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 *q=NULL;
 995
 996         if (ipfw_dyn_v == NULL)
 997                 goto done;      /* not found */
 998
 999         i = hash_packet(pkt);
1000         for (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                 q = q->next;
1030         }
1031         if (q == NULL)
1032                 goto done; /* q = NULL, not found */
1033
1034         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1035                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1036
1037 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
1038 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
1039
1040                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1041                 switch (q->state) {
1042                 case TH_SYN:                            /* opening */
1043                         q->expire = time_second + dyn_syn_lifetime;
1044                         break;
1045
1046                 case BOTH_SYN:                  /* move to established */
1047                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
1048                 case BOTH_SYN | (TH_FIN << 8) :
1049                         if (tcp) {
1050                                 uint32_t ack = ntohl(tcp->th_ack);
1051
1052 #define _SEQ_GE(a, b)   ((int)(a) - (int)(b) >= 0)
1053
1054                                 if (dir == MATCH_FORWARD) {
1055                                         if (q->ack_fwd == 0 ||
1056                                             _SEQ_GE(ack, q->ack_fwd))
1057                                                 q->ack_fwd = ack;
1058                                         else /* ignore out-of-sequence */
1059                                                 break;
1060                                 } else {
1061                                         if (q->ack_rev == 0 ||
1062                                             _SEQ_GE(ack, q->ack_rev))
1063                                                 q->ack_rev = ack;
1064                                         else /* ignore out-of-sequence */
1065                                                 break;
1066                                 }
1067 #undef _SEQ_GE
1068                         }
1069                         q->expire = time_second + dyn_ack_lifetime;
1070                         break;
1071
1072                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
1073                         KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
1074                         q->expire = time_second + dyn_fin_lifetime;
1075                         break;
1076
1077                 default:
1078 #if 0
1079                         /*
1080                          * reset or some invalid combination, but can also
1081                          * occur if we use keep-state the wrong way.
1082                          */
1083                         if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
1084                                 kprintf("invalid state: 0x%x\n", q->state);
1085 #endif
1086                         KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
1087                         q->expire = time_second + dyn_rst_lifetime;
1088                         break;
1089                 }
1090         } else if (pkt->proto == IPPROTO_UDP) {
1091                 q->expire = time_second + dyn_udp_lifetime;
1092         } else {
1093                 /* other protocols */
1094                 q->expire = time_second + dyn_short_lifetime;
1095         }
1096 done:
1097         if (match_direction)
1098                 *match_direction = dir;
1099         return q;
1100 }
1101
1102 static struct ip_fw *
1103 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
1104             uint16_t len, int *deny)
1105 {
1106         struct ip_fw *rule = NULL;
1107         ipfw_dyn_rule *q;
1108         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1109         uint32_t gen;
1110
1111         *deny = 0;
1112         gen = ctx->ipfw_gen;
1113
1114         lockmgr(&dyn_lock, LK_SHARED);
1115
1116         if (ctx->ipfw_gen != gen) {
1117                 /*
1118                  * Static rules had been change when we were waiting
1119                  * for the dynamic hash table lock; deny this packet,
1120                  * since it is _not_ known whether it is safe to keep
1121                  * iterating the static rules.
1122                  */
1123                 *deny = 1;
1124                 goto back;
1125         }
1126
1127         q = lookup_dyn_rule(pkt, match_direction, tcp);
1128         if (q == NULL) {
1129                 rule = NULL;
1130         } else {
1131                 rule = q->stub->rule[mycpuid];
1132                 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid);
1133
1134                 /* XXX */
1135                 q->pcnt++;
1136                 q->bcnt += len;
1137         }
1138 back:
1139         lockmgr(&dyn_lock, LK_RELEASE);
1140         return rule;
1141 }
1142
1143 static void
1144 realloc_dynamic_table(void)
1145 {
1146         ipfw_dyn_rule **old_dyn_v;
1147         uint32_t old_curr_dyn_buckets;
1148
1149         KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
1150                 ("invalid dyn_buckets %d", dyn_buckets));
1151
1152         /* Save the current buckets array for later error recovery */
1153         old_dyn_v = ipfw_dyn_v;
1154         old_curr_dyn_buckets = curr_dyn_buckets;
1155
1156         curr_dyn_buckets = dyn_buckets;
1157         for (;;) {
1158                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1159                                      M_IPFW, M_NOWAIT | M_ZERO);
1160                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1161                         break;
1162
1163                 curr_dyn_buckets /= 2;
1164                 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
1165                     old_dyn_v != NULL) {
1166                         /*
1167                          * Don't try allocating smaller buckets array, reuse
1168                          * the old one, which alreay contains enough buckets
1169                          */
1170                         break;
1171                 }
1172         }
1173
1174         if (ipfw_dyn_v != NULL) {
1175                 if (old_dyn_v != NULL)
1176                         kfree(old_dyn_v, M_IPFW);
1177         } else {
1178                 /* Allocation failed, restore old buckets array */
1179                 ipfw_dyn_v = old_dyn_v;
1180                 curr_dyn_buckets = old_curr_dyn_buckets;
1181         }
1182
1183         if (ipfw_dyn_v != NULL)
1184                 ++dyn_buckets_gen;
1185 }
1186
1187 /**
1188  * Install state of type 'type' for a dynamic session.
1189  * The hash table contains two type of rules:
1190  * - regular rules (O_KEEP_STATE)
1191  * - rules for sessions with limited number of sess per user
1192  *   (O_LIMIT). When they are created, the parent is
1193  *   increased by 1, and decreased on delete. In this case,
1194  *   the third parameter is the parent rule and not the chain.
1195  * - "parent" rules for the above (O_LIMIT_PARENT).
1196  */
1197 static ipfw_dyn_rule *
1198 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1199 {
1200         ipfw_dyn_rule *r;
1201         int i;
1202
1203         if (ipfw_dyn_v == NULL ||
1204             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1205                 realloc_dynamic_table();
1206                 if (ipfw_dyn_v == NULL)
1207                         return NULL; /* failed ! */
1208         }
1209         i = hash_packet(id);
1210
1211         r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1212         if (r == NULL)
1213                 return NULL;
1214
1215         /* increase refcount on parent, and set pointer */
1216         if (dyn_type == O_LIMIT) {
1217                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1218
1219                 if (parent->dyn_type != O_LIMIT_PARENT)
1220                         panic("invalid parent");
1221                 parent->count++;
1222                 r->parent = parent;
1223                 rule = parent->stub->rule[mycpuid];
1224                 KKASSERT(rule->stub == parent->stub);
1225         }
1226         KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1227
1228         r->id = *id;
1229         r->expire = time_second + dyn_syn_lifetime;
1230         r->stub = rule->stub;
1231         r->dyn_type = dyn_type;
1232         r->pcnt = r->bcnt = 0;
1233         r->count = 0;
1234
1235         r->bucket = i;
1236         r->next = ipfw_dyn_v[i];
1237         ipfw_dyn_v[i] = r;
1238         dyn_count++;
1239         dyn_buckets_gen++;
1240         DPRINTF("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1241                 dyn_type,
1242                 r->id.src_ip, r->id.src_port,
1243                 r->id.dst_ip, r->id.dst_port, dyn_count);
1244         return r;
1245 }
1246
1247 /**
1248  * lookup dynamic parent rule using pkt and rule as search keys.
1249  * If the lookup fails, then install one.
1250  */
1251 static ipfw_dyn_rule *
1252 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1253 {
1254         ipfw_dyn_rule *q;
1255         int i;
1256
1257         if (ipfw_dyn_v) {
1258                 i = hash_packet(pkt);
1259                 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1260                         if (q->dyn_type == O_LIMIT_PARENT &&
1261                             rule->stub == q->stub &&
1262                             pkt->proto == q->id.proto &&
1263                             pkt->src_ip == q->id.src_ip &&
1264                             pkt->dst_ip == q->id.dst_ip &&
1265                             pkt->src_port == q->id.src_port &&
1266                             pkt->dst_port == q->id.dst_port) {
1267                                 q->expire = time_second + dyn_short_lifetime;
1268                                 DPRINTF("lookup_dyn_parent found 0x%p\n", q);
1269                                 return q;
1270                         }
1271                 }
1272         }
1273         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1274 }
1275
1276 /**
1277  * Install dynamic state for rule type cmd->o.opcode
1278  *
1279  * Returns 1 (failure) if state is not installed because of errors or because
1280  * session limitations are enforced.
1281  */
1282 static int
1283 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1284                      struct ip_fw_args *args)
1285 {
1286         static int last_log; /* XXX */
1287
1288         ipfw_dyn_rule *q;
1289
1290         DPRINTF("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1291                 cmd->o.opcode,
1292                 args->f_id.src_ip, args->f_id.src_port,
1293                 args->f_id.dst_ip, args->f_id.dst_port);
1294
1295         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1296         if (q != NULL) { /* should never occur */
1297                 if (last_log != time_second) {
1298                         last_log = time_second;
1299                         kprintf(" install_state: entry already present, done\n");
1300                 }
1301                 return 0;
1302         }
1303
1304         if (dyn_count >= dyn_max) {
1305                 /*
1306                  * Run out of slots, try to remove any expired rule.
1307                  */
1308                 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1309                 if (dyn_count >= dyn_max) {
1310                         if (last_log != time_second) {
1311                                 last_log = time_second;
1312                                 kprintf("install_state: "
1313                                         "Too many dynamic rules\n");
1314                         }
1315                         return 1; /* cannot install, notify caller */
1316                 }
1317         }
1318
1319         switch (cmd->o.opcode) {
1320         case O_KEEP_STATE: /* bidir rule */
1321                 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1322                         return 1;
1323                 break;
1324
1325         case O_LIMIT: /* limit number of sessions */
1326                 {
1327                         uint16_t limit_mask = cmd->limit_mask;
1328                         struct ipfw_flow_id id;
1329                         ipfw_dyn_rule *parent;
1330
1331                         DPRINTF("installing dyn-limit rule %d\n",
1332                                 cmd->conn_limit);
1333
1334                         id.dst_ip = id.src_ip = 0;
1335                         id.dst_port = id.src_port = 0;
1336                         id.proto = args->f_id.proto;
1337
1338                         if (limit_mask & DYN_SRC_ADDR)
1339                                 id.src_ip = args->f_id.src_ip;
1340                         if (limit_mask & DYN_DST_ADDR)
1341                                 id.dst_ip = args->f_id.dst_ip;
1342                         if (limit_mask & DYN_SRC_PORT)
1343                                 id.src_port = args->f_id.src_port;
1344                         if (limit_mask & DYN_DST_PORT)
1345                                 id.dst_port = args->f_id.dst_port;
1346
1347                         parent = lookup_dyn_parent(&id, rule);
1348                         if (parent == NULL) {
1349                                 kprintf("add parent failed\n");
1350                                 return 1;
1351                         }
1352
1353                         if (parent->count >= cmd->conn_limit) {
1354                                 /*
1355                                  * See if we can remove some expired rule.
1356                                  */
1357                                 remove_dyn_rule_locked(rule, parent);
1358                                 if (parent->count >= cmd->conn_limit) {
1359                                         if (fw_verbose &&
1360                                             last_log != time_second) {
1361                                                 last_log = time_second;
1362                                                 log(LOG_SECURITY | LOG_DEBUG,
1363                                                     "drop session, "
1364                                                     "too many entries\n");
1365                                         }
1366                                         return 1;
1367                                 }
1368                         }
1369                         if (add_dyn_rule(&args->f_id, O_LIMIT,
1370                                          (struct ip_fw *)parent) == NULL)
1371                                 return 1;
1372                 }
1373                 break;
1374         default:
1375                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1376                 return 1;
1377         }
1378         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1379         return 0;
1380 }
1381
1382 static int
1383 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1384               struct ip_fw_args *args, int *deny)
1385 {
1386         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1387         uint32_t gen;
1388         int ret = 0;
1389
1390         *deny = 0;
1391         gen = ctx->ipfw_gen;
1392
1393         lockmgr(&dyn_lock, LK_EXCLUSIVE);
1394         if (ctx->ipfw_gen != gen) {
1395                 /* See the comment in lookup_rule() */
1396                 *deny = 1;
1397         } else {
1398                 ret = install_state_locked(rule, cmd, args);
1399         }
1400         lockmgr(&dyn_lock, LK_RELEASE);
1401
1402         return ret;
1403 }
1404
1405 /*
1406  * Transmit a TCP packet, containing either a RST or a keepalive.
1407  * When flags & TH_RST, we are sending a RST packet, because of a
1408  * "reset" action matched the packet.
1409  * Otherwise we are sending a keepalive, and flags & TH_
1410  */
1411 static void
1412 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1413 {
1414         struct mbuf *m;
1415         struct ip *ip;
1416         struct tcphdr *tcp;
1417         struct route sro;       /* fake route */
1418
1419         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1420         if (m == NULL)
1421                 return;
1422         m->m_pkthdr.rcvif = NULL;
1423         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1424         m->m_data += max_linkhdr;
1425
1426         ip = mtod(m, struct ip *);
1427         bzero(ip, m->m_len);
1428         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1429         ip->ip_p = IPPROTO_TCP;
1430         tcp->th_off = 5;
1431
1432         /*
1433          * Assume we are sending a RST (or a keepalive in the reverse
1434          * direction), swap src and destination addresses and ports.
1435          */
1436         ip->ip_src.s_addr = htonl(id->dst_ip);
1437         ip->ip_dst.s_addr = htonl(id->src_ip);
1438         tcp->th_sport = htons(id->dst_port);
1439         tcp->th_dport = htons(id->src_port);
1440         if (flags & TH_RST) {   /* we are sending a RST */
1441                 if (flags & TH_ACK) {
1442                         tcp->th_seq = htonl(ack);
1443                         tcp->th_ack = htonl(0);
1444                         tcp->th_flags = TH_RST;
1445                 } else {
1446                         if (flags & TH_SYN)
1447                                 seq++;
1448                         tcp->th_seq = htonl(0);
1449                         tcp->th_ack = htonl(seq);
1450                         tcp->th_flags = TH_RST | TH_ACK;
1451                 }
1452         } else {
1453                 /*
1454                  * We are sending a keepalive. flags & TH_SYN determines
1455                  * the direction, forward if set, reverse if clear.
1456                  * NOTE: seq and ack are always assumed to be correct
1457                  * as set by the caller. This may be confusing...
1458                  */
1459                 if (flags & TH_SYN) {
1460                         /*
1461                          * we have to rewrite the correct addresses!
1462                          */
1463                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1464                         ip->ip_src.s_addr = htonl(id->src_ip);
1465                         tcp->th_dport = htons(id->dst_port);
1466                         tcp->th_sport = htons(id->src_port);
1467                 }
1468                 tcp->th_seq = htonl(seq);
1469                 tcp->th_ack = htonl(ack);
1470                 tcp->th_flags = TH_ACK;
1471         }
1472
1473         /*
1474          * set ip_len to the payload size so we can compute
1475          * the tcp checksum on the pseudoheader
1476          * XXX check this, could save a couple of words ?
1477          */
1478         ip->ip_len = htons(sizeof(struct tcphdr));
1479         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1480
1481         /*
1482          * now fill fields left out earlier
1483          */
1484         ip->ip_ttl = ip_defttl;
1485         ip->ip_len = m->m_pkthdr.len;
1486
1487         bzero(&sro, sizeof(sro));
1488         ip_rtaddr(ip->ip_dst, &sro);
1489
1490         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1491         ip_output(m, NULL, &sro, 0, NULL, NULL);
1492         if (sro.ro_rt)
1493                 RTFREE(sro.ro_rt);
1494 }
1495
1496 /*
1497  * sends a reject message, consuming the mbuf passed as an argument.
1498  */
1499 static void
1500 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1501 {
1502         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1503                 /* We need the IP header in host order for icmp_error(). */
1504                 if (args->eh != NULL) {
1505                         struct ip *ip = mtod(args->m, struct ip *);
1506
1507                         ip->ip_len = ntohs(ip->ip_len);
1508                         ip->ip_off = ntohs(ip->ip_off);
1509                 }
1510                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1511         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1512                 struct tcphdr *const tcp =
1513                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1514
1515                 if ((tcp->th_flags & TH_RST) == 0) {
1516                         send_pkt(&args->f_id, ntohl(tcp->th_seq),
1517                                  ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1518                 }
1519                 m_freem(args->m);
1520         } else {
1521                 m_freem(args->m);
1522         }
1523         args->m = NULL;
1524 }
1525
1526 /**
1527  *
1528  * Given an ip_fw *, lookup_next_rule will return a pointer
1529  * to the next rule, which can be either the jump
1530  * target (for skipto instructions) or the next one in the list (in
1531  * all other cases including a missing jump target).
1532  * The result is also written in the "next_rule" field of the rule.
1533  * Backward jumps are not allowed, so start looking from the next
1534  * rule...
1535  *
1536  * This never returns NULL -- in case we do not have an exact match,
1537  * the next rule is returned. When the ruleset is changed,
1538  * pointers are flushed so we are always correct.
1539  */
1540
1541 static struct ip_fw *
1542 lookup_next_rule(struct ip_fw *me)
1543 {
1544         struct ip_fw *rule = NULL;
1545         ipfw_insn *cmd;
1546
1547         /* look for action, in case it is a skipto */
1548         cmd = ACTION_PTR(me);
1549         if (cmd->opcode == O_LOG)
1550                 cmd += F_LEN(cmd);
1551         if (cmd->opcode == O_SKIPTO) {
1552                 for (rule = me->next; rule; rule = rule->next) {
1553                         if (rule->rulenum >= cmd->arg1)
1554                                 break;
1555                 }
1556         }
1557         if (rule == NULL)                       /* failure or not a skipto */
1558                 rule = me->next;
1559         me->next_rule = rule;
1560         return rule;
1561 }
1562
1563 static int
1564 _ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1565                 enum ipfw_opcodes opcode, uid_t uid)
1566 {
1567         struct in_addr src_ip, dst_ip;
1568         struct inpcbinfo *pi;
1569         int wildcard;
1570         struct inpcb *pcb;
1571
1572         if (fid->proto == IPPROTO_TCP) {
1573                 wildcard = 0;
1574                 pi = &tcbinfo[mycpuid];
1575         } else if (fid->proto == IPPROTO_UDP) {
1576                 wildcard = 1;
1577                 pi = &udbinfo;
1578         } else {
1579                 return 0;
1580         }
1581
1582         /*
1583          * Values in 'fid' are in host byte order
1584          */
1585         dst_ip.s_addr = htonl(fid->dst_ip);
1586         src_ip.s_addr = htonl(fid->src_ip);
1587         if (oif) {
1588                 pcb = in_pcblookup_hash(pi,
1589                         dst_ip, htons(fid->dst_port),
1590                         src_ip, htons(fid->src_port),
1591                         wildcard, oif);
1592         } else {
1593                 pcb = in_pcblookup_hash(pi,
1594                         src_ip, htons(fid->src_port),
1595                         dst_ip, htons(fid->dst_port),
1596                         wildcard, NULL);
1597         }
1598         if (pcb == NULL || pcb->inp_socket == NULL)
1599                 return 0;
1600
1601         if (opcode == O_UID) {
1602 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1603                 return !socheckuid(pcb->inp_socket, uid);
1604 #undef socheckuid
1605         } else  {
1606                 return groupmember(uid, pcb->inp_socket->so_cred);
1607         }
1608 }
1609
1610 static int
1611 ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1612                enum ipfw_opcodes opcode, uid_t uid, int *deny)
1613 {
1614         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1615         uint32_t gen;
1616         int match = 0;
1617
1618         *deny = 0;
1619         gen = ctx->ipfw_gen;
1620
1621         get_mplock();
1622         if (gen != ctx->ipfw_gen) {
1623                 /* See the comment in lookup_rule() */
1624                 *deny = 1;
1625         } else {
1626                 match = _ipfw_match_uid(fid, oif, opcode, uid);
1627         }
1628         rel_mplock();
1629         return match;
1630 }
1631
1632 /*
1633  * The main check routine for the firewall.
1634  *
1635  * All arguments are in args so we can modify them and return them
1636  * back to the caller.
1637  *
1638  * Parameters:
1639  *
1640  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1641  *              Starts with the IP header.
1642  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1643  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1644  *              The incoming interface is in the mbuf. (in)
1645  *
1646  *      args->rule      Pointer to the last matching rule (in/out)
1647  *      args->f_id      Addresses grabbed from the packet (out)
1648  *
1649  * Return value:
1650  *
1651  *      If the packet was denied/rejected and has been dropped, *m is equal
1652  *      to NULL upon return.
1653  *
1654  *      IP_FW_DENY      the packet must be dropped.
1655  *      IP_FW_PASS      The packet is to be accepted and routed normally.
1656  *      IP_FW_DIVERT    Divert the packet to port (args->cookie)
1657  *      IP_FW_TEE       Tee the packet to port (args->cookie)
1658  *      IP_FW_DUMMYNET  Send the packet to pipe/queue (args->cookie)
1659  */
1660
1661 static int
1662 ipfw_chk(struct ip_fw_args *args)
1663 {
1664         /*
1665          * Local variables hold state during the processing of a packet.
1666          *
1667          * IMPORTANT NOTE: to speed up the processing of rules, there
1668          * are some assumption on the values of the variables, which
1669          * are documented here. Should you change them, please check
1670          * the implementation of the various instructions to make sure
1671          * that they still work.
1672          *
1673          * args->eh     The MAC header. It is non-null for a layer2
1674          *      packet, it is NULL for a layer-3 packet.
1675          *
1676          * m | args->m  Pointer to the mbuf, as received from the caller.
1677          *      It may change if ipfw_chk() does an m_pullup, or if it
1678          *      consumes the packet because it calls send_reject().
1679          *      XXX This has to change, so that ipfw_chk() never modifies
1680          *      or consumes the buffer.
1681          * ip   is simply an alias of the value of m, and it is kept
1682          *      in sync with it (the packet is  supposed to start with
1683          *      the ip header).
1684          */
1685         struct mbuf *m = args->m;
1686         struct ip *ip = mtod(m, struct ip *);
1687
1688         /*
1689          * oif | args->oif      If NULL, ipfw_chk has been called on the
1690          *      inbound path (ether_input, ip_input).
1691          *      If non-NULL, ipfw_chk has been called on the outbound path
1692          *      (ether_output, ip_output).
1693          */
1694         struct ifnet *oif = args->oif;
1695
1696         struct ip_fw *f = NULL;         /* matching rule */
1697         int retval = IP_FW_PASS;
1698         struct m_tag *mtag;
1699         struct divert_info *divinfo;
1700
1701         /*
1702          * hlen The length of the IPv4 header.
1703          *      hlen >0 means we have an IPv4 packet.
1704          */
1705         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1706
1707         /*
1708          * offset       The offset of a fragment. offset != 0 means that
1709          *      we have a fragment at this offset of an IPv4 packet.
1710          *      offset == 0 means that (if this is an IPv4 packet)
1711          *      this is the first or only fragment.
1712          */
1713         u_short offset = 0;
1714
1715         /*
1716          * Local copies of addresses. They are only valid if we have
1717          * an IP packet.
1718          *
1719          * proto        The protocol. Set to 0 for non-ip packets,
1720          *      or to the protocol read from the packet otherwise.
1721          *      proto != 0 means that we have an IPv4 packet.
1722          *
1723          * src_port, dst_port   port numbers, in HOST format. Only
1724          *      valid for TCP and UDP packets.
1725          *
1726          * src_ip, dst_ip       ip addresses, in NETWORK format.
1727          *      Only valid for IPv4 packets.
1728          */
1729         uint8_t proto;
1730         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1731         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1732         uint16_t ip_len = 0;
1733
1734         /*
1735          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1736          *      MATCH_NONE when checked and not matched (dyn_f = NULL),
1737          *      MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1738          */
1739         int dyn_dir = MATCH_UNKNOWN;
1740         struct ip_fw *dyn_f = NULL;
1741         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1742
1743         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1744                 return IP_FW_PASS;      /* accept */
1745
1746         if (args->eh == NULL ||         /* layer 3 packet */
1747             (m->m_pkthdr.len >= sizeof(struct ip) &&
1748              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1749                 hlen = ip->ip_hl << 2;
1750
1751         /*
1752          * Collect parameters into local variables for faster matching.
1753          */
1754         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1755                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1756                 goto after_ip_checks;
1757         }
1758
1759         proto = args->f_id.proto = ip->ip_p;
1760         src_ip = ip->ip_src;
1761         dst_ip = ip->ip_dst;
1762         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1763                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1764                 ip_len = ntohs(ip->ip_len);
1765         } else {
1766                 offset = ip->ip_off & IP_OFFMASK;
1767                 ip_len = ip->ip_len;
1768         }
1769
1770 #define PULLUP_TO(len)                          \
1771 do {                                            \
1772         if (m->m_len < (len)) {                 \
1773                 args->m = m = m_pullup(m, (len));\
1774                 if (m == NULL)                  \
1775                         goto pullup_failed;     \
1776                 ip = mtod(m, struct ip *);      \
1777         }                                       \
1778 } while (0)
1779
1780         if (offset == 0) {
1781                 switch (proto) {
1782                 case IPPROTO_TCP:
1783                         {
1784                                 struct tcphdr *tcp;
1785
1786                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1787                                 tcp = L3HDR(struct tcphdr, ip);
1788                                 dst_port = tcp->th_dport;
1789                                 src_port = tcp->th_sport;
1790                                 args->f_id.flags = tcp->th_flags;
1791                         }
1792                         break;
1793
1794                 case IPPROTO_UDP:
1795                         {
1796                                 struct udphdr *udp;
1797
1798                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1799                                 udp = L3HDR(struct udphdr, ip);
1800                                 dst_port = udp->uh_dport;
1801                                 src_port = udp->uh_sport;
1802                         }
1803                         break;
1804
1805                 case IPPROTO_ICMP:
1806                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1807                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1808                         break;
1809
1810                 default:
1811                         break;
1812                 }
1813         }
1814
1815 #undef PULLUP_TO
1816
1817         args->f_id.src_ip = ntohl(src_ip.s_addr);
1818         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1819         args->f_id.src_port = src_port = ntohs(src_port);
1820         args->f_id.dst_port = dst_port = ntohs(dst_port);
1821
1822 after_ip_checks:
1823         if (args->rule) {
1824                 /*
1825                  * Packet has already been tagged. Look for the next rule
1826                  * to restart processing.
1827                  *
1828                  * If fw_one_pass != 0 then just accept it.
1829                  * XXX should not happen here, but optimized out in
1830                  * the caller.
1831                  */
1832                 if (fw_one_pass)
1833                         return IP_FW_PASS;
1834
1835                 /* This rule is being/has been flushed */
1836                 if (ipfw_flushing)
1837                         return IP_FW_DENY;
1838
1839                 KASSERT(args->rule->cpuid == mycpuid,
1840                         ("rule used on cpu%d", mycpuid));
1841
1842                 /* This rule was deleted */
1843                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1844                         return IP_FW_DENY;
1845
1846                 f = args->rule->next_rule;
1847                 if (f == NULL)
1848                         f = lookup_next_rule(args->rule);
1849         } else {
1850                 /*
1851                  * Find the starting rule. It can be either the first
1852                  * one, or the one after divert_rule if asked so.
1853                  */
1854                 int skipto;
1855
1856                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1857                 if (mtag != NULL) {
1858                         divinfo = m_tag_data(mtag);
1859                         skipto = divinfo->skipto;
1860                 } else {
1861                         skipto = 0;
1862                 }
1863
1864                 f = ctx->ipfw_layer3_chain;
1865                 if (args->eh == NULL && skipto != 0) {
1866                         /* No skipto during rule flushing */
1867                         if (ipfw_flushing)
1868                                 return IP_FW_DENY;
1869
1870                         if (skipto >= IPFW_DEFAULT_RULE)
1871                                 return IP_FW_DENY; /* invalid */
1872
1873                         while (f && f->rulenum <= skipto)
1874                                 f = f->next;
1875                         if (f == NULL)  /* drop packet */
1876                                 return IP_FW_DENY;
1877                 } else if (ipfw_flushing) {
1878                         /* Rules are being flushed; skip to default rule */
1879                         f = ctx->ipfw_default_rule;
1880                 }
1881         }
1882         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1883                 m_tag_delete(m, mtag);
1884
1885         /*
1886          * Now scan the rules, and parse microinstructions for each rule.
1887          */
1888         for (; f; f = f->next) {
1889                 int l, cmdlen;
1890                 ipfw_insn *cmd;
1891                 int skip_or; /* skip rest of OR block */
1892
1893 again:
1894                 if (ctx->ipfw_set_disable & (1 << f->set))
1895                         continue;
1896
1897                 skip_or = 0;
1898                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1899                      l -= cmdlen, cmd += cmdlen) {
1900                         int match, deny;
1901
1902                         /*
1903                          * check_body is a jump target used when we find a
1904                          * CHECK_STATE, and need to jump to the body of
1905                          * the target rule.
1906                          */
1907
1908 check_body:
1909                         cmdlen = F_LEN(cmd);
1910                         /*
1911                          * An OR block (insn_1 || .. || insn_n) has the
1912                          * F_OR bit set in all but the last instruction.
1913                          * The first match will set "skip_or", and cause
1914                          * the following instructions to be skipped until
1915                          * past the one with the F_OR bit clear.
1916                          */
1917                         if (skip_or) {          /* skip this instruction */
1918                                 if ((cmd->len & F_OR) == 0)
1919                                         skip_or = 0;    /* next one is good */
1920                                 continue;
1921                         }
1922                         match = 0; /* set to 1 if we succeed */
1923
1924                         switch (cmd->opcode) {
1925                         /*
1926                          * The first set of opcodes compares the packet's
1927                          * fields with some pattern, setting 'match' if a
1928                          * match is found. At the end of the loop there is
1929                          * logic to deal with F_NOT and F_OR flags associated
1930                          * with the opcode.
1931                          */
1932                         case O_NOP:
1933                                 match = 1;
1934                                 break;
1935
1936                         case O_FORWARD_MAC:
1937                                 kprintf("ipfw: opcode %d unimplemented\n",
1938                                         cmd->opcode);
1939                                 break;
1940
1941                         case O_GID:
1942                         case O_UID:
1943                                 /*
1944                                  * We only check offset == 0 && proto != 0,
1945                                  * as this ensures that we have an IPv4
1946                                  * packet with the ports info.
1947                                  */
1948                                 if (offset!=0)
1949                                         break;
1950
1951                                 match = ipfw_match_uid(&args->f_id, oif,
1952                                         cmd->opcode,
1953                                         (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1954                                         &deny);
1955                                 if (deny)
1956                                         return IP_FW_DENY;
1957                                 break;
1958
1959                         case O_RECV:
1960                                 match = iface_match(m->m_pkthdr.rcvif,
1961                                     (ipfw_insn_if *)cmd);
1962                                 break;
1963
1964                         case O_XMIT:
1965                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1966                                 break;
1967
1968                         case O_VIA:
1969                                 match = iface_match(oif ? oif :
1970                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1971                                 break;
1972
1973                         case O_MACADDR2:
1974                                 if (args->eh != NULL) { /* have MAC header */
1975                                         uint32_t *want = (uint32_t *)
1976                                                 ((ipfw_insn_mac *)cmd)->addr;
1977                                         uint32_t *mask = (uint32_t *)
1978                                                 ((ipfw_insn_mac *)cmd)->mask;
1979                                         uint32_t *hdr = (uint32_t *)args->eh;
1980
1981                                         match =
1982                                         (want[0] == (hdr[0] & mask[0]) &&
1983                                          want[1] == (hdr[1] & mask[1]) &&
1984                                          want[2] == (hdr[2] & mask[2]));
1985                                 }
1986                                 break;
1987
1988                         case O_MAC_TYPE:
1989                                 if (args->eh != NULL) {
1990                                         uint16_t t =
1991                                             ntohs(args->eh->ether_type);
1992                                         uint16_t *p =
1993                                             ((ipfw_insn_u16 *)cmd)->ports;
1994                                         int i;
1995
1996                                         /* Special vlan handling */
1997                                         if (m->m_flags & M_VLANTAG)
1998                                                 t = ETHERTYPE_VLAN;
1999
2000                                         for (i = cmdlen - 1; !match && i > 0;
2001                                              i--, p += 2) {
2002                                                 match =
2003                                                 (t >= p[0] && t <= p[1]);
2004                                         }
2005                                 }
2006                                 break;
2007
2008                         case O_FRAG:
2009                                 match = (hlen > 0 && offset != 0);
2010                                 break;
2011
2012                         case O_IN:      /* "out" is "not in" */
2013                                 match = (oif == NULL);
2014                                 break;
2015
2016                         case O_LAYER2:
2017                                 match = (args->eh != NULL);
2018                                 break;
2019
2020                         case O_PROTO:
2021                                 /*
2022                                  * We do not allow an arg of 0 so the
2023                                  * check of "proto" only suffices.
2024                                  */
2025                                 match = (proto == cmd->arg1);
2026                                 break;
2027
2028                         case O_IP_SRC:
2029                                 match = (hlen > 0 &&
2030                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2031                                     src_ip.s_addr);
2032                                 break;
2033
2034                         case O_IP_SRC_MASK:
2035                                 match = (hlen > 0 &&
2036                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2037                                      (src_ip.s_addr &
2038                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2039                                 break;
2040
2041                         case O_IP_SRC_ME:
2042                                 if (hlen > 0) {
2043                                         struct ifnet *tif;
2044
2045                                         tif = INADDR_TO_IFP(&src_ip);
2046                                         match = (tif != NULL);
2047                                 }
2048                                 break;
2049
2050                         case O_IP_DST_SET:
2051                         case O_IP_SRC_SET:
2052                                 if (hlen > 0) {
2053                                         uint32_t *d = (uint32_t *)(cmd + 1);
2054                                         uint32_t addr =
2055                                             cmd->opcode == O_IP_DST_SET ?
2056                                                 args->f_id.dst_ip :
2057                                                 args->f_id.src_ip;
2058
2059                                         if (addr < d[0])
2060                                                 break;
2061                                         addr -= d[0]; /* subtract base */
2062                                         match =
2063                                         (addr < cmd->arg1) &&
2064                                          (d[1 + (addr >> 5)] &
2065                                           (1 << (addr & 0x1f)));
2066                                 }
2067                                 break;
2068
2069                         case O_IP_DST:
2070                                 match = (hlen > 0 &&
2071                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2072                                     dst_ip.s_addr);
2073                                 break;
2074
2075                         case O_IP_DST_MASK:
2076                                 match = (hlen > 0) &&
2077                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2078                                      (dst_ip.s_addr &
2079                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2080                                 break;
2081
2082                         case O_IP_DST_ME:
2083                                 if (hlen > 0) {
2084                                         struct ifnet *tif;
2085
2086                                         tif = INADDR_TO_IFP(&dst_ip);
2087                                         match = (tif != NULL);
2088                                 }
2089                                 break;
2090
2091                         case O_IP_SRCPORT:
2092                         case O_IP_DSTPORT:
2093                                 /*
2094                                  * offset == 0 && proto != 0 is enough
2095                                  * to guarantee that we have an IPv4
2096                                  * packet with port info.
2097                                  */
2098                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2099                                     && offset == 0) {
2100                                         uint16_t x =
2101                                             (cmd->opcode == O_IP_SRCPORT) ?
2102                                                 src_port : dst_port ;
2103                                         uint16_t *p =
2104                                             ((ipfw_insn_u16 *)cmd)->ports;
2105                                         int i;
2106
2107                                         for (i = cmdlen - 1; !match && i > 0;
2108                                              i--, p += 2) {
2109                                                 match =
2110                                                 (x >= p[0] && x <= p[1]);
2111                                         }
2112                                 }
2113                                 break;
2114
2115                         case O_ICMPTYPE:
2116                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
2117                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2118                                 break;
2119
2120                         case O_IPOPT:
2121                                 match = (hlen > 0 && ipopts_match(ip, cmd));
2122                                 break;
2123
2124                         case O_IPVER:
2125                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2126                                 break;
2127
2128                         case O_IPTTL:
2129                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2130                                 break;
2131
2132                         case O_IPID:
2133                                 match = (hlen > 0 &&
2134                                     cmd->arg1 == ntohs(ip->ip_id));
2135                                 break;
2136
2137                         case O_IPLEN:
2138                                 match = (hlen > 0 && cmd->arg1 == ip_len);
2139                                 break;
2140
2141                         case O_IPPRECEDENCE:
2142                                 match = (hlen > 0 &&
2143                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
2144                                 break;
2145
2146                         case O_IPTOS:
2147                                 match = (hlen > 0 &&
2148                                     flags_match(cmd, ip->ip_tos));
2149                                 break;
2150
2151                         case O_TCPFLAGS:
2152                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2153                                     flags_match(cmd,
2154                                         L3HDR(struct tcphdr,ip)->th_flags));
2155                                 break;
2156
2157                         case O_TCPOPTS:
2158                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2159                                     tcpopts_match(ip, cmd));
2160                                 break;
2161
2162                         case O_TCPSEQ:
2163                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2164                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2165                                         L3HDR(struct tcphdr,ip)->th_seq);
2166                                 break;
2167
2168                         case O_TCPACK:
2169                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2170                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2171                                         L3HDR(struct tcphdr,ip)->th_ack);
2172                                 break;
2173
2174                         case O_TCPWIN:
2175                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2176                                     cmd->arg1 ==
2177                                         L3HDR(struct tcphdr,ip)->th_win);
2178                                 break;
2179
2180                         case O_ESTAB:
2181                                 /* reject packets which have SYN only */
2182                                 /* XXX should i also check for TH_ACK ? */
2183                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2184                                     (L3HDR(struct tcphdr,ip)->th_flags &
2185                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2186                                 break;
2187
2188                         case O_LOG:
2189                                 if (fw_verbose)
2190                                         ipfw_log(f, hlen, args->eh, m, oif);
2191                                 match = 1;
2192                                 break;
2193
2194                         case O_PROB:
2195                                 match = (krandom() <
2196                                         ((ipfw_insn_u32 *)cmd)->d[0]);
2197                                 break;
2198
2199                         /*
2200                          * The second set of opcodes represents 'actions',
2201                          * i.e. the terminal part of a rule once the packet
2202                          * matches all previous patterns.
2203                          * Typically there is only one action for each rule,
2204                          * and the opcode is stored at the end of the rule
2205                          * (but there are exceptions -- see below).
2206                          *
2207                          * In general, here we set retval and terminate the
2208                          * outer loop (would be a 'break 3' in some language,
2209                          * but we need to do a 'goto done').
2210                          *
2211                          * Exceptions:
2212                          * O_COUNT and O_SKIPTO actions:
2213                          *   instead of terminating, we jump to the next rule
2214                          *   ('goto next_rule', equivalent to a 'break 2'),
2215                          *   or to the SKIPTO target ('goto again' after
2216                          *   having set f, cmd and l), respectively.
2217                          *
2218                          * O_LIMIT and O_KEEP_STATE: these opcodes are
2219                          *   not real 'actions', and are stored right
2220                          *   before the 'action' part of the rule.
2221                          *   These opcodes try to install an entry in the
2222                          *   state tables; if successful, we continue with
2223                          *   the next opcode (match=1; break;), otherwise
2224                          *   the packet must be dropped ('goto done' after
2225                          *   setting retval).  If static rules are changed
2226                          *   during the state installation, the packet will
2227                          *   be dropped and rule's stats will not beupdated
2228                          *   ('return IP_FW_DENY').
2229                          *
2230                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2231                          *   cause a lookup of the state table, and a jump
2232                          *   to the 'action' part of the parent rule
2233                          *   ('goto check_body') if an entry is found, or
2234                          *   (CHECK_STATE only) a jump to the next rule if
2235                          *   the entry is not found ('goto next_rule').
2236                          *   The result of the lookup is cached to make
2237                          *   further instances of these opcodes are
2238                          *   effectively NOPs.  If static rules are changed
2239                          *   during the state looking up, the packet will
2240                          *   be dropped and rule's stats will not be updated
2241                          *   ('return IP_FW_DENY').
2242                          */
2243                         case O_LIMIT:
2244                         case O_KEEP_STATE:
2245                                 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2246                                         kprintf("%s rule (%d) is not ready "
2247                                                 "on cpu%d\n",
2248                                                 cmd->opcode == O_LIMIT ?
2249                                                 "limit" : "keep state",
2250                                                 f->rulenum, f->cpuid);
2251                                         goto next_rule;
2252                                 }
2253                                 if (install_state(f,
2254                                     (ipfw_insn_limit *)cmd, args, &deny)) {
2255                                         if (deny)
2256                                                 return IP_FW_DENY;
2257
2258                                         retval = IP_FW_DENY;
2259                                         goto done; /* error/limit violation */
2260                                 }
2261                                 if (deny)
2262                                         return IP_FW_DENY;
2263                                 match = 1;
2264                                 break;
2265
2266                         case O_PROBE_STATE:
2267                         case O_CHECK_STATE:
2268                                 /*
2269                                  * dynamic rules are checked at the first
2270                                  * keep-state or check-state occurrence,
2271                                  * with the result being stored in dyn_dir.
2272                                  * The compiler introduces a PROBE_STATE
2273                                  * instruction for us when we have a
2274                                  * KEEP_STATE (because PROBE_STATE needs
2275                                  * to be run first).
2276                                  */
2277                                 if (dyn_dir == MATCH_UNKNOWN) {
2278                                         dyn_f = lookup_rule(&args->f_id,
2279                                                 &dyn_dir,
2280                                                 proto == IPPROTO_TCP ?
2281                                                 L3HDR(struct tcphdr, ip) : NULL,
2282                                                 ip_len, &deny);
2283                                         if (deny)
2284                                                 return IP_FW_DENY;
2285                                         if (dyn_f != NULL) {
2286                                                 /*
2287                                                  * Found a rule from a dynamic
2288                                                  * entry; jump to the 'action'
2289                                                  * part of the rule.
2290                                                  */
2291                                                 f = dyn_f;
2292                                                 cmd = ACTION_PTR(f);
2293                                                 l = f->cmd_len - f->act_ofs;
2294                                                 goto check_body;
2295                                         }
2296                                 }
2297                                 /*
2298                                  * Dynamic entry not found. If CHECK_STATE,
2299                                  * skip to next rule, if PROBE_STATE just
2300                                  * ignore and continue with next opcode.
2301                                  */
2302                                 if (cmd->opcode == O_CHECK_STATE)
2303                                         goto next_rule;
2304                                 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2305                                         goto next_rule; /* not ready yet */
2306                                 match = 1;
2307                                 break;
2308
2309                         case O_ACCEPT:
2310                                 retval = IP_FW_PASS;    /* accept */
2311                                 goto done;
2312
2313                         case O_PIPE:
2314                         case O_QUEUE:
2315                                 args->rule = f; /* report matching rule */
2316                                 args->cookie = cmd->arg1;
2317                                 retval = IP_FW_DUMMYNET;
2318                                 goto done;
2319
2320                         case O_DIVERT:
2321                         case O_TEE:
2322                                 if (args->eh) /* not on layer 2 */
2323                                         break;
2324
2325                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2326                                                  sizeof(*divinfo), MB_DONTWAIT);
2327                                 if (mtag == NULL) {
2328                                         retval = IP_FW_DENY;
2329                                         goto done;
2330                                 }
2331                                 divinfo = m_tag_data(mtag);
2332
2333                                 divinfo->skipto = f->rulenum;
2334                                 divinfo->port = cmd->arg1;
2335                                 divinfo->tee = (cmd->opcode == O_TEE);
2336                                 m_tag_prepend(m, mtag);
2337
2338                                 args->cookie = cmd->arg1;
2339                                 retval = (cmd->opcode == O_DIVERT) ?
2340                                          IP_FW_DIVERT : IP_FW_TEE;
2341                                 goto done;
2342
2343                         case O_COUNT:
2344                         case O_SKIPTO:
2345                                 f->pcnt++;      /* update stats */
2346                                 f->bcnt += ip_len;
2347                                 f->timestamp = time_second;
2348                                 if (cmd->opcode == O_COUNT)
2349                                         goto next_rule;
2350                                 /* handle skipto */
2351                                 if (f->next_rule == NULL)
2352                                         lookup_next_rule(f);
2353                                 f = f->next_rule;
2354                                 goto again;
2355
2356                         case O_REJECT:
2357                                 /*
2358                                  * Drop the packet and send a reject notice
2359                                  * if the packet is not ICMP (or is an ICMP
2360                                  * query), and it is not multicast/broadcast.
2361                                  */
2362                                 if (hlen > 0 &&
2363                                     (proto != IPPROTO_ICMP ||
2364                                      is_icmp_query(ip)) &&
2365                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
2366                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2367                                         /*
2368                                          * Update statistics before the possible
2369                                          * blocking 'send_reject'
2370                                          */
2371                                         f->pcnt++;
2372                                         f->bcnt += ip_len;
2373                                         f->timestamp = time_second;
2374
2375                                         send_reject(args, cmd->arg1,
2376                                             offset,ip_len);
2377                                         m = args->m;
2378
2379                                         /*
2380                                          * Return directly here, rule stats
2381                                          * have been updated above.
2382                                          */
2383                                         return IP_FW_DENY;
2384                                 }
2385                                 /* FALLTHROUGH */
2386                         case O_DENY:
2387                                 retval = IP_FW_DENY;
2388                                 goto done;
2389
2390                         case O_FORWARD_IP:
2391                                 if (args->eh)   /* not valid on layer2 pkts */
2392                                         break;
2393                                 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2394                                         struct sockaddr_in *sin;
2395
2396                                         mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2397                                                sizeof(*sin), MB_DONTWAIT);
2398                                         if (mtag == NULL) {
2399                                                 retval = IP_FW_DENY;
2400                                                 goto done;
2401                                         }
2402                                         sin = m_tag_data(mtag);
2403
2404                                         /* Structure copy */
2405                                         *sin = ((ipfw_insn_sa *)cmd)->sa;
2406
2407                                         m_tag_prepend(m, mtag);
2408                                         m->m_pkthdr.fw_flags |=
2409                                                 IPFORWARD_MBUF_TAGGED;
2410                                         m->m_pkthdr.fw_flags &=
2411                                                 ~BRIDGE_MBUF_TAGGED;
2412                                 }
2413                                 retval = IP_FW_PASS;
2414                                 goto done;
2415
2416                         default:
2417                                 panic("-- unknown opcode %d", cmd->opcode);
2418                         } /* end of switch() on opcodes */
2419
2420                         if (cmd->len & F_NOT)
2421                                 match = !match;
2422
2423                         if (match) {
2424                                 if (cmd->len & F_OR)
2425                                         skip_or = 1;
2426                         } else {
2427                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2428                                         break;          /* try next rule    */
2429                         }
2430
2431                 }       /* end of inner for, scan opcodes */
2432
2433 next_rule:;             /* try next rule                */
2434
2435         }               /* end of outer for, scan rules */
2436         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2437         return IP_FW_DENY;
2438
2439 done:
2440         /* Update statistics */
2441         f->pcnt++;
2442         f->bcnt += ip_len;
2443         f->timestamp = time_second;
2444         return retval;
2445
2446 pullup_failed:
2447         if (fw_verbose)
2448                 kprintf("pullup failed\n");
2449         return IP_FW_DENY;
2450 }
2451
2452 static void
2453 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2454 {
2455         struct m_tag *mtag;
2456         struct dn_pkt *pkt;
2457         ipfw_insn *cmd;
2458         const struct ipfw_flow_id *id;
2459         struct dn_flow_id *fid;
2460
2461         M_ASSERTPKTHDR(m);
2462
2463         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2464         if (mtag == NULL) {
2465                 m_freem(m);
2466                 return;
2467         }
2468         m_tag_prepend(m, mtag);
2469
2470         pkt = m_tag_data(mtag);
2471         bzero(pkt, sizeof(*pkt));
2472
2473         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2474         if (cmd->opcode == O_LOG)
2475                 cmd += F_LEN(cmd);
2476         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2477                 ("Rule is not PIPE or QUEUE, opcode %d", cmd->opcode));
2478
2479         pkt->dn_m = m;
2480         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2481         pkt->ifp = fwa->oif;
2482         pkt->pipe_nr = pipe_nr;
2483
2484         pkt->cpuid = mycpuid;
2485         pkt->msgport = cur_netport();
2486
2487         id = &fwa->f_id;
2488         fid = &pkt->id;
2489         fid->fid_dst_ip = id->dst_ip;
2490         fid->fid_src_ip = id->src_ip;
2491         fid->fid_dst_port = id->dst_port;
2492         fid->fid_src_port = id->src_port;
2493         fid->fid_proto = id->proto;
2494         fid->fid_flags = id->flags;
2495
2496         ipfw_ref_rule(fwa->rule);
2497         pkt->dn_priv = fwa->rule;
2498         pkt->dn_unref_priv = ipfw_unref_rule;
2499
2500         if (cmd->opcode == O_PIPE)
2501                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2502
2503         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2504 }
2505
2506 /*
2507  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2508  * These will be reconstructed on the fly as packets are matched.
2509  * Must be called at splimp().
2510  */
2511 static void
2512 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2513 {
2514         struct ip_fw *rule;
2515
2516         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2517                 rule->next_rule = NULL;
2518 }
2519
2520 static __inline void
2521 ipfw_inc_static_count(struct ip_fw *rule)
2522 {
2523         /* Static rule's counts are updated only on CPU0 */
2524         KKASSERT(mycpuid == 0);
2525
2526         static_count++;
2527         static_ioc_len += IOC_RULESIZE(rule);
2528 }
2529
2530 static __inline void
2531 ipfw_dec_static_count(struct ip_fw *rule)
2532 {
2533         int l = IOC_RULESIZE(rule);
2534
2535         /* Static rule's counts are updated only on CPU0 */
2536         KKASSERT(mycpuid == 0);
2537
2538         KASSERT(static_count > 0, ("invalid static count %u", static_count));
2539         static_count--;
2540
2541         KASSERT(static_ioc_len >= l,
2542                 ("invalid static len %u", static_ioc_len));
2543         static_ioc_len -= l;
2544 }
2545
2546 static void
2547 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2548 {
2549         if (fwmsg->sibling != NULL) {
2550                 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2551                 fwmsg->sibling->sibling = rule;
2552         }
2553         fwmsg->sibling = rule;
2554 }
2555
2556 static struct ip_fw *
2557 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2558 {
2559         struct ip_fw *rule;
2560
2561         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2562
2563         rule->act_ofs = ioc_rule->act_ofs;
2564         rule->cmd_len = ioc_rule->cmd_len;
2565         rule->rulenum = ioc_rule->rulenum;
2566         rule->set = ioc_rule->set;
2567         rule->usr_flags = ioc_rule->usr_flags;
2568
2569         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2570
2571         rule->refcnt = 1;
2572         rule->cpuid = mycpuid;
2573
2574         rule->stub = stub;
2575         if (stub != NULL)
2576                 stub->rule[mycpuid] = rule;
2577
2578         return rule;
2579 }
2580
2581 static void
2582 ipfw_add_rule_dispatch(netmsg_t nmsg)
2583 {
2584         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2585         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2586         struct ip_fw *rule;
2587
2588         rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2589
2590         /*
2591          * Bump generation after ipfw_create_rule(),
2592          * since this function is blocking
2593          */
2594         ctx->ipfw_gen++;
2595
2596         /*
2597          * Insert rule into the pre-determined position
2598          */
2599         if (fwmsg->prev_rule != NULL) {
2600                 struct ip_fw *prev, *next;
2601
2602                 prev = fwmsg->prev_rule;
2603                 KKASSERT(prev->cpuid == mycpuid);
2604
2605                 next = fwmsg->next_rule;
2606                 KKASSERT(next->cpuid == mycpuid);
2607
2608                 rule->next = next;
2609                 prev->next = rule;
2610
2611                 /*
2612                  * Move to the position on the next CPU
2613                  * before the msg is forwarded.
2614                  */
2615                 fwmsg->prev_rule = prev->sibling;
2616                 fwmsg->next_rule = next->sibling;
2617         } else {
2618                 KKASSERT(fwmsg->next_rule == NULL);
2619                 rule->next = ctx->ipfw_layer3_chain;
2620                 ctx->ipfw_layer3_chain = rule;
2621         }
2622
2623         /* Link rule CPU sibling */
2624         ipfw_link_sibling(fwmsg, rule);
2625
2626         ipfw_flush_rule_ptrs(ctx);
2627
2628         if (mycpuid == 0) {
2629                 /* Statistics only need to be updated once */
2630                 ipfw_inc_static_count(rule);
2631
2632                 /* Return the rule on CPU0 */
2633                 nmsg->lmsg.u.ms_resultp = rule;
2634         }
2635
2636         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2637 }
2638
2639 static void
2640 ipfw_enable_state_dispatch(netmsg_t nmsg)
2641 {
2642         struct lwkt_msg *lmsg = &nmsg->lmsg;
2643         struct ip_fw *rule = lmsg->u.ms_resultp;
2644         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2645
2646         ctx->ipfw_gen++;
2647
2648         KKASSERT(rule->cpuid == mycpuid);
2649         KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule);
2650         KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE));
2651         rule->rule_flags |= IPFW_RULE_F_STATE;
2652         lmsg->u.ms_resultp = rule->sibling;
2653
2654         ifnet_forwardmsg(lmsg, mycpuid + 1);
2655 }
2656
2657 /*
2658  * Add a new rule to the list.  Copy the rule into a malloc'ed area,
2659  * then possibly create a rule number and add the rule to the list.
2660  * Update the rule_number in the input struct so the caller knows
2661  * it as well.
2662  */
2663 static void
2664 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
2665 {
2666         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2667         struct netmsg_ipfw fwmsg;
2668         struct netmsg_base *nmsg;
2669         struct ip_fw *f, *prev, *rule;
2670         struct ip_fw_stub *stub;
2671
2672         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2673
2674         /*
2675          * If rulenum is 0, find highest numbered rule before the
2676          * default rule, and add rule number incremental step.
2677          */
2678         if (ioc_rule->rulenum == 0) {
2679                 int step = autoinc_step;
2680
2681                 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2682                          step <= IPFW_AUTOINC_STEP_MAX);
2683
2684                 /*
2685                  * Locate the highest numbered rule before default
2686                  */
2687                 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2688                         if (f->rulenum == IPFW_DEFAULT_RULE)
2689                                 break;
2690                         ioc_rule->rulenum = f->rulenum;
2691                 }
2692                 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2693                         ioc_rule->rulenum += step;
2694         }
2695         KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2696                 ioc_rule->rulenum != 0,
2697                 ("invalid rule num %d", ioc_rule->rulenum));
2698
2699         /*
2700          * Now find the right place for the new rule in the sorted list.
2701          */
2702         for (prev = NULL, f = ctx->ipfw_layer3_chain; f;
2703              prev = f, f = f->next) {
2704                 if (f->rulenum > ioc_rule->rulenum) {
2705                         /* Found the location */
2706                         break;
2707                 }
2708         }
2709         KASSERT(f != NULL, ("no default rule?!"));
2710
2711         if (rule_flags & IPFW_RULE_F_STATE) {
2712                 int size;
2713
2714                 /*
2715                  * If the new rule will create states, then allocate
2716                  * a rule stub, which will be referenced by states
2717                  * (dyn rules)
2718                  */
2719                 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2720                 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2721         } else {
2722                 stub = NULL;
2723         }
2724
2725         /*
2726          * Duplicate the rule onto each CPU.
2727          * The rule duplicated on CPU0 will be returned.
2728          */
2729         bzero(&fwmsg, sizeof(fwmsg));
2730         nmsg = &fwmsg.base;
2731         netmsg_init(nmsg, NULL, &curthread->td_msgport,
2732                     0, ipfw_add_rule_dispatch);
2733         fwmsg.ioc_rule = ioc_rule;
2734         fwmsg.prev_rule = prev;
2735         fwmsg.next_rule = prev == NULL ? NULL : f;
2736         fwmsg.stub = stub;
2737
2738         ifnet_domsg(&nmsg->lmsg, 0);
2739         KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2740
2741         rule = nmsg->lmsg.u.ms_resultp;
2742         KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2743
2744         if (rule_flags & IPFW_RULE_F_STATE) {
2745                 /*
2746                  * Turn on state flag, _after_ everything on all
2747                  * CPUs have been setup.
2748                  */
2749                 bzero(nmsg, sizeof(*nmsg));
2750                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
2751                             0, ipfw_enable_state_dispatch);
2752                 nmsg->lmsg.u.ms_resultp = rule;
2753
2754                 ifnet_domsg(&nmsg->lmsg, 0);
2755                 KKASSERT(nmsg->lmsg.u.ms_resultp == NULL);
2756         }
2757
2758         DPRINTF("++ installed rule %d, static count now %d\n",
2759                 rule->rulenum, static_count);
2760 }
2761
2762 /**
2763  * Free storage associated with a static rule (including derived
2764  * dynamic rules).
2765  * The caller is in charge of clearing rule pointers to avoid
2766  * dangling pointers.
2767  * @return a pointer to the next entry.
2768  * Arguments are not checked, so they better be correct.
2769  * Must be called at splimp().
2770  */
2771 static struct ip_fw *
2772 ipfw_delete_rule(struct ipfw_context *ctx,
2773                  struct ip_fw *prev, struct ip_fw *rule)
2774 {
2775         struct ip_fw *n;
2776         struct ip_fw_stub *stub;
2777
2778         ctx->ipfw_gen++;
2779
2780         /* STATE flag should have been cleared before we reach here */
2781         KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2782
2783         stub = rule->stub;
2784         n = rule->next;
2785         if (prev == NULL)
2786                 ctx->ipfw_layer3_chain = n;
2787         else
2788                 prev->next = n;
2789
2790         /* Mark the rule as invalid */
2791         rule->rule_flags |= IPFW_RULE_F_INVALID;
2792         rule->next_rule = NULL;
2793         rule->sibling = NULL;
2794         rule->stub = NULL;
2795 #ifdef foo
2796         /* Don't reset cpuid here; keep various assertion working */
2797         rule->cpuid = -1;
2798 #endif
2799
2800         /* Statistics only need to be updated once */
2801         if (mycpuid == 0)
2802                 ipfw_dec_static_count(rule);
2803
2804         /* Free 'stub' on the last CPU */
2805         if (stub != NULL && mycpuid == ncpus - 1)
2806                 kfree(stub, M_IPFW);
2807
2808         /* Try to free this rule */
2809         ipfw_free_rule(rule);
2810
2811         /* Return the next rule */
2812         return n;
2813 }
2814
2815 static void
2816 ipfw_flush_dispatch(netmsg_t nmsg)
2817 {
2818         struct lwkt_msg *lmsg = &nmsg->lmsg;
2819         int kill_default = lmsg->u.ms_result;
2820         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2821         struct ip_fw *rule;
2822
2823         ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2824
2825         while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2826                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2827                 ipfw_delete_rule(ctx, NULL, rule);
2828
2829         ifnet_forwardmsg(lmsg, mycpuid + 1);
2830 }
2831
2832 static void
2833 ipfw_disable_rule_state_dispatch(netmsg_t nmsg)
2834 {
2835         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2836         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2837         struct ip_fw *rule;
2838
2839         ctx->ipfw_gen++;
2840
2841         rule = dmsg->start_rule;
2842         if (rule != NULL) {
2843                 KKASSERT(rule->cpuid == mycpuid);
2844
2845                 /*
2846                  * Move to the position on the next CPU
2847                  * before the msg is forwarded.
2848                  */
2849                 dmsg->start_rule = rule->sibling;
2850         } else {
2851                 KKASSERT(dmsg->rulenum == 0);
2852                 rule = ctx->ipfw_layer3_chain;
2853         }
2854
2855         while (rule != NULL) {
2856                 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2857                         break;
2858                 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2859                 rule = rule->next;
2860         }
2861
2862         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2863 }
2864
2865 /*
2866  * Deletes all rules from a chain (including the default rule
2867  * if the second argument is set).
2868  * Must be called at splimp().
2869  */
2870 static void
2871 ipfw_flush(int kill_default)
2872 {
2873         struct netmsg_del dmsg;
2874         struct netmsg_base nmsg;
2875         struct lwkt_msg *lmsg;
2876         struct ip_fw *rule;
2877         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2878
2879         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2880
2881         /*
2882          * If 'kill_default' then caller has done the necessary
2883          * msgport syncing; unnecessary to do it again.
2884          */
2885         if (!kill_default) {
2886                 /*
2887                  * Let ipfw_chk() know the rules are going to
2888                  * be flushed, so it could jump directly to
2889                  * the default rule.
2890                  */
2891                 ipfw_flushing = 1;
2892                 netmsg_service_sync();
2893         }
2894
2895         /*
2896          * Clear STATE flag on rules, so no more states (dyn rules)
2897          * will be created.
2898          */
2899         bzero(&dmsg, sizeof(dmsg));
2900         netmsg_init(&dmsg.base, NULL, &curthread->td_msgport,
2901                     0, ipfw_disable_rule_state_dispatch);
2902         ifnet_domsg(&dmsg.base.lmsg, 0);
2903
2904         /*
2905          * This actually nukes all states (dyn rules)
2906          */
2907         lockmgr(&dyn_lock, LK_EXCLUSIVE);
2908         for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2909                 /*
2910                  * Can't check IPFW_RULE_F_STATE here,
2911                  * since it has been cleared previously.
2912                  * Check 'stub' instead.
2913                  */
2914                 if (rule->stub != NULL) {
2915                         /* Force removal */
2916                         remove_dyn_rule_locked(rule, NULL);
2917                 }
2918         }
2919         lockmgr(&dyn_lock, LK_RELEASE);
2920
2921         /*
2922          * Press the 'flush' button
2923          */
2924         bzero(&nmsg, sizeof(nmsg));
2925         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
2926                     0, ipfw_flush_dispatch);
2927         lmsg = &nmsg.lmsg;
2928         lmsg->u.ms_result = kill_default;
2929         ifnet_domsg(lmsg, 0);
2930
2931         KASSERT(dyn_count == 0, ("%u dyn rule remains", dyn_count));
2932
2933         if (kill_default) {
2934                 if (ipfw_dyn_v != NULL) {
2935                         /*
2936                          * Free dynamic rules(state) hash table
2937                          */
2938                         kfree(ipfw_dyn_v, M_IPFW);
2939                         ipfw_dyn_v = NULL;
2940                 }
2941
2942                 KASSERT(static_count == 0,
2943                         ("%u static rules remain", static_count));
2944                 KASSERT(static_ioc_len == 0,
2945                         ("%u bytes of static rules remain", static_ioc_len));
2946         } else {
2947                 KASSERT(static_count == 1,
2948                         ("%u static rules remain", static_count));
2949                 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2950                         ("%u bytes of static rules remain, should be %lu",
2951                          static_ioc_len,
2952                          (u_long)IOC_RULESIZE(ctx->ipfw_default_rule)));
2953         }
2954
2955         /* Flush is done */
2956         ipfw_flushing = 0;
2957 }
2958
2959 static void
2960 ipfw_alt_delete_rule_dispatch(netmsg_t nmsg)
2961 {
2962         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2963         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2964         struct ip_fw *rule, *prev;
2965
2966         rule = dmsg->start_rule;
2967         KKASSERT(rule->cpuid == mycpuid);
2968         dmsg->start_rule = rule->sibling;
2969
2970         prev = dmsg->prev_rule;
2971         if (prev != NULL) {
2972                 KKASSERT(prev->cpuid == mycpuid);
2973
2974                 /*
2975                  * Move to the position on the next CPU
2976                  * before the msg is forwarded.
2977                  */
2978                 dmsg->prev_rule = prev->sibling;
2979         }
2980
2981         /*
2982          * flush pointers outside the loop, then delete all matching
2983          * rules.  'prev' remains the same throughout the cycle.
2984          */
2985         ipfw_flush_rule_ptrs(ctx);
2986         while (rule && rule->rulenum == dmsg->rulenum)
2987                 rule = ipfw_delete_rule(ctx, prev, rule);
2988
2989         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2990 }
2991
2992 static int
2993 ipfw_alt_delete_rule(uint16_t rulenum)
2994 {
2995         struct ip_fw *prev, *rule, *f;
2996         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2997         struct netmsg_del dmsg;
2998         struct netmsg_base *nmsg;
2999         int state;
3000
3001         /*
3002          * Locate first rule to delete
3003          */
3004         for (prev = NULL, rule = ctx->ipfw_layer3_chain;
3005              rule && rule->rulenum < rulenum;
3006              prev = rule, rule = rule->next)
3007                 ; /* EMPTY */
3008         if (rule->rulenum != rulenum)
3009                 return EINVAL;
3010
3011         /*
3012          * Check whether any rules with the given number will
3013          * create states.
3014          */
3015         state = 0;
3016         for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3017                 if (f->rule_flags & IPFW_RULE_F_STATE) {
3018                         state = 1;
3019                         break;
3020                 }
3021         }
3022
3023         if (state) {
3024                 /*
3025                  * Clear the STATE flag, so no more states will be
3026                  * created based the rules numbered 'rulenum'.
3027                  */
3028                 bzero(&dmsg, sizeof(dmsg));
3029                 nmsg = &dmsg.base;
3030                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3031                             0, ipfw_disable_rule_state_dispatch);
3032                 dmsg.start_rule = rule;
3033                 dmsg.rulenum = rulenum;
3034
3035                 ifnet_domsg(&nmsg->lmsg, 0);
3036                 KKASSERT(dmsg.start_rule == NULL);
3037
3038                 /*
3039                  * Nuke all related states
3040                  */
3041                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3042                 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3043                         /*
3044                          * Can't check IPFW_RULE_F_STATE here,
3045                          * since it has been cleared previously.
3046                          * Check 'stub' instead.
3047                          */
3048                         if (f->stub != NULL) {
3049                                 /* Force removal */
3050                                 remove_dyn_rule_locked(f, NULL);
3051                         }
3052                 }
3053                 lockmgr(&dyn_lock, LK_RELEASE);
3054         }
3055
3056         /*
3057          * Get rid of the rule duplications on all CPUs
3058          */
3059         bzero(&dmsg, sizeof(dmsg));
3060         nmsg = &dmsg.base;
3061         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3062                     0, ipfw_alt_delete_rule_dispatch);
3063         dmsg.prev_rule = prev;
3064         dmsg.start_rule = rule;
3065         dmsg.rulenum = rulenum;
3066
3067         ifnet_domsg(&nmsg->lmsg, 0);
3068         KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3069         return 0;
3070 }
3071
3072 static void
3073 ipfw_alt_delete_ruleset_dispatch(netmsg_t nmsg)
3074 {
3075         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3076         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3077         struct ip_fw *prev, *rule;
3078 #ifdef INVARIANTS
3079         int del = 0;
3080 #endif
3081
3082         ipfw_flush_rule_ptrs(ctx);
3083
3084         prev = NULL;
3085         rule = ctx->ipfw_layer3_chain;
3086         while (rule != NULL) {
3087                 if (rule->set == dmsg->from_set) {
3088                         rule = ipfw_delete_rule(ctx, prev, rule);
3089 #ifdef INVARIANTS
3090                         del = 1;
3091 #endif
3092                 } else {
3093                         prev = rule;
3094                         rule = rule->next;
3095                 }
3096         }
3097         KASSERT(del, ("no match set?!"));
3098
3099         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3100 }
3101
3102 static void
3103 ipfw_disable_ruleset_state_dispatch(netmsg_t nmsg)
3104 {
3105         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3106         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3107         struct ip_fw *rule;
3108 #ifdef INVARIANTS
3109         int cleared = 0;
3110 #endif
3111
3112         ctx->ipfw_gen++;
3113
3114         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3115                 if (rule->set == dmsg->from_set) {
3116 #ifdef INVARIANTS
3117                         cleared = 1;
3118 #endif
3119                         rule->rule_flags &= ~IPFW_RULE_F_STATE;
3120                 }
3121         }
3122         KASSERT(cleared, ("no match set?!"));
3123
3124         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3125 }
3126
3127 static int
3128 ipfw_alt_delete_ruleset(uint8_t set)
3129 {
3130         struct netmsg_del dmsg;
3131         struct netmsg_base *nmsg;
3132         int state, del;
3133         struct ip_fw *rule;
3134         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3135
3136         /*
3137          * Check whether the 'set' exists.  If it exists,
3138          * then check whether any rules within the set will
3139          * try to create states.
3140          */
3141         state = 0;
3142         del = 0;
3143         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3144                 if (rule->set == set) {
3145                         del = 1;
3146                         if (rule->rule_flags & IPFW_RULE_F_STATE) {
3147                                 state = 1;
3148                                 break;
3149                         }
3150                 }
3151         }
3152         if (!del)
3153                 return 0; /* XXX EINVAL? */
3154
3155         if (state) {
3156                 /*
3157                  * Clear the STATE flag, so no more states will be
3158                  * created based the rules in this set.
3159                  */
3160                 bzero(&dmsg, sizeof(dmsg));
3161                 nmsg = &dmsg.base;
3162                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3163                             0, ipfw_disable_ruleset_state_dispatch);
3164                 dmsg.from_set = set;
3165
3166                 ifnet_domsg(&nmsg->lmsg, 0);
3167
3168                 /*
3169                  * Nuke all related states
3170                  */
3171                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3172                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3173                         if (rule->set != set)
3174                                 continue;
3175
3176                         /*
3177                          * Can't check IPFW_RULE_F_STATE here,
3178                          * since it has been cleared previously.
3179                          * Check 'stub' instead.
3180                          */
3181                         if (rule->stub != NULL) {
3182                                 /* Force removal */
3183                                 remove_dyn_rule_locked(rule, NULL);
3184                         }
3185                 }
3186                 lockmgr(&dyn_lock, LK_RELEASE);
3187         }
3188
3189         /*
3190          * Delete this set
3191          */
3192         bzero(&dmsg, sizeof(dmsg));
3193         nmsg = &dmsg.base;
3194         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3195                     0, ipfw_alt_delete_ruleset_dispatch);
3196         dmsg.from_set = set;
3197
3198         ifnet_domsg(&nmsg->lmsg, 0);
3199         return 0;
3200 }
3201
3202 static void
3203 ipfw_alt_move_rule_dispatch(netmsg_t nmsg)
3204 {
3205         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3206         struct ip_fw *rule;
3207
3208         rule = dmsg->start_rule;
3209         KKASSERT(rule->cpuid == mycpuid);
3210
3211         /*
3212          * Move to the position on the next CPU
3213          * before the msg is forwarded.
3214          */
3215         dmsg->start_rule = rule->sibling;
3216
3217         while (rule && rule->rulenum <= dmsg->rulenum) {
3218                 if (rule->rulenum == dmsg->rulenum)
3219                         rule->set = dmsg->to_set;
3220                 rule = rule->next;
3221         }
3222         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3223 }
3224
3225 static int
3226 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3227 {
3228         struct netmsg_del dmsg;
3229         struct netmsg_base *nmsg;
3230         struct ip_fw *rule;
3231         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3232
3233         /*
3234          * Locate first rule to move
3235          */
3236         for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3237              rule = rule->next) {
3238                 if (rule->rulenum == rulenum && rule->set != set)
3239                         break;
3240         }
3241         if (rule == NULL || rule->rulenum > rulenum)
3242                 return 0; /* XXX error? */
3243
3244         bzero(&dmsg, sizeof(dmsg));
3245         nmsg = &dmsg.base;
3246         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3247                     0, ipfw_alt_move_rule_dispatch);
3248         dmsg.start_rule = rule;
3249         dmsg.rulenum = rulenum;
3250         dmsg.to_set = set;
3251
3252         ifnet_domsg(&nmsg->lmsg, 0);
3253         KKASSERT(dmsg.start_rule == NULL);
3254         return 0;
3255 }
3256
3257 static void
3258 ipfw_alt_move_ruleset_dispatch(netmsg_t nmsg)
3259 {
3260         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3261         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3262         struct ip_fw *rule;
3263
3264         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3265                 if (rule->set == dmsg->from_set)
3266                         rule->set = dmsg->to_set;
3267         }
3268         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3269 }
3270
3271 static int
3272 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3273 {
3274         struct netmsg_del dmsg;
3275         struct netmsg_base *nmsg;
3276
3277         bzero(&dmsg, sizeof(dmsg));
3278         nmsg = &dmsg.base;
3279         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3280                     0, ipfw_alt_move_ruleset_dispatch);
3281         dmsg.from_set = from_set;
3282         dmsg.to_set = to_set;
3283
3284         ifnet_domsg(&nmsg->lmsg, 0);
3285         return 0;
3286 }
3287
3288 static void
3289 ipfw_alt_swap_ruleset_dispatch(netmsg_t nmsg)
3290 {
3291         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3292         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3293         struct ip_fw *rule;
3294
3295         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3296                 if (rule->set == dmsg->from_set)
3297                         rule->set = dmsg->to_set;
3298                 else if (rule->set == dmsg->to_set)
3299                         rule->set = dmsg->from_set;
3300         }
3301         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3302 }
3303
3304 static int
3305 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3306 {
3307         struct netmsg_del dmsg;
3308         struct netmsg_base *nmsg;
3309
3310         bzero(&dmsg, sizeof(dmsg));
3311         nmsg = &dmsg.base;
3312         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3313                     0, ipfw_alt_swap_ruleset_dispatch);
3314         dmsg.from_set = set1;
3315         dmsg.to_set = set2;
3316
3317         ifnet_domsg(&nmsg->lmsg, 0);
3318         return 0;
3319 }
3320
3321 /**
3322  * Remove all rules with given number, and also do set manipulation.
3323  *
3324  * The argument is an uint32_t. The low 16 bit are the rule or set number,
3325  * the next 8 bits are the new set, the top 8 bits are the command:
3326  *
3327  *      0       delete rules with given number
3328  *      1       delete rules with given set number
3329  *      2       move rules with given number to new set
3330  *      3       move rules with given set number to new set
3331  *      4       swap sets with given numbers
3332  */
3333 static int
3334 ipfw_ctl_alter(uint32_t arg)
3335 {
3336         uint16_t rulenum;
3337         uint8_t cmd, new_set;
3338         int error = 0;
3339
3340         rulenum = arg & 0xffff;
3341         cmd = (arg >> 24) & 0xff;
3342         new_set = (arg >> 16) & 0xff;
3343
3344         if (cmd > 4)
3345                 return EINVAL;
3346         if (new_set >= IPFW_DEFAULT_SET)
3347                 return EINVAL;
3348         if (cmd == 0 || cmd == 2) {
3349                 if (rulenum == IPFW_DEFAULT_RULE)
3350                         return EINVAL;
3351         } else {
3352                 if (rulenum >= IPFW_DEFAULT_SET)
3353                         return EINVAL;
3354         }
3355
3356         switch (cmd) {
3357         case 0: /* delete rules with given number */
3358                 error = ipfw_alt_delete_rule(rulenum);
3359                 break;
3360
3361         case 1: /* delete all rules with given set number */
3362                 error = ipfw_alt_delete_ruleset(rulenum);
3363                 break;
3364
3365         case 2: /* move rules with given number to new set */
3366                 error = ipfw_alt_move_rule(rulenum, new_set);
3367                 break;
3368
3369         case 3: /* move rules with given set number to new set */
3370                 error = ipfw_alt_move_ruleset(rulenum, new_set);
3371                 break;
3372
3373         case 4: /* swap two sets */
3374                 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3375                 break;
3376         }
3377         return error;
3378 }
3379
3380 /*
3381  * Clear counters for a specific rule.
3382  */
3383 static void
3384 clear_counters(struct ip_fw *rule, int log_only)
3385 {
3386         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3387
3388         if (log_only == 0) {
3389                 rule->bcnt = rule->pcnt = 0;
3390                 rule->timestamp = 0;
3391         }
3392         if (l->o.opcode == O_LOG)
3393                 l->log_left = l->max_log;
3394 }
3395
3396 static void
3397 ipfw_zero_entry_dispatch(netmsg_t nmsg)
3398 {
3399         struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3400         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3401         struct ip_fw *rule;
3402
3403         if (zmsg->rulenum == 0) {
3404                 KKASSERT(zmsg->start_rule == NULL);
3405
3406                 ctx->ipfw_norule_counter = 0;
3407                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3408                         clear_counters(rule, zmsg->log_only);
3409         } else {
3410                 struct ip_fw *start = zmsg->start_rule;
3411
3412                 KKASSERT(start->cpuid == mycpuid);
3413                 KKASSERT(start->rulenum == zmsg->rulenum);
3414
3415                 /*
3416                  * We can have multiple rules with the same number, so we
3417                  * need to clear them all.
3418                  */
3419                 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3420                      rule = rule->next)
3421                         clear_counters(rule, zmsg->log_only);
3422
3423                 /*
3424                  * Move to the position on the next CPU
3425                  * before the msg is forwarded.
3426                  */
3427                 zmsg->start_rule = start->sibling;
3428         }
3429         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3430 }
3431
3432 /**
3433  * Reset some or all counters on firewall rules.
3434  * @arg frwl is null to clear all entries, or contains a specific
3435  * rule number.
3436  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3437  */
3438 static int
3439 ipfw_ctl_zero_entry(int rulenum, int log_only)
3440 {
3441         struct netmsg_zent zmsg;
3442         struct netmsg_base *nmsg;
3443         const char *msg;
3444         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3445
3446         bzero(&zmsg, sizeof(zmsg));
3447         nmsg = &zmsg.base;
3448         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3449                     0, ipfw_zero_entry_dispatch);
3450         zmsg.log_only = log_only;
3451
3452         if (rulenum == 0) {
3453                 msg = log_only ? "ipfw: All logging counts reset.\n"
3454                                : "ipfw: Accounting cleared.\n";
3455         } else {
3456                 struct ip_fw *rule;
3457
3458                 /*
3459                  * Locate the first rule with 'rulenum'
3460                  */
3461                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3462                         if (rule->rulenum == rulenum)
3463                                 break;
3464                 }
3465                 if (rule == NULL) /* we did not find any matching rules */
3466                         return (EINVAL);
3467                 zmsg.start_rule = rule;
3468                 zmsg.rulenum = rulenum;
3469
3470                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3471                                : "ipfw: Entry %d cleared.\n";
3472         }
3473         ifnet_domsg(&nmsg->lmsg, 0);
3474         KKASSERT(zmsg.start_rule == NULL);
3475
3476         if (fw_verbose)
3477                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3478         return (0);
3479 }
3480
3481 /*
3482  * Check validity of the structure before insert.
3483  * Fortunately rules are simple, so this mostly need to check rule sizes.
3484  */
3485 static int
3486 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3487 {
3488         int l, cmdlen = 0;
3489         int have_action = 0;
3490         ipfw_insn *cmd;
3491
3492         *rule_flags = 0;
3493
3494         /* Check for valid size */
3495         if (size < sizeof(*rule)) {
3496                 kprintf("ipfw: rule too short\n");
3497                 return EINVAL;
3498         }
3499         l = IOC_RULESIZE(rule);
3500         if (l != size) {
3501                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3502                 return EINVAL;
3503         }
3504
3505         /* Check rule number */
3506         if (rule->rulenum == IPFW_DEFAULT_RULE) {
3507                 kprintf("ipfw: invalid rule number\n");
3508                 return EINVAL;
3509         }
3510
3511         /*
3512          * Now go for the individual checks. Very simple ones, basically only
3513          * instruction sizes.
3514          */
3515         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3516              l -= cmdlen, cmd += cmdlen) {
3517                 cmdlen = F_LEN(cmd);
3518                 if (cmdlen > l) {
3519                         kprintf("ipfw: opcode %d size truncated\n",
3520                                 cmd->opcode);
3521                         return EINVAL;
3522                 }
3523
3524                 DPRINTF("ipfw: opcode %d\n", cmd->opcode);
3525
3526                 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3527                         /* This rule will create states */
3528                         *rule_flags |= IPFW_RULE_F_STATE;
3529                 }
3530
3531                 switch (cmd->opcode) {
3532                 case O_NOP:
3533                 case O_PROBE_STATE:
3534                 case O_KEEP_STATE:
3535                 case O_PROTO:
3536                 case O_IP_SRC_ME:
3537                 case O_IP_DST_ME:
3538                 case O_LAYER2:
3539                 case O_IN:
3540                 case O_FRAG:
3541                 case O_IPOPT:
3542                 case O_IPLEN:
3543                 case O_IPID:
3544                 case O_IPTOS:
3545                 case O_IPPRECEDENCE:
3546                 case O_IPTTL:
3547                 case O_IPVER:
3548                 case O_TCPWIN:
3549                 case O_TCPFLAGS:
3550                 case O_TCPOPTS:
3551                 case O_ESTAB:
3552                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3553                                 goto bad_size;
3554                         break;
3555
3556                 case O_UID:
3557                 case O_GID:
3558                 case O_IP_SRC:
3559                 case O_IP_DST:
3560                 case O_TCPSEQ:
3561                 case O_TCPACK:
3562                 case O_PROB:
3563                 case O_ICMPTYPE:
3564                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3565                                 goto bad_size;
3566                         break;
3567
3568                 case O_LIMIT:
3569                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3570                                 goto bad_size;
3571                         break;
3572
3573                 case O_LOG:
3574                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3575                                 goto bad_size;
3576
3577                         ((ipfw_insn_log *)cmd)->log_left =
3578                             ((ipfw_insn_log *)cmd)->max_log;
3579
3580                         break;
3581
3582                 case O_IP_SRC_MASK:
3583                 case O_IP_DST_MASK:
3584                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3585                                 goto bad_size;
3586                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3587                                 kprintf("ipfw: opcode %d, useless rule\n",
3588                                         cmd->opcode);
3589                                 return EINVAL;
3590                         }
3591                         break;
3592
3593                 case O_IP_SRC_SET:
3594                 case O_IP_DST_SET:
3595                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3596                                 kprintf("ipfw: invalid set size %d\n",
3597                                         cmd->arg1);
3598                                 return EINVAL;
3599                         }
3600                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3601                             (cmd->arg1+31)/32 )
3602                                 goto bad_size;
3603                         break;
3604
3605                 case O_MACADDR2:
3606                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3607                                 goto bad_size;
3608                         break;
3609
3610                 case O_MAC_TYPE:
3611                 case O_IP_SRCPORT:
3612                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3613                         if (cmdlen < 2 || cmdlen > 31)
3614                                 goto bad_size;
3615                         break;
3616
3617                 case O_RECV:
3618                 case O_XMIT:
3619                 case O_VIA:
3620                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3621                                 goto bad_size;
3622                         break;
3623
3624                 case O_PIPE:
3625                 case O_QUEUE:
3626                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3627                                 goto bad_size;
3628                         goto check_action;
3629
3630                 case O_FORWARD_IP:
3631                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3632                                 goto bad_size;
3633                         } else {
3634                                 in_addr_t fwd_addr;
3635
3636                                 fwd_addr = ((ipfw_insn_sa *)cmd)->
3637                                            sa.sin_addr.s_addr;
3638                                 if (IN_MULTICAST(ntohl(fwd_addr))) {
3639                                         kprintf("ipfw: try forwarding to "
3640                                                 "multicast address\n");
3641                                         return EINVAL;
3642                                 }
3643                         }
3644                         goto check_action;
3645
3646                 case O_FORWARD_MAC: /* XXX not implemented yet */
3647                 case O_CHECK_STATE:
3648                 case O_COUNT:
3649                 case O_ACCEPT:
3650                 case O_DENY:
3651                 case O_REJECT:
3652                 case O_SKIPTO:
3653                 case O_DIVERT:
3654                 case O_TEE:
3655                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3656                                 goto bad_size;
3657 check_action:
3658                         if (have_action) {
3659                                 kprintf("ipfw: opcode %d, multiple actions"
3660                                         " not allowed\n",
3661                                         cmd->opcode);
3662                                 return EINVAL;
3663                         }
3664                         have_action = 1;
3665                         if (l != cmdlen) {
3666                                 kprintf("ipfw: opcode %d, action must be"
3667                                         " last opcode\n",
3668                                         cmd->opcode);
3669                                 return EINVAL;
3670                         }
3671                         break;
3672                 default:
3673                         kprintf("ipfw: opcode %d, unknown opcode\n",
3674                                 cmd->opcode);
3675                         return EINVAL;
3676                 }
3677         }
3678         if (have_action == 0) {
3679                 kprintf("ipfw: missing action\n");
3680                 return EINVAL;
3681         }
3682         return 0;
3683
3684 bad_size:
3685         kprintf("ipfw: opcode %d size %d wrong\n",
3686                 cmd->opcode, cmdlen);
3687         return EINVAL;
3688 }
3689
3690 static int
3691 ipfw_ctl_add_rule(struct sockopt *sopt)
3692 {
3693         struct ipfw_ioc_rule *ioc_rule;
3694         size_t size;
3695         uint32_t rule_flags;
3696         int error;
3697
3698         size = sopt->sopt_valsize;
3699         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3700             size < sizeof(*ioc_rule)) {
3701                 return EINVAL;
3702         }
3703         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3704                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3705                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3706         }
3707         ioc_rule = sopt->sopt_val;
3708
3709         error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3710         if (error)
3711                 return error;
3712
3713         ipfw_add_rule(ioc_rule, rule_flags);
3714
3715         if (sopt->sopt_dir == SOPT_GET)
3716                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3717         return 0;
3718 }
3719
3720 static void *
3721 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3722 {
3723         const struct ip_fw *sibling;
3724 #ifdef INVARIANTS
3725         int i;
3726 #endif
3727
3728         KKASSERT(rule->cpuid == IPFW_CFGCPUID);
3729
3730         ioc_rule->act_ofs = rule->act_ofs;
3731         ioc_rule->cmd_len = rule->cmd_len;
3732         ioc_rule->rulenum = rule->rulenum;
3733         ioc_rule->set = rule->set;
3734         ioc_rule->usr_flags = rule->usr_flags;
3735
3736         ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3737         ioc_rule->static_count = static_count;
3738         ioc_rule->static_len = static_ioc_len;
3739
3740         /*
3741          * Visit (read-only) all of the rule's duplications to get
3742          * the necessary statistics
3743          */
3744 #ifdef INVARIANTS
3745         i = 0;
3746 #endif
3747         ioc_rule->pcnt = 0;
3748         ioc_rule->bcnt = 0;
3749         ioc_rule->timestamp = 0;
3750         for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3751                 ioc_rule->pcnt += sibling->pcnt;
3752                 ioc_rule->bcnt += sibling->bcnt;
3753                 if (sibling->timestamp > ioc_rule->timestamp)
3754                         ioc_rule->timestamp = sibling->timestamp;
3755 #ifdef INVARIANTS
3756                 ++i;
3757 #endif
3758         }
3759         KASSERT(i == ncpus, ("static rule is not duplicated on every cpu"));
3760
3761         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3762
3763         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3764 }
3765
3766 static void
3767 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3768                 struct ipfw_ioc_state *ioc_state)
3769 {
3770         const struct ipfw_flow_id *id;
3771         struct ipfw_ioc_flowid *ioc_id;
3772
3773         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3774                             0 : dyn_rule->expire - time_second;
3775         ioc_state->pcnt = dyn_rule->pcnt;
3776         ioc_state->bcnt = dyn_rule->bcnt;
3777
3778         ioc_state->dyn_type = dyn_rule->dyn_type;
3779         ioc_state->count = dyn_rule->count;
3780
3781         ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3782
3783         id = &dyn_rule->id;
3784         ioc_id = &ioc_state->id;
3785
3786         ioc_id->type = ETHERTYPE_IP;
3787         ioc_id->u.ip.dst_ip = id->dst_ip;
3788         ioc_id->u.ip.src_ip = id->src_ip;
3789         ioc_id->u.ip.dst_port = id->dst_port;
3790         ioc_id->u.ip.src_port = id->src_port;
3791         ioc_id->u.ip.proto = id->proto;
3792 }
3793
3794 static int
3795 ipfw_ctl_get_rules(struct sockopt *sopt)
3796 {
3797         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3798         struct ip_fw *rule;
3799         void *bp;
3800         size_t size;
3801         uint32_t dcount = 0;
3802
3803         /*
3804          * pass up a copy of the current rules. Static rules
3805          * come first (the last of which has number IPFW_DEFAULT_RULE),
3806          * followed by a possibly empty list of dynamic rule.
3807          */
3808
3809         size = static_ioc_len;  /* size of static rules */
3810         if (ipfw_dyn_v) {       /* add size of dyn.rules */
3811                 dcount = dyn_count;
3812                 size += dcount * sizeof(struct ipfw_ioc_state);
3813         }
3814
3815         if (sopt->sopt_valsize < size) {
3816                 /* short length, no need to return incomplete rules */
3817                 /* XXX: if superuser, no need to zero buffer */
3818                 bzero(sopt->sopt_val, sopt->sopt_valsize);
3819                 return 0;
3820         }
3821         bp = sopt->sopt_val;
3822
3823         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3824                 bp = ipfw_copy_rule(rule, bp);
3825
3826         if (ipfw_dyn_v && dcount != 0) {
3827                 struct ipfw_ioc_state *ioc_state = bp;
3828                 uint32_t dcount2 = 0;
3829 #ifdef INVARIANTS
3830                 size_t old_size = size;
3831 #endif
3832                 int i;
3833
3834                 lockmgr(&dyn_lock, LK_SHARED);
3835
3836                 /* Check 'ipfw_dyn_v' again with lock held */
3837                 if (ipfw_dyn_v == NULL)
3838                         goto skip;
3839
3840                 for (i = 0; i < curr_dyn_buckets; i++) {
3841                         ipfw_dyn_rule *p;
3842
3843                         /*
3844                          * The # of dynamic rules may have grown after the
3845                          * snapshot of 'dyn_count' was taken, so we will have
3846                          * to check 'dcount' (snapshot of dyn_count) here to
3847                          * make sure that we don't overflow the pre-allocated
3848                          * buffer.
3849                          */
3850                         for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
3851                              p = p->next, ioc_state++, dcount--, dcount2++)
3852                                 ipfw_copy_state(p, ioc_state);
3853                 }
3854 skip:
3855                 lockmgr(&dyn_lock, LK_RELEASE);
3856
3857                 /*
3858                  * The # of dynamic rules may be shrinked after the
3859                  * snapshot of 'dyn_count' was taken.  To give user a
3860                  * correct dynamic rule count, we use the 'dcount2'
3861                  * calculated above (with shared lockmgr lock held).
3862                  */
3863                 size = static_ioc_len +
3864                        (dcount2 * sizeof(struct ipfw_ioc_state));
3865                 KKASSERT(size <= old_size);
3866         }
3867
3868         sopt->sopt_valsize = size;
3869         return 0;
3870 }
3871
3872 static void
3873 ipfw_set_disable_dispatch(netmsg_t nmsg)
3874 {
3875         struct lwkt_msg *lmsg = &nmsg->lmsg;
3876         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3877
3878         ctx->ipfw_gen++;
3879         ctx->ipfw_set_disable = lmsg->u.ms_result32;
3880
3881         ifnet_forwardmsg(lmsg, mycpuid + 1);
3882 }
3883
3884 static void
3885 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable)
3886 {
3887         struct netmsg_base nmsg;
3888         struct lwkt_msg *lmsg;
3889         uint32_t set_disable;
3890
3891         /* IPFW_DEFAULT_SET is always enabled */
3892         enable |= (1 << IPFW_DEFAULT_SET);
3893         set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable;
3894
3895         bzero(&nmsg, sizeof(nmsg));
3896         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
3897                     0, ipfw_set_disable_dispatch);
3898         lmsg = &nmsg.lmsg;
3899         lmsg->u.ms_result32 = set_disable;
3900
3901         ifnet_domsg(lmsg, 0);
3902 }
3903
3904 /**
3905  * {set|get}sockopt parser.
3906  */
3907 static int
3908 ipfw_ctl(struct sockopt *sopt)
3909 {
3910         int error, rulenum;
3911         uint32_t *masks;
3912         size_t size;
3913
3914         error = 0;
3915
3916         switch (sopt->sopt_name) {
3917         case IP_FW_GET:
3918                 error = ipfw_ctl_get_rules(sopt);
3919                 break;
3920
3921         case IP_FW_FLUSH:
3922                 ipfw_flush(0 /* keep default rule */);
3923                 break;
3924
3925         case IP_FW_ADD:
3926                 error = ipfw_ctl_add_rule(sopt);
3927                 break;
3928
3929         case IP_FW_DEL:
3930                 /*
3931                  * IP_FW_DEL is used for deleting single rules or sets,
3932                  * and (ab)used to atomically manipulate sets.
3933                  * Argument size is used to distinguish between the two:
3934                  *    sizeof(uint32_t)
3935                  *      delete single rule or set of rules,
3936                  *      or reassign rules (or sets) to a different set.
3937                  *    2 * sizeof(uint32_t)
3938                  *      atomic disable/enable sets.
3939                  *      first uint32_t contains sets to be disabled,
3940                  *      second uint32_t contains sets to be enabled.
3941                  */
3942                 masks = sopt->sopt_val;
3943                 size = sopt->sopt_valsize;
3944                 if (size == sizeof(*masks)) {
3945                         /*
3946                          * Delete or reassign static rule
3947                          */
3948                         error = ipfw_ctl_alter(masks[0]);
3949                 } else if (size == (2 * sizeof(*masks))) {
3950                         /*
3951                          * Set enable/disable
3952                          */
3953                         ipfw_ctl_set_disable(masks[0], masks[1]);
3954                 } else {
3955                         error = EINVAL;
3956                 }
3957                 break;
3958
3959         case IP_FW_ZERO:
3960         case IP_FW_RESETLOG: /* argument is an int, the rule number */
3961                 rulenum = 0;
3962
3963                 if (sopt->sopt_val != 0) {
3964                     error = soopt_to_kbuf(sopt, &rulenum,
3965                             sizeof(int), sizeof(int));
3966                     if (error)
3967                         break;
3968                 }
3969                 error = ipfw_ctl_zero_entry(rulenum,
3970                         sopt->sopt_name == IP_FW_RESETLOG);
3971                 break;
3972
3973         default:
3974                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3975                 error = EINVAL;
3976         }
3977         return error;
3978 }
3979
3980 /*
3981  * This procedure is only used to handle keepalives. It is invoked
3982  * every dyn_keepalive_period
3983  */
3984 static void
3985 ipfw_tick_dispatch(netmsg_t nmsg)
3986 {
3987         time_t keep_alive;
3988         uint32_t gen;
3989         int i;
3990
3991         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
3992         KKASSERT(IPFW_LOADED);
3993
3994         /* Reply ASAP */
3995         crit_enter();
3996         lwkt_replymsg(&nmsg->lmsg, 0);
3997         crit_exit();
3998
3999         if (ipfw_dyn_v == NULL || dyn_count == 0)
4000                 goto done;
4001
4002         keep_alive = time_second;
4003
4004         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4005 again:
4006         if (ipfw_dyn_v == NULL || dyn_count == 0) {
4007                 lockmgr(&dyn_lock, LK_RELEASE);
4008                 goto done;
4009         }
4010         gen = dyn_buckets_gen;
4011
4012         for (i = 0; i < curr_dyn_buckets; i++) {
4013                 ipfw_dyn_rule *q, *prev;
4014
4015                 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
4016                         uint32_t ack_rev, ack_fwd;
4017                         struct ipfw_flow_id id;
4018
4019                         if (q->dyn_type == O_LIMIT_PARENT)
4020                                 goto next;
4021
4022                         if (TIME_LEQ(q->expire, time_second)) {
4023                                 /* State expired */
4024                                 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
4025                                 continue;
4026                         }
4027
4028                         /*
4029                          * Keep alive processing
4030                          */
4031
4032                         if (!dyn_keepalive)
4033                                 goto next;
4034                         if (q->id.proto != IPPROTO_TCP)
4035                                 goto next;
4036                         if ((q->state & BOTH_SYN) != BOTH_SYN)
4037                                 goto next;
4038                         if (TIME_LEQ(time_second + dyn_keepalive_interval,
4039                             q->expire))
4040                                 goto next;      /* too early */
4041                         if (q->keep_alive == keep_alive)
4042                                 goto next;      /* alreay done */
4043
4044                         /*
4045                          * Save necessary information, so that they could
4046                          * survive after possible blocking in send_pkt()
4047                          */
4048                         id = q->id;
4049                         ack_rev = q->ack_rev;
4050                         ack_fwd = q->ack_fwd;
4051
4052                         /* Sending has been started */
4053                         q->keep_alive = keep_alive;
4054
4055                         /* Release lock to avoid possible dead lock */
4056                         lockmgr(&dyn_lock, LK_RELEASE);
4057                         send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
4058                         send_pkt(&id, ack_fwd - 1, ack_rev, 0);
4059                         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4060
4061                         if (gen != dyn_buckets_gen) {
4062                                 /*
4063                                  * Dyn bucket array has been changed during
4064                                  * the above two sending; reiterate.
4065                                  */
4066                                 goto again;
4067                         }
4068 next:
4069                         prev = q;
4070                         q = q->next;
4071                 }
4072         }
4073         lockmgr(&dyn_lock, LK_RELEASE);
4074 done:
4075         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
4076                       ipfw_tick, NULL);
4077 }
4078
4079 /*
4080  * This procedure is only used to handle keepalives. It is invoked
4081  * every dyn_keepalive_period
4082  */
4083 static void
4084 ipfw_tick(void *dummy __unused)
4085 {
4086         struct lwkt_msg *lmsg = &ipfw_timeout_netmsg.lmsg;
4087
4088         KKASSERT(mycpuid == IPFW_CFGCPUID);
4089
4090         crit_enter();
4091
4092         KKASSERT(lmsg->ms_flags & MSGF_DONE);
4093         if (IPFW_LOADED) {
4094                 lwkt_sendmsg(IPFW_CFGPORT, lmsg);
4095                 /* ipfw_timeout_netmsg's handler reset this callout */
4096         }
4097
4098         crit_exit();
4099 }
4100
4101 static int
4102 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4103 {
4104         struct ip_fw_args args;
4105         struct mbuf *m = *m0;
4106         struct m_tag *mtag;
4107         int tee = 0, error = 0, ret;
4108
4109         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4110                 /* Extract info from dummynet tag */
4111                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4112                 KKASSERT(mtag != NULL);
4113                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4114                 KKASSERT(args.rule != NULL);
4115
4116                 m_tag_delete(m, mtag);
4117                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4118         } else {
4119                 args.rule = NULL;
4120         }
4121
4122         args.eh = NULL;
4123         args.oif = NULL;
4124         args.m = m;
4125         ret = ipfw_chk(&args);
4126         m = args.m;
4127
4128         if (m == NULL) {
4129                 error = EACCES;
4130                 goto back;
4131         }
4132
4133         switch (ret) {
4134         case IP_FW_PASS:
4135                 break;
4136
4137         case IP_FW_DENY:
4138                 m_freem(m);
4139                 m = NULL;
4140                 error = EACCES;
4141                 break;
4142
4143         case IP_FW_DUMMYNET:
4144                 /* Send packet to the appropriate pipe */
4145                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args);
4146                 break;
4147
4148         case IP_FW_TEE:
4149                 tee = 1;
4150                 /* FALL THROUGH */
4151
4152         case IP_FW_DIVERT:
4153                 /*
4154                  * Must clear bridge tag when changing
4155                  */
4156                 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
4157                 if (ip_divert_p != NULL) {
4158                         m = ip_divert_p(m, tee, 1);
4159                 } else {
4160                         m_freem(m);
4161                         m = NULL;
4162                         /* not sure this is the right error msg */
4163                         error = EACCES;
4164                 }
4165                 break;
4166
4167         default:
4168                 panic("unknown ipfw return value: %d", ret);
4169         }
4170 back:
4171         *m0 = m;
4172         return error;
4173 }
4174
4175 static int
4176 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4177 {
4178         struct ip_fw_args args;
4179         struct mbuf *m = *m0;
4180         struct m_tag *mtag;
4181         int tee = 0, error = 0, ret;
4182
4183         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4184                 /* Extract info from dummynet tag */
4185                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4186                 KKASSERT(mtag != NULL);
4187                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4188                 KKASSERT(args.rule != NULL);
4189
4190                 m_tag_delete(m, mtag);
4191                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4192         } else {
4193                 args.rule = NULL;
4194         }
4195
4196         args.eh = NULL;
4197         args.m = m;
4198         args.oif = ifp;
4199         ret = ipfw_chk(&args);
4200         m = args.m;
4201
4202         if (m == NULL) {
4203                 error = EACCES;
4204                 goto back;
4205         }
4206
4207         switch (ret) {
4208         case IP_FW_PASS:
4209                 break;
4210
4211         case IP_FW_DENY:
4212                 m_freem(m);
4213                 m = NULL;
4214                 error = EACCES;
4215                 break;
4216
4217         case IP_FW_DUMMYNET:
4218                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args);
4219                 break;
4220
4221         case IP_FW_TEE:
4222                 tee = 1;
4223                 /* FALL THROUGH */
4224
4225         case IP_FW_DIVERT:
4226                 if (ip_divert_p != NULL) {
4227                         m = ip_divert_p(m, tee, 0);
4228                 } else {
4229                         m_freem(m);
4230                         m = NULL;
4231                         /* not sure this is the right error msg */
4232                         error = EACCES;
4233                 }
4234                 break;
4235
4236         default:
4237                 panic("unknown ipfw return value: %d", ret);
4238         }
4239 back:
4240         *m0 = m;
4241         return error;
4242 }
4243
4244 static void
4245 ipfw_hook(void)
4246 {
4247         struct pfil_head *pfh;
4248
4249         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4250
4251         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4252         if (pfh == NULL)
4253                 return;
4254
4255         pfil_add_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_MPSAFE, pfh);
4256         pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_MPSAFE, pfh);
4257 }
4258
4259 static void
4260 ipfw_dehook(void)
4261 {
4262         struct pfil_head *pfh;
4263
4264         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4265
4266         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4267         if (pfh == NULL)
4268                 return;
4269
4270         pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN, pfh);
4271         pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT, pfh);
4272 }
4273
4274 static void
4275 ipfw_sysctl_enable_dispatch(netmsg_t nmsg)
4276 {
4277         struct lwkt_msg *lmsg = &nmsg->lmsg;
4278         int enable = lmsg->u.ms_result;
4279
4280         if (fw_enable == enable)
4281                 goto reply;
4282
4283         fw_enable = enable;
4284         if (fw_enable)
4285                 ipfw_hook();
4286         else
4287                 ipfw_dehook();
4288 reply:
4289         lwkt_replymsg(lmsg, 0);
4290 }
4291
4292 static int
4293 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS)
4294 {
4295         struct netmsg_base nmsg;
4296         struct lwkt_msg *lmsg;
4297         int enable, error;
4298
4299         enable = fw_enable;
4300         error = sysctl_handle_int(oidp, &enable, 0, req);
4301         if (error || req->newptr == NULL)
4302                 return error;
4303
4304         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
4305                     0, ipfw_sysctl_enable_dispatch);
4306         lmsg = &nmsg.lmsg;
4307         lmsg->u.ms_result = enable;
4308
4309         return lwkt_domsg(IPFW_CFGPORT, lmsg, 0);
4310 }
4311
4312 static int
4313 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
4314 {
4315         return sysctl_int_range(oidp, arg1, arg2, req,
4316                IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
4317 }
4318
4319 static int
4320 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
4321 {
4322         int error, value;
4323
4324         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4325
4326         value = dyn_buckets;
4327         error = sysctl_handle_int(oidp, &value, 0, req);
4328         if (error || !req->newptr)
4329                 goto back;
4330
4331         /*
4332          * Make sure we have a power of 2 and
4333          * do not allow more than 64k entries.
4334          */
4335         error = EINVAL;
4336         if (value <= 1 || value > 65536)
4337                 goto back;
4338         if ((value & (value - 1)) != 0)
4339                 goto back;
4340
4341         error = 0;
4342         dyn_buckets = value;
4343 back:
4344         lockmgr(&dyn_lock, LK_RELEASE);
4345         return error;
4346 }
4347
4348 static int
4349 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
4350 {
4351         return sysctl_int_range(oidp, arg1, arg2, req,
4352                                 1, dyn_keepalive_period - 1);
4353 }
4354
4355 static int
4356 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
4357 {
4358         return sysctl_int_range(oidp, arg1, arg2, req,
4359                                 1, dyn_keepalive_period - 1);
4360 }
4361
4362 static void
4363 ipfw_ctx_init_dispatch(netmsg_t nmsg)
4364 {
4365         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
4366         struct ipfw_context *ctx;
4367         struct ip_fw *def_rule;
4368
4369         ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO);
4370         ipfw_ctx[mycpuid] = ctx;
4371
4372         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
4373
4374         def_rule->act_ofs = 0;
4375         def_rule->rulenum = IPFW_DEFAULT_RULE;
4376         def_rule->cmd_len = 1;
4377         def_rule->set = IPFW_DEFAULT_SET;
4378
4379         def_rule->cmd[0].len = 1;
4380 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4381         def_rule->cmd[0].opcode = O_ACCEPT;
4382 #else
4383         def_rule->cmd[0].opcode = O_DENY;
4384 #endif
4385
4386         def_rule->refcnt = 1;
4387         def_rule->cpuid = mycpuid;
4388
4389         /* Install the default rule */
4390         ctx->ipfw_default_rule = def_rule;
4391         ctx->ipfw_layer3_chain = def_rule;
4392
4393         /* Link rule CPU sibling */
4394         ipfw_link_sibling(fwmsg, def_rule);
4395
4396         /* Statistics only need to be updated once */
4397         if (mycpuid == 0)
4398                 ipfw_inc_static_count(def_rule);
4399
4400         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
4401 }
4402
4403 static void
4404 ipfw_init_dispatch(netmsg_t nmsg)
4405 {
4406         struct netmsg_ipfw fwmsg;
4407         int error = 0;
4408
4409         if (IPFW_LOADED) {
4410                 kprintf("IP firewall already loaded\n");
4411                 error = EEXIST;
4412                 goto reply;
4413         }
4414
4415         bzero(&fwmsg, sizeof(fwmsg));
4416         netmsg_init(&fwmsg.base, NULL, &curthread->td_msgport,
4417                     0, ipfw_ctx_init_dispatch);
4418         ifnet_domsg(&fwmsg.base.lmsg, 0);
4419
4420         ip_fw_chk_ptr = ipfw_chk;
4421         ip_fw_ctl_ptr = ipfw_ctl;
4422         ip_fw_dn_io_ptr = ipfw_dummynet_io;
4423
4424         kprintf("ipfw2 initialized, default to %s, logging ",
4425                 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode ==
4426                 O_ACCEPT ? "accept" : "deny");
4427
4428 #ifdef IPFIREWALL_VERBOSE
4429         fw_verbose = 1;
4430 #endif
4431 #ifdef IPFIREWALL_VERBOSE_LIMIT
4432         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4433 #endif
4434         if (fw_verbose == 0) {
4435                 kprintf("disabled\n");
4436         } else if (verbose_limit == 0) {
4437                 kprintf("unlimited\n");
4438         } else {
4439                 kprintf("limited to %d packets/entry by default\n",
4440                         verbose_limit);
4441         }
4442
4443         callout_init_mp(&ipfw_timeout_h);
4444         netmsg_init(&ipfw_timeout_netmsg, NULL, &netisr_adone_rport,
4445                     MSGF_DROPABLE | MSGF_PRIORITY,
4446                     ipfw_tick_dispatch);
4447         lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
4448
4449         ip_fw_loaded = 1;
4450         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
4451
4452         if (fw_enable)
4453                 ipfw_hook();
4454 reply:
4455         lwkt_replymsg(&nmsg->lmsg, error);
4456 }
4457
4458 static int
4459 ipfw_init(void)
4460 {
4461         struct netmsg_base smsg;
4462
4463         netmsg_init(&smsg, NULL, &curthread->td_msgport,
4464                     0, ipfw_init_dispatch);
4465         return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4466 }
4467
4468 #ifdef KLD_MODULE
4469
4470 static void
4471 ipfw_fini_dispatch(netmsg_t nmsg)
4472 {
4473         int error = 0, cpu;
4474
4475         if (ipfw_refcnt != 0) {
4476                 error = EBUSY;
4477                 goto reply;
4478         }
4479
4480         ip_fw_loaded = 0;
4481
4482         ipfw_dehook();
4483         callout_stop(&ipfw_timeout_h);
4484
4485         netmsg_service_sync();
4486
4487         crit_enter();
4488         if ((ipfw_timeout_netmsg.lmsg.ms_flags & MSGF_DONE) == 0) {
4489                 /*
4490                  * Callout message is pending; drop it
4491                  */
4492                 lwkt_dropmsg(&ipfw_timeout_netmsg.lmsg);
4493         }
4494         crit_exit();
4495
4496         ip_fw_chk_ptr = NULL;
4497         ip_fw_ctl_ptr = NULL;
4498         ip_fw_dn_io_ptr = NULL;
4499         ipfw_flush(1 /* kill default rule */);
4500
4501         /* Free pre-cpu context */
4502         for (cpu = 0; cpu < ncpus; ++cpu)
4503                 kfree(ipfw_ctx[cpu], M_IPFW);
4504
4505         kprintf("IP firewall unloaded\n");
4506 reply:
4507         lwkt_replymsg(&nmsg->lmsg, error);
4508 }
4509
4510 static int
4511 ipfw_fini(void)
4512 {
4513         struct netmsg_base smsg;
4514
4515         netmsg_init(&smsg, NULL, &curthread->td_msgport,
4516                     0, ipfw_fini_dispatch);
4517         return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4518 }
4519
4520 #endif  /* KLD_MODULE */
4521
4522 static int
4523 ipfw_modevent(module_t mod, int type, void *unused)
4524 {
4525         int err = 0;
4526
4527         switch (type) {
4528         case MOD_LOAD:
4529                 err = ipfw_init();
4530                 break;
4531
4532         case MOD_UNLOAD:
4533 #ifndef KLD_MODULE
4534                 kprintf("ipfw statically compiled, cannot unload\n");
4535                 err = EBUSY;
4536 #else
4537                 err = ipfw_fini();
4538 #endif
4539                 break;
4540         default:
4541                 break;
4542         }
4543         return err;
4544 }
4545
4546 static moduledata_t ipfwmod = {
4547         "ipfw",
4548         ipfw_modevent,
4549         0
4550 };
4551 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
4552 MODULE_VERSION(ipfw, 1);