_CHFLAGS_DEV:
3868 	case PRIV_VFS_CHOWN:
3869 	case PRIV_VFS_CHROOT:
3870 	case PRIV_VFS_RETAINSUGID:
3871 	case PRIV_VFS_FCHROOT:
3872 	case PRIV_VFS_LINK:
3873 	case PRIV_VFS_SETGID:
3874 	case PRIV_VFS_STAT:
3875 	case PRIV_VFS_STICKYFILE:
3876 
3877 		/*
3878 		 * As in the non-jail case, non-root users are expected to be
3879 		 * able to read kernel/physical memory (provided /dev/[k]mem
3880 		 * exists in the jail and they have permission to access it).
3881 		 */
3882 	case PRIV_KMEM_READ:
3883 		return (0);
3884 
3885 		/*
3886 		 * Depending on the global setting, allow privilege of
3887 		 * setting system flags.
3888 		 */
3889 	case PRIV_VFS_SYSFLAGS:
3890 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3891 			return (0);
3892 		else
3893 			return (EPERM);
3894 
3895 		/*
3896 		 * Depending on the global setting, allow privilege of
3897 		 * mounting/unmounting file systems.
3898 		 */
3899 	case PRIV_VFS_MOUNT:
3900 	case PRIV_VFS_UNMOUNT:
3901 	case PRIV_VFS_MOUNT_NONUSER:
3902 	case PRIV_VFS_MOUNT_OWNER:
3903 		pr = cred->cr_prison;
3904 		prison_lock(pr);
3905 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3906 			error = 0;
3907 		else
3908 			error = EPERM;
3909 		prison_unlock(pr);
3910 		return (error);
3911 
3912 		/*
3913 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3914 		 * policy.  priv_check_cred will not specifically allow it, and
3915 		 * we may want a MAC policy to allow it.
3916 		 */
3917 	case PRIV_VFS_READ_DIR:
3918 		return (0);
3919 
3920 		/*
3921 		 * Conditionnaly allow locking (unlocking) physical pages
3922 		 * in memory.
3923 		 */
3924 	case PRIV_VM_MLOCK:
3925 	case PRIV_VM_MUNLOCK:
3926 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3927 			return (0);
3928 		else
3929 			return (EPERM);
3930 
3931 		/*
3932 		 * Conditionally allow jailed root to bind reserved ports.
3933 		 */
3934 	case PRIV_NETINET_RESERVEDPORT:
3935 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3936 			return (0);
3937 		else
3938 			return (EPERM);
3939 
3940 		/*
3941 		 * Allow jailed root to reuse in-use ports.
3942 		 */
3943 	case PRIV_NETINET_REUSEPORT:
3944 		return (0);
3945 
3946 		/*
3947 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3948 		 */
3949 	case PRIV_NETINET_SETHDROPTS:
3950 		return (0);
3951 
3952 		/*
3953 		 * Conditionally allow creating raw sockets in jail.
3954 		 */
3955 	case PRIV_NETINET_RAW:
3956 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3957 			return (0);
3958 		else
3959 			return (EPERM);
3960 
3961 		/*
3962 		 * Since jail implements its own visibility limits on netstat
3963 		 * sysctls, allow getcred.  This allows identd to work in
3964 		 * jail.
3965 		 */
3966 	case PRIV_NETINET_GETCRED:
3967 		return (0);
3968 
3969 		/*
3970 		 * Allow jailed root to set loginclass.
3971 		 */
3972 	case PRIV_PROC_SETLOGINCLASS:
3973 		return (0);
3974 
3975 		/*
3976 		 * Do not allow a process inside a jail to read the kernel
3977 		 * message buffer unless explicitly permitted.
3978 		 */
3979 	case PRIV_MSGBUF:
3980 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3981 			return (0);
3982 		return (EPERM);
3983 
3984 	default:
3985 		/*
3986 		 * In all remaining cases, deny the privilege request.  This
3987 		 * includes almost all network privileges, many system
3988 		 * configuration privileges.
3989 		 */
3990 		return (EPERM);
3991 	}
3992 }
3993 
3994 /*
3995  * Return the part of pr2's name that is relative to pr1, or the whole name
3996  * if it does not directly follow.
3997  */
3998 
3999 char *
4000 prison_name(struct prison *pr1, struct prison *pr2)
4001 {
4002 	char *name;
4003 
4004 	/* Jails see themselves as "0" (if they see themselves at all). */
4005 	if (pr1 == pr2)
4006 		return "0";
4007 	name = pr2->pr_name;
4008 	if (prison_ischild(pr1, pr2)) {
4009 		/*
4010 		 * pr1 isn't locked (and allprison_lock may not be either)
4011 		 * so its length can't be counted on.  But the number of dots
4012 		 * can be counted on - and counted.
4013 		 */
4014 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4015 			name = strchr(name, '.') + 1;
4016 	}
4017 	return (name);
4018 }
4019 
4020 /*
4021  * Return the part of pr2's path that is relative to pr1, or the whole path
4022  * if it does not directly follow.
4023  */
4024 static char *
4025 prison_path(struct prison *pr1, struct prison *pr2)
4026 {
4027 	char *path1, *path2;
4028 	int len1;
4029 
4030 	path1 = pr1->pr_path;
4031 	path2 = pr2->pr_path;
4032 	if (!strcmp(path1, "/"))
4033 		return (path2);
4034 	len1 = strlen(path1);
4035 	if (strncmp(path1, path2, len1))
4036 		return (path2);
4037 	if (path2[len1] == '\0')
4038 		return "/";
4039 	if (path2[len1] == '/')
4040 		return (path2 + len1);
4041 	return (path2);
4042 }
4043 
4044 /*
4045  * Jail-related sysctls.
4046  */
4047 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4048     "Jails");
4049 
4050 #if defined(INET) || defined(INET6)
4051 /*
4052  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4053  * sysctl_jail_list() helper.
4054  */
4055 static void
4056 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4057 {
4058 	const size_t size = pr_families[af].size;
4059 
4060  again:
4061 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4062 	if (pr->pr_addrs[af] != NULL) {
4063 		if (*len < pr->pr_addrs[af]->ips) {
4064 			*len = pr->pr_addrs[af]->ips;
4065 			mtx_unlock(&pr->pr_mtx);
4066 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4067 			mtx_lock(&pr->pr_mtx);
4068 			goto again;
4069 		}
4070 		bcopy(pr->pr_addrs[af] + 1, *out, pr->pr_addrs[af]->ips * size);
4071 	}
4072 }
4073 #endif
4074 
4075 static int
4076 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4077 {
4078 	struct xprison *xp;
4079 	struct prison *pr, *cpr;
4080 #ifdef INET
4081 	struct in_addr *ip4 = NULL;
4082 	int ip4s = 0;
4083 #endif
4084 #ifdef INET6
4085 	struct in6_addr *ip6 = NULL;
4086 	int ip6s = 0;
4087 #endif
4088 	int descend, error;
4089 
4090 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4091 	pr = req->td->td_ucred->cr_prison;
4092 	error = 0;
4093 	sx_slock(&allprison_lock);
4094 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4095 		mtx_lock(&cpr->pr_mtx);
4096 #ifdef INET
4097 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4098 #endif
4099 #ifdef INET6
4100 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4101 #endif
4102 		bzero(xp, sizeof(*xp));
4103 		xp->pr_version = XPRISON_VERSION;
4104 		xp->pr_id = cpr->pr_id;
4105 		xp->pr_state = cpr->pr_state;
4106 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4107 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4108 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4109 #ifdef INET
4110 		xp->pr_ip4s = ip4s;
4111 #endif
4112 #ifdef INET6
4113 		xp->pr_ip6s = ip6s;
4114 #endif
4115 		mtx_unlock(&cpr->pr_mtx);
4116 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4117 		if (error)
4118 			break;
4119 #ifdef INET
4120 		if (xp->pr_ip4s > 0) {
4121 			error = SYSCTL_OUT(req, ip4,
4122 			    xp->pr_ip4s * sizeof(struct in_addr));
4123 			if (error)
4124 				break;
4125 		}
4126 #endif
4127 #ifdef INET6
4128 		if (xp->pr_ip6s > 0) {
4129 			error = SYSCTL_OUT(req, ip6,
4130 			    xp->pr_ip6s * sizeof(struct in6_addr));
4131 			if (error)
4132 				break;
4133 		}
4134 #endif
4135 	}
4136 	sx_sunlock(&allprison_lock);
4137 	free(xp, M_TEMP);
4138 #ifdef INET
4139 	free(ip4, M_TEMP);
4140 #endif
4141 #ifdef INET6
4142 	free(ip6, M_TEMP);
4143 #endif
4144 	return (error);
4145 }
4146 
4147 SYSCTL_OID(_security_jail, OID_AUTO, list,
4148     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4149     sysctl_jail_list, "S", "List of active jails");
4150 
4151 static int
4152 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4153 {
4154 	int error, injail;
4155 
4156 	injail = jailed(req->td->td_ucred);
4157 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4158 
4159 	return (error);
4160 }
4161 
4162 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4163     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4164     sysctl_jail_jailed, "I", "Process in jail?");
4165 
4166 static int
4167 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4168 {
4169 	int error, havevnet;
4170 #ifdef VIMAGE
4171 	struct ucred *cred = req->td->td_ucred;
4172 
4173 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4174 #else
4175 	havevnet = 0;
4176 #endif
4177 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4178 
4179 	return (error);
4180 }
4181 
4182 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4183     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4184     sysctl_jail_vnet, "I", "Jail owns vnet?");
4185 
4186 #if defined(INET) || defined(INET6)
4187 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4188     &jail_max_af_ips, 0,
4189     "Number of IP addresses a jail may have at most per address family (deprecated)");
4190 #endif
4191 
4192 /*
4193  * Default parameters for jail(2) compatibility.  For historical reasons,
4194  * the sysctl names have varying similarity to the parameter names.  Prisons
4195  * just see their own parameters, and can't change them.
4196  */
4197 static int
4198 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4199 {
4200 	int error, i;
4201 
4202 	/* Get the current flag value, and convert it to a boolean. */
4203 	if (req->td->td_ucred->cr_prison == &prison0) {
4204 		mtx_lock(&prison0.pr_mtx);
4205 		i = (jail_default_allow & arg2) != 0;
4206 		mtx_unlock(&prison0.pr_mtx);
4207 	} else
4208 		i = prison_allow(req->td->td_ucred, arg2);
4209 
4210 	if (arg1 != NULL)
4211 		i = !i;
4212 	error = sysctl_handle_int(oidp, &i, 0, req);
4213 	if (error || !req->newptr)
4214 		return (error);
4215 	i = i ? arg2 : 0;
4216 	if (arg1 != NULL)
4217 		i ^= arg2;
4218 	/*
4219 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4220 	 * for writing.
4221 	 */
4222 	mtx_lock(&prison0.pr_mtx);
4223 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4224 	mtx_unlock(&prison0.pr_mtx);
4225 	return (0);
4226 }
4227 
4228 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4229     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4230     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4231     "Processes in jail can set their hostnames (deprecated)");
4232 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4233     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4234     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4235     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4236 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4237     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4238     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4239     "Processes in jail can use System V IPC primitives (deprecated)");
4240 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4241     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4242     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4243     "Prison root can create raw sockets (deprecated)");
4244 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4245     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4246     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4247     "Processes in jail can alter system file flags (deprecated)");
4248 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4249     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4250     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4251     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4252 
4253 static int
4254 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4255 {
4256 	struct prison *pr;
4257 	int level, error;
4258 
4259 	pr = req->td->td_ucred->cr_prison;
4260 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4261 	error = sysctl_handle_int(oidp, &level, 0, req);
4262 	if (error || !req->newptr)
4263 		return (error);
4264 	*(int *)arg1 = level;
4265 	return (0);
4266 }
4267 
4268 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4269     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4270     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4271     sysctl_jail_default_level, "I",
4272     "Processes in jail cannot see all mounted file systems (deprecated)");
4273 
4274 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4275     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4276     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4277     sysctl_jail_default_level, "I",
4278     "Ruleset for the devfs filesystem in jail (deprecated)");
4279 
4280 /*
4281  * Nodes to describe jail parameters.  Maximum length of string parameters
4282  * is returned in the string itself, and the other parameters exist merely
4283  * to make themselves and their types known.
4284  */
4285 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4286     "Jail parameters");
4287 
4288 int
4289 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4290 {
4291 	int i;
4292 	long l;
4293 	size_t s;
4294 	char numbuf[12];
4295 
4296 	switch (oidp->oid_kind & CTLTYPE)
4297 	{
4298 	case CTLTYPE_LONG:
4299 	case CTLTYPE_ULONG:
4300 		l = 0;
4301 #ifdef SCTL_MASK32
4302 		if (!(req->flags & SCTL_MASK32))
4303 #endif
4304 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4305 	case CTLTYPE_INT:
4306 	case CTLTYPE_UINT:
4307 		i = 0;
4308 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4309 	case CTLTYPE_STRING:
4310 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4311 		return
4312 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4313 	case CTLTYPE_STRUCT:
4314 		s = (size_t)arg2;
4315 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4316 	}
4317 	return (0);
4318 }
4319 
4320 /*
4321  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4322  * jail creation time but cannot be changed in an existing jail.
4323  */
4324 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4325 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4326 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4327 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4328 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4329     "I", "Jail secure level");
4330 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4331     "Jail value for kern.osreldate and uname -K");
4332 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4333     "Jail value for kern.osrelease and uname -r");
4334 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4335     "I", "Jail cannot see all mounted file systems");
4336 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4337     "I", "Ruleset for in-jail devfs mounts");
4338 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4339     "B", "Jail persistence");
4340 #ifdef VIMAGE
4341 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4342     "E,jailsys", "Virtual network stack");
4343 #endif
4344 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4345     "B", "Jail is in the process of shutting down");
4346 
4347 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4348 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4349     "I", "Current number of child jails");
4350 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4351     "I", "Maximum number of child jails");
4352 
4353 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4354 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4355     "Jail hostname");
4356 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4357     "Jail NIS domainname");
4358 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4359     "Jail host UUID");
4360 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4361     "LU", "Jail host ID");
4362 
4363 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4364 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4365 
4366 #ifdef INET
4367 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4368     "Jail IPv4 address virtualization");
4369 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4370     "S,in_addr,a", "Jail IPv4 addresses");
4371 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4372     "B", "Do (not) use IPv4 source address selection rather than the "
4373     "primary jail IPv4 address.");
4374 #endif
4375 #ifdef INET6
4376 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4377     "Jail IPv6 address virtualization");
4378 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4379     "S,in6_addr,a", "Jail IPv6 addresses");
4380 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4381     "B", "Do (not) use IPv6 source address selection rather than the "
4382     "primary jail IPv6 address.");
4383 #endif
4384 
4385 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4386 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4387     "B", "Jail may set hostname");
4388 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4389     "B", "Jail may use SYSV IPC");
4390 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4391     "B", "Jail may create raw sockets");
4392 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4393     "B", "Jail may alter system file flags");
4394 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4395     "B", "Jail may set file quotas");
4396 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4397     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4398 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4399     "B", "Jail may lock (unlock) physical pages in memory");
4400 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4401     "B", "Jail may bind sockets to reserved ports");
4402 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4403     "B", "Jail may read the kernel message buffer");
4404 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4405     "B", "Unprivileged processes may use process debugging facilities");
4406 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4407     "B", "Processes in jail with uid 0 have privilege");
4408 
4409 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4410 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4411     "B", "Jail may mount/unmount jail-friendly file systems in general");
4412 
4413 /*
4414  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4415  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4416  * not created.
4417  */
4418 unsigned
4419 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4420     const char *descr)
4421 {
4422 	struct bool_flags *bf;
4423 	struct sysctl_oid *parent;
4424 	char *allow_name, *allow_noname, *allowed;
4425 #ifndef NO_SYSCTL_DESCR
4426 	char *descr_deprecated;
4427 #endif
4428 	u_int allow_flag;
4429 
4430 	if (prefix
4431 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4432 		< 0 ||
4433 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4434 		< 0
4435 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4436 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4437 		free(allow_name, M_PRISON);
4438 		return 0;
4439 	}
4440 
4441 	/*
4442 	 * See if this parameter has already beed added, i.e. a module was
4443 	 * previously loaded/unloaded.
4444 	 */
4445 	mtx_lock(&prison0.pr_mtx);
4446 	for (bf = pr_flag_allow;
4447 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4448 		atomic_load_int(&bf->flag) != 0;
4449 	     bf++) {
4450 		if (strcmp(bf->name, allow_name) == 0) {
4451 			allow_flag = bf->flag;
4452 			goto no_add;
4453 		}
4454 	}
4455 
4456 	/*
4457 	 * Find a free bit in pr_allow_all, failing if there are none
4458 	 * (which shouldn't happen as long as we keep track of how many
4459 	 * potential dynamic flags exist).
4460 	 */
4461 	for (allow_flag = 1;; allow_flag <<= 1) {
4462 		if (allow_flag == 0)
4463 			goto no_add;
4464 		if ((pr_allow_all & allow_flag) == 0)
4465 			break;
4466 	}
4467 
4468 	/* Note the parameter in the next open slot in pr_flag_allow. */
4469 	for (bf = pr_flag_allow; ; bf++) {
4470 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4471 			/* This should never happen, but is not fatal. */
4472 			allow_flag = 0;
4473 			goto no_add;
4474 		}
4475 		if (atomic_load_int(&bf->flag) == 0)
4476 			break;
4477 	}
4478 	bf->name = allow_name;
4479 	bf->noname = allow_noname;
4480 	pr_allow_all |= allow_flag;
4481 	/*
4482 	 * prison0 always has permission for the new parameter.
4483 	 * Other jails must have it granted to them.
4484 	 */
4485 	prison0.pr_allow |= allow_flag;
4486 	/* The flag indicates a valid entry, so make sure it is set last. */
4487 	atomic_store_rel_int(&bf->flag, allow_flag);
4488 	mtx_unlock(&prison0.pr_mtx);
4489 
4490 	/*
4491 	 * Create sysctls for the paramter, and the back-compat global
4492 	 * permission.
4493 	 */
4494 	parent = prefix
4495 	    ? SYSCTL_ADD_NODE(NULL,
4496 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4497 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4498 	    : &sysctl___security_jail_param_allow;
4499 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4500 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4501 	    NULL, 0, sysctl_jail_param, "B", descr);
4502 	if ((prefix
4503 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4504 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4505 #ifndef NO_SYSCTL_DESCR
4506 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4507 		    descr);
4508 #endif
4509 		(void)SYSCTL_ADD_PROC(NULL,
4510 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4511 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4512 		    sysctl_jail_default_allow, "I", descr_deprecated);
4513 #ifndef NO_SYSCTL_DESCR
4514 		free(descr_deprecated, M_TEMP);
4515 #endif
4516 		free(allowed, M_TEMP);
4517 	}
4518 	return allow_flag;
4519 
4520  no_add:
4521 	mtx_unlock(&prison0.pr_mtx);
4522 	free(allow_name, M_PRISON);
4523 	free(allow_noname, M_PRISON);
4524 	return allow_flag;
4525 }
4526 
4527 /*
4528  * The VFS system will register jail-aware filesystems here.  They each get
4529  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4530  * attempts to mount.
4531  */
4532 void
4533 prison_add_vfs(struct vfsconf *vfsp)
4534 {
4535 #ifdef NO_SYSCTL_DESCR
4536 
4537 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4538 	    NULL, NULL);
4539 #else
4540 	char *descr;
4541 
4542 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4543 	    vfsp->vfc_name);
4544 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4545 	    NULL, descr);
4546 	free(descr, M_TEMP);
4547 #endif
4548 }
4549 
4550 #ifdef RACCT
4551 void
4552 prison_racct_foreach(void (*callback)(struct racct *racct,
4553     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4554     void *arg2, void *arg3)
4555 {
4556 	struct prison_racct *prr;
4557 
4558 	ASSERT_RACCT_ENABLED();
4559 
4560 	sx_slock(&allprison_lock);
4561 	if (pre != NULL)
4562 		(pre)();
4563 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4564 		(callback)(prr->prr_racct, arg2, arg3);
4565 	if (post != NULL)
4566 		(post)();
4567 	sx_sunlock(&allprison_lock);
4568 }
4569 
4570 static struct prison_racct *
4571 prison_racct_find_locked(const char *name)
4572 {
4573 	struct prison_racct *prr;
4574 
4575 	ASSERT_RACCT_ENABLED();
4576 	sx_assert(&allprison_lock, SA_XLOCKED);
4577 
4578 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4579 		return (NULL);
4580 
4581 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4582 		if (strcmp(name, prr->prr_name) != 0)
4583 			continue;
4584 
4585 		/* Found prison_racct with a matching name? */
4586 		prison_racct_hold(prr);
4587 		return (prr);
4588 	}
4589 
4590 	/* Add new prison_racct. */
4591 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4592 	racct_create(&prr->prr_racct);
4593 
4594 	strcpy(prr->prr_name, name);
4595 	refcount_init(&prr->prr_refcount, 1);
4596 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4597 
4598 	return (prr);
4599 }
4600 
4601 struct prison_racct *
4602 prison_racct_find(const char *name)
4603 {
4604 	struct prison_racct *prr;
4605 
4606 	ASSERT_RACCT_ENABLED();
4607 
4608 	sx_xlock(&allprison_lock);
4609 	prr = prison_racct_find_locked(name);
4610 	sx_xunlock(&allprison_lock);
4611 	return (prr);
4612 }
4613 
4614 void
4615 prison_racct_hold(struct prison_racct *prr)
4616 {
4617 
4618 	ASSERT_RACCT_ENABLED();
4619 
4620 	refcount_acquire(&prr->prr_refcount);
4621 }
4622 
4623 static void
4624 prison_racct_free_locked(struct prison_racct *prr)
4625 {
4626 
4627 	ASSERT_RACCT_ENABLED();
4628 	sx_assert(&allprison_lock, SA_XLOCKED);
4629 
4630 	if (refcount_release(&prr->prr_refcount)) {
4631 		racct_destroy(&prr->prr_racct);
4632 		LIST_REMOVE(prr, prr_next);
4633 		free(prr, M_PRISON_RACCT);
4634 	}
4635 }
4636 
4637 void
4638 prison_racct_free(struct prison_racct *prr)
4639 {
4640 
4641 	ASSERT_RACCT_ENABLED();
4642 	sx_assert(&allprison_lock, SA_UNLOCKED);
4643 
4644 	if (refcount_release_if_not_last(&prr->prr_refcount))
4645 		return;
4646 
4647 	sx_xlock(&allprison_lock);
4648 	prison_racct_free_locked(prr);
4649 	sx_xunlock(&allprison_lock);
4650 }
4651 
4652 static void
4653 prison_racct_attach(struct prison *pr)
4654 {
4655 	struct prison_racct *prr;
4656 
4657 	ASSERT_RACCT_ENABLED();
4658 	sx_assert(&allprison_lock, SA_XLOCKED);
4659 
4660 	prr = prison_racct_find_locked(pr->pr_name);
4661 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4662 
4663 	pr->pr_prison_racct = prr;
4664 }
4665 
4666 /*
4667  * Handle jail renaming.  From the racct point of view, renaming means
4668  * moving from one prison_racct to another.
4669  */
4670 static void
4671 prison_racct_modify(struct prison *pr)
4672 {
4673 #ifdef RCTL
4674 	struct proc *p;
4675 	struct ucred *cred;
4676 #endif
4677 	struct prison_racct *oldprr;
4678 
4679 	ASSERT_RACCT_ENABLED();
4680 
4681 	sx_slock(&allproc_lock);
4682 	sx_xlock(&allprison_lock);
4683 
4684 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4685 		sx_xunlock(&allprison_lock);
4686 		sx_sunlock(&allproc_lock);
4687 		return;
4688 	}
4689 
4690 	oldprr = pr->pr_prison_racct;
4691 	pr->pr_prison_racct = NULL;
4692 
4693 	prison_racct_attach(pr);
4694 
4695 	/*
4696 	 * Move resource utilisation records.
4697 	 */
4698 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4699 
4700 #ifdef RCTL
4701 	/*
4702 	 * Force rctl to reattach rules to processes.
4703 	 */
4704 	FOREACH_PROC_IN_SYSTEM(p) {
4705 		PROC_LOCK(p);
4706 		cred = crhold(p->p_ucred);
4707 		PROC_UNLOCK(p);
4708 		rctl_proc_ucred_changed(p, cred);
4709 		crfree(cred);
4710 	}
4711 #endif
4712 
4713 	sx_sunlock(&allproc_lock);
4714 	prison_racct_free_locked(oldprr);
4715 	sx_xunlock(&allprison_lock);
4716 }
4717 
4718 static void
4719 prison_racct_detach(struct prison *pr)
4720 {
4721 
4722 	ASSERT_RACCT_ENABLED();
4723 	sx_assert(&allprison_lock, SA_UNLOCKED);
4724 
4725 	if (pr->pr_prison_racct == NULL)
4726 		return;
4727 	prison_racct_free(pr->pr_prison_racct);
4728 	pr->pr_prison_racct = NULL;
4729 }
4730 #endif /* RACCT */
4731 
4732 #ifdef DDB
4733 
4734 static void
4735 db_show_prison(struct prison *pr)
4736 {
4737 	struct bool_flags *bf;
4738 	struct jailsys_flags *jsf;
4739 #if defined(INET) || defined(INET6)
4740 	int ii;
4741 #endif
4742 	unsigned f;
4743 #ifdef INET
4744 	char ip4buf[INET_ADDRSTRLEN];
4745 #endif
4746 #ifdef INET6
4747 	char ip6buf[INET6_ADDRSTRLEN];
4748 #endif
4749 
4750 	db_printf("prison %p:\n", pr);
4751 	db_printf(" jid             = %d\n", pr->pr_id);
4752 	db_printf(" name            = %s\n", pr->pr_name);
4753 	db_printf(" parent          = %p\n", pr->pr_parent);
4754 	db_printf(" ref             = %d\n", pr->pr_ref);
4755 	db_printf(" uref            = %d\n", pr->pr_uref);
4756 	db_printf(" state           = %s\n",
4757 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4758 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
4759 	    "invalid");
4760 	db_printf(" path            = %s\n", pr->pr_path);
4761 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4762 	    ? pr->pr_cpuset->cs_id : -1);
4763 #ifdef VIMAGE
4764 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4765 #endif
4766 	db_printf(" root            = %p\n", pr->pr_root);
4767 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4768 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4769 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4770 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4771 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4772 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4773 	db_printf(" flags           = 0x%x", pr->pr_flags);
4774 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4775 		if (pr->pr_flags & bf->flag)
4776 			db_printf(" %s", bf->name);
4777 	for (jsf = pr_flag_jailsys;
4778 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4779 	     jsf++) {
4780 		f = pr->pr_flags & (jsf->disable | jsf->new);
4781 		db_printf(" %-16s= %s\n", jsf->name,
4782 		    (f != 0 && f == jsf->disable) ? "disable"
4783 		    : (f == jsf->new) ? "new"
4784 		    : "inherit");
4785 	}
4786 	db_printf(" allow           = 0x%x", pr->pr_allow);
4787 	for (bf = pr_flag_allow;
4788 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4789 		atomic_load_int(&bf->flag) != 0;
4790 	     bf++)
4791 		if (pr->pr_allow & bf->flag)
4792 			db_printf(" %s", bf->name);
4793 	db_printf("\n");
4794 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4795 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4796 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4797 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4798 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4799 #ifdef INET
4800 	if (pr->pr_addrs[PR_INET] != NULL) {
4801 		pr_family_t af = PR_INET;
4802 
4803 		db_printf(" ip4s            = %d\n", pr->pr_addrs[af]->ips);
4804 		for (ii = 0; ii < pr->pr_addrs[af]->ips; ii++)
4805 			db_printf(" %s %s\n",
4806 			    ii == 0 ? "ip4.addr        =" : "                 ",
4807 			    inet_ntoa_r(
4808 			    *(const struct in_addr *)PR_IP(pr, ii),
4809 			    ip4buf));
4810 	}
4811 #endif
4812 #ifdef INET6
4813 	if (pr->pr_addrs[PR_INET6] != NULL) {
4814 		pr_family_t af = PR_INET6;
4815 
4816 		db_printf(" ip6s            = %d\n", pr->pr_addrs[af]->ips);
4817 		for (ii = 0; ii < pr->pr_addrs[af]->ips; ii++)
4818 			db_printf(" %s %s\n",
4819 			    ii == 0 ? "ip6.addr        =" : "                 ",
4820 			    ip6_sprintf(ip6buf,
4821 			    (const struct in6_addr *)PR_IP(pr, ii)));
4822 	}
4823 #endif
4824 }
4825 
4826 DB_SHOW_COMMAND(prison, db_show_prison_command)
4827 {
4828 	struct prison *pr;
4829 
4830 	if (!have_addr) {
4831 		/*
4832 		 * Show all prisons in the list, and prison0 which is not
4833 		 * listed.
4834 		 */
4835 		db_show_prison(&prison0);
4836 		if (!db_pager_quit) {
4837 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4838 				db_show_prison(pr);
4839 				if (db_pager_quit)
4840 					break;
4841 			}
4842 		}
4843 		return;
4844 	}
4845 
4846 	if (addr == 0)
4847 		pr = &prison0;
4848 	else {
4849 		/* Look for a prison with the ID and with references. */
4850 		TAILQ_FOREACH(pr, &allprison, pr_list)
4851 			if (pr->pr_id == addr && pr->pr_ref > 0)
4852 				break;
4853 		if (pr == NULL)
4854 			/* Look again, without requiring a reference. */
4855 			TAILQ_FOREACH(pr, &allprison, pr_list)
4856 				if (pr->pr_id == addr)
4857 					break;
4858 		if (pr == NULL)
4859 			/* Assume address points to a valid prison. */
4860 			pr = (struct prison *)addr;
4861 	}
4862 	db_show_prison(pr);
4863 }
4864 
4865 #endif /* DDB */
4866