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 #ifdef INET6
 252 static __inline int
 253 hash_packet6(struct ipfw_flow_id *id)
 254 {
 255         u_int32_t i;
 256         i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
 257             (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
 258             (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
 259             (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
 260             (id->dst_port) ^ (id->src_port);
 261         return i;
 262 }
 263 #endif
 264
 265 /*
 266  * IMPORTANT: the hash function for dynamic rules must be commutative
 267  * in source and destination (ip,port), because rules are bidirectional
 268  * and we want to find both in the same bucket.
 269  */
 270 static __inline int
 271 hash_packet(struct ipfw_flow_id *id, int buckets)
 272 {
 273         u_int32_t i;
 274
 275 #ifdef INET6
 276         if (IS_IP6_FLOW_ID(id))
 277                 i = hash_packet6(id);
 278         else
 279 #endif /* INET6 */
 280         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 281         i &= (buckets - 1);
 282         return i;
 283 }
 284
 285 /**
 286  * Print customizable flow id description via log(9) facility.
 287  */
 288 static void
 289 print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags,
 290     char *prefix, char *postfix)
 291 {
 292         struct in_addr da;
 293 #ifdef INET6
 294         char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN];
 295 #else
 296         char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
 297 #endif
 298
 299 #ifdef INET6
 300         if (IS_IP6_FLOW_ID(id)) {
 301                 ip6_sprintf(src, &id->src_ip6);
 302                 ip6_sprintf(dst, &id->dst_ip6);
 303         } else
 304 #endif
 305         {
 306                 da.s_addr = htonl(id->src_ip);
 307                 inet_ntop(AF_INET, &da, src, sizeof(src));
 308                 da.s_addr = htonl(id->dst_ip);
 309                 inet_ntop(AF_INET, &da, dst, sizeof(dst));
 310         }
 311         log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n",
 312             prefix, dyn_type, src, id->src_port, dst,
 313             id->dst_port, DYN_COUNT, postfix);
 314 }
 315
 316 #define print_dyn_rule(id, dtype, prefix, postfix)      \
 317         print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix)
 318
 319 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
 320 #define TIME_LE(a,b)       ((int)((a)-(b)) < 0)
 321
 322 /*
 323  * Lookup a dynamic rule, locked version.
 324  */
 325 static ipfw_dyn_rule *
 326 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction,
 327     struct tcphdr *tcp)
 328 {
 329         /*
 330          * Stateful ipfw extensions.
 331          * Lookup into dynamic session queue.
 332          */
 333 #define MATCH_REVERSE   0
 334 #define MATCH_FORWARD   1
 335 #define MATCH_NONE      2
 336 #define MATCH_UNKNOWN   3
 337         int dir = MATCH_NONE;
 338         ipfw_dyn_rule *prev, *q = NULL;
 339
 340         IPFW_BUCK_ASSERT(i);
 341
 342         for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) {
 343                 if (q->dyn_type == O_LIMIT_PARENT && q->count)
 344                         continue;
 345
 346                 if (pkt->proto != q->id.proto || q->dyn_type == O_LIMIT_PARENT)
 347                         continue;
 348
 349                 if (IS_IP6_FLOW_ID(pkt)) {
 350                         if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) &&
 351                             IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) &&
 352                             pkt->src_port == q->id.src_port &&
 353                             pkt->dst_port == q->id.dst_port) {
 354                                 dir = MATCH_FORWARD;
 355                                 break;
 356                         }
 357                         if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) &&
 358                             IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) &&
 359                             pkt->src_port == q->id.dst_port &&
 360                             pkt->dst_port == q->id.src_port) {
 361                                 dir = MATCH_REVERSE;
 362                                 break;
 363                         }
 364                 } else {
 365                         if (pkt->src_ip == q->id.src_ip &&
 366                             pkt->dst_ip == q->id.dst_ip &&
 367                             pkt->src_port == q->id.src_port &&
 368                             pkt->dst_port == q->id.dst_port) {
 369                                 dir = MATCH_FORWARD;
 370                                 break;
 371                         }
 372                         if (pkt->src_ip == q->id.dst_ip &&
 373                             pkt->dst_ip == q->id.src_ip &&
 374                             pkt->src_port == q->id.dst_port &&
 375                             pkt->dst_port == q->id.src_port) {
 376                                 dir = MATCH_REVERSE;
 377                                 break;
 378                         }
 379                 }
 380         }
 381         if (q == NULL)
 382                 goto done;      /* q = NULL, not found */
 383
 384         if (prev != NULL) {     /* found and not in front */
 385                 prev->next = q->next;
 386                 q->next = V_ipfw_dyn_v[i].head;
 387                 V_ipfw_dyn_v[i].head = q;
 388         }
 389         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 390                 uint32_t ack;
 391                 u_char flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST);
 392
 393 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 394 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 395 #define TCP_FLAGS       (TH_FLAGS | (TH_FLAGS << 8))
 396 #define ACK_FWD         0x10000                 /* fwd ack seen */
 397 #define ACK_REV         0x20000                 /* rev ack seen */
 398
 399                 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8);
 400                 switch (q->state & TCP_FLAGS) {
 401                 case TH_SYN:                    /* opening */
 402                         q->expire = time_uptime + V_dyn_syn_lifetime;
 403                         break;
 404
 405                 case BOTH_SYN:                  /* move to established */
 406                 case BOTH_SYN | TH_FIN:         /* one side tries to close */
 407                 case BOTH_SYN | (TH_FIN << 8):
 408 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
 409                         if (tcp == NULL)
 410                                 break;
 411
 412                         ack = ntohl(tcp->th_ack);
 413                         if (dir == MATCH_FORWARD) {
 414                                 if (q->ack_fwd == 0 ||
 415                                     _SEQ_GE(ack, q->ack_fwd)) {
 416                                         q->ack_fwd = ack;
 417                                         q->state |= ACK_FWD;
 418                                 }
 419                         } else {
 420                                 if (q->ack_rev == 0 ||
 421                                     _SEQ_GE(ack, q->ack_rev)) {
 422                                         q->ack_rev = ack;
 423                                         q->state |= ACK_REV;
 424                                 }
 425                         }
 426                         if ((q->state & (ACK_FWD | ACK_REV)) ==
 427                             (ACK_FWD | ACK_REV)) {
 428                                 q->expire = time_uptime + V_dyn_ack_lifetime;
 429                                 q->state &= ~(ACK_FWD | ACK_REV);
 430                         }
 431                         break;
 432
 433                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 434                         if (V_dyn_fin_lifetime >= V_dyn_keepalive_period)
 435                                 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1;
 436                         q->expire = time_uptime + V_dyn_fin_lifetime;
 437                         break;
 438
 439                 default:
 440 #if 0
 441                         /*
 442                          * reset or some invalid combination, but can also
 443                          * occur if we use keep-state the wrong way.
 444                          */
 445                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
 446                                 printf("invalid state: 0x%x\n", q->state);
 447 #endif
 448                         if (V_dyn_rst_lifetime >= V_dyn_keepalive_period)
 449                                 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1;
 450                         q->expire = time_uptime + V_dyn_rst_lifetime;
 451                         break;
 452                 }
 453         } else if (pkt->proto == IPPROTO_UDP) {
 454                 q->expire = time_uptime + V_dyn_udp_lifetime;
 455         } else {
 456                 /* other protocols */
 457                 q->expire = time_uptime + V_dyn_short_lifetime;
 458         }
 459 done:
 460         if (match_direction != NULL)
 461                 *match_direction = dir;
 462         return (q);
 463 }
 464
 465 ipfw_dyn_rule *
 466 ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 467     struct tcphdr *tcp)
 468 {
 469         ipfw_dyn_rule *q;
 470         int i;
 471
 472         i = hash_packet(pkt, V_curr_dyn_buckets);
 473
 474         IPFW_BUCK_LOCK(i);
 475         q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp);
 476         if (q == NULL)
 477                 IPFW_BUCK_UNLOCK(i);
 478         /* NB: return table locked when q is not NULL */
 479         return q;
 480 }
 481
 482 /*
 483  * Unlock bucket mtx
 484  * @p - pointer to dynamic rule
 485  */
 486 void
 487 ipfw_dyn_unlock(ipfw_dyn_rule *q)
 488 {
 489
 490         IPFW_BUCK_UNLOCK(q->bucket);
 491 }
 492
 493 static int
 494 resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets)
 495 {
 496         int i, k, nbuckets_old;
 497         ipfw_dyn_rule *q;
 498         struct ipfw_dyn_bucket *dyn_v, *dyn_v_old;
 499
 500         /* Check if given number is power of 2 and less than 64k */
 501         if ((nbuckets > 65536) || (!powerof2(nbuckets)))
 502                 return 1;
 503
 504         CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__,
 505             V_curr_dyn_buckets, nbuckets);
 506
 507         /* Allocate and initialize new hash */
 508         dyn_v = malloc(nbuckets * sizeof(ipfw_dyn_rule), M_IPFW,
 509             M_WAITOK | M_ZERO);
 510
 511         for (i = 0 ; i < nbuckets; i++)
 512                 IPFW_BUCK_LOCK_INIT(&dyn_v[i]);
 513
 514         /*
 515          * Call upper half lock, as get_map() do to ease
 516          * read-only access to dynamic rules hash from sysctl
 517          */
 518         IPFW_UH_WLOCK(chain);
 519
 520         /*
 521          * Acquire chain write lock to permit hash access
 522          * for main traffic path without additional locks
 523          */
 524         IPFW_WLOCK(chain);
 525
 526         /* Save old values */
 527         nbuckets_old = V_curr_dyn_buckets;
 528         dyn_v_old = V_ipfw_dyn_v;
 529
 530         /* Skip relinking if array is not set up */
 531         if (V_ipfw_dyn_v == NULL)
 532                 V_curr_dyn_buckets = 0;
 533
 534         /* Re-link all dynamic states */
 535         for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
 536                 while (V_ipfw_dyn_v[i].head != NULL) {
 537                         /* Remove from current chain */
 538                         q = V_ipfw_dyn_v[i].head;
 539                         V_ipfw_dyn_v[i].head = q->next;
 540
 541                         /* Get new hash value */
 542                         k = hash_packet(&q->id, nbuckets);
 543                         q->bucket = k;
 544                         /* Add to the new head */
 545                         q->next = dyn_v[k].head;
 546                         dyn_v[k].head = q;
 547              }
 548         }
 549
 550         /* Update current pointers/buckets values */
 551         V_curr_dyn_buckets = nbuckets;
 552         V_ipfw_dyn_v = dyn_v;
 553
 554         IPFW_WUNLOCK(chain);
 555
 556         IPFW_UH_WUNLOCK(chain);
 557
 558         /* Start periodic callout on initial creation */
 559         if (dyn_v_old == NULL) {
 560                 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0);
 561                 return (0);
 562         }
 563
 564         /* Destroy all mutexes */
 565         for (i = 0 ; i < nbuckets_old ; i++)
 566                 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]);
 567
 568         /* Free old hash */
 569         free(dyn_v_old, M_IPFW);
 570
 571         return 0;
 572 }
 573
 574 /**
 575  * Install state of type 'type' for a dynamic session.
 576  * The hash table contains two type of rules:
 577  * - regular rules (O_KEEP_STATE)
 578  * - rules for sessions with limited number of sess per user
 579  *   (O_LIMIT). When they are created, the parent is
 580  *   increased by 1, and decreased on delete. In this case,
 581  *   the third parameter is the parent rule and not the chain.
 582  * - "parent" rules for the above (O_LIMIT_PARENT).
 583  */
 584 static ipfw_dyn_rule *
 585 add_dyn_rule(struct ipfw_flow_id *id, int i, u_int8_t dyn_type, struct ip_fw *rule)
 586 {
 587         ipfw_dyn_rule *r;
 588
 589         IPFW_BUCK_ASSERT(i);
 590
 591         r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
 592         if (r == NULL) {
 593                 if (last_log != time_uptime) {
 594                         last_log = time_uptime;
 595                         log(LOG_DEBUG, "ipfw: %s: Cannot allocate rule\n",
 596                             __func__);
 597                 }
 598                 return NULL;
 599         }
 600         ipfw_dyn_count++;
 601
 602         /*
 603          * refcount on parent is already incremented, so
 604          * it is safe to use parent unlocked.
 605          */
 606         if (dyn_type == O_LIMIT) {
 607                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 608                 if ( parent->dyn_type != O_LIMIT_PARENT)
 609                         panic("invalid parent");
 610                 r->parent = parent;
 611                 rule = parent->rule;
 612         }
 613
 614         r->id = *id;
 615         r->expire = time_uptime + V_dyn_syn_lifetime;
 616         r->rule = rule;
 617         r->dyn_type = dyn_type;
 618         IPFW_ZERO_DYN_COUNTER(r);
 619         r->count = 0;
 620
 621         r->bucket = i;
 622         r->next = V_ipfw_dyn_v[i].head;
 623         V_ipfw_dyn_v[i].head = r;
 624         DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");)
 625         return r;
 626 }
 627
 628 /**
 629  * lookup dynamic parent rule using pkt and rule as search keys.
 630  * If the lookup fails, then install one.
 631  */
 632 static ipfw_dyn_rule *
 633 lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule)
 634 {
 635         ipfw_dyn_rule *q;
 636         int i, is_v6;
 637
 638         is_v6 = IS_IP6_FLOW_ID(pkt);
 639         i = hash_packet( pkt, V_curr_dyn_buckets );
 640         *pindex = i;
 641         IPFW_BUCK_LOCK(i);
 642         for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next)
 643                 if (q->dyn_type == O_LIMIT_PARENT &&
 644                     rule== q->rule &&
 645                     pkt->proto == q->id.proto &&
 646                     pkt->src_port == q->id.src_port &&
 647                     pkt->dst_port == q->id.dst_port &&
 648                     (
 649                         (is_v6 &&
 650                          IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
 651                                 &(q->id.src_ip6)) &&
 652                          IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
 653                                 &(q->id.dst_ip6))) ||
 654                         (!is_v6 &&
 655                          pkt->src_ip == q->id.src_ip &&
 656                          pkt->dst_ip == q->id.dst_ip)
 657                     )
 658                 ) {
 659                         q->expire = time_uptime + V_dyn_short_lifetime;
 660                         DEB(print_dyn_rule(pkt, q->dyn_type,
 661                             "lookup_dyn_parent found", "");)
 662                         return q;
 663                 }
 664
 665         /* Add virtual limiting rule */
 666         return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule);
 667 }
 668
 669 /**
 670  * Install dynamic state for rule type cmd->o.opcode
 671  *
 672  * Returns 1 (failure) if state is not installed because of errors or because
 673  * session limitations are enforced.
 674  */
 675 int
 676 ipfw_install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
 677     struct ip_fw_args *args, uint32_t tablearg)
 678 {
 679         ipfw_dyn_rule *q;
 680         int i;
 681
 682         DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", "");)
 683
 684         i = hash_packet(&args->f_id, V_curr_dyn_buckets);
 685
 686         IPFW_BUCK_LOCK(i);
 687
 688         q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
 689
 690         if (q != NULL) {        /* should never occur */
 691                 DEB(
 692                 if (last_log != time_uptime) {
 693                         last_log = time_uptime;
 694                         printf("ipfw: %s: entry already present, done\n",
 695                             __func__);
 696                 })
 697                 IPFW_BUCK_UNLOCK(i);
 698                 return (0);
 699         }
 700
 701         /*
 702          * State limiting is done via uma(9) zone limiting.
 703          * Save pointer to newly-installed rule and reject
 704          * packet if add_dyn_rule() returned NULL.
 705          * Note q is currently set to NULL.
 706          */
 707
 708         switch (cmd->o.opcode) {
 709         case O_KEEP_STATE:      /* bidir rule */
 710                 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule);
 711                 break;
 712
 713         case O_LIMIT: {         /* limit number of sessions */
 714                 struct ipfw_flow_id id;
 715                 ipfw_dyn_rule *parent;
 716                 uint32_t conn_limit;
 717                 uint16_t limit_mask = cmd->limit_mask;
 718                 int pindex;
 719
 720                 conn_limit = IP_FW_ARG_TABLEARG(cmd->conn_limit);
 721
 722                 DEB(
 723                 if (cmd->conn_limit == IP_FW_TABLEARG)
 724                         printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
 725                             "(tablearg)\n", __func__, conn_limit);
 726                 else
 727                         printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
 728                             __func__, conn_limit);
 729                 )
 730
 731                 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
 732                 id.proto = args->f_id.proto;
 733                 id.addr_type = args->f_id.addr_type;
 734                 id.fib = M_GETFIB(args->m);
 735
 736                 if (IS_IP6_FLOW_ID (&(args->f_id))) {
 737                         if (limit_mask & DYN_SRC_ADDR)
 738                                 id.src_ip6 = args->f_id.src_ip6;
 739                         if (limit_mask & DYN_DST_ADDR)
 740                                 id.dst_ip6 = args->f_id.dst_ip6;
 741                 } else {
 742                         if (limit_mask & DYN_SRC_ADDR)
 743                                 id.src_ip = args->f_id.src_ip;
 744                         if (limit_mask & DYN_DST_ADDR)
 745                                 id.dst_ip = args->f_id.dst_ip;
 746                 }
 747                 if (limit_mask & DYN_SRC_PORT)
 748                         id.src_port = args->f_id.src_port;
 749                 if (limit_mask & DYN_DST_PORT)
 750                         id.dst_port = args->f_id.dst_port;
 751
 752                 /*
 753                  * We have to release lock for previous bucket to
 754                  * avoid possible deadlock
 755                  */
 756                 IPFW_BUCK_UNLOCK(i);
 757
 758                 if ((parent = lookup_dyn_parent(&id, &pindex, rule)) == NULL) {
 759                         printf("ipfw: %s: add parent failed\n", __func__);
 760                         IPFW_BUCK_UNLOCK(pindex);
 761                         return (1);
 762                 }
 763
 764                 if (parent->count >= conn_limit) {
 765                         if (V_fw_verbose && last_log != time_uptime) {
 766                                 last_log = time_uptime;
 767                                 char sbuf[24];
 768                                 last_log = time_uptime;
 769                                 snprintf(sbuf, sizeof(sbuf),
 770                                     "%d drop session",
 771                                     parent->rule->rulenum);
 772                                 print_dyn_rule_flags(&args->f_id,
 773                                     cmd->o.opcode,
 774                                     LOG_SECURITY | LOG_DEBUG,
 775                                     sbuf, "too many entries");
 776                         }
 777                         IPFW_BUCK_UNLOCK(pindex);
 778                         return (1);
 779                 }
 780                 /* Increment counter on parent */
 781                 parent->count++;
 782                 IPFW_BUCK_UNLOCK(pindex);
 783
 784                 IPFW_BUCK_LOCK(i);
 785                 q = add_dyn_rule(&args->f_id, i, O_LIMIT, (struct ip_fw *)parent);
 786                 if (q == NULL) {
 787                         /* Decrement index and notify caller */
 788                         IPFW_BUCK_UNLOCK(i);
 789                         IPFW_BUCK_LOCK(pindex);
 790                         parent->count--;
 791                         IPFW_BUCK_UNLOCK(pindex);
 792                         return (1);
 793                 }
 794                 break;
 795         }
 796         default:
 797                 printf("ipfw: %s: unknown dynamic rule type %u\n",
 798                     __func__, cmd->o.opcode);
 799         }
 800
 801         if (q == NULL) {
 802                 IPFW_BUCK_UNLOCK(i);
 803                 return (1);     /* Notify caller about failure */
 804         }
 805
 806         /* XXX just set lifetime */
 807         lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
 808
 809         IPFW_BUCK_UNLOCK(i);
 810         return (0);
 811 }
 812
 813 /*
 814  * Generate a TCP packet, containing either a RST or a keepalive.
 815  * When flags & TH_RST, we are sending a RST packet, because of a
 816  * "reset" action matched the packet.
 817  * Otherwise we are sending a keepalive, and flags & TH_
 818  * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
 819  * so that MAC can label the reply appropriately.
 820  */
 821 struct mbuf *
 822 ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
 823     u_int32_t ack, int flags)
 824 {
 825         struct mbuf *m = NULL;          /* stupid compiler */
 826         int len, dir;
 827         struct ip *h = NULL;            /* stupid compiler */
 828 #ifdef INET6
 829         struct ip6_hdr *h6 = NULL;
 830 #endif
 831         struct tcphdr *th = NULL;
 832
 833         MGETHDR(m, M_NOWAIT, MT_DATA);
 834         if (m == NULL)
 835                 return (NULL);
 836
 837         M_SETFIB(m, id->fib);
 838 #ifdef MAC
 839         if (replyto != NULL)
 840                 mac_netinet_firewall_reply(replyto, m);
 841         else
 842                 mac_netinet_firewall_send(m);
 843 #else
 844         (void)replyto;          /* don't warn about unused arg */
 845 #endif
 846
 847         switch (id->addr_type) {
 848         case 4:
 849                 len = sizeof(struct ip) + sizeof(struct tcphdr);
 850                 break;
 851 #ifdef INET6
 852         case 6:
 853                 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
 854                 break;
 855 #endif
 856         default:
 857                 /* XXX: log me?!? */
 858                 FREE_PKT(m);
 859                 return (NULL);
 860         }
 861         dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN);
 862
 863         m->m_data += max_linkhdr;
 864         m->m_flags |= M_SKIP_FIREWALL;
 865         m->m_pkthdr.len = m->m_len = len;
 866         m->m_pkthdr.rcvif = NULL;
 867         bzero(m->m_data, len);
 868
 869         switch (id->addr_type) {
 870         case 4:
 871                 h = mtod(m, struct ip *);
 872
 873                 /* prepare for checksum */
 874                 h->ip_p = IPPROTO_TCP;
 875                 h->ip_len = htons(sizeof(struct tcphdr));
 876                 if (dir) {
 877                         h->ip_src.s_addr = htonl(id->src_ip);
 878                         h->ip_dst.s_addr = htonl(id->dst_ip);
 879                 } else {
 880                         h->ip_src.s_addr = htonl(id->dst_ip);
 881                         h->ip_dst.s_addr = htonl(id->src_ip);
 882                 }
 883
 884                 th = (struct tcphdr *)(h + 1);
 885                 break;
 886 #ifdef INET6
 887         case 6:
 888                 h6 = mtod(m, struct ip6_hdr *);
 889
 890                 /* prepare for checksum */
 891                 h6->ip6_nxt = IPPROTO_TCP;
 892                 h6->ip6_plen = htons(sizeof(struct tcphdr));
 893                 if (dir) {
 894                         h6->ip6_src = id->src_ip6;
 895                         h6->ip6_dst = id->dst_ip6;
 896                 } else {
 897                         h6->ip6_src = id->dst_ip6;
 898                         h6->ip6_dst = id->src_ip6;
 899                 }
 900
 901                 th = (struct tcphdr *)(h6 + 1);
 902                 break;
 903 #endif
 904         }
 905
 906         if (dir) {
 907                 th->th_sport = htons(id->src_port);
 908                 th->th_dport = htons(id->dst_port);
 909         } else {
 910                 th->th_sport = htons(id->dst_port);
 911                 th->th_dport = htons(id->src_port);
 912         }
 913         th->th_off = sizeof(struct tcphdr) >> 2;
 914
 915         if (flags & TH_RST) {
 916                 if (flags & TH_ACK) {
 917                         th->th_seq = htonl(ack);
 918                         th->th_flags = TH_RST;
 919                 } else {
 920                         if (flags & TH_SYN)
 921                                 seq++;
 922                         th->th_ack = htonl(seq);
 923                         th->th_flags = TH_RST | TH_ACK;
 924                 }
 925         } else {
 926                 /*
 927                  * Keepalive - use caller provided sequence numbers
 928                  */
 929                 th->th_seq = htonl(seq);
 930                 th->th_ack = htonl(ack);
 931                 th->th_flags = TH_ACK;
 932         }
 933
 934         switch (id->addr_type) {
 935         case 4:
 936                 th->th_sum = in_cksum(m, len);
 937
 938                 /* finish the ip header */
 939                 h->ip_v = 4;
 940                 h->ip_hl = sizeof(*h) >> 2;
 941                 h->ip_tos = IPTOS_LOWDELAY;
 942                 h->ip_off = htons(0);
 943                 h->ip_len = htons(len);
 944                 h->ip_ttl = V_ip_defttl;
 945                 h->ip_sum = 0;
 946                 break;
 947 #ifdef INET6
 948         case 6:
 949                 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6),
 950                     sizeof(struct tcphdr));
 951
 952                 /* finish the ip6 header */
 953                 h6->ip6_vfc |= IPV6_VERSION;
 954                 h6->ip6_hlim = IPV6_DEFHLIM;
 955                 break;
 956 #endif
 957         }
 958
 959         return (m);
 960 }
 961
 962 /*
 963  * Queue keepalive packets for given dynamic rule
 964  */
 965 static struct mbuf **
 966 ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q)
 967 {
 968         struct mbuf *m_rev, *m_fwd;
 969
 970         m_rev = (q->state & ACK_REV) ? NULL :
 971             ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
 972         m_fwd = (q->state & ACK_FWD) ? NULL :
 973             ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0);
 974
 975         if (m_rev != NULL) {
 976                 *mtailp = m_rev;
 977                 mtailp = &(*mtailp)->m_nextpkt;
 978         }
 979         if (m_fwd != NULL) {
 980                 *mtailp = m_fwd;
 981                 mtailp = &(*mtailp)->m_nextpkt;
 982         }
 983
 984         return (mtailp);
 985 }
 986
 987 /*
 988  * This procedure is used to perform various maintance
 989  * on dynamic hash list. Currently it is called every second.
 990  */
 991 static void
 992 ipfw_dyn_tick(void * vnetx)
 993 {
 994         struct ip_fw_chain *chain;
 995         int check_ka = 0;
 996 #ifdef VIMAGE
 997         struct vnet *vp = vnetx;
 998 #endif
 999
1000         CURVNET_SET(vp);
1001
1002         chain = &V_layer3_chain;
1003
1004         /* Run keepalive checks every keepalive_period iff ka is enabled */
1005         if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) &&
1006             (V_dyn_keepalive != 0)) {
1007                 V_dyn_keepalive_last = time_uptime;
1008                 check_ka = 1;
1009         }
1010
1011         check_dyn_rules(chain, NULL, RESVD_SET, check_ka, 1);
1012
1013         callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0);
1014
1015         CURVNET_RESTORE();
1016 }
1017
1018
1019 /*
1020  * Walk thru all dynamic states doing generic maintance:
1021  * 1) free expired states
1022  * 2) free all states based on deleted rule / set
1023  * 3) send keepalives for states if needed
1024  *
1025  * @chain - pointer to current ipfw rules chain
1026  * @rule - delete all states originated by given rule if != NULL
1027  * @set - delete all states originated by any rule in set @set if != RESVD_SET
1028  * @check_ka - perform checking/sending keepalives
1029  * @timer - indicate call from timer routine.
1030  *
1031  * Timer routine must call this function unlocked to permit
1032  * sending keepalives/resizing table.
1033  *
1034  * Others has to call function with IPFW_UH_WLOCK held.
1035  * Additionally, function assume that dynamic rule/set is
1036  * ALREADY deleted so no new states can be generated by
1037  * 'deleted' rules.
1038  *
1039  * Write lock is needed to ensure that unused parent rules
1040  * are not freed by other instance (see stage 2, 3)
1041  */
1042 static void
1043 check_dyn_rules(struct ip_fw_chain *chain, struct ip_fw *rule,
1044     int set, int check_ka, int timer)
1045 {
1046         struct mbuf *m0, *m, *mnext, **mtailp;
1047         struct ip *h;
1048         int i, dyn_count, new_buckets = 0, max_buckets;
1049         int expired = 0, expired_limits = 0, parents = 0, total = 0;
1050         ipfw_dyn_rule *q, *q_prev, *q_next;
1051         ipfw_dyn_rule *exp_head, **exptailp;
1052         ipfw_dyn_rule *exp_lhead, **expltailp;
1053
1054         KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated",
1055             __func__));
1056
1057         /* Avoid possible LOR */
1058         KASSERT(!check_ka || timer, ("%s: keepalive check with lock held",
1059             __func__));
1060
1061         /*
1062          * Do not perform any checks if we currently have no dynamic states
1063          */
1064         if (DYN_COUNT == 0)
1065                 return;
1066
1067         /* Expired states */
1068         exp_head = NULL;
1069         exptailp = &exp_head;
1070
1071         /* Expired limit states */
1072         exp_lhead = NULL;
1073         expltailp = &exp_lhead;
1074
1075         /*
1076          * We make a chain of packets to go out here -- not deferring
1077          * until after we drop the IPFW dynamic rule lock would result
1078          * in a lock order reversal with the normal packet input -> ipfw
1079          * call stack.
1080          */
1081         m0 = NULL;
1082         mtailp = &m0;
1083
1084         /* Protect from hash resizing */
1085         if (timer != 0)
1086                 IPFW_UH_WLOCK(chain);
1087         else
1088                 IPFW_UH_WLOCK_ASSERT(chain);
1089
1090 #define NEXT_RULE()     { q_prev = q; q = q->next ; continue; }
1091
1092         /* Stage 1: perform requested deletion */
1093         for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1094                 IPFW_BUCK_LOCK(i);
1095                 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) {
1096                         /* account every rule */
1097                         total++;
1098
1099                         /* Skip parent rules at all */
1100                         if (q->dyn_type == O_LIMIT_PARENT) {
1101                                 parents++;
1102                                 NEXT_RULE();
1103                         }
1104
1105                         /*
1106                          * Remove rules which are:
1107                          * 1) expired
1108                          * 2) created by given rule
1109                          * 3) created by any rule in given set
1110                          */
1111                         if ((TIME_LEQ(q->expire, time_uptime)) ||
1112                             ((rule != NULL) && (q->rule == rule)) ||
1113                             ((set != RESVD_SET) && (q->rule->set == set))) {
1114                                 if (TIME_LE(time_uptime, q->expire) &&
1115                                     q->dyn_type == O_KEEP_STATE &&
1116                                     V_dyn_keep_states != 0) {
1117                                         /*
1118                                          * Do not delete state if
1119                                          * it is not expired and
1120                                          * dyn_keep_states is ON.
1121                                          * However we need to re-link it
1122                                          * to any other stable rule
1123                                          */
1124                                         q->rule = chain->default_rule;
1125                                         NEXT_RULE();
1126                                 }
1127
1128                                 /* Unlink q from current list */
1129                                 q_next = q->next;
1130                                 if (q == V_ipfw_dyn_v[i].head)
1131                                         V_ipfw_dyn_v[i].head = q_next;
1132                                 else
1133                                         q_prev->next = q_next;
1134
1135                                 q->next = NULL;
1136
1137                                 /* queue q to expire list */
1138                                 if (q->dyn_type != O_LIMIT) {
1139                                         *exptailp = q;
1140                                         exptailp = &(*exptailp)->next;
1141                                         DEB(print_dyn_rule(&q->id, q->dyn_type,
1142                                             "unlink entry", "left");
1143                                         )
1144                                 } else {
1145                                         /* Separate list for limit rules */
1146                                         *expltailp = q;
1147                                         expltailp = &(*expltailp)->next;
1148                                         expired_limits++;
1149                                         DEB(print_dyn_rule(&q->id, q->dyn_type,
1150                                             "unlink limit entry", "left");
1151                                         )
1152                                 }
1153
1154                                 q = q_next;
1155                                 expired++;
1156                                 continue;
1157                         }
1158
1159                         /*
1160                          * Check if we need to send keepalive:
1161                          * we need to ensure if is time to do KA,
1162                          * this is established TCP session, and
1163                          * expire time is within keepalive interval
1164                          */
1165                         if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) &&
1166                             ((q->state & BOTH_SYN) == BOTH_SYN) &&
1167                             (TIME_LEQ(q->expire, time_uptime +
1168                               V_dyn_keepalive_interval)))
1169                                 mtailp = ipfw_dyn_send_ka(mtailp, q);
1170
1171                         NEXT_RULE();
1172                 }
1173                 IPFW_BUCK_UNLOCK(i);
1174         }
1175
1176         /* Stage 2: decrement counters from O_LIMIT parents */
1177         if (expired_limits != 0) {
1178                 /*
1179                  * XXX: Note that deleting set with more than one
1180                  * heavily-used LIMIT rules can result in overwhelming
1181                  * locking due to lack of per-hash value sorting
1182                  *
1183                  * We should probably think about:
1184                  * 1) pre-allocating hash of size, say,
1185                  * MAX(16, V_curr_dyn_buckets / 1024)
1186                  * 2) checking if expired_limits is large enough
1187                  * 3) If yes, init hash (or its part), re-link
1188                  * current list and start decrementing procedure in
1189                  * each bucket separately
1190                  */
1191
1192                 /*
1193                  * Small optimization: do not unlock bucket until
1194                  * we see the next item resides in different bucket
1195                  */
1196                 if (exp_lhead != NULL) {
1197                         i = exp_lhead->parent->bucket;
1198                         IPFW_BUCK_LOCK(i);
1199                 }
1200                 for (q = exp_lhead; q != NULL; q = q->next) {
1201                         if (i != q->parent->bucket) {
1202                                 IPFW_BUCK_UNLOCK(i);
1203                                 i = q->parent->bucket;
1204                                 IPFW_BUCK_LOCK(i);
1205                         }
1206
1207                         /* Decrease parent refcount */
1208                         q->parent->count--;
1209                 }
1210                 if (exp_lhead != NULL)
1211                         IPFW_BUCK_UNLOCK(i);
1212         }
1213
1214         /*
1215          * We protectet ourselves from unused parent deletion
1216          * (from the timer function) by holding UH write lock.
1217          */
1218
1219         /* Stage 3: remove unused parent rules */
1220         if ((parents != 0) && (expired != 0)) {
1221                 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1222                         IPFW_BUCK_LOCK(i);
1223                         for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) {
1224                                 if (q->dyn_type != O_LIMIT_PARENT)
1225                                         NEXT_RULE();
1226
1227                                 if (q->count != 0)
1228                                         NEXT_RULE();
1229
1230                                 /* Parent rule without consumers */
1231
1232                                 /* Unlink q from current list */
1233                                 q_next = q->next;
1234                                 if (q == V_ipfw_dyn_v[i].head)
1235                                         V_ipfw_dyn_v[i].head = q_next;
1236                                 else
1237                                         q_prev->next = q_next;
1238
1239                                 q->next = NULL;
1240
1241                                 /* Add to expired list */
1242                                 *exptailp = q;
1243                                 exptailp = &(*exptailp)->next;
1244
1245                                 DEB(print_dyn_rule(&q->id, q->dyn_type,
1246                                     "unlink parent entry", "left");
1247                                 )
1248
1249                                 expired++;
1250
1251                                 q = q_next;
1252                         }
1253                         IPFW_BUCK_UNLOCK(i);
1254                 }
1255         }
1256
1257 #undef NEXT_RULE
1258
1259         if (timer != 0) {
1260                 /*
1261                  * Check if we need to resize hash:
1262                  * if current number of states exceeds number of buckes in hash,
1263                  * grow hash size to the minimum power of 2 which is bigger than
1264                  * current states count. Limit hash size by 64k.
1265                  */
1266                 max_buckets = (V_dyn_buckets_max > 65536) ?
1267                     65536 : V_dyn_buckets_max;
1268
1269                 dyn_count = DYN_COUNT;
1270
1271                 if ((dyn_count > V_curr_dyn_buckets * 2) &&
1272                     (dyn_count < max_buckets)) {
1273                         new_buckets = V_curr_dyn_buckets;
1274                         while (new_buckets < dyn_count) {
1275                                 new_buckets *= 2;
1276
1277                                 if (new_buckets >= max_buckets)
1278                                         break;
1279                         }
1280                 }
1281
1282                 IPFW_UH_WUNLOCK(chain);
1283         }
1284
1285         /* Finally delete old states ad limits if any */
1286         for (q = exp_head; q != NULL; q = q_next) {
1287                 q_next = q->next;
1288                 uma_zfree(V_ipfw_dyn_rule_zone, q);
1289                 ipfw_dyn_count--;
1290         }
1291
1292         for (q = exp_lhead; q != NULL; q = q_next) {
1293                 q_next = q->next;
1294                 uma_zfree(V_ipfw_dyn_rule_zone, q);
1295                 ipfw_dyn_count--;
1296         }
1297
1298         /*
1299          * The rest code MUST be called from timer routine only
1300          * without holding any locks
1301          */
1302         if (timer == 0)
1303                 return;
1304
1305         /* Send keepalive packets if any */
1306         for (m = m0; m != NULL; m = mnext) {
1307                 mnext = m->m_nextpkt;
1308                 m->m_nextpkt = NULL;
1309                 h = mtod(m, struct ip *);
1310                 if (h->ip_v == 4)
1311                         ip_output(m, NULL, NULL, 0, NULL, NULL);
1312 #ifdef INET6
1313                 else
1314                         ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1315 #endif
1316         }
1317
1318         /* Run table resize without holding any locks */
1319         if (new_buckets != 0)
1320                 resize_dynamic_table(chain, new_buckets);
1321 }
1322
1323 /*
1324  * Deletes all dynamic rules originated by given rule or all rules in
1325  * given set. Specify RESVD_SET to indicate set should not be used.
1326  * @chain - pointer to current ipfw rules chain
1327  * @rule - delete all states originated by given rule if != NULL
1328  * @set - delete all states originated by any rule in set @set if != RESVD_SET
1329  *
1330  * Function has to be called with IPFW_UH_WLOCK held.
1331  * Additionally, function assume that dynamic rule/set is
1332  * ALREADY deleted so no new states can be generated by
1333  * 'deleted' rules.
1334  */
1335 void
1336 ipfw_expire_dyn_rules(struct ip_fw_chain *chain, struct ip_fw *rule, int set)
1337 {
1338
1339         check_dyn_rules(chain, rule, set, 0, 0);
1340 }
1341
1342 void
1343 ipfw_dyn_init(struct ip_fw_chain *chain)
1344 {
1345
1346         V_ipfw_dyn_v = NULL;
1347         V_dyn_buckets_max = 256; /* must be power of 2 */
1348         V_curr_dyn_buckets = 256; /* must be power of 2 */
1349
1350         V_dyn_ack_lifetime = 300;
1351         V_dyn_syn_lifetime = 20;
1352         V_dyn_fin_lifetime = 1;
1353         V_dyn_rst_lifetime = 1;
1354         V_dyn_udp_lifetime = 10;
1355         V_dyn_short_lifetime = 5;
1356
1357         V_dyn_keepalive_interval = 20;
1358         V_dyn_keepalive_period = 5;
1359         V_dyn_keepalive = 1;    /* do send keepalives */
1360         V_dyn_keepalive_last = time_uptime;
1361
1362         V_dyn_max = 4096;       /* max # of dynamic rules */
1363
1364         V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule",
1365             sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
1366             UMA_ALIGN_PTR, 0);
1367
1368         /* Enforce limit on dynamic rules */
1369         uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1370
1371         callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE);
1372
1373         /*
1374          * This can potentially be done on first dynamic rule
1375          * being added to chain.
1376          */
1377         resize_dynamic_table(chain, V_curr_dyn_buckets);
1378 }
1379
1380 void
1381 ipfw_dyn_uninit(int pass)
1382 {
1383         int i;
1384
1385         if (pass == 0) {
1386                 callout_drain(&V_ipfw_timeout);
1387                 return;
1388         }
1389
1390         if (V_ipfw_dyn_v != NULL) {
1391                 /*
1392                  * Skip deleting all dynamic states -
1393                  * uma_zdestroy() does this more efficiently;
1394                  */
1395
1396                 /* Destroy all mutexes */
1397                 for (i = 0 ; i < V_curr_dyn_buckets ; i++)
1398                         IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]);
1399                 free(V_ipfw_dyn_v, M_IPFW);
1400                 V_ipfw_dyn_v = NULL;
1401         }
1402
1403         uma_zdestroy(V_ipfw_dyn_rule_zone);
1404 }
1405
1406 #ifdef SYSCTL_NODE
1407 /*
1408  * Get/set maximum number of dynamic states in given VNET instance.
1409  */
1410 static int
1411 sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS)
1412 {
1413         int error;
1414         unsigned int nstates;
1415
1416         nstates = V_dyn_max;
1417
1418         error = sysctl_handle_int(oidp, &nstates, 0, req);
1419         /* Read operation or some error */
1420         if ((error != 0) || (req->newptr == NULL))
1421                 return (error);
1422
1423         V_dyn_max = nstates;
1424         uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1425
1426         return (0);
1427 }
1428
1429 /*
1430  * Get current number of dynamic states in given VNET instance.
1431  */
1432 static int
1433 sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS)
1434 {
1435         int error;
1436         unsigned int nstates;
1437
1438         nstates = DYN_COUNT;
1439
1440         error = sysctl_handle_int(oidp, &nstates, 0, req);
1441
1442         return (error);
1443 }
1444 #endif
1445
1446 /*
1447  * Returns size of dynamic states in legacy format
1448  */
1449 int
1450 ipfw_dyn_len(void)
1451 {
1452
1453         return (V_ipfw_dyn_v == NULL) ? 0 :
1454                 (DYN_COUNT * sizeof(ipfw_dyn_rule));
1455 }
1456
1457 /*
1458  * Returns number of dynamic states.
1459  * Used by dump format v1 (current).
1460  */
1461 int
1462 ipfw_dyn_get_count(void)
1463 {
1464
1465         return (V_ipfw_dyn_v == NULL) ? 0 : DYN_COUNT;
1466 }
1467
1468 static void
1469 export_dyn_rule(ipfw_dyn_rule *src, ipfw_dyn_rule *dst)
1470 {
1471
1472         memcpy(dst, src, sizeof(*src));
1473         memcpy(&(dst->rule), &(src->rule->rulenum), sizeof(src->rule->rulenum));
1474         /*
1475          * store set number into high word of
1476          * dst->rule pointer.
1477          */
1478         memcpy((char *)&dst->rule + sizeof(src->rule->rulenum),
1479             &(src->rule->set), sizeof(src->rule->set));
1480         /*
1481          * store a non-null value in "next".
1482          * The userland code will interpret a
1483          * NULL here as a marker
1484          * for the last dynamic rule.
1485          */
1486         memcpy(&dst->next, &dst, sizeof(dst));
1487         dst->expire =
1488             TIME_LEQ(dst->expire, time_uptime) ?  0 : dst->expire - time_uptime;
1489 }
1490
1491 /*
1492  * Fills int buffer given by @sd with dynamic states.
1493  * Used by dump format v1 (current).
1494  *
1495  * Returns 0 on success.
1496  */
1497 int
1498 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd)
1499 {
1500         ipfw_dyn_rule *p;
1501         ipfw_obj_dyntlv *dst, *last;
1502         ipfw_obj_ctlv *ctlv;
1503         int i;
1504         size_t sz;
1505
1506         if (V_ipfw_dyn_v == NULL)
1507                 return (0);
1508
1509         IPFW_UH_RLOCK_ASSERT(chain);
1510
1511         ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv));
1512         if (ctlv == NULL)
1513                 return (ENOMEM);
1514         sz = sizeof(ipfw_obj_dyntlv);
1515         ctlv->head.type = IPFW_TLV_DYNSTATE_LIST;
1516         ctlv->objsize = sz;
1517         last = NULL;
1518
1519         for (i = 0 ; i < V_curr_dyn_buckets; i++) {
1520                 IPFW_BUCK_LOCK(i);
1521                 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) {
1522                         dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, sz);
1523                         if (dst == NULL) {
1524                                 IPFW_BUCK_UNLOCK(i);
1525                                 return (ENOMEM);
1526                         }
1527
1528                         export_dyn_rule(p, &dst->state);
1529                         dst->head.length = sz;
1530                         dst->head.type = IPFW_TLV_DYN_ENT;
1531                         last = dst;
1532                 }
1533                 IPFW_BUCK_UNLOCK(i);
1534         }
1535
1536         if (last != NULL) /* mark last dynamic rule */
1537                 last->head.flags = IPFW_DF_LAST;
1538
1539         return (0);
1540 }
1541
1542 /*
1543  * Fill given buffer with dynamic states (legacy format).
1544  * IPFW_UH_RLOCK has to be held while calling.
1545  */
1546 void
1547 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep)
1548 {
1549         ipfw_dyn_rule *p, *last = NULL;
1550         char *bp;
1551         int i;
1552
1553         if (V_ipfw_dyn_v == NULL)
1554                 return;
1555         bp = *pbp;
1556
1557         IPFW_UH_RLOCK_ASSERT(chain);
1558
1559         for (i = 0 ; i < V_curr_dyn_buckets; i++) {
1560                 IPFW_BUCK_LOCK(i);
1561                 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) {
1562                         if (bp + sizeof *p <= ep) {
1563                                 ipfw_dyn_rule *dst =
1564                                         (ipfw_dyn_rule *)bp;
1565
1566                                 export_dyn_rule(p, dst);
1567                                 last = dst;
1568                                 bp += sizeof(ipfw_dyn_rule);
1569                         }
1570                 }
1571                 IPFW_BUCK_UNLOCK(i);
1572         }
1573
1574         if (last != NULL) /* mark last dynamic rule */
1575                 bzero(&last->next, sizeof(last));
1576         *pbp = bp;
1577 }
1578 /* end of file */