1 /*-
2 * Copyright (c) 1999 Poul-Henning Kamp.
3 * Copyright (c) 2008 Bjoern A. Zeeb.
4 * Copyright (c) 2009 James Gritton.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD: stable/10/sys/kern/kern_jail.c 339410 2018-10-17 16:17:56Z jamie $");
31
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.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/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63
64 #include <net/if.h>
65 #include <net/vnet.h>
66
67 #include <netinet/in.h>
68
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #ifdef INET6
72 #include <netinet6/in6_var.h>
73 #endif /* INET6 */
74 #endif /* DDB */
75
76 #include <security/mac/mac_framework.h>
77
78 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
79
80 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
81 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
82
83 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
84 #ifdef INET
85 #ifdef INET6
86 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
87 #else
88 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
89 #endif
90 #else /* !INET */
91 #ifdef INET6
92 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
93 #else
94 #define _PR_IP_SADDRSEL 0
95 #endif
96 #endif
97
98 /* prison0 describes what is "real" about the system. */
99 struct prison prison0 = {
100 .pr_id = 0,
101 .pr_name = "0",
102 .pr_ref = 1,
103 .pr_uref = 1,
104 .pr_path = "/",
105 .pr_securelevel = -1,
106 .pr_devfs_rsnum = 0,
107 .pr_childmax = JAIL_MAX,
108 .pr_hostuuid = DEFAULT_HOSTUUID,
109 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
110 #ifdef VIMAGE
111 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
112 #else
113 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
114 #endif
115 .pr_allow = PR_ALLOW_ALL,
116 };
117 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
118
119 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
120 struct sx allprison_lock;
121 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
122 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
123 LIST_HEAD(, prison_racct) allprison_racct;
124 int lastprid = 0;
125
126 static int do_jail_attach(struct thread *td, struct prison *pr);
127 static void prison_complete(void *context, int pending);
128 static void prison_deref(struct prison *pr, int flags);
129 static char *prison_path(struct prison *pr1, struct prison *pr2);
130 static void prison_remove_one(struct prison *pr);
131 #ifdef RACCT
132 static void prison_racct_attach(struct prison *pr);
133 static void prison_racct_modify(struct prison *pr);
134 static void prison_racct_detach(struct prison *pr);
135 #endif
136 #ifdef INET
137 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
138 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
139 #endif
140 #ifdef INET6
141 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
142 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
143 #endif
144
145 /* Flags for prison_deref */
146 #define PD_DEREF 0x01
147 #define PD_DEUREF 0x02
148 #define PD_LOCKED 0x04
149 #define PD_LIST_SLOCKED 0x08
150 #define PD_LIST_XLOCKED 0x10
151
152 /*
153 * Parameter names corresponding to PR_* flag values. Size values are for kvm
154 * as we cannot figure out the size of a sparse array, or an array without a
155 * terminating entry.
156 */
157 static char *pr_flag_names[] = {
158 [0] = "persist",
159 #ifdef INET
160 [7] = "ip4.saddrsel",
161 #endif
162 #ifdef INET6
163 [8] = "ip6.saddrsel",
164 #endif
165 };
166 const size_t pr_flag_names_size = sizeof(pr_flag_names);
167
168 static char *pr_flag_nonames[] = {
169 [0] = "nopersist",
170 #ifdef INET
171 [7] = "ip4.nosaddrsel",
172 #endif
173 #ifdef INET6
174 [8] = "ip6.nosaddrsel",
175 #endif
176 };
177 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
178
179 struct jailsys_flags {
180 const char *name;
181 unsigned disable;
182 unsigned new;
183 } pr_flag_jailsys[] = {
184 { "host", 0, PR_HOST },
185 #ifdef VIMAGE
186 { "vnet", 0, PR_VNET },
187 #endif
188 #ifdef INET
189 { "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
190 #endif
191 #ifdef INET6
192 { "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
193 #endif
194 };
195 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
196
197 static char *pr_allow_names[] = {
198 "allow.set_hostname",
199 "allow.sysvipc",
200 "allow.raw_sockets",
201 "allow.chflags",
202 "allow.mount",
203 "allow.quotas",
204 "allow.socket_af",
205 "allow.mount.devfs",
206 "allow.mount.nullfs",
207 "allow.mount.zfs",
208 "allow.mount.procfs",
209 "allow.mount.tmpfs",
210 "allow.mount.fdescfs",
211 "allow.mount.linprocfs",
212 "allow.mount.linsysfs",
213 };
214 const size_t pr_allow_names_size = sizeof(pr_allow_names);
215
216 static char *pr_allow_nonames[] = {
217 "allow.noset_hostname",
218 "allow.nosysvipc",
219 "allow.noraw_sockets",
220 "allow.nochflags",
221 "allow.nomount",
222 "allow.noquotas",
223 "allow.nosocket_af",
224 "allow.mount.nodevfs",
225 "allow.mount.nonullfs",
226 "allow.mount.nozfs",
227 "allow.mount.noprocfs",
228 "allow.mount.notmpfs",
229 "allow.mount.nofdescfs",
230 "allow.mount.nolinprocfs",
231 "allow.mount.nolinsysfs",
232 };
233 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
234
235 #define JAIL_DEFAULT_ALLOW PR_ALLOW_SET_HOSTNAME
236 #define JAIL_DEFAULT_ENFORCE_STATFS 2
237 #define JAIL_DEFAULT_DEVFS_RSNUM 0
238 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
239 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
240 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
241 #if defined(INET) || defined(INET6)
242 static unsigned jail_max_af_ips = 255;
243 #endif
244
245 /*
246 * Initialize the parts of prison0 that can't be static-initialized with
247 * constants. This is called from proc0_init() after creating thread0 cpuset.
248 */
249 void
prison0_init(void)250 prison0_init(void)
251 {
252
253 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
254 prison0.pr_osreldate = osreldate;
255 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
256 }
257
258 #ifdef INET
259 static int
qcmp_v4(const void * ip1,const void * ip2)260 qcmp_v4(const void *ip1, const void *ip2)
261 {
262 in_addr_t iaa, iab;
263
264 /*
265 * We need to compare in HBO here to get the list sorted as expected
266 * by the result of the code. Sorting NBO addresses gives you
267 * interesting results. If you do not understand, do not try.
268 */
269 iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
270 iab = ntohl(((const struct in_addr *)ip2)->s_addr);
271
272 /*
273 * Do not simply return the difference of the two numbers, the int is
274 * not wide enough.
275 */
276 if (iaa > iab)
277 return (1);
278 else if (iaa < iab)
279 return (-1);
280 else
281 return (0);
282 }
283 #endif
284
285 #ifdef INET6
286 static int
qcmp_v6(const void * ip1,const void * ip2)287 qcmp_v6(const void *ip1, const void *ip2)
288 {
289 const struct in6_addr *ia6a, *ia6b;
290 int i, rc;
291
292 ia6a = (const struct in6_addr *)ip1;
293 ia6b = (const struct in6_addr *)ip2;
294
295 rc = 0;
296 for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
297 if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
298 rc = 1;
299 else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
300 rc = -1;
301 }
302 return (rc);
303 }
304 #endif
305
306 /*
307 * struct jail_args {
308 * struct jail *jail;
309 * };
310 */
311 int
sys_jail(struct thread * td,struct jail_args * uap)312 sys_jail(struct thread *td, struct jail_args *uap)
313 {
314 uint32_t version;
315 int error;
316 struct jail j;
317
318 error = copyin(uap->jail, &version, sizeof(uint32_t));
319 if (error)
320 return (error);
321
322 switch (version) {
323 case 0:
324 {
325 struct jail_v0 j0;
326
327 /* FreeBSD single IPv4 jails. */
328 bzero(&j, sizeof(struct jail));
329 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
330 if (error)
331 return (error);
332 j.version = j0.version;
333 j.path = j0.path;
334 j.hostname = j0.hostname;
335 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
336 break;
337 }
338
339 case 1:
340 /*
341 * Version 1 was used by multi-IPv4 jail implementations
342 * that never made it into the official kernel.
343 */
344 return (EINVAL);
345
346 case 2: /* JAIL_API_VERSION */
347 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
348 error = copyin(uap->jail, &j, sizeof(struct jail));
349 if (error)
350 return (error);
351 break;
352
353 default:
354 /* Sci-Fi jails are not supported, sorry. */
355 return (EINVAL);
356 }
357 return (kern_jail(td, &j));
358 }
359
360 int
kern_jail(struct thread * td,struct jail * j)361 kern_jail(struct thread *td, struct jail *j)
362 {
363 struct iovec optiov[2 * (4
364 + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
365 #ifdef INET
366 + 1
367 #endif
368 #ifdef INET6
369 + 1
370 #endif
371 )];
372 struct uio opt;
373 char *u_path, *u_hostname, *u_name;
374 #ifdef INET
375 uint32_t ip4s;
376 struct in_addr *u_ip4;
377 #endif
378 #ifdef INET6
379 struct in6_addr *u_ip6;
380 #endif
381 size_t tmplen;
382 int error, enforce_statfs, fi;
383
384 bzero(&optiov, sizeof(optiov));
385 opt.uio_iov = optiov;
386 opt.uio_iovcnt = 0;
387 opt.uio_offset = -1;
388 opt.uio_resid = -1;
389 opt.uio_segflg = UIO_SYSSPACE;
390 opt.uio_rw = UIO_READ;
391 opt.uio_td = td;
392
393 /* Set permissions for top-level jails from sysctls. */
394 if (!jailed(td->td_ucred)) {
395 for (fi = 0; fi < sizeof(pr_allow_names) /
396 sizeof(pr_allow_names[0]); fi++) {
397 optiov[opt.uio_iovcnt].iov_base =
398 (jail_default_allow & (1 << fi))
399 ? pr_allow_names[fi] : pr_allow_nonames[fi];
400 optiov[opt.uio_iovcnt].iov_len =
401 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
402 opt.uio_iovcnt += 2;
403 }
404 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
405 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
406 opt.uio_iovcnt++;
407 enforce_statfs = jail_default_enforce_statfs;
408 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
409 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
410 opt.uio_iovcnt++;
411 }
412
413 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
414 #ifdef INET
415 ip4s = (j->version == 0) ? 1 : j->ip4s;
416 if (ip4s > jail_max_af_ips)
417 return (EINVAL);
418 tmplen += ip4s * sizeof(struct in_addr);
419 #else
420 if (j->ip4s > 0)
421 return (EINVAL);
422 #endif
423 #ifdef INET6
424 if (j->ip6s > jail_max_af_ips)
425 return (EINVAL);
426 tmplen += j->ip6s * sizeof(struct in6_addr);
427 #else
428 if (j->ip6s > 0)
429 return (EINVAL);
430 #endif
431 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
432 u_hostname = u_path + MAXPATHLEN;
433 u_name = u_hostname + MAXHOSTNAMELEN;
434 #ifdef INET
435 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
436 #endif
437 #ifdef INET6
438 #ifdef INET
439 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
440 #else
441 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
442 #endif
443 #endif
444 optiov[opt.uio_iovcnt].iov_base = "path";
445 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
446 opt.uio_iovcnt++;
447 optiov[opt.uio_iovcnt].iov_base = u_path;
448 error = copyinstr(j->path, u_path, MAXPATHLEN,
449 &optiov[opt.uio_iovcnt].iov_len);
450 if (error) {
451 free(u_path, M_TEMP);
452 return (error);
453 }
454 opt.uio_iovcnt++;
455 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
456 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
457 opt.uio_iovcnt++;
458 optiov[opt.uio_iovcnt].iov_base = u_hostname;
459 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
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 if (j->jailname != NULL) {
467 optiov[opt.uio_iovcnt].iov_base = "name";
468 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
469 opt.uio_iovcnt++;
470 optiov[opt.uio_iovcnt].iov_base = u_name;
471 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
472 &optiov[opt.uio_iovcnt].iov_len);
473 if (error) {
474 free(u_path, M_TEMP);
475 return (error);
476 }
477 opt.uio_iovcnt++;
478 }
479 #ifdef INET
480 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
481 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
482 opt.uio_iovcnt++;
483 optiov[opt.uio_iovcnt].iov_base = u_ip4;
484 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
485 if (j->version == 0)
486 u_ip4->s_addr = j->ip4s;
487 else {
488 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
489 if (error) {
490 free(u_path, M_TEMP);
491 return (error);
492 }
493 }
494 opt.uio_iovcnt++;
495 #endif
496 #ifdef INET6
497 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
498 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
499 opt.uio_iovcnt++;
500 optiov[opt.uio_iovcnt].iov_base = u_ip6;
501 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
502 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
503 if (error) {
504 free(u_path, M_TEMP);
505 return (error);
506 }
507 opt.uio_iovcnt++;
508 #endif
509 KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
510 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
511 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
512 free(u_path, M_TEMP);
513 return (error);
514 }
515
516
517 /*
518 * struct jail_set_args {
519 * struct iovec *iovp;
520 * unsigned int iovcnt;
521 * int flags;
522 * };
523 */
524 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)525 sys_jail_set(struct thread *td, struct jail_set_args *uap)
526 {
527 struct uio *auio;
528 int error;
529
530 /* Check that we have an even number of iovecs. */
531 if (uap->iovcnt & 1)
532 return (EINVAL);
533
534 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
535 if (error)
536 return (error);
537 error = kern_jail_set(td, auio, uap->flags);
538 free(auio, M_IOV);
539 return (error);
540 }
541
542 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)543 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
544 {
545 struct nameidata nd;
546 #ifdef INET
547 struct in_addr *ip4;
548 #endif
549 #ifdef INET6
550 struct in6_addr *ip6;
551 #endif
552 struct vfsopt *opt;
553 struct vfsoptlist *opts;
554 struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
555 struct vnode *root;
556 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
557 char *g_path, *osrelstr;
558 #if defined(INET) || defined(INET6)
559 struct prison *tppr;
560 void *op;
561 #endif
562 unsigned long hid;
563 size_t namelen, onamelen, pnamelen;
564 int born, created, cuflags, descend, enforce;
565 int error, errmsg_len, errmsg_pos;
566 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
567 int fi, jid, jsys, len, level;
568 int childmax, osreldt, rsnum, slevel;
569 int fullpath_disabled;
570 #if defined(INET) || defined(INET6)
571 int ii, ij;
572 #endif
573 #ifdef INET
574 int ip4s, redo_ip4;
575 #endif
576 #ifdef INET6
577 int ip6s, redo_ip6;
578 #endif
579 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
580 unsigned tallow;
581 char numbuf[12];
582
583 error = priv_check(td, PRIV_JAIL_SET);
584 if (!error && (flags & JAIL_ATTACH))
585 error = priv_check(td, PRIV_JAIL_ATTACH);
586 if (error)
587 return (error);
588 mypr = td->td_ucred->cr_prison;
589 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
590 return (EPERM);
591 if (flags & ~JAIL_SET_MASK)
592 return (EINVAL);
593
594 /*
595 * Check all the parameters before committing to anything. Not all
596 * errors can be caught early, but we may as well try. Also, this
597 * takes care of some expensive stuff (path lookup) before getting
598 * the allprison lock.
599 *
600 * XXX Jails are not filesystems, and jail parameters are not mount
601 * options. But it makes more sense to re-use the vfsopt code
602 * than duplicate it under a different name.
603 */
604 error = vfs_buildopts(optuio, &opts);
605 if (error)
606 return (error);
607 #ifdef INET
608 ip4 = NULL;
609 #endif
610 #ifdef INET6
611 ip6 = NULL;
612 #endif
613 g_path = NULL;
614
615 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
616 if (!cuflags) {
617 error = EINVAL;
618 vfs_opterror(opts, "no valid operation (create or update)");
619 goto done_errmsg;
620 }
621
622 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
623 if (error == ENOENT)
624 jid = 0;
625 else if (error != 0)
626 goto done_free;
627
628 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
629 if (error == ENOENT)
630 gotslevel = 0;
631 else if (error != 0)
632 goto done_free;
633 else
634 gotslevel = 1;
635
636 error =
637 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
638 if (error == ENOENT)
639 gotchildmax = 0;
640 else if (error != 0)
641 goto done_free;
642 else
643 gotchildmax = 1;
644
645 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
646 if (error == ENOENT)
647 gotenforce = 0;
648 else if (error != 0)
649 goto done_free;
650 else if (enforce < 0 || enforce > 2) {
651 error = EINVAL;
652 goto done_free;
653 } else
654 gotenforce = 1;
655
656 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
657 if (error == ENOENT)
658 gotrsnum = 0;
659 else if (error != 0)
660 goto done_free;
661 else
662 gotrsnum = 1;
663
664 pr_flags = ch_flags = 0;
665 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
666 fi++) {
667 if (pr_flag_names[fi] == NULL)
668 continue;
669 vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
670 vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
671 }
672 ch_flags |= pr_flags;
673 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
674 fi++) {
675 error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
676 sizeof(jsys));
677 if (error == ENOENT)
678 continue;
679 if (error != 0)
680 goto done_free;
681 switch (jsys) {
682 case JAIL_SYS_DISABLE:
683 if (!pr_flag_jailsys[fi].disable) {
684 error = EINVAL;
685 goto done_free;
686 }
687 pr_flags |= pr_flag_jailsys[fi].disable;
688 break;
689 case JAIL_SYS_NEW:
690 pr_flags |= pr_flag_jailsys[fi].new;
691 break;
692 case JAIL_SYS_INHERIT:
693 break;
694 default:
695 error = EINVAL;
696 goto done_free;
697 }
698 ch_flags |=
699 pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
700 }
701 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
702 && !(pr_flags & PR_PERSIST)) {
703 error = EINVAL;
704 vfs_opterror(opts, "new jail must persist or attach");
705 goto done_errmsg;
706 }
707 #ifdef VIMAGE
708 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
709 error = EINVAL;
710 vfs_opterror(opts, "vnet cannot be changed after creation");
711 goto done_errmsg;
712 }
713 #endif
714 #ifdef INET
715 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
716 error = EINVAL;
717 vfs_opterror(opts, "ip4 cannot be changed after creation");
718 goto done_errmsg;
719 }
720 #endif
721 #ifdef INET6
722 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
723 error = EINVAL;
724 vfs_opterror(opts, "ip6 cannot be changed after creation");
725 goto done_errmsg;
726 }
727 #endif
728
729 pr_allow = ch_allow = 0;
730 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
731 fi++) {
732 vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
733 vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
734 }
735 ch_allow |= pr_allow;
736
737 error = vfs_getopt(opts, "name", (void **)&name, &len);
738 if (error == ENOENT)
739 name = NULL;
740 else if (error != 0)
741 goto done_free;
742 else {
743 if (len == 0 || name[len - 1] != '\0') {
744 error = EINVAL;
745 goto done_free;
746 }
747 if (len > MAXHOSTNAMELEN) {
748 error = ENAMETOOLONG;
749 goto done_free;
750 }
751 }
752
753 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
754 if (error == ENOENT)
755 host = NULL;
756 else if (error != 0)
757 goto done_free;
758 else {
759 ch_flags |= PR_HOST;
760 pr_flags |= PR_HOST;
761 if (len == 0 || host[len - 1] != '\0') {
762 error = EINVAL;
763 goto done_free;
764 }
765 if (len > MAXHOSTNAMELEN) {
766 error = ENAMETOOLONG;
767 goto done_free;
768 }
769 }
770
771 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
772 if (error == ENOENT)
773 domain = NULL;
774 else if (error != 0)
775 goto done_free;
776 else {
777 ch_flags |= PR_HOST;
778 pr_flags |= PR_HOST;
779 if (len == 0 || domain[len - 1] != '\0') {
780 error = EINVAL;
781 goto done_free;
782 }
783 if (len > MAXHOSTNAMELEN) {
784 error = ENAMETOOLONG;
785 goto done_free;
786 }
787 }
788
789 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
790 if (error == ENOENT)
791 uuid = NULL;
792 else if (error != 0)
793 goto done_free;
794 else {
795 ch_flags |= PR_HOST;
796 pr_flags |= PR_HOST;
797 if (len == 0 || uuid[len - 1] != '\0') {
798 error = EINVAL;
799 goto done_free;
800 }
801 if (len > HOSTUUIDLEN) {
802 error = ENAMETOOLONG;
803 goto done_free;
804 }
805 }
806
807 #ifdef COMPAT_FREEBSD32
808 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
809 uint32_t hid32;
810
811 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
812 hid = hid32;
813 } else
814 #endif
815 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
816 if (error == ENOENT)
817 gothid = 0;
818 else if (error != 0)
819 goto done_free;
820 else {
821 gothid = 1;
822 ch_flags |= PR_HOST;
823 pr_flags |= PR_HOST;
824 }
825
826 #ifdef INET
827 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
828 if (error == ENOENT)
829 ip4s = 0;
830 else if (error != 0)
831 goto done_free;
832 else if (ip4s & (sizeof(*ip4) - 1)) {
833 error = EINVAL;
834 goto done_free;
835 } else {
836 ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
837 if (ip4s == 0)
838 pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
839 else {
840 pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
841 ip4s /= sizeof(*ip4);
842 if (ip4s > jail_max_af_ips) {
843 error = EINVAL;
844 vfs_opterror(opts, "too many IPv4 addresses");
845 goto done_errmsg;
846 }
847 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
848 bcopy(op, ip4, ip4s * sizeof(*ip4));
849 /*
850 * IP addresses are all sorted but ip[0] to preserve
851 * the primary IP address as given from userland.
852 * This special IP is used for unbound outgoing
853 * connections as well for "loopback" traffic in case
854 * source address selection cannot find any more fitting
855 * address to connect from.
856 */
857 if (ip4s > 1)
858 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
859 /*
860 * Check for duplicate addresses and do some simple
861 * zero and broadcast checks. If users give other bogus
862 * addresses it is their problem.
863 *
864 * We do not have to care about byte order for these
865 * checks so we will do them in NBO.
866 */
867 for (ii = 0; ii < ip4s; ii++) {
868 if (ip4[ii].s_addr == INADDR_ANY ||
869 ip4[ii].s_addr == INADDR_BROADCAST) {
870 error = EINVAL;
871 goto done_free;
872 }
873 if ((ii+1) < ip4s &&
874 (ip4[0].s_addr == ip4[ii+1].s_addr ||
875 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
876 error = EINVAL;
877 goto done_free;
878 }
879 }
880 }
881 }
882 #endif
883
884 #ifdef INET6
885 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
886 if (error == ENOENT)
887 ip6s = 0;
888 else if (error != 0)
889 goto done_free;
890 else if (ip6s & (sizeof(*ip6) - 1)) {
891 error = EINVAL;
892 goto done_free;
893 } else {
894 ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
895 if (ip6s == 0)
896 pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
897 else {
898 pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
899 ip6s /= sizeof(*ip6);
900 if (ip6s > jail_max_af_ips) {
901 error = EINVAL;
902 vfs_opterror(opts, "too many IPv6 addresses");
903 goto done_errmsg;
904 }
905 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
906 bcopy(op, ip6, ip6s * sizeof(*ip6));
907 if (ip6s > 1)
908 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
909 for (ii = 0; ii < ip6s; ii++) {
910 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
911 error = EINVAL;
912 goto done_free;
913 }
914 if ((ii+1) < ip6s &&
915 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
916 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
917 {
918 error = EINVAL;
919 goto done_free;
920 }
921 }
922 }
923 }
924 #endif
925
926 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
927 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
928 error = EINVAL;
929 vfs_opterror(opts,
930 "vnet jails cannot have IP address restrictions");
931 goto done_errmsg;
932 }
933 #endif
934
935 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
936 if (error == ENOENT)
937 osrelstr = NULL;
938 else if (error != 0)
939 goto done_free;
940 else {
941 if (flags & JAIL_UPDATE) {
942 error = EINVAL;
943 vfs_opterror(opts,
944 "osrelease cannot be changed after creation");
945 goto done_errmsg;
946 }
947 if (len == 0 || len >= OSRELEASELEN) {
948 error = EINVAL;
949 vfs_opterror(opts,
950 "osrelease string must be 1-%d bytes long",
951 OSRELEASELEN - 1);
952 goto done_errmsg;
953 }
954 }
955
956 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
957 if (error == ENOENT)
958 osreldt = 0;
959 else if (error != 0)
960 goto done_free;
961 else {
962 if (flags & JAIL_UPDATE) {
963 error = EINVAL;
964 vfs_opterror(opts,
965 "osreldate cannot be changed after creation");
966 goto done_errmsg;
967 }
968 if (osreldt == 0) {
969 error = EINVAL;
970 vfs_opterror(opts, "osreldate cannot be 0");
971 goto done_errmsg;
972 }
973 }
974
975 fullpath_disabled = 0;
976 root = NULL;
977 error = vfs_getopt(opts, "path", (void **)&path, &len);
978 if (error == ENOENT)
979 path = NULL;
980 else if (error != 0)
981 goto done_free;
982 else {
983 if (flags & JAIL_UPDATE) {
984 error = EINVAL;
985 vfs_opterror(opts,
986 "path cannot be changed after creation");
987 goto done_errmsg;
988 }
989 if (len == 0 || path[len - 1] != '\0') {
990 error = EINVAL;
991 goto done_free;
992 }
993 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
994 path, td);
995 error = namei(&nd);
996 if (error)
997 goto done_free;
998 root = nd.ni_vp;
999 NDFREE(&nd, NDF_ONLY_PNBUF);
1000 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1001 strlcpy(g_path, path, MAXPATHLEN);
1002 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1003 if (error == 0)
1004 path = g_path;
1005 else if (error == ENODEV) {
1006 /* proceed if sysctl debug.disablefullpath == 1 */
1007 fullpath_disabled = 1;
1008 if (len < 2 || (len == 2 && path[0] == '/'))
1009 path = NULL;
1010 } else {
1011 /* exit on other errors */
1012 goto done_free;
1013 }
1014 if (root->v_type != VDIR) {
1015 error = ENOTDIR;
1016 vput(root);
1017 goto done_free;
1018 }
1019 VOP_UNLOCK(root, 0);
1020 if (fullpath_disabled) {
1021 /* Leave room for a real-root full pathname. */
1022 if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
1023 ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
1024 error = ENAMETOOLONG;
1025 vrele(root);
1026 goto done_free;
1027 }
1028 }
1029 }
1030
1031 /*
1032 * Find the specified jail, or at least its parent.
1033 * This abuses the file error codes ENOENT and EEXIST.
1034 */
1035 pr = NULL;
1036 ppr = mypr;
1037 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1038 namelc = strrchr(name, '.');
1039 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1040 if (*p != '\0')
1041 jid = 0;
1042 }
1043 sx_xlock(&allprison_lock);
1044 if (jid != 0) {
1045 /*
1046 * See if a requested jid already exists. There is an
1047 * information leak here if the jid exists but is not within
1048 * the caller's jail hierarchy. Jail creators will get EEXIST
1049 * even though they cannot see the jail, and CREATE | UPDATE
1050 * will return ENOENT which is not normally a valid error.
1051 */
1052 if (jid < 0) {
1053 error = EINVAL;
1054 vfs_opterror(opts, "negative jid");
1055 goto done_unlock_list;
1056 }
1057 pr = prison_find(jid);
1058 if (pr != NULL) {
1059 ppr = pr->pr_parent;
1060 /* Create: jid must not exist. */
1061 if (cuflags == JAIL_CREATE) {
1062 mtx_unlock(&pr->pr_mtx);
1063 error = EEXIST;
1064 vfs_opterror(opts, "jail %d already exists",
1065 jid);
1066 goto done_unlock_list;
1067 }
1068 if (!prison_ischild(mypr, pr)) {
1069 mtx_unlock(&pr->pr_mtx);
1070 pr = NULL;
1071 } else if (pr->pr_uref == 0) {
1072 if (!(flags & JAIL_DYING)) {
1073 mtx_unlock(&pr->pr_mtx);
1074 error = ENOENT;
1075 vfs_opterror(opts, "jail %d is dying",
1076 jid);
1077 goto done_unlock_list;
1078 } else if ((flags & JAIL_ATTACH) ||
1079 (pr_flags & PR_PERSIST)) {
1080 /*
1081 * A dying jail might be resurrected
1082 * (via attach or persist), but first
1083 * it must determine if another jail
1084 * has claimed its name. Accomplish
1085 * this by implicitly re-setting the
1086 * name.
1087 */
1088 if (name == NULL)
1089 name = prison_name(mypr, pr);
1090 }
1091 }
1092 }
1093 if (pr == NULL) {
1094 /* Update: jid must exist. */
1095 if (cuflags == JAIL_UPDATE) {
1096 error = ENOENT;
1097 vfs_opterror(opts, "jail %d not found", jid);
1098 goto done_unlock_list;
1099 }
1100 }
1101 }
1102 /*
1103 * If the caller provided a name, look for a jail by that name.
1104 * This has different semantics for creates and updates keyed by jid
1105 * (where the name must not already exist in a different jail),
1106 * and updates keyed by the name itself (where the name must exist
1107 * because that is the jail being updated).
1108 */
1109 namelc = NULL;
1110 if (name != NULL) {
1111 namelc = strrchr(name, '.');
1112 if (namelc == NULL)
1113 namelc = name;
1114 else {
1115 /*
1116 * This is a hierarchical name. Split it into the
1117 * parent and child names, and make sure the parent
1118 * exists or matches an already found jail.
1119 */
1120 if (pr != NULL) {
1121 if (strncmp(name, ppr->pr_name, namelc - name)
1122 || ppr->pr_name[namelc - name] != '\0') {
1123 mtx_unlock(&pr->pr_mtx);
1124 error = EINVAL;
1125 vfs_opterror(opts,
1126 "cannot change jail's parent");
1127 goto done_unlock_list;
1128 }
1129 } else {
1130 *namelc = '\0';
1131 ppr = prison_find_name(mypr, name);
1132 if (ppr == NULL) {
1133 error = ENOENT;
1134 vfs_opterror(opts,
1135 "jail \"%s\" not found", name);
1136 goto done_unlock_list;
1137 }
1138 mtx_unlock(&ppr->pr_mtx);
1139 *namelc = '.';
1140 }
1141 namelc++;
1142 }
1143 if (namelc[0] != '\0') {
1144 pnamelen =
1145 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1146 name_again:
1147 deadpr = NULL;
1148 FOREACH_PRISON_CHILD(ppr, tpr) {
1149 if (tpr != pr && tpr->pr_ref > 0 &&
1150 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1151 if (pr == NULL &&
1152 cuflags != JAIL_CREATE) {
1153 mtx_lock(&tpr->pr_mtx);
1154 if (tpr->pr_ref > 0) {
1155 /*
1156 * Use this jail
1157 * for updates.
1158 */
1159 if (tpr->pr_uref > 0) {
1160 pr = tpr;
1161 break;
1162 }
1163 deadpr = tpr;
1164 }
1165 mtx_unlock(&tpr->pr_mtx);
1166 } else if (tpr->pr_uref > 0) {
1167 /*
1168 * Create, or update(jid):
1169 * name must not exist in an
1170 * active sibling jail.
1171 */
1172 error = EEXIST;
1173 if (pr != NULL)
1174 mtx_unlock(&pr->pr_mtx);
1175 vfs_opterror(opts,
1176 "jail \"%s\" already exists",
1177 name);
1178 goto done_unlock_list;
1179 }
1180 }
1181 }
1182 /* If no active jail is found, use a dying one. */
1183 if (deadpr != NULL && pr == NULL) {
1184 if (flags & JAIL_DYING) {
1185 mtx_lock(&deadpr->pr_mtx);
1186 if (deadpr->pr_ref == 0) {
1187 mtx_unlock(&deadpr->pr_mtx);
1188 goto name_again;
1189 }
1190 pr = deadpr;
1191 } else if (cuflags == JAIL_UPDATE) {
1192 error = ENOENT;
1193 vfs_opterror(opts,
1194 "jail \"%s\" is dying", name);
1195 goto done_unlock_list;
1196 }
1197 }
1198 /* Update: name must exist if no jid. */
1199 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1200 error = ENOENT;
1201 vfs_opterror(opts, "jail \"%s\" not found",
1202 name);
1203 goto done_unlock_list;
1204 }
1205 }
1206 }
1207 /* Update: must provide a jid or name. */
1208 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1209 error = ENOENT;
1210 vfs_opterror(opts, "update specified no jail");
1211 goto done_unlock_list;
1212 }
1213
1214 /* If there's no prison to update, create a new one and link it in. */
1215 if (pr == NULL) {
1216 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1217 if (tpr->pr_childcount >= tpr->pr_childmax) {
1218 error = EPERM;
1219 vfs_opterror(opts, "prison limit exceeded");
1220 goto done_unlock_list;
1221 }
1222 created = 1;
1223 mtx_lock(&ppr->pr_mtx);
1224 if (ppr->pr_ref == 0) {
1225 mtx_unlock(&ppr->pr_mtx);
1226 error = ENOENT;
1227 vfs_opterror(opts, "jail \"%s\" not found",
1228 prison_name(mypr, ppr));
1229 goto done_unlock_list;
1230 }
1231 ppr->pr_ref++;
1232 ppr->pr_uref++;
1233 mtx_unlock(&ppr->pr_mtx);
1234 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1235 if (jid == 0) {
1236 /* Find the next free jid. */
1237 jid = lastprid + 1;
1238 findnext:
1239 if (jid == JAIL_MAX)
1240 jid = 1;
1241 TAILQ_FOREACH(tpr, &allprison, pr_list) {
1242 if (tpr->pr_id < jid)
1243 continue;
1244 if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1245 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1246 break;
1247 }
1248 if (jid == lastprid) {
1249 error = EAGAIN;
1250 vfs_opterror(opts,
1251 "no available jail IDs");
1252 free(pr, M_PRISON);
1253 prison_deref(ppr, PD_DEREF |
1254 PD_DEUREF | PD_LIST_XLOCKED);
1255 goto done_releroot;
1256 }
1257 jid++;
1258 goto findnext;
1259 }
1260 lastprid = jid;
1261 } else {
1262 /*
1263 * The jail already has a jid (that did not yet exist),
1264 * so just find where to insert it.
1265 */
1266 TAILQ_FOREACH(tpr, &allprison, pr_list)
1267 if (tpr->pr_id >= jid) {
1268 TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1269 break;
1270 }
1271 }
1272 if (tpr == NULL)
1273 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1274 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1275 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1276 tpr->pr_childcount++;
1277
1278 pr->pr_parent = ppr;
1279 pr->pr_id = jid;
1280
1281 /* Set some default values, and inherit some from the parent. */
1282 if (namelc == NULL)
1283 namelc = "";
1284 if (path == NULL) {
1285 path = "/";
1286 root = mypr->pr_root;
1287 vref(root);
1288 }
1289 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1290 pr->pr_flags |= PR_HOST;
1291 #if defined(INET) || defined(INET6)
1292 #ifdef VIMAGE
1293 if (!(pr_flags & PR_VNET))
1294 #endif
1295 {
1296 #ifdef INET
1297 if (!(ch_flags & PR_IP4_USER))
1298 pr->pr_flags |=
1299 PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1300 else if (!(pr_flags & PR_IP4_USER)) {
1301 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1302 if (ppr->pr_ip4 != NULL) {
1303 pr->pr_ip4s = ppr->pr_ip4s;
1304 pr->pr_ip4 = malloc(pr->pr_ip4s *
1305 sizeof(struct in_addr), M_PRISON,
1306 M_WAITOK);
1307 bcopy(ppr->pr_ip4, pr->pr_ip4,
1308 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1309 }
1310 }
1311 #endif
1312 #ifdef INET6
1313 if (!(ch_flags & PR_IP6_USER))
1314 pr->pr_flags |=
1315 PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1316 else if (!(pr_flags & PR_IP6_USER)) {
1317 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1318 if (ppr->pr_ip6 != NULL) {
1319 pr->pr_ip6s = ppr->pr_ip6s;
1320 pr->pr_ip6 = malloc(pr->pr_ip6s *
1321 sizeof(struct in6_addr), M_PRISON,
1322 M_WAITOK);
1323 bcopy(ppr->pr_ip6, pr->pr_ip6,
1324 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1325 }
1326 }
1327 #endif
1328 }
1329 #endif
1330 /* Source address selection is always on by default. */
1331 pr->pr_flags |= _PR_IP_SADDRSEL;
1332
1333 pr->pr_securelevel = ppr->pr_securelevel;
1334 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1335 pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1336 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1337
1338 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1339 if (osrelstr == NULL)
1340 strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1341 else
1342 strcpy(pr->pr_osrelease, osrelstr);
1343
1344 LIST_INIT(&pr->pr_children);
1345 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1346 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1347
1348 #ifdef VIMAGE
1349 /* Allocate a new vnet if specified. */
1350 pr->pr_vnet = (pr_flags & PR_VNET)
1351 ? vnet_alloc() : ppr->pr_vnet;
1352 #endif
1353 /*
1354 * Allocate a dedicated cpuset for each jail.
1355 * Unlike other initial settings, this may return an erorr.
1356 */
1357 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1358 if (error) {
1359 prison_deref(pr, PD_LIST_XLOCKED);
1360 goto done_releroot;
1361 }
1362
1363 mtx_lock(&pr->pr_mtx);
1364 /*
1365 * New prisons do not yet have a reference, because we do not
1366 * want others to see the incomplete prison once the
1367 * allprison_lock is downgraded.
1368 */
1369 } else {
1370 created = 0;
1371 /*
1372 * Grab a reference for existing prisons, to ensure they
1373 * continue to exist for the duration of the call.
1374 */
1375 pr->pr_ref++;
1376 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1377 if ((pr->pr_flags & PR_VNET) &&
1378 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1379 error = EINVAL;
1380 vfs_opterror(opts,
1381 "vnet jails cannot have IP address restrictions");
1382 goto done_deref_locked;
1383 }
1384 #endif
1385 #ifdef INET
1386 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1387 error = EINVAL;
1388 vfs_opterror(opts,
1389 "ip4 cannot be changed after creation");
1390 goto done_deref_locked;
1391 }
1392 #endif
1393 #ifdef INET6
1394 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1395 error = EINVAL;
1396 vfs_opterror(opts,
1397 "ip6 cannot be changed after creation");
1398 goto done_deref_locked;
1399 }
1400 #endif
1401 }
1402
1403 /* Do final error checking before setting anything. */
1404 if (gotslevel) {
1405 if (slevel < ppr->pr_securelevel) {
1406 error = EPERM;
1407 goto done_deref_locked;
1408 }
1409 }
1410 if (gotchildmax) {
1411 if (childmax >= ppr->pr_childmax) {
1412 error = EPERM;
1413 goto done_deref_locked;
1414 }
1415 }
1416 if (gotenforce) {
1417 if (enforce < ppr->pr_enforce_statfs) {
1418 error = EPERM;
1419 goto done_deref_locked;
1420 }
1421 }
1422 if (gotrsnum) {
1423 /*
1424 * devfs_rsnum is a uint16_t
1425 */
1426 if (rsnum < 0 || rsnum > 65535) {
1427 error = EINVAL;
1428 goto done_deref_locked;
1429 }
1430 /*
1431 * Nested jails always inherit parent's devfs ruleset
1432 */
1433 if (jailed(td->td_ucred)) {
1434 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1435 error = EPERM;
1436 goto done_deref_locked;
1437 } else
1438 rsnum = ppr->pr_devfs_rsnum;
1439 }
1440 }
1441 #ifdef INET
1442 if (ip4s > 0) {
1443 if (ppr->pr_flags & PR_IP4) {
1444 /*
1445 * Make sure the new set of IP addresses is a
1446 * subset of the parent's list. Don't worry
1447 * about the parent being unlocked, as any
1448 * setting is done with allprison_lock held.
1449 */
1450 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1451 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1452 break;
1453 if (ij == ppr->pr_ip4s) {
1454 error = EPERM;
1455 goto done_deref_locked;
1456 }
1457 if (ip4s > 1) {
1458 for (ii = ij = 1; ii < ip4s; ii++) {
1459 if (ip4[ii].s_addr ==
1460 ppr->pr_ip4[0].s_addr)
1461 continue;
1462 for (; ij < ppr->pr_ip4s; ij++)
1463 if (ip4[ii].s_addr ==
1464 ppr->pr_ip4[ij].s_addr)
1465 break;
1466 if (ij == ppr->pr_ip4s)
1467 break;
1468 }
1469 if (ij == ppr->pr_ip4s) {
1470 error = EPERM;
1471 goto done_deref_locked;
1472 }
1473 }
1474 }
1475 /*
1476 * Check for conflicting IP addresses. We permit them
1477 * if there is no more than one IP on each jail. If
1478 * there is a duplicate on a jail with more than one
1479 * IP stop checking and return error.
1480 */
1481 #ifdef VIMAGE
1482 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1483 if (tppr->pr_flags & PR_VNET)
1484 break;
1485 #else
1486 tppr = &prison0;
1487 #endif
1488 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1489 if (tpr == pr ||
1490 #ifdef VIMAGE
1491 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1492 #endif
1493 tpr->pr_uref == 0) {
1494 descend = 0;
1495 continue;
1496 }
1497 if (!(tpr->pr_flags & PR_IP4_USER))
1498 continue;
1499 descend = 0;
1500 if (tpr->pr_ip4 == NULL ||
1501 (ip4s == 1 && tpr->pr_ip4s == 1))
1502 continue;
1503 for (ii = 0; ii < ip4s; ii++) {
1504 if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1505 error = EADDRINUSE;
1506 vfs_opterror(opts,
1507 "IPv4 addresses clash");
1508 goto done_deref_locked;
1509 }
1510 }
1511 }
1512 }
1513 #endif
1514 #ifdef INET6
1515 if (ip6s > 0) {
1516 if (ppr->pr_flags & PR_IP6) {
1517 /*
1518 * Make sure the new set of IP addresses is a
1519 * subset of the parent's list.
1520 */
1521 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1522 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1523 &ppr->pr_ip6[ij]))
1524 break;
1525 if (ij == ppr->pr_ip6s) {
1526 error = EPERM;
1527 goto done_deref_locked;
1528 }
1529 if (ip6s > 1) {
1530 for (ii = ij = 1; ii < ip6s; ii++) {
1531 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1532 &ppr->pr_ip6[0]))
1533 continue;
1534 for (; ij < ppr->pr_ip6s; ij++)
1535 if (IN6_ARE_ADDR_EQUAL(
1536 &ip6[ii], &ppr->pr_ip6[ij]))
1537 break;
1538 if (ij == ppr->pr_ip6s)
1539 break;
1540 }
1541 if (ij == ppr->pr_ip6s) {
1542 error = EPERM;
1543 goto done_deref_locked;
1544 }
1545 }
1546 }
1547 /* Check for conflicting IP addresses. */
1548 #ifdef VIMAGE
1549 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1550 if (tppr->pr_flags & PR_VNET)
1551 break;
1552 #else
1553 tppr = &prison0;
1554 #endif
1555 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1556 if (tpr == pr ||
1557 #ifdef VIMAGE
1558 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1559 #endif
1560 tpr->pr_uref == 0) {
1561 descend = 0;
1562 continue;
1563 }
1564 if (!(tpr->pr_flags & PR_IP6_USER))
1565 continue;
1566 descend = 0;
1567 if (tpr->pr_ip6 == NULL ||
1568 (ip6s == 1 && tpr->pr_ip6s == 1))
1569 continue;
1570 for (ii = 0; ii < ip6s; ii++) {
1571 if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1572 error = EADDRINUSE;
1573 vfs_opterror(opts,
1574 "IPv6 addresses clash");
1575 goto done_deref_locked;
1576 }
1577 }
1578 }
1579 }
1580 #endif
1581 onamelen = namelen = 0;
1582 if (namelc != NULL) {
1583 /* Give a default name of the jid. Also allow the name to be
1584 * explicitly the jid - but not any other number, and only in
1585 * normal form (no leading zero/etc).
1586 */
1587 if (namelc[0] == '\0')
1588 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1589 else if ((strtoul(namelc, &p, 10) != jid ||
1590 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1591 error = EINVAL;
1592 vfs_opterror(opts,
1593 "name cannot be numeric (unless it is the jid)");
1594 goto done_deref_locked;
1595 }
1596 /*
1597 * Make sure the name isn't too long for the prison or its
1598 * children.
1599 */
1600 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1601 onamelen = strlen(pr->pr_name + pnamelen);
1602 namelen = strlen(namelc);
1603 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1604 error = ENAMETOOLONG;
1605 goto done_deref_locked;
1606 }
1607 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1608 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1609 sizeof(pr->pr_name)) {
1610 error = ENAMETOOLONG;
1611 goto done_deref_locked;
1612 }
1613 }
1614 }
1615 if (pr_allow & ~ppr->pr_allow) {
1616 error = EPERM;
1617 goto done_deref_locked;
1618 }
1619
1620 /*
1621 * Let modules check their parameters. This requires unlocking and
1622 * then re-locking the prison, but this is still a valid state as long
1623 * as allprison_lock remains xlocked.
1624 */
1625 mtx_unlock(&pr->pr_mtx);
1626 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1627 if (error != 0) {
1628 prison_deref(pr, created
1629 ? PD_LIST_XLOCKED
1630 : PD_DEREF | PD_LIST_XLOCKED);
1631 goto done_releroot;
1632 }
1633 mtx_lock(&pr->pr_mtx);
1634
1635 /* At this point, all valid parameters should have been noted. */
1636 TAILQ_FOREACH(opt, opts, link) {
1637 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1638 error = EINVAL;
1639 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1640 goto done_deref_locked;
1641 }
1642 }
1643
1644 /* Set the parameters of the prison. */
1645 #ifdef INET
1646 redo_ip4 = 0;
1647 if (pr_flags & PR_IP4_USER) {
1648 pr->pr_flags |= PR_IP4;
1649 free(pr->pr_ip4, M_PRISON);
1650 pr->pr_ip4s = ip4s;
1651 pr->pr_ip4 = ip4;
1652 ip4 = NULL;
1653 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1654 #ifdef VIMAGE
1655 if (tpr->pr_flags & PR_VNET) {
1656 descend = 0;
1657 continue;
1658 }
1659 #endif
1660 if (prison_restrict_ip4(tpr, NULL)) {
1661 redo_ip4 = 1;
1662 descend = 0;
1663 }
1664 }
1665 }
1666 #endif
1667 #ifdef INET6
1668 redo_ip6 = 0;
1669 if (pr_flags & PR_IP6_USER) {
1670 pr->pr_flags |= PR_IP6;
1671 free(pr->pr_ip6, M_PRISON);
1672 pr->pr_ip6s = ip6s;
1673 pr->pr_ip6 = ip6;
1674 ip6 = NULL;
1675 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1676 #ifdef VIMAGE
1677 if (tpr->pr_flags & PR_VNET) {
1678 descend = 0;
1679 continue;
1680 }
1681 #endif
1682 if (prison_restrict_ip6(tpr, NULL)) {
1683 redo_ip6 = 1;
1684 descend = 0;
1685 }
1686 }
1687 }
1688 #endif
1689 if (gotslevel) {
1690 pr->pr_securelevel = slevel;
1691 /* Set all child jails to be at least this level. */
1692 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1693 if (tpr->pr_securelevel < slevel)
1694 tpr->pr_securelevel = slevel;
1695 }
1696 if (gotchildmax) {
1697 pr->pr_childmax = childmax;
1698 /* Set all child jails to under this limit. */
1699 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1700 if (tpr->pr_childmax > childmax - level)
1701 tpr->pr_childmax = childmax > level
1702 ? childmax - level : 0;
1703 }
1704 if (gotenforce) {
1705 pr->pr_enforce_statfs = enforce;
1706 /* Pass this restriction on to the children. */
1707 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1708 if (tpr->pr_enforce_statfs < enforce)
1709 tpr->pr_enforce_statfs = enforce;
1710 }
1711 if (gotrsnum) {
1712 pr->pr_devfs_rsnum = rsnum;
1713 /* Pass this restriction on to the children. */
1714 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1715 tpr->pr_devfs_rsnum = rsnum;
1716 }
1717 if (namelc != NULL) {
1718 if (ppr == &prison0)
1719 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1720 else
1721 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1722 ppr->pr_name, namelc);
1723 /* Change this component of child names. */
1724 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1725 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1726 strlen(tpr->pr_name + onamelen) + 1);
1727 bcopy(pr->pr_name, tpr->pr_name, namelen);
1728 }
1729 }
1730 if (path != NULL) {
1731 /* Try to keep a real-rooted full pathname. */
1732 if (fullpath_disabled && path[0] == '/' &&
1733 strcmp(mypr->pr_path, "/"))
1734 snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1735 mypr->pr_path, path);
1736 else
1737 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1738 pr->pr_root = root;
1739 }
1740 if (PR_HOST & ch_flags & ~pr_flags) {
1741 if (pr->pr_flags & PR_HOST) {
1742 /*
1743 * Copy the parent's host info. As with pr_ip4 above,
1744 * the lack of a lock on the parent is not a problem;
1745 * it is always set with allprison_lock at least
1746 * shared, and is held exclusively here.
1747 */
1748 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1749 sizeof(pr->pr_hostname));
1750 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1751 sizeof(pr->pr_domainname));
1752 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1753 sizeof(pr->pr_hostuuid));
1754 pr->pr_hostid = pr->pr_parent->pr_hostid;
1755 }
1756 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1757 /* Set this prison, and any descendants without PR_HOST. */
1758 if (host != NULL)
1759 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1760 if (domain != NULL)
1761 strlcpy(pr->pr_domainname, domain,
1762 sizeof(pr->pr_domainname));
1763 if (uuid != NULL)
1764 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1765 if (gothid)
1766 pr->pr_hostid = hid;
1767 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1768 if (tpr->pr_flags & PR_HOST)
1769 descend = 0;
1770 else {
1771 if (host != NULL)
1772 strlcpy(tpr->pr_hostname,
1773 pr->pr_hostname,
1774 sizeof(tpr->pr_hostname));
1775 if (domain != NULL)
1776 strlcpy(tpr->pr_domainname,
1777 pr->pr_domainname,
1778 sizeof(tpr->pr_domainname));
1779 if (uuid != NULL)
1780 strlcpy(tpr->pr_hostuuid,
1781 pr->pr_hostuuid,
1782 sizeof(tpr->pr_hostuuid));
1783 if (gothid)
1784 tpr->pr_hostid = hid;
1785 }
1786 }
1787 }
1788 if ((tallow = ch_allow & ~pr_allow)) {
1789 /* Clear allow bits in all children. */
1790 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1791 tpr->pr_allow &= ~tallow;
1792 }
1793 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1794 /*
1795 * Persistent prisons get an extra reference, and prisons losing their
1796 * persist flag lose that reference. Only do this for existing prisons
1797 * for now, so new ones will remain unseen until after the module
1798 * handlers have completed.
1799 */
1800 born = pr->pr_uref == 0;
1801 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1802 if (pr_flags & PR_PERSIST) {
1803 pr->pr_ref++;
1804 pr->pr_uref++;
1805 } else {
1806 pr->pr_ref--;
1807 pr->pr_uref--;
1808 }
1809 }
1810 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1811 mtx_unlock(&pr->pr_mtx);
1812
1813 #ifdef RACCT
1814 if (racct_enable && created)
1815 prison_racct_attach(pr);
1816 #endif
1817
1818 /* Locks may have prevented a complete restriction of child IP
1819 * addresses. If so, allocate some more memory and try again.
1820 */
1821 #ifdef INET
1822 while (redo_ip4) {
1823 ip4s = pr->pr_ip4s;
1824 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1825 mtx_lock(&pr->pr_mtx);
1826 redo_ip4 = 0;
1827 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1828 #ifdef VIMAGE
1829 if (tpr->pr_flags & PR_VNET) {
1830 descend = 0;
1831 continue;
1832 }
1833 #endif
1834 if (prison_restrict_ip4(tpr, ip4)) {
1835 if (ip4 != NULL)
1836 ip4 = NULL;
1837 else
1838 redo_ip4 = 1;
1839 }
1840 }
1841 mtx_unlock(&pr->pr_mtx);
1842 }
1843 #endif
1844 #ifdef INET6
1845 while (redo_ip6) {
1846 ip6s = pr->pr_ip6s;
1847 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1848 mtx_lock(&pr->pr_mtx);
1849 redo_ip6 = 0;
1850 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1851 #ifdef VIMAGE
1852 if (tpr->pr_flags & PR_VNET) {
1853 descend = 0;
1854 continue;
1855 }
1856 #endif
1857 if (prison_restrict_ip6(tpr, ip6)) {
1858 if (ip6 != NULL)
1859 ip6 = NULL;
1860 else
1861 redo_ip6 = 1;
1862 }
1863 }
1864 mtx_unlock(&pr->pr_mtx);
1865 }
1866 #endif
1867
1868 /* Let the modules do their work. */
1869 sx_downgrade(&allprison_lock);
1870 if (born) {
1871 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1872 if (error) {
1873 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1874 prison_deref(pr, created
1875 ? PD_LIST_SLOCKED
1876 : PD_DEREF | PD_LIST_SLOCKED);
1877 goto done_errmsg;
1878 }
1879 }
1880 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1881 if (error) {
1882 if (born)
1883 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1884 prison_deref(pr, created
1885 ? PD_LIST_SLOCKED
1886 : PD_DEREF | PD_LIST_SLOCKED);
1887 goto done_errmsg;
1888 }
1889
1890 /* Attach this process to the prison if requested. */
1891 if (flags & JAIL_ATTACH) {
1892 mtx_lock(&pr->pr_mtx);
1893 error = do_jail_attach(td, pr);
1894 if (error) {
1895 vfs_opterror(opts, "attach failed");
1896 if (!created)
1897 prison_deref(pr, PD_DEREF);
1898 goto done_errmsg;
1899 }
1900 }
1901
1902 #ifdef RACCT
1903 if (racct_enable && !created) {
1904 if (!(flags & JAIL_ATTACH))
1905 sx_sunlock(&allprison_lock);
1906 prison_racct_modify(pr);
1907 if (!(flags & JAIL_ATTACH))
1908 sx_slock(&allprison_lock);
1909 }
1910 #endif
1911
1912 td->td_retval[0] = pr->pr_id;
1913
1914 /*
1915 * Now that it is all there, drop the temporary reference from existing
1916 * prisons. Or add a reference to newly created persistent prisons
1917 * (which was not done earlier so that the prison would not be publicly
1918 * visible).
1919 */
1920 if (!created) {
1921 prison_deref(pr, (flags & JAIL_ATTACH)
1922 ? PD_DEREF
1923 : PD_DEREF | PD_LIST_SLOCKED);
1924 } else {
1925 if (pr_flags & PR_PERSIST) {
1926 mtx_lock(&pr->pr_mtx);
1927 pr->pr_ref++;
1928 pr->pr_uref++;
1929 mtx_unlock(&pr->pr_mtx);
1930 }
1931 if (!(flags & JAIL_ATTACH))
1932 sx_sunlock(&allprison_lock);
1933 }
1934
1935 goto done_free;
1936
1937 done_deref_locked:
1938 prison_deref(pr, created
1939 ? PD_LOCKED | PD_LIST_XLOCKED
1940 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1941 goto done_releroot;
1942 done_unlock_list:
1943 sx_xunlock(&allprison_lock);
1944 done_releroot:
1945 if (root != NULL)
1946 vrele(root);
1947 done_errmsg:
1948 if (error) {
1949 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1950 &errmsg_len) == 0 && errmsg_len > 0) {
1951 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1952 if (optuio->uio_segflg == UIO_SYSSPACE)
1953 bcopy(errmsg,
1954 optuio->uio_iov[errmsg_pos].iov_base,
1955 errmsg_len);
1956 else
1957 copyout(errmsg,
1958 optuio->uio_iov[errmsg_pos].iov_base,
1959 errmsg_len);
1960 }
1961 }
1962 done_free:
1963 #ifdef INET
1964 free(ip4, M_PRISON);
1965 #endif
1966 #ifdef INET6
1967 free(ip6, M_PRISON);
1968 #endif
1969 if (g_path != NULL)
1970 free(g_path, M_TEMP);
1971 vfs_freeopts(opts);
1972 return (error);
1973 }
1974
1975
1976 /*
1977 * struct jail_get_args {
1978 * struct iovec *iovp;
1979 * unsigned int iovcnt;
1980 * int flags;
1981 * };
1982 */
1983 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)1984 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1985 {
1986 struct uio *auio;
1987 int error;
1988
1989 /* Check that we have an even number of iovecs. */
1990 if (uap->iovcnt & 1)
1991 return (EINVAL);
1992
1993 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1994 if (error)
1995 return (error);
1996 error = kern_jail_get(td, auio, uap->flags);
1997 if (error == 0)
1998 error = copyout(auio->uio_iov, uap->iovp,
1999 uap->iovcnt * sizeof (struct iovec));
2000 free(auio, M_IOV);
2001 return (error);
2002 }
2003
2004 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2005 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2006 {
2007 struct prison *pr, *mypr;
2008 struct vfsopt *opt;
2009 struct vfsoptlist *opts;
2010 char *errmsg, *name;
2011 int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
2012
2013 if (flags & ~JAIL_GET_MASK)
2014 return (EINVAL);
2015
2016 /* Get the parameter list. */
2017 error = vfs_buildopts(optuio, &opts);
2018 if (error)
2019 return (error);
2020 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2021 mypr = td->td_ucred->cr_prison;
2022
2023 /*
2024 * Find the prison specified by one of: lastjid, jid, name.
2025 */
2026 sx_slock(&allprison_lock);
2027 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2028 if (error == 0) {
2029 TAILQ_FOREACH(pr, &allprison, pr_list) {
2030 if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2031 mtx_lock(&pr->pr_mtx);
2032 if (pr->pr_ref > 0 &&
2033 (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2034 break;
2035 mtx_unlock(&pr->pr_mtx);
2036 }
2037 }
2038 if (pr != NULL)
2039 goto found_prison;
2040 error = ENOENT;
2041 vfs_opterror(opts, "no jail after %d", jid);
2042 goto done_unlock_list;
2043 } else if (error != ENOENT)
2044 goto done_unlock_list;
2045
2046 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2047 if (error == 0) {
2048 if (jid != 0) {
2049 pr = prison_find_child(mypr, jid);
2050 if (pr != NULL) {
2051 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2052 mtx_unlock(&pr->pr_mtx);
2053 error = ENOENT;
2054 vfs_opterror(opts, "jail %d is dying",
2055 jid);
2056 goto done_unlock_list;
2057 }
2058 goto found_prison;
2059 }
2060 error = ENOENT;
2061 vfs_opterror(opts, "jail %d not found", jid);
2062 goto done_unlock_list;
2063 }
2064 } else if (error != ENOENT)
2065 goto done_unlock_list;
2066
2067 error = vfs_getopt(opts, "name", (void **)&name, &len);
2068 if (error == 0) {
2069 if (len == 0 || name[len - 1] != '\0') {
2070 error = EINVAL;
2071 goto done_unlock_list;
2072 }
2073 pr = prison_find_name(mypr, name);
2074 if (pr != NULL) {
2075 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2076 mtx_unlock(&pr->pr_mtx);
2077 error = ENOENT;
2078 vfs_opterror(opts, "jail \"%s\" is dying",
2079 name);
2080 goto done_unlock_list;
2081 }
2082 goto found_prison;
2083 }
2084 error = ENOENT;
2085 vfs_opterror(opts, "jail \"%s\" not found", name);
2086 goto done_unlock_list;
2087 } else if (error != ENOENT)
2088 goto done_unlock_list;
2089
2090 vfs_opterror(opts, "no jail specified");
2091 error = ENOENT;
2092 goto done_unlock_list;
2093
2094 found_prison:
2095 /* Get the parameters of the prison. */
2096 pr->pr_ref++;
2097 locked = PD_LOCKED;
2098 td->td_retval[0] = pr->pr_id;
2099 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2100 if (error != 0 && error != ENOENT)
2101 goto done_deref;
2102 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2103 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2104 if (error != 0 && error != ENOENT)
2105 goto done_deref;
2106 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2107 if (error != 0 && error != ENOENT)
2108 goto done_deref;
2109 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2110 sizeof(pr->pr_cpuset->cs_id));
2111 if (error != 0 && error != ENOENT)
2112 goto done_deref;
2113 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2114 if (error != 0 && error != ENOENT)
2115 goto done_deref;
2116 #ifdef INET
2117 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2118 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2119 if (error != 0 && error != ENOENT)
2120 goto done_deref;
2121 #endif
2122 #ifdef INET6
2123 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2124 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2125 if (error != 0 && error != ENOENT)
2126 goto done_deref;
2127 #endif
2128 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2129 sizeof(pr->pr_securelevel));
2130 if (error != 0 && error != ENOENT)
2131 goto done_deref;
2132 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2133 sizeof(pr->pr_childcount));
2134 if (error != 0 && error != ENOENT)
2135 goto done_deref;
2136 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2137 sizeof(pr->pr_childmax));
2138 if (error != 0 && error != ENOENT)
2139 goto done_deref;
2140 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2141 if (error != 0 && error != ENOENT)
2142 goto done_deref;
2143 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2144 if (error != 0 && error != ENOENT)
2145 goto done_deref;
2146 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2147 if (error != 0 && error != ENOENT)
2148 goto done_deref;
2149 #ifdef COMPAT_FREEBSD32
2150 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2151 uint32_t hid32 = pr->pr_hostid;
2152
2153 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2154 } else
2155 #endif
2156 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2157 sizeof(pr->pr_hostid));
2158 if (error != 0 && error != ENOENT)
2159 goto done_deref;
2160 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2161 sizeof(pr->pr_enforce_statfs));
2162 if (error != 0 && error != ENOENT)
2163 goto done_deref;
2164 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2165 sizeof(pr->pr_devfs_rsnum));
2166 if (error != 0 && error != ENOENT)
2167 goto done_deref;
2168 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
2169 fi++) {
2170 if (pr_flag_names[fi] == NULL)
2171 continue;
2172 i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2173 error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2174 if (error != 0 && error != ENOENT)
2175 goto done_deref;
2176 i = !i;
2177 error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2178 if (error != 0 && error != ENOENT)
2179 goto done_deref;
2180 }
2181 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2182 fi++) {
2183 i = pr->pr_flags &
2184 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2185 i = pr_flag_jailsys[fi].disable &&
2186 (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2187 : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2188 : JAIL_SYS_INHERIT;
2189 error =
2190 vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2191 if (error != 0 && error != ENOENT)
2192 goto done_deref;
2193 }
2194 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2195 fi++) {
2196 if (pr_allow_names[fi] == NULL)
2197 continue;
2198 i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2199 error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2200 if (error != 0 && error != ENOENT)
2201 goto done_deref;
2202 i = !i;
2203 error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2204 if (error != 0 && error != ENOENT)
2205 goto done_deref;
2206 }
2207 i = (pr->pr_uref == 0);
2208 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2209 if (error != 0 && error != ENOENT)
2210 goto done_deref;
2211 i = !i;
2212 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2213 if (error != 0 && error != ENOENT)
2214 goto done_deref;
2215 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2216 sizeof(pr->pr_osreldate));
2217 if (error != 0 && error != ENOENT)
2218 goto done_deref;
2219 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2220 if (error != 0 && error != ENOENT)
2221 goto done_deref;
2222
2223 /* Get the module parameters. */
2224 mtx_unlock(&pr->pr_mtx);
2225 locked = 0;
2226 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2227 if (error)
2228 goto done_deref;
2229 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2230
2231 /* By now, all parameters should have been noted. */
2232 TAILQ_FOREACH(opt, opts, link) {
2233 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2234 error = EINVAL;
2235 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2236 goto done_errmsg;
2237 }
2238 }
2239
2240 /* Write the fetched parameters back to userspace. */
2241 error = 0;
2242 TAILQ_FOREACH(opt, opts, link) {
2243 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2244 pos = 2 * opt->pos + 1;
2245 optuio->uio_iov[pos].iov_len = opt->len;
2246 if (opt->value != NULL) {
2247 if (optuio->uio_segflg == UIO_SYSSPACE) {
2248 bcopy(opt->value,
2249 optuio->uio_iov[pos].iov_base,
2250 opt->len);
2251 } else {
2252 error = copyout(opt->value,
2253 optuio->uio_iov[pos].iov_base,
2254 opt->len);
2255 if (error)
2256 break;
2257 }
2258 }
2259 }
2260 }
2261 goto done_errmsg;
2262
2263 done_deref:
2264 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2265 goto done_errmsg;
2266
2267 done_unlock_list:
2268 sx_sunlock(&allprison_lock);
2269 done_errmsg:
2270 if (error && errmsg_pos >= 0) {
2271 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2272 errmsg_pos = 2 * errmsg_pos + 1;
2273 if (errmsg_len > 0) {
2274 if (optuio->uio_segflg == UIO_SYSSPACE)
2275 bcopy(errmsg,
2276 optuio->uio_iov[errmsg_pos].iov_base,
2277 errmsg_len);
2278 else
2279 copyout(errmsg,
2280 optuio->uio_iov[errmsg_pos].iov_base,
2281 errmsg_len);
2282 }
2283 }
2284 vfs_freeopts(opts);
2285 return (error);
2286 }
2287
2288
2289 /*
2290 * struct jail_remove_args {
2291 * int jid;
2292 * };
2293 */
2294 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2295 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2296 {
2297 struct prison *pr, *cpr, *lpr, *tpr;
2298 int descend, error;
2299
2300 error = priv_check(td, PRIV_JAIL_REMOVE);
2301 if (error)
2302 return (error);
2303
2304 sx_xlock(&allprison_lock);
2305 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2306 if (pr == NULL) {
2307 sx_xunlock(&allprison_lock);
2308 return (EINVAL);
2309 }
2310
2311 /* Remove all descendants of this prison, then remove this prison. */
2312 pr->pr_ref++;
2313 if (!LIST_EMPTY(&pr->pr_children)) {
2314 mtx_unlock(&pr->pr_mtx);
2315 lpr = NULL;
2316 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2317 mtx_lock(&cpr->pr_mtx);
2318 if (cpr->pr_ref > 0) {
2319 tpr = cpr;
2320 cpr->pr_ref++;
2321 } else {
2322 /* Already removed - do not do it again. */
2323 tpr = NULL;
2324 }
2325 mtx_unlock(&cpr->pr_mtx);
2326 if (lpr != NULL) {
2327 mtx_lock(&lpr->pr_mtx);
2328 prison_remove_one(lpr);
2329 sx_xlock(&allprison_lock);
2330 }
2331 lpr = tpr;
2332 }
2333 if (lpr != NULL) {
2334 mtx_lock(&lpr->pr_mtx);
2335 prison_remove_one(lpr);
2336 sx_xlock(&allprison_lock);
2337 }
2338 mtx_lock(&pr->pr_mtx);
2339 }
2340 prison_remove_one(pr);
2341 return (0);
2342 }
2343
2344 static void
prison_remove_one(struct prison * pr)2345 prison_remove_one(struct prison *pr)
2346 {
2347 struct proc *p;
2348 int deuref;
2349
2350 /* If the prison was persistent, it is not anymore. */
2351 deuref = 0;
2352 if (pr->pr_flags & PR_PERSIST) {
2353 pr->pr_ref--;
2354 deuref = PD_DEUREF;
2355 pr->pr_flags &= ~PR_PERSIST;
2356 }
2357
2358 /*
2359 * jail_remove added a reference. If that's the only one, remove
2360 * the prison now.
2361 */
2362 KASSERT(pr->pr_ref > 0,
2363 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2364 if (pr->pr_ref == 1) {
2365 prison_deref(pr,
2366 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2367 return;
2368 }
2369
2370 mtx_unlock(&pr->pr_mtx);
2371 sx_xunlock(&allprison_lock);
2372 /*
2373 * Kill all processes unfortunate enough to be attached to this prison.
2374 */
2375 sx_slock(&allproc_lock);
2376 LIST_FOREACH(p, &allproc, p_list) {
2377 PROC_LOCK(p);
2378 if (p->p_state != PRS_NEW && p->p_ucred &&
2379 p->p_ucred->cr_prison == pr)
2380 kern_psignal(p, SIGKILL);
2381 PROC_UNLOCK(p);
2382 }
2383 sx_sunlock(&allproc_lock);
2384 /* Remove the temporary reference added by jail_remove. */
2385 prison_deref(pr, deuref | PD_DEREF);
2386 }
2387
2388
2389 /*
2390 * struct jail_attach_args {
2391 * int jid;
2392 * };
2393 */
2394 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)2395 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2396 {
2397 struct prison *pr;
2398 int error;
2399
2400 error = priv_check(td, PRIV_JAIL_ATTACH);
2401 if (error)
2402 return (error);
2403
2404 /*
2405 * Start with exclusive hold on allprison_lock to ensure that a possible
2406 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2407 * But then immediately downgrade it since we don't need to stop
2408 * readers.
2409 */
2410 sx_xlock(&allprison_lock);
2411 sx_downgrade(&allprison_lock);
2412 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2413 if (pr == NULL) {
2414 sx_sunlock(&allprison_lock);
2415 return (EINVAL);
2416 }
2417
2418 /*
2419 * Do not allow a process to attach to a prison that is not
2420 * considered to be "alive".
2421 */
2422 if (pr->pr_uref == 0) {
2423 mtx_unlock(&pr->pr_mtx);
2424 sx_sunlock(&allprison_lock);
2425 return (EINVAL);
2426 }
2427
2428 return (do_jail_attach(td, pr));
2429 }
2430
2431 static int
do_jail_attach(struct thread * td,struct prison * pr)2432 do_jail_attach(struct thread *td, struct prison *pr)
2433 {
2434 struct proc *p;
2435 struct ucred *newcred, *oldcred;
2436 int error;
2437
2438 /*
2439 * XXX: Note that there is a slight race here if two threads
2440 * in the same privileged process attempt to attach to two
2441 * different jails at the same time. It is important for
2442 * user processes not to do this, or they might end up with
2443 * a process root from one prison, but attached to the jail
2444 * of another.
2445 */
2446 pr->pr_ref++;
2447 pr->pr_uref++;
2448 mtx_unlock(&pr->pr_mtx);
2449
2450 /* Let modules do whatever they need to prepare for attaching. */
2451 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2452 if (error) {
2453 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2454 return (error);
2455 }
2456 sx_sunlock(&allprison_lock);
2457
2458 /*
2459 * Reparent the newly attached process to this jail.
2460 */
2461 p = td->td_proc;
2462 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2463 if (error)
2464 goto e_revert_osd;
2465
2466 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2467 if ((error = change_dir(pr->pr_root, td)) != 0)
2468 goto e_unlock;
2469 #ifdef MAC
2470 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2471 goto e_unlock;
2472 #endif
2473 VOP_UNLOCK(pr->pr_root, 0);
2474 if ((error = change_root(pr->pr_root, td)))
2475 goto e_revert_osd;
2476
2477 newcred = crget();
2478 PROC_LOCK(p);
2479 oldcred = crcopysafe(p, newcred);
2480 newcred->cr_prison = pr;
2481 proc_set_cred(p, newcred);
2482 setsugid(p);
2483 PROC_UNLOCK(p);
2484 #ifdef RACCT
2485 racct_proc_ucred_changed(p, oldcred, newcred);
2486 #endif
2487 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2488 crfree(oldcred);
2489 return (0);
2490
2491 e_unlock:
2492 VOP_UNLOCK(pr->pr_root, 0);
2493 e_revert_osd:
2494 /* Tell modules this thread is still in its old jail after all. */
2495 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2496 prison_deref(pr, PD_DEREF | PD_DEUREF);
2497 return (error);
2498 }
2499
2500
2501 /*
2502 * Returns a locked prison instance, or NULL on failure.
2503 */
2504 struct prison *
prison_find(int prid)2505 prison_find(int prid)
2506 {
2507 struct prison *pr;
2508
2509 sx_assert(&allprison_lock, SX_LOCKED);
2510 TAILQ_FOREACH(pr, &allprison, pr_list) {
2511 if (pr->pr_id == prid) {
2512 mtx_lock(&pr->pr_mtx);
2513 if (pr->pr_ref > 0)
2514 return (pr);
2515 mtx_unlock(&pr->pr_mtx);
2516 }
2517 }
2518 return (NULL);
2519 }
2520
2521 /*
2522 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2523 */
2524 struct prison *
prison_find_child(struct prison * mypr,int prid)2525 prison_find_child(struct prison *mypr, int prid)
2526 {
2527 struct prison *pr;
2528 int descend;
2529
2530 sx_assert(&allprison_lock, SX_LOCKED);
2531 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2532 if (pr->pr_id == prid) {
2533 mtx_lock(&pr->pr_mtx);
2534 if (pr->pr_ref > 0)
2535 return (pr);
2536 mtx_unlock(&pr->pr_mtx);
2537 }
2538 }
2539 return (NULL);
2540 }
2541
2542 /*
2543 * Look for the name relative to mypr. Returns a locked prison or NULL.
2544 */
2545 struct prison *
prison_find_name(struct prison * mypr,const char * name)2546 prison_find_name(struct prison *mypr, const char *name)
2547 {
2548 struct prison *pr, *deadpr;
2549 size_t mylen;
2550 int descend;
2551
2552 sx_assert(&allprison_lock, SX_LOCKED);
2553 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2554 again:
2555 deadpr = NULL;
2556 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2557 if (!strcmp(pr->pr_name + mylen, name)) {
2558 mtx_lock(&pr->pr_mtx);
2559 if (pr->pr_ref > 0) {
2560 if (pr->pr_uref > 0)
2561 return (pr);
2562 deadpr = pr;
2563 }
2564 mtx_unlock(&pr->pr_mtx);
2565 }
2566 }
2567 /* There was no valid prison - perhaps there was a dying one. */
2568 if (deadpr != NULL) {
2569 mtx_lock(&deadpr->pr_mtx);
2570 if (deadpr->pr_ref == 0) {
2571 mtx_unlock(&deadpr->pr_mtx);
2572 goto again;
2573 }
2574 }
2575 return (deadpr);
2576 }
2577
2578 /*
2579 * See if a prison has the specific flag set.
2580 */
2581 int
prison_flag(struct ucred * cred,unsigned flag)2582 prison_flag(struct ucred *cred, unsigned flag)
2583 {
2584
2585 /* This is an atomic read, so no locking is necessary. */
2586 return (cred->cr_prison->pr_flags & flag);
2587 }
2588
2589 int
prison_allow(struct ucred * cred,unsigned flag)2590 prison_allow(struct ucred *cred, unsigned flag)
2591 {
2592
2593 /* This is an atomic read, so no locking is necessary. */
2594 return (cred->cr_prison->pr_allow & flag);
2595 }
2596
2597 /*
2598 * Remove a prison reference. If that was the last reference, remove the
2599 * prison itself - but not in this context in case there are locks held.
2600 */
2601 void
prison_free_locked(struct prison * pr)2602 prison_free_locked(struct prison *pr)
2603 {
2604 int ref;
2605
2606 mtx_assert(&pr->pr_mtx, MA_OWNED);
2607 ref = --pr->pr_ref;
2608 mtx_unlock(&pr->pr_mtx);
2609 if (ref == 0)
2610 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2611 }
2612
2613 void
prison_free(struct prison * pr)2614 prison_free(struct prison *pr)
2615 {
2616
2617 mtx_lock(&pr->pr_mtx);
2618 prison_free_locked(pr);
2619 }
2620
2621 /*
2622 * Complete a call to either prison_free or prison_proc_free.
2623 */
2624 static void
prison_complete(void * context,int pending)2625 prison_complete(void *context, int pending)
2626 {
2627 struct prison *pr = context;
2628
2629 sx_xlock(&allprison_lock);
2630 mtx_lock(&pr->pr_mtx);
2631 prison_deref(pr, pr->pr_uref
2632 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2633 : PD_LOCKED | PD_LIST_XLOCKED);
2634 }
2635
2636 /*
2637 * Remove a prison reference (usually). This internal version assumes no
2638 * mutexes are held, except perhaps the prison itself. If there are no more
2639 * references, release and delist the prison. On completion, the prison lock
2640 * and the allprison lock are both unlocked.
2641 */
2642 static void
prison_deref(struct prison * pr,int flags)2643 prison_deref(struct prison *pr, int flags)
2644 {
2645 struct prison *ppr, *tpr;
2646 int ref, lasturef;
2647
2648 if (!(flags & PD_LOCKED))
2649 mtx_lock(&pr->pr_mtx);
2650 for (;;) {
2651 if (flags & PD_DEUREF) {
2652 KASSERT(pr->pr_uref > 0,
2653 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2654 pr->pr_id));
2655 pr->pr_uref--;
2656 lasturef = pr->pr_uref == 0;
2657 if (lasturef)
2658 pr->pr_ref++;
2659 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2660 } else
2661 lasturef = 0;
2662 if (flags & PD_DEREF) {
2663 KASSERT(pr->pr_ref > 0,
2664 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2665 pr->pr_id));
2666 pr->pr_ref--;
2667 }
2668 ref = pr->pr_ref;
2669 mtx_unlock(&pr->pr_mtx);
2670
2671 /*
2672 * Tell the modules if the last user reference was removed
2673 * (even it sticks around in dying state).
2674 */
2675 if (lasturef) {
2676 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2677 sx_xlock(&allprison_lock);
2678 flags |= PD_LIST_XLOCKED;
2679 }
2680 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2681 mtx_lock(&pr->pr_mtx);
2682 ref = --pr->pr_ref;
2683 mtx_unlock(&pr->pr_mtx);
2684 }
2685
2686 /* If the prison still has references, nothing else to do. */
2687 if (ref > 0) {
2688 if (flags & PD_LIST_SLOCKED)
2689 sx_sunlock(&allprison_lock);
2690 else if (flags & PD_LIST_XLOCKED)
2691 sx_xunlock(&allprison_lock);
2692 return;
2693 }
2694
2695 if (flags & PD_LIST_SLOCKED) {
2696 if (!sx_try_upgrade(&allprison_lock)) {
2697 sx_sunlock(&allprison_lock);
2698 sx_xlock(&allprison_lock);
2699 }
2700 } else if (!(flags & PD_LIST_XLOCKED))
2701 sx_xlock(&allprison_lock);
2702
2703 TAILQ_REMOVE(&allprison, pr, pr_list);
2704 LIST_REMOVE(pr, pr_sibling);
2705 ppr = pr->pr_parent;
2706 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2707 tpr->pr_childcount--;
2708 sx_xunlock(&allprison_lock);
2709
2710 #ifdef VIMAGE
2711 if (pr->pr_vnet != ppr->pr_vnet)
2712 vnet_destroy(pr->pr_vnet);
2713 #endif
2714 if (pr->pr_root != NULL)
2715 vrele(pr->pr_root);
2716 mtx_destroy(&pr->pr_mtx);
2717 #ifdef INET
2718 free(pr->pr_ip4, M_PRISON);
2719 #endif
2720 #ifdef INET6
2721 free(pr->pr_ip6, M_PRISON);
2722 #endif
2723 if (pr->pr_cpuset != NULL)
2724 cpuset_rel(pr->pr_cpuset);
2725 osd_jail_exit(pr);
2726 #ifdef RACCT
2727 if (racct_enable)
2728 prison_racct_detach(pr);
2729 #endif
2730 free(pr, M_PRISON);
2731
2732 /* Removing a prison frees a reference on its parent. */
2733 pr = ppr;
2734 mtx_lock(&pr->pr_mtx);
2735 flags = PD_DEREF | PD_DEUREF;
2736 }
2737 }
2738
2739 void
prison_hold_locked(struct prison * pr)2740 prison_hold_locked(struct prison *pr)
2741 {
2742
2743 mtx_assert(&pr->pr_mtx, MA_OWNED);
2744 KASSERT(pr->pr_ref > 0,
2745 ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2746 pr->pr_ref++;
2747 }
2748
2749 void
prison_hold(struct prison * pr)2750 prison_hold(struct prison *pr)
2751 {
2752
2753 mtx_lock(&pr->pr_mtx);
2754 prison_hold_locked(pr);
2755 mtx_unlock(&pr->pr_mtx);
2756 }
2757
2758 void
prison_proc_hold(struct prison * pr)2759 prison_proc_hold(struct prison *pr)
2760 {
2761
2762 mtx_lock(&pr->pr_mtx);
2763 KASSERT(pr->pr_uref > 0,
2764 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2765 pr->pr_uref++;
2766 mtx_unlock(&pr->pr_mtx);
2767 }
2768
2769 void
prison_proc_free(struct prison * pr)2770 prison_proc_free(struct prison *pr)
2771 {
2772
2773 mtx_lock(&pr->pr_mtx);
2774 KASSERT(pr->pr_uref > 0,
2775 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2776 if (pr->pr_uref > 1)
2777 pr->pr_uref--;
2778 else {
2779 /*
2780 * Don't remove the last user reference in this context, which
2781 * is expected to be a process that is not only locked, but
2782 * also half dead.
2783 */
2784 pr->pr_ref++;
2785 mtx_unlock(&pr->pr_mtx);
2786 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2787 return;
2788 }
2789 mtx_unlock(&pr->pr_mtx);
2790 }
2791
2792
2793 #ifdef INET
2794 /*
2795 * Restrict a prison's IP address list with its parent's, possibly replacing
2796 * it. Return true if the replacement buffer was used (or would have been).
2797 */
2798 static int
prison_restrict_ip4(struct prison * pr,struct in_addr * newip4)2799 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2800 {
2801 int ii, ij, used;
2802 struct prison *ppr;
2803
2804 ppr = pr->pr_parent;
2805 if (!(pr->pr_flags & PR_IP4_USER)) {
2806 /* This has no user settings, so just copy the parent's list. */
2807 if (pr->pr_ip4s < ppr->pr_ip4s) {
2808 /*
2809 * There's no room for the parent's list. Use the
2810 * new list buffer, which is assumed to be big enough
2811 * (if it was passed). If there's no buffer, try to
2812 * allocate one.
2813 */
2814 used = 1;
2815 if (newip4 == NULL) {
2816 newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2817 M_PRISON, M_NOWAIT);
2818 if (newip4 != NULL)
2819 used = 0;
2820 }
2821 if (newip4 != NULL) {
2822 bcopy(ppr->pr_ip4, newip4,
2823 ppr->pr_ip4s * sizeof(*newip4));
2824 free(pr->pr_ip4, M_PRISON);
2825 pr->pr_ip4 = newip4;
2826 pr->pr_ip4s = ppr->pr_ip4s;
2827 }
2828 return (used);
2829 }
2830 pr->pr_ip4s = ppr->pr_ip4s;
2831 if (pr->pr_ip4s > 0)
2832 bcopy(ppr->pr_ip4, pr->pr_ip4,
2833 pr->pr_ip4s * sizeof(*newip4));
2834 else if (pr->pr_ip4 != NULL) {
2835 free(pr->pr_ip4, M_PRISON);
2836 pr->pr_ip4 = NULL;
2837 }
2838 } else if (pr->pr_ip4s > 0) {
2839 /* Remove addresses that aren't in the parent. */
2840 for (ij = 0; ij < ppr->pr_ip4s; ij++)
2841 if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2842 break;
2843 if (ij < ppr->pr_ip4s)
2844 ii = 1;
2845 else {
2846 bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2847 --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2848 ii = 0;
2849 }
2850 for (ij = 1; ii < pr->pr_ip4s; ) {
2851 if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2852 ii++;
2853 continue;
2854 }
2855 switch (ij >= ppr->pr_ip4s ? -1 :
2856 qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2857 case -1:
2858 bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2859 (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2860 break;
2861 case 0:
2862 ii++;
2863 ij++;
2864 break;
2865 case 1:
2866 ij++;
2867 break;
2868 }
2869 }
2870 if (pr->pr_ip4s == 0) {
2871 pr->pr_flags |= PR_IP4_DISABLE;
2872 free(pr->pr_ip4, M_PRISON);
2873 pr->pr_ip4 = NULL;
2874 }
2875 }
2876 return (0);
2877 }
2878
2879 /*
2880 * Pass back primary IPv4 address of this jail.
2881 *
2882 * If not restricted return success but do not alter the address. Caller has
2883 * to make sure to initialize it correctly (e.g. INADDR_ANY).
2884 *
2885 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2886 * Address returned in NBO.
2887 */
2888 int
prison_get_ip4(struct ucred * cred,struct in_addr * ia)2889 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2890 {
2891 struct prison *pr;
2892
2893 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2894 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2895
2896 pr = cred->cr_prison;
2897 if (!(pr->pr_flags & PR_IP4))
2898 return (0);
2899 mtx_lock(&pr->pr_mtx);
2900 if (!(pr->pr_flags & PR_IP4)) {
2901 mtx_unlock(&pr->pr_mtx);
2902 return (0);
2903 }
2904 if (pr->pr_ip4 == NULL) {
2905 mtx_unlock(&pr->pr_mtx);
2906 return (EAFNOSUPPORT);
2907 }
2908
2909 ia->s_addr = pr->pr_ip4[0].s_addr;
2910 mtx_unlock(&pr->pr_mtx);
2911 return (0);
2912 }
2913
2914 /*
2915 * Return 1 if we should do proper source address selection or are not jailed.
2916 * We will return 0 if we should bypass source address selection in favour
2917 * of the primary jail IPv4 address. Only in this case *ia will be updated and
2918 * returned in NBO.
2919 * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2920 */
2921 int
prison_saddrsel_ip4(struct ucred * cred,struct in_addr * ia)2922 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2923 {
2924 struct prison *pr;
2925 struct in_addr lia;
2926 int error;
2927
2928 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2929 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2930
2931 if (!jailed(cred))
2932 return (1);
2933
2934 pr = cred->cr_prison;
2935 if (pr->pr_flags & PR_IP4_SADDRSEL)
2936 return (1);
2937
2938 lia.s_addr = INADDR_ANY;
2939 error = prison_get_ip4(cred, &lia);
2940 if (error)
2941 return (error);
2942 if (lia.s_addr == INADDR_ANY)
2943 return (1);
2944
2945 ia->s_addr = lia.s_addr;
2946 return (0);
2947 }
2948
2949 /*
2950 * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2951 */
2952 int
prison_equal_ip4(struct prison * pr1,struct prison * pr2)2953 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2954 {
2955
2956 if (pr1 == pr2)
2957 return (1);
2958
2959 /*
2960 * No need to lock since the PR_IP4_USER flag can't be altered for
2961 * existing prisons.
2962 */
2963 while (pr1 != &prison0 &&
2964 #ifdef VIMAGE
2965 !(pr1->pr_flags & PR_VNET) &&
2966 #endif
2967 !(pr1->pr_flags & PR_IP4_USER))
2968 pr1 = pr1->pr_parent;
2969 while (pr2 != &prison0 &&
2970 #ifdef VIMAGE
2971 !(pr2->pr_flags & PR_VNET) &&
2972 #endif
2973 !(pr2->pr_flags & PR_IP4_USER))
2974 pr2 = pr2->pr_parent;
2975 return (pr1 == pr2);
2976 }
2977
2978 /*
2979 * Make sure our (source) address is set to something meaningful to this
2980 * jail.
2981 *
2982 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2983 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2984 * doesn't allow IPv4. Address passed in in NBO and returned in NBO.
2985 */
2986 int
prison_local_ip4(struct ucred * cred,struct in_addr * ia)2987 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2988 {
2989 struct prison *pr;
2990 struct in_addr ia0;
2991 int error;
2992
2993 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2994 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2995
2996 pr = cred->cr_prison;
2997 if (!(pr->pr_flags & PR_IP4))
2998 return (0);
2999 mtx_lock(&pr->pr_mtx);
3000 if (!(pr->pr_flags & PR_IP4)) {
3001 mtx_unlock(&pr->pr_mtx);
3002 return (0);
3003 }
3004 if (pr->pr_ip4 == NULL) {
3005 mtx_unlock(&pr->pr_mtx);
3006 return (EAFNOSUPPORT);
3007 }
3008
3009 ia0.s_addr = ntohl(ia->s_addr);
3010 if (ia0.s_addr == INADDR_LOOPBACK) {
3011 ia->s_addr = pr->pr_ip4[0].s_addr;
3012 mtx_unlock(&pr->pr_mtx);
3013 return (0);
3014 }
3015
3016 if (ia0.s_addr == INADDR_ANY) {
3017 /*
3018 * In case there is only 1 IPv4 address, bind directly.
3019 */
3020 if (pr->pr_ip4s == 1)
3021 ia->s_addr = pr->pr_ip4[0].s_addr;
3022 mtx_unlock(&pr->pr_mtx);
3023 return (0);
3024 }
3025
3026 error = _prison_check_ip4(pr, ia);
3027 mtx_unlock(&pr->pr_mtx);
3028 return (error);
3029 }
3030
3031 /*
3032 * Rewrite destination address in case we will connect to loopback address.
3033 *
3034 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
3035 * Address passed in in NBO and returned in NBO.
3036 */
3037 int
prison_remote_ip4(struct ucred * cred,struct in_addr * ia)3038 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
3039 {
3040 struct prison *pr;
3041
3042 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3043 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3044
3045 pr = cred->cr_prison;
3046 if (!(pr->pr_flags & PR_IP4))
3047 return (0);
3048 mtx_lock(&pr->pr_mtx);
3049 if (!(pr->pr_flags & PR_IP4)) {
3050 mtx_unlock(&pr->pr_mtx);
3051 return (0);
3052 }
3053 if (pr->pr_ip4 == NULL) {
3054 mtx_unlock(&pr->pr_mtx);
3055 return (EAFNOSUPPORT);
3056 }
3057
3058 if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
3059 ia->s_addr = pr->pr_ip4[0].s_addr;
3060 mtx_unlock(&pr->pr_mtx);
3061 return (0);
3062 }
3063
3064 /*
3065 * Return success because nothing had to be changed.
3066 */
3067 mtx_unlock(&pr->pr_mtx);
3068 return (0);
3069 }
3070
3071 /*
3072 * Check if given address belongs to the jail referenced by cred/prison.
3073 *
3074 * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
3075 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3076 * doesn't allow IPv4. Address passed in in NBO.
3077 */
3078 static int
_prison_check_ip4(struct prison * pr,struct in_addr * ia)3079 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
3080 {
3081 int i, a, z, d;
3082
3083 /*
3084 * Check the primary IP.
3085 */
3086 if (pr->pr_ip4[0].s_addr == ia->s_addr)
3087 return (0);
3088
3089 /*
3090 * All the other IPs are sorted so we can do a binary search.
3091 */
3092 a = 0;
3093 z = pr->pr_ip4s - 2;
3094 while (a <= z) {
3095 i = (a + z) / 2;
3096 d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3097 if (d > 0)
3098 z = i - 1;
3099 else if (d < 0)
3100 a = i + 1;
3101 else
3102 return (0);
3103 }
3104
3105 return (EADDRNOTAVAIL);
3106 }
3107
3108 int
prison_check_ip4(struct ucred * cred,struct in_addr * ia)3109 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
3110 {
3111 struct prison *pr;
3112 int error;
3113
3114 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3115 KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3116
3117 pr = cred->cr_prison;
3118 if (!(pr->pr_flags & PR_IP4))
3119 return (0);
3120 mtx_lock(&pr->pr_mtx);
3121 if (!(pr->pr_flags & PR_IP4)) {
3122 mtx_unlock(&pr->pr_mtx);
3123 return (0);
3124 }
3125 if (pr->pr_ip4 == NULL) {
3126 mtx_unlock(&pr->pr_mtx);
3127 return (EAFNOSUPPORT);
3128 }
3129
3130 error = _prison_check_ip4(pr, ia);
3131 mtx_unlock(&pr->pr_mtx);
3132 return (error);
3133 }
3134 #endif
3135
3136 #ifdef INET6
3137 static int
prison_restrict_ip6(struct prison * pr,struct in6_addr * newip6)3138 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3139 {
3140 int ii, ij, used;
3141 struct prison *ppr;
3142
3143 ppr = pr->pr_parent;
3144 if (!(pr->pr_flags & PR_IP6_USER)) {
3145 /* This has no user settings, so just copy the parent's list. */
3146 if (pr->pr_ip6s < ppr->pr_ip6s) {
3147 /*
3148 * There's no room for the parent's list. Use the
3149 * new list buffer, which is assumed to be big enough
3150 * (if it was passed). If there's no buffer, try to
3151 * allocate one.
3152 */
3153 used = 1;
3154 if (newip6 == NULL) {
3155 newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3156 M_PRISON, M_NOWAIT);
3157 if (newip6 != NULL)
3158 used = 0;
3159 }
3160 if (newip6 != NULL) {
3161 bcopy(ppr->pr_ip6, newip6,
3162 ppr->pr_ip6s * sizeof(*newip6));
3163 free(pr->pr_ip6, M_PRISON);
3164 pr->pr_ip6 = newip6;
3165 pr->pr_ip6s = ppr->pr_ip6s;
3166 }
3167 return (used);
3168 }
3169 pr->pr_ip6s = ppr->pr_ip6s;
3170 if (pr->pr_ip6s > 0)
3171 bcopy(ppr->pr_ip6, pr->pr_ip6,
3172 pr->pr_ip6s * sizeof(*newip6));
3173 else if (pr->pr_ip6 != NULL) {
3174 free(pr->pr_ip6, M_PRISON);
3175 pr->pr_ip6 = NULL;
3176 }
3177 } else if (pr->pr_ip6s > 0) {
3178 /* Remove addresses that aren't in the parent. */
3179 for (ij = 0; ij < ppr->pr_ip6s; ij++)
3180 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3181 &ppr->pr_ip6[ij]))
3182 break;
3183 if (ij < ppr->pr_ip6s)
3184 ii = 1;
3185 else {
3186 bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3187 --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3188 ii = 0;
3189 }
3190 for (ij = 1; ii < pr->pr_ip6s; ) {
3191 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3192 &ppr->pr_ip6[0])) {
3193 ii++;
3194 continue;
3195 }
3196 switch (ij >= ppr->pr_ip6s ? -1 :
3197 qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3198 case -1:
3199 bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3200 (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3201 break;
3202 case 0:
3203 ii++;
3204 ij++;
3205 break;
3206 case 1:
3207 ij++;
3208 break;
3209 }
3210 }
3211 if (pr->pr_ip6s == 0) {
3212 pr->pr_flags |= PR_IP6_DISABLE;
3213 free(pr->pr_ip6, M_PRISON);
3214 pr->pr_ip6 = NULL;
3215 }
3216 }
3217 return 0;
3218 }
3219
3220 /*
3221 * Pass back primary IPv6 address for this jail.
3222 *
3223 * If not restricted return success but do not alter the address. Caller has
3224 * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3225 *
3226 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3227 */
3228 int
prison_get_ip6(struct ucred * cred,struct in6_addr * ia6)3229 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3230 {
3231 struct prison *pr;
3232
3233 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3234 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3235
3236 pr = cred->cr_prison;
3237 if (!(pr->pr_flags & PR_IP6))
3238 return (0);
3239 mtx_lock(&pr->pr_mtx);
3240 if (!(pr->pr_flags & PR_IP6)) {
3241 mtx_unlock(&pr->pr_mtx);
3242 return (0);
3243 }
3244 if (pr->pr_ip6 == NULL) {
3245 mtx_unlock(&pr->pr_mtx);
3246 return (EAFNOSUPPORT);
3247 }
3248
3249 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3250 mtx_unlock(&pr->pr_mtx);
3251 return (0);
3252 }
3253
3254 /*
3255 * Return 1 if we should do proper source address selection or are not jailed.
3256 * We will return 0 if we should bypass source address selection in favour
3257 * of the primary jail IPv6 address. Only in this case *ia will be updated and
3258 * returned in NBO.
3259 * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3260 */
3261 int
prison_saddrsel_ip6(struct ucred * cred,struct in6_addr * ia6)3262 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3263 {
3264 struct prison *pr;
3265 struct in6_addr lia6;
3266 int error;
3267
3268 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3269 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3270
3271 if (!jailed(cred))
3272 return (1);
3273
3274 pr = cred->cr_prison;
3275 if (pr->pr_flags & PR_IP6_SADDRSEL)
3276 return (1);
3277
3278 lia6 = in6addr_any;
3279 error = prison_get_ip6(cred, &lia6);
3280 if (error)
3281 return (error);
3282 if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3283 return (1);
3284
3285 bcopy(&lia6, ia6, sizeof(struct in6_addr));
3286 return (0);
3287 }
3288
3289 /*
3290 * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3291 */
3292 int
prison_equal_ip6(struct prison * pr1,struct prison * pr2)3293 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3294 {
3295
3296 if (pr1 == pr2)
3297 return (1);
3298
3299 while (pr1 != &prison0 &&
3300 #ifdef VIMAGE
3301 !(pr1->pr_flags & PR_VNET) &&
3302 #endif
3303 !(pr1->pr_flags & PR_IP6_USER))
3304 pr1 = pr1->pr_parent;
3305 while (pr2 != &prison0 &&
3306 #ifdef VIMAGE
3307 !(pr2->pr_flags & PR_VNET) &&
3308 #endif
3309 !(pr2->pr_flags & PR_IP6_USER))
3310 pr2 = pr2->pr_parent;
3311 return (pr1 == pr2);
3312 }
3313
3314 /*
3315 * Make sure our (source) address is set to something meaningful to this jail.
3316 *
3317 * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3318 * when needed while binding.
3319 *
3320 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3321 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3322 * doesn't allow IPv6.
3323 */
3324 int
prison_local_ip6(struct ucred * cred,struct in6_addr * ia6,int v6only)3325 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3326 {
3327 struct prison *pr;
3328 int error;
3329
3330 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3331 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3332
3333 pr = cred->cr_prison;
3334 if (!(pr->pr_flags & PR_IP6))
3335 return (0);
3336 mtx_lock(&pr->pr_mtx);
3337 if (!(pr->pr_flags & PR_IP6)) {
3338 mtx_unlock(&pr->pr_mtx);
3339 return (0);
3340 }
3341 if (pr->pr_ip6 == NULL) {
3342 mtx_unlock(&pr->pr_mtx);
3343 return (EAFNOSUPPORT);
3344 }
3345
3346 if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3347 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3348 mtx_unlock(&pr->pr_mtx);
3349 return (0);
3350 }
3351
3352 if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3353 /*
3354 * In case there is only 1 IPv6 address, and v6only is true,
3355 * then bind directly.
3356 */
3357 if (v6only != 0 && pr->pr_ip6s == 1)
3358 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3359 mtx_unlock(&pr->pr_mtx);
3360 return (0);
3361 }
3362
3363 error = _prison_check_ip6(pr, ia6);
3364 mtx_unlock(&pr->pr_mtx);
3365 return (error);
3366 }
3367
3368 /*
3369 * Rewrite destination address in case we will connect to loopback address.
3370 *
3371 * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3372 */
3373 int
prison_remote_ip6(struct ucred * cred,struct in6_addr * ia6)3374 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3375 {
3376 struct prison *pr;
3377
3378 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3379 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3380
3381 pr = cred->cr_prison;
3382 if (!(pr->pr_flags & PR_IP6))
3383 return (0);
3384 mtx_lock(&pr->pr_mtx);
3385 if (!(pr->pr_flags & PR_IP6)) {
3386 mtx_unlock(&pr->pr_mtx);
3387 return (0);
3388 }
3389 if (pr->pr_ip6 == NULL) {
3390 mtx_unlock(&pr->pr_mtx);
3391 return (EAFNOSUPPORT);
3392 }
3393
3394 if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3395 bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3396 mtx_unlock(&pr->pr_mtx);
3397 return (0);
3398 }
3399
3400 /*
3401 * Return success because nothing had to be changed.
3402 */
3403 mtx_unlock(&pr->pr_mtx);
3404 return (0);
3405 }
3406
3407 /*
3408 * Check if given address belongs to the jail referenced by cred/prison.
3409 *
3410 * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3411 * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3412 * doesn't allow IPv6.
3413 */
3414 static int
_prison_check_ip6(struct prison * pr,struct in6_addr * ia6)3415 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3416 {
3417 int i, a, z, d;
3418
3419 /*
3420 * Check the primary IP.
3421 */
3422 if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3423 return (0);
3424
3425 /*
3426 * All the other IPs are sorted so we can do a binary search.
3427 */
3428 a = 0;
3429 z = pr->pr_ip6s - 2;
3430 while (a <= z) {
3431 i = (a + z) / 2;
3432 d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3433 if (d > 0)
3434 z = i - 1;
3435 else if (d < 0)
3436 a = i + 1;
3437 else
3438 return (0);
3439 }
3440
3441 return (EADDRNOTAVAIL);
3442 }
3443
3444 int
prison_check_ip6(struct ucred * cred,struct in6_addr * ia6)3445 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3446 {
3447 struct prison *pr;
3448 int error;
3449
3450 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3451 KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3452
3453 pr = cred->cr_prison;
3454 if (!(pr->pr_flags & PR_IP6))
3455 return (0);
3456 mtx_lock(&pr->pr_mtx);
3457 if (!(pr->pr_flags & PR_IP6)) {
3458 mtx_unlock(&pr->pr_mtx);
3459 return (0);
3460 }
3461 if (pr->pr_ip6 == NULL) {
3462 mtx_unlock(&pr->pr_mtx);
3463 return (EAFNOSUPPORT);
3464 }
3465
3466 error = _prison_check_ip6(pr, ia6);
3467 mtx_unlock(&pr->pr_mtx);
3468 return (error);
3469 }
3470 #endif
3471
3472 /*
3473 * Check if a jail supports the given address family.
3474 *
3475 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3476 * if not.
3477 */
3478 int
prison_check_af(struct ucred * cred,int af)3479 prison_check_af(struct ucred *cred, int af)
3480 {
3481 struct prison *pr;
3482 int error;
3483
3484 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3485
3486 pr = cred->cr_prison;
3487 #ifdef VIMAGE
3488 /* Prisons with their own network stack are not limited. */
3489 if (prison_owns_vnet(cred))
3490 return (0);
3491 #endif
3492
3493 error = 0;
3494 switch (af)
3495 {
3496 #ifdef INET
3497 case AF_INET:
3498 if (pr->pr_flags & PR_IP4)
3499 {
3500 mtx_lock(&pr->pr_mtx);
3501 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3502 error = EAFNOSUPPORT;
3503 mtx_unlock(&pr->pr_mtx);
3504 }
3505 break;
3506 #endif
3507 #ifdef INET6
3508 case AF_INET6:
3509 if (pr->pr_flags & PR_IP6)
3510 {
3511 mtx_lock(&pr->pr_mtx);
3512 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3513 error = EAFNOSUPPORT;
3514 mtx_unlock(&pr->pr_mtx);
3515 }
3516 break;
3517 #endif
3518 case AF_LOCAL:
3519 case AF_ROUTE:
3520 break;
3521 default:
3522 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3523 error = EAFNOSUPPORT;
3524 }
3525 return (error);
3526 }
3527
3528 /*
3529 * Check if given address belongs to the jail referenced by cred (wrapper to
3530 * prison_check_ip[46]).
3531 *
3532 * Returns 0 if jail doesn't restrict the address family or if address belongs
3533 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3534 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3535 */
3536 int
prison_if(struct ucred * cred,struct sockaddr * sa)3537 prison_if(struct ucred *cred, struct sockaddr *sa)
3538 {
3539 #ifdef INET
3540 struct sockaddr_in *sai;
3541 #endif
3542 #ifdef INET6
3543 struct sockaddr_in6 *sai6;
3544 #endif
3545 int error;
3546
3547 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3548 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3549
3550 #ifdef VIMAGE
3551 if (prison_owns_vnet(cred))
3552 return (0);
3553 #endif
3554
3555 error = 0;
3556 switch (sa->sa_family)
3557 {
3558 #ifdef INET
3559 case AF_INET:
3560 sai = (struct sockaddr_in *)sa;
3561 error = prison_check_ip4(cred, &sai->sin_addr);
3562 break;
3563 #endif
3564 #ifdef INET6
3565 case AF_INET6:
3566 sai6 = (struct sockaddr_in6 *)sa;
3567 error = prison_check_ip6(cred, &sai6->sin6_addr);
3568 break;
3569 #endif
3570 default:
3571 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3572 error = EAFNOSUPPORT;
3573 }
3574 return (error);
3575 }
3576
3577 /*
3578 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3579 */
3580 int
prison_check(struct ucred * cred1,struct ucred * cred2)3581 prison_check(struct ucred *cred1, struct ucred *cred2)
3582 {
3583
3584 return ((cred1->cr_prison == cred2->cr_prison ||
3585 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3586 }
3587
3588 /*
3589 * Return 1 if p2 is a child of p1, otherwise 0.
3590 */
3591 int
prison_ischild(struct prison * pr1,struct prison * pr2)3592 prison_ischild(struct prison *pr1, struct prison *pr2)
3593 {
3594
3595 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3596 if (pr1 == pr2)
3597 return (1);
3598 return (0);
3599 }
3600
3601 /*
3602 * Return 1 if the passed credential is in a jail, otherwise 0.
3603 */
3604 int
jailed(struct ucred * cred)3605 jailed(struct ucred *cred)
3606 {
3607
3608 return (cred->cr_prison != &prison0);
3609 }
3610
3611 /*
3612 * Return 1 if the passed credential is in a jail and that jail does not
3613 * have its own virtual network stack, otherwise 0.
3614 */
3615 int
jailed_without_vnet(struct ucred * cred)3616 jailed_without_vnet(struct ucred *cred)
3617 {
3618
3619 if (!jailed(cred))
3620 return (0);
3621 #ifdef VIMAGE
3622 if (prison_owns_vnet(cred))
3623 return (0);
3624 #endif
3625
3626 return (1);
3627 }
3628
3629 /*
3630 * Return the correct hostname (domainname, et al) for the passed credential.
3631 */
3632 void
getcredhostname(struct ucred * cred,char * buf,size_t size)3633 getcredhostname(struct ucred *cred, char *buf, size_t size)
3634 {
3635 struct prison *pr;
3636
3637 /*
3638 * A NULL credential can be used to shortcut to the physical
3639 * system's hostname.
3640 */
3641 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3642 mtx_lock(&pr->pr_mtx);
3643 strlcpy(buf, pr->pr_hostname, size);
3644 mtx_unlock(&pr->pr_mtx);
3645 }
3646
3647 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)3648 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3649 {
3650
3651 mtx_lock(&cred->cr_prison->pr_mtx);
3652 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3653 mtx_unlock(&cred->cr_prison->pr_mtx);
3654 }
3655
3656 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)3657 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3658 {
3659
3660 mtx_lock(&cred->cr_prison->pr_mtx);
3661 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3662 mtx_unlock(&cred->cr_prison->pr_mtx);
3663 }
3664
3665 void
getcredhostid(struct ucred * cred,unsigned long * hostid)3666 getcredhostid(struct ucred *cred, unsigned long *hostid)
3667 {
3668
3669 mtx_lock(&cred->cr_prison->pr_mtx);
3670 *hostid = cred->cr_prison->pr_hostid;
3671 mtx_unlock(&cred->cr_prison->pr_mtx);
3672 }
3673
3674 #ifdef VIMAGE
3675 /*
3676 * Determine whether the prison represented by cred owns
3677 * its vnet rather than having it inherited.
3678 *
3679 * Returns 1 in case the prison owns the vnet, 0 otherwise.
3680 */
3681 int
prison_owns_vnet(struct ucred * cred)3682 prison_owns_vnet(struct ucred *cred)
3683 {
3684
3685 /*
3686 * vnets cannot be added/removed after jail creation,
3687 * so no need to lock here.
3688 */
3689 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3690 }
3691 #endif
3692
3693 /*
3694 * Determine whether the subject represented by cred can "see"
3695 * status of a mount point.
3696 * Returns: 0 for permitted, ENOENT otherwise.
3697 * XXX: This function should be called cr_canseemount() and should be
3698 * placed in kern_prot.c.
3699 */
3700 int
prison_canseemount(struct ucred * cred,struct mount * mp)3701 prison_canseemount(struct ucred *cred, struct mount *mp)
3702 {
3703 struct prison *pr;
3704 struct statfs *sp;
3705 size_t len;
3706
3707 pr = cred->cr_prison;
3708 if (pr->pr_enforce_statfs == 0)
3709 return (0);
3710 if (pr->pr_root->v_mount == mp)
3711 return (0);
3712 if (pr->pr_enforce_statfs == 2)
3713 return (ENOENT);
3714 /*
3715 * If jail's chroot directory is set to "/" we should be able to see
3716 * all mount-points from inside a jail.
3717 * This is ugly check, but this is the only situation when jail's
3718 * directory ends with '/'.
3719 */
3720 if (strcmp(pr->pr_path, "/") == 0)
3721 return (0);
3722 len = strlen(pr->pr_path);
3723 sp = &mp->mnt_stat;
3724 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3725 return (ENOENT);
3726 /*
3727 * Be sure that we don't have situation where jail's root directory
3728 * is "/some/path" and mount point is "/some/pathpath".
3729 */
3730 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3731 return (ENOENT);
3732 return (0);
3733 }
3734
3735 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)3736 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3737 {
3738 char jpath[MAXPATHLEN];
3739 struct prison *pr;
3740 size_t len;
3741
3742 pr = cred->cr_prison;
3743 if (pr->pr_enforce_statfs == 0)
3744 return;
3745 if (prison_canseemount(cred, mp) != 0) {
3746 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3747 strlcpy(sp->f_mntonname, "[restricted]",
3748 sizeof(sp->f_mntonname));
3749 return;
3750 }
3751 if (pr->pr_root->v_mount == mp) {
3752 /*
3753 * Clear current buffer data, so we are sure nothing from
3754 * the valid path left there.
3755 */
3756 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3757 *sp->f_mntonname = '/';
3758 return;
3759 }
3760 /*
3761 * If jail's chroot directory is set to "/" we should be able to see
3762 * all mount-points from inside a jail.
3763 */
3764 if (strcmp(pr->pr_path, "/") == 0)
3765 return;
3766 len = strlen(pr->pr_path);
3767 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3768 /*
3769 * Clear current buffer data, so we are sure nothing from
3770 * the valid path left there.
3771 */
3772 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3773 if (*jpath == '\0') {
3774 /* Should never happen. */
3775 *sp->f_mntonname = '/';
3776 } else {
3777 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3778 }
3779 }
3780
3781 /*
3782 * Check with permission for a specific privilege is granted within jail. We
3783 * have a specific list of accepted privileges; the rest are denied.
3784 */
3785 int
prison_priv_check(struct ucred * cred,int priv)3786 prison_priv_check(struct ucred *cred, int priv)
3787 {
3788
3789 if (!jailed(cred))
3790 return (0);
3791
3792 #ifdef VIMAGE
3793 /*
3794 * Privileges specific to prisons with a virtual network stack.
3795 * There might be a duplicate entry here in case the privilege
3796 * is only granted conditionally in the legacy jail case.
3797 */
3798 switch (priv) {
3799 #ifdef notyet
3800 /*
3801 * NFS-specific privileges.
3802 */
3803 case PRIV_NFS_DAEMON:
3804 case PRIV_NFS_LOCKD:
3805 #endif
3806 /*
3807 * Network stack privileges.
3808 */
3809 case PRIV_NET_BRIDGE:
3810 case PRIV_NET_GRE:
3811 case PRIV_NET_BPF:
3812 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3813 case PRIV_NET_ROUTE:
3814 case PRIV_NET_TAP:
3815 case PRIV_NET_SETIFMTU:
3816 case PRIV_NET_SETIFFLAGS:
3817 case PRIV_NET_SETIFCAP:
3818 case PRIV_NET_SETIFDESCR:
3819 case PRIV_NET_SETIFNAME :
3820 case PRIV_NET_SETIFMETRIC:
3821 case PRIV_NET_SETIFPHYS:
3822 case PRIV_NET_SETIFMAC:
3823 case PRIV_NET_ADDMULTI:
3824 case PRIV_NET_DELMULTI:
3825 case PRIV_NET_HWIOCTL:
3826 case PRIV_NET_SETLLADDR:
3827 case PRIV_NET_ADDIFGROUP:
3828 case PRIV_NET_DELIFGROUP:
3829 case PRIV_NET_IFCREATE:
3830 case PRIV_NET_IFDESTROY:
3831 case PRIV_NET_ADDIFADDR:
3832 case PRIV_NET_DELIFADDR:
3833 case PRIV_NET_LAGG:
3834 case PRIV_NET_GIF:
3835 case PRIV_NET_SETIFVNET:
3836 case PRIV_NET_SETIFFIB:
3837
3838 /*
3839 * 802.11-related privileges.
3840 */
3841 case PRIV_NET80211_GETKEY:
3842 #ifdef notyet
3843 case PRIV_NET80211_MANAGE: /* XXX-BZ discuss with sam@ */
3844 #endif
3845
3846 #ifdef notyet
3847 /*
3848 * AppleTalk privileges.
3849 */
3850 case PRIV_NETATALK_RESERVEDPORT:
3851
3852 /*
3853 * ATM privileges.
3854 */
3855 case PRIV_NETATM_CFG:
3856 case PRIV_NETATM_ADD:
3857 case PRIV_NETATM_DEL:
3858 case PRIV_NETATM_SET:
3859
3860 /*
3861 * Bluetooth privileges.
3862 */
3863 case PRIV_NETBLUETOOTH_RAW:
3864 #endif
3865
3866 /*
3867 * Netgraph and netgraph module privileges.
3868 */
3869 case PRIV_NETGRAPH_CONTROL:
3870 #ifdef notyet
3871 case PRIV_NETGRAPH_TTY:
3872 #endif
3873
3874 /*
3875 * IPv4 and IPv6 privileges.
3876 */
3877 case PRIV_NETINET_IPFW:
3878 case PRIV_NETINET_DIVERT:
3879 case PRIV_NETINET_PF:
3880 case PRIV_NETINET_DUMMYNET:
3881 case PRIV_NETINET_CARP:
3882 case PRIV_NETINET_MROUTE:
3883 case PRIV_NETINET_RAW:
3884 case PRIV_NETINET_ADDRCTRL6:
3885 case PRIV_NETINET_ND6:
3886 case PRIV_NETINET_SCOPE6:
3887 case PRIV_NETINET_ALIFETIME6:
3888 case PRIV_NETINET_IPSEC:
3889 case PRIV_NETINET_BINDANY:
3890
3891 #ifdef notyet
3892 /*
3893 * IPX/SPX privileges.
3894 */
3895 case PRIV_NETIPX_RESERVEDPORT:
3896 case PRIV_NETIPX_RAW:
3897
3898 /*
3899 * NCP privileges.
3900 */
3901 case PRIV_NETNCP:
3902
3903 /*
3904 * SMB privileges.
3905 */
3906 case PRIV_NETSMB:
3907 #endif
3908
3909 /*
3910 * No default: or deny here.
3911 * In case of no permit fall through to next switch().
3912 */
3913 if (cred->cr_prison->pr_flags & PR_VNET)
3914 return (0);
3915 }
3916 #endif /* VIMAGE */
3917
3918 switch (priv) {
3919
3920 /*
3921 * Allow ktrace privileges for root in jail.
3922 */
3923 case PRIV_KTRACE:
3924
3925 #if 0
3926 /*
3927 * Allow jailed processes to configure audit identity and
3928 * submit audit records (login, etc). In the future we may
3929 * want to further refine the relationship between audit and
3930 * jail.
3931 */
3932 case PRIV_AUDIT_GETAUDIT:
3933 case PRIV_AUDIT_SETAUDIT:
3934 case PRIV_AUDIT_SUBMIT:
3935 #endif
3936
3937 /*
3938 * Allow jailed processes to manipulate process UNIX
3939 * credentials in any way they see fit.
3940 */
3941 case PRIV_CRED_SETUID:
3942 case PRIV_CRED_SETEUID:
3943 case PRIV_CRED_SETGID:
3944 case PRIV_CRED_SETEGID:
3945 case PRIV_CRED_SETGROUPS:
3946 case PRIV_CRED_SETREUID:
3947 case PRIV_CRED_SETREGID:
3948 case PRIV_CRED_SETRESUID:
3949 case PRIV_CRED_SETRESGID:
3950
3951 /*
3952 * Jail implements visibility constraints already, so allow
3953 * jailed root to override uid/gid-based constraints.
3954 */
3955 case PRIV_SEEOTHERGIDS:
3956 case PRIV_SEEOTHERUIDS:
3957
3958 /*
3959 * Jail implements inter-process debugging limits already, so
3960 * allow jailed root various debugging privileges.
3961 */
3962 case PRIV_DEBUG_DIFFCRED:
3963 case PRIV_DEBUG_SUGID:
3964 case PRIV_DEBUG_UNPRIV:
3965
3966 /*
3967 * Allow jail to set various resource limits and login
3968 * properties, and for now, exceed process resource limits.
3969 */
3970 case PRIV_PROC_LIMIT:
3971 case PRIV_PROC_SETLOGIN:
3972 case PRIV_PROC_SETRLIMIT:
3973
3974 /*
3975 * System V and POSIX IPC privileges are granted in jail.
3976 */
3977 case PRIV_IPC_READ:
3978 case PRIV_IPC_WRITE:
3979 case PRIV_IPC_ADMIN:
3980 case PRIV_IPC_MSGSIZE:
3981 case PRIV_MQ_ADMIN:
3982
3983 /*
3984 * Jail operations within a jail work on child jails.
3985 */
3986 case PRIV_JAIL_ATTACH:
3987 case PRIV_JAIL_SET:
3988 case PRIV_JAIL_REMOVE:
3989
3990 /*
3991 * Jail implements its own inter-process limits, so allow
3992 * root processes in jail to change scheduling on other
3993 * processes in the same jail. Likewise for signalling.
3994 */
3995 case PRIV_SCHED_DIFFCRED:
3996 case PRIV_SCHED_CPUSET:
3997 case PRIV_SIGNAL_DIFFCRED:
3998 case PRIV_SIGNAL_SUGID:
3999
4000 /*
4001 * Allow jailed processes to write to sysctls marked as jail
4002 * writable.
4003 */
4004 case PRIV_SYSCTL_WRITEJAIL:
4005
4006 /*
4007 * Allow root in jail to manage a variety of quota
4008 * properties. These should likely be conditional on a
4009 * configuration option.
4010 */
4011 case PRIV_VFS_GETQUOTA:
4012 case PRIV_VFS_SETQUOTA:
4013
4014 /*
4015 * Since Jail relies on chroot() to implement file system
4016 * protections, grant many VFS privileges to root in jail.
4017 * Be careful to exclude mount-related and NFS-related
4018 * privileges.
4019 */
4020 case PRIV_VFS_READ:
4021 case PRIV_VFS_WRITE:
4022 case PRIV_VFS_ADMIN:
4023 case PRIV_VFS_EXEC:
4024 case PRIV_VFS_LOOKUP:
4025 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4026 case PRIV_VFS_CHFLAGS_DEV:
4027 case PRIV_VFS_CHOWN:
4028 case PRIV_VFS_CHROOT:
4029 case PRIV_VFS_RETAINSUGID:
4030 case PRIV_VFS_FCHROOT:
4031 case PRIV_VFS_LINK:
4032 case PRIV_VFS_SETGID:
4033 case PRIV_VFS_STAT:
4034 case PRIV_VFS_STICKYFILE:
4035
4036 /*
4037 * As in the non-jail case, non-root users are expected to be
4038 * able to read kernel/phyiscal memory (provided /dev/[k]mem
4039 * exists in the jail and they have permission to access it).
4040 */
4041 case PRIV_KMEM_READ:
4042 return (0);
4043
4044 /*
4045 * Depending on the global setting, allow privilege of
4046 * setting system flags.
4047 */
4048 case PRIV_VFS_SYSFLAGS:
4049 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4050 return (0);
4051 else
4052 return (EPERM);
4053
4054 /*
4055 * Depending on the global setting, allow privilege of
4056 * mounting/unmounting file systems.
4057 */
4058 case PRIV_VFS_MOUNT:
4059 case PRIV_VFS_UNMOUNT:
4060 case PRIV_VFS_MOUNT_NONUSER:
4061 case PRIV_VFS_MOUNT_OWNER:
4062 if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
4063 cred->cr_prison->pr_enforce_statfs < 2)
4064 return (0);
4065 else
4066 return (EPERM);
4067
4068 /*
4069 * Allow jailed root to bind reserved ports and reuse in-use
4070 * ports.
4071 */
4072 case PRIV_NETINET_RESERVEDPORT:
4073 case PRIV_NETINET_REUSEPORT:
4074 return (0);
4075
4076 /*
4077 * Allow jailed root to set certain IPv4/6 (option) headers.
4078 */
4079 case PRIV_NETINET_SETHDROPTS:
4080 return (0);
4081
4082 /*
4083 * Conditionally allow creating raw sockets in jail.
4084 */
4085 case PRIV_NETINET_RAW:
4086 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4087 return (0);
4088 else
4089 return (EPERM);
4090
4091 /*
4092 * Since jail implements its own visibility limits on netstat
4093 * sysctls, allow getcred. This allows identd to work in
4094 * jail.
4095 */
4096 case PRIV_NETINET_GETCRED:
4097 return (0);
4098
4099 /*
4100 * Allow jailed root to set loginclass.
4101 */
4102 case PRIV_PROC_SETLOGINCLASS:
4103 return (0);
4104
4105 default:
4106 /*
4107 * In all remaining cases, deny the privilege request. This
4108 * includes almost all network privileges, many system
4109 * configuration privileges.
4110 */
4111 return (EPERM);
4112 }
4113 }
4114
4115 /*
4116 * Return the part of pr2's name that is relative to pr1, or the whole name
4117 * if it does not directly follow.
4118 */
4119
4120 char *
prison_name(struct prison * pr1,struct prison * pr2)4121 prison_name(struct prison *pr1, struct prison *pr2)
4122 {
4123 char *name;
4124
4125 /* Jails see themselves as "0" (if they see themselves at all). */
4126 if (pr1 == pr2)
4127 return "0";
4128 name = pr2->pr_name;
4129 if (prison_ischild(pr1, pr2)) {
4130 /*
4131 * pr1 isn't locked (and allprison_lock may not be either)
4132 * so its length can't be counted on. But the number of dots
4133 * can be counted on - and counted.
4134 */
4135 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4136 name = strchr(name, '.') + 1;
4137 }
4138 return (name);
4139 }
4140
4141 /*
4142 * Return the part of pr2's path that is relative to pr1, or the whole path
4143 * if it does not directly follow.
4144 */
4145 static char *
prison_path(struct prison * pr1,struct prison * pr2)4146 prison_path(struct prison *pr1, struct prison *pr2)
4147 {
4148 char *path1, *path2;
4149 int len1;
4150
4151 path1 = pr1->pr_path;
4152 path2 = pr2->pr_path;
4153 if (!strcmp(path1, "/"))
4154 return (path2);
4155 len1 = strlen(path1);
4156 if (strncmp(path1, path2, len1))
4157 return (path2);
4158 if (path2[len1] == '\0')
4159 return "/";
4160 if (path2[len1] == '/')
4161 return (path2 + len1);
4162 return (path2);
4163 }
4164
4165
4166 /*
4167 * Jail-related sysctls.
4168 */
4169 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4170 "Jails");
4171
4172 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4173 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4174 {
4175 struct xprison *xp;
4176 struct prison *pr, *cpr;
4177 #ifdef INET
4178 struct in_addr *ip4 = NULL;
4179 int ip4s = 0;
4180 #endif
4181 #ifdef INET6
4182 struct in6_addr *ip6 = NULL;
4183 int ip6s = 0;
4184 #endif
4185 int descend, error;
4186
4187 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4188 pr = req->td->td_ucred->cr_prison;
4189 error = 0;
4190 sx_slock(&allprison_lock);
4191 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4192 #if defined(INET) || defined(INET6)
4193 again:
4194 #endif
4195 mtx_lock(&cpr->pr_mtx);
4196 #ifdef INET
4197 if (cpr->pr_ip4s > 0) {
4198 if (ip4s < cpr->pr_ip4s) {
4199 ip4s = cpr->pr_ip4s;
4200 mtx_unlock(&cpr->pr_mtx);
4201 ip4 = realloc(ip4, ip4s *
4202 sizeof(struct in_addr), M_TEMP, M_WAITOK);
4203 goto again;
4204 }
4205 bcopy(cpr->pr_ip4, ip4,
4206 cpr->pr_ip4s * sizeof(struct in_addr));
4207 }
4208 #endif
4209 #ifdef INET6
4210 if (cpr->pr_ip6s > 0) {
4211 if (ip6s < cpr->pr_ip6s) {
4212 ip6s = cpr->pr_ip6s;
4213 mtx_unlock(&cpr->pr_mtx);
4214 ip6 = realloc(ip6, ip6s *
4215 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4216 goto again;
4217 }
4218 bcopy(cpr->pr_ip6, ip6,
4219 cpr->pr_ip6s * sizeof(struct in6_addr));
4220 }
4221 #endif
4222 if (cpr->pr_ref == 0) {
4223 mtx_unlock(&cpr->pr_mtx);
4224 continue;
4225 }
4226 bzero(xp, sizeof(*xp));
4227 xp->pr_version = XPRISON_VERSION;
4228 xp->pr_id = cpr->pr_id;
4229 xp->pr_state = cpr->pr_uref > 0
4230 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4231 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4232 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4233 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4234 #ifdef INET
4235 xp->pr_ip4s = cpr->pr_ip4s;
4236 #endif
4237 #ifdef INET6
4238 xp->pr_ip6s = cpr->pr_ip6s;
4239 #endif
4240 mtx_unlock(&cpr->pr_mtx);
4241 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4242 if (error)
4243 break;
4244 #ifdef INET
4245 if (xp->pr_ip4s > 0) {
4246 error = SYSCTL_OUT(req, ip4,
4247 xp->pr_ip4s * sizeof(struct in_addr));
4248 if (error)
4249 break;
4250 }
4251 #endif
4252 #ifdef INET6
4253 if (xp->pr_ip6s > 0) {
4254 error = SYSCTL_OUT(req, ip6,
4255 xp->pr_ip6s * sizeof(struct in6_addr));
4256 if (error)
4257 break;
4258 }
4259 #endif
4260 }
4261 sx_sunlock(&allprison_lock);
4262 free(xp, M_TEMP);
4263 #ifdef INET
4264 free(ip4, M_TEMP);
4265 #endif
4266 #ifdef INET6
4267 free(ip6, M_TEMP);
4268 #endif
4269 return (error);
4270 }
4271
4272 SYSCTL_OID(_security_jail, OID_AUTO, list,
4273 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4274 sysctl_jail_list, "S", "List of active jails");
4275
4276 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)4277 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4278 {
4279 int error, injail;
4280
4281 injail = jailed(req->td->td_ucred);
4282 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4283
4284 return (error);
4285 }
4286
4287 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4288 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4289 sysctl_jail_jailed, "I", "Process in jail?");
4290
4291 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)4292 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4293 {
4294 int error, havevnet;
4295 #ifdef VIMAGE
4296 struct ucred *cred = req->td->td_ucred;
4297
4298 havevnet = jailed(cred) && prison_owns_vnet(cred);
4299 #else
4300 havevnet = 0;
4301 #endif
4302 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4303
4304 return (error);
4305 }
4306
4307 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4308 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4309 sysctl_jail_vnet, "I", "Jail owns VNET?");
4310
4311 #if defined(INET) || defined(INET6)
4312 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4313 &jail_max_af_ips, 0,
4314 "Number of IP addresses a jail may have at most per address family (deprecated)");
4315 #endif
4316
4317 /*
4318 * Default parameters for jail(2) compatibility. For historical reasons,
4319 * the sysctl names have varying similarity to the parameter names. Prisons
4320 * just see their own parameters, and can't change them.
4321 */
4322 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)4323 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4324 {
4325 struct prison *pr;
4326 int allow, error, i;
4327
4328 pr = req->td->td_ucred->cr_prison;
4329 allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4330
4331 /* Get the current flag value, and convert it to a boolean. */
4332 i = (allow & arg2) ? 1 : 0;
4333 if (arg1 != NULL)
4334 i = !i;
4335 error = sysctl_handle_int(oidp, &i, 0, req);
4336 if (error || !req->newptr)
4337 return (error);
4338 i = i ? arg2 : 0;
4339 if (arg1 != NULL)
4340 i ^= arg2;
4341 /*
4342 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4343 * for writing.
4344 */
4345 mtx_lock(&prison0.pr_mtx);
4346 jail_default_allow = (jail_default_allow & ~arg2) | i;
4347 mtx_unlock(&prison0.pr_mtx);
4348 return (0);
4349 }
4350
4351 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4352 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4353 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4354 "Processes in jail can set their hostnames (deprecated)");
4355 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4356 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4357 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4358 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4359 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4360 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4361 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4362 "Processes in jail can use System V IPC primitives (deprecated)");
4363 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4364 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4365 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4366 "Prison root can create raw sockets (deprecated)");
4367 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4368 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4369 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4370 "Processes in jail can alter system file flags (deprecated)");
4371 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4372 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4373 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4374 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4375 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4376 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4377 NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4378 "Processes in jail can mount the devfs file system (deprecated)");
4379 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4380 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4381 NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4382 "Processes in jail can mount the fdescfs file system (deprecated)");
4383 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4384 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4385 NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4386 "Processes in jail can mount the nullfs file system (deprecated)");
4387 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4388 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4389 NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4390 "Processes in jail can mount the procfs file system (deprecated)");
4391 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
4392 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4393 NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4394 "Processes in jail can mount the linprocfs file system (deprecated)");
4395 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4396 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4397 NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4398 "Processes in jail can mount the linsysfs file system (deprecated)");
4399 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4400 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4401 NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4402 "Processes in jail can mount the tmpfs file system (deprecated)");
4403 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4404 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4405 NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4406 "Processes in jail can mount the zfs file system (deprecated)");
4407
4408 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)4409 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4410 {
4411 struct prison *pr;
4412 int level, error;
4413
4414 pr = req->td->td_ucred->cr_prison;
4415 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4416 error = sysctl_handle_int(oidp, &level, 0, req);
4417 if (error || !req->newptr)
4418 return (error);
4419 *(int *)arg1 = level;
4420 return (0);
4421 }
4422
4423 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4424 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4425 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4426 sysctl_jail_default_level, "I",
4427 "Processes in jail cannot see all mounted file systems (deprecated)");
4428
4429 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4430 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4431 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4432 sysctl_jail_default_level, "I",
4433 "Ruleset for the devfs filesystem in jail (deprecated)");
4434
4435 /*
4436 * Nodes to describe jail parameters. Maximum length of string parameters
4437 * is returned in the string itself, and the other parameters exist merely
4438 * to make themselves and their types known.
4439 */
4440 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4441 "Jail parameters");
4442
4443 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)4444 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4445 {
4446 int i;
4447 long l;
4448 size_t s;
4449 char numbuf[12];
4450
4451 switch (oidp->oid_kind & CTLTYPE)
4452 {
4453 case CTLTYPE_LONG:
4454 case CTLTYPE_ULONG:
4455 l = 0;
4456 #ifdef SCTL_MASK32
4457 if (!(req->flags & SCTL_MASK32))
4458 #endif
4459 return (SYSCTL_OUT(req, &l, sizeof(l)));
4460 case CTLTYPE_INT:
4461 case CTLTYPE_UINT:
4462 i = 0;
4463 return (SYSCTL_OUT(req, &i, sizeof(i)));
4464 case CTLTYPE_STRING:
4465 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4466 return
4467 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4468 case CTLTYPE_STRUCT:
4469 s = (size_t)arg2;
4470 return (SYSCTL_OUT(req, &s, sizeof(s)));
4471 }
4472 return (0);
4473 }
4474
4475 /*
4476 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4477 * jail creation time but cannot be changed in an existing jail.
4478 */
4479 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4480 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4481 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4482 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4483 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4484 "I", "Jail secure level");
4485 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4486 "Jail value for kern.osreldate and uname -K");
4487 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4488 "Jail value for kern.osrelease and uname -r");
4489 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4490 "I", "Jail cannot see all mounted file systems");
4491 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4492 "I", "Ruleset for in-jail devfs mounts");
4493 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4494 "B", "Jail persistence");
4495 #ifdef VIMAGE
4496 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4497 "E,jailsys", "Virtual network stack");
4498 #endif
4499 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4500 "B", "Jail is in the process of shutting down");
4501
4502 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4503 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4504 "I", "Current number of child jails");
4505 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4506 "I", "Maximum number of child jails");
4507
4508 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4509 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4510 "Jail hostname");
4511 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4512 "Jail NIS domainname");
4513 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4514 "Jail host UUID");
4515 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4516 "LU", "Jail host ID");
4517
4518 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4519 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4520
4521 #ifdef INET
4522 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4523 "Jail IPv4 address virtualization");
4524 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4525 "S,in_addr,a", "Jail IPv4 addresses");
4526 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4527 "B", "Do (not) use IPv4 source address selection rather than the "
4528 "primary jail IPv4 address.");
4529 #endif
4530 #ifdef INET6
4531 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4532 "Jail IPv6 address virtualization");
4533 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4534 "S,in6_addr,a", "Jail IPv6 addresses");
4535 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4536 "B", "Do (not) use IPv6 source address selection rather than the "
4537 "primary jail IPv6 address.");
4538 #endif
4539
4540 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4541 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4542 "B", "Jail may set hostname");
4543 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4544 "B", "Jail may use SYSV IPC");
4545 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4546 "B", "Jail may create raw sockets");
4547 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4548 "B", "Jail may alter system file flags");
4549 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4550 "B", "Jail may set file quotas");
4551 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4552 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4553
4554 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4555 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4556 "B", "Jail may mount/unmount jail-friendly file systems in general");
4557 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4558 "B", "Jail may mount the devfs file system");
4559 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4560 "B", "Jail may mount the fdescfs file system");
4561 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4562 "B", "Jail may mount the nullfs file system");
4563 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4564 "B", "Jail may mount the procfs file system");
4565 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4566 "B", "Jail may mount the linprocfs file system");
4567 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4568 "B", "Jail may mount the linsysfs file system");
4569 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4570 "B", "Jail may mount the tmpfs file system");
4571 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4572 "B", "Jail may mount the zfs file system");
4573
4574 #ifdef RACCT
4575 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void * arg2,void * arg3)4576 prison_racct_foreach(void (*callback)(struct racct *racct,
4577 void *arg2, void *arg3), void *arg2, void *arg3)
4578 {
4579 struct prison_racct *prr;
4580
4581 ASSERT_RACCT_ENABLED();
4582
4583 sx_slock(&allprison_lock);
4584 LIST_FOREACH(prr, &allprison_racct, prr_next)
4585 (callback)(prr->prr_racct, arg2, arg3);
4586 sx_sunlock(&allprison_lock);
4587 }
4588
4589 static struct prison_racct *
prison_racct_find_locked(const char * name)4590 prison_racct_find_locked(const char *name)
4591 {
4592 struct prison_racct *prr;
4593
4594 ASSERT_RACCT_ENABLED();
4595 sx_assert(&allprison_lock, SA_XLOCKED);
4596
4597 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4598 return (NULL);
4599
4600 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4601 if (strcmp(name, prr->prr_name) != 0)
4602 continue;
4603
4604 /* Found prison_racct with a matching name? */
4605 prison_racct_hold(prr);
4606 return (prr);
4607 }
4608
4609 /* Add new prison_racct. */
4610 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4611 racct_create(&prr->prr_racct);
4612
4613 strcpy(prr->prr_name, name);
4614 refcount_init(&prr->prr_refcount, 1);
4615 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4616
4617 return (prr);
4618 }
4619
4620 struct prison_racct *
prison_racct_find(const char * name)4621 prison_racct_find(const char *name)
4622 {
4623 struct prison_racct *prr;
4624
4625 ASSERT_RACCT_ENABLED();
4626
4627 sx_xlock(&allprison_lock);
4628 prr = prison_racct_find_locked(name);
4629 sx_xunlock(&allprison_lock);
4630 return (prr);
4631 }
4632
4633 void
prison_racct_hold(struct prison_racct * prr)4634 prison_racct_hold(struct prison_racct *prr)
4635 {
4636
4637 ASSERT_RACCT_ENABLED();
4638
4639 refcount_acquire(&prr->prr_refcount);
4640 }
4641
4642 static void
prison_racct_free_locked(struct prison_racct * prr)4643 prison_racct_free_locked(struct prison_racct *prr)
4644 {
4645
4646 ASSERT_RACCT_ENABLED();
4647 sx_assert(&allprison_lock, SA_XLOCKED);
4648
4649 if (refcount_release(&prr->prr_refcount)) {
4650 racct_destroy(&prr->prr_racct);
4651 LIST_REMOVE(prr, prr_next);
4652 free(prr, M_PRISON_RACCT);
4653 }
4654 }
4655
4656 void
prison_racct_free(struct prison_racct * prr)4657 prison_racct_free(struct prison_racct *prr)
4658 {
4659 int old;
4660
4661 ASSERT_RACCT_ENABLED();
4662 sx_assert(&allprison_lock, SA_UNLOCKED);
4663
4664 old = prr->prr_refcount;
4665 if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4666 return;
4667
4668 sx_xlock(&allprison_lock);
4669 prison_racct_free_locked(prr);
4670 sx_xunlock(&allprison_lock);
4671 }
4672
4673 static void
prison_racct_attach(struct prison * pr)4674 prison_racct_attach(struct prison *pr)
4675 {
4676 struct prison_racct *prr;
4677
4678 ASSERT_RACCT_ENABLED();
4679 sx_assert(&allprison_lock, SA_XLOCKED);
4680
4681 prr = prison_racct_find_locked(pr->pr_name);
4682 KASSERT(prr != NULL, ("cannot find prison_racct"));
4683
4684 pr->pr_prison_racct = prr;
4685 }
4686
4687 /*
4688 * Handle jail renaming. From the racct point of view, renaming means
4689 * moving from one prison_racct to another.
4690 */
4691 static void
prison_racct_modify(struct prison * pr)4692 prison_racct_modify(struct prison *pr)
4693 {
4694 struct proc *p;
4695 struct ucred *cred;
4696 struct prison_racct *oldprr;
4697
4698 ASSERT_RACCT_ENABLED();
4699
4700 sx_slock(&allproc_lock);
4701 sx_xlock(&allprison_lock);
4702
4703 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4704 sx_xunlock(&allprison_lock);
4705 sx_sunlock(&allproc_lock);
4706 return;
4707 }
4708
4709 oldprr = pr->pr_prison_racct;
4710 pr->pr_prison_racct = NULL;
4711
4712 prison_racct_attach(pr);
4713
4714 /*
4715 * Move resource utilisation records.
4716 */
4717 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4718
4719 /*
4720 * Force rctl to reattach rules to processes.
4721 */
4722 FOREACH_PROC_IN_SYSTEM(p) {
4723 PROC_LOCK(p);
4724 cred = crhold(p->p_ucred);
4725 PROC_UNLOCK(p);
4726 racct_proc_ucred_changed(p, cred, cred);
4727 crfree(cred);
4728 }
4729
4730 sx_sunlock(&allproc_lock);
4731 prison_racct_free_locked(oldprr);
4732 sx_xunlock(&allprison_lock);
4733 }
4734
4735 static void
prison_racct_detach(struct prison * pr)4736 prison_racct_detach(struct prison *pr)
4737 {
4738
4739 ASSERT_RACCT_ENABLED();
4740 sx_assert(&allprison_lock, SA_UNLOCKED);
4741
4742 if (pr->pr_prison_racct == NULL)
4743 return;
4744 prison_racct_free(pr->pr_prison_racct);
4745 pr->pr_prison_racct = NULL;
4746 }
4747 #endif /* RACCT */
4748
4749 #ifdef DDB
4750
4751 static void
db_show_prison(struct prison * pr)4752 db_show_prison(struct prison *pr)
4753 {
4754 int fi;
4755 #if defined(INET) || defined(INET6)
4756 int ii;
4757 #endif
4758 unsigned jsf;
4759 #ifdef INET6
4760 char ip6buf[INET6_ADDRSTRLEN];
4761 #endif
4762
4763 db_printf("prison %p:\n", pr);
4764 db_printf(" jid = %d\n", pr->pr_id);
4765 db_printf(" name = %s\n", pr->pr_name);
4766 db_printf(" parent = %p\n", pr->pr_parent);
4767 db_printf(" ref = %d\n", pr->pr_ref);
4768 db_printf(" uref = %d\n", pr->pr_uref);
4769 db_printf(" path = %s\n", pr->pr_path);
4770 db_printf(" cpuset = %d\n", pr->pr_cpuset
4771 ? pr->pr_cpuset->cs_id : -1);
4772 #ifdef VIMAGE
4773 db_printf(" vnet = %p\n", pr->pr_vnet);
4774 #endif
4775 db_printf(" root = %p\n", pr->pr_root);
4776 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4777 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4778 db_printf(" children.max = %d\n", pr->pr_childmax);
4779 db_printf(" children.cur = %d\n", pr->pr_childcount);
4780 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4781 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4782 db_printf(" flags = 0x%x", pr->pr_flags);
4783 for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4784 fi++)
4785 if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4786 db_printf(" %s", pr_flag_names[fi]);
4787 for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4788 fi++) {
4789 jsf = pr->pr_flags &
4790 (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4791 db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4792 pr_flag_jailsys[fi].disable &&
4793 (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4794 : (jsf == pr_flag_jailsys[fi].new) ? "new"
4795 : "inherit");
4796 }
4797 db_printf(" allow = 0x%x", pr->pr_allow);
4798 for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4799 fi++)
4800 if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4801 db_printf(" %s", pr_allow_names[fi]);
4802 db_printf("\n");
4803 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4804 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4805 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4806 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4807 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4808 #ifdef INET
4809 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4810 for (ii = 0; ii < pr->pr_ip4s; ii++)
4811 db_printf(" %s %s\n",
4812 ii == 0 ? "ip4.addr =" : " ",
4813 inet_ntoa(pr->pr_ip4[ii]));
4814 #endif
4815 #ifdef INET6
4816 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4817 for (ii = 0; ii < pr->pr_ip6s; ii++)
4818 db_printf(" %s %s\n",
4819 ii == 0 ? "ip6.addr =" : " ",
4820 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4821 #endif
4822 }
4823
DB_SHOW_COMMAND(prison,db_show_prison_command)4824 DB_SHOW_COMMAND(prison, db_show_prison_command)
4825 {
4826 struct prison *pr;
4827
4828 if (!have_addr) {
4829 /*
4830 * Show all prisons in the list, and prison0 which is not
4831 * listed.
4832 */
4833 db_show_prison(&prison0);
4834 if (!db_pager_quit) {
4835 TAILQ_FOREACH(pr, &allprison, pr_list) {
4836 db_show_prison(pr);
4837 if (db_pager_quit)
4838 break;
4839 }
4840 }
4841 return;
4842 }
4843
4844 if (addr == 0)
4845 pr = &prison0;
4846 else {
4847 /* Look for a prison with the ID and with references. */
4848 TAILQ_FOREACH(pr, &allprison, pr_list)
4849 if (pr->pr_id == addr && pr->pr_ref > 0)
4850 break;
4851 if (pr == NULL)
4852 /* Look again, without requiring a reference. */
4853 TAILQ_FOREACH(pr, &allprison, pr_list)
4854 if (pr->pr_id == addr)
4855 break;
4856 if (pr == NULL)
4857 /* Assume address points to a valid prison. */
4858 pr = (struct prison *)addr;
4859 }
4860 db_show_prison(pr);
4861 }
4862
4863 #endif /* DDB */
4864