xref: /freebsd/sys/kern/kern_jail.c (revision 8a5ceebece0311bc41180b3ca0ce7237def1e253)
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/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/epoch.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/linker.h>
52 #include <sys/lock.h>
53 #include <sys/mman.h>
54 #include <sys/mutex.h>
55 #include <sys/racct.h>
56 #include <sys/rctl.h>
57 #include <sys/refcount.h>
58 #include <sys/sx.h>
59 #include <sys/sysent.h>
60 #include <sys/namei.h>
61 #include <sys/mount.h>
62 #include <sys/queue.h>
63 #include <sys/socket.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/uuid.h>
67 #include <sys/vnode.h>
68 
69 #include <net/if.h>
70 #include <net/vnet.h>
71 
72 #include <netinet/in.h>
73 
74 #ifdef DDB
75 #include <ddb/ddb.h>
76 #endif /* DDB */
77 
78 #include <security/mac/mac_framework.h>
79 
80 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
81 
82 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
83 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
84 
85 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
86 #ifdef INET
87 #ifdef INET6
88 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
89 #else
90 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
91 #endif
92 #else /* !INET */
93 #ifdef INET6
94 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
95 #else
96 #define	_PR_IP_SADDRSEL	0
97 #endif
98 #endif
99 
100 /* prison0 describes what is "real" about the system. */
101 struct prison prison0 = {
102 	.pr_id		= 0,
103 	.pr_name	= "0",
104 	.pr_ref		= 1,
105 	.pr_uref	= 1,
106 	.pr_path	= "/",
107 	.pr_securelevel	= -1,
108 	.pr_devfs_rsnum = 0,
109 	.pr_state	= PRISON_STATE_ALIVE,
110 	.pr_childmax	= JAIL_MAX,
111 	.pr_hostuuid	= DEFAULT_HOSTUUID,
112 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
113 #ifdef VIMAGE
114 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
115 #else
116 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
117 #endif
118 	.pr_allow	= PR_ALLOW_PRISON0,
119 };
120 _Static_assert((PR_ALLOW_PRISON0 & ~PR_ALLOW_ALL_STATIC) == 0,
121     "Bits enabled in PR_ALLOW_PRISON0 that are not statically reserved");
122 
123 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
124 
125 struct bool_flags {
126 	const char	*name;
127 	const char	*noname;
128 	volatile u_int	 flag;
129 };
130 struct jailsys_flags {
131 	const char	*name;
132 	unsigned	 disable;
133 	unsigned	 new;
134 };
135 
136 /*
137  * Handle jail teardown in a dedicated thread to avoid deadlocks from
138  * vnet_destroy().
139  */
140 TASKQUEUE_DEFINE_THREAD(jail_remove);
141 
142 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
143 struct	sx allprison_lock;
144 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
145 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
146 LIST_HEAD(, prison_racct) allprison_racct;
147 int	lastprid = 0;
148 int	lastdeadid = 0;
149 
150 static int get_next_prid(struct prison **insprp);
151 static int get_next_deadid(struct prison **insprp);
152 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
153 static void prison_complete(void *context, int pending);
154 static void prison_deref(struct prison *pr, int flags);
155 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
156 static int prison_lock_xlock(struct prison *pr, int flags);
157 static void prison_cleanup(struct prison *pr);
158 static void prison_free_not_last(struct prison *pr);
159 static void prison_proc_free_not_last(struct prison *pr);
160 static void prison_proc_relink(struct prison *opr, struct prison *npr,
161     struct proc *p);
162 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
163     int enable);
164 static char *prison_path(struct prison *pr1, struct prison *pr2);
165 #ifdef RACCT
166 static void prison_racct_attach(struct prison *pr);
167 static void prison_racct_modify(struct prison *pr);
168 static void prison_racct_detach(struct prison *pr);
169 #endif
170 
171 /* Flags for prison_deref */
172 #define	PD_DEREF	0x01	/* Decrement pr_ref */
173 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
174 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
175 #define	PD_LOCKED	0x10	/* pr_mtx is held */
176 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
177 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
178 #define PD_OP_FLAGS	0x07	/* Operation flags */
179 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
180 
181 /*
182  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
183  * as we cannot figure out the size of a sparse array, or an array without a
184  * terminating entry.
185  */
186 static struct bool_flags pr_flag_bool[] = {
187 	{"persist", "nopersist", PR_PERSIST},
188 #ifdef INET
189 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
190 #endif
191 #ifdef INET6
192 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
193 #endif
194 };
195 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
196 
197 static struct jailsys_flags pr_flag_jailsys[] = {
198 	{"host", 0, PR_HOST},
199 #ifdef VIMAGE
200 	{"vnet", 0, PR_VNET},
201 #endif
202 #ifdef INET
203 	{"ip4", PR_IP4_USER, PR_IP4_USER},
204 #endif
205 #ifdef INET6
206 	{"ip6", PR_IP6_USER, PR_IP6_USER},
207 #endif
208 };
209 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
210 
211 /*
212  * Make this array full-size so dynamic parameters can be added.
213  * It is protected by prison0.mtx, but lockless reading is allowed
214  * with an atomic check of the flag values.
215  */
216 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
217 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
218 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
219 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
220 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
221 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
222 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
223 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
224 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
225 	{"allow.reserved_ports", "allow.noreserved_ports",
226 	 PR_ALLOW_RESERVED_PORTS},
227 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
228 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
229 	 PR_ALLOW_UNPRIV_DEBUG},
230 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
231 #ifdef VIMAGE
232 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
233 #endif
234 	{"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
235 	{"allow.adjtime", "allow.noadjtime", PR_ALLOW_ADJTIME},
236 	{"allow.settime", "allow.nosettime", PR_ALLOW_SETTIME},
237 	{"allow.routing", "allow.norouting", PR_ALLOW_ROUTING},
238 	{"allow.unprivileged_parent_tampering",
239 	    "allow.nounprivileged_parent_tampering",
240 	    PR_ALLOW_UNPRIV_PARENT_TAMPER},
241 };
242 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
243 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
244 
245 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
246 					 PR_ALLOW_RESERVED_PORTS | \
247 					 PR_ALLOW_UNPRIV_DEBUG | \
248 					 PR_ALLOW_SUSER)
249 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
250 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
251 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
252 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
253 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
254 #if defined(INET) || defined(INET6)
255 static unsigned jail_max_af_ips = 255;
256 #endif
257 
258 /*
259  * Initialize the parts of prison0 that can't be static-initialized with
260  * constants.  This is called from proc0_init() after creating thread0 cpuset.
261  */
262 void
prison0_init(void)263 prison0_init(void)
264 {
265 	uint8_t *file, *data;
266 	size_t size;
267 	char buf[sizeof(prison0.pr_hostuuid)];
268 	bool valid;
269 
270 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
271 	prison0.pr_osreldate = osreldate;
272 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
273 
274 	/* If we have a preloaded hostuuid, use it. */
275 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
276 	if (file != NULL) {
277 		data = preload_fetch_addr(file);
278 		size = preload_fetch_size(file);
279 		if (data != NULL) {
280 			/*
281 			 * The preloaded data may include trailing whitespace, almost
282 			 * certainly a newline; skip over any whitespace or
283 			 * non-printable characters to be safe.
284 			 */
285 			while (size > 0 && data[size - 1] <= 0x20) {
286 				size--;
287 			}
288 
289 			valid = false;
290 
291 			/*
292 			 * Not NUL-terminated when passed from loader, but
293 			 * validate_uuid requires that due to using sscanf (as
294 			 * does the subsequent strlcpy, since it still reads
295 			 * past the given size to return the true length);
296 			 * bounce to a temporary buffer to fix.
297 			 */
298 			if (size >= sizeof(buf))
299 				goto done;
300 
301 			memcpy(buf, data, size);
302 			buf[size] = '\0';
303 
304 			if (validate_uuid(buf, size, NULL, 0) != 0)
305 				goto done;
306 
307 			valid = true;
308 			(void)strlcpy(prison0.pr_hostuuid, buf,
309 			    sizeof(prison0.pr_hostuuid));
310 
311 done:
312 			if (bootverbose && !valid) {
313 				printf("hostuuid: preload data malformed: '%.*s'\n",
314 				    (int)size, data);
315 			}
316 		}
317 	}
318 	if (bootverbose)
319 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
320 }
321 
322 /*
323  * struct jail_args {
324  *	struct jail *jail;
325  * };
326  */
327 int
sys_jail(struct thread * td,struct jail_args * uap)328 sys_jail(struct thread *td, struct jail_args *uap)
329 {
330 	uint32_t version;
331 	int error;
332 	struct jail j;
333 
334 	error = copyin(uap->jail, &version, sizeof(uint32_t));
335 	if (error)
336 		return (error);
337 
338 	switch (version) {
339 	case 0:
340 	{
341 		struct jail_v0 j0;
342 
343 		/* FreeBSD single IPv4 jails. */
344 		bzero(&j, sizeof(struct jail));
345 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
346 		if (error)
347 			return (error);
348 		j.version = j0.version;
349 		j.path = j0.path;
350 		j.hostname = j0.hostname;
351 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
352 		break;
353 	}
354 
355 	case 1:
356 		/*
357 		 * Version 1 was used by multi-IPv4 jail implementations
358 		 * that never made it into the official kernel.
359 		 */
360 		return (EINVAL);
361 
362 	case 2:	/* JAIL_API_VERSION */
363 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
364 		error = copyin(uap->jail, &j, sizeof(struct jail));
365 		if (error)
366 			return (error);
367 		break;
368 
369 	default:
370 		/* Sci-Fi jails are not supported, sorry. */
371 		return (EINVAL);
372 	}
373 	return (kern_jail(td, &j));
374 }
375 
376 int
kern_jail(struct thread * td,struct jail * j)377 kern_jail(struct thread *td, struct jail *j)
378 {
379 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
380 #ifdef INET
381 			    + 1
382 #endif
383 #ifdef INET6
384 			    + 1
385 #endif
386 			    )];
387 	struct uio opt;
388 	char *u_path, *u_hostname, *u_name;
389 	struct bool_flags *bf;
390 #ifdef INET
391 	uint32_t ip4s;
392 	struct in_addr *u_ip4;
393 #endif
394 #ifdef INET6
395 	struct in6_addr *u_ip6;
396 #endif
397 	size_t tmplen;
398 	int error, enforce_statfs;
399 
400 	bzero(&optiov, sizeof(optiov));
401 	opt.uio_iov = optiov;
402 	opt.uio_iovcnt = 0;
403 	opt.uio_offset = -1;
404 	opt.uio_resid = -1;
405 	opt.uio_segflg = UIO_SYSSPACE;
406 	opt.uio_rw = UIO_READ;
407 	opt.uio_td = td;
408 
409 	/* Set permissions for top-level jails from sysctls. */
410 	if (!jailed(td->td_ucred)) {
411 		for (bf = pr_flag_allow;
412 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
413 			atomic_load_int(&bf->flag) != 0;
414 		     bf++) {
415 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
416 			    (jail_default_allow & bf->flag)
417 			    ? bf->name : bf->noname);
418 			optiov[opt.uio_iovcnt].iov_len =
419 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
420 			opt.uio_iovcnt += 2;
421 		}
422 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
423 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
424 		opt.uio_iovcnt++;
425 		enforce_statfs = jail_default_enforce_statfs;
426 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
427 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
428 		opt.uio_iovcnt++;
429 	}
430 
431 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
432 #ifdef INET
433 	ip4s = (j->version == 0) ? 1 : j->ip4s;
434 	if (ip4s > jail_max_af_ips)
435 		return (EINVAL);
436 	tmplen += ip4s * sizeof(struct in_addr);
437 #else
438 	if (j->ip4s > 0)
439 		return (EINVAL);
440 #endif
441 #ifdef INET6
442 	if (j->ip6s > jail_max_af_ips)
443 		return (EINVAL);
444 	tmplen += j->ip6s * sizeof(struct in6_addr);
445 #else
446 	if (j->ip6s > 0)
447 		return (EINVAL);
448 #endif
449 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
450 	u_hostname = u_path + MAXPATHLEN;
451 	u_name = u_hostname + MAXHOSTNAMELEN;
452 #ifdef INET
453 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
454 #endif
455 #ifdef INET6
456 #ifdef INET
457 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
458 #else
459 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
460 #endif
461 #endif
462 	optiov[opt.uio_iovcnt].iov_base = "path";
463 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
464 	opt.uio_iovcnt++;
465 	optiov[opt.uio_iovcnt].iov_base = u_path;
466 	error = copyinstr(j->path, u_path, MAXPATHLEN,
467 	    &optiov[opt.uio_iovcnt].iov_len);
468 	if (error) {
469 		free(u_path, M_TEMP);
470 		return (error);
471 	}
472 	opt.uio_iovcnt++;
473 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
474 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
475 	opt.uio_iovcnt++;
476 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
477 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
478 	    &optiov[opt.uio_iovcnt].iov_len);
479 	if (error) {
480 		free(u_path, M_TEMP);
481 		return (error);
482 	}
483 	opt.uio_iovcnt++;
484 	if (j->jailname != NULL) {
485 		optiov[opt.uio_iovcnt].iov_base = "name";
486 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
487 		opt.uio_iovcnt++;
488 		optiov[opt.uio_iovcnt].iov_base = u_name;
489 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
490 		    &optiov[opt.uio_iovcnt].iov_len);
491 		if (error) {
492 			free(u_path, M_TEMP);
493 			return (error);
494 		}
495 		opt.uio_iovcnt++;
496 	}
497 #ifdef INET
498 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
499 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
500 	opt.uio_iovcnt++;
501 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
502 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
503 	if (j->version == 0)
504 		u_ip4->s_addr = j->ip4s;
505 	else {
506 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
507 		if (error) {
508 			free(u_path, M_TEMP);
509 			return (error);
510 		}
511 	}
512 	opt.uio_iovcnt++;
513 #endif
514 #ifdef INET6
515 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
516 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
517 	opt.uio_iovcnt++;
518 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
519 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
520 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
521 	if (error) {
522 		free(u_path, M_TEMP);
523 		return (error);
524 	}
525 	opt.uio_iovcnt++;
526 #endif
527 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
528 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
529 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
530 	free(u_path, M_TEMP);
531 	return (error);
532 }
533 
534 /*
535  * struct jail_set_args {
536  *	struct iovec *iovp;
537  *	unsigned int iovcnt;
538  *	int flags;
539  * };
540  */
541 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)542 sys_jail_set(struct thread *td, struct jail_set_args *uap)
543 {
544 	struct uio *auio;
545 	int error;
546 
547 	/* Check that we have an even number of iovecs. */
548 	if (uap->iovcnt & 1)
549 		return (EINVAL);
550 
551 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
552 	if (error)
553 		return (error);
554 	error = kern_jail_set(td, auio, uap->flags);
555 	freeuio(auio);
556 	return (error);
557 }
558 
559 #if defined(INET) || defined(INET6)
560 typedef int prison_addr_cmp_t(const void *, const void *);
561 typedef bool prison_addr_valid_t(const void *);
562 static const struct pr_family {
563 	size_t			size;
564 	prison_addr_cmp_t	*cmp;
565 	prison_addr_valid_t	*valid;
566 	int			ip_flag;
567 } pr_families[PR_FAMILY_MAX] = {
568 #ifdef INET
569 	[PR_INET] = {
570 		.size = sizeof(struct in_addr),
571 		.cmp = prison_qcmp_v4,
572 		.valid = prison_valid_v4,
573 		.ip_flag = PR_IP4_USER,
574 	 },
575 #endif
576 #ifdef INET6
577 	[PR_INET6] = {
578 		.size = sizeof(struct in6_addr),
579 		.cmp = prison_qcmp_v6,
580 		.valid = prison_valid_v6,
581 		.ip_flag = PR_IP6_USER,
582 	},
583 #endif
584 };
585 
586 /*
587  * Network address lists (pr_addrs) allocation for jails.  The addresses
588  * are accessed locklessly by the network stack, thus need to be protected by
589  * the network epoch.
590  */
591 struct prison_ip {
592 	struct epoch_context ctx;
593 	uint32_t	ips;
594 #ifdef FUTURE_C
595 	/*
596 	 * XXX Variable-length automatic arrays in union may be
597 	 * supported in future C.
598 	 */
599 	union {
600 		char pr_ip[];
601 		struct in_addr pr_ip4[];
602 		struct in6_addr pr_ip6[];
603 	};
604 #else /* No future C :( */
605 	char pr_ip[];
606 #endif
607 };
608 
609 static char *
PR_IP(struct prison_ip * pip,const pr_family_t af,int idx)610 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
611 {
612 	MPASS(pip);
613 	MPASS(af < PR_FAMILY_MAX);
614 	MPASS(idx >= 0 && idx < pip->ips);
615 
616 	return (pip->pr_ip + pr_families[af].size * idx);
617 }
618 
619 static struct prison_ip *
prison_ip_alloc(const pr_family_t af,uint32_t cnt,int flags)620 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
621 {
622 	struct prison_ip *pip;
623 
624 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
625 	    M_PRISON, flags);
626 	if (pip != NULL)
627 		pip->ips = cnt;
628 	return (pip);
629 }
630 
631 /*
632  * Allocate and copyin user supplied address list, sorting and validating.
633  * kern_jail_set() helper.
634  */
635 static struct prison_ip *
prison_ip_copyin(const pr_family_t af,void * op,uint32_t cnt)636 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
637 {
638 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
639 	const size_t size = pr_families[af].size;
640 	struct prison_ip *pip;
641 
642 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
643 	bcopy(op, pip->pr_ip, cnt * size);
644 	/*
645 	 * IP addresses are all sorted but ip[0] to preserve
646 	 * the primary IP address as given from userland.
647 	 * This special IP is used for unbound outgoing
648 	 * connections as well for "loopback" traffic in case
649 	 * source address selection cannot find any more fitting
650 	 * address to connect from.
651 	 */
652 	if (cnt > 1)
653 		qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
654 	/*
655 	 * Check for duplicate addresses and do some simple
656 	 * zero and broadcast checks. If users give other bogus
657 	 * addresses it is their problem.
658 	 */
659 	for (int i = 0; i < cnt; i++) {
660 		if (!pr_families[af].valid(PR_IP(pip, af, i))) {
661 			free(pip, M_PRISON);
662 			return (NULL);
663 		}
664 		if (i + 1 < cnt &&
665 		    (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
666 		     cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
667 			free(pip, M_PRISON);
668 			return (NULL);
669 		}
670 	}
671 
672 	return (pip);
673 }
674 
675 /*
676  * Allocate and dup parent prison address list.
677  * kern_jail_set() helper.
678  */
679 static void
prison_ip_dup(struct prison * ppr,struct prison * pr,const pr_family_t af)680 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
681 {
682 	const struct prison_ip *ppip = ppr->pr_addrs[af];
683 	struct prison_ip *pip;
684 
685 	if (ppip != NULL) {
686 		pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
687 		bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
688 		pr->pr_addrs[af] = pip;
689 	}
690 }
691 
692 /*
693  * Make sure the new set of IP addresses is a subset of the parent's list.
694  * Don't worry about the parent being unlocked, as any setting is done with
695  * allprison_lock held.
696  * kern_jail_set() helper.
697  */
698 static bool
prison_ip_parent_match(struct prison_ip * ppip,struct prison_ip * pip,const pr_family_t af)699 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
700     const pr_family_t af)
701 {
702 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
703 	int i, j;
704 
705 	if (ppip == NULL)
706 		return (false);
707 
708 	for (i = 0; i < ppip->ips; i++)
709 		if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
710 			break;
711 
712 	if (i == ppip->ips)
713 		/* Main address not present in parent. */
714 		return (false);
715 
716 	if (pip->ips > 1) {
717 		for (i = j = 1; i < pip->ips; i++) {
718 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
719 				/* Equals to parent primary address. */
720 				continue;
721 			for (; j < ppip->ips; j++)
722 				if (cmp(PR_IP(pip, af, i),
723 				    PR_IP(ppip, af, j)) == 0)
724 					break;
725 			if (j == ppip->ips)
726 				break;
727 		}
728 		if (j == ppip->ips)
729 			/* Address not present in parent. */
730 			return (false);
731 	}
732 	return (true);
733 }
734 
735 /*
736  * Check for conflicting IP addresses.  We permit them if there is no more
737  * than one IP on each jail.  If there is a duplicate on a jail with more
738  * than one IP stop checking and return error.
739  * kern_jail_set() helper.
740  */
741 static bool
prison_ip_conflict_check(const struct prison * ppr,const struct prison * pr,struct prison_ip * pip,pr_family_t af)742 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
743     struct prison_ip *pip, pr_family_t af)
744 {
745 	const struct prison *tppr, *tpr;
746 	int descend;
747 
748 #ifdef VIMAGE
749 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
750 		if (tppr->pr_flags & PR_VNET)
751 			break;
752 #else
753 	tppr = &prison0;
754 #endif
755 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
756 		if (tpr == pr ||
757 #ifdef VIMAGE
758 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
759 #endif
760 		    !prison_isalive(tpr)) {
761 			descend = 0;
762 			continue;
763 		}
764 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
765 			continue;
766 		descend = 0;
767 		if (tpr->pr_addrs[af] == NULL ||
768 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
769 			continue;
770 		for (int i = 0; i < pip->ips; i++)
771 			if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
772 				return (false);
773 	}
774 
775 	return (true);
776 }
777 
778 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
779     "prison must start with epoch context");
780 static void
prison_ip_free_deferred(epoch_context_t ctx)781 prison_ip_free_deferred(epoch_context_t ctx)
782 {
783 
784 	free(ctx, M_PRISON);
785 }
786 
787 static void
prison_ip_free(struct prison_ip * pip)788 prison_ip_free(struct prison_ip *pip)
789 {
790 
791 	if (pip != NULL)
792 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
793 }
794 
795 static void
prison_ip_set(struct prison * pr,const pr_family_t af,struct prison_ip * new)796 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
797 {
798 	struct prison_ip **mem, *old;
799 
800 	mtx_assert(&pr->pr_mtx, MA_OWNED);
801 
802 	mem = &pr->pr_addrs[af];
803 
804 	old = *mem;
805 	atomic_store_ptr(mem, new);
806 	prison_ip_free(old);
807 }
808 
809 /*
810  * Restrict a prison's IP address list with its parent's, possibly replacing
811  * it.  Return true if succeed, otherwise should redo.
812  * kern_jail_set() helper.
813  */
814 static bool
prison_ip_restrict(struct prison * pr,const pr_family_t af,struct prison_ip ** newp)815 prison_ip_restrict(struct prison *pr, const pr_family_t af,
816     struct prison_ip **newp)
817 {
818 	struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
819 	struct prison_ip *pip = pr->pr_addrs[af];
820 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
821 	const size_t size = pr_families[af].size;
822 	struct prison_ip *new = newp != NULL ? *newp : NULL;
823 	uint32_t ips;
824 
825 	mtx_assert(&pr->pr_mtx, MA_OWNED);
826 
827 	/*
828 	 * Due to epoch-synchronized access to the IP address lists we always
829 	 * allocate a new list even if the old one has enough space.  We could
830 	 * atomically update an IPv4 address inside a list, but that would
831 	 * screw up sorting, and in case of IPv6 we can't even atomically write
832 	 * one.
833 	 */
834 	if (ppip == NULL) {
835 		if (pip != NULL)
836 			prison_ip_set(pr, af, NULL);
837 		return (true);
838 	}
839 
840 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
841 		if (new == NULL) {
842 			new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
843 			if (new == NULL)
844 				return (false); /* Redo */
845 		}
846 		/* This has no user settings, so just copy the parent's list. */
847 		MPASS(new->ips == ppip->ips);
848 		bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
849 		prison_ip_set(pr, af, new);
850 		if (newp != NULL)
851 			*newp = NULL; /* Used */
852 	} else if (pip != NULL) {
853 		/* Remove addresses that aren't in the parent. */
854 		int i;
855 
856 		i = 0; /* index in pip */
857 		ips = 0; /* index in new */
858 
859 		if (new == NULL) {
860 			new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
861 			if (new == NULL)
862 				return (false); /* Redo */
863 		}
864 
865 		for (int pi = 0; pi < ppip->ips; pi++)
866 			if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
867 				/* Found our primary address in parent. */
868 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
869 				    size);
870 				i++;
871 				ips++;
872 				break;
873 			}
874 		for (int pi = 1; i < pip->ips; ) {
875 			/* Check against primary, which is unsorted. */
876 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
877 				/* Matches parent's primary address. */
878 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
879 				    size);
880 				i++;
881 				ips++;
882 				continue;
883 			}
884 			/* The rest are sorted. */
885 			switch (pi >= ppip->ips ? -1 :
886 				cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
887 			case -1:
888 				i++;
889 				break;
890 			case 0:
891 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
892 				    size);
893 				i++;
894 				pi++;
895 				ips++;
896 				break;
897 			case 1:
898 				pi++;
899 				break;
900 			}
901 		}
902 		if (ips == 0) {
903 			if (newp == NULL || *newp == NULL)
904 				prison_ip_free(new);
905 			new = NULL;
906 		} else {
907 			/* Shrink to real size */
908 			KASSERT((new->ips >= ips),
909 			    ("Out-of-bounds write to prison_ip %p", new));
910 			new->ips = ips;
911 		}
912 		prison_ip_set(pr, af, new);
913 		if (newp != NULL)
914 			*newp = NULL; /* Used */
915 	}
916 	return (true);
917 }
918 
919 /*
920  * Fast-path check if an address belongs to a prison.
921  */
922 int
prison_ip_check(const struct prison * pr,const pr_family_t af,const void * addr)923 prison_ip_check(const struct prison *pr, const pr_family_t af,
924     const void *addr)
925 {
926 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
927 	struct prison_ip *pip;
928 	int i, a, z, d;
929 
930 	MPASS(mtx_owned(&pr->pr_mtx) ||
931 	    in_epoch(net_epoch_preempt) ||
932 	    sx_xlocked(&allprison_lock));
933 
934 	pip = atomic_load_ptr(&pr->pr_addrs[af]);
935 	if (__predict_false(pip == NULL))
936 		return (EAFNOSUPPORT);
937 
938 	/* Check the primary IP. */
939 	if (cmp(PR_IP(pip, af, 0), addr) == 0)
940 		return (0);
941 
942 	/*
943 	 * All the other IPs are sorted so we can do a binary search.
944 	 */
945 	a = 0;
946 	z = pip->ips - 2;
947 	while (a <= z) {
948 		i = (a + z) / 2;
949 		d = cmp(PR_IP(pip, af, i + 1), addr);
950 		if (d > 0)
951 			z = i - 1;
952 		else if (d < 0)
953 			a = i + 1;
954 		else
955 			return (0);
956 	}
957 
958 	return (EADDRNOTAVAIL);
959 }
960 
961 /*
962  * Grab primary IP.  Historically required mutex, but nothing prevents
963  * us to support epoch-protected access.  Is it used in fast path?
964  * in{6}_jail.c helper
965  */
966 const void *
prison_ip_get0(const struct prison * pr,const pr_family_t af)967 prison_ip_get0(const struct prison *pr, const pr_family_t af)
968 {
969 	const struct prison_ip *pip = pr->pr_addrs[af];
970 
971 	mtx_assert(&pr->pr_mtx, MA_OWNED);
972 	MPASS(pip);
973 
974 	return (pip->pr_ip);
975 }
976 
977 u_int
prison_ip_cnt(const struct prison * pr,const pr_family_t af)978 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
979 {
980 
981 	return (pr->pr_addrs[af]->ips);
982 }
983 #endif	/* defined(INET) || defined(INET6) */
984 
985 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)986 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
987 {
988 	struct nameidata nd;
989 #ifdef INET
990 	struct prison_ip *ip4;
991 #endif
992 #ifdef INET6
993 	struct prison_ip *ip6;
994 #endif
995 	struct vfsopt *opt;
996 	struct vfsoptlist *opts;
997 	struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
998 	struct vnode *root;
999 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
1000 	char *g_path, *osrelstr;
1001 	struct bool_flags *bf;
1002 	struct jailsys_flags *jsf;
1003 #if defined(INET) || defined(INET6)
1004 	void *op;
1005 #endif
1006 	unsigned long hid;
1007 	size_t namelen, onamelen, pnamelen;
1008 	int created, cuflags, descend, drflags, enforce;
1009 	int error, errmsg_len, errmsg_pos;
1010 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1011 	int deadid, jid, jsys, len, level;
1012 	int childmax, osreldt, rsnum, slevel;
1013 #ifdef INET
1014 	int ip4s;
1015 	bool redo_ip4;
1016 #endif
1017 #ifdef INET6
1018 	int ip6s;
1019 	bool redo_ip6;
1020 #endif
1021 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1022 	uint64_t pr_allow_diff;
1023 	unsigned tallow;
1024 	char numbuf[12];
1025 
1026 	error = priv_check(td, PRIV_JAIL_SET);
1027 	if (!error && (flags & JAIL_ATTACH))
1028 		error = priv_check(td, PRIV_JAIL_ATTACH);
1029 	if (error)
1030 		return (error);
1031 	mypr = td->td_ucred->cr_prison;
1032 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
1033 		return (EPERM);
1034 	if (flags & ~JAIL_SET_MASK)
1035 		return (EINVAL);
1036 
1037 	/*
1038 	 * Check all the parameters before committing to anything.  Not all
1039 	 * errors can be caught early, but we may as well try.  Also, this
1040 	 * takes care of some expensive stuff (path lookup) before getting
1041 	 * the allprison lock.
1042 	 *
1043 	 * XXX Jails are not filesystems, and jail parameters are not mount
1044 	 *     options.  But it makes more sense to re-use the vfsopt code
1045 	 *     than duplicate it under a different name.
1046 	 */
1047 	error = vfs_buildopts(optuio, &opts);
1048 	if (error)
1049 		return (error);
1050 #ifdef INET
1051 	ip4 = NULL;
1052 #endif
1053 #ifdef INET6
1054 	ip6 = NULL;
1055 #endif
1056 	g_path = NULL;
1057 
1058 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1059 	if (!cuflags) {
1060 		error = EINVAL;
1061 		vfs_opterror(opts, "no valid operation (create or update)");
1062 		goto done_errmsg;
1063 	}
1064 
1065 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1066 	if (error == ENOENT)
1067 		jid = 0;
1068 	else if (error != 0)
1069 		goto done_free;
1070 
1071 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1072 	if (error == ENOENT)
1073 		gotslevel = 0;
1074 	else if (error != 0)
1075 		goto done_free;
1076 	else
1077 		gotslevel = 1;
1078 
1079 	error =
1080 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1081 	if (error == ENOENT)
1082 		gotchildmax = 0;
1083 	else if (error != 0)
1084 		goto done_free;
1085 	else
1086 		gotchildmax = 1;
1087 
1088 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1089 	if (error == ENOENT)
1090 		gotenforce = 0;
1091 	else if (error != 0)
1092 		goto done_free;
1093 	else if (enforce < 0 || enforce > 2) {
1094 		error = EINVAL;
1095 		goto done_free;
1096 	} else
1097 		gotenforce = 1;
1098 
1099 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1100 	if (error == ENOENT)
1101 		gotrsnum = 0;
1102 	else if (error != 0)
1103 		goto done_free;
1104 	else
1105 		gotrsnum = 1;
1106 
1107 	pr_flags = ch_flags = 0;
1108 	for (bf = pr_flag_bool;
1109 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1110 	     bf++) {
1111 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1112 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1113 	}
1114 	ch_flags |= pr_flags;
1115 	for (jsf = pr_flag_jailsys;
1116 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1117 	     jsf++) {
1118 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1119 		if (error == ENOENT)
1120 			continue;
1121 		if (error != 0)
1122 			goto done_free;
1123 		switch (jsys) {
1124 		case JAIL_SYS_DISABLE:
1125 			if (!jsf->disable) {
1126 				error = EINVAL;
1127 				goto done_free;
1128 			}
1129 			pr_flags |= jsf->disable;
1130 			break;
1131 		case JAIL_SYS_NEW:
1132 			pr_flags |= jsf->new;
1133 			break;
1134 		case JAIL_SYS_INHERIT:
1135 			break;
1136 		default:
1137 			error = EINVAL;
1138 			goto done_free;
1139 		}
1140 		ch_flags |= jsf->new | jsf->disable;
1141 	}
1142 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1143 	    && !(pr_flags & PR_PERSIST)) {
1144 		error = EINVAL;
1145 		vfs_opterror(opts, "new jail must persist or attach");
1146 		goto done_errmsg;
1147 	}
1148 #ifdef VIMAGE
1149 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1150 		error = EINVAL;
1151 		vfs_opterror(opts, "vnet cannot be changed after creation");
1152 		goto done_errmsg;
1153 	}
1154 #endif
1155 #ifdef INET
1156 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1157 		error = EINVAL;
1158 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1159 		goto done_errmsg;
1160 	}
1161 #endif
1162 #ifdef INET6
1163 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1164 		error = EINVAL;
1165 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1166 		goto done_errmsg;
1167 	}
1168 #endif
1169 
1170 	pr_allow = ch_allow = 0;
1171 	for (bf = pr_flag_allow;
1172 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1173 		atomic_load_int(&bf->flag) != 0;
1174 	     bf++) {
1175 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1176 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1177 	}
1178 	ch_allow |= pr_allow;
1179 
1180 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1181 	if (error == ENOENT)
1182 		name = NULL;
1183 	else if (error != 0)
1184 		goto done_free;
1185 	else {
1186 		if (len == 0 || name[len - 1] != '\0') {
1187 			error = EINVAL;
1188 			goto done_free;
1189 		}
1190 		if (len > MAXHOSTNAMELEN) {
1191 			error = ENAMETOOLONG;
1192 			goto done_free;
1193 		}
1194 	}
1195 
1196 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1197 	if (error == ENOENT)
1198 		host = NULL;
1199 	else if (error != 0)
1200 		goto done_free;
1201 	else {
1202 		ch_flags |= PR_HOST;
1203 		pr_flags |= PR_HOST;
1204 		if (len == 0 || host[len - 1] != '\0') {
1205 			error = EINVAL;
1206 			goto done_free;
1207 		}
1208 		if (len > MAXHOSTNAMELEN) {
1209 			error = ENAMETOOLONG;
1210 			goto done_free;
1211 		}
1212 	}
1213 
1214 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1215 	if (error == ENOENT)
1216 		domain = NULL;
1217 	else if (error != 0)
1218 		goto done_free;
1219 	else {
1220 		ch_flags |= PR_HOST;
1221 		pr_flags |= PR_HOST;
1222 		if (len == 0 || domain[len - 1] != '\0') {
1223 			error = EINVAL;
1224 			goto done_free;
1225 		}
1226 		if (len > MAXHOSTNAMELEN) {
1227 			error = ENAMETOOLONG;
1228 			goto done_free;
1229 		}
1230 	}
1231 
1232 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1233 	if (error == ENOENT)
1234 		uuid = NULL;
1235 	else if (error != 0)
1236 		goto done_free;
1237 	else {
1238 		ch_flags |= PR_HOST;
1239 		pr_flags |= PR_HOST;
1240 		if (len == 0 || uuid[len - 1] != '\0') {
1241 			error = EINVAL;
1242 			goto done_free;
1243 		}
1244 		if (len > HOSTUUIDLEN) {
1245 			error = ENAMETOOLONG;
1246 			goto done_free;
1247 		}
1248 	}
1249 
1250 #ifdef COMPAT_FREEBSD32
1251 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1252 		uint32_t hid32;
1253 
1254 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1255 		hid = hid32;
1256 	} else
1257 #endif
1258 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1259 	if (error == ENOENT)
1260 		gothid = 0;
1261 	else if (error != 0)
1262 		goto done_free;
1263 	else {
1264 		gothid = 1;
1265 		ch_flags |= PR_HOST;
1266 		pr_flags |= PR_HOST;
1267 	}
1268 
1269 #ifdef INET
1270 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1271 	if (error == ENOENT)
1272 		ip4s = 0;
1273 	else if (error != 0)
1274 		goto done_free;
1275 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1276 		error = EINVAL;
1277 		goto done_free;
1278 	} else {
1279 		ch_flags |= PR_IP4_USER;
1280 		pr_flags |= PR_IP4_USER;
1281 		if (ip4s > 0) {
1282 			ip4s /= sizeof(struct in_addr);
1283 			if (ip4s > jail_max_af_ips) {
1284 				error = EINVAL;
1285 				vfs_opterror(opts, "too many IPv4 addresses");
1286 				goto done_errmsg;
1287 			}
1288 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1289 			if (ip4 == NULL) {
1290 				error = EINVAL;
1291 				goto done_free;
1292 			}
1293 		}
1294 	}
1295 #endif
1296 
1297 #ifdef INET6
1298 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1299 	if (error == ENOENT)
1300 		ip6s = 0;
1301 	else if (error != 0)
1302 		goto done_free;
1303 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1304 		error = EINVAL;
1305 		goto done_free;
1306 	} else {
1307 		ch_flags |= PR_IP6_USER;
1308 		pr_flags |= PR_IP6_USER;
1309 		if (ip6s > 0) {
1310 			ip6s /= sizeof(struct in6_addr);
1311 			if (ip6s > jail_max_af_ips) {
1312 				error = EINVAL;
1313 				vfs_opterror(opts, "too many IPv6 addresses");
1314 				goto done_errmsg;
1315 			}
1316 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1317 			if (ip6 == NULL) {
1318 				error = EINVAL;
1319 				goto done_free;
1320 			}
1321 		}
1322 	}
1323 #endif
1324 
1325 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1326 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1327 		error = EINVAL;
1328 		vfs_opterror(opts,
1329 		    "vnet jails cannot have IP address restrictions");
1330 		goto done_errmsg;
1331 	}
1332 #endif
1333 
1334 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1335 	if (error == ENOENT)
1336 		osrelstr = NULL;
1337 	else if (error != 0)
1338 		goto done_free;
1339 	else {
1340 		if (flags & JAIL_UPDATE) {
1341 			error = EINVAL;
1342 			vfs_opterror(opts,
1343 			    "osrelease cannot be changed after creation");
1344 			goto done_errmsg;
1345 		}
1346 		if (len == 0 || osrelstr[len - 1] != '\0') {
1347 			error = EINVAL;
1348 			goto done_free;
1349 		}
1350 		if (len >= OSRELEASELEN) {
1351 			error = ENAMETOOLONG;
1352 			vfs_opterror(opts,
1353 			    "osrelease string must be 1-%d bytes long",
1354 			    OSRELEASELEN - 1);
1355 			goto done_errmsg;
1356 		}
1357 	}
1358 
1359 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1360 	if (error == ENOENT)
1361 		osreldt = 0;
1362 	else if (error != 0)
1363 		goto done_free;
1364 	else {
1365 		if (flags & JAIL_UPDATE) {
1366 			error = EINVAL;
1367 			vfs_opterror(opts,
1368 			    "osreldate cannot be changed after creation");
1369 			goto done_errmsg;
1370 		}
1371 		if (osreldt == 0) {
1372 			error = EINVAL;
1373 			vfs_opterror(opts, "osreldate cannot be 0");
1374 			goto done_errmsg;
1375 		}
1376 	}
1377 
1378 	root = NULL;
1379 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1380 	if (error == ENOENT)
1381 		path = NULL;
1382 	else if (error != 0)
1383 		goto done_free;
1384 	else {
1385 		if (flags & JAIL_UPDATE) {
1386 			error = EINVAL;
1387 			vfs_opterror(opts,
1388 			    "path cannot be changed after creation");
1389 			goto done_errmsg;
1390 		}
1391 		if (len == 0 || path[len - 1] != '\0') {
1392 			error = EINVAL;
1393 			goto done_free;
1394 		}
1395 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1396 		error = namei(&nd);
1397 		if (error)
1398 			goto done_free;
1399 		root = nd.ni_vp;
1400 		NDFREE_PNBUF(&nd);
1401 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1402 		strlcpy(g_path, path, MAXPATHLEN);
1403 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1404 		if (error == 0) {
1405 			path = g_path;
1406 		} else {
1407 			/* exit on other errors */
1408 			goto done_free;
1409 		}
1410 		if (root->v_type != VDIR) {
1411 			error = ENOTDIR;
1412 			vput(root);
1413 			goto done_free;
1414 		}
1415 		VOP_UNLOCK(root);
1416 	}
1417 
1418 	/*
1419 	 * Find the specified jail, or at least its parent.
1420 	 * This abuses the file error codes ENOENT and EEXIST.
1421 	 */
1422 	pr = NULL;
1423 	inspr = NULL;
1424 	deadpr = NULL;
1425 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1426 		namelc = strrchr(name, '.');
1427 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1428 		if (*p != '\0')
1429 			jid = 0;
1430 	}
1431 	sx_xlock(&allprison_lock);
1432 	drflags = PD_LIST_XLOCKED;
1433 	ppr = mypr;
1434 	if (!prison_isalive(ppr)) {
1435 		/* This jail is dying.  This process will surely follow. */
1436 		error = EAGAIN;
1437 		goto done_deref;
1438 	}
1439 	if (jid != 0) {
1440 		if (jid < 0) {
1441 			error = EINVAL;
1442 			vfs_opterror(opts, "negative jid");
1443 			goto done_deref;
1444 		}
1445 		/*
1446 		 * See if a requested jid already exists.  Keep track of
1447 		 * where it can be inserted later.
1448 		 */
1449 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1450 			if (inspr->pr_id < jid)
1451 				continue;
1452 			if (inspr->pr_id > jid)
1453 				break;
1454 			if (prison_isalive(inspr)) {
1455 				pr = inspr;
1456 				mtx_lock(&pr->pr_mtx);
1457 				drflags |= PD_LOCKED;
1458 			} else {
1459 				/* Note a dying jail to handle later. */
1460 				deadpr = inspr;
1461 			}
1462 			inspr = NULL;
1463 			break;
1464 		}
1465 		if (cuflags == JAIL_CREATE && pr != NULL) {
1466 			/*
1467 			 * Even creators that cannot see the jail will
1468 			 * get EEXIST.
1469 			 */
1470 			error = EEXIST;
1471 			vfs_opterror(opts, "jail %d already exists", jid);
1472 			goto done_deref;
1473 		}
1474 		if ((pr == NULL)
1475 		    ? cuflags == JAIL_UPDATE
1476 		    : !prison_ischild(mypr, pr)) {
1477 			/*
1478 			 * Updaters get ENOENT for nonexistent jails,
1479 			 * or for jails they cannot see.  The latter
1480 			 * case is true even for CREATE | UPDATE,
1481 			 * which normally cannot give this error.
1482 			 */
1483 			error = ENOENT;
1484 			vfs_opterror(opts, "jail %d not found", jid);
1485 			goto done_deref;
1486 		}
1487 	}
1488 	/*
1489 	 * If the caller provided a name, look for a jail by that name.
1490 	 * This has different semantics for creates and updates keyed by jid
1491 	 * (where the name must not already exist in a different jail),
1492 	 * and updates keyed by the name itself (where the name must exist
1493 	 * because that is the jail being updated).
1494 	 */
1495 	namelc = NULL;
1496 	if (name != NULL) {
1497 		namelc = strrchr(name, '.');
1498 		if (namelc == NULL)
1499 			namelc = name;
1500 		else {
1501 			/*
1502 			 * This is a hierarchical name.  Split it into the
1503 			 * parent and child names, and make sure the parent
1504 			 * exists or matches an already found jail.
1505 			 */
1506 			if (pr != NULL) {
1507 				if (strncmp(name, ppr->pr_name, namelc - name)
1508 				    || ppr->pr_name[namelc - name] != '\0') {
1509 					error = EINVAL;
1510 					vfs_opterror(opts,
1511 					    "cannot change jail's parent");
1512 					goto done_deref;
1513 				}
1514 			} else {
1515 				*namelc = '\0';
1516 				ppr = prison_find_name(mypr, name);
1517 				if (ppr == NULL) {
1518 					error = ENOENT;
1519 					vfs_opterror(opts,
1520 					    "jail \"%s\" not found", name);
1521 					goto done_deref;
1522 				}
1523 				mtx_unlock(&ppr->pr_mtx);
1524 				if (!prison_isalive(ppr)) {
1525 					error = ENOENT;
1526 					vfs_opterror(opts,
1527 					    "jail \"%s\" is dying", name);
1528 					goto done_deref;
1529 				}
1530 				*namelc = '.';
1531 			}
1532 			namelc++;
1533 		}
1534 		if (namelc[0] != '\0') {
1535 			pnamelen =
1536 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1537 			FOREACH_PRISON_CHILD(ppr, tpr) {
1538 				if (tpr == pr || !prison_isalive(tpr) ||
1539 				    strcmp(tpr->pr_name + pnamelen, namelc))
1540 					continue;
1541 				if (cuflags == JAIL_CREATE || pr != NULL) {
1542 					/*
1543 					 * Create, or update(jid): name must
1544 					 * not exist in an active sibling jail.
1545 					 */
1546 					error = EEXIST;
1547 					vfs_opterror(opts,
1548 					    "jail \"%s\" already exists", name);
1549 					goto done_deref;
1550 				}
1551 				/* Use this jail for updates. */
1552 				pr = tpr;
1553 				mtx_lock(&pr->pr_mtx);
1554 				drflags |= PD_LOCKED;
1555 				break;
1556 			}
1557 			/*
1558 			 * Update: name must exist if no jid is specified.
1559 			 * As with the jid case, the jail must be currently
1560 			 * visible, or else even CREATE | UPDATE will get
1561 			 * an error.
1562 			 */
1563 			if ((pr == NULL)
1564 			    ? cuflags == JAIL_UPDATE
1565 			    : !prison_isalive(pr)) {
1566 				error = ENOENT;
1567 				vfs_opterror(opts, "jail \"%s\" not found",
1568 				    name);
1569 				goto done_deref;
1570 			}
1571 		}
1572 	}
1573 	/* Update: must provide a jid or name. */
1574 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1575 		error = ENOENT;
1576 		vfs_opterror(opts, "update specified no jail");
1577 		goto done_deref;
1578 	}
1579 
1580 	/* If there's no prison to update, create a new one and link it in. */
1581 	created = pr == NULL;
1582 	if (created) {
1583 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1584 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1585 				error = EPERM;
1586 				vfs_opterror(opts, "prison limit exceeded");
1587 				goto done_deref;
1588 			}
1589 
1590 		if (deadpr != NULL) {
1591 			/*
1592 			 * The prison being created has the same ID as a dying
1593 			 * one.  Handle this by giving the dying jail a new ID.
1594 			 * This may cause some confusion to user space, but
1595 			 * only to those listing dying jails.
1596 			 */
1597 			deadid = get_next_deadid(&dinspr);
1598 			if (deadid == 0) {
1599 				error = EAGAIN;
1600 				vfs_opterror(opts, "no available jail IDs");
1601 				goto done_deref;
1602 			}
1603 			mtx_lock(&deadpr->pr_mtx);
1604 			deadpr->pr_id = deadid;
1605 			mtx_unlock(&deadpr->pr_mtx);
1606 			if (dinspr == deadpr)
1607 				inspr = deadpr;
1608 			else {
1609 				inspr = TAILQ_NEXT(deadpr, pr_list);
1610 				TAILQ_REMOVE(&allprison, deadpr, pr_list);
1611 				if (dinspr != NULL)
1612 					TAILQ_INSERT_AFTER(&allprison, dinspr,
1613 					    deadpr, pr_list);
1614 				else
1615 					TAILQ_INSERT_HEAD(&allprison, deadpr,
1616 					    pr_list);
1617 			}
1618 		}
1619 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1620 			error = EAGAIN;
1621 			vfs_opterror(opts, "no available jail IDs");
1622 			goto done_deref;
1623 		}
1624 
1625 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1626 		pr->pr_state = PRISON_STATE_INVALID;
1627 		refcount_init(&pr->pr_ref, 1);
1628 		refcount_init(&pr->pr_uref, 0);
1629 		drflags |= PD_DEREF;
1630 		LIST_INIT(&pr->pr_children);
1631 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1632 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1633 
1634 		pr->pr_id = jid;
1635 		if (inspr != NULL)
1636 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1637 		else
1638 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1639 
1640 		pr->pr_parent = ppr;
1641 		prison_hold(ppr);
1642 		prison_proc_hold(ppr);
1643 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1644 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1645 			tpr->pr_childcount++;
1646 
1647 		/* Set some default values, and inherit some from the parent. */
1648 		if (namelc == NULL)
1649 			namelc = "";
1650 		if (path == NULL) {
1651 			path = "/";
1652 			root = mypr->pr_root;
1653 			vref(root);
1654 		}
1655 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1656 		pr->pr_flags |= PR_HOST;
1657 #if defined(INET) || defined(INET6)
1658 #ifdef VIMAGE
1659 		if (!(pr_flags & PR_VNET))
1660 #endif
1661 		{
1662 #ifdef INET
1663 			if (!(ch_flags & PR_IP4_USER))
1664 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1665 			else if (!(pr_flags & PR_IP4_USER)) {
1666 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1667 				prison_ip_dup(ppr, pr, PR_INET);
1668 			}
1669 #endif
1670 #ifdef INET6
1671 			if (!(ch_flags & PR_IP6_USER))
1672 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1673 			else if (!(pr_flags & PR_IP6_USER)) {
1674 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1675 				prison_ip_dup(ppr, pr, PR_INET6);
1676 			}
1677 #endif
1678 		}
1679 #endif
1680 		/* Source address selection is always on by default. */
1681 		pr->pr_flags |= _PR_IP_SADDRSEL;
1682 
1683 		pr->pr_securelevel = ppr->pr_securelevel;
1684 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1685 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1686 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1687 
1688 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1689 		if (osrelstr == NULL)
1690 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1691 			    sizeof(pr->pr_osrelease));
1692 		else
1693 			strlcpy(pr->pr_osrelease, osrelstr,
1694 			    sizeof(pr->pr_osrelease));
1695 
1696 #ifdef VIMAGE
1697 		/*
1698 		 * Allocate a new vnet if specified.
1699 		 *
1700 		 * Set PR_VNET now if so, so that the vnet is disposed of
1701 		 * properly when the jail is destroyed.
1702 		 */
1703 		if (pr_flags & PR_VNET) {
1704 			pr->pr_flags |= PR_VNET;
1705 			pr->pr_vnet = vnet_alloc();
1706 		} else {
1707 			pr->pr_vnet = ppr->pr_vnet;
1708 		}
1709 #endif
1710 		/*
1711 		 * Allocate a dedicated cpuset for each jail.
1712 		 * Unlike other initial settings, this may return an error.
1713 		 */
1714 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1715 		if (error)
1716 			goto done_deref;
1717 
1718 		mtx_lock(&pr->pr_mtx);
1719 		drflags |= PD_LOCKED;
1720 	} else {
1721 		/*
1722 		 * Grab a reference for existing prisons, to ensure they
1723 		 * continue to exist for the duration of the call.
1724 		 */
1725 		prison_hold(pr);
1726 		drflags |= PD_DEREF;
1727 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1728 		if ((pr->pr_flags & PR_VNET) &&
1729 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1730 			error = EINVAL;
1731 			vfs_opterror(opts,
1732 			    "vnet jails cannot have IP address restrictions");
1733 			goto done_deref;
1734 		}
1735 #endif
1736 #ifdef INET
1737 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1738 			error = EINVAL;
1739 			vfs_opterror(opts,
1740 			    "ip4 cannot be changed after creation");
1741 			goto done_deref;
1742 		}
1743 #endif
1744 #ifdef INET6
1745 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1746 			error = EINVAL;
1747 			vfs_opterror(opts,
1748 			    "ip6 cannot be changed after creation");
1749 			goto done_deref;
1750 		}
1751 #endif
1752 	}
1753 
1754 	/* Do final error checking before setting anything. */
1755 	if (gotslevel) {
1756 		if (slevel < ppr->pr_securelevel) {
1757 			error = EPERM;
1758 			goto done_deref;
1759 		}
1760 	}
1761 	if (gotchildmax) {
1762 		if (childmax >= ppr->pr_childmax) {
1763 			error = EPERM;
1764 			goto done_deref;
1765 		}
1766 	}
1767 	if (gotenforce) {
1768 		if (enforce < ppr->pr_enforce_statfs) {
1769 			error = EPERM;
1770 			goto done_deref;
1771 		}
1772 	}
1773 	if (gotrsnum) {
1774 		/*
1775 		 * devfs_rsnum is a uint16_t
1776 		 */
1777 		if (rsnum < 0 || rsnum > 65535) {
1778 			error = EINVAL;
1779 			goto done_deref;
1780 		}
1781 		/*
1782 		 * Nested jails always inherit parent's devfs ruleset
1783 		 */
1784 		if (jailed(td->td_ucred)) {
1785 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1786 				error = EPERM;
1787 				goto done_deref;
1788 			} else
1789 				rsnum = ppr->pr_devfs_rsnum;
1790 		}
1791 	}
1792 #ifdef INET
1793 	if (ip4s > 0) {
1794 		if ((ppr->pr_flags & PR_IP4) &&
1795 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1796 		    PR_INET)) {
1797 			error = EPERM;
1798 			goto done_deref;
1799 		}
1800 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1801 			error = EADDRINUSE;
1802 			vfs_opterror(opts, "IPv4 addresses clash");
1803 			goto done_deref;
1804 		}
1805 	}
1806 #endif
1807 #ifdef INET6
1808 	if (ip6s > 0) {
1809 		if ((ppr->pr_flags & PR_IP6) &&
1810 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1811 		    PR_INET6)) {
1812 			error = EPERM;
1813 			goto done_deref;
1814 		}
1815 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1816 			error = EADDRINUSE;
1817 			vfs_opterror(opts, "IPv6 addresses clash");
1818 			goto done_deref;
1819 		}
1820 	}
1821 #endif
1822 	onamelen = namelen = 0;
1823 	if (namelc != NULL) {
1824 		/* Give a default name of the jid.  Also allow the name to be
1825 		 * explicitly the jid - but not any other number, and only in
1826 		 * normal form (no leading zero/etc).
1827 		 */
1828 		if (namelc[0] == '\0')
1829 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1830 		else if ((strtoul(namelc, &p, 10) != jid ||
1831 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1832 			error = EINVAL;
1833 			vfs_opterror(opts,
1834 			    "name cannot be numeric (unless it is the jid)");
1835 			goto done_deref;
1836 		}
1837 		/*
1838 		 * Make sure the name isn't too long for the prison or its
1839 		 * children.
1840 		 */
1841 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1842 		onamelen = strlen(pr->pr_name + pnamelen);
1843 		namelen = strlen(namelc);
1844 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1845 			error = ENAMETOOLONG;
1846 			goto done_deref;
1847 		}
1848 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1849 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1850 			    sizeof(pr->pr_name)) {
1851 				error = ENAMETOOLONG;
1852 				goto done_deref;
1853 			}
1854 		}
1855 	}
1856 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1857 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1858 		error = EPERM;
1859 		goto done_deref;
1860 	}
1861 
1862 	/*
1863 	 * Let modules check their parameters.  This requires unlocking and
1864 	 * then re-locking the prison, but this is still a valid state as long
1865 	 * as allprison_lock remains xlocked.
1866 	 */
1867 	mtx_unlock(&pr->pr_mtx);
1868 	drflags &= ~PD_LOCKED;
1869 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1870 	if (error != 0)
1871 		goto done_deref;
1872 	mtx_lock(&pr->pr_mtx);
1873 	drflags |= PD_LOCKED;
1874 
1875 	/* At this point, all valid parameters should have been noted. */
1876 	TAILQ_FOREACH(opt, opts, link) {
1877 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1878 			error = EINVAL;
1879 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1880 			goto done_deref;
1881 		}
1882 	}
1883 
1884 	/* Set the parameters of the prison. */
1885 #ifdef INET
1886 	redo_ip4 = false;
1887 	if (pr_flags & PR_IP4_USER) {
1888 		pr->pr_flags |= PR_IP4;
1889 		prison_ip_set(pr, PR_INET, ip4);
1890 		ip4 = NULL;
1891 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1892 #ifdef VIMAGE
1893 			if (tpr->pr_flags & PR_VNET) {
1894 				descend = 0;
1895 				continue;
1896 			}
1897 #endif
1898 			if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
1899 				redo_ip4 = true;
1900 				descend = 0;
1901 			}
1902 		}
1903 	}
1904 #endif
1905 #ifdef INET6
1906 	redo_ip6 = false;
1907 	if (pr_flags & PR_IP6_USER) {
1908 		pr->pr_flags |= PR_IP6;
1909 		prison_ip_set(pr, PR_INET6, ip6);
1910 		ip6 = NULL;
1911 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1912 #ifdef VIMAGE
1913 			if (tpr->pr_flags & PR_VNET) {
1914 				descend = 0;
1915 				continue;
1916 			}
1917 #endif
1918 			if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
1919 				redo_ip6 = true;
1920 				descend = 0;
1921 			}
1922 		}
1923 	}
1924 #endif
1925 	if (gotslevel) {
1926 		pr->pr_securelevel = slevel;
1927 		/* Set all child jails to be at least this level. */
1928 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1929 			if (tpr->pr_securelevel < slevel)
1930 				tpr->pr_securelevel = slevel;
1931 	}
1932 	if (gotchildmax) {
1933 		pr->pr_childmax = childmax;
1934 		/* Set all child jails to under this limit. */
1935 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1936 			if (tpr->pr_childmax > childmax - level)
1937 				tpr->pr_childmax = childmax > level
1938 				    ? childmax - level : 0;
1939 	}
1940 	if (gotenforce) {
1941 		pr->pr_enforce_statfs = enforce;
1942 		/* Pass this restriction on to the children. */
1943 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1944 			if (tpr->pr_enforce_statfs < enforce)
1945 				tpr->pr_enforce_statfs = enforce;
1946 	}
1947 	if (gotrsnum) {
1948 		pr->pr_devfs_rsnum = rsnum;
1949 		/* Pass this restriction on to the children. */
1950 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1951 			tpr->pr_devfs_rsnum = rsnum;
1952 	}
1953 	if (namelc != NULL) {
1954 		if (ppr == &prison0)
1955 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1956 		else
1957 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1958 			    ppr->pr_name, namelc);
1959 		/* Change this component of child names. */
1960 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1961 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1962 			    strlen(tpr->pr_name + onamelen) + 1);
1963 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1964 		}
1965 	}
1966 	if (path != NULL) {
1967 		/* Try to keep a real-rooted full pathname. */
1968 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1969 		pr->pr_root = root;
1970 		root = NULL;
1971 	}
1972 	if (PR_HOST & ch_flags & ~pr_flags) {
1973 		if (pr->pr_flags & PR_HOST) {
1974 			/*
1975 			 * Copy the parent's host info.  As with pr_ip4 above,
1976 			 * the lack of a lock on the parent is not a problem;
1977 			 * it is always set with allprison_lock at least
1978 			 * shared, and is held exclusively here.
1979 			 */
1980 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1981 			    sizeof(pr->pr_hostname));
1982 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1983 			    sizeof(pr->pr_domainname));
1984 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1985 			    sizeof(pr->pr_hostuuid));
1986 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1987 		}
1988 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1989 		/* Set this prison, and any descendants without PR_HOST. */
1990 		if (host != NULL)
1991 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1992 		if (domain != NULL)
1993 			strlcpy(pr->pr_domainname, domain,
1994 			    sizeof(pr->pr_domainname));
1995 		if (uuid != NULL)
1996 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1997 		if (gothid)
1998 			pr->pr_hostid = hid;
1999 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2000 			if (tpr->pr_flags & PR_HOST)
2001 				descend = 0;
2002 			else {
2003 				if (host != NULL)
2004 					strlcpy(tpr->pr_hostname,
2005 					    pr->pr_hostname,
2006 					    sizeof(tpr->pr_hostname));
2007 				if (domain != NULL)
2008 					strlcpy(tpr->pr_domainname,
2009 					    pr->pr_domainname,
2010 					    sizeof(tpr->pr_domainname));
2011 				if (uuid != NULL)
2012 					strlcpy(tpr->pr_hostuuid,
2013 					    pr->pr_hostuuid,
2014 					    sizeof(tpr->pr_hostuuid));
2015 				if (gothid)
2016 					tpr->pr_hostid = hid;
2017 			}
2018 		}
2019 	}
2020 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2021 	if ((tallow = ch_allow & ~pr_allow))
2022 		prison_set_allow_locked(pr, tallow, 0);
2023 	/*
2024 	 * Persistent prisons get an extra reference, and prisons losing their
2025 	 * persist flag lose that reference.
2026 	 */
2027 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2028 		if (pr_flags & PR_PERSIST) {
2029 			prison_hold(pr);
2030 			/*
2031 			 * This may be a new prison's first user reference,
2032 			 * but wait to call it alive until after OSD calls
2033 			 * have had a chance to run (and perhaps to fail).
2034 			 */
2035 			refcount_acquire(&pr->pr_uref);
2036 		} else {
2037 			drflags |= PD_DEUREF;
2038 			prison_free_not_last(pr);
2039 		}
2040 	}
2041 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2042 	mtx_unlock(&pr->pr_mtx);
2043 	drflags &= ~PD_LOCKED;
2044 	/*
2045 	 * Any errors past this point will need to de-persist newly created
2046 	 * prisons, as well as call remove methods.
2047 	 */
2048 	if (created)
2049 		drflags |= PD_KILL;
2050 
2051 #ifdef RACCT
2052 	if (racct_enable && created)
2053 		prison_racct_attach(pr);
2054 #endif
2055 
2056 	/* Locks may have prevented a complete restriction of child IP
2057 	 * addresses.  If so, allocate some more memory and try again.
2058 	 */
2059 #ifdef INET
2060 	while (redo_ip4) {
2061 		ip4s = pr->pr_addrs[PR_INET]->ips;
2062 		MPASS(ip4 == NULL);
2063 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2064 		mtx_lock(&pr->pr_mtx);
2065 		redo_ip4 = false;
2066 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2067 #ifdef VIMAGE
2068 			if (tpr->pr_flags & PR_VNET) {
2069 				descend = 0;
2070 				continue;
2071 			}
2072 #endif
2073 			if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2074 				redo_ip4 = true;
2075 		}
2076 		mtx_unlock(&pr->pr_mtx);
2077 	}
2078 #endif
2079 #ifdef INET6
2080 	while (redo_ip6) {
2081 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2082 		MPASS(ip6 == NULL);
2083 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2084 		mtx_lock(&pr->pr_mtx);
2085 		redo_ip6 = false;
2086 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2087 #ifdef VIMAGE
2088 			if (tpr->pr_flags & PR_VNET) {
2089 				descend = 0;
2090 				continue;
2091 			}
2092 #endif
2093 			if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2094 				redo_ip6 = true;
2095 		}
2096 		mtx_unlock(&pr->pr_mtx);
2097 	}
2098 #endif
2099 
2100 	/* Let the modules do their work. */
2101 	if (created) {
2102 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2103 		if (error)
2104 			goto done_deref;
2105 	}
2106 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2107 	if (error)
2108 		goto done_deref;
2109 
2110 	/*
2111 	 * A new prison is now ready to be seen; either it has gained a user
2112 	 * reference via persistence, or is about to gain one via attachment.
2113 	 */
2114 	if (created) {
2115 		drflags = prison_lock_xlock(pr, drflags);
2116 		pr->pr_state = PRISON_STATE_ALIVE;
2117 	}
2118 
2119 	/* Attach this process to the prison if requested. */
2120 	if (flags & JAIL_ATTACH) {
2121 		error = do_jail_attach(td, pr,
2122 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2123 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2124 		if (error) {
2125 			vfs_opterror(opts, "attach failed");
2126 			goto done_deref;
2127 		}
2128 	}
2129 
2130 #ifdef RACCT
2131 	if (racct_enable && !created) {
2132 		if (drflags & PD_LOCKED) {
2133 			mtx_unlock(&pr->pr_mtx);
2134 			drflags &= ~PD_LOCKED;
2135 		}
2136 		if (drflags & PD_LIST_XLOCKED) {
2137 			sx_xunlock(&allprison_lock);
2138 			drflags &= ~PD_LIST_XLOCKED;
2139 		}
2140 		prison_racct_modify(pr);
2141 	}
2142 #endif
2143 
2144 	if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2145 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2146 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2147 		   " file system\n", pr->pr_id);
2148 
2149 	drflags &= ~PD_KILL;
2150 	td->td_retval[0] = pr->pr_id;
2151 
2152  done_deref:
2153 	/* Release any temporary prison holds and/or locks. */
2154 	if (pr != NULL)
2155 		prison_deref(pr, drflags);
2156 	else if (drflags & PD_LIST_SLOCKED)
2157 		sx_sunlock(&allprison_lock);
2158 	else if (drflags & PD_LIST_XLOCKED)
2159 		sx_xunlock(&allprison_lock);
2160 	if (root != NULL)
2161 		vrele(root);
2162  done_errmsg:
2163 	if (error) {
2164 		/* Write the error message back to userspace. */
2165 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2166 		    &errmsg_len) == 0 && errmsg_len > 0) {
2167 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2168 			if (optuio->uio_segflg == UIO_SYSSPACE)
2169 				bcopy(errmsg,
2170 				    optuio->uio_iov[errmsg_pos].iov_base,
2171 				    errmsg_len);
2172 			else
2173 				(void)copyout(errmsg,
2174 				    optuio->uio_iov[errmsg_pos].iov_base,
2175 				    errmsg_len);
2176 		}
2177 	}
2178  done_free:
2179 #ifdef INET
2180 	prison_ip_free(ip4);
2181 #endif
2182 #ifdef INET6
2183 	prison_ip_free(ip6);
2184 #endif
2185 	if (g_path != NULL)
2186 		free(g_path, M_TEMP);
2187 	vfs_freeopts(opts);
2188 	return (error);
2189 }
2190 
2191 /*
2192  * Find the next available prison ID.  Return the ID on success, or zero
2193  * on failure.  Also set a pointer to the allprison list entry the prison
2194  * should be inserted before.
2195  */
2196 static int
get_next_prid(struct prison ** insprp)2197 get_next_prid(struct prison **insprp)
2198 {
2199 	struct prison *inspr;
2200 	int jid, maxid;
2201 
2202 	jid = lastprid % JAIL_MAX + 1;
2203 	if (TAILQ_EMPTY(&allprison) ||
2204 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2205 		/*
2206 		 * A common case is for all jails to be implicitly numbered,
2207 		 * which means they'll go on the end of the list, at least
2208 		 * for the first JAIL_MAX times.
2209 		 */
2210 		inspr = NULL;
2211 	} else {
2212 		/*
2213 		 * Take two passes through the allprison list: first starting
2214 		 * with the proposed jid, then ending with it.
2215 		 */
2216 		for (maxid = JAIL_MAX; maxid != 0; ) {
2217 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2218 				if (inspr->pr_id < jid)
2219 					continue;
2220 				if (inspr->pr_id > jid) {
2221 					/* Found an opening. */
2222 					maxid = 0;
2223 					break;
2224 				}
2225 				if (++jid > maxid) {
2226 					if (lastprid == maxid || lastprid == 0)
2227 					{
2228 						/*
2229 						 * The entire legal range
2230 						 * has been traversed
2231 						 */
2232 						return 0;
2233 					}
2234 					/* Try again from the start. */
2235 					jid = 1;
2236 					maxid = lastprid;
2237 					break;
2238 				}
2239 			}
2240 			if (inspr == NULL) {
2241 				/* Found room at the end of the list. */
2242 				break;
2243 			}
2244 		}
2245 	}
2246 	*insprp = inspr;
2247 	lastprid = jid;
2248 	return (jid);
2249 }
2250 
2251 /*
2252  * Find the next available ID for a renumbered dead prison.  This is the same
2253  * as get_next_prid, but counting backward from the end of the range.
2254  */
2255 static int
get_next_deadid(struct prison ** dinsprp)2256 get_next_deadid(struct prison **dinsprp)
2257 {
2258 	struct prison *dinspr;
2259 	int deadid, minid;
2260 
2261 	deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2262 	/*
2263 	 * Take two reverse passes through the allprison list: first
2264 	 * starting with the proposed deadid, then ending with it.
2265 	 */
2266 	for (minid = 1; minid != 0; ) {
2267 		TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2268 			if (dinspr->pr_id > deadid)
2269 				continue;
2270 			if (dinspr->pr_id < deadid) {
2271 				/* Found an opening. */
2272 				minid = 0;
2273 				break;
2274 			}
2275 			if (--deadid < minid) {
2276 				if (lastdeadid == minid || lastdeadid == 0)
2277 				{
2278 					/*
2279 					 * The entire legal range
2280 					 * has been traversed
2281 					 */
2282 					return 0;
2283 				}
2284 				/* Try again from the end. */
2285 				deadid = JAIL_MAX;
2286 				minid = lastdeadid;
2287 				break;
2288 			}
2289 		}
2290 		if (dinspr == NULL) {
2291 			/* Found room at the beginning of the list. */
2292 			break;
2293 		}
2294 	}
2295 	*dinsprp = dinspr;
2296 	lastdeadid = deadid;
2297 	return (deadid);
2298 }
2299 
2300 /*
2301  * struct jail_get_args {
2302  *	struct iovec *iovp;
2303  *	unsigned int iovcnt;
2304  *	int flags;
2305  * };
2306  */
2307 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)2308 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2309 {
2310 	struct uio *auio;
2311 	int error;
2312 
2313 	/* Check that we have an even number of iovecs. */
2314 	if (uap->iovcnt & 1)
2315 		return (EINVAL);
2316 
2317 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2318 	if (error)
2319 		return (error);
2320 	error = kern_jail_get(td, auio, uap->flags);
2321 	if (error == 0)
2322 		error = copyout(auio->uio_iov, uap->iovp,
2323 		    uap->iovcnt * sizeof(struct iovec));
2324 	freeuio(auio);
2325 	return (error);
2326 }
2327 
2328 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2329 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2330 {
2331 	struct bool_flags *bf;
2332 	struct jailsys_flags *jsf;
2333 	struct prison *pr, *mypr;
2334 	struct vfsopt *opt;
2335 	struct vfsoptlist *opts;
2336 	char *errmsg, *name;
2337 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2338 	unsigned f;
2339 
2340 	if (flags & ~JAIL_GET_MASK)
2341 		return (EINVAL);
2342 
2343 	/* Get the parameter list. */
2344 	error = vfs_buildopts(optuio, &opts);
2345 	if (error)
2346 		return (error);
2347 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2348 	mypr = td->td_ucred->cr_prison;
2349 	pr = NULL;
2350 
2351 	/*
2352 	 * Find the prison specified by one of: lastjid, jid, name.
2353 	 */
2354 	sx_slock(&allprison_lock);
2355 	drflags = PD_LIST_SLOCKED;
2356 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2357 	if (error == 0) {
2358 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2359 			if (pr->pr_id > jid &&
2360 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2361 			    prison_ischild(mypr, pr)) {
2362 				mtx_lock(&pr->pr_mtx);
2363 				drflags |= PD_LOCKED;
2364 				goto found_prison;
2365 			}
2366 		}
2367 		error = ENOENT;
2368 		vfs_opterror(opts, "no jail after %d", jid);
2369 		goto done;
2370 	} else if (error != ENOENT)
2371 		goto done;
2372 
2373 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2374 	if (error == 0) {
2375 		if (jid != 0) {
2376 			pr = prison_find_child(mypr, jid);
2377 			if (pr != NULL) {
2378 				drflags |= PD_LOCKED;
2379 				if (!(prison_isalive(pr) ||
2380 				    (flags & JAIL_DYING))) {
2381 					error = ENOENT;
2382 					vfs_opterror(opts, "jail %d is dying",
2383 					    jid);
2384 					goto done;
2385 				}
2386 				goto found_prison;
2387 			}
2388 			error = ENOENT;
2389 			vfs_opterror(opts, "jail %d not found", jid);
2390 			goto done;
2391 		}
2392 	} else if (error != ENOENT)
2393 		goto done;
2394 
2395 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2396 	if (error == 0) {
2397 		if (len == 0 || name[len - 1] != '\0') {
2398 			error = EINVAL;
2399 			goto done;
2400 		}
2401 		pr = prison_find_name(mypr, name);
2402 		if (pr != NULL) {
2403 			drflags |= PD_LOCKED;
2404 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2405 				error = ENOENT;
2406 				vfs_opterror(opts, "jail \"%s\" is dying",
2407 				    name);
2408 				goto done;
2409 			}
2410 			goto found_prison;
2411 		}
2412 		error = ENOENT;
2413 		vfs_opterror(opts, "jail \"%s\" not found", name);
2414 		goto done;
2415 	} else if (error != ENOENT)
2416 		goto done;
2417 
2418 	vfs_opterror(opts, "no jail specified");
2419 	error = ENOENT;
2420 	goto done;
2421 
2422  found_prison:
2423 	/* Get the parameters of the prison. */
2424 	prison_hold(pr);
2425 	drflags |= PD_DEREF;
2426 	td->td_retval[0] = pr->pr_id;
2427 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2428 	if (error != 0 && error != ENOENT)
2429 		goto done;
2430 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2431 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2432 	if (error != 0 && error != ENOENT)
2433 		goto done;
2434 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2435 	if (error != 0 && error != ENOENT)
2436 		goto done;
2437 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2438 	    sizeof(pr->pr_cpuset->cs_id));
2439 	if (error != 0 && error != ENOENT)
2440 		goto done;
2441 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2442 	if (error != 0 && error != ENOENT)
2443 		goto done;
2444 #ifdef INET
2445 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2446 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2447 	    pr_families[PR_INET].size : 0 );
2448 	if (error != 0 && error != ENOENT)
2449 		goto done;
2450 #endif
2451 #ifdef INET6
2452 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2453 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2454 	    pr_families[PR_INET6].size : 0 );
2455 	if (error != 0 && error != ENOENT)
2456 		goto done;
2457 #endif
2458 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2459 	    sizeof(pr->pr_securelevel));
2460 	if (error != 0 && error != ENOENT)
2461 		goto done;
2462 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2463 	    sizeof(pr->pr_childcount));
2464 	if (error != 0 && error != ENOENT)
2465 		goto done;
2466 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2467 	    sizeof(pr->pr_childmax));
2468 	if (error != 0 && error != ENOENT)
2469 		goto done;
2470 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2471 	if (error != 0 && error != ENOENT)
2472 		goto done;
2473 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2474 	if (error != 0 && error != ENOENT)
2475 		goto done;
2476 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2477 	if (error != 0 && error != ENOENT)
2478 		goto done;
2479 #ifdef COMPAT_FREEBSD32
2480 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2481 		uint32_t hid32 = pr->pr_hostid;
2482 
2483 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2484 	} else
2485 #endif
2486 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2487 	    sizeof(pr->pr_hostid));
2488 	if (error != 0 && error != ENOENT)
2489 		goto done;
2490 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2491 	    sizeof(pr->pr_enforce_statfs));
2492 	if (error != 0 && error != ENOENT)
2493 		goto done;
2494 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2495 	    sizeof(pr->pr_devfs_rsnum));
2496 	if (error != 0 && error != ENOENT)
2497 		goto done;
2498 	for (bf = pr_flag_bool;
2499 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2500 	     bf++) {
2501 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2502 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2503 		if (error != 0 && error != ENOENT)
2504 			goto done;
2505 		i = !i;
2506 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2507 		if (error != 0 && error != ENOENT)
2508 			goto done;
2509 	}
2510 	for (jsf = pr_flag_jailsys;
2511 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2512 	     jsf++) {
2513 		f = pr->pr_flags & (jsf->disable | jsf->new);
2514 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2515 		    : (f == jsf->new) ? JAIL_SYS_NEW
2516 		    : JAIL_SYS_INHERIT;
2517 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2518 		if (error != 0 && error != ENOENT)
2519 			goto done;
2520 	}
2521 	for (bf = pr_flag_allow;
2522 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2523 		atomic_load_int(&bf->flag) != 0;
2524 	     bf++) {
2525 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2526 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2527 		if (error != 0 && error != ENOENT)
2528 			goto done;
2529 		i = !i;
2530 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2531 		if (error != 0 && error != ENOENT)
2532 			goto done;
2533 	}
2534 	i = !prison_isalive(pr);
2535 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2536 	if (error != 0 && error != ENOENT)
2537 		goto done;
2538 	i = !i;
2539 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2540 	if (error != 0 && error != ENOENT)
2541 		goto done;
2542 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2543 	    sizeof(pr->pr_osreldate));
2544 	if (error != 0 && error != ENOENT)
2545 		goto done;
2546 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2547 	if (error != 0 && error != ENOENT)
2548 		goto done;
2549 
2550 	/* Get the module parameters. */
2551 	mtx_unlock(&pr->pr_mtx);
2552 	drflags &= ~PD_LOCKED;
2553 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2554 	if (error)
2555 		goto done;
2556 	prison_deref(pr, drflags);
2557 	pr = NULL;
2558 	drflags = 0;
2559 
2560 	/* By now, all parameters should have been noted. */
2561 	TAILQ_FOREACH(opt, opts, link) {
2562 		if (!opt->seen &&
2563 		    (strstr(opt->name, JAIL_META_PRIVATE ".") == opt->name ||
2564 		    strstr(opt->name, JAIL_META_SHARED ".") == opt->name)) {
2565 			/* Communicate back a missing key. */
2566 			free(opt->value, M_MOUNT);
2567 			opt->value = NULL;
2568 			opt->len = 0;
2569 			continue;
2570 		}
2571 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2572 			error = EINVAL;
2573 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2574 			goto done;
2575 		}
2576 	}
2577 
2578 	/* Write the fetched parameters back to userspace. */
2579 	error = 0;
2580 	TAILQ_FOREACH(opt, opts, link) {
2581 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2582 			pos = 2 * opt->pos + 1;
2583 			optuio->uio_iov[pos].iov_len = opt->len;
2584 			if (opt->value != NULL) {
2585 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2586 					bcopy(opt->value,
2587 					    optuio->uio_iov[pos].iov_base,
2588 					    opt->len);
2589 				} else {
2590 					error = copyout(opt->value,
2591 					    optuio->uio_iov[pos].iov_base,
2592 					    opt->len);
2593 					if (error)
2594 						break;
2595 				}
2596 			}
2597 		}
2598 	}
2599 
2600  done:
2601 	/* Release any temporary prison holds and/or locks. */
2602 	if (pr != NULL)
2603 		prison_deref(pr, drflags);
2604 	else if (drflags & PD_LIST_SLOCKED)
2605 		sx_sunlock(&allprison_lock);
2606 	if (error && errmsg_pos >= 0) {
2607 		/* Write the error message back to userspace. */
2608 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2609 		errmsg_pos = 2 * errmsg_pos + 1;
2610 		if (errmsg_len > 0) {
2611 			if (optuio->uio_segflg == UIO_SYSSPACE)
2612 				bcopy(errmsg,
2613 				    optuio->uio_iov[errmsg_pos].iov_base,
2614 				    errmsg_len);
2615 			else
2616 				(void)copyout(errmsg,
2617 				    optuio->uio_iov[errmsg_pos].iov_base,
2618 				    errmsg_len);
2619 		}
2620 	}
2621 	vfs_freeopts(opts);
2622 	return (error);
2623 }
2624 
2625 /*
2626  * struct jail_remove_args {
2627  *	int jid;
2628  * };
2629  */
2630 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2631 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2632 {
2633 	struct prison *pr;
2634 	int error;
2635 
2636 	error = priv_check(td, PRIV_JAIL_REMOVE);
2637 	if (error)
2638 		return (error);
2639 
2640 	sx_xlock(&allprison_lock);
2641 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2642 	if (pr == NULL) {
2643 		sx_xunlock(&allprison_lock);
2644 		return (EINVAL);
2645 	}
2646 	if (!prison_isalive(pr)) {
2647 		/* Silently ignore already-dying prisons. */
2648 		mtx_unlock(&pr->pr_mtx);
2649 		sx_xunlock(&allprison_lock);
2650 		return (0);
2651 	}
2652 	prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2653 	return (0);
2654 }
2655 
2656 /*
2657  * struct jail_attach_args {
2658  *	int jid;
2659  * };
2660  */
2661 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)2662 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2663 {
2664 	struct prison *pr;
2665 	int error;
2666 
2667 	error = priv_check(td, PRIV_JAIL_ATTACH);
2668 	if (error)
2669 		return (error);
2670 
2671 	sx_slock(&allprison_lock);
2672 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2673 	if (pr == NULL) {
2674 		sx_sunlock(&allprison_lock);
2675 		return (EINVAL);
2676 	}
2677 
2678 	/* Do not allow a process to attach to a prison that is not alive. */
2679 	if (!prison_isalive(pr)) {
2680 		mtx_unlock(&pr->pr_mtx);
2681 		sx_sunlock(&allprison_lock);
2682 		return (EINVAL);
2683 	}
2684 
2685 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2686 }
2687 
2688 static int
do_jail_attach(struct thread * td,struct prison * pr,int drflags)2689 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2690 {
2691 	struct proc *p;
2692 	struct ucred *newcred, *oldcred;
2693 	int error;
2694 
2695 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2696 	sx_assert(&allprison_lock, SX_LOCKED);
2697 	drflags &= PD_LOCK_FLAGS;
2698 	/*
2699 	 * XXX: Note that there is a slight race here if two threads
2700 	 * in the same privileged process attempt to attach to two
2701 	 * different jails at the same time.  It is important for
2702 	 * user processes not to do this, or they might end up with
2703 	 * a process root from one prison, but attached to the jail
2704 	 * of another.
2705 	 */
2706 	prison_hold(pr);
2707 	refcount_acquire(&pr->pr_uref);
2708 	drflags |= PD_DEREF | PD_DEUREF;
2709 	mtx_unlock(&pr->pr_mtx);
2710 	drflags &= ~PD_LOCKED;
2711 
2712 	/* Let modules do whatever they need to prepare for attaching. */
2713 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2714 	if (error) {
2715 		prison_deref(pr, drflags);
2716 		return (error);
2717 	}
2718 	sx_unlock(&allprison_lock);
2719 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2720 
2721 	/*
2722 	 * Reparent the newly attached process to this jail.
2723 	 */
2724 	p = td->td_proc;
2725 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2726 	if (error)
2727 		goto e_revert_osd;
2728 
2729 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2730 	if ((error = change_dir(pr->pr_root, td)) != 0)
2731 		goto e_unlock;
2732 #ifdef MAC
2733 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2734 		goto e_unlock;
2735 #endif
2736 	VOP_UNLOCK(pr->pr_root);
2737 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2738 		goto e_revert_osd;
2739 
2740 	newcred = crget();
2741 	PROC_LOCK(p);
2742 	oldcred = crcopysafe(p, newcred);
2743 	newcred->cr_prison = pr;
2744 	proc_set_cred(p, newcred);
2745 	setsugid(p);
2746 #ifdef RACCT
2747 	racct_proc_ucred_changed(p, oldcred, newcred);
2748 	crhold(newcred);
2749 #endif
2750 	PROC_UNLOCK(p);
2751 #ifdef RCTL
2752 	rctl_proc_ucred_changed(p, newcred);
2753 	crfree(newcred);
2754 #endif
2755 	prison_proc_relink(oldcred->cr_prison, pr, p);
2756 	prison_deref(oldcred->cr_prison, drflags);
2757 	crfree(oldcred);
2758 
2759 	/*
2760 	 * If the prison was killed while changing credentials, die along
2761 	 * with it.
2762 	 */
2763 	if (!prison_isalive(pr)) {
2764 		PROC_LOCK(p);
2765 		kern_psignal(p, SIGKILL);
2766 		PROC_UNLOCK(p);
2767 	}
2768 
2769 	return (0);
2770 
2771  e_unlock:
2772 	VOP_UNLOCK(pr->pr_root);
2773  e_revert_osd:
2774 	/* Tell modules this thread is still in its old jail after all. */
2775 	sx_slock(&allprison_lock);
2776 	drflags |= PD_LIST_SLOCKED;
2777 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2778 	prison_deref(pr, drflags);
2779 	return (error);
2780 }
2781 
2782 /*
2783  * Returns a locked prison instance, or NULL on failure.
2784  */
2785 struct prison *
prison_find(int prid)2786 prison_find(int prid)
2787 {
2788 	struct prison *pr;
2789 
2790 	sx_assert(&allprison_lock, SX_LOCKED);
2791 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2792 		if (pr->pr_id < prid)
2793 			continue;
2794 		if (pr->pr_id > prid)
2795 			break;
2796 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2797 		mtx_lock(&pr->pr_mtx);
2798 		return (pr);
2799 	}
2800 	return (NULL);
2801 }
2802 
2803 /*
2804  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2805  */
2806 struct prison *
prison_find_child(struct prison * mypr,int prid)2807 prison_find_child(struct prison *mypr, int prid)
2808 {
2809 	struct prison *pr;
2810 	int descend;
2811 
2812 	sx_assert(&allprison_lock, SX_LOCKED);
2813 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2814 		if (pr->pr_id == prid) {
2815 			KASSERT(prison_isvalid(pr),
2816 			    ("Found invalid prison %p", pr));
2817 			mtx_lock(&pr->pr_mtx);
2818 			return (pr);
2819 		}
2820 	}
2821 	return (NULL);
2822 }
2823 
2824 /*
2825  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2826  */
2827 struct prison *
prison_find_name(struct prison * mypr,const char * name)2828 prison_find_name(struct prison *mypr, const char *name)
2829 {
2830 	struct prison *pr, *deadpr;
2831 	size_t mylen;
2832 	int descend;
2833 
2834 	sx_assert(&allprison_lock, SX_LOCKED);
2835 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2836 	deadpr = NULL;
2837 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2838 		if (!strcmp(pr->pr_name + mylen, name)) {
2839 			KASSERT(prison_isvalid(pr),
2840 			    ("Found invalid prison %p", pr));
2841 			if (prison_isalive(pr)) {
2842 				mtx_lock(&pr->pr_mtx);
2843 				return (pr);
2844 			}
2845 			deadpr = pr;
2846 		}
2847 	}
2848 	/* There was no valid prison - perhaps there was a dying one. */
2849 	if (deadpr != NULL)
2850 		mtx_lock(&deadpr->pr_mtx);
2851 	return (deadpr);
2852 }
2853 
2854 /*
2855  * See if a prison has the specific flag set.  The prison should be locked,
2856  * unless checking for flags that are only set at jail creation (such as
2857  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2858  * to any other prison data.
2859  */
2860 bool
prison_flag(struct ucred * cred,unsigned flag)2861 prison_flag(struct ucred *cred, unsigned flag)
2862 {
2863 
2864 	return ((cred->cr_prison->pr_flags & flag) != 0);
2865 }
2866 
2867 /*
2868  * See if a prison has the specific allow flag set.
2869  * The prison *should* be locked, or only a single bit is examined, without
2870  * regard to any other prison data.
2871  */
2872 bool
prison_allow(struct ucred * cred,unsigned flag)2873 prison_allow(struct ucred *cred, unsigned flag)
2874 {
2875 
2876 	return ((cred->cr_prison->pr_allow & flag) != 0);
2877 }
2878 
2879 /*
2880  * Hold a prison reference, by incrementing pr_ref.  It is generally
2881  * an error to hold a prison that does not already have a reference.
2882  * A prison record will remain valid as long as it has at least one
2883  * reference, and will not be removed as long as either the prison
2884  * mutex or the allprison lock is held (allprison_lock may be shared).
2885  */
2886 void
prison_hold_locked(struct prison * pr)2887 prison_hold_locked(struct prison *pr)
2888 {
2889 
2890 	/* Locking is no longer required. */
2891 	prison_hold(pr);
2892 }
2893 
2894 void
prison_hold(struct prison * pr)2895 prison_hold(struct prison *pr)
2896 {
2897 #ifdef INVARIANTS
2898 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2899 
2900 	KASSERT(was_valid,
2901 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2902 #else
2903 	refcount_acquire(&pr->pr_ref);
2904 #endif
2905 }
2906 
2907 /*
2908  * Remove a prison reference.  If that was the last reference, the
2909  * prison will be removed (at a later time).
2910  */
2911 void
prison_free_locked(struct prison * pr)2912 prison_free_locked(struct prison *pr)
2913 {
2914 
2915 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2916 	/*
2917 	 * Locking is no longer required, but unlock because the caller
2918 	 * expects it.
2919 	 */
2920 	mtx_unlock(&pr->pr_mtx);
2921 	prison_free(pr);
2922 }
2923 
2924 void
prison_free(struct prison * pr)2925 prison_free(struct prison *pr)
2926 {
2927 
2928 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2929 	    ("Trying to free dead prison %p (jid=%d).",
2930 	     pr, pr->pr_id));
2931 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
2932 		/*
2933 		 * Don't remove the last reference in this context,
2934 		 * in case there are locks held.
2935 		 */
2936 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
2937 	}
2938 }
2939 
2940 static void
prison_free_not_last(struct prison * pr)2941 prison_free_not_last(struct prison *pr)
2942 {
2943 #ifdef INVARIANTS
2944 	int lastref;
2945 
2946 	KASSERT(refcount_load(&pr->pr_ref) > 0,
2947 	    ("Trying to free dead prison %p (jid=%d).",
2948 	     pr, pr->pr_id));
2949 	lastref = refcount_release(&pr->pr_ref);
2950 	KASSERT(!lastref,
2951 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2952 	     pr, pr->pr_id));
2953 #else
2954 	refcount_release(&pr->pr_ref);
2955 #endif
2956 }
2957 
2958 /*
2959  * Hold a prison for user visibility, by incrementing pr_uref.
2960  * It is generally an error to hold a prison that isn't already
2961  * user-visible, except through the jail system calls.  It is also
2962  * an error to hold an invalid prison.  A prison record will remain
2963  * alive as long as it has at least one user reference, and will not
2964  * be set to the dying state until the prison mutex and allprison_lock
2965  * are both freed.
2966  */
2967 void
prison_proc_hold(struct prison * pr)2968 prison_proc_hold(struct prison *pr)
2969 {
2970 #ifdef INVARIANTS
2971 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2972 
2973 	KASSERT(was_alive,
2974 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2975 #else
2976 	refcount_acquire(&pr->pr_uref);
2977 #endif
2978 }
2979 
2980 /*
2981  * Remove a prison user reference.  If it was the last reference, the
2982  * prison will be considered "dying", and may be removed once all of
2983  * its references are dropped.
2984  */
2985 void
prison_proc_free(struct prison * pr)2986 prison_proc_free(struct prison *pr)
2987 {
2988 
2989 	/*
2990 	 * Locking is only required when releasing the last reference.
2991 	 * This allows assurance that a locked prison will remain alive
2992 	 * until it is unlocked.
2993 	 */
2994 	KASSERT(refcount_load(&pr->pr_uref) > 0,
2995 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2996 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
2997 		/*
2998 		 * Don't remove the last user reference in this context,
2999 		 * which is expected to be a process that is not only locked,
3000 		 * but also half dead.  Add a reference so any calls to
3001 		 * prison_free() won't re-submit the task.
3002 		 */
3003 		prison_hold(pr);
3004 		mtx_lock(&pr->pr_mtx);
3005 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
3006 		    ("Redundant last reference in prison_proc_free (jid=%d)",
3007 		     pr->pr_id));
3008 		pr->pr_flags |= PR_COMPLETE_PROC;
3009 		mtx_unlock(&pr->pr_mtx);
3010 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3011 	}
3012 }
3013 
3014 static void
prison_proc_free_not_last(struct prison * pr)3015 prison_proc_free_not_last(struct prison *pr)
3016 {
3017 #ifdef INVARIANTS
3018 	int lastref;
3019 
3020 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3021 	    ("Trying to free dead prison %p (jid=%d).",
3022 	     pr, pr->pr_id));
3023 	lastref = refcount_release(&pr->pr_uref);
3024 	KASSERT(!lastref,
3025 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3026 	     pr, pr->pr_id));
3027 #else
3028 	refcount_release(&pr->pr_uref);
3029 #endif
3030 }
3031 
3032 void
prison_proc_link(struct prison * pr,struct proc * p)3033 prison_proc_link(struct prison *pr, struct proc *p)
3034 {
3035 
3036 	sx_assert(&allproc_lock, SA_XLOCKED);
3037 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3038 }
3039 
3040 void
prison_proc_unlink(struct prison * pr,struct proc * p)3041 prison_proc_unlink(struct prison *pr, struct proc *p)
3042 {
3043 
3044 	sx_assert(&allproc_lock, SA_XLOCKED);
3045 	LIST_REMOVE(p, p_jaillist);
3046 }
3047 
3048 static void
prison_proc_relink(struct prison * opr,struct prison * npr,struct proc * p)3049 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3050 {
3051 
3052 	sx_xlock(&allproc_lock);
3053 	prison_proc_unlink(opr, p);
3054 	prison_proc_link(npr, p);
3055 	sx_xunlock(&allproc_lock);
3056 }
3057 
3058 /*
3059  * Complete a call to either prison_free or prison_proc_free.
3060  */
3061 static void
prison_complete(void * context,int pending)3062 prison_complete(void *context, int pending)
3063 {
3064 	struct prison *pr = context;
3065 	int drflags;
3066 
3067 	/*
3068 	 * This could be called to release the last reference, or the last
3069 	 * user reference (plus the reference held in prison_proc_free).
3070 	 */
3071 	drflags = prison_lock_xlock(pr, PD_DEREF);
3072 	if (pr->pr_flags & PR_COMPLETE_PROC) {
3073 		pr->pr_flags &= ~PR_COMPLETE_PROC;
3074 		drflags |= PD_DEUREF;
3075 	}
3076 	prison_deref(pr, drflags);
3077 }
3078 
3079 static void
prison_kill_processes_cb(struct proc * p,void * arg __unused)3080 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3081 {
3082 
3083 	kern_psignal(p, SIGKILL);
3084 }
3085 
3086 /*
3087  * Note the iteration does not guarantee acting on all processes.
3088  * Most notably there may be fork or jail_attach in progress.
3089  */
3090 void
prison_proc_iterate(struct prison * pr,void (* cb)(struct proc *,void *),void * cbarg)3091 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3092     void *cbarg)
3093 {
3094 	struct prison *ppr;
3095 	struct proc *p;
3096 
3097 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3098 		sx_slock(&allproc_lock);
3099 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3100 			if (p->p_state == PRS_NEW)
3101 				continue;
3102 			PROC_LOCK(p);
3103 			cb(p, cbarg);
3104 			PROC_UNLOCK(p);
3105 		}
3106 		sx_sunlock(&allproc_lock);
3107 		if (atomic_load_int(&pr->pr_childcount) == 0)
3108 			return;
3109 		/*
3110 		 * Some jails popped up during the iteration, fall through to a
3111 		 * system-wide search.
3112 		 */
3113 	}
3114 
3115 	sx_slock(&allproc_lock);
3116 	FOREACH_PROC_IN_SYSTEM(p) {
3117 		PROC_LOCK(p);
3118 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3119 			for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3120 			    ppr = ppr->pr_parent) {
3121 				if (ppr == pr) {
3122 					cb(p, cbarg);
3123 					break;
3124 				}
3125 			}
3126 		}
3127 		PROC_UNLOCK(p);
3128 	}
3129 	sx_sunlock(&allproc_lock);
3130 }
3131 
3132 /*
3133  * Remove a prison reference and/or user reference (usually).
3134  * This assumes context that allows sleeping (for allprison_lock),
3135  * with no non-sleeping locks held, except perhaps the prison itself.
3136  * If there are no more references, release and delist the prison.
3137  * On completion, the prison lock and the allprison lock are both
3138  * unlocked.
3139  */
3140 static void
prison_deref(struct prison * pr,int flags)3141 prison_deref(struct prison *pr, int flags)
3142 {
3143 	struct prisonlist freeprison;
3144 	struct prison *killpr, *rpr, *ppr, *tpr;
3145 
3146 	killpr = NULL;
3147 	TAILQ_INIT(&freeprison);
3148 	/*
3149 	 * Release this prison as requested, which may cause its parent
3150 	 * to be released, and then maybe its grandparent, etc.
3151 	 */
3152 	for (;;) {
3153 		if (flags & PD_KILL) {
3154 			/* Kill the prison and its descendents. */
3155 			KASSERT(pr != &prison0,
3156 			    ("prison_deref trying to kill prison0"));
3157 			if (!(flags & PD_DEREF)) {
3158 				prison_hold(pr);
3159 				flags |= PD_DEREF;
3160 			}
3161 			flags = prison_lock_xlock(pr, flags);
3162 			prison_deref_kill(pr, &freeprison);
3163 		}
3164 		if (flags & PD_DEUREF) {
3165 			/* Drop a user reference. */
3166 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3167 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3168 			     pr->pr_id));
3169 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3170 				if (!(flags & PD_DEREF)) {
3171 					prison_hold(pr);
3172 					flags |= PD_DEREF;
3173 				}
3174 				flags = prison_lock_xlock(pr, flags);
3175 				if (refcount_release(&pr->pr_uref) &&
3176 				    pr->pr_state == PRISON_STATE_ALIVE) {
3177 					/*
3178 					 * When the last user references goes,
3179 					 * this becomes a dying prison.
3180 					 */
3181 					KASSERT(
3182 					    refcount_load(&prison0.pr_uref) > 0,
3183 					    ("prison0 pr_uref=0"));
3184 					pr->pr_state = PRISON_STATE_DYING;
3185 					mtx_unlock(&pr->pr_mtx);
3186 					flags &= ~PD_LOCKED;
3187 					prison_cleanup(pr);
3188 				}
3189 			}
3190 		}
3191 		if (flags & PD_KILL) {
3192 			/*
3193 			 * Any remaining user references are probably processes
3194 			 * that need to be killed, either in this prison or its
3195 			 * descendants.
3196 			 */
3197 			if (refcount_load(&pr->pr_uref) > 0)
3198 				killpr = pr;
3199 			/* Make sure the parent prison doesn't get killed. */
3200 			flags &= ~PD_KILL;
3201 		}
3202 		if (flags & PD_DEREF) {
3203 			/* Drop a reference. */
3204 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3205 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3206 			     pr->pr_id));
3207 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3208 				flags = prison_lock_xlock(pr, flags);
3209 				if (refcount_release(&pr->pr_ref)) {
3210 					/*
3211 					 * When the last reference goes,
3212 					 * unlink the prison and set it aside.
3213 					 */
3214 					KASSERT(
3215 					    refcount_load(&pr->pr_uref) == 0,
3216 					    ("prison_deref: last ref, "
3217 					     "but still has %d urefs (jid=%d)",
3218 					     pr->pr_uref, pr->pr_id));
3219 					KASSERT(
3220 					    refcount_load(&prison0.pr_ref) != 0,
3221 					    ("prison0 pr_ref=0"));
3222 					pr->pr_state = PRISON_STATE_INVALID;
3223 					TAILQ_REMOVE(&allprison, pr, pr_list);
3224 					LIST_REMOVE(pr, pr_sibling);
3225 					TAILQ_INSERT_TAIL(&freeprison, pr,
3226 					    pr_list);
3227 					for (ppr = pr->pr_parent;
3228 					     ppr != NULL;
3229 					     ppr = ppr->pr_parent)
3230 						ppr->pr_childcount--;
3231 					/*
3232 					 * Removing a prison frees references
3233 					 * from its parent.
3234 					 */
3235 					ppr = pr->pr_parent;
3236 					pr->pr_parent = NULL;
3237 					mtx_unlock(&pr->pr_mtx);
3238 
3239 					pr = ppr;
3240 					flags &= ~PD_LOCKED;
3241 					flags |= PD_DEREF | PD_DEUREF;
3242 					continue;
3243 				}
3244 			}
3245 		}
3246 		break;
3247 	}
3248 
3249 	/* Release all the prison locks. */
3250 	if (flags & PD_LOCKED)
3251 		mtx_unlock(&pr->pr_mtx);
3252 	if (flags & PD_LIST_SLOCKED)
3253 		sx_sunlock(&allprison_lock);
3254 	else if (flags & PD_LIST_XLOCKED)
3255 		sx_xunlock(&allprison_lock);
3256 
3257 	/* Kill any processes attached to a killed prison. */
3258 	if (killpr != NULL)
3259 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3260 
3261 	/*
3262 	 * Finish removing any unreferenced prisons, which couldn't happen
3263 	 * while allprison_lock was held (to avoid a LOR on vrele).
3264 	 */
3265 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3266 #ifdef VIMAGE
3267 		if (rpr->pr_flags & PR_VNET)
3268 			vnet_destroy(rpr->pr_vnet);
3269 #endif
3270 		if (rpr->pr_root != NULL)
3271 			vrele(rpr->pr_root);
3272 		mtx_destroy(&rpr->pr_mtx);
3273 #ifdef INET
3274 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3275 #endif
3276 #ifdef INET6
3277 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3278 #endif
3279 		if (rpr->pr_cpuset != NULL)
3280 			cpuset_rel(rpr->pr_cpuset);
3281 		osd_jail_exit(rpr);
3282 #ifdef RACCT
3283 		if (racct_enable)
3284 			prison_racct_detach(rpr);
3285 #endif
3286 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3287 		free(rpr, M_PRISON);
3288 	}
3289 }
3290 
3291 /*
3292  * Kill the prison and its descendants.  Mark them as dying, clear the
3293  * persist flag, and call module remove methods.
3294  */
3295 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)3296 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3297 {
3298 	struct prison *cpr, *ppr, *rpr;
3299 	bool descend;
3300 
3301 	/*
3302 	 * Unlike the descendants, the target prison can be killed
3303 	 * even if it is currently dying.  This is useful for failed
3304 	 * creation in jail_set(2).
3305 	 */
3306 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3307 	    ("Trying to kill dead prison %p (jid=%d).",
3308 	     pr, pr->pr_id));
3309 	refcount_acquire(&pr->pr_uref);
3310 	pr->pr_state = PRISON_STATE_DYING;
3311 	mtx_unlock(&pr->pr_mtx);
3312 
3313 	rpr = NULL;
3314 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3315 		if (descend) {
3316 			if (!prison_isalive(cpr)) {
3317 				descend = false;
3318 				continue;
3319 			}
3320 			prison_hold(cpr);
3321 			prison_proc_hold(cpr);
3322 			mtx_lock(&cpr->pr_mtx);
3323 			cpr->pr_state = PRISON_STATE_DYING;
3324 			cpr->pr_flags |= PR_REMOVE;
3325 			mtx_unlock(&cpr->pr_mtx);
3326 			continue;
3327 		}
3328 		if (!(cpr->pr_flags & PR_REMOVE))
3329 			continue;
3330 		prison_cleanup(cpr);
3331 		mtx_lock(&cpr->pr_mtx);
3332 		cpr->pr_flags &= ~PR_REMOVE;
3333 		if (cpr->pr_flags & PR_PERSIST) {
3334 			cpr->pr_flags &= ~PR_PERSIST;
3335 			prison_proc_free_not_last(cpr);
3336 			prison_free_not_last(cpr);
3337 		}
3338 		(void)refcount_release(&cpr->pr_uref);
3339 		if (refcount_release(&cpr->pr_ref)) {
3340 			/*
3341 			 * When the last reference goes, unlink the prison
3342 			 * and set it aside for prison_deref() to handle.
3343 			 * Delay unlinking the sibling list to keep the loop
3344 			 * safe.
3345 			 */
3346 			if (rpr != NULL)
3347 				LIST_REMOVE(rpr, pr_sibling);
3348 			rpr = cpr;
3349 			rpr->pr_state = PRISON_STATE_INVALID;
3350 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3351 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3352 			/*
3353 			 * Removing a prison frees references from its parent.
3354 			 */
3355 			ppr = rpr->pr_parent;
3356 			prison_proc_free_not_last(ppr);
3357 			prison_free_not_last(ppr);
3358 			for (; ppr != NULL; ppr = ppr->pr_parent)
3359 				ppr->pr_childcount--;
3360 		}
3361 		mtx_unlock(&cpr->pr_mtx);
3362 	}
3363 	if (rpr != NULL)
3364 		LIST_REMOVE(rpr, pr_sibling);
3365 
3366 	prison_cleanup(pr);
3367 	mtx_lock(&pr->pr_mtx);
3368 	if (pr->pr_flags & PR_PERSIST) {
3369 		pr->pr_flags &= ~PR_PERSIST;
3370 		prison_proc_free_not_last(pr);
3371 		prison_free_not_last(pr);
3372 	}
3373 	(void)refcount_release(&pr->pr_uref);
3374 }
3375 
3376 /*
3377  * Given the current locking state in the flags, make sure allprison_lock
3378  * is held exclusive, and the prison is locked.  Return flags indicating
3379  * the new state.
3380  */
3381 static int
prison_lock_xlock(struct prison * pr,int flags)3382 prison_lock_xlock(struct prison *pr, int flags)
3383 {
3384 
3385 	if (!(flags & PD_LIST_XLOCKED)) {
3386 		/*
3387 		 * Get allprison_lock, which may be an upgrade,
3388 		 * and may require unlocking the prison.
3389 		 */
3390 		if (flags & PD_LOCKED) {
3391 			mtx_unlock(&pr->pr_mtx);
3392 			flags &= ~PD_LOCKED;
3393 		}
3394 		if (flags & PD_LIST_SLOCKED) {
3395 			if (!sx_try_upgrade(&allprison_lock)) {
3396 				sx_sunlock(&allprison_lock);
3397 				sx_xlock(&allprison_lock);
3398 			}
3399 			flags &= ~PD_LIST_SLOCKED;
3400 		} else
3401 			sx_xlock(&allprison_lock);
3402 		flags |= PD_LIST_XLOCKED;
3403 	}
3404 	if (!(flags & PD_LOCKED)) {
3405 		/* Lock the prison mutex. */
3406 		mtx_lock(&pr->pr_mtx);
3407 		flags |= PD_LOCKED;
3408 	}
3409 	return flags;
3410 }
3411 
3412 /*
3413  * Release a prison's resources when it starts dying (when the last user
3414  * reference is dropped, or when it is killed).
3415  */
3416 static void
prison_cleanup(struct prison * pr)3417 prison_cleanup(struct prison *pr)
3418 {
3419 	sx_assert(&allprison_lock, SA_XLOCKED);
3420 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3421 	vfs_exjail_delete(pr);
3422 	shm_remove_prison(pr);
3423 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3424 }
3425 
3426 /*
3427  * Set or clear a permission bit in the pr_allow field, passing restrictions
3428  * (cleared permission) down to child jails.
3429  */
3430 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)3431 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3432 {
3433 	struct prison *pr;
3434 
3435 	pr = cred->cr_prison;
3436 	sx_slock(&allprison_lock);
3437 	mtx_lock(&pr->pr_mtx);
3438 	prison_set_allow_locked(pr, flag, enable);
3439 	mtx_unlock(&pr->pr_mtx);
3440 	sx_sunlock(&allprison_lock);
3441 }
3442 
3443 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)3444 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3445 {
3446 	struct prison *cpr;
3447 	int descend;
3448 
3449 	if (enable != 0)
3450 		pr->pr_allow |= flag;
3451 	else {
3452 		pr->pr_allow &= ~flag;
3453 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3454 			cpr->pr_allow &= ~flag;
3455 	}
3456 }
3457 
3458 /*
3459  * Check if a jail supports the given address family.
3460  *
3461  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3462  * if not.
3463  */
3464 int
prison_check_af(struct ucred * cred,int af)3465 prison_check_af(struct ucred *cred, int af)
3466 {
3467 	struct prison *pr;
3468 	int error;
3469 
3470 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3471 
3472 	pr = cred->cr_prison;
3473 #ifdef VIMAGE
3474 	/* Prisons with their own network stack are not limited. */
3475 	if (prison_owns_vnet(pr))
3476 		return (0);
3477 #endif
3478 
3479 	error = 0;
3480 	switch (af)
3481 	{
3482 #ifdef INET
3483 	case AF_INET:
3484 		if (pr->pr_flags & PR_IP4)
3485 		{
3486 			mtx_lock(&pr->pr_mtx);
3487 			if ((pr->pr_flags & PR_IP4) &&
3488 			    pr->pr_addrs[PR_INET] == NULL)
3489 				error = EAFNOSUPPORT;
3490 			mtx_unlock(&pr->pr_mtx);
3491 		}
3492 		break;
3493 #endif
3494 #ifdef INET6
3495 	case AF_INET6:
3496 		if (pr->pr_flags & PR_IP6)
3497 		{
3498 			mtx_lock(&pr->pr_mtx);
3499 			if ((pr->pr_flags & PR_IP6) &&
3500 			    pr->pr_addrs[PR_INET6] == NULL)
3501 				error = EAFNOSUPPORT;
3502 			mtx_unlock(&pr->pr_mtx);
3503 		}
3504 		break;
3505 #endif
3506 	case AF_LOCAL:
3507 	case AF_ROUTE:
3508 	case AF_NETLINK:
3509 		break;
3510 	default:
3511 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3512 			error = EAFNOSUPPORT;
3513 	}
3514 	return (error);
3515 }
3516 
3517 /*
3518  * Check if given address belongs to the jail referenced by cred (wrapper to
3519  * prison_check_ip[46]).
3520  *
3521  * Returns 0 if jail doesn't restrict the address family or if address belongs
3522  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3523  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3524  */
3525 int
prison_if(struct ucred * cred,const struct sockaddr * sa)3526 prison_if(struct ucred *cred, const struct sockaddr *sa)
3527 {
3528 #ifdef INET
3529 	const struct sockaddr_in *sai;
3530 #endif
3531 #ifdef INET6
3532 	const struct sockaddr_in6 *sai6;
3533 #endif
3534 	int error;
3535 
3536 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3537 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3538 
3539 #ifdef VIMAGE
3540 	if (prison_owns_vnet(cred->cr_prison))
3541 		return (0);
3542 #endif
3543 
3544 	error = 0;
3545 	switch (sa->sa_family)
3546 	{
3547 #ifdef INET
3548 	case AF_INET:
3549 		sai = (const struct sockaddr_in *)sa;
3550 		error = prison_check_ip4(cred, &sai->sin_addr);
3551 		break;
3552 #endif
3553 #ifdef INET6
3554 	case AF_INET6:
3555 		sai6 = (const struct sockaddr_in6 *)sa;
3556 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3557 		break;
3558 #endif
3559 	default:
3560 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3561 			error = EAFNOSUPPORT;
3562 	}
3563 	return (error);
3564 }
3565 
3566 /*
3567  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3568  */
3569 int
prison_check(struct ucred * cred1,struct ucred * cred2)3570 prison_check(struct ucred *cred1, struct ucred *cred2)
3571 {
3572 
3573 	return ((cred1->cr_prison == cred2->cr_prison ||
3574 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3575 }
3576 
3577 /*
3578  * For mountd/nfsd to run within a prison, it must be:
3579  * - A vnet prison.
3580  * - PR_ALLOW_NFSD must be set on it.
3581  * - The root directory (pr_root) of the prison must be
3582  *   a file system mount point, so the mountd can hang
3583  *   export information on it.
3584  * - The prison's enforce_statfs cannot be 0, so that
3585  *   mountd(8) can do exports.
3586  */
3587 bool
prison_check_nfsd(struct ucred * cred)3588 prison_check_nfsd(struct ucred *cred)
3589 {
3590 
3591 	if (jailed_without_vnet(cred))
3592 		return (false);
3593 	if (!prison_allow(cred, PR_ALLOW_NFSD))
3594 		return (false);
3595 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3596 		return (false);
3597 	if (cred->cr_prison->pr_enforce_statfs == 0)
3598 		return (false);
3599 	return (true);
3600 }
3601 
3602 /*
3603  * Return true if p2 is a child of p1, otherwise false.
3604  */
3605 bool
prison_ischild(struct prison * pr1,struct prison * pr2)3606 prison_ischild(struct prison *pr1, struct prison *pr2)
3607 {
3608 
3609 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3610 		if (pr1 == pr2)
3611 			return (true);
3612 	return (false);
3613 }
3614 
3615 /*
3616  * Return true if the prison is currently alive.  A prison is alive if it
3617  * holds user references and it isn't being removed.
3618  */
3619 bool
prison_isalive(const struct prison * pr)3620 prison_isalive(const struct prison *pr)
3621 {
3622 
3623 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3624 		return (false);
3625 	return (true);
3626 }
3627 
3628 /*
3629  * Return true if the prison is currently valid.  A prison is valid if it has
3630  * been fully created, and is not being destroyed.  Note that dying prisons
3631  * are still considered valid.  Invalid prisons won't be found under normal
3632  * circumstances, as they're only put in that state by functions that have
3633  * an exclusive hold on allprison_lock.
3634  */
3635 bool
prison_isvalid(struct prison * pr)3636 prison_isvalid(struct prison *pr)
3637 {
3638 
3639 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3640 		return (false);
3641 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3642 		return (false);
3643 	return (true);
3644 }
3645 
3646 /*
3647  * Return true if the passed credential is in a jail and that jail does not
3648  * have its own virtual network stack, otherwise false.
3649  */
3650 bool
jailed_without_vnet(struct ucred * cred)3651 jailed_without_vnet(struct ucred *cred)
3652 {
3653 
3654 	if (!jailed(cred))
3655 		return (false);
3656 #ifdef VIMAGE
3657 	if (prison_owns_vnet(cred->cr_prison))
3658 		return (false);
3659 #endif
3660 
3661 	return (true);
3662 }
3663 
3664 /*
3665  * Return the correct hostname (domainname, et al) for the passed credential.
3666  */
3667 void
getcredhostname(struct ucred * cred,char * buf,size_t size)3668 getcredhostname(struct ucred *cred, char *buf, size_t size)
3669 {
3670 	struct prison *pr;
3671 
3672 	/*
3673 	 * A NULL credential can be used to shortcut to the physical
3674 	 * system's hostname.
3675 	 */
3676 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3677 	mtx_lock(&pr->pr_mtx);
3678 	strlcpy(buf, pr->pr_hostname, size);
3679 	mtx_unlock(&pr->pr_mtx);
3680 }
3681 
3682 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)3683 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3684 {
3685 
3686 	mtx_lock(&cred->cr_prison->pr_mtx);
3687 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3688 	mtx_unlock(&cred->cr_prison->pr_mtx);
3689 }
3690 
3691 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)3692 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3693 {
3694 
3695 	mtx_lock(&cred->cr_prison->pr_mtx);
3696 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3697 	mtx_unlock(&cred->cr_prison->pr_mtx);
3698 }
3699 
3700 void
getcredhostid(struct ucred * cred,unsigned long * hostid)3701 getcredhostid(struct ucred *cred, unsigned long *hostid)
3702 {
3703 
3704 	mtx_lock(&cred->cr_prison->pr_mtx);
3705 	*hostid = cred->cr_prison->pr_hostid;
3706 	mtx_unlock(&cred->cr_prison->pr_mtx);
3707 }
3708 
3709 void
getjailname(struct ucred * cred,char * name,size_t len)3710 getjailname(struct ucred *cred, char *name, size_t len)
3711 {
3712 
3713 	mtx_lock(&cred->cr_prison->pr_mtx);
3714 	strlcpy(name, cred->cr_prison->pr_name, len);
3715 	mtx_unlock(&cred->cr_prison->pr_mtx);
3716 }
3717 
3718 #ifdef VIMAGE
3719 /*
3720  * Determine whether the prison owns its VNET.
3721  */
3722 bool
prison_owns_vnet(struct prison * pr)3723 prison_owns_vnet(struct prison *pr)
3724 {
3725 
3726 	/*
3727 	 * vnets cannot be added/removed after jail creation,
3728 	 * so no need to lock here.
3729 	 */
3730 	return ((pr->pr_flags & PR_VNET) != 0);
3731 }
3732 #endif
3733 
3734 /*
3735  * Determine whether the subject represented by cred can "see"
3736  * status of a mount point.
3737  * Returns: 0 for permitted, ENOENT otherwise.
3738  * XXX: This function should be called cr_canseemount() and should be
3739  *      placed in kern_prot.c.
3740  */
3741 int
prison_canseemount(struct ucred * cred,struct mount * mp)3742 prison_canseemount(struct ucred *cred, struct mount *mp)
3743 {
3744 	struct prison *pr;
3745 	struct statfs *sp;
3746 	size_t len;
3747 
3748 	pr = cred->cr_prison;
3749 	if (pr->pr_enforce_statfs == 0)
3750 		return (0);
3751 	if (pr->pr_root->v_mount == mp)
3752 		return (0);
3753 	if (pr->pr_enforce_statfs == 2)
3754 		return (ENOENT);
3755 	/*
3756 	 * If jail's chroot directory is set to "/" we should be able to see
3757 	 * all mount-points from inside a jail.
3758 	 * This is ugly check, but this is the only situation when jail's
3759 	 * directory ends with '/'.
3760 	 */
3761 	if (strcmp(pr->pr_path, "/") == 0)
3762 		return (0);
3763 	len = strlen(pr->pr_path);
3764 	sp = &mp->mnt_stat;
3765 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3766 		return (ENOENT);
3767 	/*
3768 	 * Be sure that we don't have situation where jail's root directory
3769 	 * is "/some/path" and mount point is "/some/pathpath".
3770 	 */
3771 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3772 		return (ENOENT);
3773 	return (0);
3774 }
3775 
3776 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)3777 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3778 {
3779 	char jpath[MAXPATHLEN];
3780 	struct prison *pr;
3781 	size_t len;
3782 
3783 	pr = cred->cr_prison;
3784 	if (pr->pr_enforce_statfs == 0)
3785 		return;
3786 	if (prison_canseemount(cred, mp) != 0) {
3787 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3788 		strlcpy(sp->f_mntonname, "[restricted]",
3789 		    sizeof(sp->f_mntonname));
3790 		return;
3791 	}
3792 	if (pr->pr_root->v_mount == mp) {
3793 		/*
3794 		 * Clear current buffer data, so we are sure nothing from
3795 		 * the valid path left there.
3796 		 */
3797 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3798 		*sp->f_mntonname = '/';
3799 		return;
3800 	}
3801 	/*
3802 	 * If jail's chroot directory is set to "/" we should be able to see
3803 	 * all mount-points from inside a jail.
3804 	 */
3805 	if (strcmp(pr->pr_path, "/") == 0)
3806 		return;
3807 	len = strlen(pr->pr_path);
3808 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3809 	/*
3810 	 * Clear current buffer data, so we are sure nothing from
3811 	 * the valid path left there.
3812 	 */
3813 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3814 	if (*jpath == '\0') {
3815 		/* Should never happen. */
3816 		*sp->f_mntonname = '/';
3817 	} else {
3818 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3819 	}
3820 }
3821 
3822 /*
3823  * Check with permission for a specific privilege is granted within jail.  We
3824  * have a specific list of accepted privileges; the rest are denied.
3825  */
3826 int
prison_priv_check(struct ucred * cred,int priv)3827 prison_priv_check(struct ucred *cred, int priv)
3828 {
3829 	struct prison *pr;
3830 	int error;
3831 
3832 	/*
3833 	 * Some policies have custom handlers. This routine should not be
3834 	 * called for them. See priv_check_cred().
3835 	 */
3836 	switch (priv) {
3837 	case PRIV_VFS_LOOKUP:
3838 	case PRIV_VFS_GENERATION:
3839 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3840 		    "called for %d\n", priv));
3841 	}
3842 
3843 	if (!jailed(cred))
3844 		return (0);
3845 
3846 #ifdef VIMAGE
3847 	/*
3848 	 * Privileges specific to prisons with a virtual network stack.
3849 	 * There might be a duplicate entry here in case the privilege
3850 	 * is only granted conditionally in the legacy jail case.
3851 	 */
3852 	switch (priv) {
3853 		/*
3854 		 * NFS-specific privileges.
3855 		 */
3856 	case PRIV_NFS_DAEMON:
3857 	case PRIV_VFS_GETFH:
3858 	case PRIV_VFS_MOUNT_EXPORTED:
3859 		if (!prison_check_nfsd(cred))
3860 			return (EPERM);
3861 #ifdef notyet
3862 	case PRIV_NFS_LOCKD:
3863 #endif
3864 		/*
3865 		 * Network stack privileges.
3866 		 */
3867 	case PRIV_NET_BRIDGE:
3868 	case PRIV_NET_GRE:
3869 	case PRIV_NET_BPF:
3870 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3871 	case PRIV_NET_ROUTE:
3872 	case PRIV_NET_TAP:
3873 	case PRIV_NET_SETIFMTU:
3874 	case PRIV_NET_SETIFFLAGS:
3875 	case PRIV_NET_SETIFCAP:
3876 	case PRIV_NET_SETIFDESCR:
3877 	case PRIV_NET_SETIFNAME	:
3878 	case PRIV_NET_SETIFMETRIC:
3879 	case PRIV_NET_SETIFPHYS:
3880 	case PRIV_NET_SETIFMAC:
3881 	case PRIV_NET_SETLANPCP:
3882 	case PRIV_NET_ADDMULTI:
3883 	case PRIV_NET_DELMULTI:
3884 	case PRIV_NET_HWIOCTL:
3885 	case PRIV_NET_SETLLADDR:
3886 	case PRIV_NET_ADDIFGROUP:
3887 	case PRIV_NET_DELIFGROUP:
3888 	case PRIV_NET_IFCREATE:
3889 	case PRIV_NET_IFDESTROY:
3890 	case PRIV_NET_ADDIFADDR:
3891 	case PRIV_NET_DELIFADDR:
3892 	case PRIV_NET_LAGG:
3893 	case PRIV_NET_GIF:
3894 	case PRIV_NET_SETIFVNET:
3895 	case PRIV_NET_SETIFFIB:
3896 	case PRIV_NET_OVPN:
3897 	case PRIV_NET_ME:
3898 	case PRIV_NET_WG:
3899 
3900 		/*
3901 		 * 802.11-related privileges.
3902 		 */
3903 	case PRIV_NET80211_VAP_GETKEY:
3904 	case PRIV_NET80211_VAP_MANAGE:
3905 
3906 #ifdef notyet
3907 		/*
3908 		 * ATM privileges.
3909 		 */
3910 	case PRIV_NETATM_CFG:
3911 	case PRIV_NETATM_ADD:
3912 	case PRIV_NETATM_DEL:
3913 	case PRIV_NETATM_SET:
3914 
3915 		/*
3916 		 * Bluetooth privileges.
3917 		 */
3918 	case PRIV_NETBLUETOOTH_RAW:
3919 #endif
3920 
3921 		/*
3922 		 * Netgraph and netgraph module privileges.
3923 		 */
3924 	case PRIV_NETGRAPH_CONTROL:
3925 #ifdef notyet
3926 	case PRIV_NETGRAPH_TTY:
3927 #endif
3928 
3929 		/*
3930 		 * IPv4 and IPv6 privileges.
3931 		 */
3932 	case PRIV_NETINET_IPFW:
3933 	case PRIV_NETINET_DIVERT:
3934 	case PRIV_NETINET_PF:
3935 	case PRIV_NETINET_DUMMYNET:
3936 	case PRIV_NETINET_CARP:
3937 	case PRIV_NETINET_MROUTE:
3938 	case PRIV_NETINET_RAW:
3939 	case PRIV_NETINET_ADDRCTRL6:
3940 	case PRIV_NETINET_ND6:
3941 	case PRIV_NETINET_SCOPE6:
3942 	case PRIV_NETINET_ALIFETIME6:
3943 	case PRIV_NETINET_IPSEC:
3944 	case PRIV_NETINET_BINDANY:
3945 
3946 #ifdef notyet
3947 		/*
3948 		 * NCP privileges.
3949 		 */
3950 	case PRIV_NETNCP:
3951 
3952 		/*
3953 		 * SMB privileges.
3954 		 */
3955 	case PRIV_NETSMB:
3956 #endif
3957 
3958 	/*
3959 	 * No default: or deny here.
3960 	 * In case of no permit fall through to next switch().
3961 	 */
3962 		if (cred->cr_prison->pr_flags & PR_VNET)
3963 			return (0);
3964 	}
3965 #endif /* VIMAGE */
3966 
3967 	switch (priv) {
3968 		/*
3969 		 * Allow ktrace privileges for root in jail.
3970 		 */
3971 	case PRIV_KTRACE:
3972 
3973 #if 0
3974 		/*
3975 		 * Allow jailed processes to configure audit identity and
3976 		 * submit audit records (login, etc).  In the future we may
3977 		 * want to further refine the relationship between audit and
3978 		 * jail.
3979 		 */
3980 	case PRIV_AUDIT_GETAUDIT:
3981 	case PRIV_AUDIT_SETAUDIT:
3982 	case PRIV_AUDIT_SUBMIT:
3983 #endif
3984 
3985 		/*
3986 		 * Allow jailed processes to manipulate process UNIX
3987 		 * credentials in any way they see fit.
3988 		 */
3989 	case PRIV_CRED_SETCRED:
3990 	case PRIV_CRED_SETUID:
3991 	case PRIV_CRED_SETEUID:
3992 	case PRIV_CRED_SETGID:
3993 	case PRIV_CRED_SETEGID:
3994 	case PRIV_CRED_SETGROUPS:
3995 	case PRIV_CRED_SETREUID:
3996 	case PRIV_CRED_SETREGID:
3997 	case PRIV_CRED_SETRESUID:
3998 	case PRIV_CRED_SETRESGID:
3999 
4000 		/*
4001 		 * Jail implements visibility constraints already, so allow
4002 		 * jailed root to override uid/gid-based constraints.
4003 		 */
4004 	case PRIV_SEEOTHERGIDS:
4005 	case PRIV_SEEOTHERUIDS:
4006 	case PRIV_SEEJAILPROC:
4007 
4008 		/*
4009 		 * Jail implements inter-process debugging limits already, so
4010 		 * allow jailed root various debugging privileges.
4011 		 */
4012 	case PRIV_DEBUG_DIFFCRED:
4013 	case PRIV_DEBUG_SUGID:
4014 	case PRIV_DEBUG_UNPRIV:
4015 	case PRIV_DEBUG_DIFFJAIL:
4016 
4017 		/*
4018 		 * Allow jail to set various resource limits and login
4019 		 * properties, and for now, exceed process resource limits.
4020 		 */
4021 	case PRIV_PROC_LIMIT:
4022 	case PRIV_PROC_SETLOGIN:
4023 	case PRIV_PROC_SETRLIMIT:
4024 
4025 		/*
4026 		 * Debuggers should work in jails.
4027 		 */
4028 	case PRIV_PROC_MEM_WRITE:
4029 
4030 		/*
4031 		 * System V and POSIX IPC privileges are granted in jail.
4032 		 */
4033 	case PRIV_IPC_READ:
4034 	case PRIV_IPC_WRITE:
4035 	case PRIV_IPC_ADMIN:
4036 	case PRIV_IPC_MSGSIZE:
4037 	case PRIV_MQ_ADMIN:
4038 
4039 		/*
4040 		 * Jail operations within a jail work on child jails.
4041 		 */
4042 	case PRIV_JAIL_ATTACH:
4043 	case PRIV_JAIL_SET:
4044 	case PRIV_JAIL_REMOVE:
4045 
4046 		/*
4047 		 * Jail implements its own inter-process limits, so allow
4048 		 * root processes in jail to change scheduling on other
4049 		 * processes in the same jail.  Likewise for signalling.
4050 		 */
4051 	case PRIV_SCHED_DIFFCRED:
4052 	case PRIV_SCHED_CPUSET:
4053 	case PRIV_SCHED_DIFFJAIL:
4054 	case PRIV_SIGNAL_DIFFCRED:
4055 	case PRIV_SIGNAL_SUGID:
4056 	case PRIV_SIGNAL_DIFFJAIL:
4057 
4058 		/*
4059 		 * Allow jailed processes to write to sysctls marked as jail
4060 		 * writable.
4061 		 */
4062 	case PRIV_SYSCTL_WRITEJAIL:
4063 
4064 		/*
4065 		 * Allow root in jail to manage a variety of quota
4066 		 * properties.  These should likely be conditional on a
4067 		 * configuration option.
4068 		 */
4069 	case PRIV_VFS_GETQUOTA:
4070 	case PRIV_VFS_SETQUOTA:
4071 
4072 		/*
4073 		 * Since Jail relies on chroot() to implement file system
4074 		 * protections, grant many VFS privileges to root in jail.
4075 		 * Be careful to exclude mount-related and NFS-related
4076 		 * privileges.
4077 		 */
4078 	case PRIV_VFS_READ:
4079 	case PRIV_VFS_WRITE:
4080 	case PRIV_VFS_ADMIN:
4081 	case PRIV_VFS_EXEC:
4082 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
4083 	case PRIV_VFS_CHFLAGS_DEV:
4084 	case PRIV_VFS_CHOWN:
4085 	case PRIV_VFS_CHROOT:
4086 	case PRIV_VFS_RETAINSUGID:
4087 	case PRIV_VFS_FCHROOT:
4088 	case PRIV_VFS_LINK:
4089 	case PRIV_VFS_SETGID:
4090 	case PRIV_VFS_STAT:
4091 	case PRIV_VFS_STICKYFILE:
4092 
4093 		/*
4094 		 * As in the non-jail case, non-root users are expected to be
4095 		 * able to read kernel/physical memory (provided /dev/[k]mem
4096 		 * exists in the jail and they have permission to access it).
4097 		 */
4098 	case PRIV_KMEM_READ:
4099 		return (0);
4100 
4101 		/*
4102 		 * Depending on the global setting, allow privilege of
4103 		 * setting system flags.
4104 		 */
4105 	case PRIV_VFS_SYSFLAGS:
4106 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4107 			return (0);
4108 		else
4109 			return (EPERM);
4110 
4111 		/*
4112 		 * Depending on the global setting, allow privilege of
4113 		 * mounting/unmounting file systems.
4114 		 */
4115 	case PRIV_VFS_MOUNT:
4116 	case PRIV_VFS_UNMOUNT:
4117 	case PRIV_VFS_MOUNT_NONUSER:
4118 	case PRIV_VFS_MOUNT_OWNER:
4119 		pr = cred->cr_prison;
4120 		prison_lock(pr);
4121 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4122 			error = 0;
4123 		else
4124 			error = EPERM;
4125 		prison_unlock(pr);
4126 		return (error);
4127 
4128 		/*
4129 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4130 		 * policy.  priv_check_cred will not specifically allow it, and
4131 		 * we may want a MAC policy to allow it.
4132 		 */
4133 	case PRIV_VFS_READ_DIR:
4134 		return (0);
4135 
4136 		/*
4137 		 * Conditionally allow privileged process in the jail to
4138 		 * manipulate filesystem extended attributes in the system
4139 		 * namespace.
4140 		 */
4141 	case PRIV_VFS_EXTATTR_SYSTEM:
4142 		if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4143 			return (0);
4144 		else
4145 			return (EPERM);
4146 
4147 		/*
4148 		 * Conditionnaly allow locking (unlocking) physical pages
4149 		 * in memory.
4150 		 */
4151 	case PRIV_VM_MLOCK:
4152 	case PRIV_VM_MUNLOCK:
4153 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4154 			return (0);
4155 		else
4156 			return (EPERM);
4157 
4158 		/*
4159 		 * Conditionally allow jailed root to bind reserved ports.
4160 		 */
4161 	case PRIV_NETINET_RESERVEDPORT:
4162 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4163 			return (0);
4164 		else
4165 			return (EPERM);
4166 
4167 		/*
4168 		 * Allow jailed root to reuse in-use ports.
4169 		 */
4170 	case PRIV_NETINET_REUSEPORT:
4171 		return (0);
4172 
4173 		/*
4174 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4175 		 */
4176 	case PRIV_NETINET_SETHDROPTS:
4177 		return (0);
4178 
4179 		/*
4180 		 * Conditionally allow creating raw sockets in jail.
4181 		 */
4182 	case PRIV_NETINET_RAW:
4183 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4184 			return (0);
4185 		else
4186 			return (EPERM);
4187 
4188 		/*
4189 		 * Since jail implements its own visibility limits on netstat
4190 		 * sysctls, allow getcred.  This allows identd to work in
4191 		 * jail.
4192 		 */
4193 	case PRIV_NETINET_GETCRED:
4194 		return (0);
4195 
4196 		/*
4197 		 * Allow jailed root to set loginclass.
4198 		 */
4199 	case PRIV_PROC_SETLOGINCLASS:
4200 		return (0);
4201 
4202 		/*
4203 		 * Do not allow a process inside a jail to read the kernel
4204 		 * message buffer unless explicitly permitted.
4205 		 */
4206 	case PRIV_MSGBUF:
4207 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4208 			return (0);
4209 		return (EPERM);
4210 
4211 		/*
4212 		 * Conditionally allow privileged process in the jail adjust
4213 		 * machine time.
4214 		 */
4215 	case PRIV_ADJTIME:
4216 	case PRIV_NTP_ADJTIME:
4217 		if (cred->cr_prison->pr_allow &
4218 		    (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4219 			return (0);
4220 		}
4221 		return (EPERM);
4222 
4223 		/*
4224 		 * Conditionally allow privileged process in the jail set
4225 		 * machine time.
4226 		 */
4227 	case PRIV_SETTIMEOFDAY:
4228 	case PRIV_CLOCK_SETTIME:
4229 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4230 			return (0);
4231 		else
4232 			return (EPERM);
4233 
4234 		/*
4235 		 * Conditionally allow privileged process in the jail to modify
4236 		 * the routing table.
4237 		 */
4238 	case PRIV_NET_ROUTE:
4239 		if (cred->cr_prison->pr_allow & PR_ALLOW_ROUTING)
4240 			return (0);
4241 		else
4242 			return (EPERM);
4243 
4244 	default:
4245 		/*
4246 		 * In all remaining cases, deny the privilege request.  This
4247 		 * includes almost all network privileges, many system
4248 		 * configuration privileges.
4249 		 */
4250 		return (EPERM);
4251 	}
4252 }
4253 
4254 /*
4255  * Return the part of pr2's name that is relative to pr1, or the whole name
4256  * if it does not directly follow.
4257  */
4258 
4259 char *
prison_name(struct prison * pr1,struct prison * pr2)4260 prison_name(struct prison *pr1, struct prison *pr2)
4261 {
4262 	char *name;
4263 
4264 	/* Jails see themselves as "0" (if they see themselves at all). */
4265 	if (pr1 == pr2)
4266 		return "0";
4267 	name = pr2->pr_name;
4268 	if (prison_ischild(pr1, pr2)) {
4269 		/*
4270 		 * pr1 isn't locked (and allprison_lock may not be either)
4271 		 * so its length can't be counted on.  But the number of dots
4272 		 * can be counted on - and counted.
4273 		 */
4274 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4275 			name = strchr(name, '.') + 1;
4276 	}
4277 	return (name);
4278 }
4279 
4280 /*
4281  * Return the part of pr2's path that is relative to pr1, or the whole path
4282  * if it does not directly follow.
4283  */
4284 static char *
prison_path(struct prison * pr1,struct prison * pr2)4285 prison_path(struct prison *pr1, struct prison *pr2)
4286 {
4287 	char *path1, *path2;
4288 	int len1;
4289 
4290 	path1 = pr1->pr_path;
4291 	path2 = pr2->pr_path;
4292 	if (!strcmp(path1, "/"))
4293 		return (path2);
4294 	len1 = strlen(path1);
4295 	if (strncmp(path1, path2, len1))
4296 		return (path2);
4297 	if (path2[len1] == '\0')
4298 		return "/";
4299 	if (path2[len1] == '/')
4300 		return (path2 + len1);
4301 	return (path2);
4302 }
4303 
4304 /*
4305  * Jail-related sysctls.
4306  */
4307 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4308     "Jails");
4309 
4310 #if defined(INET) || defined(INET6)
4311 /*
4312  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4313  * sysctl_jail_list() helper.
4314  */
4315 static void
prison_ip_copyout(struct prison * pr,const pr_family_t af,void ** out,int * len)4316 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4317 {
4318 	const struct prison_ip *pip;
4319 	const size_t size = pr_families[af].size;
4320 
4321  again:
4322 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4323 	if ((pip = pr->pr_addrs[af]) != NULL) {
4324 		if (*len < pip->ips) {
4325 			*len = pip->ips;
4326 			mtx_unlock(&pr->pr_mtx);
4327 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4328 			mtx_lock(&pr->pr_mtx);
4329 			goto again;
4330 		}
4331 		bcopy(pip->pr_ip, *out, pip->ips * size);
4332 	}
4333 }
4334 #endif
4335 
4336 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4337 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4338 {
4339 	struct xprison *xp;
4340 	struct prison *pr, *cpr;
4341 #ifdef INET
4342 	struct in_addr *ip4 = NULL;
4343 	int ip4s = 0;
4344 #endif
4345 #ifdef INET6
4346 	struct in6_addr *ip6 = NULL;
4347 	int ip6s = 0;
4348 #endif
4349 	int descend, error;
4350 
4351 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4352 	pr = req->td->td_ucred->cr_prison;
4353 	error = 0;
4354 	sx_slock(&allprison_lock);
4355 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4356 		mtx_lock(&cpr->pr_mtx);
4357 #ifdef INET
4358 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4359 #endif
4360 #ifdef INET6
4361 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4362 #endif
4363 		bzero(xp, sizeof(*xp));
4364 		xp->pr_version = XPRISON_VERSION;
4365 		xp->pr_id = cpr->pr_id;
4366 		xp->pr_state = cpr->pr_state;
4367 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4368 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4369 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4370 #ifdef INET
4371 		xp->pr_ip4s = ip4s;
4372 #endif
4373 #ifdef INET6
4374 		xp->pr_ip6s = ip6s;
4375 #endif
4376 		mtx_unlock(&cpr->pr_mtx);
4377 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4378 		if (error)
4379 			break;
4380 #ifdef INET
4381 		if (xp->pr_ip4s > 0) {
4382 			error = SYSCTL_OUT(req, ip4,
4383 			    xp->pr_ip4s * sizeof(struct in_addr));
4384 			if (error)
4385 				break;
4386 		}
4387 #endif
4388 #ifdef INET6
4389 		if (xp->pr_ip6s > 0) {
4390 			error = SYSCTL_OUT(req, ip6,
4391 			    xp->pr_ip6s * sizeof(struct in6_addr));
4392 			if (error)
4393 				break;
4394 		}
4395 #endif
4396 	}
4397 	sx_sunlock(&allprison_lock);
4398 	free(xp, M_TEMP);
4399 #ifdef INET
4400 	free(ip4, M_TEMP);
4401 #endif
4402 #ifdef INET6
4403 	free(ip6, M_TEMP);
4404 #endif
4405 	return (error);
4406 }
4407 
4408 SYSCTL_OID(_security_jail, OID_AUTO, list,
4409     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4410     sysctl_jail_list, "S", "List of active jails");
4411 
4412 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)4413 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4414 {
4415 	int error, injail;
4416 
4417 	injail = jailed(req->td->td_ucred);
4418 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4419 
4420 	return (error);
4421 }
4422 
4423 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4424     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4425     sysctl_jail_jailed, "I", "Process in jail?");
4426 
4427 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)4428 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4429 {
4430 	int error, havevnet;
4431 #ifdef VIMAGE
4432 	struct ucred *cred = req->td->td_ucred;
4433 
4434 	havevnet = jailed(cred) && prison_owns_vnet(cred->cr_prison);
4435 #else
4436 	havevnet = 0;
4437 #endif
4438 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4439 
4440 	return (error);
4441 }
4442 
4443 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4444     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4445     sysctl_jail_vnet, "I", "Jail owns vnet?");
4446 
4447 #if defined(INET) || defined(INET6)
4448 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4449     &jail_max_af_ips, 0,
4450     "Number of IP addresses a jail may have at most per address family (deprecated)");
4451 #endif
4452 
4453 /*
4454  * Default parameters for jail(2) compatibility.  For historical reasons,
4455  * the sysctl names have varying similarity to the parameter names.  Prisons
4456  * just see their own parameters, and can't change them.
4457  */
4458 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)4459 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4460 {
4461 	int error, i;
4462 
4463 	/* Get the current flag value, and convert it to a boolean. */
4464 	if (req->td->td_ucred->cr_prison == &prison0) {
4465 		mtx_lock(&prison0.pr_mtx);
4466 		i = (jail_default_allow & arg2) != 0;
4467 		mtx_unlock(&prison0.pr_mtx);
4468 	} else
4469 		i = prison_allow(req->td->td_ucred, arg2);
4470 
4471 	if (arg1 != NULL)
4472 		i = !i;
4473 	error = sysctl_handle_int(oidp, &i, 0, req);
4474 	if (error || !req->newptr)
4475 		return (error);
4476 	i = i ? arg2 : 0;
4477 	if (arg1 != NULL)
4478 		i ^= arg2;
4479 	/*
4480 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4481 	 * for writing.
4482 	 */
4483 	mtx_lock(&prison0.pr_mtx);
4484 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4485 	mtx_unlock(&prison0.pr_mtx);
4486 	return (0);
4487 }
4488 
4489 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4490     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4491     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4492     "Processes in jail can set their hostnames (deprecated)");
4493 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4494     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4495     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4496     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4497 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4498     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4499     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4500     "Processes in jail can use System V IPC primitives (deprecated)");
4501 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4502     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4503     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4504     "Prison root can create raw sockets (deprecated)");
4505 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4506     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4507     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4508     "Processes in jail can alter system file flags (deprecated)");
4509 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4510     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4511     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4512     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4513 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
4514     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4515     NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
4516     "Processes in jail can lock/unlock physical pages in memory");
4517 
4518 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)4519 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4520 {
4521 	struct prison *pr;
4522 	int level, error;
4523 
4524 	pr = req->td->td_ucred->cr_prison;
4525 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4526 	error = sysctl_handle_int(oidp, &level, 0, req);
4527 	if (error || !req->newptr)
4528 		return (error);
4529 	*(int *)arg1 = level;
4530 	return (0);
4531 }
4532 
4533 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4534     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4535     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4536     sysctl_jail_default_level, "I",
4537     "Processes in jail cannot see all mounted file systems (deprecated)");
4538 
4539 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4540     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4541     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4542     sysctl_jail_default_level, "I",
4543     "Ruleset for the devfs filesystem in jail (deprecated)");
4544 
4545 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4546     "Limits and stats of child jails");
4547 
4548 static int
sysctl_jail_children(SYSCTL_HANDLER_ARGS)4549 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
4550 {
4551 	struct prison *pr;
4552 	int i;
4553 
4554 	pr = req->td->td_ucred->cr_prison;
4555 
4556 	switch (oidp->oid_kind & CTLTYPE) {
4557 	case CTLTYPE_INT:
4558 		i = *(int *)((char *)pr + arg2);
4559 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4560 	}
4561 
4562 	return (0);
4563 }
4564 
4565 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
4566     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4567     NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
4568     "I", "Maximum number of child jails");
4569 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
4570     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4571     NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
4572     "I", "Current number of child jails");
4573 
4574 /*
4575  * Nodes to describe jail parameters.  Maximum length of string parameters
4576  * is returned in the string itself, and the other parameters exist merely
4577  * to make themselves and their types known.
4578  */
4579 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4580     "Jail parameters");
4581 
4582 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)4583 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4584 {
4585 	int i;
4586 	long l;
4587 	size_t s;
4588 	char numbuf[12];
4589 
4590 	switch (oidp->oid_kind & CTLTYPE)
4591 	{
4592 	case CTLTYPE_LONG:
4593 	case CTLTYPE_ULONG:
4594 		l = 0;
4595 #ifdef SCTL_MASK32
4596 		if (!(req->flags & SCTL_MASK32))
4597 #endif
4598 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4599 	case CTLTYPE_INT:
4600 	case CTLTYPE_UINT:
4601 		i = 0;
4602 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4603 	case CTLTYPE_STRING:
4604 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4605 		return
4606 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4607 	case CTLTYPE_STRUCT:
4608 		s = (size_t)arg2;
4609 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4610 	}
4611 	return (0);
4612 }
4613 
4614 /*
4615  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4616  * jail creation time but cannot be changed in an existing jail.
4617  */
4618 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4619 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4620 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4621 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4622 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4623     "I", "Jail secure level");
4624 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4625     "Jail value for kern.osreldate and uname -K");
4626 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4627     "Jail value for kern.osrelease and uname -r");
4628 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4629     "I", "Jail cannot see all mounted file systems");
4630 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4631     "I", "Ruleset for in-jail devfs mounts");
4632 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4633     "B", "Jail persistence");
4634 #ifdef VIMAGE
4635 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4636     "E,jailsys", "Virtual network stack");
4637 #endif
4638 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4639     "B", "Jail is in the process of shutting down");
4640 
4641 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4642 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4643     "I", "Current number of child jails");
4644 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4645     "I", "Maximum number of child jails");
4646 
4647 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4648 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4649     "Jail hostname");
4650 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4651     "Jail NIS domainname");
4652 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4653     "Jail host UUID");
4654 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4655     "LU", "Jail host ID");
4656 
4657 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4658 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4659 
4660 #ifdef INET
4661 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4662     "Jail IPv4 address virtualization");
4663 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4664     "S,in_addr,a", "Jail IPv4 addresses");
4665 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4666     "B", "Do (not) use IPv4 source address selection rather than the "
4667     "primary jail IPv4 address.");
4668 #endif
4669 #ifdef INET6
4670 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4671     "Jail IPv6 address virtualization");
4672 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4673     "S,in6_addr,a", "Jail IPv6 addresses");
4674 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4675     "B", "Do (not) use IPv6 source address selection rather than the "
4676     "primary jail IPv6 address.");
4677 #endif
4678 
4679 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4680 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4681     "B", "Jail may set hostname");
4682 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4683     "B", "Jail may use SYSV IPC");
4684 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4685     "B", "Jail may create raw sockets");
4686 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4687     "B", "Jail may alter system file flags");
4688 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4689     "B", "Jail may set file quotas");
4690 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4691     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4692 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4693     "B", "Jail may lock (unlock) physical pages in memory");
4694 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4695     "B", "Jail may bind sockets to reserved ports");
4696 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4697     "B", "Jail may read the kernel message buffer");
4698 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4699     "B", "Unprivileged processes may use process debugging facilities");
4700 SYSCTL_JAIL_PARAM(_allow, unprivileged_parent_tampering,
4701     CTLTYPE_INT | CTLFLAG_RW, "B",
4702     "Unprivileged parent jail processes may tamper with same-uid processes"
4703     " (signal/debug/cpuset)");
4704 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4705     "B", "Processes in jail with uid 0 have privilege");
4706 #ifdef VIMAGE
4707 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
4708     "B", "Mountd/nfsd may run in the jail");
4709 #endif
4710 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
4711     "B", "Jail may set system-level filesystem extended attributes");
4712 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
4713     "B", "Jail may adjust system time");
4714 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
4715     "B", "Jail may set system time");
4716 SYSCTL_JAIL_PARAM(_allow, routing, CTLTYPE_INT | CTLFLAG_RW,
4717     "B", "Jail may modify routing table");
4718 
4719 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4720 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4721     "B", "Jail may mount/unmount jail-friendly file systems in general");
4722 
4723 /*
4724  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
4725  * its associated bit in the pr_allow bitmask, or zero if the parameter was
4726  * not created.
4727  */
4728 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)4729 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4730     const char *descr)
4731 {
4732 	struct bool_flags *bf;
4733 	struct sysctl_oid *parent;
4734 	char *allow_name, *allow_noname, *allowed;
4735 #ifndef NO_SYSCTL_DESCR
4736 	char *descr_deprecated;
4737 #endif
4738 	u_int allow_flag;
4739 
4740 	if (prefix
4741 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4742 		< 0 ||
4743 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4744 		< 0
4745 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4746 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4747 		free(allow_name, M_PRISON);
4748 		return 0;
4749 	}
4750 
4751 	/*
4752 	 * See if this parameter has already beed added, i.e. a module was
4753 	 * previously loaded/unloaded.
4754 	 */
4755 	mtx_lock(&prison0.pr_mtx);
4756 	for (bf = pr_flag_allow;
4757 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
4758 		atomic_load_int(&bf->flag) != 0;
4759 	     bf++) {
4760 		if (strcmp(bf->name, allow_name) == 0) {
4761 			allow_flag = bf->flag;
4762 			goto no_add;
4763 		}
4764 	}
4765 
4766 	/*
4767 	 * Find a free bit in pr_allow_all, failing if there are none
4768 	 * (which shouldn't happen as long as we keep track of how many
4769 	 * potential dynamic flags exist).
4770 	 */
4771 	for (allow_flag = 1;; allow_flag <<= 1) {
4772 		if (allow_flag == 0)
4773 			goto no_add;
4774 		if ((pr_allow_all & allow_flag) == 0)
4775 			break;
4776 	}
4777 
4778 	/* Note the parameter in the next open slot in pr_flag_allow. */
4779 	for (bf = pr_flag_allow; ; bf++) {
4780 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4781 			/* This should never happen, but is not fatal. */
4782 			allow_flag = 0;
4783 			goto no_add;
4784 		}
4785 		if (atomic_load_int(&bf->flag) == 0)
4786 			break;
4787 	}
4788 	bf->name = allow_name;
4789 	bf->noname = allow_noname;
4790 	pr_allow_all |= allow_flag;
4791 	/*
4792 	 * prison0 always has permission for the new parameter.
4793 	 * Other jails must have it granted to them.
4794 	 */
4795 	prison0.pr_allow |= allow_flag;
4796 	/* The flag indicates a valid entry, so make sure it is set last. */
4797 	atomic_store_rel_int(&bf->flag, allow_flag);
4798 	mtx_unlock(&prison0.pr_mtx);
4799 
4800 	/*
4801 	 * Create sysctls for the parameter, and the back-compat global
4802 	 * permission.
4803 	 */
4804 	parent = prefix
4805 	    ? SYSCTL_ADD_NODE(NULL,
4806 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4807 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4808 	    : &sysctl___security_jail_param_allow;
4809 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4810 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4811 	    NULL, 0, sysctl_jail_param, "B", descr);
4812 	if ((prefix
4813 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4814 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4815 #ifndef NO_SYSCTL_DESCR
4816 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4817 		    descr);
4818 #endif
4819 		(void)SYSCTL_ADD_PROC(NULL,
4820 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4821 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4822 		    sysctl_jail_default_allow, "I", descr_deprecated);
4823 #ifndef NO_SYSCTL_DESCR
4824 		free(descr_deprecated, M_TEMP);
4825 #endif
4826 		free(allowed, M_TEMP);
4827 	}
4828 	return allow_flag;
4829 
4830  no_add:
4831 	mtx_unlock(&prison0.pr_mtx);
4832 	free(allow_name, M_PRISON);
4833 	free(allow_noname, M_PRISON);
4834 	return allow_flag;
4835 }
4836 
4837 /*
4838  * The VFS system will register jail-aware filesystems here.  They each get
4839  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4840  * attempts to mount.
4841  */
4842 void
prison_add_vfs(struct vfsconf * vfsp)4843 prison_add_vfs(struct vfsconf *vfsp)
4844 {
4845 #ifdef NO_SYSCTL_DESCR
4846 
4847 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4848 	    NULL, NULL);
4849 #else
4850 	char *descr;
4851 
4852 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4853 	    vfsp->vfc_name);
4854 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4855 	    NULL, descr);
4856 	free(descr, M_TEMP);
4857 #endif
4858 }
4859 
4860 #ifdef RACCT
4861 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)4862 prison_racct_foreach(void (*callback)(struct racct *racct,
4863     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4864     void *arg2, void *arg3)
4865 {
4866 	struct prison_racct *prr;
4867 
4868 	ASSERT_RACCT_ENABLED();
4869 
4870 	sx_slock(&allprison_lock);
4871 	if (pre != NULL)
4872 		(pre)();
4873 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4874 		(callback)(prr->prr_racct, arg2, arg3);
4875 	if (post != NULL)
4876 		(post)();
4877 	sx_sunlock(&allprison_lock);
4878 }
4879 
4880 static struct prison_racct *
prison_racct_find_locked(const char * name)4881 prison_racct_find_locked(const char *name)
4882 {
4883 	struct prison_racct *prr;
4884 
4885 	ASSERT_RACCT_ENABLED();
4886 	sx_assert(&allprison_lock, SA_XLOCKED);
4887 
4888 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4889 		return (NULL);
4890 
4891 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4892 		if (strcmp(name, prr->prr_name) != 0)
4893 			continue;
4894 
4895 		/* Found prison_racct with a matching name? */
4896 		prison_racct_hold(prr);
4897 		return (prr);
4898 	}
4899 
4900 	/* Add new prison_racct. */
4901 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4902 	racct_create(&prr->prr_racct);
4903 
4904 	strcpy(prr->prr_name, name);
4905 	refcount_init(&prr->prr_refcount, 1);
4906 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4907 
4908 	return (prr);
4909 }
4910 
4911 struct prison_racct *
prison_racct_find(const char * name)4912 prison_racct_find(const char *name)
4913 {
4914 	struct prison_racct *prr;
4915 
4916 	ASSERT_RACCT_ENABLED();
4917 
4918 	sx_xlock(&allprison_lock);
4919 	prr = prison_racct_find_locked(name);
4920 	sx_xunlock(&allprison_lock);
4921 	return (prr);
4922 }
4923 
4924 void
prison_racct_hold(struct prison_racct * prr)4925 prison_racct_hold(struct prison_racct *prr)
4926 {
4927 
4928 	ASSERT_RACCT_ENABLED();
4929 
4930 	refcount_acquire(&prr->prr_refcount);
4931 }
4932 
4933 static void
prison_racct_free_locked(struct prison_racct * prr)4934 prison_racct_free_locked(struct prison_racct *prr)
4935 {
4936 
4937 	ASSERT_RACCT_ENABLED();
4938 	sx_assert(&allprison_lock, SA_XLOCKED);
4939 
4940 	if (refcount_release(&prr->prr_refcount)) {
4941 		racct_destroy(&prr->prr_racct);
4942 		LIST_REMOVE(prr, prr_next);
4943 		free(prr, M_PRISON_RACCT);
4944 	}
4945 }
4946 
4947 void
prison_racct_free(struct prison_racct * prr)4948 prison_racct_free(struct prison_racct *prr)
4949 {
4950 
4951 	ASSERT_RACCT_ENABLED();
4952 	sx_assert(&allprison_lock, SA_UNLOCKED);
4953 
4954 	if (refcount_release_if_not_last(&prr->prr_refcount))
4955 		return;
4956 
4957 	sx_xlock(&allprison_lock);
4958 	prison_racct_free_locked(prr);
4959 	sx_xunlock(&allprison_lock);
4960 }
4961 
4962 static void
prison_racct_attach(struct prison * pr)4963 prison_racct_attach(struct prison *pr)
4964 {
4965 	struct prison_racct *prr;
4966 
4967 	ASSERT_RACCT_ENABLED();
4968 	sx_assert(&allprison_lock, SA_XLOCKED);
4969 
4970 	prr = prison_racct_find_locked(pr->pr_name);
4971 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4972 
4973 	pr->pr_prison_racct = prr;
4974 }
4975 
4976 /*
4977  * Handle jail renaming.  From the racct point of view, renaming means
4978  * moving from one prison_racct to another.
4979  */
4980 static void
prison_racct_modify(struct prison * pr)4981 prison_racct_modify(struct prison *pr)
4982 {
4983 #ifdef RCTL
4984 	struct proc *p;
4985 	struct ucred *cred;
4986 #endif
4987 	struct prison_racct *oldprr;
4988 
4989 	ASSERT_RACCT_ENABLED();
4990 
4991 	sx_slock(&allproc_lock);
4992 	sx_xlock(&allprison_lock);
4993 
4994 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4995 		sx_xunlock(&allprison_lock);
4996 		sx_sunlock(&allproc_lock);
4997 		return;
4998 	}
4999 
5000 	oldprr = pr->pr_prison_racct;
5001 	pr->pr_prison_racct = NULL;
5002 
5003 	prison_racct_attach(pr);
5004 
5005 	/*
5006 	 * Move resource utilisation records.
5007 	 */
5008 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
5009 
5010 #ifdef RCTL
5011 	/*
5012 	 * Force rctl to reattach rules to processes.
5013 	 */
5014 	FOREACH_PROC_IN_SYSTEM(p) {
5015 		PROC_LOCK(p);
5016 		cred = crhold(p->p_ucred);
5017 		PROC_UNLOCK(p);
5018 		rctl_proc_ucred_changed(p, cred);
5019 		crfree(cred);
5020 	}
5021 #endif
5022 
5023 	sx_sunlock(&allproc_lock);
5024 	prison_racct_free_locked(oldprr);
5025 	sx_xunlock(&allprison_lock);
5026 }
5027 
5028 static void
prison_racct_detach(struct prison * pr)5029 prison_racct_detach(struct prison *pr)
5030 {
5031 
5032 	ASSERT_RACCT_ENABLED();
5033 	sx_assert(&allprison_lock, SA_UNLOCKED);
5034 
5035 	if (pr->pr_prison_racct == NULL)
5036 		return;
5037 	prison_racct_free(pr->pr_prison_racct);
5038 	pr->pr_prison_racct = NULL;
5039 }
5040 #endif /* RACCT */
5041 
5042 #ifdef DDB
5043 
5044 static void
db_show_prison(struct prison * pr)5045 db_show_prison(struct prison *pr)
5046 {
5047 	struct bool_flags *bf;
5048 	struct jailsys_flags *jsf;
5049 #if defined(INET) || defined(INET6)
5050 	int ii;
5051 	struct prison_ip *pip;
5052 #endif
5053 	unsigned f;
5054 #ifdef INET
5055 	char ip4buf[INET_ADDRSTRLEN];
5056 #endif
5057 #ifdef INET6
5058 	char ip6buf[INET6_ADDRSTRLEN];
5059 #endif
5060 
5061 	db_printf("prison %p:\n", pr);
5062 	db_printf(" jid             = %d\n", pr->pr_id);
5063 	db_printf(" name            = %s\n", pr->pr_name);
5064 	db_printf(" parent          = %p\n", pr->pr_parent);
5065 	db_printf(" ref             = %d\n", pr->pr_ref);
5066 	db_printf(" uref            = %d\n", pr->pr_uref);
5067 	db_printf(" state           = %s\n",
5068 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5069 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
5070 	    "invalid");
5071 	db_printf(" path            = %s\n", pr->pr_path);
5072 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
5073 	    ? pr->pr_cpuset->cs_id : -1);
5074 #ifdef VIMAGE
5075 	db_printf(" vnet            = %p\n", pr->pr_vnet);
5076 #endif
5077 	db_printf(" root            = %p\n", pr->pr_root);
5078 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
5079 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
5080 	db_printf(" children.max    = %d\n", pr->pr_childmax);
5081 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
5082 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
5083 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
5084 	db_printf(" flags           = 0x%x", pr->pr_flags);
5085 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5086 		if (pr->pr_flags & bf->flag)
5087 			db_printf(" %s", bf->name);
5088 	for (jsf = pr_flag_jailsys;
5089 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5090 	     jsf++) {
5091 		f = pr->pr_flags & (jsf->disable | jsf->new);
5092 		db_printf(" %-16s= %s\n", jsf->name,
5093 		    (f != 0 && f == jsf->disable) ? "disable"
5094 		    : (f == jsf->new) ? "new"
5095 		    : "inherit");
5096 	}
5097 	db_printf(" allow           = 0x%x", pr->pr_allow);
5098 	for (bf = pr_flag_allow;
5099 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5100 		atomic_load_int(&bf->flag) != 0;
5101 	     bf++)
5102 		if (pr->pr_allow & bf->flag)
5103 			db_printf(" %s", bf->name);
5104 	db_printf("\n");
5105 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
5106 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
5107 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
5108 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
5109 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
5110 #ifdef INET
5111 	if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5112 		db_printf(" ip4s            = %d\n", pip->ips);
5113 		for (ii = 0; ii < pip->ips; ii++)
5114 			db_printf(" %s %s\n",
5115 			    ii == 0 ? "ip4.addr        =" : "                 ",
5116 			    inet_ntoa_r(
5117 			    *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5118 			    ip4buf));
5119 	}
5120 #endif
5121 #ifdef INET6
5122 	if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5123 		db_printf(" ip6s            = %d\n", pip->ips);
5124 		for (ii = 0; ii < pip->ips; ii++)
5125 			db_printf(" %s %s\n",
5126 			    ii == 0 ? "ip6.addr        =" : "                 ",
5127 			    ip6_sprintf(ip6buf,
5128 			    (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5129 	}
5130 #endif
5131 }
5132 
DB_SHOW_COMMAND(prison,db_show_prison_command)5133 DB_SHOW_COMMAND(prison, db_show_prison_command)
5134 {
5135 	struct prison *pr;
5136 
5137 	if (!have_addr) {
5138 		/*
5139 		 * Show all prisons in the list, and prison0 which is not
5140 		 * listed.
5141 		 */
5142 		db_show_prison(&prison0);
5143 		if (!db_pager_quit) {
5144 			TAILQ_FOREACH(pr, &allprison, pr_list) {
5145 				db_show_prison(pr);
5146 				if (db_pager_quit)
5147 					break;
5148 			}
5149 		}
5150 		return;
5151 	}
5152 
5153 	if (addr == 0)
5154 		pr = &prison0;
5155 	else {
5156 		/* Look for a prison with the ID and with references. */
5157 		TAILQ_FOREACH(pr, &allprison, pr_list)
5158 			if (pr->pr_id == addr && pr->pr_ref > 0)
5159 				break;
5160 		if (pr == NULL)
5161 			/* Look again, without requiring a reference. */
5162 			TAILQ_FOREACH(pr, &allprison, pr_list)
5163 				if (pr->pr_id == addr)
5164 					break;
5165 		if (pr == NULL)
5166 			/* Assume address points to a valid prison. */
5167 			pr = (struct prison *)addr;
5168 	}
5169 	db_show_prison(pr);
5170 }
5171 
5172 #endif /* DDB */
5173