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 1.41 2008/03/07 11:34:20 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 #include <net/if.h>
  60 #include <net/route.h>
  61 #include <netinet/in.h>
  62 #include <netinet/in_systm.h>
  63 #include <netinet/in_var.h>
  64 #include <netinet/in_pcb.h>
  65 #include <netinet/ip.h>
  66 #include <netinet/ip_var.h>
  67 #include <netinet/ip_icmp.h>
  68 #include "ip_fw.h"
  69 #include <net/dummynet/ip_dummynet.h>
  70 #include <netinet/tcp.h>
  71 #include <netinet/tcp_timer.h>
  72 #include <netinet/tcp_var.h>
  73 #include <netinet/tcpip.h>
  74 #include <netinet/udp.h>
  75 #include <netinet/udp_var.h>
  76
  77 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  78
  79 /*
  80  * set_disable contains one bit per set value (0..31).
  81  * If the bit is set, all rules with the corresponding set
  82  * are disabled. Set 31 is reserved for the default rule
  83  * and CANNOT be disabled.
  84  */
  85 static uint32_t set_disable;
  86
  87 static int fw_verbose;
  88 static int verbose_limit;
  89
  90 #ifdef KLD_MODULE
  91 static int ipfw_refcnt;
  92 #endif
  93
  94 static struct callout ipfw_timeout_h;
  95 #define IPFW_DEFAULT_RULE       65535
  96
  97 /*
  98  * list of rules for layer 3
  99  */
 100 static struct ip_fw *layer3_chain;
 101
 102 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 103
 104 static int fw_debug = 1;
 105 static int autoinc_step = 100; /* bounded to 1..1000 in ipfw_add_rule() */
 106
 107 #ifdef SYSCTL_NODE
 108 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 109 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
 110     &fw_enable, 0, "Enable ipfw");
 111 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
 112     &autoinc_step, 0, "Rule number autincrement step");
 113 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 114     &fw_one_pass, 0,
 115     "Only do a single pass through ipfw when using dummynet(4)");
 116 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 117     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 118 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 119     &fw_verbose, 0, "Log matches to ipfw rules");
 120 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 121     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 122
 123 /*
 124  * Description of dynamic rules.
 125  *
 126  * Dynamic rules are stored in lists accessed through a hash table
 127  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 128  * be modified through the sysctl variable dyn_buckets which is
 129  * updated when the table becomes empty.
 130  *
 131  * XXX currently there is only one list, ipfw_dyn.
 132  *
 133  * When a packet is received, its address fields are first masked
 134  * with the mask defined for the rule, then hashed, then matched
 135  * against the entries in the corresponding list.
 136  * Dynamic rules can be used for different purposes:
 137  *  + stateful rules;
 138  *  + enforcing limits on the number of sessions;
 139  *  + in-kernel NAT (not implemented yet)
 140  *
 141  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 142  * measured in seconds and depending on the flags.
 143  *
 144  * The total number of dynamic rules is stored in dyn_count.
 145  * The max number of dynamic rules is dyn_max. When we reach
 146  * the maximum number of rules we do not create anymore. This is
 147  * done to avoid consuming too much memory, but also too much
 148  * time when searching on each packet (ideally, we should try instead
 149  * to put a limit on the length of the list on each bucket...).
 150  *
 151  * Each dynamic rule holds a pointer to the parent ipfw rule so
 152  * we know what action to perform. Dynamic rules are removed when
 153  * the parent rule is deleted. XXX we should make them survive.
 154  *
 155  * There are some limitations with dynamic rules -- we do not
 156  * obey the 'randomized match', and we do not do multiple
 157  * passes through the firewall. XXX check the latter!!!
 158  */
 159 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 160 static uint32_t dyn_buckets = 256; /* must be power of 2 */
 161 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
 162
 163 /*
 164  * Timeouts for various events in handing dynamic rules.
 165  */
 166 static uint32_t dyn_ack_lifetime = 300;
 167 static uint32_t dyn_syn_lifetime = 20;
 168 static uint32_t dyn_fin_lifetime = 1;
 169 static uint32_t dyn_rst_lifetime = 1;
 170 static uint32_t dyn_udp_lifetime = 10;
 171 static uint32_t dyn_short_lifetime = 5;
 172
 173 /*
 174  * Keepalives are sent if dyn_keepalive is set. They are sent every
 175  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 176  * seconds of lifetime of a rule.
 177  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 178  * than dyn_keepalive_period.
 179  */
 180
 181 static uint32_t dyn_keepalive_interval = 20;
 182 static uint32_t dyn_keepalive_period = 5;
 183 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
 184
 185 static uint32_t static_count;           /* # of static rules */
 186 static uint32_t static_ioc_len; /* bytes of static rules */
 187 static uint32_t dyn_count;              /* # of dynamic rules */
 188 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
 189
 190 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
 191     &dyn_buckets, 0, "Number of dyn. buckets");
 192 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 193     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 194 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 195     &dyn_count, 0, "Number of dyn. rules");
 196 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 197     &dyn_max, 0, "Max number of dyn. rules");
 198 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 199     &static_count, 0, "Number of static rules");
 200 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 201     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 202 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 203     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 204 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
 205     &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
 206 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
 207     &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
 208 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 209     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 210 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 211     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 212 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 213     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 214
 215 #endif /* SYSCTL_NODE */
 216
 217 /**
 218  * dummynet needs a reference to the default rule, because rules can be
 219  * deleted while packets hold a reference to them. When this happens,
 220  * dummynet changes the reference to the default rule (it could well be a
 221  * NULL pointer, but this way we do not need to check for the special
 222  * case, plus here he have info on the default behaviour).
 223  */
 224 struct ip_fw *ip_fw_default_rule;
 225
 226 static ip_fw_chk_t      ipfw_chk;
 227
 228 static __inline int
 229 ipfw_free_rule(struct ip_fw *rule)
 230 {
 231         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
 232         atomic_subtract_int(&rule->refcnt, 1);
 233         if (atomic_cmpset_int(&rule->refcnt, 0, 1)) {
 234                 kfree(rule, M_IPFW);
 235                 return 1;
 236         }
 237         return 0;
 238 }
 239
 240 static void
 241 ipfw_unref_rule(void *priv)
 242 {
 243         ipfw_free_rule(priv);
 244 #ifdef KLD_MODULE
 245         atomic_subtract_int(&ipfw_refcnt, 1);
 246 #endif
 247 }
 248
 249 static __inline void
 250 ipfw_ref_rule(struct ip_fw *rule)
 251 {
 252 #ifdef KLD_MODULE
 253         atomic_add_int(&ipfw_refcnt, 1);
 254 #endif
 255         atomic_add_int(&rule->refcnt, 1);
 256 }
 257
 258 /*
 259  * This macro maps an ip pointer into a layer3 header pointer of type T
 260  */
 261 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
 262
 263 static __inline int
 264 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 265 {
 266         int type = L3HDR(struct icmp,ip)->icmp_type;
 267
 268         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
 269 }
 270
 271 #define TT      ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
 272     (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
 273
 274 static int
 275 is_icmp_query(struct ip *ip)
 276 {
 277         int type = L3HDR(struct icmp, ip)->icmp_type;
 278         return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
 279 }
 280 #undef TT
 281
 282 /*
 283  * The following checks use two arrays of 8 or 16 bits to store the
 284  * bits that we want set or clear, respectively. They are in the
 285  * low and high half of cmd->arg1 or cmd->d[0].
 286  *
 287  * We scan options and store the bits we find set. We succeed if
 288  *
 289  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 290  *
 291  * The code is sometimes optimized not to store additional variables.
 292  */
 293
 294 static int
 295 flags_match(ipfw_insn *cmd, uint8_t bits)
 296 {
 297         u_char want_clear;
 298         bits = ~bits;
 299
 300         if ( ((cmd->arg1 & 0xff) & bits) != 0)
 301                 return 0; /* some bits we want set were clear */
 302         want_clear = (cmd->arg1 >> 8) & 0xff;
 303         if ( (want_clear & bits) != want_clear)
 304                 return 0; /* some bits we want clear were set */
 305         return 1;
 306 }
 307
 308 static int
 309 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 310 {
 311         int optlen, bits = 0;
 312         u_char *cp = (u_char *)(ip + 1);
 313         int x = (ip->ip_hl << 2) - sizeof (struct ip);
 314
 315         for (; x > 0; x -= optlen, cp += optlen) {
 316                 int opt = cp[IPOPT_OPTVAL];
 317
 318                 if (opt == IPOPT_EOL)
 319                         break;
 320                 if (opt == IPOPT_NOP)
 321                         optlen = 1;
 322                 else {
 323                         optlen = cp[IPOPT_OLEN];
 324                         if (optlen <= 0 || optlen > x)
 325                                 return 0; /* invalid or truncated */
 326                 }
 327                 switch (opt) {
 328
 329                 default:
 330                         break;
 331
 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         }
 349         return (flags_match(cmd, bits));
 350 }
 351
 352 static int
 353 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 354 {
 355         int optlen, bits = 0;
 356         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 357         u_char *cp = (u_char *)(tcp + 1);
 358         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 359
 360         for (; x > 0; x -= optlen, cp += optlen) {
 361                 int opt = cp[0];
 362                 if (opt == TCPOPT_EOL)
 363                         break;
 364                 if (opt == TCPOPT_NOP)
 365                         optlen = 1;
 366                 else {
 367                         optlen = cp[1];
 368                         if (optlen <= 0)
 369                                 break;
 370                 }
 371
 372                 switch (opt) {
 373
 374                 default:
 375                         break;
 376
 377                 case TCPOPT_MAXSEG:
 378                         bits |= IP_FW_TCPOPT_MSS;
 379                         break;
 380
 381                 case TCPOPT_WINDOW:
 382                         bits |= IP_FW_TCPOPT_WINDOW;
 383                         break;
 384
 385                 case TCPOPT_SACK_PERMITTED:
 386                 case TCPOPT_SACK:
 387                         bits |= IP_FW_TCPOPT_SACK;
 388                         break;
 389
 390                 case TCPOPT_TIMESTAMP:
 391                         bits |= IP_FW_TCPOPT_TS;
 392                         break;
 393
 394                 case TCPOPT_CC:
 395                 case TCPOPT_CCNEW:
 396                 case TCPOPT_CCECHO:
 397                         bits |= IP_FW_TCPOPT_CC;
 398                         break;
 399                 }
 400         }
 401         return (flags_match(cmd, bits));
 402 }
 403
 404 static int
 405 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 406 {
 407         if (ifp == NULL)        /* no iface with this packet, match fails */
 408                 return 0;
 409         /* Check by name or by IP address */
 410         if (cmd->name[0] != '\0') { /* match by name */
 411                 /* Check name */
 412                 if (cmd->p.glob) {
 413                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 414                                 return(1);
 415                 } else {
 416                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 417                                 return(1);
 418                 }
 419         } else {
 420                 struct ifaddr_container *ifac;
 421
 422                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 423                         struct ifaddr *ia = ifac->ifa;
 424
 425                         if (ia->ifa_addr == NULL)
 426                                 continue;
 427                         if (ia->ifa_addr->sa_family != AF_INET)
 428                                 continue;
 429                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 430                             (ia->ifa_addr))->sin_addr.s_addr)
 431                                 return(1);      /* match */
 432                 }
 433         }
 434         return(0);      /* no match, fail ... */
 435 }
 436
 437 static uint64_t norule_counter; /* counter for ipfw_log(NULL...) */
 438
 439 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 440 #define SNP(buf) buf, sizeof(buf)
 441
 442 /*
 443  * We enter here when we have a rule with O_LOG.
 444  * XXX this function alone takes about 2Kbytes of code!
 445  */
 446 static void
 447 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 448         struct mbuf *m, struct ifnet *oif)
 449 {
 450         char *action;
 451         int limit_reached = 0;
 452         char action2[40], proto[48], fragment[28];
 453
 454         fragment[0] = '\0';
 455         proto[0] = '\0';
 456
 457         if (f == NULL) {        /* bogus pkt */
 458                 if (verbose_limit != 0 && norule_counter >= verbose_limit)
 459                         return;
 460                 norule_counter++;
 461                 if (norule_counter == verbose_limit)
 462                         limit_reached = verbose_limit;
 463                 action = "Refuse";
 464         } else {        /* O_LOG is the first action, find the real one */
 465                 ipfw_insn *cmd = ACTION_PTR(f);
 466                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 467
 468                 if (l->max_log != 0 && l->log_left == 0)
 469                         return;
 470                 l->log_left--;
 471                 if (l->log_left == 0)
 472                         limit_reached = l->max_log;
 473                 cmd += F_LEN(cmd);      /* point to first action */
 474                 if (cmd->opcode == O_PROB)
 475                         cmd += F_LEN(cmd);
 476
 477                 action = action2;
 478                 switch (cmd->opcode) {
 479                 case O_DENY:
 480                         action = "Deny";
 481                         break;
 482
 483                 case O_REJECT:
 484                         if (cmd->arg1==ICMP_REJECT_RST)
 485                                 action = "Reset";
 486                         else if (cmd->arg1==ICMP_UNREACH_HOST)
 487                                 action = "Reject";
 488                         else
 489                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 490                                         cmd->arg1);
 491                         break;
 492
 493                 case O_ACCEPT:
 494                         action = "Accept";
 495                         break;
 496                 case O_COUNT:
 497                         action = "Count";
 498                         break;
 499                 case O_DIVERT:
 500                         ksnprintf(SNPARGS(action2, 0), "Divert %d",
 501                                 cmd->arg1);
 502                         break;
 503                 case O_TEE:
 504                         ksnprintf(SNPARGS(action2, 0), "Tee %d",
 505                                 cmd->arg1);
 506                         break;
 507                 case O_SKIPTO:
 508                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d",
 509                                 cmd->arg1);
 510                         break;
 511                 case O_PIPE:
 512                         ksnprintf(SNPARGS(action2, 0), "Pipe %d",
 513                                 cmd->arg1);
 514                         break;
 515                 case O_QUEUE:
 516                         ksnprintf(SNPARGS(action2, 0), "Queue %d",
 517                                 cmd->arg1);
 518                         break;
 519                 case O_FORWARD_IP: {
 520                         ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 521                         int len;
 522
 523                         len = ksnprintf(SNPARGS(action2, 0), "Forward to %s",
 524                                 inet_ntoa(sa->sa.sin_addr));
 525                         if (sa->sa.sin_port)
 526                                 ksnprintf(SNPARGS(action2, len), ":%d",
 527                                     sa->sa.sin_port);
 528                         }
 529                         break;
 530                 default:
 531                         action = "UNKNOWN";
 532                         break;
 533                 }
 534         }
 535
 536         if (hlen == 0) {        /* non-ip */
 537                 ksnprintf(SNPARGS(proto, 0), "MAC");
 538         } else {
 539                 struct ip *ip = mtod(m, struct ip *);
 540                 /* these three are all aliases to the same thing */
 541                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 542                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 543                 struct udphdr *const udp = (struct udphdr *)icmp;
 544
 545                 int ip_off, offset, ip_len;
 546
 547                 int len;
 548
 549                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 550                         ip_off = ntohs(ip->ip_off);
 551                         ip_len = ntohs(ip->ip_len);
 552                 } else {
 553                         ip_off = ip->ip_off;
 554                         ip_len = ip->ip_len;
 555                 }
 556                 offset = ip_off & IP_OFFMASK;
 557                 switch (ip->ip_p) {
 558                 case IPPROTO_TCP:
 559                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 560                             inet_ntoa(ip->ip_src));
 561                         if (offset == 0)
 562                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 563                                     ntohs(tcp->th_sport),
 564                                     inet_ntoa(ip->ip_dst),
 565                                     ntohs(tcp->th_dport));
 566                         else
 567                                 ksnprintf(SNPARGS(proto, len), " %s",
 568                                     inet_ntoa(ip->ip_dst));
 569                         break;
 570
 571                 case IPPROTO_UDP:
 572                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 573                                 inet_ntoa(ip->ip_src));
 574                         if (offset == 0)
 575                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 576                                     ntohs(udp->uh_sport),
 577                                     inet_ntoa(ip->ip_dst),
 578                                     ntohs(udp->uh_dport));
 579                         else
 580                                 ksnprintf(SNPARGS(proto, len), " %s",
 581                                     inet_ntoa(ip->ip_dst));
 582                         break;
 583
 584                 case IPPROTO_ICMP:
 585                         if (offset == 0)
 586                                 len = ksnprintf(SNPARGS(proto, 0),
 587                                     "ICMP:%u.%u ",
 588                                     icmp->icmp_type, icmp->icmp_code);
 589                         else
 590                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 591                         len += ksnprintf(SNPARGS(proto, len), "%s",
 592                             inet_ntoa(ip->ip_src));
 593                         ksnprintf(SNPARGS(proto, len), " %s",
 594                             inet_ntoa(ip->ip_dst));
 595                         break;
 596
 597                 default:
 598                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 599                             inet_ntoa(ip->ip_src));
 600                         ksnprintf(SNPARGS(proto, len), " %s",
 601                             inet_ntoa(ip->ip_dst));
 602                         break;
 603                 }
 604
 605                 if (ip_off & (IP_MF | IP_OFFMASK))
 606                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 607                              ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 608                              offset << 3,
 609                              (ip_off & IP_MF) ? "+" : "");
 610         }
 611         if (oif || m->m_pkthdr.rcvif)
 612                 log(LOG_SECURITY | LOG_INFO,
 613                     "ipfw: %d %s %s %s via %s%s\n",
 614                     f ? f->rulenum : -1,
 615                     action, proto, oif ? "out" : "in",
 616                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 617                     fragment);
 618         else
 619                 log(LOG_SECURITY | LOG_INFO,
 620                     "ipfw: %d %s %s [no if info]%s\n",
 621                     f ? f->rulenum : -1,
 622                     action, proto, fragment);
 623         if (limit_reached)
 624                 log(LOG_SECURITY | LOG_NOTICE,
 625                     "ipfw: limit %d reached on entry %d\n",
 626                     limit_reached, f ? f->rulenum : -1);
 627 }
 628
 629 /*
 630  * IMPORTANT: the hash function for dynamic rules must be commutative
 631  * in source and destination (ip,port), because rules are bidirectional
 632  * and we want to find both in the same bucket.
 633  */
 634 static __inline int
 635 hash_packet(struct ipfw_flow_id *id)
 636 {
 637         uint32_t i;
 638
 639         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 640         i &= (curr_dyn_buckets - 1);
 641         return i;
 642 }
 643
 644 /**
 645  * unlink a dynamic rule from a chain. prev is a pointer to
 646  * the previous one, q is a pointer to the rule to delete,
 647  * head is a pointer to the head of the queue.
 648  * Modifies q and potentially also head.
 649  */
 650 #define UNLINK_DYN_RULE(prev, head, q) {                                \
 651         ipfw_dyn_rule *old_q = q;                                       \
 652                                                                         \
 653         /* remove a refcount to the parent */                           \
 654         if (q->dyn_type == O_LIMIT)                                     \
 655                 q->parent->count--;                                     \
 656         DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",        \
 657                 (q->id.src_ip), (q->id.src_port),                       \
 658                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
 659         if (prev != NULL)                                               \
 660                 prev->next = q = q->next;                               \
 661         else                                                            \
 662                 head = q = q->next;                                     \
 663         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
 664         dyn_count--;                                                    \
 665         kfree(old_q, M_IPFW); }
 666
 667 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
 668
 669 /**
 670  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 671  *
 672  * If keep_me == NULL, rules are deleted even if not expired,
 673  * otherwise only expired rules are removed.
 674  *
 675  * The value of the second parameter is also used to point to identify
 676  * a rule we absolutely do not want to remove (e.g. because we are
 677  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 678  * rules). The pointer is only used for comparison, so any non-null
 679  * value will do.
 680  */
 681 static void
 682 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 683 {
 684         static uint32_t last_remove = 0;
 685
 686 #define FORCE (keep_me == NULL)
 687
 688         ipfw_dyn_rule *prev, *q;
 689         int i, pass = 0, max_pass = 0;
 690
 691         if (ipfw_dyn_v == NULL || dyn_count == 0)
 692                 return;
 693         /* do not expire more than once per second, it is useless */
 694         if (!FORCE && last_remove == time_second)
 695                 return;
 696         last_remove = time_second;
 697
 698         /*
 699          * because O_LIMIT refer to parent rules, during the first pass only
 700          * remove child and mark any pending LIMIT_PARENT, and remove
 701          * them in a second pass.
 702          */
 703 next_pass:
 704         for (i = 0 ; i < curr_dyn_buckets ; i++) {
 705                 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
 706                         /*
 707                          * Logic can become complex here, so we split tests.
 708                          */
 709                         if (q == keep_me)
 710                                 goto next;
 711                         if (rule != NULL && rule != q->rule)
 712                                 goto next; /* not the one we are looking for */
 713                         if (q->dyn_type == O_LIMIT_PARENT) {
 714                                 /*
 715                                  * handle parent in the second pass,
 716                                  * record we need one.
 717                                  */
 718                                 max_pass = 1;
 719                                 if (pass == 0)
 720                                         goto next;
 721                                 if (FORCE && q->count != 0 ) {
 722                                         /* XXX should not happen! */
 723                                         kprintf( "OUCH! cannot remove rule,"
 724                                              " count %d\n", q->count);
 725                                 }
 726                         } else {
 727                                 if (!FORCE &&
 728                                     !TIME_LEQ( q->expire, time_second ))
 729                                         goto next;
 730                         }
 731                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 732                         continue;
 733 next:
 734                         prev=q;
 735                         q=q->next;
 736                 }
 737         }
 738         if (pass++ < max_pass)
 739                 goto next_pass;
 740 }
 741
 742
 743 /**
 744  * lookup a dynamic rule.
 745  */
 746 static ipfw_dyn_rule *
 747 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 748         struct tcphdr *tcp)
 749 {
 750         /*
 751          * stateful ipfw extensions.
 752          * Lookup into dynamic session queue
 753          */
 754 #define MATCH_REVERSE   0
 755 #define MATCH_FORWARD   1
 756 #define MATCH_NONE      2
 757 #define MATCH_UNKNOWN   3
 758         int i, dir = MATCH_NONE;
 759         ipfw_dyn_rule *prev, *q=NULL;
 760
 761         if (ipfw_dyn_v == NULL)
 762                 goto done;      /* not found */
 763         i = hash_packet( pkt );
 764         for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
 765                 if (q->dyn_type == O_LIMIT_PARENT)
 766                         goto next;
 767                 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
 768                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 769                         continue;
 770                 }
 771                 if ( pkt->proto == q->id.proto) {
 772                         if (pkt->src_ip == q->id.src_ip &&
 773                             pkt->dst_ip == q->id.dst_ip &&
 774                             pkt->src_port == q->id.src_port &&
 775                             pkt->dst_port == q->id.dst_port ) {
 776                                 dir = MATCH_FORWARD;
 777                                 break;
 778                         }
 779                         if (pkt->src_ip == q->id.dst_ip &&
 780                             pkt->dst_ip == q->id.src_ip &&
 781                             pkt->src_port == q->id.dst_port &&
 782                             pkt->dst_port == q->id.src_port ) {
 783                                 dir = MATCH_REVERSE;
 784                                 break;
 785                         }
 786                 }
 787 next:
 788                 prev = q;
 789                 q = q->next;
 790         }
 791         if (q == NULL)
 792                 goto done; /* q = NULL, not found */
 793
 794         if ( prev != NULL) { /* found and not in front */
 795                 prev->next = q->next;
 796                 q->next = ipfw_dyn_v[i];
 797                 ipfw_dyn_v[i] = q;
 798         }
 799         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 800                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
 801
 802 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 803 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 804                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
 805                 switch (q->state) {
 806                 case TH_SYN:                            /* opening */
 807                         q->expire = time_second + dyn_syn_lifetime;
 808                         break;
 809
 810                 case BOTH_SYN:                  /* move to established */
 811                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
 812                 case BOTH_SYN | (TH_FIN << 8) :
 813                         if (tcp) {
 814 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
 815                             uint32_t ack = ntohl(tcp->th_ack);
 816                             if (dir == MATCH_FORWARD) {
 817                                 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
 818                                     q->ack_fwd = ack;
 819                                 else { /* ignore out-of-sequence */
 820                                     break;
 821                                 }
 822                             } else {
 823                                 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
 824                                     q->ack_rev = ack;
 825                                 else { /* ignore out-of-sequence */
 826                                     break;
 827                                 }
 828                             }
 829                         }
 830                         q->expire = time_second + dyn_ack_lifetime;
 831                         break;
 832
 833                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 834                         if (dyn_fin_lifetime >= dyn_keepalive_period)
 835                                 dyn_fin_lifetime = dyn_keepalive_period - 1;
 836                         q->expire = time_second + dyn_fin_lifetime;
 837                         break;
 838
 839                 default:
 840 #if 0
 841                         /*
 842                          * reset or some invalid combination, but can also
 843                          * occur if we use keep-state the wrong way.
 844                          */
 845                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
 846                                 kprintf("invalid state: 0x%x\n", q->state);
 847 #endif
 848                         if (dyn_rst_lifetime >= dyn_keepalive_period)
 849                                 dyn_rst_lifetime = dyn_keepalive_period - 1;
 850                         q->expire = time_second + dyn_rst_lifetime;
 851                         break;
 852                 }
 853         } else if (pkt->proto == IPPROTO_UDP) {
 854                 q->expire = time_second + dyn_udp_lifetime;
 855         } else {
 856                 /* other protocols */
 857                 q->expire = time_second + dyn_short_lifetime;
 858         }
 859 done:
 860         if (match_direction)
 861                 *match_direction = dir;
 862         return q;
 863 }
 864
 865 static void
 866 realloc_dynamic_table(void)
 867 {
 868         /*
 869          * Try reallocation, make sure we have a power of 2 and do
 870          * not allow more than 64k entries. In case of overflow,
 871          * default to 1024.
 872          */
 873
 874         if (dyn_buckets > 65536)
 875                 dyn_buckets = 1024;
 876         if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
 877                 dyn_buckets = curr_dyn_buckets; /* reset */
 878                 return;
 879         }
 880         curr_dyn_buckets = dyn_buckets;
 881         if (ipfw_dyn_v != NULL)
 882                 kfree(ipfw_dyn_v, M_IPFW);
 883         for (;;) {
 884                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
 885                        M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
 886                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
 887                         break;
 888                 curr_dyn_buckets /= 2;
 889         }
 890 }
 891
 892 /**
 893  * Install state of type 'type' for a dynamic session.
 894  * The hash table contains two type of rules:
 895  * - regular rules (O_KEEP_STATE)
 896  * - rules for sessions with limited number of sess per user
 897  *   (O_LIMIT). When they are created, the parent is
 898  *   increased by 1, and decreased on delete. In this case,
 899  *   the third parameter is the parent rule and not the chain.
 900  * - "parent" rules for the above (O_LIMIT_PARENT).
 901  */
 902 static ipfw_dyn_rule *
 903 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
 904 {
 905         ipfw_dyn_rule *r;
 906         int i;
 907
 908         if (ipfw_dyn_v == NULL ||
 909             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
 910                 realloc_dynamic_table();
 911                 if (ipfw_dyn_v == NULL)
 912                         return NULL; /* failed ! */
 913         }
 914         i = hash_packet(id);
 915
 916         r = kmalloc(sizeof *r, M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
 917         if (r == NULL) {
 918                 kprintf ("sorry cannot allocate state\n");
 919                 return NULL;
 920         }
 921
 922         /* increase refcount on parent, and set pointer */
 923         if (dyn_type == O_LIMIT) {
 924                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 925                 if ( parent->dyn_type != O_LIMIT_PARENT)
 926                         panic("invalid parent");
 927                 parent->count++;
 928                 r->parent = parent;
 929                 rule = parent->rule;
 930         }
 931
 932         r->id = *id;
 933         r->expire = time_second + dyn_syn_lifetime;
 934         r->rule = rule;
 935         r->dyn_type = dyn_type;
 936         r->pcnt = r->bcnt = 0;
 937         r->count = 0;
 938
 939         r->bucket = i;
 940         r->next = ipfw_dyn_v[i];
 941         ipfw_dyn_v[i] = r;
 942         dyn_count++;
 943         DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
 944            dyn_type,
 945            (r->id.src_ip), (r->id.src_port),
 946            (r->id.dst_ip), (r->id.dst_port),
 947            dyn_count ); )
 948         return r;
 949 }
 950
 951 /**
 952  * lookup dynamic parent rule using pkt and rule as search keys.
 953  * If the lookup fails, then install one.
 954  */
 955 static ipfw_dyn_rule *
 956 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
 957 {
 958         ipfw_dyn_rule *q;
 959         int i;
 960
 961         if (ipfw_dyn_v) {
 962                 i = hash_packet( pkt );
 963                 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
 964                         if (q->dyn_type == O_LIMIT_PARENT &&
 965                             rule== q->rule &&
 966                             pkt->proto == q->id.proto &&
 967                             pkt->src_ip == q->id.src_ip &&
 968                             pkt->dst_ip == q->id.dst_ip &&
 969                             pkt->src_port == q->id.src_port &&
 970                             pkt->dst_port == q->id.dst_port) {
 971                                 q->expire = time_second + dyn_short_lifetime;
 972                                 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
 973                                 return q;
 974                         }
 975         }
 976         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
 977 }
 978
 979 /**
 980  * Install dynamic state for rule type cmd->o.opcode
 981  *
 982  * Returns 1 (failure) if state is not installed because of errors or because
 983  * session limitations are enforced.
 984  */
 985 static int
 986 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
 987         struct ip_fw_args *args)
 988 {
 989         static int last_log;
 990
 991         ipfw_dyn_rule *q;
 992
 993         DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
 994             cmd->o.opcode,
 995             (args->f_id.src_ip), (args->f_id.src_port),
 996             (args->f_id.dst_ip), (args->f_id.dst_port) );)
 997
 998         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
 999
1000         if (q != NULL) { /* should never occur */
1001                 if (last_log != time_second) {
1002                         last_log = time_second;
1003                         kprintf(" install_state: entry already present, done\n");
1004                 }
1005                 return 0;
1006         }
1007
1008         if (dyn_count >= dyn_max)
1009                 /*
1010                  * Run out of slots, try to remove any expired rule.
1011                  */
1012                 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1013
1014         if (dyn_count >= dyn_max) {
1015                 if (last_log != time_second) {
1016                         last_log = time_second;
1017                         kprintf("install_state: Too many dynamic rules\n");
1018                 }
1019                 return 1; /* cannot install, notify caller */
1020         }
1021
1022         switch (cmd->o.opcode) {
1023         case O_KEEP_STATE: /* bidir rule */
1024                 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1025                 break;
1026
1027         case O_LIMIT: /* limit number of sessions */
1028             {
1029                 uint16_t limit_mask = cmd->limit_mask;
1030                 struct ipfw_flow_id id;
1031                 ipfw_dyn_rule *parent;
1032
1033                 DEB(kprintf("installing dyn-limit rule %d\n", cmd->conn_limit);)
1034
1035                 id.dst_ip = id.src_ip = 0;
1036                 id.dst_port = id.src_port = 0;
1037                 id.proto = args->f_id.proto;
1038
1039                 if (limit_mask & DYN_SRC_ADDR)
1040                         id.src_ip = args->f_id.src_ip;
1041                 if (limit_mask & DYN_DST_ADDR)
1042                         id.dst_ip = args->f_id.dst_ip;
1043                 if (limit_mask & DYN_SRC_PORT)
1044                         id.src_port = args->f_id.src_port;
1045                 if (limit_mask & DYN_DST_PORT)
1046                         id.dst_port = args->f_id.dst_port;
1047                 parent = lookup_dyn_parent(&id, rule);
1048                 if (parent == NULL) {
1049                         kprintf("add parent failed\n");
1050                         return 1;
1051                 }
1052                 if (parent->count >= cmd->conn_limit) {
1053                         /*
1054                          * See if we can remove some expired rule.
1055                          */
1056                         remove_dyn_rule(rule, parent);
1057                         if (parent->count >= cmd->conn_limit) {
1058                                 if (fw_verbose && last_log != time_second) {
1059                                         last_log = time_second;
1060                                         log(LOG_SECURITY | LOG_DEBUG,
1061                                             "drop session, too many entries\n");
1062                                 }
1063                                 return 1;
1064                         }
1065                 }
1066                 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1067             }
1068                 break;
1069         default:
1070                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1071                 return 1;
1072         }
1073         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1074         return 0;
1075 }
1076
1077 /*
1078  * Transmit a TCP packet, containing either a RST or a keepalive.
1079  * When flags & TH_RST, we are sending a RST packet, because of a
1080  * "reset" action matched the packet.
1081  * Otherwise we are sending a keepalive, and flags & TH_
1082  */
1083 static void
1084 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1085 {
1086         struct mbuf *m;
1087         struct ip *ip;
1088         struct tcphdr *tcp;
1089         struct route sro;       /* fake route */
1090
1091         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1092         if (m == 0)
1093                 return;
1094         m->m_pkthdr.rcvif = (struct ifnet *)0;
1095         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1096         m->m_data += max_linkhdr;
1097
1098         ip = mtod(m, struct ip *);
1099         bzero(ip, m->m_len);
1100         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1101         ip->ip_p = IPPROTO_TCP;
1102         tcp->th_off = 5;
1103         /*
1104          * Assume we are sending a RST (or a keepalive in the reverse
1105          * direction), swap src and destination addresses and ports.
1106          */
1107         ip->ip_src.s_addr = htonl(id->dst_ip);
1108         ip->ip_dst.s_addr = htonl(id->src_ip);
1109         tcp->th_sport = htons(id->dst_port);
1110         tcp->th_dport = htons(id->src_port);
1111         if (flags & TH_RST) {   /* we are sending a RST */
1112                 if (flags & TH_ACK) {
1113                         tcp->th_seq = htonl(ack);
1114                         tcp->th_ack = htonl(0);
1115                         tcp->th_flags = TH_RST;
1116                 } else {
1117                         if (flags & TH_SYN)
1118                                 seq++;
1119                         tcp->th_seq = htonl(0);
1120                         tcp->th_ack = htonl(seq);
1121                         tcp->th_flags = TH_RST | TH_ACK;
1122                 }
1123         } else {
1124                 /*
1125                  * We are sending a keepalive. flags & TH_SYN determines
1126                  * the direction, forward if set, reverse if clear.
1127                  * NOTE: seq and ack are always assumed to be correct
1128                  * as set by the caller. This may be confusing...
1129                  */
1130                 if (flags & TH_SYN) {
1131                         /*
1132                          * we have to rewrite the correct addresses!
1133                          */
1134                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1135                         ip->ip_src.s_addr = htonl(id->src_ip);
1136                         tcp->th_dport = htons(id->dst_port);
1137                         tcp->th_sport = htons(id->src_port);
1138                 }
1139                 tcp->th_seq = htonl(seq);
1140                 tcp->th_ack = htonl(ack);
1141                 tcp->th_flags = TH_ACK;
1142         }
1143         /*
1144          * set ip_len to the payload size so we can compute
1145          * the tcp checksum on the pseudoheader
1146          * XXX check this, could save a couple of words ?
1147          */
1148         ip->ip_len = htons(sizeof(struct tcphdr));
1149         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1150         /*
1151          * now fill fields left out earlier
1152          */
1153         ip->ip_ttl = ip_defttl;
1154         ip->ip_len = m->m_pkthdr.len;
1155         bzero (&sro, sizeof (sro));
1156         ip_rtaddr(ip->ip_dst, &sro);
1157         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1158         ip_output(m, NULL, &sro, 0, NULL, NULL);
1159         if (sro.ro_rt)
1160                 RTFREE(sro.ro_rt);
1161 }
1162
1163 /*
1164  * sends a reject message, consuming the mbuf passed as an argument.
1165  */
1166 static void
1167 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1168 {
1169
1170         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1171                 /* We need the IP header in host order for icmp_error(). */
1172                 if (args->eh != NULL) {
1173                         struct ip *ip = mtod(args->m, struct ip *);
1174                         ip->ip_len = ntohs(ip->ip_len);
1175                         ip->ip_off = ntohs(ip->ip_off);
1176                 }
1177                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1178         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1179                 struct tcphdr *const tcp =
1180                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1181                 if ( (tcp->th_flags & TH_RST) == 0)
1182                         send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1183                                 ntohl(tcp->th_ack),
1184                                 tcp->th_flags | TH_RST);
1185                 m_freem(args->m);
1186         } else
1187                 m_freem(args->m);
1188         args->m = NULL;
1189 }
1190
1191 /**
1192  *
1193  * Given an ip_fw *, lookup_next_rule will return a pointer
1194  * to the next rule, which can be either the jump
1195  * target (for skipto instructions) or the next one in the list (in
1196  * all other cases including a missing jump target).
1197  * The result is also written in the "next_rule" field of the rule.
1198  * Backward jumps are not allowed, so start looking from the next
1199  * rule...
1200  *
1201  * This never returns NULL -- in case we do not have an exact match,
1202  * the next rule is returned. When the ruleset is changed,
1203  * pointers are flushed so we are always correct.
1204  */
1205
1206 static struct ip_fw *
1207 lookup_next_rule(struct ip_fw *me)
1208 {
1209         struct ip_fw *rule = NULL;
1210         ipfw_insn *cmd;
1211
1212         /* look for action, in case it is a skipto */
1213         cmd = ACTION_PTR(me);
1214         if (cmd->opcode == O_LOG)
1215                 cmd += F_LEN(cmd);
1216         if ( cmd->opcode == O_SKIPTO )
1217                 for (rule = me->next; rule ; rule = rule->next)
1218                         if (rule->rulenum >= cmd->arg1)
1219                                 break;
1220         if (rule == NULL)                       /* failure or not a skipto */
1221                 rule = me->next;
1222         me->next_rule = rule;
1223         return rule;
1224 }
1225
1226 /*
1227  * The main check routine for the firewall.
1228  *
1229  * All arguments are in args so we can modify them and return them
1230  * back to the caller.
1231  *
1232  * Parameters:
1233  *
1234  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1235  *              Starts with the IP header.
1236  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1237  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1238  *              The incoming interface is in the mbuf. (in)
1239  *
1240  *      args->rule      Pointer to the last matching rule (in/out)
1241  *      args->next_hop  Socket we are forwarding to (out).
1242  *      args->f_id      Addresses grabbed from the packet (out)
1243  *
1244  * Return value:
1245  *
1246  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1247  *      0       The packet is to be accepted and routed normally OR
1248  *              the packet was denied/rejected and has been dropped;
1249  *              in the latter case, *m is equal to NULL upon return.
1250  *      port    Divert the packet to port, with these caveats:
1251  *
1252  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1253  *                of diverting it (ie, 'ipfw tee').
1254  *
1255  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1256  *                16 bits as a dummynet pipe number instead of diverting
1257  */
1258
1259 static int
1260 ipfw_chk(struct ip_fw_args *args)
1261 {
1262         /*
1263          * Local variables hold state during the processing of a packet.
1264          *
1265          * IMPORTANT NOTE: to speed up the processing of rules, there
1266          * are some assumption on the values of the variables, which
1267          * are documented here. Should you change them, please check
1268          * the implementation of the various instructions to make sure
1269          * that they still work.
1270          *
1271          * args->eh     The MAC header. It is non-null for a layer2
1272          *      packet, it is NULL for a layer-3 packet.
1273          *
1274          * m | args->m  Pointer to the mbuf, as received from the caller.
1275          *      It may change if ipfw_chk() does an m_pullup, or if it
1276          *      consumes the packet because it calls send_reject().
1277          *      XXX This has to change, so that ipfw_chk() never modifies
1278          *      or consumes the buffer.
1279          * ip   is simply an alias of the value of m, and it is kept
1280          *      in sync with it (the packet is  supposed to start with
1281          *      the ip header).
1282          */
1283         struct mbuf *m = args->m;
1284         struct ip *ip = mtod(m, struct ip *);
1285
1286         /*
1287          * oif | args->oif      If NULL, ipfw_chk has been called on the
1288          *      inbound path (ether_input, ip_input).
1289          *      If non-NULL, ipfw_chk has been called on the outbound path
1290          *      (ether_output, ip_output).
1291          */
1292         struct ifnet *oif = args->oif;
1293
1294         struct ip_fw *f = NULL;         /* matching rule */
1295         int retval = 0;
1296         struct m_tag *mtag;
1297
1298         /*
1299          * hlen The length of the IPv4 header.
1300          *      hlen >0 means we have an IPv4 packet.
1301          */
1302         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1303
1304         /*
1305          * offset       The offset of a fragment. offset != 0 means that
1306          *      we have a fragment at this offset of an IPv4 packet.
1307          *      offset == 0 means that (if this is an IPv4 packet)
1308          *      this is the first or only fragment.
1309          */
1310         u_short offset = 0;
1311
1312         /*
1313          * Local copies of addresses. They are only valid if we have
1314          * an IP packet.
1315          *
1316          * proto        The protocol. Set to 0 for non-ip packets,
1317          *      or to the protocol read from the packet otherwise.
1318          *      proto != 0 means that we have an IPv4 packet.
1319          *
1320          * src_port, dst_port   port numbers, in HOST format. Only
1321          *      valid for TCP and UDP packets.
1322          *
1323          * src_ip, dst_ip       ip addresses, in NETWORK format.
1324          *      Only valid for IPv4 packets.
1325          */
1326         uint8_t proto;
1327         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1328         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1329         uint16_t ip_len=0;
1330         int dyn_dir = MATCH_UNKNOWN;
1331         ipfw_dyn_rule *q = NULL;
1332
1333         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1334                 return 0;       /* accept */
1335         /*
1336          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1337          *      MATCH_NONE when checked and not matched (q = NULL),
1338          *      MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1339          */
1340
1341         if (args->eh == NULL ||         /* layer 3 packet */
1342                 ( m->m_pkthdr.len >= sizeof(struct ip) &&
1343                     ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1344                         hlen = ip->ip_hl << 2;
1345
1346         /*
1347          * Collect parameters into local variables for faster matching.
1348          */
1349         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1350                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1351                 goto after_ip_checks;
1352         }
1353
1354         proto = args->f_id.proto = ip->ip_p;
1355         src_ip = ip->ip_src;
1356         dst_ip = ip->ip_dst;
1357         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1358                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1359                 ip_len = ntohs(ip->ip_len);
1360         } else {
1361                 offset = ip->ip_off & IP_OFFMASK;
1362                 ip_len = ip->ip_len;
1363         }
1364
1365 #define PULLUP_TO(len)                                          \
1366                 do {                                            \
1367                         if ((m)->m_len < (len)) {               \
1368                             args->m = m = m_pullup(m, (len));   \
1369                             if (m == 0)                         \
1370                                 goto pullup_failed;             \
1371                             ip = mtod(m, struct ip *);          \
1372                         }                                       \
1373                 } while (0)
1374
1375         if (offset == 0) {
1376                 switch (proto) {
1377                 case IPPROTO_TCP:
1378                     {
1379                         struct tcphdr *tcp;
1380
1381                         PULLUP_TO(hlen + sizeof(struct tcphdr));
1382                         tcp = L3HDR(struct tcphdr, ip);
1383                         dst_port = tcp->th_dport;
1384                         src_port = tcp->th_sport;
1385                         args->f_id.flags = tcp->th_flags;
1386                         }
1387                         break;
1388
1389                 case IPPROTO_UDP:
1390                     {
1391                         struct udphdr *udp;
1392
1393                         PULLUP_TO(hlen + sizeof(struct udphdr));
1394                         udp = L3HDR(struct udphdr, ip);
1395                         dst_port = udp->uh_dport;
1396                         src_port = udp->uh_sport;
1397                         }
1398                         break;
1399
1400                 case IPPROTO_ICMP:
1401                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1402                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1403                         break;
1404
1405                 default:
1406                         break;
1407                 }
1408 #undef PULLUP_TO
1409         }
1410
1411         args->f_id.src_ip = ntohl(src_ip.s_addr);
1412         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1413         args->f_id.src_port = src_port = ntohs(src_port);
1414         args->f_id.dst_port = dst_port = ntohs(dst_port);
1415
1416 after_ip_checks:
1417         if (args->rule) {
1418                 /*
1419                  * Packet has already been tagged. Look for the next rule
1420                  * to restart processing.
1421                  *
1422                  * If fw_one_pass != 0 then just accept it.
1423                  * XXX should not happen here, but optimized out in
1424                  * the caller.
1425                  */
1426                 if (fw_one_pass)
1427                         return 0;
1428
1429                 /* This rule was deleted */
1430                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1431                         return IP_FW_PORT_DENY_FLAG;
1432
1433                 f = args->rule->next_rule;
1434                 if (f == NULL)
1435                         f = lookup_next_rule(args->rule);
1436         } else {
1437                 /*
1438                  * Find the starting rule. It can be either the first
1439                  * one, or the one after divert_rule if asked so.
1440                  */
1441                 int skipto;
1442
1443                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1444                 if (mtag != NULL)
1445                         skipto = *(uint16_t *)m_tag_data(mtag);
1446                 else
1447                         skipto = 0;
1448
1449                 f = layer3_chain;
1450                 if (args->eh == NULL && skipto != 0) {
1451                         if (skipto >= IPFW_DEFAULT_RULE)
1452                                 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1453                         while (f && f->rulenum <= skipto)
1454                                 f = f->next;
1455                         if (f == NULL)  /* drop packet */
1456                                 return(IP_FW_PORT_DENY_FLAG);
1457                 }
1458         }
1459         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1460                 m_tag_delete(m, mtag);
1461
1462         /*
1463          * Now scan the rules, and parse microinstructions for each rule.
1464          */
1465         for (; f; f = f->next) {
1466                 int l, cmdlen;
1467                 ipfw_insn *cmd;
1468                 int skip_or; /* skip rest of OR block */
1469
1470 again:
1471                 if (set_disable & (1 << f->set) )
1472                         continue;
1473
1474                 skip_or = 0;
1475                 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1476                     l -= cmdlen, cmd += cmdlen) {
1477                         int match;
1478
1479                         /*
1480                          * check_body is a jump target used when we find a
1481                          * CHECK_STATE, and need to jump to the body of
1482                          * the target rule.
1483                          */
1484
1485 check_body:
1486                         cmdlen = F_LEN(cmd);
1487                         /*
1488                          * An OR block (insn_1 || .. || insn_n) has the
1489                          * F_OR bit set in all but the last instruction.
1490                          * The first match will set "skip_or", and cause
1491                          * the following instructions to be skipped until
1492                          * past the one with the F_OR bit clear.
1493                          */
1494                         if (skip_or) {          /* skip this instruction */
1495                                 if ((cmd->len & F_OR) == 0)
1496                                         skip_or = 0;    /* next one is good */
1497                                 continue;
1498                         }
1499                         match = 0; /* set to 1 if we succeed */
1500
1501                         switch (cmd->opcode) {
1502                         /*
1503                          * The first set of opcodes compares the packet's
1504                          * fields with some pattern, setting 'match' if a
1505                          * match is found. At the end of the loop there is
1506                          * logic to deal with F_NOT and F_OR flags associated
1507                          * with the opcode.
1508                          */
1509                         case O_NOP:
1510                                 match = 1;
1511                                 break;
1512
1513                         case O_FORWARD_MAC:
1514                                 kprintf("ipfw: opcode %d unimplemented\n",
1515                                     cmd->opcode);
1516                                 break;
1517
1518                         case O_GID:
1519                         case O_UID:
1520                                 /*
1521                                  * We only check offset == 0 && proto != 0,
1522                                  * as this ensures that we have an IPv4
1523                                  * packet with the ports info.
1524                                  */
1525                                 if (offset!=0)
1526                                         break;
1527                             {
1528                                 struct inpcbinfo *pi;
1529                                 int wildcard;
1530                                 struct inpcb *pcb;
1531
1532                                 if (proto == IPPROTO_TCP) {
1533                                         wildcard = 0;
1534                                         pi = &tcbinfo[mycpu->gd_cpuid];
1535                                 } else if (proto == IPPROTO_UDP) {
1536                                         wildcard = 1;
1537                                         pi = &udbinfo;
1538                                 } else
1539                                         break;
1540
1541                                 pcb =  (oif) ?
1542                                         in_pcblookup_hash(pi,
1543                                             dst_ip, htons(dst_port),
1544                                             src_ip, htons(src_port),
1545                                             wildcard, oif) :
1546                                         in_pcblookup_hash(pi,
1547                                             src_ip, htons(src_port),
1548                                             dst_ip, htons(dst_port),
1549                                             wildcard, NULL);
1550
1551                                 if (pcb == NULL || pcb->inp_socket == NULL)
1552                                         break;
1553 #if defined(__DragonFly__) || (defined(__FreeBSD__) && __FreeBSD_version < 500034)
1554 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1555 #endif
1556                                 if (cmd->opcode == O_UID) {
1557                                         match =
1558                                           !socheckuid(pcb->inp_socket,
1559                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1560                                 } else  {
1561                                         match = groupmember(
1562                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1563                                             pcb->inp_socket->so_cred);
1564                                 }
1565                             }
1566                                 break;
1567
1568                         case O_RECV:
1569                                 match = iface_match(m->m_pkthdr.rcvif,
1570                                     (ipfw_insn_if *)cmd);
1571                                 break;
1572
1573                         case O_XMIT:
1574                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1575                                 break;
1576
1577                         case O_VIA:
1578                                 match = iface_match(oif ? oif :
1579                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1580                                 break;
1581
1582                         case O_MACADDR2:
1583                                 if (args->eh != NULL) { /* have MAC header */
1584                                         uint32_t *want = (uint32_t *)
1585                                                 ((ipfw_insn_mac *)cmd)->addr;
1586                                         uint32_t *mask = (uint32_t *)
1587                                                 ((ipfw_insn_mac *)cmd)->mask;
1588                                         uint32_t *hdr = (uint32_t *)args->eh;
1589
1590                                         match =
1591                                             ( want[0] == (hdr[0] & mask[0]) &&
1592                                               want[1] == (hdr[1] & mask[1]) &&
1593                                               want[2] == (hdr[2] & mask[2]) );
1594                                 }
1595                                 break;
1596
1597                         case O_MAC_TYPE:
1598                                 if (args->eh != NULL) {
1599                                         uint16_t t =
1600                                             ntohs(args->eh->ether_type);
1601                                         uint16_t *p =
1602                                             ((ipfw_insn_u16 *)cmd)->ports;
1603                                         int i;
1604
1605                                         for (i = cmdlen - 1; !match && i>0;
1606                                             i--, p += 2)
1607                                                 match = (t>=p[0] && t<=p[1]);
1608                                 }
1609                                 break;
1610
1611                         case O_FRAG:
1612                                 match = (hlen > 0 && offset != 0);
1613                                 break;
1614
1615                         case O_IN:      /* "out" is "not in" */
1616                                 match = (oif == NULL);
1617                                 break;
1618
1619                         case O_LAYER2:
1620                                 match = (args->eh != NULL);
1621                                 break;
1622
1623                         case O_PROTO:
1624                                 /*
1625                                  * We do not allow an arg of 0 so the
1626                                  * check of "proto" only suffices.
1627                                  */
1628                                 match = (proto == cmd->arg1);
1629                                 break;
1630
1631                         case O_IP_SRC:
1632                                 match = (hlen > 0 &&
1633                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1634                                     src_ip.s_addr);
1635                                 break;
1636
1637                         case O_IP_SRC_MASK:
1638                                 match = (hlen > 0 &&
1639                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1640                                      (src_ip.s_addr &
1641                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1642                                 break;
1643
1644                         case O_IP_SRC_ME:
1645                                 if (hlen > 0) {
1646                                         struct ifnet *tif;
1647
1648                                         INADDR_TO_IFP(src_ip, tif);
1649                                         match = (tif != NULL);
1650                                 }
1651                                 break;
1652
1653                         case O_IP_DST_SET:
1654                         case O_IP_SRC_SET:
1655                                 if (hlen > 0) {
1656                                         uint32_t *d = (uint32_t *)(cmd+1);
1657                                         uint32_t addr =
1658                                             cmd->opcode == O_IP_DST_SET ?
1659                                                 args->f_id.dst_ip :
1660                                                 args->f_id.src_ip;
1661
1662                                             if (addr < d[0])
1663                                                     break;
1664                                             addr -= d[0]; /* subtract base */
1665                                             match = (addr < cmd->arg1) &&
1666                                                 ( d[ 1 + (addr>>5)] &
1667                                                   (1<<(addr & 0x1f)) );
1668                                 }
1669                                 break;
1670
1671                         case O_IP_DST:
1672                                 match = (hlen > 0 &&
1673                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1674                                     dst_ip.s_addr);
1675                                 break;
1676
1677                         case O_IP_DST_MASK:
1678                                 match = (hlen > 0) &&
1679                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1680                                      (dst_ip.s_addr &
1681                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1682                                 break;
1683
1684                         case O_IP_DST_ME:
1685                                 if (hlen > 0) {
1686                                         struct ifnet *tif;
1687
1688                                         INADDR_TO_IFP(dst_ip, tif);
1689                                         match = (tif != NULL);
1690                                 }
1691                                 break;
1692
1693                         case O_IP_SRCPORT:
1694                         case O_IP_DSTPORT:
1695                                 /*
1696                                  * offset == 0 && proto != 0 is enough
1697                                  * to guarantee that we have an IPv4
1698                                  * packet with port info.
1699                                  */
1700                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1701                                     && offset == 0) {
1702                                         uint16_t x =
1703                                             (cmd->opcode == O_IP_SRCPORT) ?
1704                                                 src_port : dst_port ;
1705                                         uint16_t *p =
1706                                             ((ipfw_insn_u16 *)cmd)->ports;
1707                                         int i;
1708
1709                                         for (i = cmdlen - 1; !match && i>0;
1710                                             i--, p += 2)
1711                                                 match = (x>=p[0] && x<=p[1]);
1712                                 }
1713                                 break;
1714
1715                         case O_ICMPTYPE:
1716                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
1717                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
1718                                 break;
1719
1720                         case O_IPOPT:
1721                                 match = (hlen > 0 && ipopts_match(ip, cmd) );
1722                                 break;
1723
1724                         case O_IPVER:
1725                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1726                                 break;
1727
1728                         case O_IPTTL:
1729                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1730                                 break;
1731
1732                         case O_IPID:
1733                                 match = (hlen > 0 &&
1734                                     cmd->arg1 == ntohs(ip->ip_id));
1735                                 break;
1736
1737                         case O_IPLEN:
1738                                 match = (hlen > 0 && cmd->arg1 == ip_len);
1739                                 break;
1740
1741                         case O_IPPRECEDENCE:
1742                                 match = (hlen > 0 &&
1743                                     (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1744                                 break;
1745
1746                         case O_IPTOS:
1747                                 match = (hlen > 0 &&
1748                                     flags_match(cmd, ip->ip_tos));
1749                                 break;
1750
1751                         case O_TCPFLAGS:
1752                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1753                                     flags_match(cmd,
1754                                         L3HDR(struct tcphdr,ip)->th_flags));
1755                                 break;
1756
1757                         case O_TCPOPTS:
1758                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1759                                     tcpopts_match(ip, cmd));
1760                                 break;
1761
1762                         case O_TCPSEQ:
1763                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1764                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1765                                         L3HDR(struct tcphdr,ip)->th_seq);
1766                                 break;
1767
1768                         case O_TCPACK:
1769                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1770                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1771                                         L3HDR(struct tcphdr,ip)->th_ack);
1772                                 break;
1773
1774                         case O_TCPWIN:
1775                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1776                                     cmd->arg1 ==
1777                                         L3HDR(struct tcphdr,ip)->th_win);
1778                                 break;
1779
1780                         case O_ESTAB:
1781                                 /* reject packets which have SYN only */
1782                                 /* XXX should i also check for TH_ACK ? */
1783                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1784                                     (L3HDR(struct tcphdr,ip)->th_flags &
1785                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1786                                 break;
1787
1788                         case O_LOG:
1789                                 if (fw_verbose)
1790                                         ipfw_log(f, hlen, args->eh, m, oif);
1791                                 match = 1;
1792                                 break;
1793
1794                         case O_PROB:
1795                                 match = (krandom() <
1796                                         ((ipfw_insn_u32 *)cmd)->d[0]);
1797                                 break;
1798
1799                         /*
1800                          * The second set of opcodes represents 'actions',
1801                          * i.e. the terminal part of a rule once the packet
1802                          * matches all previous patterns.
1803                          * Typically there is only one action for each rule,
1804                          * and the opcode is stored at the end of the rule
1805                          * (but there are exceptions -- see below).
1806                          *
1807                          * In general, here we set retval and terminate the
1808                          * outer loop (would be a 'break 3' in some language,
1809                          * but we need to do a 'goto done').
1810                          *
1811                          * Exceptions:
1812                          * O_COUNT and O_SKIPTO actions:
1813                          *   instead of terminating, we jump to the next rule
1814                          *   ('goto next_rule', equivalent to a 'break 2'),
1815                          *   or to the SKIPTO target ('goto again' after
1816                          *   having set f, cmd and l), respectively.
1817                          *
1818                          * O_LIMIT and O_KEEP_STATE: these opcodes are
1819                          *   not real 'actions', and are stored right
1820                          *   before the 'action' part of the rule.
1821                          *   These opcodes try to install an entry in the
1822                          *   state tables; if successful, we continue with
1823                          *   the next opcode (match=1; break;), otherwise
1824                          *   the packet *   must be dropped
1825                          *   ('goto done' after setting retval);
1826                          *
1827                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1828                          *   cause a lookup of the state table, and a jump
1829                          *   to the 'action' part of the parent rule
1830                          *   ('goto check_body') if an entry is found, or
1831                          *   (CHECK_STATE only) a jump to the next rule if
1832                          *   the entry is not found ('goto next_rule').
1833                          *   The result of the lookup is cached to make
1834                          *   further instances of these opcodes are
1835                          *   effectively NOPs.
1836                          */
1837                         case O_LIMIT:
1838                         case O_KEEP_STATE:
1839                                 if (install_state(f,
1840                                     (ipfw_insn_limit *)cmd, args)) {
1841                                         retval = IP_FW_PORT_DENY_FLAG;
1842                                         goto done; /* error/limit violation */
1843                                 }
1844                                 match = 1;
1845                                 break;
1846
1847                         case O_PROBE_STATE:
1848                         case O_CHECK_STATE:
1849                                 /*
1850                                  * dynamic rules are checked at the first
1851                                  * keep-state or check-state occurrence,
1852                                  * with the result being stored in dyn_dir.
1853                                  * The compiler introduces a PROBE_STATE
1854                                  * instruction for us when we have a
1855                                  * KEEP_STATE (because PROBE_STATE needs
1856                                  * to be run first).
1857                                  */
1858                                 if (dyn_dir == MATCH_UNKNOWN &&
1859                                     (q = lookup_dyn_rule(&args->f_id,
1860                                      &dyn_dir, proto == IPPROTO_TCP ?
1861                                         L3HDR(struct tcphdr, ip) : NULL))
1862                                         != NULL) {
1863                                         /*
1864                                          * Found dynamic entry, update stats
1865                                          * and jump to the 'action' part of
1866                                          * the parent rule.
1867                                          */
1868                                         q->pcnt++;
1869                                         q->bcnt += ip_len;
1870                                         f = q->rule;
1871                                         cmd = ACTION_PTR(f);
1872                                         l = f->cmd_len - f->act_ofs;
1873                                         goto check_body;
1874                                 }
1875                                 /*
1876                                  * Dynamic entry not found. If CHECK_STATE,
1877                                  * skip to next rule, if PROBE_STATE just
1878                                  * ignore and continue with next opcode.
1879                                  */
1880                                 if (cmd->opcode == O_CHECK_STATE)
1881                                         goto next_rule;
1882                                 match = 1;
1883                                 break;
1884
1885                         case O_ACCEPT:
1886                                 retval = 0;     /* accept */
1887                                 goto done;
1888
1889                         case O_PIPE:
1890                         case O_QUEUE:
1891                                 args->rule = f; /* report matching rule */
1892                                 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
1893                                 goto done;
1894
1895                         case O_DIVERT:
1896                         case O_TEE:
1897                                 if (args->eh) /* not on layer 2 */
1898                                         break;
1899
1900                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
1901                                                  sizeof(uint16_t), MB_DONTWAIT);
1902                                 if (mtag == NULL) {
1903                                         retval = IP_FW_PORT_DENY_FLAG;
1904                                         goto done;
1905                                 }
1906                                 *(uint16_t *)m_tag_data(mtag) = f->rulenum;
1907                                 m_tag_prepend(m, mtag);
1908                                 retval = (cmd->opcode == O_DIVERT) ?
1909                                     cmd->arg1 :
1910                                     cmd->arg1 | IP_FW_PORT_TEE_FLAG;
1911                                 goto done;
1912
1913                         case O_COUNT:
1914                         case O_SKIPTO:
1915                                 f->pcnt++;      /* update stats */
1916                                 f->bcnt += ip_len;
1917                                 f->timestamp = time_second;
1918                                 if (cmd->opcode == O_COUNT)
1919                                         goto next_rule;
1920                                 /* handle skipto */
1921                                 if (f->next_rule == NULL)
1922                                         lookup_next_rule(f);
1923                                 f = f->next_rule;
1924                                 goto again;
1925
1926                         case O_REJECT:
1927                                 /*
1928                                  * Drop the packet and send a reject notice
1929                                  * if the packet is not ICMP (or is an ICMP
1930                                  * query), and it is not multicast/broadcast.
1931                                  */
1932                                 if (hlen > 0 &&
1933                                     (proto != IPPROTO_ICMP ||
1934                                      is_icmp_query(ip)) &&
1935                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
1936                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
1937                                         send_reject(args, cmd->arg1,
1938                                             offset,ip_len);
1939                                         m = args->m;
1940                                 }
1941                                 /* FALLTHROUGH */
1942                         case O_DENY:
1943                                 retval = IP_FW_PORT_DENY_FLAG;
1944                                 goto done;
1945
1946                         case O_FORWARD_IP:
1947                                 if (args->eh)   /* not valid on layer2 pkts */
1948                                         break;
1949                                 if (!q || dyn_dir == MATCH_FORWARD)
1950                                         args->next_hop =
1951                                             &((ipfw_insn_sa *)cmd)->sa;
1952                                 retval = 0;
1953                                 goto done;
1954
1955                         default:
1956                                 panic("-- unknown opcode %d\n", cmd->opcode);
1957                         } /* end of switch() on opcodes */
1958
1959                         if (cmd->len & F_NOT)
1960                                 match = !match;
1961
1962                         if (match) {
1963                                 if (cmd->len & F_OR)
1964                                         skip_or = 1;
1965                         } else {
1966                                 if (!(cmd->len & F_OR)) /* not an OR block, */
1967                                         break;          /* try next rule    */
1968                         }
1969
1970                 }       /* end of inner for, scan opcodes */
1971
1972 next_rule:;             /* try next rule                */
1973
1974         }               /* end of outer for, scan rules */
1975         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
1976         return(IP_FW_PORT_DENY_FLAG);
1977
1978 done:
1979         /* Update statistics */
1980         f->pcnt++;
1981         f->bcnt += ip_len;
1982         f->timestamp = time_second;
1983         return retval;
1984
1985 pullup_failed:
1986         if (fw_verbose)
1987                 kprintf("pullup failed\n");
1988         return(IP_FW_PORT_DENY_FLAG);
1989 }
1990
1991 static void
1992 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
1993 {
1994         struct m_tag *mtag;
1995         struct dn_pkt *pkt;
1996         ipfw_insn *cmd;
1997         const struct ipfw_flow_id *id;
1998         struct dn_flow_id *fid;
1999
2000         M_ASSERTPKTHDR(m);
2001
2002         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2003         if (mtag == NULL) {
2004                 m_freem(m);
2005                 return;
2006         }
2007         m_tag_prepend(m, mtag);
2008
2009         pkt = m_tag_data(mtag);
2010         bzero(pkt, sizeof(*pkt));
2011
2012         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2013         if (cmd->opcode == O_LOG)
2014                 cmd += F_LEN(cmd);
2015         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2016                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2017
2018         pkt->dn_m = m;
2019         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2020         pkt->ifp = fwa->oif;
2021         pkt->cpuid = mycpu->gd_cpuid;
2022         pkt->pipe_nr = pipe_nr;
2023
2024         id = &fwa->f_id;
2025         fid = &pkt->id;
2026         fid->fid_dst_ip = id->dst_ip;
2027         fid->fid_src_ip = id->src_ip;
2028         fid->fid_dst_port = id->dst_port;
2029         fid->fid_src_port = id->src_port;
2030         fid->fid_proto = id->proto;
2031         fid->fid_flags = id->flags;
2032
2033         ipfw_ref_rule(fwa->rule);
2034         pkt->dn_priv = fwa->rule;
2035         pkt->dn_unref_priv = ipfw_unref_rule;
2036
2037         if (cmd->opcode == O_PIPE)
2038                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2039
2040         if (dir == DN_TO_IP_OUT) {
2041                 /*
2042                  * We need to copy *ro because for ICMP pkts (and maybe
2043                  * others) the caller passed a pointer into the stack;
2044                  * dst might also be a pointer into *ro so it needs to
2045                  * be updated.
2046                  */
2047                 pkt->ro = *(fwa->ro);
2048                 if (fwa->ro->ro_rt)
2049                         fwa->ro->ro_rt->rt_refcnt++;
2050                 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) {
2051                         /* 'dst' points into 'ro' */
2052                         fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst);
2053                 }
2054                 pkt->dn_dst = fwa->dst;
2055                 pkt->flags = fwa->flags;
2056         }
2057
2058         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2059         ip_dn_queue(m);
2060 }
2061
2062 /*
2063  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2064  * These will be reconstructed on the fly as packets are matched.
2065  * Must be called at splimp().
2066  */
2067 static void
2068 flush_rule_ptrs(void)
2069 {
2070         struct ip_fw *rule;
2071
2072         for (rule = layer3_chain; rule; rule = rule->next)
2073                 rule->next_rule = NULL;
2074 }
2075
2076 static __inline void
2077 ipfw_inc_static_count(struct ip_fw *rule)
2078 {
2079         static_count++;
2080         static_ioc_len += IOC_RULESIZE(rule);
2081 }
2082
2083 static __inline void
2084 ipfw_dec_static_count(struct ip_fw *rule)
2085 {
2086         int l = IOC_RULESIZE(rule);
2087
2088         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2089         static_count--;
2090
2091         KASSERT(static_ioc_len >= l,
2092                 ("invalid static len %u\n", static_ioc_len));
2093         static_ioc_len -= l;
2094 }
2095
2096 static struct ip_fw *
2097 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule)
2098 {
2099         struct ip_fw *rule;
2100
2101         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2102
2103         rule->act_ofs = ioc_rule->act_ofs;
2104         rule->cmd_len = ioc_rule->cmd_len;
2105         rule->rulenum = ioc_rule->rulenum;
2106         rule->set = ioc_rule->set;
2107         rule->usr_flags = ioc_rule->usr_flags;
2108
2109         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2110
2111         rule->refcnt = 1;
2112
2113         return rule;
2114 }
2115
2116 /*
2117  * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2118  * possibly create a rule number and add the rule to the list.
2119  * Update the rule_number in the input struct so the caller knows it as well.
2120  */
2121 static int
2122 ipfw_add_rule(struct ip_fw **head, struct ipfw_ioc_rule *ioc_rule)
2123 {
2124         struct ip_fw *rule, *f, *prev;
2125
2126         KKASSERT(*head != NULL);
2127
2128         rule = ipfw_create_rule(ioc_rule);
2129
2130         crit_enter();
2131
2132         /*
2133          * If rulenum is 0, find highest numbered rule before the
2134          * default rule, and add autoinc_step
2135          */
2136         if (autoinc_step < 1)
2137                 autoinc_step = 1;
2138         else if (autoinc_step > 1000)
2139                 autoinc_step = 1000;
2140         if (rule->rulenum == 0) {
2141                 /*
2142                  * locate the highest numbered rule before default
2143                  */
2144                 for (f = *head; f; f = f->next) {
2145                         if (f->rulenum == IPFW_DEFAULT_RULE)
2146                                 break;
2147                         rule->rulenum = f->rulenum;
2148                 }
2149                 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2150                         rule->rulenum += autoinc_step;
2151                 ioc_rule->rulenum = rule->rulenum;
2152         }
2153
2154         /*
2155          * Now insert the new rule in the right place in the sorted list.
2156          */
2157         for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2158                 if (f->rulenum > rule->rulenum) { /* found the location */
2159                         if (prev) {
2160                                 rule->next = f;
2161                                 prev->next = rule;
2162                         } else { /* head insert */
2163                                 rule->next = *head;
2164                                 *head = rule;
2165                         }
2166                         break;
2167                 }
2168         }
2169
2170         flush_rule_ptrs();
2171         ipfw_inc_static_count(rule);
2172
2173         crit_exit();
2174
2175         DEB(kprintf("++ installed rule %d, static count now %d\n",
2176                 rule->rulenum, static_count);)
2177         return (0);
2178 }
2179
2180 /**
2181  * Free storage associated with a static rule (including derived
2182  * dynamic rules).
2183  * The caller is in charge of clearing rule pointers to avoid
2184  * dangling pointers.
2185  * @return a pointer to the next entry.
2186  * Arguments are not checked, so they better be correct.
2187  * Must be called at splimp().
2188  */
2189 static struct ip_fw *
2190 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2191 {
2192         struct ip_fw *n;
2193
2194         n = rule->next;
2195         remove_dyn_rule(rule, NULL /* force removal */);
2196         if (prev == NULL)
2197                 *head = n;
2198         else
2199                 prev->next = n;
2200         ipfw_dec_static_count(rule);
2201
2202         /* Mark the rule as invalid */
2203         rule->rule_flags |= IPFW_RULE_F_INVALID;
2204         rule->next_rule = NULL;
2205
2206         /* Try to free this rule */
2207         ipfw_free_rule(rule);
2208
2209         return n;
2210 }
2211
2212 /*
2213  * Deletes all rules from a chain (including the default rule
2214  * if the second argument is set).
2215  * Must be called at splimp().
2216  */
2217 static void
2218 free_chain(struct ip_fw **chain, int kill_default)
2219 {
2220         struct ip_fw *rule;
2221
2222         flush_rule_ptrs(); /* more efficient to do outside the loop */
2223
2224         while ( (rule = *chain) != NULL &&
2225             (kill_default || rule->rulenum != IPFW_DEFAULT_RULE) )
2226                 delete_rule(chain, NULL, rule);
2227
2228         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2229
2230         if (kill_default) {
2231                 ip_fw_default_rule = NULL;      /* Reset default rule */
2232
2233                 if (ipfw_dyn_v != NULL) {
2234                         /*
2235                          * Free dynamic rules(state) hash table
2236                          */
2237                         kfree(ipfw_dyn_v, M_IPFW);
2238                         ipfw_dyn_v = NULL;
2239                 }
2240
2241                 KASSERT(static_count == 0,
2242                         ("%u static rules remains\n", static_count));
2243                 KASSERT(static_ioc_len == 0,
2244                         ("%u bytes of static rules remains\n", static_ioc_len));
2245         } else {
2246                 KASSERT(static_count == 1,
2247                         ("%u static rules remains\n", static_count));
2248                 KASSERT(static_ioc_len == IOC_RULESIZE(ip_fw_default_rule),
2249                         ("%u bytes of static rules remains, should be %u\n",
2250                          static_ioc_len, IOC_RULESIZE(ip_fw_default_rule)));
2251         }
2252 }
2253
2254 /**
2255  * Remove all rules with given number, and also do set manipulation.
2256  *
2257  * The argument is an uint32_t. The low 16 bit are the rule or set number,
2258  * the next 8 bits are the new set, the top 8 bits are the command:
2259  *
2260  *      0       delete rules with given number
2261  *      1       delete rules with given set number
2262  *      2       move rules with given number to new set
2263  *      3       move rules with given set number to new set
2264  *      4       swap sets with given numbers
2265  */
2266 static int
2267 del_entry(struct ip_fw **chain, uint32_t arg)
2268 {
2269         struct ip_fw *prev, *rule;
2270         uint16_t rulenum;
2271         uint8_t cmd, new_set;
2272
2273         rulenum = arg & 0xffff;
2274         cmd = (arg >> 24) & 0xff;
2275         new_set = (arg >> 16) & 0xff;
2276
2277         if (cmd > 4)
2278                 return EINVAL;
2279         if (new_set > 30)
2280                 return EINVAL;
2281         if (cmd == 0 || cmd == 2) {
2282                 if (rulenum == IPFW_DEFAULT_RULE)
2283                         return EINVAL;
2284         } else {
2285                 if (rulenum > 30)
2286                         return EINVAL;
2287         }
2288
2289         switch (cmd) {
2290         case 0: /* delete rules with given number */
2291                 /*
2292                  * locate first rule to delete
2293                  */
2294                 for (prev = NULL, rule = *chain;
2295                     rule && rule->rulenum < rulenum;
2296                      prev = rule, rule = rule->next)
2297                         ;
2298                 if (rule->rulenum != rulenum)
2299                         return EINVAL;
2300
2301                 crit_enter(); /* no access to rules while removing */
2302                 /*
2303                  * flush pointers outside the loop, then delete all matching
2304                  * rules. prev remains the same throughout the cycle.
2305                  */
2306                 flush_rule_ptrs();
2307                 while (rule && rule->rulenum == rulenum)
2308                         rule = delete_rule(chain, prev, rule);
2309                 crit_exit();
2310                 break;
2311
2312         case 1: /* delete all rules with given set number */
2313                 crit_enter();
2314                 flush_rule_ptrs();
2315                 for (prev = NULL, rule = *chain; rule ; )
2316                         if (rule->set == rulenum)
2317                                 rule = delete_rule(chain, prev, rule);
2318                         else {
2319                                 prev = rule;
2320                                 rule = rule->next;
2321                         }
2322                 crit_exit();
2323                 break;
2324
2325         case 2: /* move rules with given number to new set */
2326                 crit_enter();
2327                 for (rule = *chain; rule ; rule = rule->next)
2328                         if (rule->rulenum == rulenum)
2329                                 rule->set = new_set;
2330                 crit_exit();
2331                 break;
2332
2333         case 3: /* move rules with given set number to new set */
2334                 crit_enter();
2335                 for (rule = *chain; rule ; rule = rule->next)
2336                         if (rule->set == rulenum)
2337                                 rule->set = new_set;
2338                 crit_exit();
2339                 break;
2340
2341         case 4: /* swap two sets */
2342                 crit_enter();
2343                 for (rule = *chain; rule ; rule = rule->next)
2344                         if (rule->set == rulenum)
2345                                 rule->set = new_set;
2346                         else if (rule->set == new_set)
2347                                 rule->set = rulenum;
2348                 crit_exit();
2349                 break;
2350         }
2351         return 0;
2352 }
2353
2354 /*
2355  * Clear counters for a specific rule.
2356  */
2357 static void
2358 clear_counters(struct ip_fw *rule, int log_only)
2359 {
2360         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2361
2362         if (log_only == 0) {
2363                 rule->bcnt = rule->pcnt = 0;
2364                 rule->timestamp = 0;
2365         }
2366         if (l->o.opcode == O_LOG)
2367                 l->log_left = l->max_log;
2368 }
2369
2370 /**
2371  * Reset some or all counters on firewall rules.
2372  * @arg frwl is null to clear all entries, or contains a specific
2373  * rule number.
2374  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2375  */
2376 static int
2377 zero_entry(int rulenum, int log_only)
2378 {
2379         struct ip_fw *rule;
2380         char *msg;
2381
2382         if (rulenum == 0) {
2383                 crit_enter();
2384                 norule_counter = 0;
2385                 for (rule = layer3_chain; rule; rule = rule->next)
2386                         clear_counters(rule, log_only);
2387                 crit_exit();
2388                 msg = log_only ? "ipfw: All logging counts reset.\n" :
2389                                 "ipfw: Accounting cleared.\n";
2390         } else {
2391                 int cleared = 0;
2392                 /*
2393                  * We can have multiple rules with the same number, so we
2394                  * need to clear them all.
2395                  */
2396                 for (rule = layer3_chain; rule; rule = rule->next)
2397                         if (rule->rulenum == rulenum) {
2398                                 crit_enter();
2399                                 while (rule && rule->rulenum == rulenum) {
2400                                         clear_counters(rule, log_only);
2401                                         rule = rule->next;
2402                                 }
2403                                 crit_exit();
2404                                 cleared = 1;
2405                                 break;
2406                         }
2407                 if (!cleared)   /* we did not find any matching rules */
2408                         return (EINVAL);
2409                 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2410                                 "ipfw: Entry %d cleared.\n";
2411         }
2412         if (fw_verbose)
2413                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2414         return (0);
2415 }
2416
2417 /*
2418  * Check validity of the structure before insert.
2419  * Fortunately rules are simple, so this mostly need to check rule sizes.
2420  */
2421 static int
2422 ipfw_ctl_check_rule(struct ipfw_ioc_rule *rule, int size)
2423 {
2424         int l, cmdlen = 0;
2425         int have_action = 0;
2426         ipfw_insn *cmd;
2427
2428         /* Check for valid size */
2429         if (size < sizeof(*rule)) {
2430                 kprintf("ipfw: rule too short\n");
2431                 return EINVAL;
2432         }
2433         l = IOC_RULESIZE(rule);
2434         if (l != size) {
2435                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
2436                 return EINVAL;
2437         }
2438
2439         /*
2440          * Now go for the individual checks. Very simple ones, basically only
2441          * instruction sizes.
2442          */
2443         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
2444              l -= cmdlen, cmd += cmdlen) {
2445                 cmdlen = F_LEN(cmd);
2446                 if (cmdlen > l) {
2447                         kprintf("ipfw: opcode %d size truncated\n",
2448                             cmd->opcode);
2449                         return EINVAL;
2450                 }
2451                 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
2452                 switch (cmd->opcode) {
2453                 case O_NOP:
2454                 case O_PROBE_STATE:
2455                 case O_KEEP_STATE:
2456                 case O_PROTO:
2457                 case O_IP_SRC_ME:
2458                 case O_IP_DST_ME:
2459                 case O_LAYER2:
2460                 case O_IN:
2461                 case O_FRAG:
2462                 case O_IPOPT:
2463                 case O_IPLEN:
2464                 case O_IPID:
2465                 case O_IPTOS:
2466                 case O_IPPRECEDENCE:
2467                 case O_IPTTL:
2468                 case O_IPVER:
2469                 case O_TCPWIN:
2470                 case O_TCPFLAGS:
2471                 case O_TCPOPTS:
2472                 case O_ESTAB:
2473                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2474                                 goto bad_size;
2475                         break;
2476
2477                 case O_UID:
2478                 case O_GID:
2479                 case O_IP_SRC:
2480                 case O_IP_DST:
2481                 case O_TCPSEQ:
2482                 case O_TCPACK:
2483                 case O_PROB:
2484                 case O_ICMPTYPE:
2485                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2486                                 goto bad_size;
2487                         break;
2488
2489                 case O_LIMIT:
2490                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2491                                 goto bad_size;
2492                         break;
2493
2494                 case O_LOG:
2495                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2496                                 goto bad_size;
2497
2498                         ((ipfw_insn_log *)cmd)->log_left =
2499                             ((ipfw_insn_log *)cmd)->max_log;
2500
2501                         break;
2502
2503                 case O_IP_SRC_MASK:
2504                 case O_IP_DST_MASK:
2505                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2506                                 goto bad_size;
2507                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2508                                 kprintf("ipfw: opcode %d, useless rule\n",
2509                                         cmd->opcode);
2510                                 return EINVAL;
2511                         }
2512                         break;
2513
2514                 case O_IP_SRC_SET:
2515                 case O_IP_DST_SET:
2516                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2517                                 kprintf("ipfw: invalid set size %d\n",
2518                                         cmd->arg1);
2519                                 return EINVAL;
2520                         }
2521                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2522                             (cmd->arg1+31)/32 )
2523                                 goto bad_size;
2524                         break;
2525
2526                 case O_MACADDR2:
2527                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2528                                 goto bad_size;
2529                         break;
2530
2531                 case O_MAC_TYPE:
2532                 case O_IP_SRCPORT:
2533                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2534                         if (cmdlen < 2 || cmdlen > 31)
2535                                 goto bad_size;
2536                         break;
2537
2538                 case O_RECV:
2539                 case O_XMIT:
2540                 case O_VIA:
2541                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2542                                 goto bad_size;
2543                         break;
2544
2545                 case O_PIPE:
2546                 case O_QUEUE:
2547                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2548                                 goto bad_size;
2549                         goto check_action;
2550
2551                 case O_FORWARD_IP:
2552                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2553                                 goto bad_size;
2554                         goto check_action;
2555
2556                 case O_FORWARD_MAC: /* XXX not implemented yet */
2557                 case O_CHECK_STATE:
2558                 case O_COUNT:
2559                 case O_ACCEPT:
2560                 case O_DENY:
2561                 case O_REJECT:
2562                 case O_SKIPTO:
2563                 case O_DIVERT:
2564                 case O_TEE:
2565                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2566                                 goto bad_size;
2567 check_action:
2568                         if (have_action) {
2569                                 kprintf("ipfw: opcode %d, multiple actions"
2570                                         " not allowed\n",
2571                                         cmd->opcode);
2572                                 return EINVAL;
2573                         }
2574                         have_action = 1;
2575                         if (l != cmdlen) {
2576                                 kprintf("ipfw: opcode %d, action must be"
2577                                         " last opcode\n",
2578                                         cmd->opcode);
2579                                 return EINVAL;
2580                         }
2581                         break;
2582                 default:
2583                         kprintf("ipfw: opcode %d, unknown opcode\n",
2584                                 cmd->opcode);
2585                         return EINVAL;
2586                 }
2587         }
2588         if (have_action == 0) {
2589                 kprintf("ipfw: missing action\n");
2590                 return EINVAL;
2591         }
2592         return 0;
2593
2594 bad_size:
2595         kprintf("ipfw: opcode %d size %d wrong\n",
2596                 cmd->opcode, cmdlen);
2597         return EINVAL;
2598 }
2599
2600 static int
2601 ipfw_ctl_add_rule(struct sockopt *sopt)
2602 {
2603         struct ipfw_ioc_rule *ioc_rule;
2604         uint32_t rule_buf[IPFW_RULE_SIZE_MAX];
2605         size_t size;
2606         int error;
2607
2608         ioc_rule = (struct ipfw_ioc_rule *)rule_buf;
2609         error = sooptcopyin(sopt, ioc_rule, sizeof(rule_buf),
2610                             sizeof(*ioc_rule));
2611         if (error)
2612                 return error;
2613
2614         size = sopt->sopt_valsize;
2615         error = ipfw_ctl_check_rule(ioc_rule, size);
2616         if (error)
2617                 return error;
2618
2619         error = ipfw_add_rule(&layer3_chain, ioc_rule);
2620         if (error)
2621                 return error;
2622
2623         if (sopt->sopt_dir == SOPT_GET)
2624                 error = sooptcopyout(sopt, ioc_rule, IOC_RULESIZE(ioc_rule));
2625         return error;
2626 }
2627
2628 static void *
2629 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
2630 {
2631         ioc_rule->act_ofs = rule->act_ofs;
2632         ioc_rule->cmd_len = rule->cmd_len;
2633         ioc_rule->rulenum = rule->rulenum;
2634         ioc_rule->set = rule->set;
2635         ioc_rule->usr_flags = rule->usr_flags;
2636
2637         ioc_rule->set_disable = set_disable;
2638         ioc_rule->static_count = static_count;
2639         ioc_rule->static_len = static_ioc_len;
2640
2641         ioc_rule->pcnt = rule->pcnt;
2642         ioc_rule->bcnt = rule->bcnt;
2643         ioc_rule->timestamp = rule->timestamp;
2644
2645         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
2646
2647         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
2648 }
2649
2650 static void
2651 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
2652                 struct ipfw_ioc_state *ioc_state)
2653 {
2654         const struct ipfw_flow_id *id;
2655         struct ipfw_ioc_flowid *ioc_id;
2656
2657         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
2658                             0 : dyn_rule->expire - time_second;
2659         ioc_state->pcnt = dyn_rule->pcnt;
2660         ioc_state->bcnt = dyn_rule->bcnt;
2661
2662         ioc_state->dyn_type = dyn_rule->dyn_type;
2663         ioc_state->count = dyn_rule->count;
2664
2665         ioc_state->rulenum = dyn_rule->rule->rulenum;
2666
2667         id = &dyn_rule->id;
2668         ioc_id = &ioc_state->id;
2669
2670         ioc_id->type = ETHERTYPE_IP;
2671         ioc_id->u.ip.dst_ip = id->dst_ip;
2672         ioc_id->u.ip.src_ip = id->src_ip;
2673         ioc_id->u.ip.dst_port = id->dst_port;
2674         ioc_id->u.ip.src_port = id->src_port;
2675         ioc_id->u.ip.proto = id->proto;
2676 }
2677
2678 static int
2679 ipfw_ctl_get_rules(struct sockopt *sopt)
2680 {
2681         struct ip_fw *rule;
2682         void *buf, *bp;
2683         size_t size;
2684         int error;
2685
2686         /*
2687          * pass up a copy of the current rules. Static rules
2688          * come first (the last of which has number IPFW_DEFAULT_RULE),
2689          * followed by a possibly empty list of dynamic rule.
2690          */
2691         crit_enter();
2692
2693         size = static_ioc_len;  /* size of static rules */
2694         if (ipfw_dyn_v)         /* add size of dyn.rules */
2695                 size += (dyn_count * sizeof(struct ipfw_ioc_state));
2696
2697         /*
2698          * XXX todo: if the user passes a short length just to know
2699          * how much room is needed, do not bother filling up the
2700          * buffer, just jump to the sooptcopyout.
2701          */
2702         bp = buf = kmalloc(size, M_TEMP, M_WAITOK | M_ZERO);
2703
2704         for (rule = layer3_chain; rule; rule = rule->next)
2705                 bp = ipfw_copy_rule(rule, bp);
2706
2707         if (ipfw_dyn_v) {
2708                 struct ipfw_ioc_state *ioc_state;
2709                 int i;
2710
2711                 ioc_state = bp;
2712                 for (i = 0; i < curr_dyn_buckets; i++) {
2713                         ipfw_dyn_rule *p;
2714
2715                         for (p = ipfw_dyn_v[i]; p != NULL;
2716                              p = p->next, ioc_state++)
2717                                 ipfw_copy_state(p, ioc_state);
2718                 }
2719         }
2720
2721         crit_exit();
2722
2723         error = sooptcopyout(sopt, buf, size);
2724         kfree(buf, M_TEMP);
2725         return error;
2726 }
2727
2728 /**
2729  * {set|get}sockopt parser.
2730  */
2731 static int
2732 ipfw_ctl(struct sockopt *sopt)
2733 {
2734         int error, rulenum;
2735         uint32_t masks[2];
2736         size_t size;
2737
2738         /*
2739          * Disallow modifications in really-really secure mode, but still allow
2740          * the logging counters to be reset.
2741          */
2742         if (sopt->sopt_name == IP_FW_ADD ||
2743             (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
2744                 if (securelevel >= 3)
2745                         return EPERM;
2746         }
2747
2748         error = 0;
2749
2750         switch (sopt->sopt_name) {
2751         case IP_FW_GET:
2752                 error = ipfw_ctl_get_rules(sopt);
2753                 break;
2754
2755         case IP_FW_FLUSH:
2756                 /*
2757                  * Normally we cannot release the lock on each iteration.
2758                  * We could do it here only because we start from the head all
2759                  * the times so there is no risk of missing some entries.
2760                  * On the other hand, the risk is that we end up with
2761                  * a very inconsistent ruleset, so better keep the lock
2762                  * around the whole cycle.
2763                  *
2764                  * XXX this code can be improved by resetting the head of
2765                  * the list to point to the default rule, and then freeing
2766                  * the old list without the need for a lock.
2767                  */
2768
2769                 crit_enter();
2770                 free_chain(&layer3_chain, 0 /* keep default rule */);
2771                 crit_exit();
2772                 break;
2773
2774         case IP_FW_ADD:
2775                 error = ipfw_ctl_add_rule(sopt);
2776                 break;
2777
2778         case IP_FW_DEL:
2779                 /*
2780                  * IP_FW_DEL is used for deleting single rules or sets,
2781                  * and (ab)used to atomically manipulate sets. Argument size
2782                  * is used to distinguish between the two:
2783                  *    sizeof(uint32_t)
2784                  *      delete single rule or set of rules,
2785                  *      or reassign rules (or sets) to a different set.
2786                  *    2*sizeof(uint32_t)
2787                  *      atomic disable/enable sets.
2788                  *      first uint32_t contains sets to be disabled,
2789                  *      second uint32_t contains sets to be enabled.
2790                  */
2791                 error = sooptcopyin(sopt, masks,
2792                         sizeof(masks), sizeof(masks[0]));
2793                 if (error)
2794                         break;
2795
2796                 size = sopt->sopt_valsize;
2797                 if (size == sizeof(masks[0])) {
2798                         /*
2799                          * Delete or reassign static rule
2800                          */
2801                         error = del_entry(&layer3_chain, masks[0]);
2802                 } else if (size == sizeof(masks)) {
2803                         /*
2804                          * Set enable/disable
2805                          */
2806                         crit_enter();
2807
2808                         set_disable =
2809                             (set_disable | masks[0]) & ~masks[1] &
2810                             ~(1 << 31); /* set 31 always enabled */
2811
2812                         crit_exit();
2813                 } else {
2814                         error = EINVAL;
2815                 }
2816                 break;
2817
2818         case IP_FW_ZERO:
2819         case IP_FW_RESETLOG: /* argument is an int, the rule number */
2820                 rulenum=0;
2821
2822                 if (sopt->sopt_val != 0) {
2823                     error = sooptcopyin(sopt, &rulenum,
2824                             sizeof(int), sizeof(int));
2825                     if (error)
2826                         break;
2827                 }
2828                 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2829                 break;
2830
2831         default:
2832                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
2833                 error = EINVAL;
2834         }
2835
2836         return error;
2837 }
2838
2839 /*
2840  * This procedure is only used to handle keepalives. It is invoked
2841  * every dyn_keepalive_period
2842  */
2843 static void
2844 ipfw_tick(void * __unused unused)
2845 {
2846         int i;
2847         ipfw_dyn_rule *q;
2848
2849         if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2850                 goto done;
2851
2852         crit_enter();
2853         for (i = 0 ; i < curr_dyn_buckets ; i++) {
2854                 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
2855                         if (q->dyn_type == O_LIMIT_PARENT)
2856                                 continue;
2857                         if (q->id.proto != IPPROTO_TCP)
2858                                 continue;
2859                         if ( (q->state & BOTH_SYN) != BOTH_SYN)
2860                                 continue;
2861                         if (TIME_LEQ( time_second+dyn_keepalive_interval,
2862                             q->expire))
2863                                 continue;       /* too early */
2864                         if (TIME_LEQ(q->expire, time_second))
2865                                 continue;       /* too late, rule expired */
2866
2867                         send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
2868                         send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
2869                 }
2870         }
2871         crit_exit();
2872 done:
2873         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
2874                       ipfw_tick, NULL);
2875 }
2876
2877 static void
2878 ipfw_init_default_rule(struct ip_fw **head)
2879 {
2880         struct ip_fw *def_rule;
2881
2882         KKASSERT(*head == NULL);
2883
2884         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
2885
2886         def_rule->act_ofs = 0;
2887         def_rule->rulenum = IPFW_DEFAULT_RULE;
2888         def_rule->cmd_len = 1;
2889         def_rule->set = 31;
2890
2891         def_rule->cmd[0].len = 1;
2892 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2893         def_rule->cmd[0].opcode = O_ACCEPT;
2894 #else
2895         def_rule->cmd[0].opcode = O_DENY;
2896 #endif
2897
2898         def_rule->refcnt = 1;
2899
2900         *head = def_rule;
2901         ipfw_inc_static_count(def_rule);
2902
2903         /* Install the default rule */
2904         ip_fw_default_rule = def_rule;
2905 }
2906
2907 static void
2908 ipfw_init(void)
2909 {
2910         ip_fw_chk_ptr = ipfw_chk;
2911         ip_fw_ctl_ptr = ipfw_ctl;
2912         ip_fw_dn_io_ptr = ipfw_dummynet_io;
2913
2914         layer3_chain = NULL;
2915         ipfw_init_default_rule(&layer3_chain);
2916
2917         kprintf("ipfw2 initialized, divert %s, "
2918                 "rule-based forwarding enabled, default to %s, logging ",
2919 #ifdef IPDIVERT
2920                 "enabled",
2921 #else
2922                 "disabled",
2923 #endif
2924                 ip_fw_default_rule->cmd[0].opcode == O_ACCEPT ?
2925                 "accept" : "deny");
2926
2927 #ifdef IPFIREWALL_VERBOSE
2928         fw_verbose = 1;
2929 #endif
2930 #ifdef IPFIREWALL_VERBOSE_LIMIT
2931         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2932 #endif
2933         if (fw_verbose == 0)
2934                 kprintf("disabled\n");
2935         else if (verbose_limit == 0)
2936                 kprintf("unlimited\n");
2937         else
2938                 kprintf("limited to %d packets/entry by default\n",
2939                     verbose_limit);
2940         callout_init(&ipfw_timeout_h);
2941         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
2942 }
2943
2944 static int
2945 ipfw_modevent(module_t mod, int type, void *unused)
2946 {
2947         int err = 0;
2948
2949         switch (type) {
2950         case MOD_LOAD:
2951                 crit_enter();
2952                 if (IPFW_LOADED) {
2953                         crit_exit();
2954                         kprintf("IP firewall already loaded\n");
2955                         err = EEXIST;
2956                 } else {
2957                         ipfw_init();
2958                         crit_exit();
2959                 }
2960                 break;
2961
2962         case MOD_UNLOAD:
2963 #ifndef KLD_MODULE
2964                 kprintf("ipfw statically compiled, cannot unload\n");
2965                 err = EBUSY;
2966 #else
2967                 if (ipfw_refcnt != 0) {
2968                         err = EBUSY;
2969                         break;
2970                 }
2971
2972                 crit_enter();
2973                 callout_stop(&ipfw_timeout_h);
2974                 ip_fw_chk_ptr = NULL;
2975                 ip_fw_ctl_ptr = NULL;
2976                 ip_fw_dn_io_ptr = NULL;
2977                 free_chain(&layer3_chain, 1 /* kill default rule */);
2978                 crit_exit();
2979                 kprintf("IP firewall unloaded\n");
2980 #endif
2981                 break;
2982         default:
2983                 break;
2984         }
2985         return err;
2986 }
2987
2988 static moduledata_t ipfwmod = {
2989         "ipfw",
2990         ipfw_modevent,
2991         0
2992 };
2993 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
2994 MODULE_VERSION(ipfw, 1);