xref: /freebsd/sys/kern/kern_jail.c (revision 5f7d5709e0c242d5a5fa5516d6079433ee06c347)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_ddb.h"
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_nfs.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/file.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/epoch.h>
49 #include <sys/event.h>
50 #include <sys/taskqueue.h>
51 #include <sys/fcntl.h>
52 #include <sys/jail.h>
53 #include <sys/jaildesc.h>
54 #include <sys/linker.h>
55 #include <sys/lock.h>
56 #include <sys/mman.h>
57 #include <sys/mutex.h>
58 #include <sys/racct.h>
59 #include <sys/rctl.h>
60 #include <sys/refcount.h>
61 #include <sys/sx.h>
62 #include <sys/sysent.h>
63 #include <sys/namei.h>
64 #include <sys/mount.h>
65 #include <sys/queue.h>
66 #include <sys/socket.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/uuid.h>
70 #include <sys/vnode.h>
71 
72 #include <net/if.h>
73 #include <net/vnet.h>
74 
75 #include <netinet/in.h>
76 
77 #ifdef DDB
78 #include <ddb/ddb.h>
79 #endif /* DDB */
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
84 
85 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
86 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
87 
88 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
89 #ifdef INET
90 #ifdef INET6
91 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
92 #else
93 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
94 #endif
95 #else /* !INET */
96 #ifdef INET6
97 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
98 #else
99 #define	_PR_IP_SADDRSEL	0
100 #endif
101 #endif
102 
103 /* prison0 describes what is "real" about the system. */
104 struct prison prison0 = {
105 	.pr_id		= 0,
106 	.pr_name	= "0",
107 	.pr_ref		= 1,
108 	.pr_uref	= 1,
109 	.pr_path	= "/",
110 	.pr_securelevel	= -1,
111 	.pr_devfs_rsnum = 0,
112 	.pr_state	= PRISON_STATE_ALIVE,
113 	.pr_childmax	= JAIL_MAX,
114 	.pr_hostuuid	= DEFAULT_HOSTUUID,
115 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
116 #ifdef VIMAGE
117 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
118 #else
119 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
120 #endif
121 	.pr_allow	= PR_ALLOW_PRISON0,
122 };
123 _Static_assert((PR_ALLOW_PRISON0 & ~PR_ALLOW_ALL_STATIC) == 0,
124     "Bits enabled in PR_ALLOW_PRISON0 that are not statically reserved");
125 
126 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
127 
128 struct bool_flags {
129 	const char	*name;
130 	const char	*noname;
131 	volatile u_int	 flag;
132 };
133 struct jailsys_flags {
134 	const char	*name;
135 	unsigned	 disable;
136 	unsigned	 new;
137 };
138 
139 /*
140  * Handle jail teardown in a dedicated thread to avoid deadlocks from
141  * vnet_destroy().
142  */
143 TASKQUEUE_DEFINE_THREAD(jail_remove);
144 
145 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
146 struct	sx allprison_lock;
147 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
148 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
149 LIST_HEAD(, prison_racct) allprison_racct;
150 int	lastprid = 0;
151 int	lastdeadid = 0;
152 
153 static int get_next_prid(struct prison **insprp);
154 static int get_next_deadid(struct prison **insprp);
155 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
156 static void prison_complete(void *context, int pending);
157 static void prison_deref(struct prison *pr, int flags);
158 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
159 static int prison_lock_xlock(struct prison *pr, int flags);
160 static void prison_cleanup_locked(struct prison *pr);
161 static void prison_cleanup_unlocked(struct prison *pr);
162 static void prison_free_not_last(struct prison *pr);
163 static void prison_proc_free_not_last(struct prison *pr);
164 static void prison_proc_relink(struct prison *opr, struct prison *npr,
165     struct proc *p);
166 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
167     int enable);
168 static char *prison_path(struct prison *pr1, struct prison *pr2);
169 #ifdef RACCT
170 static void prison_racct_attach(struct prison *pr);
171 static void prison_racct_modify(struct prison *pr);
172 static void prison_racct_detach(struct prison *pr);
173 #endif
174 static void prison_knote(struct prison *pr, long hint);
175 
176 /* Flags for prison_deref */
177 #define	PD_DEREF	0x01	/* Decrement pr_ref */
178 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
179 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
180 #define	PD_LOCKED	0x10	/* pr_mtx is held */
181 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
182 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
183 #define PD_OP_FLAGS	0x07	/* Operation flags */
184 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
185 
186 /*
187  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
188  * as we cannot figure out the size of a sparse array, or an array without a
189  * terminating entry.
190  */
191 static struct bool_flags pr_flag_bool[] = {
192 	{"persist", "nopersist", PR_PERSIST},
193 #ifdef INET
194 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
195 #endif
196 #ifdef INET6
197 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
198 #endif
199 };
200 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
201 
202 static struct jailsys_flags pr_flag_jailsys[] = {
203 	{"host", 0, PR_HOST},
204 #ifdef VIMAGE
205 	{"vnet", 0, PR_VNET},
206 #endif
207 #ifdef INET
208 	{"ip4", PR_IP4_USER, PR_IP4_USER},
209 #endif
210 #ifdef INET6
211 	{"ip6", PR_IP6_USER, PR_IP6_USER},
212 #endif
213 };
214 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
215 
216 /*
217  * Make this array full-size so dynamic parameters can be added.
218  * It is protected by prison0.mtx, but lockless reading is allowed
219  * with an atomic check of the flag values.
220  */
221 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
222 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
223 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
224 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
225 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
226 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
227 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
228 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
229 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
230 	{"allow.reserved_ports", "allow.noreserved_ports",
231 	 PR_ALLOW_RESERVED_PORTS},
232 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
233 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
234 	 PR_ALLOW_UNPRIV_DEBUG},
235 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
236 #ifdef VIMAGE
237 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
238 #endif
239 	{"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
240 	{"allow.adjtime", "allow.noadjtime", PR_ALLOW_ADJTIME},
241 	{"allow.settime", "allow.nosettime", PR_ALLOW_SETTIME},
242 	{"allow.routing", "allow.norouting", PR_ALLOW_ROUTING},
243 	{"allow.unprivileged_parent_tampering",
244 	    "allow.nounprivileged_parent_tampering",
245 	    PR_ALLOW_UNPRIV_PARENT_TAMPER},
246 #ifdef AUDIT
247 	{"allow.setaudit", "allow.nosetaudit", PR_ALLOW_SETAUDIT},
248 #endif
249 };
250 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
251 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
252 
253 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
254 					 PR_ALLOW_RESERVED_PORTS | \
255 					 PR_ALLOW_UNPRIV_DEBUG | \
256 					 PR_ALLOW_SUSER)
257 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
258 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
259 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
260 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
261 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
262 #if defined(INET) || defined(INET6)
263 static unsigned jail_max_af_ips = 255;
264 #endif
265 
266 /*
267  * Initialize the parts of prison0 that can't be static-initialized with
268  * constants.  This is called from proc0_init() after creating thread0 cpuset.
269  */
270 void
prison0_init(void)271 prison0_init(void)
272 {
273 	uint8_t *file, *data;
274 	size_t size;
275 	char buf[sizeof(prison0.pr_hostuuid)];
276 	bool valid;
277 
278 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
279 	prison0.pr_osreldate = osreldate;
280 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
281 
282 	/* If we have a preloaded hostuuid, use it. */
283 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
284 	if (file != NULL) {
285 		data = preload_fetch_addr(file);
286 		size = preload_fetch_size(file);
287 		if (data != NULL) {
288 			/*
289 			 * The preloaded data may include trailing whitespace, almost
290 			 * certainly a newline; skip over any whitespace or
291 			 * non-printable characters to be safe.
292 			 */
293 			while (size > 0 && data[size - 1] <= 0x20) {
294 				size--;
295 			}
296 
297 			valid = false;
298 
299 			/*
300 			 * Not NUL-terminated when passed from loader, but
301 			 * validate_uuid requires that due to using sscanf (as
302 			 * does the subsequent strlcpy, since it still reads
303 			 * past the given size to return the true length);
304 			 * bounce to a temporary buffer to fix.
305 			 */
306 			if (size >= sizeof(buf))
307 				goto done;
308 
309 			memcpy(buf, data, size);
310 			buf[size] = '\0';
311 
312 			if (validate_uuid(buf, size, NULL, 0) != 0)
313 				goto done;
314 
315 			valid = true;
316 			(void)strlcpy(prison0.pr_hostuuid, buf,
317 			    sizeof(prison0.pr_hostuuid));
318 
319 done:
320 			if (bootverbose && !valid) {
321 				printf("hostuuid: preload data malformed: '%.*s'\n",
322 				    (int)size, data);
323 			}
324 		}
325 	}
326 	if (bootverbose)
327 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
328 }
329 
330 /*
331  * struct jail_args {
332  *	struct jail *jail;
333  * };
334  */
335 int
sys_jail(struct thread * td,struct jail_args * uap)336 sys_jail(struct thread *td, struct jail_args *uap)
337 {
338 	uint32_t version;
339 	int error;
340 	struct jail j;
341 
342 	error = copyin(uap->jail, &version, sizeof(uint32_t));
343 	if (error)
344 		return (error);
345 
346 	switch (version) {
347 	case 0:
348 	{
349 		struct jail_v0 j0;
350 
351 		/* FreeBSD single IPv4 jails. */
352 		bzero(&j, sizeof(struct jail));
353 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
354 		if (error)
355 			return (error);
356 		j.version = j0.version;
357 		j.path = j0.path;
358 		j.hostname = j0.hostname;
359 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
360 		break;
361 	}
362 
363 	case 1:
364 		/*
365 		 * Version 1 was used by multi-IPv4 jail implementations
366 		 * that never made it into the official kernel.
367 		 */
368 		return (EINVAL);
369 
370 	case 2:	/* JAIL_API_VERSION */
371 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
372 		error = copyin(uap->jail, &j, sizeof(struct jail));
373 		if (error)
374 			return (error);
375 		break;
376 
377 	default:
378 		/* Sci-Fi jails are not supported, sorry. */
379 		return (EINVAL);
380 	}
381 	return (kern_jail(td, &j));
382 }
383 
384 int
kern_jail(struct thread * td,struct jail * j)385 kern_jail(struct thread *td, struct jail *j)
386 {
387 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
388 #ifdef INET
389 			    + 1
390 #endif
391 #ifdef INET6
392 			    + 1
393 #endif
394 			    )];
395 	struct uio opt;
396 	char *u_path, *u_hostname, *u_name;
397 	struct bool_flags *bf;
398 #ifdef INET
399 	uint32_t ip4s;
400 	struct in_addr *u_ip4;
401 #endif
402 #ifdef INET6
403 	struct in6_addr *u_ip6;
404 #endif
405 	size_t tmplen;
406 	int error, enforce_statfs;
407 
408 	bzero(&optiov, sizeof(optiov));
409 	opt.uio_iov = optiov;
410 	opt.uio_iovcnt = 0;
411 	opt.uio_offset = -1;
412 	opt.uio_resid = -1;
413 	opt.uio_segflg = UIO_SYSSPACE;
414 	opt.uio_rw = UIO_READ;
415 	opt.uio_td = td;
416 
417 	/* Set permissions for top-level jails from sysctls. */
418 	if (!jailed(td->td_ucred)) {
419 		for (bf = pr_flag_allow;
420 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
421 			atomic_load_int(&bf->flag) != 0;
422 		     bf++) {
423 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
424 			    (jail_default_allow & bf->flag)
425 			    ? bf->name : bf->noname);
426 			optiov[opt.uio_iovcnt].iov_len =
427 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
428 			opt.uio_iovcnt += 2;
429 		}
430 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
431 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
432 		opt.uio_iovcnt++;
433 		enforce_statfs = jail_default_enforce_statfs;
434 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
435 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
436 		opt.uio_iovcnt++;
437 	}
438 
439 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
440 #ifdef INET
441 	ip4s = (j->version == 0) ? 1 : j->ip4s;
442 	if (ip4s > jail_max_af_ips)
443 		return (EINVAL);
444 	tmplen += ip4s * sizeof(struct in_addr);
445 #else
446 	if (j->ip4s > 0)
447 		return (EINVAL);
448 #endif
449 #ifdef INET6
450 	if (j->ip6s > jail_max_af_ips)
451 		return (EINVAL);
452 	tmplen += j->ip6s * sizeof(struct in6_addr);
453 #else
454 	if (j->ip6s > 0)
455 		return (EINVAL);
456 #endif
457 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
458 	u_hostname = u_path + MAXPATHLEN;
459 	u_name = u_hostname + MAXHOSTNAMELEN;
460 #ifdef INET
461 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
462 #endif
463 #ifdef INET6
464 #ifdef INET
465 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
466 #else
467 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
468 #endif
469 #endif
470 	optiov[opt.uio_iovcnt].iov_base = "path";
471 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
472 	opt.uio_iovcnt++;
473 	optiov[opt.uio_iovcnt].iov_base = u_path;
474 	error = copyinstr(j->path, u_path, MAXPATHLEN,
475 	    &optiov[opt.uio_iovcnt].iov_len);
476 	if (error) {
477 		free(u_path, M_TEMP);
478 		return (error);
479 	}
480 	opt.uio_iovcnt++;
481 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
482 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
483 	opt.uio_iovcnt++;
484 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
485 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
486 	    &optiov[opt.uio_iovcnt].iov_len);
487 	if (error) {
488 		free(u_path, M_TEMP);
489 		return (error);
490 	}
491 	opt.uio_iovcnt++;
492 	if (j->jailname != NULL) {
493 		optiov[opt.uio_iovcnt].iov_base = "name";
494 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
495 		opt.uio_iovcnt++;
496 		optiov[opt.uio_iovcnt].iov_base = u_name;
497 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
498 		    &optiov[opt.uio_iovcnt].iov_len);
499 		if (error) {
500 			free(u_path, M_TEMP);
501 			return (error);
502 		}
503 		opt.uio_iovcnt++;
504 	}
505 #ifdef INET
506 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
507 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
508 	opt.uio_iovcnt++;
509 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
510 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
511 	if (j->version == 0)
512 		u_ip4->s_addr = j->ip4s;
513 	else {
514 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
515 		if (error) {
516 			free(u_path, M_TEMP);
517 			return (error);
518 		}
519 	}
520 	opt.uio_iovcnt++;
521 #endif
522 #ifdef INET6
523 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
524 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
525 	opt.uio_iovcnt++;
526 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
527 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
528 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
529 	if (error) {
530 		free(u_path, M_TEMP);
531 		return (error);
532 	}
533 	opt.uio_iovcnt++;
534 #endif
535 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
536 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
537 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
538 	free(u_path, M_TEMP);
539 	return (error);
540 }
541 
542 /*
543  * struct jail_set_args {
544  *	struct iovec *iovp;
545  *	unsigned int iovcnt;
546  *	int flags;
547  * };
548  */
549 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)550 sys_jail_set(struct thread *td, struct jail_set_args *uap)
551 {
552 	struct uio *auio;
553 	int error;
554 
555 	/* Check that we have an even number of iovecs. */
556 	if (uap->iovcnt & 1)
557 		return (EINVAL);
558 
559 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
560 	if (error)
561 		return (error);
562 	error = kern_jail_set(td, auio, uap->flags);
563 	freeuio(auio);
564 	return (error);
565 }
566 
567 #if defined(INET) || defined(INET6)
568 typedef int prison_addr_cmp_t(const void *, const void *);
569 typedef bool prison_addr_valid_t(const void *);
570 static const struct pr_family {
571 	size_t			size;
572 	prison_addr_cmp_t	*cmp;
573 	prison_addr_valid_t	*valid;
574 	int			ip_flag;
575 } pr_families[PR_FAMILY_MAX] = {
576 #ifdef INET
577 	[PR_INET] = {
578 		.size = sizeof(struct in_addr),
579 		.cmp = prison_qcmp_v4,
580 		.valid = prison_valid_v4,
581 		.ip_flag = PR_IP4_USER,
582 	 },
583 #endif
584 #ifdef INET6
585 	[PR_INET6] = {
586 		.size = sizeof(struct in6_addr),
587 		.cmp = prison_qcmp_v6,
588 		.valid = prison_valid_v6,
589 		.ip_flag = PR_IP6_USER,
590 	},
591 #endif
592 };
593 
594 /*
595  * Network address lists (pr_addrs) allocation for jails.  The addresses
596  * are accessed locklessly by the network stack, thus need to be protected by
597  * the network epoch.
598  */
599 struct prison_ip {
600 	struct epoch_context ctx;
601 	uint32_t	ips;
602 #ifdef FUTURE_C
603 	/*
604 	 * XXX Variable-length automatic arrays in union may be
605 	 * supported in future C.
606 	 */
607 	union {
608 		char pr_ip[];
609 		struct in_addr pr_ip4[];
610 		struct in6_addr pr_ip6[];
611 	};
612 #else /* No future C :( */
613 	char pr_ip[];
614 #endif
615 };
616 
617 static char *
PR_IP(struct prison_ip * pip,const pr_family_t af,int idx)618 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
619 {
620 	MPASS(pip);
621 	MPASS(af < PR_FAMILY_MAX);
622 	MPASS(idx >= 0 && idx < pip->ips);
623 
624 	return (pip->pr_ip + pr_families[af].size * idx);
625 }
626 
627 static struct prison_ip *
prison_ip_alloc(const pr_family_t af,uint32_t cnt,int flags)628 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
629 {
630 	struct prison_ip *pip;
631 
632 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
633 	    M_PRISON, flags);
634 	if (pip != NULL)
635 		pip->ips = cnt;
636 	return (pip);
637 }
638 
639 /*
640  * Allocate and copyin user supplied address list, sorting and validating.
641  * kern_jail_set() helper.
642  */
643 static struct prison_ip *
prison_ip_copyin(const pr_family_t af,void * op,uint32_t cnt)644 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
645 {
646 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
647 	const size_t size = pr_families[af].size;
648 	struct prison_ip *pip;
649 
650 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
651 	bcopy(op, pip->pr_ip, cnt * size);
652 	/*
653 	 * IP addresses are all sorted but ip[0] to preserve
654 	 * the primary IP address as given from userland.
655 	 * This special IP is used for unbound outgoing
656 	 * connections as well for "loopback" traffic in case
657 	 * source address selection cannot find any more fitting
658 	 * address to connect from.
659 	 */
660 	if (cnt > 1)
661 		qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
662 	/*
663 	 * Check for duplicate addresses and do some simple
664 	 * zero and broadcast checks. If users give other bogus
665 	 * addresses it is their problem.
666 	 */
667 	for (int i = 0; i < cnt; i++) {
668 		if (!pr_families[af].valid(PR_IP(pip, af, i))) {
669 			free(pip, M_PRISON);
670 			return (NULL);
671 		}
672 		if (i + 1 < cnt &&
673 		    (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
674 		     cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
675 			free(pip, M_PRISON);
676 			return (NULL);
677 		}
678 	}
679 
680 	return (pip);
681 }
682 
683 /*
684  * Allocate and dup parent prison address list.
685  * kern_jail_set() helper.
686  */
687 static void
prison_ip_dup(struct prison * ppr,struct prison * pr,const pr_family_t af)688 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
689 {
690 	const struct prison_ip *ppip = ppr->pr_addrs[af];
691 	struct prison_ip *pip;
692 
693 	if (ppip != NULL) {
694 		pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
695 		bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
696 		pr->pr_addrs[af] = pip;
697 	}
698 }
699 
700 /*
701  * Make sure the new set of IP addresses is a subset of the parent's list.
702  * Don't worry about the parent being unlocked, as any setting is done with
703  * allprison_lock held.
704  * kern_jail_set() helper.
705  */
706 static bool
prison_ip_parent_match(struct prison_ip * ppip,struct prison_ip * pip,const pr_family_t af)707 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
708     const pr_family_t af)
709 {
710 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
711 	int i, j;
712 
713 	if (ppip == NULL)
714 		return (false);
715 
716 	for (i = 0; i < ppip->ips; i++)
717 		if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
718 			break;
719 
720 	if (i == ppip->ips)
721 		/* Main address not present in parent. */
722 		return (false);
723 
724 	if (pip->ips > 1) {
725 		for (i = j = 1; i < pip->ips; i++) {
726 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
727 				/* Equals to parent primary address. */
728 				continue;
729 			for (; j < ppip->ips; j++)
730 				if (cmp(PR_IP(pip, af, i),
731 				    PR_IP(ppip, af, j)) == 0)
732 					break;
733 			if (j == ppip->ips)
734 				break;
735 		}
736 		if (j == ppip->ips)
737 			/* Address not present in parent. */
738 			return (false);
739 	}
740 	return (true);
741 }
742 
743 /*
744  * Check for conflicting IP addresses.  We permit them if there is no more
745  * than one IP on each jail.  If there is a duplicate on a jail with more
746  * than one IP stop checking and return error.
747  * kern_jail_set() helper.
748  */
749 static bool
prison_ip_conflict_check(const struct prison * ppr,const struct prison * pr,struct prison_ip * pip,pr_family_t af)750 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
751     struct prison_ip *pip, pr_family_t af)
752 {
753 	const struct prison *tppr, *tpr;
754 	int descend;
755 
756 #ifdef VIMAGE
757 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
758 		if (tppr->pr_flags & PR_VNET)
759 			break;
760 #else
761 	tppr = &prison0;
762 #endif
763 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
764 		if (tpr == pr ||
765 #ifdef VIMAGE
766 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
767 #endif
768 		    !prison_isalive(tpr)) {
769 			descend = 0;
770 			continue;
771 		}
772 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
773 			continue;
774 		descend = 0;
775 		if (tpr->pr_addrs[af] == NULL ||
776 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
777 			continue;
778 		for (int i = 0; i < pip->ips; i++)
779 			if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
780 				return (false);
781 	}
782 
783 	return (true);
784 }
785 
786 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
787     "prison must start with epoch context");
788 static void
prison_ip_free_deferred(epoch_context_t ctx)789 prison_ip_free_deferred(epoch_context_t ctx)
790 {
791 
792 	free(ctx, M_PRISON);
793 }
794 
795 static void
prison_ip_free(struct prison_ip * pip)796 prison_ip_free(struct prison_ip *pip)
797 {
798 
799 	if (pip != NULL)
800 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
801 }
802 
803 static void
prison_ip_set(struct prison * pr,const pr_family_t af,struct prison_ip * new)804 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
805 {
806 	struct prison_ip **mem, *old;
807 
808 	mtx_assert(&pr->pr_mtx, MA_OWNED);
809 
810 	mem = &pr->pr_addrs[af];
811 
812 	old = *mem;
813 	atomic_store_ptr(mem, new);
814 	prison_ip_free(old);
815 }
816 
817 /*
818  * Restrict a prison's IP address list with its parent's, possibly replacing
819  * it.  Return true if succeed, otherwise should redo.
820  * kern_jail_set() helper.
821  */
822 static bool
prison_ip_restrict(struct prison * pr,const pr_family_t af,struct prison_ip ** newp)823 prison_ip_restrict(struct prison *pr, const pr_family_t af,
824     struct prison_ip **newp)
825 {
826 	struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
827 	struct prison_ip *pip = pr->pr_addrs[af];
828 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
829 	const size_t size = pr_families[af].size;
830 	struct prison_ip *new = newp != NULL ? *newp : NULL;
831 	uint32_t ips;
832 
833 	mtx_assert(&pr->pr_mtx, MA_OWNED);
834 
835 	/*
836 	 * Due to epoch-synchronized access to the IP address lists we always
837 	 * allocate a new list even if the old one has enough space.  We could
838 	 * atomically update an IPv4 address inside a list, but that would
839 	 * screw up sorting, and in case of IPv6 we can't even atomically write
840 	 * one.
841 	 */
842 	if (ppip == NULL) {
843 		if (pip != NULL)
844 			prison_ip_set(pr, af, NULL);
845 		return (true);
846 	}
847 
848 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
849 		if (new == NULL) {
850 			new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
851 			if (new == NULL)
852 				return (false); /* Redo */
853 		}
854 		/* This has no user settings, so just copy the parent's list. */
855 		MPASS(new->ips == ppip->ips);
856 		bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
857 		prison_ip_set(pr, af, new);
858 		if (newp != NULL)
859 			*newp = NULL; /* Used */
860 	} else if (pip != NULL) {
861 		/* Remove addresses that aren't in the parent. */
862 		int i;
863 
864 		i = 0; /* index in pip */
865 		ips = 0; /* index in new */
866 
867 		if (new == NULL) {
868 			new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
869 			if (new == NULL)
870 				return (false); /* Redo */
871 		}
872 
873 		for (int pi = 0; pi < ppip->ips; pi++)
874 			if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
875 				/* Found our primary address in parent. */
876 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
877 				    size);
878 				i++;
879 				ips++;
880 				break;
881 			}
882 		for (int pi = 1; i < pip->ips; ) {
883 			/* Check against primary, which is unsorted. */
884 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
885 				/* Matches parent's primary address. */
886 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
887 				    size);
888 				i++;
889 				ips++;
890 				continue;
891 			}
892 			/* The rest are sorted. */
893 			switch (pi >= ppip->ips ? -1 :
894 				cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
895 			case -1:
896 				i++;
897 				break;
898 			case 0:
899 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
900 				    size);
901 				i++;
902 				pi++;
903 				ips++;
904 				break;
905 			case 1:
906 				pi++;
907 				break;
908 			}
909 		}
910 		if (ips == 0) {
911 			if (newp == NULL || *newp == NULL)
912 				prison_ip_free(new);
913 			new = NULL;
914 		} else {
915 			/* Shrink to real size */
916 			KASSERT((new->ips >= ips),
917 			    ("Out-of-bounds write to prison_ip %p", new));
918 			new->ips = ips;
919 		}
920 		prison_ip_set(pr, af, new);
921 		if (newp != NULL)
922 			*newp = NULL; /* Used */
923 	}
924 	return (true);
925 }
926 
927 /*
928  * Fast-path check if an address belongs to a prison.
929  */
930 int
prison_ip_check(const struct prison * pr,const pr_family_t af,const void * addr)931 prison_ip_check(const struct prison *pr, const pr_family_t af,
932     const void *addr)
933 {
934 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
935 	struct prison_ip *pip;
936 	int i, a, z, d;
937 
938 	MPASS(mtx_owned(&pr->pr_mtx) ||
939 	    in_epoch(net_epoch_preempt) ||
940 	    sx_xlocked(&allprison_lock));
941 
942 	pip = atomic_load_ptr(&pr->pr_addrs[af]);
943 	if (__predict_false(pip == NULL))
944 		return (EAFNOSUPPORT);
945 
946 	/* Check the primary IP. */
947 	if (cmp(PR_IP(pip, af, 0), addr) == 0)
948 		return (0);
949 
950 	/*
951 	 * All the other IPs are sorted so we can do a binary search.
952 	 */
953 	a = 0;
954 	z = pip->ips - 2;
955 	while (a <= z) {
956 		i = (a + z) / 2;
957 		d = cmp(PR_IP(pip, af, i + 1), addr);
958 		if (d > 0)
959 			z = i - 1;
960 		else if (d < 0)
961 			a = i + 1;
962 		else
963 			return (0);
964 	}
965 
966 	return (EADDRNOTAVAIL);
967 }
968 
969 /*
970  * Grab primary IP.  Historically required mutex, but nothing prevents
971  * us to support epoch-protected access.  Is it used in fast path?
972  * in{6}_jail.c helper
973  */
974 const void *
prison_ip_get0(const struct prison * pr,const pr_family_t af)975 prison_ip_get0(const struct prison *pr, const pr_family_t af)
976 {
977 	const struct prison_ip *pip = pr->pr_addrs[af];
978 
979 	mtx_assert(&pr->pr_mtx, MA_OWNED);
980 	MPASS(pip);
981 
982 	return (pip->pr_ip);
983 }
984 
985 u_int
prison_ip_cnt(const struct prison * pr,const pr_family_t af)986 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
987 {
988 
989 	return (pr->pr_addrs[af]->ips);
990 }
991 #endif	/* defined(INET) || defined(INET6) */
992 
993 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)994 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
995 {
996 	struct file *jfp_out;
997 	struct nameidata nd;
998 #ifdef INET
999 	struct prison_ip *ip4;
1000 #endif
1001 #ifdef INET6
1002 	struct prison_ip *ip6;
1003 #endif
1004 	struct vfsopt *opt;
1005 	struct vfsoptlist *opts;
1006 	struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
1007 	struct ucred *jdcred;
1008 	struct vnode *root;
1009 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
1010 	char *g_path, *osrelstr;
1011 	struct bool_flags *bf;
1012 	struct jailsys_flags *jsf;
1013 #if defined(INET) || defined(INET6)
1014 	void *op;
1015 #endif
1016 	unsigned long hid;
1017 	size_t namelen, onamelen, pnamelen;
1018 	int created, cuflags, descend, drflags, enforce;
1019 	int error, errmsg_len, errmsg_pos;
1020 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1021 	int deadid, jfd_in, jfd_out, jfd_pos, jid, jsys, len, level;
1022 	int childmax, osreldt, rsnum, slevel;
1023 #ifdef INET
1024 	int ip4s;
1025 	bool redo_ip4;
1026 #endif
1027 #ifdef INET6
1028 	int ip6s;
1029 	bool redo_ip6;
1030 #endif
1031 	bool maybe_changed;
1032 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1033 	uint64_t pr_allow_diff;
1034 	unsigned tallow;
1035 	char numbuf[12];
1036 
1037 	mypr = td->td_ucred->cr_prison;
1038 	if (((flags & (JAIL_CREATE | JAIL_AT_DESC)) == JAIL_CREATE) &&
1039 	    mypr->pr_childmax == 0)
1040 		return (EPERM);
1041 	if (flags & ~JAIL_SET_MASK)
1042 		return (EINVAL);
1043 	if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
1044 	    (JAIL_USE_DESC | JAIL_AT_DESC))
1045 		return (EINVAL);
1046 	prison_hold(mypr);
1047 
1048 #ifdef INET
1049 	ip4 = NULL;
1050 #endif
1051 #ifdef INET6
1052 	ip6 = NULL;
1053 #endif
1054 	g_path = NULL;
1055 	jfp_out = NULL;
1056 	jfd_out = -1;
1057 	/*
1058 	 * Check all the parameters before committing to anything.  Not all
1059 	 * errors can be caught early, but we may as well try.  Also, this
1060 	 * takes care of some expensive stuff (path lookup) before getting
1061 	 * the allprison lock.
1062 	 *
1063 	 * XXX Jails are not filesystems, and jail parameters are not mount
1064 	 *     options.  But it makes more sense to re-use the vfsopt code
1065 	 *     than duplicate it under a different name.
1066 	 */
1067 	error = vfs_buildopts(optuio, &opts);
1068 	if (error)
1069 		goto done_free;
1070 
1071 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1072 	if (!cuflags) {
1073 		error = EINVAL;
1074 		vfs_opterror(opts, "no valid operation (create or update)");
1075 		goto done_errmsg;
1076 	}
1077 
1078 	error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
1079 	if (error == ENOENT) {
1080 		if (flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1081 		    JAIL_OWN_DESC)) {
1082 			vfs_opterror(opts, "missing desc");
1083 			goto done_errmsg;
1084 		}
1085 		jfd_in = -1;
1086 	} else if (error != 0)
1087 		goto done_free;
1088 	else {
1089 		if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1090 		    JAIL_OWN_DESC))) {
1091 			error = EINVAL;
1092 			vfs_opterror(opts, "unexpected desc");
1093 			goto done_errmsg;
1094 		}
1095 		if (flags & JAIL_AT_DESC) {
1096 			/*
1097 			 * Look up and create jails based on the
1098 			 * descriptor's prison.
1099 			 */
1100 			prison_free(mypr);
1101 			error = jaildesc_find(td, jfd_in, &mypr, NULL);
1102 			if (error != 0) {
1103 				vfs_opterror(opts, error == ENOENT ?
1104 				    "descriptor to dead jail" :
1105 				    "not a jail descriptor");
1106 				goto done_errmsg;
1107 			}
1108 			if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) {
1109 				error = EPERM;
1110 				goto done_free;
1111 			}
1112 		}
1113 		if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
1114 			/* Allocate a jail descriptor to return later. */
1115 			error = jaildesc_alloc(td, &jfp_out, &jfd_out,
1116 				flags & JAIL_OWN_DESC);
1117 			if (error)
1118 				goto done_free;
1119 		}
1120 	}
1121 
1122 	/*
1123 	 * Delay the permission check if using a jail descriptor,
1124 	 * until we get the descriptor's credentials.
1125 	 */
1126 	if (!(flags & JAIL_USE_DESC)) {
1127 		error = priv_check(td, PRIV_JAIL_SET);
1128 		if (error == 0 && (flags & JAIL_ATTACH))
1129 			error = priv_check(td, PRIV_JAIL_ATTACH);
1130 		if (error)
1131 			goto done_free;
1132 	}
1133 
1134 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1135 	if (error == ENOENT)
1136 		jid = 0;
1137 	else if (error != 0)
1138 		goto done_free;
1139 
1140 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1141 	if (error == ENOENT)
1142 		gotslevel = 0;
1143 	else if (error != 0)
1144 		goto done_free;
1145 	else
1146 		gotslevel = 1;
1147 
1148 	error =
1149 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1150 	if (error == ENOENT)
1151 		gotchildmax = 0;
1152 	else if (error != 0)
1153 		goto done_free;
1154 	else
1155 		gotchildmax = 1;
1156 
1157 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1158 	if (error == ENOENT)
1159 		gotenforce = 0;
1160 	else if (error != 0)
1161 		goto done_free;
1162 	else if (enforce < 0 || enforce > 2) {
1163 		error = EINVAL;
1164 		goto done_free;
1165 	} else
1166 		gotenforce = 1;
1167 
1168 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1169 	if (error == ENOENT)
1170 		gotrsnum = 0;
1171 	else if (error != 0)
1172 		goto done_free;
1173 	else
1174 		gotrsnum = 1;
1175 
1176 	pr_flags = ch_flags = 0;
1177 	for (bf = pr_flag_bool;
1178 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1179 	     bf++) {
1180 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1181 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1182 	}
1183 	ch_flags |= pr_flags;
1184 	for (jsf = pr_flag_jailsys;
1185 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1186 	     jsf++) {
1187 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1188 		if (error == ENOENT)
1189 			continue;
1190 		if (error != 0)
1191 			goto done_free;
1192 		switch (jsys) {
1193 		case JAIL_SYS_DISABLE:
1194 			if (!jsf->disable) {
1195 				error = EINVAL;
1196 				goto done_free;
1197 			}
1198 			pr_flags |= jsf->disable;
1199 			break;
1200 		case JAIL_SYS_NEW:
1201 			pr_flags |= jsf->new;
1202 			break;
1203 		case JAIL_SYS_INHERIT:
1204 			break;
1205 		default:
1206 			error = EINVAL;
1207 			goto done_free;
1208 		}
1209 		ch_flags |= jsf->new | jsf->disable;
1210 	}
1211 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1212 	    && !(pr_flags & PR_PERSIST)) {
1213 		error = EINVAL;
1214 		vfs_opterror(opts, "new jail must persist or attach");
1215 		goto done_errmsg;
1216 	}
1217 #ifdef VIMAGE
1218 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1219 		error = EINVAL;
1220 		vfs_opterror(opts, "vnet cannot be changed after creation");
1221 		goto done_errmsg;
1222 	}
1223 #endif
1224 #ifdef INET
1225 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1226 		error = EINVAL;
1227 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1228 		goto done_errmsg;
1229 	}
1230 #endif
1231 #ifdef INET6
1232 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1233 		error = EINVAL;
1234 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1235 		goto done_errmsg;
1236 	}
1237 #endif
1238 
1239 	pr_allow = ch_allow = 0;
1240 	for (bf = pr_flag_allow;
1241 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1242 		atomic_load_int(&bf->flag) != 0;
1243 	     bf++) {
1244 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1245 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1246 	}
1247 	ch_allow |= pr_allow;
1248 
1249 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1250 	if (error == ENOENT)
1251 		name = NULL;
1252 	else if (error != 0)
1253 		goto done_free;
1254 	else {
1255 		if (len == 0 || name[len - 1] != '\0') {
1256 			error = EINVAL;
1257 			goto done_free;
1258 		}
1259 		if (len > MAXHOSTNAMELEN) {
1260 			error = ENAMETOOLONG;
1261 			goto done_free;
1262 		}
1263 	}
1264 
1265 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1266 	if (error == ENOENT)
1267 		host = NULL;
1268 	else if (error != 0)
1269 		goto done_free;
1270 	else {
1271 		ch_flags |= PR_HOST;
1272 		pr_flags |= PR_HOST;
1273 		if (len == 0 || host[len - 1] != '\0') {
1274 			error = EINVAL;
1275 			goto done_free;
1276 		}
1277 		if (len > MAXHOSTNAMELEN) {
1278 			error = ENAMETOOLONG;
1279 			goto done_free;
1280 		}
1281 	}
1282 
1283 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1284 	if (error == ENOENT)
1285 		domain = NULL;
1286 	else if (error != 0)
1287 		goto done_free;
1288 	else {
1289 		ch_flags |= PR_HOST;
1290 		pr_flags |= PR_HOST;
1291 		if (len == 0 || domain[len - 1] != '\0') {
1292 			error = EINVAL;
1293 			goto done_free;
1294 		}
1295 		if (len > MAXHOSTNAMELEN) {
1296 			error = ENAMETOOLONG;
1297 			goto done_free;
1298 		}
1299 	}
1300 
1301 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1302 	if (error == ENOENT)
1303 		uuid = NULL;
1304 	else if (error != 0)
1305 		goto done_free;
1306 	else {
1307 		ch_flags |= PR_HOST;
1308 		pr_flags |= PR_HOST;
1309 		if (len == 0 || uuid[len - 1] != '\0') {
1310 			error = EINVAL;
1311 			goto done_free;
1312 		}
1313 		if (len > HOSTUUIDLEN) {
1314 			error = ENAMETOOLONG;
1315 			goto done_free;
1316 		}
1317 	}
1318 
1319 #ifdef COMPAT_FREEBSD32
1320 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1321 		uint32_t hid32;
1322 
1323 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1324 		hid = hid32;
1325 	} else
1326 #endif
1327 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1328 	if (error == ENOENT)
1329 		gothid = 0;
1330 	else if (error != 0)
1331 		goto done_free;
1332 	else {
1333 		gothid = 1;
1334 		ch_flags |= PR_HOST;
1335 		pr_flags |= PR_HOST;
1336 	}
1337 
1338 #ifdef INET
1339 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1340 	if (error == ENOENT)
1341 		ip4s = 0;
1342 	else if (error != 0)
1343 		goto done_free;
1344 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1345 		error = EINVAL;
1346 		goto done_free;
1347 	} else {
1348 		ch_flags |= PR_IP4_USER;
1349 		pr_flags |= PR_IP4_USER;
1350 		if (ip4s > 0) {
1351 			ip4s /= sizeof(struct in_addr);
1352 			if (ip4s > jail_max_af_ips) {
1353 				error = EINVAL;
1354 				vfs_opterror(opts, "too many IPv4 addresses");
1355 				goto done_errmsg;
1356 			}
1357 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1358 			if (ip4 == NULL) {
1359 				error = EINVAL;
1360 				goto done_free;
1361 			}
1362 		}
1363 	}
1364 #endif
1365 
1366 #ifdef INET6
1367 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1368 	if (error == ENOENT)
1369 		ip6s = 0;
1370 	else if (error != 0)
1371 		goto done_free;
1372 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1373 		error = EINVAL;
1374 		goto done_free;
1375 	} else {
1376 		ch_flags |= PR_IP6_USER;
1377 		pr_flags |= PR_IP6_USER;
1378 		if (ip6s > 0) {
1379 			ip6s /= sizeof(struct in6_addr);
1380 			if (ip6s > jail_max_af_ips) {
1381 				error = EINVAL;
1382 				vfs_opterror(opts, "too many IPv6 addresses");
1383 				goto done_errmsg;
1384 			}
1385 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1386 			if (ip6 == NULL) {
1387 				error = EINVAL;
1388 				goto done_free;
1389 			}
1390 		}
1391 	}
1392 #endif
1393 
1394 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1395 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1396 		error = EINVAL;
1397 		vfs_opterror(opts,
1398 		    "vnet jails cannot have IP address restrictions");
1399 		goto done_errmsg;
1400 	}
1401 #endif
1402 
1403 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1404 	if (error == ENOENT)
1405 		osrelstr = NULL;
1406 	else if (error != 0)
1407 		goto done_free;
1408 	else {
1409 		if (flags & JAIL_UPDATE) {
1410 			error = EINVAL;
1411 			vfs_opterror(opts,
1412 			    "osrelease cannot be changed after creation");
1413 			goto done_errmsg;
1414 		}
1415 		if (len == 0 || osrelstr[len - 1] != '\0') {
1416 			error = EINVAL;
1417 			goto done_free;
1418 		}
1419 		if (len >= OSRELEASELEN) {
1420 			error = ENAMETOOLONG;
1421 			vfs_opterror(opts,
1422 			    "osrelease string must be 1-%d bytes long",
1423 			    OSRELEASELEN - 1);
1424 			goto done_errmsg;
1425 		}
1426 	}
1427 
1428 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1429 	if (error == ENOENT)
1430 		osreldt = 0;
1431 	else if (error != 0)
1432 		goto done_free;
1433 	else {
1434 		if (flags & JAIL_UPDATE) {
1435 			error = EINVAL;
1436 			vfs_opterror(opts,
1437 			    "osreldate cannot be changed after creation");
1438 			goto done_errmsg;
1439 		}
1440 		if (osreldt == 0) {
1441 			error = EINVAL;
1442 			vfs_opterror(opts, "osreldate cannot be 0");
1443 			goto done_errmsg;
1444 		}
1445 	}
1446 
1447 	root = NULL;
1448 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1449 	if (error == ENOENT)
1450 		path = NULL;
1451 	else if (error != 0)
1452 		goto done_free;
1453 	else {
1454 		if (flags & JAIL_UPDATE) {
1455 			error = EINVAL;
1456 			vfs_opterror(opts,
1457 			    "path cannot be changed after creation");
1458 			goto done_errmsg;
1459 		}
1460 		if (len == 0 || path[len - 1] != '\0') {
1461 			error = EINVAL;
1462 			goto done_free;
1463 		}
1464 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1465 		error = namei(&nd);
1466 		if (error)
1467 			goto done_free;
1468 		root = nd.ni_vp;
1469 		NDFREE_PNBUF(&nd);
1470 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1471 		strlcpy(g_path, path, MAXPATHLEN);
1472 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1473 		if (error == 0) {
1474 			path = g_path;
1475 		} else {
1476 			/* exit on other errors */
1477 			goto done_free;
1478 		}
1479 		if (root->v_type != VDIR) {
1480 			error = ENOTDIR;
1481 			vput(root);
1482 			goto done_free;
1483 		}
1484 		VOP_UNLOCK(root);
1485 	}
1486 
1487 	/*
1488 	 * Find the specified jail, or at least its parent.
1489 	 * This abuses the file error codes ENOENT and EEXIST.
1490 	 */
1491 	pr = NULL;
1492 	inspr = NULL;
1493 	deadpr = NULL;
1494 	maybe_changed = false;
1495 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1496 		namelc = strrchr(name, '.');
1497 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1498 		if (*p != '\0')
1499 			jid = 0;
1500 	}
1501 	sx_xlock(&allprison_lock);
1502 	drflags = PD_LIST_XLOCKED;
1503 	ppr = mypr;
1504 	if (!prison_isalive(ppr)) {
1505 		/* This jail is dying.  This process will surely follow. */
1506 		error = EAGAIN;
1507 		goto done_deref;
1508 	}
1509 	if (flags & JAIL_USE_DESC) {
1510 		/* Get the jail from its descriptor. */
1511 		error = jaildesc_find(td, jfd_in, &pr, &jdcred);
1512 		if (error) {
1513 			vfs_opterror(opts, error == ENOENT ?
1514 			    "descriptor to dead jail" :
1515 			    "not a jail descriptor");
1516 			goto done_deref;
1517 		}
1518 		drflags |= PD_DEREF;
1519 		error = priv_check_cred(jdcred, PRIV_JAIL_SET);
1520 		if (error == 0 && (flags & JAIL_ATTACH))
1521 			error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
1522 		crfree(jdcred);
1523 		if (error)
1524 			goto done_deref;
1525 		mtx_lock(&pr->pr_mtx);
1526 		drflags |= PD_LOCKED;
1527 		if (cuflags == JAIL_CREATE) {
1528 			error = EEXIST;
1529 			vfs_opterror(opts, "jail %d already exists",
1530 			    pr->pr_id);
1531 			goto done_deref;
1532 		}
1533 		if (!prison_isalive(pr)) {
1534 			/* While a jid can be resurrected, the prison
1535 			 * itself cannot.
1536 			 */
1537 			error = ENOENT;
1538 			vfs_opterror(opts, "jail %d is dying", pr->pr_id);
1539 			goto done_deref;
1540 		}
1541 		if (jid != 0 && jid != pr->pr_id) {
1542 			error = EINVAL;
1543 			vfs_opterror(opts, "cannot change jid");
1544 			goto done_deref;
1545 		}
1546 		jid = pr->pr_id;
1547 	} else if (jid != 0) {
1548 		if (jid < 0) {
1549 			error = EINVAL;
1550 			vfs_opterror(opts, "negative jid");
1551 			goto done_deref;
1552 		}
1553 		/*
1554 		 * See if a requested jid already exists.  Keep track of
1555 		 * where it can be inserted later.
1556 		 */
1557 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1558 			if (inspr->pr_id < jid)
1559 				continue;
1560 			if (inspr->pr_id > jid)
1561 				break;
1562 			if (prison_isalive(inspr)) {
1563 				pr = inspr;
1564 				mtx_lock(&pr->pr_mtx);
1565 				drflags |= PD_LOCKED;
1566 			} else {
1567 				/* Note a dying jail to handle later. */
1568 				deadpr = inspr;
1569 			}
1570 			inspr = NULL;
1571 			break;
1572 		}
1573 		if (cuflags == JAIL_CREATE && pr != NULL) {
1574 			/*
1575 			 * Even creators that cannot see the jail will
1576 			 * get EEXIST.
1577 			 */
1578 			error = EEXIST;
1579 			vfs_opterror(opts, "jail %d already exists", jid);
1580 			goto done_deref;
1581 		}
1582 		if ((pr == NULL)
1583 		    ? cuflags == JAIL_UPDATE
1584 		    : !prison_ischild(mypr, pr)) {
1585 			/*
1586 			 * Updaters get ENOENT for nonexistent jails,
1587 			 * or for jails they cannot see.  The latter
1588 			 * case is true even for CREATE | UPDATE,
1589 			 * which normally cannot give this error.
1590 			 */
1591 			error = ENOENT;
1592 			vfs_opterror(opts, "jail %d not found", jid);
1593 			goto done_deref;
1594 		}
1595 	}
1596 	/*
1597 	 * If the caller provided a name, look for a jail by that name.
1598 	 * This has different semantics for creates and updates keyed by jid
1599 	 * (where the name must not already exist in a different jail),
1600 	 * and updates keyed by the name itself (where the name must exist
1601 	 * because that is the jail being updated).
1602 	 */
1603 	namelc = NULL;
1604 	if (name != NULL) {
1605 		namelc = strrchr(name, '.');
1606 		if (namelc == NULL)
1607 			namelc = name;
1608 		else {
1609 			/*
1610 			 * This is a hierarchical name.  Split it into the
1611 			 * parent and child names, and make sure the parent
1612 			 * exists or matches an already found jail.
1613 			 */
1614 			if (pr != NULL) {
1615 				if (strncmp(name, ppr->pr_name, namelc - name)
1616 				    || ppr->pr_name[namelc - name] != '\0') {
1617 					error = EINVAL;
1618 					vfs_opterror(opts,
1619 					    "cannot change jail's parent");
1620 					goto done_deref;
1621 				}
1622 			} else {
1623 				*namelc = '\0';
1624 				ppr = prison_find_name(mypr, name);
1625 				if (ppr == NULL) {
1626 					error = ENOENT;
1627 					vfs_opterror(opts,
1628 					    "jail \"%s\" not found", name);
1629 					goto done_deref;
1630 				}
1631 				mtx_unlock(&ppr->pr_mtx);
1632 				if (!prison_isalive(ppr)) {
1633 					error = ENOENT;
1634 					vfs_opterror(opts,
1635 					    "jail \"%s\" is dying", name);
1636 					goto done_deref;
1637 				}
1638 				*namelc = '.';
1639 			}
1640 			namelc++;
1641 		}
1642 		if (namelc[0] != '\0') {
1643 			pnamelen =
1644 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1645 			FOREACH_PRISON_CHILD(ppr, tpr) {
1646 				if (tpr == pr || !prison_isalive(tpr) ||
1647 				    strcmp(tpr->pr_name + pnamelen, namelc))
1648 					continue;
1649 				if (cuflags == JAIL_CREATE || pr != NULL) {
1650 					/*
1651 					 * Create, or update(jid): name must
1652 					 * not exist in an active sibling jail.
1653 					 */
1654 					error = EEXIST;
1655 					vfs_opterror(opts,
1656 					    "jail \"%s\" already exists", name);
1657 					goto done_deref;
1658 				}
1659 				/* Use this jail for updates. */
1660 				pr = tpr;
1661 				mtx_lock(&pr->pr_mtx);
1662 				drflags |= PD_LOCKED;
1663 				break;
1664 			}
1665 			/*
1666 			 * Update: name must exist if no jid is specified.
1667 			 * As with the jid case, the jail must be currently
1668 			 * visible, or else even CREATE | UPDATE will get
1669 			 * an error.
1670 			 */
1671 			if ((pr == NULL)
1672 			    ? cuflags == JAIL_UPDATE
1673 			    : !prison_isalive(pr)) {
1674 				error = ENOENT;
1675 				vfs_opterror(opts, "jail \"%s\" not found",
1676 				    name);
1677 				goto done_deref;
1678 			}
1679 		}
1680 	}
1681 	/* Update: must provide a desc, jid, or name. */
1682 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1683 		error = ENOENT;
1684 		vfs_opterror(opts, "update specified no jail");
1685 		goto done_deref;
1686 	}
1687 
1688 	/* If there's no prison to update, create a new one and link it in. */
1689 	created = pr == NULL;
1690 	if (created) {
1691 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1692 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1693 				error = EPERM;
1694 				vfs_opterror(opts, "prison limit exceeded");
1695 				goto done_deref;
1696 			}
1697 
1698 		if (deadpr != NULL) {
1699 			/*
1700 			 * The prison being created has the same ID as a dying
1701 			 * one.  Handle this by giving the dying jail a new ID.
1702 			 * This may cause some confusion to user space, but
1703 			 * only to those listing dying jails.
1704 			 */
1705 			deadid = get_next_deadid(&dinspr);
1706 			if (deadid == 0) {
1707 				error = EAGAIN;
1708 				vfs_opterror(opts, "no available jail IDs");
1709 				goto done_deref;
1710 			}
1711 			mtx_lock(&deadpr->pr_mtx);
1712 			deadpr->pr_id = deadid;
1713 			mtx_unlock(&deadpr->pr_mtx);
1714 			if (dinspr == deadpr)
1715 				inspr = deadpr;
1716 			else {
1717 				inspr = TAILQ_NEXT(deadpr, pr_list);
1718 				TAILQ_REMOVE(&allprison, deadpr, pr_list);
1719 				if (dinspr != NULL)
1720 					TAILQ_INSERT_AFTER(&allprison, dinspr,
1721 					    deadpr, pr_list);
1722 				else
1723 					TAILQ_INSERT_HEAD(&allprison, deadpr,
1724 					    pr_list);
1725 			}
1726 		}
1727 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1728 			error = EAGAIN;
1729 			vfs_opterror(opts, "no available jail IDs");
1730 			goto done_deref;
1731 		}
1732 
1733 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1734 		pr->pr_state = PRISON_STATE_INVALID;
1735 		refcount_init(&pr->pr_ref, 1);
1736 		refcount_init(&pr->pr_uref, 0);
1737 		drflags |= PD_DEREF;
1738 		LIST_INIT(&pr->pr_children);
1739 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1740 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1741 
1742 		pr->pr_id = jid;
1743 		if (inspr != NULL)
1744 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1745 		else
1746 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1747 
1748 		pr->pr_parent = ppr;
1749 		prison_hold(ppr);
1750 		prison_proc_hold(ppr);
1751 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1752 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1753 			tpr->pr_childcount++;
1754 		pr->pr_klist = knlist_alloc(&pr->pr_mtx);
1755 
1756 		/* Set some default values, and inherit some from the parent. */
1757 		if (namelc == NULL)
1758 			namelc = "";
1759 		if (path == NULL) {
1760 			path = "/";
1761 			root = mypr->pr_root;
1762 			vref(root);
1763 		}
1764 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1765 		pr->pr_flags |= PR_HOST;
1766 #if defined(INET) || defined(INET6)
1767 #ifdef VIMAGE
1768 		if (!(pr_flags & PR_VNET))
1769 #endif
1770 		{
1771 #ifdef INET
1772 			if (!(ch_flags & PR_IP4_USER))
1773 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1774 			else if (!(pr_flags & PR_IP4_USER)) {
1775 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1776 				prison_ip_dup(ppr, pr, PR_INET);
1777 			}
1778 #endif
1779 #ifdef INET6
1780 			if (!(ch_flags & PR_IP6_USER))
1781 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1782 			else if (!(pr_flags & PR_IP6_USER)) {
1783 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1784 				prison_ip_dup(ppr, pr, PR_INET6);
1785 			}
1786 #endif
1787 		}
1788 #endif
1789 		/* Source address selection is always on by default. */
1790 		pr->pr_flags |= _PR_IP_SADDRSEL;
1791 
1792 		pr->pr_securelevel = ppr->pr_securelevel;
1793 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1794 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1795 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1796 
1797 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1798 		if (osrelstr == NULL)
1799 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1800 			    sizeof(pr->pr_osrelease));
1801 		else
1802 			strlcpy(pr->pr_osrelease, osrelstr,
1803 			    sizeof(pr->pr_osrelease));
1804 
1805 #ifdef VIMAGE
1806 		/*
1807 		 * Allocate a new vnet if specified.
1808 		 *
1809 		 * Set PR_VNET now if so, so that the vnet is disposed of
1810 		 * properly when the jail is destroyed.
1811 		 */
1812 		if (pr_flags & PR_VNET) {
1813 			pr->pr_flags |= PR_VNET;
1814 			pr->pr_vnet = vnet_alloc();
1815 		} else {
1816 			pr->pr_vnet = ppr->pr_vnet;
1817 		}
1818 #endif
1819 		/*
1820 		 * Allocate a dedicated cpuset for each jail.
1821 		 * Unlike other initial settings, this may return an error.
1822 		 */
1823 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1824 		if (error)
1825 			goto done_deref;
1826 
1827 		mtx_lock(&pr->pr_mtx);
1828 		drflags |= PD_LOCKED;
1829 	} else {
1830 		/*
1831 		 * Grab a reference for existing prisons, to ensure they
1832 		 * continue to exist for the duration of the call.
1833 		 */
1834 		if (!(drflags & PD_DEREF)) {
1835 			prison_hold(pr);
1836 			drflags |= PD_DEREF;
1837 		}
1838 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1839 		if ((pr->pr_flags & PR_VNET) &&
1840 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1841 			error = EINVAL;
1842 			vfs_opterror(opts,
1843 			    "vnet jails cannot have IP address restrictions");
1844 			goto done_deref;
1845 		}
1846 #endif
1847 #ifdef INET
1848 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1849 			error = EINVAL;
1850 			vfs_opterror(opts,
1851 			    "ip4 cannot be changed after creation");
1852 			goto done_deref;
1853 		}
1854 #endif
1855 #ifdef INET6
1856 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1857 			error = EINVAL;
1858 			vfs_opterror(opts,
1859 			    "ip6 cannot be changed after creation");
1860 			goto done_deref;
1861 		}
1862 #endif
1863 	}
1864 
1865 	/* Do final error checking before setting anything. */
1866 	if (gotslevel) {
1867 		if (slevel < ppr->pr_securelevel) {
1868 			error = EPERM;
1869 			goto done_deref;
1870 		}
1871 	}
1872 	if (gotchildmax) {
1873 		if (childmax >= ppr->pr_childmax) {
1874 			error = EPERM;
1875 			goto done_deref;
1876 		}
1877 	}
1878 	if (gotenforce) {
1879 		if (enforce < ppr->pr_enforce_statfs) {
1880 			error = EPERM;
1881 			goto done_deref;
1882 		}
1883 	}
1884 	if (gotrsnum) {
1885 		/*
1886 		 * devfs_rsnum is a uint16_t
1887 		 */
1888 		if (rsnum < 0 || rsnum > 65535) {
1889 			error = EINVAL;
1890 			goto done_deref;
1891 		}
1892 		/*
1893 		 * Nested jails always inherit parent's devfs ruleset
1894 		 */
1895 		if (jailed(td->td_ucred)) {
1896 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1897 				error = EPERM;
1898 				goto done_deref;
1899 			} else
1900 				rsnum = ppr->pr_devfs_rsnum;
1901 		}
1902 	}
1903 #ifdef INET
1904 	if (ip4s > 0) {
1905 		if ((ppr->pr_flags & PR_IP4) &&
1906 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1907 		    PR_INET)) {
1908 			error = EPERM;
1909 			goto done_deref;
1910 		}
1911 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1912 			error = EADDRINUSE;
1913 			vfs_opterror(opts, "IPv4 addresses clash");
1914 			goto done_deref;
1915 		}
1916 	}
1917 #endif
1918 #ifdef INET6
1919 	if (ip6s > 0) {
1920 		if ((ppr->pr_flags & PR_IP6) &&
1921 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1922 		    PR_INET6)) {
1923 			error = EPERM;
1924 			goto done_deref;
1925 		}
1926 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1927 			error = EADDRINUSE;
1928 			vfs_opterror(opts, "IPv6 addresses clash");
1929 			goto done_deref;
1930 		}
1931 	}
1932 #endif
1933 	onamelen = namelen = 0;
1934 	if (namelc != NULL) {
1935 		/* Give a default name of the jid.  Also allow the name to be
1936 		 * explicitly the jid - but not any other number, and only in
1937 		 * normal form (no leading zero/etc).
1938 		 */
1939 		if (namelc[0] == '\0')
1940 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1941 		else if ((strtoul(namelc, &p, 10) != jid ||
1942 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1943 			error = EINVAL;
1944 			vfs_opterror(opts,
1945 			    "name cannot be numeric (unless it is the jid)");
1946 			goto done_deref;
1947 		}
1948 		/*
1949 		 * Make sure the name isn't too long for the prison or its
1950 		 * children.
1951 		 */
1952 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1953 		onamelen = strlen(pr->pr_name + pnamelen);
1954 		namelen = strlen(namelc);
1955 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1956 			error = ENAMETOOLONG;
1957 			goto done_deref;
1958 		}
1959 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1960 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1961 			    sizeof(pr->pr_name)) {
1962 				error = ENAMETOOLONG;
1963 				goto done_deref;
1964 			}
1965 		}
1966 	}
1967 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1968 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1969 		error = EPERM;
1970 		goto done_deref;
1971 	}
1972 
1973 	/*
1974 	 * Let modules check their parameters.  This requires unlocking and
1975 	 * then re-locking the prison, but this is still a valid state as long
1976 	 * as allprison_lock remains xlocked.
1977 	 */
1978 	mtx_unlock(&pr->pr_mtx);
1979 	drflags &= ~PD_LOCKED;
1980 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1981 	if (error != 0)
1982 		goto done_deref;
1983 	mtx_lock(&pr->pr_mtx);
1984 	drflags |= PD_LOCKED;
1985 
1986 	/* At this point, all valid parameters should have been noted. */
1987 	TAILQ_FOREACH(opt, opts, link) {
1988 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1989 			error = EINVAL;
1990 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1991 			goto done_deref;
1992 		}
1993 	}
1994 	maybe_changed = true;
1995 
1996 	/* Set the parameters of the prison. */
1997 #ifdef INET
1998 	redo_ip4 = false;
1999 	if (pr_flags & PR_IP4_USER) {
2000 		pr->pr_flags |= PR_IP4;
2001 		prison_ip_set(pr, PR_INET, ip4);
2002 		ip4 = NULL;
2003 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2004 #ifdef VIMAGE
2005 			if (tpr->pr_flags & PR_VNET) {
2006 				descend = 0;
2007 				continue;
2008 			}
2009 #endif
2010 			if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
2011 				redo_ip4 = true;
2012 				descend = 0;
2013 			}
2014 		}
2015 	}
2016 #endif
2017 #ifdef INET6
2018 	redo_ip6 = false;
2019 	if (pr_flags & PR_IP6_USER) {
2020 		pr->pr_flags |= PR_IP6;
2021 		prison_ip_set(pr, PR_INET6, ip6);
2022 		ip6 = NULL;
2023 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2024 #ifdef VIMAGE
2025 			if (tpr->pr_flags & PR_VNET) {
2026 				descend = 0;
2027 				continue;
2028 			}
2029 #endif
2030 			if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
2031 				redo_ip6 = true;
2032 				descend = 0;
2033 			}
2034 		}
2035 	}
2036 #endif
2037 	if (gotslevel) {
2038 		pr->pr_securelevel = slevel;
2039 		/* Set all child jails to be at least this level. */
2040 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2041 			if (tpr->pr_securelevel < slevel)
2042 				tpr->pr_securelevel = slevel;
2043 	}
2044 	if (gotchildmax) {
2045 		pr->pr_childmax = childmax;
2046 		/* Set all child jails to under this limit. */
2047 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
2048 			if (tpr->pr_childmax > childmax - level)
2049 				tpr->pr_childmax = childmax > level
2050 				    ? childmax - level : 0;
2051 	}
2052 	if (gotenforce) {
2053 		pr->pr_enforce_statfs = enforce;
2054 		/* Pass this restriction on to the children. */
2055 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2056 			if (tpr->pr_enforce_statfs < enforce)
2057 				tpr->pr_enforce_statfs = enforce;
2058 	}
2059 	if (gotrsnum) {
2060 		pr->pr_devfs_rsnum = rsnum;
2061 		/* Pass this restriction on to the children. */
2062 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2063 			tpr->pr_devfs_rsnum = rsnum;
2064 	}
2065 	if (namelc != NULL) {
2066 		if (ppr == &prison0)
2067 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
2068 		else
2069 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
2070 			    ppr->pr_name, namelc);
2071 		/* Change this component of child names. */
2072 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2073 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
2074 			    strlen(tpr->pr_name + onamelen) + 1);
2075 			bcopy(pr->pr_name, tpr->pr_name, namelen);
2076 		}
2077 	}
2078 	if (path != NULL) {
2079 		/* Try to keep a real-rooted full pathname. */
2080 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
2081 		pr->pr_root = root;
2082 		root = NULL;
2083 	}
2084 	if (PR_HOST & ch_flags & ~pr_flags) {
2085 		if (pr->pr_flags & PR_HOST) {
2086 			/*
2087 			 * Copy the parent's host info.  As with pr_ip4 above,
2088 			 * the lack of a lock on the parent is not a problem;
2089 			 * it is always set with allprison_lock at least
2090 			 * shared, and is held exclusively here.
2091 			 */
2092 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
2093 			    sizeof(pr->pr_hostname));
2094 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
2095 			    sizeof(pr->pr_domainname));
2096 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
2097 			    sizeof(pr->pr_hostuuid));
2098 			pr->pr_hostid = pr->pr_parent->pr_hostid;
2099 		}
2100 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
2101 		/* Set this prison, and any descendants without PR_HOST. */
2102 		if (host != NULL)
2103 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
2104 		if (domain != NULL)
2105 			strlcpy(pr->pr_domainname, domain,
2106 			    sizeof(pr->pr_domainname));
2107 		if (uuid != NULL)
2108 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
2109 		if (gothid)
2110 			pr->pr_hostid = hid;
2111 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2112 			if (tpr->pr_flags & PR_HOST)
2113 				descend = 0;
2114 			else {
2115 				if (host != NULL)
2116 					strlcpy(tpr->pr_hostname,
2117 					    pr->pr_hostname,
2118 					    sizeof(tpr->pr_hostname));
2119 				if (domain != NULL)
2120 					strlcpy(tpr->pr_domainname,
2121 					    pr->pr_domainname,
2122 					    sizeof(tpr->pr_domainname));
2123 				if (uuid != NULL)
2124 					strlcpy(tpr->pr_hostuuid,
2125 					    pr->pr_hostuuid,
2126 					    sizeof(tpr->pr_hostuuid));
2127 				if (gothid)
2128 					tpr->pr_hostid = hid;
2129 			}
2130 		}
2131 	}
2132 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2133 	if ((tallow = ch_allow & ~pr_allow))
2134 		prison_set_allow_locked(pr, tallow, 0);
2135 	/*
2136 	 * Persistent prisons get an extra reference, and prisons losing their
2137 	 * persist flag lose that reference.
2138 	 */
2139 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2140 		if (pr_flags & PR_PERSIST) {
2141 			prison_hold(pr);
2142 			/*
2143 			 * This may be a new prison's first user reference,
2144 			 * but wait to call it alive until after OSD calls
2145 			 * have had a chance to run (and perhaps to fail).
2146 			 */
2147 			refcount_acquire(&pr->pr_uref);
2148 		} else {
2149 			drflags |= PD_DEUREF;
2150 			prison_free_not_last(pr);
2151 		}
2152 	}
2153 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2154 	mtx_unlock(&pr->pr_mtx);
2155 	drflags &= ~PD_LOCKED;
2156 	/*
2157 	 * Any errors past this point will need to de-persist newly created
2158 	 * prisons, as well as call remove methods.
2159 	 */
2160 	if (created)
2161 		drflags |= PD_KILL;
2162 
2163 #ifdef RACCT
2164 	if (racct_enable && created)
2165 		prison_racct_attach(pr);
2166 #endif
2167 
2168 	/* Locks may have prevented a complete restriction of child IP
2169 	 * addresses.  If so, allocate some more memory and try again.
2170 	 */
2171 #ifdef INET
2172 	while (redo_ip4) {
2173 		ip4s = pr->pr_addrs[PR_INET]->ips;
2174 		MPASS(ip4 == NULL);
2175 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2176 		mtx_lock(&pr->pr_mtx);
2177 		redo_ip4 = false;
2178 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2179 #ifdef VIMAGE
2180 			if (tpr->pr_flags & PR_VNET) {
2181 				descend = 0;
2182 				continue;
2183 			}
2184 #endif
2185 			if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2186 				redo_ip4 = true;
2187 		}
2188 		mtx_unlock(&pr->pr_mtx);
2189 	}
2190 #endif
2191 #ifdef INET6
2192 	while (redo_ip6) {
2193 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2194 		MPASS(ip6 == NULL);
2195 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2196 		mtx_lock(&pr->pr_mtx);
2197 		redo_ip6 = false;
2198 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2199 #ifdef VIMAGE
2200 			if (tpr->pr_flags & PR_VNET) {
2201 				descend = 0;
2202 				continue;
2203 			}
2204 #endif
2205 			if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2206 				redo_ip6 = true;
2207 		}
2208 		mtx_unlock(&pr->pr_mtx);
2209 	}
2210 #endif
2211 
2212 	/* Let the modules do their work. */
2213 	if (created) {
2214 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2215 		if (error)
2216 			goto done_deref;
2217 	}
2218 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2219 	if (error)
2220 		goto done_deref;
2221 
2222 	/*
2223 	 * A new prison is now ready to be seen; either it has gained a user
2224 	 * reference via persistence, or is about to gain one via attachment.
2225 	 */
2226 	if (created) {
2227 		sx_assert(&allprison_lock, SX_XLOCKED);
2228 		prison_knote(ppr, NOTE_JAIL_CHILD | pr->pr_id);
2229 		mtx_lock(&pr->pr_mtx);
2230 		drflags |= PD_LOCKED;
2231 		pr->pr_state = PRISON_STATE_ALIVE;
2232 	}
2233 
2234 	/* Attach this process to the prison if requested. */
2235 	if (flags & JAIL_ATTACH) {
2236 		error = do_jail_attach(td, pr,
2237 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2238 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2239 		if (error) {
2240 			vfs_opterror(opts, "attach failed");
2241 			goto done_deref;
2242 		}
2243 	}
2244 
2245 #ifdef RACCT
2246 	if (racct_enable && !created) {
2247 		if (drflags & PD_LOCKED) {
2248 			mtx_unlock(&pr->pr_mtx);
2249 			drflags &= ~PD_LOCKED;
2250 		}
2251 		if (drflags & PD_LIST_XLOCKED) {
2252 			sx_xunlock(&allprison_lock);
2253 			drflags &= ~PD_LIST_XLOCKED;
2254 		}
2255 		prison_racct_modify(pr);
2256 	}
2257 #endif
2258 
2259 	if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2260 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2261 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2262 		   " file system\n", pr->pr_id);
2263 
2264 	/*
2265 	 * Now that the prison is fully created without error, set the
2266 	 * jail descriptor if one was requested.  This is the only
2267 	 * parameter that is returned to the caller (except the error
2268 	 * message).
2269 	 */
2270 	if (jfd_out >= 0) {
2271 		if (!(drflags & PD_LOCKED)) {
2272 			mtx_lock(&pr->pr_mtx);
2273 			drflags |= PD_LOCKED;
2274 		}
2275 		jfd_pos = 2 * vfs_getopt_pos(opts, "desc") + 1;
2276 		if (optuio->uio_segflg == UIO_SYSSPACE)
2277 			*(int*)optuio->uio_iov[jfd_pos].iov_base = jfd_out;
2278 		else
2279 			(void)copyout(&jfd_out,
2280 			    optuio->uio_iov[jfd_pos].iov_base, sizeof(jfd_out));
2281 		jaildesc_set_prison(jfp_out, pr);
2282 	}
2283 
2284 	drflags &= ~PD_KILL;
2285 	td->td_retval[0] = pr->pr_id;
2286 
2287  done_deref:
2288 	/*
2289 	 * Report changes to kevent.  This can happen even if the
2290 	 * system call fails, as changes might have been made before
2291 	 * the failure.
2292 	 */
2293 	if (maybe_changed && !created)
2294 		prison_knote(pr, NOTE_JAIL_SET);
2295 	/* Release any temporary prison holds and/or locks. */
2296 	if (pr != NULL)
2297 		prison_deref(pr, drflags);
2298 	else if (drflags & PD_LIST_SLOCKED)
2299 		sx_sunlock(&allprison_lock);
2300 	else if (drflags & PD_LIST_XLOCKED)
2301 		sx_xunlock(&allprison_lock);
2302 	if (root != NULL)
2303 		vrele(root);
2304  done_errmsg:
2305 	if (error) {
2306 		/* Write the error message back to userspace. */
2307 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2308 		    &errmsg_len) == 0 && errmsg_len > 0) {
2309 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2310 			if (optuio->uio_segflg == UIO_SYSSPACE)
2311 				bcopy(errmsg,
2312 				    optuio->uio_iov[errmsg_pos].iov_base,
2313 				    errmsg_len);
2314 			else
2315 				(void)copyout(errmsg,
2316 				    optuio->uio_iov[errmsg_pos].iov_base,
2317 				    errmsg_len);
2318 		}
2319 	}
2320  done_free:
2321 	/* Clean up other resources. */
2322 #ifdef INET
2323 	prison_ip_free(ip4);
2324 #endif
2325 #ifdef INET6
2326 	prison_ip_free(ip6);
2327 #endif
2328 	if (jfp_out != NULL)
2329 		fdrop(jfp_out, td);
2330 	if (error && jfd_out >= 0)
2331 		(void)kern_close(td, jfd_out);
2332 	if (g_path != NULL)
2333 		free(g_path, M_TEMP);
2334 	vfs_freeopts(opts);
2335 	prison_free(mypr);
2336 	return (error);
2337 }
2338 
2339 /*
2340  * Find the next available prison ID.  Return the ID on success, or zero
2341  * on failure.  Also set a pointer to the allprison list entry the prison
2342  * should be inserted before.
2343  */
2344 static int
get_next_prid(struct prison ** insprp)2345 get_next_prid(struct prison **insprp)
2346 {
2347 	struct prison *inspr;
2348 	int jid, maxid;
2349 
2350 	jid = lastprid % JAIL_MAX + 1;
2351 	if (TAILQ_EMPTY(&allprison) ||
2352 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2353 		/*
2354 		 * A common case is for all jails to be implicitly numbered,
2355 		 * which means they'll go on the end of the list, at least
2356 		 * for the first JAIL_MAX times.
2357 		 */
2358 		inspr = NULL;
2359 	} else {
2360 		/*
2361 		 * Take two passes through the allprison list: first starting
2362 		 * with the proposed jid, then ending with it.
2363 		 */
2364 		for (maxid = JAIL_MAX; maxid != 0; ) {
2365 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2366 				if (inspr->pr_id < jid)
2367 					continue;
2368 				if (inspr->pr_id > jid) {
2369 					/* Found an opening. */
2370 					maxid = 0;
2371 					break;
2372 				}
2373 				if (++jid > maxid) {
2374 					if (lastprid == maxid || lastprid == 0)
2375 					{
2376 						/*
2377 						 * The entire legal range
2378 						 * has been traversed
2379 						 */
2380 						return 0;
2381 					}
2382 					/* Try again from the start. */
2383 					jid = 1;
2384 					maxid = lastprid;
2385 					break;
2386 				}
2387 			}
2388 			if (inspr == NULL) {
2389 				/* Found room at the end of the list. */
2390 				break;
2391 			}
2392 		}
2393 	}
2394 	*insprp = inspr;
2395 	lastprid = jid;
2396 	return (jid);
2397 }
2398 
2399 /*
2400  * Find the next available ID for a renumbered dead prison.  This is the same
2401  * as get_next_prid, but counting backward from the end of the range.
2402  */
2403 static int
get_next_deadid(struct prison ** dinsprp)2404 get_next_deadid(struct prison **dinsprp)
2405 {
2406 	struct prison *dinspr;
2407 	int deadid, minid;
2408 
2409 	deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2410 	/*
2411 	 * Take two reverse passes through the allprison list: first
2412 	 * starting with the proposed deadid, then ending with it.
2413 	 */
2414 	for (minid = 1; minid != 0; ) {
2415 		TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2416 			if (dinspr->pr_id > deadid)
2417 				continue;
2418 			if (dinspr->pr_id < deadid) {
2419 				/* Found an opening. */
2420 				minid = 0;
2421 				break;
2422 			}
2423 			if (--deadid < minid) {
2424 				if (lastdeadid == minid || lastdeadid == 0)
2425 				{
2426 					/*
2427 					 * The entire legal range
2428 					 * has been traversed
2429 					 */
2430 					return 0;
2431 				}
2432 				/* Try again from the end. */
2433 				deadid = JAIL_MAX;
2434 				minid = lastdeadid;
2435 				break;
2436 			}
2437 		}
2438 		if (dinspr == NULL) {
2439 			/* Found room at the beginning of the list. */
2440 			break;
2441 		}
2442 	}
2443 	*dinsprp = dinspr;
2444 	lastdeadid = deadid;
2445 	return (deadid);
2446 }
2447 
2448 /*
2449  * struct jail_get_args {
2450  *	struct iovec *iovp;
2451  *	unsigned int iovcnt;
2452  *	int flags;
2453  * };
2454  */
2455 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)2456 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2457 {
2458 	struct uio *auio;
2459 	int error;
2460 
2461 	/* Check that we have an even number of iovecs. */
2462 	if (uap->iovcnt & 1)
2463 		return (EINVAL);
2464 
2465 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2466 	if (error)
2467 		return (error);
2468 	error = kern_jail_get(td, auio, uap->flags);
2469 	if (error == 0)
2470 		error = copyout(auio->uio_iov, uap->iovp,
2471 		    uap->iovcnt * sizeof(struct iovec));
2472 	freeuio(auio);
2473 	return (error);
2474 }
2475 
2476 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2477 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2478 {
2479 	struct bool_flags *bf;
2480 	struct file *jfp_out;
2481 	struct jailsys_flags *jsf;
2482 	struct prison *pr, *mypr;
2483 	struct vfsopt *opt;
2484 	struct vfsoptlist *opts;
2485 	char *errmsg, *name;
2486 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2487 	int jfd_in, jfd_out;
2488 	unsigned f;
2489 
2490 	if (flags & ~JAIL_GET_MASK)
2491 		return (EINVAL);
2492 	if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
2493 	    (JAIL_USE_DESC | JAIL_AT_DESC))
2494 		return (EINVAL);
2495 
2496 	/* Get the parameter list. */
2497 	error = vfs_buildopts(optuio, &opts);
2498 	if (error)
2499 		return (error);
2500 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2501 	mypr = td->td_ucred->cr_prison;
2502 	prison_hold(mypr);
2503 	pr = NULL;
2504 	jfp_out = NULL;
2505 	jfd_out = -1;
2506 
2507 	/*
2508 	 * Find the prison specified by one of: desc, lastjid, jid, name.
2509 	 */
2510 	sx_slock(&allprison_lock);
2511 	drflags = PD_LIST_SLOCKED;
2512 
2513 	error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
2514 	if (error == ENOENT) {
2515 		if (flags & (JAIL_AT_DESC | JAIL_GET_DESC | JAIL_OWN_DESC)) {
2516 			vfs_opterror(opts, "missing desc");
2517 			goto done;
2518 		}
2519 	} else if (error == 0) {
2520 		if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
2521 		    JAIL_OWN_DESC))) {
2522 			error = EINVAL;
2523 			vfs_opterror(opts, "unexpected desc");
2524 			goto done;
2525 		}
2526 		if (flags & JAIL_USE_DESC) {
2527 			/* Get the jail from its descriptor. */
2528 			error = jaildesc_find(td, jfd_in, &pr, NULL);
2529 			if (error) {
2530 				vfs_opterror(opts, error == ENOENT ?
2531 				    "descriptor to dead jail" :
2532 				    "not a jail descriptor");
2533 				goto done;
2534 			}
2535 			drflags |= PD_DEREF;
2536 			mtx_lock(&pr->pr_mtx);
2537 			drflags |= PD_LOCKED;
2538 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2539 				error = ENOENT;
2540 				vfs_opterror(opts, "jail %d is dying",
2541 				    pr->pr_id);
2542 				goto done;
2543 			}
2544 			goto found_prison;
2545 		}
2546 		if (flags & JAIL_AT_DESC) {
2547 			/* Look up jails based on the descriptor's prison. */
2548 			prison_free(mypr);
2549 			error = jaildesc_find(td, jfd_in, &mypr, NULL);
2550 			if (error != 0) {
2551 				vfs_opterror(opts, error == ENOENT ?
2552 				    "descriptor to dead jail" :
2553 				    "not a jail descriptor");
2554 				goto done;
2555 			}
2556 		}
2557 		if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
2558 			/* Allocate a jail descriptor to return later. */
2559 			error = jaildesc_alloc(td, &jfp_out, &jfd_out,
2560 				flags & JAIL_OWN_DESC);
2561 			if (error)
2562 				goto done;
2563 		}
2564 	} else
2565 		goto done;
2566 
2567 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2568 	if (error == 0) {
2569 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2570 			if (pr->pr_id > jid &&
2571 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2572 			    prison_ischild(mypr, pr)) {
2573 				mtx_lock(&pr->pr_mtx);
2574 				drflags |= PD_LOCKED;
2575 				goto found_prison;
2576 			}
2577 		}
2578 		error = ENOENT;
2579 		vfs_opterror(opts, "no jail after %d", jid);
2580 		goto done;
2581 	} else if (error != ENOENT)
2582 		goto done;
2583 
2584 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2585 	if (error == 0) {
2586 		if (jid != 0) {
2587 			pr = prison_find_child(mypr, jid);
2588 			if (pr != NULL) {
2589 				drflags |= PD_LOCKED;
2590 				if (!(prison_isalive(pr) ||
2591 				    (flags & JAIL_DYING))) {
2592 					error = ENOENT;
2593 					vfs_opterror(opts, "jail %d is dying",
2594 					    jid);
2595 					goto done;
2596 				}
2597 				goto found_prison;
2598 			}
2599 			error = ENOENT;
2600 			vfs_opterror(opts, "jail %d not found", jid);
2601 			goto done;
2602 		}
2603 	} else if (error != ENOENT)
2604 		goto done;
2605 
2606 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2607 	if (error == 0) {
2608 		if (len == 0 || name[len - 1] != '\0') {
2609 			error = EINVAL;
2610 			goto done;
2611 		}
2612 		pr = prison_find_name(mypr, name);
2613 		if (pr != NULL) {
2614 			drflags |= PD_LOCKED;
2615 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2616 				error = ENOENT;
2617 				vfs_opterror(opts, "jail \"%s\" is dying",
2618 				    name);
2619 				goto done;
2620 			}
2621 			goto found_prison;
2622 		}
2623 		error = ENOENT;
2624 		vfs_opterror(opts, "jail \"%s\" not found", name);
2625 		goto done;
2626 	} else if (error != ENOENT)
2627 		goto done;
2628 
2629 	vfs_opterror(opts, "no jail specified");
2630 	error = ENOENT;
2631 	goto done;
2632 
2633  found_prison:
2634 	/* Get the parameters of the prison. */
2635 	if (!(drflags & PD_DEREF)) {
2636 		prison_hold(pr);
2637 		drflags |= PD_DEREF;
2638 	}
2639 	td->td_retval[0] = pr->pr_id;
2640 	if (jfd_out >= 0) {
2641 		error = vfs_setopt(opts, "desc", &jfd_out, sizeof(jfd_out));
2642 		if (error != 0 && error != ENOENT)
2643 			goto done;
2644 		jaildesc_set_prison(jfp_out, pr);
2645 	}
2646 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2647 	if (error != 0 && error != ENOENT)
2648 		goto done;
2649 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2650 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2651 	if (error != 0 && error != ENOENT)
2652 		goto done;
2653 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2654 	if (error != 0 && error != ENOENT)
2655 		goto done;
2656 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2657 	    sizeof(pr->pr_cpuset->cs_id));
2658 	if (error != 0 && error != ENOENT)
2659 		goto done;
2660 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2661 	if (error != 0 && error != ENOENT)
2662 		goto done;
2663 #ifdef INET
2664 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2665 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2666 	    pr_families[PR_INET].size : 0 );
2667 	if (error != 0 && error != ENOENT)
2668 		goto done;
2669 #endif
2670 #ifdef INET6
2671 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2672 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2673 	    pr_families[PR_INET6].size : 0 );
2674 	if (error != 0 && error != ENOENT)
2675 		goto done;
2676 #endif
2677 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2678 	    sizeof(pr->pr_securelevel));
2679 	if (error != 0 && error != ENOENT)
2680 		goto done;
2681 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2682 	    sizeof(pr->pr_childcount));
2683 	if (error != 0 && error != ENOENT)
2684 		goto done;
2685 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2686 	    sizeof(pr->pr_childmax));
2687 	if (error != 0 && error != ENOENT)
2688 		goto done;
2689 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2690 	if (error != 0 && error != ENOENT)
2691 		goto done;
2692 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2693 	if (error != 0 && error != ENOENT)
2694 		goto done;
2695 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2696 	if (error != 0 && error != ENOENT)
2697 		goto done;
2698 #ifdef COMPAT_FREEBSD32
2699 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2700 		uint32_t hid32 = pr->pr_hostid;
2701 
2702 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2703 	} else
2704 #endif
2705 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2706 	    sizeof(pr->pr_hostid));
2707 	if (error != 0 && error != ENOENT)
2708 		goto done;
2709 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2710 	    sizeof(pr->pr_enforce_statfs));
2711 	if (error != 0 && error != ENOENT)
2712 		goto done;
2713 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2714 	    sizeof(pr->pr_devfs_rsnum));
2715 	if (error != 0 && error != ENOENT)
2716 		goto done;
2717 	for (bf = pr_flag_bool;
2718 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2719 	     bf++) {
2720 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2721 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2722 		if (error != 0 && error != ENOENT)
2723 			goto done;
2724 		i = !i;
2725 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2726 		if (error != 0 && error != ENOENT)
2727 			goto done;
2728 	}
2729 	for (jsf = pr_flag_jailsys;
2730 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2731 	     jsf++) {
2732 		f = pr->pr_flags & (jsf->disable | jsf->new);
2733 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2734 		    : (f == jsf->new) ? JAIL_SYS_NEW
2735 		    : JAIL_SYS_INHERIT;
2736 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2737 		if (error != 0 && error != ENOENT)
2738 			goto done;
2739 	}
2740 	for (bf = pr_flag_allow;
2741 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2742 		atomic_load_int(&bf->flag) != 0;
2743 	     bf++) {
2744 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2745 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2746 		if (error != 0 && error != ENOENT)
2747 			goto done;
2748 		i = !i;
2749 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2750 		if (error != 0 && error != ENOENT)
2751 			goto done;
2752 	}
2753 	i = !prison_isalive(pr);
2754 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2755 	if (error != 0 && error != ENOENT)
2756 		goto done;
2757 	i = !i;
2758 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2759 	if (error != 0 && error != ENOENT)
2760 		goto done;
2761 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2762 	    sizeof(pr->pr_osreldate));
2763 	if (error != 0 && error != ENOENT)
2764 		goto done;
2765 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2766 	if (error != 0 && error != ENOENT)
2767 		goto done;
2768 
2769 	/* Get the module parameters. */
2770 	mtx_unlock(&pr->pr_mtx);
2771 	drflags &= ~PD_LOCKED;
2772 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2773 	if (error)
2774 		goto done;
2775 	prison_deref(pr, drflags);
2776 	pr = NULL;
2777 	drflags = 0;
2778 
2779 	/* By now, all parameters should have been noted. */
2780 	TAILQ_FOREACH(opt, opts, link) {
2781 		if (!opt->seen &&
2782 		    (strstr(opt->name, JAIL_META_PRIVATE ".") == opt->name ||
2783 		    strstr(opt->name, JAIL_META_SHARED ".") == opt->name)) {
2784 			/* Communicate back a missing key. */
2785 			free(opt->value, M_MOUNT);
2786 			opt->value = NULL;
2787 			opt->len = 0;
2788 			continue;
2789 		}
2790 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2791 			error = EINVAL;
2792 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2793 			goto done;
2794 		}
2795 	}
2796 
2797 	/* Write the fetched parameters back to userspace. */
2798 	error = 0;
2799 	TAILQ_FOREACH(opt, opts, link) {
2800 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2801 			pos = 2 * opt->pos + 1;
2802 			optuio->uio_iov[pos].iov_len = opt->len;
2803 			if (opt->value != NULL) {
2804 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2805 					bcopy(opt->value,
2806 					    optuio->uio_iov[pos].iov_base,
2807 					    opt->len);
2808 				} else {
2809 					error = copyout(opt->value,
2810 					    optuio->uio_iov[pos].iov_base,
2811 					    opt->len);
2812 					if (error)
2813 						break;
2814 				}
2815 			}
2816 		}
2817 	}
2818 
2819  done:
2820 	/* Release any temporary prison holds and/or locks. */
2821 	if (pr != NULL)
2822 		prison_deref(pr, drflags);
2823 	else if (drflags & PD_LIST_SLOCKED)
2824 		sx_sunlock(&allprison_lock);
2825 	else if (drflags & PD_LIST_XLOCKED)
2826 		sx_xunlock(&allprison_lock);
2827 	/* Clean up other resources. */
2828 	if (jfp_out != NULL)
2829 		(void)fdrop(jfp_out, td);
2830 	if (error && jfd_out >= 0)
2831 		(void)kern_close(td, jfd_out);
2832 	if (error && errmsg_pos >= 0) {
2833 		/* Write the error message back to userspace. */
2834 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2835 		errmsg_pos = 2 * errmsg_pos + 1;
2836 		if (errmsg_len > 0) {
2837 			if (optuio->uio_segflg == UIO_SYSSPACE)
2838 				bcopy(errmsg,
2839 				    optuio->uio_iov[errmsg_pos].iov_base,
2840 				    errmsg_len);
2841 			else
2842 				(void)copyout(errmsg,
2843 				    optuio->uio_iov[errmsg_pos].iov_base,
2844 				    errmsg_len);
2845 		}
2846 	}
2847 	vfs_freeopts(opts);
2848 	prison_free(mypr);
2849 	return (error);
2850 }
2851 
2852 /*
2853  * struct jail_remove_args {
2854  *	int jid;
2855  * };
2856  */
2857 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2858 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2859 {
2860 	struct prison *pr;
2861 	int error;
2862 
2863 	error = priv_check(td, PRIV_JAIL_REMOVE);
2864 	if (error)
2865 		return (error);
2866 
2867 	sx_xlock(&allprison_lock);
2868 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2869 	if (pr == NULL) {
2870 		sx_xunlock(&allprison_lock);
2871 		return (EINVAL);
2872 	}
2873 	prison_hold(pr);
2874 	prison_remove(pr);
2875 	return (0);
2876 }
2877 
2878 /*
2879  * struct jail_remove_jd_args {
2880  *	int fd;
2881  * };
2882  */
2883 int
sys_jail_remove_jd(struct thread * td,struct jail_remove_jd_args * uap)2884 sys_jail_remove_jd(struct thread *td, struct jail_remove_jd_args *uap)
2885 {
2886 	struct prison *pr;
2887 	struct ucred *jdcred;
2888 	int error;
2889 
2890 	error = jaildesc_find(td, uap->fd, &pr, &jdcred);
2891 	if (error)
2892 		return (error);
2893 	error = priv_check_cred(jdcred, PRIV_JAIL_REMOVE);
2894 	crfree(jdcred);
2895 	if (error) {
2896 		prison_free(pr);
2897 		return (error);
2898 	}
2899 	sx_xlock(&allprison_lock);
2900 	mtx_lock(&pr->pr_mtx);
2901 	prison_remove(pr);
2902 	return (0);
2903 }
2904 
2905 /*
2906  * Begin the removal process for a prison.  The allprison lock should
2907  * be held exclusively, and the prison should be both locked and held.
2908  */
2909 void
prison_remove(struct prison * pr)2910 prison_remove(struct prison *pr)
2911 {
2912 	sx_assert(&allprison_lock, SA_XLOCKED);
2913 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2914 	prison_deref(pr, PD_KILL | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2915 }
2916 
2917 /*
2918  * struct jail_attach_args {
2919  *	int jid;
2920  * };
2921  */
2922 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)2923 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2924 {
2925 	struct prison *pr;
2926 	int error;
2927 
2928 	error = priv_check(td, PRIV_JAIL_ATTACH);
2929 	if (error)
2930 		return (error);
2931 
2932 	sx_slock(&allprison_lock);
2933 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2934 	if (pr == NULL) {
2935 		sx_sunlock(&allprison_lock);
2936 		return (EINVAL);
2937 	}
2938 
2939 	/* Do not allow a process to attach to a prison that is not alive. */
2940 	if (!prison_isalive(pr)) {
2941 		mtx_unlock(&pr->pr_mtx);
2942 		sx_sunlock(&allprison_lock);
2943 		return (EINVAL);
2944 	}
2945 
2946 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2947 }
2948 
2949 /*
2950  * struct jail_attach_jd_args {
2951  *	int fd;
2952  * };
2953  */
2954 int
sys_jail_attach_jd(struct thread * td,struct jail_attach_jd_args * uap)2955 sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
2956 {
2957 	struct prison *pr;
2958 	struct ucred *jdcred;
2959 	int drflags, error;
2960 
2961 	sx_slock(&allprison_lock);
2962 	drflags = PD_LIST_SLOCKED;
2963 	error = jaildesc_find(td, uap->fd, &pr, &jdcred);
2964 	if (error)
2965 		goto fail;
2966 	drflags |= PD_DEREF;
2967 	error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
2968 	crfree(jdcred);
2969 	if (error)
2970 		goto fail;
2971 	mtx_lock(&pr->pr_mtx);
2972 	drflags |= PD_LOCKED;
2973 
2974 	/* Do not allow a process to attach to a prison that is not alive. */
2975 	if (!prison_isalive(pr)) {
2976 		error = EINVAL;
2977 		goto fail;
2978 	}
2979 
2980 	return (do_jail_attach(td, pr, drflags));
2981 
2982  fail:
2983 	prison_deref(pr, drflags);
2984 	return (error);
2985 }
2986 
2987 static int
do_jail_attach(struct thread * td,struct prison * pr,int drflags)2988 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2989 {
2990 	struct proc *p;
2991 	struct ucred *newcred, *oldcred;
2992 	int error;
2993 
2994 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2995 	sx_assert(&allprison_lock, SX_LOCKED);
2996 	drflags &= PD_LOCK_FLAGS;
2997 	/*
2998 	 * XXX: Note that there is a slight race here if two threads
2999 	 * in the same privileged process attempt to attach to two
3000 	 * different jails at the same time.  It is important for
3001 	 * user processes not to do this, or they might end up with
3002 	 * a process root from one prison, but attached to the jail
3003 	 * of another.
3004 	 */
3005 	if (!(drflags & PD_DEREF)) {
3006 		prison_hold(pr);
3007 		drflags |= PD_DEREF;
3008 	}
3009 	refcount_acquire(&pr->pr_uref);
3010 	drflags |= PD_DEUREF;
3011 	mtx_unlock(&pr->pr_mtx);
3012 	drflags &= ~PD_LOCKED;
3013 
3014 	/* Let modules do whatever they need to prepare for attaching. */
3015 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
3016 	if (error) {
3017 		prison_deref(pr, drflags);
3018 		return (error);
3019 	}
3020 	sx_unlock(&allprison_lock);
3021 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
3022 
3023 	/*
3024 	 * Reparent the newly attached process to this jail.
3025 	 */
3026 	p = td->td_proc;
3027 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
3028 	if (error)
3029 		goto e_revert_osd;
3030 
3031 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
3032 	if ((error = change_dir(pr->pr_root, td)) != 0)
3033 		goto e_unlock;
3034 #ifdef MAC
3035 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
3036 		goto e_unlock;
3037 #endif
3038 	VOP_UNLOCK(pr->pr_root);
3039 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
3040 		goto e_revert_osd;
3041 
3042 	newcred = crget();
3043 	PROC_LOCK(p);
3044 	oldcred = crcopysafe(p, newcred);
3045 	newcred->cr_prison = pr;
3046 	proc_set_cred(p, newcred);
3047 	setsugid(p);
3048 #ifdef RACCT
3049 	racct_proc_ucred_changed(p, oldcred, newcred);
3050 	crhold(newcred);
3051 #endif
3052 	PROC_UNLOCK(p);
3053 #ifdef RCTL
3054 	rctl_proc_ucred_changed(p, newcred);
3055 	crfree(newcred);
3056 #endif
3057 	prison_proc_relink(oldcred->cr_prison, pr, p);
3058 	prison_deref(oldcred->cr_prison, drflags);
3059 	crfree(oldcred);
3060 	prison_knote(pr, NOTE_JAIL_ATTACH | td->td_proc->p_pid);
3061 
3062 	/*
3063 	 * If the prison was killed while changing credentials, die along
3064 	 * with it.
3065 	 */
3066 	if (!prison_isalive(pr)) {
3067 		PROC_LOCK(p);
3068 		kern_psignal(p, SIGKILL);
3069 		PROC_UNLOCK(p);
3070 	}
3071 
3072 	return (0);
3073 
3074  e_unlock:
3075 	VOP_UNLOCK(pr->pr_root);
3076  e_revert_osd:
3077 	/* Tell modules this thread is still in its old jail after all. */
3078 	sx_slock(&allprison_lock);
3079 	drflags |= PD_LIST_SLOCKED;
3080 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
3081 	prison_deref(pr, drflags);
3082 	return (error);
3083 }
3084 
3085 /*
3086  * Returns a locked prison instance, or NULL on failure.
3087  */
3088 struct prison *
prison_find(int prid)3089 prison_find(int prid)
3090 {
3091 	struct prison *pr;
3092 
3093 	sx_assert(&allprison_lock, SX_LOCKED);
3094 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3095 		if (pr->pr_id < prid)
3096 			continue;
3097 		if (pr->pr_id > prid)
3098 			break;
3099 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
3100 		mtx_lock(&pr->pr_mtx);
3101 		return (pr);
3102 	}
3103 	return (NULL);
3104 }
3105 
3106 /*
3107  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
3108  */
3109 struct prison *
prison_find_child(struct prison * mypr,int prid)3110 prison_find_child(struct prison *mypr, int prid)
3111 {
3112 	struct prison *pr;
3113 	int descend;
3114 
3115 	sx_assert(&allprison_lock, SX_LOCKED);
3116 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3117 		if (pr->pr_id == prid) {
3118 			KASSERT(prison_isvalid(pr),
3119 			    ("Found invalid prison %p", pr));
3120 			mtx_lock(&pr->pr_mtx);
3121 			return (pr);
3122 		}
3123 	}
3124 	return (NULL);
3125 }
3126 
3127 /*
3128  * Look for the name relative to mypr.  Returns a locked prison or NULL.
3129  */
3130 struct prison *
prison_find_name(struct prison * mypr,const char * name)3131 prison_find_name(struct prison *mypr, const char *name)
3132 {
3133 	struct prison *pr, *deadpr;
3134 	size_t mylen;
3135 	int descend;
3136 
3137 	sx_assert(&allprison_lock, SX_LOCKED);
3138 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
3139 	deadpr = NULL;
3140 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3141 		if (!strcmp(pr->pr_name + mylen, name)) {
3142 			KASSERT(prison_isvalid(pr),
3143 			    ("Found invalid prison %p", pr));
3144 			if (prison_isalive(pr)) {
3145 				mtx_lock(&pr->pr_mtx);
3146 				return (pr);
3147 			}
3148 			deadpr = pr;
3149 		}
3150 	}
3151 	/* There was no valid prison - perhaps there was a dying one. */
3152 	if (deadpr != NULL)
3153 		mtx_lock(&deadpr->pr_mtx);
3154 	return (deadpr);
3155 }
3156 
3157 /*
3158  * See if a prison has the specific flag set.  The prison should be locked,
3159  * unless checking for flags that are only set at jail creation (such as
3160  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
3161  * to any other prison data.
3162  */
3163 bool
prison_flag(struct ucred * cred,unsigned flag)3164 prison_flag(struct ucred *cred, unsigned flag)
3165 {
3166 
3167 	return ((cred->cr_prison->pr_flags & flag) != 0);
3168 }
3169 
3170 /*
3171  * See if a prison has the specific allow flag set.
3172  * The prison *should* be locked, or only a single bit is examined, without
3173  * regard to any other prison data.
3174  */
3175 bool
prison_allow(struct ucred * cred,unsigned flag)3176 prison_allow(struct ucred *cred, unsigned flag)
3177 {
3178 
3179 	return ((cred->cr_prison->pr_allow & flag) != 0);
3180 }
3181 
3182 /*
3183  * Hold a prison reference, by incrementing pr_ref.  It is generally
3184  * an error to hold a prison that does not already have a reference.
3185  * A prison record will remain valid as long as it has at least one
3186  * reference, and will not be removed as long as either the prison
3187  * mutex or the allprison lock is held (allprison_lock may be shared).
3188  */
3189 void
prison_hold_locked(struct prison * pr)3190 prison_hold_locked(struct prison *pr)
3191 {
3192 
3193 	/* Locking is no longer required. */
3194 	prison_hold(pr);
3195 }
3196 
3197 void
prison_hold(struct prison * pr)3198 prison_hold(struct prison *pr)
3199 {
3200 #ifdef INVARIANTS
3201 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
3202 
3203 	KASSERT(was_valid,
3204 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
3205 #else
3206 	refcount_acquire(&pr->pr_ref);
3207 #endif
3208 }
3209 
3210 /*
3211  * Remove a prison reference.  If that was the last reference, the
3212  * prison will be removed (at a later time).
3213  */
3214 void
prison_free_locked(struct prison * pr)3215 prison_free_locked(struct prison *pr)
3216 {
3217 
3218 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3219 	/*
3220 	 * Locking is no longer required, but unlock because the caller
3221 	 * expects it.
3222 	 */
3223 	mtx_unlock(&pr->pr_mtx);
3224 	prison_free(pr);
3225 }
3226 
3227 void
prison_free(struct prison * pr)3228 prison_free(struct prison *pr)
3229 {
3230 
3231 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3232 	    ("Trying to free dead prison %p (jid=%d).",
3233 	     pr, pr->pr_id));
3234 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
3235 		/*
3236 		 * Don't remove the last reference in this context,
3237 		 * in case there are locks held.
3238 		 */
3239 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3240 	}
3241 }
3242 
3243 static void
prison_free_not_last(struct prison * pr)3244 prison_free_not_last(struct prison *pr)
3245 {
3246 #ifdef INVARIANTS
3247 	int lastref;
3248 
3249 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3250 	    ("Trying to free dead prison %p (jid=%d).",
3251 	     pr, pr->pr_id));
3252 	lastref = refcount_release(&pr->pr_ref);
3253 	KASSERT(!lastref,
3254 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
3255 	     pr, pr->pr_id));
3256 #else
3257 	refcount_release(&pr->pr_ref);
3258 #endif
3259 }
3260 
3261 /*
3262  * Hold a prison for user visibility, by incrementing pr_uref.
3263  * It is generally an error to hold a prison that isn't already
3264  * user-visible, except through the jail system calls.  It is also
3265  * an error to hold an invalid prison.  A prison record will remain
3266  * alive as long as it has at least one user reference, and will not
3267  * be set to the dying state until the prison mutex and allprison_lock
3268  * are both freed.
3269  */
3270 void
prison_proc_hold(struct prison * pr)3271 prison_proc_hold(struct prison *pr)
3272 {
3273 #ifdef INVARIANTS
3274 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
3275 
3276 	KASSERT(was_alive,
3277 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
3278 #else
3279 	refcount_acquire(&pr->pr_uref);
3280 #endif
3281 }
3282 
3283 /*
3284  * Remove a prison user reference.  If it was the last reference, the
3285  * prison will be considered "dying", and may be removed once all of
3286  * its references are dropped.
3287  */
3288 void
prison_proc_free(struct prison * pr)3289 prison_proc_free(struct prison *pr)
3290 {
3291 
3292 	/*
3293 	 * Locking is only required when releasing the last reference.
3294 	 * This allows assurance that a locked prison will remain alive
3295 	 * until it is unlocked.
3296 	 */
3297 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3298 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
3299 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
3300 		/*
3301 		 * Don't remove the last user reference in this context,
3302 		 * which is expected to be a process that is not only locked,
3303 		 * but also half dead.  Add a reference so any calls to
3304 		 * prison_free() won't re-submit the task.
3305 		 */
3306 		prison_hold(pr);
3307 		mtx_lock(&pr->pr_mtx);
3308 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
3309 		    ("Redundant last reference in prison_proc_free (jid=%d)",
3310 		     pr->pr_id));
3311 		pr->pr_flags |= PR_COMPLETE_PROC;
3312 		mtx_unlock(&pr->pr_mtx);
3313 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3314 	}
3315 }
3316 
3317 static void
prison_proc_free_not_last(struct prison * pr)3318 prison_proc_free_not_last(struct prison *pr)
3319 {
3320 #ifdef INVARIANTS
3321 	int lastref;
3322 
3323 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3324 	    ("Trying to free dead prison %p (jid=%d).",
3325 	     pr, pr->pr_id));
3326 	lastref = refcount_release(&pr->pr_uref);
3327 	KASSERT(!lastref,
3328 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3329 	     pr, pr->pr_id));
3330 #else
3331 	refcount_release(&pr->pr_uref);
3332 #endif
3333 }
3334 
3335 void
prison_proc_link(struct prison * pr,struct proc * p)3336 prison_proc_link(struct prison *pr, struct proc *p)
3337 {
3338 
3339 	sx_assert(&allproc_lock, SA_XLOCKED);
3340 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3341 }
3342 
3343 void
prison_proc_unlink(struct prison * pr,struct proc * p)3344 prison_proc_unlink(struct prison *pr, struct proc *p)
3345 {
3346 
3347 	sx_assert(&allproc_lock, SA_XLOCKED);
3348 	LIST_REMOVE(p, p_jaillist);
3349 }
3350 
3351 static void
prison_proc_relink(struct prison * opr,struct prison * npr,struct proc * p)3352 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3353 {
3354 
3355 	sx_xlock(&allproc_lock);
3356 	prison_proc_unlink(opr, p);
3357 	prison_proc_link(npr, p);
3358 	sx_xunlock(&allproc_lock);
3359 }
3360 
3361 /*
3362  * Complete a call to either prison_free or prison_proc_free.
3363  */
3364 static void
prison_complete(void * context,int pending)3365 prison_complete(void *context, int pending)
3366 {
3367 	struct prison *pr = context;
3368 	int drflags;
3369 
3370 	/*
3371 	 * This could be called to release the last reference, or the last
3372 	 * user reference (plus the reference held in prison_proc_free).
3373 	 */
3374 	drflags = prison_lock_xlock(pr, PD_DEREF);
3375 	if (pr->pr_flags & PR_COMPLETE_PROC) {
3376 		pr->pr_flags &= ~PR_COMPLETE_PROC;
3377 		drflags |= PD_DEUREF;
3378 	}
3379 	prison_deref(pr, drflags);
3380 }
3381 
3382 static void
prison_kill_processes_cb(struct proc * p,void * arg __unused)3383 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3384 {
3385 
3386 	kern_psignal(p, SIGKILL);
3387 }
3388 
3389 /*
3390  * Note the iteration does not guarantee acting on all processes.
3391  * Most notably there may be fork or jail_attach in progress.
3392  */
3393 void
prison_proc_iterate(struct prison * pr,void (* cb)(struct proc *,void *),void * cbarg)3394 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3395     void *cbarg)
3396 {
3397 	struct prison *ppr;
3398 	struct proc *p;
3399 
3400 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3401 		sx_slock(&allproc_lock);
3402 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3403 			if (p->p_state == PRS_NEW)
3404 				continue;
3405 			PROC_LOCK(p);
3406 			cb(p, cbarg);
3407 			PROC_UNLOCK(p);
3408 		}
3409 		sx_sunlock(&allproc_lock);
3410 		if (atomic_load_int(&pr->pr_childcount) == 0)
3411 			return;
3412 		/*
3413 		 * Some jails popped up during the iteration, fall through to a
3414 		 * system-wide search.
3415 		 */
3416 	}
3417 
3418 	sx_slock(&allproc_lock);
3419 	FOREACH_PROC_IN_SYSTEM(p) {
3420 		PROC_LOCK(p);
3421 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3422 			for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3423 			    ppr = ppr->pr_parent) {
3424 				if (ppr == pr) {
3425 					cb(p, cbarg);
3426 					break;
3427 				}
3428 			}
3429 		}
3430 		PROC_UNLOCK(p);
3431 	}
3432 	sx_sunlock(&allproc_lock);
3433 }
3434 
3435 /*
3436  * Remove a prison reference and/or user reference (usually).
3437  * This assumes context that allows sleeping (for allprison_lock),
3438  * with no non-sleeping locks held, except perhaps the prison itself.
3439  * If there are no more references, release and delist the prison.
3440  * On completion, the prison lock and the allprison lock are both
3441  * unlocked.
3442  */
3443 static void
prison_deref(struct prison * pr,int flags)3444 prison_deref(struct prison *pr, int flags)
3445 {
3446 	struct prisonlist freeprison;
3447 	struct prison *killpr, *rpr, *ppr, *tpr;
3448 
3449 	killpr = NULL;
3450 	TAILQ_INIT(&freeprison);
3451 	/*
3452 	 * Release this prison as requested, which may cause its parent
3453 	 * to be released, and then maybe its grandparent, etc.
3454 	 */
3455 	for (;;) {
3456 		if (flags & PD_KILL) {
3457 			/* Kill the prison and its descendents. */
3458 			KASSERT(pr != &prison0,
3459 			    ("prison_deref trying to kill prison0"));
3460 			if (!prison_isalive(pr)) {
3461 				/* Silently ignore already-dying prisons. */
3462 				flags &= ~PD_KILL;
3463 			} else {
3464 				if (!(flags & PD_DEREF)) {
3465 					prison_hold(pr);
3466 					flags |= PD_DEREF;
3467 				}
3468 				flags = prison_lock_xlock(pr, flags);
3469 				prison_deref_kill(pr, &freeprison);
3470 			}
3471 		}
3472 		if (flags & PD_DEUREF) {
3473 			/* Drop a user reference. */
3474 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3475 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3476 			     pr->pr_id));
3477 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3478 				if (!(flags & PD_DEREF)) {
3479 					prison_hold(pr);
3480 					flags |= PD_DEREF;
3481 				}
3482 				flags = prison_lock_xlock(pr, flags);
3483 				if (refcount_release(&pr->pr_uref) &&
3484 				    pr->pr_state == PRISON_STATE_ALIVE) {
3485 					/*
3486 					 * When the last user references goes,
3487 					 * this becomes a dying prison.
3488 					 */
3489 					KASSERT(
3490 					    refcount_load(&prison0.pr_uref) > 0,
3491 					    ("prison0 pr_uref=0"));
3492 					pr->pr_state = PRISON_STATE_DYING;
3493 					prison_cleanup_locked(pr);
3494 					mtx_unlock(&pr->pr_mtx);
3495 					flags &= ~PD_LOCKED;
3496 					prison_cleanup_unlocked(pr);
3497 				}
3498 			}
3499 		}
3500 		if (flags & PD_KILL) {
3501 			/*
3502 			 * Any remaining user references are probably processes
3503 			 * that need to be killed, either in this prison or its
3504 			 * descendants.
3505 			 */
3506 			if (refcount_load(&pr->pr_uref) > 0)
3507 				killpr = pr;
3508 			/* Make sure the parent prison doesn't get killed. */
3509 			flags &= ~PD_KILL;
3510 		}
3511 		if (flags & PD_DEREF) {
3512 			/* Drop a reference. */
3513 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3514 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3515 			     pr->pr_id));
3516 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3517 				flags = prison_lock_xlock(pr, flags);
3518 				if (refcount_release(&pr->pr_ref)) {
3519 					/*
3520 					 * When the last reference goes,
3521 					 * unlink the prison and set it aside.
3522 					 */
3523 					KASSERT(
3524 					    refcount_load(&pr->pr_uref) == 0,
3525 					    ("prison_deref: last ref, "
3526 					     "but still has %d urefs (jid=%d)",
3527 					     pr->pr_uref, pr->pr_id));
3528 					KASSERT(
3529 					    refcount_load(&prison0.pr_ref) != 0,
3530 					    ("prison0 pr_ref=0"));
3531 					pr->pr_state = PRISON_STATE_INVALID;
3532 					TAILQ_REMOVE(&allprison, pr, pr_list);
3533 					LIST_REMOVE(pr, pr_sibling);
3534 					TAILQ_INSERT_TAIL(&freeprison, pr,
3535 					    pr_list);
3536 					for (ppr = pr->pr_parent;
3537 					     ppr != NULL;
3538 					     ppr = ppr->pr_parent)
3539 						ppr->pr_childcount--;
3540 					/*
3541 					 * Removing a prison frees references
3542 					 * from its parent.
3543 					 */
3544 					ppr = pr->pr_parent;
3545 					pr->pr_parent = NULL;
3546 					mtx_unlock(&pr->pr_mtx);
3547 
3548 					pr = ppr;
3549 					flags &= ~PD_LOCKED;
3550 					flags |= PD_DEREF | PD_DEUREF;
3551 					continue;
3552 				}
3553 			}
3554 		}
3555 		break;
3556 	}
3557 
3558 	/* Release all the prison locks. */
3559 	if (flags & PD_LOCKED)
3560 		mtx_unlock(&pr->pr_mtx);
3561 	if (flags & PD_LIST_SLOCKED)
3562 		sx_sunlock(&allprison_lock);
3563 	else if (flags & PD_LIST_XLOCKED)
3564 		sx_xunlock(&allprison_lock);
3565 
3566 	/* Kill any processes attached to a killed prison. */
3567 	if (killpr != NULL)
3568 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3569 
3570 	/*
3571 	 * Finish removing any unreferenced prisons, which couldn't happen
3572 	 * while allprison_lock was held (to avoid a LOR on vrele).
3573 	 */
3574 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3575 #ifdef VIMAGE
3576 		if (rpr->pr_flags & PR_VNET)
3577 			vnet_destroy(rpr->pr_vnet);
3578 #endif
3579 		if (rpr->pr_root != NULL)
3580 			vrele(rpr->pr_root);
3581 		mtx_destroy(&rpr->pr_mtx);
3582 #ifdef INET
3583 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3584 #endif
3585 #ifdef INET6
3586 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3587 #endif
3588 		if (rpr->pr_cpuset != NULL)
3589 			cpuset_rel(rpr->pr_cpuset);
3590 		osd_jail_exit(rpr);
3591 #ifdef RACCT
3592 		if (racct_enable)
3593 			prison_racct_detach(rpr);
3594 #endif
3595 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3596 		free(rpr, M_PRISON);
3597 	}
3598 }
3599 
3600 /*
3601  * Kill the prison and its descendants.  Mark them as dying, clear the
3602  * persist flag, and call module remove methods.
3603  */
3604 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)3605 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3606 {
3607 	struct prison *cpr, *ppr, *rpr;
3608 	bool descend;
3609 
3610 	/*
3611 	 * Unlike the descendants, the target prison can be killed
3612 	 * even if it is currently dying.  This is useful for failed
3613 	 * creation in jail_set(2).
3614 	 */
3615 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3616 	    ("Trying to kill dead prison %p (jid=%d).",
3617 	     pr, pr->pr_id));
3618 	refcount_acquire(&pr->pr_uref);
3619 	pr->pr_state = PRISON_STATE_DYING;
3620 	mtx_unlock(&pr->pr_mtx);
3621 
3622 	rpr = NULL;
3623 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3624 		if (descend) {
3625 			if (!prison_isalive(cpr)) {
3626 				descend = false;
3627 				continue;
3628 			}
3629 			prison_hold(cpr);
3630 			prison_proc_hold(cpr);
3631 			mtx_lock(&cpr->pr_mtx);
3632 			cpr->pr_state = PRISON_STATE_DYING;
3633 			cpr->pr_flags |= PR_REMOVE;
3634 			mtx_unlock(&cpr->pr_mtx);
3635 			continue;
3636 		}
3637 		if (!(cpr->pr_flags & PR_REMOVE))
3638 			continue;
3639 		prison_cleanup_unlocked(cpr);
3640 		mtx_lock(&cpr->pr_mtx);
3641 		prison_cleanup_locked(cpr);
3642 		cpr->pr_flags &= ~PR_REMOVE;
3643 		if (cpr->pr_flags & PR_PERSIST) {
3644 			cpr->pr_flags &= ~PR_PERSIST;
3645 			prison_proc_free_not_last(cpr);
3646 			prison_free_not_last(cpr);
3647 		}
3648 		(void)refcount_release(&cpr->pr_uref);
3649 		if (refcount_release(&cpr->pr_ref)) {
3650 			/*
3651 			 * When the last reference goes, unlink the prison
3652 			 * and set it aside for prison_deref() to handle.
3653 			 * Delay unlinking the sibling list to keep the loop
3654 			 * safe.
3655 			 */
3656 			if (rpr != NULL)
3657 				LIST_REMOVE(rpr, pr_sibling);
3658 			rpr = cpr;
3659 			rpr->pr_state = PRISON_STATE_INVALID;
3660 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3661 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3662 			/*
3663 			 * Removing a prison frees references from its parent.
3664 			 */
3665 			ppr = rpr->pr_parent;
3666 			prison_proc_free_not_last(ppr);
3667 			prison_free_not_last(ppr);
3668 			for (; ppr != NULL; ppr = ppr->pr_parent)
3669 				ppr->pr_childcount--;
3670 		}
3671 		mtx_unlock(&cpr->pr_mtx);
3672 	}
3673 	if (rpr != NULL)
3674 		LIST_REMOVE(rpr, pr_sibling);
3675 
3676 	prison_cleanup_unlocked(pr);
3677 	mtx_lock(&pr->pr_mtx);
3678 	prison_cleanup_locked(pr);
3679 	if (pr->pr_flags & PR_PERSIST) {
3680 		pr->pr_flags &= ~PR_PERSIST;
3681 		prison_proc_free_not_last(pr);
3682 		prison_free_not_last(pr);
3683 	}
3684 	(void)refcount_release(&pr->pr_uref);
3685 }
3686 
3687 /*
3688  * Given the current locking state in the flags, make sure allprison_lock
3689  * is held exclusive, and the prison is locked.  Return flags indicating
3690  * the new state.
3691  */
3692 static int
prison_lock_xlock(struct prison * pr,int flags)3693 prison_lock_xlock(struct prison *pr, int flags)
3694 {
3695 
3696 	if (!(flags & PD_LIST_XLOCKED)) {
3697 		/*
3698 		 * Get allprison_lock, which may be an upgrade,
3699 		 * and may require unlocking the prison.
3700 		 */
3701 		if (flags & PD_LOCKED) {
3702 			mtx_unlock(&pr->pr_mtx);
3703 			flags &= ~PD_LOCKED;
3704 		}
3705 		if (flags & PD_LIST_SLOCKED) {
3706 			if (!sx_try_upgrade(&allprison_lock)) {
3707 				sx_sunlock(&allprison_lock);
3708 				sx_xlock(&allprison_lock);
3709 			}
3710 			flags &= ~PD_LIST_SLOCKED;
3711 		} else
3712 			sx_xlock(&allprison_lock);
3713 		flags |= PD_LIST_XLOCKED;
3714 	}
3715 	if (!(flags & PD_LOCKED)) {
3716 		/* Lock the prison mutex. */
3717 		mtx_lock(&pr->pr_mtx);
3718 		flags |= PD_LOCKED;
3719 	}
3720 	return flags;
3721 }
3722 
3723 /*
3724  * Release a prison's resources when it starts dying (when the last user
3725  * reference is dropped, or when it is killed).  Two functions are called,
3726  * for work that requires a locked prison or an unlocked one.
3727  */
3728 static void
prison_cleanup_locked(struct prison * pr)3729 prison_cleanup_locked(struct prison *pr)
3730 {
3731 	sx_assert(&allprison_lock, SA_XLOCKED);
3732 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3733 	prison_knote(pr, NOTE_JAIL_REMOVE);
3734 	knlist_detach(pr->pr_klist);
3735 	jaildesc_prison_cleanup(pr);
3736 	pr->pr_klist = NULL;
3737 }
3738 
3739 static void
prison_cleanup_unlocked(struct prison * pr)3740 prison_cleanup_unlocked(struct prison *pr)
3741 {
3742 	sx_assert(&allprison_lock, SA_XLOCKED);
3743 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3744 	vfs_exjail_delete(pr);
3745 	shm_remove_prison(pr);
3746 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3747 }
3748 
3749 /*
3750  * Set or clear a permission bit in the pr_allow field, passing restrictions
3751  * (cleared permission) down to child jails.
3752  */
3753 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)3754 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3755 {
3756 	struct prison *pr;
3757 
3758 	pr = cred->cr_prison;
3759 	sx_slock(&allprison_lock);
3760 	mtx_lock(&pr->pr_mtx);
3761 	prison_set_allow_locked(pr, flag, enable);
3762 	mtx_unlock(&pr->pr_mtx);
3763 	sx_sunlock(&allprison_lock);
3764 }
3765 
3766 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)3767 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3768 {
3769 	struct prison *cpr;
3770 	int descend;
3771 
3772 	if (enable != 0)
3773 		pr->pr_allow |= flag;
3774 	else {
3775 		pr->pr_allow &= ~flag;
3776 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3777 			cpr->pr_allow &= ~flag;
3778 	}
3779 }
3780 
3781 /*
3782  * Check if a jail supports the given address family.
3783  *
3784  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3785  * if not.
3786  */
3787 int
prison_check_af(struct ucred * cred,int af)3788 prison_check_af(struct ucred *cred, int af)
3789 {
3790 	struct prison *pr;
3791 	int error;
3792 
3793 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3794 
3795 	pr = cred->cr_prison;
3796 #ifdef VIMAGE
3797 	/* Prisons with their own network stack are not limited. */
3798 	if (prison_owns_vnet(pr))
3799 		return (0);
3800 #endif
3801 
3802 	error = 0;
3803 	switch (af)
3804 	{
3805 #ifdef INET
3806 	case AF_INET:
3807 		if (pr->pr_flags & PR_IP4)
3808 		{
3809 			mtx_lock(&pr->pr_mtx);
3810 			if ((pr->pr_flags & PR_IP4) &&
3811 			    pr->pr_addrs[PR_INET] == NULL)
3812 				error = EAFNOSUPPORT;
3813 			mtx_unlock(&pr->pr_mtx);
3814 		}
3815 		break;
3816 #endif
3817 #ifdef INET6
3818 	case AF_INET6:
3819 		if (pr->pr_flags & PR_IP6)
3820 		{
3821 			mtx_lock(&pr->pr_mtx);
3822 			if ((pr->pr_flags & PR_IP6) &&
3823 			    pr->pr_addrs[PR_INET6] == NULL)
3824 				error = EAFNOSUPPORT;
3825 			mtx_unlock(&pr->pr_mtx);
3826 		}
3827 		break;
3828 #endif
3829 	case AF_LOCAL:
3830 	case AF_ROUTE:
3831 	case AF_NETLINK:
3832 		break;
3833 	default:
3834 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3835 			error = EAFNOSUPPORT;
3836 	}
3837 	return (error);
3838 }
3839 
3840 /*
3841  * Check if given address belongs to the jail referenced by cred (wrapper to
3842  * prison_check_ip[46]).
3843  *
3844  * Returns 0 if jail doesn't restrict the address family or if address belongs
3845  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3846  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3847  */
3848 int
prison_if(struct ucred * cred,const struct sockaddr * sa)3849 prison_if(struct ucred *cred, const struct sockaddr *sa)
3850 {
3851 #ifdef INET
3852 	const struct sockaddr_in *sai;
3853 #endif
3854 #ifdef INET6
3855 	const struct sockaddr_in6 *sai6;
3856 #endif
3857 	int error;
3858 
3859 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3860 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3861 
3862 #ifdef VIMAGE
3863 	if (prison_owns_vnet(cred->cr_prison))
3864 		return (0);
3865 #endif
3866 
3867 	error = 0;
3868 	switch (sa->sa_family)
3869 	{
3870 #ifdef INET
3871 	case AF_INET:
3872 		sai = (const struct sockaddr_in *)sa;
3873 		error = prison_check_ip4(cred, &sai->sin_addr);
3874 		break;
3875 #endif
3876 #ifdef INET6
3877 	case AF_INET6:
3878 		sai6 = (const struct sockaddr_in6 *)sa;
3879 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3880 		break;
3881 #endif
3882 	default:
3883 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3884 			error = EAFNOSUPPORT;
3885 	}
3886 	return (error);
3887 }
3888 
3889 /*
3890  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3891  */
3892 int
prison_check(struct ucred * cred1,struct ucred * cred2)3893 prison_check(struct ucred *cred1, struct ucred *cred2)
3894 {
3895 
3896 	return ((cred1->cr_prison == cred2->cr_prison ||
3897 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3898 }
3899 
3900 /*
3901  * For mountd/nfsd to run within a prison, it must be:
3902  * - A vnet prison.
3903  * - PR_ALLOW_NFSD must be set on it.
3904  * - The root directory (pr_root) of the prison must be
3905  *   a file system mount point, so the mountd can hang
3906  *   export information on it.
3907  * - The prison's enforce_statfs cannot be 0, so that
3908  *   mountd(8) can do exports.
3909  */
3910 bool
prison_check_nfsd(struct ucred * cred)3911 prison_check_nfsd(struct ucred *cred)
3912 {
3913 
3914 	if (jailed_without_vnet(cred))
3915 		return (false);
3916 	if (!prison_allow(cred, PR_ALLOW_NFSD))
3917 		return (false);
3918 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3919 		return (false);
3920 	if (cred->cr_prison->pr_enforce_statfs == 0)
3921 		return (false);
3922 	return (true);
3923 }
3924 
3925 /*
3926  * Return true if p2 is a child of p1, otherwise false.
3927  */
3928 bool
prison_ischild(struct prison * pr1,struct prison * pr2)3929 prison_ischild(struct prison *pr1, struct prison *pr2)
3930 {
3931 
3932 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3933 		if (pr1 == pr2)
3934 			return (true);
3935 	return (false);
3936 }
3937 
3938 /*
3939  * Return true if the prison is currently alive.  A prison is alive if it
3940  * holds user references and it isn't being removed.
3941  */
3942 bool
prison_isalive(const struct prison * pr)3943 prison_isalive(const struct prison *pr)
3944 {
3945 
3946 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3947 		return (false);
3948 	return (true);
3949 }
3950 
3951 /*
3952  * Return true if the prison is currently valid.  A prison is valid if it has
3953  * been fully created, and is not being destroyed.  Note that dying prisons
3954  * are still considered valid.  Invalid prisons won't be found under normal
3955  * circumstances, as they're only put in that state by functions that have
3956  * an exclusive hold on allprison_lock.
3957  */
3958 bool
prison_isvalid(struct prison * pr)3959 prison_isvalid(struct prison *pr)
3960 {
3961 
3962 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3963 		return (false);
3964 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3965 		return (false);
3966 	return (true);
3967 }
3968 
3969 /*
3970  * Return true if the passed credential is in a jail and that jail does not
3971  * have its own virtual network stack, otherwise false.
3972  */
3973 bool
jailed_without_vnet(struct ucred * cred)3974 jailed_without_vnet(struct ucred *cred)
3975 {
3976 
3977 	if (!jailed(cred))
3978 		return (false);
3979 #ifdef VIMAGE
3980 	if (prison_owns_vnet(cred->cr_prison))
3981 		return (false);
3982 #endif
3983 
3984 	return (true);
3985 }
3986 
3987 /*
3988  * Return the correct hostname (domainname, et al) for the passed credential.
3989  */
3990 void
getcredhostname(struct ucred * cred,char * buf,size_t size)3991 getcredhostname(struct ucred *cred, char *buf, size_t size)
3992 {
3993 	struct prison *pr;
3994 
3995 	/*
3996 	 * A NULL credential can be used to shortcut to the physical
3997 	 * system's hostname.
3998 	 */
3999 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
4000 	mtx_lock(&pr->pr_mtx);
4001 	strlcpy(buf, pr->pr_hostname, size);
4002 	mtx_unlock(&pr->pr_mtx);
4003 }
4004 
4005 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)4006 getcreddomainname(struct ucred *cred, char *buf, size_t size)
4007 {
4008 
4009 	mtx_lock(&cred->cr_prison->pr_mtx);
4010 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
4011 	mtx_unlock(&cred->cr_prison->pr_mtx);
4012 }
4013 
4014 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)4015 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
4016 {
4017 
4018 	mtx_lock(&cred->cr_prison->pr_mtx);
4019 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
4020 	mtx_unlock(&cred->cr_prison->pr_mtx);
4021 }
4022 
4023 void
getcredhostid(struct ucred * cred,unsigned long * hostid)4024 getcredhostid(struct ucred *cred, unsigned long *hostid)
4025 {
4026 
4027 	mtx_lock(&cred->cr_prison->pr_mtx);
4028 	*hostid = cred->cr_prison->pr_hostid;
4029 	mtx_unlock(&cred->cr_prison->pr_mtx);
4030 }
4031 
4032 void
getjailname(struct ucred * cred,char * name,size_t len)4033 getjailname(struct ucred *cred, char *name, size_t len)
4034 {
4035 
4036 	mtx_lock(&cred->cr_prison->pr_mtx);
4037 	strlcpy(name, cred->cr_prison->pr_name, len);
4038 	mtx_unlock(&cred->cr_prison->pr_mtx);
4039 }
4040 
4041 #ifdef VIMAGE
4042 /*
4043  * Determine whether the prison owns its VNET.
4044  */
4045 bool
prison_owns_vnet(struct prison * pr)4046 prison_owns_vnet(struct prison *pr)
4047 {
4048 
4049 	/*
4050 	 * vnets cannot be added/removed after jail creation,
4051 	 * so no need to lock here.
4052 	 */
4053 	return ((pr->pr_flags & PR_VNET) != 0);
4054 }
4055 #endif
4056 
4057 /*
4058  * Determine whether the subject represented by cred can "see"
4059  * status of a mount point.
4060  * Returns: 0 for permitted, ENOENT otherwise.
4061  * XXX: This function should be called cr_canseemount() and should be
4062  *      placed in kern_prot.c.
4063  */
4064 int
prison_canseemount(struct ucred * cred,struct mount * mp)4065 prison_canseemount(struct ucred *cred, struct mount *mp)
4066 {
4067 	struct prison *pr;
4068 	struct statfs *sp;
4069 	size_t len;
4070 
4071 	pr = cred->cr_prison;
4072 	if (pr->pr_enforce_statfs == 0)
4073 		return (0);
4074 	if (pr->pr_root->v_mount == mp)
4075 		return (0);
4076 	if (pr->pr_enforce_statfs == 2)
4077 		return (ENOENT);
4078 	/*
4079 	 * If jail's chroot directory is set to "/" we should be able to see
4080 	 * all mount-points from inside a jail.
4081 	 * This is ugly check, but this is the only situation when jail's
4082 	 * directory ends with '/'.
4083 	 */
4084 	if (strcmp(pr->pr_path, "/") == 0)
4085 		return (0);
4086 	len = strlen(pr->pr_path);
4087 	sp = &mp->mnt_stat;
4088 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
4089 		return (ENOENT);
4090 	/*
4091 	 * Be sure that we don't have situation where jail's root directory
4092 	 * is "/some/path" and mount point is "/some/pathpath".
4093 	 */
4094 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
4095 		return (ENOENT);
4096 	return (0);
4097 }
4098 
4099 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)4100 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
4101 {
4102 	char jpath[MAXPATHLEN];
4103 	struct prison *pr;
4104 	size_t len;
4105 
4106 	pr = cred->cr_prison;
4107 	if (pr->pr_enforce_statfs == 0)
4108 		return;
4109 	if (prison_canseemount(cred, mp) != 0) {
4110 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4111 		strlcpy(sp->f_mntonname, "[restricted]",
4112 		    sizeof(sp->f_mntonname));
4113 		return;
4114 	}
4115 	if (pr->pr_root->v_mount == mp) {
4116 		/*
4117 		 * Clear current buffer data, so we are sure nothing from
4118 		 * the valid path left there.
4119 		 */
4120 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4121 		*sp->f_mntonname = '/';
4122 		return;
4123 	}
4124 	/*
4125 	 * If jail's chroot directory is set to "/" we should be able to see
4126 	 * all mount-points from inside a jail.
4127 	 */
4128 	if (strcmp(pr->pr_path, "/") == 0)
4129 		return;
4130 	len = strlen(pr->pr_path);
4131 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
4132 	/*
4133 	 * Clear current buffer data, so we are sure nothing from
4134 	 * the valid path left there.
4135 	 */
4136 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4137 	if (*jpath == '\0') {
4138 		/* Should never happen. */
4139 		*sp->f_mntonname = '/';
4140 	} else {
4141 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
4142 	}
4143 }
4144 
4145 /*
4146  * Check with permission for a specific privilege is granted within jail.  We
4147  * have a specific list of accepted privileges; the rest are denied.
4148  */
4149 int
prison_priv_check(struct ucred * cred,int priv)4150 prison_priv_check(struct ucred *cred, int priv)
4151 {
4152 	struct prison *pr;
4153 	int error;
4154 
4155 	/*
4156 	 * Some policies have custom handlers. This routine should not be
4157 	 * called for them. See priv_check_cred().
4158 	 */
4159 	switch (priv) {
4160 	case PRIV_VFS_LOOKUP:
4161 	case PRIV_VFS_GENERATION:
4162 		KASSERT(0, ("prison_priv_check instead of a custom handler "
4163 		    "called for %d\n", priv));
4164 	}
4165 
4166 	if (!jailed(cred))
4167 		return (0);
4168 
4169 #ifdef VIMAGE
4170 	/*
4171 	 * Privileges specific to prisons with a virtual network stack.
4172 	 * There might be a duplicate entry here in case the privilege
4173 	 * is only granted conditionally in the legacy jail case.
4174 	 */
4175 	switch (priv) {
4176 		/*
4177 		 * NFS-specific privileges.
4178 		 */
4179 	case PRIV_NFS_DAEMON:
4180 	case PRIV_VFS_GETFH:
4181 	case PRIV_VFS_MOUNT_EXPORTED:
4182 		if (!prison_check_nfsd(cred))
4183 			return (EPERM);
4184 #ifdef notyet
4185 	case PRIV_NFS_LOCKD:
4186 #endif
4187 		/*
4188 		 * Network stack privileges.
4189 		 */
4190 	case PRIV_NET_BRIDGE:
4191 	case PRIV_NET_GRE:
4192 	case PRIV_NET_BPF:
4193 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
4194 	case PRIV_NET_ROUTE:
4195 	case PRIV_NET_TAP:
4196 	case PRIV_NET_SETIFMTU:
4197 	case PRIV_NET_SETIFFLAGS:
4198 	case PRIV_NET_SETIFCAP:
4199 	case PRIV_NET_SETIFDESCR:
4200 	case PRIV_NET_SETIFNAME	:
4201 	case PRIV_NET_SETIFMETRIC:
4202 	case PRIV_NET_SETIFPHYS:
4203 	case PRIV_NET_SETIFMAC:
4204 	case PRIV_NET_SETLANPCP:
4205 	case PRIV_NET_ADDMULTI:
4206 	case PRIV_NET_DELMULTI:
4207 	case PRIV_NET_HWIOCTL:
4208 	case PRIV_NET_SETLLADDR:
4209 	case PRIV_NET_ADDIFGROUP:
4210 	case PRIV_NET_DELIFGROUP:
4211 	case PRIV_NET_IFCREATE:
4212 	case PRIV_NET_IFDESTROY:
4213 	case PRIV_NET_ADDIFADDR:
4214 	case PRIV_NET_DELIFADDR:
4215 	case PRIV_NET_LAGG:
4216 	case PRIV_NET_GIF:
4217 	case PRIV_NET_SETIFVNET:
4218 	case PRIV_NET_SETIFFIB:
4219 	case PRIV_NET_OVPN:
4220 	case PRIV_NET_ME:
4221 	case PRIV_NET_WG:
4222 
4223 		/*
4224 		 * 802.11-related privileges.
4225 		 */
4226 	case PRIV_NET80211_VAP_GETKEY:
4227 	case PRIV_NET80211_VAP_MANAGE:
4228 
4229 #ifdef notyet
4230 		/*
4231 		 * ATM privileges.
4232 		 */
4233 	case PRIV_NETATM_CFG:
4234 	case PRIV_NETATM_ADD:
4235 	case PRIV_NETATM_DEL:
4236 	case PRIV_NETATM_SET:
4237 
4238 		/*
4239 		 * Bluetooth privileges.
4240 		 */
4241 	case PRIV_NETBLUETOOTH_RAW:
4242 #endif
4243 
4244 		/*
4245 		 * Netgraph and netgraph module privileges.
4246 		 */
4247 	case PRIV_NETGRAPH_CONTROL:
4248 #ifdef notyet
4249 	case PRIV_NETGRAPH_TTY:
4250 #endif
4251 
4252 		/*
4253 		 * IPv4 and IPv6 privileges.
4254 		 */
4255 	case PRIV_NETINET_IPFW:
4256 	case PRIV_NETINET_DIVERT:
4257 	case PRIV_NETINET_PF:
4258 	case PRIV_NETINET_DUMMYNET:
4259 	case PRIV_NETINET_CARP:
4260 	case PRIV_NETINET_MROUTE:
4261 	case PRIV_NETINET_RAW:
4262 	case PRIV_NETINET_ADDRCTRL6:
4263 	case PRIV_NETINET_ND6:
4264 	case PRIV_NETINET_SCOPE6:
4265 	case PRIV_NETINET_ALIFETIME6:
4266 	case PRIV_NETINET_IPSEC:
4267 	case PRIV_NETINET_BINDANY:
4268 
4269 #ifdef notyet
4270 		/*
4271 		 * NCP privileges.
4272 		 */
4273 	case PRIV_NETNCP:
4274 
4275 		/*
4276 		 * SMB privileges.
4277 		 */
4278 	case PRIV_NETSMB:
4279 #endif
4280 
4281 	/*
4282 	 * No default: or deny here.
4283 	 * In case of no permit fall through to next switch().
4284 	 */
4285 		if (cred->cr_prison->pr_flags & PR_VNET)
4286 			return (0);
4287 	}
4288 #endif /* VIMAGE */
4289 
4290 	switch (priv) {
4291 		/*
4292 		 * Allow ktrace privileges for root in jail.
4293 		 */
4294 	case PRIV_KTRACE:
4295 
4296 		/*
4297 		 * Allow jailed processes to configure audit identity and
4298 		 * submit audit records (login, etc).  In the future we may
4299 		 * want to further refine the relationship between audit and
4300 		 * jail.
4301 		 */
4302 	case PRIV_AUDIT_GETAUDIT:
4303 	case PRIV_AUDIT_SETAUDIT:
4304 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETAUDIT)
4305 			return (0);
4306 		else
4307 			return (EPERM);
4308 #if 0
4309 	case PRIV_AUDIT_SUBMIT:
4310 #endif
4311 
4312 		/*
4313 		 * Allow jailed processes to manipulate process UNIX
4314 		 * credentials in any way they see fit.
4315 		 */
4316 	case PRIV_CRED_SETCRED:
4317 	case PRIV_CRED_SETUID:
4318 	case PRIV_CRED_SETEUID:
4319 	case PRIV_CRED_SETGID:
4320 	case PRIV_CRED_SETEGID:
4321 	case PRIV_CRED_SETGROUPS:
4322 	case PRIV_CRED_SETREUID:
4323 	case PRIV_CRED_SETREGID:
4324 	case PRIV_CRED_SETRESUID:
4325 	case PRIV_CRED_SETRESGID:
4326 
4327 		/*
4328 		 * Jail implements visibility constraints already, so allow
4329 		 * jailed root to override uid/gid-based constraints.
4330 		 */
4331 	case PRIV_SEEOTHERGIDS:
4332 	case PRIV_SEEOTHERUIDS:
4333 	case PRIV_SEEJAILPROC:
4334 
4335 		/*
4336 		 * Jail implements inter-process debugging limits already, so
4337 		 * allow jailed root various debugging privileges.
4338 		 */
4339 	case PRIV_DEBUG_DIFFCRED:
4340 	case PRIV_DEBUG_SUGID:
4341 	case PRIV_DEBUG_UNPRIV:
4342 	case PRIV_DEBUG_DIFFJAIL:
4343 
4344 		/*
4345 		 * Allow jail to set various resource limits and login
4346 		 * properties, and for now, exceed process resource limits.
4347 		 */
4348 	case PRIV_PROC_LIMIT:
4349 	case PRIV_PROC_SETLOGIN:
4350 	case PRIV_PROC_SETRLIMIT:
4351 
4352 		/*
4353 		 * Debuggers should work in jails.
4354 		 */
4355 	case PRIV_PROC_MEM_WRITE:
4356 
4357 		/*
4358 		 * System V and POSIX IPC privileges are granted in jail.
4359 		 */
4360 	case PRIV_IPC_READ:
4361 	case PRIV_IPC_WRITE:
4362 	case PRIV_IPC_ADMIN:
4363 	case PRIV_IPC_MSGSIZE:
4364 	case PRIV_MQ_ADMIN:
4365 
4366 		/*
4367 		 * Jail operations within a jail work on child jails.
4368 		 */
4369 	case PRIV_JAIL_ATTACH:
4370 	case PRIV_JAIL_SET:
4371 	case PRIV_JAIL_REMOVE:
4372 
4373 		/*
4374 		 * Jail implements its own inter-process limits, so allow
4375 		 * root processes in jail to change scheduling on other
4376 		 * processes in the same jail.  Likewise for signalling.
4377 		 */
4378 	case PRIV_SCHED_DIFFCRED:
4379 	case PRIV_SCHED_CPUSET:
4380 	case PRIV_SCHED_DIFFJAIL:
4381 	case PRIV_SIGNAL_DIFFCRED:
4382 	case PRIV_SIGNAL_SUGID:
4383 	case PRIV_SIGNAL_DIFFJAIL:
4384 
4385 		/*
4386 		 * Allow jailed processes to write to sysctls marked as jail
4387 		 * writable.
4388 		 */
4389 	case PRIV_SYSCTL_WRITEJAIL:
4390 
4391 		/*
4392 		 * Allow root in jail to manage a variety of quota
4393 		 * properties.  These should likely be conditional on a
4394 		 * configuration option.
4395 		 */
4396 	case PRIV_VFS_GETQUOTA:
4397 	case PRIV_VFS_SETQUOTA:
4398 
4399 		/*
4400 		 * Since Jail relies on chroot() to implement file system
4401 		 * protections, grant many VFS privileges to root in jail.
4402 		 * Be careful to exclude mount-related and NFS-related
4403 		 * privileges.
4404 		 */
4405 	case PRIV_VFS_READ:
4406 	case PRIV_VFS_WRITE:
4407 	case PRIV_VFS_ADMIN:
4408 	case PRIV_VFS_EXEC:
4409 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
4410 	case PRIV_VFS_CHFLAGS_DEV:
4411 	case PRIV_VFS_CHOWN:
4412 	case PRIV_VFS_CHROOT:
4413 	case PRIV_VFS_RETAINSUGID:
4414 	case PRIV_VFS_FCHROOT:
4415 	case PRIV_VFS_LINK:
4416 	case PRIV_VFS_SETGID:
4417 	case PRIV_VFS_STAT:
4418 	case PRIV_VFS_STICKYFILE:
4419 
4420 		/*
4421 		 * As in the non-jail case, non-root users are expected to be
4422 		 * able to read kernel/physical memory (provided /dev/[k]mem
4423 		 * exists in the jail and they have permission to access it).
4424 		 */
4425 	case PRIV_KMEM_READ:
4426 		return (0);
4427 
4428 		/*
4429 		 * Depending on the global setting, allow privilege of
4430 		 * setting system flags.
4431 		 */
4432 	case PRIV_VFS_SYSFLAGS:
4433 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4434 			return (0);
4435 		else
4436 			return (EPERM);
4437 
4438 		/*
4439 		 * Depending on the global setting, allow privilege of
4440 		 * mounting/unmounting file systems.
4441 		 */
4442 	case PRIV_VFS_MOUNT:
4443 	case PRIV_VFS_UNMOUNT:
4444 	case PRIV_VFS_MOUNT_NONUSER:
4445 	case PRIV_VFS_MOUNT_OWNER:
4446 		pr = cred->cr_prison;
4447 		prison_lock(pr);
4448 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4449 			error = 0;
4450 		else
4451 			error = EPERM;
4452 		prison_unlock(pr);
4453 		return (error);
4454 
4455 		/*
4456 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4457 		 * policy.  priv_check_cred will not specifically allow it, and
4458 		 * we may want a MAC policy to allow it.
4459 		 */
4460 	case PRIV_VFS_READ_DIR:
4461 		return (0);
4462 
4463 		/*
4464 		 * Conditionally allow privileged process in the jail to
4465 		 * manipulate filesystem extended attributes in the system
4466 		 * namespace.
4467 		 */
4468 	case PRIV_VFS_EXTATTR_SYSTEM:
4469 		if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4470 			return (0);
4471 		else
4472 			return (EPERM);
4473 
4474 		/*
4475 		 * Conditionnaly allow locking (unlocking) physical pages
4476 		 * in memory.
4477 		 */
4478 	case PRIV_VM_MLOCK:
4479 	case PRIV_VM_MUNLOCK:
4480 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4481 			return (0);
4482 		else
4483 			return (EPERM);
4484 
4485 		/*
4486 		 * Conditionally allow jailed root to bind reserved ports.
4487 		 */
4488 	case PRIV_NETINET_RESERVEDPORT:
4489 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4490 			return (0);
4491 		else
4492 			return (EPERM);
4493 
4494 		/*
4495 		 * Allow jailed root to reuse in-use ports.
4496 		 */
4497 	case PRIV_NETINET_REUSEPORT:
4498 		return (0);
4499 
4500 		/*
4501 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4502 		 */
4503 	case PRIV_NETINET_SETHDROPTS:
4504 		return (0);
4505 
4506 		/*
4507 		 * Conditionally allow creating raw sockets in jail.
4508 		 */
4509 	case PRIV_NETINET_RAW:
4510 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4511 			return (0);
4512 		else
4513 			return (EPERM);
4514 
4515 		/*
4516 		 * Since jail implements its own visibility limits on netstat
4517 		 * sysctls, allow getcred.  This allows identd to work in
4518 		 * jail.
4519 		 */
4520 	case PRIV_NETINET_GETCRED:
4521 		return (0);
4522 
4523 		/*
4524 		 * Allow jailed root to set loginclass.
4525 		 */
4526 	case PRIV_PROC_SETLOGINCLASS:
4527 		return (0);
4528 
4529 		/*
4530 		 * Do not allow a process inside a jail to read the kernel
4531 		 * message buffer unless explicitly permitted.
4532 		 */
4533 	case PRIV_MSGBUF:
4534 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4535 			return (0);
4536 		return (EPERM);
4537 
4538 		/*
4539 		 * Conditionally allow privileged process in the jail adjust
4540 		 * machine time.
4541 		 */
4542 	case PRIV_ADJTIME:
4543 	case PRIV_NTP_ADJTIME:
4544 		if (cred->cr_prison->pr_allow &
4545 		    (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4546 			return (0);
4547 		}
4548 		return (EPERM);
4549 
4550 		/*
4551 		 * Conditionally allow privileged process in the jail set
4552 		 * machine time.
4553 		 */
4554 	case PRIV_SETTIMEOFDAY:
4555 	case PRIV_CLOCK_SETTIME:
4556 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4557 			return (0);
4558 		else
4559 			return (EPERM);
4560 
4561 		/*
4562 		 * Conditionally allow privileged process in the jail to modify
4563 		 * the routing table.
4564 		 */
4565 	case PRIV_NET_ROUTE:
4566 		if (cred->cr_prison->pr_allow & PR_ALLOW_ROUTING)
4567 			return (0);
4568 		else
4569 			return (EPERM);
4570 
4571 	default:
4572 		/*
4573 		 * In all remaining cases, deny the privilege request.  This
4574 		 * includes almost all network privileges, many system
4575 		 * configuration privileges.
4576 		 */
4577 		return (EPERM);
4578 	}
4579 }
4580 
4581 /*
4582  * Return the part of pr2's name that is relative to pr1, or the whole name
4583  * if it does not directly follow.
4584  */
4585 
4586 char *
prison_name(struct prison * pr1,struct prison * pr2)4587 prison_name(struct prison *pr1, struct prison *pr2)
4588 {
4589 	char *name;
4590 
4591 	/* Jails see themselves as "0" (if they see themselves at all). */
4592 	if (pr1 == pr2)
4593 		return "0";
4594 	name = pr2->pr_name;
4595 	if (prison_ischild(pr1, pr2)) {
4596 		/*
4597 		 * pr1 isn't locked (and allprison_lock may not be either)
4598 		 * so its length can't be counted on.  But the number of dots
4599 		 * can be counted on - and counted.
4600 		 */
4601 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4602 			name = strchr(name, '.') + 1;
4603 	}
4604 	return (name);
4605 }
4606 
4607 /*
4608  * Return the part of pr2's path that is relative to pr1, or the whole path
4609  * if it does not directly follow.
4610  */
4611 static char *
prison_path(struct prison * pr1,struct prison * pr2)4612 prison_path(struct prison *pr1, struct prison *pr2)
4613 {
4614 	char *path1, *path2;
4615 	int len1;
4616 
4617 	path1 = pr1->pr_path;
4618 	path2 = pr2->pr_path;
4619 	if (!strcmp(path1, "/"))
4620 		return (path2);
4621 	len1 = strlen(path1);
4622 	if (strncmp(path1, path2, len1))
4623 		return (path2);
4624 	if (path2[len1] == '\0')
4625 		return "/";
4626 	if (path2[len1] == '/')
4627 		return (path2 + len1);
4628 	return (path2);
4629 }
4630 
4631 /*
4632  * Jail-related sysctls.
4633  */
4634 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4635     "Jails");
4636 
4637 #if defined(INET) || defined(INET6)
4638 /*
4639  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4640  * sysctl_jail_list() helper.
4641  */
4642 static void
prison_ip_copyout(struct prison * pr,const pr_family_t af,void ** out,int * len)4643 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4644 {
4645 	const struct prison_ip *pip;
4646 	const size_t size = pr_families[af].size;
4647 
4648  again:
4649 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4650 	if ((pip = pr->pr_addrs[af]) != NULL) {
4651 		if (*len < pip->ips) {
4652 			*len = pip->ips;
4653 			mtx_unlock(&pr->pr_mtx);
4654 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4655 			mtx_lock(&pr->pr_mtx);
4656 			goto again;
4657 		}
4658 		bcopy(pip->pr_ip, *out, pip->ips * size);
4659 	}
4660 }
4661 #endif
4662 
4663 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4664 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4665 {
4666 	struct xprison *xp;
4667 	struct prison *pr, *cpr;
4668 #ifdef INET
4669 	struct in_addr *ip4 = NULL;
4670 	int ip4s = 0;
4671 #endif
4672 #ifdef INET6
4673 	struct in6_addr *ip6 = NULL;
4674 	int ip6s = 0;
4675 #endif
4676 	int descend, error;
4677 
4678 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4679 	pr = req->td->td_ucred->cr_prison;
4680 	error = 0;
4681 	sx_slock(&allprison_lock);
4682 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4683 		mtx_lock(&cpr->pr_mtx);
4684 #ifdef INET
4685 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4686 #endif
4687 #ifdef INET6
4688 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4689 #endif
4690 		bzero(xp, sizeof(*xp));
4691 		xp->pr_version = XPRISON_VERSION;
4692 		xp->pr_id = cpr->pr_id;
4693 		xp->pr_state = cpr->pr_state;
4694 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4695 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4696 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4697 #ifdef INET
4698 		xp->pr_ip4s = ip4s;
4699 #endif
4700 #ifdef INET6
4701 		xp->pr_ip6s = ip6s;
4702 #endif
4703 		mtx_unlock(&cpr->pr_mtx);
4704 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4705 		if (error)
4706 			break;
4707 #ifdef INET
4708 		if (xp->pr_ip4s > 0) {
4709 			error = SYSCTL_OUT(req, ip4,
4710 			    xp->pr_ip4s * sizeof(struct in_addr));
4711 			if (error)
4712 				break;
4713 		}
4714 #endif
4715 #ifdef INET6
4716 		if (xp->pr_ip6s > 0) {
4717 			error = SYSCTL_OUT(req, ip6,
4718 			    xp->pr_ip6s * sizeof(struct in6_addr));
4719 			if (error)
4720 				break;
4721 		}
4722 #endif
4723 	}
4724 	sx_sunlock(&allprison_lock);
4725 	free(xp, M_TEMP);
4726 #ifdef INET
4727 	free(ip4, M_TEMP);
4728 #endif
4729 #ifdef INET6
4730 	free(ip6, M_TEMP);
4731 #endif
4732 	return (error);
4733 }
4734 
4735 SYSCTL_OID(_security_jail, OID_AUTO, list,
4736     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4737     sysctl_jail_list, "S", "List of active jails");
4738 
4739 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)4740 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4741 {
4742 	int error, injail;
4743 
4744 	injail = jailed(req->td->td_ucred);
4745 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4746 
4747 	return (error);
4748 }
4749 
4750 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4751     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4752     sysctl_jail_jailed, "I", "Process in jail?");
4753 
4754 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)4755 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4756 {
4757 	int error, havevnet;
4758 #ifdef VIMAGE
4759 	struct ucred *cred = req->td->td_ucred;
4760 
4761 	havevnet = jailed(cred) && prison_owns_vnet(cred->cr_prison);
4762 #else
4763 	havevnet = 0;
4764 #endif
4765 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4766 
4767 	return (error);
4768 }
4769 
4770 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4771     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4772     sysctl_jail_vnet, "I", "Jail owns vnet?");
4773 
4774 #if defined(INET) || defined(INET6)
4775 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4776     &jail_max_af_ips, 0,
4777     "Number of IP addresses a jail may have at most per address family (deprecated)");
4778 #endif
4779 
4780 /*
4781  * Default parameters for jail(2) compatibility.  For historical reasons,
4782  * the sysctl names have varying similarity to the parameter names.  Prisons
4783  * just see their own parameters, and can't change them.
4784  */
4785 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)4786 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4787 {
4788 	int error, i;
4789 
4790 	/* Get the current flag value, and convert it to a boolean. */
4791 	if (req->td->td_ucred->cr_prison == &prison0) {
4792 		mtx_lock(&prison0.pr_mtx);
4793 		i = (jail_default_allow & arg2) != 0;
4794 		mtx_unlock(&prison0.pr_mtx);
4795 	} else
4796 		i = prison_allow(req->td->td_ucred, arg2);
4797 
4798 	if (arg1 != NULL)
4799 		i = !i;
4800 	error = sysctl_handle_int(oidp, &i, 0, req);
4801 	if (error || !req->newptr)
4802 		return (error);
4803 	i = i ? arg2 : 0;
4804 	if (arg1 != NULL)
4805 		i ^= arg2;
4806 	/*
4807 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4808 	 * for writing.
4809 	 */
4810 	mtx_lock(&prison0.pr_mtx);
4811 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4812 	mtx_unlock(&prison0.pr_mtx);
4813 	return (0);
4814 }
4815 
4816 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4817     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4818     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4819     "Processes in jail can set their hostnames (deprecated)");
4820 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4821     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4822     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4823     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4824 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4825     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4826     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4827     "Processes in jail can use System V IPC primitives (deprecated)");
4828 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4829     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4830     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4831     "Prison root can create raw sockets (deprecated)");
4832 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4833     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4834     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4835     "Processes in jail can alter system file flags (deprecated)");
4836 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4837     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4838     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4839     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4840 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
4841     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4842     NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
4843     "Processes in jail can lock/unlock physical pages in memory");
4844 
4845 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)4846 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4847 {
4848 	struct prison *pr;
4849 	int level, error;
4850 
4851 	pr = req->td->td_ucred->cr_prison;
4852 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4853 	error = sysctl_handle_int(oidp, &level, 0, req);
4854 	if (error || !req->newptr)
4855 		return (error);
4856 	*(int *)arg1 = level;
4857 	return (0);
4858 }
4859 
4860 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4861     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4862     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4863     sysctl_jail_default_level, "I",
4864     "Processes in jail cannot see all mounted file systems (deprecated)");
4865 
4866 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4867     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4868     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4869     sysctl_jail_default_level, "I",
4870     "Ruleset for the devfs filesystem in jail (deprecated)");
4871 
4872 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4873     "Limits and stats of child jails");
4874 
4875 static int
sysctl_jail_children(SYSCTL_HANDLER_ARGS)4876 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
4877 {
4878 	struct prison *pr;
4879 	int i;
4880 
4881 	pr = req->td->td_ucred->cr_prison;
4882 
4883 	switch (oidp->oid_kind & CTLTYPE) {
4884 	case CTLTYPE_INT:
4885 		i = *(int *)((char *)pr + arg2);
4886 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4887 	}
4888 
4889 	return (0);
4890 }
4891 
4892 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
4893     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4894     NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
4895     "I", "Maximum number of child jails");
4896 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
4897     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4898     NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
4899     "I", "Current number of child jails");
4900 
4901 /*
4902  * Nodes to describe jail parameters.  Maximum length of string parameters
4903  * is returned in the string itself, and the other parameters exist merely
4904  * to make themselves and their types known.
4905  */
4906 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4907     "Jail parameters");
4908 
4909 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)4910 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4911 {
4912 	int i;
4913 	long l;
4914 	size_t s;
4915 	char numbuf[12];
4916 
4917 	switch (oidp->oid_kind & CTLTYPE)
4918 	{
4919 	case CTLTYPE_LONG:
4920 	case CTLTYPE_ULONG:
4921 		l = 0;
4922 #ifdef SCTL_MASK32
4923 		if (!(req->flags & SCTL_MASK32))
4924 #endif
4925 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4926 	case CTLTYPE_INT:
4927 	case CTLTYPE_UINT:
4928 		i = 0;
4929 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4930 	case CTLTYPE_STRING:
4931 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4932 		return
4933 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4934 	case CTLTYPE_STRUCT:
4935 		s = (size_t)arg2;
4936 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4937 	}
4938 	return (0);
4939 }
4940 
4941 /*
4942  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4943  * jail creation time but cannot be changed in an existing jail.
4944  */
4945 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4946 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4947 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4948 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4949 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4950     "I", "Jail secure level");
4951 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4952     "Jail value for kern.osreldate and uname -K");
4953 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4954     "Jail value for kern.osrelease and uname -r");
4955 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4956     "I", "Jail cannot see all mounted file systems");
4957 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4958     "I", "Ruleset for in-jail devfs mounts");
4959 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4960     "B", "Jail persistence");
4961 #ifdef VIMAGE
4962 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4963     "E,jailsys", "Virtual network stack");
4964 #endif
4965 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4966     "B", "Jail is in the process of shutting down");
4967 
4968 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4969 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4970     "I", "Current number of child jails");
4971 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4972     "I", "Maximum number of child jails");
4973 
4974 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4975 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4976     "Jail hostname");
4977 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4978     "Jail NIS domainname");
4979 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4980     "Jail host UUID");
4981 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4982     "LU", "Jail host ID");
4983 
4984 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4985 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4986 
4987 #ifdef INET
4988 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4989     "Jail IPv4 address virtualization");
4990 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4991     "S,in_addr,a", "Jail IPv4 addresses");
4992 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4993     "B", "Do (not) use IPv4 source address selection rather than the "
4994     "primary jail IPv4 address.");
4995 #endif
4996 #ifdef INET6
4997 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4998     "Jail IPv6 address virtualization");
4999 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
5000     "S,in6_addr,a", "Jail IPv6 addresses");
5001 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5002     "B", "Do (not) use IPv6 source address selection rather than the "
5003     "primary jail IPv6 address.");
5004 #endif
5005 
5006 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
5007 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
5008     "B", "Jail may set hostname");
5009 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
5010     "B", "Jail may use SYSV IPC");
5011 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
5012     "B", "Jail may create raw sockets");
5013 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
5014     "B", "Jail may alter system file flags");
5015 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
5016     "B", "Jail may set file quotas");
5017 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
5018     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
5019 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
5020     "B", "Jail may lock (unlock) physical pages in memory");
5021 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
5022     "B", "Jail may bind sockets to reserved ports");
5023 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
5024     "B", "Jail may read the kernel message buffer");
5025 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
5026     "B", "Unprivileged processes may use process debugging facilities");
5027 SYSCTL_JAIL_PARAM(_allow, unprivileged_parent_tampering,
5028     CTLTYPE_INT | CTLFLAG_RW, "B",
5029     "Unprivileged parent jail processes may tamper with same-uid processes"
5030     " (signal/debug/cpuset)");
5031 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
5032     "B", "Processes in jail with uid 0 have privilege");
5033 #ifdef VIMAGE
5034 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
5035     "B", "Mountd/nfsd may run in the jail");
5036 #endif
5037 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
5038     "B", "Jail may set system-level filesystem extended attributes");
5039 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
5040     "B", "Jail may adjust system time");
5041 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
5042     "B", "Jail may set system time");
5043 SYSCTL_JAIL_PARAM(_allow, routing, CTLTYPE_INT | CTLFLAG_RW,
5044     "B", "Jail may modify routing table");
5045 #ifdef AUDIT
5046 SYSCTL_JAIL_PARAM(_allow, setaudit, CTLTYPE_INT | CTLFLAG_RW,
5047     "B", "Jail may set and get audit session state");
5048 #endif
5049 
5050 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
5051 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
5052     "B", "Jail may mount/unmount jail-friendly file systems in general");
5053 
5054 /*
5055  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
5056  * its associated bit in the pr_allow bitmask, or zero if the parameter was
5057  * not created.
5058  */
5059 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)5060 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
5061     const char *descr)
5062 {
5063 	struct bool_flags *bf;
5064 	struct sysctl_oid *parent;
5065 	char *allow_name, *allow_noname, *allowed;
5066 #ifndef NO_SYSCTL_DESCR
5067 	char *descr_deprecated;
5068 #endif
5069 	u_int allow_flag;
5070 
5071 	if (prefix
5072 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
5073 		< 0 ||
5074 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
5075 		< 0
5076 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
5077 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
5078 		free(allow_name, M_PRISON);
5079 		return 0;
5080 	}
5081 
5082 	/*
5083 	 * See if this parameter has already beed added, i.e. a module was
5084 	 * previously loaded/unloaded.
5085 	 */
5086 	mtx_lock(&prison0.pr_mtx);
5087 	for (bf = pr_flag_allow;
5088 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5089 		atomic_load_int(&bf->flag) != 0;
5090 	     bf++) {
5091 		if (strcmp(bf->name, allow_name) == 0) {
5092 			allow_flag = bf->flag;
5093 			goto no_add;
5094 		}
5095 	}
5096 
5097 	/*
5098 	 * Find a free bit in pr_allow_all, failing if there are none
5099 	 * (which shouldn't happen as long as we keep track of how many
5100 	 * potential dynamic flags exist).
5101 	 */
5102 	for (allow_flag = 1;; allow_flag <<= 1) {
5103 		if (allow_flag == 0)
5104 			goto no_add;
5105 		if ((pr_allow_all & allow_flag) == 0)
5106 			break;
5107 	}
5108 
5109 	/* Note the parameter in the next open slot in pr_flag_allow. */
5110 	for (bf = pr_flag_allow; ; bf++) {
5111 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
5112 			/* This should never happen, but is not fatal. */
5113 			allow_flag = 0;
5114 			goto no_add;
5115 		}
5116 		if (atomic_load_int(&bf->flag) == 0)
5117 			break;
5118 	}
5119 	bf->name = allow_name;
5120 	bf->noname = allow_noname;
5121 	pr_allow_all |= allow_flag;
5122 	/*
5123 	 * prison0 always has permission for the new parameter.
5124 	 * Other jails must have it granted to them.
5125 	 */
5126 	prison0.pr_allow |= allow_flag;
5127 	/* The flag indicates a valid entry, so make sure it is set last. */
5128 	atomic_store_rel_int(&bf->flag, allow_flag);
5129 	mtx_unlock(&prison0.pr_mtx);
5130 
5131 	/*
5132 	 * Create sysctls for the parameter, and the back-compat global
5133 	 * permission.
5134 	 */
5135 	parent = prefix
5136 	    ? SYSCTL_ADD_NODE(NULL,
5137 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
5138 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
5139 	    : &sysctl___security_jail_param_allow;
5140 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
5141 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5142 	    NULL, 0, sysctl_jail_param, "B", descr);
5143 	if ((prefix
5144 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
5145 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
5146 #ifndef NO_SYSCTL_DESCR
5147 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
5148 		    descr);
5149 #endif
5150 		(void)SYSCTL_ADD_PROC(NULL,
5151 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
5152 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
5153 		    sysctl_jail_default_allow, "I", descr_deprecated);
5154 #ifndef NO_SYSCTL_DESCR
5155 		free(descr_deprecated, M_TEMP);
5156 #endif
5157 		free(allowed, M_TEMP);
5158 	}
5159 	return allow_flag;
5160 
5161  no_add:
5162 	mtx_unlock(&prison0.pr_mtx);
5163 	free(allow_name, M_PRISON);
5164 	free(allow_noname, M_PRISON);
5165 	return allow_flag;
5166 }
5167 
5168 /*
5169  * The VFS system will register jail-aware filesystems here.  They each get
5170  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
5171  * attempts to mount.
5172  */
5173 void
prison_add_vfs(struct vfsconf * vfsp)5174 prison_add_vfs(struct vfsconf *vfsp)
5175 {
5176 #ifdef NO_SYSCTL_DESCR
5177 
5178 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5179 	    NULL, NULL);
5180 #else
5181 	char *descr;
5182 
5183 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
5184 	    vfsp->vfc_name);
5185 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5186 	    NULL, descr);
5187 	free(descr, M_TEMP);
5188 #endif
5189 }
5190 
5191 #ifdef RACCT
5192 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)5193 prison_racct_foreach(void (*callback)(struct racct *racct,
5194     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
5195     void *arg2, void *arg3)
5196 {
5197 	struct prison_racct *prr;
5198 
5199 	ASSERT_RACCT_ENABLED();
5200 
5201 	sx_slock(&allprison_lock);
5202 	if (pre != NULL)
5203 		(pre)();
5204 	LIST_FOREACH(prr, &allprison_racct, prr_next)
5205 		(callback)(prr->prr_racct, arg2, arg3);
5206 	if (post != NULL)
5207 		(post)();
5208 	sx_sunlock(&allprison_lock);
5209 }
5210 
5211 static struct prison_racct *
prison_racct_find_locked(const char * name)5212 prison_racct_find_locked(const char *name)
5213 {
5214 	struct prison_racct *prr;
5215 
5216 	ASSERT_RACCT_ENABLED();
5217 	sx_assert(&allprison_lock, SA_XLOCKED);
5218 
5219 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
5220 		return (NULL);
5221 
5222 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
5223 		if (strcmp(name, prr->prr_name) != 0)
5224 			continue;
5225 
5226 		/* Found prison_racct with a matching name? */
5227 		prison_racct_hold(prr);
5228 		return (prr);
5229 	}
5230 
5231 	/* Add new prison_racct. */
5232 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
5233 	racct_create(&prr->prr_racct);
5234 
5235 	strcpy(prr->prr_name, name);
5236 	refcount_init(&prr->prr_refcount, 1);
5237 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
5238 
5239 	return (prr);
5240 }
5241 
5242 struct prison_racct *
prison_racct_find(const char * name)5243 prison_racct_find(const char *name)
5244 {
5245 	struct prison_racct *prr;
5246 
5247 	ASSERT_RACCT_ENABLED();
5248 
5249 	sx_xlock(&allprison_lock);
5250 	prr = prison_racct_find_locked(name);
5251 	sx_xunlock(&allprison_lock);
5252 	return (prr);
5253 }
5254 
5255 void
prison_racct_hold(struct prison_racct * prr)5256 prison_racct_hold(struct prison_racct *prr)
5257 {
5258 
5259 	ASSERT_RACCT_ENABLED();
5260 
5261 	refcount_acquire(&prr->prr_refcount);
5262 }
5263 
5264 static void
prison_racct_free_locked(struct prison_racct * prr)5265 prison_racct_free_locked(struct prison_racct *prr)
5266 {
5267 
5268 	ASSERT_RACCT_ENABLED();
5269 	sx_assert(&allprison_lock, SA_XLOCKED);
5270 
5271 	if (refcount_release(&prr->prr_refcount)) {
5272 		racct_destroy(&prr->prr_racct);
5273 		LIST_REMOVE(prr, prr_next);
5274 		free(prr, M_PRISON_RACCT);
5275 	}
5276 }
5277 
5278 void
prison_racct_free(struct prison_racct * prr)5279 prison_racct_free(struct prison_racct *prr)
5280 {
5281 
5282 	ASSERT_RACCT_ENABLED();
5283 	sx_assert(&allprison_lock, SA_UNLOCKED);
5284 
5285 	if (refcount_release_if_not_last(&prr->prr_refcount))
5286 		return;
5287 
5288 	sx_xlock(&allprison_lock);
5289 	prison_racct_free_locked(prr);
5290 	sx_xunlock(&allprison_lock);
5291 }
5292 
5293 static void
prison_racct_attach(struct prison * pr)5294 prison_racct_attach(struct prison *pr)
5295 {
5296 	struct prison_racct *prr;
5297 
5298 	ASSERT_RACCT_ENABLED();
5299 	sx_assert(&allprison_lock, SA_XLOCKED);
5300 
5301 	prr = prison_racct_find_locked(pr->pr_name);
5302 	KASSERT(prr != NULL, ("cannot find prison_racct"));
5303 
5304 	pr->pr_prison_racct = prr;
5305 }
5306 
5307 /*
5308  * Handle jail renaming.  From the racct point of view, renaming means
5309  * moving from one prison_racct to another.
5310  */
5311 static void
prison_racct_modify(struct prison * pr)5312 prison_racct_modify(struct prison *pr)
5313 {
5314 #ifdef RCTL
5315 	struct proc *p;
5316 	struct ucred *cred;
5317 #endif
5318 	struct prison_racct *oldprr;
5319 
5320 	ASSERT_RACCT_ENABLED();
5321 
5322 	sx_slock(&allproc_lock);
5323 	sx_xlock(&allprison_lock);
5324 
5325 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
5326 		sx_xunlock(&allprison_lock);
5327 		sx_sunlock(&allproc_lock);
5328 		return;
5329 	}
5330 
5331 	oldprr = pr->pr_prison_racct;
5332 	pr->pr_prison_racct = NULL;
5333 
5334 	prison_racct_attach(pr);
5335 
5336 	/*
5337 	 * Move resource utilisation records.
5338 	 */
5339 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
5340 
5341 #ifdef RCTL
5342 	/*
5343 	 * Force rctl to reattach rules to processes.
5344 	 */
5345 	FOREACH_PROC_IN_SYSTEM(p) {
5346 		PROC_LOCK(p);
5347 		cred = crhold(p->p_ucred);
5348 		PROC_UNLOCK(p);
5349 		rctl_proc_ucred_changed(p, cred);
5350 		crfree(cred);
5351 	}
5352 #endif
5353 
5354 	sx_sunlock(&allproc_lock);
5355 	prison_racct_free_locked(oldprr);
5356 	sx_xunlock(&allprison_lock);
5357 }
5358 
5359 static void
prison_racct_detach(struct prison * pr)5360 prison_racct_detach(struct prison *pr)
5361 {
5362 
5363 	ASSERT_RACCT_ENABLED();
5364 	sx_assert(&allprison_lock, SA_UNLOCKED);
5365 
5366 	if (pr->pr_prison_racct == NULL)
5367 		return;
5368 	prison_racct_free(pr->pr_prison_racct);
5369 	pr->pr_prison_racct = NULL;
5370 }
5371 #endif /* RACCT */
5372 
5373 /*
5374  * Submit a knote for a prison, locking if necessary.
5375  */
5376 static void
prison_knote(struct prison * pr,long hint)5377 prison_knote(struct prison *pr, long hint)
5378 {
5379 	int locked;
5380 
5381 	locked = mtx_owned(&pr->pr_mtx);
5382 	if (!locked)
5383 		mtx_lock(&pr->pr_mtx);
5384 	KNOTE_LOCKED(pr->pr_klist, hint);
5385 	jaildesc_knote(pr, hint);
5386 	if (!locked)
5387 		mtx_unlock(&pr->pr_mtx);
5388 }
5389 
5390 #ifdef DDB
5391 
5392 static void
db_show_prison(struct prison * pr)5393 db_show_prison(struct prison *pr)
5394 {
5395 	struct bool_flags *bf;
5396 	struct jailsys_flags *jsf;
5397 #if defined(INET) || defined(INET6)
5398 	int ii;
5399 	struct prison_ip *pip;
5400 #endif
5401 	unsigned f;
5402 #ifdef INET
5403 	char ip4buf[INET_ADDRSTRLEN];
5404 #endif
5405 #ifdef INET6
5406 	char ip6buf[INET6_ADDRSTRLEN];
5407 #endif
5408 
5409 	db_printf("prison %p:\n", pr);
5410 	db_printf(" jid             = %d\n", pr->pr_id);
5411 	db_printf(" name            = %s\n", pr->pr_name);
5412 	db_printf(" parent          = %p\n", pr->pr_parent);
5413 	db_printf(" ref             = %d\n", pr->pr_ref);
5414 	db_printf(" uref            = %d\n", pr->pr_uref);
5415 	db_printf(" state           = %s\n",
5416 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5417 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
5418 	    "invalid");
5419 	db_printf(" path            = %s\n", pr->pr_path);
5420 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
5421 	    ? pr->pr_cpuset->cs_id : -1);
5422 #ifdef VIMAGE
5423 	db_printf(" vnet            = %p\n", pr->pr_vnet);
5424 #endif
5425 	db_printf(" root            = %p\n", pr->pr_root);
5426 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
5427 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
5428 	db_printf(" children.max    = %d\n", pr->pr_childmax);
5429 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
5430 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
5431 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
5432 	db_printf(" flags           = 0x%x", pr->pr_flags);
5433 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5434 		if (pr->pr_flags & bf->flag)
5435 			db_printf(" %s", bf->name);
5436 	for (jsf = pr_flag_jailsys;
5437 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5438 	     jsf++) {
5439 		f = pr->pr_flags & (jsf->disable | jsf->new);
5440 		db_printf(" %-16s= %s\n", jsf->name,
5441 		    (f != 0 && f == jsf->disable) ? "disable"
5442 		    : (f == jsf->new) ? "new"
5443 		    : "inherit");
5444 	}
5445 	db_printf(" allow           = 0x%x", pr->pr_allow);
5446 	for (bf = pr_flag_allow;
5447 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5448 		atomic_load_int(&bf->flag) != 0;
5449 	     bf++)
5450 		if (pr->pr_allow & bf->flag)
5451 			db_printf(" %s", bf->name);
5452 	db_printf("\n");
5453 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
5454 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
5455 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
5456 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
5457 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
5458 #ifdef INET
5459 	if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5460 		db_printf(" ip4s            = %d\n", pip->ips);
5461 		for (ii = 0; ii < pip->ips; ii++)
5462 			db_printf(" %s %s\n",
5463 			    ii == 0 ? "ip4.addr        =" : "                 ",
5464 			    inet_ntoa_r(
5465 			    *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5466 			    ip4buf));
5467 	}
5468 #endif
5469 #ifdef INET6
5470 	if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5471 		db_printf(" ip6s            = %d\n", pip->ips);
5472 		for (ii = 0; ii < pip->ips; ii++)
5473 			db_printf(" %s %s\n",
5474 			    ii == 0 ? "ip6.addr        =" : "                 ",
5475 			    ip6_sprintf(ip6buf,
5476 			    (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5477 	}
5478 #endif
5479 }
5480 
DB_SHOW_COMMAND(prison,db_show_prison_command)5481 DB_SHOW_COMMAND(prison, db_show_prison_command)
5482 {
5483 	struct prison *pr;
5484 
5485 	if (!have_addr) {
5486 		/*
5487 		 * Show all prisons in the list, and prison0 which is not
5488 		 * listed.
5489 		 */
5490 		db_show_prison(&prison0);
5491 		if (!db_pager_quit) {
5492 			TAILQ_FOREACH(pr, &allprison, pr_list) {
5493 				db_show_prison(pr);
5494 				if (db_pager_quit)
5495 					break;
5496 			}
5497 		}
5498 		return;
5499 	}
5500 
5501 	if (addr == 0)
5502 		pr = &prison0;
5503 	else {
5504 		/* Look for a prison with the ID and with references. */
5505 		TAILQ_FOREACH(pr, &allprison, pr_list)
5506 			if (pr->pr_id == addr && pr->pr_ref > 0)
5507 				break;
5508 		if (pr == NULL)
5509 			/* Look again, without requiring a reference. */
5510 			TAILQ_FOREACH(pr, &allprison, pr_list)
5511 				if (pr->pr_id == addr)
5512 					break;
5513 		if (pr == NULL)
5514 			/* Assume address points to a valid prison. */
5515 			pr = (struct prison *)addr;
5516 	}
5517 	db_show_prison(pr);
5518 }
5519 
5520 #endif /* DDB */
5521