ip_fw2.c revision 100589
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 100589 2002-07-24 02:41:19Z 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 83static struct callout_handle ipfw_timeout_h; 84#define IPFW_DEFAULT_RULE 65535 85 86/* 87 * list of rules for layer 3 88 */ 89static struct ip_fw *layer3_chain; 90 91MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's"); 92 93static int fw_debug = 1; 94int fw_one_pass = 1; 95static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */ 96 97#ifdef SYSCTL_NODE 98SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); 99SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW, 100 &fw_enable, 0, "Enable ipfw"); 101SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW, 102 &autoinc_step, 0, "Rule number autincrement step"); 103SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW, 104 &fw_one_pass, 0, 105 "Only do a single pass through ipfw when using dummynet(4)"); 106SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW, 107 &fw_debug, 0, "Enable printing of debug ip_fw statements"); 108SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW, 109 &fw_verbose, 0, "Log matches to ipfw rules"); 110SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW, 111 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged"); 112 113/* 114 * Description of dynamic rules. 115 * 116 * Dynamic rules are stored in lists accessed through a hash table 117 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 118 * be modified through the sysctl variable dyn_buckets which is 119 * updated when the table becomes empty. 120 * 121 * XXX currently there is only one list, ipfw_dyn. 122 * 123 * When a packet is received, its address fields are first masked 124 * with the mask defined for the rule, then hashed, then matched 125 * against the entries in the corresponding list. 126 * Dynamic rules can be used for different purposes: 127 * + stateful rules; 128 * + enforcing limits on the number of sessions; 129 * + in-kernel NAT (not implemented yet) 130 * 131 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 132 * measured in seconds and depending on the flags. 133 * 134 * The total number of dynamic rules is stored in dyn_count. 135 * The max number of dynamic rules is dyn_max. When we reach 136 * the maximum number of rules we do not create anymore. This is 137 * done to avoid consuming too much memory, but also too much 138 * time when searching on each packet (ideally, we should try instead 139 * to put a limit on the length of the list on each bucket...). 140 * 141 * Each dynamic rule holds a pointer to the parent ipfw rule so 142 * we know what action to perform. Dynamic rules are removed when 143 * the parent rule is deleted. XXX we should make them survive. 144 * 145 * There are some limitations with dynamic rules -- we do not 146 * obey the 'randomized match', and we do not do multiple 147 * passes through the firewall. XXX check the latter!!! 148 */ 149static ipfw_dyn_rule **ipfw_dyn_v = NULL; 150static u_int32_t dyn_buckets = 256; /* must be power of 2 */ 151static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */ 152 153/* 154 * Timeouts for various events in handing dynamic rules. 155 */ 156static u_int32_t dyn_ack_lifetime = 300; 157static u_int32_t dyn_syn_lifetime = 20; 158static u_int32_t dyn_fin_lifetime = 1; 159static u_int32_t dyn_rst_lifetime = 1; 160static u_int32_t dyn_udp_lifetime = 10; 161static u_int32_t dyn_short_lifetime = 5; 162 163static u_int32_t dyn_keepalive = 1; /* do send keepalives */ 164 165static u_int32_t static_count; /* # of static rules */ 166static u_int32_t static_len; /* size in bytes of static rules */ 167static u_int32_t dyn_count; /* # of dynamic rules */ 168static u_int32_t dyn_max = 1000; /* max # of dynamic rules */ 169 170SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW, 171 &dyn_buckets, 0, "Number of dyn. buckets"); 172SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD, 173 &curr_dyn_buckets, 0, "Current Number of dyn. buckets"); 174SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD, 175 &dyn_count, 0, "Number of dyn. rules"); 176SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW, 177 &dyn_max, 0, "Max number of dyn. rules"); 178SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD, 179 &static_count, 0, "Number of static rules"); 180SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW, 181 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks"); 182SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW, 183 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn"); 184SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW, 185 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin"); 186SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW, 187 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst"); 188SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW, 189 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP"); 190SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW, 191 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations"); 192SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW, 193 &dyn_keepalive, 0, "Enable keepalives for dyn. rules"); 194 195#endif /* SYSCTL_NODE */ 196 197 198static ip_fw_chk_t ipfw_chk; 199 200ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL; /* hook into dummynet */ 201 202/* 203 * This macro maps an ip pointer into a layer3 header pointer of type T 204 */ 205#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl)) 206 207static __inline int 208icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd) 209{ 210 int type = L3HDR(struct icmp,ip)->icmp_type; 211 212 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) ); 213} 214 215#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \ 216 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) ) 217 218static int 219is_icmp_query(struct ip *ip) 220{ 221 int type = L3HDR(struct icmp, ip)->icmp_type; 222 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) ); 223} 224#undef TT 225 226/* 227 * The following checks use two arrays of 8 or 16 bits to store the 228 * bits that we want set or clear, respectively. They are in the 229 * low and high half of cmd->arg1 or cmd->d[0]. 230 * 231 * We scan options and store the bits we find set. We succeed if 232 * 233 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear 234 * 235 * The code is sometimes optimized not to store additional variables. 236 */ 237 238static int 239flags_match(ipfw_insn *cmd, u_int8_t bits) 240{ 241 u_char want_clear; 242 bits = ~bits; 243 244 if ( ((cmd->arg1 & 0xff) & bits) != 0) 245 return 0; /* some bits we want set were clear */ 246 want_clear = (cmd->arg1 >> 8) & 0xff; 247 if ( (want_clear & bits) != want_clear) 248 return 0; /* some bits we want clear were set */ 249 return 1; 250} 251 252static int 253ipopts_match(struct ip *ip, ipfw_insn *cmd) 254{ 255 int optlen, bits = 0; 256 u_char *cp = (u_char *)(ip + 1); 257 int x = (ip->ip_hl << 2) - sizeof (struct ip); 258 259 for (; x > 0; x -= optlen, cp += optlen) { 260 int opt = cp[IPOPT_OPTVAL]; 261 262 if (opt == IPOPT_EOL) 263 break; 264 if (opt == IPOPT_NOP) 265 optlen = 1; 266 else { 267 optlen = cp[IPOPT_OLEN]; 268 if (optlen <= 0 || optlen > x) 269 return 0; /* invalid or truncated */ 270 } 271 switch (opt) { 272 273 default: 274 break; 275 276 case IPOPT_LSRR: 277 bits |= IP_FW_IPOPT_LSRR; 278 break; 279 280 case IPOPT_SSRR: 281 bits |= IP_FW_IPOPT_SSRR; 282 break; 283 284 case IPOPT_RR: 285 bits |= IP_FW_IPOPT_RR; 286 break; 287 288 case IPOPT_TS: 289 bits |= IP_FW_IPOPT_TS; 290 break; 291 } 292 } 293 return (flags_match(cmd, bits)); 294} 295 296static int 297tcpopts_match(struct ip *ip, ipfw_insn *cmd) 298{ 299 int optlen, bits = 0; 300 struct tcphdr *tcp = L3HDR(struct tcphdr,ip); 301 u_char *cp = (u_char *)(tcp + 1); 302 int x = (tcp->th_off << 2) - sizeof(struct tcphdr); 303 304 for (; x > 0; x -= optlen, cp += optlen) { 305 int opt = cp[0]; 306 if (opt == TCPOPT_EOL) 307 break; 308 if (opt == TCPOPT_NOP) 309 optlen = 1; 310 else { 311 optlen = cp[1]; 312 if (optlen <= 0) 313 break; 314 } 315 316 switch (opt) { 317 318 default: 319 break; 320 321 case TCPOPT_MAXSEG: 322 bits |= IP_FW_TCPOPT_MSS; 323 break; 324 325 case TCPOPT_WINDOW: 326 bits |= IP_FW_TCPOPT_WINDOW; 327 break; 328 329 case TCPOPT_SACK_PERMITTED: 330 case TCPOPT_SACK: 331 bits |= IP_FW_TCPOPT_SACK; 332 break; 333 334 case TCPOPT_TIMESTAMP: 335 bits |= IP_FW_TCPOPT_TS; 336 break; 337 338 case TCPOPT_CC: 339 case TCPOPT_CCNEW: 340 case TCPOPT_CCECHO: 341 bits |= IP_FW_TCPOPT_CC; 342 break; 343 } 344 } 345 return (flags_match(cmd, bits)); 346} 347 348static int 349iface_match(struct ifnet *ifp, ipfw_insn_if *cmd) 350{ 351 if (ifp == NULL) /* no iface with this packet, match fails */ 352 return 0; 353 /* Check by name or by IP address */ 354 if (cmd->name[0] != '\0') { /* match by name */ 355 /* Check unit number (-1 is wildcard) */ 356 if (cmd->p.unit != -1 && cmd->p.unit != ifp->if_unit) 357 return(0); 358 /* Check name */ 359 if (!strncmp(ifp->if_name, cmd->name, IFNAMSIZ)) 360 return(1); 361 } else { 362 struct ifaddr *ia; 363 364 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) { 365 if (ia->ifa_addr == NULL) 366 continue; 367 if (ia->ifa_addr->sa_family != AF_INET) 368 continue; 369 if (cmd->p.ip.s_addr == ((struct sockaddr_in *) 370 (ia->ifa_addr))->sin_addr.s_addr) 371 return(1); /* match */ 372 } 373 } 374 return(0); /* no match, fail ... */ 375} 376 377static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */ 378 379#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0 380#define SNP(buf) buf, sizeof(buf) 381 382/* 383 * We enter here when we have a rule with O_LOG. 384 * XXX this function alone takes about 2Kbytes of code! 385 */ 386static void 387ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh, 388 struct mbuf *m, struct ifnet *oif) 389{ 390 char *action; 391 int limit_reached = 0; 392 char action2[40], proto[48], fragment[28]; 393 394 fragment[0] = '\0'; 395 proto[0] = '\0'; 396 397 if (f == NULL) { /* bogus pkt */ 398 if (verbose_limit != 0 && norule_counter >= verbose_limit) 399 return; 400 norule_counter++; 401 if (norule_counter == verbose_limit) 402 limit_reached = verbose_limit; 403 action = "Refuse"; 404 } else { /* O_LOG is the first action, find the real one */ 405 ipfw_insn *cmd = ACTION_PTR(f); 406 ipfw_insn_log *l = (ipfw_insn_log *)cmd; 407 408 if (l->max_log != 0 && l->log_left == 0) 409 return; 410 l->log_left--; 411 if (l->log_left == 0) 412 limit_reached = l->max_log; 413 cmd += F_LEN(cmd); /* point to first action */ 414 if (cmd->opcode == O_PROB) 415 cmd += F_LEN(cmd); 416 417 action = action2; 418 switch (cmd->opcode) { 419 case O_DENY: 420 action = "Deny"; 421 break; 422 423 case O_REJECT: 424 if (cmd->arg1==ICMP_REJECT_RST) 425 action = "Reset"; 426 else if (cmd->arg1==ICMP_UNREACH_HOST) 427 action = "Reject"; 428 else 429 snprintf(SNPARGS(action2, 0), "Unreach %d", 430 cmd->arg1); 431 break; 432 433 case O_ACCEPT: 434 action = "Accept"; 435 break; 436 case O_COUNT: 437 action = "Count"; 438 break; 439 case O_DIVERT: 440 snprintf(SNPARGS(action2, 0), "Divert %d", 441 cmd->arg1); 442 break; 443 case O_TEE: 444 snprintf(SNPARGS(action2, 0), "Tee %d", 445 cmd->arg1); 446 break; 447 case O_SKIPTO: 448 snprintf(SNPARGS(action2, 0), "SkipTo %d", 449 cmd->arg1); 450 break; 451 case O_PIPE: 452 snprintf(SNPARGS(action2, 0), "Pipe %d", 453 cmd->arg1); 454 break; 455 case O_QUEUE: 456 snprintf(SNPARGS(action2, 0), "Queue %d", 457 cmd->arg1); 458 break; 459 case O_FORWARD_IP: { 460 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd; 461 int len; 462 463 len = snprintf(SNPARGS(action2, 0), "Forward to %s", 464 inet_ntoa(sa->sa.sin_addr)); 465 if (sa->sa.sin_port) 466 snprintf(SNPARGS(action2, len), ":%d", 467 ntohs(sa->sa.sin_port)); 468 } 469 break; 470 default: 471 action = "UNKNOWN"; 472 break; 473 } 474 } 475 476 if (hlen == 0) { /* non-ip */ 477 snprintf(SNPARGS(proto, 0), "MAC"); 478 } else { 479 struct ip *ip = mtod(m, struct ip *); 480 /* these three are all aliases to the same thing */ 481 struct icmp *const icmp = L3HDR(struct icmp, ip); 482 struct tcphdr *const tcp = (struct tcphdr *)icmp; 483 struct udphdr *const udp = (struct udphdr *)icmp; 484 485 int ip_off, offset, ip_len; 486 487 int len; 488 489 if (eh != NULL) { /* layer 2 packets are as on the wire */ 490 ip_off = ntohs(ip->ip_off); 491 ip_len = ntohs(ip->ip_len); 492 } else { 493 ip_off = ip->ip_off; 494 ip_len = ip->ip_len; 495 } 496 offset = ip_off & IP_OFFMASK; 497 switch (ip->ip_p) { 498 case IPPROTO_TCP: 499 len = snprintf(SNPARGS(proto, 0), "TCP %s", 500 inet_ntoa(ip->ip_src)); 501 if (offset == 0) 502 snprintf(SNPARGS(proto, len), ":%d %s:%d", 503 ntohs(tcp->th_sport), 504 inet_ntoa(ip->ip_dst), 505 ntohs(tcp->th_dport)); 506 else 507 snprintf(SNPARGS(proto, len), " %s", 508 inet_ntoa(ip->ip_dst)); 509 break; 510 511 case IPPROTO_UDP: 512 len = snprintf(SNPARGS(proto, 0), "UDP %s", 513 inet_ntoa(ip->ip_src)); 514 if (offset == 0) 515 snprintf(SNPARGS(proto, len), ":%d %s:%d", 516 ntohs(udp->uh_sport), 517 inet_ntoa(ip->ip_dst), 518 ntohs(udp->uh_dport)); 519 else 520 snprintf(SNPARGS(proto, len), " %s", 521 inet_ntoa(ip->ip_dst)); 522 break; 523 524 case IPPROTO_ICMP: 525 if (offset == 0) 526 len = snprintf(SNPARGS(proto, 0), 527 "ICMP:%u.%u ", 528 icmp->icmp_type, icmp->icmp_code); 529 else 530 len = snprintf(SNPARGS(proto, 0), "ICMP "); 531 len += snprintf(SNPARGS(proto, len), "%s", 532 inet_ntoa(ip->ip_src)); 533 snprintf(SNPARGS(proto, len), " %s", 534 inet_ntoa(ip->ip_dst)); 535 break; 536 537 default: 538 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p, 539 inet_ntoa(ip->ip_src)); 540 snprintf(SNPARGS(proto, len), " %s", 541 inet_ntoa(ip->ip_dst)); 542 break; 543 } 544 545 if (ip_off & (IP_MF | IP_OFFMASK)) 546 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)", 547 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2), 548 offset << 3, 549 (ip_off & IP_MF) ? "+" : ""); 550 } 551 if (oif || m->m_pkthdr.rcvif) 552 log(LOG_SECURITY | LOG_INFO, 553 "ipfw: %d %s %s %s via %s%d%s\n", 554 f ? f->rulenum : -1, 555 action, proto, oif ? "out" : "in", 556 oif ? oif->if_name : m->m_pkthdr.rcvif->if_name, 557 oif ? oif->if_unit : m->m_pkthdr.rcvif->if_unit, 558 fragment); 559 else 560 log(LOG_SECURITY | LOG_INFO, 561 "ipfw: %d %s %s [no if info]%s\n", 562 f ? f->rulenum : -1, 563 action, proto, fragment); 564 if (limit_reached) 565 log(LOG_SECURITY | LOG_NOTICE, 566 "ipfw: limit %d reached on entry %d\n", 567 limit_reached, f ? f->rulenum : -1); 568} 569 570/* 571 * IMPORTANT: the hash function for dynamic rules must be commutative 572 * in source and destination (ip,port), because rules are bidirectional 573 * and we want to find both in the same bucket. 574 */ 575static __inline int 576hash_packet(struct ipfw_flow_id *id) 577{ 578 u_int32_t i; 579 580 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 581 i &= (curr_dyn_buckets - 1); 582 return i; 583} 584 585/** 586 * unlink a dynamic rule from a chain. prev is a pointer to 587 * the previous one, q is a pointer to the rule to delete, 588 * head is a pointer to the head of the queue. 589 * Modifies q and potentially also head. 590 */ 591#define UNLINK_DYN_RULE(prev, head, q) { \ 592 ipfw_dyn_rule *old_q = q; \ 593 \ 594 /* remove a refcount to the parent */ \ 595 if (q->dyn_type == O_LIMIT) \ 596 q->parent->count--; \ 597 DEB(printf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \ 598 (q->id.src_ip), (q->id.src_port), \ 599 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \ 600 if (prev != NULL) \ 601 prev->next = q = q->next; \ 602 else \ 603 head = q = q->next; \ 604 dyn_count--; \ 605 free(old_q, M_IPFW); } 606 607#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 608 609/** 610 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 611 * 612 * If keep_me == NULL, rules are deleted even if not expired, 613 * otherwise only expired rules are removed. 614 * 615 * The value of the second parameter is also used to point to identify 616 * a rule we absolutely do not want to remove (e.g. because we are 617 * holding a reference to it -- this is the case with O_LIMIT_PARENT 618 * rules). The pointer is only used for comparison, so any non-null 619 * value will do. 620 */ 621static void 622remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 623{ 624 static u_int32_t last_remove = 0; 625 626#define FORCE (keep_me == NULL) 627 628 ipfw_dyn_rule *prev, *q; 629 int i, pass = 0, max_pass = 0; 630 631 if (ipfw_dyn_v == NULL || dyn_count == 0) 632 return; 633 /* do not expire more than once per second, it is useless */ 634 if (!FORCE && last_remove == time_second) 635 return; 636 last_remove = time_second; 637 638 /* 639 * because O_LIMIT refer to parent rules, during the first pass only 640 * remove child and mark any pending LIMIT_PARENT, and remove 641 * them in a second pass. 642 */ 643next_pass: 644 for (i = 0 ; i < curr_dyn_buckets ; i++) { 645 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) { 646 /* 647 * Logic can become complex here, so we split tests. 648 */ 649 if (q == keep_me) 650 goto next; 651 if (rule != NULL && rule != q->rule) 652 goto next; /* not the one we are looking for */ 653 if (q->dyn_type == O_LIMIT_PARENT) { 654 /* 655 * handle parent in the second pass, 656 * record we need one. 657 */ 658 max_pass = 1; 659 if (pass == 0) 660 goto next; 661 if (FORCE && q->count != 0 ) { 662 /* XXX should not happen! */ 663 printf( "OUCH! cannot remove rule," 664 " count %d\n", q->count); 665 } 666 } else { 667 if (!FORCE && 668 !TIME_LEQ( q->expire, time_second )) 669 goto next; 670 } 671 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 672 continue; 673next: 674 prev=q; 675 q=q->next; 676 } 677 } 678 if (pass++ < max_pass) 679 goto next_pass; 680} 681 682 683/** 684 * lookup a dynamic rule. 685 */ 686static ipfw_dyn_rule * 687lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 688 struct tcphdr *tcp) 689{ 690 /* 691 * stateful ipfw extensions. 692 * Lookup into dynamic session queue 693 */ 694#define MATCH_REVERSE 0 695#define MATCH_FORWARD 1 696#define MATCH_NONE 2 697#define MATCH_UNKNOWN 3 698 int i, dir = MATCH_NONE; 699 ipfw_dyn_rule *prev, *q=NULL; 700 701 if (ipfw_dyn_v == NULL) 702 goto done; /* not found */ 703 i = hash_packet( pkt ); 704 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) { 705 if (q->dyn_type == O_LIMIT_PARENT) 706 goto next; 707 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */ 708 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 709 continue; 710 } 711 if ( pkt->proto == q->id.proto) { 712 if (pkt->src_ip == q->id.src_ip && 713 pkt->dst_ip == q->id.dst_ip && 714 pkt->src_port == q->id.src_port && 715 pkt->dst_port == q->id.dst_port ) { 716 dir = MATCH_FORWARD; 717 break; 718 } 719 if (pkt->src_ip == q->id.dst_ip && 720 pkt->dst_ip == q->id.src_ip && 721 pkt->src_port == q->id.dst_port && 722 pkt->dst_port == q->id.src_port ) { 723 dir = MATCH_REVERSE; 724 break; 725 } 726 } 727next: 728 prev = q; 729 q = q->next; 730 } 731 if (q == NULL) 732 goto done; /* q = NULL, not found */ 733 734 if ( prev != NULL) { /* found and not in front */ 735 prev->next = q->next; 736 q->next = ipfw_dyn_v[i]; 737 ipfw_dyn_v[i] = q; 738 } 739 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 740 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST); 741 742#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 743#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 744 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8); 745 switch (q->state) { 746 case TH_SYN: /* opening */ 747 q->expire = time_second + dyn_syn_lifetime; 748 break; 749 case BOTH_SYN: /* move to established */ 750 case BOTH_SYN | TH_FIN : /* one side tries to close */ 751 case BOTH_SYN | (TH_FIN << 8) : 752 if (tcp) { 753#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 754 u_int32_t ack = ntohl(tcp->th_ack); 755 if (dir == MATCH_FORWARD) { 756 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd)) 757 q->ack_fwd = ack; 758 else { /* ignore out-of-sequence */ 759 break; 760 } 761 } else { 762 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev)) 763 q->ack_rev = ack; 764 else { /* ignore out-of-sequence */ 765 break; 766 } 767 } 768 } 769 q->expire = time_second + dyn_ack_lifetime; 770 break; 771 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 772 q->expire = time_second + dyn_fin_lifetime; 773 break; 774 default: 775#if 0 776 /* 777 * reset or some invalid combination, but can also 778 * occur if we use keep-state the wrong way. 779 */ 780 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 781 printf("invalid state: 0x%x\n", q->state); 782#endif 783 q->expire = time_second + dyn_rst_lifetime; 784 break; 785 } 786 } else if (pkt->proto == IPPROTO_UDP) { 787 q->expire = time_second + dyn_udp_lifetime; 788 } else { 789 /* other protocols */ 790 q->expire = time_second + dyn_short_lifetime; 791 } 792done: 793 if (match_direction) 794 *match_direction = dir; 795 return q; 796} 797 798static void 799realloc_dynamic_table(void) 800{ 801 /* try reallocation, make sure we have a power of 2 */ 802 803 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */ 804 dyn_buckets = curr_dyn_buckets; /* reset */ 805 return; 806 } 807 curr_dyn_buckets = dyn_buckets; 808 if (ipfw_dyn_v != NULL) 809 free(ipfw_dyn_v, M_IPFW); 810 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 811 M_IPFW, M_DONTWAIT | M_ZERO); 812} 813 814/** 815 * Install state of type 'type' for a dynamic session. 816 * The hash table contains two type of rules: 817 * - regular rules (O_KEEP_STATE) 818 * - rules for sessions with limited number of sess per user 819 * (O_LIMIT). When they are created, the parent is 820 * increased by 1, and decreased on delete. In this case, 821 * the third parameter is the parent rule and not the chain. 822 * - "parent" rules for the above (O_LIMIT_PARENT). 823 */ 824static ipfw_dyn_rule * 825add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule) 826{ 827 ipfw_dyn_rule *r; 828 int i; 829 830 if (ipfw_dyn_v == NULL || 831 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) { 832 realloc_dynamic_table(); 833 if (ipfw_dyn_v == NULL) 834 return NULL; /* failed ! */ 835 } 836 i = hash_packet(id); 837 838 r = malloc(sizeof *r, M_IPFW, M_DONTWAIT | M_ZERO); 839 if (r == NULL) { 840 printf ("sorry cannot allocate state\n"); 841 return NULL; 842 } 843 844 /* increase refcount on parent, and set pointer */ 845 if (dyn_type == O_LIMIT) { 846 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 847 if ( parent->dyn_type != O_LIMIT_PARENT) 848 panic("invalid parent"); 849 parent->count++; 850 r->parent = parent; 851 rule = parent->rule; 852 } 853 854 r->id = *id; 855 r->expire = time_second + dyn_syn_lifetime; 856 r->rule = rule; 857 r->dyn_type = dyn_type; 858 r->pcnt = r->bcnt = 0; 859 r->count = 0; 860 861 r->bucket = i; 862 r->next = ipfw_dyn_v[i]; 863 ipfw_dyn_v[i] = r; 864 dyn_count++; 865 DEB(printf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n", 866 dyn_type, 867 (r->id.src_ip), (r->id.src_port), 868 (r->id.dst_ip), (r->id.dst_port), 869 dyn_count ); ) 870 return r; 871} 872 873/** 874 * lookup dynamic parent rule using pkt and rule as search keys. 875 * If the lookup fails, then install one. 876 */ 877static ipfw_dyn_rule * 878lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 879{ 880 ipfw_dyn_rule *q; 881 int i; 882 883 if (ipfw_dyn_v) { 884 i = hash_packet( pkt ); 885 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next) 886 if (q->dyn_type == O_LIMIT_PARENT && 887 rule== q->rule && 888 pkt->proto == q->id.proto && 889 pkt->src_ip == q->id.src_ip && 890 pkt->dst_ip == q->id.dst_ip && 891 pkt->src_port == q->id.src_port && 892 pkt->dst_port == q->id.dst_port) { 893 q->expire = time_second + dyn_short_lifetime; 894 DEB(printf("lookup_dyn_parent found 0x%p\n",q);) 895 return q; 896 } 897 } 898 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 899} 900 901/** 902 * Install dynamic state for rule type cmd->o.opcode 903 * 904 * Returns 1 (failure) if state is not installed because of errors or because 905 * session limitations are enforced. 906 */ 907static int 908install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 909 struct ip_fw_args *args) 910{ 911 static int last_log; 912 913 ipfw_dyn_rule *q; 914 915 DEB(printf("-- install state type %d 0x%08x %u -> 0x%08x %u\n", 916 cmd->o.opcode, 917 (args->f_id.src_ip), (args->f_id.src_port), 918 (args->f_id.dst_ip), (args->f_id.dst_port) );) 919 920 q = lookup_dyn_rule(&args->f_id, NULL, NULL); 921 922 if (q != NULL) { /* should never occur */ 923 if (last_log != time_second) { 924 last_log = time_second; 925 printf(" install_state: entry already present, done\n"); 926 } 927 return 0; 928 } 929 930 if (dyn_count >= dyn_max) 931 /* 932 * Run out of slots, try to remove any expired rule. 933 */ 934 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1); 935 936 if (dyn_count >= dyn_max) { 937 if (last_log != time_second) { 938 last_log = time_second; 939 printf("install_state: Too many dynamic rules\n"); 940 } 941 return 1; /* cannot install, notify caller */ 942 } 943 944 switch (cmd->o.opcode) { 945 case O_KEEP_STATE: /* bidir rule */ 946 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule); 947 break; 948 949 case O_LIMIT: /* limit number of sessions */ 950 { 951 u_int16_t limit_mask = cmd->limit_mask; 952 struct ipfw_flow_id id; 953 ipfw_dyn_rule *parent; 954 955 DEB(printf("installing dyn-limit rule %d\n", cmd->conn_limit);) 956 957 id.dst_ip = id.src_ip = 0; 958 id.dst_port = id.src_port = 0; 959 id.proto = args->f_id.proto; 960 961 if (limit_mask & DYN_SRC_ADDR) 962 id.src_ip = args->f_id.src_ip; 963 if (limit_mask & DYN_DST_ADDR) 964 id.dst_ip = args->f_id.dst_ip; 965 if (limit_mask & DYN_SRC_PORT) 966 id.src_port = args->f_id.src_port; 967 if (limit_mask & DYN_DST_PORT) 968 id.dst_port = args->f_id.dst_port; 969 parent = lookup_dyn_parent(&id, rule); 970 if (parent == NULL) { 971 printf("add parent failed\n"); 972 return 1; 973 } 974 if (parent->count >= cmd->conn_limit) { 975 /* 976 * See if we can remove some expired rule. 977 */ 978 remove_dyn_rule(rule, parent); 979 if (parent->count >= cmd->conn_limit) { 980 if (fw_verbose && last_log != time_second) { 981 last_log = time_second; 982 printf( 983 "drop session, too many entries\n"); 984 } 985 return 1; 986 } 987 } 988 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent); 989 } 990 break; 991 default: 992 printf("unknown dynamic rule type %u\n", cmd->o.opcode); 993 return 1; 994 } 995 lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */ 996 return 0; 997} 998 999/* 1000 * Transmit a TCP packet, containing either a RST or a keepalive. 1001 * When flags & TH_RST, we are sending a RST packet, because of a 1002 * "reset" action matched the packet. 1003 * Otherwise we are sending a keepalive, and flags & TH_ 1004 */ 1005static void 1006send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags) 1007{ 1008 struct mbuf *m; 1009 struct ip *ip; 1010 struct tcphdr *tcp; 1011 struct route sro; /* fake route */ 1012 1013 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1014 if (m == 0) 1015 return; 1016 m->m_pkthdr.rcvif = (struct ifnet *)0; 1017 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); 1018 m->m_data += max_linkhdr; 1019 1020 ip = mtod(m, struct ip *); 1021 bzero(ip, m->m_len); 1022 tcp = (struct tcphdr *)(ip + 1); /* no IP options */ 1023 ip->ip_p = IPPROTO_TCP; 1024 tcp->th_off = 5; 1025 /* 1026 * Assume we are sending a RST (or a keepalive in the reverse 1027 * direction), swap src and destination addresses and ports. 1028 */ 1029 ip->ip_src.s_addr = htonl(id->dst_ip); 1030 ip->ip_dst.s_addr = htonl(id->src_ip); 1031 tcp->th_sport = htons(id->dst_port); 1032 tcp->th_dport = htons(id->src_port); 1033 if (flags & TH_RST) { /* we are sending a RST */ 1034 if (flags & TH_ACK) { 1035 tcp->th_seq = htonl(ack); 1036 tcp->th_ack = htonl(0); 1037 tcp->th_flags = TH_RST; 1038 } else { 1039 if (flags & TH_SYN) 1040 seq++; 1041 tcp->th_seq = htonl(0); 1042 tcp->th_ack = htonl(seq); 1043 tcp->th_flags = TH_RST | TH_ACK; 1044 } 1045 } else { 1046 /* 1047 * We are sending a keepalive. flags & TH_SYN determines 1048 * the direction, forward if set, reverse if clear. 1049 * NOTE: seq and ack are always assumed to be correct 1050 * as set by the caller. This may be confusing... 1051 */ 1052 if (flags & TH_SYN) { 1053 /* 1054 * we have to rewrite the correct addresses! 1055 */ 1056 ip->ip_dst.s_addr = htonl(id->dst_ip); 1057 ip->ip_src.s_addr = htonl(id->src_ip); 1058 tcp->th_dport = htons(id->dst_port); 1059 tcp->th_sport = htons(id->src_port); 1060 } 1061 tcp->th_seq = htonl(seq); 1062 tcp->th_ack = htonl(ack); 1063 tcp->th_flags = TH_ACK; 1064 } 1065 /* 1066 * set ip_len to the payload size so we can compute 1067 * the tcp checksum on the pseudoheader 1068 * XXX check this, could save a couple of words ? 1069 */ 1070 ip->ip_len = htons(sizeof(struct tcphdr)); 1071 tcp->th_sum = in_cksum(m, m->m_pkthdr.len); 1072 /* 1073 * now fill fields left out earlier 1074 */ 1075 ip->ip_ttl = ip_defttl; 1076 ip->ip_len = m->m_pkthdr.len; 1077 bzero (&sro, sizeof (sro)); 1078 ip_rtaddr(ip->ip_dst, &sro); 1079 ip_output(m, NULL, &sro, 0, NULL); 1080 if (sro.ro_rt) 1081 RTFREE(sro.ro_rt); 1082} 1083 1084/* 1085 * sends a reject message, consuming the mbuf passed as an argument. 1086 */ 1087static void 1088send_reject(struct ip_fw_args *args, int code, int offset, int ip_len) 1089{ 1090printf("+++ send reject\n"); 1091 if (code != ICMP_REJECT_RST) /* Send an ICMP unreach */ 1092 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); 1093 else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) { 1094 struct tcphdr *const tcp = 1095 L3HDR(struct tcphdr, mtod(args->m, struct ip *)); 1096 if ( (tcp->th_flags & TH_RST) == 0) 1097 send_pkt(&(args->f_id), ntohl(tcp->th_seq), 1098 ntohl(tcp->th_ack), 1099 tcp->th_flags | TH_RST); 1100 m_freem(args->m); 1101 } else 1102 m_freem(args->m); 1103 args->m = NULL; 1104} 1105 1106/** 1107 * 1108 * Given an ip_fw *, lookup_next_rule will return a pointer 1109 * to the next rule, which can be either the jump 1110 * target (for skipto instructions) or the next one in the list (in 1111 * all other cases including a missing jump target). 1112 * The result is also written in the "next_rule" field of the rule. 1113 * Backward jumps are not allowed, so start looking from the next 1114 * rule... 1115 * 1116 * This never returns NULL -- in case we do not have an exact match, 1117 * the next rule is returned. When the ruleset is changed, 1118 * pointers are flushed so we are always correct. 1119 */ 1120 1121static struct ip_fw * 1122lookup_next_rule(struct ip_fw *me) 1123{ 1124 struct ip_fw *rule = NULL; 1125 ipfw_insn *cmd; 1126 1127 /* look for action, in case it is a skipto */ 1128 cmd = ACTION_PTR(me); 1129 if ( cmd->opcode == O_SKIPTO ) 1130 for (rule = me->next; rule ; rule = rule->next) 1131 if (rule->rulenum >= cmd->arg1) 1132 break; 1133 if (rule == NULL) /* failure or not a skipto */ 1134 rule = me->next; 1135 me->next_rule = rule; 1136 return rule; 1137} 1138 1139/* 1140 * The main check routine for the firewall. 1141 * 1142 * All arguments are in args so we can modify them and return them 1143 * back to the caller. 1144 * 1145 * Parameters: 1146 * 1147 * args->m (in/out) The packet; we set to NULL when/if we nuke it. 1148 * Starts with the IP header. 1149 * args->eh (in) Mac header if present, or NULL for layer3 packet. 1150 * args->oif Outgoing interface, or NULL if packet is incoming. 1151 * The incoming interface is in the mbuf. (in) 1152 * args->divert_rule (in/out) 1153 * Skip up to the first rule past this rule number; 1154 * upon return, non-zero port number for divert or tee. 1155 * 1156 * args->rule Pointer to the last matching rule (in/out) 1157 * args->next_hop Socket we are forwarding to (out). 1158 * args->f_id Addresses grabbed from the packet (out) 1159 * 1160 * Return value: 1161 * 1162 * IP_FW_PORT_DENY_FLAG the packet must be dropped. 1163 * 0 The packet is to be accepted and routed normally OR 1164 * the packet was denied/rejected and has been dropped; 1165 * in the latter case, *m is equal to NULL upon return. 1166 * port Divert the packet to port, with these caveats: 1167 * 1168 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead 1169 * of diverting it (ie, 'ipfw tee'). 1170 * 1171 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower 1172 * 16 bits as a dummynet pipe number instead of diverting 1173 */ 1174 1175static int 1176ipfw_chk(struct ip_fw_args *args) 1177{ 1178 /* 1179 * Local variables hold state during the processing of a packet. 1180 * 1181 * IMPORTANT NOTE: to speed up the processing of rules, there 1182 * are some assumption on the values of the variables, which 1183 * are documented here. Should you change them, please check 1184 * the implementation of the various instructions to make sure 1185 * that they still work. 1186 */ 1187 /* 1188 * args->eh The MAC header. It is non-null for a layer2 1189 * packet, it is NULL for a layer-3 packet. 1190 * 1191 * m | args->m Pointer to the mbuf, as received from the caller. 1192 * It may change if ipfw_chk() does an m_pullup, or if it 1193 * consumes the packet because it calls send_reject(). 1194 * XXX This has to change, so that ipfw_chk() never modifies 1195 * or consumes the buffer. 1196 * ip is simply an alias of the value of m, and it is kept 1197 * in sync with it (the packet is supposed to start with 1198 * the ip header). 1199 */ 1200 struct mbuf *m = args->m; 1201 struct ip *ip = mtod(m, struct ip *); 1202 1203 /* 1204 * oif | args->oif If NULL, ipfw_chk has been called on the 1205 * inbound path (ether_input, bdg_forward, ip_input). 1206 * If non-NULL, ipfw_chk has been called on the outbound path 1207 * (ether_output, ip_output). 1208 */ 1209 struct ifnet *oif = args->oif; 1210 1211 struct ip_fw *f = NULL; /* matching rule */ 1212 int retval = 0; 1213 1214 /* 1215 * hlen The length of the IPv4 header. 1216 * hlen >0 means we have an IPv4 packet. 1217 */ 1218 u_int hlen = 0; /* hlen >0 means we have an IP pkt */ 1219 1220 /* 1221 * offset The offset of a fragment. offset != 0 means that 1222 * we have a fragment at this offset of an IPv4 packet. 1223 * offset == 0 means that (if this is an IPv4 packet) 1224 * this is the first or only fragment. 1225 */ 1226 u_short offset = 0; 1227 1228 /* 1229 * Local copies of addresses. They are only valid if we have 1230 * an IP packet. 1231 * 1232 * proto The protocol. Set to 0 for non-ip packets, 1233 * or to the protocol read from the packet otherwise. 1234 * proto != 0 means that we have an IPv4 packet. 1235 * 1236 * src_port, dst_port port numbers, in HOST format. Only 1237 * valid for TCP and UDP packets. 1238 * 1239 * src_ip, dst_ip ip addresses, in NETWORK format. 1240 * Only valid for IPv4 packets. 1241 */ 1242 u_int8_t proto; 1243 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */ 1244 struct in_addr src_ip, dst_ip; /* NOTE: network format */ 1245 u_int16_t ip_len=0; 1246 int dyn_dir = MATCH_UNKNOWN; 1247 ipfw_dyn_rule *q = NULL; 1248 1249 /* 1250 * dyn_dir = MATCH_UNKNOWN when rules unchecked, 1251 * MATCH_NONE when checked and not matched (q = NULL), 1252 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL) 1253 */ 1254 1255 if (args->eh == NULL || /* layer 3 packet */ 1256 ( m->m_pkthdr.len >= sizeof(struct ip) && 1257 ntohs(args->eh->ether_type) == ETHERTYPE_IP)) 1258 hlen = ip->ip_hl << 2; 1259 1260 /* 1261 * Collect parameters into local variables for faster matching. 1262 */ 1263 if (hlen == 0) { /* do not grab addresses for non-ip pkts */ 1264 proto = args->f_id.proto = 0; /* mark f_id invalid */ 1265 goto after_ip_checks; 1266 } 1267 1268 proto = args->f_id.proto = ip->ip_p; 1269 src_ip = ip->ip_src; 1270 dst_ip = ip->ip_dst; 1271 if (args->eh != NULL) { /* layer 2 packets are as on the wire */ 1272 offset = ntohs(ip->ip_off) & IP_OFFMASK; 1273 ip_len = ntohs(ip->ip_len); 1274 } else { 1275 offset = ip->ip_off & IP_OFFMASK; 1276 ip_len = ip->ip_len; 1277 } 1278 1279#define PULLUP_TO(len) \ 1280 do { \ 1281 if ((m)->m_len < (len)) { \ 1282 args->m = m = m_pullup(m, (len)); \ 1283 if (m == 0) \ 1284 goto pullup_failed; \ 1285 ip = mtod(m, struct ip *); \ 1286 } \ 1287 } while (0) 1288 1289 if (offset == 0) { 1290 switch (proto) { 1291 case IPPROTO_TCP: 1292 { 1293 struct tcphdr *tcp; 1294 1295 PULLUP_TO(hlen + sizeof(struct tcphdr)); 1296 tcp = L3HDR(struct tcphdr, ip); 1297 dst_port = tcp->th_dport; 1298 src_port = tcp->th_sport; 1299 args->f_id.flags = tcp->th_flags; 1300 } 1301 break; 1302 1303 case IPPROTO_UDP: 1304 { 1305 struct udphdr *udp; 1306 1307 PULLUP_TO(hlen + sizeof(struct udphdr)); 1308 udp = L3HDR(struct udphdr, ip); 1309 dst_port = udp->uh_dport; 1310 src_port = udp->uh_sport; 1311 } 1312 break; 1313 1314 case IPPROTO_ICMP: 1315 PULLUP_TO(hlen + 4); /* type, code and checksum. */ 1316 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type; 1317 break; 1318 1319 default: 1320 break; 1321 } 1322#undef PULLUP_TO 1323 } 1324 1325 args->f_id.src_ip = ntohl(src_ip.s_addr); 1326 args->f_id.dst_ip = ntohl(dst_ip.s_addr); 1327 args->f_id.src_port = src_port = ntohs(src_port); 1328 args->f_id.dst_port = dst_port = ntohs(dst_port); 1329 1330after_ip_checks: 1331 if (args->rule) { 1332 /* 1333 * Packet has already been tagged. Look for the next rule 1334 * to restart processing. 1335 * 1336 * If fw_one_pass != 0 then just accept it. 1337 * XXX should not happen here, but optimized out in 1338 * the caller. 1339 */ 1340 if (fw_one_pass) 1341 return 0; 1342 1343 f = args->rule->next_rule; 1344 if (f == NULL) 1345 f = lookup_next_rule(args->rule); 1346 } else { 1347 /* 1348 * Find the starting rule. It can be either the first 1349 * one, or the one after divert_rule if asked so. 1350 */ 1351 int skipto = args->divert_rule; 1352 1353 f = layer3_chain; 1354 if (args->eh == NULL && skipto != 0) { 1355 if (skipto >= IPFW_DEFAULT_RULE) 1356 return(IP_FW_PORT_DENY_FLAG); /* invalid */ 1357 while (f && f->rulenum <= skipto) 1358 f = f->next; 1359 if (f == NULL) /* drop packet */ 1360 return(IP_FW_PORT_DENY_FLAG); 1361 } 1362 } 1363 args->divert_rule = 0; /* reset to avoid confusion later */ 1364 1365 /* 1366 * Now scan the rules, and parse microinstructions for each rule. 1367 */ 1368 for (; f; f = f->next) { 1369 int l, cmdlen; 1370 ipfw_insn *cmd; 1371 int skip_or; /* skip rest of OR block */ 1372 1373again: 1374 skip_or = 0; 1375 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ; 1376 l -= cmdlen, cmd += cmdlen) { 1377 int match; 1378 1379 /* 1380 * check_body is a jump target used when we find a 1381 * CHECK_STATE, and need to jump to the body of 1382 * the target rule. 1383 */ 1384 1385check_body: 1386 cmdlen = F_LEN(cmd); 1387 /* 1388 * An OR block (insn_1 || .. || insn_n) has the 1389 * F_OR bit set in all but the last instruction. 1390 * The first match will set "skip_or", and cause 1391 * the following instructions to be skipped until 1392 * past the one with the F_OR bit clear. 1393 */ 1394 if (skip_or) { /* skip this instruction */ 1395 if ((cmd->len & F_OR) == 0) 1396 skip_or = 0; /* next one is good */ 1397 continue; 1398 } 1399 match = 0; /* set to 1 if we succeed */ 1400 1401 switch (cmd->opcode) { 1402 /* 1403 * The first set of opcodes compares the packet's 1404 * fields with some pattern, setting 'match' if a 1405 * match is found. At the end of the loop there is 1406 * logic to deal with F_NOT and F_OR flags associated 1407 * with the opcode. 1408 */ 1409 case O_NOP: 1410 match = 1; 1411 break; 1412 1413 case O_FORWARD_MAC: 1414 printf("ipfw: opcode %d unimplemented\n", 1415 cmd->opcode); 1416 break; 1417 1418 case O_GID: 1419 case O_UID: 1420 /* 1421 * We only check offset == 0 && proto != 0, 1422 * as this ensures that we have an IPv4 1423 * packet with the ports info. 1424 */ 1425 if (offset!=0) 1426 break; 1427 { 1428 struct inpcbinfo *pi; 1429 int wildcard; 1430 struct inpcb *pcb; 1431 1432 if (proto == IPPROTO_TCP) { 1433 wildcard = 0; 1434 pi = &tcbinfo; 1435 } else if (proto == IPPROTO_UDP) { 1436 wildcard = 1; 1437 pi = &udbinfo; 1438 } else 1439 break; 1440 1441 pcb = (oif) ? 1442 in_pcblookup_hash(pi, 1443 dst_ip, htons(dst_port), 1444 src_ip, htons(src_port), 1445 wildcard, oif) : 1446 in_pcblookup_hash(pi, 1447 src_ip, htons(src_port), 1448 dst_ip, htons(dst_port), 1449 wildcard, NULL); 1450 1451 if (pcb == NULL || pcb->inp_socket == NULL) 1452 break; 1453#if __FreeBSD_version < 500034 1454#define socheckuid(a,b) ((a)->so_cred->cr_uid == (b)) 1455#endif 1456 if (cmd->opcode == O_UID) { 1457 match = 1458 socheckuid(pcb->inp_socket, 1459 (uid_t)((ipfw_insn_u32 *)cmd)->d[0]); 1460 } else { 1461 match = groupmember( 1462 (uid_t)((ipfw_insn_u32 *)cmd)->d[0], 1463 pcb->inp_socket->so_cred); 1464 } 1465 } 1466 break; 1467 1468 case O_RECV: 1469 match = iface_match(m->m_pkthdr.rcvif, 1470 (ipfw_insn_if *)cmd); 1471 break; 1472 1473 case O_XMIT: 1474 match = iface_match(oif, (ipfw_insn_if *)cmd); 1475 break; 1476 1477 case O_VIA: 1478 match = iface_match(oif ? oif : 1479 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd); 1480 break; 1481 1482 case O_MACADDR2: 1483 if (args->eh != NULL) { /* have MAC header */ 1484 u_int32_t *want = (u_int32_t *) 1485 ((ipfw_insn_mac *)cmd)->addr; 1486 u_int32_t *mask = (u_int32_t *) 1487 ((ipfw_insn_mac *)cmd)->mask; 1488 u_int32_t *hdr = (u_int32_t *)args->eh; 1489 1490 match = 1491 ( want[0] == (hdr[0] & mask[0]) && 1492 want[1] == (hdr[1] & mask[1]) && 1493 want[2] == (hdr[2] & mask[2]) ); 1494 } 1495 break; 1496 1497 case O_MAC_TYPE: 1498 if (args->eh != NULL) { 1499 u_int16_t t = 1500 ntohs(args->eh->ether_type); 1501 u_int16_t *p = 1502 ((ipfw_insn_u16 *)cmd)->ports; 1503 int i; 1504 1505 for (i = cmdlen - 1; !match && i>0; 1506 i--, p += 2) 1507 match = (t>=p[0] && t<=p[1]); 1508 } 1509 break; 1510 1511 case O_FRAG: 1512 match = (hlen > 0 && offset != 0); 1513 break; 1514 1515 case O_IN: /* "out" is "not in" */ 1516 match = (oif == NULL); 1517 break; 1518 1519 case O_LAYER2: 1520 match = (args->eh != NULL); 1521 break; 1522 1523 case O_PROTO: 1524 /* 1525 * We do not allow an arg of 0 so the 1526 * check of "proto" only suffices. 1527 */ 1528 match = (proto == cmd->arg1); 1529 break; 1530 1531 case O_IP_SRC: 1532 match = (hlen > 0 && 1533 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1534 src_ip.s_addr); 1535 break; 1536 1537 case O_IP_SRC_MASK: 1538 match = (hlen > 0 && 1539 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1540 (src_ip.s_addr & 1541 ((ipfw_insn_ip *)cmd)->mask.s_addr)); 1542 break; 1543 1544 case O_IP_SRC_ME: 1545 if (hlen > 0) { 1546 struct ifnet *tif; 1547 1548 INADDR_TO_IFP(src_ip, tif); 1549 match = (tif != NULL); 1550 } 1551 break; 1552 1553 case O_IP_DST_SET: 1554 case O_IP_SRC_SET: 1555 if (hlen > 0) { 1556 u_int32_t *d = (u_int32_t *)(cmd+1); 1557 u_int32_t addr = 1558 cmd->opcode == O_IP_DST_SET ? 1559 args->f_id.src_ip : 1560 args->f_id.dst_ip; 1561 1562 if (addr < d[0]) 1563 break; 1564 addr -= d[0]; /* subtract base */ 1565 match = (addr < cmd->arg1) && 1566 ( d[ 1 + (addr>>5)] & 1567 (1<<(addr & 0x1f)) ); 1568 } 1569 break; 1570 1571 case O_IP_DST: 1572 match = (hlen > 0 && 1573 ((ipfw_insn_ip *)cmd)->addr.s_addr == 1574 dst_ip.s_addr); 1575 break; 1576 1577 case O_IP_DST_MASK: 1578 match = (hlen > 0) && 1579 (((ipfw_insn_ip *)cmd)->addr.s_addr == 1580 (dst_ip.s_addr & 1581 ((ipfw_insn_ip *)cmd)->mask.s_addr)); 1582 break; 1583 1584 case O_IP_DST_ME: 1585 if (hlen > 0) { 1586 struct ifnet *tif; 1587 1588 INADDR_TO_IFP(dst_ip, tif); 1589 match = (tif != NULL); 1590 } 1591 break; 1592 1593 case O_IP_SRCPORT: 1594 case O_IP_DSTPORT: 1595 /* 1596 * offset == 0 && proto != 0 is enough 1597 * to guarantee that we have an IPv4 1598 * packet with port info. 1599 */ 1600 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP) 1601 && offset == 0) { 1602 u_int16_t x = 1603 (cmd->opcode == O_IP_SRCPORT) ? 1604 src_port : dst_port ; 1605 u_int16_t *p = 1606 ((ipfw_insn_u16 *)cmd)->ports; 1607 int i; 1608 1609 for (i = cmdlen - 1; !match && i>0; 1610 i--, p += 2) 1611 match = (x>=p[0] && x<=p[1]); 1612 } 1613 break; 1614 1615 case O_ICMPTYPE: 1616 match = (offset == 0 && proto==IPPROTO_ICMP && 1617 icmptype_match(ip, (ipfw_insn_u32 *)cmd) ); 1618 break; 1619 1620 case O_IPOPT: 1621 match = (hlen > 0 && ipopts_match(ip, cmd) ); 1622 break; 1623 1624 case O_IPVER: 1625 match = (hlen > 0 && cmd->arg1 == ip->ip_v); 1626 break; 1627 1628 case O_IPTTL: 1629 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl); 1630 break; 1631 1632 case O_IPID: 1633 match = (hlen > 0 && 1634 cmd->arg1 == ntohs(ip->ip_id)); 1635 break; 1636 1637 case O_IPLEN: 1638 match = (hlen > 0 && cmd->arg1 == ip_len); 1639 break; 1640 1641 case O_IPPRECEDENCE: 1642 match = (hlen > 0 && 1643 (cmd->arg1 == (ip->ip_tos & 0xe0)) ); 1644 break; 1645 1646 case O_IPTOS: 1647 match = (hlen > 0 && 1648 flags_match(cmd, ip->ip_tos)); 1649 break; 1650 1651 case O_TCPFLAGS: 1652 match = (proto == IPPROTO_TCP && offset == 0 && 1653 flags_match(cmd, 1654 L3HDR(struct tcphdr,ip)->th_flags)); 1655 break; 1656 1657 case O_TCPOPTS: 1658 match = (proto == IPPROTO_TCP && offset == 0 && 1659 tcpopts_match(ip, cmd)); 1660 break; 1661 1662 case O_TCPSEQ: 1663 match = (proto == IPPROTO_TCP && offset == 0 && 1664 ((ipfw_insn_u32 *)cmd)->d[0] == 1665 L3HDR(struct tcphdr,ip)->th_seq); 1666 break; 1667 1668 case O_TCPACK: 1669 match = (proto == IPPROTO_TCP && offset == 0 && 1670 ((ipfw_insn_u32 *)cmd)->d[0] == 1671 L3HDR(struct tcphdr,ip)->th_ack); 1672 break; 1673 1674 case O_TCPWIN: 1675 match = (proto == IPPROTO_TCP && offset == 0 && 1676 cmd->arg1 == 1677 L3HDR(struct tcphdr,ip)->th_win); 1678 break; 1679 1680 case O_ESTAB: 1681 /* reject packets which have SYN only */ 1682 /* XXX should i also check for TH_ACK ? */ 1683 match = (proto == IPPROTO_TCP && offset == 0 && 1684 (L3HDR(struct tcphdr,ip)->th_flags & 1685 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); 1686 break; 1687 1688 case O_LOG: 1689 if (fw_verbose) 1690 ipfw_log(f, hlen, args->eh, m, oif); 1691 match = 1; 1692 break; 1693 1694 case O_PROB: 1695 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]); 1696 break; 1697 1698 /* 1699 * The second set of opcodes represents 'actions', 1700 * i.e. the terminal part of a rule once the packet 1701 * matches all previous patterns. 1702 * Typically there is only one action for each rule, 1703 * and the opcode is stored at the end of the rule 1704 * (but there are exceptions -- see below). 1705 * 1706 * In general, here we set retval and terminate the 1707 * outer loop (would be a 'break 3' in some language, 1708 * but we need to do a 'goto done'). 1709 * 1710 * Exceptions: 1711 * O_COUNT and O_SKIPTO actions: 1712 * instead of terminating, we jump to the next rule 1713 * ('goto next_rule', equivalent to a 'break 2'), 1714 * or to the SKIPTO target ('goto again' after 1715 * having set f, cmd and l), respectively. 1716 * 1717 * O_LIMIT and O_KEEP_STATE: these opcodes are 1718 * not real 'actions', and are stored right 1719 * before the 'action' part of the rule. 1720 * These opcodes try to install an entry in the 1721 * state tables; if successful, we continue with 1722 * the next opcode (match=1; break;), otherwise 1723 * the packet * must be dropped 1724 * ('goto done' after setting retval); 1725 * 1726 * O_PROBE_STATE and O_CHECK_STATE: these opcodes 1727 * cause a lookup of the state table, and a jump 1728 * to the 'action' part of the parent rule 1729 * ('goto check_body') if an entry is found, or 1730 * (CHECK_STATE only) a jump to the next rule if 1731 * the entry is not found ('goto next_rule'). 1732 * The result of the lookup is cached to make 1733 * further instances of these opcodes are 1734 * effectively NOPs. 1735 */ 1736 case O_LIMIT: 1737 case O_KEEP_STATE: 1738 if (install_state(f, 1739 (ipfw_insn_limit *)cmd, args)) { 1740 retval = IP_FW_PORT_DENY_FLAG; 1741 goto done; /* error/limit violation */ 1742 } 1743 match = 1; 1744 break; 1745 1746 case O_PROBE_STATE: 1747 case O_CHECK_STATE: 1748 /* 1749 * dynamic rules are checked at the first 1750 * keep-state or check-state occurrence, 1751 * with the result being stored in dyn_dir. 1752 * The compiler introduces a PROBE_STATE 1753 * instruction for us when we have a 1754 * KEEP_STATE (because PROBE_STATE needs 1755 * to be run first). 1756 */ 1757 if (dyn_dir == MATCH_UNKNOWN && 1758 (q = lookup_dyn_rule(&args->f_id, 1759 &dyn_dir, proto == IPPROTO_TCP ? 1760 L3HDR(struct tcphdr, ip) : NULL)) 1761 != NULL) { 1762 /* 1763 * Found dynamic entry, update stats 1764 * and jump to the 'action' part of 1765 * the parent rule. 1766 */ 1767 q->pcnt++; 1768 q->bcnt += ip_len; 1769 f = q->rule; 1770 cmd = ACTION_PTR(f); 1771 l = f->cmd_len - f->act_ofs; 1772 goto check_body; 1773 } 1774 /* 1775 * Dynamic entry not found. If CHECK_STATE, 1776 * skip to next rule, if PROBE_STATE just 1777 * ignore and continue with next opcode. 1778 */ 1779 if (cmd->opcode == O_CHECK_STATE) 1780 goto next_rule; 1781 match = 1; 1782 break; 1783 1784 case O_ACCEPT: 1785 retval = 0; /* accept */ 1786 goto done; 1787 1788 case O_PIPE: 1789 case O_QUEUE: 1790 args->rule = f; /* report matching rule */ 1791 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG; 1792 goto done; 1793 1794 case O_DIVERT: 1795 case O_TEE: 1796 if (args->eh) /* not on layer 2 */ 1797 break; 1798 args->divert_rule = f->rulenum; 1799 retval = (cmd->opcode == O_DIVERT) ? 1800 cmd->arg1 : 1801 cmd->arg1 | IP_FW_PORT_TEE_FLAG; 1802 goto done; 1803 1804 case O_COUNT: 1805 case O_SKIPTO: 1806 f->pcnt++; /* update stats */ 1807 f->bcnt += ip_len; 1808 f->timestamp = time_second; 1809 if (cmd->opcode == O_COUNT) 1810 goto next_rule; 1811 /* handle skipto */ 1812 if (f->next_rule == NULL) 1813 lookup_next_rule(f); 1814 f = f->next_rule; 1815 goto again; 1816 1817 case O_REJECT: 1818 /* 1819 * Drop the packet and send a reject notice 1820 * if the packet is not ICMP (or is an ICMP 1821 * query), and it is not multicast/broadcast. 1822 */ 1823 if (hlen > 0 && 1824 (proto != IPPROTO_ICMP || 1825 is_icmp_query(ip)) && 1826 !(m->m_flags & (M_BCAST|M_MCAST)) && 1827 !IN_MULTICAST(dst_ip.s_addr)) { 1828 send_reject(args, cmd->arg1, 1829 offset,ip_len); 1830 m = args->m; 1831 } 1832 /* FALLTHROUGH */ 1833 case O_DENY: 1834 retval = IP_FW_PORT_DENY_FLAG; 1835 goto done; 1836 1837 case O_FORWARD_IP: 1838 if (args->eh) /* not valid on layer2 pkts */ 1839 break; 1840 if (!q || dyn_dir == MATCH_FORWARD) 1841 args->next_hop = 1842 &((ipfw_insn_sa *)cmd)->sa; 1843 retval = 0; 1844 goto done; 1845 1846 default: 1847 panic("-- unknown opcode %d\n", cmd->opcode); 1848 } /* end of switch() on opcodes */ 1849 1850 if (cmd->len & F_NOT) 1851 match = !match; 1852 1853 if (match) { 1854 if (cmd->len & F_OR) 1855 skip_or = 1; 1856 } else { 1857 if (!(cmd->len & F_OR)) /* not an OR block, */ 1858 break; /* try next rule */ 1859 } 1860 1861 } /* end of inner for, scan opcodes */ 1862 1863next_rule:; /* try next rule */ 1864 1865 } /* end of outer for, scan rules */ 1866 1867done: 1868 /* Update statistics */ 1869 f->pcnt++; 1870 f->bcnt += ip_len; 1871 f->timestamp = time_second; 1872 return retval; 1873 1874pullup_failed: 1875 if (fw_verbose) 1876 printf("pullup failed\n"); 1877 return(IP_FW_PORT_DENY_FLAG); 1878} 1879 1880/* 1881 * When a rule is added/deleted, clear the next_rule pointers in all rules. 1882 * These will be reconstructed on the fly as packets are matched. 1883 * Must be called at splimp(). 1884 */ 1885static void 1886flush_rule_ptrs(void) 1887{ 1888 struct ip_fw *rule; 1889 1890 for (rule = layer3_chain; rule; rule = rule->next) 1891 rule->next_rule = NULL; 1892} 1893 1894/* 1895 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given 1896 * pipe/queue, or to all of them (match == NULL). 1897 * Must be called at splimp(). 1898 */ 1899void 1900flush_pipe_ptrs(struct dn_flow_set *match) 1901{ 1902 struct ip_fw *rule; 1903 1904 for (rule = layer3_chain; rule; rule = rule->next) { 1905 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule); 1906 1907 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE) 1908 continue; 1909 1910 if (match == NULL || cmd->pipe_ptr == match) 1911 cmd->pipe_ptr = NULL; 1912 } 1913} 1914 1915/* 1916 * Add a new rule to the list. Copy the rule into a malloc'ed area, then 1917 * possibly create a rule number and add the rule to the list. 1918 * Update the rule_number in the input struct so the caller knows it as well. 1919 */ 1920static int 1921add_rule(struct ip_fw **head, struct ip_fw *input_rule) 1922{ 1923 struct ip_fw *rule, *f, *prev; 1924 int s; 1925 int l = RULESIZE(input_rule); 1926 1927 if (*head == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE) 1928 return (EINVAL); 1929 1930 rule = malloc(l, M_IPFW, M_DONTWAIT | M_ZERO); 1931 if (rule == NULL) 1932 return (ENOSPC); 1933 1934 bcopy(input_rule, rule, l); 1935 1936 rule->next = NULL; 1937 rule->next_rule = NULL; 1938 1939 rule->pcnt = 0; 1940 rule->bcnt = 0; 1941 rule->timestamp = 0; 1942 1943 s = splimp(); 1944 1945 if (*head == NULL) { /* default rule */ 1946 *head = rule; 1947 goto done; 1948 } 1949 1950 /* 1951 * If rulenum is 0, find highest numbered rule before the 1952 * default rule, and add autoinc_step 1953 */ 1954 if (autoinc_step < 1) 1955 autoinc_step = 1; 1956 else if (autoinc_step > 1000) 1957 autoinc_step = 1000; 1958 if (rule->rulenum == 0) { 1959 /* 1960 * locate the highest numbered rule before default 1961 */ 1962 for (f = *head; f; f = f->next) { 1963 if (f->rulenum == IPFW_DEFAULT_RULE) 1964 break; 1965 rule->rulenum = f->rulenum; 1966 } 1967 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step) 1968 rule->rulenum += autoinc_step; 1969 input_rule->rulenum = rule->rulenum; 1970 } 1971 1972 /* 1973 * Now insert the new rule in the right place in the sorted list. 1974 */ 1975 for (prev = NULL, f = *head; f; prev = f, f = f->next) { 1976 if (f->rulenum > rule->rulenum) { /* found the location */ 1977 if (prev) { 1978 rule->next = f; 1979 prev->next = rule; 1980 } else { /* head insert */ 1981 rule->next = *head; 1982 *head = rule; 1983 } 1984 break; 1985 } 1986 } 1987 flush_rule_ptrs(); 1988done: 1989 static_count++; 1990 static_len += l; 1991 splx(s); 1992 DEB(printf("++ installed rule %d, static count now %d\n", 1993 rule->rulenum, static_count);) 1994 return (0); 1995} 1996 1997/** 1998 * Free storage associated with a static rule (including derived 1999 * dynamic rules). 2000 * The caller is in charge of clearing rule pointers to avoid 2001 * dangling pointers. 2002 * @return a pointer to the next entry. 2003 * Arguments are not checked, so they better be correct. 2004 * Must be called at splimp(). 2005 */ 2006static struct ip_fw * 2007delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule) 2008{ 2009 struct ip_fw *n; 2010 int l = RULESIZE(rule); 2011 2012 n = rule->next; 2013 remove_dyn_rule(rule, NULL /* force removal */); 2014 if (prev == NULL) 2015 *head = n; 2016 else 2017 prev->next = n; 2018 static_count--; 2019 static_len -= l; 2020 2021 if (DUMMYNET_LOADED) 2022 ip_dn_ruledel_ptr(rule); 2023 free(rule, M_IPFW); 2024 return n; 2025} 2026 2027/* 2028 * Deletes all rules from a chain (including the default rule 2029 * if the second argument is set). 2030 * Must be called at splimp(). 2031 */ 2032static void 2033free_chain(struct ip_fw **chain, int kill_default) 2034{ 2035 struct ip_fw *rule; 2036 2037 flush_rule_ptrs(); /* more efficient to do outside the loop */ 2038 2039 while ( (rule = *chain) != NULL && 2040 (kill_default || rule->rulenum != IPFW_DEFAULT_RULE) ) 2041 delete_rule(chain, NULL, rule); 2042} 2043 2044/** 2045 * Remove all rules with given number. 2046 */ 2047static int 2048del_entry(struct ip_fw **chain, u_short rulenum) 2049{ 2050 struct ip_fw *prev, *rule; 2051 int s; 2052 2053 if (rulenum == IPFW_DEFAULT_RULE) 2054 return EINVAL; 2055 2056 /* 2057 * locate first rule to delete 2058 */ 2059 for (prev = NULL, rule = *chain; rule && rule->rulenum < rulenum; 2060 prev = rule, rule = rule->next) 2061 ; 2062 if (rule->rulenum != rulenum) 2063 return EINVAL; 2064 2065 s = splimp(); /* no access to rules while removing */ 2066 flush_rule_ptrs(); /* more efficient to do outside the loop */ 2067 /* 2068 * prev remains the same throughout the cycle 2069 */ 2070 while (rule && rule->rulenum == rulenum) 2071 rule = delete_rule(chain, prev, rule); 2072 splx(s); 2073 return 0; 2074} 2075 2076/* 2077 * Clear counters for a specific rule. 2078 */ 2079static void 2080clear_counters(struct ip_fw *rule, int log_only) 2081{ 2082 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); 2083 2084 if (log_only == 0) { 2085 rule->bcnt = rule->pcnt = 0; 2086 rule->timestamp = 0; 2087 } 2088 if (l->o.opcode == O_LOG) 2089 l->log_left = l->max_log; 2090} 2091 2092/** 2093 * Reset some or all counters on firewall rules. 2094 * @arg frwl is null to clear all entries, or contains a specific 2095 * rule number. 2096 * @arg log_only is 1 if we only want to reset logs, zero otherwise. 2097 */ 2098static int 2099zero_entry(int rulenum, int log_only) 2100{ 2101 struct ip_fw *rule; 2102 int s; 2103 char *msg; 2104 2105 if (rulenum == 0) { 2106 s = splimp(); 2107 norule_counter = 0; 2108 for (rule = layer3_chain; rule; rule = rule->next) 2109 clear_counters(rule, log_only); 2110 splx(s); 2111 msg = log_only ? "ipfw: All logging counts reset.\n" : 2112 "ipfw: Accounting cleared.\n"; 2113 } else { 2114 int cleared = 0; 2115 /* 2116 * We can have multiple rules with the same number, so we 2117 * need to clear them all. 2118 */ 2119 for (rule = layer3_chain; rule; rule = rule->next) 2120 if (rule->rulenum == rulenum) { 2121 s = splimp(); 2122 while (rule && rule->rulenum == rulenum) { 2123 clear_counters(rule, log_only); 2124 rule = rule->next; 2125 } 2126 splx(s); 2127 cleared = 1; 2128 break; 2129 } 2130 if (!cleared) /* we did not find any matching rules */ 2131 return (EINVAL); 2132 msg = log_only ? "ipfw: Entry %d logging count reset.\n" : 2133 "ipfw: Entry %d cleared.\n"; 2134 } 2135 if (fw_verbose) 2136 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum); 2137 return (0); 2138} 2139 2140/* 2141 * Check validity of the structure before insert. 2142 * Fortunately rules are simple, so this mostly need to check rule sizes. 2143 */ 2144static int 2145check_ipfw_struct(struct ip_fw *rule, int size) 2146{ 2147 int l, cmdlen = 0; 2148 int have_action=0; 2149 ipfw_insn *cmd; 2150 2151 if (size < sizeof(*rule)) { 2152 printf("ipfw: rule too short\n"); 2153 return (EINVAL); 2154 } 2155 /* first, check for valid size */ 2156 l = RULESIZE(rule); 2157 if (l != size) { 2158 printf("ipfw: size mismatch (have %d want %d)\n", size, l); 2159 return (EINVAL); 2160 } 2161 /* 2162 * Now go for the individual checks. Very simple ones, basically only 2163 * instruction sizes. 2164 */ 2165 for (l = rule->cmd_len, cmd = rule->cmd ; 2166 l > 0 ; l -= cmdlen, cmd += cmdlen) { 2167 cmdlen = F_LEN(cmd); 2168 if (cmdlen > l) { 2169 printf("ipfw: opcode %d size truncated\n", 2170 cmd->opcode); 2171 return EINVAL; 2172 } 2173 DEB(printf("ipfw: opcode %d\n", cmd->opcode);) 2174 switch (cmd->opcode) { 2175 case O_NOP: 2176 case O_PROBE_STATE: 2177 case O_KEEP_STATE: 2178 case O_PROTO: 2179 case O_IP_SRC_ME: 2180 case O_IP_DST_ME: 2181 case O_LAYER2: 2182 case O_IN: 2183 case O_FRAG: 2184 case O_IPOPT: 2185 case O_IPLEN: 2186 case O_IPID: 2187 case O_IPTOS: 2188 case O_IPPRECEDENCE: 2189 case O_IPTTL: 2190 case O_IPVER: 2191 case O_TCPWIN: 2192 case O_TCPFLAGS: 2193 case O_TCPOPTS: 2194 case O_ESTAB: 2195 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2196 goto bad_size; 2197 break; 2198 2199 case O_UID: 2200 case O_GID: 2201 case O_IP_SRC: 2202 case O_IP_DST: 2203 case O_TCPSEQ: 2204 case O_TCPACK: 2205 case O_PROB: 2206 case O_ICMPTYPE: 2207 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) 2208 goto bad_size; 2209 break; 2210 2211 case O_LIMIT: 2212 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) 2213 goto bad_size; 2214 break; 2215 2216 case O_LOG: 2217 if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) 2218 goto bad_size; 2219 2220 ((ipfw_insn_log *)cmd)->log_left = 2221 ((ipfw_insn_log *)cmd)->max_log; 2222 2223 break; 2224 2225 case O_IP_SRC_MASK: 2226 case O_IP_DST_MASK: 2227 if (cmdlen != F_INSN_SIZE(ipfw_insn_ip)) 2228 goto bad_size; 2229 if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) { 2230 printf("ipfw: opcode %d, useless rule\n", 2231 cmd->opcode); 2232 return EINVAL; 2233 } 2234 break; 2235 2236 case O_IP_SRC_SET: 2237 case O_IP_DST_SET: 2238 if (cmd->arg1 == 0 || cmd->arg1 > 256) { 2239 printf("ipfw: invalid set size %d\n", 2240 cmd->arg1); 2241 return EINVAL; 2242 } 2243 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 2244 (cmd->arg1+31)/32 ) 2245 goto bad_size; 2246 break; 2247 2248 case O_MACADDR2: 2249 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) 2250 goto bad_size; 2251 break; 2252 2253 case O_MAC_TYPE: 2254 case O_IP_SRCPORT: 2255 case O_IP_DSTPORT: /* XXX artificial limit, 15 port pairs */ 2256 if (cmdlen < 2 || cmdlen > 15) 2257 goto bad_size; 2258 break; 2259 2260 case O_RECV: 2261 case O_XMIT: 2262 case O_VIA: 2263 if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) 2264 goto bad_size; 2265 break; 2266 2267 case O_PIPE: 2268 case O_QUEUE: 2269 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe)) 2270 goto bad_size; 2271 goto check_action; 2272 2273 case O_FORWARD_IP: 2274 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) 2275 goto bad_size; 2276 goto check_action; 2277 2278 case O_FORWARD_MAC: /* XXX not implemented yet */ 2279 case O_CHECK_STATE: 2280 case O_COUNT: 2281 case O_ACCEPT: 2282 case O_DENY: 2283 case O_REJECT: 2284 case O_SKIPTO: 2285 case O_DIVERT: 2286 case O_TEE: 2287 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 2288 goto bad_size; 2289check_action: 2290 if (have_action) { 2291 printf("ipfw: opcode %d, multiple actions" 2292 " not allowed\n", 2293 cmd->opcode); 2294 return EINVAL; 2295 } 2296 have_action = 1; 2297 if (l != cmdlen) { 2298 printf("ipfw: opcode %d, action must be" 2299 " last opcode\n", 2300 cmd->opcode); 2301 return EINVAL; 2302 } 2303 break; 2304 default: 2305 printf("ipfw: opcode %d, unknown opcode\n", 2306 cmd->opcode); 2307 return EINVAL; 2308 } 2309 } 2310 if (have_action == 0) { 2311 printf("ipfw: missing action\n"); 2312 return EINVAL; 2313 } 2314 return 0; 2315 2316bad_size: 2317 printf("ipfw: opcode %d size %d wrong\n", 2318 cmd->opcode, cmdlen); 2319 return EINVAL; 2320} 2321 2322 2323/** 2324 * {set|get}sockopt parser. 2325 */ 2326static int 2327ipfw_ctl(struct sockopt *sopt) 2328{ 2329 int error, s, rulenum; 2330 size_t size; 2331 struct ip_fw *bp , *buf, *rule; 2332 2333 static u_int32_t rule_buf[255]; /* we copy the data here */ 2334 2335 /* 2336 * Disallow modifications in really-really secure mode, but still allow 2337 * the logging counters to be reset. 2338 */ 2339 if (sopt->sopt_name == IP_FW_ADD || 2340 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) { 2341#if __FreeBSD_version >= 500034 2342 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 2343 if (error) 2344 return (error); 2345#else /* FreeBSD 4.x */ 2346 if (securelevel >= 3) 2347 return (EPERM); 2348#endif 2349 } 2350 2351 error = 0; 2352 2353 switch (sopt->sopt_name) { 2354 case IP_FW_GET: 2355 /* 2356 * pass up a copy of the current rules. Static rules 2357 * come first (the last of which has number IPFW_DEFAULT_RULE), 2358 * followed by a possibly empty list of dynamic rule. 2359 * The last dynamic rule has NULL in the "next" field. 2360 */ 2361 s = splimp(); 2362 size = static_len; /* size of static rules */ 2363 if (ipfw_dyn_v) /* add size of dyn.rules */ 2364 size += (dyn_count * sizeof(ipfw_dyn_rule)); 2365 2366 /* 2367 * XXX todo: if the user passes a short length just to know 2368 * how much room is needed, do not bother filling up the 2369 * buffer, just jump to the sooptcopyout. 2370 */ 2371 buf = malloc(size, M_TEMP, M_WAITOK); 2372 if (buf == 0) { 2373 splx(s); 2374 error = ENOBUFS; 2375 break; 2376 } 2377 2378 bp = buf; 2379 for (rule = layer3_chain; rule ; rule = rule->next) { 2380 int i = RULESIZE(rule); 2381 bcopy(rule, bp, i); 2382 bp = (struct ip_fw *)((char *)bp + i); 2383 } 2384 if (ipfw_dyn_v) { 2385 int i; 2386 ipfw_dyn_rule *p, *dst, *last = NULL; 2387 2388 dst = (ipfw_dyn_rule *)bp; 2389 for (i = 0 ; i < curr_dyn_buckets ; i++ ) 2390 for ( p = ipfw_dyn_v[i] ; p != NULL ; 2391 p = p->next, dst++ ) { 2392 bcopy(p, dst, sizeof *p); 2393 (int)dst->rule = p->rule->rulenum ; 2394 /* 2395 * store a non-null value in "next". 2396 * The userland code will interpret a 2397 * NULL here as a marker 2398 * for the last dynamic rule. 2399 */ 2400 dst->next = dst ; 2401 last = dst ; 2402 dst->expire = 2403 TIME_LEQ(dst->expire, time_second) ? 2404 0 : dst->expire - time_second ; 2405 } 2406 if (last != NULL) /* mark last dynamic rule */ 2407 last->next = NULL; 2408 } 2409 splx(s); 2410 2411 error = sooptcopyout(sopt, buf, size); 2412 free(buf, M_TEMP); 2413 break; 2414 2415 case IP_FW_FLUSH: 2416 /* 2417 * Normally we cannot release the lock on each iteration. 2418 * We could do it here only because we start from the head all 2419 * the times so there is no risk of missing some entries. 2420 * On the other hand, the risk is that we end up with 2421 * a very inconsistent ruleset, so better keep the lock 2422 * around the whole cycle. 2423 * 2424 * XXX this code can be improved by resetting the head of 2425 * the list to point to the default rule, and then freeing 2426 * the old list without the need for a lock. 2427 */ 2428 2429 s = splimp(); 2430 free_chain(&layer3_chain, 0 /* keep default rule */); 2431 splx(s); 2432 break; 2433 2434 case IP_FW_ADD: 2435 rule = (struct ip_fw *)rule_buf; /* XXX do a malloc */ 2436 error = sooptcopyin(sopt, rule, sizeof(rule_buf), 2437 sizeof(struct ip_fw) ); 2438 size = sopt->sopt_valsize; 2439 if (error || (error = check_ipfw_struct(rule, size))) 2440 break; 2441 2442 error = add_rule(&layer3_chain, rule); 2443 size = RULESIZE(rule); 2444 if (!error && sopt->sopt_dir == SOPT_GET) 2445 error = sooptcopyout(sopt, rule, size); 2446 break; 2447 2448 case IP_FW_DEL: /* argument is an int, the rule number */ 2449 error = sooptcopyin(sopt, &rulenum, sizeof(int), sizeof(int)); 2450 if (error) 2451 break; 2452 if (rulenum == IPFW_DEFAULT_RULE) { 2453 if (fw_debug) 2454 printf("ipfw: can't delete rule %u\n", 2455 (unsigned)IPFW_DEFAULT_RULE); 2456 error = EINVAL; 2457 } else 2458 error = del_entry(&layer3_chain, rulenum); 2459 break; 2460 2461 case IP_FW_ZERO: 2462 case IP_FW_RESETLOG: /* argument is an int, the rule number */ 2463 rulenum=0; 2464 2465 if (sopt->sopt_val != 0) { 2466 error = sooptcopyin(sopt, &rulenum, 2467 sizeof(int), sizeof(int)); 2468 if (error) 2469 break; 2470 } 2471 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG); 2472 break; 2473 2474 default: 2475 printf("ipfw_ctl invalid option %d\n", sopt->sopt_name); 2476 error = EINVAL; 2477 } 2478 2479 return (error); 2480} 2481 2482/** 2483 * dummynet needs a reference to the default rule, because rules can be 2484 * deleted while packets hold a reference to them. When this happens, 2485 * dummynet changes the reference to the default rule (it could well be a 2486 * NULL pointer, but this way we do not need to check for the special 2487 * case, plus here he have info on the default behaviour). 2488 */ 2489struct ip_fw *ip_fw_default_rule; 2490 2491static void 2492ipfw_tick(void * __unused unused) 2493{ 2494 int i; 2495 int s; 2496 ipfw_dyn_rule *q; 2497 2498 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0) 2499 goto done; 2500 2501 s = splimp(); 2502 for (i = 0 ; i < curr_dyn_buckets ; i++) { 2503 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) { 2504 if (q->dyn_type == O_LIMIT_PARENT) 2505 continue; 2506 if (q->id.proto != IPPROTO_TCP) 2507 continue; 2508 if ( (q->state & BOTH_SYN) != BOTH_SYN) 2509 continue; 2510 if (TIME_LEQ( time_second+20, q->expire)) 2511 continue; /* too early */ 2512 if (TIME_LEQ(q->expire, time_second)) 2513 continue; /* too late, rule expired */ 2514 2515 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 2516 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0); 2517 } 2518 } 2519 splx(s); 2520done: 2521 ipfw_timeout_h = timeout(ipfw_tick, NULL, 5*hz); 2522} 2523 2524static void 2525ipfw_init(void) 2526{ 2527 struct ip_fw default_rule; 2528 2529 ip_fw_chk_ptr = ipfw_chk; 2530 ip_fw_ctl_ptr = ipfw_ctl; 2531 layer3_chain = NULL; 2532 2533 bzero(&default_rule, sizeof default_rule); 2534 2535 default_rule.act_ofs = 0; 2536 default_rule.rulenum = IPFW_DEFAULT_RULE; 2537 default_rule.cmd_len = 1; 2538 2539 default_rule.cmd[0].len = 1; 2540 default_rule.cmd[0].opcode = 2541#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT 2542 1 ? O_ACCEPT : 2543#endif 2544 O_DENY; 2545 2546 add_rule(&layer3_chain, &default_rule); 2547 2548 ip_fw_default_rule = layer3_chain; 2549 printf("IP packet filtering initialized, divert %s, " 2550 "rule-based forwarding %s, default to %s, logging ", 2551#ifdef IPDIVERT 2552 "enabled", 2553#else 2554 "disabled", 2555#endif 2556 "enabled", 2557 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny"); 2558 2559#ifdef IPFIREWALL_VERBOSE 2560 fw_verbose = 1; 2561#endif 2562#ifdef IPFIREWALL_VERBOSE_LIMIT 2563 verbose_limit = IPFIREWALL_VERBOSE_LIMIT; 2564#endif 2565 printf("logging "); 2566 if (fw_verbose == 0) 2567 printf("disabled\n"); 2568 else if (verbose_limit == 0) 2569 printf("unlimited\n"); 2570 else 2571 printf("limited to %d packets/entry by default\n", 2572 verbose_limit); 2573 bzero(&ipfw_timeout_h, sizeof(struct callout_handle)); 2574 ipfw_timeout_h = timeout(ipfw_tick, NULL, hz); 2575} 2576 2577static int 2578ipfw_modevent(module_t mod, int type, void *unused) 2579{ 2580 int s; 2581 int err = 0; 2582 2583 switch (type) { 2584 case MOD_LOAD: 2585 s = splimp(); 2586 if (IPFW_LOADED) { 2587 splx(s); 2588 printf("IP firewall already loaded\n"); 2589 err = EEXIST; 2590 } else { 2591 ipfw_init(); 2592 splx(s); 2593 } 2594 break; 2595 2596 case MOD_UNLOAD: 2597#if !defined(KLD_MODULE) 2598 printf("ipfw statically compiled, cannot unload\n"); 2599 err = EBUSY; 2600#else 2601 s = splimp(); 2602 untimeout(ipfw_tick, NULL, ipfw_timeout_h); 2603 ip_fw_chk_ptr = NULL; 2604 ip_fw_ctl_ptr = NULL; 2605 free_chain(&layer3_chain, 1 /* kill default rule */); 2606 splx(s); 2607 printf("IP firewall unloaded\n"); 2608#endif 2609 break; 2610 default: 2611 break; 2612 } 2613 return err; 2614} 2615 2616static moduledata_t ipfwmod = { 2617 "ipfw", 2618 ipfw_modevent, 2619 0 2620}; 2621DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY); 2622MODULE_VERSION(ipfw, 1); 2623#endif /* IPFW2 */ 2624