1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1999 Poul-Henning Kamp.
5 * Copyright (c) 2008 Bjoern A. Zeeb.
6 * Copyright (c) 2009 James Gritton.
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 #include "opt_ddb.h"
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_nfs.h"
36
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/ctype.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/file.h>
44 #include <sys/sysproto.h>
45 #include <sys/malloc.h>
46 #include <sys/osd.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/epoch.h>
50 #include <sys/event.h>
51 #include <sys/taskqueue.h>
52 #include <sys/fcntl.h>
53 #include <sys/jail.h>
54 #include <sys/jaildesc.h>
55 #include <sys/linker.h>
56 #include <sys/lock.h>
57 #include <sys/mman.h>
58 #include <sys/mutex.h>
59 #include <sys/racct.h>
60 #include <sys/rctl.h>
61 #include <sys/refcount.h>
62 #include <sys/sx.h>
63 #include <sys/sysent.h>
64 #include <sys/namei.h>
65 #include <sys/mount.h>
66 #include <sys/queue.h>
67 #include <sys/socket.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/uuid.h>
71 #include <sys/vnode.h>
72
73 #include <net/if.h>
74 #include <net/vnet.h>
75
76 #include <netinet/in.h>
77
78 #ifdef DDB
79 #include <ddb/ddb.h>
80 #endif /* DDB */
81
82 #include <security/mac/mac_framework.h>
83 #include <security/mac/mac_policy.h>
84 #include <security/mac/mac_syscalls.h>
85
86 #define PRISON0_HOSTUUID_MODULE "hostuuid"
87
88 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
89 #ifdef RACCT
90 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
91 #endif
92
93 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
94 #ifdef INET
95 #ifdef INET6
96 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
97 #else
98 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
99 #endif
100 #else /* !INET */
101 #ifdef INET6
102 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
103 #else
104 #define _PR_IP_SADDRSEL 0
105 #endif
106 #endif
107
108 /* prison0 describes what is "real" about the system. */
109 struct prison prison0 = {
110 .pr_id = 0,
111 .pr_name = "0",
112 .pr_ref = 1,
113 .pr_uref = 1,
114 .pr_path = "/",
115 .pr_securelevel = -1,
116 .pr_devfs_rsnum = 0,
117 .pr_state = PRISON_STATE_ALIVE,
118 .pr_childmax = JAIL_MAX,
119 .pr_hostuuid = DEFAULT_HOSTUUID,
120 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
121 #ifdef VIMAGE
122 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
123 #else
124 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
125 #endif
126 .pr_allow = PR_ALLOW_PRISON0,
127 };
128 _Static_assert((PR_ALLOW_PRISON0 & ~PR_ALLOW_ALL_STATIC) == 0,
129 "Bits enabled in PR_ALLOW_PRISON0 that are not statically reserved");
130
131 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
132
133 struct bool_flags {
134 const char *name;
135 const char *noname;
136 volatile u_int flag;
137 };
138 struct jailsys_flags {
139 const char *name;
140 unsigned disable;
141 unsigned new;
142 };
143
144 /*
145 * Handle jail teardown in a dedicated thread to avoid deadlocks from
146 * vnet_destroy().
147 */
148 TASKQUEUE_DEFINE_THREAD(jail_remove);
149
150 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
151 struct sx allprison_lock;
152 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
153 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
154 LIST_HEAD(, prison_racct) allprison_racct;
155 int lastprid = 0;
156 int lastdeadid = 0;
157
158 static int get_next_prid(struct prison **insprp);
159 static int get_next_deadid(struct prison **insprp);
160 static int do_jail_attach(struct thread *td, struct prison *pr, int *drflagsp);
161 static int prison_attach_thread_single(struct thread *td);
162 static void prison_attach_thread_single_end(struct thread *td);
163 static void prison_complete(void *context, int pending);
164 static void prison_deref(struct prison *pr, int flags);
165 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
166 static int prison_lock_xlock(struct prison *pr, int flags);
167 static void prison_cleanup_locked(struct prison *pr);
168 static void prison_cleanup_unlocked(struct prison *pr);
169 static void prison_free_not_last(struct prison *pr);
170 static void prison_proc_free_not_last(struct prison *pr);
171 static void prison_proc_relink(struct prison *opr, struct prison *npr,
172 struct proc *p);
173 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
174 int enable);
175 static char *prison_path(struct prison *pr1, struct prison *pr2);
176 #ifdef RACCT
177 static void prison_racct_attach(struct prison *pr);
178 static void prison_racct_modify(struct prison *pr);
179 static void prison_racct_detach(struct prison *pr);
180 #endif
181 static void prison_knote(struct prison *pr, long hint);
182
183 /* Flags for prison_deref */
184 #define PD_DEREF 0x01 /* Decrement pr_ref */
185 #define PD_DEUREF 0x02 /* Decrement pr_uref */
186 #define PD_KILL 0x04 /* Remove jail, kill processes, etc */
187 #define PD_LOCKED 0x10 /* pr_mtx is held */
188 #define PD_LIST_SLOCKED 0x20 /* allprison_lock is held shared */
189 #define PD_LIST_XLOCKED 0x40 /* allprison_lock is held exclusive */
190 #define PD_OP_FLAGS 0x07 /* Operation flags */
191 #define PD_LOCK_FLAGS 0x70 /* Lock status flags */
192
193 /*
194 * Parameter names corresponding to PR_* flag values. Size values are for kvm
195 * as we cannot figure out the size of a sparse array, or an array without a
196 * terminating entry.
197 */
198 static struct bool_flags pr_flag_bool[] = {
199 {JAIL_PARAM_PERSIST, JAIL_PARAM_NOPERSIST, PR_PERSIST},
200 #ifdef INET
201 {JAIL_PARAM_IP4_SADDRSEL, JAIL_PARAM_IP4_NOSADDRSEL, PR_IP4_SADDRSEL},
202 #endif
203 #ifdef INET6
204 {JAIL_PARAM_IP6_SADDRSEL, JAIL_PARAM_IP6_NOSADDRSEL, PR_IP6_SADDRSEL},
205 #endif
206 };
207 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
208
209 static struct jailsys_flags pr_flag_jailsys[] = {
210 {JAIL_PARAM_HOST, 0, PR_HOST},
211 #ifdef VIMAGE
212 {JAIL_PARAM_VNET, 0, PR_VNET},
213 #endif
214 #ifdef INET
215 {JAIL_PARAM_IP4, PR_IP4_USER, PR_IP4_USER},
216 #endif
217 #ifdef INET6
218 {JAIL_PARAM_IP6, PR_IP6_USER, PR_IP6_USER},
219 #endif
220 };
221 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
222
223 /*
224 * Make this array full-size so dynamic parameters can be added.
225 * It is protected by prison0.mtx, but lockless reading is allowed
226 * with an atomic check of the flag values.
227 */
228 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
229 {JAIL_PARAM_ALLOW_SET_HOSTNAME, JAIL_PARAM_ALLOW_NOSET_HOSTNAME,
230 PR_ALLOW_SET_HOSTNAME},
231 {JAIL_PARAM_ALLOW_SYSVIPC, JAIL_PARAM_ALLOW_NOSYSVIPC,
232 PR_ALLOW_SYSVIPC},
233 {JAIL_PARAM_ALLOW_RAW_SOCKETS, JAIL_PARAM_ALLOW_NORAW_SOCKETS,
234 PR_ALLOW_RAW_SOCKETS},
235 {JAIL_PARAM_ALLOW_CHFLAGS, JAIL_PARAM_ALLOW_NOCHFLAGS,
236 PR_ALLOW_CHFLAGS},
237 {JAIL_PARAM_ALLOW_MOUNT, JAIL_PARAM_ALLOW_NOMOUNT, PR_ALLOW_MOUNT},
238 {JAIL_PARAM_ALLOW_QUOTAS, JAIL_PARAM_ALLOW_NOQUOTAS, PR_ALLOW_QUOTAS},
239 {JAIL_PARAM_ALLOW_SOCKET_AF, JAIL_PARAM_ALLOW_NOSOCKET_AF,
240 PR_ALLOW_SOCKET_AF},
241 {JAIL_PARAM_ALLOW_MLOCK, JAIL_PARAM_ALLOW_NOMLOCK, PR_ALLOW_MLOCK},
242 {JAIL_PARAM_ALLOW_RESERVED_PORTS, JAIL_PARAM_ALLOW_NORESERVED_PORTS,
243 PR_ALLOW_RESERVED_PORTS},
244 {JAIL_PARAM_ALLOW_READ_MSGBUF, JAIL_PARAM_ALLOW_NOREAD_MSGBUF,
245 PR_ALLOW_READ_MSGBUF},
246 {JAIL_PARAM_ALLOW_UNPRIV_DEBUG, JAIL_PARAM_ALLOW_NOUNPRIV_DEBUG,
247 PR_ALLOW_UNPRIV_DEBUG},
248 {JAIL_PARAM_ALLOW_SUSER, JAIL_PARAM_ALLOW_NOSUSER, PR_ALLOW_SUSER},
249 #ifdef VIMAGE
250 {JAIL_PARAM_ALLOW_NFSD, JAIL_PARAM_ALLOW_NONFSD, PR_ALLOW_NFSD},
251 #endif
252 {JAIL_PARAM_ALLOW_EXTATTR, JAIL_PARAM_ALLOW_NOEXTATTR,
253 PR_ALLOW_EXTATTR},
254 {JAIL_PARAM_ALLOW_ADJTIME, JAIL_PARAM_ALLOW_NOADJTIME,
255 PR_ALLOW_ADJTIME},
256 {JAIL_PARAM_ALLOW_SETTIME, JAIL_PARAM_ALLOW_NOSETTIME,
257 PR_ALLOW_SETTIME},
258 {JAIL_PARAM_ALLOW_ROUTING, JAIL_PARAM_ALLOW_NOROUTING,
259 PR_ALLOW_ROUTING},
260 {JAIL_PARAM_ALLOW_UNPRIV_PARENT_TAMPER,
261 JAIL_PARAM_ALLOW_NOUNPRIV_PARENT_TAMPER,
262 PR_ALLOW_UNPRIV_PARENT_TAMPER},
263 #ifdef AUDIT
264 {JAIL_PARAM_ALLOW_SETAUDIT, JAIL_PARAM_ALLOW_NOSETAUDIT,
265 PR_ALLOW_SETAUDIT},
266 #endif
267 };
268 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
269 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
270
271 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
272 PR_ALLOW_RESERVED_PORTS | \
273 PR_ALLOW_UNPRIV_DEBUG | \
274 PR_ALLOW_SUSER)
275 #define JAIL_DEFAULT_ENFORCE_STATFS 2
276 #define JAIL_DEFAULT_DEVFS_RSNUM 0
277 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
278 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
279 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
280 #if defined(INET) || defined(INET6)
281 static unsigned jail_max_af_ips = 255;
282 #endif
283
284 /*
285 * Initialize the parts of prison0 that can't be static-initialized with
286 * constants. This is called from proc0_init() after creating thread0 cpuset.
287 */
288 void
prison0_init(void)289 prison0_init(void)
290 {
291 uint8_t *file, *data;
292 size_t size;
293 char buf[sizeof(prison0.pr_hostuuid)];
294 #ifdef MAC
295 int error __diagused;
296 #endif
297 bool valid;
298
299 #ifdef MAC
300 error = mac_prison_init(&prison0, M_WAITOK);
301 MPASS(error == 0);
302
303 mtx_unlock(&prison0.pr_mtx);
304 #endif
305 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
306 prison0.pr_osreldate = osreldate;
307 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
308
309 /* If we have a preloaded hostuuid, use it. */
310 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
311 if (file != NULL) {
312 data = preload_fetch_addr(file);
313 size = preload_fetch_size(file);
314 if (data != NULL) {
315 /*
316 * The preloaded data may include trailing whitespace, almost
317 * certainly a newline; skip over any whitespace or
318 * non-printable characters to be safe.
319 */
320 while (size > 0 && data[size - 1] <= 0x20) {
321 size--;
322 }
323
324 valid = false;
325
326 /*
327 * Not NUL-terminated when passed from loader, but
328 * validate_uuid requires that due to using sscanf (as
329 * does the subsequent strlcpy, since it still reads
330 * past the given size to return the true length);
331 * bounce to a temporary buffer to fix.
332 */
333 if (size >= sizeof(buf))
334 goto done;
335
336 memcpy(buf, data, size);
337 buf[size] = '\0';
338
339 if (validate_uuid(buf, size, NULL, 0) != 0)
340 goto done;
341
342 valid = true;
343 (void)strlcpy(prison0.pr_hostuuid, buf,
344 sizeof(prison0.pr_hostuuid));
345
346 done:
347 if (bootverbose && !valid) {
348 printf("hostuuid: preload data malformed: '%.*s'\n",
349 (int)size, data);
350 }
351 }
352 }
353 if (bootverbose)
354 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
355 }
356
357 /*
358 * struct jail_args {
359 * struct jail *jail;
360 * };
361 */
362 int
sys_jail(struct thread * td,struct jail_args * uap)363 sys_jail(struct thread *td, struct jail_args *uap)
364 {
365 uint32_t version;
366 int error;
367 struct jail j;
368
369 error = copyin(uap->jail, &version, sizeof(uint32_t));
370 if (error)
371 return (error);
372
373 switch (version) {
374 case 0:
375 {
376 struct jail_v0 j0;
377
378 /* FreeBSD single IPv4 jails. */
379 bzero(&j, sizeof(struct jail));
380 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
381 if (error)
382 return (error);
383 j.version = j0.version;
384 j.path = j0.path;
385 j.hostname = j0.hostname;
386 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
387 break;
388 }
389
390 case 1:
391 /*
392 * Version 1 was used by multi-IPv4 jail implementations
393 * that never made it into the official kernel.
394 */
395 return (EINVAL);
396
397 case 2: /* JAIL_API_VERSION */
398 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
399 error = copyin(uap->jail, &j, sizeof(struct jail));
400 if (error)
401 return (error);
402 break;
403
404 default:
405 /* Sci-Fi jails are not supported, sorry. */
406 return (EINVAL);
407 }
408 return (kern_jail(td, &j));
409 }
410
411 int
kern_jail(struct thread * td,struct jail * j)412 kern_jail(struct thread *td, struct jail *j)
413 {
414 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
415 #ifdef INET
416 + 1
417 #endif
418 #ifdef INET6
419 + 1
420 #endif
421 )];
422 struct uio opt;
423 char *u_path, *u_hostname, *u_name;
424 struct bool_flags *bf;
425 #ifdef INET
426 uint32_t ip4s;
427 struct in_addr *u_ip4;
428 #endif
429 #ifdef INET6
430 struct in6_addr *u_ip6;
431 #endif
432 size_t tmplen;
433 int error, enforce_statfs;
434
435 bzero(&optiov, sizeof(optiov));
436 opt.uio_iov = optiov;
437 opt.uio_iovcnt = 0;
438 opt.uio_offset = -1;
439 opt.uio_resid = -1;
440 opt.uio_segflg = UIO_SYSSPACE;
441 opt.uio_rw = UIO_READ;
442 opt.uio_td = td;
443
444 /* Set permissions for top-level jails from sysctls. */
445 if (!jailed(td->td_ucred)) {
446 for (bf = pr_flag_allow;
447 bf < pr_flag_allow + nitems(pr_flag_allow) &&
448 atomic_load_int(&bf->flag) != 0;
449 bf++) {
450 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
451 (jail_default_allow & bf->flag)
452 ? bf->name : bf->noname);
453 optiov[opt.uio_iovcnt].iov_len =
454 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
455 opt.uio_iovcnt += 2;
456 }
457 optiov[opt.uio_iovcnt].iov_base = JAIL_PARAM_ENFORCE_STATFS;
458 optiov[opt.uio_iovcnt].iov_len = sizeof(JAIL_PARAM_ENFORCE_STATFS);
459 opt.uio_iovcnt++;
460 enforce_statfs = jail_default_enforce_statfs;
461 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
462 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
463 opt.uio_iovcnt++;
464 }
465
466 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
467 #ifdef INET
468 ip4s = (j->version == 0) ? 1 : j->ip4s;
469 if (ip4s > jail_max_af_ips)
470 return (EINVAL);
471 tmplen += ip4s * sizeof(struct in_addr);
472 #else
473 if (j->ip4s > 0)
474 return (EINVAL);
475 #endif
476 #ifdef INET6
477 if (j->ip6s > jail_max_af_ips)
478 return (EINVAL);
479 tmplen += j->ip6s * sizeof(struct in6_addr);
480 #else
481 if (j->ip6s > 0)
482 return (EINVAL);
483 #endif
484 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
485 u_hostname = u_path + MAXPATHLEN;
486 u_name = u_hostname + MAXHOSTNAMELEN;
487 #ifdef INET
488 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
489 #endif
490 #ifdef INET6
491 #ifdef INET
492 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
493 #else
494 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
495 #endif
496 #endif
497 optiov[opt.uio_iovcnt].iov_base = JAIL_PARAM_PATH;
498 optiov[opt.uio_iovcnt].iov_len = sizeof(JAIL_PARAM_PATH);
499 opt.uio_iovcnt++;
500 optiov[opt.uio_iovcnt].iov_base = u_path;
501 error = copyinstr(j->path, u_path, MAXPATHLEN,
502 &optiov[opt.uio_iovcnt].iov_len);
503 if (error) {
504 free(u_path, M_TEMP);
505 return (error);
506 }
507 opt.uio_iovcnt++;
508 optiov[opt.uio_iovcnt].iov_base = JAIL_PARAM_HOST_HOSTNAME;
509 optiov[opt.uio_iovcnt].iov_len = sizeof(JAIL_PARAM_HOST_HOSTNAME);
510 opt.uio_iovcnt++;
511 optiov[opt.uio_iovcnt].iov_base = u_hostname;
512 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
513 &optiov[opt.uio_iovcnt].iov_len);
514 if (error) {
515 free(u_path, M_TEMP);
516 return (error);
517 }
518 opt.uio_iovcnt++;
519 if (j->jailname != NULL) {
520 optiov[opt.uio_iovcnt].iov_base = JAIL_PARAM_NAME;
521 optiov[opt.uio_iovcnt].iov_len = sizeof(JAIL_PARAM_NAME);
522 opt.uio_iovcnt++;
523 optiov[opt.uio_iovcnt].iov_base = u_name;
524 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
525 &optiov[opt.uio_iovcnt].iov_len);
526 if (error) {
527 free(u_path, M_TEMP);
528 return (error);
529 }
530 opt.uio_iovcnt++;
531 }
532 #ifdef INET
533 optiov[opt.uio_iovcnt].iov_base = JAIL_PARAM_IP4_ADDR;
534 optiov[opt.uio_iovcnt].iov_len = sizeof(JAIL_PARAM_IP4_ADDR);
535 opt.uio_iovcnt++;
536 optiov[opt.uio_iovcnt].iov_base = u_ip4;
537 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
538 if (j->version == 0)
539 u_ip4->s_addr = j->ip4s;
540 else {
541 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
542 if (error) {
543 free(u_path, M_TEMP);
544 return (error);
545 }
546 }
547 opt.uio_iovcnt++;
548 #endif
549 #ifdef INET6
550 optiov[opt.uio_iovcnt].iov_base = JAIL_PARAM_IP6_ADDR;
551 optiov[opt.uio_iovcnt].iov_len = sizeof(JAIL_PARAM_IP6_ADDR);
552 opt.uio_iovcnt++;
553 optiov[opt.uio_iovcnt].iov_base = u_ip6;
554 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
555 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
556 if (error) {
557 free(u_path, M_TEMP);
558 return (error);
559 }
560 opt.uio_iovcnt++;
561 #endif
562 KASSERT(opt.uio_iovcnt <= nitems(optiov),
563 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
564 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
565 free(u_path, M_TEMP);
566 return (error);
567 }
568
569 /*
570 * struct jail_set_args {
571 * struct iovec *iovp;
572 * unsigned int iovcnt;
573 * int flags;
574 * };
575 */
576 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)577 sys_jail_set(struct thread *td, struct jail_set_args *uap)
578 {
579 return (user_jail_set(td, uap->iovp, uap->iovcnt, uap->flags,
580 copyinuio));
581 }
582
583 int
user_jail_set(struct thread * td,struct iovec * iovp,unsigned int iovcnt,int flags,copyinuio_t * copyinuio_f)584 user_jail_set(struct thread *td, struct iovec *iovp,
585 unsigned int iovcnt, int flags, copyinuio_t *copyinuio_f)
586 {
587 struct uio *auio;
588 int error;
589
590 /* Check that we have an even number of iovecs. */
591 if (iovcnt & 1)
592 return (EINVAL);
593
594 error = copyinuio_f(iovp, iovcnt, &auio);
595 if (error)
596 return (error);
597 error = kern_jail_set(td, auio, flags);
598 freeuio(auio);
599 return (error);
600 }
601
602 #if defined(INET) || defined(INET6)
603 typedef int prison_addr_cmp_t(const void *, const void *);
604 typedef bool prison_addr_valid_t(const void *);
605 static const struct pr_family {
606 size_t size;
607 prison_addr_cmp_t *cmp;
608 prison_addr_valid_t *valid;
609 int ip_flag;
610 } pr_families[PR_FAMILY_MAX] = {
611 #ifdef INET
612 [PR_INET] = {
613 .size = sizeof(struct in_addr),
614 .cmp = prison_qcmp_v4,
615 .valid = prison_valid_v4,
616 .ip_flag = PR_IP4_USER,
617 },
618 #endif
619 #ifdef INET6
620 [PR_INET6] = {
621 .size = sizeof(struct in6_addr),
622 .cmp = prison_qcmp_v6,
623 .valid = prison_valid_v6,
624 .ip_flag = PR_IP6_USER,
625 },
626 #endif
627 };
628
629 /*
630 * Network address lists (pr_addrs) allocation for jails. The addresses
631 * are accessed locklessly by the network stack, thus need to be protected by
632 * the network epoch.
633 */
634 struct prison_ip {
635 struct epoch_context ctx;
636 uint32_t ips;
637 #ifdef FUTURE_C
638 /*
639 * XXX Variable-length automatic arrays in union may be
640 * supported in future C.
641 */
642 union {
643 char pr_ip[];
644 struct in_addr pr_ip4[];
645 struct in6_addr pr_ip6[];
646 };
647 #else /* No future C :( */
648 char pr_ip[];
649 #endif
650 };
651
652 static char *
PR_IP(struct prison_ip * pip,const pr_family_t af,int idx)653 PR_IP(struct prison_ip *pip, const pr_family_t af, int idx)
654 {
655 MPASS(pip);
656 MPASS(af < PR_FAMILY_MAX);
657 MPASS(idx >= 0 && idx < pip->ips);
658
659 return (pip->pr_ip + pr_families[af].size * idx);
660 }
661
662 static struct prison_ip *
prison_ip_alloc(const pr_family_t af,uint32_t cnt,int flags)663 prison_ip_alloc(const pr_family_t af, uint32_t cnt, int flags)
664 {
665 struct prison_ip *pip;
666
667 pip = malloc(sizeof(struct prison_ip) + cnt * pr_families[af].size,
668 M_PRISON, flags);
669 if (pip != NULL)
670 pip->ips = cnt;
671 return (pip);
672 }
673
674 /*
675 * Allocate and copyin user supplied address list, sorting and validating.
676 * kern_jail_set() helper.
677 */
678 static struct prison_ip *
prison_ip_copyin(const pr_family_t af,void * op,uint32_t cnt)679 prison_ip_copyin(const pr_family_t af, void *op, uint32_t cnt)
680 {
681 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
682 const size_t size = pr_families[af].size;
683 struct prison_ip *pip;
684
685 pip = prison_ip_alloc(af, cnt, M_WAITOK);
686 bcopy(op, pip->pr_ip, cnt * size);
687 /*
688 * IP addresses are all sorted but ip[0] to preserve
689 * the primary IP address as given from userland.
690 * This special IP is used for unbound outgoing
691 * connections as well for "loopback" traffic in case
692 * source address selection cannot find any more fitting
693 * address to connect from.
694 */
695 if (cnt > 1)
696 qsort(PR_IP(pip, af, 1), cnt - 1, size, cmp);
697 /*
698 * Check for duplicate addresses and do some simple
699 * zero and broadcast checks. If users give other bogus
700 * addresses it is their problem.
701 */
702 for (int i = 0; i < cnt; i++) {
703 if (!pr_families[af].valid(PR_IP(pip, af, i))) {
704 free(pip, M_PRISON);
705 return (NULL);
706 }
707 if (i + 1 < cnt &&
708 (cmp(PR_IP(pip, af, 0), PR_IP(pip, af, i + 1)) == 0 ||
709 cmp(PR_IP(pip, af, i), PR_IP(pip, af, i + 1)) == 0)) {
710 free(pip, M_PRISON);
711 return (NULL);
712 }
713 }
714
715 return (pip);
716 }
717
718 /*
719 * Allocate and dup parent prison address list.
720 * kern_jail_set() helper.
721 */
722 static void
prison_ip_dup(struct prison * ppr,struct prison * pr,const pr_family_t af)723 prison_ip_dup(struct prison *ppr, struct prison *pr, const pr_family_t af)
724 {
725 const struct prison_ip *ppip = ppr->pr_addrs[af];
726 struct prison_ip *pip;
727
728 if (ppip != NULL) {
729 pip = prison_ip_alloc(af, ppip->ips, M_WAITOK);
730 bcopy(ppip->pr_ip, pip->pr_ip, pip->ips * pr_families[af].size);
731 pr->pr_addrs[af] = pip;
732 }
733 }
734
735 /*
736 * Make sure the new set of IP addresses is a subset of the parent's list.
737 * Don't worry about the parent being unlocked, as any setting is done with
738 * allprison_lock held.
739 * kern_jail_set() helper.
740 */
741 static bool
prison_ip_parent_match(struct prison_ip * ppip,struct prison_ip * pip,const pr_family_t af)742 prison_ip_parent_match(struct prison_ip *ppip, struct prison_ip *pip,
743 const pr_family_t af)
744 {
745 prison_addr_cmp_t *const cmp = pr_families[af].cmp;
746 int i, j;
747
748 if (ppip == NULL)
749 return (false);
750
751 for (i = 0; i < ppip->ips; i++)
752 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, i)) == 0)
753 break;
754
755 if (i == ppip->ips)
756 /* Main address not present in parent. */
757 return (false);
758
759 if (pip->ips > 1) {
760 for (i = j = 1; i < pip->ips; i++) {
761 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0)
762 /* Equals to parent primary address. */
763 continue;
764 for (; j < ppip->ips; j++)
765 if (cmp(PR_IP(pip, af, i),
766 PR_IP(ppip, af, j)) == 0)
767 break;
768 if (j == ppip->ips)
769 break;
770 }
771 if (j == ppip->ips)
772 /* Address not present in parent. */
773 return (false);
774 }
775 return (true);
776 }
777
778 /*
779 * Check for conflicting IP addresses. We permit them if there is no more
780 * than one IP on each jail. If there is a duplicate on a jail with more
781 * than one IP stop checking and return error.
782 * kern_jail_set() helper.
783 */
784 static bool
prison_ip_conflict_check(const struct prison * ppr,const struct prison * pr,struct prison_ip * pip,pr_family_t af)785 prison_ip_conflict_check(const struct prison *ppr, const struct prison *pr,
786 struct prison_ip *pip, pr_family_t af)
787 {
788 const struct prison *tppr, *tpr;
789 int descend;
790
791 #ifdef VIMAGE
792 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
793 if (tppr->pr_flags & PR_VNET)
794 break;
795 #else
796 tppr = &prison0;
797 #endif
798 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
799 if (tpr == pr ||
800 #ifdef VIMAGE
801 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
802 #endif
803 !prison_isalive(tpr)) {
804 descend = 0;
805 continue;
806 }
807 if (!(tpr->pr_flags & pr_families[af].ip_flag))
808 continue;
809 descend = 0;
810 if (tpr->pr_addrs[af] == NULL ||
811 (pip->ips == 1 && tpr->pr_addrs[af]->ips == 1))
812 continue;
813 for (int i = 0; i < pip->ips; i++)
814 if (prison_ip_check(tpr, af, PR_IP(pip, af, i)) == 0)
815 return (false);
816 }
817
818 return (true);
819 }
820
821 _Static_assert(offsetof(struct prison_ip, ctx) == 0,
822 "prison must start with epoch context");
823 static void
prison_ip_free_deferred(epoch_context_t ctx)824 prison_ip_free_deferred(epoch_context_t ctx)
825 {
826
827 free(ctx, M_PRISON);
828 }
829
830 static void
prison_ip_free(struct prison_ip * pip)831 prison_ip_free(struct prison_ip *pip)
832 {
833
834 if (pip != NULL)
835 NET_EPOCH_CALL(prison_ip_free_deferred, &pip->ctx);
836 }
837
838 static void
prison_ip_set(struct prison * pr,const pr_family_t af,struct prison_ip * new)839 prison_ip_set(struct prison *pr, const pr_family_t af, struct prison_ip *new)
840 {
841 struct prison_ip **mem, *old;
842
843 mtx_assert(&pr->pr_mtx, MA_OWNED);
844
845 mem = &pr->pr_addrs[af];
846
847 old = *mem;
848 atomic_store_ptr(mem, new);
849 prison_ip_free(old);
850 }
851
852 /*
853 * Restrict a prison's IP address list with its parent's, possibly replacing
854 * it. Return true if succeed, otherwise should redo.
855 * kern_jail_set() helper.
856 */
857 static bool
prison_ip_restrict(struct prison * pr,const pr_family_t af,struct prison_ip ** newp)858 prison_ip_restrict(struct prison *pr, const pr_family_t af,
859 struct prison_ip **newp)
860 {
861 struct prison_ip *ppip = pr->pr_parent->pr_addrs[af];
862 struct prison_ip *pip = pr->pr_addrs[af];
863 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
864 const size_t size = pr_families[af].size;
865 struct prison_ip *new = newp != NULL ? *newp : NULL;
866 uint32_t ips;
867
868 mtx_assert(&pr->pr_mtx, MA_OWNED);
869
870 /*
871 * Due to epoch-synchronized access to the IP address lists we always
872 * allocate a new list even if the old one has enough space. We could
873 * atomically update an IPv4 address inside a list, but that would
874 * screw up sorting, and in case of IPv6 we can't even atomically write
875 * one.
876 */
877 if (ppip == NULL) {
878 if (pip != NULL)
879 prison_ip_set(pr, af, NULL);
880 return (true);
881 }
882
883 if (!(pr->pr_flags & pr_families[af].ip_flag)) {
884 if (new == NULL) {
885 new = prison_ip_alloc(af, ppip->ips, M_NOWAIT);
886 if (new == NULL)
887 return (false); /* Redo */
888 }
889 /* This has no user settings, so just copy the parent's list. */
890 MPASS(new->ips == ppip->ips);
891 bcopy(ppip->pr_ip, new->pr_ip, ppip->ips * size);
892 prison_ip_set(pr, af, new);
893 if (newp != NULL)
894 *newp = NULL; /* Used */
895 } else if (pip != NULL) {
896 /* Remove addresses that aren't in the parent. */
897 int i;
898
899 i = 0; /* index in pip */
900 ips = 0; /* index in new */
901
902 if (new == NULL) {
903 new = prison_ip_alloc(af, pip->ips, M_NOWAIT);
904 if (new == NULL)
905 return (false); /* Redo */
906 }
907
908 for (int pi = 0; pi < ppip->ips; pi++)
909 if (cmp(PR_IP(pip, af, 0), PR_IP(ppip, af, pi)) == 0) {
910 /* Found our primary address in parent. */
911 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
912 size);
913 i++;
914 ips++;
915 break;
916 }
917 for (int pi = 1; i < pip->ips; ) {
918 /* Check against primary, which is unsorted. */
919 if (cmp(PR_IP(pip, af, i), PR_IP(ppip, af, 0)) == 0) {
920 /* Matches parent's primary address. */
921 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
922 size);
923 i++;
924 ips++;
925 continue;
926 }
927 /* The rest are sorted. */
928 switch (pi >= ppip->ips ? -1 :
929 cmp(PR_IP(pip, af, i), PR_IP(ppip, af, pi))) {
930 case -1:
931 i++;
932 break;
933 case 0:
934 bcopy(PR_IP(pip, af, i), PR_IP(new, af, ips),
935 size);
936 i++;
937 pi++;
938 ips++;
939 break;
940 case 1:
941 pi++;
942 break;
943 }
944 }
945 if (ips == 0) {
946 if (newp == NULL || *newp == NULL)
947 prison_ip_free(new);
948 new = NULL;
949 } else {
950 /* Shrink to real size */
951 KASSERT((new->ips >= ips),
952 ("Out-of-bounds write to prison_ip %p", new));
953 new->ips = ips;
954 }
955 prison_ip_set(pr, af, new);
956 if (newp != NULL)
957 *newp = NULL; /* Used */
958 }
959 return (true);
960 }
961
962 /*
963 * Fast-path check if an address belongs to a prison.
964 */
965 int
prison_ip_check(const struct prison * pr,const pr_family_t af,const void * addr)966 prison_ip_check(const struct prison *pr, const pr_family_t af,
967 const void *addr)
968 {
969 int (*const cmp)(const void *, const void *) = pr_families[af].cmp;
970 struct prison_ip *pip;
971 int i, a, z, d;
972
973 MPASS(mtx_owned(&pr->pr_mtx) ||
974 in_epoch(net_epoch_preempt) ||
975 sx_xlocked(&allprison_lock));
976
977 pip = atomic_load_ptr(&pr->pr_addrs[af]);
978 if (__predict_false(pip == NULL))
979 return (EAFNOSUPPORT);
980
981 /* Check the primary IP. */
982 if (cmp(PR_IP(pip, af, 0), addr) == 0)
983 return (0);
984
985 /*
986 * All the other IPs are sorted so we can do a binary search.
987 */
988 a = 0;
989 z = pip->ips - 2;
990 while (a <= z) {
991 i = (a + z) / 2;
992 d = cmp(PR_IP(pip, af, i + 1), addr);
993 if (d > 0)
994 z = i - 1;
995 else if (d < 0)
996 a = i + 1;
997 else
998 return (0);
999 }
1000
1001 return (EADDRNOTAVAIL);
1002 }
1003
1004 /*
1005 * Grab primary IP. Historically required mutex, but nothing prevents
1006 * us to support epoch-protected access. Is it used in fast path?
1007 * in{6}_jail.c helper
1008 */
1009 const void *
prison_ip_get0(const struct prison * pr,const pr_family_t af)1010 prison_ip_get0(const struct prison *pr, const pr_family_t af)
1011 {
1012 const struct prison_ip *pip = pr->pr_addrs[af];
1013
1014 mtx_assert(&pr->pr_mtx, MA_OWNED);
1015 MPASS(pip);
1016
1017 return (pip->pr_ip);
1018 }
1019
1020 u_int
prison_ip_cnt(const struct prison * pr,const pr_family_t af)1021 prison_ip_cnt(const struct prison *pr, const pr_family_t af)
1022 {
1023
1024 return (pr->pr_addrs[af]->ips);
1025 }
1026 #endif /* defined(INET) || defined(INET6) */
1027
1028 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)1029 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
1030 {
1031 struct file *jfp_out;
1032 struct nameidata nd;
1033 #ifdef INET
1034 struct prison_ip *ip4;
1035 #endif
1036 #ifdef INET6
1037 struct prison_ip *ip6;
1038 #endif
1039 struct vfsopt *opt;
1040 struct vfsoptlist *opts;
1041 struct prison *pr, *deadpr, *dinspr, *inspr, *mypr, *ppr, *tpr;
1042 struct ucred *jdcred;
1043 struct vnode *root;
1044 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
1045 char *g_path, *osrelstr;
1046 struct bool_flags *bf;
1047 struct jailsys_flags *jsf;
1048 #if defined(INET) || defined(INET6)
1049 void *op;
1050 #endif
1051 unsigned long hid;
1052 size_t namelen, onamelen, pnamelen;
1053 #ifdef MAC
1054 void *mac_set_prison_data = NULL;
1055 int gotmaclabel;
1056 #endif
1057 int created, cuflags, descend, drflags, enforce;
1058 int error, errmsg_len, errmsg_pos;
1059 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
1060 int deadid, jfd_in, jfd_out, jfd_pos, jid, jsys, len, level;
1061 int childmax, osreldt, rsnum, slevel;
1062 #ifdef INET
1063 int ip4s;
1064 bool redo_ip4;
1065 #endif
1066 #ifdef INET6
1067 int ip6s;
1068 bool redo_ip6;
1069 #endif
1070 bool maybe_changed;
1071 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
1072 uint64_t pr_allow_diff;
1073 unsigned tallow;
1074 char numbuf[12];
1075
1076 mypr = td->td_ucred->cr_prison;
1077 if (((flags & (JAIL_CREATE | JAIL_AT_DESC)) == JAIL_CREATE) &&
1078 mypr->pr_childmax == 0)
1079 return (EPERM);
1080 if (flags & ~JAIL_SET_MASK)
1081 return (EINVAL);
1082 if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
1083 (JAIL_USE_DESC | JAIL_AT_DESC))
1084 return (EINVAL);
1085
1086 /* Only let a single thread in the process try to attach at a time. */
1087 if (flags & JAIL_ATTACH) {
1088 error = prison_attach_thread_single(td);
1089 if (error != 0)
1090 return (error);
1091 }
1092
1093 #ifdef INET
1094 ip4 = NULL;
1095 #endif
1096 #ifdef INET6
1097 ip6 = NULL;
1098 #endif
1099 g_path = NULL;
1100 jfp_out = NULL;
1101 jfd_out = -1;
1102 prison_hold(mypr);
1103 /*
1104 * Check all the parameters before committing to anything. Not all
1105 * errors can be caught early, but we may as well try. Also, this
1106 * takes care of some expensive stuff (path lookup) before getting
1107 * the allprison lock.
1108 *
1109 * XXX Jails are not filesystems, and jail parameters are not mount
1110 * options. But it makes more sense to re-use the vfsopt code
1111 * than duplicate it under a different name.
1112 */
1113 error = vfs_buildopts(optuio, &opts);
1114 if (error) {
1115 opts = NULL;
1116 goto done_free;
1117 }
1118
1119 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
1120 if (!cuflags) {
1121 error = EINVAL;
1122 vfs_opterror(opts, "no valid operation (create or update)");
1123 goto done_errmsg;
1124 }
1125
1126 error = vfs_copyopt(opts, JAIL_PARAM_DESC, &jfd_in, sizeof(jfd_in));
1127 if (error == ENOENT) {
1128 if (flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1129 JAIL_OWN_DESC)) {
1130 vfs_opterror(opts, "missing desc");
1131 goto done_errmsg;
1132 }
1133 jfd_in = -1;
1134 } else if (error != 0)
1135 goto done_free;
1136 else {
1137 if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
1138 JAIL_OWN_DESC))) {
1139 error = EINVAL;
1140 vfs_opterror(opts, "unexpected desc");
1141 goto done_errmsg;
1142 }
1143 if (flags & JAIL_AT_DESC) {
1144 /*
1145 * Look up and create jails based on the
1146 * descriptor's prison.
1147 */
1148 struct prison *jdpr;
1149
1150 error = jaildesc_find(td, jfd_in, &jdpr, NULL);
1151 if (error != 0) {
1152 vfs_opterror(opts, error == ENOENT ?
1153 "descriptor to dead jail" :
1154 "not a jail descriptor");
1155 goto done_errmsg;
1156 }
1157 prison_free(mypr);
1158 mypr = jdpr;
1159 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) {
1160 error = EPERM;
1161 goto done_free;
1162 }
1163 }
1164 if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
1165 /* Allocate a jail descriptor to return later. */
1166 error = jaildesc_alloc(td, &jfp_out, &jfd_out,
1167 flags & JAIL_OWN_DESC);
1168 if (error)
1169 goto done_free;
1170 }
1171 }
1172
1173 /*
1174 * Delay the permission check if using a jail descriptor,
1175 * until we get the descriptor's credentials.
1176 */
1177 if (!(flags & JAIL_USE_DESC)) {
1178 error = priv_check(td, PRIV_JAIL_SET);
1179 if (error == 0 && (flags & JAIL_ATTACH))
1180 error = priv_check(td, PRIV_JAIL_ATTACH);
1181 if (error)
1182 goto done_free;
1183 }
1184
1185 error = vfs_copyopt(opts, JAIL_PARAM_JID, &jid, sizeof(jid));
1186 if (error == ENOENT)
1187 jid = 0;
1188 else if (error != 0)
1189 goto done_free;
1190
1191 error = vfs_copyopt(opts, JAIL_PARAM_SECURELEVEL, &slevel, sizeof(slevel));
1192 if (error == ENOENT)
1193 gotslevel = 0;
1194 else if (error != 0)
1195 goto done_free;
1196 else
1197 gotslevel = 1;
1198
1199 error =
1200 vfs_copyopt(opts, JAIL_PARAM_CHILDREN_MAX, &childmax, sizeof(childmax));
1201 if (error == ENOENT)
1202 gotchildmax = 0;
1203 else if (error != 0)
1204 goto done_free;
1205 else
1206 gotchildmax = 1;
1207
1208 error = vfs_copyopt(opts, JAIL_PARAM_ENFORCE_STATFS, &enforce,
1209 sizeof(enforce));
1210 if (error == ENOENT)
1211 gotenforce = 0;
1212 else if (error != 0)
1213 goto done_free;
1214 else if (enforce < 0 || enforce > 2) {
1215 error = EINVAL;
1216 goto done_free;
1217 } else
1218 gotenforce = 1;
1219
1220 error = vfs_copyopt(opts, JAIL_PARAM_DEVFS_RULESET, &rsnum, sizeof(rsnum));
1221 if (error == ENOENT)
1222 gotrsnum = 0;
1223 else if (error != 0)
1224 goto done_free;
1225 else
1226 gotrsnum = 1;
1227
1228 pr_flags = ch_flags = 0;
1229 for (bf = pr_flag_bool;
1230 bf < pr_flag_bool + nitems(pr_flag_bool);
1231 bf++) {
1232 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
1233 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
1234 }
1235 ch_flags |= pr_flags;
1236 for (jsf = pr_flag_jailsys;
1237 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
1238 jsf++) {
1239 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
1240 if (error == ENOENT)
1241 continue;
1242 if (error != 0)
1243 goto done_free;
1244 switch (jsys) {
1245 case JAIL_SYS_DISABLE:
1246 if (!jsf->disable) {
1247 error = EINVAL;
1248 goto done_free;
1249 }
1250 pr_flags |= jsf->disable;
1251 break;
1252 case JAIL_SYS_NEW:
1253 pr_flags |= jsf->new;
1254 break;
1255 case JAIL_SYS_INHERIT:
1256 break;
1257 default:
1258 error = EINVAL;
1259 goto done_free;
1260 }
1261 ch_flags |= jsf->new | jsf->disable;
1262 }
1263 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
1264 && !(pr_flags & PR_PERSIST)) {
1265 error = EINVAL;
1266 vfs_opterror(opts, "new jail must persist or attach");
1267 goto done_errmsg;
1268 }
1269 #ifdef VIMAGE
1270 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
1271 error = EINVAL;
1272 vfs_opterror(opts, "vnet cannot be changed after creation");
1273 goto done_errmsg;
1274 }
1275 #endif
1276 #ifdef INET
1277 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
1278 error = EINVAL;
1279 vfs_opterror(opts, "ip4 cannot be changed after creation");
1280 goto done_errmsg;
1281 }
1282 #endif
1283 #ifdef INET6
1284 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
1285 error = EINVAL;
1286 vfs_opterror(opts, "ip6 cannot be changed after creation");
1287 goto done_errmsg;
1288 }
1289 #endif
1290
1291 pr_allow = ch_allow = 0;
1292 for (bf = pr_flag_allow;
1293 bf < pr_flag_allow + nitems(pr_flag_allow) &&
1294 atomic_load_int(&bf->flag) != 0;
1295 bf++) {
1296 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
1297 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
1298 }
1299 ch_allow |= pr_allow;
1300
1301 error = vfs_getopt(opts, JAIL_PARAM_NAME, (void **)&name, &len);
1302 if (error == ENOENT)
1303 name = NULL;
1304 else if (error != 0)
1305 goto done_free;
1306 else {
1307 if (len == 0 || name[len - 1] != '\0') {
1308 error = EINVAL;
1309 goto done_free;
1310 }
1311 if (len > MAXHOSTNAMELEN) {
1312 error = ENAMETOOLONG;
1313 goto done_free;
1314 }
1315 }
1316
1317 error = vfs_getopt(opts, JAIL_PARAM_HOST_HOSTNAME, (void **)&host, &len);
1318 if (error == ENOENT)
1319 host = NULL;
1320 else if (error != 0)
1321 goto done_free;
1322 else {
1323 ch_flags |= PR_HOST;
1324 pr_flags |= PR_HOST;
1325 if (len == 0 || host[len - 1] != '\0') {
1326 error = EINVAL;
1327 goto done_free;
1328 }
1329 if (len > MAXHOSTNAMELEN) {
1330 error = ENAMETOOLONG;
1331 goto done_free;
1332 }
1333 }
1334
1335 error = vfs_getopt(opts, JAIL_PARAM_HOST_DOMAINNAME, (void **)&domain, &len);
1336 if (error == ENOENT)
1337 domain = NULL;
1338 else if (error != 0)
1339 goto done_free;
1340 else {
1341 ch_flags |= PR_HOST;
1342 pr_flags |= PR_HOST;
1343 if (len == 0 || domain[len - 1] != '\0') {
1344 error = EINVAL;
1345 goto done_free;
1346 }
1347 if (len > MAXHOSTNAMELEN) {
1348 error = ENAMETOOLONG;
1349 goto done_free;
1350 }
1351 }
1352
1353 error = vfs_getopt(opts, JAIL_PARAM_HOST_HOSTUUID, (void **)&uuid, &len);
1354 if (error == ENOENT)
1355 uuid = NULL;
1356 else if (error != 0)
1357 goto done_free;
1358 else {
1359 ch_flags |= PR_HOST;
1360 pr_flags |= PR_HOST;
1361 if (len == 0 || uuid[len - 1] != '\0') {
1362 error = EINVAL;
1363 goto done_free;
1364 }
1365 if (len > HOSTUUIDLEN) {
1366 error = ENAMETOOLONG;
1367 goto done_free;
1368 }
1369 }
1370
1371 #ifdef COMPAT_FREEBSD32
1372 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1373 uint32_t hid32;
1374
1375 error = vfs_copyopt(opts, JAIL_PARAM_HOST_HOSTID, &hid32, sizeof(hid32));
1376 hid = hid32;
1377 } else
1378 #endif
1379 error = vfs_copyopt(opts, JAIL_PARAM_HOST_HOSTID, &hid, sizeof(hid));
1380 if (error == ENOENT)
1381 gothid = 0;
1382 else if (error != 0)
1383 goto done_free;
1384 else {
1385 gothid = 1;
1386 ch_flags |= PR_HOST;
1387 pr_flags |= PR_HOST;
1388 }
1389
1390 #ifdef MAC
1391 /* Process the mac.label vfsopt */
1392 error = mac_set_prison_prepare(td, opts, &mac_set_prison_data);
1393 if (error == ENOENT)
1394 gotmaclabel = 0;
1395 else if (error != 0)
1396 goto done_errmsg;
1397 else
1398 gotmaclabel = 1;
1399 #endif
1400
1401 #ifdef INET
1402 error = vfs_getopt(opts, JAIL_PARAM_IP4_ADDR, &op, &ip4s);
1403 if (error == ENOENT)
1404 ip4s = 0;
1405 else if (error != 0)
1406 goto done_free;
1407 else if (ip4s & (sizeof(struct in_addr) - 1)) {
1408 error = EINVAL;
1409 goto done_free;
1410 } else {
1411 ch_flags |= PR_IP4_USER;
1412 pr_flags |= PR_IP4_USER;
1413 if (ip4s > 0) {
1414 ip4s /= sizeof(struct in_addr);
1415 if (ip4s > jail_max_af_ips) {
1416 error = EINVAL;
1417 vfs_opterror(opts, "too many IPv4 addresses");
1418 goto done_errmsg;
1419 }
1420 ip4 = prison_ip_copyin(PR_INET, op, ip4s);
1421 if (ip4 == NULL) {
1422 error = EINVAL;
1423 goto done_free;
1424 }
1425 }
1426 }
1427 #endif
1428
1429 #ifdef INET6
1430 error = vfs_getopt(opts, JAIL_PARAM_IP6_ADDR, &op, &ip6s);
1431 if (error == ENOENT)
1432 ip6s = 0;
1433 else if (error != 0)
1434 goto done_free;
1435 else if (ip6s & (sizeof(struct in6_addr) - 1)) {
1436 error = EINVAL;
1437 goto done_free;
1438 } else {
1439 ch_flags |= PR_IP6_USER;
1440 pr_flags |= PR_IP6_USER;
1441 if (ip6s > 0) {
1442 ip6s /= sizeof(struct in6_addr);
1443 if (ip6s > jail_max_af_ips) {
1444 error = EINVAL;
1445 vfs_opterror(opts, "too many IPv6 addresses");
1446 goto done_errmsg;
1447 }
1448 ip6 = prison_ip_copyin(PR_INET6, op, ip6s);
1449 if (ip6 == NULL) {
1450 error = EINVAL;
1451 goto done_free;
1452 }
1453 }
1454 }
1455 #endif
1456
1457 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1458 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1459 error = EINVAL;
1460 vfs_opterror(opts,
1461 "vnet jails cannot have IP address restrictions");
1462 goto done_errmsg;
1463 }
1464 #endif
1465
1466 error = vfs_getopt(opts, JAIL_PARAM_OSRELEASE, (void **)&osrelstr, &len);
1467 if (error == ENOENT)
1468 osrelstr = NULL;
1469 else if (error != 0)
1470 goto done_free;
1471 else {
1472 if (flags & JAIL_UPDATE) {
1473 error = EINVAL;
1474 vfs_opterror(opts,
1475 "osrelease cannot be changed after creation");
1476 goto done_errmsg;
1477 }
1478 if (len == 0 || osrelstr[len - 1] != '\0') {
1479 error = EINVAL;
1480 goto done_free;
1481 }
1482 if (len >= OSRELEASELEN) {
1483 error = ENAMETOOLONG;
1484 vfs_opterror(opts,
1485 "osrelease string must be 1-%d bytes long",
1486 OSRELEASELEN - 1);
1487 goto done_errmsg;
1488 }
1489 }
1490
1491 error = vfs_copyopt(opts, JAIL_PARAM_OSRELDATE, &osreldt, sizeof(osreldt));
1492 if (error == ENOENT)
1493 osreldt = 0;
1494 else if (error != 0)
1495 goto done_free;
1496 else {
1497 if (flags & JAIL_UPDATE) {
1498 error = EINVAL;
1499 vfs_opterror(opts,
1500 "osreldate cannot be changed after creation");
1501 goto done_errmsg;
1502 }
1503 if (osreldt == 0) {
1504 error = EINVAL;
1505 vfs_opterror(opts, "osreldate cannot be 0");
1506 goto done_errmsg;
1507 }
1508 }
1509
1510 root = NULL;
1511 error = vfs_getopt(opts, JAIL_PARAM_PATH, (void **)&path, &len);
1512 if (error == ENOENT)
1513 path = NULL;
1514 else if (error != 0)
1515 goto done_free;
1516 else {
1517 if (flags & JAIL_UPDATE) {
1518 error = EINVAL;
1519 vfs_opterror(opts,
1520 "path cannot be changed after creation");
1521 goto done_errmsg;
1522 }
1523 if (len == 0 || path[len - 1] != '\0') {
1524 error = EINVAL;
1525 goto done_free;
1526 }
1527 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path);
1528 error = namei(&nd);
1529 if (error)
1530 goto done_free;
1531 root = nd.ni_vp;
1532 NDFREE_PNBUF(&nd);
1533 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1534 strlcpy(g_path, path, MAXPATHLEN);
1535 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
1536 if (error == 0) {
1537 path = g_path;
1538 } else {
1539 /* exit on other errors */
1540 goto done_free;
1541 }
1542 if (root->v_type != VDIR) {
1543 error = ENOTDIR;
1544 vput(root);
1545 goto done_free;
1546 }
1547 VOP_UNLOCK(root);
1548 }
1549
1550 /*
1551 * Find the specified jail, or at least its parent.
1552 * This abuses the file error codes ENOENT and EEXIST.
1553 */
1554 pr = NULL;
1555 inspr = NULL;
1556 deadpr = NULL;
1557 maybe_changed = false;
1558 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1559 namelc = strrchr(name, '.');
1560 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1561 if (*p != '\0')
1562 jid = 0;
1563 }
1564 sx_xlock(&allprison_lock);
1565 drflags = PD_LIST_XLOCKED;
1566 ppr = mypr;
1567 if (!prison_isalive(ppr)) {
1568 /* This jail is dying. This process will surely follow. */
1569 error = EAGAIN;
1570 goto done_deref;
1571 }
1572 if (flags & JAIL_USE_DESC) {
1573 /* Get the jail from its descriptor. */
1574 error = jaildesc_find(td, jfd_in, &pr, &jdcred);
1575 if (error) {
1576 vfs_opterror(opts, error == ENOENT ?
1577 "descriptor to dead jail" :
1578 "not a jail descriptor");
1579 goto done_deref;
1580 }
1581 drflags |= PD_DEREF;
1582 error = priv_check_cred(jdcred, PRIV_JAIL_SET);
1583 if (error == 0 && (flags & JAIL_ATTACH))
1584 error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
1585 crfree(jdcred);
1586 if (error)
1587 goto done_deref;
1588 mtx_lock(&pr->pr_mtx);
1589 drflags |= PD_LOCKED;
1590 if (cuflags == JAIL_CREATE) {
1591 error = EEXIST;
1592 vfs_opterror(opts, "jail %d already exists",
1593 pr->pr_id);
1594 goto done_deref;
1595 }
1596 if (!prison_isalive(pr)) {
1597 /* While a jid can be resurrected, the prison
1598 * itself cannot.
1599 */
1600 error = ENOENT;
1601 vfs_opterror(opts, "jail %d is dying", pr->pr_id);
1602 goto done_deref;
1603 }
1604 if (jid != 0 && jid != pr->pr_id) {
1605 error = EINVAL;
1606 vfs_opterror(opts, "cannot change jid");
1607 goto done_deref;
1608 }
1609 jid = pr->pr_id;
1610 } else if (jid != 0) {
1611 if (jid < 0) {
1612 error = EINVAL;
1613 vfs_opterror(opts, "negative jid");
1614 goto done_deref;
1615 }
1616 /*
1617 * See if a requested jid already exists. Keep track of
1618 * where it can be inserted later.
1619 */
1620 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1621 if (inspr->pr_id < jid)
1622 continue;
1623 if (inspr->pr_id > jid)
1624 break;
1625 if (prison_isalive(inspr)) {
1626 pr = inspr;
1627 mtx_lock(&pr->pr_mtx);
1628 drflags |= PD_LOCKED;
1629 } else {
1630 /* Note a dying jail to handle later. */
1631 deadpr = inspr;
1632 }
1633 inspr = NULL;
1634 break;
1635 }
1636 if (cuflags == JAIL_CREATE && pr != NULL) {
1637 /*
1638 * Even creators that cannot see the jail will
1639 * get EEXIST.
1640 */
1641 error = EEXIST;
1642 vfs_opterror(opts, "jail %d already exists", jid);
1643 goto done_deref;
1644 }
1645 if ((pr == NULL)
1646 ? cuflags == JAIL_UPDATE
1647 : !prison_ischild(mypr, pr)) {
1648 /*
1649 * Updaters get ENOENT for nonexistent jails,
1650 * or for jails they cannot see. The latter
1651 * case is true even for CREATE | UPDATE,
1652 * which normally cannot give this error.
1653 */
1654 error = ENOENT;
1655 vfs_opterror(opts, "jail %d not found", jid);
1656 goto done_deref;
1657 }
1658 }
1659 /*
1660 * If the caller provided a name, look for a jail by that name.
1661 * This has different semantics for creates and updates keyed by jid
1662 * (where the name must not already exist in a different jail),
1663 * and updates keyed by the name itself (where the name must exist
1664 * because that is the jail being updated).
1665 */
1666 namelc = NULL;
1667 if (name != NULL) {
1668 namelc = strrchr(name, '.');
1669 if (namelc == NULL)
1670 namelc = name;
1671 else {
1672 /*
1673 * This is a hierarchical name. Split it into the
1674 * parent and child names, and make sure the parent
1675 * exists or matches an already found jail.
1676 */
1677 if (pr != NULL) {
1678 if (strncmp(name, ppr->pr_name, namelc - name)
1679 || ppr->pr_name[namelc - name] != '\0') {
1680 error = EINVAL;
1681 vfs_opterror(opts,
1682 "cannot change jail's parent");
1683 goto done_deref;
1684 }
1685 } else {
1686 *namelc = '\0';
1687 ppr = prison_find_name(mypr, name);
1688 if (ppr == NULL) {
1689 error = ENOENT;
1690 vfs_opterror(opts,
1691 "jail \"%s\" not found", name);
1692 goto done_deref;
1693 }
1694 mtx_unlock(&ppr->pr_mtx);
1695 if (!prison_isalive(ppr)) {
1696 error = ENOENT;
1697 vfs_opterror(opts,
1698 "jail \"%s\" is dying", name);
1699 goto done_deref;
1700 }
1701 *namelc = '.';
1702 }
1703 namelc++;
1704 }
1705 if (namelc[0] != '\0') {
1706 pnamelen =
1707 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1708 FOREACH_PRISON_CHILD(ppr, tpr) {
1709 if (tpr == pr || !prison_isalive(tpr) ||
1710 strcmp(tpr->pr_name + pnamelen, namelc))
1711 continue;
1712 if (cuflags == JAIL_CREATE || pr != NULL) {
1713 /*
1714 * Create, or update(jid): name must
1715 * not exist in an active sibling jail.
1716 */
1717 error = EEXIST;
1718 vfs_opterror(opts,
1719 "jail \"%s\" already exists", name);
1720 goto done_deref;
1721 }
1722 /* Use this jail for updates. */
1723 pr = tpr;
1724 mtx_lock(&pr->pr_mtx);
1725 drflags |= PD_LOCKED;
1726 break;
1727 }
1728 /*
1729 * Update: name must exist if no jid is specified.
1730 * As with the jid case, the jail must be currently
1731 * visible, or else even CREATE | UPDATE will get
1732 * an error.
1733 */
1734 if ((pr == NULL)
1735 ? cuflags == JAIL_UPDATE
1736 : !prison_isalive(pr)) {
1737 error = ENOENT;
1738 vfs_opterror(opts, "jail \"%s\" not found",
1739 name);
1740 goto done_deref;
1741 }
1742 }
1743 }
1744 /* Update: must provide a desc, jid, or name. */
1745 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1746 error = ENOENT;
1747 vfs_opterror(opts, "update specified no jail");
1748 goto done_deref;
1749 }
1750
1751 /* If there's no prison to update, create a new one and link it in. */
1752 created = pr == NULL;
1753 if (created) {
1754 #ifdef MAC
1755 error = mac_prison_check_create(td->td_ucred, opts, flags);
1756 if (error != 0)
1757 goto done_deref;
1758 #endif
1759 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1760 if (tpr->pr_childcount >= tpr->pr_childmax) {
1761 error = EPERM;
1762 vfs_opterror(opts, "prison limit exceeded");
1763 goto done_deref;
1764 }
1765
1766 if (deadpr != NULL) {
1767 /*
1768 * The prison being created has the same ID as a dying
1769 * one. Handle this by giving the dying jail a new ID.
1770 * This may cause some confusion to user space, but
1771 * only to those listing dying jails.
1772 */
1773 deadid = get_next_deadid(&dinspr);
1774 if (deadid == 0) {
1775 error = EAGAIN;
1776 vfs_opterror(opts, "no available jail IDs");
1777 goto done_deref;
1778 }
1779 mtx_lock(&deadpr->pr_mtx);
1780 deadpr->pr_id = deadid;
1781 mtx_unlock(&deadpr->pr_mtx);
1782 if (dinspr == deadpr)
1783 inspr = deadpr;
1784 else {
1785 inspr = TAILQ_NEXT(deadpr, pr_list);
1786 TAILQ_REMOVE(&allprison, deadpr, pr_list);
1787 if (dinspr != NULL)
1788 TAILQ_INSERT_AFTER(&allprison, dinspr,
1789 deadpr, pr_list);
1790 else
1791 TAILQ_INSERT_HEAD(&allprison, deadpr,
1792 pr_list);
1793 }
1794 }
1795 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1796 error = EAGAIN;
1797 vfs_opterror(opts, "no available jail IDs");
1798 goto done_deref;
1799 }
1800
1801 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1802 pr->pr_state = PRISON_STATE_INVALID;
1803 refcount_init(&pr->pr_ref, 1);
1804 refcount_init(&pr->pr_uref, 0);
1805 drflags |= PD_DEREF;
1806 LIST_INIT(&pr->pr_children);
1807 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1808 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1809
1810 pr->pr_id = jid;
1811 if (inspr != NULL)
1812 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1813 else
1814 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1815
1816 pr->pr_parent = ppr;
1817 prison_hold(ppr);
1818 prison_proc_hold(ppr);
1819 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1820 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1821 tpr->pr_childcount++;
1822 pr->pr_klist = knlist_alloc(&pr->pr_mtx);
1823
1824 /* Set some default values, and inherit some from the parent. */
1825 if (namelc == NULL)
1826 namelc = "";
1827 if (path == NULL) {
1828 path = "/";
1829 root = ppr->pr_root;
1830 vref(root);
1831 }
1832 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1833 pr->pr_flags |= PR_HOST;
1834 #if defined(INET) || defined(INET6)
1835 #ifdef VIMAGE
1836 if (!(pr_flags & PR_VNET))
1837 #endif
1838 {
1839 #ifdef INET
1840 if (!(ch_flags & PR_IP4_USER))
1841 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1842 else if (!(pr_flags & PR_IP4_USER)) {
1843 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1844 prison_ip_dup(ppr, pr, PR_INET);
1845 }
1846 #endif
1847 #ifdef INET6
1848 if (!(ch_flags & PR_IP6_USER))
1849 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1850 else if (!(pr_flags & PR_IP6_USER)) {
1851 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1852 prison_ip_dup(ppr, pr, PR_INET6);
1853 }
1854 #endif
1855 }
1856 #endif
1857 /* Source address selection is always on by default. */
1858 pr->pr_flags |= _PR_IP_SADDRSEL;
1859
1860 pr->pr_securelevel = ppr->pr_securelevel;
1861 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1862 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1863 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1864
1865 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1866 if (osrelstr == NULL)
1867 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1868 sizeof(pr->pr_osrelease));
1869 else
1870 strlcpy(pr->pr_osrelease, osrelstr,
1871 sizeof(pr->pr_osrelease));
1872
1873 #ifdef VIMAGE
1874 /*
1875 * Allocate a new vnet if specified.
1876 *
1877 * Set PR_VNET now if so, so that the vnet is disposed of
1878 * properly when the jail is destroyed.
1879 */
1880 if (pr_flags & PR_VNET) {
1881 pr->pr_flags |= PR_VNET;
1882 pr->pr_vnet = vnet_alloc();
1883 } else {
1884 pr->pr_vnet = ppr->pr_vnet;
1885 }
1886 #endif
1887 /*
1888 * Allocate a dedicated cpuset for each jail.
1889 * Unlike other initial settings, this may return an error.
1890 */
1891 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1892 if (error)
1893 goto done_deref;
1894
1895 #ifdef MAC
1896 error = mac_prison_init(pr, M_WAITOK);
1897 MPASS(error == 0);
1898
1899 mtx_assert(&pr->pr_mtx, MA_OWNED);
1900 #else
1901 mtx_lock(&pr->pr_mtx);
1902 #endif
1903 drflags |= PD_LOCKED;
1904 } else {
1905 /*
1906 * Grab a reference for existing prisons, to ensure they
1907 * continue to exist for the duration of the call.
1908 */
1909 if (!(drflags & PD_DEREF)) {
1910 prison_hold(pr);
1911 drflags |= PD_DEREF;
1912 }
1913 #ifdef MAC
1914 error = mac_prison_check_set(td->td_ucred, pr, opts, flags);
1915 if (error != 0)
1916 goto done_deref;
1917 #endif
1918 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1919 if ((pr->pr_flags & PR_VNET) &&
1920 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1921 error = EINVAL;
1922 vfs_opterror(opts,
1923 "vnet jails cannot have IP address restrictions");
1924 goto done_deref;
1925 }
1926 #endif
1927 #ifdef INET
1928 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1929 error = EINVAL;
1930 vfs_opterror(opts,
1931 "ip4 cannot be changed after creation");
1932 goto done_deref;
1933 }
1934 #endif
1935 #ifdef INET6
1936 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1937 error = EINVAL;
1938 vfs_opterror(opts,
1939 "ip6 cannot be changed after creation");
1940 goto done_deref;
1941 }
1942 #endif
1943 }
1944
1945 /* Do final error checking before setting anything. */
1946 if (gotslevel) {
1947 if (slevel < ppr->pr_securelevel) {
1948 error = EPERM;
1949 goto done_deref;
1950 }
1951 }
1952 if (gotchildmax) {
1953 if (childmax >= ppr->pr_childmax) {
1954 error = EPERM;
1955 goto done_deref;
1956 }
1957 }
1958 if (gotenforce) {
1959 if (enforce < ppr->pr_enforce_statfs) {
1960 error = EPERM;
1961 goto done_deref;
1962 }
1963 }
1964 if (gotrsnum) {
1965 /*
1966 * devfs_rsnum is a uint16_t
1967 */
1968 if (rsnum < 0 || rsnum > 65535) {
1969 error = EINVAL;
1970 goto done_deref;
1971 }
1972 /*
1973 * Nested jails always inherit parent's devfs ruleset
1974 */
1975 if (jailed(td->td_ucred)) {
1976 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1977 error = EPERM;
1978 goto done_deref;
1979 } else
1980 rsnum = ppr->pr_devfs_rsnum;
1981 }
1982 }
1983 #ifdef INET
1984 if (ip4s > 0) {
1985 if ((ppr->pr_flags & PR_IP4) &&
1986 !prison_ip_parent_match(ppr->pr_addrs[PR_INET], ip4,
1987 PR_INET)) {
1988 error = EPERM;
1989 goto done_deref;
1990 }
1991 if (!prison_ip_conflict_check(ppr, pr, ip4, PR_INET)) {
1992 error = EADDRINUSE;
1993 vfs_opterror(opts, "IPv4 addresses clash");
1994 goto done_deref;
1995 }
1996 }
1997 #endif
1998 #ifdef INET6
1999 if (ip6s > 0) {
2000 if ((ppr->pr_flags & PR_IP6) &&
2001 !prison_ip_parent_match(ppr->pr_addrs[PR_INET6], ip6,
2002 PR_INET6)) {
2003 error = EPERM;
2004 goto done_deref;
2005 }
2006 if (!prison_ip_conflict_check(ppr, pr, ip6, PR_INET6)) {
2007 error = EADDRINUSE;
2008 vfs_opterror(opts, "IPv6 addresses clash");
2009 goto done_deref;
2010 }
2011 }
2012 #endif
2013 onamelen = namelen = 0;
2014 if (namelc != NULL) {
2015 /* Give a default name of the jid. Also allow the name to be
2016 * explicitly the jid - but not any other number, and only in
2017 * normal form (no leading zero/etc).
2018 */
2019 if (namelc[0] == '\0')
2020 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
2021 else if ((strtoul(namelc, &p, 10) != jid ||
2022 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
2023 error = EINVAL;
2024 vfs_opterror(opts,
2025 "name cannot be numeric (unless it is the jid)");
2026 goto done_deref;
2027 }
2028 /*
2029 * Make sure the name isn't too long for the prison or its
2030 * children.
2031 */
2032 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
2033 onamelen = strlen(pr->pr_name + pnamelen);
2034 namelen = strlen(namelc);
2035 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
2036 error = ENAMETOOLONG;
2037 goto done_deref;
2038 }
2039 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
2040 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
2041 sizeof(pr->pr_name)) {
2042 error = ENAMETOOLONG;
2043 goto done_deref;
2044 }
2045 }
2046 }
2047 pr_allow_diff = pr_allow & ~ppr->pr_allow;
2048 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
2049 error = EPERM;
2050 goto done_deref;
2051 }
2052
2053 /*
2054 * Let modules check their parameters. This requires unlocking and
2055 * then re-locking the prison, but this is still a valid state as long
2056 * as allprison_lock remains xlocked.
2057 */
2058 mtx_unlock(&pr->pr_mtx);
2059 drflags &= ~PD_LOCKED;
2060 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
2061 if (error != 0)
2062 goto done_deref;
2063 mtx_lock(&pr->pr_mtx);
2064 drflags |= PD_LOCKED;
2065
2066 /* At this point, all valid parameters should have been noted. */
2067 TAILQ_FOREACH(opt, opts, link) {
2068 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2069 error = EINVAL;
2070 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2071 goto done_deref;
2072 }
2073 }
2074 maybe_changed = true;
2075
2076 /* Set the parameters of the prison. */
2077 #ifdef INET
2078 redo_ip4 = false;
2079 if (pr_flags & PR_IP4_USER) {
2080 pr->pr_flags |= PR_IP4;
2081 prison_ip_set(pr, PR_INET, ip4);
2082 ip4 = NULL;
2083 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2084 #ifdef VIMAGE
2085 if (tpr->pr_flags & PR_VNET) {
2086 descend = 0;
2087 continue;
2088 }
2089 #endif
2090 if (!prison_ip_restrict(tpr, PR_INET, NULL)) {
2091 redo_ip4 = true;
2092 descend = 0;
2093 }
2094 }
2095 }
2096 #endif
2097 #ifdef INET6
2098 redo_ip6 = false;
2099 if (pr_flags & PR_IP6_USER) {
2100 pr->pr_flags |= PR_IP6;
2101 prison_ip_set(pr, PR_INET6, ip6);
2102 ip6 = NULL;
2103 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2104 #ifdef VIMAGE
2105 if (tpr->pr_flags & PR_VNET) {
2106 descend = 0;
2107 continue;
2108 }
2109 #endif
2110 if (!prison_ip_restrict(tpr, PR_INET6, NULL)) {
2111 redo_ip6 = true;
2112 descend = 0;
2113 }
2114 }
2115 }
2116 #endif
2117 if (gotslevel) {
2118 pr->pr_securelevel = slevel;
2119 /* Set all child jails to be at least this level. */
2120 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2121 if (tpr->pr_securelevel < slevel)
2122 tpr->pr_securelevel = slevel;
2123 }
2124 if (gotchildmax) {
2125 pr->pr_childmax = childmax;
2126 /* Set all child jails to under this limit. */
2127 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
2128 if (tpr->pr_childmax > childmax - level)
2129 tpr->pr_childmax = childmax > level
2130 ? childmax - level : 0;
2131 }
2132 if (gotenforce) {
2133 pr->pr_enforce_statfs = enforce;
2134 /* Pass this restriction on to the children. */
2135 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2136 if (tpr->pr_enforce_statfs < enforce)
2137 tpr->pr_enforce_statfs = enforce;
2138 }
2139 if (gotrsnum) {
2140 pr->pr_devfs_rsnum = rsnum;
2141 /* Pass this restriction on to the children. */
2142 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
2143 tpr->pr_devfs_rsnum = rsnum;
2144 }
2145 if (namelc != NULL) {
2146 if (ppr == &prison0)
2147 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
2148 else
2149 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
2150 ppr->pr_name, namelc);
2151 /* Change this component of child names. */
2152 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2153 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
2154 strlen(tpr->pr_name + onamelen) + 1);
2155 bcopy(pr->pr_name, tpr->pr_name, namelen);
2156 }
2157 }
2158 if (path != NULL) {
2159 /* Try to keep a real-rooted full pathname. */
2160 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
2161 pr->pr_root = root;
2162 root = NULL;
2163 }
2164 if (PR_HOST & ch_flags & ~pr_flags) {
2165 if (pr->pr_flags & PR_HOST) {
2166 /*
2167 * Copy the parent's host info. As with pr_ip4 above,
2168 * the lack of a lock on the parent is not a problem;
2169 * it is always set with allprison_lock at least
2170 * shared, and is held exclusively here.
2171 */
2172 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
2173 sizeof(pr->pr_hostname));
2174 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
2175 sizeof(pr->pr_domainname));
2176 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
2177 sizeof(pr->pr_hostuuid));
2178 pr->pr_hostid = pr->pr_parent->pr_hostid;
2179 }
2180 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
2181 /* Set this prison, and any descendants without PR_HOST. */
2182 if (host != NULL)
2183 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
2184 if (domain != NULL)
2185 strlcpy(pr->pr_domainname, domain,
2186 sizeof(pr->pr_domainname));
2187 if (uuid != NULL)
2188 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
2189 if (gothid)
2190 pr->pr_hostid = hid;
2191 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2192 if (tpr->pr_flags & PR_HOST)
2193 descend = 0;
2194 else {
2195 if (host != NULL)
2196 strlcpy(tpr->pr_hostname,
2197 pr->pr_hostname,
2198 sizeof(tpr->pr_hostname));
2199 if (domain != NULL)
2200 strlcpy(tpr->pr_domainname,
2201 pr->pr_domainname,
2202 sizeof(tpr->pr_domainname));
2203 if (uuid != NULL)
2204 strlcpy(tpr->pr_hostuuid,
2205 pr->pr_hostuuid,
2206 sizeof(tpr->pr_hostuuid));
2207 if (gothid)
2208 tpr->pr_hostid = hid;
2209 }
2210 }
2211 }
2212 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
2213 if ((tallow = ch_allow & ~pr_allow))
2214 prison_set_allow_locked(pr, tallow, 0);
2215 /*
2216 * Persistent prisons get an extra reference, and prisons losing their
2217 * persist flag lose that reference.
2218 */
2219 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
2220 if (pr_flags & PR_PERSIST) {
2221 prison_hold(pr);
2222 /*
2223 * This may be a new prison's first user reference,
2224 * but wait to call it alive until after OSD calls
2225 * have had a chance to run (and perhaps to fail).
2226 */
2227 refcount_acquire(&pr->pr_uref);
2228 } else {
2229 drflags |= PD_DEUREF;
2230 prison_free_not_last(pr);
2231 }
2232 }
2233 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
2234
2235 #ifdef MAC
2236 /* Apply any request MAC label before we let modules do their work. */
2237 if (gotmaclabel) {
2238 error = mac_set_prison_core(td, pr, mac_set_prison_data);
2239 if (error) {
2240 vfs_opterror(opts, "mac relabel denied");
2241 goto done_deref;
2242 }
2243 }
2244 #endif
2245 mtx_unlock(&pr->pr_mtx);
2246 drflags &= ~PD_LOCKED;
2247 /*
2248 * Any errors past this point will need to de-persist newly created
2249 * prisons, as well as call remove methods.
2250 */
2251 if (created)
2252 drflags |= PD_KILL;
2253
2254 #ifdef RACCT
2255 if (racct_enable && created)
2256 prison_racct_attach(pr);
2257 #endif
2258
2259 /* Locks may have prevented a complete restriction of child IP
2260 * addresses. If so, allocate some more memory and try again.
2261 */
2262 #ifdef INET
2263 while (redo_ip4) {
2264 ip4s = pr->pr_addrs[PR_INET]->ips;
2265 MPASS(ip4 == NULL);
2266 ip4 = prison_ip_alloc(PR_INET, ip4s, M_WAITOK);
2267 mtx_lock(&pr->pr_mtx);
2268 redo_ip4 = false;
2269 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2270 #ifdef VIMAGE
2271 if (tpr->pr_flags & PR_VNET) {
2272 descend = 0;
2273 continue;
2274 }
2275 #endif
2276 if (!prison_ip_restrict(tpr, PR_INET, &ip4))
2277 redo_ip4 = true;
2278 }
2279 mtx_unlock(&pr->pr_mtx);
2280 }
2281 #endif
2282 #ifdef INET6
2283 while (redo_ip6) {
2284 ip6s = pr->pr_addrs[PR_INET6]->ips;
2285 MPASS(ip6 == NULL);
2286 ip6 = prison_ip_alloc(PR_INET6, ip6s, M_WAITOK);
2287 mtx_lock(&pr->pr_mtx);
2288 redo_ip6 = false;
2289 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
2290 #ifdef VIMAGE
2291 if (tpr->pr_flags & PR_VNET) {
2292 descend = 0;
2293 continue;
2294 }
2295 #endif
2296 if (!prison_ip_restrict(tpr, PR_INET6, &ip6))
2297 redo_ip6 = true;
2298 }
2299 mtx_unlock(&pr->pr_mtx);
2300 }
2301 #endif
2302
2303 /* Let the modules do their work. */
2304 if (created) {
2305 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
2306 if (error)
2307 goto done_deref;
2308 }
2309 error = osd_jail_call(pr, PR_METHOD_SET, opts);
2310 if (error)
2311 goto done_deref;
2312
2313 /*
2314 * A new prison is now ready to be seen; either it has gained a user
2315 * reference via persistence, or is about to gain one via attachment.
2316 */
2317 if (created) {
2318 sx_assert(&allprison_lock, SX_XLOCKED);
2319 prison_knote(ppr, NOTE_JAIL_CHILD | pr->pr_id);
2320 #ifdef MAC
2321 /*
2322 * Note that mac_prison_created() assumes that it's called in a
2323 * sleepable context.
2324 */
2325 mac_prison_created(td->td_ucred, pr);
2326 #endif
2327 mtx_lock(&pr->pr_mtx);
2328 drflags |= PD_LOCKED;
2329 pr->pr_state = PRISON_STATE_ALIVE;
2330 }
2331
2332 /*
2333 * Attach this process to the prison if requested. This will
2334 * unlock allprison_lock, meaning changes are now user-visible.
2335 */
2336 if (flags & JAIL_ATTACH) {
2337 #ifdef MAC
2338 error = mac_prison_check_attach(td->td_ucred, pr);
2339 if (error != 0) {
2340 vfs_opterror(opts,
2341 "attach operation denied by MAC policy");
2342 goto done_deref;
2343 }
2344 #endif
2345 error = do_jail_attach(td, pr, &drflags);
2346 if (error) {
2347 vfs_opterror(opts, "attach failed");
2348 goto done_deref;
2349 }
2350 }
2351
2352 #ifdef RACCT
2353 if (racct_enable && !created) {
2354 if (drflags & PD_LOCKED) {
2355 mtx_unlock(&pr->pr_mtx);
2356 drflags &= ~PD_LOCKED;
2357 }
2358 if (drflags & PD_LIST_XLOCKED) {
2359 sx_xunlock(&allprison_lock);
2360 drflags &= ~PD_LIST_XLOCKED;
2361 }
2362 prison_racct_modify(pr);
2363 }
2364 #endif
2365
2366 if (created && pr != &prison0 && (pr->pr_allow & PR_ALLOW_NFSD) != 0 &&
2367 (pr->pr_root->v_vflag & VV_ROOT) == 0)
2368 printf("Warning jail jid=%d: mountd/nfsd requires a separate"
2369 " file system\n", pr->pr_id);
2370
2371 /*
2372 * Now that the prison is fully created without error, set the
2373 * jail descriptor if one was requested. This is the only
2374 * parameter that is returned to the caller (except the error
2375 * message).
2376 */
2377 if (jfd_out >= 0) {
2378 if (!(drflags & PD_LOCKED)) {
2379 mtx_lock(&pr->pr_mtx);
2380 drflags |= PD_LOCKED;
2381 }
2382 jfd_pos = 2 * vfs_getopt_pos(opts, JAIL_PARAM_DESC) + 1;
2383 if (optuio->uio_segflg == UIO_SYSSPACE)
2384 *(int*)optuio->uio_iov[jfd_pos].iov_base = jfd_out;
2385 else
2386 (void)copyout(&jfd_out,
2387 optuio->uio_iov[jfd_pos].iov_base, sizeof(jfd_out));
2388 jaildesc_set_prison(jfp_out, pr);
2389 }
2390
2391 drflags &= ~PD_KILL;
2392 td->td_retval[0] = pr->pr_id;
2393
2394 done_deref:
2395 /*
2396 * Report changes to kevent. This can happen even if the
2397 * system call fails, as changes might have been made before
2398 * the failure.
2399 */
2400 if (maybe_changed && !created)
2401 prison_knote(pr, NOTE_JAIL_SET);
2402 /* Release any temporary prison holds and/or locks. */
2403 if (pr != NULL)
2404 prison_deref(pr, drflags);
2405 else if (drflags & PD_LIST_SLOCKED)
2406 sx_sunlock(&allprison_lock);
2407 else if (drflags & PD_LIST_XLOCKED)
2408 sx_xunlock(&allprison_lock);
2409 if (root != NULL)
2410 vrele(root);
2411 done_errmsg:
2412 if (error) {
2413 /* Write the error message back to userspace. */
2414 if (vfs_getopt(opts, JAIL_PARAM_ERRMSG, (void **)&errmsg,
2415 &errmsg_len) == 0 && errmsg_len > 0) {
2416 errmsg_pos = 2 * vfs_getopt_pos(opts, JAIL_PARAM_ERRMSG) + 1;
2417 if (optuio->uio_segflg == UIO_SYSSPACE)
2418 bcopy(errmsg,
2419 optuio->uio_iov[errmsg_pos].iov_base,
2420 errmsg_len);
2421 else
2422 (void)copyout(errmsg,
2423 optuio->uio_iov[errmsg_pos].iov_base,
2424 errmsg_len);
2425 }
2426 }
2427 done_free:
2428 /* Clean up other resources. */
2429 #ifdef INET
2430 prison_ip_free(ip4);
2431 #endif
2432 #ifdef INET6
2433 prison_ip_free(ip6);
2434 #endif
2435 #ifdef MAC
2436 if (mac_set_prison_data != NULL)
2437 mac_set_prison_finish(td, error == 0, mac_set_prison_data);
2438 #endif
2439 if (jfp_out != NULL)
2440 fdrop(jfp_out, td);
2441 if (error && jfd_out >= 0)
2442 (void)kern_close(td, jfd_out);
2443 if (g_path != NULL)
2444 free(g_path, M_TEMP);
2445 if (opts != NULL)
2446 vfs_freeopts(opts);
2447 prison_free(mypr);
2448 if (flags & JAIL_ATTACH)
2449 prison_attach_thread_single_end(td);
2450 return (error);
2451 }
2452
2453 /*
2454 * Find the next available prison ID. Return the ID on success, or zero
2455 * on failure. Also set a pointer to the allprison list entry the prison
2456 * should be inserted before.
2457 */
2458 static int
get_next_prid(struct prison ** insprp)2459 get_next_prid(struct prison **insprp)
2460 {
2461 struct prison *inspr;
2462 int jid, maxid;
2463
2464 jid = lastprid % JAIL_MAX + 1;
2465 if (TAILQ_EMPTY(&allprison) ||
2466 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
2467 /*
2468 * A common case is for all jails to be implicitly numbered,
2469 * which means they'll go on the end of the list, at least
2470 * for the first JAIL_MAX times.
2471 */
2472 inspr = NULL;
2473 } else {
2474 /*
2475 * Take two passes through the allprison list: first starting
2476 * with the proposed jid, then ending with it.
2477 */
2478 for (maxid = JAIL_MAX; maxid != 0; ) {
2479 TAILQ_FOREACH(inspr, &allprison, pr_list) {
2480 if (inspr->pr_id < jid)
2481 continue;
2482 if (inspr->pr_id > jid) {
2483 /* Found an opening. */
2484 maxid = 0;
2485 break;
2486 }
2487 if (++jid > maxid) {
2488 if (lastprid == maxid || lastprid == 0)
2489 {
2490 /*
2491 * The entire legal range
2492 * has been traversed
2493 */
2494 return 0;
2495 }
2496 /* Try again from the start. */
2497 jid = 1;
2498 maxid = lastprid;
2499 break;
2500 }
2501 }
2502 if (inspr == NULL) {
2503 /* Found room at the end of the list. */
2504 break;
2505 }
2506 }
2507 }
2508 *insprp = inspr;
2509 lastprid = jid;
2510 return (jid);
2511 }
2512
2513 /*
2514 * Find the next available ID for a renumbered dead prison. This is the same
2515 * as get_next_prid, but counting backward from the end of the range.
2516 */
2517 static int
get_next_deadid(struct prison ** dinsprp)2518 get_next_deadid(struct prison **dinsprp)
2519 {
2520 struct prison *dinspr;
2521 int deadid, minid;
2522
2523 deadid = lastdeadid ? lastdeadid - 1 : JAIL_MAX;
2524 /*
2525 * Take two reverse passes through the allprison list: first
2526 * starting with the proposed deadid, then ending with it.
2527 */
2528 for (minid = 1; minid != 0; ) {
2529 TAILQ_FOREACH_REVERSE(dinspr, &allprison, prisonlist, pr_list) {
2530 if (dinspr->pr_id > deadid)
2531 continue;
2532 if (dinspr->pr_id < deadid) {
2533 /* Found an opening. */
2534 minid = 0;
2535 break;
2536 }
2537 if (--deadid < minid) {
2538 if (lastdeadid == minid || lastdeadid == 0)
2539 {
2540 /*
2541 * The entire legal range
2542 * has been traversed
2543 */
2544 return 0;
2545 }
2546 /* Try again from the end. */
2547 deadid = JAIL_MAX;
2548 minid = lastdeadid;
2549 break;
2550 }
2551 }
2552 if (dinspr == NULL) {
2553 /* Found room at the beginning of the list. */
2554 break;
2555 }
2556 }
2557 *dinsprp = dinspr;
2558 lastdeadid = deadid;
2559 return (deadid);
2560 }
2561
2562 /*
2563 * struct jail_get_args {
2564 * struct iovec *iovp;
2565 * unsigned int iovcnt;
2566 * int flags;
2567 * };
2568 */
2569 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)2570 sys_jail_get(struct thread *td, struct jail_get_args *uap)
2571 {
2572 return (user_jail_get(td, uap->iovp, uap->iovcnt, uap->flags,
2573 copyinuio, updateiov));
2574 }
2575
2576 int
user_jail_get(struct thread * td,struct iovec * iovp,unsigned int iovcnt,int flags,copyinuio_t * copyinuio_f,updateiov_t * updateiov_f)2577 user_jail_get(struct thread *td, struct iovec *iovp,
2578 unsigned int iovcnt, int flags, copyinuio_t *copyinuio_f,
2579 updateiov_t *updateiov_f)
2580 {
2581 struct uio *auio;
2582 int error;
2583
2584 /* Check that we have an even number of iovecs. */
2585 if (iovcnt & 1)
2586 return (EINVAL);
2587
2588 error = copyinuio_f(iovp, iovcnt, &auio);
2589 if (error)
2590 return (error);
2591 error = kern_jail_get(td, auio, flags);
2592 if (error == 0)
2593 error = updateiov_f(auio, iovp);
2594 freeuio(auio);
2595 return (error);
2596 }
2597
2598 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2599 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2600 {
2601 struct bool_flags *bf;
2602 struct file *jfp_out;
2603 struct jailsys_flags *jsf;
2604 struct prison *pr, *mypr;
2605 struct vfsopt *opt;
2606 struct vfsoptlist *opts;
2607 char *errmsg, *name;
2608 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2609 int jfd_in, jfd_out;
2610 unsigned f;
2611
2612 if (flags & ~JAIL_GET_MASK)
2613 return (EINVAL);
2614 if ((flags & (JAIL_USE_DESC | JAIL_AT_DESC)) ==
2615 (JAIL_USE_DESC | JAIL_AT_DESC))
2616 return (EINVAL);
2617
2618 /* Get the parameter list. */
2619 error = vfs_buildopts(optuio, &opts);
2620 if (error)
2621 return (error);
2622 errmsg_pos = vfs_getopt_pos(opts, JAIL_PARAM_ERRMSG);
2623 mypr = td->td_ucred->cr_prison;
2624 prison_hold(mypr);
2625 pr = NULL;
2626 jfp_out = NULL;
2627 jfd_out = -1;
2628
2629 /*
2630 * Find the prison specified by one of: desc, lastjid, jid, name.
2631 */
2632 sx_slock(&allprison_lock);
2633 drflags = PD_LIST_SLOCKED;
2634
2635 error = vfs_copyopt(opts, JAIL_PARAM_DESC, &jfd_in, sizeof(jfd_in));
2636 if (error == ENOENT) {
2637 if (flags & (JAIL_AT_DESC | JAIL_GET_DESC | JAIL_OWN_DESC)) {
2638 vfs_opterror(opts, "missing desc");
2639 goto done;
2640 }
2641 } else if (error == 0) {
2642 if (!(flags & (JAIL_USE_DESC | JAIL_AT_DESC | JAIL_GET_DESC |
2643 JAIL_OWN_DESC))) {
2644 error = EINVAL;
2645 vfs_opterror(opts, "unexpected desc");
2646 goto done;
2647 }
2648 if (flags & JAIL_USE_DESC) {
2649 /* Get the jail from its descriptor. */
2650 error = jaildesc_find(td, jfd_in, &pr, NULL);
2651 if (error) {
2652 vfs_opterror(opts, error == ENOENT ?
2653 "descriptor to dead jail" :
2654 "not a jail descriptor");
2655 goto done;
2656 }
2657 drflags |= PD_DEREF;
2658 mtx_lock(&pr->pr_mtx);
2659 drflags |= PD_LOCKED;
2660 goto found_prison;
2661 }
2662 if (flags & JAIL_AT_DESC) {
2663 /* Look up jails based on the descriptor's prison. */
2664 struct prison *jdpr;
2665
2666 error = jaildesc_find(td, jfd_in, &jdpr, NULL);
2667 if (error != 0) {
2668 vfs_opterror(opts, error == ENOENT ?
2669 "descriptor to dead jail" :
2670 "not a jail descriptor");
2671 goto done;
2672 }
2673 prison_free(mypr);
2674 mypr = jdpr;
2675 }
2676 if (flags & (JAIL_GET_DESC | JAIL_OWN_DESC)) {
2677 /* Allocate a jail descriptor to return later. */
2678 error = jaildesc_alloc(td, &jfp_out, &jfd_out,
2679 flags & JAIL_OWN_DESC);
2680 if (error)
2681 goto done;
2682 }
2683 } else
2684 goto done;
2685
2686 error = vfs_copyopt(opts, JAIL_PARAM_LASTJID, &jid, sizeof(jid));
2687 if (error == 0) {
2688 TAILQ_FOREACH(pr, &allprison, pr_list) {
2689 if (pr->pr_id > jid &&
2690 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2691 prison_ischild(mypr, pr)) {
2692 mtx_lock(&pr->pr_mtx);
2693 drflags |= PD_LOCKED;
2694 #ifdef MAC
2695 /*
2696 * We special-case this one check because we
2697 * don't want MAC to break jail enumeration. We
2698 * need to just move on to the next accessible
2699 * and alive prison.
2700 */
2701 error = mac_prison_check_get(td->td_ucred, pr,
2702 opts, flags);
2703 if (error != 0) {
2704 mtx_unlock(&pr->pr_mtx);
2705 drflags &= ~PD_LOCKED;
2706 continue;
2707 }
2708
2709 /*
2710 * Avoid potentially expensive trip back into
2711 * the MAC framework.
2712 */
2713 goto found_prison_nomac_alive;
2714 #else
2715 goto found_prison;
2716 #endif
2717 }
2718 }
2719 error = ENOENT;
2720 vfs_opterror(opts, "no jail after %d", jid);
2721 goto done;
2722 } else if (error != ENOENT)
2723 goto done;
2724
2725 error = vfs_copyopt(opts, JAIL_PARAM_JID, &jid, sizeof(jid));
2726 if (error == 0) {
2727 if (jid != 0) {
2728 pr = prison_find_child(mypr, jid);
2729 if (pr != NULL) {
2730 drflags |= PD_LOCKED;
2731 goto found_prison;
2732 }
2733 error = ENOENT;
2734 vfs_opterror(opts, "jail %d not found", jid);
2735 goto done;
2736 }
2737 } else if (error != ENOENT)
2738 goto done;
2739
2740 error = vfs_getopt(opts, JAIL_PARAM_NAME, (void **)&name, &len);
2741 if (error == 0) {
2742 if (len == 0 || name[len - 1] != '\0') {
2743 error = EINVAL;
2744 goto done;
2745 }
2746 pr = prison_find_name(mypr, name);
2747 if (pr != NULL) {
2748 drflags |= PD_LOCKED;
2749 goto found_prison;
2750 }
2751 error = ENOENT;
2752 vfs_opterror(opts, "jail \"%s\" not found", name);
2753 goto done;
2754 } else if (error != ENOENT)
2755 goto done;
2756
2757 vfs_opterror(opts, "no jail specified");
2758 error = ENOENT;
2759 goto done;
2760
2761 found_prison:
2762 #ifdef MAC
2763 error = mac_prison_check_get(td->td_ucred, pr, opts, flags);
2764 if (error != 0)
2765 goto done;
2766 #endif
2767 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2768 error = ENOENT;
2769 if (pr->pr_name[0] != '0' && isdigit(pr->pr_name[0])) {
2770 vfs_opterror(opts, "jail %d is dying",
2771 pr->pr_id);
2772 } else {
2773 vfs_opterror(opts, "jail \"%s\" (%d) is dying",
2774 pr->pr_name, pr->pr_id);
2775 }
2776 goto done;
2777 }
2778 #ifdef MAC
2779 found_prison_nomac_alive:
2780 #endif
2781 /* Get the parameters of the prison. */
2782 if (!(drflags & PD_DEREF)) {
2783 prison_hold(pr);
2784 drflags |= PD_DEREF;
2785 }
2786 td->td_retval[0] = pr->pr_id;
2787 if (jfd_out >= 0) {
2788 error = vfs_setopt(opts, JAIL_PARAM_DESC, &jfd_out, sizeof(jfd_out));
2789 if (error != 0 && error != ENOENT)
2790 goto done;
2791 jaildesc_set_prison(jfp_out, pr);
2792 }
2793 error = vfs_setopt(opts, JAIL_PARAM_JID, &pr->pr_id, sizeof(pr->pr_id));
2794 if (error != 0 && error != ENOENT)
2795 goto done;
2796 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2797 error = vfs_setopt(opts, JAIL_PARAM_PARENT, &i, sizeof(i));
2798 if (error != 0 && error != ENOENT)
2799 goto done;
2800 error = vfs_setopts(opts, JAIL_PARAM_NAME, prison_name(mypr, pr));
2801 if (error != 0 && error != ENOENT)
2802 goto done;
2803 error = vfs_setopt(opts, JAIL_PARAM_CPUSET_ID, &pr->pr_cpuset->cs_id,
2804 sizeof(pr->pr_cpuset->cs_id));
2805 if (error != 0 && error != ENOENT)
2806 goto done;
2807 error = vfs_setopts(opts, JAIL_PARAM_PATH, prison_path(mypr, pr));
2808 if (error != 0 && error != ENOENT)
2809 goto done;
2810 #ifdef INET
2811 error = vfs_setopt_part(opts, JAIL_PARAM_IP4_ADDR,
2812 pr->pr_addrs[PR_INET]->pr_ip,
2813 pr->pr_addrs[PR_INET] ? pr->pr_addrs[PR_INET]->ips *
2814 pr_families[PR_INET].size : 0 );
2815 if (error != 0 && error != ENOENT)
2816 goto done;
2817 #endif
2818 #ifdef INET6
2819 error = vfs_setopt_part(opts, JAIL_PARAM_IP6_ADDR,
2820 pr->pr_addrs[PR_INET6]->pr_ip,
2821 pr->pr_addrs[PR_INET6] ? pr->pr_addrs[PR_INET6]->ips *
2822 pr_families[PR_INET6].size : 0 );
2823 if (error != 0 && error != ENOENT)
2824 goto done;
2825 #endif
2826 error = vfs_setopt(opts, JAIL_PARAM_SECURELEVEL, &pr->pr_securelevel,
2827 sizeof(pr->pr_securelevel));
2828 if (error != 0 && error != ENOENT)
2829 goto done;
2830 error = vfs_setopt(opts, JAIL_PARAM_CHILDREN_CUR, &pr->pr_childcount,
2831 sizeof(pr->pr_childcount));
2832 if (error != 0 && error != ENOENT)
2833 goto done;
2834 error = vfs_setopt(opts, JAIL_PARAM_CHILDREN_MAX, &pr->pr_childmax,
2835 sizeof(pr->pr_childmax));
2836 if (error != 0 && error != ENOENT)
2837 goto done;
2838 error = vfs_setopts(opts, JAIL_PARAM_HOST_HOSTNAME, pr->pr_hostname);
2839 if (error != 0 && error != ENOENT)
2840 goto done;
2841 error = vfs_setopts(opts, JAIL_PARAM_HOST_DOMAINNAME, pr->pr_domainname);
2842 if (error != 0 && error != ENOENT)
2843 goto done;
2844 error = vfs_setopts(opts, JAIL_PARAM_HOST_HOSTUUID, pr->pr_hostuuid);
2845 if (error != 0 && error != ENOENT)
2846 goto done;
2847 #ifdef COMPAT_FREEBSD32
2848 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2849 uint32_t hid32 = pr->pr_hostid;
2850
2851 error = vfs_setopt(opts, JAIL_PARAM_HOST_HOSTID, &hid32, sizeof(hid32));
2852 } else
2853 #endif
2854 error = vfs_setopt(opts, JAIL_PARAM_HOST_HOSTID, &pr->pr_hostid,
2855 sizeof(pr->pr_hostid));
2856 if (error != 0 && error != ENOENT)
2857 goto done;
2858 error = vfs_setopt(opts, JAIL_PARAM_ENFORCE_STATFS, &pr->pr_enforce_statfs,
2859 sizeof(pr->pr_enforce_statfs));
2860 if (error != 0 && error != ENOENT)
2861 goto done;
2862 error = vfs_setopt(opts, JAIL_PARAM_DEVFS_RULESET, &pr->pr_devfs_rsnum,
2863 sizeof(pr->pr_devfs_rsnum));
2864 if (error != 0 && error != ENOENT)
2865 goto done;
2866 for (bf = pr_flag_bool;
2867 bf < pr_flag_bool + nitems(pr_flag_bool);
2868 bf++) {
2869 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2870 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2871 if (error != 0 && error != ENOENT)
2872 goto done;
2873 i = !i;
2874 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2875 if (error != 0 && error != ENOENT)
2876 goto done;
2877 }
2878 for (jsf = pr_flag_jailsys;
2879 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2880 jsf++) {
2881 f = pr->pr_flags & (jsf->disable | jsf->new);
2882 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2883 : (f == jsf->new) ? JAIL_SYS_NEW
2884 : JAIL_SYS_INHERIT;
2885 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2886 if (error != 0 && error != ENOENT)
2887 goto done;
2888 }
2889 for (bf = pr_flag_allow;
2890 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2891 atomic_load_int(&bf->flag) != 0;
2892 bf++) {
2893 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2894 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2895 if (error != 0 && error != ENOENT)
2896 goto done;
2897 i = !i;
2898 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2899 if (error != 0 && error != ENOENT)
2900 goto done;
2901 }
2902 i = !prison_isalive(pr);
2903 error = vfs_setopt(opts, JAIL_PARAM_DYING, &i, sizeof(i));
2904 if (error != 0 && error != ENOENT)
2905 goto done;
2906 i = !i;
2907 error = vfs_setopt(opts, JAIL_PARAM_NODYING, &i, sizeof(i));
2908 if (error != 0 && error != ENOENT)
2909 goto done;
2910 error = vfs_setopt(opts, JAIL_PARAM_OSRELDATE, &pr->pr_osreldate,
2911 sizeof(pr->pr_osreldate));
2912 if (error != 0 && error != ENOENT)
2913 goto done;
2914 error = vfs_setopts(opts, JAIL_PARAM_OSRELEASE, pr->pr_osrelease);
2915 if (error != 0 && error != ENOENT)
2916 goto done;
2917
2918 #ifdef MAC
2919 /*
2920 * We get the MAC label last because we'll let the MAC framework drop
2921 * pr_mtx to externalize the label.
2922 */
2923 error = mac_get_prison(td, pr, opts);
2924 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
2925 drflags &= ~PD_LOCKED;
2926 if (error != 0 && error != ENOENT)
2927 goto done;
2928 #else
2929 mtx_unlock(&pr->pr_mtx);
2930 drflags &= ~PD_LOCKED;
2931 #endif
2932
2933 /* Get the module parameters. */
2934 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2935 if (error)
2936 goto done;
2937 prison_deref(pr, drflags);
2938 pr = NULL;
2939 drflags = 0;
2940
2941 /* By now, all parameters should have been noted. */
2942 TAILQ_FOREACH(opt, opts, link) {
2943 if (!opt->seen &&
2944 (strstr(opt->name, JAIL_META_PRIVATE ".") == opt->name ||
2945 strstr(opt->name, JAIL_META_SHARED ".") == opt->name)) {
2946 /* Communicate back a missing key. */
2947 free(opt->value, M_MOUNT);
2948 opt->value = NULL;
2949 opt->len = 0;
2950 continue;
2951 }
2952 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2953 error = EINVAL;
2954 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2955 goto done;
2956 }
2957 }
2958
2959 /* Write the fetched parameters back to userspace. */
2960 error = 0;
2961 TAILQ_FOREACH(opt, opts, link) {
2962 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2963 pos = 2 * opt->pos + 1;
2964 optuio->uio_iov[pos].iov_len = opt->len;
2965 if (opt->value != NULL) {
2966 if (optuio->uio_segflg == UIO_SYSSPACE) {
2967 bcopy(opt->value,
2968 optuio->uio_iov[pos].iov_base,
2969 opt->len);
2970 } else {
2971 error = copyout(opt->value,
2972 optuio->uio_iov[pos].iov_base,
2973 opt->len);
2974 if (error)
2975 break;
2976 }
2977 }
2978 }
2979 }
2980
2981 done:
2982 /* Release any temporary prison holds and/or locks. */
2983 if (pr != NULL)
2984 prison_deref(pr, drflags);
2985 else if (drflags & PD_LIST_SLOCKED)
2986 sx_sunlock(&allprison_lock);
2987 else if (drflags & PD_LIST_XLOCKED)
2988 sx_xunlock(&allprison_lock);
2989 /* Clean up other resources. */
2990 if (jfp_out != NULL)
2991 (void)fdrop(jfp_out, td);
2992 if (error && jfd_out >= 0)
2993 (void)kern_close(td, jfd_out);
2994 if (error && errmsg_pos >= 0) {
2995 /* Write the error message back to userspace. */
2996 vfs_getopt(opts, JAIL_PARAM_ERRMSG, (void **)&errmsg, &errmsg_len);
2997 errmsg_pos = 2 * errmsg_pos + 1;
2998 if (errmsg_len > 0) {
2999 if (optuio->uio_segflg == UIO_SYSSPACE)
3000 bcopy(errmsg,
3001 optuio->uio_iov[errmsg_pos].iov_base,
3002 errmsg_len);
3003 else
3004 (void)copyout(errmsg,
3005 optuio->uio_iov[errmsg_pos].iov_base,
3006 errmsg_len);
3007 }
3008 }
3009 vfs_freeopts(opts);
3010 prison_free(mypr);
3011 return (error);
3012 }
3013
3014 /*
3015 * struct jail_remove_args {
3016 * int jid;
3017 * };
3018 */
3019 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)3020 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
3021 {
3022 struct prison *pr;
3023 int error;
3024
3025 error = priv_check(td, PRIV_JAIL_REMOVE);
3026 if (error)
3027 return (error);
3028
3029 sx_xlock(&allprison_lock);
3030 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
3031 if (pr == NULL) {
3032 sx_xunlock(&allprison_lock);
3033 return (EINVAL);
3034 }
3035 #ifdef MAC
3036 error = mac_prison_check_remove(td->td_ucred, pr);
3037 if (error != 0) {
3038 mtx_unlock(&pr->pr_mtx);
3039 sx_xunlock(&allprison_lock);
3040 return (error);
3041 }
3042 #endif
3043 prison_hold(pr);
3044 prison_remove(pr);
3045 return (0);
3046 }
3047
3048 /*
3049 * struct jail_remove_jd_args {
3050 * int fd;
3051 * };
3052 */
3053 int
sys_jail_remove_jd(struct thread * td,struct jail_remove_jd_args * uap)3054 sys_jail_remove_jd(struct thread *td, struct jail_remove_jd_args *uap)
3055 {
3056 struct prison *pr;
3057 struct ucred *jdcred;
3058 int error;
3059
3060 error = jaildesc_find(td, uap->fd, &pr, &jdcred);
3061 if (error)
3062 return (error);
3063 error = priv_check_cred(jdcred, PRIV_JAIL_REMOVE);
3064 crfree(jdcred);
3065 #ifdef MAC
3066 if (error == 0)
3067 error = mac_prison_check_remove(td->td_ucred, pr);
3068 #endif
3069 if (error) {
3070 prison_free(pr);
3071 return (error);
3072 }
3073 sx_xlock(&allprison_lock);
3074 mtx_lock(&pr->pr_mtx);
3075 prison_remove(pr);
3076 return (0);
3077 }
3078
3079 /*
3080 * Begin the removal process for a prison. The allprison lock should
3081 * be held exclusively, and the prison should be both locked and held.
3082 */
3083 void
prison_remove(struct prison * pr)3084 prison_remove(struct prison *pr)
3085 {
3086 sx_assert(&allprison_lock, SA_XLOCKED);
3087 mtx_assert(&pr->pr_mtx, MA_OWNED);
3088 prison_deref(pr, PD_KILL | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
3089 }
3090
3091 /*
3092 * struct jail_attach_args {
3093 * int jid;
3094 * };
3095 */
3096 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)3097 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
3098 {
3099 struct prison *pr;
3100 int drflags, error;
3101
3102 error = priv_check(td, PRIV_JAIL_ATTACH);
3103 if (error)
3104 return (error);
3105 /* Only let a single thread in the process try to attach at a time. */
3106 error = prison_attach_thread_single(td);
3107 if (error != 0)
3108 return (error);
3109
3110 sx_slock(&allprison_lock);
3111 drflags = PD_LIST_SLOCKED;
3112 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
3113 if (pr == NULL) {
3114 error = EINVAL;
3115 goto done;
3116 }
3117 drflags |= PD_LOCKED;
3118
3119 #ifdef MAC
3120 error = mac_prison_check_attach(td->td_ucred, pr);
3121 if (error != 0)
3122 goto done;
3123 #endif
3124
3125 /* Do not allow a process to attach to a prison that is not alive. */
3126 if (!prison_isalive(pr)) {
3127 error = EINVAL;
3128 goto done;
3129 }
3130
3131 error = do_jail_attach(td, pr, &drflags);
3132
3133 done:
3134 prison_deref(pr, drflags);
3135 prison_attach_thread_single_end(td);
3136 return (error);
3137 }
3138
3139 /*
3140 * struct jail_attach_jd_args {
3141 * int fd;
3142 * };
3143 */
3144 int
sys_jail_attach_jd(struct thread * td,struct jail_attach_jd_args * uap)3145 sys_jail_attach_jd(struct thread *td, struct jail_attach_jd_args *uap)
3146 {
3147 struct prison *pr;
3148 struct ucred *jdcred;
3149 int drflags, error;
3150
3151 /* Only let a single thread in the process try to attach at a time. */
3152 error = prison_attach_thread_single(td);
3153 if (error != 0)
3154 return (error);
3155
3156 sx_slock(&allprison_lock);
3157 drflags = PD_LIST_SLOCKED;
3158 pr = NULL;
3159 error = jaildesc_find(td, uap->fd, &pr, &jdcred);
3160 if (error)
3161 goto done;
3162 drflags |= PD_DEREF;
3163 error = priv_check_cred(jdcred, PRIV_JAIL_ATTACH);
3164 #ifdef MAC
3165 if (error == 0)
3166 error = mac_prison_check_attach(td->td_ucred, pr);
3167 #endif
3168 crfree(jdcred);
3169 if (error)
3170 goto done;
3171
3172 /* Do not allow a process to attach to a prison that is not alive. */
3173 if (!prison_isalive(pr)) {
3174 error = EINVAL;
3175 goto done;
3176 }
3177
3178 error = do_jail_attach(td, pr, &drflags);
3179
3180 done:
3181 prison_deref(pr, drflags);
3182 prison_attach_thread_single_end(td);
3183 return (error);
3184 }
3185
3186 /*
3187 * Attach the current process to a prison. On entry, the allprison
3188 * lock should be at least shared. On exit, both it and the prison
3189 * itself will be unlocked, which will be refelected in *drflagsp.
3190 */
3191 static int
do_jail_attach(struct thread * td,struct prison * pr,int * drflagsp)3192 do_jail_attach(struct thread *td, struct prison *pr, int *drflagsp)
3193 {
3194 struct proc *p;
3195 struct ucred *newcred, *oldcred;
3196 int drflags, error;
3197
3198 sx_assert(&allprison_lock, SX_LOCKED);
3199 KASSERT(prison_isvalid(pr), ("Attaching to invalid prison %p", pr));
3200 /*
3201 * Note the caller's locking state, but gain and track our own
3202 * references. The caller will see that locks have been
3203 * dropped (which isn't true now, but will be after OSD calls).
3204 */
3205 prison_hold(pr);
3206 refcount_acquire(&pr->pr_uref);
3207 drflags = PD_DEREF | PD_DEUREF | (*drflagsp & PD_LOCK_FLAGS);
3208 *drflagsp &= PD_OP_FLAGS;
3209
3210 /* Let modules do whatever they need to prepare for attaching. */
3211 if (drflags & PD_LOCKED) {
3212 mtx_unlock(&pr->pr_mtx);
3213 drflags &= ~PD_LOCKED;
3214 }
3215 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
3216 if (error)
3217 goto e_revert_osd;
3218 sx_unlock(&allprison_lock);
3219 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
3220
3221 /*
3222 * Reparent the newly attached process to this jail.
3223 */
3224 p = td->td_proc;
3225 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
3226 if (error)
3227 goto e_revert_osd;
3228
3229 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
3230 if ((error = change_dir(pr->pr_root, td)) != 0)
3231 goto e_unlock;
3232 #ifdef MAC
3233 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
3234 goto e_unlock;
3235 #endif
3236 VOP_UNLOCK(pr->pr_root);
3237 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
3238 goto e_revert_osd;
3239
3240 newcred = crget();
3241 PROC_LOCK(p);
3242 oldcred = crcopysafe(p, newcred);
3243 newcred->cr_prison = pr;
3244 #ifdef RACCT
3245 racct_proc_ucred_changed(p, oldcred, newcred);
3246 #endif
3247 #ifdef RCTL
3248 crhold(newcred);
3249 #endif
3250 /*
3251 * Takes over 'newcred''s reference, so 'newcred' must not be used
3252 * besides this point except on RCTL where we took an additional
3253 * reference above.
3254 */
3255 proc_set_cred(p, newcred);
3256 setsugid(p);
3257 PROC_UNLOCK(p);
3258 #ifdef RCTL
3259 rctl_proc_ucred_changed(p, newcred);
3260 crfree(newcred);
3261 #endif
3262 prison_proc_relink(oldcred->cr_prison, pr, p);
3263 prison_deref(oldcred->cr_prison, drflags);
3264 crfree(oldcred);
3265 prison_knote(pr, NOTE_JAIL_ATTACH | td->td_proc->p_pid);
3266 #ifdef MAC
3267 /*
3268 * Note that mac_prison_attached() assumes that it's called in a
3269 * sleepable context.
3270 */
3271 mac_prison_attached(td->td_ucred, pr, td->td_proc);
3272 #endif
3273
3274 /*
3275 * If the prison was killed while changing credentials, die along
3276 * with it.
3277 */
3278 if (!prison_isalive(pr)) {
3279 PROC_LOCK(p);
3280 kern_psignal(p, SIGKILL);
3281 PROC_UNLOCK(p);
3282 }
3283
3284 return (0);
3285
3286 e_unlock:
3287 VOP_UNLOCK(pr->pr_root);
3288 e_revert_osd:
3289 /* Tell modules this process is still in its old jail after all. */
3290 if (!(drflags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
3291 sx_slock(&allprison_lock);
3292 drflags |= PD_LIST_SLOCKED;
3293 }
3294 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
3295 prison_deref(pr, drflags);
3296 return (error);
3297 }
3298
3299 /*
3300 * Only one thread in a process should try to attach to a jail, or
3301 * they might end up with a process root from one prison, but attached
3302 * to the jail of another. Enforce this by making the process run in
3303 * single-threaded mode for the duration of the system call, which
3304 * also prevents other related calls such as chroot.
3305 */
3306 static int
prison_attach_thread_single(struct thread * td)3307 prison_attach_thread_single(struct thread *td)
3308 {
3309 struct proc *p;
3310 int error;
3311
3312 error = 0;
3313 p = td->td_proc;
3314 if ((atomic_load_int(&p->p_flag) & P_HADTHREADS) != 0) {
3315 PROC_LOCK(p);
3316 if (thread_single(p, SINGLE_BOUNDARY))
3317 error = ERESTART;
3318 PROC_UNLOCK(p);
3319 }
3320 return (error);
3321 }
3322
3323 static void
prison_attach_thread_single_end(struct thread * td)3324 prison_attach_thread_single_end(struct thread *td)
3325 {
3326 struct proc *p;
3327
3328 p = td->td_proc;
3329 if ((atomic_load_int(&p->p_flag) & P_HADTHREADS) != 0) {
3330 PROC_LOCK(p);
3331 thread_single_end(p, SINGLE_BOUNDARY);
3332 PROC_UNLOCK(p);
3333 }
3334 }
3335
3336 /*
3337 * Returns a locked prison instance, or NULL on failure.
3338 */
3339 struct prison *
prison_find(int prid)3340 prison_find(int prid)
3341 {
3342 struct prison *pr;
3343
3344 sx_assert(&allprison_lock, SX_LOCKED);
3345 TAILQ_FOREACH(pr, &allprison, pr_list) {
3346 if (pr->pr_id < prid)
3347 continue;
3348 if (pr->pr_id > prid)
3349 break;
3350 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
3351 mtx_lock(&pr->pr_mtx);
3352 return (pr);
3353 }
3354 return (NULL);
3355 }
3356
3357 /*
3358 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
3359 */
3360 struct prison *
prison_find_child(struct prison * mypr,int prid)3361 prison_find_child(struct prison *mypr, int prid)
3362 {
3363 struct prison *pr;
3364 int descend;
3365
3366 sx_assert(&allprison_lock, SX_LOCKED);
3367 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3368 if (pr->pr_id == prid) {
3369 KASSERT(prison_isvalid(pr),
3370 ("Found invalid prison %p", pr));
3371 mtx_lock(&pr->pr_mtx);
3372 return (pr);
3373 }
3374 }
3375 return (NULL);
3376 }
3377
3378 /*
3379 * Look for the name relative to mypr. Returns a locked prison or NULL.
3380 */
3381 struct prison *
prison_find_name(struct prison * mypr,const char * name)3382 prison_find_name(struct prison *mypr, const char *name)
3383 {
3384 struct prison *pr, *deadpr;
3385 size_t mylen;
3386 int descend;
3387
3388 sx_assert(&allprison_lock, SX_LOCKED);
3389 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
3390 deadpr = NULL;
3391 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
3392 if (!strcmp(pr->pr_name + mylen, name)) {
3393 KASSERT(prison_isvalid(pr),
3394 ("Found invalid prison %p", pr));
3395 if (prison_isalive(pr)) {
3396 mtx_lock(&pr->pr_mtx);
3397 return (pr);
3398 }
3399 deadpr = pr;
3400 }
3401 }
3402 /* There was no valid prison - perhaps there was a dying one. */
3403 if (deadpr != NULL)
3404 mtx_lock(&deadpr->pr_mtx);
3405 return (deadpr);
3406 }
3407
3408 /*
3409 * See if a prison has the specific flag set. The prison should be locked,
3410 * unless checking for flags that are only set at jail creation (such as
3411 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
3412 * to any other prison data.
3413 */
3414 bool
prison_flag(struct ucred * cred,unsigned flag)3415 prison_flag(struct ucred *cred, unsigned flag)
3416 {
3417
3418 return ((cred->cr_prison->pr_flags & flag) != 0);
3419 }
3420
3421 /*
3422 * See if a prison has the specific allow flag set.
3423 * The prison *should* be locked, or only a single bit is examined, without
3424 * regard to any other prison data.
3425 */
3426 bool
prison_allow(struct ucred * cred,unsigned flag)3427 prison_allow(struct ucred *cred, unsigned flag)
3428 {
3429
3430 return ((cred->cr_prison->pr_allow & flag) != 0);
3431 }
3432
3433 /*
3434 * Hold a prison reference, by incrementing pr_ref. It is generally
3435 * an error to hold a prison that does not already have a reference.
3436 * A prison record will remain valid as long as it has at least one
3437 * reference, and will not be removed as long as either the prison
3438 * mutex or the allprison lock is held (allprison_lock may be shared).
3439 */
3440 void
prison_hold_locked(struct prison * pr)3441 prison_hold_locked(struct prison *pr)
3442 {
3443
3444 /* Locking is no longer required. */
3445 prison_hold(pr);
3446 }
3447
3448 void
prison_hold(struct prison * pr)3449 prison_hold(struct prison *pr)
3450 {
3451 u_int old __diagused;
3452
3453 old = refcount_acquire(&pr->pr_ref);
3454 KASSERT(old > 0,
3455 ("cannot hold a non-alive prison (jid=%d)", pr->pr_id));
3456 }
3457
3458 /*
3459 * Remove a prison reference. If that was the last reference, the
3460 * prison will be removed (at a later time).
3461 */
3462 void
prison_free_locked(struct prison * pr)3463 prison_free_locked(struct prison *pr)
3464 {
3465
3466 mtx_assert(&pr->pr_mtx, MA_OWNED);
3467 /*
3468 * Locking is no longer required, but unlock because the caller
3469 * expects it.
3470 */
3471 mtx_unlock(&pr->pr_mtx);
3472 prison_free(pr);
3473 }
3474
3475 void
prison_free(struct prison * pr)3476 prison_free(struct prison *pr)
3477 {
3478
3479 KASSERT(refcount_load(&pr->pr_ref) > 0,
3480 ("Trying to free dead prison %p (jid=%d).",
3481 pr, pr->pr_id));
3482 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3483 /*
3484 * Don't remove the last reference in this context,
3485 * in case there are locks held.
3486 */
3487 taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3488 }
3489 }
3490
3491 static void
prison_free_not_last(struct prison * pr)3492 prison_free_not_last(struct prison *pr)
3493 {
3494 bool released __diagused;
3495
3496 released = refcount_release(&pr->pr_ref);
3497 KASSERT(!released,
3498 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
3499 pr, pr->pr_id));
3500 }
3501
3502 /*
3503 * Hold a prison for user visibility, by incrementing pr_uref.
3504 * It is generally an error to hold a prison that isn't already
3505 * user-visible, except through the jail system calls. It is also
3506 * an error to hold an invalid prison. A prison record will remain
3507 * alive as long as it has at least one user reference, and will not
3508 * be set to the dying state until the prison mutex and allprison_lock
3509 * are both freed.
3510 */
3511 void
prison_proc_hold(struct prison * pr)3512 prison_proc_hold(struct prison *pr)
3513 {
3514 u_int old __diagused;
3515
3516 old = refcount_acquire(&pr->pr_uref);
3517 KASSERT(old > 0,
3518 ("cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
3519 }
3520
3521 /*
3522 * Remove a prison user reference. If it was the last reference, the
3523 * prison will be considered "dying", and may be removed once all of
3524 * its references are dropped.
3525 */
3526 void
prison_proc_free(struct prison * pr)3527 prison_proc_free(struct prison *pr)
3528 {
3529
3530 /*
3531 * Locking is only required when releasing the last reference.
3532 * This allows assurance that a locked prison will remain alive
3533 * until it is unlocked.
3534 */
3535 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3536 /*
3537 * Don't remove the last user reference in this context,
3538 * which is expected to be a process that is not only locked,
3539 * but also half dead. Add a reference so any calls to
3540 * prison_free() won't re-submit the task.
3541 */
3542 prison_hold(pr);
3543 mtx_lock(&pr->pr_mtx);
3544 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
3545 ("Redundant last reference in prison_proc_free (jid=%d)",
3546 pr->pr_id));
3547 pr->pr_flags |= PR_COMPLETE_PROC;
3548 mtx_unlock(&pr->pr_mtx);
3549 taskqueue_enqueue(taskqueue_jail_remove, &pr->pr_task);
3550 }
3551 }
3552
3553 static void
prison_proc_free_not_last(struct prison * pr)3554 prison_proc_free_not_last(struct prison *pr)
3555 {
3556 #ifdef INVARIANTS
3557 int lastref;
3558
3559 KASSERT(refcount_load(&pr->pr_uref) > 0,
3560 ("Trying to free dead prison %p (jid=%d).",
3561 pr, pr->pr_id));
3562 lastref = refcount_release(&pr->pr_uref);
3563 KASSERT(!lastref,
3564 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
3565 pr, pr->pr_id));
3566 #else
3567 refcount_release(&pr->pr_uref);
3568 #endif
3569 }
3570
3571 void
prison_proc_link(struct prison * pr,struct proc * p)3572 prison_proc_link(struct prison *pr, struct proc *p)
3573 {
3574
3575 sx_assert(&allproc_lock, SA_XLOCKED);
3576 LIST_INSERT_HEAD(&pr->pr_proclist, p, p_jaillist);
3577 }
3578
3579 void
prison_proc_unlink(struct prison * pr,struct proc * p)3580 prison_proc_unlink(struct prison *pr, struct proc *p)
3581 {
3582
3583 sx_assert(&allproc_lock, SA_XLOCKED);
3584 LIST_REMOVE(p, p_jaillist);
3585 }
3586
3587 static void
prison_proc_relink(struct prison * opr,struct prison * npr,struct proc * p)3588 prison_proc_relink(struct prison *opr, struct prison *npr, struct proc *p)
3589 {
3590
3591 sx_xlock(&allproc_lock);
3592 prison_proc_unlink(opr, p);
3593 prison_proc_link(npr, p);
3594 sx_xunlock(&allproc_lock);
3595 }
3596
3597 /*
3598 * Complete a call to either prison_free or prison_proc_free.
3599 */
3600 static void
prison_complete(void * context,int pending)3601 prison_complete(void *context, int pending)
3602 {
3603 struct prison *pr = context;
3604 int drflags;
3605
3606 /*
3607 * This could be called to release the last reference, or the last
3608 * user reference (plus the reference held in prison_proc_free).
3609 */
3610 drflags = prison_lock_xlock(pr, PD_DEREF);
3611 if (pr->pr_flags & PR_COMPLETE_PROC) {
3612 pr->pr_flags &= ~PR_COMPLETE_PROC;
3613 drflags |= PD_DEUREF;
3614 }
3615 prison_deref(pr, drflags);
3616 }
3617
3618 static void
prison_kill_processes_cb(struct proc * p,void * arg __unused)3619 prison_kill_processes_cb(struct proc *p, void *arg __unused)
3620 {
3621
3622 kern_psignal(p, SIGKILL);
3623 }
3624
3625 /*
3626 * Note the iteration does not guarantee acting on all processes.
3627 * Most notably there may be fork or jail_attach in progress.
3628 */
3629 void
prison_proc_iterate(struct prison * pr,void (* cb)(struct proc *,void *),void * cbarg)3630 prison_proc_iterate(struct prison *pr, void (*cb)(struct proc *, void *),
3631 void *cbarg)
3632 {
3633 struct prison *ppr;
3634 struct proc *p;
3635
3636 if (atomic_load_int(&pr->pr_childcount) == 0) {
3637 sx_slock(&allproc_lock);
3638 LIST_FOREACH(p, &pr->pr_proclist, p_jaillist) {
3639 if (p->p_state == PRS_NEW)
3640 continue;
3641 PROC_LOCK(p);
3642 cb(p, cbarg);
3643 PROC_UNLOCK(p);
3644 }
3645 sx_sunlock(&allproc_lock);
3646 if (atomic_load_int(&pr->pr_childcount) == 0)
3647 return;
3648 /*
3649 * Some jails popped up during the iteration, fall through to a
3650 * system-wide search.
3651 */
3652 }
3653
3654 sx_slock(&allproc_lock);
3655 FOREACH_PROC_IN_SYSTEM(p) {
3656 PROC_LOCK(p);
3657 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
3658 for (ppr = p->p_ucred->cr_prison; ppr != NULL;
3659 ppr = ppr->pr_parent) {
3660 if (ppr == pr) {
3661 cb(p, cbarg);
3662 break;
3663 }
3664 }
3665 }
3666 PROC_UNLOCK(p);
3667 }
3668 sx_sunlock(&allproc_lock);
3669 }
3670
3671 /*
3672 * Remove a prison reference and/or user reference (usually).
3673 * This assumes context that allows sleeping (for allprison_lock),
3674 * with no non-sleeping locks held, except perhaps the prison itself.
3675 * If there are no more references, release and delist the prison.
3676 * On completion, the prison lock and the allprison lock are both
3677 * unlocked.
3678 */
3679 static void
prison_deref(struct prison * pr,int flags)3680 prison_deref(struct prison *pr, int flags)
3681 {
3682 struct prisonlist freeprison;
3683 struct prison *killpr, *rpr, *ppr, *tpr;
3684
3685 killpr = NULL;
3686 TAILQ_INIT(&freeprison);
3687 /*
3688 * Release this prison as requested, which may cause its parent
3689 * to be released, and then maybe its grandparent, etc.
3690 */
3691 for (;;) {
3692 if (flags & PD_KILL) {
3693 /* Kill the prison and its descendents. */
3694 KASSERT(pr != &prison0,
3695 ("prison_deref trying to kill prison0"));
3696 if (!prison_isalive(pr)) {
3697 /* Silently ignore already-dying prisons. */
3698 flags &= ~PD_KILL;
3699 } else {
3700 if (!(flags & PD_DEREF)) {
3701 prison_hold(pr);
3702 flags |= PD_DEREF;
3703 }
3704 flags = prison_lock_xlock(pr, flags);
3705 prison_deref_kill(pr, &freeprison);
3706 }
3707 }
3708 if (flags & PD_DEUREF) {
3709 /* Drop a user reference. */
3710 KASSERT(refcount_load(&pr->pr_uref) > 0,
3711 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
3712 pr->pr_id));
3713 if (!refcount_release_if_not_last(&pr->pr_uref)) {
3714 if (!(flags & PD_DEREF)) {
3715 prison_hold(pr);
3716 flags |= PD_DEREF;
3717 }
3718 flags = prison_lock_xlock(pr, flags);
3719 if (refcount_release(&pr->pr_uref) &&
3720 pr->pr_state == PRISON_STATE_ALIVE) {
3721 /*
3722 * When the last user references goes,
3723 * this becomes a dying prison.
3724 */
3725 KASSERT(
3726 refcount_load(&prison0.pr_uref) > 0,
3727 ("prison0 pr_uref=0"));
3728 pr->pr_state = PRISON_STATE_DYING;
3729 prison_cleanup_locked(pr);
3730 mtx_unlock(&pr->pr_mtx);
3731 flags &= ~PD_LOCKED;
3732 prison_cleanup_unlocked(pr);
3733 }
3734 }
3735 }
3736 if (flags & PD_KILL) {
3737 /*
3738 * Any remaining user references are probably processes
3739 * that need to be killed, either in this prison or its
3740 * descendants.
3741 */
3742 if (refcount_load(&pr->pr_uref) > 0) {
3743 killpr = pr;
3744 flags &= ~PD_DEREF;
3745 }
3746 /* Make sure the parent prison doesn't get killed. */
3747 flags &= ~PD_KILL;
3748 }
3749 if (flags & PD_DEREF) {
3750 /* Drop a reference. */
3751 KASSERT(refcount_load(&pr->pr_ref) > 0,
3752 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
3753 pr->pr_id));
3754 if (!refcount_release_if_not_last(&pr->pr_ref)) {
3755 flags = prison_lock_xlock(pr, flags);
3756 if (refcount_release(&pr->pr_ref)) {
3757 /*
3758 * When the last reference goes,
3759 * unlink the prison and set it aside.
3760 */
3761 KASSERT(
3762 refcount_load(&pr->pr_uref) == 0,
3763 ("prison_deref: last ref, "
3764 "but still has %d urefs (jid=%d)",
3765 pr->pr_uref, pr->pr_id));
3766 KASSERT(
3767 refcount_load(&prison0.pr_ref) != 0,
3768 ("prison0 pr_ref=0"));
3769 #ifdef MAC
3770 /*
3771 * The MAC framework will call into any
3772 * policies that want to hook
3773 * prison_destroy_label, so ideally we
3774 * call this prior to any final state
3775 * invalidation to be safe.
3776 */
3777 mac_prison_destroy(pr);
3778 #endif
3779 pr->pr_state = PRISON_STATE_INVALID;
3780 TAILQ_REMOVE(&allprison, pr, pr_list);
3781 LIST_REMOVE(pr, pr_sibling);
3782 TAILQ_INSERT_TAIL(&freeprison, pr,
3783 pr_list);
3784 for (ppr = pr->pr_parent;
3785 ppr != NULL;
3786 ppr = ppr->pr_parent)
3787 ppr->pr_childcount--;
3788 /*
3789 * Removing a prison frees references
3790 * from its parent.
3791 */
3792 ppr = pr->pr_parent;
3793 pr->pr_parent = NULL;
3794 mtx_unlock(&pr->pr_mtx);
3795
3796 pr = ppr;
3797 flags &= ~PD_LOCKED;
3798 flags |= PD_DEREF | PD_DEUREF;
3799 continue;
3800 }
3801 }
3802 }
3803 break;
3804 }
3805
3806 /* Release all the prison locks. */
3807 if (flags & PD_LOCKED)
3808 mtx_unlock(&pr->pr_mtx);
3809 if (flags & PD_LIST_SLOCKED)
3810 sx_sunlock(&allprison_lock);
3811 else if (flags & PD_LIST_XLOCKED)
3812 sx_xunlock(&allprison_lock);
3813
3814 /* Kill any processes attached to a killed prison. */
3815 if (killpr != NULL) {
3816 prison_proc_iterate(killpr, prison_kill_processes_cb, NULL);
3817 prison_free(killpr);
3818 }
3819
3820 /*
3821 * Finish removing any unreferenced prisons, which couldn't happen
3822 * while allprison_lock was held (to avoid a LOR on vrele).
3823 */
3824 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
3825 #ifdef VIMAGE
3826 if (rpr->pr_flags & PR_VNET)
3827 vnet_destroy(rpr->pr_vnet);
3828 #endif
3829 if (rpr->pr_root != NULL)
3830 vrele(rpr->pr_root);
3831 mtx_destroy(&rpr->pr_mtx);
3832 #ifdef INET
3833 prison_ip_free(rpr->pr_addrs[PR_INET]);
3834 #endif
3835 #ifdef INET6
3836 prison_ip_free(rpr->pr_addrs[PR_INET6]);
3837 #endif
3838 if (rpr->pr_cpuset != NULL)
3839 cpuset_rel(rpr->pr_cpuset);
3840 osd_jail_exit(rpr);
3841 #ifdef RACCT
3842 if (racct_enable)
3843 prison_racct_detach(rpr);
3844 #endif
3845 TAILQ_REMOVE(&freeprison, rpr, pr_list);
3846 free(rpr, M_PRISON);
3847 }
3848 }
3849
3850 /*
3851 * Kill the prison and its descendants. Mark them as dying, clear the
3852 * persist flag, and call module remove methods.
3853 */
3854 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)3855 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
3856 {
3857 struct prison *cpr, *ppr, *rpr;
3858 bool descend;
3859
3860 /*
3861 * Unlike the descendants, the target prison can be killed
3862 * even if it is currently dying. This is useful for failed
3863 * creation in jail_set(2).
3864 */
3865 KASSERT(refcount_load(&pr->pr_ref) > 0,
3866 ("Trying to kill dead prison %p (jid=%d).",
3867 pr, pr->pr_id));
3868 refcount_acquire(&pr->pr_uref);
3869 pr->pr_state = PRISON_STATE_DYING;
3870 mtx_unlock(&pr->pr_mtx);
3871
3872 rpr = NULL;
3873 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
3874 if (descend) {
3875 if (!prison_isalive(cpr)) {
3876 descend = false;
3877 continue;
3878 }
3879 prison_hold(cpr);
3880 prison_proc_hold(cpr);
3881 mtx_lock(&cpr->pr_mtx);
3882 cpr->pr_state = PRISON_STATE_DYING;
3883 cpr->pr_flags |= PR_REMOVE;
3884 mtx_unlock(&cpr->pr_mtx);
3885 continue;
3886 }
3887 if (!(cpr->pr_flags & PR_REMOVE))
3888 continue;
3889 prison_cleanup_unlocked(cpr);
3890 mtx_lock(&cpr->pr_mtx);
3891 prison_cleanup_locked(cpr);
3892 cpr->pr_flags &= ~PR_REMOVE;
3893 if (cpr->pr_flags & PR_PERSIST) {
3894 cpr->pr_flags &= ~PR_PERSIST;
3895 prison_proc_free_not_last(cpr);
3896 prison_free_not_last(cpr);
3897 }
3898 (void)refcount_release(&cpr->pr_uref);
3899 if (refcount_release(&cpr->pr_ref)) {
3900 /*
3901 * When the last reference goes, unlink the prison
3902 * and set it aside for prison_deref() to handle.
3903 * Delay unlinking the sibling list to keep the loop
3904 * safe.
3905 */
3906 if (rpr != NULL)
3907 LIST_REMOVE(rpr, pr_sibling);
3908 rpr = cpr;
3909 rpr->pr_state = PRISON_STATE_INVALID;
3910 TAILQ_REMOVE(&allprison, rpr, pr_list);
3911 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
3912 /*
3913 * Removing a prison frees references from its parent.
3914 */
3915 ppr = rpr->pr_parent;
3916 prison_proc_free_not_last(ppr);
3917 prison_free_not_last(ppr);
3918 for (; ppr != NULL; ppr = ppr->pr_parent)
3919 ppr->pr_childcount--;
3920 }
3921 mtx_unlock(&cpr->pr_mtx);
3922 }
3923 if (rpr != NULL)
3924 LIST_REMOVE(rpr, pr_sibling);
3925
3926 prison_cleanup_unlocked(pr);
3927 mtx_lock(&pr->pr_mtx);
3928 prison_cleanup_locked(pr);
3929 if (pr->pr_flags & PR_PERSIST) {
3930 pr->pr_flags &= ~PR_PERSIST;
3931 prison_proc_free_not_last(pr);
3932 prison_free_not_last(pr);
3933 }
3934 (void)refcount_release(&pr->pr_uref);
3935 }
3936
3937 /*
3938 * Given the current locking state in the flags, make sure allprison_lock
3939 * is held exclusive, and the prison is locked. Return flags indicating
3940 * the new state.
3941 */
3942 static int
prison_lock_xlock(struct prison * pr,int flags)3943 prison_lock_xlock(struct prison *pr, int flags)
3944 {
3945
3946 if (!(flags & PD_LIST_XLOCKED)) {
3947 /*
3948 * Get allprison_lock, which may be an upgrade,
3949 * and may require unlocking the prison.
3950 */
3951 if (flags & PD_LOCKED) {
3952 mtx_unlock(&pr->pr_mtx);
3953 flags &= ~PD_LOCKED;
3954 }
3955 if (flags & PD_LIST_SLOCKED) {
3956 if (!sx_try_upgrade(&allprison_lock)) {
3957 sx_sunlock(&allprison_lock);
3958 sx_xlock(&allprison_lock);
3959 }
3960 flags &= ~PD_LIST_SLOCKED;
3961 } else
3962 sx_xlock(&allprison_lock);
3963 flags |= PD_LIST_XLOCKED;
3964 }
3965 if (!(flags & PD_LOCKED)) {
3966 /* Lock the prison mutex. */
3967 mtx_lock(&pr->pr_mtx);
3968 flags |= PD_LOCKED;
3969 }
3970 return flags;
3971 }
3972
3973 /*
3974 * Release a prison's resources when it starts dying (when the last user
3975 * reference is dropped, or when it is killed). Two functions are called,
3976 * for work that requires a locked prison or an unlocked one.
3977 */
3978 static void
prison_cleanup_locked(struct prison * pr)3979 prison_cleanup_locked(struct prison *pr)
3980 {
3981 sx_assert(&allprison_lock, SA_XLOCKED);
3982 mtx_assert(&pr->pr_mtx, MA_OWNED);
3983 prison_knote(pr, NOTE_JAIL_REMOVE);
3984 knlist_detach(pr->pr_klist);
3985 jaildesc_prison_cleanup(pr);
3986 pr->pr_klist = NULL;
3987 }
3988
3989 static void
prison_cleanup_unlocked(struct prison * pr)3990 prison_cleanup_unlocked(struct prison *pr)
3991 {
3992 sx_assert(&allprison_lock, SA_XLOCKED);
3993 mtx_assert(&pr->pr_mtx, MA_NOTOWNED);
3994 vfs_exjail_delete(pr);
3995 shm_remove_prison(pr);
3996 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
3997 }
3998
3999 /*
4000 * Set or clear a permission bit in the pr_allow field, passing restrictions
4001 * (cleared permission) down to child jails.
4002 */
4003 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)4004 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
4005 {
4006 struct prison *pr;
4007
4008 pr = cred->cr_prison;
4009 sx_slock(&allprison_lock);
4010 mtx_lock(&pr->pr_mtx);
4011 prison_set_allow_locked(pr, flag, enable);
4012 mtx_unlock(&pr->pr_mtx);
4013 sx_sunlock(&allprison_lock);
4014 }
4015
4016 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)4017 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
4018 {
4019 struct prison *cpr;
4020 int descend;
4021
4022 if (enable != 0)
4023 pr->pr_allow |= flag;
4024 else {
4025 pr->pr_allow &= ~flag;
4026 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
4027 cpr->pr_allow &= ~flag;
4028 }
4029 }
4030
4031 /*
4032 * Check if a jail supports the given address family.
4033 *
4034 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
4035 * if not.
4036 */
4037 int
prison_check_af(struct ucred * cred,int af)4038 prison_check_af(struct ucred *cred, int af)
4039 {
4040 struct prison *pr;
4041 int error;
4042
4043 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
4044
4045 pr = cred->cr_prison;
4046 #ifdef VIMAGE
4047 /* Prisons with their own network stack are not limited. */
4048 if (prison_owns_vnet(pr))
4049 return (0);
4050 #endif
4051
4052 error = 0;
4053 switch (af)
4054 {
4055 #ifdef INET
4056 case AF_INET:
4057 if (pr->pr_flags & PR_IP4)
4058 {
4059 mtx_lock(&pr->pr_mtx);
4060 if ((pr->pr_flags & PR_IP4) &&
4061 pr->pr_addrs[PR_INET] == NULL)
4062 error = EAFNOSUPPORT;
4063 mtx_unlock(&pr->pr_mtx);
4064 }
4065 break;
4066 #endif
4067 #ifdef INET6
4068 case AF_INET6:
4069 if (pr->pr_flags & PR_IP6)
4070 {
4071 mtx_lock(&pr->pr_mtx);
4072 if ((pr->pr_flags & PR_IP6) &&
4073 pr->pr_addrs[PR_INET6] == NULL)
4074 error = EAFNOSUPPORT;
4075 mtx_unlock(&pr->pr_mtx);
4076 }
4077 break;
4078 #endif
4079 case AF_LOCAL:
4080 case AF_ROUTE:
4081 case AF_NETLINK:
4082 break;
4083 default:
4084 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
4085 error = EAFNOSUPPORT;
4086 }
4087 return (error);
4088 }
4089
4090 /*
4091 * Check if given address belongs to the jail referenced by cred (wrapper to
4092 * prison_check_ip[46]).
4093 *
4094 * Returns 0 if jail doesn't restrict the address family or if address belongs
4095 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
4096 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
4097 */
4098 int
prison_if(struct ucred * cred,const struct sockaddr * sa)4099 prison_if(struct ucred *cred, const struct sockaddr *sa)
4100 {
4101 #ifdef INET
4102 const struct sockaddr_in *sai;
4103 #endif
4104 #ifdef INET6
4105 const struct sockaddr_in6 *sai6;
4106 #endif
4107 int error;
4108
4109 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
4110 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
4111
4112 #ifdef VIMAGE
4113 if (prison_owns_vnet(cred->cr_prison))
4114 return (0);
4115 #endif
4116
4117 error = 0;
4118 switch (sa->sa_family)
4119 {
4120 #ifdef INET
4121 case AF_INET:
4122 sai = (const struct sockaddr_in *)sa;
4123 error = prison_check_ip4(cred, &sai->sin_addr);
4124 break;
4125 #endif
4126 #ifdef INET6
4127 case AF_INET6:
4128 sai6 = (const struct sockaddr_in6 *)sa;
4129 error = prison_check_ip6(cred, &sai6->sin6_addr);
4130 break;
4131 #endif
4132 default:
4133 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
4134 error = EAFNOSUPPORT;
4135 }
4136 return (error);
4137 }
4138
4139 /*
4140 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
4141 */
4142 int
prison_check(struct ucred * cred1,struct ucred * cred2)4143 prison_check(struct ucred *cred1, struct ucred *cred2)
4144 {
4145
4146 return ((cred1->cr_prison == cred2->cr_prison ||
4147 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
4148 }
4149
4150 /*
4151 * For mountd/nfsd to run within a prison, it must be:
4152 * - A vnet prison.
4153 * - PR_ALLOW_NFSD must be set on it.
4154 * - The root directory (pr_root) of the prison must be
4155 * a file system mount point, so the mountd can hang
4156 * export information on it.
4157 * - The prison's enforce_statfs cannot be 0, so that
4158 * mountd(8) can do exports.
4159 */
4160 bool
prison_check_nfsd(struct ucred * cred)4161 prison_check_nfsd(struct ucred *cred)
4162 {
4163
4164 if (jailed_without_vnet(cred))
4165 return (false);
4166 if (!prison_allow(cred, PR_ALLOW_NFSD))
4167 return (false);
4168 if ((cred->cr_prison->pr_root->v_vflag & VV_ROOT) == 0)
4169 return (false);
4170 if (cred->cr_prison->pr_enforce_statfs == 0)
4171 return (false);
4172 return (true);
4173 }
4174
4175 /*
4176 * Return true if p2 is a child of p1, otherwise false.
4177 */
4178 bool
prison_ischild(struct prison * pr1,struct prison * pr2)4179 prison_ischild(struct prison *pr1, struct prison *pr2)
4180 {
4181
4182 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
4183 if (pr1 == pr2)
4184 return (true);
4185 return (false);
4186 }
4187
4188 /*
4189 * Return true if the prison is currently alive. A prison is alive if it
4190 * holds user references and it isn't being removed.
4191 */
4192 bool
prison_isalive(const struct prison * pr)4193 prison_isalive(const struct prison *pr)
4194 {
4195
4196 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
4197 return (false);
4198 return (true);
4199 }
4200
4201 /*
4202 * Return true if the prison is currently valid. A prison is valid if it has
4203 * been fully created, and is not being destroyed. Note that dying prisons
4204 * are still considered valid. Invalid prisons won't be found under normal
4205 * circumstances, as they're only put in that state by functions that have
4206 * an exclusive hold on allprison_lock.
4207 */
4208 bool
prison_isvalid(struct prison * pr)4209 prison_isvalid(struct prison *pr)
4210 {
4211
4212 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
4213 return (false);
4214 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
4215 return (false);
4216 return (true);
4217 }
4218
4219 /*
4220 * Return true if the passed credential is in a jail and that jail does not
4221 * have its own virtual network stack, otherwise false.
4222 */
4223 bool
jailed_without_vnet(struct ucred * cred)4224 jailed_without_vnet(struct ucred *cred)
4225 {
4226
4227 if (!jailed(cred))
4228 return (false);
4229 #ifdef VIMAGE
4230 if (prison_owns_vnet(cred->cr_prison))
4231 return (false);
4232 #endif
4233
4234 return (true);
4235 }
4236
4237 /*
4238 * Return the correct hostname (domainname, et al) for the passed credential.
4239 */
4240 void
getcredhostname(struct ucred * cred,char * buf,size_t size)4241 getcredhostname(struct ucred *cred, char *buf, size_t size)
4242 {
4243 struct prison *pr;
4244
4245 /*
4246 * A NULL credential can be used to shortcut to the physical
4247 * system's hostname.
4248 */
4249 pr = (cred != NULL) ? cred->cr_prison : &prison0;
4250 mtx_lock(&pr->pr_mtx);
4251 strlcpy(buf, pr->pr_hostname, size);
4252 mtx_unlock(&pr->pr_mtx);
4253 }
4254
4255 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)4256 getcreddomainname(struct ucred *cred, char *buf, size_t size)
4257 {
4258
4259 mtx_lock(&cred->cr_prison->pr_mtx);
4260 strlcpy(buf, cred->cr_prison->pr_domainname, size);
4261 mtx_unlock(&cred->cr_prison->pr_mtx);
4262 }
4263
4264 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)4265 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
4266 {
4267
4268 mtx_lock(&cred->cr_prison->pr_mtx);
4269 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
4270 mtx_unlock(&cred->cr_prison->pr_mtx);
4271 }
4272
4273 void
getcredhostid(struct ucred * cred,unsigned long * hostid)4274 getcredhostid(struct ucred *cred, unsigned long *hostid)
4275 {
4276
4277 mtx_lock(&cred->cr_prison->pr_mtx);
4278 *hostid = cred->cr_prison->pr_hostid;
4279 mtx_unlock(&cred->cr_prison->pr_mtx);
4280 }
4281
4282 void
getjailname(struct ucred * cred,char * name,size_t len)4283 getjailname(struct ucred *cred, char *name, size_t len)
4284 {
4285
4286 mtx_lock(&cred->cr_prison->pr_mtx);
4287 strlcpy(name, cred->cr_prison->pr_name, len);
4288 mtx_unlock(&cred->cr_prison->pr_mtx);
4289 }
4290
4291 #ifdef VIMAGE
4292 /*
4293 * Determine whether the prison owns its VNET.
4294 */
4295 bool
prison_owns_vnet(struct prison * pr)4296 prison_owns_vnet(struct prison *pr)
4297 {
4298
4299 /*
4300 * vnets cannot be added/removed after jail creation,
4301 * so no need to lock here.
4302 */
4303 return ((pr->pr_flags & PR_VNET) != 0);
4304 }
4305 #endif
4306
4307 /*
4308 * Determine whether the subject represented by cred can "see"
4309 * status of a mount point.
4310 * Returns: 0 for permitted, ENOENT otherwise.
4311 * XXX: This function should be called cr_canseemount() and should be
4312 * placed in kern_prot.c.
4313 */
4314 int
prison_canseemount(struct ucred * cred,struct mount * mp)4315 prison_canseemount(struct ucred *cred, struct mount *mp)
4316 {
4317 struct prison *pr;
4318 struct statfs *sp;
4319 size_t len;
4320
4321 pr = cred->cr_prison;
4322 if (pr->pr_enforce_statfs == 0)
4323 return (0);
4324 if (pr->pr_root->v_mount == mp)
4325 return (0);
4326 if (pr->pr_enforce_statfs == 2)
4327 return (ENOENT);
4328 /*
4329 * If jail's chroot directory is set to "/" we should be able to see
4330 * all mount-points from inside a jail.
4331 * This is ugly check, but this is the only situation when jail's
4332 * directory ends with '/'.
4333 */
4334 if (strcmp(pr->pr_path, "/") == 0)
4335 return (0);
4336 len = strlen(pr->pr_path);
4337 sp = &mp->mnt_stat;
4338 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
4339 return (ENOENT);
4340 /*
4341 * Be sure that we don't have situation where jail's root directory
4342 * is "/some/path" and mount point is "/some/pathpath".
4343 */
4344 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
4345 return (ENOENT);
4346 return (0);
4347 }
4348
4349 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)4350 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
4351 {
4352 char jpath[MAXPATHLEN];
4353 struct prison *pr;
4354 size_t len;
4355
4356 pr = cred->cr_prison;
4357 if (pr->pr_enforce_statfs == 0)
4358 return;
4359 if (prison_canseemount(cred, mp) != 0) {
4360 bzero(&sp->f_fsid, sizeof(sp->f_fsid));
4361 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4362 strlcpy(sp->f_mntonname, "[restricted]",
4363 sizeof(sp->f_mntonname));
4364 return;
4365 }
4366 if (pr->pr_enforce_statfs > 1)
4367 bzero(&sp->f_fsid, sizeof(sp->f_fsid));
4368 if (pr->pr_root->v_mount == mp) {
4369 /*
4370 * Clear current buffer data, so we are sure nothing from
4371 * the valid path left there.
4372 */
4373 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4374 *sp->f_mntonname = '/';
4375 return;
4376 }
4377 /*
4378 * If jail's chroot directory is set to "/" we should be able to see
4379 * all mount-points from inside a jail.
4380 */
4381 if (strcmp(pr->pr_path, "/") == 0)
4382 return;
4383 len = strlen(pr->pr_path);
4384 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
4385 /*
4386 * Clear current buffer data, so we are sure nothing from
4387 * the valid path left there.
4388 */
4389 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4390 if (*jpath == '\0') {
4391 /* Should never happen. */
4392 *sp->f_mntonname = '/';
4393 } else {
4394 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
4395 }
4396 }
4397
4398 /*
4399 * Check with permission for a specific privilege is granted within jail. We
4400 * have a specific list of accepted privileges; the rest are denied.
4401 */
4402 int
prison_priv_check(struct ucred * cred,int priv)4403 prison_priv_check(struct ucred *cred, int priv)
4404 {
4405 struct prison *pr;
4406 int error;
4407
4408 /*
4409 * Some policies have custom handlers. This routine should not be
4410 * called for them. See priv_check_cred().
4411 */
4412 switch (priv) {
4413 case PRIV_VFS_LOOKUP:
4414 case PRIV_VFS_GENERATION:
4415 KASSERT(0, ("prison_priv_check instead of a custom handler "
4416 "called for %d\n", priv));
4417 }
4418
4419 if (!jailed(cred))
4420 return (0);
4421
4422 #ifdef VIMAGE
4423 /*
4424 * Privileges specific to prisons with a virtual network stack.
4425 * There might be a duplicate entry here in case the privilege
4426 * is only granted conditionally in the legacy jail case.
4427 */
4428 switch (priv) {
4429 /*
4430 * NFS-specific privileges.
4431 */
4432 case PRIV_NFS_DAEMON:
4433 case PRIV_VFS_GETFH:
4434 case PRIV_VFS_MOUNT_EXPORTED:
4435 if (!prison_check_nfsd(cred))
4436 return (EPERM);
4437 #ifdef notyet
4438 case PRIV_NFS_LOCKD:
4439 #endif
4440 /*
4441 * Network stack privileges.
4442 */
4443 case PRIV_NET_BRIDGE:
4444 case PRIV_NET_GRE:
4445 case PRIV_NET_BPF:
4446 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
4447 case PRIV_NET_ROUTE:
4448 case PRIV_NET_TAP:
4449 case PRIV_NET_SETIFMTU:
4450 case PRIV_NET_SETIFFLAGS:
4451 case PRIV_NET_SETIFCAP:
4452 case PRIV_NET_SETIFDESCR:
4453 case PRIV_NET_SETIFNAME :
4454 case PRIV_NET_SETIFMETRIC:
4455 case PRIV_NET_SETIFPHYS:
4456 case PRIV_NET_SETIFMAC:
4457 case PRIV_NET_SETLANPCP:
4458 case PRIV_NET_ADDMULTI:
4459 case PRIV_NET_DELMULTI:
4460 case PRIV_NET_HWIOCTL:
4461 case PRIV_NET_SETLLADDR:
4462 case PRIV_NET_ADDIFGROUP:
4463 case PRIV_NET_DELIFGROUP:
4464 case PRIV_NET_IFCREATE:
4465 case PRIV_NET_IFDESTROY:
4466 case PRIV_NET_ADDIFADDR:
4467 case PRIV_NET_DELIFADDR:
4468 case PRIV_NET_LAGG:
4469 case PRIV_NET_GIF:
4470 case PRIV_NET_SETIFVNET:
4471 case PRIV_NET_SETIFFIB:
4472 case PRIV_NET_OVPN:
4473 case PRIV_NET_GENEVE:
4474 case PRIV_NET_ME:
4475 case PRIV_NET_WG:
4476
4477 /*
4478 * 802.11-related privileges.
4479 */
4480 case PRIV_NET80211_VAP_GETKEY:
4481 case PRIV_NET80211_VAP_MANAGE:
4482
4483 #ifdef notyet
4484 /*
4485 * ATM privileges.
4486 */
4487 case PRIV_NETATM_CFG:
4488 case PRIV_NETATM_ADD:
4489 case PRIV_NETATM_DEL:
4490 case PRIV_NETATM_SET:
4491
4492 /*
4493 * Bluetooth privileges.
4494 */
4495 case PRIV_NETBLUETOOTH_RAW:
4496 #endif
4497
4498 /*
4499 * Netgraph and netgraph module privileges.
4500 */
4501 case PRIV_NETGRAPH_CONTROL:
4502 #ifdef notyet
4503 case PRIV_NETGRAPH_TTY:
4504 #endif
4505
4506 /*
4507 * IPv4 and IPv6 privileges.
4508 */
4509 case PRIV_NETINET_IPFW:
4510 case PRIV_NETINET_DIVERT:
4511 case PRIV_NETINET_PF:
4512 case PRIV_NETINET_DUMMYNET:
4513 case PRIV_NETINET_CARP:
4514 case PRIV_NETINET_MROUTE:
4515 case PRIV_NETINET_RAW:
4516 case PRIV_NETINET_ADDRCTRL6:
4517 case PRIV_NETINET_ND6:
4518 case PRIV_NETINET_SCOPE6:
4519 case PRIV_NETINET_ALIFETIME6:
4520 case PRIV_NETINET_IPSEC:
4521 case PRIV_NETINET_BINDANY:
4522
4523 #ifdef notyet
4524 /*
4525 * NCP privileges.
4526 */
4527 case PRIV_NETNCP:
4528
4529 /*
4530 * SMB privileges.
4531 */
4532 case PRIV_NETSMB:
4533 #endif
4534
4535 /*
4536 * No default: or deny here.
4537 * In case of no permit fall through to next switch().
4538 */
4539 if (cred->cr_prison->pr_flags & PR_VNET)
4540 return (0);
4541 }
4542 #endif /* VIMAGE */
4543
4544 switch (priv) {
4545 /*
4546 * Allow ktrace privileges for root in jail.
4547 */
4548 case PRIV_KTRACE:
4549 return (0);
4550
4551 /*
4552 * Allow jailed processes to configure audit identity and
4553 * submit audit records (login, etc). In the future we may
4554 * want to further refine the relationship between audit and
4555 * jail.
4556 */
4557 case PRIV_AUDIT_GETAUDIT:
4558 case PRIV_AUDIT_SETAUDIT:
4559 if (cred->cr_prison->pr_allow & PR_ALLOW_SETAUDIT)
4560 return (0);
4561 else
4562 return (EPERM);
4563 #if 0
4564 case PRIV_AUDIT_SUBMIT:
4565 #endif
4566
4567 /*
4568 * Allow jailed processes to manipulate process UNIX
4569 * credentials in any way they see fit.
4570 */
4571 case PRIV_CRED_SETCRED:
4572 case PRIV_CRED_SETUID:
4573 case PRIV_CRED_SETEUID:
4574 case PRIV_CRED_SETGID:
4575 case PRIV_CRED_SETEGID:
4576 case PRIV_CRED_SETGROUPS:
4577 case PRIV_CRED_SETREUID:
4578 case PRIV_CRED_SETREGID:
4579 case PRIV_CRED_SETRESUID:
4580 case PRIV_CRED_SETRESGID:
4581
4582 /*
4583 * Jail implements visibility constraints already, so allow
4584 * jailed root to override uid/gid-based constraints.
4585 */
4586 case PRIV_SEEOTHERGIDS:
4587 case PRIV_SEEOTHERUIDS:
4588 case PRIV_SEEJAILPROC:
4589
4590 /*
4591 * Jail implements inter-process debugging limits already, so
4592 * allow jailed root various debugging privileges.
4593 */
4594 case PRIV_DEBUG_DIFFCRED:
4595 case PRIV_DEBUG_SUGID:
4596 case PRIV_DEBUG_UNPRIV:
4597 case PRIV_DEBUG_DIFFJAIL:
4598
4599 /*
4600 * Allow jail to set various resource limits and login
4601 * properties, and for now, exceed process resource limits.
4602 */
4603 case PRIV_PROC_LIMIT:
4604 case PRIV_PROC_SETLOGIN:
4605 case PRIV_PROC_SETRLIMIT:
4606
4607 /*
4608 * Debuggers should work in jails.
4609 */
4610 case PRIV_PROC_MEM_WRITE:
4611
4612 /*
4613 * System V and POSIX IPC privileges are granted in jail.
4614 */
4615 case PRIV_IPC_READ:
4616 case PRIV_IPC_WRITE:
4617 case PRIV_IPC_ADMIN:
4618 case PRIV_IPC_MSGSIZE:
4619 case PRIV_MQ_ADMIN:
4620
4621 /*
4622 * Jail operations within a jail work on child jails.
4623 */
4624 case PRIV_JAIL_ATTACH:
4625 case PRIV_JAIL_SET:
4626 case PRIV_JAIL_REMOVE:
4627
4628 /*
4629 * Jail implements its own inter-process limits, so allow
4630 * root processes in jail to change scheduling on other
4631 * processes in the same jail. Likewise for signalling.
4632 */
4633 case PRIV_SCHED_DIFFCRED:
4634 case PRIV_SCHED_CPUSET:
4635 case PRIV_SCHED_DIFFJAIL:
4636 case PRIV_SIGNAL_DIFFCRED:
4637 case PRIV_SIGNAL_SUGID:
4638 case PRIV_SIGNAL_DIFFJAIL:
4639
4640 /*
4641 * Allow jailed processes to write to sysctls marked as jail
4642 * writable.
4643 */
4644 case PRIV_SYSCTL_WRITEJAIL:
4645
4646 /*
4647 * Allow root in jail to manage a variety of quota
4648 * properties. These should likely be conditional on a
4649 * configuration option.
4650 */
4651 case PRIV_VFS_GETQUOTA:
4652 case PRIV_VFS_SETQUOTA:
4653
4654 /*
4655 * Since Jail relies on chroot() to implement file system
4656 * protections, grant many VFS privileges to root in jail.
4657 * Be careful to exclude mount-related and NFS-related
4658 * privileges.
4659 */
4660 case PRIV_VFS_READ:
4661 case PRIV_VFS_WRITE:
4662 case PRIV_VFS_ADMIN:
4663 case PRIV_VFS_EXEC:
4664 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
4665 case PRIV_VFS_CHFLAGS_DEV:
4666 case PRIV_VFS_CHOWN:
4667 case PRIV_VFS_CHROOT:
4668 case PRIV_VFS_RETAINSUGID:
4669 case PRIV_VFS_FCHROOT:
4670 case PRIV_VFS_LINK:
4671 case PRIV_VFS_SETGID:
4672 case PRIV_VFS_STAT:
4673 case PRIV_VFS_STICKYFILE:
4674
4675 /*
4676 * As in the non-jail case, non-root users are expected to be
4677 * able to read kernel/physical memory (provided /dev/[k]mem
4678 * exists in the jail and they have permission to access it).
4679 */
4680 case PRIV_KMEM_READ:
4681 return (0);
4682
4683 /*
4684 * Depending on the global setting, allow privilege of
4685 * setting system flags.
4686 */
4687 case PRIV_VFS_SYSFLAGS:
4688 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4689 return (0);
4690 else
4691 return (EPERM);
4692
4693 /*
4694 * Depending on the global setting, allow privilege of
4695 * mounting/unmounting file systems.
4696 */
4697 case PRIV_VFS_MOUNT:
4698 case PRIV_VFS_UNMOUNT:
4699 case PRIV_VFS_MOUNT_NONUSER:
4700 case PRIV_VFS_MOUNT_OWNER:
4701 pr = cred->cr_prison;
4702 prison_lock(pr);
4703 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
4704 error = 0;
4705 else
4706 error = EPERM;
4707 prison_unlock(pr);
4708 return (error);
4709
4710 /*
4711 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
4712 * policy. priv_check_cred will not specifically allow it, and
4713 * we may want a MAC policy to allow it.
4714 */
4715 case PRIV_VFS_READ_DIR:
4716 return (0);
4717
4718 /*
4719 * Conditionally allow privileged process in the jail to
4720 * manipulate filesystem extended attributes in the system
4721 * namespace.
4722 */
4723 case PRIV_VFS_EXTATTR_SYSTEM:
4724 if ((cred->cr_prison->pr_allow & PR_ALLOW_EXTATTR) != 0)
4725 return (0);
4726 else
4727 return (EPERM);
4728
4729 /*
4730 * Conditionnaly allow locking (unlocking) physical pages
4731 * in memory.
4732 */
4733 case PRIV_VM_MLOCK:
4734 case PRIV_VM_MUNLOCK:
4735 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
4736 return (0);
4737 else
4738 return (EPERM);
4739
4740 /*
4741 * Conditionally allow jailed root to bind reserved ports.
4742 */
4743 case PRIV_NETINET_RESERVEDPORT:
4744 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
4745 return (0);
4746 else
4747 return (EPERM);
4748
4749 /*
4750 * Allow jailed root to reuse in-use ports.
4751 */
4752 case PRIV_NETINET_REUSEPORT:
4753 return (0);
4754
4755 /*
4756 * Allow jailed root to set certain IPv4/6 (option) headers.
4757 */
4758 case PRIV_NETINET_SETHDROPTS:
4759 return (0);
4760
4761 /*
4762 * Conditionally allow creating raw sockets in jail.
4763 */
4764 case PRIV_NETINET_RAW:
4765 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4766 return (0);
4767 else
4768 return (EPERM);
4769
4770 /*
4771 * Since jail implements its own visibility limits on netstat
4772 * sysctls, allow getcred. This allows identd to work in
4773 * jail.
4774 */
4775 case PRIV_NETINET_GETCRED:
4776 return (0);
4777
4778 /*
4779 * Allow jailed root to set loginclass.
4780 */
4781 case PRIV_PROC_SETLOGINCLASS:
4782 return (0);
4783
4784 /*
4785 * Do not allow a process inside a jail to read the kernel
4786 * message buffer unless explicitly permitted.
4787 */
4788 case PRIV_MSGBUF:
4789 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
4790 return (0);
4791 return (EPERM);
4792
4793 /*
4794 * Conditionally allow privileged process in the jail adjust
4795 * machine time.
4796 */
4797 case PRIV_ADJTIME:
4798 case PRIV_NTP_ADJTIME:
4799 if (cred->cr_prison->pr_allow &
4800 (PR_ALLOW_ADJTIME | PR_ALLOW_SETTIME)) {
4801 return (0);
4802 }
4803 return (EPERM);
4804
4805 /*
4806 * Conditionally allow privileged process in the jail set
4807 * machine time.
4808 */
4809 case PRIV_SETTIMEOFDAY:
4810 case PRIV_CLOCK_SETTIME:
4811 if (cred->cr_prison->pr_allow & PR_ALLOW_SETTIME)
4812 return (0);
4813 else
4814 return (EPERM);
4815
4816 /*
4817 * Conditionally allow privileged process in the jail to modify
4818 * the routing table.
4819 */
4820 case PRIV_NET_ROUTE:
4821 if (cred->cr_prison->pr_allow & PR_ALLOW_ROUTING)
4822 return (0);
4823 else
4824 return (EPERM);
4825
4826 case PRIV_VMM_CREATE:
4827 case PRIV_VMM_DESTROY:
4828 /*
4829 * Jailed root can create and destroy VMs; the vmm module
4830 * additionally checks for the allow.vmm flag.
4831 */
4832 return (0);
4833
4834 case PRIV_VMM_PPTDEV:
4835 /*
4836 * Allow jailed root to manage passthrough devices. vmm(4) also
4837 * checks for the dynamically added allow.vmm_ppt.
4838 */
4839 return (0);
4840
4841 default:
4842 /*
4843 * In all remaining cases, deny the privilege request. This
4844 * includes almost all network privileges, many system
4845 * configuration privileges.
4846 */
4847 return (EPERM);
4848 }
4849 }
4850
4851 /*
4852 * Return the part of pr2's name that is relative to pr1, or the whole name
4853 * if it does not directly follow.
4854 */
4855
4856 char *
prison_name(struct prison * pr1,struct prison * pr2)4857 prison_name(struct prison *pr1, struct prison *pr2)
4858 {
4859 char *name;
4860
4861 /* Jails see themselves as "0" (if they see themselves at all). */
4862 if (pr1 == pr2)
4863 return "0";
4864 name = pr2->pr_name;
4865 if (prison_ischild(pr1, pr2)) {
4866 /*
4867 * pr1 isn't locked (and allprison_lock may not be either)
4868 * so its length can't be counted on. But the number of dots
4869 * can be counted on - and counted.
4870 */
4871 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4872 name = strchr(name, '.') + 1;
4873 }
4874 return (name);
4875 }
4876
4877 /*
4878 * Return the part of pr2's path that is relative to pr1, or the whole path
4879 * if it does not directly follow.
4880 */
4881 static char *
prison_path(struct prison * pr1,struct prison * pr2)4882 prison_path(struct prison *pr1, struct prison *pr2)
4883 {
4884 char *path1, *path2;
4885 int len1;
4886
4887 path1 = pr1->pr_path;
4888 path2 = pr2->pr_path;
4889 if (!strcmp(path1, "/"))
4890 return (path2);
4891 len1 = strlen(path1);
4892 if (strncmp(path1, path2, len1))
4893 return (path2);
4894 if (path2[len1] == '\0')
4895 return "/";
4896 if (path2[len1] == '/')
4897 return (path2 + len1);
4898 return (path2);
4899 }
4900
4901 /*
4902 * Jail-related sysctls.
4903 */
4904 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4905 "Jails");
4906
4907 #if defined(INET) || defined(INET6)
4908 /*
4909 * Copy address array to memory that would be then SYSCTL_OUT-ed.
4910 * sysctl_jail_list() helper.
4911 */
4912 static void
prison_ip_copyout(struct prison * pr,const pr_family_t af,void ** out,int * len)4913 prison_ip_copyout(struct prison *pr, const pr_family_t af, void **out, int *len)
4914 {
4915 const struct prison_ip *pip;
4916 const size_t size = pr_families[af].size;
4917
4918 again:
4919 mtx_assert(&pr->pr_mtx, MA_OWNED);
4920 if ((pip = pr->pr_addrs[af]) != NULL) {
4921 if (*len < pip->ips) {
4922 *len = pip->ips;
4923 mtx_unlock(&pr->pr_mtx);
4924 *out = realloc(*out, *len * size, M_TEMP, M_WAITOK);
4925 mtx_lock(&pr->pr_mtx);
4926 goto again;
4927 }
4928 bcopy(pip->pr_ip, *out, pip->ips * size);
4929 }
4930 }
4931 #endif
4932
4933 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)4934 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4935 {
4936 struct xprison *xp;
4937 struct prison *pr, *cpr;
4938 #ifdef INET
4939 struct in_addr *ip4 = NULL;
4940 int ip4s = 0;
4941 #endif
4942 #ifdef INET6
4943 struct in6_addr *ip6 = NULL;
4944 int ip6s = 0;
4945 #endif
4946 int descend, error;
4947
4948 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4949 pr = req->td->td_ucred->cr_prison;
4950 error = 0;
4951 sx_slock(&allprison_lock);
4952 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4953 mtx_lock(&cpr->pr_mtx);
4954 #ifdef INET
4955 prison_ip_copyout(cpr, PR_INET, (void **)&ip4, &ip4s);
4956 #endif
4957 #ifdef INET6
4958 prison_ip_copyout(cpr, PR_INET6, (void **)&ip6, &ip6s);
4959 #endif
4960 bzero(xp, sizeof(*xp));
4961 xp->pr_version = XPRISON_VERSION;
4962 xp->pr_id = cpr->pr_id;
4963 xp->pr_state = cpr->pr_state;
4964 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4965 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4966 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4967 #ifdef INET
4968 xp->pr_ip4s = ip4s;
4969 #endif
4970 #ifdef INET6
4971 xp->pr_ip6s = ip6s;
4972 #endif
4973 mtx_unlock(&cpr->pr_mtx);
4974 error = SYSCTL_OUT(req, xp, sizeof(*xp));
4975 if (error)
4976 break;
4977 #ifdef INET
4978 if (xp->pr_ip4s > 0) {
4979 error = SYSCTL_OUT(req, ip4,
4980 xp->pr_ip4s * sizeof(struct in_addr));
4981 if (error)
4982 break;
4983 }
4984 #endif
4985 #ifdef INET6
4986 if (xp->pr_ip6s > 0) {
4987 error = SYSCTL_OUT(req, ip6,
4988 xp->pr_ip6s * sizeof(struct in6_addr));
4989 if (error)
4990 break;
4991 }
4992 #endif
4993 }
4994 sx_sunlock(&allprison_lock);
4995 free(xp, M_TEMP);
4996 #ifdef INET
4997 free(ip4, M_TEMP);
4998 #endif
4999 #ifdef INET6
5000 free(ip6, M_TEMP);
5001 #endif
5002 return (error);
5003 }
5004
5005 SYSCTL_OID(_security_jail, OID_AUTO, list,
5006 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
5007 sysctl_jail_list, "S", "List of active jails");
5008
5009 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)5010 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
5011 {
5012 int error, injail;
5013
5014 injail = jailed(req->td->td_ucred);
5015 error = SYSCTL_OUT(req, &injail, sizeof(injail));
5016
5017 return (error);
5018 }
5019
5020 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
5021 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
5022 sysctl_jail_jailed, "I", "Process in jail?");
5023
5024 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)5025 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
5026 {
5027 int error, havevnet;
5028 #ifdef VIMAGE
5029 struct ucred *cred = req->td->td_ucred;
5030
5031 havevnet = jailed(cred) && prison_owns_vnet(cred->cr_prison);
5032 #else
5033 havevnet = 0;
5034 #endif
5035 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
5036
5037 return (error);
5038 }
5039
5040 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
5041 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
5042 sysctl_jail_vnet, "I", "Jail owns vnet?");
5043
5044 #if defined(INET) || defined(INET6)
5045 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
5046 &jail_max_af_ips, 0,
5047 "Number of IP addresses a jail may have at most per address family (deprecated)");
5048 #endif
5049
5050 /*
5051 * Default parameters for jail(2) compatibility. For historical reasons,
5052 * the sysctl names have varying similarity to the parameter names. Prisons
5053 * just see their own parameters, and can't change them.
5054 */
5055 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)5056 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
5057 {
5058 int error, i;
5059
5060 /* Get the current flag value, and convert it to a boolean. */
5061 if (req->td->td_ucred->cr_prison == &prison0) {
5062 mtx_lock(&prison0.pr_mtx);
5063 i = (jail_default_allow & arg2) != 0;
5064 mtx_unlock(&prison0.pr_mtx);
5065 } else
5066 i = prison_allow(req->td->td_ucred, arg2);
5067
5068 if (arg1 != NULL)
5069 i = !i;
5070 error = sysctl_handle_int(oidp, &i, 0, req);
5071 if (error || !req->newptr)
5072 return (error);
5073 i = i ? arg2 : 0;
5074 if (arg1 != NULL)
5075 i ^= arg2;
5076 /*
5077 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
5078 * for writing.
5079 */
5080 mtx_lock(&prison0.pr_mtx);
5081 jail_default_allow = (jail_default_allow & ~arg2) | i;
5082 mtx_unlock(&prison0.pr_mtx);
5083 return (0);
5084 }
5085
5086 /*
5087 * Do not add more here. Use SYSCTL_JAIL_PARAM (allow flags for jails)
5088 * instead.
5089 */
5090 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
5091 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5092 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
5093 "Processes in jail can set their hostnames (deprecated)");
5094 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
5095 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5096 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
5097 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
5098 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
5099 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5100 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
5101 "Processes in jail can use System V IPC primitives (deprecated)");
5102 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
5103 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5104 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
5105 "Prison root can create raw sockets (deprecated)");
5106 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
5107 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5108 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
5109 "Processes in jail can alter system file flags (deprecated)");
5110 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
5111 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5112 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
5113 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
5114 SYSCTL_PROC(_security_jail, OID_AUTO, mlock_allowed,
5115 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5116 NULL, PR_ALLOW_MLOCK, sysctl_jail_default_allow, "I",
5117 "Processes in jail can lock/unlock physical pages in memory (deprecated)");
5118
5119 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)5120 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
5121 {
5122 struct prison *pr;
5123 int level, error;
5124
5125 pr = req->td->td_ucred->cr_prison;
5126 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
5127 error = sysctl_handle_int(oidp, &level, 0, req);
5128 if (error || !req->newptr)
5129 return (error);
5130 *(int *)arg1 = level;
5131 return (0);
5132 }
5133
5134 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
5135 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5136 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
5137 sysctl_jail_default_level, "I",
5138 "Processes in jail cannot see all mounted file systems (deprecated)");
5139
5140 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
5141 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
5142 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
5143 sysctl_jail_default_level, "I",
5144 "Ruleset for the devfs filesystem in jail (deprecated)");
5145
5146 SYSCTL_NODE(_security_jail, OID_AUTO, children, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
5147 "Limits and stats of child jails");
5148
5149 static int
sysctl_jail_children(SYSCTL_HANDLER_ARGS)5150 sysctl_jail_children(SYSCTL_HANDLER_ARGS)
5151 {
5152 struct prison *pr;
5153 int i;
5154
5155 pr = req->td->td_ucred->cr_prison;
5156
5157 switch (oidp->oid_kind & CTLTYPE) {
5158 case CTLTYPE_INT:
5159 i = *(int *)((char *)pr + arg2);
5160 return (SYSCTL_OUT(req, &i, sizeof(i)));
5161 }
5162
5163 return (0);
5164 }
5165
5166 SYSCTL_PROC(_security_jail_children, OID_AUTO, max,
5167 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
5168 NULL, offsetof(struct prison, pr_childmax), sysctl_jail_children,
5169 "I", "Maximum number of child jails");
5170 SYSCTL_PROC(_security_jail_children, OID_AUTO, cur,
5171 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
5172 NULL, offsetof(struct prison, pr_childcount), sysctl_jail_children,
5173 "I", "Current number of child jails");
5174
5175 /*
5176 * Nodes to describe jail parameters. Maximum length of string parameters
5177 * is returned in the string itself, and the other parameters exist merely
5178 * to make themselves and their types known.
5179 */
5180 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
5181 "Jail parameters");
5182
5183 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)5184 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
5185 {
5186 int i;
5187 long l;
5188 size_t s;
5189 char numbuf[12];
5190
5191 switch (oidp->oid_kind & CTLTYPE)
5192 {
5193 case CTLTYPE_LONG:
5194 case CTLTYPE_ULONG:
5195 l = 0;
5196 #ifdef SCTL_MASK32
5197 if (!(req->flags & SCTL_MASK32))
5198 #endif
5199 return (SYSCTL_OUT(req, &l, sizeof(l)));
5200 case CTLTYPE_INT:
5201 case CTLTYPE_UINT:
5202 i = 0;
5203 return (SYSCTL_OUT(req, &i, sizeof(i)));
5204 case CTLTYPE_STRING:
5205 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
5206 return
5207 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
5208 case CTLTYPE_STRUCT:
5209 s = (size_t)arg2;
5210 return (SYSCTL_OUT(req, &s, sizeof(s)));
5211 }
5212 return (0);
5213 }
5214
5215 /*
5216 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
5217 * jail creation time but cannot be changed in an existing jail.
5218 */
5219 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
5220 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
5221 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
5222 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
5223 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
5224 "I", "Jail secure level");
5225 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
5226 "Jail value for kern.osreldate and uname -K");
5227 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
5228 "Jail value for kern.osrelease and uname -r");
5229 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
5230 "I", "Jail cannot see all mounted file systems");
5231 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
5232 "I", "Ruleset for in-jail devfs mounts");
5233 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
5234 "B", "Jail persistence");
5235 #ifdef VIMAGE
5236 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
5237 "E,jailsys", "Virtual network stack");
5238 #endif
5239 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
5240 "B", "Jail is in the process of shutting down");
5241
5242 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
5243 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
5244 "I", "Current number of child jails");
5245 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
5246 "I", "Maximum number of child jails");
5247
5248 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
5249 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
5250 "Jail hostname");
5251 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
5252 "Jail NIS domainname");
5253 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
5254 "Jail host UUID");
5255 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
5256 "LU", "Jail host ID");
5257
5258 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
5259 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
5260
5261 #ifdef MAC
5262 SYSCTL_JAIL_PARAM_STRUCT(_mac, label, CTLFLAG_RW, sizeof(struct mac),
5263 "S,mac", "Jail MAC label");
5264 #endif
5265
5266 #ifdef INET
5267 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
5268 "Jail IPv4 address virtualization");
5269 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
5270 "S,in_addr,a", "Jail IPv4 addresses");
5271 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5272 "B", "Do (not) use IPv4 source address selection rather than the "
5273 "primary jail IPv4 address.");
5274 #endif
5275 #ifdef INET6
5276 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
5277 "Jail IPv6 address virtualization");
5278 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
5279 "S,in6_addr,a", "Jail IPv6 addresses");
5280 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
5281 "B", "Do (not) use IPv6 source address selection rather than the "
5282 "primary jail IPv6 address.");
5283 #endif
5284
5285 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
5286 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
5287 "B", "Jail may set hostname");
5288 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
5289 "B", "Jail may use SYSV IPC");
5290 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
5291 "B", "Jail may create raw sockets");
5292 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
5293 "B", "Jail may alter system file flags");
5294 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
5295 "B", "Jail may set file quotas");
5296 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
5297 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
5298 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
5299 "B", "Jail may lock (unlock) physical pages in memory");
5300 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
5301 "B", "Jail may bind sockets to reserved ports");
5302 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
5303 "B", "Jail may read the kernel message buffer");
5304 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
5305 "B", "Unprivileged processes may use process debugging facilities");
5306 SYSCTL_JAIL_PARAM(_allow, unprivileged_parent_tampering,
5307 CTLTYPE_INT | CTLFLAG_RW, "B",
5308 "Unprivileged parent jail processes may tamper with same-uid processes"
5309 " (signal/debug/cpuset)");
5310 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
5311 "B", "Processes in jail with uid 0 have privilege");
5312 #ifdef VIMAGE
5313 SYSCTL_JAIL_PARAM(_allow, nfsd, CTLTYPE_INT | CTLFLAG_RW,
5314 "B", "Mountd/nfsd may run in the jail");
5315 #endif
5316 SYSCTL_JAIL_PARAM(_allow, extattr, CTLTYPE_INT | CTLFLAG_RW,
5317 "B", "Jail may set system-level filesystem extended attributes");
5318 SYSCTL_JAIL_PARAM(_allow, adjtime, CTLTYPE_INT | CTLFLAG_RW,
5319 "B", "Jail may adjust system time");
5320 SYSCTL_JAIL_PARAM(_allow, settime, CTLTYPE_INT | CTLFLAG_RW,
5321 "B", "Jail may set system time");
5322 SYSCTL_JAIL_PARAM(_allow, routing, CTLTYPE_INT | CTLFLAG_RW,
5323 "B", "Jail may modify routing table");
5324 #ifdef AUDIT
5325 SYSCTL_JAIL_PARAM(_allow, setaudit, CTLTYPE_INT | CTLFLAG_RW,
5326 "B", "Jail may set and get audit session state");
5327 #endif
5328
5329 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
5330 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
5331 "B", "Jail may mount/unmount jail-friendly file systems in general");
5332
5333 /*
5334 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
5335 * its associated bit in the pr_allow bitmask, or zero if the parameter was
5336 * not created.
5337 */
5338 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)5339 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
5340 const char *descr)
5341 {
5342 struct bool_flags *bf;
5343 struct sysctl_oid *parent;
5344 char *allow_name, *allow_noname, *allowed;
5345 #ifndef NO_SYSCTL_DESCR
5346 char *descr_deprecated;
5347 #endif
5348 u_int allow_flag;
5349
5350 if (prefix
5351 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
5352 < 0 ||
5353 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
5354 < 0
5355 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
5356 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
5357 free(allow_name, M_PRISON);
5358 return 0;
5359 }
5360
5361 /*
5362 * See if this parameter has already beed added, i.e. a module was
5363 * previously loaded/unloaded.
5364 */
5365 mtx_lock(&prison0.pr_mtx);
5366 for (bf = pr_flag_allow;
5367 bf < pr_flag_allow + nitems(pr_flag_allow) &&
5368 atomic_load_int(&bf->flag) != 0;
5369 bf++) {
5370 if (strcmp(bf->name, allow_name) == 0) {
5371 allow_flag = bf->flag;
5372 goto no_add;
5373 }
5374 }
5375
5376 /*
5377 * Find a free bit in pr_allow_all, failing if there are none
5378 * (which shouldn't happen as long as we keep track of how many
5379 * potential dynamic flags exist).
5380 */
5381 for (allow_flag = 1;; allow_flag <<= 1) {
5382 if (allow_flag == 0)
5383 goto no_add;
5384 if ((pr_allow_all & allow_flag) == 0)
5385 break;
5386 }
5387
5388 /* Note the parameter in the next open slot in pr_flag_allow. */
5389 for (bf = pr_flag_allow; ; bf++) {
5390 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
5391 /* This should never happen, but is not fatal. */
5392 allow_flag = 0;
5393 goto no_add;
5394 }
5395 if (atomic_load_int(&bf->flag) == 0)
5396 break;
5397 }
5398 bf->name = allow_name;
5399 bf->noname = allow_noname;
5400 pr_allow_all |= allow_flag;
5401 /*
5402 * prison0 always has permission for the new parameter.
5403 * Other jails must have it granted to them.
5404 */
5405 prison0.pr_allow |= allow_flag;
5406 /* The flag indicates a valid entry, so make sure it is set last. */
5407 atomic_store_rel_int(&bf->flag, allow_flag);
5408 mtx_unlock(&prison0.pr_mtx);
5409
5410 /*
5411 * Create sysctls for the parameter, and the back-compat global
5412 * permission.
5413 */
5414 parent = prefix
5415 ? SYSCTL_ADD_NODE(NULL,
5416 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
5417 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
5418 : &sysctl___security_jail_param_allow;
5419 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
5420 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
5421 NULL, 0, sysctl_jail_param, "B", descr);
5422 if ((prefix
5423 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
5424 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
5425 #ifndef NO_SYSCTL_DESCR
5426 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
5427 descr);
5428 #endif
5429 (void)SYSCTL_ADD_PROC(NULL,
5430 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
5431 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
5432 sysctl_jail_default_allow, "I", descr_deprecated);
5433 #ifndef NO_SYSCTL_DESCR
5434 free(descr_deprecated, M_TEMP);
5435 #endif
5436 free(allowed, M_TEMP);
5437 }
5438 return allow_flag;
5439
5440 no_add:
5441 mtx_unlock(&prison0.pr_mtx);
5442 free(allow_name, M_PRISON);
5443 free(allow_noname, M_PRISON);
5444 return allow_flag;
5445 }
5446
5447 /*
5448 * The VFS system will register jail-aware filesystems here. They each get
5449 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
5450 * attempts to mount.
5451 */
5452 void
prison_add_vfs(struct vfsconf * vfsp)5453 prison_add_vfs(struct vfsconf *vfsp)
5454 {
5455 #ifdef NO_SYSCTL_DESCR
5456
5457 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5458 NULL, NULL);
5459 #else
5460 char *descr;
5461
5462 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
5463 vfsp->vfc_name);
5464 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
5465 NULL, descr);
5466 free(descr, M_TEMP);
5467 #endif
5468 }
5469
5470 #ifdef RACCT
5471 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)5472 prison_racct_foreach(void (*callback)(struct racct *racct,
5473 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
5474 void *arg2, void *arg3)
5475 {
5476 struct prison_racct *prr;
5477
5478 ASSERT_RACCT_ENABLED();
5479
5480 sx_slock(&allprison_lock);
5481 if (pre != NULL)
5482 (pre)();
5483 LIST_FOREACH(prr, &allprison_racct, prr_next)
5484 (callback)(prr->prr_racct, arg2, arg3);
5485 if (post != NULL)
5486 (post)();
5487 sx_sunlock(&allprison_lock);
5488 }
5489
5490 static struct prison_racct *
prison_racct_find_locked(const char * name)5491 prison_racct_find_locked(const char *name)
5492 {
5493 struct prison_racct *prr;
5494
5495 ASSERT_RACCT_ENABLED();
5496 sx_assert(&allprison_lock, SA_XLOCKED);
5497
5498 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
5499 return (NULL);
5500
5501 LIST_FOREACH(prr, &allprison_racct, prr_next) {
5502 if (strcmp(name, prr->prr_name) != 0)
5503 continue;
5504
5505 /* Found prison_racct with a matching name? */
5506 prison_racct_hold(prr);
5507 return (prr);
5508 }
5509
5510 /* Add new prison_racct. */
5511 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
5512 racct_create(&prr->prr_racct);
5513
5514 strcpy(prr->prr_name, name);
5515 refcount_init(&prr->prr_refcount, 1);
5516 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
5517
5518 return (prr);
5519 }
5520
5521 struct prison_racct *
prison_racct_find(const char * name)5522 prison_racct_find(const char *name)
5523 {
5524 struct prison_racct *prr;
5525
5526 ASSERT_RACCT_ENABLED();
5527
5528 sx_xlock(&allprison_lock);
5529 prr = prison_racct_find_locked(name);
5530 sx_xunlock(&allprison_lock);
5531 return (prr);
5532 }
5533
5534 void
prison_racct_hold(struct prison_racct * prr)5535 prison_racct_hold(struct prison_racct *prr)
5536 {
5537
5538 ASSERT_RACCT_ENABLED();
5539
5540 refcount_acquire(&prr->prr_refcount);
5541 }
5542
5543 static void
prison_racct_free_locked(struct prison_racct * prr)5544 prison_racct_free_locked(struct prison_racct *prr)
5545 {
5546
5547 ASSERT_RACCT_ENABLED();
5548 sx_assert(&allprison_lock, SA_XLOCKED);
5549
5550 if (refcount_release(&prr->prr_refcount)) {
5551 racct_destroy(&prr->prr_racct);
5552 LIST_REMOVE(prr, prr_next);
5553 free(prr, M_PRISON_RACCT);
5554 }
5555 }
5556
5557 void
prison_racct_free(struct prison_racct * prr)5558 prison_racct_free(struct prison_racct *prr)
5559 {
5560
5561 ASSERT_RACCT_ENABLED();
5562 sx_assert(&allprison_lock, SA_UNLOCKED);
5563
5564 if (refcount_release_if_not_last(&prr->prr_refcount))
5565 return;
5566
5567 sx_xlock(&allprison_lock);
5568 prison_racct_free_locked(prr);
5569 sx_xunlock(&allprison_lock);
5570 }
5571
5572 static void
prison_racct_attach(struct prison * pr)5573 prison_racct_attach(struct prison *pr)
5574 {
5575 struct prison_racct *prr;
5576
5577 ASSERT_RACCT_ENABLED();
5578 sx_assert(&allprison_lock, SA_XLOCKED);
5579
5580 prr = prison_racct_find_locked(pr->pr_name);
5581 KASSERT(prr != NULL, ("cannot find prison_racct"));
5582
5583 pr->pr_prison_racct = prr;
5584 }
5585
5586 /*
5587 * Handle jail renaming. From the racct point of view, renaming means
5588 * moving from one prison_racct to another.
5589 */
5590 static void
prison_racct_modify(struct prison * pr)5591 prison_racct_modify(struct prison *pr)
5592 {
5593 #ifdef RCTL
5594 struct proc *p;
5595 struct ucred *cred;
5596 #endif
5597 struct prison_racct *oldprr;
5598
5599 ASSERT_RACCT_ENABLED();
5600
5601 sx_slock(&allproc_lock);
5602 sx_xlock(&allprison_lock);
5603
5604 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
5605 sx_xunlock(&allprison_lock);
5606 sx_sunlock(&allproc_lock);
5607 return;
5608 }
5609
5610 oldprr = pr->pr_prison_racct;
5611 pr->pr_prison_racct = NULL;
5612
5613 prison_racct_attach(pr);
5614
5615 /*
5616 * Move resource utilisation records.
5617 */
5618 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
5619
5620 #ifdef RCTL
5621 /*
5622 * Force rctl to reattach rules to processes.
5623 */
5624 FOREACH_PROC_IN_SYSTEM(p) {
5625 PROC_LOCK(p);
5626 cred = crhold(p->p_ucred);
5627 PROC_UNLOCK(p);
5628 rctl_proc_ucred_changed(p, cred);
5629 crfree(cred);
5630 }
5631 #endif
5632
5633 sx_sunlock(&allproc_lock);
5634 prison_racct_free_locked(oldprr);
5635 sx_xunlock(&allprison_lock);
5636 }
5637
5638 static void
prison_racct_detach(struct prison * pr)5639 prison_racct_detach(struct prison *pr)
5640 {
5641
5642 ASSERT_RACCT_ENABLED();
5643 sx_assert(&allprison_lock, SA_UNLOCKED);
5644
5645 if (pr->pr_prison_racct == NULL)
5646 return;
5647 prison_racct_free(pr->pr_prison_racct);
5648 pr->pr_prison_racct = NULL;
5649 }
5650 #endif /* RACCT */
5651
5652 /*
5653 * Submit a knote for a prison, locking if necessary.
5654 */
5655 static void
prison_knote(struct prison * pr,long hint)5656 prison_knote(struct prison *pr, long hint)
5657 {
5658 int locked;
5659
5660 locked = mtx_owned(&pr->pr_mtx);
5661 if (!locked)
5662 mtx_lock(&pr->pr_mtx);
5663 KNOTE_LOCKED(pr->pr_klist, hint);
5664 jaildesc_knote(pr, hint);
5665 if (!locked)
5666 mtx_unlock(&pr->pr_mtx);
5667 }
5668
5669 #ifdef DDB
5670
5671 static void
db_show_prison(struct prison * pr)5672 db_show_prison(struct prison *pr)
5673 {
5674 struct bool_flags *bf;
5675 struct jailsys_flags *jsf;
5676 #if defined(INET) || defined(INET6)
5677 int ii;
5678 struct prison_ip *pip;
5679 #endif
5680 unsigned f;
5681 #ifdef INET
5682 char ip4buf[INET_ADDRSTRLEN];
5683 #endif
5684 #ifdef INET6
5685 char ip6buf[INET6_ADDRSTRLEN];
5686 #endif
5687
5688 db_printf("prison %p:\n", pr);
5689 db_printf(" jid = %d\n", pr->pr_id);
5690 db_printf(" name = %s\n", pr->pr_name);
5691 db_printf(" parent = %p\n", pr->pr_parent);
5692 db_printf(" ref = %d\n", pr->pr_ref);
5693 db_printf(" uref = %d\n", pr->pr_uref);
5694 db_printf(" state = %s\n",
5695 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
5696 pr->pr_state == PRISON_STATE_DYING ? "dying" :
5697 "invalid");
5698 db_printf(" path = %s\n", pr->pr_path);
5699 db_printf(" cpuset = %d\n", pr->pr_cpuset
5700 ? pr->pr_cpuset->cs_id : -1);
5701 #ifdef VIMAGE
5702 db_printf(" vnet = %p\n", pr->pr_vnet);
5703 #endif
5704 db_printf(" root = %p\n", pr->pr_root);
5705 db_printf(" securelevel = %d\n", pr->pr_securelevel);
5706 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
5707 db_printf(" children.max = %d\n", pr->pr_childmax);
5708 db_printf(" children.cur = %d\n", pr->pr_childcount);
5709 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
5710 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
5711 db_printf(" flags = 0x%x", pr->pr_flags);
5712 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
5713 if (pr->pr_flags & bf->flag)
5714 db_printf(" %s", bf->name);
5715 for (jsf = pr_flag_jailsys;
5716 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
5717 jsf++) {
5718 f = pr->pr_flags & (jsf->disable | jsf->new);
5719 db_printf(" %-16s= %s\n", jsf->name,
5720 (f != 0 && f == jsf->disable) ? "disable"
5721 : (f == jsf->new) ? "new"
5722 : "inherit");
5723 }
5724 db_printf(" allow = 0x%x", pr->pr_allow);
5725 for (bf = pr_flag_allow;
5726 bf < pr_flag_allow + nitems(pr_flag_allow) &&
5727 atomic_load_int(&bf->flag) != 0;
5728 bf++)
5729 if (pr->pr_allow & bf->flag)
5730 db_printf(" %s", bf->name);
5731 db_printf("\n");
5732 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
5733 db_printf(" host.hostname = %s\n", pr->pr_hostname);
5734 db_printf(" host.domainname = %s\n", pr->pr_domainname);
5735 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
5736 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
5737 #ifdef INET
5738 if ((pip = pr->pr_addrs[PR_INET]) != NULL) {
5739 db_printf(" ip4s = %d\n", pip->ips);
5740 for (ii = 0; ii < pip->ips; ii++)
5741 db_printf(" %s %s\n",
5742 ii == 0 ? "ip4.addr =" : " ",
5743 inet_ntoa_r(
5744 *(const struct in_addr *)PR_IP(pip, PR_INET, ii),
5745 ip4buf));
5746 }
5747 #endif
5748 #ifdef INET6
5749 if ((pip = pr->pr_addrs[PR_INET6]) != NULL) {
5750 db_printf(" ip6s = %d\n", pip->ips);
5751 for (ii = 0; ii < pip->ips; ii++)
5752 db_printf(" %s %s\n",
5753 ii == 0 ? "ip6.addr =" : " ",
5754 ip6_sprintf(ip6buf,
5755 (const struct in6_addr *)PR_IP(pip, PR_INET6, ii)));
5756 }
5757 #endif
5758 }
5759
DB_SHOW_COMMAND(prison,db_show_prison_command)5760 DB_SHOW_COMMAND(prison, db_show_prison_command)
5761 {
5762 struct prison *pr;
5763
5764 if (!have_addr) {
5765 /*
5766 * Show all prisons in the list, and prison0 which is not
5767 * listed.
5768 */
5769 db_show_prison(&prison0);
5770 if (!db_pager_quit) {
5771 TAILQ_FOREACH(pr, &allprison, pr_list) {
5772 db_show_prison(pr);
5773 if (db_pager_quit)
5774 break;
5775 }
5776 }
5777 return;
5778 }
5779
5780 if (addr == 0)
5781 pr = &prison0;
5782 else {
5783 /* Look for a prison with the ID and with references. */
5784 TAILQ_FOREACH(pr, &allprison, pr_list)
5785 if (pr->pr_id == addr && pr->pr_ref > 0)
5786 break;
5787 if (pr == NULL)
5788 /* Look again, without requiring a reference. */
5789 TAILQ_FOREACH(pr, &allprison, pr_list)
5790 if (pr->pr_id == addr)
5791 break;
5792 if (pr == NULL)
5793 /* Assume address points to a valid prison. */
5794 pr = (struct prison *)addr;
5795 }
5796 db_show_prison(pr);
5797 }
5798
5799 #endif /* DDB */
5800