xref: /src/sys/kern/kern_jail.c (revision c71354030a26900e564f0c80a8abdff7e77b3c9e)
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/ctype.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/file.h>
44 #include <sys/sysproto.h>
45 #include <sys/malloc.h>
46 #include <sys/osd.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/epoch.h>
50 #include <sys/event.h>
51 #include <sys/taskqueue.h>
52 #include <sys/fcntl.h>
53 #include <sys/jail.h>
54 #include <sys/jaildesc.h>
55 #include <sys/linker.h>
56 #include <sys/lock.h>
57 #include <sys/mman.h>
58 #include <sys/mutex.h>
59 #include <sys/racct.h>
60 #include <sys/rctl.h>
61 #include <sys/refcount.h>
62 #include <sys/sx.h>
63 #include <sys/sysent.h>
64 #include <sys/namei.h>
65 #include <sys/mount.h>
66 #include <sys/queue.h>
67 #include <sys/socket.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/uuid.h>
71 #include <sys/vnode.h>
72 
73 #include <net/if.h>
74 #include <net/vnet.h>
75 
76 #include <netinet/in.h>
77 
78 #ifdef DDB
79 #include <ddb/ddb.h>
80 #endif /* DDB */
81 
82 #include <security/mac/mac_framework.h>
83 #include <security/mac/mac_policy.h>
84 #include <security/mac/mac_syscalls.h>
85 
86 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
87 
88 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
89 #ifdef RACCT
90 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
91 #endif
92 
93 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
94 #ifdef INET
95 #ifdef INET6
96 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
97 #else
98 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
99 #endif
100 #else /* !INET */
101 #ifdef INET6
102 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
103 #else
104 #define	_PR_IP_SADDRSEL	0
105 #endif
106 #endif
107 
108 /* prison0 describes what is "real" about the system. */
109 struct prison prison0 = {
110 	.pr_id		= 0,
111 	.pr_name	= "0",
112 	.pr_ref		= 1,
113 	.pr_uref	= 1,
114 	.pr_path	= "/",
115 	.pr_securelevel	= -1,
116 	.pr_devfs_rsnum = 0,
117 	.pr_state	= PRISON_STATE_ALIVE,
118 	.pr_childmax	= JAIL_MAX,
119 	.pr_hostuuid	= DEFAULT_HOSTUUID,
120 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
121 #ifdef VIMAGE
122 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
123 #else
124 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
125 #endif
126 	.pr_allow	= PR_ALLOW_PRISON0,
127 };
128 _Static_assert((PR_ALLOW_PRISON0 & ~PR_ALLOW_ALL_STATIC) == 0,
129     "Bits enabled in PR_ALLOW_PRISON0 that are not statically reserved");
130 
131 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
132 
133 struct bool_flags {
134 	const char	*name;
135 	const char	*noname;
136 	volatile u_int	 flag;
137 };
138 struct jailsys_flags {
139 	const char	*name;
140 	unsigned	 disable;
141 	unsigned	 new;
142 };
143 
144 /*
145  * Handle jail teardown in a dedicated thread to avoid deadlocks from
146  * vnet_destroy().
147  */
148 TASKQUEUE_DEFINE_THREAD(jail_remove);
149 
150 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
151 struct	sx allprison_lock;
152 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
153 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
154 LIST_HEAD(, prison_racct) allprison_racct;
155 int	lastprid = 0;
156 int	lastdeadid = 0;
157 
158 static int get_next_prid(struct prison **insprp);
159 static int get_next_deadid(struct prison **insprp);
160 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
161 static void prison_complete(void *context, int pending);
162 static void prison_deref(struct prison *pr, int flags);
163 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
164 static int prison_lock_xlock(struct prison *pr, int flags);
165 static void prison_cleanup_locked(struct prison *pr);
166 static void prison_cleanup_unlocked(struct prison *pr);
167 static void prison_free_not_last(struct prison *pr);
168 static void prison_proc_free_not_last(struct prison *pr);
169 static void prison_proc_relink(struct prison *opr, struct prison *npr,
170     struct proc *p);
171 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
172     int enable);
173 static char *prison_path(struct prison *pr1, struct prison *pr2);
174 #ifdef RACCT
175 static void prison_racct_attach(struct prison *pr);
176 static void prison_racct_modify(struct prison *pr);
177 static void prison_racct_detach(struct prison *pr);
178 #endif
179 static void prison_knote(struct prison *pr, long hint);
180 
181 /* Flags for prison_deref */
182 #define	PD_DEREF	0x01	/* Decrement pr_ref */
183 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
184 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
185 #define	PD_LOCKED	0x10	/* pr_mtx is held */
186 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
187 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
188 #define PD_OP_FLAGS	0x07	/* Operation flags */
189 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
190 
191 /*
192  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
193  * as we cannot figure out the size of a sparse array, or an array without a
194  * terminating entry.
195  */
196 static struct bool_flags pr_flag_bool[] = {
197 	{"persist", "nopersist", PR_PERSIST},
198 #ifdef INET
199 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
200 #endif
201 #ifdef INET6
202 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
203 #endif
204 };
205 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
206 
207 static struct jailsys_flags pr_flag_jailsys[] = {
208 	{"host", 0, PR_HOST},
209 #ifdef VIMAGE
210 	{"vnet", 0, PR_VNET},
211 #endif
212 #ifdef INET
213 	{"ip4", PR_IP4_USER, PR_IP4_USER},
214 #endif
215 #ifdef INET6
216 	{"ip6", PR_IP6_USER, PR_IP6_USER},
217 #endif
218 };
219 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
220 
221 /*
222  * Make this array full-size so dynamic parameters can be added.
223  * It is protected by prison0.mtx, but lockless reading is allowed
224  * with an atomic check of the flag values.
225  */
226 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
227 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
228 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
229 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
230 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
231 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
232 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
233 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
234 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
235 	{"allow.reserved_ports", "allow.noreserved_ports",
236 	 PR_ALLOW_RESERVED_PORTS},
237 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
238 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
239 	 PR_ALLOW_UNPRIV_DEBUG},
240 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
241 #ifdef VIMAGE
242 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
243 #endif
244 	{"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
245 	{"allow.adjtime", "allow.noadjtime", PR_ALLOW_ADJTIME},
246 	{"allow.settime", "allow.nosettime", PR_ALLOW_SETTIME},
247 	{"allow.routing", "allow.norouting", PR_ALLOW_ROUTING},
248 	{"allow.unprivileged_parent_tampering",
249 	    "allow.nounprivileged_parent_tampering",
250 	    PR_ALLOW_UNPRIV_PARENT_TAMPER},
251 #ifdef AUDIT
252 	{"allow.setaudit", "allow.nosetaudit", PR_ALLOW_SETAUDIT},
253 #endif
254 };
255 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
256 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
257 
258 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
259 					 PR_ALLOW_RESERVED_PORTS | \
260 					 PR_ALLOW_UNPRIV_DEBUG | \
261 					 PR_ALLOW_SUSER)
262 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
263 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
264 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
265 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
266 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
267 #if defined(INET) || defined(INET6)
268 static unsigned jail_max_af_ips = 255;
269 #endif
270 
271 /*
272  * Initialize the parts of prison0 that can't be static-initialized with
273  * constants.  This is called from proc0_init() after creating thread0 cpuset.
274  */
275 void
prison0_init(void)276 prison0_init(void)
277 {
278 	uint8_t *file, *data;
279 	size_t size;
280 	char buf[sizeof(prison0.pr_hostuuid)];
281 #ifdef MAC
282 	int error __diagused;
283 #endif
284 	bool valid;
285 
286 #ifdef MAC
287 	error = mac_prison_init(&prison0, M_WAITOK);
288 	MPASS(error == 0);
289 
290 	mtx_unlock(&prison0.pr_mtx);
291 #endif
292 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
293 	prison0.pr_osreldate = osreldate;
294 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
295 
296 	/* If we have a preloaded hostuuid, use it. */
297 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
298 	if (file != NULL) {
299 		data = preload_fetch_addr(file);
300 		size = preload_fetch_size(file);
301 		if (data != NULL) {
302 			/*
303 			 * The preloaded data may include trailing whitespace, almost
304 			 * certainly a newline; skip over any whitespace or
305 			 * non-printable characters to be safe.
306 			 */
307 			while (size > 0 && data[size - 1] <= 0x20) {
308 				size--;
309 			}
310 
311 			valid = false;
312 
313 			/*
314 			 * Not NUL-terminated when passed from loader, but
315 			 * validate_uuid requires that due to using sscanf (as
316 			 * does the subsequent strlcpy, since it still reads
317 			 * past the given size to return the true length);
318 			 * bounce to a temporary buffer to fix.
319 			 */
320 			if (size >= sizeof(buf))
321 				goto done;
322 
323 			memcpy(buf, data, size);
324 			buf[size] = '\0';
325 
326 			if (validate_uuid(buf, size, NULL, 0) != 0)
327 				goto done;
328 
329 			valid = true;
330 			(void)strlcpy(prison0.pr_hostuuid, buf,
331 			    sizeof(prison0.pr_hostuuid));
332 
333 done:
334 			if (bootverbose && !valid) {
335 				printf("hostuuid: preload data malformed: '%.*s'\n",
336 				    (int)size, data);
337 			}
338 		}
339 	}
340 	if (bootverbose)
341 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
342 }
343 
344 /*
345  * struct jail_args {
346  *	struct jail *jail;
347  * };
348  */
349 int
sys_jail(struct thread * td,struct jail_args * uap)350 sys_jail(struct thread *td, struct jail_args *uap)
351 {
352 	uint32_t version;
353 	int error;
354 	struct jail j;
355 
356 	error = copyin(uap->jail, &version, sizeof(uint32_t));
357 	if (error)
358 		return (error);
359 
360 	switch (version) {
361 	case 0:
362 	{
363 		struct jail_v0 j0;
364 
365 		/* FreeBSD single IPv4 jails. */
366 		bzero(&j, sizeof(struct jail));
367 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
368 		if (error)
369 			return (error);
370 		j.version = j0.version;
371 		j.path = j0.path;
372 		j.hostname = j0.hostname;
373 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
374 		break;
375 	}
376 
377 	case 1:
378 		/*
379 		 * Version 1 was used by multi-IPv4 jail implementations
380 		 * that never made it into the official kernel.
381 		 */
382 		return (EINVAL);
383 
384 	case 2:	/* JAIL_API_VERSION */
385 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
386 		error = copyin(uap->jail, &j, sizeof(struct jail));
387 		if (error)
388 			return (error);
389 		break;
390 
391 	default:
392 		/* Sci-Fi jails are not supported, sorry. */
393 		return (EINVAL);
394 	}
395 	return (kern_jail(td, &j));
396 }
397 
398 int
kern_jail(struct thread * td,struct jail * j)399 kern_jail(struct thread *td, struct jail *j)
400 {
401 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
402 #ifdef INET
403 			    + 1
404 #endif
405 #ifdef INET6
406 			    + 1
407 #endif
408 			    )];
409 	struct uio opt;
410 	char *u_path, *u_hostname, *u_name;
411 	struct bool_flags *bf;
412 #ifdef INET
413 	uint32_t ip4s;
414 	struct in_addr *u_ip4;
415 #endif
416 #ifdef INET6
417 	struct in6_addr *u_ip6;
418 #endif
419 	size_t tmplen;
420 	int error, enforce_statfs;
421 
422 	bzero(&optiov, sizeof(optiov));
423 	opt.uio_iov = optiov;
424 	opt.uio_iovcnt = 0;
425 	opt.uio_offset = -1;
426 	opt.uio_resid = -1;
427 	opt.uio_segflg = UIO_SYSSPACE;
428 	opt.uio_rw = UIO_READ;
429 	opt.uio_td = td;
430 
431 	/* Set permissions for top-level jails from sysctls. */
432 	if (!jailed(td->td_ucred)) {
433 		for (bf = pr_flag_allow;
434 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
435 			atomic_load_int(&bf->flag) != 0;
436 		     bf++) {
437 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
438 			    (jail_default_allow & bf->flag)
439 			    ? bf->name : bf->noname);
440 			optiov[opt.uio_iovcnt].iov_len =
441 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
442 			opt.uio_iovcnt += 2;
443 		}
444 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
445 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
446 		opt.uio_iovcnt++;
447 		enforce_statfs = jail_default_enforce_statfs;
448 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
449 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
450 		opt.uio_iovcnt++;
451 	}
452 
453 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
454 #ifdef INET
455 	ip4s = (j->version == 0) ? 1 : j->ip4s;
456 	if (ip4s > jail_max_af_ips)
457 		return (EINVAL);
458 	tmplen += ip4s * sizeof(struct in_addr);
459 #else
460 	if (j->ip4s > 0)
461 		return (EINVAL);
462 #endif
463 #ifdef INET6
464 	if (j->ip6s > jail_max_af_ips)
465 		return (EINVAL);
466 	tmplen += j->ip6s * sizeof(struct in6_addr);
467 #else
468 	if (j->ip6s > 0)
469 		return (EINVAL);
470 #endif
471 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
472 	u_hostname = u_path + MAXPATHLEN;
473 	u_name = u_hostname + MAXHOSTNAMELEN;
474 #ifdef INET
475 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
476 #endif
477 #ifdef INET6
478 #ifdef INET
479 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
480 #else
481 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
482 #endif
483 #endif
484 	optiov[opt.uio_iovcnt].iov_base = "path";
485 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
486 	opt.uio_iovcnt++;
487 	optiov[opt.uio_iovcnt].iov_base = u_path;
488 	error = copyinstr(j->path, u_path, MAXPATHLEN,
489 	    &optiov[opt.uio_iovcnt].iov_len);
490 	if (error) {
491 		free(u_path, M_TEMP);
492 		return (error);
493 	}
494 	opt.uio_iovcnt++;
495 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
496 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
497 	opt.uio_iovcnt++;
498 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
499 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
500 	    &optiov[opt.uio_iovcnt].iov_len);
501 	if (error) {
502 		free(u_path, M_TEMP);
503 		return (error);
504 	}
505 	opt.uio_iovcnt++;
506 	if (j->jailname != NULL) {
507 		optiov[opt.uio_iovcnt].iov_base = "name";
508 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
509 		opt.uio_iovcnt++;
510 		optiov[opt.uio_iovcnt].iov_base = u_name;
511 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
512 		    &optiov[opt.uio_iovcnt].iov_len);
513 		if (error) {
514 			free(u_path, M_TEMP);
515 			return (error);
516 		}
517 		opt.uio_iovcnt++;
518 	}
519 #ifdef INET
520 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
521 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
522 	opt.uio_iovcnt++;
523 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
524 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
525 	if (j->version == 0)
526 		u_ip4->s_addr = j->ip4s;
527 	else {
528 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
529 		if (error) {
530 			free(u_path, M_TEMP);
531 			return (error);
532 		}
533 	}
534 	opt.uio_iovcnt++;
535 #endif
536 #ifdef INET6
537 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
538 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
539 	opt.uio_iovcnt++;
540 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
541 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
542 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
543 	if (error) {
544 		free(u_path, M_TEMP);
545 		return (error);
546 	}
547 	opt.uio_iovcnt++;
548 #endif
549 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
550 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
551 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
552 	free(u_path, M_TEMP);
553 	return (error);
554 }
555 
556 /*
557  * struct jail_set_args {
558  *	struct iovec *iovp;
559  *	unsigned int iovcnt;
560  *	int flags;
561  * };
562  */
563 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)564 sys_jail_set(struct thread *td, struct jail_set_args *uap)
565 {
566 	struct uio *auio;
567 	int error;
568 
569 	/* Check that we have an even number of iovecs. */
570 	if (uap->iovcnt & 1)
571 		return (EINVAL);
572 
573 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
574 	if (error)
575 		return (error);
576 	error = kern_jail_set(td, auio, uap->flags);
577 	freeuio(auio);
578 	return (error);
579 }
580 
581 #if defined(INET) || defined(INET6)
582 typedef int prison_addr_cmp_t(const void *, const void *);
583 typedef bool prison_addr_valid_t(const void *);
584 static const struct pr_family {
585 	size_t			size;
586 	prison_addr_cmp_t	*cmp;
587 	prison_addr_valid_t	*valid;
588 	int			ip_flag;
589 } pr_families[PR_FAMILY_MAX] = {
590 #ifdef INET
591 	[PR_INET] = {
592 		.size = sizeof(struct in_addr),
593 		.cmp = prison_qcmp_v4,
594 		.valid = prison_valid_v4,
595 		.ip_flag = PR_IP4_USER,
596 	 },
597 #endif
598 #ifdef INET6
599 	[PR_INET6] = {
600 		.size = sizeof(struct in6_addr),
601 		.cmp = prison_qcmp_v6,
602 		.valid = prison_valid_v6,
603 		.ip_flag = PR_IP6_USER,
604 	},
605 #endif
606 };
607 
608 /*
609  * Network address lists (pr_addrs) allocation for jails.  The addresses
610  * are accessed locklessly by the network stack, thus need to be protected by
611  * the network epoch.
612  */
613 struct prison_ip {
614 	struct epoch_context ctx;
615 	uint32_t	ips;
616 #ifdef FUTURE_C
617 	/*
618 	 * XXX Variable-length automatic arrays in union may be
619 	 * supported in future C.
620 	 */
621 	union {
622 		char pr_ip[];
623 		struct in_addr pr_ip4[];
624 		struct in6_addr pr_ip6[];
625 	};
626 #else /* No future C :( */
627 	char pr_ip[];
628 #endif
629 };
630 
631 static char *
PR_IP(struct prison_ip * pip,const pr_family_t af,int idx)632 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
633 {
634 	MPASS(pip);
635 	MPASS(af < PR_FAMILY_MAX);
636 	MPASS(idx >= 0 && idx < pip->ips);
637 
638 	return (pip->pr_ip + pr_families[af].size * idx);
639 }
640 
641 static struct prison_ip *
prison_ip_alloc(const pr_family_t af,uint32_t cnt,int flags)642 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
643 {
644 	struct prison_ip *pip;
645 
646 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
647 	    M_PRISON, flags);
648 	if (pip != NULL)
649 		pip->ips = cnt;
650 	return (pip);
651 }
652 
653 /*
654  * Allocate and copyin user supplied address list, sorting and validating.
655  * kern_jail_set() helper.
656  */
657 static struct prison_ip *
prison_ip_copyin(const pr_family_t af,void * op,uint32_t cnt)658 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
659 {
660 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
661 	const size_t size = pr_families[af].size;
662 	struct prison_ip *pip;
663 
664 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
665 	bcopy(op, pip->pr_ip, cnt * size);
666 	/*
667 	 * IP addresses are all sorted but ip[0] to preserve
668 	 * the primary IP address as given from userland.
669 	 * This special IP is used for unbound outgoing
670 	 * connections as well for "loopback" traffic in case
671 	 * source address selection cannot find any more fitting
672 	 * address to connect from.
673 	 */
674 	if (cnt > 1)
675 		qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
676 	/*
677 	 * Check for duplicate addresses and do some simple
678 	 * zero and broadcast checks. If users give other bogus
679 	 * addresses it is their problem.
680 	 */
681 	for (int i = 0; i < cnt; i++) {
682 		if (!pr_families[af].valid(PR_IP(pip, af, i))) {
683 			free(pip, M_PRISON);
684 			return (NULL);
685 		}
686 		if (i + 1 < cnt &&
687 		    (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
688 		     cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
689 			free(pip, M_PRISON);
690 			return (NULL);
691 		}
692 	}
693 
694 	return (pip);
695 }
696 
697 /*
698  * Allocate and dup parent prison address list.
699  * kern_jail_set() helper.
700  */
701 static void
prison_ip_dup(struct prison * ppr,struct prison * pr,const pr_family_t af)702 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
703 {
704 	const struct prison_ip *ppip = ppr->pr_addrs[af];
705 	struct prison_ip *pip;
706 
707 	if (ppip != NULL) {
708 		pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
709 		bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
710 		pr->pr_addrs[af] = pip;
711 	}
712 }
713 
714 /*
715  * Make sure the new set of IP addresses is a subset of the parent's list.
716  * Don't worry about the parent being unlocked, as any setting is done with
717  * allprison_lock held.
718  * kern_jail_set() helper.
719  */
720 static bool
prison_ip_parent_match(struct prison_ip * ppip,struct prison_ip * pip,const pr_family_t af)721 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
722     const pr_family_t af)
723 {
724 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
725 	int i, j;
726 
727 	if (ppip == NULL)
728 		return (false);
729 
730 	for (i = 0; i < ppip->ips; i++)
731 		if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
732 			break;
733 
734 	if (i == ppip->ips)
735 		/* Main address not present in parent. */
736 		return (false);
737 
738 	if (pip->ips > 1) {
739 		for (i = j = 1; i < pip->ips; i++) {
740 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
741 				/* Equals to parent primary address. */
742 				continue;
743 			for (; j < ppip->ips; j++)
744 				if (cmp(PR_IP(pip, af, i),
745 				    PR_IP(ppip, af, j)) == 0)
746 					break;
747 			if (j == ppip->ips)
748 				break;
749 		}
750 		if (j == ppip->ips)
751 			/* Address not present in parent. */
752 			return (false);
753 	}
754 	return (true);
755 }
756 
757 /*
758  * Check for conflicting IP addresses.  We permit them if there is no more
759  * than one IP on each jail.  If there is a duplicate on a jail with more
760  * than one IP stop checking and return error.
761  * kern_jail_set() helper.
762  */
763 static bool
prison_ip_conflict_check(const struct prison * ppr,const struct prison * pr,struct prison_ip * pip,pr_family_t af)764 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
765     struct prison_ip *pip, pr_family_t af)
766 {
767 	const struct prison *tppr, *tpr;
768 	int descend;
769 
770 #ifdef VIMAGE
771 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
772 		if (tppr->pr_flags & PR_VNET)
773 			break;
774 #else
775 	tppr = &prison0;
776 #endif
777 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
778 		if (tpr == pr ||
779 #ifdef VIMAGE
780 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
781 #endif
782 		    !prison_isalive(tpr)) {
783 			descend = 0;
784 			continue;
785 		}
786 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
787 			continue;
788 		descend = 0;
789 		if (tpr->pr_addrs[af] == NULL ||
790 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
791 			continue;
792 		for (int i = 0; i < pip->ips; i++)
793 			if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
794 				return (false);
795 	}
796 
797 	return (true);
798 }
799 
800 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
801     "prison must start with epoch context");
802 static void
prison_ip_free_deferred(epoch_context_t ctx)803 prison_ip_free_deferred(epoch_context_t ctx)
804 {
805 
806 	free(ctx, M_PRISON);
807 }
808 
809 static void
prison_ip_free(struct prison_ip * pip)810 prison_ip_free(struct prison_ip *pip)
811 {
812 
813 	if (pip != NULL)
814 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
815 }
816 
817 static void
prison_ip_set(struct prison * pr,const pr_family_t af,struct prison_ip * new)818 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
819 {
820 	struct prison_ip **mem, *old;
821 
822 	mtx_assert(&pr->pr_mtx, MA_OWNED);
823 
824 	mem = &pr->pr_addrs[af];
825 
826 	old = *mem;
827 	atomic_store_ptr(mem, new);
828 	prison_ip_free(old);
829 }
830 
831 /*
832  * Restrict a prison's IP address list with its parent's, possibly replacing
833  * it.  Return true if succeed, otherwise should redo.
834  * kern_jail_set() helper.
835  */
836 static bool
prison_ip_restrict(struct prison * pr,const pr_family_t af,struct prison_ip ** newp)837 prison_ip_restrict(struct prison *pr, const pr_family_t af,
838     struct prison_ip **newp)
839 {
840 	struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
841 	struct prison_ip *pip = pr->pr_addrs[af];
842 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
843 	const size_t size = pr_families[af].size;
844 	struct prison_ip *new = newp != NULL ? *newp : NULL;
845 	uint32_t ips;
846 
847 	mtx_assert(&pr->pr_mtx, MA_OWNED);
848 
849 	/*
850 	 * Due to epoch-synchronized access to the IP address lists we always
851 	 * allocate a new list even if the old one has enough space.  We could
852 	 * atomically update an IPv4 address inside a list, but that would
853 	 * screw up sorting, and in case of IPv6 we can't even atomically write
854 	 * one.
855 	 */
856 	if (ppip == NULL) {
857 		if (pip != NULL)
858 			prison_ip_set(pr, af, NULL);
859 		return (true);
860 	}
861 
862 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
863 		if (new == NULL) {
864 			new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
865 			if (new == NULL)
866 				return (false); /* Redo */
867 		}
868 		/* This has no user settings, so just copy the parent's list. */
869 		MPASS(new->ips == ppip->ips);
870 		bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
871 		prison_ip_set(pr, af, new);
872 		if (newp != NULL)
873 			*newp = NULL; /* Used */
874 	} else if (pip != NULL) {
875 		/* Remove addresses that aren't in the parent. */
876 		int i;
877 
878 		i = 0; /* index in pip */
879 		ips = 0; /* index in new */
880 
881 		if (new == NULL) {
882 			new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
883 			if (new == NULL)
884 				return (false); /* Redo */
885 		}
886 
887 		for (int pi = 0; pi < ppip->ips; pi++)
888 			if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
889 				/* Found our primary address in parent. */
890 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
891 				    size);
892 				i++;
893 				ips++;
894 				break;
895 			}
896 		for (int pi = 1; i < pip->ips; ) {
897 			/* Check against primary, which is unsorted. */
898 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
899 				/* Matches parent's primary address. */
900 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
901 				    size);
902 				i++;
903 				ips++;
904 				continue;
905 			}
906 			/* The rest are sorted. */
907 			switch (pi >= ppip->ips ? -1 :
908 				cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
909 			case -1:
910 				i++;
911 				break;
912 			case 0:
913 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
914 				    size);
915 				i++;
916 				pi++;
917 				ips++;
918 				break;
919 			case 1:
920 				pi++;
921 				break;
922 			}
923 		}
924 		if (ips == 0) {
925 			if (newp == NULL || *newp == NULL)
926 				prison_ip_free(new);
927 			new = NULL;
928 		} else {
929 			/* Shrink to real size */
930 			KASSERT((new->ips >= ips),
931 			    ("Out-of-bounds write to prison_ip %p", new));
932 			new->ips = ips;
933 		}
934 		prison_ip_set(pr, af, new);
935 		if (newp != NULL)
936 			*newp = NULL; /* Used */
937 	}
938 	return (true);
939 }
940 
941 /*
942  * Fast-path check if an address belongs to a prison.
943  */
944 int
prison_ip_check(const struct prison * pr,const pr_family_t af,const void * addr)945 prison_ip_check(const struct prison *pr, const pr_family_t af,
946     const void *addr)
947 {
948 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
949 	struct prison_ip *pip;
950 	int i, a, z, d;
951 
952 	MPASS(mtx_owned(&pr->pr_mtx) ||
953 	    in_epoch(net_epoch_preempt) ||
954 	    sx_xlocked(&allprison_lock));
955 
956 	pip = atomic_load_ptr(&pr->pr_addrs[af]);
957 	if (__predict_false(pip == NULL))
958 		return (EAFNOSUPPORT);
959 
960 	/* Check the primary IP. */
961 	if (cmp(PR_IP(pip, af, 0), addr) == 0)
962 		return (0);
963 
964 	/*
965 	 * All the other IPs are sorted so we can do a binary search.
966 	 */
967 	a = 0;
968 	z = pip->ips - 2;
969 	while (a <= z) {
970 		i = (a + z) / 2;
971 		d = cmp(PR_IP(pip, af, i + 1), addr);
972 		if (d > 0)
973 			z = i - 1;
974 		else if (d < 0)
975 			a = i + 1;
976 		else
977 			return (0);
978 	}
979 
980 	return (EADDRNOTAVAIL);
981 }
982 
983 /*
984  * Grab primary IP.  Historically required mutex, but nothing prevents
985  * us to support epoch-protected access.  Is it used in fast path?
986  * in{6}_jail.c helper
987  */
988 const void *
prison_ip_get0(const struct prison * pr,const pr_family_t af)989 prison_ip_get0(const struct prison *pr, const pr_family_t af)
990 {
991 	const struct prison_ip *pip = pr->pr_addrs[af];
992 
993 	mtx_assert(&pr->pr_mtx, MA_OWNED);
994 	MPASS(pip);
995 
996 	return (pip->pr_ip);
997 }
998 
999 u_int
prison_ip_cnt(const struct prison * pr,const pr_family_t af)1000 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
1001 {
1002 
1003 	return (pr->pr_addrs[af]->ips);
1004 }
1005 #endif	/* defined(INET) || defined(INET6) */
1006 
1007 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)1008 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
1009 {
1010 	struct file *jfp_out;
1011 	struct nameidata nd;
1012 #ifdef INET
1013 	struct prison_ip *ip4;
1014 #endif
1015 #ifdef INET6
1016 	struct prison_ip *ip6;
1017 #endif
1018 	struct vfsopt *opt;
1019 	struct vfsoptlist *opts;
1020 	struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
1021 	struct ucred *jdcred;
1022 	struct vnode *root;
1023 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
1024 	char *g_path, *osrelstr;
1025 	struct bool_flags *bf;
1026 	struct jailsys_flags *jsf;
1027 #if defined(INET) || defined(INET6)
1028 	void *op;
1029 #endif
1030 	unsigned long hid;
1031 	size_t namelen, onamelen, pnamelen;
1032 #ifdef MAC
1033 	void *mac_set_prison_data = NULL;
1034 	int gotmaclabel;
1035 #endif
1036 	int created, cuflags, descend, drflags, enforce;
1037 	int error, errmsg_len, errmsg_pos;
1038 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1039 	int deadid, jfd_in, jfd_out, jfd_pos, jid, jsys, len, level;
1040 	int childmax, osreldt, rsnum, slevel;
1041 #ifdef INET
1042 	int ip4s;
1043 	bool redo_ip4;
1044 #endif
1045 #ifdef INET6
1046 	int ip6s;
1047 	bool redo_ip6;
1048 #endif
1049 	bool maybe_changed;
1050 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1051 	uint64_t pr_allow_diff;
1052 	unsigned tallow;
1053 	char numbuf[12];
1054 
1055 	mypr = td->td_ucred->cr_prison;
1056 	if (((flags & (JAIL_CREATE | JAIL_AT_DESC)) == JAIL_CREATE) &&
1057 	    mypr->pr_childmax == 0)
1058 		return (EPERM);
1059 	if (flags & ~JAIL_SET_MASK)
1060 		return (EINVAL);
1061 	if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
1062 	    (JAIL_USE_DESC | JAIL_AT_DESC))
1063 		return (EINVAL);
1064 	prison_hold(mypr);
1065 
1066 #ifdef INET
1067 	ip4 = NULL;
1068 #endif
1069 #ifdef INET6
1070 	ip6 = NULL;
1071 #endif
1072 	g_path = NULL;
1073 	jfp_out = NULL;
1074 	jfd_out = -1;
1075 	/*
1076 	 * Check all the parameters before committing to anything.  Not all
1077 	 * errors can be caught early, but we may as well try.  Also, this
1078 	 * takes care of some expensive stuff (path lookup) before getting
1079 	 * the allprison lock.
1080 	 *
1081 	 * XXX Jails are not filesystems, and jail parameters are not mount
1082 	 *     options.  But it makes more sense to re-use the vfsopt code
1083 	 *     than duplicate it under a different name.
1084 	 */
1085 	error = vfs_buildopts(optuio, &opts);
1086 	if (error) {
1087 		opts = NULL;
1088 		goto done_free;
1089 	}
1090 
1091 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1092 	if (!cuflags) {
1093 		error = EINVAL;
1094 		vfs_opterror(opts, "no valid operation (create or update)");
1095 		goto done_errmsg;
1096 	}
1097 
1098 	error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
1099 	if (error == ENOENT) {
1100 		if (flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1101 		    JAIL_OWN_DESC)) {
1102 			vfs_opterror(opts, "missing desc");
1103 			goto done_errmsg;
1104 		}
1105 		jfd_in = -1;
1106 	} else if (error != 0)
1107 		goto done_free;
1108 	else {
1109 		if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1110 		    JAIL_OWN_DESC))) {
1111 			error = EINVAL;
1112 			vfs_opterror(opts, "unexpected desc");
1113 			goto done_errmsg;
1114 		}
1115 		if (flags & JAIL_AT_DESC) {
1116 			/*
1117 			 * Look up and create jails based on the
1118 			 * descriptor's prison.
1119 			 */
1120 			prison_free(mypr);
1121 			error = jaildesc_find(td, jfd_in, &mypr, NULL);
1122 			if (error != 0) {
1123 				vfs_opterror(opts, error == ENOENT ?
1124 				    "descriptor to dead jail" :
1125 				    "not a jail descriptor");
1126 				goto done_errmsg;
1127 			}
1128 			if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) {
1129 				error = EPERM;
1130 				goto done_free;
1131 			}
1132 		}
1133 		if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
1134 			/* Allocate a jail descriptor to return later. */
1135 			error = jaildesc_alloc(td, &jfp_out, &jfd_out,
1136 				flags & JAIL_OWN_DESC);
1137 			if (error)
1138 				goto done_free;
1139 		}
1140 	}
1141 
1142 	/*
1143 	 * Delay the permission check if using a jail descriptor,
1144 	 * until we get the descriptor's credentials.
1145 	 */
1146 	if (!(flags & JAIL_USE_DESC)) {
1147 		error = priv_check(td, PRIV_JAIL_SET);
1148 		if (error == 0 && (flags & JAIL_ATTACH))
1149 			error = priv_check(td, PRIV_JAIL_ATTACH);
1150 		if (error)
1151 			goto done_free;
1152 	}
1153 
1154 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1155 	if (error == ENOENT)
1156 		jid = 0;
1157 	else if (error != 0)
1158 		goto done_free;
1159 
1160 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1161 	if (error == ENOENT)
1162 		gotslevel = 0;
1163 	else if (error != 0)
1164 		goto done_free;
1165 	else
1166 		gotslevel = 1;
1167 
1168 	error =
1169 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1170 	if (error == ENOENT)
1171 		gotchildmax = 0;
1172 	else if (error != 0)
1173 		goto done_free;
1174 	else
1175 		gotchildmax = 1;
1176 
1177 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1178 	if (error == ENOENT)
1179 		gotenforce = 0;
1180 	else if (error != 0)
1181 		goto done_free;
1182 	else if (enforce < 0 || enforce > 2) {
1183 		error = EINVAL;
1184 		goto done_free;
1185 	} else
1186 		gotenforce = 1;
1187 
1188 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1189 	if (error == ENOENT)
1190 		gotrsnum = 0;
1191 	else if (error != 0)
1192 		goto done_free;
1193 	else
1194 		gotrsnum = 1;
1195 
1196 	pr_flags = ch_flags = 0;
1197 	for (bf = pr_flag_bool;
1198 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1199 	     bf++) {
1200 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1201 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1202 	}
1203 	ch_flags |= pr_flags;
1204 	for (jsf = pr_flag_jailsys;
1205 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1206 	     jsf++) {
1207 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1208 		if (error == ENOENT)
1209 			continue;
1210 		if (error != 0)
1211 			goto done_free;
1212 		switch (jsys) {
1213 		case JAIL_SYS_DISABLE:
1214 			if (!jsf->disable) {
1215 				error = EINVAL;
1216 				goto done_free;
1217 			}
1218 			pr_flags |= jsf->disable;
1219 			break;
1220 		case JAIL_SYS_NEW:
1221 			pr_flags |= jsf->new;
1222 			break;
1223 		case JAIL_SYS_INHERIT:
1224 			break;
1225 		default:
1226 			error = EINVAL;
1227 			goto done_free;
1228 		}
1229 		ch_flags |= jsf->new | jsf->disable;
1230 	}
1231 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1232 	    && !(pr_flags & PR_PERSIST)) {
1233 		error = EINVAL;
1234 		vfs_opterror(opts, "new jail must persist or attach");
1235 		goto done_errmsg;
1236 	}
1237 #ifdef VIMAGE
1238 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1239 		error = EINVAL;
1240 		vfs_opterror(opts, "vnet cannot be changed after creation");
1241 		goto done_errmsg;
1242 	}
1243 #endif
1244 #ifdef INET
1245 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1246 		error = EINVAL;
1247 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1248 		goto done_errmsg;
1249 	}
1250 #endif
1251 #ifdef INET6
1252 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1253 		error = EINVAL;
1254 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1255 		goto done_errmsg;
1256 	}
1257 #endif
1258 
1259 	pr_allow = ch_allow = 0;
1260 	for (bf = pr_flag_allow;
1261 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1262 		atomic_load_int(&bf->flag) != 0;
1263 	     bf++) {
1264 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1265 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1266 	}
1267 	ch_allow |= pr_allow;
1268 
1269 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1270 	if (error == ENOENT)
1271 		name = NULL;
1272 	else if (error != 0)
1273 		goto done_free;
1274 	else {
1275 		if (len == 0 || name[len - 1] != '\0') {
1276 			error = EINVAL;
1277 			goto done_free;
1278 		}
1279 		if (len > MAXHOSTNAMELEN) {
1280 			error = ENAMETOOLONG;
1281 			goto done_free;
1282 		}
1283 	}
1284 
1285 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1286 	if (error == ENOENT)
1287 		host = NULL;
1288 	else if (error != 0)
1289 		goto done_free;
1290 	else {
1291 		ch_flags |= PR_HOST;
1292 		pr_flags |= PR_HOST;
1293 		if (len == 0 || host[len - 1] != '\0') {
1294 			error = EINVAL;
1295 			goto done_free;
1296 		}
1297 		if (len > MAXHOSTNAMELEN) {
1298 			error = ENAMETOOLONG;
1299 			goto done_free;
1300 		}
1301 	}
1302 
1303 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1304 	if (error == ENOENT)
1305 		domain = NULL;
1306 	else if (error != 0)
1307 		goto done_free;
1308 	else {
1309 		ch_flags |= PR_HOST;
1310 		pr_flags |= PR_HOST;
1311 		if (len == 0 || domain[len - 1] != '\0') {
1312 			error = EINVAL;
1313 			goto done_free;
1314 		}
1315 		if (len > MAXHOSTNAMELEN) {
1316 			error = ENAMETOOLONG;
1317 			goto done_free;
1318 		}
1319 	}
1320 
1321 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1322 	if (error == ENOENT)
1323 		uuid = NULL;
1324 	else if (error != 0)
1325 		goto done_free;
1326 	else {
1327 		ch_flags |= PR_HOST;
1328 		pr_flags |= PR_HOST;
1329 		if (len == 0 || uuid[len - 1] != '\0') {
1330 			error = EINVAL;
1331 			goto done_free;
1332 		}
1333 		if (len > HOSTUUIDLEN) {
1334 			error = ENAMETOOLONG;
1335 			goto done_free;
1336 		}
1337 	}
1338 
1339 #ifdef COMPAT_FREEBSD32
1340 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1341 		uint32_t hid32;
1342 
1343 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1344 		hid = hid32;
1345 	} else
1346 #endif
1347 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1348 	if (error == ENOENT)
1349 		gothid = 0;
1350 	else if (error != 0)
1351 		goto done_free;
1352 	else {
1353 		gothid = 1;
1354 		ch_flags |= PR_HOST;
1355 		pr_flags |= PR_HOST;
1356 	}
1357 
1358 #ifdef MAC
1359 	/* Process the mac.label vfsopt */
1360 	error = mac_set_prison_prepare(td, opts, &mac_set_prison_data);
1361 	if (error == ENOENT)
1362 		gotmaclabel = 0;
1363 	else if (error != 0)
1364 		goto done_errmsg;
1365 	else
1366 		gotmaclabel = 1;
1367 #endif
1368 
1369 #ifdef INET
1370 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1371 	if (error == ENOENT)
1372 		ip4s = 0;
1373 	else if (error != 0)
1374 		goto done_free;
1375 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1376 		error = EINVAL;
1377 		goto done_free;
1378 	} else {
1379 		ch_flags |= PR_IP4_USER;
1380 		pr_flags |= PR_IP4_USER;
1381 		if (ip4s > 0) {
1382 			ip4s /= sizeof(struct in_addr);
1383 			if (ip4s > jail_max_af_ips) {
1384 				error = EINVAL;
1385 				vfs_opterror(opts, "too many IPv4 addresses");
1386 				goto done_errmsg;
1387 			}
1388 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1389 			if (ip4 == NULL) {
1390 				error = EINVAL;
1391 				goto done_free;
1392 			}
1393 		}
1394 	}
1395 #endif
1396 
1397 #ifdef INET6
1398 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1399 	if (error == ENOENT)
1400 		ip6s = 0;
1401 	else if (error != 0)
1402 		goto done_free;
1403 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1404 		error = EINVAL;
1405 		goto done_free;
1406 	} else {
1407 		ch_flags |= PR_IP6_USER;
1408 		pr_flags |= PR_IP6_USER;
1409 		if (ip6s > 0) {
1410 			ip6s /= sizeof(struct in6_addr);
1411 			if (ip6s > jail_max_af_ips) {
1412 				error = EINVAL;
1413 				vfs_opterror(opts, "too many IPv6 addresses");
1414 				goto done_errmsg;
1415 			}
1416 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1417 			if (ip6 == NULL) {
1418 				error = EINVAL;
1419 				goto done_free;
1420 			}
1421 		}
1422 	}
1423 #endif
1424 
1425 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1426 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1427 		error = EINVAL;
1428 		vfs_opterror(opts,
1429 		    "vnet jails cannot have IP address restrictions");
1430 		goto done_errmsg;
1431 	}
1432 #endif
1433 
1434 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1435 	if (error == ENOENT)
1436 		osrelstr = NULL;
1437 	else if (error != 0)
1438 		goto done_free;
1439 	else {
1440 		if (flags & JAIL_UPDATE) {
1441 			error = EINVAL;
1442 			vfs_opterror(opts,
1443 			    "osrelease cannot be changed after creation");
1444 			goto done_errmsg;
1445 		}
1446 		if (len == 0 || osrelstr[len - 1] != '\0') {
1447 			error = EINVAL;
1448 			goto done_free;
1449 		}
1450 		if (len >= OSRELEASELEN) {
1451 			error = ENAMETOOLONG;
1452 			vfs_opterror(opts,
1453 			    "osrelease string must be 1-%d bytes long",
1454 			    OSRELEASELEN - 1);
1455 			goto done_errmsg;
1456 		}
1457 	}
1458 
1459 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1460 	if (error == ENOENT)
1461 		osreldt = 0;
1462 	else if (error != 0)
1463 		goto done_free;
1464 	else {
1465 		if (flags & JAIL_UPDATE) {
1466 			error = EINVAL;
1467 			vfs_opterror(opts,
1468 			    "osreldate cannot be changed after creation");
1469 			goto done_errmsg;
1470 		}
1471 		if (osreldt == 0) {
1472 			error = EINVAL;
1473 			vfs_opterror(opts, "osreldate cannot be 0");
1474 			goto done_errmsg;
1475 		}
1476 	}
1477 
1478 	root = NULL;
1479 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1480 	if (error == ENOENT)
1481 		path = NULL;
1482 	else if (error != 0)
1483 		goto done_free;
1484 	else {
1485 		if (flags & JAIL_UPDATE) {
1486 			error = EINVAL;
1487 			vfs_opterror(opts,
1488 			    "path cannot be changed after creation");
1489 			goto done_errmsg;
1490 		}
1491 		if (len == 0 || path[len - 1] != '\0') {
1492 			error = EINVAL;
1493 			goto done_free;
1494 		}
1495 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1496 		error = namei(&nd);
1497 		if (error)
1498 			goto done_free;
1499 		root = nd.ni_vp;
1500 		NDFREE_PNBUF(&nd);
1501 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1502 		strlcpy(g_path, path, MAXPATHLEN);
1503 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1504 		if (error == 0) {
1505 			path = g_path;
1506 		} else {
1507 			/* exit on other errors */
1508 			goto done_free;
1509 		}
1510 		if (root->v_type != VDIR) {
1511 			error = ENOTDIR;
1512 			vput(root);
1513 			goto done_free;
1514 		}
1515 		VOP_UNLOCK(root);
1516 	}
1517 
1518 	/*
1519 	 * Find the specified jail, or at least its parent.
1520 	 * This abuses the file error codes ENOENT and EEXIST.
1521 	 */
1522 	pr = NULL;
1523 	inspr = NULL;
1524 	deadpr = NULL;
1525 	maybe_changed = false;
1526 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1527 		namelc = strrchr(name, '.');
1528 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1529 		if (*p != '\0')
1530 			jid = 0;
1531 	}
1532 	sx_xlock(&allprison_lock);
1533 	drflags = PD_LIST_XLOCKED;
1534 	ppr = mypr;
1535 	if (!prison_isalive(ppr)) {
1536 		/* This jail is dying.  This process will surely follow. */
1537 		error = EAGAIN;
1538 		goto done_deref;
1539 	}
1540 	if (flags & JAIL_USE_DESC) {
1541 		/* Get the jail from its descriptor. */
1542 		error = jaildesc_find(td, jfd_in, &pr, &jdcred);
1543 		if (error) {
1544 			vfs_opterror(opts, error == ENOENT ?
1545 			    "descriptor to dead jail" :
1546 			    "not a jail descriptor");
1547 			goto done_deref;
1548 		}
1549 		drflags |= PD_DEREF;
1550 		error = priv_check_cred(jdcred, PRIV_JAIL_SET);
1551 		if (error == 0 && (flags & JAIL_ATTACH))
1552 			error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
1553 		crfree(jdcred);
1554 		if (error)
1555 			goto done_deref;
1556 		mtx_lock(&pr->pr_mtx);
1557 		drflags |= PD_LOCKED;
1558 		if (cuflags == JAIL_CREATE) {
1559 			error = EEXIST;
1560 			vfs_opterror(opts, "jail %d already exists",
1561 			    pr->pr_id);
1562 			goto done_deref;
1563 		}
1564 		if (!prison_isalive(pr)) {
1565 			/* While a jid can be resurrected, the prison
1566 			 * itself cannot.
1567 			 */
1568 			error = ENOENT;
1569 			vfs_opterror(opts, "jail %d is dying", pr->pr_id);
1570 			goto done_deref;
1571 		}
1572 		if (jid != 0 && jid != pr->pr_id) {
1573 			error = EINVAL;
1574 			vfs_opterror(opts, "cannot change jid");
1575 			goto done_deref;
1576 		}
1577 		jid = pr->pr_id;
1578 	} else if (jid != 0) {
1579 		if (jid < 0) {
1580 			error = EINVAL;
1581 			vfs_opterror(opts, "negative jid");
1582 			goto done_deref;
1583 		}
1584 		/*
1585 		 * See if a requested jid already exists.  Keep track of
1586 		 * where it can be inserted later.
1587 		 */
1588 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1589 			if (inspr->pr_id < jid)
1590 				continue;
1591 			if (inspr->pr_id > jid)
1592 				break;
1593 			if (prison_isalive(inspr)) {
1594 				pr = inspr;
1595 				mtx_lock(&pr->pr_mtx);
1596 				drflags |= PD_LOCKED;
1597 			} else {
1598 				/* Note a dying jail to handle later. */
1599 				deadpr = inspr;
1600 			}
1601 			inspr = NULL;
1602 			break;
1603 		}
1604 		if (cuflags == JAIL_CREATE && pr != NULL) {
1605 			/*
1606 			 * Even creators that cannot see the jail will
1607 			 * get EEXIST.
1608 			 */
1609 			error = EEXIST;
1610 			vfs_opterror(opts, "jail %d already exists", jid);
1611 			goto done_deref;
1612 		}
1613 		if ((pr == NULL)
1614 		    ? cuflags == JAIL_UPDATE
1615 		    : !prison_ischild(mypr, pr)) {
1616 			/*
1617 			 * Updaters get ENOENT for nonexistent jails,
1618 			 * or for jails they cannot see.  The latter
1619 			 * case is true even for CREATE | UPDATE,
1620 			 * which normally cannot give this error.
1621 			 */
1622 			error = ENOENT;
1623 			vfs_opterror(opts, "jail %d not found", jid);
1624 			goto done_deref;
1625 		}
1626 	}
1627 	/*
1628 	 * If the caller provided a name, look for a jail by that name.
1629 	 * This has different semantics for creates and updates keyed by jid
1630 	 * (where the name must not already exist in a different jail),
1631 	 * and updates keyed by the name itself (where the name must exist
1632 	 * because that is the jail being updated).
1633 	 */
1634 	namelc = NULL;
1635 	if (name != NULL) {
1636 		namelc = strrchr(name, '.');
1637 		if (namelc == NULL)
1638 			namelc = name;
1639 		else {
1640 			/*
1641 			 * This is a hierarchical name.  Split it into the
1642 			 * parent and child names, and make sure the parent
1643 			 * exists or matches an already found jail.
1644 			 */
1645 			if (pr != NULL) {
1646 				if (strncmp(name, ppr->pr_name, namelc - name)
1647 				    || ppr->pr_name[namelc - name] != '\0') {
1648 					error = EINVAL;
1649 					vfs_opterror(opts,
1650 					    "cannot change jail's parent");
1651 					goto done_deref;
1652 				}
1653 			} else {
1654 				*namelc = '\0';
1655 				ppr = prison_find_name(mypr, name);
1656 				if (ppr == NULL) {
1657 					error = ENOENT;
1658 					vfs_opterror(opts,
1659 					    "jail \"%s\" not found", name);
1660 					goto done_deref;
1661 				}
1662 				mtx_unlock(&ppr->pr_mtx);
1663 				if (!prison_isalive(ppr)) {
1664 					error = ENOENT;
1665 					vfs_opterror(opts,
1666 					    "jail \"%s\" is dying", name);
1667 					goto done_deref;
1668 				}
1669 				*namelc = '.';
1670 			}
1671 			namelc++;
1672 		}
1673 		if (namelc[0] != '\0') {
1674 			pnamelen =
1675 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1676 			FOREACH_PRISON_CHILD(ppr, tpr) {
1677 				if (tpr == pr || !prison_isalive(tpr) ||
1678 				    strcmp(tpr->pr_name + pnamelen, namelc))
1679 					continue;
1680 				if (cuflags == JAIL_CREATE || pr != NULL) {
1681 					/*
1682 					 * Create, or update(jid): name must
1683 					 * not exist in an active sibling jail.
1684 					 */
1685 					error = EEXIST;
1686 					vfs_opterror(opts,
1687 					    "jail \"%s\" already exists", name);
1688 					goto done_deref;
1689 				}
1690 				/* Use this jail for updates. */
1691 				pr = tpr;
1692 				mtx_lock(&pr->pr_mtx);
1693 				drflags |= PD_LOCKED;
1694 				break;
1695 			}
1696 			/*
1697 			 * Update: name must exist if no jid is specified.
1698 			 * As with the jid case, the jail must be currently
1699 			 * visible, or else even CREATE | UPDATE will get
1700 			 * an error.
1701 			 */
1702 			if ((pr == NULL)
1703 			    ? cuflags == JAIL_UPDATE
1704 			    : !prison_isalive(pr)) {
1705 				error = ENOENT;
1706 				vfs_opterror(opts, "jail \"%s\" not found",
1707 				    name);
1708 				goto done_deref;
1709 			}
1710 		}
1711 	}
1712 	/* Update: must provide a desc, jid, or name. */
1713 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1714 		error = ENOENT;
1715 		vfs_opterror(opts, "update specified no jail");
1716 		goto done_deref;
1717 	}
1718 
1719 	/* If there's no prison to update, create a new one and link it in. */
1720 	created = pr == NULL;
1721 	if (created) {
1722 #ifdef MAC
1723 		error = mac_prison_check_create(td->td_ucred, opts, flags);
1724 		if (error != 0)
1725 			goto done_deref;
1726 #endif
1727 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1728 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1729 				error = EPERM;
1730 				vfs_opterror(opts, "prison limit exceeded");
1731 				goto done_deref;
1732 			}
1733 
1734 		if (deadpr != NULL) {
1735 			/*
1736 			 * The prison being created has the same ID as a dying
1737 			 * one.  Handle this by giving the dying jail a new ID.
1738 			 * This may cause some confusion to user space, but
1739 			 * only to those listing dying jails.
1740 			 */
1741 			deadid = get_next_deadid(&dinspr);
1742 			if (deadid == 0) {
1743 				error = EAGAIN;
1744 				vfs_opterror(opts, "no available jail IDs");
1745 				goto done_deref;
1746 			}
1747 			mtx_lock(&deadpr->pr_mtx);
1748 			deadpr->pr_id = deadid;
1749 			mtx_unlock(&deadpr->pr_mtx);
1750 			if (dinspr == deadpr)
1751 				inspr = deadpr;
1752 			else {
1753 				inspr = TAILQ_NEXT(deadpr, pr_list);
1754 				TAILQ_REMOVE(&allprison, deadpr, pr_list);
1755 				if (dinspr != NULL)
1756 					TAILQ_INSERT_AFTER(&allprison, dinspr,
1757 					    deadpr, pr_list);
1758 				else
1759 					TAILQ_INSERT_HEAD(&allprison, deadpr,
1760 					    pr_list);
1761 			}
1762 		}
1763 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1764 			error = EAGAIN;
1765 			vfs_opterror(opts, "no available jail IDs");
1766 			goto done_deref;
1767 		}
1768 
1769 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1770 		pr->pr_state = PRISON_STATE_INVALID;
1771 		refcount_init(&pr->pr_ref, 1);
1772 		refcount_init(&pr->pr_uref, 0);
1773 		drflags |= PD_DEREF;
1774 		LIST_INIT(&pr->pr_children);
1775 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1776 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1777 
1778 		pr->pr_id = jid;
1779 		if (inspr != NULL)
1780 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1781 		else
1782 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1783 
1784 		pr->pr_parent = ppr;
1785 		prison_hold(ppr);
1786 		prison_proc_hold(ppr);
1787 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1788 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1789 			tpr->pr_childcount++;
1790 		pr->pr_klist = knlist_alloc(&pr->pr_mtx);
1791 
1792 		/* Set some default values, and inherit some from the parent. */
1793 		if (namelc == NULL)
1794 			namelc = "";
1795 		if (path == NULL) {
1796 			path = "/";
1797 			root = mypr->pr_root;
1798 			vref(root);
1799 		}
1800 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1801 		pr->pr_flags |= PR_HOST;
1802 #if defined(INET) || defined(INET6)
1803 #ifdef VIMAGE
1804 		if (!(pr_flags & PR_VNET))
1805 #endif
1806 		{
1807 #ifdef INET
1808 			if (!(ch_flags & PR_IP4_USER))
1809 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1810 			else if (!(pr_flags & PR_IP4_USER)) {
1811 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1812 				prison_ip_dup(ppr, pr, PR_INET);
1813 			}
1814 #endif
1815 #ifdef INET6
1816 			if (!(ch_flags & PR_IP6_USER))
1817 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1818 			else if (!(pr_flags & PR_IP6_USER)) {
1819 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1820 				prison_ip_dup(ppr, pr, PR_INET6);
1821 			}
1822 #endif
1823 		}
1824 #endif
1825 		/* Source address selection is always on by default. */
1826 		pr->pr_flags |= _PR_IP_SADDRSEL;
1827 
1828 		pr->pr_securelevel = ppr->pr_securelevel;
1829 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1830 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1831 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1832 
1833 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1834 		if (osrelstr == NULL)
1835 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1836 			    sizeof(pr->pr_osrelease));
1837 		else
1838 			strlcpy(pr->pr_osrelease, osrelstr,
1839 			    sizeof(pr->pr_osrelease));
1840 
1841 #ifdef VIMAGE
1842 		/*
1843 		 * Allocate a new vnet if specified.
1844 		 *
1845 		 * Set PR_VNET now if so, so that the vnet is disposed of
1846 		 * properly when the jail is destroyed.
1847 		 */
1848 		if (pr_flags & PR_VNET) {
1849 			pr->pr_flags |= PR_VNET;
1850 			pr->pr_vnet = vnet_alloc();
1851 		} else {
1852 			pr->pr_vnet = ppr->pr_vnet;
1853 		}
1854 #endif
1855 		/*
1856 		 * Allocate a dedicated cpuset for each jail.
1857 		 * Unlike other initial settings, this may return an error.
1858 		 */
1859 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1860 		if (error)
1861 			goto done_deref;
1862 
1863 #ifdef MAC
1864 		error = mac_prison_init(pr, M_WAITOK);
1865 		MPASS(error == 0);
1866 
1867 		mtx_assert(&pr->pr_mtx, MA_OWNED);
1868 #else
1869 		mtx_lock(&pr->pr_mtx);
1870 #endif
1871 		drflags |= PD_LOCKED;
1872 	} else {
1873 		/*
1874 		 * Grab a reference for existing prisons, to ensure they
1875 		 * continue to exist for the duration of the call.
1876 		 */
1877 		if (!(drflags & PD_DEREF)) {
1878 			prison_hold(pr);
1879 			drflags |= PD_DEREF;
1880 		}
1881 #ifdef MAC
1882 		error = mac_prison_check_set(td->td_ucred, pr, opts, flags);
1883 		if (error != 0)
1884 			goto done_deref;
1885 #endif
1886 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1887 		if ((pr->pr_flags & PR_VNET) &&
1888 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1889 			error = EINVAL;
1890 			vfs_opterror(opts,
1891 			    "vnet jails cannot have IP address restrictions");
1892 			goto done_deref;
1893 		}
1894 #endif
1895 #ifdef INET
1896 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1897 			error = EINVAL;
1898 			vfs_opterror(opts,
1899 			    "ip4 cannot be changed after creation");
1900 			goto done_deref;
1901 		}
1902 #endif
1903 #ifdef INET6
1904 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1905 			error = EINVAL;
1906 			vfs_opterror(opts,
1907 			    "ip6 cannot be changed after creation");
1908 			goto done_deref;
1909 		}
1910 #endif
1911 	}
1912 
1913 	/* Do final error checking before setting anything. */
1914 	if (gotslevel) {
1915 		if (slevel < ppr->pr_securelevel) {
1916 			error = EPERM;
1917 			goto done_deref;
1918 		}
1919 	}
1920 	if (gotchildmax) {
1921 		if (childmax >= ppr->pr_childmax) {
1922 			error = EPERM;
1923 			goto done_deref;
1924 		}
1925 	}
1926 	if (gotenforce) {
1927 		if (enforce < ppr->pr_enforce_statfs) {
1928 			error = EPERM;
1929 			goto done_deref;
1930 		}
1931 	}
1932 	if (gotrsnum) {
1933 		/*
1934 		 * devfs_rsnum is a uint16_t
1935 		 */
1936 		if (rsnum < 0 || rsnum > 65535) {
1937 			error = EINVAL;
1938 			goto done_deref;
1939 		}
1940 		/*
1941 		 * Nested jails always inherit parent's devfs ruleset
1942 		 */
1943 		if (jailed(td->td_ucred)) {
1944 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1945 				error = EPERM;
1946 				goto done_deref;
1947 			} else
1948 				rsnum = ppr->pr_devfs_rsnum;
1949 		}
1950 	}
1951 #ifdef INET
1952 	if (ip4s > 0) {
1953 		if ((ppr->pr_flags & PR_IP4) &&
1954 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1955 		    PR_INET)) {
1956 			error = EPERM;
1957 			goto done_deref;
1958 		}
1959 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1960 			error = EADDRINUSE;
1961 			vfs_opterror(opts, "IPv4 addresses clash");
1962 			goto done_deref;
1963 		}
1964 	}
1965 #endif
1966 #ifdef INET6
1967 	if (ip6s > 0) {
1968 		if ((ppr->pr_flags & PR_IP6) &&
1969 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1970 		    PR_INET6)) {
1971 			error = EPERM;
1972 			goto done_deref;
1973 		}
1974 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1975 			error = EADDRINUSE;
1976 			vfs_opterror(opts, "IPv6 addresses clash");
1977 			goto done_deref;
1978 		}
1979 	}
1980 #endif
1981 	onamelen = namelen = 0;
1982 	if (namelc != NULL) {
1983 		/* Give a default name of the jid.  Also allow the name to be
1984 		 * explicitly the jid - but not any other number, and only in
1985 		 * normal form (no leading zero/etc).
1986 		 */
1987 		if (namelc[0] == '\0')
1988 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1989 		else if ((strtoul(namelc, &p, 10) != jid ||
1990 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1991 			error = EINVAL;
1992 			vfs_opterror(opts,
1993 			    "name cannot be numeric (unless it is the jid)");
1994 			goto done_deref;
1995 		}
1996 		/*
1997 		 * Make sure the name isn't too long for the prison or its
1998 		 * children.
1999 		 */
2000 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
2001 		onamelen = strlen(pr->pr_name + pnamelen);
2002 		namelen = strlen(namelc);
2003 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
2004 			error = ENAMETOOLONG;
2005 			goto done_deref;
2006 		}
2007 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
2008 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
2009 			    sizeof(pr->pr_name)) {
2010 				error = ENAMETOOLONG;
2011 				goto done_deref;
2012 			}
2013 		}
2014 	}
2015 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
2016 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
2017 		error = EPERM;
2018 		goto done_deref;
2019 	}
2020 
2021 	/*
2022 	 * Let modules check their parameters.  This requires unlocking and
2023 	 * then re-locking the prison, but this is still a valid state as long
2024 	 * as allprison_lock remains xlocked.
2025 	 */
2026 	mtx_unlock(&pr->pr_mtx);
2027 	drflags &= ~PD_LOCKED;
2028 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
2029 	if (error != 0)
2030 		goto done_deref;
2031 	mtx_lock(&pr->pr_mtx);
2032 	drflags |= PD_LOCKED;
2033 
2034 	/* At this point, all valid parameters should have been noted. */
2035 	TAILQ_FOREACH(opt, opts, link) {
2036 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2037 			error = EINVAL;
2038 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2039 			goto done_deref;
2040 		}
2041 	}
2042 	maybe_changed = true;
2043 
2044 	/* Set the parameters of the prison. */
2045 #ifdef INET
2046 	redo_ip4 = false;
2047 	if (pr_flags & PR_IP4_USER) {
2048 		pr->pr_flags |= PR_IP4;
2049 		prison_ip_set(pr, PR_INET, ip4);
2050 		ip4 = NULL;
2051 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2052 #ifdef VIMAGE
2053 			if (tpr->pr_flags & PR_VNET) {
2054 				descend = 0;
2055 				continue;
2056 			}
2057 #endif
2058 			if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
2059 				redo_ip4 = true;
2060 				descend = 0;
2061 			}
2062 		}
2063 	}
2064 #endif
2065 #ifdef INET6
2066 	redo_ip6 = false;
2067 	if (pr_flags & PR_IP6_USER) {
2068 		pr->pr_flags |= PR_IP6;
2069 		prison_ip_set(pr, PR_INET6, ip6);
2070 		ip6 = NULL;
2071 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2072 #ifdef VIMAGE
2073 			if (tpr->pr_flags & PR_VNET) {
2074 				descend = 0;
2075 				continue;
2076 			}
2077 #endif
2078 			if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
2079 				redo_ip6 = true;
2080 				descend = 0;
2081 			}
2082 		}
2083 	}
2084 #endif
2085 	if (gotslevel) {
2086 		pr->pr_securelevel = slevel;
2087 		/* Set all child jails to be at least this level. */
2088 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2089 			if (tpr->pr_securelevel < slevel)
2090 				tpr->pr_securelevel = slevel;
2091 	}
2092 	if (gotchildmax) {
2093 		pr->pr_childmax = childmax;
2094 		/* Set all child jails to under this limit. */
2095 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
2096 			if (tpr->pr_childmax > childmax - level)
2097 				tpr->pr_childmax = childmax > level
2098 				    ? childmax - level : 0;
2099 	}
2100 	if (gotenforce) {
2101 		pr->pr_enforce_statfs = enforce;
2102 		/* Pass this restriction on to the children. */
2103 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2104 			if (tpr->pr_enforce_statfs < enforce)
2105 				tpr->pr_enforce_statfs = enforce;
2106 	}
2107 	if (gotrsnum) {
2108 		pr->pr_devfs_rsnum = rsnum;
2109 		/* Pass this restriction on to the children. */
2110 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2111 			tpr->pr_devfs_rsnum = rsnum;
2112 	}
2113 	if (namelc != NULL) {
2114 		if (ppr == &prison0)
2115 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
2116 		else
2117 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
2118 			    ppr->pr_name, namelc);
2119 		/* Change this component of child names. */
2120 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2121 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
2122 			    strlen(tpr->pr_name + onamelen) + 1);
2123 			bcopy(pr->pr_name, tpr->pr_name, namelen);
2124 		}
2125 	}
2126 	if (path != NULL) {
2127 		/* Try to keep a real-rooted full pathname. */
2128 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
2129 		pr->pr_root = root;
2130 		root = NULL;
2131 	}
2132 	if (PR_HOST & ch_flags & ~pr_flags) {
2133 		if (pr->pr_flags & PR_HOST) {
2134 			/*
2135 			 * Copy the parent's host info.  As with pr_ip4 above,
2136 			 * the lack of a lock on the parent is not a problem;
2137 			 * it is always set with allprison_lock at least
2138 			 * shared, and is held exclusively here.
2139 			 */
2140 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
2141 			    sizeof(pr->pr_hostname));
2142 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
2143 			    sizeof(pr->pr_domainname));
2144 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
2145 			    sizeof(pr->pr_hostuuid));
2146 			pr->pr_hostid = pr->pr_parent->pr_hostid;
2147 		}
2148 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
2149 		/* Set this prison, and any descendants without PR_HOST. */
2150 		if (host != NULL)
2151 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
2152 		if (domain != NULL)
2153 			strlcpy(pr->pr_domainname, domain,
2154 			    sizeof(pr->pr_domainname));
2155 		if (uuid != NULL)
2156 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
2157 		if (gothid)
2158 			pr->pr_hostid = hid;
2159 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2160 			if (tpr->pr_flags & PR_HOST)
2161 				descend = 0;
2162 			else {
2163 				if (host != NULL)
2164 					strlcpy(tpr->pr_hostname,
2165 					    pr->pr_hostname,
2166 					    sizeof(tpr->pr_hostname));
2167 				if (domain != NULL)
2168 					strlcpy(tpr->pr_domainname,
2169 					    pr->pr_domainname,
2170 					    sizeof(tpr->pr_domainname));
2171 				if (uuid != NULL)
2172 					strlcpy(tpr->pr_hostuuid,
2173 					    pr->pr_hostuuid,
2174 					    sizeof(tpr->pr_hostuuid));
2175 				if (gothid)
2176 					tpr->pr_hostid = hid;
2177 			}
2178 		}
2179 	}
2180 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2181 	if ((tallow = ch_allow & ~pr_allow))
2182 		prison_set_allow_locked(pr, tallow, 0);
2183 	/*
2184 	 * Persistent prisons get an extra reference, and prisons losing their
2185 	 * persist flag lose that reference.
2186 	 */
2187 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2188 		if (pr_flags & PR_PERSIST) {
2189 			prison_hold(pr);
2190 			/*
2191 			 * This may be a new prison's first user reference,
2192 			 * but wait to call it alive until after OSD calls
2193 			 * have had a chance to run (and perhaps to fail).
2194 			 */
2195 			refcount_acquire(&pr->pr_uref);
2196 		} else {
2197 			drflags |= PD_DEUREF;
2198 			prison_free_not_last(pr);
2199 		}
2200 	}
2201 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2202 
2203 #ifdef MAC
2204 	/* Apply any request MAC label before we let modules do their work. */
2205 	if (gotmaclabel) {
2206 		error = mac_set_prison_core(td, pr, mac_set_prison_data);
2207 		if (error) {
2208 			vfs_opterror(opts, "mac relabel denied");
2209 			goto done_deref;
2210 		}
2211 	}
2212 #endif
2213 	mtx_unlock(&pr->pr_mtx);
2214 	drflags &= ~PD_LOCKED;
2215 	/*
2216 	 * Any errors past this point will need to de-persist newly created
2217 	 * prisons, as well as call remove methods.
2218 	 */
2219 	if (created)
2220 		drflags |= PD_KILL;
2221 
2222 #ifdef RACCT
2223 	if (racct_enable && created)
2224 		prison_racct_attach(pr);
2225 #endif
2226 
2227 	/* Locks may have prevented a complete restriction of child IP
2228 	 * addresses.  If so, allocate some more memory and try again.
2229 	 */
2230 #ifdef INET
2231 	while (redo_ip4) {
2232 		ip4s = pr->pr_addrs[PR_INET]->ips;
2233 		MPASS(ip4 == NULL);
2234 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2235 		mtx_lock(&pr->pr_mtx);
2236 		redo_ip4 = false;
2237 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2238 #ifdef VIMAGE
2239 			if (tpr->pr_flags & PR_VNET) {
2240 				descend = 0;
2241 				continue;
2242 			}
2243 #endif
2244 			if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2245 				redo_ip4 = true;
2246 		}
2247 		mtx_unlock(&pr->pr_mtx);
2248 	}
2249 #endif
2250 #ifdef INET6
2251 	while (redo_ip6) {
2252 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2253 		MPASS(ip6 == NULL);
2254 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2255 		mtx_lock(&pr->pr_mtx);
2256 		redo_ip6 = false;
2257 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2258 #ifdef VIMAGE
2259 			if (tpr->pr_flags & PR_VNET) {
2260 				descend = 0;
2261 				continue;
2262 			}
2263 #endif
2264 			if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2265 				redo_ip6 = true;
2266 		}
2267 		mtx_unlock(&pr->pr_mtx);
2268 	}
2269 #endif
2270 
2271 	/* Let the modules do their work. */
2272 	if (created) {
2273 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2274 		if (error)
2275 			goto done_deref;
2276 	}
2277 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2278 	if (error)
2279 		goto done_deref;
2280 
2281 	/*
2282 	 * A new prison is now ready to be seen; either it has gained a user
2283 	 * reference via persistence, or is about to gain one via attachment.
2284 	 */
2285 	if (created) {
2286 		sx_assert(&allprison_lock, SX_XLOCKED);
2287 		prison_knote(ppr, NOTE_JAIL_CHILD | pr->pr_id);
2288 #ifdef MAC
2289 		/*
2290 		 * Note that mac_prison_created() assumes that it's called in a
2291 		 * sleepable context.
2292 		 */
2293 		mac_prison_created(td->td_ucred, pr);
2294 #endif
2295 		mtx_lock(&pr->pr_mtx);
2296 		drflags |= PD_LOCKED;
2297 		pr->pr_state = PRISON_STATE_ALIVE;
2298 	}
2299 
2300 	/* Attach this process to the prison if requested. */
2301 	if (flags & JAIL_ATTACH) {
2302 #ifdef MAC
2303 		error = mac_prison_check_attach(td->td_ucred, pr);
2304 		if (error != 0) {
2305 			vfs_opterror(opts,
2306 			    "attach operation denied by MAC policy");
2307 			goto done_deref;
2308 		}
2309 #endif
2310 		error = do_jail_attach(td, pr,
2311 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2312 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2313 		if (error) {
2314 			vfs_opterror(opts, "attach failed");
2315 			goto done_deref;
2316 		}
2317 	}
2318 
2319 #ifdef RACCT
2320 	if (racct_enable && !created) {
2321 		if (drflags & PD_LOCKED) {
2322 			mtx_unlock(&pr->pr_mtx);
2323 			drflags &= ~PD_LOCKED;
2324 		}
2325 		if (drflags & PD_LIST_XLOCKED) {
2326 			sx_xunlock(&allprison_lock);
2327 			drflags &= ~PD_LIST_XLOCKED;
2328 		}
2329 		prison_racct_modify(pr);
2330 	}
2331 #endif
2332 
2333 	if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2334 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2335 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2336 		   " file system\n", pr->pr_id);
2337 
2338 	/*
2339 	 * Now that the prison is fully created without error, set the
2340 	 * jail descriptor if one was requested.  This is the only
2341 	 * parameter that is returned to the caller (except the error
2342 	 * message).
2343 	 */
2344 	if (jfd_out >= 0) {
2345 		if (!(drflags & PD_LOCKED)) {
2346 			mtx_lock(&pr->pr_mtx);
2347 			drflags |= PD_LOCKED;
2348 		}
2349 		jfd_pos = 2 * vfs_getopt_pos(opts, "desc") + 1;
2350 		if (optuio->uio_segflg == UIO_SYSSPACE)
2351 			*(int*)optuio->uio_iov[jfd_pos].iov_base = jfd_out;
2352 		else
2353 			(void)copyout(&jfd_out,
2354 			    optuio->uio_iov[jfd_pos].iov_base, sizeof(jfd_out));
2355 		jaildesc_set_prison(jfp_out, pr);
2356 	}
2357 
2358 	drflags &= ~PD_KILL;
2359 	td->td_retval[0] = pr->pr_id;
2360 
2361  done_deref:
2362 	/*
2363 	 * Report changes to kevent.  This can happen even if the
2364 	 * system call fails, as changes might have been made before
2365 	 * the failure.
2366 	 */
2367 	if (maybe_changed && !created)
2368 		prison_knote(pr, NOTE_JAIL_SET);
2369 	/* Release any temporary prison holds and/or locks. */
2370 	if (pr != NULL)
2371 		prison_deref(pr, drflags);
2372 	else if (drflags & PD_LIST_SLOCKED)
2373 		sx_sunlock(&allprison_lock);
2374 	else if (drflags & PD_LIST_XLOCKED)
2375 		sx_xunlock(&allprison_lock);
2376 	if (root != NULL)
2377 		vrele(root);
2378  done_errmsg:
2379 	if (error) {
2380 		/* Write the error message back to userspace. */
2381 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2382 		    &errmsg_len) == 0 && errmsg_len > 0) {
2383 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2384 			if (optuio->uio_segflg == UIO_SYSSPACE)
2385 				bcopy(errmsg,
2386 				    optuio->uio_iov[errmsg_pos].iov_base,
2387 				    errmsg_len);
2388 			else
2389 				(void)copyout(errmsg,
2390 				    optuio->uio_iov[errmsg_pos].iov_base,
2391 				    errmsg_len);
2392 		}
2393 	}
2394  done_free:
2395 	/* Clean up other resources. */
2396 #ifdef INET
2397 	prison_ip_free(ip4);
2398 #endif
2399 #ifdef INET6
2400 	prison_ip_free(ip6);
2401 #endif
2402 #ifdef MAC
2403 	if (mac_set_prison_data != NULL)
2404 		mac_set_prison_finish(td, error == 0, mac_set_prison_data);
2405 #endif
2406 	if (jfp_out != NULL)
2407 		fdrop(jfp_out, td);
2408 	if (error && jfd_out >= 0)
2409 		(void)kern_close(td, jfd_out);
2410 	if (g_path != NULL)
2411 		free(g_path, M_TEMP);
2412 	if (opts != NULL)
2413 		vfs_freeopts(opts);
2414 	prison_free(mypr);
2415 	return (error);
2416 }
2417 
2418 /*
2419  * Find the next available prison ID.  Return the ID on success, or zero
2420  * on failure.  Also set a pointer to the allprison list entry the prison
2421  * should be inserted before.
2422  */
2423 static int
get_next_prid(struct prison ** insprp)2424 get_next_prid(struct prison **insprp)
2425 {
2426 	struct prison *inspr;
2427 	int jid, maxid;
2428 
2429 	jid = lastprid % JAIL_MAX + 1;
2430 	if (TAILQ_EMPTY(&allprison) ||
2431 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2432 		/*
2433 		 * A common case is for all jails to be implicitly numbered,
2434 		 * which means they'll go on the end of the list, at least
2435 		 * for the first JAIL_MAX times.
2436 		 */
2437 		inspr = NULL;
2438 	} else {
2439 		/*
2440 		 * Take two passes through the allprison list: first starting
2441 		 * with the proposed jid, then ending with it.
2442 		 */
2443 		for (maxid = JAIL_MAX; maxid != 0; ) {
2444 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2445 				if (inspr->pr_id < jid)
2446 					continue;
2447 				if (inspr->pr_id > jid) {
2448 					/* Found an opening. */
2449 					maxid = 0;
2450 					break;
2451 				}
2452 				if (++jid > maxid) {
2453 					if (lastprid == maxid || lastprid == 0)
2454 					{
2455 						/*
2456 						 * The entire legal range
2457 						 * has been traversed
2458 						 */
2459 						return 0;
2460 					}
2461 					/* Try again from the start. */
2462 					jid = 1;
2463 					maxid = lastprid;
2464 					break;
2465 				}
2466 			}
2467 			if (inspr == NULL) {
2468 				/* Found room at the end of the list. */
2469 				break;
2470 			}
2471 		}
2472 	}
2473 	*insprp = inspr;
2474 	lastprid = jid;
2475 	return (jid);
2476 }
2477 
2478 /*
2479  * Find the next available ID for a renumbered dead prison.  This is the same
2480  * as get_next_prid, but counting backward from the end of the range.
2481  */
2482 static int
get_next_deadid(struct prison ** dinsprp)2483 get_next_deadid(struct prison **dinsprp)
2484 {
2485 	struct prison *dinspr;
2486 	int deadid, minid;
2487 
2488 	deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2489 	/*
2490 	 * Take two reverse passes through the allprison list: first
2491 	 * starting with the proposed deadid, then ending with it.
2492 	 */
2493 	for (minid = 1; minid != 0; ) {
2494 		TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2495 			if (dinspr->pr_id > deadid)
2496 				continue;
2497 			if (dinspr->pr_id < deadid) {
2498 				/* Found an opening. */
2499 				minid = 0;
2500 				break;
2501 			}
2502 			if (--deadid < minid) {
2503 				if (lastdeadid == minid || lastdeadid == 0)
2504 				{
2505 					/*
2506 					 * The entire legal range
2507 					 * has been traversed
2508 					 */
2509 					return 0;
2510 				}
2511 				/* Try again from the end. */
2512 				deadid = JAIL_MAX;
2513 				minid = lastdeadid;
2514 				break;
2515 			}
2516 		}
2517 		if (dinspr == NULL) {
2518 			/* Found room at the beginning of the list. */
2519 			break;
2520 		}
2521 	}
2522 	*dinsprp = dinspr;
2523 	lastdeadid = deadid;
2524 	return (deadid);
2525 }
2526 
2527 /*
2528  * struct jail_get_args {
2529  *	struct iovec *iovp;
2530  *	unsigned int iovcnt;
2531  *	int flags;
2532  * };
2533  */
2534 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)2535 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2536 {
2537 	struct uio *auio;
2538 	int error;
2539 
2540 	/* Check that we have an even number of iovecs. */
2541 	if (uap->iovcnt & 1)
2542 		return (EINVAL);
2543 
2544 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2545 	if (error)
2546 		return (error);
2547 	error = kern_jail_get(td, auio, uap->flags);
2548 	if (error == 0)
2549 		error = copyout(auio->uio_iov, uap->iovp,
2550 		    uap->iovcnt * sizeof(struct iovec));
2551 	freeuio(auio);
2552 	return (error);
2553 }
2554 
2555 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2556 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2557 {
2558 	struct bool_flags *bf;
2559 	struct file *jfp_out;
2560 	struct jailsys_flags *jsf;
2561 	struct prison *pr, *mypr;
2562 	struct vfsopt *opt;
2563 	struct vfsoptlist *opts;
2564 	char *errmsg, *name;
2565 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2566 	int jfd_in, jfd_out;
2567 	unsigned f;
2568 
2569 	if (flags & ~JAIL_GET_MASK)
2570 		return (EINVAL);
2571 	if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
2572 	    (JAIL_USE_DESC | JAIL_AT_DESC))
2573 		return (EINVAL);
2574 
2575 	/* Get the parameter list. */
2576 	error = vfs_buildopts(optuio, &opts);
2577 	if (error)
2578 		return (error);
2579 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2580 	mypr = td->td_ucred->cr_prison;
2581 	prison_hold(mypr);
2582 	pr = NULL;
2583 	jfp_out = NULL;
2584 	jfd_out = -1;
2585 
2586 	/*
2587 	 * Find the prison specified by one of: desc, lastjid, jid, name.
2588 	 */
2589 	sx_slock(&allprison_lock);
2590 	drflags = PD_LIST_SLOCKED;
2591 
2592 	error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
2593 	if (error == ENOENT) {
2594 		if (flags & (JAIL_AT_DESC | JAIL_GET_DESC | JAIL_OWN_DESC)) {
2595 			vfs_opterror(opts, "missing desc");
2596 			goto done;
2597 		}
2598 	} else if (error == 0) {
2599 		if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
2600 		    JAIL_OWN_DESC))) {
2601 			error = EINVAL;
2602 			vfs_opterror(opts, "unexpected desc");
2603 			goto done;
2604 		}
2605 		if (flags & JAIL_USE_DESC) {
2606 			/* Get the jail from its descriptor. */
2607 			error = jaildesc_find(td, jfd_in, &pr, NULL);
2608 			if (error) {
2609 				vfs_opterror(opts, error == ENOENT ?
2610 				    "descriptor to dead jail" :
2611 				    "not a jail descriptor");
2612 				goto done;
2613 			}
2614 			drflags |= PD_DEREF;
2615 			mtx_lock(&pr->pr_mtx);
2616 			drflags |= PD_LOCKED;
2617 			goto found_prison;
2618 		}
2619 		if (flags & JAIL_AT_DESC) {
2620 			/* Look up jails based on the descriptor's prison. */
2621 			prison_free(mypr);
2622 			error = jaildesc_find(td, jfd_in, &mypr, NULL);
2623 			if (error != 0) {
2624 				vfs_opterror(opts, error == ENOENT ?
2625 				    "descriptor to dead jail" :
2626 				    "not a jail descriptor");
2627 				goto done;
2628 			}
2629 		}
2630 		if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
2631 			/* Allocate a jail descriptor to return later. */
2632 			error = jaildesc_alloc(td, &jfp_out, &jfd_out,
2633 				flags & JAIL_OWN_DESC);
2634 			if (error)
2635 				goto done;
2636 		}
2637 	} else
2638 		goto done;
2639 
2640 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2641 	if (error == 0) {
2642 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2643 			if (pr->pr_id > jid &&
2644 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2645 			    prison_ischild(mypr, pr)) {
2646 				mtx_lock(&pr->pr_mtx);
2647 				drflags |= PD_LOCKED;
2648 #ifdef MAC
2649 				/*
2650 				 * We special-case this one check because we
2651 				 * don't want MAC to break jail enumeration.  We
2652 				 * need to just move on to the next accessible
2653 				 * and alive prison.
2654 				 */
2655 				error = mac_prison_check_get(td->td_ucred, pr,
2656 				    opts, flags);
2657 				if (error != 0) {
2658 					mtx_unlock(&pr->pr_mtx);
2659 					drflags &= ~PD_LOCKED;
2660 					continue;
2661 				}
2662 
2663 				/*
2664 				 * Avoid potentially expensive trip back into
2665 				 * the MAC framework.
2666 				 */
2667 				goto found_prison_nomac_alive;
2668 #else
2669 				goto found_prison;
2670 #endif
2671 			}
2672 		}
2673 		error = ENOENT;
2674 		vfs_opterror(opts, "no jail after %d", jid);
2675 		goto done;
2676 	} else if (error != ENOENT)
2677 		goto done;
2678 
2679 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2680 	if (error == 0) {
2681 		if (jid != 0) {
2682 			pr = prison_find_child(mypr, jid);
2683 			if (pr != NULL) {
2684 				drflags |= PD_LOCKED;
2685 				goto found_prison;
2686 			}
2687 			error = ENOENT;
2688 			vfs_opterror(opts, "jail %d not found", jid);
2689 			goto done;
2690 		}
2691 	} else if (error != ENOENT)
2692 		goto done;
2693 
2694 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2695 	if (error == 0) {
2696 		if (len == 0 || name[len - 1] != '\0') {
2697 			error = EINVAL;
2698 			goto done;
2699 		}
2700 		pr = prison_find_name(mypr, name);
2701 		if (pr != NULL) {
2702 			drflags |= PD_LOCKED;
2703 			goto found_prison;
2704 		}
2705 		error = ENOENT;
2706 		vfs_opterror(opts, "jail \"%s\" not found", name);
2707 		goto done;
2708 	} else if (error != ENOENT)
2709 		goto done;
2710 
2711 	vfs_opterror(opts, "no jail specified");
2712 	error = ENOENT;
2713 	goto done;
2714 
2715  found_prison:
2716 #ifdef MAC
2717 	error = mac_prison_check_get(td->td_ucred, pr, opts, flags);
2718 	if (error != 0)
2719 		goto done;
2720 #endif
2721 	if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2722 		error = ENOENT;
2723 		if (pr->pr_name[0] != '0' && isdigit(pr->pr_name[0])) {
2724 			vfs_opterror(opts, "jail %d is dying",
2725 			    pr->pr_id);
2726 		} else {
2727 			vfs_opterror(opts, "jail \"%s\" (%d) is dying",
2728 			    pr->pr_name, pr->pr_id);
2729 		}
2730 		goto done;
2731 	}
2732 #ifdef MAC
2733  found_prison_nomac_alive:
2734 #endif
2735 	/* Get the parameters of the prison. */
2736 	if (!(drflags & PD_DEREF)) {
2737 		prison_hold(pr);
2738 		drflags |= PD_DEREF;
2739 	}
2740 	td->td_retval[0] = pr->pr_id;
2741 	if (jfd_out >= 0) {
2742 		error = vfs_setopt(opts, "desc", &jfd_out, sizeof(jfd_out));
2743 		if (error != 0 && error != ENOENT)
2744 			goto done;
2745 		jaildesc_set_prison(jfp_out, pr);
2746 	}
2747 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2748 	if (error != 0 && error != ENOENT)
2749 		goto done;
2750 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2751 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2752 	if (error != 0 && error != ENOENT)
2753 		goto done;
2754 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2755 	if (error != 0 && error != ENOENT)
2756 		goto done;
2757 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2758 	    sizeof(pr->pr_cpuset->cs_id));
2759 	if (error != 0 && error != ENOENT)
2760 		goto done;
2761 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2762 	if (error != 0 && error != ENOENT)
2763 		goto done;
2764 #ifdef INET
2765 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2766 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2767 	    pr_families[PR_INET].size : 0 );
2768 	if (error != 0 && error != ENOENT)
2769 		goto done;
2770 #endif
2771 #ifdef INET6
2772 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2773 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2774 	    pr_families[PR_INET6].size : 0 );
2775 	if (error != 0 && error != ENOENT)
2776 		goto done;
2777 #endif
2778 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2779 	    sizeof(pr->pr_securelevel));
2780 	if (error != 0 && error != ENOENT)
2781 		goto done;
2782 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2783 	    sizeof(pr->pr_childcount));
2784 	if (error != 0 && error != ENOENT)
2785 		goto done;
2786 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2787 	    sizeof(pr->pr_childmax));
2788 	if (error != 0 && error != ENOENT)
2789 		goto done;
2790 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2791 	if (error != 0 && error != ENOENT)
2792 		goto done;
2793 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2794 	if (error != 0 && error != ENOENT)
2795 		goto done;
2796 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2797 	if (error != 0 && error != ENOENT)
2798 		goto done;
2799 #ifdef COMPAT_FREEBSD32
2800 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2801 		uint32_t hid32 = pr->pr_hostid;
2802 
2803 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2804 	} else
2805 #endif
2806 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2807 	    sizeof(pr->pr_hostid));
2808 	if (error != 0 && error != ENOENT)
2809 		goto done;
2810 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2811 	    sizeof(pr->pr_enforce_statfs));
2812 	if (error != 0 && error != ENOENT)
2813 		goto done;
2814 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2815 	    sizeof(pr->pr_devfs_rsnum));
2816 	if (error != 0 && error != ENOENT)
2817 		goto done;
2818 	for (bf = pr_flag_bool;
2819 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2820 	     bf++) {
2821 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2822 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2823 		if (error != 0 && error != ENOENT)
2824 			goto done;
2825 		i = !i;
2826 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2827 		if (error != 0 && error != ENOENT)
2828 			goto done;
2829 	}
2830 	for (jsf = pr_flag_jailsys;
2831 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2832 	     jsf++) {
2833 		f = pr->pr_flags & (jsf->disable | jsf->new);
2834 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2835 		    : (f == jsf->new) ? JAIL_SYS_NEW
2836 		    : JAIL_SYS_INHERIT;
2837 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2838 		if (error != 0 && error != ENOENT)
2839 			goto done;
2840 	}
2841 	for (bf = pr_flag_allow;
2842 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2843 		atomic_load_int(&bf->flag) != 0;
2844 	     bf++) {
2845 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2846 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2847 		if (error != 0 && error != ENOENT)
2848 			goto done;
2849 		i = !i;
2850 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2851 		if (error != 0 && error != ENOENT)
2852 			goto done;
2853 	}
2854 	i = !prison_isalive(pr);
2855 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2856 	if (error != 0 && error != ENOENT)
2857 		goto done;
2858 	i = !i;
2859 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2860 	if (error != 0 && error != ENOENT)
2861 		goto done;
2862 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2863 	    sizeof(pr->pr_osreldate));
2864 	if (error != 0 && error != ENOENT)
2865 		goto done;
2866 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2867 	if (error != 0 && error != ENOENT)
2868 		goto done;
2869 
2870 #ifdef MAC
2871 	/*
2872 	 * We get the MAC label last because we'll let the MAC framework drop
2873 	 * pr_mtx to externalize the label.
2874 	 */
2875 	error = mac_get_prison(td, pr, opts);
2876 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
2877 	drflags &= ~PD_LOCKED;
2878 	if (error != 0 && error != ENOENT)
2879 		goto done;
2880 #else
2881 	mtx_unlock(&pr->pr_mtx);
2882 	drflags &= ~PD_LOCKED;
2883 #endif
2884 
2885 	/* Get the module parameters. */
2886 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2887 	if (error)
2888 		goto done;
2889 	prison_deref(pr, drflags);
2890 	pr = NULL;
2891 	drflags = 0;
2892 
2893 	/* By now, all parameters should have been noted. */
2894 	TAILQ_FOREACH(opt, opts, link) {
2895 		if (!opt->seen &&
2896 		    (strstr(opt->name, JAIL_META_PRIVATE ".") == opt->name ||
2897 		    strstr(opt->name, JAIL_META_SHARED ".") == opt->name)) {
2898 			/* Communicate back a missing key. */
2899 			free(opt->value, M_MOUNT);
2900 			opt->value = NULL;
2901 			opt->len = 0;
2902 			continue;
2903 		}
2904 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2905 			error = EINVAL;
2906 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2907 			goto done;
2908 		}
2909 	}
2910 
2911 	/* Write the fetched parameters back to userspace. */
2912 	error = 0;
2913 	TAILQ_FOREACH(opt, opts, link) {
2914 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2915 			pos = 2 * opt->pos + 1;
2916 			optuio->uio_iov[pos].iov_len = opt->len;
2917 			if (opt->value != NULL) {
2918 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2919 					bcopy(opt->value,
2920 					    optuio->uio_iov[pos].iov_base,
2921 					    opt->len);
2922 				} else {
2923 					error = copyout(opt->value,
2924 					    optuio->uio_iov[pos].iov_base,
2925 					    opt->len);
2926 					if (error)
2927 						break;
2928 				}
2929 			}
2930 		}
2931 	}
2932 
2933  done:
2934 	/* Release any temporary prison holds and/or locks. */
2935 	if (pr != NULL)
2936 		prison_deref(pr, drflags);
2937 	else if (drflags & PD_LIST_SLOCKED)
2938 		sx_sunlock(&allprison_lock);
2939 	else if (drflags & PD_LIST_XLOCKED)
2940 		sx_xunlock(&allprison_lock);
2941 	/* Clean up other resources. */
2942 	if (jfp_out != NULL)
2943 		(void)fdrop(jfp_out, td);
2944 	if (error && jfd_out >= 0)
2945 		(void)kern_close(td, jfd_out);
2946 	if (error && errmsg_pos >= 0) {
2947 		/* Write the error message back to userspace. */
2948 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2949 		errmsg_pos = 2 * errmsg_pos + 1;
2950 		if (errmsg_len > 0) {
2951 			if (optuio->uio_segflg == UIO_SYSSPACE)
2952 				bcopy(errmsg,
2953 				    optuio->uio_iov[errmsg_pos].iov_base,
2954 				    errmsg_len);
2955 			else
2956 				(void)copyout(errmsg,
2957 				    optuio->uio_iov[errmsg_pos].iov_base,
2958 				    errmsg_len);
2959 		}
2960 	}
2961 	vfs_freeopts(opts);
2962 	prison_free(mypr);
2963 	return (error);
2964 }
2965 
2966 /*
2967  * struct jail_remove_args {
2968  *	int jid;
2969  * };
2970  */
2971 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2972 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2973 {
2974 	struct prison *pr;
2975 	int error;
2976 
2977 	error = priv_check(td, PRIV_JAIL_REMOVE);
2978 	if (error)
2979 		return (error);
2980 
2981 	sx_xlock(&allprison_lock);
2982 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2983 	if (pr == NULL) {
2984 		sx_xunlock(&allprison_lock);
2985 		return (EINVAL);
2986 	}
2987 #ifdef MAC
2988 	error = mac_prison_check_remove(td->td_ucred, pr);
2989 	if (error != 0) {
2990 		mtx_unlock(&pr->pr_mtx);
2991 		sx_xunlock(&allprison_lock);
2992 		return (error);
2993 	}
2994 #endif
2995 	prison_hold(pr);
2996 	prison_remove(pr);
2997 	return (0);
2998 }
2999 
3000 /*
3001  * struct jail_remove_jd_args {
3002  *	int fd;
3003  * };
3004  */
3005 int
sys_jail_remove_jd(struct thread * td,struct jail_remove_jd_args * uap)3006 sys_jail_remove_jd(struct thread *td, struct jail_remove_jd_args *uap)
3007 {
3008 	struct prison *pr;
3009 	struct ucred *jdcred;
3010 	int error;
3011 
3012 	error = jaildesc_find(td, uap->fd, &pr, &jdcred);
3013 	if (error)
3014 		return (error);
3015 	error = priv_check_cred(jdcred, PRIV_JAIL_REMOVE);
3016 	crfree(jdcred);
3017 #ifdef MAC
3018 	if (error == 0)
3019 		error = mac_prison_check_remove(td->td_ucred, pr);
3020 #endif
3021 	if (error) {
3022 		prison_free(pr);
3023 		return (error);
3024 	}
3025 	sx_xlock(&allprison_lock);
3026 	mtx_lock(&pr->pr_mtx);
3027 	prison_remove(pr);
3028 	return (0);
3029 }
3030 
3031 /*
3032  * Begin the removal process for a prison.  The allprison lock should
3033  * be held exclusively, and the prison should be both locked and held.
3034  */
3035 void
prison_remove(struct prison * pr)3036 prison_remove(struct prison *pr)
3037 {
3038 	sx_assert(&allprison_lock, SA_XLOCKED);
3039 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3040 	prison_deref(pr, PD_KILL | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
3041 }
3042 
3043 /*
3044  * struct jail_attach_args {
3045  *	int jid;
3046  * };
3047  */
3048 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)3049 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
3050 {
3051 	struct prison *pr;
3052 	int error;
3053 
3054 	error = priv_check(td, PRIV_JAIL_ATTACH);
3055 	if (error)
3056 		return (error);
3057 
3058 	sx_slock(&allprison_lock);
3059 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
3060 	if (pr == NULL) {
3061 		sx_sunlock(&allprison_lock);
3062 		return (EINVAL);
3063 	}
3064 
3065 #ifdef MAC
3066 	error = mac_prison_check_attach(td->td_ucred, pr);
3067 	if (error != 0)
3068 		goto unlock;
3069 #endif
3070 
3071 	/* Do not allow a process to attach to a prison that is not alive. */
3072 	if (!prison_isalive(pr)) {
3073 		error = EINVAL;
3074 		goto unlock;
3075 	}
3076 
3077 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
3078 
3079 unlock:
3080 
3081 	mtx_unlock(&pr->pr_mtx);
3082 	sx_sunlock(&allprison_lock);
3083 	return (error);
3084 }
3085 
3086 /*
3087  * struct jail_attach_jd_args {
3088  *	int fd;
3089  * };
3090  */
3091 int
sys_jail_attach_jd(struct thread * td,struct jail_attach_jd_args * uap)3092 sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
3093 {
3094 	struct prison *pr;
3095 	struct ucred *jdcred;
3096 	int drflags, error;
3097 
3098 	sx_slock(&allprison_lock);
3099 	drflags = PD_LIST_SLOCKED;
3100 	error = jaildesc_find(td, uap->fd, &pr, &jdcred);
3101 	if (error)
3102 		goto fail;
3103 	drflags |= PD_DEREF;
3104 	error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
3105 #ifdef MAC
3106 	if (error == 0)
3107 		error = mac_prison_check_attach(td->td_ucred, pr);
3108 #endif
3109 	crfree(jdcred);
3110 	if (error)
3111 		goto fail;
3112 	mtx_lock(&pr->pr_mtx);
3113 	drflags |= PD_LOCKED;
3114 
3115 	/* Do not allow a process to attach to a prison that is not alive. */
3116 	if (!prison_isalive(pr)) {
3117 		error = EINVAL;
3118 		goto fail;
3119 	}
3120 
3121 	return (do_jail_attach(td, pr, drflags));
3122 
3123  fail:
3124 	prison_deref(pr, drflags);
3125 	return (error);
3126 }
3127 
3128 static int
do_jail_attach(struct thread * td,struct prison * pr,int drflags)3129 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
3130 {
3131 	struct proc *p;
3132 	struct ucred *newcred, *oldcred;
3133 	int error;
3134 
3135 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3136 	sx_assert(&allprison_lock, SX_LOCKED);
3137 	drflags &= PD_LOCK_FLAGS;
3138 	/*
3139 	 * XXX: Note that there is a slight race here if two threads
3140 	 * in the same privileged process attempt to attach to two
3141 	 * different jails at the same time.  It is important for
3142 	 * user processes not to do this, or they might end up with
3143 	 * a process root from one prison, but attached to the jail
3144 	 * of another.
3145 	 */
3146 	if (!(drflags & PD_DEREF)) {
3147 		prison_hold(pr);
3148 		drflags |= PD_DEREF;
3149 	}
3150 	refcount_acquire(&pr->pr_uref);
3151 	drflags |= PD_DEUREF;
3152 	mtx_unlock(&pr->pr_mtx);
3153 	drflags &= ~PD_LOCKED;
3154 
3155 	/* Let modules do whatever they need to prepare for attaching. */
3156 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
3157 	if (error) {
3158 		prison_deref(pr, drflags);
3159 		return (error);
3160 	}
3161 	sx_unlock(&allprison_lock);
3162 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
3163 
3164 	/*
3165 	 * Reparent the newly attached process to this jail.
3166 	 */
3167 	p = td->td_proc;
3168 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
3169 	if (error)
3170 		goto e_revert_osd;
3171 
3172 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
3173 	if ((error = change_dir(pr->pr_root, td)) != 0)
3174 		goto e_unlock;
3175 #ifdef MAC
3176 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
3177 		goto e_unlock;
3178 #endif
3179 	VOP_UNLOCK(pr->pr_root);
3180 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
3181 		goto e_revert_osd;
3182 
3183 	newcred = crget();
3184 	PROC_LOCK(p);
3185 	oldcred = crcopysafe(p, newcred);
3186 	newcred->cr_prison = pr;
3187 #ifdef RACCT
3188 	racct_proc_ucred_changed(p, oldcred, newcred);
3189 #endif
3190 #ifdef RCTL
3191 	crhold(newcred);
3192 #endif
3193 	/*
3194 	 * Takes over 'newcred''s reference, so 'newcred' must not be used
3195 	 * besides this point except on RCTL where we took an additional
3196 	 * reference above.
3197 	 */
3198 	proc_set_cred(p, newcred);
3199 	setsugid(p);
3200 	PROC_UNLOCK(p);
3201 #ifdef RCTL
3202 	rctl_proc_ucred_changed(p, newcred);
3203 	crfree(newcred);
3204 #endif
3205 	prison_proc_relink(oldcred->cr_prison, pr, p);
3206 	prison_deref(oldcred->cr_prison, drflags);
3207 	crfree(oldcred);
3208 	prison_knote(pr, NOTE_JAIL_ATTACH | td->td_proc->p_pid);
3209 #ifdef MAC
3210 	/*
3211 	 * Note that mac_prison_attached() assumes that it's called in a
3212 	 * sleepable context.
3213 	 */
3214 	mac_prison_attached(td->td_ucred, pr, td->td_proc);
3215 #endif
3216 
3217 	/*
3218 	 * If the prison was killed while changing credentials, die along
3219 	 * with it.
3220 	 */
3221 	if (!prison_isalive(pr)) {
3222 		PROC_LOCK(p);
3223 		kern_psignal(p, SIGKILL);
3224 		PROC_UNLOCK(p);
3225 	}
3226 
3227 	return (0);
3228 
3229  e_unlock:
3230 	VOP_UNLOCK(pr->pr_root);
3231  e_revert_osd:
3232 	/* Tell modules this thread is still in its old jail after all. */
3233 	sx_slock(&allprison_lock);
3234 	drflags |= PD_LIST_SLOCKED;
3235 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
3236 	prison_deref(pr, drflags);
3237 	return (error);
3238 }
3239 
3240 /*
3241  * Returns a locked prison instance, or NULL on failure.
3242  */
3243 struct prison *
prison_find(int prid)3244 prison_find(int prid)
3245 {
3246 	struct prison *pr;
3247 
3248 	sx_assert(&allprison_lock, SX_LOCKED);
3249 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3250 		if (pr->pr_id < prid)
3251 			continue;
3252 		if (pr->pr_id > prid)
3253 			break;
3254 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
3255 		mtx_lock(&pr->pr_mtx);
3256 		return (pr);
3257 	}
3258 	return (NULL);
3259 }
3260 
3261 /*
3262  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
3263  */
3264 struct prison *
prison_find_child(struct prison * mypr,int prid)3265 prison_find_child(struct prison *mypr, int prid)
3266 {
3267 	struct prison *pr;
3268 	int descend;
3269 
3270 	sx_assert(&allprison_lock, SX_LOCKED);
3271 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3272 		if (pr->pr_id == prid) {
3273 			KASSERT(prison_isvalid(pr),
3274 			    ("Found invalid prison %p", pr));
3275 			mtx_lock(&pr->pr_mtx);
3276 			return (pr);
3277 		}
3278 	}
3279 	return (NULL);
3280 }
3281 
3282 /*
3283  * Look for the name relative to mypr.  Returns a locked prison or NULL.
3284  */
3285 struct prison *
prison_find_name(struct prison * mypr,const char * name)3286 prison_find_name(struct prison *mypr, const char *name)
3287 {
3288 	struct prison *pr, *deadpr;
3289 	size_t mylen;
3290 	int descend;
3291 
3292 	sx_assert(&allprison_lock, SX_LOCKED);
3293 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
3294 	deadpr = NULL;
3295 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3296 		if (!strcmp(pr->pr_name + mylen, name)) {
3297 			KASSERT(prison_isvalid(pr),
3298 			    ("Found invalid prison %p", pr));
3299 			if (prison_isalive(pr)) {
3300 				mtx_lock(&pr->pr_mtx);
3301 				return (pr);
3302 			}
3303 			deadpr = pr;
3304 		}
3305 	}
3306 	/* There was no valid prison - perhaps there was a dying one. */
3307 	if (deadpr != NULL)
3308 		mtx_lock(&deadpr->pr_mtx);
3309 	return (deadpr);
3310 }
3311 
3312 /*
3313  * See if a prison has the specific flag set.  The prison should be locked,
3314  * unless checking for flags that are only set at jail creation (such as
3315  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
3316  * to any other prison data.
3317  */
3318 bool
prison_flag(struct ucred * cred,unsigned flag)3319 prison_flag(struct ucred *cred, unsigned flag)
3320 {
3321 
3322 	return ((cred->cr_prison->pr_flags & flag) != 0);
3323 }
3324 
3325 /*
3326  * See if a prison has the specific allow flag set.
3327  * The prison *should* be locked, or only a single bit is examined, without
3328  * regard to any other prison data.
3329  */
3330 bool
prison_allow(struct ucred * cred,unsigned flag)3331 prison_allow(struct ucred *cred, unsigned flag)
3332 {
3333 
3334 	return ((cred->cr_prison->pr_allow & flag) != 0);
3335 }
3336 
3337 /*
3338  * Hold a prison reference, by incrementing pr_ref.  It is generally
3339  * an error to hold a prison that does not already have a reference.
3340  * A prison record will remain valid as long as it has at least one
3341  * reference, and will not be removed as long as either the prison
3342  * mutex or the allprison lock is held (allprison_lock may be shared).
3343  */
3344 void
prison_hold_locked(struct prison * pr)3345 prison_hold_locked(struct prison *pr)
3346 {
3347 
3348 	/* Locking is no longer required. */
3349 	prison_hold(pr);
3350 }
3351 
3352 void
prison_hold(struct prison * pr)3353 prison_hold(struct prison *pr)
3354 {
3355 #ifdef INVARIANTS
3356 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
3357 
3358 	KASSERT(was_valid,
3359 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
3360 #else
3361 	refcount_acquire(&pr->pr_ref);
3362 #endif
3363 }
3364 
3365 /*
3366  * Remove a prison reference.  If that was the last reference, the
3367  * prison will be removed (at a later time).
3368  */
3369 void
prison_free_locked(struct prison * pr)3370 prison_free_locked(struct prison *pr)
3371 {
3372 
3373 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3374 	/*
3375 	 * Locking is no longer required, but unlock because the caller
3376 	 * expects it.
3377 	 */
3378 	mtx_unlock(&pr->pr_mtx);
3379 	prison_free(pr);
3380 }
3381 
3382 void
prison_free(struct prison * pr)3383 prison_free(struct prison *pr)
3384 {
3385 
3386 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3387 	    ("Trying to free dead prison %p (jid=%d).",
3388 	     pr, pr->pr_id));
3389 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
3390 		/*
3391 		 * Don't remove the last reference in this context,
3392 		 * in case there are locks held.
3393 		 */
3394 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3395 	}
3396 }
3397 
3398 static void
prison_free_not_last(struct prison * pr)3399 prison_free_not_last(struct prison *pr)
3400 {
3401 #ifdef INVARIANTS
3402 	int lastref;
3403 
3404 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3405 	    ("Trying to free dead prison %p (jid=%d).",
3406 	     pr, pr->pr_id));
3407 	lastref = refcount_release(&pr->pr_ref);
3408 	KASSERT(!lastref,
3409 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
3410 	     pr, pr->pr_id));
3411 #else
3412 	refcount_release(&pr->pr_ref);
3413 #endif
3414 }
3415 
3416 /*
3417  * Hold a prison for user visibility, by incrementing pr_uref.
3418  * It is generally an error to hold a prison that isn't already
3419  * user-visible, except through the jail system calls.  It is also
3420  * an error to hold an invalid prison.  A prison record will remain
3421  * alive as long as it has at least one user reference, and will not
3422  * be set to the dying state until the prison mutex and allprison_lock
3423  * are both freed.
3424  */
3425 void
prison_proc_hold(struct prison * pr)3426 prison_proc_hold(struct prison *pr)
3427 {
3428 #ifdef INVARIANTS
3429 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
3430 
3431 	KASSERT(was_alive,
3432 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
3433 #else
3434 	refcount_acquire(&pr->pr_uref);
3435 #endif
3436 }
3437 
3438 /*
3439  * Remove a prison user reference.  If it was the last reference, the
3440  * prison will be considered "dying", and may be removed once all of
3441  * its references are dropped.
3442  */
3443 void
prison_proc_free(struct prison * pr)3444 prison_proc_free(struct prison *pr)
3445 {
3446 
3447 	/*
3448 	 * Locking is only required when releasing the last reference.
3449 	 * This allows assurance that a locked prison will remain alive
3450 	 * until it is unlocked.
3451 	 */
3452 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3453 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
3454 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
3455 		/*
3456 		 * Don't remove the last user reference in this context,
3457 		 * which is expected to be a process that is not only locked,
3458 		 * but also half dead.  Add a reference so any calls to
3459 		 * prison_free() won't re-submit the task.
3460 		 */
3461 		prison_hold(pr);
3462 		mtx_lock(&pr->pr_mtx);
3463 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
3464 		    ("Redundant last reference in prison_proc_free (jid=%d)",
3465 		     pr->pr_id));
3466 		pr->pr_flags |= PR_COMPLETE_PROC;
3467 		mtx_unlock(&pr->pr_mtx);
3468 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3469 	}
3470 }
3471 
3472 static void
prison_proc_free_not_last(struct prison * pr)3473 prison_proc_free_not_last(struct prison *pr)
3474 {
3475 #ifdef INVARIANTS
3476 	int lastref;
3477 
3478 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3479 	    ("Trying to free dead prison %p (jid=%d).",
3480 	     pr, pr->pr_id));
3481 	lastref = refcount_release(&pr->pr_uref);
3482 	KASSERT(!lastref,
3483 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3484 	     pr, pr->pr_id));
3485 #else
3486 	refcount_release(&pr->pr_uref);
3487 #endif
3488 }
3489 
3490 void
prison_proc_link(struct prison * pr,struct proc * p)3491 prison_proc_link(struct prison *pr, struct proc *p)
3492 {
3493 
3494 	sx_assert(&allproc_lock, SA_XLOCKED);
3495 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3496 }
3497 
3498 void
prison_proc_unlink(struct prison * pr,struct proc * p)3499 prison_proc_unlink(struct prison *pr, struct proc *p)
3500 {
3501 
3502 	sx_assert(&allproc_lock, SA_XLOCKED);
3503 	LIST_REMOVE(p, p_jaillist);
3504 }
3505 
3506 static void
prison_proc_relink(struct prison * opr,struct prison * npr,struct proc * p)3507 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3508 {
3509 
3510 	sx_xlock(&allproc_lock);
3511 	prison_proc_unlink(opr, p);
3512 	prison_proc_link(npr, p);
3513 	sx_xunlock(&allproc_lock);
3514 }
3515 
3516 /*
3517  * Complete a call to either prison_free or prison_proc_free.
3518  */
3519 static void
prison_complete(void * context,int pending)3520 prison_complete(void *context, int pending)
3521 {
3522 	struct prison *pr = context;
3523 	int drflags;
3524 
3525 	/*
3526 	 * This could be called to release the last reference, or the last
3527 	 * user reference (plus the reference held in prison_proc_free).
3528 	 */
3529 	drflags = prison_lock_xlock(pr, PD_DEREF);
3530 	if (pr->pr_flags & PR_COMPLETE_PROC) {
3531 		pr->pr_flags &= ~PR_COMPLETE_PROC;
3532 		drflags |= PD_DEUREF;
3533 	}
3534 	prison_deref(pr, drflags);
3535 }
3536 
3537 static void
prison_kill_processes_cb(struct proc * p,void * arg __unused)3538 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3539 {
3540 
3541 	kern_psignal(p, SIGKILL);
3542 }
3543 
3544 /*
3545  * Note the iteration does not guarantee acting on all processes.
3546  * Most notably there may be fork or jail_attach in progress.
3547  */
3548 void
prison_proc_iterate(struct prison * pr,void (* cb)(struct proc *,void *),void * cbarg)3549 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3550     void *cbarg)
3551 {
3552 	struct prison *ppr;
3553 	struct proc *p;
3554 
3555 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3556 		sx_slock(&allproc_lock);
3557 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3558 			if (p->p_state == PRS_NEW)
3559 				continue;
3560 			PROC_LOCK(p);
3561 			cb(p, cbarg);
3562 			PROC_UNLOCK(p);
3563 		}
3564 		sx_sunlock(&allproc_lock);
3565 		if (atomic_load_int(&pr->pr_childcount) == 0)
3566 			return;
3567 		/*
3568 		 * Some jails popped up during the iteration, fall through to a
3569 		 * system-wide search.
3570 		 */
3571 	}
3572 
3573 	sx_slock(&allproc_lock);
3574 	FOREACH_PROC_IN_SYSTEM(p) {
3575 		PROC_LOCK(p);
3576 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3577 			for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3578 			    ppr = ppr->pr_parent) {
3579 				if (ppr == pr) {
3580 					cb(p, cbarg);
3581 					break;
3582 				}
3583 			}
3584 		}
3585 		PROC_UNLOCK(p);
3586 	}
3587 	sx_sunlock(&allproc_lock);
3588 }
3589 
3590 /*
3591  * Remove a prison reference and/or user reference (usually).
3592  * This assumes context that allows sleeping (for allprison_lock),
3593  * with no non-sleeping locks held, except perhaps the prison itself.
3594  * If there are no more references, release and delist the prison.
3595  * On completion, the prison lock and the allprison lock are both
3596  * unlocked.
3597  */
3598 static void
prison_deref(struct prison * pr,int flags)3599 prison_deref(struct prison *pr, int flags)
3600 {
3601 	struct prisonlist freeprison;
3602 	struct prison *killpr, *rpr, *ppr, *tpr;
3603 
3604 	killpr = NULL;
3605 	TAILQ_INIT(&freeprison);
3606 	/*
3607 	 * Release this prison as requested, which may cause its parent
3608 	 * to be released, and then maybe its grandparent, etc.
3609 	 */
3610 	for (;;) {
3611 		if (flags & PD_KILL) {
3612 			/* Kill the prison and its descendents. */
3613 			KASSERT(pr != &prison0,
3614 			    ("prison_deref trying to kill prison0"));
3615 			if (!prison_isalive(pr)) {
3616 				/* Silently ignore already-dying prisons. */
3617 				flags &= ~PD_KILL;
3618 			} else {
3619 				if (!(flags & PD_DEREF)) {
3620 					prison_hold(pr);
3621 					flags |= PD_DEREF;
3622 				}
3623 				flags = prison_lock_xlock(pr, flags);
3624 				prison_deref_kill(pr, &freeprison);
3625 			}
3626 		}
3627 		if (flags & PD_DEUREF) {
3628 			/* Drop a user reference. */
3629 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3630 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3631 			     pr->pr_id));
3632 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3633 				if (!(flags & PD_DEREF)) {
3634 					prison_hold(pr);
3635 					flags |= PD_DEREF;
3636 				}
3637 				flags = prison_lock_xlock(pr, flags);
3638 				if (refcount_release(&pr->pr_uref) &&
3639 				    pr->pr_state == PRISON_STATE_ALIVE) {
3640 					/*
3641 					 * When the last user references goes,
3642 					 * this becomes a dying prison.
3643 					 */
3644 					KASSERT(
3645 					    refcount_load(&prison0.pr_uref) > 0,
3646 					    ("prison0 pr_uref=0"));
3647 					pr->pr_state = PRISON_STATE_DYING;
3648 					prison_cleanup_locked(pr);
3649 					mtx_unlock(&pr->pr_mtx);
3650 					flags &= ~PD_LOCKED;
3651 					prison_cleanup_unlocked(pr);
3652 				}
3653 			}
3654 		}
3655 		if (flags & PD_KILL) {
3656 			/*
3657 			 * Any remaining user references are probably processes
3658 			 * that need to be killed, either in this prison or its
3659 			 * descendants.
3660 			 */
3661 			if (refcount_load(&pr->pr_uref) > 0)
3662 				killpr = pr;
3663 			/* Make sure the parent prison doesn't get killed. */
3664 			flags &= ~PD_KILL;
3665 		}
3666 		if (flags & PD_DEREF) {
3667 			/* Drop a reference. */
3668 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3669 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3670 			     pr->pr_id));
3671 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3672 				flags = prison_lock_xlock(pr, flags);
3673 				if (refcount_release(&pr->pr_ref)) {
3674 					/*
3675 					 * When the last reference goes,
3676 					 * unlink the prison and set it aside.
3677 					 */
3678 					KASSERT(
3679 					    refcount_load(&pr->pr_uref) == 0,
3680 					    ("prison_deref: last ref, "
3681 					     "but still has %d urefs (jid=%d)",
3682 					     pr->pr_uref, pr->pr_id));
3683 					KASSERT(
3684 					    refcount_load(&prison0.pr_ref) != 0,
3685 					    ("prison0 pr_ref=0"));
3686 #ifdef MAC
3687 					/*
3688 					 * The MAC framework will call into any
3689 					 * policies that want to hook
3690 					 * prison_destroy_label, so ideally we
3691 					 * call this prior to any final state
3692 					 * invalidation to be safe.
3693 					 */
3694 					mac_prison_destroy(pr);
3695 #endif
3696 					pr->pr_state = PRISON_STATE_INVALID;
3697 					TAILQ_REMOVE(&allprison, pr, pr_list);
3698 					LIST_REMOVE(pr, pr_sibling);
3699 					TAILQ_INSERT_TAIL(&freeprison, pr,
3700 					    pr_list);
3701 					for (ppr = pr->pr_parent;
3702 					     ppr != NULL;
3703 					     ppr = ppr->pr_parent)
3704 						ppr->pr_childcount--;
3705 					/*
3706 					 * Removing a prison frees references
3707 					 * from its parent.
3708 					 */
3709 					ppr = pr->pr_parent;
3710 					pr->pr_parent = NULL;
3711 					mtx_unlock(&pr->pr_mtx);
3712 
3713 					pr = ppr;
3714 					flags &= ~PD_LOCKED;
3715 					flags |= PD_DEREF | PD_DEUREF;
3716 					continue;
3717 				}
3718 			}
3719 		}
3720 		break;
3721 	}
3722 
3723 	/* Release all the prison locks. */
3724 	if (flags & PD_LOCKED)
3725 		mtx_unlock(&pr->pr_mtx);
3726 	if (flags & PD_LIST_SLOCKED)
3727 		sx_sunlock(&allprison_lock);
3728 	else if (flags & PD_LIST_XLOCKED)
3729 		sx_xunlock(&allprison_lock);
3730 
3731 	/* Kill any processes attached to a killed prison. */
3732 	if (killpr != NULL)
3733 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3734 
3735 	/*
3736 	 * Finish removing any unreferenced prisons, which couldn't happen
3737 	 * while allprison_lock was held (to avoid a LOR on vrele).
3738 	 */
3739 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3740 #ifdef VIMAGE
3741 		if (rpr->pr_flags & PR_VNET)
3742 			vnet_destroy(rpr->pr_vnet);
3743 #endif
3744 		if (rpr->pr_root != NULL)
3745 			vrele(rpr->pr_root);
3746 		mtx_destroy(&rpr->pr_mtx);
3747 #ifdef INET
3748 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3749 #endif
3750 #ifdef INET6
3751 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3752 #endif
3753 		if (rpr->pr_cpuset != NULL)
3754 			cpuset_rel(rpr->pr_cpuset);
3755 		osd_jail_exit(rpr);
3756 #ifdef RACCT
3757 		if (racct_enable)
3758 			prison_racct_detach(rpr);
3759 #endif
3760 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3761 		free(rpr, M_PRISON);
3762 	}
3763 }
3764 
3765 /*
3766  * Kill the prison and its descendants.  Mark them as dying, clear the
3767  * persist flag, and call module remove methods.
3768  */
3769 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)3770 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3771 {
3772 	struct prison *cpr, *ppr, *rpr;
3773 	bool descend;
3774 
3775 	/*
3776 	 * Unlike the descendants, the target prison can be killed
3777 	 * even if it is currently dying.  This is useful for failed
3778 	 * creation in jail_set(2).
3779 	 */
3780 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3781 	    ("Trying to kill dead prison %p (jid=%d).",
3782 	     pr, pr->pr_id));
3783 	refcount_acquire(&pr->pr_uref);
3784 	pr->pr_state = PRISON_STATE_DYING;
3785 	mtx_unlock(&pr->pr_mtx);
3786 
3787 	rpr = NULL;
3788 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3789 		if (descend) {
3790 			if (!prison_isalive(cpr)) {
3791 				descend = false;
3792 				continue;
3793 			}
3794 			prison_hold(cpr);
3795 			prison_proc_hold(cpr);
3796 			mtx_lock(&cpr->pr_mtx);
3797 			cpr->pr_state = PRISON_STATE_DYING;
3798 			cpr->pr_flags |= PR_REMOVE;
3799 			mtx_unlock(&cpr->pr_mtx);
3800 			continue;
3801 		}
3802 		if (!(cpr->pr_flags & PR_REMOVE))
3803 			continue;
3804 		prison_cleanup_unlocked(cpr);
3805 		mtx_lock(&cpr->pr_mtx);
3806 		prison_cleanup_locked(cpr);
3807 		cpr->pr_flags &= ~PR_REMOVE;
3808 		if (cpr->pr_flags & PR_PERSIST) {
3809 			cpr->pr_flags &= ~PR_PERSIST;
3810 			prison_proc_free_not_last(cpr);
3811 			prison_free_not_last(cpr);
3812 		}
3813 		(void)refcount_release(&cpr->pr_uref);
3814 		if (refcount_release(&cpr->pr_ref)) {
3815 			/*
3816 			 * When the last reference goes, unlink the prison
3817 			 * and set it aside for prison_deref() to handle.
3818 			 * Delay unlinking the sibling list to keep the loop
3819 			 * safe.
3820 			 */
3821 			if (rpr != NULL)
3822 				LIST_REMOVE(rpr, pr_sibling);
3823 			rpr = cpr;
3824 			rpr->pr_state = PRISON_STATE_INVALID;
3825 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3826 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3827 			/*
3828 			 * Removing a prison frees references from its parent.
3829 			 */
3830 			ppr = rpr->pr_parent;
3831 			prison_proc_free_not_last(ppr);
3832 			prison_free_not_last(ppr);
3833 			for (; ppr != NULL; ppr = ppr->pr_parent)
3834 				ppr->pr_childcount--;
3835 		}
3836 		mtx_unlock(&cpr->pr_mtx);
3837 	}
3838 	if (rpr != NULL)
3839 		LIST_REMOVE(rpr, pr_sibling);
3840 
3841 	prison_cleanup_unlocked(pr);
3842 	mtx_lock(&pr->pr_mtx);
3843 	prison_cleanup_locked(pr);
3844 	if (pr->pr_flags & PR_PERSIST) {
3845 		pr->pr_flags &= ~PR_PERSIST;
3846 		prison_proc_free_not_last(pr);
3847 		prison_free_not_last(pr);
3848 	}
3849 	(void)refcount_release(&pr->pr_uref);
3850 }
3851 
3852 /*
3853  * Given the current locking state in the flags, make sure allprison_lock
3854  * is held exclusive, and the prison is locked.  Return flags indicating
3855  * the new state.
3856  */
3857 static int
prison_lock_xlock(struct prison * pr,int flags)3858 prison_lock_xlock(struct prison *pr, int flags)
3859 {
3860 
3861 	if (!(flags & PD_LIST_XLOCKED)) {
3862 		/*
3863 		 * Get allprison_lock, which may be an upgrade,
3864 		 * and may require unlocking the prison.
3865 		 */
3866 		if (flags & PD_LOCKED) {
3867 			mtx_unlock(&pr->pr_mtx);
3868 			flags &= ~PD_LOCKED;
3869 		}
3870 		if (flags & PD_LIST_SLOCKED) {
3871 			if (!sx_try_upgrade(&allprison_lock)) {
3872 				sx_sunlock(&allprison_lock);
3873 				sx_xlock(&allprison_lock);
3874 			}
3875 			flags &= ~PD_LIST_SLOCKED;
3876 		} else
3877 			sx_xlock(&allprison_lock);
3878 		flags |= PD_LIST_XLOCKED;
3879 	}
3880 	if (!(flags & PD_LOCKED)) {
3881 		/* Lock the prison mutex. */
3882 		mtx_lock(&pr->pr_mtx);
3883 		flags |= PD_LOCKED;
3884 	}
3885 	return flags;
3886 }
3887 
3888 /*
3889  * Release a prison's resources when it starts dying (when the last user
3890  * reference is dropped, or when it is killed).  Two functions are called,
3891  * for work that requires a locked prison or an unlocked one.
3892  */
3893 static void
prison_cleanup_locked(struct prison * pr)3894 prison_cleanup_locked(struct prison *pr)
3895 {
3896 	sx_assert(&allprison_lock, SA_XLOCKED);
3897 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3898 	prison_knote(pr, NOTE_JAIL_REMOVE);
3899 	knlist_detach(pr->pr_klist);
3900 	jaildesc_prison_cleanup(pr);
3901 	pr->pr_klist = NULL;
3902 }
3903 
3904 static void
prison_cleanup_unlocked(struct prison * pr)3905 prison_cleanup_unlocked(struct prison *pr)
3906 {
3907 	sx_assert(&allprison_lock, SA_XLOCKED);
3908 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3909 	vfs_exjail_delete(pr);
3910 	shm_remove_prison(pr);
3911 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3912 }
3913 
3914 /*
3915  * Set or clear a permission bit in the pr_allow field, passing restrictions
3916  * (cleared permission) down to child jails.
3917  */
3918 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)3919 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3920 {
3921 	struct prison *pr;
3922 
3923 	pr = cred->cr_prison;
3924 	sx_slock(&allprison_lock);
3925 	mtx_lock(&pr->pr_mtx);
3926 	prison_set_allow_locked(pr, flag, enable);
3927 	mtx_unlock(&pr->pr_mtx);
3928 	sx_sunlock(&allprison_lock);
3929 }
3930 
3931 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)3932 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3933 {
3934 	struct prison *cpr;
3935 	int descend;
3936 
3937 	if (enable != 0)
3938 		pr->pr_allow |= flag;
3939 	else {
3940 		pr->pr_allow &= ~flag;
3941 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3942 			cpr->pr_allow &= ~flag;
3943 	}
3944 }
3945 
3946 /*
3947  * Check if a jail supports the given address family.
3948  *
3949  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3950  * if not.
3951  */
3952 int
prison_check_af(struct ucred * cred,int af)3953 prison_check_af(struct ucred *cred, int af)
3954 {
3955 	struct prison *pr;
3956 	int error;
3957 
3958 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3959 
3960 	pr = cred->cr_prison;
3961 #ifdef VIMAGE
3962 	/* Prisons with their own network stack are not limited. */
3963 	if (prison_owns_vnet(pr))
3964 		return (0);
3965 #endif
3966 
3967 	error = 0;
3968 	switch (af)
3969 	{
3970 #ifdef INET
3971 	case AF_INET:
3972 		if (pr->pr_flags & PR_IP4)
3973 		{
3974 			mtx_lock(&pr->pr_mtx);
3975 			if ((pr->pr_flags & PR_IP4) &&
3976 			    pr->pr_addrs[PR_INET] == NULL)
3977 				error = EAFNOSUPPORT;
3978 			mtx_unlock(&pr->pr_mtx);
3979 		}
3980 		break;
3981 #endif
3982 #ifdef INET6
3983 	case AF_INET6:
3984 		if (pr->pr_flags & PR_IP6)
3985 		{
3986 			mtx_lock(&pr->pr_mtx);
3987 			if ((pr->pr_flags & PR_IP6) &&
3988 			    pr->pr_addrs[PR_INET6] == NULL)
3989 				error = EAFNOSUPPORT;
3990 			mtx_unlock(&pr->pr_mtx);
3991 		}
3992 		break;
3993 #endif
3994 	case AF_LOCAL:
3995 	case AF_ROUTE:
3996 	case AF_NETLINK:
3997 		break;
3998 	default:
3999 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
4000 			error = EAFNOSUPPORT;
4001 	}
4002 	return (error);
4003 }
4004 
4005 /*
4006  * Check if given address belongs to the jail referenced by cred (wrapper to
4007  * prison_check_ip[46]).
4008  *
4009  * Returns 0 if jail doesn't restrict the address family or if address belongs
4010  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
4011  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
4012  */
4013 int
prison_if(struct ucred * cred,const struct sockaddr * sa)4014 prison_if(struct ucred *cred, const struct sockaddr *sa)
4015 {
4016 #ifdef INET
4017 	const struct sockaddr_in *sai;
4018 #endif
4019 #ifdef INET6
4020 	const struct sockaddr_in6 *sai6;
4021 #endif
4022 	int error;
4023 
4024 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
4025 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
4026 
4027 #ifdef VIMAGE
4028 	if (prison_owns_vnet(cred->cr_prison))
4029 		return (0);
4030 #endif
4031 
4032 	error = 0;
4033 	switch (sa->sa_family)
4034 	{
4035 #ifdef INET
4036 	case AF_INET:
4037 		sai = (const struct sockaddr_in *)sa;
4038 		error = prison_check_ip4(cred, &sai->sin_addr);
4039 		break;
4040 #endif
4041 #ifdef INET6
4042 	case AF_INET6:
4043 		sai6 = (const struct sockaddr_in6 *)sa;
4044 		error = prison_check_ip6(cred, &sai6->sin6_addr);
4045 		break;
4046 #endif
4047 	default:
4048 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
4049 			error = EAFNOSUPPORT;
4050 	}
4051 	return (error);
4052 }
4053 
4054 /*
4055  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
4056  */
4057 int
prison_check(struct ucred * cred1,struct ucred * cred2)4058 prison_check(struct ucred *cred1, struct ucred *cred2)
4059 {
4060 
4061 	return ((cred1->cr_prison == cred2->cr_prison ||
4062 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
4063 }
4064 
4065 /*
4066  * For mountd/nfsd to run within a prison, it must be:
4067  * - A vnet prison.
4068  * - PR_ALLOW_NFSD must be set on it.
4069  * - The root directory (pr_root) of the prison must be
4070  *   a file system mount point, so the mountd can hang
4071  *   export information on it.
4072  * - The prison's enforce_statfs cannot be 0, so that
4073  *   mountd(8) can do exports.
4074  */
4075 bool
prison_check_nfsd(struct ucred * cred)4076 prison_check_nfsd(struct ucred *cred)
4077 {
4078 
4079 	if (jailed_without_vnet(cred))
4080 		return (false);
4081 	if (!prison_allow(cred, PR_ALLOW_NFSD))
4082 		return (false);
4083 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
4084 		return (false);
4085 	if (cred->cr_prison->pr_enforce_statfs == 0)
4086 		return (false);
4087 	return (true);
4088 }
4089 
4090 /*
4091  * Return true if p2 is a child of p1, otherwise false.
4092  */
4093 bool
prison_ischild(struct prison * pr1,struct prison * pr2)4094 prison_ischild(struct prison *pr1, struct prison *pr2)
4095 {
4096 
4097 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
4098 		if (pr1 == pr2)
4099 			return (true);
4100 	return (false);
4101 }
4102 
4103 /*
4104  * Return true if the prison is currently alive.  A prison is alive if it
4105  * holds user references and it isn't being removed.
4106  */
4107 bool
prison_isalive(const struct prison * pr)4108 prison_isalive(const struct prison *pr)
4109 {
4110 
4111 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
4112 		return (false);
4113 	return (true);
4114 }
4115 
4116 /*
4117  * Return true if the prison is currently valid.  A prison is valid if it has
4118  * been fully created, and is not being destroyed.  Note that dying prisons
4119  * are still considered valid.  Invalid prisons won't be found under normal
4120  * circumstances, as they're only put in that state by functions that have
4121  * an exclusive hold on allprison_lock.
4122  */
4123 bool
prison_isvalid(struct prison * pr)4124 prison_isvalid(struct prison *pr)
4125 {
4126 
4127 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
4128 		return (false);
4129 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
4130 		return (false);
4131 	return (true);
4132 }
4133 
4134 /*
4135  * Return true if the passed credential is in a jail and that jail does not
4136  * have its own virtual network stack, otherwise false.
4137  */
4138 bool
jailed_without_vnet(struct ucred * cred)4139 jailed_without_vnet(struct ucred *cred)
4140 {
4141 
4142 	if (!jailed(cred))
4143 		return (false);
4144 #ifdef VIMAGE
4145 	if (prison_owns_vnet(cred->cr_prison))
4146 		return (false);
4147 #endif
4148 
4149 	return (true);
4150 }
4151 
4152 /*
4153  * Return the correct hostname (domainname, et al) for the passed credential.
4154  */
4155 void
getcredhostname(struct ucred * cred,char * buf,size_t size)4156 getcredhostname(struct ucred *cred, char *buf, size_t size)
4157 {
4158 	struct prison *pr;
4159 
4160 	/*
4161 	 * A NULL credential can be used to shortcut to the physical
4162 	 * system's hostname.
4163 	 */
4164 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
4165 	mtx_lock(&pr->pr_mtx);
4166 	strlcpy(buf, pr->pr_hostname, size);
4167 	mtx_unlock(&pr->pr_mtx);
4168 }
4169 
4170 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)4171 getcreddomainname(struct ucred *cred, char *buf, size_t size)
4172 {
4173 
4174 	mtx_lock(&cred->cr_prison->pr_mtx);
4175 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
4176 	mtx_unlock(&cred->cr_prison->pr_mtx);
4177 }
4178 
4179 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)4180 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
4181 {
4182 
4183 	mtx_lock(&cred->cr_prison->pr_mtx);
4184 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
4185 	mtx_unlock(&cred->cr_prison->pr_mtx);
4186 }
4187 
4188 void
getcredhostid(struct ucred * cred,unsigned long * hostid)4189 getcredhostid(struct ucred *cred, unsigned long *hostid)
4190 {
4191 
4192 	mtx_lock(&cred->cr_prison->pr_mtx);
4193 	*hostid = cred->cr_prison->pr_hostid;
4194 	mtx_unlock(&cred->cr_prison->pr_mtx);
4195 }
4196 
4197 void
getjailname(struct ucred * cred,char * name,size_t len)4198 getjailname(struct ucred *cred, char *name, size_t len)
4199 {
4200 
4201 	mtx_lock(&cred->cr_prison->pr_mtx);
4202 	strlcpy(name, cred->cr_prison->pr_name, len);
4203 	mtx_unlock(&cred->cr_prison->pr_mtx);
4204 }
4205 
4206 #ifdef VIMAGE
4207 /*
4208  * Determine whether the prison owns its VNET.
4209  */
4210 bool
prison_owns_vnet(struct prison * pr)4211 prison_owns_vnet(struct prison *pr)
4212 {
4213 
4214 	/*
4215 	 * vnets cannot be added/removed after jail creation,
4216 	 * so no need to lock here.
4217 	 */
4218 	return ((pr->pr_flags & PR_VNET) != 0);
4219 }
4220 #endif
4221 
4222 /*
4223  * Determine whether the subject represented by cred can "see"
4224  * status of a mount point.
4225  * Returns: 0 for permitted, ENOENT otherwise.
4226  * XXX: This function should be called cr_canseemount() and should be
4227  *      placed in kern_prot.c.
4228  */
4229 int
prison_canseemount(struct ucred * cred,struct mount * mp)4230 prison_canseemount(struct ucred *cred, struct mount *mp)
4231 {
4232 	struct prison *pr;
4233 	struct statfs *sp;
4234 	size_t len;
4235 
4236 	pr = cred->cr_prison;
4237 	if (pr->pr_enforce_statfs == 0)
4238 		return (0);
4239 	if (pr->pr_root->v_mount == mp)
4240 		return (0);
4241 	if (pr->pr_enforce_statfs == 2)
4242 		return (ENOENT);
4243 	/*
4244 	 * If jail's chroot directory is set to "/" we should be able to see
4245 	 * all mount-points from inside a jail.
4246 	 * This is ugly check, but this is the only situation when jail's
4247 	 * directory ends with '/'.
4248 	 */
4249 	if (strcmp(pr->pr_path, "/") == 0)
4250 		return (0);
4251 	len = strlen(pr->pr_path);
4252 	sp = &mp->mnt_stat;
4253 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
4254 		return (ENOENT);
4255 	/*
4256 	 * Be sure that we don't have situation where jail's root directory
4257 	 * is "/some/path" and mount point is "/some/pathpath".
4258 	 */
4259 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
4260 		return (ENOENT);
4261 	return (0);
4262 }
4263 
4264 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)4265 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
4266 {
4267 	char jpath[MAXPATHLEN];
4268 	struct prison *pr;
4269 	size_t len;
4270 
4271 	pr = cred->cr_prison;
4272 	if (pr->pr_enforce_statfs == 0)
4273 		return;
4274 	if (prison_canseemount(cred, mp) != 0) {
4275 		bzero(&sp->f_fsid, sizeof(sp->f_fsid));
4276 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4277 		strlcpy(sp->f_mntonname, "[restricted]",
4278 		    sizeof(sp->f_mntonname));
4279 		return;
4280 	}
4281 	if (pr->pr_enforce_statfs > 1)
4282 		bzero(&sp->f_fsid, sizeof(sp->f_fsid));
4283 	if (pr->pr_root->v_mount == mp) {
4284 		/*
4285 		 * Clear current buffer data, so we are sure nothing from
4286 		 * the valid path left there.
4287 		 */
4288 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4289 		*sp->f_mntonname = '/';
4290 		return;
4291 	}
4292 	/*
4293 	 * If jail's chroot directory is set to "/" we should be able to see
4294 	 * all mount-points from inside a jail.
4295 	 */
4296 	if (strcmp(pr->pr_path, "/") == 0)
4297 		return;
4298 	len = strlen(pr->pr_path);
4299 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
4300 	/*
4301 	 * Clear current buffer data, so we are sure nothing from
4302 	 * the valid path left there.
4303 	 */
4304 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4305 	if (*jpath == '\0') {
4306 		/* Should never happen. */
4307 		*sp->f_mntonname = '/';
4308 	} else {
4309 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
4310 	}
4311 }
4312 
4313 /*
4314  * Check with permission for a specific privilege is granted within jail.  We
4315  * have a specific list of accepted privileges; the rest are denied.
4316  */
4317 int
prison_priv_check(struct ucred * cred,int priv)4318 prison_priv_check(struct ucred *cred, int priv)
4319 {
4320 	struct prison *pr;
4321 	int error;
4322 
4323 	/*
4324 	 * Some policies have custom handlers. This routine should not be
4325 	 * called for them. See priv_check_cred().
4326 	 */
4327 	switch (priv) {
4328 	case PRIV_VFS_LOOKUP:
4329 	case PRIV_VFS_GENERATION:
4330 		KASSERT(0, ("prison_priv_check instead of a custom handler "
4331 		    "called for %d\n", priv));
4332 	}
4333 
4334 	if (!jailed(cred))
4335 		return (0);
4336 
4337 #ifdef VIMAGE
4338 	/*
4339 	 * Privileges specific to prisons with a virtual network stack.
4340 	 * There might be a duplicate entry here in case the privilege
4341 	 * is only granted conditionally in the legacy jail case.
4342 	 */
4343 	switch (priv) {
4344 		/*
4345 		 * NFS-specific privileges.
4346 		 */
4347 	case PRIV_NFS_DAEMON:
4348 	case PRIV_VFS_GETFH:
4349 	case PRIV_VFS_MOUNT_EXPORTED:
4350 		if (!prison_check_nfsd(cred))
4351 			return (EPERM);
4352 #ifdef notyet
4353 	case PRIV_NFS_LOCKD:
4354 #endif
4355 		/*
4356 		 * Network stack privileges.
4357 		 */
4358 	case PRIV_NET_BRIDGE:
4359 	case PRIV_NET_GRE:
4360 	case PRIV_NET_BPF:
4361 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
4362 	case PRIV_NET_ROUTE:
4363 	case PRIV_NET_TAP:
4364 	case PRIV_NET_SETIFMTU:
4365 	case PRIV_NET_SETIFFLAGS:
4366 	case PRIV_NET_SETIFCAP:
4367 	case PRIV_NET_SETIFDESCR:
4368 	case PRIV_NET_SETIFNAME	:
4369 	case PRIV_NET_SETIFMETRIC:
4370 	case PRIV_NET_SETIFPHYS:
4371 	case PRIV_NET_SETIFMAC:
4372 	case PRIV_NET_SETLANPCP:
4373 	case PRIV_NET_ADDMULTI:
4374 	case PRIV_NET_DELMULTI:
4375 	case PRIV_NET_HWIOCTL:
4376 	case PRIV_NET_SETLLADDR:
4377 	case PRIV_NET_ADDIFGROUP:
4378 	case PRIV_NET_DELIFGROUP:
4379 	case PRIV_NET_IFCREATE:
4380 	case PRIV_NET_IFDESTROY:
4381 	case PRIV_NET_ADDIFADDR:
4382 	case PRIV_NET_DELIFADDR:
4383 	case PRIV_NET_LAGG:
4384 	case PRIV_NET_GIF:
4385 	case PRIV_NET_SETIFVNET:
4386 	case PRIV_NET_SETIFFIB:
4387 	case PRIV_NET_OVPN:
4388 	case PRIV_NET_ME:
4389 	case PRIV_NET_WG:
4390 
4391 		/*
4392 		 * 802.11-related privileges.
4393 		 */
4394 	case PRIV_NET80211_VAP_GETKEY:
4395 	case PRIV_NET80211_VAP_MANAGE:
4396 
4397 #ifdef notyet
4398 		/*
4399 		 * ATM privileges.
4400 		 */
4401 	case PRIV_NETATM_CFG:
4402 	case PRIV_NETATM_ADD:
4403 	case PRIV_NETATM_DEL:
4404 	case PRIV_NETATM_SET:
4405 
4406 		/*
4407 		 * Bluetooth privileges.
4408 		 */
4409 	case PRIV_NETBLUETOOTH_RAW:
4410 #endif
4411 
4412 		/*
4413 		 * Netgraph and netgraph module privileges.
4414 		 */
4415 	case PRIV_NETGRAPH_CONTROL:
4416 #ifdef notyet
4417 	case PRIV_NETGRAPH_TTY:
4418 #endif
4419 
4420 		/*
4421 		 * IPv4 and IPv6 privileges.
4422 		 */
4423 	case PRIV_NETINET_IPFW:
4424 	case PRIV_NETINET_DIVERT:
4425 	case PRIV_NETINET_PF:
4426 	case PRIV_NETINET_DUMMYNET:
4427 	case PRIV_NETINET_CARP:
4428 	case PRIV_NETINET_MROUTE:
4429 	case PRIV_NETINET_RAW:
4430 	case PRIV_NETINET_ADDRCTRL6:
4431 	case PRIV_NETINET_ND6:
4432 	case PRIV_NETINET_SCOPE6:
4433 	case PRIV_NETINET_ALIFETIME6:
4434 	case PRIV_NETINET_IPSEC:
4435 	case PRIV_NETINET_BINDANY:
4436 
4437 #ifdef notyet
4438 		/*
4439 		 * NCP privileges.
4440 		 */
4441 	case PRIV_NETNCP:
4442 
4443 		/*
4444 		 * SMB privileges.
4445 		 */
4446 	case PRIV_NETSMB:
4447 #endif
4448 
4449 	/*
4450 	 * No default: or deny here.
4451 	 * In case of no permit fall through to next switch().
4452 	 */
4453 		if (cred->cr_prison->pr_flags & PR_VNET)
4454 			return (0);
4455 	}
4456 #endif /* VIMAGE */
4457 
4458 	switch (priv) {
4459 		/*
4460 		 * Allow ktrace privileges for root in jail.
4461 		 */
4462 	case PRIV_KTRACE:
4463 
4464 		/*
4465 		 * Allow jailed processes to configure audit identity and
4466 		 * submit audit records (login, etc).  In the future we may
4467 		 * want to further refine the relationship between audit and
4468 		 * jail.
4469 		 */
4470 	case PRIV_AUDIT_GETAUDIT:
4471 	case PRIV_AUDIT_SETAUDIT:
4472 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETAUDIT)
4473 			return (0);
4474 		else
4475 			return (EPERM);
4476 #if 0
4477 	case PRIV_AUDIT_SUBMIT:
4478 #endif
4479 
4480 		/*
4481 		 * Allow jailed processes to manipulate process UNIX
4482 		 * credentials in any way they see fit.
4483 		 */
4484 	case PRIV_CRED_SETCRED:
4485 	case PRIV_CRED_SETUID:
4486 	case PRIV_CRED_SETEUID:
4487 	case PRIV_CRED_SETGID:
4488 	case PRIV_CRED_SETEGID:
4489 	case PRIV_CRED_SETGROUPS:
4490 	case PRIV_CRED_SETREUID:
4491 	case PRIV_CRED_SETREGID:
4492 	case PRIV_CRED_SETRESUID:
4493 	case PRIV_CRED_SETRESGID:
4494 
4495 		/*
4496 		 * Jail implements visibility constraints already, so allow
4497 		 * jailed root to override uid/gid-based constraints.
4498 		 */
4499 	case PRIV_SEEOTHERGIDS:
4500 	case PRIV_SEEOTHERUIDS:
4501 	case PRIV_SEEJAILPROC:
4502 
4503 		/*
4504 		 * Jail implements inter-process debugging limits already, so
4505 		 * allow jailed root various debugging privileges.
4506 		 */
4507 	case PRIV_DEBUG_DIFFCRED:
4508 	case PRIV_DEBUG_SUGID:
4509 	case PRIV_DEBUG_UNPRIV:
4510 	case PRIV_DEBUG_DIFFJAIL:
4511 
4512 		/*
4513 		 * Allow jail to set various resource limits and login
4514 		 * properties, and for now, exceed process resource limits.
4515 		 */
4516 	case PRIV_PROC_LIMIT:
4517 	case PRIV_PROC_SETLOGIN:
4518 	case PRIV_PROC_SETRLIMIT:
4519 
4520 		/*
4521 		 * Debuggers should work in jails.
4522 		 */
4523 	case PRIV_PROC_MEM_WRITE:
4524 
4525 		/*
4526 		 * System V and POSIX IPC privileges are granted in jail.
4527 		 */
4528 	case PRIV_IPC_READ:
4529 	case PRIV_IPC_WRITE:
4530 	case PRIV_IPC_ADMIN:
4531 	case PRIV_IPC_MSGSIZE:
4532 	case PRIV_MQ_ADMIN:
4533 
4534 		/*
4535 		 * Jail operations within a jail work on child jails.
4536 		 */
4537 	case PRIV_JAIL_ATTACH:
4538 	case PRIV_JAIL_SET:
4539 	case PRIV_JAIL_REMOVE:
4540 
4541 		/*
4542 		 * Jail implements its own inter-process limits, so allow
4543 		 * root processes in jail to change scheduling on other
4544 		 * processes in the same jail.  Likewise for signalling.
4545 		 */
4546 	case PRIV_SCHED_DIFFCRED:
4547 	case PRIV_SCHED_CPUSET:
4548 	case PRIV_SCHED_DIFFJAIL:
4549 	case PRIV_SIGNAL_DIFFCRED:
4550 	case PRIV_SIGNAL_SUGID:
4551 	case PRIV_SIGNAL_DIFFJAIL:
4552 
4553 		/*
4554 		 * Allow jailed processes to write to sysctls marked as jail
4555 		 * writable.
4556 		 */
4557 	case PRIV_SYSCTL_WRITEJAIL:
4558 
4559 		/*
4560 		 * Allow root in jail to manage a variety of quota
4561 		 * properties.  These should likely be conditional on a
4562 		 * configuration option.
4563 		 */
4564 	case PRIV_VFS_GETQUOTA:
4565 	case PRIV_VFS_SETQUOTA:
4566 
4567 		/*
4568 		 * Since Jail relies on chroot() to implement file system
4569 		 * protections, grant many VFS privileges to root in jail.
4570 		 * Be careful to exclude mount-related and NFS-related
4571 		 * privileges.
4572 		 */
4573 	case PRIV_VFS_READ:
4574 	case PRIV_VFS_WRITE:
4575 	case PRIV_VFS_ADMIN:
4576 	case PRIV_VFS_EXEC:
4577 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
4578 	case PRIV_VFS_CHFLAGS_DEV:
4579 	case PRIV_VFS_CHOWN:
4580 	case PRIV_VFS_CHROOT:
4581 	case PRIV_VFS_RETAINSUGID:
4582 	case PRIV_VFS_FCHROOT:
4583 	case PRIV_VFS_LINK:
4584 	case PRIV_VFS_SETGID:
4585 	case PRIV_VFS_STAT:
4586 	case PRIV_VFS_STICKYFILE:
4587 
4588 		/*
4589 		 * As in the non-jail case, non-root users are expected to be
4590 		 * able to read kernel/physical memory (provided /dev/[k]mem
4591 		 * exists in the jail and they have permission to access it).
4592 		 */
4593 	case PRIV_KMEM_READ:
4594 		return (0);
4595 
4596 		/*
4597 		 * Depending on the global setting, allow privilege of
4598 		 * setting system flags.
4599 		 */
4600 	case PRIV_VFS_SYSFLAGS:
4601 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4602 			return (0);
4603 		else
4604 			return (EPERM);
4605 
4606 		/*
4607 		 * Depending on the global setting, allow privilege of
4608 		 * mounting/unmounting file systems.
4609 		 */
4610 	case PRIV_VFS_MOUNT:
4611 	case PRIV_VFS_UNMOUNT:
4612 	case PRIV_VFS_MOUNT_NONUSER:
4613 	case PRIV_VFS_MOUNT_OWNER:
4614 		pr = cred->cr_prison;
4615 		prison_lock(pr);
4616 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4617 			error = 0;
4618 		else
4619 			error = EPERM;
4620 		prison_unlock(pr);
4621 		return (error);
4622 
4623 		/*
4624 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4625 		 * policy.  priv_check_cred will not specifically allow it, and
4626 		 * we may want a MAC policy to allow it.
4627 		 */
4628 	case PRIV_VFS_READ_DIR:
4629 		return (0);
4630 
4631 		/*
4632 		 * Conditionally allow privileged process in the jail to
4633 		 * manipulate filesystem extended attributes in the system
4634 		 * namespace.
4635 		 */
4636 	case PRIV_VFS_EXTATTR_SYSTEM:
4637 		if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4638 			return (0);
4639 		else
4640 			return (EPERM);
4641 
4642 		/*
4643 		 * Conditionnaly allow locking (unlocking) physical pages
4644 		 * in memory.
4645 		 */
4646 	case PRIV_VM_MLOCK:
4647 	case PRIV_VM_MUNLOCK:
4648 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4649 			return (0);
4650 		else
4651 			return (EPERM);
4652 
4653 		/*
4654 		 * Conditionally allow jailed root to bind reserved ports.
4655 		 */
4656 	case PRIV_NETINET_RESERVEDPORT:
4657 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4658 			return (0);
4659 		else
4660 			return (EPERM);
4661 
4662 		/*
4663 		 * Allow jailed root to reuse in-use ports.
4664 		 */
4665 	case PRIV_NETINET_REUSEPORT:
4666 		return (0);
4667 
4668 		/*
4669 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4670 		 */
4671 	case PRIV_NETINET_SETHDROPTS:
4672 		return (0);
4673 
4674 		/*
4675 		 * Conditionally allow creating raw sockets in jail.
4676 		 */
4677 	case PRIV_NETINET_RAW:
4678 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4679 			return (0);
4680 		else
4681 			return (EPERM);
4682 
4683 		/*
4684 		 * Since jail implements its own visibility limits on netstat
4685 		 * sysctls, allow getcred.  This allows identd to work in
4686 		 * jail.
4687 		 */
4688 	case PRIV_NETINET_GETCRED:
4689 		return (0);
4690 
4691 		/*
4692 		 * Allow jailed root to set loginclass.
4693 		 */
4694 	case PRIV_PROC_SETLOGINCLASS:
4695 		return (0);
4696 
4697 		/*
4698 		 * Do not allow a process inside a jail to read the kernel
4699 		 * message buffer unless explicitly permitted.
4700 		 */
4701 	case PRIV_MSGBUF:
4702 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4703 			return (0);
4704 		return (EPERM);
4705 
4706 		/*
4707 		 * Conditionally allow privileged process in the jail adjust
4708 		 * machine time.
4709 		 */
4710 	case PRIV_ADJTIME:
4711 	case PRIV_NTP_ADJTIME:
4712 		if (cred->cr_prison->pr_allow &
4713 		    (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4714 			return (0);
4715 		}
4716 		return (EPERM);
4717 
4718 		/*
4719 		 * Conditionally allow privileged process in the jail set
4720 		 * machine time.
4721 		 */
4722 	case PRIV_SETTIMEOFDAY:
4723 	case PRIV_CLOCK_SETTIME:
4724 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4725 			return (0);
4726 		else
4727 			return (EPERM);
4728 
4729 		/*
4730 		 * Conditionally allow privileged process in the jail to modify
4731 		 * the routing table.
4732 		 */
4733 	case PRIV_NET_ROUTE:
4734 		if (cred->cr_prison->pr_allow & PR_ALLOW_ROUTING)
4735 			return (0);
4736 		else
4737 			return (EPERM);
4738 
4739 	case PRIV_VMM_PPTDEV:
4740 		/*
4741 		 * Allow jailed root to manage passthrough devices.  vmm(4) also
4742 		 * checks for the dynamically added allow.vmm_ppt.
4743 		 */
4744 		return (0);
4745 
4746 	default:
4747 		/*
4748 		 * In all remaining cases, deny the privilege request.  This
4749 		 * includes almost all network privileges, many system
4750 		 * configuration privileges.
4751 		 */
4752 		return (EPERM);
4753 	}
4754 }
4755 
4756 /*
4757  * Return the part of pr2's name that is relative to pr1, or the whole name
4758  * if it does not directly follow.
4759  */
4760 
4761 char *
prison_name(struct prison * pr1,struct prison * pr2)4762 prison_name(struct prison *pr1, struct prison *pr2)
4763 {
4764 	char *name;
4765 
4766 	/* Jails see themselves as "0" (if they see themselves at all). */
4767 	if (pr1 == pr2)
4768 		return "0";
4769 	name = pr2->pr_name;
4770 	if (prison_ischild(pr1, pr2)) {
4771 		/*
4772 		 * pr1 isn't locked (and allprison_lock may not be either)
4773 		 * so its length can't be counted on.  But the number of dots
4774 		 * can be counted on - and counted.
4775 		 */
4776 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4777 			name = strchr(name, '.') + 1;
4778 	}
4779 	return (name);
4780 }
4781 
4782 /*
4783  * Return the part of pr2's path that is relative to pr1, or the whole path
4784  * if it does not directly follow.
4785  */
4786 static char *
prison_path(struct prison * pr1,struct prison * pr2)4787 prison_path(struct prison *pr1, struct prison *pr2)
4788 {
4789 	char *path1, *path2;
4790 	int len1;
4791 
4792 	path1 = pr1->pr_path;
4793 	path2 = pr2->pr_path;
4794 	if (!strcmp(path1, "/"))
4795 		return (path2);
4796 	len1 = strlen(path1);
4797 	if (strncmp(path1, path2, len1))
4798 		return (path2);
4799 	if (path2[len1] == '\0')
4800 		return "/";
4801 	if (path2[len1] == '/')
4802 		return (path2 + len1);
4803 	return (path2);
4804 }
4805 
4806 /*
4807  * Jail-related sysctls.
4808  */
4809 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4810     "Jails");
4811 
4812 #if defined(INET) || defined(INET6)
4813 /*
4814  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4815  * sysctl_jail_list() helper.
4816  */
4817 static void
prison_ip_copyout(struct prison * pr,const pr_family_t af,void ** out,int * len)4818 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4819 {
4820 	const struct prison_ip *pip;
4821 	const size_t size = pr_families[af].size;
4822 
4823  again:
4824 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4825 	if ((pip = pr->pr_addrs[af]) != NULL) {
4826 		if (*len < pip->ips) {
4827 			*len = pip->ips;
4828 			mtx_unlock(&pr->pr_mtx);
4829 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4830 			mtx_lock(&pr->pr_mtx);
4831 			goto again;
4832 		}
4833 		bcopy(pip->pr_ip, *out, pip->ips * size);
4834 	}
4835 }
4836 #endif
4837 
4838 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4839 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4840 {
4841 	struct xprison *xp;
4842 	struct prison *pr, *cpr;
4843 #ifdef INET
4844 	struct in_addr *ip4 = NULL;
4845 	int ip4s = 0;
4846 #endif
4847 #ifdef INET6
4848 	struct in6_addr *ip6 = NULL;
4849 	int ip6s = 0;
4850 #endif
4851 	int descend, error;
4852 
4853 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4854 	pr = req->td->td_ucred->cr_prison;
4855 	error = 0;
4856 	sx_slock(&allprison_lock);
4857 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4858 		mtx_lock(&cpr->pr_mtx);
4859 #ifdef INET
4860 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4861 #endif
4862 #ifdef INET6
4863 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4864 #endif
4865 		bzero(xp, sizeof(*xp));
4866 		xp->pr_version = XPRISON_VERSION;
4867 		xp->pr_id = cpr->pr_id;
4868 		xp->pr_state = cpr->pr_state;
4869 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4870 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4871 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4872 #ifdef INET
4873 		xp->pr_ip4s = ip4s;
4874 #endif
4875 #ifdef INET6
4876 		xp->pr_ip6s = ip6s;
4877 #endif
4878 		mtx_unlock(&cpr->pr_mtx);
4879 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4880 		if (error)
4881 			break;
4882 #ifdef INET
4883 		if (xp->pr_ip4s > 0) {
4884 			error = SYSCTL_OUT(req, ip4,
4885 			    xp->pr_ip4s * sizeof(struct in_addr));
4886 			if (error)
4887 				break;
4888 		}
4889 #endif
4890 #ifdef INET6
4891 		if (xp->pr_ip6s > 0) {
4892 			error = SYSCTL_OUT(req, ip6,
4893 			    xp->pr_ip6s * sizeof(struct in6_addr));
4894 			if (error)
4895 				break;
4896 		}
4897 #endif
4898 	}
4899 	sx_sunlock(&allprison_lock);
4900 	free(xp, M_TEMP);
4901 #ifdef INET
4902 	free(ip4, M_TEMP);
4903 #endif
4904 #ifdef INET6
4905 	free(ip6, M_TEMP);
4906 #endif
4907 	return (error);
4908 }
4909 
4910 SYSCTL_OID(_security_jail, OID_AUTO, list,
4911     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4912     sysctl_jail_list, "S", "List of active jails");
4913 
4914 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)4915 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4916 {
4917 	int error, injail;
4918 
4919 	injail = jailed(req->td->td_ucred);
4920 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4921 
4922 	return (error);
4923 }
4924 
4925 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4926     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4927     sysctl_jail_jailed, "I", "Process in jail?");
4928 
4929 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)4930 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4931 {
4932 	int error, havevnet;
4933 #ifdef VIMAGE
4934 	struct ucred *cred = req->td->td_ucred;
4935 
4936 	havevnet = jailed(cred) && prison_owns_vnet(cred->cr_prison);
4937 #else
4938 	havevnet = 0;
4939 #endif
4940 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4941 
4942 	return (error);
4943 }
4944 
4945 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4946     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4947     sysctl_jail_vnet, "I", "Jail owns vnet?");
4948 
4949 #if defined(INET) || defined(INET6)
4950 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4951     &jail_max_af_ips, 0,
4952     "Number of IP addresses a jail may have at most per address family (deprecated)");
4953 #endif
4954 
4955 /*
4956  * Default parameters for jail(2) compatibility.  For historical reasons,
4957  * the sysctl names have varying similarity to the parameter names.  Prisons
4958  * just see their own parameters, and can't change them.
4959  */
4960 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)4961 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4962 {
4963 	int error, i;
4964 
4965 	/* Get the current flag value, and convert it to a boolean. */
4966 	if (req->td->td_ucred->cr_prison == &prison0) {
4967 		mtx_lock(&prison0.pr_mtx);
4968 		i = (jail_default_allow & arg2) != 0;
4969 		mtx_unlock(&prison0.pr_mtx);
4970 	} else
4971 		i = prison_allow(req->td->td_ucred, arg2);
4972 
4973 	if (arg1 != NULL)
4974 		i = !i;
4975 	error = sysctl_handle_int(oidp, &i, 0, req);
4976 	if (error || !req->newptr)
4977 		return (error);
4978 	i = i ? arg2 : 0;
4979 	if (arg1 != NULL)
4980 		i ^= arg2;
4981 	/*
4982 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4983 	 * for writing.
4984 	 */
4985 	mtx_lock(&prison0.pr_mtx);
4986 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4987 	mtx_unlock(&prison0.pr_mtx);
4988 	return (0);
4989 }
4990 
4991 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4992     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4993     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4994     "Processes in jail can set their hostnames (deprecated)");
4995 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4996     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4997     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4998     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4999 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
5000     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5001     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
5002     "Processes in jail can use System V IPC primitives (deprecated)");
5003 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
5004     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5005     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
5006     "Prison root can create raw sockets (deprecated)");
5007 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
5008     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5009     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
5010     "Processes in jail can alter system file flags (deprecated)");
5011 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
5012     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5013     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
5014     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
5015 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
5016     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5017     NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
5018     "Processes in jail can lock/unlock physical pages in memory");
5019 
5020 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)5021 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
5022 {
5023 	struct prison *pr;
5024 	int level, error;
5025 
5026 	pr = req->td->td_ucred->cr_prison;
5027 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
5028 	error = sysctl_handle_int(oidp, &level, 0, req);
5029 	if (error || !req->newptr)
5030 		return (error);
5031 	*(int *)arg1 = level;
5032 	return (0);
5033 }
5034 
5035 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
5036     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5037     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
5038     sysctl_jail_default_level, "I",
5039     "Processes in jail cannot see all mounted file systems (deprecated)");
5040 
5041 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
5042     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
5043     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
5044     sysctl_jail_default_level, "I",
5045     "Ruleset for the devfs filesystem in jail (deprecated)");
5046 
5047 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
5048     "Limits and stats of child jails");
5049 
5050 static int
sysctl_jail_children(SYSCTL_HANDLER_ARGS)5051 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
5052 {
5053 	struct prison *pr;
5054 	int i;
5055 
5056 	pr = req->td->td_ucred->cr_prison;
5057 
5058 	switch (oidp->oid_kind & CTLTYPE) {
5059 	case CTLTYPE_INT:
5060 		i = *(int *)((char *)pr + arg2);
5061 		return (SYSCTL_OUT(req, &i, sizeof(i)));
5062 	}
5063 
5064 	return (0);
5065 }
5066 
5067 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
5068     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
5069     NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
5070     "I", "Maximum number of child jails");
5071 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
5072     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
5073     NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
5074     "I", "Current number of child jails");
5075 
5076 /*
5077  * Nodes to describe jail parameters.  Maximum length of string parameters
5078  * is returned in the string itself, and the other parameters exist merely
5079  * to make themselves and their types known.
5080  */
5081 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
5082     "Jail parameters");
5083 
5084 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)5085 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
5086 {
5087 	int i;
5088 	long l;
5089 	size_t s;
5090 	char numbuf[12];
5091 
5092 	switch (oidp->oid_kind & CTLTYPE)
5093 	{
5094 	case CTLTYPE_LONG:
5095 	case CTLTYPE_ULONG:
5096 		l = 0;
5097 #ifdef SCTL_MASK32
5098 		if (!(req->flags & SCTL_MASK32))
5099 #endif
5100 			return (SYSCTL_OUT(req, &l, sizeof(l)));
5101 	case CTLTYPE_INT:
5102 	case CTLTYPE_UINT:
5103 		i = 0;
5104 		return (SYSCTL_OUT(req, &i, sizeof(i)));
5105 	case CTLTYPE_STRING:
5106 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
5107 		return
5108 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
5109 	case CTLTYPE_STRUCT:
5110 		s = (size_t)arg2;
5111 		return (SYSCTL_OUT(req, &s, sizeof(s)));
5112 	}
5113 	return (0);
5114 }
5115 
5116 /*
5117  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
5118  * jail creation time but cannot be changed in an existing jail.
5119  */
5120 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
5121 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
5122 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
5123 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
5124 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
5125     "I", "Jail secure level");
5126 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
5127     "Jail value for kern.osreldate and uname -K");
5128 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
5129     "Jail value for kern.osrelease and uname -r");
5130 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
5131     "I", "Jail cannot see all mounted file systems");
5132 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
5133     "I", "Ruleset for in-jail devfs mounts");
5134 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
5135     "B", "Jail persistence");
5136 #ifdef VIMAGE
5137 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
5138     "E,jailsys", "Virtual network stack");
5139 #endif
5140 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
5141     "B", "Jail is in the process of shutting down");
5142 
5143 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
5144 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
5145     "I", "Current number of child jails");
5146 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
5147     "I", "Maximum number of child jails");
5148 
5149 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
5150 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
5151     "Jail hostname");
5152 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
5153     "Jail NIS domainname");
5154 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
5155     "Jail host UUID");
5156 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
5157     "LU", "Jail host ID");
5158 
5159 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
5160 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
5161 
5162 #ifdef MAC
5163 SYSCTL_JAIL_PARAM_STRUCT(_mac, label, CTLFLAG_RW, sizeof(struct mac),
5164     "S,mac", "Jail MAC label");
5165 #endif
5166 
5167 #ifdef INET
5168 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
5169     "Jail IPv4 address virtualization");
5170 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
5171     "S,in_addr,a", "Jail IPv4 addresses");
5172 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5173     "B", "Do (not) use IPv4 source address selection rather than the "
5174     "primary jail IPv4 address.");
5175 #endif
5176 #ifdef INET6
5177 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
5178     "Jail IPv6 address virtualization");
5179 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
5180     "S,in6_addr,a", "Jail IPv6 addresses");
5181 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5182     "B", "Do (not) use IPv6 source address selection rather than the "
5183     "primary jail IPv6 address.");
5184 #endif
5185 
5186 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
5187 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
5188     "B", "Jail may set hostname");
5189 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
5190     "B", "Jail may use SYSV IPC");
5191 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
5192     "B", "Jail may create raw sockets");
5193 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
5194     "B", "Jail may alter system file flags");
5195 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
5196     "B", "Jail may set file quotas");
5197 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
5198     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
5199 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
5200     "B", "Jail may lock (unlock) physical pages in memory");
5201 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
5202     "B", "Jail may bind sockets to reserved ports");
5203 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
5204     "B", "Jail may read the kernel message buffer");
5205 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
5206     "B", "Unprivileged processes may use process debugging facilities");
5207 SYSCTL_JAIL_PARAM(_allow, unprivileged_parent_tampering,
5208     CTLTYPE_INT | CTLFLAG_RW, "B",
5209     "Unprivileged parent jail processes may tamper with same-uid processes"
5210     " (signal/debug/cpuset)");
5211 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
5212     "B", "Processes in jail with uid 0 have privilege");
5213 #ifdef VIMAGE
5214 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
5215     "B", "Mountd/nfsd may run in the jail");
5216 #endif
5217 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
5218     "B", "Jail may set system-level filesystem extended attributes");
5219 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
5220     "B", "Jail may adjust system time");
5221 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
5222     "B", "Jail may set system time");
5223 SYSCTL_JAIL_PARAM(_allow, routing, CTLTYPE_INT | CTLFLAG_RW,
5224     "B", "Jail may modify routing table");
5225 #ifdef AUDIT
5226 SYSCTL_JAIL_PARAM(_allow, setaudit, CTLTYPE_INT | CTLFLAG_RW,
5227     "B", "Jail may set and get audit session state");
5228 #endif
5229 
5230 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
5231 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
5232     "B", "Jail may mount/unmount jail-friendly file systems in general");
5233 
5234 /*
5235  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
5236  * its associated bit in the pr_allow bitmask, or zero if the parameter was
5237  * not created.
5238  */
5239 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)5240 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
5241     const char *descr)
5242 {
5243 	struct bool_flags *bf;
5244 	struct sysctl_oid *parent;
5245 	char *allow_name, *allow_noname, *allowed;
5246 #ifndef NO_SYSCTL_DESCR
5247 	char *descr_deprecated;
5248 #endif
5249 	u_int allow_flag;
5250 
5251 	if (prefix
5252 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
5253 		< 0 ||
5254 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
5255 		< 0
5256 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
5257 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
5258 		free(allow_name, M_PRISON);
5259 		return 0;
5260 	}
5261 
5262 	/*
5263 	 * See if this parameter has already beed added, i.e. a module was
5264 	 * previously loaded/unloaded.
5265 	 */
5266 	mtx_lock(&prison0.pr_mtx);
5267 	for (bf = pr_flag_allow;
5268 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5269 		atomic_load_int(&bf->flag) != 0;
5270 	     bf++) {
5271 		if (strcmp(bf->name, allow_name) == 0) {
5272 			allow_flag = bf->flag;
5273 			goto no_add;
5274 		}
5275 	}
5276 
5277 	/*
5278 	 * Find a free bit in pr_allow_all, failing if there are none
5279 	 * (which shouldn't happen as long as we keep track of how many
5280 	 * potential dynamic flags exist).
5281 	 */
5282 	for (allow_flag = 1;; allow_flag <<= 1) {
5283 		if (allow_flag == 0)
5284 			goto no_add;
5285 		if ((pr_allow_all & allow_flag) == 0)
5286 			break;
5287 	}
5288 
5289 	/* Note the parameter in the next open slot in pr_flag_allow. */
5290 	for (bf = pr_flag_allow; ; bf++) {
5291 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
5292 			/* This should never happen, but is not fatal. */
5293 			allow_flag = 0;
5294 			goto no_add;
5295 		}
5296 		if (atomic_load_int(&bf->flag) == 0)
5297 			break;
5298 	}
5299 	bf->name = allow_name;
5300 	bf->noname = allow_noname;
5301 	pr_allow_all |= allow_flag;
5302 	/*
5303 	 * prison0 always has permission for the new parameter.
5304 	 * Other jails must have it granted to them.
5305 	 */
5306 	prison0.pr_allow |= allow_flag;
5307 	/* The flag indicates a valid entry, so make sure it is set last. */
5308 	atomic_store_rel_int(&bf->flag, allow_flag);
5309 	mtx_unlock(&prison0.pr_mtx);
5310 
5311 	/*
5312 	 * Create sysctls for the parameter, and the back-compat global
5313 	 * permission.
5314 	 */
5315 	parent = prefix
5316 	    ? SYSCTL_ADD_NODE(NULL,
5317 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
5318 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
5319 	    : &sysctl___security_jail_param_allow;
5320 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
5321 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5322 	    NULL, 0, sysctl_jail_param, "B", descr);
5323 	if ((prefix
5324 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
5325 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
5326 #ifndef NO_SYSCTL_DESCR
5327 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
5328 		    descr);
5329 #endif
5330 		(void)SYSCTL_ADD_PROC(NULL,
5331 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
5332 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
5333 		    sysctl_jail_default_allow, "I", descr_deprecated);
5334 #ifndef NO_SYSCTL_DESCR
5335 		free(descr_deprecated, M_TEMP);
5336 #endif
5337 		free(allowed, M_TEMP);
5338 	}
5339 	return allow_flag;
5340 
5341  no_add:
5342 	mtx_unlock(&prison0.pr_mtx);
5343 	free(allow_name, M_PRISON);
5344 	free(allow_noname, M_PRISON);
5345 	return allow_flag;
5346 }
5347 
5348 /*
5349  * The VFS system will register jail-aware filesystems here.  They each get
5350  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
5351  * attempts to mount.
5352  */
5353 void
prison_add_vfs(struct vfsconf * vfsp)5354 prison_add_vfs(struct vfsconf *vfsp)
5355 {
5356 #ifdef NO_SYSCTL_DESCR
5357 
5358 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5359 	    NULL, NULL);
5360 #else
5361 	char *descr;
5362 
5363 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
5364 	    vfsp->vfc_name);
5365 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5366 	    NULL, descr);
5367 	free(descr, M_TEMP);
5368 #endif
5369 }
5370 
5371 #ifdef RACCT
5372 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)5373 prison_racct_foreach(void (*callback)(struct racct *racct,
5374     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
5375     void *arg2, void *arg3)
5376 {
5377 	struct prison_racct *prr;
5378 
5379 	ASSERT_RACCT_ENABLED();
5380 
5381 	sx_slock(&allprison_lock);
5382 	if (pre != NULL)
5383 		(pre)();
5384 	LIST_FOREACH(prr, &allprison_racct, prr_next)
5385 		(callback)(prr->prr_racct, arg2, arg3);
5386 	if (post != NULL)
5387 		(post)();
5388 	sx_sunlock(&allprison_lock);
5389 }
5390 
5391 static struct prison_racct *
prison_racct_find_locked(const char * name)5392 prison_racct_find_locked(const char *name)
5393 {
5394 	struct prison_racct *prr;
5395 
5396 	ASSERT_RACCT_ENABLED();
5397 	sx_assert(&allprison_lock, SA_XLOCKED);
5398 
5399 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
5400 		return (NULL);
5401 
5402 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
5403 		if (strcmp(name, prr->prr_name) != 0)
5404 			continue;
5405 
5406 		/* Found prison_racct with a matching name? */
5407 		prison_racct_hold(prr);
5408 		return (prr);
5409 	}
5410 
5411 	/* Add new prison_racct. */
5412 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
5413 	racct_create(&prr->prr_racct);
5414 
5415 	strcpy(prr->prr_name, name);
5416 	refcount_init(&prr->prr_refcount, 1);
5417 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
5418 
5419 	return (prr);
5420 }
5421 
5422 struct prison_racct *
prison_racct_find(const char * name)5423 prison_racct_find(const char *name)
5424 {
5425 	struct prison_racct *prr;
5426 
5427 	ASSERT_RACCT_ENABLED();
5428 
5429 	sx_xlock(&allprison_lock);
5430 	prr = prison_racct_find_locked(name);
5431 	sx_xunlock(&allprison_lock);
5432 	return (prr);
5433 }
5434 
5435 void
prison_racct_hold(struct prison_racct * prr)5436 prison_racct_hold(struct prison_racct *prr)
5437 {
5438 
5439 	ASSERT_RACCT_ENABLED();
5440 
5441 	refcount_acquire(&prr->prr_refcount);
5442 }
5443 
5444 static void
prison_racct_free_locked(struct prison_racct * prr)5445 prison_racct_free_locked(struct prison_racct *prr)
5446 {
5447 
5448 	ASSERT_RACCT_ENABLED();
5449 	sx_assert(&allprison_lock, SA_XLOCKED);
5450 
5451 	if (refcount_release(&prr->prr_refcount)) {
5452 		racct_destroy(&prr->prr_racct);
5453 		LIST_REMOVE(prr, prr_next);
5454 		free(prr, M_PRISON_RACCT);
5455 	}
5456 }
5457 
5458 void
prison_racct_free(struct prison_racct * prr)5459 prison_racct_free(struct prison_racct *prr)
5460 {
5461 
5462 	ASSERT_RACCT_ENABLED();
5463 	sx_assert(&allprison_lock, SA_UNLOCKED);
5464 
5465 	if (refcount_release_if_not_last(&prr->prr_refcount))
5466 		return;
5467 
5468 	sx_xlock(&allprison_lock);
5469 	prison_racct_free_locked(prr);
5470 	sx_xunlock(&allprison_lock);
5471 }
5472 
5473 static void
prison_racct_attach(struct prison * pr)5474 prison_racct_attach(struct prison *pr)
5475 {
5476 	struct prison_racct *prr;
5477 
5478 	ASSERT_RACCT_ENABLED();
5479 	sx_assert(&allprison_lock, SA_XLOCKED);
5480 
5481 	prr = prison_racct_find_locked(pr->pr_name);
5482 	KASSERT(prr != NULL, ("cannot find prison_racct"));
5483 
5484 	pr->pr_prison_racct = prr;
5485 }
5486 
5487 /*
5488  * Handle jail renaming.  From the racct point of view, renaming means
5489  * moving from one prison_racct to another.
5490  */
5491 static void
prison_racct_modify(struct prison * pr)5492 prison_racct_modify(struct prison *pr)
5493 {
5494 #ifdef RCTL
5495 	struct proc *p;
5496 	struct ucred *cred;
5497 #endif
5498 	struct prison_racct *oldprr;
5499 
5500 	ASSERT_RACCT_ENABLED();
5501 
5502 	sx_slock(&allproc_lock);
5503 	sx_xlock(&allprison_lock);
5504 
5505 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
5506 		sx_xunlock(&allprison_lock);
5507 		sx_sunlock(&allproc_lock);
5508 		return;
5509 	}
5510 
5511 	oldprr = pr->pr_prison_racct;
5512 	pr->pr_prison_racct = NULL;
5513 
5514 	prison_racct_attach(pr);
5515 
5516 	/*
5517 	 * Move resource utilisation records.
5518 	 */
5519 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
5520 
5521 #ifdef RCTL
5522 	/*
5523 	 * Force rctl to reattach rules to processes.
5524 	 */
5525 	FOREACH_PROC_IN_SYSTEM(p) {
5526 		PROC_LOCK(p);
5527 		cred = crhold(p->p_ucred);
5528 		PROC_UNLOCK(p);
5529 		rctl_proc_ucred_changed(p, cred);
5530 		crfree(cred);
5531 	}
5532 #endif
5533 
5534 	sx_sunlock(&allproc_lock);
5535 	prison_racct_free_locked(oldprr);
5536 	sx_xunlock(&allprison_lock);
5537 }
5538 
5539 static void
prison_racct_detach(struct prison * pr)5540 prison_racct_detach(struct prison *pr)
5541 {
5542 
5543 	ASSERT_RACCT_ENABLED();
5544 	sx_assert(&allprison_lock, SA_UNLOCKED);
5545 
5546 	if (pr->pr_prison_racct == NULL)
5547 		return;
5548 	prison_racct_free(pr->pr_prison_racct);
5549 	pr->pr_prison_racct = NULL;
5550 }
5551 #endif /* RACCT */
5552 
5553 /*
5554  * Submit a knote for a prison, locking if necessary.
5555  */
5556 static void
prison_knote(struct prison * pr,long hint)5557 prison_knote(struct prison *pr, long hint)
5558 {
5559 	int locked;
5560 
5561 	locked = mtx_owned(&pr->pr_mtx);
5562 	if (!locked)
5563 		mtx_lock(&pr->pr_mtx);
5564 	KNOTE_LOCKED(pr->pr_klist, hint);
5565 	jaildesc_knote(pr, hint);
5566 	if (!locked)
5567 		mtx_unlock(&pr->pr_mtx);
5568 }
5569 
5570 #ifdef DDB
5571 
5572 static void
db_show_prison(struct prison * pr)5573 db_show_prison(struct prison *pr)
5574 {
5575 	struct bool_flags *bf;
5576 	struct jailsys_flags *jsf;
5577 #if defined(INET) || defined(INET6)
5578 	int ii;
5579 	struct prison_ip *pip;
5580 #endif
5581 	unsigned f;
5582 #ifdef INET
5583 	char ip4buf[INET_ADDRSTRLEN];
5584 #endif
5585 #ifdef INET6
5586 	char ip6buf[INET6_ADDRSTRLEN];
5587 #endif
5588 
5589 	db_printf("prison %p:\n", pr);
5590 	db_printf(" jid             = %d\n", pr->pr_id);
5591 	db_printf(" name            = %s\n", pr->pr_name);
5592 	db_printf(" parent          = %p\n", pr->pr_parent);
5593 	db_printf(" ref             = %d\n", pr->pr_ref);
5594 	db_printf(" uref            = %d\n", pr->pr_uref);
5595 	db_printf(" state           = %s\n",
5596 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5597 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
5598 	    "invalid");
5599 	db_printf(" path            = %s\n", pr->pr_path);
5600 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
5601 	    ? pr->pr_cpuset->cs_id : -1);
5602 #ifdef VIMAGE
5603 	db_printf(" vnet            = %p\n", pr->pr_vnet);
5604 #endif
5605 	db_printf(" root            = %p\n", pr->pr_root);
5606 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
5607 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
5608 	db_printf(" children.max    = %d\n", pr->pr_childmax);
5609 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
5610 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
5611 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
5612 	db_printf(" flags           = 0x%x", pr->pr_flags);
5613 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5614 		if (pr->pr_flags & bf->flag)
5615 			db_printf(" %s", bf->name);
5616 	for (jsf = pr_flag_jailsys;
5617 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5618 	     jsf++) {
5619 		f = pr->pr_flags & (jsf->disable | jsf->new);
5620 		db_printf(" %-16s= %s\n", jsf->name,
5621 		    (f != 0 && f == jsf->disable) ? "disable"
5622 		    : (f == jsf->new) ? "new"
5623 		    : "inherit");
5624 	}
5625 	db_printf(" allow           = 0x%x", pr->pr_allow);
5626 	for (bf = pr_flag_allow;
5627 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5628 		atomic_load_int(&bf->flag) != 0;
5629 	     bf++)
5630 		if (pr->pr_allow & bf->flag)
5631 			db_printf(" %s", bf->name);
5632 	db_printf("\n");
5633 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
5634 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
5635 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
5636 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
5637 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
5638 #ifdef INET
5639 	if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5640 		db_printf(" ip4s            = %d\n", pip->ips);
5641 		for (ii = 0; ii < pip->ips; ii++)
5642 			db_printf(" %s %s\n",
5643 			    ii == 0 ? "ip4.addr        =" : "                 ",
5644 			    inet_ntoa_r(
5645 			    *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5646 			    ip4buf));
5647 	}
5648 #endif
5649 #ifdef INET6
5650 	if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5651 		db_printf(" ip6s            = %d\n", pip->ips);
5652 		for (ii = 0; ii < pip->ips; ii++)
5653 			db_printf(" %s %s\n",
5654 			    ii == 0 ? "ip6.addr        =" : "                 ",
5655 			    ip6_sprintf(ip6buf,
5656 			    (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5657 	}
5658 #endif
5659 }
5660 
DB_SHOW_COMMAND(prison,db_show_prison_command)5661 DB_SHOW_COMMAND(prison, db_show_prison_command)
5662 {
5663 	struct prison *pr;
5664 
5665 	if (!have_addr) {
5666 		/*
5667 		 * Show all prisons in the list, and prison0 which is not
5668 		 * listed.
5669 		 */
5670 		db_show_prison(&prison0);
5671 		if (!db_pager_quit) {
5672 			TAILQ_FOREACH(pr, &allprison, pr_list) {
5673 				db_show_prison(pr);
5674 				if (db_pager_quit)
5675 					break;
5676 			}
5677 		}
5678 		return;
5679 	}
5680 
5681 	if (addr == 0)
5682 		pr = &prison0;
5683 	else {
5684 		/* Look for a prison with the ID and with references. */
5685 		TAILQ_FOREACH(pr, &allprison, pr_list)
5686 			if (pr->pr_id == addr && pr->pr_ref > 0)
5687 				break;
5688 		if (pr == NULL)
5689 			/* Look again, without requiring a reference. */
5690 			TAILQ_FOREACH(pr, &allprison, pr_list)
5691 				if (pr->pr_id == addr)
5692 					break;
5693 		if (pr == NULL)
5694 			/* Assume address points to a valid prison. */
5695 			pr = (struct prison *)addr;
5696 	}
5697 	db_show_prison(pr);
5698 }
5699 
5700 #endif /* DDB */
5701