1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1999 Poul-Henning Kamp.
5 * Copyright (c) 2009 James Gritton.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#ifndef _SYS_JAIL_H_
31#define _SYS_JAIL_H_
32
33#ifdef _KERNEL
34struct jail_v0 {
35 u_int32_t version;
36 char *path;
37 char *hostname;
38 u_int32_t ip_number;
39};
40#endif
41
42struct jail {
43 uint32_t version;
44 char *path;
45 char *hostname;
46 char *jailname;
47 uint32_t ip4s;
48 uint32_t ip6s;
49 struct in_addr *ip4;
50 struct in6_addr *ip6;
51};
52#define JAIL_API_VERSION 2
53
54/*
55 * For all xprison structs, always keep the pr_version an int and
56 * the first variable so userspace can easily distinguish them.
57 */
58#ifndef _KERNEL
59struct xprison_v1 {
60 int pr_version;
61 int pr_id;
62 char pr_path[MAXPATHLEN];
63 char pr_host[MAXHOSTNAMELEN];
64 u_int32_t pr_ip;
65};
66#endif
67
68struct xprison {
69 int pr_version;
70 int pr_id;
71 int pr_state;
72 cpusetid_t pr_cpusetid;
73 char pr_path[MAXPATHLEN];
74 char pr_host[MAXHOSTNAMELEN];
75 char pr_name[MAXHOSTNAMELEN];
76 uint32_t pr_ip4s;
77 uint32_t pr_ip6s;
78#if 0
79 /*
80 * sizeof(xprison) will be malloced + size needed for all
81 * IPv4 and IPv6 addesses. Offsets are based numbers of addresses.
82 */
83 struct in_addr pr_ip4[];
84 struct in6_addr pr_ip6[];
85#endif
86};
87#define XPRISON_VERSION 3
88
89enum prison_state {
90 PRISON_STATE_INVALID = 0, /* New prison, not ready to be seen */
91 PRISON_STATE_ALIVE, /* Current prison, visible to all */
92 PRISON_STATE_DYING /* Removed but holding resources, */
93}; /* optionally visible. */
94
95/*
96 * Flags for jail_set and jail_get.
97 */
98#define JAIL_CREATE 0x01 /* Create jail if it doesn't exist */
99#define JAIL_UPDATE 0x02 /* Update parameters of existing jail */
100#define JAIL_ATTACH 0x04 /* Attach to jail upon creation */
101#define JAIL_DYING 0x08 /* Allow getting a dying jail */
102#define JAIL_USE_DESC 0x10 /* Get/set jail in descriptor */
103#define JAIL_AT_DESC 0x20 /* Find/add jail under descriptor */
104#define JAIL_GET_DESC 0x40 /* Return a new jail descriptor */
105#define JAIL_OWN_DESC 0x80 /* Return a new owning jail descriptor */
106#define JAIL_SET_MASK 0xff /* JAIL_DYING is deprecated/ignored here */
107#define JAIL_GET_MASK 0xf8
108
109#define JAIL_SYS_DISABLE 0
110#define JAIL_SYS_NEW 1
111#define JAIL_SYS_INHERIT 2
112
113#ifndef _KERNEL
114
115struct iovec;
116
117__BEGIN_DECLS
118int jail(struct jail *);
119int jail_set(struct iovec *, unsigned int, int);
120int jail_get(struct iovec *, unsigned int, int);
121int jail_attach(int);
122int jail_attach_jd(int);
123int jail_remove(int);
124int jail_remove_jd(int);
125__END_DECLS
126
127#else /* _KERNEL */
128
129#include <sys/queue.h>
130#include <sys/sysctl.h>
131#include <sys/lock.h>
132#include <sys/mutex.h>
133#include <sys/_task.h>
134
135#define JAIL_MAX 999999
136
137#ifdef MALLOC_DECLARE
138MALLOC_DECLARE(M_PRISON);
139#endif
140#endif /* _KERNEL */
141
142#if defined(_KERNEL) || defined(_WANT_PRISON)
143
144#include <sys/osd.h>
145
146#define HOSTUUIDLEN 64
147#define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000"
148#define OSRELEASELEN 32
149
150#define JAIL_META_PRIVATE "meta"
151#define JAIL_META_SHARED "env"
152
153struct jaildesc;
154struct knlist;
155struct racct;
156struct prison_racct;
157
158typedef enum {
159 PR_INET = 0,
160 PR_INET6 = 1,
161 PR_FAMILY_MAX = 2,
162} pr_family_t;
163
164/*
165 * This structure describes a prison. It is pointed to by all struct
166 * ucreds's of the inmates. pr_ref keeps track of them and is used to
167 * delete the structure when the last inmate is dead.
168 *
169 * Lock key:
170 * (a) allprison_lock
171 * (A) allproc_lock
172 * (c) set only during creation before the structure is shared, no mutex
173 * required to read
174 * (m) locked by pr_mtx
175 * (p) locked by pr_mtx, and also at least shared allprison_lock required
176 * to update
177 * (q) locked by both pr_mtx and allprison_lock
178 * (r) atomic via refcount(9), pr_mtx and allprison_lock required to
179 * decrement to zero
180 * (n) read access granted with the network epoch
181 */
182struct prison {
183 TAILQ_ENTRY(prison) pr_list; /* (a) all prisons */
184 int pr_id; /* (c) prison id */
185 volatile u_int pr_ref; /* (r) refcount */
186 volatile u_int pr_uref; /* (r) user (alive) refcount */
187 unsigned pr_flags; /* (p) PR_* flags */
188 LIST_HEAD(, prison) pr_children; /* (a) list of child jails */
189 LIST_HEAD(, proc) pr_proclist; /* (A) list of jailed processes */
190 LIST_ENTRY(prison) pr_sibling; /* (a) next in parent's list */
191 struct prison *pr_parent; /* (c) containing jail */
192 struct mtx pr_mtx;
193 struct task pr_task; /* (c) destroy task */
194 struct osd pr_osd; /* (p) additional data */
195 struct cpuset *pr_cpuset; /* (p) cpuset */
196 struct vnet *pr_vnet; /* (c) network stack */
197 struct vnode *pr_root; /* (c) vnode to rdir */
198 struct prison_ip *pr_addrs[PR_FAMILY_MAX]; /* (p,n) IPs of jail */
199 struct prison_racct *pr_prison_racct; /* (c) racct jail proxy */
200 struct knlist *pr_klist; /* (m) attached knotes */
201 LIST_HEAD(, jaildesc) pr_descs; /* (a) attached descriptors */
202 void *pr_sparep;
203 int pr_childcount; /* (a) number of child jails */
204 int pr_childmax; /* (p) maximum child jails */
205 unsigned pr_allow; /* (p) PR_ALLOW_* flags */
206 int pr_securelevel; /* (p) securelevel */
207 int pr_enforce_statfs; /* (p) statfs permission */
208 int pr_devfs_rsnum; /* (p) devfs ruleset */
209 enum prison_state pr_state; /* (q) state in life cycle */
210 volatile int pr_exportcnt; /* (r) count of mount exports */
211 int pr_spare;
212 int pr_osreldate; /* (c) kern.osreldate value */
213 unsigned long pr_hostid; /* (p) jail hostid */
214 char pr_name[MAXHOSTNAMELEN]; /* (p) admin jail name */
215 char pr_path[MAXPATHLEN]; /* (c) chroot path */
216 char pr_hostname[MAXHOSTNAMELEN]; /* (p) jail hostname */
217 char pr_domainname[MAXHOSTNAMELEN]; /* (p) jail domainname */
218 char pr_hostuuid[HOSTUUIDLEN]; /* (p) jail hostuuid */
219 char pr_osrelease[OSRELEASELEN]; /* (c) kern.osrelease value */
220};
221
222struct prison_racct {
223 LIST_ENTRY(prison_racct) prr_next;
224 char prr_name[MAXHOSTNAMELEN];
225 u_int prr_refcount;
226 struct racct *prr_racct;
227};
228#endif /* _KERNEL || _WANT_PRISON */
229
230#ifdef _KERNEL
231/* Flag bits set via options */
232#define PR_PERSIST 0x00000001 /* Can exist without processes */
233#define PR_HOST 0x00000002 /* Virtualize hostname et al */
234#define PR_IP4_USER 0x00000004 /* Restrict IPv4 addresses */
235#define PR_IP6_USER 0x00000008 /* Restrict IPv6 addresses */
236#define PR_VNET 0x00000010 /* Virtual network stack */
237#define PR_IP4_SADDRSEL 0x00000080 /* Do IPv4 src addr sel. or use the */
238 /* primary jail address. */
239#define PR_IP6_SADDRSEL 0x00000100 /* Do IPv6 src addr sel. or use the */
240 /* primary jail address. */
241
242/* Internal flag bits */
243#define PR_REMOVE 0x01000000 /* In process of being removed */
244#define PR_IP4 0x02000000 /* IPv4 restricted or disabled */
245 /* by this jail or an ancestor */
246#define PR_IP6 0x04000000 /* IPv6 restricted or disabled */
247 /* by this jail or an ancestor */
248#define PR_COMPLETE_PROC 0x08000000 /* prison_complete called from */
249 /* prison_proc_free, releases uref */
250
251/*
252 * Flags for pr_allow
253 * Bits not noted here may be used for dynamic allow.mount.xxxfs.
254 */
255#define PR_ALLOW_SET_HOSTNAME 0x00000001
256#define PR_ALLOW_SYSVIPC 0x00000002
257#define PR_ALLOW_RAW_SOCKETS 0x00000004
258#define PR_ALLOW_CHFLAGS 0x00000008
259#define PR_ALLOW_MOUNT 0x00000010
260#define PR_ALLOW_QUOTAS 0x00000020
261#define PR_ALLOW_SOCKET_AF 0x00000040
262#define PR_ALLOW_MLOCK 0x00000080
263#define PR_ALLOW_READ_MSGBUF 0x00000100
264#define PR_ALLOW_UNPRIV_DEBUG 0x00000200
265#define PR_ALLOW_SUSER 0x00000400
266#define PR_ALLOW_RESERVED_PORTS 0x00008000
267#define PR_ALLOW_KMEM_ACCESS 0x00010000 /* reserved, not used yet */
268#define PR_ALLOW_NFSD 0x00020000
269#define PR_ALLOW_EXTATTR 0x00040000
270#define PR_ALLOW_ADJTIME 0x00080000
271#define PR_ALLOW_SETTIME 0x00100000
272#define PR_ALLOW_ROUTING 0x00200000
273#define PR_ALLOW_UNPRIV_PARENT_TAMPER 0x00400000
274#define PR_ALLOW_SETAUDIT 0x00800000
275
276/*
277 * PR_ALLOW_PRISON0 are the allow flags that we apply by default to prison0,
278 * while PR_ALLOW_ALL_STATIC are all of the allow bits that we have allocated at
279 * build time. PR_ALLOW_ALL_STATIC should contain any bit above that we expect
280 * to be used on the system, while PR_ALLOW_PRISON0 will be some subset of that.
281 */
282#define PR_ALLOW_ALL_STATIC 0x00ff87ff
283#define PR_ALLOW_PRISON0 \
284 (PR_ALLOW_ALL_STATIC & ~(PR_ALLOW_UNPRIV_PARENT_TAMPER))
285
286/*
287 * PR_ALLOW_DIFFERENCES determines which flags are able to be
288 * different between the parent and child jail upon creation.
289 */
290#define PR_ALLOW_DIFFERENCES \
291 (PR_ALLOW_UNPRIV_DEBUG | PR_ALLOW_UNPRIV_PARENT_TAMPER)
292
293/*
294 * OSD methods
295 */
296#define PR_METHOD_CREATE 0
297#define PR_METHOD_GET 1
298#define PR_METHOD_SET 2
299#define PR_METHOD_CHECK 3
300#define PR_METHOD_ATTACH 4
301#define PR_METHOD_REMOVE 5
302#define PR_MAXMETHOD 6
303
304/*
305 * Lock/unlock a prison.
306 * XXX These exist not so much for general convenience, but to be useable in
307 * the FOREACH_PRISON_DESCENDANT_LOCKED macro which can't handle them in
308 * non-function form as currently defined.
309 */
310static __inline void
311prison_lock(struct prison *pr)
312{
313
314 mtx_lock(&pr->pr_mtx);
315}
316
317static __inline void
318prison_unlock(struct prison *pr)
319{
320
321 mtx_unlock(&pr->pr_mtx);
322}
323
324/* Traverse a prison's immediate children. */
325#define FOREACH_PRISON_CHILD(ppr, cpr) \
326 LIST_FOREACH(cpr, &(ppr)->pr_children, pr_sibling)
327
328/*
329 * Preorder traversal of all of a prison's descendants.
330 * This ugly loop allows the macro to be followed by a single block
331 * as expected in a looping primitive.
332 */
333#define FOREACH_PRISON_DESCENDANT(ppr, cpr, descend) \
334 for ((cpr) = (ppr), (descend) = 1; \
335 ((cpr) = (((descend) && !LIST_EMPTY(&(cpr)->pr_children)) \
336 ? LIST_FIRST(&(cpr)->pr_children) \
337 : ((cpr) == (ppr) \
338 ? NULL \
339 : (((descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \
340 ? LIST_NEXT(cpr, pr_sibling) \
341 : (cpr)->pr_parent))));) \
342 if (!(descend)) \
343 ; \
344 else
345
346/*
347 * As above, but lock descendants on the way down and unlock on the way up.
348 */
349#define FOREACH_PRISON_DESCENDANT_LOCKED(ppr, cpr, descend) \
350 for ((cpr) = (ppr), (descend) = 1; \
351 ((cpr) = (((descend) && !LIST_EMPTY(&(cpr)->pr_children)) \
352 ? LIST_FIRST(&(cpr)->pr_children) \
353 : ((cpr) == (ppr) \
354 ? NULL \
355 : ((prison_unlock(cpr), \
356 (descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \
357 ? LIST_NEXT(cpr, pr_sibling) \
358 : (cpr)->pr_parent))));) \
359 if ((descend) ? (prison_lock(cpr), 0) : 1) \
360 ; \
361 else
362
363/*
364 * As above, but also keep track of the level descended to.
365 */
366#define FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(ppr, cpr, descend, level)\
367 for ((cpr) = (ppr), (descend) = 1, (level) = 0; \
368 ((cpr) = (((descend) && !LIST_EMPTY(&(cpr)->pr_children)) \
369 ? (level++, LIST_FIRST(&(cpr)->pr_children)) \
370 : ((cpr) == (ppr) \
371 ? NULL \
372 : ((prison_unlock(cpr), \
373 (descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \
374 ? LIST_NEXT(cpr, pr_sibling) \
375 : (level--, (cpr)->pr_parent)))));) \
376 if ((descend) ? (prison_lock(cpr), 0) : 1) \
377 ; \
378 else
379
380/*
381 * Traverse a prison's descendants, visiting both preorder and postorder.
382 */
383#define FOREACH_PRISON_DESCENDANT_PRE_POST(ppr, cpr, descend) \
384 for ((cpr) = (ppr), (descend) = 1; \
385 ((cpr) = (descend) \
386 ? ((descend) = !LIST_EMPTY(&(cpr)->pr_children)) \
387 ? LIST_FIRST(&(cpr)->pr_children) \
388 : (cpr) \
389 : ((descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \
390 ? LIST_NEXT(cpr, pr_sibling) \
391 : cpr->pr_parent) != (ppr);)
392
393/*
394 * Attributes of the physical system, and the root of the jail tree.
395 */
396extern struct prison prison0;
397
398TAILQ_HEAD(prisonlist, prison);
399extern struct prisonlist allprison;
400extern struct sx allprison_lock;
401
402/*
403 * Sysctls to describe jail parameters.
404 */
405SYSCTL_DECL(_security_jail);
406SYSCTL_DECL(_security_jail_param);
407
408#define SYSCTL_JAIL_PARAM_DECL(name) \
409 SYSCTL_DECL(_security_jail_param_##name)
410#define SYSCTL_JAIL_PARAM(module, param, type, fmt, descr) \
411 SYSCTL_PROC(_security_jail_param ## module, OID_AUTO, param, \
412 (type) | CTLFLAG_MPSAFE, NULL, 0, sysctl_jail_param, fmt, descr)
413#define SYSCTL_JAIL_PARAM_STRING(module, param, access, len, descr) \
414 SYSCTL_PROC(_security_jail_param ## module, OID_AUTO, param, \
415 CTLTYPE_STRING | CTLFLAG_MPSAFE | (access), NULL, len, \
416 sysctl_jail_param, "A", descr)
417#define SYSCTL_JAIL_PARAM_STRUCT(module, param, access, len, fmt, descr) \
418 SYSCTL_PROC(_security_jail_param ## module, OID_AUTO, param, \
419 CTLTYPE_STRUCT | CTLFLAG_MPSAFE | (access), NULL, len, \
420 sysctl_jail_param, fmt, descr)
421#define SYSCTL_JAIL_PARAM_NODE(module, descr) \
422 SYSCTL_NODE(_security_jail_param, OID_AUTO, module, CTLFLAG_MPSAFE, \
423 0, descr)
424#define SYSCTL_JAIL_PARAM_SUBNODE(parent, module, descr) \
425 SYSCTL_NODE(_security_jail_param_##parent, OID_AUTO, module, \
426 CTLFLAG_MPSAFE, 0, descr)
427#define SYSCTL_JAIL_PARAM_SYS_NODE(module, access, descr) \
428 SYSCTL_JAIL_PARAM_NODE(module, descr); \
429 SYSCTL_JAIL_PARAM(_##module, , CTLTYPE_INT | (access), "E,jailsys", \
430 descr)
431#define SYSCTL_JAIL_PARAM_SYS_SUBNODE(parent, module, access, descr) \
432 SYSCTL_JAIL_PARAM_SUBNODE(parent, module, descr); \
433 SYSCTL_JAIL_PARAM(_##parent##_##module, , CTLTYPE_INT | (access), \
434 "E,jailsys", descr)
435
436/*
437 * Kernel support functions for jail().
438 */
439struct knote;
440struct mount;
441struct sockaddr;
442struct statfs;
443struct ucred;
444struct vfsconf;
445
446/*
447 * Return 1 if the passed credential is in a jail, otherwise 0.
448 */
449#define jailed(cred) (cred->cr_prison != &prison0)
450
451bool jailed_without_vnet(struct ucred *);
452void getcredhostname(struct ucred *, char *, size_t);
453void getcreddomainname(struct ucred *, char *, size_t);
454void getcredhostuuid(struct ucred *, char *, size_t);
455void getcredhostid(struct ucred *, unsigned long *);
456void getjailname(struct ucred *cred, char *name, size_t len);
457void prison0_init(void);
458bool prison_allow(struct ucred *, unsigned);
459int prison_check(struct ucred *cred1, struct ucred *cred2);
460bool prison_check_nfsd(struct ucred *cred);
461bool prison_owns_vnet(struct prison *pr);
462int prison_canseemount(struct ucred *cred, struct mount *mp);
463void prison_enforce_statfs(struct ucred *cred, struct mount *mp,
464 struct statfs *sp);
465struct prison *prison_find(int prid);
466struct prison *prison_find_child(struct prison *, int);
467struct prison *prison_find_name(struct prison *, const char *);
468bool prison_flag(struct ucred *, unsigned);
469void prison_free(struct prison *pr);
470void prison_free_locked(struct prison *pr);
471void prison_hold(struct prison *pr);
472void prison_hold_locked(struct prison *pr);
473void prison_proc_hold(struct prison *);
474void prison_proc_free(struct prison *);
475void prison_proc_link(struct prison *, struct proc *);
476void prison_proc_unlink(struct prison *, struct proc *);
477void prison_proc_iterate(struct prison *, void (*)(struct proc *, void *), void *);
478void prison_remove(struct prison *);
479void prison_set_allow(struct ucred *cred, unsigned flag, int enable);
480bool prison_ischild(struct prison *, struct prison *);
481bool prison_isalive(const struct prison *);
482bool prison_isvalid(struct prison *);
483#if defined(INET) || defined(INET6)
484int prison_ip_check(const struct prison *, const pr_family_t, const void *);
485const void *prison_ip_get0(const struct prison *, const pr_family_t);
486u_int prison_ip_cnt(const struct prison *, const pr_family_t);
487#endif
488#ifdef INET
489bool prison_equal_ip4(struct prison *, struct prison *);
490int prison_get_ip4(struct ucred *cred, struct in_addr *ia);
491int prison_local_ip4(struct ucred *cred, struct in_addr *ia);
492int prison_remote_ip4(struct ucred *cred, struct in_addr *ia);
493int prison_check_ip4(const struct ucred *, const struct in_addr *);
494int prison_check_ip4_locked(const struct prison *, const struct in_addr *);
495bool prison_saddrsel_ip4(struct ucred *, struct in_addr *);
496int prison_qcmp_v4(const void *, const void *);
497bool prison_valid_v4(const void *);
498#endif
499#ifdef INET6
500bool prison_equal_ip6(struct prison *, struct prison *);
501int prison_get_ip6(struct ucred *, struct in6_addr *);
502int prison_local_ip6(struct ucred *, struct in6_addr *, int);
503int prison_remote_ip6(struct ucred *, struct in6_addr *);
504int prison_check_ip6(const struct ucred *, const struct in6_addr *);
505int prison_check_ip6_locked(const struct prison *, const struct in6_addr *);
506bool prison_saddrsel_ip6(struct ucred *, struct in6_addr *);
507int prison_qcmp_v6(const void *, const void *);
508bool prison_valid_v6(const void *);
509#endif
510int prison_check_af(struct ucred *cred, int af);
511int prison_if(struct ucred *cred, const struct sockaddr *sa);
512char *prison_name(struct prison *, struct prison *);
513int prison_priv_check(struct ucred *cred, int priv);
514int sysctl_jail_param(SYSCTL_HANDLER_ARGS);
515unsigned prison_add_allow(const char *prefix, const char *name,
516 const char *prefix_descr, const char *descr);
517void prison_add_vfs(struct vfsconf *vfsp);
518void prison_racct_foreach(void (*callback)(struct racct *racct,
519 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
520 void *arg2, void *arg3);
521struct prison_racct *prison_racct_find(const char *name);
522void prison_racct_hold(struct prison_racct *prr);
523void prison_racct_free(struct prison_racct *prr);
524
525#endif /* _KERNEL */
526#endif /* !_SYS_JAIL_H_ */