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 1.16 2005/06/15 18:46:54 joerg Exp $
  27  */
  28
  29 #define        DEB(x)
  30 #define        DDB(x) x
  31
  32 /*
  33  * Implement IP packet firewall (new version)
  34  */
  35
  36 #if !defined(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 #if IPFW2
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/malloc.h>
  50 #include <sys/mbuf.h>
  51 #include <sys/kernel.h>
  52 #include <sys/proc.h>
  53 #include <sys/socket.h>
  54 #include <sys/socketvar.h>
  55 #include <sys/sysctl.h>
  56 #include <sys/syslog.h>
  57 #include <sys/thread2.h>
  58 #include <sys/ucred.h>
  59 #include <sys/in_cksum.h>
  60 #include <net/if.h>
  61 #include <net/route.h>
  62 #include <netinet/in.h>
  63 #include <netinet/in_systm.h>
  64 #include <netinet/in_var.h>
  65 #include <netinet/in_pcb.h>
  66 #include <netinet/ip.h>
  67 #include <netinet/ip_var.h>
  68 #include <netinet/ip_icmp.h>
  69 #include "ip_fw.h"
  70 #include <net/dummynet/ip_dummynet.h>
  71 #include <netinet/tcp.h>
  72 #include <netinet/tcp_timer.h>
  73 #include <netinet/tcp_var.h>
  74 #include <netinet/tcpip.h>
  75 #include <netinet/udp.h>
  76 #include <netinet/udp_var.h>
  77
  78 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  79
  80 /*
  81  * XXX This one should go in sys/mbuf.h. It is used to avoid that
  82  * a firewall-generated packet loops forever through the firewall.
  83  */
  84 #ifndef M_SKIP_FIREWALL
  85 #define M_SKIP_FIREWALL         0x4000
  86 #endif
  87
  88 /*
  89  * set_disable contains one bit per set value (0..31).
  90  * If the bit is set, all rules with the corresponding set
  91  * are disabled. Set 31 is reserved for the default rule
  92  * and CANNOT be disabled.
  93  */
  94 static u_int32_t set_disable;
  95
  96 static int fw_verbose;
  97 static int verbose_limit;
  98
  99 static struct callout ipfw_timeout_h;
 100 #define IPFW_DEFAULT_RULE       65535
 101
 102 /*
 103  * list of rules for layer 3
 104  */
 105 static struct ip_fw *layer3_chain;
 106
 107 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 108
 109 static int fw_debug = 1;
 110 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
 111
 112 #ifdef SYSCTL_NODE
 113 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 114 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
 115     &fw_enable, 0, "Enable ipfw");
 116 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
 117     &autoinc_step, 0, "Rule number autincrement step");
 118 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 119     &fw_one_pass, 0,
 120     "Only do a single pass through ipfw when using dummynet(4)");
 121 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 122     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 123 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 124     &fw_verbose, 0, "Log matches to ipfw rules");
 125 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 126     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 127
 128 /*
 129  * Description of dynamic rules.
 130  *
 131  * Dynamic rules are stored in lists accessed through a hash table
 132  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 133  * be modified through the sysctl variable dyn_buckets which is
 134  * updated when the table becomes empty.
 135  *
 136  * XXX currently there is only one list, ipfw_dyn.
 137  *
 138  * When a packet is received, its address fields are first masked
 139  * with the mask defined for the rule, then hashed, then matched
 140  * against the entries in the corresponding list.
 141  * Dynamic rules can be used for different purposes:
 142  *  + stateful rules;
 143  *  + enforcing limits on the number of sessions;
 144  *  + in-kernel NAT (not implemented yet)
 145  *
 146  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 147  * measured in seconds and depending on the flags.
 148  *
 149  * The total number of dynamic rules is stored in dyn_count.
 150  * The max number of dynamic rules is dyn_max. When we reach
 151  * the maximum number of rules we do not create anymore. This is
 152  * done to avoid consuming too much memory, but also too much
 153  * time when searching on each packet (ideally, we should try instead
 154  * to put a limit on the length of the list on each bucket...).
 155  *
 156  * Each dynamic rule holds a pointer to the parent ipfw rule so
 157  * we know what action to perform. Dynamic rules are removed when
 158  * the parent rule is deleted. XXX we should make them survive.
 159  *
 160  * There are some limitations with dynamic rules -- we do not
 161  * obey the 'randomized match', and we do not do multiple
 162  * passes through the firewall. XXX check the latter!!!
 163  */
 164 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 165 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
 166 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
 167
 168 /*
 169  * Timeouts for various events in handing dynamic rules.
 170  */
 171 static u_int32_t dyn_ack_lifetime = 300;
 172 static u_int32_t dyn_syn_lifetime = 20;
 173 static u_int32_t dyn_fin_lifetime = 1;
 174 static u_int32_t dyn_rst_lifetime = 1;
 175 static u_int32_t dyn_udp_lifetime = 10;
 176 static u_int32_t dyn_short_lifetime = 5;
 177
 178 /*
 179  * Keepalives are sent if dyn_keepalive is set. They are sent every
 180  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 181  * seconds of lifetime of a rule.
 182  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 183  * than dyn_keepalive_period.
 184  */
 185
 186 static u_int32_t dyn_keepalive_interval = 20;
 187 static u_int32_t dyn_keepalive_period = 5;
 188 static u_int32_t dyn_keepalive = 1;     /* do send keepalives */
 189
 190 static u_int32_t static_count;  /* # of static rules */
 191 static u_int32_t static_len;    /* size in bytes of static rules */
 192 static u_int32_t dyn_count;             /* # of dynamic rules */
 193 static u_int32_t dyn_max = 4096;        /* max # of dynamic rules */
 194
 195 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
 196     &dyn_buckets, 0, "Number of dyn. buckets");
 197 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 198     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 199 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 200     &dyn_count, 0, "Number of dyn. rules");
 201 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 202     &dyn_max, 0, "Max number of dyn. rules");
 203 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 204     &static_count, 0, "Number of static rules");
 205 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 206     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 207 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 208     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 209 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
 210     &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
 211 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
 212     &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
 213 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 214     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 215 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 216     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 217 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 218     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 219
 220 #endif /* SYSCTL_NODE */
 221
 222
 223 static ip_fw_chk_t      ipfw_chk;
 224
 225 ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL;      /* hook into dummynet */
 226
 227 /*
 228  * This macro maps an ip pointer into a layer3 header pointer of type T
 229  */
 230 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
 231
 232 static __inline int
 233 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 234 {
 235         int type = L3HDR(struct icmp,ip)->icmp_type;
 236
 237         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
 238 }
 239
 240 #define TT      ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
 241     (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
 242
 243 static int
 244 is_icmp_query(struct ip *ip)
 245 {
 246         int type = L3HDR(struct icmp, ip)->icmp_type;
 247         return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
 248 }
 249 #undef TT
 250
 251 /*
 252  * The following checks use two arrays of 8 or 16 bits to store the
 253  * bits that we want set or clear, respectively. They are in the
 254  * low and high half of cmd->arg1 or cmd->d[0].
 255  *
 256  * We scan options and store the bits we find set. We succeed if
 257  *
 258  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 259  *
 260  * The code is sometimes optimized not to store additional variables.
 261  */
 262
 263 static int
 264 flags_match(ipfw_insn *cmd, u_int8_t bits)
 265 {
 266         u_char want_clear;
 267         bits = ~bits;
 268
 269         if ( ((cmd->arg1 & 0xff) & bits) != 0)
 270                 return 0; /* some bits we want set were clear */
 271         want_clear = (cmd->arg1 >> 8) & 0xff;
 272         if ( (want_clear & bits) != want_clear)
 273                 return 0; /* some bits we want clear were set */
 274         return 1;
 275 }
 276
 277 static int
 278 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 279 {
 280         int optlen, bits = 0;
 281         u_char *cp = (u_char *)(ip + 1);
 282         int x = (ip->ip_hl << 2) - sizeof (struct ip);
 283
 284         for (; x > 0; x -= optlen, cp += optlen) {
 285                 int opt = cp[IPOPT_OPTVAL];
 286
 287                 if (opt == IPOPT_EOL)
 288                         break;
 289                 if (opt == IPOPT_NOP)
 290                         optlen = 1;
 291                 else {
 292                         optlen = cp[IPOPT_OLEN];
 293                         if (optlen <= 0 || optlen > x)
 294                                 return 0; /* invalid or truncated */
 295                 }
 296                 switch (opt) {
 297
 298                 default:
 299                         break;
 300
 301                 case IPOPT_LSRR:
 302                         bits |= IP_FW_IPOPT_LSRR;
 303                         break;
 304
 305                 case IPOPT_SSRR:
 306                         bits |= IP_FW_IPOPT_SSRR;
 307                         break;
 308
 309                 case IPOPT_RR:
 310                         bits |= IP_FW_IPOPT_RR;
 311                         break;
 312
 313                 case IPOPT_TS:
 314                         bits |= IP_FW_IPOPT_TS;
 315                         break;
 316                 }
 317         }
 318         return (flags_match(cmd, bits));
 319 }
 320
 321 static int
 322 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 323 {
 324         int optlen, bits = 0;
 325         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 326         u_char *cp = (u_char *)(tcp + 1);
 327         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 328
 329         for (; x > 0; x -= optlen, cp += optlen) {
 330                 int opt = cp[0];
 331                 if (opt == TCPOPT_EOL)
 332                         break;
 333                 if (opt == TCPOPT_NOP)
 334                         optlen = 1;
 335                 else {
 336                         optlen = cp[1];
 337                         if (optlen <= 0)
 338                                 break;
 339                 }
 340
 341                 switch (opt) {
 342
 343                 default:
 344                         break;
 345
 346                 case TCPOPT_MAXSEG:
 347                         bits |= IP_FW_TCPOPT_MSS;
 348                         break;
 349
 350                 case TCPOPT_WINDOW:
 351                         bits |= IP_FW_TCPOPT_WINDOW;
 352                         break;
 353
 354                 case TCPOPT_SACK_PERMITTED:
 355                 case TCPOPT_SACK:
 356                         bits |= IP_FW_TCPOPT_SACK;
 357                         break;
 358
 359                 case TCPOPT_TIMESTAMP:
 360                         bits |= IP_FW_TCPOPT_TS;
 361                         break;
 362
 363                 case TCPOPT_CC:
 364                 case TCPOPT_CCNEW:
 365                 case TCPOPT_CCECHO:
 366                         bits |= IP_FW_TCPOPT_CC;
 367                         break;
 368                 }
 369         }
 370         return (flags_match(cmd, bits));
 371 }
 372
 373 static int
 374 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 375 {
 376         if (ifp == NULL)        /* no iface with this packet, match fails */
 377                 return 0;
 378         /* Check by name or by IP address */
 379         if (cmd->name[0] != '\0') { /* match by name */
 380                 /* Check name */
 381                 if (cmd->p.glob) {
 382                         if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
 383                                 return(1);
 384                 } else {
 385                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 386                                 return(1);
 387                 }
 388         } else {
 389                 struct ifaddr *ia;
 390
 391                 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
 392                         if (ia->ifa_addr == NULL)
 393                                 continue;
 394                         if (ia->ifa_addr->sa_family != AF_INET)
 395                                 continue;
 396                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 397                             (ia->ifa_addr))->sin_addr.s_addr)
 398                                 return(1);      /* match */
 399                 }
 400         }
 401         return(0);      /* no match, fail ... */
 402 }
 403
 404 static u_int64_t norule_counter;        /* counter for ipfw_log(NULL...) */
 405
 406 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 407 #define SNP(buf) buf, sizeof(buf)
 408
 409 /*
 410  * We enter here when we have a rule with O_LOG.
 411  * XXX this function alone takes about 2Kbytes of code!
 412  */
 413 static void
 414 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 415         struct mbuf *m, struct ifnet *oif)
 416 {
 417         char *action;
 418         int limit_reached = 0;
 419         char action2[40], proto[48], fragment[28];
 420
 421         fragment[0] = '\0';
 422         proto[0] = '\0';
 423
 424         if (f == NULL) {        /* bogus pkt */
 425                 if (verbose_limit != 0 && norule_counter >= verbose_limit)
 426                         return;
 427                 norule_counter++;
 428                 if (norule_counter == verbose_limit)
 429                         limit_reached = verbose_limit;
 430                 action = "Refuse";
 431         } else {        /* O_LOG is the first action, find the real one */
 432                 ipfw_insn *cmd = ACTION_PTR(f);
 433                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 434
 435                 if (l->max_log != 0 && l->log_left == 0)
 436                         return;
 437                 l->log_left--;
 438                 if (l->log_left == 0)
 439                         limit_reached = l->max_log;
 440                 cmd += F_LEN(cmd);      /* point to first action */
 441                 if (cmd->opcode == O_PROB)
 442                         cmd += F_LEN(cmd);
 443
 444                 action = action2;
 445                 switch (cmd->opcode) {
 446                 case O_DENY:
 447                         action = "Deny";
 448                         break;
 449
 450                 case O_REJECT:
 451                         if (cmd->arg1==ICMP_REJECT_RST)
 452                                 action = "Reset";
 453                         else if (cmd->arg1==ICMP_UNREACH_HOST)
 454                                 action = "Reject";
 455                         else
 456                                 snprintf(SNPARGS(action2, 0), "Unreach %d",
 457                                         cmd->arg1);
 458                         break;
 459
 460                 case O_ACCEPT:
 461                         action = "Accept";
 462                         break;
 463                 case O_COUNT:
 464                         action = "Count";
 465                         break;
 466                 case O_DIVERT:
 467                         snprintf(SNPARGS(action2, 0), "Divert %d",
 468                                 cmd->arg1);
 469                         break;
 470                 case O_TEE:
 471                         snprintf(SNPARGS(action2, 0), "Tee %d",
 472                                 cmd->arg1);
 473                         break;
 474                 case O_SKIPTO:
 475                         snprintf(SNPARGS(action2, 0), "SkipTo %d",
 476                                 cmd->arg1);
 477                         break;
 478                 case O_PIPE:
 479                         snprintf(SNPARGS(action2, 0), "Pipe %d",
 480                                 cmd->arg1);
 481                         break;
 482                 case O_QUEUE:
 483                         snprintf(SNPARGS(action2, 0), "Queue %d",
 484                                 cmd->arg1);
 485                         break;
 486                 case O_FORWARD_IP: {
 487                         ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 488                         int len;
 489
 490                         len = snprintf(SNPARGS(action2, 0), "Forward to %s",
 491                                 inet_ntoa(sa->sa.sin_addr));
 492                         if (sa->sa.sin_port)
 493                                 snprintf(SNPARGS(action2, len), ":%d",
 494                                     sa->sa.sin_port);
 495                         }
 496                         break;
 497                 default:
 498                         action = "UNKNOWN";
 499                         break;
 500                 }
 501         }
 502
 503         if (hlen == 0) {        /* non-ip */
 504                 snprintf(SNPARGS(proto, 0), "MAC");
 505         } else {
 506                 struct ip *ip = mtod(m, struct ip *);
 507                 /* these three are all aliases to the same thing */
 508                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 509                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 510                 struct udphdr *const udp = (struct udphdr *)icmp;
 511
 512                 int ip_off, offset, ip_len;
 513
 514                 int len;
 515
 516                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 517                         ip_off = ntohs(ip->ip_off);
 518                         ip_len = ntohs(ip->ip_len);
 519                 } else {
 520                         ip_off = ip->ip_off;
 521                         ip_len = ip->ip_len;
 522                 }
 523                 offset = ip_off & IP_OFFMASK;
 524                 switch (ip->ip_p) {
 525                 case IPPROTO_TCP:
 526                         len = snprintf(SNPARGS(proto, 0), "TCP %s",
 527                             inet_ntoa(ip->ip_src));
 528                         if (offset == 0)
 529                                 snprintf(SNPARGS(proto, len), ":%d %s:%d",
 530                                     ntohs(tcp->th_sport),
 531                                     inet_ntoa(ip->ip_dst),
 532                                     ntohs(tcp->th_dport));
 533                         else
 534                                 snprintf(SNPARGS(proto, len), " %s",
 535                                     inet_ntoa(ip->ip_dst));
 536                         break;
 537
 538                 case IPPROTO_UDP:
 539                         len = snprintf(SNPARGS(proto, 0), "UDP %s",
 540                                 inet_ntoa(ip->ip_src));
 541                         if (offset == 0)
 542                                 snprintf(SNPARGS(proto, len), ":%d %s:%d",
 543                                     ntohs(udp->uh_sport),
 544                                     inet_ntoa(ip->ip_dst),
 545                                     ntohs(udp->uh_dport));
 546                         else
 547                                 snprintf(SNPARGS(proto, len), " %s",
 548                                     inet_ntoa(ip->ip_dst));
 549                         break;
 550
 551                 case IPPROTO_ICMP:
 552                         if (offset == 0)
 553                                 len = snprintf(SNPARGS(proto, 0),
 554                                     "ICMP:%u.%u ",
 555                                     icmp->icmp_type, icmp->icmp_code);
 556                         else
 557                                 len = snprintf(SNPARGS(proto, 0), "ICMP ");
 558                         len += snprintf(SNPARGS(proto, len), "%s",
 559                             inet_ntoa(ip->ip_src));
 560                         snprintf(SNPARGS(proto, len), " %s",
 561                             inet_ntoa(ip->ip_dst));
 562                         break;
 563
 564                 default:
 565                         len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 566                             inet_ntoa(ip->ip_src));
 567                         snprintf(SNPARGS(proto, len), " %s",
 568                             inet_ntoa(ip->ip_dst));
 569                         break;
 570                 }
 571
 572                 if (ip_off & (IP_MF | IP_OFFMASK))
 573                         snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 574                              ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 575                              offset << 3,
 576                              (ip_off & IP_MF) ? "+" : "");
 577         }
 578         if (oif || m->m_pkthdr.rcvif)
 579                 log(LOG_SECURITY | LOG_INFO,
 580                     "ipfw: %d %s %s %s via %s%s\n",
 581                     f ? f->rulenum : -1,
 582                     action, proto, oif ? "out" : "in",
 583                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 584                     fragment);
 585         else
 586                 log(LOG_SECURITY | LOG_INFO,
 587                     "ipfw: %d %s %s [no if info]%s\n",
 588                     f ? f->rulenum : -1,
 589                     action, proto, fragment);
 590         if (limit_reached)
 591                 log(LOG_SECURITY | LOG_NOTICE,
 592                     "ipfw: limit %d reached on entry %d\n",
 593                     limit_reached, f ? f->rulenum : -1);
 594 }
 595
 596 /*
 597  * IMPORTANT: the hash function for dynamic rules must be commutative
 598  * in source and destination (ip,port), because rules are bidirectional
 599  * and we want to find both in the same bucket.
 600  */
 601 static __inline int
 602 hash_packet(struct ipfw_flow_id *id)
 603 {
 604         u_int32_t i;
 605
 606         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 607         i &= (curr_dyn_buckets - 1);
 608         return i;
 609 }
 610
 611 /**
 612  * unlink a dynamic rule from a chain. prev is a pointer to
 613  * the previous one, q is a pointer to the rule to delete,
 614  * head is a pointer to the head of the queue.
 615  * Modifies q and potentially also head.
 616  */
 617 #define UNLINK_DYN_RULE(prev, head, q) {                                \
 618         ipfw_dyn_rule *old_q = q;                                       \
 619                                                                         \
 620         /* remove a refcount to the parent */                           \
 621         if (q->dyn_type == O_LIMIT)                                     \
 622                 q->parent->count--;                                     \
 623         DEB(printf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \
 624                 (q->id.src_ip), (q->id.src_port),                       \
 625                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
 626         if (prev != NULL)                                               \
 627                 prev->next = q = q->next;                               \
 628         else                                                            \
 629                 head = q = q->next;                                     \
 630         dyn_count--;                                                    \
 631         free(old_q, M_IPFW); }
 632
 633 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
 634
 635 /**
 636  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 637  *
 638  * If keep_me == NULL, rules are deleted even if not expired,
 639  * otherwise only expired rules are removed.
 640  *
 641  * The value of the second parameter is also used to point to identify
 642  * a rule we absolutely do not want to remove (e.g. because we are
 643  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 644  * rules). The pointer is only used for comparison, so any non-null
 645  * value will do.
 646  */
 647 static void
 648 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 649 {
 650         static u_int32_t last_remove = 0;
 651
 652 #define FORCE (keep_me == NULL)
 653
 654         ipfw_dyn_rule *prev, *q;
 655         int i, pass = 0, max_pass = 0;
 656
 657         if (ipfw_dyn_v == NULL || dyn_count == 0)
 658                 return;
 659         /* do not expire more than once per second, it is useless */
 660         if (!FORCE && last_remove == time_second)
 661                 return;
 662         last_remove = time_second;
 663
 664         /*
 665          * because O_LIMIT refer to parent rules, during the first pass only
 666          * remove child and mark any pending LIMIT_PARENT, and remove
 667          * them in a second pass.
 668          */
 669 next_pass:
 670         for (i = 0 ; i < curr_dyn_buckets ; i++) {
 671                 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
 672                         /*
 673                          * Logic can become complex here, so we split tests.
 674                          */
 675                         if (q == keep_me)
 676                                 goto next;
 677                         if (rule != NULL && rule != q->rule)
 678                                 goto next; /* not the one we are looking for */
 679                         if (q->dyn_type == O_LIMIT_PARENT) {
 680                                 /*
 681                                  * handle parent in the second pass,
 682                                  * record we need one.
 683                                  */
 684                                 max_pass = 1;
 685                                 if (pass == 0)
 686                                         goto next;
 687                                 if (FORCE && q->count != 0 ) {
 688                                         /* XXX should not happen! */
 689                                         printf( "OUCH! cannot remove rule,"
 690                                              " count %d\n", q->count);
 691                                 }
 692                         } else {
 693                                 if (!FORCE &&
 694                                     !TIME_LEQ( q->expire, time_second ))
 695                                         goto next;
 696                         }
 697                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 698                         continue;
 699 next:
 700                         prev=q;
 701                         q=q->next;
 702                 }
 703         }
 704         if (pass++ < max_pass)
 705                 goto next_pass;
 706 }
 707
 708
 709 /**
 710  * lookup a dynamic rule.
 711  */
 712 static ipfw_dyn_rule *
 713 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 714         struct tcphdr *tcp)
 715 {
 716         /*
 717          * stateful ipfw extensions.
 718          * Lookup into dynamic session queue
 719          */
 720 #define MATCH_REVERSE   0
 721 #define MATCH_FORWARD   1
 722 #define MATCH_NONE      2
 723 #define MATCH_UNKNOWN   3
 724         int i, dir = MATCH_NONE;
 725         ipfw_dyn_rule *prev, *q=NULL;
 726
 727         if (ipfw_dyn_v == NULL)
 728                 goto done;      /* not found */
 729         i = hash_packet( pkt );
 730         for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
 731                 if (q->dyn_type == O_LIMIT_PARENT)
 732                         goto next;
 733                 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
 734                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 735                         continue;
 736                 }
 737                 if ( pkt->proto == q->id.proto) {
 738                         if (pkt->src_ip == q->id.src_ip &&
 739                             pkt->dst_ip == q->id.dst_ip &&
 740                             pkt->src_port == q->id.src_port &&
 741                             pkt->dst_port == q->id.dst_port ) {
 742                                 dir = MATCH_FORWARD;
 743                                 break;
 744                         }
 745                         if (pkt->src_ip == q->id.dst_ip &&
 746                             pkt->dst_ip == q->id.src_ip &&
 747                             pkt->src_port == q->id.dst_port &&
 748                             pkt->dst_port == q->id.src_port ) {
 749                                 dir = MATCH_REVERSE;
 750                                 break;
 751                         }
 752                 }
 753 next:
 754                 prev = q;
 755                 q = q->next;
 756         }
 757         if (q == NULL)
 758                 goto done; /* q = NULL, not found */
 759
 760         if ( prev != NULL) { /* found and not in front */
 761                 prev->next = q->next;
 762                 q->next = ipfw_dyn_v[i];
 763                 ipfw_dyn_v[i] = q;
 764         }
 765         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 766                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
 767
 768 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 769 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 770                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
 771                 switch (q->state) {
 772                 case TH_SYN:                            /* opening */
 773                         q->expire = time_second + dyn_syn_lifetime;
 774                         break;
 775
 776                 case BOTH_SYN:                  /* move to established */
 777                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
 778                 case BOTH_SYN | (TH_FIN << 8) :
 779                         if (tcp) {
 780 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
 781                             u_int32_t ack = ntohl(tcp->th_ack);
 782                             if (dir == MATCH_FORWARD) {
 783                                 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
 784                                     q->ack_fwd = ack;
 785                                 else { /* ignore out-of-sequence */
 786                                     break;
 787                                 }
 788                             } else {
 789                                 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
 790                                     q->ack_rev = ack;
 791                                 else { /* ignore out-of-sequence */
 792                                     break;
 793                                 }
 794                             }
 795                         }
 796                         q->expire = time_second + dyn_ack_lifetime;
 797                         break;
 798
 799                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 800                         if (dyn_fin_lifetime >= dyn_keepalive_period)
 801                                 dyn_fin_lifetime = dyn_keepalive_period - 1;
 802                         q->expire = time_second + dyn_fin_lifetime;
 803                         break;
 804
 805                 default:
 806 #if 0
 807                         /*
 808                          * reset or some invalid combination, but can also
 809                          * occur if we use keep-state the wrong way.
 810                          */
 811                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
 812                                 printf("invalid state: 0x%x\n", q->state);
 813 #endif
 814                         if (dyn_rst_lifetime >= dyn_keepalive_period)
 815                                 dyn_rst_lifetime = dyn_keepalive_period - 1;
 816                         q->expire = time_second + dyn_rst_lifetime;
 817                         break;
 818                 }
 819         } else if (pkt->proto == IPPROTO_UDP) {
 820                 q->expire = time_second + dyn_udp_lifetime;
 821         } else {
 822                 /* other protocols */
 823                 q->expire = time_second + dyn_short_lifetime;
 824         }
 825 done:
 826         if (match_direction)
 827                 *match_direction = dir;
 828         return q;
 829 }
 830
 831 static void
 832 realloc_dynamic_table(void)
 833 {
 834         /*
 835          * Try reallocation, make sure we have a power of 2 and do
 836          * not allow more than 64k entries. In case of overflow,
 837          * default to 1024.
 838          */
 839
 840         if (dyn_buckets > 65536)
 841                 dyn_buckets = 1024;
 842         if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
 843                 dyn_buckets = curr_dyn_buckets; /* reset */
 844                 return;
 845         }
 846         curr_dyn_buckets = dyn_buckets;
 847         if (ipfw_dyn_v != NULL)
 848                 free(ipfw_dyn_v, M_IPFW);
 849         for (;;) {
 850                 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
 851                        M_IPFW, M_WAITOK | M_ZERO);
 852                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
 853                         break;
 854                 curr_dyn_buckets /= 2;
 855         }
 856 }
 857
 858 /**
 859  * Install state of type 'type' for a dynamic session.
 860  * The hash table contains two type of rules:
 861  * - regular rules (O_KEEP_STATE)
 862  * - rules for sessions with limited number of sess per user
 863  *   (O_LIMIT). When they are created, the parent is
 864  *   increased by 1, and decreased on delete. In this case,
 865  *   the third parameter is the parent rule and not the chain.
 866  * - "parent" rules for the above (O_LIMIT_PARENT).
 867  */
 868 static ipfw_dyn_rule *
 869 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
 870 {
 871         ipfw_dyn_rule *r;
 872         int i;
 873
 874         if (ipfw_dyn_v == NULL ||
 875             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
 876                 realloc_dynamic_table();
 877                 if (ipfw_dyn_v == NULL)
 878                         return NULL; /* failed ! */
 879         }
 880         i = hash_packet(id);
 881
 882         r = malloc(sizeof *r, M_IPFW, M_WAITOK | M_ZERO);
 883         if (r == NULL) {
 884                 printf ("sorry cannot allocate state\n");
 885                 return NULL;
 886         }
 887
 888         /* increase refcount on parent, and set pointer */
 889         if (dyn_type == O_LIMIT) {
 890                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 891                 if ( parent->dyn_type != O_LIMIT_PARENT)
 892                         panic("invalid parent");
 893                 parent->count++;
 894                 r->parent = parent;
 895                 rule = parent->rule;
 896         }
 897
 898         r->id = *id;
 899         r->expire = time_second + dyn_syn_lifetime;
 900         r->rule = rule;
 901         r->dyn_type = dyn_type;
 902         r->pcnt = r->bcnt = 0;
 903         r->count = 0;
 904
 905         r->bucket = i;
 906         r->next = ipfw_dyn_v[i];
 907         ipfw_dyn_v[i] = r;
 908         dyn_count++;
 909         DEB(printf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
 910            dyn_type,
 911            (r->id.src_ip), (r->id.src_port),
 912            (r->id.dst_ip), (r->id.dst_port),
 913            dyn_count ); )
 914         return r;
 915 }
 916
 917 /**
 918  * lookup dynamic parent rule using pkt and rule as search keys.
 919  * If the lookup fails, then install one.
 920  */
 921 static ipfw_dyn_rule *
 922 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
 923 {
 924         ipfw_dyn_rule *q;
 925         int i;
 926
 927         if (ipfw_dyn_v) {
 928                 i = hash_packet( pkt );
 929                 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
 930                         if (q->dyn_type == O_LIMIT_PARENT &&
 931                             rule== q->rule &&
 932                             pkt->proto == q->id.proto &&
 933                             pkt->src_ip == q->id.src_ip &&
 934                             pkt->dst_ip == q->id.dst_ip &&
 935                             pkt->src_port == q->id.src_port &&
 936                             pkt->dst_port == q->id.dst_port) {
 937                                 q->expire = time_second + dyn_short_lifetime;
 938                                 DEB(printf("lookup_dyn_parent found 0x%p\n",q);)
 939                                 return q;
 940                         }
 941         }
 942         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
 943 }
 944
 945 /**
 946  * Install dynamic state for rule type cmd->o.opcode
 947  *
 948  * Returns 1 (failure) if state is not installed because of errors or because
 949  * session limitations are enforced.
 950  */
 951 static int
 952 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
 953         struct ip_fw_args *args)
 954 {
 955         static int last_log;
 956
 957         ipfw_dyn_rule *q;
 958
 959         DEB(printf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
 960             cmd->o.opcode,
 961             (args->f_id.src_ip), (args->f_id.src_port),
 962             (args->f_id.dst_ip), (args->f_id.dst_port) );)
 963
 964         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
 965
 966         if (q != NULL) { /* should never occur */
 967                 if (last_log != time_second) {
 968                         last_log = time_second;
 969                         printf(" install_state: entry already present, done\n");
 970                 }
 971                 return 0;
 972         }
 973
 974         if (dyn_count >= dyn_max)
 975                 /*
 976                  * Run out of slots, try to remove any expired rule.
 977                  */
 978                 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
 979
 980         if (dyn_count >= dyn_max) {
 981                 if (last_log != time_second) {
 982                         last_log = time_second;
 983                         printf("install_state: Too many dynamic rules\n");
 984                 }
 985                 return 1; /* cannot install, notify caller */
 986         }
 987
 988         switch (cmd->o.opcode) {
 989         case O_KEEP_STATE: /* bidir rule */
 990                 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
 991                 break;
 992
 993         case O_LIMIT: /* limit number of sessions */
 994             {
 995                 u_int16_t limit_mask = cmd->limit_mask;
 996                 struct ipfw_flow_id id;
 997                 ipfw_dyn_rule *parent;
 998
 999                 DEB(printf("installing dyn-limit rule %d\n", cmd->conn_limit);)
1000
1001                 id.dst_ip = id.src_ip = 0;
1002                 id.dst_port = id.src_port = 0;
1003                 id.proto = args->f_id.proto;
1004
1005                 if (limit_mask & DYN_SRC_ADDR)
1006                         id.src_ip = args->f_id.src_ip;
1007                 if (limit_mask & DYN_DST_ADDR)
1008                         id.dst_ip = args->f_id.dst_ip;
1009                 if (limit_mask & DYN_SRC_PORT)
1010                         id.src_port = args->f_id.src_port;
1011                 if (limit_mask & DYN_DST_PORT)
1012                         id.dst_port = args->f_id.dst_port;
1013                 parent = lookup_dyn_parent(&id, rule);
1014                 if (parent == NULL) {
1015                         printf("add parent failed\n");
1016                         return 1;
1017                 }
1018                 if (parent->count >= cmd->conn_limit) {
1019                         /*
1020                          * See if we can remove some expired rule.
1021                          */
1022                         remove_dyn_rule(rule, parent);
1023                         if (parent->count >= cmd->conn_limit) {
1024                                 if (fw_verbose && last_log != time_second) {
1025                                         last_log = time_second;
1026                                         log(LOG_SECURITY | LOG_DEBUG,
1027                                             "drop session, too many entries\n");
1028                                 }
1029                                 return 1;
1030                         }
1031                 }
1032                 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1033             }
1034                 break;
1035         default:
1036                 printf("unknown dynamic rule type %u\n", cmd->o.opcode);
1037                 return 1;
1038         }
1039         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1040         return 0;
1041 }
1042
1043 /*
1044  * Transmit a TCP packet, containing either a RST or a keepalive.
1045  * When flags & TH_RST, we are sending a RST packet, because of a
1046  * "reset" action matched the packet.
1047  * Otherwise we are sending a keepalive, and flags & TH_
1048  */
1049 static void
1050 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1051 {
1052         struct mbuf *m;
1053         struct ip *ip;
1054         struct tcphdr *tcp;
1055         struct route sro;       /* fake route */
1056
1057         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1058         if (m == 0)
1059                 return;
1060         m->m_pkthdr.rcvif = (struct ifnet *)0;
1061         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1062         m->m_data += max_linkhdr;
1063
1064         ip = mtod(m, struct ip *);
1065         bzero(ip, m->m_len);
1066         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1067         ip->ip_p = IPPROTO_TCP;
1068         tcp->th_off = 5;
1069         /*
1070          * Assume we are sending a RST (or a keepalive in the reverse
1071          * direction), swap src and destination addresses and ports.
1072          */
1073         ip->ip_src.s_addr = htonl(id->dst_ip);
1074         ip->ip_dst.s_addr = htonl(id->src_ip);
1075         tcp->th_sport = htons(id->dst_port);
1076         tcp->th_dport = htons(id->src_port);
1077         if (flags & TH_RST) {   /* we are sending a RST */
1078                 if (flags & TH_ACK) {
1079                         tcp->th_seq = htonl(ack);
1080                         tcp->th_ack = htonl(0);
1081                         tcp->th_flags = TH_RST;
1082                 } else {
1083                         if (flags & TH_SYN)
1084                                 seq++;
1085                         tcp->th_seq = htonl(0);
1086                         tcp->th_ack = htonl(seq);
1087                         tcp->th_flags = TH_RST | TH_ACK;
1088                 }
1089         } else {
1090                 /*
1091                  * We are sending a keepalive. flags & TH_SYN determines
1092                  * the direction, forward if set, reverse if clear.
1093                  * NOTE: seq and ack are always assumed to be correct
1094                  * as set by the caller. This may be confusing...
1095                  */
1096                 if (flags & TH_SYN) {
1097                         /*
1098                          * we have to rewrite the correct addresses!
1099                          */
1100                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1101                         ip->ip_src.s_addr = htonl(id->src_ip);
1102                         tcp->th_dport = htons(id->dst_port);
1103                         tcp->th_sport = htons(id->src_port);
1104                 }
1105                 tcp->th_seq = htonl(seq);
1106                 tcp->th_ack = htonl(ack);
1107                 tcp->th_flags = TH_ACK;
1108         }
1109         /*
1110          * set ip_len to the payload size so we can compute
1111          * the tcp checksum on the pseudoheader
1112          * XXX check this, could save a couple of words ?
1113          */
1114         ip->ip_len = htons(sizeof(struct tcphdr));
1115         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1116         /*
1117          * now fill fields left out earlier
1118          */
1119         ip->ip_ttl = ip_defttl;
1120         ip->ip_len = m->m_pkthdr.len;
1121         bzero (&sro, sizeof (sro));
1122         ip_rtaddr(ip->ip_dst, &sro);
1123         m->m_flags |= M_SKIP_FIREWALL;
1124         ip_output(m, NULL, &sro, 0, NULL, NULL);
1125         if (sro.ro_rt)
1126                 RTFREE(sro.ro_rt);
1127 }
1128
1129 /*
1130  * sends a reject message, consuming the mbuf passed as an argument.
1131  */
1132 static void
1133 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1134 {
1135
1136         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1137                 /* We need the IP header in host order for icmp_error(). */
1138                 if (args->eh != NULL) {
1139                         struct ip *ip = mtod(args->m, struct ip *);
1140                         ip->ip_len = ntohs(ip->ip_len);
1141                         ip->ip_off = ntohs(ip->ip_off);
1142                 }
1143                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1144         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1145                 struct tcphdr *const tcp =
1146                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1147                 if ( (tcp->th_flags & TH_RST) == 0)
1148                         send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1149                                 ntohl(tcp->th_ack),
1150                                 tcp->th_flags | TH_RST);
1151                 m_freem(args->m);
1152         } else
1153                 m_freem(args->m);
1154         args->m = NULL;
1155 }
1156
1157 /**
1158  *
1159  * Given an ip_fw *, lookup_next_rule will return a pointer
1160  * to the next rule, which can be either the jump
1161  * target (for skipto instructions) or the next one in the list (in
1162  * all other cases including a missing jump target).
1163  * The result is also written in the "next_rule" field of the rule.
1164  * Backward jumps are not allowed, so start looking from the next
1165  * rule...
1166  *
1167  * This never returns NULL -- in case we do not have an exact match,
1168  * the next rule is returned. When the ruleset is changed,
1169  * pointers are flushed so we are always correct.
1170  */
1171
1172 static struct ip_fw *
1173 lookup_next_rule(struct ip_fw *me)
1174 {
1175         struct ip_fw *rule = NULL;
1176         ipfw_insn *cmd;
1177
1178         /* look for action, in case it is a skipto */
1179         cmd = ACTION_PTR(me);
1180         if (cmd->opcode == O_LOG)
1181                 cmd += F_LEN(cmd);
1182         if ( cmd->opcode == O_SKIPTO )
1183                 for (rule = me->next; rule ; rule = rule->next)
1184                         if (rule->rulenum >= cmd->arg1)
1185                                 break;
1186         if (rule == NULL)                       /* failure or not a skipto */
1187                 rule = me->next;
1188         me->next_rule = rule;
1189         return rule;
1190 }
1191
1192 /*
1193  * The main check routine for the firewall.
1194  *
1195  * All arguments are in args so we can modify them and return them
1196  * back to the caller.
1197  *
1198  * Parameters:
1199  *
1200  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1201  *              Starts with the IP header.
1202  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1203  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1204  *              The incoming interface is in the mbuf. (in)
1205  *
1206  *      args->rule      Pointer to the last matching rule (in/out)
1207  *      args->next_hop  Socket we are forwarding to (out).
1208  *      args->f_id      Addresses grabbed from the packet (out)
1209  *
1210  * Return value:
1211  *
1212  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1213  *      0       The packet is to be accepted and routed normally OR
1214  *              the packet was denied/rejected and has been dropped;
1215  *              in the latter case, *m is equal to NULL upon return.
1216  *      port    Divert the packet to port, with these caveats:
1217  *
1218  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1219  *                of diverting it (ie, 'ipfw tee').
1220  *
1221  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1222  *                16 bits as a dummynet pipe number instead of diverting
1223  */
1224
1225 static int
1226 ipfw_chk(struct ip_fw_args *args)
1227 {
1228         /*
1229          * Local variables hold state during the processing of a packet.
1230          *
1231          * IMPORTANT NOTE: to speed up the processing of rules, there
1232          * are some assumption on the values of the variables, which
1233          * are documented here. Should you change them, please check
1234          * the implementation of the various instructions to make sure
1235          * that they still work.
1236          *
1237          * args->eh     The MAC header. It is non-null for a layer2
1238          *      packet, it is NULL for a layer-3 packet.
1239          *
1240          * m | args->m  Pointer to the mbuf, as received from the caller.
1241          *      It may change if ipfw_chk() does an m_pullup, or if it
1242          *      consumes the packet because it calls send_reject().
1243          *      XXX This has to change, so that ipfw_chk() never modifies
1244          *      or consumes the buffer.
1245          * ip   is simply an alias of the value of m, and it is kept
1246          *      in sync with it (the packet is  supposed to start with
1247          *      the ip header).
1248          */
1249         struct mbuf *m = args->m;
1250         struct ip *ip = mtod(m, struct ip *);
1251
1252         /*
1253          * oif | args->oif      If NULL, ipfw_chk has been called on the
1254          *      inbound path (ether_input, bdg_forward, ip_input).
1255          *      If non-NULL, ipfw_chk has been called on the outbound path
1256          *      (ether_output, ip_output).
1257          */
1258         struct ifnet *oif = args->oif;
1259
1260         struct ip_fw *f = NULL;         /* matching rule */
1261         int retval = 0;
1262         struct m_tag *mtag;
1263
1264         /*
1265          * hlen The length of the IPv4 header.
1266          *      hlen >0 means we have an IPv4 packet.
1267          */
1268         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1269
1270         /*
1271          * offset       The offset of a fragment. offset != 0 means that
1272          *      we have a fragment at this offset of an IPv4 packet.
1273          *      offset == 0 means that (if this is an IPv4 packet)
1274          *      this is the first or only fragment.
1275          */
1276         u_short offset = 0;
1277
1278         /*
1279          * Local copies of addresses. They are only valid if we have
1280          * an IP packet.
1281          *
1282          * proto        The protocol. Set to 0 for non-ip packets,
1283          *      or to the protocol read from the packet otherwise.
1284          *      proto != 0 means that we have an IPv4 packet.
1285          *
1286          * src_port, dst_port   port numbers, in HOST format. Only
1287          *      valid for TCP and UDP packets.
1288          *
1289          * src_ip, dst_ip       ip addresses, in NETWORK format.
1290          *      Only valid for IPv4 packets.
1291          */
1292         u_int8_t proto;
1293         u_int16_t src_port = 0, dst_port = 0;   /* NOTE: host format    */
1294         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1295         u_int16_t ip_len=0;
1296         int dyn_dir = MATCH_UNKNOWN;
1297         ipfw_dyn_rule *q = NULL;
1298
1299         if (m->m_flags & M_SKIP_FIREWALL)
1300                 return 0;       /* accept */
1301         /*
1302          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1303          *      MATCH_NONE when checked and not matched (q = NULL),
1304          *      MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1305          */
1306
1307         if (args->eh == NULL ||         /* layer 3 packet */
1308                 ( m->m_pkthdr.len >= sizeof(struct ip) &&
1309                     ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1310                         hlen = ip->ip_hl << 2;
1311
1312         /*
1313          * Collect parameters into local variables for faster matching.
1314          */
1315         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1316                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1317                 goto after_ip_checks;
1318         }
1319
1320         proto = args->f_id.proto = ip->ip_p;
1321         src_ip = ip->ip_src;
1322         dst_ip = ip->ip_dst;
1323         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1324                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1325                 ip_len = ntohs(ip->ip_len);
1326         } else {
1327                 offset = ip->ip_off & IP_OFFMASK;
1328                 ip_len = ip->ip_len;
1329         }
1330
1331 #define PULLUP_TO(len)                                          \
1332                 do {                                            \
1333                         if ((m)->m_len < (len)) {               \
1334                             args->m = m = m_pullup(m, (len));   \
1335                             if (m == 0)                         \
1336                                 goto pullup_failed;             \
1337                             ip = mtod(m, struct ip *);          \
1338                         }                                       \
1339                 } while (0)
1340
1341         if (offset == 0) {
1342                 switch (proto) {
1343                 case IPPROTO_TCP:
1344                     {
1345                         struct tcphdr *tcp;
1346
1347                         PULLUP_TO(hlen + sizeof(struct tcphdr));
1348                         tcp = L3HDR(struct tcphdr, ip);
1349                         dst_port = tcp->th_dport;
1350                         src_port = tcp->th_sport;
1351                         args->f_id.flags = tcp->th_flags;
1352                         }
1353                         break;
1354
1355                 case IPPROTO_UDP:
1356                     {
1357                         struct udphdr *udp;
1358
1359                         PULLUP_TO(hlen + sizeof(struct udphdr));
1360                         udp = L3HDR(struct udphdr, ip);
1361                         dst_port = udp->uh_dport;
1362                         src_port = udp->uh_sport;
1363                         }
1364                         break;
1365
1366                 case IPPROTO_ICMP:
1367                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1368                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1369                         break;
1370
1371                 default:
1372                         break;
1373                 }
1374 #undef PULLUP_TO
1375         }
1376
1377         args->f_id.src_ip = ntohl(src_ip.s_addr);
1378         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1379         args->f_id.src_port = src_port = ntohs(src_port);
1380         args->f_id.dst_port = dst_port = ntohs(dst_port);
1381
1382 after_ip_checks:
1383         if (args->rule) {
1384                 /*
1385                  * Packet has already been tagged. Look for the next rule
1386                  * to restart processing.
1387                  *
1388                  * If fw_one_pass != 0 then just accept it.
1389                  * XXX should not happen here, but optimized out in
1390                  * the caller.
1391                  */
1392                 if (fw_one_pass)
1393                         return 0;
1394
1395                 f = args->rule->next_rule;
1396                 if (f == NULL)
1397                         f = lookup_next_rule(args->rule);
1398         } else {
1399                 /*
1400                  * Find the starting rule. It can be either the first
1401                  * one, or the one after divert_rule if asked so.
1402                  */
1403                 int skipto;
1404
1405                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1406                 if (mtag != NULL)
1407                         skipto = *(u_int16_t *)(mtag + 1);
1408                 else
1409                         skipto = 0;
1410
1411                 f = layer3_chain;
1412                 if (args->eh == NULL && skipto != 0) {
1413                         if (skipto >= IPFW_DEFAULT_RULE)
1414                                 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1415                         while (f && f->rulenum <= skipto)
1416                                 f = f->next;
1417                         if (f == NULL)  /* drop packet */
1418                                 return(IP_FW_PORT_DENY_FLAG);
1419                 }
1420         }
1421         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1422                 m_tag_delete(m, mtag);
1423
1424         /*
1425          * Now scan the rules, and parse microinstructions for each rule.
1426          */
1427         for (; f; f = f->next) {
1428                 int l, cmdlen;
1429                 ipfw_insn *cmd;
1430                 int skip_or; /* skip rest of OR block */
1431
1432 again:
1433                 if (set_disable & (1 << f->set) )
1434                         continue;
1435
1436                 skip_or = 0;
1437                 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1438                     l -= cmdlen, cmd += cmdlen) {
1439                         int match;
1440
1441                         /*
1442                          * check_body is a jump target used when we find a
1443                          * CHECK_STATE, and need to jump to the body of
1444                          * the target rule.
1445                          */
1446
1447 check_body:
1448                         cmdlen = F_LEN(cmd);
1449                         /*
1450                          * An OR block (insn_1 || .. || insn_n) has the
1451                          * F_OR bit set in all but the last instruction.
1452                          * The first match will set "skip_or", and cause
1453                          * the following instructions to be skipped until
1454                          * past the one with the F_OR bit clear.
1455                          */
1456                         if (skip_or) {          /* skip this instruction */
1457                                 if ((cmd->len & F_OR) == 0)
1458                                         skip_or = 0;    /* next one is good */
1459                                 continue;
1460                         }
1461                         match = 0; /* set to 1 if we succeed */
1462
1463                         switch (cmd->opcode) {
1464                         /*
1465                          * The first set of opcodes compares the packet's
1466                          * fields with some pattern, setting 'match' if a
1467                          * match is found. At the end of the loop there is
1468                          * logic to deal with F_NOT and F_OR flags associated
1469                          * with the opcode.
1470                          */
1471                         case O_NOP:
1472                                 match = 1;
1473                                 break;
1474
1475                         case O_FORWARD_MAC:
1476                                 printf("ipfw: opcode %d unimplemented\n",
1477                                     cmd->opcode);
1478                                 break;
1479
1480                         case O_GID:
1481                         case O_UID:
1482                                 /*
1483                                  * We only check offset == 0 && proto != 0,
1484                                  * as this ensures that we have an IPv4
1485                                  * packet with the ports info.
1486                                  */
1487                                 if (offset!=0)
1488                                         break;
1489                             {
1490                                 struct inpcbinfo *pi;
1491                                 int wildcard;
1492                                 struct inpcb *pcb;
1493
1494                                 if (proto == IPPROTO_TCP) {
1495                                         wildcard = 0;
1496                                         pi = &tcbinfo[mycpu->gd_cpuid];
1497                                 } else if (proto == IPPROTO_UDP) {
1498                                         wildcard = 1;
1499                                         pi = &udbinfo;
1500                                 } else
1501                                         break;
1502
1503                                 pcb =  (oif) ?
1504                                         in_pcblookup_hash(pi,
1505                                             dst_ip, htons(dst_port),
1506                                             src_ip, htons(src_port),
1507                                             wildcard, oif) :
1508                                         in_pcblookup_hash(pi,
1509                                             src_ip, htons(src_port),
1510                                             dst_ip, htons(dst_port),
1511                                             wildcard, NULL);
1512
1513                                 if (pcb == NULL || pcb->inp_socket == NULL)
1514                                         break;
1515 #if defined(__DragonFly__) || (defined(__FreeBSD__) && __FreeBSD_version < 500034)
1516 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1517 #endif
1518                                 if (cmd->opcode == O_UID) {
1519                                         match =
1520                                           !socheckuid(pcb->inp_socket,
1521                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1522                                 } else  {
1523                                         match = groupmember(
1524                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1525                                             pcb->inp_socket->so_cred);
1526                                 }
1527                             }
1528                                 break;
1529
1530                         case O_RECV:
1531                                 match = iface_match(m->m_pkthdr.rcvif,
1532                                     (ipfw_insn_if *)cmd);
1533                                 break;
1534
1535                         case O_XMIT:
1536                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1537                                 break;
1538
1539                         case O_VIA:
1540                                 match = iface_match(oif ? oif :
1541                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1542                                 break;
1543
1544                         case O_MACADDR2:
1545                                 if (args->eh != NULL) { /* have MAC header */
1546                                         u_int32_t *want = (u_int32_t *)
1547                                                 ((ipfw_insn_mac *)cmd)->addr;
1548                                         u_int32_t *mask = (u_int32_t *)
1549                                                 ((ipfw_insn_mac *)cmd)->mask;
1550                                         u_int32_t *hdr = (u_int32_t *)args->eh;
1551
1552                                         match =
1553                                             ( want[0] == (hdr[0] & mask[0]) &&
1554                                               want[1] == (hdr[1] & mask[1]) &&
1555                                               want[2] == (hdr[2] & mask[2]) );
1556                                 }
1557                                 break;
1558
1559                         case O_MAC_TYPE:
1560                                 if (args->eh != NULL) {
1561                                         u_int16_t t =
1562                                             ntohs(args->eh->ether_type);
1563                                         u_int16_t *p =
1564                                             ((ipfw_insn_u16 *)cmd)->ports;
1565                                         int i;
1566
1567                                         for (i = cmdlen - 1; !match && i>0;
1568                                             i--, p += 2)
1569                                                 match = (t>=p[0] && t<=p[1]);
1570                                 }
1571                                 break;
1572
1573                         case O_FRAG:
1574                                 match = (hlen > 0 && offset != 0);
1575                                 break;
1576
1577                         case O_IN:      /* "out" is "not in" */
1578                                 match = (oif == NULL);
1579                                 break;
1580
1581                         case O_LAYER2:
1582                                 match = (args->eh != NULL);
1583                                 break;
1584
1585                         case O_PROTO:
1586                                 /*
1587                                  * We do not allow an arg of 0 so the
1588                                  * check of "proto" only suffices.
1589                                  */
1590                                 match = (proto == cmd->arg1);
1591                                 break;
1592
1593                         case O_IP_SRC:
1594                                 match = (hlen > 0 &&
1595                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1596                                     src_ip.s_addr);
1597                                 break;
1598
1599                         case O_IP_SRC_MASK:
1600                                 match = (hlen > 0 &&
1601                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1602                                      (src_ip.s_addr &
1603                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1604                                 break;
1605
1606                         case O_IP_SRC_ME:
1607                                 if (hlen > 0) {
1608                                         struct ifnet *tif;
1609
1610                                         INADDR_TO_IFP(src_ip, tif);
1611                                         match = (tif != NULL);
1612                                 }
1613                                 break;
1614
1615                         case O_IP_DST_SET:
1616                         case O_IP_SRC_SET:
1617                                 if (hlen > 0) {
1618                                         u_int32_t *d = (u_int32_t *)(cmd+1);
1619                                         u_int32_t addr =
1620                                             cmd->opcode == O_IP_DST_SET ?
1621                                                 args->f_id.dst_ip :
1622                                                 args->f_id.src_ip;
1623
1624                                             if (addr < d[0])
1625                                                     break;
1626                                             addr -= d[0]; /* subtract base */
1627                                             match = (addr < cmd->arg1) &&
1628                                                 ( d[ 1 + (addr>>5)] &
1629                                                   (1<<(addr & 0x1f)) );
1630                                 }
1631                                 break;
1632
1633                         case O_IP_DST:
1634                                 match = (hlen > 0 &&
1635                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1636                                     dst_ip.s_addr);
1637                                 break;
1638
1639                         case O_IP_DST_MASK:
1640                                 match = (hlen > 0) &&
1641                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1642                                      (dst_ip.s_addr &
1643                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1644                                 break;
1645
1646                         case O_IP_DST_ME:
1647                                 if (hlen > 0) {
1648                                         struct ifnet *tif;
1649
1650                                         INADDR_TO_IFP(dst_ip, tif);
1651                                         match = (tif != NULL);
1652                                 }
1653                                 break;
1654
1655                         case O_IP_SRCPORT:
1656                         case O_IP_DSTPORT:
1657                                 /*
1658                                  * offset == 0 && proto != 0 is enough
1659                                  * to guarantee that we have an IPv4
1660                                  * packet with port info.
1661                                  */
1662                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1663                                     && offset == 0) {
1664                                         u_int16_t x =
1665                                             (cmd->opcode == O_IP_SRCPORT) ?
1666                                                 src_port : dst_port ;
1667                                         u_int16_t *p =
1668                                             ((ipfw_insn_u16 *)cmd)->ports;
1669                                         int i;
1670
1671                                         for (i = cmdlen - 1; !match && i>0;
1672                                             i--, p += 2)
1673                                                 match = (x>=p[0] && x<=p[1]);
1674                                 }
1675                                 break;
1676
1677                         case O_ICMPTYPE:
1678                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
1679                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
1680                                 break;
1681
1682                         case O_IPOPT:
1683                                 match = (hlen > 0 && ipopts_match(ip, cmd) );
1684                                 break;
1685
1686                         case O_IPVER:
1687                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1688                                 break;
1689
1690                         case O_IPTTL:
1691                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1692                                 break;
1693
1694                         case O_IPID:
1695                                 match = (hlen > 0 &&
1696                                     cmd->arg1 == ntohs(ip->ip_id));
1697                                 break;
1698
1699                         case O_IPLEN:
1700                                 match = (hlen > 0 && cmd->arg1 == ip_len);
1701                                 break;
1702
1703                         case O_IPPRECEDENCE:
1704                                 match = (hlen > 0 &&
1705                                     (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1706                                 break;
1707
1708                         case O_IPTOS:
1709                                 match = (hlen > 0 &&
1710                                     flags_match(cmd, ip->ip_tos));
1711                                 break;
1712
1713                         case O_TCPFLAGS:
1714                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1715                                     flags_match(cmd,
1716                                         L3HDR(struct tcphdr,ip)->th_flags));
1717                                 break;
1718
1719                         case O_TCPOPTS:
1720                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1721                                     tcpopts_match(ip, cmd));
1722                                 break;
1723
1724                         case O_TCPSEQ:
1725                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1726                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1727                                         L3HDR(struct tcphdr,ip)->th_seq);
1728                                 break;
1729
1730                         case O_TCPACK:
1731                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1732                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1733                                         L3HDR(struct tcphdr,ip)->th_ack);
1734                                 break;
1735
1736                         case O_TCPWIN:
1737                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1738                                     cmd->arg1 ==
1739                                         L3HDR(struct tcphdr,ip)->th_win);
1740                                 break;
1741
1742                         case O_ESTAB:
1743                                 /* reject packets which have SYN only */
1744                                 /* XXX should i also check for TH_ACK ? */
1745                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1746                                     (L3HDR(struct tcphdr,ip)->th_flags &
1747                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1748                                 break;
1749
1750                         case O_LOG:
1751                                 if (fw_verbose)
1752                                         ipfw_log(f, hlen, args->eh, m, oif);
1753                                 match = 1;
1754                                 break;
1755
1756                         case O_PROB:
1757                                 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
1758                                 break;
1759
1760                         /*
1761                          * The second set of opcodes represents 'actions',
1762                          * i.e. the terminal part of a rule once the packet
1763                          * matches all previous patterns.
1764                          * Typically there is only one action for each rule,
1765                          * and the opcode is stored at the end of the rule
1766                          * (but there are exceptions -- see below).
1767                          *
1768                          * In general, here we set retval and terminate the
1769                          * outer loop (would be a 'break 3' in some language,
1770                          * but we need to do a 'goto done').
1771                          *
1772                          * Exceptions:
1773                          * O_COUNT and O_SKIPTO actions:
1774                          *   instead of terminating, we jump to the next rule
1775                          *   ('goto next_rule', equivalent to a 'break 2'),
1776                          *   or to the SKIPTO target ('goto again' after
1777                          *   having set f, cmd and l), respectively.
1778                          *
1779                          * O_LIMIT and O_KEEP_STATE: these opcodes are
1780                          *   not real 'actions', and are stored right
1781                          *   before the 'action' part of the rule.
1782                          *   These opcodes try to install an entry in the
1783                          *   state tables; if successful, we continue with
1784                          *   the next opcode (match=1; break;), otherwise
1785                          *   the packet *   must be dropped
1786                          *   ('goto done' after setting retval);
1787                          *
1788                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1789                          *   cause a lookup of the state table, and a jump
1790                          *   to the 'action' part of the parent rule
1791                          *   ('goto check_body') if an entry is found, or
1792                          *   (CHECK_STATE only) a jump to the next rule if
1793                          *   the entry is not found ('goto next_rule').
1794                          *   The result of the lookup is cached to make
1795                          *   further instances of these opcodes are
1796                          *   effectively NOPs.
1797                          */
1798                         case O_LIMIT:
1799                         case O_KEEP_STATE:
1800                                 if (install_state(f,
1801                                     (ipfw_insn_limit *)cmd, args)) {
1802                                         retval = IP_FW_PORT_DENY_FLAG;
1803                                         goto done; /* error/limit violation */
1804                                 }
1805                                 match = 1;
1806                                 break;
1807
1808                         case O_PROBE_STATE:
1809                         case O_CHECK_STATE:
1810                                 /*
1811                                  * dynamic rules are checked at the first
1812                                  * keep-state or check-state occurrence,
1813                                  * with the result being stored in dyn_dir.
1814                                  * The compiler introduces a PROBE_STATE
1815                                  * instruction for us when we have a
1816                                  * KEEP_STATE (because PROBE_STATE needs
1817                                  * to be run first).
1818                                  */
1819                                 if (dyn_dir == MATCH_UNKNOWN &&
1820                                     (q = lookup_dyn_rule(&args->f_id,
1821                                      &dyn_dir, proto == IPPROTO_TCP ?
1822                                         L3HDR(struct tcphdr, ip) : NULL))
1823                                         != NULL) {
1824                                         /*
1825                                          * Found dynamic entry, update stats
1826                                          * and jump to the 'action' part of
1827                                          * the parent rule.
1828                                          */
1829                                         q->pcnt++;
1830                                         q->bcnt += ip_len;
1831                                         f = q->rule;
1832                                         cmd = ACTION_PTR(f);
1833                                         l = f->cmd_len - f->act_ofs;
1834                                         goto check_body;
1835                                 }
1836                                 /*
1837                                  * Dynamic entry not found. If CHECK_STATE,
1838                                  * skip to next rule, if PROBE_STATE just
1839                                  * ignore and continue with next opcode.
1840                                  */
1841                                 if (cmd->opcode == O_CHECK_STATE)
1842                                         goto next_rule;
1843                                 match = 1;
1844                                 break;
1845
1846                         case O_ACCEPT:
1847                                 retval = 0;     /* accept */
1848                                 goto done;
1849
1850                         case O_PIPE:
1851                         case O_QUEUE:
1852                                 args->rule = f; /* report matching rule */
1853                                 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
1854                                 goto done;
1855
1856                         case O_DIVERT:
1857                         case O_TEE:
1858                                 if (args->eh) /* not on layer 2 */
1859                                         break;
1860
1861                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
1862                                     sizeof(u_int16_t), M_NOWAIT);
1863                                 if (mtag == NULL) {
1864                                         retval = IP_FW_PORT_DENY_FLAG;
1865                                         goto done;
1866                                 }
1867                                 *(u_int16_t *)mtag = f->rulenum;
1868                                 m_tag_prepend(m, mtag);
1869                                 retval = (cmd->opcode == O_DIVERT) ?
1870                                     cmd->arg1 :
1871                                     cmd->arg1 | IP_FW_PORT_TEE_FLAG;
1872                                 goto done;
1873
1874                         case O_COUNT:
1875                         case O_SKIPTO:
1876                                 f->pcnt++;      /* update stats */
1877                                 f->bcnt += ip_len;
1878                                 f->timestamp = time_second;
1879                                 if (cmd->opcode == O_COUNT)
1880                                         goto next_rule;
1881                                 /* handle skipto */
1882                                 if (f->next_rule == NULL)
1883                                         lookup_next_rule(f);
1884                                 f = f->next_rule;
1885                                 goto again;
1886
1887                         case O_REJECT:
1888                                 /*
1889                                  * Drop the packet and send a reject notice
1890                                  * if the packet is not ICMP (or is an ICMP
1891                                  * query), and it is not multicast/broadcast.
1892                                  */
1893                                 if (hlen > 0 &&
1894                                     (proto != IPPROTO_ICMP ||
1895                                      is_icmp_query(ip)) &&
1896                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
1897                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
1898                                         send_reject(args, cmd->arg1,
1899                                             offset,ip_len);
1900                                         m = args->m;
1901                                 }
1902                                 /* FALLTHROUGH */
1903                         case O_DENY:
1904                                 retval = IP_FW_PORT_DENY_FLAG;
1905                                 goto done;
1906
1907                         case O_FORWARD_IP:
1908                                 if (args->eh)   /* not valid on layer2 pkts */
1909                                         break;
1910                                 if (!q || dyn_dir == MATCH_FORWARD)
1911                                         args->next_hop =
1912                                             &((ipfw_insn_sa *)cmd)->sa;
1913                                 retval = 0;
1914                                 goto done;
1915
1916                         default:
1917                                 panic("-- unknown opcode %d\n", cmd->opcode);
1918                         } /* end of switch() on opcodes */
1919
1920                         if (cmd->len & F_NOT)
1921                                 match = !match;
1922
1923                         if (match) {
1924                                 if (cmd->len & F_OR)
1925                                         skip_or = 1;
1926                         } else {
1927                                 if (!(cmd->len & F_OR)) /* not an OR block, */
1928                                         break;          /* try next rule    */
1929                         }
1930
1931                 }       /* end of inner for, scan opcodes */
1932
1933 next_rule:;             /* try next rule                */
1934
1935         }               /* end of outer for, scan rules */
1936         printf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
1937         return(IP_FW_PORT_DENY_FLAG);
1938
1939 done:
1940         /* Update statistics */
1941         f->pcnt++;
1942         f->bcnt += ip_len;
1943         f->timestamp = time_second;
1944         return retval;
1945
1946 pullup_failed:
1947         if (fw_verbose)
1948                 printf("pullup failed\n");
1949         return(IP_FW_PORT_DENY_FLAG);
1950 }
1951
1952 /*
1953  * When a rule is added/deleted, clear the next_rule pointers in all rules.
1954  * These will be reconstructed on the fly as packets are matched.
1955  * Must be called at splimp().
1956  */
1957 static void
1958 flush_rule_ptrs(void)
1959 {
1960         struct ip_fw *rule;
1961
1962         for (rule = layer3_chain; rule; rule = rule->next)
1963                 rule->next_rule = NULL;
1964 }
1965
1966 /*
1967  * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
1968  * pipe/queue, or to all of them (match == NULL).
1969  * Must be called at splimp().
1970  */
1971 void
1972 flush_pipe_ptrs(struct dn_flow_set *match)
1973 {
1974         struct ip_fw *rule;
1975
1976         for (rule = layer3_chain; rule; rule = rule->next) {
1977                 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
1978
1979                 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
1980                         continue;
1981                 if (match == NULL || cmd->pipe_ptr == match)
1982                         cmd->pipe_ptr = NULL;
1983         }
1984 }
1985
1986 /*
1987  * Add a new rule to the list. Copy the rule into a malloc'ed area, then
1988  * possibly create a rule number and add the rule to the list.
1989  * Update the rule_number in the input struct so the caller knows it as well.
1990  */
1991 static int
1992 add_rule(struct ip_fw **head, struct ip_fw *input_rule)
1993 {
1994         struct ip_fw *rule, *f, *prev;
1995         int l = RULESIZE(input_rule);
1996
1997         if (*head == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
1998                 return (EINVAL);
1999
2000         rule = malloc(l, M_IPFW, M_WAITOK | M_ZERO);
2001         if (rule == NULL)
2002                 return (ENOSPC);
2003
2004         bcopy(input_rule, rule, l);
2005
2006         rule->next = NULL;
2007         rule->next_rule = NULL;
2008
2009         rule->pcnt = 0;
2010         rule->bcnt = 0;
2011         rule->timestamp = 0;
2012
2013         crit_enter();
2014
2015         if (*head == NULL) {    /* default rule */
2016                 *head = rule;
2017                 goto done;
2018         }
2019
2020         /*
2021          * If rulenum is 0, find highest numbered rule before the
2022          * default rule, and add autoinc_step
2023          */
2024         if (autoinc_step < 1)
2025                 autoinc_step = 1;
2026         else if (autoinc_step > 1000)
2027                 autoinc_step = 1000;
2028         if (rule->rulenum == 0) {
2029                 /*
2030                  * locate the highest numbered rule before default
2031                  */
2032                 for (f = *head; f; f = f->next) {
2033                         if (f->rulenum == IPFW_DEFAULT_RULE)
2034                                 break;
2035                         rule->rulenum = f->rulenum;
2036                 }
2037                 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2038                         rule->rulenum += autoinc_step;
2039                 input_rule->rulenum = rule->rulenum;
2040         }
2041
2042         /*
2043          * Now insert the new rule in the right place in the sorted list.
2044          */
2045         for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2046                 if (f->rulenum > rule->rulenum) { /* found the location */
2047                         if (prev) {
2048                                 rule->next = f;
2049                                 prev->next = rule;
2050                         } else { /* head insert */
2051                                 rule->next = *head;
2052                                 *head = rule;
2053                         }
2054                         break;
2055                 }
2056         }
2057         flush_rule_ptrs();
2058 done:
2059         static_count++;
2060         static_len += l;
2061         crit_exit();
2062         DEB(printf("++ installed rule %d, static count now %d\n",
2063                 rule->rulenum, static_count);)
2064         return (0);
2065 }
2066
2067 /**
2068  * Free storage associated with a static rule (including derived
2069  * dynamic rules).
2070  * The caller is in charge of clearing rule pointers to avoid
2071  * dangling pointers.
2072  * @return a pointer to the next entry.
2073  * Arguments are not checked, so they better be correct.
2074  * Must be called at splimp().
2075  */
2076 static struct ip_fw *
2077 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2078 {
2079         struct ip_fw *n;
2080         int l = RULESIZE(rule);
2081
2082         n = rule->next;
2083         remove_dyn_rule(rule, NULL /* force removal */);
2084         if (prev == NULL)
2085                 *head = n;
2086         else
2087                 prev->next = n;
2088         static_count--;
2089         static_len -= l;
2090
2091         if (DUMMYNET_LOADED)
2092                 ip_dn_ruledel_ptr(rule);
2093         free(rule, M_IPFW);
2094         return n;
2095 }
2096
2097 /*
2098  * Deletes all rules from a chain (including the default rule
2099  * if the second argument is set).
2100  * Must be called at splimp().
2101  */
2102 static void
2103 free_chain(struct ip_fw **chain, int kill_default)
2104 {
2105         struct ip_fw *rule;
2106
2107         flush_rule_ptrs(); /* more efficient to do outside the loop */
2108
2109         while ( (rule = *chain) != NULL &&
2110             (kill_default || rule->rulenum != IPFW_DEFAULT_RULE) )
2111                 delete_rule(chain, NULL, rule);
2112 }
2113
2114 /**
2115  * Remove all rules with given number, and also do set manipulation.
2116  *
2117  * The argument is an u_int32_t. The low 16 bit are the rule or set number,
2118  * the next 8 bits are the new set, the top 8 bits are the command:
2119  *
2120  *      0       delete rules with given number
2121  *      1       delete rules with given set number
2122  *      2       move rules with given number to new set
2123  *      3       move rules with given set number to new set
2124  *      4       swap sets with given numbers
2125  */
2126 static int
2127 del_entry(struct ip_fw **chain, u_int32_t arg)
2128 {
2129         struct ip_fw *prev, *rule;
2130         u_int16_t rulenum;
2131         u_int8_t cmd, new_set;
2132
2133         rulenum = arg & 0xffff;
2134         cmd = (arg >> 24) & 0xff;
2135         new_set = (arg >> 16) & 0xff;
2136
2137         if (cmd > 4)
2138                 return EINVAL;
2139         if (new_set > 30)
2140                 return EINVAL;
2141         if (cmd == 0 || cmd == 2) {
2142                 if (rulenum == IPFW_DEFAULT_RULE)
2143                         return EINVAL;
2144         } else {
2145                 if (rulenum > 30)
2146                         return EINVAL;
2147         }
2148
2149         switch (cmd) {
2150         case 0: /* delete rules with given number */
2151                 /*
2152                  * locate first rule to delete
2153                  */
2154                 for (prev = NULL, rule = *chain;
2155                     rule && rule->rulenum < rulenum;
2156                      prev = rule, rule = rule->next)
2157                         ;
2158                 if (rule->rulenum != rulenum)
2159                         return EINVAL;
2160
2161                 crit_enter(); /* no access to rules while removing */
2162                 /*
2163                  * flush pointers outside the loop, then delete all matching
2164                  * rules. prev remains the same throughout the cycle.
2165                  */
2166                 flush_rule_ptrs();
2167                 while (rule && rule->rulenum == rulenum)
2168                         rule = delete_rule(chain, prev, rule);
2169                 crit_exit();
2170                 break;
2171
2172         case 1: /* delete all rules with given set number */
2173                 crit_enter();
2174                 flush_rule_ptrs();
2175                 for (prev = NULL, rule = *chain; rule ; )
2176                         if (rule->set == rulenum)
2177                                 rule = delete_rule(chain, prev, rule);
2178                         else {
2179                                 prev = rule;
2180                                 rule = rule->next;
2181                         }
2182                 crit_exit();
2183                 break;
2184
2185         case 2: /* move rules with given number to new set */
2186                 crit_enter();
2187                 for (rule = *chain; rule ; rule = rule->next)
2188                         if (rule->rulenum == rulenum)
2189                                 rule->set = new_set;
2190                 crit_exit();
2191                 break;
2192
2193         case 3: /* move rules with given set number to new set */
2194                 crit_enter();
2195                 for (rule = *chain; rule ; rule = rule->next)
2196                         if (rule->set == rulenum)
2197                                 rule->set = new_set;
2198                 crit_exit();
2199                 break;
2200
2201         case 4: /* swap two sets */
2202                 crit_enter();
2203                 for (rule = *chain; rule ; rule = rule->next)
2204                         if (rule->set == rulenum)
2205                                 rule->set = new_set;
2206                         else if (rule->set == new_set)
2207                                 rule->set = rulenum;
2208                 crit_exit();
2209                 break;
2210         }
2211         return 0;
2212 }
2213
2214 /*
2215  * Clear counters for a specific rule.
2216  */
2217 static void
2218 clear_counters(struct ip_fw *rule, int log_only)
2219 {
2220         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2221
2222         if (log_only == 0) {
2223                 rule->bcnt = rule->pcnt = 0;
2224                 rule->timestamp = 0;
2225         }
2226         if (l->o.opcode == O_LOG)
2227                 l->log_left = l->max_log;
2228 }
2229
2230 /**
2231  * Reset some or all counters on firewall rules.
2232  * @arg frwl is null to clear all entries, or contains a specific
2233  * rule number.
2234  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2235  */
2236 static int
2237 zero_entry(int rulenum, int log_only)
2238 {
2239         struct ip_fw *rule;
2240         char *msg;
2241
2242         if (rulenum == 0) {
2243                 crit_enter();
2244                 norule_counter = 0;
2245                 for (rule = layer3_chain; rule; rule = rule->next)
2246                         clear_counters(rule, log_only);
2247                 crit_exit();
2248                 msg = log_only ? "ipfw: All logging counts reset.\n" :
2249                                 "ipfw: Accounting cleared.\n";
2250         } else {
2251                 int cleared = 0;
2252                 /*
2253                  * We can have multiple rules with the same number, so we
2254                  * need to clear them all.
2255                  */
2256                 for (rule = layer3_chain; rule; rule = rule->next)
2257                         if (rule->rulenum == rulenum) {
2258                                 crit_enter();
2259                                 while (rule && rule->rulenum == rulenum) {
2260                                         clear_counters(rule, log_only);
2261                                         rule = rule->next;
2262                                 }
2263                                 crit_exit();
2264                                 cleared = 1;
2265                                 break;
2266                         }
2267                 if (!cleared)   /* we did not find any matching rules */
2268                         return (EINVAL);
2269                 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2270                                 "ipfw: Entry %d cleared.\n";
2271         }
2272         if (fw_verbose)
2273                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2274         return (0);
2275 }
2276
2277 /*
2278  * Check validity of the structure before insert.
2279  * Fortunately rules are simple, so this mostly need to check rule sizes.
2280  */
2281 static int
2282 check_ipfw_struct(struct ip_fw *rule, int size)
2283 {
2284         int l, cmdlen = 0;
2285         int have_action=0;
2286         ipfw_insn *cmd;
2287
2288         if (size < sizeof(*rule)) {
2289                 printf("ipfw: rule too short\n");
2290                 return (EINVAL);
2291         }
2292         /* first, check for valid size */
2293         l = RULESIZE(rule);
2294         if (l != size) {
2295                 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
2296                 return (EINVAL);
2297         }
2298         /*
2299          * Now go for the individual checks. Very simple ones, basically only
2300          * instruction sizes.
2301          */
2302         for (l = rule->cmd_len, cmd = rule->cmd ;
2303                         l > 0 ; l -= cmdlen, cmd += cmdlen) {
2304                 cmdlen = F_LEN(cmd);
2305                 if (cmdlen > l) {
2306                         printf("ipfw: opcode %d size truncated\n",
2307                             cmd->opcode);
2308                         return EINVAL;
2309                 }
2310                 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
2311                 switch (cmd->opcode) {
2312                 case O_NOP:
2313                 case O_PROBE_STATE:
2314                 case O_KEEP_STATE:
2315                 case O_PROTO:
2316                 case O_IP_SRC_ME:
2317                 case O_IP_DST_ME:
2318                 case O_LAYER2:
2319                 case O_IN:
2320                 case O_FRAG:
2321                 case O_IPOPT:
2322                 case O_IPLEN:
2323                 case O_IPID:
2324                 case O_IPTOS:
2325                 case O_IPPRECEDENCE:
2326                 case O_IPTTL:
2327                 case O_IPVER:
2328                 case O_TCPWIN:
2329                 case O_TCPFLAGS:
2330                 case O_TCPOPTS:
2331                 case O_ESTAB:
2332                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2333                                 goto bad_size;
2334                         break;
2335
2336                 case O_UID:
2337                 case O_GID:
2338                 case O_IP_SRC:
2339                 case O_IP_DST:
2340                 case O_TCPSEQ:
2341                 case O_TCPACK:
2342                 case O_PROB:
2343                 case O_ICMPTYPE:
2344                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2345                                 goto bad_size;
2346                         break;
2347
2348                 case O_LIMIT:
2349                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2350                                 goto bad_size;
2351                         break;
2352
2353                 case O_LOG:
2354                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2355                                 goto bad_size;
2356
2357                         ((ipfw_insn_log *)cmd)->log_left =
2358                             ((ipfw_insn_log *)cmd)->max_log;
2359
2360                         break;
2361
2362                 case O_IP_SRC_MASK:
2363                 case O_IP_DST_MASK:
2364                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2365                                 goto bad_size;
2366                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2367                                 printf("ipfw: opcode %d, useless rule\n",
2368                                         cmd->opcode);
2369                                 return EINVAL;
2370                         }
2371                         break;
2372
2373                 case O_IP_SRC_SET:
2374                 case O_IP_DST_SET:
2375                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2376                                 printf("ipfw: invalid set size %d\n",
2377                                         cmd->arg1);
2378                                 return EINVAL;
2379                         }
2380                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2381                             (cmd->arg1+31)/32 )
2382                                 goto bad_size;
2383                         break;
2384
2385                 case O_MACADDR2:
2386                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2387                                 goto bad_size;
2388                         break;
2389
2390                 case O_MAC_TYPE:
2391                 case O_IP_SRCPORT:
2392                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2393                         if (cmdlen < 2 || cmdlen > 31)
2394                                 goto bad_size;
2395                         break;
2396
2397                 case O_RECV:
2398                 case O_XMIT:
2399                 case O_VIA:
2400                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2401                                 goto bad_size;
2402                         break;
2403
2404                 case O_PIPE:
2405                 case O_QUEUE:
2406                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2407                                 goto bad_size;
2408                         goto check_action;
2409
2410                 case O_FORWARD_IP:
2411                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2412                                 goto bad_size;
2413                         goto check_action;
2414
2415                 case O_FORWARD_MAC: /* XXX not implemented yet */
2416                 case O_CHECK_STATE:
2417                 case O_COUNT:
2418                 case O_ACCEPT:
2419                 case O_DENY:
2420                 case O_REJECT:
2421                 case O_SKIPTO:
2422                 case O_DIVERT:
2423                 case O_TEE:
2424                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2425                                 goto bad_size;
2426 check_action:
2427                         if (have_action) {
2428                                 printf("ipfw: opcode %d, multiple actions"
2429                                         " not allowed\n",
2430                                         cmd->opcode);
2431                                 return EINVAL;
2432                         }
2433                         have_action = 1;
2434                         if (l != cmdlen) {
2435                                 printf("ipfw: opcode %d, action must be"
2436                                         " last opcode\n",
2437                                         cmd->opcode);
2438                                 return EINVAL;
2439                         }
2440                         break;
2441                 default:
2442                         printf("ipfw: opcode %d, unknown opcode\n",
2443                                 cmd->opcode);
2444                         return EINVAL;
2445                 }
2446         }
2447         if (have_action == 0) {
2448                 printf("ipfw: missing action\n");
2449                 return EINVAL;
2450         }
2451         return 0;
2452
2453 bad_size:
2454         printf("ipfw: opcode %d size %d wrong\n",
2455                 cmd->opcode, cmdlen);
2456         return EINVAL;
2457 }
2458
2459
2460 /**
2461  * {set|get}sockopt parser.
2462  */
2463 static int
2464 ipfw_ctl(struct sockopt *sopt)
2465 {
2466         int error, rulenum;
2467         size_t size;
2468         struct ip_fw *bp , *buf, *rule;
2469
2470         static u_int32_t rule_buf[255]; /* we copy the data here */
2471
2472         /*
2473          * Disallow modifications in really-really secure mode, but still allow
2474          * the logging counters to be reset.
2475          */
2476         if (sopt->sopt_name == IP_FW_ADD ||
2477             (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
2478 #if defined(__FreeBSD__) && __FreeBSD_version >= 500034
2479                 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2480                 if (error)
2481                         return (error);
2482 #else /* FreeBSD 4.x */
2483                 if (securelevel >= 3)
2484                         return (EPERM);
2485 #endif
2486         }
2487
2488         error = 0;
2489
2490         switch (sopt->sopt_name) {
2491         case IP_FW_GET:
2492                 /*
2493                  * pass up a copy of the current rules. Static rules
2494                  * come first (the last of which has number IPFW_DEFAULT_RULE),
2495                  * followed by a possibly empty list of dynamic rule.
2496                  * The last dynamic rule has NULL in the "next" field.
2497                  */
2498                 crit_enter();
2499                 size = static_len;      /* size of static rules */
2500                 if (ipfw_dyn_v)         /* add size of dyn.rules */
2501                         size += (dyn_count * sizeof(ipfw_dyn_rule));
2502
2503                 /*
2504                  * XXX todo: if the user passes a short length just to know
2505                  * how much room is needed, do not bother filling up the
2506                  * buffer, just jump to the sooptcopyout.
2507                  */
2508                 buf = malloc(size, M_TEMP, M_WAITOK);
2509
2510                 bp = buf;
2511                 for (rule = layer3_chain; rule ; rule = rule->next) {
2512                         int i = RULESIZE(rule);
2513                         bcopy(rule, bp, i);
2514                         /*
2515                          * abuse 'next_rule' to store the set_disable word
2516                          */
2517                         (u_int32_t)(((struct ip_fw *)bp)->next_rule) =
2518                                 set_disable;
2519                         bp = (struct ip_fw *)((char *)bp + i);
2520                 }
2521                 if (ipfw_dyn_v) {
2522                         int i;
2523                         ipfw_dyn_rule *p, *dst, *last = NULL;
2524
2525                         dst = (ipfw_dyn_rule *)bp;
2526                         for (i = 0 ; i < curr_dyn_buckets ; i++ )
2527                                 for ( p = ipfw_dyn_v[i] ; p != NULL ;
2528                                     p = p->next, dst++ ) {
2529                                         bcopy(p, dst, sizeof *p);
2530                                         (int)dst->rule = p->rule->rulenum ;
2531                                         /*
2532                                          * store a non-null value in "next".
2533                                          * The userland code will interpret a
2534                                          * NULL here as a marker
2535                                          * for the last dynamic rule.
2536                                          */
2537                                         dst->next = dst ;
2538                                         last = dst ;
2539                                         dst->expire =
2540                                             TIME_LEQ(dst->expire, time_second) ?
2541                                                 0 : dst->expire - time_second ;
2542                                 }
2543                         if (last != NULL) /* mark last dynamic rule */
2544                                 last->next = NULL;
2545                 }
2546                 crit_exit();
2547
2548                 error = sooptcopyout(sopt, buf, size);
2549                 free(buf, M_TEMP);
2550                 break;
2551
2552         case IP_FW_FLUSH:
2553                 /*
2554                  * Normally we cannot release the lock on each iteration.
2555                  * We could do it here only because we start from the head all
2556                  * the times so there is no risk of missing some entries.
2557                  * On the other hand, the risk is that we end up with
2558                  * a very inconsistent ruleset, so better keep the lock
2559                  * around the whole cycle.
2560                  *
2561                  * XXX this code can be improved by resetting the head of
2562                  * the list to point to the default rule, and then freeing
2563                  * the old list without the need for a lock.
2564                  */
2565
2566                 crit_enter();
2567                 free_chain(&layer3_chain, 0 /* keep default rule */);
2568                 crit_exit();
2569                 break;
2570
2571         case IP_FW_ADD:
2572                 rule = (struct ip_fw *)rule_buf; /* XXX do a malloc */
2573                 error = sooptcopyin(sopt, rule, sizeof(rule_buf),
2574                         sizeof(struct ip_fw) );
2575                 size = sopt->sopt_valsize;
2576                 if (error || (error = check_ipfw_struct(rule, size)))
2577                         break;
2578
2579                 error = add_rule(&layer3_chain, rule);
2580                 size = RULESIZE(rule);
2581                 if (!error && sopt->sopt_dir == SOPT_GET)
2582                         error = sooptcopyout(sopt, rule, size);
2583                 break;
2584
2585         case IP_FW_DEL:
2586                 /*
2587                  * IP_FW_DEL is used for deleting single rules or sets,
2588                  * and (ab)used to atomically manipulate sets. Argument size
2589                  * is used to distinguish between the two:
2590                  *    sizeof(u_int32_t)
2591                  *      delete single rule or set of rules,
2592                  *      or reassign rules (or sets) to a different set.
2593                  *    2*sizeof(u_int32_t)
2594                  *      atomic disable/enable sets.
2595                  *      first u_int32_t contains sets to be disabled,
2596                  *      second u_int32_t contains sets to be enabled.
2597                  */
2598                 error = sooptcopyin(sopt, rule_buf,
2599                         2*sizeof(u_int32_t), sizeof(u_int32_t));
2600                 if (error)
2601                         break;
2602                 size = sopt->sopt_valsize;
2603                 if (size == sizeof(u_int32_t))  /* delete or reassign */
2604                         error = del_entry(&layer3_chain, rule_buf[0]);
2605                 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
2606                         set_disable =
2607                             (set_disable | rule_buf[0]) & ~rule_buf[1] &
2608                             ~(1<<31); /* set 31 always enabled */
2609                 else
2610                         error = EINVAL;
2611                 break;
2612
2613         case IP_FW_ZERO:
2614         case IP_FW_RESETLOG: /* argument is an int, the rule number */
2615                 rulenum=0;
2616
2617                 if (sopt->sopt_val != 0) {
2618                     error = sooptcopyin(sopt, &rulenum,
2619                             sizeof(int), sizeof(int));
2620                     if (error)
2621                         break;
2622                 }
2623                 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2624                 break;
2625
2626         default:
2627                 printf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
2628                 error = EINVAL;
2629         }
2630
2631         return (error);
2632 }
2633
2634 /**
2635  * dummynet needs a reference to the default rule, because rules can be
2636  * deleted while packets hold a reference to them. When this happens,
2637  * dummynet changes the reference to the default rule (it could well be a
2638  * NULL pointer, but this way we do not need to check for the special
2639  * case, plus here he have info on the default behaviour).
2640  */
2641 struct ip_fw *ip_fw_default_rule;
2642
2643 /*
2644  * This procedure is only used to handle keepalives. It is invoked
2645  * every dyn_keepalive_period
2646  */
2647 static void
2648 ipfw_tick(void * __unused unused)
2649 {
2650         int i;
2651         ipfw_dyn_rule *q;
2652
2653         if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2654                 goto done;
2655
2656         crit_enter();
2657         for (i = 0 ; i < curr_dyn_buckets ; i++) {
2658                 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
2659                         if (q->dyn_type == O_LIMIT_PARENT)
2660                                 continue;
2661                         if (q->id.proto != IPPROTO_TCP)
2662                                 continue;
2663                         if ( (q->state & BOTH_SYN) != BOTH_SYN)
2664                                 continue;
2665                         if (TIME_LEQ( time_second+dyn_keepalive_interval,
2666                             q->expire))
2667                                 continue;       /* too early */
2668                         if (TIME_LEQ(q->expire, time_second))
2669                                 continue;       /* too late, rule expired */
2670
2671                         send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
2672                         send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
2673                 }
2674         }
2675         crit_exit();
2676 done:
2677         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
2678                       ipfw_tick, NULL);
2679 }
2680
2681 static void
2682 ipfw_init(void)
2683 {
2684         struct ip_fw default_rule;
2685
2686         ip_fw_chk_ptr = ipfw_chk;
2687         ip_fw_ctl_ptr = ipfw_ctl;
2688         layer3_chain = NULL;
2689
2690         bzero(&default_rule, sizeof default_rule);
2691
2692         default_rule.act_ofs = 0;
2693         default_rule.rulenum = IPFW_DEFAULT_RULE;
2694         default_rule.cmd_len = 1;
2695         default_rule.set = 31;
2696
2697         default_rule.cmd[0].len = 1;
2698         default_rule.cmd[0].opcode =
2699 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2700                                 1 ? O_ACCEPT :
2701 #endif
2702                                 O_DENY;
2703
2704         add_rule(&layer3_chain, &default_rule);
2705
2706         ip_fw_default_rule = layer3_chain;
2707         printf("ipfw2 initialized, divert %s, "
2708                 "rule-based forwarding enabled, default to %s, logging ",
2709 #ifdef IPDIVERT
2710                 "enabled",
2711 #else
2712                 "disabled",
2713 #endif
2714                 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
2715
2716 #ifdef IPFIREWALL_VERBOSE
2717         fw_verbose = 1;
2718 #endif
2719 #ifdef IPFIREWALL_VERBOSE_LIMIT
2720         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2721 #endif
2722         if (fw_verbose == 0)
2723                 printf("disabled\n");
2724         else if (verbose_limit == 0)
2725                 printf("unlimited\n");
2726         else
2727                 printf("limited to %d packets/entry by default\n",
2728                     verbose_limit);
2729         callout_init(&ipfw_timeout_h);
2730         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
2731 }
2732
2733 static int
2734 ipfw_modevent(module_t mod, int type, void *unused)
2735 {
2736         int err = 0;
2737
2738         switch (type) {
2739         case MOD_LOAD:
2740                 crit_enter();
2741                 if (IPFW_LOADED) {
2742                         crit_exit();
2743                         printf("IP firewall already loaded\n");
2744                         err = EEXIST;
2745                 } else {
2746                         ipfw_init();
2747                         crit_exit();
2748                 }
2749                 break;
2750
2751         case MOD_UNLOAD:
2752 #if !defined(KLD_MODULE)
2753                 printf("ipfw statically compiled, cannot unload\n");
2754                 err = EBUSY;
2755 #else
2756                 crit_enter();
2757                 callout_stop(&ipfw_timeout_h);
2758                 ip_fw_chk_ptr = NULL;
2759                 ip_fw_ctl_ptr = NULL;
2760                 free_chain(&layer3_chain, 1 /* kill default rule */);
2761                 crit_exit();
2762                 printf("IP firewall unloaded\n");
2763 #endif
2764                 break;
2765         default:
2766                 break;
2767         }
2768         return err;
2769 }
2770
2771 static moduledata_t ipfwmod = {
2772         "ipfw",
2773         ipfw_modevent,
2774         0
2775 };
2776 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY);
2777 MODULE_VERSION(ipfw, 1);
2778 #endif /* IPFW2 */