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 1.38 2007/11/16 02:45:45 sephe Exp $
  27  */
  28
  29 #define        DEB(x)
  30 #define        DDB(x) x
  31
  32 /*
  33  * Implement IP packet firewall (new version)
  34  */
  35
  36 #ifndef KLD_MODULE
  37 #include "opt_ipfw.h"
  38 #include "opt_ipdn.h"
  39 #include "opt_ipdivert.h"
  40 #include "opt_inet.h"
  41 #ifndef INET
  42 #error IPFIREWALL requires INET.
  43 #endif /* INET */
  44 #endif
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/kernel.h>
  51 #include <sys/proc.h>
  52 #include <sys/socket.h>
  53 #include <sys/socketvar.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/syslog.h>
  56 #include <sys/thread2.h>
  57 #include <sys/ucred.h>
  58 #include <sys/in_cksum.h>
  59 #include <net/if.h>
  60 #include <net/route.h>
  61 #include <netinet/in.h>
  62 #include <netinet/in_systm.h>
  63 #include <netinet/in_var.h>
  64 #include <netinet/in_pcb.h>
  65 #include <netinet/ip.h>
  66 #include <netinet/ip_var.h>
  67 #include <netinet/ip_icmp.h>
  68 #include "ip_fw.h"
  69 #include <net/dummynet/ip_dummynet.h>
  70 #include <netinet/tcp.h>
  71 #include <netinet/tcp_timer.h>
  72 #include <netinet/tcp_var.h>
  73 #include <netinet/tcpip.h>
  74 #include <netinet/udp.h>
  75 #include <netinet/udp_var.h>
  76
  77 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  78
  79 /*
  80  * set_disable contains one bit per set value (0..31).
  81  * If the bit is set, all rules with the corresponding set
  82  * are disabled. Set 31 is reserved for the default rule
  83  * and CANNOT be disabled.
  84  */
  85 static uint32_t set_disable;
  86
  87 static int fw_verbose;
  88 static int verbose_limit;
  89
  90 #ifdef KLD_MODULE
  91 static int ipfw_refcnt;
  92 #endif
  93
  94 static struct callout ipfw_timeout_h;
  95 #define IPFW_DEFAULT_RULE       65535
  96
  97 /*
  98  * list of rules for layer 3
  99  */
 100 static struct ip_fw *layer3_chain;
 101
 102 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 103
 104 static int fw_debug = 1;
 105 static int autoinc_step = 100; /* bounded to 1..1000 in ipfw_add_rule() */
 106
 107 #ifdef SYSCTL_NODE
 108 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 109 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
 110     &fw_enable, 0, "Enable ipfw");
 111 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
 112     &autoinc_step, 0, "Rule number autincrement step");
 113 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 114     &fw_one_pass, 0,
 115     "Only do a single pass through ipfw when using dummynet(4)");
 116 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 117     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 118 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 119     &fw_verbose, 0, "Log matches to ipfw rules");
 120 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 121     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 122
 123 /*
 124  * Description of dynamic rules.
 125  *
 126  * Dynamic rules are stored in lists accessed through a hash table
 127  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 128  * be modified through the sysctl variable dyn_buckets which is
 129  * updated when the table becomes empty.
 130  *
 131  * XXX currently there is only one list, ipfw_dyn.
 132  *
 133  * When a packet is received, its address fields are first masked
 134  * with the mask defined for the rule, then hashed, then matched
 135  * against the entries in the corresponding list.
 136  * Dynamic rules can be used for different purposes:
 137  *  + stateful rules;
 138  *  + enforcing limits on the number of sessions;
 139  *  + in-kernel NAT (not implemented yet)
 140  *
 141  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 142  * measured in seconds and depending on the flags.
 143  *
 144  * The total number of dynamic rules is stored in dyn_count.
 145  * The max number of dynamic rules is dyn_max. When we reach
 146  * the maximum number of rules we do not create anymore. This is
 147  * done to avoid consuming too much memory, but also too much
 148  * time when searching on each packet (ideally, we should try instead
 149  * to put a limit on the length of the list on each bucket...).
 150  *
 151  * Each dynamic rule holds a pointer to the parent ipfw rule so
 152  * we know what action to perform. Dynamic rules are removed when
 153  * the parent rule is deleted. XXX we should make them survive.
 154  *
 155  * There are some limitations with dynamic rules -- we do not
 156  * obey the 'randomized match', and we do not do multiple
 157  * passes through the firewall. XXX check the latter!!!
 158  */
 159 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 160 static uint32_t dyn_buckets = 256; /* must be power of 2 */
 161 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
 162
 163 /*
 164  * Timeouts for various events in handing dynamic rules.
 165  */
 166 static uint32_t dyn_ack_lifetime = 300;
 167 static uint32_t dyn_syn_lifetime = 20;
 168 static uint32_t dyn_fin_lifetime = 1;
 169 static uint32_t dyn_rst_lifetime = 1;
 170 static uint32_t dyn_udp_lifetime = 10;
 171 static uint32_t dyn_short_lifetime = 5;
 172
 173 /*
 174  * Keepalives are sent if dyn_keepalive is set. They are sent every
 175  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 176  * seconds of lifetime of a rule.
 177  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 178  * than dyn_keepalive_period.
 179  */
 180
 181 static uint32_t dyn_keepalive_interval = 20;
 182 static uint32_t dyn_keepalive_period = 5;
 183 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
 184
 185 static uint32_t static_count;           /* # of static rules */
 186 static uint32_t static_ioc_len; /* bytes of static rules */
 187 static uint32_t dyn_count;              /* # of dynamic rules */
 188 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
 189
 190 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
 191     &dyn_buckets, 0, "Number of dyn. buckets");
 192 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 193     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 194 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 195     &dyn_count, 0, "Number of dyn. rules");
 196 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 197     &dyn_max, 0, "Max number of dyn. rules");
 198 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 199     &static_count, 0, "Number of static rules");
 200 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 201     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 202 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 203     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 204 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
 205     &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
 206 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
 207     &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
 208 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 209     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 210 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 211     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 212 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 213     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 214
 215 #endif /* SYSCTL_NODE */
 216
 217 /**
 218  * dummynet needs a reference to the default rule, because rules can be
 219  * deleted while packets hold a reference to them. When this happens,
 220  * dummynet changes the reference to the default rule (it could well be a
 221  * NULL pointer, but this way we do not need to check for the special
 222  * case, plus here he have info on the default behaviour).
 223  */
 224 struct ip_fw *ip_fw_default_rule;
 225
 226 static ip_fw_chk_t      ipfw_chk;
 227
 228 static __inline int
 229 ipfw_free_rule(struct ip_fw *rule)
 230 {
 231         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
 232         atomic_subtract_int(&rule->refcnt, 1);
 233         if (atomic_cmpset_int(&rule->refcnt, 0, 1)) {
 234                 kfree(rule, M_IPFW);
 235                 return 1;
 236         }
 237         return 0;
 238 }
 239
 240 static void
 241 ipfw_unref_rule(void *priv)
 242 {
 243         ipfw_free_rule(priv);
 244 #ifdef KLD_MODULE
 245         atomic_subtract_int(&ipfw_refcnt, 1);
 246 #endif
 247 }
 248
 249 static __inline void
 250 ipfw_ref_rule(struct ip_fw *rule)
 251 {
 252 #ifdef KLD_MODULE
 253         atomic_add_int(&ipfw_refcnt, 1);
 254 #endif
 255         atomic_add_int(&rule->refcnt, 1);
 256 }
 257
 258 /*
 259  * This macro maps an ip pointer into a layer3 header pointer of type T
 260  */
 261 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
 262
 263 static __inline int
 264 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 265 {
 266         int type = L3HDR(struct icmp,ip)->icmp_type;
 267
 268         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
 269 }
 270
 271 #define TT      ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
 272     (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
 273
 274 static int
 275 is_icmp_query(struct ip *ip)
 276 {
 277         int type = L3HDR(struct icmp, ip)->icmp_type;
 278         return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
 279 }
 280 #undef TT
 281
 282 /*
 283  * The following checks use two arrays of 8 or 16 bits to store the
 284  * bits that we want set or clear, respectively. They are in the
 285  * low and high half of cmd->arg1 or cmd->d[0].
 286  *
 287  * We scan options and store the bits we find set. We succeed if
 288  *
 289  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 290  *
 291  * The code is sometimes optimized not to store additional variables.
 292  */
 293
 294 static int
 295 flags_match(ipfw_insn *cmd, uint8_t bits)
 296 {
 297         u_char want_clear;
 298         bits = ~bits;
 299
 300         if ( ((cmd->arg1 & 0xff) & bits) != 0)
 301                 return 0; /* some bits we want set were clear */
 302         want_clear = (cmd->arg1 >> 8) & 0xff;
 303         if ( (want_clear & bits) != want_clear)
 304                 return 0; /* some bits we want clear were set */
 305         return 1;
 306 }
 307
 308 static int
 309 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 310 {
 311         int optlen, bits = 0;
 312         u_char *cp = (u_char *)(ip + 1);
 313         int x = (ip->ip_hl << 2) - sizeof (struct ip);
 314
 315         for (; x > 0; x -= optlen, cp += optlen) {
 316                 int opt = cp[IPOPT_OPTVAL];
 317
 318                 if (opt == IPOPT_EOL)
 319                         break;
 320                 if (opt == IPOPT_NOP)
 321                         optlen = 1;
 322                 else {
 323                         optlen = cp[IPOPT_OLEN];
 324                         if (optlen <= 0 || optlen > x)
 325                                 return 0; /* invalid or truncated */
 326                 }
 327                 switch (opt) {
 328
 329                 default:
 330                         break;
 331
 332                 case IPOPT_LSRR:
 333                         bits |= IP_FW_IPOPT_LSRR;
 334                         break;
 335
 336                 case IPOPT_SSRR:
 337                         bits |= IP_FW_IPOPT_SSRR;
 338                         break;
 339
 340                 case IPOPT_RR:
 341                         bits |= IP_FW_IPOPT_RR;
 342                         break;
 343
 344                 case IPOPT_TS:
 345                         bits |= IP_FW_IPOPT_TS;
 346                         break;
 347                 }
 348         }
 349         return (flags_match(cmd, bits));
 350 }
 351
 352 static int
 353 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 354 {
 355         int optlen, bits = 0;
 356         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 357         u_char *cp = (u_char *)(tcp + 1);
 358         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 359
 360         for (; x > 0; x -= optlen, cp += optlen) {
 361                 int opt = cp[0];
 362                 if (opt == TCPOPT_EOL)
 363                         break;
 364                 if (opt == TCPOPT_NOP)
 365                         optlen = 1;
 366                 else {
 367                         optlen = cp[1];
 368                         if (optlen <= 0)
 369                                 break;
 370                 }
 371
 372                 switch (opt) {
 373
 374                 default:
 375                         break;
 376
 377                 case TCPOPT_MAXSEG:
 378                         bits |= IP_FW_TCPOPT_MSS;
 379                         break;
 380
 381                 case TCPOPT_WINDOW:
 382                         bits |= IP_FW_TCPOPT_WINDOW;
 383                         break;
 384
 385                 case TCPOPT_SACK_PERMITTED:
 386                 case TCPOPT_SACK:
 387                         bits |= IP_FW_TCPOPT_SACK;
 388                         break;
 389
 390                 case TCPOPT_TIMESTAMP:
 391                         bits |= IP_FW_TCPOPT_TS;
 392                         break;
 393
 394                 case TCPOPT_CC:
 395                 case TCPOPT_CCNEW:
 396                 case TCPOPT_CCECHO:
 397                         bits |= IP_FW_TCPOPT_CC;
 398                         break;
 399                 }
 400         }
 401         return (flags_match(cmd, bits));
 402 }
 403
 404 static int
 405 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 406 {
 407         if (ifp == NULL)        /* no iface with this packet, match fails */
 408                 return 0;
 409         /* Check by name or by IP address */
 410         if (cmd->name[0] != '\0') { /* match by name */
 411                 /* Check name */
 412                 if (cmd->p.glob) {
 413                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 414                                 return(1);
 415                 } else {
 416                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 417                                 return(1);
 418                 }
 419         } else {
 420                 struct ifaddr *ia;
 421
 422                 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
 423                         if (ia->ifa_addr == NULL)
 424                                 continue;
 425                         if (ia->ifa_addr->sa_family != AF_INET)
 426                                 continue;
 427                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 428                             (ia->ifa_addr))->sin_addr.s_addr)
 429                                 return(1);      /* match */
 430                 }
 431         }
 432         return(0);      /* no match, fail ... */
 433 }
 434
 435 static uint64_t norule_counter; /* counter for ipfw_log(NULL...) */
 436
 437 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 438 #define SNP(buf) buf, sizeof(buf)
 439
 440 /*
 441  * We enter here when we have a rule with O_LOG.
 442  * XXX this function alone takes about 2Kbytes of code!
 443  */
 444 static void
 445 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 446         struct mbuf *m, struct ifnet *oif)
 447 {
 448         char *action;
 449         int limit_reached = 0;
 450         char action2[40], proto[48], fragment[28];
 451
 452         fragment[0] = '\0';
 453         proto[0] = '\0';
 454
 455         if (f == NULL) {        /* bogus pkt */
 456                 if (verbose_limit != 0 && norule_counter >= verbose_limit)
 457                         return;
 458                 norule_counter++;
 459                 if (norule_counter == verbose_limit)
 460                         limit_reached = verbose_limit;
 461                 action = "Refuse";
 462         } else {        /* O_LOG is the first action, find the real one */
 463                 ipfw_insn *cmd = ACTION_PTR(f);
 464                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 465
 466                 if (l->max_log != 0 && l->log_left == 0)
 467                         return;
 468                 l->log_left--;
 469                 if (l->log_left == 0)
 470                         limit_reached = l->max_log;
 471                 cmd += F_LEN(cmd);      /* point to first action */
 472                 if (cmd->opcode == O_PROB)
 473                         cmd += F_LEN(cmd);
 474
 475                 action = action2;
 476                 switch (cmd->opcode) {
 477                 case O_DENY:
 478                         action = "Deny";
 479                         break;
 480
 481                 case O_REJECT:
 482                         if (cmd->arg1==ICMP_REJECT_RST)
 483                                 action = "Reset";
 484                         else if (cmd->arg1==ICMP_UNREACH_HOST)
 485                                 action = "Reject";
 486                         else
 487                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 488                                         cmd->arg1);
 489                         break;
 490
 491                 case O_ACCEPT:
 492                         action = "Accept";
 493                         break;
 494                 case O_COUNT:
 495                         action = "Count";
 496                         break;
 497                 case O_DIVERT:
 498                         ksnprintf(SNPARGS(action2, 0), "Divert %d",
 499                                 cmd->arg1);
 500                         break;
 501                 case O_TEE:
 502                         ksnprintf(SNPARGS(action2, 0), "Tee %d",
 503                                 cmd->arg1);
 504                         break;
 505                 case O_SKIPTO:
 506                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d",
 507                                 cmd->arg1);
 508                         break;
 509                 case O_PIPE:
 510                         ksnprintf(SNPARGS(action2, 0), "Pipe %d",
 511                                 cmd->arg1);
 512                         break;
 513                 case O_QUEUE:
 514                         ksnprintf(SNPARGS(action2, 0), "Queue %d",
 515                                 cmd->arg1);
 516                         break;
 517                 case O_FORWARD_IP: {
 518                         ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 519                         int len;
 520
 521                         len = ksnprintf(SNPARGS(action2, 0), "Forward to %s",
 522                                 inet_ntoa(sa->sa.sin_addr));
 523                         if (sa->sa.sin_port)
 524                                 ksnprintf(SNPARGS(action2, len), ":%d",
 525                                     sa->sa.sin_port);
 526                         }
 527                         break;
 528                 default:
 529                         action = "UNKNOWN";
 530                         break;
 531                 }
 532         }
 533
 534         if (hlen == 0) {        /* non-ip */
 535                 ksnprintf(SNPARGS(proto, 0), "MAC");
 536         } else {
 537                 struct ip *ip = mtod(m, struct ip *);
 538                 /* these three are all aliases to the same thing */
 539                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 540                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 541                 struct udphdr *const udp = (struct udphdr *)icmp;
 542
 543                 int ip_off, offset, ip_len;
 544
 545                 int len;
 546
 547                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 548                         ip_off = ntohs(ip->ip_off);
 549                         ip_len = ntohs(ip->ip_len);
 550                 } else {
 551                         ip_off = ip->ip_off;
 552                         ip_len = ip->ip_len;
 553                 }
 554                 offset = ip_off & IP_OFFMASK;
 555                 switch (ip->ip_p) {
 556                 case IPPROTO_TCP:
 557                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 558                             inet_ntoa(ip->ip_src));
 559                         if (offset == 0)
 560                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 561                                     ntohs(tcp->th_sport),
 562                                     inet_ntoa(ip->ip_dst),
 563                                     ntohs(tcp->th_dport));
 564                         else
 565                                 ksnprintf(SNPARGS(proto, len), " %s",
 566                                     inet_ntoa(ip->ip_dst));
 567                         break;
 568
 569                 case IPPROTO_UDP:
 570                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 571                                 inet_ntoa(ip->ip_src));
 572                         if (offset == 0)
 573                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 574                                     ntohs(udp->uh_sport),
 575                                     inet_ntoa(ip->ip_dst),
 576                                     ntohs(udp->uh_dport));
 577                         else
 578                                 ksnprintf(SNPARGS(proto, len), " %s",
 579                                     inet_ntoa(ip->ip_dst));
 580                         break;
 581
 582                 case IPPROTO_ICMP:
 583                         if (offset == 0)
 584                                 len = ksnprintf(SNPARGS(proto, 0),
 585                                     "ICMP:%u.%u ",
 586                                     icmp->icmp_type, icmp->icmp_code);
 587                         else
 588                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 589                         len += ksnprintf(SNPARGS(proto, len), "%s",
 590                             inet_ntoa(ip->ip_src));
 591                         ksnprintf(SNPARGS(proto, len), " %s",
 592                             inet_ntoa(ip->ip_dst));
 593                         break;
 594
 595                 default:
 596                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 597                             inet_ntoa(ip->ip_src));
 598                         ksnprintf(SNPARGS(proto, len), " %s",
 599                             inet_ntoa(ip->ip_dst));
 600                         break;
 601                 }
 602
 603                 if (ip_off & (IP_MF | IP_OFFMASK))
 604                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 605                              ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 606                              offset << 3,
 607                              (ip_off & IP_MF) ? "+" : "");
 608         }
 609         if (oif || m->m_pkthdr.rcvif)
 610                 log(LOG_SECURITY | LOG_INFO,
 611                     "ipfw: %d %s %s %s via %s%s\n",
 612                     f ? f->rulenum : -1,
 613                     action, proto, oif ? "out" : "in",
 614                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 615                     fragment);
 616         else
 617                 log(LOG_SECURITY | LOG_INFO,
 618                     "ipfw: %d %s %s [no if info]%s\n",
 619                     f ? f->rulenum : -1,
 620                     action, proto, fragment);
 621         if (limit_reached)
 622                 log(LOG_SECURITY | LOG_NOTICE,
 623                     "ipfw: limit %d reached on entry %d\n",
 624                     limit_reached, f ? f->rulenum : -1);
 625 }
 626
 627 /*
 628  * IMPORTANT: the hash function for dynamic rules must be commutative
 629  * in source and destination (ip,port), because rules are bidirectional
 630  * and we want to find both in the same bucket.
 631  */
 632 static __inline int
 633 hash_packet(struct ipfw_flow_id *id)
 634 {
 635         uint32_t i;
 636
 637         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 638         i &= (curr_dyn_buckets - 1);
 639         return i;
 640 }
 641
 642 /**
 643  * unlink a dynamic rule from a chain. prev is a pointer to
 644  * the previous one, q is a pointer to the rule to delete,
 645  * head is a pointer to the head of the queue.
 646  * Modifies q and potentially also head.
 647  */
 648 #define UNLINK_DYN_RULE(prev, head, q) {                                \
 649         ipfw_dyn_rule *old_q = q;                                       \
 650                                                                         \
 651         /* remove a refcount to the parent */                           \
 652         if (q->dyn_type == O_LIMIT)                                     \
 653                 q->parent->count--;                                     \
 654         DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",        \
 655                 (q->id.src_ip), (q->id.src_port),                       \
 656                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
 657         if (prev != NULL)                                               \
 658                 prev->next = q = q->next;                               \
 659         else                                                            \
 660                 head = q = q->next;                                     \
 661         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
 662         dyn_count--;                                                    \
 663         kfree(old_q, M_IPFW); }
 664
 665 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
 666
 667 /**
 668  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 669  *
 670  * If keep_me == NULL, rules are deleted even if not expired,
 671  * otherwise only expired rules are removed.
 672  *
 673  * The value of the second parameter is also used to point to identify
 674  * a rule we absolutely do not want to remove (e.g. because we are
 675  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 676  * rules). The pointer is only used for comparison, so any non-null
 677  * value will do.
 678  */
 679 static void
 680 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 681 {
 682         static uint32_t last_remove = 0;
 683
 684 #define FORCE (keep_me == NULL)
 685
 686         ipfw_dyn_rule *prev, *q;
 687         int i, pass = 0, max_pass = 0;
 688
 689         if (ipfw_dyn_v == NULL || dyn_count == 0)
 690                 return;
 691         /* do not expire more than once per second, it is useless */
 692         if (!FORCE && last_remove == time_second)
 693                 return;
 694         last_remove = time_second;
 695
 696         /*
 697          * because O_LIMIT refer to parent rules, during the first pass only
 698          * remove child and mark any pending LIMIT_PARENT, and remove
 699          * them in a second pass.
 700          */
 701 next_pass:
 702         for (i = 0 ; i < curr_dyn_buckets ; i++) {
 703                 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
 704                         /*
 705                          * Logic can become complex here, so we split tests.
 706                          */
 707                         if (q == keep_me)
 708                                 goto next;
 709                         if (rule != NULL && rule != q->rule)
 710                                 goto next; /* not the one we are looking for */
 711                         if (q->dyn_type == O_LIMIT_PARENT) {
 712                                 /*
 713                                  * handle parent in the second pass,
 714                                  * record we need one.
 715                                  */
 716                                 max_pass = 1;
 717                                 if (pass == 0)
 718                                         goto next;
 719                                 if (FORCE && q->count != 0 ) {
 720                                         /* XXX should not happen! */
 721                                         kprintf( "OUCH! cannot remove rule,"
 722                                              " count %d\n", q->count);
 723                                 }
 724                         } else {
 725                                 if (!FORCE &&
 726                                     !TIME_LEQ( q->expire, time_second ))
 727                                         goto next;
 728                         }
 729                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 730                         continue;
 731 next:
 732                         prev=q;
 733                         q=q->next;
 734                 }
 735         }
 736         if (pass++ < max_pass)
 737                 goto next_pass;
 738 }
 739
 740
 741 /**
 742  * lookup a dynamic rule.
 743  */
 744 static ipfw_dyn_rule *
 745 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 746         struct tcphdr *tcp)
 747 {
 748         /*
 749          * stateful ipfw extensions.
 750          * Lookup into dynamic session queue
 751          */
 752 #define MATCH_REVERSE   0
 753 #define MATCH_FORWARD   1
 754 #define MATCH_NONE      2
 755 #define MATCH_UNKNOWN   3
 756         int i, dir = MATCH_NONE;
 757         ipfw_dyn_rule *prev, *q=NULL;
 758
 759         if (ipfw_dyn_v == NULL)
 760                 goto done;      /* not found */
 761         i = hash_packet( pkt );
 762         for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
 763                 if (q->dyn_type == O_LIMIT_PARENT)
 764                         goto next;
 765                 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
 766                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 767                         continue;
 768                 }
 769                 if ( pkt->proto == q->id.proto) {
 770                         if (pkt->src_ip == q->id.src_ip &&
 771                             pkt->dst_ip == q->id.dst_ip &&
 772                             pkt->src_port == q->id.src_port &&
 773                             pkt->dst_port == q->id.dst_port ) {
 774                                 dir = MATCH_FORWARD;
 775                                 break;
 776                         }
 777                         if (pkt->src_ip == q->id.dst_ip &&
 778                             pkt->dst_ip == q->id.src_ip &&
 779                             pkt->src_port == q->id.dst_port &&
 780                             pkt->dst_port == q->id.src_port ) {
 781                                 dir = MATCH_REVERSE;
 782                                 break;
 783                         }
 784                 }
 785 next:
 786                 prev = q;
 787                 q = q->next;
 788         }
 789         if (q == NULL)
 790                 goto done; /* q = NULL, not found */
 791
 792         if ( prev != NULL) { /* found and not in front */
 793                 prev->next = q->next;
 794                 q->next = ipfw_dyn_v[i];
 795                 ipfw_dyn_v[i] = q;
 796         }
 797         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
 798                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
 799
 800 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
 801 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
 802                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
 803                 switch (q->state) {
 804                 case TH_SYN:                            /* opening */
 805                         q->expire = time_second + dyn_syn_lifetime;
 806                         break;
 807
 808                 case BOTH_SYN:                  /* move to established */
 809                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
 810                 case BOTH_SYN | (TH_FIN << 8) :
 811                         if (tcp) {
 812 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
 813                             uint32_t ack = ntohl(tcp->th_ack);
 814                             if (dir == MATCH_FORWARD) {
 815                                 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
 816                                     q->ack_fwd = ack;
 817                                 else { /* ignore out-of-sequence */
 818                                     break;
 819                                 }
 820                             } else {
 821                                 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
 822                                     q->ack_rev = ack;
 823                                 else { /* ignore out-of-sequence */
 824                                     break;
 825                                 }
 826                             }
 827                         }
 828                         q->expire = time_second + dyn_ack_lifetime;
 829                         break;
 830
 831                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
 832                         if (dyn_fin_lifetime >= dyn_keepalive_period)
 833                                 dyn_fin_lifetime = dyn_keepalive_period - 1;
 834                         q->expire = time_second + dyn_fin_lifetime;
 835                         break;
 836
 837                 default:
 838 #if 0
 839                         /*
 840                          * reset or some invalid combination, but can also
 841                          * occur if we use keep-state the wrong way.
 842                          */
 843                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
 844                                 kprintf("invalid state: 0x%x\n", q->state);
 845 #endif
 846                         if (dyn_rst_lifetime >= dyn_keepalive_period)
 847                                 dyn_rst_lifetime = dyn_keepalive_period - 1;
 848                         q->expire = time_second + dyn_rst_lifetime;
 849                         break;
 850                 }
 851         } else if (pkt->proto == IPPROTO_UDP) {
 852                 q->expire = time_second + dyn_udp_lifetime;
 853         } else {
 854                 /* other protocols */
 855                 q->expire = time_second + dyn_short_lifetime;
 856         }
 857 done:
 858         if (match_direction)
 859                 *match_direction = dir;
 860         return q;
 861 }
 862
 863 static void
 864 realloc_dynamic_table(void)
 865 {
 866         /*
 867          * Try reallocation, make sure we have a power of 2 and do
 868          * not allow more than 64k entries. In case of overflow,
 869          * default to 1024.
 870          */
 871
 872         if (dyn_buckets > 65536)
 873                 dyn_buckets = 1024;
 874         if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
 875                 dyn_buckets = curr_dyn_buckets; /* reset */
 876                 return;
 877         }
 878         curr_dyn_buckets = dyn_buckets;
 879         if (ipfw_dyn_v != NULL)
 880                 kfree(ipfw_dyn_v, M_IPFW);
 881         for (;;) {
 882                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
 883                        M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
 884                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
 885                         break;
 886                 curr_dyn_buckets /= 2;
 887         }
 888 }
 889
 890 /**
 891  * Install state of type 'type' for a dynamic session.
 892  * The hash table contains two type of rules:
 893  * - regular rules (O_KEEP_STATE)
 894  * - rules for sessions with limited number of sess per user
 895  *   (O_LIMIT). When they are created, the parent is
 896  *   increased by 1, and decreased on delete. In this case,
 897  *   the third parameter is the parent rule and not the chain.
 898  * - "parent" rules for the above (O_LIMIT_PARENT).
 899  */
 900 static ipfw_dyn_rule *
 901 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
 902 {
 903         ipfw_dyn_rule *r;
 904         int i;
 905
 906         if (ipfw_dyn_v == NULL ||
 907             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
 908                 realloc_dynamic_table();
 909                 if (ipfw_dyn_v == NULL)
 910                         return NULL; /* failed ! */
 911         }
 912         i = hash_packet(id);
 913
 914         r = kmalloc(sizeof *r, M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
 915         if (r == NULL) {
 916                 kprintf ("sorry cannot allocate state\n");
 917                 return NULL;
 918         }
 919
 920         /* increase refcount on parent, and set pointer */
 921         if (dyn_type == O_LIMIT) {
 922                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
 923                 if ( parent->dyn_type != O_LIMIT_PARENT)
 924                         panic("invalid parent");
 925                 parent->count++;
 926                 r->parent = parent;
 927                 rule = parent->rule;
 928         }
 929
 930         r->id = *id;
 931         r->expire = time_second + dyn_syn_lifetime;
 932         r->rule = rule;
 933         r->dyn_type = dyn_type;
 934         r->pcnt = r->bcnt = 0;
 935         r->count = 0;
 936
 937         r->bucket = i;
 938         r->next = ipfw_dyn_v[i];
 939         ipfw_dyn_v[i] = r;
 940         dyn_count++;
 941         DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
 942            dyn_type,
 943            (r->id.src_ip), (r->id.src_port),
 944            (r->id.dst_ip), (r->id.dst_port),
 945            dyn_count ); )
 946         return r;
 947 }
 948
 949 /**
 950  * lookup dynamic parent rule using pkt and rule as search keys.
 951  * If the lookup fails, then install one.
 952  */
 953 static ipfw_dyn_rule *
 954 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
 955 {
 956         ipfw_dyn_rule *q;
 957         int i;
 958
 959         if (ipfw_dyn_v) {
 960                 i = hash_packet( pkt );
 961                 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
 962                         if (q->dyn_type == O_LIMIT_PARENT &&
 963                             rule== q->rule &&
 964                             pkt->proto == q->id.proto &&
 965                             pkt->src_ip == q->id.src_ip &&
 966                             pkt->dst_ip == q->id.dst_ip &&
 967                             pkt->src_port == q->id.src_port &&
 968                             pkt->dst_port == q->id.dst_port) {
 969                                 q->expire = time_second + dyn_short_lifetime;
 970                                 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
 971                                 return q;
 972                         }
 973         }
 974         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
 975 }
 976
 977 /**
 978  * Install dynamic state for rule type cmd->o.opcode
 979  *
 980  * Returns 1 (failure) if state is not installed because of errors or because
 981  * session limitations are enforced.
 982  */
 983 static int
 984 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
 985         struct ip_fw_args *args)
 986 {
 987         static int last_log;
 988
 989         ipfw_dyn_rule *q;
 990
 991         DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
 992             cmd->o.opcode,
 993             (args->f_id.src_ip), (args->f_id.src_port),
 994             (args->f_id.dst_ip), (args->f_id.dst_port) );)
 995
 996         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
 997
 998         if (q != NULL) { /* should never occur */
 999                 if (last_log != time_second) {
1000                         last_log = time_second;
1001                         kprintf(" install_state: entry already present, done\n");
1002                 }
1003                 return 0;
1004         }
1005
1006         if (dyn_count >= dyn_max)
1007                 /*
1008                  * Run out of slots, try to remove any expired rule.
1009                  */
1010                 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1011
1012         if (dyn_count >= dyn_max) {
1013                 if (last_log != time_second) {
1014                         last_log = time_second;
1015                         kprintf("install_state: Too many dynamic rules\n");
1016                 }
1017                 return 1; /* cannot install, notify caller */
1018         }
1019
1020         switch (cmd->o.opcode) {
1021         case O_KEEP_STATE: /* bidir rule */
1022                 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1023                 break;
1024
1025         case O_LIMIT: /* limit number of sessions */
1026             {
1027                 uint16_t limit_mask = cmd->limit_mask;
1028                 struct ipfw_flow_id id;
1029                 ipfw_dyn_rule *parent;
1030
1031                 DEB(kprintf("installing dyn-limit rule %d\n", cmd->conn_limit);)
1032
1033                 id.dst_ip = id.src_ip = 0;
1034                 id.dst_port = id.src_port = 0;
1035                 id.proto = args->f_id.proto;
1036
1037                 if (limit_mask & DYN_SRC_ADDR)
1038                         id.src_ip = args->f_id.src_ip;
1039                 if (limit_mask & DYN_DST_ADDR)
1040                         id.dst_ip = args->f_id.dst_ip;
1041                 if (limit_mask & DYN_SRC_PORT)
1042                         id.src_port = args->f_id.src_port;
1043                 if (limit_mask & DYN_DST_PORT)
1044                         id.dst_port = args->f_id.dst_port;
1045                 parent = lookup_dyn_parent(&id, rule);
1046                 if (parent == NULL) {
1047                         kprintf("add parent failed\n");
1048                         return 1;
1049                 }
1050                 if (parent->count >= cmd->conn_limit) {
1051                         /*
1052                          * See if we can remove some expired rule.
1053                          */
1054                         remove_dyn_rule(rule, parent);
1055                         if (parent->count >= cmd->conn_limit) {
1056                                 if (fw_verbose && last_log != time_second) {
1057                                         last_log = time_second;
1058                                         log(LOG_SECURITY | LOG_DEBUG,
1059                                             "drop session, too many entries\n");
1060                                 }
1061                                 return 1;
1062                         }
1063                 }
1064                 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1065             }
1066                 break;
1067         default:
1068                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1069                 return 1;
1070         }
1071         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1072         return 0;
1073 }
1074
1075 /*
1076  * Transmit a TCP packet, containing either a RST or a keepalive.
1077  * When flags & TH_RST, we are sending a RST packet, because of a
1078  * "reset" action matched the packet.
1079  * Otherwise we are sending a keepalive, and flags & TH_
1080  */
1081 static void
1082 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1083 {
1084         struct mbuf *m;
1085         struct ip *ip;
1086         struct tcphdr *tcp;
1087         struct route sro;       /* fake route */
1088
1089         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1090         if (m == 0)
1091                 return;
1092         m->m_pkthdr.rcvif = (struct ifnet *)0;
1093         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1094         m->m_data += max_linkhdr;
1095
1096         ip = mtod(m, struct ip *);
1097         bzero(ip, m->m_len);
1098         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1099         ip->ip_p = IPPROTO_TCP;
1100         tcp->th_off = 5;
1101         /*
1102          * Assume we are sending a RST (or a keepalive in the reverse
1103          * direction), swap src and destination addresses and ports.
1104          */
1105         ip->ip_src.s_addr = htonl(id->dst_ip);
1106         ip->ip_dst.s_addr = htonl(id->src_ip);
1107         tcp->th_sport = htons(id->dst_port);
1108         tcp->th_dport = htons(id->src_port);
1109         if (flags & TH_RST) {   /* we are sending a RST */
1110                 if (flags & TH_ACK) {
1111                         tcp->th_seq = htonl(ack);
1112                         tcp->th_ack = htonl(0);
1113                         tcp->th_flags = TH_RST;
1114                 } else {
1115                         if (flags & TH_SYN)
1116                                 seq++;
1117                         tcp->th_seq = htonl(0);
1118                         tcp->th_ack = htonl(seq);
1119                         tcp->th_flags = TH_RST | TH_ACK;
1120                 }
1121         } else {
1122                 /*
1123                  * We are sending a keepalive. flags & TH_SYN determines
1124                  * the direction, forward if set, reverse if clear.
1125                  * NOTE: seq and ack are always assumed to be correct
1126                  * as set by the caller. This may be confusing...
1127                  */
1128                 if (flags & TH_SYN) {
1129                         /*
1130                          * we have to rewrite the correct addresses!
1131                          */
1132                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1133                         ip->ip_src.s_addr = htonl(id->src_ip);
1134                         tcp->th_dport = htons(id->dst_port);
1135                         tcp->th_sport = htons(id->src_port);
1136                 }
1137                 tcp->th_seq = htonl(seq);
1138                 tcp->th_ack = htonl(ack);
1139                 tcp->th_flags = TH_ACK;
1140         }
1141         /*
1142          * set ip_len to the payload size so we can compute
1143          * the tcp checksum on the pseudoheader
1144          * XXX check this, could save a couple of words ?
1145          */
1146         ip->ip_len = htons(sizeof(struct tcphdr));
1147         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1148         /*
1149          * now fill fields left out earlier
1150          */
1151         ip->ip_ttl = ip_defttl;
1152         ip->ip_len = m->m_pkthdr.len;
1153         bzero (&sro, sizeof (sro));
1154         ip_rtaddr(ip->ip_dst, &sro);
1155         m->m_pkthdr.fw_flags |= IPFW_MBUF_SKIP_FIREWALL;
1156         ip_output(m, NULL, &sro, 0, NULL, NULL);
1157         if (sro.ro_rt)
1158                 RTFREE(sro.ro_rt);
1159 }
1160
1161 /*
1162  * sends a reject message, consuming the mbuf passed as an argument.
1163  */
1164 static void
1165 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1166 {
1167
1168         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1169                 /* We need the IP header in host order for icmp_error(). */
1170                 if (args->eh != NULL) {
1171                         struct ip *ip = mtod(args->m, struct ip *);
1172                         ip->ip_len = ntohs(ip->ip_len);
1173                         ip->ip_off = ntohs(ip->ip_off);
1174                 }
1175                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1176         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1177                 struct tcphdr *const tcp =
1178                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1179                 if ( (tcp->th_flags & TH_RST) == 0)
1180                         send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1181                                 ntohl(tcp->th_ack),
1182                                 tcp->th_flags | TH_RST);
1183                 m_freem(args->m);
1184         } else
1185                 m_freem(args->m);
1186         args->m = NULL;
1187 }
1188
1189 /**
1190  *
1191  * Given an ip_fw *, lookup_next_rule will return a pointer
1192  * to the next rule, which can be either the jump
1193  * target (for skipto instructions) or the next one in the list (in
1194  * all other cases including a missing jump target).
1195  * The result is also written in the "next_rule" field of the rule.
1196  * Backward jumps are not allowed, so start looking from the next
1197  * rule...
1198  *
1199  * This never returns NULL -- in case we do not have an exact match,
1200  * the next rule is returned. When the ruleset is changed,
1201  * pointers are flushed so we are always correct.
1202  */
1203
1204 static struct ip_fw *
1205 lookup_next_rule(struct ip_fw *me)
1206 {
1207         struct ip_fw *rule = NULL;
1208         ipfw_insn *cmd;
1209
1210         /* look for action, in case it is a skipto */
1211         cmd = ACTION_PTR(me);
1212         if (cmd->opcode == O_LOG)
1213                 cmd += F_LEN(cmd);
1214         if ( cmd->opcode == O_SKIPTO )
1215                 for (rule = me->next; rule ; rule = rule->next)
1216                         if (rule->rulenum >= cmd->arg1)
1217                                 break;
1218         if (rule == NULL)                       /* failure or not a skipto */
1219                 rule = me->next;
1220         me->next_rule = rule;
1221         return rule;
1222 }
1223
1224 /*
1225  * The main check routine for the firewall.
1226  *
1227  * All arguments are in args so we can modify them and return them
1228  * back to the caller.
1229  *
1230  * Parameters:
1231  *
1232  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1233  *              Starts with the IP header.
1234  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1235  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1236  *              The incoming interface is in the mbuf. (in)
1237  *
1238  *      args->rule      Pointer to the last matching rule (in/out)
1239  *      args->next_hop  Socket we are forwarding to (out).
1240  *      args->f_id      Addresses grabbed from the packet (out)
1241  *
1242  * Return value:
1243  *
1244  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1245  *      0       The packet is to be accepted and routed normally OR
1246  *              the packet was denied/rejected and has been dropped;
1247  *              in the latter case, *m is equal to NULL upon return.
1248  *      port    Divert the packet to port, with these caveats:
1249  *
1250  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1251  *                of diverting it (ie, 'ipfw tee').
1252  *
1253  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1254  *                16 bits as a dummynet pipe number instead of diverting
1255  */
1256
1257 static int
1258 ipfw_chk(struct ip_fw_args *args)
1259 {
1260         /*
1261          * Local variables hold state during the processing of a packet.
1262          *
1263          * IMPORTANT NOTE: to speed up the processing of rules, there
1264          * are some assumption on the values of the variables, which
1265          * are documented here. Should you change them, please check
1266          * the implementation of the various instructions to make sure
1267          * that they still work.
1268          *
1269          * args->eh     The MAC header. It is non-null for a layer2
1270          *      packet, it is NULL for a layer-3 packet.
1271          *
1272          * m | args->m  Pointer to the mbuf, as received from the caller.
1273          *      It may change if ipfw_chk() does an m_pullup, or if it
1274          *      consumes the packet because it calls send_reject().
1275          *      XXX This has to change, so that ipfw_chk() never modifies
1276          *      or consumes the buffer.
1277          * ip   is simply an alias of the value of m, and it is kept
1278          *      in sync with it (the packet is  supposed to start with
1279          *      the ip header).
1280          */
1281         struct mbuf *m = args->m;
1282         struct ip *ip = mtod(m, struct ip *);
1283
1284         /*
1285          * oif | args->oif      If NULL, ipfw_chk has been called on the
1286          *      inbound path (ether_input, ip_input).
1287          *      If non-NULL, ipfw_chk has been called on the outbound path
1288          *      (ether_output, ip_output).
1289          */
1290         struct ifnet *oif = args->oif;
1291
1292         struct ip_fw *f = NULL;         /* matching rule */
1293         int retval = 0;
1294         struct m_tag *mtag;
1295
1296         /*
1297          * hlen The length of the IPv4 header.
1298          *      hlen >0 means we have an IPv4 packet.
1299          */
1300         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1301
1302         /*
1303          * offset       The offset of a fragment. offset != 0 means that
1304          *      we have a fragment at this offset of an IPv4 packet.
1305          *      offset == 0 means that (if this is an IPv4 packet)
1306          *      this is the first or only fragment.
1307          */
1308         u_short offset = 0;
1309
1310         /*
1311          * Local copies of addresses. They are only valid if we have
1312          * an IP packet.
1313          *
1314          * proto        The protocol. Set to 0 for non-ip packets,
1315          *      or to the protocol read from the packet otherwise.
1316          *      proto != 0 means that we have an IPv4 packet.
1317          *
1318          * src_port, dst_port   port numbers, in HOST format. Only
1319          *      valid for TCP and UDP packets.
1320          *
1321          * src_ip, dst_ip       ip addresses, in NETWORK format.
1322          *      Only valid for IPv4 packets.
1323          */
1324         uint8_t proto;
1325         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1326         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1327         uint16_t ip_len=0;
1328         int dyn_dir = MATCH_UNKNOWN;
1329         ipfw_dyn_rule *q = NULL;
1330
1331         if (m->m_pkthdr.fw_flags & IPFW_MBUF_SKIP_FIREWALL)
1332                 return 0;       /* accept */
1333         /*
1334          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1335          *      MATCH_NONE when checked and not matched (q = NULL),
1336          *      MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1337          */
1338
1339         if (args->eh == NULL ||         /* layer 3 packet */
1340                 ( m->m_pkthdr.len >= sizeof(struct ip) &&
1341                     ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1342                         hlen = ip->ip_hl << 2;
1343
1344         /*
1345          * Collect parameters into local variables for faster matching.
1346          */
1347         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1348                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1349                 goto after_ip_checks;
1350         }
1351
1352         proto = args->f_id.proto = ip->ip_p;
1353         src_ip = ip->ip_src;
1354         dst_ip = ip->ip_dst;
1355         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1356                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1357                 ip_len = ntohs(ip->ip_len);
1358         } else {
1359                 offset = ip->ip_off & IP_OFFMASK;
1360                 ip_len = ip->ip_len;
1361         }
1362
1363 #define PULLUP_TO(len)                                          \
1364                 do {                                            \
1365                         if ((m)->m_len < (len)) {               \
1366                             args->m = m = m_pullup(m, (len));   \
1367                             if (m == 0)                         \
1368                                 goto pullup_failed;             \
1369                             ip = mtod(m, struct ip *);          \
1370                         }                                       \
1371                 } while (0)
1372
1373         if (offset == 0) {
1374                 switch (proto) {
1375                 case IPPROTO_TCP:
1376                     {
1377                         struct tcphdr *tcp;
1378
1379                         PULLUP_TO(hlen + sizeof(struct tcphdr));
1380                         tcp = L3HDR(struct tcphdr, ip);
1381                         dst_port = tcp->th_dport;
1382                         src_port = tcp->th_sport;
1383                         args->f_id.flags = tcp->th_flags;
1384                         }
1385                         break;
1386
1387                 case IPPROTO_UDP:
1388                     {
1389                         struct udphdr *udp;
1390
1391                         PULLUP_TO(hlen + sizeof(struct udphdr));
1392                         udp = L3HDR(struct udphdr, ip);
1393                         dst_port = udp->uh_dport;
1394                         src_port = udp->uh_sport;
1395                         }
1396                         break;
1397
1398                 case IPPROTO_ICMP:
1399                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1400                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1401                         break;
1402
1403                 default:
1404                         break;
1405                 }
1406 #undef PULLUP_TO
1407         }
1408
1409         args->f_id.src_ip = ntohl(src_ip.s_addr);
1410         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1411         args->f_id.src_port = src_port = ntohs(src_port);
1412         args->f_id.dst_port = dst_port = ntohs(dst_port);
1413
1414 after_ip_checks:
1415         if (args->rule) {
1416                 /*
1417                  * Packet has already been tagged. Look for the next rule
1418                  * to restart processing.
1419                  *
1420                  * If fw_one_pass != 0 then just accept it.
1421                  * XXX should not happen here, but optimized out in
1422                  * the caller.
1423                  */
1424                 if (fw_one_pass)
1425                         return 0;
1426
1427                 /* This rule was deleted */
1428                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1429                         return IP_FW_PORT_DENY_FLAG;
1430
1431                 f = args->rule->next_rule;
1432                 if (f == NULL)
1433                         f = lookup_next_rule(args->rule);
1434         } else {
1435                 /*
1436                  * Find the starting rule. It can be either the first
1437                  * one, or the one after divert_rule if asked so.
1438                  */
1439                 int skipto;
1440
1441                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1442                 if (mtag != NULL)
1443                         skipto = *(uint16_t *)m_tag_data(mtag);
1444                 else
1445                         skipto = 0;
1446
1447                 f = layer3_chain;
1448                 if (args->eh == NULL && skipto != 0) {
1449                         if (skipto >= IPFW_DEFAULT_RULE)
1450                                 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1451                         while (f && f->rulenum <= skipto)
1452                                 f = f->next;
1453                         if (f == NULL)  /* drop packet */
1454                                 return(IP_FW_PORT_DENY_FLAG);
1455                 }
1456         }
1457         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1458                 m_tag_delete(m, mtag);
1459
1460         /*
1461          * Now scan the rules, and parse microinstructions for each rule.
1462          */
1463         for (; f; f = f->next) {
1464                 int l, cmdlen;
1465                 ipfw_insn *cmd;
1466                 int skip_or; /* skip rest of OR block */
1467
1468 again:
1469                 if (set_disable & (1 << f->set) )
1470                         continue;
1471
1472                 skip_or = 0;
1473                 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1474                     l -= cmdlen, cmd += cmdlen) {
1475                         int match;
1476
1477                         /*
1478                          * check_body is a jump target used when we find a
1479                          * CHECK_STATE, and need to jump to the body of
1480                          * the target rule.
1481                          */
1482
1483 check_body:
1484                         cmdlen = F_LEN(cmd);
1485                         /*
1486                          * An OR block (insn_1 || .. || insn_n) has the
1487                          * F_OR bit set in all but the last instruction.
1488                          * The first match will set "skip_or", and cause
1489                          * the following instructions to be skipped until
1490                          * past the one with the F_OR bit clear.
1491                          */
1492                         if (skip_or) {          /* skip this instruction */
1493                                 if ((cmd->len & F_OR) == 0)
1494                                         skip_or = 0;    /* next one is good */
1495                                 continue;
1496                         }
1497                         match = 0; /* set to 1 if we succeed */
1498
1499                         switch (cmd->opcode) {
1500                         /*
1501                          * The first set of opcodes compares the packet's
1502                          * fields with some pattern, setting 'match' if a
1503                          * match is found. At the end of the loop there is
1504                          * logic to deal with F_NOT and F_OR flags associated
1505                          * with the opcode.
1506                          */
1507                         case O_NOP:
1508                                 match = 1;
1509                                 break;
1510
1511                         case O_FORWARD_MAC:
1512                                 kprintf("ipfw: opcode %d unimplemented\n",
1513                                     cmd->opcode);
1514                                 break;
1515
1516                         case O_GID:
1517                         case O_UID:
1518                                 /*
1519                                  * We only check offset == 0 && proto != 0,
1520                                  * as this ensures that we have an IPv4
1521                                  * packet with the ports info.
1522                                  */
1523                                 if (offset!=0)
1524                                         break;
1525                             {
1526                                 struct inpcbinfo *pi;
1527                                 int wildcard;
1528                                 struct inpcb *pcb;
1529
1530                                 if (proto == IPPROTO_TCP) {
1531                                         wildcard = 0;
1532                                         pi = &tcbinfo[mycpu->gd_cpuid];
1533                                 } else if (proto == IPPROTO_UDP) {
1534                                         wildcard = 1;
1535                                         pi = &udbinfo;
1536                                 } else
1537                                         break;
1538
1539                                 pcb =  (oif) ?
1540                                         in_pcblookup_hash(pi,
1541                                             dst_ip, htons(dst_port),
1542                                             src_ip, htons(src_port),
1543                                             wildcard, oif) :
1544                                         in_pcblookup_hash(pi,
1545                                             src_ip, htons(src_port),
1546                                             dst_ip, htons(dst_port),
1547                                             wildcard, NULL);
1548
1549                                 if (pcb == NULL || pcb->inp_socket == NULL)
1550                                         break;
1551 #if defined(__DragonFly__) || (defined(__FreeBSD__) && __FreeBSD_version < 500034)
1552 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1553 #endif
1554                                 if (cmd->opcode == O_UID) {
1555                                         match =
1556                                           !socheckuid(pcb->inp_socket,
1557                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1558                                 } else  {
1559                                         match = groupmember(
1560                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1561                                             pcb->inp_socket->so_cred);
1562                                 }
1563                             }
1564                                 break;
1565
1566                         case O_RECV:
1567                                 match = iface_match(m->m_pkthdr.rcvif,
1568                                     (ipfw_insn_if *)cmd);
1569                                 break;
1570
1571                         case O_XMIT:
1572                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1573                                 break;
1574
1575                         case O_VIA:
1576                                 match = iface_match(oif ? oif :
1577                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1578                                 break;
1579
1580                         case O_MACADDR2:
1581                                 if (args->eh != NULL) { /* have MAC header */
1582                                         uint32_t *want = (uint32_t *)
1583                                                 ((ipfw_insn_mac *)cmd)->addr;
1584                                         uint32_t *mask = (uint32_t *)
1585                                                 ((ipfw_insn_mac *)cmd)->mask;
1586                                         uint32_t *hdr = (uint32_t *)args->eh;
1587
1588                                         match =
1589                                             ( want[0] == (hdr[0] & mask[0]) &&
1590                                               want[1] == (hdr[1] & mask[1]) &&
1591                                               want[2] == (hdr[2] & mask[2]) );
1592                                 }
1593                                 break;
1594
1595                         case O_MAC_TYPE:
1596                                 if (args->eh != NULL) {
1597                                         uint16_t t =
1598                                             ntohs(args->eh->ether_type);
1599                                         uint16_t *p =
1600                                             ((ipfw_insn_u16 *)cmd)->ports;
1601                                         int i;
1602
1603                                         for (i = cmdlen - 1; !match && i>0;
1604                                             i--, p += 2)
1605                                                 match = (t>=p[0] && t<=p[1]);
1606                                 }
1607                                 break;
1608
1609                         case O_FRAG:
1610                                 match = (hlen > 0 && offset != 0);
1611                                 break;
1612
1613                         case O_IN:      /* "out" is "not in" */
1614                                 match = (oif == NULL);
1615                                 break;
1616
1617                         case O_LAYER2:
1618                                 match = (args->eh != NULL);
1619                                 break;
1620
1621                         case O_PROTO:
1622                                 /*
1623                                  * We do not allow an arg of 0 so the
1624                                  * check of "proto" only suffices.
1625                                  */
1626                                 match = (proto == cmd->arg1);
1627                                 break;
1628
1629                         case O_IP_SRC:
1630                                 match = (hlen > 0 &&
1631                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1632                                     src_ip.s_addr);
1633                                 break;
1634
1635                         case O_IP_SRC_MASK:
1636                                 match = (hlen > 0 &&
1637                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1638                                      (src_ip.s_addr &
1639                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1640                                 break;
1641
1642                         case O_IP_SRC_ME:
1643                                 if (hlen > 0) {
1644                                         struct ifnet *tif;
1645
1646                                         INADDR_TO_IFP(src_ip, tif);
1647                                         match = (tif != NULL);
1648                                 }
1649                                 break;
1650
1651                         case O_IP_DST_SET:
1652                         case O_IP_SRC_SET:
1653                                 if (hlen > 0) {
1654                                         uint32_t *d = (uint32_t *)(cmd+1);
1655                                         uint32_t addr =
1656                                             cmd->opcode == O_IP_DST_SET ?
1657                                                 args->f_id.dst_ip :
1658                                                 args->f_id.src_ip;
1659
1660                                             if (addr < d[0])
1661                                                     break;
1662                                             addr -= d[0]; /* subtract base */
1663                                             match = (addr < cmd->arg1) &&
1664                                                 ( d[ 1 + (addr>>5)] &
1665                                                   (1<<(addr & 0x1f)) );
1666                                 }
1667                                 break;
1668
1669                         case O_IP_DST:
1670                                 match = (hlen > 0 &&
1671                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1672                                     dst_ip.s_addr);
1673                                 break;
1674
1675                         case O_IP_DST_MASK:
1676                                 match = (hlen > 0) &&
1677                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1678                                      (dst_ip.s_addr &
1679                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1680                                 break;
1681
1682                         case O_IP_DST_ME:
1683                                 if (hlen > 0) {
1684                                         struct ifnet *tif;
1685
1686                                         INADDR_TO_IFP(dst_ip, tif);
1687                                         match = (tif != NULL);
1688                                 }
1689                                 break;
1690
1691                         case O_IP_SRCPORT:
1692                         case O_IP_DSTPORT:
1693                                 /*
1694                                  * offset == 0 && proto != 0 is enough
1695                                  * to guarantee that we have an IPv4
1696                                  * packet with port info.
1697                                  */
1698                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1699                                     && offset == 0) {
1700                                         uint16_t x =
1701                                             (cmd->opcode == O_IP_SRCPORT) ?
1702                                                 src_port : dst_port ;
1703                                         uint16_t *p =
1704                                             ((ipfw_insn_u16 *)cmd)->ports;
1705                                         int i;
1706
1707                                         for (i = cmdlen - 1; !match && i>0;
1708                                             i--, p += 2)
1709                                                 match = (x>=p[0] && x<=p[1]);
1710                                 }
1711                                 break;
1712
1713                         case O_ICMPTYPE:
1714                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
1715                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
1716                                 break;
1717
1718                         case O_IPOPT:
1719                                 match = (hlen > 0 && ipopts_match(ip, cmd) );
1720                                 break;
1721
1722                         case O_IPVER:
1723                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1724                                 break;
1725
1726                         case O_IPTTL:
1727                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1728                                 break;
1729
1730                         case O_IPID:
1731                                 match = (hlen > 0 &&
1732                                     cmd->arg1 == ntohs(ip->ip_id));
1733                                 break;
1734
1735                         case O_IPLEN:
1736                                 match = (hlen > 0 && cmd->arg1 == ip_len);
1737                                 break;
1738
1739                         case O_IPPRECEDENCE:
1740                                 match = (hlen > 0 &&
1741                                     (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1742                                 break;
1743
1744                         case O_IPTOS:
1745                                 match = (hlen > 0 &&
1746                                     flags_match(cmd, ip->ip_tos));
1747                                 break;
1748
1749                         case O_TCPFLAGS:
1750                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1751                                     flags_match(cmd,
1752                                         L3HDR(struct tcphdr,ip)->th_flags));
1753                                 break;
1754
1755                         case O_TCPOPTS:
1756                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1757                                     tcpopts_match(ip, cmd));
1758                                 break;
1759
1760                         case O_TCPSEQ:
1761                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1762                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1763                                         L3HDR(struct tcphdr,ip)->th_seq);
1764                                 break;
1765
1766                         case O_TCPACK:
1767                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1768                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1769                                         L3HDR(struct tcphdr,ip)->th_ack);
1770                                 break;
1771
1772                         case O_TCPWIN:
1773                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1774                                     cmd->arg1 ==
1775                                         L3HDR(struct tcphdr,ip)->th_win);
1776                                 break;
1777
1778                         case O_ESTAB:
1779                                 /* reject packets which have SYN only */
1780                                 /* XXX should i also check for TH_ACK ? */
1781                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1782                                     (L3HDR(struct tcphdr,ip)->th_flags &
1783                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1784                                 break;
1785
1786                         case O_LOG:
1787                                 if (fw_verbose)
1788                                         ipfw_log(f, hlen, args->eh, m, oif);
1789                                 match = 1;
1790                                 break;
1791
1792                         case O_PROB:
1793                                 match = (krandom() <
1794                                         ((ipfw_insn_u32 *)cmd)->d[0]);
1795                                 break;
1796
1797                         /*
1798                          * The second set of opcodes represents 'actions',
1799                          * i.e. the terminal part of a rule once the packet
1800                          * matches all previous patterns.
1801                          * Typically there is only one action for each rule,
1802                          * and the opcode is stored at the end of the rule
1803                          * (but there are exceptions -- see below).
1804                          *
1805                          * In general, here we set retval and terminate the
1806                          * outer loop (would be a 'break 3' in some language,
1807                          * but we need to do a 'goto done').
1808                          *
1809                          * Exceptions:
1810                          * O_COUNT and O_SKIPTO actions:
1811                          *   instead of terminating, we jump to the next rule
1812                          *   ('goto next_rule', equivalent to a 'break 2'),
1813                          *   or to the SKIPTO target ('goto again' after
1814                          *   having set f, cmd and l), respectively.
1815                          *
1816                          * O_LIMIT and O_KEEP_STATE: these opcodes are
1817                          *   not real 'actions', and are stored right
1818                          *   before the 'action' part of the rule.
1819                          *   These opcodes try to install an entry in the
1820                          *   state tables; if successful, we continue with
1821                          *   the next opcode (match=1; break;), otherwise
1822                          *   the packet *   must be dropped
1823                          *   ('goto done' after setting retval);
1824                          *
1825                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1826                          *   cause a lookup of the state table, and a jump
1827                          *   to the 'action' part of the parent rule
1828                          *   ('goto check_body') if an entry is found, or
1829                          *   (CHECK_STATE only) a jump to the next rule if
1830                          *   the entry is not found ('goto next_rule').
1831                          *   The result of the lookup is cached to make
1832                          *   further instances of these opcodes are
1833                          *   effectively NOPs.
1834                          */
1835                         case O_LIMIT:
1836                         case O_KEEP_STATE:
1837                                 if (install_state(f,
1838                                     (ipfw_insn_limit *)cmd, args)) {
1839                                         retval = IP_FW_PORT_DENY_FLAG;
1840                                         goto done; /* error/limit violation */
1841                                 }
1842                                 match = 1;
1843                                 break;
1844
1845                         case O_PROBE_STATE:
1846                         case O_CHECK_STATE:
1847                                 /*
1848                                  * dynamic rules are checked at the first
1849                                  * keep-state or check-state occurrence,
1850                                  * with the result being stored in dyn_dir.
1851                                  * The compiler introduces a PROBE_STATE
1852                                  * instruction for us when we have a
1853                                  * KEEP_STATE (because PROBE_STATE needs
1854                                  * to be run first).
1855                                  */
1856                                 if (dyn_dir == MATCH_UNKNOWN &&
1857                                     (q = lookup_dyn_rule(&args->f_id,
1858                                      &dyn_dir, proto == IPPROTO_TCP ?
1859                                         L3HDR(struct tcphdr, ip) : NULL))
1860                                         != NULL) {
1861                                         /*
1862                                          * Found dynamic entry, update stats
1863                                          * and jump to the 'action' part of
1864                                          * the parent rule.
1865                                          */
1866                                         q->pcnt++;
1867                                         q->bcnt += ip_len;
1868                                         f = q->rule;
1869                                         cmd = ACTION_PTR(f);
1870                                         l = f->cmd_len - f->act_ofs;
1871                                         goto check_body;
1872                                 }
1873                                 /*
1874                                  * Dynamic entry not found. If CHECK_STATE,
1875                                  * skip to next rule, if PROBE_STATE just
1876                                  * ignore and continue with next opcode.
1877                                  */
1878                                 if (cmd->opcode == O_CHECK_STATE)
1879                                         goto next_rule;
1880                                 match = 1;
1881                                 break;
1882
1883                         case O_ACCEPT:
1884                                 retval = 0;     /* accept */
1885                                 goto done;
1886
1887                         case O_PIPE:
1888                         case O_QUEUE:
1889                                 args->rule = f; /* report matching rule */
1890                                 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
1891                                 goto done;
1892
1893                         case O_DIVERT:
1894                         case O_TEE:
1895                                 if (args->eh) /* not on layer 2 */
1896                                         break;
1897
1898                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
1899                                                  sizeof(uint16_t), MB_DONTWAIT);
1900                                 if (mtag == NULL) {
1901                                         retval = IP_FW_PORT_DENY_FLAG;
1902                                         goto done;
1903                                 }
1904                                 *(uint16_t *)m_tag_data(mtag) = f->rulenum;
1905                                 m_tag_prepend(m, mtag);
1906                                 retval = (cmd->opcode == O_DIVERT) ?
1907                                     cmd->arg1 :
1908                                     cmd->arg1 | IP_FW_PORT_TEE_FLAG;
1909                                 goto done;
1910
1911                         case O_COUNT:
1912                         case O_SKIPTO:
1913                                 f->pcnt++;      /* update stats */
1914                                 f->bcnt += ip_len;
1915                                 f->timestamp = time_second;
1916                                 if (cmd->opcode == O_COUNT)
1917                                         goto next_rule;
1918                                 /* handle skipto */
1919                                 if (f->next_rule == NULL)
1920                                         lookup_next_rule(f);
1921                                 f = f->next_rule;
1922                                 goto again;
1923
1924                         case O_REJECT:
1925                                 /*
1926                                  * Drop the packet and send a reject notice
1927                                  * if the packet is not ICMP (or is an ICMP
1928                                  * query), and it is not multicast/broadcast.
1929                                  */
1930                                 if (hlen > 0 &&
1931                                     (proto != IPPROTO_ICMP ||
1932                                      is_icmp_query(ip)) &&
1933                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
1934                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
1935                                         send_reject(args, cmd->arg1,
1936                                             offset,ip_len);
1937                                         m = args->m;
1938                                 }
1939                                 /* FALLTHROUGH */
1940                         case O_DENY:
1941                                 retval = IP_FW_PORT_DENY_FLAG;
1942                                 goto done;
1943
1944                         case O_FORWARD_IP:
1945                                 if (args->eh)   /* not valid on layer2 pkts */
1946                                         break;
1947                                 if (!q || dyn_dir == MATCH_FORWARD)
1948                                         args->next_hop =
1949                                             &((ipfw_insn_sa *)cmd)->sa;
1950                                 retval = 0;
1951                                 goto done;
1952
1953                         default:
1954                                 panic("-- unknown opcode %d\n", cmd->opcode);
1955                         } /* end of switch() on opcodes */
1956
1957                         if (cmd->len & F_NOT)
1958                                 match = !match;
1959
1960                         if (match) {
1961                                 if (cmd->len & F_OR)
1962                                         skip_or = 1;
1963                         } else {
1964                                 if (!(cmd->len & F_OR)) /* not an OR block, */
1965                                         break;          /* try next rule    */
1966                         }
1967
1968                 }       /* end of inner for, scan opcodes */
1969
1970 next_rule:;             /* try next rule                */
1971
1972         }               /* end of outer for, scan rules */
1973         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
1974         return(IP_FW_PORT_DENY_FLAG);
1975
1976 done:
1977         /* Update statistics */
1978         f->pcnt++;
1979         f->bcnt += ip_len;
1980         f->timestamp = time_second;
1981         return retval;
1982
1983 pullup_failed:
1984         if (fw_verbose)
1985                 kprintf("pullup failed\n");
1986         return(IP_FW_PORT_DENY_FLAG);
1987 }
1988
1989 static void
1990 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
1991 {
1992         struct m_tag *mtag;
1993         struct dn_pkt *pkt;
1994         ipfw_insn *cmd;
1995         const struct ipfw_flow_id *id;
1996         struct dn_flow_id *fid;
1997
1998         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
1999         if (mtag == NULL) {
2000                 m_freem(m);
2001                 return;
2002         }
2003         m_tag_prepend(m, mtag);
2004
2005         pkt = m_tag_data(mtag);
2006         bzero(pkt, sizeof(*pkt));
2007
2008         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2009         if (cmd->opcode == O_LOG)
2010                 cmd += F_LEN(cmd);
2011         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2012                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2013
2014         pkt->dn_m = m;
2015         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2016         pkt->ifp = fwa->oif;
2017         pkt->cpuid = mycpu->gd_cpuid;
2018         pkt->pipe_nr = pipe_nr;
2019
2020         id = &fwa->f_id;
2021         fid = &pkt->id;
2022         fid->fid_dst_ip = id->dst_ip;
2023         fid->fid_src_ip = id->src_ip;
2024         fid->fid_dst_port = id->dst_port;
2025         fid->fid_src_port = id->src_port;
2026         fid->fid_proto = id->proto;
2027         fid->fid_flags = id->flags;
2028
2029         ipfw_ref_rule(fwa->rule);
2030         pkt->dn_priv = fwa->rule;
2031         pkt->dn_unref_priv = ipfw_unref_rule;
2032
2033         if (cmd->opcode == O_PIPE)
2034                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2035
2036         if (dir == DN_TO_IP_OUT) {
2037                 /*
2038                  * We need to copy *ro because for ICMP pkts (and maybe
2039                  * others) the caller passed a pointer into the stack;
2040                  * dst might also be a pointer into *ro so it needs to
2041                  * be updated.
2042                  */
2043                 pkt->ro = *(fwa->ro);
2044                 if (fwa->ro->ro_rt)
2045                         fwa->ro->ro_rt->rt_refcnt++;
2046                 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) {
2047                         /* 'dst' points into 'ro' */
2048                         fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst);
2049                 }
2050                 pkt->dn_dst = fwa->dst;
2051                 pkt->flags = fwa->flags;
2052         }
2053
2054         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2055         ip_dn_queue(m);
2056 }
2057
2058 /*
2059  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2060  * These will be reconstructed on the fly as packets are matched.
2061  * Must be called at splimp().
2062  */
2063 static void
2064 flush_rule_ptrs(void)
2065 {
2066         struct ip_fw *rule;
2067
2068         for (rule = layer3_chain; rule; rule = rule->next)
2069                 rule->next_rule = NULL;
2070 }
2071
2072 static __inline void
2073 ipfw_inc_static_count(struct ip_fw *rule)
2074 {
2075         static_count++;
2076         static_ioc_len += IOC_RULESIZE(rule);
2077 }
2078
2079 static __inline void
2080 ipfw_dec_static_count(struct ip_fw *rule)
2081 {
2082         int l = IOC_RULESIZE(rule);
2083
2084         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2085         static_count--;
2086
2087         KASSERT(static_ioc_len >= l,
2088                 ("invalid static len %u\n", static_ioc_len));
2089         static_ioc_len -= l;
2090 }
2091
2092 static struct ip_fw *
2093 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule)
2094 {
2095         struct ip_fw *rule;
2096
2097         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2098
2099         rule->act_ofs = ioc_rule->act_ofs;
2100         rule->cmd_len = ioc_rule->cmd_len;
2101         rule->rulenum = ioc_rule->rulenum;
2102         rule->set = ioc_rule->set;
2103         rule->usr_flags = ioc_rule->usr_flags;
2104
2105         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2106
2107         rule->refcnt = 1;
2108
2109         return rule;
2110 }
2111
2112 /*
2113  * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2114  * possibly create a rule number and add the rule to the list.
2115  * Update the rule_number in the input struct so the caller knows it as well.
2116  */
2117 static int
2118 ipfw_add_rule(struct ip_fw **head, struct ipfw_ioc_rule *ioc_rule)
2119 {
2120         struct ip_fw *rule, *f, *prev;
2121
2122         KKASSERT(*head != NULL);
2123
2124         rule = ipfw_create_rule(ioc_rule);
2125
2126         crit_enter();
2127
2128         /*
2129          * If rulenum is 0, find highest numbered rule before the
2130          * default rule, and add autoinc_step
2131          */
2132         if (autoinc_step < 1)
2133                 autoinc_step = 1;
2134         else if (autoinc_step > 1000)
2135                 autoinc_step = 1000;
2136         if (rule->rulenum == 0) {
2137                 /*
2138                  * locate the highest numbered rule before default
2139                  */
2140                 for (f = *head; f; f = f->next) {
2141                         if (f->rulenum == IPFW_DEFAULT_RULE)
2142                                 break;
2143                         rule->rulenum = f->rulenum;
2144                 }
2145                 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2146                         rule->rulenum += autoinc_step;
2147                 ioc_rule->rulenum = rule->rulenum;
2148         }
2149
2150         /*
2151          * Now insert the new rule in the right place in the sorted list.
2152          */
2153         for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2154                 if (f->rulenum > rule->rulenum) { /* found the location */
2155                         if (prev) {
2156                                 rule->next = f;
2157                                 prev->next = rule;
2158                         } else { /* head insert */
2159                                 rule->next = *head;
2160                                 *head = rule;
2161                         }
2162                         break;
2163                 }
2164         }
2165
2166         flush_rule_ptrs();
2167         ipfw_inc_static_count(rule);
2168
2169         crit_exit();
2170
2171         DEB(kprintf("++ installed rule %d, static count now %d\n",
2172                 rule->rulenum, static_count);)
2173         return (0);
2174 }
2175
2176 /**
2177  * Free storage associated with a static rule (including derived
2178  * dynamic rules).
2179  * The caller is in charge of clearing rule pointers to avoid
2180  * dangling pointers.
2181  * @return a pointer to the next entry.
2182  * Arguments are not checked, so they better be correct.
2183  * Must be called at splimp().
2184  */
2185 static struct ip_fw *
2186 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2187 {
2188         struct ip_fw *n;
2189
2190         n = rule->next;
2191         remove_dyn_rule(rule, NULL /* force removal */);
2192         if (prev == NULL)
2193                 *head = n;
2194         else
2195                 prev->next = n;
2196         ipfw_dec_static_count(rule);
2197
2198         /* Mark the rule as invalid */
2199         rule->rule_flags |= IPFW_RULE_F_INVALID;
2200         rule->next_rule = NULL;
2201
2202         /* Try to free this rule */
2203         ipfw_free_rule(rule);
2204
2205         return n;
2206 }
2207
2208 /*
2209  * Deletes all rules from a chain (including the default rule
2210  * if the second argument is set).
2211  * Must be called at splimp().
2212  */
2213 static void
2214 free_chain(struct ip_fw **chain, int kill_default)
2215 {
2216         struct ip_fw *rule;
2217
2218         flush_rule_ptrs(); /* more efficient to do outside the loop */
2219
2220         while ( (rule = *chain) != NULL &&
2221             (kill_default || rule->rulenum != IPFW_DEFAULT_RULE) )
2222                 delete_rule(chain, NULL, rule);
2223
2224         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2225
2226         if (kill_default) {
2227                 ip_fw_default_rule = NULL;      /* Reset default rule */
2228
2229                 if (ipfw_dyn_v != NULL) {
2230                         /*
2231                          * Free dynamic rules(state) hash table
2232                          */
2233                         kfree(ipfw_dyn_v, M_IPFW);
2234                         ipfw_dyn_v = NULL;
2235                 }
2236
2237                 KASSERT(static_count == 0,
2238                         ("%u static rules remains\n", static_count));
2239                 KASSERT(static_ioc_len == 0,
2240                         ("%u bytes of static rules remains\n", static_ioc_len));
2241         } else {
2242                 KASSERT(static_count == 1,
2243                         ("%u static rules remains\n", static_count));
2244                 KASSERT(static_ioc_len == IOC_RULESIZE(ip_fw_default_rule),
2245                         ("%u bytes of static rules remains, should be %u\n",
2246                          static_ioc_len, IOC_RULESIZE(ip_fw_default_rule)));
2247         }
2248 }
2249
2250 /**
2251  * Remove all rules with given number, and also do set manipulation.
2252  *
2253  * The argument is an uint32_t. The low 16 bit are the rule or set number,
2254  * the next 8 bits are the new set, the top 8 bits are the command:
2255  *
2256  *      0       delete rules with given number
2257  *      1       delete rules with given set number
2258  *      2       move rules with given number to new set
2259  *      3       move rules with given set number to new set
2260  *      4       swap sets with given numbers
2261  */
2262 static int
2263 del_entry(struct ip_fw **chain, uint32_t arg)
2264 {
2265         struct ip_fw *prev, *rule;
2266         uint16_t rulenum;
2267         uint8_t cmd, new_set;
2268
2269         rulenum = arg & 0xffff;
2270         cmd = (arg >> 24) & 0xff;
2271         new_set = (arg >> 16) & 0xff;
2272
2273         if (cmd > 4)
2274                 return EINVAL;
2275         if (new_set > 30)
2276                 return EINVAL;
2277         if (cmd == 0 || cmd == 2) {
2278                 if (rulenum == IPFW_DEFAULT_RULE)
2279                         return EINVAL;
2280         } else {
2281                 if (rulenum > 30)
2282                         return EINVAL;
2283         }
2284
2285         switch (cmd) {
2286         case 0: /* delete rules with given number */
2287                 /*
2288                  * locate first rule to delete
2289                  */
2290                 for (prev = NULL, rule = *chain;
2291                     rule && rule->rulenum < rulenum;
2292                      prev = rule, rule = rule->next)
2293                         ;
2294                 if (rule->rulenum != rulenum)
2295                         return EINVAL;
2296
2297                 crit_enter(); /* no access to rules while removing */
2298                 /*
2299                  * flush pointers outside the loop, then delete all matching
2300                  * rules. prev remains the same throughout the cycle.
2301                  */
2302                 flush_rule_ptrs();
2303                 while (rule && rule->rulenum == rulenum)
2304                         rule = delete_rule(chain, prev, rule);
2305                 crit_exit();
2306                 break;
2307
2308         case 1: /* delete all rules with given set number */
2309                 crit_enter();
2310                 flush_rule_ptrs();
2311                 for (prev = NULL, rule = *chain; rule ; )
2312                         if (rule->set == rulenum)
2313                                 rule = delete_rule(chain, prev, rule);
2314                         else {
2315                                 prev = rule;
2316                                 rule = rule->next;
2317                         }
2318                 crit_exit();
2319                 break;
2320
2321         case 2: /* move rules with given number to new set */
2322                 crit_enter();
2323                 for (rule = *chain; rule ; rule = rule->next)
2324                         if (rule->rulenum == rulenum)
2325                                 rule->set = new_set;
2326                 crit_exit();
2327                 break;
2328
2329         case 3: /* move rules with given set number to new set */
2330                 crit_enter();
2331                 for (rule = *chain; rule ; rule = rule->next)
2332                         if (rule->set == rulenum)
2333                                 rule->set = new_set;
2334                 crit_exit();
2335                 break;
2336
2337         case 4: /* swap two sets */
2338                 crit_enter();
2339                 for (rule = *chain; rule ; rule = rule->next)
2340                         if (rule->set == rulenum)
2341                                 rule->set = new_set;
2342                         else if (rule->set == new_set)
2343                                 rule->set = rulenum;
2344                 crit_exit();
2345                 break;
2346         }
2347         return 0;
2348 }
2349
2350 /*
2351  * Clear counters for a specific rule.
2352  */
2353 static void
2354 clear_counters(struct ip_fw *rule, int log_only)
2355 {
2356         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2357
2358         if (log_only == 0) {
2359                 rule->bcnt = rule->pcnt = 0;
2360                 rule->timestamp = 0;
2361         }
2362         if (l->o.opcode == O_LOG)
2363                 l->log_left = l->max_log;
2364 }
2365
2366 /**
2367  * Reset some or all counters on firewall rules.
2368  * @arg frwl is null to clear all entries, or contains a specific
2369  * rule number.
2370  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2371  */
2372 static int
2373 zero_entry(int rulenum, int log_only)
2374 {
2375         struct ip_fw *rule;
2376         char *msg;
2377
2378         if (rulenum == 0) {
2379                 crit_enter();
2380                 norule_counter = 0;
2381                 for (rule = layer3_chain; rule; rule = rule->next)
2382                         clear_counters(rule, log_only);
2383                 crit_exit();
2384                 msg = log_only ? "ipfw: All logging counts reset.\n" :
2385                                 "ipfw: Accounting cleared.\n";
2386         } else {
2387                 int cleared = 0;
2388                 /*
2389                  * We can have multiple rules with the same number, so we
2390                  * need to clear them all.
2391                  */
2392                 for (rule = layer3_chain; rule; rule = rule->next)
2393                         if (rule->rulenum == rulenum) {
2394                                 crit_enter();
2395                                 while (rule && rule->rulenum == rulenum) {
2396                                         clear_counters(rule, log_only);
2397                                         rule = rule->next;
2398                                 }
2399                                 crit_exit();
2400                                 cleared = 1;
2401                                 break;
2402                         }
2403                 if (!cleared)   /* we did not find any matching rules */
2404                         return (EINVAL);
2405                 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2406                                 "ipfw: Entry %d cleared.\n";
2407         }
2408         if (fw_verbose)
2409                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2410         return (0);
2411 }
2412
2413 /*
2414  * Check validity of the structure before insert.
2415  * Fortunately rules are simple, so this mostly need to check rule sizes.
2416  */
2417 static int
2418 ipfw_ctl_check_rule(struct ipfw_ioc_rule *rule, int size)
2419 {
2420         int l, cmdlen = 0;
2421         int have_action = 0;
2422         ipfw_insn *cmd;
2423
2424         /* Check for valid size */
2425         if (size < sizeof(*rule)) {
2426                 kprintf("ipfw: rule too short\n");
2427                 return EINVAL;
2428         }
2429         l = IOC_RULESIZE(rule);
2430         if (l != size) {
2431                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
2432                 return EINVAL;
2433         }
2434
2435         /*
2436          * Now go for the individual checks. Very simple ones, basically only
2437          * instruction sizes.
2438          */
2439         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
2440              l -= cmdlen, cmd += cmdlen) {
2441                 cmdlen = F_LEN(cmd);
2442                 if (cmdlen > l) {
2443                         kprintf("ipfw: opcode %d size truncated\n",
2444                             cmd->opcode);
2445                         return EINVAL;
2446                 }
2447                 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
2448                 switch (cmd->opcode) {
2449                 case O_NOP:
2450                 case O_PROBE_STATE:
2451                 case O_KEEP_STATE:
2452                 case O_PROTO:
2453                 case O_IP_SRC_ME:
2454                 case O_IP_DST_ME:
2455                 case O_LAYER2:
2456                 case O_IN:
2457                 case O_FRAG:
2458                 case O_IPOPT:
2459                 case O_IPLEN:
2460                 case O_IPID:
2461                 case O_IPTOS:
2462                 case O_IPPRECEDENCE:
2463                 case O_IPTTL:
2464                 case O_IPVER:
2465                 case O_TCPWIN:
2466                 case O_TCPFLAGS:
2467                 case O_TCPOPTS:
2468                 case O_ESTAB:
2469                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2470                                 goto bad_size;
2471                         break;
2472
2473                 case O_UID:
2474                 case O_GID:
2475                 case O_IP_SRC:
2476                 case O_IP_DST:
2477                 case O_TCPSEQ:
2478                 case O_TCPACK:
2479                 case O_PROB:
2480                 case O_ICMPTYPE:
2481                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2482                                 goto bad_size;
2483                         break;
2484
2485                 case O_LIMIT:
2486                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2487                                 goto bad_size;
2488                         break;
2489
2490                 case O_LOG:
2491                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2492                                 goto bad_size;
2493
2494                         ((ipfw_insn_log *)cmd)->log_left =
2495                             ((ipfw_insn_log *)cmd)->max_log;
2496
2497                         break;
2498
2499                 case O_IP_SRC_MASK:
2500                 case O_IP_DST_MASK:
2501                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2502                                 goto bad_size;
2503                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2504                                 kprintf("ipfw: opcode %d, useless rule\n",
2505                                         cmd->opcode);
2506                                 return EINVAL;
2507                         }
2508                         break;
2509
2510                 case O_IP_SRC_SET:
2511                 case O_IP_DST_SET:
2512                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2513                                 kprintf("ipfw: invalid set size %d\n",
2514                                         cmd->arg1);
2515                                 return EINVAL;
2516                         }
2517                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2518                             (cmd->arg1+31)/32 )
2519                                 goto bad_size;
2520                         break;
2521
2522                 case O_MACADDR2:
2523                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2524                                 goto bad_size;
2525                         break;
2526
2527                 case O_MAC_TYPE:
2528                 case O_IP_SRCPORT:
2529                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2530                         if (cmdlen < 2 || cmdlen > 31)
2531                                 goto bad_size;
2532                         break;
2533
2534                 case O_RECV:
2535                 case O_XMIT:
2536                 case O_VIA:
2537                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2538                                 goto bad_size;
2539                         break;
2540
2541                 case O_PIPE:
2542                 case O_QUEUE:
2543                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2544                                 goto bad_size;
2545                         goto check_action;
2546
2547                 case O_FORWARD_IP:
2548                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2549                                 goto bad_size;
2550                         goto check_action;
2551
2552                 case O_FORWARD_MAC: /* XXX not implemented yet */
2553                 case O_CHECK_STATE:
2554                 case O_COUNT:
2555                 case O_ACCEPT:
2556                 case O_DENY:
2557                 case O_REJECT:
2558                 case O_SKIPTO:
2559                 case O_DIVERT:
2560                 case O_TEE:
2561                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2562                                 goto bad_size;
2563 check_action:
2564                         if (have_action) {
2565                                 kprintf("ipfw: opcode %d, multiple actions"
2566                                         " not allowed\n",
2567                                         cmd->opcode);
2568                                 return EINVAL;
2569                         }
2570                         have_action = 1;
2571                         if (l != cmdlen) {
2572                                 kprintf("ipfw: opcode %d, action must be"
2573                                         " last opcode\n",
2574                                         cmd->opcode);
2575                                 return EINVAL;
2576                         }
2577                         break;
2578                 default:
2579                         kprintf("ipfw: opcode %d, unknown opcode\n",
2580                                 cmd->opcode);
2581                         return EINVAL;
2582                 }
2583         }
2584         if (have_action == 0) {
2585                 kprintf("ipfw: missing action\n");
2586                 return EINVAL;
2587         }
2588         return 0;
2589
2590 bad_size:
2591         kprintf("ipfw: opcode %d size %d wrong\n",
2592                 cmd->opcode, cmdlen);
2593         return EINVAL;
2594 }
2595
2596 static int
2597 ipfw_ctl_add_rule(struct sockopt *sopt)
2598 {
2599         struct ipfw_ioc_rule *ioc_rule;
2600         uint32_t rule_buf[IPFW_RULE_SIZE_MAX];
2601         size_t size;
2602         int error;
2603
2604         ioc_rule = (struct ipfw_ioc_rule *)rule_buf;
2605         error = sooptcopyin(sopt, ioc_rule, sizeof(rule_buf),
2606                             sizeof(*ioc_rule));
2607         if (error)
2608                 return error;
2609
2610         size = sopt->sopt_valsize;
2611         error = ipfw_ctl_check_rule(ioc_rule, size);
2612         if (error)
2613                 return error;
2614
2615         error = ipfw_add_rule(&layer3_chain, ioc_rule);
2616         if (error)
2617                 return error;
2618
2619         if (sopt->sopt_dir == SOPT_GET)
2620                 error = sooptcopyout(sopt, ioc_rule, IOC_RULESIZE(ioc_rule));
2621         return error;
2622 }
2623
2624 static void *
2625 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
2626 {
2627         ioc_rule->act_ofs = rule->act_ofs;
2628         ioc_rule->cmd_len = rule->cmd_len;
2629         ioc_rule->rulenum = rule->rulenum;
2630         ioc_rule->set = rule->set;
2631         ioc_rule->usr_flags = rule->usr_flags;
2632
2633         ioc_rule->set_disable = set_disable;
2634         ioc_rule->static_count = static_count;
2635         ioc_rule->static_len = static_ioc_len;
2636
2637         ioc_rule->pcnt = rule->pcnt;
2638         ioc_rule->bcnt = rule->bcnt;
2639         ioc_rule->timestamp = rule->timestamp;
2640
2641         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
2642
2643         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
2644 }
2645
2646 static void
2647 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
2648                 struct ipfw_ioc_state *ioc_state)
2649 {
2650         const struct ipfw_flow_id *id;
2651         struct ipfw_ioc_flowid *ioc_id;
2652
2653         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
2654                             0 : dyn_rule->expire - time_second;
2655         ioc_state->pcnt = dyn_rule->pcnt;
2656         ioc_state->bcnt = dyn_rule->bcnt;
2657
2658         ioc_state->dyn_type = dyn_rule->dyn_type;
2659         ioc_state->count = dyn_rule->count;
2660
2661         ioc_state->rulenum = dyn_rule->rule->rulenum;
2662
2663         id = &dyn_rule->id;
2664         ioc_id = &ioc_state->id;
2665
2666         ioc_id->type = ETHERTYPE_IP;
2667         ioc_id->u.ip.dst_ip = id->dst_ip;
2668         ioc_id->u.ip.src_ip = id->src_ip;
2669         ioc_id->u.ip.dst_port = id->dst_port;
2670         ioc_id->u.ip.src_port = id->src_port;
2671         ioc_id->u.ip.proto = id->proto;
2672 }
2673
2674 static int
2675 ipfw_ctl_get_rules(struct sockopt *sopt)
2676 {
2677         struct ip_fw *rule;
2678         void *buf, *bp;
2679         size_t size;
2680         int error;
2681
2682         /*
2683          * pass up a copy of the current rules. Static rules
2684          * come first (the last of which has number IPFW_DEFAULT_RULE),
2685          * followed by a possibly empty list of dynamic rule.
2686          */
2687         crit_enter();
2688
2689         size = static_ioc_len;  /* size of static rules */
2690         if (ipfw_dyn_v)         /* add size of dyn.rules */
2691                 size += (dyn_count * sizeof(struct ipfw_ioc_state));
2692
2693         /*
2694          * XXX todo: if the user passes a short length just to know
2695          * how much room is needed, do not bother filling up the
2696          * buffer, just jump to the sooptcopyout.
2697          */
2698         bp = buf = kmalloc(size, M_TEMP, M_WAITOK | M_ZERO);
2699
2700         for (rule = layer3_chain; rule; rule = rule->next)
2701                 bp = ipfw_copy_rule(rule, bp);
2702
2703         if (ipfw_dyn_v) {
2704                 struct ipfw_ioc_state *ioc_state;
2705                 int i;
2706
2707                 ioc_state = bp;
2708                 for (i = 0; i < curr_dyn_buckets; i++) {
2709                         ipfw_dyn_rule *p;
2710
2711                         for (p = ipfw_dyn_v[i]; p != NULL;
2712                              p = p->next, ioc_state++)
2713                                 ipfw_copy_state(p, ioc_state);
2714                 }
2715         }
2716
2717         crit_exit();
2718
2719         error = sooptcopyout(sopt, buf, size);
2720         kfree(buf, M_TEMP);
2721         return error;
2722 }
2723
2724 /**
2725  * {set|get}sockopt parser.
2726  */
2727 static int
2728 ipfw_ctl(struct sockopt *sopt)
2729 {
2730         int error, rulenum;
2731         uint32_t masks[2];
2732         size_t size;
2733
2734         /*
2735          * Disallow modifications in really-really secure mode, but still allow
2736          * the logging counters to be reset.
2737          */
2738         if (sopt->sopt_name == IP_FW_ADD ||
2739             (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
2740                 if (securelevel >= 3)
2741                         return EPERM;
2742         }
2743
2744         error = 0;
2745
2746         switch (sopt->sopt_name) {
2747         case IP_FW_GET:
2748                 error = ipfw_ctl_get_rules(sopt);
2749                 break;
2750
2751         case IP_FW_FLUSH:
2752                 /*
2753                  * Normally we cannot release the lock on each iteration.
2754                  * We could do it here only because we start from the head all
2755                  * the times so there is no risk of missing some entries.
2756                  * On the other hand, the risk is that we end up with
2757                  * a very inconsistent ruleset, so better keep the lock
2758                  * around the whole cycle.
2759                  *
2760                  * XXX this code can be improved by resetting the head of
2761                  * the list to point to the default rule, and then freeing
2762                  * the old list without the need for a lock.
2763                  */
2764
2765                 crit_enter();
2766                 free_chain(&layer3_chain, 0 /* keep default rule */);
2767                 crit_exit();
2768                 break;
2769
2770         case IP_FW_ADD:
2771                 error = ipfw_ctl_add_rule(sopt);
2772                 break;
2773
2774         case IP_FW_DEL:
2775                 /*
2776                  * IP_FW_DEL is used for deleting single rules or sets,
2777                  * and (ab)used to atomically manipulate sets. Argument size
2778                  * is used to distinguish between the two:
2779                  *    sizeof(uint32_t)
2780                  *      delete single rule or set of rules,
2781                  *      or reassign rules (or sets) to a different set.
2782                  *    2*sizeof(uint32_t)
2783                  *      atomic disable/enable sets.
2784                  *      first uint32_t contains sets to be disabled,
2785                  *      second uint32_t contains sets to be enabled.
2786                  */
2787                 error = sooptcopyin(sopt, masks,
2788                         sizeof(masks), sizeof(masks[0]));
2789                 if (error)
2790                         break;
2791
2792                 size = sopt->sopt_valsize;
2793                 if (size == sizeof(masks[0])) {
2794                         /*
2795                          * Delete or reassign static rule
2796                          */
2797                         error = del_entry(&layer3_chain, masks[0]);
2798                 } else if (size == sizeof(masks)) {
2799                         /*
2800                          * Set enable/disable
2801                          */
2802                         crit_enter();
2803
2804                         set_disable =
2805                             (set_disable | masks[0]) & ~masks[1] &
2806                             ~(1 << 31); /* set 31 always enabled */
2807
2808                         crit_exit();
2809                 } else {
2810                         error = EINVAL;
2811                 }
2812                 break;
2813
2814         case IP_FW_ZERO:
2815         case IP_FW_RESETLOG: /* argument is an int, the rule number */
2816                 rulenum=0;
2817
2818                 if (sopt->sopt_val != 0) {
2819                     error = sooptcopyin(sopt, &rulenum,
2820                             sizeof(int), sizeof(int));
2821                     if (error)
2822                         break;
2823                 }
2824                 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2825                 break;
2826
2827         default:
2828                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
2829                 error = EINVAL;
2830         }
2831
2832         return error;
2833 }
2834
2835 /*
2836  * This procedure is only used to handle keepalives. It is invoked
2837  * every dyn_keepalive_period
2838  */
2839 static void
2840 ipfw_tick(void * __unused unused)
2841 {
2842         int i;
2843         ipfw_dyn_rule *q;
2844
2845         if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2846                 goto done;
2847
2848         crit_enter();
2849         for (i = 0 ; i < curr_dyn_buckets ; i++) {
2850                 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
2851                         if (q->dyn_type == O_LIMIT_PARENT)
2852                                 continue;
2853                         if (q->id.proto != IPPROTO_TCP)
2854                                 continue;
2855                         if ( (q->state & BOTH_SYN) != BOTH_SYN)
2856                                 continue;
2857                         if (TIME_LEQ( time_second+dyn_keepalive_interval,
2858                             q->expire))
2859                                 continue;       /* too early */
2860                         if (TIME_LEQ(q->expire, time_second))
2861                                 continue;       /* too late, rule expired */
2862
2863                         send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
2864                         send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
2865                 }
2866         }
2867         crit_exit();
2868 done:
2869         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
2870                       ipfw_tick, NULL);
2871 }
2872
2873 static void
2874 ipfw_init_default_rule(struct ip_fw **head)
2875 {
2876         struct ip_fw *def_rule;
2877
2878         KKASSERT(*head == NULL);
2879
2880         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
2881
2882         def_rule->act_ofs = 0;
2883         def_rule->rulenum = IPFW_DEFAULT_RULE;
2884         def_rule->cmd_len = 1;
2885         def_rule->set = 31;
2886
2887         def_rule->cmd[0].len = 1;
2888 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2889         def_rule->cmd[0].opcode = O_ACCEPT;
2890 #else
2891         def_rule->cmd[0].opcode = O_DENY;
2892 #endif
2893
2894         def_rule->refcnt = 1;
2895
2896         *head = def_rule;
2897         ipfw_inc_static_count(def_rule);
2898
2899         /* Install the default rule */
2900         ip_fw_default_rule = def_rule;
2901 }
2902
2903 static void
2904 ipfw_init(void)
2905 {
2906         ip_fw_chk_ptr = ipfw_chk;
2907         ip_fw_ctl_ptr = ipfw_ctl;
2908         ip_fw_dn_io_ptr = ipfw_dummynet_io;
2909
2910         layer3_chain = NULL;
2911         ipfw_init_default_rule(&layer3_chain);
2912
2913         kprintf("ipfw2 initialized, divert %s, "
2914                 "rule-based forwarding enabled, default to %s, logging ",
2915 #ifdef IPDIVERT
2916                 "enabled",
2917 #else
2918                 "disabled",
2919 #endif
2920                 ip_fw_default_rule->cmd[0].opcode == O_ACCEPT ?
2921                 "accept" : "deny");
2922
2923 #ifdef IPFIREWALL_VERBOSE
2924         fw_verbose = 1;
2925 #endif
2926 #ifdef IPFIREWALL_VERBOSE_LIMIT
2927         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2928 #endif
2929         if (fw_verbose == 0)
2930                 kprintf("disabled\n");
2931         else if (verbose_limit == 0)
2932                 kprintf("unlimited\n");
2933         else
2934                 kprintf("limited to %d packets/entry by default\n",
2935                     verbose_limit);
2936         callout_init(&ipfw_timeout_h);
2937         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
2938 }
2939
2940 static int
2941 ipfw_modevent(module_t mod, int type, void *unused)
2942 {
2943         int err = 0;
2944
2945         switch (type) {
2946         case MOD_LOAD:
2947                 crit_enter();
2948                 if (IPFW_LOADED) {
2949                         crit_exit();
2950                         kprintf("IP firewall already loaded\n");
2951                         err = EEXIST;
2952                 } else {
2953                         ipfw_init();
2954                         crit_exit();
2955                 }
2956                 break;
2957
2958         case MOD_UNLOAD:
2959 #ifndef KLD_MODULE
2960                 kprintf("ipfw statically compiled, cannot unload\n");
2961                 err = EBUSY;
2962 #else
2963                 if (ipfw_refcnt != 0) {
2964                         err = EBUSY;
2965                         break;
2966                 }
2967
2968                 crit_enter();
2969                 callout_stop(&ipfw_timeout_h);
2970                 ip_fw_chk_ptr = NULL;
2971                 ip_fw_ctl_ptr = NULL;
2972                 ip_fw_dn_io_ptr = NULL;
2973                 free_chain(&layer3_chain, 1 /* kill default rule */);
2974                 crit_exit();
2975                 kprintf("IP firewall unloaded\n");
2976 #endif
2977                 break;
2978         default:
2979                 break;
2980         }
2981         return err;
2982 }
2983
2984 static moduledata_t ipfwmod = {
2985         "ipfw",
2986         ipfw_modevent,
2987         0
2988 };
2989 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
2990 MODULE_VERSION(ipfw, 1);