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