ip_fw2.c revision 99622
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 99622 2002-07-08 22:46:01Z luigi $ 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#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */ 77 78#include <machine/in_cksum.h> /* XXX for in_cksum */ 79 80static int fw_verbose; 81static int verbose_limit; 82 83#define IPFW_DEFAULT_RULE 65535 84 85/* 86 * list of rules for layer 3 87 */ 88static struct ip_fw *layer3_chain; 89 90MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's"); 91 92static int fw_debug = 1; 93int fw_one_pass = 1; 94static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */ 95 96#ifdef SYSCTL_NODE 97SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); 98SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW, 99 &fw_enable, 0, "Enable ipfw"); 100SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW, 101 &autoinc_step, 0, "Rule number autincrement step"); 102SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW, 103 &fw_one_pass, 0, 104 "Only do a single pass through ipfw when using dummynet(4)"); 105SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW, 106 &fw_debug, 0, "Enable printing of debug ip_fw statements"); 107SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW, 108 &fw_verbose, 0, "Log matches to ipfw rules"); 109SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW, 110 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged"); 111 112/* 113 * Description of dynamic rules. 114 * 115 * Dynamic rules are stored in lists accessed through a hash table 116 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 117 * be modified through the sysctl variable dyn_buckets which is 118 * updated when the table becomes empty. 119 * 120 * XXX currently there is only one list, ipfw_dyn. 121 * 122 * When a packet is received, its address fields are first masked 123 * with the mask defined for the rule, then hashed, then matched 124 * against the entries in the corresponding list. 125 * Dynamic rules can be used for different purposes: 126 * + stateful rules; 127 * + enforcing limits on the number of sessions; 128 * + in-kernel NAT (not implemented yet) 129 * 130 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 131 * measured in seconds and depending on the flags. 132 * 133 * The total number of dynamic rules is stored in dyn_count. 134 * The max number of dynamic rules is dyn_max. When we reach 135 * the maximum number of rules we do not create anymore. This is 136 * done to avoid consuming too much memory, but also too much 137 * time when searching on each packet (ideally, we should try instead 138 * to put a limit on the length of the list on each bucket...). 139 * 140 * Each dynamic rule holds a pointer to the parent ipfw rule so 141 * we know what action to perform. Dynamic rules are removed when 142 * the parent rule is deleted. XXX we should make them survive. 143 * 144 * There are some limitations with dynamic rules -- we do not 145 * obey the 'randomized match', and we do not do multiple 146 * passes through the firewall. XXX check the latter!!! 147 */ 148static ipfw_dyn_rule **ipfw_dyn_v = NULL; 149static u_int32_t dyn_buckets = 256; /* must be power of 2 */ 150static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */ 151 152/* 153 * Timeouts for various events in handing dynamic rules. 154 */ 155static u_int32_t dyn_ack_lifetime = 300; 156static u_int32_t dyn_syn_lifetime = 20; 157static u_int32_t dyn_fin_lifetime = 1; 158static u_int32_t dyn_rst_lifetime = 1; 159static u_int32_t dyn_udp_lifetime = 10; 160static u_int32_t dyn_short_lifetime = 5; 161 162/* 163 * After reaching 0, dynamic rules are considered still valid for 164 * an additional grace time, unless there is lack of resources. 165 * XXX not implemented yet. 166 */ 167static u_int32_t dyn_grace_time = 10; 168 169static u_int32_t static_count; /* # of static rules */ 170static u_int32_t static_len; /* size in bytes of static rules */ 171static u_int32_t dyn_count; /* # of dynamic rules */ 172static u_int32_t dyn_max = 1000; /* max # of dynamic rules */ 173 174SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW, 175 &dyn_buckets, 0, "Number of dyn. buckets"); 176SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD, 177 &curr_dyn_buckets, 0, "Current Number of dyn. buckets"); 178SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD, 179 &dyn_count, 0, "Number of dyn. rules"); 180SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW, 181 &dyn_max, 0, "Max number of dyn. rules"); 182SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD, 183 &static_count, 0, "Number of static rules"); 184SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW, 185 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks"); 186SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW, 187 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn"); 188SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW, 189 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin"); 190SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW, 191 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst"); 192SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW, 193 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP"); 194SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW, 195 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations"); 196SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_grace_time, CTLFLAG_RD, 197 &dyn_grace_time, 0, "Grace time for dyn. rules"); 198 199#endif /* SYSCTL_NODE */ 200 201 202static ip_fw_chk_t ipfw_chk; 203 204ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL; /* hook into dummynet */ 205 206/* 207 * This macro maps an ip pointer into a layer3 header pointer of type T 208 */ 209#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl)) 210 211static __inline int 212icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd) 213{ 214 int type = L3HDR(struct icmp,ip)->icmp_type; 215 216 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) ); 217} 218 219#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \ 220 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) ) 221 222static int 223is_icmp_query(struct ip *ip) 224{ 225 int type = L3HDR(struct icmp, ip)->icmp_type; 226 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) ); 227} 228#undef TT 229 230/* 231 * The following checks use two arrays of 8 or 16 bits to store the 232 * bits that we want set or clear, respectively. They are in the 233 * low and high half of cmd->arg1 or cmd->d[0]. 234 * 235 * We scan options and store the bits we find set. We succeed if 236 * 237 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear 238 * 239 * The code is sometimes optimized not to store additional variables. 240 */ 241 242static int 243flags_match(ipfw_insn *cmd, u_int8_t bits) 244{ 245 u_char want_clear; 246 bits = ~bits; 247 248 if ( ((cmd->arg1 & 0xff) & bits) != 0) 249 return 0; /* some bits we want set were clear */ 250 want_clear = (cmd->arg1 >> 8) & 0xff; 251 if ( (want_clear & bits) != want_clear) 252 return 0; /* some bits we want clear were set */ 253 return 1; 254} 255 256static int 257ipopts_match(struct ip *ip, ipfw_insn *cmd) 258{ 259 int optlen, bits = 0; 260 u_char *cp = (u_char *)(ip + 1); 261 int x = (ip->ip_hl << 2) - sizeof (struct ip); 262 263 for (; x > 0; x -= optlen, cp += optlen) { 264 int opt = cp[IPOPT_OPTVAL]; 265 266 if (opt == IPOPT_EOL) 267 break; 268 if (opt == IPOPT_NOP) 269 optlen = 1; 270 else { 271 optlen = cp[IPOPT_OLEN]; 272 if (optlen <= 0 || optlen > x) 273 return 0; /* invalid or truncated */ 274 } 275 switch (opt) { 276 277 default: 278 break; 279 280 case IPOPT_LSRR: 281 bits |= IP_FW_IPOPT_LSRR; 282 break; 283 284 case IPOPT_SSRR: 285 bits |= IP_FW_IPOPT_SSRR; 286 break; 287 288 case IPOPT_RR: 289 bits |= IP_FW_IPOPT_RR; 290 break; 291 292 case IPOPT_TS: 293 bits |= IP_FW_IPOPT_TS; 294 break; 295 } 296 } 297 return (flags_match(cmd, bits)); 298} 299 300static int 301tcpopts_match(struct ip *ip, ipfw_insn *cmd) 302{ 303 int optlen, bits = 0; 304 struct tcphdr *tcp = L3HDR(struct tcphdr,ip); 305 u_char *cp = (u_char *)(tcp + 1); 306 int x = (tcp->th_off << 2) - sizeof(struct tcphdr); 307 308 for (; x > 0; x -= optlen, cp += optlen) { 309 int opt = cp[0]; 310 if (opt == TCPOPT_EOL) 311 break; 312 if (opt == TCPOPT_NOP) 313 optlen = 1; 314 else { 315 optlen = cp[1]; 316 if (optlen <= 0) 317 break; 318 } 319 320 switch (opt) { 321 322 default: 323 break; 324 325 case TCPOPT_MAXSEG: 326 bits |= IP_FW_TCPOPT_MSS; 327 break; 328 329 case TCPOPT_WINDOW: 330 bits |= IP_FW_TCPOPT_WINDOW; 331 break; 332 333 case TCPOPT_SACK_PERMITTED: 334 case TCPOPT_SACK: 335 bits |= IP_FW_TCPOPT_SACK; 336 break; 337 338 case TCPOPT_TIMESTAMP: 339 bits |= IP_FW_TCPOPT_TS; 340 break; 341 342 case TCPOPT_CC: 343 case TCPOPT_CCNEW: 344 case TCPOPT_CCECHO: 345 bits |= IP_FW_TCPOPT_CC; 346 break; 347 } 348 } 349 return (flags_match(cmd, bits)); 350} 351 352static int 353iface_match(struct ifnet *ifp, ipfw_insn_if *cmd) 354{ 355 if (ifp == NULL) /* no iface with this packet, match fails */ 356 return 0; 357 /* Check by name or by IP address */ 358 if (cmd->name[0] != '\0') { /* match by name */ 359 /* Check unit number (-1 is wildcard) */ 360 if (cmd->p.unit != -1 && cmd->p.unit != ifp->if_unit) 361 return(0); 362 /* Check name */ 363 if (!strncmp(ifp->if_name, cmd->name, IFNAMSIZ)) 364 return(1); 365 } else { 366 struct ifaddr *ia; 367 368 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) { 369 if (ia->ifa_addr == NULL) 370 continue; 371 if (ia->ifa_addr->sa_family != AF_INET) 372 continue; 373 if (cmd->p.ip.s_addr == ((struct sockaddr_in *) 374 (ia->ifa_addr))->sin_addr.s_addr) 375 return(1); /* match */ 376 } 377 } 378 return(0); /* no match, fail ... */ 379} 380 381static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */ 382 383#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0 384#define SNP(buf) buf, sizeof(buf) 385/* 386 * We enter here when we have a rule with O_LOG. 387 * XXX this function alone takes about 2Kbytes of code! 388 */ 389static void 390ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh, 391 struct mbuf *m, struct ifnet *oif) 392{ 393 char *action; 394 int limit_reached = 0; 395 char action2[40], proto[48], fragment[28]; 396 397 fragment[0] = '\0'; 398 proto[0] = '\0'; 399 400 if (f == NULL) { /* bogus pkt */ 401 if (verbose_limit != 0 && norule_counter >= verbose_limit) 402 return; 403 norule_counter++; 404 if (norule_counter == verbose_limit) 405 limit_reached = verbose_limit; 406 action = "Refuse"; 407 } else { /* O_LOG is the first action, find the real one */ 408 ipfw_insn *cmd = ACTION_PTR(f); 409 ipfw_insn_log *l = (ipfw_insn_log *)cmd; 410 411 if (l->max_log != 0 && l->log_left == 0) 412 return; 413 l->log_left--; 414 if (l->log_left == 0) 415 limit_reached = l->max_log; 416 cmd += F_LEN(cmd); /* point to first action */ 417 if (cmd->opcode == O_PROB) 418 cmd += F_LEN(cmd); 419 420 action = action2; 421 switch (cmd->opcode) { 422 case O_DENY: 423 action = "Deny"; 424 break; 425 426 case O_REJECT: 427 if (cmd->arg1==ICMP_REJECT_RST) 428 action = "Reset"; 429 else if (cmd->arg1==ICMP_UNREACH_HOST) 430 action = "Reject"; 431 else 432 snprintf(SNPARGS(action2, 0), "Unreach %d", 433 cmd->arg1); 434 break; 435 436 case O_ACCEPT: 437 action = "Accept"; 438 break; 439 case O_COUNT: 440 action = "Count"; 441 break; 442 case O_DIVERT: 443 snprintf(SNPARGS(action2, 0), "Divert %d", 444 cmd->arg1); 445 break; 446 case O_TEE: 447 snprintf(SNPARGS(action2, 0), "Tee %d", 448 cmd->arg1); 449 break; 450 case O_SKIPTO: 451 snprintf(SNPARGS(action2, 0), "SkipTo %d", 452 cmd->arg1); 453 break; 454 case O_PIPE: 455 snprintf(SNPARGS(action2, 0), "Pipe %d", 456 cmd->arg1); 457 break; 458 case O_QUEUE: 459 snprintf(SNPARGS(action2, 0), "Queue %d", 460 cmd->arg1); 461 break; 462 case O_FORWARD_IP: { 463 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd; 464 465 if (sa->sa.sin_port) 466 snprintf(SNPARGS(action2, 0), 467 "Forward to %s:%d", 468 inet_ntoa(sa->sa.sin_addr), 469 sa->sa.sin_port); 470 else 471 snprintf(SNPARGS(action2, 0), "Forward to %s", 472 inet_ntoa(sa->sa.sin_addr)); 473 } 474 break; 475 default: 476 action = "UNKNOWN"; 477 break; 478 } 479 } 480 481 if (hlen == 0) { /* non-ip */ 482 snprintf(SNPARGS(proto, 0), "MAC"); 483 } else { 484 struct ip *ip = mtod(m, struct ip *); 485 /* these three are all aliases to the same thing */ 486 struct icmp *const icmp = L3HDR(struct icmp, ip); 487 struct tcphdr *const tcp = (struct tcphdr *)icmp; 488 struct udphdr *const udp = (struct udphdr *)icmp; 489 490 int ip_off, offset, ip_len; 491 492 int len; 493 494 if (eh != NULL) { /* layer 2 packets are as on the wire */ 495 ip_off = ntohs(ip->ip_off); 496 ip_len = ntohs(ip->ip_len); 497 } else { 498 ip_off = ip->ip_off; 499 ip_len = ip->ip_len; 500 } 501 offset = ip_off & IP_OFFMASK; 502 switch (ip->ip_p) { 503 case IPPROTO_TCP: 504 len = snprintf(SNPARGS(proto, 0), "TCP %s", 505 inet_ntoa(ip->ip_src)); 506 if (offset == 0) 507 len += snprintf(SNPARGS(proto, len), ":%d ", 508 ntohs(tcp->th_sport)); 509 else 510 len += snprintf(SNPARGS(proto, len), " "); 511 len += snprintf(SNPARGS(proto, len), "%s", 512 inet_ntoa(ip->ip_dst)); 513 if (offset == 0) 514 snprintf(SNPARGS(proto, len), ":%d", 515 ntohs(tcp->th_dport)); 516 break; 517 518 case IPPROTO_UDP: 519 len = snprintf(SNPARGS(proto, 0), "UDP %s", 520 inet_ntoa(ip->ip_src)); 521 if (offset == 0) 522 len += snprintf(SNPARGS(proto, len), ":%d ", 523 ntohs(udp->uh_sport)); 524 else 525 len += snprintf(SNPARGS(proto, len), " "); 526 len += snprintf(SNPARGS(proto, len), "%s", 527 inet_ntoa(ip->ip_dst)); 528 if (offset == 0) 529 snprintf(SNPARGS(proto, len), ":%d", 530 ntohs(udp->uh_dport)); 531 break; 532 533 case IPPROTO_ICMP: 534 if (offset == 0) 535 len = snprintf(SNPARGS(proto, 0), 536 "ICMP:%u.%u ", 537 icmp->icmp_type, icmp->icmp_code); 538 else 539 len = snprintf(SNPARGS(proto, 0), "ICMP "); 540 len += snprintf(SNPARGS(proto, len), "%s", 541 inet_ntoa(ip->ip_src)); 542 snprintf(SNPARGS(proto, len), " %s", 543 inet_ntoa(ip->ip_dst)); 544 break; 545 546 default: 547 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p, 548 inet_ntoa(ip->ip_src)); 549 snprintf(SNPARGS(proto, len), " %s", 550 inet_ntoa(ip->ip_dst)); 551 break; 552 } 553 554 if (ip_off & (IP_MF | IP_OFFMASK)) 555 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)", 556 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2), 557 offset << 3, 558 (ip_off & IP_MF) ? "+" : ""); 559 } 560 if (oif || m->m_pkthdr.rcvif) 561 log(LOG_SECURITY | LOG_INFO, 562 "ipfw: %d %s %s %s via %s%d%s\n", 563 f ? f->rulenum : -1, 564 action, proto, oif ? "out" : "in", 565 oif ? oif->if_name : m->m_pkthdr.rcvif->if_name, 566 oif ? oif->if_unit : m->m_pkthdr.rcvif->if_unit, 567 fragment); 568 else 569 log(LOG_SECURITY | LOG_INFO, 570 "ipfw: %d %s %s [no if info]%s\n", 571 f ? f->rulenum : -1, 572 action, proto, fragment); 573 if (limit_reached) 574 log(LOG_SECURITY | LOG_NOTICE, 575 "ipfw: limit %d reached on entry %d\n", 576 limit_reached, f ? f->rulenum : -1); 577} 578 579/* 580 * IMPORTANT: the hash function for dynamic rules must be commutative 581 * in source and destination (ip,port), because rules are bidirectional 582 * and we want to find both in the same bucket. 583 */ 584static __inline int 585hash_packet(struct ipfw_flow_id *id) 586{ 587 u_int32_t i; 588 589 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 590 i &= (curr_dyn_buckets - 1); 591 return i; 592} 593 594/** 595 * unlink a dynamic rule from a chain. prev is a pointer to 596 * the previous one, q is a pointer to the rule to delete, 597 * head is a pointer to the head of the queue. 598 * Modifies q and potentially also head. 599 */ 600#define UNLINK_DYN_RULE(prev, head, q) { \ 601 ipfw_dyn_rule *old_q = q; \ 602 \ 603 /* remove a refcount to the parent */ \ 604 if (q->dyn_type == O_LIMIT) \ 605 q->parent->count--; \ 606 DEB(printf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \ 607 (q->id.src_ip), (q->id.src_port), \ 608 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \ 609 if (prev != NULL) \ 610 prev->next = q = q->next; \ 611 else \ 612 head = q = q->next; \ 613 dyn_count--; \ 614 free(old_q, M_IPFW); } 615 616#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 617 618/** 619 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 620 * 621 * If keep_me == NULL, rules are deleted even if not expired, 622 * otherwise only expired rules are removed. 623 * 624 * The value of the second parameter is also used to point to identify 625 * a rule we absolutely do not want to remove (e.g. because we are 626 * holding a reference to it -- this is the case with O_LIMIT_PARENT 627 * rules). The pointer is only used for comparison, so any non-null 628 * value will do. 629 */ 630static void 631remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 632{ 633 static u_int32_t last_remove = 0; 634 635#define FORCE (keep_me == NULL) 636 637 ipfw_dyn_rule *prev, *q; 638 int i, pass = 0, max_pass = 0; 639 640 if (ipfw_dyn_v == NULL || dyn_count == 0) 641 return; 642 /* do not expire more than once per second, it is useless */ 643 if (!FORCE && last_remove == time_second) 644 return; 645 last_remove = time_second; 646 647 /* 648 * because O_LIMIT refer to parent rules, during the first pass only 649 * remove child and mark any pending LIMIT_PARENT, and remove 650 * them in a second pass. 651 */ 652next_pass: 653 for (i = 0 ; i < curr_dyn_buckets ; i++) { 654 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) { 655 /* 656 * Logic can become complex here, so we split tests. 657 */ 658 if (q == keep_me) 659 goto next; 660 if (rule != NULL && rule != q->rule) 661 goto next; /* not the one we are looking for */ 662 if (q->dyn_type == O_LIMIT_PARENT) { 663 /* 664 * handle parent in the second pass, 665 * record we need one. 666 */ 667 max_pass = 1; 668 if (pass == 0) 669 goto next; 670 if (FORCE && q->count != 0 ) { 671 /* XXX should not happen! */ 672 printf( "OUCH! cannot remove rule," 673 " count %d\n", q->count); 674 } 675 } else { 676 if (!FORCE && 677 !TIME_LEQ( q->expire, time_second )) 678 goto next; 679 } 680 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 681 continue; 682next: 683 prev=q; 684 q=q->next; 685 } 686 } 687 if (pass++ < max_pass) 688 goto next_pass; 689} 690 691 692/** 693 * lookup a dynamic rule. 694 */ 695static ipfw_dyn_rule * 696lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction) 697{ 698 /* 699 * stateful ipfw extensions. 700 * Lookup into dynamic session queue 701 */ 702#define MATCH_REVERSE 0 703#define MATCH_FORWARD 1 704#define MATCH_NONE 2 705#define MATCH_UNKNOWN 3 706 int i, dir = MATCH_NONE; 707 ipfw_dyn_rule *prev, *q=NULL; 708 709 if (ipfw_dyn_v == NULL) 710 goto done; /* not found */ 711 i = hash_packet( pkt ); 712 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) { 713 if (q->dyn_type == O_LIMIT_PARENT) 714 goto next; 715 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */ 716 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 717 continue; 718 } 719 if ( pkt->proto == q->id.proto) { 720 if (pkt->src_ip == q->id.src_ip && 721 pkt->dst_ip == q->id.dst_ip && 722 pkt->src_port == q->id.src_port && 723 pkt->dst_port == q->id.dst_port ) { 724 dir = MATCH_FORWARD; 725 break; 726 } 727 if (pkt->src_ip == q->id.dst_ip && 728 pkt->dst_ip == q->id.src_ip && 729 pkt->src_port == q->id.dst_port && 730 pkt->dst_port == q->id.src_port ) { 731 dir = MATCH_REVERSE; 732 break; 733 } 734 } 735next: 736 prev = q; 737 q = q->next; 738 } 739 if (q == NULL) 740 goto done; /* q = NULL, not found */ 741 742 if ( prev != NULL) { /* found and not in front */ 743 prev->next = q->next; 744 q->next = ipfw_dyn_v[i]; 745 ipfw_dyn_v[i] = q; 746 } 747 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 748 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST); 749 750#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 751#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 752 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8); 753 switch (q->state) { 754 case TH_SYN: /* opening */ 755 q->expire = time_second + dyn_syn_lifetime; 756 break; 757 case BOTH_SYN: /* move to established */ 758 q->expire = time_second + dyn_ack_lifetime; 759 break; 760 case BOTH_SYN | TH_FIN : /* one side tries to close */ 761 case BOTH_SYN | (TH_FIN << 8) : 762 q->expire = time_second + dyn_ack_lifetime; 763 break; 764 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 765 q->expire = time_second + dyn_fin_lifetime; 766 break; 767 default: 768#if 0 769 /* 770 * reset or some invalid combination, but can also 771 * occur if we use keep-state the wrong way. 772 */ 773 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 774 printf("invalid state: 0x%x\n", q->state); 775#endif 776 q->expire = time_second + dyn_rst_lifetime; 777 break; 778 } 779 } else if (pkt->proto == IPPROTO_UDP) { 780 q->expire = time_second + dyn_udp_lifetime; 781 } else { 782 /* other protocols */ 783 q->expire = time_second + dyn_short_lifetime; 784 } 785done: 786 if (match_direction) 787 *match_direction = dir; 788 return q; 789} 790 791static void 792realloc_dynamic_table(void) 793{ 794 /* try reallocation, make sure we have a power of 2 */ 795 796 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */ 797 dyn_buckets = curr_dyn_buckets; /* reset */ 798 return; 799 } 800 curr_dyn_buckets = dyn_buckets; 801 if (ipfw_dyn_v != NULL) 802 free(ipfw_dyn_v, M_IPFW); 803 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 804 M_IPFW, M_DONTWAIT | M_ZERO); 805} 806 807/** 808 * Install state of type 'type' for a dynamic session. 809 * The hash table contains two type of rules: 810 * - regular rules (O_KEEP_STATE) 811 * - rules for sessions with limited number of sess per user 812 * (O_LIMIT). When they are created, the parent is 813 * increased by 1, and decreased on delete. In this case, 814 * the third parameter is the parent rule and not the chain. 815 * - "parent" rules for the above (O_LIMIT_PARENT). 816 */ 817static ipfw_dyn_rule * 818add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule) 819{ 820 ipfw_dyn_rule *r; 821 int i; 822 823 if (ipfw_dyn_v == NULL || 824 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) { 825 realloc_dynamic_table(); 826 if (ipfw_dyn_v == NULL) 827 return NULL; /* failed ! */ 828 } 829 i = hash_packet(id); 830 831 r = malloc(sizeof *r, M_IPFW, M_DONTWAIT | M_ZERO); 832 if (r == NULL) { 833 printf ("sorry cannot allocate state\n"); 834 return NULL; 835 } 836 837 /* increase refcount on parent, and set pointer */ 838 if (dyn_type == O_LIMIT) { 839 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 840 if ( parent->dyn_type != O_LIMIT_PARENT) 841 panic("invalid parent"); 842 parent->count++; 843 r->parent = parent; 844 rule = parent->rule; 845 } 846 847 r->id = *id; 848 r->expire = time_second + dyn_syn_lifetime; 849 r->rule = rule; 850 r->dyn_type = dyn_type; 851 r->pcnt = r->bcnt = 0; 852 r->count = 0; 853 854 r->bucket = i; 855 r->next = ipfw_dyn_v[i]; 856 ipfw_dyn_v[i] = r; 857 dyn_count++; 858 DEB(printf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n", 859 dyn_type, 860 (r->id.src_ip), (r->id.src_port), 861 (r->id.dst_ip), (r->id.dst_port), 862 dyn_count ); ) 863 return r; 864} 865 866/** 867 * lookup dynamic parent rule using pkt and rule as search keys. 868 * If the lookup fails, then install one. 869 */ 870static ipfw_dyn_rule * 871lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 872{ 873 ipfw_dyn_rule *q; 874 int i; 875 876 if (ipfw_dyn_v) { 877 i = hash_packet( pkt ); 878 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next) 879 if (q->dyn_type == O_LIMIT_PARENT && 880 rule== q->rule && 881 pkt->proto == q->id.proto && 882 pkt->src_ip == q->id.src_ip && 883 pkt->dst_ip == q->id.dst_ip && 884 pkt->src_port == q->id.src_port && 885 pkt->dst_port == q->id.dst_port) { 886 q->expire = time_second + dyn_short_lifetime; 887 DEB(printf("lookup_dyn_parent found 0x%p\n",q);) 888 return q; 889 } 890 } 891 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 892} 893 894/** 895 * Install dynamic state for rule type cmd->o.opcode 896 * 897 * Returns 1 (failure) if state is not installed because of errors or because 898 * session limitations are enforced. 899 */ 900static int 901install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 902 struct ip_fw_args *args) 903{ 904 static int last_log; 905 906 ipfw_dyn_rule *q; 907 908 DEB(printf("-- install state type %d 0x%08x %u -> 0x%08x %u\n", 909 cmd->o.opcode, 910 (args->f_id.src_ip), (args->f_id.src_port), 911 (args->f_id.dst_ip), (args->f_id.dst_port) );) 912 913 q = lookup_dyn_rule(&args->f_id, NULL); 914 915 if (q != NULL) { /* should never occur */ 916 if (last_log != time_second) { 917 last_log = time_second; 918 printf(" install_state: entry already present, done\n"); 919 } 920 return 0; 921 } 922 923 if (dyn_count >= dyn_max) 924 /* 925 * Run out of slots, try to remove any expired rule. 926 */ 927 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1); 928 929 if (dyn_count >= dyn_max) { 930 if (last_log != time_second) { 931 last_log = time_second; 932 printf("install_state: Too many dynamic rules\n"); 933 } 934 return 1; /* cannot install, notify caller */ 935 } 936 937 switch (cmd->o.opcode) { 938 case O_KEEP_STATE: /* bidir rule */ 939 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule); 940 break; 941 942 case O_LIMIT: /* limit number of sessions */ 943 { 944 u_int16_t limit_mask = cmd->limit_mask; 945 struct ipfw_flow_id id; 946 ipfw_dyn_rule *parent; 947 948 DEB(printf("installing dyn-limit rule %d\n", cmd->conn_limit);) 949 950 id.dst_ip = id.src_ip = 0; 951 id.dst_port = id.src_port = 0; 952 id.proto = args->f_id.proto; 953 954 if (limit_mask & DYN_SRC_ADDR) 955 id.src_ip = args->f_id.src_ip; 956 if (limit_mask & DYN_DST_ADDR) 957 id.dst_ip = args->f_id.dst_ip; 958 if (limit_mask & DYN_SRC_PORT) 959 id.src_port = args->f_id.src_port; 960 if (limit_mask & DYN_DST_PORT) 961 id.dst_port = args->f_id.dst_port; 962 parent = lookup_dyn_parent(&id, rule); 963 if (parent == NULL) { 964 printf("add parent failed\n"); 965 return 1; 966 } 967 if (parent->count >= cmd->conn_limit) { 968 /* 969 * See if we can remove some expired rule. 970 */ 971 remove_dyn_rule(rule, parent); 972 if (parent->count >= cmd->conn_limit) { 973 if (fw_verbose && last_log != time_second) { 974 last_log = time_second; 975 printf( 976 "drop session, too many entries\n"); 977 } 978 return 1; 979 } 980 } 981 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent); 982 } 983 break; 984 default: 985 printf("unknown dynamic rule type %u\n", cmd->o.opcode); 986 return 1; 987 } 988 lookup_dyn_rule(&args->f_id, NULL); /* XXX just set the lifetime */ 989 return 0; 990} 991 992#if 0 993static void 994send_pkt(struct ip_fw_args *args, u_int32_t seq, u_int32_t ack, int flags) 995{ 996 struct mbuf *m; 997 struct ip *ip = mtod(args->m, struct ip *); 998 struct tcphdr *tcp; 999 struct route sro; /* fake route */ 1000 1001 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1002 if (m == 0) 1003 return; 1004 m->m_pkthdr.rcvif = (struct ifnet *)0; 1005 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); 1006 m->m_data += max_linkhdr; 1007 1008 ip->ip_v = 4; 1009 ip->ip_hl = 5; 1010 ip->ip_tos = 0; 1011 ip->ip_len = 0; /* set later */ 1012#ifdef RANDOM_IP_ID 1013 ip->ip_id = ip_randomid(); 1014#else 1015 ip->ip_id = htons(ip_id++); 1016#endif 1017 ip->ip_off = 0; 1018 ip->ip_ttl = 0; /* set later */ 1019 ip->ip_p = IPPROTO_TCP; 1020 ip->ip_sum = 0; 1021 ip->ip_src.s_addr = args->f_id.dst_ip; 1022 ip->ip_dst.s_addr = args->f_id.src_ip; 1023 1024 tcp = L3HDR(struct tcphdr, ip); 1025 tcp->th_sport = htons(args->f_id.dst_port); /* swap ports */ 1026 tcp->th_dport = htons(args->f_id.src_port); 1027 tcp->th_off = 4; 1028 tcp->th_x2 = 0; 1029 tcp->th_win = 0; 1030 tcp->th_sum = 0; 1031 tcp->th_urp = 0; 1032 if (flags & TH_ACK) { 1033 tcp->th_seq = htonl(ack); 1034 tcp->th_ack = htonl(0); 1035 tcp->th_flags = TH_RST; 1036 } else { 1037 if (flags & TH_SYN) 1038 seq++; 1039 tcp->th_seq = htonl(0); 1040 tcp->th_ack = htonl(seq); 1041 tcp->th_flags = TH_RST | TH_ACK; 1042 } 1043 /* compute TCP checksum... */ 1044 tcp->th_sum = in_cksum(m, m->m_pkthdr.len); 1045 ip->ip_ttl = ip_defttl; 1046 ip->ip_len = m->m_pkthdr.len; 1047 bzero (&sro, sizeof (sro)); 1048 ip_rtaddr(ip->ip_dst, &sro); 1049 ip_output(m, NULL, &sro, 0, NULL); 1050 if (sro.ro_rt) 1051 RTFREE(sro.ro_rt); 1052} 1053#endif 1054 1055/* 1056 * sends a reject message, consuming the mbuf passed as an argument. 1057 */ 1058static void 1059send_reject(struct ip_fw_args *args, int code, int offset, int ip_len) 1060{ 1061 if (code != ICMP_REJECT_RST) /* Send an ICMP unreach */ 1062 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); 1063 else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) { 1064#if 0 1065 struct tcphdr *const tcp = 1066 L3HDR(struct tcphdr, mtod(args->m, struct ip *)); 1067 if ( (tcp->th_flags & TH_RST) == 0) 1068 send_pkt(args, tcp->th_seq, tcp->th_ack, tcp->th_flags); 1069 m_freem(args->m); 1070#else 1071 /* XXX warning, this code writes into the mbuf */ 1072 struct ip *ip = mtod(args->m, struct ip *); 1073 struct tcphdr *const tcp = L3HDR(struct tcphdr, ip); 1074 struct tcpiphdr ti, *const tip = (struct tcpiphdr *) ip; 1075 int hlen = ip->ip_hl << 2; 1076 1077 if (tcp->th_flags & TH_RST) { 1078 m_freem(args->m); /* free the mbuf */ 1079 args->m = NULL; 1080 return; 1081 } 1082 ti.ti_i = *((struct ipovly *) ip); 1083 ti.ti_t = *tcp; 1084 bcopy(&ti, ip, sizeof(ti)); 1085 tip->ti_seq = ntohl(tip->ti_seq); 1086 tip->ti_ack = ntohl(tip->ti_ack); 1087 tip->ti_len = ip_len - hlen - (tip->ti_off << 2); 1088 if (tcp->th_flags & TH_ACK) { 1089 tcp_respond(NULL, (void *)ip, tcp, args->m, 1090 0, tcp->th_ack, TH_RST); 1091 } else { 1092 if (tcp->th_flags & TH_SYN) 1093 tip->ti_len++; 1094 tcp_respond(NULL, (void *)ip, tcp, args->m, 1095 tip->ti_seq + tip->ti_len, 0, TH_RST|TH_ACK); 1096 } 1097#endif 1098 } else 1099 m_freem(args->m); 1100 args->m = NULL; 1101} 1102 1103/** 1104 * 1105 * Given an ip_fw *, lookup_next_rule will return a pointer 1106 * to the next rule, which can be either the jump 1107 * target (for skipto instructions) or the next one in the list (in 1108 * all other cases including a missing jump target). 1109 * The result is also written in the "next_rule" field of the rule. 1110 * Backward jumps are not allowed, so start looking from the next 1111 * rule... 1112 * 1113 * This never returns NULL -- in case we do not have an exact match, 1114 * the next rule is returned. When the ruleset is changed, 1115 * pointers are flushed so we are always correct. 1116 */ 1117 1118static struct ip_fw * 1119lookup_next_rule(struct ip_fw *me) 1120{ 1121 struct ip_fw *rule = NULL; 1122 ipfw_insn *cmd; 1123 1124 /* look for action, in case it is a skipto */ 1125 cmd = ACTION_PTR(me); 1126 if ( cmd->opcode == O_SKIPTO ) 1127 for (rule = me->next; rule ; rule = rule->next) 1128 if (rule->rulenum >= cmd->arg1) 1129 break; 1130 if (rule == NULL) /* failure or not a skipto */ 1131 rule = me->next; 1132 me->next_rule = rule; 1133 return rule; 1134} 1135 1136/* 1137 * The main check routine for the firewall. 1138 * 1139 * All arguments are in args so we can modify them and return them 1140 * back to the caller. 1141 * 1142 * Parameters: 1143 * 1144 * args->m (in/out) The packet; we set to NULL when/if we nuke it. 1145 * Starts with the IP header. 1146 * args->eh (in) Mac header if present, or NULL for layer3 packet. 1147 * args->oif Outgoing interface, or NULL if packet is incoming. 1148 * The incoming interface is in the mbuf. (in) 1149 * args->divert_rule (in/out) 1150 * Skip up to the first rule past this rule number; 1151 * upon return, non-zero port number for divert or tee. 1152 * 1153 * args->rule Pointer to the last matching rule (in/out) 1154 * args->next_hop Socket we are forwarding to (out). 1155 * args->f_id Addresses grabbed from the packet (out) 1156 * 1157 * Return value: 1158 * 1159 * IP_FW_PORT_DENY_FLAG the packet must be dropped. 1160 * 0 The packet is to be accepted and routed normally OR 1161 * the packet was denied/rejected and has been dropped; 1162 * in the latter case, *m is equal to NULL upon return. 1163 * port Divert the packet to port, with these caveats: 1164 * 1165 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead 1166 * of diverting it (ie, 'ipfw tee'). 1167 * 1168 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower 1169 * 16 bits as a dummynet pipe number instead of diverting 1170 */ 1171 1172static int 1173ipfw_chk(struct ip_fw_args *args) 1174{ 1175 /* 1176 * Local variables hold state during the processing of a packet. 1177 * 1178 * IMPORTANT NOTE: to speed up the processing of rules, there 1179 * are some assumption on the values of the variables, which 1180 * are documented here. Should you change them, please check 1181 * the implementation of the various instructions to make sure 1182 * that they still work. 1183 */ 1184 /* 1185 * args->eh The MAC header. It is non-null for a layer2 1186 * packet, it is NULL for a layer-3 packet. 1187 * 1188 * m | args->m Pointer to the mbuf, as received from the caller. 1189 * It may change if ipfw_chk() does an m_pullup, or if it 1190 * consumes the packet because it calls send_reject(). 1191 * XXX This has to change, so that ipfw_chk() never modifies 1192 * or consumes the buffer. 1193 * ip is simply an alias of the value of m, and it is kept 1194 * in sync with it (the packet is supposed to start with 1195 * the ip header). 1196 */ 1197 struct mbuf *m = args->m; 1198 struct ip *ip = mtod(m, struct ip *); 1199 1200 /* 1201 * oif | args->oif If NULL, ipfw_chk has been called on the 1202 * inbound path (ether_input, bdg_forward, ip_input). 1203 * If non-NULL, ipfw_chk has been called on the outbound path 1204 * (ether_output, ip_output). 1205 */ 1206 struct ifnet *oif = args->oif; 1207 1208 struct ip_fw *f = NULL; /* matching rule */ 1209 int retval = 0; 1210 1211 /* 1212 * hlen The length of the IPv4 header. 1213 * hlen >0 means we have an IPv4 packet. 1214 */ 1215 u_int hlen = 0; /* hlen >0 means we have an IP pkt */ 1216 1217 /* 1218 * offset The offset of a fragment. offset != 0 means that 1219 * we have a fragment at this offset of an IPv4 packet. 1220 * offset == 0 means that (if this is an IPv4 packet) 1221 * this is the first or only fragment. 1222 */ 1223 u_short offset = 0; 1224 1225 /* 1226 * Local copies of addresses. They are only valid if we have 1227 * an IP packet. 1228 * 1229 * proto The protocol. Set to 0 for non-ip packets, 1230 * or to the protocol read from the packet otherwise. 1231 * proto != 0 means that we have an IPv4 packet. 1232 * 1233 * src_port, dst_port port numbers, in HOST format. Only 1234 * valid for TCP and UDP packets. 1235 * 1236 * src_ip, dst_ip ip addresses, in NETWORK format. 1237 * Only valid for IPv4 packets. 1238 */ 1239 u_int8_t proto; 1240 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */ 1241 struct in_addr src_ip, dst_ip; /* NOTE: network format */ 1242 u_int16_t ip_len=0; 1243 int dyn_dir = MATCH_UNKNOWN; 1244 ipfw_dyn_rule *q = NULL; 1245 1246 /* 1247 * dyn_dir = MATCH_UNKNOWN when rules unchecked, 1248 * MATCH_NONE when checked and not matched (q = NULL), 1249 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL) 1250 */ 1251 1252 if (args->eh == NULL || /* layer 3 packet */ 1253 ( m->m_pkthdr.len >= sizeof(struct ip) && 1254 ntohs(args->eh->ether_type) == ETHERTYPE_IP)) 1255 hlen = ip->ip_hl << 2; 1256 1257 /* 1258 * Collect parameters into local variables for faster matching. 1259 */ 1260 if (hlen == 0) { /* do not grab addresses for non-ip pkts */ 1261 proto = args->f_id.proto = 0; /* mark f_id invalid */ 1262 goto after_ip_checks; 1263 } 1264 1265 proto = args->f_id.proto = ip->ip_p; 1266 src_ip = ip->ip_src; 1267 dst_ip = ip->ip_dst; 1268 if (args->eh != NULL) { /* layer 2 packets are as on the wire */ 1269 offset = ntohs(ip->ip_off) & IP_OFFMASK; 1270 ip_len = ntohs(ip->ip_len); 1271 } else { 1272 offset = ip->ip_off & IP_OFFMASK; 1273 ip_len = ip->ip_len; 1274 } 1275 1276#define PULLUP_TO(len) \ 1277 do { \ 1278 if ((m)->m_len < (len)) { \ 1279 args->m = m = m_pullup(m, (len)); \ 1280 if (m == 0) \ 1281 goto pullup_failed; \ 1282 ip = mtod(m, struct ip *); \ 1283 } \ 1284 } while (0) 1285 1286 if (offset == 0) { 1287 switch (proto) { 1288 case IPPROTO_TCP: 1289 { 1290 struct tcphdr *tcp; 1291 1292 PULLUP_TO(hlen + sizeof(struct tcphdr)); 1293 tcp = L3HDR(struct tcphdr, ip); 1294 dst_port = tcp->th_dport; 1295 src_port = tcp->th_sport; 1296 args->f_id.flags = tcp->th_flags; 1297 } 1298 break; 1299 1300 case IPPROTO_UDP: 1301 { 1302 struct udphdr *udp; 1303 1304 PULLUP_TO(hlen + sizeof(struct udphdr)); 1305 udp = L3HDR(struct udphdr, ip); 1306 dst_port = udp->uh_dport; 1307 src_port = udp->uh_sport; 1308 } 1309 break; 1310 1311 case IPPROTO_ICMP: 1312 PULLUP_TO(hlen + 4); /* type, code and checksum. */ 1313 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type; 1314 break; 1315 1316 default: 1317 break; 1318 } 1319#undef PULLUP_TO 1320 } 1321 1322 args->f_id.src_ip = ntohl(src_ip.s_addr); 1323 args->f_id.dst_ip = ntohl(dst_ip.s_addr); 1324 args->f_id.src_port = src_port = ntohs(src_port); 1325 args->f_id.dst_port = dst_port = ntohs(dst_port); 1326 1327after_ip_checks: 1328 if (args->rule) { 1329 /* 1330 * Packet has already been tagged. Look for the next rule 1331 * to restart processing. 1332 * 1333 * If fw_one_pass != 0 then just accept it. 1334 * XXX should not happen here, but optimized out in 1335 * the caller. 1336 */ 1337 if (fw_one_pass) 1338 return 0; 1339 1340 f = args->rule->next_rule; 1341 if (f == NULL) 1342 f = lookup_next_rule(args->rule); 1343 } else { 1344 /* 1345 * Find the starting rule. It can be either the first 1346 * one, or the one after divert_rule if asked so. 1347 */ 1348 int skipto = args->divert_rule; 1349 1350 f = layer3_chain; 1351 if (args->eh == NULL && skipto != 0) { 1352 if (skipto >= IPFW_DEFAULT_RULE) 1353 return(IP_FW_PORT_DENY_FLAG); /* invalid */ 1354 while (f && f->rulenum <= skipto) 1355 f = f->next; 1356 if (f == NULL) /* drop packet */ 1357 return(IP_FW_PORT_DENY_FLAG); 1358 } 1359 } 1360 args->divert_rule = 0; /* reset to avoid confusion later */ 1361 1362 /* 1363 * Now scan the rules, and parse microinstructions for each rule. 1364 */ 1365 for (; f; f = f->next) { 1366 int l, cmdlen; 1367 ipfw_insn *cmd; 1368 int skip_or; /* skip rest of OR block */ 1369 1370again: 1371 skip_or = 0; 1372 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ; 1373 l -= cmdlen, cmd += cmdlen) { 1374 int match; 1375 1376 /* 1377 * check_body is a jump target used when we find a 1378 * CHECK_STATE, and need to jump to the body of 1379 * the target rule. 1380 */ 1381 1382check_body: 1383 cmdlen = F_LEN(cmd); 1384 /* 1385 * An OR block (insn_1 || .. || insn_n) has the 1386 * F_OR bit set in all but the last instruction. 1387 * The first match will set "skip_or", and cause 1388 * the following instructions to be skipped until 1389 * past the one with the F_OR bit clear. 1390 */ 1391 if (skip_or) { /* skip this instruction */ 1392 if ((cmd->len & F_OR) == 0) 1393 skip_or = 0; /* next one is good */ 1394 continue; 1395 } 1396 match = 0; /* set to 1 if we succeed */ 1397 1398 switch (cmd->opcode) { 1399 /* 1400 * The first set of opcodes compares the packet's 1401 * fields with some pattern, setting 'match' if a 1402 * match is found. At the end of the loop there is 1403 * logic to deal with F_NOT and F_OR flags associated 1404 * with the opcode. 1405 */ 1406 case O_NOP: 1407 match = 1; 1408 break; 1409 1410 case O_FORWARD_MAC: 1411 printf("ipfw: opcode %d unimplemented\n", 1412 cmd->opcode); 1413 break; 1414 1415 case O_GID: 1416 case O_UID: 1417 /* 1418 * We only check offset == 0 && proto != 0, 1419 * as this ensures that we have an IPv4 1420 * packet with the ports info. 1421 */ 1422 if (offset!=0) 1423 break; 1424 { 1425 struct inpcbinfo *pi; 1426 int wildcard; 1427 struct inpcb *pcb; 1428 1429 if (proto == IPPROTO_TCP) { 1430 wildcard = 0; 1431 pi = &tcbinfo; 1432 } else if (proto == IPPROTO_UDP) { 1433 wildcard = 1; 1434 pi = &udbinfo; 1435 } else 1436 break; 1437 1438 pcb = (oif) ? 1439 in_pcblookup_hash(pi, 1440 dst_ip, htons(dst_port), 1441 src_ip, htons(src_port), 1442 wildcard, oif) : 1443 in_pcblookup_hash(pi, 1444 src_ip, htons(src_port), 1445 dst_ip, htons(dst_port), 1446 wildcard, NULL); 1447 1448 if (pcb == NULL || pcb->inp_socket == NULL) 1449 break; 1450#if __FreeBSD_version < 500034 1451#define socheckuid(a,b) ((a)->so_cred->cr_uid == (b)) 1452#endif 1453 if (cmd->opcode == O_UID) { 1454 match = 1455 socheckuid(pcb->inp_socket, 1456 (uid_t)((ipfw_insn_u32 *)cmd)->d[0]); 1457 } else { 1458 match = groupmember( 1459 (uid_t)((ipfw_insn_u32 *)cmd)->d[0], 1460 pcb->inp_socket->so_cred); 1461 } 1462 } 1463 break; 1464 1465 case O_RECV: 1466 match = iface_match(m->m_pkthdr.rcvif, 1467 (ipfw_insn_if *)cmd); 1468 break; 1469 1470 case O_XMIT: 1471 match = iface_match(oif, (ipfw_insn_if *)cmd); 1472 break; 1473 1474 case O_VIA: 1475 match = iface_match(oif ? oif : 1476 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd); 1477 break; 1478 1479 case O_MACADDR2: 1480 if (args->eh != NULL) { /* have MAC header */ 1481 u_int32_t *want = (u_int32_t *) 1482 ((ipfw_insn_mac *)cmd)->addr; 1483 u_int32_t *mask = (u_int32_t *) 1484 ((ipfw_insn_mac *)cmd)->mask; 1485 u_int32_t *hdr = (u_int32_t *)args->eh; 1486 1487 match = 1488 ( want[0] == (hdr[0] & mask[0]) && 1489 want[1] == (hdr[1] & mask[1]) && 1490 want[2] == (hdr[2] & mask[2]) ); 1491 } 1492 break; 1493 1494 case O_MAC_TYPE: 1495 if (args->eh != NULL) { 1496 u_int16_t t = 1497 ntohs(args->eh->ether_type); 1498 u_int16_t *p = 1499 ((ipfw_insn_u16 *)cmd)->ports; 1500 int i; 1501 1502 for (i = cmdlen - 1; !match && i>0; 1503 i--, p += 2) 1504 match = (t>=p[0] && t<=p[1]); 1505 } 1506 break; 1507 1508 case O_FRAG: 1509 match = (hlen > 0 && offset != 0); 1510 break; 1511 1512 case O_IN: /* "out" is "not in" */ 1513 match = (oif == NULL); 1514 break; 1515 1516 case O_LAYER2: 1517 match = (args->eh != NULL); 1518 break; 1519 1520 case O_PROTO: 1521 /* 1522 * We do not allow an arg of 0 so the 1523 * check of "proto" only suffices. 1524 */ 1525 match = (proto == cmd->arg1); 1526 break; 1527 1528 case O_IP_SRC: 1529 match = (hlen > 0 && 1530 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1531 src_ip.s_addr); 1532 break; 1533 1534 case O_IP_SRC_MASK: 1535 match = (hlen > 0 && 1536 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1537 (src_ip.s_addr & 1538 ((ipfw_insn_ip *)cmd)->mask.s_addr)); 1539 break; 1540 1541 case O_IP_SRC_ME: 1542 if (hlen > 0) { 1543 struct ifnet *tif; 1544 1545 INADDR_TO_IFP(src_ip, tif); 1546 match = (tif != NULL); 1547 } 1548 break; 1549 1550 case O_IP_DST_SET: 1551 case O_IP_SRC_SET: 1552 if (hlen > 0) { 1553 u_int32_t *d = (u_int32_t *)(cmd+1); 1554 u_int32_t addr = 1555 cmd->opcode == O_IP_DST_SET ? 1556 args->f_id.src_ip : 1557 args->f_id.dst_ip; 1558 1559 if (addr < d[0]) 1560 break; 1561 addr -= d[0]; /* subtract base */ 1562 match = (addr < cmd->arg1) && 1563 ( d[ 1 + (addr>>5)] & 1564 (1<<(addr & 0x1f)) ); 1565 } 1566 break; 1567 1568 case O_IP_DST: 1569 match = (hlen > 0 && 1570 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1571 dst_ip.s_addr); 1572 break; 1573 1574 case O_IP_DST_MASK: 1575 match = (hlen > 0) && 1576 (((ipfw_insn_ip *)cmd)->addr.s_addr == 1577 (dst_ip.s_addr & 1578 ((ipfw_insn_ip *)cmd)->mask.s_addr)); 1579 break; 1580 1581 case O_IP_DST_ME: 1582 if (hlen > 0) { 1583 struct ifnet *tif; 1584 1585 INADDR_TO_IFP(dst_ip, tif); 1586 match = (tif != NULL); 1587 } 1588 break; 1589 1590 case O_IP_SRCPORT: 1591 case O_IP_DSTPORT: 1592 /* 1593 * offset == 0 && proto != 0 is enough 1594 * to guarantee that we have an IPv4 1595 * packet with port info. 1596 */ 1597 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP) 1598 && offset == 0) { 1599 u_int16_t x = 1600 (cmd->opcode == O_IP_SRCPORT) ? 1601 src_port : dst_port ; 1602 u_int16_t *p = 1603 ((ipfw_insn_u16 *)cmd)->ports; 1604 int i; 1605 1606 for (i = cmdlen - 1; !match && i>0; 1607 i--, p += 2) 1608 match = (x>=p[0] && x<=p[1]); 1609 } 1610 break; 1611 1612 case O_ICMPTYPE: 1613 match = (offset == 0 && proto==IPPROTO_ICMP && 1614 icmptype_match(ip, (ipfw_insn_u32 *)cmd) ); 1615 break; 1616 1617 case O_IPOPT: 1618 match = (hlen > 0 && ipopts_match(ip, cmd) ); 1619 break; 1620 1621 case O_IPVER: 1622 match = (hlen > 0 && cmd->arg1 == ip->ip_v); 1623 break; 1624 1625 case O_IPTTL: 1626 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl); 1627 break; 1628 1629 case O_IPID: 1630 match = (hlen > 0 && 1631 cmd->arg1 == ntohs(ip->ip_id)); 1632 break; 1633 1634 case O_IPLEN: 1635 match = (hlen > 0 && cmd->arg1 == ip_len); 1636 break; 1637 1638 case O_IPPRECEDENCE: 1639 match = (hlen > 0 && 1640 (cmd->arg1 == (ip->ip_tos & 0xe0)) ); 1641 break; 1642 1643 case O_IPTOS: 1644 match = (hlen > 0 && 1645 flags_match(cmd, ip->ip_tos)); 1646 break; 1647 1648 case O_TCPFLAGS: 1649 match = (proto == IPPROTO_TCP && offset == 0 && 1650 flags_match(cmd, 1651 L3HDR(struct tcphdr,ip)->th_flags)); 1652 break; 1653 1654 case O_TCPOPTS: 1655 match = (proto == IPPROTO_TCP && offset == 0 && 1656 tcpopts_match(ip, cmd)); 1657 break; 1658 1659 case O_TCPSEQ: 1660 match = (proto == IPPROTO_TCP && offset == 0 && 1661 ((ipfw_insn_u32 *)cmd)->d[0] == 1662 L3HDR(struct tcphdr,ip)->th_seq); 1663 break; 1664 1665 case O_TCPACK: 1666 match = (proto == IPPROTO_TCP && offset == 0 && 1667 ((ipfw_insn_u32 *)cmd)->d[0] == 1668 L3HDR(struct tcphdr,ip)->th_ack); 1669 break; 1670 1671 case O_TCPWIN: 1672 match = (proto == IPPROTO_TCP && offset == 0 && 1673 cmd->arg1 == 1674 L3HDR(struct tcphdr,ip)->th_win); 1675 break; 1676 1677 case O_ESTAB: 1678 /* reject packets which have SYN only */ 1679 /* XXX should i also check for TH_ACK ? */ 1680 match = (proto == IPPROTO_TCP && offset == 0 && 1681 (L3HDR(struct tcphdr,ip)->th_flags & 1682 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); 1683 break; 1684 1685 case O_LOG: 1686 ipfw_log(f, hlen, args->eh, m, oif); 1687 match = 1; 1688 break; 1689 1690 case O_PROB: 1691 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]); 1692 break; 1693 1694 /* 1695 * The second set of opcodes represents 'actions', 1696 * i.e. the terminal part of a rule once the packet 1697 * matches all previous patterns. 1698 * Typically there is only one action for each rule, 1699 * and the opcode is stored at the end of the rule 1700 * (but there are exceptions -- see below). 1701 * 1702 * In general, here we set retval and terminate the 1703 * outer loop (would be a 'break 3' in some language, 1704 * but we need to do a 'goto done'). 1705 * 1706 * Exceptions: 1707 * O_COUNT and O_SKIPTO actions: 1708 * instead of terminating, we jump to the next rule 1709 * ('goto next_rule', equivalent to a 'break 2'), 1710 * or to the SKIPTO target ('goto again' after 1711 * having set f, cmd and l), respectively. 1712 * 1713 * O_LIMIT and O_KEEP_STATE: these opcodes are 1714 * not real 'actions', and are stored right 1715 * before the 'action' part of the rule. 1716 * These opcodes try to install an entry in the 1717 * state tables; if successful, we continue with 1718 * the next opcode (match=1; break;), otherwise 1719 * the packet * must be dropped 1720 * ('goto done' after setting retval); 1721 * 1722 * O_PROBE_STATE and O_CHECK_STATE: these opcodes 1723 * cause a lookup of the state table, and a jump 1724 * to the 'action' part of the parent rule 1725 * ('goto check_body') if an entry is found, or 1726 * (CHECK_STATE only) a jump to the next rule if 1727 * the entry is not found ('goto next_rule'). 1728 * The result of the lookup is cached to make 1729 * further instances of these opcodes are 1730 * effectively NOPs. 1731 */ 1732 case O_LIMIT: 1733 case O_KEEP_STATE: 1734 if (install_state(f, 1735 (ipfw_insn_limit *)cmd, args)) { 1736 retval = IP_FW_PORT_DENY_FLAG; 1737 goto done; /* error/limit violation */ 1738 } 1739 match = 1; 1740 break; 1741 1742 case O_PROBE_STATE: 1743 case O_CHECK_STATE: 1744 /* 1745 * dynamic rules are checked at the first 1746 * keep-state or check-state occurrence, 1747 * with the result being stored in dyn_dir. 1748 * The compiler introduces a PROBE_STATE 1749 * instruction for us when we have a 1750 * KEEP_STATE (because PROBE_STATE needs 1751 * to be run first). 1752 */ 1753 if (dyn_dir == MATCH_UNKNOWN && 1754 (q = lookup_dyn_rule(&args->f_id, 1755 &dyn_dir)) != NULL) { 1756 /* 1757 * Found dynamic entry, update stats 1758 * and jump to the 'action' part of 1759 * the parent rule. 1760 */ 1761 q->pcnt++; 1762 q->bcnt += ip_len; 1763 f = q->rule; 1764 cmd = ACTION_PTR(f); 1765 l = f->cmd_len - f->act_ofs; 1766 goto check_body; 1767 } 1768 /* 1769 * Dynamic entry not found. If CHECK_STATE, 1770 * skip to next rule, if PROBE_STATE just 1771 * ignore and continue with next opcode. 1772 */ 1773 if (cmd->opcode == O_CHECK_STATE) 1774 goto next_rule; 1775 match = 1; 1776 break; 1777 1778 case O_ACCEPT: 1779 retval = 0; /* accept */ 1780 goto done; 1781 1782 case O_PIPE: 1783 case O_QUEUE: 1784 args->rule = f; /* report matching rule */ 1785 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG; 1786 goto done; 1787 1788 case O_DIVERT: 1789 case O_TEE: 1790 if (args->eh) /* not on layer 2 */ 1791 break; 1792 args->divert_rule = f->rulenum; 1793 retval = (cmd->opcode == O_DIVERT) ? 1794 cmd->arg1 : 1795 cmd->arg1 | IP_FW_PORT_TEE_FLAG; 1796 goto done; 1797 1798 case O_COUNT: 1799 case O_SKIPTO: 1800 f->pcnt++; /* update stats */ 1801 f->bcnt += ip_len; 1802 f->timestamp = time_second; 1803 if (cmd->opcode == O_COUNT) 1804 goto next_rule; 1805 /* handle skipto */ 1806 if (f->next_rule == NULL) 1807 lookup_next_rule(f); 1808 f = f->next_rule; 1809 goto again; 1810 1811 case O_REJECT: 1812 /* 1813 * Drop the packet and send a reject notice 1814 * if the packet is not ICMP (or is an ICMP 1815 * query), and it is not multicast/broadcast. 1816 */ 1817 if (hlen > 0 && 1818 (proto != IPPROTO_ICMP || 1819 is_icmp_query(ip)) && 1820 !(m->m_flags & (M_BCAST|M_MCAST)) && 1821 !IN_MULTICAST(dst_ip.s_addr)) { 1822 send_reject(args, cmd->arg1, 1823 offset,ip_len); 1824 m = args->m; 1825 } 1826 /* FALLTHROUGH */ 1827 case O_DENY: 1828 retval = IP_FW_PORT_DENY_FLAG; 1829 goto done; 1830 1831 case O_FORWARD_IP: 1832 if (args->eh) /* not valid on layer2 pkts */ 1833 break; 1834 if (!q || dyn_dir == MATCH_FORWARD) 1835 args->next_hop = 1836 &((ipfw_insn_sa *)cmd)->sa; 1837 retval = 0; 1838 goto done; 1839 1840 default: 1841 panic("-- unknown opcode %d\n", cmd->opcode); 1842 } /* end of switch() on opcodes */ 1843 1844 if (cmd->len & F_NOT) 1845 match = !match; 1846 1847 if (match) { 1848 if (cmd->len & F_OR) 1849 skip_or = 1; 1850 } else { 1851 if (!(cmd->len & F_OR)) /* not an OR block, */ 1852 break; /* try next rule */ 1853 } 1854 1855 } /* end of inner for, scan opcodes */ 1856 1857next_rule:; /* try next rule */ 1858 1859 } /* end of outer for, scan rules */ 1860 1861done: 1862 /* Update statistics */ 1863 f->pcnt++; 1864 f->bcnt += ip_len; 1865 f->timestamp = time_second; 1866 return retval; 1867 1868pullup_failed: 1869 if (fw_verbose) 1870 printf("pullup failed\n"); 1871 return(IP_FW_PORT_DENY_FLAG); 1872} 1873 1874/* 1875 * When a rule is added/deleted, clear the next_rule pointers in all rules. 1876 * These will be reconstructed on the fly as packets are matched. 1877 * Must be called at splimp(). 1878 */ 1879static void 1880flush_rule_ptrs(void) 1881{ 1882 struct ip_fw *rule; 1883 1884 for (rule = layer3_chain; rule; rule = rule->next) 1885 rule->next_rule = NULL; 1886} 1887 1888/* 1889 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given 1890 * pipe/queue, or to all of them (match == NULL). 1891 * Must be called at splimp(). 1892 */ 1893void 1894flush_pipe_ptrs(struct dn_flow_set *match) 1895{ 1896 struct ip_fw *rule; 1897 1898 for (rule = layer3_chain; rule; rule = rule->next) { 1899 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule); 1900 1901 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE) 1902 continue; 1903 1904 if (match == NULL || cmd->pipe_ptr == match) 1905 cmd->pipe_ptr = NULL; 1906 } 1907} 1908 1909/* 1910 * Add a new rule to the list. Copy the rule into a malloc'ed area, then 1911 * possibly create a rule number and add the rule to the list. 1912 * Update the rule_number in the input struct so the caller knows it as well. 1913 */ 1914static int 1915add_rule(struct ip_fw **head, struct ip_fw *input_rule) 1916{ 1917 struct ip_fw *rule, *f, *prev; 1918 int s; 1919 int l = RULESIZE(input_rule); 1920 1921 if (*head == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE) 1922 return (EINVAL); 1923 1924 rule = malloc(l, M_IPFW, M_DONTWAIT | M_ZERO); 1925 if (rule == NULL) 1926 return (ENOSPC); 1927 1928 bcopy(input_rule, rule, l); 1929 1930 rule->next = NULL; 1931 rule->next_rule = NULL; 1932 1933 rule->pcnt = 0; 1934 rule->bcnt = 0; 1935 rule->timestamp = 0; 1936 1937 s = splimp(); 1938 1939 if (*head == NULL) { /* default rule */ 1940 *head = rule; 1941 goto done; 1942 } 1943 1944 /* 1945 * If rulenum is 0, find highest numbered rule before the 1946 * default rule, and add autoinc_step 1947 */ 1948 if (autoinc_step < 1) 1949 autoinc_step = 1; 1950 else if (autoinc_step > 1000) 1951 autoinc_step = 1000; 1952 if (rule->rulenum == 0) { 1953 /* 1954 * locate the highest numbered rule before default 1955 */ 1956 for (f = *head; f; f = f->next) { 1957 if (f->rulenum == IPFW_DEFAULT_RULE) 1958 break; 1959 rule->rulenum = f->rulenum; 1960 } 1961 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step) 1962 rule->rulenum += autoinc_step; 1963 input_rule->rulenum = rule->rulenum; 1964 } 1965 1966 /* 1967 * Now insert the new rule in the right place in the sorted list. 1968 */ 1969 for (prev = NULL, f = *head; f; prev = f, f = f->next) { 1970 if (f->rulenum > rule->rulenum) { /* found the location */ 1971 if (prev) { 1972 rule->next = f; 1973 prev->next = rule; 1974 } else { /* head insert */ 1975 rule->next = *head; 1976 *head = rule; 1977 } 1978 break; 1979 } 1980 } 1981 flush_rule_ptrs(); 1982done: 1983 static_count++; 1984 static_len += l; 1985 splx(s); 1986 DEB(printf("++ installed rule %d, static count now %d\n", 1987 rule->rulenum, static_count);) 1988 return (0); 1989} 1990 1991/** 1992 * Free storage associated with a static rule (including derived 1993 * dynamic rules). 1994 * The caller is in charge of clearing rule pointers to avoid 1995 * dangling pointers. 1996 * @return a pointer to the next entry. 1997 * Arguments are not checked, so they better be correct. 1998 * Must be called at splimp(). 1999 */ 2000static struct ip_fw * 2001delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule) 2002{ 2003 struct ip_fw *n; 2004 int l = RULESIZE(rule); 2005 2006 n = rule->next; 2007 remove_dyn_rule(rule, NULL /* force removal */); 2008 if (prev == NULL) 2009 *head = n; 2010 else 2011 prev->next = n; 2012 static_count--; 2013 static_len -= l; 2014 2015 if (DUMMYNET_LOADED) 2016 ip_dn_ruledel_ptr(rule); 2017 free(rule, M_IPFW); 2018 return n; 2019} 2020 2021/* 2022 * Deletes all rules from a chain (including the default rule 2023 * if the second argument is set). 2024 * Must be called at splimp(). 2025 */ 2026static void 2027free_chain(struct ip_fw **chain, int kill_default) 2028{ 2029 struct ip_fw *rule; 2030 2031 flush_rule_ptrs(); /* more efficient to do outside the loop */ 2032 2033 while ( (rule = *chain) != NULL && 2034 (kill_default || rule->rulenum != IPFW_DEFAULT_RULE) ) 2035 delete_rule(chain, NULL, rule); 2036} 2037 2038/** 2039 * Remove all rules with given number. 2040 */ 2041static int 2042del_entry(struct ip_fw **chain, u_short rulenum) 2043{ 2044 struct ip_fw *prev, *rule; 2045 int s; 2046 2047 if (rulenum == IPFW_DEFAULT_RULE) 2048 return EINVAL; 2049 2050 /* 2051 * locate first rule to delete 2052 */ 2053 for (prev = NULL, rule = *chain; rule && rule->rulenum < rulenum; 2054 prev = rule, rule = rule->next) 2055 ; 2056 if (rule->rulenum != rulenum) 2057 return EINVAL; 2058 2059 s = splimp(); /* no access to rules while removing */ 2060 flush_rule_ptrs(); /* more efficient to do outside the loop */ 2061 /* 2062 * prev remains the same throughout the cycle 2063 */ 2064 while (rule && rule->rulenum == rulenum) 2065 rule = delete_rule(chain, prev, rule); 2066 splx(s); 2067 return 0; 2068} 2069 2070/* 2071 * Clear counters for a specific rule. 2072 */ 2073static void 2074clear_counters(struct ip_fw *rule, int log_only) 2075{ 2076 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); 2077 2078 if (log_only == 0) { 2079 rule->bcnt = rule->pcnt = 0; 2080 rule->timestamp = 0; 2081 } 2082 if (l->o.opcode == O_LOG) 2083 l->log_left = l->max_log; 2084} 2085 2086/** 2087 * Reset some or all counters on firewall rules. 2088 * @arg frwl is null to clear all entries, or contains a specific 2089 * rule number. 2090 * @arg log_only is 1 if we only want to reset logs, zero otherwise. 2091 */ 2092static int 2093zero_entry(int rulenum, int log_only) 2094{ 2095 struct ip_fw *rule; 2096 int s; 2097 char *msg; 2098 2099 if (rulenum == 0) { 2100 s = splimp(); 2101 norule_counter = 0; 2102 for (rule = layer3_chain; rule; rule = rule->next) 2103 clear_counters(rule, log_only); 2104 splx(s); 2105 msg = log_only ? "ipfw: All logging counts reset.\n" : 2106 "ipfw: Accounting cleared.\n"; 2107 } else { 2108 int cleared = 0; 2109 /* 2110 * We can have multiple rules with the same number, so we 2111 * need to clear them all. 2112 */ 2113 for (rule = layer3_chain; rule; rule = rule->next) 2114 if (rule->rulenum == rulenum) { 2115 s = splimp(); 2116 while (rule && rule->rulenum == rulenum) { 2117 clear_counters(rule, log_only); 2118 rule = rule->next; 2119 } 2120 splx(s); 2121 cleared = 1; 2122 break; 2123 } 2124 if (!cleared) /* we did not find any matching rules */ 2125 return (EINVAL); 2126 msg = log_only ? "ipfw: Entry %d logging count reset.\n" : 2127 "ipfw: Entry %d cleared.\n"; 2128 } 2129 if (fw_verbose) 2130 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum); 2131 return (0); 2132} 2133 2134/* 2135 * Check validity of the structure before insert. 2136 * Fortunately rules are simple, so this mostly need to check rule sizes. 2137 */ 2138static int 2139check_ipfw_struct(struct ip_fw *rule, int size) 2140{ 2141 int l, cmdlen = 0; 2142 int have_action=0; 2143 ipfw_insn *cmd; 2144 2145 if (size < sizeof(*rule)) { 2146 printf("ipfw: rule too short\n"); 2147 return (EINVAL); 2148 } 2149 /* first, check for valid size */ 2150 l = RULESIZE(rule); 2151 if (l != size) { 2152 printf("ipfw: size mismatch (have %d want %d)\n", size, l); 2153 return (EINVAL); 2154 } 2155 /* 2156 * Now go for the individual checks. Very simple ones, basically only 2157 * instruction sizes. 2158 */ 2159 for (l = rule->cmd_len, cmd = rule->cmd ; 2160 l > 0 ; l -= cmdlen, cmd += cmdlen) { 2161 cmdlen = F_LEN(cmd); 2162 if (cmdlen > l) { 2163 printf("ipfw: opcode %d size truncated\n", 2164 cmd->opcode); 2165 return EINVAL; 2166 } 2167 DEB(printf("ipfw: opcode %d\n", cmd->opcode);) 2168 switch (cmd->opcode) { 2169 case O_NOP: 2170 case O_PROBE_STATE: 2171 case O_KEEP_STATE: 2172 case O_PROTO: 2173 case O_IP_SRC_ME: 2174 case O_IP_DST_ME: 2175 case O_LAYER2: 2176 case O_IN: 2177 case O_FRAG: 2178 case O_IPOPT: 2179 case O_IPLEN: 2180 case O_IPID: 2181 case O_IPTOS: 2182 case O_IPPRECEDENCE: 2183 case O_IPTTL: 2184 case O_IPVER: 2185 case O_TCPWIN: 2186 case O_TCPFLAGS: 2187 case O_TCPOPTS: 2188 case O_ESTAB: 2189 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2190 goto bad_size; 2191 break; 2192 2193 case O_UID: 2194 case O_GID: 2195 case O_IP_SRC: 2196 case O_IP_DST: 2197 case O_TCPSEQ: 2198 case O_TCPACK: 2199 case O_PROB: 2200 case O_ICMPTYPE: 2201 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) 2202 goto bad_size; 2203 break; 2204 2205 case O_LIMIT: 2206 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) 2207 goto bad_size; 2208 break; 2209 2210 case O_LOG: 2211 if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) 2212 goto bad_size; 2213 2214 ((ipfw_insn_log *)cmd)->log_left = 2215 ((ipfw_insn_log *)cmd)->max_log; 2216 2217 break; 2218 2219 case O_IP_SRC_MASK: 2220 case O_IP_DST_MASK: 2221 if (cmdlen != F_INSN_SIZE(ipfw_insn_ip)) 2222 goto bad_size; 2223 if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) { 2224 printf("ipfw: opcode %d, useless rule\n", 2225 cmd->opcode); 2226 return EINVAL; 2227 } 2228 break; 2229 2230 case O_IP_SRC_SET: 2231 case O_IP_DST_SET: 2232 if (cmd->arg1 == 0 || cmd->arg1 > 256) { 2233 printf("ipfw: invalid set size %d\n", 2234 cmd->arg1); 2235 return EINVAL; 2236 } 2237 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 2238 (cmd->arg1+31)/32 ) 2239 goto bad_size; 2240 break; 2241 2242 case O_MACADDR2: 2243 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) 2244 goto bad_size; 2245 break; 2246 2247 case O_MAC_TYPE: 2248 case O_IP_SRCPORT: 2249 case O_IP_DSTPORT: /* XXX artificial limit, 15 port pairs */ 2250 if (cmdlen < 2 || cmdlen > 15) 2251 goto bad_size; 2252 break; 2253 2254 case O_RECV: 2255 case O_XMIT: 2256 case O_VIA: 2257 if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) 2258 goto bad_size; 2259 break; 2260 2261 case O_PIPE: 2262 case O_QUEUE: 2263 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe)) 2264 goto bad_size; 2265 goto check_action; 2266 2267 case O_FORWARD_IP: 2268 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) 2269 goto bad_size; 2270 goto check_action; 2271 2272 case O_FORWARD_MAC: /* XXX not implemented yet */ 2273 case O_CHECK_STATE: 2274 case O_COUNT: 2275 case O_ACCEPT: 2276 case O_DENY: 2277 case O_REJECT: 2278 case O_SKIPTO: 2279 case O_DIVERT: 2280 case O_TEE: 2281 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2282 goto bad_size; 2283check_action: 2284 if (have_action) { 2285 printf("ipfw: opcode %d, multiple actions" 2286 " not allowed\n", 2287 cmd->opcode); 2288 return EINVAL; 2289 } 2290 have_action = 1; 2291 if (l != cmdlen) { 2292 printf("ipfw: opcode %d, action must be" 2293 " last opcode\n", 2294 cmd->opcode); 2295 return EINVAL; 2296 } 2297 break; 2298 default: 2299 printf("ipfw: opcode %d, unknown opcode\n", 2300 cmd->opcode); 2301 return EINVAL; 2302 } 2303 } 2304 if (have_action == 0) { 2305 printf("ipfw: missing action\n"); 2306 return EINVAL; 2307 } 2308 return 0; 2309 2310bad_size: 2311 printf("ipfw: opcode %d size %d wrong\n", 2312 cmd->opcode, cmdlen); 2313 return EINVAL; 2314} 2315 2316 2317/** 2318 * {set|get}sockopt parser. 2319 */ 2320static int 2321ipfw_ctl(struct sockopt *sopt) 2322{ 2323 int error, s, rulenum; 2324 size_t size; 2325 struct ip_fw *bp , *buf, *rule; 2326 2327 static u_int32_t rule_buf[255]; /* we copy the data here */ 2328 2329 /* 2330 * Disallow modifications in really-really secure mode, but still allow 2331 * the logging counters to be reset. 2332 */ 2333 if (sopt->sopt_name == IP_FW_ADD || 2334 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) { 2335#if __FreeBSD_version >= 500034 2336 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 2337 if (error) 2338 return (error); 2339#else /* FreeBSD 4.x */ 2340 if (securelevel >= 3) 2341 return (EPERM); 2342#endif 2343 } 2344 2345 error = 0; 2346 2347 switch (sopt->sopt_name) { 2348 case IP_FW_GET: 2349 /* 2350 * pass up a copy of the current rules. Static rules 2351 * come first (the last of which has number IPFW_DEFAULT_RULE), 2352 * followed by a possibly empty list of dynamic rule. 2353 * The last dynamic rule has NULL in the "next" field. 2354 */ 2355 s = splimp(); 2356 size = static_len; /* size of static rules */ 2357 if (ipfw_dyn_v) /* add size of dyn.rules */ 2358 size += (dyn_count * sizeof(ipfw_dyn_rule)); 2359 2360 /* 2361 * XXX todo: if the user passes a short length just to know 2362 * how much room is needed, do not bother filling up the 2363 * buffer, just jump to the sooptcopyout. 2364 */ 2365 buf = malloc(size, M_TEMP, M_WAITOK); 2366 if (buf == 0) { 2367 splx(s); 2368 error = ENOBUFS; 2369 break; 2370 } 2371 2372 bp = buf; 2373 for (rule = layer3_chain; rule ; rule = rule->next) { 2374 int i = RULESIZE(rule); 2375 bcopy(rule, bp, i); 2376 bp = (struct ip_fw *)((char *)bp + i); 2377 } 2378 if (ipfw_dyn_v) { 2379 int i; 2380 ipfw_dyn_rule *p, *dst, *last = NULL; 2381 2382 dst = (ipfw_dyn_rule *)bp; 2383 for (i = 0 ; i < curr_dyn_buckets ; i++ ) 2384 for ( p = ipfw_dyn_v[i] ; p != NULL ; 2385 p = p->next, dst++ ) { 2386 bcopy(p, dst, sizeof *p); 2387 (int)dst->rule = p->rule->rulenum ; 2388 /* 2389 * store a non-null value in "next". 2390 * The userland code will interpret a 2391 * NULL here as a marker 2392 * for the last dynamic rule. 2393 */ 2394 dst->next = dst ; 2395 last = dst ; 2396 dst->expire = 2397 TIME_LEQ(dst->expire, time_second) ? 2398 0 : dst->expire - time_second ; 2399 } 2400 if (last != NULL) /* mark last dynamic rule */ 2401 last->next = NULL; 2402 } 2403 splx(s); 2404 2405 error = sooptcopyout(sopt, buf, size); 2406 free(buf, M_TEMP); 2407 break; 2408 2409 case IP_FW_FLUSH: 2410 /* 2411 * Normally we cannot release the lock on each iteration. 2412 * We could do it here only because we start from the head all 2413 * the times so there is no risk of missing some entries. 2414 * On the other hand, the risk is that we end up with 2415 * a very inconsistent ruleset, so better keep the lock 2416 * around the whole cycle. 2417 * 2418 * XXX this code can be improved by resetting the head of 2419 * the list to point to the default rule, and then freeing 2420 * the old list without the need for a lock. 2421 */ 2422 2423 s = splimp(); 2424 free_chain(&layer3_chain, 0 /* keep default rule */); 2425 splx(s); 2426 break; 2427 2428 case IP_FW_ADD: 2429 rule = (struct ip_fw *)rule_buf; /* XXX do a malloc */ 2430 error = sooptcopyin(sopt, rule, sizeof(rule_buf), 2431 sizeof(struct ip_fw) ); 2432 size = sopt->sopt_valsize; 2433 if (error || (error = check_ipfw_struct(rule, size))) 2434 break; 2435 2436 error = add_rule(&layer3_chain, rule); 2437 size = RULESIZE(rule); 2438 if (!error && sopt->sopt_dir == SOPT_GET) 2439 error = sooptcopyout(sopt, rule, size); 2440 break; 2441 2442 case IP_FW_DEL: /* argument is an int, the rule number */ 2443 error = sooptcopyin(sopt, &rulenum, sizeof(int), sizeof(int)); 2444 if (error) 2445 break; 2446 if (rulenum == IPFW_DEFAULT_RULE) { 2447 if (fw_debug) 2448 printf("ipfw: can't delete rule %u\n", 2449 (unsigned)IPFW_DEFAULT_RULE); 2450 error = EINVAL; 2451 } else 2452 error = del_entry(&layer3_chain, rulenum); 2453 break; 2454 2455 case IP_FW_ZERO: 2456 case IP_FW_RESETLOG: /* argument is an int, the rule number */ 2457 rulenum=0; 2458 2459 if (sopt->sopt_val != 0) { 2460 error = sooptcopyin(sopt, &rulenum, 2461 sizeof(int), sizeof(int)); 2462 if (error) 2463 break; 2464 } 2465 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG); 2466 break; 2467 2468 default: 2469 printf("ipfw_ctl invalid option %d\n", sopt->sopt_name); 2470 error = EINVAL; 2471 } 2472 2473 return (error); 2474} 2475 2476/** 2477 * dummynet needs a reference to the default rule, because rules can be 2478 * deleted while packets hold a reference to them. When this happens, 2479 * dummynet changes the reference to the default rule (it could well be a 2480 * NULL pointer, but this way we do not need to check for the special 2481 * case, plus here he have info on the default behaviour). 2482 */ 2483struct ip_fw *ip_fw_default_rule; 2484 2485static void 2486ipfw_init(void) 2487{ 2488 struct ip_fw default_rule; 2489 2490 ip_fw_chk_ptr = ipfw_chk; 2491 ip_fw_ctl_ptr = ipfw_ctl; 2492 layer3_chain = NULL; 2493 2494 bzero(&default_rule, sizeof default_rule); 2495 2496 default_rule.act_ofs = 0; 2497 default_rule.rulenum = IPFW_DEFAULT_RULE; 2498 default_rule.cmd_len = 1; 2499 2500 default_rule.cmd[0].len = 1; 2501 default_rule.cmd[0].opcode = 2502#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT 2503 1 ? O_ACCEPT : 2504#endif 2505 O_DENY; 2506 2507 add_rule(&layer3_chain, &default_rule); 2508 2509 ip_fw_default_rule = layer3_chain; 2510 printf("IP packet filtering initialized, divert %s, " 2511 "rule-based forwarding %s, default to %s, logging ", 2512#ifdef IPDIVERT 2513 "enabled", 2514#else 2515 "disabled", 2516#endif 2517 "enabled", 2518 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny"); 2519 2520#ifdef IPFIREWALL_VERBOSE 2521 fw_verbose = 1; 2522#endif 2523#ifdef IPFIREWALL_VERBOSE_LIMIT 2524 verbose_limit = IPFIREWALL_VERBOSE_LIMIT; 2525#endif 2526 printf("logging "); 2527 if (fw_verbose == 0) 2528 printf("disabled\n"); 2529 else if (verbose_limit == 0) 2530 printf("unlimited\n"); 2531 else 2532 printf("limited to %d packets/entry by default\n", 2533 verbose_limit); 2534} 2535 2536static int 2537ipfw_modevent(module_t mod, int type, void *unused) 2538{ 2539 int s; 2540 int err = 0; 2541 2542 switch (type) { 2543 case MOD_LOAD: 2544 s = splimp(); 2545 if (IPFW_LOADED) { 2546 splx(s); 2547 printf("IP firewall already loaded\n"); 2548 err = EEXIST; 2549 } else { 2550 ipfw_init(); 2551 splx(s); 2552 } 2553 break; 2554 2555 case MOD_UNLOAD: 2556#if !defined(KLD_MODULE) 2557 printf("ipfw statically compiled, cannot unload\n"); 2558 err = EBUSY; 2559#else 2560 s = splimp(); 2561 ip_fw_chk_ptr = NULL; 2562 ip_fw_ctl_ptr = NULL; 2563 free_chain(&layer3_chain, 1 /* kill default rule */); 2564 splx(s); 2565 printf("IP firewall unloaded\n"); 2566#endif 2567 break; 2568 default: 2569 break; 2570 } 2571 return err; 2572} 2573 2574static moduledata_t ipfwmod = { 2575 "ipfw", 2576 ipfw_modevent, 2577 0 2578}; 2579DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY); 2580MODULE_VERSION(ipfw, 1); 2581#endif /* IPFW2 */ 2582