sys/net/ipfw/ip_fw2.c

   1 /*
   2  * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
   3  *
   4  * Redistribution and use in source and binary forms, with or without
   5  * modification, are permitted provided that the following conditions
   6  * are met:
   7  * 1. Redistributions of source code must retain the above copyright
   8  *    notice, this list of conditions and the following disclaimer.
   9  * 2. Redistributions in binary form must reproduce the above copyright
  10  *    notice, this list of conditions and the following disclaimer in the
  11  *    documentation and/or other materials provided with the distribution.
  12  *
  13  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  16  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  23  * SUCH DAMAGE.
  24  *
  25  * $FreeBSD: src/sys/netinet/ip_fw2.c,v 1.6.2.12 2003/04/08 10:42:32 maxim Exp $
  26  * $DragonFly: src/sys/net/ipfw/ip_fw2.c,v 1.100 2008/11/22 11:03:35 sephe Exp $
  27  */
  28
  29 /*
  30  * Implement IP packet firewall (new version)
  31  */
  32
  33 #include "opt_ipfw.h"
  34 #include "opt_inet.h"
  35 #ifndef INET
  36 #error IPFIREWALL requires INET.
  37 #endif /* INET */
  38
  39 #include <sys/param.h>
  40 #include <sys/systm.h>
  41 #include <sys/malloc.h>
  42 #include <sys/mbuf.h>
  43 #include <sys/kernel.h>
  44 #include <sys/proc.h>
  45 #include <sys/socket.h>
  46 #include <sys/socketvar.h>
  47 #include <sys/sysctl.h>
  48 #include <sys/syslog.h>
  49 #include <sys/ucred.h>
  50 #include <sys/in_cksum.h>
  51 #include <sys/lock.h>
  52
  53 #include <net/if.h>
  54 #include <net/route.h>
  55 #include <net/pfil.h>
  56 #include <net/dummynet/ip_dummynet.h>
  57
  58 #include <sys/thread2.h>
  59 #include <sys/mplock2.h>
  60 #include <net/netmsg2.h>
  61
  62 #include <netinet/in.h>
  63 #include <netinet/in_systm.h>
  64 #include <netinet/in_var.h>
  65 #include <netinet/in_pcb.h>
  66 #include <netinet/ip.h>
  67 #include <netinet/ip_var.h>
  68 #include <netinet/ip_icmp.h>
  69 #include <netinet/tcp.h>
  70 #include <netinet/tcp_timer.h>
  71 #include <netinet/tcp_var.h>
  72 #include <netinet/tcpip.h>
  73 #include <netinet/udp.h>
  74 #include <netinet/udp_var.h>
  75 #include <netinet/ip_divert.h>
  76 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  77
  78 #include <net/ipfw/ip_fw2.h>
  79
  80 #ifdef IPFIREWALL_DEBUG
  81 #define DPRINTF(fmt, ...) \
  82 do { \
  83         if (fw_debug > 0) \
  84                 kprintf(fmt, __VA_ARGS__); \
  85 } while (0)
  86 #else
  87 #define DPRINTF(fmt, ...)       ((void)0)
  88 #endif
  89
  90 /*
  91  * Description about per-CPU rule duplication:
  92  *
  93  * Module loading/unloading and all ioctl operations are serialized
  94  * by netisr0, so we don't have any ordering or locking problems.
  95  *
  96  * Following graph shows how operation on per-CPU rule list is
  97  * performed [2 CPU case]:
  98  *
  99  *   CPU0                 CPU1
 100  *
 101  * netisr0 <------------------------------------+
 102  *  domsg                                       |
 103  *    |                                         |
 104  *    | netmsg                                  |
 105  *    |                                         |
 106  *    V                                         |
 107  *  ifnet0                                      |
 108  *    :                                         | netmsg
 109  *    :(delete/add...)                          |
 110  *    :                                         |
 111  *    :         netmsg                          |
 112  *  forwardmsg---------->ifnet1                 |
 113  *                          :                   |
 114  *                          :(delete/add...)    |
 115  *                          :                   |
 116  *                          :                   |
 117  *                        replymsg--------------+
 118  *
 119  *
 120  *
 121  *
 122  * Rules which will not create states (dyn rules) [2 CPU case]
 123  *
 124  *    CPU0               CPU1
 125  * layer3_chain       layer3_chain
 126  *     |                  |
 127  *     V                  V
 128  * +-------+ sibling  +-------+ sibling
 129  * | rule1 |--------->| rule1 |--------->NULL
 130  * +-------+          +-------+
 131  *     |                  |
 132  *     |next              |next
 133  *     V                  V
 134  * +-------+ sibling  +-------+ sibling
 135  * | rule2 |--------->| rule2 |--------->NULL
 136  * +-------+          +-------+
 137  *
 138  * ip_fw.sibling:
 139  * 1) Ease statistics calculation during IP_FW_GET.  We only need to
 140  *    iterate layer3_chain on CPU0; the current rule's duplication on
 141  *    the other CPUs could safely be read-only accessed by using
 142  *    ip_fw.sibling
 143  * 2) Accelerate rule insertion and deletion, e.g. rule insertion:
 144  *    a) In netisr0 (on CPU0) rule3 is determined to be inserted between
 145  *       rule1 and rule2.  To make this decision we need to iterate the
 146  *       layer3_chain on CPU0.  The netmsg, which is used to insert the
 147  *       rule, will contain rule1 on CPU0 as prev_rule and rule2 on CPU0
 148  *       as next_rule
 149  *    b) After the insertion on CPU0 is done, we will move on to CPU1.
 150  *       But instead of relocating the rule3's position on CPU1 by
 151  *       iterating the layer3_chain on CPU1, we set the netmsg's prev_rule
 152  *       to rule1->sibling and next_rule to rule2->sibling before the
 153  *       netmsg is forwarded to CPU1 from CPU0
 154  *
 155  *
 156  *
 157  * Rules which will create states (dyn rules) [2 CPU case]
 158  * (unnecessary parts are omitted; they are same as in the previous figure)
 159  *
 160  *   CPU0                       CPU1
 161  *
 162  * +-------+                  +-------+
 163  * | rule1 |                  | rule1 |
 164  * +-------+                  +-------+
 165  *   ^   |                      |   ^
 166  *   |   |stub              stub|   |
 167  *   |   |                      |   |
 168  *   |   +----+            +----+   |
 169  *   |        |            |        |
 170  *   |        V            V        |
 171  *   |    +--------------------+    |
 172  *   |    |     rule_stub      |    |
 173  *   |    | (read-only shared) |    |
 174  *   |    |                    |    |
 175  *   |    | back pointer array |    |
 176  *   |    | (indexed by cpuid) |    |
 177  *   |    |                    |    |
 178  *   +----|---------[0]        |    |
 179  *        |         [1]--------|----+
 180  *        |                    |
 181  *        +--------------------+
 182  *          ^            ^
 183  *          |            |
 184  *  ........|............|............
 185  *  :       |            |           :
 186  *  :       |stub        |stub       :
 187  *  :       |            |           :
 188  *  :  +---------+  +---------+      :
 189  *  :  | state1a |  | state1b | .... :
 190  *  :  +---------+  +---------+      :
 191  *  :                                :
 192  *  :           states table         :
 193  *  :            (shared)            :
 194  *  :      (protected by dyn_lock)   :
 195  *  ..................................
 196  *
 197  * [state1a and state1b are states created by rule1]
 198  *
 199  * ip_fw_stub:
 200  * This structure is introduced so that shared (locked) state table could
 201  * work with per-CPU (duplicated) static rules.  It mainly bridges states
 202  * and static rules and serves as static rule's place holder (a read-only
 203  * shared part of duplicated rules) from states point of view.
 204  *
 205  * IPFW_RULE_F_STATE (only for rules which create states):
 206  * o  During rule installation, this flag is turned on after rule's
 207  *    duplications reach all CPUs, to avoid at least following race:
 208  *    1) rule1 is duplicated on CPU0 and is not duplicated on CPU1 yet
 209  *    2) rule1 creates state1
 210  *    3) state1 is located on CPU1 by check-state
 211  *    But rule1 is not duplicated on CPU1 yet
 212  * o  During rule deletion, this flag is turned off before deleting states
 213  *    created by the rule and before deleting the rule itself, so no
 214  *    more states will be created by the to-be-deleted rule even when its
 215  *    duplication on certain CPUs are not eliminated yet.
 216  */
 217
 218 #define IPFW_AUTOINC_STEP_MIN   1
 219 #define IPFW_AUTOINC_STEP_MAX   1000
 220 #define IPFW_AUTOINC_STEP_DEF   100
 221
 222 #define IPFW_DEFAULT_RULE       65535   /* rulenum for the default rule */
 223 #define IPFW_DEFAULT_SET        31      /* set number for the default rule */
 224
 225 struct netmsg_ipfw {
 226         struct netmsg_base base;
 227         const struct ipfw_ioc_rule *ioc_rule;
 228         struct ip_fw    *next_rule;
 229         struct ip_fw    *prev_rule;
 230         struct ip_fw    *sibling;
 231         struct ip_fw_stub *stub;
 232 };
 233
 234 struct netmsg_del {
 235         struct netmsg_base base;
 236         struct ip_fw    *start_rule;
 237         struct ip_fw    *prev_rule;
 238         uint16_t        rulenum;
 239         uint8_t         from_set;
 240         uint8_t         to_set;
 241 };
 242
 243 struct netmsg_zent {
 244         struct netmsg_base base;
 245         struct ip_fw    *start_rule;
 246         uint16_t        rulenum;
 247         uint16_t        log_only;
 248 };
 249
 250 struct ipfw_context {
 251         struct ip_fw    *ipfw_layer3_chain;     /* list of rules for layer3 */
 252         struct ip_fw    *ipfw_default_rule;     /* default rule */
 253         uint64_t        ipfw_norule_counter;    /* counter for ipfw_log(NULL) */
 254
 255         /*
 256          * ipfw_set_disable contains one bit per set value (0..31).
 257          * If the bit is set, all rules with the corresponding set
 258          * are disabled.  Set IPDW_DEFAULT_SET is reserved for the
 259          * default rule and CANNOT be disabled.
 260          */
 261         uint32_t        ipfw_set_disable;
 262         uint32_t        ipfw_gen;               /* generation of rule list */
 263 };
 264
 265 static struct ipfw_context      *ipfw_ctx[MAXCPU];
 266
 267 #ifdef KLD_MODULE
 268 /*
 269  * Module can not be unloaded, if there are references to
 270  * certains rules of ipfw(4), e.g. dummynet(4)
 271  */
 272 static int ipfw_refcnt;
 273 #endif
 274
 275 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 276
 277 /*
 278  * Following two global variables are accessed and
 279  * updated only on CPU0
 280  */
 281 static uint32_t static_count;   /* # of static rules */
 282 static uint32_t static_ioc_len; /* bytes of static rules */
 283
 284 /*
 285  * If 1, then ipfw static rules are being flushed,
 286  * ipfw_chk() will skip to the default rule.
 287  */
 288 static int ipfw_flushing;
 289
 290 static int fw_verbose;
 291 static int verbose_limit;
 292
 293 static int fw_debug;
 294 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
 295
 296 static int      ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS);
 297 static int      ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS);
 298 static int      ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS);
 299 static int      ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS);
 300 static int      ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS);
 301
 302 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 303 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
 304     &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw");
 305 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
 306     &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
 307     "Rule number autincrement step");
 308 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 309     &fw_one_pass, 0,
 310     "Only do a single pass through ipfw when using dummynet(4)");
 311 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 312     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 313 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 314     &fw_verbose, 0, "Log matches to ipfw rules");
 315 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 316     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 317
 318 /*
 319  * Description of dynamic rules.
 320  *
 321  * Dynamic rules are stored in lists accessed through a hash table
 322  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 323  * be modified through the sysctl variable dyn_buckets which is
 324  * updated when the table becomes empty.
 325  *
 326  * XXX currently there is only one list, ipfw_dyn.
 327  *
 328  * When a packet is received, its address fields are first masked
 329  * with the mask defined for the rule, then hashed, then matched
 330  * against the entries in the corresponding list.
 331  * Dynamic rules can be used for different purposes:
 332  *  + stateful rules;
 333  *  + enforcing limits on the number of sessions;
 334  *  + in-kernel NAT (not implemented yet)
 335  *
 336  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 337  * measured in seconds and depending on the flags.
 338  *
 339  * The total number of dynamic rules is stored in dyn_count.
 340  * The max number of dynamic rules is dyn_max. When we reach
 341  * the maximum number of rules we do not create anymore. This is
 342  * done to avoid consuming too much memory, but also too much
 343  * time when searching on each packet (ideally, we should try instead
 344  * to put a limit on the length of the list on each bucket...).
 345  *
 346  * Each dynamic rule holds a pointer to the parent ipfw rule so
 347  * we know what action to perform. Dynamic rules are removed when
 348  * the parent rule is deleted. XXX we should make them survive.
 349  *
 350  * There are some limitations with dynamic rules -- we do not
 351  * obey the 'randomized match', and we do not do multiple
 352  * passes through the firewall. XXX check the latter!!!
 353  *
 354  * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
 355  * Only TCP state transition will change dynamic rule's state and ack
 356  * sequences, while all packets of one TCP connection only goes through
 357  * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
 358  * rule looking up.  The keep alive callout uses exclusive lockmgr lock
 359  * when it tries to find suitable dynamic rules to send keep alive, so
 360  * it will not see half updated state and ack sequences.  Though the expire
 361  * field updating looks racy for other protocols, the resolution (second)
 362  * of expire field makes this kind of race harmless.
 363  * XXX statistics' updating is _not_ MPsafe!!!
 364  * XXX once UDP output path is fixed, we could use lockless dynamic rule
 365  *     hash table
 366  */
 367 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 368 static uint32_t dyn_buckets = 256; /* must be power of 2 */
 369 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
 370 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
 371 static struct lock dyn_lock; /* dynamic rules' hash table lock */
 372
 373 static struct netmsg_base ipfw_timeout_netmsg; /* schedule ipfw timeout */
 374 static struct callout ipfw_timeout_h;
 375
 376 /*
 377  * Timeouts for various events in handing dynamic rules.
 378  */
 379 static uint32_t dyn_ack_lifetime = 300;
 380 static uint32_t dyn_syn_lifetime = 20;
 381 static uint32_t dyn_fin_lifetime = 1;
 382 static uint32_t dyn_rst_lifetime = 1;
 383 static uint32_t dyn_udp_lifetime = 10;
 384 static uint32_t dyn_short_lifetime = 5;
 385
 386 /*
 387  * Keepalives are sent if dyn_keepalive is set. They are sent every
 388  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 389  * seconds of lifetime of a rule.
 390  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 391  * than dyn_keepalive_period.
 392  */
 393
 394 static uint32_t dyn_keepalive_interval = 20;
 395 static uint32_t dyn_keepalive_period = 5;
 396 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
 397
 398 static uint32_t dyn_count;              /* # of dynamic rules */
 399 static uint32_t dyn_max = 4096;         /* max # of dynamic rules */
 400
 401 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
 402     &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
 403 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 404     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 405 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 406     &dyn_count, 0, "Number of dyn. rules");
 407 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 408     &dyn_max, 0, "Max number of dyn. rules");
 409 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 410     &static_count, 0, "Number of static rules");
 411 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 412     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 413 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 414     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 415 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
 416     CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
 417     "Lifetime of dyn. rules for fin");
 418 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
 419     CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
 420     "Lifetime of dyn. rules for rst");
 421 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 422     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 423 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 424     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 425 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 426     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 427
 428 static ip_fw_chk_t      ipfw_chk;
 429 static void             ipfw_tick(void *);
 430
 431 static __inline int
 432 ipfw_free_rule(struct ip_fw *rule)
 433 {
 434         KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d\n", mycpuid));
 435         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
 436         rule->refcnt--;
 437         if (rule->refcnt == 0) {
 438                 kfree(rule, M_IPFW);
 439                 return 1;
 440         }
 441         return 0;
 442 }
 443
 444 static void
 445 ipfw_unref_rule(void *priv)
 446 {
 447         ipfw_free_rule(priv);
 448 #ifdef KLD_MODULE
 449         atomic_subtract_int(&ipfw_refcnt, 1);
 450 #endif
 451 }
 452
 453 static __inline void
 454 ipfw_ref_rule(struct ip_fw *rule)
 455 {
 456         KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d\n", mycpuid));
 457 #ifdef KLD_MODULE
 458         atomic_add_int(&ipfw_refcnt, 1);
 459 #endif
 460         rule->refcnt++;
 461 }
 462
 463 /*
 464  * This macro maps an ip pointer into a layer3 header pointer of type T
 465  */
 466 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
 467
 468 static __inline int
 469 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 470 {
 471         int type = L3HDR(struct icmp,ip)->icmp_type;
 472
 473         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
 474 }
 475
 476 #define TT      ((1 << ICMP_ECHO) | \
 477                  (1 << ICMP_ROUTERSOLICIT) | \
 478                  (1 << ICMP_TSTAMP) | \
 479                  (1 << ICMP_IREQ) | \
 480                  (1 << ICMP_MASKREQ))
 481
 482 static int
 483 is_icmp_query(struct ip *ip)
 484 {
 485         int type = L3HDR(struct icmp, ip)->icmp_type;
 486
 487         return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
 488 }
 489
 490 #undef TT
 491
 492 /*
 493  * The following checks use two arrays of 8 or 16 bits to store the
 494  * bits that we want set or clear, respectively. They are in the
 495  * low and high half of cmd->arg1 or cmd->d[0].
 496  *
 497  * We scan options and store the bits we find set. We succeed if
 498  *
 499  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 500  *
 501  * The code is sometimes optimized not to store additional variables.
 502  */
 503
 504 static int
 505 flags_match(ipfw_insn *cmd, uint8_t bits)
 506 {
 507         u_char want_clear;
 508         bits = ~bits;
 509
 510         if (((cmd->arg1 & 0xff) & bits) != 0)
 511                 return 0; /* some bits we want set were clear */
 512
 513         want_clear = (cmd->arg1 >> 8) & 0xff;
 514         if ((want_clear & bits) != want_clear)
 515                 return 0; /* some bits we want clear were set */
 516         return 1;
 517 }
 518
 519 static int
 520 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 521 {
 522         int optlen, bits = 0;
 523         u_char *cp = (u_char *)(ip + 1);
 524         int x = (ip->ip_hl << 2) - sizeof(struct ip);
 525
 526         for (; x > 0; x -= optlen, cp += optlen) {
 527                 int opt = cp[IPOPT_OPTVAL];
 528
 529                 if (opt == IPOPT_EOL)
 530                         break;
 531
 532                 if (opt == IPOPT_NOP) {
 533                         optlen = 1;
 534                 } else {
 535                         optlen = cp[IPOPT_OLEN];
 536                         if (optlen <= 0 || optlen > x)
 537                                 return 0; /* invalid or truncated */
 538                 }
 539
 540                 switch (opt) {
 541                 case IPOPT_LSRR:
 542                         bits |= IP_FW_IPOPT_LSRR;
 543                         break;
 544
 545                 case IPOPT_SSRR:
 546                         bits |= IP_FW_IPOPT_SSRR;
 547                         break;
 548
 549                 case IPOPT_RR:
 550                         bits |= IP_FW_IPOPT_RR;
 551                         break;
 552
 553                 case IPOPT_TS:
 554                         bits |= IP_FW_IPOPT_TS;
 555                         break;
 556
 557                 default:
 558                         break;
 559                 }
 560         }
 561         return (flags_match(cmd, bits));
 562 }
 563
 564 static int
 565 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 566 {
 567         int optlen, bits = 0;
 568         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 569         u_char *cp = (u_char *)(tcp + 1);
 570         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 571
 572         for (; x > 0; x -= optlen, cp += optlen) {
 573                 int opt = cp[0];
 574
 575                 if (opt == TCPOPT_EOL)
 576                         break;
 577
 578                 if (opt == TCPOPT_NOP) {
 579                         optlen = 1;
 580                 } else {
 581                         optlen = cp[1];
 582                         if (optlen <= 0)
 583                                 break;
 584                 }
 585
 586                 switch (opt) {
 587                 case TCPOPT_MAXSEG:
 588                         bits |= IP_FW_TCPOPT_MSS;
 589                         break;
 590
 591                 case TCPOPT_WINDOW:
 592                         bits |= IP_FW_TCPOPT_WINDOW;
 593                         break;
 594
 595                 case TCPOPT_SACK_PERMITTED:
 596                 case TCPOPT_SACK:
 597                         bits |= IP_FW_TCPOPT_SACK;
 598                         break;
 599
 600                 case TCPOPT_TIMESTAMP:
 601                         bits |= IP_FW_TCPOPT_TS;
 602                         break;
 603
 604                 case TCPOPT_CC:
 605                 case TCPOPT_CCNEW:
 606                 case TCPOPT_CCECHO:
 607                         bits |= IP_FW_TCPOPT_CC;
 608                         break;
 609
 610                 default:
 611                         break;
 612                 }
 613         }
 614         return (flags_match(cmd, bits));
 615 }
 616
 617 static int
 618 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 619 {
 620         if (ifp == NULL)        /* no iface with this packet, match fails */
 621                 return 0;
 622
 623         /* Check by name or by IP address */
 624         if (cmd->name[0] != '\0') { /* match by name */
 625                 /* Check name */
 626                 if (cmd->p.glob) {
 627                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 628                                 return(1);
 629                 } else {
 630                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 631                                 return(1);
 632                 }
 633         } else {
 634                 struct ifaddr_container *ifac;
 635
 636                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 637                         struct ifaddr *ia = ifac->ifa;
 638
 639                         if (ia->ifa_addr == NULL)
 640                                 continue;
 641                         if (ia->ifa_addr->sa_family != AF_INET)
 642                                 continue;
 643                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 644                             (ia->ifa_addr))->sin_addr.s_addr)
 645                                 return(1);      /* match */
 646                 }
 647         }
 648         return(0);      /* no match, fail ... */
 649 }
 650
 651 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 652
 653 /*
 654  * We enter here when we have a rule with O_LOG.
 655  * XXX this function alone takes about 2Kbytes of code!
 656  */
 657 static void
 658 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 659          struct mbuf *m, struct ifnet *oif)
 660 {
 661         char *action;
 662         int limit_reached = 0;
 663         char action2[40], proto[48], fragment[28];
 664
 665         fragment[0] = '\0';
 666         proto[0] = '\0';
 667
 668         if (f == NULL) {        /* bogus pkt */
 669                 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
 670
 671                 if (verbose_limit != 0 &&
 672                     ctx->ipfw_norule_counter >= verbose_limit)
 673                         return;
 674                 ctx->ipfw_norule_counter++;
 675                 if (ctx->ipfw_norule_counter == verbose_limit)
 676                         limit_reached = verbose_limit;
 677                 action = "Refuse";
 678         } else {        /* O_LOG is the first action, find the real one */
 679                 ipfw_insn *cmd = ACTION_PTR(f);
 680                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 681
 682                 if (l->max_log != 0 && l->log_left == 0)
 683                         return;
 684                 l->log_left--;
 685                 if (l->log_left == 0)
 686                         limit_reached = l->max_log;
 687                 cmd += F_LEN(cmd);      /* point to first action */
 688                 if (cmd->opcode == O_PROB)
 689                         cmd += F_LEN(cmd);
 690
 691                 action = action2;
 692                 switch (cmd->opcode) {
 693                 case O_DENY:
 694                         action = "Deny";
 695                         break;
 696
 697                 case O_REJECT:
 698                         if (cmd->arg1==ICMP_REJECT_RST) {
 699                                 action = "Reset";
 700                         } else if (cmd->arg1==ICMP_UNREACH_HOST) {
 701                                 action = "Reject";
 702                         } else {
 703                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 704                                           cmd->arg1);
 705                         }
 706                         break;
 707
 708                 case O_ACCEPT:
 709                         action = "Accept";
 710                         break;
 711
 712                 case O_COUNT:
 713                         action = "Count";
 714                         break;
 715
 716                 case O_DIVERT:
 717                         ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
 718                         break;
 719
 720                 case O_TEE:
 721                         ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
 722                         break;
 723
 724                 case O_SKIPTO:
 725                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
 726                         break;
 727
 728                 case O_PIPE:
 729                         ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
 730                         break;
 731
 732                 case O_QUEUE:
 733                         ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
 734                         break;
 735
 736                 case O_FORWARD_IP:
 737                         {
 738                                 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 739                                 int len;
 740
 741                                 len = ksnprintf(SNPARGS(action2, 0),
 742                                                 "Forward to %s",
 743                                                 inet_ntoa(sa->sa.sin_addr));
 744                                 if (sa->sa.sin_port) {
 745                                         ksnprintf(SNPARGS(action2, len), ":%d",
 746                                                   sa->sa.sin_port);
 747                                 }
 748                         }
 749                         break;
 750
 751                 default:
 752                         action = "UNKNOWN";
 753                         break;
 754                 }
 755         }
 756
 757         if (hlen == 0) {        /* non-ip */
 758                 ksnprintf(SNPARGS(proto, 0), "MAC");
 759         } else {
 760                 struct ip *ip = mtod(m, struct ip *);
 761                 /* these three are all aliases to the same thing */
 762                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 763                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 764                 struct udphdr *const udp = (struct udphdr *)icmp;
 765
 766                 int ip_off, offset, ip_len;
 767                 int len;
 768
 769                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 770                         ip_off = ntohs(ip->ip_off);
 771                         ip_len = ntohs(ip->ip_len);
 772                 } else {
 773                         ip_off = ip->ip_off;
 774                         ip_len = ip->ip_len;
 775                 }
 776                 offset = ip_off & IP_OFFMASK;
 777                 switch (ip->ip_p) {
 778                 case IPPROTO_TCP:
 779                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 780                                         inet_ntoa(ip->ip_src));
 781                         if (offset == 0) {
 782                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 783                                           ntohs(tcp->th_sport),
 784                                           inet_ntoa(ip->ip_dst),
 785                                           ntohs(tcp->th_dport));
 786                         } else {
 787                                 ksnprintf(SNPARGS(proto, len), " %s",
 788                                           inet_ntoa(ip->ip_dst));
 789                         }
 790                         break;
 791
 792                 case IPPROTO_UDP:
 793                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 794                                         inet_ntoa(ip->ip_src));
 795                         if (offset == 0) {
 796                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 797                                           ntohs(udp->uh_sport),
 798                                           inet_ntoa(ip->ip_dst),
 799                                           ntohs(udp->uh_dport));
 800                         } else {
 801                                 ksnprintf(SNPARGS(proto, len), " %s",
 802                                           inet_ntoa(ip->ip_dst));
 803                         }
 804                         break;
 805
 806                 case IPPROTO_ICMP:
 807                         if (offset == 0) {
 808                                 len = ksnprintf(SNPARGS(proto, 0),
 809                                                 "ICMP:%u.%u ",
 810                                                 icmp->icmp_type,
 811                                                 icmp->icmp_code);
 812                         } else {
 813                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 814                         }
 815                         len += ksnprintf(SNPARGS(proto, len), "%s",
 816                                          inet_ntoa(ip->ip_src));
 817                         ksnprintf(SNPARGS(proto, len), " %s",
 818                                   inet_ntoa(ip->ip_dst));
 819                         break;
 820
 821                 default:
 822                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 823                                         inet_ntoa(ip->ip_src));
 824                         ksnprintf(SNPARGS(proto, len), " %s",
 825                                   inet_ntoa(ip->ip_dst));
 826                         break;
 827                 }
 828
 829                 if (ip_off & (IP_MF | IP_OFFMASK)) {
 830                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 831                                   ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 832                                   offset << 3, (ip_off & IP_MF) ? "+" : "");
 833                 }
 834         }
 835
 836         if (oif || m->m_pkthdr.rcvif) {
 837                 log(LOG_SECURITY | LOG_INFO,
 838                     "ipfw: %d %s %s %s via %s%s\n",
 839                     f ? f->rulenum : -1,
 840                     action, proto, oif ? "out" : "in",
 841                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 842                     fragment);
 843         } else {
 844                 log(LOG_SECURITY | LOG_INFO,
 845                     "ipfw: %d %s %s [no if info]%s\n",
 846                     f ? f->rulenum : -1,
 847                     action, proto, fragment);
 848         }
 849
 850         if (limit_reached) {
 851                 log(LOG_SECURITY | LOG_NOTICE,
 852                     "ipfw: limit %d reached on entry %d\n",
 853                     limit_reached, f ? f->rulenum : -1);
 854         }
 855 }
 856
 857 #undef SNPARGS
 858
 859 /*
 860  * IMPORTANT: the hash function for dynamic rules must be commutative
 861  * in source and destination (ip,port), because rules are bidirectional
 862  * and we want to find both in the same bucket.
 863  */
 864 static __inline int
 865 hash_packet(struct ipfw_flow_id *id)
 866 {
 867         uint32_t i;
 868
 869         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 870         i &= (curr_dyn_buckets - 1);
 871         return i;
 872 }
 873
 874 /**
 875  * unlink a dynamic rule from a chain. prev is a pointer to
 876  * the previous one, q is a pointer to the rule to delete,
 877  * head is a pointer to the head of the queue.
 878  * Modifies q and potentially also head.
 879  */
 880 #define UNLINK_DYN_RULE(prev, head, q)                                  \
 881 do {                                                                    \
 882         ipfw_dyn_rule *old_q = q;                                       \
 883                                                                         \
 884         /* remove a refcount to the parent */                           \
 885         if (q->dyn_type == O_LIMIT)                                     \
 886                 q->parent->count--;                                     \
 887         DPRINTF("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",    \
 888                 q->id.src_ip, q->id.src_port,                           \
 889                 q->id.dst_ip, q->id.dst_port, dyn_count - 1);           \
 890         if (prev != NULL)                                               \
 891                 prev->next = q = q->next;                               \
 892         else                                                            \
 893                 head = q = q->next;                                     \
 894         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
 895         dyn_count--;                                                    \
 896         kfree(old_q, M_IPFW);                                           \
 897 } while (0)
 898
 899 #define TIME_LEQ(a, b)  ((int)((a) - (b)) <= 0)
 900
 901 /**
 902  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 903  *
 904  * If keep_me == NULL, rules are deleted even if not expired,
 905  * otherwise only expired rules are removed.
 906  *
 907  * The value of the second parameter is also used to point to identify
 908  * a rule we absolutely do not want to remove (e.g. because we are
 909  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 910  * rules). The pointer is only used for comparison, so any non-null
 911  * value will do.
 912  */
 913 static void
 914 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 915 {
 916         static uint32_t last_remove = 0; /* XXX */
 917
 918 #define FORCE   (keep_me == NULL)
 919
 920         ipfw_dyn_rule *prev, *q;
 921         int i, pass = 0, max_pass = 0, unlinked = 0;
 922
 923         if (ipfw_dyn_v == NULL || dyn_count == 0)
 924                 return;
 925         /* do not expire more than once per second, it is useless */
 926         if (!FORCE && last_remove == time_second)
 927                 return;
 928         last_remove = time_second;
 929
 930         /*
 931          * because O_LIMIT refer to parent rules, during the first pass only
 932          * remove child and mark any pending LIMIT_PARENT, and remove
 933          * them in a second pass.
 934          */
 935 next_pass:
 936         for (i = 0; i < curr_dyn_buckets; i++) {
 937                 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
 938                         /*
 939                          * Logic can become complex here, so we split tests.
 940                          */
 941                         if (q == keep_me)
 942                                 goto next;
 943                         if (rule != NULL && rule->stub != q->stub)
 944                                 goto next; /* not the one we are looking for */
 945                         if (q->dyn_type == O_LIMIT_PARENT) {
 946                                 /*
 947                                  * handle parent in the second pass,
 948                                  * record we need one.
 949                                  */
 950                                 max_pass = 1;
 951                                 if (pass == 0)
 952                                         goto next;
 953                                 if (FORCE && q->count != 0) {
 954                                         /* XXX should not happen! */
 955                                         kprintf("OUCH! cannot remove rule, "
 956                                                 "count %d\n", q->count);
 957                                 }
 958                         } else {
 959                                 if (!FORCE && !TIME_LEQ(q->expire, time_second))
 960                                         goto next;
 961                         }
 962                         unlinked = 1;
 963                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 964                         continue;
 965 next:
 966                         prev = q;
 967                         q = q->next;
 968                 }
 969         }
 970         if (pass++ < max_pass)
 971                 goto next_pass;
 972
 973         if (unlinked)
 974                 ++dyn_buckets_gen;
 975
 976 #undef FORCE
 977 }
 978
 979 /**
 980  * lookup a dynamic rule.
 981  */
 982 static ipfw_dyn_rule *
 983 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 984                 struct tcphdr *tcp)
 985 {
 986         /*
 987          * stateful ipfw extensions.
 988          * Lookup into dynamic session queue
 989          */
 990 #define MATCH_REVERSE   0
 991 #define MATCH_FORWARD   1
 992 #define MATCH_NONE      2
 993 #define MATCH_UNKNOWN   3
 994         int i, dir = MATCH_NONE;
 995         ipfw_dyn_rule *prev, *q=NULL;
 996
 997         if (ipfw_dyn_v == NULL)
 998                 goto done;      /* not found */
 999
1000         i = hash_packet(pkt);
1001         for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
1002                 if (q->dyn_type == O_LIMIT_PARENT)
1003                         goto next;
1004
1005                 if (TIME_LEQ(q->expire, time_second)) {
1006                         /*
1007                          * Entry expired; skip.
1008                          * Let ipfw_tick() take care of it
1009                          */
1010                         goto next;
1011                 }
1012
1013                 if (pkt->proto == q->id.proto) {
1014                         if (pkt->src_ip == q->id.src_ip &&
1015                             pkt->dst_ip == q->id.dst_ip &&
1016                             pkt->src_port == q->id.src_port &&
1017                             pkt->dst_port == q->id.dst_port) {
1018                                 dir = MATCH_FORWARD;
1019                                 break;
1020                         }
1021                         if (pkt->src_ip == q->id.dst_ip &&
1022                             pkt->dst_ip == q->id.src_ip &&
1023                             pkt->src_port == q->id.dst_port &&
1024                             pkt->dst_port == q->id.src_port) {
1025                                 dir = MATCH_REVERSE;
1026                                 break;
1027                         }
1028                 }
1029 next:
1030                 prev = q;
1031                 q = q->next;
1032         }
1033         if (q == NULL)
1034                 goto done; /* q = NULL, not found */
1035
1036         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1037                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1038
1039 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
1040 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
1041
1042                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1043                 switch (q->state) {
1044                 case TH_SYN:                            /* opening */
1045                         q->expire = time_second + dyn_syn_lifetime;
1046                         break;
1047
1048                 case BOTH_SYN:                  /* move to established */
1049                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
1050                 case BOTH_SYN | (TH_FIN << 8) :
1051                         if (tcp) {
1052                                 uint32_t ack = ntohl(tcp->th_ack);
1053
1054 #define _SEQ_GE(a, b)   ((int)(a) - (int)(b) >= 0)
1055
1056                                 if (dir == MATCH_FORWARD) {
1057                                         if (q->ack_fwd == 0 ||
1058                                             _SEQ_GE(ack, q->ack_fwd))
1059                                                 q->ack_fwd = ack;
1060                                         else /* ignore out-of-sequence */
1061                                                 break;
1062                                 } else {
1063                                         if (q->ack_rev == 0 ||
1064                                             _SEQ_GE(ack, q->ack_rev))
1065                                                 q->ack_rev = ack;
1066                                         else /* ignore out-of-sequence */
1067                                                 break;
1068                                 }
1069 #undef _SEQ_GE
1070                         }
1071                         q->expire = time_second + dyn_ack_lifetime;
1072                         break;
1073
1074                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
1075                         KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
1076                         q->expire = time_second + dyn_fin_lifetime;
1077                         break;
1078
1079                 default:
1080 #if 0
1081                         /*
1082                          * reset or some invalid combination, but can also
1083                          * occur if we use keep-state the wrong way.
1084                          */
1085                         if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
1086                                 kprintf("invalid state: 0x%x\n", q->state);
1087 #endif
1088                         KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
1089                         q->expire = time_second + dyn_rst_lifetime;
1090                         break;
1091                 }
1092         } else if (pkt->proto == IPPROTO_UDP) {
1093                 q->expire = time_second + dyn_udp_lifetime;
1094         } else {
1095                 /* other protocols */
1096                 q->expire = time_second + dyn_short_lifetime;
1097         }
1098 done:
1099         if (match_direction)
1100                 *match_direction = dir;
1101         return q;
1102 }
1103
1104 static struct ip_fw *
1105 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
1106             uint16_t len, int *deny)
1107 {
1108         struct ip_fw *rule = NULL;
1109         ipfw_dyn_rule *q;
1110         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1111         uint32_t gen;
1112
1113         *deny = 0;
1114         gen = ctx->ipfw_gen;
1115
1116         lockmgr(&dyn_lock, LK_SHARED);
1117
1118         if (ctx->ipfw_gen != gen) {
1119                 /*
1120                  * Static rules had been change when we were waiting
1121                  * for the dynamic hash table lock; deny this packet,
1122                  * since it is _not_ known whether it is safe to keep
1123                  * iterating the static rules.
1124                  */
1125                 *deny = 1;
1126                 goto back;
1127         }
1128
1129         q = lookup_dyn_rule(pkt, match_direction, tcp);
1130         if (q == NULL) {
1131                 rule = NULL;
1132         } else {
1133                 rule = q->stub->rule[mycpuid];
1134                 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid);
1135
1136                 /* XXX */
1137                 q->pcnt++;
1138                 q->bcnt += len;
1139         }
1140 back:
1141         lockmgr(&dyn_lock, LK_RELEASE);
1142         return rule;
1143 }
1144
1145 static void
1146 realloc_dynamic_table(void)
1147 {
1148         ipfw_dyn_rule **old_dyn_v;
1149         uint32_t old_curr_dyn_buckets;
1150
1151         KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
1152                 ("invalid dyn_buckets %d\n", dyn_buckets));
1153
1154         /* Save the current buckets array for later error recovery */
1155         old_dyn_v = ipfw_dyn_v;
1156         old_curr_dyn_buckets = curr_dyn_buckets;
1157
1158         curr_dyn_buckets = dyn_buckets;
1159         for (;;) {
1160                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1161                                      M_IPFW, M_NOWAIT | M_ZERO);
1162                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1163                         break;
1164
1165                 curr_dyn_buckets /= 2;
1166                 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
1167                     old_dyn_v != NULL) {
1168                         /*
1169                          * Don't try allocating smaller buckets array, reuse
1170                          * the old one, which alreay contains enough buckets
1171                          */
1172                         break;
1173                 }
1174         }
1175
1176         if (ipfw_dyn_v != NULL) {
1177                 if (old_dyn_v != NULL)
1178                         kfree(old_dyn_v, M_IPFW);
1179         } else {
1180                 /* Allocation failed, restore old buckets array */
1181                 ipfw_dyn_v = old_dyn_v;
1182                 curr_dyn_buckets = old_curr_dyn_buckets;
1183         }
1184
1185         if (ipfw_dyn_v != NULL)
1186                 ++dyn_buckets_gen;
1187 }
1188
1189 /**
1190  * Install state of type 'type' for a dynamic session.
1191  * The hash table contains two type of rules:
1192  * - regular rules (O_KEEP_STATE)
1193  * - rules for sessions with limited number of sess per user
1194  *   (O_LIMIT). When they are created, the parent is
1195  *   increased by 1, and decreased on delete. In this case,
1196  *   the third parameter is the parent rule and not the chain.
1197  * - "parent" rules for the above (O_LIMIT_PARENT).
1198  */
1199 static ipfw_dyn_rule *
1200 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1201 {
1202         ipfw_dyn_rule *r;
1203         int i;
1204
1205         if (ipfw_dyn_v == NULL ||
1206             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1207                 realloc_dynamic_table();
1208                 if (ipfw_dyn_v == NULL)
1209                         return NULL; /* failed ! */
1210         }
1211         i = hash_packet(id);
1212
1213         r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1214         if (r == NULL) {
1215                 kprintf ("sorry cannot allocate state\n");
1216                 return NULL;
1217         }
1218
1219         /* increase refcount on parent, and set pointer */
1220         if (dyn_type == O_LIMIT) {
1221                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1222
1223                 if (parent->dyn_type != O_LIMIT_PARENT)
1224                         panic("invalid parent");
1225                 parent->count++;
1226                 r->parent = parent;
1227                 rule = parent->stub->rule[mycpuid];
1228                 KKASSERT(rule->stub == parent->stub);
1229         }
1230         KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1231
1232         r->id = *id;
1233         r->expire = time_second + dyn_syn_lifetime;
1234         r->stub = rule->stub;
1235         r->dyn_type = dyn_type;
1236         r->pcnt = r->bcnt = 0;
1237         r->count = 0;
1238
1239         r->bucket = i;
1240         r->next = ipfw_dyn_v[i];
1241         ipfw_dyn_v[i] = r;
1242         dyn_count++;
1243         dyn_buckets_gen++;
1244         DPRINTF("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1245                 dyn_type,
1246                 r->id.src_ip, r->id.src_port,
1247                 r->id.dst_ip, r->id.dst_port, dyn_count);
1248         return r;
1249 }
1250
1251 /**
1252  * lookup dynamic parent rule using pkt and rule as search keys.
1253  * If the lookup fails, then install one.
1254  */
1255 static ipfw_dyn_rule *
1256 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1257 {
1258         ipfw_dyn_rule *q;
1259         int i;
1260
1261         if (ipfw_dyn_v) {
1262                 i = hash_packet(pkt);
1263                 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1264                         if (q->dyn_type == O_LIMIT_PARENT &&
1265                             rule->stub == q->stub &&
1266                             pkt->proto == q->id.proto &&
1267                             pkt->src_ip == q->id.src_ip &&
1268                             pkt->dst_ip == q->id.dst_ip &&
1269                             pkt->src_port == q->id.src_port &&
1270                             pkt->dst_port == q->id.dst_port) {
1271                                 q->expire = time_second + dyn_short_lifetime;
1272                                 DPRINTF("lookup_dyn_parent found 0x%p\n", q);
1273                                 return q;
1274                         }
1275                 }
1276         }
1277         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1278 }
1279
1280 /**
1281  * Install dynamic state for rule type cmd->o.opcode
1282  *
1283  * Returns 1 (failure) if state is not installed because of errors or because
1284  * session limitations are enforced.
1285  */
1286 static int
1287 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1288                      struct ip_fw_args *args)
1289 {
1290         static int last_log; /* XXX */
1291
1292         ipfw_dyn_rule *q;
1293
1294         DPRINTF("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1295                 cmd->o.opcode,
1296                 args->f_id.src_ip, args->f_id.src_port,
1297                 args->f_id.dst_ip, args->f_id.dst_port);
1298
1299         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1300         if (q != NULL) { /* should never occur */
1301                 if (last_log != time_second) {
1302                         last_log = time_second;
1303                         kprintf(" install_state: entry already present, done\n");
1304                 }
1305                 return 0;
1306         }
1307
1308         if (dyn_count >= dyn_max) {
1309                 /*
1310                  * Run out of slots, try to remove any expired rule.
1311                  */
1312                 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1313                 if (dyn_count >= dyn_max) {
1314                         if (last_log != time_second) {
1315                                 last_log = time_second;
1316                                 kprintf("install_state: "
1317                                         "Too many dynamic rules\n");
1318                         }
1319                         return 1; /* cannot install, notify caller */
1320                 }
1321         }
1322
1323         switch (cmd->o.opcode) {
1324         case O_KEEP_STATE: /* bidir rule */
1325                 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1326                         return 1;
1327                 break;
1328
1329         case O_LIMIT: /* limit number of sessions */
1330                 {
1331                         uint16_t limit_mask = cmd->limit_mask;
1332                         struct ipfw_flow_id id;
1333                         ipfw_dyn_rule *parent;
1334
1335                         DPRINTF("installing dyn-limit rule %d\n",
1336                                 cmd->conn_limit);
1337
1338                         id.dst_ip = id.src_ip = 0;
1339                         id.dst_port = id.src_port = 0;
1340                         id.proto = args->f_id.proto;
1341
1342                         if (limit_mask & DYN_SRC_ADDR)
1343                                 id.src_ip = args->f_id.src_ip;
1344                         if (limit_mask & DYN_DST_ADDR)
1345                                 id.dst_ip = args->f_id.dst_ip;
1346                         if (limit_mask & DYN_SRC_PORT)
1347                                 id.src_port = args->f_id.src_port;
1348                         if (limit_mask & DYN_DST_PORT)
1349                                 id.dst_port = args->f_id.dst_port;
1350
1351                         parent = lookup_dyn_parent(&id, rule);
1352                         if (parent == NULL) {
1353                                 kprintf("add parent failed\n");
1354                                 return 1;
1355                         }
1356
1357                         if (parent->count >= cmd->conn_limit) {
1358                                 /*
1359                                  * See if we can remove some expired rule.
1360                                  */
1361                                 remove_dyn_rule_locked(rule, parent);
1362                                 if (parent->count >= cmd->conn_limit) {
1363                                         if (fw_verbose &&
1364                                             last_log != time_second) {
1365                                                 last_log = time_second;
1366                                                 log(LOG_SECURITY | LOG_DEBUG,
1367                                                     "drop session, "
1368                                                     "too many entries\n");
1369                                         }
1370                                         return 1;
1371                                 }
1372                         }
1373                         if (add_dyn_rule(&args->f_id, O_LIMIT,
1374                                          (struct ip_fw *)parent) == NULL)
1375                                 return 1;
1376                 }
1377                 break;
1378         default:
1379                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1380                 return 1;
1381         }
1382         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1383         return 0;
1384 }
1385
1386 static int
1387 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1388               struct ip_fw_args *args, int *deny)
1389 {
1390         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1391         uint32_t gen;
1392         int ret = 0;
1393
1394         *deny = 0;
1395         gen = ctx->ipfw_gen;
1396
1397         lockmgr(&dyn_lock, LK_EXCLUSIVE);
1398         if (ctx->ipfw_gen != gen) {
1399                 /* See the comment in lookup_rule() */
1400                 *deny = 1;
1401         } else {
1402                 ret = install_state_locked(rule, cmd, args);
1403         }
1404         lockmgr(&dyn_lock, LK_RELEASE);
1405
1406         return ret;
1407 }
1408
1409 /*
1410  * Transmit a TCP packet, containing either a RST or a keepalive.
1411  * When flags & TH_RST, we are sending a RST packet, because of a
1412  * "reset" action matched the packet.
1413  * Otherwise we are sending a keepalive, and flags & TH_
1414  */
1415 static void
1416 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1417 {
1418         struct mbuf *m;
1419         struct ip *ip;
1420         struct tcphdr *tcp;
1421         struct route sro;       /* fake route */
1422
1423         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1424         if (m == NULL)
1425                 return;
1426         m->m_pkthdr.rcvif = NULL;
1427         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1428         m->m_data += max_linkhdr;
1429
1430         ip = mtod(m, struct ip *);
1431         bzero(ip, m->m_len);
1432         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1433         ip->ip_p = IPPROTO_TCP;
1434         tcp->th_off = 5;
1435
1436         /*
1437          * Assume we are sending a RST (or a keepalive in the reverse
1438          * direction), swap src and destination addresses and ports.
1439          */
1440         ip->ip_src.s_addr = htonl(id->dst_ip);
1441         ip->ip_dst.s_addr = htonl(id->src_ip);
1442         tcp->th_sport = htons(id->dst_port);
1443         tcp->th_dport = htons(id->src_port);
1444         if (flags & TH_RST) {   /* we are sending a RST */
1445                 if (flags & TH_ACK) {
1446                         tcp->th_seq = htonl(ack);
1447                         tcp->th_ack = htonl(0);
1448                         tcp->th_flags = TH_RST;
1449                 } else {
1450                         if (flags & TH_SYN)
1451                                 seq++;
1452                         tcp->th_seq = htonl(0);
1453                         tcp->th_ack = htonl(seq);
1454                         tcp->th_flags = TH_RST | TH_ACK;
1455                 }
1456         } else {
1457                 /*
1458                  * We are sending a keepalive. flags & TH_SYN determines
1459                  * the direction, forward if set, reverse if clear.
1460                  * NOTE: seq and ack are always assumed to be correct
1461                  * as set by the caller. This may be confusing...
1462                  */
1463                 if (flags & TH_SYN) {
1464                         /*
1465                          * we have to rewrite the correct addresses!
1466                          */
1467                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1468                         ip->ip_src.s_addr = htonl(id->src_ip);
1469                         tcp->th_dport = htons(id->dst_port);
1470                         tcp->th_sport = htons(id->src_port);
1471                 }
1472                 tcp->th_seq = htonl(seq);
1473                 tcp->th_ack = htonl(ack);
1474                 tcp->th_flags = TH_ACK;
1475         }
1476
1477         /*
1478          * set ip_len to the payload size so we can compute
1479          * the tcp checksum on the pseudoheader
1480          * XXX check this, could save a couple of words ?
1481          */
1482         ip->ip_len = htons(sizeof(struct tcphdr));
1483         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1484
1485         /*
1486          * now fill fields left out earlier
1487          */
1488         ip->ip_ttl = ip_defttl;
1489         ip->ip_len = m->m_pkthdr.len;
1490
1491         bzero(&sro, sizeof(sro));
1492         ip_rtaddr(ip->ip_dst, &sro);
1493
1494         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1495         ip_output(m, NULL, &sro, 0, NULL, NULL);
1496         if (sro.ro_rt)
1497                 RTFREE(sro.ro_rt);
1498 }
1499
1500 /*
1501  * sends a reject message, consuming the mbuf passed as an argument.
1502  */
1503 static void
1504 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1505 {
1506         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1507                 /* We need the IP header in host order for icmp_error(). */
1508                 if (args->eh != NULL) {
1509                         struct ip *ip = mtod(args->m, struct ip *);
1510
1511                         ip->ip_len = ntohs(ip->ip_len);
1512                         ip->ip_off = ntohs(ip->ip_off);
1513                 }
1514                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1515         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1516                 struct tcphdr *const tcp =
1517                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1518
1519                 if ((tcp->th_flags & TH_RST) == 0) {
1520                         send_pkt(&args->f_id, ntohl(tcp->th_seq),
1521                                  ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1522                 }
1523                 m_freem(args->m);
1524         } else {
1525                 m_freem(args->m);
1526         }
1527         args->m = NULL;
1528 }
1529
1530 /**
1531  *
1532  * Given an ip_fw *, lookup_next_rule will return a pointer
1533  * to the next rule, which can be either the jump
1534  * target (for skipto instructions) or the next one in the list (in
1535  * all other cases including a missing jump target).
1536  * The result is also written in the "next_rule" field of the rule.
1537  * Backward jumps are not allowed, so start looking from the next
1538  * rule...
1539  *
1540  * This never returns NULL -- in case we do not have an exact match,
1541  * the next rule is returned. When the ruleset is changed,
1542  * pointers are flushed so we are always correct.
1543  */
1544
1545 static struct ip_fw *
1546 lookup_next_rule(struct ip_fw *me)
1547 {
1548         struct ip_fw *rule = NULL;
1549         ipfw_insn *cmd;
1550
1551         /* look for action, in case it is a skipto */
1552         cmd = ACTION_PTR(me);
1553         if (cmd->opcode == O_LOG)
1554                 cmd += F_LEN(cmd);
1555         if (cmd->opcode == O_SKIPTO) {
1556                 for (rule = me->next; rule; rule = rule->next) {
1557                         if (rule->rulenum >= cmd->arg1)
1558                                 break;
1559                 }
1560         }
1561         if (rule == NULL)                       /* failure or not a skipto */
1562                 rule = me->next;
1563         me->next_rule = rule;
1564         return rule;
1565 }
1566
1567 static int
1568 _ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1569                 enum ipfw_opcodes opcode, uid_t uid)
1570 {
1571         struct in_addr src_ip, dst_ip;
1572         struct inpcbinfo *pi;
1573         int wildcard;
1574         struct inpcb *pcb;
1575
1576         if (fid->proto == IPPROTO_TCP) {
1577                 wildcard = 0;
1578                 pi = &tcbinfo[mycpuid];
1579         } else if (fid->proto == IPPROTO_UDP) {
1580                 wildcard = 1;
1581                 pi = &udbinfo;
1582         } else {
1583                 return 0;
1584         }
1585
1586         /*
1587          * Values in 'fid' are in host byte order
1588          */
1589         dst_ip.s_addr = htonl(fid->dst_ip);
1590         src_ip.s_addr = htonl(fid->src_ip);
1591         if (oif) {
1592                 pcb = in_pcblookup_hash(pi,
1593                         dst_ip, htons(fid->dst_port),
1594                         src_ip, htons(fid->src_port),
1595                         wildcard, oif);
1596         } else {
1597                 pcb = in_pcblookup_hash(pi,
1598                         src_ip, htons(fid->src_port),
1599                         dst_ip, htons(fid->dst_port),
1600                         wildcard, NULL);
1601         }
1602         if (pcb == NULL || pcb->inp_socket == NULL)
1603                 return 0;
1604
1605         if (opcode == O_UID) {
1606 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1607                 return !socheckuid(pcb->inp_socket, uid);
1608 #undef socheckuid
1609         } else  {
1610                 return groupmember(uid, pcb->inp_socket->so_cred);
1611         }
1612 }
1613
1614 static int
1615 ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1616                enum ipfw_opcodes opcode, uid_t uid, int *deny)
1617 {
1618         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1619         uint32_t gen;
1620         int match = 0;
1621
1622         *deny = 0;
1623         gen = ctx->ipfw_gen;
1624
1625         get_mplock();
1626         if (gen != ctx->ipfw_gen) {
1627                 /* See the comment in lookup_rule() */
1628                 *deny = 1;
1629         } else {
1630                 match = _ipfw_match_uid(fid, oif, opcode, uid);
1631         }
1632         rel_mplock();
1633         return match;
1634 }
1635
1636 /*
1637  * The main check routine for the firewall.
1638  *
1639  * All arguments are in args so we can modify them and return them
1640  * back to the caller.
1641  *
1642  * Parameters:
1643  *
1644  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1645  *              Starts with the IP header.
1646  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1647  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1648  *              The incoming interface is in the mbuf. (in)
1649  *
1650  *      args->rule      Pointer to the last matching rule (in/out)
1651  *      args->f_id      Addresses grabbed from the packet (out)
1652  *
1653  * Return value:
1654  *
1655  *      If the packet was denied/rejected and has been dropped, *m is equal
1656  *      to NULL upon return.
1657  *
1658  *      IP_FW_DENY      the packet must be dropped.
1659  *      IP_FW_PASS      The packet is to be accepted and routed normally.
1660  *      IP_FW_DIVERT    Divert the packet to port (args->cookie)
1661  *      IP_FW_TEE       Tee the packet to port (args->cookie)
1662  *      IP_FW_DUMMYNET  Send the packet to pipe/queue (args->cookie)
1663  */
1664
1665 static int
1666 ipfw_chk(struct ip_fw_args *args)
1667 {
1668         /*
1669          * Local variables hold state during the processing of a packet.
1670          *
1671          * IMPORTANT NOTE: to speed up the processing of rules, there
1672          * are some assumption on the values of the variables, which
1673          * are documented here. Should you change them, please check
1674          * the implementation of the various instructions to make sure
1675          * that they still work.
1676          *
1677          * args->eh     The MAC header. It is non-null for a layer2
1678          *      packet, it is NULL for a layer-3 packet.
1679          *
1680          * m | args->m  Pointer to the mbuf, as received from the caller.
1681          *      It may change if ipfw_chk() does an m_pullup, or if it
1682          *      consumes the packet because it calls send_reject().
1683          *      XXX This has to change, so that ipfw_chk() never modifies
1684          *      or consumes the buffer.
1685          * ip   is simply an alias of the value of m, and it is kept
1686          *      in sync with it (the packet is  supposed to start with
1687          *      the ip header).
1688          */
1689         struct mbuf *m = args->m;
1690         struct ip *ip = mtod(m, struct ip *);
1691
1692         /*
1693          * oif | args->oif      If NULL, ipfw_chk has been called on the
1694          *      inbound path (ether_input, ip_input).
1695          *      If non-NULL, ipfw_chk has been called on the outbound path
1696          *      (ether_output, ip_output).
1697          */
1698         struct ifnet *oif = args->oif;
1699
1700         struct ip_fw *f = NULL;         /* matching rule */
1701         int retval = IP_FW_PASS;
1702         struct m_tag *mtag;
1703         struct divert_info *divinfo;
1704
1705         /*
1706          * hlen The length of the IPv4 header.
1707          *      hlen >0 means we have an IPv4 packet.
1708          */
1709         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1710
1711         /*
1712          * offset       The offset of a fragment. offset != 0 means that
1713          *      we have a fragment at this offset of an IPv4 packet.
1714          *      offset == 0 means that (if this is an IPv4 packet)
1715          *      this is the first or only fragment.
1716          */
1717         u_short offset = 0;
1718
1719         /*
1720          * Local copies of addresses. They are only valid if we have
1721          * an IP packet.
1722          *
1723          * proto        The protocol. Set to 0 for non-ip packets,
1724          *      or to the protocol read from the packet otherwise.
1725          *      proto != 0 means that we have an IPv4 packet.
1726          *
1727          * src_port, dst_port   port numbers, in HOST format. Only
1728          *      valid for TCP and UDP packets.
1729          *
1730          * src_ip, dst_ip       ip addresses, in NETWORK format.
1731          *      Only valid for IPv4 packets.
1732          */
1733         uint8_t proto;
1734         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1735         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1736         uint16_t ip_len = 0;
1737
1738         /*
1739          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1740          *      MATCH_NONE when checked and not matched (dyn_f = NULL),
1741          *      MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1742          */
1743         int dyn_dir = MATCH_UNKNOWN;
1744         struct ip_fw *dyn_f = NULL;
1745         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1746
1747         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1748                 return IP_FW_PASS;      /* accept */
1749
1750         if (args->eh == NULL ||         /* layer 3 packet */
1751             (m->m_pkthdr.len >= sizeof(struct ip) &&
1752              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1753                 hlen = ip->ip_hl << 2;
1754
1755         /*
1756          * Collect parameters into local variables for faster matching.
1757          */
1758         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1759                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1760                 goto after_ip_checks;
1761         }
1762
1763         proto = args->f_id.proto = ip->ip_p;
1764         src_ip = ip->ip_src;
1765         dst_ip = ip->ip_dst;
1766         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1767                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1768                 ip_len = ntohs(ip->ip_len);
1769         } else {
1770                 offset = ip->ip_off & IP_OFFMASK;
1771                 ip_len = ip->ip_len;
1772         }
1773
1774 #define PULLUP_TO(len)                          \
1775 do {                                            \
1776         if (m->m_len < (len)) {                 \
1777                 args->m = m = m_pullup(m, (len));\
1778                 if (m == NULL)                  \
1779                         goto pullup_failed;     \
1780                 ip = mtod(m, struct ip *);      \
1781         }                                       \
1782 } while (0)
1783
1784         if (offset == 0) {
1785                 switch (proto) {
1786                 case IPPROTO_TCP:
1787                         {
1788                                 struct tcphdr *tcp;
1789
1790                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1791                                 tcp = L3HDR(struct tcphdr, ip);
1792                                 dst_port = tcp->th_dport;
1793                                 src_port = tcp->th_sport;
1794                                 args->f_id.flags = tcp->th_flags;
1795                         }
1796                         break;
1797
1798                 case IPPROTO_UDP:
1799                         {
1800                                 struct udphdr *udp;
1801
1802                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1803                                 udp = L3HDR(struct udphdr, ip);
1804                                 dst_port = udp->uh_dport;
1805                                 src_port = udp->uh_sport;
1806                         }
1807                         break;
1808
1809                 case IPPROTO_ICMP:
1810                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1811                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1812                         break;
1813
1814                 default:
1815                         break;
1816                 }
1817         }
1818
1819 #undef PULLUP_TO
1820
1821         args->f_id.src_ip = ntohl(src_ip.s_addr);
1822         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1823         args->f_id.src_port = src_port = ntohs(src_port);
1824         args->f_id.dst_port = dst_port = ntohs(dst_port);
1825
1826 after_ip_checks:
1827         if (args->rule) {
1828                 /*
1829                  * Packet has already been tagged. Look for the next rule
1830                  * to restart processing.
1831                  *
1832                  * If fw_one_pass != 0 then just accept it.
1833                  * XXX should not happen here, but optimized out in
1834                  * the caller.
1835                  */
1836                 if (fw_one_pass)
1837                         return IP_FW_PASS;
1838
1839                 /* This rule is being/has been flushed */
1840                 if (ipfw_flushing)
1841                         return IP_FW_DENY;
1842
1843                 KASSERT(args->rule->cpuid == mycpuid,
1844                         ("rule used on cpu%d\n", mycpuid));
1845
1846                 /* This rule was deleted */
1847                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1848                         return IP_FW_DENY;
1849
1850                 f = args->rule->next_rule;
1851                 if (f == NULL)
1852                         f = lookup_next_rule(args->rule);
1853         } else {
1854                 /*
1855                  * Find the starting rule. It can be either the first
1856                  * one, or the one after divert_rule if asked so.
1857                  */
1858                 int skipto;
1859
1860                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1861                 if (mtag != NULL) {
1862                         divinfo = m_tag_data(mtag);
1863                         skipto = divinfo->skipto;
1864                 } else {
1865                         skipto = 0;
1866                 }
1867
1868                 f = ctx->ipfw_layer3_chain;
1869                 if (args->eh == NULL && skipto != 0) {
1870                         /* No skipto during rule flushing */
1871                         if (ipfw_flushing)
1872                                 return IP_FW_DENY;
1873
1874                         if (skipto >= IPFW_DEFAULT_RULE)
1875                                 return IP_FW_DENY; /* invalid */
1876
1877                         while (f && f->rulenum <= skipto)
1878                                 f = f->next;
1879                         if (f == NULL)  /* drop packet */
1880                                 return IP_FW_DENY;
1881                 } else if (ipfw_flushing) {
1882                         /* Rules are being flushed; skip to default rule */
1883                         f = ctx->ipfw_default_rule;
1884                 }
1885         }
1886         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1887                 m_tag_delete(m, mtag);
1888
1889         /*
1890          * Now scan the rules, and parse microinstructions for each rule.
1891          */
1892         for (; f; f = f->next) {
1893                 int l, cmdlen;
1894                 ipfw_insn *cmd;
1895                 int skip_or; /* skip rest of OR block */
1896
1897 again:
1898                 if (ctx->ipfw_set_disable & (1 << f->set))
1899                         continue;
1900
1901                 skip_or = 0;
1902                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1903                      l -= cmdlen, cmd += cmdlen) {
1904                         int match, deny;
1905
1906                         /*
1907                          * check_body is a jump target used when we find a
1908                          * CHECK_STATE, and need to jump to the body of
1909                          * the target rule.
1910                          */
1911
1912 check_body:
1913                         cmdlen = F_LEN(cmd);
1914                         /*
1915                          * An OR block (insn_1 || .. || insn_n) has the
1916                          * F_OR bit set in all but the last instruction.
1917                          * The first match will set "skip_or", and cause
1918                          * the following instructions to be skipped until
1919                          * past the one with the F_OR bit clear.
1920                          */
1921                         if (skip_or) {          /* skip this instruction */
1922                                 if ((cmd->len & F_OR) == 0)
1923                                         skip_or = 0;    /* next one is good */
1924                                 continue;
1925                         }
1926                         match = 0; /* set to 1 if we succeed */
1927
1928                         switch (cmd->opcode) {
1929                         /*
1930                          * The first set of opcodes compares the packet's
1931                          * fields with some pattern, setting 'match' if a
1932                          * match is found. At the end of the loop there is
1933                          * logic to deal with F_NOT and F_OR flags associated
1934                          * with the opcode.
1935                          */
1936                         case O_NOP:
1937                                 match = 1;
1938                                 break;
1939
1940                         case O_FORWARD_MAC:
1941                                 kprintf("ipfw: opcode %d unimplemented\n",
1942                                         cmd->opcode);
1943                                 break;
1944
1945                         case O_GID:
1946                         case O_UID:
1947                                 /*
1948                                  * We only check offset == 0 && proto != 0,
1949                                  * as this ensures that we have an IPv4
1950                                  * packet with the ports info.
1951                                  */
1952                                 if (offset!=0)
1953                                         break;
1954
1955                                 match = ipfw_match_uid(&args->f_id, oif,
1956                                         cmd->opcode,
1957                                         (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1958                                         &deny);
1959                                 if (deny)
1960                                         return IP_FW_DENY;
1961                                 break;
1962
1963                         case O_RECV:
1964                                 match = iface_match(m->m_pkthdr.rcvif,
1965                                     (ipfw_insn_if *)cmd);
1966                                 break;
1967
1968                         case O_XMIT:
1969                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1970                                 break;
1971
1972                         case O_VIA:
1973                                 match = iface_match(oif ? oif :
1974                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1975                                 break;
1976
1977                         case O_MACADDR2:
1978                                 if (args->eh != NULL) { /* have MAC header */
1979                                         uint32_t *want = (uint32_t *)
1980                                                 ((ipfw_insn_mac *)cmd)->addr;
1981                                         uint32_t *mask = (uint32_t *)
1982                                                 ((ipfw_insn_mac *)cmd)->mask;
1983                                         uint32_t *hdr = (uint32_t *)args->eh;
1984
1985                                         match =
1986                                         (want[0] == (hdr[0] & mask[0]) &&
1987                                          want[1] == (hdr[1] & mask[1]) &&
1988                                          want[2] == (hdr[2] & mask[2]));
1989                                 }
1990                                 break;
1991
1992                         case O_MAC_TYPE:
1993                                 if (args->eh != NULL) {
1994                                         uint16_t t =
1995                                             ntohs(args->eh->ether_type);
1996                                         uint16_t *p =
1997                                             ((ipfw_insn_u16 *)cmd)->ports;
1998                                         int i;
1999
2000                                         /* Special vlan handling */
2001                                         if (m->m_flags & M_VLANTAG)
2002                                                 t = ETHERTYPE_VLAN;
2003
2004                                         for (i = cmdlen - 1; !match && i > 0;
2005                                              i--, p += 2) {
2006                                                 match =
2007                                                 (t >= p[0] && t <= p[1]);
2008                                         }
2009                                 }
2010                                 break;
2011
2012                         case O_FRAG:
2013                                 match = (hlen > 0 && offset != 0);
2014                                 break;
2015
2016                         case O_IN:      /* "out" is "not in" */
2017                                 match = (oif == NULL);
2018                                 break;
2019
2020                         case O_LAYER2:
2021                                 match = (args->eh != NULL);
2022                                 break;
2023
2024                         case O_PROTO:
2025                                 /*
2026                                  * We do not allow an arg of 0 so the
2027                                  * check of "proto" only suffices.
2028                                  */
2029                                 match = (proto == cmd->arg1);
2030                                 break;
2031
2032                         case O_IP_SRC:
2033                                 match = (hlen > 0 &&
2034                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2035                                     src_ip.s_addr);
2036                                 break;
2037
2038                         case O_IP_SRC_MASK:
2039                                 match = (hlen > 0 &&
2040                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2041                                      (src_ip.s_addr &
2042                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2043                                 break;
2044
2045                         case O_IP_SRC_ME:
2046                                 if (hlen > 0) {
2047                                         struct ifnet *tif;
2048
2049                                         tif = INADDR_TO_IFP(&src_ip);
2050                                         match = (tif != NULL);
2051                                 }
2052                                 break;
2053
2054                         case O_IP_DST_SET:
2055                         case O_IP_SRC_SET:
2056                                 if (hlen > 0) {
2057                                         uint32_t *d = (uint32_t *)(cmd + 1);
2058                                         uint32_t addr =
2059                                             cmd->opcode == O_IP_DST_SET ?
2060                                                 args->f_id.dst_ip :
2061                                                 args->f_id.src_ip;
2062
2063                                         if (addr < d[0])
2064                                                 break;
2065                                         addr -= d[0]; /* subtract base */
2066                                         match =
2067                                         (addr < cmd->arg1) &&
2068                                          (d[1 + (addr >> 5)] &
2069                                           (1 << (addr & 0x1f)));
2070                                 }
2071                                 break;
2072
2073                         case O_IP_DST:
2074                                 match = (hlen > 0 &&
2075                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2076                                     dst_ip.s_addr);
2077                                 break;
2078
2079                         case O_IP_DST_MASK:
2080                                 match = (hlen > 0) &&
2081                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2082                                      (dst_ip.s_addr &
2083                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2084                                 break;
2085
2086                         case O_IP_DST_ME:
2087                                 if (hlen > 0) {
2088                                         struct ifnet *tif;
2089
2090                                         tif = INADDR_TO_IFP(&dst_ip);
2091                                         match = (tif != NULL);
2092                                 }
2093                                 break;
2094
2095                         case O_IP_SRCPORT:
2096                         case O_IP_DSTPORT:
2097                                 /*
2098                                  * offset == 0 && proto != 0 is enough
2099                                  * to guarantee that we have an IPv4
2100                                  * packet with port info.
2101                                  */
2102                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2103                                     && offset == 0) {
2104                                         uint16_t x =
2105                                             (cmd->opcode == O_IP_SRCPORT) ?
2106                                                 src_port : dst_port ;
2107                                         uint16_t *p =
2108                                             ((ipfw_insn_u16 *)cmd)->ports;
2109                                         int i;
2110
2111                                         for (i = cmdlen - 1; !match && i > 0;
2112                                              i--, p += 2) {
2113                                                 match =
2114                                                 (x >= p[0] && x <= p[1]);
2115                                         }
2116                                 }
2117                                 break;
2118
2119                         case O_ICMPTYPE:
2120                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
2121                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2122                                 break;
2123
2124                         case O_IPOPT:
2125                                 match = (hlen > 0 && ipopts_match(ip, cmd));
2126                                 break;
2127
2128                         case O_IPVER:
2129                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2130                                 break;
2131
2132                         case O_IPTTL:
2133                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2134                                 break;
2135
2136                         case O_IPID:
2137                                 match = (hlen > 0 &&
2138                                     cmd->arg1 == ntohs(ip->ip_id));
2139                                 break;
2140
2141                         case O_IPLEN:
2142                                 match = (hlen > 0 && cmd->arg1 == ip_len);
2143                                 break;
2144
2145                         case O_IPPRECEDENCE:
2146                                 match = (hlen > 0 &&
2147                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
2148                                 break;
2149
2150                         case O_IPTOS:
2151                                 match = (hlen > 0 &&
2152                                     flags_match(cmd, ip->ip_tos));
2153                                 break;
2154
2155                         case O_TCPFLAGS:
2156                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2157                                     flags_match(cmd,
2158                                         L3HDR(struct tcphdr,ip)->th_flags));
2159                                 break;
2160
2161                         case O_TCPOPTS:
2162                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2163                                     tcpopts_match(ip, cmd));
2164                                 break;
2165
2166                         case O_TCPSEQ:
2167                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2168                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2169                                         L3HDR(struct tcphdr,ip)->th_seq);
2170                                 break;
2171
2172                         case O_TCPACK:
2173                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2174                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2175                                         L3HDR(struct tcphdr,ip)->th_ack);
2176                                 break;
2177
2178                         case O_TCPWIN:
2179                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2180                                     cmd->arg1 ==
2181                                         L3HDR(struct tcphdr,ip)->th_win);
2182                                 break;
2183
2184                         case O_ESTAB:
2185                                 /* reject packets which have SYN only */
2186                                 /* XXX should i also check for TH_ACK ? */
2187                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2188                                     (L3HDR(struct tcphdr,ip)->th_flags &
2189                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2190                                 break;
2191
2192                         case O_LOG:
2193                                 if (fw_verbose)
2194                                         ipfw_log(f, hlen, args->eh, m, oif);
2195                                 match = 1;
2196                                 break;
2197
2198                         case O_PROB:
2199                                 match = (krandom() <
2200                                         ((ipfw_insn_u32 *)cmd)->d[0]);
2201                                 break;
2202
2203                         /*
2204                          * The second set of opcodes represents 'actions',
2205                          * i.e. the terminal part of a rule once the packet
2206                          * matches all previous patterns.
2207                          * Typically there is only one action for each rule,
2208                          * and the opcode is stored at the end of the rule
2209                          * (but there are exceptions -- see below).
2210                          *
2211                          * In general, here we set retval and terminate the
2212                          * outer loop (would be a 'break 3' in some language,
2213                          * but we need to do a 'goto done').
2214                          *
2215                          * Exceptions:
2216                          * O_COUNT and O_SKIPTO actions:
2217                          *   instead of terminating, we jump to the next rule
2218                          *   ('goto next_rule', equivalent to a 'break 2'),
2219                          *   or to the SKIPTO target ('goto again' after
2220                          *   having set f, cmd and l), respectively.
2221                          *
2222                          * O_LIMIT and O_KEEP_STATE: these opcodes are
2223                          *   not real 'actions', and are stored right
2224                          *   before the 'action' part of the rule.
2225                          *   These opcodes try to install an entry in the
2226                          *   state tables; if successful, we continue with
2227                          *   the next opcode (match=1; break;), otherwise
2228                          *   the packet must be dropped ('goto done' after
2229                          *   setting retval).  If static rules are changed
2230                          *   during the state installation, the packet will
2231                          *   be dropped and rule's stats will not beupdated
2232                          *   ('return IP_FW_DENY').
2233                          *
2234                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2235                          *   cause a lookup of the state table, and a jump
2236                          *   to the 'action' part of the parent rule
2237                          *   ('goto check_body') if an entry is found, or
2238                          *   (CHECK_STATE only) a jump to the next rule if
2239                          *   the entry is not found ('goto next_rule').
2240                          *   The result of the lookup is cached to make
2241                          *   further instances of these opcodes are
2242                          *   effectively NOPs.  If static rules are changed
2243                          *   during the state looking up, the packet will
2244                          *   be dropped and rule's stats will not be updated
2245                          *   ('return IP_FW_DENY').
2246                          */
2247                         case O_LIMIT:
2248                         case O_KEEP_STATE:
2249                                 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2250                                         kprintf("%s rule (%d) is not ready "
2251                                                 "on cpu%d\n",
2252                                                 cmd->opcode == O_LIMIT ?
2253                                                 "limit" : "keep state",
2254                                                 f->rulenum, f->cpuid);
2255                                         goto next_rule;
2256                                 }
2257                                 if (install_state(f,
2258                                     (ipfw_insn_limit *)cmd, args, &deny)) {
2259                                         if (deny)
2260                                                 return IP_FW_DENY;
2261
2262                                         retval = IP_FW_DENY;
2263                                         goto done; /* error/limit violation */
2264                                 }
2265                                 if (deny)
2266                                         return IP_FW_DENY;
2267                                 match = 1;
2268                                 break;
2269
2270                         case O_PROBE_STATE:
2271                         case O_CHECK_STATE:
2272                                 /*
2273                                  * dynamic rules are checked at the first
2274                                  * keep-state or check-state occurrence,
2275                                  * with the result being stored in dyn_dir.
2276                                  * The compiler introduces a PROBE_STATE
2277                                  * instruction for us when we have a
2278                                  * KEEP_STATE (because PROBE_STATE needs
2279                                  * to be run first).
2280                                  */
2281                                 if (dyn_dir == MATCH_UNKNOWN) {
2282                                         dyn_f = lookup_rule(&args->f_id,
2283                                                 &dyn_dir,
2284                                                 proto == IPPROTO_TCP ?
2285                                                 L3HDR(struct tcphdr, ip) : NULL,
2286                                                 ip_len, &deny);
2287                                         if (deny)
2288                                                 return IP_FW_DENY;
2289                                         if (dyn_f != NULL) {
2290                                                 /*
2291                                                  * Found a rule from a dynamic
2292                                                  * entry; jump to the 'action'
2293                                                  * part of the rule.
2294                                                  */
2295                                                 f = dyn_f;
2296                                                 cmd = ACTION_PTR(f);
2297                                                 l = f->cmd_len - f->act_ofs;
2298                                                 goto check_body;
2299                                         }
2300                                 }
2301                                 /*
2302                                  * Dynamic entry not found. If CHECK_STATE,
2303                                  * skip to next rule, if PROBE_STATE just
2304                                  * ignore and continue with next opcode.
2305                                  */
2306                                 if (cmd->opcode == O_CHECK_STATE)
2307                                         goto next_rule;
2308                                 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2309                                         goto next_rule; /* not ready yet */
2310                                 match = 1;
2311                                 break;
2312
2313                         case O_ACCEPT:
2314                                 retval = IP_FW_PASS;    /* accept */
2315                                 goto done;
2316
2317                         case O_PIPE:
2318                         case O_QUEUE:
2319                                 args->rule = f; /* report matching rule */
2320                                 args->cookie = cmd->arg1;
2321                                 retval = IP_FW_DUMMYNET;
2322                                 goto done;
2323
2324                         case O_DIVERT:
2325                         case O_TEE:
2326                                 if (args->eh) /* not on layer 2 */
2327                                         break;
2328
2329                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2330                                                  sizeof(*divinfo), MB_DONTWAIT);
2331                                 if (mtag == NULL) {
2332                                         retval = IP_FW_DENY;
2333                                         goto done;
2334                                 }
2335                                 divinfo = m_tag_data(mtag);
2336
2337                                 divinfo->skipto = f->rulenum;
2338                                 divinfo->port = cmd->arg1;
2339                                 divinfo->tee = (cmd->opcode == O_TEE);
2340                                 m_tag_prepend(m, mtag);
2341
2342                                 args->cookie = cmd->arg1;
2343                                 retval = (cmd->opcode == O_DIVERT) ?
2344                                          IP_FW_DIVERT : IP_FW_TEE;
2345                                 goto done;
2346
2347                         case O_COUNT:
2348                         case O_SKIPTO:
2349                                 f->pcnt++;      /* update stats */
2350                                 f->bcnt += ip_len;
2351                                 f->timestamp = time_second;
2352                                 if (cmd->opcode == O_COUNT)
2353                                         goto next_rule;
2354                                 /* handle skipto */
2355                                 if (f->next_rule == NULL)
2356                                         lookup_next_rule(f);
2357                                 f = f->next_rule;
2358                                 goto again;
2359
2360                         case O_REJECT:
2361                                 /*
2362                                  * Drop the packet and send a reject notice
2363                                  * if the packet is not ICMP (or is an ICMP
2364                                  * query), and it is not multicast/broadcast.
2365                                  */
2366                                 if (hlen > 0 &&
2367                                     (proto != IPPROTO_ICMP ||
2368                                      is_icmp_query(ip)) &&
2369                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
2370                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2371                                         /*
2372                                          * Update statistics before the possible
2373                                          * blocking 'send_reject'
2374                                          */
2375                                         f->pcnt++;
2376                                         f->bcnt += ip_len;
2377                                         f->timestamp = time_second;
2378
2379                                         send_reject(args, cmd->arg1,
2380                                             offset,ip_len);
2381                                         m = args->m;
2382
2383                                         /*
2384                                          * Return directly here, rule stats
2385                                          * have been updated above.
2386                                          */
2387                                         return IP_FW_DENY;
2388                                 }
2389                                 /* FALLTHROUGH */
2390                         case O_DENY:
2391                                 retval = IP_FW_DENY;
2392                                 goto done;
2393
2394                         case O_FORWARD_IP:
2395                                 if (args->eh)   /* not valid on layer2 pkts */
2396                                         break;
2397                                 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2398                                         struct sockaddr_in *sin;
2399
2400                                         mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2401                                                sizeof(*sin), MB_DONTWAIT);
2402                                         if (mtag == NULL) {
2403                                                 retval = IP_FW_DENY;
2404                                                 goto done;
2405                                         }
2406                                         sin = m_tag_data(mtag);
2407
2408                                         /* Structure copy */
2409                                         *sin = ((ipfw_insn_sa *)cmd)->sa;
2410
2411                                         m_tag_prepend(m, mtag);
2412                                         m->m_pkthdr.fw_flags |=
2413                                                 IPFORWARD_MBUF_TAGGED;
2414                                 }
2415                                 retval = IP_FW_PASS;
2416                                 goto done;
2417
2418                         default:
2419                                 panic("-- unknown opcode %d\n", cmd->opcode);
2420                         } /* end of switch() on opcodes */
2421
2422                         if (cmd->len & F_NOT)
2423                                 match = !match;
2424
2425                         if (match) {
2426                                 if (cmd->len & F_OR)
2427                                         skip_or = 1;
2428                         } else {
2429                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2430                                         break;          /* try next rule    */
2431                         }
2432
2433                 }       /* end of inner for, scan opcodes */
2434
2435 next_rule:;             /* try next rule                */
2436
2437         }               /* end of outer for, scan rules */
2438         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2439         return IP_FW_DENY;
2440
2441 done:
2442         /* Update statistics */
2443         f->pcnt++;
2444         f->bcnt += ip_len;
2445         f->timestamp = time_second;
2446         return retval;
2447
2448 pullup_failed:
2449         if (fw_verbose)
2450                 kprintf("pullup failed\n");
2451         return IP_FW_DENY;
2452 }
2453
2454 static void
2455 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2456 {
2457         struct m_tag *mtag;
2458         struct dn_pkt *pkt;
2459         ipfw_insn *cmd;
2460         const struct ipfw_flow_id *id;
2461         struct dn_flow_id *fid;
2462
2463         M_ASSERTPKTHDR(m);
2464
2465         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2466         if (mtag == NULL) {
2467                 m_freem(m);
2468                 return;
2469         }
2470         m_tag_prepend(m, mtag);
2471
2472         pkt = m_tag_data(mtag);
2473         bzero(pkt, sizeof(*pkt));
2474
2475         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2476         if (cmd->opcode == O_LOG)
2477                 cmd += F_LEN(cmd);
2478         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2479                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2480
2481         pkt->dn_m = m;
2482         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2483         pkt->ifp = fwa->oif;
2484         pkt->pipe_nr = pipe_nr;
2485
2486         pkt->cpuid = mycpuid;
2487         pkt->msgport = cur_netport();
2488
2489         id = &fwa->f_id;
2490         fid = &pkt->id;
2491         fid->fid_dst_ip = id->dst_ip;
2492         fid->fid_src_ip = id->src_ip;
2493         fid->fid_dst_port = id->dst_port;
2494         fid->fid_src_port = id->src_port;
2495         fid->fid_proto = id->proto;
2496         fid->fid_flags = id->flags;
2497
2498         ipfw_ref_rule(fwa->rule);
2499         pkt->dn_priv = fwa->rule;
2500         pkt->dn_unref_priv = ipfw_unref_rule;
2501
2502         if (cmd->opcode == O_PIPE)
2503                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2504
2505         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2506 }
2507
2508 /*
2509  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2510  * These will be reconstructed on the fly as packets are matched.
2511  * Must be called at splimp().
2512  */
2513 static void
2514 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2515 {
2516         struct ip_fw *rule;
2517
2518         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2519                 rule->next_rule = NULL;
2520 }
2521
2522 static __inline void
2523 ipfw_inc_static_count(struct ip_fw *rule)
2524 {
2525         /* Static rule's counts are updated only on CPU0 */
2526         KKASSERT(mycpuid == 0);
2527
2528         static_count++;
2529         static_ioc_len += IOC_RULESIZE(rule);
2530 }
2531
2532 static __inline void
2533 ipfw_dec_static_count(struct ip_fw *rule)
2534 {
2535         int l = IOC_RULESIZE(rule);
2536
2537         /* Static rule's counts are updated only on CPU0 */
2538         KKASSERT(mycpuid == 0);
2539
2540         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2541         static_count--;
2542
2543         KASSERT(static_ioc_len >= l,
2544                 ("invalid static len %u\n", static_ioc_len));
2545         static_ioc_len -= l;
2546 }
2547
2548 static void
2549 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2550 {
2551         if (fwmsg->sibling != NULL) {
2552                 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2553                 fwmsg->sibling->sibling = rule;
2554         }
2555         fwmsg->sibling = rule;
2556 }
2557
2558 static struct ip_fw *
2559 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2560 {
2561         struct ip_fw *rule;
2562
2563         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2564
2565         rule->act_ofs = ioc_rule->act_ofs;
2566         rule->cmd_len = ioc_rule->cmd_len;
2567         rule->rulenum = ioc_rule->rulenum;
2568         rule->set = ioc_rule->set;
2569         rule->usr_flags = ioc_rule->usr_flags;
2570
2571         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2572
2573         rule->refcnt = 1;
2574         rule->cpuid = mycpuid;
2575
2576         rule->stub = stub;
2577         if (stub != NULL)
2578                 stub->rule[mycpuid] = rule;
2579
2580         return rule;
2581 }
2582
2583 static void
2584 ipfw_add_rule_dispatch(netmsg_t nmsg)
2585 {
2586         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2587         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2588         struct ip_fw *rule;
2589
2590         rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2591
2592         /*
2593          * Bump generation after ipfw_create_rule(),
2594          * since this function is blocking
2595          */
2596         ctx->ipfw_gen++;
2597
2598         /*
2599          * Insert rule into the pre-determined position
2600          */
2601         if (fwmsg->prev_rule != NULL) {
2602                 struct ip_fw *prev, *next;
2603
2604                 prev = fwmsg->prev_rule;
2605                 KKASSERT(prev->cpuid == mycpuid);
2606
2607                 next = fwmsg->next_rule;
2608                 KKASSERT(next->cpuid == mycpuid);
2609
2610                 rule->next = next;
2611                 prev->next = rule;
2612
2613                 /*
2614                  * Move to the position on the next CPU
2615                  * before the msg is forwarded.
2616                  */
2617                 fwmsg->prev_rule = prev->sibling;
2618                 fwmsg->next_rule = next->sibling;
2619         } else {
2620                 KKASSERT(fwmsg->next_rule == NULL);
2621                 rule->next = ctx->ipfw_layer3_chain;
2622                 ctx->ipfw_layer3_chain = rule;
2623         }
2624
2625         /* Link rule CPU sibling */
2626         ipfw_link_sibling(fwmsg, rule);
2627
2628         ipfw_flush_rule_ptrs(ctx);
2629
2630         if (mycpuid == 0) {
2631                 /* Statistics only need to be updated once */
2632                 ipfw_inc_static_count(rule);
2633
2634                 /* Return the rule on CPU0 */
2635                 nmsg->lmsg.u.ms_resultp = rule;
2636         }
2637
2638         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2639 }
2640
2641 static void
2642 ipfw_enable_state_dispatch(netmsg_t nmsg)
2643 {
2644         struct lwkt_msg *lmsg = &nmsg->lmsg;
2645         struct ip_fw *rule = lmsg->u.ms_resultp;
2646         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2647
2648         ctx->ipfw_gen++;
2649
2650         KKASSERT(rule->cpuid == mycpuid);
2651         KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule);
2652         KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE));
2653         rule->rule_flags |= IPFW_RULE_F_STATE;
2654         lmsg->u.ms_resultp = rule->sibling;
2655
2656         ifnet_forwardmsg(lmsg, mycpuid + 1);
2657 }
2658
2659 /*
2660  * Add a new rule to the list.  Copy the rule into a malloc'ed area,
2661  * then possibly create a rule number and add the rule to the list.
2662  * Update the rule_number in the input struct so the caller knows
2663  * it as well.
2664  */
2665 static void
2666 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
2667 {
2668         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2669         struct netmsg_ipfw fwmsg;
2670         struct netmsg_base *nmsg;
2671         struct ip_fw *f, *prev, *rule;
2672         struct ip_fw_stub *stub;
2673
2674         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2675
2676         /*
2677          * If rulenum is 0, find highest numbered rule before the
2678          * default rule, and add rule number incremental step.
2679          */
2680         if (ioc_rule->rulenum == 0) {
2681                 int step = autoinc_step;
2682
2683                 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2684                          step <= IPFW_AUTOINC_STEP_MAX);
2685
2686                 /*
2687                  * Locate the highest numbered rule before default
2688                  */
2689                 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2690                         if (f->rulenum == IPFW_DEFAULT_RULE)
2691                                 break;
2692                         ioc_rule->rulenum = f->rulenum;
2693                 }
2694                 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2695                         ioc_rule->rulenum += step;
2696         }
2697         KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2698                 ioc_rule->rulenum != 0,
2699                 ("invalid rule num %d\n", ioc_rule->rulenum));
2700
2701         /*
2702          * Now find the right place for the new rule in the sorted list.
2703          */
2704         for (prev = NULL, f = ctx->ipfw_layer3_chain; f;
2705              prev = f, f = f->next) {
2706                 if (f->rulenum > ioc_rule->rulenum) {
2707                         /* Found the location */
2708                         break;
2709                 }
2710         }
2711         KASSERT(f != NULL, ("no default rule?!\n"));
2712
2713         if (rule_flags & IPFW_RULE_F_STATE) {
2714                 int size;
2715
2716                 /*
2717                  * If the new rule will create states, then allocate
2718                  * a rule stub, which will be referenced by states
2719                  * (dyn rules)
2720                  */
2721                 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2722                 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2723         } else {
2724                 stub = NULL;
2725         }
2726
2727         /*
2728          * Duplicate the rule onto each CPU.
2729          * The rule duplicated on CPU0 will be returned.
2730          */
2731         bzero(&fwmsg, sizeof(fwmsg));
2732         nmsg = &fwmsg.base;
2733         netmsg_init(nmsg, NULL, &curthread->td_msgport,
2734                     0, ipfw_add_rule_dispatch);
2735         fwmsg.ioc_rule = ioc_rule;
2736         fwmsg.prev_rule = prev;
2737         fwmsg.next_rule = prev == NULL ? NULL : f;
2738         fwmsg.stub = stub;
2739
2740         ifnet_domsg(&nmsg->lmsg, 0);
2741         KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2742
2743         rule = nmsg->lmsg.u.ms_resultp;
2744         KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2745
2746         if (rule_flags & IPFW_RULE_F_STATE) {
2747                 /*
2748                  * Turn on state flag, _after_ everything on all
2749                  * CPUs have been setup.
2750                  */
2751                 bzero(nmsg, sizeof(*nmsg));
2752                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
2753                             0, ipfw_enable_state_dispatch);
2754                 nmsg->lmsg.u.ms_resultp = rule;
2755
2756                 ifnet_domsg(&nmsg->lmsg, 0);
2757                 KKASSERT(nmsg->lmsg.u.ms_resultp == NULL);
2758         }
2759
2760         DPRINTF("++ installed rule %d, static count now %d\n",
2761                 rule->rulenum, static_count);
2762 }
2763
2764 /**
2765  * Free storage associated with a static rule (including derived
2766  * dynamic rules).
2767  * The caller is in charge of clearing rule pointers to avoid
2768  * dangling pointers.
2769  * @return a pointer to the next entry.
2770  * Arguments are not checked, so they better be correct.
2771  * Must be called at splimp().
2772  */
2773 static struct ip_fw *
2774 ipfw_delete_rule(struct ipfw_context *ctx,
2775                  struct ip_fw *prev, struct ip_fw *rule)
2776 {
2777         struct ip_fw *n;
2778         struct ip_fw_stub *stub;
2779
2780         ctx->ipfw_gen++;
2781
2782         /* STATE flag should have been cleared before we reach here */
2783         KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2784
2785         stub = rule->stub;
2786         n = rule->next;
2787         if (prev == NULL)
2788                 ctx->ipfw_layer3_chain = n;
2789         else
2790                 prev->next = n;
2791
2792         /* Mark the rule as invalid */
2793         rule->rule_flags |= IPFW_RULE_F_INVALID;
2794         rule->next_rule = NULL;
2795         rule->sibling = NULL;
2796         rule->stub = NULL;
2797 #ifdef foo
2798         /* Don't reset cpuid here; keep various assertion working */
2799         rule->cpuid = -1;
2800 #endif
2801
2802         /* Statistics only need to be updated once */
2803         if (mycpuid == 0)
2804                 ipfw_dec_static_count(rule);
2805
2806         /* Free 'stub' on the last CPU */
2807         if (stub != NULL && mycpuid == ncpus - 1)
2808                 kfree(stub, M_IPFW);
2809
2810         /* Try to free this rule */
2811         ipfw_free_rule(rule);
2812
2813         /* Return the next rule */
2814         return n;
2815 }
2816
2817 static void
2818 ipfw_flush_dispatch(netmsg_t nmsg)
2819 {
2820         struct lwkt_msg *lmsg = &nmsg->lmsg;
2821         int kill_default = lmsg->u.ms_result;
2822         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2823         struct ip_fw *rule;
2824
2825         ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2826
2827         while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2828                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2829                 ipfw_delete_rule(ctx, NULL, rule);
2830
2831         ifnet_forwardmsg(lmsg, mycpuid + 1);
2832 }
2833
2834 static void
2835 ipfw_disable_rule_state_dispatch(netmsg_t nmsg)
2836 {
2837         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2838         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2839         struct ip_fw *rule;
2840
2841         ctx->ipfw_gen++;
2842
2843         rule = dmsg->start_rule;
2844         if (rule != NULL) {
2845                 KKASSERT(rule->cpuid == mycpuid);
2846
2847                 /*
2848                  * Move to the position on the next CPU
2849                  * before the msg is forwarded.
2850                  */
2851                 dmsg->start_rule = rule->sibling;
2852         } else {
2853                 KKASSERT(dmsg->rulenum == 0);
2854                 rule = ctx->ipfw_layer3_chain;
2855         }
2856
2857         while (rule != NULL) {
2858                 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2859                         break;
2860                 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2861                 rule = rule->next;
2862         }
2863
2864         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2865 }
2866
2867 /*
2868  * Deletes all rules from a chain (including the default rule
2869  * if the second argument is set).
2870  * Must be called at splimp().
2871  */
2872 static void
2873 ipfw_flush(int kill_default)
2874 {
2875         struct netmsg_del dmsg;
2876         struct netmsg_base nmsg;
2877         struct lwkt_msg *lmsg;
2878         struct ip_fw *rule;
2879         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2880
2881         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2882
2883         /*
2884          * If 'kill_default' then caller has done the necessary
2885          * msgport syncing; unnecessary to do it again.
2886          */
2887         if (!kill_default) {
2888                 /*
2889                  * Let ipfw_chk() know the rules are going to
2890                  * be flushed, so it could jump directly to
2891                  * the default rule.
2892                  */
2893                 ipfw_flushing = 1;
2894                 netmsg_service_sync();
2895         }
2896
2897         /*
2898          * Clear STATE flag on rules, so no more states (dyn rules)
2899          * will be created.
2900          */
2901         bzero(&dmsg, sizeof(dmsg));
2902         netmsg_init(&dmsg.base, NULL, &curthread->td_msgport,
2903                     0, ipfw_disable_rule_state_dispatch);
2904         ifnet_domsg(&dmsg.base.lmsg, 0);
2905
2906         /*
2907          * This actually nukes all states (dyn rules)
2908          */
2909         lockmgr(&dyn_lock, LK_EXCLUSIVE);
2910         for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2911                 /*
2912                  * Can't check IPFW_RULE_F_STATE here,
2913                  * since it has been cleared previously.
2914                  * Check 'stub' instead.
2915                  */
2916                 if (rule->stub != NULL) {
2917                         /* Force removal */
2918                         remove_dyn_rule_locked(rule, NULL);
2919                 }
2920         }
2921         lockmgr(&dyn_lock, LK_RELEASE);
2922
2923         /*
2924          * Press the 'flush' button
2925          */
2926         bzero(&nmsg, sizeof(nmsg));
2927         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
2928                     0, ipfw_flush_dispatch);
2929         lmsg = &nmsg.lmsg;
2930         lmsg->u.ms_result = kill_default;
2931         ifnet_domsg(lmsg, 0);
2932
2933         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2934
2935         if (kill_default) {
2936                 if (ipfw_dyn_v != NULL) {
2937                         /*
2938                          * Free dynamic rules(state) hash table
2939                          */
2940                         kfree(ipfw_dyn_v, M_IPFW);
2941                         ipfw_dyn_v = NULL;
2942                 }
2943
2944                 KASSERT(static_count == 0,
2945                         ("%u static rules remains\n", static_count));
2946                 KASSERT(static_ioc_len == 0,
2947                         ("%u bytes of static rules remains\n", static_ioc_len));
2948         } else {
2949                 KASSERT(static_count == 1,
2950                         ("%u static rules remains\n", static_count));
2951                 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2952                         ("%u bytes of static rules remains, should be %lu\n",
2953                          static_ioc_len,
2954                          (u_long)IOC_RULESIZE(ctx->ipfw_default_rule)));
2955         }
2956
2957         /* Flush is done */
2958         ipfw_flushing = 0;
2959 }
2960
2961 static void
2962 ipfw_alt_delete_rule_dispatch(netmsg_t nmsg)
2963 {
2964         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2965         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2966         struct ip_fw *rule, *prev;
2967
2968         rule = dmsg->start_rule;
2969         KKASSERT(rule->cpuid == mycpuid);
2970         dmsg->start_rule = rule->sibling;
2971
2972         prev = dmsg->prev_rule;
2973         if (prev != NULL) {
2974                 KKASSERT(prev->cpuid == mycpuid);
2975
2976                 /*
2977                  * Move to the position on the next CPU
2978                  * before the msg is forwarded.
2979                  */
2980                 dmsg->prev_rule = prev->sibling;
2981         }
2982
2983         /*
2984          * flush pointers outside the loop, then delete all matching
2985          * rules.  'prev' remains the same throughout the cycle.
2986          */
2987         ipfw_flush_rule_ptrs(ctx);
2988         while (rule && rule->rulenum == dmsg->rulenum)
2989                 rule = ipfw_delete_rule(ctx, prev, rule);
2990
2991         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2992 }
2993
2994 static int
2995 ipfw_alt_delete_rule(uint16_t rulenum)
2996 {
2997         struct ip_fw *prev, *rule, *f;
2998         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2999         struct netmsg_del dmsg;
3000         struct netmsg_base *nmsg;
3001         int state;
3002
3003         /*
3004          * Locate first rule to delete
3005          */
3006         for (prev = NULL, rule = ctx->ipfw_layer3_chain;
3007              rule && rule->rulenum < rulenum;
3008              prev = rule, rule = rule->next)
3009                 ; /* EMPTY */
3010         if (rule->rulenum != rulenum)
3011                 return EINVAL;
3012
3013         /*
3014          * Check whether any rules with the given number will
3015          * create states.
3016          */
3017         state = 0;
3018         for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3019                 if (f->rule_flags & IPFW_RULE_F_STATE) {
3020                         state = 1;
3021                         break;
3022                 }
3023         }
3024
3025         if (state) {
3026                 /*
3027                  * Clear the STATE flag, so no more states will be
3028                  * created based the rules numbered 'rulenum'.
3029                  */
3030                 bzero(&dmsg, sizeof(dmsg));
3031                 nmsg = &dmsg.base;
3032                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3033                             0, ipfw_disable_rule_state_dispatch);
3034                 dmsg.start_rule = rule;
3035                 dmsg.rulenum = rulenum;
3036
3037                 ifnet_domsg(&nmsg->lmsg, 0);
3038                 KKASSERT(dmsg.start_rule == NULL);
3039
3040                 /*
3041                  * Nuke all related states
3042                  */
3043                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3044                 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3045                         /*
3046                          * Can't check IPFW_RULE_F_STATE here,
3047                          * since it has been cleared previously.
3048                          * Check 'stub' instead.
3049                          */
3050                         if (f->stub != NULL) {
3051                                 /* Force removal */
3052                                 remove_dyn_rule_locked(f, NULL);
3053                         }
3054                 }
3055                 lockmgr(&dyn_lock, LK_RELEASE);
3056         }
3057
3058         /*
3059          * Get rid of the rule duplications on all CPUs
3060          */
3061         bzero(&dmsg, sizeof(dmsg));
3062         nmsg = &dmsg.base;
3063         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3064                     0, ipfw_alt_delete_rule_dispatch);
3065         dmsg.prev_rule = prev;
3066         dmsg.start_rule = rule;
3067         dmsg.rulenum = rulenum;
3068
3069         ifnet_domsg(&nmsg->lmsg, 0);
3070         KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3071         return 0;
3072 }
3073
3074 static void
3075 ipfw_alt_delete_ruleset_dispatch(netmsg_t nmsg)
3076 {
3077         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3078         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3079         struct ip_fw *prev, *rule;
3080 #ifdef INVARIANTS
3081         int del = 0;
3082 #endif
3083
3084         ipfw_flush_rule_ptrs(ctx);
3085
3086         prev = NULL;
3087         rule = ctx->ipfw_layer3_chain;
3088         while (rule != NULL) {
3089                 if (rule->set == dmsg->from_set) {
3090                         rule = ipfw_delete_rule(ctx, prev, rule);
3091 #ifdef INVARIANTS
3092                         del = 1;
3093 #endif
3094                 } else {
3095                         prev = rule;
3096                         rule = rule->next;
3097                 }
3098         }
3099         KASSERT(del, ("no match set?!\n"));
3100
3101         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3102 }
3103
3104 static void
3105 ipfw_disable_ruleset_state_dispatch(netmsg_t nmsg)
3106 {
3107         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3108         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3109         struct ip_fw *rule;
3110 #ifdef INVARIANTS
3111         int cleared = 0;
3112 #endif
3113
3114         ctx->ipfw_gen++;
3115
3116         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3117                 if (rule->set == dmsg->from_set) {
3118 #ifdef INVARIANTS
3119                         cleared = 1;
3120 #endif
3121                         rule->rule_flags &= ~IPFW_RULE_F_STATE;
3122                 }
3123         }
3124         KASSERT(cleared, ("no match set?!\n"));
3125
3126         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3127 }
3128
3129 static int
3130 ipfw_alt_delete_ruleset(uint8_t set)
3131 {
3132         struct netmsg_del dmsg;
3133         struct netmsg_base *nmsg;
3134         int state, del;
3135         struct ip_fw *rule;
3136         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3137
3138         /*
3139          * Check whether the 'set' exists.  If it exists,
3140          * then check whether any rules within the set will
3141          * try to create states.
3142          */
3143         state = 0;
3144         del = 0;
3145         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3146                 if (rule->set == set) {
3147                         del = 1;
3148                         if (rule->rule_flags & IPFW_RULE_F_STATE) {
3149                                 state = 1;
3150                                 break;
3151                         }
3152                 }
3153         }
3154         if (!del)
3155                 return 0; /* XXX EINVAL? */
3156
3157         if (state) {
3158                 /*
3159                  * Clear the STATE flag, so no more states will be
3160                  * created based the rules in this set.
3161                  */
3162                 bzero(&dmsg, sizeof(dmsg));
3163                 nmsg = &dmsg.base;
3164                 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3165                             0, ipfw_disable_ruleset_state_dispatch);
3166                 dmsg.from_set = set;
3167
3168                 ifnet_domsg(&nmsg->lmsg, 0);
3169
3170                 /*
3171                  * Nuke all related states
3172                  */
3173                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3174                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3175                         if (rule->set != set)
3176                                 continue;
3177
3178                         /*
3179                          * Can't check IPFW_RULE_F_STATE here,
3180                          * since it has been cleared previously.
3181                          * Check 'stub' instead.
3182                          */
3183                         if (rule->stub != NULL) {
3184                                 /* Force removal */
3185                                 remove_dyn_rule_locked(rule, NULL);
3186                         }
3187                 }
3188                 lockmgr(&dyn_lock, LK_RELEASE);
3189         }
3190
3191         /*
3192          * Delete this set
3193          */
3194         bzero(&dmsg, sizeof(dmsg));
3195         nmsg = &dmsg.base;
3196         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3197                     0, ipfw_alt_delete_ruleset_dispatch);
3198         dmsg.from_set = set;
3199
3200         ifnet_domsg(&nmsg->lmsg, 0);
3201         return 0;
3202 }
3203
3204 static void
3205 ipfw_alt_move_rule_dispatch(netmsg_t nmsg)
3206 {
3207         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3208         struct ip_fw *rule;
3209
3210         rule = dmsg->start_rule;
3211         KKASSERT(rule->cpuid == mycpuid);
3212
3213         /*
3214          * Move to the position on the next CPU
3215          * before the msg is forwarded.
3216          */
3217         dmsg->start_rule = rule->sibling;
3218
3219         while (rule && rule->rulenum <= dmsg->rulenum) {
3220                 if (rule->rulenum == dmsg->rulenum)
3221                         rule->set = dmsg->to_set;
3222                 rule = rule->next;
3223         }
3224         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3225 }
3226
3227 static int
3228 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3229 {
3230         struct netmsg_del dmsg;
3231         struct netmsg_base *nmsg;
3232         struct ip_fw *rule;
3233         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3234
3235         /*
3236          * Locate first rule to move
3237          */
3238         for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3239              rule = rule->next) {
3240                 if (rule->rulenum == rulenum && rule->set != set)
3241                         break;
3242         }
3243         if (rule == NULL || rule->rulenum > rulenum)
3244                 return 0; /* XXX error? */
3245
3246         bzero(&dmsg, sizeof(dmsg));
3247         nmsg = &dmsg.base;
3248         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3249                     0, ipfw_alt_move_rule_dispatch);
3250         dmsg.start_rule = rule;
3251         dmsg.rulenum = rulenum;
3252         dmsg.to_set = set;
3253
3254         ifnet_domsg(&nmsg->lmsg, 0);
3255         KKASSERT(dmsg.start_rule == NULL);
3256         return 0;
3257 }
3258
3259 static void
3260 ipfw_alt_move_ruleset_dispatch(netmsg_t nmsg)
3261 {
3262         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3263         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3264         struct ip_fw *rule;
3265
3266         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3267                 if (rule->set == dmsg->from_set)
3268                         rule->set = dmsg->to_set;
3269         }
3270         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3271 }
3272
3273 static int
3274 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3275 {
3276         struct netmsg_del dmsg;
3277         struct netmsg_base *nmsg;
3278
3279         bzero(&dmsg, sizeof(dmsg));
3280         nmsg = &dmsg.base;
3281         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3282                     0, ipfw_alt_move_ruleset_dispatch);
3283         dmsg.from_set = from_set;
3284         dmsg.to_set = to_set;
3285
3286         ifnet_domsg(&nmsg->lmsg, 0);
3287         return 0;
3288 }
3289
3290 static void
3291 ipfw_alt_swap_ruleset_dispatch(netmsg_t nmsg)
3292 {
3293         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3294         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3295         struct ip_fw *rule;
3296
3297         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3298                 if (rule->set == dmsg->from_set)
3299                         rule->set = dmsg->to_set;
3300                 else if (rule->set == dmsg->to_set)
3301                         rule->set = dmsg->from_set;
3302         }
3303         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3304 }
3305
3306 static int
3307 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3308 {
3309         struct netmsg_del dmsg;
3310         struct netmsg_base *nmsg;
3311
3312         bzero(&dmsg, sizeof(dmsg));
3313         nmsg = &dmsg.base;
3314         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3315                     0, ipfw_alt_swap_ruleset_dispatch);
3316         dmsg.from_set = set1;
3317         dmsg.to_set = set2;
3318
3319         ifnet_domsg(&nmsg->lmsg, 0);
3320         return 0;
3321 }
3322
3323 /**
3324  * Remove all rules with given number, and also do set manipulation.
3325  *
3326  * The argument is an uint32_t. The low 16 bit are the rule or set number,
3327  * the next 8 bits are the new set, the top 8 bits are the command:
3328  *
3329  *      0       delete rules with given number
3330  *      1       delete rules with given set number
3331  *      2       move rules with given number to new set
3332  *      3       move rules with given set number to new set
3333  *      4       swap sets with given numbers
3334  */
3335 static int
3336 ipfw_ctl_alter(uint32_t arg)
3337 {
3338         uint16_t rulenum;
3339         uint8_t cmd, new_set;
3340         int error = 0;
3341
3342         rulenum = arg & 0xffff;
3343         cmd = (arg >> 24) & 0xff;
3344         new_set = (arg >> 16) & 0xff;
3345
3346         if (cmd > 4)
3347                 return EINVAL;
3348         if (new_set >= IPFW_DEFAULT_SET)
3349                 return EINVAL;
3350         if (cmd == 0 || cmd == 2) {
3351                 if (rulenum == IPFW_DEFAULT_RULE)
3352                         return EINVAL;
3353         } else {
3354                 if (rulenum >= IPFW_DEFAULT_SET)
3355                         return EINVAL;
3356         }
3357
3358         switch (cmd) {
3359         case 0: /* delete rules with given number */
3360                 error = ipfw_alt_delete_rule(rulenum);
3361                 break;
3362
3363         case 1: /* delete all rules with given set number */
3364                 error = ipfw_alt_delete_ruleset(rulenum);
3365                 break;
3366
3367         case 2: /* move rules with given number to new set */
3368                 error = ipfw_alt_move_rule(rulenum, new_set);
3369                 break;
3370
3371         case 3: /* move rules with given set number to new set */
3372                 error = ipfw_alt_move_ruleset(rulenum, new_set);
3373                 break;
3374
3375         case 4: /* swap two sets */
3376                 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3377                 break;
3378         }
3379         return error;
3380 }
3381
3382 /*
3383  * Clear counters for a specific rule.
3384  */
3385 static void
3386 clear_counters(struct ip_fw *rule, int log_only)
3387 {
3388         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3389
3390         if (log_only == 0) {
3391                 rule->bcnt = rule->pcnt = 0;
3392                 rule->timestamp = 0;
3393         }
3394         if (l->o.opcode == O_LOG)
3395                 l->log_left = l->max_log;
3396 }
3397
3398 static void
3399 ipfw_zero_entry_dispatch(netmsg_t nmsg)
3400 {
3401         struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3402         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3403         struct ip_fw *rule;
3404
3405         if (zmsg->rulenum == 0) {
3406                 KKASSERT(zmsg->start_rule == NULL);
3407
3408                 ctx->ipfw_norule_counter = 0;
3409                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3410                         clear_counters(rule, zmsg->log_only);
3411         } else {
3412                 struct ip_fw *start = zmsg->start_rule;
3413
3414                 KKASSERT(start->cpuid == mycpuid);
3415                 KKASSERT(start->rulenum == zmsg->rulenum);
3416
3417                 /*
3418                  * We can have multiple rules with the same number, so we
3419                  * need to clear them all.
3420                  */
3421                 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3422                      rule = rule->next)
3423                         clear_counters(rule, zmsg->log_only);
3424
3425                 /*
3426                  * Move to the position on the next CPU
3427                  * before the msg is forwarded.
3428                  */
3429                 zmsg->start_rule = start->sibling;
3430         }
3431         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3432 }
3433
3434 /**
3435  * Reset some or all counters on firewall rules.
3436  * @arg frwl is null to clear all entries, or contains a specific
3437  * rule number.
3438  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3439  */
3440 static int
3441 ipfw_ctl_zero_entry(int rulenum, int log_only)
3442 {
3443         struct netmsg_zent zmsg;
3444         struct netmsg_base *nmsg;
3445         const char *msg;
3446         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3447
3448         bzero(&zmsg, sizeof(zmsg));
3449         nmsg = &zmsg.base;
3450         netmsg_init(nmsg, NULL, &curthread->td_msgport,
3451                     0, ipfw_zero_entry_dispatch);
3452         zmsg.log_only = log_only;
3453
3454         if (rulenum == 0) {
3455                 msg = log_only ? "ipfw: All logging counts reset.\n"
3456                                : "ipfw: Accounting cleared.\n";
3457         } else {
3458                 struct ip_fw *rule;
3459
3460                 /*
3461                  * Locate the first rule with 'rulenum'
3462                  */
3463                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3464                         if (rule->rulenum == rulenum)
3465                                 break;
3466                 }
3467                 if (rule == NULL) /* we did not find any matching rules */
3468                         return (EINVAL);
3469                 zmsg.start_rule = rule;
3470                 zmsg.rulenum = rulenum;
3471
3472                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3473                                : "ipfw: Entry %d cleared.\n";
3474         }
3475         ifnet_domsg(&nmsg->lmsg, 0);
3476         KKASSERT(zmsg.start_rule == NULL);
3477
3478         if (fw_verbose)
3479                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3480         return (0);
3481 }
3482
3483 /*
3484  * Check validity of the structure before insert.
3485  * Fortunately rules are simple, so this mostly need to check rule sizes.
3486  */
3487 static int
3488 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3489 {
3490         int l, cmdlen = 0;
3491         int have_action = 0;
3492         ipfw_insn *cmd;
3493
3494         *rule_flags = 0;
3495
3496         /* Check for valid size */
3497         if (size < sizeof(*rule)) {
3498                 kprintf("ipfw: rule too short\n");
3499                 return EINVAL;
3500         }
3501         l = IOC_RULESIZE(rule);
3502         if (l != size) {
3503                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3504                 return EINVAL;
3505         }
3506
3507         /* Check rule number */
3508         if (rule->rulenum == IPFW_DEFAULT_RULE) {
3509                 kprintf("ipfw: invalid rule number\n");
3510                 return EINVAL;
3511         }
3512
3513         /*
3514          * Now go for the individual checks. Very simple ones, basically only
3515          * instruction sizes.
3516          */
3517         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3518              l -= cmdlen, cmd += cmdlen) {
3519                 cmdlen = F_LEN(cmd);
3520                 if (cmdlen > l) {
3521                         kprintf("ipfw: opcode %d size truncated\n",
3522                                 cmd->opcode);
3523                         return EINVAL;
3524                 }
3525
3526                 DPRINTF("ipfw: opcode %d\n", cmd->opcode);
3527
3528                 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3529                         /* This rule will create states */
3530                         *rule_flags |= IPFW_RULE_F_STATE;
3531                 }
3532
3533                 switch (cmd->opcode) {
3534                 case O_NOP:
3535                 case O_PROBE_STATE:
3536                 case O_KEEP_STATE:
3537                 case O_PROTO:
3538                 case O_IP_SRC_ME:
3539                 case O_IP_DST_ME:
3540                 case O_LAYER2:
3541                 case O_IN:
3542                 case O_FRAG:
3543                 case O_IPOPT:
3544                 case O_IPLEN:
3545                 case O_IPID:
3546                 case O_IPTOS:
3547                 case O_IPPRECEDENCE:
3548                 case O_IPTTL:
3549                 case O_IPVER:
3550                 case O_TCPWIN:
3551                 case O_TCPFLAGS:
3552                 case O_TCPOPTS:
3553                 case O_ESTAB:
3554                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3555                                 goto bad_size;
3556                         break;
3557
3558                 case O_UID:
3559                 case O_GID:
3560                 case O_IP_SRC:
3561                 case O_IP_DST:
3562                 case O_TCPSEQ:
3563                 case O_TCPACK:
3564                 case O_PROB:
3565                 case O_ICMPTYPE:
3566                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3567                                 goto bad_size;
3568                         break;
3569
3570                 case O_LIMIT:
3571                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3572                                 goto bad_size;
3573                         break;
3574
3575                 case O_LOG:
3576                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3577                                 goto bad_size;
3578
3579                         ((ipfw_insn_log *)cmd)->log_left =
3580                             ((ipfw_insn_log *)cmd)->max_log;
3581
3582                         break;
3583
3584                 case O_IP_SRC_MASK:
3585                 case O_IP_DST_MASK:
3586                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3587                                 goto bad_size;
3588                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3589                                 kprintf("ipfw: opcode %d, useless rule\n",
3590                                         cmd->opcode);
3591                                 return EINVAL;
3592                         }
3593                         break;
3594
3595                 case O_IP_SRC_SET:
3596                 case O_IP_DST_SET:
3597                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3598                                 kprintf("ipfw: invalid set size %d\n",
3599                                         cmd->arg1);
3600                                 return EINVAL;
3601                         }
3602                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3603                             (cmd->arg1+31)/32 )
3604                                 goto bad_size;
3605                         break;
3606
3607                 case O_MACADDR2:
3608                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3609                                 goto bad_size;
3610                         break;
3611
3612                 case O_MAC_TYPE:
3613                 case O_IP_SRCPORT:
3614                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3615                         if (cmdlen < 2 || cmdlen > 31)
3616                                 goto bad_size;
3617                         break;
3618
3619                 case O_RECV:
3620                 case O_XMIT:
3621                 case O_VIA:
3622                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3623                                 goto bad_size;
3624                         break;
3625
3626                 case O_PIPE:
3627                 case O_QUEUE:
3628                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3629                                 goto bad_size;
3630                         goto check_action;
3631
3632                 case O_FORWARD_IP:
3633                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3634                                 goto bad_size;
3635                         } else {
3636                                 in_addr_t fwd_addr;
3637
3638                                 fwd_addr = ((ipfw_insn_sa *)cmd)->
3639                                            sa.sin_addr.s_addr;
3640                                 if (IN_MULTICAST(ntohl(fwd_addr))) {
3641                                         kprintf("ipfw: try forwarding to "
3642                                                 "multicast address\n");
3643                                         return EINVAL;
3644                                 }
3645                         }
3646                         goto check_action;
3647
3648                 case O_FORWARD_MAC: /* XXX not implemented yet */
3649                 case O_CHECK_STATE:
3650                 case O_COUNT:
3651                 case O_ACCEPT:
3652                 case O_DENY:
3653                 case O_REJECT:
3654                 case O_SKIPTO:
3655                 case O_DIVERT:
3656                 case O_TEE:
3657                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3658                                 goto bad_size;
3659 check_action:
3660                         if (have_action) {
3661                                 kprintf("ipfw: opcode %d, multiple actions"
3662                                         " not allowed\n",
3663                                         cmd->opcode);
3664                                 return EINVAL;
3665                         }
3666                         have_action = 1;
3667                         if (l != cmdlen) {
3668                                 kprintf("ipfw: opcode %d, action must be"
3669                                         " last opcode\n",
3670                                         cmd->opcode);
3671                                 return EINVAL;
3672                         }
3673                         break;
3674                 default:
3675                         kprintf("ipfw: opcode %d, unknown opcode\n",
3676                                 cmd->opcode);
3677                         return EINVAL;
3678                 }
3679         }
3680         if (have_action == 0) {
3681                 kprintf("ipfw: missing action\n");
3682                 return EINVAL;
3683         }
3684         return 0;
3685
3686 bad_size:
3687         kprintf("ipfw: opcode %d size %d wrong\n",
3688                 cmd->opcode, cmdlen);
3689         return EINVAL;
3690 }
3691
3692 static int
3693 ipfw_ctl_add_rule(struct sockopt *sopt)
3694 {
3695         struct ipfw_ioc_rule *ioc_rule;
3696         size_t size;
3697         uint32_t rule_flags;
3698         int error;
3699
3700         size = sopt->sopt_valsize;
3701         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3702             size < sizeof(*ioc_rule)) {
3703                 return EINVAL;
3704         }
3705         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3706                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3707                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3708         }
3709         ioc_rule = sopt->sopt_val;
3710
3711         error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3712         if (error)
3713                 return error;
3714
3715         ipfw_add_rule(ioc_rule, rule_flags);
3716
3717         if (sopt->sopt_dir == SOPT_GET)
3718                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3719         return 0;
3720 }
3721
3722 static void *
3723 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3724 {
3725         const struct ip_fw *sibling;
3726 #ifdef INVARIANTS
3727         int i;
3728 #endif
3729
3730         KKASSERT(rule->cpuid == IPFW_CFGCPUID);
3731
3732         ioc_rule->act_ofs = rule->act_ofs;
3733         ioc_rule->cmd_len = rule->cmd_len;
3734         ioc_rule->rulenum = rule->rulenum;
3735         ioc_rule->set = rule->set;
3736         ioc_rule->usr_flags = rule->usr_flags;
3737
3738         ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3739         ioc_rule->static_count = static_count;
3740         ioc_rule->static_len = static_ioc_len;
3741
3742         /*
3743          * Visit (read-only) all of the rule's duplications to get
3744          * the necessary statistics
3745          */
3746 #ifdef INVARIANTS
3747         i = 0;
3748 #endif
3749         ioc_rule->pcnt = 0;
3750         ioc_rule->bcnt = 0;
3751         ioc_rule->timestamp = 0;
3752         for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3753                 ioc_rule->pcnt += sibling->pcnt;
3754                 ioc_rule->bcnt += sibling->bcnt;
3755                 if (sibling->timestamp > ioc_rule->timestamp)
3756                         ioc_rule->timestamp = sibling->timestamp;
3757 #ifdef INVARIANTS
3758                 ++i;
3759 #endif
3760         }
3761         KASSERT(i == ncpus, ("static rule is not duplicated on every cpu\n"));
3762
3763         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3764
3765         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3766 }
3767
3768 static void
3769 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3770                 struct ipfw_ioc_state *ioc_state)
3771 {
3772         const struct ipfw_flow_id *id;
3773         struct ipfw_ioc_flowid *ioc_id;
3774
3775         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3776                             0 : dyn_rule->expire - time_second;
3777         ioc_state->pcnt = dyn_rule->pcnt;
3778         ioc_state->bcnt = dyn_rule->bcnt;
3779
3780         ioc_state->dyn_type = dyn_rule->dyn_type;
3781         ioc_state->count = dyn_rule->count;
3782
3783         ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3784
3785         id = &dyn_rule->id;
3786         ioc_id = &ioc_state->id;
3787
3788         ioc_id->type = ETHERTYPE_IP;
3789         ioc_id->u.ip.dst_ip = id->dst_ip;
3790         ioc_id->u.ip.src_ip = id->src_ip;
3791         ioc_id->u.ip.dst_port = id->dst_port;
3792         ioc_id->u.ip.src_port = id->src_port;
3793         ioc_id->u.ip.proto = id->proto;
3794 }
3795
3796 static int
3797 ipfw_ctl_get_rules(struct sockopt *sopt)
3798 {
3799         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3800         struct ip_fw *rule;
3801         void *bp;
3802         size_t size;
3803         uint32_t dcount = 0;
3804
3805         /*
3806          * pass up a copy of the current rules. Static rules
3807          * come first (the last of which has number IPFW_DEFAULT_RULE),
3808          * followed by a possibly empty list of dynamic rule.
3809          */
3810
3811         size = static_ioc_len;  /* size of static rules */
3812         if (ipfw_dyn_v) {       /* add size of dyn.rules */
3813                 dcount = dyn_count;
3814                 size += dcount * sizeof(struct ipfw_ioc_state);
3815         }
3816
3817         if (sopt->sopt_valsize < size) {
3818                 /* short length, no need to return incomplete rules */
3819                 /* XXX: if superuser, no need to zero buffer */
3820                 bzero(sopt->sopt_val, sopt->sopt_valsize);
3821                 return 0;
3822         }
3823         bp = sopt->sopt_val;
3824
3825         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3826                 bp = ipfw_copy_rule(rule, bp);
3827
3828         if (ipfw_dyn_v && dcount != 0) {
3829                 struct ipfw_ioc_state *ioc_state = bp;
3830                 uint32_t dcount2 = 0;
3831 #ifdef INVARIANTS
3832                 size_t old_size = size;
3833 #endif
3834                 int i;
3835
3836                 lockmgr(&dyn_lock, LK_SHARED);
3837
3838                 /* Check 'ipfw_dyn_v' again with lock held */
3839                 if (ipfw_dyn_v == NULL)
3840                         goto skip;
3841
3842                 for (i = 0; i < curr_dyn_buckets; i++) {
3843                         ipfw_dyn_rule *p;
3844
3845                         /*
3846                          * The # of dynamic rules may have grown after the
3847                          * snapshot of 'dyn_count' was taken, so we will have
3848                          * to check 'dcount' (snapshot of dyn_count) here to
3849                          * make sure that we don't overflow the pre-allocated
3850                          * buffer.
3851                          */
3852                         for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
3853                              p = p->next, ioc_state++, dcount--, dcount2++)
3854                                 ipfw_copy_state(p, ioc_state);
3855                 }
3856 skip:
3857                 lockmgr(&dyn_lock, LK_RELEASE);
3858
3859                 /*
3860                  * The # of dynamic rules may be shrinked after the
3861                  * snapshot of 'dyn_count' was taken.  To give user a
3862                  * correct dynamic rule count, we use the 'dcount2'
3863                  * calculated above (with shared lockmgr lock held).
3864                  */
3865                 size = static_ioc_len +
3866                        (dcount2 * sizeof(struct ipfw_ioc_state));
3867                 KKASSERT(size <= old_size);
3868         }
3869
3870         sopt->sopt_valsize = size;
3871         return 0;
3872 }
3873
3874 static void
3875 ipfw_set_disable_dispatch(netmsg_t nmsg)
3876 {
3877         struct lwkt_msg *lmsg = &nmsg->lmsg;
3878         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3879
3880         ctx->ipfw_gen++;
3881         ctx->ipfw_set_disable = lmsg->u.ms_result32;
3882
3883         ifnet_forwardmsg(lmsg, mycpuid + 1);
3884 }
3885
3886 static void
3887 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable)
3888 {
3889         struct netmsg_base nmsg;
3890         struct lwkt_msg *lmsg;
3891         uint32_t set_disable;
3892
3893         /* IPFW_DEFAULT_SET is always enabled */
3894         enable |= (1 << IPFW_DEFAULT_SET);
3895         set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable;
3896
3897         bzero(&nmsg, sizeof(nmsg));
3898         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
3899                     0, ipfw_set_disable_dispatch);
3900         lmsg = &nmsg.lmsg;
3901         lmsg->u.ms_result32 = set_disable;
3902
3903         ifnet_domsg(lmsg, 0);
3904 }
3905
3906 /**
3907  * {set|get}sockopt parser.
3908  */
3909 static int
3910 ipfw_ctl(struct sockopt *sopt)
3911 {
3912         int error, rulenum;
3913         uint32_t *masks;
3914         size_t size;
3915
3916         error = 0;
3917
3918         switch (sopt->sopt_name) {
3919         case IP_FW_GET:
3920                 error = ipfw_ctl_get_rules(sopt);
3921                 break;
3922
3923         case IP_FW_FLUSH:
3924                 ipfw_flush(0 /* keep default rule */);
3925                 break;
3926
3927         case IP_FW_ADD:
3928                 error = ipfw_ctl_add_rule(sopt);
3929                 break;
3930
3931         case IP_FW_DEL:
3932                 /*
3933                  * IP_FW_DEL is used for deleting single rules or sets,
3934                  * and (ab)used to atomically manipulate sets.
3935                  * Argument size is used to distinguish between the two:
3936                  *    sizeof(uint32_t)
3937                  *      delete single rule or set of rules,
3938                  *      or reassign rules (or sets) to a different set.
3939                  *    2 * sizeof(uint32_t)
3940                  *      atomic disable/enable sets.
3941                  *      first uint32_t contains sets to be disabled,
3942                  *      second uint32_t contains sets to be enabled.
3943                  */
3944                 masks = sopt->sopt_val;
3945                 size = sopt->sopt_valsize;
3946                 if (size == sizeof(*masks)) {
3947                         /*
3948                          * Delete or reassign static rule
3949                          */
3950                         error = ipfw_ctl_alter(masks[0]);
3951                 } else if (size == (2 * sizeof(*masks))) {
3952                         /*
3953                          * Set enable/disable
3954                          */
3955                         ipfw_ctl_set_disable(masks[0], masks[1]);
3956                 } else {
3957                         error = EINVAL;
3958                 }
3959                 break;
3960
3961         case IP_FW_ZERO:
3962         case IP_FW_RESETLOG: /* argument is an int, the rule number */
3963                 rulenum = 0;
3964
3965                 if (sopt->sopt_val != 0) {
3966                     error = soopt_to_kbuf(sopt, &rulenum,
3967                             sizeof(int), sizeof(int));
3968                     if (error)
3969                         break;
3970                 }
3971                 error = ipfw_ctl_zero_entry(rulenum,
3972                         sopt->sopt_name == IP_FW_RESETLOG);
3973                 break;
3974
3975         default:
3976                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3977                 error = EINVAL;
3978         }
3979         return error;
3980 }
3981
3982 /*
3983  * This procedure is only used to handle keepalives. It is invoked
3984  * every dyn_keepalive_period
3985  */
3986 static void
3987 ipfw_tick_dispatch(netmsg_t nmsg)
3988 {
3989         time_t keep_alive;
3990         uint32_t gen;
3991         int i;
3992
3993         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
3994         KKASSERT(IPFW_LOADED);
3995
3996         /* Reply ASAP */
3997         crit_enter();
3998         lwkt_replymsg(&nmsg->lmsg, 0);
3999         crit_exit();
4000
4001         if (ipfw_dyn_v == NULL || dyn_count == 0)
4002                 goto done;
4003
4004         keep_alive = time_second;
4005
4006         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4007 again:
4008         if (ipfw_dyn_v == NULL || dyn_count == 0) {
4009                 lockmgr(&dyn_lock, LK_RELEASE);
4010                 goto done;
4011         }
4012         gen = dyn_buckets_gen;
4013
4014         for (i = 0; i < curr_dyn_buckets; i++) {
4015                 ipfw_dyn_rule *q, *prev;
4016
4017                 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
4018                         uint32_t ack_rev, ack_fwd;
4019                         struct ipfw_flow_id id;
4020
4021                         if (q->dyn_type == O_LIMIT_PARENT)
4022                                 goto next;
4023
4024                         if (TIME_LEQ(q->expire, time_second)) {
4025                                 /* State expired */
4026                                 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
4027                                 continue;
4028                         }
4029
4030                         /*
4031                          * Keep alive processing
4032                          */
4033
4034                         if (!dyn_keepalive)
4035                                 goto next;
4036                         if (q->id.proto != IPPROTO_TCP)
4037                                 goto next;
4038                         if ((q->state & BOTH_SYN) != BOTH_SYN)
4039                                 goto next;
4040                         if (TIME_LEQ(time_second + dyn_keepalive_interval,
4041                             q->expire))
4042                                 goto next;      /* too early */
4043                         if (q->keep_alive == keep_alive)
4044                                 goto next;      /* alreay done */
4045
4046                         /*
4047                          * Save necessary information, so that they could
4048                          * survive after possible blocking in send_pkt()
4049                          */
4050                         id = q->id;
4051                         ack_rev = q->ack_rev;
4052                         ack_fwd = q->ack_fwd;
4053
4054                         /* Sending has been started */
4055                         q->keep_alive = keep_alive;
4056
4057                         /* Release lock to avoid possible dead lock */
4058                         lockmgr(&dyn_lock, LK_RELEASE);
4059                         send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
4060                         send_pkt(&id, ack_fwd - 1, ack_rev, 0);
4061                         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4062
4063                         if (gen != dyn_buckets_gen) {
4064                                 /*
4065                                  * Dyn bucket array has been changed during
4066                                  * the above two sending; reiterate.
4067                                  */
4068                                 goto again;
4069                         }
4070 next:
4071                         prev = q;
4072                         q = q->next;
4073                 }
4074         }
4075         lockmgr(&dyn_lock, LK_RELEASE);
4076 done:
4077         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
4078                       ipfw_tick, NULL);
4079 }
4080
4081 /*
4082  * This procedure is only used to handle keepalives. It is invoked
4083  * every dyn_keepalive_period
4084  */
4085 static void
4086 ipfw_tick(void *dummy __unused)
4087 {
4088         struct lwkt_msg *lmsg = &ipfw_timeout_netmsg.lmsg;
4089
4090         KKASSERT(mycpuid == IPFW_CFGCPUID);
4091
4092         crit_enter();
4093
4094         KKASSERT(lmsg->ms_flags & MSGF_DONE);
4095         if (IPFW_LOADED) {
4096                 lwkt_sendmsg(IPFW_CFGPORT, lmsg);
4097                 /* ipfw_timeout_netmsg's handler reset this callout */
4098         }
4099
4100         crit_exit();
4101 }
4102
4103 static int
4104 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4105 {
4106         struct ip_fw_args args;
4107         struct mbuf *m = *m0;
4108         struct m_tag *mtag;
4109         int tee = 0, error = 0, ret;
4110
4111         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4112                 /* Extract info from dummynet tag */
4113                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4114                 KKASSERT(mtag != NULL);
4115                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4116                 KKASSERT(args.rule != NULL);
4117
4118                 m_tag_delete(m, mtag);
4119                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4120         } else {
4121                 args.rule = NULL;
4122         }
4123
4124         args.eh = NULL;
4125         args.oif = NULL;
4126         args.m = m;
4127         ret = ipfw_chk(&args);
4128         m = args.m;
4129
4130         if (m == NULL) {
4131                 error = EACCES;
4132                 goto back;
4133         }
4134
4135         switch (ret) {
4136         case IP_FW_PASS:
4137                 break;
4138
4139         case IP_FW_DENY:
4140                 m_freem(m);
4141                 m = NULL;
4142                 error = EACCES;
4143                 break;
4144
4145         case IP_FW_DUMMYNET:
4146                 /* Send packet to the appropriate pipe */
4147                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args);
4148                 break;
4149
4150         case IP_FW_TEE:
4151                 tee = 1;
4152                 /* FALL THROUGH */
4153
4154         case IP_FW_DIVERT:
4155                 if (ip_divert_p != NULL) {
4156                         m = ip_divert_p(m, tee, 1);
4157                 } else {
4158                         m_freem(m);
4159                         m = NULL;
4160                         /* not sure this is the right error msg */
4161                         error = EACCES;
4162                 }
4163                 break;
4164
4165         default:
4166                 panic("unknown ipfw return value: %d\n", ret);
4167         }
4168 back:
4169         *m0 = m;
4170         return error;
4171 }
4172
4173 static int
4174 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4175 {
4176         struct ip_fw_args args;
4177         struct mbuf *m = *m0;
4178         struct m_tag *mtag;
4179         int tee = 0, error = 0, ret;
4180
4181         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4182                 /* Extract info from dummynet tag */
4183                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4184                 KKASSERT(mtag != NULL);
4185                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4186                 KKASSERT(args.rule != NULL);
4187
4188                 m_tag_delete(m, mtag);
4189                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4190         } else {
4191                 args.rule = NULL;
4192         }
4193
4194         args.eh = NULL;
4195         args.m = m;
4196         args.oif = ifp;
4197         ret = ipfw_chk(&args);
4198         m = args.m;
4199
4200         if (m == NULL) {
4201                 error = EACCES;
4202                 goto back;
4203         }
4204
4205         switch (ret) {
4206         case IP_FW_PASS:
4207                 break;
4208
4209         case IP_FW_DENY:
4210                 m_freem(m);
4211                 m = NULL;
4212                 error = EACCES;
4213                 break;
4214
4215         case IP_FW_DUMMYNET:
4216                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args);
4217                 break;
4218
4219         case IP_FW_TEE:
4220                 tee = 1;
4221                 /* FALL THROUGH */
4222
4223         case IP_FW_DIVERT:
4224                 if (ip_divert_p != NULL) {
4225                         m = ip_divert_p(m, tee, 0);
4226                 } else {
4227                         m_freem(m);
4228                         m = NULL;
4229                         /* not sure this is the right error msg */
4230                         error = EACCES;
4231                 }
4232                 break;
4233
4234         default:
4235                 panic("unknown ipfw return value: %d\n", ret);
4236         }
4237 back:
4238         *m0 = m;
4239         return error;
4240 }
4241
4242 static void
4243 ipfw_hook(void)
4244 {
4245         struct pfil_head *pfh;
4246
4247         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4248
4249         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4250         if (pfh == NULL)
4251                 return;
4252
4253         pfil_add_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_MPSAFE, pfh);
4254         pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_MPSAFE, pfh);
4255 }
4256
4257 static void
4258 ipfw_dehook(void)
4259 {
4260         struct pfil_head *pfh;
4261
4262         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4263
4264         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4265         if (pfh == NULL)
4266                 return;
4267
4268         pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN, pfh);
4269         pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT, pfh);
4270 }
4271
4272 static void
4273 ipfw_sysctl_enable_dispatch(netmsg_t nmsg)
4274 {
4275         struct lwkt_msg *lmsg = &nmsg->lmsg;
4276         int enable = lmsg->u.ms_result;
4277
4278         if (fw_enable == enable)
4279                 goto reply;
4280
4281         fw_enable = enable;
4282         if (fw_enable)
4283                 ipfw_hook();
4284         else
4285                 ipfw_dehook();
4286 reply:
4287         lwkt_replymsg(lmsg, 0);
4288 }
4289
4290 static int
4291 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS)
4292 {
4293         struct netmsg_base nmsg;
4294         struct lwkt_msg *lmsg;
4295         int enable, error;
4296
4297         enable = fw_enable;
4298         error = sysctl_handle_int(oidp, &enable, 0, req);
4299         if (error || req->newptr == NULL)
4300                 return error;
4301
4302         netmsg_init(&nmsg, NULL, &curthread->td_msgport,
4303                     0, ipfw_sysctl_enable_dispatch);
4304         lmsg = &nmsg.lmsg;
4305         lmsg->u.ms_result = enable;
4306
4307         return lwkt_domsg(IPFW_CFGPORT, lmsg, 0);
4308 }
4309
4310 static int
4311 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
4312 {
4313         return sysctl_int_range(oidp, arg1, arg2, req,
4314                IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
4315 }
4316
4317 static int
4318 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
4319 {
4320         int error, value;
4321
4322         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4323
4324         value = dyn_buckets;
4325         error = sysctl_handle_int(oidp, &value, 0, req);
4326         if (error || !req->newptr)
4327                 goto back;
4328
4329         /*
4330          * Make sure we have a power of 2 and
4331          * do not allow more than 64k entries.
4332          */
4333         error = EINVAL;
4334         if (value <= 1 || value > 65536)
4335                 goto back;
4336         if ((value & (value - 1)) != 0)
4337                 goto back;
4338
4339         error = 0;
4340         dyn_buckets = value;
4341 back:
4342         lockmgr(&dyn_lock, LK_RELEASE);
4343         return error;
4344 }
4345
4346 static int
4347 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
4348 {
4349         return sysctl_int_range(oidp, arg1, arg2, req,
4350                                 1, dyn_keepalive_period - 1);
4351 }
4352
4353 static int
4354 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
4355 {
4356         return sysctl_int_range(oidp, arg1, arg2, req,
4357                                 1, dyn_keepalive_period - 1);
4358 }
4359
4360 static void
4361 ipfw_ctx_init_dispatch(netmsg_t nmsg)
4362 {
4363         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
4364         struct ipfw_context *ctx;
4365         struct ip_fw *def_rule;
4366
4367         ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO);
4368         ipfw_ctx[mycpuid] = ctx;
4369
4370         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
4371
4372         def_rule->act_ofs = 0;
4373         def_rule->rulenum = IPFW_DEFAULT_RULE;
4374         def_rule->cmd_len = 1;
4375         def_rule->set = IPFW_DEFAULT_SET;
4376
4377         def_rule->cmd[0].len = 1;
4378 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4379         def_rule->cmd[0].opcode = O_ACCEPT;
4380 #else
4381         def_rule->cmd[0].opcode = O_DENY;
4382 #endif
4383
4384         def_rule->refcnt = 1;
4385         def_rule->cpuid = mycpuid;
4386
4387         /* Install the default rule */
4388         ctx->ipfw_default_rule = def_rule;
4389         ctx->ipfw_layer3_chain = def_rule;
4390
4391         /* Link rule CPU sibling */
4392         ipfw_link_sibling(fwmsg, def_rule);
4393
4394         /* Statistics only need to be updated once */
4395         if (mycpuid == 0)
4396                 ipfw_inc_static_count(def_rule);
4397
4398         ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
4399 }
4400
4401 static void
4402 ipfw_init_dispatch(netmsg_t nmsg)
4403 {
4404         struct netmsg_ipfw fwmsg;
4405         int error = 0;
4406
4407         if (IPFW_LOADED) {
4408                 kprintf("IP firewall already loaded\n");
4409                 error = EEXIST;
4410                 goto reply;
4411         }
4412
4413         bzero(&fwmsg, sizeof(fwmsg));
4414         netmsg_init(&fwmsg.base, NULL, &curthread->td_msgport,
4415                     0, ipfw_ctx_init_dispatch);
4416         ifnet_domsg(&fwmsg.base.lmsg, 0);
4417
4418         ip_fw_chk_ptr = ipfw_chk;
4419         ip_fw_ctl_ptr = ipfw_ctl;
4420         ip_fw_dn_io_ptr = ipfw_dummynet_io;
4421
4422         kprintf("ipfw2 initialized, default to %s, logging ",
4423                 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode ==
4424                 O_ACCEPT ? "accept" : "deny");
4425
4426 #ifdef IPFIREWALL_VERBOSE
4427         fw_verbose = 1;
4428 #endif
4429 #ifdef IPFIREWALL_VERBOSE_LIMIT
4430         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4431 #endif
4432         if (fw_verbose == 0) {
4433                 kprintf("disabled\n");
4434         } else if (verbose_limit == 0) {
4435                 kprintf("unlimited\n");
4436         } else {
4437                 kprintf("limited to %d packets/entry by default\n",
4438                         verbose_limit);
4439         }
4440
4441         callout_init_mp(&ipfw_timeout_h);
4442         netmsg_init(&ipfw_timeout_netmsg, NULL, &netisr_adone_rport,
4443                     MSGF_DROPABLE | MSGF_PRIORITY,
4444                     ipfw_tick_dispatch);
4445         lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
4446
4447         ip_fw_loaded = 1;
4448         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
4449
4450         if (fw_enable)
4451                 ipfw_hook();
4452 reply:
4453         lwkt_replymsg(&nmsg->lmsg, error);
4454 }
4455
4456 static int
4457 ipfw_init(void)
4458 {
4459         struct netmsg_base smsg;
4460
4461         netmsg_init(&smsg, NULL, &curthread->td_msgport,
4462                     0, ipfw_init_dispatch);
4463         return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4464 }
4465
4466 #ifdef KLD_MODULE
4467
4468 static void
4469 ipfw_fini_dispatch(netmsg_t nmsg)
4470 {
4471         int error = 0, cpu;
4472
4473         if (ipfw_refcnt != 0) {
4474                 error = EBUSY;
4475                 goto reply;
4476         }
4477
4478         ip_fw_loaded = 0;
4479
4480         ipfw_dehook();
4481         callout_stop(&ipfw_timeout_h);
4482
4483         netmsg_service_sync();
4484
4485         crit_enter();
4486         if ((ipfw_timeout_netmsg.lmsg.ms_flags & MSGF_DONE) == 0) {
4487                 /*
4488                  * Callout message is pending; drop it
4489                  */
4490                 lwkt_dropmsg(&ipfw_timeout_netmsg.lmsg);
4491         }
4492         crit_exit();
4493
4494         ip_fw_chk_ptr = NULL;
4495         ip_fw_ctl_ptr = NULL;
4496         ip_fw_dn_io_ptr = NULL;
4497         ipfw_flush(1 /* kill default rule */);
4498
4499         /* Free pre-cpu context */
4500         for (cpu = 0; cpu < ncpus; ++cpu)
4501                 kfree(ipfw_ctx[cpu], M_IPFW);
4502
4503         kprintf("IP firewall unloaded\n");
4504 reply:
4505         lwkt_replymsg(&nmsg->lmsg, error);
4506 }
4507
4508 static int
4509 ipfw_fini(void)
4510 {
4511         struct netmsg_base smsg;
4512
4513         netmsg_init(&smsg, NULL, &curthread->td_msgport,
4514                     0, ipfw_fini_dispatch);
4515         return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4516 }
4517
4518 #endif  /* KLD_MODULE */
4519
4520 static int
4521 ipfw_modevent(module_t mod, int type, void *unused)
4522 {
4523         int err = 0;
4524
4525         switch (type) {
4526         case MOD_LOAD:
4527                 err = ipfw_init();
4528                 break;
4529
4530         case MOD_UNLOAD:
4531 #ifndef KLD_MODULE
4532                 kprintf("ipfw statically compiled, cannot unload\n");
4533                 err = EBUSY;
4534 #else
4535                 err = ipfw_fini();
4536 #endif
4537                 break;
4538         default:
4539                 break;
4540         }
4541         return err;
4542 }
4543
4544 static moduledata_t ipfwmod = {
4545         "ipfw",
4546         ipfw_modevent,
4547         0
4548 };
4549 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
4550 MODULE_VERSION(ipfw, 1);