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("$FreeBSD$"); 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/ktr.h> 49 #include <sys/lock.h> 50 #include <sys/rmlock.h> 51 #include <sys/socket.h> 52 #include <sys/sysctl.h> 53 #include <sys/syslog.h> 54 #include <net/ethernet.h> /* for ETHERTYPE_IP */ 55 #include <net/if.h> 56 #include <net/if_var.h> 57 #include <net/vnet.h> 58 59 #include <netinet/in.h> 60 #include <netinet/ip.h> 61 #include <netinet/ip_var.h> /* ip_defttl */ 62 #include <netinet/ip_fw.h> 63 #include <netinet/tcp_var.h> 64 #include <netinet/udp.h> 65 66 #include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */ 67 #ifdef INET6 68 #include <netinet6/in6_var.h> 69 #include <netinet6/ip6_var.h> 70 #endif 71 72 #include <netpfil/ipfw/ip_fw_private.h> 73 74 #include <machine/in_cksum.h> /* XXX for in_cksum */ 75 76 #ifdef MAC 77 #include <security/mac/mac_framework.h> 78 #endif 79 80 /* 81 * Description of dynamic rules. 82 * 83 * Dynamic rules are stored in lists accessed through a hash table 84 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 85 * be modified through the sysctl variable dyn_buckets which is 86 * updated when the table becomes empty. 87 * 88 * XXX currently there is only one list, ipfw_dyn. 89 * 90 * When a packet is received, its address fields are first masked 91 * with the mask defined for the rule, then hashed, then matched 92 * against the entries in the corresponding list. 93 * Dynamic rules can be used for different purposes: 94 * + stateful rules; 95 * + enforcing limits on the number of sessions; 96 * + in-kernel NAT (not implemented yet) 97 * 98 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 99 * measured in seconds and depending on the flags. 100 * 101 * The total number of dynamic rules is equal to UMA zone items count. 102 * The max number of dynamic rules is dyn_max. When we reach 103 * the maximum number of rules we do not create anymore. This is 104 * done to avoid consuming too much memory, but also too much 105 * time when searching on each packet (ideally, we should try instead 106 * to put a limit on the length of the list on each bucket...). 107 * 108 * Each dynamic rule holds a pointer to the parent ipfw rule so 109 * we know what action to perform. Dynamic rules are removed when 110 * the parent rule is deleted. This can be changed by dyn_keep_states 111 * sysctl. 112 * 113 * There are some limitations with dynamic rules -- we do not 114 * obey the 'randomized match', and we do not do multiple 115 * passes through the firewall. XXX check the latter!!! 116 */ 117 118 struct ipfw_dyn_bucket { 119 struct mtx mtx; /* Bucket protecting lock */ 120 ipfw_dyn_rule *head; /* Pointer to first rule */ 121 }; 122 123 /* 124 * Static variables followed by global ones 125 */ 126 static VNET_DEFINE(struct ipfw_dyn_bucket *, ipfw_dyn_v); 127 static VNET_DEFINE(u_int32_t, dyn_buckets_max); 128 static VNET_DEFINE(u_int32_t, curr_dyn_buckets); 129 static VNET_DEFINE(struct callout, ipfw_timeout); 130 #define V_ipfw_dyn_v VNET(ipfw_dyn_v) 131 #define V_dyn_buckets_max VNET(dyn_buckets_max) 132 #define V_curr_dyn_buckets VNET(curr_dyn_buckets) 133 #define V_ipfw_timeout VNET(ipfw_timeout) 134 135 static VNET_DEFINE(uma_zone_t, ipfw_dyn_rule_zone); 136 #define V_ipfw_dyn_rule_zone VNET(ipfw_dyn_rule_zone) 137 138 #define IPFW_BUCK_LOCK_INIT(b) \ 139 mtx_init(&(b)->mtx, "IPFW dynamic bucket", NULL, MTX_DEF) 140 #define IPFW_BUCK_LOCK_DESTROY(b) \ 141 mtx_destroy(&(b)->mtx) 142 #define IPFW_BUCK_LOCK(i) mtx_lock(&V_ipfw_dyn_v[(i)].mtx) 143 #define IPFW_BUCK_UNLOCK(i) mtx_unlock(&V_ipfw_dyn_v[(i)].mtx) 144 #define IPFW_BUCK_ASSERT(i) mtx_assert(&V_ipfw_dyn_v[(i)].mtx, MA_OWNED) 145 146 147 static VNET_DEFINE(int, dyn_keep_states); 148 #define V_dyn_keep_states VNET(dyn_keep_states) 149 150 /* 151 * Timeouts for various events in handing dynamic rules. 152 */ 153 static VNET_DEFINE(u_int32_t, dyn_ack_lifetime); 154 static VNET_DEFINE(u_int32_t, dyn_syn_lifetime); 155 static VNET_DEFINE(u_int32_t, dyn_fin_lifetime); 156 static VNET_DEFINE(u_int32_t, dyn_rst_lifetime); 157 static VNET_DEFINE(u_int32_t, dyn_udp_lifetime); 158 static VNET_DEFINE(u_int32_t, dyn_short_lifetime); 159 160 #define V_dyn_ack_lifetime VNET(dyn_ack_lifetime) 161 #define V_dyn_syn_lifetime VNET(dyn_syn_lifetime) 162 #define V_dyn_fin_lifetime VNET(dyn_fin_lifetime) 163 #define V_dyn_rst_lifetime VNET(dyn_rst_lifetime) 164 #define V_dyn_udp_lifetime VNET(dyn_udp_lifetime) 165 #define V_dyn_short_lifetime VNET(dyn_short_lifetime) 166 167 /* 168 * Keepalives are sent if dyn_keepalive is set. They are sent every 169 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 170 * seconds of lifetime of a rule. 171 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 172 * than dyn_keepalive_period. 173 */ 174 175 static VNET_DEFINE(u_int32_t, dyn_keepalive_interval); 176 static VNET_DEFINE(u_int32_t, dyn_keepalive_period); 177 static VNET_DEFINE(u_int32_t, dyn_keepalive); 178 static VNET_DEFINE(time_t, dyn_keepalive_last); 179 180 #define V_dyn_keepalive_interval VNET(dyn_keepalive_interval) 181 #define V_dyn_keepalive_period VNET(dyn_keepalive_period) 182 #define V_dyn_keepalive VNET(dyn_keepalive) 183 #define V_dyn_keepalive_last VNET(dyn_keepalive_last) 184 185 static VNET_DEFINE(u_int32_t, dyn_max); /* max # of dynamic rules */ 186 187 #define DYN_COUNT uma_zone_get_cur(V_ipfw_dyn_rule_zone) 188 #define V_dyn_max VNET(dyn_max) 189 190 /* for userspace, we emulate the uma_zone_counter with ipfw_dyn_count */ 191 static int ipfw_dyn_count; /* number of objects */ 192 193 #ifdef USERSPACE /* emulation of UMA object counters for userspace */ 194 #define uma_zone_get_cur(x) ipfw_dyn_count 195 #endif /* USERSPACE */ 196 197 static int last_log; /* Log ratelimiting */ 198 199 static void ipfw_dyn_tick(void *vnetx); 200 static void check_dyn_rules(struct ip_fw_chain *, ipfw_range_tlv *, int, int); 201 #ifdef SYSCTL_NODE 202 203 static int sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS); 204 static int sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS); 205 206 SYSBEGIN(f2) 207 208 SYSCTL_DECL(_net_inet_ip_fw); 209 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, 210 CTLFLAG_RW, &VNET_NAME(dyn_buckets_max), 0, 211 "Max number of dyn. buckets"); 212 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, 213 CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0, 214 "Current Number of dyn. buckets"); 215 SYSCTL_VNET_PROC(_net_inet_ip_fw, OID_AUTO, dyn_count, 216 CTLTYPE_UINT|CTLFLAG_RD, 0, 0, sysctl_ipfw_dyn_count, "IU", 217 "Number of dyn. rules"); 218 SYSCTL_VNET_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max, 219 CTLTYPE_UINT|CTLFLAG_RW, 0, 0, sysctl_ipfw_dyn_max, "IU", 220 "Max number of dyn. rules"); 221 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, 222 CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0, 223 "Lifetime of dyn. rules for acks"); 224 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, 225 CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0, 226 "Lifetime of dyn. rules for syn"); 227 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, 228 CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0, 229 "Lifetime of dyn. rules for fin"); 230 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, 231 CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0, 232 "Lifetime of dyn. rules for rst"); 233 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, 234 CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0, 235 "Lifetime of dyn. rules for UDP"); 236 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, 237 CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0, 238 "Lifetime of dyn. rules for other situations"); 239 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, 240 CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0, 241 "Enable keepalives for dyn. rules"); 242 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keep_states, 243 CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0, 244 "Do not flush dynamic states on rule deletion"); 245 246 SYSEND 247 248 #endif /* SYSCTL_NODE */ 249 250 251 #ifdef INET6 252 static __inline int 253 hash_packet6(struct ipfw_flow_id *id) 254 { 255 u_int32_t i; 256 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^ 257 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^ 258 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^ 259 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^ 260 (id->dst_port) ^ (id->src_port); 261 return i; 262 } 263 #endif 264 265 /* 266 * IMPORTANT: the hash function for dynamic rules must be commutative 267 * in source and destination (ip,port), because rules are bidirectional 268 * and we want to find both in the same bucket. 269 */ 270 static __inline int 271 hash_packet(struct ipfw_flow_id *id, int buckets) 272 { 273 u_int32_t i; 274 275 #ifdef INET6 276 if (IS_IP6_FLOW_ID(id)) 277 i = hash_packet6(id); 278 else 279 #endif /* INET6 */ 280 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 281 i &= (buckets - 1); 282 return i; 283 } 284 285 /** 286 * Print customizable flow id description via log(9) facility. 287 */ 288 static void 289 print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags, 290 char *prefix, char *postfix) 291 { 292 struct in_addr da; 293 #ifdef INET6 294 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; 295 #else 296 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 297 #endif 298 299 #ifdef INET6 300 if (IS_IP6_FLOW_ID(id)) { 301 ip6_sprintf(src, &id->src_ip6); 302 ip6_sprintf(dst, &id->dst_ip6); 303 } else 304 #endif 305 { 306 da.s_addr = htonl(id->src_ip); 307 inet_ntop(AF_INET, &da, src, sizeof(src)); 308 da.s_addr = htonl(id->dst_ip); 309 inet_ntop(AF_INET, &da, dst, sizeof(dst)); 310 } 311 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", 312 prefix, dyn_type, src, id->src_port, dst, 313 id->dst_port, DYN_COUNT, postfix); 314 } 315 316 #define print_dyn_rule(id, dtype, prefix, postfix) \ 317 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) 318 319 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 320 #define TIME_LE(a,b) ((int)((a)-(b)) < 0) 321 322 /* 323 * Lookup a dynamic rule, locked version. 324 */ 325 static ipfw_dyn_rule * 326 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction, 327 struct tcphdr *tcp) 328 { 329 /* 330 * Stateful ipfw extensions. 331 * Lookup into dynamic session queue. 332 */ 333 #define MATCH_REVERSE 0 334 #define MATCH_FORWARD 1 335 #define MATCH_NONE 2 336 #define MATCH_UNKNOWN 3 337 int dir = MATCH_NONE; 338 ipfw_dyn_rule *prev, *q = NULL; 339 340 IPFW_BUCK_ASSERT(i); 341 342 for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) { 343 if (q->dyn_type == O_LIMIT_PARENT && q->count) 344 continue; 345 346 if (pkt->proto != q->id.proto || q->dyn_type == O_LIMIT_PARENT) 347 continue; 348 349 if (IS_IP6_FLOW_ID(pkt)) { 350 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) && 351 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) && 352 pkt->src_port == q->id.src_port && 353 pkt->dst_port == q->id.dst_port) { 354 dir = MATCH_FORWARD; 355 break; 356 } 357 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) && 358 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) && 359 pkt->src_port == q->id.dst_port && 360 pkt->dst_port == q->id.src_port) { 361 dir = MATCH_REVERSE; 362 break; 363 } 364 } else { 365 if (pkt->src_ip == q->id.src_ip && 366 pkt->dst_ip == q->id.dst_ip && 367 pkt->src_port == q->id.src_port && 368 pkt->dst_port == q->id.dst_port) { 369 dir = MATCH_FORWARD; 370 break; 371 } 372 if (pkt->src_ip == q->id.dst_ip && 373 pkt->dst_ip == q->id.src_ip && 374 pkt->src_port == q->id.dst_port && 375 pkt->dst_port == q->id.src_port) { 376 dir = MATCH_REVERSE; 377 break; 378 } 379 } 380 } 381 if (q == NULL) 382 goto done; /* q = NULL, not found */ 383 384 if (prev != NULL) { /* found and not in front */ 385 prev->next = q->next; 386 q->next = V_ipfw_dyn_v[i].head; 387 V_ipfw_dyn_v[i].head = q; 388 } 389 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 390 uint32_t ack; 391 u_char flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST); 392 393 #define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 394 #define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 395 #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) 396 #define ACK_FWD 0x10000 /* fwd ack seen */ 397 #define ACK_REV 0x20000 /* rev ack seen */ 398 399 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8); 400 switch (q->state & TCP_FLAGS) { 401 case TH_SYN: /* opening */ 402 q->expire = time_uptime + V_dyn_syn_lifetime; 403 break; 404 405 case BOTH_SYN: /* move to established */ 406 case BOTH_SYN | TH_FIN: /* one side tries to close */ 407 case BOTH_SYN | (TH_FIN << 8): 408 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 409 if (tcp == NULL) 410 break; 411 412 ack = ntohl(tcp->th_ack); 413 if (dir == MATCH_FORWARD) { 414 if (q->ack_fwd == 0 || 415 _SEQ_GE(ack, q->ack_fwd)) { 416 q->ack_fwd = ack; 417 q->state |= ACK_FWD; 418 } 419 } else { 420 if (q->ack_rev == 0 || 421 _SEQ_GE(ack, q->ack_rev)) { 422 q->ack_rev = ack; 423 q->state |= ACK_REV; 424 } 425 } 426 if ((q->state & (ACK_FWD | ACK_REV)) == 427 (ACK_FWD | ACK_REV)) { 428 q->expire = time_uptime + V_dyn_ack_lifetime; 429 q->state &= ~(ACK_FWD | ACK_REV); 430 } 431 break; 432 433 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 434 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) 435 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1; 436 q->expire = time_uptime + V_dyn_fin_lifetime; 437 break; 438 439 default: 440 #if 0 441 /* 442 * reset or some invalid combination, but can also 443 * occur if we use keep-state the wrong way. 444 */ 445 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 446 printf("invalid state: 0x%x\n", q->state); 447 #endif 448 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period) 449 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1; 450 q->expire = time_uptime + V_dyn_rst_lifetime; 451 break; 452 } 453 } else if (pkt->proto == IPPROTO_UDP) { 454 q->expire = time_uptime + V_dyn_udp_lifetime; 455 } else { 456 /* other protocols */ 457 q->expire = time_uptime + V_dyn_short_lifetime; 458 } 459 done: 460 if (match_direction != NULL) 461 *match_direction = dir; 462 return (q); 463 } 464 465 ipfw_dyn_rule * 466 ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 467 struct tcphdr *tcp) 468 { 469 ipfw_dyn_rule *q; 470 int i; 471 472 i = hash_packet(pkt, V_curr_dyn_buckets); 473 474 IPFW_BUCK_LOCK(i); 475 q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp); 476 if (q == NULL) 477 IPFW_BUCK_UNLOCK(i); 478 /* NB: return table locked when q is not NULL */ 479 return q; 480 } 481 482 /* 483 * Unlock bucket mtx 484 * @p - pointer to dynamic rule 485 */ 486 void 487 ipfw_dyn_unlock(ipfw_dyn_rule *q) 488 { 489 490 IPFW_BUCK_UNLOCK(q->bucket); 491 } 492 493 static int 494 resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets) 495 { 496 int i, k, nbuckets_old; 497 ipfw_dyn_rule *q; 498 struct ipfw_dyn_bucket *dyn_v, *dyn_v_old; 499 500 /* Check if given number is power of 2 and less than 64k */ 501 if ((nbuckets > 65536) || (!powerof2(nbuckets))) 502 return 1; 503 504 CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__, 505 V_curr_dyn_buckets, nbuckets); 506 507 /* Allocate and initialize new hash */ 508 dyn_v = malloc(nbuckets * sizeof(ipfw_dyn_rule), M_IPFW, 509 M_WAITOK | M_ZERO); 510 511 for (i = 0 ; i < nbuckets; i++) 512 IPFW_BUCK_LOCK_INIT(&dyn_v[i]); 513 514 /* 515 * Call upper half lock, as get_map() do to ease 516 * read-only access to dynamic rules hash from sysctl 517 */ 518 IPFW_UH_WLOCK(chain); 519 520 /* 521 * Acquire chain write lock to permit hash access 522 * for main traffic path without additional locks 523 */ 524 IPFW_WLOCK(chain); 525 526 /* Save old values */ 527 nbuckets_old = V_curr_dyn_buckets; 528 dyn_v_old = V_ipfw_dyn_v; 529 530 /* Skip relinking if array is not set up */ 531 if (V_ipfw_dyn_v == NULL) 532 V_curr_dyn_buckets = 0; 533 534 /* Re-link all dynamic states */ 535 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 536 while (V_ipfw_dyn_v[i].head != NULL) { 537 /* Remove from current chain */ 538 q = V_ipfw_dyn_v[i].head; 539 V_ipfw_dyn_v[i].head = q->next; 540 541 /* Get new hash value */ 542 k = hash_packet(&q->id, nbuckets); 543 q->bucket = k; 544 /* Add to the new head */ 545 q->next = dyn_v[k].head; 546 dyn_v[k].head = q; 547 } 548 } 549 550 /* Update current pointers/buckets values */ 551 V_curr_dyn_buckets = nbuckets; 552 V_ipfw_dyn_v = dyn_v; 553 554 IPFW_WUNLOCK(chain); 555 556 IPFW_UH_WUNLOCK(chain); 557 558 /* Start periodic callout on initial creation */ 559 if (dyn_v_old == NULL) { 560 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0); 561 return (0); 562 } 563 564 /* Destroy all mutexes */ 565 for (i = 0 ; i < nbuckets_old ; i++) 566 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]); 567 568 /* Free old hash */ 569 free(dyn_v_old, M_IPFW); 570 571 return 0; 572 } 573 574 /** 575 * Install state of type 'type' for a dynamic session. 576 * The hash table contains two type of rules: 577 * - regular rules (O_KEEP_STATE) 578 * - rules for sessions with limited number of sess per user 579 * (O_LIMIT). When they are created, the parent is 580 * increased by 1, and decreased on delete. In this case, 581 * the third parameter is the parent rule and not the chain. 582 * - "parent" rules for the above (O_LIMIT_PARENT). 583 */ 584 static ipfw_dyn_rule * 585 add_dyn_rule(struct ipfw_flow_id *id, int i, u_int8_t dyn_type, struct ip_fw *rule) 586 { 587 ipfw_dyn_rule *r; 588 589 IPFW_BUCK_ASSERT(i); 590 591 r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO); 592 if (r == NULL) { 593 if (last_log != time_uptime) { 594 last_log = time_uptime; 595 log(LOG_DEBUG, "ipfw: %s: Cannot allocate rule\n", 596 __func__); 597 } 598 return NULL; 599 } 600 ipfw_dyn_count++; 601 602 /* 603 * refcount on parent is already incremented, so 604 * it is safe to use parent unlocked. 605 */ 606 if (dyn_type == O_LIMIT) { 607 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 608 if ( parent->dyn_type != O_LIMIT_PARENT) 609 panic("invalid parent"); 610 r->parent = parent; 611 rule = parent->rule; 612 } 613 614 r->id = *id; 615 r->expire = time_uptime + V_dyn_syn_lifetime; 616 r->rule = rule; 617 r->dyn_type = dyn_type; 618 IPFW_ZERO_DYN_COUNTER(r); 619 r->count = 0; 620 621 r->bucket = i; 622 r->next = V_ipfw_dyn_v[i].head; 623 V_ipfw_dyn_v[i].head = r; 624 DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");) 625 return r; 626 } 627 628 /** 629 * lookup dynamic parent rule using pkt and rule as search keys. 630 * If the lookup fails, then install one. 631 */ 632 static ipfw_dyn_rule * 633 lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule) 634 { 635 ipfw_dyn_rule *q; 636 int i, is_v6; 637 638 is_v6 = IS_IP6_FLOW_ID(pkt); 639 i = hash_packet( pkt, V_curr_dyn_buckets ); 640 *pindex = i; 641 IPFW_BUCK_LOCK(i); 642 for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next) 643 if (q->dyn_type == O_LIMIT_PARENT && 644 rule== q->rule && 645 pkt->proto == q->id.proto && 646 pkt->src_port == q->id.src_port && 647 pkt->dst_port == q->id.dst_port && 648 ( 649 (is_v6 && 650 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6), 651 &(q->id.src_ip6)) && 652 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6), 653 &(q->id.dst_ip6))) || 654 (!is_v6 && 655 pkt->src_ip == q->id.src_ip && 656 pkt->dst_ip == q->id.dst_ip) 657 ) 658 ) { 659 q->expire = time_uptime + V_dyn_short_lifetime; 660 DEB(print_dyn_rule(pkt, q->dyn_type, 661 "lookup_dyn_parent found", "");) 662 return q; 663 } 664 665 /* Add virtual limiting rule */ 666 return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule); 667 } 668 669 /** 670 * Install dynamic state for rule type cmd->o.opcode 671 * 672 * Returns 1 (failure) if state is not installed because of errors or because 673 * session limitations are enforced. 674 */ 675 int 676 ipfw_install_state(struct ip_fw_chain *chain, struct ip_fw *rule, 677 ipfw_insn_limit *cmd, struct ip_fw_args *args, uint32_t tablearg) 678 { 679 ipfw_dyn_rule *q; 680 int i; 681 682 DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", "");) 683 684 i = hash_packet(&args->f_id, V_curr_dyn_buckets); 685 686 IPFW_BUCK_LOCK(i); 687 688 q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL); 689 690 if (q != NULL) { /* should never occur */ 691 DEB( 692 if (last_log != time_uptime) { 693 last_log = time_uptime; 694 printf("ipfw: %s: entry already present, done\n", 695 __func__); 696 }) 697 IPFW_BUCK_UNLOCK(i); 698 return (0); 699 } 700 701 /* 702 * State limiting is done via uma(9) zone limiting. 703 * Save pointer to newly-installed rule and reject 704 * packet if add_dyn_rule() returned NULL. 705 * Note q is currently set to NULL. 706 */ 707 708 switch (cmd->o.opcode) { 709 case O_KEEP_STATE: /* bidir rule */ 710 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule); 711 break; 712 713 case O_LIMIT: { /* limit number of sessions */ 714 struct ipfw_flow_id id; 715 ipfw_dyn_rule *parent; 716 uint32_t conn_limit; 717 uint16_t limit_mask = cmd->limit_mask; 718 int pindex; 719 720 conn_limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit); 721 722 DEB( 723 if (cmd->conn_limit == IP_FW_TARG) 724 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u " 725 "(tablearg)\n", __func__, conn_limit); 726 else 727 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n", 728 __func__, conn_limit); 729 ) 730 731 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0; 732 id.proto = args->f_id.proto; 733 id.addr_type = args->f_id.addr_type; 734 id.fib = M_GETFIB(args->m); 735 736 if (IS_IP6_FLOW_ID (&(args->f_id))) { 737 if (limit_mask & DYN_SRC_ADDR) 738 id.src_ip6 = args->f_id.src_ip6; 739 if (limit_mask & DYN_DST_ADDR) 740 id.dst_ip6 = args->f_id.dst_ip6; 741 } else { 742 if (limit_mask & DYN_SRC_ADDR) 743 id.src_ip = args->f_id.src_ip; 744 if (limit_mask & DYN_DST_ADDR) 745 id.dst_ip = args->f_id.dst_ip; 746 } 747 if (limit_mask & DYN_SRC_PORT) 748 id.src_port = args->f_id.src_port; 749 if (limit_mask & DYN_DST_PORT) 750 id.dst_port = args->f_id.dst_port; 751 752 /* 753 * We have to release lock for previous bucket to 754 * avoid possible deadlock 755 */ 756 IPFW_BUCK_UNLOCK(i); 757 758 if ((parent = lookup_dyn_parent(&id, &pindex, rule)) == NULL) { 759 printf("ipfw: %s: add parent failed\n", __func__); 760 IPFW_BUCK_UNLOCK(pindex); 761 return (1); 762 } 763 764 if (parent->count >= conn_limit) { 765 if (V_fw_verbose && last_log != time_uptime) { 766 last_log = time_uptime; 767 char sbuf[24]; 768 last_log = time_uptime; 769 snprintf(sbuf, sizeof(sbuf), 770 "%d drop session", 771 parent->rule->rulenum); 772 print_dyn_rule_flags(&args->f_id, 773 cmd->o.opcode, 774 LOG_SECURITY | LOG_DEBUG, 775 sbuf, "too many entries"); 776 } 777 IPFW_BUCK_UNLOCK(pindex); 778 return (1); 779 } 780 /* Increment counter on parent */ 781 parent->count++; 782 IPFW_BUCK_UNLOCK(pindex); 783 784 IPFW_BUCK_LOCK(i); 785 q = add_dyn_rule(&args->f_id, i, O_LIMIT, (struct ip_fw *)parent); 786 if (q == NULL) { 787 /* Decrement index and notify caller */ 788 IPFW_BUCK_UNLOCK(i); 789 IPFW_BUCK_LOCK(pindex); 790 parent->count--; 791 IPFW_BUCK_UNLOCK(pindex); 792 return (1); 793 } 794 break; 795 } 796 default: 797 printf("ipfw: %s: unknown dynamic rule type %u\n", 798 __func__, cmd->o.opcode); 799 } 800 801 if (q == NULL) { 802 IPFW_BUCK_UNLOCK(i); 803 return (1); /* Notify caller about failure */ 804 } 805 806 /* XXX just set lifetime */ 807 lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL); 808 809 IPFW_BUCK_UNLOCK(i); 810 return (0); 811 } 812 813 /* 814 * Generate a TCP packet, containing either a RST or a keepalive. 815 * When flags & TH_RST, we are sending a RST packet, because of a 816 * "reset" action matched the packet. 817 * Otherwise we are sending a keepalive, and flags & TH_ 818 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required 819 * so that MAC can label the reply appropriately. 820 */ 821 struct mbuf * 822 ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq, 823 u_int32_t ack, int flags) 824 { 825 struct mbuf *m = NULL; /* stupid compiler */ 826 int len, dir; 827 struct ip *h = NULL; /* stupid compiler */ 828 #ifdef INET6 829 struct ip6_hdr *h6 = NULL; 830 #endif 831 struct tcphdr *th = NULL; 832 833 MGETHDR(m, M_NOWAIT, MT_DATA); 834 if (m == NULL) 835 return (NULL); 836 837 M_SETFIB(m, id->fib); 838 #ifdef MAC 839 if (replyto != NULL) 840 mac_netinet_firewall_reply(replyto, m); 841 else 842 mac_netinet_firewall_send(m); 843 #else 844 (void)replyto; /* don't warn about unused arg */ 845 #endif 846 847 switch (id->addr_type) { 848 case 4: 849 len = sizeof(struct ip) + sizeof(struct tcphdr); 850 break; 851 #ifdef INET6 852 case 6: 853 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 854 break; 855 #endif 856 default: 857 /* XXX: log me?!? */ 858 FREE_PKT(m); 859 return (NULL); 860 } 861 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN); 862 863 m->m_data += max_linkhdr; 864 m->m_flags |= M_SKIP_FIREWALL; 865 m->m_pkthdr.len = m->m_len = len; 866 m->m_pkthdr.rcvif = NULL; 867 bzero(m->m_data, len); 868 869 switch (id->addr_type) { 870 case 4: 871 h = mtod(m, struct ip *); 872 873 /* prepare for checksum */ 874 h->ip_p = IPPROTO_TCP; 875 h->ip_len = htons(sizeof(struct tcphdr)); 876 if (dir) { 877 h->ip_src.s_addr = htonl(id->src_ip); 878 h->ip_dst.s_addr = htonl(id->dst_ip); 879 } else { 880 h->ip_src.s_addr = htonl(id->dst_ip); 881 h->ip_dst.s_addr = htonl(id->src_ip); 882 } 883 884 th = (struct tcphdr *)(h + 1); 885 break; 886 #ifdef INET6 887 case 6: 888 h6 = mtod(m, struct ip6_hdr *); 889 890 /* prepare for checksum */ 891 h6->ip6_nxt = IPPROTO_TCP; 892 h6->ip6_plen = htons(sizeof(struct tcphdr)); 893 if (dir) { 894 h6->ip6_src = id->src_ip6; 895 h6->ip6_dst = id->dst_ip6; 896 } else { 897 h6->ip6_src = id->dst_ip6; 898 h6->ip6_dst = id->src_ip6; 899 } 900 901 th = (struct tcphdr *)(h6 + 1); 902 break; 903 #endif 904 } 905 906 if (dir) { 907 th->th_sport = htons(id->src_port); 908 th->th_dport = htons(id->dst_port); 909 } else { 910 th->th_sport = htons(id->dst_port); 911 th->th_dport = htons(id->src_port); 912 } 913 th->th_off = sizeof(struct tcphdr) >> 2; 914 915 if (flags & TH_RST) { 916 if (flags & TH_ACK) { 917 th->th_seq = htonl(ack); 918 th->th_flags = TH_RST; 919 } else { 920 if (flags & TH_SYN) 921 seq++; 922 th->th_ack = htonl(seq); 923 th->th_flags = TH_RST | TH_ACK; 924 } 925 } else { 926 /* 927 * Keepalive - use caller provided sequence numbers 928 */ 929 th->th_seq = htonl(seq); 930 th->th_ack = htonl(ack); 931 th->th_flags = TH_ACK; 932 } 933 934 switch (id->addr_type) { 935 case 4: 936 th->th_sum = in_cksum(m, len); 937 938 /* finish the ip header */ 939 h->ip_v = 4; 940 h->ip_hl = sizeof(*h) >> 2; 941 h->ip_tos = IPTOS_LOWDELAY; 942 h->ip_off = htons(0); 943 h->ip_len = htons(len); 944 h->ip_ttl = V_ip_defttl; 945 h->ip_sum = 0; 946 break; 947 #ifdef INET6 948 case 6: 949 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6), 950 sizeof(struct tcphdr)); 951 952 /* finish the ip6 header */ 953 h6->ip6_vfc |= IPV6_VERSION; 954 h6->ip6_hlim = IPV6_DEFHLIM; 955 break; 956 #endif 957 } 958 959 return (m); 960 } 961 962 /* 963 * Queue keepalive packets for given dynamic rule 964 */ 965 static struct mbuf ** 966 ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q) 967 { 968 struct mbuf *m_rev, *m_fwd; 969 970 m_rev = (q->state & ACK_REV) ? NULL : 971 ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 972 m_fwd = (q->state & ACK_FWD) ? NULL : 973 ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0); 974 975 if (m_rev != NULL) { 976 *mtailp = m_rev; 977 mtailp = &(*mtailp)->m_nextpkt; 978 } 979 if (m_fwd != NULL) { 980 *mtailp = m_fwd; 981 mtailp = &(*mtailp)->m_nextpkt; 982 } 983 984 return (mtailp); 985 } 986 987 /* 988 * This procedure is used to perform various maintance 989 * on dynamic hash list. Currently it is called every second. 990 */ 991 static void 992 ipfw_dyn_tick(void * vnetx) 993 { 994 struct ip_fw_chain *chain; 995 int check_ka = 0; 996 #ifdef VIMAGE 997 struct vnet *vp = vnetx; 998 #endif 999 1000 CURVNET_SET(vp); 1001 1002 chain = &V_layer3_chain; 1003 1004 /* Run keepalive checks every keepalive_period iff ka is enabled */ 1005 if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) && 1006 (V_dyn_keepalive != 0)) { 1007 V_dyn_keepalive_last = time_uptime; 1008 check_ka = 1; 1009 } 1010 1011 check_dyn_rules(chain, NULL, check_ka, 1); 1012 1013 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0); 1014 1015 CURVNET_RESTORE(); 1016 } 1017 1018 1019 /* 1020 * Walk thru all dynamic states doing generic maintance: 1021 * 1) free expired states 1022 * 2) free all states based on deleted rule / set 1023 * 3) send keepalives for states if needed 1024 * 1025 * @chain - pointer to current ipfw rules chain 1026 * @rule - delete all states originated by given rule if != NULL 1027 * @set - delete all states originated by any rule in set @set if != RESVD_SET 1028 * @check_ka - perform checking/sending keepalives 1029 * @timer - indicate call from timer routine. 1030 * 1031 * Timer routine must call this function unlocked to permit 1032 * sending keepalives/resizing table. 1033 * 1034 * Others has to call function with IPFW_UH_WLOCK held. 1035 * Additionally, function assume that dynamic rule/set is 1036 * ALREADY deleted so no new states can be generated by 1037 * 'deleted' rules. 1038 * 1039 * Write lock is needed to ensure that unused parent rules 1040 * are not freed by other instance (see stage 2, 3) 1041 */ 1042 static void 1043 check_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt, 1044 int check_ka, int timer) 1045 { 1046 struct mbuf *m0, *m, *mnext, **mtailp; 1047 struct ip *h; 1048 int i, dyn_count, new_buckets = 0, max_buckets; 1049 int expired = 0, expired_limits = 0, parents = 0, total = 0; 1050 ipfw_dyn_rule *q, *q_prev, *q_next; 1051 ipfw_dyn_rule *exp_head, **exptailp; 1052 ipfw_dyn_rule *exp_lhead, **expltailp; 1053 1054 KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated", 1055 __func__)); 1056 1057 /* Avoid possible LOR */ 1058 KASSERT(!check_ka || timer, ("%s: keepalive check with lock held", 1059 __func__)); 1060 1061 /* 1062 * Do not perform any checks if we currently have no dynamic states 1063 */ 1064 if (DYN_COUNT == 0) 1065 return; 1066 1067 /* Expired states */ 1068 exp_head = NULL; 1069 exptailp = &exp_head; 1070 1071 /* Expired limit states */ 1072 exp_lhead = NULL; 1073 expltailp = &exp_lhead; 1074 1075 /* 1076 * We make a chain of packets to go out here -- not deferring 1077 * until after we drop the IPFW dynamic rule lock would result 1078 * in a lock order reversal with the normal packet input -> ipfw 1079 * call stack. 1080 */ 1081 m0 = NULL; 1082 mtailp = &m0; 1083 1084 /* Protect from hash resizing */ 1085 if (timer != 0) 1086 IPFW_UH_WLOCK(chain); 1087 else 1088 IPFW_UH_WLOCK_ASSERT(chain); 1089 1090 #define NEXT_RULE() { q_prev = q; q = q->next ; continue; } 1091 1092 /* Stage 1: perform requested deletion */ 1093 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1094 IPFW_BUCK_LOCK(i); 1095 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) { 1096 /* account every rule */ 1097 total++; 1098 1099 /* Skip parent rules at all */ 1100 if (q->dyn_type == O_LIMIT_PARENT) { 1101 parents++; 1102 NEXT_RULE(); 1103 } 1104 1105 /* 1106 * Remove rules which are: 1107 * 1) expired 1108 * 2) matches deletion range 1109 */ 1110 if ((TIME_LEQ(q->expire, time_uptime)) || 1111 (rt != NULL && ipfw_match_range(q->rule, rt))) { 1112 if (TIME_LE(time_uptime, q->expire) && 1113 q->dyn_type == O_KEEP_STATE && 1114 V_dyn_keep_states != 0) { 1115 /* 1116 * Do not delete state if 1117 * it is not expired and 1118 * dyn_keep_states is ON. 1119 * However we need to re-link it 1120 * to any other stable rule 1121 */ 1122 q->rule = chain->default_rule; 1123 NEXT_RULE(); 1124 } 1125 1126 /* Unlink q from current list */ 1127 q_next = q->next; 1128 if (q == V_ipfw_dyn_v[i].head) 1129 V_ipfw_dyn_v[i].head = q_next; 1130 else 1131 q_prev->next = q_next; 1132 1133 q->next = NULL; 1134 1135 /* queue q to expire list */ 1136 if (q->dyn_type != O_LIMIT) { 1137 *exptailp = q; 1138 exptailp = &(*exptailp)->next; 1139 DEB(print_dyn_rule(&q->id, q->dyn_type, 1140 "unlink entry", "left"); 1141 ) 1142 } else { 1143 /* Separate list for limit rules */ 1144 *expltailp = q; 1145 expltailp = &(*expltailp)->next; 1146 expired_limits++; 1147 DEB(print_dyn_rule(&q->id, q->dyn_type, 1148 "unlink limit entry", "left"); 1149 ) 1150 } 1151 1152 q = q_next; 1153 expired++; 1154 continue; 1155 } 1156 1157 /* 1158 * Check if we need to send keepalive: 1159 * we need to ensure if is time to do KA, 1160 * this is established TCP session, and 1161 * expire time is within keepalive interval 1162 */ 1163 if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) && 1164 ((q->state & BOTH_SYN) == BOTH_SYN) && 1165 (TIME_LEQ(q->expire, time_uptime + 1166 V_dyn_keepalive_interval))) 1167 mtailp = ipfw_dyn_send_ka(mtailp, q); 1168 1169 NEXT_RULE(); 1170 } 1171 IPFW_BUCK_UNLOCK(i); 1172 } 1173 1174 /* Stage 2: decrement counters from O_LIMIT parents */ 1175 if (expired_limits != 0) { 1176 /* 1177 * XXX: Note that deleting set with more than one 1178 * heavily-used LIMIT rules can result in overwhelming 1179 * locking due to lack of per-hash value sorting 1180 * 1181 * We should probably think about: 1182 * 1) pre-allocating hash of size, say, 1183 * MAX(16, V_curr_dyn_buckets / 1024) 1184 * 2) checking if expired_limits is large enough 1185 * 3) If yes, init hash (or its part), re-link 1186 * current list and start decrementing procedure in 1187 * each bucket separately 1188 */ 1189 1190 /* 1191 * Small optimization: do not unlock bucket until 1192 * we see the next item resides in different bucket 1193 */ 1194 if (exp_lhead != NULL) { 1195 i = exp_lhead->parent->bucket; 1196 IPFW_BUCK_LOCK(i); 1197 } 1198 for (q = exp_lhead; q != NULL; q = q->next) { 1199 if (i != q->parent->bucket) { 1200 IPFW_BUCK_UNLOCK(i); 1201 i = q->parent->bucket; 1202 IPFW_BUCK_LOCK(i); 1203 } 1204 1205 /* Decrease parent refcount */ 1206 q->parent->count--; 1207 } 1208 if (exp_lhead != NULL) 1209 IPFW_BUCK_UNLOCK(i); 1210 } 1211 1212 /* 1213 * We protectet ourselves from unused parent deletion 1214 * (from the timer function) by holding UH write lock. 1215 */ 1216 1217 /* Stage 3: remove unused parent rules */ 1218 if ((parents != 0) && (expired != 0)) { 1219 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1220 IPFW_BUCK_LOCK(i); 1221 for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) { 1222 if (q->dyn_type != O_LIMIT_PARENT) 1223 NEXT_RULE(); 1224 1225 if (q->count != 0) 1226 NEXT_RULE(); 1227 1228 /* Parent rule without consumers */ 1229 1230 /* Unlink q from current list */ 1231 q_next = q->next; 1232 if (q == V_ipfw_dyn_v[i].head) 1233 V_ipfw_dyn_v[i].head = q_next; 1234 else 1235 q_prev->next = q_next; 1236 1237 q->next = NULL; 1238 1239 /* Add to expired list */ 1240 *exptailp = q; 1241 exptailp = &(*exptailp)->next; 1242 1243 DEB(print_dyn_rule(&q->id, q->dyn_type, 1244 "unlink parent entry", "left"); 1245 ) 1246 1247 expired++; 1248 1249 q = q_next; 1250 } 1251 IPFW_BUCK_UNLOCK(i); 1252 } 1253 } 1254 1255 #undef NEXT_RULE 1256 1257 if (timer != 0) { 1258 /* 1259 * Check if we need to resize hash: 1260 * if current number of states exceeds number of buckes in hash, 1261 * grow hash size to the minimum power of 2 which is bigger than 1262 * current states count. Limit hash size by 64k. 1263 */ 1264 max_buckets = (V_dyn_buckets_max > 65536) ? 1265 65536 : V_dyn_buckets_max; 1266 1267 dyn_count = DYN_COUNT; 1268 1269 if ((dyn_count > V_curr_dyn_buckets * 2) && 1270 (dyn_count < max_buckets)) { 1271 new_buckets = V_curr_dyn_buckets; 1272 while (new_buckets < dyn_count) { 1273 new_buckets *= 2; 1274 1275 if (new_buckets >= max_buckets) 1276 break; 1277 } 1278 } 1279 1280 IPFW_UH_WUNLOCK(chain); 1281 } 1282 1283 /* Finally delete old states ad limits if any */ 1284 for (q = exp_head; q != NULL; q = q_next) { 1285 q_next = q->next; 1286 uma_zfree(V_ipfw_dyn_rule_zone, q); 1287 ipfw_dyn_count--; 1288 } 1289 1290 for (q = exp_lhead; q != NULL; q = q_next) { 1291 q_next = q->next; 1292 uma_zfree(V_ipfw_dyn_rule_zone, q); 1293 ipfw_dyn_count--; 1294 } 1295 1296 /* 1297 * The rest code MUST be called from timer routine only 1298 * without holding any locks 1299 */ 1300 if (timer == 0) 1301 return; 1302 1303 /* Send keepalive packets if any */ 1304 for (m = m0; m != NULL; m = mnext) { 1305 mnext = m->m_nextpkt; 1306 m->m_nextpkt = NULL; 1307 h = mtod(m, struct ip *); 1308 if (h->ip_v == 4) 1309 ip_output(m, NULL, NULL, 0, NULL, NULL); 1310 #ifdef INET6 1311 else 1312 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1313 #endif 1314 } 1315 1316 /* Run table resize without holding any locks */ 1317 if (new_buckets != 0) 1318 resize_dynamic_table(chain, new_buckets); 1319 } 1320 1321 /* 1322 * Deletes all dynamic rules originated by given rule or all rules in 1323 * given set. Specify RESVD_SET to indicate set should not be used. 1324 * @chain - pointer to current ipfw rules chain 1325 * @rr - delete all states originated by rules in matched range. 1326 * 1327 * Function has to be called with IPFW_UH_WLOCK held. 1328 * Additionally, function assume that dynamic rule/set is 1329 * ALREADY deleted so no new states can be generated by 1330 * 'deleted' rules. 1331 */ 1332 void 1333 ipfw_expire_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt) 1334 { 1335 1336 check_dyn_rules(chain, rt, 0, 0); 1337 } 1338 1339 /* 1340 * Check if rule contains at least one dynamic opcode. 1341 * 1342 * Returns 1 if such opcode is found, 0 otherwise. 1343 */ 1344 int 1345 ipfw_is_dyn_rule(struct ip_fw *rule) 1346 { 1347 int cmdlen, l; 1348 ipfw_insn *cmd; 1349 1350 l = rule->cmd_len; 1351 cmd = rule->cmd; 1352 cmdlen = 0; 1353 for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { 1354 cmdlen = F_LEN(cmd); 1355 1356 switch (cmd->opcode) { 1357 case O_LIMIT: 1358 case O_KEEP_STATE: 1359 case O_PROBE_STATE: 1360 case O_CHECK_STATE: 1361 return (1); 1362 } 1363 } 1364 1365 return (0); 1366 } 1367 1368 void 1369 ipfw_dyn_init(struct ip_fw_chain *chain) 1370 { 1371 1372 V_ipfw_dyn_v = NULL; 1373 V_dyn_buckets_max = 256; /* must be power of 2 */ 1374 V_curr_dyn_buckets = 256; /* must be power of 2 */ 1375 1376 V_dyn_ack_lifetime = 300; 1377 V_dyn_syn_lifetime = 20; 1378 V_dyn_fin_lifetime = 1; 1379 V_dyn_rst_lifetime = 1; 1380 V_dyn_udp_lifetime = 10; 1381 V_dyn_short_lifetime = 5; 1382 1383 V_dyn_keepalive_interval = 20; 1384 V_dyn_keepalive_period = 5; 1385 V_dyn_keepalive = 1; /* do send keepalives */ 1386 V_dyn_keepalive_last = time_uptime; 1387 1388 V_dyn_max = 4096; /* max # of dynamic rules */ 1389 1390 V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule", 1391 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL, 1392 UMA_ALIGN_PTR, 0); 1393 1394 /* Enforce limit on dynamic rules */ 1395 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max); 1396 1397 callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE); 1398 1399 /* 1400 * This can potentially be done on first dynamic rule 1401 * being added to chain. 1402 */ 1403 resize_dynamic_table(chain, V_curr_dyn_buckets); 1404 } 1405 1406 void 1407 ipfw_dyn_uninit(int pass) 1408 { 1409 int i; 1410 1411 if (pass == 0) { 1412 callout_drain(&V_ipfw_timeout); 1413 return; 1414 } 1415 1416 if (V_ipfw_dyn_v != NULL) { 1417 /* 1418 * Skip deleting all dynamic states - 1419 * uma_zdestroy() does this more efficiently; 1420 */ 1421 1422 /* Destroy all mutexes */ 1423 for (i = 0 ; i < V_curr_dyn_buckets ; i++) 1424 IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]); 1425 free(V_ipfw_dyn_v, M_IPFW); 1426 V_ipfw_dyn_v = NULL; 1427 } 1428 1429 uma_zdestroy(V_ipfw_dyn_rule_zone); 1430 } 1431 1432 #ifdef SYSCTL_NODE 1433 /* 1434 * Get/set maximum number of dynamic states in given VNET instance. 1435 */ 1436 static int 1437 sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS) 1438 { 1439 int error; 1440 unsigned int nstates; 1441 1442 nstates = V_dyn_max; 1443 1444 error = sysctl_handle_int(oidp, &nstates, 0, req); 1445 /* Read operation or some error */ 1446 if ((error != 0) || (req->newptr == NULL)) 1447 return (error); 1448 1449 V_dyn_max = nstates; 1450 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max); 1451 1452 return (0); 1453 } 1454 1455 /* 1456 * Get current number of dynamic states in given VNET instance. 1457 */ 1458 static int 1459 sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS) 1460 { 1461 int error; 1462 unsigned int nstates; 1463 1464 nstates = DYN_COUNT; 1465 1466 error = sysctl_handle_int(oidp, &nstates, 0, req); 1467 1468 return (error); 1469 } 1470 #endif 1471 1472 /* 1473 * Returns size of dynamic states in legacy format 1474 */ 1475 int 1476 ipfw_dyn_len(void) 1477 { 1478 1479 return (V_ipfw_dyn_v == NULL) ? 0 : 1480 (DYN_COUNT * sizeof(ipfw_dyn_rule)); 1481 } 1482 1483 /* 1484 * Returns number of dynamic states. 1485 * Used by dump format v1 (current). 1486 */ 1487 int 1488 ipfw_dyn_get_count(void) 1489 { 1490 1491 return (V_ipfw_dyn_v == NULL) ? 0 : DYN_COUNT; 1492 } 1493 1494 static void 1495 export_dyn_rule(ipfw_dyn_rule *src, ipfw_dyn_rule *dst) 1496 { 1497 1498 memcpy(dst, src, sizeof(*src)); 1499 memcpy(&(dst->rule), &(src->rule->rulenum), sizeof(src->rule->rulenum)); 1500 /* 1501 * store set number into high word of 1502 * dst->rule pointer. 1503 */ 1504 memcpy((char *)&dst->rule + sizeof(src->rule->rulenum), 1505 &(src->rule->set), sizeof(src->rule->set)); 1506 /* 1507 * store a non-null value in "next". 1508 * The userland code will interpret a 1509 * NULL here as a marker 1510 * for the last dynamic rule. 1511 */ 1512 memcpy(&dst->next, &dst, sizeof(dst)); 1513 dst->expire = 1514 TIME_LEQ(dst->expire, time_uptime) ? 0 : dst->expire - time_uptime; 1515 } 1516 1517 /* 1518 * Fills int buffer given by @sd with dynamic states. 1519 * Used by dump format v1 (current). 1520 * 1521 * Returns 0 on success. 1522 */ 1523 int 1524 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd) 1525 { 1526 ipfw_dyn_rule *p; 1527 ipfw_obj_dyntlv *dst, *last; 1528 ipfw_obj_ctlv *ctlv; 1529 int i; 1530 size_t sz; 1531 1532 if (V_ipfw_dyn_v == NULL) 1533 return (0); 1534 1535 IPFW_UH_RLOCK_ASSERT(chain); 1536 1537 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); 1538 if (ctlv == NULL) 1539 return (ENOMEM); 1540 sz = sizeof(ipfw_obj_dyntlv); 1541 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST; 1542 ctlv->objsize = sz; 1543 last = NULL; 1544 1545 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1546 IPFW_BUCK_LOCK(i); 1547 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1548 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, sz); 1549 if (dst == NULL) { 1550 IPFW_BUCK_UNLOCK(i); 1551 return (ENOMEM); 1552 } 1553 1554 export_dyn_rule(p, &dst->state); 1555 dst->head.length = sz; 1556 dst->head.type = IPFW_TLV_DYN_ENT; 1557 last = dst; 1558 } 1559 IPFW_BUCK_UNLOCK(i); 1560 } 1561 1562 if (last != NULL) /* mark last dynamic rule */ 1563 last->head.flags = IPFW_DF_LAST; 1564 1565 return (0); 1566 } 1567 1568 /* 1569 * Fill given buffer with dynamic states (legacy format). 1570 * IPFW_UH_RLOCK has to be held while calling. 1571 */ 1572 void 1573 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep) 1574 { 1575 ipfw_dyn_rule *p, *last = NULL; 1576 char *bp; 1577 int i; 1578 1579 if (V_ipfw_dyn_v == NULL) 1580 return; 1581 bp = *pbp; 1582 1583 IPFW_UH_RLOCK_ASSERT(chain); 1584 1585 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1586 IPFW_BUCK_LOCK(i); 1587 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1588 if (bp + sizeof *p <= ep) { 1589 ipfw_dyn_rule *dst = 1590 (ipfw_dyn_rule *)bp; 1591 1592 export_dyn_rule(p, dst); 1593 last = dst; 1594 bp += sizeof(ipfw_dyn_rule); 1595 } 1596 } 1597 IPFW_BUCK_UNLOCK(i); 1598 } 1599 1600 if (last != NULL) /* mark last dynamic rule */ 1601 bzero(&last->next, sizeof(last)); 1602 *pbp = bp; 1603 } 1604 /* end of file */ 1605