ip_fw2.c revision 122966
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 122966 2003-11-23 18:13:41Z sam $ 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) 470 return 0; 471 if ((ifp == NULL) || (ro.ro_rt->rt_ifp->if_index != ifp->if_index)) { 472 RTFREE(ro.ro_rt); 473 return 0; 474 } 475 RTFREE(ro.ro_rt); 476 return 1; 477} 478 479 480static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */ 481 482#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0 483#define SNP(buf) buf, sizeof(buf) 484 485/* 486 * We enter here when we have a rule with O_LOG. 487 * XXX this function alone takes about 2Kbytes of code! 488 */ 489static void 490ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh, 491 struct mbuf *m, struct ifnet *oif) 492{ 493 char *action; 494 int limit_reached = 0; 495 char action2[40], proto[48], fragment[28]; 496 497 fragment[0] = '\0'; 498 proto[0] = '\0'; 499 500 if (f == NULL) { /* bogus pkt */ 501 if (verbose_limit != 0 && norule_counter >= verbose_limit) 502 return; 503 norule_counter++; 504 if (norule_counter == verbose_limit) 505 limit_reached = verbose_limit; 506 action = "Refuse"; 507 } else { /* O_LOG is the first action, find the real one */ 508 ipfw_insn *cmd = ACTION_PTR(f); 509 ipfw_insn_log *l = (ipfw_insn_log *)cmd; 510 511 if (l->max_log != 0 && l->log_left == 0) 512 return; 513 l->log_left--; 514 if (l->log_left == 0) 515 limit_reached = l->max_log; 516 cmd += F_LEN(cmd); /* point to first action */ 517 if (cmd->opcode == O_PROB) 518 cmd += F_LEN(cmd); 519 520 action = action2; 521 switch (cmd->opcode) { 522 case O_DENY: 523 action = "Deny"; 524 break; 525 526 case O_REJECT: 527 if (cmd->arg1==ICMP_REJECT_RST) 528 action = "Reset"; 529 else if (cmd->arg1==ICMP_UNREACH_HOST) 530 action = "Reject"; 531 else 532 snprintf(SNPARGS(action2, 0), "Unreach %d", 533 cmd->arg1); 534 break; 535 536 case O_ACCEPT: 537 action = "Accept"; 538 break; 539 case O_COUNT: 540 action = "Count"; 541 break; 542 case O_DIVERT: 543 snprintf(SNPARGS(action2, 0), "Divert %d", 544 cmd->arg1); 545 break; 546 case O_TEE: 547 snprintf(SNPARGS(action2, 0), "Tee %d", 548 cmd->arg1); 549 break; 550 case O_SKIPTO: 551 snprintf(SNPARGS(action2, 0), "SkipTo %d", 552 cmd->arg1); 553 break; 554 case O_PIPE: 555 snprintf(SNPARGS(action2, 0), "Pipe %d", 556 cmd->arg1); 557 break; 558 case O_QUEUE: 559 snprintf(SNPARGS(action2, 0), "Queue %d", 560 cmd->arg1); 561 break; 562 case O_FORWARD_IP: { 563 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd; 564 int len; 565 566 len = snprintf(SNPARGS(action2, 0), "Forward to %s", 567 inet_ntoa(sa->sa.sin_addr)); 568 if (sa->sa.sin_port) 569 snprintf(SNPARGS(action2, len), ":%d", 570 sa->sa.sin_port); 571 } 572 break; 573 default: 574 action = "UNKNOWN"; 575 break; 576 } 577 } 578 579 if (hlen == 0) { /* non-ip */ 580 snprintf(SNPARGS(proto, 0), "MAC"); 581 } else { 582 struct ip *ip = mtod(m, struct ip *); 583 /* these three are all aliases to the same thing */ 584 struct icmp *const icmp = L3HDR(struct icmp, ip); 585 struct tcphdr *const tcp = (struct tcphdr *)icmp; 586 struct udphdr *const udp = (struct udphdr *)icmp; 587 588 int ip_off, offset, ip_len; 589 590 int len; 591 592 if (eh != NULL) { /* layer 2 packets are as on the wire */ 593 ip_off = ntohs(ip->ip_off); 594 ip_len = ntohs(ip->ip_len); 595 } else { 596 ip_off = ip->ip_off; 597 ip_len = ip->ip_len; 598 } 599 offset = ip_off & IP_OFFMASK; 600 switch (ip->ip_p) { 601 case IPPROTO_TCP: 602 len = snprintf(SNPARGS(proto, 0), "TCP %s", 603 inet_ntoa(ip->ip_src)); 604 if (offset == 0) 605 snprintf(SNPARGS(proto, len), ":%d %s:%d", 606 ntohs(tcp->th_sport), 607 inet_ntoa(ip->ip_dst), 608 ntohs(tcp->th_dport)); 609 else 610 snprintf(SNPARGS(proto, len), " %s", 611 inet_ntoa(ip->ip_dst)); 612 break; 613 614 case IPPROTO_UDP: 615 len = snprintf(SNPARGS(proto, 0), "UDP %s", 616 inet_ntoa(ip->ip_src)); 617 if (offset == 0) 618 snprintf(SNPARGS(proto, len), ":%d %s:%d", 619 ntohs(udp->uh_sport), 620 inet_ntoa(ip->ip_dst), 621 ntohs(udp->uh_dport)); 622 else 623 snprintf(SNPARGS(proto, len), " %s", 624 inet_ntoa(ip->ip_dst)); 625 break; 626 627 case IPPROTO_ICMP: 628 if (offset == 0) 629 len = snprintf(SNPARGS(proto, 0), 630 "ICMP:%u.%u ", 631 icmp->icmp_type, icmp->icmp_code); 632 else 633 len = snprintf(SNPARGS(proto, 0), "ICMP "); 634 len += snprintf(SNPARGS(proto, len), "%s", 635 inet_ntoa(ip->ip_src)); 636 snprintf(SNPARGS(proto, len), " %s", 637 inet_ntoa(ip->ip_dst)); 638 break; 639 640 default: 641 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p, 642 inet_ntoa(ip->ip_src)); 643 snprintf(SNPARGS(proto, len), " %s", 644 inet_ntoa(ip->ip_dst)); 645 break; 646 } 647 648 if (ip_off & (IP_MF | IP_OFFMASK)) 649 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)", 650 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2), 651 offset << 3, 652 (ip_off & IP_MF) ? "+" : ""); 653 } 654 if (oif || m->m_pkthdr.rcvif) 655 log(LOG_SECURITY | LOG_INFO, 656 "ipfw: %d %s %s %s via %s%s\n", 657 f ? f->rulenum : -1, 658 action, proto, oif ? "out" : "in", 659 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname, 660 fragment); 661 else 662 log(LOG_SECURITY | LOG_INFO, 663 "ipfw: %d %s %s [no if info]%s\n", 664 f ? f->rulenum : -1, 665 action, proto, fragment); 666 if (limit_reached) 667 log(LOG_SECURITY | LOG_NOTICE, 668 "ipfw: limit %d reached on entry %d\n", 669 limit_reached, f ? f->rulenum : -1); 670} 671 672/* 673 * IMPORTANT: the hash function for dynamic rules must be commutative 674 * in source and destination (ip,port), because rules are bidirectional 675 * and we want to find both in the same bucket. 676 */ 677static __inline int 678hash_packet(struct ipfw_flow_id *id) 679{ 680 u_int32_t i; 681 682 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 683 i &= (curr_dyn_buckets - 1); 684 return i; 685} 686 687/** 688 * unlink a dynamic rule from a chain. prev is a pointer to 689 * the previous one, q is a pointer to the rule to delete, 690 * head is a pointer to the head of the queue. 691 * Modifies q and potentially also head. 692 */ 693#define UNLINK_DYN_RULE(prev, head, q) { \ 694 ipfw_dyn_rule *old_q = q; \ 695 \ 696 /* remove a refcount to the parent */ \ 697 if (q->dyn_type == O_LIMIT) \ 698 q->parent->count--; \ 699 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\ 700 (q->id.src_ip), (q->id.src_port), \ 701 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \ 702 if (prev != NULL) \ 703 prev->next = q = q->next; \ 704 else \ 705 head = q = q->next; \ 706 dyn_count--; \ 707 free(old_q, M_IPFW); } 708 709#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 710 711/** 712 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 713 * 714 * If keep_me == NULL, rules are deleted even if not expired, 715 * otherwise only expired rules are removed. 716 * 717 * The value of the second parameter is also used to point to identify 718 * a rule we absolutely do not want to remove (e.g. because we are 719 * holding a reference to it -- this is the case with O_LIMIT_PARENT 720 * rules). The pointer is only used for comparison, so any non-null 721 * value will do. 722 */ 723static void 724remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 725{ 726 static u_int32_t last_remove = 0; 727 728#define FORCE (keep_me == NULL) 729 730 ipfw_dyn_rule *prev, *q; 731 int i, pass = 0, max_pass = 0; 732 733 IPFW_DYN_LOCK_ASSERT(); 734 735 if (ipfw_dyn_v == NULL || dyn_count == 0) 736 return; 737 /* do not expire more than once per second, it is useless */ 738 if (!FORCE && last_remove == time_second) 739 return; 740 last_remove = time_second; 741 742 /* 743 * because O_LIMIT refer to parent rules, during the first pass only 744 * remove child and mark any pending LIMIT_PARENT, and remove 745 * them in a second pass. 746 */ 747next_pass: 748 for (i = 0 ; i < curr_dyn_buckets ; i++) { 749 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) { 750 /* 751 * Logic can become complex here, so we split tests. 752 */ 753 if (q == keep_me) 754 goto next; 755 if (rule != NULL && rule != q->rule) 756 goto next; /* not the one we are looking for */ 757 if (q->dyn_type == O_LIMIT_PARENT) { 758 /* 759 * handle parent in the second pass, 760 * record we need one. 761 */ 762 max_pass = 1; 763 if (pass == 0) 764 goto next; 765 if (FORCE && q->count != 0 ) { 766 /* XXX should not happen! */ 767 printf("ipfw: OUCH! cannot remove rule," 768 " count %d\n", q->count); 769 } 770 } else { 771 if (!FORCE && 772 !TIME_LEQ( q->expire, time_second )) 773 goto next; 774 } 775 if (q->dyn_type != O_LIMIT_PARENT || !q->count) { 776 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 777 continue; 778 } 779next: 780 prev=q; 781 q=q->next; 782 } 783 } 784 if (pass++ < max_pass) 785 goto next_pass; 786} 787 788 789/** 790 * lookup a dynamic rule. 791 */ 792static ipfw_dyn_rule * 793lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction, 794 struct tcphdr *tcp) 795{ 796 /* 797 * stateful ipfw extensions. 798 * Lookup into dynamic session queue 799 */ 800#define MATCH_REVERSE 0 801#define MATCH_FORWARD 1 802#define MATCH_NONE 2 803#define MATCH_UNKNOWN 3 804 int i, dir = MATCH_NONE; 805 ipfw_dyn_rule *prev, *q=NULL; 806 807 IPFW_DYN_LOCK_ASSERT(); 808 809 if (ipfw_dyn_v == NULL) 810 goto done; /* not found */ 811 i = hash_packet( pkt ); 812 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) { 813 if (q->dyn_type == O_LIMIT_PARENT && q->count) 814 goto next; 815 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */ 816 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 817 continue; 818 } 819 if (pkt->proto == q->id.proto && 820 q->dyn_type != O_LIMIT_PARENT) { 821 if (pkt->src_ip == q->id.src_ip && 822 pkt->dst_ip == q->id.dst_ip && 823 pkt->src_port == q->id.src_port && 824 pkt->dst_port == q->id.dst_port ) { 825 dir = MATCH_FORWARD; 826 break; 827 } 828 if (pkt->src_ip == q->id.dst_ip && 829 pkt->dst_ip == q->id.src_ip && 830 pkt->src_port == q->id.dst_port && 831 pkt->dst_port == q->id.src_port ) { 832 dir = MATCH_REVERSE; 833 break; 834 } 835 } 836next: 837 prev = q; 838 q = q->next; 839 } 840 if (q == NULL) 841 goto done; /* q = NULL, not found */ 842 843 if ( prev != NULL) { /* found and not in front */ 844 prev->next = q->next; 845 q->next = ipfw_dyn_v[i]; 846 ipfw_dyn_v[i] = q; 847 } 848 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 849 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST); 850 851#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 852#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 853 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8); 854 switch (q->state) { 855 case TH_SYN: /* opening */ 856 q->expire = time_second + dyn_syn_lifetime; 857 break; 858 859 case BOTH_SYN: /* move to established */ 860 case BOTH_SYN | TH_FIN : /* one side tries to close */ 861 case BOTH_SYN | (TH_FIN << 8) : 862 if (tcp) { 863#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 864 u_int32_t ack = ntohl(tcp->th_ack); 865 if (dir == MATCH_FORWARD) { 866 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd)) 867 q->ack_fwd = ack; 868 else { /* ignore out-of-sequence */ 869 break; 870 } 871 } else { 872 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev)) 873 q->ack_rev = ack; 874 else { /* ignore out-of-sequence */ 875 break; 876 } 877 } 878 } 879 q->expire = time_second + dyn_ack_lifetime; 880 break; 881 882 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 883 if (dyn_fin_lifetime >= dyn_keepalive_period) 884 dyn_fin_lifetime = dyn_keepalive_period - 1; 885 q->expire = time_second + dyn_fin_lifetime; 886 break; 887 888 default: 889#if 0 890 /* 891 * reset or some invalid combination, but can also 892 * occur if we use keep-state the wrong way. 893 */ 894 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 895 printf("invalid state: 0x%x\n", q->state); 896#endif 897 if (dyn_rst_lifetime >= dyn_keepalive_period) 898 dyn_rst_lifetime = dyn_keepalive_period - 1; 899 q->expire = time_second + dyn_rst_lifetime; 900 break; 901 } 902 } else if (pkt->proto == IPPROTO_UDP) { 903 q->expire = time_second + dyn_udp_lifetime; 904 } else { 905 /* other protocols */ 906 q->expire = time_second + dyn_short_lifetime; 907 } 908done: 909 if (match_direction) 910 *match_direction = dir; 911 return q; 912} 913 914static ipfw_dyn_rule * 915lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 916 struct tcphdr *tcp) 917{ 918 ipfw_dyn_rule *q; 919 920 IPFW_DYN_LOCK(); 921 q = lookup_dyn_rule_locked(pkt, match_direction, tcp); 922 if (q == NULL) 923 IPFW_DYN_UNLOCK(); 924 /* NB: return table locked when q is not NULL */ 925 return q; 926} 927 928static void 929realloc_dynamic_table(void) 930{ 931 IPFW_DYN_LOCK_ASSERT(); 932 933 /* 934 * Try reallocation, make sure we have a power of 2 and do 935 * not allow more than 64k entries. In case of overflow, 936 * default to 1024. 937 */ 938 939 if (dyn_buckets > 65536) 940 dyn_buckets = 1024; 941 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */ 942 dyn_buckets = curr_dyn_buckets; /* reset */ 943 return; 944 } 945 curr_dyn_buckets = dyn_buckets; 946 if (ipfw_dyn_v != NULL) 947 free(ipfw_dyn_v, M_IPFW); 948 for (;;) { 949 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 950 M_IPFW, M_NOWAIT | M_ZERO); 951 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2) 952 break; 953 curr_dyn_buckets /= 2; 954 } 955} 956 957/** 958 * Install state of type 'type' for a dynamic session. 959 * The hash table contains two type of rules: 960 * - regular rules (O_KEEP_STATE) 961 * - rules for sessions with limited number of sess per user 962 * (O_LIMIT). When they are created, the parent is 963 * increased by 1, and decreased on delete. In this case, 964 * the third parameter is the parent rule and not the chain. 965 * - "parent" rules for the above (O_LIMIT_PARENT). 966 */ 967static ipfw_dyn_rule * 968add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule) 969{ 970 ipfw_dyn_rule *r; 971 int i; 972 973 IPFW_DYN_LOCK_ASSERT(); 974 975 if (ipfw_dyn_v == NULL || 976 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) { 977 realloc_dynamic_table(); 978 if (ipfw_dyn_v == NULL) 979 return NULL; /* failed ! */ 980 } 981 i = hash_packet(id); 982 983 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO); 984 if (r == NULL) { 985 printf ("ipfw: sorry cannot allocate state\n"); 986 return NULL; 987 } 988 989 /* increase refcount on parent, and set pointer */ 990 if (dyn_type == O_LIMIT) { 991 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 992 if ( parent->dyn_type != O_LIMIT_PARENT) 993 panic("invalid parent"); 994 parent->count++; 995 r->parent = parent; 996 rule = parent->rule; 997 } 998 999 r->id = *id; 1000 r->expire = time_second + dyn_syn_lifetime; 1001 r->rule = rule; 1002 r->dyn_type = dyn_type; 1003 r->pcnt = r->bcnt = 0; 1004 r->count = 0; 1005 1006 r->bucket = i; 1007 r->next = ipfw_dyn_v[i]; 1008 ipfw_dyn_v[i] = r; 1009 dyn_count++; 1010 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n", 1011 dyn_type, 1012 (r->id.src_ip), (r->id.src_port), 1013 (r->id.dst_ip), (r->id.dst_port), 1014 dyn_count ); ) 1015 return r; 1016} 1017 1018/** 1019 * lookup dynamic parent rule using pkt and rule as search keys. 1020 * If the lookup fails, then install one. 1021 */ 1022static ipfw_dyn_rule * 1023lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 1024{ 1025 ipfw_dyn_rule *q; 1026 int i; 1027 1028 IPFW_DYN_LOCK_ASSERT(); 1029 1030 if (ipfw_dyn_v) { 1031 i = hash_packet( pkt ); 1032 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next) 1033 if (q->dyn_type == O_LIMIT_PARENT && 1034 rule== q->rule && 1035 pkt->proto == q->id.proto && 1036 pkt->src_ip == q->id.src_ip && 1037 pkt->dst_ip == q->id.dst_ip && 1038 pkt->src_port == q->id.src_port && 1039 pkt->dst_port == q->id.dst_port) { 1040 q->expire = time_second + dyn_short_lifetime; 1041 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);) 1042 return q; 1043 } 1044 } 1045 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 1046} 1047 1048/** 1049 * Install dynamic state for rule type cmd->o.opcode 1050 * 1051 * Returns 1 (failure) if state is not installed because of errors or because 1052 * session limitations are enforced. 1053 */ 1054static int 1055install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 1056 struct ip_fw_args *args) 1057{ 1058 static int last_log; 1059 1060 ipfw_dyn_rule *q; 1061 1062 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n", 1063 cmd->o.opcode, 1064 (args->f_id.src_ip), (args->f_id.src_port), 1065 (args->f_id.dst_ip), (args->f_id.dst_port) );) 1066 1067 IPFW_DYN_LOCK(); 1068 1069 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL); 1070 1071 if (q != NULL) { /* should never occur */ 1072 if (last_log != time_second) { 1073 last_log = time_second; 1074 printf("ipfw: install_state: entry already present, done\n"); 1075 } 1076 IPFW_DYN_UNLOCK(); 1077 return 0; 1078 } 1079 1080 if (dyn_count >= dyn_max) 1081 /* 1082 * Run out of slots, try to remove any expired rule. 1083 */ 1084 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1); 1085 1086 if (dyn_count >= dyn_max) { 1087 if (last_log != time_second) { 1088 last_log = time_second; 1089 printf("ipfw: install_state: Too many dynamic rules\n"); 1090 } 1091 IPFW_DYN_UNLOCK(); 1092 return 1; /* cannot install, notify caller */ 1093 } 1094 1095 switch (cmd->o.opcode) { 1096 case O_KEEP_STATE: /* bidir rule */ 1097 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule); 1098 break; 1099 1100 case O_LIMIT: /* limit number of sessions */ 1101 { 1102 u_int16_t limit_mask = cmd->limit_mask; 1103 struct ipfw_flow_id id; 1104 ipfw_dyn_rule *parent; 1105 1106 DEB(printf("ipfw: installing dyn-limit rule %d\n", 1107 cmd->conn_limit);) 1108 1109 id.dst_ip = id.src_ip = 0; 1110 id.dst_port = id.src_port = 0; 1111 id.proto = args->f_id.proto; 1112 1113 if (limit_mask & DYN_SRC_ADDR) 1114 id.src_ip = args->f_id.src_ip; 1115 if (limit_mask & DYN_DST_ADDR) 1116 id.dst_ip = args->f_id.dst_ip; 1117 if (limit_mask & DYN_SRC_PORT) 1118 id.src_port = args->f_id.src_port; 1119 if (limit_mask & DYN_DST_PORT) 1120 id.dst_port = args->f_id.dst_port; 1121 parent = lookup_dyn_parent(&id, rule); 1122 if (parent == NULL) { 1123 printf("ipfw: add parent failed\n"); 1124 return 1; 1125 } 1126 if (parent->count >= cmd->conn_limit) { 1127 /* 1128 * See if we can remove some expired rule. 1129 */ 1130 remove_dyn_rule(rule, parent); 1131 if (parent->count >= cmd->conn_limit) { 1132 if (fw_verbose && last_log != time_second) { 1133 last_log = time_second; 1134 log(LOG_SECURITY | LOG_DEBUG, 1135 "drop session, too many entries\n"); 1136 } 1137 IPFW_DYN_UNLOCK(); 1138 return 1; 1139 } 1140 } 1141 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent); 1142 } 1143 break; 1144 default: 1145 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode); 1146 IPFW_DYN_UNLOCK(); 1147 return 1; 1148 } 1149 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */ 1150 IPFW_DYN_UNLOCK(); 1151 return 0; 1152} 1153 1154/* 1155 * Transmit a TCP packet, containing either a RST or a keepalive. 1156 * When flags & TH_RST, we are sending a RST packet, because of a 1157 * "reset" action matched the packet. 1158 * Otherwise we are sending a keepalive, and flags & TH_ 1159 */ 1160static void 1161send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags) 1162{ 1163 struct mbuf *m; 1164 struct ip *ip; 1165 struct tcphdr *tcp; 1166 1167 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1168 if (m == 0) 1169 return; 1170 m->m_pkthdr.rcvif = (struct ifnet *)0; 1171 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); 1172 m->m_data += max_linkhdr; 1173 1174 ip = mtod(m, struct ip *); 1175 bzero(ip, m->m_len); 1176 tcp = (struct tcphdr *)(ip + 1); /* no IP options */ 1177 ip->ip_p = IPPROTO_TCP; 1178 tcp->th_off = 5; 1179 /* 1180 * Assume we are sending a RST (or a keepalive in the reverse 1181 * direction), swap src and destination addresses and ports. 1182 */ 1183 ip->ip_src.s_addr = htonl(id->dst_ip); 1184 ip->ip_dst.s_addr = htonl(id->src_ip); 1185 tcp->th_sport = htons(id->dst_port); 1186 tcp->th_dport = htons(id->src_port); 1187 if (flags & TH_RST) { /* we are sending a RST */ 1188 if (flags & TH_ACK) { 1189 tcp->th_seq = htonl(ack); 1190 tcp->th_ack = htonl(0); 1191 tcp->th_flags = TH_RST; 1192 } else { 1193 if (flags & TH_SYN) 1194 seq++; 1195 tcp->th_seq = htonl(0); 1196 tcp->th_ack = htonl(seq); 1197 tcp->th_flags = TH_RST | TH_ACK; 1198 } 1199 } else { 1200 /* 1201 * We are sending a keepalive. flags & TH_SYN determines 1202 * the direction, forward if set, reverse if clear. 1203 * NOTE: seq and ack are always assumed to be correct 1204 * as set by the caller. This may be confusing... 1205 */ 1206 if (flags & TH_SYN) { 1207 /* 1208 * we have to rewrite the correct addresses! 1209 */ 1210 ip->ip_dst.s_addr = htonl(id->dst_ip); 1211 ip->ip_src.s_addr = htonl(id->src_ip); 1212 tcp->th_dport = htons(id->dst_port); 1213 tcp->th_sport = htons(id->src_port); 1214 } 1215 tcp->th_seq = htonl(seq); 1216 tcp->th_ack = htonl(ack); 1217 tcp->th_flags = TH_ACK; 1218 } 1219 /* 1220 * set ip_len to the payload size so we can compute 1221 * the tcp checksum on the pseudoheader 1222 * XXX check this, could save a couple of words ? 1223 */ 1224 ip->ip_len = htons(sizeof(struct tcphdr)); 1225 tcp->th_sum = in_cksum(m, m->m_pkthdr.len); 1226 /* 1227 * now fill fields left out earlier 1228 */ 1229 ip->ip_ttl = ip_defttl; 1230 ip->ip_len = m->m_pkthdr.len; 1231 m->m_flags |= M_SKIP_FIREWALL; 1232 ip_output(m, NULL, NULL, 0, NULL, NULL); 1233} 1234 1235/* 1236 * sends a reject message, consuming the mbuf passed as an argument. 1237 */ 1238static void 1239send_reject(struct ip_fw_args *args, int code, int offset, int ip_len) 1240{ 1241 1242 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */ 1243 /* We need the IP header in host order for icmp_error(). */ 1244 if (args->eh != NULL) { 1245 struct ip *ip = mtod(args->m, struct ip *); 1246 ip->ip_len = ntohs(ip->ip_len); 1247 ip->ip_off = ntohs(ip->ip_off); 1248 } 1249 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); 1250 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) { 1251 struct tcphdr *const tcp = 1252 L3HDR(struct tcphdr, mtod(args->m, struct ip *)); 1253 if ( (tcp->th_flags & TH_RST) == 0) 1254 send_pkt(&(args->f_id), ntohl(tcp->th_seq), 1255 ntohl(tcp->th_ack), 1256 tcp->th_flags | TH_RST); 1257 m_freem(args->m); 1258 } else 1259 m_freem(args->m); 1260 args->m = NULL; 1261} 1262 1263/** 1264 * 1265 * Given an ip_fw *, lookup_next_rule will return a pointer 1266 * to the next rule, which can be either the jump 1267 * target (for skipto instructions) or the next one in the list (in 1268 * all other cases including a missing jump target). 1269 * The result is also written in the "next_rule" field of the rule. 1270 * Backward jumps are not allowed, so start looking from the next 1271 * rule... 1272 * 1273 * This never returns NULL -- in case we do not have an exact match, 1274 * the next rule is returned. When the ruleset is changed, 1275 * pointers are flushed so we are always correct. 1276 */ 1277 1278static struct ip_fw * 1279lookup_next_rule(struct ip_fw *me) 1280{ 1281 struct ip_fw *rule = NULL; 1282 ipfw_insn *cmd; 1283 1284 /* look for action, in case it is a skipto */ 1285 cmd = ACTION_PTR(me); 1286 if (cmd->opcode == O_LOG) 1287 cmd += F_LEN(cmd); 1288 if ( cmd->opcode == O_SKIPTO ) 1289 for (rule = me->next; rule ; rule = rule->next) 1290 if (rule->rulenum >= cmd->arg1) 1291 break; 1292 if (rule == NULL) /* failure or not a skipto */ 1293 rule = me->next; 1294 me->next_rule = rule; 1295 return rule; 1296} 1297 1298static int 1299check_uidgid(ipfw_insn_u32 *insn, 1300 int proto, struct ifnet *oif, 1301 struct in_addr dst_ip, u_int16_t dst_port, 1302 struct in_addr src_ip, u_int16_t src_port) 1303{ 1304 struct inpcbinfo *pi; 1305 int wildcard; 1306 struct inpcb *pcb; 1307 int match; 1308 1309 if (proto == IPPROTO_TCP) { 1310 wildcard = 0; 1311 pi = &tcbinfo; 1312 } else if (proto == IPPROTO_UDP) { 1313 wildcard = 1; 1314 pi = &udbinfo; 1315 } else 1316 return 0; 1317 1318 match = 0; 1319 1320 INP_INFO_RLOCK(pi); /* XXX LOR with IPFW */ 1321 pcb = (oif) ? 1322 in_pcblookup_hash(pi, 1323 dst_ip, htons(dst_port), 1324 src_ip, htons(src_port), 1325 wildcard, oif) : 1326 in_pcblookup_hash(pi, 1327 src_ip, htons(src_port), 1328 dst_ip, htons(dst_port), 1329 wildcard, NULL); 1330 if (pcb != NULL) { 1331 INP_LOCK(pcb); 1332 if (pcb->inp_socket != NULL) { 1333#if __FreeBSD_version < 500034 1334#define socheckuid(a,b) ((a)->so_cred->cr_uid != (b)) 1335#endif 1336 if (insn->o.opcode == O_UID) { 1337 match = !socheckuid(pcb->inp_socket, 1338 (uid_t)insn->d[0]); 1339 } else { 1340 match = groupmember((uid_t)insn->d[0], 1341 pcb->inp_socket->so_cred); 1342 } 1343 } 1344 INP_UNLOCK(pcb); 1345 } 1346 INP_INFO_RUNLOCK(pi); 1347 1348 return match; 1349} 1350 1351/* 1352 * The main check routine for the firewall. 1353 * 1354 * All arguments are in args so we can modify them and return them 1355 * back to the caller. 1356 * 1357 * Parameters: 1358 * 1359 * args->m (in/out) The packet; we set to NULL when/if we nuke it. 1360 * Starts with the IP header. 1361 * args->eh (in) Mac header if present, or NULL for layer3 packet. 1362 * args->oif Outgoing interface, or NULL if packet is incoming. 1363 * The incoming interface is in the mbuf. (in) 1364 * args->divert_rule (in/out) 1365 * Skip up to the first rule past this rule number; 1366 * upon return, non-zero port number for divert or tee. 1367 * 1368 * args->rule Pointer to the last matching rule (in/out) 1369 * args->next_hop Socket we are forwarding to (out). 1370 * args->f_id Addresses grabbed from the packet (out) 1371 * 1372 * Return value: 1373 * 1374 * IP_FW_PORT_DENY_FLAG the packet must be dropped. 1375 * 0 The packet is to be accepted and routed normally OR 1376 * the packet was denied/rejected and has been dropped; 1377 * in the latter case, *m is equal to NULL upon return. 1378 * port Divert the packet to port, with these caveats: 1379 * 1380 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead 1381 * of diverting it (ie, 'ipfw tee'). 1382 * 1383 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower 1384 * 16 bits as a dummynet pipe number instead of diverting 1385 */ 1386 1387static int 1388ipfw_chk(struct ip_fw_args *args) 1389{ 1390 /* 1391 * Local variables hold state during the processing of a packet. 1392 * 1393 * IMPORTANT NOTE: to speed up the processing of rules, there 1394 * are some assumption on the values of the variables, which 1395 * are documented here. Should you change them, please check 1396 * the implementation of the various instructions to make sure 1397 * that they still work. 1398 * 1399 * args->eh The MAC header. It is non-null for a layer2 1400 * packet, it is NULL for a layer-3 packet. 1401 * 1402 * m | args->m Pointer to the mbuf, as received from the caller. 1403 * It may change if ipfw_chk() does an m_pullup, or if it 1404 * consumes the packet because it calls send_reject(). 1405 * XXX This has to change, so that ipfw_chk() never modifies 1406 * or consumes the buffer. 1407 * ip is simply an alias of the value of m, and it is kept 1408 * in sync with it (the packet is supposed to start with 1409 * the ip header). 1410 */ 1411 struct mbuf *m = args->m; 1412 struct ip *ip = mtod(m, struct ip *); 1413 1414 /* 1415 * oif | args->oif If NULL, ipfw_chk has been called on the 1416 * inbound path (ether_input, bdg_forward, ip_input). 1417 * If non-NULL, ipfw_chk has been called on the outbound path 1418 * (ether_output, ip_output). 1419 */ 1420 struct ifnet *oif = args->oif; 1421 1422 struct ip_fw *f = NULL; /* matching rule */ 1423 int retval = 0; 1424 1425 /* 1426 * hlen The length of the IPv4 header. 1427 * hlen >0 means we have an IPv4 packet. 1428 */ 1429 u_int hlen = 0; /* hlen >0 means we have an IP pkt */ 1430 1431 /* 1432 * offset The offset of a fragment. offset != 0 means that 1433 * we have a fragment at this offset of an IPv4 packet. 1434 * offset == 0 means that (if this is an IPv4 packet) 1435 * this is the first or only fragment. 1436 */ 1437 u_short offset = 0; 1438 1439 /* 1440 * Local copies of addresses. They are only valid if we have 1441 * an IP packet. 1442 * 1443 * proto The protocol. Set to 0 for non-ip packets, 1444 * or to the protocol read from the packet otherwise. 1445 * proto != 0 means that we have an IPv4 packet. 1446 * 1447 * src_port, dst_port port numbers, in HOST format. Only 1448 * valid for TCP and UDP packets. 1449 * 1450 * src_ip, dst_ip ip addresses, in NETWORK format. 1451 * Only valid for IPv4 packets. 1452 */ 1453 u_int8_t proto; 1454 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */ 1455 struct in_addr src_ip, dst_ip; /* NOTE: network format */ 1456 u_int16_t ip_len=0; 1457 int pktlen; 1458 int dyn_dir = MATCH_UNKNOWN; 1459 ipfw_dyn_rule *q = NULL; 1460 struct ip_fw_chain *chain = &layer3_chain; 1461 1462 if (m->m_flags & M_SKIP_FIREWALL) 1463 return 0; /* accept */ 1464 /* 1465 * dyn_dir = MATCH_UNKNOWN when rules unchecked, 1466 * MATCH_NONE when checked and not matched (q = NULL), 1467 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL) 1468 */ 1469 1470 pktlen = m->m_pkthdr.len; 1471 if (args->eh == NULL || /* layer 3 packet */ 1472 ( m->m_pkthdr.len >= sizeof(struct ip) && 1473 ntohs(args->eh->ether_type) == ETHERTYPE_IP)) 1474 hlen = ip->ip_hl << 2; 1475 1476 /* 1477 * Collect parameters into local variables for faster matching. 1478 */ 1479 if (hlen == 0) { /* do not grab addresses for non-ip pkts */ 1480 proto = args->f_id.proto = 0; /* mark f_id invalid */ 1481 goto after_ip_checks; 1482 } 1483 1484 proto = args->f_id.proto = ip->ip_p; 1485 src_ip = ip->ip_src; 1486 dst_ip = ip->ip_dst; 1487 if (args->eh != NULL) { /* layer 2 packets are as on the wire */ 1488 offset = ntohs(ip->ip_off) & IP_OFFMASK; 1489 ip_len = ntohs(ip->ip_len); 1490 } else { 1491 offset = ip->ip_off & IP_OFFMASK; 1492 ip_len = ip->ip_len; 1493 } 1494 pktlen = ip_len < pktlen ? ip_len : pktlen; 1495 1496#define PULLUP_TO(len) \ 1497 do { \ 1498 if ((m)->m_len < (len)) { \ 1499 args->m = m = m_pullup(m, (len)); \ 1500 if (m == 0) \ 1501 goto pullup_failed; \ 1502 ip = mtod(m, struct ip *); \ 1503 } \ 1504 } while (0) 1505 1506 if (offset == 0) { 1507 switch (proto) { 1508 case IPPROTO_TCP: 1509 { 1510 struct tcphdr *tcp; 1511 1512 PULLUP_TO(hlen + sizeof(struct tcphdr)); 1513 tcp = L3HDR(struct tcphdr, ip); 1514 dst_port = tcp->th_dport; 1515 src_port = tcp->th_sport; 1516 args->f_id.flags = tcp->th_flags; 1517 } 1518 break; 1519 1520 case IPPROTO_UDP: 1521 { 1522 struct udphdr *udp; 1523 1524 PULLUP_TO(hlen + sizeof(struct udphdr)); 1525 udp = L3HDR(struct udphdr, ip); 1526 dst_port = udp->uh_dport; 1527 src_port = udp->uh_sport; 1528 } 1529 break; 1530 1531 case IPPROTO_ICMP: 1532 PULLUP_TO(hlen + 4); /* type, code and checksum. */ 1533 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type; 1534 break; 1535 1536 default: 1537 break; 1538 } 1539#undef PULLUP_TO 1540 } 1541 1542 args->f_id.src_ip = ntohl(src_ip.s_addr); 1543 args->f_id.dst_ip = ntohl(dst_ip.s_addr); 1544 args->f_id.src_port = src_port = ntohs(src_port); 1545 args->f_id.dst_port = dst_port = ntohs(dst_port); 1546 1547after_ip_checks: 1548 IPFW_LOCK(chain); /* XXX expensive? can we run lock free? */ 1549 if (args->rule) { 1550 /* 1551 * Packet has already been tagged. Look for the next rule 1552 * to restart processing. 1553 * 1554 * If fw_one_pass != 0 then just accept it. 1555 * XXX should not happen here, but optimized out in 1556 * the caller. 1557 */ 1558 if (fw_one_pass) { 1559 IPFW_UNLOCK(chain); /* XXX optimize */ 1560 return 0; 1561 } 1562 1563 f = args->rule->next_rule; 1564 if (f == NULL) 1565 f = lookup_next_rule(args->rule); 1566 } else { 1567 /* 1568 * Find the starting rule. It can be either the first 1569 * one, or the one after divert_rule if asked so. 1570 */ 1571 int skipto = args->divert_rule; 1572 1573 f = chain->rules; 1574 if (args->eh == NULL && skipto != 0) { 1575 if (skipto >= IPFW_DEFAULT_RULE) { 1576 IPFW_UNLOCK(chain); 1577 return(IP_FW_PORT_DENY_FLAG); /* invalid */ 1578 } 1579 while (f && f->rulenum <= skipto) 1580 f = f->next; 1581 if (f == NULL) { /* drop packet */ 1582 IPFW_UNLOCK(chain); 1583 return(IP_FW_PORT_DENY_FLAG); 1584 } 1585 } 1586 } 1587 args->divert_rule = 0; /* reset to avoid confusion later */ 1588 1589 /* 1590 * Now scan the rules, and parse microinstructions for each rule. 1591 */ 1592 for (; f; f = f->next) { 1593 int l, cmdlen; 1594 ipfw_insn *cmd; 1595 int skip_or; /* skip rest of OR block */ 1596 1597again: 1598 if (set_disable & (1 << f->set) ) 1599 continue; 1600 1601 skip_or = 0; 1602 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ; 1603 l -= cmdlen, cmd += cmdlen) { 1604 int match; 1605 1606 /* 1607 * check_body is a jump target used when we find a 1608 * CHECK_STATE, and need to jump to the body of 1609 * the target rule. 1610 */ 1611 1612check_body: 1613 cmdlen = F_LEN(cmd); 1614 /* 1615 * An OR block (insn_1 || .. || insn_n) has the 1616 * F_OR bit set in all but the last instruction. 1617 * The first match will set "skip_or", and cause 1618 * the following instructions to be skipped until 1619 * past the one with the F_OR bit clear. 1620 */ 1621 if (skip_or) { /* skip this instruction */ 1622 if ((cmd->len & F_OR) == 0) 1623 skip_or = 0; /* next one is good */ 1624 continue; 1625 } 1626 match = 0; /* set to 1 if we succeed */ 1627 1628 switch (cmd->opcode) { 1629 /* 1630 * The first set of opcodes compares the packet's 1631 * fields with some pattern, setting 'match' if a 1632 * match is found. At the end of the loop there is 1633 * logic to deal with F_NOT and F_OR flags associated 1634 * with the opcode. 1635 */ 1636 case O_NOP: 1637 match = 1; 1638 break; 1639 1640 case O_FORWARD_MAC: 1641 printf("ipfw: opcode %d unimplemented\n", 1642 cmd->opcode); 1643 break; 1644 1645 case O_GID: 1646 case O_UID: 1647 /* 1648 * We only check offset == 0 && proto != 0, 1649 * as this ensures that we have an IPv4 1650 * packet with the ports info. 1651 */ 1652 if (offset!=0) 1653 break; 1654 if (proto == IPPROTO_TCP || 1655 proto == IPPROTO_UDP) 1656 match = check_uidgid( 1657 (ipfw_insn_u32 *)cmd, 1658 proto, oif, 1659 dst_ip, dst_port, 1660 src_ip, src_port); 1661 break; 1662 1663 case O_RECV: 1664 match = iface_match(m->m_pkthdr.rcvif, 1665 (ipfw_insn_if *)cmd); 1666 break; 1667 1668 case O_XMIT: 1669 match = iface_match(oif, (ipfw_insn_if *)cmd); 1670 break; 1671 1672 case O_VIA: 1673 match = iface_match(oif ? oif : 1674 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd); 1675 break; 1676 1677 case O_MACADDR2: 1678 if (args->eh != NULL) { /* have MAC header */ 1679 u_int32_t *want = (u_int32_t *) 1680 ((ipfw_insn_mac *)cmd)->addr; 1681 u_int32_t *mask = (u_int32_t *) 1682 ((ipfw_insn_mac *)cmd)->mask; 1683 u_int32_t *hdr = (u_int32_t *)args->eh; 1684 1685 match = 1686 ( want[0] == (hdr[0] & mask[0]) && 1687 want[1] == (hdr[1] & mask[1]) && 1688 want[2] == (hdr[2] & mask[2]) ); 1689 } 1690 break; 1691 1692 case O_MAC_TYPE: 1693 if (args->eh != NULL) { 1694 u_int16_t t = 1695 ntohs(args->eh->ether_type); 1696 u_int16_t *p = 1697 ((ipfw_insn_u16 *)cmd)->ports; 1698 int i; 1699 1700 for (i = cmdlen - 1; !match && i>0; 1701 i--, p += 2) 1702 match = (t>=p[0] && t<=p[1]); 1703 } 1704 break; 1705 1706 case O_FRAG: 1707 match = (hlen > 0 && offset != 0); 1708 break; 1709 1710 case O_IN: /* "out" is "not in" */ 1711 match = (oif == NULL); 1712 break; 1713 1714 case O_LAYER2: 1715 match = (args->eh != NULL); 1716 break; 1717 1718 case O_PROTO: 1719 /* 1720 * We do not allow an arg of 0 so the 1721 * check of "proto" only suffices. 1722 */ 1723 match = (proto == cmd->arg1); 1724 break; 1725 1726 case O_IP_SRC: 1727 match = (hlen > 0 && 1728 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1729 src_ip.s_addr); 1730 break; 1731 1732 case O_IP_SRC_MASK: 1733 case O_IP_DST_MASK: 1734 if (hlen > 0) { 1735 uint32_t a = 1736 (cmd->opcode == O_IP_DST_MASK) ? 1737 dst_ip.s_addr : src_ip.s_addr; 1738 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d; 1739 int i = cmdlen-1; 1740 1741 for (; !match && i>0; i-= 2, p+= 2) 1742 match = (p[0] == (a & p[1])); 1743 } 1744 break; 1745 1746 case O_IP_SRC_ME: 1747 if (hlen > 0) { 1748 struct ifnet *tif; 1749 1750 INADDR_TO_IFP(src_ip, tif); 1751 match = (tif != NULL); 1752 } 1753 break; 1754 1755 case O_IP_DST_SET: 1756 case O_IP_SRC_SET: 1757 if (hlen > 0) { 1758 u_int32_t *d = (u_int32_t *)(cmd+1); 1759 u_int32_t addr = 1760 cmd->opcode == O_IP_DST_SET ? 1761 args->f_id.dst_ip : 1762 args->f_id.src_ip; 1763 1764 if (addr < d[0]) 1765 break; 1766 addr -= d[0]; /* subtract base */ 1767 match = (addr < cmd->arg1) && 1768 ( d[ 1 + (addr>>5)] & 1769 (1<<(addr & 0x1f)) ); 1770 } 1771 break; 1772 1773 case O_IP_DST: 1774 match = (hlen > 0 && 1775 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1776 dst_ip.s_addr); 1777 break; 1778 1779 case O_IP_DST_ME: 1780 if (hlen > 0) { 1781 struct ifnet *tif; 1782 1783 INADDR_TO_IFP(dst_ip, tif); 1784 match = (tif != NULL); 1785 } 1786 break; 1787 1788 case O_IP_SRCPORT: 1789 case O_IP_DSTPORT: 1790 /* 1791 * offset == 0 && proto != 0 is enough 1792 * to guarantee that we have an IPv4 1793 * packet with port info. 1794 */ 1795 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP) 1796 && offset == 0) { 1797 u_int16_t x = 1798 (cmd->opcode == O_IP_SRCPORT) ? 1799 src_port : dst_port ; 1800 u_int16_t *p = 1801 ((ipfw_insn_u16 *)cmd)->ports; 1802 int i; 1803 1804 for (i = cmdlen - 1; !match && i>0; 1805 i--, p += 2) 1806 match = (x>=p[0] && x<=p[1]); 1807 } 1808 break; 1809 1810 case O_ICMPTYPE: 1811 match = (offset == 0 && proto==IPPROTO_ICMP && 1812 icmptype_match(ip, (ipfw_insn_u32 *)cmd) ); 1813 break; 1814 1815 case O_IPOPT: 1816 match = (hlen > 0 && ipopts_match(ip, cmd) ); 1817 break; 1818 1819 case O_IPVER: 1820 match = (hlen > 0 && cmd->arg1 == ip->ip_v); 1821 break; 1822 1823 case O_IPID: 1824 case O_IPLEN: 1825 case O_IPTTL: 1826 if (hlen > 0) { /* only for IP packets */ 1827 uint16_t x; 1828 uint16_t *p; 1829 int i; 1830 1831 if (cmd->opcode == O_IPLEN) 1832 x = ip_len; 1833 else if (cmd->opcode == O_IPTTL) 1834 x = ip->ip_ttl; 1835 else /* must be IPID */ 1836 x = ntohs(ip->ip_id); 1837 if (cmdlen == 1) { 1838 match = (cmd->arg1 == x); 1839 break; 1840 } 1841 /* otherwise we have ranges */ 1842 p = ((ipfw_insn_u16 *)cmd)->ports; 1843 i = cmdlen - 1; 1844 for (; !match && i>0; i--, p += 2) 1845 match = (x >= p[0] && x <= p[1]); 1846 } 1847 break; 1848 1849 case O_IPPRECEDENCE: 1850 match = (hlen > 0 && 1851 (cmd->arg1 == (ip->ip_tos & 0xe0)) ); 1852 break; 1853 1854 case O_IPTOS: 1855 match = (hlen > 0 && 1856 flags_match(cmd, ip->ip_tos)); 1857 break; 1858 1859 case O_TCPFLAGS: 1860 match = (proto == IPPROTO_TCP && offset == 0 && 1861 flags_match(cmd, 1862 L3HDR(struct tcphdr,ip)->th_flags)); 1863 break; 1864 1865 case O_TCPOPTS: 1866 match = (proto == IPPROTO_TCP && offset == 0 && 1867 tcpopts_match(ip, cmd)); 1868 break; 1869 1870 case O_TCPSEQ: 1871 match = (proto == IPPROTO_TCP && offset == 0 && 1872 ((ipfw_insn_u32 *)cmd)->d[0] == 1873 L3HDR(struct tcphdr,ip)->th_seq); 1874 break; 1875 1876 case O_TCPACK: 1877 match = (proto == IPPROTO_TCP && offset == 0 && 1878 ((ipfw_insn_u32 *)cmd)->d[0] == 1879 L3HDR(struct tcphdr,ip)->th_ack); 1880 break; 1881 1882 case O_TCPWIN: 1883 match = (proto == IPPROTO_TCP && offset == 0 && 1884 cmd->arg1 == 1885 L3HDR(struct tcphdr,ip)->th_win); 1886 break; 1887 1888 case O_ESTAB: 1889 /* reject packets which have SYN only */ 1890 /* XXX should i also check for TH_ACK ? */ 1891 match = (proto == IPPROTO_TCP && offset == 0 && 1892 (L3HDR(struct tcphdr,ip)->th_flags & 1893 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); 1894 break; 1895 1896 case O_LOG: 1897 if (fw_verbose) 1898 ipfw_log(f, hlen, args->eh, m, oif); 1899 match = 1; 1900 break; 1901 1902 case O_PROB: 1903 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]); 1904 break; 1905 1906 case O_VERREVPATH: 1907 /* Outgoing packets automatically pass/match */ 1908 match = ((oif != NULL) || 1909 (m->m_pkthdr.rcvif == NULL) || 1910 verify_rev_path(src_ip, m->m_pkthdr.rcvif)); 1911 break; 1912 1913 case O_IPSEC: 1914#ifdef FAST_IPSEC 1915 match = (m_tag_find(m, 1916 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL); 1917#endif 1918#ifdef IPSEC 1919 match = (ipsec_getnhist(m) != NULL); 1920#endif 1921 /* otherwise no match */ 1922 break; 1923 1924 /* 1925 * The second set of opcodes represents 'actions', 1926 * i.e. the terminal part of a rule once the packet 1927 * matches all previous patterns. 1928 * Typically there is only one action for each rule, 1929 * and the opcode is stored at the end of the rule 1930 * (but there are exceptions -- see below). 1931 * 1932 * In general, here we set retval and terminate the 1933 * outer loop (would be a 'break 3' in some language, 1934 * but we need to do a 'goto done'). 1935 * 1936 * Exceptions: 1937 * O_COUNT and O_SKIPTO actions: 1938 * instead of terminating, we jump to the next rule 1939 * ('goto next_rule', equivalent to a 'break 2'), 1940 * or to the SKIPTO target ('goto again' after 1941 * having set f, cmd and l), respectively. 1942 * 1943 * O_LIMIT and O_KEEP_STATE: these opcodes are 1944 * not real 'actions', and are stored right 1945 * before the 'action' part of the rule. 1946 * These opcodes try to install an entry in the 1947 * state tables; if successful, we continue with 1948 * the next opcode (match=1; break;), otherwise 1949 * the packet * must be dropped 1950 * ('goto done' after setting retval); 1951 * 1952 * O_PROBE_STATE and O_CHECK_STATE: these opcodes 1953 * cause a lookup of the state table, and a jump 1954 * to the 'action' part of the parent rule 1955 * ('goto check_body') if an entry is found, or 1956 * (CHECK_STATE only) a jump to the next rule if 1957 * the entry is not found ('goto next_rule'). 1958 * The result of the lookup is cached to make 1959 * further instances of these opcodes are 1960 * effectively NOPs. 1961 */ 1962 case O_LIMIT: 1963 case O_KEEP_STATE: 1964 if (install_state(f, 1965 (ipfw_insn_limit *)cmd, args)) { 1966 retval = IP_FW_PORT_DENY_FLAG; 1967 goto done; /* error/limit violation */ 1968 } 1969 match = 1; 1970 break; 1971 1972 case O_PROBE_STATE: 1973 case O_CHECK_STATE: 1974 /* 1975 * dynamic rules are checked at the first 1976 * keep-state or check-state occurrence, 1977 * with the result being stored in dyn_dir. 1978 * The compiler introduces a PROBE_STATE 1979 * instruction for us when we have a 1980 * KEEP_STATE (because PROBE_STATE needs 1981 * to be run first). 1982 */ 1983 if (dyn_dir == MATCH_UNKNOWN && 1984 (q = lookup_dyn_rule(&args->f_id, 1985 &dyn_dir, proto == IPPROTO_TCP ? 1986 L3HDR(struct tcphdr, ip) : NULL)) 1987 != NULL) { 1988 /* 1989 * Found dynamic entry, update stats 1990 * and jump to the 'action' part of 1991 * the parent rule. 1992 */ 1993 q->pcnt++; 1994 q->bcnt += pktlen; 1995 f = q->rule; 1996 cmd = ACTION_PTR(f); 1997 l = f->cmd_len - f->act_ofs; 1998 IPFW_DYN_UNLOCK(); 1999 goto check_body; 2000 } 2001 /* 2002 * Dynamic entry not found. If CHECK_STATE, 2003 * skip to next rule, if PROBE_STATE just 2004 * ignore and continue with next opcode. 2005 */ 2006 if (cmd->opcode == O_CHECK_STATE) 2007 goto next_rule; 2008 match = 1; 2009 break; 2010 2011 case O_ACCEPT: 2012 retval = 0; /* accept */ 2013 goto done; 2014 2015 case O_PIPE: 2016 case O_QUEUE: 2017 args->rule = f; /* report matching rule */ 2018 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG; 2019 goto done; 2020 2021 case O_DIVERT: 2022 case O_TEE: 2023 if (args->eh) /* not on layer 2 */ 2024 break; 2025 args->divert_rule = f->rulenum; 2026 retval = (cmd->opcode == O_DIVERT) ? 2027 cmd->arg1 : 2028 cmd->arg1 | IP_FW_PORT_TEE_FLAG; 2029 goto done; 2030 2031 case O_COUNT: 2032 case O_SKIPTO: 2033 f->pcnt++; /* update stats */ 2034 f->bcnt += pktlen; 2035 f->timestamp = time_second; 2036 if (cmd->opcode == O_COUNT) 2037 goto next_rule; 2038 /* handle skipto */ 2039 if (f->next_rule == NULL) 2040 lookup_next_rule(f); 2041 f = f->next_rule; 2042 goto again; 2043 2044 case O_REJECT: 2045 /* 2046 * Drop the packet and send a reject notice 2047 * if the packet is not ICMP (or is an ICMP 2048 * query), and it is not multicast/broadcast. 2049 */ 2050 if (hlen > 0 && 2051 (proto != IPPROTO_ICMP || 2052 is_icmp_query(ip)) && 2053 !(m->m_flags & (M_BCAST|M_MCAST)) && 2054 !IN_MULTICAST(dst_ip.s_addr)) { 2055 send_reject(args, cmd->arg1, 2056 offset,ip_len); 2057 m = args->m; 2058 } 2059 /* FALLTHROUGH */ 2060 case O_DENY: 2061 retval = IP_FW_PORT_DENY_FLAG; 2062 goto done; 2063 2064 case O_FORWARD_IP: 2065 if (args->eh) /* not valid on layer2 pkts */ 2066 break; 2067 if (!q || dyn_dir == MATCH_FORWARD) 2068 args->next_hop = 2069 &((ipfw_insn_sa *)cmd)->sa; 2070 retval = 0; 2071 goto done; 2072 2073 default: 2074 panic("-- unknown opcode %d\n", cmd->opcode); 2075 } /* end of switch() on opcodes */ 2076 2077 if (cmd->len & F_NOT) 2078 match = !match; 2079 2080 if (match) { 2081 if (cmd->len & F_OR) 2082 skip_or = 1; 2083 } else { 2084 if (!(cmd->len & F_OR)) /* not an OR block, */ 2085 break; /* try next rule */ 2086 } 2087 2088 } /* end of inner for, scan opcodes */ 2089 2090next_rule:; /* try next rule */ 2091 2092 } /* end of outer for, scan rules */ 2093 printf("ipfw: ouch!, skip past end of rules, denying packet\n"); 2094 IPFW_UNLOCK(chain); 2095 return(IP_FW_PORT_DENY_FLAG); 2096 2097done: 2098 /* Update statistics */ 2099 f->pcnt++; 2100 f->bcnt += pktlen; 2101 f->timestamp = time_second; 2102 IPFW_UNLOCK(chain); 2103 return retval; 2104 2105pullup_failed: 2106 if (fw_verbose) 2107 printf("ipfw: pullup failed\n"); 2108 return(IP_FW_PORT_DENY_FLAG); 2109} 2110 2111/* 2112 * When a rule is added/deleted, clear the next_rule pointers in all rules. 2113 * These will be reconstructed on the fly as packets are matched. 2114 */ 2115static void 2116flush_rule_ptrs(struct ip_fw_chain *chain) 2117{ 2118 struct ip_fw *rule; 2119 2120 IPFW_LOCK_ASSERT(chain); 2121 2122 for (rule = chain->rules; rule; rule = rule->next) 2123 rule->next_rule = NULL; 2124} 2125 2126/* 2127 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given 2128 * pipe/queue, or to all of them (match == NULL). 2129 */ 2130void 2131flush_pipe_ptrs(struct dn_flow_set *match) 2132{ 2133 struct ip_fw *rule; 2134 2135 IPFW_LOCK(&layer3_chain); 2136 for (rule = layer3_chain.rules; rule; rule = rule->next) { 2137 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule); 2138 2139 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE) 2140 continue; 2141 /* 2142 * XXX Use bcmp/bzero to handle pipe_ptr to overcome 2143 * possible alignment problems on 64-bit architectures. 2144 * This code is seldom used so we do not worry too 2145 * much about efficiency. 2146 */ 2147 if (match == NULL || 2148 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) ) 2149 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr)); 2150 } 2151 IPFW_UNLOCK(&layer3_chain); 2152} 2153 2154/* 2155 * Add a new rule to the list. Copy the rule into a malloc'ed area, then 2156 * possibly create a rule number and add the rule to the list. 2157 * Update the rule_number in the input struct so the caller knows it as well. 2158 */ 2159static int 2160add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule) 2161{ 2162 struct ip_fw *rule, *f, *prev; 2163 int l = RULESIZE(input_rule); 2164 2165 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE) 2166 return (EINVAL); 2167 2168 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO); 2169 if (rule == NULL) 2170 return (ENOSPC); 2171 2172 bcopy(input_rule, rule, l); 2173 2174 rule->next = NULL; 2175 rule->next_rule = NULL; 2176 2177 rule->pcnt = 0; 2178 rule->bcnt = 0; 2179 rule->timestamp = 0; 2180 2181 IPFW_LOCK(chain); 2182 2183 if (chain->rules == NULL) { /* default rule */ 2184 chain->rules = rule; 2185 goto done; 2186 } 2187 2188 /* 2189 * If rulenum is 0, find highest numbered rule before the 2190 * default rule, and add autoinc_step 2191 */ 2192 if (autoinc_step < 1) 2193 autoinc_step = 1; 2194 else if (autoinc_step > 1000) 2195 autoinc_step = 1000; 2196 if (rule->rulenum == 0) { 2197 /* 2198 * locate the highest numbered rule before default 2199 */ 2200 for (f = chain->rules; f; f = f->next) { 2201 if (f->rulenum == IPFW_DEFAULT_RULE) 2202 break; 2203 rule->rulenum = f->rulenum; 2204 } 2205 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step) 2206 rule->rulenum += autoinc_step; 2207 input_rule->rulenum = rule->rulenum; 2208 } 2209 2210 /* 2211 * Now insert the new rule in the right place in the sorted list. 2212 */ 2213 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) { 2214 if (f->rulenum > rule->rulenum) { /* found the location */ 2215 if (prev) { 2216 rule->next = f; 2217 prev->next = rule; 2218 } else { /* head insert */ 2219 rule->next = chain->rules; 2220 chain->rules = rule; 2221 } 2222 break; 2223 } 2224 } 2225 flush_rule_ptrs(chain); 2226done: 2227 static_count++; 2228 static_len += l; 2229 IPFW_UNLOCK(chain); 2230 DEB(printf("ipfw: installed rule %d, static count now %d\n", 2231 rule->rulenum, static_count);) 2232 return (0); 2233} 2234 2235/** 2236 * Remove a static rule (including derived * dynamic rules) 2237 * and place it on the ``reap list'' for later reclamation. 2238 * The caller is in charge of clearing rule pointers to avoid 2239 * dangling pointers. 2240 * @return a pointer to the next entry. 2241 * Arguments are not checked, so they better be correct. 2242 */ 2243static struct ip_fw * 2244remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev) 2245{ 2246 struct ip_fw *n; 2247 int l = RULESIZE(rule); 2248 2249 IPFW_LOCK_ASSERT(chain); 2250 2251 n = rule->next; 2252 IPFW_DYN_LOCK(); 2253 remove_dyn_rule(rule, NULL /* force removal */); 2254 IPFW_DYN_UNLOCK(); 2255 if (prev == NULL) 2256 chain->rules = n; 2257 else 2258 prev->next = n; 2259 static_count--; 2260 static_len -= l; 2261 2262 rule->next = chain->reap; 2263 chain->reap = rule; 2264 2265 return n; 2266} 2267 2268/** 2269 * Reclaim storage associated with a list of rules. This is 2270 * typically the list created using remove_rule. 2271 */ 2272static void 2273reap_rules(struct ip_fw *head) 2274{ 2275 struct ip_fw *rule; 2276 2277 while ((rule = head) != NULL) { 2278 head = head->next; 2279 if (DUMMYNET_LOADED) 2280 ip_dn_ruledel_ptr(rule); 2281 free(rule, M_IPFW); 2282 } 2283} 2284 2285/* 2286 * Remove all rules from a chain (except rules in set RESVD_SET 2287 * unless kill_default = 1). The caller is responsible for 2288 * reclaiming storage for the rules left in chain->reap. 2289 */ 2290static void 2291free_chain(struct ip_fw_chain *chain, int kill_default) 2292{ 2293 struct ip_fw *prev, *rule; 2294 2295 IPFW_LOCK_ASSERT(chain); 2296 2297 flush_rule_ptrs(chain); /* more efficient to do outside the loop */ 2298 for (prev = NULL, rule = chain->rules; rule ; ) 2299 if (kill_default || rule->set != RESVD_SET) 2300 rule = remove_rule(chain, rule, prev); 2301 else { 2302 prev = rule; 2303 rule = rule->next; 2304 } 2305} 2306 2307/** 2308 * Remove all rules with given number, and also do set manipulation. 2309 * Assumes chain != NULL && *chain != NULL. 2310 * 2311 * The argument is an u_int32_t. The low 16 bit are the rule or set number, 2312 * the next 8 bits are the new set, the top 8 bits are the command: 2313 * 2314 * 0 delete rules with given number 2315 * 1 delete rules with given set number 2316 * 2 move rules with given number to new set 2317 * 3 move rules with given set number to new set 2318 * 4 swap sets with given numbers 2319 */ 2320static int 2321del_entry(struct ip_fw_chain *chain, u_int32_t arg) 2322{ 2323 struct ip_fw *prev = NULL, *rule; 2324 u_int16_t rulenum; /* rule or old_set */ 2325 u_int8_t cmd, new_set; 2326 2327 rulenum = arg & 0xffff; 2328 cmd = (arg >> 24) & 0xff; 2329 new_set = (arg >> 16) & 0xff; 2330 2331 if (cmd > 4) 2332 return EINVAL; 2333 if (new_set > RESVD_SET) 2334 return EINVAL; 2335 if (cmd == 0 || cmd == 2) { 2336 if (rulenum >= IPFW_DEFAULT_RULE) 2337 return EINVAL; 2338 } else { 2339 if (rulenum > RESVD_SET) /* old_set */ 2340 return EINVAL; 2341 } 2342 2343 IPFW_LOCK(chain); 2344 rule = chain->rules; 2345 chain->reap = NULL; 2346 switch (cmd) { 2347 case 0: /* delete rules with given number */ 2348 /* 2349 * locate first rule to delete 2350 */ 2351 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next) 2352 ; 2353 if (rule->rulenum != rulenum) { 2354 IPFW_UNLOCK(chain); 2355 return EINVAL; 2356 } 2357 2358 /* 2359 * flush pointers outside the loop, then delete all matching 2360 * rules. prev remains the same throughout the cycle. 2361 */ 2362 flush_rule_ptrs(chain); 2363 while (rule->rulenum == rulenum) 2364 rule = remove_rule(chain, rule, prev); 2365 break; 2366 2367 case 1: /* delete all rules with given set number */ 2368 flush_rule_ptrs(chain); 2369 rule = chain->rules; 2370 while (rule->rulenum < IPFW_DEFAULT_RULE) 2371 if (rule->set == rulenum) 2372 rule = remove_rule(chain, rule, prev); 2373 else { 2374 prev = rule; 2375 rule = rule->next; 2376 } 2377 break; 2378 2379 case 2: /* move rules with given number to new set */ 2380 rule = chain->rules; 2381 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2382 if (rule->rulenum == rulenum) 2383 rule->set = new_set; 2384 break; 2385 2386 case 3: /* move rules with given set number to new set */ 2387 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2388 if (rule->set == rulenum) 2389 rule->set = new_set; 2390 break; 2391 2392 case 4: /* swap two sets */ 2393 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2394 if (rule->set == rulenum) 2395 rule->set = new_set; 2396 else if (rule->set == new_set) 2397 rule->set = rulenum; 2398 break; 2399 } 2400 /* 2401 * Look for rules to reclaim. We grab the list before 2402 * releasing the lock then reclaim them w/o the lock to 2403 * avoid a LOR with dummynet. 2404 */ 2405 rule = chain->reap; 2406 chain->reap = NULL; 2407 IPFW_UNLOCK(chain); 2408 if (rule) 2409 reap_rules(rule); 2410 return 0; 2411} 2412 2413/* 2414 * Clear counters for a specific rule. 2415 * The enclosing "table" is assumed locked. 2416 */ 2417static void 2418clear_counters(struct ip_fw *rule, int log_only) 2419{ 2420 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); 2421 2422 if (log_only == 0) { 2423 rule->bcnt = rule->pcnt = 0; 2424 rule->timestamp = 0; 2425 } 2426 if (l->o.opcode == O_LOG) 2427 l->log_left = l->max_log; 2428} 2429 2430/** 2431 * Reset some or all counters on firewall rules. 2432 * @arg frwl is null to clear all entries, or contains a specific 2433 * rule number. 2434 * @arg log_only is 1 if we only want to reset logs, zero otherwise. 2435 */ 2436static int 2437zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only) 2438{ 2439 struct ip_fw *rule; 2440 char *msg; 2441 2442 IPFW_LOCK(chain); 2443 if (rulenum == 0) { 2444 norule_counter = 0; 2445 for (rule = chain->rules; rule; rule = rule->next) 2446 clear_counters(rule, log_only); 2447 msg = log_only ? "ipfw: All logging counts reset.\n" : 2448 "ipfw: Accounting cleared.\n"; 2449 } else { 2450 int cleared = 0; 2451 /* 2452 * We can have multiple rules with the same number, so we 2453 * need to clear them all. 2454 */ 2455 for (rule = chain->rules; rule; rule = rule->next) 2456 if (rule->rulenum == rulenum) { 2457 while (rule && rule->rulenum == rulenum) { 2458 clear_counters(rule, log_only); 2459 rule = rule->next; 2460 } 2461 cleared = 1; 2462 break; 2463 } 2464 if (!cleared) { /* we did not find any matching rules */ 2465 IPFW_UNLOCK(chain); 2466 return (EINVAL); 2467 } 2468 msg = log_only ? "ipfw: Entry %d logging count reset.\n" : 2469 "ipfw: Entry %d cleared.\n"; 2470 } 2471 IPFW_UNLOCK(chain); 2472 2473 if (fw_verbose) 2474 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum); 2475 return (0); 2476} 2477 2478/* 2479 * Check validity of the structure before insert. 2480 * Fortunately rules are simple, so this mostly need to check rule sizes. 2481 */ 2482static int 2483check_ipfw_struct(struct ip_fw *rule, int size) 2484{ 2485 int l, cmdlen = 0; 2486 int have_action=0; 2487 ipfw_insn *cmd; 2488 2489 if (size < sizeof(*rule)) { 2490 printf("ipfw: rule too short\n"); 2491 return (EINVAL); 2492 } 2493 /* first, check for valid size */ 2494 l = RULESIZE(rule); 2495 if (l != size) { 2496 printf("ipfw: size mismatch (have %d want %d)\n", size, l); 2497 return (EINVAL); 2498 } 2499 /* 2500 * Now go for the individual checks. Very simple ones, basically only 2501 * instruction sizes. 2502 */ 2503 for (l = rule->cmd_len, cmd = rule->cmd ; 2504 l > 0 ; l -= cmdlen, cmd += cmdlen) { 2505 cmdlen = F_LEN(cmd); 2506 if (cmdlen > l) { 2507 printf("ipfw: opcode %d size truncated\n", 2508 cmd->opcode); 2509 return EINVAL; 2510 } 2511 DEB(printf("ipfw: opcode %d\n", cmd->opcode);) 2512 switch (cmd->opcode) { 2513 case O_PROBE_STATE: 2514 case O_KEEP_STATE: 2515 case O_PROTO: 2516 case O_IP_SRC_ME: 2517 case O_IP_DST_ME: 2518 case O_LAYER2: 2519 case O_IN: 2520 case O_FRAG: 2521 case O_IPOPT: 2522 case O_IPTOS: 2523 case O_IPPRECEDENCE: 2524 case O_IPVER: 2525 case O_TCPWIN: 2526 case O_TCPFLAGS: 2527 case O_TCPOPTS: 2528 case O_ESTAB: 2529 case O_VERREVPATH: 2530 case O_IPSEC: 2531 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2532 goto bad_size; 2533 break; 2534 2535 case O_UID: 2536 case O_GID: 2537 case O_IP_SRC: 2538 case O_IP_DST: 2539 case O_TCPSEQ: 2540 case O_TCPACK: 2541 case O_PROB: 2542 case O_ICMPTYPE: 2543 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) 2544 goto bad_size; 2545 break; 2546 2547 case O_LIMIT: 2548 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) 2549 goto bad_size; 2550 break; 2551 2552 case O_LOG: 2553 if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) 2554 goto bad_size; 2555 2556 ((ipfw_insn_log *)cmd)->log_left = 2557 ((ipfw_insn_log *)cmd)->max_log; 2558 2559 break; 2560 2561 case O_IP_SRC_MASK: 2562 case O_IP_DST_MASK: 2563 /* only odd command lengths */ 2564 if ( !(cmdlen & 1) || cmdlen > 31) 2565 goto bad_size; 2566 break; 2567 2568 case O_IP_SRC_SET: 2569 case O_IP_DST_SET: 2570 if (cmd->arg1 == 0 || cmd->arg1 > 256) { 2571 printf("ipfw: invalid set size %d\n", 2572 cmd->arg1); 2573 return EINVAL; 2574 } 2575 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 2576 (cmd->arg1+31)/32 ) 2577 goto bad_size; 2578 break; 2579 2580 case O_MACADDR2: 2581 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) 2582 goto bad_size; 2583 break; 2584 2585 case O_NOP: 2586 case O_IPID: 2587 case O_IPTTL: 2588 case O_IPLEN: 2589 if (cmdlen < 1 || cmdlen > 31) 2590 goto bad_size; 2591 break; 2592 2593 case O_MAC_TYPE: 2594 case O_IP_SRCPORT: 2595 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */ 2596 if (cmdlen < 2 || cmdlen > 31) 2597 goto bad_size; 2598 break; 2599 2600 case O_RECV: 2601 case O_XMIT: 2602 case O_VIA: 2603 if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) 2604 goto bad_size; 2605 break; 2606 2607 case O_PIPE: 2608 case O_QUEUE: 2609 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe)) 2610 goto bad_size; 2611 goto check_action; 2612 2613 case O_FORWARD_IP: 2614 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) 2615 goto bad_size; 2616 goto check_action; 2617 2618 case O_FORWARD_MAC: /* XXX not implemented yet */ 2619 case O_CHECK_STATE: 2620 case O_COUNT: 2621 case O_ACCEPT: 2622 case O_DENY: 2623 case O_REJECT: 2624 case O_SKIPTO: 2625 case O_DIVERT: 2626 case O_TEE: 2627 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2628 goto bad_size; 2629check_action: 2630 if (have_action) { 2631 printf("ipfw: opcode %d, multiple actions" 2632 " not allowed\n", 2633 cmd->opcode); 2634 return EINVAL; 2635 } 2636 have_action = 1; 2637 if (l != cmdlen) { 2638 printf("ipfw: opcode %d, action must be" 2639 " last opcode\n", 2640 cmd->opcode); 2641 return EINVAL; 2642 } 2643 break; 2644 default: 2645 printf("ipfw: opcode %d, unknown opcode\n", 2646 cmd->opcode); 2647 return EINVAL; 2648 } 2649 } 2650 if (have_action == 0) { 2651 printf("ipfw: missing action\n"); 2652 return EINVAL; 2653 } 2654 return 0; 2655 2656bad_size: 2657 printf("ipfw: opcode %d size %d wrong\n", 2658 cmd->opcode, cmdlen); 2659 return EINVAL; 2660} 2661 2662/* 2663 * Copy the static and dynamic rules to the supplied buffer 2664 * and return the amount of space actually used. 2665 */ 2666static size_t 2667ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space) 2668{ 2669 char *bp = buf; 2670 char *ep = bp + space; 2671 struct ip_fw *rule; 2672 int i; 2673 2674 /* XXX this can take a long time and locking will block packet flow */ 2675 IPFW_LOCK(chain); 2676 for (rule = chain->rules; rule ; rule = rule->next) { 2677 /* 2678 * Verify the entry fits in the buffer in case the 2679 * rules changed between calculating buffer space and 2680 * now. This would be better done using a generation 2681 * number but should suffice for now. 2682 */ 2683 i = RULESIZE(rule); 2684 if (bp + i <= ep) { 2685 bcopy(rule, bp, i); 2686 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule), 2687 sizeof(set_disable)); 2688 bp += i; 2689 } 2690 } 2691 IPFW_UNLOCK(chain); 2692 if (ipfw_dyn_v) { 2693 ipfw_dyn_rule *p, *last = NULL; 2694 2695 IPFW_DYN_LOCK(); 2696 for (i = 0 ; i < curr_dyn_buckets; i++) 2697 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) { 2698 if (bp + sizeof *p <= ep) { 2699 ipfw_dyn_rule *dst = 2700 (ipfw_dyn_rule *)bp; 2701 bcopy(p, dst, sizeof *p); 2702 bcopy(&(p->rule->rulenum), &(dst->rule), 2703 sizeof(p->rule->rulenum)); 2704 /* 2705 * store a non-null value in "next". 2706 * The userland code will interpret a 2707 * NULL here as a marker 2708 * for the last dynamic rule. 2709 */ 2710 bcopy(&dst, &dst->next, sizeof(dst)); 2711 last = dst; 2712 dst->expire = 2713 TIME_LEQ(dst->expire, time_second) ? 2714 0 : dst->expire - time_second ; 2715 bp += sizeof(ipfw_dyn_rule); 2716 } 2717 } 2718 IPFW_DYN_UNLOCK(); 2719 if (last != NULL) /* mark last dynamic rule */ 2720 bzero(&last->next, sizeof(last)); 2721 } 2722 return (bp - (char *)buf); 2723} 2724 2725 2726/** 2727 * {set|get}sockopt parser. 2728 */ 2729static int 2730ipfw_ctl(struct sockopt *sopt) 2731{ 2732#define RULE_MAXSIZE (256*sizeof(u_int32_t)) 2733 int error, rule_num; 2734 size_t size; 2735 struct ip_fw *buf, *rule; 2736 u_int32_t rulenum[2]; 2737 2738 /* 2739 * Disallow modifications in really-really secure mode, but still allow 2740 * the logging counters to be reset. 2741 */ 2742 if (sopt->sopt_name == IP_FW_ADD || 2743 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) { 2744#if __FreeBSD_version >= 500034 2745 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 2746 if (error) 2747 return (error); 2748#else /* FreeBSD 4.x */ 2749 if (securelevel >= 3) 2750 return (EPERM); 2751#endif 2752 } 2753 2754 error = 0; 2755 2756 switch (sopt->sopt_name) { 2757 case IP_FW_GET: 2758 /* 2759 * pass up a copy of the current rules. Static rules 2760 * come first (the last of which has number IPFW_DEFAULT_RULE), 2761 * followed by a possibly empty list of dynamic rule. 2762 * The last dynamic rule has NULL in the "next" field. 2763 * 2764 * Note that the calculated size is used to bound the 2765 * amount of data returned to the user. The rule set may 2766 * change between calculating the size and returning the 2767 * data in which case we'll just return what fits. 2768 */ 2769 size = static_len; /* size of static rules */ 2770 if (ipfw_dyn_v) /* add size of dyn.rules */ 2771 size += (dyn_count * sizeof(ipfw_dyn_rule)); 2772 2773 /* 2774 * XXX todo: if the user passes a short length just to know 2775 * how much room is needed, do not bother filling up the 2776 * buffer, just jump to the sooptcopyout. 2777 */ 2778 buf = malloc(size, M_TEMP, M_WAITOK); 2779 error = sooptcopyout(sopt, buf, 2780 ipfw_getrules(&layer3_chain, buf, size)); 2781 free(buf, M_TEMP); 2782 break; 2783 2784 case IP_FW_FLUSH: 2785 /* 2786 * Normally we cannot release the lock on each iteration. 2787 * We could do it here only because we start from the head all 2788 * the times so there is no risk of missing some entries. 2789 * On the other hand, the risk is that we end up with 2790 * a very inconsistent ruleset, so better keep the lock 2791 * around the whole cycle. 2792 * 2793 * XXX this code can be improved by resetting the head of 2794 * the list to point to the default rule, and then freeing 2795 * the old list without the need for a lock. 2796 */ 2797 2798 IPFW_LOCK(&layer3_chain); 2799 layer3_chain.reap = NULL; 2800 free_chain(&layer3_chain, 0 /* keep default rule */); 2801 rule = layer3_chain.reap, layer3_chain.reap = NULL; 2802 IPFW_UNLOCK(&layer3_chain); 2803 if (layer3_chain.reap != NULL) 2804 reap_rules(rule); 2805 break; 2806 2807 case IP_FW_ADD: 2808 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK); 2809 error = sooptcopyin(sopt, rule, RULE_MAXSIZE, 2810 sizeof(struct ip_fw) ); 2811 if (error == 0) 2812 error = check_ipfw_struct(rule, sopt->sopt_valsize); 2813 if (error == 0) { 2814 error = add_rule(&layer3_chain, rule); 2815 size = RULESIZE(rule); 2816 if (!error && sopt->sopt_dir == SOPT_GET) 2817 error = sooptcopyout(sopt, rule, size); 2818 } 2819 free(rule, M_TEMP); 2820 break; 2821 2822 case IP_FW_DEL: 2823 /* 2824 * IP_FW_DEL is used for deleting single rules or sets, 2825 * and (ab)used to atomically manipulate sets. Argument size 2826 * is used to distinguish between the two: 2827 * sizeof(u_int32_t) 2828 * delete single rule or set of rules, 2829 * or reassign rules (or sets) to a different set. 2830 * 2*sizeof(u_int32_t) 2831 * atomic disable/enable sets. 2832 * first u_int32_t contains sets to be disabled, 2833 * second u_int32_t contains sets to be enabled. 2834 */ 2835 error = sooptcopyin(sopt, rulenum, 2836 2*sizeof(u_int32_t), sizeof(u_int32_t)); 2837 if (error) 2838 break; 2839 size = sopt->sopt_valsize; 2840 if (size == sizeof(u_int32_t)) /* delete or reassign */ 2841 error = del_entry(&layer3_chain, rulenum[0]); 2842 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */ 2843 set_disable = 2844 (set_disable | rulenum[0]) & ~rulenum[1] & 2845 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */ 2846 else 2847 error = EINVAL; 2848 break; 2849 2850 case IP_FW_ZERO: 2851 case IP_FW_RESETLOG: /* argument is an int, the rule number */ 2852 rule_num = 0; 2853 if (sopt->sopt_val != 0) { 2854 error = sooptcopyin(sopt, &rule_num, 2855 sizeof(int), sizeof(int)); 2856 if (error) 2857 break; 2858 } 2859 error = zero_entry(&layer3_chain, rule_num, 2860 sopt->sopt_name == IP_FW_RESETLOG); 2861 break; 2862 2863 default: 2864 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name); 2865 error = EINVAL; 2866 } 2867 2868 return (error); 2869#undef RULE_MAXSIZE 2870} 2871 2872/** 2873 * dummynet needs a reference to the default rule, because rules can be 2874 * deleted while packets hold a reference to them. When this happens, 2875 * dummynet changes the reference to the default rule (it could well be a 2876 * NULL pointer, but this way we do not need to check for the special 2877 * case, plus here he have info on the default behaviour). 2878 */ 2879struct ip_fw *ip_fw_default_rule; 2880 2881/* 2882 * This procedure is only used to handle keepalives. It is invoked 2883 * every dyn_keepalive_period 2884 */ 2885static void 2886ipfw_tick(void * __unused unused) 2887{ 2888 int i; 2889 ipfw_dyn_rule *q; 2890 2891 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0) 2892 goto done; 2893 2894 IPFW_DYN_LOCK(); 2895 for (i = 0 ; i < curr_dyn_buckets ; i++) { 2896 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) { 2897 if (q->dyn_type == O_LIMIT_PARENT) 2898 continue; 2899 if (q->id.proto != IPPROTO_TCP) 2900 continue; 2901 if ( (q->state & BOTH_SYN) != BOTH_SYN) 2902 continue; 2903 if (TIME_LEQ( time_second+dyn_keepalive_interval, 2904 q->expire)) 2905 continue; /* too early */ 2906 if (TIME_LEQ(q->expire, time_second)) 2907 continue; /* too late, rule expired */ 2908 2909 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 2910 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0); 2911 } 2912 } 2913 IPFW_DYN_UNLOCK(); 2914done: 2915 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL); 2916} 2917 2918static int 2919ipfw_init(void) 2920{ 2921 struct ip_fw default_rule; 2922 int error; 2923 2924 layer3_chain.rules = NULL; 2925 IPFW_LOCK_INIT(&layer3_chain); 2926 IPFW_DYN_LOCK_INIT(); 2927 callout_init(&ipfw_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0); 2928 2929 bzero(&default_rule, sizeof default_rule); 2930 2931 default_rule.act_ofs = 0; 2932 default_rule.rulenum = IPFW_DEFAULT_RULE; 2933 default_rule.cmd_len = 1; 2934 default_rule.set = RESVD_SET; 2935 2936 default_rule.cmd[0].len = 1; 2937 default_rule.cmd[0].opcode = 2938#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT 2939 1 ? O_ACCEPT : 2940#endif 2941 O_DENY; 2942 2943 error = add_rule(&layer3_chain, &default_rule); 2944 if (error != 0) { 2945 printf("ipfw2: error %u initializing default rule " 2946 "(support disabled)\n", error); 2947 IPFW_DYN_LOCK_DESTROY(); 2948 IPFW_LOCK_DESTROY(&layer3_chain); 2949 return (error); 2950 } 2951 2952 ip_fw_default_rule = layer3_chain.rules; 2953 printf("ipfw2 initialized, divert %s, " 2954 "rule-based forwarding enabled, default to %s, logging ", 2955#ifdef IPDIVERT 2956 "enabled", 2957#else 2958 "disabled", 2959#endif 2960 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny"); 2961 2962#ifdef IPFIREWALL_VERBOSE 2963 fw_verbose = 1; 2964#endif 2965#ifdef IPFIREWALL_VERBOSE_LIMIT 2966 verbose_limit = IPFIREWALL_VERBOSE_LIMIT; 2967#endif 2968 if (fw_verbose == 0) 2969 printf("disabled\n"); 2970 else if (verbose_limit == 0) 2971 printf("unlimited\n"); 2972 else 2973 printf("limited to %d packets/entry by default\n", 2974 verbose_limit); 2975 2976 ip_fw_chk_ptr = ipfw_chk; 2977 ip_fw_ctl_ptr = ipfw_ctl; 2978 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL); 2979 2980 return (0); 2981} 2982 2983static void 2984ipfw_destroy(void) 2985{ 2986 struct ip_fw *reap; 2987 2988 IPFW_LOCK(&layer3_chain); 2989 callout_stop(&ipfw_timeout); 2990 ip_fw_chk_ptr = NULL; 2991 ip_fw_ctl_ptr = NULL; 2992 layer3_chain.reap = NULL; 2993 free_chain(&layer3_chain, 1 /* kill default rule */); 2994 reap = layer3_chain.reap, layer3_chain.reap = NULL; 2995 IPFW_UNLOCK(&layer3_chain); 2996 if (reap != NULL) 2997 reap_rules(reap); 2998 2999 IPFW_DYN_LOCK_DESTROY(); 3000 IPFW_LOCK_DESTROY(&layer3_chain); 3001 printf("IP firewall unloaded\n"); 3002} 3003 3004static int 3005ipfw_modevent(module_t mod, int type, void *unused) 3006{ 3007 int err = 0; 3008 3009 switch (type) { 3010 case MOD_LOAD: 3011 if (IPFW_LOADED) { 3012 printf("IP firewall already loaded\n"); 3013 err = EEXIST; 3014 } else { 3015 err = ipfw_init(); 3016 } 3017 break; 3018 3019 case MOD_UNLOAD: 3020 ipfw_destroy(); 3021 err = 0; 3022 break; 3023 default: 3024 break; 3025 } 3026 return err; 3027} 3028 3029static moduledata_t ipfwmod = { 3030 "ipfw", 3031 ipfw_modevent, 3032 0 3033}; 3034DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3035MODULE_VERSION(ipfw, 1); 3036#endif /* IPFW2 */ 3037