1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1999 Poul-Henning Kamp. 5 * Copyright (c) 2008 Bjoern A. Zeeb. 6 * Copyright (c) 2009 James Gritton. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_ddb.h" 35 #include "opt_inet.h" 36 #include "opt_inet6.h" 37 38 #include <sys/param.h> 39 #include <sys/types.h> 40 #include <sys/kernel.h> 41 #include <sys/systm.h> 42 #include <sys/errno.h> 43 #include <sys/sysproto.h> 44 #include <sys/malloc.h> 45 #include <sys/osd.h> 46 #include <sys/priv.h> 47 #include <sys/proc.h> 48 #include <sys/taskqueue.h> 49 #include <sys/fcntl.h> 50 #include <sys/jail.h> 51 #include <sys/linker.h> 52 #include <sys/lock.h> 53 #include <sys/mutex.h> 54 #include <sys/racct.h> 55 #include <sys/rctl.h> 56 #include <sys/refcount.h> 57 #include <sys/sx.h> 58 #include <sys/sysent.h> 59 #include <sys/namei.h> 60 #include <sys/mount.h> 61 #include <sys/queue.h> 62 #include <sys/socket.h> 63 #include <sys/syscallsubr.h> 64 #include <sys/sysctl.h> 65 #include <sys/uuid.h> 66 #include <sys/vnode.h> 67 68 #include <net/if.h> 69 #include <net/vnet.h> 70 71 #include <netinet/in.h> 72 73 #ifdef DDB 74 #include <ddb/ddb.h> 75 #endif /* DDB */ 76 77 #include <security/mac/mac_framework.h> 78 79 #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000" 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_childmax = JAIL_MAX, 110 .pr_hostuuid = DEFAULT_HOSTUUID, 111 .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children), 112 #ifdef VIMAGE 113 .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL, 114 #else 115 .pr_flags = PR_HOST|_PR_IP_SADDRSEL, 116 #endif 117 .pr_allow = PR_ALLOW_ALL_STATIC, 118 }; 119 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF); 120 121 struct bool_flags { 122 const char *name; 123 const char *noname; 124 volatile u_int flag; 125 }; 126 struct jailsys_flags { 127 const char *name; 128 unsigned disable; 129 unsigned new; 130 }; 131 132 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */ 133 struct sx allprison_lock; 134 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison"); 135 struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison); 136 LIST_HEAD(, prison_racct) allprison_racct; 137 int lastprid = 0; 138 139 static int get_next_prid(struct prison **insprp); 140 static int do_jail_attach(struct thread *td, struct prison *pr); 141 static void prison_complete(void *context, int pending); 142 static void prison_deref(struct prison *pr, int flags); 143 static void prison_set_allow_locked(struct prison *pr, unsigned flag, 144 int enable); 145 static char *prison_path(struct prison *pr1, struct prison *pr2); 146 static void prison_remove_one(struct prison *pr); 147 #ifdef RACCT 148 static void prison_racct_attach(struct prison *pr); 149 static void prison_racct_modify(struct prison *pr); 150 static void prison_racct_detach(struct prison *pr); 151 #endif 152 153 /* Flags for prison_deref */ 154 #define PD_DEREF 0x01 155 #define PD_DEUREF 0x02 156 #define PD_LOCKED 0x04 157 #define PD_LIST_SLOCKED 0x08 158 #define PD_LIST_XLOCKED 0x10 159 160 /* 161 * Parameter names corresponding to PR_* flag values. Size values are for kvm 162 * as we cannot figure out the size of a sparse array, or an array without a 163 * terminating entry. 164 */ 165 static struct bool_flags pr_flag_bool[] = { 166 {"persist", "nopersist", PR_PERSIST}, 167 #ifdef INET 168 {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL}, 169 #endif 170 #ifdef INET6 171 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL}, 172 #endif 173 }; 174 const size_t pr_flag_bool_size = sizeof(pr_flag_bool); 175 176 static struct jailsys_flags pr_flag_jailsys[] = { 177 {"host", 0, PR_HOST}, 178 #ifdef VIMAGE 179 {"vnet", 0, PR_VNET}, 180 #endif 181 #ifdef INET 182 {"ip4", PR_IP4_USER, PR_IP4_USER}, 183 #endif 184 #ifdef INET6 185 {"ip6", PR_IP6_USER, PR_IP6_USER}, 186 #endif 187 }; 188 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys); 189 190 /* 191 * Make this array full-size so dynamic parameters can be added. 192 * It is protected by prison0.mtx, but lockless reading is allowed 193 * with an atomic check of the flag values. 194 */ 195 static struct bool_flags pr_flag_allow[NBBY * NBPW] = { 196 {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME}, 197 {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC}, 198 {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS}, 199 {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS}, 200 {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT}, 201 {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS}, 202 {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF}, 203 {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK}, 204 {"allow.reserved_ports", "allow.noreserved_ports", 205 PR_ALLOW_RESERVED_PORTS}, 206 {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF}, 207 {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug", 208 PR_ALLOW_UNPRIV_DEBUG}, 209 {"allow.suser", "allow.nosuser", PR_ALLOW_SUSER}, 210 }; 211 static unsigned pr_allow_all = PR_ALLOW_ALL_STATIC; 212 const size_t pr_flag_allow_size = sizeof(pr_flag_allow); 213 214 #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \ 215 PR_ALLOW_RESERVED_PORTS | \ 216 PR_ALLOW_UNPRIV_DEBUG | \ 217 PR_ALLOW_SUSER) 218 #define JAIL_DEFAULT_ENFORCE_STATFS 2 219 #define JAIL_DEFAULT_DEVFS_RSNUM 0 220 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW; 221 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS; 222 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM; 223 #if defined(INET) || defined(INET6) 224 static unsigned jail_max_af_ips = 255; 225 #endif 226 227 /* 228 * Initialize the parts of prison0 that can't be static-initialized with 229 * constants. This is called from proc0_init() after creating thread0 cpuset. 230 */ 231 void 232 prison0_init(void) 233 { 234 uint8_t *file, *data; 235 size_t size; 236 237 prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset); 238 prison0.pr_osreldate = osreldate; 239 strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease)); 240 241 /* If we have a preloaded hostuuid, use it. */ 242 file = preload_search_by_type(PRISON0_HOSTUUID_MODULE); 243 if (file != NULL) { 244 data = preload_fetch_addr(file); 245 size = preload_fetch_size(file); 246 if (data != NULL) { 247 /* 248 * The preloaded data may include trailing whitespace, almost 249 * certainly a newline; skip over any whitespace or 250 * non-printable characters to be safe. 251 */ 252 while (size > 0 && data[size - 1] <= 0x20) { 253 data[size--] = '\0'; 254 } 255 if (validate_uuid(data, size, NULL, 0) == 0) { 256 (void)strlcpy(prison0.pr_hostuuid, data, 257 size + 1); 258 } else if (bootverbose) { 259 printf("hostuuid: preload data malformed: '%s'", 260 data); 261 } 262 } 263 } 264 if (bootverbose) 265 printf("hostuuid: using %s\n", prison0.pr_hostuuid); 266 } 267 268 /* 269 * struct jail_args { 270 * struct jail *jail; 271 * }; 272 */ 273 int 274 sys_jail(struct thread *td, struct jail_args *uap) 275 { 276 uint32_t version; 277 int error; 278 struct jail j; 279 280 error = copyin(uap->jail, &version, sizeof(uint32_t)); 281 if (error) 282 return (error); 283 284 switch (version) { 285 case 0: 286 { 287 struct jail_v0 j0; 288 289 /* FreeBSD single IPv4 jails. */ 290 bzero(&j, sizeof(struct jail)); 291 error = copyin(uap->jail, &j0, sizeof(struct jail_v0)); 292 if (error) 293 return (error); 294 j.version = j0.version; 295 j.path = j0.path; 296 j.hostname = j0.hostname; 297 j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */ 298 break; 299 } 300 301 case 1: 302 /* 303 * Version 1 was used by multi-IPv4 jail implementations 304 * that never made it into the official kernel. 305 */ 306 return (EINVAL); 307 308 case 2: /* JAIL_API_VERSION */ 309 /* FreeBSD multi-IPv4/IPv6,noIP jails. */ 310 error = copyin(uap->jail, &j, sizeof(struct jail)); 311 if (error) 312 return (error); 313 break; 314 315 default: 316 /* Sci-Fi jails are not supported, sorry. */ 317 return (EINVAL); 318 } 319 return (kern_jail(td, &j)); 320 } 321 322 int 323 kern_jail(struct thread *td, struct jail *j) 324 { 325 struct iovec optiov[2 * (4 + nitems(pr_flag_allow) 326 #ifdef INET 327 + 1 328 #endif 329 #ifdef INET6 330 + 1 331 #endif 332 )]; 333 struct uio opt; 334 char *u_path, *u_hostname, *u_name; 335 struct bool_flags *bf; 336 #ifdef INET 337 uint32_t ip4s; 338 struct in_addr *u_ip4; 339 #endif 340 #ifdef INET6 341 struct in6_addr *u_ip6; 342 #endif 343 size_t tmplen; 344 int error, enforce_statfs; 345 346 bzero(&optiov, sizeof(optiov)); 347 opt.uio_iov = optiov; 348 opt.uio_iovcnt = 0; 349 opt.uio_offset = -1; 350 opt.uio_resid = -1; 351 opt.uio_segflg = UIO_SYSSPACE; 352 opt.uio_rw = UIO_READ; 353 opt.uio_td = td; 354 355 /* Set permissions for top-level jails from sysctls. */ 356 if (!jailed(td->td_ucred)) { 357 for (bf = pr_flag_allow; 358 bf < pr_flag_allow + nitems(pr_flag_allow) && 359 atomic_load_int(&bf->flag) != 0; 360 bf++) { 361 optiov[opt.uio_iovcnt].iov_base = __DECONST(char *, 362 (jail_default_allow & bf->flag) 363 ? bf->name : bf->noname); 364 optiov[opt.uio_iovcnt].iov_len = 365 strlen(optiov[opt.uio_iovcnt].iov_base) + 1; 366 opt.uio_iovcnt += 2; 367 } 368 optiov[opt.uio_iovcnt].iov_base = "enforce_statfs"; 369 optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs"); 370 opt.uio_iovcnt++; 371 enforce_statfs = jail_default_enforce_statfs; 372 optiov[opt.uio_iovcnt].iov_base = &enforce_statfs; 373 optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs); 374 opt.uio_iovcnt++; 375 } 376 377 tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN; 378 #ifdef INET 379 ip4s = (j->version == 0) ? 1 : j->ip4s; 380 if (ip4s > jail_max_af_ips) 381 return (EINVAL); 382 tmplen += ip4s * sizeof(struct in_addr); 383 #else 384 if (j->ip4s > 0) 385 return (EINVAL); 386 #endif 387 #ifdef INET6 388 if (j->ip6s > jail_max_af_ips) 389 return (EINVAL); 390 tmplen += j->ip6s * sizeof(struct in6_addr); 391 #else 392 if (j->ip6s > 0) 393 return (EINVAL); 394 #endif 395 u_path = malloc(tmplen, M_TEMP, M_WAITOK); 396 u_hostname = u_path + MAXPATHLEN; 397 u_name = u_hostname + MAXHOSTNAMELEN; 398 #ifdef INET 399 u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN); 400 #endif 401 #ifdef INET6 402 #ifdef INET 403 u_ip6 = (struct in6_addr *)(u_ip4 + ip4s); 404 #else 405 u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN); 406 #endif 407 #endif 408 optiov[opt.uio_iovcnt].iov_base = "path"; 409 optiov[opt.uio_iovcnt].iov_len = sizeof("path"); 410 opt.uio_iovcnt++; 411 optiov[opt.uio_iovcnt].iov_base = u_path; 412 error = copyinstr(j->path, u_path, MAXPATHLEN, 413 &optiov[opt.uio_iovcnt].iov_len); 414 if (error) { 415 free(u_path, M_TEMP); 416 return (error); 417 } 418 opt.uio_iovcnt++; 419 optiov[opt.uio_iovcnt].iov_base = "host.hostname"; 420 optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname"); 421 opt.uio_iovcnt++; 422 optiov[opt.uio_iovcnt].iov_base = u_hostname; 423 error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN, 424 &optiov[opt.uio_iovcnt].iov_len); 425 if (error) { 426 free(u_path, M_TEMP); 427 return (error); 428 } 429 opt.uio_iovcnt++; 430 if (j->jailname != NULL) { 431 optiov[opt.uio_iovcnt].iov_base = "name"; 432 optiov[opt.uio_iovcnt].iov_len = sizeof("name"); 433 opt.uio_iovcnt++; 434 optiov[opt.uio_iovcnt].iov_base = u_name; 435 error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN, 436 &optiov[opt.uio_iovcnt].iov_len); 437 if (error) { 438 free(u_path, M_TEMP); 439 return (error); 440 } 441 opt.uio_iovcnt++; 442 } 443 #ifdef INET 444 optiov[opt.uio_iovcnt].iov_base = "ip4.addr"; 445 optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr"); 446 opt.uio_iovcnt++; 447 optiov[opt.uio_iovcnt].iov_base = u_ip4; 448 optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr); 449 if (j->version == 0) 450 u_ip4->s_addr = j->ip4s; 451 else { 452 error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len); 453 if (error) { 454 free(u_path, M_TEMP); 455 return (error); 456 } 457 } 458 opt.uio_iovcnt++; 459 #endif 460 #ifdef INET6 461 optiov[opt.uio_iovcnt].iov_base = "ip6.addr"; 462 optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr"); 463 opt.uio_iovcnt++; 464 optiov[opt.uio_iovcnt].iov_base = u_ip6; 465 optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr); 466 error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len); 467 if (error) { 468 free(u_path, M_TEMP); 469 return (error); 470 } 471 opt.uio_iovcnt++; 472 #endif 473 KASSERT(opt.uio_iovcnt <= nitems(optiov), 474 ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt)); 475 error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH); 476 free(u_path, M_TEMP); 477 return (error); 478 } 479 480 /* 481 * struct jail_set_args { 482 * struct iovec *iovp; 483 * unsigned int iovcnt; 484 * int flags; 485 * }; 486 */ 487 int 488 sys_jail_set(struct thread *td, struct jail_set_args *uap) 489 { 490 struct uio *auio; 491 int error; 492 493 /* Check that we have an even number of iovecs. */ 494 if (uap->iovcnt & 1) 495 return (EINVAL); 496 497 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 498 if (error) 499 return (error); 500 error = kern_jail_set(td, auio, uap->flags); 501 free(auio, M_IOV); 502 return (error); 503 } 504 505 int 506 kern_jail_set(struct thread *td, struct uio *optuio, int flags) 507 { 508 struct nameidata nd; 509 #ifdef INET 510 struct in_addr *ip4; 511 #endif 512 #ifdef INET6 513 struct in6_addr *ip6; 514 #endif 515 struct vfsopt *opt; 516 struct vfsoptlist *opts; 517 struct prison *pr, *deadpr, *inspr, *mypr, *ppr, *tpr; 518 struct vnode *root; 519 char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid; 520 char *g_path, *osrelstr; 521 struct bool_flags *bf; 522 struct jailsys_flags *jsf; 523 #if defined(INET) || defined(INET6) 524 struct prison *tppr; 525 void *op; 526 #endif 527 unsigned long hid; 528 size_t namelen, onamelen, pnamelen; 529 int born, created, cuflags, descend, enforce; 530 int error, errmsg_len, errmsg_pos; 531 int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel; 532 int jid, jsys, len, level; 533 int childmax, osreldt, rsnum, slevel; 534 #if defined(INET) || defined(INET6) 535 int ii, ij; 536 #endif 537 #ifdef INET 538 int ip4s, redo_ip4; 539 #endif 540 #ifdef INET6 541 int ip6s, redo_ip6; 542 #endif 543 uint64_t pr_allow, ch_allow, pr_flags, ch_flags; 544 uint64_t pr_allow_diff; 545 unsigned tallow; 546 char numbuf[12]; 547 548 error = priv_check(td, PRIV_JAIL_SET); 549 if (!error && (flags & JAIL_ATTACH)) 550 error = priv_check(td, PRIV_JAIL_ATTACH); 551 if (error) 552 return (error); 553 mypr = td->td_ucred->cr_prison; 554 if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) 555 return (EPERM); 556 if (flags & ~JAIL_SET_MASK) 557 return (EINVAL); 558 559 /* 560 * Check all the parameters before committing to anything. Not all 561 * errors can be caught early, but we may as well try. Also, this 562 * takes care of some expensive stuff (path lookup) before getting 563 * the allprison lock. 564 * 565 * XXX Jails are not filesystems, and jail parameters are not mount 566 * options. But it makes more sense to re-use the vfsopt code 567 * than duplicate it under a different name. 568 */ 569 error = vfs_buildopts(optuio, &opts); 570 if (error) 571 return (error); 572 #ifdef INET 573 ip4 = NULL; 574 #endif 575 #ifdef INET6 576 ip6 = NULL; 577 #endif 578 g_path = NULL; 579 580 cuflags = flags & (JAIL_CREATE | JAIL_UPDATE); 581 if (!cuflags) { 582 error = EINVAL; 583 vfs_opterror(opts, "no valid operation (create or update)"); 584 goto done_errmsg; 585 } 586 587 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid)); 588 if (error == ENOENT) 589 jid = 0; 590 else if (error != 0) 591 goto done_free; 592 593 error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel)); 594 if (error == ENOENT) 595 gotslevel = 0; 596 else if (error != 0) 597 goto done_free; 598 else 599 gotslevel = 1; 600 601 error = 602 vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax)); 603 if (error == ENOENT) 604 gotchildmax = 0; 605 else if (error != 0) 606 goto done_free; 607 else 608 gotchildmax = 1; 609 610 error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce)); 611 if (error == ENOENT) 612 gotenforce = 0; 613 else if (error != 0) 614 goto done_free; 615 else if (enforce < 0 || enforce > 2) { 616 error = EINVAL; 617 goto done_free; 618 } else 619 gotenforce = 1; 620 621 error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum)); 622 if (error == ENOENT) 623 gotrsnum = 0; 624 else if (error != 0) 625 goto done_free; 626 else 627 gotrsnum = 1; 628 629 pr_flags = ch_flags = 0; 630 for (bf = pr_flag_bool; 631 bf < pr_flag_bool + nitems(pr_flag_bool); 632 bf++) { 633 vfs_flagopt(opts, bf->name, &pr_flags, bf->flag); 634 vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag); 635 } 636 ch_flags |= pr_flags; 637 for (jsf = pr_flag_jailsys; 638 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys); 639 jsf++) { 640 error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys)); 641 if (error == ENOENT) 642 continue; 643 if (error != 0) 644 goto done_free; 645 switch (jsys) { 646 case JAIL_SYS_DISABLE: 647 if (!jsf->disable) { 648 error = EINVAL; 649 goto done_free; 650 } 651 pr_flags |= jsf->disable; 652 break; 653 case JAIL_SYS_NEW: 654 pr_flags |= jsf->new; 655 break; 656 case JAIL_SYS_INHERIT: 657 break; 658 default: 659 error = EINVAL; 660 goto done_free; 661 } 662 ch_flags |= jsf->new | jsf->disable; 663 } 664 if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE 665 && !(pr_flags & PR_PERSIST)) { 666 error = EINVAL; 667 vfs_opterror(opts, "new jail must persist or attach"); 668 goto done_errmsg; 669 } 670 #ifdef VIMAGE 671 if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) { 672 error = EINVAL; 673 vfs_opterror(opts, "vnet cannot be changed after creation"); 674 goto done_errmsg; 675 } 676 #endif 677 #ifdef INET 678 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) { 679 error = EINVAL; 680 vfs_opterror(opts, "ip4 cannot be changed after creation"); 681 goto done_errmsg; 682 } 683 #endif 684 #ifdef INET6 685 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) { 686 error = EINVAL; 687 vfs_opterror(opts, "ip6 cannot be changed after creation"); 688 goto done_errmsg; 689 } 690 #endif 691 692 pr_allow = ch_allow = 0; 693 for (bf = pr_flag_allow; 694 bf < pr_flag_allow + nitems(pr_flag_allow) && 695 atomic_load_int(&bf->flag) != 0; 696 bf++) { 697 vfs_flagopt(opts, bf->name, &pr_allow, bf->flag); 698 vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag); 699 } 700 ch_allow |= pr_allow; 701 702 error = vfs_getopt(opts, "name", (void **)&name, &len); 703 if (error == ENOENT) 704 name = NULL; 705 else if (error != 0) 706 goto done_free; 707 else { 708 if (len == 0 || name[len - 1] != '\0') { 709 error = EINVAL; 710 goto done_free; 711 } 712 if (len > MAXHOSTNAMELEN) { 713 error = ENAMETOOLONG; 714 goto done_free; 715 } 716 } 717 718 error = vfs_getopt(opts, "host.hostname", (void **)&host, &len); 719 if (error == ENOENT) 720 host = NULL; 721 else if (error != 0) 722 goto done_free; 723 else { 724 ch_flags |= PR_HOST; 725 pr_flags |= PR_HOST; 726 if (len == 0 || host[len - 1] != '\0') { 727 error = EINVAL; 728 goto done_free; 729 } 730 if (len > MAXHOSTNAMELEN) { 731 error = ENAMETOOLONG; 732 goto done_free; 733 } 734 } 735 736 error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len); 737 if (error == ENOENT) 738 domain = NULL; 739 else if (error != 0) 740 goto done_free; 741 else { 742 ch_flags |= PR_HOST; 743 pr_flags |= PR_HOST; 744 if (len == 0 || domain[len - 1] != '\0') { 745 error = EINVAL; 746 goto done_free; 747 } 748 if (len > MAXHOSTNAMELEN) { 749 error = ENAMETOOLONG; 750 goto done_free; 751 } 752 } 753 754 error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len); 755 if (error == ENOENT) 756 uuid = NULL; 757 else if (error != 0) 758 goto done_free; 759 else { 760 ch_flags |= PR_HOST; 761 pr_flags |= PR_HOST; 762 if (len == 0 || uuid[len - 1] != '\0') { 763 error = EINVAL; 764 goto done_free; 765 } 766 if (len > HOSTUUIDLEN) { 767 error = ENAMETOOLONG; 768 goto done_free; 769 } 770 } 771 772 #ifdef COMPAT_FREEBSD32 773 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 774 uint32_t hid32; 775 776 error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32)); 777 hid = hid32; 778 } else 779 #endif 780 error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid)); 781 if (error == ENOENT) 782 gothid = 0; 783 else if (error != 0) 784 goto done_free; 785 else { 786 gothid = 1; 787 ch_flags |= PR_HOST; 788 pr_flags |= PR_HOST; 789 } 790 791 #ifdef INET 792 error = vfs_getopt(opts, "ip4.addr", &op, &ip4s); 793 if (error == ENOENT) 794 ip4s = 0; 795 else if (error != 0) 796 goto done_free; 797 else if (ip4s & (sizeof(*ip4) - 1)) { 798 error = EINVAL; 799 goto done_free; 800 } else { 801 ch_flags |= PR_IP4_USER; 802 pr_flags |= PR_IP4_USER; 803 if (ip4s > 0) { 804 ip4s /= sizeof(*ip4); 805 if (ip4s > jail_max_af_ips) { 806 error = EINVAL; 807 vfs_opterror(opts, "too many IPv4 addresses"); 808 goto done_errmsg; 809 } 810 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK); 811 bcopy(op, ip4, ip4s * sizeof(*ip4)); 812 /* 813 * IP addresses are all sorted but ip[0] to preserve 814 * the primary IP address as given from userland. 815 * This special IP is used for unbound outgoing 816 * connections as well for "loopback" traffic in case 817 * source address selection cannot find any more fitting 818 * address to connect from. 819 */ 820 if (ip4s > 1) 821 qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), 822 prison_qcmp_v4); 823 /* 824 * Check for duplicate addresses and do some simple 825 * zero and broadcast checks. If users give other bogus 826 * addresses it is their problem. 827 * 828 * We do not have to care about byte order for these 829 * checks so we will do them in NBO. 830 */ 831 for (ii = 0; ii < ip4s; ii++) { 832 if (ip4[ii].s_addr == INADDR_ANY || 833 ip4[ii].s_addr == INADDR_BROADCAST) { 834 error = EINVAL; 835 goto done_free; 836 } 837 if ((ii+1) < ip4s && 838 (ip4[0].s_addr == ip4[ii+1].s_addr || 839 ip4[ii].s_addr == ip4[ii+1].s_addr)) { 840 error = EINVAL; 841 goto done_free; 842 } 843 } 844 } 845 } 846 #endif 847 848 #ifdef INET6 849 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s); 850 if (error == ENOENT) 851 ip6s = 0; 852 else if (error != 0) 853 goto done_free; 854 else if (ip6s & (sizeof(*ip6) - 1)) { 855 error = EINVAL; 856 goto done_free; 857 } else { 858 ch_flags |= PR_IP6_USER; 859 pr_flags |= PR_IP6_USER; 860 if (ip6s > 0) { 861 ip6s /= sizeof(*ip6); 862 if (ip6s > jail_max_af_ips) { 863 error = EINVAL; 864 vfs_opterror(opts, "too many IPv6 addresses"); 865 goto done_errmsg; 866 } 867 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK); 868 bcopy(op, ip6, ip6s * sizeof(*ip6)); 869 if (ip6s > 1) 870 qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), 871 prison_qcmp_v6); 872 for (ii = 0; ii < ip6s; ii++) { 873 if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) { 874 error = EINVAL; 875 goto done_free; 876 } 877 if ((ii+1) < ip6s && 878 (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) || 879 IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1]))) 880 { 881 error = EINVAL; 882 goto done_free; 883 } 884 } 885 } 886 } 887 #endif 888 889 #if defined(VIMAGE) && (defined(INET) || defined(INET6)) 890 if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) { 891 error = EINVAL; 892 vfs_opterror(opts, 893 "vnet jails cannot have IP address restrictions"); 894 goto done_errmsg; 895 } 896 #endif 897 898 error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len); 899 if (error == ENOENT) 900 osrelstr = NULL; 901 else if (error != 0) 902 goto done_free; 903 else { 904 if (flags & JAIL_UPDATE) { 905 error = EINVAL; 906 vfs_opterror(opts, 907 "osrelease cannot be changed after creation"); 908 goto done_errmsg; 909 } 910 if (len == 0 || osrelstr[len - 1] != '\0') { 911 error = EINVAL; 912 goto done_free; 913 } 914 if (len >= OSRELEASELEN) { 915 error = ENAMETOOLONG; 916 vfs_opterror(opts, 917 "osrelease string must be 1-%d bytes long", 918 OSRELEASELEN - 1); 919 goto done_errmsg; 920 } 921 } 922 923 error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt)); 924 if (error == ENOENT) 925 osreldt = 0; 926 else if (error != 0) 927 goto done_free; 928 else { 929 if (flags & JAIL_UPDATE) { 930 error = EINVAL; 931 vfs_opterror(opts, 932 "osreldate cannot be changed after creation"); 933 goto done_errmsg; 934 } 935 if (osreldt == 0) { 936 error = EINVAL; 937 vfs_opterror(opts, "osreldate cannot be 0"); 938 goto done_errmsg; 939 } 940 } 941 942 root = NULL; 943 error = vfs_getopt(opts, "path", (void **)&path, &len); 944 if (error == ENOENT) 945 path = NULL; 946 else if (error != 0) 947 goto done_free; 948 else { 949 if (flags & JAIL_UPDATE) { 950 error = EINVAL; 951 vfs_opterror(opts, 952 "path cannot be changed after creation"); 953 goto done_errmsg; 954 } 955 if (len == 0 || path[len - 1] != '\0') { 956 error = EINVAL; 957 goto done_free; 958 } 959 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, 960 path, td); 961 error = namei(&nd); 962 if (error) 963 goto done_free; 964 root = nd.ni_vp; 965 NDFREE(&nd, NDF_ONLY_PNBUF); 966 g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 967 strlcpy(g_path, path, MAXPATHLEN); 968 error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN); 969 if (error == 0) { 970 path = g_path; 971 } else { 972 /* exit on other errors */ 973 goto done_free; 974 } 975 if (root->v_type != VDIR) { 976 error = ENOTDIR; 977 vput(root); 978 goto done_free; 979 } 980 VOP_UNLOCK(root); 981 } 982 983 /* 984 * Find the specified jail, or at least its parent. 985 * This abuses the file error codes ENOENT and EEXIST. 986 */ 987 pr = NULL; 988 ppr = mypr; 989 inspr = NULL; 990 if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) { 991 namelc = strrchr(name, '.'); 992 jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10); 993 if (*p != '\0') 994 jid = 0; 995 } 996 sx_xlock(&allprison_lock); 997 if (jid != 0) { 998 if (jid < 0) { 999 error = EINVAL; 1000 vfs_opterror(opts, "negative jid"); 1001 goto done_unlock_list; 1002 } 1003 /* 1004 * See if a requested jid already exists. Keep track of 1005 * where it can be inserted later. 1006 */ 1007 TAILQ_FOREACH(inspr, &allprison, pr_list) { 1008 if (inspr->pr_id == jid) { 1009 mtx_lock(&inspr->pr_mtx); 1010 if (inspr->pr_ref > 0) { 1011 pr = inspr; 1012 inspr = NULL; 1013 } else 1014 mtx_unlock(&inspr->pr_mtx); 1015 break; 1016 } 1017 if (inspr->pr_id > jid) 1018 break; 1019 } 1020 if (pr != NULL) { 1021 ppr = pr->pr_parent; 1022 /* Create: jid must not exist. */ 1023 if (cuflags == JAIL_CREATE) { 1024 /* 1025 * Even creators that cannot see the jail will 1026 * get EEXIST. 1027 */ 1028 mtx_unlock(&pr->pr_mtx); 1029 error = EEXIST; 1030 vfs_opterror(opts, "jail %d already exists", 1031 jid); 1032 goto done_unlock_list; 1033 } 1034 if (!prison_ischild(mypr, pr)) { 1035 /* 1036 * Updaters get ENOENT if they cannot see the 1037 * jail. This is true even for CREATE | UPDATE, 1038 * which normally cannot give this error. 1039 */ 1040 mtx_unlock(&pr->pr_mtx); 1041 pr = NULL; 1042 } else if (pr->pr_uref == 0) { 1043 if (!(flags & JAIL_DYING)) { 1044 mtx_unlock(&pr->pr_mtx); 1045 error = ENOENT; 1046 vfs_opterror(opts, "jail %d is dying", 1047 jid); 1048 goto done_unlock_list; 1049 } else if ((flags & JAIL_ATTACH) || 1050 (pr_flags & PR_PERSIST)) { 1051 /* 1052 * A dying jail might be resurrected 1053 * (via attach or persist), but first 1054 * it must determine if another jail 1055 * has claimed its name. Accomplish 1056 * this by implicitly re-setting the 1057 * name. 1058 */ 1059 if (name == NULL) 1060 name = prison_name(mypr, pr); 1061 } 1062 } 1063 } 1064 if (pr == NULL) { 1065 /* Update: jid must exist. */ 1066 if (cuflags == JAIL_UPDATE) { 1067 error = ENOENT; 1068 vfs_opterror(opts, "jail %d not found", jid); 1069 goto done_unlock_list; 1070 } 1071 } 1072 } 1073 /* 1074 * If the caller provided a name, look for a jail by that name. 1075 * This has different semantics for creates and updates keyed by jid 1076 * (where the name must not already exist in a different jail), 1077 * and updates keyed by the name itself (where the name must exist 1078 * because that is the jail being updated). 1079 */ 1080 namelc = NULL; 1081 if (name != NULL) { 1082 namelc = strrchr(name, '.'); 1083 if (namelc == NULL) 1084 namelc = name; 1085 else { 1086 /* 1087 * This is a hierarchical name. Split it into the 1088 * parent and child names, and make sure the parent 1089 * exists or matches an already found jail. 1090 */ 1091 if (pr != NULL) { 1092 if (strncmp(name, ppr->pr_name, namelc - name) 1093 || ppr->pr_name[namelc - name] != '\0') { 1094 mtx_unlock(&pr->pr_mtx); 1095 error = EINVAL; 1096 vfs_opterror(opts, 1097 "cannot change jail's parent"); 1098 goto done_unlock_list; 1099 } 1100 } else { 1101 *namelc = '\0'; 1102 ppr = prison_find_name(mypr, name); 1103 if (ppr == NULL) { 1104 error = ENOENT; 1105 vfs_opterror(opts, 1106 "jail \"%s\" not found", name); 1107 goto done_unlock_list; 1108 } 1109 mtx_unlock(&ppr->pr_mtx); 1110 *namelc = '.'; 1111 } 1112 namelc++; 1113 } 1114 if (namelc[0] != '\0') { 1115 pnamelen = 1116 (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1; 1117 name_again: 1118 deadpr = NULL; 1119 FOREACH_PRISON_CHILD(ppr, tpr) { 1120 if (tpr != pr && tpr->pr_ref > 0 && 1121 !strcmp(tpr->pr_name + pnamelen, namelc)) { 1122 if (pr == NULL && 1123 cuflags != JAIL_CREATE) { 1124 mtx_lock(&tpr->pr_mtx); 1125 if (tpr->pr_ref > 0) { 1126 /* 1127 * Use this jail 1128 * for updates. 1129 */ 1130 if (tpr->pr_uref > 0) { 1131 pr = tpr; 1132 break; 1133 } 1134 deadpr = tpr; 1135 } 1136 mtx_unlock(&tpr->pr_mtx); 1137 } else if (tpr->pr_uref > 0) { 1138 /* 1139 * Create, or update(jid): 1140 * name must not exist in an 1141 * active sibling jail. 1142 */ 1143 error = EEXIST; 1144 if (pr != NULL) 1145 mtx_unlock(&pr->pr_mtx); 1146 vfs_opterror(opts, 1147 "jail \"%s\" already exists", 1148 name); 1149 goto done_unlock_list; 1150 } 1151 } 1152 } 1153 /* If no active jail is found, use a dying one. */ 1154 if (deadpr != NULL && pr == NULL) { 1155 if (flags & JAIL_DYING) { 1156 mtx_lock(&deadpr->pr_mtx); 1157 if (deadpr->pr_ref == 0) { 1158 mtx_unlock(&deadpr->pr_mtx); 1159 goto name_again; 1160 } 1161 pr = deadpr; 1162 } else if (cuflags == JAIL_UPDATE) { 1163 error = ENOENT; 1164 vfs_opterror(opts, 1165 "jail \"%s\" is dying", name); 1166 goto done_unlock_list; 1167 } 1168 } 1169 /* Update: name must exist if no jid. */ 1170 else if (cuflags == JAIL_UPDATE && pr == NULL) { 1171 error = ENOENT; 1172 vfs_opterror(opts, "jail \"%s\" not found", 1173 name); 1174 goto done_unlock_list; 1175 } 1176 } 1177 } 1178 /* Update: must provide a jid or name. */ 1179 else if (cuflags == JAIL_UPDATE && pr == NULL) { 1180 error = ENOENT; 1181 vfs_opterror(opts, "update specified no jail"); 1182 goto done_unlock_list; 1183 } 1184 1185 /* If there's no prison to update, create a new one and link it in. */ 1186 if (pr == NULL) { 1187 for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent) 1188 if (tpr->pr_childcount >= tpr->pr_childmax) { 1189 error = EPERM; 1190 vfs_opterror(opts, "prison limit exceeded"); 1191 goto done_unlock_list; 1192 } 1193 created = 1; 1194 mtx_lock(&ppr->pr_mtx); 1195 if (ppr->pr_ref == 0) { 1196 mtx_unlock(&ppr->pr_mtx); 1197 error = ENOENT; 1198 vfs_opterror(opts, "jail \"%s\" not found", 1199 prison_name(mypr, ppr)); 1200 goto done_unlock_list; 1201 } 1202 ppr->pr_ref++; 1203 ppr->pr_uref++; 1204 mtx_unlock(&ppr->pr_mtx); 1205 pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO); 1206 1207 if (jid == 0 && (jid = get_next_prid(&inspr)) == 0) { 1208 error = EAGAIN; 1209 vfs_opterror(opts, "no available jail IDs"); 1210 free(pr, M_PRISON); 1211 prison_deref(ppr, 1212 PD_DEREF | PD_DEUREF | PD_LIST_XLOCKED); 1213 goto done_releroot; 1214 } 1215 pr->pr_id = jid; 1216 if (inspr != NULL) 1217 TAILQ_INSERT_BEFORE(inspr, pr, pr_list); 1218 else 1219 TAILQ_INSERT_TAIL(&allprison, pr, pr_list); 1220 1221 pr->pr_parent = ppr; 1222 LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling); 1223 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent) 1224 tpr->pr_childcount++; 1225 1226 /* Set some default values, and inherit some from the parent. */ 1227 if (namelc == NULL) 1228 namelc = ""; 1229 if (path == NULL) { 1230 path = "/"; 1231 root = mypr->pr_root; 1232 vref(root); 1233 } 1234 strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN); 1235 pr->pr_flags |= PR_HOST; 1236 #if defined(INET) || defined(INET6) 1237 #ifdef VIMAGE 1238 if (!(pr_flags & PR_VNET)) 1239 #endif 1240 { 1241 #ifdef INET 1242 if (!(ch_flags & PR_IP4_USER)) 1243 pr->pr_flags |= PR_IP4 | PR_IP4_USER; 1244 else if (!(pr_flags & PR_IP4_USER)) { 1245 pr->pr_flags |= ppr->pr_flags & PR_IP4; 1246 if (ppr->pr_ip4 != NULL) { 1247 pr->pr_ip4s = ppr->pr_ip4s; 1248 pr->pr_ip4 = malloc(pr->pr_ip4s * 1249 sizeof(struct in_addr), M_PRISON, 1250 M_WAITOK); 1251 bcopy(ppr->pr_ip4, pr->pr_ip4, 1252 pr->pr_ip4s * sizeof(*pr->pr_ip4)); 1253 } 1254 } 1255 #endif 1256 #ifdef INET6 1257 if (!(ch_flags & PR_IP6_USER)) 1258 pr->pr_flags |= PR_IP6 | PR_IP6_USER; 1259 else if (!(pr_flags & PR_IP6_USER)) { 1260 pr->pr_flags |= ppr->pr_flags & PR_IP6; 1261 if (ppr->pr_ip6 != NULL) { 1262 pr->pr_ip6s = ppr->pr_ip6s; 1263 pr->pr_ip6 = malloc(pr->pr_ip6s * 1264 sizeof(struct in6_addr), M_PRISON, 1265 M_WAITOK); 1266 bcopy(ppr->pr_ip6, pr->pr_ip6, 1267 pr->pr_ip6s * sizeof(*pr->pr_ip6)); 1268 } 1269 } 1270 #endif 1271 } 1272 #endif 1273 /* Source address selection is always on by default. */ 1274 pr->pr_flags |= _PR_IP_SADDRSEL; 1275 1276 pr->pr_securelevel = ppr->pr_securelevel; 1277 pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow; 1278 pr->pr_enforce_statfs = jail_default_enforce_statfs; 1279 pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum; 1280 1281 pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate; 1282 if (osrelstr == NULL) 1283 strlcpy(pr->pr_osrelease, ppr->pr_osrelease, 1284 sizeof(pr->pr_osrelease)); 1285 else 1286 strlcpy(pr->pr_osrelease, osrelstr, 1287 sizeof(pr->pr_osrelease)); 1288 1289 LIST_INIT(&pr->pr_children); 1290 mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK); 1291 TASK_INIT(&pr->pr_task, 0, prison_complete, pr); 1292 1293 #ifdef VIMAGE 1294 /* Allocate a new vnet if specified. */ 1295 pr->pr_vnet = (pr_flags & PR_VNET) 1296 ? vnet_alloc() : ppr->pr_vnet; 1297 #endif 1298 /* 1299 * Allocate a dedicated cpuset for each jail. 1300 * Unlike other initial settings, this may return an erorr. 1301 */ 1302 error = cpuset_create_root(ppr, &pr->pr_cpuset); 1303 if (error) { 1304 prison_deref(pr, PD_LIST_XLOCKED); 1305 goto done_releroot; 1306 } 1307 1308 mtx_lock(&pr->pr_mtx); 1309 /* 1310 * New prisons do not yet have a reference, because we do not 1311 * want others to see the incomplete prison once the 1312 * allprison_lock is downgraded. 1313 */ 1314 } else { 1315 created = 0; 1316 /* 1317 * Grab a reference for existing prisons, to ensure they 1318 * continue to exist for the duration of the call. 1319 */ 1320 pr->pr_ref++; 1321 #if defined(VIMAGE) && (defined(INET) || defined(INET6)) 1322 if ((pr->pr_flags & PR_VNET) && 1323 (ch_flags & (PR_IP4_USER | PR_IP6_USER))) { 1324 error = EINVAL; 1325 vfs_opterror(opts, 1326 "vnet jails cannot have IP address restrictions"); 1327 goto done_deref_locked; 1328 } 1329 #endif 1330 #ifdef INET 1331 if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) { 1332 error = EINVAL; 1333 vfs_opterror(opts, 1334 "ip4 cannot be changed after creation"); 1335 goto done_deref_locked; 1336 } 1337 #endif 1338 #ifdef INET6 1339 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) { 1340 error = EINVAL; 1341 vfs_opterror(opts, 1342 "ip6 cannot be changed after creation"); 1343 goto done_deref_locked; 1344 } 1345 #endif 1346 } 1347 1348 /* Do final error checking before setting anything. */ 1349 if (gotslevel) { 1350 if (slevel < ppr->pr_securelevel) { 1351 error = EPERM; 1352 goto done_deref_locked; 1353 } 1354 } 1355 if (gotchildmax) { 1356 if (childmax >= ppr->pr_childmax) { 1357 error = EPERM; 1358 goto done_deref_locked; 1359 } 1360 } 1361 if (gotenforce) { 1362 if (enforce < ppr->pr_enforce_statfs) { 1363 error = EPERM; 1364 goto done_deref_locked; 1365 } 1366 } 1367 if (gotrsnum) { 1368 /* 1369 * devfs_rsnum is a uint16_t 1370 */ 1371 if (rsnum < 0 || rsnum > 65535) { 1372 error = EINVAL; 1373 goto done_deref_locked; 1374 } 1375 /* 1376 * Nested jails always inherit parent's devfs ruleset 1377 */ 1378 if (jailed(td->td_ucred)) { 1379 if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) { 1380 error = EPERM; 1381 goto done_deref_locked; 1382 } else 1383 rsnum = ppr->pr_devfs_rsnum; 1384 } 1385 } 1386 #ifdef INET 1387 if (ip4s > 0) { 1388 if (ppr->pr_flags & PR_IP4) { 1389 /* 1390 * Make sure the new set of IP addresses is a 1391 * subset of the parent's list. Don't worry 1392 * about the parent being unlocked, as any 1393 * setting is done with allprison_lock held. 1394 */ 1395 for (ij = 0; ij < ppr->pr_ip4s; ij++) 1396 if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr) 1397 break; 1398 if (ij == ppr->pr_ip4s) { 1399 error = EPERM; 1400 goto done_deref_locked; 1401 } 1402 if (ip4s > 1) { 1403 for (ii = ij = 1; ii < ip4s; ii++) { 1404 if (ip4[ii].s_addr == 1405 ppr->pr_ip4[0].s_addr) 1406 continue; 1407 for (; ij < ppr->pr_ip4s; ij++) 1408 if (ip4[ii].s_addr == 1409 ppr->pr_ip4[ij].s_addr) 1410 break; 1411 if (ij == ppr->pr_ip4s) 1412 break; 1413 } 1414 if (ij == ppr->pr_ip4s) { 1415 error = EPERM; 1416 goto done_deref_locked; 1417 } 1418 } 1419 } 1420 /* 1421 * Check for conflicting IP addresses. We permit them 1422 * if there is no more than one IP on each jail. If 1423 * there is a duplicate on a jail with more than one 1424 * IP stop checking and return error. 1425 */ 1426 #ifdef VIMAGE 1427 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent) 1428 if (tppr->pr_flags & PR_VNET) 1429 break; 1430 #else 1431 tppr = &prison0; 1432 #endif 1433 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) { 1434 if (tpr == pr || 1435 #ifdef VIMAGE 1436 (tpr != tppr && (tpr->pr_flags & PR_VNET)) || 1437 #endif 1438 tpr->pr_uref == 0) { 1439 descend = 0; 1440 continue; 1441 } 1442 if (!(tpr->pr_flags & PR_IP4_USER)) 1443 continue; 1444 descend = 0; 1445 if (tpr->pr_ip4 == NULL || 1446 (ip4s == 1 && tpr->pr_ip4s == 1)) 1447 continue; 1448 for (ii = 0; ii < ip4s; ii++) { 1449 if (prison_check_ip4_locked(tpr, &ip4[ii]) == 1450 0) { 1451 error = EADDRINUSE; 1452 vfs_opterror(opts, 1453 "IPv4 addresses clash"); 1454 goto done_deref_locked; 1455 } 1456 } 1457 } 1458 } 1459 #endif 1460 #ifdef INET6 1461 if (ip6s > 0) { 1462 if (ppr->pr_flags & PR_IP6) { 1463 /* 1464 * Make sure the new set of IP addresses is a 1465 * subset of the parent's list. 1466 */ 1467 for (ij = 0; ij < ppr->pr_ip6s; ij++) 1468 if (IN6_ARE_ADDR_EQUAL(&ip6[0], 1469 &ppr->pr_ip6[ij])) 1470 break; 1471 if (ij == ppr->pr_ip6s) { 1472 error = EPERM; 1473 goto done_deref_locked; 1474 } 1475 if (ip6s > 1) { 1476 for (ii = ij = 1; ii < ip6s; ii++) { 1477 if (IN6_ARE_ADDR_EQUAL(&ip6[ii], 1478 &ppr->pr_ip6[0])) 1479 continue; 1480 for (; ij < ppr->pr_ip6s; ij++) 1481 if (IN6_ARE_ADDR_EQUAL( 1482 &ip6[ii], &ppr->pr_ip6[ij])) 1483 break; 1484 if (ij == ppr->pr_ip6s) 1485 break; 1486 } 1487 if (ij == ppr->pr_ip6s) { 1488 error = EPERM; 1489 goto done_deref_locked; 1490 } 1491 } 1492 } 1493 /* Check for conflicting IP addresses. */ 1494 #ifdef VIMAGE 1495 for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent) 1496 if (tppr->pr_flags & PR_VNET) 1497 break; 1498 #else 1499 tppr = &prison0; 1500 #endif 1501 FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) { 1502 if (tpr == pr || 1503 #ifdef VIMAGE 1504 (tpr != tppr && (tpr->pr_flags & PR_VNET)) || 1505 #endif 1506 tpr->pr_uref == 0) { 1507 descend = 0; 1508 continue; 1509 } 1510 if (!(tpr->pr_flags & PR_IP6_USER)) 1511 continue; 1512 descend = 0; 1513 if (tpr->pr_ip6 == NULL || 1514 (ip6s == 1 && tpr->pr_ip6s == 1)) 1515 continue; 1516 for (ii = 0; ii < ip6s; ii++) { 1517 if (prison_check_ip6_locked(tpr, &ip6[ii]) == 1518 0) { 1519 error = EADDRINUSE; 1520 vfs_opterror(opts, 1521 "IPv6 addresses clash"); 1522 goto done_deref_locked; 1523 } 1524 } 1525 } 1526 } 1527 #endif 1528 onamelen = namelen = 0; 1529 if (namelc != NULL) { 1530 /* Give a default name of the jid. Also allow the name to be 1531 * explicitly the jid - but not any other number, and only in 1532 * normal form (no leading zero/etc). 1533 */ 1534 if (namelc[0] == '\0') 1535 snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid); 1536 else if ((strtoul(namelc, &p, 10) != jid || 1537 namelc[0] < '1' || namelc[0] > '9') && *p == '\0') { 1538 error = EINVAL; 1539 vfs_opterror(opts, 1540 "name cannot be numeric (unless it is the jid)"); 1541 goto done_deref_locked; 1542 } 1543 /* 1544 * Make sure the name isn't too long for the prison or its 1545 * children. 1546 */ 1547 pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1; 1548 onamelen = strlen(pr->pr_name + pnamelen); 1549 namelen = strlen(namelc); 1550 if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) { 1551 error = ENAMETOOLONG; 1552 goto done_deref_locked; 1553 } 1554 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1555 if (strlen(tpr->pr_name) + (namelen - onamelen) >= 1556 sizeof(pr->pr_name)) { 1557 error = ENAMETOOLONG; 1558 goto done_deref_locked; 1559 } 1560 } 1561 } 1562 pr_allow_diff = pr_allow & ~ppr->pr_allow; 1563 if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) { 1564 error = EPERM; 1565 goto done_deref_locked; 1566 } 1567 1568 /* 1569 * Let modules check their parameters. This requires unlocking and 1570 * then re-locking the prison, but this is still a valid state as long 1571 * as allprison_lock remains xlocked. 1572 */ 1573 mtx_unlock(&pr->pr_mtx); 1574 error = osd_jail_call(pr, PR_METHOD_CHECK, opts); 1575 if (error != 0) { 1576 prison_deref(pr, created 1577 ? PD_LIST_XLOCKED 1578 : PD_DEREF | PD_LIST_XLOCKED); 1579 goto done_releroot; 1580 } 1581 mtx_lock(&pr->pr_mtx); 1582 1583 /* At this point, all valid parameters should have been noted. */ 1584 TAILQ_FOREACH(opt, opts, link) { 1585 if (!opt->seen && strcmp(opt->name, "errmsg")) { 1586 error = EINVAL; 1587 vfs_opterror(opts, "unknown parameter: %s", opt->name); 1588 goto done_deref_locked; 1589 } 1590 } 1591 1592 /* Set the parameters of the prison. */ 1593 #ifdef INET 1594 redo_ip4 = 0; 1595 if (pr_flags & PR_IP4_USER) { 1596 pr->pr_flags |= PR_IP4; 1597 free(pr->pr_ip4, M_PRISON); 1598 pr->pr_ip4s = ip4s; 1599 pr->pr_ip4 = ip4; 1600 ip4 = NULL; 1601 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { 1602 #ifdef VIMAGE 1603 if (tpr->pr_flags & PR_VNET) { 1604 descend = 0; 1605 continue; 1606 } 1607 #endif 1608 if (prison_restrict_ip4(tpr, NULL)) { 1609 redo_ip4 = 1; 1610 descend = 0; 1611 } 1612 } 1613 } 1614 #endif 1615 #ifdef INET6 1616 redo_ip6 = 0; 1617 if (pr_flags & PR_IP6_USER) { 1618 pr->pr_flags |= PR_IP6; 1619 free(pr->pr_ip6, M_PRISON); 1620 pr->pr_ip6s = ip6s; 1621 pr->pr_ip6 = ip6; 1622 ip6 = NULL; 1623 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { 1624 #ifdef VIMAGE 1625 if (tpr->pr_flags & PR_VNET) { 1626 descend = 0; 1627 continue; 1628 } 1629 #endif 1630 if (prison_restrict_ip6(tpr, NULL)) { 1631 redo_ip6 = 1; 1632 descend = 0; 1633 } 1634 } 1635 } 1636 #endif 1637 if (gotslevel) { 1638 pr->pr_securelevel = slevel; 1639 /* Set all child jails to be at least this level. */ 1640 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) 1641 if (tpr->pr_securelevel < slevel) 1642 tpr->pr_securelevel = slevel; 1643 } 1644 if (gotchildmax) { 1645 pr->pr_childmax = childmax; 1646 /* Set all child jails to under this limit. */ 1647 FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level) 1648 if (tpr->pr_childmax > childmax - level) 1649 tpr->pr_childmax = childmax > level 1650 ? childmax - level : 0; 1651 } 1652 if (gotenforce) { 1653 pr->pr_enforce_statfs = enforce; 1654 /* Pass this restriction on to the children. */ 1655 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) 1656 if (tpr->pr_enforce_statfs < enforce) 1657 tpr->pr_enforce_statfs = enforce; 1658 } 1659 if (gotrsnum) { 1660 pr->pr_devfs_rsnum = rsnum; 1661 /* Pass this restriction on to the children. */ 1662 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) 1663 tpr->pr_devfs_rsnum = rsnum; 1664 } 1665 if (namelc != NULL) { 1666 if (ppr == &prison0) 1667 strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name)); 1668 else 1669 snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s", 1670 ppr->pr_name, namelc); 1671 /* Change this component of child names. */ 1672 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { 1673 bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen, 1674 strlen(tpr->pr_name + onamelen) + 1); 1675 bcopy(pr->pr_name, tpr->pr_name, namelen); 1676 } 1677 } 1678 if (path != NULL) { 1679 /* Try to keep a real-rooted full pathname. */ 1680 strlcpy(pr->pr_path, path, sizeof(pr->pr_path)); 1681 pr->pr_root = root; 1682 } 1683 if (PR_HOST & ch_flags & ~pr_flags) { 1684 if (pr->pr_flags & PR_HOST) { 1685 /* 1686 * Copy the parent's host info. As with pr_ip4 above, 1687 * the lack of a lock on the parent is not a problem; 1688 * it is always set with allprison_lock at least 1689 * shared, and is held exclusively here. 1690 */ 1691 strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname, 1692 sizeof(pr->pr_hostname)); 1693 strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname, 1694 sizeof(pr->pr_domainname)); 1695 strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid, 1696 sizeof(pr->pr_hostuuid)); 1697 pr->pr_hostid = pr->pr_parent->pr_hostid; 1698 } 1699 } else if (host != NULL || domain != NULL || uuid != NULL || gothid) { 1700 /* Set this prison, and any descendants without PR_HOST. */ 1701 if (host != NULL) 1702 strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname)); 1703 if (domain != NULL) 1704 strlcpy(pr->pr_domainname, domain, 1705 sizeof(pr->pr_domainname)); 1706 if (uuid != NULL) 1707 strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid)); 1708 if (gothid) 1709 pr->pr_hostid = hid; 1710 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { 1711 if (tpr->pr_flags & PR_HOST) 1712 descend = 0; 1713 else { 1714 if (host != NULL) 1715 strlcpy(tpr->pr_hostname, 1716 pr->pr_hostname, 1717 sizeof(tpr->pr_hostname)); 1718 if (domain != NULL) 1719 strlcpy(tpr->pr_domainname, 1720 pr->pr_domainname, 1721 sizeof(tpr->pr_domainname)); 1722 if (uuid != NULL) 1723 strlcpy(tpr->pr_hostuuid, 1724 pr->pr_hostuuid, 1725 sizeof(tpr->pr_hostuuid)); 1726 if (gothid) 1727 tpr->pr_hostid = hid; 1728 } 1729 } 1730 } 1731 pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow; 1732 if ((tallow = ch_allow & ~pr_allow)) 1733 prison_set_allow_locked(pr, tallow, 0); 1734 /* 1735 * Persistent prisons get an extra reference, and prisons losing their 1736 * persist flag lose that reference. Only do this for existing prisons 1737 * for now, so new ones will remain unseen until after the module 1738 * handlers have completed. 1739 */ 1740 born = pr->pr_uref == 0; 1741 if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) { 1742 if (pr_flags & PR_PERSIST) { 1743 pr->pr_ref++; 1744 pr->pr_uref++; 1745 } else { 1746 pr->pr_ref--; 1747 pr->pr_uref--; 1748 } 1749 } 1750 pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags; 1751 mtx_unlock(&pr->pr_mtx); 1752 1753 #ifdef RACCT 1754 if (racct_enable && created) 1755 prison_racct_attach(pr); 1756 #endif 1757 1758 /* Locks may have prevented a complete restriction of child IP 1759 * addresses. If so, allocate some more memory and try again. 1760 */ 1761 #ifdef INET 1762 while (redo_ip4) { 1763 ip4s = pr->pr_ip4s; 1764 ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK); 1765 mtx_lock(&pr->pr_mtx); 1766 redo_ip4 = 0; 1767 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { 1768 #ifdef VIMAGE 1769 if (tpr->pr_flags & PR_VNET) { 1770 descend = 0; 1771 continue; 1772 } 1773 #endif 1774 if (prison_restrict_ip4(tpr, ip4)) { 1775 if (ip4 != NULL) 1776 ip4 = NULL; 1777 else 1778 redo_ip4 = 1; 1779 } 1780 } 1781 mtx_unlock(&pr->pr_mtx); 1782 } 1783 #endif 1784 #ifdef INET6 1785 while (redo_ip6) { 1786 ip6s = pr->pr_ip6s; 1787 ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK); 1788 mtx_lock(&pr->pr_mtx); 1789 redo_ip6 = 0; 1790 FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { 1791 #ifdef VIMAGE 1792 if (tpr->pr_flags & PR_VNET) { 1793 descend = 0; 1794 continue; 1795 } 1796 #endif 1797 if (prison_restrict_ip6(tpr, ip6)) { 1798 if (ip6 != NULL) 1799 ip6 = NULL; 1800 else 1801 redo_ip6 = 1; 1802 } 1803 } 1804 mtx_unlock(&pr->pr_mtx); 1805 } 1806 #endif 1807 1808 /* Let the modules do their work. */ 1809 sx_downgrade(&allprison_lock); 1810 if (born) { 1811 error = osd_jail_call(pr, PR_METHOD_CREATE, opts); 1812 if (error) { 1813 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); 1814 prison_deref(pr, created 1815 ? PD_LIST_SLOCKED 1816 : PD_DEREF | PD_LIST_SLOCKED); 1817 goto done_errmsg; 1818 } 1819 } 1820 error = osd_jail_call(pr, PR_METHOD_SET, opts); 1821 if (error) { 1822 if (born) 1823 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); 1824 prison_deref(pr, created 1825 ? PD_LIST_SLOCKED 1826 : PD_DEREF | PD_LIST_SLOCKED); 1827 goto done_errmsg; 1828 } 1829 1830 /* Attach this process to the prison if requested. */ 1831 if (flags & JAIL_ATTACH) { 1832 mtx_lock(&pr->pr_mtx); 1833 error = do_jail_attach(td, pr); 1834 if (error) { 1835 vfs_opterror(opts, "attach failed"); 1836 if (!created) 1837 prison_deref(pr, PD_DEREF); 1838 goto done_errmsg; 1839 } 1840 } 1841 1842 #ifdef RACCT 1843 if (racct_enable && !created) { 1844 if (!(flags & JAIL_ATTACH)) 1845 sx_sunlock(&allprison_lock); 1846 prison_racct_modify(pr); 1847 if (!(flags & JAIL_ATTACH)) 1848 sx_slock(&allprison_lock); 1849 } 1850 #endif 1851 1852 td->td_retval[0] = pr->pr_id; 1853 1854 /* 1855 * Now that it is all there, drop the temporary reference from existing 1856 * prisons. Or add a reference to newly created persistent prisons 1857 * (which was not done earlier so that the prison would not be publicly 1858 * visible). 1859 */ 1860 if (!created) { 1861 prison_deref(pr, (flags & JAIL_ATTACH) 1862 ? PD_DEREF 1863 : PD_DEREF | PD_LIST_SLOCKED); 1864 } else { 1865 if (pr_flags & PR_PERSIST) { 1866 mtx_lock(&pr->pr_mtx); 1867 pr->pr_ref++; 1868 pr->pr_uref++; 1869 mtx_unlock(&pr->pr_mtx); 1870 } 1871 if (!(flags & JAIL_ATTACH)) 1872 sx_sunlock(&allprison_lock); 1873 } 1874 1875 goto done_free; 1876 1877 done_deref_locked: 1878 prison_deref(pr, created 1879 ? PD_LOCKED | PD_LIST_XLOCKED 1880 : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED); 1881 goto done_releroot; 1882 done_unlock_list: 1883 sx_xunlock(&allprison_lock); 1884 done_releroot: 1885 if (root != NULL) 1886 vrele(root); 1887 done_errmsg: 1888 if (error) { 1889 if (vfs_getopt(opts, "errmsg", (void **)&errmsg, 1890 &errmsg_len) == 0 && errmsg_len > 0) { 1891 errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1; 1892 if (optuio->uio_segflg == UIO_SYSSPACE) 1893 bcopy(errmsg, 1894 optuio->uio_iov[errmsg_pos].iov_base, 1895 errmsg_len); 1896 else 1897 copyout(errmsg, 1898 optuio->uio_iov[errmsg_pos].iov_base, 1899 errmsg_len); 1900 } 1901 } 1902 done_free: 1903 #ifdef INET 1904 free(ip4, M_PRISON); 1905 #endif 1906 #ifdef INET6 1907 free(ip6, M_PRISON); 1908 #endif 1909 if (g_path != NULL) 1910 free(g_path, M_TEMP); 1911 vfs_freeopts(opts); 1912 return (error); 1913 } 1914 1915 /* 1916 * Find the next available prison ID. Return the ID on success, or zero 1917 * on failure. Also set a pointer to the allprison list entry the prison 1918 * should be inserted before. 1919 */ 1920 static int 1921 get_next_prid(struct prison **insprp) 1922 { 1923 struct prison *inspr; 1924 int jid, maxid; 1925 1926 jid = lastprid % JAIL_MAX + 1; 1927 if (TAILQ_EMPTY(&allprison) || 1928 TAILQ_LAST(&allprison, prisonlist)->pr_id < jid) { 1929 /* 1930 * A common case is for all jails to be implicitly numbered, 1931 * which means they'll go on the end of the list, at least 1932 * for the first JAIL_MAX times. 1933 */ 1934 inspr = NULL; 1935 } else { 1936 /* 1937 * Take two passes through the allprison list: first starting 1938 * with the proposed jid, then ending with it. 1939 */ 1940 for (maxid = JAIL_MAX; maxid != 0; ) { 1941 TAILQ_FOREACH(inspr, &allprison, pr_list) { 1942 if (inspr->pr_id < jid) 1943 continue; 1944 if (inspr->pr_id > jid || inspr->pr_ref == 0) { 1945 /* 1946 * Found an opening. This may be a gap 1947 * in the list, or a dead jail with the 1948 * same ID. 1949 */ 1950 maxid = 0; 1951 break; 1952 } 1953 if (++jid > maxid) { 1954 if (lastprid == maxid || lastprid == 0) 1955 { 1956 /* 1957 * The entire legal range 1958 * has been traversed 1959 */ 1960 return 0; 1961 } 1962 /* Try again from the start. */ 1963 jid = 1; 1964 maxid = lastprid; 1965 break; 1966 } 1967 } 1968 if (inspr == NULL) { 1969 /* Found room at the end of the list. */ 1970 break; 1971 } 1972 } 1973 } 1974 *insprp = inspr; 1975 lastprid = jid; 1976 return (jid); 1977 } 1978 1979 /* 1980 * struct jail_get_args { 1981 * struct iovec *iovp; 1982 * unsigned int iovcnt; 1983 * int flags; 1984 * }; 1985 */ 1986 int 1987 sys_jail_get(struct thread *td, struct jail_get_args *uap) 1988 { 1989 struct uio *auio; 1990 int error; 1991 1992 /* Check that we have an even number of iovecs. */ 1993 if (uap->iovcnt & 1) 1994 return (EINVAL); 1995 1996 error = copyinuio(uap->iovp, uap->iovcnt, &auio); 1997 if (error) 1998 return (error); 1999 error = kern_jail_get(td, auio, uap->flags); 2000 if (error == 0) 2001 error = copyout(auio->uio_iov, uap->iovp, 2002 uap->iovcnt * sizeof (struct iovec)); 2003 free(auio, M_IOV); 2004 return (error); 2005 } 2006 2007 int 2008 kern_jail_get(struct thread *td, struct uio *optuio, int flags) 2009 { 2010 struct bool_flags *bf; 2011 struct jailsys_flags *jsf; 2012 struct prison *pr, *mypr; 2013 struct vfsopt *opt; 2014 struct vfsoptlist *opts; 2015 char *errmsg, *name; 2016 int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos; 2017 unsigned f; 2018 2019 if (flags & ~JAIL_GET_MASK) 2020 return (EINVAL); 2021 2022 /* Get the parameter list. */ 2023 error = vfs_buildopts(optuio, &opts); 2024 if (error) 2025 return (error); 2026 errmsg_pos = vfs_getopt_pos(opts, "errmsg"); 2027 mypr = td->td_ucred->cr_prison; 2028 2029 /* 2030 * Find the prison specified by one of: lastjid, jid, name. 2031 */ 2032 sx_slock(&allprison_lock); 2033 error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid)); 2034 if (error == 0) { 2035 TAILQ_FOREACH(pr, &allprison, pr_list) { 2036 if (pr->pr_id > jid && prison_ischild(mypr, pr)) { 2037 mtx_lock(&pr->pr_mtx); 2038 if (pr->pr_ref > 0 && 2039 (pr->pr_uref > 0 || (flags & JAIL_DYING))) 2040 break; 2041 mtx_unlock(&pr->pr_mtx); 2042 } 2043 } 2044 if (pr != NULL) 2045 goto found_prison; 2046 error = ENOENT; 2047 vfs_opterror(opts, "no jail after %d", jid); 2048 goto done_unlock_list; 2049 } else if (error != ENOENT) 2050 goto done_unlock_list; 2051 2052 error = vfs_copyopt(opts, "jid", &jid, sizeof(jid)); 2053 if (error == 0) { 2054 if (jid != 0) { 2055 pr = prison_find_child(mypr, jid); 2056 if (pr != NULL) { 2057 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) { 2058 mtx_unlock(&pr->pr_mtx); 2059 error = ENOENT; 2060 vfs_opterror(opts, "jail %d is dying", 2061 jid); 2062 goto done_unlock_list; 2063 } 2064 goto found_prison; 2065 } 2066 error = ENOENT; 2067 vfs_opterror(opts, "jail %d not found", jid); 2068 goto done_unlock_list; 2069 } 2070 } else if (error != ENOENT) 2071 goto done_unlock_list; 2072 2073 error = vfs_getopt(opts, "name", (void **)&name, &len); 2074 if (error == 0) { 2075 if (len == 0 || name[len - 1] != '\0') { 2076 error = EINVAL; 2077 goto done_unlock_list; 2078 } 2079 pr = prison_find_name(mypr, name); 2080 if (pr != NULL) { 2081 if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) { 2082 mtx_unlock(&pr->pr_mtx); 2083 error = ENOENT; 2084 vfs_opterror(opts, "jail \"%s\" is dying", 2085 name); 2086 goto done_unlock_list; 2087 } 2088 goto found_prison; 2089 } 2090 error = ENOENT; 2091 vfs_opterror(opts, "jail \"%s\" not found", name); 2092 goto done_unlock_list; 2093 } else if (error != ENOENT) 2094 goto done_unlock_list; 2095 2096 vfs_opterror(opts, "no jail specified"); 2097 error = ENOENT; 2098 goto done_unlock_list; 2099 2100 found_prison: 2101 /* Get the parameters of the prison. */ 2102 pr->pr_ref++; 2103 locked = PD_LOCKED; 2104 td->td_retval[0] = pr->pr_id; 2105 error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id)); 2106 if (error != 0 && error != ENOENT) 2107 goto done_deref; 2108 i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id; 2109 error = vfs_setopt(opts, "parent", &i, sizeof(i)); 2110 if (error != 0 && error != ENOENT) 2111 goto done_deref; 2112 error = vfs_setopts(opts, "name", prison_name(mypr, pr)); 2113 if (error != 0 && error != ENOENT) 2114 goto done_deref; 2115 error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id, 2116 sizeof(pr->pr_cpuset->cs_id)); 2117 if (error != 0 && error != ENOENT) 2118 goto done_deref; 2119 error = vfs_setopts(opts, "path", prison_path(mypr, pr)); 2120 if (error != 0 && error != ENOENT) 2121 goto done_deref; 2122 #ifdef INET 2123 error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4, 2124 pr->pr_ip4s * sizeof(*pr->pr_ip4)); 2125 if (error != 0 && error != ENOENT) 2126 goto done_deref; 2127 #endif 2128 #ifdef INET6 2129 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6, 2130 pr->pr_ip6s * sizeof(*pr->pr_ip6)); 2131 if (error != 0 && error != ENOENT) 2132 goto done_deref; 2133 #endif 2134 error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel, 2135 sizeof(pr->pr_securelevel)); 2136 if (error != 0 && error != ENOENT) 2137 goto done_deref; 2138 error = vfs_setopt(opts, "children.cur", &pr->pr_childcount, 2139 sizeof(pr->pr_childcount)); 2140 if (error != 0 && error != ENOENT) 2141 goto done_deref; 2142 error = vfs_setopt(opts, "children.max", &pr->pr_childmax, 2143 sizeof(pr->pr_childmax)); 2144 if (error != 0 && error != ENOENT) 2145 goto done_deref; 2146 error = vfs_setopts(opts, "host.hostname", pr->pr_hostname); 2147 if (error != 0 && error != ENOENT) 2148 goto done_deref; 2149 error = vfs_setopts(opts, "host.domainname", pr->pr_domainname); 2150 if (error != 0 && error != ENOENT) 2151 goto done_deref; 2152 error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid); 2153 if (error != 0 && error != ENOENT) 2154 goto done_deref; 2155 #ifdef COMPAT_FREEBSD32 2156 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 2157 uint32_t hid32 = pr->pr_hostid; 2158 2159 error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32)); 2160 } else 2161 #endif 2162 error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid, 2163 sizeof(pr->pr_hostid)); 2164 if (error != 0 && error != ENOENT) 2165 goto done_deref; 2166 error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs, 2167 sizeof(pr->pr_enforce_statfs)); 2168 if (error != 0 && error != ENOENT) 2169 goto done_deref; 2170 error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum, 2171 sizeof(pr->pr_devfs_rsnum)); 2172 if (error != 0 && error != ENOENT) 2173 goto done_deref; 2174 for (bf = pr_flag_bool; 2175 bf < pr_flag_bool + nitems(pr_flag_bool); 2176 bf++) { 2177 i = (pr->pr_flags & bf->flag) ? 1 : 0; 2178 error = vfs_setopt(opts, bf->name, &i, sizeof(i)); 2179 if (error != 0 && error != ENOENT) 2180 goto done_deref; 2181 i = !i; 2182 error = vfs_setopt(opts, bf->noname, &i, sizeof(i)); 2183 if (error != 0 && error != ENOENT) 2184 goto done_deref; 2185 } 2186 for (jsf = pr_flag_jailsys; 2187 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys); 2188 jsf++) { 2189 f = pr->pr_flags & (jsf->disable | jsf->new); 2190 i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE 2191 : (f == jsf->new) ? JAIL_SYS_NEW 2192 : JAIL_SYS_INHERIT; 2193 error = vfs_setopt(opts, jsf->name, &i, sizeof(i)); 2194 if (error != 0 && error != ENOENT) 2195 goto done_deref; 2196 } 2197 for (bf = pr_flag_allow; 2198 bf < pr_flag_allow + nitems(pr_flag_allow) && 2199 atomic_load_int(&bf->flag) != 0; 2200 bf++) { 2201 i = (pr->pr_allow & bf->flag) ? 1 : 0; 2202 error = vfs_setopt(opts, bf->name, &i, sizeof(i)); 2203 if (error != 0 && error != ENOENT) 2204 goto done_deref; 2205 i = !i; 2206 error = vfs_setopt(opts, bf->noname, &i, sizeof(i)); 2207 if (error != 0 && error != ENOENT) 2208 goto done_deref; 2209 } 2210 i = (pr->pr_uref == 0); 2211 error = vfs_setopt(opts, "dying", &i, sizeof(i)); 2212 if (error != 0 && error != ENOENT) 2213 goto done_deref; 2214 i = !i; 2215 error = vfs_setopt(opts, "nodying", &i, sizeof(i)); 2216 if (error != 0 && error != ENOENT) 2217 goto done_deref; 2218 error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate, 2219 sizeof(pr->pr_osreldate)); 2220 if (error != 0 && error != ENOENT) 2221 goto done_deref; 2222 error = vfs_setopts(opts, "osrelease", pr->pr_osrelease); 2223 if (error != 0 && error != ENOENT) 2224 goto done_deref; 2225 2226 /* Get the module parameters. */ 2227 mtx_unlock(&pr->pr_mtx); 2228 locked = 0; 2229 error = osd_jail_call(pr, PR_METHOD_GET, opts); 2230 if (error) 2231 goto done_deref; 2232 prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED); 2233 2234 /* By now, all parameters should have been noted. */ 2235 TAILQ_FOREACH(opt, opts, link) { 2236 if (!opt->seen && strcmp(opt->name, "errmsg")) { 2237 error = EINVAL; 2238 vfs_opterror(opts, "unknown parameter: %s", opt->name); 2239 goto done_errmsg; 2240 } 2241 } 2242 2243 /* Write the fetched parameters back to userspace. */ 2244 error = 0; 2245 TAILQ_FOREACH(opt, opts, link) { 2246 if (opt->pos >= 0 && opt->pos != errmsg_pos) { 2247 pos = 2 * opt->pos + 1; 2248 optuio->uio_iov[pos].iov_len = opt->len; 2249 if (opt->value != NULL) { 2250 if (optuio->uio_segflg == UIO_SYSSPACE) { 2251 bcopy(opt->value, 2252 optuio->uio_iov[pos].iov_base, 2253 opt->len); 2254 } else { 2255 error = copyout(opt->value, 2256 optuio->uio_iov[pos].iov_base, 2257 opt->len); 2258 if (error) 2259 break; 2260 } 2261 } 2262 } 2263 } 2264 goto done_errmsg; 2265 2266 done_deref: 2267 prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED); 2268 goto done_errmsg; 2269 2270 done_unlock_list: 2271 sx_sunlock(&allprison_lock); 2272 done_errmsg: 2273 if (error && errmsg_pos >= 0) { 2274 vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len); 2275 errmsg_pos = 2 * errmsg_pos + 1; 2276 if (errmsg_len > 0) { 2277 if (optuio->uio_segflg == UIO_SYSSPACE) 2278 bcopy(errmsg, 2279 optuio->uio_iov[errmsg_pos].iov_base, 2280 errmsg_len); 2281 else 2282 copyout(errmsg, 2283 optuio->uio_iov[errmsg_pos].iov_base, 2284 errmsg_len); 2285 } 2286 } 2287 vfs_freeopts(opts); 2288 return (error); 2289 } 2290 2291 /* 2292 * struct jail_remove_args { 2293 * int jid; 2294 * }; 2295 */ 2296 int 2297 sys_jail_remove(struct thread *td, struct jail_remove_args *uap) 2298 { 2299 struct prison *pr, *cpr, *lpr, *tpr; 2300 int descend, error; 2301 2302 error = priv_check(td, PRIV_JAIL_REMOVE); 2303 if (error) 2304 return (error); 2305 2306 sx_xlock(&allprison_lock); 2307 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid); 2308 if (pr == NULL) { 2309 sx_xunlock(&allprison_lock); 2310 return (EINVAL); 2311 } 2312 2313 /* Remove all descendants of this prison, then remove this prison. */ 2314 pr->pr_ref++; 2315 if (!LIST_EMPTY(&pr->pr_children)) { 2316 mtx_unlock(&pr->pr_mtx); 2317 lpr = NULL; 2318 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) { 2319 mtx_lock(&cpr->pr_mtx); 2320 if (cpr->pr_ref > 0) { 2321 tpr = cpr; 2322 cpr->pr_ref++; 2323 } else { 2324 /* Already removed - do not do it again. */ 2325 tpr = NULL; 2326 } 2327 mtx_unlock(&cpr->pr_mtx); 2328 if (lpr != NULL) { 2329 mtx_lock(&lpr->pr_mtx); 2330 prison_remove_one(lpr); 2331 sx_xlock(&allprison_lock); 2332 } 2333 lpr = tpr; 2334 } 2335 if (lpr != NULL) { 2336 mtx_lock(&lpr->pr_mtx); 2337 prison_remove_one(lpr); 2338 sx_xlock(&allprison_lock); 2339 } 2340 mtx_lock(&pr->pr_mtx); 2341 } 2342 prison_remove_one(pr); 2343 return (0); 2344 } 2345 2346 static void 2347 prison_remove_one(struct prison *pr) 2348 { 2349 struct proc *p; 2350 int deuref; 2351 2352 /* If the prison was persistent, it is not anymore. */ 2353 deuref = 0; 2354 if (pr->pr_flags & PR_PERSIST) { 2355 pr->pr_ref--; 2356 deuref = PD_DEUREF; 2357 pr->pr_flags &= ~PR_PERSIST; 2358 } 2359 2360 /* 2361 * jail_remove added a reference. If that's the only one, remove 2362 * the prison now. 2363 */ 2364 KASSERT(pr->pr_ref > 0, 2365 ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id)); 2366 if (pr->pr_ref == 1) { 2367 prison_deref(pr, 2368 deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED); 2369 return; 2370 } 2371 2372 mtx_unlock(&pr->pr_mtx); 2373 sx_xunlock(&allprison_lock); 2374 /* 2375 * Kill all processes unfortunate enough to be attached to this prison. 2376 */ 2377 sx_slock(&allproc_lock); 2378 FOREACH_PROC_IN_SYSTEM(p) { 2379 PROC_LOCK(p); 2380 if (p->p_state != PRS_NEW && p->p_ucred && 2381 p->p_ucred->cr_prison == pr) 2382 kern_psignal(p, SIGKILL); 2383 PROC_UNLOCK(p); 2384 } 2385 sx_sunlock(&allproc_lock); 2386 /* Remove the temporary reference added by jail_remove. */ 2387 prison_deref(pr, deuref | PD_DEREF); 2388 } 2389 2390 /* 2391 * struct jail_attach_args { 2392 * int jid; 2393 * }; 2394 */ 2395 int 2396 sys_jail_attach(struct thread *td, struct jail_attach_args *uap) 2397 { 2398 struct prison *pr; 2399 int error; 2400 2401 error = priv_check(td, PRIV_JAIL_ATTACH); 2402 if (error) 2403 return (error); 2404 2405 /* 2406 * Start with exclusive hold on allprison_lock to ensure that a possible 2407 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove. 2408 * But then immediately downgrade it since we don't need to stop 2409 * readers. 2410 */ 2411 sx_xlock(&allprison_lock); 2412 sx_downgrade(&allprison_lock); 2413 pr = prison_find_child(td->td_ucred->cr_prison, uap->jid); 2414 if (pr == NULL) { 2415 sx_sunlock(&allprison_lock); 2416 return (EINVAL); 2417 } 2418 2419 /* 2420 * Do not allow a process to attach to a prison that is not 2421 * considered to be "alive". 2422 */ 2423 if (pr->pr_uref == 0) { 2424 mtx_unlock(&pr->pr_mtx); 2425 sx_sunlock(&allprison_lock); 2426 return (EINVAL); 2427 } 2428 2429 return (do_jail_attach(td, pr)); 2430 } 2431 2432 static int 2433 do_jail_attach(struct thread *td, struct prison *pr) 2434 { 2435 struct proc *p; 2436 struct ucred *newcred, *oldcred; 2437 int error; 2438 2439 /* 2440 * XXX: Note that there is a slight race here if two threads 2441 * in the same privileged process attempt to attach to two 2442 * different jails at the same time. It is important for 2443 * user processes not to do this, or they might end up with 2444 * a process root from one prison, but attached to the jail 2445 * of another. 2446 */ 2447 pr->pr_ref++; 2448 pr->pr_uref++; 2449 mtx_unlock(&pr->pr_mtx); 2450 2451 /* Let modules do whatever they need to prepare for attaching. */ 2452 error = osd_jail_call(pr, PR_METHOD_ATTACH, td); 2453 if (error) { 2454 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED); 2455 return (error); 2456 } 2457 sx_sunlock(&allprison_lock); 2458 2459 /* 2460 * Reparent the newly attached process to this jail. 2461 */ 2462 p = td->td_proc; 2463 error = cpuset_setproc_update_set(p, pr->pr_cpuset); 2464 if (error) 2465 goto e_revert_osd; 2466 2467 vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY); 2468 if ((error = change_dir(pr->pr_root, td)) != 0) 2469 goto e_unlock; 2470 #ifdef MAC 2471 if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root))) 2472 goto e_unlock; 2473 #endif 2474 VOP_UNLOCK(pr->pr_root); 2475 if ((error = pwd_chroot(td, pr->pr_root))) 2476 goto e_revert_osd; 2477 2478 newcred = crget(); 2479 PROC_LOCK(p); 2480 oldcred = crcopysafe(p, newcred); 2481 newcred->cr_prison = pr; 2482 proc_set_cred(p, newcred); 2483 setsugid(p); 2484 #ifdef RACCT 2485 racct_proc_ucred_changed(p, oldcred, newcred); 2486 crhold(newcred); 2487 #endif 2488 PROC_UNLOCK(p); 2489 #ifdef RCTL 2490 rctl_proc_ucred_changed(p, newcred); 2491 crfree(newcred); 2492 #endif 2493 prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF); 2494 crfree(oldcred); 2495 return (0); 2496 2497 e_unlock: 2498 VOP_UNLOCK(pr->pr_root); 2499 e_revert_osd: 2500 /* Tell modules this thread is still in its old jail after all. */ 2501 sx_slock(&allprison_lock); 2502 (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td); 2503 prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED); 2504 return (error); 2505 } 2506 2507 /* 2508 * Returns a locked prison instance, or NULL on failure. 2509 */ 2510 struct prison * 2511 prison_find(int prid) 2512 { 2513 struct prison *pr; 2514 2515 sx_assert(&allprison_lock, SX_LOCKED); 2516 TAILQ_FOREACH(pr, &allprison, pr_list) { 2517 if (pr->pr_id == prid) { 2518 mtx_lock(&pr->pr_mtx); 2519 if (pr->pr_ref > 0) 2520 return (pr); 2521 /* 2522 * Any active prison with the same ID would have 2523 * been inserted before a dead one. 2524 */ 2525 mtx_unlock(&pr->pr_mtx); 2526 break; 2527 } 2528 if (pr->pr_id > prid) 2529 break; 2530 } 2531 return (NULL); 2532 } 2533 2534 /* 2535 * Find a prison that is a descendant of mypr. Returns a locked prison or NULL. 2536 */ 2537 struct prison * 2538 prison_find_child(struct prison *mypr, int prid) 2539 { 2540 struct prison *pr; 2541 int descend; 2542 2543 sx_assert(&allprison_lock, SX_LOCKED); 2544 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) { 2545 if (pr->pr_id == prid) { 2546 mtx_lock(&pr->pr_mtx); 2547 if (pr->pr_ref > 0) 2548 return (pr); 2549 mtx_unlock(&pr->pr_mtx); 2550 } 2551 } 2552 return (NULL); 2553 } 2554 2555 /* 2556 * Look for the name relative to mypr. Returns a locked prison or NULL. 2557 */ 2558 struct prison * 2559 prison_find_name(struct prison *mypr, const char *name) 2560 { 2561 struct prison *pr, *deadpr; 2562 size_t mylen; 2563 int descend; 2564 2565 sx_assert(&allprison_lock, SX_LOCKED); 2566 mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1; 2567 again: 2568 deadpr = NULL; 2569 FOREACH_PRISON_DESCENDANT(mypr, pr, descend) { 2570 if (!strcmp(pr->pr_name + mylen, name)) { 2571 mtx_lock(&pr->pr_mtx); 2572 if (pr->pr_ref > 0) { 2573 if (pr->pr_uref > 0) 2574 return (pr); 2575 deadpr = pr; 2576 } 2577 mtx_unlock(&pr->pr_mtx); 2578 } 2579 } 2580 /* There was no valid prison - perhaps there was a dying one. */ 2581 if (deadpr != NULL) { 2582 mtx_lock(&deadpr->pr_mtx); 2583 if (deadpr->pr_ref == 0) { 2584 mtx_unlock(&deadpr->pr_mtx); 2585 goto again; 2586 } 2587 } 2588 return (deadpr); 2589 } 2590 2591 /* 2592 * See if a prison has the specific flag set. The prison should be locked, 2593 * unless checking for flags that are only set at jail creation (such as 2594 * PR_IP4 and PR_IP6), or only the single bit is examined, without regard 2595 * to any other prison data. 2596 */ 2597 int 2598 prison_flag(struct ucred *cred, unsigned flag) 2599 { 2600 2601 return (cred->cr_prison->pr_flags & flag); 2602 } 2603 2604 int 2605 prison_allow(struct ucred *cred, unsigned flag) 2606 { 2607 2608 return ((cred->cr_prison->pr_allow & flag) != 0); 2609 } 2610 2611 /* 2612 * Remove a prison reference. If that was the last reference, remove the 2613 * prison itself - but not in this context in case there are locks held. 2614 */ 2615 void 2616 prison_free_locked(struct prison *pr) 2617 { 2618 int ref; 2619 2620 mtx_assert(&pr->pr_mtx, MA_OWNED); 2621 ref = --pr->pr_ref; 2622 mtx_unlock(&pr->pr_mtx); 2623 if (ref == 0) 2624 taskqueue_enqueue(taskqueue_thread, &pr->pr_task); 2625 } 2626 2627 void 2628 prison_free(struct prison *pr) 2629 { 2630 2631 mtx_lock(&pr->pr_mtx); 2632 prison_free_locked(pr); 2633 } 2634 2635 /* 2636 * Complete a call to either prison_free or prison_proc_free. 2637 */ 2638 static void 2639 prison_complete(void *context, int pending) 2640 { 2641 struct prison *pr = context; 2642 2643 sx_xlock(&allprison_lock); 2644 mtx_lock(&pr->pr_mtx); 2645 prison_deref(pr, pr->pr_uref 2646 ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED 2647 : PD_LOCKED | PD_LIST_XLOCKED); 2648 } 2649 2650 /* 2651 * Remove a prison reference (usually). This internal version assumes no 2652 * mutexes are held, except perhaps the prison itself. If there are no more 2653 * references, release and delist the prison. On completion, the prison lock 2654 * and the allprison lock are both unlocked. 2655 */ 2656 static void 2657 prison_deref(struct prison *pr, int flags) 2658 { 2659 struct prison *ppr, *tpr; 2660 int ref, lasturef; 2661 2662 if (!(flags & PD_LOCKED)) 2663 mtx_lock(&pr->pr_mtx); 2664 for (;;) { 2665 if (flags & PD_DEUREF) { 2666 KASSERT(pr->pr_uref > 0, 2667 ("prison_deref PD_DEUREF on a dead prison (jid=%d)", 2668 pr->pr_id)); 2669 pr->pr_uref--; 2670 lasturef = pr->pr_uref == 0; 2671 if (lasturef) 2672 pr->pr_ref++; 2673 KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0")); 2674 } else 2675 lasturef = 0; 2676 if (flags & PD_DEREF) { 2677 KASSERT(pr->pr_ref > 0, 2678 ("prison_deref PD_DEREF on a dead prison (jid=%d)", 2679 pr->pr_id)); 2680 pr->pr_ref--; 2681 } 2682 ref = pr->pr_ref; 2683 mtx_unlock(&pr->pr_mtx); 2684 2685 /* 2686 * Tell the modules if the last user reference was removed 2687 * (even it sticks around in dying state). 2688 */ 2689 if (lasturef) { 2690 if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) { 2691 if (ref > 1) { 2692 sx_slock(&allprison_lock); 2693 flags |= PD_LIST_SLOCKED; 2694 } else { 2695 sx_xlock(&allprison_lock); 2696 flags |= PD_LIST_XLOCKED; 2697 } 2698 } 2699 (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); 2700 mtx_lock(&pr->pr_mtx); 2701 ref = --pr->pr_ref; 2702 mtx_unlock(&pr->pr_mtx); 2703 } 2704 2705 /* If the prison still has references, nothing else to do. */ 2706 if (ref > 0) { 2707 if (flags & PD_LIST_SLOCKED) 2708 sx_sunlock(&allprison_lock); 2709 else if (flags & PD_LIST_XLOCKED) 2710 sx_xunlock(&allprison_lock); 2711 return; 2712 } 2713 2714 if (flags & PD_LIST_SLOCKED) { 2715 if (!sx_try_upgrade(&allprison_lock)) { 2716 sx_sunlock(&allprison_lock); 2717 sx_xlock(&allprison_lock); 2718 } 2719 } else if (!(flags & PD_LIST_XLOCKED)) 2720 sx_xlock(&allprison_lock); 2721 2722 TAILQ_REMOVE(&allprison, pr, pr_list); 2723 LIST_REMOVE(pr, pr_sibling); 2724 ppr = pr->pr_parent; 2725 for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent) 2726 tpr->pr_childcount--; 2727 sx_xunlock(&allprison_lock); 2728 2729 #ifdef VIMAGE 2730 if (pr->pr_vnet != ppr->pr_vnet) 2731 vnet_destroy(pr->pr_vnet); 2732 #endif 2733 if (pr->pr_root != NULL) 2734 vrele(pr->pr_root); 2735 mtx_destroy(&pr->pr_mtx); 2736 #ifdef INET 2737 free(pr->pr_ip4, M_PRISON); 2738 #endif 2739 #ifdef INET6 2740 free(pr->pr_ip6, M_PRISON); 2741 #endif 2742 if (pr->pr_cpuset != NULL) 2743 cpuset_rel(pr->pr_cpuset); 2744 osd_jail_exit(pr); 2745 #ifdef RACCT 2746 if (racct_enable) 2747 prison_racct_detach(pr); 2748 #endif 2749 free(pr, M_PRISON); 2750 2751 /* Removing a prison frees a reference on its parent. */ 2752 pr = ppr; 2753 mtx_lock(&pr->pr_mtx); 2754 flags = PD_DEREF | PD_DEUREF; 2755 } 2756 } 2757 2758 void 2759 prison_hold_locked(struct prison *pr) 2760 { 2761 2762 mtx_assert(&pr->pr_mtx, MA_OWNED); 2763 KASSERT(pr->pr_ref > 0, 2764 ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id)); 2765 pr->pr_ref++; 2766 } 2767 2768 void 2769 prison_hold(struct prison *pr) 2770 { 2771 2772 mtx_lock(&pr->pr_mtx); 2773 prison_hold_locked(pr); 2774 mtx_unlock(&pr->pr_mtx); 2775 } 2776 2777 void 2778 prison_proc_hold(struct prison *pr) 2779 { 2780 2781 mtx_lock(&pr->pr_mtx); 2782 KASSERT(pr->pr_uref > 0, 2783 ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id)); 2784 pr->pr_uref++; 2785 mtx_unlock(&pr->pr_mtx); 2786 } 2787 2788 void 2789 prison_proc_free(struct prison *pr) 2790 { 2791 2792 mtx_lock(&pr->pr_mtx); 2793 KASSERT(pr->pr_uref > 0, 2794 ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id)); 2795 if (pr->pr_uref > 1) 2796 pr->pr_uref--; 2797 else { 2798 /* 2799 * Don't remove the last user reference in this context, which 2800 * is expected to be a process that is not only locked, but 2801 * also half dead. 2802 */ 2803 pr->pr_ref++; 2804 mtx_unlock(&pr->pr_mtx); 2805 taskqueue_enqueue(taskqueue_thread, &pr->pr_task); 2806 return; 2807 } 2808 mtx_unlock(&pr->pr_mtx); 2809 } 2810 2811 /* 2812 * Set or clear a permission bit in the pr_allow field, passing restrictions 2813 * (cleared permission) down to child jails. 2814 */ 2815 void 2816 prison_set_allow(struct ucred *cred, unsigned flag, int enable) 2817 { 2818 struct prison *pr; 2819 2820 pr = cred->cr_prison; 2821 sx_slock(&allprison_lock); 2822 mtx_lock(&pr->pr_mtx); 2823 prison_set_allow_locked(pr, flag, enable); 2824 mtx_unlock(&pr->pr_mtx); 2825 sx_sunlock(&allprison_lock); 2826 } 2827 2828 static void 2829 prison_set_allow_locked(struct prison *pr, unsigned flag, int enable) 2830 { 2831 struct prison *cpr; 2832 int descend; 2833 2834 if (enable != 0) 2835 pr->pr_allow |= flag; 2836 else { 2837 pr->pr_allow &= ~flag; 2838 FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend) 2839 cpr->pr_allow &= ~flag; 2840 } 2841 } 2842 2843 /* 2844 * Check if a jail supports the given address family. 2845 * 2846 * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT 2847 * if not. 2848 */ 2849 int 2850 prison_check_af(struct ucred *cred, int af) 2851 { 2852 struct prison *pr; 2853 int error; 2854 2855 KASSERT(cred != NULL, ("%s: cred is NULL", __func__)); 2856 2857 pr = cred->cr_prison; 2858 #ifdef VIMAGE 2859 /* Prisons with their own network stack are not limited. */ 2860 if (prison_owns_vnet(cred)) 2861 return (0); 2862 #endif 2863 2864 error = 0; 2865 switch (af) 2866 { 2867 #ifdef INET 2868 case AF_INET: 2869 if (pr->pr_flags & PR_IP4) 2870 { 2871 mtx_lock(&pr->pr_mtx); 2872 if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL) 2873 error = EAFNOSUPPORT; 2874 mtx_unlock(&pr->pr_mtx); 2875 } 2876 break; 2877 #endif 2878 #ifdef INET6 2879 case AF_INET6: 2880 if (pr->pr_flags & PR_IP6) 2881 { 2882 mtx_lock(&pr->pr_mtx); 2883 if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL) 2884 error = EAFNOSUPPORT; 2885 mtx_unlock(&pr->pr_mtx); 2886 } 2887 break; 2888 #endif 2889 case AF_LOCAL: 2890 case AF_ROUTE: 2891 break; 2892 default: 2893 if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF)) 2894 error = EAFNOSUPPORT; 2895 } 2896 return (error); 2897 } 2898 2899 /* 2900 * Check if given address belongs to the jail referenced by cred (wrapper to 2901 * prison_check_ip[46]). 2902 * 2903 * Returns 0 if jail doesn't restrict the address family or if address belongs 2904 * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if 2905 * the jail doesn't allow the address family. IPv4 Address passed in in NBO. 2906 */ 2907 int 2908 prison_if(struct ucred *cred, const struct sockaddr *sa) 2909 { 2910 #ifdef INET 2911 const struct sockaddr_in *sai; 2912 #endif 2913 #ifdef INET6 2914 const struct sockaddr_in6 *sai6; 2915 #endif 2916 int error; 2917 2918 KASSERT(cred != NULL, ("%s: cred is NULL", __func__)); 2919 KASSERT(sa != NULL, ("%s: sa is NULL", __func__)); 2920 2921 #ifdef VIMAGE 2922 if (prison_owns_vnet(cred)) 2923 return (0); 2924 #endif 2925 2926 error = 0; 2927 switch (sa->sa_family) 2928 { 2929 #ifdef INET 2930 case AF_INET: 2931 sai = (const struct sockaddr_in *)sa; 2932 error = prison_check_ip4(cred, &sai->sin_addr); 2933 break; 2934 #endif 2935 #ifdef INET6 2936 case AF_INET6: 2937 sai6 = (const struct sockaddr_in6 *)sa; 2938 error = prison_check_ip6(cred, &sai6->sin6_addr); 2939 break; 2940 #endif 2941 default: 2942 if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF)) 2943 error = EAFNOSUPPORT; 2944 } 2945 return (error); 2946 } 2947 2948 /* 2949 * Return 0 if jails permit p1 to frob p2, otherwise ESRCH. 2950 */ 2951 int 2952 prison_check(struct ucred *cred1, struct ucred *cred2) 2953 { 2954 2955 return ((cred1->cr_prison == cred2->cr_prison || 2956 prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH); 2957 } 2958 2959 /* 2960 * Return 1 if p2 is a child of p1, otherwise 0. 2961 */ 2962 int 2963 prison_ischild(struct prison *pr1, struct prison *pr2) 2964 { 2965 2966 for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent) 2967 if (pr1 == pr2) 2968 return (1); 2969 return (0); 2970 } 2971 2972 /* 2973 * Return 1 if the passed credential is in a jail and that jail does not 2974 * have its own virtual network stack, otherwise 0. 2975 */ 2976 int 2977 jailed_without_vnet(struct ucred *cred) 2978 { 2979 2980 if (!jailed(cred)) 2981 return (0); 2982 #ifdef VIMAGE 2983 if (prison_owns_vnet(cred)) 2984 return (0); 2985 #endif 2986 2987 return (1); 2988 } 2989 2990 /* 2991 * Return the correct hostname (domainname, et al) for the passed credential. 2992 */ 2993 void 2994 getcredhostname(struct ucred *cred, char *buf, size_t size) 2995 { 2996 struct prison *pr; 2997 2998 /* 2999 * A NULL credential can be used to shortcut to the physical 3000 * system's hostname. 3001 */ 3002 pr = (cred != NULL) ? cred->cr_prison : &prison0; 3003 mtx_lock(&pr->pr_mtx); 3004 strlcpy(buf, pr->pr_hostname, size); 3005 mtx_unlock(&pr->pr_mtx); 3006 } 3007 3008 void 3009 getcreddomainname(struct ucred *cred, char *buf, size_t size) 3010 { 3011 3012 mtx_lock(&cred->cr_prison->pr_mtx); 3013 strlcpy(buf, cred->cr_prison->pr_domainname, size); 3014 mtx_unlock(&cred->cr_prison->pr_mtx); 3015 } 3016 3017 void 3018 getcredhostuuid(struct ucred *cred, char *buf, size_t size) 3019 { 3020 3021 mtx_lock(&cred->cr_prison->pr_mtx); 3022 strlcpy(buf, cred->cr_prison->pr_hostuuid, size); 3023 mtx_unlock(&cred->cr_prison->pr_mtx); 3024 } 3025 3026 void 3027 getcredhostid(struct ucred *cred, unsigned long *hostid) 3028 { 3029 3030 mtx_lock(&cred->cr_prison->pr_mtx); 3031 *hostid = cred->cr_prison->pr_hostid; 3032 mtx_unlock(&cred->cr_prison->pr_mtx); 3033 } 3034 3035 void 3036 getjailname(struct ucred *cred, char *name, size_t len) 3037 { 3038 3039 mtx_lock(&cred->cr_prison->pr_mtx); 3040 strlcpy(name, cred->cr_prison->pr_name, len); 3041 mtx_unlock(&cred->cr_prison->pr_mtx); 3042 } 3043 3044 #ifdef VIMAGE 3045 /* 3046 * Determine whether the prison represented by cred owns 3047 * its vnet rather than having it inherited. 3048 * 3049 * Returns 1 in case the prison owns the vnet, 0 otherwise. 3050 */ 3051 int 3052 prison_owns_vnet(struct ucred *cred) 3053 { 3054 3055 /* 3056 * vnets cannot be added/removed after jail creation, 3057 * so no need to lock here. 3058 */ 3059 return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0); 3060 } 3061 #endif 3062 3063 /* 3064 * Determine whether the subject represented by cred can "see" 3065 * status of a mount point. 3066 * Returns: 0 for permitted, ENOENT otherwise. 3067 * XXX: This function should be called cr_canseemount() and should be 3068 * placed in kern_prot.c. 3069 */ 3070 int 3071 prison_canseemount(struct ucred *cred, struct mount *mp) 3072 { 3073 struct prison *pr; 3074 struct statfs *sp; 3075 size_t len; 3076 3077 pr = cred->cr_prison; 3078 if (pr->pr_enforce_statfs == 0) 3079 return (0); 3080 if (pr->pr_root->v_mount == mp) 3081 return (0); 3082 if (pr->pr_enforce_statfs == 2) 3083 return (ENOENT); 3084 /* 3085 * If jail's chroot directory is set to "/" we should be able to see 3086 * all mount-points from inside a jail. 3087 * This is ugly check, but this is the only situation when jail's 3088 * directory ends with '/'. 3089 */ 3090 if (strcmp(pr->pr_path, "/") == 0) 3091 return (0); 3092 len = strlen(pr->pr_path); 3093 sp = &mp->mnt_stat; 3094 if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0) 3095 return (ENOENT); 3096 /* 3097 * Be sure that we don't have situation where jail's root directory 3098 * is "/some/path" and mount point is "/some/pathpath". 3099 */ 3100 if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/') 3101 return (ENOENT); 3102 return (0); 3103 } 3104 3105 void 3106 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp) 3107 { 3108 char jpath[MAXPATHLEN]; 3109 struct prison *pr; 3110 size_t len; 3111 3112 pr = cred->cr_prison; 3113 if (pr->pr_enforce_statfs == 0) 3114 return; 3115 if (prison_canseemount(cred, mp) != 0) { 3116 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 3117 strlcpy(sp->f_mntonname, "[restricted]", 3118 sizeof(sp->f_mntonname)); 3119 return; 3120 } 3121 if (pr->pr_root->v_mount == mp) { 3122 /* 3123 * Clear current buffer data, so we are sure nothing from 3124 * the valid path left there. 3125 */ 3126 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 3127 *sp->f_mntonname = '/'; 3128 return; 3129 } 3130 /* 3131 * If jail's chroot directory is set to "/" we should be able to see 3132 * all mount-points from inside a jail. 3133 */ 3134 if (strcmp(pr->pr_path, "/") == 0) 3135 return; 3136 len = strlen(pr->pr_path); 3137 strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath)); 3138 /* 3139 * Clear current buffer data, so we are sure nothing from 3140 * the valid path left there. 3141 */ 3142 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 3143 if (*jpath == '\0') { 3144 /* Should never happen. */ 3145 *sp->f_mntonname = '/'; 3146 } else { 3147 strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname)); 3148 } 3149 } 3150 3151 /* 3152 * Check with permission for a specific privilege is granted within jail. We 3153 * have a specific list of accepted privileges; the rest are denied. 3154 */ 3155 int 3156 prison_priv_check(struct ucred *cred, int priv) 3157 { 3158 struct prison *pr; 3159 int error; 3160 3161 /* 3162 * Some policies have custom handlers. This routine should not be 3163 * called for them. See priv_check_cred(). 3164 */ 3165 switch (priv) { 3166 case PRIV_VFS_LOOKUP: 3167 case PRIV_VFS_GENERATION: 3168 KASSERT(0, ("prison_priv_check instead of a custom handler " 3169 "called for %d\n", priv)); 3170 } 3171 3172 if (!jailed(cred)) 3173 return (0); 3174 3175 #ifdef VIMAGE 3176 /* 3177 * Privileges specific to prisons with a virtual network stack. 3178 * There might be a duplicate entry here in case the privilege 3179 * is only granted conditionally in the legacy jail case. 3180 */ 3181 switch (priv) { 3182 #ifdef notyet 3183 /* 3184 * NFS-specific privileges. 3185 */ 3186 case PRIV_NFS_DAEMON: 3187 case PRIV_NFS_LOCKD: 3188 #endif 3189 /* 3190 * Network stack privileges. 3191 */ 3192 case PRIV_NET_BRIDGE: 3193 case PRIV_NET_GRE: 3194 case PRIV_NET_BPF: 3195 case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */ 3196 case PRIV_NET_ROUTE: 3197 case PRIV_NET_TAP: 3198 case PRIV_NET_SETIFMTU: 3199 case PRIV_NET_SETIFFLAGS: 3200 case PRIV_NET_SETIFCAP: 3201 case PRIV_NET_SETIFDESCR: 3202 case PRIV_NET_SETIFNAME : 3203 case PRIV_NET_SETIFMETRIC: 3204 case PRIV_NET_SETIFPHYS: 3205 case PRIV_NET_SETIFMAC: 3206 case PRIV_NET_SETLANPCP: 3207 case PRIV_NET_ADDMULTI: 3208 case PRIV_NET_DELMULTI: 3209 case PRIV_NET_HWIOCTL: 3210 case PRIV_NET_SETLLADDR: 3211 case PRIV_NET_ADDIFGROUP: 3212 case PRIV_NET_DELIFGROUP: 3213 case PRIV_NET_IFCREATE: 3214 case PRIV_NET_IFDESTROY: 3215 case PRIV_NET_ADDIFADDR: 3216 case PRIV_NET_DELIFADDR: 3217 case PRIV_NET_LAGG: 3218 case PRIV_NET_GIF: 3219 case PRIV_NET_SETIFVNET: 3220 case PRIV_NET_SETIFFIB: 3221 3222 /* 3223 * 802.11-related privileges. 3224 */ 3225 case PRIV_NET80211_VAP_GETKEY: 3226 case PRIV_NET80211_VAP_MANAGE: 3227 3228 #ifdef notyet 3229 /* 3230 * ATM privileges. 3231 */ 3232 case PRIV_NETATM_CFG: 3233 case PRIV_NETATM_ADD: 3234 case PRIV_NETATM_DEL: 3235 case PRIV_NETATM_SET: 3236 3237 /* 3238 * Bluetooth privileges. 3239 */ 3240 case PRIV_NETBLUETOOTH_RAW: 3241 #endif 3242 3243 /* 3244 * Netgraph and netgraph module privileges. 3245 */ 3246 case PRIV_NETGRAPH_CONTROL: 3247 #ifdef notyet 3248 case PRIV_NETGRAPH_TTY: 3249 #endif 3250 3251 /* 3252 * IPv4 and IPv6 privileges. 3253 */ 3254 case PRIV_NETINET_IPFW: 3255 case PRIV_NETINET_DIVERT: 3256 case PRIV_NETINET_PF: 3257 case PRIV_NETINET_DUMMYNET: 3258 case PRIV_NETINET_CARP: 3259 case PRIV_NETINET_MROUTE: 3260 case PRIV_NETINET_RAW: 3261 case PRIV_NETINET_ADDRCTRL6: 3262 case PRIV_NETINET_ND6: 3263 case PRIV_NETINET_SCOPE6: 3264 case PRIV_NETINET_ALIFETIME6: 3265 case PRIV_NETINET_IPSEC: 3266 case PRIV_NETINET_BINDANY: 3267 3268 #ifdef notyet 3269 /* 3270 * NCP privileges. 3271 */ 3272 case PRIV_NETNCP: 3273 3274 /* 3275 * SMB privileges. 3276 */ 3277 case PRIV_NETSMB: 3278 #endif 3279 3280 /* 3281 * No default: or deny here. 3282 * In case of no permit fall through to next switch(). 3283 */ 3284 if (cred->cr_prison->pr_flags & PR_VNET) 3285 return (0); 3286 } 3287 #endif /* VIMAGE */ 3288 3289 switch (priv) { 3290 /* 3291 * Allow ktrace privileges for root in jail. 3292 */ 3293 case PRIV_KTRACE: 3294 3295 #if 0 3296 /* 3297 * Allow jailed processes to configure audit identity and 3298 * submit audit records (login, etc). In the future we may 3299 * want to further refine the relationship between audit and 3300 * jail. 3301 */ 3302 case PRIV_AUDIT_GETAUDIT: 3303 case PRIV_AUDIT_SETAUDIT: 3304 case PRIV_AUDIT_SUBMIT: 3305 #endif 3306 3307 /* 3308 * Allow jailed processes to manipulate process UNIX 3309 * credentials in any way they see fit. 3310 */ 3311 case PRIV_CRED_SETUID: 3312 case PRIV_CRED_SETEUID: 3313 case PRIV_CRED_SETGID: 3314 case PRIV_CRED_SETEGID: 3315 case PRIV_CRED_SETGROUPS: 3316 case PRIV_CRED_SETREUID: 3317 case PRIV_CRED_SETREGID: 3318 case PRIV_CRED_SETRESUID: 3319 case PRIV_CRED_SETRESGID: 3320 3321 /* 3322 * Jail implements visibility constraints already, so allow 3323 * jailed root to override uid/gid-based constraints. 3324 */ 3325 case PRIV_SEEOTHERGIDS: 3326 case PRIV_SEEOTHERUIDS: 3327 3328 /* 3329 * Jail implements inter-process debugging limits already, so 3330 * allow jailed root various debugging privileges. 3331 */ 3332 case PRIV_DEBUG_DIFFCRED: 3333 case PRIV_DEBUG_SUGID: 3334 case PRIV_DEBUG_UNPRIV: 3335 3336 /* 3337 * Allow jail to set various resource limits and login 3338 * properties, and for now, exceed process resource limits. 3339 */ 3340 case PRIV_PROC_LIMIT: 3341 case PRIV_PROC_SETLOGIN: 3342 case PRIV_PROC_SETRLIMIT: 3343 3344 /* 3345 * System V and POSIX IPC privileges are granted in jail. 3346 */ 3347 case PRIV_IPC_READ: 3348 case PRIV_IPC_WRITE: 3349 case PRIV_IPC_ADMIN: 3350 case PRIV_IPC_MSGSIZE: 3351 case PRIV_MQ_ADMIN: 3352 3353 /* 3354 * Jail operations within a jail work on child jails. 3355 */ 3356 case PRIV_JAIL_ATTACH: 3357 case PRIV_JAIL_SET: 3358 case PRIV_JAIL_REMOVE: 3359 3360 /* 3361 * Jail implements its own inter-process limits, so allow 3362 * root processes in jail to change scheduling on other 3363 * processes in the same jail. Likewise for signalling. 3364 */ 3365 case PRIV_SCHED_DIFFCRED: 3366 case PRIV_SCHED_CPUSET: 3367 case PRIV_SIGNAL_DIFFCRED: 3368 case PRIV_SIGNAL_SUGID: 3369 3370 /* 3371 * Allow jailed processes to write to sysctls marked as jail 3372 * writable. 3373 */ 3374 case PRIV_SYSCTL_WRITEJAIL: 3375 3376 /* 3377 * Allow root in jail to manage a variety of quota 3378 * properties. These should likely be conditional on a 3379 * configuration option. 3380 */ 3381 case PRIV_VFS_GETQUOTA: 3382 case PRIV_VFS_SETQUOTA: 3383 3384 /* 3385 * Since Jail relies on chroot() to implement file system 3386 * protections, grant many VFS privileges to root in jail. 3387 * Be careful to exclude mount-related and NFS-related 3388 * privileges. 3389 */ 3390 case PRIV_VFS_READ: 3391 case PRIV_VFS_WRITE: 3392 case PRIV_VFS_ADMIN: 3393 case PRIV_VFS_EXEC: 3394 case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */ 3395 case PRIV_VFS_CHFLAGS_DEV: 3396 case PRIV_VFS_CHOWN: 3397 case PRIV_VFS_CHROOT: 3398 case PRIV_VFS_RETAINSUGID: 3399 case PRIV_VFS_FCHROOT: 3400 case PRIV_VFS_LINK: 3401 case PRIV_VFS_SETGID: 3402 case PRIV_VFS_STAT: 3403 case PRIV_VFS_STICKYFILE: 3404 3405 /* 3406 * As in the non-jail case, non-root users are expected to be 3407 * able to read kernel/phyiscal memory (provided /dev/[k]mem 3408 * exists in the jail and they have permission to access it). 3409 */ 3410 case PRIV_KMEM_READ: 3411 return (0); 3412 3413 /* 3414 * Depending on the global setting, allow privilege of 3415 * setting system flags. 3416 */ 3417 case PRIV_VFS_SYSFLAGS: 3418 if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS) 3419 return (0); 3420 else 3421 return (EPERM); 3422 3423 /* 3424 * Depending on the global setting, allow privilege of 3425 * mounting/unmounting file systems. 3426 */ 3427 case PRIV_VFS_MOUNT: 3428 case PRIV_VFS_UNMOUNT: 3429 case PRIV_VFS_MOUNT_NONUSER: 3430 case PRIV_VFS_MOUNT_OWNER: 3431 pr = cred->cr_prison; 3432 prison_lock(pr); 3433 if (pr->pr_allow & PR_ALLOW_MOUNT && pr->pr_enforce_statfs < 2) 3434 error = 0; 3435 else 3436 error = EPERM; 3437 prison_unlock(pr); 3438 return (error); 3439 3440 /* 3441 * Jails should hold no disposition on the PRIV_VFS_READ_DIR 3442 * policy. priv_check_cred will not specifically allow it, and 3443 * we may want a MAC policy to allow it. 3444 */ 3445 case PRIV_VFS_READ_DIR: 3446 return (0); 3447 3448 /* 3449 * Conditionnaly allow locking (unlocking) physical pages 3450 * in memory. 3451 */ 3452 case PRIV_VM_MLOCK: 3453 case PRIV_VM_MUNLOCK: 3454 if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK) 3455 return (0); 3456 else 3457 return (EPERM); 3458 3459 /* 3460 * Conditionally allow jailed root to bind reserved ports. 3461 */ 3462 case PRIV_NETINET_RESERVEDPORT: 3463 if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS) 3464 return (0); 3465 else 3466 return (EPERM); 3467 3468 /* 3469 * Allow jailed root to reuse in-use ports. 3470 */ 3471 case PRIV_NETINET_REUSEPORT: 3472 return (0); 3473 3474 /* 3475 * Allow jailed root to set certain IPv4/6 (option) headers. 3476 */ 3477 case PRIV_NETINET_SETHDROPTS: 3478 return (0); 3479 3480 /* 3481 * Conditionally allow creating raw sockets in jail. 3482 */ 3483 case PRIV_NETINET_RAW: 3484 if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS) 3485 return (0); 3486 else 3487 return (EPERM); 3488 3489 /* 3490 * Since jail implements its own visibility limits on netstat 3491 * sysctls, allow getcred. This allows identd to work in 3492 * jail. 3493 */ 3494 case PRIV_NETINET_GETCRED: 3495 return (0); 3496 3497 /* 3498 * Allow jailed root to set loginclass. 3499 */ 3500 case PRIV_PROC_SETLOGINCLASS: 3501 return (0); 3502 3503 /* 3504 * Do not allow a process inside a jail to read the kernel 3505 * message buffer unless explicitly permitted. 3506 */ 3507 case PRIV_MSGBUF: 3508 if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF) 3509 return (0); 3510 return (EPERM); 3511 3512 default: 3513 /* 3514 * In all remaining cases, deny the privilege request. This 3515 * includes almost all network privileges, many system 3516 * configuration privileges. 3517 */ 3518 return (EPERM); 3519 } 3520 } 3521 3522 /* 3523 * Return the part of pr2's name that is relative to pr1, or the whole name 3524 * if it does not directly follow. 3525 */ 3526 3527 char * 3528 prison_name(struct prison *pr1, struct prison *pr2) 3529 { 3530 char *name; 3531 3532 /* Jails see themselves as "0" (if they see themselves at all). */ 3533 if (pr1 == pr2) 3534 return "0"; 3535 name = pr2->pr_name; 3536 if (prison_ischild(pr1, pr2)) { 3537 /* 3538 * pr1 isn't locked (and allprison_lock may not be either) 3539 * so its length can't be counted on. But the number of dots 3540 * can be counted on - and counted. 3541 */ 3542 for (; pr1 != &prison0; pr1 = pr1->pr_parent) 3543 name = strchr(name, '.') + 1; 3544 } 3545 return (name); 3546 } 3547 3548 /* 3549 * Return the part of pr2's path that is relative to pr1, or the whole path 3550 * if it does not directly follow. 3551 */ 3552 static char * 3553 prison_path(struct prison *pr1, struct prison *pr2) 3554 { 3555 char *path1, *path2; 3556 int len1; 3557 3558 path1 = pr1->pr_path; 3559 path2 = pr2->pr_path; 3560 if (!strcmp(path1, "/")) 3561 return (path2); 3562 len1 = strlen(path1); 3563 if (strncmp(path1, path2, len1)) 3564 return (path2); 3565 if (path2[len1] == '\0') 3566 return "/"; 3567 if (path2[len1] == '/') 3568 return (path2 + len1); 3569 return (path2); 3570 } 3571 3572 /* 3573 * Jail-related sysctls. 3574 */ 3575 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 3576 "Jails"); 3577 3578 static int 3579 sysctl_jail_list(SYSCTL_HANDLER_ARGS) 3580 { 3581 struct xprison *xp; 3582 struct prison *pr, *cpr; 3583 #ifdef INET 3584 struct in_addr *ip4 = NULL; 3585 int ip4s = 0; 3586 #endif 3587 #ifdef INET6 3588 struct in6_addr *ip6 = NULL; 3589 int ip6s = 0; 3590 #endif 3591 int descend, error; 3592 3593 xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK); 3594 pr = req->td->td_ucred->cr_prison; 3595 error = 0; 3596 sx_slock(&allprison_lock); 3597 FOREACH_PRISON_DESCENDANT(pr, cpr, descend) { 3598 #if defined(INET) || defined(INET6) 3599 again: 3600 #endif 3601 mtx_lock(&cpr->pr_mtx); 3602 #ifdef INET 3603 if (cpr->pr_ip4s > 0) { 3604 if (ip4s < cpr->pr_ip4s) { 3605 ip4s = cpr->pr_ip4s; 3606 mtx_unlock(&cpr->pr_mtx); 3607 ip4 = realloc(ip4, ip4s * 3608 sizeof(struct in_addr), M_TEMP, M_WAITOK); 3609 goto again; 3610 } 3611 bcopy(cpr->pr_ip4, ip4, 3612 cpr->pr_ip4s * sizeof(struct in_addr)); 3613 } 3614 #endif 3615 #ifdef INET6 3616 if (cpr->pr_ip6s > 0) { 3617 if (ip6s < cpr->pr_ip6s) { 3618 ip6s = cpr->pr_ip6s; 3619 mtx_unlock(&cpr->pr_mtx); 3620 ip6 = realloc(ip6, ip6s * 3621 sizeof(struct in6_addr), M_TEMP, M_WAITOK); 3622 goto again; 3623 } 3624 bcopy(cpr->pr_ip6, ip6, 3625 cpr->pr_ip6s * sizeof(struct in6_addr)); 3626 } 3627 #endif 3628 if (cpr->pr_ref == 0) { 3629 mtx_unlock(&cpr->pr_mtx); 3630 continue; 3631 } 3632 bzero(xp, sizeof(*xp)); 3633 xp->pr_version = XPRISON_VERSION; 3634 xp->pr_id = cpr->pr_id; 3635 xp->pr_state = cpr->pr_uref > 0 3636 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING; 3637 strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path)); 3638 strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host)); 3639 strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name)); 3640 #ifdef INET 3641 xp->pr_ip4s = cpr->pr_ip4s; 3642 #endif 3643 #ifdef INET6 3644 xp->pr_ip6s = cpr->pr_ip6s; 3645 #endif 3646 mtx_unlock(&cpr->pr_mtx); 3647 error = SYSCTL_OUT(req, xp, sizeof(*xp)); 3648 if (error) 3649 break; 3650 #ifdef INET 3651 if (xp->pr_ip4s > 0) { 3652 error = SYSCTL_OUT(req, ip4, 3653 xp->pr_ip4s * sizeof(struct in_addr)); 3654 if (error) 3655 break; 3656 } 3657 #endif 3658 #ifdef INET6 3659 if (xp->pr_ip6s > 0) { 3660 error = SYSCTL_OUT(req, ip6, 3661 xp->pr_ip6s * sizeof(struct in6_addr)); 3662 if (error) 3663 break; 3664 } 3665 #endif 3666 } 3667 sx_sunlock(&allprison_lock); 3668 free(xp, M_TEMP); 3669 #ifdef INET 3670 free(ip4, M_TEMP); 3671 #endif 3672 #ifdef INET6 3673 free(ip6, M_TEMP); 3674 #endif 3675 return (error); 3676 } 3677 3678 SYSCTL_OID(_security_jail, OID_AUTO, list, 3679 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 3680 sysctl_jail_list, "S", "List of active jails"); 3681 3682 static int 3683 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS) 3684 { 3685 int error, injail; 3686 3687 injail = jailed(req->td->td_ucred); 3688 error = SYSCTL_OUT(req, &injail, sizeof(injail)); 3689 3690 return (error); 3691 } 3692 3693 SYSCTL_PROC(_security_jail, OID_AUTO, jailed, 3694 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 3695 sysctl_jail_jailed, "I", "Process in jail?"); 3696 3697 static int 3698 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS) 3699 { 3700 int error, havevnet; 3701 #ifdef VIMAGE 3702 struct ucred *cred = req->td->td_ucred; 3703 3704 havevnet = jailed(cred) && prison_owns_vnet(cred); 3705 #else 3706 havevnet = 0; 3707 #endif 3708 error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet)); 3709 3710 return (error); 3711 } 3712 3713 SYSCTL_PROC(_security_jail, OID_AUTO, vnet, 3714 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 3715 sysctl_jail_vnet, "I", "Jail owns vnet?"); 3716 3717 #if defined(INET) || defined(INET6) 3718 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW, 3719 &jail_max_af_ips, 0, 3720 "Number of IP addresses a jail may have at most per address family (deprecated)"); 3721 #endif 3722 3723 /* 3724 * Default parameters for jail(2) compatibility. For historical reasons, 3725 * the sysctl names have varying similarity to the parameter names. Prisons 3726 * just see their own parameters, and can't change them. 3727 */ 3728 static int 3729 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS) 3730 { 3731 int error, i; 3732 3733 /* Get the current flag value, and convert it to a boolean. */ 3734 if (req->td->td_ucred->cr_prison == &prison0) { 3735 mtx_lock(&prison0.pr_mtx); 3736 i = (jail_default_allow & arg2) != 0; 3737 mtx_unlock(&prison0.pr_mtx); 3738 } else 3739 i = prison_allow(req->td->td_ucred, arg2); 3740 3741 if (arg1 != NULL) 3742 i = !i; 3743 error = sysctl_handle_int(oidp, &i, 0, req); 3744 if (error || !req->newptr) 3745 return (error); 3746 i = i ? arg2 : 0; 3747 if (arg1 != NULL) 3748 i ^= arg2; 3749 /* 3750 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0 3751 * for writing. 3752 */ 3753 mtx_lock(&prison0.pr_mtx); 3754 jail_default_allow = (jail_default_allow & ~arg2) | i; 3755 mtx_unlock(&prison0.pr_mtx); 3756 return (0); 3757 } 3758 3759 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed, 3760 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3761 NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I", 3762 "Processes in jail can set their hostnames (deprecated)"); 3763 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only, 3764 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3765 (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I", 3766 "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)"); 3767 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed, 3768 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3769 NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I", 3770 "Processes in jail can use System V IPC primitives (deprecated)"); 3771 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets, 3772 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3773 NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I", 3774 "Prison root can create raw sockets (deprecated)"); 3775 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed, 3776 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3777 NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I", 3778 "Processes in jail can alter system file flags (deprecated)"); 3779 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed, 3780 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3781 NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I", 3782 "Processes in jail can mount/unmount jail-friendly file systems (deprecated)"); 3783 3784 static int 3785 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS) 3786 { 3787 struct prison *pr; 3788 int level, error; 3789 3790 pr = req->td->td_ucred->cr_prison; 3791 level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2); 3792 error = sysctl_handle_int(oidp, &level, 0, req); 3793 if (error || !req->newptr) 3794 return (error); 3795 *(int *)arg1 = level; 3796 return (0); 3797 } 3798 3799 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs, 3800 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 3801 &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs), 3802 sysctl_jail_default_level, "I", 3803 "Processes in jail cannot see all mounted file systems (deprecated)"); 3804 3805 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset, 3806 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, 3807 &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum), 3808 sysctl_jail_default_level, "I", 3809 "Ruleset for the devfs filesystem in jail (deprecated)"); 3810 3811 /* 3812 * Nodes to describe jail parameters. Maximum length of string parameters 3813 * is returned in the string itself, and the other parameters exist merely 3814 * to make themselves and their types known. 3815 */ 3816 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 3817 "Jail parameters"); 3818 3819 int 3820 sysctl_jail_param(SYSCTL_HANDLER_ARGS) 3821 { 3822 int i; 3823 long l; 3824 size_t s; 3825 char numbuf[12]; 3826 3827 switch (oidp->oid_kind & CTLTYPE) 3828 { 3829 case CTLTYPE_LONG: 3830 case CTLTYPE_ULONG: 3831 l = 0; 3832 #ifdef SCTL_MASK32 3833 if (!(req->flags & SCTL_MASK32)) 3834 #endif 3835 return (SYSCTL_OUT(req, &l, sizeof(l))); 3836 case CTLTYPE_INT: 3837 case CTLTYPE_UINT: 3838 i = 0; 3839 return (SYSCTL_OUT(req, &i, sizeof(i))); 3840 case CTLTYPE_STRING: 3841 snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2); 3842 return 3843 (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req)); 3844 case CTLTYPE_STRUCT: 3845 s = (size_t)arg2; 3846 return (SYSCTL_OUT(req, &s, sizeof(s))); 3847 } 3848 return (0); 3849 } 3850 3851 /* 3852 * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at 3853 * jail creation time but cannot be changed in an existing jail. 3854 */ 3855 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID"); 3856 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID"); 3857 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name"); 3858 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path"); 3859 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW, 3860 "I", "Jail secure level"); 3861 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I", 3862 "Jail value for kern.osreldate and uname -K"); 3863 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN, 3864 "Jail value for kern.osrelease and uname -r"); 3865 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW, 3866 "I", "Jail cannot see all mounted file systems"); 3867 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW, 3868 "I", "Ruleset for in-jail devfs mounts"); 3869 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW, 3870 "B", "Jail persistence"); 3871 #ifdef VIMAGE 3872 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN, 3873 "E,jailsys", "Virtual network stack"); 3874 #endif 3875 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD, 3876 "B", "Jail is in the process of shutting down"); 3877 3878 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails"); 3879 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD, 3880 "I", "Current number of child jails"); 3881 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW, 3882 "I", "Maximum number of child jails"); 3883 3884 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info"); 3885 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN, 3886 "Jail hostname"); 3887 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN, 3888 "Jail NIS domainname"); 3889 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN, 3890 "Jail host UUID"); 3891 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW, 3892 "LU", "Jail host ID"); 3893 3894 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset"); 3895 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID"); 3896 3897 #ifdef INET 3898 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN, 3899 "Jail IPv4 address virtualization"); 3900 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr), 3901 "S,in_addr,a", "Jail IPv4 addresses"); 3902 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW, 3903 "B", "Do (not) use IPv4 source address selection rather than the " 3904 "primary jail IPv4 address."); 3905 #endif 3906 #ifdef INET6 3907 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN, 3908 "Jail IPv6 address virtualization"); 3909 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr), 3910 "S,in6_addr,a", "Jail IPv6 addresses"); 3911 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW, 3912 "B", "Do (not) use IPv6 source address selection rather than the " 3913 "primary jail IPv6 address."); 3914 #endif 3915 3916 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags"); 3917 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW, 3918 "B", "Jail may set hostname"); 3919 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW, 3920 "B", "Jail may use SYSV IPC"); 3921 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW, 3922 "B", "Jail may create raw sockets"); 3923 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW, 3924 "B", "Jail may alter system file flags"); 3925 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW, 3926 "B", "Jail may set file quotas"); 3927 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW, 3928 "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route"); 3929 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW, 3930 "B", "Jail may lock (unlock) physical pages in memory"); 3931 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW, 3932 "B", "Jail may bind sockets to reserved ports"); 3933 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW, 3934 "B", "Jail may read the kernel message buffer"); 3935 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW, 3936 "B", "Unprivileged processes may use process debugging facilities"); 3937 SYSCTL_JAIL_PARAM(_allow, suser, CTLTYPE_INT | CTLFLAG_RW, 3938 "B", "Processes in jail with uid 0 have privilege"); 3939 3940 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags"); 3941 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW, 3942 "B", "Jail may mount/unmount jail-friendly file systems in general"); 3943 3944 /* 3945 * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>. Return 3946 * its associated bit in the pr_allow bitmask, or zero if the parameter was 3947 * not created. 3948 */ 3949 unsigned 3950 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr, 3951 const char *descr) 3952 { 3953 struct bool_flags *bf; 3954 struct sysctl_oid *parent; 3955 char *allow_name, *allow_noname, *allowed; 3956 #ifndef NO_SYSCTL_DESCR 3957 char *descr_deprecated; 3958 #endif 3959 u_int allow_flag; 3960 3961 if (prefix 3962 ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name) 3963 < 0 || 3964 asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name) 3965 < 0 3966 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 || 3967 asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) { 3968 free(allow_name, M_PRISON); 3969 return 0; 3970 } 3971 3972 /* 3973 * See if this parameter has already beed added, i.e. a module was 3974 * previously loaded/unloaded. 3975 */ 3976 mtx_lock(&prison0.pr_mtx); 3977 for (bf = pr_flag_allow; 3978 bf < pr_flag_allow + nitems(pr_flag_allow) && 3979 atomic_load_int(&bf->flag) != 0; 3980 bf++) { 3981 if (strcmp(bf->name, allow_name) == 0) { 3982 allow_flag = bf->flag; 3983 goto no_add; 3984 } 3985 } 3986 3987 /* 3988 * Find a free bit in pr_allow_all, failing if there are none 3989 * (which shouldn't happen as long as we keep track of how many 3990 * potential dynamic flags exist). 3991 */ 3992 for (allow_flag = 1;; allow_flag <<= 1) { 3993 if (allow_flag == 0) 3994 goto no_add; 3995 if ((pr_allow_all & allow_flag) == 0) 3996 break; 3997 } 3998 3999 /* Note the parameter in the next open slot in pr_flag_allow. */ 4000 for (bf = pr_flag_allow; ; bf++) { 4001 if (bf == pr_flag_allow + nitems(pr_flag_allow)) { 4002 /* This should never happen, but is not fatal. */ 4003 allow_flag = 0; 4004 goto no_add; 4005 } 4006 if (atomic_load_int(&bf->flag) == 0) 4007 break; 4008 } 4009 bf->name = allow_name; 4010 bf->noname = allow_noname; 4011 pr_allow_all |= allow_flag; 4012 /* 4013 * prison0 always has permission for the new parameter. 4014 * Other jails must have it granted to them. 4015 */ 4016 prison0.pr_allow |= allow_flag; 4017 /* The flag indicates a valid entry, so make sure it is set last. */ 4018 atomic_store_rel_int(&bf->flag, allow_flag); 4019 mtx_unlock(&prison0.pr_mtx); 4020 4021 /* 4022 * Create sysctls for the paramter, and the back-compat global 4023 * permission. 4024 */ 4025 parent = prefix 4026 ? SYSCTL_ADD_NODE(NULL, 4027 SYSCTL_CHILDREN(&sysctl___security_jail_param_allow), 4028 OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr) 4029 : &sysctl___security_jail_param_allow; 4030 (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO, 4031 name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 4032 NULL, 0, sysctl_jail_param, "B", descr); 4033 if ((prefix 4034 ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name) 4035 : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) { 4036 #ifndef NO_SYSCTL_DESCR 4037 (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)", 4038 descr); 4039 #endif 4040 (void)SYSCTL_ADD_PROC(NULL, 4041 SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed, 4042 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag, 4043 sysctl_jail_default_allow, "I", descr_deprecated); 4044 #ifndef NO_SYSCTL_DESCR 4045 free(descr_deprecated, M_TEMP); 4046 #endif 4047 free(allowed, M_TEMP); 4048 } 4049 return allow_flag; 4050 4051 no_add: 4052 mtx_unlock(&prison0.pr_mtx); 4053 free(allow_name, M_PRISON); 4054 free(allow_noname, M_PRISON); 4055 return allow_flag; 4056 } 4057 4058 /* 4059 * The VFS system will register jail-aware filesystems here. They each get 4060 * a parameter allow.mount.xxxfs and a flag to check when a jailed user 4061 * attempts to mount. 4062 */ 4063 void 4064 prison_add_vfs(struct vfsconf *vfsp) 4065 { 4066 #ifdef NO_SYSCTL_DESCR 4067 4068 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name, 4069 NULL, NULL); 4070 #else 4071 char *descr; 4072 4073 (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system", 4074 vfsp->vfc_name); 4075 vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name, 4076 NULL, descr); 4077 free(descr, M_TEMP); 4078 #endif 4079 } 4080 4081 #ifdef RACCT 4082 void 4083 prison_racct_foreach(void (*callback)(struct racct *racct, 4084 void *arg2, void *arg3), void (*pre)(void), void (*post)(void), 4085 void *arg2, void *arg3) 4086 { 4087 struct prison_racct *prr; 4088 4089 ASSERT_RACCT_ENABLED(); 4090 4091 sx_slock(&allprison_lock); 4092 if (pre != NULL) 4093 (pre)(); 4094 LIST_FOREACH(prr, &allprison_racct, prr_next) 4095 (callback)(prr->prr_racct, arg2, arg3); 4096 if (post != NULL) 4097 (post)(); 4098 sx_sunlock(&allprison_lock); 4099 } 4100 4101 static struct prison_racct * 4102 prison_racct_find_locked(const char *name) 4103 { 4104 struct prison_racct *prr; 4105 4106 ASSERT_RACCT_ENABLED(); 4107 sx_assert(&allprison_lock, SA_XLOCKED); 4108 4109 if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN) 4110 return (NULL); 4111 4112 LIST_FOREACH(prr, &allprison_racct, prr_next) { 4113 if (strcmp(name, prr->prr_name) != 0) 4114 continue; 4115 4116 /* Found prison_racct with a matching name? */ 4117 prison_racct_hold(prr); 4118 return (prr); 4119 } 4120 4121 /* Add new prison_racct. */ 4122 prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK); 4123 racct_create(&prr->prr_racct); 4124 4125 strcpy(prr->prr_name, name); 4126 refcount_init(&prr->prr_refcount, 1); 4127 LIST_INSERT_HEAD(&allprison_racct, prr, prr_next); 4128 4129 return (prr); 4130 } 4131 4132 struct prison_racct * 4133 prison_racct_find(const char *name) 4134 { 4135 struct prison_racct *prr; 4136 4137 ASSERT_RACCT_ENABLED(); 4138 4139 sx_xlock(&allprison_lock); 4140 prr = prison_racct_find_locked(name); 4141 sx_xunlock(&allprison_lock); 4142 return (prr); 4143 } 4144 4145 void 4146 prison_racct_hold(struct prison_racct *prr) 4147 { 4148 4149 ASSERT_RACCT_ENABLED(); 4150 4151 refcount_acquire(&prr->prr_refcount); 4152 } 4153 4154 static void 4155 prison_racct_free_locked(struct prison_racct *prr) 4156 { 4157 4158 ASSERT_RACCT_ENABLED(); 4159 sx_assert(&allprison_lock, SA_XLOCKED); 4160 4161 if (refcount_release(&prr->prr_refcount)) { 4162 racct_destroy(&prr->prr_racct); 4163 LIST_REMOVE(prr, prr_next); 4164 free(prr, M_PRISON_RACCT); 4165 } 4166 } 4167 4168 void 4169 prison_racct_free(struct prison_racct *prr) 4170 { 4171 4172 ASSERT_RACCT_ENABLED(); 4173 sx_assert(&allprison_lock, SA_UNLOCKED); 4174 4175 if (refcount_release_if_not_last(&prr->prr_refcount)) 4176 return; 4177 4178 sx_xlock(&allprison_lock); 4179 prison_racct_free_locked(prr); 4180 sx_xunlock(&allprison_lock); 4181 } 4182 4183 static void 4184 prison_racct_attach(struct prison *pr) 4185 { 4186 struct prison_racct *prr; 4187 4188 ASSERT_RACCT_ENABLED(); 4189 sx_assert(&allprison_lock, SA_XLOCKED); 4190 4191 prr = prison_racct_find_locked(pr->pr_name); 4192 KASSERT(prr != NULL, ("cannot find prison_racct")); 4193 4194 pr->pr_prison_racct = prr; 4195 } 4196 4197 /* 4198 * Handle jail renaming. From the racct point of view, renaming means 4199 * moving from one prison_racct to another. 4200 */ 4201 static void 4202 prison_racct_modify(struct prison *pr) 4203 { 4204 #ifdef RCTL 4205 struct proc *p; 4206 struct ucred *cred; 4207 #endif 4208 struct prison_racct *oldprr; 4209 4210 ASSERT_RACCT_ENABLED(); 4211 4212 sx_slock(&allproc_lock); 4213 sx_xlock(&allprison_lock); 4214 4215 if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) { 4216 sx_xunlock(&allprison_lock); 4217 sx_sunlock(&allproc_lock); 4218 return; 4219 } 4220 4221 oldprr = pr->pr_prison_racct; 4222 pr->pr_prison_racct = NULL; 4223 4224 prison_racct_attach(pr); 4225 4226 /* 4227 * Move resource utilisation records. 4228 */ 4229 racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct); 4230 4231 #ifdef RCTL 4232 /* 4233 * Force rctl to reattach rules to processes. 4234 */ 4235 FOREACH_PROC_IN_SYSTEM(p) { 4236 PROC_LOCK(p); 4237 cred = crhold(p->p_ucred); 4238 PROC_UNLOCK(p); 4239 rctl_proc_ucred_changed(p, cred); 4240 crfree(cred); 4241 } 4242 #endif 4243 4244 sx_sunlock(&allproc_lock); 4245 prison_racct_free_locked(oldprr); 4246 sx_xunlock(&allprison_lock); 4247 } 4248 4249 static void 4250 prison_racct_detach(struct prison *pr) 4251 { 4252 4253 ASSERT_RACCT_ENABLED(); 4254 sx_assert(&allprison_lock, SA_UNLOCKED); 4255 4256 if (pr->pr_prison_racct == NULL) 4257 return; 4258 prison_racct_free(pr->pr_prison_racct); 4259 pr->pr_prison_racct = NULL; 4260 } 4261 #endif /* RACCT */ 4262 4263 #ifdef DDB 4264 4265 static void 4266 db_show_prison(struct prison *pr) 4267 { 4268 struct bool_flags *bf; 4269 struct jailsys_flags *jsf; 4270 #if defined(INET) || defined(INET6) 4271 int ii; 4272 #endif 4273 unsigned f; 4274 #ifdef INET 4275 char ip4buf[INET_ADDRSTRLEN]; 4276 #endif 4277 #ifdef INET6 4278 char ip6buf[INET6_ADDRSTRLEN]; 4279 #endif 4280 4281 db_printf("prison %p:\n", pr); 4282 db_printf(" jid = %d\n", pr->pr_id); 4283 db_printf(" name = %s\n", pr->pr_name); 4284 db_printf(" parent = %p\n", pr->pr_parent); 4285 db_printf(" ref = %d\n", pr->pr_ref); 4286 db_printf(" uref = %d\n", pr->pr_uref); 4287 db_printf(" path = %s\n", pr->pr_path); 4288 db_printf(" cpuset = %d\n", pr->pr_cpuset 4289 ? pr->pr_cpuset->cs_id : -1); 4290 #ifdef VIMAGE 4291 db_printf(" vnet = %p\n", pr->pr_vnet); 4292 #endif 4293 db_printf(" root = %p\n", pr->pr_root); 4294 db_printf(" securelevel = %d\n", pr->pr_securelevel); 4295 db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum); 4296 db_printf(" children.max = %d\n", pr->pr_childmax); 4297 db_printf(" children.cur = %d\n", pr->pr_childcount); 4298 db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children)); 4299 db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling)); 4300 db_printf(" flags = 0x%x", pr->pr_flags); 4301 for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++) 4302 if (pr->pr_flags & bf->flag) 4303 db_printf(" %s", bf->name); 4304 for (jsf = pr_flag_jailsys; 4305 jsf < pr_flag_jailsys + nitems(pr_flag_jailsys); 4306 jsf++) { 4307 f = pr->pr_flags & (jsf->disable | jsf->new); 4308 db_printf(" %-16s= %s\n", jsf->name, 4309 (f != 0 && f == jsf->disable) ? "disable" 4310 : (f == jsf->new) ? "new" 4311 : "inherit"); 4312 } 4313 db_printf(" allow = 0x%x", pr->pr_allow); 4314 for (bf = pr_flag_allow; 4315 bf < pr_flag_allow + nitems(pr_flag_allow) && 4316 atomic_load_int(&bf->flag) != 0; 4317 bf++) 4318 if (pr->pr_allow & bf->flag) 4319 db_printf(" %s", bf->name); 4320 db_printf("\n"); 4321 db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs); 4322 db_printf(" host.hostname = %s\n", pr->pr_hostname); 4323 db_printf(" host.domainname = %s\n", pr->pr_domainname); 4324 db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid); 4325 db_printf(" host.hostid = %lu\n", pr->pr_hostid); 4326 #ifdef INET 4327 db_printf(" ip4s = %d\n", pr->pr_ip4s); 4328 for (ii = 0; ii < pr->pr_ip4s; ii++) 4329 db_printf(" %s %s\n", 4330 ii == 0 ? "ip4.addr =" : " ", 4331 inet_ntoa_r(pr->pr_ip4[ii], ip4buf)); 4332 #endif 4333 #ifdef INET6 4334 db_printf(" ip6s = %d\n", pr->pr_ip6s); 4335 for (ii = 0; ii < pr->pr_ip6s; ii++) 4336 db_printf(" %s %s\n", 4337 ii == 0 ? "ip6.addr =" : " ", 4338 ip6_sprintf(ip6buf, &pr->pr_ip6[ii])); 4339 #endif 4340 } 4341 4342 DB_SHOW_COMMAND(prison, db_show_prison_command) 4343 { 4344 struct prison *pr; 4345 4346 if (!have_addr) { 4347 /* 4348 * Show all prisons in the list, and prison0 which is not 4349 * listed. 4350 */ 4351 db_show_prison(&prison0); 4352 if (!db_pager_quit) { 4353 TAILQ_FOREACH(pr, &allprison, pr_list) { 4354 db_show_prison(pr); 4355 if (db_pager_quit) 4356 break; 4357 } 4358 } 4359 return; 4360 } 4361 4362 if (addr == 0) 4363 pr = &prison0; 4364 else { 4365 /* Look for a prison with the ID and with references. */ 4366 TAILQ_FOREACH(pr, &allprison, pr_list) 4367 if (pr->pr_id == addr && pr->pr_ref > 0) 4368 break; 4369 if (pr == NULL) 4370 /* Look again, without requiring a reference. */ 4371 TAILQ_FOREACH(pr, &allprison, pr_list) 4372 if (pr->pr_id == addr) 4373 break; 4374 if (pr == NULL) 4375 /* Assume address points to a valid prison. */ 4376 pr = (struct prison *)addr; 4377 } 4378 db_show_prison(pr); 4379 } 4380 4381 #endif /* DDB */ 4382