sys/net/ipfw/ip_fw2.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  * $FreeBSD: src/sys/netinet/ip_fw2.c,v 1.6.2.12 2003/04/08 10:42:32 maxim Exp $
  26  * $DragonFly: src/sys/net/ipfw/ip_fw2.c,v 1.57 2008/07/31 12:35:15 sephe Exp $
  27  */
  28
  29 #define        DEB(x)
  30 #define        DDB(x) x
  31
  32 /*
  33  * Implement IP packet firewall (new version)
  34  */
  35
  36 #ifndef KLD_MODULE
  37 #include "opt_ipfw.h"
  38 #include "opt_ipdn.h"
  39 #include "opt_ipdivert.h"
  40 #include "opt_inet.h"
  41 #ifndef INET
  42 #error IPFIREWALL requires INET.
  43 #endif /* INET */
  44 #endif
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/kernel.h>
  51 #include <sys/proc.h>
  52 #include <sys/socket.h>
  53 #include <sys/socketvar.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/syslog.h>
  56 #include <sys/thread2.h>
  57 #include <sys/ucred.h>
  58 #include <sys/in_cksum.h>
  59
  60 #include <net/if.h>
  61 #include <net/route.h>
  62 #include <net/netmsg2.h>
  63
  64 #include <netinet/in.h>
  65 #include <netinet/in_systm.h>
  66 #include <netinet/in_var.h>
  67 #include <netinet/in_pcb.h>
  68 #include <netinet/ip.h>
  69 #include <netinet/ip_var.h>
  70 #include <netinet/ip_icmp.h>
  71 #include "ip_fw.h"
  72 #include <net/dummynet/ip_dummynet.h>
  73 #include <netinet/tcp.h>
  74 #include <netinet/tcp_timer.h>
  75 #include <netinet/tcp_var.h>
  76 #include <netinet/tcpip.h>
  77 #include <netinet/udp.h>
  78 #include <netinet/udp_var.h>
  79
  80 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  81
  82 /*
  83  * set_disable contains one bit per set value (0..31).
  84  * If the bit is set, all rules with the corresponding set
  85  * are disabled. Set 31 is reserved for the default rule
  86  * and CANNOT be disabled.
  87  */
  88 static uint32_t set_disable;
  89
  90 static int fw_verbose;
  91 static int verbose_limit;
  92
  93 #ifdef KLD_MODULE
  94 static int ipfw_refcnt;
  95 #endif
  96
  97 static struct callout ipfw_timeout_h;
  98 #define IPFW_DEFAULT_RULE       65535
  99
 100 /*
 101  * list of rules for layer 3
 102  */
 103 static struct ip_fw *layer3_chain;
 104
 105 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 106
 107 static int fw_debug = 1;
 108 static int autoinc_step = 100; /* bounded to 1..1000 in ipfw_add_rule() */
 109
 110 #ifdef SYSCTL_NODE
 111 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 112 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
 113     &fw_enable, 0, "Enable ipfw");
 114 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
 115     &autoinc_step, 0, "Rule number autincrement step");
 116 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 117     &fw_one_pass, 0,
 118     "Only do a single pass through ipfw when using dummynet(4)");
 119 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 120     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 121 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 122     &fw_verbose, 0, "Log matches to ipfw rules");
 123 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 124     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 125
 126 /*
 127  * Description of dynamic rules.
 128  *
 129  * Dynamic rules are stored in lists accessed through a hash table
 130  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 131  * be modified through the sysctl variable dyn_buckets which is
 132  * updated when the table becomes empty.
 133  *
 134  * XXX currently there is only one list, ipfw_dyn.
 135  *
 136  * When a packet is received, its address fields are first masked
 137  * with the mask defined for the rule, then hashed, then matched
 138  * against the entries in the corresponding list.
 139  * Dynamic rules can be used for different purposes:
 140  *  + stateful rules;
 141  *  + enforcing limits on the number of sessions;
 142  *  + in-kernel NAT (not implemented yet)
 143  *
 144  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 145  * measured in seconds and depending on the flags.
 146  *
 147  * The total number of dynamic rules is stored in dyn_count.
 148  * The max number of dynamic rules is dyn_max. When we reach
 149  * the maximum number of rules we do not create anymore. This is
 150  * done to avoid consuming too much memory, but also too much
 151  * time when searching on each packet (ideally, we should try instead
 152  * to put a limit on the length of the list on each bucket...).
 153  *
 154  * Each dynamic rule holds a pointer to the parent ipfw rule so
 155  * we know what action to perform. Dynamic rules are removed when
 156  * the parent rule is deleted. XXX we should make them survive.
 157  *
 158  * There are some limitations with dynamic rules -- we do not
 159  * obey the 'randomized match', and we do not do multiple
 160  * passes through the firewall. XXX check the latter!!!
 161  */
 162 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 163 static uint32_t dyn_buckets = 256; /* must be power of 2 */
 164 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
 165
 166 /*
 167  * Timeouts for various events in handing dynamic rules.
 168  */
 169 static uint32_t dyn_ack_lifetime = 300;
 170 static uint32_t dyn_syn_lifetime = 20;
 171 static uint32_t dyn_fin_lifetime = 1;
 172 static uint32_t dyn_rst_lifetime = 1;
 173 static uint32_t dyn_udp_lifetime = 10;
 174 static uint32_t dyn_short_lifetime = 5;
 175
 176 /*
 177  * Keepalives are sent if dyn_keepalive is set. They are sent every
 178  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 179  * seconds of lifetime of a rule.
 180  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 181  * than dyn_keepalive_period.
 182  */
 183
 184 static uint32_t dyn_keepalive_interval = 20;
 185 static uint32_t dyn_keepalive_period = 5;
 186 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
 187
 188 static uint32_t static_count;           /* # of static rules */
 189 static uint32_t static_ioc_len; /* bytes of static rules */
 190 static uint32_t dyn_count;              /* # of dynamic rules */
 191 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
 192
 193 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
 194     &dyn_buckets, 0, "Number of dyn. buckets");
 195 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 196     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 197 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 198     &dyn_count, 0, "Number of dyn. rules");
 199 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 200     &dyn_max, 0, "Max number of dyn. rules");
 201 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 202     &static_count, 0, "Number of static rules");
 203 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 204     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 205 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 206     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 207 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
 208     &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
 209 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
 210     &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
 211 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 212     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 213 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 214     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 215 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 216     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 217
 218 #endif /* SYSCTL_NODE */
 219
 220 static struct ip_fw *ip_fw_default_rule;
 221
 222 static ip_fw_chk_t      ipfw_chk;
 223
 224 static __inline int
 225 ipfw_free_rule(struct ip_fw *rule)
 226 {
 227         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
 228         atomic_subtract_int(&rule->refcnt, 1);
 229         if (atomic_cmpset_int(&rule->refcnt, 0, 1)) {
 230                 kfree(rule, M_IPFW);
 231                 return 1;
 232         }
 233         return 0;
 234 }
 235
 236 static void
 237 ipfw_unref_rule(void *priv)
 238 {
 239         ipfw_free_rule(priv);
 240 #ifdef KLD_MODULE
 241         atomic_subtract_int(&ipfw_refcnt, 1);
 242 #endif
 243 }
 244
 245 static __inline void
 246 ipfw_ref_rule(struct ip_fw *rule)
 247 {
 248 #ifdef KLD_MODULE
 249         atomic_add_int(&ipfw_refcnt, 1);
 250 #endif
 251         atomic_add_int(&rule->refcnt, 1);
 252 }
 253
 254 /*
 255  * This macro maps an ip pointer into a layer3 header pointer of type T
 256  */
 257 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
 258
 259 static __inline int
 260 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 261 {
 262         int type = L3HDR(struct icmp,ip)->icmp_type;
 263
 264         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
 265 }
 266
 267 #define TT      ((1 << ICMP_ECHO) | \
 268                  (1 << ICMP_ROUTERSOLICIT) | \
 269                  (1 << ICMP_TSTAMP) | \
 270                  (1 << ICMP_IREQ) | \
 271                  (1 << ICMP_MASKREQ))
 272
 273 static int
 274 is_icmp_query(struct ip *ip)
 275 {
 276         int type = L3HDR(struct icmp, ip)->icmp_type;
 277
 278         return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
 279 }
 280
 281 #undef TT
 282
 283 /*
 284  * The following checks use two arrays of 8 or 16 bits to store the
 285  * bits that we want set or clear, respectively. They are in the
 286  * low and high half of cmd->arg1 or cmd->d[0].
 287  *
 288  * We scan options and store the bits we find set. We succeed if
 289  *
 290  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 291  *
 292  * The code is sometimes optimized not to store additional variables.
 293  */
 294
 295 static int
 296 flags_match(ipfw_insn *cmd, uint8_t bits)
 297 {
 298         u_char want_clear;
 299         bits = ~bits;
 300
 301         if (((cmd->arg1 & 0xff) & bits) != 0)
 302                 return 0; /* some bits we want set were clear */
 303
 304         want_clear = (cmd->arg1 >> 8) & 0xff;
 305         if ((want_clear & bits) != want_clear)
 306                 return 0; /* some bits we want clear were set */
 307         return 1;
 308 }
 309
 310 static int
 311 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 312 {
 313         int optlen, bits = 0;
 314         u_char *cp = (u_char *)(ip + 1);
 315         int x = (ip->ip_hl << 2) - sizeof(struct ip);
 316
 317         for (; x > 0; x -= optlen, cp += optlen) {
 318                 int opt = cp[IPOPT_OPTVAL];
 319
 320                 if (opt == IPOPT_EOL)
 321                         break;
 322
 323                 if (opt == IPOPT_NOP) {
 324                         optlen = 1;
 325                 } else {
 326                         optlen = cp[IPOPT_OLEN];
 327                         if (optlen <= 0 || optlen > x)
 328                                 return 0; /* invalid or truncated */
 329                 }
 330
 331                 switch (opt) {
 332                 case IPOPT_LSRR:
 333                         bits |= IP_FW_IPOPT_LSRR;
 334                         break;
 335
 336                 case IPOPT_SSRR:
 337                         bits |= IP_FW_IPOPT_SSRR;
 338                         break;
 339
 340                 case IPOPT_RR:
 341                         bits |= IP_FW_IPOPT_RR;
 342                         break;
 343
 344                 case IPOPT_TS:
 345                         bits |= IP_FW_IPOPT_TS;
 346                         break;
 347
 348                 default:
 349                         break;
 350                 }
 351         }
 352         return (flags_match(cmd, bits));
 353 }
 354
 355 static int
 356 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 357 {
 358         int optlen, bits = 0;
 359         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 360         u_char *cp = (u_char *)(tcp + 1);
 361         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 362
 363         for (; x > 0; x -= optlen, cp += optlen) {
 364                 int opt = cp[0];
 365
 366                 if (opt == TCPOPT_EOL)
 367                         break;
 368
 369                 if (opt == TCPOPT_NOP) {
 370                         optlen = 1;
 371                 } else {
 372                         optlen = cp[1];
 373                         if (optlen <= 0)
 374                                 break;
 375                 }
 376
 377                 switch (opt) {
 378                 case TCPOPT_MAXSEG:
 379                         bits |= IP_FW_TCPOPT_MSS;
 380                         break;
 381
 382                 case TCPOPT_WINDOW:
 383                         bits |= IP_FW_TCPOPT_WINDOW;
 384                         break;
 385
 386                 case TCPOPT_SACK_PERMITTED:
 387                 case TCPOPT_SACK:
 388                         bits |= IP_FW_TCPOPT_SACK;
 389                         break;
 390
 391                 case TCPOPT_TIMESTAMP:
 392                         bits |= IP_FW_TCPOPT_TS;
 393                         break;
 394
 395                 case TCPOPT_CC:
 396                 case TCPOPT_CCNEW:
 397                 case TCPOPT_CCECHO:
 398                         bits |= IP_FW_TCPOPT_CC;
 399                         break;
 400
 401                 default:
 402                         break;
 403                 }
 404         }
 405         return (flags_match(cmd, bits));
 406 }
 407
 408 static int
 409 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 410 {
 411         if (ifp == NULL)        /* no iface with this packet, match fails */
 412                 return 0;
 413
 414         /* Check by name or by IP address */
 415         if (cmd->name[0] != '\0') { /* match by name */
 416                 /* Check name */
 417                 if (cmd->p.glob) {
 418                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 419                                 return(1);
 420                 } else {
 421                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 422                                 return(1);
 423                 }
 424         } else {
 425                 struct ifaddr_container *ifac;
 426
 427                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 428                         struct ifaddr *ia = ifac->ifa;
 429
 430                         if (ia->ifa_addr == NULL)
 431                                 continue;
 432                         if (ia->ifa_addr->sa_family != AF_INET)
 433                                 continue;
 434                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 435                             (ia->ifa_addr))->sin_addr.s_addr)
 436                                 return(1);      /* match */
 437                 }
 438         }
 439         return(0);      /* no match, fail ... */
 440 }
 441
 442 static uint64_t norule_counter; /* counter for ipfw_log(NULL...) */
 443
 444 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 445
 446 /*
 447  * We enter here when we have a rule with O_LOG.
 448  * XXX this function alone takes about 2Kbytes of code!
 449  */
 450 static void
 451 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 452          struct mbuf *m, struct ifnet *oif)
 453 {
 454         char *action;
 455         int limit_reached = 0;
 456         char action2[40], proto[48], fragment[28];
 457
 458         fragment[0] = '\0';
 459         proto[0] = '\0';
 460
 461         if (f == NULL) {        /* bogus pkt */
 462                 if (verbose_limit != 0 && norule_counter >= verbose_limit)
 463                         return;
 464                 norule_counter++;
 465                 if (norule_counter == verbose_limit)
 466                         limit_reached = verbose_limit;
 467                 action = "Refuse";
 468         } else {        /* O_LOG is the first action, find the real one */
 469                 ipfw_insn *cmd = ACTION_PTR(f);
 470                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 471
 472                 if (l->max_log != 0 && l->log_left == 0)
 473                         return;
 474                 l->log_left--;
 475                 if (l->log_left == 0)
 476                         limit_reached = l->max_log;
 477                 cmd += F_LEN(cmd);      /* point to first action */
 478                 if (cmd->opcode == O_PROB)
 479                         cmd += F_LEN(cmd);
 480
 481                 action = action2;
 482                 switch (cmd->opcode) {
 483                 case O_DENY:
 484                         action = "Deny";
 485                         break;
 486
 487                 case O_REJECT:
 488                         if (cmd->arg1==ICMP_REJECT_RST) {
 489                                 action = "Reset";
 490                         } else if (cmd->arg1==ICMP_UNREACH_HOST) {
 491                                 action = "Reject";
 492                         } else {
 493                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 494                                           cmd->arg1);
 495                         }
 496                         break;
 497
 498                 case O_ACCEPT:
 499                         action = "Accept";
 500                         break;
 501
 502                 case O_COUNT:
 503                         action = "Count";
 504                         break;
 505
 506                 case O_DIVERT:
 507                         ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
 508                         break;
 509
 510                 case O_TEE:
 511                         ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
 512                         break;
 513
 514                 case O_SKIPTO:
 515                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
 516                         break;
 517
 518                 case O_PIPE:
 519                         ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
 520                         break;
 521
 522                 case O_QUEUE:
 523                         ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
 524                         break;
 525
 526                 case O_FORWARD_IP:
 527                         {
 528                                 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 529                                 int len;
 530
 531                                 len = ksnprintf(SNPARGS(action2, 0),
 532                                                 "Forward to %s",
 533                                                 inet_ntoa(sa->sa.sin_addr));
 534                                 if (sa->sa.sin_port) {
 535                                         ksnprintf(SNPARGS(action2, len), ":%d",
 536                                                   sa->sa.sin_port);
 537                                 }
 538                         }
 539                         break;
 540
 541                 default:
 542                         action = "UNKNOWN";
 543                         break;
 544                 }
 545         }
 546
 547         if (hlen == 0) {        /* non-ip */
 548                 ksnprintf(SNPARGS(proto, 0), "MAC");
 549         } else {
 550                 struct ip *ip = mtod(m, struct ip *);
 551                 /* these three are all aliases to the same thing */
 552                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 553                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 554                 struct udphdr *const udp = (struct udphdr *)icmp;
 555
 556                 int ip_off, offset, ip_len;
 557                 int len;
 558
 559                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 560                         ip_off = ntohs(ip->ip_off);
 561                         ip_len = ntohs(ip->ip_len);
 562                 } else {
 563                         ip_off = ip->ip_off;
 564                         ip_len = ip->ip_len;
 565                 }
 566                 offset = ip_off & IP_OFFMASK;
 567                 switch (ip->ip_p) {
 568                 case IPPROTO_TCP:
 569                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 570                                         inet_ntoa(ip->ip_src));
 571                         if (offset == 0) {
 572                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 573                                           ntohs(tcp->th_sport),
 574                                           inet_ntoa(ip->ip_dst),
 575                                           ntohs(tcp->th_dport));
 576                         } else {
 577                                 ksnprintf(SNPARGS(proto, len), " %s",
 578                                           inet_ntoa(ip->ip_dst));
 579                         }
 580                         break;
 581
 582                 case IPPROTO_UDP:
 583                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 584                                         inet_ntoa(ip->ip_src));
 585                         if (offset == 0) {
 586                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 587                                           ntohs(udp->uh_sport),
 588                                           inet_ntoa(ip->ip_dst),
 589                                           ntohs(udp->uh_dport));
 590                         } else {
 591                                 ksnprintf(SNPARGS(proto, len), " %s",
 592                                           inet_ntoa(ip->ip_dst));
 593                         }
 594                         break;
 595
 596                 case IPPROTO_ICMP:
 597                         if (offset == 0) {
 598                                 len = ksnprintf(SNPARGS(proto, 0),
 599                                                 "ICMP:%u.%u ",
 600                                                 icmp->icmp_type,
 601                                                 icmp->icmp_code);
 602                         } else {
 603                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 604                         }
 605                         len += ksnprintf(SNPARGS(proto, len), "%s",
 606                                          inet_ntoa(ip->ip_src));
 607                         ksnprintf(SNPARGS(proto, len), " %s",
 608                                   inet_ntoa(ip->ip_dst));
 609                         break;
 610
 611                 default:
 612                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 613                                         inet_ntoa(ip->ip_src));
 614                         ksnprintf(SNPARGS(proto, len), " %s",
 615                                   inet_ntoa(ip->ip_dst));
 616                         break;
 617                 }
 618
 619                 if (ip_off & (IP_MF | IP_OFFMASK)) {
 620                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 621                                   ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 622                                   offset << 3, (ip_off & IP_MF) ? "+" : "");
 623                 }
 624         }
 625
 626         if (oif || m->m_pkthdr.rcvif) {
 627                 log(LOG_SECURITY | LOG_INFO,
 628                     "ipfw: %d %s %s %s via %s%s\n",
 629                     f ? f->rulenum : -1,
 630                     action, proto, oif ? "out" : "in",
 631                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 632                     fragment);
 633         } else {
 634                 log(LOG_SECURITY | LOG_INFO,
 635                     "ipfw: %d %s %s [no if info]%s\n",
 636                     f ? f->rulenum : -1,
 637                     action, proto, fragment);
 638         }
 639
 640         if (limit_reached) {
 641                 log(LOG_SECURITY | LOG_NOTICE,
 642                     "ipfw: limit %d reached on entry %d\n",
 643                     limit_reached, f ? f->rulenum : -1);
 644         }
 645 }
 646
 647 #undef SNPARGS
 648
 649 /*
 650  * IMPORTANT: the hash function for dynamic rules must be commutative
 651  * in source and destination (ip,port), because rules are bidirectional
 652  * and we want to find both in the same bucket.
 653  */
 654 static __inline int
 655 hash_packet(struct ipfw_flow_id *id)
 656 {
 657         uint32_t i;
 658
 659         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 660         i &= (curr_dyn_buckets - 1);
 661         return i;
 662 }
 663
 664 /**
 665  * unlink a dynamic rule from a chain. prev is a pointer to
 666  * the previous one, q is a pointer to the rule to delete,
 667  * head is a pointer to the head of the queue.
 668  * Modifies q and potentially also head.
 669  */
 670 #define UNLINK_DYN_RULE(prev, head, q)                                  \
 671 do {                                                                    \
 672         ipfw_dyn_rule *old_q = q;                                       \
 673                                                                         \
 674         /* remove a refcount to the parent */                           \
 675         if (q->dyn_type == O_LIMIT)                                     \
 676                 q->parent->count--;                                     \
 677         DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \
 678                 (q->id.src_ip), (q->id.src_port),                       \
 679                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
 680         if (prev != NULL)                                               \
 681                 prev->next = q = q->next;                               \
 682         else                                                            \
 683                 head = q = q->next;                                     \
 684         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
 685         dyn_count--;                                                    \
 686         kfree(old_q, M_IPFW);                                           \
 687 } while (0)
 688
 689 #define TIME_LEQ(a, b)  ((int)((a) - (b)) <= 0)
 690
 691 /**
 692  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 693  *
 694  * If keep_me == NULL, rules are deleted even if not expired,
 695  * otherwise only expired rules are removed.
 696  *
 697  * The value of the second parameter is also used to point to identify
 698  * a rule we absolutely do not want to remove (e.g. because we are
 699  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 700  * rules). The pointer is only used for comparison, so any non-null
 701  * value will do.
 702  */
 703 static void
 704 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 705 {
 706         static uint32_t last_remove = 0;
 707
 708 #define FORCE   (keep_me == NULL)
 709
 710         ipfw_dyn_rule *prev, *q;
 711         int i, pass = 0, max_pass = 0;
 712
 713         if (ipfw_dyn_v == NULL || dyn_count == 0)
 714                 return;
 715         /* do not expire more than once per second, it is useless */
 716         if (!FORCE && last_remove == time_second)
 717                 return;
 718         last_remove = time_second;
 719
 720         /*
 721          * because O_LIMIT refer to parent rules, during the first pass only
 722          * remove child and mark any pending LIMIT_PARENT, and remove
 723          * them in a second pass.
 724          */
 725 next_pass:
 726         for (i = 0; i < curr_dyn_buckets; i++) {
 727                 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
 728                         /*
 729                          * Logic can become complex here, so we split tests.
 730                          */
 731                         if (q == keep_me)
 732                                 goto next;
 733                         if (rule != NULL && rule != q->rule)
 734                                 goto next; /* not the one we are looking for */
 735                         if (q->dyn_type == O_LIMIT_PARENT) {
 736                                 /*
 737                                  * handle parent in the second pass,
 738                                  * record we need one.
 739                                  */
 740                                 max_pass = 1;
 741                                 if (pass == 0)
 742                                         goto next;
 743                                 if (FORCE && q->count != 0) {
 744                                         /* XXX should not happen! */
 745                                         kprintf("OUCH! cannot remove rule, "
 746                                                 "count %d\n", q->count);
 747                                 }
 748                         } else {
 749                                 if (!FORCE && !TIME_LEQ(q->expire, time_second))
 750                                         goto next;
 751                         }
 752                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 753                         continue;
 754 next:
 755                         prev = q;
 756                         q = q->next;
 757                 }
 758         }
 759         if (pass++ < max_pass)
 760                 goto next_pass;
 761
 762 #undef FORCE
 763 }
 764
 765
 766 /**
 767  * lookup a dynamic rule.
 768  */
 769 static ipfw_dyn_rule *
 770 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 771                 struct tcphdr *tcp)
 772 {
 773         /*
 774          * stateful ipfw extensions.
 775          * Lookup into dynamic session queue
 776          */
 777 #define MATCH_REVERSE   0
 778 #define MATCH_FORWARD   1
 779 #define MATCH_NONE      2
 780 #define MATCH_UNKNOWN   3
 781         int i, dir = MATCH_NONE;
 782         ipfw_dyn_rule *prev, *q=NULL;
 783
 784         if (ipfw_dyn_v == NULL)
 785                 goto done;      /* not found */
 786
 787         i = hash_packet(pkt);
 788         for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
 789                 if (q->dyn_type == O_LIMIT_PARENT)
 790                         goto next;
 791
 792                 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
 793                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 794                         continue;
 795                 }
 796                 if (pkt->proto == q->id.proto) {
 797                         if (pkt->src_ip == q->id.src_ip &&
 798                             pkt->dst_ip == q->id.dst_ip &&
 799                             pkt->src_port == q->id.src_port &&
 800                             pkt->dst_port == q->id.dst_port) {
 801                                 dir = MATCH_FORWARD;
 802                                 break;
 803                         }
 804                         if (pkt->src_ip == q->id.dst_ip &&
 805                             pkt->dst_ip == q->id.src_ip &&
 806                             pkt->src_port == q->id.dst_port &&
 807                             pkt->dst_port == q->id.src_port) {
 808                                 dir = MATCH_REVERSE;
 809                                 break;
 810                         }
 811                 }
 812 next:
 813                 prev = q;
 814                 q = q->next;
 815         }
 816         if (q == NULL)
 817                 goto done; /* q = NULL, not found */
 818
 819         if (prev != NULL) { /* found and not in front */
 820                 prev->next = q->next;
 821                 q->next = ipfw_dyn_v[i];
 822                 ipfw_dyn_v[i] = q;
 823         }
 824
 825         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 826                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
 827
 828 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 829 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 830
 831                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
 832                 switch (q->state) {
 833                 case TH_SYN:                            /* opening */
 834                         q->expire = time_second + dyn_syn_lifetime;
 835                         break;
 836
 837                 case BOTH_SYN:                  /* move to established */
 838                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
 839                 case BOTH_SYN | (TH_FIN << 8) :
 840                         if (tcp) {
 841                                 uint32_t ack = ntohl(tcp->th_ack);
 842
 843 #define _SEQ_GE(a, b)   ((int)(a) - (int)(b) >= 0)
 844
 845                                 if (dir == MATCH_FORWARD) {
 846                                         if (q->ack_fwd == 0 ||
 847                                             _SEQ_GE(ack, q->ack_fwd))
 848                                                 q->ack_fwd = ack;
 849                                         else /* ignore out-of-sequence */
 850                                                 break;
 851                                 } else {
 852                                         if (q->ack_rev == 0 ||
 853                                             _SEQ_GE(ack, q->ack_rev))
 854                                                 q->ack_rev = ack;
 855                                         else /* ignore out-of-sequence */
 856                                                 break;
 857                                 }
 858 #undef _SEQ_GE
 859                         }
 860                         q->expire = time_second + dyn_ack_lifetime;
 861                         break;
 862
 863                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 864                         if (dyn_fin_lifetime >= dyn_keepalive_period)
 865                                 dyn_fin_lifetime = dyn_keepalive_period - 1;
 866                         q->expire = time_second + dyn_fin_lifetime;
 867                         break;
 868
 869                 default:
 870 #if 0
 871                         /*
 872                          * reset or some invalid combination, but can also
 873                          * occur if we use keep-state the wrong way.
 874                          */
 875                         if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
 876                                 kprintf("invalid state: 0x%x\n", q->state);
 877 #endif
 878                         if (dyn_rst_lifetime >= dyn_keepalive_period)
 879                                 dyn_rst_lifetime = dyn_keepalive_period - 1;
 880                         q->expire = time_second + dyn_rst_lifetime;
 881                         break;
 882                 }
 883         } else if (pkt->proto == IPPROTO_UDP) {
 884                 q->expire = time_second + dyn_udp_lifetime;
 885         } else {
 886                 /* other protocols */
 887                 q->expire = time_second + dyn_short_lifetime;
 888         }
 889 done:
 890         if (match_direction)
 891                 *match_direction = dir;
 892         return q;
 893 }
 894
 895 static void
 896 realloc_dynamic_table(void)
 897 {
 898         /*
 899          * Try reallocation, make sure we have a power of 2 and do
 900          * not allow more than 64k entries. In case of overflow,
 901          * default to 1024.
 902          */
 903
 904         if (dyn_buckets > 65536)
 905                 dyn_buckets = 1024;
 906         if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
 907                 dyn_buckets = curr_dyn_buckets; /* reset */
 908                 return;
 909         }
 910         curr_dyn_buckets = dyn_buckets;
 911
 912         if (ipfw_dyn_v != NULL)
 913                 kfree(ipfw_dyn_v, M_IPFW);
 914
 915         for (;;) {
 916                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
 917                                      M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
 918                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
 919                         break;
 920                 curr_dyn_buckets /= 2;
 921         }
 922 }
 923
 924 /**
 925  * Install state of type 'type' for a dynamic session.
 926  * The hash table contains two type of rules:
 927  * - regular rules (O_KEEP_STATE)
 928  * - rules for sessions with limited number of sess per user
 929  *   (O_LIMIT). When they are created, the parent is
 930  *   increased by 1, and decreased on delete. In this case,
 931  *   the third parameter is the parent rule and not the chain.
 932  * - "parent" rules for the above (O_LIMIT_PARENT).
 933  */
 934 static ipfw_dyn_rule *
 935 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
 936 {
 937         ipfw_dyn_rule *r;
 938         int i;
 939
 940         if (ipfw_dyn_v == NULL ||
 941             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
 942                 realloc_dynamic_table();
 943                 if (ipfw_dyn_v == NULL)
 944                         return NULL; /* failed ! */
 945         }
 946         i = hash_packet(id);
 947
 948         r = kmalloc(sizeof(*r), M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
 949         if (r == NULL) {
 950                 kprintf ("sorry cannot allocate state\n");
 951                 return NULL;
 952         }
 953
 954         /* increase refcount on parent, and set pointer */
 955         if (dyn_type == O_LIMIT) {
 956                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 957
 958                 if (parent->dyn_type != O_LIMIT_PARENT)
 959                         panic("invalid parent");
 960                 parent->count++;
 961                 r->parent = parent;
 962                 rule = parent->rule;
 963         }
 964
 965         r->id = *id;
 966         r->expire = time_second + dyn_syn_lifetime;
 967         r->rule = rule;
 968         r->dyn_type = dyn_type;
 969         r->pcnt = r->bcnt = 0;
 970         r->count = 0;
 971
 972         r->bucket = i;
 973         r->next = ipfw_dyn_v[i];
 974         ipfw_dyn_v[i] = r;
 975         dyn_count++;
 976         DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
 977            dyn_type,
 978            (r->id.src_ip), (r->id.src_port),
 979            (r->id.dst_ip), (r->id.dst_port),
 980            dyn_count );)
 981         return r;
 982 }
 983
 984 /**
 985  * lookup dynamic parent rule using pkt and rule as search keys.
 986  * If the lookup fails, then install one.
 987  */
 988 static ipfw_dyn_rule *
 989 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
 990 {
 991         ipfw_dyn_rule *q;
 992         int i;
 993
 994         if (ipfw_dyn_v) {
 995                 i = hash_packet(pkt);
 996                 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
 997                         if (q->dyn_type == O_LIMIT_PARENT &&
 998                             rule== q->rule &&
 999                             pkt->proto == q->id.proto &&
1000                             pkt->src_ip == q->id.src_ip &&
1001                             pkt->dst_ip == q->id.dst_ip &&
1002                             pkt->src_port == q->id.src_port &&
1003                             pkt->dst_port == q->id.dst_port) {
1004                                 q->expire = time_second + dyn_short_lifetime;
1005                                 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
1006                                 return q;
1007                         }
1008                 }
1009         }
1010         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1011 }
1012
1013 /**
1014  * Install dynamic state for rule type cmd->o.opcode
1015  *
1016  * Returns 1 (failure) if state is not installed because of errors or because
1017  * session limitations are enforced.
1018  */
1019 static int
1020 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1021               struct ip_fw_args *args)
1022 {
1023         static int last_log;
1024
1025         ipfw_dyn_rule *q;
1026
1027         DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1028             cmd->o.opcode,
1029             (args->f_id.src_ip), (args->f_id.src_port),
1030             (args->f_id.dst_ip), (args->f_id.dst_port) );)
1031
1032         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1033         if (q != NULL) { /* should never occur */
1034                 if (last_log != time_second) {
1035                         last_log = time_second;
1036                         kprintf(" install_state: entry already present, done\n");
1037                 }
1038                 return 0;
1039         }
1040
1041         if (dyn_count >= dyn_max) {
1042                 /*
1043                  * Run out of slots, try to remove any expired rule.
1044                  */
1045                 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1046         }
1047
1048         if (dyn_count >= dyn_max) {
1049                 if (last_log != time_second) {
1050                         last_log = time_second;
1051                         kprintf("install_state: Too many dynamic rules\n");
1052                 }
1053                 return 1; /* cannot install, notify caller */
1054         }
1055
1056         switch (cmd->o.opcode) {
1057         case O_KEEP_STATE: /* bidir rule */
1058                 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1059                 break;
1060
1061         case O_LIMIT: /* limit number of sessions */
1062                 {
1063                         uint16_t limit_mask = cmd->limit_mask;
1064                         struct ipfw_flow_id id;
1065                         ipfw_dyn_rule *parent;
1066
1067                         DEB(kprintf("installing dyn-limit rule %d\n",
1068                             cmd->conn_limit);)
1069
1070                         id.dst_ip = id.src_ip = 0;
1071                         id.dst_port = id.src_port = 0;
1072                         id.proto = args->f_id.proto;
1073
1074                         if (limit_mask & DYN_SRC_ADDR)
1075                                 id.src_ip = args->f_id.src_ip;
1076                         if (limit_mask & DYN_DST_ADDR)
1077                                 id.dst_ip = args->f_id.dst_ip;
1078                         if (limit_mask & DYN_SRC_PORT)
1079                                 id.src_port = args->f_id.src_port;
1080                         if (limit_mask & DYN_DST_PORT)
1081                                 id.dst_port = args->f_id.dst_port;
1082
1083                         parent = lookup_dyn_parent(&id, rule);
1084                         if (parent == NULL) {
1085                                 kprintf("add parent failed\n");
1086                                 return 1;
1087                         }
1088
1089                         if (parent->count >= cmd->conn_limit) {
1090                                 /*
1091                                  * See if we can remove some expired rule.
1092                                  */
1093                                 remove_dyn_rule(rule, parent);
1094                                 if (parent->count >= cmd->conn_limit) {
1095                                         if (fw_verbose &&
1096                                             last_log != time_second) {
1097                                                 last_log = time_second;
1098                                                 log(LOG_SECURITY | LOG_DEBUG,
1099                                                     "drop session, "
1100                                                     "too many entries\n");
1101                                         }
1102                                         return 1;
1103                                 }
1104                         }
1105                         add_dyn_rule(&args->f_id, O_LIMIT,
1106                                      (struct ip_fw *)parent);
1107                 }
1108                 break;
1109         default:
1110                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1111                 return 1;
1112         }
1113         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1114         return 0;
1115 }
1116
1117 /*
1118  * Transmit a TCP packet, containing either a RST or a keepalive.
1119  * When flags & TH_RST, we are sending a RST packet, because of a
1120  * "reset" action matched the packet.
1121  * Otherwise we are sending a keepalive, and flags & TH_
1122  */
1123 static void
1124 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1125 {
1126         struct mbuf *m;
1127         struct ip *ip;
1128         struct tcphdr *tcp;
1129         struct route sro;       /* fake route */
1130
1131         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1132         if (m == 0)
1133                 return;
1134         m->m_pkthdr.rcvif = NULL;
1135         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1136         m->m_data += max_linkhdr;
1137
1138         ip = mtod(m, struct ip *);
1139         bzero(ip, m->m_len);
1140         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1141         ip->ip_p = IPPROTO_TCP;
1142         tcp->th_off = 5;
1143
1144         /*
1145          * Assume we are sending a RST (or a keepalive in the reverse
1146          * direction), swap src and destination addresses and ports.
1147          */
1148         ip->ip_src.s_addr = htonl(id->dst_ip);
1149         ip->ip_dst.s_addr = htonl(id->src_ip);
1150         tcp->th_sport = htons(id->dst_port);
1151         tcp->th_dport = htons(id->src_port);
1152         if (flags & TH_RST) {   /* we are sending a RST */
1153                 if (flags & TH_ACK) {
1154                         tcp->th_seq = htonl(ack);
1155                         tcp->th_ack = htonl(0);
1156                         tcp->th_flags = TH_RST;
1157                 } else {
1158                         if (flags & TH_SYN)
1159                                 seq++;
1160                         tcp->th_seq = htonl(0);
1161                         tcp->th_ack = htonl(seq);
1162                         tcp->th_flags = TH_RST | TH_ACK;
1163                 }
1164         } else {
1165                 /*
1166                  * We are sending a keepalive. flags & TH_SYN determines
1167                  * the direction, forward if set, reverse if clear.
1168                  * NOTE: seq and ack are always assumed to be correct
1169                  * as set by the caller. This may be confusing...
1170                  */
1171                 if (flags & TH_SYN) {
1172                         /*
1173                          * we have to rewrite the correct addresses!
1174                          */
1175                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1176                         ip->ip_src.s_addr = htonl(id->src_ip);
1177                         tcp->th_dport = htons(id->dst_port);
1178                         tcp->th_sport = htons(id->src_port);
1179                 }
1180                 tcp->th_seq = htonl(seq);
1181                 tcp->th_ack = htonl(ack);
1182                 tcp->th_flags = TH_ACK;
1183         }
1184
1185         /*
1186          * set ip_len to the payload size so we can compute
1187          * the tcp checksum on the pseudoheader
1188          * XXX check this, could save a couple of words ?
1189          */
1190         ip->ip_len = htons(sizeof(struct tcphdr));
1191         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1192
1193         /*
1194          * now fill fields left out earlier
1195          */
1196         ip->ip_ttl = ip_defttl;
1197         ip->ip_len = m->m_pkthdr.len;
1198
1199         bzero(&sro, sizeof(sro));
1200         ip_rtaddr(ip->ip_dst, &sro);
1201
1202         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1203         ip_output(m, NULL, &sro, 0, NULL, NULL);
1204         if (sro.ro_rt)
1205                 RTFREE(sro.ro_rt);
1206 }
1207
1208 /*
1209  * sends a reject message, consuming the mbuf passed as an argument.
1210  */
1211 static void
1212 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1213 {
1214
1215         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1216                 /* We need the IP header in host order for icmp_error(). */
1217                 if (args->eh != NULL) {
1218                         struct ip *ip = mtod(args->m, struct ip *);
1219                         ip->ip_len = ntohs(ip->ip_len);
1220                         ip->ip_off = ntohs(ip->ip_off);
1221                 }
1222                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1223         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1224                 struct tcphdr *const tcp =
1225                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1226                 if ( (tcp->th_flags & TH_RST) == 0)
1227                         send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1228                                 ntohl(tcp->th_ack),
1229                                 tcp->th_flags | TH_RST);
1230                 m_freem(args->m);
1231         } else
1232                 m_freem(args->m);
1233         args->m = NULL;
1234 }
1235
1236 /**
1237  *
1238  * Given an ip_fw *, lookup_next_rule will return a pointer
1239  * to the next rule, which can be either the jump
1240  * target (for skipto instructions) or the next one in the list (in
1241  * all other cases including a missing jump target).
1242  * The result is also written in the "next_rule" field of the rule.
1243  * Backward jumps are not allowed, so start looking from the next
1244  * rule...
1245  *
1246  * This never returns NULL -- in case we do not have an exact match,
1247  * the next rule is returned. When the ruleset is changed,
1248  * pointers are flushed so we are always correct.
1249  */
1250
1251 static struct ip_fw *
1252 lookup_next_rule(struct ip_fw *me)
1253 {
1254         struct ip_fw *rule = NULL;
1255         ipfw_insn *cmd;
1256
1257         /* look for action, in case it is a skipto */
1258         cmd = ACTION_PTR(me);
1259         if (cmd->opcode == O_LOG)
1260                 cmd += F_LEN(cmd);
1261         if ( cmd->opcode == O_SKIPTO )
1262                 for (rule = me->next; rule ; rule = rule->next)
1263                         if (rule->rulenum >= cmd->arg1)
1264                                 break;
1265         if (rule == NULL)                       /* failure or not a skipto */
1266                 rule = me->next;
1267         me->next_rule = rule;
1268         return rule;
1269 }
1270
1271 /*
1272  * The main check routine for the firewall.
1273  *
1274  * All arguments are in args so we can modify them and return them
1275  * back to the caller.
1276  *
1277  * Parameters:
1278  *
1279  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1280  *              Starts with the IP header.
1281  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1282  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1283  *              The incoming interface is in the mbuf. (in)
1284  *
1285  *      args->rule      Pointer to the last matching rule (in/out)
1286  *      args->next_hop  Socket we are forwarding to (out).
1287  *      args->f_id      Addresses grabbed from the packet (out)
1288  *
1289  * Return value:
1290  *
1291  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1292  *      0       The packet is to be accepted and routed normally OR
1293  *              the packet was denied/rejected and has been dropped;
1294  *              in the latter case, *m is equal to NULL upon return.
1295  *      port    Divert the packet to port, with these caveats:
1296  *
1297  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1298  *                of diverting it (ie, 'ipfw tee').
1299  *
1300  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1301  *                16 bits as a dummynet pipe number instead of diverting
1302  */
1303
1304 static int
1305 ipfw_chk(struct ip_fw_args *args)
1306 {
1307         /*
1308          * Local variables hold state during the processing of a packet.
1309          *
1310          * IMPORTANT NOTE: to speed up the processing of rules, there
1311          * are some assumption on the values of the variables, which
1312          * are documented here. Should you change them, please check
1313          * the implementation of the various instructions to make sure
1314          * that they still work.
1315          *
1316          * args->eh     The MAC header. It is non-null for a layer2
1317          *      packet, it is NULL for a layer-3 packet.
1318          *
1319          * m | args->m  Pointer to the mbuf, as received from the caller.
1320          *      It may change if ipfw_chk() does an m_pullup, or if it
1321          *      consumes the packet because it calls send_reject().
1322          *      XXX This has to change, so that ipfw_chk() never modifies
1323          *      or consumes the buffer.
1324          * ip   is simply an alias of the value of m, and it is kept
1325          *      in sync with it (the packet is  supposed to start with
1326          *      the ip header).
1327          */
1328         struct mbuf *m = args->m;
1329         struct ip *ip = mtod(m, struct ip *);
1330
1331         /*
1332          * oif | args->oif      If NULL, ipfw_chk has been called on the
1333          *      inbound path (ether_input, ip_input).
1334          *      If non-NULL, ipfw_chk has been called on the outbound path
1335          *      (ether_output, ip_output).
1336          */
1337         struct ifnet *oif = args->oif;
1338
1339         struct ip_fw *f = NULL;         /* matching rule */
1340         int retval = 0;
1341         struct m_tag *mtag;
1342
1343         /*
1344          * hlen The length of the IPv4 header.
1345          *      hlen >0 means we have an IPv4 packet.
1346          */
1347         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1348
1349         /*
1350          * offset       The offset of a fragment. offset != 0 means that
1351          *      we have a fragment at this offset of an IPv4 packet.
1352          *      offset == 0 means that (if this is an IPv4 packet)
1353          *      this is the first or only fragment.
1354          */
1355         u_short offset = 0;
1356
1357         /*
1358          * Local copies of addresses. They are only valid if we have
1359          * an IP packet.
1360          *
1361          * proto        The protocol. Set to 0 for non-ip packets,
1362          *      or to the protocol read from the packet otherwise.
1363          *      proto != 0 means that we have an IPv4 packet.
1364          *
1365          * src_port, dst_port   port numbers, in HOST format. Only
1366          *      valid for TCP and UDP packets.
1367          *
1368          * src_ip, dst_ip       ip addresses, in NETWORK format.
1369          *      Only valid for IPv4 packets.
1370          */
1371         uint8_t proto;
1372         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1373         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1374         uint16_t ip_len = 0;
1375         int dyn_dir = MATCH_UNKNOWN;
1376         ipfw_dyn_rule *q = NULL;
1377
1378         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1379                 return 0;       /* accept */
1380         /*
1381          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1382          *      MATCH_NONE when checked and not matched (q = NULL),
1383          *      MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1384          */
1385
1386         if (args->eh == NULL ||         /* layer 3 packet */
1387             (m->m_pkthdr.len >= sizeof(struct ip) &&
1388              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1389                 hlen = ip->ip_hl << 2;
1390
1391         /*
1392          * Collect parameters into local variables for faster matching.
1393          */
1394         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1395                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1396                 goto after_ip_checks;
1397         }
1398
1399         proto = args->f_id.proto = ip->ip_p;
1400         src_ip = ip->ip_src;
1401         dst_ip = ip->ip_dst;
1402         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1403                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1404                 ip_len = ntohs(ip->ip_len);
1405         } else {
1406                 offset = ip->ip_off & IP_OFFMASK;
1407                 ip_len = ip->ip_len;
1408         }
1409
1410 #define PULLUP_TO(len)                          \
1411 do {                                            \
1412         if (m->m_len < (len)) {                 \
1413                 args->m = m = m_pullup(m, (len));\
1414                 if (m == NULL)                  \
1415                         goto pullup_failed;     \
1416                 ip = mtod(m, struct ip *);      \
1417         }                                       \
1418 } while (0)
1419
1420         if (offset == 0) {
1421                 switch (proto) {
1422                 case IPPROTO_TCP:
1423                         {
1424                                 struct tcphdr *tcp;
1425
1426                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1427                                 tcp = L3HDR(struct tcphdr, ip);
1428                                 dst_port = tcp->th_dport;
1429                                 src_port = tcp->th_sport;
1430                                 args->f_id.flags = tcp->th_flags;
1431                         }
1432                         break;
1433
1434                 case IPPROTO_UDP:
1435                         {
1436                                 struct udphdr *udp;
1437
1438                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1439                                 udp = L3HDR(struct udphdr, ip);
1440                                 dst_port = udp->uh_dport;
1441                                 src_port = udp->uh_sport;
1442                         }
1443                         break;
1444
1445                 case IPPROTO_ICMP:
1446                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1447                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1448                         break;
1449
1450                 default:
1451                         break;
1452                 }
1453         }
1454
1455 #undef PULLUP_TO
1456
1457         args->f_id.src_ip = ntohl(src_ip.s_addr);
1458         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1459         args->f_id.src_port = src_port = ntohs(src_port);
1460         args->f_id.dst_port = dst_port = ntohs(dst_port);
1461
1462 after_ip_checks:
1463         if (args->rule) {
1464                 /*
1465                  * Packet has already been tagged. Look for the next rule
1466                  * to restart processing.
1467                  *
1468                  * If fw_one_pass != 0 then just accept it.
1469                  * XXX should not happen here, but optimized out in
1470                  * the caller.
1471                  */
1472                 if (fw_one_pass)
1473                         return 0;
1474
1475                 /* This rule was deleted */
1476                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1477                         return IP_FW_PORT_DENY_FLAG;
1478
1479                 f = args->rule->next_rule;
1480                 if (f == NULL)
1481                         f = lookup_next_rule(args->rule);
1482         } else {
1483                 /*
1484                  * Find the starting rule. It can be either the first
1485                  * one, or the one after divert_rule if asked so.
1486                  */
1487                 int skipto;
1488
1489                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1490                 if (mtag != NULL)
1491                         skipto = *(uint16_t *)m_tag_data(mtag);
1492                 else
1493                         skipto = 0;
1494
1495                 f = layer3_chain;
1496                 if (args->eh == NULL && skipto != 0) {
1497                         if (skipto >= IPFW_DEFAULT_RULE)
1498                                 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1499                         while (f && f->rulenum <= skipto)
1500                                 f = f->next;
1501                         if (f == NULL)  /* drop packet */
1502                                 return(IP_FW_PORT_DENY_FLAG);
1503                 }
1504         }
1505         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1506                 m_tag_delete(m, mtag);
1507
1508         /*
1509          * Now scan the rules, and parse microinstructions for each rule.
1510          */
1511         for (; f; f = f->next) {
1512                 int l, cmdlen;
1513                 ipfw_insn *cmd;
1514                 int skip_or; /* skip rest of OR block */
1515
1516 again:
1517                 if (set_disable & (1 << f->set))
1518                         continue;
1519
1520                 skip_or = 0;
1521                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1522                      l -= cmdlen, cmd += cmdlen) {
1523                         int match;
1524
1525                         /*
1526                          * check_body is a jump target used when we find a
1527                          * CHECK_STATE, and need to jump to the body of
1528                          * the target rule.
1529                          */
1530
1531 check_body:
1532                         cmdlen = F_LEN(cmd);
1533                         /*
1534                          * An OR block (insn_1 || .. || insn_n) has the
1535                          * F_OR bit set in all but the last instruction.
1536                          * The first match will set "skip_or", and cause
1537                          * the following instructions to be skipped until
1538                          * past the one with the F_OR bit clear.
1539                          */
1540                         if (skip_or) {          /* skip this instruction */
1541                                 if ((cmd->len & F_OR) == 0)
1542                                         skip_or = 0;    /* next one is good */
1543                                 continue;
1544                         }
1545                         match = 0; /* set to 1 if we succeed */
1546
1547                         switch (cmd->opcode) {
1548                         /*
1549                          * The first set of opcodes compares the packet's
1550                          * fields with some pattern, setting 'match' if a
1551                          * match is found. At the end of the loop there is
1552                          * logic to deal with F_NOT and F_OR flags associated
1553                          * with the opcode.
1554                          */
1555                         case O_NOP:
1556                                 match = 1;
1557                                 break;
1558
1559                         case O_FORWARD_MAC:
1560                                 kprintf("ipfw: opcode %d unimplemented\n",
1561                                         cmd->opcode);
1562                                 break;
1563
1564                         case O_GID:
1565                         case O_UID:
1566                                 /*
1567                                  * We only check offset == 0 && proto != 0,
1568                                  * as this ensures that we have an IPv4
1569                                  * packet with the ports info.
1570                                  */
1571                                 if (offset!=0)
1572                                         break;
1573                             {
1574                                 struct inpcbinfo *pi;
1575                                 int wildcard;
1576                                 struct inpcb *pcb;
1577
1578                                 if (proto == IPPROTO_TCP) {
1579                                         wildcard = 0;
1580                                         pi = &tcbinfo[mycpu->gd_cpuid];
1581                                 } else if (proto == IPPROTO_UDP) {
1582                                         wildcard = 1;
1583                                         pi = &udbinfo;
1584                                 } else
1585                                         break;
1586
1587                                 pcb =  (oif) ?
1588                                         in_pcblookup_hash(pi,
1589                                             dst_ip, htons(dst_port),
1590                                             src_ip, htons(src_port),
1591                                             wildcard, oif) :
1592                                         in_pcblookup_hash(pi,
1593                                             src_ip, htons(src_port),
1594                                             dst_ip, htons(dst_port),
1595                                             wildcard, NULL);
1596
1597                                 if (pcb == NULL || pcb->inp_socket == NULL)
1598                                         break;
1599
1600                                 if (cmd->opcode == O_UID) {
1601 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1602                                         match =
1603                                           !socheckuid(pcb->inp_socket,
1604                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1605 #undef socheckuid
1606                                 } else  {
1607                                         match = groupmember(
1608                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1609                                             pcb->inp_socket->so_cred);
1610                                 }
1611                             }
1612                                 break;
1613
1614                         case O_RECV:
1615                                 match = iface_match(m->m_pkthdr.rcvif,
1616                                     (ipfw_insn_if *)cmd);
1617                                 break;
1618
1619                         case O_XMIT:
1620                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1621                                 break;
1622
1623                         case O_VIA:
1624                                 match = iface_match(oif ? oif :
1625                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1626                                 break;
1627
1628                         case O_MACADDR2:
1629                                 if (args->eh != NULL) { /* have MAC header */
1630                                         uint32_t *want = (uint32_t *)
1631                                                 ((ipfw_insn_mac *)cmd)->addr;
1632                                         uint32_t *mask = (uint32_t *)
1633                                                 ((ipfw_insn_mac *)cmd)->mask;
1634                                         uint32_t *hdr = (uint32_t *)args->eh;
1635
1636                                         match =
1637                                         (want[0] == (hdr[0] & mask[0]) &&
1638                                          want[1] == (hdr[1] & mask[1]) &&
1639                                          want[2] == (hdr[2] & mask[2]));
1640                                 }
1641                                 break;
1642
1643                         case O_MAC_TYPE:
1644                                 if (args->eh != NULL) {
1645                                         uint16_t t =
1646                                             ntohs(args->eh->ether_type);
1647                                         uint16_t *p =
1648                                             ((ipfw_insn_u16 *)cmd)->ports;
1649                                         int i;
1650
1651                                         /* Special vlan handling */
1652                                         if (m->m_flags & M_VLANTAG)
1653                                                 t = ETHERTYPE_VLAN;
1654
1655                                         for (i = cmdlen - 1; !match && i > 0;
1656                                              i--, p += 2) {
1657                                                 match =
1658                                                 (t >= p[0] && t <= p[1]);
1659                                         }
1660                                 }
1661                                 break;
1662
1663                         case O_FRAG:
1664                                 match = (hlen > 0 && offset != 0);
1665                                 break;
1666
1667                         case O_IN:      /* "out" is "not in" */
1668                                 match = (oif == NULL);
1669                                 break;
1670
1671                         case O_LAYER2:
1672                                 match = (args->eh != NULL);
1673                                 break;
1674
1675                         case O_PROTO:
1676                                 /*
1677                                  * We do not allow an arg of 0 so the
1678                                  * check of "proto" only suffices.
1679                                  */
1680                                 match = (proto == cmd->arg1);
1681                                 break;
1682
1683                         case O_IP_SRC:
1684                                 match = (hlen > 0 &&
1685                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1686                                     src_ip.s_addr);
1687                                 break;
1688
1689                         case O_IP_SRC_MASK:
1690                                 match = (hlen > 0 &&
1691                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1692                                      (src_ip.s_addr &
1693                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1694                                 break;
1695
1696                         case O_IP_SRC_ME:
1697                                 if (hlen > 0) {
1698                                         struct ifnet *tif;
1699
1700                                         tif = INADDR_TO_IFP(&src_ip);
1701                                         match = (tif != NULL);
1702                                 }
1703                                 break;
1704
1705                         case O_IP_DST_SET:
1706                         case O_IP_SRC_SET:
1707                                 if (hlen > 0) {
1708                                         uint32_t *d = (uint32_t *)(cmd+1);
1709                                         uint32_t addr =
1710                                             cmd->opcode == O_IP_DST_SET ?
1711                                                 args->f_id.dst_ip :
1712                                                 args->f_id.src_ip;
1713
1714                                         if (addr < d[0])
1715                                                 break;
1716                                         addr -= d[0]; /* subtract base */
1717                                         match =
1718                                         (addr < cmd->arg1) &&
1719                                          (d[1 + (addr >> 5)] &
1720                                           (1 << (addr & 0x1f)));
1721                                 }
1722                                 break;
1723
1724                         case O_IP_DST:
1725                                 match = (hlen > 0 &&
1726                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1727                                     dst_ip.s_addr);
1728                                 break;
1729
1730                         case O_IP_DST_MASK:
1731                                 match = (hlen > 0) &&
1732                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1733                                      (dst_ip.s_addr &
1734                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1735                                 break;
1736
1737                         case O_IP_DST_ME:
1738                                 if (hlen > 0) {
1739                                         struct ifnet *tif;
1740
1741                                         tif = INADDR_TO_IFP(&dst_ip);
1742                                         match = (tif != NULL);
1743                                 }
1744                                 break;
1745
1746                         case O_IP_SRCPORT:
1747                         case O_IP_DSTPORT:
1748                                 /*
1749                                  * offset == 0 && proto != 0 is enough
1750                                  * to guarantee that we have an IPv4
1751                                  * packet with port info.
1752                                  */
1753                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1754                                     && offset == 0) {
1755                                         uint16_t x =
1756                                             (cmd->opcode == O_IP_SRCPORT) ?
1757                                                 src_port : dst_port ;
1758                                         uint16_t *p =
1759                                             ((ipfw_insn_u16 *)cmd)->ports;
1760                                         int i;
1761
1762                                         for (i = cmdlen - 1; !match && i > 0;
1763                                              i--, p += 2) {
1764                                                 match =
1765                                                 (x >= p[0] && x <= p[1]);
1766                                         }
1767                                 }
1768                                 break;
1769
1770                         case O_ICMPTYPE:
1771                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
1772                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
1773                                 break;
1774
1775                         case O_IPOPT:
1776                                 match = (hlen > 0 && ipopts_match(ip, cmd));
1777                                 break;
1778
1779                         case O_IPVER:
1780                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1781                                 break;
1782
1783                         case O_IPTTL:
1784                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1785                                 break;
1786
1787                         case O_IPID:
1788                                 match = (hlen > 0 &&
1789                                     cmd->arg1 == ntohs(ip->ip_id));
1790                                 break;
1791
1792                         case O_IPLEN:
1793                                 match = (hlen > 0 && cmd->arg1 == ip_len);
1794                                 break;
1795
1796                         case O_IPPRECEDENCE:
1797                                 match = (hlen > 0 &&
1798                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
1799                                 break;
1800
1801                         case O_IPTOS:
1802                                 match = (hlen > 0 &&
1803                                     flags_match(cmd, ip->ip_tos));
1804                                 break;
1805
1806                         case O_TCPFLAGS:
1807                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1808                                     flags_match(cmd,
1809                                         L3HDR(struct tcphdr,ip)->th_flags));
1810                                 break;
1811
1812                         case O_TCPOPTS:
1813                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1814                                     tcpopts_match(ip, cmd));
1815                                 break;
1816
1817                         case O_TCPSEQ:
1818                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1819                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1820                                         L3HDR(struct tcphdr,ip)->th_seq);
1821                                 break;
1822
1823                         case O_TCPACK:
1824                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1825                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1826                                         L3HDR(struct tcphdr,ip)->th_ack);
1827                                 break;
1828
1829                         case O_TCPWIN:
1830                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1831                                     cmd->arg1 ==
1832                                         L3HDR(struct tcphdr,ip)->th_win);
1833                                 break;
1834
1835                         case O_ESTAB:
1836                                 /* reject packets which have SYN only */
1837                                 /* XXX should i also check for TH_ACK ? */
1838                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1839                                     (L3HDR(struct tcphdr,ip)->th_flags &
1840                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1841                                 break;
1842
1843                         case O_LOG:
1844                                 if (fw_verbose)
1845                                         ipfw_log(f, hlen, args->eh, m, oif);
1846                                 match = 1;
1847                                 break;
1848
1849                         case O_PROB:
1850                                 match = (krandom() <
1851                                         ((ipfw_insn_u32 *)cmd)->d[0]);
1852                                 break;
1853
1854                         /*
1855                          * The second set of opcodes represents 'actions',
1856                          * i.e. the terminal part of a rule once the packet
1857                          * matches all previous patterns.
1858                          * Typically there is only one action for each rule,
1859                          * and the opcode is stored at the end of the rule
1860                          * (but there are exceptions -- see below).
1861                          *
1862                          * In general, here we set retval and terminate the
1863                          * outer loop (would be a 'break 3' in some language,
1864                          * but we need to do a 'goto done').
1865                          *
1866                          * Exceptions:
1867                          * O_COUNT and O_SKIPTO actions:
1868                          *   instead of terminating, we jump to the next rule
1869                          *   ('goto next_rule', equivalent to a 'break 2'),
1870                          *   or to the SKIPTO target ('goto again' after
1871                          *   having set f, cmd and l), respectively.
1872                          *
1873                          * O_LIMIT and O_KEEP_STATE: these opcodes are
1874                          *   not real 'actions', and are stored right
1875                          *   before the 'action' part of the rule.
1876                          *   These opcodes try to install an entry in the
1877                          *   state tables; if successful, we continue with
1878                          *   the next opcode (match=1; break;), otherwise
1879                          *   the packet *   must be dropped
1880                          *   ('goto done' after setting retval);
1881                          *
1882                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1883                          *   cause a lookup of the state table, and a jump
1884                          *   to the 'action' part of the parent rule
1885                          *   ('goto check_body') if an entry is found, or
1886                          *   (CHECK_STATE only) a jump to the next rule if
1887                          *   the entry is not found ('goto next_rule').
1888                          *   The result of the lookup is cached to make
1889                          *   further instances of these opcodes are
1890                          *   effectively NOPs.
1891                          */
1892                         case O_LIMIT:
1893                         case O_KEEP_STATE:
1894                                 if (install_state(f,
1895                                     (ipfw_insn_limit *)cmd, args)) {
1896                                         retval = IP_FW_PORT_DENY_FLAG;
1897                                         goto done; /* error/limit violation */
1898                                 }
1899                                 match = 1;
1900                                 break;
1901
1902                         case O_PROBE_STATE:
1903                         case O_CHECK_STATE:
1904                                 /*
1905                                  * dynamic rules are checked at the first
1906                                  * keep-state or check-state occurrence,
1907                                  * with the result being stored in dyn_dir.
1908                                  * The compiler introduces a PROBE_STATE
1909                                  * instruction for us when we have a
1910                                  * KEEP_STATE (because PROBE_STATE needs
1911                                  * to be run first).
1912                                  */
1913                                 if (dyn_dir == MATCH_UNKNOWN &&
1914                                     (q = lookup_dyn_rule(&args->f_id,
1915                                      &dyn_dir, proto == IPPROTO_TCP ?
1916                                         L3HDR(struct tcphdr, ip) : NULL))
1917                                         != NULL) {
1918                                         /*
1919                                          * Found dynamic entry, update stats
1920                                          * and jump to the 'action' part of
1921                                          * the parent rule.
1922                                          */
1923                                         q->pcnt++;
1924                                         q->bcnt += ip_len;
1925                                         f = q->rule;
1926                                         cmd = ACTION_PTR(f);
1927                                         l = f->cmd_len - f->act_ofs;
1928                                         goto check_body;
1929                                 }
1930                                 /*
1931                                  * Dynamic entry not found. If CHECK_STATE,
1932                                  * skip to next rule, if PROBE_STATE just
1933                                  * ignore and continue with next opcode.
1934                                  */
1935                                 if (cmd->opcode == O_CHECK_STATE)
1936                                         goto next_rule;
1937                                 match = 1;
1938                                 break;
1939
1940                         case O_ACCEPT:
1941                                 retval = 0;     /* accept */
1942                                 goto done;
1943
1944                         case O_PIPE:
1945                         case O_QUEUE:
1946                                 args->rule = f; /* report matching rule */
1947                                 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
1948                                 goto done;
1949
1950                         case O_DIVERT:
1951                         case O_TEE:
1952                                 if (args->eh) /* not on layer 2 */
1953                                         break;
1954
1955                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
1956                                                  sizeof(uint16_t), MB_DONTWAIT);
1957                                 if (mtag == NULL) {
1958                                         retval = IP_FW_PORT_DENY_FLAG;
1959                                         goto done;
1960                                 }
1961                                 *(uint16_t *)m_tag_data(mtag) = f->rulenum;
1962                                 m_tag_prepend(m, mtag);
1963                                 retval = (cmd->opcode == O_DIVERT) ?
1964                                     cmd->arg1 :
1965                                     cmd->arg1 | IP_FW_PORT_TEE_FLAG;
1966                                 goto done;
1967
1968                         case O_COUNT:
1969                         case O_SKIPTO:
1970                                 f->pcnt++;      /* update stats */
1971                                 f->bcnt += ip_len;
1972                                 f->timestamp = time_second;
1973                                 if (cmd->opcode == O_COUNT)
1974                                         goto next_rule;
1975                                 /* handle skipto */
1976                                 if (f->next_rule == NULL)
1977                                         lookup_next_rule(f);
1978                                 f = f->next_rule;
1979                                 goto again;
1980
1981                         case O_REJECT:
1982                                 /*
1983                                  * Drop the packet and send a reject notice
1984                                  * if the packet is not ICMP (or is an ICMP
1985                                  * query), and it is not multicast/broadcast.
1986                                  */
1987                                 if (hlen > 0 &&
1988                                     (proto != IPPROTO_ICMP ||
1989                                      is_icmp_query(ip)) &&
1990                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
1991                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
1992                                         send_reject(args, cmd->arg1,
1993                                             offset,ip_len);
1994                                         m = args->m;
1995                                 }
1996                                 /* FALLTHROUGH */
1997                         case O_DENY:
1998                                 retval = IP_FW_PORT_DENY_FLAG;
1999                                 goto done;
2000
2001                         case O_FORWARD_IP:
2002                                 if (args->eh)   /* not valid on layer2 pkts */
2003                                         break;
2004                                 if (!q || dyn_dir == MATCH_FORWARD) {
2005                                         args->next_hop =
2006                                             &((ipfw_insn_sa *)cmd)->sa;
2007                                 }
2008                                 retval = 0;
2009                                 goto done;
2010
2011                         default:
2012                                 panic("-- unknown opcode %d\n", cmd->opcode);
2013                         } /* end of switch() on opcodes */
2014
2015                         if (cmd->len & F_NOT)
2016                                 match = !match;
2017
2018                         if (match) {
2019                                 if (cmd->len & F_OR)
2020                                         skip_or = 1;
2021                         } else {
2022                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2023                                         break;          /* try next rule    */
2024                         }
2025
2026                 }       /* end of inner for, scan opcodes */
2027
2028 next_rule:;             /* try next rule                */
2029
2030         }               /* end of outer for, scan rules */
2031         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2032         return(IP_FW_PORT_DENY_FLAG);
2033
2034 done:
2035         /* Update statistics */
2036         f->pcnt++;
2037         f->bcnt += ip_len;
2038         f->timestamp = time_second;
2039         return retval;
2040
2041 pullup_failed:
2042         if (fw_verbose)
2043                 kprintf("pullup failed\n");
2044         return(IP_FW_PORT_DENY_FLAG);
2045 }
2046
2047 static void
2048 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2049 {
2050         struct m_tag *mtag;
2051         struct dn_pkt *pkt;
2052         ipfw_insn *cmd;
2053         const struct ipfw_flow_id *id;
2054         struct dn_flow_id *fid;
2055
2056         M_ASSERTPKTHDR(m);
2057
2058         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2059         if (mtag == NULL) {
2060                 m_freem(m);
2061                 return;
2062         }
2063         m_tag_prepend(m, mtag);
2064
2065         pkt = m_tag_data(mtag);
2066         bzero(pkt, sizeof(*pkt));
2067
2068         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2069         if (cmd->opcode == O_LOG)
2070                 cmd += F_LEN(cmd);
2071         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2072                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2073
2074         pkt->dn_m = m;
2075         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2076         pkt->ifp = fwa->oif;
2077         pkt->cpuid = mycpu->gd_cpuid;
2078         pkt->pipe_nr = pipe_nr;
2079
2080         id = &fwa->f_id;
2081         fid = &pkt->id;
2082         fid->fid_dst_ip = id->dst_ip;
2083         fid->fid_src_ip = id->src_ip;
2084         fid->fid_dst_port = id->dst_port;
2085         fid->fid_src_port = id->src_port;
2086         fid->fid_proto = id->proto;
2087         fid->fid_flags = id->flags;
2088
2089         ipfw_ref_rule(fwa->rule);
2090         pkt->dn_priv = fwa->rule;
2091         pkt->dn_unref_priv = ipfw_unref_rule;
2092
2093         if (cmd->opcode == O_PIPE)
2094                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2095
2096         if (dir == DN_TO_IP_OUT) {
2097                 /*
2098                  * We need to copy *ro because for ICMP pkts (and maybe
2099                  * others) the caller passed a pointer into the stack;
2100                  * dst might also be a pointer into *ro so it needs to
2101                  * be updated.
2102                  */
2103                 pkt->ro = *(fwa->ro);
2104                 if (fwa->ro->ro_rt)
2105                         fwa->ro->ro_rt->rt_refcnt++;
2106                 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) {
2107                         /* 'dst' points into 'ro' */
2108                         fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst);
2109                 }
2110                 pkt->dn_dst = fwa->dst;
2111                 pkt->flags = fwa->flags;
2112         }
2113
2114         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2115         ip_dn_queue(m);
2116 }
2117
2118 /*
2119  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2120  * These will be reconstructed on the fly as packets are matched.
2121  * Must be called at splimp().
2122  */
2123 static void
2124 flush_rule_ptrs(void)
2125 {
2126         struct ip_fw *rule;
2127
2128         for (rule = layer3_chain; rule; rule = rule->next)
2129                 rule->next_rule = NULL;
2130 }
2131
2132 static __inline void
2133 ipfw_inc_static_count(struct ip_fw *rule)
2134 {
2135         static_count++;
2136         static_ioc_len += IOC_RULESIZE(rule);
2137 }
2138
2139 static __inline void
2140 ipfw_dec_static_count(struct ip_fw *rule)
2141 {
2142         int l = IOC_RULESIZE(rule);
2143
2144         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2145         static_count--;
2146
2147         KASSERT(static_ioc_len >= l,
2148                 ("invalid static len %u\n", static_ioc_len));
2149         static_ioc_len -= l;
2150 }
2151
2152 static struct ip_fw *
2153 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule)
2154 {
2155         struct ip_fw *rule;
2156
2157         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2158
2159         rule->act_ofs = ioc_rule->act_ofs;
2160         rule->cmd_len = ioc_rule->cmd_len;
2161         rule->rulenum = ioc_rule->rulenum;
2162         rule->set = ioc_rule->set;
2163         rule->usr_flags = ioc_rule->usr_flags;
2164
2165         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2166
2167         rule->refcnt = 1;
2168
2169         return rule;
2170 }
2171
2172 /*
2173  * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2174  * possibly create a rule number and add the rule to the list.
2175  * Update the rule_number in the input struct so the caller knows it as well.
2176  */
2177 static int
2178 ipfw_add_rule(struct ip_fw **head, struct ipfw_ioc_rule *ioc_rule)
2179 {
2180         struct ip_fw *rule, *f, *prev;
2181
2182         KKASSERT(*head != NULL);
2183
2184         rule = ipfw_create_rule(ioc_rule);
2185
2186         crit_enter();
2187
2188         /*
2189          * If rulenum is 0, find highest numbered rule before the
2190          * default rule, and add autoinc_step
2191          */
2192         if (autoinc_step < 1)
2193                 autoinc_step = 1;
2194         else if (autoinc_step > 1000)
2195                 autoinc_step = 1000;
2196         if (rule->rulenum == 0) {
2197                 /*
2198                  * locate the highest numbered rule before default
2199                  */
2200                 for (f = *head; f; f = f->next) {
2201                         if (f->rulenum == IPFW_DEFAULT_RULE)
2202                                 break;
2203                         rule->rulenum = f->rulenum;
2204                 }
2205                 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2206                         rule->rulenum += autoinc_step;
2207                 ioc_rule->rulenum = rule->rulenum;
2208         }
2209
2210         /*
2211          * Now insert the new rule in the right place in the sorted list.
2212          */
2213         for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2214                 if (f->rulenum > rule->rulenum) { /* found the location */
2215                         if (prev) {
2216                                 rule->next = f;
2217                                 prev->next = rule;
2218                         } else { /* head insert */
2219                                 rule->next = *head;
2220                                 *head = rule;
2221                         }
2222                         break;
2223                 }
2224         }
2225
2226         flush_rule_ptrs();
2227         ipfw_inc_static_count(rule);
2228
2229         crit_exit();
2230
2231         DEB(kprintf("++ installed rule %d, static count now %d\n",
2232                 rule->rulenum, static_count);)
2233         return (0);
2234 }
2235
2236 /**
2237  * Free storage associated with a static rule (including derived
2238  * dynamic rules).
2239  * The caller is in charge of clearing rule pointers to avoid
2240  * dangling pointers.
2241  * @return a pointer to the next entry.
2242  * Arguments are not checked, so they better be correct.
2243  * Must be called at splimp().
2244  */
2245 static struct ip_fw *
2246 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2247 {
2248         struct ip_fw *n;
2249
2250         n = rule->next;
2251         remove_dyn_rule(rule, NULL /* force removal */);
2252         if (prev == NULL)
2253                 *head = n;
2254         else
2255                 prev->next = n;
2256         ipfw_dec_static_count(rule);
2257
2258         /* Mark the rule as invalid */
2259         rule->rule_flags |= IPFW_RULE_F_INVALID;
2260         rule->next_rule = NULL;
2261
2262         /* Try to free this rule */
2263         ipfw_free_rule(rule);
2264
2265         return n;
2266 }
2267
2268 /*
2269  * Deletes all rules from a chain (including the default rule
2270  * if the second argument is set).
2271  * Must be called at splimp().
2272  */
2273 static void
2274 free_chain(struct ip_fw **chain, int kill_default)
2275 {
2276         struct ip_fw *rule;
2277
2278         flush_rule_ptrs(); /* more efficient to do outside the loop */
2279
2280         while ((rule = *chain) != NULL &&
2281                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2282                 delete_rule(chain, NULL, rule);
2283
2284         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2285
2286         if (kill_default) {
2287                 ip_fw_default_rule = NULL;      /* Reset default rule */
2288
2289                 if (ipfw_dyn_v != NULL) {
2290                         /*
2291                          * Free dynamic rules(state) hash table
2292                          */
2293                         kfree(ipfw_dyn_v, M_IPFW);
2294                         ipfw_dyn_v = NULL;
2295                 }
2296
2297                 KASSERT(static_count == 0,
2298                         ("%u static rules remains\n", static_count));
2299                 KASSERT(static_ioc_len == 0,
2300                         ("%u bytes of static rules remains\n", static_ioc_len));
2301         } else {
2302                 KASSERT(static_count == 1,
2303                         ("%u static rules remains\n", static_count));
2304                 KASSERT(static_ioc_len == IOC_RULESIZE(ip_fw_default_rule),
2305                         ("%u bytes of static rules remains, should be %u\n",
2306                          static_ioc_len, IOC_RULESIZE(ip_fw_default_rule)));
2307         }
2308 }
2309
2310 /**
2311  * Remove all rules with given number, and also do set manipulation.
2312  *
2313  * The argument is an uint32_t. The low 16 bit are the rule or set number,
2314  * the next 8 bits are the new set, the top 8 bits are the command:
2315  *
2316  *      0       delete rules with given number
2317  *      1       delete rules with given set number
2318  *      2       move rules with given number to new set
2319  *      3       move rules with given set number to new set
2320  *      4       swap sets with given numbers
2321  */
2322 static int
2323 del_entry(struct ip_fw **chain, uint32_t arg)
2324 {
2325         struct ip_fw *prev, *rule;
2326         uint16_t rulenum;
2327         uint8_t cmd, new_set;
2328
2329         rulenum = arg & 0xffff;
2330         cmd = (arg >> 24) & 0xff;
2331         new_set = (arg >> 16) & 0xff;
2332
2333         if (cmd > 4)
2334                 return EINVAL;
2335         if (new_set > 30)
2336                 return EINVAL;
2337         if (cmd == 0 || cmd == 2) {
2338                 if (rulenum == IPFW_DEFAULT_RULE)
2339                         return EINVAL;
2340         } else {
2341                 if (rulenum > 30)
2342                         return EINVAL;
2343         }
2344
2345         switch (cmd) {
2346         case 0: /* delete rules with given number */
2347                 /*
2348                  * locate first rule to delete
2349                  */
2350                 for (prev = NULL, rule = *chain;
2351                      rule && rule->rulenum < rulenum;
2352                      prev = rule, rule = rule->next)
2353                         ;
2354                 if (rule->rulenum != rulenum)
2355                         return EINVAL;
2356
2357                 crit_enter(); /* no access to rules while removing */
2358                 /*
2359                  * flush pointers outside the loop, then delete all matching
2360                  * rules. prev remains the same throughout the cycle.
2361                  */
2362                 flush_rule_ptrs();
2363                 while (rule && rule->rulenum == rulenum)
2364                         rule = delete_rule(chain, prev, rule);
2365                 crit_exit();
2366                 break;
2367
2368         case 1: /* delete all rules with given set number */
2369                 crit_enter();
2370                 flush_rule_ptrs();
2371                 for (prev = NULL, rule = *chain; rule;) {
2372                         if (rule->set == rulenum) {
2373                                 rule = delete_rule(chain, prev, rule);
2374                         } else {
2375                                 prev = rule;
2376                                 rule = rule->next;
2377                         }
2378                 }
2379                 crit_exit();
2380                 break;
2381
2382         case 2: /* move rules with given number to new set */
2383                 crit_enter();
2384                 for (rule = *chain; rule; rule = rule->next) {
2385                         if (rule->rulenum == rulenum)
2386                                 rule->set = new_set;
2387                 }
2388                 crit_exit();
2389                 break;
2390
2391         case 3: /* move rules with given set number to new set */
2392                 crit_enter();
2393                 for (rule = *chain; rule; rule = rule->next) {
2394                         if (rule->set == rulenum)
2395                                 rule->set = new_set;
2396                 }
2397                 crit_exit();
2398                 break;
2399
2400         case 4: /* swap two sets */
2401                 crit_enter();
2402                 for (rule = *chain; rule; rule = rule->next) {
2403                         if (rule->set == rulenum)
2404                                 rule->set = new_set;
2405                         else if (rule->set == new_set)
2406                                 rule->set = rulenum;
2407                 }
2408                 crit_exit();
2409                 break;
2410         }
2411         return 0;
2412 }
2413
2414 /*
2415  * Clear counters for a specific rule.
2416  */
2417 static void
2418 clear_counters(struct ip_fw *rule, int log_only)
2419 {
2420         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2421
2422         if (log_only == 0) {
2423                 rule->bcnt = rule->pcnt = 0;
2424                 rule->timestamp = 0;
2425         }
2426         if (l->o.opcode == O_LOG)
2427                 l->log_left = l->max_log;
2428 }
2429
2430 /**
2431  * Reset some or all counters on firewall rules.
2432  * @arg frwl is null to clear all entries, or contains a specific
2433  * rule number.
2434  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2435  */
2436 static int
2437 zero_entry(int rulenum, int log_only)
2438 {
2439         struct ip_fw *rule;
2440         char *msg;
2441
2442         if (rulenum == 0) {
2443                 crit_enter();
2444                 norule_counter = 0;
2445                 for (rule = layer3_chain; rule; rule = rule->next)
2446                         clear_counters(rule, log_only);
2447                 crit_exit();
2448                 msg = log_only ? "ipfw: All logging counts reset.\n"
2449                                : "ipfw: Accounting cleared.\n";
2450         } else {
2451                 int cleared = 0;
2452
2453                 /*
2454                  * We can have multiple rules with the same number, so we
2455                  * need to clear them all.
2456                  */
2457                 for (rule = layer3_chain; rule; rule = rule->next) {
2458                         if (rule->rulenum == rulenum) {
2459                                 crit_enter();
2460                                 while (rule && rule->rulenum == rulenum) {
2461                                         clear_counters(rule, log_only);
2462                                         rule = rule->next;
2463                                 }
2464                                 crit_exit();
2465                                 cleared = 1;
2466                                 break;
2467                         }
2468                 }
2469                 if (!cleared)   /* we did not find any matching rules */
2470                         return (EINVAL);
2471                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
2472                                : "ipfw: Entry %d cleared.\n";
2473         }
2474         if (fw_verbose)
2475                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2476         return (0);
2477 }
2478
2479 /*
2480  * Check validity of the structure before insert.
2481  * Fortunately rules are simple, so this mostly need to check rule sizes.
2482  */
2483 static int
2484 ipfw_ctl_check_rule(struct ipfw_ioc_rule *rule, int size)
2485 {
2486         int l, cmdlen = 0;
2487         int have_action = 0;
2488         ipfw_insn *cmd;
2489
2490         /* Check for valid size */
2491         if (size < sizeof(*rule)) {
2492                 kprintf("ipfw: rule too short\n");
2493                 return EINVAL;
2494         }
2495         l = IOC_RULESIZE(rule);
2496         if (l != size) {
2497                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
2498                 return EINVAL;
2499         }
2500
2501         /*
2502          * Now go for the individual checks. Very simple ones, basically only
2503          * instruction sizes.
2504          */
2505         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
2506              l -= cmdlen, cmd += cmdlen) {
2507                 cmdlen = F_LEN(cmd);
2508                 if (cmdlen > l) {
2509                         kprintf("ipfw: opcode %d size truncated\n",
2510                                 cmd->opcode);
2511                         return EINVAL;
2512                 }
2513                 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
2514                 switch (cmd->opcode) {
2515                 case O_NOP:
2516                 case O_PROBE_STATE:
2517                 case O_KEEP_STATE:
2518                 case O_PROTO:
2519                 case O_IP_SRC_ME:
2520                 case O_IP_DST_ME:
2521                 case O_LAYER2:
2522                 case O_IN:
2523                 case O_FRAG:
2524                 case O_IPOPT:
2525                 case O_IPLEN:
2526                 case O_IPID:
2527                 case O_IPTOS:
2528                 case O_IPPRECEDENCE:
2529                 case O_IPTTL:
2530                 case O_IPVER:
2531                 case O_TCPWIN:
2532                 case O_TCPFLAGS:
2533                 case O_TCPOPTS:
2534                 case O_ESTAB:
2535                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2536                                 goto bad_size;
2537                         break;
2538
2539                 case O_UID:
2540                 case O_GID:
2541                 case O_IP_SRC:
2542                 case O_IP_DST:
2543                 case O_TCPSEQ:
2544                 case O_TCPACK:
2545                 case O_PROB:
2546                 case O_ICMPTYPE:
2547                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2548                                 goto bad_size;
2549                         break;
2550
2551                 case O_LIMIT:
2552                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2553                                 goto bad_size;
2554                         break;
2555
2556                 case O_LOG:
2557                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2558                                 goto bad_size;
2559
2560                         ((ipfw_insn_log *)cmd)->log_left =
2561                             ((ipfw_insn_log *)cmd)->max_log;
2562
2563                         break;
2564
2565                 case O_IP_SRC_MASK:
2566                 case O_IP_DST_MASK:
2567                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2568                                 goto bad_size;
2569                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2570                                 kprintf("ipfw: opcode %d, useless rule\n",
2571                                         cmd->opcode);
2572                                 return EINVAL;
2573                         }
2574                         break;
2575
2576                 case O_IP_SRC_SET:
2577                 case O_IP_DST_SET:
2578                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2579                                 kprintf("ipfw: invalid set size %d\n",
2580                                         cmd->arg1);
2581                                 return EINVAL;
2582                         }
2583                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2584                             (cmd->arg1+31)/32 )
2585                                 goto bad_size;
2586                         break;
2587
2588                 case O_MACADDR2:
2589                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2590                                 goto bad_size;
2591                         break;
2592
2593                 case O_MAC_TYPE:
2594                 case O_IP_SRCPORT:
2595                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2596                         if (cmdlen < 2 || cmdlen > 31)
2597                                 goto bad_size;
2598                         break;
2599
2600                 case O_RECV:
2601                 case O_XMIT:
2602                 case O_VIA:
2603                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2604                                 goto bad_size;
2605                         break;
2606
2607                 case O_PIPE:
2608                 case O_QUEUE:
2609                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2610                                 goto bad_size;
2611                         goto check_action;
2612
2613                 case O_FORWARD_IP:
2614                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2615                                 goto bad_size;
2616                         goto check_action;
2617
2618                 case O_FORWARD_MAC: /* XXX not implemented yet */
2619                 case O_CHECK_STATE:
2620                 case O_COUNT:
2621                 case O_ACCEPT:
2622                 case O_DENY:
2623                 case O_REJECT:
2624                 case O_SKIPTO:
2625                 case O_DIVERT:
2626                 case O_TEE:
2627                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2628                                 goto bad_size;
2629 check_action:
2630                         if (have_action) {
2631                                 kprintf("ipfw: opcode %d, multiple actions"
2632                                         " not allowed\n",
2633                                         cmd->opcode);
2634                                 return EINVAL;
2635                         }
2636                         have_action = 1;
2637                         if (l != cmdlen) {
2638                                 kprintf("ipfw: opcode %d, action must be"
2639                                         " last opcode\n",
2640                                         cmd->opcode);
2641                                 return EINVAL;
2642                         }
2643                         break;
2644                 default:
2645                         kprintf("ipfw: opcode %d, unknown opcode\n",
2646                                 cmd->opcode);
2647                         return EINVAL;
2648                 }
2649         }
2650         if (have_action == 0) {
2651                 kprintf("ipfw: missing action\n");
2652                 return EINVAL;
2653         }
2654         return 0;
2655
2656 bad_size:
2657         kprintf("ipfw: opcode %d size %d wrong\n",
2658                 cmd->opcode, cmdlen);
2659         return EINVAL;
2660 }
2661
2662 static int
2663 ipfw_ctl_add_rule(struct sockopt *sopt)
2664 {
2665         struct ipfw_ioc_rule *ioc_rule;
2666         size_t size;
2667         int error;
2668
2669         size = sopt->sopt_valsize;
2670         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
2671             size < sizeof(*ioc_rule)) {
2672                 return EINVAL;
2673         }
2674         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
2675                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
2676                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
2677         }
2678         ioc_rule = sopt->sopt_val;
2679
2680         error = ipfw_ctl_check_rule(ioc_rule, size);
2681         if (error)
2682                 return error;
2683
2684         error = ipfw_add_rule(&layer3_chain, ioc_rule);
2685         if (error)
2686                 return error;
2687
2688         if (sopt->sopt_dir == SOPT_GET)
2689                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
2690         return 0;
2691 }
2692
2693 static void *
2694 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
2695 {
2696         ioc_rule->act_ofs = rule->act_ofs;
2697         ioc_rule->cmd_len = rule->cmd_len;
2698         ioc_rule->rulenum = rule->rulenum;
2699         ioc_rule->set = rule->set;
2700         ioc_rule->usr_flags = rule->usr_flags;
2701
2702         ioc_rule->set_disable = set_disable;
2703         ioc_rule->static_count = static_count;
2704         ioc_rule->static_len = static_ioc_len;
2705
2706         ioc_rule->pcnt = rule->pcnt;
2707         ioc_rule->bcnt = rule->bcnt;
2708         ioc_rule->timestamp = rule->timestamp;
2709
2710         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
2711
2712         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
2713 }
2714
2715 static void
2716 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
2717                 struct ipfw_ioc_state *ioc_state)
2718 {
2719         const struct ipfw_flow_id *id;
2720         struct ipfw_ioc_flowid *ioc_id;
2721
2722         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
2723                             0 : dyn_rule->expire - time_second;
2724         ioc_state->pcnt = dyn_rule->pcnt;
2725         ioc_state->bcnt = dyn_rule->bcnt;
2726
2727         ioc_state->dyn_type = dyn_rule->dyn_type;
2728         ioc_state->count = dyn_rule->count;
2729
2730         ioc_state->rulenum = dyn_rule->rule->rulenum;
2731
2732         id = &dyn_rule->id;
2733         ioc_id = &ioc_state->id;
2734
2735         ioc_id->type = ETHERTYPE_IP;
2736         ioc_id->u.ip.dst_ip = id->dst_ip;
2737         ioc_id->u.ip.src_ip = id->src_ip;
2738         ioc_id->u.ip.dst_port = id->dst_port;
2739         ioc_id->u.ip.src_port = id->src_port;
2740         ioc_id->u.ip.proto = id->proto;
2741 }
2742
2743 static int
2744 ipfw_ctl_get_rules(struct sockopt *sopt)
2745 {
2746         struct ip_fw *rule;
2747         void *bp;
2748         size_t size;
2749
2750         /*
2751          * pass up a copy of the current rules. Static rules
2752          * come first (the last of which has number IPFW_DEFAULT_RULE),
2753          * followed by a possibly empty list of dynamic rule.
2754          */
2755         crit_enter();
2756
2757         size = static_ioc_len;  /* size of static rules */
2758         if (ipfw_dyn_v)         /* add size of dyn.rules */
2759                 size += (dyn_count * sizeof(struct ipfw_ioc_state));
2760
2761         if (sopt->sopt_valsize < size) {
2762                 /* short length, no need to return incomplete rules */
2763                 /* XXX: if superuser, no need to zero buffer */
2764                 bzero(sopt->sopt_val, sopt->sopt_valsize);
2765                 return 0;
2766         }
2767         bp = sopt->sopt_val;
2768
2769         for (rule = layer3_chain; rule; rule = rule->next)
2770                 bp = ipfw_copy_rule(rule, bp);
2771
2772         if (ipfw_dyn_v) {
2773                 struct ipfw_ioc_state *ioc_state;
2774                 int i;
2775
2776                 ioc_state = bp;
2777                 for (i = 0; i < curr_dyn_buckets; i++) {
2778                         ipfw_dyn_rule *p;
2779
2780                         for (p = ipfw_dyn_v[i]; p != NULL;
2781                              p = p->next, ioc_state++)
2782                                 ipfw_copy_state(p, ioc_state);
2783                 }
2784         }
2785
2786         crit_exit();
2787
2788         sopt->sopt_valsize = size;
2789         return 0;
2790 }
2791
2792 /**
2793  * {set|get}sockopt parser.
2794  */
2795 static int
2796 ipfw_ctl(struct sockopt *sopt)
2797 {
2798         int error, rulenum;
2799         uint32_t *masks;
2800         size_t size;
2801
2802         error = 0;
2803
2804         switch (sopt->sopt_name) {
2805         case IP_FW_GET:
2806                 error = ipfw_ctl_get_rules(sopt);
2807                 break;
2808
2809         case IP_FW_FLUSH:
2810                 /*
2811                  * Normally we cannot release the lock on each iteration.
2812                  * We could do it here only because we start from the head all
2813                  * the times so there is no risk of missing some entries.
2814                  * On the other hand, the risk is that we end up with
2815                  * a very inconsistent ruleset, so better keep the lock
2816                  * around the whole cycle.
2817                  *
2818                  * XXX this code can be improved by resetting the head of
2819                  * the list to point to the default rule, and then freeing
2820                  * the old list without the need for a lock.
2821                  */
2822
2823                 crit_enter();
2824                 free_chain(&layer3_chain, 0 /* keep default rule */);
2825                 crit_exit();
2826                 break;
2827
2828         case IP_FW_ADD:
2829                 error = ipfw_ctl_add_rule(sopt);
2830                 break;
2831
2832         case IP_FW_DEL:
2833                 /*
2834                  * IP_FW_DEL is used for deleting single rules or sets,
2835                  * and (ab)used to atomically manipulate sets. Argument size
2836                  * is used to distinguish between the two:
2837                  *    sizeof(uint32_t)
2838                  *      delete single rule or set of rules,
2839                  *      or reassign rules (or sets) to a different set.
2840                  *    2*sizeof(uint32_t)
2841                  *      atomic disable/enable sets.
2842                  *      first uint32_t contains sets to be disabled,
2843                  *      second uint32_t contains sets to be enabled.
2844                  */
2845                 masks = sopt->sopt_val;
2846                 size = sopt->sopt_valsize;
2847                 if (size == sizeof(*masks)) {
2848                         /*
2849                          * Delete or reassign static rule
2850                          */
2851                         error = del_entry(&layer3_chain, masks[0]);
2852                 } else if (size == (2 * sizeof(*masks))) {
2853                         /*
2854                          * Set enable/disable
2855                          */
2856                         crit_enter();
2857
2858                         set_disable =
2859                             (set_disable | masks[0]) & ~masks[1] &
2860                             ~(1 << 31); /* set 31 always enabled */
2861
2862                         crit_exit();
2863                 } else {
2864                         error = EINVAL;
2865                 }
2866                 break;
2867
2868         case IP_FW_ZERO:
2869         case IP_FW_RESETLOG: /* argument is an int, the rule number */
2870                 rulenum=0;
2871
2872                 if (sopt->sopt_val != 0) {
2873                     error = soopt_to_kbuf(sopt, &rulenum,
2874                             sizeof(int), sizeof(int));
2875                     if (error)
2876                         break;
2877                 }
2878                 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2879                 break;
2880
2881         default:
2882                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
2883                 error = EINVAL;
2884         }
2885         return error;
2886 }
2887
2888 /*
2889  * This procedure is only used to handle keepalives. It is invoked
2890  * every dyn_keepalive_period
2891  */
2892 static void
2893 ipfw_tick(void *unused __unused)
2894 {
2895         int i;
2896         ipfw_dyn_rule *q;
2897
2898         if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2899                 goto done;
2900
2901         crit_enter();
2902         for (i = 0; i < curr_dyn_buckets; i++) {
2903                 for (q = ipfw_dyn_v[i]; q; q = q->next) {
2904                         if (q->dyn_type == O_LIMIT_PARENT)
2905                                 continue;
2906                         if (q->id.proto != IPPROTO_TCP)
2907                                 continue;
2908                         if ((q->state & BOTH_SYN) != BOTH_SYN)
2909                                 continue;
2910                         if (TIME_LEQ(time_second + dyn_keepalive_interval,
2911                             q->expire))
2912                                 continue;       /* too early */
2913                         if (TIME_LEQ(q->expire, time_second))
2914                                 continue;       /* too late, rule expired */
2915
2916                         send_pkt(&q->id, q->ack_rev - 1, q->ack_fwd, TH_SYN);
2917                         send_pkt(&q->id, q->ack_fwd - 1, q->ack_rev, 0);
2918                 }
2919         }
2920         crit_exit();
2921 done:
2922         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
2923                       ipfw_tick, NULL);
2924 }
2925
2926 static void
2927 ipfw_init_default_rule(struct ip_fw **head)
2928 {
2929         struct ip_fw *def_rule;
2930
2931         KKASSERT(*head == NULL);
2932
2933         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
2934
2935         def_rule->act_ofs = 0;
2936         def_rule->rulenum = IPFW_DEFAULT_RULE;
2937         def_rule->cmd_len = 1;
2938         def_rule->set = 31;
2939
2940         def_rule->cmd[0].len = 1;
2941 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2942         def_rule->cmd[0].opcode = O_ACCEPT;
2943 #else
2944         def_rule->cmd[0].opcode = O_DENY;
2945 #endif
2946
2947         def_rule->refcnt = 1;
2948
2949         *head = def_rule;
2950         ipfw_inc_static_count(def_rule);
2951
2952         /* Install the default rule */
2953         ip_fw_default_rule = def_rule;
2954 }
2955
2956 static void
2957 ipfw_init_dispatch(struct netmsg *nmsg)
2958 {
2959         int error = 0;
2960
2961         crit_enter();
2962
2963         if (IPFW_LOADED) {
2964                 kprintf("IP firewall already loaded\n");
2965                 error = EEXIST;
2966                 goto reply;
2967         }
2968
2969         ip_fw_chk_ptr = ipfw_chk;
2970         ip_fw_ctl_ptr = ipfw_ctl;
2971         ip_fw_dn_io_ptr = ipfw_dummynet_io;
2972
2973         layer3_chain = NULL;
2974         ipfw_init_default_rule(&layer3_chain);
2975
2976         kprintf("ipfw2 initialized, divert %s, "
2977                 "rule-based forwarding enabled, default to %s, logging ",
2978 #ifdef IPDIVERT
2979                 "enabled",
2980 #else
2981                 "disabled",
2982 #endif
2983                 ip_fw_default_rule->cmd[0].opcode == O_ACCEPT ?
2984                 "accept" : "deny");
2985
2986 #ifdef IPFIREWALL_VERBOSE
2987         fw_verbose = 1;
2988 #endif
2989 #ifdef IPFIREWALL_VERBOSE_LIMIT
2990         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2991 #endif
2992         if (fw_verbose == 0) {
2993                 kprintf("disabled\n");
2994         } else if (verbose_limit == 0) {
2995                 kprintf("unlimited\n");
2996         } else {
2997                 kprintf("limited to %d packets/entry by default\n",
2998                         verbose_limit);
2999         }
3000         callout_init(&ipfw_timeout_h);
3001
3002         ip_fw_loaded = 1;
3003         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
3004 reply:
3005         crit_exit();
3006         lwkt_replymsg(&nmsg->nm_lmsg, error);
3007 }
3008
3009 static int
3010 ipfw_init(void)
3011 {
3012         struct netmsg smsg;
3013
3014         netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_init_dispatch);
3015         return lwkt_domsg(cpu_portfn(0), &smsg.nm_lmsg, 0);
3016 }
3017
3018 #ifdef KLD_MODULE
3019
3020 static void
3021 ipfw_fini_dispatch(struct netmsg *nmsg)
3022 {
3023         int error = 0;
3024
3025         crit_enter();
3026
3027         if (ipfw_refcnt != 0) {
3028                 error = EBUSY;
3029                 goto reply;
3030         }
3031
3032         callout_stop(&ipfw_timeout_h);
3033
3034         ip_fw_loaded = 0;
3035         netmsg_service_sync();
3036
3037         ip_fw_chk_ptr = NULL;
3038         ip_fw_ctl_ptr = NULL;
3039         ip_fw_dn_io_ptr = NULL;
3040         free_chain(&layer3_chain, 1 /* kill default rule */);
3041
3042         kprintf("IP firewall unloaded\n");
3043 reply:
3044         crit_exit();
3045         lwkt_replymsg(&nmsg->nm_lmsg, error);
3046 }
3047
3048 static int
3049 ipfw_fini(void)
3050 {
3051         struct netmsg smsg;
3052
3053         netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_fini_dispatch);
3054         return lwkt_domsg(cpu_portfn(0), &smsg.nm_lmsg, 0);
3055 }
3056
3057 #endif  /* KLD_MODULE */
3058
3059 static int
3060 ipfw_modevent(module_t mod, int type, void *unused)
3061 {
3062         int err = 0;
3063
3064         switch (type) {
3065         case MOD_LOAD:
3066                 err = ipfw_init();
3067                 break;
3068
3069         case MOD_UNLOAD:
3070 #ifndef KLD_MODULE
3071                 kprintf("ipfw statically compiled, cannot unload\n");
3072                 err = EBUSY;
3073 #else
3074                 err = ipfw_fini();
3075 #endif
3076                 break;
3077         default:
3078                 break;
3079         }
3080         return err;
3081 }
3082
3083 static moduledata_t ipfwmod = {
3084         "ipfw",
3085         ipfw_modevent,
3086         0
3087 };
3088 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
3089 MODULE_VERSION(ipfw, 1);