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  */
  27
  28 /*
  29  * Implement IP packet firewall (new version)
  30  */
  31
  32 #include "opt_ipfw.h"
  33 #include "opt_inet.h"
  34 #ifndef INET
  35 #error IPFIREWALL requires INET.
  36 #endif /* INET */
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/malloc.h>
  41 #include <sys/mbuf.h>
  42 #include <sys/kernel.h>
  43 #include <sys/proc.h>
  44 #include <sys/socket.h>
  45 #include <sys/socketvar.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/syslog.h>
  48 #include <sys/ucred.h>
  49 #include <sys/in_cksum.h>
  50 #include <sys/lock.h>
  51
  52 #include <net/if.h>
  53 #include <net/route.h>
  54 #include <net/pfil.h>
  55 #include <net/dummynet/ip_dummynet.h>
  56
  57 #include <sys/thread2.h>
  58 #include <sys/mplock2.h>
  59 #include <net/netmsg2.h>
  60
  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 <netinet/tcp.h>
  69 #include <netinet/tcp_timer.h>
  70 #include <netinet/tcp_var.h>
  71 #include <netinet/tcpip.h>
  72 #include <netinet/udp.h>
  73 #include <netinet/udp_var.h>
  74 #include <netinet/ip_divert.h>
  75 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  76
  77 #include <net/ipfw/ip_fw2.h>
  78
  79 #ifdef IPFIREWALL_DEBUG
  80 #define DPRINTF(fmt, ...) \
  81 do { \
  82         if (fw_debug > 0) \
  83                 kprintf(fmt, __VA_ARGS__); \
  84 } while (0)
  85 #else
  86 #define DPRINTF(fmt, ...)       ((void)0)
  87 #endif
  88
  89 /*
  90  * Description about per-CPU rule duplication:
  91  *
  92  * Module loading/unloading and all ioctl operations are serialized
  93  * by netisr0, so we don't have any ordering or locking problems.
  94  *
  95  * Following graph shows how operation on per-CPU rule list is
  96  * performed [2 CPU case]:
  97  *
  98  *   CPU0                 CPU1
  99  *
 100  * netisr0 <------------------------------------+
 101  *  domsg                                       |
 102  *    |                                         |
 103  *    | netmsg                                  |
 104  *    |                                         |
 105  *    V                                         |
 106  *  ifnet0                                      |
 107  *    :                                         | netmsg
 108  *    :(delete/add...)                          |
 109  *    :                                         |
 110  *    :         netmsg                          |
 111  *  forwardmsg---------->ifnet1                 |
 112  *                          :                   |
 113  *                          :(delete/add...)    |
 114  *                          :                   |
 115  *                          :                   |
 116  *                        replymsg--------------+
 117  *
 118  *
 119  *
 120  *
 121  * Rules which will not create states (dyn rules) [2 CPU case]
 122  *
 123  *    CPU0               CPU1
 124  * layer3_chain       layer3_chain
 125  *     |                  |
 126  *     V                  V
 127  * +-------+ sibling  +-------+ sibling
 128  * | rule1 |--------->| rule1 |--------->NULL
 129  * +-------+          +-------+
 130  *     |                  |
 131  *     |next              |next
 132  *     V                  V
 133  * +-------+ sibling  +-------+ sibling
 134  * | rule2 |--------->| rule2 |--------->NULL
 135  * +-------+          +-------+
 136  *
 137  * ip_fw.sibling:
 138  * 1) Ease statistics calculation during IP_FW_GET.  We only need to
 139  *    iterate layer3_chain on CPU0; the current rule's duplication on
 140  *    the other CPUs could safely be read-only accessed by using
 141  *    ip_fw.sibling
 142  * 2) Accelerate rule insertion and deletion, e.g. rule insertion:
 143  *    a) In netisr0 (on CPU0) rule3 is determined to be inserted between
 144  *       rule1 and rule2.  To make this decision we need to iterate the
 145  *       layer3_chain on CPU0.  The netmsg, which is used to insert the
 146  *       rule, will contain rule1 on CPU0 as prev_rule and rule2 on CPU0
 147  *       as next_rule
 148  *    b) After the insertion on CPU0 is done, we will move on to CPU1.
 149  *       But instead of relocating the rule3's position on CPU1 by
 150  *       iterating the layer3_chain on CPU1, we set the netmsg's prev_rule
 151  *       to rule1->sibling and next_rule to rule2->sibling before the
 152  *       netmsg is forwarded to CPU1 from CPU0
 153  *
 154  *
 155  *
 156  * Rules which will create states (dyn rules) [2 CPU case]
 157  * (unnecessary parts are omitted; they are same as in the previous figure)
 158  *
 159  *   CPU0                       CPU1
 160  *
 161  * +-------+                  +-------+
 162  * | rule1 |                  | rule1 |
 163  * +-------+                  +-------+
 164  *   ^   |                      |   ^
 165  *   |   |stub              stub|   |
 166  *   |   |                      |   |
 167  *   |   +----+            +----+   |
 168  *   |        |            |        |
 169  *   |        V            V        |
 170  *   |    +--------------------+    |
 171  *   |    |     rule_stub      |    |
 172  *   |    | (read-only shared) |    |
 173  *   |    |                    |    |
 174  *   |    | back pointer array |    |
 175  *   |    | (indexed by cpuid) |    |
 176  *   |    |                    |    |
 177  *   +----|---------[0]        |    |
 178  *        |         [1]--------|----+
 179  *        |                    |
 180  *        +--------------------+
 181  *          ^            ^
 182  *          |            |
 183  *  ........|............|............
 184  *  :       |            |           :
 185  *  :       |stub        |stub       :
 186  *  :       |            |           :
 187  *  :  +---------+  +---------+      :
 188  *  :  | state1a |  | state1b | .... :
 189  *  :  +---------+  +---------+      :
 190  *  :                                :
 191  *  :           states table         :
 192  *  :            (shared)            :
 193  *  :      (protected by dyn_lock)   :
 194  *  ..................................
 195  *
 196  * [state1a and state1b are states created by rule1]
 197  *
 198  * ip_fw_stub:
 199  * This structure is introduced so that shared (locked) state table could
 200  * work with per-CPU (duplicated) static rules.  It mainly bridges states
 201  * and static rules and serves as static rule's place holder (a read-only
 202  * shared part of duplicated rules) from states point of view.
 203  *
 204  * IPFW_RULE_F_STATE (only for rules which create states):
 205  * o  During rule installation, this flag is turned on after rule's
 206  *    duplications reach all CPUs, to avoid at least following race:
 207  *    1) rule1 is duplicated on CPU0 and is not duplicated on CPU1 yet
 208  *    2) rule1 creates state1
 209  *    3) state1 is located on CPU1 by check-state
 210  *    But rule1 is not duplicated on CPU1 yet
 211  * o  During rule deletion, this flag is turned off before deleting states
 212  *    created by the rule and before deleting the rule itself, so no
 213  *    more states will be created by the to-be-deleted rule even when its
 214  *    duplication on certain CPUs are not eliminated yet.
 215  */
 216
 217 #define IPFW_AUTOINC_STEP_MIN   1
 218 #define IPFW_AUTOINC_STEP_MAX   1000
 219 #define IPFW_AUTOINC_STEP_DEF   100
 220
 221 #define IPFW_DEFAULT_RULE       65535   /* rulenum for the default rule */
 222 #define IPFW_DEFAULT_SET        31      /* set number for the default rule */
 223
 224 struct netmsg_ipfw {
 225         struct netmsg_base base;
 226         const struct ipfw_ioc_rule *ioc_rule;
 227         struct ip_fw    *next_rule;
 228         struct ip_fw    *prev_rule;
 229         struct ip_fw    *sibling;
 230         struct ip_fw_stub *stub;
 231 };
 232
 233 struct netmsg_del {
 234         struct netmsg_base base;
 235         struct ip_fw    *start_rule;
 236         struct ip_fw    *prev_rule;
 237         uint16_t        rulenum;
 238         uint8_t         from_set;
 239         uint8_t         to_set;
 240 };
 241
 242 struct netmsg_zent {
 243         struct netmsg_base base;
 244         struct ip_fw    *start_rule;
 245         uint16_t        rulenum;
 246         uint16_t        log_only;
 247 };
 248
 249 struct ipfw_context {
 250         struct ip_fw    *ipfw_layer3_chain;     /* list of rules for layer3 */
 251         struct ip_fw    *ipfw_default_rule;     /* default rule */
 252         uint64_t        ipfw_norule_counter;    /* counter for ipfw_log(NULL) */
 253
 254         /*
 255          * ipfw_set_disable contains one bit per set value (0..31).
 256          * If the bit is set, all rules with the corresponding set
 257          * are disabled.  Set IPDW_DEFAULT_SET is reserved for the
 258          * default rule and CANNOT be disabled.
 259          */
 260         uint32_t        ipfw_set_disable;
 261         uint32_t        ipfw_gen;               /* generation of rule list */
 262 };
 263
 264 static struct ipfw_context      *ipfw_ctx[MAXCPU];
 265
 266 #ifdef KLD_MODULE
 267 /*
 268  * Module can not be unloaded, if there are references to
 269  * certains rules of ipfw(4), e.g. dummynet(4)
 270  */
 271 static int ipfw_refcnt;
 272 #endif
 273
 274 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 275
 276 /*
 277  * Following two global variables are accessed and
 278  * updated only on CPU0
 279  */
 280 static uint32_t static_count;   /* # of static rules */
 281 static uint32_t static_ioc_len; /* bytes of static rules */
 282
 283 /*
 284  * If 1, then ipfw static rules are being flushed,
 285  * ipfw_chk() will skip to the default rule.
 286  */
 287 static int ipfw_flushing;
 288
 289 static int fw_verbose;
 290 static int verbose_limit;
 291
 292 static int fw_debug;
 293 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
 294
 295 static int      ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS);
 296 static int      ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS);
 297 static int      ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS);
 298 static int      ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS);
 299 static int      ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS);
 300
 301 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 302 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
 303     &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw");
 304 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
 305     &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
 306     "Rule number autincrement step");
 307 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 308     &fw_one_pass, 0,
 309     "Only do a single pass through ipfw when using dummynet(4)");
 310 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 311     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 312 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 313     &fw_verbose, 0, "Log matches to ipfw rules");
 314 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 315     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 316
 317 /*
 318  * Description of dynamic rules.
 319  *
 320  * Dynamic rules are stored in lists accessed through a hash table
 321  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 322  * be modified through the sysctl variable dyn_buckets which is
 323  * updated when the table becomes empty.
 324  *
 325  * XXX currently there is only one list, ipfw_dyn.
 326  *
 327  * When a packet is received, its address fields are first masked
 328  * with the mask defined for the rule, then hashed, then matched
 329  * against the entries in the corresponding list.
 330  * Dynamic rules can be used for different purposes:
 331  *  + stateful rules;
 332  *  + enforcing limits on the number of sessions;
 333  *  + in-kernel NAT (not implemented yet)
 334  *
 335  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 336  * measured in seconds and depending on the flags.
 337  *
 338  * The total number of dynamic rules is stored in dyn_count.
 339  * The max number of dynamic rules is dyn_max. When we reach
 340  * the maximum number of rules we do not create anymore. This is
 341  * done to avoid consuming too much memory, but also too much
 342  * time when searching on each packet (ideally, we should try instead
 343  * to put a limit on the length of the list on each bucket...).
 344  *
 345  * Each dynamic rule holds a pointer to the parent ipfw rule so
 346  * we know what action to perform. Dynamic rules are removed when
 347  * the parent rule is deleted. XXX we should make them survive.
 348  *
 349  * There are some limitations with dynamic rules -- we do not
 350  * obey the 'randomized match', and we do not do multiple
 351  * passes through the firewall. XXX check the latter!!!
 352  *
 353  * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
 354  * Only TCP state transition will change dynamic rule's state and ack
 355  * sequences, while all packets of one TCP connection only goes through
 356  * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
 357  * rule looking up.  The keep alive callout uses exclusive lockmgr lock
 358  * when it tries to find suitable dynamic rules to send keep alive, so
 359  * it will not see half updated state and ack sequences.  Though the expire
 360  * field updating looks racy for other protocols, the resolution (second)
 361  * of expire field makes this kind of race harmless.
 362  * XXX statistics' updating is _not_ MPsafe!!!
 363  * XXX once UDP output path is fixed, we could use lockless dynamic rule
 364  *     hash table
 365  */
 366 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 367 static uint32_t dyn_buckets = 256; /* must be power of 2 */
 368 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
 369 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
 370 static struct lock dyn_lock; /* dynamic rules' hash table lock */
 371
 372 static struct netmsg_base ipfw_timeout_netmsg; /* schedule ipfw timeout */
 373 static struct callout ipfw_timeout_h;
 374
 375 /*
 376  * Timeouts for various events in handing dynamic rules.
 377  */
 378 static uint32_t dyn_ack_lifetime = 300;
 379 static uint32_t dyn_syn_lifetime = 20;
 380 static uint32_t dyn_fin_lifetime = 1;
 381 static uint32_t dyn_rst_lifetime = 1;
 382 static uint32_t dyn_udp_lifetime = 10;
 383 static uint32_t dyn_short_lifetime = 5;
 384
 385 /*
 386  * Keepalives are sent if dyn_keepalive is set. They are sent every
 387  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 388  * seconds of lifetime of a rule.
 389  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 390  * than dyn_keepalive_period.
 391  */
 392
 393 static uint32_t dyn_keepalive_interval = 20;
 394 static uint32_t dyn_keepalive_period = 5;
 395 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
 396
 397 static uint32_t dyn_count;              /* # of dynamic rules */
 398 static uint32_t dyn_max = 4096;         /* max # of dynamic rules */
 399
 400 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
 401     &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
 402 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 403     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 404 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 405     &dyn_count, 0, "Number of dyn. rules");
 406 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 407     &dyn_max, 0, "Max number of dyn. rules");
 408 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 409     &static_count, 0, "Number of static rules");
 410 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 411     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 412 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 413     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 414 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
 415     CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
 416     "Lifetime of dyn. rules for fin");
 417 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
 418     CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
 419     "Lifetime of dyn. rules for rst");
 420 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 421     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 422 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 423     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 424 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 425     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 426
 427 static ip_fw_chk_t      ipfw_chk;
 428 static void             ipfw_tick(void *);
 429
 430 static __inline int
 431 ipfw_free_rule(struct ip_fw *rule)
 432 {
 433         KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d", mycpuid));
 434         KASSERT(rule->refcnt > 0, ("invalid refcnt %u", rule->refcnt));
 435         rule->refcnt--;
 436         if (rule->refcnt == 0) {
 437                 kfree(rule, M_IPFW);
 438                 return 1;
 439         }
 440         return 0;
 441 }
 442
 443 static void
 444 ipfw_unref_rule(void *priv)
 445 {
 446         ipfw_free_rule(priv);
 447 #ifdef KLD_MODULE
 448         atomic_subtract_int(&ipfw_refcnt, 1);
 449 #endif
 450 }
 451
 452 static __inline void
 453 ipfw_ref_rule(struct ip_fw *rule)
 454 {
 455         KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d", mycpuid));
 456 #ifdef KLD_MODULE
 457         atomic_add_int(&ipfw_refcnt, 1);
 458 #endif
 459         rule->refcnt++;
 460 }
 461
 462 /*
 463  * This macro maps an ip pointer into a layer3 header pointer of type T
 464  */
 465 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
 466
 467 static __inline int
 468 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 469 {
 470         int type = L3HDR(struct icmp,ip)->icmp_type;
 471
 472         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
 473 }
 474
 475 #define TT      ((1 << ICMP_ECHO) | \
 476                  (1 << ICMP_ROUTERSOLICIT) | \
 477                  (1 << ICMP_TSTAMP) | \
 478                  (1 << ICMP_IREQ) | \
 479                  (1 << ICMP_MASKREQ))
 480
 481 static int
 482 is_icmp_query(struct ip *ip)
 483 {
 484         int type = L3HDR(struct icmp, ip)->icmp_type;
 485
 486         return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
 487 }
 488
 489 #undef TT
 490
 491 /*
 492  * The following checks use two arrays of 8 or 16 bits to store the
 493  * bits that we want set or clear, respectively. They are in the
 494  * low and high half of cmd->arg1 or cmd->d[0].
 495  *
 496  * We scan options and store the bits we find set. We succeed if
 497  *
 498  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 499  *
 500  * The code is sometimes optimized not to store additional variables.
 501  */
 502
 503 static int
 504 flags_match(ipfw_insn *cmd, uint8_t bits)
 505 {
 506         u_char want_clear;
 507         bits = ~bits;
 508
 509         if (((cmd->arg1 & 0xff) & bits) != 0)
 510                 return 0; /* some bits we want set were clear */
 511
 512         want_clear = (cmd->arg1 >> 8) & 0xff;
 513         if ((want_clear & bits) != want_clear)
 514                 return 0; /* some bits we want clear were set */
 515         return 1;
 516 }
 517
 518 static int
 519 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 520 {
 521         int optlen, bits = 0;
 522         u_char *cp = (u_char *)(ip + 1);
 523         int x = (ip->ip_hl << 2) - sizeof(struct ip);
 524
 525         for (; x > 0; x -= optlen, cp += optlen) {
 526                 int opt = cp[IPOPT_OPTVAL];
 527
 528                 if (opt == IPOPT_EOL)
 529                         break;
 530
 531                 if (opt == IPOPT_NOP) {
 532                         optlen = 1;
 533                 } else {
 534                         optlen = cp[IPOPT_OLEN];
 535                         if (optlen <= 0 || optlen > x)
 536                                 return 0; /* invalid or truncated */
 537                 }
 538
 539                 switch (opt) {
 540                 case IPOPT_LSRR:
 541                         bits |= IP_FW_IPOPT_LSRR;
 542                         break;
 543
 544                 case IPOPT_SSRR:
 545                         bits |= IP_FW_IPOPT_SSRR;
 546                         break;
 547
 548                 case IPOPT_RR:
 549                         bits |= IP_FW_IPOPT_RR;
 550                         break;
 551
 552                 case IPOPT_TS:
 553                         bits |= IP_FW_IPOPT_TS;
 554                         break;
 555
 556                 default:
 557                         break;
 558                 }
 559         }
 560         return (flags_match(cmd, bits));
 561 }
 562
 563 static int
 564 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 565 {
 566         int optlen, bits = 0;
 567         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 568         u_char *cp = (u_char *)(tcp + 1);
 569         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 570
 571         for (; x > 0; x -= optlen, cp += optlen) {
 572                 int opt = cp[0];
 573
 574                 if (opt == TCPOPT_EOL)
 575                         break;
 576
 577                 if (opt == TCPOPT_NOP) {
 578                         optlen = 1;
 579                 } else {
 580                         optlen = cp[1];
 581                         if (optlen <= 0)
 582                                 break;
 583                 }
 584
 585                 switch (opt) {
 586                 case TCPOPT_MAXSEG:
 587                         bits |= IP_FW_TCPOPT_MSS;
 588                         break;
 589
 590                 case TCPOPT_WINDOW:
 591                         bits |= IP_FW_TCPOPT_WINDOW;
 592                         break;
 593
 594                 case TCPOPT_SACK_PERMITTED:
 595                 case TCPOPT_SACK:
 596                         bits |= IP_FW_TCPOPT_SACK;
 597                         break;
 598
 599                 case TCPOPT_TIMESTAMP:
 600                         bits |= IP_FW_TCPOPT_TS;
 601                         break;
 602
 603                 case TCPOPT_CC:
 604                 case TCPOPT_CCNEW:
 605                 case TCPOPT_CCECHO:
 606                         bits |= IP_FW_TCPOPT_CC;
 607                         break;
 608
 609                 default:
 610                         break;
 611                 }
 612         }
 613         return (flags_match(cmd, bits));
 614 }
 615
 616 static int
 617 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 618 {
 619         if (ifp == NULL)        /* no iface with this packet, match fails */
 620                 return 0;
 621
 622         /* Check by name or by IP address */
 623         if (cmd->name[0] != '\0') { /* match by name */
 624                 /* Check name */
 625                 if (cmd->p.glob) {
 626                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 627                                 return(1);
 628                 } else {
 629                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 630                                 return(1);
 631                 }
 632         } else {
 633                 struct ifaddr_container *ifac;
 634
 635                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 636                         struct ifaddr *ia = ifac->ifa;
 637
 638                         if (ia->ifa_addr == NULL)
 639                                 continue;
 640                         if (ia->ifa_addr->sa_family != AF_INET)
 641                                 continue;
 642                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 643                             (ia->ifa_addr))->sin_addr.s_addr)
 644                                 return(1);      /* match */
 645                 }
 646         }
 647         return(0);      /* no match, fail ... */
 648 }
 649
 650 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 651
 652 /*
 653  * We enter here when we have a rule with O_LOG.
 654  * XXX this function alone takes about 2Kbytes of code!
 655  */
 656 static void
 657 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 658          struct mbuf *m, struct ifnet *oif)
 659 {
 660         char *action;
 661         int limit_reached = 0;
 662         char action2[40], proto[48], fragment[28];
 663
 664         fragment[0] = '\0';
 665         proto[0] = '\0';
 666
 667         if (f == NULL) {        /* bogus pkt */
 668                 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
 669
 670                 if (verbose_limit != 0 &&
 671                     ctx->ipfw_norule_counter >= verbose_limit)
 672                         return;
 673                 ctx->ipfw_norule_counter++;
 674                 if (ctx->ipfw_norule_counter == verbose_limit)
 675                         limit_reached = verbose_limit;
 676                 action = "Refuse";
 677         } else {        /* O_LOG is the first action, find the real one */
 678                 ipfw_insn *cmd = ACTION_PTR(f);
 679                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 680
 681                 if (l->max_log != 0 && l->log_left == 0)
 682                         return;
 683                 l->log_left--;
 684                 if (l->log_left == 0)
 685                         limit_reached = l->max_log;
 686                 cmd += F_LEN(cmd);      /* point to first action */
 687                 if (cmd->opcode == O_PROB)
 688                         cmd += F_LEN(cmd);
 689
 690                 action = action2;
 691                 switch (cmd->opcode) {
 692                 case O_DENY:
 693                         action = "Deny";
 694                         break;
 695
 696                 case O_REJECT:
 697                         if (cmd->arg1==ICMP_REJECT_RST) {
 698                                 action = "Reset";
 699                         } else if (cmd->arg1==ICMP_UNREACH_HOST) {
 700                                 action = "Reject";
 701                         } else {
 702                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 703                                           cmd->arg1);
 704                         }
 705                         break;
 706
 707                 case O_ACCEPT:
 708                         action = "Accept";
 709                         break;
 710
 711                 case O_COUNT:
 712                         action = "Count";
 713                         break;
 714
 715                 case O_DIVERT:
 716                         ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
 717                         break;
 718
 719                 case O_TEE:
 720                         ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
 721                         break;
 722
 723                 case O_SKIPTO:
 724                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
 725                         break;
 726
 727                 case O_PIPE:
 728                         ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
 729                         break;
 730
 731                 case O_QUEUE:
 732                         ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
 733                         break;
 734
 735                 case O_FORWARD_IP:
 736                         {
 737                                 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 738                                 int len;
 739
 740                                 len = ksnprintf(SNPARGS(action2, 0),
 741                                                 "Forward to %s",
 742                                                 inet_ntoa(sa->sa.sin_addr));
 743                                 if (sa->sa.sin_port) {
 744                                         ksnprintf(SNPARGS(action2, len), ":%d",
 745                                                   sa->sa.sin_port);
 746                                 }
 747                         }
 748                         break;
 749
 750                 default:
 751                         action = "UNKNOWN";
 752                         break;
 753                 }
 754         }
 755
 756         if (hlen == 0) {        /* non-ip */
 757                 ksnprintf(SNPARGS(proto, 0), "MAC");
 758         } else {
 759                 struct ip *ip = mtod(m, struct ip *);
 760                 /* these three are all aliases to the same thing */
 761                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 762                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 763                 struct udphdr *const udp = (struct udphdr *)icmp;
 764
 765                 int ip_off, offset, ip_len;
 766                 int len;
 767
 768                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 769                         ip_off = ntohs(ip->ip_off);
 770                         ip_len = ntohs(ip->ip_len);
 771                 } else {
 772                         ip_off = ip->ip_off;
 773                         ip_len = ip->ip_len;
 774                 }
 775                 offset = ip_off & IP_OFFMASK;
 776                 switch (ip->ip_p) {
 777                 case IPPROTO_TCP:
 778                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 779                                         inet_ntoa(ip->ip_src));
 780                         if (offset == 0) {
 781                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 782                                           ntohs(tcp->th_sport),
 783                                           inet_ntoa(ip->ip_dst),
 784                                           ntohs(tcp->th_dport));
 785                         } else {
 786                                 ksnprintf(SNPARGS(proto, len), " %s",
 787                                           inet_ntoa(ip->ip_dst));
 788                         }
 789                         break;
 790
 791                 case IPPROTO_UDP:
 792                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 793                                         inet_ntoa(ip->ip_src));
 794                         if (offset == 0) {
 795                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 796                                           ntohs(udp->uh_sport),
 797                                           inet_ntoa(ip->ip_dst),
 798                                           ntohs(udp->uh_dport));
 799                         } else {
 800                                 ksnprintf(SNPARGS(proto, len), " %s",
 801                                           inet_ntoa(ip->ip_dst));
 802                         }
 803                         break;
 804
 805                 case IPPROTO_ICMP:
 806                         if (offset == 0) {
 807                                 len = ksnprintf(SNPARGS(proto, 0),
 808                                                 "ICMP:%u.%u ",
 809                                                 icmp->icmp_type,
 810                                                 icmp->icmp_code);
 811                         } else {
 812                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 813                         }
 814                         len += ksnprintf(SNPARGS(proto, len), "%s",
 815                                          inet_ntoa(ip->ip_src));
 816                         ksnprintf(SNPARGS(proto, len), " %s",
 817                                   inet_ntoa(ip->ip_dst));
 818                         break;
 819
 820                 default:
 821                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 822                                         inet_ntoa(ip->ip_src));
 823                         ksnprintf(SNPARGS(proto, len), " %s",
 824                                   inet_ntoa(ip->ip_dst));
 825                         break;
 826                 }
 827
 828                 if (ip_off & (IP_MF | IP_OFFMASK)) {
 829                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 830                                   ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 831                                   offset << 3, (ip_off & IP_MF) ? "+" : "");
 832                 }
 833         }
 834
 835         if (oif || m->m_pkthdr.rcvif) {
 836                 log(LOG_SECURITY | LOG_INFO,
 837                     "ipfw: %d %s %s %s via %s%s\n",
 838                     f ? f->rulenum : -1,
 839                     action, proto, oif ? "out" : "in",
 840                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 841                     fragment);
 842         } else {
 843                 log(LOG_SECURITY | LOG_INFO,
 844                     "ipfw: %d %s %s [no if info]%s\n",
 845                     f ? f->rulenum : -1,
 846                     action, proto, fragment);
 847         }
 848
 849         if (limit_reached) {
 850                 log(LOG_SECURITY | LOG_NOTICE,
 851                     "ipfw: limit %d reached on entry %d\n",
 852                     limit_reached, f ? f->rulenum : -1);
 853         }
 854 }
 855
 856 #undef SNPARGS
 857
 858 /*
 859  * IMPORTANT: the hash function for dynamic rules must be commutative
 860  * in source and destination (ip,port), because rules are bidirectional
 861  * and we want to find both in the same bucket.
 862  */
 863 static __inline int
 864 hash_packet(struct ipfw_flow_id *id)
 865 {
 866         uint32_t i;
 867
 868         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 869         i &= (curr_dyn_buckets - 1);
 870         return i;
 871 }
 872
 873 /**
 874  * unlink a dynamic rule from a chain. prev is a pointer to
 875  * the previous one, q is a pointer to the rule to delete,
 876  * head is a pointer to the head of the queue.
 877  * Modifies q and potentially also head.
 878  */
 879 #define UNLINK_DYN_RULE(prev, head, q)                                  \
 880 do {                                                                    \
 881         ipfw_dyn_rule *old_q = q;                                       \
 882                                                                         \
 883         /* remove a refcount to the parent */                           \
 884         if (q->dyn_type == O_LIMIT)                                     \
 885                 q->parent->count--;                                     \
 886         DPRINTF("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",    \
 887                 q->id.src_ip, q->id.src_port,                           \
 888                 q->id.dst_ip, q->id.dst_port, dyn_count - 1);           \
 889         if (prev != NULL)                                               \
 890                 prev->next = q = q->next;                               \
 891         else                                                            \
 892                 head = q = q->next;                                     \
 893         KASSERT(dyn_count > 0, ("invalid dyn count %u", dyn_count));    \
 894         dyn_count--;                                                    \
 895         kfree(old_q, M_IPFW);                                           \
 896 } while (0)
 897
 898 #define TIME_LEQ(a, b)  ((int)((a) - (b)) <= 0)
 899
 900 /**
 901  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 902  *
 903  * If keep_me == NULL, rules are deleted even if not expired,
 904  * otherwise only expired rules are removed.
 905  *
 906  * The value of the second parameter is also used to point to identify
 907  * a rule we absolutely do not want to remove (e.g. because we are
 908  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 909  * rules). The pointer is only used for comparison, so any non-null
 910  * value will do.
 911  */
 912 static void
 913 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 914 {
 915         static uint32_t last_remove = 0; /* XXX */
 916
 917 #define FORCE   (keep_me == NULL)
 918
 919         ipfw_dyn_rule *prev, *q;
 920         int i, pass = 0, max_pass = 0, unlinked = 0;
 921
 922         if (ipfw_dyn_v == NULL || dyn_count == 0)
 923                 return;
 924         /* do not expire more than once per second, it is useless */
 925         if (!FORCE && last_remove == time_second)
 926                 return;
 927         last_remove = time_second;
 928
 929         /*
 930          * because O_LIMIT refer to parent rules, during the first pass only
 931          * remove child and mark any pending LIMIT_PARENT, and remove
 932          * them in a second pass.
 933          */
 934 next_pass:
 935         for (i = 0; i < curr_dyn_buckets; i++) {
 936                 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
 937                         /*
 938                          * Logic can become complex here, so we split tests.
 939                          */
 940                         if (q == keep_me)
 941                                 goto next;
 942                         if (rule != NULL && rule->stub != q->stub)
 943                                 goto next; /* not the one we are looking for */
 944                         if (q->dyn_type == O_LIMIT_PARENT) {
 945                                 /*
 946                                  * handle parent in the second pass,
 947                                  * record we need one.
 948                                  */
 949                                 max_pass = 1;
 950                                 if (pass == 0)
 951                                         goto next;
 952                                 if (FORCE && q->count != 0) {
 953                                         /* XXX should not happen! */
 954                                         kprintf("OUCH! cannot remove rule, "
 955                                                 "count %d\n", q->count);
 956                                 }
 957                         } else {
 958                                 if (!FORCE && !TIME_LEQ(q->expire, time_second))
 959                                         goto next;
 960                         }
 961                         unlinked = 1;
 962                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 963                         continue;
 964 next:
 965                         prev = q;
 966                         q = q->next;
 967                 }
 968         }
 969         if (pass++ < max_pass)
 970                 goto next_pass;
 971
 972         if (unlinked)
 973                 ++dyn_buckets_gen;
 974
 975 #undef FORCE
 976 }
 977
 978 /**
 979  * lookup a dynamic rule.
 980  */
 981 static ipfw_dyn_rule *
 982 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 983                 struct tcphdr *tcp)
 984 {
 985         /*
 986          * stateful ipfw extensions.
 987          * Lookup into dynamic session queue
 988          */
 989 #define MATCH_REVERSE   0
 990 #define MATCH_FORWARD   1
 991 #define MATCH_NONE      2
 992 #define MATCH_UNKNOWN   3
 993         int i, dir = MATCH_NONE;
 994         ipfw_dyn_rule *prev, *q=NULL;
 995
 996         if (ipfw_dyn_v == NULL)
 997                 goto done;      /* not found */
 998
 999         i = hash_packet(pkt);
1000         for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
1001                 if (q->dyn_type == O_LIMIT_PARENT)
1002                         goto next;
1003
1004                 if (TIME_LEQ(q->expire, time_second)) {
1005                         /*
1006                          * Entry expired; skip.
1007                          * Let ipfw_tick() take care of it
1008                          */
1009                         goto next;
1010                 }
1011
1012                 if (pkt->proto == q->id.proto) {
1013                         if (pkt->src_ip == q->id.src_ip &&
1014                             pkt->dst_ip == q->id.dst_ip &&
1015                             pkt->src_port == q->id.src_port &&
1016                             pkt->dst_port == q->id.dst_port) {
1017                                 dir = MATCH_FORWARD;
1018                                 break;
1019                         }
1020                         if (pkt->src_ip == q->id.dst_ip &&
1021                             pkt->dst_ip == q->id.src_ip &&
1022                             pkt->src_port == q->id.dst_port &&
1023                             pkt->dst_port == q->id.src_port) {
1024                                 dir = MATCH_REVERSE;
1025                                 break;
1026                         }
1027                 }
1028 next:
1029                 prev = q;
1030                 q = q->next;
1031         }
1032         if (q == NULL)
1033                 goto done; /* q = NULL, not found */
1034
1035         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1036                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1037
1038 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
1039 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
1040
1041                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1042                 switch (q->state) {
1043                 case TH_SYN:                            /* opening */
1044                         q->expire = time_second + dyn_syn_lifetime;
1045                         break;
1046
1047                 case BOTH_SYN:                  /* move to established */
1048                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
1049                 case BOTH_SYN | (TH_FIN << 8) :
1050                         if (tcp) {
1051                                 uint32_t ack = ntohl(tcp->th_ack);
1052
1053 #define _SEQ_GE(a, b)   ((int)(a) - (int)(b) >= 0)
1054
1055                                 if (dir == MATCH_FORWARD) {
1056                                         if (q->ack_fwd == 0 ||
1057                                             _SEQ_GE(ack, q->ack_fwd))
1058                                                 q->ack_fwd = ack;
1059                                         else /* ignore out-of-sequence */
1060                                                 break;
1061                                 } else {
1062                                         if (q->ack_rev == 0 ||
1063                                             _SEQ_GE(ack, q->ack_rev))
1064                                                 q->ack_rev = ack;
1065                                         else /* ignore out-of-sequence */
1066                                                 break;
1067                                 }
1068 #undef _SEQ_GE
1069                         }
1070                         q->expire = time_second + dyn_ack_lifetime;
1071                         break;
1072
1073                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
1074                         KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
1075                         q->expire = time_second + dyn_fin_lifetime;
1076                         break;
1077
1078                 default:
1079 #if 0
1080                         /*
1081                          * reset or some invalid combination, but can also
1082                          * occur if we use keep-state the wrong way.
1083                          */
1084                         if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
1085                                 kprintf("invalid state: 0x%x\n", q->state);
1086 #endif
1087                         KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
1088                         q->expire = time_second + dyn_rst_lifetime;
1089                         break;
1090                 }
1091         } else if (pkt->proto == IPPROTO_UDP) {
1092                 q->expire = time_second + dyn_udp_lifetime;
1093         } else {
1094                 /* other protocols */
1095                 q->expire = time_second + dyn_short_lifetime;
1096         }
1097 done:
1098         if (match_direction)
1099                 *match_direction = dir;
1100         return q;
1101 }
1102
1103 static struct ip_fw *
1104 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
1105             uint16_t len, int *deny)
1106 {
1107         struct ip_fw *rule = NULL;
1108         ipfw_dyn_rule *q;
1109         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1110         uint32_t gen;
1111
1112         *deny = 0;
1113         gen = ctx->ipfw_gen;
1114
1115         lockmgr(&dyn_lock, LK_SHARED);
1116
1117         if (ctx->ipfw_gen != gen) {
1118                 /*
1119                  * Static rules had been change when we were waiting
1120                  * for the dynamic hash table lock; deny this packet,
1121                  * since it is _not_ known whether it is safe to keep
1122                  * iterating the static rules.
1123                  */
1124                 *deny = 1;
1125                 goto back;
1126         }
1127
1128         q = lookup_dyn_rule(pkt, match_direction, tcp);
1129         if (q == NULL) {
1130                 rule = NULL;
1131         } else {
1132                 rule = q->stub->rule[mycpuid];
1133                 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid);
1134
1135                 /* XXX */
1136                 q->pcnt++;
1137                 q->bcnt += len;
1138         }
1139 back:
1140         lockmgr(&dyn_lock, LK_RELEASE);
1141         return rule;
1142 }
1143
1144 static void
1145 realloc_dynamic_table(void)
1146 {
1147         ipfw_dyn_rule **old_dyn_v;
1148         uint32_t old_curr_dyn_buckets;
1149
1150         KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
1151                 ("invalid dyn_buckets %d", dyn_buckets));
1152
1153         /* Save the current buckets array for later error recovery */
1154         old_dyn_v = ipfw_dyn_v;
1155         old_curr_dyn_buckets = curr_dyn_buckets;
1156
1157         curr_dyn_buckets = dyn_buckets;
1158         for (;;) {
1159                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1160                                      M_IPFW, M_NOWAIT | M_ZERO);
1161                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1162                         break;
1163
1164                 curr_dyn_buckets /= 2;
1165                 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
1166                     old_dyn_v != NULL) {
1167                         /*
1168                          * Don't try allocating smaller buckets array, reuse
1169                          * the old one, which alreay contains enough buckets
1170                          */
1171                         break;
1172                 }
1173         }
1174
1175         if (ipfw_dyn_v != NULL) {
1176                 if (old_dyn_v != NULL)
1177                         kfree(old_dyn_v, M_IPFW);
1178         } else {
1179                 /* Allocation failed, restore old buckets array */
1180                 ipfw_dyn_v = old_dyn_v;
1181                 curr_dyn_buckets = old_curr_dyn_buckets;
1182         }
1183
1184         if (ipfw_dyn_v != NULL)
1185                 ++dyn_buckets_gen;
1186 }
1187
1188 /**
1189  * Install state of type 'type' for a dynamic session.
1190  * The hash table contains two type of rules:
1191  * - regular rules (O_KEEP_STATE)
1192  * - rules for sessions with limited number of sess per user
1193  *   (O_LIMIT). When they are created, the parent is
1194  *   increased by 1, and decreased on delete. In this case,
1195  *   the third parameter is the parent rule and not the chain.
1196  * - "parent" rules for the above (O_LIMIT_PARENT).
1197  */
1198 static ipfw_dyn_rule *
1199 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1200 {
1201         ipfw_dyn_rule *r;
1202         int i;
1203
1204         if (ipfw_dyn_v == NULL ||
1205             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1206                 realloc_dynamic_table();
1207                 if (ipfw_dyn_v == NULL)
1208                         return NULL; /* failed ! */
1209         }
1210         i = hash_packet(id);
1211
1212         r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1213         if (r == NULL) {
1214                 kprintf ("sorry cannot allocate state\n");
1215                 return NULL;
1216         }
1217
1218         /* increase refcount on parent, and set pointer */
1219         if (dyn_type == O_LIMIT) {
1220                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1221
1222                 if (parent->dyn_type != O_LIMIT_PARENT)
1223                         panic("invalid parent");
1224                 parent->count++;
1225                 r->parent = parent;
1226                 rule = parent->stub->rule[mycpuid];
1227                 KKASSERT(rule->stub == parent->stub);
1228         }
1229         KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1230
1231         r->id = *id;
1232         r->expire = time_second + dyn_syn_lifetime;
1233         r->stub = rule->stub;
1234         r->dyn_type = dyn_type;
1235         r->pcnt = r->bcnt = 0;
1236         r->count = 0;
1237
1238         r->bucket = i;
1239         r->next = ipfw_dyn_v[i];
1240         ipfw_dyn_v[i] = r;
1241         dyn_count++;
1242         dyn_buckets_gen++;
1243         DPRINTF("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1244                 dyn_type,
1245                 r->id.src_ip, r->id.src_port,
1246                 r->id.dst_ip, r->id.dst_port, dyn_count);
1247         return r;
1248 }
1249
1250 /**
1251  * lookup dynamic parent rule using pkt and rule as search keys.
1252  * If the lookup fails, then install one.
1253  */
1254 static ipfw_dyn_rule *
1255 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1256 {
1257         ipfw_dyn_rule *q;
1258         int i;
1259
1260         if (ipfw_dyn_v) {
1261                 i = hash_packet(pkt);
1262                 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1263                         if (q->dyn_type == O_LIMIT_PARENT &&
1264                             rule->stub == q->stub &&
1265                             pkt->proto == q->id.proto &&
1266                             pkt->src_ip == q->id.src_ip &&
1267                             pkt->dst_ip == q->id.dst_ip &&
1268                             pkt->src_port == q->id.src_port &&
1269                             pkt->dst_port == q->id.dst_port) {
1270                                 q->expire = time_second + dyn_short_lifetime;
1271                                 DPRINTF("lookup_dyn_parent found 0x%p\n", q);
1272                                 return q;
1273                         }
1274                 }
1275         }
1276         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1277 }
1278
1279 /**
1280  * Install dynamic state for rule type cmd->o.opcode
1281  *
1282  * Returns 1 (failure) if state is not installed because of errors or because
1283  * session limitations are enforced.
1284  */
1285 static int
1286 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1287                      struct ip_fw_args *args)
1288 {
1289         static int last_log; /* XXX */
1290
1291         ipfw_dyn_rule *q;
1292
1293         DPRINTF("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1294                 cmd->o.opcode,
1295                 args->f_id.src_ip, args->f_id.src_port,
1296                 args->f_id.dst_ip, args->f_id.dst_port);
1297
1298         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1299         if (q != NULL) { /* should never occur */
1300                 if (last_log != time_second) {
1301                         last_log = time_second;
1302                         kprintf(" install_state: entry already present, done\n");
1303                 }
1304                 return 0;
1305         }
1306
1307         if (dyn_count >= dyn_max) {
1308                 /*
1309                  * Run out of slots, try to remove any expired rule.
1310                  */
1311                 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1312                 if (dyn_count >= dyn_max) {
1313                         if (last_log != time_second) {
1314                                 last_log = time_second;
1315                                 kprintf("install_state: "
1316                                         "Too many dynamic rules\n");
1317                         }
1318                         return 1; /* cannot install, notify caller */
1319                 }
1320         }
1321
1322         switch (cmd->o.opcode) {
1323         case O_KEEP_STATE: /* bidir rule */
1324                 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1325                         return 1;
1326                 break;
1327
1328         case O_LIMIT: /* limit number of sessions */
1329                 {
1330                         uint16_t limit_mask = cmd->limit_mask;
1331                         struct ipfw_flow_id id;
1332                         ipfw_dyn_rule *parent;
1333
1334                         DPRINTF("installing dyn-limit rule %d\n",
1335                                 cmd->conn_limit);
1336
1337                         id.dst_ip = id.src_ip = 0;
1338                         id.dst_port = id.src_port = 0;
1339                         id.proto = args->f_id.proto;
1340
1341                         if (limit_mask & DYN_SRC_ADDR)
1342                                 id.src_ip = args->f_id.src_ip;
1343                         if (limit_mask & DYN_DST_ADDR)
1344                                 id.dst_ip = args->f_id.dst_ip;
1345                         if (limit_mask & DYN_SRC_PORT)
1346                                 id.src_port = args->f_id.src_port;
1347                         if (limit_mask & DYN_DST_PORT)
1348                                 id.dst_port = args->f_id.dst_port;
1349
1350                         parent = lookup_dyn_parent(&id, rule);
1351                         if (parent == NULL) {
1352                                 kprintf("add parent failed\n");
1353                                 return 1;
1354                         }
1355
1356                         if (parent->count >= cmd->conn_limit) {
1357                                 /*
1358                                  * See if we can remove some expired rule.
1359                                  */
1360                                 remove_dyn_rule_locked(rule, parent);
1361                                 if (parent->count >= cmd->conn_limit) {
1362                                         if (fw_verbose &&
1363                                             last_log != time_second) {
1364                                                 last_log = time_second;
1365                                                 log(LOG_SECURITY | LOG_DEBUG,
1366                                                     "drop session, "
1367                                                     "too many entries\n");
1368                                         }
1369                                         return 1;
1370                                 }
1371                         }
1372                         if (add_dyn_rule(&args->f_id, O_LIMIT,
1373                                          (struct ip_fw *)parent) == NULL)
1374                                 return 1;
1375                 }
1376                 break;
1377         default:
1378                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1379                 return 1;
1380         }
1381         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1382         return 0;
1383 }
1384
1385 static int
1386 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1387               struct ip_fw_args *args, int *deny)
1388 {
1389         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1390         uint32_t gen;
1391         int ret = 0;
1392
1393         *deny = 0;
1394         gen = ctx->ipfw_gen;
1395
1396         lockmgr(&dyn_lock, LK_EXCLUSIVE);
1397         if (ctx->ipfw_gen != gen) {
1398                 /* See the comment in lookup_rule() */
1399                 *deny = 1;
1400         } else {
1401                 ret = install_state_locked(rule, cmd, args);
1402         }
1403         lockmgr(&dyn_lock, LK_RELEASE);
1404
1405         return ret;
1406 }
1407
1408 /*
1409  * Transmit a TCP packet, containing either a RST or a keepalive.
1410  * When flags & TH_RST, we are sending a RST packet, because of a
1411  * "reset" action matched the packet.
1412  * Otherwise we are sending a keepalive, and flags & TH_
1413  */
1414 static void
1415 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1416 {
1417         struct mbuf *m;
1418         struct ip *ip;
1419         struct tcphdr *tcp;
1420         struct route sro;       /* fake route */
1421
1422         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1423         if (m == NULL)
1424                 return;
1425         m->m_pkthdr.rcvif = NULL;
1426         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1427         m->m_data += max_linkhdr;
1428
1429         ip = mtod(m, struct ip *);
1430         bzero(ip, m->m_len);
1431         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1432         ip->ip_p = IPPROTO_TCP;
1433         tcp->th_off = 5;
1434
1435         /*
1436          * Assume we are sending a RST (or a keepalive in the reverse
1437          * direction), swap src and destination addresses and ports.
1438          */
1439         ip->ip_src.s_addr = htonl(id->dst_ip);
1440         ip->ip_dst.s_addr = htonl(id->src_ip);
1441         tcp->th_sport = htons(id->dst_port);
1442         tcp->th_dport = htons(id->src_port);
1443         if (flags & TH_RST) {   /* we are sending a RST */
1444                 if (flags & TH_ACK) {
1445                         tcp->th_seq = htonl(ack);
1446                         tcp->th_ack = htonl(0);
1447                         tcp->th_flags = TH_RST;
1448                 } else {
1449                         if (flags & TH_SYN)
1450                                 seq++;
1451                         tcp->th_seq = htonl(0);
1452                         tcp->th_ack = htonl(seq);
1453                         tcp->th_flags = TH_RST | TH_ACK;
1454                 }
1455         } else {
1456                 /*
1457                  * We are sending a keepalive. flags & TH_SYN determines
1458                  * the direction, forward if set, reverse if clear.
1459                  * NOTE: seq and ack are always assumed to be correct
1460                  * as set by the caller. This may be confusing...
1461                  */
1462                 if (flags & TH_SYN) {
1463                         /*
1464                          * we have to rewrite the correct addresses!
1465                          */
1466                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1467                         ip->ip_src.s_addr = htonl(id->src_ip);
1468                         tcp->th_dport = htons(id->dst_port);
1469                         tcp->th_sport = htons(id->src_port);
1470                 }
1471                 tcp->th_seq = htonl(seq);
1472                 tcp->th_ack = htonl(ack);
1473                 tcp->th_flags = TH_ACK;
1474         }
1475
1476         /*
1477          * set ip_len to the payload size so we can compute
1478          * the tcp checksum on the pseudoheader
1479          * XXX check this, could save a couple of words ?
1480          */
1481         ip->ip_len = htons(sizeof(struct tcphdr));
1482         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1483
1484         /*
1485          * now fill fields left out earlier
1486          */
1487         ip->ip_ttl = ip_defttl;
1488         ip->ip_len = m->m_pkthdr.len;
1489
1490         bzero(&sro, sizeof(sro));
1491         ip_rtaddr(ip->ip_dst, &sro);
1492
1493         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1494         ip_output(m, NULL, &sro, 0, NULL, NULL);
1495         if (sro.ro_rt)
1496                 RTFREE(sro.ro_rt);
1497 }
1498
1499 /*
1500  * sends a reject message, consuming the mbuf passed as an argument.
1501  */
1502 static void
1503 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1504 {
1505         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1506                 /* We need the IP header in host order for icmp_error(). */
1507                 if (args->eh != NULL) {
1508                         struct ip *ip = mtod(args->m, struct ip *);
1509
1510                         ip->ip_len = ntohs(ip->ip_len);
1511                         ip->ip_off = ntohs(ip->ip_off);
1512                 }
1513                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1514         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1515                 struct tcphdr *const tcp =
1516                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1517
1518                 if ((tcp->th_flags & TH_RST) == 0) {
1519                         send_pkt(&args->f_id, ntohl(tcp->th_seq),
1520                                  ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1521                 }
1522                 m_freem(args->m);
1523         } else {
1524                 m_freem(args->m);
1525         }
1526         args->m = NULL;
1527 }
1528
1529 /**
1530  *
1531  * Given an ip_fw *, lookup_next_rule will return a pointer
1532  * to the next rule, which can be either the jump
1533  * target (for skipto instructions) or the next one in the list (in
1534  * all other cases including a missing jump target).
1535  * The result is also written in the "next_rule" field of the rule.
1536  * Backward jumps are not allowed, so start looking from the next
1537  * rule...
1538  *
1539  * This never returns NULL -- in case we do not have an exact match,
1540  * the next rule is returned. When the ruleset is changed,
1541  * pointers are flushed so we are always correct.
1542  */
1543
1544 static struct ip_fw *
1545 lookup_next_rule(struct ip_fw *me)
1546 {
1547         struct ip_fw *rule = NULL;
1548         ipfw_insn *cmd;
1549
1550         /* look for action, in case it is a skipto */
1551         cmd = ACTION_PTR(me);
1552         if (cmd->opcode == O_LOG)
1553                 cmd += F_LEN(cmd);
1554         if (cmd->opcode == O_SKIPTO) {
1555                 for (rule = me->next; rule; rule = rule->next) {
1556                         if (rule->rulenum >= cmd->arg1)
1557                                 break;
1558                 }
1559         }
1560         if (rule == NULL)                       /* failure or not a skipto */
1561                 rule = me->next;
1562         me->next_rule = rule;
1563         return rule;
1564 }
1565
1566 static int
1567 _ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1568                 enum ipfw_opcodes opcode, uid_t uid)
1569 {
1570         struct in_addr src_ip, dst_ip;
1571         struct inpcbinfo *pi;
1572         int wildcard;
1573         struct inpcb *pcb;
1574
1575         if (fid->proto == IPPROTO_TCP) {
1576                 wildcard = 0;
1577                 pi = &tcbinfo[mycpuid];
1578         } else if (fid->proto == IPPROTO_UDP) {
1579                 wildcard = 1;
1580                 pi = &udbinfo;
1581         } else {
1582                 return 0;
1583         }
1584
1585         /*
1586          * Values in 'fid' are in host byte order
1587          */
1588         dst_ip.s_addr = htonl(fid->dst_ip);
1589         src_ip.s_addr = htonl(fid->src_ip);
1590         if (oif) {
1591                 pcb = in_pcblookup_hash(pi,
1592                         dst_ip, htons(fid->dst_port),
1593                         src_ip, htons(fid->src_port),
1594                         wildcard, oif);
1595         } else {
1596                 pcb = in_pcblookup_hash(pi,
1597                         src_ip, htons(fid->src_port),
1598                         dst_ip, htons(fid->dst_port),
1599                         wildcard, NULL);
1600         }
1601         if (pcb == NULL || pcb->inp_socket == NULL)
1602                 return 0;
1603
1604         if (opcode == O_UID) {
1605 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1606                 return !socheckuid(pcb->inp_socket, uid);
1607 #undef socheckuid
1608         } else  {
1609                 return groupmember(uid, pcb->inp_socket->so_cred);
1610         }
1611 }
1612
1613 static int
1614 ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1615                enum ipfw_opcodes opcode, uid_t uid, int *deny)
1616 {
1617         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1618         uint32_t gen;
1619         int match = 0;
1620
1621         *deny = 0;
1622         gen = ctx->ipfw_gen;
1623
1624         get_mplock();
1625         if (gen != ctx->ipfw_gen) {
1626                 /* See the comment in lookup_rule() */
1627                 *deny = 1;
1628         } else {
1629                 match = _ipfw_match_uid(fid, oif, opcode, uid);
1630         }
1631         rel_mplock();
1632         return match;
1633 }
1634
1635 /*
1636  * The main check routine for the firewall.
1637  *
1638  * All arguments are in args so we can modify them and return them
1639  * back to the caller.
1640  *
1641  * Parameters:
1642  *
1643  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1644  *              Starts with the IP header.
1645  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1646  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1647  *              The incoming interface is in the mbuf. (in)
1648  *
1649  *      args->rule      Pointer to the last matching rule (in/out)
1650  *      args->f_id      Addresses grabbed from the packet (out)
1651  *
1652  * Return value:
1653  *
1654  *      If the packet was denied/rejected and has been dropped, *m is equal
1655  *      to NULL upon return.
1656  *
1657  *      IP_FW_DENY      the packet must be dropped.
1658  *      IP_FW_PASS      The packet is to be accepted and routed normally.
1659  *      IP_FW_DIVERT    Divert the packet to port (args->cookie)
1660  *      IP_FW_TEE       Tee the packet to port (args->cookie)
1661  *      IP_FW_DUMMYNET  Send the packet to pipe/queue (args->cookie)
1662  */
1663
1664 static int
1665 ipfw_chk(struct ip_fw_args *args)
1666 {
1667         /*
1668          * Local variables hold state during the processing of a packet.
1669          *
1670          * IMPORTANT NOTE: to speed up the processing of rules, there
1671          * are some assumption on the values of the variables, which
1672          * are documented here. Should you change them, please check
1673          * the implementation of the various instructions to make sure
1674          * that they still work.
1675          *
1676          * args->eh     The MAC header. It is non-null for a layer2
1677          *      packet, it is NULL for a layer-3 packet.
1678          *
1679          * m | args->m  Pointer to the mbuf, as received from the caller.
1680          *      It may change if ipfw_chk() does an m_pullup, or if it
1681          *      consumes the packet because it calls send_reject().
1682          *      XXX This has to change, so that ipfw_chk() never modifies
1683          *      or consumes the buffer.
1684          * ip   is simply an alias of the value of m, and it is kept
1685          *      in sync with it (the packet is  supposed to start with
1686          *      the ip header).
1687          */
1688         struct mbuf *m = args->m;
1689         struct ip *ip = mtod(m, struct ip *);
1690
1691         /*
1692          * oif | args->oif      If NULL, ipfw_chk has been called on the
1693          *      inbound path (ether_input, ip_input).
1694          *      If non-NULL, ipfw_chk has been called on the outbound path
1695          *      (ether_output, ip_output).
1696          */
1697         struct ifnet *oif = args->oif;
1698
1699         struct ip_fw *f = NULL;         /* matching rule */
1700         int retval = IP_FW_PASS;
1701         struct m_tag *mtag;
1702         struct divert_info *divinfo;
1703
1704         /*
1705          * hlen The length of the IPv4 header.
1706          *      hlen >0 means we have an IPv4 packet.
1707          */
1708         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1709
1710         /*
1711          * offset       The offset of a fragment. offset != 0 means that
1712          *      we have a fragment at this offset of an IPv4 packet.
1713          *      offset == 0 means that (if this is an IPv4 packet)
1714          *      this is the first or only fragment.
1715          */
1716         u_short offset = 0;
1717
1718         /*
1719          * Local copies of addresses. They are only valid if we have
1720          * an IP packet.
1721          *
1722          * proto        The protocol. Set to 0 for non-ip packets,
1723          *      or to the protocol read from the packet otherwise.
1724          *      proto != 0 means that we have an IPv4 packet.
1725          *
1726          * src_port, dst_port   port numbers, in HOST format. Only
1727          *      valid for TCP and UDP packets.
1728          *
1729          * src_ip, dst_ip       ip addresses, in NETWORK format.
1730          *      Only valid for IPv4 packets.
1731          */
1732         uint8_t proto;
1733         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1734         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1735         uint16_t ip_len = 0;
1736
1737         /*
1738          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1739          *      MATCH_NONE when checked and not matched (dyn_f = NULL),
1740          *      MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1741          */
1742         int dyn_dir = MATCH_UNKNOWN;
1743         struct ip_fw *dyn_f = NULL;
1744         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1745
1746         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1747                 return IP_FW_PASS;      /* accept */
1748
1749         if (args->eh == NULL ||         /* layer 3 packet */
1750             (m->m_pkthdr.len >= sizeof(struct ip) &&
1751              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1752                 hlen = ip->ip_hl << 2;
1753
1754         /*
1755          * Collect parameters into local variables for faster matching.
1756          */
1757         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1758                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1759                 goto after_ip_checks;
1760         }
1761
1762         proto = args->f_id.proto = ip->ip_p;
1763         src_ip = ip->ip_src;
1764         dst_ip = ip->ip_dst;
1765         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1766                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1767                 ip_len = ntohs(ip->ip_len);
1768         } else {
1769                 offset = ip->ip_off & IP_OFFMASK;
1770                 ip_len = ip->ip_len;
1771         }
1772
1773 #define PULLUP_TO(len)                          \
1774 do {                                            \
1775         if (m->m_len < (len)) {                 \
1776                 args->m = m = m_pullup(m, (len));\
1777                 if (m == NULL)                  \
1778                         goto pullup_failed;     \
1779                 ip = mtod(m, struct ip *);      \
1780         }                                       \
1781 } while (0)
1782
1783         if (offset == 0) {
1784                 switch (proto) {
1785                 case IPPROTO_TCP:
1786                         {
1787                                 struct tcphdr *tcp;
1788
1789                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1790                                 tcp = L3HDR(struct tcphdr, ip);
1791                                 dst_port = tcp->th_dport;
1792                                 src_port = tcp->th_sport;
1793                                 args->f_id.flags = tcp->th_flags;
1794                         }
1795                         break;
1796
1797                 case IPPROTO_UDP:
1798                         {
1799                                 struct udphdr *udp;
1800
1801                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1802                                 udp = L3HDR(struct udphdr, ip);
1803                                 dst_port = udp->uh_dport;
1804                                 src_port = udp->uh_sport;
1805                         }
1806                         break;
1807
1808                 case IPPROTO_ICMP:
1809                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1810                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1811                         break;
1812
1813                 default:
1814                         break;
1815                 }
1816         }
1817
1818 #undef PULLUP_TO
1819
1820         args->f_id.src_ip = ntohl(src_ip.s_addr);
1821         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1822         args->f_id.src_port = src_port = ntohs(src_port);
1823         args->f_id.dst_port = dst_port = ntohs(dst_port);
1824
1825 after_ip_checks:
1826         if (args->rule) {
1827                 /*
1828                  * Packet has already been tagged. Look for the next rule
1829                  * to restart processing.
1830                  *
1831                  * If fw_one_pass != 0 then just accept it.
1832                  * XXX should not happen here, but optimized out in
1833                  * the caller.
1834                  */
1835                 if (fw_one_pass)
1836                         return IP_FW_PASS;
1837
1838                 /* This rule is being/has been flushed */
1839                 if (ipfw_flushing)
1840                         return IP_FW_DENY;
1841
1842                 KASSERT(args->rule->cpuid == mycpuid,
1843                         ("rule used on cpu%d", mycpuid));
1844
1845                 /* This rule was deleted */
1846                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1847                         return IP_FW_DENY;
1848
1849                 f = args->rule->next_rule;
1850                 if (f == NULL)
1851                         f = lookup_next_rule(args->rule);
1852         } else {
1853                 /*
1854                  * Find the starting rule. It can be either the first
1855                  * one, or the one after divert_rule if asked so.
1856                  */
1857                 int skipto;
1858
1859                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1860                 if (mtag != NULL) {
1861                         divinfo = m_tag_data(mtag);
1862                         skipto = divinfo->skipto;
1863                 } else {
1864                         skipto = 0;
1865                 }
1866
1867                 f = ctx->ipfw_layer3_chain;
1868                 if (args->eh == NULL && skipto != 0) {
1869                         /* No skipto during rule flushing */
1870                         if (ipfw_flushing)
1871                                 return IP_FW_DENY;
1872
1873                         if (skipto >= IPFW_DEFAULT_RULE)
1874                                 return IP_FW_DENY; /* invalid */
1875
1876                         while (f && f->rulenum <= skipto)
1877                                 f = f->next;
1878                         if (f == NULL)  /* drop packet */
1879                                 return IP_FW_DENY;
1880                 } else if (ipfw_flushing) {
1881                         /* Rules are being flushed; skip to default rule */
1882                         f = ctx->ipfw_default_rule;
1883                 }
1884         }
1885         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1886                 m_tag_delete(m, mtag);
1887
1888         /*
1889          * Now scan the rules, and parse microinstructions for each rule.
1890          */
1891         for (; f; f = f->next) {
1892                 int l, cmdlen;
1893                 ipfw_insn *cmd;
1894                 int skip_or; /* skip rest of OR block */
1895
1896 again:
1897                 if (ctx->ipfw_set_disable & (1 << f->set))
1898                         continue;
1899
1900                 skip_or = 0;
1901                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1902                      l -= cmdlen, cmd += cmdlen) {
1903                         int match, deny;
1904
1905                         /*
1906                          * check_body is a jump target used when we find a
1907                          * CHECK_STATE, and need to jump to the body of
1908                          * the target rule.
1909                          */
1910
1911 check_body:
1912                         cmdlen = F_LEN(cmd);
1913                         /*
1914                          * An OR block (insn_1 || .. || insn_n) has the
1915                          * F_OR bit set in all but the last instruction.
1916                          * The first match will set "skip_or", and cause
1917                          * the following instructions to be skipped until
1918                          * past the one with the F_OR bit clear.
1919                          */
1920                         if (skip_or) {          /* skip this instruction */
1921                                 if ((cmd->len & F_OR) == 0)
1922                                         skip_or = 0;    /* next one is good */
1923                                 continue;
1924                         }
1925                         match = 0; /* set to 1 if we succeed */
1926
1927                         switch (cmd->opcode) {
1928                         /*
1929                          * The first set of opcodes compares the packet's
1930                          * fields with some pattern, setting 'match' if a
1931                          * match is found. At the end of the loop there is
1932                          * logic to deal with F_NOT and F_OR flags associated
1933                          * with the opcode.
1934                          */
1935                         case O_NOP:
1936                                 match = 1;
1937                                 break;
1938
1939                         case O_FORWARD_MAC:
1940                                 kprintf("ipfw: opcode %d unimplemented\n",
1941                                         cmd->opcode);
1942                                 break;
1943
1944                         case O_GID:
1945                         case O_UID:
1946                                 /*
1947                                  * We only check offset == 0 && proto != 0,
1948                                  * as this ensures that we have an IPv4
1949                                  * packet with the ports info.
1950                                  */
1951                                 if (offset!=0)
1952                                         break;
1953
1954                                 match = ipfw_match_uid(&args->f_id, oif,
1955                                         cmd->opcode,
1956                                         (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1957                                         &deny);
1958                                 if (deny)
1959                                         return IP_FW_DENY;
1960                                 break;
1961
1962                         case O_RECV:
1963                                 match = iface_match(m->m_pkthdr.rcvif,
1964                                     (ipfw_insn_if *)cmd);
1965                                 break;
1966
1967                         case O_XMIT:
1968                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1969                                 break;
1970
1971                         case O_VIA:
1972                                 match = iface_match(oif ? oif :
1973                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1974                                 break;
1975
1976                         case O_MACADDR2:
1977                                 if (args->eh != NULL) { /* have MAC header */
1978                                         uint32_t *want = (uint32_t *)
1979                                                 ((ipfw_insn_mac *)cmd)->addr;
1980                                         uint32_t *mask = (uint32_t *)
1981                                                 ((ipfw_insn_mac *)cmd)->mask;
1982                                         uint32_t *hdr = (uint32_t *)args->eh;
1983
1984                                         match =
1985                                         (want[0] == (hdr[0] & mask[0]) &&
1986                                          want[1] == (hdr[1] & mask[1]) &&
1987                                          want[2] == (hdr[2] & mask[2]));
1988                                 }
1989                                 break;
1990
1991                         case O_MAC_TYPE:
1992                                 if (args->eh != NULL) {
1993                                         uint16_t t =
1994                                             ntohs(args->eh->ether_type);
1995                                         uint16_t *p =
1996                                             ((ipfw_insn_u16 *)cmd)->ports;
1997                                         int i;
1998
1999                                         /* Special vlan handling */
2000                                         if (m->m_flags & M_VLANTAG)
2001                                                 t = ETHERTYPE_VLAN;
2002
2003                                         for (i = cmdlen - 1; !match && i > 0;
2004                                              i--, p += 2) {
2005                                                 match =
2006                                                 (t >= p[0] && t <= p[1]);
2007                                         }
2008                                 }
2009                                 break;
2010
2011                         case O_FRAG:
2012                                 match = (hlen > 0 && offset != 0);
2013                                 break;
2014
2015                         case O_IN:      /* "out" is "not in" */
2016                                 match = (oif == NULL);
2017                                 break;
2018
2019                         case O_LAYER2:
2020                                 match = (args->eh != NULL);
2021                                 break;
2022
2023                         case O_PROTO:
2024                                 /*
2025                                  * We do not allow an arg of 0 so the
2026                                  * check of "proto" only suffices.
2027                                  */
2028                                 match = (proto == cmd->arg1);
2029                                 break;
2030
2031                         case O_IP_SRC:
2032                                 match = (hlen > 0 &&
2033                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2034                                     src_ip.s_addr);
2035                                 break;
2036
2037                         case O_IP_SRC_MASK:
2038                                 match = (hlen > 0 &&
2039                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2040                                      (src_ip.s_addr &
2041                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2042                                 break;
2043
2044                         case O_IP_SRC_ME:
2045                                 if (hlen > 0) {
2046                                         struct ifnet *tif;
2047
2048                                         tif = INADDR_TO_IFP(&src_ip);
2049                                         match = (tif != NULL);
2050                                 }
2051                                 break;
2052
2053                         case O_IP_DST_SET:
2054                         case O_IP_SRC_SET:
2055                                 if (hlen > 0) {
2056                                         uint32_t *d = (uint32_t *)(cmd + 1);
2057                                         uint32_t addr =
2058                                             cmd->opcode == O_IP_DST_SET ?
2059                                                 args->f_id.dst_ip :
2060                                                 args->f_id.src_ip;
2061
2062                                         if (addr < d[0])
2063                                                 break;
2064                                         addr -= d[0]; /* subtract base */
2065                                         match =
2066                                         (addr < cmd->arg1) &&
2067                                          (d[1 + (addr >> 5)] &
2068                                           (1 << (addr & 0x1f)));
2069                                 }
2070                                 break;
2071
2072                         case O_IP_DST:
2073                                 match = (hlen > 0 &&
2074                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2075                                     dst_ip.s_addr);
2076                                 break;
2077
2078                         case O_IP_DST_MASK:
2079                                 match = (hlen > 0) &&
2080                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2081                                      (dst_ip.s_addr &
2082                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2083                                 break;
2084
2085                         case O_IP_DST_ME:
2086                                 if (hlen > 0) {
2087                                         struct ifnet *tif;
2088
2089                                         tif = INADDR_TO_IFP(&dst_ip);
2090                                         match = (tif != NULL);
2091                                 }
2092                                 break;
2093
2094                         case O_IP_SRCPORT:
2095                         case O_IP_DSTPORT:
2096                                 /*
2097                                  * offset == 0 && proto != 0 is enough
2098                                  * to guarantee that we have an IPv4
2099                                  * packet with port info.
2100                                  */
2101                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2102                                     && offset == 0) {
2103                                         uint16_t x =
2104                                             (cmd->opcode == O_IP_SRCPORT) ?
2105                                                 src_port : dst_port ;
2106                                         uint16_t *p =
2107                                             ((ipfw_insn_u16 *)cmd)->ports;
2108                                         int i;
2109
2110                                         for (i = cmdlen - 1; !match && i > 0;
2111                                              i--, p += 2) {
2112                                                 match =
2113                                                 (x >= p[0] && x <= p[1]);
2114                                         }
2115                                 }
2116                                 break;
2117
2118                         case O_ICMPTYPE:
2119                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
2120                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2121                                 break;
2122
2123                         case O_IPOPT:
2124                                 match = (hlen > 0 && ipopts_match(ip, cmd));
2125                                 break;
2126
2127                         case O_IPVER:
2128                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2129                                 break;
2130
2131                         case O_IPTTL:
2132                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2133                                 break;
2134
2135                         case O_IPID:
2136                                 match = (hlen > 0 &&
2137                                     cmd->arg1 == ntohs(ip->ip_id));
2138                                 break;
2139
2140                         case O_IPLEN:
2141                                 match = (hlen > 0 && cmd->arg1 == ip_len);
2142                                 break;
2143
2144                         case O_IPPRECEDENCE:
2145                                 match = (hlen > 0 &&
2146                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
2147                                 break;
2148
2149                         case O_IPTOS:
2150                                 match = (hlen > 0 &&
2151                                     flags_match(cmd, ip->ip_tos));
2152                                 break;
2153
2154                         case O_TCPFLAGS:
2155                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2156                                     flags_match(cmd,
2157                                         L3HDR(struct tcphdr,ip)->th_flags));
2158                                 break;
2159
2160                         case O_TCPOPTS:
2161                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2162                                     tcpopts_match(ip, cmd));
2163                                 break;
2164
2165                         case O_TCPSEQ:
2166                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2167                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2168                                         L3HDR(struct tcphdr,ip)->th_seq);
2169                                 break;
2170
2171                         case O_TCPACK:
2172                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2173                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2174                                         L3HDR(struct tcphdr,ip)->th_ack);
2175                                 break;
2176
2177                         case O_TCPWIN:
2178                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2179                                     cmd->arg1 ==
2180                                         L3HDR(struct tcphdr,ip)->th_win);
2181                                 break;
2182
2183                         case O_ESTAB:
2184                                 /* reject packets which have SYN only */
2185                                 /* XXX should i also check for TH_ACK ? */
2186                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2187                                     (L3HDR(struct tcphdr,ip)->th_flags &
2188                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2189                                 break;
2190
2191                         case O_LOG:
2192                                 if (fw_verbose)
2193                                         ipfw_log(f, hlen, args->eh, m, oif);
2194                                 match = 1;
2195                                 break;
2196
2197                         case O_PROB:
2198                                 match = (krandom() <
2199                                         ((ipfw_insn_u32 *)cmd)->d[0]);
2200                                 break;
2201
2202                         /*
2203                          * The second set of opcodes represents 'actions',
2204                          * i.e. the terminal part of a rule once the packet
2205                          * matches all previous patterns.
2206                          * Typically there is only one action for each rule,
2207                          * and the opcode is stored at the end of the rule
2208                          * (but there are exceptions -- see below).
2209                          *
2210                          * In general, here we set retval and terminate the
2211                          * outer loop (would be a 'break 3' in some language,
2212                          * but we need to do a 'goto done').
2213                          *
2214                          * Exceptions:
2215                          * O_COUNT and O_SKIPTO actions:
2216                          *   instead of terminating, we jump to the next rule
2217                          *   ('goto next_rule', equivalent to a 'break 2'),
2218                          *   or to the SKIPTO target ('goto again' after
2219                          *   having set f, cmd and l), respectively.
2220                          *
2221                          * O_LIMIT and O_KEEP_STATE: these opcodes are
2222                          *   not real 'actions', and are stored right
2223                          *   before the 'action' part of the rule.
2224                          *   These opcodes try to install an entry in the
2225                          *   state tables; if successful, we continue with
2226                          *   the next opcode (match=1; break;), otherwise
2227                          *   the packet must be dropped ('goto done' after
2228                          *   setting retval).  If static rules are changed
2229                          *   during the state installation, the packet will
2230                          *   be dropped and rule's stats will not beupdated
2231                          *   ('return IP_FW_DENY').
2232                          *
2233                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2234                          *   cause a lookup of the state table, and a jump
2235                          *   to the 'action' part of the parent rule
2236                          *   ('goto check_body') if an entry is found, or
2237                          *   (CHECK_STATE only) a jump to the next rule if
2238                          *   the entry is not found ('goto next_rule').
2239                          *   The result of the lookup is cached to make
2240                          *   further instances of these opcodes are
2241                          *   effectively NOPs.  If static rules are changed
2242                          *   during the state looking up, the packet will
2243                          *   be dropped and rule's stats will not be updated
2244                          *   ('return IP_FW_DENY').
2245                          */
2246                         case O_LIMIT:
2247                         case O_KEEP_STATE:
2248                                 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2249                                         kprintf("%s rule (%d) is not ready "
2250                                                 "on cpu%d\n",
2251                                                 cmd->opcode == O_LIMIT ?
2252                                                 "limit" : "keep state",
2253                                                 f->rulenum, f->cpuid);
2254                                         goto next_rule;
2255                                 }
2256                                 if (install_state(f,
2257                                     (ipfw_insn_limit *)cmd, args, &deny)) {
2258                                         if (deny)
2259                                                 return IP_FW_DENY;
2260
2261                                         retval = IP_FW_DENY;
2262                                         goto done; /* error/limit violation */
2263                                 }
2264                                 if (deny)
2265                                         return IP_FW_DENY;
2266                                 match = 1;
2267                                 break;
2268
2269                         case O_PROBE_STATE:
2270                         case O_CHECK_STATE:
2271                                 /*
2272                                  * dynamic rules are checked at the first
2273                                  * keep-state or check-state occurrence,
2274                                  * with the result being stored in dyn_dir.
2275                                  * The compiler introduces a PROBE_STATE
2276                                  * instruction for us when we have a
2277                                  * KEEP_STATE (because PROBE_STATE needs
2278                                  * to be run first).
2279                                  */
2280                                 if (dyn_dir == MATCH_UNKNOWN) {
2281                                         dyn_f = lookup_rule(&args->f_id,
2282                                                 &dyn_dir,
2283                                                 proto == IPPROTO_TCP ?
2284                                                 L3HDR(struct tcphdr, ip) : NULL,
2285                                                 ip_len, &deny);
2286                                         if (deny)
2287                                                 return IP_FW_DENY;
2288                                         if (dyn_f != NULL) {
2289                                                 /*
2290                                                  * Found a rule from a dynamic
2291                                                  * entry; jump to the 'action'
2292                                                  * part of the rule.
2293                                                  */
2294                                                 f = dyn_f;
2295                                                 cmd = ACTION_PTR(f);
2296                                                 l = f->cmd_len - f->act_ofs;
2297                                                 goto check_body;
2298                                         }
2299                                 }
2300                                 /*
2301                                  * Dynamic entry not found. If CHECK_STATE,
2302                                  * skip to next rule, if PROBE_STATE just
2303                                  * ignore and continue with next opcode.
2304                                  */
2305                                 if (cmd->opcode == O_CHECK_STATE)
2306                                         goto next_rule;
2307                                 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2308                                         goto next_rule; /* not ready yet */
2309                                 match = 1;
2310                                 break;
2311
2312                         case O_ACCEPT:
2313                                 retval = IP_FW_PASS;    /* accept */
2314                                 goto done;
2315
2316                         case O_PIPE:
2317                         case O_QUEUE:
2318                                 args->rule = f; /* report matching rule */
2319                                 args->cookie = cmd->arg1;
2320                                 retval = IP_FW_DUMMYNET;
2321                                 goto done;
2322
2323                         case O_DIVERT:
2324                         case O_TEE:
2325                                 if (args->eh) /* not on layer 2 */
2326                                         break;
2327
2328                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2329                                                  sizeof(*divinfo), MB_DONTWAIT);
2330                                 if (mtag == NULL) {
2331                                         retval = IP_FW_DENY;
2332                                         goto done;
2333                                 }
2334                                 divinfo = m_tag_data(mtag);
2335
2336                                 divinfo->skipto = f->rulenum;
2337                                 divinfo->port = cmd->arg1;
2338                                 divinfo->tee = (cmd->opcode == O_TEE);
2339                                 m_tag_prepend(m, mtag);
2340
2341                                 args->cookie = cmd->arg1;
2342                                 retval = (cmd->opcode == O_DIVERT) ?
2343                                          IP_FW_DIVERT : IP_FW_TEE;
2344                                 goto done;
2345
2346                         case O_COUNT:
2347                         case O_SKIPTO:
2348                                 f->pcnt++;      /* update stats */
2349                                 f->bcnt += ip_len;
2350                                 f->timestamp = time_second;
2351                                 if (cmd->opcode == O_COUNT)
2352                                         goto next_rule;
2353                                 /* handle skipto */
2354                                 if (f->next_rule == NULL)
2355                                         lookup_next_rule(f);
2356                                 f = f->next_rule;
2357                                 goto again;
2358
2359                         case O_REJECT:
2360                                 /*
2361                                  * Drop the packet and send a reject notice
2362                                  * if the packet is not ICMP (or is an ICMP
2363                                  * query), and it is not multicast/broadcast.
2364                                  */
2365                                 if (hlen > 0 &&
2366                                     (proto != IPPROTO_ICMP ||
2367                                      is_icmp_query(ip)) &&
2368                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
2369                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2370                                         /*
2371                                          * Update statistics before the possible
2372                                          * blocking 'send_reject'
2373                                          */
2374                                         f->pcnt++;
2375                                         f->bcnt += ip_len;
2376                                         f->timestamp = time_second;
2377
2378                                         send_reject(args, cmd->arg1,
2379                                             offset,ip_len);
2380                                         m = args->m;
2381
2382                                         /*
2383                                          * Return directly here, rule stats
2384                                          * have been updated above.
2385                                          */
2386                                         return IP_FW_DENY;
2387                                 }
2388                                 /* FALLTHROUGH */
2389                         case O_DENY:
2390                                 retval = IP_FW_DENY;
2391                                 goto done;
2392
2393                         case O_FORWARD_IP:
2394                                 if (args->eh)   /* not valid on layer2 pkts */
2395                                         break;
2396                                 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2397                                         struct sockaddr_in *sin;
2398
2399                                         mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2400                                                sizeof(*sin), MB_DONTWAIT);
2401                                         if (mtag == NULL) {
2402                                                 retval = IP_FW_DENY;
2403                                                 goto done;
2404                                         }
2405                                         sin = m_tag_data(mtag);
2406
2407                                         /* Structure copy */
2408                                         *sin = ((ipfw_insn_sa *)cmd)->sa;
2409
2410                                         m_tag_prepend(m, mtag);
2411                                         m->m_pkthdr.fw_flags |=
2412                                                 IPFORWARD_MBUF_TAGGED;
2413                                         m->m_pkthdr.fw_flags &=
2414                                                 ~BRIDGE_MBUF_TAGGED;
2415                                 }
2416                                 retval = IP_FW_PASS;
2417                                 goto done;
2418
2419                         default:
2420                                 panic("-- unknown opcode %d", cmd->opcode);
2421                         } /* end of switch() on opcodes */
2422
2423                         if (cmd->len & F_NOT)
2424                                 match = !match;
2425
2426                         if (match) {
2427                                 if (cmd->len & F_OR)
2428                                         skip_or = 1;
2429                         } else {
2430                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2431                                         break;          /* try next rule    */
2432                         }
2433
2434                 }       /* end of inner for, scan opcodes */
2435
2436 next_rule:;             /* try next rule                */
2437
2438         }               /* end of outer for, scan rules */
2439         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2440         return IP_FW_DENY;
2441
2442 done:
2443         /* Update statistics */
2444         f->pcnt++;
2445         f->bcnt += ip_len;
2446         f->timestamp = time_second;
2447         return retval;
2448
2449 pullup_failed:
2450         if (fw_verbose)
2451                 kprintf("pullup failed\n");
2452         return IP_FW_DENY;
2453 }
2454
2455 static void
2456 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2457 {
2458         struct m_tag *mtag;
2459         struct dn_pkt *pkt;
2460         ipfw_insn *cmd;
2461         const struct ipfw_flow_id *id;
2462         struct dn_flow_id *fid;
2463
2464         M_ASSERTPKTHDR(m);
2465
2466         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2467         if (mtag == NULL) {
2468                 m_freem(m);
2469                 return;
2470         }
2471         m_tag_prepend(m, mtag);
2472
2473         pkt = m_tag_data(mtag);
2474         bzero(pkt, sizeof(*pkt));
2475
2476         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2477         if (cmd->opcode == O_LOG)
2478                 cmd += F_LEN(cmd);
2479         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2480                 ("Rule is not PIPE or QUEUE, opcode %d", cmd->opcode));
2481
2482         pkt->dn_m = m;
2483         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2484         pkt->ifp = fwa->oif;
2485         pkt->pipe_nr = pipe_nr;
2486
2487         pkt->cpuid = mycpuid;
2488         pkt->msgport = cur_netport();
2489
2490         id = &fwa->f_id;
2491         fid = &pkt->id;
2492         fid->fid_dst_ip = id->dst_ip;
2493         fid->fid_src_ip = id->src_ip;
2494         fid->fid_dst_port = id->dst_port;
2495         fid->fid_src_port = id->src_port;
2496         fid->fid_proto = id->proto;
2497         fid->fid_flags = id->flags;
2498
2499         ipfw_ref_rule(fwa->rule);
2500         pkt->dn_priv = fwa->rule;
2501         pkt->dn_unref_priv = ipfw_unref_rule;
2502
2503         if (cmd->opcode == O_PIPE)
2504                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2505
2506         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2507 }
2508
2509 /*
2510  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2511  * These will be reconstructed on the fly as packets are matched.
2512  * Must be called at splimp().
2513  */
2514 static void
2515 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2516 {
2517         struct ip_fw *rule;
2518
2519         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2520                 rule->next_rule = NULL;
2521 }
2522
2523 static __inline void
2524 ipfw_inc_static_count(struct ip_fw *rule)
2525 {
2526         /* Static rule's counts are updated only on CPU0 */
2527         KKASSERT(mycpuid == 0);
2528
2529         static_count++;
2530         static_ioc_len += IOC_RULESIZE(rule);
2531 }
2532
2533 static __inline void
2534 ipfw_dec_static_count(struct ip_fw *rule)
2535 {
2536         int l = IOC_RULESIZE(rule);
2537
2538         /* Static rule's counts are updated only on CPU0 */
2539         KKASSERT(mycpuid == 0);
2540
2541         KASSERT(static_count > 0, ("invalid static count %u", static_count));
2542         static_count--;
2543
2544         KASSERT(static_ioc_len >= l,
2545                 ("invalid static len %u", static_ioc_len));
2546         static_ioc_len -= l;
2547 }
2548
2549 static void
2550 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2551 {
2552         if (fwmsg->sibling != NULL) {
2553                 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2554                 fwmsg->sibling->sibling = rule;
2555         }
2556         fwmsg->sibling = rule;
2557 }
2558
2559 static struct ip_fw *
2560 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2561 {
2562         struct ip_fw *rule;
2563
2564         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2565
2566         rule->act_ofs = ioc_rule->act_ofs;
2567         rule->cmd_len = ioc_rule->cmd_len;
2568         rule->rulenum = ioc_rule->rulenum;
2569         rule->set = ioc_rule->set;
2570         rule->usr_flags = ioc_rule->usr_flags;
2571
2572         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2573
2574         rule->refcnt = 1;
2575         rule->cpuid = mycpuid;
2576
2577         rule->stub = stub;
2578         if (stub != NULL)
2579                 stub->rule[mycpuid] = rule;
2580
2581         return rule;
2582 }
2583
2584 static void
2585 ipfw_add_rule_dispatch(netmsg_t nmsg)
2586 {
2587         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2588         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2589         struct ip_fw *rule;
2590
2591         rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2592
2593         /*
2594          * Bump generation after ipfw_create_rule(),
2595          * since this function is blocking
2596          */
2597         ctx->ipfw_gen++;
2598
2599         /*
2600          * Insert rule into the pre-determined position
2601          */
2602         if (fwmsg->prev_rule != NULL) {
2603                 struct ip_fw *prev, *next;
2604
2605                 prev = fwmsg->prev_rule;
2606                 KKASSERT(prev->cpuid == mycpuid);
2607
2608                 next = fwmsg->next_rule;
2609                 KKASSERT(next->cpuid == mycpuid);
2610
2611                 rule->next = next;
2612                 prev->next = rule;
2613
2614                 /*
2615                  * Move to the position on the next CPU
2616                  * before the msg is forwarded.
2617                  */
2618                 fwmsg->prev_rule = prev->sibling;
2619                 fwmsg->next_rule = next->sibling;
2620         } else {
2621                 KKASSERT(fwmsg->next_rule == NULL);
2622                 rule->next = ctx->ipfw_layer3_chain;
2623                 ctx->ipfw_layer3_chain = rule;
2624         }
2625
2626         /* Link rule CPU sibling */
2627         ipfw_link_sibling(fwmsg, rule);
2628
2629         ipfw_flush_rule_ptrs(ctx);
2630
2631         if (mycpuid == 0) {
2632                 /* Statistics only need to be updated once */
2633                 ipfw_inc_static_count(rule);
2634
2635                 /* Return the rule on CPU0 */
2636                 nmsg->lmsg.u.ms_resultp = rule;
2637         }
2638
2639         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2640 }
2641
2642 static void
2643 ipfw_enable_state_dispatch(netmsg_t nmsg)
2644 {
2645         struct lwkt_msg *lmsg = &nmsg->lmsg;
2646         struct ip_fw *rule = lmsg->u.ms_resultp;
2647         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2648
2649         ctx->ipfw_gen++;
2650
2651         KKASSERT(rule->cpuid == mycpuid);
2652         KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule);
2653         KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE));
2654         rule->rule_flags |= IPFW_RULE_F_STATE;
2655         lmsg->u.ms_resultp = rule->sibling;
2656
2657         ifnet_forwardmsg(lmsg, mycpuid + 1);
2658 }
2659
2660 /*
2661  * Add a new rule to the list.  Copy the rule into a malloc'ed area,
2662  * then possibly create a rule number and add the rule to the list.
2663  * Update the rule_number in the input struct so the caller knows
2664  * it as well.
2665  */
2666 static void
2667 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
2668 {
2669         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2670         struct netmsg_ipfw fwmsg;
2671         struct netmsg_base *nmsg;
2672         struct ip_fw *f, *prev, *rule;
2673         struct ip_fw_stub *stub;
2674
2675         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2676
2677         /*
2678          * If rulenum is 0, find highest numbered rule before the
2679          * default rule, and add rule number incremental step.
2680          */
2681         if (ioc_rule->rulenum == 0) {
2682                 int step = autoinc_step;
2683
2684                 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2685                          step <= IPFW_AUTOINC_STEP_MAX);
2686
2687                 /*
2688                  * Locate the highest numbered rule before default
2689                  */
2690                 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2691                         if (f->rulenum == IPFW_DEFAULT_RULE)
2692                                 break;
2693                         ioc_rule->rulenum = f->rulenum;
2694                 }
2695                 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2696                         ioc_rule->rulenum += step;
2697         }
2698         KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2699                 ioc_rule->rulenum != 0,
2700                 ("invalid rule num %d", ioc_rule->rulenum));
2701
2702         /*
2703          * Now find the right place for the new rule in the sorted list.
2704          */
2705         for (prev = NULL, f = ctx->ipfw_layer3_chain; f;
2706              prev = f, f = f->next) {
2707                 if (f->rulenum > ioc_rule->rulenum) {
2708                         /* Found the location */
2709                         break;
2710                 }
2711         }
2712         KASSERT(f != NULL, ("no default rule?!"));
2713
2714         if (rule_flags & IPFW_RULE_F_STATE) {
2715                 int size;
2716
2717                 /*
2718                  * If the new rule will create states, then allocate
2719                  * a rule stub, which will be referenced by states
2720                  * (dyn rules)
2721                  */
2722                 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2723                 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2724         } else {
2725                 stub = NULL;
2726         }
2727
2728         /*
2729          * Duplicate the rule onto each CPU.
2730          * The rule duplicated on CPU0 will be returned.
2731          */
2732         bzero(&fwmsg, sizeof(fwmsg));
2733         nmsg = &fwmsg.base;
2734         netmsg_init(nmsg, NULL, &curthread->td_msgport,
2735                     0, ipfw_add_rule_dispatch);
2736         fwmsg.ioc_rule = ioc_rule;
2737         fwmsg.prev_rule = prev;
2738         fwmsg.next_rule = prev == NULL ? NULL : f;
2739         fwmsg.stub = stub;
2740
2741         ifnet_domsg(&nmsg->lmsg, 0);
2742         KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2743
2744         rule = nmsg->lmsg.u.ms_resultp;
2745         KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2746
2747         if (rule_flags & IPFW_RULE_F_STATE) {
2748                 /*
2749                  * Turn on state flag, _after_ everything on all
2750                  * CPUs have been setup.
2751                  */
2752                 bzero(nmsg, sizeof(*nmsg));
2753                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
2754                             0, ipfw_enable_state_dispatch);
2755                 nmsg->lmsg.u.ms_resultp = rule;
2756
2757                 ifnet_domsg(&nmsg->lmsg, 0);
2758                 KKASSERT(nmsg->lmsg.u.ms_resultp == NULL);
2759         }
2760
2761         DPRINTF("++ installed rule %d, static count now %d\n",
2762                 rule->rulenum, static_count);
2763 }
2764
2765 /**
2766  * Free storage associated with a static rule (including derived
2767  * dynamic rules).
2768  * The caller is in charge of clearing rule pointers to avoid
2769  * dangling pointers.
2770  * @return a pointer to the next entry.
2771  * Arguments are not checked, so they better be correct.
2772  * Must be called at splimp().
2773  */
2774 static struct ip_fw *
2775 ipfw_delete_rule(struct ipfw_context *ctx,
2776                  struct ip_fw *prev, struct ip_fw *rule)
2777 {
2778         struct ip_fw *n;
2779         struct ip_fw_stub *stub;
2780
2781         ctx->ipfw_gen++;
2782
2783         /* STATE flag should have been cleared before we reach here */
2784         KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2785
2786         stub = rule->stub;
2787         n = rule->next;
2788         if (prev == NULL)
2789                 ctx->ipfw_layer3_chain = n;
2790         else
2791                 prev->next = n;
2792
2793         /* Mark the rule as invalid */
2794         rule->rule_flags |= IPFW_RULE_F_INVALID;
2795         rule->next_rule = NULL;
2796         rule->sibling = NULL;
2797         rule->stub = NULL;
2798 #ifdef foo
2799         /* Don't reset cpuid here; keep various assertion working */
2800         rule->cpuid = -1;
2801 #endif
2802
2803         /* Statistics only need to be updated once */
2804         if (mycpuid == 0)
2805                 ipfw_dec_static_count(rule);
2806
2807         /* Free 'stub' on the last CPU */
2808         if (stub != NULL && mycpuid == ncpus - 1)
2809                 kfree(stub, M_IPFW);
2810
2811         /* Try to free this rule */
2812         ipfw_free_rule(rule);
2813
2814         /* Return the next rule */
2815         return n;
2816 }
2817
2818 static void
2819 ipfw_flush_dispatch(netmsg_t nmsg)
2820 {
2821         struct lwkt_msg *lmsg = &nmsg->lmsg;
2822         int kill_default = lmsg->u.ms_result;
2823         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2824         struct ip_fw *rule;
2825
2826         ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2827
2828         while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2829                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2830                 ipfw_delete_rule(ctx, NULL, rule);
2831
2832         ifnet_forwardmsg(lmsg, mycpuid + 1);
2833 }
2834
2835 static void
2836 ipfw_disable_rule_state_dispatch(netmsg_t nmsg)
2837 {
2838         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2839         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2840         struct ip_fw *rule;
2841
2842         ctx->ipfw_gen++;
2843
2844         rule = dmsg->start_rule;
2845         if (rule != NULL) {
2846                 KKASSERT(rule->cpuid == mycpuid);
2847
2848                 /*
2849                  * Move to the position on the next CPU
2850                  * before the msg is forwarded.
2851                  */
2852                 dmsg->start_rule = rule->sibling;
2853         } else {
2854                 KKASSERT(dmsg->rulenum == 0);
2855                 rule = ctx->ipfw_layer3_chain;
2856         }
2857
2858         while (rule != NULL) {
2859                 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2860                         break;
2861                 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2862                 rule = rule->next;
2863         }
2864
2865         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2866 }
2867
2868 /*
2869  * Deletes all rules from a chain (including the default rule
2870  * if the second argument is set).
2871  * Must be called at splimp().
2872  */
2873 static void
2874 ipfw_flush(int kill_default)
2875 {
2876         struct netmsg_del dmsg;
2877         struct netmsg_base nmsg;
2878         struct lwkt_msg *lmsg;
2879         struct ip_fw *rule;
2880         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2881
2882         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2883
2884         /*
2885          * If 'kill_default' then caller has done the necessary
2886          * msgport syncing; unnecessary to do it again.
2887          */
2888         if (!kill_default) {
2889                 /*
2890                  * Let ipfw_chk() know the rules are going to
2891                  * be flushed, so it could jump directly to
2892                  * the default rule.
2893                  */
2894                 ipfw_flushing = 1;
2895                 netmsg_service_sync();
2896         }
2897
2898         /*
2899          * Clear STATE flag on rules, so no more states (dyn rules)
2900          * will be created.
2901          */
2902         bzero(&dmsg, sizeof(dmsg));
2903         netmsg_init(&dmsg.base, NULL, &curthread->td_msgport,
2904                     0, ipfw_disable_rule_state_dispatch);
2905         ifnet_domsg(&dmsg.base.lmsg, 0);
2906
2907         /*
2908          * This actually nukes all states (dyn rules)
2909          */
2910         lockmgr(&dyn_lock, LK_EXCLUSIVE);
2911         for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2912                 /*
2913                  * Can't check IPFW_RULE_F_STATE here,
2914                  * since it has been cleared previously.
2915                  * Check 'stub' instead.
2916                  */
2917                 if (rule->stub != NULL) {
2918                         /* Force removal */
2919                         remove_dyn_rule_locked(rule, NULL);
2920                 }
2921         }
2922         lockmgr(&dyn_lock, LK_RELEASE);
2923
2924         /*
2925          * Press the 'flush' button
2926          */
2927         bzero(&nmsg, sizeof(nmsg));
2928         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
2929                     0, ipfw_flush_dispatch);
2930         lmsg = &nmsg.lmsg;
2931         lmsg->u.ms_result = kill_default;
2932         ifnet_domsg(lmsg, 0);
2933
2934         KASSERT(dyn_count == 0, ("%u dyn rule remains", dyn_count));
2935
2936         if (kill_default) {
2937                 if (ipfw_dyn_v != NULL) {
2938                         /*
2939                          * Free dynamic rules(state) hash table
2940                          */
2941                         kfree(ipfw_dyn_v, M_IPFW);
2942                         ipfw_dyn_v = NULL;
2943                 }
2944
2945                 KASSERT(static_count == 0,
2946                         ("%u static rules remain", static_count));
2947                 KASSERT(static_ioc_len == 0,
2948                         ("%u bytes of static rules remain", static_ioc_len));
2949         } else {
2950                 KASSERT(static_count == 1,
2951                         ("%u static rules remain", static_count));
2952                 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2953                         ("%u bytes of static rules remain, should be %lu",
2954                          static_ioc_len,
2955                          (u_long)IOC_RULESIZE(ctx->ipfw_default_rule)));
2956         }
2957
2958         /* Flush is done */
2959         ipfw_flushing = 0;
2960 }
2961
2962 static void
2963 ipfw_alt_delete_rule_dispatch(netmsg_t nmsg)
2964 {
2965         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2966         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2967         struct ip_fw *rule, *prev;
2968
2969         rule = dmsg->start_rule;
2970         KKASSERT(rule->cpuid == mycpuid);
2971         dmsg->start_rule = rule->sibling;
2972
2973         prev = dmsg->prev_rule;
2974         if (prev != NULL) {
2975                 KKASSERT(prev->cpuid == mycpuid);
2976
2977                 /*
2978                  * Move to the position on the next CPU
2979                  * before the msg is forwarded.
2980                  */
2981                 dmsg->prev_rule = prev->sibling;
2982         }
2983
2984         /*
2985          * flush pointers outside the loop, then delete all matching
2986          * rules.  'prev' remains the same throughout the cycle.
2987          */
2988         ipfw_flush_rule_ptrs(ctx);
2989         while (rule && rule->rulenum == dmsg->rulenum)
2990                 rule = ipfw_delete_rule(ctx, prev, rule);
2991
2992         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2993 }
2994
2995 static int
2996 ipfw_alt_delete_rule(uint16_t rulenum)
2997 {
2998         struct ip_fw *prev, *rule, *f;
2999         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3000         struct netmsg_del dmsg;
3001         struct netmsg_base *nmsg;
3002         int state;
3003
3004         /*
3005          * Locate first rule to delete
3006          */
3007         for (prev = NULL, rule = ctx->ipfw_layer3_chain;
3008              rule && rule->rulenum < rulenum;
3009              prev = rule, rule = rule->next)
3010                 ; /* EMPTY */
3011         if (rule->rulenum != rulenum)
3012                 return EINVAL;
3013
3014         /*
3015          * Check whether any rules with the given number will
3016          * create states.
3017          */
3018         state = 0;
3019         for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3020                 if (f->rule_flags & IPFW_RULE_F_STATE) {
3021                         state = 1;
3022                         break;
3023                 }
3024         }
3025
3026         if (state) {
3027                 /*
3028                  * Clear the STATE flag, so no more states will be
3029                  * created based the rules numbered 'rulenum'.
3030                  */
3031                 bzero(&dmsg, sizeof(dmsg));
3032                 nmsg = &dmsg.base;
3033                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3034                             0, ipfw_disable_rule_state_dispatch);
3035                 dmsg.start_rule = rule;
3036                 dmsg.rulenum = rulenum;
3037
3038                 ifnet_domsg(&nmsg->lmsg, 0);
3039                 KKASSERT(dmsg.start_rule == NULL);
3040
3041                 /*
3042                  * Nuke all related states
3043                  */
3044                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3045                 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3046                         /*
3047                          * Can't check IPFW_RULE_F_STATE here,
3048                          * since it has been cleared previously.
3049                          * Check 'stub' instead.
3050                          */
3051                         if (f->stub != NULL) {
3052                                 /* Force removal */
3053                                 remove_dyn_rule_locked(f, NULL);
3054                         }
3055                 }
3056                 lockmgr(&dyn_lock, LK_RELEASE);
3057         }
3058
3059         /*
3060          * Get rid of the rule duplications on all CPUs
3061          */
3062         bzero(&dmsg, sizeof(dmsg));
3063         nmsg = &dmsg.base;
3064         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3065                     0, ipfw_alt_delete_rule_dispatch);
3066         dmsg.prev_rule = prev;
3067         dmsg.start_rule = rule;
3068         dmsg.rulenum = rulenum;
3069
3070         ifnet_domsg(&nmsg->lmsg, 0);
3071         KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3072         return 0;
3073 }
3074
3075 static void
3076 ipfw_alt_delete_ruleset_dispatch(netmsg_t nmsg)
3077 {
3078         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3079         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3080         struct ip_fw *prev, *rule;
3081 #ifdef INVARIANTS
3082         int del = 0;
3083 #endif
3084
3085         ipfw_flush_rule_ptrs(ctx);
3086
3087         prev = NULL;
3088         rule = ctx->ipfw_layer3_chain;
3089         while (rule != NULL) {
3090                 if (rule->set == dmsg->from_set) {
3091                         rule = ipfw_delete_rule(ctx, prev, rule);
3092 #ifdef INVARIANTS
3093                         del = 1;
3094 #endif
3095                 } else {
3096                         prev = rule;
3097                         rule = rule->next;
3098                 }
3099         }
3100         KASSERT(del, ("no match set?!"));
3101
3102         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3103 }
3104
3105 static void
3106 ipfw_disable_ruleset_state_dispatch(netmsg_t nmsg)
3107 {
3108         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3109         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3110         struct ip_fw *rule;
3111 #ifdef INVARIANTS
3112         int cleared = 0;
3113 #endif
3114
3115         ctx->ipfw_gen++;
3116
3117         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3118                 if (rule->set == dmsg->from_set) {
3119 #ifdef INVARIANTS
3120                         cleared = 1;
3121 #endif
3122                         rule->rule_flags &= ~IPFW_RULE_F_STATE;
3123                 }
3124         }
3125         KASSERT(cleared, ("no match set?!"));
3126
3127         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3128 }
3129
3130 static int
3131 ipfw_alt_delete_ruleset(uint8_t set)
3132 {
3133         struct netmsg_del dmsg;
3134         struct netmsg_base *nmsg;
3135         int state, del;
3136         struct ip_fw *rule;
3137         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3138
3139         /*
3140          * Check whether the 'set' exists.  If it exists,
3141          * then check whether any rules within the set will
3142          * try to create states.
3143          */
3144         state = 0;
3145         del = 0;
3146         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3147                 if (rule->set == set) {
3148                         del = 1;
3149                         if (rule->rule_flags & IPFW_RULE_F_STATE) {
3150                                 state = 1;
3151                                 break;
3152                         }
3153                 }
3154         }
3155         if (!del)
3156                 return 0; /* XXX EINVAL? */
3157
3158         if (state) {
3159                 /*
3160                  * Clear the STATE flag, so no more states will be
3161                  * created based the rules in this set.
3162                  */
3163                 bzero(&dmsg, sizeof(dmsg));
3164                 nmsg = &dmsg.base;
3165                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3166                             0, ipfw_disable_ruleset_state_dispatch);
3167                 dmsg.from_set = set;
3168
3169                 ifnet_domsg(&nmsg->lmsg, 0);
3170
3171                 /*
3172                  * Nuke all related states
3173                  */
3174                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3175                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3176                         if (rule->set != set)
3177                                 continue;
3178
3179                         /*
3180                          * Can't check IPFW_RULE_F_STATE here,
3181                          * since it has been cleared previously.
3182                          * Check 'stub' instead.
3183                          */
3184                         if (rule->stub != NULL) {
3185                                 /* Force removal */
3186                                 remove_dyn_rule_locked(rule, NULL);
3187                         }
3188                 }
3189                 lockmgr(&dyn_lock, LK_RELEASE);
3190         }
3191
3192         /*
3193          * Delete this set
3194          */
3195         bzero(&dmsg, sizeof(dmsg));
3196         nmsg = &dmsg.base;
3197         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3198                     0, ipfw_alt_delete_ruleset_dispatch);
3199         dmsg.from_set = set;
3200
3201         ifnet_domsg(&nmsg->lmsg, 0);
3202         return 0;
3203 }
3204
3205 static void
3206 ipfw_alt_move_rule_dispatch(netmsg_t nmsg)
3207 {
3208         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3209         struct ip_fw *rule;
3210
3211         rule = dmsg->start_rule;
3212         KKASSERT(rule->cpuid == mycpuid);
3213
3214         /*
3215          * Move to the position on the next CPU
3216          * before the msg is forwarded.
3217          */
3218         dmsg->start_rule = rule->sibling;
3219
3220         while (rule && rule->rulenum <= dmsg->rulenum) {
3221                 if (rule->rulenum == dmsg->rulenum)
3222                         rule->set = dmsg->to_set;
3223                 rule = rule->next;
3224         }
3225         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3226 }
3227
3228 static int
3229 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3230 {
3231         struct netmsg_del dmsg;
3232         struct netmsg_base *nmsg;
3233         struct ip_fw *rule;
3234         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3235
3236         /*
3237          * Locate first rule to move
3238          */
3239         for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3240              rule = rule->next) {
3241                 if (rule->rulenum == rulenum && rule->set != set)
3242                         break;
3243         }
3244         if (rule == NULL || rule->rulenum > rulenum)
3245                 return 0; /* XXX error? */
3246
3247         bzero(&dmsg, sizeof(dmsg));
3248         nmsg = &dmsg.base;
3249         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3250                     0, ipfw_alt_move_rule_dispatch);
3251         dmsg.start_rule = rule;
3252         dmsg.rulenum = rulenum;
3253         dmsg.to_set = set;
3254
3255         ifnet_domsg(&nmsg->lmsg, 0);
3256         KKASSERT(dmsg.start_rule == NULL);
3257         return 0;
3258 }
3259
3260 static void
3261 ipfw_alt_move_ruleset_dispatch(netmsg_t nmsg)
3262 {
3263         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3264         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3265         struct ip_fw *rule;
3266
3267         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3268                 if (rule->set == dmsg->from_set)
3269                         rule->set = dmsg->to_set;
3270         }
3271         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3272 }
3273
3274 static int
3275 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3276 {
3277         struct netmsg_del dmsg;
3278         struct netmsg_base *nmsg;
3279
3280         bzero(&dmsg, sizeof(dmsg));
3281         nmsg = &dmsg.base;
3282         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3283                     0, ipfw_alt_move_ruleset_dispatch);
3284         dmsg.from_set = from_set;
3285         dmsg.to_set = to_set;
3286
3287         ifnet_domsg(&nmsg->lmsg, 0);
3288         return 0;
3289 }
3290
3291 static void
3292 ipfw_alt_swap_ruleset_dispatch(netmsg_t nmsg)
3293 {
3294         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3295         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3296         struct ip_fw *rule;
3297
3298         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3299                 if (rule->set == dmsg->from_set)
3300                         rule->set = dmsg->to_set;
3301                 else if (rule->set == dmsg->to_set)
3302                         rule->set = dmsg->from_set;
3303         }
3304         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3305 }
3306
3307 static int
3308 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3309 {
3310         struct netmsg_del dmsg;
3311         struct netmsg_base *nmsg;
3312
3313         bzero(&dmsg, sizeof(dmsg));
3314         nmsg = &dmsg.base;
3315         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3316                     0, ipfw_alt_swap_ruleset_dispatch);
3317         dmsg.from_set = set1;
3318         dmsg.to_set = set2;
3319
3320         ifnet_domsg(&nmsg->lmsg, 0);
3321         return 0;
3322 }
3323
3324 /**
3325  * Remove all rules with given number, and also do set manipulation.
3326  *
3327  * The argument is an uint32_t. The low 16 bit are the rule or set number,
3328  * the next 8 bits are the new set, the top 8 bits are the command:
3329  *
3330  *      0       delete rules with given number
3331  *      1       delete rules with given set number
3332  *      2       move rules with given number to new set
3333  *      3       move rules with given set number to new set
3334  *      4       swap sets with given numbers
3335  */
3336 static int
3337 ipfw_ctl_alter(uint32_t arg)
3338 {
3339         uint16_t rulenum;
3340         uint8_t cmd, new_set;
3341         int error = 0;
3342
3343         rulenum = arg & 0xffff;
3344         cmd = (arg >> 24) & 0xff;
3345         new_set = (arg >> 16) & 0xff;
3346
3347         if (cmd > 4)
3348                 return EINVAL;
3349         if (new_set >= IPFW_DEFAULT_SET)
3350                 return EINVAL;
3351         if (cmd == 0 || cmd == 2) {
3352                 if (rulenum == IPFW_DEFAULT_RULE)
3353                         return EINVAL;
3354         } else {
3355                 if (rulenum >= IPFW_DEFAULT_SET)
3356                         return EINVAL;
3357         }
3358
3359         switch (cmd) {
3360         case 0: /* delete rules with given number */
3361                 error = ipfw_alt_delete_rule(rulenum);
3362                 break;
3363
3364         case 1: /* delete all rules with given set number */
3365                 error = ipfw_alt_delete_ruleset(rulenum);
3366                 break;
3367
3368         case 2: /* move rules with given number to new set */
3369                 error = ipfw_alt_move_rule(rulenum, new_set);
3370                 break;
3371
3372         case 3: /* move rules with given set number to new set */
3373                 error = ipfw_alt_move_ruleset(rulenum, new_set);
3374                 break;
3375
3376         case 4: /* swap two sets */
3377                 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3378                 break;
3379         }
3380         return error;
3381 }
3382
3383 /*
3384  * Clear counters for a specific rule.
3385  */
3386 static void
3387 clear_counters(struct ip_fw *rule, int log_only)
3388 {
3389         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3390
3391         if (log_only == 0) {
3392                 rule->bcnt = rule->pcnt = 0;
3393                 rule->timestamp = 0;
3394         }
3395         if (l->o.opcode == O_LOG)
3396                 l->log_left = l->max_log;
3397 }
3398
3399 static void
3400 ipfw_zero_entry_dispatch(netmsg_t nmsg)
3401 {
3402         struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3403         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3404         struct ip_fw *rule;
3405
3406         if (zmsg->rulenum == 0) {
3407                 KKASSERT(zmsg->start_rule == NULL);
3408
3409                 ctx->ipfw_norule_counter = 0;
3410                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3411                         clear_counters(rule, zmsg->log_only);
3412         } else {
3413                 struct ip_fw *start = zmsg->start_rule;
3414
3415                 KKASSERT(start->cpuid == mycpuid);
3416                 KKASSERT(start->rulenum == zmsg->rulenum);
3417
3418                 /*
3419                  * We can have multiple rules with the same number, so we
3420                  * need to clear them all.
3421                  */
3422                 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3423                      rule = rule->next)
3424                         clear_counters(rule, zmsg->log_only);
3425
3426                 /*
3427                  * Move to the position on the next CPU
3428                  * before the msg is forwarded.
3429                  */
3430                 zmsg->start_rule = start->sibling;
3431         }
3432         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3433 }
3434
3435 /**
3436  * Reset some or all counters on firewall rules.
3437  * @arg frwl is null to clear all entries, or contains a specific
3438  * rule number.
3439  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3440  */
3441 static int
3442 ipfw_ctl_zero_entry(int rulenum, int log_only)
3443 {
3444         struct netmsg_zent zmsg;
3445         struct netmsg_base *nmsg;
3446         const char *msg;
3447         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3448
3449         bzero(&zmsg, sizeof(zmsg));
3450         nmsg = &zmsg.base;
3451         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3452                     0, ipfw_zero_entry_dispatch);
3453         zmsg.log_only = log_only;
3454
3455         if (rulenum == 0) {
3456                 msg = log_only ? "ipfw: All logging counts reset.\n"
3457                                : "ipfw: Accounting cleared.\n";
3458         } else {
3459                 struct ip_fw *rule;
3460
3461                 /*
3462                  * Locate the first rule with 'rulenum'
3463                  */
3464                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3465                         if (rule->rulenum == rulenum)
3466                                 break;
3467                 }
3468                 if (rule == NULL) /* we did not find any matching rules */
3469                         return (EINVAL);
3470                 zmsg.start_rule = rule;
3471                 zmsg.rulenum = rulenum;
3472
3473                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3474                                : "ipfw: Entry %d cleared.\n";
3475         }
3476         ifnet_domsg(&nmsg->lmsg, 0);
3477         KKASSERT(zmsg.start_rule == NULL);
3478
3479         if (fw_verbose)
3480                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3481         return (0);
3482 }
3483
3484 /*
3485  * Check validity of the structure before insert.
3486  * Fortunately rules are simple, so this mostly need to check rule sizes.
3487  */
3488 static int
3489 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3490 {
3491         int l, cmdlen = 0;
3492         int have_action = 0;
3493         ipfw_insn *cmd;
3494
3495         *rule_flags = 0;
3496
3497         /* Check for valid size */
3498         if (size < sizeof(*rule)) {
3499                 kprintf("ipfw: rule too short\n");
3500                 return EINVAL;
3501         }
3502         l = IOC_RULESIZE(rule);
3503         if (l != size) {
3504                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3505                 return EINVAL;
3506         }
3507
3508         /* Check rule number */
3509         if (rule->rulenum == IPFW_DEFAULT_RULE) {
3510                 kprintf("ipfw: invalid rule number\n");
3511                 return EINVAL;
3512         }
3513
3514         /*
3515          * Now go for the individual checks. Very simple ones, basically only
3516          * instruction sizes.
3517          */
3518         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3519              l -= cmdlen, cmd += cmdlen) {
3520                 cmdlen = F_LEN(cmd);
3521                 if (cmdlen > l) {
3522                         kprintf("ipfw: opcode %d size truncated\n",
3523                                 cmd->opcode);
3524                         return EINVAL;
3525                 }
3526
3527                 DPRINTF("ipfw: opcode %d\n", cmd->opcode);
3528
3529                 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3530                         /* This rule will create states */
3531                         *rule_flags |= IPFW_RULE_F_STATE;
3532                 }
3533
3534                 switch (cmd->opcode) {
3535                 case O_NOP:
3536                 case O_PROBE_STATE:
3537                 case O_KEEP_STATE:
3538                 case O_PROTO:
3539                 case O_IP_SRC_ME:
3540                 case O_IP_DST_ME:
3541                 case O_LAYER2:
3542                 case O_IN:
3543                 case O_FRAG:
3544                 case O_IPOPT:
3545                 case O_IPLEN:
3546                 case O_IPID:
3547                 case O_IPTOS:
3548                 case O_IPPRECEDENCE:
3549                 case O_IPTTL:
3550                 case O_IPVER:
3551                 case O_TCPWIN:
3552                 case O_TCPFLAGS:
3553                 case O_TCPOPTS:
3554                 case O_ESTAB:
3555                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3556                                 goto bad_size;
3557                         break;
3558
3559                 case O_UID:
3560                 case O_GID:
3561                 case O_IP_SRC:
3562                 case O_IP_DST:
3563                 case O_TCPSEQ:
3564                 case O_TCPACK:
3565                 case O_PROB:
3566                 case O_ICMPTYPE:
3567                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3568                                 goto bad_size;
3569                         break;
3570
3571                 case O_LIMIT:
3572                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3573                                 goto bad_size;
3574                         break;
3575
3576                 case O_LOG:
3577                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3578                                 goto bad_size;
3579
3580                         ((ipfw_insn_log *)cmd)->log_left =
3581                             ((ipfw_insn_log *)cmd)->max_log;
3582
3583                         break;
3584
3585                 case O_IP_SRC_MASK:
3586                 case O_IP_DST_MASK:
3587                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3588                                 goto bad_size;
3589                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3590                                 kprintf("ipfw: opcode %d, useless rule\n",
3591                                         cmd->opcode);
3592                                 return EINVAL;
3593                         }
3594                         break;
3595
3596                 case O_IP_SRC_SET:
3597                 case O_IP_DST_SET:
3598                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3599                                 kprintf("ipfw: invalid set size %d\n",
3600                                         cmd->arg1);
3601                                 return EINVAL;
3602                         }
3603                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3604                             (cmd->arg1+31)/32 )
3605                                 goto bad_size;
3606                         break;
3607
3608                 case O_MACADDR2:
3609                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3610                                 goto bad_size;
3611                         break;
3612
3613                 case O_MAC_TYPE:
3614                 case O_IP_SRCPORT:
3615                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3616                         if (cmdlen < 2 || cmdlen > 31)
3617                                 goto bad_size;
3618                         break;
3619
3620                 case O_RECV:
3621                 case O_XMIT:
3622                 case O_VIA:
3623                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3624                                 goto bad_size;
3625                         break;
3626
3627                 case O_PIPE:
3628                 case O_QUEUE:
3629                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3630                                 goto bad_size;
3631                         goto check_action;
3632
3633                 case O_FORWARD_IP:
3634                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3635                                 goto bad_size;
3636                         } else {
3637                                 in_addr_t fwd_addr;
3638
3639                                 fwd_addr = ((ipfw_insn_sa *)cmd)->
3640                                            sa.sin_addr.s_addr;
3641                                 if (IN_MULTICAST(ntohl(fwd_addr))) {
3642                                         kprintf("ipfw: try forwarding to "
3643                                                 "multicast address\n");
3644                                         return EINVAL;
3645                                 }
3646                         }
3647                         goto check_action;
3648
3649                 case O_FORWARD_MAC: /* XXX not implemented yet */
3650                 case O_CHECK_STATE:
3651                 case O_COUNT:
3652                 case O_ACCEPT:
3653                 case O_DENY:
3654                 case O_REJECT:
3655                 case O_SKIPTO:
3656                 case O_DIVERT:
3657                 case O_TEE:
3658                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3659                                 goto bad_size;
3660 check_action:
3661                         if (have_action) {
3662                                 kprintf("ipfw: opcode %d, multiple actions"
3663                                         " not allowed\n",
3664                                         cmd->opcode);
3665                                 return EINVAL;
3666                         }
3667                         have_action = 1;
3668                         if (l != cmdlen) {
3669                                 kprintf("ipfw: opcode %d, action must be"
3670                                         " last opcode\n",
3671                                         cmd->opcode);
3672                                 return EINVAL;
3673                         }
3674                         break;
3675                 default:
3676                         kprintf("ipfw: opcode %d, unknown opcode\n",
3677                                 cmd->opcode);
3678                         return EINVAL;
3679                 }
3680         }
3681         if (have_action == 0) {
3682                 kprintf("ipfw: missing action\n");
3683                 return EINVAL;
3684         }
3685         return 0;
3686
3687 bad_size:
3688         kprintf("ipfw: opcode %d size %d wrong\n",
3689                 cmd->opcode, cmdlen);
3690         return EINVAL;
3691 }
3692
3693 static int
3694 ipfw_ctl_add_rule(struct sockopt *sopt)
3695 {
3696         struct ipfw_ioc_rule *ioc_rule;
3697         size_t size;
3698         uint32_t rule_flags;
3699         int error;
3700
3701         size = sopt->sopt_valsize;
3702         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3703             size < sizeof(*ioc_rule)) {
3704                 return EINVAL;
3705         }
3706         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3707                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3708                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3709         }
3710         ioc_rule = sopt->sopt_val;
3711
3712         error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3713         if (error)
3714                 return error;
3715
3716         ipfw_add_rule(ioc_rule, rule_flags);
3717
3718         if (sopt->sopt_dir == SOPT_GET)
3719                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3720         return 0;
3721 }
3722
3723 static void *
3724 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3725 {
3726         const struct ip_fw *sibling;
3727 #ifdef INVARIANTS
3728         int i;
3729 #endif
3730
3731         KKASSERT(rule->cpuid == IPFW_CFGCPUID);
3732
3733         ioc_rule->act_ofs = rule->act_ofs;
3734         ioc_rule->cmd_len = rule->cmd_len;
3735         ioc_rule->rulenum = rule->rulenum;
3736         ioc_rule->set = rule->set;
3737         ioc_rule->usr_flags = rule->usr_flags;
3738
3739         ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3740         ioc_rule->static_count = static_count;
3741         ioc_rule->static_len = static_ioc_len;
3742
3743         /*
3744          * Visit (read-only) all of the rule's duplications to get
3745          * the necessary statistics
3746          */
3747 #ifdef INVARIANTS
3748         i = 0;
3749 #endif
3750         ioc_rule->pcnt = 0;
3751         ioc_rule->bcnt = 0;
3752         ioc_rule->timestamp = 0;
3753         for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3754                 ioc_rule->pcnt += sibling->pcnt;
3755                 ioc_rule->bcnt += sibling->bcnt;
3756                 if (sibling->timestamp > ioc_rule->timestamp)
3757                         ioc_rule->timestamp = sibling->timestamp;
3758 #ifdef INVARIANTS
3759                 ++i;
3760 #endif
3761         }
3762         KASSERT(i == ncpus, ("static rule is not duplicated on every cpu"));
3763
3764         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3765
3766         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3767 }
3768
3769 static void
3770 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3771                 struct ipfw_ioc_state *ioc_state)
3772 {
3773         const struct ipfw_flow_id *id;
3774         struct ipfw_ioc_flowid *ioc_id;
3775
3776         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3777                             0 : dyn_rule->expire - time_second;
3778         ioc_state->pcnt = dyn_rule->pcnt;
3779         ioc_state->bcnt = dyn_rule->bcnt;
3780
3781         ioc_state->dyn_type = dyn_rule->dyn_type;
3782         ioc_state->count = dyn_rule->count;
3783
3784         ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3785
3786         id = &dyn_rule->id;
3787         ioc_id = &ioc_state->id;
3788
3789         ioc_id->type = ETHERTYPE_IP;
3790         ioc_id->u.ip.dst_ip = id->dst_ip;
3791         ioc_id->u.ip.src_ip = id->src_ip;
3792         ioc_id->u.ip.dst_port = id->dst_port;
3793         ioc_id->u.ip.src_port = id->src_port;
3794         ioc_id->u.ip.proto = id->proto;
3795 }
3796
3797 static int
3798 ipfw_ctl_get_rules(struct sockopt *sopt)
3799 {
3800         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3801         struct ip_fw *rule;
3802         void *bp;
3803         size_t size;
3804         uint32_t dcount = 0;
3805
3806         /*
3807          * pass up a copy of the current rules. Static rules
3808          * come first (the last of which has number IPFW_DEFAULT_RULE),
3809          * followed by a possibly empty list of dynamic rule.
3810          */
3811
3812         size = static_ioc_len;  /* size of static rules */
3813         if (ipfw_dyn_v) {       /* add size of dyn.rules */
3814                 dcount = dyn_count;
3815                 size += dcount * sizeof(struct ipfw_ioc_state);
3816         }
3817
3818         if (sopt->sopt_valsize < size) {
3819                 /* short length, no need to return incomplete rules */
3820                 /* XXX: if superuser, no need to zero buffer */
3821                 bzero(sopt->sopt_val, sopt->sopt_valsize);
3822                 return 0;
3823         }
3824         bp = sopt->sopt_val;
3825
3826         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3827                 bp = ipfw_copy_rule(rule, bp);
3828
3829         if (ipfw_dyn_v && dcount != 0) {
3830                 struct ipfw_ioc_state *ioc_state = bp;
3831                 uint32_t dcount2 = 0;
3832 #ifdef INVARIANTS
3833                 size_t old_size = size;
3834 #endif
3835                 int i;
3836
3837                 lockmgr(&dyn_lock, LK_SHARED);
3838
3839                 /* Check 'ipfw_dyn_v' again with lock held */
3840                 if (ipfw_dyn_v == NULL)
3841                         goto skip;
3842
3843                 for (i = 0; i < curr_dyn_buckets; i++) {
3844                         ipfw_dyn_rule *p;
3845
3846                         /*
3847                          * The # of dynamic rules may have grown after the
3848                          * snapshot of 'dyn_count' was taken, so we will have
3849                          * to check 'dcount' (snapshot of dyn_count) here to
3850                          * make sure that we don't overflow the pre-allocated
3851                          * buffer.
3852                          */
3853                         for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
3854                              p = p->next, ioc_state++, dcount--, dcount2++)
3855                                 ipfw_copy_state(p, ioc_state);
3856                 }
3857 skip:
3858                 lockmgr(&dyn_lock, LK_RELEASE);
3859
3860                 /*
3861                  * The # of dynamic rules may be shrinked after the
3862                  * snapshot of 'dyn_count' was taken.  To give user a
3863                  * correct dynamic rule count, we use the 'dcount2'
3864                  * calculated above (with shared lockmgr lock held).
3865                  */
3866                 size = static_ioc_len +
3867                        (dcount2 * sizeof(struct ipfw_ioc_state));
3868                 KKASSERT(size <= old_size);
3869         }
3870
3871         sopt->sopt_valsize = size;
3872         return 0;
3873 }
3874
3875 static void
3876 ipfw_set_disable_dispatch(netmsg_t nmsg)
3877 {
3878         struct lwkt_msg *lmsg = &nmsg->lmsg;
3879         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3880
3881         ctx->ipfw_gen++;
3882         ctx->ipfw_set_disable = lmsg->u.ms_result32;
3883
3884         ifnet_forwardmsg(lmsg, mycpuid + 1);
3885 }
3886
3887 static void
3888 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable)
3889 {
3890         struct netmsg_base nmsg;
3891         struct lwkt_msg *lmsg;
3892         uint32_t set_disable;
3893
3894         /* IPFW_DEFAULT_SET is always enabled */
3895         enable |= (1 << IPFW_DEFAULT_SET);
3896         set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable;
3897
3898         bzero(&nmsg, sizeof(nmsg));
3899         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
3900                     0, ipfw_set_disable_dispatch);
3901         lmsg = &nmsg.lmsg;
3902         lmsg->u.ms_result32 = set_disable;
3903
3904         ifnet_domsg(lmsg, 0);
3905 }
3906
3907 /**
3908  * {set|get}sockopt parser.
3909  */
3910 static int
3911 ipfw_ctl(struct sockopt *sopt)
3912 {
3913         int error, rulenum;
3914         uint32_t *masks;
3915         size_t size;
3916
3917         error = 0;
3918
3919         switch (sopt->sopt_name) {
3920         case IP_FW_GET:
3921                 error = ipfw_ctl_get_rules(sopt);
3922                 break;
3923
3924         case IP_FW_FLUSH:
3925                 ipfw_flush(0 /* keep default rule */);
3926                 break;
3927
3928         case IP_FW_ADD:
3929                 error = ipfw_ctl_add_rule(sopt);
3930                 break;
3931
3932         case IP_FW_DEL:
3933                 /*
3934                  * IP_FW_DEL is used for deleting single rules or sets,
3935                  * and (ab)used to atomically manipulate sets.
3936                  * Argument size is used to distinguish between the two:
3937                  *    sizeof(uint32_t)
3938                  *      delete single rule or set of rules,
3939                  *      or reassign rules (or sets) to a different set.
3940                  *    2 * sizeof(uint32_t)
3941                  *      atomic disable/enable sets.
3942                  *      first uint32_t contains sets to be disabled,
3943                  *      second uint32_t contains sets to be enabled.
3944                  */
3945                 masks = sopt->sopt_val;
3946                 size = sopt->sopt_valsize;
3947                 if (size == sizeof(*masks)) {
3948                         /*
3949                          * Delete or reassign static rule
3950                          */
3951                         error = ipfw_ctl_alter(masks[0]);
3952                 } else if (size == (2 * sizeof(*masks))) {
3953                         /*
3954                          * Set enable/disable
3955                          */
3956                         ipfw_ctl_set_disable(masks[0], masks[1]);
3957                 } else {
3958                         error = EINVAL;
3959                 }
3960                 break;
3961
3962         case IP_FW_ZERO:
3963         case IP_FW_RESETLOG: /* argument is an int, the rule number */
3964                 rulenum = 0;
3965
3966                 if (sopt->sopt_val != 0) {
3967                     error = soopt_to_kbuf(sopt, &rulenum,
3968                             sizeof(int), sizeof(int));
3969                     if (error)
3970                         break;
3971                 }
3972                 error = ipfw_ctl_zero_entry(rulenum,
3973                         sopt->sopt_name == IP_FW_RESETLOG);
3974                 break;
3975
3976         default:
3977                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3978                 error = EINVAL;
3979         }
3980         return error;
3981 }
3982
3983 /*
3984  * This procedure is only used to handle keepalives. It is invoked
3985  * every dyn_keepalive_period
3986  */
3987 static void
3988 ipfw_tick_dispatch(netmsg_t nmsg)
3989 {
3990         time_t keep_alive;
3991         uint32_t gen;
3992         int i;
3993
3994         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
3995         KKASSERT(IPFW_LOADED);
3996
3997         /* Reply ASAP */
3998         crit_enter();
3999         lwkt_replymsg(&nmsg->lmsg, 0);
4000         crit_exit();
4001
4002         if (ipfw_dyn_v == NULL || dyn_count == 0)
4003                 goto done;
4004
4005         keep_alive = time_second;
4006
4007         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4008 again:
4009         if (ipfw_dyn_v == NULL || dyn_count == 0) {
4010                 lockmgr(&dyn_lock, LK_RELEASE);
4011                 goto done;
4012         }
4013         gen = dyn_buckets_gen;
4014
4015         for (i = 0; i < curr_dyn_buckets; i++) {
4016                 ipfw_dyn_rule *q, *prev;
4017
4018                 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
4019                         uint32_t ack_rev, ack_fwd;
4020                         struct ipfw_flow_id id;
4021
4022                         if (q->dyn_type == O_LIMIT_PARENT)
4023                                 goto next;
4024
4025                         if (TIME_LEQ(q->expire, time_second)) {
4026                                 /* State expired */
4027                                 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
4028                                 continue;
4029                         }
4030
4031                         /*
4032                          * Keep alive processing
4033                          */
4034
4035                         if (!dyn_keepalive)
4036                                 goto next;
4037                         if (q->id.proto != IPPROTO_TCP)
4038                                 goto next;
4039                         if ((q->state & BOTH_SYN) != BOTH_SYN)
4040                                 goto next;
4041                         if (TIME_LEQ(time_second + dyn_keepalive_interval,
4042                             q->expire))
4043                                 goto next;      /* too early */
4044                         if (q->keep_alive == keep_alive)
4045                                 goto next;      /* alreay done */
4046
4047                         /*
4048                          * Save necessary information, so that they could
4049                          * survive after possible blocking in send_pkt()
4050                          */
4051                         id = q->id;
4052                         ack_rev = q->ack_rev;
4053                         ack_fwd = q->ack_fwd;
4054
4055                         /* Sending has been started */
4056                         q->keep_alive = keep_alive;
4057
4058                         /* Release lock to avoid possible dead lock */
4059                         lockmgr(&dyn_lock, LK_RELEASE);
4060                         send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
4061                         send_pkt(&id, ack_fwd - 1, ack_rev, 0);
4062                         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4063
4064                         if (gen != dyn_buckets_gen) {
4065                                 /*
4066                                  * Dyn bucket array has been changed during
4067                                  * the above two sending; reiterate.
4068                                  */
4069                                 goto again;
4070                         }
4071 next:
4072                         prev = q;
4073                         q = q->next;
4074                 }
4075         }
4076         lockmgr(&dyn_lock, LK_RELEASE);
4077 done:
4078         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
4079                       ipfw_tick, NULL);
4080 }
4081
4082 /*
4083  * This procedure is only used to handle keepalives. It is invoked
4084  * every dyn_keepalive_period
4085  */
4086 static void
4087 ipfw_tick(void *dummy __unused)
4088 {
4089         struct lwkt_msg *lmsg = &ipfw_timeout_netmsg.lmsg;
4090
4091         KKASSERT(mycpuid == IPFW_CFGCPUID);
4092
4093         crit_enter();
4094
4095         KKASSERT(lmsg->ms_flags & MSGF_DONE);
4096         if (IPFW_LOADED) {
4097                 lwkt_sendmsg(IPFW_CFGPORT, lmsg);
4098                 /* ipfw_timeout_netmsg's handler reset this callout */
4099         }
4100
4101         crit_exit();
4102 }
4103
4104 static int
4105 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4106 {
4107         struct ip_fw_args args;
4108         struct mbuf *m = *m0;
4109         struct m_tag *mtag;
4110         int tee = 0, error = 0, ret;
4111
4112         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4113                 /* Extract info from dummynet tag */
4114                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4115                 KKASSERT(mtag != NULL);
4116                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4117                 KKASSERT(args.rule != NULL);
4118
4119                 m_tag_delete(m, mtag);
4120                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4121         } else {
4122                 args.rule = NULL;
4123         }
4124
4125         args.eh = NULL;
4126         args.oif = NULL;
4127         args.m = m;
4128         ret = ipfw_chk(&args);
4129         m = args.m;
4130
4131         if (m == NULL) {
4132                 error = EACCES;
4133                 goto back;
4134         }
4135
4136         switch (ret) {
4137         case IP_FW_PASS:
4138                 break;
4139
4140         case IP_FW_DENY:
4141                 m_freem(m);
4142                 m = NULL;
4143                 error = EACCES;
4144                 break;
4145
4146         case IP_FW_DUMMYNET:
4147                 /* Send packet to the appropriate pipe */
4148                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args);
4149                 break;
4150
4151         case IP_FW_TEE:
4152                 tee = 1;
4153                 /* FALL THROUGH */
4154
4155         case IP_FW_DIVERT:
4156                 /*
4157                  * Must clear bridge tag when changing
4158                  */
4159                 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
4160                 if (ip_divert_p != NULL) {
4161                         m = ip_divert_p(m, tee, 1);
4162                 } else {
4163                         m_freem(m);
4164                         m = NULL;
4165                         /* not sure this is the right error msg */
4166                         error = EACCES;
4167                 }
4168                 break;
4169
4170         default:
4171                 panic("unknown ipfw return value: %d", ret);
4172         }
4173 back:
4174         *m0 = m;
4175         return error;
4176 }
4177
4178 static int
4179 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4180 {
4181         struct ip_fw_args args;
4182         struct mbuf *m = *m0;
4183         struct m_tag *mtag;
4184         int tee = 0, error = 0, ret;
4185
4186         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4187                 /* Extract info from dummynet tag */
4188                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4189                 KKASSERT(mtag != NULL);
4190                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4191                 KKASSERT(args.rule != NULL);
4192
4193                 m_tag_delete(m, mtag);
4194                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4195         } else {
4196                 args.rule = NULL;
4197         }
4198
4199         args.eh = NULL;
4200         args.m = m;
4201         args.oif = ifp;
4202         ret = ipfw_chk(&args);
4203         m = args.m;
4204
4205         if (m == NULL) {
4206                 error = EACCES;
4207                 goto back;
4208         }
4209
4210         switch (ret) {
4211         case IP_FW_PASS:
4212                 break;
4213
4214         case IP_FW_DENY:
4215                 m_freem(m);
4216                 m = NULL;
4217                 error = EACCES;
4218                 break;
4219
4220         case IP_FW_DUMMYNET:
4221                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args);
4222                 break;
4223
4224         case IP_FW_TEE:
4225                 tee = 1;
4226                 /* FALL THROUGH */
4227
4228         case IP_FW_DIVERT:
4229                 if (ip_divert_p != NULL) {
4230                         m = ip_divert_p(m, tee, 0);
4231                 } else {
4232                         m_freem(m);
4233                         m = NULL;
4234                         /* not sure this is the right error msg */
4235                         error = EACCES;
4236                 }
4237                 break;
4238
4239         default:
4240                 panic("unknown ipfw return value: %d", ret);
4241         }
4242 back:
4243         *m0 = m;
4244         return error;
4245 }
4246
4247 static void
4248 ipfw_hook(void)
4249 {
4250         struct pfil_head *pfh;
4251
4252         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4253
4254         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4255         if (pfh == NULL)
4256                 return;
4257
4258         pfil_add_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_MPSAFE, pfh);
4259         pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_MPSAFE, pfh);
4260 }
4261
4262 static void
4263 ipfw_dehook(void)
4264 {
4265         struct pfil_head *pfh;
4266
4267         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4268
4269         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4270         if (pfh == NULL)
4271                 return;
4272
4273         pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN, pfh);
4274         pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT, pfh);
4275 }
4276
4277 static void
4278 ipfw_sysctl_enable_dispatch(netmsg_t nmsg)
4279 {
4280         struct lwkt_msg *lmsg = &nmsg->lmsg;
4281         int enable = lmsg->u.ms_result;
4282
4283         if (fw_enable == enable)
4284                 goto reply;
4285
4286         fw_enable = enable;
4287         if (fw_enable)
4288                 ipfw_hook();
4289         else
4290                 ipfw_dehook();
4291 reply:
4292         lwkt_replymsg(lmsg, 0);
4293 }
4294
4295 static int
4296 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS)
4297 {
4298         struct netmsg_base nmsg;
4299         struct lwkt_msg *lmsg;
4300         int enable, error;
4301
4302         enable = fw_enable;
4303         error = sysctl_handle_int(oidp, &enable, 0, req);
4304         if (error || req->newptr == NULL)
4305                 return error;
4306
4307         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
4308                     0, ipfw_sysctl_enable_dispatch);
4309         lmsg = &nmsg.lmsg;
4310         lmsg->u.ms_result = enable;
4311
4312         return lwkt_domsg(IPFW_CFGPORT, lmsg, 0);
4313 }
4314
4315 static int
4316 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
4317 {
4318         return sysctl_int_range(oidp, arg1, arg2, req,
4319                IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
4320 }
4321
4322 static int
4323 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
4324 {
4325         int error, value;
4326
4327         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4328
4329         value = dyn_buckets;
4330         error = sysctl_handle_int(oidp, &value, 0, req);
4331         if (error || !req->newptr)
4332                 goto back;
4333
4334         /*
4335          * Make sure we have a power of 2 and
4336          * do not allow more than 64k entries.
4337          */
4338         error = EINVAL;
4339         if (value <= 1 || value > 65536)
4340                 goto back;
4341         if ((value & (value - 1)) != 0)
4342                 goto back;
4343
4344         error = 0;
4345         dyn_buckets = value;
4346 back:
4347         lockmgr(&dyn_lock, LK_RELEASE);
4348         return error;
4349 }
4350
4351 static int
4352 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
4353 {
4354         return sysctl_int_range(oidp, arg1, arg2, req,
4355                                 1, dyn_keepalive_period - 1);
4356 }
4357
4358 static int
4359 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
4360 {
4361         return sysctl_int_range(oidp, arg1, arg2, req,
4362                                 1, dyn_keepalive_period - 1);
4363 }
4364
4365 static void
4366 ipfw_ctx_init_dispatch(netmsg_t nmsg)
4367 {
4368         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
4369         struct ipfw_context *ctx;
4370         struct ip_fw *def_rule;
4371
4372         ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO);
4373         ipfw_ctx[mycpuid] = ctx;
4374
4375         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
4376
4377         def_rule->act_ofs = 0;
4378         def_rule->rulenum = IPFW_DEFAULT_RULE;
4379         def_rule->cmd_len = 1;
4380         def_rule->set = IPFW_DEFAULT_SET;
4381
4382         def_rule->cmd[0].len = 1;
4383 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4384         def_rule->cmd[0].opcode = O_ACCEPT;
4385 #else
4386         def_rule->cmd[0].opcode = O_DENY;
4387 #endif
4388
4389         def_rule->refcnt = 1;
4390         def_rule->cpuid = mycpuid;
4391
4392         /* Install the default rule */
4393         ctx->ipfw_default_rule = def_rule;
4394         ctx->ipfw_layer3_chain = def_rule;
4395
4396         /* Link rule CPU sibling */
4397         ipfw_link_sibling(fwmsg, def_rule);
4398
4399         /* Statistics only need to be updated once */
4400         if (mycpuid == 0)
4401                 ipfw_inc_static_count(def_rule);
4402
4403         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
4404 }
4405
4406 static void
4407 ipfw_init_dispatch(netmsg_t nmsg)
4408 {
4409         struct netmsg_ipfw fwmsg;
4410         int error = 0;
4411
4412         if (IPFW_LOADED) {
4413                 kprintf("IP firewall already loaded\n");
4414                 error = EEXIST;
4415                 goto reply;
4416         }
4417
4418         bzero(&fwmsg, sizeof(fwmsg));
4419         netmsg_init(&fwmsg.base, NULL, &curthread->td_msgport,
4420                     0, ipfw_ctx_init_dispatch);
4421         ifnet_domsg(&fwmsg.base.lmsg, 0);
4422
4423         ip_fw_chk_ptr = ipfw_chk;
4424         ip_fw_ctl_ptr = ipfw_ctl;
4425         ip_fw_dn_io_ptr = ipfw_dummynet_io;
4426
4427         kprintf("ipfw2 initialized, default to %s, logging ",
4428                 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode ==
4429                 O_ACCEPT ? "accept" : "deny");
4430
4431 #ifdef IPFIREWALL_VERBOSE
4432         fw_verbose = 1;
4433 #endif
4434 #ifdef IPFIREWALL_VERBOSE_LIMIT
4435         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4436 #endif
4437         if (fw_verbose == 0) {
4438                 kprintf("disabled\n");
4439         } else if (verbose_limit == 0) {
4440                 kprintf("unlimited\n");
4441         } else {
4442                 kprintf("limited to %d packets/entry by default\n",
4443                         verbose_limit);
4444         }
4445
4446         callout_init_mp(&ipfw_timeout_h);
4447         netmsg_init(&ipfw_timeout_netmsg, NULL, &netisr_adone_rport,
4448                     MSGF_DROPABLE | MSGF_PRIORITY,
4449                     ipfw_tick_dispatch);
4450         lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
4451
4452         ip_fw_loaded = 1;
4453         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
4454
4455         if (fw_enable)
4456                 ipfw_hook();
4457 reply:
4458         lwkt_replymsg(&nmsg->lmsg, error);
4459 }
4460
4461 static int
4462 ipfw_init(void)
4463 {
4464         struct netmsg_base smsg;
4465
4466         netmsg_init(&smsg, NULL, &curthread->td_msgport,
4467                     0, ipfw_init_dispatch);
4468         return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4469 }
4470
4471 #ifdef KLD_MODULE
4472
4473 static void
4474 ipfw_fini_dispatch(netmsg_t nmsg)
4475 {
4476         int error = 0, cpu;
4477
4478         if (ipfw_refcnt != 0) {
4479                 error = EBUSY;
4480                 goto reply;
4481         }
4482
4483         ip_fw_loaded = 0;
4484
4485         ipfw_dehook();
4486         callout_stop(&ipfw_timeout_h);
4487
4488         netmsg_service_sync();
4489
4490         crit_enter();
4491         if ((ipfw_timeout_netmsg.lmsg.ms_flags & MSGF_DONE) == 0) {
4492                 /*
4493                  * Callout message is pending; drop it
4494                  */
4495                 lwkt_dropmsg(&ipfw_timeout_netmsg.lmsg);
4496         }
4497         crit_exit();
4498
4499         ip_fw_chk_ptr = NULL;
4500         ip_fw_ctl_ptr = NULL;
4501         ip_fw_dn_io_ptr = NULL;
4502         ipfw_flush(1 /* kill default rule */);
4503
4504         /* Free pre-cpu context */
4505         for (cpu = 0; cpu < ncpus; ++cpu)
4506                 kfree(ipfw_ctx[cpu], M_IPFW);
4507
4508         kprintf("IP firewall unloaded\n");
4509 reply:
4510         lwkt_replymsg(&nmsg->lmsg, error);
4511 }
4512
4513 static int
4514 ipfw_fini(void)
4515 {
4516         struct netmsg_base smsg;
4517
4518         netmsg_init(&smsg, NULL, &curthread->td_msgport,
4519                     0, ipfw_fini_dispatch);
4520         return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4521 }
4522
4523 #endif  /* KLD_MODULE */
4524
4525 static int
4526 ipfw_modevent(module_t mod, int type, void *unused)
4527 {
4528         int err = 0;
4529
4530         switch (type) {
4531         case MOD_LOAD:
4532                 err = ipfw_init();
4533                 break;
4534
4535         case MOD_UNLOAD:
4536 #ifndef KLD_MODULE
4537                 kprintf("ipfw statically compiled, cannot unload\n");
4538                 err = EBUSY;
4539 #else
4540                 err = ipfw_fini();
4541 #endif
4542                 break;
4543         default:
4544                 break;
4545         }
4546         return err;
4547 }
4548
4549 static moduledata_t ipfwmod = {
4550         "ipfw",
4551         ipfw_modevent,
4552         0
4553 };
4554 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
4555 MODULE_VERSION(ipfw, 1);