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 1.27 2007/09/02 13:27:23 sephe Exp $
  27  */
  28
  29 #define        DEB(x)
  30 #define        DDB(x) x
  31
  32 /*
  33  * Implement IP packet firewall (new version)
  34  */
  35
  36 #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 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/kernel.h>
  51 #include <sys/proc.h>
  52 #include <sys/socket.h>
  53 #include <sys/socketvar.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/syslog.h>
  56 #include <sys/thread2.h>
  57 #include <sys/ucred.h>
  58 #include <sys/in_cksum.h>
  59 #include <net/if.h>
  60 #include <net/route.h>
  61 #include <netinet/in.h>
  62 #include <netinet/in_systm.h>
  63 #include <netinet/in_var.h>
  64 #include <netinet/in_pcb.h>
  65 #include <netinet/ip.h>
  66 #include <netinet/ip_var.h>
  67 #include <netinet/ip_icmp.h>
  68 #include "ip_fw.h"
  69 #include <net/dummynet/ip_dummynet.h>
  70 #include <netinet/tcp.h>
  71 #include <netinet/tcp_timer.h>
  72 #include <netinet/tcp_var.h>
  73 #include <netinet/tcpip.h>
  74 #include <netinet/udp.h>
  75 #include <netinet/udp_var.h>
  76
  77 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  78
  79 /*
  80  * set_disable contains one bit per set value (0..31).
  81  * If the bit is set, all rules with the corresponding set
  82  * are disabled. Set 31 is reserved for the default rule
  83  * and CANNOT be disabled.
  84  */
  85 static u_int32_t set_disable;
  86
  87 static int fw_verbose;
  88 static int verbose_limit;
  89
  90 static struct callout ipfw_timeout_h;
  91 #define IPFW_DEFAULT_RULE       65535
  92
  93 /*
  94  * list of rules for layer 3
  95  */
  96 static struct ip_fw *layer3_chain;
  97
  98 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
  99
 100 static int fw_debug = 1;
 101 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
 102
 103 #ifdef SYSCTL_NODE
 104 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 105 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
 106     &fw_enable, 0, "Enable ipfw");
 107 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
 108     &autoinc_step, 0, "Rule number autincrement step");
 109 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 110     &fw_one_pass, 0,
 111     "Only do a single pass through ipfw when using dummynet(4)");
 112 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 113     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 114 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 115     &fw_verbose, 0, "Log matches to ipfw rules");
 116 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 117     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 118
 119 /*
 120  * Description of dynamic rules.
 121  *
 122  * Dynamic rules are stored in lists accessed through a hash table
 123  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 124  * be modified through the sysctl variable dyn_buckets which is
 125  * updated when the table becomes empty.
 126  *
 127  * XXX currently there is only one list, ipfw_dyn.
 128  *
 129  * When a packet is received, its address fields are first masked
 130  * with the mask defined for the rule, then hashed, then matched
 131  * against the entries in the corresponding list.
 132  * Dynamic rules can be used for different purposes:
 133  *  + stateful rules;
 134  *  + enforcing limits on the number of sessions;
 135  *  + in-kernel NAT (not implemented yet)
 136  *
 137  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 138  * measured in seconds and depending on the flags.
 139  *
 140  * The total number of dynamic rules is stored in dyn_count.
 141  * The max number of dynamic rules is dyn_max. When we reach
 142  * the maximum number of rules we do not create anymore. This is
 143  * done to avoid consuming too much memory, but also too much
 144  * time when searching on each packet (ideally, we should try instead
 145  * to put a limit on the length of the list on each bucket...).
 146  *
 147  * Each dynamic rule holds a pointer to the parent ipfw rule so
 148  * we know what action to perform. Dynamic rules are removed when
 149  * the parent rule is deleted. XXX we should make them survive.
 150  *
 151  * There are some limitations with dynamic rules -- we do not
 152  * obey the 'randomized match', and we do not do multiple
 153  * passes through the firewall. XXX check the latter!!!
 154  */
 155 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 156 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
 157 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
 158
 159 /*
 160  * Timeouts for various events in handing dynamic rules.
 161  */
 162 static u_int32_t dyn_ack_lifetime = 300;
 163 static u_int32_t dyn_syn_lifetime = 20;
 164 static u_int32_t dyn_fin_lifetime = 1;
 165 static u_int32_t dyn_rst_lifetime = 1;
 166 static u_int32_t dyn_udp_lifetime = 10;
 167 static u_int32_t dyn_short_lifetime = 5;
 168
 169 /*
 170  * Keepalives are sent if dyn_keepalive is set. They are sent every
 171  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 172  * seconds of lifetime of a rule.
 173  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 174  * than dyn_keepalive_period.
 175  */
 176
 177 static u_int32_t dyn_keepalive_interval = 20;
 178 static u_int32_t dyn_keepalive_period = 5;
 179 static u_int32_t dyn_keepalive = 1;     /* do send keepalives */
 180
 181 static u_int32_t static_count;  /* # of static rules */
 182 static u_int32_t static_len;    /* size in bytes of static rules */
 183 static u_int32_t dyn_count;             /* # of dynamic rules */
 184 static u_int32_t dyn_max = 4096;        /* max # of dynamic rules */
 185
 186 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
 187     &dyn_buckets, 0, "Number of dyn. buckets");
 188 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 189     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 190 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 191     &dyn_count, 0, "Number of dyn. rules");
 192 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 193     &dyn_max, 0, "Max number of dyn. rules");
 194 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 195     &static_count, 0, "Number of static rules");
 196 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 197     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 198 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 199     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 200 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
 201     &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
 202 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
 203     &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
 204 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 205     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 206 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 207     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 208 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 209     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 210
 211 #endif /* SYSCTL_NODE */
 212
 213
 214 static ip_fw_chk_t      ipfw_chk;
 215
 216 ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL;      /* hook into dummynet */
 217
 218 /*
 219  * This macro maps an ip pointer into a layer3 header pointer of type T
 220  */
 221 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
 222
 223 static __inline int
 224 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 225 {
 226         int type = L3HDR(struct icmp,ip)->icmp_type;
 227
 228         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
 229 }
 230
 231 #define TT      ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
 232     (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
 233
 234 static int
 235 is_icmp_query(struct ip *ip)
 236 {
 237         int type = L3HDR(struct icmp, ip)->icmp_type;
 238         return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
 239 }
 240 #undef TT
 241
 242 /*
 243  * The following checks use two arrays of 8 or 16 bits to store the
 244  * bits that we want set or clear, respectively. They are in the
 245  * low and high half of cmd->arg1 or cmd->d[0].
 246  *
 247  * We scan options and store the bits we find set. We succeed if
 248  *
 249  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 250  *
 251  * The code is sometimes optimized not to store additional variables.
 252  */
 253
 254 static int
 255 flags_match(ipfw_insn *cmd, u_int8_t bits)
 256 {
 257         u_char want_clear;
 258         bits = ~bits;
 259
 260         if ( ((cmd->arg1 & 0xff) & bits) != 0)
 261                 return 0; /* some bits we want set were clear */
 262         want_clear = (cmd->arg1 >> 8) & 0xff;
 263         if ( (want_clear & bits) != want_clear)
 264                 return 0; /* some bits we want clear were set */
 265         return 1;
 266 }
 267
 268 static int
 269 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 270 {
 271         int optlen, bits = 0;
 272         u_char *cp = (u_char *)(ip + 1);
 273         int x = (ip->ip_hl << 2) - sizeof (struct ip);
 274
 275         for (; x > 0; x -= optlen, cp += optlen) {
 276                 int opt = cp[IPOPT_OPTVAL];
 277
 278                 if (opt == IPOPT_EOL)
 279                         break;
 280                 if (opt == IPOPT_NOP)
 281                         optlen = 1;
 282                 else {
 283                         optlen = cp[IPOPT_OLEN];
 284                         if (optlen <= 0 || optlen > x)
 285                                 return 0; /* invalid or truncated */
 286                 }
 287                 switch (opt) {
 288
 289                 default:
 290                         break;
 291
 292                 case IPOPT_LSRR:
 293                         bits |= IP_FW_IPOPT_LSRR;
 294                         break;
 295
 296                 case IPOPT_SSRR:
 297                         bits |= IP_FW_IPOPT_SSRR;
 298                         break;
 299
 300                 case IPOPT_RR:
 301                         bits |= IP_FW_IPOPT_RR;
 302                         break;
 303
 304                 case IPOPT_TS:
 305                         bits |= IP_FW_IPOPT_TS;
 306                         break;
 307                 }
 308         }
 309         return (flags_match(cmd, bits));
 310 }
 311
 312 static int
 313 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 314 {
 315         int optlen, bits = 0;
 316         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 317         u_char *cp = (u_char *)(tcp + 1);
 318         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 319
 320         for (; x > 0; x -= optlen, cp += optlen) {
 321                 int opt = cp[0];
 322                 if (opt == TCPOPT_EOL)
 323                         break;
 324                 if (opt == TCPOPT_NOP)
 325                         optlen = 1;
 326                 else {
 327                         optlen = cp[1];
 328                         if (optlen <= 0)
 329                                 break;
 330                 }
 331
 332                 switch (opt) {
 333
 334                 default:
 335                         break;
 336
 337                 case TCPOPT_MAXSEG:
 338                         bits |= IP_FW_TCPOPT_MSS;
 339                         break;
 340
 341                 case TCPOPT_WINDOW:
 342                         bits |= IP_FW_TCPOPT_WINDOW;
 343                         break;
 344
 345                 case TCPOPT_SACK_PERMITTED:
 346                 case TCPOPT_SACK:
 347                         bits |= IP_FW_TCPOPT_SACK;
 348                         break;
 349
 350                 case TCPOPT_TIMESTAMP:
 351                         bits |= IP_FW_TCPOPT_TS;
 352                         break;
 353
 354                 case TCPOPT_CC:
 355                 case TCPOPT_CCNEW:
 356                 case TCPOPT_CCECHO:
 357                         bits |= IP_FW_TCPOPT_CC;
 358                         break;
 359                 }
 360         }
 361         return (flags_match(cmd, bits));
 362 }
 363
 364 static int
 365 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 366 {
 367         if (ifp == NULL)        /* no iface with this packet, match fails */
 368                 return 0;
 369         /* Check by name or by IP address */
 370         if (cmd->name[0] != '\0') { /* match by name */
 371                 /* Check name */
 372                 if (cmd->p.glob) {
 373                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 374                                 return(1);
 375                 } else {
 376                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 377                                 return(1);
 378                 }
 379         } else {
 380                 struct ifaddr *ia;
 381
 382                 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
 383                         if (ia->ifa_addr == NULL)
 384                                 continue;
 385                         if (ia->ifa_addr->sa_family != AF_INET)
 386                                 continue;
 387                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 388                             (ia->ifa_addr))->sin_addr.s_addr)
 389                                 return(1);      /* match */
 390                 }
 391         }
 392         return(0);      /* no match, fail ... */
 393 }
 394
 395 static u_int64_t norule_counter;        /* counter for ipfw_log(NULL...) */
 396
 397 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 398 #define SNP(buf) buf, sizeof(buf)
 399
 400 /*
 401  * We enter here when we have a rule with O_LOG.
 402  * XXX this function alone takes about 2Kbytes of code!
 403  */
 404 static void
 405 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 406         struct mbuf *m, struct ifnet *oif)
 407 {
 408         char *action;
 409         int limit_reached = 0;
 410         char action2[40], proto[48], fragment[28];
 411
 412         fragment[0] = '\0';
 413         proto[0] = '\0';
 414
 415         if (f == NULL) {        /* bogus pkt */
 416                 if (verbose_limit != 0 && norule_counter >= verbose_limit)
 417                         return;
 418                 norule_counter++;
 419                 if (norule_counter == verbose_limit)
 420                         limit_reached = verbose_limit;
 421                 action = "Refuse";
 422         } else {        /* O_LOG is the first action, find the real one */
 423                 ipfw_insn *cmd = ACTION_PTR(f);
 424                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 425
 426                 if (l->max_log != 0 && l->log_left == 0)
 427                         return;
 428                 l->log_left--;
 429                 if (l->log_left == 0)
 430                         limit_reached = l->max_log;
 431                 cmd += F_LEN(cmd);      /* point to first action */
 432                 if (cmd->opcode == O_PROB)
 433                         cmd += F_LEN(cmd);
 434
 435                 action = action2;
 436                 switch (cmd->opcode) {
 437                 case O_DENY:
 438                         action = "Deny";
 439                         break;
 440
 441                 case O_REJECT:
 442                         if (cmd->arg1==ICMP_REJECT_RST)
 443                                 action = "Reset";
 444                         else if (cmd->arg1==ICMP_UNREACH_HOST)
 445                                 action = "Reject";
 446                         else
 447                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 448                                         cmd->arg1);
 449                         break;
 450
 451                 case O_ACCEPT:
 452                         action = "Accept";
 453                         break;
 454                 case O_COUNT:
 455                         action = "Count";
 456                         break;
 457                 case O_DIVERT:
 458                         ksnprintf(SNPARGS(action2, 0), "Divert %d",
 459                                 cmd->arg1);
 460                         break;
 461                 case O_TEE:
 462                         ksnprintf(SNPARGS(action2, 0), "Tee %d",
 463                                 cmd->arg1);
 464                         break;
 465                 case O_SKIPTO:
 466                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d",
 467                                 cmd->arg1);
 468                         break;
 469                 case O_PIPE:
 470                         ksnprintf(SNPARGS(action2, 0), "Pipe %d",
 471                                 cmd->arg1);
 472                         break;
 473                 case O_QUEUE:
 474                         ksnprintf(SNPARGS(action2, 0), "Queue %d",
 475                                 cmd->arg1);
 476                         break;
 477                 case O_FORWARD_IP: {
 478                         ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 479                         int len;
 480
 481                         len = ksnprintf(SNPARGS(action2, 0), "Forward to %s",
 482                                 inet_ntoa(sa->sa.sin_addr));
 483                         if (sa->sa.sin_port)
 484                                 ksnprintf(SNPARGS(action2, len), ":%d",
 485                                     sa->sa.sin_port);
 486                         }
 487                         break;
 488                 default:
 489                         action = "UNKNOWN";
 490                         break;
 491                 }
 492         }
 493
 494         if (hlen == 0) {        /* non-ip */
 495                 ksnprintf(SNPARGS(proto, 0), "MAC");
 496         } else {
 497                 struct ip *ip = mtod(m, struct ip *);
 498                 /* these three are all aliases to the same thing */
 499                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 500                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 501                 struct udphdr *const udp = (struct udphdr *)icmp;
 502
 503                 int ip_off, offset, ip_len;
 504
 505                 int len;
 506
 507                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 508                         ip_off = ntohs(ip->ip_off);
 509                         ip_len = ntohs(ip->ip_len);
 510                 } else {
 511                         ip_off = ip->ip_off;
 512                         ip_len = ip->ip_len;
 513                 }
 514                 offset = ip_off & IP_OFFMASK;
 515                 switch (ip->ip_p) {
 516                 case IPPROTO_TCP:
 517                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 518                             inet_ntoa(ip->ip_src));
 519                         if (offset == 0)
 520                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 521                                     ntohs(tcp->th_sport),
 522                                     inet_ntoa(ip->ip_dst),
 523                                     ntohs(tcp->th_dport));
 524                         else
 525                                 ksnprintf(SNPARGS(proto, len), " %s",
 526                                     inet_ntoa(ip->ip_dst));
 527                         break;
 528
 529                 case IPPROTO_UDP:
 530                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 531                                 inet_ntoa(ip->ip_src));
 532                         if (offset == 0)
 533                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 534                                     ntohs(udp->uh_sport),
 535                                     inet_ntoa(ip->ip_dst),
 536                                     ntohs(udp->uh_dport));
 537                         else
 538                                 ksnprintf(SNPARGS(proto, len), " %s",
 539                                     inet_ntoa(ip->ip_dst));
 540                         break;
 541
 542                 case IPPROTO_ICMP:
 543                         if (offset == 0)
 544                                 len = ksnprintf(SNPARGS(proto, 0),
 545                                     "ICMP:%u.%u ",
 546                                     icmp->icmp_type, icmp->icmp_code);
 547                         else
 548                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 549                         len += ksnprintf(SNPARGS(proto, len), "%s",
 550                             inet_ntoa(ip->ip_src));
 551                         ksnprintf(SNPARGS(proto, len), " %s",
 552                             inet_ntoa(ip->ip_dst));
 553                         break;
 554
 555                 default:
 556                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 557                             inet_ntoa(ip->ip_src));
 558                         ksnprintf(SNPARGS(proto, len), " %s",
 559                             inet_ntoa(ip->ip_dst));
 560                         break;
 561                 }
 562
 563                 if (ip_off & (IP_MF | IP_OFFMASK))
 564                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 565                              ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 566                              offset << 3,
 567                              (ip_off & IP_MF) ? "+" : "");
 568         }
 569         if (oif || m->m_pkthdr.rcvif)
 570                 log(LOG_SECURITY | LOG_INFO,
 571                     "ipfw: %d %s %s %s via %s%s\n",
 572                     f ? f->rulenum : -1,
 573                     action, proto, oif ? "out" : "in",
 574                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 575                     fragment);
 576         else
 577                 log(LOG_SECURITY | LOG_INFO,
 578                     "ipfw: %d %s %s [no if info]%s\n",
 579                     f ? f->rulenum : -1,
 580                     action, proto, fragment);
 581         if (limit_reached)
 582                 log(LOG_SECURITY | LOG_NOTICE,
 583                     "ipfw: limit %d reached on entry %d\n",
 584                     limit_reached, f ? f->rulenum : -1);
 585 }
 586
 587 /*
 588  * IMPORTANT: the hash function for dynamic rules must be commutative
 589  * in source and destination (ip,port), because rules are bidirectional
 590  * and we want to find both in the same bucket.
 591  */
 592 static __inline int
 593 hash_packet(struct ipfw_flow_id *id)
 594 {
 595         u_int32_t i;
 596
 597         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 598         i &= (curr_dyn_buckets - 1);
 599         return i;
 600 }
 601
 602 /**
 603  * unlink a dynamic rule from a chain. prev is a pointer to
 604  * the previous one, q is a pointer to the rule to delete,
 605  * head is a pointer to the head of the queue.
 606  * Modifies q and potentially also head.
 607  */
 608 #define UNLINK_DYN_RULE(prev, head, q) {                                \
 609         ipfw_dyn_rule *old_q = q;                                       \
 610                                                                         \
 611         /* remove a refcount to the parent */                           \
 612         if (q->dyn_type == O_LIMIT)                                     \
 613                 q->parent->count--;                                     \
 614         DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",        \
 615                 (q->id.src_ip), (q->id.src_port),                       \
 616                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
 617         if (prev != NULL)                                               \
 618                 prev->next = q = q->next;                               \
 619         else                                                            \
 620                 head = q = q->next;                                     \
 621         dyn_count--;                                                    \
 622         kfree(old_q, M_IPFW); }
 623
 624 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
 625
 626 /**
 627  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 628  *
 629  * If keep_me == NULL, rules are deleted even if not expired,
 630  * otherwise only expired rules are removed.
 631  *
 632  * The value of the second parameter is also used to point to identify
 633  * a rule we absolutely do not want to remove (e.g. because we are
 634  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 635  * rules). The pointer is only used for comparison, so any non-null
 636  * value will do.
 637  */
 638 static void
 639 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 640 {
 641         static u_int32_t last_remove = 0;
 642
 643 #define FORCE (keep_me == NULL)
 644
 645         ipfw_dyn_rule *prev, *q;
 646         int i, pass = 0, max_pass = 0;
 647
 648         if (ipfw_dyn_v == NULL || dyn_count == 0)
 649                 return;
 650         /* do not expire more than once per second, it is useless */
 651         if (!FORCE && last_remove == time_second)
 652                 return;
 653         last_remove = time_second;
 654
 655         /*
 656          * because O_LIMIT refer to parent rules, during the first pass only
 657          * remove child and mark any pending LIMIT_PARENT, and remove
 658          * them in a second pass.
 659          */
 660 next_pass:
 661         for (i = 0 ; i < curr_dyn_buckets ; i++) {
 662                 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
 663                         /*
 664                          * Logic can become complex here, so we split tests.
 665                          */
 666                         if (q == keep_me)
 667                                 goto next;
 668                         if (rule != NULL && rule != q->rule)
 669                                 goto next; /* not the one we are looking for */
 670                         if (q->dyn_type == O_LIMIT_PARENT) {
 671                                 /*
 672                                  * handle parent in the second pass,
 673                                  * record we need one.
 674                                  */
 675                                 max_pass = 1;
 676                                 if (pass == 0)
 677                                         goto next;
 678                                 if (FORCE && q->count != 0 ) {
 679                                         /* XXX should not happen! */
 680                                         kprintf( "OUCH! cannot remove rule,"
 681                                              " count %d\n", q->count);
 682                                 }
 683                         } else {
 684                                 if (!FORCE &&
 685                                     !TIME_LEQ( q->expire, time_second ))
 686                                         goto next;
 687                         }
 688                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 689                         continue;
 690 next:
 691                         prev=q;
 692                         q=q->next;
 693                 }
 694         }
 695         if (pass++ < max_pass)
 696                 goto next_pass;
 697 }
 698
 699
 700 /**
 701  * lookup a dynamic rule.
 702  */
 703 static ipfw_dyn_rule *
 704 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 705         struct tcphdr *tcp)
 706 {
 707         /*
 708          * stateful ipfw extensions.
 709          * Lookup into dynamic session queue
 710          */
 711 #define MATCH_REVERSE   0
 712 #define MATCH_FORWARD   1
 713 #define MATCH_NONE      2
 714 #define MATCH_UNKNOWN   3
 715         int i, dir = MATCH_NONE;
 716         ipfw_dyn_rule *prev, *q=NULL;
 717
 718         if (ipfw_dyn_v == NULL)
 719                 goto done;      /* not found */
 720         i = hash_packet( pkt );
 721         for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
 722                 if (q->dyn_type == O_LIMIT_PARENT)
 723                         goto next;
 724                 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
 725                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 726                         continue;
 727                 }
 728                 if ( pkt->proto == q->id.proto) {
 729                         if (pkt->src_ip == q->id.src_ip &&
 730                             pkt->dst_ip == q->id.dst_ip &&
 731                             pkt->src_port == q->id.src_port &&
 732                             pkt->dst_port == q->id.dst_port ) {
 733                                 dir = MATCH_FORWARD;
 734                                 break;
 735                         }
 736                         if (pkt->src_ip == q->id.dst_ip &&
 737                             pkt->dst_ip == q->id.src_ip &&
 738                             pkt->src_port == q->id.dst_port &&
 739                             pkt->dst_port == q->id.src_port ) {
 740                                 dir = MATCH_REVERSE;
 741                                 break;
 742                         }
 743                 }
 744 next:
 745                 prev = q;
 746                 q = q->next;
 747         }
 748         if (q == NULL)
 749                 goto done; /* q = NULL, not found */
 750
 751         if ( prev != NULL) { /* found and not in front */
 752                 prev->next = q->next;
 753                 q->next = ipfw_dyn_v[i];
 754                 ipfw_dyn_v[i] = q;
 755         }
 756         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 757                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
 758
 759 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 760 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 761                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
 762                 switch (q->state) {
 763                 case TH_SYN:                            /* opening */
 764                         q->expire = time_second + dyn_syn_lifetime;
 765                         break;
 766
 767                 case BOTH_SYN:                  /* move to established */
 768                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
 769                 case BOTH_SYN | (TH_FIN << 8) :
 770                         if (tcp) {
 771 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
 772                             u_int32_t ack = ntohl(tcp->th_ack);
 773                             if (dir == MATCH_FORWARD) {
 774                                 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
 775                                     q->ack_fwd = ack;
 776                                 else { /* ignore out-of-sequence */
 777                                     break;
 778                                 }
 779                             } else {
 780                                 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
 781                                     q->ack_rev = ack;
 782                                 else { /* ignore out-of-sequence */
 783                                     break;
 784                                 }
 785                             }
 786                         }
 787                         q->expire = time_second + dyn_ack_lifetime;
 788                         break;
 789
 790                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 791                         if (dyn_fin_lifetime >= dyn_keepalive_period)
 792                                 dyn_fin_lifetime = dyn_keepalive_period - 1;
 793                         q->expire = time_second + dyn_fin_lifetime;
 794                         break;
 795
 796                 default:
 797 #if 0
 798                         /*
 799                          * reset or some invalid combination, but can also
 800                          * occur if we use keep-state the wrong way.
 801                          */
 802                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
 803                                 kprintf("invalid state: 0x%x\n", q->state);
 804 #endif
 805                         if (dyn_rst_lifetime >= dyn_keepalive_period)
 806                                 dyn_rst_lifetime = dyn_keepalive_period - 1;
 807                         q->expire = time_second + dyn_rst_lifetime;
 808                         break;
 809                 }
 810         } else if (pkt->proto == IPPROTO_UDP) {
 811                 q->expire = time_second + dyn_udp_lifetime;
 812         } else {
 813                 /* other protocols */
 814                 q->expire = time_second + dyn_short_lifetime;
 815         }
 816 done:
 817         if (match_direction)
 818                 *match_direction = dir;
 819         return q;
 820 }
 821
 822 static void
 823 realloc_dynamic_table(void)
 824 {
 825         /*
 826          * Try reallocation, make sure we have a power of 2 and do
 827          * not allow more than 64k entries. In case of overflow,
 828          * default to 1024.
 829          */
 830
 831         if (dyn_buckets > 65536)
 832                 dyn_buckets = 1024;
 833         if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
 834                 dyn_buckets = curr_dyn_buckets; /* reset */
 835                 return;
 836         }
 837         curr_dyn_buckets = dyn_buckets;
 838         if (ipfw_dyn_v != NULL)
 839                 kfree(ipfw_dyn_v, M_IPFW);
 840         for (;;) {
 841                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
 842                        M_IPFW, M_WAITOK | M_ZERO);
 843                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
 844                         break;
 845                 curr_dyn_buckets /= 2;
 846         }
 847 }
 848
 849 /**
 850  * Install state of type 'type' for a dynamic session.
 851  * The hash table contains two type of rules:
 852  * - regular rules (O_KEEP_STATE)
 853  * - rules for sessions with limited number of sess per user
 854  *   (O_LIMIT). When they are created, the parent is
 855  *   increased by 1, and decreased on delete. In this case,
 856  *   the third parameter is the parent rule and not the chain.
 857  * - "parent" rules for the above (O_LIMIT_PARENT).
 858  */
 859 static ipfw_dyn_rule *
 860 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
 861 {
 862         ipfw_dyn_rule *r;
 863         int i;
 864
 865         if (ipfw_dyn_v == NULL ||
 866             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
 867                 realloc_dynamic_table();
 868                 if (ipfw_dyn_v == NULL)
 869                         return NULL; /* failed ! */
 870         }
 871         i = hash_packet(id);
 872
 873         r = kmalloc(sizeof *r, M_IPFW, M_WAITOK | M_ZERO);
 874         if (r == NULL) {
 875                 kprintf ("sorry cannot allocate state\n");
 876                 return NULL;
 877         }
 878
 879         /* increase refcount on parent, and set pointer */
 880         if (dyn_type == O_LIMIT) {
 881                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 882                 if ( parent->dyn_type != O_LIMIT_PARENT)
 883                         panic("invalid parent");
 884                 parent->count++;
 885                 r->parent = parent;
 886                 rule = parent->rule;
 887         }
 888
 889         r->id = *id;
 890         r->expire = time_second + dyn_syn_lifetime;
 891         r->rule = rule;
 892         r->dyn_type = dyn_type;
 893         r->pcnt = r->bcnt = 0;
 894         r->count = 0;
 895
 896         r->bucket = i;
 897         r->next = ipfw_dyn_v[i];
 898         ipfw_dyn_v[i] = r;
 899         dyn_count++;
 900         DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
 901            dyn_type,
 902            (r->id.src_ip), (r->id.src_port),
 903            (r->id.dst_ip), (r->id.dst_port),
 904            dyn_count ); )
 905         return r;
 906 }
 907
 908 /**
 909  * lookup dynamic parent rule using pkt and rule as search keys.
 910  * If the lookup fails, then install one.
 911  */
 912 static ipfw_dyn_rule *
 913 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
 914 {
 915         ipfw_dyn_rule *q;
 916         int i;
 917
 918         if (ipfw_dyn_v) {
 919                 i = hash_packet( pkt );
 920                 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
 921                         if (q->dyn_type == O_LIMIT_PARENT &&
 922                             rule== q->rule &&
 923                             pkt->proto == q->id.proto &&
 924                             pkt->src_ip == q->id.src_ip &&
 925                             pkt->dst_ip == q->id.dst_ip &&
 926                             pkt->src_port == q->id.src_port &&
 927                             pkt->dst_port == q->id.dst_port) {
 928                                 q->expire = time_second + dyn_short_lifetime;
 929                                 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
 930                                 return q;
 931                         }
 932         }
 933         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
 934 }
 935
 936 /**
 937  * Install dynamic state for rule type cmd->o.opcode
 938  *
 939  * Returns 1 (failure) if state is not installed because of errors or because
 940  * session limitations are enforced.
 941  */
 942 static int
 943 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
 944         struct ip_fw_args *args)
 945 {
 946         static int last_log;
 947
 948         ipfw_dyn_rule *q;
 949
 950         DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
 951             cmd->o.opcode,
 952             (args->f_id.src_ip), (args->f_id.src_port),
 953             (args->f_id.dst_ip), (args->f_id.dst_port) );)
 954
 955         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
 956
 957         if (q != NULL) { /* should never occur */
 958                 if (last_log != time_second) {
 959                         last_log = time_second;
 960                         kprintf(" install_state: entry already present, done\n");
 961                 }
 962                 return 0;
 963         }
 964
 965         if (dyn_count >= dyn_max)
 966                 /*
 967                  * Run out of slots, try to remove any expired rule.
 968                  */
 969                 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
 970
 971         if (dyn_count >= dyn_max) {
 972                 if (last_log != time_second) {
 973                         last_log = time_second;
 974                         kprintf("install_state: Too many dynamic rules\n");
 975                 }
 976                 return 1; /* cannot install, notify caller */
 977         }
 978
 979         switch (cmd->o.opcode) {
 980         case O_KEEP_STATE: /* bidir rule */
 981                 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
 982                 break;
 983
 984         case O_LIMIT: /* limit number of sessions */
 985             {
 986                 u_int16_t limit_mask = cmd->limit_mask;
 987                 struct ipfw_flow_id id;
 988                 ipfw_dyn_rule *parent;
 989
 990                 DEB(kprintf("installing dyn-limit rule %d\n", cmd->conn_limit);)
 991
 992                 id.dst_ip = id.src_ip = 0;
 993                 id.dst_port = id.src_port = 0;
 994                 id.proto = args->f_id.proto;
 995
 996                 if (limit_mask & DYN_SRC_ADDR)
 997                         id.src_ip = args->f_id.src_ip;
 998                 if (limit_mask & DYN_DST_ADDR)
 999                         id.dst_ip = args->f_id.dst_ip;
1000                 if (limit_mask & DYN_SRC_PORT)
1001                         id.src_port = args->f_id.src_port;
1002                 if (limit_mask & DYN_DST_PORT)
1003                         id.dst_port = args->f_id.dst_port;
1004                 parent = lookup_dyn_parent(&id, rule);
1005                 if (parent == NULL) {
1006                         kprintf("add parent failed\n");
1007                         return 1;
1008                 }
1009                 if (parent->count >= cmd->conn_limit) {
1010                         /*
1011                          * See if we can remove some expired rule.
1012                          */
1013                         remove_dyn_rule(rule, parent);
1014                         if (parent->count >= cmd->conn_limit) {
1015                                 if (fw_verbose && last_log != time_second) {
1016                                         last_log = time_second;
1017                                         log(LOG_SECURITY | LOG_DEBUG,
1018                                             "drop session, too many entries\n");
1019                                 }
1020                                 return 1;
1021                         }
1022                 }
1023                 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1024             }
1025                 break;
1026         default:
1027                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1028                 return 1;
1029         }
1030         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1031         return 0;
1032 }
1033
1034 /*
1035  * Transmit a TCP packet, containing either a RST or a keepalive.
1036  * When flags & TH_RST, we are sending a RST packet, because of a
1037  * "reset" action matched the packet.
1038  * Otherwise we are sending a keepalive, and flags & TH_
1039  */
1040 static void
1041 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1042 {
1043         struct mbuf *m;
1044         struct ip *ip;
1045         struct tcphdr *tcp;
1046         struct route sro;       /* fake route */
1047
1048         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1049         if (m == 0)
1050                 return;
1051         m->m_pkthdr.rcvif = (struct ifnet *)0;
1052         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1053         m->m_data += max_linkhdr;
1054
1055         ip = mtod(m, struct ip *);
1056         bzero(ip, m->m_len);
1057         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1058         ip->ip_p = IPPROTO_TCP;
1059         tcp->th_off = 5;
1060         /*
1061          * Assume we are sending a RST (or a keepalive in the reverse
1062          * direction), swap src and destination addresses and ports.
1063          */
1064         ip->ip_src.s_addr = htonl(id->dst_ip);
1065         ip->ip_dst.s_addr = htonl(id->src_ip);
1066         tcp->th_sport = htons(id->dst_port);
1067         tcp->th_dport = htons(id->src_port);
1068         if (flags & TH_RST) {   /* we are sending a RST */
1069                 if (flags & TH_ACK) {
1070                         tcp->th_seq = htonl(ack);
1071                         tcp->th_ack = htonl(0);
1072                         tcp->th_flags = TH_RST;
1073                 } else {
1074                         if (flags & TH_SYN)
1075                                 seq++;
1076                         tcp->th_seq = htonl(0);
1077                         tcp->th_ack = htonl(seq);
1078                         tcp->th_flags = TH_RST | TH_ACK;
1079                 }
1080         } else {
1081                 /*
1082                  * We are sending a keepalive. flags & TH_SYN determines
1083                  * the direction, forward if set, reverse if clear.
1084                  * NOTE: seq and ack are always assumed to be correct
1085                  * as set by the caller. This may be confusing...
1086                  */
1087                 if (flags & TH_SYN) {
1088                         /*
1089                          * we have to rewrite the correct addresses!
1090                          */
1091                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1092                         ip->ip_src.s_addr = htonl(id->src_ip);
1093                         tcp->th_dport = htons(id->dst_port);
1094                         tcp->th_sport = htons(id->src_port);
1095                 }
1096                 tcp->th_seq = htonl(seq);
1097                 tcp->th_ack = htonl(ack);
1098                 tcp->th_flags = TH_ACK;
1099         }
1100         /*
1101          * set ip_len to the payload size so we can compute
1102          * the tcp checksum on the pseudoheader
1103          * XXX check this, could save a couple of words ?
1104          */
1105         ip->ip_len = htons(sizeof(struct tcphdr));
1106         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1107         /*
1108          * now fill fields left out earlier
1109          */
1110         ip->ip_ttl = ip_defttl;
1111         ip->ip_len = m->m_pkthdr.len;
1112         bzero (&sro, sizeof (sro));
1113         ip_rtaddr(ip->ip_dst, &sro);
1114         m->m_pkthdr.fw_flags |= IPFW_MBUF_SKIP_FIREWALL;
1115         ip_output(m, NULL, &sro, 0, NULL, NULL);
1116         if (sro.ro_rt)
1117                 RTFREE(sro.ro_rt);
1118 }
1119
1120 /*
1121  * sends a reject message, consuming the mbuf passed as an argument.
1122  */
1123 static void
1124 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1125 {
1126
1127         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1128                 /* We need the IP header in host order for icmp_error(). */
1129                 if (args->eh != NULL) {
1130                         struct ip *ip = mtod(args->m, struct ip *);
1131                         ip->ip_len = ntohs(ip->ip_len);
1132                         ip->ip_off = ntohs(ip->ip_off);
1133                 }
1134                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1135         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1136                 struct tcphdr *const tcp =
1137                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1138                 if ( (tcp->th_flags & TH_RST) == 0)
1139                         send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1140                                 ntohl(tcp->th_ack),
1141                                 tcp->th_flags | TH_RST);
1142                 m_freem(args->m);
1143         } else
1144                 m_freem(args->m);
1145         args->m = NULL;
1146 }
1147
1148 /**
1149  *
1150  * Given an ip_fw *, lookup_next_rule will return a pointer
1151  * to the next rule, which can be either the jump
1152  * target (for skipto instructions) or the next one in the list (in
1153  * all other cases including a missing jump target).
1154  * The result is also written in the "next_rule" field of the rule.
1155  * Backward jumps are not allowed, so start looking from the next
1156  * rule...
1157  *
1158  * This never returns NULL -- in case we do not have an exact match,
1159  * the next rule is returned. When the ruleset is changed,
1160  * pointers are flushed so we are always correct.
1161  */
1162
1163 static struct ip_fw *
1164 lookup_next_rule(struct ip_fw *me)
1165 {
1166         struct ip_fw *rule = NULL;
1167         ipfw_insn *cmd;
1168
1169         /* look for action, in case it is a skipto */
1170         cmd = ACTION_PTR(me);
1171         if (cmd->opcode == O_LOG)
1172                 cmd += F_LEN(cmd);
1173         if ( cmd->opcode == O_SKIPTO )
1174                 for (rule = me->next; rule ; rule = rule->next)
1175                         if (rule->rulenum >= cmd->arg1)
1176                                 break;
1177         if (rule == NULL)                       /* failure or not a skipto */
1178                 rule = me->next;
1179         me->next_rule = rule;
1180         return rule;
1181 }
1182
1183 /*
1184  * The main check routine for the firewall.
1185  *
1186  * All arguments are in args so we can modify them and return them
1187  * back to the caller.
1188  *
1189  * Parameters:
1190  *
1191  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1192  *              Starts with the IP header.
1193  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1194  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1195  *              The incoming interface is in the mbuf. (in)
1196  *
1197  *      args->rule      Pointer to the last matching rule (in/out)
1198  *      args->next_hop  Socket we are forwarding to (out).
1199  *      args->f_id      Addresses grabbed from the packet (out)
1200  *
1201  * Return value:
1202  *
1203  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1204  *      0       The packet is to be accepted and routed normally OR
1205  *              the packet was denied/rejected and has been dropped;
1206  *              in the latter case, *m is equal to NULL upon return.
1207  *      port    Divert the packet to port, with these caveats:
1208  *
1209  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1210  *                of diverting it (ie, 'ipfw tee').
1211  *
1212  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1213  *                16 bits as a dummynet pipe number instead of diverting
1214  */
1215
1216 static int
1217 ipfw_chk(struct ip_fw_args *args)
1218 {
1219         /*
1220          * Local variables hold state during the processing of a packet.
1221          *
1222          * IMPORTANT NOTE: to speed up the processing of rules, there
1223          * are some assumption on the values of the variables, which
1224          * are documented here. Should you change them, please check
1225          * the implementation of the various instructions to make sure
1226          * that they still work.
1227          *
1228          * args->eh     The MAC header. It is non-null for a layer2
1229          *      packet, it is NULL for a layer-3 packet.
1230          *
1231          * m | args->m  Pointer to the mbuf, as received from the caller.
1232          *      It may change if ipfw_chk() does an m_pullup, or if it
1233          *      consumes the packet because it calls send_reject().
1234          *      XXX This has to change, so that ipfw_chk() never modifies
1235          *      or consumes the buffer.
1236          * ip   is simply an alias of the value of m, and it is kept
1237          *      in sync with it (the packet is  supposed to start with
1238          *      the ip header).
1239          */
1240         struct mbuf *m = args->m;
1241         struct ip *ip = mtod(m, struct ip *);
1242
1243         /*
1244          * oif | args->oif      If NULL, ipfw_chk has been called on the
1245          *      inbound path (ether_input, ip_input).
1246          *      If non-NULL, ipfw_chk has been called on the outbound path
1247          *      (ether_output, ip_output).
1248          */
1249         struct ifnet *oif = args->oif;
1250
1251         struct ip_fw *f = NULL;         /* matching rule */
1252         int retval = 0;
1253         struct m_tag *mtag;
1254
1255         /*
1256          * hlen The length of the IPv4 header.
1257          *      hlen >0 means we have an IPv4 packet.
1258          */
1259         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1260
1261         /*
1262          * offset       The offset of a fragment. offset != 0 means that
1263          *      we have a fragment at this offset of an IPv4 packet.
1264          *      offset == 0 means that (if this is an IPv4 packet)
1265          *      this is the first or only fragment.
1266          */
1267         u_short offset = 0;
1268
1269         /*
1270          * Local copies of addresses. They are only valid if we have
1271          * an IP packet.
1272          *
1273          * proto        The protocol. Set to 0 for non-ip packets,
1274          *      or to the protocol read from the packet otherwise.
1275          *      proto != 0 means that we have an IPv4 packet.
1276          *
1277          * src_port, dst_port   port numbers, in HOST format. Only
1278          *      valid for TCP and UDP packets.
1279          *
1280          * src_ip, dst_ip       ip addresses, in NETWORK format.
1281          *      Only valid for IPv4 packets.
1282          */
1283         u_int8_t proto;
1284         u_int16_t src_port = 0, dst_port = 0;   /* NOTE: host format    */
1285         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1286         u_int16_t ip_len=0;
1287         int dyn_dir = MATCH_UNKNOWN;
1288         ipfw_dyn_rule *q = NULL;
1289
1290         if (m->m_pkthdr.fw_flags & IPFW_MBUF_SKIP_FIREWALL)
1291                 return 0;       /* accept */
1292         /*
1293          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1294          *      MATCH_NONE when checked and not matched (q = NULL),
1295          *      MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1296          */
1297
1298         if (args->eh == NULL ||         /* layer 3 packet */
1299                 ( m->m_pkthdr.len >= sizeof(struct ip) &&
1300                     ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1301                         hlen = ip->ip_hl << 2;
1302
1303         /*
1304          * Collect parameters into local variables for faster matching.
1305          */
1306         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1307                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1308                 goto after_ip_checks;
1309         }
1310
1311         proto = args->f_id.proto = ip->ip_p;
1312         src_ip = ip->ip_src;
1313         dst_ip = ip->ip_dst;
1314         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1315                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1316                 ip_len = ntohs(ip->ip_len);
1317         } else {
1318                 offset = ip->ip_off & IP_OFFMASK;
1319                 ip_len = ip->ip_len;
1320         }
1321
1322 #define PULLUP_TO(len)                                          \
1323                 do {                                            \
1324                         if ((m)->m_len < (len)) {               \
1325                             args->m = m = m_pullup(m, (len));   \
1326                             if (m == 0)                         \
1327                                 goto pullup_failed;             \
1328                             ip = mtod(m, struct ip *);          \
1329                         }                                       \
1330                 } while (0)
1331
1332         if (offset == 0) {
1333                 switch (proto) {
1334                 case IPPROTO_TCP:
1335                     {
1336                         struct tcphdr *tcp;
1337
1338                         PULLUP_TO(hlen + sizeof(struct tcphdr));
1339                         tcp = L3HDR(struct tcphdr, ip);
1340                         dst_port = tcp->th_dport;
1341                         src_port = tcp->th_sport;
1342                         args->f_id.flags = tcp->th_flags;
1343                         }
1344                         break;
1345
1346                 case IPPROTO_UDP:
1347                     {
1348                         struct udphdr *udp;
1349
1350                         PULLUP_TO(hlen + sizeof(struct udphdr));
1351                         udp = L3HDR(struct udphdr, ip);
1352                         dst_port = udp->uh_dport;
1353                         src_port = udp->uh_sport;
1354                         }
1355                         break;
1356
1357                 case IPPROTO_ICMP:
1358                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1359                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1360                         break;
1361
1362                 default:
1363                         break;
1364                 }
1365 #undef PULLUP_TO
1366         }
1367
1368         args->f_id.src_ip = ntohl(src_ip.s_addr);
1369         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1370         args->f_id.src_port = src_port = ntohs(src_port);
1371         args->f_id.dst_port = dst_port = ntohs(dst_port);
1372
1373 after_ip_checks:
1374         if (args->rule) {
1375                 /*
1376                  * Packet has already been tagged. Look for the next rule
1377                  * to restart processing.
1378                  *
1379                  * If fw_one_pass != 0 then just accept it.
1380                  * XXX should not happen here, but optimized out in
1381                  * the caller.
1382                  */
1383                 if (fw_one_pass)
1384                         return 0;
1385
1386                 f = args->rule->next_rule;
1387                 if (f == NULL)
1388                         f = lookup_next_rule(args->rule);
1389         } else {
1390                 /*
1391                  * Find the starting rule. It can be either the first
1392                  * one, or the one after divert_rule if asked so.
1393                  */
1394                 int skipto;
1395
1396                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1397                 if (mtag != NULL)
1398                         skipto = *(u_int16_t *)m_tag_data(mtag);
1399                 else
1400                         skipto = 0;
1401
1402                 f = layer3_chain;
1403                 if (args->eh == NULL && skipto != 0) {
1404                         if (skipto >= IPFW_DEFAULT_RULE)
1405                                 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1406                         while (f && f->rulenum <= skipto)
1407                                 f = f->next;
1408                         if (f == NULL)  /* drop packet */
1409                                 return(IP_FW_PORT_DENY_FLAG);
1410                 }
1411         }
1412         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1413                 m_tag_delete(m, mtag);
1414
1415         /*
1416          * Now scan the rules, and parse microinstructions for each rule.
1417          */
1418         for (; f; f = f->next) {
1419                 int l, cmdlen;
1420                 ipfw_insn *cmd;
1421                 int skip_or; /* skip rest of OR block */
1422
1423 again:
1424                 if (set_disable & (1 << f->set) )
1425                         continue;
1426
1427                 skip_or = 0;
1428                 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1429                     l -= cmdlen, cmd += cmdlen) {
1430                         int match;
1431
1432                         /*
1433                          * check_body is a jump target used when we find a
1434                          * CHECK_STATE, and need to jump to the body of
1435                          * the target rule.
1436                          */
1437
1438 check_body:
1439                         cmdlen = F_LEN(cmd);
1440                         /*
1441                          * An OR block (insn_1 || .. || insn_n) has the
1442                          * F_OR bit set in all but the last instruction.
1443                          * The first match will set "skip_or", and cause
1444                          * the following instructions to be skipped until
1445                          * past the one with the F_OR bit clear.
1446                          */
1447                         if (skip_or) {          /* skip this instruction */
1448                                 if ((cmd->len & F_OR) == 0)
1449                                         skip_or = 0;    /* next one is good */
1450                                 continue;
1451                         }
1452                         match = 0; /* set to 1 if we succeed */
1453
1454                         switch (cmd->opcode) {
1455                         /*
1456                          * The first set of opcodes compares the packet's
1457                          * fields with some pattern, setting 'match' if a
1458                          * match is found. At the end of the loop there is
1459                          * logic to deal with F_NOT and F_OR flags associated
1460                          * with the opcode.
1461                          */
1462                         case O_NOP:
1463                                 match = 1;
1464                                 break;
1465
1466                         case O_FORWARD_MAC:
1467                                 kprintf("ipfw: opcode %d unimplemented\n",
1468                                     cmd->opcode);
1469                                 break;
1470
1471                         case O_GID:
1472                         case O_UID:
1473                                 /*
1474                                  * We only check offset == 0 && proto != 0,
1475                                  * as this ensures that we have an IPv4
1476                                  * packet with the ports info.
1477                                  */
1478                                 if (offset!=0)
1479                                         break;
1480                             {
1481                                 struct inpcbinfo *pi;
1482                                 int wildcard;
1483                                 struct inpcb *pcb;
1484
1485                                 if (proto == IPPROTO_TCP) {
1486                                         wildcard = 0;
1487                                         pi = &tcbinfo[mycpu->gd_cpuid];
1488                                 } else if (proto == IPPROTO_UDP) {
1489                                         wildcard = 1;
1490                                         pi = &udbinfo;
1491                                 } else
1492                                         break;
1493
1494                                 pcb =  (oif) ?
1495                                         in_pcblookup_hash(pi,
1496                                             dst_ip, htons(dst_port),
1497                                             src_ip, htons(src_port),
1498                                             wildcard, oif) :
1499                                         in_pcblookup_hash(pi,
1500                                             src_ip, htons(src_port),
1501                                             dst_ip, htons(dst_port),
1502                                             wildcard, NULL);
1503
1504                                 if (pcb == NULL || pcb->inp_socket == NULL)
1505                                         break;
1506 #if defined(__DragonFly__) || (defined(__FreeBSD__) && __FreeBSD_version < 500034)
1507 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1508 #endif
1509                                 if (cmd->opcode == O_UID) {
1510                                         match =
1511                                           !socheckuid(pcb->inp_socket,
1512                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1513                                 } else  {
1514                                         match = groupmember(
1515                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1516                                             pcb->inp_socket->so_cred);
1517                                 }
1518                             }
1519                                 break;
1520
1521                         case O_RECV:
1522                                 match = iface_match(m->m_pkthdr.rcvif,
1523                                     (ipfw_insn_if *)cmd);
1524                                 break;
1525
1526                         case O_XMIT:
1527                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1528                                 break;
1529
1530                         case O_VIA:
1531                                 match = iface_match(oif ? oif :
1532                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1533                                 break;
1534
1535                         case O_MACADDR2:
1536                                 if (args->eh != NULL) { /* have MAC header */
1537                                         u_int32_t *want = (u_int32_t *)
1538                                                 ((ipfw_insn_mac *)cmd)->addr;
1539                                         u_int32_t *mask = (u_int32_t *)
1540                                                 ((ipfw_insn_mac *)cmd)->mask;
1541                                         u_int32_t *hdr = (u_int32_t *)args->eh;
1542
1543                                         match =
1544                                             ( want[0] == (hdr[0] & mask[0]) &&
1545                                               want[1] == (hdr[1] & mask[1]) &&
1546                                               want[2] == (hdr[2] & mask[2]) );
1547                                 }
1548                                 break;
1549
1550                         case O_MAC_TYPE:
1551                                 if (args->eh != NULL) {
1552                                         u_int16_t t =
1553                                             ntohs(args->eh->ether_type);
1554                                         u_int16_t *p =
1555                                             ((ipfw_insn_u16 *)cmd)->ports;
1556                                         int i;
1557
1558                                         for (i = cmdlen - 1; !match && i>0;
1559                                             i--, p += 2)
1560                                                 match = (t>=p[0] && t<=p[1]);
1561                                 }
1562                                 break;
1563
1564                         case O_FRAG:
1565                                 match = (hlen > 0 && offset != 0);
1566                                 break;
1567
1568                         case O_IN:      /* "out" is "not in" */
1569                                 match = (oif == NULL);
1570                                 break;
1571
1572                         case O_LAYER2:
1573                                 match = (args->eh != NULL);
1574                                 break;
1575
1576                         case O_PROTO:
1577                                 /*
1578                                  * We do not allow an arg of 0 so the
1579                                  * check of "proto" only suffices.
1580                                  */
1581                                 match = (proto == cmd->arg1);
1582                                 break;
1583
1584                         case O_IP_SRC:
1585                                 match = (hlen > 0 &&
1586                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1587                                     src_ip.s_addr);
1588                                 break;
1589
1590                         case O_IP_SRC_MASK:
1591                                 match = (hlen > 0 &&
1592                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1593                                      (src_ip.s_addr &
1594                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1595                                 break;
1596
1597                         case O_IP_SRC_ME:
1598                                 if (hlen > 0) {
1599                                         struct ifnet *tif;
1600
1601                                         INADDR_TO_IFP(src_ip, tif);
1602                                         match = (tif != NULL);
1603                                 }
1604                                 break;
1605
1606                         case O_IP_DST_SET:
1607                         case O_IP_SRC_SET:
1608                                 if (hlen > 0) {
1609                                         u_int32_t *d = (u_int32_t *)(cmd+1);
1610                                         u_int32_t addr =
1611                                             cmd->opcode == O_IP_DST_SET ?
1612                                                 args->f_id.dst_ip :
1613                                                 args->f_id.src_ip;
1614
1615                                             if (addr < d[0])
1616                                                     break;
1617                                             addr -= d[0]; /* subtract base */
1618                                             match = (addr < cmd->arg1) &&
1619                                                 ( d[ 1 + (addr>>5)] &
1620                                                   (1<<(addr & 0x1f)) );
1621                                 }
1622                                 break;
1623
1624                         case O_IP_DST:
1625                                 match = (hlen > 0 &&
1626                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1627                                     dst_ip.s_addr);
1628                                 break;
1629
1630                         case O_IP_DST_MASK:
1631                                 match = (hlen > 0) &&
1632                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1633                                      (dst_ip.s_addr &
1634                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1635                                 break;
1636
1637                         case O_IP_DST_ME:
1638                                 if (hlen > 0) {
1639                                         struct ifnet *tif;
1640
1641                                         INADDR_TO_IFP(dst_ip, tif);
1642                                         match = (tif != NULL);
1643                                 }
1644                                 break;
1645
1646                         case O_IP_SRCPORT:
1647                         case O_IP_DSTPORT:
1648                                 /*
1649                                  * offset == 0 && proto != 0 is enough
1650                                  * to guarantee that we have an IPv4
1651                                  * packet with port info.
1652                                  */
1653                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1654                                     && offset == 0) {
1655                                         u_int16_t x =
1656                                             (cmd->opcode == O_IP_SRCPORT) ?
1657                                                 src_port : dst_port ;
1658                                         u_int16_t *p =
1659                                             ((ipfw_insn_u16 *)cmd)->ports;
1660                                         int i;
1661
1662                                         for (i = cmdlen - 1; !match && i>0;
1663                                             i--, p += 2)
1664                                                 match = (x>=p[0] && x<=p[1]);
1665                                 }
1666                                 break;
1667
1668                         case O_ICMPTYPE:
1669                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
1670                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
1671                                 break;
1672
1673                         case O_IPOPT:
1674                                 match = (hlen > 0 && ipopts_match(ip, cmd) );
1675                                 break;
1676
1677                         case O_IPVER:
1678                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1679                                 break;
1680
1681                         case O_IPTTL:
1682                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1683                                 break;
1684
1685                         case O_IPID:
1686                                 match = (hlen > 0 &&
1687                                     cmd->arg1 == ntohs(ip->ip_id));
1688                                 break;
1689
1690                         case O_IPLEN:
1691                                 match = (hlen > 0 && cmd->arg1 == ip_len);
1692                                 break;
1693
1694                         case O_IPPRECEDENCE:
1695                                 match = (hlen > 0 &&
1696                                     (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1697                                 break;
1698
1699                         case O_IPTOS:
1700                                 match = (hlen > 0 &&
1701                                     flags_match(cmd, ip->ip_tos));
1702                                 break;
1703
1704                         case O_TCPFLAGS:
1705                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1706                                     flags_match(cmd,
1707                                         L3HDR(struct tcphdr,ip)->th_flags));
1708                                 break;
1709
1710                         case O_TCPOPTS:
1711                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1712                                     tcpopts_match(ip, cmd));
1713                                 break;
1714
1715                         case O_TCPSEQ:
1716                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1717                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1718                                         L3HDR(struct tcphdr,ip)->th_seq);
1719                                 break;
1720
1721                         case O_TCPACK:
1722                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1723                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1724                                         L3HDR(struct tcphdr,ip)->th_ack);
1725                                 break;
1726
1727                         case O_TCPWIN:
1728                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1729                                     cmd->arg1 ==
1730                                         L3HDR(struct tcphdr,ip)->th_win);
1731                                 break;
1732
1733                         case O_ESTAB:
1734                                 /* reject packets which have SYN only */
1735                                 /* XXX should i also check for TH_ACK ? */
1736                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1737                                     (L3HDR(struct tcphdr,ip)->th_flags &
1738                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1739                                 break;
1740
1741                         case O_LOG:
1742                                 if (fw_verbose)
1743                                         ipfw_log(f, hlen, args->eh, m, oif);
1744                                 match = 1;
1745                                 break;
1746
1747                         case O_PROB:
1748                                 match = (krandom() <
1749                                         ((ipfw_insn_u32 *)cmd)->d[0]);
1750                                 break;
1751
1752                         /*
1753                          * The second set of opcodes represents 'actions',
1754                          * i.e. the terminal part of a rule once the packet
1755                          * matches all previous patterns.
1756                          * Typically there is only one action for each rule,
1757                          * and the opcode is stored at the end of the rule
1758                          * (but there are exceptions -- see below).
1759                          *
1760                          * In general, here we set retval and terminate the
1761                          * outer loop (would be a 'break 3' in some language,
1762                          * but we need to do a 'goto done').
1763                          *
1764                          * Exceptions:
1765                          * O_COUNT and O_SKIPTO actions:
1766                          *   instead of terminating, we jump to the next rule
1767                          *   ('goto next_rule', equivalent to a 'break 2'),
1768                          *   or to the SKIPTO target ('goto again' after
1769                          *   having set f, cmd and l), respectively.
1770                          *
1771                          * O_LIMIT and O_KEEP_STATE: these opcodes are
1772                          *   not real 'actions', and are stored right
1773                          *   before the 'action' part of the rule.
1774                          *   These opcodes try to install an entry in the
1775                          *   state tables; if successful, we continue with
1776                          *   the next opcode (match=1; break;), otherwise
1777                          *   the packet *   must be dropped
1778                          *   ('goto done' after setting retval);
1779                          *
1780                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1781                          *   cause a lookup of the state table, and a jump
1782                          *   to the 'action' part of the parent rule
1783                          *   ('goto check_body') if an entry is found, or
1784                          *   (CHECK_STATE only) a jump to the next rule if
1785                          *   the entry is not found ('goto next_rule').
1786                          *   The result of the lookup is cached to make
1787                          *   further instances of these opcodes are
1788                          *   effectively NOPs.
1789                          */
1790                         case O_LIMIT:
1791                         case O_KEEP_STATE:
1792                                 if (install_state(f,
1793                                     (ipfw_insn_limit *)cmd, args)) {
1794                                         retval = IP_FW_PORT_DENY_FLAG;
1795                                         goto done; /* error/limit violation */
1796                                 }
1797                                 match = 1;
1798                                 break;
1799
1800                         case O_PROBE_STATE:
1801                         case O_CHECK_STATE:
1802                                 /*
1803                                  * dynamic rules are checked at the first
1804                                  * keep-state or check-state occurrence,
1805                                  * with the result being stored in dyn_dir.
1806                                  * The compiler introduces a PROBE_STATE
1807                                  * instruction for us when we have a
1808                                  * KEEP_STATE (because PROBE_STATE needs
1809                                  * to be run first).
1810                                  */
1811                                 if (dyn_dir == MATCH_UNKNOWN &&
1812                                     (q = lookup_dyn_rule(&args->f_id,
1813                                      &dyn_dir, proto == IPPROTO_TCP ?
1814                                         L3HDR(struct tcphdr, ip) : NULL))
1815                                         != NULL) {
1816                                         /*
1817                                          * Found dynamic entry, update stats
1818                                          * and jump to the 'action' part of
1819                                          * the parent rule.
1820                                          */
1821                                         q->pcnt++;
1822                                         q->bcnt += ip_len;
1823                                         f = q->rule;
1824                                         cmd = ACTION_PTR(f);
1825                                         l = f->cmd_len - f->act_ofs;
1826                                         goto check_body;
1827                                 }
1828                                 /*
1829                                  * Dynamic entry not found. If CHECK_STATE,
1830                                  * skip to next rule, if PROBE_STATE just
1831                                  * ignore and continue with next opcode.
1832                                  */
1833                                 if (cmd->opcode == O_CHECK_STATE)
1834                                         goto next_rule;
1835                                 match = 1;
1836                                 break;
1837
1838                         case O_ACCEPT:
1839                                 retval = 0;     /* accept */
1840                                 goto done;
1841
1842                         case O_PIPE:
1843                         case O_QUEUE:
1844                                 args->rule = f; /* report matching rule */
1845                                 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
1846                                 goto done;
1847
1848                         case O_DIVERT:
1849                         case O_TEE:
1850                                 if (args->eh) /* not on layer 2 */
1851                                         break;
1852
1853                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
1854                                                 sizeof(u_int16_t), M_NOWAIT);
1855                                 if (mtag == NULL) {
1856                                         retval = IP_FW_PORT_DENY_FLAG;
1857                                         goto done;
1858                                 }
1859                                 *(u_int16_t *)m_tag_data(mtag) = f->rulenum;
1860                                 m_tag_prepend(m, mtag);
1861                                 retval = (cmd->opcode == O_DIVERT) ?
1862                                     cmd->arg1 :
1863                                     cmd->arg1 | IP_FW_PORT_TEE_FLAG;
1864                                 goto done;
1865
1866                         case O_COUNT:
1867                         case O_SKIPTO:
1868                                 f->pcnt++;      /* update stats */
1869                                 f->bcnt += ip_len;
1870                                 f->timestamp = time_second;
1871                                 if (cmd->opcode == O_COUNT)
1872                                         goto next_rule;
1873                                 /* handle skipto */
1874                                 if (f->next_rule == NULL)
1875                                         lookup_next_rule(f);
1876                                 f = f->next_rule;
1877                                 goto again;
1878
1879                         case O_REJECT:
1880                                 /*
1881                                  * Drop the packet and send a reject notice
1882                                  * if the packet is not ICMP (or is an ICMP
1883                                  * query), and it is not multicast/broadcast.
1884                                  */
1885                                 if (hlen > 0 &&
1886                                     (proto != IPPROTO_ICMP ||
1887                                      is_icmp_query(ip)) &&
1888                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
1889                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
1890                                         send_reject(args, cmd->arg1,
1891                                             offset,ip_len);
1892                                         m = args->m;
1893                                 }
1894                                 /* FALLTHROUGH */
1895                         case O_DENY:
1896                                 retval = IP_FW_PORT_DENY_FLAG;
1897                                 goto done;
1898
1899                         case O_FORWARD_IP:
1900                                 if (args->eh)   /* not valid on layer2 pkts */
1901                                         break;
1902                                 if (!q || dyn_dir == MATCH_FORWARD)
1903                                         args->next_hop =
1904                                             &((ipfw_insn_sa *)cmd)->sa;
1905                                 retval = 0;
1906                                 goto done;
1907
1908                         default:
1909                                 panic("-- unknown opcode %d\n", cmd->opcode);
1910                         } /* end of switch() on opcodes */
1911
1912                         if (cmd->len & F_NOT)
1913                                 match = !match;
1914
1915                         if (match) {
1916                                 if (cmd->len & F_OR)
1917                                         skip_or = 1;
1918                         } else {
1919                                 if (!(cmd->len & F_OR)) /* not an OR block, */
1920                                         break;          /* try next rule    */
1921                         }
1922
1923                 }       /* end of inner for, scan opcodes */
1924
1925 next_rule:;             /* try next rule                */
1926
1927         }               /* end of outer for, scan rules */
1928         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
1929         return(IP_FW_PORT_DENY_FLAG);
1930
1931 done:
1932         /* Update statistics */
1933         f->pcnt++;
1934         f->bcnt += ip_len;
1935         f->timestamp = time_second;
1936         return retval;
1937
1938 pullup_failed:
1939         if (fw_verbose)
1940                 kprintf("pullup failed\n");
1941         return(IP_FW_PORT_DENY_FLAG);
1942 }
1943
1944 /*
1945  * When a rule is added/deleted, clear the next_rule pointers in all rules.
1946  * These will be reconstructed on the fly as packets are matched.
1947  * Must be called at splimp().
1948  */
1949 static void
1950 flush_rule_ptrs(void)
1951 {
1952         struct ip_fw *rule;
1953
1954         for (rule = layer3_chain; rule; rule = rule->next)
1955                 rule->next_rule = NULL;
1956 }
1957
1958 /*
1959  * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
1960  * pipe/queue, or to all of them (match == NULL).
1961  * Must be called at splimp().
1962  */
1963 void
1964 flush_pipe_ptrs(struct dn_flow_set *match)
1965 {
1966         struct ip_fw *rule;
1967
1968         for (rule = layer3_chain; rule; rule = rule->next) {
1969                 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
1970
1971                 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
1972                         continue;
1973                 if (match == NULL || cmd->pipe_ptr == match)
1974                         cmd->pipe_ptr = NULL;
1975         }
1976 }
1977
1978 /*
1979  * Add a new rule to the list. Copy the rule into a malloc'ed area, then
1980  * possibly create a rule number and add the rule to the list.
1981  * Update the rule_number in the input struct so the caller knows it as well.
1982  */
1983 static int
1984 add_rule(struct ip_fw **head, struct ip_fw *input_rule)
1985 {
1986         struct ip_fw *rule, *f, *prev;
1987         int l = RULESIZE(input_rule);
1988
1989         if (*head == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
1990                 return (EINVAL);
1991
1992         rule = kmalloc(l, M_IPFW, M_WAITOK | M_ZERO);
1993         if (rule == NULL)
1994                 return (ENOSPC);
1995
1996         bcopy(input_rule, rule, l);
1997
1998         rule->next = NULL;
1999         rule->next_rule = NULL;
2000
2001         rule->pcnt = 0;
2002         rule->bcnt = 0;
2003         rule->timestamp = 0;
2004
2005         crit_enter();
2006
2007         if (*head == NULL) {    /* default rule */
2008                 *head = rule;
2009                 goto done;
2010         }
2011
2012         /*
2013          * If rulenum is 0, find highest numbered rule before the
2014          * default rule, and add autoinc_step
2015          */
2016         if (autoinc_step < 1)
2017                 autoinc_step = 1;
2018         else if (autoinc_step > 1000)
2019                 autoinc_step = 1000;
2020         if (rule->rulenum == 0) {
2021                 /*
2022                  * locate the highest numbered rule before default
2023                  */
2024                 for (f = *head; f; f = f->next) {
2025                         if (f->rulenum == IPFW_DEFAULT_RULE)
2026                                 break;
2027                         rule->rulenum = f->rulenum;
2028                 }
2029                 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2030                         rule->rulenum += autoinc_step;
2031                 input_rule->rulenum = rule->rulenum;
2032         }
2033
2034         /*
2035          * Now insert the new rule in the right place in the sorted list.
2036          */
2037         for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2038                 if (f->rulenum > rule->rulenum) { /* found the location */
2039                         if (prev) {
2040                                 rule->next = f;
2041                                 prev->next = rule;
2042                         } else { /* head insert */
2043                                 rule->next = *head;
2044                                 *head = rule;
2045                         }
2046                         break;
2047                 }
2048         }
2049         flush_rule_ptrs();
2050 done:
2051         static_count++;
2052         static_len += l;
2053         crit_exit();
2054         DEB(kprintf("++ installed rule %d, static count now %d\n",
2055                 rule->rulenum, static_count);)
2056         return (0);
2057 }
2058
2059 /**
2060  * Free storage associated with a static rule (including derived
2061  * dynamic rules).
2062  * The caller is in charge of clearing rule pointers to avoid
2063  * dangling pointers.
2064  * @return a pointer to the next entry.
2065  * Arguments are not checked, so they better be correct.
2066  * Must be called at splimp().
2067  */
2068 static struct ip_fw *
2069 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2070 {
2071         struct ip_fw *n;
2072         int l = RULESIZE(rule);
2073
2074         n = rule->next;
2075         remove_dyn_rule(rule, NULL /* force removal */);
2076         if (prev == NULL)
2077                 *head = n;
2078         else
2079                 prev->next = n;
2080         static_count--;
2081         static_len -= l;
2082
2083         if (DUMMYNET_LOADED)
2084                 ip_dn_ruledel_ptr(rule);
2085         kfree(rule, M_IPFW);
2086         return n;
2087 }
2088
2089 /*
2090  * Deletes all rules from a chain (including the default rule
2091  * if the second argument is set).
2092  * Must be called at splimp().
2093  */
2094 static void
2095 free_chain(struct ip_fw **chain, int kill_default)
2096 {
2097         struct ip_fw *rule;
2098
2099         flush_rule_ptrs(); /* more efficient to do outside the loop */
2100
2101         while ( (rule = *chain) != NULL &&
2102             (kill_default || rule->rulenum != IPFW_DEFAULT_RULE) )
2103                 delete_rule(chain, NULL, rule);
2104 }
2105
2106 /**
2107  * Remove all rules with given number, and also do set manipulation.
2108  *
2109  * The argument is an u_int32_t. The low 16 bit are the rule or set number,
2110  * the next 8 bits are the new set, the top 8 bits are the command:
2111  *
2112  *      0       delete rules with given number
2113  *      1       delete rules with given set number
2114  *      2       move rules with given number to new set
2115  *      3       move rules with given set number to new set
2116  *      4       swap sets with given numbers
2117  */
2118 static int
2119 del_entry(struct ip_fw **chain, u_int32_t arg)
2120 {
2121         struct ip_fw *prev, *rule;
2122         u_int16_t rulenum;
2123         u_int8_t cmd, new_set;
2124
2125         rulenum = arg & 0xffff;
2126         cmd = (arg >> 24) & 0xff;
2127         new_set = (arg >> 16) & 0xff;
2128
2129         if (cmd > 4)
2130                 return EINVAL;
2131         if (new_set > 30)
2132                 return EINVAL;
2133         if (cmd == 0 || cmd == 2) {
2134                 if (rulenum == IPFW_DEFAULT_RULE)
2135                         return EINVAL;
2136         } else {
2137                 if (rulenum > 30)
2138                         return EINVAL;
2139         }
2140
2141         switch (cmd) {
2142         case 0: /* delete rules with given number */
2143                 /*
2144                  * locate first rule to delete
2145                  */
2146                 for (prev = NULL, rule = *chain;
2147                     rule && rule->rulenum < rulenum;
2148                      prev = rule, rule = rule->next)
2149                         ;
2150                 if (rule->rulenum != rulenum)
2151                         return EINVAL;
2152
2153                 crit_enter(); /* no access to rules while removing */
2154                 /*
2155                  * flush pointers outside the loop, then delete all matching
2156                  * rules. prev remains the same throughout the cycle.
2157                  */
2158                 flush_rule_ptrs();
2159                 while (rule && rule->rulenum == rulenum)
2160                         rule = delete_rule(chain, prev, rule);
2161                 crit_exit();
2162                 break;
2163
2164         case 1: /* delete all rules with given set number */
2165                 crit_enter();
2166                 flush_rule_ptrs();
2167                 for (prev = NULL, rule = *chain; rule ; )
2168                         if (rule->set == rulenum)
2169                                 rule = delete_rule(chain, prev, rule);
2170                         else {
2171                                 prev = rule;
2172                                 rule = rule->next;
2173                         }
2174                 crit_exit();
2175                 break;
2176
2177         case 2: /* move rules with given number to new set */
2178                 crit_enter();
2179                 for (rule = *chain; rule ; rule = rule->next)
2180                         if (rule->rulenum == rulenum)
2181                                 rule->set = new_set;
2182                 crit_exit();
2183                 break;
2184
2185         case 3: /* move rules with given set number to new set */
2186                 crit_enter();
2187                 for (rule = *chain; rule ; rule = rule->next)
2188                         if (rule->set == rulenum)
2189                                 rule->set = new_set;
2190                 crit_exit();
2191                 break;
2192
2193         case 4: /* swap two sets */
2194                 crit_enter();
2195                 for (rule = *chain; rule ; rule = rule->next)
2196                         if (rule->set == rulenum)
2197                                 rule->set = new_set;
2198                         else if (rule->set == new_set)
2199                                 rule->set = rulenum;
2200                 crit_exit();
2201                 break;
2202         }
2203         return 0;
2204 }
2205
2206 /*
2207  * Clear counters for a specific rule.
2208  */
2209 static void
2210 clear_counters(struct ip_fw *rule, int log_only)
2211 {
2212         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2213
2214         if (log_only == 0) {
2215                 rule->bcnt = rule->pcnt = 0;
2216                 rule->timestamp = 0;
2217         }
2218         if (l->o.opcode == O_LOG)
2219                 l->log_left = l->max_log;
2220 }
2221
2222 /**
2223  * Reset some or all counters on firewall rules.
2224  * @arg frwl is null to clear all entries, or contains a specific
2225  * rule number.
2226  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2227  */
2228 static int
2229 zero_entry(int rulenum, int log_only)
2230 {
2231         struct ip_fw *rule;
2232         char *msg;
2233
2234         if (rulenum == 0) {
2235                 crit_enter();
2236                 norule_counter = 0;
2237                 for (rule = layer3_chain; rule; rule = rule->next)
2238                         clear_counters(rule, log_only);
2239                 crit_exit();
2240                 msg = log_only ? "ipfw: All logging counts reset.\n" :
2241                                 "ipfw: Accounting cleared.\n";
2242         } else {
2243                 int cleared = 0;
2244                 /*
2245                  * We can have multiple rules with the same number, so we
2246                  * need to clear them all.
2247                  */
2248                 for (rule = layer3_chain; rule; rule = rule->next)
2249                         if (rule->rulenum == rulenum) {
2250                                 crit_enter();
2251                                 while (rule && rule->rulenum == rulenum) {
2252                                         clear_counters(rule, log_only);
2253                                         rule = rule->next;
2254                                 }
2255                                 crit_exit();
2256                                 cleared = 1;
2257                                 break;
2258                         }
2259                 if (!cleared)   /* we did not find any matching rules */
2260                         return (EINVAL);
2261                 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2262                                 "ipfw: Entry %d cleared.\n";
2263         }
2264         if (fw_verbose)
2265                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2266         return (0);
2267 }
2268
2269 /*
2270  * Check validity of the structure before insert.
2271  * Fortunately rules are simple, so this mostly need to check rule sizes.
2272  */
2273 static int
2274 check_ipfw_struct(struct ip_fw *rule, int size)
2275 {
2276         int l, cmdlen = 0;
2277         int have_action=0;
2278         ipfw_insn *cmd;
2279
2280         if (size < sizeof(*rule)) {
2281                 kprintf("ipfw: rule too short\n");
2282                 return (EINVAL);
2283         }
2284         /* first, check for valid size */
2285         l = RULESIZE(rule);
2286         if (l != size) {
2287                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
2288                 return (EINVAL);
2289         }
2290         /*
2291          * Now go for the individual checks. Very simple ones, basically only
2292          * instruction sizes.
2293          */
2294         for (l = rule->cmd_len, cmd = rule->cmd ;
2295                         l > 0 ; l -= cmdlen, cmd += cmdlen) {
2296                 cmdlen = F_LEN(cmd);
2297                 if (cmdlen > l) {
2298                         kprintf("ipfw: opcode %d size truncated\n",
2299                             cmd->opcode);
2300                         return EINVAL;
2301                 }
2302                 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
2303                 switch (cmd->opcode) {
2304                 case O_NOP:
2305                 case O_PROBE_STATE:
2306                 case O_KEEP_STATE:
2307                 case O_PROTO:
2308                 case O_IP_SRC_ME:
2309                 case O_IP_DST_ME:
2310                 case O_LAYER2:
2311                 case O_IN:
2312                 case O_FRAG:
2313                 case O_IPOPT:
2314                 case O_IPLEN:
2315                 case O_IPID:
2316                 case O_IPTOS:
2317                 case O_IPPRECEDENCE:
2318                 case O_IPTTL:
2319                 case O_IPVER:
2320                 case O_TCPWIN:
2321                 case O_TCPFLAGS:
2322                 case O_TCPOPTS:
2323                 case O_ESTAB:
2324                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2325                                 goto bad_size;
2326                         break;
2327
2328                 case O_UID:
2329                 case O_GID:
2330                 case O_IP_SRC:
2331                 case O_IP_DST:
2332                 case O_TCPSEQ:
2333                 case O_TCPACK:
2334                 case O_PROB:
2335                 case O_ICMPTYPE:
2336                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2337                                 goto bad_size;
2338                         break;
2339
2340                 case O_LIMIT:
2341                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2342                                 goto bad_size;
2343                         break;
2344
2345                 case O_LOG:
2346                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2347                                 goto bad_size;
2348
2349                         ((ipfw_insn_log *)cmd)->log_left =
2350                             ((ipfw_insn_log *)cmd)->max_log;
2351
2352                         break;
2353
2354                 case O_IP_SRC_MASK:
2355                 case O_IP_DST_MASK:
2356                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2357                                 goto bad_size;
2358                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2359                                 kprintf("ipfw: opcode %d, useless rule\n",
2360                                         cmd->opcode);
2361                                 return EINVAL;
2362                         }
2363                         break;
2364
2365                 case O_IP_SRC_SET:
2366                 case O_IP_DST_SET:
2367                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2368                                 kprintf("ipfw: invalid set size %d\n",
2369                                         cmd->arg1);
2370                                 return EINVAL;
2371                         }
2372                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2373                             (cmd->arg1+31)/32 )
2374                                 goto bad_size;
2375                         break;
2376
2377                 case O_MACADDR2:
2378                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2379                                 goto bad_size;
2380                         break;
2381
2382                 case O_MAC_TYPE:
2383                 case O_IP_SRCPORT:
2384                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2385                         if (cmdlen < 2 || cmdlen > 31)
2386                                 goto bad_size;
2387                         break;
2388
2389                 case O_RECV:
2390                 case O_XMIT:
2391                 case O_VIA:
2392                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2393                                 goto bad_size;
2394                         break;
2395
2396                 case O_PIPE:
2397                 case O_QUEUE:
2398                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2399                                 goto bad_size;
2400                         goto check_action;
2401
2402                 case O_FORWARD_IP:
2403                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2404                                 goto bad_size;
2405                         goto check_action;
2406
2407                 case O_FORWARD_MAC: /* XXX not implemented yet */
2408                 case O_CHECK_STATE:
2409                 case O_COUNT:
2410                 case O_ACCEPT:
2411                 case O_DENY:
2412                 case O_REJECT:
2413                 case O_SKIPTO:
2414                 case O_DIVERT:
2415                 case O_TEE:
2416                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2417                                 goto bad_size;
2418 check_action:
2419                         if (have_action) {
2420                                 kprintf("ipfw: opcode %d, multiple actions"
2421                                         " not allowed\n",
2422                                         cmd->opcode);
2423                                 return EINVAL;
2424                         }
2425                         have_action = 1;
2426                         if (l != cmdlen) {
2427                                 kprintf("ipfw: opcode %d, action must be"
2428                                         " last opcode\n",
2429                                         cmd->opcode);
2430                                 return EINVAL;
2431                         }
2432                         break;
2433                 default:
2434                         kprintf("ipfw: opcode %d, unknown opcode\n",
2435                                 cmd->opcode);
2436                         return EINVAL;
2437                 }
2438         }
2439         if (have_action == 0) {
2440                 kprintf("ipfw: missing action\n");
2441                 return EINVAL;
2442         }
2443         return 0;
2444
2445 bad_size:
2446         kprintf("ipfw: opcode %d size %d wrong\n",
2447                 cmd->opcode, cmdlen);
2448         return EINVAL;
2449 }
2450
2451
2452 /**
2453  * {set|get}sockopt parser.
2454  */
2455 static int
2456 ipfw_ctl(struct sockopt *sopt)
2457 {
2458         int error, rulenum;
2459         size_t size;
2460         struct ip_fw *bp , *buf, *rule;
2461
2462         static u_int32_t rule_buf[255]; /* we copy the data here */
2463
2464         /*
2465          * Disallow modifications in really-really secure mode, but still allow
2466          * the logging counters to be reset.
2467          */
2468         if (sopt->sopt_name == IP_FW_ADD ||
2469             (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
2470 #if defined(__FreeBSD__) && __FreeBSD_version >= 500034
2471                 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2472                 if (error)
2473                         return (error);
2474 #else /* FreeBSD 4.x */
2475                 if (securelevel >= 3)
2476                         return (EPERM);
2477 #endif
2478         }
2479
2480         error = 0;
2481
2482         switch (sopt->sopt_name) {
2483         case IP_FW_GET:
2484                 /*
2485                  * pass up a copy of the current rules. Static rules
2486                  * come first (the last of which has number IPFW_DEFAULT_RULE),
2487                  * followed by a possibly empty list of dynamic rule.
2488                  * The last dynamic rule has NULL in the "next" field.
2489                  */
2490                 crit_enter();
2491                 size = static_len;      /* size of static rules */
2492                 if (ipfw_dyn_v)         /* add size of dyn.rules */
2493                         size += (dyn_count * sizeof(ipfw_dyn_rule));
2494
2495                 /*
2496                  * XXX todo: if the user passes a short length just to know
2497                  * how much room is needed, do not bother filling up the
2498                  * buffer, just jump to the sooptcopyout.
2499                  */
2500                 buf = kmalloc(size, M_TEMP, M_WAITOK);
2501
2502                 bp = buf;
2503                 for (rule = layer3_chain; rule ; rule = rule->next) {
2504                         int i = RULESIZE(rule);
2505                         bcopy(rule, bp, i);
2506                         /*
2507                          * abuse 'next_rule' to store the set_disable word
2508                          */
2509                         bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
2510                             sizeof(set_disable));
2511                         bp = (struct ip_fw *)((char *)bp + i);
2512                 }
2513                 if (ipfw_dyn_v) {
2514                         int i;
2515                         ipfw_dyn_rule *p, *dst, *last = NULL;
2516
2517                         dst = (ipfw_dyn_rule *)bp;
2518                         for (i = 0 ; i < curr_dyn_buckets ; i++ )
2519                                 for ( p = ipfw_dyn_v[i] ; p != NULL ;
2520                                     p = p->next, dst++ ) {
2521                                         bcopy(p, dst, sizeof *p);
2522                                         bcopy(&(p->rule->rulenum), &(dst->rule),
2523                                             sizeof(p->rule->rulenum));
2524                                         /*
2525                                          * store a non-null value in "next".
2526                                          * The userland code will interpret a
2527                                          * NULL here as a marker
2528                                          * for the last dynamic rule.
2529                                          */
2530                                         dst->next = dst ;
2531                                         last = dst ;
2532                                         dst->expire =
2533                                             TIME_LEQ(dst->expire, time_second) ?
2534                                                 0 : dst->expire - time_second ;
2535                                 }
2536                         if (last != NULL) /* mark last dynamic rule */
2537                                 last->next = NULL;
2538                 }
2539                 crit_exit();
2540
2541                 error = sooptcopyout(sopt, buf, size);
2542                 kfree(buf, M_TEMP);
2543                 break;
2544
2545         case IP_FW_FLUSH:
2546                 /*
2547                  * Normally we cannot release the lock on each iteration.
2548                  * We could do it here only because we start from the head all
2549                  * the times so there is no risk of missing some entries.
2550                  * On the other hand, the risk is that we end up with
2551                  * a very inconsistent ruleset, so better keep the lock
2552                  * around the whole cycle.
2553                  *
2554                  * XXX this code can be improved by resetting the head of
2555                  * the list to point to the default rule, and then freeing
2556                  * the old list without the need for a lock.
2557                  */
2558
2559                 crit_enter();
2560                 free_chain(&layer3_chain, 0 /* keep default rule */);
2561                 crit_exit();
2562                 break;
2563
2564         case IP_FW_ADD:
2565                 rule = (struct ip_fw *)rule_buf; /* XXX do a malloc */
2566                 error = sooptcopyin(sopt, rule, sizeof(rule_buf),
2567                         sizeof(struct ip_fw) );
2568                 size = sopt->sopt_valsize;
2569                 if (error || (error = check_ipfw_struct(rule, size)))
2570                         break;
2571
2572                 error = add_rule(&layer3_chain, rule);
2573                 size = RULESIZE(rule);
2574                 if (!error && sopt->sopt_dir == SOPT_GET)
2575                         error = sooptcopyout(sopt, rule, size);
2576                 break;
2577
2578         case IP_FW_DEL:
2579                 /*
2580                  * IP_FW_DEL is used for deleting single rules or sets,
2581                  * and (ab)used to atomically manipulate sets. Argument size
2582                  * is used to distinguish between the two:
2583                  *    sizeof(u_int32_t)
2584                  *      delete single rule or set of rules,
2585                  *      or reassign rules (or sets) to a different set.
2586                  *    2*sizeof(u_int32_t)
2587                  *      atomic disable/enable sets.
2588                  *      first u_int32_t contains sets to be disabled,
2589                  *      second u_int32_t contains sets to be enabled.
2590                  */
2591                 error = sooptcopyin(sopt, rule_buf,
2592                         2*sizeof(u_int32_t), sizeof(u_int32_t));
2593                 if (error)
2594                         break;
2595                 size = sopt->sopt_valsize;
2596                 if (size == sizeof(u_int32_t))  /* delete or reassign */
2597                         error = del_entry(&layer3_chain, rule_buf[0]);
2598                 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
2599                         set_disable =
2600                             (set_disable | rule_buf[0]) & ~rule_buf[1] &
2601                             ~(1<<31); /* set 31 always enabled */
2602                 else
2603                         error = EINVAL;
2604                 break;
2605
2606         case IP_FW_ZERO:
2607         case IP_FW_RESETLOG: /* argument is an int, the rule number */
2608                 rulenum=0;
2609
2610                 if (sopt->sopt_val != 0) {
2611                     error = sooptcopyin(sopt, &rulenum,
2612                             sizeof(int), sizeof(int));
2613                     if (error)
2614                         break;
2615                 }
2616                 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2617                 break;
2618
2619         default:
2620                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
2621                 error = EINVAL;
2622         }
2623
2624         return (error);
2625 }
2626
2627 /**
2628  * dummynet needs a reference to the default rule, because rules can be
2629  * deleted while packets hold a reference to them. When this happens,
2630  * dummynet changes the reference to the default rule (it could well be a
2631  * NULL pointer, but this way we do not need to check for the special
2632  * case, plus here he have info on the default behaviour).
2633  */
2634 struct ip_fw *ip_fw_default_rule;
2635
2636 /*
2637  * This procedure is only used to handle keepalives. It is invoked
2638  * every dyn_keepalive_period
2639  */
2640 static void
2641 ipfw_tick(void * __unused unused)
2642 {
2643         int i;
2644         ipfw_dyn_rule *q;
2645
2646         if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2647                 goto done;
2648
2649         crit_enter();
2650         for (i = 0 ; i < curr_dyn_buckets ; i++) {
2651                 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
2652                         if (q->dyn_type == O_LIMIT_PARENT)
2653                                 continue;
2654                         if (q->id.proto != IPPROTO_TCP)
2655                                 continue;
2656                         if ( (q->state & BOTH_SYN) != BOTH_SYN)
2657                                 continue;
2658                         if (TIME_LEQ( time_second+dyn_keepalive_interval,
2659                             q->expire))
2660                                 continue;       /* too early */
2661                         if (TIME_LEQ(q->expire, time_second))
2662                                 continue;       /* too late, rule expired */
2663
2664                         send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
2665                         send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
2666                 }
2667         }
2668         crit_exit();
2669 done:
2670         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
2671                       ipfw_tick, NULL);
2672 }
2673
2674 static void
2675 ipfw_init(void)
2676 {
2677         struct ip_fw default_rule;
2678
2679         ip_fw_chk_ptr = ipfw_chk;
2680         ip_fw_ctl_ptr = ipfw_ctl;
2681         layer3_chain = NULL;
2682
2683         bzero(&default_rule, sizeof default_rule);
2684
2685         default_rule.act_ofs = 0;
2686         default_rule.rulenum = IPFW_DEFAULT_RULE;
2687         default_rule.cmd_len = 1;
2688         default_rule.set = 31;
2689
2690         default_rule.cmd[0].len = 1;
2691         default_rule.cmd[0].opcode =
2692 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2693                                 1 ? O_ACCEPT :
2694 #endif
2695                                 O_DENY;
2696
2697         add_rule(&layer3_chain, &default_rule);
2698
2699         ip_fw_default_rule = layer3_chain;
2700         kprintf("ipfw2 initialized, divert %s, "
2701                 "rule-based forwarding enabled, default to %s, logging ",
2702 #ifdef IPDIVERT
2703                 "enabled",
2704 #else
2705                 "disabled",
2706 #endif
2707                 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
2708
2709 #ifdef IPFIREWALL_VERBOSE
2710         fw_verbose = 1;
2711 #endif
2712 #ifdef IPFIREWALL_VERBOSE_LIMIT
2713         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2714 #endif
2715         if (fw_verbose == 0)
2716                 kprintf("disabled\n");
2717         else if (verbose_limit == 0)
2718                 kprintf("unlimited\n");
2719         else
2720                 kprintf("limited to %d packets/entry by default\n",
2721                     verbose_limit);
2722         callout_init(&ipfw_timeout_h);
2723         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
2724 }
2725
2726 static int
2727 ipfw_modevent(module_t mod, int type, void *unused)
2728 {
2729         int err = 0;
2730
2731         switch (type) {
2732         case MOD_LOAD:
2733                 crit_enter();
2734                 if (IPFW_LOADED) {
2735                         crit_exit();
2736                         kprintf("IP firewall already loaded\n");
2737                         err = EEXIST;
2738                 } else {
2739                         ipfw_init();
2740                         crit_exit();
2741                 }
2742                 break;
2743
2744         case MOD_UNLOAD:
2745 #if !defined(KLD_MODULE)
2746                 kprintf("ipfw statically compiled, cannot unload\n");
2747                 err = EBUSY;
2748 #else
2749                 crit_enter();
2750                 callout_stop(&ipfw_timeout_h);
2751                 ip_fw_chk_ptr = NULL;
2752                 ip_fw_ctl_ptr = NULL;
2753                 free_chain(&layer3_chain, 1 /* kill default rule */);
2754                 crit_exit();
2755                 kprintf("IP firewall unloaded\n");
2756 #endif
2757                 break;
2758         default:
2759                 break;
2760         }
2761         return err;
2762 }
2763
2764 static moduledata_t ipfwmod = {
2765         "ipfw",
2766         ipfw_modevent,
2767         0
2768 };
2769 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY);
2770 MODULE_VERSION(ipfw, 1);