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_UINT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, 210 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_buckets_max), 0, 211 "Max number of dyn. buckets"); 212 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, 213 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0, 214 "Current Number of dyn. buckets"); 215 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_count, 216 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RD, 0, 0, sysctl_ipfw_dyn_count, "IU", 217 "Number of dyn. rules"); 218 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max, 219 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 0, 0, sysctl_ipfw_dyn_max, "IU", 220 "Max number of dyn. rules"); 221 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, 222 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0, 223 "Lifetime of dyn. rules for acks"); 224 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, 225 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0, 226 "Lifetime of dyn. rules for syn"); 227 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, 228 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0, 229 "Lifetime of dyn. rules for fin"); 230 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, 231 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0, 232 "Lifetime of dyn. rules for rst"); 233 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, 234 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0, 235 "Lifetime of dyn. rules for UDP"); 236 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, 237 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0, 238 "Lifetime of dyn. rules for other situations"); 239 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, 240 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0, 241 "Enable keepalives for dyn. rules"); 242 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keep_states, 243 CTLFLAG_VNET | 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 return ntohl(i); 261 } 262 #endif 263 264 /* 265 * IMPORTANT: the hash function for dynamic rules must be commutative 266 * in source and destination (ip,port), because rules are bidirectional 267 * and we want to find both in the same bucket. 268 */ 269 static __inline int 270 hash_packet(struct ipfw_flow_id *id, int buckets) 271 { 272 u_int32_t i; 273 274 #ifdef INET6 275 if (IS_IP6_FLOW_ID(id)) 276 i = hash_packet6(id); 277 else 278 #endif /* INET6 */ 279 i = (id->dst_ip) ^ (id->src_ip); 280 i ^= (id->dst_port) ^ (id->src_port); 281 return (i & (buckets - 1)); 282 } 283 284 #if 0 285 #define DYN_DEBUG(fmt, ...) do { \ 286 printf("%s: " fmt "\n", __func__, __VA_ARGS__); \ 287 } while (0) 288 #else 289 #define DYN_DEBUG(fmt, ...) 290 #endif 291 292 static char *default_state_name = "default"; 293 struct dyn_state_obj { 294 struct named_object no; 295 char name[64]; 296 }; 297 298 #define DYN_STATE_OBJ(ch, cmd) \ 299 ((struct dyn_state_obj *)SRV_OBJECT(ch, (cmd)->arg1)) 300 /* 301 * Classifier callback. 302 * Return 0 if opcode contains object that should be referenced 303 * or rewritten. 304 */ 305 static int 306 dyn_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) 307 { 308 309 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); 310 /* Don't rewrite "check-state any" */ 311 if (cmd->arg1 == 0 && 312 cmd->opcode == O_CHECK_STATE) 313 return (1); 314 315 *puidx = cmd->arg1; 316 *ptype = 0; 317 return (0); 318 } 319 320 static void 321 dyn_update(ipfw_insn *cmd, uint16_t idx) 322 { 323 324 cmd->arg1 = idx; 325 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); 326 } 327 328 static int 329 dyn_findbyname(struct ip_fw_chain *ch, struct tid_info *ti, 330 struct named_object **pno) 331 { 332 ipfw_obj_ntlv *ntlv; 333 const char *name; 334 335 DYN_DEBUG("uidx %d", ti->uidx); 336 if (ti->uidx != 0) { 337 if (ti->tlvs == NULL) 338 return (EINVAL); 339 /* Search ntlv in the buffer provided by user */ 340 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, 341 IPFW_TLV_STATE_NAME); 342 if (ntlv == NULL) 343 return (EINVAL); 344 name = ntlv->name; 345 } else 346 name = default_state_name; 347 /* 348 * Search named object with corresponding name. 349 * Since states objects are global - ignore the set value 350 * and use zero instead. 351 */ 352 *pno = ipfw_objhash_lookup_name_type(CHAIN_TO_SRV(ch), 0, 353 IPFW_TLV_STATE_NAME, name); 354 /* 355 * We always return success here. 356 * The caller will check *pno and mark object as unresolved, 357 * then it will automatically create "default" object. 358 */ 359 return (0); 360 } 361 362 static struct named_object * 363 dyn_findbykidx(struct ip_fw_chain *ch, uint16_t idx) 364 { 365 366 DYN_DEBUG("kidx %d", idx); 367 return (ipfw_objhash_lookup_kidx(CHAIN_TO_SRV(ch), idx)); 368 } 369 370 static int 371 dyn_create(struct ip_fw_chain *ch, struct tid_info *ti, 372 uint16_t *pkidx) 373 { 374 struct namedobj_instance *ni; 375 struct dyn_state_obj *obj; 376 struct named_object *no; 377 ipfw_obj_ntlv *ntlv; 378 char *name; 379 380 DYN_DEBUG("uidx %d", ti->uidx); 381 if (ti->uidx != 0) { 382 if (ti->tlvs == NULL) 383 return (EINVAL); 384 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, 385 IPFW_TLV_STATE_NAME); 386 if (ntlv == NULL) 387 return (EINVAL); 388 name = ntlv->name; 389 } else 390 name = default_state_name; 391 392 ni = CHAIN_TO_SRV(ch); 393 obj = malloc(sizeof(*obj), M_IPFW, M_WAITOK | M_ZERO); 394 obj->no.name = obj->name; 395 obj->no.etlv = IPFW_TLV_STATE_NAME; 396 strlcpy(obj->name, name, sizeof(obj->name)); 397 398 IPFW_UH_WLOCK(ch); 399 no = ipfw_objhash_lookup_name_type(ni, 0, 400 IPFW_TLV_STATE_NAME, name); 401 if (no != NULL) { 402 /* 403 * Object is already created. 404 * Just return its kidx and bump refcount. 405 */ 406 *pkidx = no->kidx; 407 no->refcnt++; 408 IPFW_UH_WUNLOCK(ch); 409 free(obj, M_IPFW); 410 DYN_DEBUG("\tfound kidx %d", *pkidx); 411 return (0); 412 } 413 if (ipfw_objhash_alloc_idx(ni, &obj->no.kidx) != 0) { 414 DYN_DEBUG("\talloc_idx failed for %s", name); 415 IPFW_UH_WUNLOCK(ch); 416 free(obj, M_IPFW); 417 return (ENOSPC); 418 } 419 ipfw_objhash_add(ni, &obj->no); 420 IPFW_WLOCK(ch); 421 SRV_OBJECT(ch, obj->no.kidx) = obj; 422 IPFW_WUNLOCK(ch); 423 obj->no.refcnt++; 424 *pkidx = obj->no.kidx; 425 IPFW_UH_WUNLOCK(ch); 426 DYN_DEBUG("\tcreated kidx %d", *pkidx); 427 return (0); 428 } 429 430 static void 431 dyn_destroy(struct ip_fw_chain *ch, struct named_object *no) 432 { 433 struct dyn_state_obj *obj; 434 435 IPFW_UH_WLOCK_ASSERT(ch); 436 437 KASSERT(no->refcnt == 1, 438 ("Destroying object '%s' (type %u, idx %u) with refcnt %u", 439 no->name, no->etlv, no->kidx, no->refcnt)); 440 441 DYN_DEBUG("kidx %d", no->kidx); 442 IPFW_WLOCK(ch); 443 obj = SRV_OBJECT(ch, no->kidx); 444 SRV_OBJECT(ch, no->kidx) = NULL; 445 IPFW_WUNLOCK(ch); 446 ipfw_objhash_del(CHAIN_TO_SRV(ch), no); 447 ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), no->kidx); 448 449 free(obj, M_IPFW); 450 } 451 452 static struct opcode_obj_rewrite dyn_opcodes[] = { 453 { 454 O_KEEP_STATE, IPFW_TLV_STATE_NAME, 455 dyn_classify, dyn_update, 456 dyn_findbyname, dyn_findbykidx, 457 dyn_create, dyn_destroy 458 }, 459 { 460 O_CHECK_STATE, IPFW_TLV_STATE_NAME, 461 dyn_classify, dyn_update, 462 dyn_findbyname, dyn_findbykidx, 463 dyn_create, dyn_destroy 464 }, 465 { 466 O_PROBE_STATE, IPFW_TLV_STATE_NAME, 467 dyn_classify, dyn_update, 468 dyn_findbyname, dyn_findbykidx, 469 dyn_create, dyn_destroy 470 }, 471 { 472 O_LIMIT, IPFW_TLV_STATE_NAME, 473 dyn_classify, dyn_update, 474 dyn_findbyname, dyn_findbykidx, 475 dyn_create, dyn_destroy 476 }, 477 }; 478 /** 479 * Print customizable flow id description via log(9) facility. 480 */ 481 static void 482 print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags, 483 char *prefix, char *postfix) 484 { 485 struct in_addr da; 486 #ifdef INET6 487 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; 488 #else 489 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 490 #endif 491 492 #ifdef INET6 493 if (IS_IP6_FLOW_ID(id)) { 494 ip6_sprintf(src, &id->src_ip6); 495 ip6_sprintf(dst, &id->dst_ip6); 496 } else 497 #endif 498 { 499 da.s_addr = htonl(id->src_ip); 500 inet_ntop(AF_INET, &da, src, sizeof(src)); 501 da.s_addr = htonl(id->dst_ip); 502 inet_ntop(AF_INET, &da, dst, sizeof(dst)); 503 } 504 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", 505 prefix, dyn_type, src, id->src_port, dst, 506 id->dst_port, DYN_COUNT, postfix); 507 } 508 509 #define print_dyn_rule(id, dtype, prefix, postfix) \ 510 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) 511 512 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 513 #define TIME_LE(a,b) ((int)((a)-(b)) < 0) 514 515 static void 516 dyn_update_proto_state(ipfw_dyn_rule *q, const struct ipfw_flow_id *id, 517 const struct tcphdr *tcp, int dir) 518 { 519 uint32_t ack; 520 u_char flags; 521 522 if (id->proto == IPPROTO_TCP) { 523 flags = id->_flags & (TH_FIN | TH_SYN | TH_RST); 524 #define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 525 #define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 526 #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) 527 #define ACK_FWD 0x10000 /* fwd ack seen */ 528 #define ACK_REV 0x20000 /* rev ack seen */ 529 530 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8); 531 switch (q->state & TCP_FLAGS) { 532 case TH_SYN: /* opening */ 533 q->expire = time_uptime + V_dyn_syn_lifetime; 534 break; 535 536 case BOTH_SYN: /* move to established */ 537 case BOTH_SYN | TH_FIN: /* one side tries to close */ 538 case BOTH_SYN | (TH_FIN << 8): 539 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 540 if (tcp == NULL) 541 break; 542 543 ack = ntohl(tcp->th_ack); 544 if (dir == MATCH_FORWARD) { 545 if (q->ack_fwd == 0 || 546 _SEQ_GE(ack, q->ack_fwd)) { 547 q->ack_fwd = ack; 548 q->state |= ACK_FWD; 549 } 550 } else { 551 if (q->ack_rev == 0 || 552 _SEQ_GE(ack, q->ack_rev)) { 553 q->ack_rev = ack; 554 q->state |= ACK_REV; 555 } 556 } 557 if ((q->state & (ACK_FWD | ACK_REV)) == 558 (ACK_FWD | ACK_REV)) { 559 q->expire = time_uptime + V_dyn_ack_lifetime; 560 q->state &= ~(ACK_FWD | ACK_REV); 561 } 562 break; 563 564 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 565 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) 566 V_dyn_fin_lifetime = 567 V_dyn_keepalive_period - 1; 568 q->expire = time_uptime + V_dyn_fin_lifetime; 569 break; 570 571 default: 572 #if 0 573 /* 574 * reset or some invalid combination, but can also 575 * occur if we use keep-state the wrong way. 576 */ 577 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 578 printf("invalid state: 0x%x\n", q->state); 579 #endif 580 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period) 581 V_dyn_rst_lifetime = 582 V_dyn_keepalive_period - 1; 583 q->expire = time_uptime + V_dyn_rst_lifetime; 584 break; 585 } 586 } else if (id->proto == IPPROTO_UDP) { 587 q->expire = time_uptime + V_dyn_udp_lifetime; 588 } else { 589 /* other protocols */ 590 q->expire = time_uptime + V_dyn_short_lifetime; 591 } 592 } 593 594 /* 595 * Lookup a dynamic rule, locked version. 596 */ 597 static ipfw_dyn_rule * 598 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction, 599 struct tcphdr *tcp, uint16_t kidx) 600 { 601 /* 602 * Stateful ipfw extensions. 603 * Lookup into dynamic session queue. 604 */ 605 ipfw_dyn_rule *prev, *q = NULL; 606 int dir; 607 608 IPFW_BUCK_ASSERT(i); 609 610 dir = MATCH_NONE; 611 for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) { 612 if (q->dyn_type == O_LIMIT_PARENT) 613 continue; 614 615 if (pkt->proto != q->id.proto) 616 continue; 617 618 if (kidx != 0 && kidx != q->kidx) 619 continue; 620 621 if (IS_IP6_FLOW_ID(pkt)) { 622 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) && 623 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) && 624 pkt->src_port == q->id.src_port && 625 pkt->dst_port == q->id.dst_port) { 626 dir = MATCH_FORWARD; 627 break; 628 } 629 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) && 630 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) && 631 pkt->src_port == q->id.dst_port && 632 pkt->dst_port == q->id.src_port) { 633 dir = MATCH_REVERSE; 634 break; 635 } 636 } else { 637 if (pkt->src_ip == q->id.src_ip && 638 pkt->dst_ip == q->id.dst_ip && 639 pkt->src_port == q->id.src_port && 640 pkt->dst_port == q->id.dst_port) { 641 dir = MATCH_FORWARD; 642 break; 643 } 644 if (pkt->src_ip == q->id.dst_ip && 645 pkt->dst_ip == q->id.src_ip && 646 pkt->src_port == q->id.dst_port && 647 pkt->dst_port == q->id.src_port) { 648 dir = MATCH_REVERSE; 649 break; 650 } 651 } 652 } 653 if (q == NULL) 654 goto done; /* q = NULL, not found */ 655 656 if (prev != NULL) { /* found and not in front */ 657 prev->next = q->next; 658 q->next = V_ipfw_dyn_v[i].head; 659 V_ipfw_dyn_v[i].head = q; 660 } 661 662 /* update state according to flags */ 663 dyn_update_proto_state(q, pkt, tcp, dir); 664 done: 665 if (match_direction != NULL) 666 *match_direction = dir; 667 return (q); 668 } 669 670 ipfw_dyn_rule * 671 ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 672 struct tcphdr *tcp, uint16_t kidx) 673 { 674 ipfw_dyn_rule *q; 675 int i; 676 677 i = hash_packet(pkt, V_curr_dyn_buckets); 678 679 IPFW_BUCK_LOCK(i); 680 q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp, kidx); 681 if (q == NULL) 682 IPFW_BUCK_UNLOCK(i); 683 /* NB: return table locked when q is not NULL */ 684 return q; 685 } 686 687 /* 688 * Unlock bucket mtx 689 * @p - pointer to dynamic rule 690 */ 691 void 692 ipfw_dyn_unlock(ipfw_dyn_rule *q) 693 { 694 695 IPFW_BUCK_UNLOCK(q->bucket); 696 } 697 698 static int 699 resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets) 700 { 701 int i, k, nbuckets_old; 702 ipfw_dyn_rule *q; 703 struct ipfw_dyn_bucket *dyn_v, *dyn_v_old; 704 705 /* Check if given number is power of 2 and less than 64k */ 706 if ((nbuckets > 65536) || (!powerof2(nbuckets))) 707 return 1; 708 709 CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__, 710 V_curr_dyn_buckets, nbuckets); 711 712 /* Allocate and initialize new hash */ 713 dyn_v = malloc(nbuckets * sizeof(*dyn_v), M_IPFW, 714 M_WAITOK | M_ZERO); 715 716 for (i = 0 ; i < nbuckets; i++) 717 IPFW_BUCK_LOCK_INIT(&dyn_v[i]); 718 719 /* 720 * Call upper half lock, as get_map() do to ease 721 * read-only access to dynamic rules hash from sysctl 722 */ 723 IPFW_UH_WLOCK(chain); 724 725 /* 726 * Acquire chain write lock to permit hash access 727 * for main traffic path without additional locks 728 */ 729 IPFW_WLOCK(chain); 730 731 /* Save old values */ 732 nbuckets_old = V_curr_dyn_buckets; 733 dyn_v_old = V_ipfw_dyn_v; 734 735 /* Skip relinking if array is not set up */ 736 if (V_ipfw_dyn_v == NULL) 737 V_curr_dyn_buckets = 0; 738 739 /* Re-link all dynamic states */ 740 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 741 while (V_ipfw_dyn_v[i].head != NULL) { 742 /* Remove from current chain */ 743 q = V_ipfw_dyn_v[i].head; 744 V_ipfw_dyn_v[i].head = q->next; 745 746 /* Get new hash value */ 747 k = hash_packet(&q->id, nbuckets); 748 q->bucket = k; 749 /* Add to the new head */ 750 q->next = dyn_v[k].head; 751 dyn_v[k].head = q; 752 } 753 } 754 755 /* Update current pointers/buckets values */ 756 V_curr_dyn_buckets = nbuckets; 757 V_ipfw_dyn_v = dyn_v; 758 759 IPFW_WUNLOCK(chain); 760 761 IPFW_UH_WUNLOCK(chain); 762 763 /* Start periodic callout on initial creation */ 764 if (dyn_v_old == NULL) { 765 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0); 766 return (0); 767 } 768 769 /* Destroy all mutexes */ 770 for (i = 0 ; i < nbuckets_old ; i++) 771 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]); 772 773 /* Free old hash */ 774 free(dyn_v_old, M_IPFW); 775 776 return 0; 777 } 778 779 /** 780 * Install state of type 'type' for a dynamic session. 781 * The hash table contains two type of rules: 782 * - regular rules (O_KEEP_STATE) 783 * - rules for sessions with limited number of sess per user 784 * (O_LIMIT). When they are created, the parent is 785 * increased by 1, and decreased on delete. In this case, 786 * the third parameter is the parent rule and not the chain. 787 * - "parent" rules for the above (O_LIMIT_PARENT). 788 */ 789 static ipfw_dyn_rule * 790 add_dyn_rule(struct ipfw_flow_id *id, int i, uint8_t dyn_type, 791 struct ip_fw *rule, uint16_t kidx) 792 { 793 ipfw_dyn_rule *r; 794 795 IPFW_BUCK_ASSERT(i); 796 797 r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO); 798 if (r == NULL) { 799 if (last_log != time_uptime) { 800 last_log = time_uptime; 801 log(LOG_DEBUG, 802 "ipfw: Cannot allocate dynamic state, " 803 "consider increasing net.inet.ip.fw.dyn_max\n"); 804 } 805 return NULL; 806 } 807 ipfw_dyn_count++; 808 809 /* 810 * refcount on parent is already incremented, so 811 * it is safe to use parent unlocked. 812 */ 813 if (dyn_type == O_LIMIT) { 814 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 815 if ( parent->dyn_type != O_LIMIT_PARENT) 816 panic("invalid parent"); 817 r->parent = parent; 818 rule = parent->rule; 819 } 820 821 r->id = *id; 822 r->expire = time_uptime + V_dyn_syn_lifetime; 823 r->rule = rule; 824 r->dyn_type = dyn_type; 825 IPFW_ZERO_DYN_COUNTER(r); 826 r->count = 0; 827 r->kidx = kidx; 828 r->bucket = i; 829 r->next = V_ipfw_dyn_v[i].head; 830 V_ipfw_dyn_v[i].head = r; 831 DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");) 832 return r; 833 } 834 835 /** 836 * lookup dynamic parent rule using pkt and rule as search keys. 837 * If the lookup fails, then install one. 838 */ 839 static ipfw_dyn_rule * 840 lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule, 841 uint16_t kidx) 842 { 843 ipfw_dyn_rule *q; 844 int i, is_v6; 845 846 is_v6 = IS_IP6_FLOW_ID(pkt); 847 i = hash_packet( pkt, V_curr_dyn_buckets ); 848 *pindex = i; 849 IPFW_BUCK_LOCK(i); 850 for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next) 851 if (q->dyn_type == O_LIMIT_PARENT && 852 kidx == q->kidx && 853 rule == q->rule && 854 pkt->proto == q->id.proto && 855 pkt->src_port == q->id.src_port && 856 pkt->dst_port == q->id.dst_port && 857 ( 858 (is_v6 && 859 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6), 860 &(q->id.src_ip6)) && 861 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6), 862 &(q->id.dst_ip6))) || 863 (!is_v6 && 864 pkt->src_ip == q->id.src_ip && 865 pkt->dst_ip == q->id.dst_ip) 866 ) 867 ) { 868 q->expire = time_uptime + V_dyn_short_lifetime; 869 DEB(print_dyn_rule(pkt, q->dyn_type, 870 "lookup_dyn_parent found", "");) 871 return q; 872 } 873 874 /* Add virtual limiting rule */ 875 return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule, kidx); 876 } 877 878 /** 879 * Install dynamic state for rule type cmd->o.opcode 880 * 881 * Returns 1 (failure) if state is not installed because of errors or because 882 * session limitations are enforced. 883 */ 884 int 885 ipfw_install_state(struct ip_fw_chain *chain, struct ip_fw *rule, 886 ipfw_insn_limit *cmd, struct ip_fw_args *args, uint32_t tablearg) 887 { 888 ipfw_dyn_rule *q; 889 int i; 890 891 DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", 892 (cmd->o.arg1 == 0 ? "": DYN_STATE_OBJ(chain, &cmd->o)->name));) 893 894 i = hash_packet(&args->f_id, V_curr_dyn_buckets); 895 896 IPFW_BUCK_LOCK(i); 897 898 q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL, cmd->o.arg1); 899 if (q != NULL) { /* should never occur */ 900 DEB( 901 if (last_log != time_uptime) { 902 last_log = time_uptime; 903 printf("ipfw: %s: entry already present, done\n", 904 __func__); 905 }) 906 IPFW_BUCK_UNLOCK(i); 907 return (0); 908 } 909 910 /* 911 * State limiting is done via uma(9) zone limiting. 912 * Save pointer to newly-installed rule and reject 913 * packet if add_dyn_rule() returned NULL. 914 * Note q is currently set to NULL. 915 */ 916 917 switch (cmd->o.opcode) { 918 case O_KEEP_STATE: /* bidir rule */ 919 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule, 920 cmd->o.arg1); 921 break; 922 923 case O_LIMIT: { /* limit number of sessions */ 924 struct ipfw_flow_id id; 925 ipfw_dyn_rule *parent; 926 uint32_t conn_limit; 927 uint16_t limit_mask = cmd->limit_mask; 928 int pindex; 929 930 conn_limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit); 931 932 DEB( 933 if (cmd->conn_limit == IP_FW_TARG) 934 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u " 935 "(tablearg)\n", __func__, conn_limit); 936 else 937 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n", 938 __func__, conn_limit); 939 ) 940 941 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0; 942 id.proto = args->f_id.proto; 943 id.addr_type = args->f_id.addr_type; 944 id.fib = M_GETFIB(args->m); 945 946 if (IS_IP6_FLOW_ID (&(args->f_id))) { 947 bzero(&id.src_ip6, sizeof(id.src_ip6)); 948 bzero(&id.dst_ip6, sizeof(id.dst_ip6)); 949 950 if (limit_mask & DYN_SRC_ADDR) 951 id.src_ip6 = args->f_id.src_ip6; 952 if (limit_mask & DYN_DST_ADDR) 953 id.dst_ip6 = args->f_id.dst_ip6; 954 } else { 955 if (limit_mask & DYN_SRC_ADDR) 956 id.src_ip = args->f_id.src_ip; 957 if (limit_mask & DYN_DST_ADDR) 958 id.dst_ip = args->f_id.dst_ip; 959 } 960 if (limit_mask & DYN_SRC_PORT) 961 id.src_port = args->f_id.src_port; 962 if (limit_mask & DYN_DST_PORT) 963 id.dst_port = args->f_id.dst_port; 964 965 /* 966 * We have to release lock for previous bucket to 967 * avoid possible deadlock 968 */ 969 IPFW_BUCK_UNLOCK(i); 970 971 parent = lookup_dyn_parent(&id, &pindex, rule, cmd->o.arg1); 972 if (parent == NULL) { 973 printf("ipfw: %s: add parent failed\n", __func__); 974 IPFW_BUCK_UNLOCK(pindex); 975 return (1); 976 } 977 978 if (parent->count >= conn_limit) { 979 if (V_fw_verbose && last_log != time_uptime) { 980 last_log = time_uptime; 981 char sbuf[24]; 982 last_log = time_uptime; 983 snprintf(sbuf, sizeof(sbuf), 984 "%d drop session", 985 parent->rule->rulenum); 986 print_dyn_rule_flags(&args->f_id, 987 cmd->o.opcode, 988 LOG_SECURITY | LOG_DEBUG, 989 sbuf, "too many entries"); 990 } 991 IPFW_BUCK_UNLOCK(pindex); 992 return (1); 993 } 994 /* Increment counter on parent */ 995 parent->count++; 996 IPFW_BUCK_UNLOCK(pindex); 997 998 IPFW_BUCK_LOCK(i); 999 q = add_dyn_rule(&args->f_id, i, O_LIMIT, 1000 (struct ip_fw *)parent, cmd->o.arg1); 1001 if (q == NULL) { 1002 /* Decrement index and notify caller */ 1003 IPFW_BUCK_UNLOCK(i); 1004 IPFW_BUCK_LOCK(pindex); 1005 parent->count--; 1006 IPFW_BUCK_UNLOCK(pindex); 1007 return (1); 1008 } 1009 break; 1010 } 1011 default: 1012 printf("ipfw: %s: unknown dynamic rule type %u\n", 1013 __func__, cmd->o.opcode); 1014 } 1015 1016 if (q == NULL) { 1017 IPFW_BUCK_UNLOCK(i); 1018 return (1); /* Notify caller about failure */ 1019 } 1020 1021 dyn_update_proto_state(q, &args->f_id, NULL, MATCH_FORWARD); 1022 IPFW_BUCK_UNLOCK(i); 1023 return (0); 1024 } 1025 1026 /* 1027 * Generate a TCP packet, containing either a RST or a keepalive. 1028 * When flags & TH_RST, we are sending a RST packet, because of a 1029 * "reset" action matched the packet. 1030 * Otherwise we are sending a keepalive, and flags & TH_ 1031 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required 1032 * so that MAC can label the reply appropriately. 1033 */ 1034 struct mbuf * 1035 ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq, 1036 u_int32_t ack, int flags) 1037 { 1038 struct mbuf *m = NULL; /* stupid compiler */ 1039 int len, dir; 1040 struct ip *h = NULL; /* stupid compiler */ 1041 #ifdef INET6 1042 struct ip6_hdr *h6 = NULL; 1043 #endif 1044 struct tcphdr *th = NULL; 1045 1046 MGETHDR(m, M_NOWAIT, MT_DATA); 1047 if (m == NULL) 1048 return (NULL); 1049 1050 M_SETFIB(m, id->fib); 1051 #ifdef MAC 1052 if (replyto != NULL) 1053 mac_netinet_firewall_reply(replyto, m); 1054 else 1055 mac_netinet_firewall_send(m); 1056 #else 1057 (void)replyto; /* don't warn about unused arg */ 1058 #endif 1059 1060 switch (id->addr_type) { 1061 case 4: 1062 len = sizeof(struct ip) + sizeof(struct tcphdr); 1063 break; 1064 #ifdef INET6 1065 case 6: 1066 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1067 break; 1068 #endif 1069 default: 1070 /* XXX: log me?!? */ 1071 FREE_PKT(m); 1072 return (NULL); 1073 } 1074 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN); 1075 1076 m->m_data += max_linkhdr; 1077 m->m_flags |= M_SKIP_FIREWALL; 1078 m->m_pkthdr.len = m->m_len = len; 1079 m->m_pkthdr.rcvif = NULL; 1080 bzero(m->m_data, len); 1081 1082 switch (id->addr_type) { 1083 case 4: 1084 h = mtod(m, struct ip *); 1085 1086 /* prepare for checksum */ 1087 h->ip_p = IPPROTO_TCP; 1088 h->ip_len = htons(sizeof(struct tcphdr)); 1089 if (dir) { 1090 h->ip_src.s_addr = htonl(id->src_ip); 1091 h->ip_dst.s_addr = htonl(id->dst_ip); 1092 } else { 1093 h->ip_src.s_addr = htonl(id->dst_ip); 1094 h->ip_dst.s_addr = htonl(id->src_ip); 1095 } 1096 1097 th = (struct tcphdr *)(h + 1); 1098 break; 1099 #ifdef INET6 1100 case 6: 1101 h6 = mtod(m, struct ip6_hdr *); 1102 1103 /* prepare for checksum */ 1104 h6->ip6_nxt = IPPROTO_TCP; 1105 h6->ip6_plen = htons(sizeof(struct tcphdr)); 1106 if (dir) { 1107 h6->ip6_src = id->src_ip6; 1108 h6->ip6_dst = id->dst_ip6; 1109 } else { 1110 h6->ip6_src = id->dst_ip6; 1111 h6->ip6_dst = id->src_ip6; 1112 } 1113 1114 th = (struct tcphdr *)(h6 + 1); 1115 break; 1116 #endif 1117 } 1118 1119 if (dir) { 1120 th->th_sport = htons(id->src_port); 1121 th->th_dport = htons(id->dst_port); 1122 } else { 1123 th->th_sport = htons(id->dst_port); 1124 th->th_dport = htons(id->src_port); 1125 } 1126 th->th_off = sizeof(struct tcphdr) >> 2; 1127 1128 if (flags & TH_RST) { 1129 if (flags & TH_ACK) { 1130 th->th_seq = htonl(ack); 1131 th->th_flags = TH_RST; 1132 } else { 1133 if (flags & TH_SYN) 1134 seq++; 1135 th->th_ack = htonl(seq); 1136 th->th_flags = TH_RST | TH_ACK; 1137 } 1138 } else { 1139 /* 1140 * Keepalive - use caller provided sequence numbers 1141 */ 1142 th->th_seq = htonl(seq); 1143 th->th_ack = htonl(ack); 1144 th->th_flags = TH_ACK; 1145 } 1146 1147 switch (id->addr_type) { 1148 case 4: 1149 th->th_sum = in_cksum(m, len); 1150 1151 /* finish the ip header */ 1152 h->ip_v = 4; 1153 h->ip_hl = sizeof(*h) >> 2; 1154 h->ip_tos = IPTOS_LOWDELAY; 1155 h->ip_off = htons(0); 1156 h->ip_len = htons(len); 1157 h->ip_ttl = V_ip_defttl; 1158 h->ip_sum = 0; 1159 break; 1160 #ifdef INET6 1161 case 6: 1162 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6), 1163 sizeof(struct tcphdr)); 1164 1165 /* finish the ip6 header */ 1166 h6->ip6_vfc |= IPV6_VERSION; 1167 h6->ip6_hlim = IPV6_DEFHLIM; 1168 break; 1169 #endif 1170 } 1171 1172 return (m); 1173 } 1174 1175 /* 1176 * Queue keepalive packets for given dynamic rule 1177 */ 1178 static struct mbuf ** 1179 ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q) 1180 { 1181 struct mbuf *m_rev, *m_fwd; 1182 1183 m_rev = (q->state & ACK_REV) ? NULL : 1184 ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 1185 m_fwd = (q->state & ACK_FWD) ? NULL : 1186 ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0); 1187 1188 if (m_rev != NULL) { 1189 *mtailp = m_rev; 1190 mtailp = &(*mtailp)->m_nextpkt; 1191 } 1192 if (m_fwd != NULL) { 1193 *mtailp = m_fwd; 1194 mtailp = &(*mtailp)->m_nextpkt; 1195 } 1196 1197 return (mtailp); 1198 } 1199 1200 /* 1201 * This procedure is used to perform various maintenance 1202 * on dynamic hash list. Currently it is called every second. 1203 */ 1204 static void 1205 ipfw_dyn_tick(void * vnetx) 1206 { 1207 struct ip_fw_chain *chain; 1208 int check_ka = 0; 1209 #ifdef VIMAGE 1210 struct vnet *vp = vnetx; 1211 #endif 1212 1213 CURVNET_SET(vp); 1214 1215 chain = &V_layer3_chain; 1216 1217 /* Run keepalive checks every keepalive_period iff ka is enabled */ 1218 if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) && 1219 (V_dyn_keepalive != 0)) { 1220 V_dyn_keepalive_last = time_uptime; 1221 check_ka = 1; 1222 } 1223 1224 check_dyn_rules(chain, NULL, check_ka, 1); 1225 1226 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0); 1227 1228 CURVNET_RESTORE(); 1229 } 1230 1231 1232 /* 1233 * Walk through all dynamic states doing generic maintenance: 1234 * 1) free expired states 1235 * 2) free all states based on deleted rule / set 1236 * 3) send keepalives for states if needed 1237 * 1238 * @chain - pointer to current ipfw rules chain 1239 * @rule - delete all states originated by given rule if != NULL 1240 * @set - delete all states originated by any rule in set @set if != RESVD_SET 1241 * @check_ka - perform checking/sending keepalives 1242 * @timer - indicate call from timer routine. 1243 * 1244 * Timer routine must call this function unlocked to permit 1245 * sending keepalives/resizing table. 1246 * 1247 * Others has to call function with IPFW_UH_WLOCK held. 1248 * Additionally, function assume that dynamic rule/set is 1249 * ALREADY deleted so no new states can be generated by 1250 * 'deleted' rules. 1251 * 1252 * Write lock is needed to ensure that unused parent rules 1253 * are not freed by other instance (see stage 2, 3) 1254 */ 1255 static void 1256 check_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt, 1257 int check_ka, int timer) 1258 { 1259 struct mbuf *m0, *m, *mnext, **mtailp; 1260 struct ip *h; 1261 int i, dyn_count, new_buckets = 0, max_buckets; 1262 int expired = 0, expired_limits = 0, parents = 0, total = 0; 1263 ipfw_dyn_rule *q, *q_prev, *q_next; 1264 ipfw_dyn_rule *exp_head, **exptailp; 1265 ipfw_dyn_rule *exp_lhead, **expltailp; 1266 1267 KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated", 1268 __func__)); 1269 1270 /* Avoid possible LOR */ 1271 KASSERT(!check_ka || timer, ("%s: keepalive check with lock held", 1272 __func__)); 1273 1274 /* 1275 * Do not perform any checks if we currently have no dynamic states 1276 */ 1277 if (DYN_COUNT == 0) 1278 return; 1279 1280 /* Expired states */ 1281 exp_head = NULL; 1282 exptailp = &exp_head; 1283 1284 /* Expired limit states */ 1285 exp_lhead = NULL; 1286 expltailp = &exp_lhead; 1287 1288 /* 1289 * We make a chain of packets to go out here -- not deferring 1290 * until after we drop the IPFW dynamic rule lock would result 1291 * in a lock order reversal with the normal packet input -> ipfw 1292 * call stack. 1293 */ 1294 m0 = NULL; 1295 mtailp = &m0; 1296 1297 /* Protect from hash resizing */ 1298 if (timer != 0) 1299 IPFW_UH_WLOCK(chain); 1300 else 1301 IPFW_UH_WLOCK_ASSERT(chain); 1302 1303 #define NEXT_RULE() { q_prev = q; q = q->next ; continue; } 1304 1305 /* Stage 1: perform requested deletion */ 1306 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1307 IPFW_BUCK_LOCK(i); 1308 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) { 1309 /* account every rule */ 1310 total++; 1311 1312 /* Skip parent rules at all */ 1313 if (q->dyn_type == O_LIMIT_PARENT) { 1314 parents++; 1315 NEXT_RULE(); 1316 } 1317 1318 /* 1319 * Remove rules which are: 1320 * 1) expired 1321 * 2) matches deletion range 1322 */ 1323 if ((TIME_LEQ(q->expire, time_uptime)) || 1324 (rt != NULL && ipfw_match_range(q->rule, rt))) { 1325 if (TIME_LE(time_uptime, q->expire) && 1326 q->dyn_type == O_KEEP_STATE && 1327 V_dyn_keep_states != 0) { 1328 /* 1329 * Do not delete state if 1330 * it is not expired and 1331 * dyn_keep_states is ON. 1332 * However we need to re-link it 1333 * to any other stable rule 1334 */ 1335 q->rule = chain->default_rule; 1336 NEXT_RULE(); 1337 } 1338 1339 /* Unlink q from current list */ 1340 q_next = q->next; 1341 if (q == V_ipfw_dyn_v[i].head) 1342 V_ipfw_dyn_v[i].head = q_next; 1343 else 1344 q_prev->next = q_next; 1345 1346 q->next = NULL; 1347 1348 /* queue q to expire list */ 1349 if (q->dyn_type != O_LIMIT) { 1350 *exptailp = q; 1351 exptailp = &(*exptailp)->next; 1352 DEB(print_dyn_rule(&q->id, q->dyn_type, 1353 "unlink entry", "left"); 1354 ) 1355 } else { 1356 /* Separate list for limit rules */ 1357 *expltailp = q; 1358 expltailp = &(*expltailp)->next; 1359 expired_limits++; 1360 DEB(print_dyn_rule(&q->id, q->dyn_type, 1361 "unlink limit entry", "left"); 1362 ) 1363 } 1364 1365 q = q_next; 1366 expired++; 1367 continue; 1368 } 1369 1370 /* 1371 * Check if we need to send keepalive: 1372 * we need to ensure if is time to do KA, 1373 * this is established TCP session, and 1374 * expire time is within keepalive interval 1375 */ 1376 if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) && 1377 ((q->state & BOTH_SYN) == BOTH_SYN) && 1378 (TIME_LEQ(q->expire, time_uptime + 1379 V_dyn_keepalive_interval))) 1380 mtailp = ipfw_dyn_send_ka(mtailp, q); 1381 1382 NEXT_RULE(); 1383 } 1384 IPFW_BUCK_UNLOCK(i); 1385 } 1386 1387 /* Stage 2: decrement counters from O_LIMIT parents */ 1388 if (expired_limits != 0) { 1389 /* 1390 * XXX: Note that deleting set with more than one 1391 * heavily-used LIMIT rules can result in overwhelming 1392 * locking due to lack of per-hash value sorting 1393 * 1394 * We should probably think about: 1395 * 1) pre-allocating hash of size, say, 1396 * MAX(16, V_curr_dyn_buckets / 1024) 1397 * 2) checking if expired_limits is large enough 1398 * 3) If yes, init hash (or its part), re-link 1399 * current list and start decrementing procedure in 1400 * each bucket separately 1401 */ 1402 1403 /* 1404 * Small optimization: do not unlock bucket until 1405 * we see the next item resides in different bucket 1406 */ 1407 if (exp_lhead != NULL) { 1408 i = exp_lhead->parent->bucket; 1409 IPFW_BUCK_LOCK(i); 1410 } 1411 for (q = exp_lhead; q != NULL; q = q->next) { 1412 if (i != q->parent->bucket) { 1413 IPFW_BUCK_UNLOCK(i); 1414 i = q->parent->bucket; 1415 IPFW_BUCK_LOCK(i); 1416 } 1417 1418 /* Decrease parent refcount */ 1419 q->parent->count--; 1420 } 1421 if (exp_lhead != NULL) 1422 IPFW_BUCK_UNLOCK(i); 1423 } 1424 1425 /* 1426 * We protectet ourselves from unused parent deletion 1427 * (from the timer function) by holding UH write lock. 1428 */ 1429 1430 /* Stage 3: remove unused parent rules */ 1431 if ((parents != 0) && (expired != 0)) { 1432 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1433 IPFW_BUCK_LOCK(i); 1434 for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) { 1435 if (q->dyn_type != O_LIMIT_PARENT) 1436 NEXT_RULE(); 1437 1438 if (q->count != 0) 1439 NEXT_RULE(); 1440 1441 /* Parent rule without consumers */ 1442 1443 /* Unlink q from current list */ 1444 q_next = q->next; 1445 if (q == V_ipfw_dyn_v[i].head) 1446 V_ipfw_dyn_v[i].head = q_next; 1447 else 1448 q_prev->next = q_next; 1449 1450 q->next = NULL; 1451 1452 /* Add to expired list */ 1453 *exptailp = q; 1454 exptailp = &(*exptailp)->next; 1455 1456 DEB(print_dyn_rule(&q->id, q->dyn_type, 1457 "unlink parent entry", "left"); 1458 ) 1459 1460 expired++; 1461 1462 q = q_next; 1463 } 1464 IPFW_BUCK_UNLOCK(i); 1465 } 1466 } 1467 1468 #undef NEXT_RULE 1469 1470 if (timer != 0) { 1471 /* 1472 * Check if we need to resize hash: 1473 * if current number of states exceeds number of buckes in hash, 1474 * grow hash size to the minimum power of 2 which is bigger than 1475 * current states count. Limit hash size by 64k. 1476 */ 1477 max_buckets = (V_dyn_buckets_max > 65536) ? 1478 65536 : V_dyn_buckets_max; 1479 1480 dyn_count = DYN_COUNT; 1481 1482 if ((dyn_count > V_curr_dyn_buckets * 2) && 1483 (dyn_count < max_buckets)) { 1484 new_buckets = V_curr_dyn_buckets; 1485 while (new_buckets < dyn_count) { 1486 new_buckets *= 2; 1487 1488 if (new_buckets >= max_buckets) 1489 break; 1490 } 1491 } 1492 1493 IPFW_UH_WUNLOCK(chain); 1494 } 1495 1496 /* Finally delete old states ad limits if any */ 1497 for (q = exp_head; q != NULL; q = q_next) { 1498 q_next = q->next; 1499 uma_zfree(V_ipfw_dyn_rule_zone, q); 1500 ipfw_dyn_count--; 1501 } 1502 1503 for (q = exp_lhead; q != NULL; q = q_next) { 1504 q_next = q->next; 1505 uma_zfree(V_ipfw_dyn_rule_zone, q); 1506 ipfw_dyn_count--; 1507 } 1508 1509 /* 1510 * The rest code MUST be called from timer routine only 1511 * without holding any locks 1512 */ 1513 if (timer == 0) 1514 return; 1515 1516 /* Send keepalive packets if any */ 1517 for (m = m0; m != NULL; m = mnext) { 1518 mnext = m->m_nextpkt; 1519 m->m_nextpkt = NULL; 1520 h = mtod(m, struct ip *); 1521 if (h->ip_v == 4) 1522 ip_output(m, NULL, NULL, 0, NULL, NULL); 1523 #ifdef INET6 1524 else 1525 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1526 #endif 1527 } 1528 1529 /* Run table resize without holding any locks */ 1530 if (new_buckets != 0) 1531 resize_dynamic_table(chain, new_buckets); 1532 } 1533 1534 /* 1535 * Deletes all dynamic rules originated by given rule or all rules in 1536 * given set. Specify RESVD_SET to indicate set should not be used. 1537 * @chain - pointer to current ipfw rules chain 1538 * @rr - delete all states originated by rules in matched range. 1539 * 1540 * Function has to be called with IPFW_UH_WLOCK held. 1541 * Additionally, function assume that dynamic rule/set is 1542 * ALREADY deleted so no new states can be generated by 1543 * 'deleted' rules. 1544 */ 1545 void 1546 ipfw_expire_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt) 1547 { 1548 1549 check_dyn_rules(chain, rt, 0, 0); 1550 } 1551 1552 /* 1553 * Check if rule contains at least one dynamic opcode. 1554 * 1555 * Returns 1 if such opcode is found, 0 otherwise. 1556 */ 1557 int 1558 ipfw_is_dyn_rule(struct ip_fw *rule) 1559 { 1560 int cmdlen, l; 1561 ipfw_insn *cmd; 1562 1563 l = rule->cmd_len; 1564 cmd = rule->cmd; 1565 cmdlen = 0; 1566 for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { 1567 cmdlen = F_LEN(cmd); 1568 1569 switch (cmd->opcode) { 1570 case O_LIMIT: 1571 case O_KEEP_STATE: 1572 case O_PROBE_STATE: 1573 case O_CHECK_STATE: 1574 return (1); 1575 } 1576 } 1577 1578 return (0); 1579 } 1580 1581 void 1582 ipfw_dyn_init(struct ip_fw_chain *chain) 1583 { 1584 1585 V_ipfw_dyn_v = NULL; 1586 V_dyn_buckets_max = 256; /* must be power of 2 */ 1587 V_curr_dyn_buckets = 256; /* must be power of 2 */ 1588 1589 V_dyn_ack_lifetime = 300; 1590 V_dyn_syn_lifetime = 20; 1591 V_dyn_fin_lifetime = 1; 1592 V_dyn_rst_lifetime = 1; 1593 V_dyn_udp_lifetime = 10; 1594 V_dyn_short_lifetime = 5; 1595 1596 V_dyn_keepalive_interval = 20; 1597 V_dyn_keepalive_period = 5; 1598 V_dyn_keepalive = 1; /* do send keepalives */ 1599 V_dyn_keepalive_last = time_uptime; 1600 1601 V_dyn_max = 16384; /* max # of dynamic rules */ 1602 1603 V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule", 1604 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL, 1605 UMA_ALIGN_PTR, 0); 1606 1607 /* Enforce limit on dynamic rules */ 1608 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max); 1609 1610 callout_init(&V_ipfw_timeout, 1); 1611 1612 /* 1613 * This can potentially be done on first dynamic rule 1614 * being added to chain. 1615 */ 1616 resize_dynamic_table(chain, V_curr_dyn_buckets); 1617 IPFW_ADD_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); 1618 } 1619 1620 void 1621 ipfw_dyn_uninit(int pass) 1622 { 1623 int i; 1624 1625 if (pass == 0) { 1626 callout_drain(&V_ipfw_timeout); 1627 return; 1628 } 1629 IPFW_DEL_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); 1630 1631 if (V_ipfw_dyn_v != NULL) { 1632 /* 1633 * Skip deleting all dynamic states - 1634 * uma_zdestroy() does this more efficiently; 1635 */ 1636 1637 /* Destroy all mutexes */ 1638 for (i = 0 ; i < V_curr_dyn_buckets ; i++) 1639 IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]); 1640 free(V_ipfw_dyn_v, M_IPFW); 1641 V_ipfw_dyn_v = NULL; 1642 } 1643 1644 uma_zdestroy(V_ipfw_dyn_rule_zone); 1645 } 1646 1647 #ifdef SYSCTL_NODE 1648 /* 1649 * Get/set maximum number of dynamic states in given VNET instance. 1650 */ 1651 static int 1652 sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS) 1653 { 1654 int error; 1655 unsigned int nstates; 1656 1657 nstates = V_dyn_max; 1658 1659 error = sysctl_handle_int(oidp, &nstates, 0, req); 1660 /* Read operation or some error */ 1661 if ((error != 0) || (req->newptr == NULL)) 1662 return (error); 1663 1664 V_dyn_max = nstates; 1665 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max); 1666 1667 return (0); 1668 } 1669 1670 /* 1671 * Get current number of dynamic states in given VNET instance. 1672 */ 1673 static int 1674 sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS) 1675 { 1676 int error; 1677 unsigned int nstates; 1678 1679 nstates = DYN_COUNT; 1680 1681 error = sysctl_handle_int(oidp, &nstates, 0, req); 1682 1683 return (error); 1684 } 1685 #endif 1686 1687 /* 1688 * Returns size of dynamic states in legacy format 1689 */ 1690 int 1691 ipfw_dyn_len(void) 1692 { 1693 1694 return (V_ipfw_dyn_v == NULL) ? 0 : 1695 (DYN_COUNT * sizeof(ipfw_dyn_rule)); 1696 } 1697 1698 /* 1699 * Returns number of dynamic states. 1700 * Used by dump format v1 (current). 1701 */ 1702 int 1703 ipfw_dyn_get_count(void) 1704 { 1705 1706 return (V_ipfw_dyn_v == NULL) ? 0 : DYN_COUNT; 1707 } 1708 1709 static void 1710 export_dyn_rule(ipfw_dyn_rule *src, ipfw_dyn_rule *dst) 1711 { 1712 1713 memcpy(dst, src, sizeof(*src)); 1714 memcpy(&(dst->rule), &(src->rule->rulenum), sizeof(src->rule->rulenum)); 1715 /* 1716 * store set number into high word of 1717 * dst->rule pointer. 1718 */ 1719 memcpy((char *)&dst->rule + sizeof(src->rule->rulenum), 1720 &(src->rule->set), sizeof(src->rule->set)); 1721 /* 1722 * store a non-null value in "next". 1723 * The userland code will interpret a 1724 * NULL here as a marker 1725 * for the last dynamic rule. 1726 */ 1727 memcpy(&dst->next, &dst, sizeof(dst)); 1728 dst->expire = 1729 TIME_LEQ(dst->expire, time_uptime) ? 0 : dst->expire - time_uptime; 1730 } 1731 1732 /* 1733 * Fills int buffer given by @sd with dynamic states. 1734 * Used by dump format v1 (current). 1735 * 1736 * Returns 0 on success. 1737 */ 1738 int 1739 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd) 1740 { 1741 ipfw_dyn_rule *p; 1742 ipfw_obj_dyntlv *dst, *last; 1743 ipfw_obj_ctlv *ctlv; 1744 int i; 1745 size_t sz; 1746 1747 if (V_ipfw_dyn_v == NULL) 1748 return (0); 1749 1750 IPFW_UH_RLOCK_ASSERT(chain); 1751 1752 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); 1753 if (ctlv == NULL) 1754 return (ENOMEM); 1755 sz = sizeof(ipfw_obj_dyntlv); 1756 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST; 1757 ctlv->objsize = sz; 1758 last = NULL; 1759 1760 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1761 IPFW_BUCK_LOCK(i); 1762 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1763 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, sz); 1764 if (dst == NULL) { 1765 IPFW_BUCK_UNLOCK(i); 1766 return (ENOMEM); 1767 } 1768 1769 export_dyn_rule(p, &dst->state); 1770 dst->head.length = sz; 1771 dst->head.type = IPFW_TLV_DYN_ENT; 1772 last = dst; 1773 } 1774 IPFW_BUCK_UNLOCK(i); 1775 } 1776 1777 if (last != NULL) /* mark last dynamic rule */ 1778 last->head.flags = IPFW_DF_LAST; 1779 1780 return (0); 1781 } 1782 1783 /* 1784 * Fill given buffer with dynamic states (legacy format). 1785 * IPFW_UH_RLOCK has to be held while calling. 1786 */ 1787 void 1788 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep) 1789 { 1790 ipfw_dyn_rule *p, *last = NULL; 1791 char *bp; 1792 int i; 1793 1794 if (V_ipfw_dyn_v == NULL) 1795 return; 1796 bp = *pbp; 1797 1798 IPFW_UH_RLOCK_ASSERT(chain); 1799 1800 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1801 IPFW_BUCK_LOCK(i); 1802 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1803 if (bp + sizeof *p <= ep) { 1804 ipfw_dyn_rule *dst = 1805 (ipfw_dyn_rule *)bp; 1806 1807 export_dyn_rule(p, dst); 1808 last = dst; 1809 bp += sizeof(ipfw_dyn_rule); 1810 } 1811 } 1812 IPFW_BUCK_UNLOCK(i); 1813 } 1814 1815 if (last != NULL) /* mark last dynamic rule */ 1816 bzero(&last->next, sizeof(last)); 1817 *pbp = bp; 1818 } 1819 /* end of file */ 1820