ip_fw2.c revision 122921
1/* 2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 * 25 * $FreeBSD: head/sys/netinet/ip_fw2.c 122921 2003-11-20 19:47:31Z andre $ 26 */ 27 28#define DEB(x) 29#define DDB(x) x 30 31/* 32 * Implement IP packet firewall (new version) 33 */ 34 35#if !defined(KLD_MODULE) 36#include "opt_ipfw.h" 37#include "opt_ipdn.h" 38#include "opt_ipdivert.h" 39#include "opt_inet.h" 40#ifndef INET 41#error IPFIREWALL requires INET. 42#endif /* INET */ 43#endif 44 45#define IPFW2 1 46#if IPFW2 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/malloc.h> 50#include <sys/mbuf.h> 51#include <sys/kernel.h> 52#include <sys/proc.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55#include <sys/sysctl.h> 56#include <sys/syslog.h> 57#include <sys/ucred.h> 58#include <net/if.h> 59#include <net/route.h> 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/in_var.h> 63#include <netinet/in_pcb.h> 64#include <netinet/ip.h> 65#include <netinet/ip_var.h> 66#include <netinet/ip_icmp.h> 67#include <netinet/ip_fw.h> 68#include <netinet/ip_dummynet.h> 69#include <netinet/tcp.h> 70#include <netinet/tcp_timer.h> 71#include <netinet/tcp_var.h> 72#include <netinet/tcpip.h> 73#include <netinet/udp.h> 74#include <netinet/udp_var.h> 75 76#ifdef IPSEC 77#include <netinet6/ipsec.h> 78#endif 79 80#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */ 81 82#include <machine/in_cksum.h> /* XXX for in_cksum */ 83 84/* 85 * XXX This one should go in sys/mbuf.h. It is used to avoid that 86 * a firewall-generated packet loops forever through the firewall. 87 */ 88#ifndef M_SKIP_FIREWALL 89#define M_SKIP_FIREWALL 0x4000 90#endif 91 92/* 93 * set_disable contains one bit per set value (0..31). 94 * If the bit is set, all rules with the corresponding set 95 * are disabled. Set RESVD_SET(31) is reserved for the default rule 96 * and rules that are not deleted by the flush command, 97 * and CANNOT be disabled. 98 * Rules in set RESVD_SET can only be deleted explicitly. 99 */ 100static u_int32_t set_disable; 101 102static int fw_verbose; 103static int verbose_limit; 104 105static struct callout ipfw_timeout; 106#define IPFW_DEFAULT_RULE 65535 107 108struct ip_fw_chain { 109 struct ip_fw *rules; /* list of rules */ 110 struct ip_fw *reap; /* list of rules to reap */ 111 struct mtx mtx; /* lock guarding rule list */ 112}; 113#define IPFW_LOCK_INIT(_chain) \ 114 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \ 115 MTX_DEF | MTX_RECURSE) 116#define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx) 117#define IPFW_LOCK(_chain) mtx_lock(&(_chain)->mtx) 118#define IPFW_UNLOCK(_chain) mtx_unlock(&(_chain)->mtx) 119#define IPFW_LOCK_ASSERT(_chain) mtx_assert(&(_chain)->mtx, MA_OWNED) 120 121/* 122 * list of rules for layer 3 123 */ 124static struct ip_fw_chain layer3_chain; 125 126MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's"); 127 128static int fw_debug = 1; 129static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */ 130 131#ifdef SYSCTL_NODE 132SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); 133SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, 134 CTLFLAG_RW | CTLFLAG_SECURE3, 135 &fw_enable, 0, "Enable ipfw"); 136SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW, 137 &autoinc_step, 0, "Rule number autincrement step"); 138SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass, 139 CTLFLAG_RW | CTLFLAG_SECURE3, 140 &fw_one_pass, 0, 141 "Only do a single pass through ipfw when using dummynet(4)"); 142SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW, 143 &fw_debug, 0, "Enable printing of debug ip_fw statements"); 144SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, 145 CTLFLAG_RW | CTLFLAG_SECURE3, 146 &fw_verbose, 0, "Log matches to ipfw rules"); 147SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW, 148 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged"); 149 150/* 151 * Description of dynamic rules. 152 * 153 * Dynamic rules are stored in lists accessed through a hash table 154 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 155 * be modified through the sysctl variable dyn_buckets which is 156 * updated when the table becomes empty. 157 * 158 * XXX currently there is only one list, ipfw_dyn. 159 * 160 * When a packet is received, its address fields are first masked 161 * with the mask defined for the rule, then hashed, then matched 162 * against the entries in the corresponding list. 163 * Dynamic rules can be used for different purposes: 164 * + stateful rules; 165 * + enforcing limits on the number of sessions; 166 * + in-kernel NAT (not implemented yet) 167 * 168 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 169 * measured in seconds and depending on the flags. 170 * 171 * The total number of dynamic rules is stored in dyn_count. 172 * The max number of dynamic rules is dyn_max. When we reach 173 * the maximum number of rules we do not create anymore. This is 174 * done to avoid consuming too much memory, but also too much 175 * time when searching on each packet (ideally, we should try instead 176 * to put a limit on the length of the list on each bucket...). 177 * 178 * Each dynamic rule holds a pointer to the parent ipfw rule so 179 * we know what action to perform. Dynamic rules are removed when 180 * the parent rule is deleted. XXX we should make them survive. 181 * 182 * There are some limitations with dynamic rules -- we do not 183 * obey the 'randomized match', and we do not do multiple 184 * passes through the firewall. XXX check the latter!!! 185 */ 186static ipfw_dyn_rule **ipfw_dyn_v = NULL; 187static u_int32_t dyn_buckets = 256; /* must be power of 2 */ 188static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */ 189 190static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */ 191#define IPFW_DYN_LOCK_INIT() \ 192 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF) 193#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx) 194#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx) 195#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx) 196#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED) 197 198/* 199 * Timeouts for various events in handing dynamic rules. 200 */ 201static u_int32_t dyn_ack_lifetime = 300; 202static u_int32_t dyn_syn_lifetime = 20; 203static u_int32_t dyn_fin_lifetime = 1; 204static u_int32_t dyn_rst_lifetime = 1; 205static u_int32_t dyn_udp_lifetime = 10; 206static u_int32_t dyn_short_lifetime = 5; 207 208/* 209 * Keepalives are sent if dyn_keepalive is set. They are sent every 210 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 211 * seconds of lifetime of a rule. 212 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 213 * than dyn_keepalive_period. 214 */ 215 216static u_int32_t dyn_keepalive_interval = 20; 217static u_int32_t dyn_keepalive_period = 5; 218static u_int32_t dyn_keepalive = 1; /* do send keepalives */ 219 220static u_int32_t static_count; /* # of static rules */ 221static u_int32_t static_len; /* size in bytes of static rules */ 222static u_int32_t dyn_count; /* # of dynamic rules */ 223static u_int32_t dyn_max = 4096; /* max # of dynamic rules */ 224 225SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW, 226 &dyn_buckets, 0, "Number of dyn. buckets"); 227SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD, 228 &curr_dyn_buckets, 0, "Current Number of dyn. buckets"); 229SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD, 230 &dyn_count, 0, "Number of dyn. rules"); 231SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW, 232 &dyn_max, 0, "Max number of dyn. rules"); 233SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD, 234 &static_count, 0, "Number of static rules"); 235SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW, 236 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks"); 237SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW, 238 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn"); 239SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW, 240 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin"); 241SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW, 242 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst"); 243SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW, 244 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP"); 245SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW, 246 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations"); 247SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW, 248 &dyn_keepalive, 0, "Enable keepalives for dyn. rules"); 249 250#endif /* SYSCTL_NODE */ 251 252 253static ip_fw_chk_t ipfw_chk; 254 255ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL; /* hook into dummynet */ 256 257/* 258 * This macro maps an ip pointer into a layer3 header pointer of type T 259 */ 260#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl)) 261 262static __inline int 263icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd) 264{ 265 int type = L3HDR(struct icmp,ip)->icmp_type; 266 267 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) ); 268} 269 270#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \ 271 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) ) 272 273static int 274is_icmp_query(struct ip *ip) 275{ 276 int type = L3HDR(struct icmp, ip)->icmp_type; 277 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) ); 278} 279#undef TT 280 281/* 282 * The following checks use two arrays of 8 or 16 bits to store the 283 * bits that we want set or clear, respectively. They are in the 284 * low and high half of cmd->arg1 or cmd->d[0]. 285 * 286 * We scan options and store the bits we find set. We succeed if 287 * 288 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear 289 * 290 * The code is sometimes optimized not to store additional variables. 291 */ 292 293static int 294flags_match(ipfw_insn *cmd, u_int8_t bits) 295{ 296 u_char want_clear; 297 bits = ~bits; 298 299 if ( ((cmd->arg1 & 0xff) & bits) != 0) 300 return 0; /* some bits we want set were clear */ 301 want_clear = (cmd->arg1 >> 8) & 0xff; 302 if ( (want_clear & bits) != want_clear) 303 return 0; /* some bits we want clear were set */ 304 return 1; 305} 306 307static int 308ipopts_match(struct ip *ip, ipfw_insn *cmd) 309{ 310 int optlen, bits = 0; 311 u_char *cp = (u_char *)(ip + 1); 312 int x = (ip->ip_hl << 2) - sizeof (struct ip); 313 314 for (; x > 0; x -= optlen, cp += optlen) { 315 int opt = cp[IPOPT_OPTVAL]; 316 317 if (opt == IPOPT_EOL) 318 break; 319 if (opt == IPOPT_NOP) 320 optlen = 1; 321 else { 322 optlen = cp[IPOPT_OLEN]; 323 if (optlen <= 0 || optlen > x) 324 return 0; /* invalid or truncated */ 325 } 326 switch (opt) { 327 328 default: 329 break; 330 331 case IPOPT_LSRR: 332 bits |= IP_FW_IPOPT_LSRR; 333 break; 334 335 case IPOPT_SSRR: 336 bits |= IP_FW_IPOPT_SSRR; 337 break; 338 339 case IPOPT_RR: 340 bits |= IP_FW_IPOPT_RR; 341 break; 342 343 case IPOPT_TS: 344 bits |= IP_FW_IPOPT_TS; 345 break; 346 } 347 } 348 return (flags_match(cmd, bits)); 349} 350 351static int 352tcpopts_match(struct ip *ip, ipfw_insn *cmd) 353{ 354 int optlen, bits = 0; 355 struct tcphdr *tcp = L3HDR(struct tcphdr,ip); 356 u_char *cp = (u_char *)(tcp + 1); 357 int x = (tcp->th_off << 2) - sizeof(struct tcphdr); 358 359 for (; x > 0; x -= optlen, cp += optlen) { 360 int opt = cp[0]; 361 if (opt == TCPOPT_EOL) 362 break; 363 if (opt == TCPOPT_NOP) 364 optlen = 1; 365 else { 366 optlen = cp[1]; 367 if (optlen <= 0) 368 break; 369 } 370 371 switch (opt) { 372 373 default: 374 break; 375 376 case TCPOPT_MAXSEG: 377 bits |= IP_FW_TCPOPT_MSS; 378 break; 379 380 case TCPOPT_WINDOW: 381 bits |= IP_FW_TCPOPT_WINDOW; 382 break; 383 384 case TCPOPT_SACK_PERMITTED: 385 case TCPOPT_SACK: 386 bits |= IP_FW_TCPOPT_SACK; 387 break; 388 389 case TCPOPT_TIMESTAMP: 390 bits |= IP_FW_TCPOPT_TS; 391 break; 392 393 case TCPOPT_CC: 394 case TCPOPT_CCNEW: 395 case TCPOPT_CCECHO: 396 bits |= IP_FW_TCPOPT_CC; 397 break; 398 } 399 } 400 return (flags_match(cmd, bits)); 401} 402 403static int 404iface_match(struct ifnet *ifp, ipfw_insn_if *cmd) 405{ 406 if (ifp == NULL) /* no iface with this packet, match fails */ 407 return 0; 408 /* Check by name or by IP address */ 409 if (cmd->name[0] != '\0') { /* match by name */ 410 /* Check name */ 411 if (cmd->p.glob) { 412 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0) 413 return(1); 414 } else { 415 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0) 416 return(1); 417 } 418 } else { 419 struct ifaddr *ia; 420 421 /* XXX lock? */ 422 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) { 423 if (ia->ifa_addr == NULL) 424 continue; 425 if (ia->ifa_addr->sa_family != AF_INET) 426 continue; 427 if (cmd->p.ip.s_addr == ((struct sockaddr_in *) 428 (ia->ifa_addr))->sin_addr.s_addr) 429 return(1); /* match */ 430 } 431 } 432 return(0); /* no match, fail ... */ 433} 434 435/* 436 * The 'verrevpath' option checks that the interface that an IP packet 437 * arrives on is the same interface that traffic destined for the 438 * packet's source address would be routed out of. This is a measure 439 * to block forged packets. This is also commonly known as "anti-spoofing" 440 * or Unicast Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The 441 * name of the knob is purposely reminisent of the Cisco IOS command, 442 * 443 * ip verify unicast reverse-path 444 * 445 * which implements the same functionality. But note that syntax is 446 * misleading. The check may be performed on all IP packets whether unicast, 447 * multicast, or broadcast. 448 */ 449static int 450verify_rev_path(struct in_addr src, struct ifnet *ifp) 451{ 452 static struct route ro; 453 struct sockaddr_in *dst; 454 455 dst = (struct sockaddr_in *)&(ro.ro_dst); 456 457 /* Check if we've cached the route from the previous call. */ 458 if (src.s_addr != dst->sin_addr.s_addr) { 459 ro.ro_rt = NULL; 460 461 bzero(dst, sizeof(*dst)); 462 dst->sin_family = AF_INET; 463 dst->sin_len = sizeof(*dst); 464 dst->sin_addr = src; 465 466 rtalloc_ign(&ro, RTF_CLONING); 467 } 468 469 if ((ro.ro_rt == NULL) || (ifp == NULL) || 470 (ro.ro_rt->rt_ifp->if_index != ifp->if_index)) 471 return 0; 472 473 return 1; 474} 475 476 477static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */ 478 479#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0 480#define SNP(buf) buf, sizeof(buf) 481 482/* 483 * We enter here when we have a rule with O_LOG. 484 * XXX this function alone takes about 2Kbytes of code! 485 */ 486static void 487ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh, 488 struct mbuf *m, struct ifnet *oif) 489{ 490 char *action; 491 int limit_reached = 0; 492 char action2[40], proto[48], fragment[28]; 493 494 fragment[0] = '\0'; 495 proto[0] = '\0'; 496 497 if (f == NULL) { /* bogus pkt */ 498 if (verbose_limit != 0 && norule_counter >= verbose_limit) 499 return; 500 norule_counter++; 501 if (norule_counter == verbose_limit) 502 limit_reached = verbose_limit; 503 action = "Refuse"; 504 } else { /* O_LOG is the first action, find the real one */ 505 ipfw_insn *cmd = ACTION_PTR(f); 506 ipfw_insn_log *l = (ipfw_insn_log *)cmd; 507 508 if (l->max_log != 0 && l->log_left == 0) 509 return; 510 l->log_left--; 511 if (l->log_left == 0) 512 limit_reached = l->max_log; 513 cmd += F_LEN(cmd); /* point to first action */ 514 if (cmd->opcode == O_PROB) 515 cmd += F_LEN(cmd); 516 517 action = action2; 518 switch (cmd->opcode) { 519 case O_DENY: 520 action = "Deny"; 521 break; 522 523 case O_REJECT: 524 if (cmd->arg1==ICMP_REJECT_RST) 525 action = "Reset"; 526 else if (cmd->arg1==ICMP_UNREACH_HOST) 527 action = "Reject"; 528 else 529 snprintf(SNPARGS(action2, 0), "Unreach %d", 530 cmd->arg1); 531 break; 532 533 case O_ACCEPT: 534 action = "Accept"; 535 break; 536 case O_COUNT: 537 action = "Count"; 538 break; 539 case O_DIVERT: 540 snprintf(SNPARGS(action2, 0), "Divert %d", 541 cmd->arg1); 542 break; 543 case O_TEE: 544 snprintf(SNPARGS(action2, 0), "Tee %d", 545 cmd->arg1); 546 break; 547 case O_SKIPTO: 548 snprintf(SNPARGS(action2, 0), "SkipTo %d", 549 cmd->arg1); 550 break; 551 case O_PIPE: 552 snprintf(SNPARGS(action2, 0), "Pipe %d", 553 cmd->arg1); 554 break; 555 case O_QUEUE: 556 snprintf(SNPARGS(action2, 0), "Queue %d", 557 cmd->arg1); 558 break; 559 case O_FORWARD_IP: { 560 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd; 561 int len; 562 563 len = snprintf(SNPARGS(action2, 0), "Forward to %s", 564 inet_ntoa(sa->sa.sin_addr)); 565 if (sa->sa.sin_port) 566 snprintf(SNPARGS(action2, len), ":%d", 567 sa->sa.sin_port); 568 } 569 break; 570 default: 571 action = "UNKNOWN"; 572 break; 573 } 574 } 575 576 if (hlen == 0) { /* non-ip */ 577 snprintf(SNPARGS(proto, 0), "MAC"); 578 } else { 579 struct ip *ip = mtod(m, struct ip *); 580 /* these three are all aliases to the same thing */ 581 struct icmp *const icmp = L3HDR(struct icmp, ip); 582 struct tcphdr *const tcp = (struct tcphdr *)icmp; 583 struct udphdr *const udp = (struct udphdr *)icmp; 584 585 int ip_off, offset, ip_len; 586 587 int len; 588 589 if (eh != NULL) { /* layer 2 packets are as on the wire */ 590 ip_off = ntohs(ip->ip_off); 591 ip_len = ntohs(ip->ip_len); 592 } else { 593 ip_off = ip->ip_off; 594 ip_len = ip->ip_len; 595 } 596 offset = ip_off & IP_OFFMASK; 597 switch (ip->ip_p) { 598 case IPPROTO_TCP: 599 len = snprintf(SNPARGS(proto, 0), "TCP %s", 600 inet_ntoa(ip->ip_src)); 601 if (offset == 0) 602 snprintf(SNPARGS(proto, len), ":%d %s:%d", 603 ntohs(tcp->th_sport), 604 inet_ntoa(ip->ip_dst), 605 ntohs(tcp->th_dport)); 606 else 607 snprintf(SNPARGS(proto, len), " %s", 608 inet_ntoa(ip->ip_dst)); 609 break; 610 611 case IPPROTO_UDP: 612 len = snprintf(SNPARGS(proto, 0), "UDP %s", 613 inet_ntoa(ip->ip_src)); 614 if (offset == 0) 615 snprintf(SNPARGS(proto, len), ":%d %s:%d", 616 ntohs(udp->uh_sport), 617 inet_ntoa(ip->ip_dst), 618 ntohs(udp->uh_dport)); 619 else 620 snprintf(SNPARGS(proto, len), " %s", 621 inet_ntoa(ip->ip_dst)); 622 break; 623 624 case IPPROTO_ICMP: 625 if (offset == 0) 626 len = snprintf(SNPARGS(proto, 0), 627 "ICMP:%u.%u ", 628 icmp->icmp_type, icmp->icmp_code); 629 else 630 len = snprintf(SNPARGS(proto, 0), "ICMP "); 631 len += snprintf(SNPARGS(proto, len), "%s", 632 inet_ntoa(ip->ip_src)); 633 snprintf(SNPARGS(proto, len), " %s", 634 inet_ntoa(ip->ip_dst)); 635 break; 636 637 default: 638 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p, 639 inet_ntoa(ip->ip_src)); 640 snprintf(SNPARGS(proto, len), " %s", 641 inet_ntoa(ip->ip_dst)); 642 break; 643 } 644 645 if (ip_off & (IP_MF | IP_OFFMASK)) 646 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)", 647 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2), 648 offset << 3, 649 (ip_off & IP_MF) ? "+" : ""); 650 } 651 if (oif || m->m_pkthdr.rcvif) 652 log(LOG_SECURITY | LOG_INFO, 653 "ipfw: %d %s %s %s via %s%s\n", 654 f ? f->rulenum : -1, 655 action, proto, oif ? "out" : "in", 656 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname, 657 fragment); 658 else 659 log(LOG_SECURITY | LOG_INFO, 660 "ipfw: %d %s %s [no if info]%s\n", 661 f ? f->rulenum : -1, 662 action, proto, fragment); 663 if (limit_reached) 664 log(LOG_SECURITY | LOG_NOTICE, 665 "ipfw: limit %d reached on entry %d\n", 666 limit_reached, f ? f->rulenum : -1); 667} 668 669/* 670 * IMPORTANT: the hash function for dynamic rules must be commutative 671 * in source and destination (ip,port), because rules are bidirectional 672 * and we want to find both in the same bucket. 673 */ 674static __inline int 675hash_packet(struct ipfw_flow_id *id) 676{ 677 u_int32_t i; 678 679 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 680 i &= (curr_dyn_buckets - 1); 681 return i; 682} 683 684/** 685 * unlink a dynamic rule from a chain. prev is a pointer to 686 * the previous one, q is a pointer to the rule to delete, 687 * head is a pointer to the head of the queue. 688 * Modifies q and potentially also head. 689 */ 690#define UNLINK_DYN_RULE(prev, head, q) { \ 691 ipfw_dyn_rule *old_q = q; \ 692 \ 693 /* remove a refcount to the parent */ \ 694 if (q->dyn_type == O_LIMIT) \ 695 q->parent->count--; \ 696 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\ 697 (q->id.src_ip), (q->id.src_port), \ 698 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \ 699 if (prev != NULL) \ 700 prev->next = q = q->next; \ 701 else \ 702 head = q = q->next; \ 703 dyn_count--; \ 704 free(old_q, M_IPFW); } 705 706#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 707 708/** 709 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 710 * 711 * If keep_me == NULL, rules are deleted even if not expired, 712 * otherwise only expired rules are removed. 713 * 714 * The value of the second parameter is also used to point to identify 715 * a rule we absolutely do not want to remove (e.g. because we are 716 * holding a reference to it -- this is the case with O_LIMIT_PARENT 717 * rules). The pointer is only used for comparison, so any non-null 718 * value will do. 719 */ 720static void 721remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 722{ 723 static u_int32_t last_remove = 0; 724 725#define FORCE (keep_me == NULL) 726 727 ipfw_dyn_rule *prev, *q; 728 int i, pass = 0, max_pass = 0; 729 730 IPFW_DYN_LOCK_ASSERT(); 731 732 if (ipfw_dyn_v == NULL || dyn_count == 0) 733 return; 734 /* do not expire more than once per second, it is useless */ 735 if (!FORCE && last_remove == time_second) 736 return; 737 last_remove = time_second; 738 739 /* 740 * because O_LIMIT refer to parent rules, during the first pass only 741 * remove child and mark any pending LIMIT_PARENT, and remove 742 * them in a second pass. 743 */ 744next_pass: 745 for (i = 0 ; i < curr_dyn_buckets ; i++) { 746 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) { 747 /* 748 * Logic can become complex here, so we split tests. 749 */ 750 if (q == keep_me) 751 goto next; 752 if (rule != NULL && rule != q->rule) 753 goto next; /* not the one we are looking for */ 754 if (q->dyn_type == O_LIMIT_PARENT) { 755 /* 756 * handle parent in the second pass, 757 * record we need one. 758 */ 759 max_pass = 1; 760 if (pass == 0) 761 goto next; 762 if (FORCE && q->count != 0 ) { 763 /* XXX should not happen! */ 764 printf("ipfw: OUCH! cannot remove rule," 765 " count %d\n", q->count); 766 } 767 } else { 768 if (!FORCE && 769 !TIME_LEQ( q->expire, time_second )) 770 goto next; 771 } 772 if (q->dyn_type != O_LIMIT_PARENT || !q->count) { 773 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 774 continue; 775 } 776next: 777 prev=q; 778 q=q->next; 779 } 780 } 781 if (pass++ < max_pass) 782 goto next_pass; 783} 784 785 786/** 787 * lookup a dynamic rule. 788 */ 789static ipfw_dyn_rule * 790lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction, 791 struct tcphdr *tcp) 792{ 793 /* 794 * stateful ipfw extensions. 795 * Lookup into dynamic session queue 796 */ 797#define MATCH_REVERSE 0 798#define MATCH_FORWARD 1 799#define MATCH_NONE 2 800#define MATCH_UNKNOWN 3 801 int i, dir = MATCH_NONE; 802 ipfw_dyn_rule *prev, *q=NULL; 803 804 IPFW_DYN_LOCK_ASSERT(); 805 806 if (ipfw_dyn_v == NULL) 807 goto done; /* not found */ 808 i = hash_packet( pkt ); 809 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) { 810 if (q->dyn_type == O_LIMIT_PARENT && q->count) 811 goto next; 812 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */ 813 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 814 continue; 815 } 816 if (pkt->proto == q->id.proto && 817 q->dyn_type != O_LIMIT_PARENT) { 818 if (pkt->src_ip == q->id.src_ip && 819 pkt->dst_ip == q->id.dst_ip && 820 pkt->src_port == q->id.src_port && 821 pkt->dst_port == q->id.dst_port ) { 822 dir = MATCH_FORWARD; 823 break; 824 } 825 if (pkt->src_ip == q->id.dst_ip && 826 pkt->dst_ip == q->id.src_ip && 827 pkt->src_port == q->id.dst_port && 828 pkt->dst_port == q->id.src_port ) { 829 dir = MATCH_REVERSE; 830 break; 831 } 832 } 833next: 834 prev = q; 835 q = q->next; 836 } 837 if (q == NULL) 838 goto done; /* q = NULL, not found */ 839 840 if ( prev != NULL) { /* found and not in front */ 841 prev->next = q->next; 842 q->next = ipfw_dyn_v[i]; 843 ipfw_dyn_v[i] = q; 844 } 845 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 846 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST); 847 848#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 849#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 850 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8); 851 switch (q->state) { 852 case TH_SYN: /* opening */ 853 q->expire = time_second + dyn_syn_lifetime; 854 break; 855 856 case BOTH_SYN: /* move to established */ 857 case BOTH_SYN | TH_FIN : /* one side tries to close */ 858 case BOTH_SYN | (TH_FIN << 8) : 859 if (tcp) { 860#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 861 u_int32_t ack = ntohl(tcp->th_ack); 862 if (dir == MATCH_FORWARD) { 863 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd)) 864 q->ack_fwd = ack; 865 else { /* ignore out-of-sequence */ 866 break; 867 } 868 } else { 869 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev)) 870 q->ack_rev = ack; 871 else { /* ignore out-of-sequence */ 872 break; 873 } 874 } 875 } 876 q->expire = time_second + dyn_ack_lifetime; 877 break; 878 879 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 880 if (dyn_fin_lifetime >= dyn_keepalive_period) 881 dyn_fin_lifetime = dyn_keepalive_period - 1; 882 q->expire = time_second + dyn_fin_lifetime; 883 break; 884 885 default: 886#if 0 887 /* 888 * reset or some invalid combination, but can also 889 * occur if we use keep-state the wrong way. 890 */ 891 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 892 printf("invalid state: 0x%x\n", q->state); 893#endif 894 if (dyn_rst_lifetime >= dyn_keepalive_period) 895 dyn_rst_lifetime = dyn_keepalive_period - 1; 896 q->expire = time_second + dyn_rst_lifetime; 897 break; 898 } 899 } else if (pkt->proto == IPPROTO_UDP) { 900 q->expire = time_second + dyn_udp_lifetime; 901 } else { 902 /* other protocols */ 903 q->expire = time_second + dyn_short_lifetime; 904 } 905done: 906 if (match_direction) 907 *match_direction = dir; 908 return q; 909} 910 911static ipfw_dyn_rule * 912lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 913 struct tcphdr *tcp) 914{ 915 ipfw_dyn_rule *q; 916 917 IPFW_DYN_LOCK(); 918 q = lookup_dyn_rule_locked(pkt, match_direction, tcp); 919 if (q == NULL) 920 IPFW_DYN_UNLOCK(); 921 /* NB: return table locked when q is not NULL */ 922 return q; 923} 924 925static void 926realloc_dynamic_table(void) 927{ 928 IPFW_DYN_LOCK_ASSERT(); 929 930 /* 931 * Try reallocation, make sure we have a power of 2 and do 932 * not allow more than 64k entries. In case of overflow, 933 * default to 1024. 934 */ 935 936 if (dyn_buckets > 65536) 937 dyn_buckets = 1024; 938 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */ 939 dyn_buckets = curr_dyn_buckets; /* reset */ 940 return; 941 } 942 curr_dyn_buckets = dyn_buckets; 943 if (ipfw_dyn_v != NULL) 944 free(ipfw_dyn_v, M_IPFW); 945 for (;;) { 946 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 947 M_IPFW, M_NOWAIT | M_ZERO); 948 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2) 949 break; 950 curr_dyn_buckets /= 2; 951 } 952} 953 954/** 955 * Install state of type 'type' for a dynamic session. 956 * The hash table contains two type of rules: 957 * - regular rules (O_KEEP_STATE) 958 * - rules for sessions with limited number of sess per user 959 * (O_LIMIT). When they are created, the parent is 960 * increased by 1, and decreased on delete. In this case, 961 * the third parameter is the parent rule and not the chain. 962 * - "parent" rules for the above (O_LIMIT_PARENT). 963 */ 964static ipfw_dyn_rule * 965add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule) 966{ 967 ipfw_dyn_rule *r; 968 int i; 969 970 IPFW_DYN_LOCK_ASSERT(); 971 972 if (ipfw_dyn_v == NULL || 973 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) { 974 realloc_dynamic_table(); 975 if (ipfw_dyn_v == NULL) 976 return NULL; /* failed ! */ 977 } 978 i = hash_packet(id); 979 980 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO); 981 if (r == NULL) { 982 printf ("ipfw: sorry cannot allocate state\n"); 983 return NULL; 984 } 985 986 /* increase refcount on parent, and set pointer */ 987 if (dyn_type == O_LIMIT) { 988 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 989 if ( parent->dyn_type != O_LIMIT_PARENT) 990 panic("invalid parent"); 991 parent->count++; 992 r->parent = parent; 993 rule = parent->rule; 994 } 995 996 r->id = *id; 997 r->expire = time_second + dyn_syn_lifetime; 998 r->rule = rule; 999 r->dyn_type = dyn_type; 1000 r->pcnt = r->bcnt = 0; 1001 r->count = 0; 1002 1003 r->bucket = i; 1004 r->next = ipfw_dyn_v[i]; 1005 ipfw_dyn_v[i] = r; 1006 dyn_count++; 1007 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n", 1008 dyn_type, 1009 (r->id.src_ip), (r->id.src_port), 1010 (r->id.dst_ip), (r->id.dst_port), 1011 dyn_count ); ) 1012 return r; 1013} 1014 1015/** 1016 * lookup dynamic parent rule using pkt and rule as search keys. 1017 * If the lookup fails, then install one. 1018 */ 1019static ipfw_dyn_rule * 1020lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 1021{ 1022 ipfw_dyn_rule *q; 1023 int i; 1024 1025 IPFW_DYN_LOCK_ASSERT(); 1026 1027 if (ipfw_dyn_v) { 1028 i = hash_packet( pkt ); 1029 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next) 1030 if (q->dyn_type == O_LIMIT_PARENT && 1031 rule== q->rule && 1032 pkt->proto == q->id.proto && 1033 pkt->src_ip == q->id.src_ip && 1034 pkt->dst_ip == q->id.dst_ip && 1035 pkt->src_port == q->id.src_port && 1036 pkt->dst_port == q->id.dst_port) { 1037 q->expire = time_second + dyn_short_lifetime; 1038 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);) 1039 return q; 1040 } 1041 } 1042 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 1043} 1044 1045/** 1046 * Install dynamic state for rule type cmd->o.opcode 1047 * 1048 * Returns 1 (failure) if state is not installed because of errors or because 1049 * session limitations are enforced. 1050 */ 1051static int 1052install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 1053 struct ip_fw_args *args) 1054{ 1055 static int last_log; 1056 1057 ipfw_dyn_rule *q; 1058 1059 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n", 1060 cmd->o.opcode, 1061 (args->f_id.src_ip), (args->f_id.src_port), 1062 (args->f_id.dst_ip), (args->f_id.dst_port) );) 1063 1064 IPFW_DYN_LOCK(); 1065 1066 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL); 1067 1068 if (q != NULL) { /* should never occur */ 1069 if (last_log != time_second) { 1070 last_log = time_second; 1071 printf("ipfw: install_state: entry already present, done\n"); 1072 } 1073 IPFW_DYN_UNLOCK(); 1074 return 0; 1075 } 1076 1077 if (dyn_count >= dyn_max) 1078 /* 1079 * Run out of slots, try to remove any expired rule. 1080 */ 1081 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1); 1082 1083 if (dyn_count >= dyn_max) { 1084 if (last_log != time_second) { 1085 last_log = time_second; 1086 printf("ipfw: install_state: Too many dynamic rules\n"); 1087 } 1088 IPFW_DYN_UNLOCK(); 1089 return 1; /* cannot install, notify caller */ 1090 } 1091 1092 switch (cmd->o.opcode) { 1093 case O_KEEP_STATE: /* bidir rule */ 1094 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule); 1095 break; 1096 1097 case O_LIMIT: /* limit number of sessions */ 1098 { 1099 u_int16_t limit_mask = cmd->limit_mask; 1100 struct ipfw_flow_id id; 1101 ipfw_dyn_rule *parent; 1102 1103 DEB(printf("ipfw: installing dyn-limit rule %d\n", 1104 cmd->conn_limit);) 1105 1106 id.dst_ip = id.src_ip = 0; 1107 id.dst_port = id.src_port = 0; 1108 id.proto = args->f_id.proto; 1109 1110 if (limit_mask & DYN_SRC_ADDR) 1111 id.src_ip = args->f_id.src_ip; 1112 if (limit_mask & DYN_DST_ADDR) 1113 id.dst_ip = args->f_id.dst_ip; 1114 if (limit_mask & DYN_SRC_PORT) 1115 id.src_port = args->f_id.src_port; 1116 if (limit_mask & DYN_DST_PORT) 1117 id.dst_port = args->f_id.dst_port; 1118 parent = lookup_dyn_parent(&id, rule); 1119 if (parent == NULL) { 1120 printf("ipfw: add parent failed\n"); 1121 return 1; 1122 } 1123 if (parent->count >= cmd->conn_limit) { 1124 /* 1125 * See if we can remove some expired rule. 1126 */ 1127 remove_dyn_rule(rule, parent); 1128 if (parent->count >= cmd->conn_limit) { 1129 if (fw_verbose && last_log != time_second) { 1130 last_log = time_second; 1131 log(LOG_SECURITY | LOG_DEBUG, 1132 "drop session, too many entries\n"); 1133 } 1134 IPFW_DYN_UNLOCK(); 1135 return 1; 1136 } 1137 } 1138 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent); 1139 } 1140 break; 1141 default: 1142 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode); 1143 IPFW_DYN_UNLOCK(); 1144 return 1; 1145 } 1146 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */ 1147 IPFW_DYN_UNLOCK(); 1148 return 0; 1149} 1150 1151/* 1152 * Transmit a TCP packet, containing either a RST or a keepalive. 1153 * When flags & TH_RST, we are sending a RST packet, because of a 1154 * "reset" action matched the packet. 1155 * Otherwise we are sending a keepalive, and flags & TH_ 1156 */ 1157static void 1158send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags) 1159{ 1160 struct mbuf *m; 1161 struct ip *ip; 1162 struct tcphdr *tcp; 1163 1164 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1165 if (m == 0) 1166 return; 1167 m->m_pkthdr.rcvif = (struct ifnet *)0; 1168 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); 1169 m->m_data += max_linkhdr; 1170 1171 ip = mtod(m, struct ip *); 1172 bzero(ip, m->m_len); 1173 tcp = (struct tcphdr *)(ip + 1); /* no IP options */ 1174 ip->ip_p = IPPROTO_TCP; 1175 tcp->th_off = 5; 1176 /* 1177 * Assume we are sending a RST (or a keepalive in the reverse 1178 * direction), swap src and destination addresses and ports. 1179 */ 1180 ip->ip_src.s_addr = htonl(id->dst_ip); 1181 ip->ip_dst.s_addr = htonl(id->src_ip); 1182 tcp->th_sport = htons(id->dst_port); 1183 tcp->th_dport = htons(id->src_port); 1184 if (flags & TH_RST) { /* we are sending a RST */ 1185 if (flags & TH_ACK) { 1186 tcp->th_seq = htonl(ack); 1187 tcp->th_ack = htonl(0); 1188 tcp->th_flags = TH_RST; 1189 } else { 1190 if (flags & TH_SYN) 1191 seq++; 1192 tcp->th_seq = htonl(0); 1193 tcp->th_ack = htonl(seq); 1194 tcp->th_flags = TH_RST | TH_ACK; 1195 } 1196 } else { 1197 /* 1198 * We are sending a keepalive. flags & TH_SYN determines 1199 * the direction, forward if set, reverse if clear. 1200 * NOTE: seq and ack are always assumed to be correct 1201 * as set by the caller. This may be confusing... 1202 */ 1203 if (flags & TH_SYN) { 1204 /* 1205 * we have to rewrite the correct addresses! 1206 */ 1207 ip->ip_dst.s_addr = htonl(id->dst_ip); 1208 ip->ip_src.s_addr = htonl(id->src_ip); 1209 tcp->th_dport = htons(id->dst_port); 1210 tcp->th_sport = htons(id->src_port); 1211 } 1212 tcp->th_seq = htonl(seq); 1213 tcp->th_ack = htonl(ack); 1214 tcp->th_flags = TH_ACK; 1215 } 1216 /* 1217 * set ip_len to the payload size so we can compute 1218 * the tcp checksum on the pseudoheader 1219 * XXX check this, could save a couple of words ? 1220 */ 1221 ip->ip_len = htons(sizeof(struct tcphdr)); 1222 tcp->th_sum = in_cksum(m, m->m_pkthdr.len); 1223 /* 1224 * now fill fields left out earlier 1225 */ 1226 ip->ip_ttl = ip_defttl; 1227 ip->ip_len = m->m_pkthdr.len; 1228 m->m_flags |= M_SKIP_FIREWALL; 1229 ip_output(m, NULL, NULL, 0, NULL, NULL); 1230} 1231 1232/* 1233 * sends a reject message, consuming the mbuf passed as an argument. 1234 */ 1235static void 1236send_reject(struct ip_fw_args *args, int code, int offset, int ip_len) 1237{ 1238 1239 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */ 1240 /* We need the IP header in host order for icmp_error(). */ 1241 if (args->eh != NULL) { 1242 struct ip *ip = mtod(args->m, struct ip *); 1243 ip->ip_len = ntohs(ip->ip_len); 1244 ip->ip_off = ntohs(ip->ip_off); 1245 } 1246 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); 1247 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) { 1248 struct tcphdr *const tcp = 1249 L3HDR(struct tcphdr, mtod(args->m, struct ip *)); 1250 if ( (tcp->th_flags & TH_RST) == 0) 1251 send_pkt(&(args->f_id), ntohl(tcp->th_seq), 1252 ntohl(tcp->th_ack), 1253 tcp->th_flags | TH_RST); 1254 m_freem(args->m); 1255 } else 1256 m_freem(args->m); 1257 args->m = NULL; 1258} 1259 1260/** 1261 * 1262 * Given an ip_fw *, lookup_next_rule will return a pointer 1263 * to the next rule, which can be either the jump 1264 * target (for skipto instructions) or the next one in the list (in 1265 * all other cases including a missing jump target). 1266 * The result is also written in the "next_rule" field of the rule. 1267 * Backward jumps are not allowed, so start looking from the next 1268 * rule... 1269 * 1270 * This never returns NULL -- in case we do not have an exact match, 1271 * the next rule is returned. When the ruleset is changed, 1272 * pointers are flushed so we are always correct. 1273 */ 1274 1275static struct ip_fw * 1276lookup_next_rule(struct ip_fw *me) 1277{ 1278 struct ip_fw *rule = NULL; 1279 ipfw_insn *cmd; 1280 1281 /* look for action, in case it is a skipto */ 1282 cmd = ACTION_PTR(me); 1283 if (cmd->opcode == O_LOG) 1284 cmd += F_LEN(cmd); 1285 if ( cmd->opcode == O_SKIPTO ) 1286 for (rule = me->next; rule ; rule = rule->next) 1287 if (rule->rulenum >= cmd->arg1) 1288 break; 1289 if (rule == NULL) /* failure or not a skipto */ 1290 rule = me->next; 1291 me->next_rule = rule; 1292 return rule; 1293} 1294 1295static int 1296check_uidgid(ipfw_insn_u32 *insn, 1297 int proto, struct ifnet *oif, 1298 struct in_addr dst_ip, u_int16_t dst_port, 1299 struct in_addr src_ip, u_int16_t src_port) 1300{ 1301 struct inpcbinfo *pi; 1302 int wildcard; 1303 struct inpcb *pcb; 1304 int match; 1305 1306 if (proto == IPPROTO_TCP) { 1307 wildcard = 0; 1308 pi = &tcbinfo; 1309 } else if (proto == IPPROTO_UDP) { 1310 wildcard = 1; 1311 pi = &udbinfo; 1312 } else 1313 return 0; 1314 1315 match = 0; 1316 1317 INP_INFO_RLOCK(pi); /* XXX LOR with IPFW */ 1318 pcb = (oif) ? 1319 in_pcblookup_hash(pi, 1320 dst_ip, htons(dst_port), 1321 src_ip, htons(src_port), 1322 wildcard, oif) : 1323 in_pcblookup_hash(pi, 1324 src_ip, htons(src_port), 1325 dst_ip, htons(dst_port), 1326 wildcard, NULL); 1327 if (pcb != NULL) { 1328 INP_LOCK(pcb); 1329 if (pcb->inp_socket != NULL) { 1330#if __FreeBSD_version < 500034 1331#define socheckuid(a,b) ((a)->so_cred->cr_uid != (b)) 1332#endif 1333 if (insn->o.opcode == O_UID) { 1334 match = !socheckuid(pcb->inp_socket, 1335 (uid_t)insn->d[0]); 1336 } else { 1337 match = groupmember((uid_t)insn->d[0], 1338 pcb->inp_socket->so_cred); 1339 } 1340 } 1341 INP_UNLOCK(pcb); 1342 } 1343 INP_INFO_RUNLOCK(pi); 1344 1345 return match; 1346} 1347 1348/* 1349 * The main check routine for the firewall. 1350 * 1351 * All arguments are in args so we can modify them and return them 1352 * back to the caller. 1353 * 1354 * Parameters: 1355 * 1356 * args->m (in/out) The packet; we set to NULL when/if we nuke it. 1357 * Starts with the IP header. 1358 * args->eh (in) Mac header if present, or NULL for layer3 packet. 1359 * args->oif Outgoing interface, or NULL if packet is incoming. 1360 * The incoming interface is in the mbuf. (in) 1361 * args->divert_rule (in/out) 1362 * Skip up to the first rule past this rule number; 1363 * upon return, non-zero port number for divert or tee. 1364 * 1365 * args->rule Pointer to the last matching rule (in/out) 1366 * args->next_hop Socket we are forwarding to (out). 1367 * args->f_id Addresses grabbed from the packet (out) 1368 * 1369 * Return value: 1370 * 1371 * IP_FW_PORT_DENY_FLAG the packet must be dropped. 1372 * 0 The packet is to be accepted and routed normally OR 1373 * the packet was denied/rejected and has been dropped; 1374 * in the latter case, *m is equal to NULL upon return. 1375 * port Divert the packet to port, with these caveats: 1376 * 1377 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead 1378 * of diverting it (ie, 'ipfw tee'). 1379 * 1380 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower 1381 * 16 bits as a dummynet pipe number instead of diverting 1382 */ 1383 1384static int 1385ipfw_chk(struct ip_fw_args *args) 1386{ 1387 /* 1388 * Local variables hold state during the processing of a packet. 1389 * 1390 * IMPORTANT NOTE: to speed up the processing of rules, there 1391 * are some assumption on the values of the variables, which 1392 * are documented here. Should you change them, please check 1393 * the implementation of the various instructions to make sure 1394 * that they still work. 1395 * 1396 * args->eh The MAC header. It is non-null for a layer2 1397 * packet, it is NULL for a layer-3 packet. 1398 * 1399 * m | args->m Pointer to the mbuf, as received from the caller. 1400 * It may change if ipfw_chk() does an m_pullup, or if it 1401 * consumes the packet because it calls send_reject(). 1402 * XXX This has to change, so that ipfw_chk() never modifies 1403 * or consumes the buffer. 1404 * ip is simply an alias of the value of m, and it is kept 1405 * in sync with it (the packet is supposed to start with 1406 * the ip header). 1407 */ 1408 struct mbuf *m = args->m; 1409 struct ip *ip = mtod(m, struct ip *); 1410 1411 /* 1412 * oif | args->oif If NULL, ipfw_chk has been called on the 1413 * inbound path (ether_input, bdg_forward, ip_input). 1414 * If non-NULL, ipfw_chk has been called on the outbound path 1415 * (ether_output, ip_output). 1416 */ 1417 struct ifnet *oif = args->oif; 1418 1419 struct ip_fw *f = NULL; /* matching rule */ 1420 int retval = 0; 1421 1422 /* 1423 * hlen The length of the IPv4 header. 1424 * hlen >0 means we have an IPv4 packet. 1425 */ 1426 u_int hlen = 0; /* hlen >0 means we have an IP pkt */ 1427 1428 /* 1429 * offset The offset of a fragment. offset != 0 means that 1430 * we have a fragment at this offset of an IPv4 packet. 1431 * offset == 0 means that (if this is an IPv4 packet) 1432 * this is the first or only fragment. 1433 */ 1434 u_short offset = 0; 1435 1436 /* 1437 * Local copies of addresses. They are only valid if we have 1438 * an IP packet. 1439 * 1440 * proto The protocol. Set to 0 for non-ip packets, 1441 * or to the protocol read from the packet otherwise. 1442 * proto != 0 means that we have an IPv4 packet. 1443 * 1444 * src_port, dst_port port numbers, in HOST format. Only 1445 * valid for TCP and UDP packets. 1446 * 1447 * src_ip, dst_ip ip addresses, in NETWORK format. 1448 * Only valid for IPv4 packets. 1449 */ 1450 u_int8_t proto; 1451 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */ 1452 struct in_addr src_ip, dst_ip; /* NOTE: network format */ 1453 u_int16_t ip_len=0; 1454 int pktlen; 1455 int dyn_dir = MATCH_UNKNOWN; 1456 ipfw_dyn_rule *q = NULL; 1457 struct ip_fw_chain *chain = &layer3_chain; 1458 1459 if (m->m_flags & M_SKIP_FIREWALL) 1460 return 0; /* accept */ 1461 /* 1462 * dyn_dir = MATCH_UNKNOWN when rules unchecked, 1463 * MATCH_NONE when checked and not matched (q = NULL), 1464 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL) 1465 */ 1466 1467 pktlen = m->m_pkthdr.len; 1468 if (args->eh == NULL || /* layer 3 packet */ 1469 ( m->m_pkthdr.len >= sizeof(struct ip) && 1470 ntohs(args->eh->ether_type) == ETHERTYPE_IP)) 1471 hlen = ip->ip_hl << 2; 1472 1473 /* 1474 * Collect parameters into local variables for faster matching. 1475 */ 1476 if (hlen == 0) { /* do not grab addresses for non-ip pkts */ 1477 proto = args->f_id.proto = 0; /* mark f_id invalid */ 1478 goto after_ip_checks; 1479 } 1480 1481 proto = args->f_id.proto = ip->ip_p; 1482 src_ip = ip->ip_src; 1483 dst_ip = ip->ip_dst; 1484 if (args->eh != NULL) { /* layer 2 packets are as on the wire */ 1485 offset = ntohs(ip->ip_off) & IP_OFFMASK; 1486 ip_len = ntohs(ip->ip_len); 1487 } else { 1488 offset = ip->ip_off & IP_OFFMASK; 1489 ip_len = ip->ip_len; 1490 } 1491 pktlen = ip_len < pktlen ? ip_len : pktlen; 1492 1493#define PULLUP_TO(len) \ 1494 do { \ 1495 if ((m)->m_len < (len)) { \ 1496 args->m = m = m_pullup(m, (len)); \ 1497 if (m == 0) \ 1498 goto pullup_failed; \ 1499 ip = mtod(m, struct ip *); \ 1500 } \ 1501 } while (0) 1502 1503 if (offset == 0) { 1504 switch (proto) { 1505 case IPPROTO_TCP: 1506 { 1507 struct tcphdr *tcp; 1508 1509 PULLUP_TO(hlen + sizeof(struct tcphdr)); 1510 tcp = L3HDR(struct tcphdr, ip); 1511 dst_port = tcp->th_dport; 1512 src_port = tcp->th_sport; 1513 args->f_id.flags = tcp->th_flags; 1514 } 1515 break; 1516 1517 case IPPROTO_UDP: 1518 { 1519 struct udphdr *udp; 1520 1521 PULLUP_TO(hlen + sizeof(struct udphdr)); 1522 udp = L3HDR(struct udphdr, ip); 1523 dst_port = udp->uh_dport; 1524 src_port = udp->uh_sport; 1525 } 1526 break; 1527 1528 case IPPROTO_ICMP: 1529 PULLUP_TO(hlen + 4); /* type, code and checksum. */ 1530 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type; 1531 break; 1532 1533 default: 1534 break; 1535 } 1536#undef PULLUP_TO 1537 } 1538 1539 args->f_id.src_ip = ntohl(src_ip.s_addr); 1540 args->f_id.dst_ip = ntohl(dst_ip.s_addr); 1541 args->f_id.src_port = src_port = ntohs(src_port); 1542 args->f_id.dst_port = dst_port = ntohs(dst_port); 1543 1544after_ip_checks: 1545 IPFW_LOCK(chain); /* XXX expensive? can we run lock free? */ 1546 if (args->rule) { 1547 /* 1548 * Packet has already been tagged. Look for the next rule 1549 * to restart processing. 1550 * 1551 * If fw_one_pass != 0 then just accept it. 1552 * XXX should not happen here, but optimized out in 1553 * the caller. 1554 */ 1555 if (fw_one_pass) { 1556 IPFW_UNLOCK(chain); /* XXX optimize */ 1557 return 0; 1558 } 1559 1560 f = args->rule->next_rule; 1561 if (f == NULL) 1562 f = lookup_next_rule(args->rule); 1563 } else { 1564 /* 1565 * Find the starting rule. It can be either the first 1566 * one, or the one after divert_rule if asked so. 1567 */ 1568 int skipto = args->divert_rule; 1569 1570 f = chain->rules; 1571 if (args->eh == NULL && skipto != 0) { 1572 if (skipto >= IPFW_DEFAULT_RULE) { 1573 IPFW_UNLOCK(chain); 1574 return(IP_FW_PORT_DENY_FLAG); /* invalid */ 1575 } 1576 while (f && f->rulenum <= skipto) 1577 f = f->next; 1578 if (f == NULL) { /* drop packet */ 1579 IPFW_UNLOCK(chain); 1580 return(IP_FW_PORT_DENY_FLAG); 1581 } 1582 } 1583 } 1584 args->divert_rule = 0; /* reset to avoid confusion later */ 1585 1586 /* 1587 * Now scan the rules, and parse microinstructions for each rule. 1588 */ 1589 for (; f; f = f->next) { 1590 int l, cmdlen; 1591 ipfw_insn *cmd; 1592 int skip_or; /* skip rest of OR block */ 1593 1594again: 1595 if (set_disable & (1 << f->set) ) 1596 continue; 1597 1598 skip_or = 0; 1599 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ; 1600 l -= cmdlen, cmd += cmdlen) { 1601 int match; 1602 1603 /* 1604 * check_body is a jump target used when we find a 1605 * CHECK_STATE, and need to jump to the body of 1606 * the target rule. 1607 */ 1608 1609check_body: 1610 cmdlen = F_LEN(cmd); 1611 /* 1612 * An OR block (insn_1 || .. || insn_n) has the 1613 * F_OR bit set in all but the last instruction. 1614 * The first match will set "skip_or", and cause 1615 * the following instructions to be skipped until 1616 * past the one with the F_OR bit clear. 1617 */ 1618 if (skip_or) { /* skip this instruction */ 1619 if ((cmd->len & F_OR) == 0) 1620 skip_or = 0; /* next one is good */ 1621 continue; 1622 } 1623 match = 0; /* set to 1 if we succeed */ 1624 1625 switch (cmd->opcode) { 1626 /* 1627 * The first set of opcodes compares the packet's 1628 * fields with some pattern, setting 'match' if a 1629 * match is found. At the end of the loop there is 1630 * logic to deal with F_NOT and F_OR flags associated 1631 * with the opcode. 1632 */ 1633 case O_NOP: 1634 match = 1; 1635 break; 1636 1637 case O_FORWARD_MAC: 1638 printf("ipfw: opcode %d unimplemented\n", 1639 cmd->opcode); 1640 break; 1641 1642 case O_GID: 1643 case O_UID: 1644 /* 1645 * We only check offset == 0 && proto != 0, 1646 * as this ensures that we have an IPv4 1647 * packet with the ports info. 1648 */ 1649 if (offset!=0) 1650 break; 1651 if (proto == IPPROTO_TCP || 1652 proto == IPPROTO_UDP) 1653 match = check_uidgid( 1654 (ipfw_insn_u32 *)cmd, 1655 proto, oif, 1656 dst_ip, dst_port, 1657 src_ip, src_port); 1658 break; 1659 1660 case O_RECV: 1661 match = iface_match(m->m_pkthdr.rcvif, 1662 (ipfw_insn_if *)cmd); 1663 break; 1664 1665 case O_XMIT: 1666 match = iface_match(oif, (ipfw_insn_if *)cmd); 1667 break; 1668 1669 case O_VIA: 1670 match = iface_match(oif ? oif : 1671 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd); 1672 break; 1673 1674 case O_MACADDR2: 1675 if (args->eh != NULL) { /* have MAC header */ 1676 u_int32_t *want = (u_int32_t *) 1677 ((ipfw_insn_mac *)cmd)->addr; 1678 u_int32_t *mask = (u_int32_t *) 1679 ((ipfw_insn_mac *)cmd)->mask; 1680 u_int32_t *hdr = (u_int32_t *)args->eh; 1681 1682 match = 1683 ( want[0] == (hdr[0] & mask[0]) && 1684 want[1] == (hdr[1] & mask[1]) && 1685 want[2] == (hdr[2] & mask[2]) ); 1686 } 1687 break; 1688 1689 case O_MAC_TYPE: 1690 if (args->eh != NULL) { 1691 u_int16_t t = 1692 ntohs(args->eh->ether_type); 1693 u_int16_t *p = 1694 ((ipfw_insn_u16 *)cmd)->ports; 1695 int i; 1696 1697 for (i = cmdlen - 1; !match && i>0; 1698 i--, p += 2) 1699 match = (t>=p[0] && t<=p[1]); 1700 } 1701 break; 1702 1703 case O_FRAG: 1704 match = (hlen > 0 && offset != 0); 1705 break; 1706 1707 case O_IN: /* "out" is "not in" */ 1708 match = (oif == NULL); 1709 break; 1710 1711 case O_LAYER2: 1712 match = (args->eh != NULL); 1713 break; 1714 1715 case O_PROTO: 1716 /* 1717 * We do not allow an arg of 0 so the 1718 * check of "proto" only suffices. 1719 */ 1720 match = (proto == cmd->arg1); 1721 break; 1722 1723 case O_IP_SRC: 1724 match = (hlen > 0 && 1725 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1726 src_ip.s_addr); 1727 break; 1728 1729 case O_IP_SRC_MASK: 1730 case O_IP_DST_MASK: 1731 if (hlen > 0) { 1732 uint32_t a = 1733 (cmd->opcode == O_IP_DST_MASK) ? 1734 dst_ip.s_addr : src_ip.s_addr; 1735 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d; 1736 int i = cmdlen-1; 1737 1738 for (; !match && i>0; i-= 2, p+= 2) 1739 match = (p[0] == (a & p[1])); 1740 } 1741 break; 1742 1743 case O_IP_SRC_ME: 1744 if (hlen > 0) { 1745 struct ifnet *tif; 1746 1747 INADDR_TO_IFP(src_ip, tif); 1748 match = (tif != NULL); 1749 } 1750 break; 1751 1752 case O_IP_DST_SET: 1753 case O_IP_SRC_SET: 1754 if (hlen > 0) { 1755 u_int32_t *d = (u_int32_t *)(cmd+1); 1756 u_int32_t addr = 1757 cmd->opcode == O_IP_DST_SET ? 1758 args->f_id.dst_ip : 1759 args->f_id.src_ip; 1760 1761 if (addr < d[0]) 1762 break; 1763 addr -= d[0]; /* subtract base */ 1764 match = (addr < cmd->arg1) && 1765 ( d[ 1 + (addr>>5)] & 1766 (1<<(addr & 0x1f)) ); 1767 } 1768 break; 1769 1770 case O_IP_DST: 1771 match = (hlen > 0 && 1772 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1773 dst_ip.s_addr); 1774 break; 1775 1776 case O_IP_DST_ME: 1777 if (hlen > 0) { 1778 struct ifnet *tif; 1779 1780 INADDR_TO_IFP(dst_ip, tif); 1781 match = (tif != NULL); 1782 } 1783 break; 1784 1785 case O_IP_SRCPORT: 1786 case O_IP_DSTPORT: 1787 /* 1788 * offset == 0 && proto != 0 is enough 1789 * to guarantee that we have an IPv4 1790 * packet with port info. 1791 */ 1792 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP) 1793 && offset == 0) { 1794 u_int16_t x = 1795 (cmd->opcode == O_IP_SRCPORT) ? 1796 src_port : dst_port ; 1797 u_int16_t *p = 1798 ((ipfw_insn_u16 *)cmd)->ports; 1799 int i; 1800 1801 for (i = cmdlen - 1; !match && i>0; 1802 i--, p += 2) 1803 match = (x>=p[0] && x<=p[1]); 1804 } 1805 break; 1806 1807 case O_ICMPTYPE: 1808 match = (offset == 0 && proto==IPPROTO_ICMP && 1809 icmptype_match(ip, (ipfw_insn_u32 *)cmd) ); 1810 break; 1811 1812 case O_IPOPT: 1813 match = (hlen > 0 && ipopts_match(ip, cmd) ); 1814 break; 1815 1816 case O_IPVER: 1817 match = (hlen > 0 && cmd->arg1 == ip->ip_v); 1818 break; 1819 1820 case O_IPID: 1821 case O_IPLEN: 1822 case O_IPTTL: 1823 if (hlen > 0) { /* only for IP packets */ 1824 uint16_t x; 1825 uint16_t *p; 1826 int i; 1827 1828 if (cmd->opcode == O_IPLEN) 1829 x = ip_len; 1830 else if (cmd->opcode == O_IPTTL) 1831 x = ip->ip_ttl; 1832 else /* must be IPID */ 1833 x = ntohs(ip->ip_id); 1834 if (cmdlen == 1) { 1835 match = (cmd->arg1 == x); 1836 break; 1837 } 1838 /* otherwise we have ranges */ 1839 p = ((ipfw_insn_u16 *)cmd)->ports; 1840 i = cmdlen - 1; 1841 for (; !match && i>0; i--, p += 2) 1842 match = (x >= p[0] && x <= p[1]); 1843 } 1844 break; 1845 1846 case O_IPPRECEDENCE: 1847 match = (hlen > 0 && 1848 (cmd->arg1 == (ip->ip_tos & 0xe0)) ); 1849 break; 1850 1851 case O_IPTOS: 1852 match = (hlen > 0 && 1853 flags_match(cmd, ip->ip_tos)); 1854 break; 1855 1856 case O_TCPFLAGS: 1857 match = (proto == IPPROTO_TCP && offset == 0 && 1858 flags_match(cmd, 1859 L3HDR(struct tcphdr,ip)->th_flags)); 1860 break; 1861 1862 case O_TCPOPTS: 1863 match = (proto == IPPROTO_TCP && offset == 0 && 1864 tcpopts_match(ip, cmd)); 1865 break; 1866 1867 case O_TCPSEQ: 1868 match = (proto == IPPROTO_TCP && offset == 0 && 1869 ((ipfw_insn_u32 *)cmd)->d[0] == 1870 L3HDR(struct tcphdr,ip)->th_seq); 1871 break; 1872 1873 case O_TCPACK: 1874 match = (proto == IPPROTO_TCP && offset == 0 && 1875 ((ipfw_insn_u32 *)cmd)->d[0] == 1876 L3HDR(struct tcphdr,ip)->th_ack); 1877 break; 1878 1879 case O_TCPWIN: 1880 match = (proto == IPPROTO_TCP && offset == 0 && 1881 cmd->arg1 == 1882 L3HDR(struct tcphdr,ip)->th_win); 1883 break; 1884 1885 case O_ESTAB: 1886 /* reject packets which have SYN only */ 1887 /* XXX should i also check for TH_ACK ? */ 1888 match = (proto == IPPROTO_TCP && offset == 0 && 1889 (L3HDR(struct tcphdr,ip)->th_flags & 1890 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); 1891 break; 1892 1893 case O_LOG: 1894 if (fw_verbose) 1895 ipfw_log(f, hlen, args->eh, m, oif); 1896 match = 1; 1897 break; 1898 1899 case O_PROB: 1900 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]); 1901 break; 1902 1903 case O_VERREVPATH: 1904 /* Outgoing packets automatically pass/match */ 1905 match = ((oif != NULL) || 1906 (m->m_pkthdr.rcvif == NULL) || 1907 verify_rev_path(src_ip, m->m_pkthdr.rcvif)); 1908 break; 1909 1910 case O_IPSEC: 1911#ifdef FAST_IPSEC 1912 match = (m_tag_find(m, 1913 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL); 1914#endif 1915#ifdef IPSEC 1916 match = (ipsec_getnhist(m) != NULL); 1917#endif 1918 /* otherwise no match */ 1919 break; 1920 1921 /* 1922 * The second set of opcodes represents 'actions', 1923 * i.e. the terminal part of a rule once the packet 1924 * matches all previous patterns. 1925 * Typically there is only one action for each rule, 1926 * and the opcode is stored at the end of the rule 1927 * (but there are exceptions -- see below). 1928 * 1929 * In general, here we set retval and terminate the 1930 * outer loop (would be a 'break 3' in some language, 1931 * but we need to do a 'goto done'). 1932 * 1933 * Exceptions: 1934 * O_COUNT and O_SKIPTO actions: 1935 * instead of terminating, we jump to the next rule 1936 * ('goto next_rule', equivalent to a 'break 2'), 1937 * or to the SKIPTO target ('goto again' after 1938 * having set f, cmd and l), respectively. 1939 * 1940 * O_LIMIT and O_KEEP_STATE: these opcodes are 1941 * not real 'actions', and are stored right 1942 * before the 'action' part of the rule. 1943 * These opcodes try to install an entry in the 1944 * state tables; if successful, we continue with 1945 * the next opcode (match=1; break;), otherwise 1946 * the packet * must be dropped 1947 * ('goto done' after setting retval); 1948 * 1949 * O_PROBE_STATE and O_CHECK_STATE: these opcodes 1950 * cause a lookup of the state table, and a jump 1951 * to the 'action' part of the parent rule 1952 * ('goto check_body') if an entry is found, or 1953 * (CHECK_STATE only) a jump to the next rule if 1954 * the entry is not found ('goto next_rule'). 1955 * The result of the lookup is cached to make 1956 * further instances of these opcodes are 1957 * effectively NOPs. 1958 */ 1959 case O_LIMIT: 1960 case O_KEEP_STATE: 1961 if (install_state(f, 1962 (ipfw_insn_limit *)cmd, args)) { 1963 retval = IP_FW_PORT_DENY_FLAG; 1964 goto done; /* error/limit violation */ 1965 } 1966 match = 1; 1967 break; 1968 1969 case O_PROBE_STATE: 1970 case O_CHECK_STATE: 1971 /* 1972 * dynamic rules are checked at the first 1973 * keep-state or check-state occurrence, 1974 * with the result being stored in dyn_dir. 1975 * The compiler introduces a PROBE_STATE 1976 * instruction for us when we have a 1977 * KEEP_STATE (because PROBE_STATE needs 1978 * to be run first). 1979 */ 1980 if (dyn_dir == MATCH_UNKNOWN && 1981 (q = lookup_dyn_rule(&args->f_id, 1982 &dyn_dir, proto == IPPROTO_TCP ? 1983 L3HDR(struct tcphdr, ip) : NULL)) 1984 != NULL) { 1985 /* 1986 * Found dynamic entry, update stats 1987 * and jump to the 'action' part of 1988 * the parent rule. 1989 */ 1990 q->pcnt++; 1991 q->bcnt += pktlen; 1992 f = q->rule; 1993 cmd = ACTION_PTR(f); 1994 l = f->cmd_len - f->act_ofs; 1995 IPFW_DYN_UNLOCK(); 1996 goto check_body; 1997 } 1998 /* 1999 * Dynamic entry not found. If CHECK_STATE, 2000 * skip to next rule, if PROBE_STATE just 2001 * ignore and continue with next opcode. 2002 */ 2003 if (cmd->opcode == O_CHECK_STATE) 2004 goto next_rule; 2005 match = 1; 2006 break; 2007 2008 case O_ACCEPT: 2009 retval = 0; /* accept */ 2010 goto done; 2011 2012 case O_PIPE: 2013 case O_QUEUE: 2014 args->rule = f; /* report matching rule */ 2015 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG; 2016 goto done; 2017 2018 case O_DIVERT: 2019 case O_TEE: 2020 if (args->eh) /* not on layer 2 */ 2021 break; 2022 args->divert_rule = f->rulenum; 2023 retval = (cmd->opcode == O_DIVERT) ? 2024 cmd->arg1 : 2025 cmd->arg1 | IP_FW_PORT_TEE_FLAG; 2026 goto done; 2027 2028 case O_COUNT: 2029 case O_SKIPTO: 2030 f->pcnt++; /* update stats */ 2031 f->bcnt += pktlen; 2032 f->timestamp = time_second; 2033 if (cmd->opcode == O_COUNT) 2034 goto next_rule; 2035 /* handle skipto */ 2036 if (f->next_rule == NULL) 2037 lookup_next_rule(f); 2038 f = f->next_rule; 2039 goto again; 2040 2041 case O_REJECT: 2042 /* 2043 * Drop the packet and send a reject notice 2044 * if the packet is not ICMP (or is an ICMP 2045 * query), and it is not multicast/broadcast. 2046 */ 2047 if (hlen > 0 && 2048 (proto != IPPROTO_ICMP || 2049 is_icmp_query(ip)) && 2050 !(m->m_flags & (M_BCAST|M_MCAST)) && 2051 !IN_MULTICAST(dst_ip.s_addr)) { 2052 send_reject(args, cmd->arg1, 2053 offset,ip_len); 2054 m = args->m; 2055 } 2056 /* FALLTHROUGH */ 2057 case O_DENY: 2058 retval = IP_FW_PORT_DENY_FLAG; 2059 goto done; 2060 2061 case O_FORWARD_IP: 2062 if (args->eh) /* not valid on layer2 pkts */ 2063 break; 2064 if (!q || dyn_dir == MATCH_FORWARD) 2065 args->next_hop = 2066 &((ipfw_insn_sa *)cmd)->sa; 2067 retval = 0; 2068 goto done; 2069 2070 default: 2071 panic("-- unknown opcode %d\n", cmd->opcode); 2072 } /* end of switch() on opcodes */ 2073 2074 if (cmd->len & F_NOT) 2075 match = !match; 2076 2077 if (match) { 2078 if (cmd->len & F_OR) 2079 skip_or = 1; 2080 } else { 2081 if (!(cmd->len & F_OR)) /* not an OR block, */ 2082 break; /* try next rule */ 2083 } 2084 2085 } /* end of inner for, scan opcodes */ 2086 2087next_rule:; /* try next rule */ 2088 2089 } /* end of outer for, scan rules */ 2090 printf("ipfw: ouch!, skip past end of rules, denying packet\n"); 2091 IPFW_UNLOCK(chain); 2092 return(IP_FW_PORT_DENY_FLAG); 2093 2094done: 2095 /* Update statistics */ 2096 f->pcnt++; 2097 f->bcnt += pktlen; 2098 f->timestamp = time_second; 2099 IPFW_UNLOCK(chain); 2100 return retval; 2101 2102pullup_failed: 2103 if (fw_verbose) 2104 printf("ipfw: pullup failed\n"); 2105 return(IP_FW_PORT_DENY_FLAG); 2106} 2107 2108/* 2109 * When a rule is added/deleted, clear the next_rule pointers in all rules. 2110 * These will be reconstructed on the fly as packets are matched. 2111 */ 2112static void 2113flush_rule_ptrs(struct ip_fw_chain *chain) 2114{ 2115 struct ip_fw *rule; 2116 2117 IPFW_LOCK_ASSERT(chain); 2118 2119 for (rule = chain->rules; rule; rule = rule->next) 2120 rule->next_rule = NULL; 2121} 2122 2123/* 2124 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given 2125 * pipe/queue, or to all of them (match == NULL). 2126 */ 2127void 2128flush_pipe_ptrs(struct dn_flow_set *match) 2129{ 2130 struct ip_fw *rule; 2131 2132 IPFW_LOCK(&layer3_chain); 2133 for (rule = layer3_chain.rules; rule; rule = rule->next) { 2134 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule); 2135 2136 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE) 2137 continue; 2138 /* 2139 * XXX Use bcmp/bzero to handle pipe_ptr to overcome 2140 * possible alignment problems on 64-bit architectures. 2141 * This code is seldom used so we do not worry too 2142 * much about efficiency. 2143 */ 2144 if (match == NULL || 2145 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) ) 2146 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr)); 2147 } 2148 IPFW_UNLOCK(&layer3_chain); 2149} 2150 2151/* 2152 * Add a new rule to the list. Copy the rule into a malloc'ed area, then 2153 * possibly create a rule number and add the rule to the list. 2154 * Update the rule_number in the input struct so the caller knows it as well. 2155 */ 2156static int 2157add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule) 2158{ 2159 struct ip_fw *rule, *f, *prev; 2160 int l = RULESIZE(input_rule); 2161 2162 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE) 2163 return (EINVAL); 2164 2165 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO); 2166 if (rule == NULL) 2167 return (ENOSPC); 2168 2169 bcopy(input_rule, rule, l); 2170 2171 rule->next = NULL; 2172 rule->next_rule = NULL; 2173 2174 rule->pcnt = 0; 2175 rule->bcnt = 0; 2176 rule->timestamp = 0; 2177 2178 IPFW_LOCK(chain); 2179 2180 if (chain->rules == NULL) { /* default rule */ 2181 chain->rules = rule; 2182 goto done; 2183 } 2184 2185 /* 2186 * If rulenum is 0, find highest numbered rule before the 2187 * default rule, and add autoinc_step 2188 */ 2189 if (autoinc_step < 1) 2190 autoinc_step = 1; 2191 else if (autoinc_step > 1000) 2192 autoinc_step = 1000; 2193 if (rule->rulenum == 0) { 2194 /* 2195 * locate the highest numbered rule before default 2196 */ 2197 for (f = chain->rules; f; f = f->next) { 2198 if (f->rulenum == IPFW_DEFAULT_RULE) 2199 break; 2200 rule->rulenum = f->rulenum; 2201 } 2202 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step) 2203 rule->rulenum += autoinc_step; 2204 input_rule->rulenum = rule->rulenum; 2205 } 2206 2207 /* 2208 * Now insert the new rule in the right place in the sorted list. 2209 */ 2210 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) { 2211 if (f->rulenum > rule->rulenum) { /* found the location */ 2212 if (prev) { 2213 rule->next = f; 2214 prev->next = rule; 2215 } else { /* head insert */ 2216 rule->next = chain->rules; 2217 chain->rules = rule; 2218 } 2219 break; 2220 } 2221 } 2222 flush_rule_ptrs(chain); 2223done: 2224 static_count++; 2225 static_len += l; 2226 IPFW_UNLOCK(chain); 2227 DEB(printf("ipfw: installed rule %d, static count now %d\n", 2228 rule->rulenum, static_count);) 2229 return (0); 2230} 2231 2232/** 2233 * Remove a static rule (including derived * dynamic rules) 2234 * and place it on the ``reap list'' for later reclamation. 2235 * The caller is in charge of clearing rule pointers to avoid 2236 * dangling pointers. 2237 * @return a pointer to the next entry. 2238 * Arguments are not checked, so they better be correct. 2239 */ 2240static struct ip_fw * 2241remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev) 2242{ 2243 struct ip_fw *n; 2244 int l = RULESIZE(rule); 2245 2246 IPFW_LOCK_ASSERT(chain); 2247 2248 n = rule->next; 2249 IPFW_DYN_LOCK(); 2250 remove_dyn_rule(rule, NULL /* force removal */); 2251 IPFW_DYN_UNLOCK(); 2252 if (prev == NULL) 2253 chain->rules = n; 2254 else 2255 prev->next = n; 2256 static_count--; 2257 static_len -= l; 2258 2259 rule->next = chain->reap; 2260 chain->reap = rule; 2261 2262 return n; 2263} 2264 2265/** 2266 * Reclaim storage associated with a list of rules. This is 2267 * typically the list created using remove_rule. 2268 */ 2269static void 2270reap_rules(struct ip_fw *head) 2271{ 2272 struct ip_fw *rule; 2273 2274 while ((rule = head) != NULL) { 2275 head = head->next; 2276 if (DUMMYNET_LOADED) 2277 ip_dn_ruledel_ptr(rule); 2278 free(rule, M_IPFW); 2279 } 2280} 2281 2282/* 2283 * Remove all rules from a chain (except rules in set RESVD_SET 2284 * unless kill_default = 1). The caller is responsible for 2285 * reclaiming storage for the rules left in chain->reap. 2286 */ 2287static void 2288free_chain(struct ip_fw_chain *chain, int kill_default) 2289{ 2290 struct ip_fw *prev, *rule; 2291 2292 IPFW_LOCK_ASSERT(chain); 2293 2294 flush_rule_ptrs(chain); /* more efficient to do outside the loop */ 2295 for (prev = NULL, rule = chain->rules; rule ; ) 2296 if (kill_default || rule->set != RESVD_SET) 2297 rule = remove_rule(chain, rule, prev); 2298 else { 2299 prev = rule; 2300 rule = rule->next; 2301 } 2302} 2303 2304/** 2305 * Remove all rules with given number, and also do set manipulation. 2306 * Assumes chain != NULL && *chain != NULL. 2307 * 2308 * The argument is an u_int32_t. The low 16 bit are the rule or set number, 2309 * the next 8 bits are the new set, the top 8 bits are the command: 2310 * 2311 * 0 delete rules with given number 2312 * 1 delete rules with given set number 2313 * 2 move rules with given number to new set 2314 * 3 move rules with given set number to new set 2315 * 4 swap sets with given numbers 2316 */ 2317static int 2318del_entry(struct ip_fw_chain *chain, u_int32_t arg) 2319{ 2320 struct ip_fw *prev = NULL, *rule; 2321 u_int16_t rulenum; /* rule or old_set */ 2322 u_int8_t cmd, new_set; 2323 2324 rulenum = arg & 0xffff; 2325 cmd = (arg >> 24) & 0xff; 2326 new_set = (arg >> 16) & 0xff; 2327 2328 if (cmd > 4) 2329 return EINVAL; 2330 if (new_set > RESVD_SET) 2331 return EINVAL; 2332 if (cmd == 0 || cmd == 2) { 2333 if (rulenum >= IPFW_DEFAULT_RULE) 2334 return EINVAL; 2335 } else { 2336 if (rulenum > RESVD_SET) /* old_set */ 2337 return EINVAL; 2338 } 2339 2340 IPFW_LOCK(chain); 2341 rule = chain->rules; 2342 chain->reap = NULL; 2343 switch (cmd) { 2344 case 0: /* delete rules with given number */ 2345 /* 2346 * locate first rule to delete 2347 */ 2348 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next) 2349 ; 2350 if (rule->rulenum != rulenum) { 2351 IPFW_UNLOCK(chain); 2352 return EINVAL; 2353 } 2354 2355 /* 2356 * flush pointers outside the loop, then delete all matching 2357 * rules. prev remains the same throughout the cycle. 2358 */ 2359 flush_rule_ptrs(chain); 2360 while (rule->rulenum == rulenum) 2361 rule = remove_rule(chain, rule, prev); 2362 break; 2363 2364 case 1: /* delete all rules with given set number */ 2365 flush_rule_ptrs(chain); 2366 rule = chain->rules; 2367 while (rule->rulenum < IPFW_DEFAULT_RULE) 2368 if (rule->set == rulenum) 2369 rule = remove_rule(chain, rule, prev); 2370 else { 2371 prev = rule; 2372 rule = rule->next; 2373 } 2374 break; 2375 2376 case 2: /* move rules with given number to new set */ 2377 rule = chain->rules; 2378 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2379 if (rule->rulenum == rulenum) 2380 rule->set = new_set; 2381 break; 2382 2383 case 3: /* move rules with given set number to new set */ 2384 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2385 if (rule->set == rulenum) 2386 rule->set = new_set; 2387 break; 2388 2389 case 4: /* swap two sets */ 2390 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2391 if (rule->set == rulenum) 2392 rule->set = new_set; 2393 else if (rule->set == new_set) 2394 rule->set = rulenum; 2395 break; 2396 } 2397 /* 2398 * Look for rules to reclaim. We grab the list before 2399 * releasing the lock then reclaim them w/o the lock to 2400 * avoid a LOR with dummynet. 2401 */ 2402 rule = chain->reap; 2403 chain->reap = NULL; 2404 IPFW_UNLOCK(chain); 2405 if (rule) 2406 reap_rules(rule); 2407 return 0; 2408} 2409 2410/* 2411 * Clear counters for a specific rule. 2412 * The enclosing "table" is assumed locked. 2413 */ 2414static void 2415clear_counters(struct ip_fw *rule, int log_only) 2416{ 2417 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); 2418 2419 if (log_only == 0) { 2420 rule->bcnt = rule->pcnt = 0; 2421 rule->timestamp = 0; 2422 } 2423 if (l->o.opcode == O_LOG) 2424 l->log_left = l->max_log; 2425} 2426 2427/** 2428 * Reset some or all counters on firewall rules. 2429 * @arg frwl is null to clear all entries, or contains a specific 2430 * rule number. 2431 * @arg log_only is 1 if we only want to reset logs, zero otherwise. 2432 */ 2433static int 2434zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only) 2435{ 2436 struct ip_fw *rule; 2437 char *msg; 2438 2439 IPFW_LOCK(chain); 2440 if (rulenum == 0) { 2441 norule_counter = 0; 2442 for (rule = chain->rules; rule; rule = rule->next) 2443 clear_counters(rule, log_only); 2444 msg = log_only ? "ipfw: All logging counts reset.\n" : 2445 "ipfw: Accounting cleared.\n"; 2446 } else { 2447 int cleared = 0; 2448 /* 2449 * We can have multiple rules with the same number, so we 2450 * need to clear them all. 2451 */ 2452 for (rule = chain->rules; rule; rule = rule->next) 2453 if (rule->rulenum == rulenum) { 2454 while (rule && rule->rulenum == rulenum) { 2455 clear_counters(rule, log_only); 2456 rule = rule->next; 2457 } 2458 cleared = 1; 2459 break; 2460 } 2461 if (!cleared) { /* we did not find any matching rules */ 2462 IPFW_UNLOCK(chain); 2463 return (EINVAL); 2464 } 2465 msg = log_only ? "ipfw: Entry %d logging count reset.\n" : 2466 "ipfw: Entry %d cleared.\n"; 2467 } 2468 IPFW_UNLOCK(chain); 2469 2470 if (fw_verbose) 2471 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum); 2472 return (0); 2473} 2474 2475/* 2476 * Check validity of the structure before insert. 2477 * Fortunately rules are simple, so this mostly need to check rule sizes. 2478 */ 2479static int 2480check_ipfw_struct(struct ip_fw *rule, int size) 2481{ 2482 int l, cmdlen = 0; 2483 int have_action=0; 2484 ipfw_insn *cmd; 2485 2486 if (size < sizeof(*rule)) { 2487 printf("ipfw: rule too short\n"); 2488 return (EINVAL); 2489 } 2490 /* first, check for valid size */ 2491 l = RULESIZE(rule); 2492 if (l != size) { 2493 printf("ipfw: size mismatch (have %d want %d)\n", size, l); 2494 return (EINVAL); 2495 } 2496 /* 2497 * Now go for the individual checks. Very simple ones, basically only 2498 * instruction sizes. 2499 */ 2500 for (l = rule->cmd_len, cmd = rule->cmd ; 2501 l > 0 ; l -= cmdlen, cmd += cmdlen) { 2502 cmdlen = F_LEN(cmd); 2503 if (cmdlen > l) { 2504 printf("ipfw: opcode %d size truncated\n", 2505 cmd->opcode); 2506 return EINVAL; 2507 } 2508 DEB(printf("ipfw: opcode %d\n", cmd->opcode);) 2509 switch (cmd->opcode) { 2510 case O_PROBE_STATE: 2511 case O_KEEP_STATE: 2512 case O_PROTO: 2513 case O_IP_SRC_ME: 2514 case O_IP_DST_ME: 2515 case O_LAYER2: 2516 case O_IN: 2517 case O_FRAG: 2518 case O_IPOPT: 2519 case O_IPTOS: 2520 case O_IPPRECEDENCE: 2521 case O_IPVER: 2522 case O_TCPWIN: 2523 case O_TCPFLAGS: 2524 case O_TCPOPTS: 2525 case O_ESTAB: 2526 case O_VERREVPATH: 2527 case O_IPSEC: 2528 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2529 goto bad_size; 2530 break; 2531 2532 case O_UID: 2533 case O_GID: 2534 case O_IP_SRC: 2535 case O_IP_DST: 2536 case O_TCPSEQ: 2537 case O_TCPACK: 2538 case O_PROB: 2539 case O_ICMPTYPE: 2540 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) 2541 goto bad_size; 2542 break; 2543 2544 case O_LIMIT: 2545 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) 2546 goto bad_size; 2547 break; 2548 2549 case O_LOG: 2550 if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) 2551 goto bad_size; 2552 2553 ((ipfw_insn_log *)cmd)->log_left = 2554 ((ipfw_insn_log *)cmd)->max_log; 2555 2556 break; 2557 2558 case O_IP_SRC_MASK: 2559 case O_IP_DST_MASK: 2560 /* only odd command lengths */ 2561 if ( !(cmdlen & 1) || cmdlen > 31) 2562 goto bad_size; 2563 break; 2564 2565 case O_IP_SRC_SET: 2566 case O_IP_DST_SET: 2567 if (cmd->arg1 == 0 || cmd->arg1 > 256) { 2568 printf("ipfw: invalid set size %d\n", 2569 cmd->arg1); 2570 return EINVAL; 2571 } 2572 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 2573 (cmd->arg1+31)/32 ) 2574 goto bad_size; 2575 break; 2576 2577 case O_MACADDR2: 2578 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) 2579 goto bad_size; 2580 break; 2581 2582 case O_NOP: 2583 case O_IPID: 2584 case O_IPTTL: 2585 case O_IPLEN: 2586 if (cmdlen < 1 || cmdlen > 31) 2587 goto bad_size; 2588 break; 2589 2590 case O_MAC_TYPE: 2591 case O_IP_SRCPORT: 2592 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */ 2593 if (cmdlen < 2 || cmdlen > 31) 2594 goto bad_size; 2595 break; 2596 2597 case O_RECV: 2598 case O_XMIT: 2599 case O_VIA: 2600 if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) 2601 goto bad_size; 2602 break; 2603 2604 case O_PIPE: 2605 case O_QUEUE: 2606 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe)) 2607 goto bad_size; 2608 goto check_action; 2609 2610 case O_FORWARD_IP: 2611 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) 2612 goto bad_size; 2613 goto check_action; 2614 2615 case O_FORWARD_MAC: /* XXX not implemented yet */ 2616 case O_CHECK_STATE: 2617 case O_COUNT: 2618 case O_ACCEPT: 2619 case O_DENY: 2620 case O_REJECT: 2621 case O_SKIPTO: 2622 case O_DIVERT: 2623 case O_TEE: 2624 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2625 goto bad_size; 2626check_action: 2627 if (have_action) { 2628 printf("ipfw: opcode %d, multiple actions" 2629 " not allowed\n", 2630 cmd->opcode); 2631 return EINVAL; 2632 } 2633 have_action = 1; 2634 if (l != cmdlen) { 2635 printf("ipfw: opcode %d, action must be" 2636 " last opcode\n", 2637 cmd->opcode); 2638 return EINVAL; 2639 } 2640 break; 2641 default: 2642 printf("ipfw: opcode %d, unknown opcode\n", 2643 cmd->opcode); 2644 return EINVAL; 2645 } 2646 } 2647 if (have_action == 0) { 2648 printf("ipfw: missing action\n"); 2649 return EINVAL; 2650 } 2651 return 0; 2652 2653bad_size: 2654 printf("ipfw: opcode %d size %d wrong\n", 2655 cmd->opcode, cmdlen); 2656 return EINVAL; 2657} 2658 2659/* 2660 * Copy the static and dynamic rules to the supplied buffer 2661 * and return the amount of space actually used. 2662 */ 2663static size_t 2664ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space) 2665{ 2666 char *bp = buf; 2667 char *ep = bp + space; 2668 struct ip_fw *rule; 2669 int i; 2670 2671 /* XXX this can take a long time and locking will block packet flow */ 2672 IPFW_LOCK(chain); 2673 for (rule = chain->rules; rule ; rule = rule->next) { 2674 /* 2675 * Verify the entry fits in the buffer in case the 2676 * rules changed between calculating buffer space and 2677 * now. This would be better done using a generation 2678 * number but should suffice for now. 2679 */ 2680 i = RULESIZE(rule); 2681 if (bp + i <= ep) { 2682 bcopy(rule, bp, i); 2683 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule), 2684 sizeof(set_disable)); 2685 bp += i; 2686 } 2687 } 2688 IPFW_UNLOCK(chain); 2689 if (ipfw_dyn_v) { 2690 ipfw_dyn_rule *p, *last = NULL; 2691 2692 IPFW_DYN_LOCK(); 2693 for (i = 0 ; i < curr_dyn_buckets; i++) 2694 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) { 2695 if (bp + sizeof *p <= ep) { 2696 ipfw_dyn_rule *dst = 2697 (ipfw_dyn_rule *)bp; 2698 bcopy(p, dst, sizeof *p); 2699 bcopy(&(p->rule->rulenum), &(dst->rule), 2700 sizeof(p->rule->rulenum)); 2701 /* 2702 * store a non-null value in "next". 2703 * The userland code will interpret a 2704 * NULL here as a marker 2705 * for the last dynamic rule. 2706 */ 2707 bcopy(&dst, &dst->next, sizeof(dst)); 2708 last = dst; 2709 dst->expire = 2710 TIME_LEQ(dst->expire, time_second) ? 2711 0 : dst->expire - time_second ; 2712 bp += sizeof(ipfw_dyn_rule); 2713 } 2714 } 2715 IPFW_DYN_UNLOCK(); 2716 if (last != NULL) /* mark last dynamic rule */ 2717 bzero(&last->next, sizeof(last)); 2718 } 2719 return (bp - (char *)buf); 2720} 2721 2722 2723/** 2724 * {set|get}sockopt parser. 2725 */ 2726static int 2727ipfw_ctl(struct sockopt *sopt) 2728{ 2729#define RULE_MAXSIZE (256*sizeof(u_int32_t)) 2730 int error, rule_num; 2731 size_t size; 2732 struct ip_fw *buf, *rule; 2733 u_int32_t rulenum[2]; 2734 2735 /* 2736 * Disallow modifications in really-really secure mode, but still allow 2737 * the logging counters to be reset. 2738 */ 2739 if (sopt->sopt_name == IP_FW_ADD || 2740 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) { 2741#if __FreeBSD_version >= 500034 2742 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 2743 if (error) 2744 return (error); 2745#else /* FreeBSD 4.x */ 2746 if (securelevel >= 3) 2747 return (EPERM); 2748#endif 2749 } 2750 2751 error = 0; 2752 2753 switch (sopt->sopt_name) { 2754 case IP_FW_GET: 2755 /* 2756 * pass up a copy of the current rules. Static rules 2757 * come first (the last of which has number IPFW_DEFAULT_RULE), 2758 * followed by a possibly empty list of dynamic rule. 2759 * The last dynamic rule has NULL in the "next" field. 2760 * 2761 * Note that the calculated size is used to bound the 2762 * amount of data returned to the user. The rule set may 2763 * change between calculating the size and returning the 2764 * data in which case we'll just return what fits. 2765 */ 2766 size = static_len; /* size of static rules */ 2767 if (ipfw_dyn_v) /* add size of dyn.rules */ 2768 size += (dyn_count * sizeof(ipfw_dyn_rule)); 2769 2770 /* 2771 * XXX todo: if the user passes a short length just to know 2772 * how much room is needed, do not bother filling up the 2773 * buffer, just jump to the sooptcopyout. 2774 */ 2775 buf = malloc(size, M_TEMP, M_WAITOK); 2776 error = sooptcopyout(sopt, buf, 2777 ipfw_getrules(&layer3_chain, buf, size)); 2778 free(buf, M_TEMP); 2779 break; 2780 2781 case IP_FW_FLUSH: 2782 /* 2783 * Normally we cannot release the lock on each iteration. 2784 * We could do it here only because we start from the head all 2785 * the times so there is no risk of missing some entries. 2786 * On the other hand, the risk is that we end up with 2787 * a very inconsistent ruleset, so better keep the lock 2788 * around the whole cycle. 2789 * 2790 * XXX this code can be improved by resetting the head of 2791 * the list to point to the default rule, and then freeing 2792 * the old list without the need for a lock. 2793 */ 2794 2795 IPFW_LOCK(&layer3_chain); 2796 layer3_chain.reap = NULL; 2797 free_chain(&layer3_chain, 0 /* keep default rule */); 2798 rule = layer3_chain.reap, layer3_chain.reap = NULL; 2799 IPFW_UNLOCK(&layer3_chain); 2800 if (layer3_chain.reap != NULL) 2801 reap_rules(rule); 2802 break; 2803 2804 case IP_FW_ADD: 2805 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK); 2806 error = sooptcopyin(sopt, rule, RULE_MAXSIZE, 2807 sizeof(struct ip_fw) ); 2808 if (error == 0) 2809 error = check_ipfw_struct(rule, sopt->sopt_valsize); 2810 if (error == 0) { 2811 error = add_rule(&layer3_chain, rule); 2812 size = RULESIZE(rule); 2813 if (!error && sopt->sopt_dir == SOPT_GET) 2814 error = sooptcopyout(sopt, rule, size); 2815 } 2816 free(rule, M_TEMP); 2817 break; 2818 2819 case IP_FW_DEL: 2820 /* 2821 * IP_FW_DEL is used for deleting single rules or sets, 2822 * and (ab)used to atomically manipulate sets. Argument size 2823 * is used to distinguish between the two: 2824 * sizeof(u_int32_t) 2825 * delete single rule or set of rules, 2826 * or reassign rules (or sets) to a different set. 2827 * 2*sizeof(u_int32_t) 2828 * atomic disable/enable sets. 2829 * first u_int32_t contains sets to be disabled, 2830 * second u_int32_t contains sets to be enabled. 2831 */ 2832 error = sooptcopyin(sopt, rulenum, 2833 2*sizeof(u_int32_t), sizeof(u_int32_t)); 2834 if (error) 2835 break; 2836 size = sopt->sopt_valsize; 2837 if (size == sizeof(u_int32_t)) /* delete or reassign */ 2838 error = del_entry(&layer3_chain, rulenum[0]); 2839 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */ 2840 set_disable = 2841 (set_disable | rulenum[0]) & ~rulenum[1] & 2842 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */ 2843 else 2844 error = EINVAL; 2845 break; 2846 2847 case IP_FW_ZERO: 2848 case IP_FW_RESETLOG: /* argument is an int, the rule number */ 2849 rule_num = 0; 2850 if (sopt->sopt_val != 0) { 2851 error = sooptcopyin(sopt, &rule_num, 2852 sizeof(int), sizeof(int)); 2853 if (error) 2854 break; 2855 } 2856 error = zero_entry(&layer3_chain, rule_num, 2857 sopt->sopt_name == IP_FW_RESETLOG); 2858 break; 2859 2860 default: 2861 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name); 2862 error = EINVAL; 2863 } 2864 2865 return (error); 2866#undef RULE_MAXSIZE 2867} 2868 2869/** 2870 * dummynet needs a reference to the default rule, because rules can be 2871 * deleted while packets hold a reference to them. When this happens, 2872 * dummynet changes the reference to the default rule (it could well be a 2873 * NULL pointer, but this way we do not need to check for the special 2874 * case, plus here he have info on the default behaviour). 2875 */ 2876struct ip_fw *ip_fw_default_rule; 2877 2878/* 2879 * This procedure is only used to handle keepalives. It is invoked 2880 * every dyn_keepalive_period 2881 */ 2882static void 2883ipfw_tick(void * __unused unused) 2884{ 2885 int i; 2886 ipfw_dyn_rule *q; 2887 2888 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0) 2889 goto done; 2890 2891 IPFW_DYN_LOCK(); 2892 for (i = 0 ; i < curr_dyn_buckets ; i++) { 2893 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) { 2894 if (q->dyn_type == O_LIMIT_PARENT) 2895 continue; 2896 if (q->id.proto != IPPROTO_TCP) 2897 continue; 2898 if ( (q->state & BOTH_SYN) != BOTH_SYN) 2899 continue; 2900 if (TIME_LEQ( time_second+dyn_keepalive_interval, 2901 q->expire)) 2902 continue; /* too early */ 2903 if (TIME_LEQ(q->expire, time_second)) 2904 continue; /* too late, rule expired */ 2905 2906 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 2907 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0); 2908 } 2909 } 2910 IPFW_DYN_UNLOCK(); 2911done: 2912 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL); 2913} 2914 2915static int 2916ipfw_init(void) 2917{ 2918 struct ip_fw default_rule; 2919 int error; 2920 2921 layer3_chain.rules = NULL; 2922 IPFW_LOCK_INIT(&layer3_chain); 2923 IPFW_DYN_LOCK_INIT(); 2924 callout_init(&ipfw_timeout, CALLOUT_MPSAFE); 2925 2926 bzero(&default_rule, sizeof default_rule); 2927 2928 default_rule.act_ofs = 0; 2929 default_rule.rulenum = IPFW_DEFAULT_RULE; 2930 default_rule.cmd_len = 1; 2931 default_rule.set = RESVD_SET; 2932 2933 default_rule.cmd[0].len = 1; 2934 default_rule.cmd[0].opcode = 2935#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT 2936 1 ? O_ACCEPT : 2937#endif 2938 O_DENY; 2939 2940 error = add_rule(&layer3_chain, &default_rule); 2941 if (error != 0) { 2942 printf("ipfw2: error %u initializing default rule " 2943 "(support disabled)\n", error); 2944 IPFW_DYN_LOCK_DESTROY(); 2945 IPFW_LOCK_DESTROY(&layer3_chain); 2946 return (error); 2947 } 2948 2949 ip_fw_default_rule = layer3_chain.rules; 2950 printf("ipfw2 initialized, divert %s, " 2951 "rule-based forwarding enabled, default to %s, logging ", 2952#ifdef IPDIVERT 2953 "enabled", 2954#else 2955 "disabled", 2956#endif 2957 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny"); 2958 2959#ifdef IPFIREWALL_VERBOSE 2960 fw_verbose = 1; 2961#endif 2962#ifdef IPFIREWALL_VERBOSE_LIMIT 2963 verbose_limit = IPFIREWALL_VERBOSE_LIMIT; 2964#endif 2965 if (fw_verbose == 0) 2966 printf("disabled\n"); 2967 else if (verbose_limit == 0) 2968 printf("unlimited\n"); 2969 else 2970 printf("limited to %d packets/entry by default\n", 2971 verbose_limit); 2972 2973 ip_fw_chk_ptr = ipfw_chk; 2974 ip_fw_ctl_ptr = ipfw_ctl; 2975 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL); 2976 2977 return (0); 2978} 2979 2980static void 2981ipfw_destroy(void) 2982{ 2983 struct ip_fw *reap; 2984 2985 IPFW_LOCK(&layer3_chain); 2986 callout_stop(&ipfw_timeout); 2987 ip_fw_chk_ptr = NULL; 2988 ip_fw_ctl_ptr = NULL; 2989 layer3_chain.reap = NULL; 2990 free_chain(&layer3_chain, 1 /* kill default rule */); 2991 reap = layer3_chain.reap, layer3_chain.reap = NULL; 2992 IPFW_UNLOCK(&layer3_chain); 2993 if (reap != NULL) 2994 reap_rules(reap); 2995 2996 IPFW_DYN_LOCK_DESTROY(); 2997 IPFW_LOCK_DESTROY(&layer3_chain); 2998 printf("IP firewall unloaded\n"); 2999} 3000 3001static int 3002ipfw_modevent(module_t mod, int type, void *unused) 3003{ 3004 int err = 0; 3005 3006 switch (type) { 3007 case MOD_LOAD: 3008 if (IPFW_LOADED) { 3009 printf("IP firewall already loaded\n"); 3010 err = EEXIST; 3011 } else { 3012 err = ipfw_init(); 3013 } 3014 break; 3015 3016 case MOD_UNLOAD: 3017 ipfw_destroy(); 3018 err = 0; 3019 break; 3020 default: 3021 break; 3022 } 3023 return err; 3024} 3025 3026static moduledata_t ipfwmod = { 3027 "ipfw", 3028 ipfw_modevent, 3029 0 3030}; 3031DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3032MODULE_VERSION(ipfw, 1); 3033#endif /* IPFW2 */ 3034