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 * $FreeBSD$ 30 */ 31 32 #ifndef _SYS_JAIL_H_ 33 #define _SYS_JAIL_H_ 34 35 #ifdef _KERNEL 36 struct jail_v0 { 37 u_int32_t version; 38 char *path; 39 char *hostname; 40 u_int32_t ip_number; 41 }; 42 #endif 43 44 struct jail { 45 uint32_t version; 46 char *path; 47 char *hostname; 48 char *jailname; 49 uint32_t ip4s; 50 uint32_t ip6s; 51 struct in_addr *ip4; 52 struct in6_addr *ip6; 53 }; 54 #define JAIL_API_VERSION 2 55 56 /* 57 * For all xprison structs, always keep the pr_version an int and 58 * the first variable so userspace can easily distinguish them. 59 */ 60 #ifndef _KERNEL 61 struct xprison_v1 { 62 int pr_version; 63 int pr_id; 64 char pr_path[MAXPATHLEN]; 65 char pr_host[MAXHOSTNAMELEN]; 66 u_int32_t pr_ip; 67 }; 68 #endif 69 70 struct xprison { 71 int pr_version; 72 int pr_id; 73 int pr_state; 74 cpusetid_t pr_cpusetid; 75 char pr_path[MAXPATHLEN]; 76 char pr_host[MAXHOSTNAMELEN]; 77 char pr_name[MAXHOSTNAMELEN]; 78 uint32_t pr_ip4s; 79 uint32_t pr_ip6s; 80 #if 0 81 /* 82 * sizeof(xprison) will be malloced + size needed for all 83 * IPv4 and IPv6 addesses. Offsets are based numbers of addresses. 84 */ 85 struct in_addr pr_ip4[]; 86 struct in6_addr pr_ip6[]; 87 #endif 88 }; 89 #define XPRISON_VERSION 3 90 91 enum prison_state { 92 PRISON_STATE_INVALID = 0, /* New prison, not ready to be seen */ 93 PRISON_STATE_ALIVE, /* Current prison, visible to all */ 94 PRISON_STATE_DYING /* Removed but holding resources, */ 95 }; /* optionally visible. */ 96 97 /* 98 * Flags for jail_set and jail_get. 99 */ 100 #define JAIL_CREATE 0x01 /* Create jail if it doesn't exist */ 101 #define JAIL_UPDATE 0x02 /* Update parameters of existing jail */ 102 #define JAIL_ATTACH 0x04 /* Attach to jail upon creation */ 103 #define JAIL_DYING 0x08 /* Allow getting a dying jail */ 104 #define JAIL_SET_MASK 0x0f 105 #define JAIL_GET_MASK 0x08 106 107 #define JAIL_SYS_DISABLE 0 108 #define JAIL_SYS_NEW 1 109 #define JAIL_SYS_INHERIT 2 110 111 #ifndef _KERNEL 112 113 struct iovec; 114 115 __BEGIN_DECLS 116 int jail(struct jail *); 117 int jail_set(struct iovec *, unsigned int, int); 118 int jail_get(struct iovec *, unsigned int, int); 119 int jail_attach(int); 120 int jail_remove(int); 121 __END_DECLS 122 123 #else /* _KERNEL */ 124 125 #include <sys/queue.h> 126 #include <sys/sysctl.h> 127 #include <sys/lock.h> 128 #include <sys/mutex.h> 129 #include <sys/_task.h> 130 131 #define JAIL_MAX 999999 132 133 #ifdef MALLOC_DECLARE 134 MALLOC_DECLARE(M_PRISON); 135 #endif 136 #endif /* _KERNEL */ 137 138 #if defined(_KERNEL) || defined(_WANT_PRISON) 139 140 #include <sys/osd.h> 141 142 #define HOSTUUIDLEN 64 143 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000" 144 #define OSRELEASELEN 32 145 146 struct racct; 147 struct prison_racct; 148 149 typedef enum { 150 PR_INET = 0, 151 PR_INET6 = 1, 152 PR_FAMILY_MAX = 2, 153 } pr_family_t; 154 155 /* 156 * This structure describes a prison. It is pointed to by all struct 157 * ucreds's of the inmates. pr_ref keeps track of them and is used to 158 * delete the structure when the last inmate is dead. 159 * 160 * Lock key: 161 * (a) allprison_lock 162 * (A) allproc_lock 163 * (c) set only during creation before the structure is shared, no mutex 164 * required to read 165 * (m) locked by pr_mtx 166 * (p) locked by pr_mtx, and also at least shared allprison_lock required 167 * to update 168 * (q) locked by both pr_mtx and allprison_lock 169 * (r) atomic via refcount(9), pr_mtx and allprison_lock required to 170 * decrement to zero 171 * (n) read access granted with the network epoch 172 */ 173 struct prison { 174 TAILQ_ENTRY(prison) pr_list; /* (a) all prisons */ 175 int pr_id; /* (c) prison id */ 176 volatile u_int pr_ref; /* (r) refcount */ 177 volatile u_int pr_uref; /* (r) user (alive) refcount */ 178 unsigned pr_flags; /* (p) PR_* flags */ 179 LIST_HEAD(, prison) pr_children; /* (a) list of child jails */ 180 LIST_HEAD(, proc) pr_proclist; /* (A) list of jailed processes */ 181 LIST_ENTRY(prison) pr_sibling; /* (a) next in parent's list */ 182 struct prison *pr_parent; /* (c) containing jail */ 183 struct mtx pr_mtx; 184 struct task pr_task; /* (c) destroy task */ 185 struct osd pr_osd; /* (p) additional data */ 186 struct cpuset *pr_cpuset; /* (p) cpuset */ 187 struct vnet *pr_vnet; /* (c) network stack */ 188 struct vnode *pr_root; /* (c) vnode to rdir */ 189 struct prison_ip *pr_addrs[PR_FAMILY_MAX]; /* (p,n) IPs of jail */ 190 struct prison_racct *pr_prison_racct; /* (c) racct jail proxy */ 191 void *pr_sparep[3]; 192 int pr_childcount; /* (a) number of child jails */ 193 int pr_childmax; /* (p) maximum child jails */ 194 unsigned pr_allow; /* (p) PR_ALLOW_* flags */ 195 int pr_securelevel; /* (p) securelevel */ 196 int pr_enforce_statfs; /* (p) statfs permission */ 197 int pr_devfs_rsnum; /* (p) devfs ruleset */ 198 enum prison_state pr_state; /* (q) state in life cycle */ 199 volatile int pr_exportcnt; /* (r) count of mount exports */ 200 int pr_spare; 201 int pr_osreldate; /* (c) kern.osreldate value */ 202 unsigned long pr_hostid; /* (p) jail hostid */ 203 char pr_name[MAXHOSTNAMELEN]; /* (p) admin jail name */ 204 char pr_path[MAXPATHLEN]; /* (c) chroot path */ 205 char pr_hostname[MAXHOSTNAMELEN]; /* (p) jail hostname */ 206 char pr_domainname[MAXHOSTNAMELEN]; /* (p) jail domainname */ 207 char pr_hostuuid[HOSTUUIDLEN]; /* (p) jail hostuuid */ 208 char pr_osrelease[OSRELEASELEN]; /* (c) kern.osrelease value */ 209 }; 210 211 struct prison_racct { 212 LIST_ENTRY(prison_racct) prr_next; 213 char prr_name[MAXHOSTNAMELEN]; 214 u_int prr_refcount; 215 struct racct *prr_racct; 216 }; 217 #endif /* _KERNEL || _WANT_PRISON */ 218 219 #ifdef _KERNEL 220 /* Flag bits set via options */ 221 #define PR_PERSIST 0x00000001 /* Can exist without processes */ 222 #define PR_HOST 0x00000002 /* Virtualize hostname et al */ 223 #define PR_IP4_USER 0x00000004 /* Restrict IPv4 addresses */ 224 #define PR_IP6_USER 0x00000008 /* Restrict IPv6 addresses */ 225 #define PR_VNET 0x00000010 /* Virtual network stack */ 226 #define PR_IP4_SADDRSEL 0x00000080 /* Do IPv4 src addr sel. or use the */ 227 /* primary jail address. */ 228 #define PR_IP6_SADDRSEL 0x00000100 /* Do IPv6 src addr sel. or use the */ 229 /* primary jail address. */ 230 231 /* Internal flag bits */ 232 #define PR_REMOVE 0x01000000 /* In process of being removed */ 233 #define PR_IP4 0x02000000 /* IPv4 restricted or disabled */ 234 /* by this jail or an ancestor */ 235 #define PR_IP6 0x04000000 /* IPv6 restricted or disabled */ 236 /* by this jail or an ancestor */ 237 #define PR_COMPLETE_PROC 0x08000000 /* prison_complete called from */ 238 /* prison_proc_free, releases uref */ 239 240 /* 241 * Flags for pr_allow 242 * Bits not noted here may be used for dynamic allow.mount.xxxfs. 243 */ 244 #define PR_ALLOW_SET_HOSTNAME 0x00000001 245 #define PR_ALLOW_SYSVIPC 0x00000002 246 #define PR_ALLOW_RAW_SOCKETS 0x00000004 247 #define PR_ALLOW_CHFLAGS 0x00000008 248 #define PR_ALLOW_MOUNT 0x00000010 249 #define PR_ALLOW_QUOTAS 0x00000020 250 #define PR_ALLOW_SOCKET_AF 0x00000040 251 #define PR_ALLOW_MLOCK 0x00000080 252 #define PR_ALLOW_READ_MSGBUF 0x00000100 253 #define PR_ALLOW_UNPRIV_DEBUG 0x00000200 254 #define PR_ALLOW_SUSER 0x00000400 255 #define PR_ALLOW_RESERVED_PORTS 0x00008000 256 #define PR_ALLOW_KMEM_ACCESS 0x00010000 /* reserved, not used yet */ 257 #define PR_ALLOW_NFSD 0x00020000 258 #define PR_ALLOW_ALL_STATIC 0x000387ff 259 260 /* 261 * PR_ALLOW_DIFFERENCES determines which flags are able to be 262 * different between the parent and child jail upon creation. 263 */ 264 #define PR_ALLOW_DIFFERENCES (PR_ALLOW_UNPRIV_DEBUG) 265 266 /* 267 * OSD methods 268 */ 269 #define PR_METHOD_CREATE 0 270 #define PR_METHOD_GET 1 271 #define PR_METHOD_SET 2 272 #define PR_METHOD_CHECK 3 273 #define PR_METHOD_ATTACH 4 274 #define PR_METHOD_REMOVE 5 275 #define PR_MAXMETHOD 6 276 277 /* 278 * Lock/unlock a prison. 279 * XXX These exist not so much for general convenience, but to be useable in 280 * the FOREACH_PRISON_DESCENDANT_LOCKED macro which can't handle them in 281 * non-function form as currently defined. 282 */ 283 static __inline void 284 prison_lock(struct prison *pr) 285 { 286 287 mtx_lock(&pr->pr_mtx); 288 } 289 290 static __inline void 291 prison_unlock(struct prison *pr) 292 { 293 294 mtx_unlock(&pr->pr_mtx); 295 } 296 297 /* Traverse a prison's immediate children. */ 298 #define FOREACH_PRISON_CHILD(ppr, cpr) \ 299 LIST_FOREACH(cpr, &(ppr)->pr_children, pr_sibling) 300 301 /* 302 * Preorder traversal of all of a prison's descendants. 303 * This ugly loop allows the macro to be followed by a single block 304 * as expected in a looping primitive. 305 */ 306 #define FOREACH_PRISON_DESCENDANT(ppr, cpr, descend) \ 307 for ((cpr) = (ppr), (descend) = 1; \ 308 ((cpr) = (((descend) && !LIST_EMPTY(&(cpr)->pr_children)) \ 309 ? LIST_FIRST(&(cpr)->pr_children) \ 310 : ((cpr) == (ppr) \ 311 ? NULL \ 312 : (((descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \ 313 ? LIST_NEXT(cpr, pr_sibling) \ 314 : (cpr)->pr_parent))));) \ 315 if (!(descend)) \ 316 ; \ 317 else 318 319 /* 320 * As above, but lock descendants on the way down and unlock on the way up. 321 */ 322 #define FOREACH_PRISON_DESCENDANT_LOCKED(ppr, cpr, descend) \ 323 for ((cpr) = (ppr), (descend) = 1; \ 324 ((cpr) = (((descend) && !LIST_EMPTY(&(cpr)->pr_children)) \ 325 ? LIST_FIRST(&(cpr)->pr_children) \ 326 : ((cpr) == (ppr) \ 327 ? NULL \ 328 : ((prison_unlock(cpr), \ 329 (descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \ 330 ? LIST_NEXT(cpr, pr_sibling) \ 331 : (cpr)->pr_parent))));) \ 332 if ((descend) ? (prison_lock(cpr), 0) : 1) \ 333 ; \ 334 else 335 336 /* 337 * As above, but also keep track of the level descended to. 338 */ 339 #define FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(ppr, cpr, descend, level)\ 340 for ((cpr) = (ppr), (descend) = 1, (level) = 0; \ 341 ((cpr) = (((descend) && !LIST_EMPTY(&(cpr)->pr_children)) \ 342 ? (level++, LIST_FIRST(&(cpr)->pr_children)) \ 343 : ((cpr) == (ppr) \ 344 ? NULL \ 345 : ((prison_unlock(cpr), \ 346 (descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \ 347 ? LIST_NEXT(cpr, pr_sibling) \ 348 : (level--, (cpr)->pr_parent)))));) \ 349 if ((descend) ? (prison_lock(cpr), 0) : 1) \ 350 ; \ 351 else 352 353 /* 354 * Traverse a prison's descendants, visiting both preorder and postorder. 355 */ 356 #define FOREACH_PRISON_DESCENDANT_PRE_POST(ppr, cpr, descend) \ 357 for ((cpr) = (ppr), (descend) = 1; \ 358 ((cpr) = (descend) \ 359 ? ((descend) = !LIST_EMPTY(&(cpr)->pr_children)) \ 360 ? LIST_FIRST(&(cpr)->pr_children) \ 361 : (cpr) \ 362 : ((descend) = LIST_NEXT(cpr, pr_sibling) != NULL) \ 363 ? LIST_NEXT(cpr, pr_sibling) \ 364 : cpr->pr_parent) != (ppr);) 365 366 /* 367 * Attributes of the physical system, and the root of the jail tree. 368 */ 369 extern struct prison prison0; 370 371 TAILQ_HEAD(prisonlist, prison); 372 extern struct prisonlist allprison; 373 extern struct sx allprison_lock; 374 375 /* 376 * Sysctls to describe jail parameters. 377 */ 378 SYSCTL_DECL(_security_jail_param); 379 380 #define SYSCTL_JAIL_PARAM(module, param, type, fmt, descr) \ 381 SYSCTL_PROC(_security_jail_param ## module, OID_AUTO, param, \ 382 (type) | CTLFLAG_MPSAFE, NULL, 0, sysctl_jail_param, fmt, descr) 383 #define SYSCTL_JAIL_PARAM_STRING(module, param, access, len, descr) \ 384 SYSCTL_PROC(_security_jail_param ## module, OID_AUTO, param, \ 385 CTLTYPE_STRING | CTLFLAG_MPSAFE | (access), NULL, len, \ 386 sysctl_jail_param, "A", descr) 387 #define SYSCTL_JAIL_PARAM_STRUCT(module, param, access, len, fmt, descr)\ 388 SYSCTL_PROC(_security_jail_param ## module, OID_AUTO, param, \ 389 CTLTYPE_STRUCT | CTLFLAG_MPSAFE | (access), NULL, len, \ 390 sysctl_jail_param, fmt, descr) 391 #define SYSCTL_JAIL_PARAM_NODE(module, descr) \ 392 SYSCTL_NODE(_security_jail_param, OID_AUTO, module, CTLFLAG_MPSAFE, \ 393 0, descr) 394 #define SYSCTL_JAIL_PARAM_SUBNODE(parent, module, descr) \ 395 SYSCTL_NODE(_security_jail_param_##parent, OID_AUTO, module, \ 396 CTLFLAG_MPSAFE, 0, descr) 397 #define SYSCTL_JAIL_PARAM_SYS_NODE(module, access, descr) \ 398 SYSCTL_JAIL_PARAM_NODE(module, descr); \ 399 SYSCTL_JAIL_PARAM(_##module, , CTLTYPE_INT | (access), "E,jailsys", \ 400 descr) 401 402 /* 403 * Kernel support functions for jail(). 404 */ 405 struct ucred; 406 struct mount; 407 struct sockaddr; 408 struct statfs; 409 struct vfsconf; 410 411 /* 412 * Return 1 if the passed credential is in a jail, otherwise 0. 413 */ 414 #define jailed(cred) (cred->cr_prison != &prison0) 415 416 bool jailed_without_vnet(struct ucred *); 417 void getcredhostname(struct ucred *, char *, size_t); 418 void getcreddomainname(struct ucred *, char *, size_t); 419 void getcredhostuuid(struct ucred *, char *, size_t); 420 void getcredhostid(struct ucred *, unsigned long *); 421 void getjailname(struct ucred *cred, char *name, size_t len); 422 void prison0_init(void); 423 bool prison_allow(struct ucred *, unsigned); 424 int prison_check(struct ucred *cred1, struct ucred *cred2); 425 bool prison_check_nfsd(struct ucred *cred); 426 bool prison_owns_vnet(struct ucred *); 427 int prison_canseemount(struct ucred *cred, struct mount *mp); 428 void prison_enforce_statfs(struct ucred *cred, struct mount *mp, 429 struct statfs *sp); 430 struct prison *prison_find(int prid); 431 struct prison *prison_find_child(struct prison *, int); 432 struct prison *prison_find_name(struct prison *, const char *); 433 bool prison_flag(struct ucred *, unsigned); 434 void prison_free(struct prison *pr); 435 void prison_free_locked(struct prison *pr); 436 void prison_hold(struct prison *pr); 437 void prison_hold_locked(struct prison *pr); 438 void prison_proc_hold(struct prison *); 439 void prison_proc_free(struct prison *); 440 void prison_proc_link(struct prison *, struct proc *); 441 void prison_proc_unlink(struct prison *, struct proc *); 442 void prison_proc_iterate(struct prison *, void (*)(struct proc *, void *), void *); 443 void prison_set_allow(struct ucred *cred, unsigned flag, int enable); 444 bool prison_ischild(struct prison *, struct prison *); 445 bool prison_isalive(const struct prison *); 446 bool prison_isvalid(struct prison *); 447 #if defined(INET) || defined(INET6) 448 int prison_ip_check(const struct prison *, const pr_family_t, const void *); 449 const void *prison_ip_get0(const struct prison *, const pr_family_t); 450 u_int prison_ip_cnt(const struct prison *, const pr_family_t); 451 #endif 452 #ifdef INET 453 bool prison_equal_ip4(struct prison *, struct prison *); 454 int prison_get_ip4(struct ucred *cred, struct in_addr *ia); 455 int prison_local_ip4(struct ucred *cred, struct in_addr *ia); 456 int prison_remote_ip4(struct ucred *cred, struct in_addr *ia); 457 int prison_check_ip4(const struct ucred *, const struct in_addr *); 458 int prison_check_ip4_locked(const struct prison *, const struct in_addr *); 459 bool prison_saddrsel_ip4(struct ucred *, struct in_addr *); 460 int prison_qcmp_v4(const void *, const void *); 461 bool prison_valid_v4(const void *); 462 #endif 463 #ifdef INET6 464 bool prison_equal_ip6(struct prison *, struct prison *); 465 int prison_get_ip6(struct ucred *, struct in6_addr *); 466 int prison_local_ip6(struct ucred *, struct in6_addr *, int); 467 int prison_remote_ip6(struct ucred *, struct in6_addr *); 468 int prison_check_ip6(const struct ucred *, const struct in6_addr *); 469 int prison_check_ip6_locked(const struct prison *, const struct in6_addr *); 470 bool prison_saddrsel_ip6(struct ucred *, struct in6_addr *); 471 int prison_qcmp_v6(const void *, const void *); 472 bool prison_valid_v6(const void *); 473 #endif 474 int prison_check_af(struct ucred *cred, int af); 475 int prison_if(struct ucred *cred, const struct sockaddr *sa); 476 char *prison_name(struct prison *, struct prison *); 477 int prison_priv_check(struct ucred *cred, int priv); 478 int sysctl_jail_param(SYSCTL_HANDLER_ARGS); 479 unsigned prison_add_allow(const char *prefix, const char *name, 480 const char *prefix_descr, const char *descr); 481 void prison_add_vfs(struct vfsconf *vfsp); 482 void prison_racct_foreach(void (*callback)(struct racct *racct, 483 void *arg2, void *arg3), void (*pre)(void), void (*post)(void), 484 void *arg2, void *arg3); 485 struct prison_racct *prison_racct_find(const char *name); 486 void prison_racct_hold(struct prison_racct *prr); 487 void prison_racct_free(struct prison_racct *prr); 488 489 #endif /* _KERNEL */ 490 #endif /* !_SYS_JAIL_H_ */ 491