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                         if (limit_mask & DYN_SRC_ADDR)
 719                                 id.src_ip6 = args->f_id.src_ip6;
 720                         if (limit_mask & DYN_DST_ADDR)
 721                                 id.dst_ip6 = args->f_id.dst_ip6;
 722                 } else {
 723                         if (limit_mask & DYN_SRC_ADDR)
 724                                 id.src_ip = args->f_id.src_ip;
 725                         if (limit_mask & DYN_DST_ADDR)
 726                                 id.dst_ip = args->f_id.dst_ip;
 727                 }
 728                 if (limit_mask & DYN_SRC_PORT)
 729                         id.src_port = args->f_id.src_port;
 730                 if (limit_mask & DYN_DST_PORT)
 731                         id.dst_port = args->f_id.dst_port;
 732
 733                 /*
 734                  * We have to release lock for previous bucket to
 735                  * avoid possible deadlock
 736                  */
 737                 IPFW_BUCK_UNLOCK(i);
 738
 739                 if ((parent = lookup_dyn_parent(&id, &pindex, rule)) == NULL) {
 740                         printf("ipfw: %s: add parent failed\n", __func__);
 741                         IPFW_BUCK_UNLOCK(pindex);
 742                         return (1);
 743                 }
 744
 745                 if (parent->count >= conn_limit) {
 746                         if (V_fw_verbose && last_log != time_uptime) {
 747                                 last_log = time_uptime;
 748                                 char sbuf[24];
 749                                 last_log = time_uptime;
 750                                 snprintf(sbuf, sizeof(sbuf),
 751                                     "%d drop session",
 752                                     parent->rule->rulenum);
 753                                 print_dyn_rule_flags(&args->f_id,
 754                                     cmd->o.opcode,
 755                                     LOG_SECURITY | LOG_DEBUG,
 756                                     sbuf, "too many entries");
 757                         }
 758                         IPFW_BUCK_UNLOCK(pindex);
 759                         return (1);
 760                 }
 761                 /* Increment counter on parent */
 762                 parent->count++;
 763                 IPFW_BUCK_UNLOCK(pindex);
 764
 765                 IPFW_BUCK_LOCK(i);
 766                 q = add_dyn_rule(&args->f_id, i, O_LIMIT, (struct ip_fw *)parent);
 767                 if (q == NULL) {
 768                         /* Decrement index and notify caller */
 769                         IPFW_BUCK_UNLOCK(i);
 770                         IPFW_BUCK_LOCK(pindex);
 771                         parent->count--;
 772                         IPFW_BUCK_UNLOCK(pindex);
 773                         return (1);
 774                 }
 775                 break;
 776         }
 777         default:
 778                 printf("ipfw: %s: unknown dynamic rule type %u\n",
 779                     __func__, cmd->o.opcode);
 780         }
 781
 782         if (q == NULL) {
 783                 IPFW_BUCK_UNLOCK(i);
 784                 return (1);     /* Notify caller about failure */
 785         }
 786
 787         /* XXX just set lifetime */
 788         lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
 789
 790         IPFW_BUCK_UNLOCK(i);
 791         return (0);
 792 }
 793
 794 /*
 795  * Generate a TCP packet, containing either a RST or a keepalive.
 796  * When flags & TH_RST, we are sending a RST packet, because of a
 797  * "reset" action matched the packet.
 798  * Otherwise we are sending a keepalive, and flags & TH_
 799  * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
 800  * so that MAC can label the reply appropriately.
 801  */
 802 struct mbuf *
 803 ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
 804     u_int32_t ack, int flags)
 805 {
 806         struct mbuf *m = NULL;          /* stupid compiler */
 807         int len, dir;
 808         struct ip *h = NULL;            /* stupid compiler */
 809 #ifdef INET6
 810         struct ip6_hdr *h6 = NULL;
 811 #endif
 812         struct tcphdr *th = NULL;
 813
 814         MGETHDR(m, M_DONTWAIT, MT_DATA);
 815         if (m == NULL)
 816                 return (NULL);
 817
 818         M_SETFIB(m, id->fib);
 819 #ifdef MAC
 820         if (replyto != NULL)
 821                 mac_netinet_firewall_reply(replyto, m);
 822         else
 823                 mac_netinet_firewall_send(m);
 824 #else
 825         (void)replyto;          /* don't warn about unused arg */
 826 #endif
 827
 828         switch (id->addr_type) {
 829         case 4:
 830                 len = sizeof(struct ip) + sizeof(struct tcphdr);
 831                 break;
 832 #ifdef INET6
 833         case 6:
 834                 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
 835                 break;
 836 #endif
 837         default:
 838                 /* XXX: log me?!? */
 839                 FREE_PKT(m);
 840                 return (NULL);
 841         }
 842         dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN);
 843
 844         m->m_data += max_linkhdr;
 845         m->m_flags |= M_SKIP_FIREWALL;
 846         m->m_pkthdr.len = m->m_len = len;
 847         m->m_pkthdr.rcvif = NULL;
 848         bzero(m->m_data, len);
 849
 850         switch (id->addr_type) {
 851         case 4:
 852                 h = mtod(m, struct ip *);
 853
 854                 /* prepare for checksum */
 855                 h->ip_p = IPPROTO_TCP;
 856                 h->ip_len = htons(sizeof(struct tcphdr));
 857                 if (dir) {
 858                         h->ip_src.s_addr = htonl(id->src_ip);
 859                         h->ip_dst.s_addr = htonl(id->dst_ip);
 860                 } else {
 861                         h->ip_src.s_addr = htonl(id->dst_ip);
 862                         h->ip_dst.s_addr = htonl(id->src_ip);
 863                 }
 864
 865                 th = (struct tcphdr *)(h + 1);
 866                 break;
 867 #ifdef INET6
 868         case 6:
 869                 h6 = mtod(m, struct ip6_hdr *);
 870
 871                 /* prepare for checksum */
 872                 h6->ip6_nxt = IPPROTO_TCP;
 873                 h6->ip6_plen = htons(sizeof(struct tcphdr));
 874                 if (dir) {
 875                         h6->ip6_src = id->src_ip6;
 876                         h6->ip6_dst = id->dst_ip6;
 877                 } else {
 878                         h6->ip6_src = id->dst_ip6;
 879                         h6->ip6_dst = id->src_ip6;
 880                 }
 881
 882                 th = (struct tcphdr *)(h6 + 1);
 883                 break;
 884 #endif
 885         }
 886
 887         if (dir) {
 888                 th->th_sport = htons(id->src_port);
 889                 th->th_dport = htons(id->dst_port);
 890         } else {
 891                 th->th_sport = htons(id->dst_port);
 892                 th->th_dport = htons(id->src_port);
 893         }
 894         th->th_off = sizeof(struct tcphdr) >> 2;
 895
 896         if (flags & TH_RST) {
 897                 if (flags & TH_ACK) {
 898                         th->th_seq = htonl(ack);
 899                         th->th_flags = TH_RST;
 900                 } else {
 901                         if (flags & TH_SYN)
 902                                 seq++;
 903                         th->th_ack = htonl(seq);
 904                         th->th_flags = TH_RST | TH_ACK;
 905                 }
 906         } else {
 907                 /*
 908                  * Keepalive - use caller provided sequence numbers
 909                  */
 910                 th->th_seq = htonl(seq);
 911                 th->th_ack = htonl(ack);
 912                 th->th_flags = TH_ACK;
 913         }
 914
 915         switch (id->addr_type) {
 916         case 4:
 917                 th->th_sum = in_cksum(m, len);
 918
 919                 /* finish the ip header */
 920                 h->ip_v = 4;
 921                 h->ip_hl = sizeof(*h) >> 2;
 922                 h->ip_tos = IPTOS_LOWDELAY;
 923                 h->ip_off = 0;
 924                 /* ip_len must be in host format for ip_output */
 925                 h->ip_len = len;
 926                 h->ip_ttl = V_ip_defttl;
 927                 h->ip_sum = 0;
 928                 break;
 929 #ifdef INET6
 930         case 6:
 931                 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6),
 932                     sizeof(struct tcphdr));
 933
 934                 /* finish the ip6 header */
 935                 h6->ip6_vfc |= IPV6_VERSION;
 936                 h6->ip6_hlim = IPV6_DEFHLIM;
 937                 break;
 938 #endif
 939         }
 940
 941         return (m);
 942 }
 943
 944 /*
 945  * Queue keepalive packets for given dynamic rule
 946  */
 947 static struct mbuf **
 948 ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q)
 949 {
 950         struct mbuf *m_rev, *m_fwd;
 951
 952         m_rev = (q->state & ACK_REV) ? NULL :
 953             ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
 954         m_fwd = (q->state & ACK_FWD) ? NULL :
 955             ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0);
 956
 957         if (m_rev != NULL) {
 958                 *mtailp = m_rev;
 959                 mtailp = &(*mtailp)->m_nextpkt;
 960         }
 961         if (m_fwd != NULL) {
 962                 *mtailp = m_fwd;
 963                 mtailp = &(*mtailp)->m_nextpkt;
 964         }
 965
 966         return (mtailp);
 967 }
 968
 969 /*
 970  * This procedure is used to perform various maintance
 971  * on dynamic hash list. Currently it is called every second.
 972  */
 973 static void
 974 ipfw_dyn_tick(void * vnetx)
 975 {
 976         struct ip_fw_chain *chain;
 977         int check_ka = 0;
 978 #ifdef VIMAGE
 979         struct vnet *vp = vnetx;
 980 #endif
 981
 982         CURVNET_SET(vp);
 983
 984         chain = &V_layer3_chain;
 985
 986         /* Run keepalive checks every keepalive_period iff ka is enabled */
 987         if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) &&
 988             (V_dyn_keepalive != 0)) {
 989                 V_dyn_keepalive_last = time_uptime;
 990                 check_ka = 1;
 991         }
 992
 993         check_dyn_rules(chain, NULL, RESVD_SET, check_ka, 1);
 994
 995         callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0);
 996
 997         CURVNET_RESTORE();
 998 }
 999
1000
1001 /*
1002  * Walk thru all dynamic states doing generic maintance:
1003  * 1) free expired states
1004  * 2) free all states based on deleted rule / set
1005  * 3) send keepalives for states if needed
1006  *
1007  * @chain - pointer to current ipfw rules chain
1008  * @rule - delete all states originated by given rule if != NULL
1009  * @set - delete all states originated by any rule in set @set if != RESVD_SET
1010  * @check_ka - perform checking/sending keepalives
1011  * @timer - indicate call from timer routine.
1012  *
1013  * Timer routine must call this function unlocked to permit
1014  * sending keepalives/resizing table.
1015  *
1016  * Others has to call function with IPFW_UH_WLOCK held.
1017  * Additionally, function assume that dynamic rule/set is
1018  * ALREADY deleted so no new states can be generated by
1019  * 'deleted' rules.
1020  *
1021  * Write lock is needed to ensure that unused parent rules
1022  * are not freed by other instance (see stage 2, 3)
1023  */
1024 static void
1025 check_dyn_rules(struct ip_fw_chain *chain, struct ip_fw *rule,
1026     int set, int check_ka, int timer)
1027 {
1028         struct mbuf *m0, *m, *mnext, **mtailp;
1029         struct ip *h;
1030         int i, dyn_count, new_buckets = 0, max_buckets;
1031         int expired = 0, expired_limits = 0, parents = 0, total = 0;
1032         ipfw_dyn_rule *q, *q_prev, *q_next;
1033         ipfw_dyn_rule *exp_head, **exptailp;
1034         ipfw_dyn_rule *exp_lhead, **expltailp;
1035
1036         KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated",
1037             __func__));
1038
1039         /* Avoid possible LOR */
1040         KASSERT(!check_ka || timer, ("%s: keepalive check with lock held",
1041             __func__));
1042
1043         /*
1044          * Do not perform any checks if we currently have no dynamic states
1045          */
1046         if (DYN_COUNT == 0)
1047                 return;
1048
1049         /* Expired states */
1050         exp_head = NULL;
1051         exptailp = &exp_head;
1052
1053         /* Expired limit states */
1054         exp_lhead = NULL;
1055         expltailp = &exp_lhead;
1056
1057         /*
1058          * We make a chain of packets to go out here -- not deferring
1059          * until after we drop the IPFW dynamic rule lock would result
1060          * in a lock order reversal with the normal packet input -> ipfw
1061          * call stack.
1062          */
1063         m0 = NULL;
1064         mtailp = &m0;
1065
1066         /* Protect from hash resizing */
1067         if (timer != 0)
1068                 IPFW_UH_WLOCK(chain);
1069         else
1070                 IPFW_UH_WLOCK_ASSERT(chain);
1071
1072 #define NEXT_RULE()     { q_prev = q; q = q->next ; continue; }
1073
1074         /* Stage 1: perform requested deletion */
1075         for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1076                 IPFW_BUCK_LOCK(i);
1077                 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) {
1078                         /* account every rule */
1079                         total++;
1080
1081                         /* Skip parent rules at all */
1082                         if (q->dyn_type == O_LIMIT_PARENT) {
1083                                 parents++;
1084                                 NEXT_RULE();
1085                         }
1086
1087                         /*
1088                          * Remove rules which are:
1089                          * 1) expired
1090                          * 2) created by given rule
1091                          * 3) created by any rule in given set
1092                          */
1093                         if ((TIME_LEQ(q->expire, time_uptime)) ||
1094                             ((rule != NULL) && (q->rule == rule)) ||
1095                             ((set != RESVD_SET) && (q->rule->set == set))) {
1096                                 /* Unlink q from current list */
1097                                 q_next = q->next;
1098                                 if (q == V_ipfw_dyn_v[i].head)
1099                                         V_ipfw_dyn_v[i].head = q_next;
1100                                 else
1101                                         q_prev->next = q_next;
1102
1103                                 q->next = NULL;
1104
1105                                 /* queue q to expire list */
1106                                 if (q->dyn_type != O_LIMIT) {
1107                                         *exptailp = q;
1108                                         exptailp = &(*exptailp)->next;
1109                                         DEB(print_dyn_rule(&q->id, q->dyn_type,
1110                                             "unlink entry", "left");
1111                                         )
1112                                 } else {
1113                                         /* Separate list for limit rules */
1114                                         *expltailp = q;
1115                                         expltailp = &(*expltailp)->next;
1116                                         expired_limits++;
1117                                         DEB(print_dyn_rule(&q->id, q->dyn_type,
1118                                             "unlink limit entry", "left");
1119                                         )
1120                                 }
1121
1122                                 q = q_next;
1123                                 expired++;
1124                                 continue;
1125                         }
1126
1127                         /*
1128                          * Check if we need to send keepalive:
1129                          * we need to ensure if is time to do KA,
1130                          * this is established TCP session, and
1131                          * expire time is within keepalive interval
1132                          */
1133                         if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) &&
1134                             ((q->state & BOTH_SYN) == BOTH_SYN) &&
1135                             (TIME_LEQ(q->expire, time_uptime +
1136                               V_dyn_keepalive_interval)))
1137                                 mtailp = ipfw_dyn_send_ka(mtailp, q);
1138
1139                         NEXT_RULE();
1140                 }
1141                 IPFW_BUCK_UNLOCK(i);
1142         }
1143
1144         /* Stage 2: decrement counters from O_LIMIT parents */
1145         if (expired_limits != 0) {
1146                 /*
1147                  * XXX: Note that deleting set with more than one
1148                  * heavily-used LIMIT rules can result in overwhelming
1149                  * locking due to lack of per-hash value sorting
1150                  *
1151                  * We should probably think about:
1152                  * 1) pre-allocating hash of size, say,
1153                  * MAX(16, V_curr_dyn_buckets / 1024)
1154                  * 2) checking if expired_limits is large enough
1155                  * 3) If yes, init hash (or its part), re-link
1156                  * current list and start decrementing procedure in
1157                  * each bucket separately
1158                  */
1159
1160                 /*
1161                  * Small optimization: do not unlock bucket until
1162                  * we see the next item resides in different bucket
1163                  */
1164                 if (exp_lhead != NULL) {
1165                         i = exp_lhead->parent->bucket;
1166                         IPFW_BUCK_LOCK(i);
1167                 }
1168                 for (q = exp_lhead; q != NULL; q = q->next) {
1169                         if (i != q->parent->bucket) {
1170                                 IPFW_BUCK_UNLOCK(i);
1171                                 i = q->parent->bucket;
1172                                 IPFW_BUCK_LOCK(i);
1173                         }
1174
1175                         /* Decrease parent refcount */
1176                         q->parent->count--;
1177                 }
1178                 if (exp_lhead != NULL)
1179                         IPFW_BUCK_UNLOCK(i);
1180         }
1181
1182         /*
1183          * We protectet ourselves from unused parent deletion
1184          * (from the timer function) by holding UH write lock.
1185          */
1186
1187         /* Stage 3: remove unused parent rules */
1188         if ((parents != 0) && (expired != 0)) {
1189                 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1190                         IPFW_BUCK_LOCK(i);
1191                         for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) {
1192                                 if (q->dyn_type != O_LIMIT_PARENT)
1193                                         NEXT_RULE();
1194
1195                                 if (q->count != 0)
1196                                         NEXT_RULE();
1197
1198                                 /* Parent rule without consumers */
1199
1200                                 /* Unlink q from current list */
1201                                 q_next = q->next;
1202                                 if (q == V_ipfw_dyn_v[i].head)
1203                                         V_ipfw_dyn_v[i].head = q_next;
1204                                 else
1205                                         q_prev->next = q_next;
1206
1207                                 q->next = NULL;
1208
1209                                 /* Add to expired list */
1210                                 *exptailp = q;
1211                                 exptailp = &(*exptailp)->next;
1212
1213                                 DEB(print_dyn_rule(&q->id, q->dyn_type,
1214                                     "unlink parent entry", "left");
1215                                 )
1216
1217                                 expired++;
1218
1219                                 q = q_next;
1220                         }
1221                         IPFW_BUCK_UNLOCK(i);
1222                 }
1223         }
1224
1225 #undef NEXT_RULE
1226
1227         if (timer != 0) {
1228                 /*
1229                  * Check if we need to resize hash:
1230                  * if current number of states exceeds number of buckes in hash,
1231                  * grow hash size to the minimum power of 2 which is bigger than
1232                  * current states count. Limit hash size by 64k.
1233                  */
1234                 max_buckets = (V_dyn_buckets_max > 65536) ?
1235                     65536 : V_dyn_buckets_max;
1236
1237                 dyn_count = DYN_COUNT;
1238
1239                 if ((dyn_count > V_curr_dyn_buckets * 2) &&
1240                     (dyn_count < max_buckets)) {
1241                         new_buckets = V_curr_dyn_buckets;
1242                         while (new_buckets < dyn_count) {
1243                                 new_buckets *= 2;
1244
1245                                 if (new_buckets >= max_buckets)
1246                                         break;
1247                         }
1248                 }
1249
1250                 IPFW_UH_WUNLOCK(chain);
1251         }
1252
1253         /* Finally delete old states ad limits if any */
1254         for (q = exp_head; q != NULL; q = q_next) {
1255                 q_next = q->next;
1256                 uma_zfree(V_ipfw_dyn_rule_zone, q);
1257         }
1258
1259         for (q = exp_lhead; q != NULL; q = q_next) {
1260                 q_next = q->next;
1261                 uma_zfree(V_ipfw_dyn_rule_zone, q);
1262         }
1263
1264         /*
1265          * The rest code MUST be called from timer routine only
1266          * without holding any locks
1267          */
1268         if (timer == 0)
1269                 return;
1270
1271         /* Send keepalive packets if any */
1272         for (m = m0; m != NULL; m = mnext) {
1273                 mnext = m->m_nextpkt;
1274                 m->m_nextpkt = NULL;
1275                 h = mtod(m, struct ip *);
1276                 if (h->ip_v == 4)
1277                         ip_output(m, NULL, NULL, 0, NULL, NULL);
1278 #ifdef INET6
1279                 else
1280                         ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1281 #endif
1282         }
1283
1284         /* Run table resize without holding any locks */
1285         if (new_buckets != 0)
1286                 resize_dynamic_table(chain, new_buckets);
1287 }
1288
1289 /*
1290  * Deletes all dynamic rules originated by given rule or all rules in
1291  * given set. Specify RESVD_SET to indicate set should not be used.
1292  * @chain - pointer to current ipfw rules chain
1293  * @rule - delete all states originated by given rule if != NULL
1294  * @set - delete all states originated by any rule in set @set if != RESVD_SET
1295  *
1296  * Function has to be called with IPFW_UH_WLOCK held.
1297  * Additionally, function assume that dynamic rule/set is
1298  * ALREADY deleted so no new states can be generated by
1299  * 'deleted' rules.
1300  */
1301 void
1302 ipfw_expire_dyn_rules(struct ip_fw_chain *chain, struct ip_fw *rule, int set)
1303 {
1304
1305         check_dyn_rules(chain, rule, set, 0, 0);
1306 }
1307
1308 void
1309 ipfw_dyn_init(struct ip_fw_chain *chain)
1310 {
1311
1312         V_ipfw_dyn_v = NULL;
1313         V_dyn_buckets_max = 256; /* must be power of 2 */
1314         V_curr_dyn_buckets = 256; /* must be power of 2 */
1315
1316         V_dyn_ack_lifetime = 300;
1317         V_dyn_syn_lifetime = 20;
1318         V_dyn_fin_lifetime = 1;
1319         V_dyn_rst_lifetime = 1;
1320         V_dyn_udp_lifetime = 10;
1321         V_dyn_short_lifetime = 5;
1322
1323         V_dyn_keepalive_interval = 20;
1324         V_dyn_keepalive_period = 5;
1325         V_dyn_keepalive = 1;    /* do send keepalives */
1326         V_dyn_keepalive_last = time_uptime;
1327
1328         V_dyn_max = 4096;       /* max # of dynamic rules */
1329
1330         V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule",
1331             sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
1332             UMA_ALIGN_PTR, 0);
1333
1334         /* Enforce limit on dynamic rules */
1335         uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1336
1337         callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE);
1338
1339         /*
1340          * This can potentially be done on first dynamic rule
1341          * being added to chain.
1342          */
1343         resize_dynamic_table(chain, V_curr_dyn_buckets);
1344 }
1345
1346 void
1347 ipfw_dyn_uninit(int pass)
1348 {
1349         int i;
1350
1351         if (pass == 0) {
1352                 callout_drain(&V_ipfw_timeout);
1353                 return;
1354         }
1355
1356         if (V_ipfw_dyn_v != NULL) {
1357                 /*
1358                  * Skip deleting all dynamic states -
1359                  * uma_zdestroy() does this more efficiently;
1360                  */
1361
1362                 /* Destroy all mutexes */
1363                 for (i = 0 ; i < V_curr_dyn_buckets ; i++)
1364                         IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]);
1365                 free(V_ipfw_dyn_v, M_IPFW);
1366                 V_ipfw_dyn_v = NULL;
1367         }
1368
1369         uma_zdestroy(V_ipfw_dyn_rule_zone);
1370 }
1371
1372 #ifdef SYSCTL_NODE
1373 /*
1374  * Get/set maximum number of dynamic states in given VNET instance.
1375  */
1376 static int
1377 sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS)
1378 {
1379         int error;
1380         unsigned int nstates;
1381
1382         nstates = V_dyn_max;
1383
1384         error = sysctl_handle_int(oidp, &nstates, 0, req);
1385         /* Read operation or some error */
1386         if ((error != 0) || (req->newptr == NULL))
1387                 return (error);
1388
1389         V_dyn_max = nstates;
1390         uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1391
1392         return (0);
1393 }
1394
1395 /*
1396  * Get current number of dynamic states in given VNET instance.
1397  */
1398 static int
1399 sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS)
1400 {
1401         int error;
1402         unsigned int nstates;
1403
1404         nstates = DYN_COUNT;
1405
1406         error = sysctl_handle_int(oidp, &nstates, 0, req);
1407
1408         return (error);
1409 }
1410 #endif
1411
1412 /*
1413  * Returns number of dynamic rules.
1414  */
1415 int
1416 ipfw_dyn_len(void)
1417 {
1418
1419         return (V_ipfw_dyn_v == NULL) ? 0 :
1420                 (DYN_COUNT * sizeof(ipfw_dyn_rule));
1421 }
1422
1423 /*
1424  * Fill given buffer with dynamic states.
1425  * IPFW_UH_RLOCK has to be held while calling.
1426  */
1427 void
1428 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep)
1429 {
1430         ipfw_dyn_rule *p, *last = NULL;
1431         char *bp;
1432         int i;
1433
1434         if (V_ipfw_dyn_v == NULL)
1435                 return;
1436         bp = *pbp;
1437
1438         IPFW_UH_RLOCK_ASSERT(chain);
1439
1440         for (i = 0 ; i < V_curr_dyn_buckets; i++) {
1441                 IPFW_BUCK_LOCK(i);
1442                 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) {
1443                         if (bp + sizeof *p <= ep) {
1444                                 ipfw_dyn_rule *dst =
1445                                         (ipfw_dyn_rule *)bp;
1446                                 bcopy(p, dst, sizeof *p);
1447                                 bcopy(&(p->rule->rulenum), &(dst->rule),
1448                                     sizeof(p->rule->rulenum));
1449                                 /*
1450                                  * store set number into high word of
1451                                  * dst->rule pointer.
1452                                  */
1453                                 bcopy(&(p->rule->set),
1454                                     (char *)&dst->rule +
1455                                     sizeof(p->rule->rulenum),
1456                                     sizeof(p->rule->set));
1457                                 /*
1458                                  * store a non-null value in "next".
1459                                  * The userland code will interpret a
1460                                  * NULL here as a marker
1461                                  * for the last dynamic rule.
1462                                  */
1463                                 bcopy(&dst, &dst->next, sizeof(dst));
1464                                 last = dst;
1465                                 dst->expire =
1466                                     TIME_LEQ(dst->expire, time_uptime) ?
1467                                         0 : dst->expire - time_uptime ;
1468                                 bp += sizeof(ipfw_dyn_rule);
1469                         }
1470                 }
1471                 IPFW_BUCK_UNLOCK(i);
1472         }
1473
1474         if (last != NULL) /* mark last dynamic rule */
1475                 bzero(&last->next, sizeof(last));
1476         *pbp = bp;
1477 }
1478 /* end of file */