ip_fw_dynamic.c revision 324046
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: stable/11/sys/netpfil/ipfw/ip_fw_dynamic.c 324046 2017-09-27 01:46:14Z ae $"); 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 119struct 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 */ 127static VNET_DEFINE(struct ipfw_dyn_bucket *, ipfw_dyn_v); 128static VNET_DEFINE(u_int32_t, dyn_buckets_max); 129static VNET_DEFINE(u_int32_t, curr_dyn_buckets); 130static 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 136static 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 148static 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 */ 154static VNET_DEFINE(u_int32_t, dyn_ack_lifetime); 155static VNET_DEFINE(u_int32_t, dyn_syn_lifetime); 156static VNET_DEFINE(u_int32_t, dyn_fin_lifetime); 157static VNET_DEFINE(u_int32_t, dyn_rst_lifetime); 158static VNET_DEFINE(u_int32_t, dyn_udp_lifetime); 159static 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 176static VNET_DEFINE(u_int32_t, dyn_keepalive_interval); 177static VNET_DEFINE(u_int32_t, dyn_keepalive_period); 178static VNET_DEFINE(u_int32_t, dyn_keepalive); 179static 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 186static 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 */ 192static 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 198static int last_log; /* Log ratelimiting */ 199 200static void ipfw_dyn_tick(void *vnetx); 201static void check_dyn_rules(struct ip_fw_chain *, ipfw_range_tlv *, int, int); 202#ifdef SYSCTL_NODE 203 204static int sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS); 205static int sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS); 206 207SYSBEGIN(f2) 208 209SYSCTL_DECL(_net_inet_ip_fw); 210SYSCTL_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"); 213SYSCTL_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"); 216SYSCTL_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"); 219SYSCTL_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"); 222SYSCTL_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"); 225SYSCTL_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"); 228SYSCTL_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"); 231SYSCTL_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"); 234SYSCTL_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"); 237SYSCTL_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"); 240SYSCTL_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"); 243SYSCTL_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 247SYSEND 248 249#endif /* SYSCTL_NODE */ 250 251 252#ifdef INET6 253static __inline int 254hash_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 */ 270static __inline int 271hash_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 293static char *default_state_name = "default"; 294struct 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 */ 306static int 307dyn_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 321static void 322dyn_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 329static int 330dyn_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 363static struct named_object * 364dyn_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 371static int 372dyn_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 SRV_OBJECT(ch, obj->no.kidx) = obj; 422 obj->no.refcnt++; 423 *pkidx = obj->no.kidx; 424 IPFW_UH_WUNLOCK(ch); 425 DYN_DEBUG("\tcreated kidx %d", *pkidx); 426 return (0); 427} 428 429static void 430dyn_destroy(struct ip_fw_chain *ch, struct named_object *no) 431{ 432 struct dyn_state_obj *obj; 433 434 IPFW_UH_WLOCK_ASSERT(ch); 435 436 KASSERT(no->refcnt == 1, 437 ("Destroying object '%s' (type %u, idx %u) with refcnt %u", 438 no->name, no->etlv, no->kidx, no->refcnt)); 439 440 DYN_DEBUG("kidx %d", no->kidx); 441 obj = SRV_OBJECT(ch, no->kidx); 442 SRV_OBJECT(ch, no->kidx) = NULL; 443 ipfw_objhash_del(CHAIN_TO_SRV(ch), no); 444 ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), no->kidx); 445 446 free(obj, M_IPFW); 447} 448 449static struct opcode_obj_rewrite dyn_opcodes[] = { 450 { 451 O_KEEP_STATE, IPFW_TLV_STATE_NAME, 452 dyn_classify, dyn_update, 453 dyn_findbyname, dyn_findbykidx, 454 dyn_create, dyn_destroy 455 }, 456 { 457 O_CHECK_STATE, IPFW_TLV_STATE_NAME, 458 dyn_classify, dyn_update, 459 dyn_findbyname, dyn_findbykidx, 460 dyn_create, dyn_destroy 461 }, 462 { 463 O_PROBE_STATE, IPFW_TLV_STATE_NAME, 464 dyn_classify, dyn_update, 465 dyn_findbyname, dyn_findbykidx, 466 dyn_create, dyn_destroy 467 }, 468 { 469 O_LIMIT, IPFW_TLV_STATE_NAME, 470 dyn_classify, dyn_update, 471 dyn_findbyname, dyn_findbykidx, 472 dyn_create, dyn_destroy 473 }, 474}; 475/** 476 * Print customizable flow id description via log(9) facility. 477 */ 478static void 479print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags, 480 char *prefix, char *postfix) 481{ 482 struct in_addr da; 483#ifdef INET6 484 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; 485#else 486 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 487#endif 488 489#ifdef INET6 490 if (IS_IP6_FLOW_ID(id)) { 491 ip6_sprintf(src, &id->src_ip6); 492 ip6_sprintf(dst, &id->dst_ip6); 493 } else 494#endif 495 { 496 da.s_addr = htonl(id->src_ip); 497 inet_ntop(AF_INET, &da, src, sizeof(src)); 498 da.s_addr = htonl(id->dst_ip); 499 inet_ntop(AF_INET, &da, dst, sizeof(dst)); 500 } 501 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", 502 prefix, dyn_type, src, id->src_port, dst, 503 id->dst_port, DYN_COUNT, postfix); 504} 505 506#define print_dyn_rule(id, dtype, prefix, postfix) \ 507 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) 508 509#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 510#define TIME_LE(a,b) ((int)((a)-(b)) < 0) 511 512static void 513dyn_update_proto_state(ipfw_dyn_rule *q, const struct ipfw_flow_id *id, 514 const struct tcphdr *tcp, int dir) 515{ 516 uint32_t ack; 517 u_char flags; 518 519 if (id->proto == IPPROTO_TCP) { 520 flags = id->_flags & (TH_FIN | TH_SYN | TH_RST); 521#define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 522#define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 523#define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) 524#define ACK_FWD 0x10000 /* fwd ack seen */ 525#define ACK_REV 0x20000 /* rev ack seen */ 526 527 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8); 528 switch (q->state & TCP_FLAGS) { 529 case TH_SYN: /* opening */ 530 q->expire = time_uptime + V_dyn_syn_lifetime; 531 break; 532 533 case BOTH_SYN: /* move to established */ 534 case BOTH_SYN | TH_FIN: /* one side tries to close */ 535 case BOTH_SYN | (TH_FIN << 8): 536#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0) 537 if (tcp == NULL) 538 break; 539 540 ack = ntohl(tcp->th_ack); 541 if (dir == MATCH_FORWARD) { 542 if (q->ack_fwd == 0 || 543 _SEQ_GE(ack, q->ack_fwd)) { 544 q->ack_fwd = ack; 545 q->state |= ACK_FWD; 546 } 547 } else { 548 if (q->ack_rev == 0 || 549 _SEQ_GE(ack, q->ack_rev)) { 550 q->ack_rev = ack; 551 q->state |= ACK_REV; 552 } 553 } 554 if ((q->state & (ACK_FWD | ACK_REV)) == 555 (ACK_FWD | ACK_REV)) { 556 q->expire = time_uptime + V_dyn_ack_lifetime; 557 q->state &= ~(ACK_FWD | ACK_REV); 558 } 559 break; 560 561 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 562 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) 563 V_dyn_fin_lifetime = 564 V_dyn_keepalive_period - 1; 565 q->expire = time_uptime + V_dyn_fin_lifetime; 566 break; 567 568 default: 569#if 0 570 /* 571 * reset or some invalid combination, but can also 572 * occur if we use keep-state the wrong way. 573 */ 574 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0) 575 printf("invalid state: 0x%x\n", q->state); 576#endif 577 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period) 578 V_dyn_rst_lifetime = 579 V_dyn_keepalive_period - 1; 580 q->expire = time_uptime + V_dyn_rst_lifetime; 581 break; 582 } 583 } else if (id->proto == IPPROTO_UDP) { 584 q->expire = time_uptime + V_dyn_udp_lifetime; 585 } else { 586 /* other protocols */ 587 q->expire = time_uptime + V_dyn_short_lifetime; 588 } 589} 590 591/* 592 * Lookup a dynamic rule, locked version. 593 */ 594static ipfw_dyn_rule * 595lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction, 596 struct tcphdr *tcp, uint16_t kidx) 597{ 598 /* 599 * Stateful ipfw extensions. 600 * Lookup into dynamic session queue. 601 */ 602 ipfw_dyn_rule *prev, *q = NULL; 603 int dir; 604 605 IPFW_BUCK_ASSERT(i); 606 607 dir = MATCH_NONE; 608 for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) { 609 if (q->dyn_type == O_LIMIT_PARENT) 610 continue; 611 612 if (pkt->proto != q->id.proto) 613 continue; 614 615 if (kidx != 0 && kidx != q->kidx) 616 continue; 617 618 if (IS_IP6_FLOW_ID(pkt)) { 619 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) && 620 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) && 621 pkt->src_port == q->id.src_port && 622 pkt->dst_port == q->id.dst_port) { 623 dir = MATCH_FORWARD; 624 break; 625 } 626 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) && 627 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) && 628 pkt->src_port == q->id.dst_port && 629 pkt->dst_port == q->id.src_port) { 630 dir = MATCH_REVERSE; 631 break; 632 } 633 } else { 634 if (pkt->src_ip == q->id.src_ip && 635 pkt->dst_ip == q->id.dst_ip && 636 pkt->src_port == q->id.src_port && 637 pkt->dst_port == q->id.dst_port) { 638 dir = MATCH_FORWARD; 639 break; 640 } 641 if (pkt->src_ip == q->id.dst_ip && 642 pkt->dst_ip == q->id.src_ip && 643 pkt->src_port == q->id.dst_port && 644 pkt->dst_port == q->id.src_port) { 645 dir = MATCH_REVERSE; 646 break; 647 } 648 } 649 } 650 if (q == NULL) 651 goto done; /* q = NULL, not found */ 652 653 if (prev != NULL) { /* found and not in front */ 654 prev->next = q->next; 655 q->next = V_ipfw_dyn_v[i].head; 656 V_ipfw_dyn_v[i].head = q; 657 } 658 659 /* update state according to flags */ 660 dyn_update_proto_state(q, pkt, tcp, dir); 661done: 662 if (match_direction != NULL) 663 *match_direction = dir; 664 return (q); 665} 666 667ipfw_dyn_rule * 668ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 669 struct tcphdr *tcp, uint16_t kidx) 670{ 671 ipfw_dyn_rule *q; 672 int i; 673 674 i = hash_packet(pkt, V_curr_dyn_buckets); 675 676 IPFW_BUCK_LOCK(i); 677 q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp, kidx); 678 if (q == NULL) 679 IPFW_BUCK_UNLOCK(i); 680 /* NB: return table locked when q is not NULL */ 681 return q; 682} 683 684/* 685 * Unlock bucket mtx 686 * @p - pointer to dynamic rule 687 */ 688void 689ipfw_dyn_unlock(ipfw_dyn_rule *q) 690{ 691 692 IPFW_BUCK_UNLOCK(q->bucket); 693} 694 695static int 696resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets) 697{ 698 int i, k, nbuckets_old; 699 ipfw_dyn_rule *q; 700 struct ipfw_dyn_bucket *dyn_v, *dyn_v_old; 701 702 /* Check if given number is power of 2 and less than 64k */ 703 if ((nbuckets > 65536) || (!powerof2(nbuckets))) 704 return 1; 705 706 CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__, 707 V_curr_dyn_buckets, nbuckets); 708 709 /* Allocate and initialize new hash */ 710 dyn_v = malloc(nbuckets * sizeof(*dyn_v), M_IPFW, 711 M_WAITOK | M_ZERO); 712 713 for (i = 0 ; i < nbuckets; i++) 714 IPFW_BUCK_LOCK_INIT(&dyn_v[i]); 715 716 /* 717 * Call upper half lock, as get_map() do to ease 718 * read-only access to dynamic rules hash from sysctl 719 */ 720 IPFW_UH_WLOCK(chain); 721 722 /* 723 * Acquire chain write lock to permit hash access 724 * for main traffic path without additional locks 725 */ 726 IPFW_WLOCK(chain); 727 728 /* Save old values */ 729 nbuckets_old = V_curr_dyn_buckets; 730 dyn_v_old = V_ipfw_dyn_v; 731 732 /* Skip relinking if array is not set up */ 733 if (V_ipfw_dyn_v == NULL) 734 V_curr_dyn_buckets = 0; 735 736 /* Re-link all dynamic states */ 737 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 738 while (V_ipfw_dyn_v[i].head != NULL) { 739 /* Remove from current chain */ 740 q = V_ipfw_dyn_v[i].head; 741 V_ipfw_dyn_v[i].head = q->next; 742 743 /* Get new hash value */ 744 k = hash_packet(&q->id, nbuckets); 745 q->bucket = k; 746 /* Add to the new head */ 747 q->next = dyn_v[k].head; 748 dyn_v[k].head = q; 749 } 750 } 751 752 /* Update current pointers/buckets values */ 753 V_curr_dyn_buckets = nbuckets; 754 V_ipfw_dyn_v = dyn_v; 755 756 IPFW_WUNLOCK(chain); 757 758 IPFW_UH_WUNLOCK(chain); 759 760 /* Start periodic callout on initial creation */ 761 if (dyn_v_old == NULL) { 762 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0); 763 return (0); 764 } 765 766 /* Destroy all mutexes */ 767 for (i = 0 ; i < nbuckets_old ; i++) 768 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]); 769 770 /* Free old hash */ 771 free(dyn_v_old, M_IPFW); 772 773 return 0; 774} 775 776/** 777 * Install state of type 'type' for a dynamic session. 778 * The hash table contains two type of rules: 779 * - regular rules (O_KEEP_STATE) 780 * - rules for sessions with limited number of sess per user 781 * (O_LIMIT). When they are created, the parent is 782 * increased by 1, and decreased on delete. In this case, 783 * the third parameter is the parent rule and not the chain. 784 * - "parent" rules for the above (O_LIMIT_PARENT). 785 */ 786static ipfw_dyn_rule * 787add_dyn_rule(struct ipfw_flow_id *id, int i, uint8_t dyn_type, 788 struct ip_fw *rule, uint16_t kidx) 789{ 790 ipfw_dyn_rule *r; 791 792 IPFW_BUCK_ASSERT(i); 793 794 r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO); 795 if (r == NULL) { 796 if (last_log != time_uptime) { 797 last_log = time_uptime; 798 log(LOG_DEBUG, 799 "ipfw: Cannot allocate dynamic state, " 800 "consider increasing net.inet.ip.fw.dyn_max\n"); 801 } 802 return NULL; 803 } 804 ipfw_dyn_count++; 805 806 /* 807 * refcount on parent is already incremented, so 808 * it is safe to use parent unlocked. 809 */ 810 if (dyn_type == O_LIMIT) { 811 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 812 if ( parent->dyn_type != O_LIMIT_PARENT) 813 panic("invalid parent"); 814 r->parent = parent; 815 rule = parent->rule; 816 } 817 818 r->id = *id; 819 r->expire = time_uptime + V_dyn_syn_lifetime; 820 r->rule = rule; 821 r->dyn_type = dyn_type; 822 IPFW_ZERO_DYN_COUNTER(r); 823 r->count = 0; 824 r->kidx = kidx; 825 r->bucket = i; 826 r->next = V_ipfw_dyn_v[i].head; 827 V_ipfw_dyn_v[i].head = r; 828 DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");) 829 return r; 830} 831 832/** 833 * lookup dynamic parent rule using pkt and rule as search keys. 834 * If the lookup fails, then install one. 835 */ 836static ipfw_dyn_rule * 837lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule, 838 uint16_t kidx) 839{ 840 ipfw_dyn_rule *q; 841 int i, is_v6; 842 843 is_v6 = IS_IP6_FLOW_ID(pkt); 844 i = hash_packet( pkt, V_curr_dyn_buckets ); 845 *pindex = i; 846 IPFW_BUCK_LOCK(i); 847 for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next) 848 if (q->dyn_type == O_LIMIT_PARENT && 849 kidx == q->kidx && 850 rule == q->rule && 851 pkt->proto == q->id.proto && 852 pkt->src_port == q->id.src_port && 853 pkt->dst_port == q->id.dst_port && 854 ( 855 (is_v6 && 856 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6), 857 &(q->id.src_ip6)) && 858 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6), 859 &(q->id.dst_ip6))) || 860 (!is_v6 && 861 pkt->src_ip == q->id.src_ip && 862 pkt->dst_ip == q->id.dst_ip) 863 ) 864 ) { 865 q->expire = time_uptime + V_dyn_short_lifetime; 866 DEB(print_dyn_rule(pkt, q->dyn_type, 867 "lookup_dyn_parent found", "");) 868 return q; 869 } 870 871 /* Add virtual limiting rule */ 872 return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule, kidx); 873} 874 875/** 876 * Install dynamic state for rule type cmd->o.opcode 877 * 878 * Returns 1 (failure) if state is not installed because of errors or because 879 * session limitations are enforced. 880 */ 881int 882ipfw_install_state(struct ip_fw_chain *chain, struct ip_fw *rule, 883 ipfw_insn_limit *cmd, struct ip_fw_args *args, uint32_t tablearg) 884{ 885 ipfw_dyn_rule *q; 886 int i; 887 888 DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", 889 (cmd->o.arg1 == 0 ? "": DYN_STATE_OBJ(chain, &cmd->o)->name));) 890 891 i = hash_packet(&args->f_id, V_curr_dyn_buckets); 892 893 IPFW_BUCK_LOCK(i); 894 895 q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL, cmd->o.arg1); 896 if (q != NULL) { /* should never occur */ 897 DEB( 898 if (last_log != time_uptime) { 899 last_log = time_uptime; 900 printf("ipfw: %s: entry already present, done\n", 901 __func__); 902 }) 903 IPFW_BUCK_UNLOCK(i); 904 return (0); 905 } 906 907 /* 908 * State limiting is done via uma(9) zone limiting. 909 * Save pointer to newly-installed rule and reject 910 * packet if add_dyn_rule() returned NULL. 911 * Note q is currently set to NULL. 912 */ 913 914 switch (cmd->o.opcode) { 915 case O_KEEP_STATE: /* bidir rule */ 916 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule, 917 cmd->o.arg1); 918 break; 919 920 case O_LIMIT: { /* limit number of sessions */ 921 struct ipfw_flow_id id; 922 ipfw_dyn_rule *parent; 923 uint32_t conn_limit; 924 uint16_t limit_mask = cmd->limit_mask; 925 int pindex; 926 927 conn_limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit); 928 929 DEB( 930 if (cmd->conn_limit == IP_FW_TARG) 931 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u " 932 "(tablearg)\n", __func__, conn_limit); 933 else 934 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n", 935 __func__, conn_limit); 936 ) 937 938 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0; 939 id.proto = args->f_id.proto; 940 id.addr_type = args->f_id.addr_type; 941 id.fib = M_GETFIB(args->m); 942 943 if (IS_IP6_FLOW_ID (&(args->f_id))) { 944 bzero(&id.src_ip6, sizeof(id.src_ip6)); 945 bzero(&id.dst_ip6, sizeof(id.dst_ip6)); 946 947 if (limit_mask & DYN_SRC_ADDR) 948 id.src_ip6 = args->f_id.src_ip6; 949 if (limit_mask & DYN_DST_ADDR) 950 id.dst_ip6 = args->f_id.dst_ip6; 951 } else { 952 if (limit_mask & DYN_SRC_ADDR) 953 id.src_ip = args->f_id.src_ip; 954 if (limit_mask & DYN_DST_ADDR) 955 id.dst_ip = args->f_id.dst_ip; 956 } 957 if (limit_mask & DYN_SRC_PORT) 958 id.src_port = args->f_id.src_port; 959 if (limit_mask & DYN_DST_PORT) 960 id.dst_port = args->f_id.dst_port; 961 962 /* 963 * We have to release lock for previous bucket to 964 * avoid possible deadlock 965 */ 966 IPFW_BUCK_UNLOCK(i); 967 968 parent = lookup_dyn_parent(&id, &pindex, rule, cmd->o.arg1); 969 if (parent == NULL) { 970 printf("ipfw: %s: add parent failed\n", __func__); 971 IPFW_BUCK_UNLOCK(pindex); 972 return (1); 973 } 974 975 if (parent->count >= conn_limit) { 976 if (V_fw_verbose && last_log != time_uptime) { 977 last_log = time_uptime; 978 char sbuf[24]; 979 last_log = time_uptime; 980 snprintf(sbuf, sizeof(sbuf), 981 "%d drop session", 982 parent->rule->rulenum); 983 print_dyn_rule_flags(&args->f_id, 984 cmd->o.opcode, 985 LOG_SECURITY | LOG_DEBUG, 986 sbuf, "too many entries"); 987 } 988 IPFW_BUCK_UNLOCK(pindex); 989 return (1); 990 } 991 /* Increment counter on parent */ 992 parent->count++; 993 IPFW_BUCK_UNLOCK(pindex); 994 995 IPFW_BUCK_LOCK(i); 996 q = add_dyn_rule(&args->f_id, i, O_LIMIT, 997 (struct ip_fw *)parent, cmd->o.arg1); 998 if (q == NULL) { 999 /* Decrement index and notify caller */ 1000 IPFW_BUCK_UNLOCK(i); 1001 IPFW_BUCK_LOCK(pindex); 1002 parent->count--; 1003 IPFW_BUCK_UNLOCK(pindex); 1004 return (1); 1005 } 1006 break; 1007 } 1008 default: 1009 printf("ipfw: %s: unknown dynamic rule type %u\n", 1010 __func__, cmd->o.opcode); 1011 } 1012 1013 if (q == NULL) { 1014 IPFW_BUCK_UNLOCK(i); 1015 return (1); /* Notify caller about failure */ 1016 } 1017 1018 dyn_update_proto_state(q, &args->f_id, NULL, MATCH_FORWARD); 1019 IPFW_BUCK_UNLOCK(i); 1020 return (0); 1021} 1022 1023/* 1024 * Generate a TCP packet, containing either a RST or a keepalive. 1025 * When flags & TH_RST, we are sending a RST packet, because of a 1026 * "reset" action matched the packet. 1027 * Otherwise we are sending a keepalive, and flags & TH_ 1028 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required 1029 * so that MAC can label the reply appropriately. 1030 */ 1031struct mbuf * 1032ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq, 1033 u_int32_t ack, int flags) 1034{ 1035 struct mbuf *m = NULL; /* stupid compiler */ 1036 int len, dir; 1037 struct ip *h = NULL; /* stupid compiler */ 1038#ifdef INET6 1039 struct ip6_hdr *h6 = NULL; 1040#endif 1041 struct tcphdr *th = NULL; 1042 1043 MGETHDR(m, M_NOWAIT, MT_DATA); 1044 if (m == NULL) 1045 return (NULL); 1046 1047 M_SETFIB(m, id->fib); 1048#ifdef MAC 1049 if (replyto != NULL) 1050 mac_netinet_firewall_reply(replyto, m); 1051 else 1052 mac_netinet_firewall_send(m); 1053#else 1054 (void)replyto; /* don't warn about unused arg */ 1055#endif 1056 1057 switch (id->addr_type) { 1058 case 4: 1059 len = sizeof(struct ip) + sizeof(struct tcphdr); 1060 break; 1061#ifdef INET6 1062 case 6: 1063 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1064 break; 1065#endif 1066 default: 1067 /* XXX: log me?!? */ 1068 FREE_PKT(m); 1069 return (NULL); 1070 } 1071 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN); 1072 1073 m->m_data += max_linkhdr; 1074 m->m_flags |= M_SKIP_FIREWALL; 1075 m->m_pkthdr.len = m->m_len = len; 1076 m->m_pkthdr.rcvif = NULL; 1077 bzero(m->m_data, len); 1078 1079 switch (id->addr_type) { 1080 case 4: 1081 h = mtod(m, struct ip *); 1082 1083 /* prepare for checksum */ 1084 h->ip_p = IPPROTO_TCP; 1085 h->ip_len = htons(sizeof(struct tcphdr)); 1086 if (dir) { 1087 h->ip_src.s_addr = htonl(id->src_ip); 1088 h->ip_dst.s_addr = htonl(id->dst_ip); 1089 } else { 1090 h->ip_src.s_addr = htonl(id->dst_ip); 1091 h->ip_dst.s_addr = htonl(id->src_ip); 1092 } 1093 1094 th = (struct tcphdr *)(h + 1); 1095 break; 1096#ifdef INET6 1097 case 6: 1098 h6 = mtod(m, struct ip6_hdr *); 1099 1100 /* prepare for checksum */ 1101 h6->ip6_nxt = IPPROTO_TCP; 1102 h6->ip6_plen = htons(sizeof(struct tcphdr)); 1103 if (dir) { 1104 h6->ip6_src = id->src_ip6; 1105 h6->ip6_dst = id->dst_ip6; 1106 } else { 1107 h6->ip6_src = id->dst_ip6; 1108 h6->ip6_dst = id->src_ip6; 1109 } 1110 1111 th = (struct tcphdr *)(h6 + 1); 1112 break; 1113#endif 1114 } 1115 1116 if (dir) { 1117 th->th_sport = htons(id->src_port); 1118 th->th_dport = htons(id->dst_port); 1119 } else { 1120 th->th_sport = htons(id->dst_port); 1121 th->th_dport = htons(id->src_port); 1122 } 1123 th->th_off = sizeof(struct tcphdr) >> 2; 1124 1125 if (flags & TH_RST) { 1126 if (flags & TH_ACK) { 1127 th->th_seq = htonl(ack); 1128 th->th_flags = TH_RST; 1129 } else { 1130 if (flags & TH_SYN) 1131 seq++; 1132 th->th_ack = htonl(seq); 1133 th->th_flags = TH_RST | TH_ACK; 1134 } 1135 } else { 1136 /* 1137 * Keepalive - use caller provided sequence numbers 1138 */ 1139 th->th_seq = htonl(seq); 1140 th->th_ack = htonl(ack); 1141 th->th_flags = TH_ACK; 1142 } 1143 1144 switch (id->addr_type) { 1145 case 4: 1146 th->th_sum = in_cksum(m, len); 1147 1148 /* finish the ip header */ 1149 h->ip_v = 4; 1150 h->ip_hl = sizeof(*h) >> 2; 1151 h->ip_tos = IPTOS_LOWDELAY; 1152 h->ip_off = htons(0); 1153 h->ip_len = htons(len); 1154 h->ip_ttl = V_ip_defttl; 1155 h->ip_sum = 0; 1156 break; 1157#ifdef INET6 1158 case 6: 1159 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6), 1160 sizeof(struct tcphdr)); 1161 1162 /* finish the ip6 header */ 1163 h6->ip6_vfc |= IPV6_VERSION; 1164 h6->ip6_hlim = IPV6_DEFHLIM; 1165 break; 1166#endif 1167 } 1168 1169 return (m); 1170} 1171 1172/* 1173 * Queue keepalive packets for given dynamic rule 1174 */ 1175static struct mbuf ** 1176ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q) 1177{ 1178 struct mbuf *m_rev, *m_fwd; 1179 1180 m_rev = (q->state & ACK_REV) ? NULL : 1181 ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN); 1182 m_fwd = (q->state & ACK_FWD) ? NULL : 1183 ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0); 1184 1185 if (m_rev != NULL) { 1186 *mtailp = m_rev; 1187 mtailp = &(*mtailp)->m_nextpkt; 1188 } 1189 if (m_fwd != NULL) { 1190 *mtailp = m_fwd; 1191 mtailp = &(*mtailp)->m_nextpkt; 1192 } 1193 1194 return (mtailp); 1195} 1196 1197/* 1198 * This procedure is used to perform various maintenance 1199 * on dynamic hash list. Currently it is called every second. 1200 */ 1201static void 1202ipfw_dyn_tick(void * vnetx) 1203{ 1204 struct ip_fw_chain *chain; 1205 int check_ka = 0; 1206#ifdef VIMAGE 1207 struct vnet *vp = vnetx; 1208#endif 1209 1210 CURVNET_SET(vp); 1211 1212 chain = &V_layer3_chain; 1213 1214 /* Run keepalive checks every keepalive_period iff ka is enabled */ 1215 if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) && 1216 (V_dyn_keepalive != 0)) { 1217 V_dyn_keepalive_last = time_uptime; 1218 check_ka = 1; 1219 } 1220 1221 check_dyn_rules(chain, NULL, check_ka, 1); 1222 1223 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0); 1224 1225 CURVNET_RESTORE(); 1226} 1227 1228 1229/* 1230 * Walk through all dynamic states doing generic maintenance: 1231 * 1) free expired states 1232 * 2) free all states based on deleted rule / set 1233 * 3) send keepalives for states if needed 1234 * 1235 * @chain - pointer to current ipfw rules chain 1236 * @rule - delete all states originated by given rule if != NULL 1237 * @set - delete all states originated by any rule in set @set if != RESVD_SET 1238 * @check_ka - perform checking/sending keepalives 1239 * @timer - indicate call from timer routine. 1240 * 1241 * Timer routine must call this function unlocked to permit 1242 * sending keepalives/resizing table. 1243 * 1244 * Others has to call function with IPFW_UH_WLOCK held. 1245 * Additionally, function assume that dynamic rule/set is 1246 * ALREADY deleted so no new states can be generated by 1247 * 'deleted' rules. 1248 * 1249 * Write lock is needed to ensure that unused parent rules 1250 * are not freed by other instance (see stage 2, 3) 1251 */ 1252static void 1253check_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt, 1254 int check_ka, int timer) 1255{ 1256 struct mbuf *m0, *m, *mnext, **mtailp; 1257 struct ip *h; 1258 int i, dyn_count, new_buckets = 0, max_buckets; 1259 int expired = 0, expired_limits = 0, parents = 0, total = 0; 1260 ipfw_dyn_rule *q, *q_prev, *q_next; 1261 ipfw_dyn_rule *exp_head, **exptailp; 1262 ipfw_dyn_rule *exp_lhead, **expltailp; 1263 1264 KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated", 1265 __func__)); 1266 1267 /* Avoid possible LOR */ 1268 KASSERT(!check_ka || timer, ("%s: keepalive check with lock held", 1269 __func__)); 1270 1271 /* 1272 * Do not perform any checks if we currently have no dynamic states 1273 */ 1274 if (DYN_COUNT == 0) 1275 return; 1276 1277 /* Expired states */ 1278 exp_head = NULL; 1279 exptailp = &exp_head; 1280 1281 /* Expired limit states */ 1282 exp_lhead = NULL; 1283 expltailp = &exp_lhead; 1284 1285 /* 1286 * We make a chain of packets to go out here -- not deferring 1287 * until after we drop the IPFW dynamic rule lock would result 1288 * in a lock order reversal with the normal packet input -> ipfw 1289 * call stack. 1290 */ 1291 m0 = NULL; 1292 mtailp = &m0; 1293 1294 /* Protect from hash resizing */ 1295 if (timer != 0) 1296 IPFW_UH_WLOCK(chain); 1297 else 1298 IPFW_UH_WLOCK_ASSERT(chain); 1299 1300#define NEXT_RULE() { q_prev = q; q = q->next ; continue; } 1301 1302 /* Stage 1: perform requested deletion */ 1303 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1304 IPFW_BUCK_LOCK(i); 1305 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) { 1306 /* account every rule */ 1307 total++; 1308 1309 /* Skip parent rules at all */ 1310 if (q->dyn_type == O_LIMIT_PARENT) { 1311 parents++; 1312 NEXT_RULE(); 1313 } 1314 1315 /* 1316 * Remove rules which are: 1317 * 1) expired 1318 * 2) matches deletion range 1319 */ 1320 if ((TIME_LEQ(q->expire, time_uptime)) || 1321 (rt != NULL && ipfw_match_range(q->rule, rt))) { 1322 if (TIME_LE(time_uptime, q->expire) && 1323 q->dyn_type == O_KEEP_STATE && 1324 V_dyn_keep_states != 0) { 1325 /* 1326 * Do not delete state if 1327 * it is not expired and 1328 * dyn_keep_states is ON. 1329 * However we need to re-link it 1330 * to any other stable rule 1331 */ 1332 q->rule = chain->default_rule; 1333 NEXT_RULE(); 1334 } 1335 1336 /* Unlink q from current list */ 1337 q_next = q->next; 1338 if (q == V_ipfw_dyn_v[i].head) 1339 V_ipfw_dyn_v[i].head = q_next; 1340 else 1341 q_prev->next = q_next; 1342 1343 q->next = NULL; 1344 1345 /* queue q to expire list */ 1346 if (q->dyn_type != O_LIMIT) { 1347 *exptailp = q; 1348 exptailp = &(*exptailp)->next; 1349 DEB(print_dyn_rule(&q->id, q->dyn_type, 1350 "unlink entry", "left"); 1351 ) 1352 } else { 1353 /* Separate list for limit rules */ 1354 *expltailp = q; 1355 expltailp = &(*expltailp)->next; 1356 expired_limits++; 1357 DEB(print_dyn_rule(&q->id, q->dyn_type, 1358 "unlink limit entry", "left"); 1359 ) 1360 } 1361 1362 q = q_next; 1363 expired++; 1364 continue; 1365 } 1366 1367 /* 1368 * Check if we need to send keepalive: 1369 * we need to ensure if is time to do KA, 1370 * this is established TCP session, and 1371 * expire time is within keepalive interval 1372 */ 1373 if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) && 1374 ((q->state & BOTH_SYN) == BOTH_SYN) && 1375 (TIME_LEQ(q->expire, time_uptime + 1376 V_dyn_keepalive_interval))) 1377 mtailp = ipfw_dyn_send_ka(mtailp, q); 1378 1379 NEXT_RULE(); 1380 } 1381 IPFW_BUCK_UNLOCK(i); 1382 } 1383 1384 /* Stage 2: decrement counters from O_LIMIT parents */ 1385 if (expired_limits != 0) { 1386 /* 1387 * XXX: Note that deleting set with more than one 1388 * heavily-used LIMIT rules can result in overwhelming 1389 * locking due to lack of per-hash value sorting 1390 * 1391 * We should probably think about: 1392 * 1) pre-allocating hash of size, say, 1393 * MAX(16, V_curr_dyn_buckets / 1024) 1394 * 2) checking if expired_limits is large enough 1395 * 3) If yes, init hash (or its part), re-link 1396 * current list and start decrementing procedure in 1397 * each bucket separately 1398 */ 1399 1400 /* 1401 * Small optimization: do not unlock bucket until 1402 * we see the next item resides in different bucket 1403 */ 1404 if (exp_lhead != NULL) { 1405 i = exp_lhead->parent->bucket; 1406 IPFW_BUCK_LOCK(i); 1407 } 1408 for (q = exp_lhead; q != NULL; q = q->next) { 1409 if (i != q->parent->bucket) { 1410 IPFW_BUCK_UNLOCK(i); 1411 i = q->parent->bucket; 1412 IPFW_BUCK_LOCK(i); 1413 } 1414 1415 /* Decrease parent refcount */ 1416 q->parent->count--; 1417 } 1418 if (exp_lhead != NULL) 1419 IPFW_BUCK_UNLOCK(i); 1420 } 1421 1422 /* 1423 * We protectet ourselves from unused parent deletion 1424 * (from the timer function) by holding UH write lock. 1425 */ 1426 1427 /* Stage 3: remove unused parent rules */ 1428 if ((parents != 0) && (expired != 0)) { 1429 for (i = 0 ; i < V_curr_dyn_buckets ; i++) { 1430 IPFW_BUCK_LOCK(i); 1431 for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) { 1432 if (q->dyn_type != O_LIMIT_PARENT) 1433 NEXT_RULE(); 1434 1435 if (q->count != 0) 1436 NEXT_RULE(); 1437 1438 /* Parent rule without consumers */ 1439 1440 /* Unlink q from current list */ 1441 q_next = q->next; 1442 if (q == V_ipfw_dyn_v[i].head) 1443 V_ipfw_dyn_v[i].head = q_next; 1444 else 1445 q_prev->next = q_next; 1446 1447 q->next = NULL; 1448 1449 /* Add to expired list */ 1450 *exptailp = q; 1451 exptailp = &(*exptailp)->next; 1452 1453 DEB(print_dyn_rule(&q->id, q->dyn_type, 1454 "unlink parent entry", "left"); 1455 ) 1456 1457 expired++; 1458 1459 q = q_next; 1460 } 1461 IPFW_BUCK_UNLOCK(i); 1462 } 1463 } 1464 1465#undef NEXT_RULE 1466 1467 if (timer != 0) { 1468 /* 1469 * Check if we need to resize hash: 1470 * if current number of states exceeds number of buckes in hash, 1471 * grow hash size to the minimum power of 2 which is bigger than 1472 * current states count. Limit hash size by 64k. 1473 */ 1474 max_buckets = (V_dyn_buckets_max > 65536) ? 1475 65536 : V_dyn_buckets_max; 1476 1477 dyn_count = DYN_COUNT; 1478 1479 if ((dyn_count > V_curr_dyn_buckets * 2) && 1480 (dyn_count < max_buckets)) { 1481 new_buckets = V_curr_dyn_buckets; 1482 while (new_buckets < dyn_count) { 1483 new_buckets *= 2; 1484 1485 if (new_buckets >= max_buckets) 1486 break; 1487 } 1488 } 1489 1490 IPFW_UH_WUNLOCK(chain); 1491 } 1492 1493 /* Finally delete old states ad limits if any */ 1494 for (q = exp_head; q != NULL; q = q_next) { 1495 q_next = q->next; 1496 uma_zfree(V_ipfw_dyn_rule_zone, q); 1497 ipfw_dyn_count--; 1498 } 1499 1500 for (q = exp_lhead; q != NULL; q = q_next) { 1501 q_next = q->next; 1502 uma_zfree(V_ipfw_dyn_rule_zone, q); 1503 ipfw_dyn_count--; 1504 } 1505 1506 /* 1507 * The rest code MUST be called from timer routine only 1508 * without holding any locks 1509 */ 1510 if (timer == 0) 1511 return; 1512 1513 /* Send keepalive packets if any */ 1514 for (m = m0; m != NULL; m = mnext) { 1515 mnext = m->m_nextpkt; 1516 m->m_nextpkt = NULL; 1517 h = mtod(m, struct ip *); 1518 if (h->ip_v == 4) 1519 ip_output(m, NULL, NULL, 0, NULL, NULL); 1520#ifdef INET6 1521 else 1522 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 1523#endif 1524 } 1525 1526 /* Run table resize without holding any locks */ 1527 if (new_buckets != 0) 1528 resize_dynamic_table(chain, new_buckets); 1529} 1530 1531/* 1532 * Deletes all dynamic rules originated by given rule or all rules in 1533 * given set. Specify RESVD_SET to indicate set should not be used. 1534 * @chain - pointer to current ipfw rules chain 1535 * @rr - delete all states originated by rules in matched range. 1536 * 1537 * Function has to be called with IPFW_UH_WLOCK held. 1538 * Additionally, function assume that dynamic rule/set is 1539 * ALREADY deleted so no new states can be generated by 1540 * 'deleted' rules. 1541 */ 1542void 1543ipfw_expire_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt) 1544{ 1545 1546 check_dyn_rules(chain, rt, 0, 0); 1547} 1548 1549/* 1550 * Check if rule contains at least one dynamic opcode. 1551 * 1552 * Returns 1 if such opcode is found, 0 otherwise. 1553 */ 1554int 1555ipfw_is_dyn_rule(struct ip_fw *rule) 1556{ 1557 int cmdlen, l; 1558 ipfw_insn *cmd; 1559 1560 l = rule->cmd_len; 1561 cmd = rule->cmd; 1562 cmdlen = 0; 1563 for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { 1564 cmdlen = F_LEN(cmd); 1565 1566 switch (cmd->opcode) { 1567 case O_LIMIT: 1568 case O_KEEP_STATE: 1569 case O_PROBE_STATE: 1570 case O_CHECK_STATE: 1571 return (1); 1572 } 1573 } 1574 1575 return (0); 1576} 1577 1578void 1579ipfw_dyn_init(struct ip_fw_chain *chain) 1580{ 1581 1582 V_ipfw_dyn_v = NULL; 1583 V_dyn_buckets_max = 256; /* must be power of 2 */ 1584 V_curr_dyn_buckets = 256; /* must be power of 2 */ 1585 1586 V_dyn_ack_lifetime = 300; 1587 V_dyn_syn_lifetime = 20; 1588 V_dyn_fin_lifetime = 1; 1589 V_dyn_rst_lifetime = 1; 1590 V_dyn_udp_lifetime = 10; 1591 V_dyn_short_lifetime = 5; 1592 1593 V_dyn_keepalive_interval = 20; 1594 V_dyn_keepalive_period = 5; 1595 V_dyn_keepalive = 1; /* do send keepalives */ 1596 V_dyn_keepalive_last = time_uptime; 1597 1598 V_dyn_max = 16384; /* max # of dynamic rules */ 1599 1600 V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule", 1601 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL, 1602 UMA_ALIGN_PTR, 0); 1603 1604 /* Enforce limit on dynamic rules */ 1605 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max); 1606 1607 callout_init(&V_ipfw_timeout, 1); 1608 1609 /* 1610 * This can potentially be done on first dynamic rule 1611 * being added to chain. 1612 */ 1613 resize_dynamic_table(chain, V_curr_dyn_buckets); 1614 IPFW_ADD_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); 1615} 1616 1617void 1618ipfw_dyn_uninit(int pass) 1619{ 1620 int i; 1621 1622 if (pass == 0) { 1623 callout_drain(&V_ipfw_timeout); 1624 return; 1625 } 1626 IPFW_DEL_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); 1627 1628 if (V_ipfw_dyn_v != NULL) { 1629 /* 1630 * Skip deleting all dynamic states - 1631 * uma_zdestroy() does this more efficiently; 1632 */ 1633 1634 /* Destroy all mutexes */ 1635 for (i = 0 ; i < V_curr_dyn_buckets ; i++) 1636 IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]); 1637 free(V_ipfw_dyn_v, M_IPFW); 1638 V_ipfw_dyn_v = NULL; 1639 } 1640 1641 uma_zdestroy(V_ipfw_dyn_rule_zone); 1642} 1643 1644#ifdef SYSCTL_NODE 1645/* 1646 * Get/set maximum number of dynamic states in given VNET instance. 1647 */ 1648static int 1649sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS) 1650{ 1651 int error; 1652 unsigned int nstates; 1653 1654 nstates = V_dyn_max; 1655 1656 error = sysctl_handle_int(oidp, &nstates, 0, req); 1657 /* Read operation or some error */ 1658 if ((error != 0) || (req->newptr == NULL)) 1659 return (error); 1660 1661 V_dyn_max = nstates; 1662 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max); 1663 1664 return (0); 1665} 1666 1667/* 1668 * Get current number of dynamic states in given VNET instance. 1669 */ 1670static int 1671sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS) 1672{ 1673 int error; 1674 unsigned int nstates; 1675 1676 nstates = DYN_COUNT; 1677 1678 error = sysctl_handle_int(oidp, &nstates, 0, req); 1679 1680 return (error); 1681} 1682#endif 1683 1684/* 1685 * Returns size of dynamic states in legacy format 1686 */ 1687int 1688ipfw_dyn_len(void) 1689{ 1690 1691 return (V_ipfw_dyn_v == NULL) ? 0 : 1692 (DYN_COUNT * sizeof(ipfw_dyn_rule)); 1693} 1694 1695/* 1696 * Returns number of dynamic states. 1697 * Used by dump format v1 (current). 1698 */ 1699int 1700ipfw_dyn_get_count(void) 1701{ 1702 1703 return (V_ipfw_dyn_v == NULL) ? 0 : DYN_COUNT; 1704} 1705 1706static void 1707export_dyn_rule(ipfw_dyn_rule *src, ipfw_dyn_rule *dst) 1708{ 1709 uint16_t rulenum; 1710 1711 rulenum = (uint16_t)src->rule->rulenum; 1712 memcpy(dst, src, sizeof(*src)); 1713 memcpy(&dst->rule, &rulenum, sizeof(rulenum)); 1714 /* 1715 * store set number into high word of 1716 * dst->rule pointer. 1717 */ 1718 memcpy((char *)&dst->rule + sizeof(rulenum), &src->rule->set, 1719 sizeof(src->rule->set)); 1720 /* 1721 * store a non-null value in "next". 1722 * The userland code will interpret a 1723 * NULL here as a marker 1724 * for the last dynamic rule. 1725 */ 1726 memcpy(&dst->next, &dst, sizeof(dst)); 1727 dst->expire = TIME_LEQ(dst->expire, time_uptime) ? 0: 1728 dst->expire - time_uptime; 1729} 1730 1731/* 1732 * Fills int buffer given by @sd with dynamic states. 1733 * Used by dump format v1 (current). 1734 * 1735 * Returns 0 on success. 1736 */ 1737int 1738ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd) 1739{ 1740 ipfw_dyn_rule *p; 1741 ipfw_obj_dyntlv *dst, *last; 1742 ipfw_obj_ctlv *ctlv; 1743 int i; 1744 size_t sz; 1745 1746 if (V_ipfw_dyn_v == NULL) 1747 return (0); 1748 1749 IPFW_UH_RLOCK_ASSERT(chain); 1750 1751 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); 1752 if (ctlv == NULL) 1753 return (ENOMEM); 1754 sz = sizeof(ipfw_obj_dyntlv); 1755 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST; 1756 ctlv->objsize = sz; 1757 last = NULL; 1758 1759 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1760 IPFW_BUCK_LOCK(i); 1761 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1762 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, sz); 1763 if (dst == NULL) { 1764 IPFW_BUCK_UNLOCK(i); 1765 return (ENOMEM); 1766 } 1767 1768 export_dyn_rule(p, &dst->state); 1769 dst->head.length = sz; 1770 dst->head.type = IPFW_TLV_DYN_ENT; 1771 last = dst; 1772 } 1773 IPFW_BUCK_UNLOCK(i); 1774 } 1775 1776 if (last != NULL) /* mark last dynamic rule */ 1777 last->head.flags = IPFW_DF_LAST; 1778 1779 return (0); 1780} 1781 1782/* 1783 * Fill given buffer with dynamic states (legacy format). 1784 * IPFW_UH_RLOCK has to be held while calling. 1785 */ 1786void 1787ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep) 1788{ 1789 ipfw_dyn_rule *p, *last = NULL; 1790 char *bp; 1791 int i; 1792 1793 if (V_ipfw_dyn_v == NULL) 1794 return; 1795 bp = *pbp; 1796 1797 IPFW_UH_RLOCK_ASSERT(chain); 1798 1799 for (i = 0 ; i < V_curr_dyn_buckets; i++) { 1800 IPFW_BUCK_LOCK(i); 1801 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) { 1802 if (bp + sizeof *p <= ep) { 1803 ipfw_dyn_rule *dst = 1804 (ipfw_dyn_rule *)bp; 1805 1806 export_dyn_rule(p, dst); 1807 last = dst; 1808 bp += sizeof(ipfw_dyn_rule); 1809 } 1810 } 1811 IPFW_BUCK_UNLOCK(i); 1812 } 1813 1814 if (last != NULL) /* mark last dynamic rule */ 1815 bzero(&last->next, sizeof(last)); 1816 *pbp = bp; 1817} 1818/* end of file */ 1819