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. This can be changed by dyn_keep_states
 110  * sysctl.
 111  *
 112  * There are some limitations with dynamic rules -- we do not
 113  * obey the 'randomized match', and we do not do multiple
 114  * passes through the firewall. XXX check the latter!!!
 115  */
 116
 117 struct ipfw_dyn_bucket {
 118         struct mtx      mtx;            /* Bucket protecting lock */
 119         ipfw_dyn_rule   *head;          /* Pointer to first rule */
 120 };
 121
 122 /*
 123  * Static variables followed by global ones
 124  */
 125 static VNET_DEFINE(struct ipfw_dyn_bucket *, ipfw_dyn_v);
 126 static VNET_DEFINE(u_int32_t, dyn_buckets_max);
 127 static VNET_DEFINE(u_int32_t, curr_dyn_buckets);
 128 static VNET_DEFINE(struct callout, ipfw_timeout);
 129 #define V_ipfw_dyn_v                    VNET(ipfw_dyn_v)
 130 #define V_dyn_buckets_max               VNET(dyn_buckets_max)
 131 #define V_curr_dyn_buckets              VNET(curr_dyn_buckets)
 132 #define V_ipfw_timeout                  VNET(ipfw_timeout)
 133
 134 static VNET_DEFINE(uma_zone_t, ipfw_dyn_rule_zone);
 135 #define V_ipfw_dyn_rule_zone            VNET(ipfw_dyn_rule_zone)
 136
 137 #define IPFW_BUCK_LOCK_INIT(b)  \
 138         mtx_init(&(b)->mtx, "IPFW dynamic bucket", NULL, MTX_DEF)
 139 #define IPFW_BUCK_LOCK_DESTROY(b)       \
 140         mtx_destroy(&(b)->mtx)
 141 #define IPFW_BUCK_LOCK(i)       mtx_lock(&V_ipfw_dyn_v[(i)].mtx)
 142 #define IPFW_BUCK_UNLOCK(i)     mtx_unlock(&V_ipfw_dyn_v[(i)].mtx)
 143 #define IPFW_BUCK_ASSERT(i)     mtx_assert(&V_ipfw_dyn_v[(i)].mtx, MA_OWNED)
 144
 145
 146 static VNET_DEFINE(int, dyn_keep_states);
 147 #define V_dyn_keep_states               VNET(dyn_keep_states)
 148
 149 /*
 150  * Timeouts for various events in handing dynamic rules.
 151  */
 152 static VNET_DEFINE(u_int32_t, dyn_ack_lifetime);
 153 static VNET_DEFINE(u_int32_t, dyn_syn_lifetime);
 154 static VNET_DEFINE(u_int32_t, dyn_fin_lifetime);
 155 static VNET_DEFINE(u_int32_t, dyn_rst_lifetime);
 156 static VNET_DEFINE(u_int32_t, dyn_udp_lifetime);
 157 static VNET_DEFINE(u_int32_t, dyn_short_lifetime);
 158
 159 #define V_dyn_ack_lifetime              VNET(dyn_ack_lifetime)
 160 #define V_dyn_syn_lifetime              VNET(dyn_syn_lifetime)
 161 #define V_dyn_fin_lifetime              VNET(dyn_fin_lifetime)
 162 #define V_dyn_rst_lifetime              VNET(dyn_rst_lifetime)
 163 #define V_dyn_udp_lifetime              VNET(dyn_udp_lifetime)
 164 #define V_dyn_short_lifetime            VNET(dyn_short_lifetime)
 165
 166 /*
 167  * Keepalives are sent if dyn_keepalive is set. They are sent every
 168  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 169  * seconds of lifetime of a rule.
 170  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 171  * than dyn_keepalive_period.
 172  */
 173
 174 static VNET_DEFINE(u_int32_t, dyn_keepalive_interval);
 175 static VNET_DEFINE(u_int32_t, dyn_keepalive_period);
 176 static VNET_DEFINE(u_int32_t, dyn_keepalive);
 177 static VNET_DEFINE(time_t, dyn_keepalive_last);
 178
 179 #define V_dyn_keepalive_interval        VNET(dyn_keepalive_interval)
 180 #define V_dyn_keepalive_period          VNET(dyn_keepalive_period)
 181 #define V_dyn_keepalive                 VNET(dyn_keepalive)
 182 #define V_dyn_keepalive_last            VNET(dyn_keepalive_last)
 183
 184 static VNET_DEFINE(u_int32_t, dyn_max);         /* max # of dynamic rules */
 185
 186 #define DYN_COUNT                       uma_zone_get_cur(V_ipfw_dyn_rule_zone)
 187 #define V_dyn_max                       VNET(dyn_max)
 188
 189 /* for userspace, we emulate the uma_zone_counter with ipfw_dyn_count */
 190 static int ipfw_dyn_count;      /* number of objects */
 191
 192 #ifdef USERSPACE /* emulation of UMA object counters for userspace */
 193 #define uma_zone_get_cur(x)     ipfw_dyn_count
 194 #endif /* USERSPACE */
 195
 196 static int last_log;    /* Log ratelimiting */
 197
 198 static void ipfw_dyn_tick(void *vnetx);
 199 static void check_dyn_rules(struct ip_fw_chain *, struct ip_fw *,
 200     int, int, int);
 201 #ifdef SYSCTL_NODE
 202
 203 static int sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS);
 204 static int sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS);
 205
 206 SYSBEGIN(f2)
 207
 208 SYSCTL_DECL(_net_inet_ip_fw);
 209 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_buckets,
 210     CTLFLAG_RW, &VNET_NAME(dyn_buckets_max), 0,
 211     "Max number of dyn. buckets");
 212 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets,
 213     CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0,
 214     "Current Number of dyn. buckets");
 215 SYSCTL_VNET_PROC(_net_inet_ip_fw, OID_AUTO, dyn_count,
 216     CTLTYPE_UINT|CTLFLAG_RD, 0, 0, sysctl_ipfw_dyn_count, "IU",
 217     "Number of dyn. rules");
 218 SYSCTL_VNET_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max,
 219     CTLTYPE_UINT|CTLFLAG_RW, 0, 0, sysctl_ipfw_dyn_max, "IU",
 220     "Max number of dyn. rules");
 221 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime,
 222     CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0,
 223     "Lifetime of dyn. rules for acks");
 224 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime,
 225     CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0,
 226     "Lifetime of dyn. rules for syn");
 227 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
 228     CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0,
 229     "Lifetime of dyn. rules for fin");
 230 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
 231     CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0,
 232     "Lifetime of dyn. rules for rst");
 233 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime,
 234     CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0,
 235     "Lifetime of dyn. rules for UDP");
 236 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime,
 237     CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0,
 238     "Lifetime of dyn. rules for other situations");
 239 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive,
 240     CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0,
 241     "Enable keepalives for dyn. rules");
 242 SYSCTL_VNET_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keep_states,
 243     CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0,
 244     "Do not flush dynamic states on rule deletion");
 245
 246 SYSEND
 247
 248 #endif /* SYSCTL_NODE */
 249
 250
 251 static __inline int
 252 hash_packet6(struct ipfw_flow_id *id)
 253 {
 254         u_int32_t i;
 255         i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
 256             (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
 257             (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
 258             (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
 259             (id->dst_port) ^ (id->src_port);
 260         return i;
 261 }
 262
 263 /*
 264  * IMPORTANT: the hash function for dynamic rules must be commutative
 265  * in source and destination (ip,port), because rules are bidirectional
 266  * and we want to find both in the same bucket.
 267  */
 268 static __inline int
 269 hash_packet(struct ipfw_flow_id *id, int buckets)
 270 {
 271         u_int32_t i;
 272
 273 #ifdef INET6
 274         if (IS_IP6_FLOW_ID(id))
 275                 i = hash_packet6(id);
 276         else
 277 #endif /* INET6 */
 278         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 279         i &= (buckets - 1);
 280         return i;
 281 }
 282
 283 /**
 284  * Print customizable flow id description via log(9) facility.
 285  */
 286 static void
 287 print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags,
 288     char *prefix, char *postfix)
 289 {
 290         struct in_addr da;
 291 #ifdef INET6
 292         char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN];
 293 #else
 294         char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
 295 #endif
 296
 297 #ifdef INET6
 298         if (IS_IP6_FLOW_ID(id)) {
 299                 ip6_sprintf(src, &id->src_ip6);
 300                 ip6_sprintf(dst, &id->dst_ip6);
 301         } else
 302 #endif
 303         {
 304                 da.s_addr = htonl(id->src_ip);
 305                 inet_ntop(AF_INET, &da, src, sizeof(src));
 306                 da.s_addr = htonl(id->dst_ip);
 307                 inet_ntop(AF_INET, &da, dst, sizeof(dst));
 308         }
 309         log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n",
 310             prefix, dyn_type, src, id->src_port, dst,
 311             id->dst_port, DYN_COUNT, postfix);
 312 }
 313
 314 #define print_dyn_rule(id, dtype, prefix, postfix)      \
 315         print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix)
 316
 317 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
 318 #define TIME_LE(a,b)       ((int)((a)-(b)) < 0)
 319
 320 /*
 321  * Lookup a dynamic rule, locked version.
 322  */
 323 static ipfw_dyn_rule *
 324 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction,
 325     struct tcphdr *tcp)
 326 {
 327         /*
 328          * Stateful ipfw extensions.
 329          * Lookup into dynamic session queue.
 330          */
 331 #define MATCH_REVERSE   0
 332 #define MATCH_FORWARD   1
 333 #define MATCH_NONE      2
 334 #define MATCH_UNKNOWN   3
 335         int dir = MATCH_NONE;
 336         ipfw_dyn_rule *prev, *q = NULL;
 337
 338         IPFW_BUCK_ASSERT(i);
 339
 340         for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) {
 341                 if (q->dyn_type == O_LIMIT_PARENT && q->count)
 342                         continue;
 343
 344                 if (pkt->proto != q->id.proto || q->dyn_type == O_LIMIT_PARENT)
 345                         continue;
 346
 347                 if (IS_IP6_FLOW_ID(pkt)) {
 348                         if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) &&
 349                             IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) &&
 350                             pkt->src_port == q->id.src_port &&
 351                             pkt->dst_port == q->id.dst_port) {
 352                                 dir = MATCH_FORWARD;
 353                                 break;
 354                         }
 355                         if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) &&
 356                             IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) &&
 357                             pkt->src_port == q->id.dst_port &&
 358                             pkt->dst_port == q->id.src_port) {
 359                                 dir = MATCH_REVERSE;
 360                                 break;
 361                         }
 362                 } else {
 363                         if (pkt->src_ip == q->id.src_ip &&
 364                             pkt->dst_ip == q->id.dst_ip &&
 365                             pkt->src_port == q->id.src_port &&
 366                             pkt->dst_port == q->id.dst_port) {
 367                                 dir = MATCH_FORWARD;
 368                                 break;
 369                         }
 370                         if (pkt->src_ip == q->id.dst_ip &&
 371                             pkt->dst_ip == q->id.src_ip &&
 372                             pkt->src_port == q->id.dst_port &&
 373                             pkt->dst_port == q->id.src_port) {
 374                                 dir = MATCH_REVERSE;
 375                                 break;
 376                         }
 377                 }
 378         }
 379         if (q == NULL)
 380                 goto done;      /* q = NULL, not found */
 381
 382         if (prev != NULL) {     /* found and not in front */
 383                 prev->next = q->next;
 384                 q->next = V_ipfw_dyn_v[i].head;
 385                 V_ipfw_dyn_v[i].head = q;
 386         }
 387         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 388                 uint32_t ack;
 389                 u_char flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST);
 390
 391 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 392 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 393 #define TCP_FLAGS       (TH_FLAGS | (TH_FLAGS << 8))
 394 #define ACK_FWD         0x10000                 /* fwd ack seen */
 395 #define ACK_REV         0x20000                 /* rev ack seen */
 396
 397                 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8);
 398                 switch (q->state & TCP_FLAGS) {
 399                 case TH_SYN:                    /* opening */
 400                         q->expire = time_uptime + V_dyn_syn_lifetime;
 401                         break;
 402
 403                 case BOTH_SYN:                  /* move to established */
 404                 case BOTH_SYN | TH_FIN:         /* one side tries to close */
 405                 case BOTH_SYN | (TH_FIN << 8):
 406 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
 407                         if (tcp == NULL)
 408                                 break;
 409
 410                         ack = ntohl(tcp->th_ack);
 411                         if (dir == MATCH_FORWARD) {
 412                                 if (q->ack_fwd == 0 ||
 413                                     _SEQ_GE(ack, q->ack_fwd)) {
 414                                         q->ack_fwd = ack;
 415                                         q->state |= ACK_FWD;
 416                                 }
 417                         } else {
 418                                 if (q->ack_rev == 0 ||
 419                                     _SEQ_GE(ack, q->ack_rev)) {
 420                                         q->ack_rev = ack;
 421                                         q->state |= ACK_REV;
 422                                 }
 423                         }
 424                         if ((q->state & (ACK_FWD | ACK_REV)) ==
 425                             (ACK_FWD | ACK_REV)) {
 426                                 q->expire = time_uptime + V_dyn_ack_lifetime;
 427                                 q->state &= ~(ACK_FWD | ACK_REV);
 428                         }
 429                         break;
 430
 431                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 432                         if (V_dyn_fin_lifetime >= V_dyn_keepalive_period)
 433                                 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1;
 434                         q->expire = time_uptime + V_dyn_fin_lifetime;
 435                         break;
 436
 437                 default:
 438 #if 0
 439                         /*
 440                          * reset or some invalid combination, but can also
 441                          * occur if we use keep-state the wrong way.
 442                          */
 443                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
 444                                 printf("invalid state: 0x%x\n", q->state);
 445 #endif
 446                         if (V_dyn_rst_lifetime >= V_dyn_keepalive_period)
 447                                 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1;
 448                         q->expire = time_uptime + V_dyn_rst_lifetime;
 449                         break;
 450                 }
 451         } else if (pkt->proto == IPPROTO_UDP) {
 452                 q->expire = time_uptime + V_dyn_udp_lifetime;
 453         } else {
 454                 /* other protocols */
 455                 q->expire = time_uptime + V_dyn_short_lifetime;
 456         }
 457 done:
 458         if (match_direction != NULL)
 459                 *match_direction = dir;
 460         return (q);
 461 }
 462
 463 ipfw_dyn_rule *
 464 ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 465     struct tcphdr *tcp)
 466 {
 467         ipfw_dyn_rule *q;
 468         int i;
 469
 470         i = hash_packet(pkt, V_curr_dyn_buckets);
 471
 472         IPFW_BUCK_LOCK(i);
 473         q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp);
 474         if (q == NULL)
 475                 IPFW_BUCK_UNLOCK(i);
 476         /* NB: return table locked when q is not NULL */
 477         return q;
 478 }
 479
 480 /*
 481  * Unlock bucket mtx
 482  * @p - pointer to dynamic rule
 483  */
 484 void
 485 ipfw_dyn_unlock(ipfw_dyn_rule *q)
 486 {
 487
 488         IPFW_BUCK_UNLOCK(q->bucket);
 489 }
 490
 491 static int
 492 resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets)
 493 {
 494         int i, k, nbuckets_old;
 495         ipfw_dyn_rule *q;
 496         struct ipfw_dyn_bucket *dyn_v, *dyn_v_old;
 497
 498         /* Check if given number is power of 2 and less than 64k */
 499         if ((nbuckets > 65536) || (!powerof2(nbuckets)))
 500                 return 1;
 501
 502         CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__,
 503             V_curr_dyn_buckets, nbuckets);
 504
 505         /* Allocate and initialize new hash */
 506         dyn_v = malloc(nbuckets * sizeof(ipfw_dyn_rule), M_IPFW,
 507             M_WAITOK | M_ZERO);
 508
 509         for (i = 0 ; i < nbuckets; i++)
 510                 IPFW_BUCK_LOCK_INIT(&dyn_v[i]);
 511
 512         /*
 513          * Call upper half lock, as get_map() do to ease
 514          * read-only access to dynamic rules hash from sysctl
 515          */
 516         IPFW_UH_WLOCK(chain);
 517
 518         /*
 519          * Acquire chain write lock to permit hash access
 520          * for main traffic path without additional locks
 521          */
 522         IPFW_WLOCK(chain);
 523
 524         /* Save old values */
 525         nbuckets_old = V_curr_dyn_buckets;
 526         dyn_v_old = V_ipfw_dyn_v;
 527
 528         /* Skip relinking if array is not set up */
 529         if (V_ipfw_dyn_v == NULL)
 530                 V_curr_dyn_buckets = 0;
 531
 532         /* Re-link all dynamic states */
 533         for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
 534                 while (V_ipfw_dyn_v[i].head != NULL) {
 535                         /* Remove from current chain */
 536                         q = V_ipfw_dyn_v[i].head;
 537                         V_ipfw_dyn_v[i].head = q->next;
 538
 539                         /* Get new hash value */
 540                         k = hash_packet(&q->id, nbuckets);
 541                         q->bucket = k;
 542                         /* Add to the new head */
 543                         q->next = dyn_v[k].head;
 544                         dyn_v[k].head = q;
 545              }
 546         }
 547
 548         /* Update current pointers/buckets values */
 549         V_curr_dyn_buckets = nbuckets;
 550         V_ipfw_dyn_v = dyn_v;
 551
 552         IPFW_WUNLOCK(chain);
 553
 554         IPFW_UH_WUNLOCK(chain);
 555
 556         /* Start periodic callout on initial creation */
 557         if (dyn_v_old == NULL) {
 558                 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0);
 559                 return (0);
 560         }
 561
 562         /* Destroy all mutexes */
 563         for (i = 0 ; i < nbuckets_old ; i++)
 564                 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]);
 565
 566         /* Free old hash */
 567         free(dyn_v_old, M_IPFW);
 568
 569         return 0;
 570 }
 571
 572 /**
 573  * Install state of type 'type' for a dynamic session.
 574  * The hash table contains two type of rules:
 575  * - regular rules (O_KEEP_STATE)
 576  * - rules for sessions with limited number of sess per user
 577  *   (O_LIMIT). When they are created, the parent is
 578  *   increased by 1, and decreased on delete. In this case,
 579  *   the third parameter is the parent rule and not the chain.
 580  * - "parent" rules for the above (O_LIMIT_PARENT).
 581  */
 582 static ipfw_dyn_rule *
 583 add_dyn_rule(struct ipfw_flow_id *id, int i, u_int8_t dyn_type, struct ip_fw *rule)
 584 {
 585         ipfw_dyn_rule *r;
 586
 587         IPFW_BUCK_ASSERT(i);
 588
 589         r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
 590         if (r == NULL) {
 591                 if (last_log != time_uptime) {
 592                         last_log = time_uptime;
 593                         log(LOG_DEBUG, "ipfw: %s: Cannot allocate rule\n",
 594                             __func__);
 595                 }
 596                 return NULL;
 597         }
 598         ipfw_dyn_count++;
 599
 600         /*
 601          * refcount on parent is already incremented, so
 602          * it is safe to use parent unlocked.
 603          */
 604         if (dyn_type == O_LIMIT) {
 605                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 606                 if ( parent->dyn_type != O_LIMIT_PARENT)
 607                         panic("invalid parent");
 608                 r->parent = parent;
 609                 rule = parent->rule;
 610         }
 611
 612         r->id = *id;
 613         r->expire = time_uptime + V_dyn_syn_lifetime;
 614         r->rule = rule;
 615         r->dyn_type = dyn_type;
 616         IPFW_ZERO_DYN_COUNTER(r);
 617         r->count = 0;
 618
 619         r->bucket = i;
 620         r->next = V_ipfw_dyn_v[i].head;
 621         V_ipfw_dyn_v[i].head = r;
 622         DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");)
 623         return r;
 624 }
 625
 626 /**
 627  * lookup dynamic parent rule using pkt and rule as search keys.
 628  * If the lookup fails, then install one.
 629  */
 630 static ipfw_dyn_rule *
 631 lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule)
 632 {
 633         ipfw_dyn_rule *q;
 634         int i, is_v6;
 635
 636         is_v6 = IS_IP6_FLOW_ID(pkt);
 637         i = hash_packet( pkt, V_curr_dyn_buckets );
 638         *pindex = i;
 639         IPFW_BUCK_LOCK(i);
 640         for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next)
 641                 if (q->dyn_type == O_LIMIT_PARENT &&
 642                     rule== q->rule &&
 643                     pkt->proto == q->id.proto &&
 644                     pkt->src_port == q->id.src_port &&
 645                     pkt->dst_port == q->id.dst_port &&
 646                     (
 647                         (is_v6 &&
 648                          IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
 649                                 &(q->id.src_ip6)) &&
 650                          IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
 651                                 &(q->id.dst_ip6))) ||
 652                         (!is_v6 &&
 653                          pkt->src_ip == q->id.src_ip &&
 654                          pkt->dst_ip == q->id.dst_ip)
 655                     )
 656                 ) {
 657                         q->expire = time_uptime + V_dyn_short_lifetime;
 658                         DEB(print_dyn_rule(pkt, q->dyn_type,
 659                             "lookup_dyn_parent found", "");)
 660                         return q;
 661                 }
 662
 663         /* Add virtual limiting rule */
 664         return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule);
 665 }
 666
 667 /**
 668  * Install dynamic state for rule type cmd->o.opcode
 669  *
 670  * Returns 1 (failure) if state is not installed because of errors or because
 671  * session limitations are enforced.
 672  */
 673 int
 674 ipfw_install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
 675     struct ip_fw_args *args, uint32_t tablearg)
 676 {
 677         ipfw_dyn_rule *q;
 678         int i;
 679
 680         DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", "");)
 681
 682         i = hash_packet(&args->f_id, V_curr_dyn_buckets);
 683
 684         IPFW_BUCK_LOCK(i);
 685
 686         q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
 687
 688         if (q != NULL) {        /* should never occur */
 689                 DEB(
 690                 if (last_log != time_uptime) {
 691                         last_log = time_uptime;
 692                         printf("ipfw: %s: entry already present, done\n",
 693                             __func__);
 694                 })
 695                 IPFW_BUCK_UNLOCK(i);
 696                 return (0);
 697         }
 698
 699         /*
 700          * State limiting is done via uma(9) zone limiting.
 701          * Save pointer to newly-installed rule and reject
 702          * packet if add_dyn_rule() returned NULL.
 703          * Note q is currently set to NULL.
 704          */
 705
 706         switch (cmd->o.opcode) {
 707         case O_KEEP_STATE:      /* bidir rule */
 708                 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule);
 709                 break;
 710
 711         case O_LIMIT: {         /* limit number of sessions */
 712                 struct ipfw_flow_id id;
 713                 ipfw_dyn_rule *parent;
 714                 uint32_t conn_limit;
 715                 uint16_t limit_mask = cmd->limit_mask;
 716                 int pindex;
 717
 718                 conn_limit = IP_FW_ARG_TABLEARG(cmd->conn_limit);
 719
 720                 DEB(
 721                 if (cmd->conn_limit == IP_FW_TABLEARG)
 722                         printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
 723                             "(tablearg)\n", __func__, conn_limit);
 724                 else
 725                         printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
 726                             __func__, conn_limit);
 727                 )
 728
 729                 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
 730                 id.proto = args->f_id.proto;
 731                 id.addr_type = args->f_id.addr_type;
 732                 id.fib = M_GETFIB(args->m);
 733
 734                 if (IS_IP6_FLOW_ID (&(args->f_id))) {
 735                         if (limit_mask & DYN_SRC_ADDR)
 736                                 id.src_ip6 = args->f_id.src_ip6;
 737                         if (limit_mask & DYN_DST_ADDR)
 738                                 id.dst_ip6 = args->f_id.dst_ip6;
 739                 } else {
 740                         if (limit_mask & DYN_SRC_ADDR)
 741                                 id.src_ip = args->f_id.src_ip;
 742                         if (limit_mask & DYN_DST_ADDR)
 743                                 id.dst_ip = args->f_id.dst_ip;
 744                 }
 745                 if (limit_mask & DYN_SRC_PORT)
 746                         id.src_port = args->f_id.src_port;
 747                 if (limit_mask & DYN_DST_PORT)
 748                         id.dst_port = args->f_id.dst_port;
 749
 750                 /*
 751                  * We have to release lock for previous bucket to
 752                  * avoid possible deadlock
 753                  */
 754                 IPFW_BUCK_UNLOCK(i);
 755
 756                 if ((parent = lookup_dyn_parent(&id, &pindex, rule)) == NULL) {
 757                         printf("ipfw: %s: add parent failed\n", __func__);
 758                         IPFW_BUCK_UNLOCK(pindex);
 759                         return (1);
 760                 }
 761
 762                 if (parent->count >= conn_limit) {
 763                         if (V_fw_verbose && last_log != time_uptime) {
 764                                 last_log = time_uptime;
 765                                 char sbuf[24];
 766                                 last_log = time_uptime;
 767                                 snprintf(sbuf, sizeof(sbuf),
 768                                     "%d drop session",
 769                                     parent->rule->rulenum);
 770                                 print_dyn_rule_flags(&args->f_id,
 771                                     cmd->o.opcode,
 772                                     LOG_SECURITY | LOG_DEBUG,
 773                                     sbuf, "too many entries");
 774                         }
 775                         IPFW_BUCK_UNLOCK(pindex);
 776                         return (1);
 777                 }
 778                 /* Increment counter on parent */
 779                 parent->count++;
 780                 IPFW_BUCK_UNLOCK(pindex);
 781
 782                 IPFW_BUCK_LOCK(i);
 783                 q = add_dyn_rule(&args->f_id, i, O_LIMIT, (struct ip_fw *)parent);
 784                 if (q == NULL) {
 785                         /* Decrement index and notify caller */
 786                         IPFW_BUCK_UNLOCK(i);
 787                         IPFW_BUCK_LOCK(pindex);
 788                         parent->count--;
 789                         IPFW_BUCK_UNLOCK(pindex);
 790                         return (1);
 791                 }
 792                 break;
 793         }
 794         default:
 795                 printf("ipfw: %s: unknown dynamic rule type %u\n",
 796                     __func__, cmd->o.opcode);
 797         }
 798
 799         if (q == NULL) {
 800                 IPFW_BUCK_UNLOCK(i);
 801                 return (1);     /* Notify caller about failure */
 802         }
 803
 804         /* XXX just set lifetime */
 805         lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
 806
 807         IPFW_BUCK_UNLOCK(i);
 808         return (0);
 809 }
 810
 811 /*
 812  * Generate a TCP packet, containing either a RST or a keepalive.
 813  * When flags & TH_RST, we are sending a RST packet, because of a
 814  * "reset" action matched the packet.
 815  * Otherwise we are sending a keepalive, and flags & TH_
 816  * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
 817  * so that MAC can label the reply appropriately.
 818  */
 819 struct mbuf *
 820 ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
 821     u_int32_t ack, int flags)
 822 {
 823         struct mbuf *m = NULL;          /* stupid compiler */
 824         int len, dir;
 825         struct ip *h = NULL;            /* stupid compiler */
 826 #ifdef INET6
 827         struct ip6_hdr *h6 = NULL;
 828 #endif
 829         struct tcphdr *th = NULL;
 830
 831         MGETHDR(m, M_NOWAIT, MT_DATA);
 832         if (m == NULL)
 833                 return (NULL);
 834
 835         M_SETFIB(m, id->fib);
 836 #ifdef MAC
 837         if (replyto != NULL)
 838                 mac_netinet_firewall_reply(replyto, m);
 839         else
 840                 mac_netinet_firewall_send(m);
 841 #else
 842         (void)replyto;          /* don't warn about unused arg */
 843 #endif
 844
 845         switch (id->addr_type) {
 846         case 4:
 847                 len = sizeof(struct ip) + sizeof(struct tcphdr);
 848                 break;
 849 #ifdef INET6
 850         case 6:
 851                 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
 852                 break;
 853 #endif
 854         default:
 855                 /* XXX: log me?!? */
 856                 FREE_PKT(m);
 857                 return (NULL);
 858         }
 859         dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN);
 860
 861         m->m_data += max_linkhdr;
 862         m->m_flags |= M_SKIP_FIREWALL;
 863         m->m_pkthdr.len = m->m_len = len;
 864         m->m_pkthdr.rcvif = NULL;
 865         bzero(m->m_data, len);
 866
 867         switch (id->addr_type) {
 868         case 4:
 869                 h = mtod(m, struct ip *);
 870
 871                 /* prepare for checksum */
 872                 h->ip_p = IPPROTO_TCP;
 873                 h->ip_len = htons(sizeof(struct tcphdr));
 874                 if (dir) {
 875                         h->ip_src.s_addr = htonl(id->src_ip);
 876                         h->ip_dst.s_addr = htonl(id->dst_ip);
 877                 } else {
 878                         h->ip_src.s_addr = htonl(id->dst_ip);
 879                         h->ip_dst.s_addr = htonl(id->src_ip);
 880                 }
 881
 882                 th = (struct tcphdr *)(h + 1);
 883                 break;
 884 #ifdef INET6
 885         case 6:
 886                 h6 = mtod(m, struct ip6_hdr *);
 887
 888                 /* prepare for checksum */
 889                 h6->ip6_nxt = IPPROTO_TCP;
 890                 h6->ip6_plen = htons(sizeof(struct tcphdr));
 891                 if (dir) {
 892                         h6->ip6_src = id->src_ip6;
 893                         h6->ip6_dst = id->dst_ip6;
 894                 } else {
 895                         h6->ip6_src = id->dst_ip6;
 896                         h6->ip6_dst = id->src_ip6;
 897                 }
 898
 899                 th = (struct tcphdr *)(h6 + 1);
 900                 break;
 901 #endif
 902         }
 903
 904         if (dir) {
 905                 th->th_sport = htons(id->src_port);
 906                 th->th_dport = htons(id->dst_port);
 907         } else {
 908                 th->th_sport = htons(id->dst_port);
 909                 th->th_dport = htons(id->src_port);
 910         }
 911         th->th_off = sizeof(struct tcphdr) >> 2;
 912
 913         if (flags & TH_RST) {
 914                 if (flags & TH_ACK) {
 915                         th->th_seq = htonl(ack);
 916                         th->th_flags = TH_RST;
 917                 } else {
 918                         if (flags & TH_SYN)
 919                                 seq++;
 920                         th->th_ack = htonl(seq);
 921                         th->th_flags = TH_RST | TH_ACK;
 922                 }
 923         } else {
 924                 /*
 925                  * Keepalive - use caller provided sequence numbers
 926                  */
 927                 th->th_seq = htonl(seq);
 928                 th->th_ack = htonl(ack);
 929                 th->th_flags = TH_ACK;
 930         }
 931
 932         switch (id->addr_type) {
 933         case 4:
 934                 th->th_sum = in_cksum(m, len);
 935
 936                 /* finish the ip header */
 937                 h->ip_v = 4;
 938                 h->ip_hl = sizeof(*h) >> 2;
 939                 h->ip_tos = IPTOS_LOWDELAY;
 940                 h->ip_off = htons(0);
 941                 h->ip_len = htons(len);
 942                 h->ip_ttl = V_ip_defttl;
 943                 h->ip_sum = 0;
 944                 break;
 945 #ifdef INET6
 946         case 6:
 947                 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6),
 948                     sizeof(struct tcphdr));
 949
 950                 /* finish the ip6 header */
 951                 h6->ip6_vfc |= IPV6_VERSION;
 952                 h6->ip6_hlim = IPV6_DEFHLIM;
 953                 break;
 954 #endif
 955         }
 956
 957         return (m);
 958 }
 959
 960 /*
 961  * Queue keepalive packets for given dynamic rule
 962  */
 963 static struct mbuf **
 964 ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q)
 965 {
 966         struct mbuf *m_rev, *m_fwd;
 967
 968         m_rev = (q->state & ACK_REV) ? NULL :
 969             ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
 970         m_fwd = (q->state & ACK_FWD) ? NULL :
 971             ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0);
 972
 973         if (m_rev != NULL) {
 974                 *mtailp = m_rev;
 975                 mtailp = &(*mtailp)->m_nextpkt;
 976         }
 977         if (m_fwd != NULL) {
 978                 *mtailp = m_fwd;
 979                 mtailp = &(*mtailp)->m_nextpkt;
 980         }
 981
 982         return (mtailp);
 983 }
 984
 985 /*
 986  * This procedure is used to perform various maintance
 987  * on dynamic hash list. Currently it is called every second.
 988  */
 989 static void
 990 ipfw_dyn_tick(void * vnetx)
 991 {
 992         struct ip_fw_chain *chain;
 993         int check_ka = 0;
 994 #ifdef VIMAGE
 995         struct vnet *vp = vnetx;
 996 #endif
 997
 998         CURVNET_SET(vp);
 999
1000         chain = &V_layer3_chain;
1001
1002         /* Run keepalive checks every keepalive_period iff ka is enabled */
1003         if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) &&
1004             (V_dyn_keepalive != 0)) {
1005                 V_dyn_keepalive_last = time_uptime;
1006                 check_ka = 1;
1007         }
1008
1009         check_dyn_rules(chain, NULL, RESVD_SET, check_ka, 1);
1010
1011         callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0);
1012
1013         CURVNET_RESTORE();
1014 }
1015
1016
1017 /*
1018  * Walk thru all dynamic states doing generic maintance:
1019  * 1) free expired states
1020  * 2) free all states based on deleted rule / set
1021  * 3) send keepalives for states if needed
1022  *
1023  * @chain - pointer to current ipfw rules chain
1024  * @rule - delete all states originated by given rule if != NULL
1025  * @set - delete all states originated by any rule in set @set if != RESVD_SET
1026  * @check_ka - perform checking/sending keepalives
1027  * @timer - indicate call from timer routine.
1028  *
1029  * Timer routine must call this function unlocked to permit
1030  * sending keepalives/resizing table.
1031  *
1032  * Others has to call function with IPFW_UH_WLOCK held.
1033  * Additionally, function assume that dynamic rule/set is
1034  * ALREADY deleted so no new states can be generated by
1035  * 'deleted' rules.
1036  *
1037  * Write lock is needed to ensure that unused parent rules
1038  * are not freed by other instance (see stage 2, 3)
1039  */
1040 static void
1041 check_dyn_rules(struct ip_fw_chain *chain, struct ip_fw *rule,
1042     int set, int check_ka, int timer)
1043 {
1044         struct mbuf *m0, *m, *mnext, **mtailp;
1045         struct ip *h;
1046         int i, dyn_count, new_buckets = 0, max_buckets;
1047         int expired = 0, expired_limits = 0, parents = 0, total = 0;
1048         ipfw_dyn_rule *q, *q_prev, *q_next;
1049         ipfw_dyn_rule *exp_head, **exptailp;
1050         ipfw_dyn_rule *exp_lhead, **expltailp;
1051
1052         KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated",
1053             __func__));
1054
1055         /* Avoid possible LOR */
1056         KASSERT(!check_ka || timer, ("%s: keepalive check with lock held",
1057             __func__));
1058
1059         /*
1060          * Do not perform any checks if we currently have no dynamic states
1061          */
1062         if (DYN_COUNT == 0)
1063                 return;
1064
1065         /* Expired states */
1066         exp_head = NULL;
1067         exptailp = &exp_head;
1068
1069         /* Expired limit states */
1070         exp_lhead = NULL;
1071         expltailp = &exp_lhead;
1072
1073         /*
1074          * We make a chain of packets to go out here -- not deferring
1075          * until after we drop the IPFW dynamic rule lock would result
1076          * in a lock order reversal with the normal packet input -> ipfw
1077          * call stack.
1078          */
1079         m0 = NULL;
1080         mtailp = &m0;
1081
1082         /* Protect from hash resizing */
1083         if (timer != 0)
1084                 IPFW_UH_WLOCK(chain);
1085         else
1086                 IPFW_UH_WLOCK_ASSERT(chain);
1087
1088 #define NEXT_RULE()     { q_prev = q; q = q->next ; continue; }
1089
1090         /* Stage 1: perform requested deletion */
1091         for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1092                 IPFW_BUCK_LOCK(i);
1093                 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) {
1094                         /* account every rule */
1095                         total++;
1096
1097                         /* Skip parent rules at all */
1098                         if (q->dyn_type == O_LIMIT_PARENT) {
1099                                 parents++;
1100                                 NEXT_RULE();
1101                         }
1102
1103                         /*
1104                          * Remove rules which are:
1105                          * 1) expired
1106                          * 2) created by given rule
1107                          * 3) created by any rule in given set
1108                          */
1109                         if ((TIME_LEQ(q->expire, time_uptime)) ||
1110                             ((rule != NULL) && (q->rule == rule)) ||
1111                             ((set != RESVD_SET) && (q->rule->set == set))) {
1112                                 if (TIME_LE(time_uptime, q->expire) &&
1113                                     q->dyn_type == O_KEEP_STATE &&
1114                                     V_dyn_keep_states != 0) {
1115                                         /*
1116                                          * Do not delete state if
1117                                          * it is not expired and
1118                                          * dyn_keep_states is ON.
1119                                          * However we need to re-link it
1120                                          * to any other stable rule
1121                                          */
1122                                         q->rule = chain->default_rule;
1123                                         NEXT_RULE();
1124                                 }
1125
1126                                 /* Unlink q from current list */
1127                                 q_next = q->next;
1128                                 if (q == V_ipfw_dyn_v[i].head)
1129                                         V_ipfw_dyn_v[i].head = q_next;
1130                                 else
1131                                         q_prev->next = q_next;
1132
1133                                 q->next = NULL;
1134
1135                                 /* queue q to expire list */
1136                                 if (q->dyn_type != O_LIMIT) {
1137                                         *exptailp = q;
1138                                         exptailp = &(*exptailp)->next;
1139                                         DEB(print_dyn_rule(&q->id, q->dyn_type,
1140                                             "unlink entry", "left");
1141                                         )
1142                                 } else {
1143                                         /* Separate list for limit rules */
1144                                         *expltailp = q;
1145                                         expltailp = &(*expltailp)->next;
1146                                         expired_limits++;
1147                                         DEB(print_dyn_rule(&q->id, q->dyn_type,
1148                                             "unlink limit entry", "left");
1149                                         )
1150                                 }
1151
1152                                 q = q_next;
1153                                 expired++;
1154                                 continue;
1155                         }
1156
1157                         /*
1158                          * Check if we need to send keepalive:
1159                          * we need to ensure if is time to do KA,
1160                          * this is established TCP session, and
1161                          * expire time is within keepalive interval
1162                          */
1163                         if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) &&
1164                             ((q->state & BOTH_SYN) == BOTH_SYN) &&
1165                             (TIME_LEQ(q->expire, time_uptime +
1166                               V_dyn_keepalive_interval)))
1167                                 mtailp = ipfw_dyn_send_ka(mtailp, q);
1168
1169                         NEXT_RULE();
1170                 }
1171                 IPFW_BUCK_UNLOCK(i);
1172         }
1173
1174         /* Stage 2: decrement counters from O_LIMIT parents */
1175         if (expired_limits != 0) {
1176                 /*
1177                  * XXX: Note that deleting set with more than one
1178                  * heavily-used LIMIT rules can result in overwhelming
1179                  * locking due to lack of per-hash value sorting
1180                  *
1181                  * We should probably think about:
1182                  * 1) pre-allocating hash of size, say,
1183                  * MAX(16, V_curr_dyn_buckets / 1024)
1184                  * 2) checking if expired_limits is large enough
1185                  * 3) If yes, init hash (or its part), re-link
1186                  * current list and start decrementing procedure in
1187                  * each bucket separately
1188                  */
1189
1190                 /*
1191                  * Small optimization: do not unlock bucket until
1192                  * we see the next item resides in different bucket
1193                  */
1194                 if (exp_lhead != NULL) {
1195                         i = exp_lhead->parent->bucket;
1196                         IPFW_BUCK_LOCK(i);
1197                 }
1198                 for (q = exp_lhead; q != NULL; q = q->next) {
1199                         if (i != q->parent->bucket) {
1200                                 IPFW_BUCK_UNLOCK(i);
1201                                 i = q->parent->bucket;
1202                                 IPFW_BUCK_LOCK(i);
1203                         }
1204
1205                         /* Decrease parent refcount */
1206                         q->parent->count--;
1207                 }
1208                 if (exp_lhead != NULL)
1209                         IPFW_BUCK_UNLOCK(i);
1210         }
1211
1212         /*
1213          * We protectet ourselves from unused parent deletion
1214          * (from the timer function) by holding UH write lock.
1215          */
1216
1217         /* Stage 3: remove unused parent rules */
1218         if ((parents != 0) && (expired != 0)) {
1219                 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1220                         IPFW_BUCK_LOCK(i);
1221                         for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) {
1222                                 if (q->dyn_type != O_LIMIT_PARENT)
1223                                         NEXT_RULE();
1224
1225                                 if (q->count != 0)
1226                                         NEXT_RULE();
1227
1228                                 /* Parent rule without consumers */
1229
1230                                 /* Unlink q from current list */
1231                                 q_next = q->next;
1232                                 if (q == V_ipfw_dyn_v[i].head)
1233                                         V_ipfw_dyn_v[i].head = q_next;
1234                                 else
1235                                         q_prev->next = q_next;
1236
1237                                 q->next = NULL;
1238
1239                                 /* Add to expired list */
1240                                 *exptailp = q;
1241                                 exptailp = &(*exptailp)->next;
1242
1243                                 DEB(print_dyn_rule(&q->id, q->dyn_type,
1244                                     "unlink parent entry", "left");
1245                                 )
1246
1247                                 expired++;
1248
1249                                 q = q_next;
1250                         }
1251                         IPFW_BUCK_UNLOCK(i);
1252                 }
1253         }
1254
1255 #undef NEXT_RULE
1256
1257         if (timer != 0) {
1258                 /*
1259                  * Check if we need to resize hash:
1260                  * if current number of states exceeds number of buckes in hash,
1261                  * grow hash size to the minimum power of 2 which is bigger than
1262                  * current states count. Limit hash size by 64k.
1263                  */
1264                 max_buckets = (V_dyn_buckets_max > 65536) ?
1265                     65536 : V_dyn_buckets_max;
1266
1267                 dyn_count = DYN_COUNT;
1268
1269                 if ((dyn_count > V_curr_dyn_buckets * 2) &&
1270                     (dyn_count < max_buckets)) {
1271                         new_buckets = V_curr_dyn_buckets;
1272                         while (new_buckets < dyn_count) {
1273                                 new_buckets *= 2;
1274
1275                                 if (new_buckets >= max_buckets)
1276                                         break;
1277                         }
1278                 }
1279
1280                 IPFW_UH_WUNLOCK(chain);
1281         }
1282
1283         /* Finally delete old states ad limits if any */
1284         for (q = exp_head; q != NULL; q = q_next) {
1285                 q_next = q->next;
1286                 uma_zfree(V_ipfw_dyn_rule_zone, q);
1287                 ipfw_dyn_count--;
1288         }
1289
1290         for (q = exp_lhead; q != NULL; q = q_next) {
1291                 q_next = q->next;
1292                 uma_zfree(V_ipfw_dyn_rule_zone, q);
1293                 ipfw_dyn_count--;
1294         }
1295
1296         /*
1297          * The rest code MUST be called from timer routine only
1298          * without holding any locks
1299          */
1300         if (timer == 0)
1301                 return;
1302
1303         /* Send keepalive packets if any */
1304         for (m = m0; m != NULL; m = mnext) {
1305                 mnext = m->m_nextpkt;
1306                 m->m_nextpkt = NULL;
1307                 h = mtod(m, struct ip *);
1308                 if (h->ip_v == 4)
1309                         ip_output(m, NULL, NULL, 0, NULL, NULL);
1310 #ifdef INET6
1311                 else
1312                         ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1313 #endif
1314         }
1315
1316         /* Run table resize without holding any locks */
1317         if (new_buckets != 0)
1318                 resize_dynamic_table(chain, new_buckets);
1319 }
1320
1321 /*
1322  * Deletes all dynamic rules originated by given rule or all rules in
1323  * given set. Specify RESVD_SET to indicate set should not be used.
1324  * @chain - pointer to current ipfw rules chain
1325  * @rule - delete all states originated by given rule if != NULL
1326  * @set - delete all states originated by any rule in set @set if != RESVD_SET
1327  *
1328  * Function has to be called with IPFW_UH_WLOCK held.
1329  * Additionally, function assume that dynamic rule/set is
1330  * ALREADY deleted so no new states can be generated by
1331  * 'deleted' rules.
1332  */
1333 void
1334 ipfw_expire_dyn_rules(struct ip_fw_chain *chain, struct ip_fw *rule, int set)
1335 {
1336
1337         check_dyn_rules(chain, rule, set, 0, 0);
1338 }
1339
1340 void
1341 ipfw_dyn_init(struct ip_fw_chain *chain)
1342 {
1343
1344         V_ipfw_dyn_v = NULL;
1345         V_dyn_buckets_max = 256; /* must be power of 2 */
1346         V_curr_dyn_buckets = 256; /* must be power of 2 */
1347
1348         V_dyn_ack_lifetime = 300;
1349         V_dyn_syn_lifetime = 20;
1350         V_dyn_fin_lifetime = 1;
1351         V_dyn_rst_lifetime = 1;
1352         V_dyn_udp_lifetime = 10;
1353         V_dyn_short_lifetime = 5;
1354
1355         V_dyn_keepalive_interval = 20;
1356         V_dyn_keepalive_period = 5;
1357         V_dyn_keepalive = 1;    /* do send keepalives */
1358         V_dyn_keepalive_last = time_uptime;
1359
1360         V_dyn_max = 4096;       /* max # of dynamic rules */
1361
1362         V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule",
1363             sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
1364             UMA_ALIGN_PTR, 0);
1365
1366         /* Enforce limit on dynamic rules */
1367         uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1368
1369         callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE);
1370
1371         /*
1372          * This can potentially be done on first dynamic rule
1373          * being added to chain.
1374          */
1375         resize_dynamic_table(chain, V_curr_dyn_buckets);
1376 }
1377
1378 void
1379 ipfw_dyn_uninit(int pass)
1380 {
1381         int i;
1382
1383         if (pass == 0) {
1384                 callout_drain(&V_ipfw_timeout);
1385                 return;
1386         }
1387
1388         if (V_ipfw_dyn_v != NULL) {
1389                 /*
1390                  * Skip deleting all dynamic states -
1391                  * uma_zdestroy() does this more efficiently;
1392                  */
1393
1394                 /* Destroy all mutexes */
1395                 for (i = 0 ; i < V_curr_dyn_buckets ; i++)
1396                         IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]);
1397                 free(V_ipfw_dyn_v, M_IPFW);
1398                 V_ipfw_dyn_v = NULL;
1399         }
1400
1401         uma_zdestroy(V_ipfw_dyn_rule_zone);
1402 }
1403
1404 #ifdef SYSCTL_NODE
1405 /*
1406  * Get/set maximum number of dynamic states in given VNET instance.
1407  */
1408 static int
1409 sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS)
1410 {
1411         int error;
1412         unsigned int nstates;
1413
1414         nstates = V_dyn_max;
1415
1416         error = sysctl_handle_int(oidp, &nstates, 0, req);
1417         /* Read operation or some error */
1418         if ((error != 0) || (req->newptr == NULL))
1419                 return (error);
1420
1421         V_dyn_max = nstates;
1422         uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1423
1424         return (0);
1425 }
1426
1427 /*
1428  * Get current number of dynamic states in given VNET instance.
1429  */
1430 static int
1431 sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS)
1432 {
1433         int error;
1434         unsigned int nstates;
1435
1436         nstates = DYN_COUNT;
1437
1438         error = sysctl_handle_int(oidp, &nstates, 0, req);
1439
1440         return (error);
1441 }
1442 #endif
1443
1444 /*
1445  * Returns number of dynamic rules.
1446  */
1447 int
1448 ipfw_dyn_len(void)
1449 {
1450
1451         return (V_ipfw_dyn_v == NULL) ? 0 :
1452                 (DYN_COUNT * sizeof(ipfw_dyn_rule));
1453 }
1454
1455 /*
1456  * Fill given buffer with dynamic states.
1457  * IPFW_UH_RLOCK has to be held while calling.
1458  */
1459 void
1460 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep)
1461 {
1462         ipfw_dyn_rule *p, *last = NULL;
1463         char *bp;
1464         int i;
1465
1466         if (V_ipfw_dyn_v == NULL)
1467                 return;
1468         bp = *pbp;
1469
1470         IPFW_UH_RLOCK_ASSERT(chain);
1471
1472         for (i = 0 ; i < V_curr_dyn_buckets; i++) {
1473                 IPFW_BUCK_LOCK(i);
1474                 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) {
1475                         if (bp + sizeof *p <= ep) {
1476                                 ipfw_dyn_rule *dst =
1477                                         (ipfw_dyn_rule *)bp;
1478                                 bcopy(p, dst, sizeof *p);
1479                                 bcopy(&(p->rule->rulenum), &(dst->rule),
1480                                     sizeof(p->rule->rulenum));
1481                                 /*
1482                                  * store set number into high word of
1483                                  * dst->rule pointer.
1484                                  */
1485                                 bcopy(&(p->rule->set),
1486                                     (char *)&dst->rule +
1487                                     sizeof(p->rule->rulenum),
1488                                     sizeof(p->rule->set));
1489                                 /*
1490                                  * store a non-null value in "next".
1491                                  * The userland code will interpret a
1492                                  * NULL here as a marker
1493                                  * for the last dynamic rule.
1494                                  */
1495                                 bcopy(&dst, &dst->next, sizeof(dst));
1496                                 last = dst;
1497                                 dst->expire =
1498                                     TIME_LEQ(dst->expire, time_uptime) ?
1499                                         0 : dst->expire - time_uptime ;
1500                                 bp += sizeof(ipfw_dyn_rule);
1501                         }
1502                 }
1503                 IPFW_BUCK_UNLOCK(i);
1504         }
1505
1506         if (last != NULL) /* mark last dynamic rule */
1507                 bzero(&last->next, sizeof(last));
1508         *pbp = bp;
1509 }
1510 /* end of file */