ip_fw_dynamic.c revision 242834
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 26#include <sys/cdefs.h> 27__FBSDID("$FreeBSD: head/sys/netpfil/ipfw/ip_fw_dynamic.c 242834 2012-11-09 18:23:38Z melifaro $"); 28 29#define DEB(x) 30#define DDB(x) x 31 32/* 33 * Dynamic rule support for ipfw 34 */ 35 36#include "opt_ipfw.h" 37#include "opt_inet.h" 38#ifndef INET 39#error IPFIREWALL requires INET. 40#endif /* INET */ 41#include "opt_inet6.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/kernel.h> 48#include <sys/lock.h> 49#include <sys/socket.h> 50#include <sys/sysctl.h> 51#include <sys/syslog.h> 52#include <net/ethernet.h> /* for ETHERTYPE_IP */ 53#include <net/if.h> 54#include <net/vnet.h> 55 56#include <netinet/in.h> 57#include <netinet/ip.h> 58#include <netinet/ip_var.h> /* ip_defttl */ 59#include <netinet/ip_fw.h> 60#include <netinet/tcp_var.h> 61#include <netinet/udp.h> 62 63#include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */ 64#ifdef INET6 65#include <netinet6/in6_var.h> 66#include <netinet6/ip6_var.h> 67#endif 68 69#include <netpfil/ipfw/ip_fw_private.h> 70 71#include <machine/in_cksum.h> /* XXX for in_cksum */ 72 73#ifdef MAC 74#include <security/mac/mac_framework.h> 75#endif 76 77/* 78 * Description of dynamic rules. 79 * 80 * Dynamic rules are stored in lists accessed through a hash table 81 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 82 * be modified through the sysctl variable dyn_buckets which is 83 * updated when the table becomes empty. 84 * 85 * XXX currently there is only one list, ipfw_dyn. 86 * 87 * When a packet is received, its address fields are first masked 88 * with the mask defined for the rule, then hashed, then matched 89 * against the entries in the corresponding list. 90 * Dynamic rules can be used for different purposes: 91 * + stateful rules; 92 * + enforcing limits on the number of sessions; 93 * + in-kernel NAT (not implemented yet) 94 * 95 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 96 * measured in seconds and depending on the flags. 97 * 98 * The total number of dynamic rules is stored in dyn_count. 99 * The max number of dynamic rules is dyn_max. When we reach 100 * the maximum number of rules we do not create anymore. This is 101 * done to avoid consuming too much memory, but also too much 102 * time when searching on each packet (ideally, we should try instead 103 * to put a limit on the length of the list on each bucket...). 104 * 105 * Each dynamic rule holds a pointer to the parent ipfw rule so 106 * we know what action to perform. Dynamic rules are removed when 107 * the parent rule is deleted. XXX we should make them survive. 108 * 109 * There are some limitations with dynamic rules -- we do not 110 * obey the 'randomized match', and we do not do multiple 111 * passes through the firewall. XXX check the latter!!! 112 */ 113 114/* 115 * Static variables followed by global ones 116 */ 117static VNET_DEFINE(ipfw_dyn_rule **, ipfw_dyn_v); 118static VNET_DEFINE(u_int32_t, dyn_buckets); 119static VNET_DEFINE(u_int32_t, curr_dyn_buckets); 120static VNET_DEFINE(struct callout, ipfw_timeout); 121#define V_ipfw_dyn_v VNET(ipfw_dyn_v) 122#define V_dyn_buckets VNET(dyn_buckets) 123#define V_curr_dyn_buckets VNET(curr_dyn_buckets) 124#define V_ipfw_timeout VNET(ipfw_timeout) 125 126static uma_zone_t ipfw_dyn_rule_zone; 127#ifndef __FreeBSD__ 128DEFINE_SPINLOCK(ipfw_dyn_mtx); 129#else 130static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */ 131#endif 132 133#define IPFW_DYN_LOCK_INIT() \ 134 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF) 135#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx) 136#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx) 137#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx) 138#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED) 139 140void 141ipfw_dyn_unlock(void) 142{ 143 IPFW_DYN_UNLOCK(); 144} 145 146/* 147 * Timeouts for various events in handing dynamic rules. 148 */ 149static VNET_DEFINE(u_int32_t, dyn_ack_lifetime); 150static VNET_DEFINE(u_int32_t, dyn_syn_lifetime); 151static VNET_DEFINE(u_int32_t, dyn_fin_lifetime); 152static VNET_DEFINE(u_int32_t, dyn_rst_lifetime); 153static VNET_DEFINE(u_int32_t, dyn_udp_lifetime); 154static VNET_DEFINE(u_int32_t, dyn_short_lifetime); 155 156#define V_dyn_ack_lifetime VNET(dyn_ack_lifetime) 157#define V_dyn_syn_lifetime VNET(dyn_syn_lifetime) 158#define V_dyn_fin_lifetime VNET(dyn_fin_lifetime) 159#define V_dyn_rst_lifetime VNET(dyn_rst_lifetime) 160#define V_dyn_udp_lifetime VNET(dyn_udp_lifetime) 161#define V_dyn_short_lifetime VNET(dyn_short_lifetime) 162 163/* 164 * Keepalives are sent if dyn_keepalive is set. They are sent every 165 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 166 * seconds of lifetime of a rule. 167 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 168 * than dyn_keepalive_period. 169 */ 170 171static VNET_DEFINE(u_int32_t, dyn_keepalive_interval); 172static VNET_DEFINE(u_int32_t, dyn_keepalive_period); 173static VNET_DEFINE(u_int32_t, dyn_keepalive); 174 175#define V_dyn_keepalive_interval VNET(dyn_keepalive_interval) 176#define V_dyn_keepalive_period VNET(dyn_keepalive_period) 177#define V_dyn_keepalive VNET(dyn_keepalive) 178 179static VNET_DEFINE(u_int32_t, dyn_count); /* # of dynamic rules */ 180static VNET_DEFINE(u_int32_t, dyn_max); /* max # of dynamic rules */ 181 182#define V_dyn_count VNET(dyn_count) 183#define V_dyn_max VNET(dyn_max) 184 185#ifdef SYSCTL_NODE 186 187SYSBEGIN(f2) 188 189SYSCTL_DECL(_net_inet_ip_fw); 190SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, 191 CTLFLAG_RW, &VNET_NAME(dyn_buckets), 0, 192 "Number of dyn. buckets"); 193SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, 194 CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0, 195 "Current Number of dyn. buckets"); 196SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_count, 197 CTLFLAG_RD, &VNET_NAME(dyn_count), 0, 198 "Number of dyn. rules"); 199SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_max, 200 CTLFLAG_RW, &VNET_NAME(dyn_max), 0, 201 "Max number of dyn. rules"); 202SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, 203 CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0, 204 "Lifetime of dyn. rules for acks"); 205SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, 206 CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0, 207 "Lifetime of dyn. rules for syn"); 208SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, 209 CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0, 210 "Lifetime of dyn. rules for fin"); 211SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, 212 CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0, 213 "Lifetime of dyn. rules for rst"); 214SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, 215 CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0, 216 "Lifetime of dyn. rules for UDP"); 217SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, 218 CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0, 219 "Lifetime of dyn. rules for other situations"); 220SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, 221 CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0, 222 "Enable keepalives for dyn. rules"); 223 224SYSEND 225 226#endif /* SYSCTL_NODE */ 227 228 229static __inline int 230hash_packet6(struct ipfw_flow_id *id) 231{ 232 u_int32_t i; 233 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^ 234 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^ 235 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^ 236 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^ 237 (id->dst_port) ^ (id->src_port); 238 return i; 239} 240 241/* 242 * IMPORTANT: the hash function for dynamic rules must be commutative 243 * in source and destination (ip,port), because rules are bidirectional 244 * and we want to find both in the same bucket. 245 */ 246static __inline int 247hash_packet(struct ipfw_flow_id *id) 248{ 249 u_int32_t i; 250 251#ifdef INET6 252 if (IS_IP6_FLOW_ID(id)) 253 i = hash_packet6(id); 254 else 255#endif /* INET6 */ 256 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 257 i &= (V_curr_dyn_buckets - 1); 258 return i; 259} 260 261/** 262 * Print customizable flow id description via log(9) facility. 263 */ 264static void 265print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags, 266 char *prefix, char *postfix) 267{ 268 struct in_addr da; 269#ifdef INET6 270 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; 271#else 272 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 273#endif 274 275#ifdef INET6 276 if (IS_IP6_FLOW_ID(id)) { 277 ip6_sprintf(src, &id->src_ip6); 278 ip6_sprintf(dst, &id->dst_ip6); 279 } else 280#endif 281 { 282 da.s_addr = htonl(id->src_ip); 283 inet_ntop(AF_INET, &da, src, sizeof(src)); 284 da.s_addr = htonl(id->dst_ip); 285 inet_ntop(AF_INET, &da, dst, sizeof(dst)); 286 } 287 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", 288 prefix, dyn_type, src, id->src_port, dst, 289 id->dst_port, V_dyn_count, postfix); 290} 291 292#define print_dyn_rule(id, dtype, prefix, postfix) \ 293 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) 294 295/** 296 * unlink a dynamic rule from a chain. prev is a pointer to 297 * the previous one, q is a pointer to the rule to delete, 298 * head is a pointer to the head of the queue. 299 * Modifies q and potentially also head. 300 */ 301#define UNLINK_DYN_RULE(prev, head, q) { \ 302 ipfw_dyn_rule *old_q = q; \ 303 \ 304 /* remove a refcount to the parent */ \ 305 if (q->dyn_type == O_LIMIT) \ 306 q->parent->count--; \ 307 V_dyn_count--; \ 308 DEB(print_dyn_rule(&q->id, q->dyn_type, "unlink entry", "left");) \ 309 if (prev != NULL) \ 310 prev->next = q = q->next; \ 311 else \ 312 head = q = q->next; \ 313 uma_zfree(ipfw_dyn_rule_zone, old_q); } 314 315#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 316 317/** 318 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 319 * 320 * If keep_me == NULL, rules are deleted even if not expired, 321 * otherwise only expired rules are removed. 322 * 323 * The value of the second parameter is also used to point to identify 324 * a rule we absolutely do not want to remove (e.g. because we are 325 * holding a reference to it -- this is the case with O_LIMIT_PARENT 326 * rules). The pointer is only used for comparison, so any non-null 327 * value will do. 328 */ 329static void 330remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 331{ 332 static u_int32_t last_remove = 0; 333 334#define FORCE (keep_me == NULL) 335 336 ipfw_dyn_rule *prev, *q; 337 int i, pass = 0, max_pass = 0; 338 339 IPFW_DYN_LOCK_ASSERT(); 340 341 if (V_ipfw_dyn_v == NULL || V_dyn_count == 0) 342 return; 343 /* do not expire more than once per second, it is useless */ 344 if (!FORCE && last_remove == time_uptime) 345 return; 346 last_remove = time_uptime; 347 348 /* 349 * because O_LIMIT refer to parent rules, during the first pass only 350 * remove child and mark any pending LIMIT_PARENT, and remove 351 * them in a second pass. 352 */ 353next_pass: 354 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 355 for (prev=NULL, q = V_ipfw_dyn_v[i] ; q ; ) { 356 /* 357 * Logic can become complex here, so we split tests. 358 */ 359 if (q == keep_me) 360 goto next; 361 if (rule != NULL && rule != q->rule) 362 goto next; /* not the one we are looking for */ 363 if (q->dyn_type == O_LIMIT_PARENT) { 364 /* 365 * handle parent in the second pass, 366 * record we need one. 367 */ 368 max_pass = 1; 369 if (pass == 0) 370 goto next; 371 if (FORCE && q->count != 0 ) { 372 /* XXX should not happen! */ 373 printf("ipfw: OUCH! cannot remove rule," 374 " count %d\n", q->count); 375 } 376 } else { 377 if (!FORCE && 378 !TIME_LEQ( q->expire, time_uptime )) 379 goto next; 380 } 381 if (q->dyn_type != O_LIMIT_PARENT || !q->count) { 382 UNLINK_DYN_RULE(prev, V_ipfw_dyn_v[i], q); 383 continue; 384 } 385next: 386 prev=q; 387 q=q->next; 388 } 389 } 390 if (pass++ < max_pass) 391 goto next_pass; 392} 393 394void 395ipfw_remove_dyn_children(struct ip_fw *rule) 396{ 397 IPFW_DYN_LOCK(); 398 remove_dyn_rule(rule, NULL /* force removal */); 399 IPFW_DYN_UNLOCK(); 400} 401 402/* 403 * Lookup a dynamic rule, locked version. 404 */ 405static ipfw_dyn_rule * 406lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction, 407 struct tcphdr *tcp) 408{ 409 /* 410 * Stateful ipfw extensions. 411 * Lookup into dynamic session queue. 412 */ 413#define MATCH_REVERSE 0 414#define MATCH_FORWARD 1 415#define MATCH_NONE 2 416#define MATCH_UNKNOWN 3 417 int i, dir = MATCH_NONE; 418 ipfw_dyn_rule *prev, *q = NULL; 419 420 IPFW_DYN_LOCK_ASSERT(); 421 422 if (V_ipfw_dyn_v == NULL) 423 goto done; /* not found */ 424 i = hash_packet(pkt); 425 for (prev = NULL, q = V_ipfw_dyn_v[i]; q != NULL;) { 426 if (q->dyn_type == O_LIMIT_PARENT && q->count) 427 goto next; 428 if (TIME_LEQ(q->expire, time_uptime)) { /* expire entry */ 429 UNLINK_DYN_RULE(prev, V_ipfw_dyn_v[i], q); 430 continue; 431 } 432 if (pkt->proto != q->id.proto || q->dyn_type == O_LIMIT_PARENT) 433 goto next; 434 435 if (IS_IP6_FLOW_ID(pkt)) { 436 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) && 437 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) && 438 pkt->src_port == q->id.src_port && 439 pkt->dst_port == q->id.dst_port) { 440 dir = MATCH_FORWARD; 441 break; 442 } 443 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) && 444 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) && 445 pkt->src_port == q->id.dst_port && 446 pkt->dst_port == q->id.src_port) { 447 dir = MATCH_REVERSE; 448 break; 449 } 450 } else { 451 if (pkt->src_ip == q->id.src_ip && 452 pkt->dst_ip == q->id.dst_ip && 453 pkt->src_port == q->id.src_port && 454 pkt->dst_port == q->id.dst_port) { 455 dir = MATCH_FORWARD; 456 break; 457 } 458 if (pkt->src_ip == q->id.dst_ip && 459 pkt->dst_ip == q->id.src_ip && 460 pkt->src_port == q->id.dst_port && 461 pkt->dst_port == q->id.src_port) { 462 dir = MATCH_REVERSE; 463 break; 464 } 465 } 466next: 467 prev = q; 468 q = q->next; 469 } 470 if (q == NULL) 471 goto done; /* q = NULL, not found */ 472 473 if (prev != NULL) { /* found and not in front */ 474 prev->next = q->next; 475 q->next = V_ipfw_dyn_v[i]; 476 V_ipfw_dyn_v[i] = q; 477 } 478 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 479 uint32_t ack; 480 u_char flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST); 481 482#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 483#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 484#define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) 485#define ACK_FWD 0x10000 /* fwd ack seen */ 486#define ACK_REV 0x20000 /* rev ack seen */ 487 488 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8); 489 switch (q->state & TCP_FLAGS) { 490 case TH_SYN: /* opening */ 491 q->expire = time_uptime + V_dyn_syn_lifetime; 492 break; 493 494 case BOTH_SYN: /* move to established */ 495 case BOTH_SYN | TH_FIN: /* one side tries to close */ 496 case BOTH_SYN | (TH_FIN << 8): 497#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 498 if (tcp == NULL) 499 break; 500 501 ack = ntohl(tcp->th_ack); 502 if (dir == MATCH_FORWARD) { 503 if (q->ack_fwd == 0 || 504 _SEQ_GE(ack, q->ack_fwd)) { 505 q->ack_fwd = ack; 506 q->state |= ACK_FWD; 507 } 508 } else { 509 if (q->ack_rev == 0 || 510 _SEQ_GE(ack, q->ack_rev)) { 511 q->ack_rev = ack; 512 q->state |= ACK_REV; 513 } 514 } 515 if ((q->state & (ACK_FWD | ACK_REV)) == 516 (ACK_FWD | ACK_REV)) { 517 q->expire = time_uptime + V_dyn_ack_lifetime; 518 q->state &= ~(ACK_FWD | ACK_REV); 519 } 520 break; 521 522 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 523 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) 524 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1; 525 q->expire = time_uptime + V_dyn_fin_lifetime; 526 break; 527 528 default: 529#if 0 530 /* 531 * reset or some invalid combination, but can also 532 * occur if we use keep-state the wrong way. 533 */ 534 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 535 printf("invalid state: 0x%x\n", q->state); 536#endif 537 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period) 538 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1; 539 q->expire = time_uptime + V_dyn_rst_lifetime; 540 break; 541 } 542 } else if (pkt->proto == IPPROTO_UDP) { 543 q->expire = time_uptime + V_dyn_udp_lifetime; 544 } else { 545 /* other protocols */ 546 q->expire = time_uptime + V_dyn_short_lifetime; 547 } 548done: 549 if (match_direction != NULL) 550 *match_direction = dir; 551 return (q); 552} 553 554ipfw_dyn_rule * 555ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 556 struct tcphdr *tcp) 557{ 558 ipfw_dyn_rule *q; 559 560 IPFW_DYN_LOCK(); 561 q = lookup_dyn_rule_locked(pkt, match_direction, tcp); 562 if (q == NULL) 563 IPFW_DYN_UNLOCK(); 564 /* NB: return table locked when q is not NULL */ 565 return q; 566} 567 568static void 569realloc_dynamic_table(void) 570{ 571 IPFW_DYN_LOCK_ASSERT(); 572 573 /* 574 * Try reallocation, make sure we have a power of 2 and do 575 * not allow more than 64k entries. In case of overflow, 576 * default to 1024. 577 */ 578 579 if (V_dyn_buckets > 65536) 580 V_dyn_buckets = 1024; 581 if ((V_dyn_buckets & (V_dyn_buckets-1)) != 0) { /* not a power of 2 */ 582 V_dyn_buckets = V_curr_dyn_buckets; /* reset */ 583 return; 584 } 585 V_curr_dyn_buckets = V_dyn_buckets; 586 if (V_ipfw_dyn_v != NULL) 587 free(V_ipfw_dyn_v, M_IPFW); 588 for (;;) { 589 V_ipfw_dyn_v = malloc(V_curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 590 M_IPFW, M_NOWAIT | M_ZERO); 591 if (V_ipfw_dyn_v != NULL || V_curr_dyn_buckets <= 2) 592 break; 593 V_curr_dyn_buckets /= 2; 594 } 595} 596 597/** 598 * Install state of type 'type' for a dynamic session. 599 * The hash table contains two type of rules: 600 * - regular rules (O_KEEP_STATE) 601 * - rules for sessions with limited number of sess per user 602 * (O_LIMIT). When they are created, the parent is 603 * increased by 1, and decreased on delete. In this case, 604 * the third parameter is the parent rule and not the chain. 605 * - "parent" rules for the above (O_LIMIT_PARENT). 606 */ 607static ipfw_dyn_rule * 608add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule) 609{ 610 ipfw_dyn_rule *r; 611 int i; 612 613 IPFW_DYN_LOCK_ASSERT(); 614 615 if (V_ipfw_dyn_v == NULL || 616 (V_dyn_count == 0 && V_dyn_buckets != V_curr_dyn_buckets)) { 617 realloc_dynamic_table(); 618 if (V_ipfw_dyn_v == NULL) 619 return NULL; /* failed ! */ 620 } 621 i = hash_packet(id); 622 623 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO); 624 if (r == NULL) { 625 printf ("ipfw: sorry cannot allocate state\n"); 626 return NULL; 627 } 628 629 /* increase refcount on parent, and set pointer */ 630 if (dyn_type == O_LIMIT) { 631 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 632 if ( parent->dyn_type != O_LIMIT_PARENT) 633 panic("invalid parent"); 634 parent->count++; 635 r->parent = parent; 636 rule = parent->rule; 637 } 638 639 r->id = *id; 640 r->expire = time_uptime + V_dyn_syn_lifetime; 641 r->rule = rule; 642 r->dyn_type = dyn_type; 643 r->pcnt = r->bcnt = 0; 644 r->count = 0; 645 646 r->bucket = i; 647 r->next = V_ipfw_dyn_v[i]; 648 V_ipfw_dyn_v[i] = r; 649 V_dyn_count++; 650 DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");) 651 return r; 652} 653 654/** 655 * lookup dynamic parent rule using pkt and rule as search keys. 656 * If the lookup fails, then install one. 657 */ 658static ipfw_dyn_rule * 659lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 660{ 661 ipfw_dyn_rule *q; 662 int i; 663 664 IPFW_DYN_LOCK_ASSERT(); 665 666 if (V_ipfw_dyn_v) { 667 int is_v6 = IS_IP6_FLOW_ID(pkt); 668 i = hash_packet( pkt ); 669 for (q = V_ipfw_dyn_v[i] ; q != NULL ; q=q->next) 670 if (q->dyn_type == O_LIMIT_PARENT && 671 rule== q->rule && 672 pkt->proto == q->id.proto && 673 pkt->src_port == q->id.src_port && 674 pkt->dst_port == q->id.dst_port && 675 ( 676 (is_v6 && 677 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6), 678 &(q->id.src_ip6)) && 679 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6), 680 &(q->id.dst_ip6))) || 681 (!is_v6 && 682 pkt->src_ip == q->id.src_ip && 683 pkt->dst_ip == q->id.dst_ip) 684 ) 685 ) { 686 q->expire = time_uptime + V_dyn_short_lifetime; 687 DEB(print_dyn_rule(pkt, q->dyn_type, 688 "lookup_dyn_parent found", "");) 689 return q; 690 } 691 } 692 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 693} 694 695/** 696 * Install dynamic state for rule type cmd->o.opcode 697 * 698 * Returns 1 (failure) if state is not installed because of errors or because 699 * session limitations are enforced. 700 */ 701int 702ipfw_install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 703 struct ip_fw_args *args, uint32_t tablearg) 704{ 705 static int last_log; 706 ipfw_dyn_rule *q; 707 708 DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", "");) 709 710 IPFW_DYN_LOCK(); 711 712 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL); 713 714 if (q != NULL) { /* should never occur */ 715 DEB( 716 if (last_log != time_uptime) { 717 last_log = time_uptime; 718 printf("ipfw: %s: entry already present, done\n", 719 __func__); 720 }) 721 IPFW_DYN_UNLOCK(); 722 return (0); 723 } 724 725 if (V_dyn_count >= V_dyn_max) 726 /* Run out of slots, try to remove any expired rule. */ 727 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1); 728 729 if (V_dyn_count >= V_dyn_max) { 730 if (last_log != time_uptime) { 731 last_log = time_uptime; 732 printf("ipfw: %s: Too many dynamic rules\n", __func__); 733 } 734 IPFW_DYN_UNLOCK(); 735 return (1); /* cannot install, notify caller */ 736 } 737 738 switch (cmd->o.opcode) { 739 case O_KEEP_STATE: /* bidir rule */ 740 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule); 741 break; 742 743 case O_LIMIT: { /* limit number of sessions */ 744 struct ipfw_flow_id id; 745 ipfw_dyn_rule *parent; 746 uint32_t conn_limit; 747 uint16_t limit_mask = cmd->limit_mask; 748 749 conn_limit = (cmd->conn_limit == IP_FW_TABLEARG) ? 750 tablearg : cmd->conn_limit; 751 752 DEB( 753 if (cmd->conn_limit == IP_FW_TABLEARG) 754 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u " 755 "(tablearg)\n", __func__, conn_limit); 756 else 757 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n", 758 __func__, conn_limit); 759 ) 760 761 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0; 762 id.proto = args->f_id.proto; 763 id.addr_type = args->f_id.addr_type; 764 id.fib = M_GETFIB(args->m); 765 766 if (IS_IP6_FLOW_ID (&(args->f_id))) { 767 if (limit_mask & DYN_SRC_ADDR) 768 id.src_ip6 = args->f_id.src_ip6; 769 if (limit_mask & DYN_DST_ADDR) 770 id.dst_ip6 = args->f_id.dst_ip6; 771 } else { 772 if (limit_mask & DYN_SRC_ADDR) 773 id.src_ip = args->f_id.src_ip; 774 if (limit_mask & DYN_DST_ADDR) 775 id.dst_ip = args->f_id.dst_ip; 776 } 777 if (limit_mask & DYN_SRC_PORT) 778 id.src_port = args->f_id.src_port; 779 if (limit_mask & DYN_DST_PORT) 780 id.dst_port = args->f_id.dst_port; 781 if ((parent = lookup_dyn_parent(&id, rule)) == NULL) { 782 printf("ipfw: %s: add parent failed\n", __func__); 783 IPFW_DYN_UNLOCK(); 784 return (1); 785 } 786 787 if (parent->count >= conn_limit) { 788 /* See if we can remove some expired rule. */ 789 remove_dyn_rule(rule, parent); 790 if (parent->count >= conn_limit) { 791 if (V_fw_verbose && last_log != time_uptime) { 792 last_log = time_uptime; 793 char sbuf[24]; 794 last_log = time_uptime; 795 snprintf(sbuf, sizeof(sbuf), 796 "%d drop session", 797 parent->rule->rulenum); 798 print_dyn_rule_flags(&args->f_id, 799 cmd->o.opcode, 800 LOG_SECURITY | LOG_DEBUG, 801 sbuf, "too many entries"); 802 } 803 IPFW_DYN_UNLOCK(); 804 return (1); 805 } 806 } 807 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent); 808 break; 809 } 810 default: 811 printf("ipfw: %s: unknown dynamic rule type %u\n", 812 __func__, cmd->o.opcode); 813 IPFW_DYN_UNLOCK(); 814 return (1); 815 } 816 817 /* XXX just set lifetime */ 818 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); 819 820 IPFW_DYN_UNLOCK(); 821 return (0); 822} 823 824/* 825 * Generate a TCP packet, containing either a RST or a keepalive. 826 * When flags & TH_RST, we are sending a RST packet, because of a 827 * "reset" action matched the packet. 828 * Otherwise we are sending a keepalive, and flags & TH_ 829 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required 830 * so that MAC can label the reply appropriately. 831 */ 832struct mbuf * 833ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq, 834 u_int32_t ack, int flags) 835{ 836 struct mbuf *m = NULL; /* stupid compiler */ 837 int len, dir; 838 struct ip *h = NULL; /* stupid compiler */ 839#ifdef INET6 840 struct ip6_hdr *h6 = NULL; 841#endif 842 struct tcphdr *th = NULL; 843 844 MGETHDR(m, M_DONTWAIT, MT_DATA); 845 if (m == NULL) 846 return (NULL); 847 848 M_SETFIB(m, id->fib); 849#ifdef MAC 850 if (replyto != NULL) 851 mac_netinet_firewall_reply(replyto, m); 852 else 853 mac_netinet_firewall_send(m); 854#else 855 (void)replyto; /* don't warn about unused arg */ 856#endif 857 858 switch (id->addr_type) { 859 case 4: 860 len = sizeof(struct ip) + sizeof(struct tcphdr); 861 break; 862#ifdef INET6 863 case 6: 864 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 865 break; 866#endif 867 default: 868 /* XXX: log me?!? */ 869 FREE_PKT(m); 870 return (NULL); 871 } 872 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN); 873 874 m->m_data += max_linkhdr; 875 m->m_flags |= M_SKIP_FIREWALL; 876 m->m_pkthdr.len = m->m_len = len; 877 m->m_pkthdr.rcvif = NULL; 878 bzero(m->m_data, len); 879 880 switch (id->addr_type) { 881 case 4: 882 h = mtod(m, struct ip *); 883 884 /* prepare for checksum */ 885 h->ip_p = IPPROTO_TCP; 886 h->ip_len = htons(sizeof(struct tcphdr)); 887 if (dir) { 888 h->ip_src.s_addr = htonl(id->src_ip); 889 h->ip_dst.s_addr = htonl(id->dst_ip); 890 } else { 891 h->ip_src.s_addr = htonl(id->dst_ip); 892 h->ip_dst.s_addr = htonl(id->src_ip); 893 } 894 895 th = (struct tcphdr *)(h + 1); 896 break; 897#ifdef INET6 898 case 6: 899 h6 = mtod(m, struct ip6_hdr *); 900 901 /* prepare for checksum */ 902 h6->ip6_nxt = IPPROTO_TCP; 903 h6->ip6_plen = htons(sizeof(struct tcphdr)); 904 if (dir) { 905 h6->ip6_src = id->src_ip6; 906 h6->ip6_dst = id->dst_ip6; 907 } else { 908 h6->ip6_src = id->dst_ip6; 909 h6->ip6_dst = id->src_ip6; 910 } 911 912 th = (struct tcphdr *)(h6 + 1); 913 break; 914#endif 915 } 916 917 if (dir) { 918 th->th_sport = htons(id->src_port); 919 th->th_dport = htons(id->dst_port); 920 } else { 921 th->th_sport = htons(id->dst_port); 922 th->th_dport = htons(id->src_port); 923 } 924 th->th_off = sizeof(struct tcphdr) >> 2; 925 926 if (flags & TH_RST) { 927 if (flags & TH_ACK) { 928 th->th_seq = htonl(ack); 929 th->th_flags = TH_RST; 930 } else { 931 if (flags & TH_SYN) 932 seq++; 933 th->th_ack = htonl(seq); 934 th->th_flags = TH_RST | TH_ACK; 935 } 936 } else { 937 /* 938 * Keepalive - use caller provided sequence numbers 939 */ 940 th->th_seq = htonl(seq); 941 th->th_ack = htonl(ack); 942 th->th_flags = TH_ACK; 943 } 944 945 switch (id->addr_type) { 946 case 4: 947 th->th_sum = in_cksum(m, len); 948 949 /* finish the ip header */ 950 h->ip_v = 4; 951 h->ip_hl = sizeof(*h) >> 2; 952 h->ip_tos = IPTOS_LOWDELAY; 953 h->ip_off = htons(0); 954 h->ip_len = htons(len); 955 h->ip_ttl = V_ip_defttl; 956 h->ip_sum = 0; 957 break; 958#ifdef INET6 959 case 6: 960 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6), 961 sizeof(struct tcphdr)); 962 963 /* finish the ip6 header */ 964 h6->ip6_vfc |= IPV6_VERSION; 965 h6->ip6_hlim = IPV6_DEFHLIM; 966 break; 967#endif 968 } 969 970 return (m); 971} 972 973/* 974 * Queue keepalive packets for given dynamic rule 975 */ 976static struct mbuf ** 977ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q) 978{ 979 struct mbuf *m_rev, *m_fwd; 980 981 m_rev = (q->state & ACK_REV) ? NULL : 982 ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 983 m_fwd = (q->state & ACK_FWD) ? NULL : 984 ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0); 985 986 if (m_rev != NULL) { 987 *mtailp = m_rev; 988 mtailp = &(*mtailp)->m_nextpkt; 989 } 990 if (m_fwd != NULL) { 991 *mtailp = m_fwd; 992 mtailp = &(*mtailp)->m_nextpkt; 993 } 994 995 return (mtailp); 996} 997 998/* 999 * This procedure is only used to handle keepalives. It is invoked 1000 * every dyn_keepalive_period 1001 */ 1002static void 1003ipfw_tick(void * vnetx) 1004{ 1005 struct mbuf *m0, *m, *mnext, **mtailp; 1006 struct ip *h; 1007 int i; 1008 ipfw_dyn_rule *q; 1009#ifdef VIMAGE 1010 struct vnet *vp = vnetx; 1011#endif 1012 1013 CURVNET_SET(vp); 1014 if (V_dyn_keepalive == 0 || V_ipfw_dyn_v == NULL || V_dyn_count == 0) 1015 goto done; 1016 1017 /* 1018 * We make a chain of packets to go out here -- not deferring 1019 * until after we drop the IPFW dynamic rule lock would result 1020 * in a lock order reversal with the normal packet input -> ipfw 1021 * call stack. 1022 */ 1023 m0 = NULL; 1024 mtailp = &m0; 1025 IPFW_DYN_LOCK(); 1026 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1027 for (q = V_ipfw_dyn_v[i] ; q ; q = q->next ) { 1028 if (q->dyn_type == O_LIMIT_PARENT) 1029 continue; 1030 if (TIME_LEQ(q->expire, time_uptime)) 1031 continue; /* too late, rule expired */ 1032 1033 if (q->id.proto != IPPROTO_TCP) 1034 continue; 1035 if ( (q->state & BOTH_SYN) != BOTH_SYN) 1036 continue; 1037 if (TIME_LEQ(time_uptime + V_dyn_keepalive_interval, 1038 q->expire)) 1039 continue; /* too early */ 1040 1041 mtailp = ipfw_dyn_send_ka(mtailp, q); 1042 } 1043 } 1044 IPFW_DYN_UNLOCK(); 1045 1046 /* Send keepalive packets if any */ 1047 for (m = m0; m != NULL; m = mnext) { 1048 mnext = m->m_nextpkt; 1049 m->m_nextpkt = NULL; 1050 h = mtod(m, struct ip *); 1051 if (h->ip_v == 4) 1052 ip_output(m, NULL, NULL, 0, NULL, NULL); 1053#ifdef INET6 1054 else 1055 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1056#endif 1057 } 1058done: 1059 callout_reset_on(&V_ipfw_timeout, V_dyn_keepalive_period * hz, 1060 ipfw_tick, vnetx, 0); 1061 CURVNET_RESTORE(); 1062} 1063 1064void 1065ipfw_dyn_attach(void) 1066{ 1067 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule", 1068 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL, 1069 UMA_ALIGN_PTR, 0); 1070 1071 IPFW_DYN_LOCK_INIT(); 1072} 1073 1074void 1075ipfw_dyn_detach(void) 1076{ 1077 uma_zdestroy(ipfw_dyn_rule_zone); 1078 IPFW_DYN_LOCK_DESTROY(); 1079} 1080 1081void 1082ipfw_dyn_init(void) 1083{ 1084 V_ipfw_dyn_v = NULL; 1085 V_dyn_buckets = 256; /* must be power of 2 */ 1086 V_curr_dyn_buckets = 256; /* must be power of 2 */ 1087 1088 V_dyn_ack_lifetime = 300; 1089 V_dyn_syn_lifetime = 20; 1090 V_dyn_fin_lifetime = 1; 1091 V_dyn_rst_lifetime = 1; 1092 V_dyn_udp_lifetime = 10; 1093 V_dyn_short_lifetime = 5; 1094 1095 V_dyn_keepalive_interval = 20; 1096 V_dyn_keepalive_period = 5; 1097 V_dyn_keepalive = 1; /* do send keepalives */ 1098 1099 V_dyn_max = 4096; /* max # of dynamic rules */ 1100 callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE); 1101 callout_reset_on(&V_ipfw_timeout, hz, ipfw_tick, curvnet, 0); 1102} 1103 1104void 1105ipfw_dyn_uninit(int pass) 1106{ 1107 if (pass == 0) 1108 callout_drain(&V_ipfw_timeout); 1109 else { 1110 if (V_ipfw_dyn_v != NULL) 1111 free(V_ipfw_dyn_v, M_IPFW); 1112 } 1113} 1114 1115int 1116ipfw_dyn_len(void) 1117{ 1118 return (V_ipfw_dyn_v == NULL) ? 0 : 1119 (V_dyn_count * sizeof(ipfw_dyn_rule)); 1120} 1121 1122void 1123ipfw_get_dynamic(char **pbp, const char *ep) 1124{ 1125 ipfw_dyn_rule *p, *last = NULL; 1126 char *bp; 1127 int i; 1128 1129 if (V_ipfw_dyn_v == NULL) 1130 return; 1131 bp = *pbp; 1132 1133 IPFW_DYN_LOCK(); 1134 for (i = 0 ; i < V_curr_dyn_buckets; i++) 1135 for (p = V_ipfw_dyn_v[i] ; p != NULL; p = p->next) { 1136 if (bp + sizeof *p <= ep) { 1137 ipfw_dyn_rule *dst = 1138 (ipfw_dyn_rule *)bp; 1139 bcopy(p, dst, sizeof *p); 1140 bcopy(&(p->rule->rulenum), &(dst->rule), 1141 sizeof(p->rule->rulenum)); 1142 /* 1143 * store set number into high word of 1144 * dst->rule pointer. 1145 */ 1146 bcopy(&(p->rule->set), 1147 (char *)&dst->rule + 1148 sizeof(p->rule->rulenum), 1149 sizeof(p->rule->set)); 1150 /* 1151 * store a non-null value in "next". 1152 * The userland code will interpret a 1153 * NULL here as a marker 1154 * for the last dynamic rule. 1155 */ 1156 bcopy(&dst, &dst->next, sizeof(dst)); 1157 last = dst; 1158 dst->expire = 1159 TIME_LEQ(dst->expire, time_uptime) ? 1160 0 : dst->expire - time_uptime ; 1161 bp += sizeof(ipfw_dyn_rule); 1162 } 1163 } 1164 IPFW_DYN_UNLOCK(); 1165 if (last != NULL) /* mark last dynamic rule */ 1166 bzero(&last->next, sizeof(last)); 1167 *pbp = bp; 1168} 1169/* end of file */ 1170