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