1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 __FBSDID("$FreeBSD$");
33
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/linker.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/racct.h>
55 #include <sys/rctl.h>
56 #include <sys/refcount.h>
57 #include <sys/sx.h>
58 #include <sys/sysent.h>
59 #include <sys/namei.h>
60 #include <sys/mount.h>
61 #include <sys/queue.h>
62 #include <sys/socket.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysctl.h>
65 #include <sys/uuid.h>
66 #include <sys/vnode.h>
67
68 #include <net/if.h>
69 #include <net/vnet.h>
70
71 #include <netinet/in.h>
72
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif /* DDB */
76
77 #include <security/mac/mac_framework.h>
78
79 #define PRISON0_HOSTUUID_MODULE "hostuuid"
80
81 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
82 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
83
84 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
85 #ifdef INET
86 #ifdef INET6
87 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
88 #else
89 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL
90 #endif
91 #else /* !INET */
92 #ifdef INET6
93 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL
94 #else
95 #define _PR_IP_SADDRSEL 0
96 #endif
97 #endif
98
99 /* prison0 describes what is "real" about the system. */
100 struct prison prison0 = {
101 .pr_id = 0,
102 .pr_name = "0",
103 .pr_ref = 1,
104 .pr_uref = 1,
105 .pr_path = "/",
106 .pr_securelevel = -1,
107 .pr_devfs_rsnum = 0,
108 .pr_state = PRISON_STATE_ALIVE,
109 .pr_childmax = JAIL_MAX,
110 .pr_hostuuid = DEFAULT_HOSTUUID,
111 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children),
112 #ifdef VIMAGE
113 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
114 #else
115 .pr_flags = PR_HOST|_PR_IP_SADDRSEL,
116 #endif
117 .pr_allow = PR_ALLOW_ALL_STATIC,
118 };
119 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
120
121 struct bool_flags {
122 const char *name;
123 const char *noname;
124 volatile u_int flag;
125 };
126 struct jailsys_flags {
127 const char *name;
128 unsigned disable;
129 unsigned new;
130 };
131
132 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
133 struct sx allprison_lock;
134 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
135 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
136 LIST_HEAD(, prison_racct) allprison_racct;
137 int lastprid = 0;
138
139 static int get_next_prid(struct prison **insprp);
140 static int do_jail_attach(struct thread *td, struct prison *pr, int drflags);
141 static void prison_complete(void *context, int pending);
142 static void prison_deref(struct prison *pr, int flags);
143 static void prison_deref_kill(struct prison *pr, struct prisonlist *freeprison);
144 static int prison_lock_xlock(struct prison *pr, int flags);
145 static void prison_free_not_last(struct prison *pr);
146 static void prison_proc_free_not_last(struct prison *pr);
147 static void prison_set_allow_locked(struct prison *pr, unsigned flag,
148 int enable);
149 static char *prison_path(struct prison *pr1, struct prison *pr2);
150 #ifdef RACCT
151 static void prison_racct_attach(struct prison *pr);
152 static void prison_racct_modify(struct prison *pr);
153 static void prison_racct_detach(struct prison *pr);
154 #endif
155
156 /* Flags for prison_deref */
157 #define PD_DEREF 0x01 /* Decrement pr_ref */
158 #define PD_DEUREF 0x02 /* Decrement pr_uref */
159 #define PD_KILL 0x04 /* Remove jail, kill processes, etc */
160 #define PD_LOCKED 0x10 /* pr_mtx is held */
161 #define PD_LIST_SLOCKED 0x20 /* allprison_lock is held shared */
162 #define PD_LIST_XLOCKED 0x40 /* allprison_lock is held exclusive */
163 #define PD_OP_FLAGS 0x07 /* Operation flags */
164 #define PD_LOCK_FLAGS 0x70 /* Lock status flags */
165
166 /*
167 * Parameter names corresponding to PR_* flag values. Size values are for kvm
168 * as we cannot figure out the size of a sparse array, or an array without a
169 * terminating entry.
170 */
171 static struct bool_flags pr_flag_bool[] = {
172 {"persist", "nopersist", PR_PERSIST},
173 #ifdef INET
174 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
175 #endif
176 #ifdef INET6
177 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
178 #endif
179 };
180 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
181
182 static struct jailsys_flags pr_flag_jailsys[] = {
183 {"host", 0, PR_HOST},
184 #ifdef VIMAGE
185 {"vnet", 0, PR_VNET},
186 #endif
187 #ifdef INET
188 {"ip4", PR_IP4_USER, PR_IP4_USER},
189 #endif
190 #ifdef INET6
191 {"ip6", PR_IP6_USER, PR_IP6_USER},
192 #endif
193 };
194 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
195
196 /*
197 * Make this array full-size so dynamic parameters can be added.
198 * It is protected by prison0.mtx, but lockless reading is allowed
199 * with an atomic check of the flag values.
200 */
201 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
202 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
203 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
204 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
205 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
206 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
207 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
208 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
209 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
210 {"allow.reserved_ports", "allow.noreserved_ports",
211 PR_ALLOW_RESERVED_PORTS},
212 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
213 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
214 PR_ALLOW_UNPRIV_DEBUG},
215 {"allow.suser", "allow.nosuser", PR_ALLOW_SUSER},
216 };
217 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC;
218 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
219
220 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \
221 PR_ALLOW_RESERVED_PORTS | \
222 PR_ALLOW_UNPRIV_DEBUG | \
223 PR_ALLOW_SUSER)
224 #define JAIL_DEFAULT_ENFORCE_STATFS 2
225 #define JAIL_DEFAULT_DEVFS_RSNUM 0
226 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
227 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
228 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
229 #if defined(INET) || defined(INET6)
230 static unsigned jail_max_af_ips = 255;
231 #endif
232
233 /*
234 * Initialize the parts of prison0 that can't be static-initialized with
235 * constants. This is called from proc0_init() after creating thread0 cpuset.
236 */
237 void
prison0_init(void)238 prison0_init(void)
239 {
240 uint8_t *file, *data;
241 size_t size;
242 char buf[sizeof(prison0.pr_hostuuid)];
243 bool valid;
244
245 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
246 prison0.pr_osreldate = osreldate;
247 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
248
249 /* If we have a preloaded hostuuid, use it. */
250 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE);
251 if (file != NULL) {
252 data = preload_fetch_addr(file);
253 size = preload_fetch_size(file);
254 if (data != NULL) {
255 /*
256 * The preloaded data may include trailing whitespace, almost
257 * certainly a newline; skip over any whitespace or
258 * non-printable characters to be safe.
259 */
260 while (size > 0 && data[size - 1] <= 0x20) {
261 size--;
262 }
263
264 valid = false;
265
266 /*
267 * Not NUL-terminated when passed from loader, but
268 * validate_uuid requires that due to using sscanf (as
269 * does the subsequent strlcpy, since it still reads
270 * past the given size to return the true length);
271 * bounce to a temporary buffer to fix.
272 */
273 if (size >= sizeof(buf))
274 goto done;
275
276 memcpy(buf, data, size);
277 buf[size] = '\0';
278
279 if (validate_uuid(buf, size, NULL, 0) != 0)
280 goto done;
281
282 valid = true;
283 (void)strlcpy(prison0.pr_hostuuid, buf,
284 sizeof(prison0.pr_hostuuid));
285
286 done:
287 if (bootverbose && !valid) {
288 printf("hostuuid: preload data malformed: '%.*s'\n",
289 (int)size, data);
290 }
291 }
292 }
293 if (bootverbose)
294 printf("hostuuid: using %s\n", prison0.pr_hostuuid);
295 }
296
297 /*
298 * struct jail_args {
299 * struct jail *jail;
300 * };
301 */
302 int
sys_jail(struct thread * td,struct jail_args * uap)303 sys_jail(struct thread *td, struct jail_args *uap)
304 {
305 uint32_t version;
306 int error;
307 struct jail j;
308
309 error = copyin(uap->jail, &version, sizeof(uint32_t));
310 if (error)
311 return (error);
312
313 switch (version) {
314 case 0:
315 {
316 struct jail_v0 j0;
317
318 /* FreeBSD single IPv4 jails. */
319 bzero(&j, sizeof(struct jail));
320 error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
321 if (error)
322 return (error);
323 j.version = j0.version;
324 j.path = j0.path;
325 j.hostname = j0.hostname;
326 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */
327 break;
328 }
329
330 case 1:
331 /*
332 * Version 1 was used by multi-IPv4 jail implementations
333 * that never made it into the official kernel.
334 */
335 return (EINVAL);
336
337 case 2: /* JAIL_API_VERSION */
338 /* FreeBSD multi-IPv4/IPv6,noIP jails. */
339 error = copyin(uap->jail, &j, sizeof(struct jail));
340 if (error)
341 return (error);
342 break;
343
344 default:
345 /* Sci-Fi jails are not supported, sorry. */
346 return (EINVAL);
347 }
348 return (kern_jail(td, &j));
349 }
350
351 int
kern_jail(struct thread * td,struct jail * j)352 kern_jail(struct thread *td, struct jail *j)
353 {
354 struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
355 #ifdef INET
356 + 1
357 #endif
358 #ifdef INET6
359 + 1
360 #endif
361 )];
362 struct uio opt;
363 char *u_path, *u_hostname, *u_name;
364 struct bool_flags *bf;
365 #ifdef INET
366 uint32_t ip4s;
367 struct in_addr *u_ip4;
368 #endif
369 #ifdef INET6
370 struct in6_addr *u_ip6;
371 #endif
372 size_t tmplen;
373 int error, enforce_statfs;
374
375 bzero(&optiov, sizeof(optiov));
376 opt.uio_iov = optiov;
377 opt.uio_iovcnt = 0;
378 opt.uio_offset = -1;
379 opt.uio_resid = -1;
380 opt.uio_segflg = UIO_SYSSPACE;
381 opt.uio_rw = UIO_READ;
382 opt.uio_td = td;
383
384 /* Set permissions for top-level jails from sysctls. */
385 if (!jailed(td->td_ucred)) {
386 for (bf = pr_flag_allow;
387 bf < pr_flag_allow + nitems(pr_flag_allow) &&
388 atomic_load_int(&bf->flag) != 0;
389 bf++) {
390 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
391 (jail_default_allow & bf->flag)
392 ? bf->name : bf->noname);
393 optiov[opt.uio_iovcnt].iov_len =
394 strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
395 opt.uio_iovcnt += 2;
396 }
397 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
398 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
399 opt.uio_iovcnt++;
400 enforce_statfs = jail_default_enforce_statfs;
401 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
402 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
403 opt.uio_iovcnt++;
404 }
405
406 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
407 #ifdef INET
408 ip4s = (j->version == 0) ? 1 : j->ip4s;
409 if (ip4s > jail_max_af_ips)
410 return (EINVAL);
411 tmplen += ip4s * sizeof(struct in_addr);
412 #else
413 if (j->ip4s > 0)
414 return (EINVAL);
415 #endif
416 #ifdef INET6
417 if (j->ip6s > jail_max_af_ips)
418 return (EINVAL);
419 tmplen += j->ip6s * sizeof(struct in6_addr);
420 #else
421 if (j->ip6s > 0)
422 return (EINVAL);
423 #endif
424 u_path = malloc(tmplen, M_TEMP, M_WAITOK);
425 u_hostname = u_path + MAXPATHLEN;
426 u_name = u_hostname + MAXHOSTNAMELEN;
427 #ifdef INET
428 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
429 #endif
430 #ifdef INET6
431 #ifdef INET
432 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
433 #else
434 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
435 #endif
436 #endif
437 optiov[opt.uio_iovcnt].iov_base = "path";
438 optiov[opt.uio_iovcnt].iov_len = sizeof("path");
439 opt.uio_iovcnt++;
440 optiov[opt.uio_iovcnt].iov_base = u_path;
441 error = copyinstr(j->path, u_path, MAXPATHLEN,
442 &optiov[opt.uio_iovcnt].iov_len);
443 if (error) {
444 free(u_path, M_TEMP);
445 return (error);
446 }
447 opt.uio_iovcnt++;
448 optiov[opt.uio_iovcnt].iov_base = "host.hostname";
449 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
450 opt.uio_iovcnt++;
451 optiov[opt.uio_iovcnt].iov_base = u_hostname;
452 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
453 &optiov[opt.uio_iovcnt].iov_len);
454 if (error) {
455 free(u_path, M_TEMP);
456 return (error);
457 }
458 opt.uio_iovcnt++;
459 if (j->jailname != NULL) {
460 optiov[opt.uio_iovcnt].iov_base = "name";
461 optiov[opt.uio_iovcnt].iov_len = sizeof("name");
462 opt.uio_iovcnt++;
463 optiov[opt.uio_iovcnt].iov_base = u_name;
464 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
465 &optiov[opt.uio_iovcnt].iov_len);
466 if (error) {
467 free(u_path, M_TEMP);
468 return (error);
469 }
470 opt.uio_iovcnt++;
471 }
472 #ifdef INET
473 optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
474 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
475 opt.uio_iovcnt++;
476 optiov[opt.uio_iovcnt].iov_base = u_ip4;
477 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
478 if (j->version == 0)
479 u_ip4->s_addr = j->ip4s;
480 else {
481 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
482 if (error) {
483 free(u_path, M_TEMP);
484 return (error);
485 }
486 }
487 opt.uio_iovcnt++;
488 #endif
489 #ifdef INET6
490 optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
491 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
492 opt.uio_iovcnt++;
493 optiov[opt.uio_iovcnt].iov_base = u_ip6;
494 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
495 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
496 if (error) {
497 free(u_path, M_TEMP);
498 return (error);
499 }
500 opt.uio_iovcnt++;
501 #endif
502 KASSERT(opt.uio_iovcnt <= nitems(optiov),
503 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
504 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
505 free(u_path, M_TEMP);
506 return (error);
507 }
508
509 /*
510 * struct jail_set_args {
511 * struct iovec *iovp;
512 * unsigned int iovcnt;
513 * int flags;
514 * };
515 */
516 int
sys_jail_set(struct thread * td,struct jail_set_args * uap)517 sys_jail_set(struct thread *td, struct jail_set_args *uap)
518 {
519 struct uio *auio;
520 int error;
521
522 /* Check that we have an even number of iovecs. */
523 if (uap->iovcnt & 1)
524 return (EINVAL);
525
526 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
527 if (error)
528 return (error);
529 error = kern_jail_set(td, auio, uap->flags);
530 free(auio, M_IOV);
531 return (error);
532 }
533
534 int
kern_jail_set(struct thread * td,struct uio * optuio,int flags)535 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
536 {
537 struct nameidata nd;
538 #ifdef INET
539 struct in_addr *ip4;
540 #endif
541 #ifdef INET6
542 struct in6_addr *ip6;
543 #endif
544 struct vfsopt *opt;
545 struct vfsoptlist *opts;
546 struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr;
547 struct vnode *root;
548 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
549 char *g_path, *osrelstr;
550 struct bool_flags *bf;
551 struct jailsys_flags *jsf;
552 #if defined(INET) || defined(INET6)
553 struct prison *tppr;
554 void *op;
555 #endif
556 unsigned long hid;
557 size_t namelen, onamelen, pnamelen;
558 int born, created, cuflags, descend, drflags, enforce;
559 int error, errmsg_len, errmsg_pos;
560 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
561 int jid, jsys, len, level;
562 int childmax, osreldt, rsnum, slevel;
563 #if defined(INET) || defined(INET6)
564 int ii, ij;
565 #endif
566 #ifdef INET
567 int ip4s, redo_ip4;
568 #endif
569 #ifdef INET6
570 int ip6s, redo_ip6;
571 #endif
572 uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
573 uint64_t pr_allow_diff;
574 unsigned tallow;
575 char numbuf[12];
576
577 error = priv_check(td, PRIV_JAIL_SET);
578 if (!error && (flags & JAIL_ATTACH))
579 error = priv_check(td, PRIV_JAIL_ATTACH);
580 if (error)
581 return (error);
582 mypr = td->td_ucred->cr_prison;
583 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
584 return (EPERM);
585 if (flags & ~JAIL_SET_MASK)
586 return (EINVAL);
587
588 /*
589 * Check all the parameters before committing to anything. Not all
590 * errors can be caught early, but we may as well try. Also, this
591 * takes care of some expensive stuff (path lookup) before getting
592 * the allprison lock.
593 *
594 * XXX Jails are not filesystems, and jail parameters are not mount
595 * options. But it makes more sense to re-use the vfsopt code
596 * than duplicate it under a different name.
597 */
598 error = vfs_buildopts(optuio, &opts);
599 if (error)
600 return (error);
601 #ifdef INET
602 ip4 = NULL;
603 #endif
604 #ifdef INET6
605 ip6 = NULL;
606 #endif
607 g_path = NULL;
608
609 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
610 if (!cuflags) {
611 error = EINVAL;
612 vfs_opterror(opts, "no valid operation (create or update)");
613 goto done_errmsg;
614 }
615
616 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
617 if (error == ENOENT)
618 jid = 0;
619 else if (error != 0)
620 goto done_free;
621
622 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
623 if (error == ENOENT)
624 gotslevel = 0;
625 else if (error != 0)
626 goto done_free;
627 else
628 gotslevel = 1;
629
630 error =
631 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
632 if (error == ENOENT)
633 gotchildmax = 0;
634 else if (error != 0)
635 goto done_free;
636 else
637 gotchildmax = 1;
638
639 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
640 if (error == ENOENT)
641 gotenforce = 0;
642 else if (error != 0)
643 goto done_free;
644 else if (enforce < 0 || enforce > 2) {
645 error = EINVAL;
646 goto done_free;
647 } else
648 gotenforce = 1;
649
650 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
651 if (error == ENOENT)
652 gotrsnum = 0;
653 else if (error != 0)
654 goto done_free;
655 else
656 gotrsnum = 1;
657
658 pr_flags = ch_flags = 0;
659 for (bf = pr_flag_bool;
660 bf < pr_flag_bool + nitems(pr_flag_bool);
661 bf++) {
662 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
663 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
664 }
665 ch_flags |= pr_flags;
666 for (jsf = pr_flag_jailsys;
667 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
668 jsf++) {
669 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
670 if (error == ENOENT)
671 continue;
672 if (error != 0)
673 goto done_free;
674 switch (jsys) {
675 case JAIL_SYS_DISABLE:
676 if (!jsf->disable) {
677 error = EINVAL;
678 goto done_free;
679 }
680 pr_flags |= jsf->disable;
681 break;
682 case JAIL_SYS_NEW:
683 pr_flags |= jsf->new;
684 break;
685 case JAIL_SYS_INHERIT:
686 break;
687 default:
688 error = EINVAL;
689 goto done_free;
690 }
691 ch_flags |= jsf->new | jsf->disable;
692 }
693 if ((flags & (JAIL_CREATE | JAIL_ATTACH)) == JAIL_CREATE
694 && !(pr_flags & PR_PERSIST)) {
695 error = EINVAL;
696 vfs_opterror(opts, "new jail must persist or attach");
697 goto done_errmsg;
698 }
699 #ifdef VIMAGE
700 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
701 error = EINVAL;
702 vfs_opterror(opts, "vnet cannot be changed after creation");
703 goto done_errmsg;
704 }
705 #endif
706 #ifdef INET
707 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
708 error = EINVAL;
709 vfs_opterror(opts, "ip4 cannot be changed after creation");
710 goto done_errmsg;
711 }
712 #endif
713 #ifdef INET6
714 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
715 error = EINVAL;
716 vfs_opterror(opts, "ip6 cannot be changed after creation");
717 goto done_errmsg;
718 }
719 #endif
720
721 pr_allow = ch_allow = 0;
722 for (bf = pr_flag_allow;
723 bf < pr_flag_allow + nitems(pr_flag_allow) &&
724 atomic_load_int(&bf->flag) != 0;
725 bf++) {
726 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
727 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
728 }
729 ch_allow |= pr_allow;
730
731 error = vfs_getopt(opts, "name", (void **)&name, &len);
732 if (error == ENOENT)
733 name = NULL;
734 else if (error != 0)
735 goto done_free;
736 else {
737 if (len == 0 || name[len - 1] != '\0') {
738 error = EINVAL;
739 goto done_free;
740 }
741 if (len > MAXHOSTNAMELEN) {
742 error = ENAMETOOLONG;
743 goto done_free;
744 }
745 }
746
747 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
748 if (error == ENOENT)
749 host = NULL;
750 else if (error != 0)
751 goto done_free;
752 else {
753 ch_flags |= PR_HOST;
754 pr_flags |= PR_HOST;
755 if (len == 0 || host[len - 1] != '\0') {
756 error = EINVAL;
757 goto done_free;
758 }
759 if (len > MAXHOSTNAMELEN) {
760 error = ENAMETOOLONG;
761 goto done_free;
762 }
763 }
764
765 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
766 if (error == ENOENT)
767 domain = NULL;
768 else if (error != 0)
769 goto done_free;
770 else {
771 ch_flags |= PR_HOST;
772 pr_flags |= PR_HOST;
773 if (len == 0 || domain[len - 1] != '\0') {
774 error = EINVAL;
775 goto done_free;
776 }
777 if (len > MAXHOSTNAMELEN) {
778 error = ENAMETOOLONG;
779 goto done_free;
780 }
781 }
782
783 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
784 if (error == ENOENT)
785 uuid = NULL;
786 else if (error != 0)
787 goto done_free;
788 else {
789 ch_flags |= PR_HOST;
790 pr_flags |= PR_HOST;
791 if (len == 0 || uuid[len - 1] != '\0') {
792 error = EINVAL;
793 goto done_free;
794 }
795 if (len > HOSTUUIDLEN) {
796 error = ENAMETOOLONG;
797 goto done_free;
798 }
799 }
800
801 #ifdef COMPAT_FREEBSD32
802 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
803 uint32_t hid32;
804
805 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
806 hid = hid32;
807 } else
808 #endif
809 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
810 if (error == ENOENT)
811 gothid = 0;
812 else if (error != 0)
813 goto done_free;
814 else {
815 gothid = 1;
816 ch_flags |= PR_HOST;
817 pr_flags |= PR_HOST;
818 }
819
820 #ifdef INET
821 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
822 if (error == ENOENT)
823 ip4s = 0;
824 else if (error != 0)
825 goto done_free;
826 else if (ip4s & (sizeof(*ip4) - 1)) {
827 error = EINVAL;
828 goto done_free;
829 } else {
830 ch_flags |= PR_IP4_USER;
831 pr_flags |= PR_IP4_USER;
832 if (ip4s > 0) {
833 ip4s /= sizeof(*ip4);
834 if (ip4s > jail_max_af_ips) {
835 error = EINVAL;
836 vfs_opterror(opts, "too many IPv4 addresses");
837 goto done_errmsg;
838 }
839 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
840 bcopy(op, ip4, ip4s * sizeof(*ip4));
841 /*
842 * IP addresses are all sorted but ip[0] to preserve
843 * the primary IP address as given from userland.
844 * This special IP is used for unbound outgoing
845 * connections as well for "loopback" traffic in case
846 * source address selection cannot find any more fitting
847 * address to connect from.
848 */
849 if (ip4s > 1)
850 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
851 prison_qcmp_v4);
852 /*
853 * Check for duplicate addresses and do some simple
854 * zero and broadcast checks. If users give other bogus
855 * addresses it is their problem.
856 *
857 * We do not have to care about byte order for these
858 * checks so we will do them in NBO.
859 */
860 for (ii = 0; ii < ip4s; ii++) {
861 if (ip4[ii].s_addr == INADDR_ANY ||
862 ip4[ii].s_addr == INADDR_BROADCAST) {
863 error = EINVAL;
864 goto done_free;
865 }
866 if ((ii+1) < ip4s &&
867 (ip4[0].s_addr == ip4[ii+1].s_addr ||
868 ip4[ii].s_addr == ip4[ii+1].s_addr)) {
869 error = EINVAL;
870 goto done_free;
871 }
872 }
873 }
874 }
875 #endif
876
877 #ifdef INET6
878 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
879 if (error == ENOENT)
880 ip6s = 0;
881 else if (error != 0)
882 goto done_free;
883 else if (ip6s & (sizeof(*ip6) - 1)) {
884 error = EINVAL;
885 goto done_free;
886 } else {
887 ch_flags |= PR_IP6_USER;
888 pr_flags |= PR_IP6_USER;
889 if (ip6s > 0) {
890 ip6s /= sizeof(*ip6);
891 if (ip6s > jail_max_af_ips) {
892 error = EINVAL;
893 vfs_opterror(opts, "too many IPv6 addresses");
894 goto done_errmsg;
895 }
896 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
897 bcopy(op, ip6, ip6s * sizeof(*ip6));
898 if (ip6s > 1)
899 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
900 prison_qcmp_v6);
901 for (ii = 0; ii < ip6s; ii++) {
902 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
903 error = EINVAL;
904 goto done_free;
905 }
906 if ((ii+1) < ip6s &&
907 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
908 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
909 {
910 error = EINVAL;
911 goto done_free;
912 }
913 }
914 }
915 }
916 #endif
917
918 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
919 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
920 error = EINVAL;
921 vfs_opterror(opts,
922 "vnet jails cannot have IP address restrictions");
923 goto done_errmsg;
924 }
925 #endif
926
927 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
928 if (error == ENOENT)
929 osrelstr = NULL;
930 else if (error != 0)
931 goto done_free;
932 else {
933 if (flags & JAIL_UPDATE) {
934 error = EINVAL;
935 vfs_opterror(opts,
936 "osrelease cannot be changed after creation");
937 goto done_errmsg;
938 }
939 if (len == 0 || osrelstr[len - 1] != '\0') {
940 error = EINVAL;
941 goto done_free;
942 }
943 if (len >= OSRELEASELEN) {
944 error = ENAMETOOLONG;
945 vfs_opterror(opts,
946 "osrelease string must be 1-%d bytes long",
947 OSRELEASELEN - 1);
948 goto done_errmsg;
949 }
950 }
951
952 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
953 if (error == ENOENT)
954 osreldt = 0;
955 else if (error != 0)
956 goto done_free;
957 else {
958 if (flags & JAIL_UPDATE) {
959 error = EINVAL;
960 vfs_opterror(opts,
961 "osreldate cannot be changed after creation");
962 goto done_errmsg;
963 }
964 if (osreldt == 0) {
965 error = EINVAL;
966 vfs_opterror(opts, "osreldate cannot be 0");
967 goto done_errmsg;
968 }
969 }
970
971 root = NULL;
972 error = vfs_getopt(opts, "path", (void **)&path, &len);
973 if (error == ENOENT)
974 path = NULL;
975 else if (error != 0)
976 goto done_free;
977 else {
978 if (flags & JAIL_UPDATE) {
979 error = EINVAL;
980 vfs_opterror(opts,
981 "path cannot be changed after creation");
982 goto done_errmsg;
983 }
984 if (len == 0 || path[len - 1] != '\0') {
985 error = EINVAL;
986 goto done_free;
987 }
988 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
989 path, td);
990 error = namei(&nd);
991 if (error)
992 goto done_free;
993 root = nd.ni_vp;
994 NDFREE(&nd, NDF_ONLY_PNBUF);
995 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
996 strlcpy(g_path, path, MAXPATHLEN);
997 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
998 if (error == 0) {
999 path = g_path;
1000 } else {
1001 /* exit on other errors */
1002 goto done_free;
1003 }
1004 if (root->v_type != VDIR) {
1005 error = ENOTDIR;
1006 vput(root);
1007 goto done_free;
1008 }
1009 VOP_UNLOCK(root);
1010 }
1011
1012 /*
1013 * Find the specified jail, or at least its parent.
1014 * This abuses the file error codes ENOENT and EEXIST.
1015 */
1016 pr = NULL;
1017 inspr = NULL;
1018 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1019 namelc = strrchr(name, '.');
1020 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1021 if (*p != '\0')
1022 jid = 0;
1023 }
1024 sx_xlock(&allprison_lock);
1025 drflags = PD_LIST_XLOCKED;
1026 ppr = mypr;
1027 if (!prison_isalive(ppr)) {
1028 /* This jail is dying. This process will surely follow. */
1029 error = EAGAIN;
1030 goto done_deref;
1031 }
1032 if (jid != 0) {
1033 if (jid < 0) {
1034 error = EINVAL;
1035 vfs_opterror(opts, "negative jid");
1036 goto done_deref;
1037 }
1038 /*
1039 * See if a requested jid already exists. Keep track of
1040 * where it can be inserted later.
1041 */
1042 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1043 if (inspr->pr_id < jid)
1044 continue;
1045 if (inspr->pr_id > jid)
1046 break;
1047 pr = inspr;
1048 mtx_lock(&pr->pr_mtx);
1049 drflags |= PD_LOCKED;
1050 inspr = NULL;
1051 break;
1052 }
1053 if (pr != NULL) {
1054 /* Create: jid must not exist. */
1055 if (cuflags == JAIL_CREATE) {
1056 /*
1057 * Even creators that cannot see the jail will
1058 * get EEXIST.
1059 */
1060 error = EEXIST;
1061 vfs_opterror(opts, "jail %d already exists",
1062 jid);
1063 goto done_deref;
1064 }
1065 if (!prison_ischild(mypr, pr)) {
1066 /*
1067 * Updaters get ENOENT if they cannot see the
1068 * jail. This is true even for CREATE | UPDATE,
1069 * which normally cannot give this error.
1070 */
1071 error = ENOENT;
1072 vfs_opterror(opts, "jail %d not found", jid);
1073 goto done_deref;
1074 }
1075 ppr = pr->pr_parent;
1076 if (!prison_isalive(ppr)) {
1077 error = ENOENT;
1078 vfs_opterror(opts, "jail %d is dying",
1079 ppr->pr_id);
1080 goto done_deref;
1081 }
1082 if (!prison_isalive(pr)) {
1083 if (!(flags & JAIL_DYING)) {
1084 error = ENOENT;
1085 vfs_opterror(opts, "jail %d is dying",
1086 jid);
1087 goto done_deref;
1088 }
1089 if ((flags & JAIL_ATTACH) ||
1090 (pr_flags & PR_PERSIST)) {
1091 /*
1092 * A dying jail might be resurrected
1093 * (via attach or persist), but first
1094 * it must determine if another jail
1095 * has claimed its name. Accomplish
1096 * this by implicitly re-setting the
1097 * name.
1098 */
1099 if (name == NULL)
1100 name = prison_name(mypr, pr);
1101 }
1102 }
1103 } else {
1104 /* Update: jid must exist. */
1105 if (cuflags == JAIL_UPDATE) {
1106 error = ENOENT;
1107 vfs_opterror(opts, "jail %d not found", jid);
1108 goto done_deref;
1109 }
1110 }
1111 }
1112 /*
1113 * If the caller provided a name, look for a jail by that name.
1114 * This has different semantics for creates and updates keyed by jid
1115 * (where the name must not already exist in a different jail),
1116 * and updates keyed by the name itself (where the name must exist
1117 * because that is the jail being updated).
1118 */
1119 namelc = NULL;
1120 if (name != NULL) {
1121 namelc = strrchr(name, '.');
1122 if (namelc == NULL)
1123 namelc = name;
1124 else {
1125 /*
1126 * This is a hierarchical name. Split it into the
1127 * parent and child names, and make sure the parent
1128 * exists or matches an already found jail.
1129 */
1130 if (pr != NULL) {
1131 if (strncmp(name, ppr->pr_name, namelc - name)
1132 || ppr->pr_name[namelc - name] != '\0') {
1133 error = EINVAL;
1134 vfs_opterror(opts,
1135 "cannot change jail's parent");
1136 goto done_deref;
1137 }
1138 } else {
1139 *namelc = '\0';
1140 ppr = prison_find_name(mypr, name);
1141 if (ppr == NULL) {
1142 error = ENOENT;
1143 vfs_opterror(opts,
1144 "jail \"%s\" not found", name);
1145 goto done_deref;
1146 }
1147 mtx_unlock(&ppr->pr_mtx);
1148 if (!prison_isalive(ppr)) {
1149 error = ENOENT;
1150 vfs_opterror(opts,
1151 "jail \"%s\" is dying", name);
1152 goto done_deref;
1153 }
1154 *namelc = '.';
1155 }
1156 namelc++;
1157 }
1158 if (namelc[0] != '\0') {
1159 pnamelen =
1160 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1161 deadpr = NULL;
1162 FOREACH_PRISON_CHILD(ppr, tpr) {
1163 if (tpr != pr &&
1164 !strcmp(tpr->pr_name + pnamelen, namelc)) {
1165 if (prison_isalive(tpr)) {
1166 if (pr == NULL &&
1167 cuflags != JAIL_CREATE) {
1168 /*
1169 * Use this jail
1170 * for updates.
1171 */
1172 pr = tpr;
1173 mtx_lock(&pr->pr_mtx);
1174 drflags |= PD_LOCKED;
1175 break;
1176 }
1177 /*
1178 * Create, or update(jid):
1179 * name must not exist in an
1180 * active sibling jail.
1181 */
1182 error = EEXIST;
1183 vfs_opterror(opts,
1184 "jail \"%s\" already exists",
1185 name);
1186 goto done_deref;
1187 }
1188 if (pr == NULL &&
1189 cuflags != JAIL_CREATE) {
1190 deadpr = tpr;
1191 }
1192 }
1193 }
1194 /* If no active jail is found, use a dying one. */
1195 if (deadpr != NULL && pr == NULL) {
1196 if (flags & JAIL_DYING) {
1197 pr = deadpr;
1198 mtx_lock(&pr->pr_mtx);
1199 drflags |= PD_LOCKED;
1200 } else if (cuflags == JAIL_UPDATE) {
1201 error = ENOENT;
1202 vfs_opterror(opts,
1203 "jail \"%s\" is dying", name);
1204 goto done_deref;
1205 }
1206 }
1207 /* Update: name must exist if no jid. */
1208 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1209 error = ENOENT;
1210 vfs_opterror(opts, "jail \"%s\" not found",
1211 name);
1212 goto done_deref;
1213 }
1214 }
1215 }
1216 /* Update: must provide a jid or name. */
1217 else if (cuflags == JAIL_UPDATE && pr == NULL) {
1218 error = ENOENT;
1219 vfs_opterror(opts, "update specified no jail");
1220 goto done_deref;
1221 }
1222
1223 /* If there's no prison to update, create a new one and link it in. */
1224 created = pr == NULL;
1225 if (created) {
1226 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1227 if (tpr->pr_childcount >= tpr->pr_childmax) {
1228 error = EPERM;
1229 vfs_opterror(opts, "prison limit exceeded");
1230 goto done_deref;
1231 }
1232 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) {
1233 error = EAGAIN;
1234 vfs_opterror(opts, "no available jail IDs");
1235 goto done_deref;
1236 }
1237
1238 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1239 pr->pr_state = PRISON_STATE_INVALID;
1240 refcount_init(&pr->pr_ref, 1);
1241 refcount_init(&pr->pr_uref, 0);
1242 drflags |= PD_DEREF;
1243 LIST_INIT(&pr->pr_children);
1244 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1245 TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1246
1247 pr->pr_id = jid;
1248 if (inspr != NULL)
1249 TAILQ_INSERT_BEFORE(inspr, pr, pr_list);
1250 else
1251 TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1252
1253 pr->pr_parent = ppr;
1254 prison_hold(ppr);
1255 prison_proc_hold(ppr);
1256 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1257 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1258 tpr->pr_childcount++;
1259
1260 /* Set some default values, and inherit some from the parent. */
1261 if (namelc == NULL)
1262 namelc = "";
1263 if (path == NULL) {
1264 path = "/";
1265 root = mypr->pr_root;
1266 vref(root);
1267 }
1268 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1269 pr->pr_flags |= PR_HOST;
1270 #if defined(INET) || defined(INET6)
1271 #ifdef VIMAGE
1272 if (!(pr_flags & PR_VNET))
1273 #endif
1274 {
1275 #ifdef INET
1276 if (!(ch_flags & PR_IP4_USER))
1277 pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1278 else if (!(pr_flags & PR_IP4_USER)) {
1279 pr->pr_flags |= ppr->pr_flags & PR_IP4;
1280 if (ppr->pr_ip4 != NULL) {
1281 pr->pr_ip4s = ppr->pr_ip4s;
1282 pr->pr_ip4 = malloc(pr->pr_ip4s *
1283 sizeof(struct in_addr), M_PRISON,
1284 M_WAITOK);
1285 bcopy(ppr->pr_ip4, pr->pr_ip4,
1286 pr->pr_ip4s * sizeof(*pr->pr_ip4));
1287 }
1288 }
1289 #endif
1290 #ifdef INET6
1291 if (!(ch_flags & PR_IP6_USER))
1292 pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1293 else if (!(pr_flags & PR_IP6_USER)) {
1294 pr->pr_flags |= ppr->pr_flags & PR_IP6;
1295 if (ppr->pr_ip6 != NULL) {
1296 pr->pr_ip6s = ppr->pr_ip6s;
1297 pr->pr_ip6 = malloc(pr->pr_ip6s *
1298 sizeof(struct in6_addr), M_PRISON,
1299 M_WAITOK);
1300 bcopy(ppr->pr_ip6, pr->pr_ip6,
1301 pr->pr_ip6s * sizeof(*pr->pr_ip6));
1302 }
1303 }
1304 #endif
1305 }
1306 #endif
1307 /* Source address selection is always on by default. */
1308 pr->pr_flags |= _PR_IP_SADDRSEL;
1309
1310 pr->pr_securelevel = ppr->pr_securelevel;
1311 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1312 pr->pr_enforce_statfs = jail_default_enforce_statfs;
1313 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1314
1315 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1316 if (osrelstr == NULL)
1317 strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1318 sizeof(pr->pr_osrelease));
1319 else
1320 strlcpy(pr->pr_osrelease, osrelstr,
1321 sizeof(pr->pr_osrelease));
1322
1323 #ifdef VIMAGE
1324 /* Allocate a new vnet if specified. */
1325 pr->pr_vnet = (pr_flags & PR_VNET)
1326 ? vnet_alloc() : ppr->pr_vnet;
1327 #endif
1328 /*
1329 * Allocate a dedicated cpuset for each jail.
1330 * Unlike other initial settings, this may return an error.
1331 */
1332 error = cpuset_create_root(ppr, &pr->pr_cpuset);
1333 if (error)
1334 goto done_deref;
1335
1336 mtx_lock(&pr->pr_mtx);
1337 drflags |= PD_LOCKED;
1338 } else {
1339 /*
1340 * Grab a reference for existing prisons, to ensure they
1341 * continue to exist for the duration of the call.
1342 */
1343 prison_hold(pr);
1344 drflags |= PD_DEREF;
1345 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1346 if ((pr->pr_flags & PR_VNET) &&
1347 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1348 error = EINVAL;
1349 vfs_opterror(opts,
1350 "vnet jails cannot have IP address restrictions");
1351 goto done_deref;
1352 }
1353 #endif
1354 #ifdef INET
1355 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1356 error = EINVAL;
1357 vfs_opterror(opts,
1358 "ip4 cannot be changed after creation");
1359 goto done_deref;
1360 }
1361 #endif
1362 #ifdef INET6
1363 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1364 error = EINVAL;
1365 vfs_opterror(opts,
1366 "ip6 cannot be changed after creation");
1367 goto done_deref;
1368 }
1369 #endif
1370 }
1371
1372 /* Do final error checking before setting anything. */
1373 if (gotslevel) {
1374 if (slevel < ppr->pr_securelevel) {
1375 error = EPERM;
1376 goto done_deref;
1377 }
1378 }
1379 if (gotchildmax) {
1380 if (childmax >= ppr->pr_childmax) {
1381 error = EPERM;
1382 goto done_deref;
1383 }
1384 }
1385 if (gotenforce) {
1386 if (enforce < ppr->pr_enforce_statfs) {
1387 error = EPERM;
1388 goto done_deref;
1389 }
1390 }
1391 if (gotrsnum) {
1392 /*
1393 * devfs_rsnum is a uint16_t
1394 */
1395 if (rsnum < 0 || rsnum > 65535) {
1396 error = EINVAL;
1397 goto done_deref;
1398 }
1399 /*
1400 * Nested jails always inherit parent's devfs ruleset
1401 */
1402 if (jailed(td->td_ucred)) {
1403 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1404 error = EPERM;
1405 goto done_deref;
1406 } else
1407 rsnum = ppr->pr_devfs_rsnum;
1408 }
1409 }
1410 #ifdef INET
1411 if (ip4s > 0) {
1412 if (ppr->pr_flags & PR_IP4) {
1413 /*
1414 * Make sure the new set of IP addresses is a
1415 * subset of the parent's list. Don't worry
1416 * about the parent being unlocked, as any
1417 * setting is done with allprison_lock held.
1418 */
1419 for (ij = 0; ij < ppr->pr_ip4s; ij++)
1420 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1421 break;
1422 if (ij == ppr->pr_ip4s) {
1423 error = EPERM;
1424 goto done_deref;
1425 }
1426 if (ip4s > 1) {
1427 for (ii = ij = 1; ii < ip4s; ii++) {
1428 if (ip4[ii].s_addr ==
1429 ppr->pr_ip4[0].s_addr)
1430 continue;
1431 for (; ij < ppr->pr_ip4s; ij++)
1432 if (ip4[ii].s_addr ==
1433 ppr->pr_ip4[ij].s_addr)
1434 break;
1435 if (ij == ppr->pr_ip4s)
1436 break;
1437 }
1438 if (ij == ppr->pr_ip4s) {
1439 error = EPERM;
1440 goto done_deref;
1441 }
1442 }
1443 }
1444 /*
1445 * Check for conflicting IP addresses. We permit them
1446 * if there is no more than one IP on each jail. If
1447 * there is a duplicate on a jail with more than one
1448 * IP stop checking and return error.
1449 */
1450 #ifdef VIMAGE
1451 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1452 if (tppr->pr_flags & PR_VNET)
1453 break;
1454 #else
1455 tppr = &prison0;
1456 #endif
1457 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1458 if (tpr == pr ||
1459 #ifdef VIMAGE
1460 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1461 #endif
1462 !prison_isalive(tpr)) {
1463 descend = 0;
1464 continue;
1465 }
1466 if (!(tpr->pr_flags & PR_IP4_USER))
1467 continue;
1468 descend = 0;
1469 if (tpr->pr_ip4 == NULL ||
1470 (ip4s == 1 && tpr->pr_ip4s == 1))
1471 continue;
1472 for (ii = 0; ii < ip4s; ii++) {
1473 if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1474 0) {
1475 error = EADDRINUSE;
1476 vfs_opterror(opts,
1477 "IPv4 addresses clash");
1478 goto done_deref;
1479 }
1480 }
1481 }
1482 }
1483 #endif
1484 #ifdef INET6
1485 if (ip6s > 0) {
1486 if (ppr->pr_flags & PR_IP6) {
1487 /*
1488 * Make sure the new set of IP addresses is a
1489 * subset of the parent's list.
1490 */
1491 for (ij = 0; ij < ppr->pr_ip6s; ij++)
1492 if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1493 &ppr->pr_ip6[ij]))
1494 break;
1495 if (ij == ppr->pr_ip6s) {
1496 error = EPERM;
1497 goto done_deref;
1498 }
1499 if (ip6s > 1) {
1500 for (ii = ij = 1; ii < ip6s; ii++) {
1501 if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1502 &ppr->pr_ip6[0]))
1503 continue;
1504 for (; ij < ppr->pr_ip6s; ij++)
1505 if (IN6_ARE_ADDR_EQUAL(
1506 &ip6[ii], &ppr->pr_ip6[ij]))
1507 break;
1508 if (ij == ppr->pr_ip6s)
1509 break;
1510 }
1511 if (ij == ppr->pr_ip6s) {
1512 error = EPERM;
1513 goto done_deref;
1514 }
1515 }
1516 }
1517 /* Check for conflicting IP addresses. */
1518 #ifdef VIMAGE
1519 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1520 if (tppr->pr_flags & PR_VNET)
1521 break;
1522 #else
1523 tppr = &prison0;
1524 #endif
1525 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1526 if (tpr == pr ||
1527 #ifdef VIMAGE
1528 (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1529 #endif
1530 !prison_isalive(tpr)) {
1531 descend = 0;
1532 continue;
1533 }
1534 if (!(tpr->pr_flags & PR_IP6_USER))
1535 continue;
1536 descend = 0;
1537 if (tpr->pr_ip6 == NULL ||
1538 (ip6s == 1 && tpr->pr_ip6s == 1))
1539 continue;
1540 for (ii = 0; ii < ip6s; ii++) {
1541 if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1542 0) {
1543 error = EADDRINUSE;
1544 vfs_opterror(opts,
1545 "IPv6 addresses clash");
1546 goto done_deref;
1547 }
1548 }
1549 }
1550 }
1551 #endif
1552 onamelen = namelen = 0;
1553 if (namelc != NULL) {
1554 /* Give a default name of the jid. Also allow the name to be
1555 * explicitly the jid - but not any other number, and only in
1556 * normal form (no leading zero/etc).
1557 */
1558 if (namelc[0] == '\0')
1559 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1560 else if ((strtoul(namelc, &p, 10) != jid ||
1561 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1562 error = EINVAL;
1563 vfs_opterror(opts,
1564 "name cannot be numeric (unless it is the jid)");
1565 goto done_deref;
1566 }
1567 /*
1568 * Make sure the name isn't too long for the prison or its
1569 * children.
1570 */
1571 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1572 onamelen = strlen(pr->pr_name + pnamelen);
1573 namelen = strlen(namelc);
1574 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1575 error = ENAMETOOLONG;
1576 goto done_deref;
1577 }
1578 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1579 if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1580 sizeof(pr->pr_name)) {
1581 error = ENAMETOOLONG;
1582 goto done_deref;
1583 }
1584 }
1585 }
1586 pr_allow_diff = pr_allow & ~ppr->pr_allow;
1587 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1588 error = EPERM;
1589 goto done_deref;
1590 }
1591
1592 /*
1593 * Let modules check their parameters. This requires unlocking and
1594 * then re-locking the prison, but this is still a valid state as long
1595 * as allprison_lock remains xlocked.
1596 */
1597 mtx_unlock(&pr->pr_mtx);
1598 drflags &= ~PD_LOCKED;
1599 error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1600 if (error != 0)
1601 goto done_deref;
1602 mtx_lock(&pr->pr_mtx);
1603 drflags |= PD_LOCKED;
1604
1605 /* At this point, all valid parameters should have been noted. */
1606 TAILQ_FOREACH(opt, opts, link) {
1607 if (!opt->seen && strcmp(opt->name, "errmsg")) {
1608 error = EINVAL;
1609 vfs_opterror(opts, "unknown parameter: %s", opt->name);
1610 goto done_deref;
1611 }
1612 }
1613
1614 /* Set the parameters of the prison. */
1615 #ifdef INET
1616 redo_ip4 = 0;
1617 if (pr_flags & PR_IP4_USER) {
1618 pr->pr_flags |= PR_IP4;
1619 free(pr->pr_ip4, M_PRISON);
1620 pr->pr_ip4s = ip4s;
1621 pr->pr_ip4 = ip4;
1622 ip4 = NULL;
1623 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1624 #ifdef VIMAGE
1625 if (tpr->pr_flags & PR_VNET) {
1626 descend = 0;
1627 continue;
1628 }
1629 #endif
1630 if (prison_restrict_ip4(tpr, NULL)) {
1631 redo_ip4 = 1;
1632 descend = 0;
1633 }
1634 }
1635 }
1636 #endif
1637 #ifdef INET6
1638 redo_ip6 = 0;
1639 if (pr_flags & PR_IP6_USER) {
1640 pr->pr_flags |= PR_IP6;
1641 free(pr->pr_ip6, M_PRISON);
1642 pr->pr_ip6s = ip6s;
1643 pr->pr_ip6 = ip6;
1644 ip6 = NULL;
1645 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1646 #ifdef VIMAGE
1647 if (tpr->pr_flags & PR_VNET) {
1648 descend = 0;
1649 continue;
1650 }
1651 #endif
1652 if (prison_restrict_ip6(tpr, NULL)) {
1653 redo_ip6 = 1;
1654 descend = 0;
1655 }
1656 }
1657 }
1658 #endif
1659 if (gotslevel) {
1660 pr->pr_securelevel = slevel;
1661 /* Set all child jails to be at least this level. */
1662 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1663 if (tpr->pr_securelevel < slevel)
1664 tpr->pr_securelevel = slevel;
1665 }
1666 if (gotchildmax) {
1667 pr->pr_childmax = childmax;
1668 /* Set all child jails to under this limit. */
1669 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1670 if (tpr->pr_childmax > childmax - level)
1671 tpr->pr_childmax = childmax > level
1672 ? childmax - level : 0;
1673 }
1674 if (gotenforce) {
1675 pr->pr_enforce_statfs = enforce;
1676 /* Pass this restriction on to the children. */
1677 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1678 if (tpr->pr_enforce_statfs < enforce)
1679 tpr->pr_enforce_statfs = enforce;
1680 }
1681 if (gotrsnum) {
1682 pr->pr_devfs_rsnum = rsnum;
1683 /* Pass this restriction on to the children. */
1684 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1685 tpr->pr_devfs_rsnum = rsnum;
1686 }
1687 if (namelc != NULL) {
1688 if (ppr == &prison0)
1689 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1690 else
1691 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1692 ppr->pr_name, namelc);
1693 /* Change this component of child names. */
1694 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1695 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1696 strlen(tpr->pr_name + onamelen) + 1);
1697 bcopy(pr->pr_name, tpr->pr_name, namelen);
1698 }
1699 }
1700 if (path != NULL) {
1701 /* Try to keep a real-rooted full pathname. */
1702 strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1703 pr->pr_root = root;
1704 root = NULL;
1705 }
1706 if (PR_HOST & ch_flags & ~pr_flags) {
1707 if (pr->pr_flags & PR_HOST) {
1708 /*
1709 * Copy the parent's host info. As with pr_ip4 above,
1710 * the lack of a lock on the parent is not a problem;
1711 * it is always set with allprison_lock at least
1712 * shared, and is held exclusively here.
1713 */
1714 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1715 sizeof(pr->pr_hostname));
1716 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1717 sizeof(pr->pr_domainname));
1718 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1719 sizeof(pr->pr_hostuuid));
1720 pr->pr_hostid = pr->pr_parent->pr_hostid;
1721 }
1722 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1723 /* Set this prison, and any descendants without PR_HOST. */
1724 if (host != NULL)
1725 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1726 if (domain != NULL)
1727 strlcpy(pr->pr_domainname, domain,
1728 sizeof(pr->pr_domainname));
1729 if (uuid != NULL)
1730 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1731 if (gothid)
1732 pr->pr_hostid = hid;
1733 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1734 if (tpr->pr_flags & PR_HOST)
1735 descend = 0;
1736 else {
1737 if (host != NULL)
1738 strlcpy(tpr->pr_hostname,
1739 pr->pr_hostname,
1740 sizeof(tpr->pr_hostname));
1741 if (domain != NULL)
1742 strlcpy(tpr->pr_domainname,
1743 pr->pr_domainname,
1744 sizeof(tpr->pr_domainname));
1745 if (uuid != NULL)
1746 strlcpy(tpr->pr_hostuuid,
1747 pr->pr_hostuuid,
1748 sizeof(tpr->pr_hostuuid));
1749 if (gothid)
1750 tpr->pr_hostid = hid;
1751 }
1752 }
1753 }
1754 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1755 if ((tallow = ch_allow & ~pr_allow))
1756 prison_set_allow_locked(pr, tallow, 0);
1757 /*
1758 * Persistent prisons get an extra reference, and prisons losing their
1759 * persist flag lose that reference.
1760 */
1761 born = !prison_isalive(pr);
1762 if (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags)) {
1763 if (pr_flags & PR_PERSIST) {
1764 prison_hold(pr);
1765 /*
1766 * This may make a dead prison alive again, but wait
1767 * to label it as such until after OSD calls have had
1768 * a chance to run (and perhaps to fail).
1769 */
1770 refcount_acquire(&pr->pr_uref);
1771 } else {
1772 drflags |= PD_DEUREF;
1773 prison_free_not_last(pr);
1774 }
1775 }
1776 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1777 mtx_unlock(&pr->pr_mtx);
1778 drflags &= ~PD_LOCKED;
1779 /*
1780 * Any errors past this point will need to de-persist newly created
1781 * prisons, as well as call remove methods.
1782 */
1783 if (born)
1784 drflags |= PD_KILL;
1785
1786 #ifdef RACCT
1787 if (racct_enable && created)
1788 prison_racct_attach(pr);
1789 #endif
1790
1791 /* Locks may have prevented a complete restriction of child IP
1792 * addresses. If so, allocate some more memory and try again.
1793 */
1794 #ifdef INET
1795 while (redo_ip4) {
1796 ip4s = pr->pr_ip4s;
1797 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1798 mtx_lock(&pr->pr_mtx);
1799 redo_ip4 = 0;
1800 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1801 #ifdef VIMAGE
1802 if (tpr->pr_flags & PR_VNET) {
1803 descend = 0;
1804 continue;
1805 }
1806 #endif
1807 if (prison_restrict_ip4(tpr, ip4)) {
1808 if (ip4 != NULL)
1809 ip4 = NULL;
1810 else
1811 redo_ip4 = 1;
1812 }
1813 }
1814 mtx_unlock(&pr->pr_mtx);
1815 }
1816 #endif
1817 #ifdef INET6
1818 while (redo_ip6) {
1819 ip6s = pr->pr_ip6s;
1820 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1821 mtx_lock(&pr->pr_mtx);
1822 redo_ip6 = 0;
1823 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1824 #ifdef VIMAGE
1825 if (tpr->pr_flags & PR_VNET) {
1826 descend = 0;
1827 continue;
1828 }
1829 #endif
1830 if (prison_restrict_ip6(tpr, ip6)) {
1831 if (ip6 != NULL)
1832 ip6 = NULL;
1833 else
1834 redo_ip6 = 1;
1835 }
1836 }
1837 mtx_unlock(&pr->pr_mtx);
1838 }
1839 #endif
1840
1841 /* Let the modules do their work. */
1842 if (born) {
1843 error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1844 if (error)
1845 goto done_deref;
1846 }
1847 error = osd_jail_call(pr, PR_METHOD_SET, opts);
1848 if (error)
1849 goto done_deref;
1850
1851 /*
1852 * A new prison is now ready to be seen; either it has gained a user
1853 * reference via persistence, or is about to gain one via attachment.
1854 */
1855 if (born) {
1856 drflags = prison_lock_xlock(pr, drflags);
1857 pr->pr_state = PRISON_STATE_ALIVE;
1858 }
1859
1860 /* Attach this process to the prison if requested. */
1861 if (flags & JAIL_ATTACH) {
1862 error = do_jail_attach(td, pr,
1863 prison_lock_xlock(pr, drflags & PD_LOCK_FLAGS));
1864 drflags &= ~(PD_LOCKED | PD_LIST_XLOCKED);
1865 if (error) {
1866 vfs_opterror(opts, "attach failed");
1867 goto done_deref;
1868 }
1869 }
1870
1871 #ifdef RACCT
1872 if (racct_enable && !created) {
1873 if (drflags & PD_LOCKED) {
1874 mtx_unlock(&pr->pr_mtx);
1875 drflags &= ~PD_LOCKED;
1876 }
1877 if (drflags & PD_LIST_XLOCKED) {
1878 sx_xunlock(&allprison_lock);
1879 drflags &= ~PD_LIST_XLOCKED;
1880 }
1881 prison_racct_modify(pr);
1882 }
1883 #endif
1884
1885 drflags &= ~PD_KILL;
1886 td->td_retval[0] = pr->pr_id;
1887
1888 done_deref:
1889 /* Release any temporary prison holds and/or locks. */
1890 if (pr != NULL)
1891 prison_deref(pr, drflags);
1892 else if (drflags & PD_LIST_SLOCKED)
1893 sx_sunlock(&allprison_lock);
1894 else if (drflags & PD_LIST_XLOCKED)
1895 sx_xunlock(&allprison_lock);
1896 if (root != NULL)
1897 vrele(root);
1898 done_errmsg:
1899 if (error) {
1900 /* Write the error message back to userspace. */
1901 if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1902 &errmsg_len) == 0 && errmsg_len > 0) {
1903 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1904 if (optuio->uio_segflg == UIO_SYSSPACE)
1905 bcopy(errmsg,
1906 optuio->uio_iov[errmsg_pos].iov_base,
1907 errmsg_len);
1908 else
1909 copyout(errmsg,
1910 optuio->uio_iov[errmsg_pos].iov_base,
1911 errmsg_len);
1912 }
1913 }
1914 done_free:
1915 #ifdef INET
1916 free(ip4, M_PRISON);
1917 #endif
1918 #ifdef INET6
1919 free(ip6, M_PRISON);
1920 #endif
1921 if (g_path != NULL)
1922 free(g_path, M_TEMP);
1923 vfs_freeopts(opts);
1924 return (error);
1925 }
1926
1927 /*
1928 * Find the next available prison ID. Return the ID on success, or zero
1929 * on failure. Also set a pointer to the allprison list entry the prison
1930 * should be inserted before.
1931 */
1932 static int
get_next_prid(struct prison ** insprp)1933 get_next_prid(struct prison **insprp)
1934 {
1935 struct prison *inspr;
1936 int jid, maxid;
1937
1938 jid = lastprid % JAIL_MAX + 1;
1939 if (TAILQ_EMPTY(&allprison) ||
1940 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) {
1941 /*
1942 * A common case is for all jails to be implicitly numbered,
1943 * which means they'll go on the end of the list, at least
1944 * for the first JAIL_MAX times.
1945 */
1946 inspr = NULL;
1947 } else {
1948 /*
1949 * Take two passes through the allprison list: first starting
1950 * with the proposed jid, then ending with it.
1951 */
1952 for (maxid = JAIL_MAX; maxid != 0; ) {
1953 TAILQ_FOREACH(inspr, &allprison, pr_list) {
1954 if (inspr->pr_id < jid)
1955 continue;
1956 if (inspr->pr_id > jid) {
1957 /* Found an opening. */
1958 maxid = 0;
1959 break;
1960 }
1961 if (++jid > maxid) {
1962 if (lastprid == maxid || lastprid == 0)
1963 {
1964 /*
1965 * The entire legal range
1966 * has been traversed
1967 */
1968 return 0;
1969 }
1970 /* Try again from the start. */
1971 jid = 1;
1972 maxid = lastprid;
1973 break;
1974 }
1975 }
1976 if (inspr == NULL) {
1977 /* Found room at the end of the list. */
1978 break;
1979 }
1980 }
1981 }
1982 *insprp = inspr;
1983 lastprid = jid;
1984 return (jid);
1985 }
1986
1987 /*
1988 * struct jail_get_args {
1989 * struct iovec *iovp;
1990 * unsigned int iovcnt;
1991 * int flags;
1992 * };
1993 */
1994 int
sys_jail_get(struct thread * td,struct jail_get_args * uap)1995 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1996 {
1997 struct uio *auio;
1998 int error;
1999
2000 /* Check that we have an even number of iovecs. */
2001 if (uap->iovcnt & 1)
2002 return (EINVAL);
2003
2004 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
2005 if (error)
2006 return (error);
2007 error = kern_jail_get(td, auio, uap->flags);
2008 if (error == 0)
2009 error = copyout(auio->uio_iov, uap->iovp,
2010 uap->iovcnt * sizeof (struct iovec));
2011 free(auio, M_IOV);
2012 return (error);
2013 }
2014
2015 int
kern_jail_get(struct thread * td,struct uio * optuio,int flags)2016 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
2017 {
2018 struct bool_flags *bf;
2019 struct jailsys_flags *jsf;
2020 struct prison *pr, *mypr;
2021 struct vfsopt *opt;
2022 struct vfsoptlist *opts;
2023 char *errmsg, *name;
2024 int drflags, error, errmsg_len, errmsg_pos, i, jid, len, pos;
2025 unsigned f;
2026
2027 if (flags & ~JAIL_GET_MASK)
2028 return (EINVAL);
2029
2030 /* Get the parameter list. */
2031 error = vfs_buildopts(optuio, &opts);
2032 if (error)
2033 return (error);
2034 errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2035 mypr = td->td_ucred->cr_prison;
2036 pr = NULL;
2037
2038 /*
2039 * Find the prison specified by one of: lastjid, jid, name.
2040 */
2041 sx_slock(&allprison_lock);
2042 drflags = PD_LIST_SLOCKED;
2043 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2044 if (error == 0) {
2045 TAILQ_FOREACH(pr, &allprison, pr_list) {
2046 if (pr->pr_id > jid &&
2047 ((flags & JAIL_DYING) || prison_isalive(pr)) &&
2048 prison_ischild(mypr, pr)) {
2049 mtx_lock(&pr->pr_mtx);
2050 drflags |= PD_LOCKED;
2051 goto found_prison;
2052 }
2053 }
2054 error = ENOENT;
2055 vfs_opterror(opts, "no jail after %d", jid);
2056 goto done;
2057 } else if (error != ENOENT)
2058 goto done;
2059
2060 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2061 if (error == 0) {
2062 if (jid != 0) {
2063 pr = prison_find_child(mypr, jid);
2064 if (pr != NULL) {
2065 drflags |= PD_LOCKED;
2066 if (!(prison_isalive(pr) ||
2067 (flags & JAIL_DYING))) {
2068 error = ENOENT;
2069 vfs_opterror(opts, "jail %d is dying",
2070 jid);
2071 goto done;
2072 }
2073 goto found_prison;
2074 }
2075 error = ENOENT;
2076 vfs_opterror(opts, "jail %d not found", jid);
2077 goto done;
2078 }
2079 } else if (error != ENOENT)
2080 goto done;
2081
2082 error = vfs_getopt(opts, "name", (void **)&name, &len);
2083 if (error == 0) {
2084 if (len == 0 || name[len - 1] != '\0') {
2085 error = EINVAL;
2086 goto done;
2087 }
2088 pr = prison_find_name(mypr, name);
2089 if (pr != NULL) {
2090 drflags |= PD_LOCKED;
2091 if (!(prison_isalive(pr) || (flags & JAIL_DYING))) {
2092 error = ENOENT;
2093 vfs_opterror(opts, "jail \"%s\" is dying",
2094 name);
2095 goto done;
2096 }
2097 goto found_prison;
2098 }
2099 error = ENOENT;
2100 vfs_opterror(opts, "jail \"%s\" not found", name);
2101 goto done;
2102 } else if (error != ENOENT)
2103 goto done;
2104
2105 vfs_opterror(opts, "no jail specified");
2106 error = ENOENT;
2107 goto done;
2108
2109 found_prison:
2110 /* Get the parameters of the prison. */
2111 prison_hold(pr);
2112 drflags |= PD_DEREF;
2113 td->td_retval[0] = pr->pr_id;
2114 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2115 if (error != 0 && error != ENOENT)
2116 goto done;
2117 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2118 error = vfs_setopt(opts, "parent", &i, sizeof(i));
2119 if (error != 0 && error != ENOENT)
2120 goto done;
2121 error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2122 if (error != 0 && error != ENOENT)
2123 goto done;
2124 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2125 sizeof(pr->pr_cpuset->cs_id));
2126 if (error != 0 && error != ENOENT)
2127 goto done;
2128 error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2129 if (error != 0 && error != ENOENT)
2130 goto done;
2131 #ifdef INET
2132 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2133 pr->pr_ip4s * sizeof(*pr->pr_ip4));
2134 if (error != 0 && error != ENOENT)
2135 goto done;
2136 #endif
2137 #ifdef INET6
2138 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2139 pr->pr_ip6s * sizeof(*pr->pr_ip6));
2140 if (error != 0 && error != ENOENT)
2141 goto done;
2142 #endif
2143 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2144 sizeof(pr->pr_securelevel));
2145 if (error != 0 && error != ENOENT)
2146 goto done;
2147 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2148 sizeof(pr->pr_childcount));
2149 if (error != 0 && error != ENOENT)
2150 goto done;
2151 error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2152 sizeof(pr->pr_childmax));
2153 if (error != 0 && error != ENOENT)
2154 goto done;
2155 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2156 if (error != 0 && error != ENOENT)
2157 goto done;
2158 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2159 if (error != 0 && error != ENOENT)
2160 goto done;
2161 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2162 if (error != 0 && error != ENOENT)
2163 goto done;
2164 #ifdef COMPAT_FREEBSD32
2165 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2166 uint32_t hid32 = pr->pr_hostid;
2167
2168 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2169 } else
2170 #endif
2171 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2172 sizeof(pr->pr_hostid));
2173 if (error != 0 && error != ENOENT)
2174 goto done;
2175 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2176 sizeof(pr->pr_enforce_statfs));
2177 if (error != 0 && error != ENOENT)
2178 goto done;
2179 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2180 sizeof(pr->pr_devfs_rsnum));
2181 if (error != 0 && error != ENOENT)
2182 goto done;
2183 for (bf = pr_flag_bool;
2184 bf < pr_flag_bool + nitems(pr_flag_bool);
2185 bf++) {
2186 i = (pr->pr_flags & bf->flag) ? 1 : 0;
2187 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2188 if (error != 0 && error != ENOENT)
2189 goto done;
2190 i = !i;
2191 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2192 if (error != 0 && error != ENOENT)
2193 goto done;
2194 }
2195 for (jsf = pr_flag_jailsys;
2196 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2197 jsf++) {
2198 f = pr->pr_flags & (jsf->disable | jsf->new);
2199 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2200 : (f == jsf->new) ? JAIL_SYS_NEW
2201 : JAIL_SYS_INHERIT;
2202 error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2203 if (error != 0 && error != ENOENT)
2204 goto done;
2205 }
2206 for (bf = pr_flag_allow;
2207 bf < pr_flag_allow + nitems(pr_flag_allow) &&
2208 atomic_load_int(&bf->flag) != 0;
2209 bf++) {
2210 i = (pr->pr_allow & bf->flag) ? 1 : 0;
2211 error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2212 if (error != 0 && error != ENOENT)
2213 goto done;
2214 i = !i;
2215 error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2216 if (error != 0 && error != ENOENT)
2217 goto done;
2218 }
2219 i = !prison_isalive(pr);
2220 error = vfs_setopt(opts, "dying", &i, sizeof(i));
2221 if (error != 0 && error != ENOENT)
2222 goto done;
2223 i = !i;
2224 error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2225 if (error != 0 && error != ENOENT)
2226 goto done;
2227 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2228 sizeof(pr->pr_osreldate));
2229 if (error != 0 && error != ENOENT)
2230 goto done;
2231 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2232 if (error != 0 && error != ENOENT)
2233 goto done;
2234
2235 /* Get the module parameters. */
2236 mtx_unlock(&pr->pr_mtx);
2237 drflags &= ~PD_LOCKED;
2238 error = osd_jail_call(pr, PR_METHOD_GET, opts);
2239 if (error)
2240 goto done;
2241 prison_deref(pr, drflags);
2242 pr = NULL;
2243 drflags = 0;
2244
2245 /* By now, all parameters should have been noted. */
2246 TAILQ_FOREACH(opt, opts, link) {
2247 if (!opt->seen && strcmp(opt->name, "errmsg")) {
2248 error = EINVAL;
2249 vfs_opterror(opts, "unknown parameter: %s", opt->name);
2250 goto done;
2251 }
2252 }
2253
2254 /* Write the fetched parameters back to userspace. */
2255 error = 0;
2256 TAILQ_FOREACH(opt, opts, link) {
2257 if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2258 pos = 2 * opt->pos + 1;
2259 optuio->uio_iov[pos].iov_len = opt->len;
2260 if (opt->value != NULL) {
2261 if (optuio->uio_segflg == UIO_SYSSPACE) {
2262 bcopy(opt->value,
2263 optuio->uio_iov[pos].iov_base,
2264 opt->len);
2265 } else {
2266 error = copyout(opt->value,
2267 optuio->uio_iov[pos].iov_base,
2268 opt->len);
2269 if (error)
2270 break;
2271 }
2272 }
2273 }
2274 }
2275
2276 done:
2277 /* Release any temporary prison holds and/or locks. */
2278 if (pr != NULL)
2279 prison_deref(pr, drflags);
2280 else if (drflags & PD_LIST_SLOCKED)
2281 sx_sunlock(&allprison_lock);
2282 if (error && errmsg_pos >= 0) {
2283 /* Write the error message back to userspace. */
2284 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2285 errmsg_pos = 2 * errmsg_pos + 1;
2286 if (errmsg_len > 0) {
2287 if (optuio->uio_segflg == UIO_SYSSPACE)
2288 bcopy(errmsg,
2289 optuio->uio_iov[errmsg_pos].iov_base,
2290 errmsg_len);
2291 else
2292 copyout(errmsg,
2293 optuio->uio_iov[errmsg_pos].iov_base,
2294 errmsg_len);
2295 }
2296 }
2297 vfs_freeopts(opts);
2298 return (error);
2299 }
2300
2301 /*
2302 * struct jail_remove_args {
2303 * int jid;
2304 * };
2305 */
2306 int
sys_jail_remove(struct thread * td,struct jail_remove_args * uap)2307 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2308 {
2309 struct prison *pr;
2310 int error;
2311
2312 error = priv_check(td, PRIV_JAIL_REMOVE);
2313 if (error)
2314 return (error);
2315
2316 sx_xlock(&allprison_lock);
2317 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2318 if (pr == NULL) {
2319 sx_xunlock(&allprison_lock);
2320 return (EINVAL);
2321 }
2322 if (!prison_isalive(pr)) {
2323 /* Silently ignore already-dying prisons. */
2324 mtx_unlock(&pr->pr_mtx);
2325 sx_xunlock(&allprison_lock);
2326 return (0);
2327 }
2328 prison_deref(pr, PD_KILL | PD_LOCKED | PD_LIST_XLOCKED);
2329 return (0);
2330 }
2331
2332 /*
2333 * struct jail_attach_args {
2334 * int jid;
2335 * };
2336 */
2337 int
sys_jail_attach(struct thread * td,struct jail_attach_args * uap)2338 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2339 {
2340 struct prison *pr;
2341 int error;
2342
2343 error = priv_check(td, PRIV_JAIL_ATTACH);
2344 if (error)
2345 return (error);
2346
2347 sx_slock(&allprison_lock);
2348 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2349 if (pr == NULL) {
2350 sx_sunlock(&allprison_lock);
2351 return (EINVAL);
2352 }
2353
2354 /* Do not allow a process to attach to a prison that is not alive. */
2355 if (!prison_isalive(pr)) {
2356 mtx_unlock(&pr->pr_mtx);
2357 sx_sunlock(&allprison_lock);
2358 return (EINVAL);
2359 }
2360
2361 return (do_jail_attach(td, pr, PD_LOCKED | PD_LIST_SLOCKED));
2362 }
2363
2364 static int
do_jail_attach(struct thread * td,struct prison * pr,int drflags)2365 do_jail_attach(struct thread *td, struct prison *pr, int drflags)
2366 {
2367 struct proc *p;
2368 struct ucred *newcred, *oldcred;
2369 int error;
2370
2371 mtx_assert(&pr->pr_mtx, MA_OWNED);
2372 sx_assert(&allprison_lock, SX_LOCKED);
2373 drflags &= PD_LOCK_FLAGS;
2374 /*
2375 * XXX: Note that there is a slight race here if two threads
2376 * in the same privileged process attempt to attach to two
2377 * different jails at the same time. It is important for
2378 * user processes not to do this, or they might end up with
2379 * a process root from one prison, but attached to the jail
2380 * of another.
2381 */
2382 prison_hold(pr);
2383 refcount_acquire(&pr->pr_uref);
2384 drflags |= PD_DEREF | PD_DEUREF;
2385 mtx_unlock(&pr->pr_mtx);
2386 drflags &= ~PD_LOCKED;
2387
2388 /* Let modules do whatever they need to prepare for attaching. */
2389 error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2390 if (error) {
2391 prison_deref(pr, drflags);
2392 return (error);
2393 }
2394 sx_unlock(&allprison_lock);
2395 drflags &= ~(PD_LIST_SLOCKED | PD_LIST_XLOCKED);
2396
2397 /*
2398 * Reparent the newly attached process to this jail.
2399 */
2400 p = td->td_proc;
2401 error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2402 if (error)
2403 goto e_revert_osd;
2404
2405 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2406 if ((error = change_dir(pr->pr_root, td)) != 0)
2407 goto e_unlock;
2408 #ifdef MAC
2409 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2410 goto e_unlock;
2411 #endif
2412 VOP_UNLOCK(pr->pr_root);
2413 if ((error = pwd_chroot_chdir(td, pr->pr_root)))
2414 goto e_revert_osd;
2415
2416 newcred = crget();
2417 PROC_LOCK(p);
2418 oldcred = crcopysafe(p, newcred);
2419 newcred->cr_prison = pr;
2420 proc_set_cred(p, newcred);
2421 setsugid(p);
2422 #ifdef RACCT
2423 racct_proc_ucred_changed(p, oldcred, newcred);
2424 crhold(newcred);
2425 #endif
2426 PROC_UNLOCK(p);
2427 #ifdef RCTL
2428 rctl_proc_ucred_changed(p, newcred);
2429 crfree(newcred);
2430 #endif
2431 prison_deref(oldcred->cr_prison, drflags);
2432 crfree(oldcred);
2433
2434 /*
2435 * If the prison was killed while changing credentials, die along
2436 * with it.
2437 */
2438 if (!prison_isalive(pr)) {
2439 PROC_LOCK(p);
2440 kern_psignal(p, SIGKILL);
2441 PROC_UNLOCK(p);
2442 }
2443
2444 return (0);
2445
2446 e_unlock:
2447 VOP_UNLOCK(pr->pr_root);
2448 e_revert_osd:
2449 /* Tell modules this thread is still in its old jail after all. */
2450 sx_slock(&allprison_lock);
2451 drflags |= PD_LIST_SLOCKED;
2452 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2453 prison_deref(pr, drflags);
2454 return (error);
2455 }
2456
2457 /*
2458 * Returns a locked prison instance, or NULL on failure.
2459 */
2460 struct prison *
prison_find(int prid)2461 prison_find(int prid)
2462 {
2463 struct prison *pr;
2464
2465 sx_assert(&allprison_lock, SX_LOCKED);
2466 TAILQ_FOREACH(pr, &allprison, pr_list) {
2467 if (pr->pr_id < prid)
2468 continue;
2469 if (pr->pr_id > prid)
2470 break;
2471 KASSERT(prison_isvalid(pr), ("Found invalid prison %p", pr));
2472 mtx_lock(&pr->pr_mtx);
2473 return (pr);
2474 }
2475 return (NULL);
2476 }
2477
2478 /*
2479 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL.
2480 */
2481 struct prison *
prison_find_child(struct prison * mypr,int prid)2482 prison_find_child(struct prison *mypr, int prid)
2483 {
2484 struct prison *pr;
2485 int descend;
2486
2487 sx_assert(&allprison_lock, SX_LOCKED);
2488 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2489 if (pr->pr_id == prid) {
2490 KASSERT(prison_isvalid(pr),
2491 ("Found invalid prison %p", pr));
2492 mtx_lock(&pr->pr_mtx);
2493 return (pr);
2494 }
2495 }
2496 return (NULL);
2497 }
2498
2499 /*
2500 * Look for the name relative to mypr. Returns a locked prison or NULL.
2501 */
2502 struct prison *
prison_find_name(struct prison * mypr,const char * name)2503 prison_find_name(struct prison *mypr, const char *name)
2504 {
2505 struct prison *pr, *deadpr;
2506 size_t mylen;
2507 int descend;
2508
2509 sx_assert(&allprison_lock, SX_LOCKED);
2510 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2511 deadpr = NULL;
2512 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2513 if (!strcmp(pr->pr_name + mylen, name)) {
2514 KASSERT(prison_isvalid(pr),
2515 ("Found invalid prison %p", pr));
2516 if (prison_isalive(pr)) {
2517 mtx_lock(&pr->pr_mtx);
2518 return (pr);
2519 }
2520 deadpr = pr;
2521 }
2522 }
2523 /* There was no valid prison - perhaps there was a dying one. */
2524 if (deadpr != NULL)
2525 mtx_lock(&deadpr->pr_mtx);
2526 return (deadpr);
2527 }
2528
2529 /*
2530 * See if a prison has the specific flag set. The prison should be locked,
2531 * unless checking for flags that are only set at jail creation (such as
2532 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard
2533 * to any other prison data.
2534 */
2535 int
prison_flag(struct ucred * cred,unsigned flag)2536 prison_flag(struct ucred *cred, unsigned flag)
2537 {
2538
2539 return (cred->cr_prison->pr_flags & flag);
2540 }
2541
2542 int
prison_allow(struct ucred * cred,unsigned flag)2543 prison_allow(struct ucred *cred, unsigned flag)
2544 {
2545
2546 return ((cred->cr_prison->pr_allow & flag) != 0);
2547 }
2548
2549 /*
2550 * Hold a prison reference, by incrementing pr_ref. It is generally
2551 * an error to hold a prison that does not already have a reference.
2552 * A prison record will remain valid as long as it has at least one
2553 * reference, and will not be removed as long as either the prison
2554 * mutex or the allprison lock is held (allprison_lock may be shared).
2555 */
2556 void
prison_hold_locked(struct prison * pr)2557 prison_hold_locked(struct prison *pr)
2558 {
2559
2560 /* Locking is no longer required. */
2561 prison_hold(pr);
2562 }
2563
2564 void
prison_hold(struct prison * pr)2565 prison_hold(struct prison *pr)
2566 {
2567 #ifdef INVARIANTS
2568 int was_valid = refcount_acquire_if_not_zero(&pr->pr_ref);
2569
2570 KASSERT(was_valid,
2571 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2572 #else
2573 refcount_acquire(&pr->pr_ref);
2574 #endif
2575 }
2576
2577 /*
2578 * Remove a prison reference. If that was the last reference, the
2579 * prison will be removed (at a later time).
2580 */
2581 void
prison_free_locked(struct prison * pr)2582 prison_free_locked(struct prison *pr)
2583 {
2584
2585 mtx_assert(&pr->pr_mtx, MA_OWNED);
2586 /*
2587 * Locking is no longer required, but unlock because the caller
2588 * expects it.
2589 */
2590 mtx_unlock(&pr->pr_mtx);
2591 prison_free(pr);
2592 }
2593
2594 void
prison_free(struct prison * pr)2595 prison_free(struct prison *pr)
2596 {
2597
2598 KASSERT(refcount_load(&pr->pr_ref) > 0,
2599 ("Trying to free dead prison %p (jid=%d).",
2600 pr, pr->pr_id));
2601 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2602 /*
2603 * Don't remove the last reference in this context,
2604 * in case there are locks held.
2605 */
2606 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2607 }
2608 }
2609
2610 static void
prison_free_not_last(struct prison * pr)2611 prison_free_not_last(struct prison *pr)
2612 {
2613 #ifdef INVARIANTS
2614 int lastref;
2615
2616 KASSERT(refcount_load(&pr->pr_ref) > 0,
2617 ("Trying to free dead prison %p (jid=%d).",
2618 pr, pr->pr_id));
2619 lastref = refcount_release(&pr->pr_ref);
2620 KASSERT(!lastref,
2621 ("prison_free_not_last freed last ref on prison %p (jid=%d).",
2622 pr, pr->pr_id));
2623 #else
2624 refcount_release(&pr->pr_ref);
2625 #endif
2626 }
2627
2628 /*
2629 * Hold a a prison for user visibility, by incrementing pr_uref.
2630 * It is generally an error to hold a prison that isn't already
2631 * user-visible, except through the the jail system calls. It is also
2632 * an error to hold an invalid prison. A prison record will remain
2633 * alive as long as it has at least one user reference, and will not
2634 * be set to the dying state until the prison mutex and allprison_lock
2635 * are both freed.
2636 */
2637 void
prison_proc_hold(struct prison * pr)2638 prison_proc_hold(struct prison *pr)
2639 {
2640 #ifdef INVARIANTS
2641 int was_alive = refcount_acquire_if_not_zero(&pr->pr_uref);
2642
2643 KASSERT(was_alive,
2644 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2645 #else
2646 refcount_acquire(&pr->pr_uref);
2647 #endif
2648 }
2649
2650 /*
2651 * Remove a prison user reference. If it was the last reference, the
2652 * prison will be considered "dying", and may be removed once all of
2653 * its references are dropped.
2654 */
2655 void
prison_proc_free(struct prison * pr)2656 prison_proc_free(struct prison *pr)
2657 {
2658
2659 /*
2660 * Locking is only required when releasing the last reference.
2661 * This allows assurance that a locked prison will remain alive
2662 * until it is unlocked.
2663 */
2664 KASSERT(refcount_load(&pr->pr_uref) > 0,
2665 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2666 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2667 /*
2668 * Don't remove the last user reference in this context,
2669 * which is expected to be a process that is not only locked,
2670 * but also half dead. Add a reference so any calls to
2671 * prison_free() won't re-submit the task.
2672 */
2673 prison_hold(pr);
2674 mtx_lock(&pr->pr_mtx);
2675 KASSERT(!(pr->pr_flags & PR_COMPLETE_PROC),
2676 ("Redundant last reference in prison_proc_free (jid=%d)",
2677 pr->pr_id));
2678 pr->pr_flags |= PR_COMPLETE_PROC;
2679 mtx_unlock(&pr->pr_mtx);
2680 taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2681 }
2682 }
2683
2684 static void
prison_proc_free_not_last(struct prison * pr)2685 prison_proc_free_not_last(struct prison *pr)
2686 {
2687 #ifdef INVARIANTS
2688 int lastref;
2689
2690 KASSERT(refcount_load(&pr->pr_uref) > 0,
2691 ("Trying to free dead prison %p (jid=%d).",
2692 pr, pr->pr_id));
2693 lastref = refcount_release(&pr->pr_uref);
2694 KASSERT(!lastref,
2695 ("prison_proc_free_not_last freed last uref on prison %p (jid=%d).",
2696 pr, pr->pr_id));
2697 #else
2698 refcount_release(&pr->pr_uref);
2699 #endif
2700 }
2701
2702 /*
2703 * Complete a call to either prison_free or prison_proc_free.
2704 */
2705 static void
prison_complete(void * context,int pending)2706 prison_complete(void *context, int pending)
2707 {
2708 struct prison *pr = context;
2709 int drflags;
2710
2711 /*
2712 * This could be called to release the last reference, or the last
2713 * user reference (plus the reference held in prison_proc_free).
2714 */
2715 drflags = prison_lock_xlock(pr, PD_DEREF);
2716 if (pr->pr_flags & PR_COMPLETE_PROC) {
2717 pr->pr_flags &= ~PR_COMPLETE_PROC;
2718 drflags |= PD_DEUREF;
2719 }
2720 prison_deref(pr, drflags);
2721 }
2722
2723 /*
2724 * Remove a prison reference and/or user reference (usually).
2725 * This assumes context that allows sleeping (for allprison_lock),
2726 * with no non-sleeping locks held, except perhaps the prison itself.
2727 * If there are no more references, release and delist the prison.
2728 * On completion, the prison lock and the allprison lock are both
2729 * unlocked.
2730 */
2731 static void
prison_deref(struct prison * pr,int flags)2732 prison_deref(struct prison *pr, int flags)
2733 {
2734 struct prisonlist freeprison;
2735 struct prison *killpr, *rpr, *ppr, *tpr;
2736 struct proc *p;
2737
2738 killpr = NULL;
2739 TAILQ_INIT(&freeprison);
2740 /*
2741 * Release this prison as requested, which may cause its parent
2742 * to be released, and then maybe its grandparent, etc.
2743 */
2744 for (;;) {
2745 if (flags & PD_KILL) {
2746 /* Kill the prison and its descendents. */
2747 KASSERT(pr != &prison0,
2748 ("prison_deref trying to kill prison0"));
2749 if (!(flags & PD_DEREF)) {
2750 prison_hold(pr);
2751 flags |= PD_DEREF;
2752 }
2753 flags = prison_lock_xlock(pr, flags);
2754 prison_deref_kill(pr, &freeprison);
2755 }
2756 if (flags & PD_DEUREF) {
2757 /* Drop a user reference. */
2758 KASSERT(refcount_load(&pr->pr_uref) > 0,
2759 ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2760 pr->pr_id));
2761 if (!refcount_release_if_not_last(&pr->pr_uref)) {
2762 if (!(flags & PD_DEREF)) {
2763 prison_hold(pr);
2764 flags |= PD_DEREF;
2765 }
2766 flags = prison_lock_xlock(pr, flags);
2767 if (refcount_release(&pr->pr_uref) &&
2768 pr->pr_state == PRISON_STATE_ALIVE) {
2769 /*
2770 * When the last user references goes,
2771 * this becomes a dying prison.
2772 */
2773 KASSERT(
2774 refcount_load(&prison0.pr_uref) > 0,
2775 ("prison0 pr_uref=0"));
2776 pr->pr_state = PRISON_STATE_DYING;
2777 mtx_unlock(&pr->pr_mtx);
2778 flags &= ~PD_LOCKED;
2779 (void)osd_jail_call(pr,
2780 PR_METHOD_REMOVE, NULL);
2781 }
2782 }
2783 }
2784 if (flags & PD_KILL) {
2785 /*
2786 * Any remaining user references are probably processes
2787 * that need to be killed, either in this prison or its
2788 * descendants.
2789 */
2790 if (refcount_load(&pr->pr_uref) > 0)
2791 killpr = pr;
2792 /* Make sure the parent prison doesn't get killed. */
2793 flags &= ~PD_KILL;
2794 }
2795 if (flags & PD_DEREF) {
2796 /* Drop a reference. */
2797 KASSERT(refcount_load(&pr->pr_ref) > 0,
2798 ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2799 pr->pr_id));
2800 if (!refcount_release_if_not_last(&pr->pr_ref)) {
2801 flags = prison_lock_xlock(pr, flags);
2802 if (refcount_release(&pr->pr_ref)) {
2803 /*
2804 * When the last reference goes,
2805 * unlink the prison and set it aside.
2806 */
2807 KASSERT(
2808 refcount_load(&pr->pr_uref) == 0,
2809 ("prison_deref: last ref, "
2810 "but still has %d urefs (jid=%d)",
2811 pr->pr_uref, pr->pr_id));
2812 KASSERT(
2813 refcount_load(&prison0.pr_ref) != 0,
2814 ("prison0 pr_ref=0"));
2815 pr->pr_state = PRISON_STATE_INVALID;
2816 TAILQ_REMOVE(&allprison, pr, pr_list);
2817 LIST_REMOVE(pr, pr_sibling);
2818 TAILQ_INSERT_TAIL(&freeprison, pr,
2819 pr_list);
2820 for (ppr = pr->pr_parent;
2821 ppr != NULL;
2822 ppr = ppr->pr_parent)
2823 ppr->pr_childcount--;
2824 /*
2825 * Removing a prison frees references
2826 * from its parent.
2827 */
2828 mtx_unlock(&pr->pr_mtx);
2829 flags &= ~PD_LOCKED;
2830 pr = pr->pr_parent;
2831 flags |= PD_DEREF | PD_DEUREF;
2832 continue;
2833 }
2834 }
2835 }
2836 break;
2837 }
2838
2839 /* Release all the prison locks. */
2840 if (flags & PD_LOCKED)
2841 mtx_unlock(&pr->pr_mtx);
2842 if (flags & PD_LIST_SLOCKED)
2843 sx_sunlock(&allprison_lock);
2844 else if (flags & PD_LIST_XLOCKED)
2845 sx_xunlock(&allprison_lock);
2846
2847 /* Kill any processes attached to a killed prison. */
2848 if (killpr != NULL) {
2849 sx_slock(&allproc_lock);
2850 FOREACH_PROC_IN_SYSTEM(p) {
2851 PROC_LOCK(p);
2852 if (p->p_state != PRS_NEW && p->p_ucred != NULL) {
2853 for (ppr = p->p_ucred->cr_prison;
2854 ppr != &prison0;
2855 ppr = ppr->pr_parent)
2856 if (ppr == killpr) {
2857 kern_psignal(p, SIGKILL);
2858 break;
2859 }
2860 }
2861 PROC_UNLOCK(p);
2862 }
2863 sx_sunlock(&allproc_lock);
2864 }
2865
2866 /*
2867 * Finish removing any unreferenced prisons, which couldn't happen
2868 * while allprison_lock was held (to avoid a LOR on vrele).
2869 */
2870 TAILQ_FOREACH_SAFE(rpr, &freeprison, pr_list, tpr) {
2871 #ifdef VIMAGE
2872 if (rpr->pr_vnet != rpr->pr_parent->pr_vnet)
2873 vnet_destroy(rpr->pr_vnet);
2874 #endif
2875 if (rpr->pr_root != NULL)
2876 vrele(rpr->pr_root);
2877 mtx_destroy(&rpr->pr_mtx);
2878 #ifdef INET
2879 free(rpr->pr_ip4, M_PRISON);
2880 #endif
2881 #ifdef INET6
2882 free(rpr->pr_ip6, M_PRISON);
2883 #endif
2884 if (rpr->pr_cpuset != NULL)
2885 cpuset_rel(rpr->pr_cpuset);
2886 osd_jail_exit(rpr);
2887 #ifdef RACCT
2888 if (racct_enable)
2889 prison_racct_detach(rpr);
2890 #endif
2891 TAILQ_REMOVE(&freeprison, rpr, pr_list);
2892 free(rpr, M_PRISON);
2893 }
2894 }
2895
2896 /*
2897 * Kill the prison and its descendants. Mark them as dying, clear the
2898 * persist flag, and call module remove methods.
2899 */
2900 static void
prison_deref_kill(struct prison * pr,struct prisonlist * freeprison)2901 prison_deref_kill(struct prison *pr, struct prisonlist *freeprison)
2902 {
2903 struct prison *cpr, *ppr, *rpr;
2904 bool descend;
2905
2906 /*
2907 * Unlike the descendants, the target prison can be killed
2908 * even if it is currently dying. This is useful for failed
2909 * creation in jail_set(2).
2910 */
2911 KASSERT(refcount_load(&pr->pr_ref) > 0,
2912 ("Trying to kill dead prison %p (jid=%d).",
2913 pr, pr->pr_id));
2914 refcount_acquire(&pr->pr_uref);
2915 pr->pr_state = PRISON_STATE_DYING;
2916 mtx_unlock(&pr->pr_mtx);
2917
2918 rpr = NULL;
2919 FOREACH_PRISON_DESCENDANT_PRE_POST(pr, cpr, descend) {
2920 if (descend) {
2921 if (!prison_isalive(cpr)) {
2922 descend = false;
2923 continue;
2924 }
2925 prison_hold(cpr);
2926 prison_proc_hold(cpr);
2927 mtx_lock(&cpr->pr_mtx);
2928 cpr->pr_state = PRISON_STATE_DYING;
2929 cpr->pr_flags |= PR_REMOVE;
2930 mtx_unlock(&cpr->pr_mtx);
2931 continue;
2932 }
2933 if (!(cpr->pr_flags & PR_REMOVE))
2934 continue;
2935 (void)osd_jail_call(cpr, PR_METHOD_REMOVE, NULL);
2936 mtx_lock(&cpr->pr_mtx);
2937 cpr->pr_flags &= ~PR_REMOVE;
2938 if (cpr->pr_flags & PR_PERSIST) {
2939 cpr->pr_flags &= ~PR_PERSIST;
2940 prison_proc_free_not_last(cpr);
2941 prison_free_not_last(cpr);
2942 }
2943 (void)refcount_release(&cpr->pr_uref);
2944 if (refcount_release(&cpr->pr_ref)) {
2945 /*
2946 * When the last reference goes, unlink the prison
2947 * and set it aside for prison_deref() to handle.
2948 * Delay unlinking the sibling list to keep the loop
2949 * safe.
2950 */
2951 if (rpr != NULL)
2952 LIST_REMOVE(rpr, pr_sibling);
2953 rpr = cpr;
2954 rpr->pr_state = PRISON_STATE_INVALID;
2955 TAILQ_REMOVE(&allprison, rpr, pr_list);
2956 TAILQ_INSERT_TAIL(freeprison, rpr, pr_list);
2957 /*
2958 * Removing a prison frees references from its parent.
2959 */
2960 ppr = rpr->pr_parent;
2961 prison_proc_free_not_last(ppr);
2962 prison_free_not_last(ppr);
2963 for (; ppr != NULL; ppr = ppr->pr_parent)
2964 ppr->pr_childcount--;
2965 }
2966 mtx_unlock(&cpr->pr_mtx);
2967 }
2968 if (rpr != NULL)
2969 LIST_REMOVE(rpr, pr_sibling);
2970
2971 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2972 mtx_lock(&pr->pr_mtx);
2973 if (pr->pr_flags & PR_PERSIST) {
2974 pr->pr_flags &= ~PR_PERSIST;
2975 prison_proc_free_not_last(pr);
2976 prison_free_not_last(pr);
2977 }
2978 (void)refcount_release(&pr->pr_uref);
2979 }
2980
2981 /*
2982 * Given the current locking state in the flags, make sure allprison_lock
2983 * is held exclusive, and the prison is locked. Return flags indicating
2984 * the new state.
2985 */
2986 static int
prison_lock_xlock(struct prison * pr,int flags)2987 prison_lock_xlock(struct prison *pr, int flags)
2988 {
2989
2990 if (!(flags & PD_LIST_XLOCKED)) {
2991 /*
2992 * Get allprison_lock, which may be an upgrade,
2993 * and may require unlocking the prison.
2994 */
2995 if (flags & PD_LOCKED) {
2996 mtx_unlock(&pr->pr_mtx);
2997 flags &= ~PD_LOCKED;
2998 }
2999 if (flags & PD_LIST_SLOCKED) {
3000 if (!sx_try_upgrade(&allprison_lock)) {
3001 sx_sunlock(&allprison_lock);
3002 sx_xlock(&allprison_lock);
3003 }
3004 flags &= ~PD_LIST_SLOCKED;
3005 } else
3006 sx_xlock(&allprison_lock);
3007 flags |= PD_LIST_XLOCKED;
3008 }
3009 if (!(flags & PD_LOCKED)) {
3010 /* Lock the prison mutex. */
3011 mtx_lock(&pr->pr_mtx);
3012 flags |= PD_LOCKED;
3013 }
3014 return flags;
3015 }
3016
3017 /*
3018 * Set or clear a permission bit in the pr_allow field, passing restrictions
3019 * (cleared permission) down to child jails.
3020 */
3021 void
prison_set_allow(struct ucred * cred,unsigned flag,int enable)3022 prison_set_allow(struct ucred *cred, unsigned flag, int enable)
3023 {
3024 struct prison *pr;
3025
3026 pr = cred->cr_prison;
3027 sx_slock(&allprison_lock);
3028 mtx_lock(&pr->pr_mtx);
3029 prison_set_allow_locked(pr, flag, enable);
3030 mtx_unlock(&pr->pr_mtx);
3031 sx_sunlock(&allprison_lock);
3032 }
3033
3034 static void
prison_set_allow_locked(struct prison * pr,unsigned flag,int enable)3035 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable)
3036 {
3037 struct prison *cpr;
3038 int descend;
3039
3040 if (enable != 0)
3041 pr->pr_allow |= flag;
3042 else {
3043 pr->pr_allow &= ~flag;
3044 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend)
3045 cpr->pr_allow &= ~flag;
3046 }
3047 }
3048
3049 /*
3050 * Check if a jail supports the given address family.
3051 *
3052 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3053 * if not.
3054 */
3055 int
prison_check_af(struct ucred * cred,int af)3056 prison_check_af(struct ucred *cred, int af)
3057 {
3058 struct prison *pr;
3059 int error;
3060
3061 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3062
3063 pr = cred->cr_prison;
3064 #ifdef VIMAGE
3065 /* Prisons with their own network stack are not limited. */
3066 if (prison_owns_vnet(cred))
3067 return (0);
3068 #endif
3069
3070 error = 0;
3071 switch (af)
3072 {
3073 #ifdef INET
3074 case AF_INET:
3075 if (pr->pr_flags & PR_IP4)
3076 {
3077 mtx_lock(&pr->pr_mtx);
3078 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3079 error = EAFNOSUPPORT;
3080 mtx_unlock(&pr->pr_mtx);
3081 }
3082 break;
3083 #endif
3084 #ifdef INET6
3085 case AF_INET6:
3086 if (pr->pr_flags & PR_IP6)
3087 {
3088 mtx_lock(&pr->pr_mtx);
3089 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3090 error = EAFNOSUPPORT;
3091 mtx_unlock(&pr->pr_mtx);
3092 }
3093 break;
3094 #endif
3095 case AF_LOCAL:
3096 case AF_ROUTE:
3097 break;
3098 default:
3099 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3100 error = EAFNOSUPPORT;
3101 }
3102 return (error);
3103 }
3104
3105 /*
3106 * Check if given address belongs to the jail referenced by cred (wrapper to
3107 * prison_check_ip[46]).
3108 *
3109 * Returns 0 if jail doesn't restrict the address family or if address belongs
3110 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3111 * the jail doesn't allow the address family. IPv4 Address passed in in NBO.
3112 */
3113 int
prison_if(struct ucred * cred,const struct sockaddr * sa)3114 prison_if(struct ucred *cred, const struct sockaddr *sa)
3115 {
3116 #ifdef INET
3117 const struct sockaddr_in *sai;
3118 #endif
3119 #ifdef INET6
3120 const struct sockaddr_in6 *sai6;
3121 #endif
3122 int error;
3123
3124 KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3125 KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3126
3127 #ifdef VIMAGE
3128 if (prison_owns_vnet(cred))
3129 return (0);
3130 #endif
3131
3132 error = 0;
3133 switch (sa->sa_family)
3134 {
3135 #ifdef INET
3136 case AF_INET:
3137 sai = (const struct sockaddr_in *)sa;
3138 error = prison_check_ip4(cred, &sai->sin_addr);
3139 break;
3140 #endif
3141 #ifdef INET6
3142 case AF_INET6:
3143 sai6 = (const struct sockaddr_in6 *)sa;
3144 error = prison_check_ip6(cred, &sai6->sin6_addr);
3145 break;
3146 #endif
3147 default:
3148 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3149 error = EAFNOSUPPORT;
3150 }
3151 return (error);
3152 }
3153
3154 /*
3155 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3156 */
3157 int
prison_check(struct ucred * cred1,struct ucred * cred2)3158 prison_check(struct ucred *cred1, struct ucred *cred2)
3159 {
3160
3161 return ((cred1->cr_prison == cred2->cr_prison ||
3162 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3163 }
3164
3165 /*
3166 * Return 1 if p2 is a child of p1, otherwise 0.
3167 */
3168 int
prison_ischild(struct prison * pr1,struct prison * pr2)3169 prison_ischild(struct prison *pr1, struct prison *pr2)
3170 {
3171
3172 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3173 if (pr1 == pr2)
3174 return (1);
3175 return (0);
3176 }
3177
3178 /*
3179 * Return true if the prison is currently alive. A prison is alive if it
3180 * holds user references and it isn't being removed.
3181 */
3182 bool
prison_isalive(struct prison * pr)3183 prison_isalive(struct prison *pr)
3184 {
3185
3186 if (__predict_false(pr->pr_state != PRISON_STATE_ALIVE))
3187 return (false);
3188 return (true);
3189 }
3190
3191 /*
3192 * Return true if the prison is currently valid. A prison is valid if it has
3193 * been fully created, and is not being destroyed. Note that dying prisons
3194 * are still considered valid. Invalid prisons won't be found under normal
3195 * circumstances, as they're only put in that state by functions that have
3196 * an exclusive hold on allprison_lock.
3197 */
3198 bool
prison_isvalid(struct prison * pr)3199 prison_isvalid(struct prison *pr)
3200 {
3201
3202 if (__predict_false(pr->pr_state == PRISON_STATE_INVALID))
3203 return (false);
3204 if (__predict_false(refcount_load(&pr->pr_ref) == 0))
3205 return (false);
3206 return (true);
3207 }
3208
3209 /*
3210 * Return 1 if the passed credential is in a jail and that jail does not
3211 * have its own virtual network stack, otherwise 0.
3212 */
3213 int
jailed_without_vnet(struct ucred * cred)3214 jailed_without_vnet(struct ucred *cred)
3215 {
3216
3217 if (!jailed(cred))
3218 return (0);
3219 #ifdef VIMAGE
3220 if (prison_owns_vnet(cred))
3221 return (0);
3222 #endif
3223
3224 return (1);
3225 }
3226
3227 /*
3228 * Return the correct hostname (domainname, et al) for the passed credential.
3229 */
3230 void
getcredhostname(struct ucred * cred,char * buf,size_t size)3231 getcredhostname(struct ucred *cred, char *buf, size_t size)
3232 {
3233 struct prison *pr;
3234
3235 /*
3236 * A NULL credential can be used to shortcut to the physical
3237 * system's hostname.
3238 */
3239 pr = (cred != NULL) ? cred->cr_prison : &prison0;
3240 mtx_lock(&pr->pr_mtx);
3241 strlcpy(buf, pr->pr_hostname, size);
3242 mtx_unlock(&pr->pr_mtx);
3243 }
3244
3245 void
getcreddomainname(struct ucred * cred,char * buf,size_t size)3246 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3247 {
3248
3249 mtx_lock(&cred->cr_prison->pr_mtx);
3250 strlcpy(buf, cred->cr_prison->pr_domainname, size);
3251 mtx_unlock(&cred->cr_prison->pr_mtx);
3252 }
3253
3254 void
getcredhostuuid(struct ucred * cred,char * buf,size_t size)3255 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3256 {
3257
3258 mtx_lock(&cred->cr_prison->pr_mtx);
3259 strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3260 mtx_unlock(&cred->cr_prison->pr_mtx);
3261 }
3262
3263 void
getcredhostid(struct ucred * cred,unsigned long * hostid)3264 getcredhostid(struct ucred *cred, unsigned long *hostid)
3265 {
3266
3267 mtx_lock(&cred->cr_prison->pr_mtx);
3268 *hostid = cred->cr_prison->pr_hostid;
3269 mtx_unlock(&cred->cr_prison->pr_mtx);
3270 }
3271
3272 void
getjailname(struct ucred * cred,char * name,size_t len)3273 getjailname(struct ucred *cred, char *name, size_t len)
3274 {
3275
3276 mtx_lock(&cred->cr_prison->pr_mtx);
3277 strlcpy(name, cred->cr_prison->pr_name, len);
3278 mtx_unlock(&cred->cr_prison->pr_mtx);
3279 }
3280
3281 #ifdef VIMAGE
3282 /*
3283 * Determine whether the prison represented by cred owns
3284 * its vnet rather than having it inherited.
3285 *
3286 * Returns 1 in case the prison owns the vnet, 0 otherwise.
3287 */
3288 int
prison_owns_vnet(struct ucred * cred)3289 prison_owns_vnet(struct ucred *cred)
3290 {
3291
3292 /*
3293 * vnets cannot be added/removed after jail creation,
3294 * so no need to lock here.
3295 */
3296 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3297 }
3298 #endif
3299
3300 /*
3301 * Determine whether the subject represented by cred can "see"
3302 * status of a mount point.
3303 * Returns: 0 for permitted, ENOENT otherwise.
3304 * XXX: This function should be called cr_canseemount() and should be
3305 * placed in kern_prot.c.
3306 */
3307 int
prison_canseemount(struct ucred * cred,struct mount * mp)3308 prison_canseemount(struct ucred *cred, struct mount *mp)
3309 {
3310 struct prison *pr;
3311 struct statfs *sp;
3312 size_t len;
3313
3314 pr = cred->cr_prison;
3315 if (pr->pr_enforce_statfs == 0)
3316 return (0);
3317 if (pr->pr_root->v_mount == mp)
3318 return (0);
3319 if (pr->pr_enforce_statfs == 2)
3320 return (ENOENT);
3321 /*
3322 * If jail's chroot directory is set to "/" we should be able to see
3323 * all mount-points from inside a jail.
3324 * This is ugly check, but this is the only situation when jail's
3325 * directory ends with '/'.
3326 */
3327 if (strcmp(pr->pr_path, "/") == 0)
3328 return (0);
3329 len = strlen(pr->pr_path);
3330 sp = &mp->mnt_stat;
3331 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3332 return (ENOENT);
3333 /*
3334 * Be sure that we don't have situation where jail's root directory
3335 * is "/some/path" and mount point is "/some/pathpath".
3336 */
3337 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3338 return (ENOENT);
3339 return (0);
3340 }
3341
3342 void
prison_enforce_statfs(struct ucred * cred,struct mount * mp,struct statfs * sp)3343 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3344 {
3345 char jpath[MAXPATHLEN];
3346 struct prison *pr;
3347 size_t len;
3348
3349 pr = cred->cr_prison;
3350 if (pr->pr_enforce_statfs == 0)
3351 return;
3352 if (prison_canseemount(cred, mp) != 0) {
3353 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3354 strlcpy(sp->f_mntonname, "[restricted]",
3355 sizeof(sp->f_mntonname));
3356 return;
3357 }
3358 if (pr->pr_root->v_mount == mp) {
3359 /*
3360 * Clear current buffer data, so we are sure nothing from
3361 * the valid path left there.
3362 */
3363 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3364 *sp->f_mntonname = '/';
3365 return;
3366 }
3367 /*
3368 * If jail's chroot directory is set to "/" we should be able to see
3369 * all mount-points from inside a jail.
3370 */
3371 if (strcmp(pr->pr_path, "/") == 0)
3372 return;
3373 len = strlen(pr->pr_path);
3374 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3375 /*
3376 * Clear current buffer data, so we are sure nothing from
3377 * the valid path left there.
3378 */
3379 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3380 if (*jpath == '\0') {
3381 /* Should never happen. */
3382 *sp->f_mntonname = '/';
3383 } else {
3384 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3385 }
3386 }
3387
3388 /*
3389 * Check with permission for a specific privilege is granted within jail. We
3390 * have a specific list of accepted privileges; the rest are denied.
3391 */
3392 int
prison_priv_check(struct ucred * cred,int priv)3393 prison_priv_check(struct ucred *cred, int priv)
3394 {
3395 struct prison *pr;
3396 int error;
3397
3398 /*
3399 * Some policies have custom handlers. This routine should not be
3400 * called for them. See priv_check_cred().
3401 */
3402 switch (priv) {
3403 case PRIV_VFS_LOOKUP:
3404 case PRIV_VFS_GENERATION:
3405 KASSERT(0, ("prison_priv_check instead of a custom handler "
3406 "called for %d\n", priv));
3407 }
3408
3409 if (!jailed(cred))
3410 return (0);
3411
3412 #ifdef VIMAGE
3413 /*
3414 * Privileges specific to prisons with a virtual network stack.
3415 * There might be a duplicate entry here in case the privilege
3416 * is only granted conditionally in the legacy jail case.
3417 */
3418 switch (priv) {
3419 #ifdef notyet
3420 /*
3421 * NFS-specific privileges.
3422 */
3423 case PRIV_NFS_DAEMON:
3424 case PRIV_NFS_LOCKD:
3425 #endif
3426 /*
3427 * Network stack privileges.
3428 */
3429 case PRIV_NET_BRIDGE:
3430 case PRIV_NET_GRE:
3431 case PRIV_NET_BPF:
3432 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */
3433 case PRIV_NET_ROUTE:
3434 case PRIV_NET_TAP:
3435 case PRIV_NET_SETIFMTU:
3436 case PRIV_NET_SETIFFLAGS:
3437 case PRIV_NET_SETIFCAP:
3438 case PRIV_NET_SETIFDESCR:
3439 case PRIV_NET_SETIFNAME :
3440 case PRIV_NET_SETIFMETRIC:
3441 case PRIV_NET_SETIFPHYS:
3442 case PRIV_NET_SETIFMAC:
3443 case PRIV_NET_SETLANPCP:
3444 case PRIV_NET_ADDMULTI:
3445 case PRIV_NET_DELMULTI:
3446 case PRIV_NET_HWIOCTL:
3447 case PRIV_NET_SETLLADDR:
3448 case PRIV_NET_ADDIFGROUP:
3449 case PRIV_NET_DELIFGROUP:
3450 case PRIV_NET_IFCREATE:
3451 case PRIV_NET_IFDESTROY:
3452 case PRIV_NET_ADDIFADDR:
3453 case PRIV_NET_DELIFADDR:
3454 case PRIV_NET_LAGG:
3455 case PRIV_NET_GIF:
3456 case PRIV_NET_SETIFVNET:
3457 case PRIV_NET_SETIFFIB:
3458
3459 /*
3460 * 802.11-related privileges.
3461 */
3462 case PRIV_NET80211_VAP_GETKEY:
3463 case PRIV_NET80211_VAP_MANAGE:
3464
3465 #ifdef notyet
3466 /*
3467 * ATM privileges.
3468 */
3469 case PRIV_NETATM_CFG:
3470 case PRIV_NETATM_ADD:
3471 case PRIV_NETATM_DEL:
3472 case PRIV_NETATM_SET:
3473
3474 /*
3475 * Bluetooth privileges.
3476 */
3477 case PRIV_NETBLUETOOTH_RAW:
3478 #endif
3479
3480 /*
3481 * Netgraph and netgraph module privileges.
3482 */
3483 case PRIV_NETGRAPH_CONTROL:
3484 #ifdef notyet
3485 case PRIV_NETGRAPH_TTY:
3486 #endif
3487
3488 /*
3489 * IPv4 and IPv6 privileges.
3490 */
3491 case PRIV_NETINET_IPFW:
3492 case PRIV_NETINET_DIVERT:
3493 case PRIV_NETINET_PF:
3494 case PRIV_NETINET_DUMMYNET:
3495 case PRIV_NETINET_CARP:
3496 case PRIV_NETINET_MROUTE:
3497 case PRIV_NETINET_RAW:
3498 case PRIV_NETINET_ADDRCTRL6:
3499 case PRIV_NETINET_ND6:
3500 case PRIV_NETINET_SCOPE6:
3501 case PRIV_NETINET_ALIFETIME6:
3502 case PRIV_NETINET_IPSEC:
3503 case PRIV_NETINET_BINDANY:
3504
3505 #ifdef notyet
3506 /*
3507 * NCP privileges.
3508 */
3509 case PRIV_NETNCP:
3510
3511 /*
3512 * SMB privileges.
3513 */
3514 case PRIV_NETSMB:
3515 #endif
3516
3517 /*
3518 * No default: or deny here.
3519 * In case of no permit fall through to next switch().
3520 */
3521 if (cred->cr_prison->pr_flags & PR_VNET)
3522 return (0);
3523 }
3524 #endif /* VIMAGE */
3525
3526 switch (priv) {
3527 /*
3528 * Allow ktrace privileges for root in jail.
3529 */
3530 case PRIV_KTRACE:
3531
3532 #if 0
3533 /*
3534 * Allow jailed processes to configure audit identity and
3535 * submit audit records (login, etc). In the future we may
3536 * want to further refine the relationship between audit and
3537 * jail.
3538 */
3539 case PRIV_AUDIT_GETAUDIT:
3540 case PRIV_AUDIT_SETAUDIT:
3541 case PRIV_AUDIT_SUBMIT:
3542 #endif
3543
3544 /*
3545 * Allow jailed processes to manipulate process UNIX
3546 * credentials in any way they see fit.
3547 */
3548 case PRIV_CRED_SETUID:
3549 case PRIV_CRED_SETEUID:
3550 case PRIV_CRED_SETGID:
3551 case PRIV_CRED_SETEGID:
3552 case PRIV_CRED_SETGROUPS:
3553 case PRIV_CRED_SETREUID:
3554 case PRIV_CRED_SETREGID:
3555 case PRIV_CRED_SETRESUID:
3556 case PRIV_CRED_SETRESGID:
3557
3558 /*
3559 * Jail implements visibility constraints already, so allow
3560 * jailed root to override uid/gid-based constraints.
3561 */
3562 case PRIV_SEEOTHERGIDS:
3563 case PRIV_SEEOTHERUIDS:
3564
3565 /*
3566 * Jail implements inter-process debugging limits already, so
3567 * allow jailed root various debugging privileges.
3568 */
3569 case PRIV_DEBUG_DIFFCRED:
3570 case PRIV_DEBUG_SUGID:
3571 case PRIV_DEBUG_UNPRIV:
3572
3573 /*
3574 * Allow jail to set various resource limits and login
3575 * properties, and for now, exceed process resource limits.
3576 */
3577 case PRIV_PROC_LIMIT:
3578 case PRIV_PROC_SETLOGIN:
3579 case PRIV_PROC_SETRLIMIT:
3580
3581 /*
3582 * System V and POSIX IPC privileges are granted in jail.
3583 */
3584 case PRIV_IPC_READ:
3585 case PRIV_IPC_WRITE:
3586 case PRIV_IPC_ADMIN:
3587 case PRIV_IPC_MSGSIZE:
3588 case PRIV_MQ_ADMIN:
3589
3590 /*
3591 * Jail operations within a jail work on child jails.
3592 */
3593 case PRIV_JAIL_ATTACH:
3594 case PRIV_JAIL_SET:
3595 case PRIV_JAIL_REMOVE:
3596
3597 /*
3598 * Jail implements its own inter-process limits, so allow
3599 * root processes in jail to change scheduling on other
3600 * processes in the same jail. Likewise for signalling.
3601 */
3602 case PRIV_SCHED_DIFFCRED:
3603 case PRIV_SCHED_CPUSET:
3604 case PRIV_SIGNAL_DIFFCRED:
3605 case PRIV_SIGNAL_SUGID:
3606
3607 /*
3608 * Allow jailed processes to write to sysctls marked as jail
3609 * writable.
3610 */
3611 case PRIV_SYSCTL_WRITEJAIL:
3612
3613 /*
3614 * Allow root in jail to manage a variety of quota
3615 * properties. These should likely be conditional on a
3616 * configuration option.
3617 */
3618 case PRIV_VFS_GETQUOTA:
3619 case PRIV_VFS_SETQUOTA:
3620
3621 /*
3622 * Since Jail relies on chroot() to implement file system
3623 * protections, grant many VFS privileges to root in jail.
3624 * Be careful to exclude mount-related and NFS-related
3625 * privileges.
3626 */
3627 case PRIV_VFS_READ:
3628 case PRIV_VFS_WRITE:
3629 case PRIV_VFS_ADMIN:
3630 case PRIV_VFS_EXEC:
3631 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */
3632 case PRIV_VFS_CHFLAGS_DEV:
3633 case PRIV_VFS_CHOWN:
3634 case PRIV_VFS_CHROOT:
3635 case PRIV_VFS_RETAINSUGID:
3636 case PRIV_VFS_FCHROOT:
3637 case PRIV_VFS_LINK:
3638 case PRIV_VFS_SETGID:
3639 case PRIV_VFS_STAT:
3640 case PRIV_VFS_STICKYFILE:
3641
3642 /*
3643 * As in the non-jail case, non-root users are expected to be
3644 * able to read kernel/physical memory (provided /dev/[k]mem
3645 * exists in the jail and they have permission to access it).
3646 */
3647 case PRIV_KMEM_READ:
3648 return (0);
3649
3650 /*
3651 * Depending on the global setting, allow privilege of
3652 * setting system flags.
3653 */
3654 case PRIV_VFS_SYSFLAGS:
3655 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3656 return (0);
3657 else
3658 return (EPERM);
3659
3660 /*
3661 * Depending on the global setting, allow privilege of
3662 * mounting/unmounting file systems.
3663 */
3664 case PRIV_VFS_MOUNT:
3665 case PRIV_VFS_UNMOUNT:
3666 case PRIV_VFS_MOUNT_NONUSER:
3667 case PRIV_VFS_MOUNT_OWNER:
3668 pr = cred->cr_prison;
3669 prison_lock(pr);
3670 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2)
3671 error = 0;
3672 else
3673 error = EPERM;
3674 prison_unlock(pr);
3675 return (error);
3676
3677 /*
3678 * Jails should hold no disposition on the PRIV_VFS_READ_DIR
3679 * policy. priv_check_cred will not specifically allow it, and
3680 * we may want a MAC policy to allow it.
3681 */
3682 case PRIV_VFS_READ_DIR:
3683 return (0);
3684
3685 /*
3686 * Conditionnaly allow locking (unlocking) physical pages
3687 * in memory.
3688 */
3689 case PRIV_VM_MLOCK:
3690 case PRIV_VM_MUNLOCK:
3691 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3692 return (0);
3693 else
3694 return (EPERM);
3695
3696 /*
3697 * Conditionally allow jailed root to bind reserved ports.
3698 */
3699 case PRIV_NETINET_RESERVEDPORT:
3700 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3701 return (0);
3702 else
3703 return (EPERM);
3704
3705 /*
3706 * Allow jailed root to reuse in-use ports.
3707 */
3708 case PRIV_NETINET_REUSEPORT:
3709 return (0);
3710
3711 /*
3712 * Allow jailed root to set certain IPv4/6 (option) headers.
3713 */
3714 case PRIV_NETINET_SETHDROPTS:
3715 return (0);
3716
3717 /*
3718 * Conditionally allow creating raw sockets in jail.
3719 */
3720 case PRIV_NETINET_RAW:
3721 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3722 return (0);
3723 else
3724 return (EPERM);
3725
3726 /*
3727 * Since jail implements its own visibility limits on netstat
3728 * sysctls, allow getcred. This allows identd to work in
3729 * jail.
3730 */
3731 case PRIV_NETINET_GETCRED:
3732 return (0);
3733
3734 /*
3735 * Allow jailed root to set loginclass.
3736 */
3737 case PRIV_PROC_SETLOGINCLASS:
3738 return (0);
3739
3740 /*
3741 * Do not allow a process inside a jail to read the kernel
3742 * message buffer unless explicitly permitted.
3743 */
3744 case PRIV_MSGBUF:
3745 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3746 return (0);
3747 return (EPERM);
3748
3749 default:
3750 /*
3751 * In all remaining cases, deny the privilege request. This
3752 * includes almost all network privileges, many system
3753 * configuration privileges.
3754 */
3755 return (EPERM);
3756 }
3757 }
3758
3759 /*
3760 * Return the part of pr2's name that is relative to pr1, or the whole name
3761 * if it does not directly follow.
3762 */
3763
3764 char *
prison_name(struct prison * pr1,struct prison * pr2)3765 prison_name(struct prison *pr1, struct prison *pr2)
3766 {
3767 char *name;
3768
3769 /* Jails see themselves as "0" (if they see themselves at all). */
3770 if (pr1 == pr2)
3771 return "0";
3772 name = pr2->pr_name;
3773 if (prison_ischild(pr1, pr2)) {
3774 /*
3775 * pr1 isn't locked (and allprison_lock may not be either)
3776 * so its length can't be counted on. But the number of dots
3777 * can be counted on - and counted.
3778 */
3779 for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3780 name = strchr(name, '.') + 1;
3781 }
3782 return (name);
3783 }
3784
3785 /*
3786 * Return the part of pr2's path that is relative to pr1, or the whole path
3787 * if it does not directly follow.
3788 */
3789 static char *
prison_path(struct prison * pr1,struct prison * pr2)3790 prison_path(struct prison *pr1, struct prison *pr2)
3791 {
3792 char *path1, *path2;
3793 int len1;
3794
3795 path1 = pr1->pr_path;
3796 path2 = pr2->pr_path;
3797 if (!strcmp(path1, "/"))
3798 return (path2);
3799 len1 = strlen(path1);
3800 if (strncmp(path1, path2, len1))
3801 return (path2);
3802 if (path2[len1] == '\0')
3803 return "/";
3804 if (path2[len1] == '/')
3805 return (path2 + len1);
3806 return (path2);
3807 }
3808
3809 /*
3810 * Jail-related sysctls.
3811 */
3812 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3813 "Jails");
3814
3815 static int
sysctl_jail_list(SYSCTL_HANDLER_ARGS)3816 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3817 {
3818 struct xprison *xp;
3819 struct prison *pr, *cpr;
3820 #ifdef INET
3821 struct in_addr *ip4 = NULL;
3822 int ip4s = 0;
3823 #endif
3824 #ifdef INET6
3825 struct in6_addr *ip6 = NULL;
3826 int ip6s = 0;
3827 #endif
3828 int descend, error;
3829
3830 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3831 pr = req->td->td_ucred->cr_prison;
3832 error = 0;
3833 sx_slock(&allprison_lock);
3834 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3835 #if defined(INET) || defined(INET6)
3836 again:
3837 #endif
3838 mtx_lock(&cpr->pr_mtx);
3839 #ifdef INET
3840 if (cpr->pr_ip4s > 0) {
3841 if (ip4s < cpr->pr_ip4s) {
3842 ip4s = cpr->pr_ip4s;
3843 mtx_unlock(&cpr->pr_mtx);
3844 ip4 = realloc(ip4, ip4s *
3845 sizeof(struct in_addr), M_TEMP, M_WAITOK);
3846 goto again;
3847 }
3848 bcopy(cpr->pr_ip4, ip4,
3849 cpr->pr_ip4s * sizeof(struct in_addr));
3850 }
3851 #endif
3852 #ifdef INET6
3853 if (cpr->pr_ip6s > 0) {
3854 if (ip6s < cpr->pr_ip6s) {
3855 ip6s = cpr->pr_ip6s;
3856 mtx_unlock(&cpr->pr_mtx);
3857 ip6 = realloc(ip6, ip6s *
3858 sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3859 goto again;
3860 }
3861 bcopy(cpr->pr_ip6, ip6,
3862 cpr->pr_ip6s * sizeof(struct in6_addr));
3863 }
3864 #endif
3865 bzero(xp, sizeof(*xp));
3866 xp->pr_version = XPRISON_VERSION;
3867 xp->pr_id = cpr->pr_id;
3868 xp->pr_state = cpr->pr_state;
3869 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3870 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3871 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3872 #ifdef INET
3873 xp->pr_ip4s = cpr->pr_ip4s;
3874 #endif
3875 #ifdef INET6
3876 xp->pr_ip6s = cpr->pr_ip6s;
3877 #endif
3878 mtx_unlock(&cpr->pr_mtx);
3879 error = SYSCTL_OUT(req, xp, sizeof(*xp));
3880 if (error)
3881 break;
3882 #ifdef INET
3883 if (xp->pr_ip4s > 0) {
3884 error = SYSCTL_OUT(req, ip4,
3885 xp->pr_ip4s * sizeof(struct in_addr));
3886 if (error)
3887 break;
3888 }
3889 #endif
3890 #ifdef INET6
3891 if (xp->pr_ip6s > 0) {
3892 error = SYSCTL_OUT(req, ip6,
3893 xp->pr_ip6s * sizeof(struct in6_addr));
3894 if (error)
3895 break;
3896 }
3897 #endif
3898 }
3899 sx_sunlock(&allprison_lock);
3900 free(xp, M_TEMP);
3901 #ifdef INET
3902 free(ip4, M_TEMP);
3903 #endif
3904 #ifdef INET6
3905 free(ip6, M_TEMP);
3906 #endif
3907 return (error);
3908 }
3909
3910 SYSCTL_OID(_security_jail, OID_AUTO, list,
3911 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3912 sysctl_jail_list, "S", "List of active jails");
3913
3914 static int
sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)3915 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3916 {
3917 int error, injail;
3918
3919 injail = jailed(req->td->td_ucred);
3920 error = SYSCTL_OUT(req, &injail, sizeof(injail));
3921
3922 return (error);
3923 }
3924
3925 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3926 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3927 sysctl_jail_jailed, "I", "Process in jail?");
3928
3929 static int
sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)3930 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3931 {
3932 int error, havevnet;
3933 #ifdef VIMAGE
3934 struct ucred *cred = req->td->td_ucred;
3935
3936 havevnet = jailed(cred) && prison_owns_vnet(cred);
3937 #else
3938 havevnet = 0;
3939 #endif
3940 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3941
3942 return (error);
3943 }
3944
3945 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3946 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3947 sysctl_jail_vnet, "I", "Jail owns vnet?");
3948
3949 #if defined(INET) || defined(INET6)
3950 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3951 &jail_max_af_ips, 0,
3952 "Number of IP addresses a jail may have at most per address family (deprecated)");
3953 #endif
3954
3955 /*
3956 * Default parameters for jail(2) compatibility. For historical reasons,
3957 * the sysctl names have varying similarity to the parameter names. Prisons
3958 * just see their own parameters, and can't change them.
3959 */
3960 static int
sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)3961 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3962 {
3963 int error, i;
3964
3965 /* Get the current flag value, and convert it to a boolean. */
3966 if (req->td->td_ucred->cr_prison == &prison0) {
3967 mtx_lock(&prison0.pr_mtx);
3968 i = (jail_default_allow & arg2) != 0;
3969 mtx_unlock(&prison0.pr_mtx);
3970 } else
3971 i = prison_allow(req->td->td_ucred, arg2);
3972
3973 if (arg1 != NULL)
3974 i = !i;
3975 error = sysctl_handle_int(oidp, &i, 0, req);
3976 if (error || !req->newptr)
3977 return (error);
3978 i = i ? arg2 : 0;
3979 if (arg1 != NULL)
3980 i ^= arg2;
3981 /*
3982 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3983 * for writing.
3984 */
3985 mtx_lock(&prison0.pr_mtx);
3986 jail_default_allow = (jail_default_allow & ~arg2) | i;
3987 mtx_unlock(&prison0.pr_mtx);
3988 return (0);
3989 }
3990
3991 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3992 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3993 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3994 "Processes in jail can set their hostnames (deprecated)");
3995 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3996 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3997 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3998 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3999 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4000 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4001 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4002 "Processes in jail can use System V IPC primitives (deprecated)");
4003 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4004 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4005 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4006 "Prison root can create raw sockets (deprecated)");
4007 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4008 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4009 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4010 "Processes in jail can alter system file flags (deprecated)");
4011 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4012 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4013 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4014 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4015
4016 static int
sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)4017 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4018 {
4019 struct prison *pr;
4020 int level, error;
4021
4022 pr = req->td->td_ucred->cr_prison;
4023 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4024 error = sysctl_handle_int(oidp, &level, 0, req);
4025 if (error || !req->newptr)
4026 return (error);
4027 *(int *)arg1 = level;
4028 return (0);
4029 }
4030
4031 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4032 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4033 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4034 sysctl_jail_default_level, "I",
4035 "Processes in jail cannot see all mounted file systems (deprecated)");
4036
4037 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4038 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4039 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4040 sysctl_jail_default_level, "I",
4041 "Ruleset for the devfs filesystem in jail (deprecated)");
4042
4043 /*
4044 * Nodes to describe jail parameters. Maximum length of string parameters
4045 * is returned in the string itself, and the other parameters exist merely
4046 * to make themselves and their types known.
4047 */
4048 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
4049 "Jail parameters");
4050
4051 int
sysctl_jail_param(SYSCTL_HANDLER_ARGS)4052 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4053 {
4054 int i;
4055 long l;
4056 size_t s;
4057 char numbuf[12];
4058
4059 switch (oidp->oid_kind & CTLTYPE)
4060 {
4061 case CTLTYPE_LONG:
4062 case CTLTYPE_ULONG:
4063 l = 0;
4064 #ifdef SCTL_MASK32
4065 if (!(req->flags & SCTL_MASK32))
4066 #endif
4067 return (SYSCTL_OUT(req, &l, sizeof(l)));
4068 case CTLTYPE_INT:
4069 case CTLTYPE_UINT:
4070 i = 0;
4071 return (SYSCTL_OUT(req, &i, sizeof(i)));
4072 case CTLTYPE_STRING:
4073 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4074 return
4075 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4076 case CTLTYPE_STRUCT:
4077 s = (size_t)arg2;
4078 return (SYSCTL_OUT(req, &s, sizeof(s)));
4079 }
4080 return (0);
4081 }
4082
4083 /*
4084 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4085 * jail creation time but cannot be changed in an existing jail.
4086 */
4087 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4088 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4089 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4090 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4091 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4092 "I", "Jail secure level");
4093 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4094 "Jail value for kern.osreldate and uname -K");
4095 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4096 "Jail value for kern.osrelease and uname -r");
4097 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4098 "I", "Jail cannot see all mounted file systems");
4099 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4100 "I", "Ruleset for in-jail devfs mounts");
4101 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4102 "B", "Jail persistence");
4103 #ifdef VIMAGE
4104 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4105 "E,jailsys", "Virtual network stack");
4106 #endif
4107 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4108 "B", "Jail is in the process of shutting down");
4109
4110 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4111 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4112 "I", "Current number of child jails");
4113 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4114 "I", "Maximum number of child jails");
4115
4116 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4117 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4118 "Jail hostname");
4119 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4120 "Jail NIS domainname");
4121 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4122 "Jail host UUID");
4123 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4124 "LU", "Jail host ID");
4125
4126 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4127 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4128
4129 #ifdef INET
4130 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4131 "Jail IPv4 address virtualization");
4132 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4133 "S,in_addr,a", "Jail IPv4 addresses");
4134 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4135 "B", "Do (not) use IPv4 source address selection rather than the "
4136 "primary jail IPv4 address.");
4137 #endif
4138 #ifdef INET6
4139 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4140 "Jail IPv6 address virtualization");
4141 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4142 "S,in6_addr,a", "Jail IPv6 addresses");
4143 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4144 "B", "Do (not) use IPv6 source address selection rather than the "
4145 "primary jail IPv6 address.");
4146 #endif
4147
4148 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4149 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4150 "B", "Jail may set hostname");
4151 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4152 "B", "Jail may use SYSV IPC");
4153 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4154 "B", "Jail may create raw sockets");
4155 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4156 "B", "Jail may alter system file flags");
4157 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4158 "B", "Jail may set file quotas");
4159 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4160 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4161 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
4162 "B", "Jail may lock (unlock) physical pages in memory");
4163 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
4164 "B", "Jail may bind sockets to reserved ports");
4165 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
4166 "B", "Jail may read the kernel message buffer");
4167 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
4168 "B", "Unprivileged processes may use process debugging facilities");
4169 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW,
4170 "B", "Processes in jail with uid 0 have privilege");
4171
4172 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4173 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4174 "B", "Jail may mount/unmount jail-friendly file systems in general");
4175
4176 /*
4177 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return
4178 * its associated bit in the pr_allow bitmask, or zero if the parameter was
4179 * not created.
4180 */
4181 unsigned
prison_add_allow(const char * prefix,const char * name,const char * prefix_descr,const char * descr)4182 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
4183 const char *descr)
4184 {
4185 struct bool_flags *bf;
4186 struct sysctl_oid *parent;
4187 char *allow_name, *allow_noname, *allowed;
4188 #ifndef NO_SYSCTL_DESCR
4189 char *descr_deprecated;
4190 #endif
4191 u_int allow_flag;
4192
4193 if (prefix
4194 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
4195 < 0 ||
4196 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
4197 < 0
4198 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
4199 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
4200 free(allow_name, M_PRISON);
4201 return 0;
4202 }
4203
4204 /*
4205 * See if this parameter has already beed added, i.e. a module was
4206 * previously loaded/unloaded.
4207 */
4208 mtx_lock(&prison0.pr_mtx);
4209 for (bf = pr_flag_allow;
4210 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4211 atomic_load_int(&bf->flag) != 0;
4212 bf++) {
4213 if (strcmp(bf->name, allow_name) == 0) {
4214 allow_flag = bf->flag;
4215 goto no_add;
4216 }
4217 }
4218
4219 /*
4220 * Find a free bit in pr_allow_all, failing if there are none
4221 * (which shouldn't happen as long as we keep track of how many
4222 * potential dynamic flags exist).
4223 */
4224 for (allow_flag = 1;; allow_flag <<= 1) {
4225 if (allow_flag == 0)
4226 goto no_add;
4227 if ((pr_allow_all & allow_flag) == 0)
4228 break;
4229 }
4230
4231 /* Note the parameter in the next open slot in pr_flag_allow. */
4232 for (bf = pr_flag_allow; ; bf++) {
4233 if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
4234 /* This should never happen, but is not fatal. */
4235 allow_flag = 0;
4236 goto no_add;
4237 }
4238 if (atomic_load_int(&bf->flag) == 0)
4239 break;
4240 }
4241 bf->name = allow_name;
4242 bf->noname = allow_noname;
4243 pr_allow_all |= allow_flag;
4244 /*
4245 * prison0 always has permission for the new parameter.
4246 * Other jails must have it granted to them.
4247 */
4248 prison0.pr_allow |= allow_flag;
4249 /* The flag indicates a valid entry, so make sure it is set last. */
4250 atomic_store_rel_int(&bf->flag, allow_flag);
4251 mtx_unlock(&prison0.pr_mtx);
4252
4253 /*
4254 * Create sysctls for the paramter, and the back-compat global
4255 * permission.
4256 */
4257 parent = prefix
4258 ? SYSCTL_ADD_NODE(NULL,
4259 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
4260 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
4261 : &sysctl___security_jail_param_allow;
4262 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
4263 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4264 NULL, 0, sysctl_jail_param, "B", descr);
4265 if ((prefix
4266 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
4267 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
4268 #ifndef NO_SYSCTL_DESCR
4269 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
4270 descr);
4271 #endif
4272 (void)SYSCTL_ADD_PROC(NULL,
4273 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
4274 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
4275 sysctl_jail_default_allow, "I", descr_deprecated);
4276 #ifndef NO_SYSCTL_DESCR
4277 free(descr_deprecated, M_TEMP);
4278 #endif
4279 free(allowed, M_TEMP);
4280 }
4281 return allow_flag;
4282
4283 no_add:
4284 mtx_unlock(&prison0.pr_mtx);
4285 free(allow_name, M_PRISON);
4286 free(allow_noname, M_PRISON);
4287 return allow_flag;
4288 }
4289
4290 /*
4291 * The VFS system will register jail-aware filesystems here. They each get
4292 * a parameter allow.mount.xxxfs and a flag to check when a jailed user
4293 * attempts to mount.
4294 */
4295 void
prison_add_vfs(struct vfsconf * vfsp)4296 prison_add_vfs(struct vfsconf *vfsp)
4297 {
4298 #ifdef NO_SYSCTL_DESCR
4299
4300 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4301 NULL, NULL);
4302 #else
4303 char *descr;
4304
4305 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
4306 vfsp->vfc_name);
4307 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
4308 NULL, descr);
4309 free(descr, M_TEMP);
4310 #endif
4311 }
4312
4313 #ifdef RACCT
4314 void
prison_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)4315 prison_racct_foreach(void (*callback)(struct racct *racct,
4316 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4317 void *arg2, void *arg3)
4318 {
4319 struct prison_racct *prr;
4320
4321 ASSERT_RACCT_ENABLED();
4322
4323 sx_slock(&allprison_lock);
4324 if (pre != NULL)
4325 (pre)();
4326 LIST_FOREACH(prr, &allprison_racct, prr_next)
4327 (callback)(prr->prr_racct, arg2, arg3);
4328 if (post != NULL)
4329 (post)();
4330 sx_sunlock(&allprison_lock);
4331 }
4332
4333 static struct prison_racct *
prison_racct_find_locked(const char * name)4334 prison_racct_find_locked(const char *name)
4335 {
4336 struct prison_racct *prr;
4337
4338 ASSERT_RACCT_ENABLED();
4339 sx_assert(&allprison_lock, SA_XLOCKED);
4340
4341 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4342 return (NULL);
4343
4344 LIST_FOREACH(prr, &allprison_racct, prr_next) {
4345 if (strcmp(name, prr->prr_name) != 0)
4346 continue;
4347
4348 /* Found prison_racct with a matching name? */
4349 prison_racct_hold(prr);
4350 return (prr);
4351 }
4352
4353 /* Add new prison_racct. */
4354 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4355 racct_create(&prr->prr_racct);
4356
4357 strcpy(prr->prr_name, name);
4358 refcount_init(&prr->prr_refcount, 1);
4359 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4360
4361 return (prr);
4362 }
4363
4364 struct prison_racct *
prison_racct_find(const char * name)4365 prison_racct_find(const char *name)
4366 {
4367 struct prison_racct *prr;
4368
4369 ASSERT_RACCT_ENABLED();
4370
4371 sx_xlock(&allprison_lock);
4372 prr = prison_racct_find_locked(name);
4373 sx_xunlock(&allprison_lock);
4374 return (prr);
4375 }
4376
4377 void
prison_racct_hold(struct prison_racct * prr)4378 prison_racct_hold(struct prison_racct *prr)
4379 {
4380
4381 ASSERT_RACCT_ENABLED();
4382
4383 refcount_acquire(&prr->prr_refcount);
4384 }
4385
4386 static void
prison_racct_free_locked(struct prison_racct * prr)4387 prison_racct_free_locked(struct prison_racct *prr)
4388 {
4389
4390 ASSERT_RACCT_ENABLED();
4391 sx_assert(&allprison_lock, SA_XLOCKED);
4392
4393 if (refcount_release(&prr->prr_refcount)) {
4394 racct_destroy(&prr->prr_racct);
4395 LIST_REMOVE(prr, prr_next);
4396 free(prr, M_PRISON_RACCT);
4397 }
4398 }
4399
4400 void
prison_racct_free(struct prison_racct * prr)4401 prison_racct_free(struct prison_racct *prr)
4402 {
4403
4404 ASSERT_RACCT_ENABLED();
4405 sx_assert(&allprison_lock, SA_UNLOCKED);
4406
4407 if (refcount_release_if_not_last(&prr->prr_refcount))
4408 return;
4409
4410 sx_xlock(&allprison_lock);
4411 prison_racct_free_locked(prr);
4412 sx_xunlock(&allprison_lock);
4413 }
4414
4415 static void
prison_racct_attach(struct prison * pr)4416 prison_racct_attach(struct prison *pr)
4417 {
4418 struct prison_racct *prr;
4419
4420 ASSERT_RACCT_ENABLED();
4421 sx_assert(&allprison_lock, SA_XLOCKED);
4422
4423 prr = prison_racct_find_locked(pr->pr_name);
4424 KASSERT(prr != NULL, ("cannot find prison_racct"));
4425
4426 pr->pr_prison_racct = prr;
4427 }
4428
4429 /*
4430 * Handle jail renaming. From the racct point of view, renaming means
4431 * moving from one prison_racct to another.
4432 */
4433 static void
prison_racct_modify(struct prison * pr)4434 prison_racct_modify(struct prison *pr)
4435 {
4436 #ifdef RCTL
4437 struct proc *p;
4438 struct ucred *cred;
4439 #endif
4440 struct prison_racct *oldprr;
4441
4442 ASSERT_RACCT_ENABLED();
4443
4444 sx_slock(&allproc_lock);
4445 sx_xlock(&allprison_lock);
4446
4447 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4448 sx_xunlock(&allprison_lock);
4449 sx_sunlock(&allproc_lock);
4450 return;
4451 }
4452
4453 oldprr = pr->pr_prison_racct;
4454 pr->pr_prison_racct = NULL;
4455
4456 prison_racct_attach(pr);
4457
4458 /*
4459 * Move resource utilisation records.
4460 */
4461 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4462
4463 #ifdef RCTL
4464 /*
4465 * Force rctl to reattach rules to processes.
4466 */
4467 FOREACH_PROC_IN_SYSTEM(p) {
4468 PROC_LOCK(p);
4469 cred = crhold(p->p_ucred);
4470 PROC_UNLOCK(p);
4471 rctl_proc_ucred_changed(p, cred);
4472 crfree(cred);
4473 }
4474 #endif
4475
4476 sx_sunlock(&allproc_lock);
4477 prison_racct_free_locked(oldprr);
4478 sx_xunlock(&allprison_lock);
4479 }
4480
4481 static void
prison_racct_detach(struct prison * pr)4482 prison_racct_detach(struct prison *pr)
4483 {
4484
4485 ASSERT_RACCT_ENABLED();
4486 sx_assert(&allprison_lock, SA_UNLOCKED);
4487
4488 if (pr->pr_prison_racct == NULL)
4489 return;
4490 prison_racct_free(pr->pr_prison_racct);
4491 pr->pr_prison_racct = NULL;
4492 }
4493 #endif /* RACCT */
4494
4495 #ifdef DDB
4496
4497 static void
db_show_prison(struct prison * pr)4498 db_show_prison(struct prison *pr)
4499 {
4500 struct bool_flags *bf;
4501 struct jailsys_flags *jsf;
4502 #if defined(INET) || defined(INET6)
4503 int ii;
4504 #endif
4505 unsigned f;
4506 #ifdef INET
4507 char ip4buf[INET_ADDRSTRLEN];
4508 #endif
4509 #ifdef INET6
4510 char ip6buf[INET6_ADDRSTRLEN];
4511 #endif
4512
4513 db_printf("prison %p:\n", pr);
4514 db_printf(" jid = %d\n", pr->pr_id);
4515 db_printf(" name = %s\n", pr->pr_name);
4516 db_printf(" parent = %p\n", pr->pr_parent);
4517 db_printf(" ref = %d\n", pr->pr_ref);
4518 db_printf(" uref = %d\n", pr->pr_uref);
4519 db_printf(" state = %s\n",
4520 pr->pr_state == PRISON_STATE_ALIVE ? "alive" :
4521 pr->pr_state == PRISON_STATE_DYING ? "dying" :
4522 "invalid");
4523 db_printf(" path = %s\n", pr->pr_path);
4524 db_printf(" cpuset = %d\n", pr->pr_cpuset
4525 ? pr->pr_cpuset->cs_id : -1);
4526 #ifdef VIMAGE
4527 db_printf(" vnet = %p\n", pr->pr_vnet);
4528 #endif
4529 db_printf(" root = %p\n", pr->pr_root);
4530 db_printf(" securelevel = %d\n", pr->pr_securelevel);
4531 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum);
4532 db_printf(" children.max = %d\n", pr->pr_childmax);
4533 db_printf(" children.cur = %d\n", pr->pr_childcount);
4534 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children));
4535 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling));
4536 db_printf(" flags = 0x%x", pr->pr_flags);
4537 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4538 if (pr->pr_flags & bf->flag)
4539 db_printf(" %s", bf->name);
4540 for (jsf = pr_flag_jailsys;
4541 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4542 jsf++) {
4543 f = pr->pr_flags & (jsf->disable | jsf->new);
4544 db_printf(" %-16s= %s\n", jsf->name,
4545 (f != 0 && f == jsf->disable) ? "disable"
4546 : (f == jsf->new) ? "new"
4547 : "inherit");
4548 }
4549 db_printf(" allow = 0x%x", pr->pr_allow);
4550 for (bf = pr_flag_allow;
4551 bf < pr_flag_allow + nitems(pr_flag_allow) &&
4552 atomic_load_int(&bf->flag) != 0;
4553 bf++)
4554 if (pr->pr_allow & bf->flag)
4555 db_printf(" %s", bf->name);
4556 db_printf("\n");
4557 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs);
4558 db_printf(" host.hostname = %s\n", pr->pr_hostname);
4559 db_printf(" host.domainname = %s\n", pr->pr_domainname);
4560 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid);
4561 db_printf(" host.hostid = %lu\n", pr->pr_hostid);
4562 #ifdef INET
4563 db_printf(" ip4s = %d\n", pr->pr_ip4s);
4564 for (ii = 0; ii < pr->pr_ip4s; ii++)
4565 db_printf(" %s %s\n",
4566 ii == 0 ? "ip4.addr =" : " ",
4567 inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4568 #endif
4569 #ifdef INET6
4570 db_printf(" ip6s = %d\n", pr->pr_ip6s);
4571 for (ii = 0; ii < pr->pr_ip6s; ii++)
4572 db_printf(" %s %s\n",
4573 ii == 0 ? "ip6.addr =" : " ",
4574 ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4575 #endif
4576 }
4577
DB_SHOW_COMMAND(prison,db_show_prison_command)4578 DB_SHOW_COMMAND(prison, db_show_prison_command)
4579 {
4580 struct prison *pr;
4581
4582 if (!have_addr) {
4583 /*
4584 * Show all prisons in the list, and prison0 which is not
4585 * listed.
4586 */
4587 db_show_prison(&prison0);
4588 if (!db_pager_quit) {
4589 TAILQ_FOREACH(pr, &allprison, pr_list) {
4590 db_show_prison(pr);
4591 if (db_pager_quit)
4592 break;
4593 }
4594 }
4595 return;
4596 }
4597
4598 if (addr == 0)
4599 pr = &prison0;
4600 else {
4601 /* Look for a prison with the ID and with references. */
4602 TAILQ_FOREACH(pr, &allprison, pr_list)
4603 if (pr->pr_id == addr && pr->pr_ref > 0)
4604 break;
4605 if (pr == NULL)
4606 /* Look again, without requiring a reference. */
4607 TAILQ_FOREACH(pr, &allprison, pr_list)
4608 if (pr->pr_id == addr)
4609 break;
4610 if (pr == NULL)
4611 /* Assume address points to a valid prison. */
4612 pr = (struct prison *)addr;
4613 }
4614 db_show_prison(pr);
4615 }
4616
4617 #endif /* DDB */
4618