1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1999 Poul-Henning Kamp.
5 * Copyright (c) 2008 Bjoern A. Zeeb.
6 * Copyright (c) 2009 James Gritton.
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 #include "opt_ddb.h"
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_nfs.h"
36
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/file.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/epoch.h>
49 #include <sys/event.h>
50 #include <sys/taskqueue.h>
51 #include <sys/fcntl.h>
52 #include <sys/jail.h>
53 #include <sys/jaildesc.h>
54 #include <sys/linker.h>
55 #include <sys/lock.h>
56 #include <sys/mman.h>
57 #include <sys/mutex.h>
58 #include <sys/racct.h>
59 #include <sys/rctl.h>
60 #include <sys/refcount.h>
61 #include <sys/sx.h>
62 #include <sys/sysent.h>
63 #include <sys/namei.h>
64 #include <sys/mount.h>
65 #include <sys/queue.h>
66 #include <sys/socket.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/uuid.h>
70 #include <sys/vnode.h>
71
72 #include <net/if.h>
73 #include <net/vnet.h>
74
75 #include <netinet/in.h>
76
77 #ifdef DDB
78 #include <ddb/ddb.h>
79 #endif /* DDB */
80
81 #include <security/mac/mac_framework.h>
82
83 #define PRISON0_HOSTUUID_MODULE "hostuuid"
84
85 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
86 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
87
88 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
89 #ifdef INET
90 #ifdef INET6
91 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
92 #else
93 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
94 #endif
95 #else /* !INET */
96 #ifdef INET6
97 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
98 #else
99 #define _PR_IP_SADDRSEL 0
100 #endif
101 #endif
102
103 /* prison0 describes what is "real" about the system. */
104 struct prison prison0 = {
105 .pr_id = 0,
106 .pr_name = "0",
107 .pr_ref = 1,
108 .pr_uref = 1,
109 .pr_path = "/",
110 .pr_securelevel = -1,
111 .pr_devfs_rsnum = 0,
112 .pr_state = PRISON_STATE_ALIVE,
113 .pr_childmax = JAIL_MAX,
114 .pr_hostuuid = DEFAULT_HOSTUUID,
115 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
116 #ifdef VIMAGE
117 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
118 #else
119 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
120 #endif
121 .pr_allow = PR_ALLOW_PRISON0,
122 };
123 _Static_assert((PR_ALLOW_PRISON0 & ~PR_ALLOW_ALL_STATIC) == 0,
124 "Bits enabled in PR_ALLOW_PRISON0 that are not statically reserved");
125
126 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
127
128 struct bool_flags {
129 const char *name;
130 const char *noname;
131 volatile u_int flag;
132 };
133 struct jailsys_flags {
134 const char *name;
135 unsigned disable;
136 unsigned new;
137 };
138
139 /*
140 * Handle jail teardown in a dedicated thread to avoid deadlocks from
141 * vnet_destroy().
142 */
143 TASKQUEUE_DEFINE_THREAD(jail_remove);
144
145 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
146 struct sx allprison_lock;
147 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
148 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
149 LIST_HEAD(, prison_racct) allprison_racct;
150 int lastprid = 0;
151 int lastdeadid = 0;
152
153 static int get_next_prid(struct prison **insprp);
154 static int get_next_deadid(struct prison **insprp);
155 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
156 static void prison_complete(void *context, int pending);
157 static void prison_deref(struct prison *pr, int flags);
158 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
159 static int prison_lock_xlock(struct prison *pr, int flags);
160 static void prison_cleanup_locked(struct prison *pr);
161 static void prison_cleanup_unlocked(struct prison *pr);
162 static void prison_free_not_last(struct prison *pr);
163 static void prison_proc_free_not_last(struct prison *pr);
164 static void prison_proc_relink(struct prison *opr, struct prison *npr,
165 struct proc *p);
166 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
167 int enable);
168 static char *prison_path(struct prison *pr1, struct prison *pr2);
169 #ifdef RACCT
170 static void prison_racct_attach(struct prison *pr);
171 static void prison_racct_modify(struct prison *pr);
172 static void prison_racct_detach(struct prison *pr);
173 #endif
174 static void prison_knote(struct prison *pr, long hint);
175
176 /* Flags for prison_deref */
177 #define PD_DEREF 0x01 /* Decrement pr_ref */
178 #define PD_DEUREF 0x02 /* Decrement pr_uref */
179 #define PD_KILL 0x04 /* Remove jail, kill processes, etc */
180 #define PD_LOCKED 0x10 /* pr_mtx is held */
181 #define PD_LIST_SLOCKED 0x20 /* allprison_lock is held shared */
182 #define PD_LIST_XLOCKED 0x40 /* allprison_lock is held exclusive */
183 #define PD_OP_FLAGS 0x07 /* Operation flags */
184 #define PD_LOCK_FLAGS 0x70 /* Lock status flags */
185
186 /*
187 * Parameter names corresponding to PR_* flag values. Size values are for kvm
188 * as we cannot figure out the size of a sparse array, or an array without a
189 * terminating entry.
190 */
191 static struct bool_flags pr_flag_bool[] = {
192 {"persist", "nopersist", PR_PERSIST},
193 #ifdef INET
194 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
195 #endif
196 #ifdef INET6
197 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
198 #endif
199 };
200 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
201
202 static struct jailsys_flags pr_flag_jailsys[] = {
203 {"host", 0, PR_HOST},
204 #ifdef VIMAGE
205 {"vnet", 0, PR_VNET},
206 #endif
207 #ifdef INET
208 {"ip4", PR_IP4_USER, PR_IP4_USER},
209 #endif
210 #ifdef INET6
211 {"ip6", PR_IP6_USER, PR_IP6_USER},
212 #endif
213 };
214 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
215
216 /*
217 * Make this array full-size so dynamic parameters can be added.
218 * It is protected by prison0.mtx, but lockless reading is allowed
219 * with an atomic check of the flag values.
220 */
221 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
222 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
223 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
224 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
225 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
226 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
227 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
228 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
229 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
230 {"allow.reserved_ports", "allow.noreserved_ports",
231 PR_ALLOW_RESERVED_PORTS},
232 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
233 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
234 PR_ALLOW_UNPRIV_DEBUG},
235 {"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
236 #ifdef VIMAGE
237 {"allow.nfsd", "allow.nonfsd", PR_ALLOW_NFSD},
238 #endif
239 {"allow.extattr", "allow.noextattr", PR_ALLOW_EXTATTR},
240 {"allow.adjtime", "allow.noadjtime", PR_ALLOW_ADJTIME},
241 {"allow.settime", "allow.nosettime", PR_ALLOW_SETTIME},
242 {"allow.routing", "allow.norouting", PR_ALLOW_ROUTING},
243 {"allow.unprivileged_parent_tampering",
244 "allow.nounprivileged_parent_tampering",
245 PR_ALLOW_UNPRIV_PARENT_TAMPER},
246 #ifdef AUDIT
247 {"allow.setaudit", "allow.nosetaudit", PR_ALLOW_SETAUDIT},
248 #endif
249 };
250 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
251 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
252
253 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
254 PR_ALLOW_RESERVED_PORTS | \
255 PR_ALLOW_UNPRIV_DEBUG | \
256 PR_ALLOW_SUSER)
257 #define JAIL_DEFAULT_ENFORCE_STATFS 2
258 #define JAIL_DEFAULT_DEVFS_RSNUM 0
259 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
260 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
261 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
262 #if defined(INET) || defined(INET6)
263 static unsigned jail_max_af_ips = 255;
264 #endif
265
266 /*
267 * Initialize the parts of prison0 that can't be static-initialized with
268 * constants. This is called from proc0_init() after creating thread0 cpuset.
269 */
270 void
prison0_init(void)271 prison0_init(void)
272 {
273 uint8_t *file, *data;
274 size_t size;
275 char buf[sizeof(prison0.pr_hostuuid)];
276 bool valid;
277
278 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
279 prison0.pr_osreldate = osreldate;
280 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
281
282 /* If we have a preloaded hostuuid, use it. */
283 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
284 if (file != NULL) {
285 data = preload_fetch_addr(file);
286 size = preload_fetch_size(file);
287 if (data != NULL) {
288 /*
289 * The preloaded data may include trailing whitespace, almost
290 * certainly a newline; skip over any whitespace or
291 * non-printable characters to be safe.
292 */
293 while (size > 0 && data[size - 1] <= 0x20) {
294 size--;
295 }
296
297 valid = false;
298
299 /*
300 * Not NUL-terminated when passed from loader, but
301 * validate_uuid requires that due to using sscanf (as
302 * does the subsequent strlcpy, since it still reads
303 * past the given size to return the true length);
304 * bounce to a temporary buffer to fix.
305 */
306 if (size >= sizeof(buf))
307 goto done;
308
309 memcpy(buf, data, size);
310 buf[size] = '\0';
311
312 if (validate_uuid(buf, size, NULL, 0) != 0)
313 goto done;
314
315 valid = true;
316 (void)strlcpy(prison0.pr_hostuuid, buf,
317 sizeof(prison0.pr_hostuuid));
318
319 done:
320 if (bootverbose && !valid) {
321 printf("hostuuid: preload data malformed: '%.*s'\n",
322 (int)size, data);
323 }
324 }
325 }
326 if (bootverbose)
327 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
328 }
329
330 /*
331 * struct jail_args {
332 * struct jail *jail;
333 * };
334 */
335 int
sys_jail(struct thread * td,struct jail_args * uap)336 sys_jail(struct thread *td, struct jail_args *uap)
337 {
338 uint32_t version;
339 int error;
340 struct jail j;
341
342 error = copyin(uap->jail, &version, sizeof(uint32_t));
343 if (error)
344 return (error);
345
346 switch (version) {
347 case 0:
348 {
349 struct jail_v0 j0;
350
351 /* FreeBSD single IPv4 jails. */
352 bzero(&j, sizeof(struct jail));
353 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
354 if (error)
355 return (error);
356 j.version = j0.version;
357 j.path = j0.path;
358 j.hostname = j0.hostname;
359 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
360 break;
361 }
362
363 case 1:
364 /*
365 * Version 1 was used by multi-IPv4 jail implementations
366 * that never made it into the official kernel.
367 */
368 return (EINVAL);
369
370 case 2: /* JAIL_API_VERSION */
371 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
372 error = copyin(uap->jail, &j, sizeof(struct jail));
373 if (error)
374 return (error);
375 break;
376
377 default:
378 /* Sci-Fi jails are not supported, sorry. */
379 return (EINVAL);
380 }
381 return (kern_jail(td, &j));
382 }
383
384 int
kern_jail(struct thread * td,struct jail * j)385 kern_jail(struct thread *td, struct jail *j)
386 {
387 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
388 #ifdef INET
389 + 1
390 #endif
391 #ifdef INET6
392 + 1
393 #endif
394 )];
395 struct uio opt;
396 char *u_path, *u_hostname, *u_name;
397 struct bool_flags *bf;
398 #ifdef INET
399 uint32_t ip4s;
400 struct in_addr *u_ip4;
401 #endif
402 #ifdef INET6
403 struct in6_addr *u_ip6;
404 #endif
405 size_t tmplen;
406 int error, enforce_statfs;
407
408 bzero(&optiov, sizeof(optiov));
409 opt.uio_iov = optiov;
410 opt.uio_iovcnt = 0;
411 opt.uio_offset = -1;
412 opt.uio_resid = -1;
413 opt.uio_segflg = UIO_SYSSPACE;
414 opt.uio_rw = UIO_READ;
415 opt.uio_td = td;
416
417 /* Set permissions for top-level jails from sysctls. */
418 if (!jailed(td->td_ucred)) {
419 for (bf = pr_flag_allow;
420 bf < pr_flag_allow + nitems(pr_flag_allow) &&
421 atomic_load_int(&bf->flag) != 0;
422 bf++) {
423 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
424 (jail_default_allow & bf->flag)
425 ? bf->name : bf->noname);
426 optiov[opt.uio_iovcnt].iov_len =
427 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
428 opt.uio_iovcnt += 2;
429 }
430 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
431 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
432 opt.uio_iovcnt++;
433 enforce_statfs = jail_default_enforce_statfs;
434 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
435 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
436 opt.uio_iovcnt++;
437 }
438
439 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
440 #ifdef INET
441 ip4s = (j->version == 0) ? 1 : j->ip4s;
442 if (ip4s > jail_max_af_ips)
443 return (EINVAL);
444 tmplen += ip4s * sizeof(struct in_addr);
445 #else
446 if (j->ip4s > 0)
447 return (EINVAL);
448 #endif
449 #ifdef INET6
450 if (j->ip6s > jail_max_af_ips)
451 return (EINVAL);
452 tmplen += j->ip6s * sizeof(struct in6_addr);
453 #else
454 if (j->ip6s > 0)
455 return (EINVAL);
456 #endif
457 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
458 u_hostname = u_path + MAXPATHLEN;
459 u_name = u_hostname + MAXHOSTNAMELEN;
460 #ifdef INET
461 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
462 #endif
463 #ifdef INET6
464 #ifdef INET
465 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
466 #else
467 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
468 #endif
469 #endif
470 optiov[opt.uio_iovcnt].iov_base = "path";
471 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
472 opt.uio_iovcnt++;
473 optiov[opt.uio_iovcnt].iov_base = u_path;
474 error = copyinstr(j->path, u_path, MAXPATHLEN,
475 &optiov[opt.uio_iovcnt].iov_len);
476 if (error) {
477 free(u_path, M_TEMP);
478 return (error);
479 }
480 opt.uio_iovcnt++;
481 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
482 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
483 opt.uio_iovcnt++;
484 optiov[opt.uio_iovcnt].iov_base = u_hostname;
485 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
486 &optiov[opt.uio_iovcnt].iov_len);
487 if (error) {
488 free(u_path, M_TEMP);
489 return (error);
490 }
491 opt.uio_iovcnt++;
492 if (j->jailname != NULL) {
493 optiov[opt.uio_iovcnt].iov_base = "name";
494 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
495 opt.uio_iovcnt++;
496 optiov[opt.uio_iovcnt].iov_base = u_name;
497 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
498 &optiov[opt.uio_iovcnt].iov_len);
499 if (error) {
500 free(u_path, M_TEMP);
501 return (error);
502 }
503 opt.uio_iovcnt++;
504 }
505 #ifdef INET
506 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
507 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
508 opt.uio_iovcnt++;
509 optiov[opt.uio_iovcnt].iov_base = u_ip4;
510 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
511 if (j->version == 0)
512 u_ip4->s_addr = j->ip4s;
513 else {
514 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
515 if (error) {
516 free(u_path, M_TEMP);
517 return (error);
518 }
519 }
520 opt.uio_iovcnt++;
521 #endif
522 #ifdef INET6
523 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
524 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
525 opt.uio_iovcnt++;
526 optiov[opt.uio_iovcnt].iov_base = u_ip6;
527 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
528 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
529 if (error) {
530 free(u_path, M_TEMP);
531 return (error);
532 }
533 opt.uio_iovcnt++;
534 #endif
535 KASSERT(opt.uio_iovcnt <= nitems(optiov),
536 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
537 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
538 free(u_path, M_TEMP);
539 return (error);
540 }
541
542 /*
543 * struct jail_set_args {
544 * struct iovec *iovp;
545 * unsigned int iovcnt;
546 * int flags;
547 * };
548 */
549 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)550 sys_jail_set(struct thread *td, struct jail_set_args *uap)
551 {
552 struct uio *auio;
553 int error;
554
555 /* Check that we have an even number of iovecs. */
556 if (uap->iovcnt & 1)
557 return (EINVAL);
558
559 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
560 if (error)
561 return (error);
562 error = kern_jail_set(td, auio, uap->flags);
563 freeuio(auio);
564 return (error);
565 }
566
567 #if defined(INET) || defined(INET6)
568 typedef int prison_addr_cmp_t(const void *, const void *);
569 typedef bool prison_addr_valid_t(const void *);
570 static const struct pr_family {
571 size_t size;
572 prison_addr_cmp_t *cmp;
573 prison_addr_valid_t *valid;
574 int ip_flag;
575 } pr_families[PR_FAMILY_MAX] = {
576 #ifdef INET
577 [PR_INET] = {
578 .size = sizeof(struct in_addr),
579 .cmp = prison_qcmp_v4,
580 .valid = prison_valid_v4,
581 .ip_flag = PR_IP4_USER,
582 },
583 #endif
584 #ifdef INET6
585 [PR_INET6] = {
586 .size = sizeof(struct in6_addr),
587 .cmp = prison_qcmp_v6,
588 .valid = prison_valid_v6,
589 .ip_flag = PR_IP6_USER,
590 },
591 #endif
592 };
593
594 /*
595 * Network address lists (pr_addrs) allocation for jails. The addresses
596 * are accessed locklessly by the network stack, thus need to be protected by
597 * the network epoch.
598 */
599 struct prison_ip {
600 struct epoch_context ctx;
601 uint32_t ips;
602 #ifdef FUTURE_C
603 /*
604 * XXX Variable-length automatic arrays in union may be
605 * supported in future C.
606 */
607 union {
608 char pr_ip[];
609 struct in_addr pr_ip4[];
610 struct in6_addr pr_ip6[];
611 };
612 #else /* No future C :( */
613 char pr_ip[];
614 #endif
615 };
616
617 static char *
PR_IP(struct prison_ip * pip,const pr_family_t af,int idx)618 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
619 {
620 MPASS(pip);
621 MPASS(af < PR_FAMILY_MAX);
622 MPASS(idx >= 0 && idx < pip->ips);
623
624 return (pip->pr_ip + pr_families[af].size * idx);
625 }
626
627 static struct prison_ip *
prison_ip_alloc(const pr_family_t af,uint32_t cnt,int flags)628 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
629 {
630 struct prison_ip *pip;
631
632 pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
633 M_PRISON, flags);
634 if (pip != NULL)
635 pip->ips = cnt;
636 return (pip);
637 }
638
639 /*
640 * Allocate and copyin user supplied address list, sorting and validating.
641 * kern_jail_set() helper.
642 */
643 static struct prison_ip *
prison_ip_copyin(const pr_family_t af,void * op,uint32_t cnt)644 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
645 {
646 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
647 const size_t size = pr_families[af].size;
648 struct prison_ip *pip;
649
650 pip = prison_ip_alloc(af, cnt, M_WAITOK);
651 bcopy(op, pip->pr_ip, cnt * size);
652 /*
653 * IP addresses are all sorted but ip[0] to preserve
654 * the primary IP address as given from userland.
655 * This special IP is used for unbound outgoing
656 * connections as well for "loopback" traffic in case
657 * source address selection cannot find any more fitting
658 * address to connect from.
659 */
660 if (cnt > 1)
661 qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
662 /*
663 * Check for duplicate addresses and do some simple
664 * zero and broadcast checks. If users give other bogus
665 * addresses it is their problem.
666 */
667 for (int i = 0; i < cnt; i++) {
668 if (!pr_families[af].valid(PR_IP(pip, af, i))) {
669 free(pip, M_PRISON);
670 return (NULL);
671 }
672 if (i + 1 < cnt &&
673 (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
674 cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
675 free(pip, M_PRISON);
676 return (NULL);
677 }
678 }
679
680 return (pip);
681 }
682
683 /*
684 * Allocate and dup parent prison address list.
685 * kern_jail_set() helper.
686 */
687 static void
prison_ip_dup(struct prison * ppr,struct prison * pr,const pr_family_t af)688 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
689 {
690 const struct prison_ip *ppip = ppr->pr_addrs[af];
691 struct prison_ip *pip;
692
693 if (ppip != NULL) {
694 pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
695 bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
696 pr->pr_addrs[af] = pip;
697 }
698 }
699
700 /*
701 * Make sure the new set of IP addresses is a subset of the parent's list.
702 * Don't worry about the parent being unlocked, as any setting is done with
703 * allprison_lock held.
704 * kern_jail_set() helper.
705 */
706 static bool
prison_ip_parent_match(struct prison_ip * ppip,struct prison_ip * pip,const pr_family_t af)707 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
708 const pr_family_t af)
709 {
710 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
711 int i, j;
712
713 if (ppip == NULL)
714 return (false);
715
716 for (i = 0; i < ppip->ips; i++)
717 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
718 break;
719
720 if (i == ppip->ips)
721 /* Main address not present in parent. */
722 return (false);
723
724 if (pip->ips > 1) {
725 for (i = j = 1; i < pip->ips; i++) {
726 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
727 /* Equals to parent primary address. */
728 continue;
729 for (; j < ppip->ips; j++)
730 if (cmp(PR_IP(pip, af, i),
731 PR_IP(ppip, af, j)) == 0)
732 break;
733 if (j == ppip->ips)
734 break;
735 }
736 if (j == ppip->ips)
737 /* Address not present in parent. */
738 return (false);
739 }
740 return (true);
741 }
742
743 /*
744 * Check for conflicting IP addresses. We permit them if there is no more
745 * than one IP on each jail. If there is a duplicate on a jail with more
746 * than one IP stop checking and return error.
747 * kern_jail_set() helper.
748 */
749 static bool
prison_ip_conflict_check(const struct prison * ppr,const struct prison * pr,struct prison_ip * pip,pr_family_t af)750 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
751 struct prison_ip *pip, pr_family_t af)
752 {
753 const struct prison *tppr, *tpr;
754 int descend;
755
756 #ifdef VIMAGE
757 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
758 if (tppr->pr_flags & PR_VNET)
759 break;
760 #else
761 tppr = &prison0;
762 #endif
763 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
764 if (tpr == pr ||
765 #ifdef VIMAGE
766 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
767 #endif
768 !prison_isalive(tpr)) {
769 descend = 0;
770 continue;
771 }
772 if (!(tpr->pr_flags & pr_families[af].ip_flag))
773 continue;
774 descend = 0;
775 if (tpr->pr_addrs[af] == NULL ||
776 (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
777 continue;
778 for (int i = 0; i < pip->ips; i++)
779 if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
780 return (false);
781 }
782
783 return (true);
784 }
785
786 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
787 "prison must start with epoch context");
788 static void
prison_ip_free_deferred(epoch_context_t ctx)789 prison_ip_free_deferred(epoch_context_t ctx)
790 {
791
792 free(ctx, M_PRISON);
793 }
794
795 static void
prison_ip_free(struct prison_ip * pip)796 prison_ip_free(struct prison_ip *pip)
797 {
798
799 if (pip != NULL)
800 NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
801 }
802
803 static void
prison_ip_set(struct prison * pr,const pr_family_t af,struct prison_ip * new)804 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
805 {
806 struct prison_ip **mem, *old;
807
808 mtx_assert(&pr->pr_mtx, MA_OWNED);
809
810 mem = &pr->pr_addrs[af];
811
812 old = *mem;
813 atomic_store_ptr(mem, new);
814 prison_ip_free(old);
815 }
816
817 /*
818 * Restrict a prison's IP address list with its parent's, possibly replacing
819 * it. Return true if succeed, otherwise should redo.
820 * kern_jail_set() helper.
821 */
822 static bool
prison_ip_restrict(struct prison * pr,const pr_family_t af,struct prison_ip ** newp)823 prison_ip_restrict(struct prison *pr, const pr_family_t af,
824 struct prison_ip **newp)
825 {
826 struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
827 struct prison_ip *pip = pr->pr_addrs[af];
828 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
829 const size_t size = pr_families[af].size;
830 struct prison_ip *new = newp != NULL ? *newp : NULL;
831 uint32_t ips;
832
833 mtx_assert(&pr->pr_mtx, MA_OWNED);
834
835 /*
836 * Due to epoch-synchronized access to the IP address lists we always
837 * allocate a new list even if the old one has enough space. We could
838 * atomically update an IPv4 address inside a list, but that would
839 * screw up sorting, and in case of IPv6 we can't even atomically write
840 * one.
841 */
842 if (ppip == NULL) {
843 if (pip != NULL)
844 prison_ip_set(pr, af, NULL);
845 return (true);
846 }
847
848 if (!(pr->pr_flags & pr_families[af].ip_flag)) {
849 if (new == NULL) {
850 new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
851 if (new == NULL)
852 return (false); /* Redo */
853 }
854 /* This has no user settings, so just copy the parent's list. */
855 MPASS(new->ips == ppip->ips);
856 bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
857 prison_ip_set(pr, af, new);
858 if (newp != NULL)
859 *newp = NULL; /* Used */
860 } else if (pip != NULL) {
861 /* Remove addresses that aren't in the parent. */
862 int i;
863
864 i = 0; /* index in pip */
865 ips = 0; /* index in new */
866
867 if (new == NULL) {
868 new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
869 if (new == NULL)
870 return (false); /* Redo */
871 }
872
873 for (int pi = 0; pi < ppip->ips; pi++)
874 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
875 /* Found our primary address in parent. */
876 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
877 size);
878 i++;
879 ips++;
880 break;
881 }
882 for (int pi = 1; i < pip->ips; ) {
883 /* Check against primary, which is unsorted. */
884 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
885 /* Matches parent's primary address. */
886 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
887 size);
888 i++;
889 ips++;
890 continue;
891 }
892 /* The rest are sorted. */
893 switch (pi >= ppip->ips ? -1 :
894 cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
895 case -1:
896 i++;
897 break;
898 case 0:
899 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
900 size);
901 i++;
902 pi++;
903 ips++;
904 break;
905 case 1:
906 pi++;
907 break;
908 }
909 }
910 if (ips == 0) {
911 if (newp == NULL || *newp == NULL)
912 prison_ip_free(new);
913 new = NULL;
914 } else {
915 /* Shrink to real size */
916 KASSERT((new->ips >= ips),
917 ("Out-of-bounds write to prison_ip %p", new));
918 new->ips = ips;
919 }
920 prison_ip_set(pr, af, new);
921 if (newp != NULL)
922 *newp = NULL; /* Used */
923 }
924 return (true);
925 }
926
927 /*
928 * Fast-path check if an address belongs to a prison.
929 */
930 int
prison_ip_check(const struct prison * pr,const pr_family_t af,const void * addr)931 prison_ip_check(const struct prison *pr, const pr_family_t af,
932 const void *addr)
933 {
934 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
935 struct prison_ip *pip;
936 int i, a, z, d;
937
938 MPASS(mtx_owned(&pr->pr_mtx) ||
939 in_epoch(net_epoch_preempt) ||
940 sx_xlocked(&allprison_lock));
941
942 pip = atomic_load_ptr(&pr->pr_addrs[af]);
943 if (__predict_false(pip == NULL))
944 return (EAFNOSUPPORT);
945
946 /* Check the primary IP. */
947 if (cmp(PR_IP(pip, af, 0), addr) == 0)
948 return (0);
949
950 /*
951 * All the other IPs are sorted so we can do a binary search.
952 */
953 a = 0;
954 z = pip->ips - 2;
955 while (a <= z) {
956 i = (a + z) / 2;
957 d = cmp(PR_IP(pip, af, i + 1), addr);
958 if (d > 0)
959 z = i - 1;
960 else if (d < 0)
961 a = i + 1;
962 else
963 return (0);
964 }
965
966 return (EADDRNOTAVAIL);
967 }
968
969 /*
970 * Grab primary IP. Historically required mutex, but nothing prevents
971 * us to support epoch-protected access. Is it used in fast path?
972 * in{6}_jail.c helper
973 */
974 const void *
prison_ip_get0(const struct prison * pr,const pr_family_t af)975 prison_ip_get0(const struct prison *pr, const pr_family_t af)
976 {
977 const struct prison_ip *pip = pr->pr_addrs[af];
978
979 mtx_assert(&pr->pr_mtx, MA_OWNED);
980 MPASS(pip);
981
982 return (pip->pr_ip);
983 }
984
985 u_int
prison_ip_cnt(const struct prison * pr,const pr_family_t af)986 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
987 {
988
989 return (pr->pr_addrs[af]->ips);
990 }
991 #endif /* defined(INET) || defined(INET6) */
992
993 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)994 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
995 {
996 struct file *jfp_out;
997 struct nameidata nd;
998 #ifdef INET
999 struct prison_ip *ip4;
1000 #endif
1001 #ifdef INET6
1002 struct prison_ip *ip6;
1003 #endif
1004 struct vfsopt *opt;
1005 struct vfsoptlist *opts;
1006 struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
1007 struct ucred *jdcred;
1008 struct vnode *root;
1009 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
1010 char *g_path, *osrelstr;
1011 struct bool_flags *bf;
1012 struct jailsys_flags *jsf;
1013 #if defined(INET) || defined(INET6)
1014 void *op;
1015 #endif
1016 unsigned long hid;
1017 size_t namelen, onamelen, pnamelen;
1018 int created, cuflags, descend, drflags, enforce;
1019 int error, errmsg_len, errmsg_pos;
1020 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1021 int deadid, jfd_in, jfd_out, jfd_pos, jid, jsys, len, level;
1022 int childmax, osreldt, rsnum, slevel;
1023 #ifdef INET
1024 int ip4s;
1025 bool redo_ip4;
1026 #endif
1027 #ifdef INET6
1028 int ip6s;
1029 bool redo_ip6;
1030 #endif
1031 bool maybe_changed;
1032 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1033 uint64_t pr_allow_diff;
1034 unsigned tallow;
1035 char numbuf[12];
1036
1037 mypr = td->td_ucred->cr_prison;
1038 if (((flags & (JAIL_CREATE | JAIL_AT_DESC)) == JAIL_CREATE) &&
1039 mypr->pr_childmax == 0)
1040 return (EPERM);
1041 if (flags & ~JAIL_SET_MASK)
1042 return (EINVAL);
1043 if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
1044 (JAIL_USE_DESC | JAIL_AT_DESC))
1045 return (EINVAL);
1046 prison_hold(mypr);
1047
1048 #ifdef INET
1049 ip4 = NULL;
1050 #endif
1051 #ifdef INET6
1052 ip6 = NULL;
1053 #endif
1054 g_path = NULL;
1055 jfp_out = NULL;
1056 jfd_out = -1;
1057 /*
1058 * Check all the parameters before committing to anything. Not all
1059 * errors can be caught early, but we may as well try. Also, this
1060 * takes care of some expensive stuff (path lookup) before getting
1061 * the allprison lock.
1062 *
1063 * XXX Jails are not filesystems, and jail parameters are not mount
1064 * options. But it makes more sense to re-use the vfsopt code
1065 * than duplicate it under a different name.
1066 */
1067 error = vfs_buildopts(optuio, &opts);
1068 if (error) {
1069 opts = NULL;
1070 goto done_free;
1071 }
1072
1073 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1074 if (!cuflags) {
1075 error = EINVAL;
1076 vfs_opterror(opts, "no valid operation (create or update)");
1077 goto done_errmsg;
1078 }
1079
1080 error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
1081 if (error == ENOENT) {
1082 if (flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1083 JAIL_OWN_DESC)) {
1084 vfs_opterror(opts, "missing desc");
1085 goto done_errmsg;
1086 }
1087 jfd_in = -1;
1088 } else if (error != 0)
1089 goto done_free;
1090 else {
1091 if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1092 JAIL_OWN_DESC))) {
1093 error = EINVAL;
1094 vfs_opterror(opts, "unexpected desc");
1095 goto done_errmsg;
1096 }
1097 if (flags & JAIL_AT_DESC) {
1098 /*
1099 * Look up and create jails based on the
1100 * descriptor's prison.
1101 */
1102 prison_free(mypr);
1103 error = jaildesc_find(td, jfd_in, &mypr, NULL);
1104 if (error != 0) {
1105 vfs_opterror(opts, error == ENOENT ?
1106 "descriptor to dead jail" :
1107 "not a jail descriptor");
1108 goto done_errmsg;
1109 }
1110 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) {
1111 error = EPERM;
1112 goto done_free;
1113 }
1114 }
1115 if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
1116 /* Allocate a jail descriptor to return later. */
1117 error = jaildesc_alloc(td, &jfp_out, &jfd_out,
1118 flags & JAIL_OWN_DESC);
1119 if (error)
1120 goto done_free;
1121 }
1122 }
1123
1124 /*
1125 * Delay the permission check if using a jail descriptor,
1126 * until we get the descriptor's credentials.
1127 */
1128 if (!(flags & JAIL_USE_DESC)) {
1129 error = priv_check(td, PRIV_JAIL_SET);
1130 if (error == 0 && (flags & JAIL_ATTACH))
1131 error = priv_check(td, PRIV_JAIL_ATTACH);
1132 if (error)
1133 goto done_free;
1134 }
1135
1136 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1137 if (error == ENOENT)
1138 jid = 0;
1139 else if (error != 0)
1140 goto done_free;
1141
1142 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
1143 if (error == ENOENT)
1144 gotslevel = 0;
1145 else if (error != 0)
1146 goto done_free;
1147 else
1148 gotslevel = 1;
1149
1150 error =
1151 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
1152 if (error == ENOENT)
1153 gotchildmax = 0;
1154 else if (error != 0)
1155 goto done_free;
1156 else
1157 gotchildmax = 1;
1158
1159 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
1160 if (error == ENOENT)
1161 gotenforce = 0;
1162 else if (error != 0)
1163 goto done_free;
1164 else if (enforce < 0 || enforce > 2) {
1165 error = EINVAL;
1166 goto done_free;
1167 } else
1168 gotenforce = 1;
1169
1170 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
1171 if (error == ENOENT)
1172 gotrsnum = 0;
1173 else if (error != 0)
1174 goto done_free;
1175 else
1176 gotrsnum = 1;
1177
1178 pr_flags = ch_flags = 0;
1179 for (bf = pr_flag_bool;
1180 bf < pr_flag_bool + nitems(pr_flag_bool);
1181 bf++) {
1182 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1183 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1184 }
1185 ch_flags |= pr_flags;
1186 for (jsf = pr_flag_jailsys;
1187 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1188 jsf++) {
1189 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1190 if (error == ENOENT)
1191 continue;
1192 if (error != 0)
1193 goto done_free;
1194 switch (jsys) {
1195 case JAIL_SYS_DISABLE:
1196 if (!jsf->disable) {
1197 error = EINVAL;
1198 goto done_free;
1199 }
1200 pr_flags |= jsf->disable;
1201 break;
1202 case JAIL_SYS_NEW:
1203 pr_flags |= jsf->new;
1204 break;
1205 case JAIL_SYS_INHERIT:
1206 break;
1207 default:
1208 error = EINVAL;
1209 goto done_free;
1210 }
1211 ch_flags |= jsf->new | jsf->disable;
1212 }
1213 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1214 && !(pr_flags & PR_PERSIST)) {
1215 error = EINVAL;
1216 vfs_opterror(opts, "new jail must persist or attach");
1217 goto done_errmsg;
1218 }
1219 #ifdef VIMAGE
1220 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1221 error = EINVAL;
1222 vfs_opterror(opts, "vnet cannot be changed after creation");
1223 goto done_errmsg;
1224 }
1225 #endif
1226 #ifdef INET
1227 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1228 error = EINVAL;
1229 vfs_opterror(opts, "ip4 cannot be changed after creation");
1230 goto done_errmsg;
1231 }
1232 #endif
1233 #ifdef INET6
1234 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1235 error = EINVAL;
1236 vfs_opterror(opts, "ip6 cannot be changed after creation");
1237 goto done_errmsg;
1238 }
1239 #endif
1240
1241 pr_allow = ch_allow = 0;
1242 for (bf = pr_flag_allow;
1243 bf < pr_flag_allow + nitems(pr_flag_allow) &&
1244 atomic_load_int(&bf->flag) != 0;
1245 bf++) {
1246 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1247 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1248 }
1249 ch_allow |= pr_allow;
1250
1251 error = vfs_getopt(opts, "name", (void **)&name, &len);
1252 if (error == ENOENT)
1253 name = NULL;
1254 else if (error != 0)
1255 goto done_free;
1256 else {
1257 if (len == 0 || name[len - 1] != '\0') {
1258 error = EINVAL;
1259 goto done_free;
1260 }
1261 if (len > MAXHOSTNAMELEN) {
1262 error = ENAMETOOLONG;
1263 goto done_free;
1264 }
1265 }
1266
1267 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
1268 if (error == ENOENT)
1269 host = NULL;
1270 else if (error != 0)
1271 goto done_free;
1272 else {
1273 ch_flags |= PR_HOST;
1274 pr_flags |= PR_HOST;
1275 if (len == 0 || host[len - 1] != '\0') {
1276 error = EINVAL;
1277 goto done_free;
1278 }
1279 if (len > MAXHOSTNAMELEN) {
1280 error = ENAMETOOLONG;
1281 goto done_free;
1282 }
1283 }
1284
1285 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
1286 if (error == ENOENT)
1287 domain = NULL;
1288 else if (error != 0)
1289 goto done_free;
1290 else {
1291 ch_flags |= PR_HOST;
1292 pr_flags |= PR_HOST;
1293 if (len == 0 || domain[len - 1] != '\0') {
1294 error = EINVAL;
1295 goto done_free;
1296 }
1297 if (len > MAXHOSTNAMELEN) {
1298 error = ENAMETOOLONG;
1299 goto done_free;
1300 }
1301 }
1302
1303 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
1304 if (error == ENOENT)
1305 uuid = NULL;
1306 else if (error != 0)
1307 goto done_free;
1308 else {
1309 ch_flags |= PR_HOST;
1310 pr_flags |= PR_HOST;
1311 if (len == 0 || uuid[len - 1] != '\0') {
1312 error = EINVAL;
1313 goto done_free;
1314 }
1315 if (len > HOSTUUIDLEN) {
1316 error = ENAMETOOLONG;
1317 goto done_free;
1318 }
1319 }
1320
1321 #ifdef COMPAT_FREEBSD32
1322 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1323 uint32_t hid32;
1324
1325 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
1326 hid = hid32;
1327 } else
1328 #endif
1329 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
1330 if (error == ENOENT)
1331 gothid = 0;
1332 else if (error != 0)
1333 goto done_free;
1334 else {
1335 gothid = 1;
1336 ch_flags |= PR_HOST;
1337 pr_flags |= PR_HOST;
1338 }
1339
1340 #ifdef INET
1341 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
1342 if (error == ENOENT)
1343 ip4s = 0;
1344 else if (error != 0)
1345 goto done_free;
1346 else if (ip4s & (sizeof(struct in_addr) - 1)) {
1347 error = EINVAL;
1348 goto done_free;
1349 } else {
1350 ch_flags |= PR_IP4_USER;
1351 pr_flags |= PR_IP4_USER;
1352 if (ip4s > 0) {
1353 ip4s /= sizeof(struct in_addr);
1354 if (ip4s > jail_max_af_ips) {
1355 error = EINVAL;
1356 vfs_opterror(opts, "too many IPv4 addresses");
1357 goto done_errmsg;
1358 }
1359 ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1360 if (ip4 == NULL) {
1361 error = EINVAL;
1362 goto done_free;
1363 }
1364 }
1365 }
1366 #endif
1367
1368 #ifdef INET6
1369 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
1370 if (error == ENOENT)
1371 ip6s = 0;
1372 else if (error != 0)
1373 goto done_free;
1374 else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1375 error = EINVAL;
1376 goto done_free;
1377 } else {
1378 ch_flags |= PR_IP6_USER;
1379 pr_flags |= PR_IP6_USER;
1380 if (ip6s > 0) {
1381 ip6s /= sizeof(struct in6_addr);
1382 if (ip6s > jail_max_af_ips) {
1383 error = EINVAL;
1384 vfs_opterror(opts, "too many IPv6 addresses");
1385 goto done_errmsg;
1386 }
1387 ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1388 if (ip6 == NULL) {
1389 error = EINVAL;
1390 goto done_free;
1391 }
1392 }
1393 }
1394 #endif
1395
1396 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1397 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1398 error = EINVAL;
1399 vfs_opterror(opts,
1400 "vnet jails cannot have IP address restrictions");
1401 goto done_errmsg;
1402 }
1403 #endif
1404
1405 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
1406 if (error == ENOENT)
1407 osrelstr = NULL;
1408 else if (error != 0)
1409 goto done_free;
1410 else {
1411 if (flags & JAIL_UPDATE) {
1412 error = EINVAL;
1413 vfs_opterror(opts,
1414 "osrelease cannot be changed after creation");
1415 goto done_errmsg;
1416 }
1417 if (len == 0 || osrelstr[len - 1] != '\0') {
1418 error = EINVAL;
1419 goto done_free;
1420 }
1421 if (len >= OSRELEASELEN) {
1422 error = ENAMETOOLONG;
1423 vfs_opterror(opts,
1424 "osrelease string must be 1-%d bytes long",
1425 OSRELEASELEN - 1);
1426 goto done_errmsg;
1427 }
1428 }
1429
1430 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
1431 if (error == ENOENT)
1432 osreldt = 0;
1433 else if (error != 0)
1434 goto done_free;
1435 else {
1436 if (flags & JAIL_UPDATE) {
1437 error = EINVAL;
1438 vfs_opterror(opts,
1439 "osreldate cannot be changed after creation");
1440 goto done_errmsg;
1441 }
1442 if (osreldt == 0) {
1443 error = EINVAL;
1444 vfs_opterror(opts, "osreldate cannot be 0");
1445 goto done_errmsg;
1446 }
1447 }
1448
1449 root = NULL;
1450 error = vfs_getopt(opts, "path", (void **)&path, &len);
1451 if (error == ENOENT)
1452 path = NULL;
1453 else if (error != 0)
1454 goto done_free;
1455 else {
1456 if (flags & JAIL_UPDATE) {
1457 error = EINVAL;
1458 vfs_opterror(opts,
1459 "path cannot be changed after creation");
1460 goto done_errmsg;
1461 }
1462 if (len == 0 || path[len - 1] != '\0') {
1463 error = EINVAL;
1464 goto done_free;
1465 }
1466 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1467 error = namei(&nd);
1468 if (error)
1469 goto done_free;
1470 root = nd.ni_vp;
1471 NDFREE_PNBUF(&nd);
1472 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1473 strlcpy(g_path, path, MAXPATHLEN);
1474 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1475 if (error == 0) {
1476 path = g_path;
1477 } else {
1478 /* exit on other errors */
1479 goto done_free;
1480 }
1481 if (root->v_type != VDIR) {
1482 error = ENOTDIR;
1483 vput(root);
1484 goto done_free;
1485 }
1486 VOP_UNLOCK(root);
1487 }
1488
1489 /*
1490 * Find the specified jail, or at least its parent.
1491 * This abuses the file error codes ENOENT and EEXIST.
1492 */
1493 pr = NULL;
1494 inspr = NULL;
1495 deadpr = NULL;
1496 maybe_changed = false;
1497 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1498 namelc = strrchr(name, '.');
1499 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1500 if (*p != '\0')
1501 jid = 0;
1502 }
1503 sx_xlock(&allprison_lock);
1504 drflags = PD_LIST_XLOCKED;
1505 ppr = mypr;
1506 if (!prison_isalive(ppr)) {
1507 /* This jail is dying. This process will surely follow. */
1508 error = EAGAIN;
1509 goto done_deref;
1510 }
1511 if (flags & JAIL_USE_DESC) {
1512 /* Get the jail from its descriptor. */
1513 error = jaildesc_find(td, jfd_in, &pr, &jdcred);
1514 if (error) {
1515 vfs_opterror(opts, error == ENOENT ?
1516 "descriptor to dead jail" :
1517 "not a jail descriptor");
1518 goto done_deref;
1519 }
1520 drflags |= PD_DEREF;
1521 error = priv_check_cred(jdcred, PRIV_JAIL_SET);
1522 if (error == 0 && (flags & JAIL_ATTACH))
1523 error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
1524 crfree(jdcred);
1525 if (error)
1526 goto done_deref;
1527 mtx_lock(&pr->pr_mtx);
1528 drflags |= PD_LOCKED;
1529 if (cuflags == JAIL_CREATE) {
1530 error = EEXIST;
1531 vfs_opterror(opts, "jail %d already exists",
1532 pr->pr_id);
1533 goto done_deref;
1534 }
1535 if (!prison_isalive(pr)) {
1536 /* While a jid can be resurrected, the prison
1537 * itself cannot.
1538 */
1539 error = ENOENT;
1540 vfs_opterror(opts, "jail %d is dying", pr->pr_id);
1541 goto done_deref;
1542 }
1543 if (jid != 0 && jid != pr->pr_id) {
1544 error = EINVAL;
1545 vfs_opterror(opts, "cannot change jid");
1546 goto done_deref;
1547 }
1548 jid = pr->pr_id;
1549 } else if (jid != 0) {
1550 if (jid < 0) {
1551 error = EINVAL;
1552 vfs_opterror(opts, "negative jid");
1553 goto done_deref;
1554 }
1555 /*
1556 * See if a requested jid already exists. Keep track of
1557 * where it can be inserted later.
1558 */
1559 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1560 if (inspr->pr_id < jid)
1561 continue;
1562 if (inspr->pr_id > jid)
1563 break;
1564 if (prison_isalive(inspr)) {
1565 pr = inspr;
1566 mtx_lock(&pr->pr_mtx);
1567 drflags |= PD_LOCKED;
1568 } else {
1569 /* Note a dying jail to handle later. */
1570 deadpr = inspr;
1571 }
1572 inspr = NULL;
1573 break;
1574 }
1575 if (cuflags == JAIL_CREATE && pr != NULL) {
1576 /*
1577 * Even creators that cannot see the jail will
1578 * get EEXIST.
1579 */
1580 error = EEXIST;
1581 vfs_opterror(opts, "jail %d already exists", jid);
1582 goto done_deref;
1583 }
1584 if ((pr == NULL)
1585 ? cuflags == JAIL_UPDATE
1586 : !prison_ischild(mypr, pr)) {
1587 /*
1588 * Updaters get ENOENT for nonexistent jails,
1589 * or for jails they cannot see. The latter
1590 * case is true even for CREATE | UPDATE,
1591 * which normally cannot give this error.
1592 */
1593 error = ENOENT;
1594 vfs_opterror(opts, "jail %d not found", jid);
1595 goto done_deref;
1596 }
1597 }
1598 /*
1599 * If the caller provided a name, look for a jail by that name.
1600 * This has different semantics for creates and updates keyed by jid
1601 * (where the name must not already exist in a different jail),
1602 * and updates keyed by the name itself (where the name must exist
1603 * because that is the jail being updated).
1604 */
1605 namelc = NULL;
1606 if (name != NULL) {
1607 namelc = strrchr(name, '.');
1608 if (namelc == NULL)
1609 namelc = name;
1610 else {
1611 /*
1612 * This is a hierarchical name. Split it into the
1613 * parent and child names, and make sure the parent
1614 * exists or matches an already found jail.
1615 */
1616 if (pr != NULL) {
1617 if (strncmp(name, ppr->pr_name, namelc - name)
1618 || ppr->pr_name[namelc - name] != '\0') {
1619 error = EINVAL;
1620 vfs_opterror(opts,
1621 "cannot change jail's parent");
1622 goto done_deref;
1623 }
1624 } else {
1625 *namelc = '\0';
1626 ppr = prison_find_name(mypr, name);
1627 if (ppr == NULL) {
1628 error = ENOENT;
1629 vfs_opterror(opts,
1630 "jail \"%s\" not found", name);
1631 goto done_deref;
1632 }
1633 mtx_unlock(&ppr->pr_mtx);
1634 if (!prison_isalive(ppr)) {
1635 error = ENOENT;
1636 vfs_opterror(opts,
1637 "jail \"%s\" is dying", name);
1638 goto done_deref;
1639 }
1640 *namelc = '.';
1641 }
1642 namelc++;
1643 }
1644 if (namelc[0] != '\0') {
1645 pnamelen =
1646 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1647 FOREACH_PRISON_CHILD(ppr, tpr) {
1648 if (tpr == pr || !prison_isalive(tpr) ||
1649 strcmp(tpr->pr_name + pnamelen, namelc))
1650 continue;
1651 if (cuflags == JAIL_CREATE || pr != NULL) {
1652 /*
1653 * Create, or update(jid): name must
1654 * not exist in an active sibling jail.
1655 */
1656 error = EEXIST;
1657 vfs_opterror(opts,
1658 "jail \"%s\" already exists", name);
1659 goto done_deref;
1660 }
1661 /* Use this jail for updates. */
1662 pr = tpr;
1663 mtx_lock(&pr->pr_mtx);
1664 drflags |= PD_LOCKED;
1665 break;
1666 }
1667 /*
1668 * Update: name must exist if no jid is specified.
1669 * As with the jid case, the jail must be currently
1670 * visible, or else even CREATE | UPDATE will get
1671 * an error.
1672 */
1673 if ((pr == NULL)
1674 ? cuflags == JAIL_UPDATE
1675 : !prison_isalive(pr)) {
1676 error = ENOENT;
1677 vfs_opterror(opts, "jail \"%s\" not found",
1678 name);
1679 goto done_deref;
1680 }
1681 }
1682 }
1683 /* Update: must provide a desc, jid, or name. */
1684 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1685 error = ENOENT;
1686 vfs_opterror(opts, "update specified no jail");
1687 goto done_deref;
1688 }
1689
1690 /* If there's no prison to update, create a new one and link it in. */
1691 created = pr == NULL;
1692 if (created) {
1693 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1694 if (tpr->pr_childcount >= tpr->pr_childmax) {
1695 error = EPERM;
1696 vfs_opterror(opts, "prison limit exceeded");
1697 goto done_deref;
1698 }
1699
1700 if (deadpr != NULL) {
1701 /*
1702 * The prison being created has the same ID as a dying
1703 * one. Handle this by giving the dying jail a new ID.
1704 * This may cause some confusion to user space, but
1705 * only to those listing dying jails.
1706 */
1707 deadid = get_next_deadid(&dinspr);
1708 if (deadid == 0) {
1709 error = EAGAIN;
1710 vfs_opterror(opts, "no available jail IDs");
1711 goto done_deref;
1712 }
1713 mtx_lock(&deadpr->pr_mtx);
1714 deadpr->pr_id = deadid;
1715 mtx_unlock(&deadpr->pr_mtx);
1716 if (dinspr == deadpr)
1717 inspr = deadpr;
1718 else {
1719 inspr = TAILQ_NEXT(deadpr, pr_list);
1720 TAILQ_REMOVE(&allprison, deadpr, pr_list);
1721 if (dinspr != NULL)
1722 TAILQ_INSERT_AFTER(&allprison, dinspr,
1723 deadpr, pr_list);
1724 else
1725 TAILQ_INSERT_HEAD(&allprison, deadpr,
1726 pr_list);
1727 }
1728 }
1729 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1730 error = EAGAIN;
1731 vfs_opterror(opts, "no available jail IDs");
1732 goto done_deref;
1733 }
1734
1735 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1736 pr->pr_state = PRISON_STATE_INVALID;
1737 refcount_init(&pr->pr_ref, 1);
1738 refcount_init(&pr->pr_uref, 0);
1739 drflags |= PD_DEREF;
1740 LIST_INIT(&pr->pr_children);
1741 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1742 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1743
1744 pr->pr_id = jid;
1745 if (inspr != NULL)
1746 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1747 else
1748 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1749
1750 pr->pr_parent = ppr;
1751 prison_hold(ppr);
1752 prison_proc_hold(ppr);
1753 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1754 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1755 tpr->pr_childcount++;
1756 pr->pr_klist = knlist_alloc(&pr->pr_mtx);
1757
1758 /* Set some default values, and inherit some from the parent. */
1759 if (namelc == NULL)
1760 namelc = "";
1761 if (path == NULL) {
1762 path = "/";
1763 root = mypr->pr_root;
1764 vref(root);
1765 }
1766 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1767 pr->pr_flags |= PR_HOST;
1768 #if defined(INET) || defined(INET6)
1769 #ifdef VIMAGE
1770 if (!(pr_flags & PR_VNET))
1771 #endif
1772 {
1773 #ifdef INET
1774 if (!(ch_flags & PR_IP4_USER))
1775 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1776 else if (!(pr_flags & PR_IP4_USER)) {
1777 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1778 prison_ip_dup(ppr, pr, PR_INET);
1779 }
1780 #endif
1781 #ifdef INET6
1782 if (!(ch_flags & PR_IP6_USER))
1783 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1784 else if (!(pr_flags & PR_IP6_USER)) {
1785 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1786 prison_ip_dup(ppr, pr, PR_INET6);
1787 }
1788 #endif
1789 }
1790 #endif
1791 /* Source address selection is always on by default. */
1792 pr->pr_flags |= _PR_IP_SADDRSEL;
1793
1794 pr->pr_securelevel = ppr->pr_securelevel;
1795 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1796 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1797 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1798
1799 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1800 if (osrelstr == NULL)
1801 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1802 sizeof(pr->pr_osrelease));
1803 else
1804 strlcpy(pr->pr_osrelease, osrelstr,
1805 sizeof(pr->pr_osrelease));
1806
1807 #ifdef VIMAGE
1808 /*
1809 * Allocate a new vnet if specified.
1810 *
1811 * Set PR_VNET now if so, so that the vnet is disposed of
1812 * properly when the jail is destroyed.
1813 */
1814 if (pr_flags & PR_VNET) {
1815 pr->pr_flags |= PR_VNET;
1816 pr->pr_vnet = vnet_alloc();
1817 } else {
1818 pr->pr_vnet = ppr->pr_vnet;
1819 }
1820 #endif
1821 /*
1822 * Allocate a dedicated cpuset for each jail.
1823 * Unlike other initial settings, this may return an error.
1824 */
1825 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1826 if (error)
1827 goto done_deref;
1828
1829 mtx_lock(&pr->pr_mtx);
1830 drflags |= PD_LOCKED;
1831 } else {
1832 /*
1833 * Grab a reference for existing prisons, to ensure they
1834 * continue to exist for the duration of the call.
1835 */
1836 if (!(drflags & PD_DEREF)) {
1837 prison_hold(pr);
1838 drflags |= PD_DEREF;
1839 }
1840 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1841 if ((pr->pr_flags & PR_VNET) &&
1842 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1843 error = EINVAL;
1844 vfs_opterror(opts,
1845 "vnet jails cannot have IP address restrictions");
1846 goto done_deref;
1847 }
1848 #endif
1849 #ifdef INET
1850 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1851 error = EINVAL;
1852 vfs_opterror(opts,
1853 "ip4 cannot be changed after creation");
1854 goto done_deref;
1855 }
1856 #endif
1857 #ifdef INET6
1858 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1859 error = EINVAL;
1860 vfs_opterror(opts,
1861 "ip6 cannot be changed after creation");
1862 goto done_deref;
1863 }
1864 #endif
1865 }
1866
1867 /* Do final error checking before setting anything. */
1868 if (gotslevel) {
1869 if (slevel < ppr->pr_securelevel) {
1870 error = EPERM;
1871 goto done_deref;
1872 }
1873 }
1874 if (gotchildmax) {
1875 if (childmax >= ppr->pr_childmax) {
1876 error = EPERM;
1877 goto done_deref;
1878 }
1879 }
1880 if (gotenforce) {
1881 if (enforce < ppr->pr_enforce_statfs) {
1882 error = EPERM;
1883 goto done_deref;
1884 }
1885 }
1886 if (gotrsnum) {
1887 /*
1888 * devfs_rsnum is a uint16_t
1889 */
1890 if (rsnum < 0 || rsnum > 65535) {
1891 error = EINVAL;
1892 goto done_deref;
1893 }
1894 /*
1895 * Nested jails always inherit parent's devfs ruleset
1896 */
1897 if (jailed(td->td_ucred)) {
1898 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1899 error = EPERM;
1900 goto done_deref;
1901 } else
1902 rsnum = ppr->pr_devfs_rsnum;
1903 }
1904 }
1905 #ifdef INET
1906 if (ip4s > 0) {
1907 if ((ppr->pr_flags & PR_IP4) &&
1908 !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1909 PR_INET)) {
1910 error = EPERM;
1911 goto done_deref;
1912 }
1913 if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1914 error = EADDRINUSE;
1915 vfs_opterror(opts, "IPv4 addresses clash");
1916 goto done_deref;
1917 }
1918 }
1919 #endif
1920 #ifdef INET6
1921 if (ip6s > 0) {
1922 if ((ppr->pr_flags & PR_IP6) &&
1923 !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
1924 PR_INET6)) {
1925 error = EPERM;
1926 goto done_deref;
1927 }
1928 if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
1929 error = EADDRINUSE;
1930 vfs_opterror(opts, "IPv6 addresses clash");
1931 goto done_deref;
1932 }
1933 }
1934 #endif
1935 onamelen = namelen = 0;
1936 if (namelc != NULL) {
1937 /* Give a default name of the jid. Also allow the name to be
1938 * explicitly the jid - but not any other number, and only in
1939 * normal form (no leading zero/etc).
1940 */
1941 if (namelc[0] == '\0')
1942 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1943 else if ((strtoul(namelc, &p, 10) != jid ||
1944 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1945 error = EINVAL;
1946 vfs_opterror(opts,
1947 "name cannot be numeric (unless it is the jid)");
1948 goto done_deref;
1949 }
1950 /*
1951 * Make sure the name isn't too long for the prison or its
1952 * children.
1953 */
1954 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1955 onamelen = strlen(pr->pr_name + pnamelen);
1956 namelen = strlen(namelc);
1957 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1958 error = ENAMETOOLONG;
1959 goto done_deref;
1960 }
1961 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1962 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1963 sizeof(pr->pr_name)) {
1964 error = ENAMETOOLONG;
1965 goto done_deref;
1966 }
1967 }
1968 }
1969 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1970 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1971 error = EPERM;
1972 goto done_deref;
1973 }
1974
1975 /*
1976 * Let modules check their parameters. This requires unlocking and
1977 * then re-locking the prison, but this is still a valid state as long
1978 * as allprison_lock remains xlocked.
1979 */
1980 mtx_unlock(&pr->pr_mtx);
1981 drflags &= ~PD_LOCKED;
1982 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1983 if (error != 0)
1984 goto done_deref;
1985 mtx_lock(&pr->pr_mtx);
1986 drflags |= PD_LOCKED;
1987
1988 /* At this point, all valid parameters should have been noted. */
1989 TAILQ_FOREACH(opt, opts, link) {
1990 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1991 error = EINVAL;
1992 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1993 goto done_deref;
1994 }
1995 }
1996 maybe_changed = true;
1997
1998 /* Set the parameters of the prison. */
1999 #ifdef INET
2000 redo_ip4 = false;
2001 if (pr_flags & PR_IP4_USER) {
2002 pr->pr_flags |= PR_IP4;
2003 prison_ip_set(pr, PR_INET, ip4);
2004 ip4 = NULL;
2005 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2006 #ifdef VIMAGE
2007 if (tpr->pr_flags & PR_VNET) {
2008 descend = 0;
2009 continue;
2010 }
2011 #endif
2012 if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
2013 redo_ip4 = true;
2014 descend = 0;
2015 }
2016 }
2017 }
2018 #endif
2019 #ifdef INET6
2020 redo_ip6 = false;
2021 if (pr_flags & PR_IP6_USER) {
2022 pr->pr_flags |= PR_IP6;
2023 prison_ip_set(pr, PR_INET6, ip6);
2024 ip6 = NULL;
2025 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2026 #ifdef VIMAGE
2027 if (tpr->pr_flags & PR_VNET) {
2028 descend = 0;
2029 continue;
2030 }
2031 #endif
2032 if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
2033 redo_ip6 = true;
2034 descend = 0;
2035 }
2036 }
2037 }
2038 #endif
2039 if (gotslevel) {
2040 pr->pr_securelevel = slevel;
2041 /* Set all child jails to be at least this level. */
2042 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2043 if (tpr->pr_securelevel < slevel)
2044 tpr->pr_securelevel = slevel;
2045 }
2046 if (gotchildmax) {
2047 pr->pr_childmax = childmax;
2048 /* Set all child jails to under this limit. */
2049 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
2050 if (tpr->pr_childmax > childmax - level)
2051 tpr->pr_childmax = childmax > level
2052 ? childmax - level : 0;
2053 }
2054 if (gotenforce) {
2055 pr->pr_enforce_statfs = enforce;
2056 /* Pass this restriction on to the children. */
2057 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2058 if (tpr->pr_enforce_statfs < enforce)
2059 tpr->pr_enforce_statfs = enforce;
2060 }
2061 if (gotrsnum) {
2062 pr->pr_devfs_rsnum = rsnum;
2063 /* Pass this restriction on to the children. */
2064 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2065 tpr->pr_devfs_rsnum = rsnum;
2066 }
2067 if (namelc != NULL) {
2068 if (ppr == &prison0)
2069 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
2070 else
2071 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
2072 ppr->pr_name, namelc);
2073 /* Change this component of child names. */
2074 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2075 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
2076 strlen(tpr->pr_name + onamelen) + 1);
2077 bcopy(pr->pr_name, tpr->pr_name, namelen);
2078 }
2079 }
2080 if (path != NULL) {
2081 /* Try to keep a real-rooted full pathname. */
2082 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
2083 pr->pr_root = root;
2084 root = NULL;
2085 }
2086 if (PR_HOST & ch_flags & ~pr_flags) {
2087 if (pr->pr_flags & PR_HOST) {
2088 /*
2089 * Copy the parent's host info. As with pr_ip4 above,
2090 * the lack of a lock on the parent is not a problem;
2091 * it is always set with allprison_lock at least
2092 * shared, and is held exclusively here.
2093 */
2094 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
2095 sizeof(pr->pr_hostname));
2096 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
2097 sizeof(pr->pr_domainname));
2098 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
2099 sizeof(pr->pr_hostuuid));
2100 pr->pr_hostid = pr->pr_parent->pr_hostid;
2101 }
2102 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
2103 /* Set this prison, and any descendants without PR_HOST. */
2104 if (host != NULL)
2105 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
2106 if (domain != NULL)
2107 strlcpy(pr->pr_domainname, domain,
2108 sizeof(pr->pr_domainname));
2109 if (uuid != NULL)
2110 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
2111 if (gothid)
2112 pr->pr_hostid = hid;
2113 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2114 if (tpr->pr_flags & PR_HOST)
2115 descend = 0;
2116 else {
2117 if (host != NULL)
2118 strlcpy(tpr->pr_hostname,
2119 pr->pr_hostname,
2120 sizeof(tpr->pr_hostname));
2121 if (domain != NULL)
2122 strlcpy(tpr->pr_domainname,
2123 pr->pr_domainname,
2124 sizeof(tpr->pr_domainname));
2125 if (uuid != NULL)
2126 strlcpy(tpr->pr_hostuuid,
2127 pr->pr_hostuuid,
2128 sizeof(tpr->pr_hostuuid));
2129 if (gothid)
2130 tpr->pr_hostid = hid;
2131 }
2132 }
2133 }
2134 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2135 if ((tallow = ch_allow & ~pr_allow))
2136 prison_set_allow_locked(pr, tallow, 0);
2137 /*
2138 * Persistent prisons get an extra reference, and prisons losing their
2139 * persist flag lose that reference.
2140 */
2141 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2142 if (pr_flags & PR_PERSIST) {
2143 prison_hold(pr);
2144 /*
2145 * This may be a new prison's first user reference,
2146 * but wait to call it alive until after OSD calls
2147 * have had a chance to run (and perhaps to fail).
2148 */
2149 refcount_acquire(&pr->pr_uref);
2150 } else {
2151 drflags |= PD_DEUREF;
2152 prison_free_not_last(pr);
2153 }
2154 }
2155 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2156 mtx_unlock(&pr->pr_mtx);
2157 drflags &= ~PD_LOCKED;
2158 /*
2159 * Any errors past this point will need to de-persist newly created
2160 * prisons, as well as call remove methods.
2161 */
2162 if (created)
2163 drflags |= PD_KILL;
2164
2165 #ifdef RACCT
2166 if (racct_enable && created)
2167 prison_racct_attach(pr);
2168 #endif
2169
2170 /* Locks may have prevented a complete restriction of child IP
2171 * addresses. If so, allocate some more memory and try again.
2172 */
2173 #ifdef INET
2174 while (redo_ip4) {
2175 ip4s = pr->pr_addrs[PR_INET]->ips;
2176 MPASS(ip4 == NULL);
2177 ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2178 mtx_lock(&pr->pr_mtx);
2179 redo_ip4 = false;
2180 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2181 #ifdef VIMAGE
2182 if (tpr->pr_flags & PR_VNET) {
2183 descend = 0;
2184 continue;
2185 }
2186 #endif
2187 if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2188 redo_ip4 = true;
2189 }
2190 mtx_unlock(&pr->pr_mtx);
2191 }
2192 #endif
2193 #ifdef INET6
2194 while (redo_ip6) {
2195 ip6s = pr->pr_addrs[PR_INET6]->ips;
2196 MPASS(ip6 == NULL);
2197 ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2198 mtx_lock(&pr->pr_mtx);
2199 redo_ip6 = false;
2200 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2201 #ifdef VIMAGE
2202 if (tpr->pr_flags & PR_VNET) {
2203 descend = 0;
2204 continue;
2205 }
2206 #endif
2207 if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2208 redo_ip6 = true;
2209 }
2210 mtx_unlock(&pr->pr_mtx);
2211 }
2212 #endif
2213
2214 /* Let the modules do their work. */
2215 if (created) {
2216 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2217 if (error)
2218 goto done_deref;
2219 }
2220 error = osd_jail_call(pr, PR_METHOD_SET, opts);
2221 if (error)
2222 goto done_deref;
2223
2224 /*
2225 * A new prison is now ready to be seen; either it has gained a user
2226 * reference via persistence, or is about to gain one via attachment.
2227 */
2228 if (created) {
2229 sx_assert(&allprison_lock, SX_XLOCKED);
2230 prison_knote(ppr, NOTE_JAIL_CHILD | pr->pr_id);
2231 mtx_lock(&pr->pr_mtx);
2232 drflags |= PD_LOCKED;
2233 pr->pr_state = PRISON_STATE_ALIVE;
2234 }
2235
2236 /* Attach this process to the prison if requested. */
2237 if (flags & JAIL_ATTACH) {
2238 error = do_jail_attach(td, pr,
2239 prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
2240 drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
2241 if (error) {
2242 vfs_opterror(opts, "attach failed");
2243 goto done_deref;
2244 }
2245 }
2246
2247 #ifdef RACCT
2248 if (racct_enable && !created) {
2249 if (drflags & PD_LOCKED) {
2250 mtx_unlock(&pr->pr_mtx);
2251 drflags &= ~PD_LOCKED;
2252 }
2253 if (drflags & PD_LIST_XLOCKED) {
2254 sx_xunlock(&allprison_lock);
2255 drflags &= ~PD_LIST_XLOCKED;
2256 }
2257 prison_racct_modify(pr);
2258 }
2259 #endif
2260
2261 if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2262 (pr->pr_root->v_vflag & VV_ROOT) == 0)
2263 printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2264 " file system\n", pr->pr_id);
2265
2266 /*
2267 * Now that the prison is fully created without error, set the
2268 * jail descriptor if one was requested. This is the only
2269 * parameter that is returned to the caller (except the error
2270 * message).
2271 */
2272 if (jfd_out >= 0) {
2273 if (!(drflags & PD_LOCKED)) {
2274 mtx_lock(&pr->pr_mtx);
2275 drflags |= PD_LOCKED;
2276 }
2277 jfd_pos = 2 * vfs_getopt_pos(opts, "desc") + 1;
2278 if (optuio->uio_segflg == UIO_SYSSPACE)
2279 *(int*)optuio->uio_iov[jfd_pos].iov_base = jfd_out;
2280 else
2281 (void)copyout(&jfd_out,
2282 optuio->uio_iov[jfd_pos].iov_base, sizeof(jfd_out));
2283 jaildesc_set_prison(jfp_out, pr);
2284 }
2285
2286 drflags &= ~PD_KILL;
2287 td->td_retval[0] = pr->pr_id;
2288
2289 done_deref:
2290 /*
2291 * Report changes to kevent. This can happen even if the
2292 * system call fails, as changes might have been made before
2293 * the failure.
2294 */
2295 if (maybe_changed && !created)
2296 prison_knote(pr, NOTE_JAIL_SET);
2297 /* Release any temporary prison holds and/or locks. */
2298 if (pr != NULL)
2299 prison_deref(pr, drflags);
2300 else if (drflags & PD_LIST_SLOCKED)
2301 sx_sunlock(&allprison_lock);
2302 else if (drflags & PD_LIST_XLOCKED)
2303 sx_xunlock(&allprison_lock);
2304 if (root != NULL)
2305 vrele(root);
2306 done_errmsg:
2307 if (error) {
2308 /* Write the error message back to userspace. */
2309 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
2310 &errmsg_len) == 0 && errmsg_len > 0) {
2311 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
2312 if (optuio->uio_segflg == UIO_SYSSPACE)
2313 bcopy(errmsg,
2314 optuio->uio_iov[errmsg_pos].iov_base,
2315 errmsg_len);
2316 else
2317 (void)copyout(errmsg,
2318 optuio->uio_iov[errmsg_pos].iov_base,
2319 errmsg_len);
2320 }
2321 }
2322 done_free:
2323 /* Clean up other resources. */
2324 #ifdef INET
2325 prison_ip_free(ip4);
2326 #endif
2327 #ifdef INET6
2328 prison_ip_free(ip6);
2329 #endif
2330 if (jfp_out != NULL)
2331 fdrop(jfp_out, td);
2332 if (error && jfd_out >= 0)
2333 (void)kern_close(td, jfd_out);
2334 if (g_path != NULL)
2335 free(g_path, M_TEMP);
2336 if (opts != NULL)
2337 vfs_freeopts(opts);
2338 prison_free(mypr);
2339 return (error);
2340 }
2341
2342 /*
2343 * Find the next available prison ID. Return the ID on success, or zero
2344 * on failure. Also set a pointer to the allprison list entry the prison
2345 * should be inserted before.
2346 */
2347 static int
get_next_prid(struct prison ** insprp)2348 get_next_prid(struct prison **insprp)
2349 {
2350 struct prison *inspr;
2351 int jid, maxid;
2352
2353 jid = lastprid % JAIL_MAX + 1;
2354 if (TAILQ_EMPTY(&allprison) ||
2355 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2356 /*
2357 * A common case is for all jails to be implicitly numbered,
2358 * which means they'll go on the end of the list, at least
2359 * for the first JAIL_MAX times.
2360 */
2361 inspr = NULL;
2362 } else {
2363 /*
2364 * Take two passes through the allprison list: first starting
2365 * with the proposed jid, then ending with it.
2366 */
2367 for (maxid = JAIL_MAX; maxid != 0; ) {
2368 TAILQ_FOREACH(inspr, &allprison, pr_list) {
2369 if (inspr->pr_id < jid)
2370 continue;
2371 if (inspr->pr_id > jid) {
2372 /* Found an opening. */
2373 maxid = 0;
2374 break;
2375 }
2376 if (++jid > maxid) {
2377 if (lastprid == maxid || lastprid == 0)
2378 {
2379 /*
2380 * The entire legal range
2381 * has been traversed
2382 */
2383 return 0;
2384 }
2385 /* Try again from the start. */
2386 jid = 1;
2387 maxid = lastprid;
2388 break;
2389 }
2390 }
2391 if (inspr == NULL) {
2392 /* Found room at the end of the list. */
2393 break;
2394 }
2395 }
2396 }
2397 *insprp = inspr;
2398 lastprid = jid;
2399 return (jid);
2400 }
2401
2402 /*
2403 * Find the next available ID for a renumbered dead prison. This is the same
2404 * as get_next_prid, but counting backward from the end of the range.
2405 */
2406 static int
get_next_deadid(struct prison ** dinsprp)2407 get_next_deadid(struct prison **dinsprp)
2408 {
2409 struct prison *dinspr;
2410 int deadid, minid;
2411
2412 deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2413 /*
2414 * Take two reverse passes through the allprison list: first
2415 * starting with the proposed deadid, then ending with it.
2416 */
2417 for (minid = 1; minid != 0; ) {
2418 TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2419 if (dinspr->pr_id > deadid)
2420 continue;
2421 if (dinspr->pr_id < deadid) {
2422 /* Found an opening. */
2423 minid = 0;
2424 break;
2425 }
2426 if (--deadid < minid) {
2427 if (lastdeadid == minid || lastdeadid == 0)
2428 {
2429 /*
2430 * The entire legal range
2431 * has been traversed
2432 */
2433 return 0;
2434 }
2435 /* Try again from the end. */
2436 deadid = JAIL_MAX;
2437 minid = lastdeadid;
2438 break;
2439 }
2440 }
2441 if (dinspr == NULL) {
2442 /* Found room at the beginning of the list. */
2443 break;
2444 }
2445 }
2446 *dinsprp = dinspr;
2447 lastdeadid = deadid;
2448 return (deadid);
2449 }
2450
2451 /*
2452 * struct jail_get_args {
2453 * struct iovec *iovp;
2454 * unsigned int iovcnt;
2455 * int flags;
2456 * };
2457 */
2458 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)2459 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2460 {
2461 struct uio *auio;
2462 int error;
2463
2464 /* Check that we have an even number of iovecs. */
2465 if (uap->iovcnt & 1)
2466 return (EINVAL);
2467
2468 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2469 if (error)
2470 return (error);
2471 error = kern_jail_get(td, auio, uap->flags);
2472 if (error == 0)
2473 error = copyout(auio->uio_iov, uap->iovp,
2474 uap->iovcnt * sizeof(struct iovec));
2475 freeuio(auio);
2476 return (error);
2477 }
2478
2479 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2480 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2481 {
2482 struct bool_flags *bf;
2483 struct file *jfp_out;
2484 struct jailsys_flags *jsf;
2485 struct prison *pr, *mypr;
2486 struct vfsopt *opt;
2487 struct vfsoptlist *opts;
2488 char *errmsg, *name;
2489 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2490 int jfd_in, jfd_out;
2491 unsigned f;
2492
2493 if (flags & ~JAIL_GET_MASK)
2494 return (EINVAL);
2495 if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
2496 (JAIL_USE_DESC | JAIL_AT_DESC))
2497 return (EINVAL);
2498
2499 /* Get the parameter list. */
2500 error = vfs_buildopts(optuio, &opts);
2501 if (error)
2502 return (error);
2503 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2504 mypr = td->td_ucred->cr_prison;
2505 prison_hold(mypr);
2506 pr = NULL;
2507 jfp_out = NULL;
2508 jfd_out = -1;
2509
2510 /*
2511 * Find the prison specified by one of: desc, lastjid, jid, name.
2512 */
2513 sx_slock(&allprison_lock);
2514 drflags = PD_LIST_SLOCKED;
2515
2516 error = vfs_copyopt(opts, "desc", &jfd_in, sizeof(jfd_in));
2517 if (error == ENOENT) {
2518 if (flags & (JAIL_AT_DESC | JAIL_GET_DESC | JAIL_OWN_DESC)) {
2519 vfs_opterror(opts, "missing desc");
2520 goto done;
2521 }
2522 } else if (error == 0) {
2523 if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
2524 JAIL_OWN_DESC))) {
2525 error = EINVAL;
2526 vfs_opterror(opts, "unexpected desc");
2527 goto done;
2528 }
2529 if (flags & JAIL_USE_DESC) {
2530 /* Get the jail from its descriptor. */
2531 error = jaildesc_find(td, jfd_in, &pr, NULL);
2532 if (error) {
2533 vfs_opterror(opts, error == ENOENT ?
2534 "descriptor to dead jail" :
2535 "not a jail descriptor");
2536 goto done;
2537 }
2538 drflags |= PD_DEREF;
2539 mtx_lock(&pr->pr_mtx);
2540 drflags |= PD_LOCKED;
2541 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2542 error = ENOENT;
2543 vfs_opterror(opts, "jail %d is dying",
2544 pr->pr_id);
2545 goto done;
2546 }
2547 goto found_prison;
2548 }
2549 if (flags & JAIL_AT_DESC) {
2550 /* Look up jails based on the descriptor's prison. */
2551 prison_free(mypr);
2552 error = jaildesc_find(td, jfd_in, &mypr, NULL);
2553 if (error != 0) {
2554 vfs_opterror(opts, error == ENOENT ?
2555 "descriptor to dead jail" :
2556 "not a jail descriptor");
2557 goto done;
2558 }
2559 }
2560 if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
2561 /* Allocate a jail descriptor to return later. */
2562 error = jaildesc_alloc(td, &jfp_out, &jfd_out,
2563 flags & JAIL_OWN_DESC);
2564 if (error)
2565 goto done;
2566 }
2567 } else
2568 goto done;
2569
2570 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2571 if (error == 0) {
2572 TAILQ_FOREACH(pr, &allprison, pr_list) {
2573 if (pr->pr_id > jid &&
2574 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2575 prison_ischild(mypr, pr)) {
2576 mtx_lock(&pr->pr_mtx);
2577 drflags |= PD_LOCKED;
2578 goto found_prison;
2579 }
2580 }
2581 error = ENOENT;
2582 vfs_opterror(opts, "no jail after %d", jid);
2583 goto done;
2584 } else if (error != ENOENT)
2585 goto done;
2586
2587 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2588 if (error == 0) {
2589 if (jid != 0) {
2590 pr = prison_find_child(mypr, jid);
2591 if (pr != NULL) {
2592 drflags |= PD_LOCKED;
2593 if (!(prison_isalive(pr) ||
2594 (flags & JAIL_DYING))) {
2595 error = ENOENT;
2596 vfs_opterror(opts, "jail %d is dying",
2597 jid);
2598 goto done;
2599 }
2600 goto found_prison;
2601 }
2602 error = ENOENT;
2603 vfs_opterror(opts, "jail %d not found", jid);
2604 goto done;
2605 }
2606 } else if (error != ENOENT)
2607 goto done;
2608
2609 error = vfs_getopt(opts, "name", (void **)&name, &len);
2610 if (error == 0) {
2611 if (len == 0 || name[len - 1] != '\0') {
2612 error = EINVAL;
2613 goto done;
2614 }
2615 pr = prison_find_name(mypr, name);
2616 if (pr != NULL) {
2617 drflags |= PD_LOCKED;
2618 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2619 error = ENOENT;
2620 vfs_opterror(opts, "jail \"%s\" is dying",
2621 name);
2622 goto done;
2623 }
2624 goto found_prison;
2625 }
2626 error = ENOENT;
2627 vfs_opterror(opts, "jail \"%s\" not found", name);
2628 goto done;
2629 } else if (error != ENOENT)
2630 goto done;
2631
2632 vfs_opterror(opts, "no jail specified");
2633 error = ENOENT;
2634 goto done;
2635
2636 found_prison:
2637 /* Get the parameters of the prison. */
2638 if (!(drflags & PD_DEREF)) {
2639 prison_hold(pr);
2640 drflags |= PD_DEREF;
2641 }
2642 td->td_retval[0] = pr->pr_id;
2643 if (jfd_out >= 0) {
2644 error = vfs_setopt(opts, "desc", &jfd_out, sizeof(jfd_out));
2645 if (error != 0 && error != ENOENT)
2646 goto done;
2647 jaildesc_set_prison(jfp_out, pr);
2648 }
2649 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2650 if (error != 0 && error != ENOENT)
2651 goto done;
2652 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2653 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2654 if (error != 0 && error != ENOENT)
2655 goto done;
2656 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2657 if (error != 0 && error != ENOENT)
2658 goto done;
2659 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2660 sizeof(pr->pr_cpuset->cs_id));
2661 if (error != 0 && error != ENOENT)
2662 goto done;
2663 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2664 if (error != 0 && error != ENOENT)
2665 goto done;
2666 #ifdef INET
2667 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_addrs[PR_INET]->pr_ip,
2668 pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2669 pr_families[PR_INET].size : 0 );
2670 if (error != 0 && error != ENOENT)
2671 goto done;
2672 #endif
2673 #ifdef INET6
2674 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_addrs[PR_INET6]->pr_ip,
2675 pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2676 pr_families[PR_INET6].size : 0 );
2677 if (error != 0 && error != ENOENT)
2678 goto done;
2679 #endif
2680 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2681 sizeof(pr->pr_securelevel));
2682 if (error != 0 && error != ENOENT)
2683 goto done;
2684 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2685 sizeof(pr->pr_childcount));
2686 if (error != 0 && error != ENOENT)
2687 goto done;
2688 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2689 sizeof(pr->pr_childmax));
2690 if (error != 0 && error != ENOENT)
2691 goto done;
2692 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2693 if (error != 0 && error != ENOENT)
2694 goto done;
2695 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2696 if (error != 0 && error != ENOENT)
2697 goto done;
2698 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2699 if (error != 0 && error != ENOENT)
2700 goto done;
2701 #ifdef COMPAT_FREEBSD32
2702 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2703 uint32_t hid32 = pr->pr_hostid;
2704
2705 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2706 } else
2707 #endif
2708 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2709 sizeof(pr->pr_hostid));
2710 if (error != 0 && error != ENOENT)
2711 goto done;
2712 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2713 sizeof(pr->pr_enforce_statfs));
2714 if (error != 0 && error != ENOENT)
2715 goto done;
2716 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2717 sizeof(pr->pr_devfs_rsnum));
2718 if (error != 0 && error != ENOENT)
2719 goto done;
2720 for (bf = pr_flag_bool;
2721 bf < pr_flag_bool + nitems(pr_flag_bool);
2722 bf++) {
2723 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2724 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2725 if (error != 0 && error != ENOENT)
2726 goto done;
2727 i = !i;
2728 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2729 if (error != 0 && error != ENOENT)
2730 goto done;
2731 }
2732 for (jsf = pr_flag_jailsys;
2733 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2734 jsf++) {
2735 f = pr->pr_flags & (jsf->disable | jsf->new);
2736 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2737 : (f == jsf->new) ? JAIL_SYS_NEW
2738 : JAIL_SYS_INHERIT;
2739 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2740 if (error != 0 && error != ENOENT)
2741 goto done;
2742 }
2743 for (bf = pr_flag_allow;
2744 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2745 atomic_load_int(&bf->flag) != 0;
2746 bf++) {
2747 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2748 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2749 if (error != 0 && error != ENOENT)
2750 goto done;
2751 i = !i;
2752 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2753 if (error != 0 && error != ENOENT)
2754 goto done;
2755 }
2756 i = !prison_isalive(pr);
2757 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2758 if (error != 0 && error != ENOENT)
2759 goto done;
2760 i = !i;
2761 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2762 if (error != 0 && error != ENOENT)
2763 goto done;
2764 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2765 sizeof(pr->pr_osreldate));
2766 if (error != 0 && error != ENOENT)
2767 goto done;
2768 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2769 if (error != 0 && error != ENOENT)
2770 goto done;
2771
2772 /* Get the module parameters. */
2773 mtx_unlock(&pr->pr_mtx);
2774 drflags &= ~PD_LOCKED;
2775 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2776 if (error)
2777 goto done;
2778 prison_deref(pr, drflags);
2779 pr = NULL;
2780 drflags = 0;
2781
2782 /* By now, all parameters should have been noted. */
2783 TAILQ_FOREACH(opt, opts, link) {
2784 if (!opt->seen &&
2785 (strstr(opt->name, JAIL_META_PRIVATE ".") == opt->name ||
2786 strstr(opt->name, JAIL_META_SHARED ".") == opt->name)) {
2787 /* Communicate back a missing key. */
2788 free(opt->value, M_MOUNT);
2789 opt->value = NULL;
2790 opt->len = 0;
2791 continue;
2792 }
2793 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2794 error = EINVAL;
2795 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2796 goto done;
2797 }
2798 }
2799
2800 /* Write the fetched parameters back to userspace. */
2801 error = 0;
2802 TAILQ_FOREACH(opt, opts, link) {
2803 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2804 pos = 2 * opt->pos + 1;
2805 optuio->uio_iov[pos].iov_len = opt->len;
2806 if (opt->value != NULL) {
2807 if (optuio->uio_segflg == UIO_SYSSPACE) {
2808 bcopy(opt->value,
2809 optuio->uio_iov[pos].iov_base,
2810 opt->len);
2811 } else {
2812 error = copyout(opt->value,
2813 optuio->uio_iov[pos].iov_base,
2814 opt->len);
2815 if (error)
2816 break;
2817 }
2818 }
2819 }
2820 }
2821
2822 done:
2823 /* Release any temporary prison holds and/or locks. */
2824 if (pr != NULL)
2825 prison_deref(pr, drflags);
2826 else if (drflags & PD_LIST_SLOCKED)
2827 sx_sunlock(&allprison_lock);
2828 else if (drflags & PD_LIST_XLOCKED)
2829 sx_xunlock(&allprison_lock);
2830 /* Clean up other resources. */
2831 if (jfp_out != NULL)
2832 (void)fdrop(jfp_out, td);
2833 if (error && jfd_out >= 0)
2834 (void)kern_close(td, jfd_out);
2835 if (error && errmsg_pos >= 0) {
2836 /* Write the error message back to userspace. */
2837 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2838 errmsg_pos = 2 * errmsg_pos + 1;
2839 if (errmsg_len > 0) {
2840 if (optuio->uio_segflg == UIO_SYSSPACE)
2841 bcopy(errmsg,
2842 optuio->uio_iov[errmsg_pos].iov_base,
2843 errmsg_len);
2844 else
2845 (void)copyout(errmsg,
2846 optuio->uio_iov[errmsg_pos].iov_base,
2847 errmsg_len);
2848 }
2849 }
2850 vfs_freeopts(opts);
2851 prison_free(mypr);
2852 return (error);
2853 }
2854
2855 /*
2856 * struct jail_remove_args {
2857 * int jid;
2858 * };
2859 */
2860 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2861 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2862 {
2863 struct prison *pr;
2864 int error;
2865
2866 error = priv_check(td, PRIV_JAIL_REMOVE);
2867 if (error)
2868 return (error);
2869
2870 sx_xlock(&allprison_lock);
2871 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2872 if (pr == NULL) {
2873 sx_xunlock(&allprison_lock);
2874 return (EINVAL);
2875 }
2876 prison_hold(pr);
2877 prison_remove(pr);
2878 return (0);
2879 }
2880
2881 /*
2882 * struct jail_remove_jd_args {
2883 * int fd;
2884 * };
2885 */
2886 int
sys_jail_remove_jd(struct thread * td,struct jail_remove_jd_args * uap)2887 sys_jail_remove_jd(struct thread *td, struct jail_remove_jd_args *uap)
2888 {
2889 struct prison *pr;
2890 struct ucred *jdcred;
2891 int error;
2892
2893 error = jaildesc_find(td, uap->fd, &pr, &jdcred);
2894 if (error)
2895 return (error);
2896 error = priv_check_cred(jdcred, PRIV_JAIL_REMOVE);
2897 crfree(jdcred);
2898 if (error) {
2899 prison_free(pr);
2900 return (error);
2901 }
2902 sx_xlock(&allprison_lock);
2903 mtx_lock(&pr->pr_mtx);
2904 prison_remove(pr);
2905 return (0);
2906 }
2907
2908 /*
2909 * Begin the removal process for a prison. The allprison lock should
2910 * be held exclusively, and the prison should be both locked and held.
2911 */
2912 void
prison_remove(struct prison * pr)2913 prison_remove(struct prison *pr)
2914 {
2915 sx_assert(&allprison_lock, SA_XLOCKED);
2916 mtx_assert(&pr->pr_mtx, MA_OWNED);
2917 prison_deref(pr, PD_KILL | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2918 }
2919
2920 /*
2921 * struct jail_attach_args {
2922 * int jid;
2923 * };
2924 */
2925 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)2926 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2927 {
2928 struct prison *pr;
2929 int error;
2930
2931 error = priv_check(td, PRIV_JAIL_ATTACH);
2932 if (error)
2933 return (error);
2934
2935 sx_slock(&allprison_lock);
2936 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2937 if (pr == NULL) {
2938 sx_sunlock(&allprison_lock);
2939 return (EINVAL);
2940 }
2941
2942 /* Do not allow a process to attach to a prison that is not alive. */
2943 if (!prison_isalive(pr)) {
2944 mtx_unlock(&pr->pr_mtx);
2945 sx_sunlock(&allprison_lock);
2946 return (EINVAL);
2947 }
2948
2949 return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2950 }
2951
2952 /*
2953 * struct jail_attach_jd_args {
2954 * int fd;
2955 * };
2956 */
2957 int
sys_jail_attach_jd(struct thread * td,struct jail_attach_jd_args * uap)2958 sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
2959 {
2960 struct prison *pr;
2961 struct ucred *jdcred;
2962 int drflags, error;
2963
2964 sx_slock(&allprison_lock);
2965 drflags = PD_LIST_SLOCKED;
2966 error = jaildesc_find(td, uap->fd, &pr, &jdcred);
2967 if (error)
2968 goto fail;
2969 drflags |= PD_DEREF;
2970 error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
2971 crfree(jdcred);
2972 if (error)
2973 goto fail;
2974 mtx_lock(&pr->pr_mtx);
2975 drflags |= PD_LOCKED;
2976
2977 /* Do not allow a process to attach to a prison that is not alive. */
2978 if (!prison_isalive(pr)) {
2979 error = EINVAL;
2980 goto fail;
2981 }
2982
2983 return (do_jail_attach(td, pr, drflags));
2984
2985 fail:
2986 prison_deref(pr, drflags);
2987 return (error);
2988 }
2989
2990 static int
do_jail_attach(struct thread * td,struct prison * pr,int drflags)2991 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2992 {
2993 struct proc *p;
2994 struct ucred *newcred, *oldcred;
2995 int error;
2996
2997 mtx_assert(&pr->pr_mtx, MA_OWNED);
2998 sx_assert(&allprison_lock, SX_LOCKED);
2999 drflags &= PD_LOCK_FLAGS;
3000 /*
3001 * XXX: Note that there is a slight race here if two threads
3002 * in the same privileged process attempt to attach to two
3003 * different jails at the same time. It is important for
3004 * user processes not to do this, or they might end up with
3005 * a process root from one prison, but attached to the jail
3006 * of another.
3007 */
3008 if (!(drflags & PD_DEREF)) {
3009 prison_hold(pr);
3010 drflags |= PD_DEREF;
3011 }
3012 refcount_acquire(&pr->pr_uref);
3013 drflags |= PD_DEUREF;
3014 mtx_unlock(&pr->pr_mtx);
3015 drflags &= ~PD_LOCKED;
3016
3017 /* Let modules do whatever they need to prepare for attaching. */
3018 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
3019 if (error) {
3020 prison_deref(pr, drflags);
3021 return (error);
3022 }
3023 sx_unlock(&allprison_lock);
3024 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
3025
3026 /*
3027 * Reparent the newly attached process to this jail.
3028 */
3029 p = td->td_proc;
3030 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
3031 if (error)
3032 goto e_revert_osd;
3033
3034 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
3035 if ((error = change_dir(pr->pr_root, td)) != 0)
3036 goto e_unlock;
3037 #ifdef MAC
3038 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
3039 goto e_unlock;
3040 #endif
3041 VOP_UNLOCK(pr->pr_root);
3042 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
3043 goto e_revert_osd;
3044
3045 newcred = crget();
3046 PROC_LOCK(p);
3047 oldcred = crcopysafe(p, newcred);
3048 newcred->cr_prison = pr;
3049 #ifdef RACCT
3050 racct_proc_ucred_changed(p, oldcred, newcred);
3051 #endif
3052 #ifdef RCTL
3053 crhold(newcred);
3054 #endif
3055 /*
3056 * Takes over 'newcred''s reference, so 'newcred' must not be used
3057 * besides this point except on RCTL where we took an additional
3058 * reference above.
3059 */
3060 proc_set_cred(p, newcred);
3061 setsugid(p);
3062 PROC_UNLOCK(p);
3063 #ifdef RCTL
3064 rctl_proc_ucred_changed(p, newcred);
3065 crfree(newcred);
3066 #endif
3067 prison_proc_relink(oldcred->cr_prison, pr, p);
3068 prison_deref(oldcred->cr_prison, drflags);
3069 crfree(oldcred);
3070 prison_knote(pr, NOTE_JAIL_ATTACH | td->td_proc->p_pid);
3071
3072 /*
3073 * If the prison was killed while changing credentials, die along
3074 * with it.
3075 */
3076 if (!prison_isalive(pr)) {
3077 PROC_LOCK(p);
3078 kern_psignal(p, SIGKILL);
3079 PROC_UNLOCK(p);
3080 }
3081
3082 return (0);
3083
3084 e_unlock:
3085 VOP_UNLOCK(pr->pr_root);
3086 e_revert_osd:
3087 /* Tell modules this thread is still in its old jail after all. */
3088 sx_slock(&allprison_lock);
3089 drflags |= PD_LIST_SLOCKED;
3090 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
3091 prison_deref(pr, drflags);
3092 return (error);
3093 }
3094
3095 /*
3096 * Returns a locked prison instance, or NULL on failure.
3097 */
3098 struct prison *
prison_find(int prid)3099 prison_find(int prid)
3100 {
3101 struct prison *pr;
3102
3103 sx_assert(&allprison_lock, SX_LOCKED);
3104 TAILQ_FOREACH(pr, &allprison, pr_list) {
3105 if (pr->pr_id < prid)
3106 continue;
3107 if (pr->pr_id > prid)
3108 break;
3109 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
3110 mtx_lock(&pr->pr_mtx);
3111 return (pr);
3112 }
3113 return (NULL);
3114 }
3115
3116 /*
3117 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
3118 */
3119 struct prison *
prison_find_child(struct prison * mypr,int prid)3120 prison_find_child(struct prison *mypr, int prid)
3121 {
3122 struct prison *pr;
3123 int descend;
3124
3125 sx_assert(&allprison_lock, SX_LOCKED);
3126 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3127 if (pr->pr_id == prid) {
3128 KASSERT(prison_isvalid(pr),
3129 ("Found invalid prison %p", pr));
3130 mtx_lock(&pr->pr_mtx);
3131 return (pr);
3132 }
3133 }
3134 return (NULL);
3135 }
3136
3137 /*
3138 * Look for the name relative to mypr. Returns a locked prison or NULL.
3139 */
3140 struct prison *
prison_find_name(struct prison * mypr,const char * name)3141 prison_find_name(struct prison *mypr, const char *name)
3142 {
3143 struct prison *pr, *deadpr;
3144 size_t mylen;
3145 int descend;
3146
3147 sx_assert(&allprison_lock, SX_LOCKED);
3148 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
3149 deadpr = NULL;
3150 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3151 if (!strcmp(pr->pr_name + mylen, name)) {
3152 KASSERT(prison_isvalid(pr),
3153 ("Found invalid prison %p", pr));
3154 if (prison_isalive(pr)) {
3155 mtx_lock(&pr->pr_mtx);
3156 return (pr);
3157 }
3158 deadpr = pr;
3159 }
3160 }
3161 /* There was no valid prison - perhaps there was a dying one. */
3162 if (deadpr != NULL)
3163 mtx_lock(&deadpr->pr_mtx);
3164 return (deadpr);
3165 }
3166
3167 /*
3168 * See if a prison has the specific flag set. The prison should be locked,
3169 * unless checking for flags that are only set at jail creation (such as
3170 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
3171 * to any other prison data.
3172 */
3173 bool
prison_flag(struct ucred * cred,unsigned flag)3174 prison_flag(struct ucred *cred, unsigned flag)
3175 {
3176
3177 return ((cred->cr_prison->pr_flags & flag) != 0);
3178 }
3179
3180 /*
3181 * See if a prison has the specific allow flag set.
3182 * The prison *should* be locked, or only a single bit is examined, without
3183 * regard to any other prison data.
3184 */
3185 bool
prison_allow(struct ucred * cred,unsigned flag)3186 prison_allow(struct ucred *cred, unsigned flag)
3187 {
3188
3189 return ((cred->cr_prison->pr_allow & flag) != 0);
3190 }
3191
3192 /*
3193 * Hold a prison reference, by incrementing pr_ref. It is generally
3194 * an error to hold a prison that does not already have a reference.
3195 * A prison record will remain valid as long as it has at least one
3196 * reference, and will not be removed as long as either the prison
3197 * mutex or the allprison lock is held (allprison_lock may be shared).
3198 */
3199 void
prison_hold_locked(struct prison * pr)3200 prison_hold_locked(struct prison *pr)
3201 {
3202
3203 /* Locking is no longer required. */
3204 prison_hold(pr);
3205 }
3206
3207 void
prison_hold(struct prison * pr)3208 prison_hold(struct prison *pr)
3209 {
3210 #ifdef INVARIANTS
3211 int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
3212
3213 KASSERT(was_valid,
3214 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
3215 #else
3216 refcount_acquire(&pr->pr_ref);
3217 #endif
3218 }
3219
3220 /*
3221 * Remove a prison reference. If that was the last reference, the
3222 * prison will be removed (at a later time).
3223 */
3224 void
prison_free_locked(struct prison * pr)3225 prison_free_locked(struct prison *pr)
3226 {
3227
3228 mtx_assert(&pr->pr_mtx, MA_OWNED);
3229 /*
3230 * Locking is no longer required, but unlock because the caller
3231 * expects it.
3232 */
3233 mtx_unlock(&pr->pr_mtx);
3234 prison_free(pr);
3235 }
3236
3237 void
prison_free(struct prison * pr)3238 prison_free(struct prison *pr)
3239 {
3240
3241 KASSERT(refcount_load(&pr->pr_ref) > 0,
3242 ("Trying to free dead prison %p (jid=%d).",
3243 pr, pr->pr_id));
3244 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3245 /*
3246 * Don't remove the last reference in this context,
3247 * in case there are locks held.
3248 */
3249 taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3250 }
3251 }
3252
3253 static void
prison_free_not_last(struct prison * pr)3254 prison_free_not_last(struct prison *pr)
3255 {
3256 #ifdef INVARIANTS
3257 int lastref;
3258
3259 KASSERT(refcount_load(&pr->pr_ref) > 0,
3260 ("Trying to free dead prison %p (jid=%d).",
3261 pr, pr->pr_id));
3262 lastref = refcount_release(&pr->pr_ref);
3263 KASSERT(!lastref,
3264 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
3265 pr, pr->pr_id));
3266 #else
3267 refcount_release(&pr->pr_ref);
3268 #endif
3269 }
3270
3271 /*
3272 * Hold a prison for user visibility, by incrementing pr_uref.
3273 * It is generally an error to hold a prison that isn't already
3274 * user-visible, except through the jail system calls. It is also
3275 * an error to hold an invalid prison. A prison record will remain
3276 * alive as long as it has at least one user reference, and will not
3277 * be set to the dying state until the prison mutex and allprison_lock
3278 * are both freed.
3279 */
3280 void
prison_proc_hold(struct prison * pr)3281 prison_proc_hold(struct prison *pr)
3282 {
3283 #ifdef INVARIANTS
3284 int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
3285
3286 KASSERT(was_alive,
3287 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
3288 #else
3289 refcount_acquire(&pr->pr_uref);
3290 #endif
3291 }
3292
3293 /*
3294 * Remove a prison user reference. If it was the last reference, the
3295 * prison will be considered "dying", and may be removed once all of
3296 * its references are dropped.
3297 */
3298 void
prison_proc_free(struct prison * pr)3299 prison_proc_free(struct prison *pr)
3300 {
3301
3302 /*
3303 * Locking is only required when releasing the last reference.
3304 * This allows assurance that a locked prison will remain alive
3305 * until it is unlocked.
3306 */
3307 KASSERT(refcount_load(&pr->pr_uref) > 0,
3308 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
3309 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3310 /*
3311 * Don't remove the last user reference in this context,
3312 * which is expected to be a process that is not only locked,
3313 * but also half dead. Add a reference so any calls to
3314 * prison_free() won't re-submit the task.
3315 */
3316 prison_hold(pr);
3317 mtx_lock(&pr->pr_mtx);
3318 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
3319 ("Redundant last reference in prison_proc_free (jid=%d)",
3320 pr->pr_id));
3321 pr->pr_flags |= PR_COMPLETE_PROC;
3322 mtx_unlock(&pr->pr_mtx);
3323 taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3324 }
3325 }
3326
3327 static void
prison_proc_free_not_last(struct prison * pr)3328 prison_proc_free_not_last(struct prison *pr)
3329 {
3330 #ifdef INVARIANTS
3331 int lastref;
3332
3333 KASSERT(refcount_load(&pr->pr_uref) > 0,
3334 ("Trying to free dead prison %p (jid=%d).",
3335 pr, pr->pr_id));
3336 lastref = refcount_release(&pr->pr_uref);
3337 KASSERT(!lastref,
3338 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3339 pr, pr->pr_id));
3340 #else
3341 refcount_release(&pr->pr_uref);
3342 #endif
3343 }
3344
3345 void
prison_proc_link(struct prison * pr,struct proc * p)3346 prison_proc_link(struct prison *pr, struct proc *p)
3347 {
3348
3349 sx_assert(&allproc_lock, SA_XLOCKED);
3350 LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3351 }
3352
3353 void
prison_proc_unlink(struct prison * pr,struct proc * p)3354 prison_proc_unlink(struct prison *pr, struct proc *p)
3355 {
3356
3357 sx_assert(&allproc_lock, SA_XLOCKED);
3358 LIST_REMOVE(p, p_jaillist);
3359 }
3360
3361 static void
prison_proc_relink(struct prison * opr,struct prison * npr,struct proc * p)3362 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3363 {
3364
3365 sx_xlock(&allproc_lock);
3366 prison_proc_unlink(opr, p);
3367 prison_proc_link(npr, p);
3368 sx_xunlock(&allproc_lock);
3369 }
3370
3371 /*
3372 * Complete a call to either prison_free or prison_proc_free.
3373 */
3374 static void
prison_complete(void * context,int pending)3375 prison_complete(void *context, int pending)
3376 {
3377 struct prison *pr = context;
3378 int drflags;
3379
3380 /*
3381 * This could be called to release the last reference, or the last
3382 * user reference (plus the reference held in prison_proc_free).
3383 */
3384 drflags = prison_lock_xlock(pr, PD_DEREF);
3385 if (pr->pr_flags & PR_COMPLETE_PROC) {
3386 pr->pr_flags &= ~PR_COMPLETE_PROC;
3387 drflags |= PD_DEUREF;
3388 }
3389 prison_deref(pr, drflags);
3390 }
3391
3392 static void
prison_kill_processes_cb(struct proc * p,void * arg __unused)3393 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3394 {
3395
3396 kern_psignal(p, SIGKILL);
3397 }
3398
3399 /*
3400 * Note the iteration does not guarantee acting on all processes.
3401 * Most notably there may be fork or jail_attach in progress.
3402 */
3403 void
prison_proc_iterate(struct prison * pr,void (* cb)(struct proc *,void *),void * cbarg)3404 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3405 void *cbarg)
3406 {
3407 struct prison *ppr;
3408 struct proc *p;
3409
3410 if (atomic_load_int(&pr->pr_childcount) == 0) {
3411 sx_slock(&allproc_lock);
3412 LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3413 if (p->p_state == PRS_NEW)
3414 continue;
3415 PROC_LOCK(p);
3416 cb(p, cbarg);
3417 PROC_UNLOCK(p);
3418 }
3419 sx_sunlock(&allproc_lock);
3420 if (atomic_load_int(&pr->pr_childcount) == 0)
3421 return;
3422 /*
3423 * Some jails popped up during the iteration, fall through to a
3424 * system-wide search.
3425 */
3426 }
3427
3428 sx_slock(&allproc_lock);
3429 FOREACH_PROC_IN_SYSTEM(p) {
3430 PROC_LOCK(p);
3431 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3432 for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3433 ppr = ppr->pr_parent) {
3434 if (ppr == pr) {
3435 cb(p, cbarg);
3436 break;
3437 }
3438 }
3439 }
3440 PROC_UNLOCK(p);
3441 }
3442 sx_sunlock(&allproc_lock);
3443 }
3444
3445 /*
3446 * Remove a prison reference and/or user reference (usually).
3447 * This assumes context that allows sleeping (for allprison_lock),
3448 * with no non-sleeping locks held, except perhaps the prison itself.
3449 * If there are no more references, release and delist the prison.
3450 * On completion, the prison lock and the allprison lock are both
3451 * unlocked.
3452 */
3453 static void
prison_deref(struct prison * pr,int flags)3454 prison_deref(struct prison *pr, int flags)
3455 {
3456 struct prisonlist freeprison;
3457 struct prison *killpr, *rpr, *ppr, *tpr;
3458
3459 killpr = NULL;
3460 TAILQ_INIT(&freeprison);
3461 /*
3462 * Release this prison as requested, which may cause its parent
3463 * to be released, and then maybe its grandparent, etc.
3464 */
3465 for (;;) {
3466 if (flags & PD_KILL) {
3467 /* Kill the prison and its descendents. */
3468 KASSERT(pr != &prison0,
3469 ("prison_deref trying to kill prison0"));
3470 if (!prison_isalive(pr)) {
3471 /* Silently ignore already-dying prisons. */
3472 flags &= ~PD_KILL;
3473 } else {
3474 if (!(flags & PD_DEREF)) {
3475 prison_hold(pr);
3476 flags |= PD_DEREF;
3477 }
3478 flags = prison_lock_xlock(pr, flags);
3479 prison_deref_kill(pr, &freeprison);
3480 }
3481 }
3482 if (flags & PD_DEUREF) {
3483 /* Drop a user reference. */
3484 KASSERT(refcount_load(&pr->pr_uref) > 0,
3485 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3486 pr->pr_id));
3487 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3488 if (!(flags & PD_DEREF)) {
3489 prison_hold(pr);
3490 flags |= PD_DEREF;
3491 }
3492 flags = prison_lock_xlock(pr, flags);
3493 if (refcount_release(&pr->pr_uref) &&
3494 pr->pr_state == PRISON_STATE_ALIVE) {
3495 /*
3496 * When the last user references goes,
3497 * this becomes a dying prison.
3498 */
3499 KASSERT(
3500 refcount_load(&prison0.pr_uref) > 0,
3501 ("prison0 pr_uref=0"));
3502 pr->pr_state = PRISON_STATE_DYING;
3503 prison_cleanup_locked(pr);
3504 mtx_unlock(&pr->pr_mtx);
3505 flags &= ~PD_LOCKED;
3506 prison_cleanup_unlocked(pr);
3507 }
3508 }
3509 }
3510 if (flags & PD_KILL) {
3511 /*
3512 * Any remaining user references are probably processes
3513 * that need to be killed, either in this prison or its
3514 * descendants.
3515 */
3516 if (refcount_load(&pr->pr_uref) > 0)
3517 killpr = pr;
3518 /* Make sure the parent prison doesn't get killed. */
3519 flags &= ~PD_KILL;
3520 }
3521 if (flags & PD_DEREF) {
3522 /* Drop a reference. */
3523 KASSERT(refcount_load(&pr->pr_ref) > 0,
3524 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3525 pr->pr_id));
3526 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3527 flags = prison_lock_xlock(pr, flags);
3528 if (refcount_release(&pr->pr_ref)) {
3529 /*
3530 * When the last reference goes,
3531 * unlink the prison and set it aside.
3532 */
3533 KASSERT(
3534 refcount_load(&pr->pr_uref) == 0,
3535 ("prison_deref: last ref, "
3536 "but still has %d urefs (jid=%d)",
3537 pr->pr_uref, pr->pr_id));
3538 KASSERT(
3539 refcount_load(&prison0.pr_ref) != 0,
3540 ("prison0 pr_ref=0"));
3541 pr->pr_state = PRISON_STATE_INVALID;
3542 TAILQ_REMOVE(&allprison, pr, pr_list);
3543 LIST_REMOVE(pr, pr_sibling);
3544 TAILQ_INSERT_TAIL(&freeprison, pr,
3545 pr_list);
3546 for (ppr = pr->pr_parent;
3547 ppr != NULL;
3548 ppr = ppr->pr_parent)
3549 ppr->pr_childcount--;
3550 /*
3551 * Removing a prison frees references
3552 * from its parent.
3553 */
3554 ppr = pr->pr_parent;
3555 pr->pr_parent = NULL;
3556 mtx_unlock(&pr->pr_mtx);
3557
3558 pr = ppr;
3559 flags &= ~PD_LOCKED;
3560 flags |= PD_DEREF | PD_DEUREF;
3561 continue;
3562 }
3563 }
3564 }
3565 break;
3566 }
3567
3568 /* Release all the prison locks. */
3569 if (flags & PD_LOCKED)
3570 mtx_unlock(&pr->pr_mtx);
3571 if (flags & PD_LIST_SLOCKED)
3572 sx_sunlock(&allprison_lock);
3573 else if (flags & PD_LIST_XLOCKED)
3574 sx_xunlock(&allprison_lock);
3575
3576 /* Kill any processes attached to a killed prison. */
3577 if (killpr != NULL)
3578 prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3579
3580 /*
3581 * Finish removing any unreferenced prisons, which couldn't happen
3582 * while allprison_lock was held (to avoid a LOR on vrele).
3583 */
3584 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3585 #ifdef VIMAGE
3586 if (rpr->pr_flags & PR_VNET)
3587 vnet_destroy(rpr->pr_vnet);
3588 #endif
3589 if (rpr->pr_root != NULL)
3590 vrele(rpr->pr_root);
3591 mtx_destroy(&rpr->pr_mtx);
3592 #ifdef INET
3593 prison_ip_free(rpr->pr_addrs[PR_INET]);
3594 #endif
3595 #ifdef INET6
3596 prison_ip_free(rpr->pr_addrs[PR_INET6]);
3597 #endif
3598 if (rpr->pr_cpuset != NULL)
3599 cpuset_rel(rpr->pr_cpuset);
3600 osd_jail_exit(rpr);
3601 #ifdef RACCT
3602 if (racct_enable)
3603 prison_racct_detach(rpr);
3604 #endif
3605 TAILQ_REMOVE(&freeprison, rpr, pr_list);
3606 free(rpr, M_PRISON);
3607 }
3608 }
3609
3610 /*
3611 * Kill the prison and its descendants. Mark them as dying, clear the
3612 * persist flag, and call module remove methods.
3613 */
3614 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)3615 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3616 {
3617 struct prison *cpr, *ppr, *rpr;
3618 bool descend;
3619
3620 /*
3621 * Unlike the descendants, the target prison can be killed
3622 * even if it is currently dying. This is useful for failed
3623 * creation in jail_set(2).
3624 */
3625 KASSERT(refcount_load(&pr->pr_ref) > 0,
3626 ("Trying to kill dead prison %p (jid=%d).",
3627 pr, pr->pr_id));
3628 refcount_acquire(&pr->pr_uref);
3629 pr->pr_state = PRISON_STATE_DYING;
3630 mtx_unlock(&pr->pr_mtx);
3631
3632 rpr = NULL;
3633 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3634 if (descend) {
3635 if (!prison_isalive(cpr)) {
3636 descend = false;
3637 continue;
3638 }
3639 prison_hold(cpr);
3640 prison_proc_hold(cpr);
3641 mtx_lock(&cpr->pr_mtx);
3642 cpr->pr_state = PRISON_STATE_DYING;
3643 cpr->pr_flags |= PR_REMOVE;
3644 mtx_unlock(&cpr->pr_mtx);
3645 continue;
3646 }
3647 if (!(cpr->pr_flags & PR_REMOVE))
3648 continue;
3649 prison_cleanup_unlocked(cpr);
3650 mtx_lock(&cpr->pr_mtx);
3651 prison_cleanup_locked(cpr);
3652 cpr->pr_flags &= ~PR_REMOVE;
3653 if (cpr->pr_flags & PR_PERSIST) {
3654 cpr->pr_flags &= ~PR_PERSIST;
3655 prison_proc_free_not_last(cpr);
3656 prison_free_not_last(cpr);
3657 }
3658 (void)refcount_release(&cpr->pr_uref);
3659 if (refcount_release(&cpr->pr_ref)) {
3660 /*
3661 * When the last reference goes, unlink the prison
3662 * and set it aside for prison_deref() to handle.
3663 * Delay unlinking the sibling list to keep the loop
3664 * safe.
3665 */
3666 if (rpr != NULL)
3667 LIST_REMOVE(rpr, pr_sibling);
3668 rpr = cpr;
3669 rpr->pr_state = PRISON_STATE_INVALID;
3670 TAILQ_REMOVE(&allprison, rpr, pr_list);
3671 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3672 /*
3673 * Removing a prison frees references from its parent.
3674 */
3675 ppr = rpr->pr_parent;
3676 prison_proc_free_not_last(ppr);
3677 prison_free_not_last(ppr);
3678 for (; ppr != NULL; ppr = ppr->pr_parent)
3679 ppr->pr_childcount--;
3680 }
3681 mtx_unlock(&cpr->pr_mtx);
3682 }
3683 if (rpr != NULL)
3684 LIST_REMOVE(rpr, pr_sibling);
3685
3686 prison_cleanup_unlocked(pr);
3687 mtx_lock(&pr->pr_mtx);
3688 prison_cleanup_locked(pr);
3689 if (pr->pr_flags & PR_PERSIST) {
3690 pr->pr_flags &= ~PR_PERSIST;
3691 prison_proc_free_not_last(pr);
3692 prison_free_not_last(pr);
3693 }
3694 (void)refcount_release(&pr->pr_uref);
3695 }
3696
3697 /*
3698 * Given the current locking state in the flags, make sure allprison_lock
3699 * is held exclusive, and the prison is locked. Return flags indicating
3700 * the new state.
3701 */
3702 static int
prison_lock_xlock(struct prison * pr,int flags)3703 prison_lock_xlock(struct prison *pr, int flags)
3704 {
3705
3706 if (!(flags & PD_LIST_XLOCKED)) {
3707 /*
3708 * Get allprison_lock, which may be an upgrade,
3709 * and may require unlocking the prison.
3710 */
3711 if (flags & PD_LOCKED) {
3712 mtx_unlock(&pr->pr_mtx);
3713 flags &= ~PD_LOCKED;
3714 }
3715 if (flags & PD_LIST_SLOCKED) {
3716 if (!sx_try_upgrade(&allprison_lock)) {
3717 sx_sunlock(&allprison_lock);
3718 sx_xlock(&allprison_lock);
3719 }
3720 flags &= ~PD_LIST_SLOCKED;
3721 } else
3722 sx_xlock(&allprison_lock);
3723 flags |= PD_LIST_XLOCKED;
3724 }
3725 if (!(flags & PD_LOCKED)) {
3726 /* Lock the prison mutex. */
3727 mtx_lock(&pr->pr_mtx);
3728 flags |= PD_LOCKED;
3729 }
3730 return flags;
3731 }
3732
3733 /*
3734 * Release a prison's resources when it starts dying (when the last user
3735 * reference is dropped, or when it is killed). Two functions are called,
3736 * for work that requires a locked prison or an unlocked one.
3737 */
3738 static void
prison_cleanup_locked(struct prison * pr)3739 prison_cleanup_locked(struct prison *pr)
3740 {
3741 sx_assert(&allprison_lock, SA_XLOCKED);
3742 mtx_assert(&pr->pr_mtx, MA_OWNED);
3743 prison_knote(pr, NOTE_JAIL_REMOVE);
3744 knlist_detach(pr->pr_klist);
3745 jaildesc_prison_cleanup(pr);
3746 pr->pr_klist = NULL;
3747 }
3748
3749 static void
prison_cleanup_unlocked(struct prison * pr)3750 prison_cleanup_unlocked(struct prison *pr)
3751 {
3752 sx_assert(&allprison_lock, SA_XLOCKED);
3753 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3754 vfs_exjail_delete(pr);
3755 shm_remove_prison(pr);
3756 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3757 }
3758
3759 /*
3760 * Set or clear a permission bit in the pr_allow field, passing restrictions
3761 * (cleared permission) down to child jails.
3762 */
3763 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)3764 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3765 {
3766 struct prison *pr;
3767
3768 pr = cred->cr_prison;
3769 sx_slock(&allprison_lock);
3770 mtx_lock(&pr->pr_mtx);
3771 prison_set_allow_locked(pr, flag, enable);
3772 mtx_unlock(&pr->pr_mtx);
3773 sx_sunlock(&allprison_lock);
3774 }
3775
3776 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)3777 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3778 {
3779 struct prison *cpr;
3780 int descend;
3781
3782 if (enable != 0)
3783 pr->pr_allow |= flag;
3784 else {
3785 pr->pr_allow &= ~flag;
3786 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3787 cpr->pr_allow &= ~flag;
3788 }
3789 }
3790
3791 /*
3792 * Check if a jail supports the given address family.
3793 *
3794 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3795 * if not.
3796 */
3797 int
prison_check_af(struct ucred * cred,int af)3798 prison_check_af(struct ucred *cred, int af)
3799 {
3800 struct prison *pr;
3801 int error;
3802
3803 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3804
3805 pr = cred->cr_prison;
3806 #ifdef VIMAGE
3807 /* Prisons with their own network stack are not limited. */
3808 if (prison_owns_vnet(pr))
3809 return (0);
3810 #endif
3811
3812 error = 0;
3813 switch (af)
3814 {
3815 #ifdef INET
3816 case AF_INET:
3817 if (pr->pr_flags & PR_IP4)
3818 {
3819 mtx_lock(&pr->pr_mtx);
3820 if ((pr->pr_flags & PR_IP4) &&
3821 pr->pr_addrs[PR_INET] == NULL)
3822 error = EAFNOSUPPORT;
3823 mtx_unlock(&pr->pr_mtx);
3824 }
3825 break;
3826 #endif
3827 #ifdef INET6
3828 case AF_INET6:
3829 if (pr->pr_flags & PR_IP6)
3830 {
3831 mtx_lock(&pr->pr_mtx);
3832 if ((pr->pr_flags & PR_IP6) &&
3833 pr->pr_addrs[PR_INET6] == NULL)
3834 error = EAFNOSUPPORT;
3835 mtx_unlock(&pr->pr_mtx);
3836 }
3837 break;
3838 #endif
3839 case AF_LOCAL:
3840 case AF_ROUTE:
3841 case AF_NETLINK:
3842 break;
3843 default:
3844 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3845 error = EAFNOSUPPORT;
3846 }
3847 return (error);
3848 }
3849
3850 /*
3851 * Check if given address belongs to the jail referenced by cred (wrapper to
3852 * prison_check_ip[46]).
3853 *
3854 * Returns 0 if jail doesn't restrict the address family or if address belongs
3855 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3856 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3857 */
3858 int
prison_if(struct ucred * cred,const struct sockaddr * sa)3859 prison_if(struct ucred *cred, const struct sockaddr *sa)
3860 {
3861 #ifdef INET
3862 const struct sockaddr_in *sai;
3863 #endif
3864 #ifdef INET6
3865 const struct sockaddr_in6 *sai6;
3866 #endif
3867 int error;
3868
3869 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3870 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3871
3872 #ifdef VIMAGE
3873 if (prison_owns_vnet(cred->cr_prison))
3874 return (0);
3875 #endif
3876
3877 error = 0;
3878 switch (sa->sa_family)
3879 {
3880 #ifdef INET
3881 case AF_INET:
3882 sai = (const struct sockaddr_in *)sa;
3883 error = prison_check_ip4(cred, &sai->sin_addr);
3884 break;
3885 #endif
3886 #ifdef INET6
3887 case AF_INET6:
3888 sai6 = (const struct sockaddr_in6 *)sa;
3889 error = prison_check_ip6(cred, &sai6->sin6_addr);
3890 break;
3891 #endif
3892 default:
3893 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3894 error = EAFNOSUPPORT;
3895 }
3896 return (error);
3897 }
3898
3899 /*
3900 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3901 */
3902 int
prison_check(struct ucred * cred1,struct ucred * cred2)3903 prison_check(struct ucred *cred1, struct ucred *cred2)
3904 {
3905
3906 return ((cred1->cr_prison == cred2->cr_prison ||
3907 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3908 }
3909
3910 /*
3911 * For mountd/nfsd to run within a prison, it must be:
3912 * - A vnet prison.
3913 * - PR_ALLOW_NFSD must be set on it.
3914 * - The root directory (pr_root) of the prison must be
3915 * a file system mount point, so the mountd can hang
3916 * export information on it.
3917 * - The prison's enforce_statfs cannot be 0, so that
3918 * mountd(8) can do exports.
3919 */
3920 bool
prison_check_nfsd(struct ucred * cred)3921 prison_check_nfsd(struct ucred *cred)
3922 {
3923
3924 if (jailed_without_vnet(cred))
3925 return (false);
3926 if (!prison_allow(cred, PR_ALLOW_NFSD))
3927 return (false);
3928 if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
3929 return (false);
3930 if (cred->cr_prison->pr_enforce_statfs == 0)
3931 return (false);
3932 return (true);
3933 }
3934
3935 /*
3936 * Return true if p2 is a child of p1, otherwise false.
3937 */
3938 bool
prison_ischild(struct prison * pr1,struct prison * pr2)3939 prison_ischild(struct prison *pr1, struct prison *pr2)
3940 {
3941
3942 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3943 if (pr1 == pr2)
3944 return (true);
3945 return (false);
3946 }
3947
3948 /*
3949 * Return true if the prison is currently alive. A prison is alive if it
3950 * holds user references and it isn't being removed.
3951 */
3952 bool
prison_isalive(const struct prison * pr)3953 prison_isalive(const struct prison *pr)
3954 {
3955
3956 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3957 return (false);
3958 return (true);
3959 }
3960
3961 /*
3962 * Return true if the prison is currently valid. A prison is valid if it has
3963 * been fully created, and is not being destroyed. Note that dying prisons
3964 * are still considered valid. Invalid prisons won't be found under normal
3965 * circumstances, as they're only put in that state by functions that have
3966 * an exclusive hold on allprison_lock.
3967 */
3968 bool
prison_isvalid(struct prison * pr)3969 prison_isvalid(struct prison *pr)
3970 {
3971
3972 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3973 return (false);
3974 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3975 return (false);
3976 return (true);
3977 }
3978
3979 /*
3980 * Return true if the passed credential is in a jail and that jail does not
3981 * have its own virtual network stack, otherwise false.
3982 */
3983 bool
jailed_without_vnet(struct ucred * cred)3984 jailed_without_vnet(struct ucred *cred)
3985 {
3986
3987 if (!jailed(cred))
3988 return (false);
3989 #ifdef VIMAGE
3990 if (prison_owns_vnet(cred->cr_prison))
3991 return (false);
3992 #endif
3993
3994 return (true);
3995 }
3996
3997 /*
3998 * Return the correct hostname (domainname, et al) for the passed credential.
3999 */
4000 void
getcredhostname(struct ucred * cred,char * buf,size_t size)4001 getcredhostname(struct ucred *cred, char *buf, size_t size)
4002 {
4003 struct prison *pr;
4004
4005 /*
4006 * A NULL credential can be used to shortcut to the physical
4007 * system's hostname.
4008 */
4009 pr = (cred != NULL) ? cred->cr_prison : &prison0;
4010 mtx_lock(&pr->pr_mtx);
4011 strlcpy(buf, pr->pr_hostname, size);
4012 mtx_unlock(&pr->pr_mtx);
4013 }
4014
4015 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)4016 getcreddomainname(struct ucred *cred, char *buf, size_t size)
4017 {
4018
4019 mtx_lock(&cred->cr_prison->pr_mtx);
4020 strlcpy(buf, cred->cr_prison->pr_domainname, size);
4021 mtx_unlock(&cred->cr_prison->pr_mtx);
4022 }
4023
4024 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)4025 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
4026 {
4027
4028 mtx_lock(&cred->cr_prison->pr_mtx);
4029 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
4030 mtx_unlock(&cred->cr_prison->pr_mtx);
4031 }
4032
4033 void
getcredhostid(struct ucred * cred,unsigned long * hostid)4034 getcredhostid(struct ucred *cred, unsigned long *hostid)
4035 {
4036
4037 mtx_lock(&cred->cr_prison->pr_mtx);
4038 *hostid = cred->cr_prison->pr_hostid;
4039 mtx_unlock(&cred->cr_prison->pr_mtx);
4040 }
4041
4042 void
getjailname(struct ucred * cred,char * name,size_t len)4043 getjailname(struct ucred *cred, char *name, size_t len)
4044 {
4045
4046 mtx_lock(&cred->cr_prison->pr_mtx);
4047 strlcpy(name, cred->cr_prison->pr_name, len);
4048 mtx_unlock(&cred->cr_prison->pr_mtx);
4049 }
4050
4051 #ifdef VIMAGE
4052 /*
4053 * Determine whether the prison owns its VNET.
4054 */
4055 bool
prison_owns_vnet(struct prison * pr)4056 prison_owns_vnet(struct prison *pr)
4057 {
4058
4059 /*
4060 * vnets cannot be added/removed after jail creation,
4061 * so no need to lock here.
4062 */
4063 return ((pr->pr_flags & PR_VNET) != 0);
4064 }
4065 #endif
4066
4067 /*
4068 * Determine whether the subject represented by cred can "see"
4069 * status of a mount point.
4070 * Returns: 0 for permitted, ENOENT otherwise.
4071 * XXX: This function should be called cr_canseemount() and should be
4072 * placed in kern_prot.c.
4073 */
4074 int
prison_canseemount(struct ucred * cred,struct mount * mp)4075 prison_canseemount(struct ucred *cred, struct mount *mp)
4076 {
4077 struct prison *pr;
4078 struct statfs *sp;
4079 size_t len;
4080
4081 pr = cred->cr_prison;
4082 if (pr->pr_enforce_statfs == 0)
4083 return (0);
4084 if (pr->pr_root->v_mount == mp)
4085 return (0);
4086 if (pr->pr_enforce_statfs == 2)
4087 return (ENOENT);
4088 /*
4089 * If jail's chroot directory is set to "/" we should be able to see
4090 * all mount-points from inside a jail.
4091 * This is ugly check, but this is the only situation when jail's
4092 * directory ends with '/'.
4093 */
4094 if (strcmp(pr->pr_path, "/") == 0)
4095 return (0);
4096 len = strlen(pr->pr_path);
4097 sp = &mp->mnt_stat;
4098 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
4099 return (ENOENT);
4100 /*
4101 * Be sure that we don't have situation where jail's root directory
4102 * is "/some/path" and mount point is "/some/pathpath".
4103 */
4104 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
4105 return (ENOENT);
4106 return (0);
4107 }
4108
4109 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)4110 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
4111 {
4112 char jpath[MAXPATHLEN];
4113 struct prison *pr;
4114 size_t len;
4115
4116 pr = cred->cr_prison;
4117 if (pr->pr_enforce_statfs == 0)
4118 return;
4119 if (prison_canseemount(cred, mp) != 0) {
4120 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4121 strlcpy(sp->f_mntonname, "[restricted]",
4122 sizeof(sp->f_mntonname));
4123 return;
4124 }
4125 if (pr->pr_root->v_mount == mp) {
4126 /*
4127 * Clear current buffer data, so we are sure nothing from
4128 * the valid path left there.
4129 */
4130 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4131 *sp->f_mntonname = '/';
4132 return;
4133 }
4134 /*
4135 * If jail's chroot directory is set to "/" we should be able to see
4136 * all mount-points from inside a jail.
4137 */
4138 if (strcmp(pr->pr_path, "/") == 0)
4139 return;
4140 len = strlen(pr->pr_path);
4141 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
4142 /*
4143 * Clear current buffer data, so we are sure nothing from
4144 * the valid path left there.
4145 */
4146 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4147 if (*jpath == '\0') {
4148 /* Should never happen. */
4149 *sp->f_mntonname = '/';
4150 } else {
4151 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
4152 }
4153 }
4154
4155 /*
4156 * Check with permission for a specific privilege is granted within jail. We
4157 * have a specific list of accepted privileges; the rest are denied.
4158 */
4159 int
prison_priv_check(struct ucred * cred,int priv)4160 prison_priv_check(struct ucred *cred, int priv)
4161 {
4162 struct prison *pr;
4163 int error;
4164
4165 /*
4166 * Some policies have custom handlers. This routine should not be
4167 * called for them. See priv_check_cred().
4168 */
4169 switch (priv) {
4170 case PRIV_VFS_LOOKUP:
4171 case PRIV_VFS_GENERATION:
4172 KASSERT(0, ("prison_priv_check instead of a custom handler "
4173 "called for %d\n", priv));
4174 }
4175
4176 if (!jailed(cred))
4177 return (0);
4178
4179 #ifdef VIMAGE
4180 /*
4181 * Privileges specific to prisons with a virtual network stack.
4182 * There might be a duplicate entry here in case the privilege
4183 * is only granted conditionally in the legacy jail case.
4184 */
4185 switch (priv) {
4186 /*
4187 * NFS-specific privileges.
4188 */
4189 case PRIV_NFS_DAEMON:
4190 case PRIV_VFS_GETFH:
4191 case PRIV_VFS_MOUNT_EXPORTED:
4192 if (!prison_check_nfsd(cred))
4193 return (EPERM);
4194 #ifdef notyet
4195 case PRIV_NFS_LOCKD:
4196 #endif
4197 /*
4198 * Network stack privileges.
4199 */
4200 case PRIV_NET_BRIDGE:
4201 case PRIV_NET_GRE:
4202 case PRIV_NET_BPF:
4203 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
4204 case PRIV_NET_ROUTE:
4205 case PRIV_NET_TAP:
4206 case PRIV_NET_SETIFMTU:
4207 case PRIV_NET_SETIFFLAGS:
4208 case PRIV_NET_SETIFCAP:
4209 case PRIV_NET_SETIFDESCR:
4210 case PRIV_NET_SETIFNAME :
4211 case PRIV_NET_SETIFMETRIC:
4212 case PRIV_NET_SETIFPHYS:
4213 case PRIV_NET_SETIFMAC:
4214 case PRIV_NET_SETLANPCP:
4215 case PRIV_NET_ADDMULTI:
4216 case PRIV_NET_DELMULTI:
4217 case PRIV_NET_HWIOCTL:
4218 case PRIV_NET_SETLLADDR:
4219 case PRIV_NET_ADDIFGROUP:
4220 case PRIV_NET_DELIFGROUP:
4221 case PRIV_NET_IFCREATE:
4222 case PRIV_NET_IFDESTROY:
4223 case PRIV_NET_ADDIFADDR:
4224 case PRIV_NET_DELIFADDR:
4225 case PRIV_NET_LAGG:
4226 case PRIV_NET_GIF:
4227 case PRIV_NET_SETIFVNET:
4228 case PRIV_NET_SETIFFIB:
4229 case PRIV_NET_OVPN:
4230 case PRIV_NET_ME:
4231 case PRIV_NET_WG:
4232
4233 /*
4234 * 802.11-related privileges.
4235 */
4236 case PRIV_NET80211_VAP_GETKEY:
4237 case PRIV_NET80211_VAP_MANAGE:
4238
4239 #ifdef notyet
4240 /*
4241 * ATM privileges.
4242 */
4243 case PRIV_NETATM_CFG:
4244 case PRIV_NETATM_ADD:
4245 case PRIV_NETATM_DEL:
4246 case PRIV_NETATM_SET:
4247
4248 /*
4249 * Bluetooth privileges.
4250 */
4251 case PRIV_NETBLUETOOTH_RAW:
4252 #endif
4253
4254 /*
4255 * Netgraph and netgraph module privileges.
4256 */
4257 case PRIV_NETGRAPH_CONTROL:
4258 #ifdef notyet
4259 case PRIV_NETGRAPH_TTY:
4260 #endif
4261
4262 /*
4263 * IPv4 and IPv6 privileges.
4264 */
4265 case PRIV_NETINET_IPFW:
4266 case PRIV_NETINET_DIVERT:
4267 case PRIV_NETINET_PF:
4268 case PRIV_NETINET_DUMMYNET:
4269 case PRIV_NETINET_CARP:
4270 case PRIV_NETINET_MROUTE:
4271 case PRIV_NETINET_RAW:
4272 case PRIV_NETINET_ADDRCTRL6:
4273 case PRIV_NETINET_ND6:
4274 case PRIV_NETINET_SCOPE6:
4275 case PRIV_NETINET_ALIFETIME6:
4276 case PRIV_NETINET_IPSEC:
4277 case PRIV_NETINET_BINDANY:
4278
4279 #ifdef notyet
4280 /*
4281 * NCP privileges.
4282 */
4283 case PRIV_NETNCP:
4284
4285 /*
4286 * SMB privileges.
4287 */
4288 case PRIV_NETSMB:
4289 #endif
4290
4291 /*
4292 * No default: or deny here.
4293 * In case of no permit fall through to next switch().
4294 */
4295 if (cred->cr_prison->pr_flags & PR_VNET)
4296 return (0);
4297 }
4298 #endif /* VIMAGE */
4299
4300 switch (priv) {
4301 /*
4302 * Allow ktrace privileges for root in jail.
4303 */
4304 case PRIV_KTRACE:
4305
4306 /*
4307 * Allow jailed processes to configure audit identity and
4308 * submit audit records (login, etc). In the future we may
4309 * want to further refine the relationship between audit and
4310 * jail.
4311 */
4312 case PRIV_AUDIT_GETAUDIT:
4313 case PRIV_AUDIT_SETAUDIT:
4314 if (cred->cr_prison->pr_allow & PR_ALLOW_SETAUDIT)
4315 return (0);
4316 else
4317 return (EPERM);
4318 #if 0
4319 case PRIV_AUDIT_SUBMIT:
4320 #endif
4321
4322 /*
4323 * Allow jailed processes to manipulate process UNIX
4324 * credentials in any way they see fit.
4325 */
4326 case PRIV_CRED_SETCRED:
4327 case PRIV_CRED_SETUID:
4328 case PRIV_CRED_SETEUID:
4329 case PRIV_CRED_SETGID:
4330 case PRIV_CRED_SETEGID:
4331 case PRIV_CRED_SETGROUPS:
4332 case PRIV_CRED_SETREUID:
4333 case PRIV_CRED_SETREGID:
4334 case PRIV_CRED_SETRESUID:
4335 case PRIV_CRED_SETRESGID:
4336
4337 /*
4338 * Jail implements visibility constraints already, so allow
4339 * jailed root to override uid/gid-based constraints.
4340 */
4341 case PRIV_SEEOTHERGIDS:
4342 case PRIV_SEEOTHERUIDS:
4343 case PRIV_SEEJAILPROC:
4344
4345 /*
4346 * Jail implements inter-process debugging limits already, so
4347 * allow jailed root various debugging privileges.
4348 */
4349 case PRIV_DEBUG_DIFFCRED:
4350 case PRIV_DEBUG_SUGID:
4351 case PRIV_DEBUG_UNPRIV:
4352 case PRIV_DEBUG_DIFFJAIL:
4353
4354 /*
4355 * Allow jail to set various resource limits and login
4356 * properties, and for now, exceed process resource limits.
4357 */
4358 case PRIV_PROC_LIMIT:
4359 case PRIV_PROC_SETLOGIN:
4360 case PRIV_PROC_SETRLIMIT:
4361
4362 /*
4363 * Debuggers should work in jails.
4364 */
4365 case PRIV_PROC_MEM_WRITE:
4366
4367 /*
4368 * System V and POSIX IPC privileges are granted in jail.
4369 */
4370 case PRIV_IPC_READ:
4371 case PRIV_IPC_WRITE:
4372 case PRIV_IPC_ADMIN:
4373 case PRIV_IPC_MSGSIZE:
4374 case PRIV_MQ_ADMIN:
4375
4376 /*
4377 * Jail operations within a jail work on child jails.
4378 */
4379 case PRIV_JAIL_ATTACH:
4380 case PRIV_JAIL_SET:
4381 case PRIV_JAIL_REMOVE:
4382
4383 /*
4384 * Jail implements its own inter-process limits, so allow
4385 * root processes in jail to change scheduling on other
4386 * processes in the same jail. Likewise for signalling.
4387 */
4388 case PRIV_SCHED_DIFFCRED:
4389 case PRIV_SCHED_CPUSET:
4390 case PRIV_SCHED_DIFFJAIL:
4391 case PRIV_SIGNAL_DIFFCRED:
4392 case PRIV_SIGNAL_SUGID:
4393 case PRIV_SIGNAL_DIFFJAIL:
4394
4395 /*
4396 * Allow jailed processes to write to sysctls marked as jail
4397 * writable.
4398 */
4399 case PRIV_SYSCTL_WRITEJAIL:
4400
4401 /*
4402 * Allow root in jail to manage a variety of quota
4403 * properties. These should likely be conditional on a
4404 * configuration option.
4405 */
4406 case PRIV_VFS_GETQUOTA:
4407 case PRIV_VFS_SETQUOTA:
4408
4409 /*
4410 * Since Jail relies on chroot() to implement file system
4411 * protections, grant many VFS privileges to root in jail.
4412 * Be careful to exclude mount-related and NFS-related
4413 * privileges.
4414 */
4415 case PRIV_VFS_READ:
4416 case PRIV_VFS_WRITE:
4417 case PRIV_VFS_ADMIN:
4418 case PRIV_VFS_EXEC:
4419 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4420 case PRIV_VFS_CHFLAGS_DEV:
4421 case PRIV_VFS_CHOWN:
4422 case PRIV_VFS_CHROOT:
4423 case PRIV_VFS_RETAINSUGID:
4424 case PRIV_VFS_FCHROOT:
4425 case PRIV_VFS_LINK:
4426 case PRIV_VFS_SETGID:
4427 case PRIV_VFS_STAT:
4428 case PRIV_VFS_STICKYFILE:
4429
4430 /*
4431 * As in the non-jail case, non-root users are expected to be
4432 * able to read kernel/physical memory (provided /dev/[k]mem
4433 * exists in the jail and they have permission to access it).
4434 */
4435 case PRIV_KMEM_READ:
4436 return (0);
4437
4438 /*
4439 * Depending on the global setting, allow privilege of
4440 * setting system flags.
4441 */
4442 case PRIV_VFS_SYSFLAGS:
4443 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4444 return (0);
4445 else
4446 return (EPERM);
4447
4448 /*
4449 * Depending on the global setting, allow privilege of
4450 * mounting/unmounting file systems.
4451 */
4452 case PRIV_VFS_MOUNT:
4453 case PRIV_VFS_UNMOUNT:
4454 case PRIV_VFS_MOUNT_NONUSER:
4455 case PRIV_VFS_MOUNT_OWNER:
4456 pr = cred->cr_prison;
4457 prison_lock(pr);
4458 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4459 error = 0;
4460 else
4461 error = EPERM;
4462 prison_unlock(pr);
4463 return (error);
4464
4465 /*
4466 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4467 * policy. priv_check_cred will not specifically allow it, and
4468 * we may want a MAC policy to allow it.
4469 */
4470 case PRIV_VFS_READ_DIR:
4471 return (0);
4472
4473 /*
4474 * Conditionally allow privileged process in the jail to
4475 * manipulate filesystem extended attributes in the system
4476 * namespace.
4477 */
4478 case PRIV_VFS_EXTATTR_SYSTEM:
4479 if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4480 return (0);
4481 else
4482 return (EPERM);
4483
4484 /*
4485 * Conditionnaly allow locking (unlocking) physical pages
4486 * in memory.
4487 */
4488 case PRIV_VM_MLOCK:
4489 case PRIV_VM_MUNLOCK:
4490 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4491 return (0);
4492 else
4493 return (EPERM);
4494
4495 /*
4496 * Conditionally allow jailed root to bind reserved ports.
4497 */
4498 case PRIV_NETINET_RESERVEDPORT:
4499 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4500 return (0);
4501 else
4502 return (EPERM);
4503
4504 /*
4505 * Allow jailed root to reuse in-use ports.
4506 */
4507 case PRIV_NETINET_REUSEPORT:
4508 return (0);
4509
4510 /*
4511 * Allow jailed root to set certain IPv4/6 (option) headers.
4512 */
4513 case PRIV_NETINET_SETHDROPTS:
4514 return (0);
4515
4516 /*
4517 * Conditionally allow creating raw sockets in jail.
4518 */
4519 case PRIV_NETINET_RAW:
4520 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4521 return (0);
4522 else
4523 return (EPERM);
4524
4525 /*
4526 * Since jail implements its own visibility limits on netstat
4527 * sysctls, allow getcred. This allows identd to work in
4528 * jail.
4529 */
4530 case PRIV_NETINET_GETCRED:
4531 return (0);
4532
4533 /*
4534 * Allow jailed root to set loginclass.
4535 */
4536 case PRIV_PROC_SETLOGINCLASS:
4537 return (0);
4538
4539 /*
4540 * Do not allow a process inside a jail to read the kernel
4541 * message buffer unless explicitly permitted.
4542 */
4543 case PRIV_MSGBUF:
4544 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4545 return (0);
4546 return (EPERM);
4547
4548 /*
4549 * Conditionally allow privileged process in the jail adjust
4550 * machine time.
4551 */
4552 case PRIV_ADJTIME:
4553 case PRIV_NTP_ADJTIME:
4554 if (cred->cr_prison->pr_allow &
4555 (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4556 return (0);
4557 }
4558 return (EPERM);
4559
4560 /*
4561 * Conditionally allow privileged process in the jail set
4562 * machine time.
4563 */
4564 case PRIV_SETTIMEOFDAY:
4565 case PRIV_CLOCK_SETTIME:
4566 if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4567 return (0);
4568 else
4569 return (EPERM);
4570
4571 /*
4572 * Conditionally allow privileged process in the jail to modify
4573 * the routing table.
4574 */
4575 case PRIV_NET_ROUTE:
4576 if (cred->cr_prison->pr_allow & PR_ALLOW_ROUTING)
4577 return (0);
4578 else
4579 return (EPERM);
4580
4581 default:
4582 /*
4583 * In all remaining cases, deny the privilege request. This
4584 * includes almost all network privileges, many system
4585 * configuration privileges.
4586 */
4587 return (EPERM);
4588 }
4589 }
4590
4591 /*
4592 * Return the part of pr2's name that is relative to pr1, or the whole name
4593 * if it does not directly follow.
4594 */
4595
4596 char *
prison_name(struct prison * pr1,struct prison * pr2)4597 prison_name(struct prison *pr1, struct prison *pr2)
4598 {
4599 char *name;
4600
4601 /* Jails see themselves as "0" (if they see themselves at all). */
4602 if (pr1 == pr2)
4603 return "0";
4604 name = pr2->pr_name;
4605 if (prison_ischild(pr1, pr2)) {
4606 /*
4607 * pr1 isn't locked (and allprison_lock may not be either)
4608 * so its length can't be counted on. But the number of dots
4609 * can be counted on - and counted.
4610 */
4611 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4612 name = strchr(name, '.') + 1;
4613 }
4614 return (name);
4615 }
4616
4617 /*
4618 * Return the part of pr2's path that is relative to pr1, or the whole path
4619 * if it does not directly follow.
4620 */
4621 static char *
prison_path(struct prison * pr1,struct prison * pr2)4622 prison_path(struct prison *pr1, struct prison *pr2)
4623 {
4624 char *path1, *path2;
4625 int len1;
4626
4627 path1 = pr1->pr_path;
4628 path2 = pr2->pr_path;
4629 if (!strcmp(path1, "/"))
4630 return (path2);
4631 len1 = strlen(path1);
4632 if (strncmp(path1, path2, len1))
4633 return (path2);
4634 if (path2[len1] == '\0')
4635 return "/";
4636 if (path2[len1] == '/')
4637 return (path2 + len1);
4638 return (path2);
4639 }
4640
4641 /*
4642 * Jail-related sysctls.
4643 */
4644 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4645 "Jails");
4646
4647 #if defined(INET) || defined(INET6)
4648 /*
4649 * Copy address array to memory that would be then SYSCTL_OUT-ed.
4650 * sysctl_jail_list() helper.
4651 */
4652 static void
prison_ip_copyout(struct prison * pr,const pr_family_t af,void ** out,int * len)4653 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4654 {
4655 const struct prison_ip *pip;
4656 const size_t size = pr_families[af].size;
4657
4658 again:
4659 mtx_assert(&pr->pr_mtx, MA_OWNED);
4660 if ((pip = pr->pr_addrs[af]) != NULL) {
4661 if (*len < pip->ips) {
4662 *len = pip->ips;
4663 mtx_unlock(&pr->pr_mtx);
4664 *out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4665 mtx_lock(&pr->pr_mtx);
4666 goto again;
4667 }
4668 bcopy(pip->pr_ip, *out, pip->ips * size);
4669 }
4670 }
4671 #endif
4672
4673 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4674 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4675 {
4676 struct xprison *xp;
4677 struct prison *pr, *cpr;
4678 #ifdef INET
4679 struct in_addr *ip4 = NULL;
4680 int ip4s = 0;
4681 #endif
4682 #ifdef INET6
4683 struct in6_addr *ip6 = NULL;
4684 int ip6s = 0;
4685 #endif
4686 int descend, error;
4687
4688 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4689 pr = req->td->td_ucred->cr_prison;
4690 error = 0;
4691 sx_slock(&allprison_lock);
4692 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4693 mtx_lock(&cpr->pr_mtx);
4694 #ifdef INET
4695 prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4696 #endif
4697 #ifdef INET6
4698 prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4699 #endif
4700 bzero(xp, sizeof(*xp));
4701 xp->pr_version = XPRISON_VERSION;
4702 xp->pr_id = cpr->pr_id;
4703 xp->pr_state = cpr->pr_state;
4704 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4705 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4706 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4707 #ifdef INET
4708 xp->pr_ip4s = ip4s;
4709 #endif
4710 #ifdef INET6
4711 xp->pr_ip6s = ip6s;
4712 #endif
4713 mtx_unlock(&cpr->pr_mtx);
4714 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4715 if (error)
4716 break;
4717 #ifdef INET
4718 if (xp->pr_ip4s > 0) {
4719 error = SYSCTL_OUT(req, ip4,
4720 xp->pr_ip4s * sizeof(struct in_addr));
4721 if (error)
4722 break;
4723 }
4724 #endif
4725 #ifdef INET6
4726 if (xp->pr_ip6s > 0) {
4727 error = SYSCTL_OUT(req, ip6,
4728 xp->pr_ip6s * sizeof(struct in6_addr));
4729 if (error)
4730 break;
4731 }
4732 #endif
4733 }
4734 sx_sunlock(&allprison_lock);
4735 free(xp, M_TEMP);
4736 #ifdef INET
4737 free(ip4, M_TEMP);
4738 #endif
4739 #ifdef INET6
4740 free(ip6, M_TEMP);
4741 #endif
4742 return (error);
4743 }
4744
4745 SYSCTL_OID(_security_jail, OID_AUTO, list,
4746 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4747 sysctl_jail_list, "S", "List of active jails");
4748
4749 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)4750 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4751 {
4752 int error, injail;
4753
4754 injail = jailed(req->td->td_ucred);
4755 error = SYSCTL_OUT(req, &injail, sizeof(injail));
4756
4757 return (error);
4758 }
4759
4760 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4761 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4762 sysctl_jail_jailed, "I", "Process in jail?");
4763
4764 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)4765 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4766 {
4767 int error, havevnet;
4768 #ifdef VIMAGE
4769 struct ucred *cred = req->td->td_ucred;
4770
4771 havevnet = jailed(cred) && prison_owns_vnet(cred->cr_prison);
4772 #else
4773 havevnet = 0;
4774 #endif
4775 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4776
4777 return (error);
4778 }
4779
4780 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4781 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4782 sysctl_jail_vnet, "I", "Jail owns vnet?");
4783
4784 #if defined(INET) || defined(INET6)
4785 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4786 &jail_max_af_ips, 0,
4787 "Number of IP addresses a jail may have at most per address family (deprecated)");
4788 #endif
4789
4790 /*
4791 * Default parameters for jail(2) compatibility. For historical reasons,
4792 * the sysctl names have varying similarity to the parameter names. Prisons
4793 * just see their own parameters, and can't change them.
4794 */
4795 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)4796 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4797 {
4798 int error, i;
4799
4800 /* Get the current flag value, and convert it to a boolean. */
4801 if (req->td->td_ucred->cr_prison == &prison0) {
4802 mtx_lock(&prison0.pr_mtx);
4803 i = (jail_default_allow & arg2) != 0;
4804 mtx_unlock(&prison0.pr_mtx);
4805 } else
4806 i = prison_allow(req->td->td_ucred, arg2);
4807
4808 if (arg1 != NULL)
4809 i = !i;
4810 error = sysctl_handle_int(oidp, &i, 0, req);
4811 if (error || !req->newptr)
4812 return (error);
4813 i = i ? arg2 : 0;
4814 if (arg1 != NULL)
4815 i ^= arg2;
4816 /*
4817 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4818 * for writing.
4819 */
4820 mtx_lock(&prison0.pr_mtx);
4821 jail_default_allow = (jail_default_allow & ~arg2) | i;
4822 mtx_unlock(&prison0.pr_mtx);
4823 return (0);
4824 }
4825
4826 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4827 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4828 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4829 "Processes in jail can set their hostnames (deprecated)");
4830 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4831 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4832 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4833 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4834 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4835 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4836 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4837 "Processes in jail can use System V IPC primitives (deprecated)");
4838 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4839 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4840 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4841 "Prison root can create raw sockets (deprecated)");
4842 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4843 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4844 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4845 "Processes in jail can alter system file flags (deprecated)");
4846 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4847 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4848 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4849 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4850 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
4851 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4852 NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
4853 "Processes in jail can lock/unlock physical pages in memory");
4854
4855 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)4856 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4857 {
4858 struct prison *pr;
4859 int level, error;
4860
4861 pr = req->td->td_ucred->cr_prison;
4862 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4863 error = sysctl_handle_int(oidp, &level, 0, req);
4864 if (error || !req->newptr)
4865 return (error);
4866 *(int *)arg1 = level;
4867 return (0);
4868 }
4869
4870 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4871 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4872 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4873 sysctl_jail_default_level, "I",
4874 "Processes in jail cannot see all mounted file systems (deprecated)");
4875
4876 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4877 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4878 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4879 sysctl_jail_default_level, "I",
4880 "Ruleset for the devfs filesystem in jail (deprecated)");
4881
4882 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4883 "Limits and stats of child jails");
4884
4885 static int
sysctl_jail_children(SYSCTL_HANDLER_ARGS)4886 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
4887 {
4888 struct prison *pr;
4889 int i;
4890
4891 pr = req->td->td_ucred->cr_prison;
4892
4893 switch (oidp->oid_kind & CTLTYPE) {
4894 case CTLTYPE_INT:
4895 i = *(int *)((char *)pr + arg2);
4896 return (SYSCTL_OUT(req, &i, sizeof(i)));
4897 }
4898
4899 return (0);
4900 }
4901
4902 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
4903 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4904 NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
4905 "I", "Maximum number of child jails");
4906 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
4907 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4908 NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
4909 "I", "Current number of child jails");
4910
4911 /*
4912 * Nodes to describe jail parameters. Maximum length of string parameters
4913 * is returned in the string itself, and the other parameters exist merely
4914 * to make themselves and their types known.
4915 */
4916 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4917 "Jail parameters");
4918
4919 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)4920 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4921 {
4922 int i;
4923 long l;
4924 size_t s;
4925 char numbuf[12];
4926
4927 switch (oidp->oid_kind & CTLTYPE)
4928 {
4929 case CTLTYPE_LONG:
4930 case CTLTYPE_ULONG:
4931 l = 0;
4932 #ifdef SCTL_MASK32
4933 if (!(req->flags & SCTL_MASK32))
4934 #endif
4935 return (SYSCTL_OUT(req, &l, sizeof(l)));
4936 case CTLTYPE_INT:
4937 case CTLTYPE_UINT:
4938 i = 0;
4939 return (SYSCTL_OUT(req, &i, sizeof(i)));
4940 case CTLTYPE_STRING:
4941 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4942 return
4943 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4944 case CTLTYPE_STRUCT:
4945 s = (size_t)arg2;
4946 return (SYSCTL_OUT(req, &s, sizeof(s)));
4947 }
4948 return (0);
4949 }
4950
4951 /*
4952 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4953 * jail creation time but cannot be changed in an existing jail.
4954 */
4955 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4956 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4957 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4958 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4959 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4960 "I", "Jail secure level");
4961 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4962 "Jail value for kern.osreldate and uname -K");
4963 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4964 "Jail value for kern.osrelease and uname -r");
4965 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4966 "I", "Jail cannot see all mounted file systems");
4967 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4968 "I", "Ruleset for in-jail devfs mounts");
4969 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4970 "B", "Jail persistence");
4971 #ifdef VIMAGE
4972 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4973 "E,jailsys", "Virtual network stack");
4974 #endif
4975 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4976 "B", "Jail is in the process of shutting down");
4977
4978 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4979 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4980 "I", "Current number of child jails");
4981 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4982 "I", "Maximum number of child jails");
4983
4984 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4985 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4986 "Jail hostname");
4987 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4988 "Jail NIS domainname");
4989 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4990 "Jail host UUID");
4991 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4992 "LU", "Jail host ID");
4993
4994 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4995 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4996
4997 #ifdef INET
4998 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4999 "Jail IPv4 address virtualization");
5000 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
5001 "S,in_addr,a", "Jail IPv4 addresses");
5002 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5003 "B", "Do (not) use IPv4 source address selection rather than the "
5004 "primary jail IPv4 address.");
5005 #endif
5006 #ifdef INET6
5007 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
5008 "Jail IPv6 address virtualization");
5009 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
5010 "S,in6_addr,a", "Jail IPv6 addresses");
5011 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5012 "B", "Do (not) use IPv6 source address selection rather than the "
5013 "primary jail IPv6 address.");
5014 #endif
5015
5016 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
5017 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
5018 "B", "Jail may set hostname");
5019 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
5020 "B", "Jail may use SYSV IPC");
5021 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
5022 "B", "Jail may create raw sockets");
5023 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
5024 "B", "Jail may alter system file flags");
5025 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
5026 "B", "Jail may set file quotas");
5027 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
5028 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
5029 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
5030 "B", "Jail may lock (unlock) physical pages in memory");
5031 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
5032 "B", "Jail may bind sockets to reserved ports");
5033 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
5034 "B", "Jail may read the kernel message buffer");
5035 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
5036 "B", "Unprivileged processes may use process debugging facilities");
5037 SYSCTL_JAIL_PARAM(_allow, unprivileged_parent_tampering,
5038 CTLTYPE_INT | CTLFLAG_RW, "B",
5039 "Unprivileged parent jail processes may tamper with same-uid processes"
5040 " (signal/debug/cpuset)");
5041 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
5042 "B", "Processes in jail with uid 0 have privilege");
5043 #ifdef VIMAGE
5044 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
5045 "B", "Mountd/nfsd may run in the jail");
5046 #endif
5047 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
5048 "B", "Jail may set system-level filesystem extended attributes");
5049 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
5050 "B", "Jail may adjust system time");
5051 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
5052 "B", "Jail may set system time");
5053 SYSCTL_JAIL_PARAM(_allow, routing, CTLTYPE_INT | CTLFLAG_RW,
5054 "B", "Jail may modify routing table");
5055 #ifdef AUDIT
5056 SYSCTL_JAIL_PARAM(_allow, setaudit, CTLTYPE_INT | CTLFLAG_RW,
5057 "B", "Jail may set and get audit session state");
5058 #endif
5059
5060 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
5061 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
5062 "B", "Jail may mount/unmount jail-friendly file systems in general");
5063
5064 /*
5065 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
5066 * its associated bit in the pr_allow bitmask, or zero if the parameter was
5067 * not created.
5068 */
5069 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)5070 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
5071 const char *descr)
5072 {
5073 struct bool_flags *bf;
5074 struct sysctl_oid *parent;
5075 char *allow_name, *allow_noname, *allowed;
5076 #ifndef NO_SYSCTL_DESCR
5077 char *descr_deprecated;
5078 #endif
5079 u_int allow_flag;
5080
5081 if (prefix
5082 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
5083 < 0 ||
5084 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
5085 < 0
5086 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
5087 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
5088 free(allow_name, M_PRISON);
5089 return 0;
5090 }
5091
5092 /*
5093 * See if this parameter has already beed added, i.e. a module was
5094 * previously loaded/unloaded.
5095 */
5096 mtx_lock(&prison0.pr_mtx);
5097 for (bf = pr_flag_allow;
5098 bf < pr_flag_allow + nitems(pr_flag_allow) &&
5099 atomic_load_int(&bf->flag) != 0;
5100 bf++) {
5101 if (strcmp(bf->name, allow_name) == 0) {
5102 allow_flag = bf->flag;
5103 goto no_add;
5104 }
5105 }
5106
5107 /*
5108 * Find a free bit in pr_allow_all, failing if there are none
5109 * (which shouldn't happen as long as we keep track of how many
5110 * potential dynamic flags exist).
5111 */
5112 for (allow_flag = 1;; allow_flag <<= 1) {
5113 if (allow_flag == 0)
5114 goto no_add;
5115 if ((pr_allow_all & allow_flag) == 0)
5116 break;
5117 }
5118
5119 /* Note the parameter in the next open slot in pr_flag_allow. */
5120 for (bf = pr_flag_allow; ; bf++) {
5121 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
5122 /* This should never happen, but is not fatal. */
5123 allow_flag = 0;
5124 goto no_add;
5125 }
5126 if (atomic_load_int(&bf->flag) == 0)
5127 break;
5128 }
5129 bf->name = allow_name;
5130 bf->noname = allow_noname;
5131 pr_allow_all |= allow_flag;
5132 /*
5133 * prison0 always has permission for the new parameter.
5134 * Other jails must have it granted to them.
5135 */
5136 prison0.pr_allow |= allow_flag;
5137 /* The flag indicates a valid entry, so make sure it is set last. */
5138 atomic_store_rel_int(&bf->flag, allow_flag);
5139 mtx_unlock(&prison0.pr_mtx);
5140
5141 /*
5142 * Create sysctls for the parameter, and the back-compat global
5143 * permission.
5144 */
5145 parent = prefix
5146 ? SYSCTL_ADD_NODE(NULL,
5147 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
5148 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
5149 : &sysctl___security_jail_param_allow;
5150 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
5151 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5152 NULL, 0, sysctl_jail_param, "B", descr);
5153 if ((prefix
5154 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
5155 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
5156 #ifndef NO_SYSCTL_DESCR
5157 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
5158 descr);
5159 #endif
5160 (void)SYSCTL_ADD_PROC(NULL,
5161 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
5162 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
5163 sysctl_jail_default_allow, "I", descr_deprecated);
5164 #ifndef NO_SYSCTL_DESCR
5165 free(descr_deprecated, M_TEMP);
5166 #endif
5167 free(allowed, M_TEMP);
5168 }
5169 return allow_flag;
5170
5171 no_add:
5172 mtx_unlock(&prison0.pr_mtx);
5173 free(allow_name, M_PRISON);
5174 free(allow_noname, M_PRISON);
5175 return allow_flag;
5176 }
5177
5178 /*
5179 * The VFS system will register jail-aware filesystems here. They each get
5180 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
5181 * attempts to mount.
5182 */
5183 void
prison_add_vfs(struct vfsconf * vfsp)5184 prison_add_vfs(struct vfsconf *vfsp)
5185 {
5186 #ifdef NO_SYSCTL_DESCR
5187
5188 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5189 NULL, NULL);
5190 #else
5191 char *descr;
5192
5193 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
5194 vfsp->vfc_name);
5195 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5196 NULL, descr);
5197 free(descr, M_TEMP);
5198 #endif
5199 }
5200
5201 #ifdef RACCT
5202 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)5203 prison_racct_foreach(void (*callback)(struct racct *racct,
5204 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
5205 void *arg2, void *arg3)
5206 {
5207 struct prison_racct *prr;
5208
5209 ASSERT_RACCT_ENABLED();
5210
5211 sx_slock(&allprison_lock);
5212 if (pre != NULL)
5213 (pre)();
5214 LIST_FOREACH(prr, &allprison_racct, prr_next)
5215 (callback)(prr->prr_racct, arg2, arg3);
5216 if (post != NULL)
5217 (post)();
5218 sx_sunlock(&allprison_lock);
5219 }
5220
5221 static struct prison_racct *
prison_racct_find_locked(const char * name)5222 prison_racct_find_locked(const char *name)
5223 {
5224 struct prison_racct *prr;
5225
5226 ASSERT_RACCT_ENABLED();
5227 sx_assert(&allprison_lock, SA_XLOCKED);
5228
5229 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
5230 return (NULL);
5231
5232 LIST_FOREACH(prr, &allprison_racct, prr_next) {
5233 if (strcmp(name, prr->prr_name) != 0)
5234 continue;
5235
5236 /* Found prison_racct with a matching name? */
5237 prison_racct_hold(prr);
5238 return (prr);
5239 }
5240
5241 /* Add new prison_racct. */
5242 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
5243 racct_create(&prr->prr_racct);
5244
5245 strcpy(prr->prr_name, name);
5246 refcount_init(&prr->prr_refcount, 1);
5247 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
5248
5249 return (prr);
5250 }
5251
5252 struct prison_racct *
prison_racct_find(const char * name)5253 prison_racct_find(const char *name)
5254 {
5255 struct prison_racct *prr;
5256
5257 ASSERT_RACCT_ENABLED();
5258
5259 sx_xlock(&allprison_lock);
5260 prr = prison_racct_find_locked(name);
5261 sx_xunlock(&allprison_lock);
5262 return (prr);
5263 }
5264
5265 void
prison_racct_hold(struct prison_racct * prr)5266 prison_racct_hold(struct prison_racct *prr)
5267 {
5268
5269 ASSERT_RACCT_ENABLED();
5270
5271 refcount_acquire(&prr->prr_refcount);
5272 }
5273
5274 static void
prison_racct_free_locked(struct prison_racct * prr)5275 prison_racct_free_locked(struct prison_racct *prr)
5276 {
5277
5278 ASSERT_RACCT_ENABLED();
5279 sx_assert(&allprison_lock, SA_XLOCKED);
5280
5281 if (refcount_release(&prr->prr_refcount)) {
5282 racct_destroy(&prr->prr_racct);
5283 LIST_REMOVE(prr, prr_next);
5284 free(prr, M_PRISON_RACCT);
5285 }
5286 }
5287
5288 void
prison_racct_free(struct prison_racct * prr)5289 prison_racct_free(struct prison_racct *prr)
5290 {
5291
5292 ASSERT_RACCT_ENABLED();
5293 sx_assert(&allprison_lock, SA_UNLOCKED);
5294
5295 if (refcount_release_if_not_last(&prr->prr_refcount))
5296 return;
5297
5298 sx_xlock(&allprison_lock);
5299 prison_racct_free_locked(prr);
5300 sx_xunlock(&allprison_lock);
5301 }
5302
5303 static void
prison_racct_attach(struct prison * pr)5304 prison_racct_attach(struct prison *pr)
5305 {
5306 struct prison_racct *prr;
5307
5308 ASSERT_RACCT_ENABLED();
5309 sx_assert(&allprison_lock, SA_XLOCKED);
5310
5311 prr = prison_racct_find_locked(pr->pr_name);
5312 KASSERT(prr != NULL, ("cannot find prison_racct"));
5313
5314 pr->pr_prison_racct = prr;
5315 }
5316
5317 /*
5318 * Handle jail renaming. From the racct point of view, renaming means
5319 * moving from one prison_racct to another.
5320 */
5321 static void
prison_racct_modify(struct prison * pr)5322 prison_racct_modify(struct prison *pr)
5323 {
5324 #ifdef RCTL
5325 struct proc *p;
5326 struct ucred *cred;
5327 #endif
5328 struct prison_racct *oldprr;
5329
5330 ASSERT_RACCT_ENABLED();
5331
5332 sx_slock(&allproc_lock);
5333 sx_xlock(&allprison_lock);
5334
5335 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
5336 sx_xunlock(&allprison_lock);
5337 sx_sunlock(&allproc_lock);
5338 return;
5339 }
5340
5341 oldprr = pr->pr_prison_racct;
5342 pr->pr_prison_racct = NULL;
5343
5344 prison_racct_attach(pr);
5345
5346 /*
5347 * Move resource utilisation records.
5348 */
5349 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
5350
5351 #ifdef RCTL
5352 /*
5353 * Force rctl to reattach rules to processes.
5354 */
5355 FOREACH_PROC_IN_SYSTEM(p) {
5356 PROC_LOCK(p);
5357 cred = crhold(p->p_ucred);
5358 PROC_UNLOCK(p);
5359 rctl_proc_ucred_changed(p, cred);
5360 crfree(cred);
5361 }
5362 #endif
5363
5364 sx_sunlock(&allproc_lock);
5365 prison_racct_free_locked(oldprr);
5366 sx_xunlock(&allprison_lock);
5367 }
5368
5369 static void
prison_racct_detach(struct prison * pr)5370 prison_racct_detach(struct prison *pr)
5371 {
5372
5373 ASSERT_RACCT_ENABLED();
5374 sx_assert(&allprison_lock, SA_UNLOCKED);
5375
5376 if (pr->pr_prison_racct == NULL)
5377 return;
5378 prison_racct_free(pr->pr_prison_racct);
5379 pr->pr_prison_racct = NULL;
5380 }
5381 #endif /* RACCT */
5382
5383 /*
5384 * Submit a knote for a prison, locking if necessary.
5385 */
5386 static void
prison_knote(struct prison * pr,long hint)5387 prison_knote(struct prison *pr, long hint)
5388 {
5389 int locked;
5390
5391 locked = mtx_owned(&pr->pr_mtx);
5392 if (!locked)
5393 mtx_lock(&pr->pr_mtx);
5394 KNOTE_LOCKED(pr->pr_klist, hint);
5395 jaildesc_knote(pr, hint);
5396 if (!locked)
5397 mtx_unlock(&pr->pr_mtx);
5398 }
5399
5400 #ifdef DDB
5401
5402 static void
db_show_prison(struct prison * pr)5403 db_show_prison(struct prison *pr)
5404 {
5405 struct bool_flags *bf;
5406 struct jailsys_flags *jsf;
5407 #if defined(INET) || defined(INET6)
5408 int ii;
5409 struct prison_ip *pip;
5410 #endif
5411 unsigned f;
5412 #ifdef INET
5413 char ip4buf[INET_ADDRSTRLEN];
5414 #endif
5415 #ifdef INET6
5416 char ip6buf[INET6_ADDRSTRLEN];
5417 #endif
5418
5419 db_printf("prison %p:\n", pr);
5420 db_printf(" jid = %d\n", pr->pr_id);
5421 db_printf(" name = %s\n", pr->pr_name);
5422 db_printf(" parent = %p\n", pr->pr_parent);
5423 db_printf(" ref = %d\n", pr->pr_ref);
5424 db_printf(" uref = %d\n", pr->pr_uref);
5425 db_printf(" state = %s\n",
5426 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5427 pr->pr_state == PRISON_STATE_DYING ? "dying" :
5428 "invalid");
5429 db_printf(" path = %s\n", pr->pr_path);
5430 db_printf(" cpuset = %d\n", pr->pr_cpuset
5431 ? pr->pr_cpuset->cs_id : -1);
5432 #ifdef VIMAGE
5433 db_printf(" vnet = %p\n", pr->pr_vnet);
5434 #endif
5435 db_printf(" root = %p\n", pr->pr_root);
5436 db_printf(" securelevel = %d\n", pr->pr_securelevel);
5437 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
5438 db_printf(" children.max = %d\n", pr->pr_childmax);
5439 db_printf(" children.cur = %d\n", pr->pr_childcount);
5440 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
5441 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
5442 db_printf(" flags = 0x%x", pr->pr_flags);
5443 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5444 if (pr->pr_flags & bf->flag)
5445 db_printf(" %s", bf->name);
5446 for (jsf = pr_flag_jailsys;
5447 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5448 jsf++) {
5449 f = pr->pr_flags & (jsf->disable | jsf->new);
5450 db_printf(" %-16s= %s\n", jsf->name,
5451 (f != 0 && f == jsf->disable) ? "disable"
5452 : (f == jsf->new) ? "new"
5453 : "inherit");
5454 }
5455 db_printf(" allow = 0x%x", pr->pr_allow);
5456 for (bf = pr_flag_allow;
5457 bf < pr_flag_allow + nitems(pr_flag_allow) &&
5458 atomic_load_int(&bf->flag) != 0;
5459 bf++)
5460 if (pr->pr_allow & bf->flag)
5461 db_printf(" %s", bf->name);
5462 db_printf("\n");
5463 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
5464 db_printf(" host.hostname = %s\n", pr->pr_hostname);
5465 db_printf(" host.domainname = %s\n", pr->pr_domainname);
5466 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
5467 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
5468 #ifdef INET
5469 if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5470 db_printf(" ip4s = %d\n", pip->ips);
5471 for (ii = 0; ii < pip->ips; ii++)
5472 db_printf(" %s %s\n",
5473 ii == 0 ? "ip4.addr =" : " ",
5474 inet_ntoa_r(
5475 *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5476 ip4buf));
5477 }
5478 #endif
5479 #ifdef INET6
5480 if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5481 db_printf(" ip6s = %d\n", pip->ips);
5482 for (ii = 0; ii < pip->ips; ii++)
5483 db_printf(" %s %s\n",
5484 ii == 0 ? "ip6.addr =" : " ",
5485 ip6_sprintf(ip6buf,
5486 (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5487 }
5488 #endif
5489 }
5490
DB_SHOW_COMMAND(prison,db_show_prison_command)5491 DB_SHOW_COMMAND(prison, db_show_prison_command)
5492 {
5493 struct prison *pr;
5494
5495 if (!have_addr) {
5496 /*
5497 * Show all prisons in the list, and prison0 which is not
5498 * listed.
5499 */
5500 db_show_prison(&prison0);
5501 if (!db_pager_quit) {
5502 TAILQ_FOREACH(pr, &allprison, pr_list) {
5503 db_show_prison(pr);
5504 if (db_pager_quit)
5505 break;
5506 }
5507 }
5508 return;
5509 }
5510
5511 if (addr == 0)
5512 pr = &prison0;
5513 else {
5514 /* Look for a prison with the ID and with references. */
5515 TAILQ_FOREACH(pr, &allprison, pr_list)
5516 if (pr->pr_id == addr && pr->pr_ref > 0)
5517 break;
5518 if (pr == NULL)
5519 /* Look again, without requiring a reference. */
5520 TAILQ_FOREACH(pr, &allprison, pr_list)
5521 if (pr->pr_id == addr)
5522 break;
5523 if (pr == NULL)
5524 /* Assume address points to a valid prison. */
5525 pr = (struct prison *)addr;
5526 }
5527 db_show_prison(pr);
5528 }
5529
5530 #endif /* DDB */
5531