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