1 /*- 2 * Copyright (c) 2016 Yandex LLC 3 * Copyright (c) 2016 Andrey V. Elsukov <ae@FreeBSD.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/counter.h> 34 #include <sys/eventhandler.h> 35 #include <sys/errno.h> 36 #include <sys/kernel.h> 37 #include <sys/lock.h> 38 #include <sys/malloc.h> 39 #include <sys/mbuf.h> 40 #include <sys/module.h> 41 #include <sys/rmlock.h> 42 #include <sys/rwlock.h> 43 #include <sys/socket.h> 44 #include <sys/queue.h> 45 #include <sys/syslog.h> 46 #include <sys/sysctl.h> 47 48 #include <net/if.h> 49 #include <net/if_var.h> 50 #include <net/netisr.h> 51 #include <net/pfil.h> 52 #include <net/vnet.h> 53 54 #include <netinet/in.h> 55 #include <netinet/ip_var.h> 56 #include <netinet/ip_fw.h> 57 #include <netinet/ip6.h> 58 #include <netinet/icmp6.h> 59 #include <netinet6/in6_var.h> 60 #include <netinet6/ip6_var.h> 61 62 #include <netpfil/ipfw/ip_fw_private.h> 63 #include <netpfil/ipfw/nptv6/nptv6.h> 64 65 VNET_DEFINE_STATIC(uint16_t, nptv6_eid) = 0; 66 #define V_nptv6_eid VNET(nptv6_eid) 67 #define IPFW_TLV_NPTV6_NAME IPFW_TLV_EACTION_NAME(V_nptv6_eid) 68 69 static eventhandler_tag nptv6_ifaddr_event; 70 71 static struct nptv6_cfg *nptv6_alloc_config(const char *name, uint8_t set); 72 static void nptv6_free_config(struct nptv6_cfg *cfg); 73 static struct nptv6_cfg *nptv6_find(struct namedobj_instance *ni, 74 const char *name, uint8_t set); 75 static int nptv6_rewrite_internal(struct nptv6_cfg *cfg, struct mbuf **mp, 76 int offset); 77 static int nptv6_rewrite_external(struct nptv6_cfg *cfg, struct mbuf **mp, 78 int offset); 79 80 #define NPTV6_LOOKUP(chain, cmd) \ 81 (struct nptv6_cfg *)SRV_OBJECT((chain), (cmd)->arg1) 82 83 #ifndef IN6_MASK_ADDR 84 #define IN6_MASK_ADDR(a, m) do { \ 85 (a)->s6_addr32[0] &= (m)->s6_addr32[0]; \ 86 (a)->s6_addr32[1] &= (m)->s6_addr32[1]; \ 87 (a)->s6_addr32[2] &= (m)->s6_addr32[2]; \ 88 (a)->s6_addr32[3] &= (m)->s6_addr32[3]; \ 89 } while (0) 90 #endif 91 #ifndef IN6_ARE_MASKED_ADDR_EQUAL 92 #define IN6_ARE_MASKED_ADDR_EQUAL(d, a, m) ( \ 93 (((d)->s6_addr32[0] ^ (a)->s6_addr32[0]) & (m)->s6_addr32[0]) == 0 && \ 94 (((d)->s6_addr32[1] ^ (a)->s6_addr32[1]) & (m)->s6_addr32[1]) == 0 && \ 95 (((d)->s6_addr32[2] ^ (a)->s6_addr32[2]) & (m)->s6_addr32[2]) == 0 && \ 96 (((d)->s6_addr32[3] ^ (a)->s6_addr32[3]) & (m)->s6_addr32[3]) == 0 ) 97 #endif 98 99 #if 0 100 #define NPTV6_DEBUG(fmt, ...) do { \ 101 printf("%s: " fmt "\n", __func__, ## __VA_ARGS__); \ 102 } while (0) 103 #define NPTV6_IPDEBUG(fmt, ...) do { \ 104 char _s[INET6_ADDRSTRLEN], _d[INET6_ADDRSTRLEN]; \ 105 printf("%s: " fmt "\n", __func__, ## __VA_ARGS__); \ 106 } while (0) 107 #else 108 #define NPTV6_DEBUG(fmt, ...) 109 #define NPTV6_IPDEBUG(fmt, ...) 110 #endif 111 112 static int 113 nptv6_getlasthdr(struct nptv6_cfg *cfg, struct mbuf *m, int *offset) 114 { 115 struct ip6_hdr *ip6; 116 struct ip6_hbh *hbh; 117 int proto, hlen; 118 119 hlen = (offset == NULL) ? 0: *offset; 120 if (m->m_len < hlen) 121 return (-1); 122 ip6 = mtodo(m, hlen); 123 hlen += sizeof(*ip6); 124 proto = ip6->ip6_nxt; 125 while (proto == IPPROTO_HOPOPTS || proto == IPPROTO_ROUTING || 126 proto == IPPROTO_DSTOPTS) { 127 hbh = mtodo(m, hlen); 128 if (m->m_len < hlen) 129 return (-1); 130 proto = hbh->ip6h_nxt; 131 hlen += (hbh->ip6h_len + 1) << 3; 132 } 133 if (offset != NULL) 134 *offset = hlen; 135 return (proto); 136 } 137 138 static int 139 nptv6_translate_icmpv6(struct nptv6_cfg *cfg, struct mbuf **mp, int offset) 140 { 141 struct icmp6_hdr *icmp6; 142 struct ip6_hdr *ip6; 143 struct mbuf *m; 144 145 m = *mp; 146 if (offset > m->m_len) 147 return (-1); 148 icmp6 = mtodo(m, offset); 149 NPTV6_DEBUG("ICMPv6 type %d", icmp6->icmp6_type); 150 switch (icmp6->icmp6_type) { 151 case ICMP6_DST_UNREACH: 152 case ICMP6_PACKET_TOO_BIG: 153 case ICMP6_TIME_EXCEEDED: 154 case ICMP6_PARAM_PROB: 155 break; 156 case ICMP6_ECHO_REQUEST: 157 case ICMP6_ECHO_REPLY: 158 /* nothing to translate */ 159 return (0); 160 default: 161 /* 162 * XXX: We can add some checks to not translate NDP and MLD 163 * messages. Currently user must explicitly allow these message 164 * types, otherwise packets will be dropped. 165 */ 166 return (-1); 167 } 168 offset += sizeof(*icmp6); 169 if (offset + sizeof(*ip6) > m->m_pkthdr.len) 170 return (-1); 171 if (offset + sizeof(*ip6) > m->m_len) 172 *mp = m = m_pullup(m, offset + sizeof(*ip6)); 173 if (m == NULL) 174 return (-1); 175 ip6 = mtodo(m, offset); 176 NPTV6_IPDEBUG("offset %d, %s -> %s %d", offset, 177 inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), 178 inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), 179 ip6->ip6_nxt); 180 if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_src, 181 &cfg->external, &cfg->mask)) 182 return (nptv6_rewrite_external(cfg, mp, offset)); 183 else if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_dst, 184 &cfg->internal, &cfg->mask)) 185 return (nptv6_rewrite_internal(cfg, mp, offset)); 186 /* 187 * Addresses in the inner IPv6 header doesn't matched to 188 * our prefixes. 189 */ 190 return (-1); 191 } 192 193 static int 194 nptv6_search_index(struct nptv6_cfg *cfg, struct in6_addr *a) 195 { 196 int idx; 197 198 if (cfg->flags & NPTV6_48PLEN) 199 return (3); 200 201 /* Search suitable word index for adjustment */ 202 for (idx = 4; idx < 8; idx++) 203 if (a->s6_addr16[idx] != 0xffff) 204 break; 205 /* 206 * RFC 6296 p3.7: If an NPTv6 Translator discovers a datagram with 207 * an IID of all-zeros while performing address mapping, that 208 * datagram MUST be dropped, and an ICMPv6 Parameter Problem error 209 * SHOULD be generated. 210 */ 211 if (idx == 8 || 212 (a->s6_addr32[2] == 0 && a->s6_addr32[3] == 0)) 213 return (-1); 214 return (idx); 215 } 216 217 static void 218 nptv6_copy_addr(struct in6_addr *src, struct in6_addr *dst, 219 struct in6_addr *mask) 220 { 221 int i; 222 223 for (i = 0; i < 8 && mask->s6_addr8[i] != 0; i++) { 224 dst->s6_addr8[i] &= ~mask->s6_addr8[i]; 225 dst->s6_addr8[i] |= src->s6_addr8[i] & mask->s6_addr8[i]; 226 } 227 } 228 229 static int 230 nptv6_rewrite_internal(struct nptv6_cfg *cfg, struct mbuf **mp, int offset) 231 { 232 struct in6_addr *addr; 233 struct ip6_hdr *ip6; 234 int idx, proto; 235 uint16_t adj; 236 237 ip6 = mtodo(*mp, offset); 238 NPTV6_IPDEBUG("offset %d, %s -> %s %d", offset, 239 inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), 240 inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), 241 ip6->ip6_nxt); 242 if (offset == 0) 243 addr = &ip6->ip6_src; 244 else { 245 /* 246 * When we rewriting inner IPv6 header, we need to rewrite 247 * destination address back to external prefix. The datagram in 248 * the ICMPv6 payload should looks like it was send from 249 * external prefix. 250 */ 251 addr = &ip6->ip6_dst; 252 } 253 idx = nptv6_search_index(cfg, addr); 254 if (idx < 0) { 255 /* 256 * Do not send ICMPv6 error when offset isn't zero. 257 * This means we are rewriting inner IPv6 header in the 258 * ICMPv6 error message. 259 */ 260 if (offset == 0) { 261 icmp6_error2(*mp, ICMP6_DST_UNREACH, 262 ICMP6_DST_UNREACH_ADDR, 0, (*mp)->m_pkthdr.rcvif); 263 *mp = NULL; 264 } 265 return (IP_FW_DENY); 266 } 267 adj = addr->s6_addr16[idx]; 268 nptv6_copy_addr(&cfg->external, addr, &cfg->mask); 269 adj = cksum_add(adj, cfg->adjustment); 270 if (adj == 0xffff) 271 adj = 0; 272 addr->s6_addr16[idx] = adj; 273 if (offset == 0) { 274 /* 275 * We may need to translate addresses in the inner IPv6 276 * header for ICMPv6 error messages. 277 */ 278 proto = nptv6_getlasthdr(cfg, *mp, &offset); 279 if (proto < 0 || (proto == IPPROTO_ICMPV6 && 280 nptv6_translate_icmpv6(cfg, mp, offset) != 0)) 281 return (IP_FW_DENY); 282 NPTV6STAT_INC(cfg, in2ex); 283 } 284 return (0); 285 } 286 287 static int 288 nptv6_rewrite_external(struct nptv6_cfg *cfg, struct mbuf **mp, int offset) 289 { 290 struct in6_addr *addr; 291 struct ip6_hdr *ip6; 292 int idx, proto; 293 uint16_t adj; 294 295 ip6 = mtodo(*mp, offset); 296 NPTV6_IPDEBUG("offset %d, %s -> %s %d", offset, 297 inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), 298 inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), 299 ip6->ip6_nxt); 300 if (offset == 0) 301 addr = &ip6->ip6_dst; 302 else { 303 /* 304 * When we rewriting inner IPv6 header, we need to rewrite 305 * source address back to internal prefix. The datagram in 306 * the ICMPv6 payload should looks like it was send from 307 * internal prefix. 308 */ 309 addr = &ip6->ip6_src; 310 } 311 idx = nptv6_search_index(cfg, addr); 312 if (idx < 0) { 313 /* 314 * Do not send ICMPv6 error when offset isn't zero. 315 * This means we are rewriting inner IPv6 header in the 316 * ICMPv6 error message. 317 */ 318 if (offset == 0) { 319 icmp6_error2(*mp, ICMP6_DST_UNREACH, 320 ICMP6_DST_UNREACH_ADDR, 0, (*mp)->m_pkthdr.rcvif); 321 *mp = NULL; 322 } 323 return (IP_FW_DENY); 324 } 325 adj = addr->s6_addr16[idx]; 326 nptv6_copy_addr(&cfg->internal, addr, &cfg->mask); 327 adj = cksum_add(adj, ~cfg->adjustment); 328 if (adj == 0xffff) 329 adj = 0; 330 addr->s6_addr16[idx] = adj; 331 if (offset == 0) { 332 /* 333 * We may need to translate addresses in the inner IPv6 334 * header for ICMPv6 error messages. 335 */ 336 proto = nptv6_getlasthdr(cfg, *mp, &offset); 337 if (proto < 0 || (proto == IPPROTO_ICMPV6 && 338 nptv6_translate_icmpv6(cfg, mp, offset) != 0)) 339 return (IP_FW_DENY); 340 NPTV6STAT_INC(cfg, ex2in); 341 } 342 return (0); 343 } 344 345 /* 346 * ipfw external action handler. 347 */ 348 static int 349 ipfw_nptv6(struct ip_fw_chain *chain, struct ip_fw_args *args, 350 ipfw_insn *cmd, int *done) 351 { 352 struct ip6_hdr *ip6; 353 struct nptv6_cfg *cfg; 354 ipfw_insn *icmd; 355 int ret; 356 357 *done = 0; /* try next rule if not matched */ 358 ret = IP_FW_DENY; 359 icmd = cmd + 1; 360 if (cmd->opcode != O_EXTERNAL_ACTION || 361 cmd->arg1 != V_nptv6_eid || 362 icmd->opcode != O_EXTERNAL_INSTANCE || 363 (cfg = NPTV6_LOOKUP(chain, icmd)) == NULL || 364 (cfg->flags & NPTV6_READY) == 0) 365 return (ret); 366 /* 367 * We need act as router, so when forwarding is disabled - 368 * do nothing. 369 */ 370 if (V_ip6_forwarding == 0 || args->f_id.addr_type != 6) 371 return (ret); 372 /* 373 * NOTE: we expect ipfw_chk() did m_pullup() up to upper level 374 * protocol's headers. Also we skip some checks, that ip6_input(), 375 * ip6_forward(), ip6_fastfwd() and ipfw_chk() already did. 376 */ 377 ip6 = mtod(args->m, struct ip6_hdr *); 378 NPTV6_IPDEBUG("eid %u, oid %u, %s -> %s %d", 379 cmd->arg1, icmd->arg1, 380 inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), 381 inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), 382 ip6->ip6_nxt); 383 if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_src, 384 &cfg->internal, &cfg->mask)) { 385 /* 386 * XXX: Do not translate packets when both src and dst 387 * are from internal prefix. 388 */ 389 if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_dst, 390 &cfg->internal, &cfg->mask)) 391 return (ret); 392 ret = nptv6_rewrite_internal(cfg, &args->m, 0); 393 } else if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_dst, 394 &cfg->external, &cfg->mask)) 395 ret = nptv6_rewrite_external(cfg, &args->m, 0); 396 else 397 return (ret); 398 /* 399 * If address wasn't rewrited - free mbuf and terminate the search. 400 */ 401 if (ret != 0) { 402 if (args->m != NULL) { 403 m_freem(args->m); 404 args->m = NULL; /* mark mbuf as consumed */ 405 } 406 NPTV6STAT_INC(cfg, dropped); 407 *done = 1; 408 } else { 409 /* Terminate the search if one_pass is set */ 410 *done = V_fw_one_pass; 411 /* Update args->f_id when one_pass is off */ 412 if (*done == 0) { 413 ip6 = mtod(args->m, struct ip6_hdr *); 414 args->f_id.src_ip6 = ip6->ip6_src; 415 args->f_id.dst_ip6 = ip6->ip6_dst; 416 } 417 } 418 return (ret); 419 } 420 421 static struct nptv6_cfg * 422 nptv6_alloc_config(const char *name, uint8_t set) 423 { 424 struct nptv6_cfg *cfg; 425 426 cfg = malloc(sizeof(struct nptv6_cfg), M_IPFW, M_WAITOK | M_ZERO); 427 COUNTER_ARRAY_ALLOC(cfg->stats, NPTV6STATS, M_WAITOK); 428 cfg->no.name = cfg->name; 429 cfg->no.etlv = IPFW_TLV_NPTV6_NAME; 430 cfg->no.set = set; 431 strlcpy(cfg->name, name, sizeof(cfg->name)); 432 return (cfg); 433 } 434 435 static void 436 nptv6_free_config(struct nptv6_cfg *cfg) 437 { 438 439 COUNTER_ARRAY_FREE(cfg->stats, NPTV6STATS); 440 free(cfg, M_IPFW); 441 } 442 443 static void 444 nptv6_export_config(struct ip_fw_chain *ch, struct nptv6_cfg *cfg, 445 ipfw_nptv6_cfg *uc) 446 { 447 448 uc->internal = cfg->internal; 449 if (cfg->flags & NPTV6_DYNAMIC_PREFIX) 450 memcpy(uc->if_name, cfg->if_name, IF_NAMESIZE); 451 else 452 uc->external = cfg->external; 453 uc->plen = cfg->plen; 454 uc->flags = cfg->flags & NPTV6_FLAGSMASK; 455 uc->set = cfg->no.set; 456 strlcpy(uc->name, cfg->no.name, sizeof(uc->name)); 457 } 458 459 struct nptv6_dump_arg { 460 struct ip_fw_chain *ch; 461 struct sockopt_data *sd; 462 }; 463 464 static int 465 export_config_cb(struct namedobj_instance *ni, struct named_object *no, 466 void *arg) 467 { 468 struct nptv6_dump_arg *da = (struct nptv6_dump_arg *)arg; 469 ipfw_nptv6_cfg *uc; 470 471 uc = (ipfw_nptv6_cfg *)ipfw_get_sopt_space(da->sd, sizeof(*uc)); 472 nptv6_export_config(da->ch, (struct nptv6_cfg *)no, uc); 473 return (0); 474 } 475 476 static struct nptv6_cfg * 477 nptv6_find(struct namedobj_instance *ni, const char *name, uint8_t set) 478 { 479 struct nptv6_cfg *cfg; 480 481 cfg = (struct nptv6_cfg *)ipfw_objhash_lookup_name_type(ni, set, 482 IPFW_TLV_NPTV6_NAME, name); 483 484 return (cfg); 485 } 486 487 static void 488 nptv6_calculate_adjustment(struct nptv6_cfg *cfg) 489 { 490 uint16_t i, e; 491 uint16_t *p; 492 493 /* Calculate checksum of internal prefix */ 494 for (i = 0, p = (uint16_t *)&cfg->internal; 495 p < (uint16_t *)(&cfg->internal + 1); p++) 496 i = cksum_add(i, *p); 497 498 /* Calculate checksum of external prefix */ 499 for (e = 0, p = (uint16_t *)&cfg->external; 500 p < (uint16_t *)(&cfg->external + 1); p++) 501 e = cksum_add(e, *p); 502 503 /* Adjustment value for Int->Ext direction */ 504 cfg->adjustment = cksum_add(~e, i); 505 } 506 507 static int 508 nptv6_check_prefix(const struct in6_addr *addr) 509 { 510 511 if (IN6_IS_ADDR_MULTICAST(addr) || 512 IN6_IS_ADDR_LINKLOCAL(addr) || 513 IN6_IS_ADDR_LOOPBACK(addr) || 514 IN6_IS_ADDR_UNSPECIFIED(addr)) 515 return (EINVAL); 516 return (0); 517 } 518 519 static void 520 nptv6_set_external(struct nptv6_cfg *cfg, struct in6_addr *addr) 521 { 522 523 cfg->external = *addr; 524 IN6_MASK_ADDR(&cfg->external, &cfg->mask); 525 nptv6_calculate_adjustment(cfg); 526 cfg->flags |= NPTV6_READY; 527 } 528 529 /* 530 * Try to determine what prefix to use as external for 531 * configured interface name. 532 */ 533 static void 534 nptv6_find_prefix(struct ip_fw_chain *ch, struct nptv6_cfg *cfg, 535 struct ifnet *ifp) 536 { 537 struct ifaddr *ifa; 538 struct in6_ifaddr *ia; 539 540 MPASS(cfg->flags & NPTV6_DYNAMIC_PREFIX); 541 IPFW_UH_WLOCK_ASSERT(ch); 542 543 if (ifp == NULL) { 544 ifp = ifunit_ref(cfg->if_name); 545 if (ifp == NULL) 546 return; 547 } 548 if_addr_rlock(ifp); 549 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 550 if (ifa->ifa_addr->sa_family != AF_INET6) 551 continue; 552 ia = (struct in6_ifaddr *)ifa; 553 if (nptv6_check_prefix(&ia->ia_addr.sin6_addr) || 554 IN6_ARE_MASKED_ADDR_EQUAL(&ia->ia_addr.sin6_addr, 555 &cfg->internal, &cfg->mask)) 556 continue; 557 /* Suitable address is found. */ 558 nptv6_set_external(cfg, &ia->ia_addr.sin6_addr); 559 break; 560 } 561 if_addr_runlock(ifp); 562 if_rele(ifp); 563 } 564 565 struct ifaddr_event_args { 566 struct ifnet *ifp; 567 const struct in6_addr *addr; 568 int event; 569 }; 570 571 static int 572 ifaddr_cb(struct namedobj_instance *ni, struct named_object *no, 573 void *arg) 574 { 575 struct ifaddr_event_args *args; 576 struct ip_fw_chain *ch; 577 struct nptv6_cfg *cfg; 578 579 ch = &V_layer3_chain; 580 cfg = (struct nptv6_cfg *)SRV_OBJECT(ch, no->kidx); 581 if ((cfg->flags & NPTV6_DYNAMIC_PREFIX) == 0) 582 return (0); 583 584 args = arg; 585 /* If interface name doesn't match, ignore */ 586 if (strncmp(args->ifp->if_xname, cfg->if_name, IF_NAMESIZE)) 587 return (0); 588 if (args->ifp->if_flags & IFF_DYING) { /* XXX: is it possible? */ 589 cfg->flags &= ~NPTV6_READY; 590 return (0); 591 } 592 if (args->event == IFADDR_EVENT_DEL) { 593 /* If instance is not ready, ignore */ 594 if ((cfg->flags & NPTV6_READY) == 0) 595 return (0); 596 /* If address does not match the external prefix, ignore */ 597 if (IN6_ARE_MASKED_ADDR_EQUAL(&cfg->external, args->addr, 598 &cfg->mask) != 0) 599 return (0); 600 /* Otherwise clear READY flag */ 601 cfg->flags &= ~NPTV6_READY; 602 } else {/* IFADDR_EVENT_ADD */ 603 /* If instance is already ready, ignore */ 604 if (cfg->flags & NPTV6_READY) 605 return (0); 606 /* If address is not suitable for prefix, ignore */ 607 if (nptv6_check_prefix(args->addr) || 608 IN6_ARE_MASKED_ADDR_EQUAL(args->addr, &cfg->internal, 609 &cfg->mask)) 610 return (0); 611 /* FALLTHROUGH */ 612 } 613 MPASS(!(cfg->flags & NPTV6_READY)); 614 /* Try to determine the prefix */ 615 if_ref(args->ifp); 616 nptv6_find_prefix(ch, cfg, args->ifp); 617 return (0); 618 } 619 620 static void 621 nptv6_ifaddrevent_handler(void *arg __unused, struct ifnet *ifp, 622 struct ifaddr *ifa, int event) 623 { 624 struct ifaddr_event_args args; 625 struct ip_fw_chain *ch; 626 627 if (ifa->ifa_addr->sa_family != AF_INET6) 628 return; 629 630 args.ifp = ifp; 631 args.addr = &((struct sockaddr_in6 *)ifa->ifa_addr)->sin6_addr; 632 args.event = event; 633 634 ch = &V_layer3_chain; 635 IPFW_UH_WLOCK(ch); 636 ipfw_objhash_foreach_type(CHAIN_TO_SRV(ch), ifaddr_cb, &args, 637 IPFW_TLV_NPTV6_NAME); 638 IPFW_UH_WUNLOCK(ch); 639 } 640 641 /* 642 * Creates new NPTv6 instance. 643 * Data layout (v0)(current): 644 * Request: [ ipfw_obj_lheader ipfw_nptv6_cfg ] 645 * 646 * Returns 0 on success 647 */ 648 static int 649 nptv6_create(struct ip_fw_chain *ch, ip_fw3_opheader *op3, 650 struct sockopt_data *sd) 651 { 652 struct in6_addr mask; 653 ipfw_obj_lheader *olh; 654 ipfw_nptv6_cfg *uc; 655 struct namedobj_instance *ni; 656 struct nptv6_cfg *cfg; 657 658 if (sd->valsize != sizeof(*olh) + sizeof(*uc)) 659 return (EINVAL); 660 661 olh = (ipfw_obj_lheader *)sd->kbuf; 662 uc = (ipfw_nptv6_cfg *)(olh + 1); 663 if (ipfw_check_object_name_generic(uc->name) != 0) 664 return (EINVAL); 665 if (uc->plen < 8 || uc->plen > 64 || uc->set >= IPFW_MAX_SETS) 666 return (EINVAL); 667 if (nptv6_check_prefix(&uc->internal)) 668 return (EINVAL); 669 in6_prefixlen2mask(&mask, uc->plen); 670 if ((uc->flags & NPTV6_DYNAMIC_PREFIX) == 0 && ( 671 nptv6_check_prefix(&uc->external) || 672 IN6_ARE_MASKED_ADDR_EQUAL(&uc->external, &uc->internal, &mask))) 673 return (EINVAL); 674 675 ni = CHAIN_TO_SRV(ch); 676 IPFW_UH_RLOCK(ch); 677 if (nptv6_find(ni, uc->name, uc->set) != NULL) { 678 IPFW_UH_RUNLOCK(ch); 679 return (EEXIST); 680 } 681 IPFW_UH_RUNLOCK(ch); 682 683 cfg = nptv6_alloc_config(uc->name, uc->set); 684 cfg->plen = uc->plen; 685 cfg->flags = uc->flags & NPTV6_FLAGSMASK; 686 if (cfg->plen <= 48) 687 cfg->flags |= NPTV6_48PLEN; 688 cfg->mask = mask; 689 cfg->internal = uc->internal; 690 IN6_MASK_ADDR(&cfg->internal, &mask); 691 if (cfg->flags & NPTV6_DYNAMIC_PREFIX) 692 memcpy(cfg->if_name, uc->if_name, IF_NAMESIZE); 693 else 694 nptv6_set_external(cfg, &uc->external); 695 696 if ((uc->flags & NPTV6_DYNAMIC_PREFIX) != 0 && 697 nptv6_ifaddr_event == NULL) 698 nptv6_ifaddr_event = EVENTHANDLER_REGISTER( 699 ifaddr_event_ext, nptv6_ifaddrevent_handler, NULL, 700 EVENTHANDLER_PRI_ANY); 701 702 IPFW_UH_WLOCK(ch); 703 if (ipfw_objhash_alloc_idx(ni, &cfg->no.kidx) != 0) { 704 IPFW_UH_WUNLOCK(ch); 705 nptv6_free_config(cfg); 706 return (ENOSPC); 707 } 708 ipfw_objhash_add(ni, &cfg->no); 709 SRV_OBJECT(ch, cfg->no.kidx) = cfg; 710 if (cfg->flags & NPTV6_DYNAMIC_PREFIX) 711 nptv6_find_prefix(ch, cfg, NULL); 712 IPFW_UH_WUNLOCK(ch); 713 714 return (0); 715 } 716 717 /* 718 * Destroys NPTv6 instance. 719 * Data layout (v0)(current): 720 * Request: [ ipfw_obj_header ] 721 * 722 * Returns 0 on success 723 */ 724 static int 725 nptv6_destroy(struct ip_fw_chain *ch, ip_fw3_opheader *op3, 726 struct sockopt_data *sd) 727 { 728 ipfw_obj_header *oh; 729 struct nptv6_cfg *cfg; 730 731 if (sd->valsize != sizeof(*oh)) 732 return (EINVAL); 733 734 oh = (ipfw_obj_header *)sd->kbuf; 735 if (ipfw_check_object_name_generic(oh->ntlv.name) != 0) 736 return (EINVAL); 737 738 IPFW_UH_WLOCK(ch); 739 cfg = nptv6_find(CHAIN_TO_SRV(ch), oh->ntlv.name, oh->ntlv.set); 740 if (cfg == NULL) { 741 IPFW_UH_WUNLOCK(ch); 742 return (ESRCH); 743 } 744 if (cfg->no.refcnt > 0) { 745 IPFW_UH_WUNLOCK(ch); 746 return (EBUSY); 747 } 748 749 SRV_OBJECT(ch, cfg->no.kidx) = NULL; 750 ipfw_objhash_del(CHAIN_TO_SRV(ch), &cfg->no); 751 ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), cfg->no.kidx); 752 IPFW_UH_WUNLOCK(ch); 753 754 nptv6_free_config(cfg); 755 return (0); 756 } 757 758 /* 759 * Get or change nptv6 instance config. 760 * Request: [ ipfw_obj_header [ ipfw_nptv6_cfg ] ] 761 */ 762 static int 763 nptv6_config(struct ip_fw_chain *chain, ip_fw3_opheader *op, 764 struct sockopt_data *sd) 765 { 766 767 return (EOPNOTSUPP); 768 } 769 770 /* 771 * Lists all NPTv6 instances currently available in kernel. 772 * Data layout (v0)(current): 773 * Request: [ ipfw_obj_lheader ] 774 * Reply: [ ipfw_obj_lheader ipfw_nptv6_cfg x N ] 775 * 776 * Returns 0 on success 777 */ 778 static int 779 nptv6_list(struct ip_fw_chain *ch, ip_fw3_opheader *op3, 780 struct sockopt_data *sd) 781 { 782 ipfw_obj_lheader *olh; 783 struct nptv6_dump_arg da; 784 785 /* Check minimum header size */ 786 if (sd->valsize < sizeof(ipfw_obj_lheader)) 787 return (EINVAL); 788 789 olh = (ipfw_obj_lheader *)ipfw_get_sopt_header(sd, sizeof(*olh)); 790 791 IPFW_UH_RLOCK(ch); 792 olh->count = ipfw_objhash_count_type(CHAIN_TO_SRV(ch), 793 IPFW_TLV_NPTV6_NAME); 794 olh->objsize = sizeof(ipfw_nptv6_cfg); 795 olh->size = sizeof(*olh) + olh->count * olh->objsize; 796 797 if (sd->valsize < olh->size) { 798 IPFW_UH_RUNLOCK(ch); 799 return (ENOMEM); 800 } 801 memset(&da, 0, sizeof(da)); 802 da.ch = ch; 803 da.sd = sd; 804 ipfw_objhash_foreach_type(CHAIN_TO_SRV(ch), export_config_cb, 805 &da, IPFW_TLV_NPTV6_NAME); 806 IPFW_UH_RUNLOCK(ch); 807 808 return (0); 809 } 810 811 #define __COPY_STAT_FIELD(_cfg, _stats, _field) \ 812 (_stats)->_field = NPTV6STAT_FETCH(_cfg, _field) 813 static void 814 export_stats(struct ip_fw_chain *ch, struct nptv6_cfg *cfg, 815 struct ipfw_nptv6_stats *stats) 816 { 817 818 __COPY_STAT_FIELD(cfg, stats, in2ex); 819 __COPY_STAT_FIELD(cfg, stats, ex2in); 820 __COPY_STAT_FIELD(cfg, stats, dropped); 821 } 822 823 /* 824 * Get NPTv6 statistics. 825 * Data layout (v0)(current): 826 * Request: [ ipfw_obj_header ] 827 * Reply: [ ipfw_obj_header ipfw_obj_ctlv [ uint64_t x N ]] 828 * 829 * Returns 0 on success 830 */ 831 static int 832 nptv6_stats(struct ip_fw_chain *ch, ip_fw3_opheader *op, 833 struct sockopt_data *sd) 834 { 835 struct ipfw_nptv6_stats stats; 836 struct nptv6_cfg *cfg; 837 ipfw_obj_header *oh; 838 ipfw_obj_ctlv *ctlv; 839 size_t sz; 840 841 sz = sizeof(ipfw_obj_header) + sizeof(ipfw_obj_ctlv) + sizeof(stats); 842 if (sd->valsize % sizeof(uint64_t)) 843 return (EINVAL); 844 if (sd->valsize < sz) 845 return (ENOMEM); 846 oh = (ipfw_obj_header *)ipfw_get_sopt_header(sd, sz); 847 if (oh == NULL) 848 return (EINVAL); 849 if (ipfw_check_object_name_generic(oh->ntlv.name) != 0 || 850 oh->ntlv.set >= IPFW_MAX_SETS) 851 return (EINVAL); 852 memset(&stats, 0, sizeof(stats)); 853 854 IPFW_UH_RLOCK(ch); 855 cfg = nptv6_find(CHAIN_TO_SRV(ch), oh->ntlv.name, oh->ntlv.set); 856 if (cfg == NULL) { 857 IPFW_UH_RUNLOCK(ch); 858 return (ESRCH); 859 } 860 export_stats(ch, cfg, &stats); 861 IPFW_UH_RUNLOCK(ch); 862 863 ctlv = (ipfw_obj_ctlv *)(oh + 1); 864 memset(ctlv, 0, sizeof(*ctlv)); 865 ctlv->head.type = IPFW_TLV_COUNTERS; 866 ctlv->head.length = sz - sizeof(ipfw_obj_header); 867 ctlv->count = sizeof(stats) / sizeof(uint64_t); 868 ctlv->objsize = sizeof(uint64_t); 869 ctlv->version = 1; 870 memcpy(ctlv + 1, &stats, sizeof(stats)); 871 return (0); 872 } 873 874 /* 875 * Reset NPTv6 statistics. 876 * Data layout (v0)(current): 877 * Request: [ ipfw_obj_header ] 878 * 879 * Returns 0 on success 880 */ 881 static int 882 nptv6_reset_stats(struct ip_fw_chain *ch, ip_fw3_opheader *op, 883 struct sockopt_data *sd) 884 { 885 struct nptv6_cfg *cfg; 886 ipfw_obj_header *oh; 887 888 if (sd->valsize != sizeof(*oh)) 889 return (EINVAL); 890 oh = (ipfw_obj_header *)sd->kbuf; 891 if (ipfw_check_object_name_generic(oh->ntlv.name) != 0 || 892 oh->ntlv.set >= IPFW_MAX_SETS) 893 return (EINVAL); 894 895 IPFW_UH_WLOCK(ch); 896 cfg = nptv6_find(CHAIN_TO_SRV(ch), oh->ntlv.name, oh->ntlv.set); 897 if (cfg == NULL) { 898 IPFW_UH_WUNLOCK(ch); 899 return (ESRCH); 900 } 901 COUNTER_ARRAY_ZERO(cfg->stats, NPTV6STATS); 902 IPFW_UH_WUNLOCK(ch); 903 return (0); 904 } 905 906 static struct ipfw_sopt_handler scodes[] = { 907 { IP_FW_NPTV6_CREATE, 0, HDIR_SET, nptv6_create }, 908 { IP_FW_NPTV6_DESTROY,0, HDIR_SET, nptv6_destroy }, 909 { IP_FW_NPTV6_CONFIG, 0, HDIR_BOTH, nptv6_config }, 910 { IP_FW_NPTV6_LIST, 0, HDIR_GET, nptv6_list }, 911 { IP_FW_NPTV6_STATS, 0, HDIR_GET, nptv6_stats }, 912 { IP_FW_NPTV6_RESET_STATS,0, HDIR_SET, nptv6_reset_stats }, 913 }; 914 915 static int 916 nptv6_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) 917 { 918 ipfw_insn *icmd; 919 920 icmd = cmd - 1; 921 NPTV6_DEBUG("opcode %d, arg1 %d, opcode0 %d, arg1 %d", 922 cmd->opcode, cmd->arg1, icmd->opcode, icmd->arg1); 923 if (icmd->opcode != O_EXTERNAL_ACTION || 924 icmd->arg1 != V_nptv6_eid) 925 return (1); 926 927 *puidx = cmd->arg1; 928 *ptype = 0; 929 return (0); 930 } 931 932 static void 933 nptv6_update_arg1(ipfw_insn *cmd, uint16_t idx) 934 { 935 936 cmd->arg1 = idx; 937 NPTV6_DEBUG("opcode %d, arg1 -> %d", cmd->opcode, cmd->arg1); 938 } 939 940 static int 941 nptv6_findbyname(struct ip_fw_chain *ch, struct tid_info *ti, 942 struct named_object **pno) 943 { 944 int err; 945 946 err = ipfw_objhash_find_type(CHAIN_TO_SRV(ch), ti, 947 IPFW_TLV_NPTV6_NAME, pno); 948 NPTV6_DEBUG("uidx %u, type %u, err %d", ti->uidx, ti->type, err); 949 return (err); 950 } 951 952 static struct named_object * 953 nptv6_findbykidx(struct ip_fw_chain *ch, uint16_t idx) 954 { 955 struct namedobj_instance *ni; 956 struct named_object *no; 957 958 IPFW_UH_WLOCK_ASSERT(ch); 959 ni = CHAIN_TO_SRV(ch); 960 no = ipfw_objhash_lookup_kidx(ni, idx); 961 KASSERT(no != NULL, ("NPT with index %d not found", idx)); 962 963 NPTV6_DEBUG("kidx %u -> %s", idx, no->name); 964 return (no); 965 } 966 967 static int 968 nptv6_manage_sets(struct ip_fw_chain *ch, uint16_t set, uint8_t new_set, 969 enum ipfw_sets_cmd cmd) 970 { 971 972 return (ipfw_obj_manage_sets(CHAIN_TO_SRV(ch), IPFW_TLV_NPTV6_NAME, 973 set, new_set, cmd)); 974 } 975 976 static struct opcode_obj_rewrite opcodes[] = { 977 { 978 .opcode = O_EXTERNAL_INSTANCE, 979 .etlv = IPFW_TLV_EACTION /* just show it isn't table */, 980 .classifier = nptv6_classify, 981 .update = nptv6_update_arg1, 982 .find_byname = nptv6_findbyname, 983 .find_bykidx = nptv6_findbykidx, 984 .manage_sets = nptv6_manage_sets, 985 }, 986 }; 987 988 static int 989 destroy_config_cb(struct namedobj_instance *ni, struct named_object *no, 990 void *arg) 991 { 992 struct nptv6_cfg *cfg; 993 struct ip_fw_chain *ch; 994 995 ch = (struct ip_fw_chain *)arg; 996 IPFW_UH_WLOCK_ASSERT(ch); 997 998 cfg = (struct nptv6_cfg *)SRV_OBJECT(ch, no->kidx); 999 SRV_OBJECT(ch, no->kidx) = NULL; 1000 ipfw_objhash_del(ni, &cfg->no); 1001 ipfw_objhash_free_idx(ni, cfg->no.kidx); 1002 nptv6_free_config(cfg); 1003 return (0); 1004 } 1005 1006 int 1007 nptv6_init(struct ip_fw_chain *ch, int first) 1008 { 1009 1010 V_nptv6_eid = ipfw_add_eaction(ch, ipfw_nptv6, "nptv6"); 1011 if (V_nptv6_eid == 0) 1012 return (ENXIO); 1013 IPFW_ADD_SOPT_HANDLER(first, scodes); 1014 IPFW_ADD_OBJ_REWRITER(first, opcodes); 1015 return (0); 1016 } 1017 1018 void 1019 nptv6_uninit(struct ip_fw_chain *ch, int last) 1020 { 1021 1022 if (last && nptv6_ifaddr_event != NULL) 1023 EVENTHANDLER_DEREGISTER(ifaddr_event_ext, nptv6_ifaddr_event); 1024 IPFW_DEL_OBJ_REWRITER(last, opcodes); 1025 IPFW_DEL_SOPT_HANDLER(last, scodes); 1026 ipfw_del_eaction(ch, V_nptv6_eid); 1027 /* 1028 * Since we already have deregistered external action, 1029 * our named objects become unaccessible via rules, because 1030 * all rules were truncated by ipfw_del_eaction(). 1031 * So, we can unlink and destroy our named objects without holding 1032 * IPFW_WLOCK(). 1033 */ 1034 IPFW_UH_WLOCK(ch); 1035 ipfw_objhash_foreach_type(CHAIN_TO_SRV(ch), destroy_config_cb, ch, 1036 IPFW_TLV_NPTV6_NAME); 1037 V_nptv6_eid = 0; 1038 IPFW_UH_WUNLOCK(ch); 1039 } 1040 1041