xref: /freebsd/sys/kern/kern_jail.c (revision 246d7e9fc23928be22db38220f5439f5cdee5264)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_ddb.h"
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_nfs.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/file.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/epoch.h>
49 #include <sys/event.h>
50 #include <sys/taskqueue.h>
51 #include <sys/fcntl.h>
52 #include <sys/jail.h>
53 #include <sys/jaildesc.h>
54 #include <sys/linker.h>
55 #include <sys/lock.h>
56 #include <sys/mman.h>
57 #include <sys/mutex.h>
58 #include <sys/racct.h>
59 #include <sys/rctl.h>
60 #include <sys/refcount.h>
61 #include <sys/sx.h>
62 #include <sys/sysent.h>
63 #include <sys/namei.h>
64 #include <sys/mount.h>
65 #include <sys/queue.h>
66 #include <sys/socket.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/uuid.h>
70 #include <sys/vnode.h>
71 
72 #include <net/if.h>
73 #include <net/vnet.h>
74 
75 #include <netinet/in.h>
76 
77 #ifdef DDB
78 #include <ddb/ddb.h>
79 #endif /* DDB */
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define	PRISON0_HOSTUUID_MODULE	"hostuuid"
84 
85 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
86 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
87 
88 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
89 #ifdef INET
90 #ifdef INET6
91 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
92 #else
93 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
94 #endif
95 #else /* !INET */
96 #ifdef INET6
97 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
98 #else
99 #define	_PR_IP_SADDRSEL	0
100 #endif
101 #endif
102 
103 /* prison0 describes what is "real" about the system. */
104 struct prison prison0 = {
105 	.pr_id		= 0,
106 	.pr_name	= "0",
107 	.pr_ref		= 1,
108 	.pr_uref	= 1,
109 	.pr_path	= "/",
110 	.pr_securelevel	= -1,
111 	.pr_devfs_rsnum = 0,
112 	.pr_state	= PRISON_STATE_ALIVE,
113 	.pr_childmax	= JAIL_MAX,
114 	.pr_hostuuid	= DEFAULT_HOSTUUID,
115 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
116 #ifdef VIMAGE
117 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
118 #else
119 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
120 #endif
121 	.pr_allow	= PR_ALLOW_PRISON0,
122 };
123 _Static_assert((PR_ALLOW_PRISON0 & ~PR_ALLOW_ALL_STATIC) == 0,
124     "Bits enabled in PR_ALLOW_PRISON0 that are not statically reserved");
125 
126 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
127 
128 struct bool_flags {
129 	const char	*name;
130 	const char	*noname;
131 	volatile u_int	 flag;
132 };
133 struct jailsys_flags {
134 	const char	*name;
135 	unsigned	 disable;
136 	unsigned	 new;
137 };
138 
139 /*
140  * Handle jail teardown in a dedicated thread to avoid deadlocks from
141  * vnet_destroy().
142  */
143 TASKQUEUE_DEFINE_THREAD(jail_remove);
144 
145 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
146 struct	sx allprison_lock;
147 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
148 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
149 LIST_HEAD(, prison_racct) allprison_racct;
150 int	lastprid = 0;
151 int	lastdeadid = 0;
152 
153 static int get_next_prid(struct prison **insprp);
154 static int get_next_deadid(struct prison **insprp);
155 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
156 static void prison_complete(void *context, int pending);
157 static void prison_deref(struct prison *pr, int flags);
158 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
159 static int prison_lock_xlock(struct prison *pr, int flags);
160 static void prison_cleanup_locked(struct prison *pr);
161 static void prison_cleanup_unlocked(struct prison *pr);
162 static void prison_free_not_last(struct prison *pr);
163 static void prison_proc_free_not_last(struct prison *pr);
164 static void prison_proc_relink(struct prison *opr, struct prison *npr,
165     struct proc *p);
166 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
167     int enable);
168 static char *prison_path(struct prison *pr1, struct prison *pr2);
169 #ifdef RACCT
170 static void prison_racct_attach(struct prison *pr);
171 static void prison_racct_modify(struct prison *pr);
172 static void prison_racct_detach(struct prison *pr);
173 #endif
174 static void prison_knote(struct prison *pr, long hint);
175 
176 /* Flags for prison_deref */
177 #define	PD_DEREF	0x01	/* Decrement pr_ref */
178 #define	PD_DEUREF	0x02	/* Decrement pr_uref */
179 #define	PD_KILL		0x04	/* Remove jail, kill processes, etc */
180 #define	PD_LOCKED	0x10	/* pr_mtx is held */
181 #define	PD_LIST_SLOCKED	0x20	/* allprison_lock is held shared */
182 #define	PD_LIST_XLOCKED	0x40	/* allprison_lock is held exclusive */
183 #define PD_OP_FLAGS	0x07	/* Operation flags */
184 #define PD_LOCK_FLAGS	0x70	/* Lock status flags */
185 
186 /*
187  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
188  * as we cannot figure out the size of a sparse array, or an array without a
189  * terminating entry.
190  */
191 static struct bool_flags pr_flag_bool[] = {
192 	{"persist", "nopersist", PR_PERSIST},
193 #ifdef INET
194 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
195 #endif
196 #ifdef INET6
197 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
198 #endif
199 };
200 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
201 
202 static struct jailsys_flags pr_flag_jailsys[] = {
203 	{"host", 0, PR_HOST},
204 #ifdef VIMAGE
205 	{"vnet", 0, PR_VNET},
206 #endif
207 #ifdef INET
208 	{"ip4", PR_IP4_USER, PR_IP4_USER},
209 #endif
210 #ifdef INET6
211 	{"ip6", PR_IP6_USER, PR_IP6_USER},
212 #endif
213 };
214 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
215 
216 /*
217  * Make this array full-size so dynamic parameters can be added.
218  * It is protected by prison0.mtx, but lockless reading is allowed
219  * with an atomic check of the flag values.
220  */
221 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
222 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
223 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
224 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
225 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
226 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
227 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
228 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
229 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
230 	{"allow.reserved_ports", "allow.noreserved_ports",
231 	 PR_ALLOW_RESERVED_PORTS},
232 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
233 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
234 	 PR_ALLOW_UNPRIV_DEBUG},
235 	{"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
236 #ifdef VIMAGE
237 	{"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
238 #endif
239 	{"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
240 	{"allow.adjtime", "allow.noadjtime", PR_ALLOW_ADJTIME},
241 	{"allow.settime", "allow.nosettime", PR_ALLOW_SETTIME},
242 	{"allow.routing", "allow.norouting", PR_ALLOW_ROUTING},
243 	{"allow.unprivileged_parent_tampering",
244 	    "allow.nounprivileged_parent_tampering",
245 	    PR_ALLOW_UNPRIV_PARENT_TAMPER},
246 #ifdef AUDIT
247 	{"allow.setaudit", "allow.nosetaudit", PR_ALLOW_SETAUDIT},
248 #endif
249 };
250 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
251 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
252 
253 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
254 					 PR_ALLOW_RESERVED_PORTS | \
255 					 PR_ALLOW_UNPRIV_DEBUG | \
256 					 PR_ALLOW_SUSER)
257 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
258 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
259 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
260 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
261 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
262 #if defined(INET) || defined(INET6)
263 static unsigned jail_max_af_ips = 255;
264 #endif
265 
266 /*
267  * Initialize the parts of prison0 that can't be static-initialized with
268  * constants.  This is called from proc0_init() after creating thread0 cpuset.
269  */
270 void
prison0_init(void)271 prison0_init(void)
272 {
273 	uint8_t *file, *data;
274 	size_t size;
275 	char buf[sizeof(prison0.pr_hostuuid)];
276 	bool valid;
277 
278 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
279 	prison0.pr_osreldate = osreldate;
280 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
281 
282 	/* If we have a preloaded hostuuid, use it. */
283 	file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
284 	if (file != NULL) {
285 		data = preload_fetch_addr(file);
286 		size = preload_fetch_size(file);
287 		if (data != NULL) {
288 			/*
289 			 * The preloaded data may include trailing whitespace, almost
290 			 * certainly a newline; skip over any whitespace or
291 			 * non-printable characters to be safe.
292 			 */
293 			while (size > 0 && data[size - 1] <= 0x20) {
294 				size--;
295 			}
296 
297 			valid = false;
298 
299 			/*
300 			 * Not NUL-terminated when passed from loader, but
301 			 * validate_uuid requires that due to using sscanf (as
302 			 * does the subsequent strlcpy, since it still reads
303 			 * past the given size to return the true length);
304 			 * bounce to a temporary buffer to fix.
305 			 */
306 			if (size >= sizeof(buf))
307 				goto done;
308 
309 			memcpy(buf, data, size);
310 			buf[size] = '\0';
311 
312 			if (validate_uuid(buf, size, NULL, 0) != 0)
313 				goto done;
314 
315 			valid = true;
316 			(void)strlcpy(prison0.pr_hostuuid, buf,
317 			    sizeof(prison0.pr_hostuuid));
318 
319 done:
320 			if (bootverbose && !valid) {
321 				printf("hostuuid: preload data malformed: '%.*s'\n",
322 				    (int)size, data);
323 			}
324 		}
325 	}
326 	if (bootverbose)
327 		printf("hostuuid: using %s\n", prison0.pr_hostuuid);
328 }
329 
330 /*
331  * struct jail_args {
332  *	struct jail *jail;
333  * };
334  */
335 int
sys_jail(struct thread * td,struct jail_args * uap)336 sys_jail(struct thread *td, struct jail_args *uap)
337 {
338 	uint32_t version;
339 	int error;
340 	struct jail j;
341 
342 	error = copyin(uap->jail, &version, sizeof(uint32_t));
343 	if (error)
344 		return (error);
345 
346 	switch (version) {
347 	case 0:
348 	{
349 		struct jail_v0 j0;
350 
351 		/* FreeBSD single IPv4 jails. */
352 		bzero(&j, sizeof(struct jail));
353 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
354 		if (error)
355 			return (error);
356 		j.version = j0.version;
357 		j.path = j0.path;
358 		j.hostname = j0.hostname;
359 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
360 		break;
361 	}
362 
363 	case 1:
364 		/*
365 		 * Version 1 was used by multi-IPv4 jail implementations
366 		 * that never made it into the official kernel.
367 		 */
368 		return (EINVAL);
369 
370 	case 2:	/* JAIL_API_VERSION */
371 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
372 		error = copyin(uap->jail, &j, sizeof(struct jail));
373 		if (error)
374 			return (error);
375 		break;
376 
377 	default:
378 		/* Sci-Fi jails are not supported, sorry. */
379 		return (EINVAL);
380 	}
381 	return (kern_jail(td, &j));
382 }
383 
384 int
kern_jail(struct thread * td,struct jail * j)385 kern_jail(struct thread *td, struct jail *j)
386 {
387 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
388 #ifdef INET
389 			    + 1
390 #endif
391 #ifdef INET6
392 			    + 1
393 #endif
394 			    )];
395 	struct uio opt;
396 	char *u_path, *u_hostname, *u_name;
397 	struct bool_flags *bf;
398 #ifdef INET
399 	uint32_t ip4s;
400 	struct in_addr *u_ip4;
401 #endif
402 #ifdef INET6
403 	struct in6_addr *u_ip6;
404 #endif
405 	size_t tmplen;
406 	int error, enforce_statfs;
407 
408 	bzero(&optiov, sizeof(optiov));
409 	opt.uio_iov = optiov;
410 	opt.uio_iovcnt = 0;
411 	opt.uio_offset = -1;
412 	opt.uio_resid = -1;
413 	opt.uio_segflg = UIO_SYSSPACE;
414 	opt.uio_rw = UIO_READ;
415 	opt.uio_td = td;
416 
417 	/* Set permissions for top-level jails from sysctls. */
418 	if (!jailed(td->td_ucred)) {
419 		for (bf = pr_flag_allow;
420 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
421 			atomic_load_int(&bf->flag) != 0;
422 		     bf++) {
423 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
424 			    (jail_default_allow & bf->flag)
425 			    ? bf->name : bf->noname);
426 			optiov[opt.uio_iovcnt].iov_len =
427 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
428 			opt.uio_iovcnt += 2;
429 		}
430 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
431 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
432 		opt.uio_iovcnt++;
433 		enforce_statfs = jail_default_enforce_statfs;
434 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
435 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
436 		opt.uio_iovcnt++;
437 	}
438 
439 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
440 #ifdef INET
441 	ip4s = (j->version == 0) ? 1 : j->ip4s;
442 	if (ip4s > jail_max_af_ips)
443 		return (EINVAL);
444 	tmplen += ip4s * sizeof(struct in_addr);
445 #else
446 	if (j->ip4s > 0)
447 		return (EINVAL);
448 #endif
449 #ifdef INET6
450 	if (j->ip6s > jail_max_af_ips)
451 		return (EINVAL);
452 	tmplen += j->ip6s * sizeof(struct in6_addr);
453 #else
454 	if (j->ip6s > 0)
455 		return (EINVAL);
456 #endif
457 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
458 	u_hostname = u_path + MAXPATHLEN;
459 	u_name = u_hostname + MAXHOSTNAMELEN;
460 #ifdef INET
461 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
462 #endif
463 #ifdef INET6
464 #ifdef INET
465 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
466 #else
467 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
468 #endif
469 #endif
470 	optiov[opt.uio_iovcnt].iov_base = "path";
471 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
472 	opt.uio_iovcnt++;
473 	optiov[opt.uio_iovcnt].iov_base = u_path;
474 	error = copyinstr(j->path, u_path, MAXPATHLEN,
475 	    &optiov[opt.uio_iovcnt].iov_len);
476 	if (error) {
477 		free(u_path, M_TEMP);
478 		return (error);
479 	}
480 	opt.uio_iovcnt++;
481 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
482 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
483 	opt.uio_iovcnt++;
484 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
485 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
486 	    &optiov[opt.uio_iovcnt].iov_len);
487 	if (error) {
488 		free(u_path, M_TEMP);
489 		return (error);
490 	}
491 	opt.uio_iovcnt++;
492 	if (j->jailname != NULL) {
493 		optiov[opt.uio_iovcnt].iov_base = "name";
494 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
495 		opt.uio_iovcnt++;
496 		optiov[opt.uio_iovcnt].iov_base = u_name;
497 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
498 		    &optiov[opt.uio_iovcnt].iov_len);
499 		if (error) {
500 			free(u_path, M_TEMP);
501 			return (error);
502 		}
503 		opt.uio_iovcnt++;
504 	}
505 #ifdef INET
506 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
507 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
508 	opt.uio_iovcnt++;
509 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
510 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
511 	if (j->version == 0)
512 		u_ip4->s_addr = j->ip4s;
513 	else {
514 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
515 		if (error) {
516 			free(u_path, M_TEMP);
517 			return (error);
518 		}
519 	}
520 	opt.uio_iovcnt++;
521 #endif
522 #ifdef INET6
523 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
524 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
525 	opt.uio_iovcnt++;
526 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
527 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
528 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
529 	if (error) {
530 		free(u_path, M_TEMP);
531 		return (error);
532 	}
533 	opt.uio_iovcnt++;
534 #endif
535 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
536 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
537 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
538 	free(u_path, M_TEMP);
539 	return (error);
540 }
541 
542 /*
543  * struct jail_set_args {
544  *	struct iovec *iovp;
545  *	unsigned int iovcnt;
546  *	int flags;
547  * };
548  */
549 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)550 sys_jail_set(struct thread *td, struct jail_set_args *uap)
551 {
552 	struct uio *auio;
553 	int error;
554 
555 	/* Check that we have an even number of iovecs. */
556 	if (uap->iovcnt & 1)
557 		return (EINVAL);
558 
559 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
560 	if (error)
561 		return (error);
562 	error = kern_jail_set(td, auio, uap->flags);
563 	freeuio(auio);
564 	return (error);
565 }
566 
567 #if defined(INET) || defined(INET6)
568 typedef int prison_addr_cmp_t(const void *, const void *);
569 typedef bool prison_addr_valid_t(const void *);
570 static const struct pr_family {
571 	size_t			size;
572 	prison_addr_cmp_t	*cmp;
573 	prison_addr_valid_t	*valid;
574 	int			ip_flag;
575 } pr_families[PR_FAMILY_MAX] = {
576 #ifdef INET
577 	[PR_INET] = {
578 		.size = sizeof(struct in_addr),
579 		.cmp = prison_qcmp_v4,
580 		.valid = prison_valid_v4,
581 		.ip_flag = PR_IP4_USER,
582 	 },
583 #endif
584 #ifdef INET6
585 	[PR_INET6] = {
586 		.size = sizeof(struct in6_addr),
587 		.cmp = prison_qcmp_v6,
588 		.valid = prison_valid_v6,
589 		.ip_flag = PR_IP6_USER,
590 	},
591 #endif
592 };
593 
594 /*
595  * Network address lists (pr_addrs) allocation for jails.  The addresses
596  * are accessed locklessly by the network stack, thus need to be protected by
597  * the network epoch.
598  */
599 struct prison_ip {
600 	struct epoch_context ctx;
601 	uint32_t	ips;
602 #ifdef FUTURE_C
603 	/*
604 	 * XXX Variable-length automatic arrays in union may be
605 	 * supported in future C.
606 	 */
607 	union {
608 		char pr_ip[];
609 		struct in_addr pr_ip4[];
610 		struct in6_addr pr_ip6[];
611 	};
612 #else /* No future C :( */
613 	char pr_ip[];
614 #endif
615 };
616 
617 static char *
PR_IP(struct prison_ip * pip,const pr_family_t af,int idx)618 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
619 {
620 	MPASS(pip);
621 	MPASS(af < PR_FAMILY_MAX);
622 	MPASS(idx >= 0 && idx < pip->ips);
623 
624 	return (pip->pr_ip + pr_families[af].size * idx);
625 }
626 
627 static struct prison_ip *
prison_ip_alloc(const pr_family_t af,uint32_t cnt,int flags)628 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
629 {
630 	struct prison_ip *pip;
631 
632 	pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
633 	    M_PRISON, flags);
634 	if (pip != NULL)
635 		pip->ips = cnt;
636 	return (pip);
637 }
638 
639 /*
640  * Allocate and copyin user supplied address list, sorting and validating.
641  * kern_jail_set() helper.
642  */
643 static struct prison_ip *
prison_ip_copyin(const pr_family_t af,void * op,uint32_t cnt)644 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
645 {
646 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
647 	const size_t size = pr_families[af].size;
648 	struct prison_ip *pip;
649 
650 	pip = prison_ip_alloc(af, cnt, M_WAITOK);
651 	bcopy(op, pip->pr_ip, cnt * size);
652 	/*
653 	 * IP addresses are all sorted but ip[0] to preserve
654 	 * the primary IP address as given from userland.
655 	 * This special IP is used for unbound outgoing
656 	 * connections as well for "loopback" traffic in case
657 	 * source address selection cannot find any more fitting
658 	 * address to connect from.
659 	 */
660 	if (cnt > 1)
661 		qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
662 	/*
663 	 * Check for duplicate addresses and do some simple
664 	 * zero and broadcast checks. If users give other bogus
665 	 * addresses it is their problem.
666 	 */
667 	for (int i = 0; i < cnt; i++) {
668 		if (!pr_families[af].valid(PR_IP(pip, af, i))) {
669 			free(pip, M_PRISON);
670 			return (NULL);
671 		}
672 		if (i + 1 < cnt &&
673 		    (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
674 		     cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
675 			free(pip, M_PRISON);
676 			return (NULL);
677 		}
678 	}
679 
680 	return (pip);
681 }
682 
683 /*
684  * Allocate and dup parent prison address list.
685  * kern_jail_set() helper.
686  */
687 static void
prison_ip_dup(struct prison * ppr,struct prison * pr,const pr_family_t af)688 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
689 {
690 	const struct prison_ip *ppip = ppr->pr_addrs[af];
691 	struct prison_ip *pip;
692 
693 	if (ppip != NULL) {
694 		pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
695 		bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
696 		pr->pr_addrs[af] = pip;
697 	}
698 }
699 
700 /*
701  * Make sure the new set of IP addresses is a subset of the parent's list.
702  * Don't worry about the parent being unlocked, as any setting is done with
703  * allprison_lock held.
704  * kern_jail_set() helper.
705  */
706 static bool
prison_ip_parent_match(struct prison_ip * ppip,struct prison_ip * pip,const pr_family_t af)707 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
708     const pr_family_t af)
709 {
710 	prison_addr_cmp_t *const cmp = pr_families[af].cmp;
711 	int i, j;
712 
713 	if (ppip == NULL)
714 		return (false);
715 
716 	for (i = 0; i < ppip->ips; i++)
717 		if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
718 			break;
719 
720 	if (i == ppip->ips)
721 		/* Main address not present in parent. */
722 		return (false);
723 
724 	if (pip->ips > 1) {
725 		for (i = j = 1; i < pip->ips; i++) {
726 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
727 				/* Equals to parent primary address. */
728 				continue;
729 			for (; j < ppip->ips; j++)
730 				if (cmp(PR_IP(pip, af, i),
731 				    PR_IP(ppip, af, j)) == 0)
732 					break;
733 			if (j == ppip->ips)
734 				break;
735 		}
736 		if (j == ppip->ips)
737 			/* Address not present in parent. */
738 			return (false);
739 	}
740 	return (true);
741 }
742 
743 /*
744  * Check for conflicting IP addresses.  We permit them if there is no more
745  * than one IP on each jail.  If there is a duplicate on a jail with more
746  * than one IP stop checking and return error.
747  * kern_jail_set() helper.
748  */
749 static bool
prison_ip_conflict_check(const struct prison * ppr,const struct prison * pr,struct prison_ip * pip,pr_family_t af)750 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
751     struct prison_ip *pip, pr_family_t af)
752 {
753 	const struct prison *tppr, *tpr;
754 	int descend;
755 
756 #ifdef VIMAGE
757 	for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
758 		if (tppr->pr_flags & PR_VNET)
759 			break;
760 #else
761 	tppr = &prison0;
762 #endif
763 	FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
764 		if (tpr == pr ||
765 #ifdef VIMAGE
766 		    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
767 #endif
768 		    !prison_isalive(tpr)) {
769 			descend = 0;
770 			continue;
771 		}
772 		if (!(tpr->pr_flags & pr_families[af].ip_flag))
773 			continue;
774 		descend = 0;
775 		if (tpr->pr_addrs[af] == NULL ||
776 		    (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
777 			continue;
778 		for (int i = 0; i < pip->ips; i++)
779 			if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
780 				return (false);
781 	}
782 
783 	return (true);
784 }
785 
786 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
787     "prison must start with epoch context");
788 static void
prison_ip_free_deferred(epoch_context_t ctx)789 prison_ip_free_deferred(epoch_context_t ctx)
790 {
791 
792 	free(ctx, M_PRISON);
793 }
794 
795 static void
prison_ip_free(struct prison_ip * pip)796 prison_ip_free(struct prison_ip *pip)
797 {
798 
799 	if (pip != NULL)
800 		NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
801 }
802 
803 static void
prison_ip_set(struct prison * pr,const pr_family_t af,struct prison_ip * new)804 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
805 {
806 	struct prison_ip **mem, *old;
807 
808 	mtx_assert(&pr->pr_mtx, MA_OWNED);
809 
810 	mem = &pr->pr_addrs[af];
811 
812 	old = *mem;
813 	atomic_store_ptr(mem, new);
814 	prison_ip_free(old);
815 }
816 
817 /*
818  * Restrict a prison's IP address list with its parent's, possibly replacing
819  * it.  Return true if succeed, otherwise should redo.
820  * kern_jail_set() helper.
821  */
822 static bool
prison_ip_restrict(struct prison * pr,const pr_family_t af,struct prison_ip ** newp)823 prison_ip_restrict(struct prison *pr, const pr_family_t af,
824     struct prison_ip **newp)
825 {
826 	struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
827 	struct prison_ip *pip = pr->pr_addrs[af];
828 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
829 	const size_t size = pr_families[af].size;
830 	struct prison_ip *new = newp != NULL ? *newp : NULL;
831 	uint32_t ips;
832 
833 	mtx_assert(&pr->pr_mtx, MA_OWNED);
834 
835 	/*
836 	 * Due to epoch-synchronized access to the IP address lists we always
837 	 * allocate a new list even if the old one has enough space.  We could
838 	 * atomically update an IPv4 address inside a list, but that would
839 	 * screw up sorting, and in case of IPv6 we can't even atomically write
840 	 * one.
841 	 */
842 	if (ppip == NULL) {
843 		if (pip != NULL)
844 			prison_ip_set(pr, af, NULL);
845 		return (true);
846 	}
847 
848 	if (!(pr->pr_flags & pr_families[af].ip_flag)) {
849 		if (new == NULL) {
850 			new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
851 			if (new == NULL)
852 				return (false); /* Redo */
853 		}
854 		/* This has no user settings, so just copy the parent's list. */
855 		MPASS(new->ips == ppip->ips);
856 		bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
857 		prison_ip_set(pr, af, new);
858 		if (newp != NULL)
859 			*newp = NULL; /* Used */
860 	} else if (pip != NULL) {
861 		/* Remove addresses that aren't in the parent. */
862 		int i;
863 
864 		i = 0; /* index in pip */
865 		ips = 0; /* index in new */
866 
867 		if (new == NULL) {
868 			new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
869 			if (new == NULL)
870 				return (false); /* Redo */
871 		}
872 
873 		for (int pi = 0; pi < ppip->ips; pi++)
874 			if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
875 				/* Found our primary address in parent. */
876 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
877 				    size);
878 				i++;
879 				ips++;
880 				break;
881 			}
882 		for (int pi = 1; i < pip->ips; ) {
883 			/* Check against primary, which is unsorted. */
884 			if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
885 				/* Matches parent's primary address. */
886 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
887 				    size);
888 				i++;
889 				ips++;
890 				continue;
891 			}
892 			/* The rest are sorted. */
893 			switch (pi >= ppip->ips ? -1 :
894 				cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
895 			case -1:
896 				i++;
897 				break;
898 			case 0:
899 				bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
900 				    size);
901 				i++;
902 				pi++;
903 				ips++;
904 				break;
905 			case 1:
906 				pi++;
907 				break;
908 			}
909 		}
910 		if (ips == 0) {
911 			if (newp == NULL || *newp == NULL)
912 				prison_ip_free(new);
913 			new = NULL;
914 		} else {
915 			/* Shrink to real size */
916 			KASSERT((new->ips >= ips),
917 			    ("Out-of-bounds write to prison_ip %p", new));
918 			new->ips = ips;
919 		}
920 		prison_ip_set(pr, af, new);
921 		if (newp != NULL)
922 			*newp = NULL; /* Used */
923 	}
924 	return (true);
925 }
926 
927 /*
928  * Fast-path check if an address belongs to a prison.
929  */
930 int
prison_ip_check(const struct prison * pr,const pr_family_t af,const void * addr)931 prison_ip_check(const struct prison *pr, const pr_family_t af,
932     const void *addr)
933 {
934 	int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
935 	struct prison_ip *pip;
936 	int i, a, z, d;
937 
938 	MPASS(mtx_owned(&pr->pr_mtx) ||
939 	    in_epoch(net_epoch_preempt) ||
940 	    sx_xlocked(&allprison_lock));
941 
942 	pip = atomic_load_ptr(&pr->pr_addrs[af]);
943 	if (__predict_false(pip == NULL))
944 		return (EAFNOSUPPORT);
945 
946 	/* Check the primary IP. */
947 	if (cmp(PR_IP(pip, af, 0), addr) == 0)
948 		return (0);
949 
950 	/*
951 	 * All the other IPs are sorted so we can do a binary search.
952 	 */
953 	a = 0;
954 	z = pip->ips - 2;
955 	while (a <= z) {
956 		i = (a + z) / 2;
957 		d = cmp(PR_IP(pip, af, i + 1), addr);
958 		if (d > 0)
959 			z = i - 1;
960 		else if (d < 0)
961 			a = i + 1;
962 		else
963 			return (0);
964 	}
965 
966 	return (EADDRNOTAVAIL);
967 }
968 
969 /*
970  * Grab primary IP.  Historically required mutex, but nothing prevents
971  * us to support epoch-protected access.  Is it used in fast path?
972  * in{6}_jail.c helper
973  */
974 const void *
prison_ip_get0(const struct prison * pr,const pr_family_t af)975 prison_ip_get0(const struct prison *pr, const pr_family_t af)
976 {
977 	const struct prison_ip *pip = pr->pr_addrs[af];
978 
979 	mtx_assert(&pr->pr_mtx, MA_OWNED);
980 	MPASS(pip);
981 
982 	return (pip->pr_ip);
983 }
984 
985 u_int
prison_ip_cnt(const struct prison * pr,const pr_family_t af)986 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
987 {
988 
989 	return (pr->pr_addrs[af]->ips);
990 }
991 #endif	/* defined(INET) || defined(INET6) */
992 
993 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)994 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
995 {
996 	struct file *jfp_out;
997 	struct nameidata nd;
998 #ifdef INET
999 	struct prison_ip *ip4;
1000 #endif
1001 #ifdef INET6
1002 	struct prison_ip *ip6;
1003 #endif
1004 	struct vfsopt *opt;
1005 	struct vfsoptlist *opts;
1006 	struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
1007 	struct ucred *jdcred;
1008 	struct vnode *root;
1009 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
1010 	char *g_path, *osrelstr;
1011 	struct bool_flags *bf;
1012 	struct jailsys_flags *jsf;
1013 #if defined(INET) || defined(INET6)
1014 	void *op;
1015 #endif
1016 	unsigned long hid;
1017 	size_t namelen, onamelen, pnamelen;
1018 	int created, cuflags, descend, drflags, enforce;
1019 	int error, errmsg_len, errmsg_pos;
1020 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1021 	int deadid, jfd_in, jfd_out, jfd_pos, jid, jsys, len, level;
1022 	int childmax, osreldt, rsnum, slevel;
1023 #ifdef INET
1024 	int ip4s;
1025 	bool redo_ip4;
1026 #endif
1027 #ifdef INET6
1028 	int ip6s;
1029 	bool redo_ip6;
1030 #endif
1031 	bool maybe_changed;
1032 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1033 	uint64_t pr_allow_diff;
1034 	unsigned tallow;
1035 	char numbuf[12];
1036 
1037 	mypr = td->td_ucred->cr_prison;
1038 	if (((flags & (JAIL_CREATE | JAIL_AT_DESC)) == JAIL_CREATE) &&
1039 	    mypr->pr_childmax == 0)
1040 		return (EPERM);
1041 	if (flags & ~JAIL_SET_MASK)
1042 		return (EINVAL);
1043 	if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
1044 	    (JAIL_USE_DESC | JAIL_AT_DESC))
1045 		return (EINVAL);
1046 	prison_hold(mypr);
1047 
1048 #ifdef INET
1049 	ip4 = NULL;
1050 #endif
1051 #ifdef INET6
1052 	ip6 = NULL;
1053 #endif
1054 	g_path = NULL;
1055 	jfp_out = NULL;
1056 	jfd_out = -1;
1057 	/*
1058 	 * Check all the parameters before committing to anything.  Not all
1059 	 * errors can be caught early, but we may as well try.  Also, this
1060 	 * takes care of some expensive stuff (path lookup) before getting
1061 	 * the allprison lock.
1062 	 *
1063 	 * XXX Jails are not filesystems, and jail parameters are not mount
1064 	 *     options.  But it makes more sense to re-use the vfsopt code
1065 	 *     than duplicate it under a different name.
1066 	 */
1067 	error = vfs_buildopts(optuio, &opts);
1068 	if (error)
1069 		goto done_free;
1070 
1071 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1072 	if (!cuflags) {
1073 		error = EINVAL;
1074 		vfs_opterror(opts, "no valid operation (create or update)");
1075 		goto done_errmsg;
1076 	}
1077 
1078 	error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
1079 	if (error == ENOENT) {
1080 		if (flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1081 		    JAIL_OWN_DESC)) {
1082 			vfs_opterror(opts, "missing desc");
1083 			goto done_errmsg;
1084 		}
1085 		jfd_in = -1;
1086 	} else if (error != 0)
1087 		goto done_free;
1088 	else {
1089 		if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1090 		    JAIL_OWN_DESC))) {
1091 			vfs_opterror(opts, "unexpected desc");
1092 			goto done_errmsg;
1093 		}
1094 		if (flags & JAIL_AT_DESC) {
1095 			/*
1096 			 * Look up and create jails based on the
1097 			 * descriptor's prison.
1098 			 */
1099 			prison_free(mypr);
1100 			error = jaildesc_find(td, jfd_in, &mypr, NULL);
1101 			if (error != 0) {
1102 				vfs_opterror(opts, error == ENOENT ?
1103 				    "descriptor to dead jail" :
1104 				    "not a jail descriptor");
1105 				goto done_errmsg;
1106 			}
1107 			if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) {
1108 				error = EPERM;
1109 				goto done_free;
1110 			}
1111 		}
1112 		if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
1113 			/* Allocate a jail descriptor to return later. */
1114 			error = jaildesc_alloc(td, &jfp_out, &jfd_out,
1115 				flags & JAIL_OWN_DESC);
1116 			if (error)
1117 				goto done_free;
1118 		}
1119 	}
1120 
1121 	/*
1122 	 * Delay the permission check if using a jail descriptor,
1123 	 * until we get the descriptor's credentials.
1124 	 */
1125 	if (!(flags & JAIL_USE_DESC)) {
1126 		error = priv_check(td, PRIV_JAIL_SET);
1127 		if (error == 0 && (flags & JAIL_ATTACH))
1128 			error = priv_check(td, PRIV_JAIL_ATTACH);
1129 		if (error)
1130 			goto done_free;
1131 	}
1132 
1133 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1134 	if (error == ENOENT)
1135 		jid = 0;
1136 	else if (error != 0)
1137 		goto done_free;
1138 
1139 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1140 	if (error == ENOENT)
1141 		gotslevel = 0;
1142 	else if (error != 0)
1143 		goto done_free;
1144 	else
1145 		gotslevel = 1;
1146 
1147 	error =
1148 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1149 	if (error == ENOENT)
1150 		gotchildmax = 0;
1151 	else if (error != 0)
1152 		goto done_free;
1153 	else
1154 		gotchildmax = 1;
1155 
1156 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1157 	if (error == ENOENT)
1158 		gotenforce = 0;
1159 	else if (error != 0)
1160 		goto done_free;
1161 	else if (enforce < 0 || enforce > 2) {
1162 		error = EINVAL;
1163 		goto done_free;
1164 	} else
1165 		gotenforce = 1;
1166 
1167 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1168 	if (error == ENOENT)
1169 		gotrsnum = 0;
1170 	else if (error != 0)
1171 		goto done_free;
1172 	else
1173 		gotrsnum = 1;
1174 
1175 	pr_flags = ch_flags = 0;
1176 	for (bf = pr_flag_bool;
1177 	     bf < pr_flag_bool + nitems(pr_flag_bool);
1178 	     bf++) {
1179 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1180 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1181 	}
1182 	ch_flags |= pr_flags;
1183 	for (jsf = pr_flag_jailsys;
1184 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1185 	     jsf++) {
1186 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1187 		if (error == ENOENT)
1188 			continue;
1189 		if (error != 0)
1190 			goto done_free;
1191 		switch (jsys) {
1192 		case JAIL_SYS_DISABLE:
1193 			if (!jsf->disable) {
1194 				error = EINVAL;
1195 				goto done_free;
1196 			}
1197 			pr_flags |= jsf->disable;
1198 			break;
1199 		case JAIL_SYS_NEW:
1200 			pr_flags |= jsf->new;
1201 			break;
1202 		case JAIL_SYS_INHERIT:
1203 			break;
1204 		default:
1205 			error = EINVAL;
1206 			goto done_free;
1207 		}
1208 		ch_flags |= jsf->new | jsf->disable;
1209 	}
1210 	if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1211 	    && !(pr_flags & PR_PERSIST)) {
1212 		error = EINVAL;
1213 		vfs_opterror(opts, "new jail must persist or attach");
1214 		goto done_errmsg;
1215 	}
1216 #ifdef VIMAGE
1217 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1218 		error = EINVAL;
1219 		vfs_opterror(opts, "vnet cannot be changed after creation");
1220 		goto done_errmsg;
1221 	}
1222 #endif
1223 #ifdef INET
1224 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1225 		error = EINVAL;
1226 		vfs_opterror(opts, "ip4 cannot be changed after creation");
1227 		goto done_errmsg;
1228 	}
1229 #endif
1230 #ifdef INET6
1231 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1232 		error = EINVAL;
1233 		vfs_opterror(opts, "ip6 cannot be changed after creation");
1234 		goto done_errmsg;
1235 	}
1236 #endif
1237 
1238 	pr_allow = ch_allow = 0;
1239 	for (bf = pr_flag_allow;
1240 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
1241 		atomic_load_int(&bf->flag) != 0;
1242 	     bf++) {
1243 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1244 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1245 	}
1246 	ch_allow |= pr_allow;
1247 
1248 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1249 	if (error == ENOENT)
1250 		name = NULL;
1251 	else if (error != 0)
1252 		goto done_free;
1253 	else {
1254 		if (len == 0 || name[len - 1] != '\0') {
1255 			error = EINVAL;
1256 			goto done_free;
1257 		}
1258 		if (len > MAXHOSTNAMELEN) {
1259 			error = ENAMETOOLONG;
1260 			goto done_free;
1261 		}
1262 	}
1263 
1264 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1265 	if (error == ENOENT)
1266 		host = NULL;
1267 	else if (error != 0)
1268 		goto done_free;
1269 	else {
1270 		ch_flags |= PR_HOST;
1271 		pr_flags |= PR_HOST;
1272 		if (len == 0 || host[len - 1] != '\0') {
1273 			error = EINVAL;
1274 			goto done_free;
1275 		}
1276 		if (len > MAXHOSTNAMELEN) {
1277 			error = ENAMETOOLONG;
1278 			goto done_free;
1279 		}
1280 	}
1281 
1282 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1283 	if (error == ENOENT)
1284 		domain = NULL;
1285 	else if (error != 0)
1286 		goto done_free;
1287 	else {
1288 		ch_flags |= PR_HOST;
1289 		pr_flags |= PR_HOST;
1290 		if (len == 0 || domain[len - 1] != '\0') {
1291 			error = EINVAL;
1292 			goto done_free;
1293 		}
1294 		if (len > MAXHOSTNAMELEN) {
1295 			error = ENAMETOOLONG;
1296 			goto done_free;
1297 		}
1298 	}
1299 
1300 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1301 	if (error == ENOENT)
1302 		uuid = NULL;
1303 	else if (error != 0)
1304 		goto done_free;
1305 	else {
1306 		ch_flags |= PR_HOST;
1307 		pr_flags |= PR_HOST;
1308 		if (len == 0 || uuid[len - 1] != '\0') {
1309 			error = EINVAL;
1310 			goto done_free;
1311 		}
1312 		if (len > HOSTUUIDLEN) {
1313 			error = ENAMETOOLONG;
1314 			goto done_free;
1315 		}
1316 	}
1317 
1318 #ifdef COMPAT_FREEBSD32
1319 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1320 		uint32_t hid32;
1321 
1322 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1323 		hid = hid32;
1324 	} else
1325 #endif
1326 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1327 	if (error == ENOENT)
1328 		gothid = 0;
1329 	else if (error != 0)
1330 		goto done_free;
1331 	else {
1332 		gothid = 1;
1333 		ch_flags |= PR_HOST;
1334 		pr_flags |= PR_HOST;
1335 	}
1336 
1337 #ifdef INET
1338 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1339 	if (error == ENOENT)
1340 		ip4s = 0;
1341 	else if (error != 0)
1342 		goto done_free;
1343 	else if (ip4s & (sizeof(struct in_addr) - 1)) {
1344 		error = EINVAL;
1345 		goto done_free;
1346 	} else {
1347 		ch_flags |= PR_IP4_USER;
1348 		pr_flags |= PR_IP4_USER;
1349 		if (ip4s > 0) {
1350 			ip4s /= sizeof(struct in_addr);
1351 			if (ip4s > jail_max_af_ips) {
1352 				error = EINVAL;
1353 				vfs_opterror(opts, "too many IPv4 addresses");
1354 				goto done_errmsg;
1355 			}
1356 			ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1357 			if (ip4 == NULL) {
1358 				error = EINVAL;
1359 				goto done_free;
1360 			}
1361 		}
1362 	}
1363 #endif
1364 
1365 #ifdef INET6
1366 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1367 	if (error == ENOENT)
1368 		ip6s = 0;
1369 	else if (error != 0)
1370 		goto done_free;
1371 	else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1372 		error = EINVAL;
1373 		goto done_free;
1374 	} else {
1375 		ch_flags |= PR_IP6_USER;
1376 		pr_flags |= PR_IP6_USER;
1377 		if (ip6s > 0) {
1378 			ip6s /= sizeof(struct in6_addr);
1379 			if (ip6s > jail_max_af_ips) {
1380 				error = EINVAL;
1381 				vfs_opterror(opts, "too many IPv6 addresses");
1382 				goto done_errmsg;
1383 			}
1384 			ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1385 			if (ip6 == NULL) {
1386 				error = EINVAL;
1387 				goto done_free;
1388 			}
1389 		}
1390 	}
1391 #endif
1392 
1393 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1394 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1395 		error = EINVAL;
1396 		vfs_opterror(opts,
1397 		    "vnet jails cannot have IP address restrictions");
1398 		goto done_errmsg;
1399 	}
1400 #endif
1401 
1402 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1403 	if (error == ENOENT)
1404 		osrelstr = NULL;
1405 	else if (error != 0)
1406 		goto done_free;
1407 	else {
1408 		if (flags & JAIL_UPDATE) {
1409 			error = EINVAL;
1410 			vfs_opterror(opts,
1411 			    "osrelease cannot be changed after creation");
1412 			goto done_errmsg;
1413 		}
1414 		if (len == 0 || osrelstr[len - 1] != '\0') {
1415 			error = EINVAL;
1416 			goto done_free;
1417 		}
1418 		if (len >= OSRELEASELEN) {
1419 			error = ENAMETOOLONG;
1420 			vfs_opterror(opts,
1421 			    "osrelease string must be 1-%d bytes long",
1422 			    OSRELEASELEN - 1);
1423 			goto done_errmsg;
1424 		}
1425 	}
1426 
1427 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1428 	if (error == ENOENT)
1429 		osreldt = 0;
1430 	else if (error != 0)
1431 		goto done_free;
1432 	else {
1433 		if (flags & JAIL_UPDATE) {
1434 			error = EINVAL;
1435 			vfs_opterror(opts,
1436 			    "osreldate cannot be changed after creation");
1437 			goto done_errmsg;
1438 		}
1439 		if (osreldt == 0) {
1440 			error = EINVAL;
1441 			vfs_opterror(opts, "osreldate cannot be 0");
1442 			goto done_errmsg;
1443 		}
1444 	}
1445 
1446 	root = NULL;
1447 	error = vfs_getopt(opts, "path", (void **)&path, &len);
1448 	if (error == ENOENT)
1449 		path = NULL;
1450 	else if (error != 0)
1451 		goto done_free;
1452 	else {
1453 		if (flags & JAIL_UPDATE) {
1454 			error = EINVAL;
1455 			vfs_opterror(opts,
1456 			    "path cannot be changed after creation");
1457 			goto done_errmsg;
1458 		}
1459 		if (len == 0 || path[len - 1] != '\0') {
1460 			error = EINVAL;
1461 			goto done_free;
1462 		}
1463 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1464 		error = namei(&nd);
1465 		if (error)
1466 			goto done_free;
1467 		root = nd.ni_vp;
1468 		NDFREE_PNBUF(&nd);
1469 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1470 		strlcpy(g_path, path, MAXPATHLEN);
1471 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1472 		if (error == 0) {
1473 			path = g_path;
1474 		} else {
1475 			/* exit on other errors */
1476 			goto done_free;
1477 		}
1478 		if (root->v_type != VDIR) {
1479 			error = ENOTDIR;
1480 			vput(root);
1481 			goto done_free;
1482 		}
1483 		VOP_UNLOCK(root);
1484 	}
1485 
1486 	/*
1487 	 * Find the specified jail, or at least its parent.
1488 	 * This abuses the file error codes ENOENT and EEXIST.
1489 	 */
1490 	pr = NULL;
1491 	inspr = NULL;
1492 	deadpr = NULL;
1493 	maybe_changed = false;
1494 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1495 		namelc = strrchr(name, '.');
1496 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1497 		if (*p != '\0')
1498 			jid = 0;
1499 	}
1500 	sx_xlock(&allprison_lock);
1501 	drflags = PD_LIST_XLOCKED;
1502 	ppr = mypr;
1503 	if (!prison_isalive(ppr)) {
1504 		/* This jail is dying.  This process will surely follow. */
1505 		error = EAGAIN;
1506 		goto done_deref;
1507 	}
1508 	if (flags & JAIL_USE_DESC) {
1509 		/* Get the jail from its descriptor. */
1510 		error = jaildesc_find(td, jfd_in, &pr, &jdcred);
1511 		if (error) {
1512 			vfs_opterror(opts, error == ENOENT ?
1513 			    "descriptor to dead jail" :
1514 			    "not a jail descriptor");
1515 			goto done_deref;
1516 		}
1517 		drflags |= PD_DEREF;
1518 		error = priv_check_cred(jdcred, PRIV_JAIL_SET);
1519 		if (error == 0 && (flags & JAIL_ATTACH))
1520 			error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
1521 		crfree(jdcred);
1522 		if (error)
1523 			goto done_deref;
1524 		mtx_lock(&pr->pr_mtx);
1525 		drflags |= PD_LOCKED;
1526 		if (cuflags == JAIL_CREATE) {
1527 			error = EEXIST;
1528 			vfs_opterror(opts, "jail %d already exists",
1529 			    pr->pr_id);
1530 			goto done_deref;
1531 		}
1532 		if (!prison_isalive(pr)) {
1533 			/* While a jid can be resurrected, the prison
1534 			 * itself cannot.
1535 			 */
1536 			error = ENOENT;
1537 			vfs_opterror(opts, "jail %d is dying", pr->pr_id);
1538 			goto done_deref;
1539 		}
1540 		if (jid != 0 && jid != pr->pr_id) {
1541 			error = EINVAL;
1542 			vfs_opterror(opts, "cannot change jid");
1543 			goto done_deref;
1544 		}
1545 		jid = pr->pr_id;
1546 	} else if (jid != 0) {
1547 		if (jid < 0) {
1548 			error = EINVAL;
1549 			vfs_opterror(opts, "negative jid");
1550 			goto done_deref;
1551 		}
1552 		/*
1553 		 * See if a requested jid already exists.  Keep track of
1554 		 * where it can be inserted later.
1555 		 */
1556 		TAILQ_FOREACH(inspr, &allprison, pr_list) {
1557 			if (inspr->pr_id < jid)
1558 				continue;
1559 			if (inspr->pr_id > jid)
1560 				break;
1561 			if (prison_isalive(inspr)) {
1562 				pr = inspr;
1563 				mtx_lock(&pr->pr_mtx);
1564 				drflags |= PD_LOCKED;
1565 			} else {
1566 				/* Note a dying jail to handle later. */
1567 				deadpr = inspr;
1568 			}
1569 			inspr = NULL;
1570 			break;
1571 		}
1572 		if (cuflags == JAIL_CREATE && pr != NULL) {
1573 			/*
1574 			 * Even creators that cannot see the jail will
1575 			 * get EEXIST.
1576 			 */
1577 			error = EEXIST;
1578 			vfs_opterror(opts, "jail %d already exists", jid);
1579 			goto done_deref;
1580 		}
1581 		if ((pr == NULL)
1582 		    ? cuflags == JAIL_UPDATE
1583 		    : !prison_ischild(mypr, pr)) {
1584 			/*
1585 			 * Updaters get ENOENT for nonexistent jails,
1586 			 * or for jails they cannot see.  The latter
1587 			 * case is true even for CREATE | UPDATE,
1588 			 * which normally cannot give this error.
1589 			 */
1590 			error = ENOENT;
1591 			vfs_opterror(opts, "jail %d not found", jid);
1592 			goto done_deref;
1593 		}
1594 	}
1595 	/*
1596 	 * If the caller provided a name, look for a jail by that name.
1597 	 * This has different semantics for creates and updates keyed by jid
1598 	 * (where the name must not already exist in a different jail),
1599 	 * and updates keyed by the name itself (where the name must exist
1600 	 * because that is the jail being updated).
1601 	 */
1602 	namelc = NULL;
1603 	if (name != NULL) {
1604 		namelc = strrchr(name, '.');
1605 		if (namelc == NULL)
1606 			namelc = name;
1607 		else {
1608 			/*
1609 			 * This is a hierarchical name.  Split it into the
1610 			 * parent and child names, and make sure the parent
1611 			 * exists or matches an already found jail.
1612 			 */
1613 			if (pr != NULL) {
1614 				if (strncmp(name, ppr->pr_name, namelc - name)
1615 				    || ppr->pr_name[namelc - name] != '\0') {
1616 					error = EINVAL;
1617 					vfs_opterror(opts,
1618 					    "cannot change jail's parent");
1619 					goto done_deref;
1620 				}
1621 			} else {
1622 				*namelc = '\0';
1623 				ppr = prison_find_name(mypr, name);
1624 				if (ppr == NULL) {
1625 					error = ENOENT;
1626 					vfs_opterror(opts,
1627 					    "jail \"%s\" not found", name);
1628 					goto done_deref;
1629 				}
1630 				mtx_unlock(&ppr->pr_mtx);
1631 				if (!prison_isalive(ppr)) {
1632 					error = ENOENT;
1633 					vfs_opterror(opts,
1634 					    "jail \"%s\" is dying", name);
1635 					goto done_deref;
1636 				}
1637 				*namelc = '.';
1638 			}
1639 			namelc++;
1640 		}
1641 		if (namelc[0] != '\0') {
1642 			pnamelen =
1643 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1644 			FOREACH_PRISON_CHILD(ppr, tpr) {
1645 				if (tpr == pr || !prison_isalive(tpr) ||
1646 				    strcmp(tpr->pr_name + pnamelen, namelc))
1647 					continue;
1648 				if (cuflags == JAIL_CREATE || pr != NULL) {
1649 					/*
1650 					 * Create, or update(jid): name must
1651 					 * not exist in an active sibling jail.
1652 					 */
1653 					error = EEXIST;
1654 					vfs_opterror(opts,
1655 					    "jail \"%s\" already exists", name);
1656 					goto done_deref;
1657 				}
1658 				/* Use this jail for updates. */
1659 				pr = tpr;
1660 				mtx_lock(&pr->pr_mtx);
1661 				drflags |= PD_LOCKED;
1662 				break;
1663 			}
1664 			/*
1665 			 * Update: name must exist if no jid is specified.
1666 			 * As with the jid case, the jail must be currently
1667 			 * visible, or else even CREATE | UPDATE will get
1668 			 * an error.
1669 			 */
1670 			if ((pr == NULL)
1671 			    ? cuflags == JAIL_UPDATE
1672 			    : !prison_isalive(pr)) {
1673 				error = ENOENT;
1674 				vfs_opterror(opts, "jail \"%s\" not found",
1675 				    name);
1676 				goto done_deref;
1677 			}
1678 		}
1679 	}
1680 	/* Update: must provide a desc, jid, or name. */
1681 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1682 		error = ENOENT;
1683 		vfs_opterror(opts, "update specified no jail");
1684 		goto done_deref;
1685 	}
1686 
1687 	/* If there's no prison to update, create a new one and link it in. */
1688 	created = pr == NULL;
1689 	if (created) {
1690 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1691 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1692 				error = EPERM;
1693 				vfs_opterror(opts, "prison limit exceeded");
1694 				goto done_deref;
1695 			}
1696 
1697 		if (deadpr != NULL) {
1698 			/*
1699 			 * The prison being created has the same ID as a dying
1700 			 * one.  Handle this by giving the dying jail a new ID.
1701 			 * This may cause some confusion to user space, but
1702 			 * only to those listing dying jails.
1703 			 */
1704 			deadid = get_next_deadid(&dinspr);
1705 			if (deadid == 0) {
1706 				error = EAGAIN;
1707 				vfs_opterror(opts, "no available jail IDs");
1708 				goto done_deref;
1709 			}
1710 			mtx_lock(&deadpr->pr_mtx);
1711 			deadpr->pr_id = deadid;
1712 			mtx_unlock(&deadpr->pr_mtx);
1713 			if (dinspr == deadpr)
1714 				inspr = deadpr;
1715 			else {
1716 				inspr = TAILQ_NEXT(deadpr, pr_list);
1717 				TAILQ_REMOVE(&allprison, deadpr, pr_list);
1718 				if (dinspr != NULL)
1719 					TAILQ_INSERT_AFTER(&allprison, dinspr,
1720 					    deadpr, pr_list);
1721 				else
1722 					TAILQ_INSERT_HEAD(&allprison, deadpr,
1723 					    pr_list);
1724 			}
1725 		}
1726 		if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1727 			error = EAGAIN;
1728 			vfs_opterror(opts, "no available jail IDs");
1729 			goto done_deref;
1730 		}
1731 
1732 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1733 		pr->pr_state = PRISON_STATE_INVALID;
1734 		refcount_init(&pr->pr_ref, 1);
1735 		refcount_init(&pr->pr_uref, 0);
1736 		drflags |= PD_DEREF;
1737 		LIST_INIT(&pr->pr_children);
1738 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1739 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1740 
1741 		pr->pr_id = jid;
1742 		if (inspr != NULL)
1743 			TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1744 		else
1745 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1746 
1747 		pr->pr_parent = ppr;
1748 		prison_hold(ppr);
1749 		prison_proc_hold(ppr);
1750 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1751 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1752 			tpr->pr_childcount++;
1753 		pr->pr_klist = knlist_alloc(&pr->pr_mtx);
1754 
1755 		/* Set some default values, and inherit some from the parent. */
1756 		if (namelc == NULL)
1757 			namelc = "";
1758 		if (path == NULL) {
1759 			path = "/";
1760 			root = mypr->pr_root;
1761 			vref(root);
1762 		}
1763 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1764 		pr->pr_flags |= PR_HOST;
1765 #if defined(INET) || defined(INET6)
1766 #ifdef VIMAGE
1767 		if (!(pr_flags & PR_VNET))
1768 #endif
1769 		{
1770 #ifdef INET
1771 			if (!(ch_flags & PR_IP4_USER))
1772 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1773 			else if (!(pr_flags & PR_IP4_USER)) {
1774 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1775 				prison_ip_dup(ppr, pr, PR_INET);
1776 			}
1777 #endif
1778 #ifdef INET6
1779 			if (!(ch_flags & PR_IP6_USER))
1780 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1781 			else if (!(pr_flags & PR_IP6_USER)) {
1782 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1783 				prison_ip_dup(ppr, pr, PR_INET6);
1784 			}
1785 #endif
1786 		}
1787 #endif
1788 		/* Source address selection is always on by default. */
1789 		pr->pr_flags |= _PR_IP_SADDRSEL;
1790 
1791 		pr->pr_securelevel = ppr->pr_securelevel;
1792 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1793 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1794 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1795 
1796 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1797 		if (osrelstr == NULL)
1798 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1799 			    sizeof(pr->pr_osrelease));
1800 		else
1801 			strlcpy(pr->pr_osrelease, osrelstr,
1802 			    sizeof(pr->pr_osrelease));
1803 
1804 #ifdef VIMAGE
1805 		/*
1806 		 * Allocate a new vnet if specified.
1807 		 *
1808 		 * Set PR_VNET now if so, so that the vnet is disposed of
1809 		 * properly when the jail is destroyed.
1810 		 */
1811 		if (pr_flags & PR_VNET) {
1812 			pr->pr_flags |= PR_VNET;
1813 			pr->pr_vnet = vnet_alloc();
1814 		} else {
1815 			pr->pr_vnet = ppr->pr_vnet;
1816 		}
1817 #endif
1818 		/*
1819 		 * Allocate a dedicated cpuset for each jail.
1820 		 * Unlike other initial settings, this may return an error.
1821 		 */
1822 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1823 		if (error)
1824 			goto done_deref;
1825 
1826 		mtx_lock(&pr->pr_mtx);
1827 		drflags |= PD_LOCKED;
1828 	} else {
1829 		/*
1830 		 * Grab a reference for existing prisons, to ensure they
1831 		 * continue to exist for the duration of the call.
1832 		 */
1833 		if (!(drflags & PD_DEREF)) {
1834 			prison_hold(pr);
1835 			drflags |= PD_DEREF;
1836 		}
1837 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1838 		if ((pr->pr_flags & PR_VNET) &&
1839 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1840 			error = EINVAL;
1841 			vfs_opterror(opts,
1842 			    "vnet jails cannot have IP address restrictions");
1843 			goto done_deref;
1844 		}
1845 #endif
1846 #ifdef INET
1847 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1848 			error = EINVAL;
1849 			vfs_opterror(opts,
1850 			    "ip4 cannot be changed after creation");
1851 			goto done_deref;
1852 		}
1853 #endif
1854 #ifdef INET6
1855 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1856 			error = EINVAL;
1857 			vfs_opterror(opts,
1858 			    "ip6 cannot be changed after creation");
1859 			goto done_deref;
1860 		}
1861 #endif
1862 	}
1863 
1864 	/* Do final error checking before setting anything. */
1865 	if (gotslevel) {
1866 		if (slevel < ppr->pr_securelevel) {
1867 			error = EPERM;
1868 			goto done_deref;
1869 		}
1870 	}
1871 	if (gotchildmax) {
1872 		if (childmax >= ppr->pr_childmax) {
1873 			error = EPERM;
1874 			goto done_deref;
1875 		}
1876 	}
1877 	if (gotenforce) {
1878 		if (enforce < ppr->pr_enforce_statfs) {
1879 			error = EPERM;
1880 			goto done_deref;
1881 		}
1882 	}
1883 	if (gotrsnum) {
1884 		/*
1885 		 * devfs_rsnum is a uint16_t
1886 		 */
1887 		if (rsnum < 0 || rsnum > 65535) {
1888 			error = EINVAL;
1889 			goto done_deref;
1890 		}
1891 		/*
1892 		 * Nested jails always inherit parent's devfs ruleset
1893 		 */
1894 		if (jailed(td->td_ucred)) {
1895 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1896 				error = EPERM;
1897 				goto done_deref;
1898 			} else
1899 				rsnum = ppr->pr_devfs_rsnum;
1900 		}
1901 	}
1902 #ifdef INET
1903 	if (ip4s > 0) {
1904 		if ((ppr->pr_flags & PR_IP4) &&
1905 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1906 		    PR_INET)) {
1907 			error = EPERM;
1908 			goto done_deref;
1909 		}
1910 		if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1911 			error = EADDRINUSE;
1912 			vfs_opterror(opts, "IPv4 addresses clash");
1913 			goto done_deref;
1914 		}
1915 	}
1916 #endif
1917 #ifdef INET6
1918 	if (ip6s > 0) {
1919 		if ((ppr->pr_flags & PR_IP6) &&
1920 		    !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1921 		    PR_INET6)) {
1922 			error = EPERM;
1923 			goto done_deref;
1924 		}
1925 		if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1926 			error = EADDRINUSE;
1927 			vfs_opterror(opts, "IPv6 addresses clash");
1928 			goto done_deref;
1929 		}
1930 	}
1931 #endif
1932 	onamelen = namelen = 0;
1933 	if (namelc != NULL) {
1934 		/* Give a default name of the jid.  Also allow the name to be
1935 		 * explicitly the jid - but not any other number, and only in
1936 		 * normal form (no leading zero/etc).
1937 		 */
1938 		if (namelc[0] == '\0')
1939 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1940 		else if ((strtoul(namelc, &p, 10) != jid ||
1941 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1942 			error = EINVAL;
1943 			vfs_opterror(opts,
1944 			    "name cannot be numeric (unless it is the jid)");
1945 			goto done_deref;
1946 		}
1947 		/*
1948 		 * Make sure the name isn't too long for the prison or its
1949 		 * children.
1950 		 */
1951 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1952 		onamelen = strlen(pr->pr_name + pnamelen);
1953 		namelen = strlen(namelc);
1954 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1955 			error = ENAMETOOLONG;
1956 			goto done_deref;
1957 		}
1958 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1959 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1960 			    sizeof(pr->pr_name)) {
1961 				error = ENAMETOOLONG;
1962 				goto done_deref;
1963 			}
1964 		}
1965 	}
1966 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1967 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1968 		error = EPERM;
1969 		goto done_deref;
1970 	}
1971 
1972 	/*
1973 	 * Let modules check their parameters.  This requires unlocking and
1974 	 * then re-locking the prison, but this is still a valid state as long
1975 	 * as allprison_lock remains xlocked.
1976 	 */
1977 	mtx_unlock(&pr->pr_mtx);
1978 	drflags &= ~PD_LOCKED;
1979 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1980 	if (error != 0)
1981 		goto done_deref;
1982 	mtx_lock(&pr->pr_mtx);
1983 	drflags |= PD_LOCKED;
1984 
1985 	/* At this point, all valid parameters should have been noted. */
1986 	TAILQ_FOREACH(opt, opts, link) {
1987 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1988 			error = EINVAL;
1989 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1990 			goto done_deref;
1991 		}
1992 	}
1993 	maybe_changed = true;
1994 
1995 	/* Set the parameters of the prison. */
1996 #ifdef INET
1997 	redo_ip4 = false;
1998 	if (pr_flags & PR_IP4_USER) {
1999 		pr->pr_flags |= PR_IP4;
2000 		prison_ip_set(pr, PR_INET, ip4);
2001 		ip4 = NULL;
2002 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2003 #ifdef VIMAGE
2004 			if (tpr->pr_flags & PR_VNET) {
2005 				descend = 0;
2006 				continue;
2007 			}
2008 #endif
2009 			if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
2010 				redo_ip4 = true;
2011 				descend = 0;
2012 			}
2013 		}
2014 	}
2015 #endif
2016 #ifdef INET6
2017 	redo_ip6 = false;
2018 	if (pr_flags & PR_IP6_USER) {
2019 		pr->pr_flags |= PR_IP6;
2020 		prison_ip_set(pr, PR_INET6, ip6);
2021 		ip6 = NULL;
2022 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2023 #ifdef VIMAGE
2024 			if (tpr->pr_flags & PR_VNET) {
2025 				descend = 0;
2026 				continue;
2027 			}
2028 #endif
2029 			if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
2030 				redo_ip6 = true;
2031 				descend = 0;
2032 			}
2033 		}
2034 	}
2035 #endif
2036 	if (gotslevel) {
2037 		pr->pr_securelevel = slevel;
2038 		/* Set all child jails to be at least this level. */
2039 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2040 			if (tpr->pr_securelevel < slevel)
2041 				tpr->pr_securelevel = slevel;
2042 	}
2043 	if (gotchildmax) {
2044 		pr->pr_childmax = childmax;
2045 		/* Set all child jails to under this limit. */
2046 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
2047 			if (tpr->pr_childmax > childmax - level)
2048 				tpr->pr_childmax = childmax > level
2049 				    ? childmax - level : 0;
2050 	}
2051 	if (gotenforce) {
2052 		pr->pr_enforce_statfs = enforce;
2053 		/* Pass this restriction on to the children. */
2054 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2055 			if (tpr->pr_enforce_statfs < enforce)
2056 				tpr->pr_enforce_statfs = enforce;
2057 	}
2058 	if (gotrsnum) {
2059 		pr->pr_devfs_rsnum = rsnum;
2060 		/* Pass this restriction on to the children. */
2061 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2062 			tpr->pr_devfs_rsnum = rsnum;
2063 	}
2064 	if (namelc != NULL) {
2065 		if (ppr == &prison0)
2066 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
2067 		else
2068 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
2069 			    ppr->pr_name, namelc);
2070 		/* Change this component of child names. */
2071 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2072 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
2073 			    strlen(tpr->pr_name + onamelen) + 1);
2074 			bcopy(pr->pr_name, tpr->pr_name, namelen);
2075 		}
2076 	}
2077 	if (path != NULL) {
2078 		/* Try to keep a real-rooted full pathname. */
2079 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
2080 		pr->pr_root = root;
2081 		root = NULL;
2082 	}
2083 	if (PR_HOST & ch_flags & ~pr_flags) {
2084 		if (pr->pr_flags & PR_HOST) {
2085 			/*
2086 			 * Copy the parent's host info.  As with pr_ip4 above,
2087 			 * the lack of a lock on the parent is not a problem;
2088 			 * it is always set with allprison_lock at least
2089 			 * shared, and is held exclusively here.
2090 			 */
2091 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
2092 			    sizeof(pr->pr_hostname));
2093 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
2094 			    sizeof(pr->pr_domainname));
2095 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
2096 			    sizeof(pr->pr_hostuuid));
2097 			pr->pr_hostid = pr->pr_parent->pr_hostid;
2098 		}
2099 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
2100 		/* Set this prison, and any descendants without PR_HOST. */
2101 		if (host != NULL)
2102 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
2103 		if (domain != NULL)
2104 			strlcpy(pr->pr_domainname, domain,
2105 			    sizeof(pr->pr_domainname));
2106 		if (uuid != NULL)
2107 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
2108 		if (gothid)
2109 			pr->pr_hostid = hid;
2110 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2111 			if (tpr->pr_flags & PR_HOST)
2112 				descend = 0;
2113 			else {
2114 				if (host != NULL)
2115 					strlcpy(tpr->pr_hostname,
2116 					    pr->pr_hostname,
2117 					    sizeof(tpr->pr_hostname));
2118 				if (domain != NULL)
2119 					strlcpy(tpr->pr_domainname,
2120 					    pr->pr_domainname,
2121 					    sizeof(tpr->pr_domainname));
2122 				if (uuid != NULL)
2123 					strlcpy(tpr->pr_hostuuid,
2124 					    pr->pr_hostuuid,
2125 					    sizeof(tpr->pr_hostuuid));
2126 				if (gothid)
2127 					tpr->pr_hostid = hid;
2128 			}
2129 		}
2130 	}
2131 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2132 	if ((tallow = ch_allow & ~pr_allow))
2133 		prison_set_allow_locked(pr, tallow, 0);
2134 	/*
2135 	 * Persistent prisons get an extra reference, and prisons losing their
2136 	 * persist flag lose that reference.
2137 	 */
2138 	if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2139 		if (pr_flags & PR_PERSIST) {
2140 			prison_hold(pr);
2141 			/*
2142 			 * This may be a new prison's first user reference,
2143 			 * but wait to call it alive until after OSD calls
2144 			 * have had a chance to run (and perhaps to fail).
2145 			 */
2146 			refcount_acquire(&pr->pr_uref);
2147 		} else {
2148 			drflags |= PD_DEUREF;
2149 			prison_free_not_last(pr);
2150 		}
2151 	}
2152 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2153 	mtx_unlock(&pr->pr_mtx);
2154 	drflags &= ~PD_LOCKED;
2155 	/*
2156 	 * Any errors past this point will need to de-persist newly created
2157 	 * prisons, as well as call remove methods.
2158 	 */
2159 	if (created)
2160 		drflags |= PD_KILL;
2161 
2162 #ifdef RACCT
2163 	if (racct_enable && created)
2164 		prison_racct_attach(pr);
2165 #endif
2166 
2167 	/* Locks may have prevented a complete restriction of child IP
2168 	 * addresses.  If so, allocate some more memory and try again.
2169 	 */
2170 #ifdef INET
2171 	while (redo_ip4) {
2172 		ip4s = pr->pr_addrs[PR_INET]->ips;
2173 		MPASS(ip4 == NULL);
2174 		ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2175 		mtx_lock(&pr->pr_mtx);
2176 		redo_ip4 = false;
2177 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2178 #ifdef VIMAGE
2179 			if (tpr->pr_flags & PR_VNET) {
2180 				descend = 0;
2181 				continue;
2182 			}
2183 #endif
2184 			if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2185 				redo_ip4 = true;
2186 		}
2187 		mtx_unlock(&pr->pr_mtx);
2188 	}
2189 #endif
2190 #ifdef INET6
2191 	while (redo_ip6) {
2192 		ip6s = pr->pr_addrs[PR_INET6]->ips;
2193 		MPASS(ip6 == NULL);
2194 		ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2195 		mtx_lock(&pr->pr_mtx);
2196 		redo_ip6 = false;
2197 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2198 #ifdef VIMAGE
2199 			if (tpr->pr_flags & PR_VNET) {
2200 				descend = 0;
2201 				continue;
2202 			}
2203 #endif
2204 			if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2205 				redo_ip6 = true;
2206 		}
2207 		mtx_unlock(&pr->pr_mtx);
2208 	}
2209 #endif
2210 
2211 	/* Let the modules do their work. */
2212 	if (created) {
2213 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2214 		if (error)
2215 			goto done_deref;
2216 	}
2217 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
2218 	if (error)
2219 		goto done_deref;
2220 
2221 	/*
2222 	 * A new prison is now ready to be seen; either it has gained a user
2223 	 * reference via persistence, or is about to gain one via attachment.
2224 	 */
2225 	if (created) {
2226 		sx_assert(&allprison_lock, SX_XLOCKED);
2227 		prison_knote(ppr, NOTE_JAIL_CHILD | pr->pr_id);
2228 		mtx_lock(&pr->pr_mtx);
2229 		drflags |= PD_LOCKED;
2230 		pr->pr_state = PRISON_STATE_ALIVE;
2231 	}
2232 
2233 	/* Attach this process to the prison if requested. */
2234 	if (flags & JAIL_ATTACH) {
2235 		error = do_jail_attach(td, pr,
2236 		    prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2237 		drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2238 		if (error) {
2239 			vfs_opterror(opts, "attach failed");
2240 			goto done_deref;
2241 		}
2242 	}
2243 
2244 #ifdef RACCT
2245 	if (racct_enable && !created) {
2246 		if (drflags & PD_LOCKED) {
2247 			mtx_unlock(&pr->pr_mtx);
2248 			drflags &= ~PD_LOCKED;
2249 		}
2250 		if (drflags & PD_LIST_XLOCKED) {
2251 			sx_xunlock(&allprison_lock);
2252 			drflags &= ~PD_LIST_XLOCKED;
2253 		}
2254 		prison_racct_modify(pr);
2255 	}
2256 #endif
2257 
2258 	if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2259 	    (pr->pr_root->v_vflag & VV_ROOT) == 0)
2260 		printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2261 		   " file system\n", pr->pr_id);
2262 
2263 	/*
2264 	 * Now that the prison is fully created without error, set the
2265 	 * jail descriptor if one was requested.  This is the only
2266 	 * parameter that is returned to the caller (except the error
2267 	 * message).
2268 	 */
2269 	if (jfd_out >= 0) {
2270 		if (!(drflags & PD_LOCKED)) {
2271 			mtx_lock(&pr->pr_mtx);
2272 			drflags |= PD_LOCKED;
2273 		}
2274 		jfd_pos = 2 * vfs_getopt_pos(opts, "desc") + 1;
2275 		if (optuio->uio_segflg == UIO_SYSSPACE)
2276 			*(int*)optuio->uio_iov[jfd_pos].iov_base = jfd_out;
2277 		else
2278 			(void)copyout(&jfd_out,
2279 			    optuio->uio_iov[jfd_pos].iov_base, sizeof(jfd_out));
2280 		jaildesc_set_prison(jfp_out, pr);
2281 	}
2282 
2283 	drflags &= ~PD_KILL;
2284 	td->td_retval[0] = pr->pr_id;
2285 
2286  done_deref:
2287 	/*
2288 	 * Report changes to kevent.  This can happen even if the
2289 	 * system call fails, as changes might have been made before
2290 	 * the failure.
2291 	 */
2292 	if (maybe_changed && !created)
2293 		prison_knote(pr, NOTE_JAIL_SET);
2294 	/* Release any temporary prison holds and/or locks. */
2295 	if (pr != NULL)
2296 		prison_deref(pr, drflags);
2297 	else if (drflags & PD_LIST_SLOCKED)
2298 		sx_sunlock(&allprison_lock);
2299 	else if (drflags & PD_LIST_XLOCKED)
2300 		sx_xunlock(&allprison_lock);
2301 	if (root != NULL)
2302 		vrele(root);
2303  done_errmsg:
2304 	if (error) {
2305 		/* Write the error message back to userspace. */
2306 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2307 		    &errmsg_len) == 0 && errmsg_len > 0) {
2308 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2309 			if (optuio->uio_segflg == UIO_SYSSPACE)
2310 				bcopy(errmsg,
2311 				    optuio->uio_iov[errmsg_pos].iov_base,
2312 				    errmsg_len);
2313 			else
2314 				(void)copyout(errmsg,
2315 				    optuio->uio_iov[errmsg_pos].iov_base,
2316 				    errmsg_len);
2317 		}
2318 	}
2319  done_free:
2320 	/* Clean up other resources. */
2321 #ifdef INET
2322 	prison_ip_free(ip4);
2323 #endif
2324 #ifdef INET6
2325 	prison_ip_free(ip6);
2326 #endif
2327 	if (jfp_out != NULL)
2328 		fdrop(jfp_out, td);
2329 	if (error && jfd_out >= 0)
2330 		(void)kern_close(td, jfd_out);
2331 	if (g_path != NULL)
2332 		free(g_path, M_TEMP);
2333 	vfs_freeopts(opts);
2334 	prison_free(mypr);
2335 	return (error);
2336 }
2337 
2338 /*
2339  * Find the next available prison ID.  Return the ID on success, or zero
2340  * on failure.  Also set a pointer to the allprison list entry the prison
2341  * should be inserted before.
2342  */
2343 static int
get_next_prid(struct prison ** insprp)2344 get_next_prid(struct prison **insprp)
2345 {
2346 	struct prison *inspr;
2347 	int jid, maxid;
2348 
2349 	jid = lastprid % JAIL_MAX + 1;
2350 	if (TAILQ_EMPTY(&allprison) ||
2351 	    TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2352 		/*
2353 		 * A common case is for all jails to be implicitly numbered,
2354 		 * which means they'll go on the end of the list, at least
2355 		 * for the first JAIL_MAX times.
2356 		 */
2357 		inspr = NULL;
2358 	} else {
2359 		/*
2360 		 * Take two passes through the allprison list: first starting
2361 		 * with the proposed jid, then ending with it.
2362 		 */
2363 		for (maxid = JAIL_MAX; maxid != 0; ) {
2364 			TAILQ_FOREACH(inspr, &allprison, pr_list) {
2365 				if (inspr->pr_id < jid)
2366 					continue;
2367 				if (inspr->pr_id > jid) {
2368 					/* Found an opening. */
2369 					maxid = 0;
2370 					break;
2371 				}
2372 				if (++jid > maxid) {
2373 					if (lastprid == maxid || lastprid == 0)
2374 					{
2375 						/*
2376 						 * The entire legal range
2377 						 * has been traversed
2378 						 */
2379 						return 0;
2380 					}
2381 					/* Try again from the start. */
2382 					jid = 1;
2383 					maxid = lastprid;
2384 					break;
2385 				}
2386 			}
2387 			if (inspr == NULL) {
2388 				/* Found room at the end of the list. */
2389 				break;
2390 			}
2391 		}
2392 	}
2393 	*insprp = inspr;
2394 	lastprid = jid;
2395 	return (jid);
2396 }
2397 
2398 /*
2399  * Find the next available ID for a renumbered dead prison.  This is the same
2400  * as get_next_prid, but counting backward from the end of the range.
2401  */
2402 static int
get_next_deadid(struct prison ** dinsprp)2403 get_next_deadid(struct prison **dinsprp)
2404 {
2405 	struct prison *dinspr;
2406 	int deadid, minid;
2407 
2408 	deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2409 	/*
2410 	 * Take two reverse passes through the allprison list: first
2411 	 * starting with the proposed deadid, then ending with it.
2412 	 */
2413 	for (minid = 1; minid != 0; ) {
2414 		TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2415 			if (dinspr->pr_id > deadid)
2416 				continue;
2417 			if (dinspr->pr_id < deadid) {
2418 				/* Found an opening. */
2419 				minid = 0;
2420 				break;
2421 			}
2422 			if (--deadid < minid) {
2423 				if (lastdeadid == minid || lastdeadid == 0)
2424 				{
2425 					/*
2426 					 * The entire legal range
2427 					 * has been traversed
2428 					 */
2429 					return 0;
2430 				}
2431 				/* Try again from the end. */
2432 				deadid = JAIL_MAX;
2433 				minid = lastdeadid;
2434 				break;
2435 			}
2436 		}
2437 		if (dinspr == NULL) {
2438 			/* Found room at the beginning of the list. */
2439 			break;
2440 		}
2441 	}
2442 	*dinsprp = dinspr;
2443 	lastdeadid = deadid;
2444 	return (deadid);
2445 }
2446 
2447 /*
2448  * struct jail_get_args {
2449  *	struct iovec *iovp;
2450  *	unsigned int iovcnt;
2451  *	int flags;
2452  * };
2453  */
2454 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)2455 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2456 {
2457 	struct uio *auio;
2458 	int error;
2459 
2460 	/* Check that we have an even number of iovecs. */
2461 	if (uap->iovcnt & 1)
2462 		return (EINVAL);
2463 
2464 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2465 	if (error)
2466 		return (error);
2467 	error = kern_jail_get(td, auio, uap->flags);
2468 	if (error == 0)
2469 		error = copyout(auio->uio_iov, uap->iovp,
2470 		    uap->iovcnt * sizeof(struct iovec));
2471 	freeuio(auio);
2472 	return (error);
2473 }
2474 
2475 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2476 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2477 {
2478 	struct bool_flags *bf;
2479 	struct file *jfp_out;
2480 	struct jailsys_flags *jsf;
2481 	struct prison *pr, *mypr;
2482 	struct vfsopt *opt;
2483 	struct vfsoptlist *opts;
2484 	char *errmsg, *name;
2485 	int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2486 	int jfd_in, jfd_out;
2487 	unsigned f;
2488 
2489 	if (flags & ~JAIL_GET_MASK)
2490 		return (EINVAL);
2491 	if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
2492 	    (JAIL_USE_DESC | JAIL_AT_DESC))
2493 		return (EINVAL);
2494 
2495 	/* Get the parameter list. */
2496 	error = vfs_buildopts(optuio, &opts);
2497 	if (error)
2498 		return (error);
2499 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2500 	mypr = td->td_ucred->cr_prison;
2501 	prison_hold(mypr);
2502 	pr = NULL;
2503 	jfp_out = NULL;
2504 	jfd_out = -1;
2505 
2506 	/*
2507 	 * Find the prison specified by one of: desc, lastjid, jid, name.
2508 	 */
2509 	sx_slock(&allprison_lock);
2510 	drflags = PD_LIST_SLOCKED;
2511 
2512 	error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
2513 	if (error == ENOENT) {
2514 		if (flags & (JAIL_AT_DESC | JAIL_GET_DESC | JAIL_OWN_DESC)) {
2515 			vfs_opterror(opts, "missing desc");
2516 			goto done;
2517 		}
2518 	} else if (error == 0) {
2519 		if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
2520 		    JAIL_OWN_DESC))) {
2521 			vfs_opterror(opts, "unexpected desc");
2522 			goto done;
2523 		}
2524 		if (flags & JAIL_USE_DESC) {
2525 			/* Get the jail from its descriptor. */
2526 			error = jaildesc_find(td, jfd_in, &pr, NULL);
2527 			if (error) {
2528 				vfs_opterror(opts, error == ENOENT ?
2529 				    "descriptor to dead jail" :
2530 				    "not a jail descriptor");
2531 				goto done;
2532 			}
2533 			drflags |= PD_DEREF;
2534 			mtx_lock(&pr->pr_mtx);
2535 			drflags |= PD_LOCKED;
2536 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2537 				error = ENOENT;
2538 				vfs_opterror(opts, "jail %d is dying",
2539 				    pr->pr_id);
2540 				goto done;
2541 			}
2542 			goto found_prison;
2543 		}
2544 		if (flags & JAIL_AT_DESC) {
2545 			/* Look up jails based on the descriptor's prison. */
2546 			prison_free(mypr);
2547 			error = jaildesc_find(td, jfd_in, &mypr, NULL);
2548 			if (error != 0) {
2549 				vfs_opterror(opts, error == ENOENT ?
2550 				    "descriptor to dead jail" :
2551 				    "not a jail descriptor");
2552 				goto done;
2553 			}
2554 		}
2555 		if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
2556 			/* Allocate a jail descriptor to return later. */
2557 			error = jaildesc_alloc(td, &jfp_out, &jfd_out,
2558 				flags & JAIL_OWN_DESC);
2559 			if (error)
2560 				goto done;
2561 		}
2562 	} else
2563 		goto done;
2564 
2565 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2566 	if (error == 0) {
2567 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2568 			if (pr->pr_id > jid &&
2569 			    ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2570 			    prison_ischild(mypr, pr)) {
2571 				mtx_lock(&pr->pr_mtx);
2572 				drflags |= PD_LOCKED;
2573 				goto found_prison;
2574 			}
2575 		}
2576 		error = ENOENT;
2577 		vfs_opterror(opts, "no jail after %d", jid);
2578 		goto done;
2579 	} else if (error != ENOENT)
2580 		goto done;
2581 
2582 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2583 	if (error == 0) {
2584 		if (jid != 0) {
2585 			pr = prison_find_child(mypr, jid);
2586 			if (pr != NULL) {
2587 				drflags |= PD_LOCKED;
2588 				if (!(prison_isalive(pr) ||
2589 				    (flags & JAIL_DYING))) {
2590 					error = ENOENT;
2591 					vfs_opterror(opts, "jail %d is dying",
2592 					    jid);
2593 					goto done;
2594 				}
2595 				goto found_prison;
2596 			}
2597 			error = ENOENT;
2598 			vfs_opterror(opts, "jail %d not found", jid);
2599 			goto done;
2600 		}
2601 	} else if (error != ENOENT)
2602 		goto done;
2603 
2604 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2605 	if (error == 0) {
2606 		if (len == 0 || name[len - 1] != '\0') {
2607 			error = EINVAL;
2608 			goto done;
2609 		}
2610 		pr = prison_find_name(mypr, name);
2611 		if (pr != NULL) {
2612 			drflags |= PD_LOCKED;
2613 			if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2614 				error = ENOENT;
2615 				vfs_opterror(opts, "jail \"%s\" is dying",
2616 				    name);
2617 				goto done;
2618 			}
2619 			goto found_prison;
2620 		}
2621 		error = ENOENT;
2622 		vfs_opterror(opts, "jail \"%s\" not found", name);
2623 		goto done;
2624 	} else if (error != ENOENT)
2625 		goto done;
2626 
2627 	vfs_opterror(opts, "no jail specified");
2628 	error = ENOENT;
2629 	goto done;
2630 
2631  found_prison:
2632 	/* Get the parameters of the prison. */
2633 	if (!(drflags & PD_DEREF)) {
2634 		prison_hold(pr);
2635 		drflags |= PD_DEREF;
2636 	}
2637 	td->td_retval[0] = pr->pr_id;
2638 	if (jfd_out >= 0) {
2639 		error = vfs_setopt(opts, "desc", &jfd_out, sizeof(jfd_out));
2640 		if (error != 0 && error != ENOENT)
2641 			goto done;
2642 		jaildesc_set_prison(jfp_out, pr);
2643 	}
2644 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2645 	if (error != 0 && error != ENOENT)
2646 		goto done;
2647 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2648 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2649 	if (error != 0 && error != ENOENT)
2650 		goto done;
2651 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2652 	if (error != 0 && error != ENOENT)
2653 		goto done;
2654 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2655 	    sizeof(pr->pr_cpuset->cs_id));
2656 	if (error != 0 && error != ENOENT)
2657 		goto done;
2658 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2659 	if (error != 0 && error != ENOENT)
2660 		goto done;
2661 #ifdef INET
2662 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2663 	    pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2664 	    pr_families[PR_INET].size : 0 );
2665 	if (error != 0 && error != ENOENT)
2666 		goto done;
2667 #endif
2668 #ifdef INET6
2669 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2670 	    pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2671 	    pr_families[PR_INET6].size : 0 );
2672 	if (error != 0 && error != ENOENT)
2673 		goto done;
2674 #endif
2675 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2676 	    sizeof(pr->pr_securelevel));
2677 	if (error != 0 && error != ENOENT)
2678 		goto done;
2679 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2680 	    sizeof(pr->pr_childcount));
2681 	if (error != 0 && error != ENOENT)
2682 		goto done;
2683 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2684 	    sizeof(pr->pr_childmax));
2685 	if (error != 0 && error != ENOENT)
2686 		goto done;
2687 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2688 	if (error != 0 && error != ENOENT)
2689 		goto done;
2690 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2691 	if (error != 0 && error != ENOENT)
2692 		goto done;
2693 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2694 	if (error != 0 && error != ENOENT)
2695 		goto done;
2696 #ifdef COMPAT_FREEBSD32
2697 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2698 		uint32_t hid32 = pr->pr_hostid;
2699 
2700 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2701 	} else
2702 #endif
2703 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2704 	    sizeof(pr->pr_hostid));
2705 	if (error != 0 && error != ENOENT)
2706 		goto done;
2707 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2708 	    sizeof(pr->pr_enforce_statfs));
2709 	if (error != 0 && error != ENOENT)
2710 		goto done;
2711 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2712 	    sizeof(pr->pr_devfs_rsnum));
2713 	if (error != 0 && error != ENOENT)
2714 		goto done;
2715 	for (bf = pr_flag_bool;
2716 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2717 	     bf++) {
2718 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2719 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2720 		if (error != 0 && error != ENOENT)
2721 			goto done;
2722 		i = !i;
2723 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2724 		if (error != 0 && error != ENOENT)
2725 			goto done;
2726 	}
2727 	for (jsf = pr_flag_jailsys;
2728 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2729 	     jsf++) {
2730 		f = pr->pr_flags & (jsf->disable | jsf->new);
2731 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2732 		    : (f == jsf->new) ? JAIL_SYS_NEW
2733 		    : JAIL_SYS_INHERIT;
2734 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2735 		if (error != 0 && error != ENOENT)
2736 			goto done;
2737 	}
2738 	for (bf = pr_flag_allow;
2739 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
2740 		atomic_load_int(&bf->flag) != 0;
2741 	     bf++) {
2742 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2743 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2744 		if (error != 0 && error != ENOENT)
2745 			goto done;
2746 		i = !i;
2747 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2748 		if (error != 0 && error != ENOENT)
2749 			goto done;
2750 	}
2751 	i = !prison_isalive(pr);
2752 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2753 	if (error != 0 && error != ENOENT)
2754 		goto done;
2755 	i = !i;
2756 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2757 	if (error != 0 && error != ENOENT)
2758 		goto done;
2759 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2760 	    sizeof(pr->pr_osreldate));
2761 	if (error != 0 && error != ENOENT)
2762 		goto done;
2763 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2764 	if (error != 0 && error != ENOENT)
2765 		goto done;
2766 
2767 	/* Get the module parameters. */
2768 	mtx_unlock(&pr->pr_mtx);
2769 	drflags &= ~PD_LOCKED;
2770 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2771 	if (error)
2772 		goto done;
2773 	prison_deref(pr, drflags);
2774 	pr = NULL;
2775 	drflags = 0;
2776 
2777 	/* By now, all parameters should have been noted. */
2778 	TAILQ_FOREACH(opt, opts, link) {
2779 		if (!opt->seen &&
2780 		    (strstr(opt->name, JAIL_META_PRIVATE ".") == opt->name ||
2781 		    strstr(opt->name, JAIL_META_SHARED ".") == opt->name)) {
2782 			/* Communicate back a missing key. */
2783 			free(opt->value, M_MOUNT);
2784 			opt->value = NULL;
2785 			opt->len = 0;
2786 			continue;
2787 		}
2788 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2789 			error = EINVAL;
2790 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2791 			goto done;
2792 		}
2793 	}
2794 
2795 	/* Write the fetched parameters back to userspace. */
2796 	error = 0;
2797 	TAILQ_FOREACH(opt, opts, link) {
2798 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2799 			pos = 2 * opt->pos + 1;
2800 			optuio->uio_iov[pos].iov_len = opt->len;
2801 			if (opt->value != NULL) {
2802 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2803 					bcopy(opt->value,
2804 					    optuio->uio_iov[pos].iov_base,
2805 					    opt->len);
2806 				} else {
2807 					error = copyout(opt->value,
2808 					    optuio->uio_iov[pos].iov_base,
2809 					    opt->len);
2810 					if (error)
2811 						break;
2812 				}
2813 			}
2814 		}
2815 	}
2816 
2817  done:
2818 	/* Release any temporary prison holds and/or locks. */
2819 	if (pr != NULL)
2820 		prison_deref(pr, drflags);
2821 	else if (drflags & PD_LIST_SLOCKED)
2822 		sx_sunlock(&allprison_lock);
2823 	else if (drflags & PD_LIST_XLOCKED)
2824 		sx_xunlock(&allprison_lock);
2825 	/* Clean up other resources. */
2826 	if (jfp_out != NULL)
2827 		(void)fdrop(jfp_out, td);
2828 	if (error && jfd_out >= 0)
2829 		(void)kern_close(td, jfd_out);
2830 	if (error && errmsg_pos >= 0) {
2831 		/* Write the error message back to userspace. */
2832 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2833 		errmsg_pos = 2 * errmsg_pos + 1;
2834 		if (errmsg_len > 0) {
2835 			if (optuio->uio_segflg == UIO_SYSSPACE)
2836 				bcopy(errmsg,
2837 				    optuio->uio_iov[errmsg_pos].iov_base,
2838 				    errmsg_len);
2839 			else
2840 				(void)copyout(errmsg,
2841 				    optuio->uio_iov[errmsg_pos].iov_base,
2842 				    errmsg_len);
2843 		}
2844 	}
2845 	vfs_freeopts(opts);
2846 	prison_free(mypr);
2847 	return (error);
2848 }
2849 
2850 /*
2851  * struct jail_remove_args {
2852  *	int jid;
2853  * };
2854  */
2855 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2856 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2857 {
2858 	struct prison *pr;
2859 	int error;
2860 
2861 	error = priv_check(td, PRIV_JAIL_REMOVE);
2862 	if (error)
2863 		return (error);
2864 
2865 	sx_xlock(&allprison_lock);
2866 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2867 	if (pr == NULL) {
2868 		sx_xunlock(&allprison_lock);
2869 		return (EINVAL);
2870 	}
2871 	prison_hold(pr);
2872 	prison_remove(pr);
2873 	return (0);
2874 }
2875 
2876 /*
2877  * struct jail_remove_jd_args {
2878  *	int fd;
2879  * };
2880  */
2881 int
sys_jail_remove_jd(struct thread * td,struct jail_remove_jd_args * uap)2882 sys_jail_remove_jd(struct thread *td, struct jail_remove_jd_args *uap)
2883 {
2884 	struct prison *pr;
2885 	struct ucred *jdcred;
2886 	int error;
2887 
2888 	error = jaildesc_find(td, uap->fd, &pr, &jdcred);
2889 	if (error)
2890 		return (error);
2891 	error = priv_check_cred(jdcred, PRIV_JAIL_REMOVE);
2892 	crfree(jdcred);
2893 	if (error) {
2894 		prison_free(pr);
2895 		return (error);
2896 	}
2897 	sx_xlock(&allprison_lock);
2898 	mtx_lock(&pr->pr_mtx);
2899 	prison_remove(pr);
2900 	return (0);
2901 }
2902 
2903 /*
2904  * Begin the removal process for a prison.  The allprison lock should
2905  * be held exclusively, and the prison should be both locked and held.
2906  */
2907 void
prison_remove(struct prison * pr)2908 prison_remove(struct prison *pr)
2909 {
2910 	sx_assert(&allprison_lock, SA_XLOCKED);
2911 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2912 	if (!prison_isalive(pr)) {
2913 		/* Silently ignore already-dying prisons. */
2914 		mtx_unlock(&pr->pr_mtx);
2915 		sx_xunlock(&allprison_lock);
2916 		return;
2917 	}
2918 	prison_deref(pr, PD_KILL | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2919 }
2920 
2921 /*
2922  * struct jail_attach_args {
2923  *	int jid;
2924  * };
2925  */
2926 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)2927 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2928 {
2929 	struct prison *pr;
2930 	int error;
2931 
2932 	error = priv_check(td, PRIV_JAIL_ATTACH);
2933 	if (error)
2934 		return (error);
2935 
2936 	sx_slock(&allprison_lock);
2937 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2938 	if (pr == NULL) {
2939 		sx_sunlock(&allprison_lock);
2940 		return (EINVAL);
2941 	}
2942 
2943 	/* Do not allow a process to attach to a prison that is not alive. */
2944 	if (!prison_isalive(pr)) {
2945 		mtx_unlock(&pr->pr_mtx);
2946 		sx_sunlock(&allprison_lock);
2947 		return (EINVAL);
2948 	}
2949 
2950 	return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2951 }
2952 
2953 /*
2954  * struct jail_attach_jd_args {
2955  *	int fd;
2956  * };
2957  */
2958 int
sys_jail_attach_jd(struct thread * td,struct jail_attach_jd_args * uap)2959 sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
2960 {
2961 	struct prison *pr;
2962 	struct ucred *jdcred;
2963 	int drflags, error;
2964 
2965 	sx_slock(&allprison_lock);
2966 	drflags = PD_LIST_SLOCKED;
2967 	error = jaildesc_find(td, uap->fd, &pr, &jdcred);
2968 	if (error)
2969 		goto fail;
2970 	drflags |= PD_DEREF;
2971 	error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
2972 	crfree(jdcred);
2973 	if (error)
2974 		goto fail;
2975 	mtx_lock(&pr->pr_mtx);
2976 	drflags |= PD_LOCKED;
2977 
2978 	/* Do not allow a process to attach to a prison that is not alive. */
2979 	if (!prison_isalive(pr)) {
2980 		error = EINVAL;
2981 		goto fail;
2982 	}
2983 
2984 	return (do_jail_attach(td, pr, drflags));
2985 
2986  fail:
2987 	prison_deref(pr, drflags);
2988 	return (error);
2989 }
2990 
2991 static int
do_jail_attach(struct thread * td,struct prison * pr,int drflags)2992 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2993 {
2994 	struct proc *p;
2995 	struct ucred *newcred, *oldcred;
2996 	int error;
2997 
2998 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2999 	sx_assert(&allprison_lock, SX_LOCKED);
3000 	drflags &= PD_LOCK_FLAGS;
3001 	/*
3002 	 * XXX: Note that there is a slight race here if two threads
3003 	 * in the same privileged process attempt to attach to two
3004 	 * different jails at the same time.  It is important for
3005 	 * user processes not to do this, or they might end up with
3006 	 * a process root from one prison, but attached to the jail
3007 	 * of another.
3008 	 */
3009 	if (!(drflags & PD_DEREF)) {
3010 		prison_hold(pr);
3011 		drflags |= PD_DEREF;
3012 	}
3013 	refcount_acquire(&pr->pr_uref);
3014 	drflags |= PD_DEUREF;
3015 	mtx_unlock(&pr->pr_mtx);
3016 	drflags &= ~PD_LOCKED;
3017 
3018 	/* Let modules do whatever they need to prepare for attaching. */
3019 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
3020 	if (error) {
3021 		prison_deref(pr, drflags);
3022 		return (error);
3023 	}
3024 	sx_unlock(&allprison_lock);
3025 	drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
3026 
3027 	/*
3028 	 * Reparent the newly attached process to this jail.
3029 	 */
3030 	p = td->td_proc;
3031 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
3032 	if (error)
3033 		goto e_revert_osd;
3034 
3035 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
3036 	if ((error = change_dir(pr->pr_root, td)) != 0)
3037 		goto e_unlock;
3038 #ifdef MAC
3039 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
3040 		goto e_unlock;
3041 #endif
3042 	VOP_UNLOCK(pr->pr_root);
3043 	if ((error = pwd_chroot_chdir(td, pr->pr_root)))
3044 		goto e_revert_osd;
3045 
3046 	newcred = crget();
3047 	PROC_LOCK(p);
3048 	oldcred = crcopysafe(p, newcred);
3049 	newcred->cr_prison = pr;
3050 	proc_set_cred(p, newcred);
3051 	setsugid(p);
3052 #ifdef RACCT
3053 	racct_proc_ucred_changed(p, oldcred, newcred);
3054 	crhold(newcred);
3055 #endif
3056 	PROC_UNLOCK(p);
3057 #ifdef RCTL
3058 	rctl_proc_ucred_changed(p, newcred);
3059 	crfree(newcred);
3060 #endif
3061 	prison_proc_relink(oldcred->cr_prison, pr, p);
3062 	prison_deref(oldcred->cr_prison, drflags);
3063 	crfree(oldcred);
3064 	prison_knote(pr, NOTE_JAIL_ATTACH | td->td_proc->p_pid);
3065 
3066 	/*
3067 	 * If the prison was killed while changing credentials, die along
3068 	 * with it.
3069 	 */
3070 	if (!prison_isalive(pr)) {
3071 		PROC_LOCK(p);
3072 		kern_psignal(p, SIGKILL);
3073 		PROC_UNLOCK(p);
3074 	}
3075 
3076 	return (0);
3077 
3078  e_unlock:
3079 	VOP_UNLOCK(pr->pr_root);
3080  e_revert_osd:
3081 	/* Tell modules this thread is still in its old jail after all. */
3082 	sx_slock(&allprison_lock);
3083 	drflags |= PD_LIST_SLOCKED;
3084 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
3085 	prison_deref(pr, drflags);
3086 	return (error);
3087 }
3088 
3089 /*
3090  * Returns a locked prison instance, or NULL on failure.
3091  */
3092 struct prison *
prison_find(int prid)3093 prison_find(int prid)
3094 {
3095 	struct prison *pr;
3096 
3097 	sx_assert(&allprison_lock, SX_LOCKED);
3098 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3099 		if (pr->pr_id < prid)
3100 			continue;
3101 		if (pr->pr_id > prid)
3102 			break;
3103 		KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
3104 		mtx_lock(&pr->pr_mtx);
3105 		return (pr);
3106 	}
3107 	return (NULL);
3108 }
3109 
3110 /*
3111  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
3112  */
3113 struct prison *
prison_find_child(struct prison * mypr,int prid)3114 prison_find_child(struct prison *mypr, int prid)
3115 {
3116 	struct prison *pr;
3117 	int descend;
3118 
3119 	sx_assert(&allprison_lock, SX_LOCKED);
3120 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3121 		if (pr->pr_id == prid) {
3122 			KASSERT(prison_isvalid(pr),
3123 			    ("Found invalid prison %p", pr));
3124 			mtx_lock(&pr->pr_mtx);
3125 			return (pr);
3126 		}
3127 	}
3128 	return (NULL);
3129 }
3130 
3131 /*
3132  * Look for the name relative to mypr.  Returns a locked prison or NULL.
3133  */
3134 struct prison *
prison_find_name(struct prison * mypr,const char * name)3135 prison_find_name(struct prison *mypr, const char *name)
3136 {
3137 	struct prison *pr, *deadpr;
3138 	size_t mylen;
3139 	int descend;
3140 
3141 	sx_assert(&allprison_lock, SX_LOCKED);
3142 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
3143 	deadpr = NULL;
3144 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3145 		if (!strcmp(pr->pr_name + mylen, name)) {
3146 			KASSERT(prison_isvalid(pr),
3147 			    ("Found invalid prison %p", pr));
3148 			if (prison_isalive(pr)) {
3149 				mtx_lock(&pr->pr_mtx);
3150 				return (pr);
3151 			}
3152 			deadpr = pr;
3153 		}
3154 	}
3155 	/* There was no valid prison - perhaps there was a dying one. */
3156 	if (deadpr != NULL)
3157 		mtx_lock(&deadpr->pr_mtx);
3158 	return (deadpr);
3159 }
3160 
3161 /*
3162  * See if a prison has the specific flag set.  The prison should be locked,
3163  * unless checking for flags that are only set at jail creation (such as
3164  * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
3165  * to any other prison data.
3166  */
3167 bool
prison_flag(struct ucred * cred,unsigned flag)3168 prison_flag(struct ucred *cred, unsigned flag)
3169 {
3170 
3171 	return ((cred->cr_prison->pr_flags & flag) != 0);
3172 }
3173 
3174 /*
3175  * See if a prison has the specific allow flag set.
3176  * The prison *should* be locked, or only a single bit is examined, without
3177  * regard to any other prison data.
3178  */
3179 bool
prison_allow(struct ucred * cred,unsigned flag)3180 prison_allow(struct ucred *cred, unsigned flag)
3181 {
3182 
3183 	return ((cred->cr_prison->pr_allow & flag) != 0);
3184 }
3185 
3186 /*
3187  * Hold a prison reference, by incrementing pr_ref.  It is generally
3188  * an error to hold a prison that does not already have a reference.
3189  * A prison record will remain valid as long as it has at least one
3190  * reference, and will not be removed as long as either the prison
3191  * mutex or the allprison lock is held (allprison_lock may be shared).
3192  */
3193 void
prison_hold_locked(struct prison * pr)3194 prison_hold_locked(struct prison *pr)
3195 {
3196 
3197 	/* Locking is no longer required. */
3198 	prison_hold(pr);
3199 }
3200 
3201 void
prison_hold(struct prison * pr)3202 prison_hold(struct prison *pr)
3203 {
3204 #ifdef INVARIANTS
3205 	int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
3206 
3207 	KASSERT(was_valid,
3208 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
3209 #else
3210 	refcount_acquire(&pr->pr_ref);
3211 #endif
3212 }
3213 
3214 /*
3215  * Remove a prison reference.  If that was the last reference, the
3216  * prison will be removed (at a later time).
3217  */
3218 void
prison_free_locked(struct prison * pr)3219 prison_free_locked(struct prison *pr)
3220 {
3221 
3222 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3223 	/*
3224 	 * Locking is no longer required, but unlock because the caller
3225 	 * expects it.
3226 	 */
3227 	mtx_unlock(&pr->pr_mtx);
3228 	prison_free(pr);
3229 }
3230 
3231 void
prison_free(struct prison * pr)3232 prison_free(struct prison *pr)
3233 {
3234 
3235 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3236 	    ("Trying to free dead prison %p (jid=%d).",
3237 	     pr, pr->pr_id));
3238 	if (!refcount_release_if_not_last(&pr->pr_ref)) {
3239 		/*
3240 		 * Don't remove the last reference in this context,
3241 		 * in case there are locks held.
3242 		 */
3243 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3244 	}
3245 }
3246 
3247 static void
prison_free_not_last(struct prison * pr)3248 prison_free_not_last(struct prison *pr)
3249 {
3250 #ifdef INVARIANTS
3251 	int lastref;
3252 
3253 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3254 	    ("Trying to free dead prison %p (jid=%d).",
3255 	     pr, pr->pr_id));
3256 	lastref = refcount_release(&pr->pr_ref);
3257 	KASSERT(!lastref,
3258 	    ("prison_free_not_last freed last ref on prison %p (jid=%d).",
3259 	     pr, pr->pr_id));
3260 #else
3261 	refcount_release(&pr->pr_ref);
3262 #endif
3263 }
3264 
3265 /*
3266  * Hold a prison for user visibility, by incrementing pr_uref.
3267  * It is generally an error to hold a prison that isn't already
3268  * user-visible, except through the jail system calls.  It is also
3269  * an error to hold an invalid prison.  A prison record will remain
3270  * alive as long as it has at least one user reference, and will not
3271  * be set to the dying state until the prison mutex and allprison_lock
3272  * are both freed.
3273  */
3274 void
prison_proc_hold(struct prison * pr)3275 prison_proc_hold(struct prison *pr)
3276 {
3277 #ifdef INVARIANTS
3278 	int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
3279 
3280 	KASSERT(was_alive,
3281 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
3282 #else
3283 	refcount_acquire(&pr->pr_uref);
3284 #endif
3285 }
3286 
3287 /*
3288  * Remove a prison user reference.  If it was the last reference, the
3289  * prison will be considered "dying", and may be removed once all of
3290  * its references are dropped.
3291  */
3292 void
prison_proc_free(struct prison * pr)3293 prison_proc_free(struct prison *pr)
3294 {
3295 
3296 	/*
3297 	 * Locking is only required when releasing the last reference.
3298 	 * This allows assurance that a locked prison will remain alive
3299 	 * until it is unlocked.
3300 	 */
3301 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3302 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
3303 	if (!refcount_release_if_not_last(&pr->pr_uref)) {
3304 		/*
3305 		 * Don't remove the last user reference in this context,
3306 		 * which is expected to be a process that is not only locked,
3307 		 * but also half dead.  Add a reference so any calls to
3308 		 * prison_free() won't re-submit the task.
3309 		 */
3310 		prison_hold(pr);
3311 		mtx_lock(&pr->pr_mtx);
3312 		KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
3313 		    ("Redundant last reference in prison_proc_free (jid=%d)",
3314 		     pr->pr_id));
3315 		pr->pr_flags |= PR_COMPLETE_PROC;
3316 		mtx_unlock(&pr->pr_mtx);
3317 		taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3318 	}
3319 }
3320 
3321 static void
prison_proc_free_not_last(struct prison * pr)3322 prison_proc_free_not_last(struct prison *pr)
3323 {
3324 #ifdef INVARIANTS
3325 	int lastref;
3326 
3327 	KASSERT(refcount_load(&pr->pr_uref) > 0,
3328 	    ("Trying to free dead prison %p (jid=%d).",
3329 	     pr, pr->pr_id));
3330 	lastref = refcount_release(&pr->pr_uref);
3331 	KASSERT(!lastref,
3332 	    ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3333 	     pr, pr->pr_id));
3334 #else
3335 	refcount_release(&pr->pr_uref);
3336 #endif
3337 }
3338 
3339 void
prison_proc_link(struct prison * pr,struct proc * p)3340 prison_proc_link(struct prison *pr, struct proc *p)
3341 {
3342 
3343 	sx_assert(&allproc_lock, SA_XLOCKED);
3344 	LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3345 }
3346 
3347 void
prison_proc_unlink(struct prison * pr,struct proc * p)3348 prison_proc_unlink(struct prison *pr, struct proc *p)
3349 {
3350 
3351 	sx_assert(&allproc_lock, SA_XLOCKED);
3352 	LIST_REMOVE(p, p_jaillist);
3353 }
3354 
3355 static void
prison_proc_relink(struct prison * opr,struct prison * npr,struct proc * p)3356 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3357 {
3358 
3359 	sx_xlock(&allproc_lock);
3360 	prison_proc_unlink(opr, p);
3361 	prison_proc_link(npr, p);
3362 	sx_xunlock(&allproc_lock);
3363 }
3364 
3365 /*
3366  * Complete a call to either prison_free or prison_proc_free.
3367  */
3368 static void
prison_complete(void * context,int pending)3369 prison_complete(void *context, int pending)
3370 {
3371 	struct prison *pr = context;
3372 	int drflags;
3373 
3374 	/*
3375 	 * This could be called to release the last reference, or the last
3376 	 * user reference (plus the reference held in prison_proc_free).
3377 	 */
3378 	drflags = prison_lock_xlock(pr, PD_DEREF);
3379 	if (pr->pr_flags & PR_COMPLETE_PROC) {
3380 		pr->pr_flags &= ~PR_COMPLETE_PROC;
3381 		drflags |= PD_DEUREF;
3382 	}
3383 	prison_deref(pr, drflags);
3384 }
3385 
3386 static void
prison_kill_processes_cb(struct proc * p,void * arg __unused)3387 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3388 {
3389 
3390 	kern_psignal(p, SIGKILL);
3391 }
3392 
3393 /*
3394  * Note the iteration does not guarantee acting on all processes.
3395  * Most notably there may be fork or jail_attach in progress.
3396  */
3397 void
prison_proc_iterate(struct prison * pr,void (* cb)(struct proc *,void *),void * cbarg)3398 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3399     void *cbarg)
3400 {
3401 	struct prison *ppr;
3402 	struct proc *p;
3403 
3404 	if (atomic_load_int(&pr->pr_childcount) == 0) {
3405 		sx_slock(&allproc_lock);
3406 		LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3407 			if (p->p_state == PRS_NEW)
3408 				continue;
3409 			PROC_LOCK(p);
3410 			cb(p, cbarg);
3411 			PROC_UNLOCK(p);
3412 		}
3413 		sx_sunlock(&allproc_lock);
3414 		if (atomic_load_int(&pr->pr_childcount) == 0)
3415 			return;
3416 		/*
3417 		 * Some jails popped up during the iteration, fall through to a
3418 		 * system-wide search.
3419 		 */
3420 	}
3421 
3422 	sx_slock(&allproc_lock);
3423 	FOREACH_PROC_IN_SYSTEM(p) {
3424 		PROC_LOCK(p);
3425 		if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3426 			for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3427 			    ppr = ppr->pr_parent) {
3428 				if (ppr == pr) {
3429 					cb(p, cbarg);
3430 					break;
3431 				}
3432 			}
3433 		}
3434 		PROC_UNLOCK(p);
3435 	}
3436 	sx_sunlock(&allproc_lock);
3437 }
3438 
3439 /*
3440  * Remove a prison reference and/or user reference (usually).
3441  * This assumes context that allows sleeping (for allprison_lock),
3442  * with no non-sleeping locks held, except perhaps the prison itself.
3443  * If there are no more references, release and delist the prison.
3444  * On completion, the prison lock and the allprison lock are both
3445  * unlocked.
3446  */
3447 static void
prison_deref(struct prison * pr,int flags)3448 prison_deref(struct prison *pr, int flags)
3449 {
3450 	struct prisonlist freeprison;
3451 	struct prison *killpr, *rpr, *ppr, *tpr;
3452 
3453 	killpr = NULL;
3454 	TAILQ_INIT(&freeprison);
3455 	/*
3456 	 * Release this prison as requested, which may cause its parent
3457 	 * to be released, and then maybe its grandparent, etc.
3458 	 */
3459 	for (;;) {
3460 		if (flags & PD_KILL) {
3461 			/* Kill the prison and its descendents. */
3462 			KASSERT(pr != &prison0,
3463 			    ("prison_deref trying to kill prison0"));
3464 			if (!(flags & PD_DEREF)) {
3465 				prison_hold(pr);
3466 				flags |= PD_DEREF;
3467 			}
3468 			flags = prison_lock_xlock(pr, flags);
3469 			prison_deref_kill(pr, &freeprison);
3470 		}
3471 		if (flags & PD_DEUREF) {
3472 			/* Drop a user reference. */
3473 			KASSERT(refcount_load(&pr->pr_uref) > 0,
3474 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3475 			     pr->pr_id));
3476 			if (!refcount_release_if_not_last(&pr->pr_uref)) {
3477 				if (!(flags & PD_DEREF)) {
3478 					prison_hold(pr);
3479 					flags |= PD_DEREF;
3480 				}
3481 				flags = prison_lock_xlock(pr, flags);
3482 				if (refcount_release(&pr->pr_uref) &&
3483 				    pr->pr_state == PRISON_STATE_ALIVE) {
3484 					/*
3485 					 * When the last user references goes,
3486 					 * this becomes a dying prison.
3487 					 */
3488 					KASSERT(
3489 					    refcount_load(&prison0.pr_uref) > 0,
3490 					    ("prison0 pr_uref=0"));
3491 					pr->pr_state = PRISON_STATE_DYING;
3492 					prison_cleanup_locked(pr);
3493 					mtx_unlock(&pr->pr_mtx);
3494 					flags &= ~PD_LOCKED;
3495 					prison_cleanup_unlocked(pr);
3496 				}
3497 			}
3498 		}
3499 		if (flags & PD_KILL) {
3500 			/*
3501 			 * Any remaining user references are probably processes
3502 			 * that need to be killed, either in this prison or its
3503 			 * descendants.
3504 			 */
3505 			if (refcount_load(&pr->pr_uref) > 0)
3506 				killpr = pr;
3507 			/* Make sure the parent prison doesn't get killed. */
3508 			flags &= ~PD_KILL;
3509 		}
3510 		if (flags & PD_DEREF) {
3511 			/* Drop a reference. */
3512 			KASSERT(refcount_load(&pr->pr_ref) > 0,
3513 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3514 			     pr->pr_id));
3515 			if (!refcount_release_if_not_last(&pr->pr_ref)) {
3516 				flags = prison_lock_xlock(pr, flags);
3517 				if (refcount_release(&pr->pr_ref)) {
3518 					/*
3519 					 * When the last reference goes,
3520 					 * unlink the prison and set it aside.
3521 					 */
3522 					KASSERT(
3523 					    refcount_load(&pr->pr_uref) == 0,
3524 					    ("prison_deref: last ref, "
3525 					     "but still has %d urefs (jid=%d)",
3526 					     pr->pr_uref, pr->pr_id));
3527 					KASSERT(
3528 					    refcount_load(&prison0.pr_ref) != 0,
3529 					    ("prison0 pr_ref=0"));
3530 					pr->pr_state = PRISON_STATE_INVALID;
3531 					TAILQ_REMOVE(&allprison, pr, pr_list);
3532 					LIST_REMOVE(pr, pr_sibling);
3533 					TAILQ_INSERT_TAIL(&freeprison, pr,
3534 					    pr_list);
3535 					for (ppr = pr->pr_parent;
3536 					     ppr != NULL;
3537 					     ppr = ppr->pr_parent)
3538 						ppr->pr_childcount--;
3539 					/*
3540 					 * Removing a prison frees references
3541 					 * from its parent.
3542 					 */
3543 					ppr = pr->pr_parent;
3544 					pr->pr_parent = NULL;
3545 					mtx_unlock(&pr->pr_mtx);
3546 
3547 					pr = ppr;
3548 					flags &= ~PD_LOCKED;
3549 					flags |= PD_DEREF | PD_DEUREF;
3550 					continue;
3551 				}
3552 			}
3553 		}
3554 		break;
3555 	}
3556 
3557 	/* Release all the prison locks. */
3558 	if (flags & PD_LOCKED)
3559 		mtx_unlock(&pr->pr_mtx);
3560 	if (flags & PD_LIST_SLOCKED)
3561 		sx_sunlock(&allprison_lock);
3562 	else if (flags & PD_LIST_XLOCKED)
3563 		sx_xunlock(&allprison_lock);
3564 
3565 	/* Kill any processes attached to a killed prison. */
3566 	if (killpr != NULL)
3567 		prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3568 
3569 	/*
3570 	 * Finish removing any unreferenced prisons, which couldn't happen
3571 	 * while allprison_lock was held (to avoid a LOR on vrele).
3572 	 */
3573 	TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3574 #ifdef VIMAGE
3575 		if (rpr->pr_flags & PR_VNET)
3576 			vnet_destroy(rpr->pr_vnet);
3577 #endif
3578 		if (rpr->pr_root != NULL)
3579 			vrele(rpr->pr_root);
3580 		mtx_destroy(&rpr->pr_mtx);
3581 #ifdef INET
3582 		prison_ip_free(rpr->pr_addrs[PR_INET]);
3583 #endif
3584 #ifdef INET6
3585 		prison_ip_free(rpr->pr_addrs[PR_INET6]);
3586 #endif
3587 		if (rpr->pr_cpuset != NULL)
3588 			cpuset_rel(rpr->pr_cpuset);
3589 		osd_jail_exit(rpr);
3590 #ifdef RACCT
3591 		if (racct_enable)
3592 			prison_racct_detach(rpr);
3593 #endif
3594 		TAILQ_REMOVE(&freeprison, rpr, pr_list);
3595 		free(rpr, M_PRISON);
3596 	}
3597 }
3598 
3599 /*
3600  * Kill the prison and its descendants.  Mark them as dying, clear the
3601  * persist flag, and call module remove methods.
3602  */
3603 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)3604 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3605 {
3606 	struct prison *cpr, *ppr, *rpr;
3607 	bool descend;
3608 
3609 	/*
3610 	 * Unlike the descendants, the target prison can be killed
3611 	 * even if it is currently dying.  This is useful for failed
3612 	 * creation in jail_set(2).
3613 	 */
3614 	KASSERT(refcount_load(&pr->pr_ref) > 0,
3615 	    ("Trying to kill dead prison %p (jid=%d).",
3616 	     pr, pr->pr_id));
3617 	refcount_acquire(&pr->pr_uref);
3618 	pr->pr_state = PRISON_STATE_DYING;
3619 	mtx_unlock(&pr->pr_mtx);
3620 
3621 	rpr = NULL;
3622 	FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3623 		if (descend) {
3624 			if (!prison_isalive(cpr)) {
3625 				descend = false;
3626 				continue;
3627 			}
3628 			prison_hold(cpr);
3629 			prison_proc_hold(cpr);
3630 			mtx_lock(&cpr->pr_mtx);
3631 			cpr->pr_state = PRISON_STATE_DYING;
3632 			cpr->pr_flags |= PR_REMOVE;
3633 			mtx_unlock(&cpr->pr_mtx);
3634 			continue;
3635 		}
3636 		if (!(cpr->pr_flags & PR_REMOVE))
3637 			continue;
3638 		prison_cleanup_unlocked(cpr);
3639 		mtx_lock(&cpr->pr_mtx);
3640 		prison_cleanup_locked(cpr);
3641 		cpr->pr_flags &= ~PR_REMOVE;
3642 		if (cpr->pr_flags & PR_PERSIST) {
3643 			cpr->pr_flags &= ~PR_PERSIST;
3644 			prison_proc_free_not_last(cpr);
3645 			prison_free_not_last(cpr);
3646 		}
3647 		(void)refcount_release(&cpr->pr_uref);
3648 		if (refcount_release(&cpr->pr_ref)) {
3649 			/*
3650 			 * When the last reference goes, unlink the prison
3651 			 * and set it aside for prison_deref() to handle.
3652 			 * Delay unlinking the sibling list to keep the loop
3653 			 * safe.
3654 			 */
3655 			if (rpr != NULL)
3656 				LIST_REMOVE(rpr, pr_sibling);
3657 			rpr = cpr;
3658 			rpr->pr_state = PRISON_STATE_INVALID;
3659 			TAILQ_REMOVE(&allprison, rpr, pr_list);
3660 			TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3661 			/*
3662 			 * Removing a prison frees references from its parent.
3663 			 */
3664 			ppr = rpr->pr_parent;
3665 			prison_proc_free_not_last(ppr);
3666 			prison_free_not_last(ppr);
3667 			for (; ppr != NULL; ppr = ppr->pr_parent)
3668 				ppr->pr_childcount--;
3669 		}
3670 		mtx_unlock(&cpr->pr_mtx);
3671 	}
3672 	if (rpr != NULL)
3673 		LIST_REMOVE(rpr, pr_sibling);
3674 
3675 	prison_cleanup_unlocked(pr);
3676 	mtx_lock(&pr->pr_mtx);
3677 	prison_cleanup_locked(pr);
3678 	if (pr->pr_flags & PR_PERSIST) {
3679 		pr->pr_flags &= ~PR_PERSIST;
3680 		prison_proc_free_not_last(pr);
3681 		prison_free_not_last(pr);
3682 	}
3683 	(void)refcount_release(&pr->pr_uref);
3684 }
3685 
3686 /*
3687  * Given the current locking state in the flags, make sure allprison_lock
3688  * is held exclusive, and the prison is locked.  Return flags indicating
3689  * the new state.
3690  */
3691 static int
prison_lock_xlock(struct prison * pr,int flags)3692 prison_lock_xlock(struct prison *pr, int flags)
3693 {
3694 
3695 	if (!(flags & PD_LIST_XLOCKED)) {
3696 		/*
3697 		 * Get allprison_lock, which may be an upgrade,
3698 		 * and may require unlocking the prison.
3699 		 */
3700 		if (flags & PD_LOCKED) {
3701 			mtx_unlock(&pr->pr_mtx);
3702 			flags &= ~PD_LOCKED;
3703 		}
3704 		if (flags & PD_LIST_SLOCKED) {
3705 			if (!sx_try_upgrade(&allprison_lock)) {
3706 				sx_sunlock(&allprison_lock);
3707 				sx_xlock(&allprison_lock);
3708 			}
3709 			flags &= ~PD_LIST_SLOCKED;
3710 		} else
3711 			sx_xlock(&allprison_lock);
3712 		flags |= PD_LIST_XLOCKED;
3713 	}
3714 	if (!(flags & PD_LOCKED)) {
3715 		/* Lock the prison mutex. */
3716 		mtx_lock(&pr->pr_mtx);
3717 		flags |= PD_LOCKED;
3718 	}
3719 	return flags;
3720 }
3721 
3722 /*
3723  * Release a prison's resources when it starts dying (when the last user
3724  * reference is dropped, or when it is killed).  Two functions are called,
3725  * for work that requires a locked prison or an unlocked one.
3726  */
3727 static void
prison_cleanup_locked(struct prison * pr)3728 prison_cleanup_locked(struct prison *pr)
3729 {
3730 	sx_assert(&allprison_lock, SA_XLOCKED);
3731 	mtx_assert(&pr->pr_mtx, MA_OWNED);
3732 	prison_knote(pr, NOTE_JAIL_REMOVE);
3733 	knlist_detach(pr->pr_klist);
3734 	jaildesc_prison_cleanup(pr);
3735 	pr->pr_klist = NULL;
3736 }
3737 
3738 static void
prison_cleanup_unlocked(struct prison * pr)3739 prison_cleanup_unlocked(struct prison *pr)
3740 {
3741 	sx_assert(&allprison_lock, SA_XLOCKED);
3742 	mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3743 	vfs_exjail_delete(pr);
3744 	shm_remove_prison(pr);
3745 	(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3746 }
3747 
3748 /*
3749  * Set or clear a permission bit in the pr_allow field, passing restrictions
3750  * (cleared permission) down to child jails.
3751  */
3752 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)3753 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3754 {
3755 	struct prison *pr;
3756 
3757 	pr = cred->cr_prison;
3758 	sx_slock(&allprison_lock);
3759 	mtx_lock(&pr->pr_mtx);
3760 	prison_set_allow_locked(pr, flag, enable);
3761 	mtx_unlock(&pr->pr_mtx);
3762 	sx_sunlock(&allprison_lock);
3763 }
3764 
3765 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)3766 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3767 {
3768 	struct prison *cpr;
3769 	int descend;
3770 
3771 	if (enable != 0)
3772 		pr->pr_allow |= flag;
3773 	else {
3774 		pr->pr_allow &= ~flag;
3775 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3776 			cpr->pr_allow &= ~flag;
3777 	}
3778 }
3779 
3780 /*
3781  * Check if a jail supports the given address family.
3782  *
3783  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3784  * if not.
3785  */
3786 int
prison_check_af(struct ucred * cred,int af)3787 prison_check_af(struct ucred *cred, int af)
3788 {
3789 	struct prison *pr;
3790 	int error;
3791 
3792 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3793 
3794 	pr = cred->cr_prison;
3795 #ifdef VIMAGE
3796 	/* Prisons with their own network stack are not limited. */
3797 	if (prison_owns_vnet(pr))
3798 		return (0);
3799 #endif
3800 
3801 	error = 0;
3802 	switch (af)
3803 	{
3804 #ifdef INET
3805 	case AF_INET:
3806 		if (pr->pr_flags & PR_IP4)
3807 		{
3808 			mtx_lock(&pr->pr_mtx);
3809 			if ((pr->pr_flags & PR_IP4) &&
3810 			    pr->pr_addrs[PR_INET] == NULL)
3811 				error = EAFNOSUPPORT;
3812 			mtx_unlock(&pr->pr_mtx);
3813 		}
3814 		break;
3815 #endif
3816 #ifdef INET6
3817 	case AF_INET6:
3818 		if (pr->pr_flags & PR_IP6)
3819 		{
3820 			mtx_lock(&pr->pr_mtx);
3821 			if ((pr->pr_flags & PR_IP6) &&
3822 			    pr->pr_addrs[PR_INET6] == NULL)
3823 				error = EAFNOSUPPORT;
3824 			mtx_unlock(&pr->pr_mtx);
3825 		}
3826 		break;
3827 #endif
3828 	case AF_LOCAL:
3829 	case AF_ROUTE:
3830 	case AF_NETLINK:
3831 		break;
3832 	default:
3833 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3834 			error = EAFNOSUPPORT;
3835 	}
3836 	return (error);
3837 }
3838 
3839 /*
3840  * Check if given address belongs to the jail referenced by cred (wrapper to
3841  * prison_check_ip[46]).
3842  *
3843  * Returns 0 if jail doesn't restrict the address family or if address belongs
3844  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3845  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3846  */
3847 int
prison_if(struct ucred * cred,const struct sockaddr * sa)3848 prison_if(struct ucred *cred, const struct sockaddr *sa)
3849 {
3850 #ifdef INET
3851 	const struct sockaddr_in *sai;
3852 #endif
3853 #ifdef INET6
3854 	const struct sockaddr_in6 *sai6;
3855 #endif
3856 	int error;
3857 
3858 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3859 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3860 
3861 #ifdef VIMAGE
3862 	if (prison_owns_vnet(cred->cr_prison))
3863 		return (0);
3864 #endif
3865 
3866 	error = 0;
3867 	switch (sa->sa_family)
3868 	{
3869 #ifdef INET
3870 	case AF_INET:
3871 		sai = (const struct sockaddr_in *)sa;
3872 		error = prison_check_ip4(cred, &sai->sin_addr);
3873 		break;
3874 #endif
3875 #ifdef INET6
3876 	case AF_INET6:
3877 		sai6 = (const struct sockaddr_in6 *)sa;
3878 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3879 		break;
3880 #endif
3881 	default:
3882 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3883 			error = EAFNOSUPPORT;
3884 	}
3885 	return (error);
3886 }
3887 
3888 /*
3889  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3890  */
3891 int
prison_check(struct ucred * cred1,struct ucred * cred2)3892 prison_check(struct ucred *cred1, struct ucred *cred2)
3893 {
3894 
3895 	return ((cred1->cr_prison == cred2->cr_prison ||
3896 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3897 }
3898 
3899 /*
3900  * For mountd/nfsd to run within a prison, it must be:
3901  * - A vnet prison.
3902  * - PR_ALLOW_NFSD must be set on it.
3903  * - The root directory (pr_root) of the prison must be
3904  *   a file system mount point, so the mountd can hang
3905  *   export information on it.
3906  * - The prison's enforce_statfs cannot be 0, so that
3907  *   mountd(8) can do exports.
3908  */
3909 bool
prison_check_nfsd(struct ucred * cred)3910 prison_check_nfsd(struct ucred *cred)
3911 {
3912 
3913 	if (jailed_without_vnet(cred))
3914 		return (false);
3915 	if (!prison_allow(cred, PR_ALLOW_NFSD))
3916 		return (false);
3917 	if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3918 		return (false);
3919 	if (cred->cr_prison->pr_enforce_statfs == 0)
3920 		return (false);
3921 	return (true);
3922 }
3923 
3924 /*
3925  * Return true if p2 is a child of p1, otherwise false.
3926  */
3927 bool
prison_ischild(struct prison * pr1,struct prison * pr2)3928 prison_ischild(struct prison *pr1, struct prison *pr2)
3929 {
3930 
3931 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3932 		if (pr1 == pr2)
3933 			return (true);
3934 	return (false);
3935 }
3936 
3937 /*
3938  * Return true if the prison is currently alive.  A prison is alive if it
3939  * holds user references and it isn't being removed.
3940  */
3941 bool
prison_isalive(const struct prison * pr)3942 prison_isalive(const struct prison *pr)
3943 {
3944 
3945 	if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3946 		return (false);
3947 	return (true);
3948 }
3949 
3950 /*
3951  * Return true if the prison is currently valid.  A prison is valid if it has
3952  * been fully created, and is not being destroyed.  Note that dying prisons
3953  * are still considered valid.  Invalid prisons won't be found under normal
3954  * circumstances, as they're only put in that state by functions that have
3955  * an exclusive hold on allprison_lock.
3956  */
3957 bool
prison_isvalid(struct prison * pr)3958 prison_isvalid(struct prison *pr)
3959 {
3960 
3961 	if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3962 		return (false);
3963 	if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3964 		return (false);
3965 	return (true);
3966 }
3967 
3968 /*
3969  * Return true if the passed credential is in a jail and that jail does not
3970  * have its own virtual network stack, otherwise false.
3971  */
3972 bool
jailed_without_vnet(struct ucred * cred)3973 jailed_without_vnet(struct ucred *cred)
3974 {
3975 
3976 	if (!jailed(cred))
3977 		return (false);
3978 #ifdef VIMAGE
3979 	if (prison_owns_vnet(cred->cr_prison))
3980 		return (false);
3981 #endif
3982 
3983 	return (true);
3984 }
3985 
3986 /*
3987  * Return the correct hostname (domainname, et al) for the passed credential.
3988  */
3989 void
getcredhostname(struct ucred * cred,char * buf,size_t size)3990 getcredhostname(struct ucred *cred, char *buf, size_t size)
3991 {
3992 	struct prison *pr;
3993 
3994 	/*
3995 	 * A NULL credential can be used to shortcut to the physical
3996 	 * system's hostname.
3997 	 */
3998 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3999 	mtx_lock(&pr->pr_mtx);
4000 	strlcpy(buf, pr->pr_hostname, size);
4001 	mtx_unlock(&pr->pr_mtx);
4002 }
4003 
4004 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)4005 getcreddomainname(struct ucred *cred, char *buf, size_t size)
4006 {
4007 
4008 	mtx_lock(&cred->cr_prison->pr_mtx);
4009 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
4010 	mtx_unlock(&cred->cr_prison->pr_mtx);
4011 }
4012 
4013 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)4014 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
4015 {
4016 
4017 	mtx_lock(&cred->cr_prison->pr_mtx);
4018 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
4019 	mtx_unlock(&cred->cr_prison->pr_mtx);
4020 }
4021 
4022 void
getcredhostid(struct ucred * cred,unsigned long * hostid)4023 getcredhostid(struct ucred *cred, unsigned long *hostid)
4024 {
4025 
4026 	mtx_lock(&cred->cr_prison->pr_mtx);
4027 	*hostid = cred->cr_prison->pr_hostid;
4028 	mtx_unlock(&cred->cr_prison->pr_mtx);
4029 }
4030 
4031 void
getjailname(struct ucred * cred,char * name,size_t len)4032 getjailname(struct ucred *cred, char *name, size_t len)
4033 {
4034 
4035 	mtx_lock(&cred->cr_prison->pr_mtx);
4036 	strlcpy(name, cred->cr_prison->pr_name, len);
4037 	mtx_unlock(&cred->cr_prison->pr_mtx);
4038 }
4039 
4040 #ifdef VIMAGE
4041 /*
4042  * Determine whether the prison owns its VNET.
4043  */
4044 bool
prison_owns_vnet(struct prison * pr)4045 prison_owns_vnet(struct prison *pr)
4046 {
4047 
4048 	/*
4049 	 * vnets cannot be added/removed after jail creation,
4050 	 * so no need to lock here.
4051 	 */
4052 	return ((pr->pr_flags & PR_VNET) != 0);
4053 }
4054 #endif
4055 
4056 /*
4057  * Determine whether the subject represented by cred can "see"
4058  * status of a mount point.
4059  * Returns: 0 for permitted, ENOENT otherwise.
4060  * XXX: This function should be called cr_canseemount() and should be
4061  *      placed in kern_prot.c.
4062  */
4063 int
prison_canseemount(struct ucred * cred,struct mount * mp)4064 prison_canseemount(struct ucred *cred, struct mount *mp)
4065 {
4066 	struct prison *pr;
4067 	struct statfs *sp;
4068 	size_t len;
4069 
4070 	pr = cred->cr_prison;
4071 	if (pr->pr_enforce_statfs == 0)
4072 		return (0);
4073 	if (pr->pr_root->v_mount == mp)
4074 		return (0);
4075 	if (pr->pr_enforce_statfs == 2)
4076 		return (ENOENT);
4077 	/*
4078 	 * If jail's chroot directory is set to "/" we should be able to see
4079 	 * all mount-points from inside a jail.
4080 	 * This is ugly check, but this is the only situation when jail's
4081 	 * directory ends with '/'.
4082 	 */
4083 	if (strcmp(pr->pr_path, "/") == 0)
4084 		return (0);
4085 	len = strlen(pr->pr_path);
4086 	sp = &mp->mnt_stat;
4087 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
4088 		return (ENOENT);
4089 	/*
4090 	 * Be sure that we don't have situation where jail's root directory
4091 	 * is "/some/path" and mount point is "/some/pathpath".
4092 	 */
4093 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
4094 		return (ENOENT);
4095 	return (0);
4096 }
4097 
4098 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)4099 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
4100 {
4101 	char jpath[MAXPATHLEN];
4102 	struct prison *pr;
4103 	size_t len;
4104 
4105 	pr = cred->cr_prison;
4106 	if (pr->pr_enforce_statfs == 0)
4107 		return;
4108 	if (prison_canseemount(cred, mp) != 0) {
4109 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4110 		strlcpy(sp->f_mntonname, "[restricted]",
4111 		    sizeof(sp->f_mntonname));
4112 		return;
4113 	}
4114 	if (pr->pr_root->v_mount == mp) {
4115 		/*
4116 		 * Clear current buffer data, so we are sure nothing from
4117 		 * the valid path left there.
4118 		 */
4119 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4120 		*sp->f_mntonname = '/';
4121 		return;
4122 	}
4123 	/*
4124 	 * If jail's chroot directory is set to "/" we should be able to see
4125 	 * all mount-points from inside a jail.
4126 	 */
4127 	if (strcmp(pr->pr_path, "/") == 0)
4128 		return;
4129 	len = strlen(pr->pr_path);
4130 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
4131 	/*
4132 	 * Clear current buffer data, so we are sure nothing from
4133 	 * the valid path left there.
4134 	 */
4135 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4136 	if (*jpath == '\0') {
4137 		/* Should never happen. */
4138 		*sp->f_mntonname = '/';
4139 	} else {
4140 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
4141 	}
4142 }
4143 
4144 /*
4145  * Check with permission for a specific privilege is granted within jail.  We
4146  * have a specific list of accepted privileges; the rest are denied.
4147  */
4148 int
prison_priv_check(struct ucred * cred,int priv)4149 prison_priv_check(struct ucred *cred, int priv)
4150 {
4151 	struct prison *pr;
4152 	int error;
4153 
4154 	/*
4155 	 * Some policies have custom handlers. This routine should not be
4156 	 * called for them. See priv_check_cred().
4157 	 */
4158 	switch (priv) {
4159 	case PRIV_VFS_LOOKUP:
4160 	case PRIV_VFS_GENERATION:
4161 		KASSERT(0, ("prison_priv_check instead of a custom handler "
4162 		    "called for %d\n", priv));
4163 	}
4164 
4165 	if (!jailed(cred))
4166 		return (0);
4167 
4168 #ifdef VIMAGE
4169 	/*
4170 	 * Privileges specific to prisons with a virtual network stack.
4171 	 * There might be a duplicate entry here in case the privilege
4172 	 * is only granted conditionally in the legacy jail case.
4173 	 */
4174 	switch (priv) {
4175 		/*
4176 		 * NFS-specific privileges.
4177 		 */
4178 	case PRIV_NFS_DAEMON:
4179 	case PRIV_VFS_GETFH:
4180 	case PRIV_VFS_MOUNT_EXPORTED:
4181 		if (!prison_check_nfsd(cred))
4182 			return (EPERM);
4183 #ifdef notyet
4184 	case PRIV_NFS_LOCKD:
4185 #endif
4186 		/*
4187 		 * Network stack privileges.
4188 		 */
4189 	case PRIV_NET_BRIDGE:
4190 	case PRIV_NET_GRE:
4191 	case PRIV_NET_BPF:
4192 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
4193 	case PRIV_NET_ROUTE:
4194 	case PRIV_NET_TAP:
4195 	case PRIV_NET_SETIFMTU:
4196 	case PRIV_NET_SETIFFLAGS:
4197 	case PRIV_NET_SETIFCAP:
4198 	case PRIV_NET_SETIFDESCR:
4199 	case PRIV_NET_SETIFNAME	:
4200 	case PRIV_NET_SETIFMETRIC:
4201 	case PRIV_NET_SETIFPHYS:
4202 	case PRIV_NET_SETIFMAC:
4203 	case PRIV_NET_SETLANPCP:
4204 	case PRIV_NET_ADDMULTI:
4205 	case PRIV_NET_DELMULTI:
4206 	case PRIV_NET_HWIOCTL:
4207 	case PRIV_NET_SETLLADDR:
4208 	case PRIV_NET_ADDIFGROUP:
4209 	case PRIV_NET_DELIFGROUP:
4210 	case PRIV_NET_IFCREATE:
4211 	case PRIV_NET_IFDESTROY:
4212 	case PRIV_NET_ADDIFADDR:
4213 	case PRIV_NET_DELIFADDR:
4214 	case PRIV_NET_LAGG:
4215 	case PRIV_NET_GIF:
4216 	case PRIV_NET_SETIFVNET:
4217 	case PRIV_NET_SETIFFIB:
4218 	case PRIV_NET_OVPN:
4219 	case PRIV_NET_ME:
4220 	case PRIV_NET_WG:
4221 
4222 		/*
4223 		 * 802.11-related privileges.
4224 		 */
4225 	case PRIV_NET80211_VAP_GETKEY:
4226 	case PRIV_NET80211_VAP_MANAGE:
4227 
4228 #ifdef notyet
4229 		/*
4230 		 * ATM privileges.
4231 		 */
4232 	case PRIV_NETATM_CFG:
4233 	case PRIV_NETATM_ADD:
4234 	case PRIV_NETATM_DEL:
4235 	case PRIV_NETATM_SET:
4236 
4237 		/*
4238 		 * Bluetooth privileges.
4239 		 */
4240 	case PRIV_NETBLUETOOTH_RAW:
4241 #endif
4242 
4243 		/*
4244 		 * Netgraph and netgraph module privileges.
4245 		 */
4246 	case PRIV_NETGRAPH_CONTROL:
4247 #ifdef notyet
4248 	case PRIV_NETGRAPH_TTY:
4249 #endif
4250 
4251 		/*
4252 		 * IPv4 and IPv6 privileges.
4253 		 */
4254 	case PRIV_NETINET_IPFW:
4255 	case PRIV_NETINET_DIVERT:
4256 	case PRIV_NETINET_PF:
4257 	case PRIV_NETINET_DUMMYNET:
4258 	case PRIV_NETINET_CARP:
4259 	case PRIV_NETINET_MROUTE:
4260 	case PRIV_NETINET_RAW:
4261 	case PRIV_NETINET_ADDRCTRL6:
4262 	case PRIV_NETINET_ND6:
4263 	case PRIV_NETINET_SCOPE6:
4264 	case PRIV_NETINET_ALIFETIME6:
4265 	case PRIV_NETINET_IPSEC:
4266 	case PRIV_NETINET_BINDANY:
4267 
4268 #ifdef notyet
4269 		/*
4270 		 * NCP privileges.
4271 		 */
4272 	case PRIV_NETNCP:
4273 
4274 		/*
4275 		 * SMB privileges.
4276 		 */
4277 	case PRIV_NETSMB:
4278 #endif
4279 
4280 	/*
4281 	 * No default: or deny here.
4282 	 * In case of no permit fall through to next switch().
4283 	 */
4284 		if (cred->cr_prison->pr_flags & PR_VNET)
4285 			return (0);
4286 	}
4287 #endif /* VIMAGE */
4288 
4289 	switch (priv) {
4290 		/*
4291 		 * Allow ktrace privileges for root in jail.
4292 		 */
4293 	case PRIV_KTRACE:
4294 
4295 		/*
4296 		 * Allow jailed processes to configure audit identity and
4297 		 * submit audit records (login, etc).  In the future we may
4298 		 * want to further refine the relationship between audit and
4299 		 * jail.
4300 		 */
4301 	case PRIV_AUDIT_GETAUDIT:
4302 	case PRIV_AUDIT_SETAUDIT:
4303 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETAUDIT)
4304 			return (0);
4305 		else
4306 			return (EPERM);
4307 #if 0
4308 	case PRIV_AUDIT_SUBMIT:
4309 #endif
4310 
4311 		/*
4312 		 * Allow jailed processes to manipulate process UNIX
4313 		 * credentials in any way they see fit.
4314 		 */
4315 	case PRIV_CRED_SETCRED:
4316 	case PRIV_CRED_SETUID:
4317 	case PRIV_CRED_SETEUID:
4318 	case PRIV_CRED_SETGID:
4319 	case PRIV_CRED_SETEGID:
4320 	case PRIV_CRED_SETGROUPS:
4321 	case PRIV_CRED_SETREUID:
4322 	case PRIV_CRED_SETREGID:
4323 	case PRIV_CRED_SETRESUID:
4324 	case PRIV_CRED_SETRESGID:
4325 
4326 		/*
4327 		 * Jail implements visibility constraints already, so allow
4328 		 * jailed root to override uid/gid-based constraints.
4329 		 */
4330 	case PRIV_SEEOTHERGIDS:
4331 	case PRIV_SEEOTHERUIDS:
4332 	case PRIV_SEEJAILPROC:
4333 
4334 		/*
4335 		 * Jail implements inter-process debugging limits already, so
4336 		 * allow jailed root various debugging privileges.
4337 		 */
4338 	case PRIV_DEBUG_DIFFCRED:
4339 	case PRIV_DEBUG_SUGID:
4340 	case PRIV_DEBUG_UNPRIV:
4341 	case PRIV_DEBUG_DIFFJAIL:
4342 
4343 		/*
4344 		 * Allow jail to set various resource limits and login
4345 		 * properties, and for now, exceed process resource limits.
4346 		 */
4347 	case PRIV_PROC_LIMIT:
4348 	case PRIV_PROC_SETLOGIN:
4349 	case PRIV_PROC_SETRLIMIT:
4350 
4351 		/*
4352 		 * Debuggers should work in jails.
4353 		 */
4354 	case PRIV_PROC_MEM_WRITE:
4355 
4356 		/*
4357 		 * System V and POSIX IPC privileges are granted in jail.
4358 		 */
4359 	case PRIV_IPC_READ:
4360 	case PRIV_IPC_WRITE:
4361 	case PRIV_IPC_ADMIN:
4362 	case PRIV_IPC_MSGSIZE:
4363 	case PRIV_MQ_ADMIN:
4364 
4365 		/*
4366 		 * Jail operations within a jail work on child jails.
4367 		 */
4368 	case PRIV_JAIL_ATTACH:
4369 	case PRIV_JAIL_SET:
4370 	case PRIV_JAIL_REMOVE:
4371 
4372 		/*
4373 		 * Jail implements its own inter-process limits, so allow
4374 		 * root processes in jail to change scheduling on other
4375 		 * processes in the same jail.  Likewise for signalling.
4376 		 */
4377 	case PRIV_SCHED_DIFFCRED:
4378 	case PRIV_SCHED_CPUSET:
4379 	case PRIV_SCHED_DIFFJAIL:
4380 	case PRIV_SIGNAL_DIFFCRED:
4381 	case PRIV_SIGNAL_SUGID:
4382 	case PRIV_SIGNAL_DIFFJAIL:
4383 
4384 		/*
4385 		 * Allow jailed processes to write to sysctls marked as jail
4386 		 * writable.
4387 		 */
4388 	case PRIV_SYSCTL_WRITEJAIL:
4389 
4390 		/*
4391 		 * Allow root in jail to manage a variety of quota
4392 		 * properties.  These should likely be conditional on a
4393 		 * configuration option.
4394 		 */
4395 	case PRIV_VFS_GETQUOTA:
4396 	case PRIV_VFS_SETQUOTA:
4397 
4398 		/*
4399 		 * Since Jail relies on chroot() to implement file system
4400 		 * protections, grant many VFS privileges to root in jail.
4401 		 * Be careful to exclude mount-related and NFS-related
4402 		 * privileges.
4403 		 */
4404 	case PRIV_VFS_READ:
4405 	case PRIV_VFS_WRITE:
4406 	case PRIV_VFS_ADMIN:
4407 	case PRIV_VFS_EXEC:
4408 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
4409 	case PRIV_VFS_CHFLAGS_DEV:
4410 	case PRIV_VFS_CHOWN:
4411 	case PRIV_VFS_CHROOT:
4412 	case PRIV_VFS_RETAINSUGID:
4413 	case PRIV_VFS_FCHROOT:
4414 	case PRIV_VFS_LINK:
4415 	case PRIV_VFS_SETGID:
4416 	case PRIV_VFS_STAT:
4417 	case PRIV_VFS_STICKYFILE:
4418 
4419 		/*
4420 		 * As in the non-jail case, non-root users are expected to be
4421 		 * able to read kernel/physical memory (provided /dev/[k]mem
4422 		 * exists in the jail and they have permission to access it).
4423 		 */
4424 	case PRIV_KMEM_READ:
4425 		return (0);
4426 
4427 		/*
4428 		 * Depending on the global setting, allow privilege of
4429 		 * setting system flags.
4430 		 */
4431 	case PRIV_VFS_SYSFLAGS:
4432 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4433 			return (0);
4434 		else
4435 			return (EPERM);
4436 
4437 		/*
4438 		 * Depending on the global setting, allow privilege of
4439 		 * mounting/unmounting file systems.
4440 		 */
4441 	case PRIV_VFS_MOUNT:
4442 	case PRIV_VFS_UNMOUNT:
4443 	case PRIV_VFS_MOUNT_NONUSER:
4444 	case PRIV_VFS_MOUNT_OWNER:
4445 		pr = cred->cr_prison;
4446 		prison_lock(pr);
4447 		if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4448 			error = 0;
4449 		else
4450 			error = EPERM;
4451 		prison_unlock(pr);
4452 		return (error);
4453 
4454 		/*
4455 		 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4456 		 * policy.  priv_check_cred will not specifically allow it, and
4457 		 * we may want a MAC policy to allow it.
4458 		 */
4459 	case PRIV_VFS_READ_DIR:
4460 		return (0);
4461 
4462 		/*
4463 		 * Conditionally allow privileged process in the jail to
4464 		 * manipulate filesystem extended attributes in the system
4465 		 * namespace.
4466 		 */
4467 	case PRIV_VFS_EXTATTR_SYSTEM:
4468 		if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4469 			return (0);
4470 		else
4471 			return (EPERM);
4472 
4473 		/*
4474 		 * Conditionnaly allow locking (unlocking) physical pages
4475 		 * in memory.
4476 		 */
4477 	case PRIV_VM_MLOCK:
4478 	case PRIV_VM_MUNLOCK:
4479 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4480 			return (0);
4481 		else
4482 			return (EPERM);
4483 
4484 		/*
4485 		 * Conditionally allow jailed root to bind reserved ports.
4486 		 */
4487 	case PRIV_NETINET_RESERVEDPORT:
4488 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4489 			return (0);
4490 		else
4491 			return (EPERM);
4492 
4493 		/*
4494 		 * Allow jailed root to reuse in-use ports.
4495 		 */
4496 	case PRIV_NETINET_REUSEPORT:
4497 		return (0);
4498 
4499 		/*
4500 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4501 		 */
4502 	case PRIV_NETINET_SETHDROPTS:
4503 		return (0);
4504 
4505 		/*
4506 		 * Conditionally allow creating raw sockets in jail.
4507 		 */
4508 	case PRIV_NETINET_RAW:
4509 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4510 			return (0);
4511 		else
4512 			return (EPERM);
4513 
4514 		/*
4515 		 * Since jail implements its own visibility limits on netstat
4516 		 * sysctls, allow getcred.  This allows identd to work in
4517 		 * jail.
4518 		 */
4519 	case PRIV_NETINET_GETCRED:
4520 		return (0);
4521 
4522 		/*
4523 		 * Allow jailed root to set loginclass.
4524 		 */
4525 	case PRIV_PROC_SETLOGINCLASS:
4526 		return (0);
4527 
4528 		/*
4529 		 * Do not allow a process inside a jail to read the kernel
4530 		 * message buffer unless explicitly permitted.
4531 		 */
4532 	case PRIV_MSGBUF:
4533 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4534 			return (0);
4535 		return (EPERM);
4536 
4537 		/*
4538 		 * Conditionally allow privileged process in the jail adjust
4539 		 * machine time.
4540 		 */
4541 	case PRIV_ADJTIME:
4542 	case PRIV_NTP_ADJTIME:
4543 		if (cred->cr_prison->pr_allow &
4544 		    (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4545 			return (0);
4546 		}
4547 		return (EPERM);
4548 
4549 		/*
4550 		 * Conditionally allow privileged process in the jail set
4551 		 * machine time.
4552 		 */
4553 	case PRIV_SETTIMEOFDAY:
4554 	case PRIV_CLOCK_SETTIME:
4555 		if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4556 			return (0);
4557 		else
4558 			return (EPERM);
4559 
4560 		/*
4561 		 * Conditionally allow privileged process in the jail to modify
4562 		 * the routing table.
4563 		 */
4564 	case PRIV_NET_ROUTE:
4565 		if (cred->cr_prison->pr_allow & PR_ALLOW_ROUTING)
4566 			return (0);
4567 		else
4568 			return (EPERM);
4569 
4570 	default:
4571 		/*
4572 		 * In all remaining cases, deny the privilege request.  This
4573 		 * includes almost all network privileges, many system
4574 		 * configuration privileges.
4575 		 */
4576 		return (EPERM);
4577 	}
4578 }
4579 
4580 /*
4581  * Return the part of pr2's name that is relative to pr1, or the whole name
4582  * if it does not directly follow.
4583  */
4584 
4585 char *
prison_name(struct prison * pr1,struct prison * pr2)4586 prison_name(struct prison *pr1, struct prison *pr2)
4587 {
4588 	char *name;
4589 
4590 	/* Jails see themselves as "0" (if they see themselves at all). */
4591 	if (pr1 == pr2)
4592 		return "0";
4593 	name = pr2->pr_name;
4594 	if (prison_ischild(pr1, pr2)) {
4595 		/*
4596 		 * pr1 isn't locked (and allprison_lock may not be either)
4597 		 * so its length can't be counted on.  But the number of dots
4598 		 * can be counted on - and counted.
4599 		 */
4600 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4601 			name = strchr(name, '.') + 1;
4602 	}
4603 	return (name);
4604 }
4605 
4606 /*
4607  * Return the part of pr2's path that is relative to pr1, or the whole path
4608  * if it does not directly follow.
4609  */
4610 static char *
prison_path(struct prison * pr1,struct prison * pr2)4611 prison_path(struct prison *pr1, struct prison *pr2)
4612 {
4613 	char *path1, *path2;
4614 	int len1;
4615 
4616 	path1 = pr1->pr_path;
4617 	path2 = pr2->pr_path;
4618 	if (!strcmp(path1, "/"))
4619 		return (path2);
4620 	len1 = strlen(path1);
4621 	if (strncmp(path1, path2, len1))
4622 		return (path2);
4623 	if (path2[len1] == '\0')
4624 		return "/";
4625 	if (path2[len1] == '/')
4626 		return (path2 + len1);
4627 	return (path2);
4628 }
4629 
4630 /*
4631  * Jail-related sysctls.
4632  */
4633 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4634     "Jails");
4635 
4636 #if defined(INET) || defined(INET6)
4637 /*
4638  * Copy address array to memory that would be then SYSCTL_OUT-ed.
4639  * sysctl_jail_list() helper.
4640  */
4641 static void
prison_ip_copyout(struct prison * pr,const pr_family_t af,void ** out,int * len)4642 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4643 {
4644 	const struct prison_ip *pip;
4645 	const size_t size = pr_families[af].size;
4646 
4647  again:
4648 	mtx_assert(&pr->pr_mtx, MA_OWNED);
4649 	if ((pip = pr->pr_addrs[af]) != NULL) {
4650 		if (*len < pip->ips) {
4651 			*len = pip->ips;
4652 			mtx_unlock(&pr->pr_mtx);
4653 			*out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4654 			mtx_lock(&pr->pr_mtx);
4655 			goto again;
4656 		}
4657 		bcopy(pip->pr_ip, *out, pip->ips * size);
4658 	}
4659 }
4660 #endif
4661 
4662 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4663 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4664 {
4665 	struct xprison *xp;
4666 	struct prison *pr, *cpr;
4667 #ifdef INET
4668 	struct in_addr *ip4 = NULL;
4669 	int ip4s = 0;
4670 #endif
4671 #ifdef INET6
4672 	struct in6_addr *ip6 = NULL;
4673 	int ip6s = 0;
4674 #endif
4675 	int descend, error;
4676 
4677 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4678 	pr = req->td->td_ucred->cr_prison;
4679 	error = 0;
4680 	sx_slock(&allprison_lock);
4681 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4682 		mtx_lock(&cpr->pr_mtx);
4683 #ifdef INET
4684 		prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4685 #endif
4686 #ifdef INET6
4687 		prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4688 #endif
4689 		bzero(xp, sizeof(*xp));
4690 		xp->pr_version = XPRISON_VERSION;
4691 		xp->pr_id = cpr->pr_id;
4692 		xp->pr_state = cpr->pr_state;
4693 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4694 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4695 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4696 #ifdef INET
4697 		xp->pr_ip4s = ip4s;
4698 #endif
4699 #ifdef INET6
4700 		xp->pr_ip6s = ip6s;
4701 #endif
4702 		mtx_unlock(&cpr->pr_mtx);
4703 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4704 		if (error)
4705 			break;
4706 #ifdef INET
4707 		if (xp->pr_ip4s > 0) {
4708 			error = SYSCTL_OUT(req, ip4,
4709 			    xp->pr_ip4s * sizeof(struct in_addr));
4710 			if (error)
4711 				break;
4712 		}
4713 #endif
4714 #ifdef INET6
4715 		if (xp->pr_ip6s > 0) {
4716 			error = SYSCTL_OUT(req, ip6,
4717 			    xp->pr_ip6s * sizeof(struct in6_addr));
4718 			if (error)
4719 				break;
4720 		}
4721 #endif
4722 	}
4723 	sx_sunlock(&allprison_lock);
4724 	free(xp, M_TEMP);
4725 #ifdef INET
4726 	free(ip4, M_TEMP);
4727 #endif
4728 #ifdef INET6
4729 	free(ip6, M_TEMP);
4730 #endif
4731 	return (error);
4732 }
4733 
4734 SYSCTL_OID(_security_jail, OID_AUTO, list,
4735     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4736     sysctl_jail_list, "S", "List of active jails");
4737 
4738 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)4739 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4740 {
4741 	int error, injail;
4742 
4743 	injail = jailed(req->td->td_ucred);
4744 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4745 
4746 	return (error);
4747 }
4748 
4749 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4750     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4751     sysctl_jail_jailed, "I", "Process in jail?");
4752 
4753 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)4754 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4755 {
4756 	int error, havevnet;
4757 #ifdef VIMAGE
4758 	struct ucred *cred = req->td->td_ucred;
4759 
4760 	havevnet = jailed(cred) && prison_owns_vnet(cred->cr_prison);
4761 #else
4762 	havevnet = 0;
4763 #endif
4764 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4765 
4766 	return (error);
4767 }
4768 
4769 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4770     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4771     sysctl_jail_vnet, "I", "Jail owns vnet?");
4772 
4773 #if defined(INET) || defined(INET6)
4774 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4775     &jail_max_af_ips, 0,
4776     "Number of IP addresses a jail may have at most per address family (deprecated)");
4777 #endif
4778 
4779 /*
4780  * Default parameters for jail(2) compatibility.  For historical reasons,
4781  * the sysctl names have varying similarity to the parameter names.  Prisons
4782  * just see their own parameters, and can't change them.
4783  */
4784 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)4785 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4786 {
4787 	int error, i;
4788 
4789 	/* Get the current flag value, and convert it to a boolean. */
4790 	if (req->td->td_ucred->cr_prison == &prison0) {
4791 		mtx_lock(&prison0.pr_mtx);
4792 		i = (jail_default_allow & arg2) != 0;
4793 		mtx_unlock(&prison0.pr_mtx);
4794 	} else
4795 		i = prison_allow(req->td->td_ucred, arg2);
4796 
4797 	if (arg1 != NULL)
4798 		i = !i;
4799 	error = sysctl_handle_int(oidp, &i, 0, req);
4800 	if (error || !req->newptr)
4801 		return (error);
4802 	i = i ? arg2 : 0;
4803 	if (arg1 != NULL)
4804 		i ^= arg2;
4805 	/*
4806 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4807 	 * for writing.
4808 	 */
4809 	mtx_lock(&prison0.pr_mtx);
4810 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4811 	mtx_unlock(&prison0.pr_mtx);
4812 	return (0);
4813 }
4814 
4815 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4816     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4817     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4818     "Processes in jail can set their hostnames (deprecated)");
4819 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4820     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4821     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4822     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4823 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4824     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4825     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4826     "Processes in jail can use System V IPC primitives (deprecated)");
4827 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4828     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4829     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4830     "Prison root can create raw sockets (deprecated)");
4831 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4832     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4833     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4834     "Processes in jail can alter system file flags (deprecated)");
4835 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4836     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4837     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4838     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4839 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
4840     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4841     NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
4842     "Processes in jail can lock/unlock physical pages in memory");
4843 
4844 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)4845 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4846 {
4847 	struct prison *pr;
4848 	int level, error;
4849 
4850 	pr = req->td->td_ucred->cr_prison;
4851 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4852 	error = sysctl_handle_int(oidp, &level, 0, req);
4853 	if (error || !req->newptr)
4854 		return (error);
4855 	*(int *)arg1 = level;
4856 	return (0);
4857 }
4858 
4859 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4860     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4861     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4862     sysctl_jail_default_level, "I",
4863     "Processes in jail cannot see all mounted file systems (deprecated)");
4864 
4865 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4866     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4867     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4868     sysctl_jail_default_level, "I",
4869     "Ruleset for the devfs filesystem in jail (deprecated)");
4870 
4871 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4872     "Limits and stats of child jails");
4873 
4874 static int
sysctl_jail_children(SYSCTL_HANDLER_ARGS)4875 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
4876 {
4877 	struct prison *pr;
4878 	int i;
4879 
4880 	pr = req->td->td_ucred->cr_prison;
4881 
4882 	switch (oidp->oid_kind & CTLTYPE) {
4883 	case CTLTYPE_INT:
4884 		i = *(int *)((char *)pr + arg2);
4885 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4886 	}
4887 
4888 	return (0);
4889 }
4890 
4891 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
4892     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4893     NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
4894     "I", "Maximum number of child jails");
4895 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
4896     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4897     NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
4898     "I", "Current number of child jails");
4899 
4900 /*
4901  * Nodes to describe jail parameters.  Maximum length of string parameters
4902  * is returned in the string itself, and the other parameters exist merely
4903  * to make themselves and their types known.
4904  */
4905 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4906     "Jail parameters");
4907 
4908 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)4909 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4910 {
4911 	int i;
4912 	long l;
4913 	size_t s;
4914 	char numbuf[12];
4915 
4916 	switch (oidp->oid_kind & CTLTYPE)
4917 	{
4918 	case CTLTYPE_LONG:
4919 	case CTLTYPE_ULONG:
4920 		l = 0;
4921 #ifdef SCTL_MASK32
4922 		if (!(req->flags & SCTL_MASK32))
4923 #endif
4924 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4925 	case CTLTYPE_INT:
4926 	case CTLTYPE_UINT:
4927 		i = 0;
4928 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4929 	case CTLTYPE_STRING:
4930 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4931 		return
4932 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4933 	case CTLTYPE_STRUCT:
4934 		s = (size_t)arg2;
4935 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4936 	}
4937 	return (0);
4938 }
4939 
4940 /*
4941  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4942  * jail creation time but cannot be changed in an existing jail.
4943  */
4944 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4945 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4946 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4947 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4948 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4949     "I", "Jail secure level");
4950 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4951     "Jail value for kern.osreldate and uname -K");
4952 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4953     "Jail value for kern.osrelease and uname -r");
4954 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4955     "I", "Jail cannot see all mounted file systems");
4956 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4957     "I", "Ruleset for in-jail devfs mounts");
4958 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4959     "B", "Jail persistence");
4960 #ifdef VIMAGE
4961 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4962     "E,jailsys", "Virtual network stack");
4963 #endif
4964 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4965     "B", "Jail is in the process of shutting down");
4966 
4967 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4968 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4969     "I", "Current number of child jails");
4970 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4971     "I", "Maximum number of child jails");
4972 
4973 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4974 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4975     "Jail hostname");
4976 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4977     "Jail NIS domainname");
4978 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4979     "Jail host UUID");
4980 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4981     "LU", "Jail host ID");
4982 
4983 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4984 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4985 
4986 #ifdef INET
4987 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4988     "Jail IPv4 address virtualization");
4989 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4990     "S,in_addr,a", "Jail IPv4 addresses");
4991 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4992     "B", "Do (not) use IPv4 source address selection rather than the "
4993     "primary jail IPv4 address.");
4994 #endif
4995 #ifdef INET6
4996 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4997     "Jail IPv6 address virtualization");
4998 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4999     "S,in6_addr,a", "Jail IPv6 addresses");
5000 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5001     "B", "Do (not) use IPv6 source address selection rather than the "
5002     "primary jail IPv6 address.");
5003 #endif
5004 
5005 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
5006 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
5007     "B", "Jail may set hostname");
5008 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
5009     "B", "Jail may use SYSV IPC");
5010 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
5011     "B", "Jail may create raw sockets");
5012 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
5013     "B", "Jail may alter system file flags");
5014 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
5015     "B", "Jail may set file quotas");
5016 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
5017     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
5018 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
5019     "B", "Jail may lock (unlock) physical pages in memory");
5020 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
5021     "B", "Jail may bind sockets to reserved ports");
5022 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
5023     "B", "Jail may read the kernel message buffer");
5024 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
5025     "B", "Unprivileged processes may use process debugging facilities");
5026 SYSCTL_JAIL_PARAM(_allow, unprivileged_parent_tampering,
5027     CTLTYPE_INT | CTLFLAG_RW, "B",
5028     "Unprivileged parent jail processes may tamper with same-uid processes"
5029     " (signal/debug/cpuset)");
5030 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
5031     "B", "Processes in jail with uid 0 have privilege");
5032 #ifdef VIMAGE
5033 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
5034     "B", "Mountd/nfsd may run in the jail");
5035 #endif
5036 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
5037     "B", "Jail may set system-level filesystem extended attributes");
5038 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
5039     "B", "Jail may adjust system time");
5040 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
5041     "B", "Jail may set system time");
5042 SYSCTL_JAIL_PARAM(_allow, routing, CTLTYPE_INT | CTLFLAG_RW,
5043     "B", "Jail may modify routing table");
5044 #ifdef AUDIT
5045 SYSCTL_JAIL_PARAM(_allow, setaudit, CTLTYPE_INT | CTLFLAG_RW,
5046     "B", "Jail may set and get audit session state");
5047 #endif
5048 
5049 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
5050 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
5051     "B", "Jail may mount/unmount jail-friendly file systems in general");
5052 
5053 /*
5054  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
5055  * its associated bit in the pr_allow bitmask, or zero if the parameter was
5056  * not created.
5057  */
5058 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)5059 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
5060     const char *descr)
5061 {
5062 	struct bool_flags *bf;
5063 	struct sysctl_oid *parent;
5064 	char *allow_name, *allow_noname, *allowed;
5065 #ifndef NO_SYSCTL_DESCR
5066 	char *descr_deprecated;
5067 #endif
5068 	u_int allow_flag;
5069 
5070 	if (prefix
5071 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
5072 		< 0 ||
5073 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
5074 		< 0
5075 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
5076 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
5077 		free(allow_name, M_PRISON);
5078 		return 0;
5079 	}
5080 
5081 	/*
5082 	 * See if this parameter has already beed added, i.e. a module was
5083 	 * previously loaded/unloaded.
5084 	 */
5085 	mtx_lock(&prison0.pr_mtx);
5086 	for (bf = pr_flag_allow;
5087 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5088 		atomic_load_int(&bf->flag) != 0;
5089 	     bf++) {
5090 		if (strcmp(bf->name, allow_name) == 0) {
5091 			allow_flag = bf->flag;
5092 			goto no_add;
5093 		}
5094 	}
5095 
5096 	/*
5097 	 * Find a free bit in pr_allow_all, failing if there are none
5098 	 * (which shouldn't happen as long as we keep track of how many
5099 	 * potential dynamic flags exist).
5100 	 */
5101 	for (allow_flag = 1;; allow_flag <<= 1) {
5102 		if (allow_flag == 0)
5103 			goto no_add;
5104 		if ((pr_allow_all & allow_flag) == 0)
5105 			break;
5106 	}
5107 
5108 	/* Note the parameter in the next open slot in pr_flag_allow. */
5109 	for (bf = pr_flag_allow; ; bf++) {
5110 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
5111 			/* This should never happen, but is not fatal. */
5112 			allow_flag = 0;
5113 			goto no_add;
5114 		}
5115 		if (atomic_load_int(&bf->flag) == 0)
5116 			break;
5117 	}
5118 	bf->name = allow_name;
5119 	bf->noname = allow_noname;
5120 	pr_allow_all |= allow_flag;
5121 	/*
5122 	 * prison0 always has permission for the new parameter.
5123 	 * Other jails must have it granted to them.
5124 	 */
5125 	prison0.pr_allow |= allow_flag;
5126 	/* The flag indicates a valid entry, so make sure it is set last. */
5127 	atomic_store_rel_int(&bf->flag, allow_flag);
5128 	mtx_unlock(&prison0.pr_mtx);
5129 
5130 	/*
5131 	 * Create sysctls for the parameter, and the back-compat global
5132 	 * permission.
5133 	 */
5134 	parent = prefix
5135 	    ? SYSCTL_ADD_NODE(NULL,
5136 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
5137 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
5138 	    : &sysctl___security_jail_param_allow;
5139 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
5140 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5141 	    NULL, 0, sysctl_jail_param, "B", descr);
5142 	if ((prefix
5143 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
5144 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
5145 #ifndef NO_SYSCTL_DESCR
5146 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
5147 		    descr);
5148 #endif
5149 		(void)SYSCTL_ADD_PROC(NULL,
5150 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
5151 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
5152 		    sysctl_jail_default_allow, "I", descr_deprecated);
5153 #ifndef NO_SYSCTL_DESCR
5154 		free(descr_deprecated, M_TEMP);
5155 #endif
5156 		free(allowed, M_TEMP);
5157 	}
5158 	return allow_flag;
5159 
5160  no_add:
5161 	mtx_unlock(&prison0.pr_mtx);
5162 	free(allow_name, M_PRISON);
5163 	free(allow_noname, M_PRISON);
5164 	return allow_flag;
5165 }
5166 
5167 /*
5168  * The VFS system will register jail-aware filesystems here.  They each get
5169  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
5170  * attempts to mount.
5171  */
5172 void
prison_add_vfs(struct vfsconf * vfsp)5173 prison_add_vfs(struct vfsconf *vfsp)
5174 {
5175 #ifdef NO_SYSCTL_DESCR
5176 
5177 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5178 	    NULL, NULL);
5179 #else
5180 	char *descr;
5181 
5182 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
5183 	    vfsp->vfc_name);
5184 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5185 	    NULL, descr);
5186 	free(descr, M_TEMP);
5187 #endif
5188 }
5189 
5190 #ifdef RACCT
5191 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)5192 prison_racct_foreach(void (*callback)(struct racct *racct,
5193     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
5194     void *arg2, void *arg3)
5195 {
5196 	struct prison_racct *prr;
5197 
5198 	ASSERT_RACCT_ENABLED();
5199 
5200 	sx_slock(&allprison_lock);
5201 	if (pre != NULL)
5202 		(pre)();
5203 	LIST_FOREACH(prr, &allprison_racct, prr_next)
5204 		(callback)(prr->prr_racct, arg2, arg3);
5205 	if (post != NULL)
5206 		(post)();
5207 	sx_sunlock(&allprison_lock);
5208 }
5209 
5210 static struct prison_racct *
prison_racct_find_locked(const char * name)5211 prison_racct_find_locked(const char *name)
5212 {
5213 	struct prison_racct *prr;
5214 
5215 	ASSERT_RACCT_ENABLED();
5216 	sx_assert(&allprison_lock, SA_XLOCKED);
5217 
5218 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
5219 		return (NULL);
5220 
5221 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
5222 		if (strcmp(name, prr->prr_name) != 0)
5223 			continue;
5224 
5225 		/* Found prison_racct with a matching name? */
5226 		prison_racct_hold(prr);
5227 		return (prr);
5228 	}
5229 
5230 	/* Add new prison_racct. */
5231 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
5232 	racct_create(&prr->prr_racct);
5233 
5234 	strcpy(prr->prr_name, name);
5235 	refcount_init(&prr->prr_refcount, 1);
5236 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
5237 
5238 	return (prr);
5239 }
5240 
5241 struct prison_racct *
prison_racct_find(const char * name)5242 prison_racct_find(const char *name)
5243 {
5244 	struct prison_racct *prr;
5245 
5246 	ASSERT_RACCT_ENABLED();
5247 
5248 	sx_xlock(&allprison_lock);
5249 	prr = prison_racct_find_locked(name);
5250 	sx_xunlock(&allprison_lock);
5251 	return (prr);
5252 }
5253 
5254 void
prison_racct_hold(struct prison_racct * prr)5255 prison_racct_hold(struct prison_racct *prr)
5256 {
5257 
5258 	ASSERT_RACCT_ENABLED();
5259 
5260 	refcount_acquire(&prr->prr_refcount);
5261 }
5262 
5263 static void
prison_racct_free_locked(struct prison_racct * prr)5264 prison_racct_free_locked(struct prison_racct *prr)
5265 {
5266 
5267 	ASSERT_RACCT_ENABLED();
5268 	sx_assert(&allprison_lock, SA_XLOCKED);
5269 
5270 	if (refcount_release(&prr->prr_refcount)) {
5271 		racct_destroy(&prr->prr_racct);
5272 		LIST_REMOVE(prr, prr_next);
5273 		free(prr, M_PRISON_RACCT);
5274 	}
5275 }
5276 
5277 void
prison_racct_free(struct prison_racct * prr)5278 prison_racct_free(struct prison_racct *prr)
5279 {
5280 
5281 	ASSERT_RACCT_ENABLED();
5282 	sx_assert(&allprison_lock, SA_UNLOCKED);
5283 
5284 	if (refcount_release_if_not_last(&prr->prr_refcount))
5285 		return;
5286 
5287 	sx_xlock(&allprison_lock);
5288 	prison_racct_free_locked(prr);
5289 	sx_xunlock(&allprison_lock);
5290 }
5291 
5292 static void
prison_racct_attach(struct prison * pr)5293 prison_racct_attach(struct prison *pr)
5294 {
5295 	struct prison_racct *prr;
5296 
5297 	ASSERT_RACCT_ENABLED();
5298 	sx_assert(&allprison_lock, SA_XLOCKED);
5299 
5300 	prr = prison_racct_find_locked(pr->pr_name);
5301 	KASSERT(prr != NULL, ("cannot find prison_racct"));
5302 
5303 	pr->pr_prison_racct = prr;
5304 }
5305 
5306 /*
5307  * Handle jail renaming.  From the racct point of view, renaming means
5308  * moving from one prison_racct to another.
5309  */
5310 static void
prison_racct_modify(struct prison * pr)5311 prison_racct_modify(struct prison *pr)
5312 {
5313 #ifdef RCTL
5314 	struct proc *p;
5315 	struct ucred *cred;
5316 #endif
5317 	struct prison_racct *oldprr;
5318 
5319 	ASSERT_RACCT_ENABLED();
5320 
5321 	sx_slock(&allproc_lock);
5322 	sx_xlock(&allprison_lock);
5323 
5324 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
5325 		sx_xunlock(&allprison_lock);
5326 		sx_sunlock(&allproc_lock);
5327 		return;
5328 	}
5329 
5330 	oldprr = pr->pr_prison_racct;
5331 	pr->pr_prison_racct = NULL;
5332 
5333 	prison_racct_attach(pr);
5334 
5335 	/*
5336 	 * Move resource utilisation records.
5337 	 */
5338 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
5339 
5340 #ifdef RCTL
5341 	/*
5342 	 * Force rctl to reattach rules to processes.
5343 	 */
5344 	FOREACH_PROC_IN_SYSTEM(p) {
5345 		PROC_LOCK(p);
5346 		cred = crhold(p->p_ucred);
5347 		PROC_UNLOCK(p);
5348 		rctl_proc_ucred_changed(p, cred);
5349 		crfree(cred);
5350 	}
5351 #endif
5352 
5353 	sx_sunlock(&allproc_lock);
5354 	prison_racct_free_locked(oldprr);
5355 	sx_xunlock(&allprison_lock);
5356 }
5357 
5358 static void
prison_racct_detach(struct prison * pr)5359 prison_racct_detach(struct prison *pr)
5360 {
5361 
5362 	ASSERT_RACCT_ENABLED();
5363 	sx_assert(&allprison_lock, SA_UNLOCKED);
5364 
5365 	if (pr->pr_prison_racct == NULL)
5366 		return;
5367 	prison_racct_free(pr->pr_prison_racct);
5368 	pr->pr_prison_racct = NULL;
5369 }
5370 #endif /* RACCT */
5371 
5372 /*
5373  * Submit a knote for a prison, locking if necessary.
5374  */
5375 static void
prison_knote(struct prison * pr,long hint)5376 prison_knote(struct prison *pr, long hint)
5377 {
5378 	int locked;
5379 
5380 	locked = mtx_owned(&pr->pr_mtx);
5381 	if (!locked)
5382 		mtx_lock(&pr->pr_mtx);
5383 	KNOTE_LOCKED(pr->pr_klist, hint);
5384 	jaildesc_knote(pr, hint);
5385 	if (!locked)
5386 		mtx_unlock(&pr->pr_mtx);
5387 }
5388 
5389 #ifdef DDB
5390 
5391 static void
db_show_prison(struct prison * pr)5392 db_show_prison(struct prison *pr)
5393 {
5394 	struct bool_flags *bf;
5395 	struct jailsys_flags *jsf;
5396 #if defined(INET) || defined(INET6)
5397 	int ii;
5398 	struct prison_ip *pip;
5399 #endif
5400 	unsigned f;
5401 #ifdef INET
5402 	char ip4buf[INET_ADDRSTRLEN];
5403 #endif
5404 #ifdef INET6
5405 	char ip6buf[INET6_ADDRSTRLEN];
5406 #endif
5407 
5408 	db_printf("prison %p:\n", pr);
5409 	db_printf(" jid             = %d\n", pr->pr_id);
5410 	db_printf(" name            = %s\n", pr->pr_name);
5411 	db_printf(" parent          = %p\n", pr->pr_parent);
5412 	db_printf(" ref             = %d\n", pr->pr_ref);
5413 	db_printf(" uref            = %d\n", pr->pr_uref);
5414 	db_printf(" state           = %s\n",
5415 	    pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5416 	    pr->pr_state == PRISON_STATE_DYING ? "dying" :
5417 	    "invalid");
5418 	db_printf(" path            = %s\n", pr->pr_path);
5419 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
5420 	    ? pr->pr_cpuset->cs_id : -1);
5421 #ifdef VIMAGE
5422 	db_printf(" vnet            = %p\n", pr->pr_vnet);
5423 #endif
5424 	db_printf(" root            = %p\n", pr->pr_root);
5425 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
5426 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
5427 	db_printf(" children.max    = %d\n", pr->pr_childmax);
5428 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
5429 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
5430 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
5431 	db_printf(" flags           = 0x%x", pr->pr_flags);
5432 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5433 		if (pr->pr_flags & bf->flag)
5434 			db_printf(" %s", bf->name);
5435 	for (jsf = pr_flag_jailsys;
5436 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5437 	     jsf++) {
5438 		f = pr->pr_flags & (jsf->disable | jsf->new);
5439 		db_printf(" %-16s= %s\n", jsf->name,
5440 		    (f != 0 && f == jsf->disable) ? "disable"
5441 		    : (f == jsf->new) ? "new"
5442 		    : "inherit");
5443 	}
5444 	db_printf(" allow           = 0x%x", pr->pr_allow);
5445 	for (bf = pr_flag_allow;
5446 	     bf < pr_flag_allow + nitems(pr_flag_allow) &&
5447 		atomic_load_int(&bf->flag) != 0;
5448 	     bf++)
5449 		if (pr->pr_allow & bf->flag)
5450 			db_printf(" %s", bf->name);
5451 	db_printf("\n");
5452 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
5453 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
5454 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
5455 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
5456 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
5457 #ifdef INET
5458 	if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5459 		db_printf(" ip4s            = %d\n", pip->ips);
5460 		for (ii = 0; ii < pip->ips; ii++)
5461 			db_printf(" %s %s\n",
5462 			    ii == 0 ? "ip4.addr        =" : "                 ",
5463 			    inet_ntoa_r(
5464 			    *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5465 			    ip4buf));
5466 	}
5467 #endif
5468 #ifdef INET6
5469 	if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5470 		db_printf(" ip6s            = %d\n", pip->ips);
5471 		for (ii = 0; ii < pip->ips; ii++)
5472 			db_printf(" %s %s\n",
5473 			    ii == 0 ? "ip6.addr        =" : "                 ",
5474 			    ip6_sprintf(ip6buf,
5475 			    (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5476 	}
5477 #endif
5478 }
5479 
DB_SHOW_COMMAND(prison,db_show_prison_command)5480 DB_SHOW_COMMAND(prison, db_show_prison_command)
5481 {
5482 	struct prison *pr;
5483 
5484 	if (!have_addr) {
5485 		/*
5486 		 * Show all prisons in the list, and prison0 which is not
5487 		 * listed.
5488 		 */
5489 		db_show_prison(&prison0);
5490 		if (!db_pager_quit) {
5491 			TAILQ_FOREACH(pr, &allprison, pr_list) {
5492 				db_show_prison(pr);
5493 				if (db_pager_quit)
5494 					break;
5495 			}
5496 		}
5497 		return;
5498 	}
5499 
5500 	if (addr == 0)
5501 		pr = &prison0;
5502 	else {
5503 		/* Look for a prison with the ID and with references. */
5504 		TAILQ_FOREACH(pr, &allprison, pr_list)
5505 			if (pr->pr_id == addr && pr->pr_ref > 0)
5506 				break;
5507 		if (pr == NULL)
5508 			/* Look again, without requiring a reference. */
5509 			TAILQ_FOREACH(pr, &allprison, pr_list)
5510 				if (pr->pr_id == addr)
5511 					break;
5512 		if (pr == NULL)
5513 			/* Assume address points to a valid prison. */
5514 			pr = (struct prison *)addr;
5515 	}
5516 	db_show_prison(pr);
5517 }
5518 
5519 #endif /* DDB */
5520