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/pfil.h> 58 #include <net/vnet.h> 59 60 #include <netinet/in.h> 61 #include <netinet/ip.h> 62 #include <netinet/ip_var.h> /* ip_defttl */ 63 #include <netinet/ip_fw.h> 64 #include <netinet/tcp_var.h> 65 #include <netinet/udp.h> 66 67 #include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */ 68 #ifdef INET6 69 #include <netinet6/in6_var.h> 70 #include <netinet6/ip6_var.h> 71 #endif 72 73 #include <netpfil/ipfw/ip_fw_private.h> 74 75 #include <machine/in_cksum.h> /* XXX for in_cksum */ 76 77 #ifdef MAC 78 #include <security/mac/mac_framework.h> 79 #endif 80 81 /* 82 * Description of dynamic rules. 83 * 84 * Dynamic rules are stored in lists accessed through a hash table 85 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 86 * be modified through the sysctl variable dyn_buckets which is 87 * updated when the table becomes empty. 88 * 89 * XXX currently there is only one list, ipfw_dyn. 90 * 91 * When a packet is received, its address fields are first masked 92 * with the mask defined for the rule, then hashed, then matched 93 * against the entries in the corresponding list. 94 * Dynamic rules can be used for different purposes: 95 * + stateful rules; 96 * + enforcing limits on the number of sessions; 97 * + in-kernel NAT (not implemented yet) 98 * 99 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 100 * measured in seconds and depending on the flags. 101 * 102 * The total number of dynamic rules is equal to UMA zone items count. 103 * The max number of dynamic rules is dyn_max. When we reach 104 * the maximum number of rules we do not create anymore. This is 105 * done to avoid consuming too much memory, but also too much 106 * time when searching on each packet (ideally, we should try instead 107 * to put a limit on the length of the list on each bucket...). 108 * 109 * Each dynamic rule holds a pointer to the parent ipfw rule so 110 * we know what action to perform. Dynamic rules are removed when 111 * the parent rule is deleted. This can be changed by dyn_keep_states 112 * sysctl. 113 * 114 * There are some limitations with dynamic rules -- we do not 115 * obey the 'randomized match', and we do not do multiple 116 * passes through the firewall. XXX check the latter!!! 117 */ 118 119 struct ipfw_dyn_bucket { 120 struct mtx mtx; /* Bucket protecting lock */ 121 ipfw_dyn_rule *head; /* Pointer to first rule */ 122 }; 123 124 /* 125 * Static variables followed by global ones 126 */ 127 static VNET_DEFINE(struct ipfw_dyn_bucket *, ipfw_dyn_v); 128 static VNET_DEFINE(u_int32_t, dyn_buckets_max); 129 static VNET_DEFINE(u_int32_t, curr_dyn_buckets); 130 static VNET_DEFINE(struct callout, ipfw_timeout); 131 #define V_ipfw_dyn_v VNET(ipfw_dyn_v) 132 #define V_dyn_buckets_max VNET(dyn_buckets_max) 133 #define V_curr_dyn_buckets VNET(curr_dyn_buckets) 134 #define V_ipfw_timeout VNET(ipfw_timeout) 135 136 static VNET_DEFINE(uma_zone_t, ipfw_dyn_rule_zone); 137 #define V_ipfw_dyn_rule_zone VNET(ipfw_dyn_rule_zone) 138 139 #define IPFW_BUCK_LOCK_INIT(b) \ 140 mtx_init(&(b)->mtx, "IPFW dynamic bucket", NULL, MTX_DEF) 141 #define IPFW_BUCK_LOCK_DESTROY(b) \ 142 mtx_destroy(&(b)->mtx) 143 #define IPFW_BUCK_LOCK(i) mtx_lock(&V_ipfw_dyn_v[(i)].mtx) 144 #define IPFW_BUCK_UNLOCK(i) mtx_unlock(&V_ipfw_dyn_v[(i)].mtx) 145 #define IPFW_BUCK_ASSERT(i) mtx_assert(&V_ipfw_dyn_v[(i)].mtx, MA_OWNED) 146 147 148 static VNET_DEFINE(int, dyn_keep_states); 149 #define V_dyn_keep_states VNET(dyn_keep_states) 150 151 /* 152 * Timeouts for various events in handing dynamic rules. 153 */ 154 static VNET_DEFINE(u_int32_t, dyn_ack_lifetime); 155 static VNET_DEFINE(u_int32_t, dyn_syn_lifetime); 156 static VNET_DEFINE(u_int32_t, dyn_fin_lifetime); 157 static VNET_DEFINE(u_int32_t, dyn_rst_lifetime); 158 static VNET_DEFINE(u_int32_t, dyn_udp_lifetime); 159 static VNET_DEFINE(u_int32_t, dyn_short_lifetime); 160 161 #define V_dyn_ack_lifetime VNET(dyn_ack_lifetime) 162 #define V_dyn_syn_lifetime VNET(dyn_syn_lifetime) 163 #define V_dyn_fin_lifetime VNET(dyn_fin_lifetime) 164 #define V_dyn_rst_lifetime VNET(dyn_rst_lifetime) 165 #define V_dyn_udp_lifetime VNET(dyn_udp_lifetime) 166 #define V_dyn_short_lifetime VNET(dyn_short_lifetime) 167 168 /* 169 * Keepalives are sent if dyn_keepalive is set. They are sent every 170 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 171 * seconds of lifetime of a rule. 172 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 173 * than dyn_keepalive_period. 174 */ 175 176 static VNET_DEFINE(u_int32_t, dyn_keepalive_interval); 177 static VNET_DEFINE(u_int32_t, dyn_keepalive_period); 178 static VNET_DEFINE(u_int32_t, dyn_keepalive); 179 static VNET_DEFINE(time_t, dyn_keepalive_last); 180 181 #define V_dyn_keepalive_interval VNET(dyn_keepalive_interval) 182 #define V_dyn_keepalive_period VNET(dyn_keepalive_period) 183 #define V_dyn_keepalive VNET(dyn_keepalive) 184 #define V_dyn_keepalive_last VNET(dyn_keepalive_last) 185 186 static VNET_DEFINE(u_int32_t, dyn_max); /* max # of dynamic rules */ 187 188 #define DYN_COUNT uma_zone_get_cur(V_ipfw_dyn_rule_zone) 189 #define V_dyn_max VNET(dyn_max) 190 191 /* for userspace, we emulate the uma_zone_counter with ipfw_dyn_count */ 192 static int ipfw_dyn_count; /* number of objects */ 193 194 #ifdef USERSPACE /* emulation of UMA object counters for userspace */ 195 #define uma_zone_get_cur(x) ipfw_dyn_count 196 #endif /* USERSPACE */ 197 198 static int last_log; /* Log ratelimiting */ 199 200 static void ipfw_dyn_tick(void *vnetx); 201 static void check_dyn_rules(struct ip_fw_chain *, ipfw_range_tlv *, int, int); 202 #ifdef SYSCTL_NODE 203 204 static int sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS); 205 static int sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS); 206 207 SYSBEGIN(f2) 208 209 SYSCTL_DECL(_net_inet_ip_fw); 210 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, 211 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_buckets_max), 0, 212 "Max number of dyn. buckets"); 213 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, 214 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0, 215 "Current Number of dyn. buckets"); 216 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_count, 217 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RD, 0, 0, sysctl_ipfw_dyn_count, "IU", 218 "Number of dyn. rules"); 219 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max, 220 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 0, 0, sysctl_ipfw_dyn_max, "IU", 221 "Max number of dyn. rules"); 222 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, 223 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0, 224 "Lifetime of dyn. rules for acks"); 225 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, 226 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0, 227 "Lifetime of dyn. rules for syn"); 228 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, 229 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0, 230 "Lifetime of dyn. rules for fin"); 231 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, 232 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0, 233 "Lifetime of dyn. rules for rst"); 234 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, 235 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0, 236 "Lifetime of dyn. rules for UDP"); 237 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, 238 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0, 239 "Lifetime of dyn. rules for other situations"); 240 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, 241 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0, 242 "Enable keepalives for dyn. rules"); 243 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keep_states, 244 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0, 245 "Do not flush dynamic states on rule deletion"); 246 247 SYSEND 248 249 #endif /* SYSCTL_NODE */ 250 251 252 #ifdef INET6 253 static __inline int 254 hash_packet6(struct ipfw_flow_id *id) 255 { 256 u_int32_t i; 257 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^ 258 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^ 259 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^ 260 (id->src_ip6.__u6_addr.__u6_addr32[3]); 261 return ntohl(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); 281 i ^= (id->dst_port) ^ (id->src_port); 282 return (i & (buckets - 1)); 283 } 284 285 #if 0 286 #define DYN_DEBUG(fmt, ...) do { \ 287 printf("%s: " fmt "\n", __func__, __VA_ARGS__); \ 288 } while (0) 289 #else 290 #define DYN_DEBUG(fmt, ...) 291 #endif 292 293 static char *default_state_name = "default"; 294 struct dyn_state_obj { 295 struct named_object no; 296 char name[64]; 297 }; 298 299 #define DYN_STATE_OBJ(ch, cmd) \ 300 ((struct dyn_state_obj *)SRV_OBJECT(ch, (cmd)->arg1)) 301 /* 302 * Classifier callback. 303 * Return 0 if opcode contains object that should be referenced 304 * or rewritten. 305 */ 306 static int 307 dyn_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) 308 { 309 310 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); 311 /* Don't rewrite "check-state any" */ 312 if (cmd->arg1 == 0 && 313 cmd->opcode == O_CHECK_STATE) 314 return (1); 315 316 *puidx = cmd->arg1; 317 *ptype = 0; 318 return (0); 319 } 320 321 static void 322 dyn_update(ipfw_insn *cmd, uint16_t idx) 323 { 324 325 cmd->arg1 = idx; 326 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); 327 } 328 329 static int 330 dyn_findbyname(struct ip_fw_chain *ch, struct tid_info *ti, 331 struct named_object **pno) 332 { 333 ipfw_obj_ntlv *ntlv; 334 const char *name; 335 336 DYN_DEBUG("uidx %d", ti->uidx); 337 if (ti->uidx != 0) { 338 if (ti->tlvs == NULL) 339 return (EINVAL); 340 /* Search ntlv in the buffer provided by user */ 341 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, 342 IPFW_TLV_STATE_NAME); 343 if (ntlv == NULL) 344 return (EINVAL); 345 name = ntlv->name; 346 } else 347 name = default_state_name; 348 /* 349 * Search named object with corresponding name. 350 * Since states objects are global - ignore the set value 351 * and use zero instead. 352 */ 353 *pno = ipfw_objhash_lookup_name_type(CHAIN_TO_SRV(ch), 0, 354 IPFW_TLV_STATE_NAME, name); 355 /* 356 * We always return success here. 357 * The caller will check *pno and mark object as unresolved, 358 * then it will automatically create "default" object. 359 */ 360 return (0); 361 } 362 363 static struct named_object * 364 dyn_findbykidx(struct ip_fw_chain *ch, uint16_t idx) 365 { 366 367 DYN_DEBUG("kidx %d", idx); 368 return (ipfw_objhash_lookup_kidx(CHAIN_TO_SRV(ch), idx)); 369 } 370 371 static int 372 dyn_create(struct ip_fw_chain *ch, struct tid_info *ti, 373 uint16_t *pkidx) 374 { 375 struct namedobj_instance *ni; 376 struct dyn_state_obj *obj; 377 struct named_object *no; 378 ipfw_obj_ntlv *ntlv; 379 char *name; 380 381 DYN_DEBUG("uidx %d", ti->uidx); 382 if (ti->uidx != 0) { 383 if (ti->tlvs == NULL) 384 return (EINVAL); 385 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, 386 IPFW_TLV_STATE_NAME); 387 if (ntlv == NULL) 388 return (EINVAL); 389 name = ntlv->name; 390 } else 391 name = default_state_name; 392 393 ni = CHAIN_TO_SRV(ch); 394 obj = malloc(sizeof(*obj), M_IPFW, M_WAITOK | M_ZERO); 395 obj->no.name = obj->name; 396 obj->no.etlv = IPFW_TLV_STATE_NAME; 397 strlcpy(obj->name, name, sizeof(obj->name)); 398 399 IPFW_UH_WLOCK(ch); 400 no = ipfw_objhash_lookup_name_type(ni, 0, 401 IPFW_TLV_STATE_NAME, name); 402 if (no != NULL) { 403 /* 404 * Object is already created. 405 * Just return its kidx and bump refcount. 406 */ 407 *pkidx = no->kidx; 408 no->refcnt++; 409 IPFW_UH_WUNLOCK(ch); 410 free(obj, M_IPFW); 411 DYN_DEBUG("\tfound kidx %d", *pkidx); 412 return (0); 413 } 414 if (ipfw_objhash_alloc_idx(ni, &obj->no.kidx) != 0) { 415 DYN_DEBUG("\talloc_idx failed for %s", name); 416 IPFW_UH_WUNLOCK(ch); 417 free(obj, M_IPFW); 418 return (ENOSPC); 419 } 420 ipfw_objhash_add(ni, &obj->no); 421 IPFW_WLOCK(ch); 422 SRV_OBJECT(ch, obj->no.kidx) = obj; 423 IPFW_WUNLOCK(ch); 424 obj->no.refcnt++; 425 *pkidx = obj->no.kidx; 426 IPFW_UH_WUNLOCK(ch); 427 DYN_DEBUG("\tcreated kidx %d", *pkidx); 428 return (0); 429 } 430 431 static void 432 dyn_destroy(struct ip_fw_chain *ch, struct named_object *no) 433 { 434 struct dyn_state_obj *obj; 435 436 IPFW_UH_WLOCK_ASSERT(ch); 437 438 KASSERT(no->refcnt == 1, 439 ("Destroying object '%s' (type %u, idx %u) with refcnt %u", 440 no->name, no->etlv, no->kidx, no->refcnt)); 441 442 DYN_DEBUG("kidx %d", no->kidx); 443 IPFW_WLOCK(ch); 444 obj = SRV_OBJECT(ch, no->kidx); 445 SRV_OBJECT(ch, no->kidx) = NULL; 446 IPFW_WUNLOCK(ch); 447 ipfw_objhash_del(CHAIN_TO_SRV(ch), no); 448 ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), no->kidx); 449 450 free(obj, M_IPFW); 451 } 452 453 static struct opcode_obj_rewrite dyn_opcodes[] = { 454 { 455 O_KEEP_STATE, IPFW_TLV_STATE_NAME, 456 dyn_classify, dyn_update, 457 dyn_findbyname, dyn_findbykidx, 458 dyn_create, dyn_destroy 459 }, 460 { 461 O_CHECK_STATE, IPFW_TLV_STATE_NAME, 462 dyn_classify, dyn_update, 463 dyn_findbyname, dyn_findbykidx, 464 dyn_create, dyn_destroy 465 }, 466 { 467 O_PROBE_STATE, IPFW_TLV_STATE_NAME, 468 dyn_classify, dyn_update, 469 dyn_findbyname, dyn_findbykidx, 470 dyn_create, dyn_destroy 471 }, 472 { 473 O_LIMIT, IPFW_TLV_STATE_NAME, 474 dyn_classify, dyn_update, 475 dyn_findbyname, dyn_findbykidx, 476 dyn_create, dyn_destroy 477 }, 478 }; 479 /** 480 * Print customizable flow id description via log(9) facility. 481 */ 482 static void 483 print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags, 484 char *prefix, char *postfix) 485 { 486 struct in_addr da; 487 #ifdef INET6 488 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; 489 #else 490 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 491 #endif 492 493 #ifdef INET6 494 if (IS_IP6_FLOW_ID(id)) { 495 ip6_sprintf(src, &id->src_ip6); 496 ip6_sprintf(dst, &id->dst_ip6); 497 } else 498 #endif 499 { 500 da.s_addr = htonl(id->src_ip); 501 inet_ntop(AF_INET, &da, src, sizeof(src)); 502 da.s_addr = htonl(id->dst_ip); 503 inet_ntop(AF_INET, &da, dst, sizeof(dst)); 504 } 505 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", 506 prefix, dyn_type, src, id->src_port, dst, 507 id->dst_port, DYN_COUNT, postfix); 508 } 509 510 #define print_dyn_rule(id, dtype, prefix, postfix) \ 511 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) 512 513 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 514 #define TIME_LE(a,b) ((int)((a)-(b)) < 0) 515 516 static void 517 dyn_update_proto_state(ipfw_dyn_rule *q, const struct ipfw_flow_id *id, 518 const struct tcphdr *tcp, int dir) 519 { 520 uint32_t ack; 521 u_char flags; 522 523 if (id->proto == IPPROTO_TCP) { 524 flags = id->_flags & (TH_FIN | TH_SYN | TH_RST); 525 #define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 526 #define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 527 #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) 528 #define ACK_FWD 0x10000 /* fwd ack seen */ 529 #define ACK_REV 0x20000 /* rev ack seen */ 530 531 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8); 532 switch (q->state & TCP_FLAGS) { 533 case TH_SYN: /* opening */ 534 q->expire = time_uptime + V_dyn_syn_lifetime; 535 break; 536 537 case BOTH_SYN: /* move to established */ 538 case BOTH_SYN | TH_FIN: /* one side tries to close */ 539 case BOTH_SYN | (TH_FIN << 8): 540 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 541 if (tcp == NULL) 542 break; 543 544 ack = ntohl(tcp->th_ack); 545 if (dir == MATCH_FORWARD) { 546 if (q->ack_fwd == 0 || 547 _SEQ_GE(ack, q->ack_fwd)) { 548 q->ack_fwd = ack; 549 q->state |= ACK_FWD; 550 } 551 } else { 552 if (q->ack_rev == 0 || 553 _SEQ_GE(ack, q->ack_rev)) { 554 q->ack_rev = ack; 555 q->state |= ACK_REV; 556 } 557 } 558 if ((q->state & (ACK_FWD | ACK_REV)) == 559 (ACK_FWD | ACK_REV)) { 560 q->expire = time_uptime + V_dyn_ack_lifetime; 561 q->state &= ~(ACK_FWD | ACK_REV); 562 } 563 break; 564 565 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 566 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) 567 V_dyn_fin_lifetime = 568 V_dyn_keepalive_period - 1; 569 q->expire = time_uptime + V_dyn_fin_lifetime; 570 break; 571 572 default: 573 #if 0 574 /* 575 * reset or some invalid combination, but can also 576 * occur if we use keep-state the wrong way. 577 */ 578 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 579 printf("invalid state: 0x%x\n", q->state); 580 #endif 581 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period) 582 V_dyn_rst_lifetime = 583 V_dyn_keepalive_period - 1; 584 q->expire = time_uptime + V_dyn_rst_lifetime; 585 break; 586 } 587 } else if (id->proto == IPPROTO_UDP) { 588 q->expire = time_uptime + V_dyn_udp_lifetime; 589 } else { 590 /* other protocols */ 591 q->expire = time_uptime + V_dyn_short_lifetime; 592 } 593 } 594 595 /* 596 * Lookup a dynamic rule, locked version. 597 */ 598 static ipfw_dyn_rule * 599 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction, 600 struct tcphdr *tcp, uint16_t kidx) 601 { 602 /* 603 * Stateful ipfw extensions. 604 * Lookup into dynamic session queue. 605 */ 606 ipfw_dyn_rule *prev, *q = NULL; 607 int dir; 608 609 IPFW_BUCK_ASSERT(i); 610 611 dir = MATCH_NONE; 612 for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) { 613 if (q->dyn_type == O_LIMIT_PARENT) 614 continue; 615 616 if (pkt->proto != q->id.proto) 617 continue; 618 619 if (kidx != 0 && kidx != q->kidx) 620 continue; 621 622 if (IS_IP6_FLOW_ID(pkt)) { 623 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) && 624 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) && 625 pkt->src_port == q->id.src_port && 626 pkt->dst_port == q->id.dst_port) { 627 dir = MATCH_FORWARD; 628 break; 629 } 630 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) && 631 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) && 632 pkt->src_port == q->id.dst_port && 633 pkt->dst_port == q->id.src_port) { 634 dir = MATCH_REVERSE; 635 break; 636 } 637 } else { 638 if (pkt->src_ip == q->id.src_ip && 639 pkt->dst_ip == q->id.dst_ip && 640 pkt->src_port == q->id.src_port && 641 pkt->dst_port == q->id.dst_port) { 642 dir = MATCH_FORWARD; 643 break; 644 } 645 if (pkt->src_ip == q->id.dst_ip && 646 pkt->dst_ip == q->id.src_ip && 647 pkt->src_port == q->id.dst_port && 648 pkt->dst_port == q->id.src_port) { 649 dir = MATCH_REVERSE; 650 break; 651 } 652 } 653 } 654 if (q == NULL) 655 goto done; /* q = NULL, not found */ 656 657 if (prev != NULL) { /* found and not in front */ 658 prev->next = q->next; 659 q->next = V_ipfw_dyn_v[i].head; 660 V_ipfw_dyn_v[i].head = q; 661 } 662 663 /* update state according to flags */ 664 dyn_update_proto_state(q, pkt, tcp, dir); 665 done: 666 if (match_direction != NULL) 667 *match_direction = dir; 668 return (q); 669 } 670 671 ipfw_dyn_rule * 672 ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 673 struct tcphdr *tcp, uint16_t kidx) 674 { 675 ipfw_dyn_rule *q; 676 int i; 677 678 i = hash_packet(pkt, V_curr_dyn_buckets); 679 680 IPFW_BUCK_LOCK(i); 681 q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp, kidx); 682 if (q == NULL) 683 IPFW_BUCK_UNLOCK(i); 684 /* NB: return table locked when q is not NULL */ 685 return q; 686 } 687 688 /* 689 * Unlock bucket mtx 690 * @p - pointer to dynamic rule 691 */ 692 void 693 ipfw_dyn_unlock(ipfw_dyn_rule *q) 694 { 695 696 IPFW_BUCK_UNLOCK(q->bucket); 697 } 698 699 static int 700 resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets) 701 { 702 int i, k, nbuckets_old; 703 ipfw_dyn_rule *q; 704 struct ipfw_dyn_bucket *dyn_v, *dyn_v_old; 705 706 /* Check if given number is power of 2 and less than 64k */ 707 if ((nbuckets > 65536) || (!powerof2(nbuckets))) 708 return 1; 709 710 CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__, 711 V_curr_dyn_buckets, nbuckets); 712 713 /* Allocate and initialize new hash */ 714 dyn_v = malloc(nbuckets * sizeof(*dyn_v), M_IPFW, 715 M_WAITOK | M_ZERO); 716 717 for (i = 0 ; i < nbuckets; i++) 718 IPFW_BUCK_LOCK_INIT(&dyn_v[i]); 719 720 /* 721 * Call upper half lock, as get_map() do to ease 722 * read-only access to dynamic rules hash from sysctl 723 */ 724 IPFW_UH_WLOCK(chain); 725 726 /* 727 * Acquire chain write lock to permit hash access 728 * for main traffic path without additional locks 729 */ 730 IPFW_WLOCK(chain); 731 732 /* Save old values */ 733 nbuckets_old = V_curr_dyn_buckets; 734 dyn_v_old = V_ipfw_dyn_v; 735 736 /* Skip relinking if array is not set up */ 737 if (V_ipfw_dyn_v == NULL) 738 V_curr_dyn_buckets = 0; 739 740 /* Re-link all dynamic states */ 741 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 742 while (V_ipfw_dyn_v[i].head != NULL) { 743 /* Remove from current chain */ 744 q = V_ipfw_dyn_v[i].head; 745 V_ipfw_dyn_v[i].head = q->next; 746 747 /* Get new hash value */ 748 k = hash_packet(&q->id, nbuckets); 749 q->bucket = k; 750 /* Add to the new head */ 751 q->next = dyn_v[k].head; 752 dyn_v[k].head = q; 753 } 754 } 755 756 /* Update current pointers/buckets values */ 757 V_curr_dyn_buckets = nbuckets; 758 V_ipfw_dyn_v = dyn_v; 759 760 IPFW_WUNLOCK(chain); 761 762 IPFW_UH_WUNLOCK(chain); 763 764 /* Start periodic callout on initial creation */ 765 if (dyn_v_old == NULL) { 766 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0); 767 return (0); 768 } 769 770 /* Destroy all mutexes */ 771 for (i = 0 ; i < nbuckets_old ; i++) 772 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]); 773 774 /* Free old hash */ 775 free(dyn_v_old, M_IPFW); 776 777 return 0; 778 } 779 780 /** 781 * Install state of type 'type' for a dynamic session. 782 * The hash table contains two type of rules: 783 * - regular rules (O_KEEP_STATE) 784 * - rules for sessions with limited number of sess per user 785 * (O_LIMIT). When they are created, the parent is 786 * increased by 1, and decreased on delete. In this case, 787 * the third parameter is the parent rule and not the chain. 788 * - "parent" rules for the above (O_LIMIT_PARENT). 789 */ 790 static ipfw_dyn_rule * 791 add_dyn_rule(struct ipfw_flow_id *id, int i, uint8_t dyn_type, 792 struct ip_fw *rule, uint16_t kidx) 793 { 794 ipfw_dyn_rule *r; 795 796 IPFW_BUCK_ASSERT(i); 797 798 r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO); 799 if (r == NULL) { 800 if (last_log != time_uptime) { 801 last_log = time_uptime; 802 log(LOG_DEBUG, 803 "ipfw: Cannot allocate dynamic state, " 804 "consider increasing net.inet.ip.fw.dyn_max\n"); 805 } 806 return NULL; 807 } 808 ipfw_dyn_count++; 809 810 /* 811 * refcount on parent is already incremented, so 812 * it is safe to use parent unlocked. 813 */ 814 if (dyn_type == O_LIMIT) { 815 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 816 if ( parent->dyn_type != O_LIMIT_PARENT) 817 panic("invalid parent"); 818 r->parent = parent; 819 rule = parent->rule; 820 } 821 822 r->id = *id; 823 r->expire = time_uptime + V_dyn_syn_lifetime; 824 r->rule = rule; 825 r->dyn_type = dyn_type; 826 IPFW_ZERO_DYN_COUNTER(r); 827 r->count = 0; 828 r->kidx = kidx; 829 r->bucket = i; 830 r->next = V_ipfw_dyn_v[i].head; 831 V_ipfw_dyn_v[i].head = r; 832 DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");) 833 return r; 834 } 835 836 /** 837 * lookup dynamic parent rule using pkt and rule as search keys. 838 * If the lookup fails, then install one. 839 */ 840 static ipfw_dyn_rule * 841 lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule, 842 uint16_t kidx) 843 { 844 ipfw_dyn_rule *q; 845 int i, is_v6; 846 847 is_v6 = IS_IP6_FLOW_ID(pkt); 848 i = hash_packet( pkt, V_curr_dyn_buckets ); 849 *pindex = i; 850 IPFW_BUCK_LOCK(i); 851 for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next) 852 if (q->dyn_type == O_LIMIT_PARENT && 853 kidx == q->kidx && 854 rule == q->rule && 855 pkt->proto == q->id.proto && 856 pkt->src_port == q->id.src_port && 857 pkt->dst_port == q->id.dst_port && 858 ( 859 (is_v6 && 860 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6), 861 &(q->id.src_ip6)) && 862 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6), 863 &(q->id.dst_ip6))) || 864 (!is_v6 && 865 pkt->src_ip == q->id.src_ip && 866 pkt->dst_ip == q->id.dst_ip) 867 ) 868 ) { 869 q->expire = time_uptime + V_dyn_short_lifetime; 870 DEB(print_dyn_rule(pkt, q->dyn_type, 871 "lookup_dyn_parent found", "");) 872 return q; 873 } 874 875 /* Add virtual limiting rule */ 876 return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule, kidx); 877 } 878 879 /** 880 * Install dynamic state for rule type cmd->o.opcode 881 * 882 * Returns 1 (failure) if state is not installed because of errors or because 883 * session limitations are enforced. 884 */ 885 int 886 ipfw_install_state(struct ip_fw_chain *chain, struct ip_fw *rule, 887 ipfw_insn_limit *cmd, struct ip_fw_args *args, uint32_t tablearg) 888 { 889 ipfw_dyn_rule *q; 890 int i; 891 892 DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", 893 (cmd->o.arg1 == 0 ? "": DYN_STATE_OBJ(chain, &cmd->o)->name));) 894 895 i = hash_packet(&args->f_id, V_curr_dyn_buckets); 896 897 IPFW_BUCK_LOCK(i); 898 899 q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL, cmd->o.arg1); 900 if (q != NULL) { /* should never occur */ 901 DEB( 902 if (last_log != time_uptime) { 903 last_log = time_uptime; 904 printf("ipfw: %s: entry already present, done\n", 905 __func__); 906 }) 907 IPFW_BUCK_UNLOCK(i); 908 return (0); 909 } 910 911 /* 912 * State limiting is done via uma(9) zone limiting. 913 * Save pointer to newly-installed rule and reject 914 * packet if add_dyn_rule() returned NULL. 915 * Note q is currently set to NULL. 916 */ 917 918 switch (cmd->o.opcode) { 919 case O_KEEP_STATE: /* bidir rule */ 920 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule, 921 cmd->o.arg1); 922 break; 923 924 case O_LIMIT: { /* limit number of sessions */ 925 struct ipfw_flow_id id; 926 ipfw_dyn_rule *parent; 927 uint32_t conn_limit; 928 uint16_t limit_mask = cmd->limit_mask; 929 int pindex; 930 931 conn_limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit); 932 933 DEB( 934 if (cmd->conn_limit == IP_FW_TARG) 935 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u " 936 "(tablearg)\n", __func__, conn_limit); 937 else 938 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n", 939 __func__, conn_limit); 940 ) 941 942 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0; 943 id.proto = args->f_id.proto; 944 id.addr_type = args->f_id.addr_type; 945 id.fib = M_GETFIB(args->m); 946 947 if (IS_IP6_FLOW_ID (&(args->f_id))) { 948 bzero(&id.src_ip6, sizeof(id.src_ip6)); 949 bzero(&id.dst_ip6, sizeof(id.dst_ip6)); 950 951 if (limit_mask & DYN_SRC_ADDR) 952 id.src_ip6 = args->f_id.src_ip6; 953 if (limit_mask & DYN_DST_ADDR) 954 id.dst_ip6 = args->f_id.dst_ip6; 955 } else { 956 if (limit_mask & DYN_SRC_ADDR) 957 id.src_ip = args->f_id.src_ip; 958 if (limit_mask & DYN_DST_ADDR) 959 id.dst_ip = args->f_id.dst_ip; 960 } 961 if (limit_mask & DYN_SRC_PORT) 962 id.src_port = args->f_id.src_port; 963 if (limit_mask & DYN_DST_PORT) 964 id.dst_port = args->f_id.dst_port; 965 966 /* 967 * We have to release lock for previous bucket to 968 * avoid possible deadlock 969 */ 970 IPFW_BUCK_UNLOCK(i); 971 972 parent = lookup_dyn_parent(&id, &pindex, rule, cmd->o.arg1); 973 if (parent == NULL) { 974 printf("ipfw: %s: add parent failed\n", __func__); 975 IPFW_BUCK_UNLOCK(pindex); 976 return (1); 977 } 978 979 if (parent->count >= conn_limit) { 980 if (V_fw_verbose && 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 uint16_t rulenum; 1713 1714 rulenum = (uint16_t)src->rule->rulenum; 1715 memcpy(dst, src, sizeof(*src)); 1716 memcpy(&dst->rule, &rulenum, sizeof(rulenum)); 1717 /* 1718 * store set number into high word of 1719 * dst->rule pointer. 1720 */ 1721 memcpy((char *)&dst->rule + sizeof(rulenum), &src->rule->set, 1722 sizeof(src->rule->set)); 1723 /* 1724 * store a non-null value in "next". 1725 * The userland code will interpret a 1726 * NULL here as a marker 1727 * for the last dynamic rule. 1728 */ 1729 memcpy(&dst->next, &dst, sizeof(dst)); 1730 dst->expire = TIME_LEQ(dst->expire, time_uptime) ? 0: 1731 dst->expire - time_uptime; 1732 } 1733 1734 /* 1735 * Fills int buffer given by @sd with dynamic states. 1736 * Used by dump format v1 (current). 1737 * 1738 * Returns 0 on success. 1739 */ 1740 int 1741 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd) 1742 { 1743 ipfw_dyn_rule *p; 1744 ipfw_obj_dyntlv *dst, *last; 1745 ipfw_obj_ctlv *ctlv; 1746 int i; 1747 size_t sz; 1748 1749 if (V_ipfw_dyn_v == NULL) 1750 return (0); 1751 1752 IPFW_UH_RLOCK_ASSERT(chain); 1753 1754 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); 1755 if (ctlv == NULL) 1756 return (ENOMEM); 1757 sz = sizeof(ipfw_obj_dyntlv); 1758 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST; 1759 ctlv->objsize = sz; 1760 last = NULL; 1761 1762 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1763 IPFW_BUCK_LOCK(i); 1764 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1765 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, sz); 1766 if (dst == NULL) { 1767 IPFW_BUCK_UNLOCK(i); 1768 return (ENOMEM); 1769 } 1770 1771 export_dyn_rule(p, &dst->state); 1772 dst->head.length = sz; 1773 dst->head.type = IPFW_TLV_DYN_ENT; 1774 last = dst; 1775 } 1776 IPFW_BUCK_UNLOCK(i); 1777 } 1778 1779 if (last != NULL) /* mark last dynamic rule */ 1780 last->head.flags = IPFW_DF_LAST; 1781 1782 return (0); 1783 } 1784 1785 /* 1786 * Fill given buffer with dynamic states (legacy format). 1787 * IPFW_UH_RLOCK has to be held while calling. 1788 */ 1789 void 1790 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep) 1791 { 1792 ipfw_dyn_rule *p, *last = NULL; 1793 char *bp; 1794 int i; 1795 1796 if (V_ipfw_dyn_v == NULL) 1797 return; 1798 bp = *pbp; 1799 1800 IPFW_UH_RLOCK_ASSERT(chain); 1801 1802 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1803 IPFW_BUCK_LOCK(i); 1804 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1805 if (bp + sizeof *p <= ep) { 1806 ipfw_dyn_rule *dst = 1807 (ipfw_dyn_rule *)bp; 1808 1809 export_dyn_rule(p, dst); 1810 last = dst; 1811 bp += sizeof(ipfw_dyn_rule); 1812 } 1813 } 1814 IPFW_BUCK_UNLOCK(i); 1815 } 1816 1817 if (last != NULL) /* mark last dynamic rule */ 1818 bzero(&last->next, sizeof(last)); 1819 *pbp = bp; 1820 } 1821 /* end of file */ 1822