sys/netpfil/ipfw/ip_fw_dynamic.c

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