xref: /freebsd/sys/kern/kern_jail.c (revision 3110d4ebd6c0848cf5e25890d01791bb407e2a9b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/linker.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/racct.h>
55 #include <sys/rctl.h>
56 #include <sys/refcount.h>
57 #include <sys/sx.h>
58 #include <sys/sysent.h>
59 #include <sys/namei.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysctl.h>
65 #include <sys/uuid.h>
66 #include <sys/vnode.h>
67 
68 #include <net/if.h>
69 #include <net/vnet.h>
70 
71 #include <netinet/in.h>
72 
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif /* DDB */
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
80 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
81 
82 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
83 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
84 
85 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
86 #ifdef INET
87 #ifdef INET6
88 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
89 #else
90 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
91 #endif
92 #else /* !INET */
93 #ifdef INET6
94 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
95 #else
96 #define	_PR_IP_SADDRSEL	0
97 #endif
98 #endif
99 
100 /* prison0 describes what is "real" about the system. */
101 struct prison prison0 = {
102 	.pr_id		= 0,
103 	.pr_name	= "0",
104 	.pr_ref		= 1,
105 	.pr_uref	= 1,
106 	.pr_path	= "/",
107 	.pr_securelevel	= -1,
108 	.pr_devfs_rsnum = 0,
109 	.pr_childmax	= JAIL_MAX,
110 	.pr_hostuuid	= DEFAULT_HOSTUUID,
111 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
112 #ifdef VIMAGE
113 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
114 #else
115 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
116 #endif
117 	.pr_allow	= PR_ALLOW_ALL_STATIC,
118 };
119 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
120 
121 struct bool_flags {
122 	const char	*name;
123 	const char	*noname;
124 	volatile u_int	 flag;
125 };
126 struct jailsys_flags {
127 	const char	*name;
128 	unsigned	 disable;
129 	unsigned	 new;
130 };
131 
132 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
133 struct	sx allprison_lock;
134 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
135 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
136 LIST_HEAD(, prison_racct) allprison_racct;
137 int	lastprid = 0;
138 
139 static int get_next_prid(struct prison **insprp);
140 static int do_jail_attach(struct thread *td, struct prison *pr);
141 static void prison_complete(void *context, int pending);
142 static void prison_deref(struct prison *pr, int flags);
143 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
144     int enable);
145 static char *prison_path(struct prison *pr1, struct prison *pr2);
146 static void prison_remove_one(struct prison *pr);
147 #ifdef RACCT
148 static void prison_racct_attach(struct prison *pr);
149 static void prison_racct_modify(struct prison *pr);
150 static void prison_racct_detach(struct prison *pr);
151 #endif
152 
153 /* Flags for prison_deref */
154 #define	PD_DEREF	0x01	/* Decrement pr_ref */
155 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
156 #define	PD_LOCKED	0x04	/* pr_mtx is held */
157 #define	PD_LIST_SLOCKED	0x08	/* allprison_lock is held shared */
158 #define	PD_LIST_XLOCKED	0x10	/* allprison_lock is held exclusive */
159 
160 /*
161  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
162  * as we cannot figure out the size of a sparse array, or an array without a
163  * terminating entry.
164  */
165 static struct bool_flags pr_flag_bool[] = {
166 	{"persist", "nopersist", PR_PERSIST},
167 #ifdef INET
168 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
169 #endif
170 #ifdef INET6
171 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
172 #endif
173 };
174 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
175 
176 static struct jailsys_flags pr_flag_jailsys[] = {
177 	{"host", 0, PR_HOST},
178 #ifdef VIMAGE
179 	{"vnet", 0, PR_VNET},
180 #endif
181 #ifdef INET
182 	{"ip4", PR_IP4_USER, PR_IP4_USER},
183 #endif
184 #ifdef INET6
185 	{"ip6", PR_IP6_USER, PR_IP6_USER},
186 #endif
187 };
188 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
189 
190 /*
191  * Make this array full-size so dynamic parameters can be added.
192  * It is protected by prison0.mtx, but lockless reading is allowed
193  * with an atomic check of the flag values.
194  */
195 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
196 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
197 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
198 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
199 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
200 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
201 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
202 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
203 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
204 	{"allow.reserved_ports", "allow.noreserved_ports",
205 	 PR_ALLOW_RESERVED_PORTS},
206 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
207 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
208 	 PR_ALLOW_UNPRIV_DEBUG},
209 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
210 };
211 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
212 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
213 
214 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
215 					 PR_ALLOW_RESERVED_PORTS | \
216 					 PR_ALLOW_UNPRIV_DEBUG | \
217 					 PR_ALLOW_SUSER)
218 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
219 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
220 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
221 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
222 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
223 #if defined(INET) || defined(INET6)
224 static unsigned jail_max_af_ips = 255;
225 #endif
226 
227 /*
228  * Initialize the parts of prison0 that can't be static-initialized with
229  * constants.  This is called from proc0_init() after creating thread0 cpuset.
230  */
231 void
232 prison0_init(void)
233 {
234 	uint8_t *file, *data;
235 	size_t size;
236 
237 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
238 	prison0.pr_osreldate = osreldate;
239 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
240 
241 	/* If we have a preloaded hostuuid, use it. */
242 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
243 	if (file != NULL) {
244 		data = preload_fetch_addr(file);
245 		size = preload_fetch_size(file);
246 		if (data != NULL) {
247 			/*
248 			 * The preloaded data may include trailing whitespace, almost
249 			 * certainly a newline; skip over any whitespace or
250 			 * non-printable characters to be safe.
251 			 */
252 			while (size > 0 && data[size - 1] <= 0x20) {
253 				data[size--] = '\0';
254 			}
255 			if (validate_uuid(data, size, NULL, 0) == 0) {
256 				(void)strlcpy(prison0.pr_hostuuid, data,
257 				    size + 1);
258 			} else if (bootverbose) {
259 				printf("hostuuid: preload data malformed: '%s'",
260 				    data);
261 			}
262 		}
263 	}
264 	if (bootverbose)
265 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
266 }
267 
268 /*
269  * struct jail_args {
270  *	struct jail *jail;
271  * };
272  */
273 int
274 sys_jail(struct thread *td, struct jail_args *uap)
275 {
276 	uint32_t version;
277 	int error;
278 	struct jail j;
279 
280 	error = copyin(uap->jail, &version, sizeof(uint32_t));
281 	if (error)
282 		return (error);
283 
284 	switch (version) {
285 	case 0:
286 	{
287 		struct jail_v0 j0;
288 
289 		/* FreeBSD single IPv4 jails. */
290 		bzero(&j, sizeof(struct jail));
291 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
292 		if (error)
293 			return (error);
294 		j.version = j0.version;
295 		j.path = j0.path;
296 		j.hostname = j0.hostname;
297 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
298 		break;
299 	}
300 
301 	case 1:
302 		/*
303 		 * Version 1 was used by multi-IPv4 jail implementations
304 		 * that never made it into the official kernel.
305 		 */
306 		return (EINVAL);
307 
308 	case 2:	/* JAIL_API_VERSION */
309 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
310 		error = copyin(uap->jail, &j, sizeof(struct jail));
311 		if (error)
312 			return (error);
313 		break;
314 
315 	default:
316 		/* Sci-Fi jails are not supported, sorry. */
317 		return (EINVAL);
318 	}
319 	return (kern_jail(td, &j));
320 }
321 
322 int
323 kern_jail(struct thread *td, struct jail *j)
324 {
325 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
326 #ifdef INET
327 			    + 1
328 #endif
329 #ifdef INET6
330 			    + 1
331 #endif
332 			    )];
333 	struct uio opt;
334 	char *u_path, *u_hostname, *u_name;
335 	struct bool_flags *bf;
336 #ifdef INET
337 	uint32_t ip4s;
338 	struct in_addr *u_ip4;
339 #endif
340 #ifdef INET6
341 	struct in6_addr *u_ip6;
342 #endif
343 	size_t tmplen;
344 	int error, enforce_statfs;
345 
346 	bzero(&optiov, sizeof(optiov));
347 	opt.uio_iov = optiov;
348 	opt.uio_iovcnt = 0;
349 	opt.uio_offset = -1;
350 	opt.uio_resid = -1;
351 	opt.uio_segflg = UIO_SYSSPACE;
352 	opt.uio_rw = UIO_READ;
353 	opt.uio_td = td;
354 
355 	/* Set permissions for top-level jails from sysctls. */
356 	if (!jailed(td->td_ucred)) {
357 		for (bf = pr_flag_allow;
358 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
359 			atomic_load_int(&bf->flag) != 0;
360 		     bf++) {
361 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
362 			    (jail_default_allow & bf->flag)
363 			    ? bf->name : bf->noname);
364 			optiov[opt.uio_iovcnt].iov_len =
365 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
366 			opt.uio_iovcnt += 2;
367 		}
368 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
369 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
370 		opt.uio_iovcnt++;
371 		enforce_statfs = jail_default_enforce_statfs;
372 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
373 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
374 		opt.uio_iovcnt++;
375 	}
376 
377 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
378 #ifdef INET
379 	ip4s = (j->version == 0) ? 1 : j->ip4s;
380 	if (ip4s > jail_max_af_ips)
381 		return (EINVAL);
382 	tmplen += ip4s * sizeof(struct in_addr);
383 #else
384 	if (j->ip4s > 0)
385 		return (EINVAL);
386 #endif
387 #ifdef INET6
388 	if (j->ip6s > jail_max_af_ips)
389 		return (EINVAL);
390 	tmplen += j->ip6s * sizeof(struct in6_addr);
391 #else
392 	if (j->ip6s > 0)
393 		return (EINVAL);
394 #endif
395 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
396 	u_hostname = u_path + MAXPATHLEN;
397 	u_name = u_hostname + MAXHOSTNAMELEN;
398 #ifdef INET
399 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
400 #endif
401 #ifdef INET6
402 #ifdef INET
403 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
404 #else
405 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
406 #endif
407 #endif
408 	optiov[opt.uio_iovcnt].iov_base = "path";
409 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
410 	opt.uio_iovcnt++;
411 	optiov[opt.uio_iovcnt].iov_base = u_path;
412 	error = copyinstr(j->path, u_path, MAXPATHLEN,
413 	    &optiov[opt.uio_iovcnt].iov_len);
414 	if (error) {
415 		free(u_path, M_TEMP);
416 		return (error);
417 	}
418 	opt.uio_iovcnt++;
419 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
420 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
421 	opt.uio_iovcnt++;
422 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
423 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
424 	    &optiov[opt.uio_iovcnt].iov_len);
425 	if (error) {
426 		free(u_path, M_TEMP);
427 		return (error);
428 	}
429 	opt.uio_iovcnt++;
430 	if (j->jailname != NULL) {
431 		optiov[opt.uio_iovcnt].iov_base = "name";
432 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
433 		opt.uio_iovcnt++;
434 		optiov[opt.uio_iovcnt].iov_base = u_name;
435 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
436 		    &optiov[opt.uio_iovcnt].iov_len);
437 		if (error) {
438 			free(u_path, M_TEMP);
439 			return (error);
440 		}
441 		opt.uio_iovcnt++;
442 	}
443 #ifdef INET
444 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
445 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
446 	opt.uio_iovcnt++;
447 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
448 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
449 	if (j->version == 0)
450 		u_ip4->s_addr = j->ip4s;
451 	else {
452 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
453 		if (error) {
454 			free(u_path, M_TEMP);
455 			return (error);
456 		}
457 	}
458 	opt.uio_iovcnt++;
459 #endif
460 #ifdef INET6
461 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
462 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
463 	opt.uio_iovcnt++;
464 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
465 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
466 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
467 	if (error) {
468 		free(u_path, M_TEMP);
469 		return (error);
470 	}
471 	opt.uio_iovcnt++;
472 #endif
473 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
474 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
475 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
476 	free(u_path, M_TEMP);
477 	return (error);
478 }
479 
480 /*
481  * struct jail_set_args {
482  *	struct iovec *iovp;
483  *	unsigned int iovcnt;
484  *	int flags;
485  * };
486  */
487 int
488 sys_jail_set(struct thread *td, struct jail_set_args *uap)
489 {
490 	struct uio *auio;
491 	int error;
492 
493 	/* Check that we have an even number of iovecs. */
494 	if (uap->iovcnt & 1)
495 		return (EINVAL);
496 
497 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
498 	if (error)
499 		return (error);
500 	error = kern_jail_set(td, auio, uap->flags);
501 	free(auio, M_IOV);
502 	return (error);
503 }
504 
505 int
506 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
507 {
508 	struct nameidata nd;
509 #ifdef INET
510 	struct in_addr *ip4;
511 #endif
512 #ifdef INET6
513 	struct in6_addr *ip6;
514 #endif
515 	struct vfsopt *opt;
516 	struct vfsoptlist *opts;
517 	struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
518 	struct vnode *root;
519 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
520 	char *g_path, *osrelstr;
521 	struct bool_flags *bf;
522 	struct jailsys_flags *jsf;
523 #if defined(INET) || defined(INET6)
524 	struct prison *tppr;
525 	void *op;
526 #endif
527 	unsigned long hid;
528 	size_t namelen, onamelen, pnamelen;
529 	int born, created, cuflags, descend, drflags, enforce;
530 	int error, errmsg_len, errmsg_pos;
531 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
532 	int jid, jsys, len, level;
533 	int childmax, osreldt, rsnum, slevel;
534 #if defined(INET) || defined(INET6)
535 	int ii, ij;
536 #endif
537 #ifdef INET
538 	int ip4s, redo_ip4;
539 #endif
540 #ifdef INET6
541 	int ip6s, redo_ip6;
542 #endif
543 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
544 	uint64_t pr_allow_diff;
545 	unsigned tallow;
546 	char numbuf[12];
547 
548 	error = priv_check(td, PRIV_JAIL_SET);
549 	if (!error && (flags & JAIL_ATTACH))
550 		error = priv_check(td, PRIV_JAIL_ATTACH);
551 	if (error)
552 		return (error);
553 	mypr = td->td_ucred->cr_prison;
554 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
555 		return (EPERM);
556 	if (flags & ~JAIL_SET_MASK)
557 		return (EINVAL);
558 
559 	/*
560 	 * Check all the parameters before committing to anything.  Not all
561 	 * errors can be caught early, but we may as well try.  Also, this
562 	 * takes care of some expensive stuff (path lookup) before getting
563 	 * the allprison lock.
564 	 *
565 	 * XXX Jails are not filesystems, and jail parameters are not mount
566 	 *     options.  But it makes more sense to re-use the vfsopt code
567 	 *     than duplicate it under a different name.
568 	 */
569 	error = vfs_buildopts(optuio, &opts);
570 	if (error)
571 		return (error);
572 #ifdef INET
573 	ip4 = NULL;
574 #endif
575 #ifdef INET6
576 	ip6 = NULL;
577 #endif
578 	g_path = NULL;
579 
580 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
581 	if (!cuflags) {
582 		error = EINVAL;
583 		vfs_opterror(opts, "no valid operation (create or update)");
584 		goto done_errmsg;
585 	}
586 
587 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
588 	if (error == ENOENT)
589 		jid = 0;
590 	else if (error != 0)
591 		goto done_free;
592 
593 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
594 	if (error == ENOENT)
595 		gotslevel = 0;
596 	else if (error != 0)
597 		goto done_free;
598 	else
599 		gotslevel = 1;
600 
601 	error =
602 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
603 	if (error == ENOENT)
604 		gotchildmax = 0;
605 	else if (error != 0)
606 		goto done_free;
607 	else
608 		gotchildmax = 1;
609 
610 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
611 	if (error == ENOENT)
612 		gotenforce = 0;
613 	else if (error != 0)
614 		goto done_free;
615 	else if (enforce < 0 || enforce > 2) {
616 		error = EINVAL;
617 		goto done_free;
618 	} else
619 		gotenforce = 1;
620 
621 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
622 	if (error == ENOENT)
623 		gotrsnum = 0;
624 	else if (error != 0)
625 		goto done_free;
626 	else
627 		gotrsnum = 1;
628 
629 	pr_flags = ch_flags = 0;
630 	for (bf = pr_flag_bool;
631 	     bf < pr_flag_bool + nitems(pr_flag_bool);
632 	     bf++) {
633 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
634 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
635 	}
636 	ch_flags |= pr_flags;
637 	for (jsf = pr_flag_jailsys;
638 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
639 	     jsf++) {
640 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
641 		if (error == ENOENT)
642 			continue;
643 		if (error != 0)
644 			goto done_free;
645 		switch (jsys) {
646 		case JAIL_SYS_DISABLE:
647 			if (!jsf->disable) {
648 				error = EINVAL;
649 				goto done_free;
650 			}
651 			pr_flags |= jsf->disable;
652 			break;
653 		case JAIL_SYS_NEW:
654 			pr_flags |= jsf->new;
655 			break;
656 		case JAIL_SYS_INHERIT:
657 			break;
658 		default:
659 			error = EINVAL;
660 			goto done_free;
661 		}
662 		ch_flags |= jsf->new | jsf->disable;
663 	}
664 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
665 	    && !(pr_flags & PR_PERSIST)) {
666 		error = EINVAL;
667 		vfs_opterror(opts, "new jail must persist or attach");
668 		goto done_errmsg;
669 	}
670 #ifdef VIMAGE
671 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
672 		error = EINVAL;
673 		vfs_opterror(opts, "vnet cannot be changed after creation");
674 		goto done_errmsg;
675 	}
676 #endif
677 #ifdef INET
678 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
679 		error = EINVAL;
680 		vfs_opterror(opts, "ip4 cannot be changed after creation");
681 		goto done_errmsg;
682 	}
683 #endif
684 #ifdef INET6
685 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
686 		error = EINVAL;
687 		vfs_opterror(opts, "ip6 cannot be changed after creation");
688 		goto done_errmsg;
689 	}
690 #endif
691 
692 	pr_allow = ch_allow = 0;
693 	for (bf = pr_flag_allow;
694 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
695 		atomic_load_int(&bf->flag) != 0;
696 	     bf++) {
697 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
698 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
699 	}
700 	ch_allow |= pr_allow;
701 
702 	error = vfs_getopt(opts, "name", (void **)&name, &len);
703 	if (error == ENOENT)
704 		name = NULL;
705 	else if (error != 0)
706 		goto done_free;
707 	else {
708 		if (len == 0 || name[len - 1] != '\0') {
709 			error = EINVAL;
710 			goto done_free;
711 		}
712 		if (len > MAXHOSTNAMELEN) {
713 			error = ENAMETOOLONG;
714 			goto done_free;
715 		}
716 	}
717 
718 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
719 	if (error == ENOENT)
720 		host = NULL;
721 	else if (error != 0)
722 		goto done_free;
723 	else {
724 		ch_flags |= PR_HOST;
725 		pr_flags |= PR_HOST;
726 		if (len == 0 || host[len - 1] != '\0') {
727 			error = EINVAL;
728 			goto done_free;
729 		}
730 		if (len > MAXHOSTNAMELEN) {
731 			error = ENAMETOOLONG;
732 			goto done_free;
733 		}
734 	}
735 
736 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
737 	if (error == ENOENT)
738 		domain = NULL;
739 	else if (error != 0)
740 		goto done_free;
741 	else {
742 		ch_flags |= PR_HOST;
743 		pr_flags |= PR_HOST;
744 		if (len == 0 || domain[len - 1] != '\0') {
745 			error = EINVAL;
746 			goto done_free;
747 		}
748 		if (len > MAXHOSTNAMELEN) {
749 			error = ENAMETOOLONG;
750 			goto done_free;
751 		}
752 	}
753 
754 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
755 	if (error == ENOENT)
756 		uuid = NULL;
757 	else if (error != 0)
758 		goto done_free;
759 	else {
760 		ch_flags |= PR_HOST;
761 		pr_flags |= PR_HOST;
762 		if (len == 0 || uuid[len - 1] != '\0') {
763 			error = EINVAL;
764 			goto done_free;
765 		}
766 		if (len > HOSTUUIDLEN) {
767 			error = ENAMETOOLONG;
768 			goto done_free;
769 		}
770 	}
771 
772 #ifdef COMPAT_FREEBSD32
773 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
774 		uint32_t hid32;
775 
776 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
777 		hid = hid32;
778 	} else
779 #endif
780 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
781 	if (error == ENOENT)
782 		gothid = 0;
783 	else if (error != 0)
784 		goto done_free;
785 	else {
786 		gothid = 1;
787 		ch_flags |= PR_HOST;
788 		pr_flags |= PR_HOST;
789 	}
790 
791 #ifdef INET
792 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
793 	if (error == ENOENT)
794 		ip4s = 0;
795 	else if (error != 0)
796 		goto done_free;
797 	else if (ip4s & (sizeof(*ip4) - 1)) {
798 		error = EINVAL;
799 		goto done_free;
800 	} else {
801 		ch_flags |= PR_IP4_USER;
802 		pr_flags |= PR_IP4_USER;
803 		if (ip4s > 0) {
804 			ip4s /= sizeof(*ip4);
805 			if (ip4s > jail_max_af_ips) {
806 				error = EINVAL;
807 				vfs_opterror(opts, "too many IPv4 addresses");
808 				goto done_errmsg;
809 			}
810 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
811 			bcopy(op, ip4, ip4s * sizeof(*ip4));
812 			/*
813 			 * IP addresses are all sorted but ip[0] to preserve
814 			 * the primary IP address as given from userland.
815 			 * This special IP is used for unbound outgoing
816 			 * connections as well for "loopback" traffic in case
817 			 * source address selection cannot find any more fitting
818 			 * address to connect from.
819 			 */
820 			if (ip4s > 1)
821 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
822 				    prison_qcmp_v4);
823 			/*
824 			 * Check for duplicate addresses and do some simple
825 			 * zero and broadcast checks. If users give other bogus
826 			 * addresses it is their problem.
827 			 *
828 			 * We do not have to care about byte order for these
829 			 * checks so we will do them in NBO.
830 			 */
831 			for (ii = 0; ii < ip4s; ii++) {
832 				if (ip4[ii].s_addr == INADDR_ANY ||
833 				    ip4[ii].s_addr == INADDR_BROADCAST) {
834 					error = EINVAL;
835 					goto done_free;
836 				}
837 				if ((ii+1) < ip4s &&
838 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
839 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
840 					error = EINVAL;
841 					goto done_free;
842 				}
843 			}
844 		}
845 	}
846 #endif
847 
848 #ifdef INET6
849 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
850 	if (error == ENOENT)
851 		ip6s = 0;
852 	else if (error != 0)
853 		goto done_free;
854 	else if (ip6s & (sizeof(*ip6) - 1)) {
855 		error = EINVAL;
856 		goto done_free;
857 	} else {
858 		ch_flags |= PR_IP6_USER;
859 		pr_flags |= PR_IP6_USER;
860 		if (ip6s > 0) {
861 			ip6s /= sizeof(*ip6);
862 			if (ip6s > jail_max_af_ips) {
863 				error = EINVAL;
864 				vfs_opterror(opts, "too many IPv6 addresses");
865 				goto done_errmsg;
866 			}
867 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
868 			bcopy(op, ip6, ip6s * sizeof(*ip6));
869 			if (ip6s > 1)
870 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
871 				    prison_qcmp_v6);
872 			for (ii = 0; ii < ip6s; ii++) {
873 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
874 					error = EINVAL;
875 					goto done_free;
876 				}
877 				if ((ii+1) < ip6s &&
878 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
879 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
880 				{
881 					error = EINVAL;
882 					goto done_free;
883 				}
884 			}
885 		}
886 	}
887 #endif
888 
889 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
890 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
891 		error = EINVAL;
892 		vfs_opterror(opts,
893 		    "vnet jails cannot have IP address restrictions");
894 		goto done_errmsg;
895 	}
896 #endif
897 
898 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
899 	if (error == ENOENT)
900 		osrelstr = NULL;
901 	else if (error != 0)
902 		goto done_free;
903 	else {
904 		if (flags & JAIL_UPDATE) {
905 			error = EINVAL;
906 			vfs_opterror(opts,
907 			    "osrelease cannot be changed after creation");
908 			goto done_errmsg;
909 		}
910 		if (len == 0 || osrelstr[len - 1] != '\0') {
911 			error = EINVAL;
912 			goto done_free;
913 		}
914 		if (len >= OSRELEASELEN) {
915 			error = ENAMETOOLONG;
916 			vfs_opterror(opts,
917 			    "osrelease string must be 1-%d bytes long",
918 			    OSRELEASELEN - 1);
919 			goto done_errmsg;
920 		}
921 	}
922 
923 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
924 	if (error == ENOENT)
925 		osreldt = 0;
926 	else if (error != 0)
927 		goto done_free;
928 	else {
929 		if (flags & JAIL_UPDATE) {
930 			error = EINVAL;
931 			vfs_opterror(opts,
932 			    "osreldate cannot be changed after creation");
933 			goto done_errmsg;
934 		}
935 		if (osreldt == 0) {
936 			error = EINVAL;
937 			vfs_opterror(opts, "osreldate cannot be 0");
938 			goto done_errmsg;
939 		}
940 	}
941 
942 	root = NULL;
943 	error = vfs_getopt(opts, "path", (void **)&path, &len);
944 	if (error == ENOENT)
945 		path = NULL;
946 	else if (error != 0)
947 		goto done_free;
948 	else {
949 		if (flags & JAIL_UPDATE) {
950 			error = EINVAL;
951 			vfs_opterror(opts,
952 			    "path cannot be changed after creation");
953 			goto done_errmsg;
954 		}
955 		if (len == 0 || path[len - 1] != '\0') {
956 			error = EINVAL;
957 			goto done_free;
958 		}
959 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
960 		    path, td);
961 		error = namei(&nd);
962 		if (error)
963 			goto done_free;
964 		root = nd.ni_vp;
965 		NDFREE(&nd, NDF_ONLY_PNBUF);
966 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
967 		strlcpy(g_path, path, MAXPATHLEN);
968 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
969 		if (error == 0) {
970 			path = g_path;
971 		} else {
972 			/* exit on other errors */
973 			goto done_free;
974 		}
975 		if (root->v_type != VDIR) {
976 			error = ENOTDIR;
977 			vput(root);
978 			goto done_free;
979 		}
980 		VOP_UNLOCK(root);
981 	}
982 
983 	/*
984 	 * Find the specified jail, or at least its parent.
985 	 * This abuses the file error codes ENOENT and EEXIST.
986 	 */
987 	pr = NULL;
988 	ppr = mypr;
989 	inspr = NULL;
990 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
991 		namelc = strrchr(name, '.');
992 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
993 		if (*p != '\0')
994 			jid = 0;
995 	}
996 	sx_xlock(&allprison_lock);
997 	drflags = PD_LIST_XLOCKED;
998 	if (jid != 0) {
999 		if (jid < 0) {
1000 			error = EINVAL;
1001 			vfs_opterror(opts, "negative jid");
1002 			goto done_deref;
1003 		}
1004 		/*
1005 		 * See if a requested jid already exists.  Keep track of
1006 		 * where it can be inserted later.
1007 		 */
1008 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1009 			if (inspr->pr_id == jid) {
1010 				mtx_lock(&inspr->pr_mtx);
1011 				if (prison_isvalid(inspr)) {
1012 					pr = inspr;
1013 					drflags |= PD_LOCKED;
1014 					inspr = NULL;
1015 				} else
1016 					mtx_unlock(&inspr->pr_mtx);
1017 				break;
1018 			}
1019 			if (inspr->pr_id > jid)
1020 				break;
1021 		}
1022 		if (pr != NULL) {
1023 			ppr = pr->pr_parent;
1024 			/* Create: jid must not exist. */
1025 			if (cuflags == JAIL_CREATE) {
1026 				/*
1027 				 * Even creators that cannot see the jail will
1028 				 * get EEXIST.
1029 				 */
1030 				error = EEXIST;
1031 				vfs_opterror(opts, "jail %d already exists",
1032 				    jid);
1033 				goto done_deref;
1034 			}
1035 			if (!prison_ischild(mypr, pr)) {
1036 				/*
1037 				 * Updaters get ENOENT if they cannot see the
1038 				 * jail.  This is true even for CREATE | UPDATE,
1039 				 * which normally cannot give this error.
1040 				 */
1041 				error = ENOENT;
1042 				vfs_opterror(opts, "jail %d not found", jid);
1043 				goto done_deref;
1044 			} else if (!prison_isalive(pr)) {
1045 				if (!(flags & JAIL_DYING)) {
1046 					error = ENOENT;
1047 					vfs_opterror(opts, "jail %d is dying",
1048 					    jid);
1049 					goto done_deref;
1050 				} else if ((flags & JAIL_ATTACH) ||
1051 				    (pr_flags & PR_PERSIST)) {
1052 					/*
1053 					 * A dying jail might be resurrected
1054 					 * (via attach or persist), but first
1055 					 * it must determine if another jail
1056 					 * has claimed its name.  Accomplish
1057 					 * this by implicitly re-setting the
1058 					 * name.
1059 					 */
1060 					if (name == NULL)
1061 						name = prison_name(mypr, pr);
1062 				}
1063 			}
1064 		} else {
1065 			/* Update: jid must exist. */
1066 			if (cuflags == JAIL_UPDATE) {
1067 				error = ENOENT;
1068 				vfs_opterror(opts, "jail %d not found", jid);
1069 				goto done_deref;
1070 			}
1071 		}
1072 	}
1073 	/*
1074 	 * If the caller provided a name, look for a jail by that name.
1075 	 * This has different semantics for creates and updates keyed by jid
1076 	 * (where the name must not already exist in a different jail),
1077 	 * and updates keyed by the name itself (where the name must exist
1078 	 * because that is the jail being updated).
1079 	 */
1080 	namelc = NULL;
1081 	if (name != NULL) {
1082 		namelc = strrchr(name, '.');
1083 		if (namelc == NULL)
1084 			namelc = name;
1085 		else {
1086 			/*
1087 			 * This is a hierarchical name.  Split it into the
1088 			 * parent and child names, and make sure the parent
1089 			 * exists or matches an already found jail.
1090 			 */
1091 			if (pr != NULL) {
1092 				if (strncmp(name, ppr->pr_name, namelc - name)
1093 				    || ppr->pr_name[namelc - name] != '\0') {
1094 					error = EINVAL;
1095 					vfs_opterror(opts,
1096 					    "cannot change jail's parent");
1097 					goto done_deref;
1098 				}
1099 			} else {
1100 				*namelc = '\0';
1101 				ppr = prison_find_name(mypr, name);
1102 				if (ppr == NULL) {
1103 					error = ENOENT;
1104 					vfs_opterror(opts,
1105 					    "jail \"%s\" not found", name);
1106 					goto done_deref;
1107 				}
1108 				mtx_unlock(&ppr->pr_mtx);
1109 				*namelc = '.';
1110 			}
1111 			namelc++;
1112 		}
1113 		if (namelc[0] != '\0') {
1114 			pnamelen =
1115 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1116  name_again:
1117 			deadpr = NULL;
1118 			FOREACH_PRISON_CHILD(ppr, tpr) {
1119 				if (tpr != pr &&
1120 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1121 					mtx_lock(&tpr->pr_mtx);
1122 					if (prison_isalive(tpr)) {
1123 						if (pr == NULL &&
1124 						    cuflags != JAIL_CREATE) {
1125 							/*
1126 							 * Use this jail
1127 							 * for updates.
1128 							 */
1129 							pr = tpr;
1130 							drflags |= PD_LOCKED;
1131 							break;
1132 						}
1133 						/*
1134 						 * Create, or update(jid):
1135 						 * name must not exist in an
1136 						 * active sibling jail.
1137 						 */
1138 						error = EEXIST;
1139 						mtx_unlock(&tpr->pr_mtx);
1140 						vfs_opterror(opts,
1141 						   "jail \"%s\" already exists",
1142 						   name);
1143 						goto done_deref;
1144 					}
1145 					if (pr == NULL &&
1146 					    cuflags != JAIL_CREATE &&
1147 					    prison_isvalid(tpr))
1148 						deadpr = tpr;
1149 					mtx_unlock(&tpr->pr_mtx);
1150 				}
1151 			}
1152 			/* If no active jail is found, use a dying one. */
1153 			if (deadpr != NULL && pr == NULL) {
1154 				if (flags & JAIL_DYING) {
1155 					mtx_lock(&deadpr->pr_mtx);
1156 					if (!prison_isvalid(deadpr)) {
1157 						mtx_unlock(&deadpr->pr_mtx);
1158 						goto name_again;
1159 					}
1160 					pr = deadpr;
1161 					drflags |= PD_LOCKED;
1162 				} else if (cuflags == JAIL_UPDATE) {
1163 					error = ENOENT;
1164 					vfs_opterror(opts,
1165 					    "jail \"%s\" is dying", name);
1166 					goto done_deref;
1167 				}
1168 			}
1169 			/* Update: name must exist if no jid. */
1170 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1171 				error = ENOENT;
1172 				vfs_opterror(opts, "jail \"%s\" not found",
1173 				    name);
1174 				goto done_deref;
1175 			}
1176 		}
1177 	}
1178 	/* Update: must provide a jid or name. */
1179 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1180 		error = ENOENT;
1181 		vfs_opterror(opts, "update specified no jail");
1182 		goto done_deref;
1183 	}
1184 
1185 	/* If there's no prison to update, create a new one and link it in. */
1186 	created = pr == NULL;
1187 	if (created) {
1188 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1189 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1190 				error = EPERM;
1191 				vfs_opterror(opts, "prison limit exceeded");
1192 				goto done_deref;
1193 			}
1194 		mtx_lock(&ppr->pr_mtx);
1195 		if (!prison_isvalid(ppr)) {
1196 			mtx_unlock(&ppr->pr_mtx);
1197 			error = ENOENT;
1198 			vfs_opterror(opts, "jail \"%s\" not found",
1199 			    prison_name(mypr, ppr));
1200 			goto done_deref;
1201 		}
1202 		prison_hold(ppr);
1203 		refcount_acquire(&ppr->pr_uref);
1204 		mtx_unlock(&ppr->pr_mtx);
1205 
1206 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1207 			error = EAGAIN;
1208 			vfs_opterror(opts, "no available jail IDs");
1209 			pr = ppr;
1210 			drflags |= PD_DEREF | PD_DEUREF;
1211 			goto done_deref;
1212 		}
1213 
1214 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1215 		LIST_INIT(&pr->pr_children);
1216 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1217 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1218 
1219 		pr->pr_id = jid;
1220 		if (inspr != NULL)
1221 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1222 		else
1223 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1224 
1225 		pr->pr_parent = ppr;
1226 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1227 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1228 			tpr->pr_childcount++;
1229 
1230 		/* Set some default values, and inherit some from the parent. */
1231 		if (namelc == NULL)
1232 			namelc = "";
1233 		if (path == NULL) {
1234 			path = "/";
1235 			root = mypr->pr_root;
1236 			vref(root);
1237 		}
1238 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1239 		pr->pr_flags |= PR_HOST;
1240 #if defined(INET) || defined(INET6)
1241 #ifdef VIMAGE
1242 		if (!(pr_flags & PR_VNET))
1243 #endif
1244 		{
1245 #ifdef INET
1246 			if (!(ch_flags & PR_IP4_USER))
1247 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1248 			else if (!(pr_flags & PR_IP4_USER)) {
1249 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1250 				if (ppr->pr_ip4 != NULL) {
1251 					pr->pr_ip4s = ppr->pr_ip4s;
1252 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1253 					    sizeof(struct in_addr), M_PRISON,
1254 					    M_WAITOK);
1255 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1256 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1257 				}
1258 			}
1259 #endif
1260 #ifdef INET6
1261 			if (!(ch_flags & PR_IP6_USER))
1262 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1263 			else if (!(pr_flags & PR_IP6_USER)) {
1264 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1265 				if (ppr->pr_ip6 != NULL) {
1266 					pr->pr_ip6s = ppr->pr_ip6s;
1267 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1268 					    sizeof(struct in6_addr), M_PRISON,
1269 					    M_WAITOK);
1270 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1271 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1272 				}
1273 			}
1274 #endif
1275 		}
1276 #endif
1277 		/* Source address selection is always on by default. */
1278 		pr->pr_flags |= _PR_IP_SADDRSEL;
1279 
1280 		pr->pr_securelevel = ppr->pr_securelevel;
1281 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1282 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1283 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1284 
1285 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1286 		if (osrelstr == NULL)
1287 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1288 			    sizeof(pr->pr_osrelease));
1289 		else
1290 			strlcpy(pr->pr_osrelease, osrelstr,
1291 			    sizeof(pr->pr_osrelease));
1292 
1293 #ifdef VIMAGE
1294 		/* Allocate a new vnet if specified. */
1295 		pr->pr_vnet = (pr_flags & PR_VNET)
1296 		    ? vnet_alloc() : ppr->pr_vnet;
1297 #endif
1298 		/*
1299 		 * Allocate a dedicated cpuset for each jail.
1300 		 * Unlike other initial settings, this may return an erorr.
1301 		 */
1302 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1303 		if (error)
1304 			goto done_deref;
1305 
1306 		mtx_lock(&pr->pr_mtx);
1307 		drflags |= PD_LOCKED;
1308 		/*
1309 		 * New prisons do not yet have a reference, because we do not
1310 		 * want others to see the incomplete prison once the
1311 		 * allprison_lock is downgraded.
1312 		 */
1313 	} else {
1314 		/*
1315 		 * Grab a reference for existing prisons, to ensure they
1316 		 * continue to exist for the duration of the call.
1317 		 */
1318 		prison_hold(pr);
1319 		drflags |= PD_DEREF;
1320 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1321 		if ((pr->pr_flags & PR_VNET) &&
1322 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1323 			error = EINVAL;
1324 			vfs_opterror(opts,
1325 			    "vnet jails cannot have IP address restrictions");
1326 			goto done_deref;
1327 		}
1328 #endif
1329 #ifdef INET
1330 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1331 			error = EINVAL;
1332 			vfs_opterror(opts,
1333 			    "ip4 cannot be changed after creation");
1334 			goto done_deref;
1335 		}
1336 #endif
1337 #ifdef INET6
1338 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1339 			error = EINVAL;
1340 			vfs_opterror(opts,
1341 			    "ip6 cannot be changed after creation");
1342 			goto done_deref;
1343 		}
1344 #endif
1345 	}
1346 
1347 	/* Do final error checking before setting anything. */
1348 	if (gotslevel) {
1349 		if (slevel < ppr->pr_securelevel) {
1350 			error = EPERM;
1351 			goto done_deref;
1352 		}
1353 	}
1354 	if (gotchildmax) {
1355 		if (childmax >= ppr->pr_childmax) {
1356 			error = EPERM;
1357 			goto done_deref;
1358 		}
1359 	}
1360 	if (gotenforce) {
1361 		if (enforce < ppr->pr_enforce_statfs) {
1362 			error = EPERM;
1363 			goto done_deref;
1364 		}
1365 	}
1366 	if (gotrsnum) {
1367 		/*
1368 		 * devfs_rsnum is a uint16_t
1369 		 */
1370 		if (rsnum < 0 || rsnum > 65535) {
1371 			error = EINVAL;
1372 			goto done_deref;
1373 		}
1374 		/*
1375 		 * Nested jails always inherit parent's devfs ruleset
1376 		 */
1377 		if (jailed(td->td_ucred)) {
1378 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1379 				error = EPERM;
1380 				goto done_deref;
1381 			} else
1382 				rsnum = ppr->pr_devfs_rsnum;
1383 		}
1384 	}
1385 #ifdef INET
1386 	if (ip4s > 0) {
1387 		if (ppr->pr_flags & PR_IP4) {
1388 			/*
1389 			 * Make sure the new set of IP addresses is a
1390 			 * subset of the parent's list.  Don't worry
1391 			 * about the parent being unlocked, as any
1392 			 * setting is done with allprison_lock held.
1393 			 */
1394 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1395 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1396 					break;
1397 			if (ij == ppr->pr_ip4s) {
1398 				error = EPERM;
1399 				goto done_deref;
1400 			}
1401 			if (ip4s > 1) {
1402 				for (ii = ij = 1; ii < ip4s; ii++) {
1403 					if (ip4[ii].s_addr ==
1404 					    ppr->pr_ip4[0].s_addr)
1405 						continue;
1406 					for (; ij < ppr->pr_ip4s; ij++)
1407 						if (ip4[ii].s_addr ==
1408 						    ppr->pr_ip4[ij].s_addr)
1409 							break;
1410 					if (ij == ppr->pr_ip4s)
1411 						break;
1412 				}
1413 				if (ij == ppr->pr_ip4s) {
1414 					error = EPERM;
1415 					goto done_deref;
1416 				}
1417 			}
1418 		}
1419 		/*
1420 		 * Check for conflicting IP addresses.  We permit them
1421 		 * if there is no more than one IP on each jail.  If
1422 		 * there is a duplicate on a jail with more than one
1423 		 * IP stop checking and return error.
1424 		 */
1425 #ifdef VIMAGE
1426 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1427 			if (tppr->pr_flags & PR_VNET)
1428 				break;
1429 #else
1430 		tppr = &prison0;
1431 #endif
1432 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1433 			if (tpr == pr ||
1434 #ifdef VIMAGE
1435 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1436 #endif
1437 			    refcount_load(&tpr->pr_uref) == 0) {
1438 				descend = 0;
1439 				continue;
1440 			}
1441 			if (!(tpr->pr_flags & PR_IP4_USER))
1442 				continue;
1443 			descend = 0;
1444 			if (tpr->pr_ip4 == NULL ||
1445 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1446 				continue;
1447 			for (ii = 0; ii < ip4s; ii++) {
1448 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1449 				    0) {
1450 					error = EADDRINUSE;
1451 					vfs_opterror(opts,
1452 					    "IPv4 addresses clash");
1453 					goto done_deref;
1454 				}
1455 			}
1456 		}
1457 	}
1458 #endif
1459 #ifdef INET6
1460 	if (ip6s > 0) {
1461 		if (ppr->pr_flags & PR_IP6) {
1462 			/*
1463 			 * Make sure the new set of IP addresses is a
1464 			 * subset of the parent's list.
1465 			 */
1466 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1467 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1468 				    &ppr->pr_ip6[ij]))
1469 					break;
1470 			if (ij == ppr->pr_ip6s) {
1471 				error = EPERM;
1472 				goto done_deref;
1473 			}
1474 			if (ip6s > 1) {
1475 				for (ii = ij = 1; ii < ip6s; ii++) {
1476 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1477 					     &ppr->pr_ip6[0]))
1478 						continue;
1479 					for (; ij < ppr->pr_ip6s; ij++)
1480 						if (IN6_ARE_ADDR_EQUAL(
1481 						    &ip6[ii], &ppr->pr_ip6[ij]))
1482 							break;
1483 					if (ij == ppr->pr_ip6s)
1484 						break;
1485 				}
1486 				if (ij == ppr->pr_ip6s) {
1487 					error = EPERM;
1488 					goto done_deref;
1489 				}
1490 			}
1491 		}
1492 		/* Check for conflicting IP addresses. */
1493 #ifdef VIMAGE
1494 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1495 			if (tppr->pr_flags & PR_VNET)
1496 				break;
1497 #else
1498 		tppr = &prison0;
1499 #endif
1500 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1501 			if (tpr == pr ||
1502 #ifdef VIMAGE
1503 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1504 #endif
1505 			    refcount_load(&tpr->pr_uref) == 0) {
1506 				descend = 0;
1507 				continue;
1508 			}
1509 			if (!(tpr->pr_flags & PR_IP6_USER))
1510 				continue;
1511 			descend = 0;
1512 			if (tpr->pr_ip6 == NULL ||
1513 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1514 				continue;
1515 			for (ii = 0; ii < ip6s; ii++) {
1516 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1517 				    0) {
1518 					error = EADDRINUSE;
1519 					vfs_opterror(opts,
1520 					    "IPv6 addresses clash");
1521 					goto done_deref;
1522 				}
1523 			}
1524 		}
1525 	}
1526 #endif
1527 	onamelen = namelen = 0;
1528 	if (namelc != NULL) {
1529 		/* Give a default name of the jid.  Also allow the name to be
1530 		 * explicitly the jid - but not any other number, and only in
1531 		 * normal form (no leading zero/etc).
1532 		 */
1533 		if (namelc[0] == '\0')
1534 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1535 		else if ((strtoul(namelc, &p, 10) != jid ||
1536 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1537 			error = EINVAL;
1538 			vfs_opterror(opts,
1539 			    "name cannot be numeric (unless it is the jid)");
1540 			goto done_deref;
1541 		}
1542 		/*
1543 		 * Make sure the name isn't too long for the prison or its
1544 		 * children.
1545 		 */
1546 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1547 		onamelen = strlen(pr->pr_name + pnamelen);
1548 		namelen = strlen(namelc);
1549 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1550 			error = ENAMETOOLONG;
1551 			goto done_deref;
1552 		}
1553 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1554 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1555 			    sizeof(pr->pr_name)) {
1556 				error = ENAMETOOLONG;
1557 				goto done_deref;
1558 			}
1559 		}
1560 	}
1561 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1562 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1563 		error = EPERM;
1564 		goto done_deref;
1565 	}
1566 
1567 	/*
1568 	 * Let modules check their parameters.  This requires unlocking and
1569 	 * then re-locking the prison, but this is still a valid state as long
1570 	 * as allprison_lock remains xlocked.
1571 	 */
1572 	mtx_unlock(&pr->pr_mtx);
1573 	drflags &= ~PD_LOCKED;
1574 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1575 	if (error != 0)
1576 		goto done_deref;
1577 	mtx_lock(&pr->pr_mtx);
1578 	drflags |= PD_LOCKED;
1579 
1580 	/* At this point, all valid parameters should have been noted. */
1581 	TAILQ_FOREACH(opt, opts, link) {
1582 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1583 			error = EINVAL;
1584 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1585 			goto done_deref;
1586 		}
1587 	}
1588 
1589 	/* Set the parameters of the prison. */
1590 #ifdef INET
1591 	redo_ip4 = 0;
1592 	if (pr_flags & PR_IP4_USER) {
1593 		pr->pr_flags |= PR_IP4;
1594 		free(pr->pr_ip4, M_PRISON);
1595 		pr->pr_ip4s = ip4s;
1596 		pr->pr_ip4 = ip4;
1597 		ip4 = NULL;
1598 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1599 #ifdef VIMAGE
1600 			if (tpr->pr_flags & PR_VNET) {
1601 				descend = 0;
1602 				continue;
1603 			}
1604 #endif
1605 			if (prison_restrict_ip4(tpr, NULL)) {
1606 				redo_ip4 = 1;
1607 				descend = 0;
1608 			}
1609 		}
1610 	}
1611 #endif
1612 #ifdef INET6
1613 	redo_ip6 = 0;
1614 	if (pr_flags & PR_IP6_USER) {
1615 		pr->pr_flags |= PR_IP6;
1616 		free(pr->pr_ip6, M_PRISON);
1617 		pr->pr_ip6s = ip6s;
1618 		pr->pr_ip6 = ip6;
1619 		ip6 = NULL;
1620 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1621 #ifdef VIMAGE
1622 			if (tpr->pr_flags & PR_VNET) {
1623 				descend = 0;
1624 				continue;
1625 			}
1626 #endif
1627 			if (prison_restrict_ip6(tpr, NULL)) {
1628 				redo_ip6 = 1;
1629 				descend = 0;
1630 			}
1631 		}
1632 	}
1633 #endif
1634 	if (gotslevel) {
1635 		pr->pr_securelevel = slevel;
1636 		/* Set all child jails to be at least this level. */
1637 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1638 			if (tpr->pr_securelevel < slevel)
1639 				tpr->pr_securelevel = slevel;
1640 	}
1641 	if (gotchildmax) {
1642 		pr->pr_childmax = childmax;
1643 		/* Set all child jails to under this limit. */
1644 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1645 			if (tpr->pr_childmax > childmax - level)
1646 				tpr->pr_childmax = childmax > level
1647 				    ? childmax - level : 0;
1648 	}
1649 	if (gotenforce) {
1650 		pr->pr_enforce_statfs = enforce;
1651 		/* Pass this restriction on to the children. */
1652 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1653 			if (tpr->pr_enforce_statfs < enforce)
1654 				tpr->pr_enforce_statfs = enforce;
1655 	}
1656 	if (gotrsnum) {
1657 		pr->pr_devfs_rsnum = rsnum;
1658 		/* Pass this restriction on to the children. */
1659 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1660 			tpr->pr_devfs_rsnum = rsnum;
1661 	}
1662 	if (namelc != NULL) {
1663 		if (ppr == &prison0)
1664 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1665 		else
1666 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1667 			    ppr->pr_name, namelc);
1668 		/* Change this component of child names. */
1669 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1670 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1671 			    strlen(tpr->pr_name + onamelen) + 1);
1672 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1673 		}
1674 	}
1675 	if (path != NULL) {
1676 		/* Try to keep a real-rooted full pathname. */
1677 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1678 		pr->pr_root = root;
1679 		root = NULL;
1680 	}
1681 	if (PR_HOST & ch_flags & ~pr_flags) {
1682 		if (pr->pr_flags & PR_HOST) {
1683 			/*
1684 			 * Copy the parent's host info.  As with pr_ip4 above,
1685 			 * the lack of a lock on the parent is not a problem;
1686 			 * it is always set with allprison_lock at least
1687 			 * shared, and is held exclusively here.
1688 			 */
1689 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1690 			    sizeof(pr->pr_hostname));
1691 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1692 			    sizeof(pr->pr_domainname));
1693 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1694 			    sizeof(pr->pr_hostuuid));
1695 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1696 		}
1697 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1698 		/* Set this prison, and any descendants without PR_HOST. */
1699 		if (host != NULL)
1700 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1701 		if (domain != NULL)
1702 			strlcpy(pr->pr_domainname, domain,
1703 			    sizeof(pr->pr_domainname));
1704 		if (uuid != NULL)
1705 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1706 		if (gothid)
1707 			pr->pr_hostid = hid;
1708 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1709 			if (tpr->pr_flags & PR_HOST)
1710 				descend = 0;
1711 			else {
1712 				if (host != NULL)
1713 					strlcpy(tpr->pr_hostname,
1714 					    pr->pr_hostname,
1715 					    sizeof(tpr->pr_hostname));
1716 				if (domain != NULL)
1717 					strlcpy(tpr->pr_domainname,
1718 					    pr->pr_domainname,
1719 					    sizeof(tpr->pr_domainname));
1720 				if (uuid != NULL)
1721 					strlcpy(tpr->pr_hostuuid,
1722 					    pr->pr_hostuuid,
1723 					    sizeof(tpr->pr_hostuuid));
1724 				if (gothid)
1725 					tpr->pr_hostid = hid;
1726 			}
1727 		}
1728 	}
1729 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1730 	if ((tallow = ch_allow & ~pr_allow))
1731 		prison_set_allow_locked(pr, tallow, 0);
1732 	/*
1733 	 * Persistent prisons get an extra reference, and prisons losing their
1734 	 * persist flag lose that reference.  Only do this for existing prisons
1735 	 * for now, so new ones will remain unseen until after the module
1736 	 * handlers have completed.
1737 	 */
1738 	born = !prison_isalive(pr);
1739 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1740 		if (pr_flags & PR_PERSIST) {
1741 			prison_hold(pr);
1742 			refcount_acquire(&pr->pr_uref);
1743 		} else {
1744 			refcount_release(&pr->pr_ref);
1745 			drflags |= PD_DEUREF;
1746 		}
1747 	}
1748 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1749 	mtx_unlock(&pr->pr_mtx);
1750 	drflags &= ~PD_LOCKED;
1751 
1752 #ifdef RACCT
1753 	if (racct_enable && created)
1754 		prison_racct_attach(pr);
1755 #endif
1756 
1757 	/* Locks may have prevented a complete restriction of child IP
1758 	 * addresses.  If so, allocate some more memory and try again.
1759 	 */
1760 #ifdef INET
1761 	while (redo_ip4) {
1762 		ip4s = pr->pr_ip4s;
1763 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1764 		mtx_lock(&pr->pr_mtx);
1765 		redo_ip4 = 0;
1766 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1767 #ifdef VIMAGE
1768 			if (tpr->pr_flags & PR_VNET) {
1769 				descend = 0;
1770 				continue;
1771 			}
1772 #endif
1773 			if (prison_restrict_ip4(tpr, ip4)) {
1774 				if (ip4 != NULL)
1775 					ip4 = NULL;
1776 				else
1777 					redo_ip4 = 1;
1778 			}
1779 		}
1780 		mtx_unlock(&pr->pr_mtx);
1781 	}
1782 #endif
1783 #ifdef INET6
1784 	while (redo_ip6) {
1785 		ip6s = pr->pr_ip6s;
1786 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1787 		mtx_lock(&pr->pr_mtx);
1788 		redo_ip6 = 0;
1789 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1790 #ifdef VIMAGE
1791 			if (tpr->pr_flags & PR_VNET) {
1792 				descend = 0;
1793 				continue;
1794 			}
1795 #endif
1796 			if (prison_restrict_ip6(tpr, ip6)) {
1797 				if (ip6 != NULL)
1798 					ip6 = NULL;
1799 				else
1800 					redo_ip6 = 1;
1801 			}
1802 		}
1803 		mtx_unlock(&pr->pr_mtx);
1804 	}
1805 #endif
1806 
1807 	/* Let the modules do their work. */
1808 	sx_downgrade(&allprison_lock);
1809 	drflags = (drflags & ~PD_LIST_XLOCKED) | PD_LIST_SLOCKED;
1810 	if (born) {
1811 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1812 		if (error) {
1813 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1814 			goto done_deref;
1815 		}
1816 	}
1817 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1818 	if (error) {
1819 		if (born)
1820 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1821 		goto done_deref;
1822 	}
1823 
1824 	/* Attach this process to the prison if requested. */
1825 	if (flags & JAIL_ATTACH) {
1826 		mtx_lock(&pr->pr_mtx);
1827 		error = do_jail_attach(td, pr);
1828 		drflags &= ~PD_LIST_SLOCKED;
1829 		if (error) {
1830 			if (created) {
1831 				/* do_jail_attach has removed the prison. */
1832 				pr = NULL;
1833 			}
1834 			vfs_opterror(opts, "attach failed");
1835 			goto done_deref;
1836 		}
1837 	}
1838 
1839 #ifdef RACCT
1840 	if (racct_enable && !created) {
1841 		if (drflags & PD_LIST_SLOCKED) {
1842 			sx_sunlock(&allprison_lock);
1843 			drflags &= ~PD_LIST_SLOCKED;
1844 		}
1845 		prison_racct_modify(pr);
1846 	}
1847 #endif
1848 
1849 	td->td_retval[0] = pr->pr_id;
1850 
1851 	if (created) {
1852 		/*
1853 		 * Add a reference to newly created persistent prisons
1854 		 * (which was not done earlier so that the prison would
1855 		 * not be publicly visible).
1856 		 */
1857 		if (pr_flags & PR_PERSIST) {
1858 			mtx_lock(&pr->pr_mtx);
1859 			drflags |= PD_LOCKED;
1860 			refcount_acquire(&pr->pr_ref);
1861 			refcount_acquire(&pr->pr_uref);
1862 		} else {
1863 			/* Non-persistent jails need no further changes. */
1864 			pr = NULL;
1865 		}
1866 	}
1867 
1868  done_deref:
1869 	/* Release any temporary prison holds and/or locks. */
1870 	if (pr != NULL)
1871 		prison_deref(pr, drflags);
1872 	else if (drflags & PD_LIST_SLOCKED)
1873 		sx_sunlock(&allprison_lock);
1874 	else if (drflags & PD_LIST_XLOCKED)
1875 		sx_xunlock(&allprison_lock);
1876 	if (root != NULL)
1877 		vrele(root);
1878  done_errmsg:
1879 	if (error) {
1880 		/* Write the error message back to userspace. */
1881 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1882 		    &errmsg_len) == 0 && errmsg_len > 0) {
1883 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1884 			if (optuio->uio_segflg == UIO_SYSSPACE)
1885 				bcopy(errmsg,
1886 				    optuio->uio_iov[errmsg_pos].iov_base,
1887 				    errmsg_len);
1888 			else
1889 				copyout(errmsg,
1890 				    optuio->uio_iov[errmsg_pos].iov_base,
1891 				    errmsg_len);
1892 		}
1893 	}
1894  done_free:
1895 #ifdef INET
1896 	free(ip4, M_PRISON);
1897 #endif
1898 #ifdef INET6
1899 	free(ip6, M_PRISON);
1900 #endif
1901 	if (g_path != NULL)
1902 		free(g_path, M_TEMP);
1903 	vfs_freeopts(opts);
1904 	return (error);
1905 }
1906 
1907 /*
1908  * Find the next available prison ID.  Return the ID on success, or zero
1909  * on failure.  Also set a pointer to the allprison list entry the prison
1910  * should be inserted before.
1911  */
1912 static int
1913 get_next_prid(struct prison **insprp)
1914 {
1915 	struct prison *inspr;
1916 	int jid, maxid;
1917 
1918 	jid = lastprid % JAIL_MAX + 1;
1919 	if (TAILQ_EMPTY(&allprison) ||
1920 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
1921 		/*
1922 		 * A common case is for all jails to be implicitly numbered,
1923 		 * which means they'll go on the end of the list, at least
1924 		 * for the first JAIL_MAX times.
1925 		 */
1926 		inspr = NULL;
1927 	} else {
1928 		/*
1929 		 * Take two passes through the allprison list: first starting
1930 		 * with the proposed jid, then ending with it.
1931 		 */
1932 		for (maxid = JAIL_MAX; maxid != 0; ) {
1933 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
1934 				if (inspr->pr_id < jid)
1935 					continue;
1936 				if (inspr->pr_id > jid ||
1937 				    refcount_load(&inspr->pr_ref) == 0) {
1938 					/*
1939 					 * Found an opening.  This may be a gap
1940 					 * in the list, or a dead jail with the
1941 					 * same ID.
1942 					 */
1943 					maxid = 0;
1944 					break;
1945 				}
1946 				if (++jid > maxid) {
1947 					if (lastprid == maxid || lastprid == 0)
1948 					{
1949 						/*
1950 						 * The entire legal range
1951 						 * has been traversed
1952 						 */
1953 						return 0;
1954 					}
1955 					/* Try again from the start. */
1956 					jid = 1;
1957 					maxid = lastprid;
1958 					break;
1959 				}
1960 			}
1961 			if (inspr == NULL) {
1962 				/* Found room at the end of the list. */
1963 				break;
1964 			}
1965 		}
1966 	}
1967 	*insprp = inspr;
1968 	lastprid = jid;
1969 	return (jid);
1970 }
1971 
1972 /*
1973  * struct jail_get_args {
1974  *	struct iovec *iovp;
1975  *	unsigned int iovcnt;
1976  *	int flags;
1977  * };
1978  */
1979 int
1980 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1981 {
1982 	struct uio *auio;
1983 	int error;
1984 
1985 	/* Check that we have an even number of iovecs. */
1986 	if (uap->iovcnt & 1)
1987 		return (EINVAL);
1988 
1989 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1990 	if (error)
1991 		return (error);
1992 	error = kern_jail_get(td, auio, uap->flags);
1993 	if (error == 0)
1994 		error = copyout(auio->uio_iov, uap->iovp,
1995 		    uap->iovcnt * sizeof (struct iovec));
1996 	free(auio, M_IOV);
1997 	return (error);
1998 }
1999 
2000 int
2001 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2002 {
2003 	struct bool_flags *bf;
2004 	struct jailsys_flags *jsf;
2005 	struct prison *pr, *mypr;
2006 	struct vfsopt *opt;
2007 	struct vfsoptlist *opts;
2008 	char *errmsg, *name;
2009 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2010 	unsigned f;
2011 
2012 	if (flags & ~JAIL_GET_MASK)
2013 		return (EINVAL);
2014 
2015 	/* Get the parameter list. */
2016 	error = vfs_buildopts(optuio, &opts);
2017 	if (error)
2018 		return (error);
2019 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2020 	mypr = td->td_ucred->cr_prison;
2021 	pr = NULL;
2022 
2023 	/*
2024 	 * Find the prison specified by one of: lastjid, jid, name.
2025 	 */
2026 	sx_slock(&allprison_lock);
2027 	drflags = PD_LIST_SLOCKED;
2028 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2029 	if (error == 0) {
2030 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2031 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2032 				mtx_lock(&pr->pr_mtx);
2033 				if ((flags & JAIL_DYING)
2034 				    ? prison_isvalid(pr) : prison_isalive(pr))
2035 					break;
2036 				mtx_unlock(&pr->pr_mtx);
2037 			}
2038 		}
2039 		if (pr != NULL) {
2040 			drflags |= PD_LOCKED;
2041 			goto found_prison;
2042 		}
2043 		error = ENOENT;
2044 		vfs_opterror(opts, "no jail after %d", jid);
2045 		goto done;
2046 	} else if (error != ENOENT)
2047 		goto done;
2048 
2049 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2050 	if (error == 0) {
2051 		if (jid != 0) {
2052 			pr = prison_find_child(mypr, jid);
2053 			if (pr != NULL) {
2054 				drflags |= PD_LOCKED;
2055 				if (!(prison_isalive(pr) ||
2056 				    (flags & JAIL_DYING))) {
2057 					error = ENOENT;
2058 					vfs_opterror(opts, "jail %d is dying",
2059 					    jid);
2060 					goto done;
2061 				}
2062 				goto found_prison;
2063 			}
2064 			error = ENOENT;
2065 			vfs_opterror(opts, "jail %d not found", jid);
2066 			goto done;
2067 		}
2068 	} else if (error != ENOENT)
2069 		goto done;
2070 
2071 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2072 	if (error == 0) {
2073 		if (len == 0 || name[len - 1] != '\0') {
2074 			error = EINVAL;
2075 			goto done;
2076 		}
2077 		pr = prison_find_name(mypr, name);
2078 		if (pr != NULL) {
2079 			drflags |= PD_LOCKED;
2080 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2081 				error = ENOENT;
2082 				vfs_opterror(opts, "jail \"%s\" is dying",
2083 				    name);
2084 				goto done;
2085 			}
2086 			goto found_prison;
2087 		}
2088 		error = ENOENT;
2089 		vfs_opterror(opts, "jail \"%s\" not found", name);
2090 		goto done;
2091 	} else if (error != ENOENT)
2092 		goto done;
2093 
2094 	vfs_opterror(opts, "no jail specified");
2095 	error = ENOENT;
2096 	goto done;
2097 
2098  found_prison:
2099 	/* Get the parameters of the prison. */
2100 	prison_hold(pr);
2101 	drflags |= PD_DEREF;
2102 	td->td_retval[0] = pr->pr_id;
2103 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2104 	if (error != 0 && error != ENOENT)
2105 		goto done;
2106 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2107 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2108 	if (error != 0 && error != ENOENT)
2109 		goto done;
2110 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2111 	if (error != 0 && error != ENOENT)
2112 		goto done;
2113 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2114 	    sizeof(pr->pr_cpuset->cs_id));
2115 	if (error != 0 && error != ENOENT)
2116 		goto done;
2117 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2118 	if (error != 0 && error != ENOENT)
2119 		goto done;
2120 #ifdef INET
2121 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2122 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2123 	if (error != 0 && error != ENOENT)
2124 		goto done;
2125 #endif
2126 #ifdef INET6
2127 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2128 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2129 	if (error != 0 && error != ENOENT)
2130 		goto done;
2131 #endif
2132 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2133 	    sizeof(pr->pr_securelevel));
2134 	if (error != 0 && error != ENOENT)
2135 		goto done;
2136 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2137 	    sizeof(pr->pr_childcount));
2138 	if (error != 0 && error != ENOENT)
2139 		goto done;
2140 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2141 	    sizeof(pr->pr_childmax));
2142 	if (error != 0 && error != ENOENT)
2143 		goto done;
2144 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2145 	if (error != 0 && error != ENOENT)
2146 		goto done;
2147 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2148 	if (error != 0 && error != ENOENT)
2149 		goto done;
2150 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2151 	if (error != 0 && error != ENOENT)
2152 		goto done;
2153 #ifdef COMPAT_FREEBSD32
2154 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2155 		uint32_t hid32 = pr->pr_hostid;
2156 
2157 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2158 	} else
2159 #endif
2160 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2161 	    sizeof(pr->pr_hostid));
2162 	if (error != 0 && error != ENOENT)
2163 		goto done;
2164 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2165 	    sizeof(pr->pr_enforce_statfs));
2166 	if (error != 0 && error != ENOENT)
2167 		goto done;
2168 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2169 	    sizeof(pr->pr_devfs_rsnum));
2170 	if (error != 0 && error != ENOENT)
2171 		goto done;
2172 	for (bf = pr_flag_bool;
2173 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2174 	     bf++) {
2175 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2176 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2177 		if (error != 0 && error != ENOENT)
2178 			goto done;
2179 		i = !i;
2180 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2181 		if (error != 0 && error != ENOENT)
2182 			goto done;
2183 	}
2184 	for (jsf = pr_flag_jailsys;
2185 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2186 	     jsf++) {
2187 		f = pr->pr_flags & (jsf->disable | jsf->new);
2188 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2189 		    : (f == jsf->new) ? JAIL_SYS_NEW
2190 		    : JAIL_SYS_INHERIT;
2191 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2192 		if (error != 0 && error != ENOENT)
2193 			goto done;
2194 	}
2195 	for (bf = pr_flag_allow;
2196 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2197 		atomic_load_int(&bf->flag) != 0;
2198 	     bf++) {
2199 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2200 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2201 		if (error != 0 && error != ENOENT)
2202 			goto done;
2203 		i = !i;
2204 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2205 		if (error != 0 && error != ENOENT)
2206 			goto done;
2207 	}
2208 	i = !prison_isalive(pr);
2209 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2210 	if (error != 0 && error != ENOENT)
2211 		goto done;
2212 	i = !i;
2213 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2214 	if (error != 0 && error != ENOENT)
2215 		goto done;
2216 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2217 	    sizeof(pr->pr_osreldate));
2218 	if (error != 0 && error != ENOENT)
2219 		goto done;
2220 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2221 	if (error != 0 && error != ENOENT)
2222 		goto done;
2223 
2224 	/* Get the module parameters. */
2225 	mtx_unlock(&pr->pr_mtx);
2226 	drflags &= ~PD_LOCKED;
2227 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2228 	if (error)
2229 		goto done;
2230 	prison_deref(pr, drflags);
2231 	pr = NULL;
2232 	drflags = 0;
2233 
2234 	/* By now, all parameters should have been noted. */
2235 	TAILQ_FOREACH(opt, opts, link) {
2236 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2237 			error = EINVAL;
2238 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2239 			goto done;
2240 		}
2241 	}
2242 
2243 	/* Write the fetched parameters back to userspace. */
2244 	error = 0;
2245 	TAILQ_FOREACH(opt, opts, link) {
2246 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2247 			pos = 2 * opt->pos + 1;
2248 			optuio->uio_iov[pos].iov_len = opt->len;
2249 			if (opt->value != NULL) {
2250 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2251 					bcopy(opt->value,
2252 					    optuio->uio_iov[pos].iov_base,
2253 					    opt->len);
2254 				} else {
2255 					error = copyout(opt->value,
2256 					    optuio->uio_iov[pos].iov_base,
2257 					    opt->len);
2258 					if (error)
2259 						break;
2260 				}
2261 			}
2262 		}
2263 	}
2264 
2265  done:
2266 	/* Release any temporary prison holds and/or locks. */
2267 	if (pr != NULL)
2268 		prison_deref(pr, drflags);
2269 	else if (drflags & PD_LIST_SLOCKED)
2270 		sx_sunlock(&allprison_lock);
2271 	if (error && errmsg_pos >= 0) {
2272 		/* Write the error message back to userspace. */
2273 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2274 		errmsg_pos = 2 * errmsg_pos + 1;
2275 		if (errmsg_len > 0) {
2276 			if (optuio->uio_segflg == UIO_SYSSPACE)
2277 				bcopy(errmsg,
2278 				    optuio->uio_iov[errmsg_pos].iov_base,
2279 				    errmsg_len);
2280 			else
2281 				copyout(errmsg,
2282 				    optuio->uio_iov[errmsg_pos].iov_base,
2283 				    errmsg_len);
2284 		}
2285 	}
2286 	vfs_freeopts(opts);
2287 	return (error);
2288 }
2289 
2290 /*
2291  * struct jail_remove_args {
2292  *	int jid;
2293  * };
2294  */
2295 int
2296 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2297 {
2298 	struct prison *pr, *cpr, *lpr, *tpr;
2299 	int descend, error;
2300 
2301 	error = priv_check(td, PRIV_JAIL_REMOVE);
2302 	if (error)
2303 		return (error);
2304 
2305 	sx_xlock(&allprison_lock);
2306 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2307 	if (pr == NULL) {
2308 		sx_xunlock(&allprison_lock);
2309 		return (EINVAL);
2310 	}
2311 
2312 	/* Remove all descendants of this prison, then remove this prison. */
2313 	prison_hold(pr);
2314 	if (!LIST_EMPTY(&pr->pr_children)) {
2315 		mtx_unlock(&pr->pr_mtx);
2316 		lpr = NULL;
2317 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2318 			mtx_lock(&cpr->pr_mtx);
2319 			if (prison_isvalid(cpr)) {
2320 				tpr = cpr;
2321 				prison_hold(cpr);
2322 			} else {
2323 				/* Already removed - do not do it again. */
2324 				tpr = NULL;
2325 			}
2326 			mtx_unlock(&cpr->pr_mtx);
2327 			if (lpr != NULL) {
2328 				mtx_lock(&lpr->pr_mtx);
2329 				prison_remove_one(lpr);
2330 				sx_xlock(&allprison_lock);
2331 			}
2332 			lpr = tpr;
2333 		}
2334 		if (lpr != NULL) {
2335 			mtx_lock(&lpr->pr_mtx);
2336 			prison_remove_one(lpr);
2337 			sx_xlock(&allprison_lock);
2338 		}
2339 		mtx_lock(&pr->pr_mtx);
2340 	}
2341 	prison_remove_one(pr);
2342 	return (0);
2343 }
2344 
2345 static void
2346 prison_remove_one(struct prison *pr)
2347 {
2348 	struct proc *p;
2349 	int drflags;
2350 
2351 	drflags = PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED;
2352 
2353 	/* If the prison was persistent, it is not anymore. */
2354 	if (pr->pr_flags & PR_PERSIST) {
2355 		refcount_release(&pr->pr_ref);
2356 		drflags |= PD_DEUREF;
2357 		pr->pr_flags &= ~PR_PERSIST;
2358 	}
2359 
2360 	/*
2361 	 * jail_remove added a reference.  If that's the only one, remove
2362 	 * the prison now.  refcount(9) doesn't guarantee the cache coherence
2363 	 * of non-zero counters, so force it here.
2364 	 */
2365 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2366 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2367 	if (atomic_load_acq_int(&pr->pr_ref) == 1) {
2368 		prison_deref(pr, drflags);
2369 		return;
2370 	}
2371 
2372 	mtx_unlock(&pr->pr_mtx);
2373 	sx_xunlock(&allprison_lock);
2374 	drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2375 	/*
2376 	 * Kill all processes unfortunate enough to be attached to this prison.
2377 	 */
2378 	sx_slock(&allproc_lock);
2379 	FOREACH_PROC_IN_SYSTEM(p) {
2380 		PROC_LOCK(p);
2381 		if (p->p_state != PRS_NEW && p->p_ucred &&
2382 		    p->p_ucred->cr_prison == pr)
2383 			kern_psignal(p, SIGKILL);
2384 		PROC_UNLOCK(p);
2385 	}
2386 	sx_sunlock(&allproc_lock);
2387 	/* Remove the temporary reference added by jail_remove. */
2388 	prison_deref(pr, drflags);
2389 }
2390 
2391 /*
2392  * struct jail_attach_args {
2393  *	int jid;
2394  * };
2395  */
2396 int
2397 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2398 {
2399 	struct prison *pr;
2400 	int error;
2401 
2402 	error = priv_check(td, PRIV_JAIL_ATTACH);
2403 	if (error)
2404 		return (error);
2405 
2406 	/*
2407 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2408 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2409 	 * But then immediately downgrade it since we don't need to stop
2410 	 * readers.
2411 	 */
2412 	sx_xlock(&allprison_lock);
2413 	sx_downgrade(&allprison_lock);
2414 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2415 	if (pr == NULL) {
2416 		sx_sunlock(&allprison_lock);
2417 		return (EINVAL);
2418 	}
2419 
2420 	/* Do not allow a process to attach to a prison that is not alive. */
2421 	if (!prison_isalive(pr)) {
2422 		mtx_unlock(&pr->pr_mtx);
2423 		sx_sunlock(&allprison_lock);
2424 		return (EINVAL);
2425 	}
2426 
2427 	return (do_jail_attach(td, pr));
2428 }
2429 
2430 static int
2431 do_jail_attach(struct thread *td, struct prison *pr)
2432 {
2433 	struct proc *p;
2434 	struct ucred *newcred, *oldcred;
2435 	int error;
2436 
2437 	/*
2438 	 * XXX: Note that there is a slight race here if two threads
2439 	 * in the same privileged process attempt to attach to two
2440 	 * different jails at the same time.  It is important for
2441 	 * user processes not to do this, or they might end up with
2442 	 * a process root from one prison, but attached to the jail
2443 	 * of another.
2444 	 */
2445 	refcount_acquire(&pr->pr_ref);
2446 	refcount_acquire(&pr->pr_uref);
2447 	mtx_unlock(&pr->pr_mtx);
2448 
2449 	/* Let modules do whatever they need to prepare for attaching. */
2450 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2451 	if (error) {
2452 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2453 		return (error);
2454 	}
2455 	sx_sunlock(&allprison_lock);
2456 
2457 	/*
2458 	 * Reparent the newly attached process to this jail.
2459 	 */
2460 	p = td->td_proc;
2461 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2462 	if (error)
2463 		goto e_revert_osd;
2464 
2465 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2466 	if ((error = change_dir(pr->pr_root, td)) != 0)
2467 		goto e_unlock;
2468 #ifdef MAC
2469 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2470 		goto e_unlock;
2471 #endif
2472 	VOP_UNLOCK(pr->pr_root);
2473 	if ((error = pwd_chroot(td, pr->pr_root)))
2474 		goto e_revert_osd;
2475 
2476 	newcred = crget();
2477 	PROC_LOCK(p);
2478 	oldcred = crcopysafe(p, newcred);
2479 	newcred->cr_prison = pr;
2480 	proc_set_cred(p, newcred);
2481 	setsugid(p);
2482 #ifdef RACCT
2483 	racct_proc_ucred_changed(p, oldcred, newcred);
2484 	crhold(newcred);
2485 #endif
2486 	PROC_UNLOCK(p);
2487 #ifdef RCTL
2488 	rctl_proc_ucred_changed(p, newcred);
2489 	crfree(newcred);
2490 #endif
2491 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2492 	crfree(oldcred);
2493 	return (0);
2494 
2495  e_unlock:
2496 	VOP_UNLOCK(pr->pr_root);
2497  e_revert_osd:
2498 	/* Tell modules this thread is still in its old jail after all. */
2499 	sx_slock(&allprison_lock);
2500 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2501 	prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2502 	return (error);
2503 }
2504 
2505 /*
2506  * Returns a locked prison instance, or NULL on failure.
2507  */
2508 struct prison *
2509 prison_find(int prid)
2510 {
2511 	struct prison *pr;
2512 
2513 	sx_assert(&allprison_lock, SX_LOCKED);
2514 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2515 		if (pr->pr_id == prid) {
2516 			mtx_lock(&pr->pr_mtx);
2517 			if (prison_isvalid(pr))
2518 				return (pr);
2519 			/*
2520 			 * Any active prison with the same ID would have
2521 			 * been inserted before a dead one.
2522 			 */
2523 			mtx_unlock(&pr->pr_mtx);
2524 			break;
2525 		}
2526 		if (pr->pr_id > prid)
2527 			break;
2528 	}
2529 	return (NULL);
2530 }
2531 
2532 /*
2533  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2534  */
2535 struct prison *
2536 prison_find_child(struct prison *mypr, int prid)
2537 {
2538 	struct prison *pr;
2539 	int descend;
2540 
2541 	sx_assert(&allprison_lock, SX_LOCKED);
2542 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2543 		if (pr->pr_id == prid) {
2544 			mtx_lock(&pr->pr_mtx);
2545 			if (prison_isvalid(pr))
2546 				return (pr);
2547 			mtx_unlock(&pr->pr_mtx);
2548 		}
2549 	}
2550 	return (NULL);
2551 }
2552 
2553 /*
2554  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2555  */
2556 struct prison *
2557 prison_find_name(struct prison *mypr, const char *name)
2558 {
2559 	struct prison *pr, *deadpr;
2560 	size_t mylen;
2561 	int descend;
2562 
2563 	sx_assert(&allprison_lock, SX_LOCKED);
2564 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2565  again:
2566 	deadpr = NULL;
2567 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2568 		if (!strcmp(pr->pr_name + mylen, name)) {
2569 			mtx_lock(&pr->pr_mtx);
2570 			if (prison_isalive(pr))
2571 				return (pr);
2572 			if (prison_isvalid(pr))
2573 				deadpr = pr;
2574 			mtx_unlock(&pr->pr_mtx);
2575 		}
2576 	}
2577 	/* There was no valid prison - perhaps there was a dying one. */
2578 	if (deadpr != NULL) {
2579 		mtx_lock(&deadpr->pr_mtx);
2580 		if (!prison_isvalid(deadpr)) {
2581 			mtx_unlock(&deadpr->pr_mtx);
2582 			goto again;
2583 		}
2584 	}
2585 	return (deadpr);
2586 }
2587 
2588 /*
2589  * See if a prison has the specific flag set.  The prison should be locked,
2590  * unless checking for flags that are only set at jail creation (such as
2591  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2592  * to any other prison data.
2593  */
2594 int
2595 prison_flag(struct ucred *cred, unsigned flag)
2596 {
2597 
2598 	return (cred->cr_prison->pr_flags & flag);
2599 }
2600 
2601 int
2602 prison_allow(struct ucred *cred, unsigned flag)
2603 {
2604 
2605 	return ((cred->cr_prison->pr_allow & flag) != 0);
2606 }
2607 
2608 /*
2609  * Hold a prison reference, by incrementing pr_ref.  It is generally
2610  * an error to hold a prison that does not already have a reference.
2611  * A prison record will remain valid as long as it has at least one
2612  * reference, and will not be removed as long as either the prison
2613  * mutex or the allprison lock is held (allprison_lock may be shared).
2614  */
2615 void
2616 prison_hold_locked(struct prison *pr)
2617 {
2618 
2619 	/* Locking is no longer required. */
2620 	prison_hold(pr);
2621 }
2622 
2623 void
2624 prison_hold(struct prison *pr)
2625 {
2626 #ifdef INVARIANTS
2627 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2628 
2629 	KASSERT(was_valid,
2630 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2631 #else
2632 	refcount_acquire(&pr->pr_ref);
2633 #endif
2634 }
2635 
2636 /*
2637  * Remove a prison reference.  If that was the last reference, the
2638  * prison will be removed (at a later time).  Return with the prison
2639  * unlocked.
2640  */
2641 void
2642 prison_free_locked(struct prison *pr)
2643 {
2644 	int lastref;
2645 
2646 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2647 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2648 	    ("Trying to free dead prison %p (jid=%d).",
2649 	     pr, pr->pr_id));
2650 	lastref = refcount_release(&pr->pr_ref);
2651 	mtx_unlock(&pr->pr_mtx);
2652 	if (lastref) {
2653 		/*
2654 		 * Don't remove the prison itself in this context,
2655 		 * in case there are locks held.
2656 		 */
2657 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2658 	}
2659 }
2660 
2661 void
2662 prison_free(struct prison *pr)
2663 {
2664 
2665 	/*
2666 	 * Locking is only required when releasing the last reference.
2667 	 * This allows assurance that a locked prison will remain valid
2668 	 * until it is unlocked.
2669 	 */
2670 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2671 	    ("Trying to free dead prison %p (jid=%d).",
2672 	     pr, pr->pr_id));
2673 	if (refcount_release_if_not_last(&pr->pr_ref))
2674 		return;
2675 	mtx_lock(&pr->pr_mtx);
2676 	prison_free_locked(pr);
2677 }
2678 
2679 /*
2680  * Hold a a prison for user visibility, by incrementing pr_uref.
2681  * It is generally an error to hold a prison that isn't already
2682  * user-visible, except through the the jail system calls.  It is also
2683  * an error to hold an invalid prison.  A prison record will remain
2684  * alive as long as it has at least one user reference, and will not
2685  * be set to the dying state was long as the prison mutex is held.
2686  */
2687 void
2688 prison_proc_hold(struct prison *pr)
2689 {
2690 #ifdef INVARIANTS
2691 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2692 
2693 	KASSERT(was_alive,
2694 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2695 #else
2696 	refcount_acquire(&pr->pr_uref);
2697 #endif
2698 }
2699 
2700 /*
2701  * Remove a prison user reference.  If it was the last reference, the
2702  * prison will be considered "dying", and may be removed once all of
2703  * its references are dropped.
2704  */
2705 void
2706 prison_proc_free(struct prison *pr)
2707 {
2708 
2709 	/*
2710 	 * Locking is only required when releasing the last reference.
2711 	 * This allows assurance that a locked prison will remain alive
2712 	 * until it is unlocked.
2713 	 */
2714 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2715 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2716 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2717 		/*
2718 		 * Don't remove the last user reference in this context,
2719 		 * which is expected to be a process that is not only locked,
2720 		 * but also half dead.  Add a reference so any calls to
2721 		 * prison_free() won't re-submit the task.
2722 		 */
2723 		refcount_acquire(&pr->pr_ref);
2724 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2725 	}
2726 }
2727 
2728 /*
2729  * Complete a call to either prison_free or prison_proc_free.
2730  */
2731 static void
2732 prison_complete(void *context, int pending)
2733 {
2734 	struct prison *pr = context;
2735 
2736 	sx_xlock(&allprison_lock);
2737 	mtx_lock(&pr->pr_mtx);
2738 	/*
2739 	 * If this is completing a call to prison_proc_free, there will still
2740 	 * be a user reference held; clear that as well as the reference that
2741 	 * was added.  No references are expected if this is completing a call
2742 	 * to prison_free, but prison_deref is still called for the cleanup.
2743 	 */
2744 	prison_deref(pr, refcount_load(&pr->pr_uref) > 0
2745 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2746 	    : PD_LOCKED | PD_LIST_XLOCKED);
2747 }
2748 
2749 /*
2750  * Remove a prison reference and/or user reference (usually).
2751  * This assumes context that allows sleeping (for allprison_lock),
2752  * with no non-sleeping locks held, except perhaps the prison itself.
2753  * If there are no more references, release and delist the prison.
2754  * On completion, the prison lock and the allprison lock are both
2755  * unlocked.
2756  */
2757 static void
2758 prison_deref(struct prison *pr, int flags)
2759 {
2760 	struct prison *ppr, *tpr;
2761 	int lastref, lasturef;
2762 
2763 	if (!(flags & PD_LOCKED))
2764 		mtx_lock(&pr->pr_mtx);
2765 	for (;;) {
2766 		if (flags & PD_DEUREF) {
2767 			KASSERT(refcount_load(&pr->pr_uref) > 0,
2768 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2769 			     pr->pr_id));
2770 			lasturef = refcount_release(&pr->pr_uref);
2771 			if (lasturef)
2772 				refcount_acquire(&pr->pr_ref);
2773 			KASSERT(refcount_load(&prison0.pr_uref) > 0,
2774 			    ("prison0 pr_uref=0"));
2775 		} else
2776 			lasturef = 0;
2777 		if (flags & PD_DEREF) {
2778 			KASSERT(refcount_load(&pr->pr_ref) > 0,
2779 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2780 			     pr->pr_id));
2781 			lastref = refcount_release(&pr->pr_ref);
2782 		}
2783 		else
2784 			lastref = refcount_load(&pr->pr_ref) == 0;
2785 		mtx_unlock(&pr->pr_mtx);
2786 
2787 		/*
2788 		 * Tell the modules if the last user reference was removed
2789 		 * (even it sticks around in dying state).
2790 		 */
2791 		if (lasturef) {
2792 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2793 				if (atomic_load_acq_int(&pr->pr_ref) > 1) {
2794 					sx_slock(&allprison_lock);
2795 					flags |= PD_LIST_SLOCKED;
2796 				} else {
2797 					sx_xlock(&allprison_lock);
2798 					flags |= PD_LIST_XLOCKED;
2799 				}
2800 			}
2801 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2802 			mtx_lock(&pr->pr_mtx);
2803 			lastref = refcount_release(&pr->pr_ref);
2804 			mtx_unlock(&pr->pr_mtx);
2805 		}
2806 
2807 		/* If the prison still has references, nothing else to do. */
2808 		if (!lastref) {
2809 			if (flags & PD_LIST_SLOCKED)
2810 				sx_sunlock(&allprison_lock);
2811 			else if (flags & PD_LIST_XLOCKED)
2812 				sx_xunlock(&allprison_lock);
2813 			return;
2814 		}
2815 
2816 		if (flags & PD_LIST_SLOCKED) {
2817 			if (!sx_try_upgrade(&allprison_lock)) {
2818 				sx_sunlock(&allprison_lock);
2819 				sx_xlock(&allprison_lock);
2820 			}
2821 		} else if (!(flags & PD_LIST_XLOCKED))
2822 			sx_xlock(&allprison_lock);
2823 
2824 		TAILQ_REMOVE(&allprison, pr, pr_list);
2825 		LIST_REMOVE(pr, pr_sibling);
2826 		ppr = pr->pr_parent;
2827 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2828 			tpr->pr_childcount--;
2829 		sx_xunlock(&allprison_lock);
2830 
2831 #ifdef VIMAGE
2832 		if (pr->pr_vnet != ppr->pr_vnet)
2833 			vnet_destroy(pr->pr_vnet);
2834 #endif
2835 		if (pr->pr_root != NULL)
2836 			vrele(pr->pr_root);
2837 		mtx_destroy(&pr->pr_mtx);
2838 #ifdef INET
2839 		free(pr->pr_ip4, M_PRISON);
2840 #endif
2841 #ifdef INET6
2842 		free(pr->pr_ip6, M_PRISON);
2843 #endif
2844 		if (pr->pr_cpuset != NULL)
2845 			cpuset_rel(pr->pr_cpuset);
2846 		osd_jail_exit(pr);
2847 #ifdef RACCT
2848 		if (racct_enable)
2849 			prison_racct_detach(pr);
2850 #endif
2851 		free(pr, M_PRISON);
2852 
2853 		/* Removing a prison frees a reference on its parent. */
2854 		pr = ppr;
2855 		mtx_lock(&pr->pr_mtx);
2856 		flags = PD_DEREF | PD_DEUREF;
2857 	}
2858 }
2859 
2860 /*
2861  * Set or clear a permission bit in the pr_allow field, passing restrictions
2862  * (cleared permission) down to child jails.
2863  */
2864 void
2865 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
2866 {
2867 	struct prison *pr;
2868 
2869 	pr = cred->cr_prison;
2870 	sx_slock(&allprison_lock);
2871 	mtx_lock(&pr->pr_mtx);
2872 	prison_set_allow_locked(pr, flag, enable);
2873 	mtx_unlock(&pr->pr_mtx);
2874 	sx_sunlock(&allprison_lock);
2875 }
2876 
2877 static void
2878 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
2879 {
2880 	struct prison *cpr;
2881 	int descend;
2882 
2883 	if (enable != 0)
2884 		pr->pr_allow |= flag;
2885 	else {
2886 		pr->pr_allow &= ~flag;
2887 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
2888 			cpr->pr_allow &= ~flag;
2889 	}
2890 }
2891 
2892 /*
2893  * Check if a jail supports the given address family.
2894  *
2895  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2896  * if not.
2897  */
2898 int
2899 prison_check_af(struct ucred *cred, int af)
2900 {
2901 	struct prison *pr;
2902 	int error;
2903 
2904 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2905 
2906 	pr = cred->cr_prison;
2907 #ifdef VIMAGE
2908 	/* Prisons with their own network stack are not limited. */
2909 	if (prison_owns_vnet(cred))
2910 		return (0);
2911 #endif
2912 
2913 	error = 0;
2914 	switch (af)
2915 	{
2916 #ifdef INET
2917 	case AF_INET:
2918 		if (pr->pr_flags & PR_IP4)
2919 		{
2920 			mtx_lock(&pr->pr_mtx);
2921 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2922 				error = EAFNOSUPPORT;
2923 			mtx_unlock(&pr->pr_mtx);
2924 		}
2925 		break;
2926 #endif
2927 #ifdef INET6
2928 	case AF_INET6:
2929 		if (pr->pr_flags & PR_IP6)
2930 		{
2931 			mtx_lock(&pr->pr_mtx);
2932 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2933 				error = EAFNOSUPPORT;
2934 			mtx_unlock(&pr->pr_mtx);
2935 		}
2936 		break;
2937 #endif
2938 	case AF_LOCAL:
2939 	case AF_ROUTE:
2940 		break;
2941 	default:
2942 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2943 			error = EAFNOSUPPORT;
2944 	}
2945 	return (error);
2946 }
2947 
2948 /*
2949  * Check if given address belongs to the jail referenced by cred (wrapper to
2950  * prison_check_ip[46]).
2951  *
2952  * Returns 0 if jail doesn't restrict the address family or if address belongs
2953  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2954  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2955  */
2956 int
2957 prison_if(struct ucred *cred, const struct sockaddr *sa)
2958 {
2959 #ifdef INET
2960 	const struct sockaddr_in *sai;
2961 #endif
2962 #ifdef INET6
2963 	const struct sockaddr_in6 *sai6;
2964 #endif
2965 	int error;
2966 
2967 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2968 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2969 
2970 #ifdef VIMAGE
2971 	if (prison_owns_vnet(cred))
2972 		return (0);
2973 #endif
2974 
2975 	error = 0;
2976 	switch (sa->sa_family)
2977 	{
2978 #ifdef INET
2979 	case AF_INET:
2980 		sai = (const struct sockaddr_in *)sa;
2981 		error = prison_check_ip4(cred, &sai->sin_addr);
2982 		break;
2983 #endif
2984 #ifdef INET6
2985 	case AF_INET6:
2986 		sai6 = (const struct sockaddr_in6 *)sa;
2987 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2988 		break;
2989 #endif
2990 	default:
2991 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2992 			error = EAFNOSUPPORT;
2993 	}
2994 	return (error);
2995 }
2996 
2997 /*
2998  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2999  */
3000 int
3001 prison_check(struct ucred *cred1, struct ucred *cred2)
3002 {
3003 
3004 	return ((cred1->cr_prison == cred2->cr_prison ||
3005 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3006 }
3007 
3008 /*
3009  * Return 1 if p2 is a child of p1, otherwise 0.
3010  */
3011 int
3012 prison_ischild(struct prison *pr1, struct prison *pr2)
3013 {
3014 
3015 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3016 		if (pr1 == pr2)
3017 			return (1);
3018 	return (0);
3019 }
3020 
3021 /*
3022  * Return true if the prison is currently alive.  A prison is alive if it is
3023  * valid and it holds user references.
3024  */
3025 bool
3026 prison_isalive(struct prison *pr)
3027 {
3028 
3029 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3030 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3031 		return (false);
3032 	if (__predict_false(refcount_load(&pr->pr_uref) == 0))
3033 		return (false);
3034 	return (true);
3035 }
3036 
3037 /*
3038  * Return true if the prison is currently valid.  A prison is valid if it has
3039  * been fully created, and is not being destroyed.  Note that dying prisons
3040  * are still considered valid.
3041  */
3042 bool
3043 prison_isvalid(struct prison *pr)
3044 {
3045 
3046 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3047 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3048 		return (false);
3049 	return (true);
3050 }
3051 
3052 /*
3053  * Return 1 if the passed credential is in a jail and that jail does not
3054  * have its own virtual network stack, otherwise 0.
3055  */
3056 int
3057 jailed_without_vnet(struct ucred *cred)
3058 {
3059 
3060 	if (!jailed(cred))
3061 		return (0);
3062 #ifdef VIMAGE
3063 	if (prison_owns_vnet(cred))
3064 		return (0);
3065 #endif
3066 
3067 	return (1);
3068 }
3069 
3070 /*
3071  * Return the correct hostname (domainname, et al) for the passed credential.
3072  */
3073 void
3074 getcredhostname(struct ucred *cred, char *buf, size_t size)
3075 {
3076 	struct prison *pr;
3077 
3078 	/*
3079 	 * A NULL credential can be used to shortcut to the physical
3080 	 * system's hostname.
3081 	 */
3082 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3083 	mtx_lock(&pr->pr_mtx);
3084 	strlcpy(buf, pr->pr_hostname, size);
3085 	mtx_unlock(&pr->pr_mtx);
3086 }
3087 
3088 void
3089 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3090 {
3091 
3092 	mtx_lock(&cred->cr_prison->pr_mtx);
3093 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3094 	mtx_unlock(&cred->cr_prison->pr_mtx);
3095 }
3096 
3097 void
3098 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3099 {
3100 
3101 	mtx_lock(&cred->cr_prison->pr_mtx);
3102 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3103 	mtx_unlock(&cred->cr_prison->pr_mtx);
3104 }
3105 
3106 void
3107 getcredhostid(struct ucred *cred, unsigned long *hostid)
3108 {
3109 
3110 	mtx_lock(&cred->cr_prison->pr_mtx);
3111 	*hostid = cred->cr_prison->pr_hostid;
3112 	mtx_unlock(&cred->cr_prison->pr_mtx);
3113 }
3114 
3115 void
3116 getjailname(struct ucred *cred, char *name, size_t len)
3117 {
3118 
3119 	mtx_lock(&cred->cr_prison->pr_mtx);
3120 	strlcpy(name, cred->cr_prison->pr_name, len);
3121 	mtx_unlock(&cred->cr_prison->pr_mtx);
3122 }
3123 
3124 #ifdef VIMAGE
3125 /*
3126  * Determine whether the prison represented by cred owns
3127  * its vnet rather than having it inherited.
3128  *
3129  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3130  */
3131 int
3132 prison_owns_vnet(struct ucred *cred)
3133 {
3134 
3135 	/*
3136 	 * vnets cannot be added/removed after jail creation,
3137 	 * so no need to lock here.
3138 	 */
3139 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3140 }
3141 #endif
3142 
3143 /*
3144  * Determine whether the subject represented by cred can "see"
3145  * status of a mount point.
3146  * Returns: 0 for permitted, ENOENT otherwise.
3147  * XXX: This function should be called cr_canseemount() and should be
3148  *      placed in kern_prot.c.
3149  */
3150 int
3151 prison_canseemount(struct ucred *cred, struct mount *mp)
3152 {
3153 	struct prison *pr;
3154 	struct statfs *sp;
3155 	size_t len;
3156 
3157 	pr = cred->cr_prison;
3158 	if (pr->pr_enforce_statfs == 0)
3159 		return (0);
3160 	if (pr->pr_root->v_mount == mp)
3161 		return (0);
3162 	if (pr->pr_enforce_statfs == 2)
3163 		return (ENOENT);
3164 	/*
3165 	 * If jail's chroot directory is set to "/" we should be able to see
3166 	 * all mount-points from inside a jail.
3167 	 * This is ugly check, but this is the only situation when jail's
3168 	 * directory ends with '/'.
3169 	 */
3170 	if (strcmp(pr->pr_path, "/") == 0)
3171 		return (0);
3172 	len = strlen(pr->pr_path);
3173 	sp = &mp->mnt_stat;
3174 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3175 		return (ENOENT);
3176 	/*
3177 	 * Be sure that we don't have situation where jail's root directory
3178 	 * is "/some/path" and mount point is "/some/pathpath".
3179 	 */
3180 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3181 		return (ENOENT);
3182 	return (0);
3183 }
3184 
3185 void
3186 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3187 {
3188 	char jpath[MAXPATHLEN];
3189 	struct prison *pr;
3190 	size_t len;
3191 
3192 	pr = cred->cr_prison;
3193 	if (pr->pr_enforce_statfs == 0)
3194 		return;
3195 	if (prison_canseemount(cred, mp) != 0) {
3196 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3197 		strlcpy(sp->f_mntonname, "[restricted]",
3198 		    sizeof(sp->f_mntonname));
3199 		return;
3200 	}
3201 	if (pr->pr_root->v_mount == mp) {
3202 		/*
3203 		 * Clear current buffer data, so we are sure nothing from
3204 		 * the valid path left there.
3205 		 */
3206 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3207 		*sp->f_mntonname = '/';
3208 		return;
3209 	}
3210 	/*
3211 	 * If jail's chroot directory is set to "/" we should be able to see
3212 	 * all mount-points from inside a jail.
3213 	 */
3214 	if (strcmp(pr->pr_path, "/") == 0)
3215 		return;
3216 	len = strlen(pr->pr_path);
3217 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3218 	/*
3219 	 * Clear current buffer data, so we are sure nothing from
3220 	 * the valid path left there.
3221 	 */
3222 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3223 	if (*jpath == '\0') {
3224 		/* Should never happen. */
3225 		*sp->f_mntonname = '/';
3226 	} else {
3227 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3228 	}
3229 }
3230 
3231 /*
3232  * Check with permission for a specific privilege is granted within jail.  We
3233  * have a specific list of accepted privileges; the rest are denied.
3234  */
3235 int
3236 prison_priv_check(struct ucred *cred, int priv)
3237 {
3238 	struct prison *pr;
3239 	int error;
3240 
3241 	/*
3242 	 * Some policies have custom handlers. This routine should not be
3243 	 * called for them. See priv_check_cred().
3244 	 */
3245 	switch (priv) {
3246 	case PRIV_VFS_LOOKUP:
3247 	case PRIV_VFS_GENERATION:
3248 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3249 		    "called for %d\n", priv));
3250 	}
3251 
3252 	if (!jailed(cred))
3253 		return (0);
3254 
3255 #ifdef VIMAGE
3256 	/*
3257 	 * Privileges specific to prisons with a virtual network stack.
3258 	 * There might be a duplicate entry here in case the privilege
3259 	 * is only granted conditionally in the legacy jail case.
3260 	 */
3261 	switch (priv) {
3262 #ifdef notyet
3263 		/*
3264 		 * NFS-specific privileges.
3265 		 */
3266 	case PRIV_NFS_DAEMON:
3267 	case PRIV_NFS_LOCKD:
3268 #endif
3269 		/*
3270 		 * Network stack privileges.
3271 		 */
3272 	case PRIV_NET_BRIDGE:
3273 	case PRIV_NET_GRE:
3274 	case PRIV_NET_BPF:
3275 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3276 	case PRIV_NET_ROUTE:
3277 	case PRIV_NET_TAP:
3278 	case PRIV_NET_SETIFMTU:
3279 	case PRIV_NET_SETIFFLAGS:
3280 	case PRIV_NET_SETIFCAP:
3281 	case PRIV_NET_SETIFDESCR:
3282 	case PRIV_NET_SETIFNAME	:
3283 	case PRIV_NET_SETIFMETRIC:
3284 	case PRIV_NET_SETIFPHYS:
3285 	case PRIV_NET_SETIFMAC:
3286 	case PRIV_NET_SETLANPCP:
3287 	case PRIV_NET_ADDMULTI:
3288 	case PRIV_NET_DELMULTI:
3289 	case PRIV_NET_HWIOCTL:
3290 	case PRIV_NET_SETLLADDR:
3291 	case PRIV_NET_ADDIFGROUP:
3292 	case PRIV_NET_DELIFGROUP:
3293 	case PRIV_NET_IFCREATE:
3294 	case PRIV_NET_IFDESTROY:
3295 	case PRIV_NET_ADDIFADDR:
3296 	case PRIV_NET_DELIFADDR:
3297 	case PRIV_NET_LAGG:
3298 	case PRIV_NET_GIF:
3299 	case PRIV_NET_SETIFVNET:
3300 	case PRIV_NET_SETIFFIB:
3301 
3302 		/*
3303 		 * 802.11-related privileges.
3304 		 */
3305 	case PRIV_NET80211_VAP_GETKEY:
3306 	case PRIV_NET80211_VAP_MANAGE:
3307 
3308 #ifdef notyet
3309 		/*
3310 		 * ATM privileges.
3311 		 */
3312 	case PRIV_NETATM_CFG:
3313 	case PRIV_NETATM_ADD:
3314 	case PRIV_NETATM_DEL:
3315 	case PRIV_NETATM_SET:
3316 
3317 		/*
3318 		 * Bluetooth privileges.
3319 		 */
3320 	case PRIV_NETBLUETOOTH_RAW:
3321 #endif
3322 
3323 		/*
3324 		 * Netgraph and netgraph module privileges.
3325 		 */
3326 	case PRIV_NETGRAPH_CONTROL:
3327 #ifdef notyet
3328 	case PRIV_NETGRAPH_TTY:
3329 #endif
3330 
3331 		/*
3332 		 * IPv4 and IPv6 privileges.
3333 		 */
3334 	case PRIV_NETINET_IPFW:
3335 	case PRIV_NETINET_DIVERT:
3336 	case PRIV_NETINET_PF:
3337 	case PRIV_NETINET_DUMMYNET:
3338 	case PRIV_NETINET_CARP:
3339 	case PRIV_NETINET_MROUTE:
3340 	case PRIV_NETINET_RAW:
3341 	case PRIV_NETINET_ADDRCTRL6:
3342 	case PRIV_NETINET_ND6:
3343 	case PRIV_NETINET_SCOPE6:
3344 	case PRIV_NETINET_ALIFETIME6:
3345 	case PRIV_NETINET_IPSEC:
3346 	case PRIV_NETINET_BINDANY:
3347 
3348 #ifdef notyet
3349 		/*
3350 		 * NCP privileges.
3351 		 */
3352 	case PRIV_NETNCP:
3353 
3354 		/*
3355 		 * SMB privileges.
3356 		 */
3357 	case PRIV_NETSMB:
3358 #endif
3359 
3360 	/*
3361 	 * No default: or deny here.
3362 	 * In case of no permit fall through to next switch().
3363 	 */
3364 		if (cred->cr_prison->pr_flags & PR_VNET)
3365 			return (0);
3366 	}
3367 #endif /* VIMAGE */
3368 
3369 	switch (priv) {
3370 		/*
3371 		 * Allow ktrace privileges for root in jail.
3372 		 */
3373 	case PRIV_KTRACE:
3374 
3375 #if 0
3376 		/*
3377 		 * Allow jailed processes to configure audit identity and
3378 		 * submit audit records (login, etc).  In the future we may
3379 		 * want to further refine the relationship between audit and
3380 		 * jail.
3381 		 */
3382 	case PRIV_AUDIT_GETAUDIT:
3383 	case PRIV_AUDIT_SETAUDIT:
3384 	case PRIV_AUDIT_SUBMIT:
3385 #endif
3386 
3387 		/*
3388 		 * Allow jailed processes to manipulate process UNIX
3389 		 * credentials in any way they see fit.
3390 		 */
3391 	case PRIV_CRED_SETUID:
3392 	case PRIV_CRED_SETEUID:
3393 	case PRIV_CRED_SETGID:
3394 	case PRIV_CRED_SETEGID:
3395 	case PRIV_CRED_SETGROUPS:
3396 	case PRIV_CRED_SETREUID:
3397 	case PRIV_CRED_SETREGID:
3398 	case PRIV_CRED_SETRESUID:
3399 	case PRIV_CRED_SETRESGID:
3400 
3401 		/*
3402 		 * Jail implements visibility constraints already, so allow
3403 		 * jailed root to override uid/gid-based constraints.
3404 		 */
3405 	case PRIV_SEEOTHERGIDS:
3406 	case PRIV_SEEOTHERUIDS:
3407 
3408 		/*
3409 		 * Jail implements inter-process debugging limits already, so
3410 		 * allow jailed root various debugging privileges.
3411 		 */
3412 	case PRIV_DEBUG_DIFFCRED:
3413 	case PRIV_DEBUG_SUGID:
3414 	case PRIV_DEBUG_UNPRIV:
3415 
3416 		/*
3417 		 * Allow jail to set various resource limits and login
3418 		 * properties, and for now, exceed process resource limits.
3419 		 */
3420 	case PRIV_PROC_LIMIT:
3421 	case PRIV_PROC_SETLOGIN:
3422 	case PRIV_PROC_SETRLIMIT:
3423 
3424 		/*
3425 		 * System V and POSIX IPC privileges are granted in jail.
3426 		 */
3427 	case PRIV_IPC_READ:
3428 	case PRIV_IPC_WRITE:
3429 	case PRIV_IPC_ADMIN:
3430 	case PRIV_IPC_MSGSIZE:
3431 	case PRIV_MQ_ADMIN:
3432 
3433 		/*
3434 		 * Jail operations within a jail work on child jails.
3435 		 */
3436 	case PRIV_JAIL_ATTACH:
3437 	case PRIV_JAIL_SET:
3438 	case PRIV_JAIL_REMOVE:
3439 
3440 		/*
3441 		 * Jail implements its own inter-process limits, so allow
3442 		 * root processes in jail to change scheduling on other
3443 		 * processes in the same jail.  Likewise for signalling.
3444 		 */
3445 	case PRIV_SCHED_DIFFCRED:
3446 	case PRIV_SCHED_CPUSET:
3447 	case PRIV_SIGNAL_DIFFCRED:
3448 	case PRIV_SIGNAL_SUGID:
3449 
3450 		/*
3451 		 * Allow jailed processes to write to sysctls marked as jail
3452 		 * writable.
3453 		 */
3454 	case PRIV_SYSCTL_WRITEJAIL:
3455 
3456 		/*
3457 		 * Allow root in jail to manage a variety of quota
3458 		 * properties.  These should likely be conditional on a
3459 		 * configuration option.
3460 		 */
3461 	case PRIV_VFS_GETQUOTA:
3462 	case PRIV_VFS_SETQUOTA:
3463 
3464 		/*
3465 		 * Since Jail relies on chroot() to implement file system
3466 		 * protections, grant many VFS privileges to root in jail.
3467 		 * Be careful to exclude mount-related and NFS-related
3468 		 * privileges.
3469 		 */
3470 	case PRIV_VFS_READ:
3471 	case PRIV_VFS_WRITE:
3472 	case PRIV_VFS_ADMIN:
3473 	case PRIV_VFS_EXEC:
3474 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3475 	case PRIV_VFS_CHFLAGS_DEV:
3476 	case PRIV_VFS_CHOWN:
3477 	case PRIV_VFS_CHROOT:
3478 	case PRIV_VFS_RETAINSUGID:
3479 	case PRIV_VFS_FCHROOT:
3480 	case PRIV_VFS_LINK:
3481 	case PRIV_VFS_SETGID:
3482 	case PRIV_VFS_STAT:
3483 	case PRIV_VFS_STICKYFILE:
3484 
3485 		/*
3486 		 * As in the non-jail case, non-root users are expected to be
3487 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3488 		 * exists in the jail and they have permission to access it).
3489 		 */
3490 	case PRIV_KMEM_READ:
3491 		return (0);
3492 
3493 		/*
3494 		 * Depending on the global setting, allow privilege of
3495 		 * setting system flags.
3496 		 */
3497 	case PRIV_VFS_SYSFLAGS:
3498 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3499 			return (0);
3500 		else
3501 			return (EPERM);
3502 
3503 		/*
3504 		 * Depending on the global setting, allow privilege of
3505 		 * mounting/unmounting file systems.
3506 		 */
3507 	case PRIV_VFS_MOUNT:
3508 	case PRIV_VFS_UNMOUNT:
3509 	case PRIV_VFS_MOUNT_NONUSER:
3510 	case PRIV_VFS_MOUNT_OWNER:
3511 		pr = cred->cr_prison;
3512 		prison_lock(pr);
3513 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3514 			error = 0;
3515 		else
3516 			error = EPERM;
3517 		prison_unlock(pr);
3518 		return (error);
3519 
3520 		/*
3521 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3522 		 * policy.  priv_check_cred will not specifically allow it, and
3523 		 * we may want a MAC policy to allow it.
3524 		 */
3525 	case PRIV_VFS_READ_DIR:
3526 		return (0);
3527 
3528 		/*
3529 		 * Conditionnaly allow locking (unlocking) physical pages
3530 		 * in memory.
3531 		 */
3532 	case PRIV_VM_MLOCK:
3533 	case PRIV_VM_MUNLOCK:
3534 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3535 			return (0);
3536 		else
3537 			return (EPERM);
3538 
3539 		/*
3540 		 * Conditionally allow jailed root to bind reserved ports.
3541 		 */
3542 	case PRIV_NETINET_RESERVEDPORT:
3543 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3544 			return (0);
3545 		else
3546 			return (EPERM);
3547 
3548 		/*
3549 		 * Allow jailed root to reuse in-use ports.
3550 		 */
3551 	case PRIV_NETINET_REUSEPORT:
3552 		return (0);
3553 
3554 		/*
3555 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3556 		 */
3557 	case PRIV_NETINET_SETHDROPTS:
3558 		return (0);
3559 
3560 		/*
3561 		 * Conditionally allow creating raw sockets in jail.
3562 		 */
3563 	case PRIV_NETINET_RAW:
3564 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3565 			return (0);
3566 		else
3567 			return (EPERM);
3568 
3569 		/*
3570 		 * Since jail implements its own visibility limits on netstat
3571 		 * sysctls, allow getcred.  This allows identd to work in
3572 		 * jail.
3573 		 */
3574 	case PRIV_NETINET_GETCRED:
3575 		return (0);
3576 
3577 		/*
3578 		 * Allow jailed root to set loginclass.
3579 		 */
3580 	case PRIV_PROC_SETLOGINCLASS:
3581 		return (0);
3582 
3583 		/*
3584 		 * Do not allow a process inside a jail to read the kernel
3585 		 * message buffer unless explicitly permitted.
3586 		 */
3587 	case PRIV_MSGBUF:
3588 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3589 			return (0);
3590 		return (EPERM);
3591 
3592 	default:
3593 		/*
3594 		 * In all remaining cases, deny the privilege request.  This
3595 		 * includes almost all network privileges, many system
3596 		 * configuration privileges.
3597 		 */
3598 		return (EPERM);
3599 	}
3600 }
3601 
3602 /*
3603  * Return the part of pr2's name that is relative to pr1, or the whole name
3604  * if it does not directly follow.
3605  */
3606 
3607 char *
3608 prison_name(struct prison *pr1, struct prison *pr2)
3609 {
3610 	char *name;
3611 
3612 	/* Jails see themselves as "0" (if they see themselves at all). */
3613 	if (pr1 == pr2)
3614 		return "0";
3615 	name = pr2->pr_name;
3616 	if (prison_ischild(pr1, pr2)) {
3617 		/*
3618 		 * pr1 isn't locked (and allprison_lock may not be either)
3619 		 * so its length can't be counted on.  But the number of dots
3620 		 * can be counted on - and counted.
3621 		 */
3622 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3623 			name = strchr(name, '.') + 1;
3624 	}
3625 	return (name);
3626 }
3627 
3628 /*
3629  * Return the part of pr2's path that is relative to pr1, or the whole path
3630  * if it does not directly follow.
3631  */
3632 static char *
3633 prison_path(struct prison *pr1, struct prison *pr2)
3634 {
3635 	char *path1, *path2;
3636 	int len1;
3637 
3638 	path1 = pr1->pr_path;
3639 	path2 = pr2->pr_path;
3640 	if (!strcmp(path1, "/"))
3641 		return (path2);
3642 	len1 = strlen(path1);
3643 	if (strncmp(path1, path2, len1))
3644 		return (path2);
3645 	if (path2[len1] == '\0')
3646 		return "/";
3647 	if (path2[len1] == '/')
3648 		return (path2 + len1);
3649 	return (path2);
3650 }
3651 
3652 /*
3653  * Jail-related sysctls.
3654  */
3655 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3656     "Jails");
3657 
3658 static int
3659 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3660 {
3661 	struct xprison *xp;
3662 	struct prison *pr, *cpr;
3663 #ifdef INET
3664 	struct in_addr *ip4 = NULL;
3665 	int ip4s = 0;
3666 #endif
3667 #ifdef INET6
3668 	struct in6_addr *ip6 = NULL;
3669 	int ip6s = 0;
3670 #endif
3671 	int descend, error;
3672 
3673 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3674 	pr = req->td->td_ucred->cr_prison;
3675 	error = 0;
3676 	sx_slock(&allprison_lock);
3677 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3678 #if defined(INET) || defined(INET6)
3679  again:
3680 #endif
3681 		mtx_lock(&cpr->pr_mtx);
3682 #ifdef INET
3683 		if (cpr->pr_ip4s > 0) {
3684 			if (ip4s < cpr->pr_ip4s) {
3685 				ip4s = cpr->pr_ip4s;
3686 				mtx_unlock(&cpr->pr_mtx);
3687 				ip4 = realloc(ip4, ip4s *
3688 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3689 				goto again;
3690 			}
3691 			bcopy(cpr->pr_ip4, ip4,
3692 			    cpr->pr_ip4s * sizeof(struct in_addr));
3693 		}
3694 #endif
3695 #ifdef INET6
3696 		if (cpr->pr_ip6s > 0) {
3697 			if (ip6s < cpr->pr_ip6s) {
3698 				ip6s = cpr->pr_ip6s;
3699 				mtx_unlock(&cpr->pr_mtx);
3700 				ip6 = realloc(ip6, ip6s *
3701 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3702 				goto again;
3703 			}
3704 			bcopy(cpr->pr_ip6, ip6,
3705 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3706 		}
3707 #endif
3708 		if (!prison_isvalid(cpr)) {
3709 			mtx_unlock(&cpr->pr_mtx);
3710 			continue;
3711 		}
3712 		bzero(xp, sizeof(*xp));
3713 		xp->pr_version = XPRISON_VERSION;
3714 		xp->pr_id = cpr->pr_id;
3715 		xp->pr_state = prison_isalive(cpr)
3716 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3717 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3718 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3719 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3720 #ifdef INET
3721 		xp->pr_ip4s = cpr->pr_ip4s;
3722 #endif
3723 #ifdef INET6
3724 		xp->pr_ip6s = cpr->pr_ip6s;
3725 #endif
3726 		mtx_unlock(&cpr->pr_mtx);
3727 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3728 		if (error)
3729 			break;
3730 #ifdef INET
3731 		if (xp->pr_ip4s > 0) {
3732 			error = SYSCTL_OUT(req, ip4,
3733 			    xp->pr_ip4s * sizeof(struct in_addr));
3734 			if (error)
3735 				break;
3736 		}
3737 #endif
3738 #ifdef INET6
3739 		if (xp->pr_ip6s > 0) {
3740 			error = SYSCTL_OUT(req, ip6,
3741 			    xp->pr_ip6s * sizeof(struct in6_addr));
3742 			if (error)
3743 				break;
3744 		}
3745 #endif
3746 	}
3747 	sx_sunlock(&allprison_lock);
3748 	free(xp, M_TEMP);
3749 #ifdef INET
3750 	free(ip4, M_TEMP);
3751 #endif
3752 #ifdef INET6
3753 	free(ip6, M_TEMP);
3754 #endif
3755 	return (error);
3756 }
3757 
3758 SYSCTL_OID(_security_jail, OID_AUTO, list,
3759     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3760     sysctl_jail_list, "S", "List of active jails");
3761 
3762 static int
3763 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3764 {
3765 	int error, injail;
3766 
3767 	injail = jailed(req->td->td_ucred);
3768 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3769 
3770 	return (error);
3771 }
3772 
3773 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3774     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3775     sysctl_jail_jailed, "I", "Process in jail?");
3776 
3777 static int
3778 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3779 {
3780 	int error, havevnet;
3781 #ifdef VIMAGE
3782 	struct ucred *cred = req->td->td_ucred;
3783 
3784 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3785 #else
3786 	havevnet = 0;
3787 #endif
3788 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3789 
3790 	return (error);
3791 }
3792 
3793 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3794     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3795     sysctl_jail_vnet, "I", "Jail owns vnet?");
3796 
3797 #if defined(INET) || defined(INET6)
3798 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3799     &jail_max_af_ips, 0,
3800     "Number of IP addresses a jail may have at most per address family (deprecated)");
3801 #endif
3802 
3803 /*
3804  * Default parameters for jail(2) compatibility.  For historical reasons,
3805  * the sysctl names have varying similarity to the parameter names.  Prisons
3806  * just see their own parameters, and can't change them.
3807  */
3808 static int
3809 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3810 {
3811 	int error, i;
3812 
3813 	/* Get the current flag value, and convert it to a boolean. */
3814 	if (req->td->td_ucred->cr_prison == &prison0) {
3815 		mtx_lock(&prison0.pr_mtx);
3816 		i = (jail_default_allow & arg2) != 0;
3817 		mtx_unlock(&prison0.pr_mtx);
3818 	} else
3819 		i = prison_allow(req->td->td_ucred, arg2);
3820 
3821 	if (arg1 != NULL)
3822 		i = !i;
3823 	error = sysctl_handle_int(oidp, &i, 0, req);
3824 	if (error || !req->newptr)
3825 		return (error);
3826 	i = i ? arg2 : 0;
3827 	if (arg1 != NULL)
3828 		i ^= arg2;
3829 	/*
3830 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3831 	 * for writing.
3832 	 */
3833 	mtx_lock(&prison0.pr_mtx);
3834 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3835 	mtx_unlock(&prison0.pr_mtx);
3836 	return (0);
3837 }
3838 
3839 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3840     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3841     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3842     "Processes in jail can set their hostnames (deprecated)");
3843 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3844     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3845     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3846     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3847 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3848     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3849     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3850     "Processes in jail can use System V IPC primitives (deprecated)");
3851 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3852     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3853     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3854     "Prison root can create raw sockets (deprecated)");
3855 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3856     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3857     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3858     "Processes in jail can alter system file flags (deprecated)");
3859 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3860     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3861     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3862     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3863 
3864 static int
3865 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3866 {
3867 	struct prison *pr;
3868 	int level, error;
3869 
3870 	pr = req->td->td_ucred->cr_prison;
3871 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3872 	error = sysctl_handle_int(oidp, &level, 0, req);
3873 	if (error || !req->newptr)
3874 		return (error);
3875 	*(int *)arg1 = level;
3876 	return (0);
3877 }
3878 
3879 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3880     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3881     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3882     sysctl_jail_default_level, "I",
3883     "Processes in jail cannot see all mounted file systems (deprecated)");
3884 
3885 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3886     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3887     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3888     sysctl_jail_default_level, "I",
3889     "Ruleset for the devfs filesystem in jail (deprecated)");
3890 
3891 /*
3892  * Nodes to describe jail parameters.  Maximum length of string parameters
3893  * is returned in the string itself, and the other parameters exist merely
3894  * to make themselves and their types known.
3895  */
3896 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3897     "Jail parameters");
3898 
3899 int
3900 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3901 {
3902 	int i;
3903 	long l;
3904 	size_t s;
3905 	char numbuf[12];
3906 
3907 	switch (oidp->oid_kind & CTLTYPE)
3908 	{
3909 	case CTLTYPE_LONG:
3910 	case CTLTYPE_ULONG:
3911 		l = 0;
3912 #ifdef SCTL_MASK32
3913 		if (!(req->flags & SCTL_MASK32))
3914 #endif
3915 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3916 	case CTLTYPE_INT:
3917 	case CTLTYPE_UINT:
3918 		i = 0;
3919 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3920 	case CTLTYPE_STRING:
3921 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3922 		return
3923 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3924 	case CTLTYPE_STRUCT:
3925 		s = (size_t)arg2;
3926 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3927 	}
3928 	return (0);
3929 }
3930 
3931 /*
3932  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3933  * jail creation time but cannot be changed in an existing jail.
3934  */
3935 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3936 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3937 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3938 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3939 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3940     "I", "Jail secure level");
3941 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3942     "Jail value for kern.osreldate and uname -K");
3943 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3944     "Jail value for kern.osrelease and uname -r");
3945 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3946     "I", "Jail cannot see all mounted file systems");
3947 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3948     "I", "Ruleset for in-jail devfs mounts");
3949 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3950     "B", "Jail persistence");
3951 #ifdef VIMAGE
3952 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3953     "E,jailsys", "Virtual network stack");
3954 #endif
3955 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3956     "B", "Jail is in the process of shutting down");
3957 
3958 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3959 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3960     "I", "Current number of child jails");
3961 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3962     "I", "Maximum number of child jails");
3963 
3964 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3965 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3966     "Jail hostname");
3967 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3968     "Jail NIS domainname");
3969 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3970     "Jail host UUID");
3971 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3972     "LU", "Jail host ID");
3973 
3974 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3975 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3976 
3977 #ifdef INET
3978 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3979     "Jail IPv4 address virtualization");
3980 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3981     "S,in_addr,a", "Jail IPv4 addresses");
3982 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3983     "B", "Do (not) use IPv4 source address selection rather than the "
3984     "primary jail IPv4 address.");
3985 #endif
3986 #ifdef INET6
3987 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3988     "Jail IPv6 address virtualization");
3989 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3990     "S,in6_addr,a", "Jail IPv6 addresses");
3991 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3992     "B", "Do (not) use IPv6 source address selection rather than the "
3993     "primary jail IPv6 address.");
3994 #endif
3995 
3996 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3997 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3998     "B", "Jail may set hostname");
3999 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4000     "B", "Jail may use SYSV IPC");
4001 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4002     "B", "Jail may create raw sockets");
4003 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4004     "B", "Jail may alter system file flags");
4005 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4006     "B", "Jail may set file quotas");
4007 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4008     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4009 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4010     "B", "Jail may lock (unlock) physical pages in memory");
4011 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4012     "B", "Jail may bind sockets to reserved ports");
4013 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4014     "B", "Jail may read the kernel message buffer");
4015 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4016     "B", "Unprivileged processes may use process debugging facilities");
4017 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4018     "B", "Processes in jail with uid 0 have privilege");
4019 
4020 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4021 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4022     "B", "Jail may mount/unmount jail-friendly file systems in general");
4023 
4024 /*
4025  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4026  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4027  * not created.
4028  */
4029 unsigned
4030 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4031     const char *descr)
4032 {
4033 	struct bool_flags *bf;
4034 	struct sysctl_oid *parent;
4035 	char *allow_name, *allow_noname, *allowed;
4036 #ifndef NO_SYSCTL_DESCR
4037 	char *descr_deprecated;
4038 #endif
4039 	u_int allow_flag;
4040 
4041 	if (prefix
4042 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4043 		< 0 ||
4044 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4045 		< 0
4046 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4047 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4048 		free(allow_name, M_PRISON);
4049 		return 0;
4050 	}
4051 
4052 	/*
4053 	 * See if this parameter has already beed added, i.e. a module was
4054 	 * previously loaded/unloaded.
4055 	 */
4056 	mtx_lock(&prison0.pr_mtx);
4057 	for (bf = pr_flag_allow;
4058 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4059 		atomic_load_int(&bf->flag) != 0;
4060 	     bf++) {
4061 		if (strcmp(bf->name, allow_name) == 0) {
4062 			allow_flag = bf->flag;
4063 			goto no_add;
4064 		}
4065 	}
4066 
4067 	/*
4068 	 * Find a free bit in pr_allow_all, failing if there are none
4069 	 * (which shouldn't happen as long as we keep track of how many
4070 	 * potential dynamic flags exist).
4071 	 */
4072 	for (allow_flag = 1;; allow_flag <<= 1) {
4073 		if (allow_flag == 0)
4074 			goto no_add;
4075 		if ((pr_allow_all & allow_flag) == 0)
4076 			break;
4077 	}
4078 
4079 	/* Note the parameter in the next open slot in pr_flag_allow. */
4080 	for (bf = pr_flag_allow; ; bf++) {
4081 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4082 			/* This should never happen, but is not fatal. */
4083 			allow_flag = 0;
4084 			goto no_add;
4085 		}
4086 		if (atomic_load_int(&bf->flag) == 0)
4087 			break;
4088 	}
4089 	bf->name = allow_name;
4090 	bf->noname = allow_noname;
4091 	pr_allow_all |= allow_flag;
4092 	/*
4093 	 * prison0 always has permission for the new parameter.
4094 	 * Other jails must have it granted to them.
4095 	 */
4096 	prison0.pr_allow |= allow_flag;
4097 	/* The flag indicates a valid entry, so make sure it is set last. */
4098 	atomic_store_rel_int(&bf->flag, allow_flag);
4099 	mtx_unlock(&prison0.pr_mtx);
4100 
4101 	/*
4102 	 * Create sysctls for the paramter, and the back-compat global
4103 	 * permission.
4104 	 */
4105 	parent = prefix
4106 	    ? SYSCTL_ADD_NODE(NULL,
4107 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4108 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4109 	    : &sysctl___security_jail_param_allow;
4110 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4111 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4112 	    NULL, 0, sysctl_jail_param, "B", descr);
4113 	if ((prefix
4114 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4115 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4116 #ifndef NO_SYSCTL_DESCR
4117 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4118 		    descr);
4119 #endif
4120 		(void)SYSCTL_ADD_PROC(NULL,
4121 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4122 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4123 		    sysctl_jail_default_allow, "I", descr_deprecated);
4124 #ifndef NO_SYSCTL_DESCR
4125 		free(descr_deprecated, M_TEMP);
4126 #endif
4127 		free(allowed, M_TEMP);
4128 	}
4129 	return allow_flag;
4130 
4131  no_add:
4132 	mtx_unlock(&prison0.pr_mtx);
4133 	free(allow_name, M_PRISON);
4134 	free(allow_noname, M_PRISON);
4135 	return allow_flag;
4136 }
4137 
4138 /*
4139  * The VFS system will register jail-aware filesystems here.  They each get
4140  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4141  * attempts to mount.
4142  */
4143 void
4144 prison_add_vfs(struct vfsconf *vfsp)
4145 {
4146 #ifdef NO_SYSCTL_DESCR
4147 
4148 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4149 	    NULL, NULL);
4150 #else
4151 	char *descr;
4152 
4153 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4154 	    vfsp->vfc_name);
4155 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4156 	    NULL, descr);
4157 	free(descr, M_TEMP);
4158 #endif
4159 }
4160 
4161 #ifdef RACCT
4162 void
4163 prison_racct_foreach(void (*callback)(struct racct *racct,
4164     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4165     void *arg2, void *arg3)
4166 {
4167 	struct prison_racct *prr;
4168 
4169 	ASSERT_RACCT_ENABLED();
4170 
4171 	sx_slock(&allprison_lock);
4172 	if (pre != NULL)
4173 		(pre)();
4174 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4175 		(callback)(prr->prr_racct, arg2, arg3);
4176 	if (post != NULL)
4177 		(post)();
4178 	sx_sunlock(&allprison_lock);
4179 }
4180 
4181 static struct prison_racct *
4182 prison_racct_find_locked(const char *name)
4183 {
4184 	struct prison_racct *prr;
4185 
4186 	ASSERT_RACCT_ENABLED();
4187 	sx_assert(&allprison_lock, SA_XLOCKED);
4188 
4189 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4190 		return (NULL);
4191 
4192 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4193 		if (strcmp(name, prr->prr_name) != 0)
4194 			continue;
4195 
4196 		/* Found prison_racct with a matching name? */
4197 		prison_racct_hold(prr);
4198 		return (prr);
4199 	}
4200 
4201 	/* Add new prison_racct. */
4202 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4203 	racct_create(&prr->prr_racct);
4204 
4205 	strcpy(prr->prr_name, name);
4206 	refcount_init(&prr->prr_refcount, 1);
4207 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4208 
4209 	return (prr);
4210 }
4211 
4212 struct prison_racct *
4213 prison_racct_find(const char *name)
4214 {
4215 	struct prison_racct *prr;
4216 
4217 	ASSERT_RACCT_ENABLED();
4218 
4219 	sx_xlock(&allprison_lock);
4220 	prr = prison_racct_find_locked(name);
4221 	sx_xunlock(&allprison_lock);
4222 	return (prr);
4223 }
4224 
4225 void
4226 prison_racct_hold(struct prison_racct *prr)
4227 {
4228 
4229 	ASSERT_RACCT_ENABLED();
4230 
4231 	refcount_acquire(&prr->prr_refcount);
4232 }
4233 
4234 static void
4235 prison_racct_free_locked(struct prison_racct *prr)
4236 {
4237 
4238 	ASSERT_RACCT_ENABLED();
4239 	sx_assert(&allprison_lock, SA_XLOCKED);
4240 
4241 	if (refcount_release(&prr->prr_refcount)) {
4242 		racct_destroy(&prr->prr_racct);
4243 		LIST_REMOVE(prr, prr_next);
4244 		free(prr, M_PRISON_RACCT);
4245 	}
4246 }
4247 
4248 void
4249 prison_racct_free(struct prison_racct *prr)
4250 {
4251 
4252 	ASSERT_RACCT_ENABLED();
4253 	sx_assert(&allprison_lock, SA_UNLOCKED);
4254 
4255 	if (refcount_release_if_not_last(&prr->prr_refcount))
4256 		return;
4257 
4258 	sx_xlock(&allprison_lock);
4259 	prison_racct_free_locked(prr);
4260 	sx_xunlock(&allprison_lock);
4261 }
4262 
4263 static void
4264 prison_racct_attach(struct prison *pr)
4265 {
4266 	struct prison_racct *prr;
4267 
4268 	ASSERT_RACCT_ENABLED();
4269 	sx_assert(&allprison_lock, SA_XLOCKED);
4270 
4271 	prr = prison_racct_find_locked(pr->pr_name);
4272 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4273 
4274 	pr->pr_prison_racct = prr;
4275 }
4276 
4277 /*
4278  * Handle jail renaming.  From the racct point of view, renaming means
4279  * moving from one prison_racct to another.
4280  */
4281 static void
4282 prison_racct_modify(struct prison *pr)
4283 {
4284 #ifdef RCTL
4285 	struct proc *p;
4286 	struct ucred *cred;
4287 #endif
4288 	struct prison_racct *oldprr;
4289 
4290 	ASSERT_RACCT_ENABLED();
4291 
4292 	sx_slock(&allproc_lock);
4293 	sx_xlock(&allprison_lock);
4294 
4295 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4296 		sx_xunlock(&allprison_lock);
4297 		sx_sunlock(&allproc_lock);
4298 		return;
4299 	}
4300 
4301 	oldprr = pr->pr_prison_racct;
4302 	pr->pr_prison_racct = NULL;
4303 
4304 	prison_racct_attach(pr);
4305 
4306 	/*
4307 	 * Move resource utilisation records.
4308 	 */
4309 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4310 
4311 #ifdef RCTL
4312 	/*
4313 	 * Force rctl to reattach rules to processes.
4314 	 */
4315 	FOREACH_PROC_IN_SYSTEM(p) {
4316 		PROC_LOCK(p);
4317 		cred = crhold(p->p_ucred);
4318 		PROC_UNLOCK(p);
4319 		rctl_proc_ucred_changed(p, cred);
4320 		crfree(cred);
4321 	}
4322 #endif
4323 
4324 	sx_sunlock(&allproc_lock);
4325 	prison_racct_free_locked(oldprr);
4326 	sx_xunlock(&allprison_lock);
4327 }
4328 
4329 static void
4330 prison_racct_detach(struct prison *pr)
4331 {
4332 
4333 	ASSERT_RACCT_ENABLED();
4334 	sx_assert(&allprison_lock, SA_UNLOCKED);
4335 
4336 	if (pr->pr_prison_racct == NULL)
4337 		return;
4338 	prison_racct_free(pr->pr_prison_racct);
4339 	pr->pr_prison_racct = NULL;
4340 }
4341 #endif /* RACCT */
4342 
4343 #ifdef DDB
4344 
4345 static void
4346 db_show_prison(struct prison *pr)
4347 {
4348 	struct bool_flags *bf;
4349 	struct jailsys_flags *jsf;
4350 #if defined(INET) || defined(INET6)
4351 	int ii;
4352 #endif
4353 	unsigned f;
4354 #ifdef INET
4355 	char ip4buf[INET_ADDRSTRLEN];
4356 #endif
4357 #ifdef INET6
4358 	char ip6buf[INET6_ADDRSTRLEN];
4359 #endif
4360 
4361 	db_printf("prison %p:\n", pr);
4362 	db_printf(" jid             = %d\n", pr->pr_id);
4363 	db_printf(" name            = %s\n", pr->pr_name);
4364 	db_printf(" parent          = %p\n", pr->pr_parent);
4365 	db_printf(" ref             = %d\n", pr->pr_ref);
4366 	db_printf(" uref            = %d\n", pr->pr_uref);
4367 	db_printf(" path            = %s\n", pr->pr_path);
4368 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4369 	    ? pr->pr_cpuset->cs_id : -1);
4370 #ifdef VIMAGE
4371 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4372 #endif
4373 	db_printf(" root            = %p\n", pr->pr_root);
4374 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4375 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4376 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4377 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4378 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4379 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4380 	db_printf(" flags           = 0x%x", pr->pr_flags);
4381 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4382 		if (pr->pr_flags & bf->flag)
4383 			db_printf(" %s", bf->name);
4384 	for (jsf = pr_flag_jailsys;
4385 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4386 	     jsf++) {
4387 		f = pr->pr_flags & (jsf->disable | jsf->new);
4388 		db_printf(" %-16s= %s\n", jsf->name,
4389 		    (f != 0 && f == jsf->disable) ? "disable"
4390 		    : (f == jsf->new) ? "new"
4391 		    : "inherit");
4392 	}
4393 	db_printf(" allow           = 0x%x", pr->pr_allow);
4394 	for (bf = pr_flag_allow;
4395 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4396 		atomic_load_int(&bf->flag) != 0;
4397 	     bf++)
4398 		if (pr->pr_allow & bf->flag)
4399 			db_printf(" %s", bf->name);
4400 	db_printf("\n");
4401 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4402 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4403 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4404 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4405 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4406 #ifdef INET
4407 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4408 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4409 		db_printf(" %s %s\n",
4410 		    ii == 0 ? "ip4.addr        =" : "                 ",
4411 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4412 #endif
4413 #ifdef INET6
4414 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4415 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4416 		db_printf(" %s %s\n",
4417 		    ii == 0 ? "ip6.addr        =" : "                 ",
4418 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4419 #endif
4420 }
4421 
4422 DB_SHOW_COMMAND(prison, db_show_prison_command)
4423 {
4424 	struct prison *pr;
4425 
4426 	if (!have_addr) {
4427 		/*
4428 		 * Show all prisons in the list, and prison0 which is not
4429 		 * listed.
4430 		 */
4431 		db_show_prison(&prison0);
4432 		if (!db_pager_quit) {
4433 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4434 				db_show_prison(pr);
4435 				if (db_pager_quit)
4436 					break;
4437 			}
4438 		}
4439 		return;
4440 	}
4441 
4442 	if (addr == 0)
4443 		pr = &prison0;
4444 	else {
4445 		/* Look for a prison with the ID and with references. */
4446 		TAILQ_FOREACH(pr, &allprison, pr_list)
4447 			if (pr->pr_id == addr && pr->pr_ref > 0)
4448 				break;
4449 		if (pr == NULL)
4450 			/* Look again, without requiring a reference. */
4451 			TAILQ_FOREACH(pr, &allprison, pr_list)
4452 				if (pr->pr_id == addr)
4453 					break;
4454 		if (pr == NULL)
4455 			/* Assume address points to a valid prison. */
4456 			pr = (struct prison *)addr;
4457 	}
4458 	db_show_prison(pr);
4459 }
4460 
4461 #endif /* DDB */
4462