ip_fw2.c revision 129876
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 129876 2004-05-30 17:57:46Z phk $ 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#include "opt_ipsec.h" 41#ifndef INET 42#error IPFIREWALL requires INET. 43#endif /* INET */ 44#endif 45 46#define IPFW2 1 47#if IPFW2 48#include <sys/param.h> 49#include <sys/systm.h> 50#include <sys/malloc.h> 51#include <sys/mbuf.h> 52#include <sys/kernel.h> 53#include <sys/module.h> 54#include <sys/proc.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/sysctl.h> 58#include <sys/syslog.h> 59#include <sys/ucred.h> 60#include <net/if.h> 61#include <net/route.h> 62#include <netinet/in.h> 63#include <netinet/in_systm.h> 64#include <netinet/in_var.h> 65#include <netinet/in_pcb.h> 66#include <netinet/ip.h> 67#include <netinet/ip_var.h> 68#include <netinet/ip_icmp.h> 69#include <netinet/ip_fw.h> 70#include <netinet/ip_divert.h> 71#include <netinet/ip_dummynet.h> 72#include <netinet/tcp.h> 73#include <netinet/tcp_timer.h> 74#include <netinet/tcp_var.h> 75#include <netinet/tcpip.h> 76#include <netinet/udp.h> 77#include <netinet/udp_var.h> 78 79#ifdef IPSEC 80#include <netinet6/ipsec.h> 81#endif 82 83#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */ 84 85#include <machine/in_cksum.h> /* XXX for in_cksum */ 86 87/* 88 * This is used to avoid that a firewall-generated packet 89 * loops forever through the firewall. Note that it must 90 * be a flag that is unused by other protocols that might 91 * be called from ip_output (e.g. IPsec) and it must be 92 * listed in M_COPYFLAGS in mbuf.h so that if the mbuf chain 93 * is altered on the way through ip_output it is not lost. 94 * It might be better to add an m_tag since the this happens 95 * infrequently. 96 */ 97#define M_SKIP_FIREWALL M_PROTO6 98 99/* 100 * set_disable contains one bit per set value (0..31). 101 * If the bit is set, all rules with the corresponding set 102 * are disabled. Set RESVD_SET(31) is reserved for the default rule 103 * and rules that are not deleted by the flush command, 104 * and CANNOT be disabled. 105 * Rules in set RESVD_SET can only be deleted explicitly. 106 */ 107static u_int32_t set_disable; 108 109static int fw_verbose; 110static int verbose_limit; 111 112static struct callout ipfw_timeout; 113#define IPFW_DEFAULT_RULE 65535 114 115struct ip_fw_chain { 116 struct ip_fw *rules; /* list of rules */ 117 struct ip_fw *reap; /* list of rules to reap */ 118 struct mtx mtx; /* lock guarding rule list */ 119}; 120#define IPFW_LOCK_INIT(_chain) \ 121 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \ 122 MTX_DEF | MTX_RECURSE) 123#define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx) 124#define IPFW_LOCK(_chain) mtx_lock(&(_chain)->mtx) 125#define IPFW_UNLOCK(_chain) mtx_unlock(&(_chain)->mtx) 126#define IPFW_LOCK_ASSERT(_chain) mtx_assert(&(_chain)->mtx, MA_OWNED) 127 128/* 129 * list of rules for layer 3 130 */ 131static struct ip_fw_chain layer3_chain; 132 133MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's"); 134 135static int fw_debug = 1; 136static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */ 137 138#ifdef SYSCTL_NODE 139SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); 140SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, 141 CTLFLAG_RW | CTLFLAG_SECURE3, 142 &fw_enable, 0, "Enable ipfw"); 143SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW, 144 &autoinc_step, 0, "Rule number autincrement step"); 145SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass, 146 CTLFLAG_RW | CTLFLAG_SECURE3, 147 &fw_one_pass, 0, 148 "Only do a single pass through ipfw when using dummynet(4)"); 149SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW, 150 &fw_debug, 0, "Enable printing of debug ip_fw statements"); 151SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, 152 CTLFLAG_RW | CTLFLAG_SECURE3, 153 &fw_verbose, 0, "Log matches to ipfw rules"); 154SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW, 155 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged"); 156 157/* 158 * Description of dynamic rules. 159 * 160 * Dynamic rules are stored in lists accessed through a hash table 161 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 162 * be modified through the sysctl variable dyn_buckets which is 163 * updated when the table becomes empty. 164 * 165 * XXX currently there is only one list, ipfw_dyn. 166 * 167 * When a packet is received, its address fields are first masked 168 * with the mask defined for the rule, then hashed, then matched 169 * against the entries in the corresponding list. 170 * Dynamic rules can be used for different purposes: 171 * + stateful rules; 172 * + enforcing limits on the number of sessions; 173 * + in-kernel NAT (not implemented yet) 174 * 175 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 176 * measured in seconds and depending on the flags. 177 * 178 * The total number of dynamic rules is stored in dyn_count. 179 * The max number of dynamic rules is dyn_max. When we reach 180 * the maximum number of rules we do not create anymore. This is 181 * done to avoid consuming too much memory, but also too much 182 * time when searching on each packet (ideally, we should try instead 183 * to put a limit on the length of the list on each bucket...). 184 * 185 * Each dynamic rule holds a pointer to the parent ipfw rule so 186 * we know what action to perform. Dynamic rules are removed when 187 * the parent rule is deleted. XXX we should make them survive. 188 * 189 * There are some limitations with dynamic rules -- we do not 190 * obey the 'randomized match', and we do not do multiple 191 * passes through the firewall. XXX check the latter!!! 192 */ 193static ipfw_dyn_rule **ipfw_dyn_v = NULL; 194static u_int32_t dyn_buckets = 256; /* must be power of 2 */ 195static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */ 196 197static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */ 198#define IPFW_DYN_LOCK_INIT() \ 199 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF) 200#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx) 201#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx) 202#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx) 203#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED) 204 205/* 206 * Timeouts for various events in handing dynamic rules. 207 */ 208static u_int32_t dyn_ack_lifetime = 300; 209static u_int32_t dyn_syn_lifetime = 20; 210static u_int32_t dyn_fin_lifetime = 1; 211static u_int32_t dyn_rst_lifetime = 1; 212static u_int32_t dyn_udp_lifetime = 10; 213static u_int32_t dyn_short_lifetime = 5; 214 215/* 216 * Keepalives are sent if dyn_keepalive is set. They are sent every 217 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 218 * seconds of lifetime of a rule. 219 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 220 * than dyn_keepalive_period. 221 */ 222 223static u_int32_t dyn_keepalive_interval = 20; 224static u_int32_t dyn_keepalive_period = 5; 225static u_int32_t dyn_keepalive = 1; /* do send keepalives */ 226 227static u_int32_t static_count; /* # of static rules */ 228static u_int32_t static_len; /* size in bytes of static rules */ 229static u_int32_t dyn_count; /* # of dynamic rules */ 230static u_int32_t dyn_max = 4096; /* max # of dynamic rules */ 231 232SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW, 233 &dyn_buckets, 0, "Number of dyn. buckets"); 234SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD, 235 &curr_dyn_buckets, 0, "Current Number of dyn. buckets"); 236SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD, 237 &dyn_count, 0, "Number of dyn. rules"); 238SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW, 239 &dyn_max, 0, "Max number of dyn. rules"); 240SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD, 241 &static_count, 0, "Number of static rules"); 242SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW, 243 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks"); 244SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW, 245 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn"); 246SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW, 247 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin"); 248SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW, 249 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst"); 250SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW, 251 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP"); 252SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW, 253 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations"); 254SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW, 255 &dyn_keepalive, 0, "Enable keepalives for dyn. rules"); 256 257#endif /* SYSCTL_NODE */ 258 259 260static ip_fw_chk_t ipfw_chk; 261 262ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL; /* hook into dummynet */ 263 264/* 265 * This macro maps an ip pointer into a layer3 header pointer of type T 266 */ 267#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl)) 268 269static __inline int 270icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd) 271{ 272 int type = L3HDR(struct icmp,ip)->icmp_type; 273 274 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) ); 275} 276 277#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \ 278 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) ) 279 280static int 281is_icmp_query(struct ip *ip) 282{ 283 int type = L3HDR(struct icmp, ip)->icmp_type; 284 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) ); 285} 286#undef TT 287 288/* 289 * The following checks use two arrays of 8 or 16 bits to store the 290 * bits that we want set or clear, respectively. They are in the 291 * low and high half of cmd->arg1 or cmd->d[0]. 292 * 293 * We scan options and store the bits we find set. We succeed if 294 * 295 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear 296 * 297 * The code is sometimes optimized not to store additional variables. 298 */ 299 300static int 301flags_match(ipfw_insn *cmd, u_int8_t bits) 302{ 303 u_char want_clear; 304 bits = ~bits; 305 306 if ( ((cmd->arg1 & 0xff) & bits) != 0) 307 return 0; /* some bits we want set were clear */ 308 want_clear = (cmd->arg1 >> 8) & 0xff; 309 if ( (want_clear & bits) != want_clear) 310 return 0; /* some bits we want clear were set */ 311 return 1; 312} 313 314static int 315ipopts_match(struct ip *ip, ipfw_insn *cmd) 316{ 317 int optlen, bits = 0; 318 u_char *cp = (u_char *)(ip + 1); 319 int x = (ip->ip_hl << 2) - sizeof (struct ip); 320 321 for (; x > 0; x -= optlen, cp += optlen) { 322 int opt = cp[IPOPT_OPTVAL]; 323 324 if (opt == IPOPT_EOL) 325 break; 326 if (opt == IPOPT_NOP) 327 optlen = 1; 328 else { 329 optlen = cp[IPOPT_OLEN]; 330 if (optlen <= 0 || optlen > x) 331 return 0; /* invalid or truncated */ 332 } 333 switch (opt) { 334 335 default: 336 break; 337 338 case IPOPT_LSRR: 339 bits |= IP_FW_IPOPT_LSRR; 340 break; 341 342 case IPOPT_SSRR: 343 bits |= IP_FW_IPOPT_SSRR; 344 break; 345 346 case IPOPT_RR: 347 bits |= IP_FW_IPOPT_RR; 348 break; 349 350 case IPOPT_TS: 351 bits |= IP_FW_IPOPT_TS; 352 break; 353 } 354 } 355 return (flags_match(cmd, bits)); 356} 357 358static int 359tcpopts_match(struct ip *ip, ipfw_insn *cmd) 360{ 361 int optlen, bits = 0; 362 struct tcphdr *tcp = L3HDR(struct tcphdr,ip); 363 u_char *cp = (u_char *)(tcp + 1); 364 int x = (tcp->th_off << 2) - sizeof(struct tcphdr); 365 366 for (; x > 0; x -= optlen, cp += optlen) { 367 int opt = cp[0]; 368 if (opt == TCPOPT_EOL) 369 break; 370 if (opt == TCPOPT_NOP) 371 optlen = 1; 372 else { 373 optlen = cp[1]; 374 if (optlen <= 0) 375 break; 376 } 377 378 switch (opt) { 379 380 default: 381 break; 382 383 case TCPOPT_MAXSEG: 384 bits |= IP_FW_TCPOPT_MSS; 385 break; 386 387 case TCPOPT_WINDOW: 388 bits |= IP_FW_TCPOPT_WINDOW; 389 break; 390 391 case TCPOPT_SACK_PERMITTED: 392 case TCPOPT_SACK: 393 bits |= IP_FW_TCPOPT_SACK; 394 break; 395 396 case TCPOPT_TIMESTAMP: 397 bits |= IP_FW_TCPOPT_TS; 398 break; 399 400 case TCPOPT_CC: 401 case TCPOPT_CCNEW: 402 case TCPOPT_CCECHO: 403 bits |= IP_FW_TCPOPT_CC; 404 break; 405 } 406 } 407 return (flags_match(cmd, bits)); 408} 409 410static int 411iface_match(struct ifnet *ifp, ipfw_insn_if *cmd) 412{ 413 if (ifp == NULL) /* no iface with this packet, match fails */ 414 return 0; 415 /* Check by name or by IP address */ 416 if (cmd->name[0] != '\0') { /* match by name */ 417 /* Check name */ 418 if (cmd->p.glob) { 419 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0) 420 return(1); 421 } else { 422 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0) 423 return(1); 424 } 425 } else { 426 struct ifaddr *ia; 427 428 /* XXX lock? */ 429 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) { 430 if (ia->ifa_addr == NULL) 431 continue; 432 if (ia->ifa_addr->sa_family != AF_INET) 433 continue; 434 if (cmd->p.ip.s_addr == ((struct sockaddr_in *) 435 (ia->ifa_addr))->sin_addr.s_addr) 436 return(1); /* match */ 437 } 438 } 439 return(0); /* no match, fail ... */ 440} 441 442/* 443 * The verify_path function checks if a route to the src exists and 444 * if it is reachable via ifp (when provided). 445 * 446 * The 'verrevpath' option checks that the interface that an IP packet 447 * arrives on is the same interface that traffic destined for the 448 * packet's source address would be routed out of. The 'versrcreach' 449 * option just checks that the source address is reachable via any route 450 * (except default) in the routing table. These two are a measure to block 451 * forged packets. This is also commonly known as "anti-spoofing" or Unicast 452 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs 453 * is purposely reminiscent of the Cisco IOS command, 454 * 455 * ip verify unicast reverse-path 456 * ip verify unicast source reachable-via any 457 * 458 * which implements the same functionality. But note that syntax is 459 * misleading. The check may be performed on all IP packets whether unicast, 460 * multicast, or broadcast. 461 */ 462static int 463verify_path(struct in_addr src, struct ifnet *ifp) 464{ 465 struct route ro; 466 struct sockaddr_in *dst; 467 468 bzero(&ro, sizeof(ro)); 469 470 dst = (struct sockaddr_in *)&(ro.ro_dst); 471 dst->sin_family = AF_INET; 472 dst->sin_len = sizeof(*dst); 473 dst->sin_addr = src; 474 rtalloc_ign(&ro, RTF_CLONING); 475 476 if (ro.ro_rt == NULL) 477 return 0; 478 479 /* if ifp is provided, check for equality with rtentry */ 480 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) { 481 RTFREE(ro.ro_rt); 482 return 0; 483 } 484 485 /* if no ifp provided, check if rtentry is not default route */ 486 if (ifp == NULL && 487 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) { 488 RTFREE(ro.ro_rt); 489 return 0; 490 } 491 492 /* found valid route */ 493 RTFREE(ro.ro_rt); 494 return 1; 495} 496 497 498static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */ 499 500#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0 501#define SNP(buf) buf, sizeof(buf) 502 503/* 504 * We enter here when we have a rule with O_LOG. 505 * XXX this function alone takes about 2Kbytes of code! 506 */ 507static void 508ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh, 509 struct mbuf *m, struct ifnet *oif) 510{ 511 char *action; 512 int limit_reached = 0; 513 char action2[40], proto[48], fragment[28]; 514 515 fragment[0] = '\0'; 516 proto[0] = '\0'; 517 518 if (f == NULL) { /* bogus pkt */ 519 if (verbose_limit != 0 && norule_counter >= verbose_limit) 520 return; 521 norule_counter++; 522 if (norule_counter == verbose_limit) 523 limit_reached = verbose_limit; 524 action = "Refuse"; 525 } else { /* O_LOG is the first action, find the real one */ 526 ipfw_insn *cmd = ACTION_PTR(f); 527 ipfw_insn_log *l = (ipfw_insn_log *)cmd; 528 529 if (l->max_log != 0 && l->log_left == 0) 530 return; 531 l->log_left--; 532 if (l->log_left == 0) 533 limit_reached = l->max_log; 534 cmd += F_LEN(cmd); /* point to first action */ 535 if (cmd->opcode == O_PROB) 536 cmd += F_LEN(cmd); 537 538 action = action2; 539 switch (cmd->opcode) { 540 case O_DENY: 541 action = "Deny"; 542 break; 543 544 case O_REJECT: 545 if (cmd->arg1==ICMP_REJECT_RST) 546 action = "Reset"; 547 else if (cmd->arg1==ICMP_UNREACH_HOST) 548 action = "Reject"; 549 else 550 snprintf(SNPARGS(action2, 0), "Unreach %d", 551 cmd->arg1); 552 break; 553 554 case O_ACCEPT: 555 action = "Accept"; 556 break; 557 case O_COUNT: 558 action = "Count"; 559 break; 560 case O_DIVERT: 561 snprintf(SNPARGS(action2, 0), "Divert %d", 562 cmd->arg1); 563 break; 564 case O_TEE: 565 snprintf(SNPARGS(action2, 0), "Tee %d", 566 cmd->arg1); 567 break; 568 case O_SKIPTO: 569 snprintf(SNPARGS(action2, 0), "SkipTo %d", 570 cmd->arg1); 571 break; 572 case O_PIPE: 573 snprintf(SNPARGS(action2, 0), "Pipe %d", 574 cmd->arg1); 575 break; 576 case O_QUEUE: 577 snprintf(SNPARGS(action2, 0), "Queue %d", 578 cmd->arg1); 579 break; 580 case O_FORWARD_IP: { 581 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd; 582 int len; 583 584 len = snprintf(SNPARGS(action2, 0), "Forward to %s", 585 inet_ntoa(sa->sa.sin_addr)); 586 if (sa->sa.sin_port) 587 snprintf(SNPARGS(action2, len), ":%d", 588 sa->sa.sin_port); 589 } 590 break; 591 default: 592 action = "UNKNOWN"; 593 break; 594 } 595 } 596 597 if (hlen == 0) { /* non-ip */ 598 snprintf(SNPARGS(proto, 0), "MAC"); 599 } else { 600 struct ip *ip = mtod(m, struct ip *); 601 /* these three are all aliases to the same thing */ 602 struct icmp *const icmp = L3HDR(struct icmp, ip); 603 struct tcphdr *const tcp = (struct tcphdr *)icmp; 604 struct udphdr *const udp = (struct udphdr *)icmp; 605 606 int ip_off, offset, ip_len; 607 608 int len; 609 610 if (eh != NULL) { /* layer 2 packets are as on the wire */ 611 ip_off = ntohs(ip->ip_off); 612 ip_len = ntohs(ip->ip_len); 613 } else { 614 ip_off = ip->ip_off; 615 ip_len = ip->ip_len; 616 } 617 offset = ip_off & IP_OFFMASK; 618 switch (ip->ip_p) { 619 case IPPROTO_TCP: 620 len = snprintf(SNPARGS(proto, 0), "TCP %s", 621 inet_ntoa(ip->ip_src)); 622 if (offset == 0) 623 snprintf(SNPARGS(proto, len), ":%d %s:%d", 624 ntohs(tcp->th_sport), 625 inet_ntoa(ip->ip_dst), 626 ntohs(tcp->th_dport)); 627 else 628 snprintf(SNPARGS(proto, len), " %s", 629 inet_ntoa(ip->ip_dst)); 630 break; 631 632 case IPPROTO_UDP: 633 len = snprintf(SNPARGS(proto, 0), "UDP %s", 634 inet_ntoa(ip->ip_src)); 635 if (offset == 0) 636 snprintf(SNPARGS(proto, len), ":%d %s:%d", 637 ntohs(udp->uh_sport), 638 inet_ntoa(ip->ip_dst), 639 ntohs(udp->uh_dport)); 640 else 641 snprintf(SNPARGS(proto, len), " %s", 642 inet_ntoa(ip->ip_dst)); 643 break; 644 645 case IPPROTO_ICMP: 646 if (offset == 0) 647 len = snprintf(SNPARGS(proto, 0), 648 "ICMP:%u.%u ", 649 icmp->icmp_type, icmp->icmp_code); 650 else 651 len = snprintf(SNPARGS(proto, 0), "ICMP "); 652 len += snprintf(SNPARGS(proto, len), "%s", 653 inet_ntoa(ip->ip_src)); 654 snprintf(SNPARGS(proto, len), " %s", 655 inet_ntoa(ip->ip_dst)); 656 break; 657 658 default: 659 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p, 660 inet_ntoa(ip->ip_src)); 661 snprintf(SNPARGS(proto, len), " %s", 662 inet_ntoa(ip->ip_dst)); 663 break; 664 } 665 666 if (ip_off & (IP_MF | IP_OFFMASK)) 667 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)", 668 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2), 669 offset << 3, 670 (ip_off & IP_MF) ? "+" : ""); 671 } 672 if (oif || m->m_pkthdr.rcvif) 673 log(LOG_SECURITY | LOG_INFO, 674 "ipfw: %d %s %s %s via %s%s\n", 675 f ? f->rulenum : -1, 676 action, proto, oif ? "out" : "in", 677 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname, 678 fragment); 679 else 680 log(LOG_SECURITY | LOG_INFO, 681 "ipfw: %d %s %s [no if info]%s\n", 682 f ? f->rulenum : -1, 683 action, proto, fragment); 684 if (limit_reached) 685 log(LOG_SECURITY | LOG_NOTICE, 686 "ipfw: limit %d reached on entry %d\n", 687 limit_reached, f ? f->rulenum : -1); 688} 689 690/* 691 * IMPORTANT: the hash function for dynamic rules must be commutative 692 * in source and destination (ip,port), because rules are bidirectional 693 * and we want to find both in the same bucket. 694 */ 695static __inline int 696hash_packet(struct ipfw_flow_id *id) 697{ 698 u_int32_t i; 699 700 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 701 i &= (curr_dyn_buckets - 1); 702 return i; 703} 704 705/** 706 * unlink a dynamic rule from a chain. prev is a pointer to 707 * the previous one, q is a pointer to the rule to delete, 708 * head is a pointer to the head of the queue. 709 * Modifies q and potentially also head. 710 */ 711#define UNLINK_DYN_RULE(prev, head, q) { \ 712 ipfw_dyn_rule *old_q = q; \ 713 \ 714 /* remove a refcount to the parent */ \ 715 if (q->dyn_type == O_LIMIT) \ 716 q->parent->count--; \ 717 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\ 718 (q->id.src_ip), (q->id.src_port), \ 719 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \ 720 if (prev != NULL) \ 721 prev->next = q = q->next; \ 722 else \ 723 head = q = q->next; \ 724 dyn_count--; \ 725 free(old_q, M_IPFW); } 726 727#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 728 729/** 730 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 731 * 732 * If keep_me == NULL, rules are deleted even if not expired, 733 * otherwise only expired rules are removed. 734 * 735 * The value of the second parameter is also used to point to identify 736 * a rule we absolutely do not want to remove (e.g. because we are 737 * holding a reference to it -- this is the case with O_LIMIT_PARENT 738 * rules). The pointer is only used for comparison, so any non-null 739 * value will do. 740 */ 741static void 742remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 743{ 744 static u_int32_t last_remove = 0; 745 746#define FORCE (keep_me == NULL) 747 748 ipfw_dyn_rule *prev, *q; 749 int i, pass = 0, max_pass = 0; 750 751 IPFW_DYN_LOCK_ASSERT(); 752 753 if (ipfw_dyn_v == NULL || dyn_count == 0) 754 return; 755 /* do not expire more than once per second, it is useless */ 756 if (!FORCE && last_remove == time_second) 757 return; 758 last_remove = time_second; 759 760 /* 761 * because O_LIMIT refer to parent rules, during the first pass only 762 * remove child and mark any pending LIMIT_PARENT, and remove 763 * them in a second pass. 764 */ 765next_pass: 766 for (i = 0 ; i < curr_dyn_buckets ; i++) { 767 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) { 768 /* 769 * Logic can become complex here, so we split tests. 770 */ 771 if (q == keep_me) 772 goto next; 773 if (rule != NULL && rule != q->rule) 774 goto next; /* not the one we are looking for */ 775 if (q->dyn_type == O_LIMIT_PARENT) { 776 /* 777 * handle parent in the second pass, 778 * record we need one. 779 */ 780 max_pass = 1; 781 if (pass == 0) 782 goto next; 783 if (FORCE && q->count != 0 ) { 784 /* XXX should not happen! */ 785 printf("ipfw: OUCH! cannot remove rule," 786 " count %d\n", q->count); 787 } 788 } else { 789 if (!FORCE && 790 !TIME_LEQ( q->expire, time_second )) 791 goto next; 792 } 793 if (q->dyn_type != O_LIMIT_PARENT || !q->count) { 794 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 795 continue; 796 } 797next: 798 prev=q; 799 q=q->next; 800 } 801 } 802 if (pass++ < max_pass) 803 goto next_pass; 804} 805 806 807/** 808 * lookup a dynamic rule. 809 */ 810static ipfw_dyn_rule * 811lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction, 812 struct tcphdr *tcp) 813{ 814 /* 815 * stateful ipfw extensions. 816 * Lookup into dynamic session queue 817 */ 818#define MATCH_REVERSE 0 819#define MATCH_FORWARD 1 820#define MATCH_NONE 2 821#define MATCH_UNKNOWN 3 822 int i, dir = MATCH_NONE; 823 ipfw_dyn_rule *prev, *q=NULL; 824 825 IPFW_DYN_LOCK_ASSERT(); 826 827 if (ipfw_dyn_v == NULL) 828 goto done; /* not found */ 829 i = hash_packet( pkt ); 830 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) { 831 if (q->dyn_type == O_LIMIT_PARENT && q->count) 832 goto next; 833 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */ 834 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 835 continue; 836 } 837 if (pkt->proto == q->id.proto && 838 q->dyn_type != O_LIMIT_PARENT) { 839 if (pkt->src_ip == q->id.src_ip && 840 pkt->dst_ip == q->id.dst_ip && 841 pkt->src_port == q->id.src_port && 842 pkt->dst_port == q->id.dst_port ) { 843 dir = MATCH_FORWARD; 844 break; 845 } 846 if (pkt->src_ip == q->id.dst_ip && 847 pkt->dst_ip == q->id.src_ip && 848 pkt->src_port == q->id.dst_port && 849 pkt->dst_port == q->id.src_port ) { 850 dir = MATCH_REVERSE; 851 break; 852 } 853 } 854next: 855 prev = q; 856 q = q->next; 857 } 858 if (q == NULL) 859 goto done; /* q = NULL, not found */ 860 861 if ( prev != NULL) { /* found and not in front */ 862 prev->next = q->next; 863 q->next = ipfw_dyn_v[i]; 864 ipfw_dyn_v[i] = q; 865 } 866 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 867 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST); 868 869#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 870#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 871 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8); 872 switch (q->state) { 873 case TH_SYN: /* opening */ 874 q->expire = time_second + dyn_syn_lifetime; 875 break; 876 877 case BOTH_SYN: /* move to established */ 878 case BOTH_SYN | TH_FIN : /* one side tries to close */ 879 case BOTH_SYN | (TH_FIN << 8) : 880 if (tcp) { 881#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 882 u_int32_t ack = ntohl(tcp->th_ack); 883 if (dir == MATCH_FORWARD) { 884 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd)) 885 q->ack_fwd = ack; 886 else { /* ignore out-of-sequence */ 887 break; 888 } 889 } else { 890 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev)) 891 q->ack_rev = ack; 892 else { /* ignore out-of-sequence */ 893 break; 894 } 895 } 896 } 897 q->expire = time_second + dyn_ack_lifetime; 898 break; 899 900 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 901 if (dyn_fin_lifetime >= dyn_keepalive_period) 902 dyn_fin_lifetime = dyn_keepalive_period - 1; 903 q->expire = time_second + dyn_fin_lifetime; 904 break; 905 906 default: 907#if 0 908 /* 909 * reset or some invalid combination, but can also 910 * occur if we use keep-state the wrong way. 911 */ 912 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 913 printf("invalid state: 0x%x\n", q->state); 914#endif 915 if (dyn_rst_lifetime >= dyn_keepalive_period) 916 dyn_rst_lifetime = dyn_keepalive_period - 1; 917 q->expire = time_second + dyn_rst_lifetime; 918 break; 919 } 920 } else if (pkt->proto == IPPROTO_UDP) { 921 q->expire = time_second + dyn_udp_lifetime; 922 } else { 923 /* other protocols */ 924 q->expire = time_second + dyn_short_lifetime; 925 } 926done: 927 if (match_direction) 928 *match_direction = dir; 929 return q; 930} 931 932static ipfw_dyn_rule * 933lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 934 struct tcphdr *tcp) 935{ 936 ipfw_dyn_rule *q; 937 938 IPFW_DYN_LOCK(); 939 q = lookup_dyn_rule_locked(pkt, match_direction, tcp); 940 if (q == NULL) 941 IPFW_DYN_UNLOCK(); 942 /* NB: return table locked when q is not NULL */ 943 return q; 944} 945 946static void 947realloc_dynamic_table(void) 948{ 949 IPFW_DYN_LOCK_ASSERT(); 950 951 /* 952 * Try reallocation, make sure we have a power of 2 and do 953 * not allow more than 64k entries. In case of overflow, 954 * default to 1024. 955 */ 956 957 if (dyn_buckets > 65536) 958 dyn_buckets = 1024; 959 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */ 960 dyn_buckets = curr_dyn_buckets; /* reset */ 961 return; 962 } 963 curr_dyn_buckets = dyn_buckets; 964 if (ipfw_dyn_v != NULL) 965 free(ipfw_dyn_v, M_IPFW); 966 for (;;) { 967 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 968 M_IPFW, M_NOWAIT | M_ZERO); 969 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2) 970 break; 971 curr_dyn_buckets /= 2; 972 } 973} 974 975/** 976 * Install state of type 'type' for a dynamic session. 977 * The hash table contains two type of rules: 978 * - regular rules (O_KEEP_STATE) 979 * - rules for sessions with limited number of sess per user 980 * (O_LIMIT). When they are created, the parent is 981 * increased by 1, and decreased on delete. In this case, 982 * the third parameter is the parent rule and not the chain. 983 * - "parent" rules for the above (O_LIMIT_PARENT). 984 */ 985static ipfw_dyn_rule * 986add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule) 987{ 988 ipfw_dyn_rule *r; 989 int i; 990 991 IPFW_DYN_LOCK_ASSERT(); 992 993 if (ipfw_dyn_v == NULL || 994 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) { 995 realloc_dynamic_table(); 996 if (ipfw_dyn_v == NULL) 997 return NULL; /* failed ! */ 998 } 999 i = hash_packet(id); 1000 1001 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO); 1002 if (r == NULL) { 1003 printf ("ipfw: sorry cannot allocate state\n"); 1004 return NULL; 1005 } 1006 1007 /* increase refcount on parent, and set pointer */ 1008 if (dyn_type == O_LIMIT) { 1009 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 1010 if ( parent->dyn_type != O_LIMIT_PARENT) 1011 panic("invalid parent"); 1012 parent->count++; 1013 r->parent = parent; 1014 rule = parent->rule; 1015 } 1016 1017 r->id = *id; 1018 r->expire = time_second + dyn_syn_lifetime; 1019 r->rule = rule; 1020 r->dyn_type = dyn_type; 1021 r->pcnt = r->bcnt = 0; 1022 r->count = 0; 1023 1024 r->bucket = i; 1025 r->next = ipfw_dyn_v[i]; 1026 ipfw_dyn_v[i] = r; 1027 dyn_count++; 1028 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n", 1029 dyn_type, 1030 (r->id.src_ip), (r->id.src_port), 1031 (r->id.dst_ip), (r->id.dst_port), 1032 dyn_count ); ) 1033 return r; 1034} 1035 1036/** 1037 * lookup dynamic parent rule using pkt and rule as search keys. 1038 * If the lookup fails, then install one. 1039 */ 1040static ipfw_dyn_rule * 1041lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 1042{ 1043 ipfw_dyn_rule *q; 1044 int i; 1045 1046 IPFW_DYN_LOCK_ASSERT(); 1047 1048 if (ipfw_dyn_v) { 1049 i = hash_packet( pkt ); 1050 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next) 1051 if (q->dyn_type == O_LIMIT_PARENT && 1052 rule== q->rule && 1053 pkt->proto == q->id.proto && 1054 pkt->src_ip == q->id.src_ip && 1055 pkt->dst_ip == q->id.dst_ip && 1056 pkt->src_port == q->id.src_port && 1057 pkt->dst_port == q->id.dst_port) { 1058 q->expire = time_second + dyn_short_lifetime; 1059 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);) 1060 return q; 1061 } 1062 } 1063 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 1064} 1065 1066/** 1067 * Install dynamic state for rule type cmd->o.opcode 1068 * 1069 * Returns 1 (failure) if state is not installed because of errors or because 1070 * session limitations are enforced. 1071 */ 1072static int 1073install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 1074 struct ip_fw_args *args) 1075{ 1076 static int last_log; 1077 1078 ipfw_dyn_rule *q; 1079 1080 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n", 1081 cmd->o.opcode, 1082 (args->f_id.src_ip), (args->f_id.src_port), 1083 (args->f_id.dst_ip), (args->f_id.dst_port) );) 1084 1085 IPFW_DYN_LOCK(); 1086 1087 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL); 1088 1089 if (q != NULL) { /* should never occur */ 1090 if (last_log != time_second) { 1091 last_log = time_second; 1092 printf("ipfw: install_state: entry already present, done\n"); 1093 } 1094 IPFW_DYN_UNLOCK(); 1095 return 0; 1096 } 1097 1098 if (dyn_count >= dyn_max) 1099 /* 1100 * Run out of slots, try to remove any expired rule. 1101 */ 1102 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1); 1103 1104 if (dyn_count >= dyn_max) { 1105 if (last_log != time_second) { 1106 last_log = time_second; 1107 printf("ipfw: install_state: Too many dynamic rules\n"); 1108 } 1109 IPFW_DYN_UNLOCK(); 1110 return 1; /* cannot install, notify caller */ 1111 } 1112 1113 switch (cmd->o.opcode) { 1114 case O_KEEP_STATE: /* bidir rule */ 1115 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule); 1116 break; 1117 1118 case O_LIMIT: /* limit number of sessions */ 1119 { 1120 u_int16_t limit_mask = cmd->limit_mask; 1121 struct ipfw_flow_id id; 1122 ipfw_dyn_rule *parent; 1123 1124 DEB(printf("ipfw: installing dyn-limit rule %d\n", 1125 cmd->conn_limit);) 1126 1127 id.dst_ip = id.src_ip = 0; 1128 id.dst_port = id.src_port = 0; 1129 id.proto = args->f_id.proto; 1130 1131 if (limit_mask & DYN_SRC_ADDR) 1132 id.src_ip = args->f_id.src_ip; 1133 if (limit_mask & DYN_DST_ADDR) 1134 id.dst_ip = args->f_id.dst_ip; 1135 if (limit_mask & DYN_SRC_PORT) 1136 id.src_port = args->f_id.src_port; 1137 if (limit_mask & DYN_DST_PORT) 1138 id.dst_port = args->f_id.dst_port; 1139 parent = lookup_dyn_parent(&id, rule); 1140 if (parent == NULL) { 1141 printf("ipfw: add parent failed\n"); 1142 return 1; 1143 } 1144 if (parent->count >= cmd->conn_limit) { 1145 /* 1146 * See if we can remove some expired rule. 1147 */ 1148 remove_dyn_rule(rule, parent); 1149 if (parent->count >= cmd->conn_limit) { 1150 if (fw_verbose && last_log != time_second) { 1151 last_log = time_second; 1152 log(LOG_SECURITY | LOG_DEBUG, 1153 "drop session, too many entries\n"); 1154 } 1155 IPFW_DYN_UNLOCK(); 1156 return 1; 1157 } 1158 } 1159 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent); 1160 } 1161 break; 1162 default: 1163 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode); 1164 IPFW_DYN_UNLOCK(); 1165 return 1; 1166 } 1167 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */ 1168 IPFW_DYN_UNLOCK(); 1169 return 0; 1170} 1171 1172/* 1173 * Transmit a TCP packet, containing either a RST or a keepalive. 1174 * When flags & TH_RST, we are sending a RST packet, because of a 1175 * "reset" action matched the packet. 1176 * Otherwise we are sending a keepalive, and flags & TH_ 1177 */ 1178static void 1179send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags) 1180{ 1181 struct mbuf *m; 1182 struct ip *ip; 1183 struct tcphdr *tcp; 1184 1185 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1186 if (m == 0) 1187 return; 1188 m->m_pkthdr.rcvif = (struct ifnet *)0; 1189 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); 1190 m->m_data += max_linkhdr; 1191 1192 ip = mtod(m, struct ip *); 1193 bzero(ip, m->m_len); 1194 tcp = (struct tcphdr *)(ip + 1); /* no IP options */ 1195 ip->ip_p = IPPROTO_TCP; 1196 tcp->th_off = 5; 1197 /* 1198 * Assume we are sending a RST (or a keepalive in the reverse 1199 * direction), swap src and destination addresses and ports. 1200 */ 1201 ip->ip_src.s_addr = htonl(id->dst_ip); 1202 ip->ip_dst.s_addr = htonl(id->src_ip); 1203 tcp->th_sport = htons(id->dst_port); 1204 tcp->th_dport = htons(id->src_port); 1205 if (flags & TH_RST) { /* we are sending a RST */ 1206 if (flags & TH_ACK) { 1207 tcp->th_seq = htonl(ack); 1208 tcp->th_ack = htonl(0); 1209 tcp->th_flags = TH_RST; 1210 } else { 1211 if (flags & TH_SYN) 1212 seq++; 1213 tcp->th_seq = htonl(0); 1214 tcp->th_ack = htonl(seq); 1215 tcp->th_flags = TH_RST | TH_ACK; 1216 } 1217 } else { 1218 /* 1219 * We are sending a keepalive. flags & TH_SYN determines 1220 * the direction, forward if set, reverse if clear. 1221 * NOTE: seq and ack are always assumed to be correct 1222 * as set by the caller. This may be confusing... 1223 */ 1224 if (flags & TH_SYN) { 1225 /* 1226 * we have to rewrite the correct addresses! 1227 */ 1228 ip->ip_dst.s_addr = htonl(id->dst_ip); 1229 ip->ip_src.s_addr = htonl(id->src_ip); 1230 tcp->th_dport = htons(id->dst_port); 1231 tcp->th_sport = htons(id->src_port); 1232 } 1233 tcp->th_seq = htonl(seq); 1234 tcp->th_ack = htonl(ack); 1235 tcp->th_flags = TH_ACK; 1236 } 1237 /* 1238 * set ip_len to the payload size so we can compute 1239 * the tcp checksum on the pseudoheader 1240 * XXX check this, could save a couple of words ? 1241 */ 1242 ip->ip_len = htons(sizeof(struct tcphdr)); 1243 tcp->th_sum = in_cksum(m, m->m_pkthdr.len); 1244 /* 1245 * now fill fields left out earlier 1246 */ 1247 ip->ip_ttl = ip_defttl; 1248 ip->ip_len = m->m_pkthdr.len; 1249 m->m_flags |= M_SKIP_FIREWALL; 1250 ip_output(m, NULL, NULL, 0, NULL, NULL); 1251} 1252 1253/* 1254 * sends a reject message, consuming the mbuf passed as an argument. 1255 */ 1256static void 1257send_reject(struct ip_fw_args *args, int code, int offset, int ip_len) 1258{ 1259 1260 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */ 1261 /* We need the IP header in host order for icmp_error(). */ 1262 if (args->eh != NULL) { 1263 struct ip *ip = mtod(args->m, struct ip *); 1264 ip->ip_len = ntohs(ip->ip_len); 1265 ip->ip_off = ntohs(ip->ip_off); 1266 } 1267 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); 1268 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) { 1269 struct tcphdr *const tcp = 1270 L3HDR(struct tcphdr, mtod(args->m, struct ip *)); 1271 if ( (tcp->th_flags & TH_RST) == 0) 1272 send_pkt(&(args->f_id), ntohl(tcp->th_seq), 1273 ntohl(tcp->th_ack), 1274 tcp->th_flags | TH_RST); 1275 m_freem(args->m); 1276 } else 1277 m_freem(args->m); 1278 args->m = NULL; 1279} 1280 1281/** 1282 * 1283 * Given an ip_fw *, lookup_next_rule will return a pointer 1284 * to the next rule, which can be either the jump 1285 * target (for skipto instructions) or the next one in the list (in 1286 * all other cases including a missing jump target). 1287 * The result is also written in the "next_rule" field of the rule. 1288 * Backward jumps are not allowed, so start looking from the next 1289 * rule... 1290 * 1291 * This never returns NULL -- in case we do not have an exact match, 1292 * the next rule is returned. When the ruleset is changed, 1293 * pointers are flushed so we are always correct. 1294 */ 1295 1296static struct ip_fw * 1297lookup_next_rule(struct ip_fw *me) 1298{ 1299 struct ip_fw *rule = NULL; 1300 ipfw_insn *cmd; 1301 1302 /* look for action, in case it is a skipto */ 1303 cmd = ACTION_PTR(me); 1304 if (cmd->opcode == O_LOG) 1305 cmd += F_LEN(cmd); 1306 if ( cmd->opcode == O_SKIPTO ) 1307 for (rule = me->next; rule ; rule = rule->next) 1308 if (rule->rulenum >= cmd->arg1) 1309 break; 1310 if (rule == NULL) /* failure or not a skipto */ 1311 rule = me->next; 1312 me->next_rule = rule; 1313 return rule; 1314} 1315 1316static int 1317check_uidgid(ipfw_insn_u32 *insn, 1318 int proto, struct ifnet *oif, 1319 struct in_addr dst_ip, u_int16_t dst_port, 1320 struct in_addr src_ip, u_int16_t src_port) 1321{ 1322 struct inpcbinfo *pi; 1323 int wildcard; 1324 struct inpcb *pcb; 1325 int match; 1326 1327 if (proto == IPPROTO_TCP) { 1328 wildcard = 0; 1329 pi = &tcbinfo; 1330 } else if (proto == IPPROTO_UDP) { 1331 wildcard = 1; 1332 pi = &udbinfo; 1333 } else 1334 return 0; 1335 1336 match = 0; 1337 1338 INP_INFO_RLOCK(pi); /* XXX LOR with IPFW */ 1339 pcb = (oif) ? 1340 in_pcblookup_hash(pi, 1341 dst_ip, htons(dst_port), 1342 src_ip, htons(src_port), 1343 wildcard, oif) : 1344 in_pcblookup_hash(pi, 1345 src_ip, htons(src_port), 1346 dst_ip, htons(dst_port), 1347 wildcard, NULL); 1348 if (pcb != NULL) { 1349 INP_LOCK(pcb); 1350 if (pcb->inp_socket != NULL) { 1351#if __FreeBSD_version < 500034 1352#define socheckuid(a,b) ((a)->so_cred->cr_uid != (b)) 1353#endif 1354 if (insn->o.opcode == O_UID) { 1355 match = !socheckuid(pcb->inp_socket, 1356 (uid_t)insn->d[0]); 1357 } else { 1358 match = groupmember((uid_t)insn->d[0], 1359 pcb->inp_socket->so_cred); 1360 } 1361 } 1362 INP_UNLOCK(pcb); 1363 } 1364 INP_INFO_RUNLOCK(pi); 1365 1366 return match; 1367} 1368 1369/* 1370 * The main check routine for the firewall. 1371 * 1372 * All arguments are in args so we can modify them and return them 1373 * back to the caller. 1374 * 1375 * Parameters: 1376 * 1377 * args->m (in/out) The packet; we set to NULL when/if we nuke it. 1378 * Starts with the IP header. 1379 * args->eh (in) Mac header if present, or NULL for layer3 packet. 1380 * args->oif Outgoing interface, or NULL if packet is incoming. 1381 * The incoming interface is in the mbuf. (in) 1382 * args->divert_rule (in/out) 1383 * Skip up to the first rule past this rule number; 1384 * upon return, non-zero port number for divert or tee. 1385 * 1386 * args->rule Pointer to the last matching rule (in/out) 1387 * args->next_hop Socket we are forwarding to (out). 1388 * args->f_id Addresses grabbed from the packet (out) 1389 * 1390 * Return value: 1391 * 1392 * IP_FW_PORT_DENY_FLAG the packet must be dropped. 1393 * 0 The packet is to be accepted and routed normally OR 1394 * the packet was denied/rejected and has been dropped; 1395 * in the latter case, *m is equal to NULL upon return. 1396 * port Divert the packet to port, with these caveats: 1397 * 1398 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead 1399 * of diverting it (ie, 'ipfw tee'). 1400 * 1401 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower 1402 * 16 bits as a dummynet pipe number instead of diverting 1403 */ 1404 1405static int 1406ipfw_chk(struct ip_fw_args *args) 1407{ 1408 /* 1409 * Local variables hold state during the processing of a packet. 1410 * 1411 * IMPORTANT NOTE: to speed up the processing of rules, there 1412 * are some assumption on the values of the variables, which 1413 * are documented here. Should you change them, please check 1414 * the implementation of the various instructions to make sure 1415 * that they still work. 1416 * 1417 * args->eh The MAC header. It is non-null for a layer2 1418 * packet, it is NULL for a layer-3 packet. 1419 * 1420 * m | args->m Pointer to the mbuf, as received from the caller. 1421 * It may change if ipfw_chk() does an m_pullup, or if it 1422 * consumes the packet because it calls send_reject(). 1423 * XXX This has to change, so that ipfw_chk() never modifies 1424 * or consumes the buffer. 1425 * ip is simply an alias of the value of m, and it is kept 1426 * in sync with it (the packet is supposed to start with 1427 * the ip header). 1428 */ 1429 struct mbuf *m = args->m; 1430 struct ip *ip = mtod(m, struct ip *); 1431 1432 /* 1433 * oif | args->oif If NULL, ipfw_chk has been called on the 1434 * inbound path (ether_input, bdg_forward, ip_input). 1435 * If non-NULL, ipfw_chk has been called on the outbound path 1436 * (ether_output, ip_output). 1437 */ 1438 struct ifnet *oif = args->oif; 1439 1440 struct ip_fw *f = NULL; /* matching rule */ 1441 int retval = 0; 1442 1443 /* 1444 * hlen The length of the IPv4 header. 1445 * hlen >0 means we have an IPv4 packet. 1446 */ 1447 u_int hlen = 0; /* hlen >0 means we have an IP pkt */ 1448 1449 /* 1450 * offset The offset of a fragment. offset != 0 means that 1451 * we have a fragment at this offset of an IPv4 packet. 1452 * offset == 0 means that (if this is an IPv4 packet) 1453 * this is the first or only fragment. 1454 */ 1455 u_short offset = 0; 1456 1457 /* 1458 * Local copies of addresses. They are only valid if we have 1459 * an IP packet. 1460 * 1461 * proto The protocol. Set to 0 for non-ip packets, 1462 * or to the protocol read from the packet otherwise. 1463 * proto != 0 means that we have an IPv4 packet. 1464 * 1465 * src_port, dst_port port numbers, in HOST format. Only 1466 * valid for TCP and UDP packets. 1467 * 1468 * src_ip, dst_ip ip addresses, in NETWORK format. 1469 * Only valid for IPv4 packets. 1470 */ 1471 u_int8_t proto; 1472 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */ 1473 struct in_addr src_ip, dst_ip; /* NOTE: network format */ 1474 u_int16_t ip_len=0; 1475 int pktlen; 1476 int dyn_dir = MATCH_UNKNOWN; 1477 ipfw_dyn_rule *q = NULL; 1478 struct ip_fw_chain *chain = &layer3_chain; 1479 struct m_tag *mtag; 1480 1481 if (m->m_flags & M_SKIP_FIREWALL) 1482 return 0; /* accept */ 1483 /* 1484 * dyn_dir = MATCH_UNKNOWN when rules unchecked, 1485 * MATCH_NONE when checked and not matched (q = NULL), 1486 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL) 1487 */ 1488 1489 pktlen = m->m_pkthdr.len; 1490 if (args->eh == NULL || /* layer 3 packet */ 1491 ( m->m_pkthdr.len >= sizeof(struct ip) && 1492 ntohs(args->eh->ether_type) == ETHERTYPE_IP)) 1493 hlen = ip->ip_hl << 2; 1494 1495 /* 1496 * Collect parameters into local variables for faster matching. 1497 */ 1498 if (hlen == 0) { /* do not grab addresses for non-ip pkts */ 1499 proto = args->f_id.proto = 0; /* mark f_id invalid */ 1500 goto after_ip_checks; 1501 } 1502 1503 proto = args->f_id.proto = ip->ip_p; 1504 src_ip = ip->ip_src; 1505 dst_ip = ip->ip_dst; 1506 if (args->eh != NULL) { /* layer 2 packets are as on the wire */ 1507 offset = ntohs(ip->ip_off) & IP_OFFMASK; 1508 ip_len = ntohs(ip->ip_len); 1509 } else { 1510 offset = ip->ip_off & IP_OFFMASK; 1511 ip_len = ip->ip_len; 1512 } 1513 pktlen = ip_len < pktlen ? ip_len : pktlen; 1514 1515#define PULLUP_TO(len) \ 1516 do { \ 1517 if ((m)->m_len < (len)) { \ 1518 args->m = m = m_pullup(m, (len)); \ 1519 if (m == 0) \ 1520 goto pullup_failed; \ 1521 ip = mtod(m, struct ip *); \ 1522 } \ 1523 } while (0) 1524 1525 if (offset == 0) { 1526 switch (proto) { 1527 case IPPROTO_TCP: 1528 { 1529 struct tcphdr *tcp; 1530 1531 PULLUP_TO(hlen + sizeof(struct tcphdr)); 1532 tcp = L3HDR(struct tcphdr, ip); 1533 dst_port = tcp->th_dport; 1534 src_port = tcp->th_sport; 1535 args->f_id.flags = tcp->th_flags; 1536 } 1537 break; 1538 1539 case IPPROTO_UDP: 1540 { 1541 struct udphdr *udp; 1542 1543 PULLUP_TO(hlen + sizeof(struct udphdr)); 1544 udp = L3HDR(struct udphdr, ip); 1545 dst_port = udp->uh_dport; 1546 src_port = udp->uh_sport; 1547 } 1548 break; 1549 1550 case IPPROTO_ICMP: 1551 PULLUP_TO(hlen + 4); /* type, code and checksum. */ 1552 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type; 1553 break; 1554 1555 default: 1556 break; 1557 } 1558#undef PULLUP_TO 1559 } 1560 1561 args->f_id.src_ip = ntohl(src_ip.s_addr); 1562 args->f_id.dst_ip = ntohl(dst_ip.s_addr); 1563 args->f_id.src_port = src_port = ntohs(src_port); 1564 args->f_id.dst_port = dst_port = ntohs(dst_port); 1565 1566after_ip_checks: 1567 IPFW_LOCK(chain); /* XXX expensive? can we run lock free? */ 1568 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 1569 if (args->rule) { 1570 /* 1571 * Packet has already been tagged. Look for the next rule 1572 * to restart processing. 1573 * 1574 * If fw_one_pass != 0 then just accept it. 1575 * XXX should not happen here, but optimized out in 1576 * the caller. 1577 */ 1578 if (fw_one_pass) { 1579 IPFW_UNLOCK(chain); /* XXX optimize */ 1580 return 0; 1581 } 1582 1583 f = args->rule->next_rule; 1584 if (f == NULL) 1585 f = lookup_next_rule(args->rule); 1586 } else { 1587 /* 1588 * Find the starting rule. It can be either the first 1589 * one, or the one after divert_rule if asked so. 1590 */ 1591 int skipto = mtag ? divert_cookie(mtag) : 0; 1592 1593 f = chain->rules; 1594 if (args->eh == NULL && skipto != 0) { 1595 if (skipto >= IPFW_DEFAULT_RULE) { 1596 IPFW_UNLOCK(chain); 1597 return(IP_FW_PORT_DENY_FLAG); /* invalid */ 1598 } 1599 while (f && f->rulenum <= skipto) 1600 f = f->next; 1601 if (f == NULL) { /* drop packet */ 1602 IPFW_UNLOCK(chain); 1603 return(IP_FW_PORT_DENY_FLAG); 1604 } 1605 } 1606 } 1607 /* reset divert rule to avoid confusion later */ 1608 if (mtag) 1609 m_tag_delete(m, mtag); 1610 1611 /* 1612 * Now scan the rules, and parse microinstructions for each rule. 1613 */ 1614 for (; f; f = f->next) { 1615 int l, cmdlen; 1616 ipfw_insn *cmd; 1617 int skip_or; /* skip rest of OR block */ 1618 1619again: 1620 if (set_disable & (1 << f->set) ) 1621 continue; 1622 1623 skip_or = 0; 1624 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ; 1625 l -= cmdlen, cmd += cmdlen) { 1626 int match; 1627 1628 /* 1629 * check_body is a jump target used when we find a 1630 * CHECK_STATE, and need to jump to the body of 1631 * the target rule. 1632 */ 1633 1634check_body: 1635 cmdlen = F_LEN(cmd); 1636 /* 1637 * An OR block (insn_1 || .. || insn_n) has the 1638 * F_OR bit set in all but the last instruction. 1639 * The first match will set "skip_or", and cause 1640 * the following instructions to be skipped until 1641 * past the one with the F_OR bit clear. 1642 */ 1643 if (skip_or) { /* skip this instruction */ 1644 if ((cmd->len & F_OR) == 0) 1645 skip_or = 0; /* next one is good */ 1646 continue; 1647 } 1648 match = 0; /* set to 1 if we succeed */ 1649 1650 switch (cmd->opcode) { 1651 /* 1652 * The first set of opcodes compares the packet's 1653 * fields with some pattern, setting 'match' if a 1654 * match is found. At the end of the loop there is 1655 * logic to deal with F_NOT and F_OR flags associated 1656 * with the opcode. 1657 */ 1658 case O_NOP: 1659 match = 1; 1660 break; 1661 1662 case O_FORWARD_MAC: 1663 printf("ipfw: opcode %d unimplemented\n", 1664 cmd->opcode); 1665 break; 1666 1667 case O_GID: 1668 case O_UID: 1669 /* 1670 * We only check offset == 0 && proto != 0, 1671 * as this ensures that we have an IPv4 1672 * packet with the ports info. 1673 */ 1674 if (offset!=0) 1675 break; 1676 if (proto == IPPROTO_TCP || 1677 proto == IPPROTO_UDP) 1678 match = check_uidgid( 1679 (ipfw_insn_u32 *)cmd, 1680 proto, oif, 1681 dst_ip, dst_port, 1682 src_ip, src_port); 1683 break; 1684 1685 case O_RECV: 1686 match = iface_match(m->m_pkthdr.rcvif, 1687 (ipfw_insn_if *)cmd); 1688 break; 1689 1690 case O_XMIT: 1691 match = iface_match(oif, (ipfw_insn_if *)cmd); 1692 break; 1693 1694 case O_VIA: 1695 match = iface_match(oif ? oif : 1696 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd); 1697 break; 1698 1699 case O_MACADDR2: 1700 if (args->eh != NULL) { /* have MAC header */ 1701 u_int32_t *want = (u_int32_t *) 1702 ((ipfw_insn_mac *)cmd)->addr; 1703 u_int32_t *mask = (u_int32_t *) 1704 ((ipfw_insn_mac *)cmd)->mask; 1705 u_int32_t *hdr = (u_int32_t *)args->eh; 1706 1707 match = 1708 ( want[0] == (hdr[0] & mask[0]) && 1709 want[1] == (hdr[1] & mask[1]) && 1710 want[2] == (hdr[2] & mask[2]) ); 1711 } 1712 break; 1713 1714 case O_MAC_TYPE: 1715 if (args->eh != NULL) { 1716 u_int16_t t = 1717 ntohs(args->eh->ether_type); 1718 u_int16_t *p = 1719 ((ipfw_insn_u16 *)cmd)->ports; 1720 int i; 1721 1722 for (i = cmdlen - 1; !match && i>0; 1723 i--, p += 2) 1724 match = (t>=p[0] && t<=p[1]); 1725 } 1726 break; 1727 1728 case O_FRAG: 1729 match = (hlen > 0 && offset != 0); 1730 break; 1731 1732 case O_IN: /* "out" is "not in" */ 1733 match = (oif == NULL); 1734 break; 1735 1736 case O_LAYER2: 1737 match = (args->eh != NULL); 1738 break; 1739 1740 case O_PROTO: 1741 /* 1742 * We do not allow an arg of 0 so the 1743 * check of "proto" only suffices. 1744 */ 1745 match = (proto == cmd->arg1); 1746 break; 1747 1748 case O_IP_SRC: 1749 match = (hlen > 0 && 1750 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1751 src_ip.s_addr); 1752 break; 1753 1754 case O_IP_SRC_MASK: 1755 case O_IP_DST_MASK: 1756 if (hlen > 0) { 1757 uint32_t a = 1758 (cmd->opcode == O_IP_DST_MASK) ? 1759 dst_ip.s_addr : src_ip.s_addr; 1760 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d; 1761 int i = cmdlen-1; 1762 1763 for (; !match && i>0; i-= 2, p+= 2) 1764 match = (p[0] == (a & p[1])); 1765 } 1766 break; 1767 1768 case O_IP_SRC_ME: 1769 if (hlen > 0) { 1770 struct ifnet *tif; 1771 1772 INADDR_TO_IFP(src_ip, tif); 1773 match = (tif != NULL); 1774 } 1775 break; 1776 1777 case O_IP_DST_SET: 1778 case O_IP_SRC_SET: 1779 if (hlen > 0) { 1780 u_int32_t *d = (u_int32_t *)(cmd+1); 1781 u_int32_t addr = 1782 cmd->opcode == O_IP_DST_SET ? 1783 args->f_id.dst_ip : 1784 args->f_id.src_ip; 1785 1786 if (addr < d[0]) 1787 break; 1788 addr -= d[0]; /* subtract base */ 1789 match = (addr < cmd->arg1) && 1790 ( d[ 1 + (addr>>5)] & 1791 (1<<(addr & 0x1f)) ); 1792 } 1793 break; 1794 1795 case O_IP_DST: 1796 match = (hlen > 0 && 1797 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1798 dst_ip.s_addr); 1799 break; 1800 1801 case O_IP_DST_ME: 1802 if (hlen > 0) { 1803 struct ifnet *tif; 1804 1805 INADDR_TO_IFP(dst_ip, tif); 1806 match = (tif != NULL); 1807 } 1808 break; 1809 1810 case O_IP_SRCPORT: 1811 case O_IP_DSTPORT: 1812 /* 1813 * offset == 0 && proto != 0 is enough 1814 * to guarantee that we have an IPv4 1815 * packet with port info. 1816 */ 1817 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP) 1818 && offset == 0) { 1819 u_int16_t x = 1820 (cmd->opcode == O_IP_SRCPORT) ? 1821 src_port : dst_port ; 1822 u_int16_t *p = 1823 ((ipfw_insn_u16 *)cmd)->ports; 1824 int i; 1825 1826 for (i = cmdlen - 1; !match && i>0; 1827 i--, p += 2) 1828 match = (x>=p[0] && x<=p[1]); 1829 } 1830 break; 1831 1832 case O_ICMPTYPE: 1833 match = (offset == 0 && proto==IPPROTO_ICMP && 1834 icmptype_match(ip, (ipfw_insn_u32 *)cmd) ); 1835 break; 1836 1837 case O_IPOPT: 1838 match = (hlen > 0 && ipopts_match(ip, cmd) ); 1839 break; 1840 1841 case O_IPVER: 1842 match = (hlen > 0 && cmd->arg1 == ip->ip_v); 1843 break; 1844 1845 case O_IPID: 1846 case O_IPLEN: 1847 case O_IPTTL: 1848 if (hlen > 0) { /* only for IP packets */ 1849 uint16_t x; 1850 uint16_t *p; 1851 int i; 1852 1853 if (cmd->opcode == O_IPLEN) 1854 x = ip_len; 1855 else if (cmd->opcode == O_IPTTL) 1856 x = ip->ip_ttl; 1857 else /* must be IPID */ 1858 x = ntohs(ip->ip_id); 1859 if (cmdlen == 1) { 1860 match = (cmd->arg1 == x); 1861 break; 1862 } 1863 /* otherwise we have ranges */ 1864 p = ((ipfw_insn_u16 *)cmd)->ports; 1865 i = cmdlen - 1; 1866 for (; !match && i>0; i--, p += 2) 1867 match = (x >= p[0] && x <= p[1]); 1868 } 1869 break; 1870 1871 case O_IPPRECEDENCE: 1872 match = (hlen > 0 && 1873 (cmd->arg1 == (ip->ip_tos & 0xe0)) ); 1874 break; 1875 1876 case O_IPTOS: 1877 match = (hlen > 0 && 1878 flags_match(cmd, ip->ip_tos)); 1879 break; 1880 1881 case O_TCPFLAGS: 1882 match = (proto == IPPROTO_TCP && offset == 0 && 1883 flags_match(cmd, 1884 L3HDR(struct tcphdr,ip)->th_flags)); 1885 break; 1886 1887 case O_TCPOPTS: 1888 match = (proto == IPPROTO_TCP && offset == 0 && 1889 tcpopts_match(ip, cmd)); 1890 break; 1891 1892 case O_TCPSEQ: 1893 match = (proto == IPPROTO_TCP && offset == 0 && 1894 ((ipfw_insn_u32 *)cmd)->d[0] == 1895 L3HDR(struct tcphdr,ip)->th_seq); 1896 break; 1897 1898 case O_TCPACK: 1899 match = (proto == IPPROTO_TCP && offset == 0 && 1900 ((ipfw_insn_u32 *)cmd)->d[0] == 1901 L3HDR(struct tcphdr,ip)->th_ack); 1902 break; 1903 1904 case O_TCPWIN: 1905 match = (proto == IPPROTO_TCP && offset == 0 && 1906 cmd->arg1 == 1907 L3HDR(struct tcphdr,ip)->th_win); 1908 break; 1909 1910 case O_ESTAB: 1911 /* reject packets which have SYN only */ 1912 /* XXX should i also check for TH_ACK ? */ 1913 match = (proto == IPPROTO_TCP && offset == 0 && 1914 (L3HDR(struct tcphdr,ip)->th_flags & 1915 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); 1916 break; 1917 1918 case O_LOG: 1919 if (fw_verbose) 1920 ipfw_log(f, hlen, args->eh, m, oif); 1921 match = 1; 1922 break; 1923 1924 case O_PROB: 1925 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]); 1926 break; 1927 1928 case O_VERREVPATH: 1929 /* Outgoing packets automatically pass/match */ 1930 match = ((oif != NULL) || 1931 (m->m_pkthdr.rcvif == NULL) || 1932 verify_path(src_ip, m->m_pkthdr.rcvif)); 1933 break; 1934 1935 case O_VERSRCREACH: 1936 /* Outgoing packets automatically pass/match */ 1937 match = ((oif != NULL) || 1938 verify_path(src_ip, NULL)); 1939 break; 1940 1941 case O_IPSEC: 1942#ifdef FAST_IPSEC 1943 match = (m_tag_find(m, 1944 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL); 1945#endif 1946#ifdef IPSEC 1947 match = (ipsec_getnhist(m) != 0); 1948#endif 1949 /* otherwise no match */ 1950 break; 1951 1952 /* 1953 * The second set of opcodes represents 'actions', 1954 * i.e. the terminal part of a rule once the packet 1955 * matches all previous patterns. 1956 * Typically there is only one action for each rule, 1957 * and the opcode is stored at the end of the rule 1958 * (but there are exceptions -- see below). 1959 * 1960 * In general, here we set retval and terminate the 1961 * outer loop (would be a 'break 3' in some language, 1962 * but we need to do a 'goto done'). 1963 * 1964 * Exceptions: 1965 * O_COUNT and O_SKIPTO actions: 1966 * instead of terminating, we jump to the next rule 1967 * ('goto next_rule', equivalent to a 'break 2'), 1968 * or to the SKIPTO target ('goto again' after 1969 * having set f, cmd and l), respectively. 1970 * 1971 * O_LIMIT and O_KEEP_STATE: these opcodes are 1972 * not real 'actions', and are stored right 1973 * before the 'action' part of the rule. 1974 * These opcodes try to install an entry in the 1975 * state tables; if successful, we continue with 1976 * the next opcode (match=1; break;), otherwise 1977 * the packet * must be dropped 1978 * ('goto done' after setting retval); 1979 * 1980 * O_PROBE_STATE and O_CHECK_STATE: these opcodes 1981 * cause a lookup of the state table, and a jump 1982 * to the 'action' part of the parent rule 1983 * ('goto check_body') if an entry is found, or 1984 * (CHECK_STATE only) a jump to the next rule if 1985 * the entry is not found ('goto next_rule'). 1986 * The result of the lookup is cached to make 1987 * further instances of these opcodes are 1988 * effectively NOPs. 1989 */ 1990 case O_LIMIT: 1991 case O_KEEP_STATE: 1992 if (install_state(f, 1993 (ipfw_insn_limit *)cmd, args)) { 1994 retval = IP_FW_PORT_DENY_FLAG; 1995 goto done; /* error/limit violation */ 1996 } 1997 match = 1; 1998 break; 1999 2000 case O_PROBE_STATE: 2001 case O_CHECK_STATE: 2002 /* 2003 * dynamic rules are checked at the first 2004 * keep-state or check-state occurrence, 2005 * with the result being stored in dyn_dir. 2006 * The compiler introduces a PROBE_STATE 2007 * instruction for us when we have a 2008 * KEEP_STATE (because PROBE_STATE needs 2009 * to be run first). 2010 */ 2011 if (dyn_dir == MATCH_UNKNOWN && 2012 (q = lookup_dyn_rule(&args->f_id, 2013 &dyn_dir, proto == IPPROTO_TCP ? 2014 L3HDR(struct tcphdr, ip) : NULL)) 2015 != NULL) { 2016 /* 2017 * Found dynamic entry, update stats 2018 * and jump to the 'action' part of 2019 * the parent rule. 2020 */ 2021 q->pcnt++; 2022 q->bcnt += pktlen; 2023 f = q->rule; 2024 cmd = ACTION_PTR(f); 2025 l = f->cmd_len - f->act_ofs; 2026 IPFW_DYN_UNLOCK(); 2027 goto check_body; 2028 } 2029 /* 2030 * Dynamic entry not found. If CHECK_STATE, 2031 * skip to next rule, if PROBE_STATE just 2032 * ignore and continue with next opcode. 2033 */ 2034 if (cmd->opcode == O_CHECK_STATE) 2035 goto next_rule; 2036 match = 1; 2037 break; 2038 2039 case O_ACCEPT: 2040 retval = 0; /* accept */ 2041 goto done; 2042 2043 case O_PIPE: 2044 case O_QUEUE: 2045 args->rule = f; /* report matching rule */ 2046 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG; 2047 goto done; 2048 2049 case O_DIVERT: 2050 case O_TEE: { 2051 struct divert_tag *dt; 2052 2053 if (args->eh) /* not on layer 2 */ 2054 break; 2055 mtag = m_tag_get(PACKET_TAG_DIVERT, 2056 sizeof(struct divert_tag), 2057 M_NOWAIT); 2058 if (mtag == NULL) { 2059 /* XXX statistic */ 2060 /* drop packet */ 2061 IPFW_UNLOCK(chain); 2062 return IP_FW_PORT_DENY_FLAG; 2063 } 2064 dt = (struct divert_tag *)(mtag+1); 2065 dt->cookie = f->rulenum; 2066 dt->info = (cmd->opcode == O_DIVERT) ? 2067 cmd->arg1 : 2068 cmd->arg1 | IP_FW_PORT_TEE_FLAG; 2069 m_tag_prepend(m, mtag); 2070 retval = dt->info; 2071 goto done; 2072 } 2073 2074 case O_COUNT: 2075 case O_SKIPTO: 2076 f->pcnt++; /* update stats */ 2077 f->bcnt += pktlen; 2078 f->timestamp = time_second; 2079 if (cmd->opcode == O_COUNT) 2080 goto next_rule; 2081 /* handle skipto */ 2082 if (f->next_rule == NULL) 2083 lookup_next_rule(f); 2084 f = f->next_rule; 2085 goto again; 2086 2087 case O_REJECT: 2088 /* 2089 * Drop the packet and send a reject notice 2090 * if the packet is not ICMP (or is an ICMP 2091 * query), and it is not multicast/broadcast. 2092 */ 2093 if (hlen > 0 && 2094 (proto != IPPROTO_ICMP || 2095 is_icmp_query(ip)) && 2096 !(m->m_flags & (M_BCAST|M_MCAST)) && 2097 !IN_MULTICAST(ntohl(dst_ip.s_addr))) { 2098 send_reject(args, cmd->arg1, 2099 offset,ip_len); 2100 m = args->m; 2101 } 2102 /* FALLTHROUGH */ 2103 case O_DENY: 2104 retval = IP_FW_PORT_DENY_FLAG; 2105 goto done; 2106 2107 case O_FORWARD_IP: 2108 if (args->eh) /* not valid on layer2 pkts */ 2109 break; 2110 if (!q || dyn_dir == MATCH_FORWARD) 2111 args->next_hop = 2112 &((ipfw_insn_sa *)cmd)->sa; 2113 retval = 0; 2114 goto done; 2115 2116 default: 2117 panic("-- unknown opcode %d\n", cmd->opcode); 2118 } /* end of switch() on opcodes */ 2119 2120 if (cmd->len & F_NOT) 2121 match = !match; 2122 2123 if (match) { 2124 if (cmd->len & F_OR) 2125 skip_or = 1; 2126 } else { 2127 if (!(cmd->len & F_OR)) /* not an OR block, */ 2128 break; /* try next rule */ 2129 } 2130 2131 } /* end of inner for, scan opcodes */ 2132 2133next_rule:; /* try next rule */ 2134 2135 } /* end of outer for, scan rules */ 2136 printf("ipfw: ouch!, skip past end of rules, denying packet\n"); 2137 IPFW_UNLOCK(chain); 2138 return(IP_FW_PORT_DENY_FLAG); 2139 2140done: 2141 /* Update statistics */ 2142 f->pcnt++; 2143 f->bcnt += pktlen; 2144 f->timestamp = time_second; 2145 IPFW_UNLOCK(chain); 2146 return retval; 2147 2148pullup_failed: 2149 if (fw_verbose) 2150 printf("ipfw: pullup failed\n"); 2151 return(IP_FW_PORT_DENY_FLAG); 2152} 2153 2154/* 2155 * When a rule is added/deleted, clear the next_rule pointers in all rules. 2156 * These will be reconstructed on the fly as packets are matched. 2157 */ 2158static void 2159flush_rule_ptrs(struct ip_fw_chain *chain) 2160{ 2161 struct ip_fw *rule; 2162 2163 IPFW_LOCK_ASSERT(chain); 2164 2165 for (rule = chain->rules; rule; rule = rule->next) 2166 rule->next_rule = NULL; 2167} 2168 2169/* 2170 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given 2171 * pipe/queue, or to all of them (match == NULL). 2172 */ 2173void 2174flush_pipe_ptrs(struct dn_flow_set *match) 2175{ 2176 struct ip_fw *rule; 2177 2178 IPFW_LOCK(&layer3_chain); 2179 for (rule = layer3_chain.rules; rule; rule = rule->next) { 2180 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule); 2181 2182 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE) 2183 continue; 2184 /* 2185 * XXX Use bcmp/bzero to handle pipe_ptr to overcome 2186 * possible alignment problems on 64-bit architectures. 2187 * This code is seldom used so we do not worry too 2188 * much about efficiency. 2189 */ 2190 if (match == NULL || 2191 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) ) 2192 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr)); 2193 } 2194 IPFW_UNLOCK(&layer3_chain); 2195} 2196 2197/* 2198 * Add a new rule to the list. Copy the rule into a malloc'ed area, then 2199 * possibly create a rule number and add the rule to the list. 2200 * Update the rule_number in the input struct so the caller knows it as well. 2201 */ 2202static int 2203add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule) 2204{ 2205 struct ip_fw *rule, *f, *prev; 2206 int l = RULESIZE(input_rule); 2207 2208 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE) 2209 return (EINVAL); 2210 2211 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO); 2212 if (rule == NULL) 2213 return (ENOSPC); 2214 2215 bcopy(input_rule, rule, l); 2216 2217 rule->next = NULL; 2218 rule->next_rule = NULL; 2219 2220 rule->pcnt = 0; 2221 rule->bcnt = 0; 2222 rule->timestamp = 0; 2223 2224 IPFW_LOCK(chain); 2225 2226 if (chain->rules == NULL) { /* default rule */ 2227 chain->rules = rule; 2228 goto done; 2229 } 2230 2231 /* 2232 * If rulenum is 0, find highest numbered rule before the 2233 * default rule, and add autoinc_step 2234 */ 2235 if (autoinc_step < 1) 2236 autoinc_step = 1; 2237 else if (autoinc_step > 1000) 2238 autoinc_step = 1000; 2239 if (rule->rulenum == 0) { 2240 /* 2241 * locate the highest numbered rule before default 2242 */ 2243 for (f = chain->rules; f; f = f->next) { 2244 if (f->rulenum == IPFW_DEFAULT_RULE) 2245 break; 2246 rule->rulenum = f->rulenum; 2247 } 2248 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step) 2249 rule->rulenum += autoinc_step; 2250 input_rule->rulenum = rule->rulenum; 2251 } 2252 2253 /* 2254 * Now insert the new rule in the right place in the sorted list. 2255 */ 2256 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) { 2257 if (f->rulenum > rule->rulenum) { /* found the location */ 2258 if (prev) { 2259 rule->next = f; 2260 prev->next = rule; 2261 } else { /* head insert */ 2262 rule->next = chain->rules; 2263 chain->rules = rule; 2264 } 2265 break; 2266 } 2267 } 2268 flush_rule_ptrs(chain); 2269done: 2270 static_count++; 2271 static_len += l; 2272 IPFW_UNLOCK(chain); 2273 DEB(printf("ipfw: installed rule %d, static count now %d\n", 2274 rule->rulenum, static_count);) 2275 return (0); 2276} 2277 2278/** 2279 * Remove a static rule (including derived * dynamic rules) 2280 * and place it on the ``reap list'' for later reclamation. 2281 * The caller is in charge of clearing rule pointers to avoid 2282 * dangling pointers. 2283 * @return a pointer to the next entry. 2284 * Arguments are not checked, so they better be correct. 2285 */ 2286static struct ip_fw * 2287remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev) 2288{ 2289 struct ip_fw *n; 2290 int l = RULESIZE(rule); 2291 2292 IPFW_LOCK_ASSERT(chain); 2293 2294 n = rule->next; 2295 IPFW_DYN_LOCK(); 2296 remove_dyn_rule(rule, NULL /* force removal */); 2297 IPFW_DYN_UNLOCK(); 2298 if (prev == NULL) 2299 chain->rules = n; 2300 else 2301 prev->next = n; 2302 static_count--; 2303 static_len -= l; 2304 2305 rule->next = chain->reap; 2306 chain->reap = rule; 2307 2308 return n; 2309} 2310 2311/** 2312 * Reclaim storage associated with a list of rules. This is 2313 * typically the list created using remove_rule. 2314 */ 2315static void 2316reap_rules(struct ip_fw *head) 2317{ 2318 struct ip_fw *rule; 2319 2320 while ((rule = head) != NULL) { 2321 head = head->next; 2322 if (DUMMYNET_LOADED) 2323 ip_dn_ruledel_ptr(rule); 2324 free(rule, M_IPFW); 2325 } 2326} 2327 2328/* 2329 * Remove all rules from a chain (except rules in set RESVD_SET 2330 * unless kill_default = 1). The caller is responsible for 2331 * reclaiming storage for the rules left in chain->reap. 2332 */ 2333static void 2334free_chain(struct ip_fw_chain *chain, int kill_default) 2335{ 2336 struct ip_fw *prev, *rule; 2337 2338 IPFW_LOCK_ASSERT(chain); 2339 2340 flush_rule_ptrs(chain); /* more efficient to do outside the loop */ 2341 for (prev = NULL, rule = chain->rules; rule ; ) 2342 if (kill_default || rule->set != RESVD_SET) 2343 rule = remove_rule(chain, rule, prev); 2344 else { 2345 prev = rule; 2346 rule = rule->next; 2347 } 2348} 2349 2350/** 2351 * Remove all rules with given number, and also do set manipulation. 2352 * Assumes chain != NULL && *chain != NULL. 2353 * 2354 * The argument is an u_int32_t. The low 16 bit are the rule or set number, 2355 * the next 8 bits are the new set, the top 8 bits are the command: 2356 * 2357 * 0 delete rules with given number 2358 * 1 delete rules with given set number 2359 * 2 move rules with given number to new set 2360 * 3 move rules with given set number to new set 2361 * 4 swap sets with given numbers 2362 */ 2363static int 2364del_entry(struct ip_fw_chain *chain, u_int32_t arg) 2365{ 2366 struct ip_fw *prev = NULL, *rule; 2367 u_int16_t rulenum; /* rule or old_set */ 2368 u_int8_t cmd, new_set; 2369 2370 rulenum = arg & 0xffff; 2371 cmd = (arg >> 24) & 0xff; 2372 new_set = (arg >> 16) & 0xff; 2373 2374 if (cmd > 4) 2375 return EINVAL; 2376 if (new_set > RESVD_SET) 2377 return EINVAL; 2378 if (cmd == 0 || cmd == 2) { 2379 if (rulenum >= IPFW_DEFAULT_RULE) 2380 return EINVAL; 2381 } else { 2382 if (rulenum > RESVD_SET) /* old_set */ 2383 return EINVAL; 2384 } 2385 2386 IPFW_LOCK(chain); 2387 rule = chain->rules; 2388 chain->reap = NULL; 2389 switch (cmd) { 2390 case 0: /* delete rules with given number */ 2391 /* 2392 * locate first rule to delete 2393 */ 2394 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next) 2395 ; 2396 if (rule->rulenum != rulenum) { 2397 IPFW_UNLOCK(chain); 2398 return EINVAL; 2399 } 2400 2401 /* 2402 * flush pointers outside the loop, then delete all matching 2403 * rules. prev remains the same throughout the cycle. 2404 */ 2405 flush_rule_ptrs(chain); 2406 while (rule->rulenum == rulenum) 2407 rule = remove_rule(chain, rule, prev); 2408 break; 2409 2410 case 1: /* delete all rules with given set number */ 2411 flush_rule_ptrs(chain); 2412 rule = chain->rules; 2413 while (rule->rulenum < IPFW_DEFAULT_RULE) 2414 if (rule->set == rulenum) 2415 rule = remove_rule(chain, rule, prev); 2416 else { 2417 prev = rule; 2418 rule = rule->next; 2419 } 2420 break; 2421 2422 case 2: /* move rules with given number to new set */ 2423 rule = chain->rules; 2424 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2425 if (rule->rulenum == rulenum) 2426 rule->set = new_set; 2427 break; 2428 2429 case 3: /* move rules with given set number to new set */ 2430 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2431 if (rule->set == rulenum) 2432 rule->set = new_set; 2433 break; 2434 2435 case 4: /* swap two sets */ 2436 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next) 2437 if (rule->set == rulenum) 2438 rule->set = new_set; 2439 else if (rule->set == new_set) 2440 rule->set = rulenum; 2441 break; 2442 } 2443 /* 2444 * Look for rules to reclaim. We grab the list before 2445 * releasing the lock then reclaim them w/o the lock to 2446 * avoid a LOR with dummynet. 2447 */ 2448 rule = chain->reap; 2449 chain->reap = NULL; 2450 IPFW_UNLOCK(chain); 2451 if (rule) 2452 reap_rules(rule); 2453 return 0; 2454} 2455 2456/* 2457 * Clear counters for a specific rule. 2458 * The enclosing "table" is assumed locked. 2459 */ 2460static void 2461clear_counters(struct ip_fw *rule, int log_only) 2462{ 2463 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); 2464 2465 if (log_only == 0) { 2466 rule->bcnt = rule->pcnt = 0; 2467 rule->timestamp = 0; 2468 } 2469 if (l->o.opcode == O_LOG) 2470 l->log_left = l->max_log; 2471} 2472 2473/** 2474 * Reset some or all counters on firewall rules. 2475 * @arg frwl is null to clear all entries, or contains a specific 2476 * rule number. 2477 * @arg log_only is 1 if we only want to reset logs, zero otherwise. 2478 */ 2479static int 2480zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only) 2481{ 2482 struct ip_fw *rule; 2483 char *msg; 2484 2485 IPFW_LOCK(chain); 2486 if (rulenum == 0) { 2487 norule_counter = 0; 2488 for (rule = chain->rules; rule; rule = rule->next) 2489 clear_counters(rule, log_only); 2490 msg = log_only ? "ipfw: All logging counts reset.\n" : 2491 "ipfw: Accounting cleared.\n"; 2492 } else { 2493 int cleared = 0; 2494 /* 2495 * We can have multiple rules with the same number, so we 2496 * need to clear them all. 2497 */ 2498 for (rule = chain->rules; rule; rule = rule->next) 2499 if (rule->rulenum == rulenum) { 2500 while (rule && rule->rulenum == rulenum) { 2501 clear_counters(rule, log_only); 2502 rule = rule->next; 2503 } 2504 cleared = 1; 2505 break; 2506 } 2507 if (!cleared) { /* we did not find any matching rules */ 2508 IPFW_UNLOCK(chain); 2509 return (EINVAL); 2510 } 2511 msg = log_only ? "ipfw: Entry %d logging count reset.\n" : 2512 "ipfw: Entry %d cleared.\n"; 2513 } 2514 IPFW_UNLOCK(chain); 2515 2516 if (fw_verbose) 2517 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum); 2518 return (0); 2519} 2520 2521/* 2522 * Check validity of the structure before insert. 2523 * Fortunately rules are simple, so this mostly need to check rule sizes. 2524 */ 2525static int 2526check_ipfw_struct(struct ip_fw *rule, int size) 2527{ 2528 int l, cmdlen = 0; 2529 int have_action=0; 2530 ipfw_insn *cmd; 2531 2532 if (size < sizeof(*rule)) { 2533 printf("ipfw: rule too short\n"); 2534 return (EINVAL); 2535 } 2536 /* first, check for valid size */ 2537 l = RULESIZE(rule); 2538 if (l != size) { 2539 printf("ipfw: size mismatch (have %d want %d)\n", size, l); 2540 return (EINVAL); 2541 } 2542 /* 2543 * Now go for the individual checks. Very simple ones, basically only 2544 * instruction sizes. 2545 */ 2546 for (l = rule->cmd_len, cmd = rule->cmd ; 2547 l > 0 ; l -= cmdlen, cmd += cmdlen) { 2548 cmdlen = F_LEN(cmd); 2549 if (cmdlen > l) { 2550 printf("ipfw: opcode %d size truncated\n", 2551 cmd->opcode); 2552 return EINVAL; 2553 } 2554 DEB(printf("ipfw: opcode %d\n", cmd->opcode);) 2555 switch (cmd->opcode) { 2556 case O_PROBE_STATE: 2557 case O_KEEP_STATE: 2558 case O_PROTO: 2559 case O_IP_SRC_ME: 2560 case O_IP_DST_ME: 2561 case O_LAYER2: 2562 case O_IN: 2563 case O_FRAG: 2564 case O_IPOPT: 2565 case O_IPTOS: 2566 case O_IPPRECEDENCE: 2567 case O_IPVER: 2568 case O_TCPWIN: 2569 case O_TCPFLAGS: 2570 case O_TCPOPTS: 2571 case O_ESTAB: 2572 case O_VERREVPATH: 2573 case O_VERSRCREACH: 2574 case O_IPSEC: 2575 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2576 goto bad_size; 2577 break; 2578 2579 case O_UID: 2580 case O_GID: 2581 case O_IP_SRC: 2582 case O_IP_DST: 2583 case O_TCPSEQ: 2584 case O_TCPACK: 2585 case O_PROB: 2586 case O_ICMPTYPE: 2587 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) 2588 goto bad_size; 2589 break; 2590 2591 case O_LIMIT: 2592 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) 2593 goto bad_size; 2594 break; 2595 2596 case O_LOG: 2597 if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) 2598 goto bad_size; 2599 2600 ((ipfw_insn_log *)cmd)->log_left = 2601 ((ipfw_insn_log *)cmd)->max_log; 2602 2603 break; 2604 2605 case O_IP_SRC_MASK: 2606 case O_IP_DST_MASK: 2607 /* only odd command lengths */ 2608 if ( !(cmdlen & 1) || cmdlen > 31) 2609 goto bad_size; 2610 break; 2611 2612 case O_IP_SRC_SET: 2613 case O_IP_DST_SET: 2614 if (cmd->arg1 == 0 || cmd->arg1 > 256) { 2615 printf("ipfw: invalid set size %d\n", 2616 cmd->arg1); 2617 return EINVAL; 2618 } 2619 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 2620 (cmd->arg1+31)/32 ) 2621 goto bad_size; 2622 break; 2623 2624 case O_MACADDR2: 2625 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) 2626 goto bad_size; 2627 break; 2628 2629 case O_NOP: 2630 case O_IPID: 2631 case O_IPTTL: 2632 case O_IPLEN: 2633 if (cmdlen < 1 || cmdlen > 31) 2634 goto bad_size; 2635 break; 2636 2637 case O_MAC_TYPE: 2638 case O_IP_SRCPORT: 2639 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */ 2640 if (cmdlen < 2 || cmdlen > 31) 2641 goto bad_size; 2642 break; 2643 2644 case O_RECV: 2645 case O_XMIT: 2646 case O_VIA: 2647 if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) 2648 goto bad_size; 2649 break; 2650 2651 case O_PIPE: 2652 case O_QUEUE: 2653 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe)) 2654 goto bad_size; 2655 goto check_action; 2656 2657 case O_FORWARD_IP: 2658 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) 2659 goto bad_size; 2660 goto check_action; 2661 2662 case O_FORWARD_MAC: /* XXX not implemented yet */ 2663 case O_CHECK_STATE: 2664 case O_COUNT: 2665 case O_ACCEPT: 2666 case O_DENY: 2667 case O_REJECT: 2668 case O_SKIPTO: 2669 case O_DIVERT: 2670 case O_TEE: 2671 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2672 goto bad_size; 2673check_action: 2674 if (have_action) { 2675 printf("ipfw: opcode %d, multiple actions" 2676 " not allowed\n", 2677 cmd->opcode); 2678 return EINVAL; 2679 } 2680 have_action = 1; 2681 if (l != cmdlen) { 2682 printf("ipfw: opcode %d, action must be" 2683 " last opcode\n", 2684 cmd->opcode); 2685 return EINVAL; 2686 } 2687 break; 2688 default: 2689 printf("ipfw: opcode %d, unknown opcode\n", 2690 cmd->opcode); 2691 return EINVAL; 2692 } 2693 } 2694 if (have_action == 0) { 2695 printf("ipfw: missing action\n"); 2696 return EINVAL; 2697 } 2698 return 0; 2699 2700bad_size: 2701 printf("ipfw: opcode %d size %d wrong\n", 2702 cmd->opcode, cmdlen); 2703 return EINVAL; 2704} 2705 2706/* 2707 * Copy the static and dynamic rules to the supplied buffer 2708 * and return the amount of space actually used. 2709 */ 2710static size_t 2711ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space) 2712{ 2713 char *bp = buf; 2714 char *ep = bp + space; 2715 struct ip_fw *rule; 2716 int i; 2717 2718 /* XXX this can take a long time and locking will block packet flow */ 2719 IPFW_LOCK(chain); 2720 for (rule = chain->rules; rule ; rule = rule->next) { 2721 /* 2722 * Verify the entry fits in the buffer in case the 2723 * rules changed between calculating buffer space and 2724 * now. This would be better done using a generation 2725 * number but should suffice for now. 2726 */ 2727 i = RULESIZE(rule); 2728 if (bp + i <= ep) { 2729 bcopy(rule, bp, i); 2730 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule), 2731 sizeof(set_disable)); 2732 bp += i; 2733 } 2734 } 2735 IPFW_UNLOCK(chain); 2736 if (ipfw_dyn_v) { 2737 ipfw_dyn_rule *p, *last = NULL; 2738 2739 IPFW_DYN_LOCK(); 2740 for (i = 0 ; i < curr_dyn_buckets; i++) 2741 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) { 2742 if (bp + sizeof *p <= ep) { 2743 ipfw_dyn_rule *dst = 2744 (ipfw_dyn_rule *)bp; 2745 bcopy(p, dst, sizeof *p); 2746 bcopy(&(p->rule->rulenum), &(dst->rule), 2747 sizeof(p->rule->rulenum)); 2748 /* 2749 * store a non-null value in "next". 2750 * The userland code will interpret a 2751 * NULL here as a marker 2752 * for the last dynamic rule. 2753 */ 2754 bcopy(&dst, &dst->next, sizeof(dst)); 2755 last = dst; 2756 dst->expire = 2757 TIME_LEQ(dst->expire, time_second) ? 2758 0 : dst->expire - time_second ; 2759 bp += sizeof(ipfw_dyn_rule); 2760 } 2761 } 2762 IPFW_DYN_UNLOCK(); 2763 if (last != NULL) /* mark last dynamic rule */ 2764 bzero(&last->next, sizeof(last)); 2765 } 2766 return (bp - (char *)buf); 2767} 2768 2769 2770/** 2771 * {set|get}sockopt parser. 2772 */ 2773static int 2774ipfw_ctl(struct sockopt *sopt) 2775{ 2776#define RULE_MAXSIZE (256*sizeof(u_int32_t)) 2777 int error, rule_num; 2778 size_t size; 2779 struct ip_fw *buf, *rule; 2780 u_int32_t rulenum[2]; 2781 2782 error = suser(sopt->sopt_td); 2783 if (error) 2784 return (error); 2785 2786 /* 2787 * Disallow modifications in really-really secure mode, but still allow 2788 * the logging counters to be reset. 2789 */ 2790 if (sopt->sopt_name == IP_FW_ADD || 2791 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) { 2792#if __FreeBSD_version >= 500034 2793 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 2794 if (error) 2795 return (error); 2796#else /* FreeBSD 4.x */ 2797 if (securelevel >= 3) 2798 return (EPERM); 2799#endif 2800 } 2801 2802 error = 0; 2803 2804 switch (sopt->sopt_name) { 2805 case IP_FW_GET: 2806 /* 2807 * pass up a copy of the current rules. Static rules 2808 * come first (the last of which has number IPFW_DEFAULT_RULE), 2809 * followed by a possibly empty list of dynamic rule. 2810 * The last dynamic rule has NULL in the "next" field. 2811 * 2812 * Note that the calculated size is used to bound the 2813 * amount of data returned to the user. The rule set may 2814 * change between calculating the size and returning the 2815 * data in which case we'll just return what fits. 2816 */ 2817 size = static_len; /* size of static rules */ 2818 if (ipfw_dyn_v) /* add size of dyn.rules */ 2819 size += (dyn_count * sizeof(ipfw_dyn_rule)); 2820 2821 /* 2822 * XXX todo: if the user passes a short length just to know 2823 * how much room is needed, do not bother filling up the 2824 * buffer, just jump to the sooptcopyout. 2825 */ 2826 buf = malloc(size, M_TEMP, M_WAITOK); 2827 error = sooptcopyout(sopt, buf, 2828 ipfw_getrules(&layer3_chain, buf, size)); 2829 free(buf, M_TEMP); 2830 break; 2831 2832 case IP_FW_FLUSH: 2833 /* 2834 * Normally we cannot release the lock on each iteration. 2835 * We could do it here only because we start from the head all 2836 * the times so there is no risk of missing some entries. 2837 * On the other hand, the risk is that we end up with 2838 * a very inconsistent ruleset, so better keep the lock 2839 * around the whole cycle. 2840 * 2841 * XXX this code can be improved by resetting the head of 2842 * the list to point to the default rule, and then freeing 2843 * the old list without the need for a lock. 2844 */ 2845 2846 IPFW_LOCK(&layer3_chain); 2847 layer3_chain.reap = NULL; 2848 free_chain(&layer3_chain, 0 /* keep default rule */); 2849 rule = layer3_chain.reap, layer3_chain.reap = NULL; 2850 IPFW_UNLOCK(&layer3_chain); 2851 if (layer3_chain.reap != NULL) 2852 reap_rules(rule); 2853 break; 2854 2855 case IP_FW_ADD: 2856 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK); 2857 error = sooptcopyin(sopt, rule, RULE_MAXSIZE, 2858 sizeof(struct ip_fw) ); 2859 if (error == 0) 2860 error = check_ipfw_struct(rule, sopt->sopt_valsize); 2861 if (error == 0) { 2862 error = add_rule(&layer3_chain, rule); 2863 size = RULESIZE(rule); 2864 if (!error && sopt->sopt_dir == SOPT_GET) 2865 error = sooptcopyout(sopt, rule, size); 2866 } 2867 free(rule, M_TEMP); 2868 break; 2869 2870 case IP_FW_DEL: 2871 /* 2872 * IP_FW_DEL is used for deleting single rules or sets, 2873 * and (ab)used to atomically manipulate sets. Argument size 2874 * is used to distinguish between the two: 2875 * sizeof(u_int32_t) 2876 * delete single rule or set of rules, 2877 * or reassign rules (or sets) to a different set. 2878 * 2*sizeof(u_int32_t) 2879 * atomic disable/enable sets. 2880 * first u_int32_t contains sets to be disabled, 2881 * second u_int32_t contains sets to be enabled. 2882 */ 2883 error = sooptcopyin(sopt, rulenum, 2884 2*sizeof(u_int32_t), sizeof(u_int32_t)); 2885 if (error) 2886 break; 2887 size = sopt->sopt_valsize; 2888 if (size == sizeof(u_int32_t)) /* delete or reassign */ 2889 error = del_entry(&layer3_chain, rulenum[0]); 2890 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */ 2891 set_disable = 2892 (set_disable | rulenum[0]) & ~rulenum[1] & 2893 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */ 2894 else 2895 error = EINVAL; 2896 break; 2897 2898 case IP_FW_ZERO: 2899 case IP_FW_RESETLOG: /* argument is an int, the rule number */ 2900 rule_num = 0; 2901 if (sopt->sopt_val != 0) { 2902 error = sooptcopyin(sopt, &rule_num, 2903 sizeof(int), sizeof(int)); 2904 if (error) 2905 break; 2906 } 2907 error = zero_entry(&layer3_chain, rule_num, 2908 sopt->sopt_name == IP_FW_RESETLOG); 2909 break; 2910 2911 default: 2912 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name); 2913 error = EINVAL; 2914 } 2915 2916 return (error); 2917#undef RULE_MAXSIZE 2918} 2919 2920/** 2921 * dummynet needs a reference to the default rule, because rules can be 2922 * deleted while packets hold a reference to them. When this happens, 2923 * dummynet changes the reference to the default rule (it could well be a 2924 * NULL pointer, but this way we do not need to check for the special 2925 * case, plus here he have info on the default behaviour). 2926 */ 2927struct ip_fw *ip_fw_default_rule; 2928 2929/* 2930 * This procedure is only used to handle keepalives. It is invoked 2931 * every dyn_keepalive_period 2932 */ 2933static void 2934ipfw_tick(void * __unused unused) 2935{ 2936 int i; 2937 ipfw_dyn_rule *q; 2938 2939 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0) 2940 goto done; 2941 2942 IPFW_DYN_LOCK(); 2943 for (i = 0 ; i < curr_dyn_buckets ; i++) { 2944 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) { 2945 if (q->dyn_type == O_LIMIT_PARENT) 2946 continue; 2947 if (q->id.proto != IPPROTO_TCP) 2948 continue; 2949 if ( (q->state & BOTH_SYN) != BOTH_SYN) 2950 continue; 2951 if (TIME_LEQ( time_second+dyn_keepalive_interval, 2952 q->expire)) 2953 continue; /* too early */ 2954 if (TIME_LEQ(q->expire, time_second)) 2955 continue; /* too late, rule expired */ 2956 2957 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 2958 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0); 2959 } 2960 } 2961 IPFW_DYN_UNLOCK(); 2962done: 2963 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL); 2964} 2965 2966static int 2967ipfw_init(void) 2968{ 2969 struct ip_fw default_rule; 2970 int error; 2971 2972 layer3_chain.rules = NULL; 2973 IPFW_LOCK_INIT(&layer3_chain); 2974 IPFW_DYN_LOCK_INIT(); 2975 callout_init(&ipfw_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0); 2976 2977 bzero(&default_rule, sizeof default_rule); 2978 2979 default_rule.act_ofs = 0; 2980 default_rule.rulenum = IPFW_DEFAULT_RULE; 2981 default_rule.cmd_len = 1; 2982 default_rule.set = RESVD_SET; 2983 2984 default_rule.cmd[0].len = 1; 2985 default_rule.cmd[0].opcode = 2986#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT 2987 1 ? O_ACCEPT : 2988#endif 2989 O_DENY; 2990 2991 error = add_rule(&layer3_chain, &default_rule); 2992 if (error != 0) { 2993 printf("ipfw2: error %u initializing default rule " 2994 "(support disabled)\n", error); 2995 IPFW_DYN_LOCK_DESTROY(); 2996 IPFW_LOCK_DESTROY(&layer3_chain); 2997 return (error); 2998 } 2999 3000 ip_fw_default_rule = layer3_chain.rules; 3001 printf("ipfw2 initialized, divert %s, " 3002 "rule-based forwarding enabled, default to %s, logging ", 3003#ifdef IPDIVERT 3004 "enabled", 3005#else 3006 "disabled", 3007#endif 3008 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny"); 3009 3010#ifdef IPFIREWALL_VERBOSE 3011 fw_verbose = 1; 3012#endif 3013#ifdef IPFIREWALL_VERBOSE_LIMIT 3014 verbose_limit = IPFIREWALL_VERBOSE_LIMIT; 3015#endif 3016 if (fw_verbose == 0) 3017 printf("disabled\n"); 3018 else if (verbose_limit == 0) 3019 printf("unlimited\n"); 3020 else 3021 printf("limited to %d packets/entry by default\n", 3022 verbose_limit); 3023 3024 ip_fw_chk_ptr = ipfw_chk; 3025 ip_fw_ctl_ptr = ipfw_ctl; 3026 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL); 3027 3028 return (0); 3029} 3030 3031static void 3032ipfw_destroy(void) 3033{ 3034 struct ip_fw *reap; 3035 3036 IPFW_LOCK(&layer3_chain); 3037 callout_stop(&ipfw_timeout); 3038 ip_fw_chk_ptr = NULL; 3039 ip_fw_ctl_ptr = NULL; 3040 layer3_chain.reap = NULL; 3041 free_chain(&layer3_chain, 1 /* kill default rule */); 3042 reap = layer3_chain.reap, layer3_chain.reap = NULL; 3043 IPFW_UNLOCK(&layer3_chain); 3044 if (reap != NULL) 3045 reap_rules(reap); 3046 3047 IPFW_DYN_LOCK_DESTROY(); 3048 IPFW_LOCK_DESTROY(&layer3_chain); 3049 printf("IP firewall unloaded\n"); 3050} 3051 3052static int 3053ipfw_modevent(module_t mod, int type, void *unused) 3054{ 3055 int err = 0; 3056 3057 switch (type) { 3058 case MOD_LOAD: 3059 if (IPFW_LOADED) { 3060 printf("IP firewall already loaded\n"); 3061 err = EEXIST; 3062 } else { 3063 err = ipfw_init(); 3064 } 3065 break; 3066 3067 case MOD_UNLOAD: 3068 ipfw_destroy(); 3069 err = 0; 3070 break; 3071 default: 3072 break; 3073 } 3074 return err; 3075} 3076 3077static moduledata_t ipfwmod = { 3078 "ipfw", 3079 ipfw_modevent, 3080 0 3081}; 3082DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3083MODULE_VERSION(ipfw, 1); 3084#endif /* IPFW2 */ 3085