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 1.86 2008/09/16 11:28:31 sephe Exp $
  27  */
  28
  29 #define        DEB(x)
  30 #define        DDB(x) x
  31
  32 /*
  33  * Implement IP packet firewall (new version)
  34  */
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_inet.h"
  38 #ifndef INET
  39 #error IPFIREWALL requires INET.
  40 #endif /* INET */
  41
  42 #include <sys/param.h>
  43 #include <sys/systm.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mbuf.h>
  46 #include <sys/kernel.h>
  47 #include <sys/proc.h>
  48 #include <sys/socket.h>
  49 #include <sys/socketvar.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/syslog.h>
  52 #include <sys/thread2.h>
  53 #include <sys/ucred.h>
  54 #include <sys/in_cksum.h>
  55 #include <sys/lock.h>
  56
  57 #include <net/if.h>
  58 #include <net/route.h>
  59 #include <net/netmsg2.h>
  60 #include <net/pfil.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 "ip_fw.h"
  70 #include <net/dummynet/ip_dummynet.h>
  71 #include <netinet/tcp.h>
  72 #include <netinet/tcp_timer.h>
  73 #include <netinet/tcp_var.h>
  74 #include <netinet/tcpip.h>
  75 #include <netinet/udp.h>
  76 #include <netinet/udp_var.h>
  77 #include <netinet/ip_divert.h>
  78
  79 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
  80
  81 /*
  82  * Description about per-CPU rule duplication:
  83  *
  84  * Module loading/unloading and all ioctl operations are serialized
  85  * by netisr0, so we don't have any ordering or locking problems.
  86  *
  87  * Following graph shows how operation on per-CPU rule list is
  88  * performed [2 CPU case]:
  89  *
  90  *   CPU0                 CPU1
  91  *
  92  * netisr0 <------------------------------------+
  93  *  domsg                                       |
  94  *    |                                         |
  95  *    | netmsg                                  |
  96  *    |                                         |
  97  *    V                                         |
  98  *  ifnet0                                      |
  99  *    :                                         | netmsg
 100  *    :(delete/add...)                          |
 101  *    :                                         |
 102  *    :         netmsg                          |
 103  *  forwardmsg---------->ifnet1                 |
 104  *                          :                   |
 105  *                          :(delete/add...)    |
 106  *                          :                   |
 107  *                          :                   |
 108  *                        replymsg--------------+
 109  *
 110  *
 111  *
 112  *
 113  * Rules which will not create states (dyn rules) [2 CPU case]
 114  *
 115  *    CPU0               CPU1
 116  * layer3_chain       layer3_chain
 117  *     |                  |
 118  *     V                  V
 119  * +-------+ sibling  +-------+ sibling
 120  * | rule1 |--------->| rule1 |--------->NULL
 121  * +-------+          +-------+
 122  *     |                  |
 123  *     |next              |next
 124  *     V                  V
 125  * +-------+ sibling  +-------+ sibling
 126  * | rule2 |--------->| rule2 |--------->NULL
 127  * +-------+          +-------+
 128  *
 129  * ip_fw.sibling:
 130  * 1) Ease statistics calculation during IP_FW_GET.  We only need to
 131  *    iterate layer3_chain on CPU0; the current rule's duplication on
 132  *    the other CPUs could safely be read-only accessed by using
 133  *    ip_fw.sibling
 134  * 2) Accelerate rule insertion and deletion, e.g. rule insertion:
 135  *    a) In netisr0 (on CPU0) rule3 is determined to be inserted between
 136  *       rule1 and rule2.  To make this decision we need to iterate the
 137  *       layer3_chain on CPU0.  The netmsg, which is used to insert the
 138  *       rule, will contain rule1 on CPU0 as prev_rule and rule2 on CPU0
 139  *       as next_rule
 140  *    b) After the insertion on CPU0 is done, we will move on to CPU1.
 141  *       But instead of relocating the rule3's position on CPU1 by
 142  *       iterating the layer3_chain on CPU1, we set the netmsg's prev_rule
 143  *       to rule1->sibling and next_rule to rule2->sibling before the
 144  *       netmsg is forwarded to CPU1 from CPU0
 145  *
 146  *
 147  *
 148  * Rules which will create states (dyn rules) [2 CPU case]
 149  * (unnecessary parts are omitted; they are same as in the previous figure)
 150  *
 151  *   CPU0                       CPU1
 152  *
 153  * +-------+                  +-------+
 154  * | rule1 |                  | rule1 |
 155  * +-------+                  +-------+
 156  *   ^   |                      |   ^
 157  *   |   |stub              stub|   |
 158  *   |   |                      |   |
 159  *   |   +----+            +----+   |
 160  *   |        |            |        |
 161  *   |        V            V        |
 162  *   |    +--------------------+    |
 163  *   |    |     rule_stub      |    |
 164  *   |    | (read-only shared) |    |
 165  *   |    |                    |    |
 166  *   |    | back pointer array |    |
 167  *   |    | (indexed by cpuid) |    |
 168  *   |    |                    |    |
 169  *   +----|---------[0]        |    |
 170  *        |         [1]--------|----+
 171  *        |                    |
 172  *        +--------------------+
 173  *          ^            ^
 174  *          |            |
 175  *  ........|............|............
 176  *  :       |            |           :
 177  *  :       |stub        |stub       :
 178  *  :       |            |           :
 179  *  :  +---------+  +---------+      :
 180  *  :  | state1a |  | state1b | .... :
 181  *  :  +---------+  +---------+      :
 182  *  :                                :
 183  *  :           states table         :
 184  *  :            (shared)            :
 185  *  :      (protected by dyn_lock)   :
 186  *  ..................................
 187  *
 188  * [state1a and state1b are states created by rule1]
 189  *
 190  * ip_fw_stub:
 191  * This structure is introduced so that shared (locked) state table could
 192  * work with per-CPU (duplicated) static rules.  It mainly bridges states
 193  * and static rules and serves as static rule's place holder (a read-only
 194  * shared part of duplicated rules) from states point of view.
 195  *
 196  * IPFW_RULE_F_STATE (only for rules which create states):
 197  * o  During rule installation, this flag is turned on after rule's
 198  *    duplications reach all CPUs, to avoid at least following race:
 199  *    1) rule1 is duplicated on CPU0 and is not duplicated on CPU1 yet
 200  *    2) rule1 creates state1
 201  *    3) state1 is located on CPU1 by check-state
 202  *    But rule1 is not duplicated on CPU1 yet
 203  * o  During rule deletion, this flag is turned off before deleting states
 204  *    created by the rule and before deleting the rule itself, so no
 205  *    more states will be created by the to-be-deleted rule even when its
 206  *    duplication on certain CPUs are not eliminated yet.
 207  */
 208
 209 #define IPFW_AUTOINC_STEP_MIN   1
 210 #define IPFW_AUTOINC_STEP_MAX   1000
 211 #define IPFW_AUTOINC_STEP_DEF   100
 212
 213 #define IPFW_DEFAULT_RULE       65535   /* rulenum for the default rule */
 214 #define IPFW_DEFAULT_SET        31      /* set number for the default rule */
 215
 216 struct netmsg_ipfw {
 217         struct netmsg   nmsg;
 218         const struct ipfw_ioc_rule *ioc_rule;
 219         struct ip_fw    *next_rule;
 220         struct ip_fw    *prev_rule;
 221         struct ip_fw    *sibling;
 222         struct ip_fw_stub *stub;
 223 };
 224
 225 struct netmsg_del {
 226         struct netmsg   nmsg;
 227         struct ip_fw    *start_rule;
 228         struct ip_fw    *prev_rule;
 229         uint16_t        rulenum;
 230         uint8_t         from_set;
 231         uint8_t         to_set;
 232 };
 233
 234 struct netmsg_zent {
 235         struct netmsg   nmsg;
 236         struct ip_fw    *start_rule;
 237         uint16_t        rulenum;
 238         uint16_t        log_only;
 239 };
 240
 241 struct ipfw_context {
 242         struct ip_fw    *ipfw_layer3_chain;     /* list of rules for layer3 */
 243         struct ip_fw    *ipfw_default_rule;     /* default rule */
 244         uint64_t        ipfw_norule_counter;    /* counter for ipfw_log(NULL) */
 245
 246         /*
 247          * ipfw_set_disable contains one bit per set value (0..31).
 248          * If the bit is set, all rules with the corresponding set
 249          * are disabled.  Set IPDW_DEFAULT_SET is reserved for the
 250          * default rule and CANNOT be disabled.
 251          */
 252         uint32_t        ipfw_set_disable;
 253         uint32_t        ipfw_gen;               /* generation of rule list */
 254 };
 255
 256 static struct ipfw_context      *ipfw_ctx[MAXCPU];
 257
 258 #ifdef KLD_MODULE
 259 /*
 260  * Module can not be unloaded, if there are references to
 261  * certains rules of ipfw(4), e.g. dummynet(4)
 262  */
 263 static int ipfw_refcnt;
 264 #endif
 265
 266 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
 267
 268 /*
 269  * Following two global variables are accessed and
 270  * updated only on CPU0
 271  */
 272 static uint32_t static_count;   /* # of static rules */
 273 static uint32_t static_ioc_len; /* bytes of static rules */
 274
 275 /*
 276  * If 1, then ipfw static rules are being flushed,
 277  * ipfw_chk() will skip to the default rule.
 278  */
 279 static int ipfw_flushing;
 280
 281 static int fw_verbose;
 282 static int verbose_limit;
 283
 284 static int fw_debug = 1;
 285 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
 286
 287 static int      ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS);
 288 static int      ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS);
 289 static int      ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS);
 290 static int      ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS);
 291 static int      ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS);
 292
 293 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
 294 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
 295     &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw");
 296 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
 297     &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
 298     "Rule number autincrement step");
 299 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
 300     &fw_one_pass, 0,
 301     "Only do a single pass through ipfw when using dummynet(4)");
 302 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
 303     &fw_debug, 0, "Enable printing of debug ip_fw statements");
 304 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
 305     &fw_verbose, 0, "Log matches to ipfw rules");
 306 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
 307     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
 308
 309 /*
 310  * Description of dynamic rules.
 311  *
 312  * Dynamic rules are stored in lists accessed through a hash table
 313  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
 314  * be modified through the sysctl variable dyn_buckets which is
 315  * updated when the table becomes empty.
 316  *
 317  * XXX currently there is only one list, ipfw_dyn.
 318  *
 319  * When a packet is received, its address fields are first masked
 320  * with the mask defined for the rule, then hashed, then matched
 321  * against the entries in the corresponding list.
 322  * Dynamic rules can be used for different purposes:
 323  *  + stateful rules;
 324  *  + enforcing limits on the number of sessions;
 325  *  + in-kernel NAT (not implemented yet)
 326  *
 327  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
 328  * measured in seconds and depending on the flags.
 329  *
 330  * The total number of dynamic rules is stored in dyn_count.
 331  * The max number of dynamic rules is dyn_max. When we reach
 332  * the maximum number of rules we do not create anymore. This is
 333  * done to avoid consuming too much memory, but also too much
 334  * time when searching on each packet (ideally, we should try instead
 335  * to put a limit on the length of the list on each bucket...).
 336  *
 337  * Each dynamic rule holds a pointer to the parent ipfw rule so
 338  * we know what action to perform. Dynamic rules are removed when
 339  * the parent rule is deleted. XXX we should make them survive.
 340  *
 341  * There are some limitations with dynamic rules -- we do not
 342  * obey the 'randomized match', and we do not do multiple
 343  * passes through the firewall. XXX check the latter!!!
 344  *
 345  * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
 346  * Only TCP state transition will change dynamic rule's state and ack
 347  * sequences, while all packets of one TCP connection only goes through
 348  * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
 349  * rule looking up.  The keep alive callout uses exclusive lockmgr lock
 350  * when it tries to find suitable dynamic rules to send keep alive, so
 351  * it will not see half updated state and ack sequences.  Though the expire
 352  * field updating looks racy for other protocols, the resolution (second)
 353  * of expire field makes this kind of race harmless.
 354  * XXX statistics' updating is _not_ MPsafe!!!
 355  * XXX once UDP output path is fixed, we could use lockless dynamic rule
 356  *     hash table
 357  */
 358 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
 359 static uint32_t dyn_buckets = 256; /* must be power of 2 */
 360 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
 361 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
 362 static struct lock dyn_lock; /* dynamic rules' hash table lock */
 363 static struct callout ipfw_timeout_h;
 364
 365 /*
 366  * Timeouts for various events in handing dynamic rules.
 367  */
 368 static uint32_t dyn_ack_lifetime = 300;
 369 static uint32_t dyn_syn_lifetime = 20;
 370 static uint32_t dyn_fin_lifetime = 1;
 371 static uint32_t dyn_rst_lifetime = 1;
 372 static uint32_t dyn_udp_lifetime = 10;
 373 static uint32_t dyn_short_lifetime = 5;
 374
 375 /*
 376  * Keepalives are sent if dyn_keepalive is set. They are sent every
 377  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
 378  * seconds of lifetime of a rule.
 379  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
 380  * than dyn_keepalive_period.
 381  */
 382
 383 static uint32_t dyn_keepalive_interval = 20;
 384 static uint32_t dyn_keepalive_period = 5;
 385 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
 386
 387 static uint32_t dyn_count;              /* # of dynamic rules */
 388 static uint32_t dyn_max = 4096;         /* max # of dynamic rules */
 389
 390 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
 391     &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
 392 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
 393     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
 394 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
 395     &dyn_count, 0, "Number of dyn. rules");
 396 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
 397     &dyn_max, 0, "Max number of dyn. rules");
 398 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
 399     &static_count, 0, "Number of static rules");
 400 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
 401     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
 402 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
 403     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
 404 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
 405     CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
 406     "Lifetime of dyn. rules for fin");
 407 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
 408     CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
 409     "Lifetime of dyn. rules for rst");
 410 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
 411     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
 412 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
 413     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
 414 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
 415     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
 416
 417 static ip_fw_chk_t      ipfw_chk;
 418
 419 static __inline int
 420 ipfw_free_rule(struct ip_fw *rule)
 421 {
 422         KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d\n", mycpuid));
 423         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
 424         rule->refcnt--;
 425         if (rule->refcnt == 0) {
 426                 kfree(rule, M_IPFW);
 427                 return 1;
 428         }
 429         return 0;
 430 }
 431
 432 static void
 433 ipfw_unref_rule(void *priv)
 434 {
 435         ipfw_free_rule(priv);
 436 #ifdef KLD_MODULE
 437         atomic_subtract_int(&ipfw_refcnt, 1);
 438 #endif
 439 }
 440
 441 static __inline void
 442 ipfw_ref_rule(struct ip_fw *rule)
 443 {
 444         KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d\n", mycpuid));
 445 #ifdef KLD_MODULE
 446         atomic_add_int(&ipfw_refcnt, 1);
 447 #endif
 448         rule->refcnt++;
 449 }
 450
 451 /*
 452  * This macro maps an ip pointer into a layer3 header pointer of type T
 453  */
 454 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
 455
 456 static __inline int
 457 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
 458 {
 459         int type = L3HDR(struct icmp,ip)->icmp_type;
 460
 461         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
 462 }
 463
 464 #define TT      ((1 << ICMP_ECHO) | \
 465                  (1 << ICMP_ROUTERSOLICIT) | \
 466                  (1 << ICMP_TSTAMP) | \
 467                  (1 << ICMP_IREQ) | \
 468                  (1 << ICMP_MASKREQ))
 469
 470 static int
 471 is_icmp_query(struct ip *ip)
 472 {
 473         int type = L3HDR(struct icmp, ip)->icmp_type;
 474
 475         return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
 476 }
 477
 478 #undef TT
 479
 480 /*
 481  * The following checks use two arrays of 8 or 16 bits to store the
 482  * bits that we want set or clear, respectively. They are in the
 483  * low and high half of cmd->arg1 or cmd->d[0].
 484  *
 485  * We scan options and store the bits we find set. We succeed if
 486  *
 487  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
 488  *
 489  * The code is sometimes optimized not to store additional variables.
 490  */
 491
 492 static int
 493 flags_match(ipfw_insn *cmd, uint8_t bits)
 494 {
 495         u_char want_clear;
 496         bits = ~bits;
 497
 498         if (((cmd->arg1 & 0xff) & bits) != 0)
 499                 return 0; /* some bits we want set were clear */
 500
 501         want_clear = (cmd->arg1 >> 8) & 0xff;
 502         if ((want_clear & bits) != want_clear)
 503                 return 0; /* some bits we want clear were set */
 504         return 1;
 505 }
 506
 507 static int
 508 ipopts_match(struct ip *ip, ipfw_insn *cmd)
 509 {
 510         int optlen, bits = 0;
 511         u_char *cp = (u_char *)(ip + 1);
 512         int x = (ip->ip_hl << 2) - sizeof(struct ip);
 513
 514         for (; x > 0; x -= optlen, cp += optlen) {
 515                 int opt = cp[IPOPT_OPTVAL];
 516
 517                 if (opt == IPOPT_EOL)
 518                         break;
 519
 520                 if (opt == IPOPT_NOP) {
 521                         optlen = 1;
 522                 } else {
 523                         optlen = cp[IPOPT_OLEN];
 524                         if (optlen <= 0 || optlen > x)
 525                                 return 0; /* invalid or truncated */
 526                 }
 527
 528                 switch (opt) {
 529                 case IPOPT_LSRR:
 530                         bits |= IP_FW_IPOPT_LSRR;
 531                         break;
 532
 533                 case IPOPT_SSRR:
 534                         bits |= IP_FW_IPOPT_SSRR;
 535                         break;
 536
 537                 case IPOPT_RR:
 538                         bits |= IP_FW_IPOPT_RR;
 539                         break;
 540
 541                 case IPOPT_TS:
 542                         bits |= IP_FW_IPOPT_TS;
 543                         break;
 544
 545                 default:
 546                         break;
 547                 }
 548         }
 549         return (flags_match(cmd, bits));
 550 }
 551
 552 static int
 553 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
 554 {
 555         int optlen, bits = 0;
 556         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
 557         u_char *cp = (u_char *)(tcp + 1);
 558         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
 559
 560         for (; x > 0; x -= optlen, cp += optlen) {
 561                 int opt = cp[0];
 562
 563                 if (opt == TCPOPT_EOL)
 564                         break;
 565
 566                 if (opt == TCPOPT_NOP) {
 567                         optlen = 1;
 568                 } else {
 569                         optlen = cp[1];
 570                         if (optlen <= 0)
 571                                 break;
 572                 }
 573
 574                 switch (opt) {
 575                 case TCPOPT_MAXSEG:
 576                         bits |= IP_FW_TCPOPT_MSS;
 577                         break;
 578
 579                 case TCPOPT_WINDOW:
 580                         bits |= IP_FW_TCPOPT_WINDOW;
 581                         break;
 582
 583                 case TCPOPT_SACK_PERMITTED:
 584                 case TCPOPT_SACK:
 585                         bits |= IP_FW_TCPOPT_SACK;
 586                         break;
 587
 588                 case TCPOPT_TIMESTAMP:
 589                         bits |= IP_FW_TCPOPT_TS;
 590                         break;
 591
 592                 case TCPOPT_CC:
 593                 case TCPOPT_CCNEW:
 594                 case TCPOPT_CCECHO:
 595                         bits |= IP_FW_TCPOPT_CC;
 596                         break;
 597
 598                 default:
 599                         break;
 600                 }
 601         }
 602         return (flags_match(cmd, bits));
 603 }
 604
 605 static int
 606 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
 607 {
 608         if (ifp == NULL)        /* no iface with this packet, match fails */
 609                 return 0;
 610
 611         /* Check by name or by IP address */
 612         if (cmd->name[0] != '\0') { /* match by name */
 613                 /* Check name */
 614                 if (cmd->p.glob) {
 615                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
 616                                 return(1);
 617                 } else {
 618                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
 619                                 return(1);
 620                 }
 621         } else {
 622                 struct ifaddr_container *ifac;
 623
 624                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 625                         struct ifaddr *ia = ifac->ifa;
 626
 627                         if (ia->ifa_addr == NULL)
 628                                 continue;
 629                         if (ia->ifa_addr->sa_family != AF_INET)
 630                                 continue;
 631                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
 632                             (ia->ifa_addr))->sin_addr.s_addr)
 633                                 return(1);      /* match */
 634                 }
 635         }
 636         return(0);      /* no match, fail ... */
 637 }
 638
 639 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
 640
 641 /*
 642  * We enter here when we have a rule with O_LOG.
 643  * XXX this function alone takes about 2Kbytes of code!
 644  */
 645 static void
 646 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
 647          struct mbuf *m, struct ifnet *oif)
 648 {
 649         char *action;
 650         int limit_reached = 0;
 651         char action2[40], proto[48], fragment[28];
 652
 653         fragment[0] = '\0';
 654         proto[0] = '\0';
 655
 656         if (f == NULL) {        /* bogus pkt */
 657                 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
 658
 659                 if (verbose_limit != 0 &&
 660                     ctx->ipfw_norule_counter >= verbose_limit)
 661                         return;
 662                 ctx->ipfw_norule_counter++;
 663                 if (ctx->ipfw_norule_counter == verbose_limit)
 664                         limit_reached = verbose_limit;
 665                 action = "Refuse";
 666         } else {        /* O_LOG is the first action, find the real one */
 667                 ipfw_insn *cmd = ACTION_PTR(f);
 668                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
 669
 670                 if (l->max_log != 0 && l->log_left == 0)
 671                         return;
 672                 l->log_left--;
 673                 if (l->log_left == 0)
 674                         limit_reached = l->max_log;
 675                 cmd += F_LEN(cmd);      /* point to first action */
 676                 if (cmd->opcode == O_PROB)
 677                         cmd += F_LEN(cmd);
 678
 679                 action = action2;
 680                 switch (cmd->opcode) {
 681                 case O_DENY:
 682                         action = "Deny";
 683                         break;
 684
 685                 case O_REJECT:
 686                         if (cmd->arg1==ICMP_REJECT_RST) {
 687                                 action = "Reset";
 688                         } else if (cmd->arg1==ICMP_UNREACH_HOST) {
 689                                 action = "Reject";
 690                         } else {
 691                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
 692                                           cmd->arg1);
 693                         }
 694                         break;
 695
 696                 case O_ACCEPT:
 697                         action = "Accept";
 698                         break;
 699
 700                 case O_COUNT:
 701                         action = "Count";
 702                         break;
 703
 704                 case O_DIVERT:
 705                         ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
 706                         break;
 707
 708                 case O_TEE:
 709                         ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
 710                         break;
 711
 712                 case O_SKIPTO:
 713                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
 714                         break;
 715
 716                 case O_PIPE:
 717                         ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
 718                         break;
 719
 720                 case O_QUEUE:
 721                         ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
 722                         break;
 723
 724                 case O_FORWARD_IP:
 725                         {
 726                                 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
 727                                 int len;
 728
 729                                 len = ksnprintf(SNPARGS(action2, 0),
 730                                                 "Forward to %s",
 731                                                 inet_ntoa(sa->sa.sin_addr));
 732                                 if (sa->sa.sin_port) {
 733                                         ksnprintf(SNPARGS(action2, len), ":%d",
 734                                                   sa->sa.sin_port);
 735                                 }
 736                         }
 737                         break;
 738
 739                 default:
 740                         action = "UNKNOWN";
 741                         break;
 742                 }
 743         }
 744
 745         if (hlen == 0) {        /* non-ip */
 746                 ksnprintf(SNPARGS(proto, 0), "MAC");
 747         } else {
 748                 struct ip *ip = mtod(m, struct ip *);
 749                 /* these three are all aliases to the same thing */
 750                 struct icmp *const icmp = L3HDR(struct icmp, ip);
 751                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
 752                 struct udphdr *const udp = (struct udphdr *)icmp;
 753
 754                 int ip_off, offset, ip_len;
 755                 int len;
 756
 757                 if (eh != NULL) { /* layer 2 packets are as on the wire */
 758                         ip_off = ntohs(ip->ip_off);
 759                         ip_len = ntohs(ip->ip_len);
 760                 } else {
 761                         ip_off = ip->ip_off;
 762                         ip_len = ip->ip_len;
 763                 }
 764                 offset = ip_off & IP_OFFMASK;
 765                 switch (ip->ip_p) {
 766                 case IPPROTO_TCP:
 767                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
 768                                         inet_ntoa(ip->ip_src));
 769                         if (offset == 0) {
 770                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 771                                           ntohs(tcp->th_sport),
 772                                           inet_ntoa(ip->ip_dst),
 773                                           ntohs(tcp->th_dport));
 774                         } else {
 775                                 ksnprintf(SNPARGS(proto, len), " %s",
 776                                           inet_ntoa(ip->ip_dst));
 777                         }
 778                         break;
 779
 780                 case IPPROTO_UDP:
 781                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
 782                                         inet_ntoa(ip->ip_src));
 783                         if (offset == 0) {
 784                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
 785                                           ntohs(udp->uh_sport),
 786                                           inet_ntoa(ip->ip_dst),
 787                                           ntohs(udp->uh_dport));
 788                         } else {
 789                                 ksnprintf(SNPARGS(proto, len), " %s",
 790                                           inet_ntoa(ip->ip_dst));
 791                         }
 792                         break;
 793
 794                 case IPPROTO_ICMP:
 795                         if (offset == 0) {
 796                                 len = ksnprintf(SNPARGS(proto, 0),
 797                                                 "ICMP:%u.%u ",
 798                                                 icmp->icmp_type,
 799                                                 icmp->icmp_code);
 800                         } else {
 801                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
 802                         }
 803                         len += ksnprintf(SNPARGS(proto, len), "%s",
 804                                          inet_ntoa(ip->ip_src));
 805                         ksnprintf(SNPARGS(proto, len), " %s",
 806                                   inet_ntoa(ip->ip_dst));
 807                         break;
 808
 809                 default:
 810                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
 811                                         inet_ntoa(ip->ip_src));
 812                         ksnprintf(SNPARGS(proto, len), " %s",
 813                                   inet_ntoa(ip->ip_dst));
 814                         break;
 815                 }
 816
 817                 if (ip_off & (IP_MF | IP_OFFMASK)) {
 818                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
 819                                   ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
 820                                   offset << 3, (ip_off & IP_MF) ? "+" : "");
 821                 }
 822         }
 823
 824         if (oif || m->m_pkthdr.rcvif) {
 825                 log(LOG_SECURITY | LOG_INFO,
 826                     "ipfw: %d %s %s %s via %s%s\n",
 827                     f ? f->rulenum : -1,
 828                     action, proto, oif ? "out" : "in",
 829                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
 830                     fragment);
 831         } else {
 832                 log(LOG_SECURITY | LOG_INFO,
 833                     "ipfw: %d %s %s [no if info]%s\n",
 834                     f ? f->rulenum : -1,
 835                     action, proto, fragment);
 836         }
 837
 838         if (limit_reached) {
 839                 log(LOG_SECURITY | LOG_NOTICE,
 840                     "ipfw: limit %d reached on entry %d\n",
 841                     limit_reached, f ? f->rulenum : -1);
 842         }
 843 }
 844
 845 #undef SNPARGS
 846
 847 /*
 848  * IMPORTANT: the hash function for dynamic rules must be commutative
 849  * in source and destination (ip,port), because rules are bidirectional
 850  * and we want to find both in the same bucket.
 851  */
 852 static __inline int
 853 hash_packet(struct ipfw_flow_id *id)
 854 {
 855         uint32_t i;
 856
 857         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
 858         i &= (curr_dyn_buckets - 1);
 859         return i;
 860 }
 861
 862 /**
 863  * unlink a dynamic rule from a chain. prev is a pointer to
 864  * the previous one, q is a pointer to the rule to delete,
 865  * head is a pointer to the head of the queue.
 866  * Modifies q and potentially also head.
 867  */
 868 #define UNLINK_DYN_RULE(prev, head, q)                                  \
 869 do {                                                                    \
 870         ipfw_dyn_rule *old_q = q;                                       \
 871                                                                         \
 872         /* remove a refcount to the parent */                           \
 873         if (q->dyn_type == O_LIMIT)                                     \
 874                 q->parent->count--;                                     \
 875         DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \
 876                 (q->id.src_ip), (q->id.src_port),                       \
 877                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
 878         if (prev != NULL)                                               \
 879                 prev->next = q = q->next;                               \
 880         else                                                            \
 881                 head = q = q->next;                                     \
 882         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
 883         dyn_count--;                                                    \
 884         kfree(old_q, M_IPFW);                                           \
 885 } while (0)
 886
 887 #define TIME_LEQ(a, b)  ((int)((a) - (b)) <= 0)
 888
 889 /**
 890  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
 891  *
 892  * If keep_me == NULL, rules are deleted even if not expired,
 893  * otherwise only expired rules are removed.
 894  *
 895  * The value of the second parameter is also used to point to identify
 896  * a rule we absolutely do not want to remove (e.g. because we are
 897  * holding a reference to it -- this is the case with O_LIMIT_PARENT
 898  * rules). The pointer is only used for comparison, so any non-null
 899  * value will do.
 900  */
 901 static void
 902 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
 903 {
 904         static uint32_t last_remove = 0; /* XXX */
 905
 906 #define FORCE   (keep_me == NULL)
 907
 908         ipfw_dyn_rule *prev, *q;
 909         int i, pass = 0, max_pass = 0, unlinked = 0;
 910
 911         if (ipfw_dyn_v == NULL || dyn_count == 0)
 912                 return;
 913         /* do not expire more than once per second, it is useless */
 914         if (!FORCE && last_remove == time_second)
 915                 return;
 916         last_remove = time_second;
 917
 918         /*
 919          * because O_LIMIT refer to parent rules, during the first pass only
 920          * remove child and mark any pending LIMIT_PARENT, and remove
 921          * them in a second pass.
 922          */
 923 next_pass:
 924         for (i = 0; i < curr_dyn_buckets; i++) {
 925                 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
 926                         /*
 927                          * Logic can become complex here, so we split tests.
 928                          */
 929                         if (q == keep_me)
 930                                 goto next;
 931                         if (rule != NULL && rule->stub != q->stub)
 932                                 goto next; /* not the one we are looking for */
 933                         if (q->dyn_type == O_LIMIT_PARENT) {
 934                                 /*
 935                                  * handle parent in the second pass,
 936                                  * record we need one.
 937                                  */
 938                                 max_pass = 1;
 939                                 if (pass == 0)
 940                                         goto next;
 941                                 if (FORCE && q->count != 0) {
 942                                         /* XXX should not happen! */
 943                                         kprintf("OUCH! cannot remove rule, "
 944                                                 "count %d\n", q->count);
 945                                 }
 946                         } else {
 947                                 if (!FORCE && !TIME_LEQ(q->expire, time_second))
 948                                         goto next;
 949                         }
 950                         unlinked = 1;
 951                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
 952                         continue;
 953 next:
 954                         prev = q;
 955                         q = q->next;
 956                 }
 957         }
 958         if (pass++ < max_pass)
 959                 goto next_pass;
 960
 961         if (unlinked)
 962                 ++dyn_buckets_gen;
 963
 964 #undef FORCE
 965 }
 966
 967 /**
 968  * lookup a dynamic rule.
 969  */
 970 static ipfw_dyn_rule *
 971 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
 972                 struct tcphdr *tcp)
 973 {
 974         /*
 975          * stateful ipfw extensions.
 976          * Lookup into dynamic session queue
 977          */
 978 #define MATCH_REVERSE   0
 979 #define MATCH_FORWARD   1
 980 #define MATCH_NONE      2
 981 #define MATCH_UNKNOWN   3
 982         int i, dir = MATCH_NONE;
 983         ipfw_dyn_rule *prev, *q=NULL;
 984
 985         if (ipfw_dyn_v == NULL)
 986                 goto done;      /* not found */
 987
 988         i = hash_packet(pkt);
 989         for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
 990                 if (q->dyn_type == O_LIMIT_PARENT)
 991                         goto next;
 992
 993                 if (TIME_LEQ(q->expire, time_second)) {
 994                         /*
 995                          * Entry expired; skip.
 996                          * Let ipfw_tick() take care of it
 997                          */
 998                         goto next;
 999                 }
1000
1001                 if (pkt->proto == q->id.proto) {
1002                         if (pkt->src_ip == q->id.src_ip &&
1003                             pkt->dst_ip == q->id.dst_ip &&
1004                             pkt->src_port == q->id.src_port &&
1005                             pkt->dst_port == q->id.dst_port) {
1006                                 dir = MATCH_FORWARD;
1007                                 break;
1008                         }
1009                         if (pkt->src_ip == q->id.dst_ip &&
1010                             pkt->dst_ip == q->id.src_ip &&
1011                             pkt->src_port == q->id.dst_port &&
1012                             pkt->dst_port == q->id.src_port) {
1013                                 dir = MATCH_REVERSE;
1014                                 break;
1015                         }
1016                 }
1017 next:
1018                 prev = q;
1019                 q = q->next;
1020         }
1021         if (q == NULL)
1022                 goto done; /* q = NULL, not found */
1023
1024         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1025                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1026
1027 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
1028 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
1029
1030                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1031                 switch (q->state) {
1032                 case TH_SYN:                            /* opening */
1033                         q->expire = time_second + dyn_syn_lifetime;
1034                         break;
1035
1036                 case BOTH_SYN:                  /* move to established */
1037                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
1038                 case BOTH_SYN | (TH_FIN << 8) :
1039                         if (tcp) {
1040                                 uint32_t ack = ntohl(tcp->th_ack);
1041
1042 #define _SEQ_GE(a, b)   ((int)(a) - (int)(b) >= 0)
1043
1044                                 if (dir == MATCH_FORWARD) {
1045                                         if (q->ack_fwd == 0 ||
1046                                             _SEQ_GE(ack, q->ack_fwd))
1047                                                 q->ack_fwd = ack;
1048                                         else /* ignore out-of-sequence */
1049                                                 break;
1050                                 } else {
1051                                         if (q->ack_rev == 0 ||
1052                                             _SEQ_GE(ack, q->ack_rev))
1053                                                 q->ack_rev = ack;
1054                                         else /* ignore out-of-sequence */
1055                                                 break;
1056                                 }
1057 #undef _SEQ_GE
1058                         }
1059                         q->expire = time_second + dyn_ack_lifetime;
1060                         break;
1061
1062                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
1063                         KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
1064                         q->expire = time_second + dyn_fin_lifetime;
1065                         break;
1066
1067                 default:
1068 #if 0
1069                         /*
1070                          * reset or some invalid combination, but can also
1071                          * occur if we use keep-state the wrong way.
1072                          */
1073                         if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
1074                                 kprintf("invalid state: 0x%x\n", q->state);
1075 #endif
1076                         KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
1077                         q->expire = time_second + dyn_rst_lifetime;
1078                         break;
1079                 }
1080         } else if (pkt->proto == IPPROTO_UDP) {
1081                 q->expire = time_second + dyn_udp_lifetime;
1082         } else {
1083                 /* other protocols */
1084                 q->expire = time_second + dyn_short_lifetime;
1085         }
1086 done:
1087         if (match_direction)
1088                 *match_direction = dir;
1089         return q;
1090 }
1091
1092 static struct ip_fw *
1093 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
1094             uint16_t len, int *deny)
1095 {
1096         struct ip_fw *rule = NULL;
1097         ipfw_dyn_rule *q;
1098         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1099         uint32_t gen;
1100
1101         *deny = 0;
1102         gen = ctx->ipfw_gen;
1103
1104         lockmgr(&dyn_lock, LK_SHARED);
1105
1106         if (ctx->ipfw_gen != gen) {
1107                 /*
1108                  * Static rules had been change when we were waiting
1109                  * for the dynamic hash table lock; deny this packet,
1110                  * since it is _not_ known whether it is safe to keep
1111                  * iterating the static rules.
1112                  */
1113                 *deny = 1;
1114                 goto back;
1115         }
1116
1117         q = lookup_dyn_rule(pkt, match_direction, tcp);
1118         if (q == NULL) {
1119                 rule = NULL;
1120         } else {
1121                 rule = q->stub->rule[mycpuid];
1122                 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid);
1123
1124                 /* XXX */
1125                 q->pcnt++;
1126                 q->bcnt += len;
1127         }
1128 back:
1129         lockmgr(&dyn_lock, LK_RELEASE);
1130         return rule;
1131 }
1132
1133 static void
1134 realloc_dynamic_table(void)
1135 {
1136         ipfw_dyn_rule **old_dyn_v;
1137         uint32_t old_curr_dyn_buckets;
1138
1139         KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
1140                 ("invalid dyn_buckets %d\n", dyn_buckets));
1141
1142         /* Save the current buckets array for later error recovery */
1143         old_dyn_v = ipfw_dyn_v;
1144         old_curr_dyn_buckets = curr_dyn_buckets;
1145
1146         curr_dyn_buckets = dyn_buckets;
1147         for (;;) {
1148                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1149                                      M_IPFW, M_NOWAIT | M_ZERO);
1150                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1151                         break;
1152
1153                 curr_dyn_buckets /= 2;
1154                 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
1155                     old_dyn_v != NULL) {
1156                         /*
1157                          * Don't try allocating smaller buckets array, reuse
1158                          * the old one, which alreay contains enough buckets
1159                          */
1160                         break;
1161                 }
1162         }
1163
1164         if (ipfw_dyn_v != NULL) {
1165                 if (old_dyn_v != NULL)
1166                         kfree(old_dyn_v, M_IPFW);
1167         } else {
1168                 /* Allocation failed, restore old buckets array */
1169                 ipfw_dyn_v = old_dyn_v;
1170                 curr_dyn_buckets = old_curr_dyn_buckets;
1171         }
1172
1173         if (ipfw_dyn_v != NULL)
1174                 ++dyn_buckets_gen;
1175 }
1176
1177 /**
1178  * Install state of type 'type' for a dynamic session.
1179  * The hash table contains two type of rules:
1180  * - regular rules (O_KEEP_STATE)
1181  * - rules for sessions with limited number of sess per user
1182  *   (O_LIMIT). When they are created, the parent is
1183  *   increased by 1, and decreased on delete. In this case,
1184  *   the third parameter is the parent rule and not the chain.
1185  * - "parent" rules for the above (O_LIMIT_PARENT).
1186  */
1187 static ipfw_dyn_rule *
1188 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1189 {
1190         ipfw_dyn_rule *r;
1191         int i;
1192
1193         if (ipfw_dyn_v == NULL ||
1194             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1195                 realloc_dynamic_table();
1196                 if (ipfw_dyn_v == NULL)
1197                         return NULL; /* failed ! */
1198         }
1199         i = hash_packet(id);
1200
1201         r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1202         if (r == NULL) {
1203                 kprintf ("sorry cannot allocate state\n");
1204                 return NULL;
1205         }
1206
1207         /* increase refcount on parent, and set pointer */
1208         if (dyn_type == O_LIMIT) {
1209                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1210
1211                 if (parent->dyn_type != O_LIMIT_PARENT)
1212                         panic("invalid parent");
1213                 parent->count++;
1214                 r->parent = parent;
1215                 rule = parent->stub->rule[mycpuid];
1216                 KKASSERT(rule->stub == parent->stub);
1217         }
1218         KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1219
1220         r->id = *id;
1221         r->expire = time_second + dyn_syn_lifetime;
1222         r->stub = rule->stub;
1223         r->dyn_type = dyn_type;
1224         r->pcnt = r->bcnt = 0;
1225         r->count = 0;
1226
1227         r->bucket = i;
1228         r->next = ipfw_dyn_v[i];
1229         ipfw_dyn_v[i] = r;
1230         dyn_count++;
1231         dyn_buckets_gen++;
1232         DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1233            dyn_type,
1234            (r->id.src_ip), (r->id.src_port),
1235            (r->id.dst_ip), (r->id.dst_port),
1236            dyn_count );)
1237         return r;
1238 }
1239
1240 /**
1241  * lookup dynamic parent rule using pkt and rule as search keys.
1242  * If the lookup fails, then install one.
1243  */
1244 static ipfw_dyn_rule *
1245 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1246 {
1247         ipfw_dyn_rule *q;
1248         int i;
1249
1250         if (ipfw_dyn_v) {
1251                 i = hash_packet(pkt);
1252                 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1253                         if (q->dyn_type == O_LIMIT_PARENT &&
1254                             rule->stub == q->stub &&
1255                             pkt->proto == q->id.proto &&
1256                             pkt->src_ip == q->id.src_ip &&
1257                             pkt->dst_ip == q->id.dst_ip &&
1258                             pkt->src_port == q->id.src_port &&
1259                             pkt->dst_port == q->id.dst_port) {
1260                                 q->expire = time_second + dyn_short_lifetime;
1261                                 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
1262                                 return q;
1263                         }
1264                 }
1265         }
1266         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1267 }
1268
1269 /**
1270  * Install dynamic state for rule type cmd->o.opcode
1271  *
1272  * Returns 1 (failure) if state is not installed because of errors or because
1273  * session limitations are enforced.
1274  */
1275 static int
1276 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1277                      struct ip_fw_args *args)
1278 {
1279         static int last_log; /* XXX */
1280
1281         ipfw_dyn_rule *q;
1282
1283         DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1284             cmd->o.opcode,
1285             (args->f_id.src_ip), (args->f_id.src_port),
1286             (args->f_id.dst_ip), (args->f_id.dst_port) );)
1287
1288         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1289         if (q != NULL) { /* should never occur */
1290                 if (last_log != time_second) {
1291                         last_log = time_second;
1292                         kprintf(" install_state: entry already present, done\n");
1293                 }
1294                 return 0;
1295         }
1296
1297         if (dyn_count >= dyn_max) {
1298                 /*
1299                  * Run out of slots, try to remove any expired rule.
1300                  */
1301                 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1302                 if (dyn_count >= dyn_max) {
1303                         if (last_log != time_second) {
1304                                 last_log = time_second;
1305                                 kprintf("install_state: "
1306                                         "Too many dynamic rules\n");
1307                         }
1308                         return 1; /* cannot install, notify caller */
1309                 }
1310         }
1311
1312         switch (cmd->o.opcode) {
1313         case O_KEEP_STATE: /* bidir rule */
1314                 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1315                         return 1;
1316                 break;
1317
1318         case O_LIMIT: /* limit number of sessions */
1319                 {
1320                         uint16_t limit_mask = cmd->limit_mask;
1321                         struct ipfw_flow_id id;
1322                         ipfw_dyn_rule *parent;
1323
1324                         DEB(kprintf("installing dyn-limit rule %d\n",
1325                             cmd->conn_limit);)
1326
1327                         id.dst_ip = id.src_ip = 0;
1328                         id.dst_port = id.src_port = 0;
1329                         id.proto = args->f_id.proto;
1330
1331                         if (limit_mask & DYN_SRC_ADDR)
1332                                 id.src_ip = args->f_id.src_ip;
1333                         if (limit_mask & DYN_DST_ADDR)
1334                                 id.dst_ip = args->f_id.dst_ip;
1335                         if (limit_mask & DYN_SRC_PORT)
1336                                 id.src_port = args->f_id.src_port;
1337                         if (limit_mask & DYN_DST_PORT)
1338                                 id.dst_port = args->f_id.dst_port;
1339
1340                         parent = lookup_dyn_parent(&id, rule);
1341                         if (parent == NULL) {
1342                                 kprintf("add parent failed\n");
1343                                 return 1;
1344                         }
1345
1346                         if (parent->count >= cmd->conn_limit) {
1347                                 /*
1348                                  * See if we can remove some expired rule.
1349                                  */
1350                                 remove_dyn_rule_locked(rule, parent);
1351                                 if (parent->count >= cmd->conn_limit) {
1352                                         if (fw_verbose &&
1353                                             last_log != time_second) {
1354                                                 last_log = time_second;
1355                                                 log(LOG_SECURITY | LOG_DEBUG,
1356                                                     "drop session, "
1357                                                     "too many entries\n");
1358                                         }
1359                                         return 1;
1360                                 }
1361                         }
1362                         if (add_dyn_rule(&args->f_id, O_LIMIT,
1363                                          (struct ip_fw *)parent) == NULL)
1364                                 return 1;
1365                 }
1366                 break;
1367         default:
1368                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1369                 return 1;
1370         }
1371         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1372         return 0;
1373 }
1374
1375 static int
1376 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1377               struct ip_fw_args *args, int *deny)
1378 {
1379         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1380         uint32_t gen;
1381         int ret = 0;
1382
1383         *deny = 0;
1384         gen = ctx->ipfw_gen;
1385
1386         lockmgr(&dyn_lock, LK_EXCLUSIVE);
1387         if (ctx->ipfw_gen != gen) {
1388                 /* See the comment in lookup_rule() */
1389                 *deny = 1;
1390         } else {
1391                 ret = install_state_locked(rule, cmd, args);
1392         }
1393         lockmgr(&dyn_lock, LK_RELEASE);
1394
1395         return ret;
1396 }
1397
1398 /*
1399  * Transmit a TCP packet, containing either a RST or a keepalive.
1400  * When flags & TH_RST, we are sending a RST packet, because of a
1401  * "reset" action matched the packet.
1402  * Otherwise we are sending a keepalive, and flags & TH_
1403  */
1404 static void
1405 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1406 {
1407         struct mbuf *m;
1408         struct ip *ip;
1409         struct tcphdr *tcp;
1410         struct route sro;       /* fake route */
1411
1412         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1413         if (m == NULL)
1414                 return;
1415         m->m_pkthdr.rcvif = NULL;
1416         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1417         m->m_data += max_linkhdr;
1418
1419         ip = mtod(m, struct ip *);
1420         bzero(ip, m->m_len);
1421         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1422         ip->ip_p = IPPROTO_TCP;
1423         tcp->th_off = 5;
1424
1425         /*
1426          * Assume we are sending a RST (or a keepalive in the reverse
1427          * direction), swap src and destination addresses and ports.
1428          */
1429         ip->ip_src.s_addr = htonl(id->dst_ip);
1430         ip->ip_dst.s_addr = htonl(id->src_ip);
1431         tcp->th_sport = htons(id->dst_port);
1432         tcp->th_dport = htons(id->src_port);
1433         if (flags & TH_RST) {   /* we are sending a RST */
1434                 if (flags & TH_ACK) {
1435                         tcp->th_seq = htonl(ack);
1436                         tcp->th_ack = htonl(0);
1437                         tcp->th_flags = TH_RST;
1438                 } else {
1439                         if (flags & TH_SYN)
1440                                 seq++;
1441                         tcp->th_seq = htonl(0);
1442                         tcp->th_ack = htonl(seq);
1443                         tcp->th_flags = TH_RST | TH_ACK;
1444                 }
1445         } else {
1446                 /*
1447                  * We are sending a keepalive. flags & TH_SYN determines
1448                  * the direction, forward if set, reverse if clear.
1449                  * NOTE: seq and ack are always assumed to be correct
1450                  * as set by the caller. This may be confusing...
1451                  */
1452                 if (flags & TH_SYN) {
1453                         /*
1454                          * we have to rewrite the correct addresses!
1455                          */
1456                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1457                         ip->ip_src.s_addr = htonl(id->src_ip);
1458                         tcp->th_dport = htons(id->dst_port);
1459                         tcp->th_sport = htons(id->src_port);
1460                 }
1461                 tcp->th_seq = htonl(seq);
1462                 tcp->th_ack = htonl(ack);
1463                 tcp->th_flags = TH_ACK;
1464         }
1465
1466         /*
1467          * set ip_len to the payload size so we can compute
1468          * the tcp checksum on the pseudoheader
1469          * XXX check this, could save a couple of words ?
1470          */
1471         ip->ip_len = htons(sizeof(struct tcphdr));
1472         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1473
1474         /*
1475          * now fill fields left out earlier
1476          */
1477         ip->ip_ttl = ip_defttl;
1478         ip->ip_len = m->m_pkthdr.len;
1479
1480         bzero(&sro, sizeof(sro));
1481         ip_rtaddr(ip->ip_dst, &sro);
1482
1483         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1484         ip_output(m, NULL, &sro, 0, NULL, NULL);
1485         if (sro.ro_rt)
1486                 RTFREE(sro.ro_rt);
1487 }
1488
1489 /*
1490  * sends a reject message, consuming the mbuf passed as an argument.
1491  */
1492 static void
1493 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1494 {
1495         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1496                 /* We need the IP header in host order for icmp_error(). */
1497                 if (args->eh != NULL) {
1498                         struct ip *ip = mtod(args->m, struct ip *);
1499
1500                         ip->ip_len = ntohs(ip->ip_len);
1501                         ip->ip_off = ntohs(ip->ip_off);
1502                 }
1503                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1504         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1505                 struct tcphdr *const tcp =
1506                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1507
1508                 if ((tcp->th_flags & TH_RST) == 0) {
1509                         send_pkt(&args->f_id, ntohl(tcp->th_seq),
1510                                  ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1511                 }
1512                 m_freem(args->m);
1513         } else {
1514                 m_freem(args->m);
1515         }
1516         args->m = NULL;
1517 }
1518
1519 /**
1520  *
1521  * Given an ip_fw *, lookup_next_rule will return a pointer
1522  * to the next rule, which can be either the jump
1523  * target (for skipto instructions) or the next one in the list (in
1524  * all other cases including a missing jump target).
1525  * The result is also written in the "next_rule" field of the rule.
1526  * Backward jumps are not allowed, so start looking from the next
1527  * rule...
1528  *
1529  * This never returns NULL -- in case we do not have an exact match,
1530  * the next rule is returned. When the ruleset is changed,
1531  * pointers are flushed so we are always correct.
1532  */
1533
1534 static struct ip_fw *
1535 lookup_next_rule(struct ip_fw *me)
1536 {
1537         struct ip_fw *rule = NULL;
1538         ipfw_insn *cmd;
1539
1540         /* look for action, in case it is a skipto */
1541         cmd = ACTION_PTR(me);
1542         if (cmd->opcode == O_LOG)
1543                 cmd += F_LEN(cmd);
1544         if (cmd->opcode == O_SKIPTO) {
1545                 for (rule = me->next; rule; rule = rule->next) {
1546                         if (rule->rulenum >= cmd->arg1)
1547                                 break;
1548                 }
1549         }
1550         if (rule == NULL)                       /* failure or not a skipto */
1551                 rule = me->next;
1552         me->next_rule = rule;
1553         return rule;
1554 }
1555
1556 /*
1557  * The main check routine for the firewall.
1558  *
1559  * All arguments are in args so we can modify them and return them
1560  * back to the caller.
1561  *
1562  * Parameters:
1563  *
1564  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1565  *              Starts with the IP header.
1566  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1567  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1568  *              The incoming interface is in the mbuf. (in)
1569  *
1570  *      args->rule      Pointer to the last matching rule (in/out)
1571  *      args->f_id      Addresses grabbed from the packet (out)
1572  *
1573  * Return value:
1574  *
1575  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1576  *      0       The packet is to be accepted and routed normally OR
1577  *              the packet was denied/rejected and has been dropped;
1578  *              in the latter case, *m is equal to NULL upon return.
1579  *      port    Divert the packet to port, with these caveats:
1580  *
1581  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1582  *                of diverting it (ie, 'ipfw tee').
1583  *
1584  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1585  *                16 bits as a dummynet pipe number instead of diverting
1586  */
1587
1588 static int
1589 ipfw_chk(struct ip_fw_args *args)
1590 {
1591         /*
1592          * Local variables hold state during the processing of a packet.
1593          *
1594          * IMPORTANT NOTE: to speed up the processing of rules, there
1595          * are some assumption on the values of the variables, which
1596          * are documented here. Should you change them, please check
1597          * the implementation of the various instructions to make sure
1598          * that they still work.
1599          *
1600          * args->eh     The MAC header. It is non-null for a layer2
1601          *      packet, it is NULL for a layer-3 packet.
1602          *
1603          * m | args->m  Pointer to the mbuf, as received from the caller.
1604          *      It may change if ipfw_chk() does an m_pullup, or if it
1605          *      consumes the packet because it calls send_reject().
1606          *      XXX This has to change, so that ipfw_chk() never modifies
1607          *      or consumes the buffer.
1608          * ip   is simply an alias of the value of m, and it is kept
1609          *      in sync with it (the packet is  supposed to start with
1610          *      the ip header).
1611          */
1612         struct mbuf *m = args->m;
1613         struct ip *ip = mtod(m, struct ip *);
1614
1615         /*
1616          * oif | args->oif      If NULL, ipfw_chk has been called on the
1617          *      inbound path (ether_input, ip_input).
1618          *      If non-NULL, ipfw_chk has been called on the outbound path
1619          *      (ether_output, ip_output).
1620          */
1621         struct ifnet *oif = args->oif;
1622
1623         struct ip_fw *f = NULL;         /* matching rule */
1624         int retval = IP_FW_PASS;
1625         struct m_tag *mtag;
1626         struct divert_info *divinfo;
1627
1628         /*
1629          * hlen The length of the IPv4 header.
1630          *      hlen >0 means we have an IPv4 packet.
1631          */
1632         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1633
1634         /*
1635          * offset       The offset of a fragment. offset != 0 means that
1636          *      we have a fragment at this offset of an IPv4 packet.
1637          *      offset == 0 means that (if this is an IPv4 packet)
1638          *      this is the first or only fragment.
1639          */
1640         u_short offset = 0;
1641
1642         /*
1643          * Local copies of addresses. They are only valid if we have
1644          * an IP packet.
1645          *
1646          * proto        The protocol. Set to 0 for non-ip packets,
1647          *      or to the protocol read from the packet otherwise.
1648          *      proto != 0 means that we have an IPv4 packet.
1649          *
1650          * src_port, dst_port   port numbers, in HOST format. Only
1651          *      valid for TCP and UDP packets.
1652          *
1653          * src_ip, dst_ip       ip addresses, in NETWORK format.
1654          *      Only valid for IPv4 packets.
1655          */
1656         uint8_t proto;
1657         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1658         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1659         uint16_t ip_len = 0;
1660
1661         /*
1662          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1663          *      MATCH_NONE when checked and not matched (dyn_f = NULL),
1664          *      MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1665          */
1666         int dyn_dir = MATCH_UNKNOWN;
1667         struct ip_fw *dyn_f = NULL;
1668         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1669
1670         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1671                 return IP_FW_PASS;      /* accept */
1672
1673         if (args->eh == NULL ||         /* layer 3 packet */
1674             (m->m_pkthdr.len >= sizeof(struct ip) &&
1675              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1676                 hlen = ip->ip_hl << 2;
1677
1678         /*
1679          * Collect parameters into local variables for faster matching.
1680          */
1681         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1682                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1683                 goto after_ip_checks;
1684         }
1685
1686         proto = args->f_id.proto = ip->ip_p;
1687         src_ip = ip->ip_src;
1688         dst_ip = ip->ip_dst;
1689         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1690                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1691                 ip_len = ntohs(ip->ip_len);
1692         } else {
1693                 offset = ip->ip_off & IP_OFFMASK;
1694                 ip_len = ip->ip_len;
1695         }
1696
1697 #define PULLUP_TO(len)                          \
1698 do {                                            \
1699         if (m->m_len < (len)) {                 \
1700                 args->m = m = m_pullup(m, (len));\
1701                 if (m == NULL)                  \
1702                         goto pullup_failed;     \
1703                 ip = mtod(m, struct ip *);      \
1704         }                                       \
1705 } while (0)
1706
1707         if (offset == 0) {
1708                 switch (proto) {
1709                 case IPPROTO_TCP:
1710                         {
1711                                 struct tcphdr *tcp;
1712
1713                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1714                                 tcp = L3HDR(struct tcphdr, ip);
1715                                 dst_port = tcp->th_dport;
1716                                 src_port = tcp->th_sport;
1717                                 args->f_id.flags = tcp->th_flags;
1718                         }
1719                         break;
1720
1721                 case IPPROTO_UDP:
1722                         {
1723                                 struct udphdr *udp;
1724
1725                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1726                                 udp = L3HDR(struct udphdr, ip);
1727                                 dst_port = udp->uh_dport;
1728                                 src_port = udp->uh_sport;
1729                         }
1730                         break;
1731
1732                 case IPPROTO_ICMP:
1733                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1734                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1735                         break;
1736
1737                 default:
1738                         break;
1739                 }
1740         }
1741
1742 #undef PULLUP_TO
1743
1744         args->f_id.src_ip = ntohl(src_ip.s_addr);
1745         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1746         args->f_id.src_port = src_port = ntohs(src_port);
1747         args->f_id.dst_port = dst_port = ntohs(dst_port);
1748
1749 after_ip_checks:
1750         if (args->rule) {
1751                 /*
1752                  * Packet has already been tagged. Look for the next rule
1753                  * to restart processing.
1754                  *
1755                  * If fw_one_pass != 0 then just accept it.
1756                  * XXX should not happen here, but optimized out in
1757                  * the caller.
1758                  */
1759                 if (fw_one_pass)
1760                         return IP_FW_PASS;
1761
1762                 /* This rule is being/has been flushed */
1763                 if (ipfw_flushing)
1764                         return IP_FW_DENY;
1765
1766                 KASSERT(args->rule->cpuid == mycpuid,
1767                         ("rule used on cpu%d\n", mycpuid));
1768
1769                 /* This rule was deleted */
1770                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1771                         return IP_FW_DENY;
1772
1773                 f = args->rule->next_rule;
1774                 if (f == NULL)
1775                         f = lookup_next_rule(args->rule);
1776         } else {
1777                 /*
1778                  * Find the starting rule. It can be either the first
1779                  * one, or the one after divert_rule if asked so.
1780                  */
1781                 int skipto;
1782
1783                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1784                 if (mtag != NULL) {
1785                         divinfo = m_tag_data(mtag);
1786                         skipto = divinfo->skipto;
1787                 } else {
1788                         skipto = 0;
1789                 }
1790
1791                 f = ctx->ipfw_layer3_chain;
1792                 if (args->eh == NULL && skipto != 0) {
1793                         /* No skipto during rule flushing */
1794                         if (ipfw_flushing)
1795                                 return IP_FW_DENY;
1796
1797                         if (skipto >= IPFW_DEFAULT_RULE)
1798                                 return IP_FW_DENY; /* invalid */
1799
1800                         while (f && f->rulenum <= skipto)
1801                                 f = f->next;
1802                         if (f == NULL)  /* drop packet */
1803                                 return IP_FW_DENY;
1804                 } else if (ipfw_flushing) {
1805                         /* Rules are being flushed; skip to default rule */
1806                         f = ctx->ipfw_default_rule;
1807                 }
1808         }
1809         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1810                 m_tag_delete(m, mtag);
1811
1812         /*
1813          * Now scan the rules, and parse microinstructions for each rule.
1814          */
1815         for (; f; f = f->next) {
1816                 int l, cmdlen;
1817                 ipfw_insn *cmd;
1818                 int skip_or; /* skip rest of OR block */
1819
1820 again:
1821                 if (ctx->ipfw_set_disable & (1 << f->set))
1822                         continue;
1823
1824                 skip_or = 0;
1825                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1826                      l -= cmdlen, cmd += cmdlen) {
1827                         int match, deny;
1828
1829                         /*
1830                          * check_body is a jump target used when we find a
1831                          * CHECK_STATE, and need to jump to the body of
1832                          * the target rule.
1833                          */
1834
1835 check_body:
1836                         cmdlen = F_LEN(cmd);
1837                         /*
1838                          * An OR block (insn_1 || .. || insn_n) has the
1839                          * F_OR bit set in all but the last instruction.
1840                          * The first match will set "skip_or", and cause
1841                          * the following instructions to be skipped until
1842                          * past the one with the F_OR bit clear.
1843                          */
1844                         if (skip_or) {          /* skip this instruction */
1845                                 if ((cmd->len & F_OR) == 0)
1846                                         skip_or = 0;    /* next one is good */
1847                                 continue;
1848                         }
1849                         match = 0; /* set to 1 if we succeed */
1850
1851                         switch (cmd->opcode) {
1852                         /*
1853                          * The first set of opcodes compares the packet's
1854                          * fields with some pattern, setting 'match' if a
1855                          * match is found. At the end of the loop there is
1856                          * logic to deal with F_NOT and F_OR flags associated
1857                          * with the opcode.
1858                          */
1859                         case O_NOP:
1860                                 match = 1;
1861                                 break;
1862
1863                         case O_FORWARD_MAC:
1864                                 kprintf("ipfw: opcode %d unimplemented\n",
1865                                         cmd->opcode);
1866                                 break;
1867
1868                         case O_GID:
1869                         case O_UID:
1870                                 /*
1871                                  * We only check offset == 0 && proto != 0,
1872                                  * as this ensures that we have an IPv4
1873                                  * packet with the ports info.
1874                                  */
1875                                 if (offset!=0)
1876                                         break;
1877                             {
1878                                 struct inpcbinfo *pi;
1879                                 int wildcard;
1880                                 struct inpcb *pcb;
1881
1882                                 if (proto == IPPROTO_TCP) {
1883                                         wildcard = 0;
1884                                         pi = &tcbinfo[mycpu->gd_cpuid];
1885                                 } else if (proto == IPPROTO_UDP) {
1886                                         wildcard = 1;
1887                                         pi = &udbinfo;
1888                                 } else
1889                                         break;
1890
1891                                 pcb =  (oif) ?
1892                                         in_pcblookup_hash(pi,
1893                                             dst_ip, htons(dst_port),
1894                                             src_ip, htons(src_port),
1895                                             wildcard, oif) :
1896                                         in_pcblookup_hash(pi,
1897                                             src_ip, htons(src_port),
1898                                             dst_ip, htons(dst_port),
1899                                             wildcard, NULL);
1900
1901                                 if (pcb == NULL || pcb->inp_socket == NULL)
1902                                         break;
1903
1904                                 if (cmd->opcode == O_UID) {
1905 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1906                                         match =
1907                                           !socheckuid(pcb->inp_socket,
1908                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1909 #undef socheckuid
1910                                 } else  {
1911                                         match = groupmember(
1912                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1913                                             pcb->inp_socket->so_cred);
1914                                 }
1915                             }
1916                                 break;
1917
1918                         case O_RECV:
1919                                 match = iface_match(m->m_pkthdr.rcvif,
1920                                     (ipfw_insn_if *)cmd);
1921                                 break;
1922
1923                         case O_XMIT:
1924                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1925                                 break;
1926
1927                         case O_VIA:
1928                                 match = iface_match(oif ? oif :
1929                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1930                                 break;
1931
1932                         case O_MACADDR2:
1933                                 if (args->eh != NULL) { /* have MAC header */
1934                                         uint32_t *want = (uint32_t *)
1935                                                 ((ipfw_insn_mac *)cmd)->addr;
1936                                         uint32_t *mask = (uint32_t *)
1937                                                 ((ipfw_insn_mac *)cmd)->mask;
1938                                         uint32_t *hdr = (uint32_t *)args->eh;
1939
1940                                         match =
1941                                         (want[0] == (hdr[0] & mask[0]) &&
1942                                          want[1] == (hdr[1] & mask[1]) &&
1943                                          want[2] == (hdr[2] & mask[2]));
1944                                 }
1945                                 break;
1946
1947                         case O_MAC_TYPE:
1948                                 if (args->eh != NULL) {
1949                                         uint16_t t =
1950                                             ntohs(args->eh->ether_type);
1951                                         uint16_t *p =
1952                                             ((ipfw_insn_u16 *)cmd)->ports;
1953                                         int i;
1954
1955                                         /* Special vlan handling */
1956                                         if (m->m_flags & M_VLANTAG)
1957                                                 t = ETHERTYPE_VLAN;
1958
1959                                         for (i = cmdlen - 1; !match && i > 0;
1960                                              i--, p += 2) {
1961                                                 match =
1962                                                 (t >= p[0] && t <= p[1]);
1963                                         }
1964                                 }
1965                                 break;
1966
1967                         case O_FRAG:
1968                                 match = (hlen > 0 && offset != 0);
1969                                 break;
1970
1971                         case O_IN:      /* "out" is "not in" */
1972                                 match = (oif == NULL);
1973                                 break;
1974
1975                         case O_LAYER2:
1976                                 match = (args->eh != NULL);
1977                                 break;
1978
1979                         case O_PROTO:
1980                                 /*
1981                                  * We do not allow an arg of 0 so the
1982                                  * check of "proto" only suffices.
1983                                  */
1984                                 match = (proto == cmd->arg1);
1985                                 break;
1986
1987                         case O_IP_SRC:
1988                                 match = (hlen > 0 &&
1989                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1990                                     src_ip.s_addr);
1991                                 break;
1992
1993                         case O_IP_SRC_MASK:
1994                                 match = (hlen > 0 &&
1995                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1996                                      (src_ip.s_addr &
1997                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1998                                 break;
1999
2000                         case O_IP_SRC_ME:
2001                                 if (hlen > 0) {
2002                                         struct ifnet *tif;
2003
2004                                         tif = INADDR_TO_IFP(&src_ip);
2005                                         match = (tif != NULL);
2006                                 }
2007                                 break;
2008
2009                         case O_IP_DST_SET:
2010                         case O_IP_SRC_SET:
2011                                 if (hlen > 0) {
2012                                         uint32_t *d = (uint32_t *)(cmd + 1);
2013                                         uint32_t addr =
2014                                             cmd->opcode == O_IP_DST_SET ?
2015                                                 args->f_id.dst_ip :
2016                                                 args->f_id.src_ip;
2017
2018                                         if (addr < d[0])
2019                                                 break;
2020                                         addr -= d[0]; /* subtract base */
2021                                         match =
2022                                         (addr < cmd->arg1) &&
2023                                          (d[1 + (addr >> 5)] &
2024                                           (1 << (addr & 0x1f)));
2025                                 }
2026                                 break;
2027
2028                         case O_IP_DST:
2029                                 match = (hlen > 0 &&
2030                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2031                                     dst_ip.s_addr);
2032                                 break;
2033
2034                         case O_IP_DST_MASK:
2035                                 match = (hlen > 0) &&
2036                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2037                                      (dst_ip.s_addr &
2038                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2039                                 break;
2040
2041                         case O_IP_DST_ME:
2042                                 if (hlen > 0) {
2043                                         struct ifnet *tif;
2044
2045                                         tif = INADDR_TO_IFP(&dst_ip);
2046                                         match = (tif != NULL);
2047                                 }
2048                                 break;
2049
2050                         case O_IP_SRCPORT:
2051                         case O_IP_DSTPORT:
2052                                 /*
2053                                  * offset == 0 && proto != 0 is enough
2054                                  * to guarantee that we have an IPv4
2055                                  * packet with port info.
2056                                  */
2057                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2058                                     && offset == 0) {
2059                                         uint16_t x =
2060                                             (cmd->opcode == O_IP_SRCPORT) ?
2061                                                 src_port : dst_port ;
2062                                         uint16_t *p =
2063                                             ((ipfw_insn_u16 *)cmd)->ports;
2064                                         int i;
2065
2066                                         for (i = cmdlen - 1; !match && i > 0;
2067                                              i--, p += 2) {
2068                                                 match =
2069                                                 (x >= p[0] && x <= p[1]);
2070                                         }
2071                                 }
2072                                 break;
2073
2074                         case O_ICMPTYPE:
2075                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
2076                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2077                                 break;
2078
2079                         case O_IPOPT:
2080                                 match = (hlen > 0 && ipopts_match(ip, cmd));
2081                                 break;
2082
2083                         case O_IPVER:
2084                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2085                                 break;
2086
2087                         case O_IPTTL:
2088                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2089                                 break;
2090
2091                         case O_IPID:
2092                                 match = (hlen > 0 &&
2093                                     cmd->arg1 == ntohs(ip->ip_id));
2094                                 break;
2095
2096                         case O_IPLEN:
2097                                 match = (hlen > 0 && cmd->arg1 == ip_len);
2098                                 break;
2099
2100                         case O_IPPRECEDENCE:
2101                                 match = (hlen > 0 &&
2102                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
2103                                 break;
2104
2105                         case O_IPTOS:
2106                                 match = (hlen > 0 &&
2107                                     flags_match(cmd, ip->ip_tos));
2108                                 break;
2109
2110                         case O_TCPFLAGS:
2111                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2112                                     flags_match(cmd,
2113                                         L3HDR(struct tcphdr,ip)->th_flags));
2114                                 break;
2115
2116                         case O_TCPOPTS:
2117                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2118                                     tcpopts_match(ip, cmd));
2119                                 break;
2120
2121                         case O_TCPSEQ:
2122                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2123                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2124                                         L3HDR(struct tcphdr,ip)->th_seq);
2125                                 break;
2126
2127                         case O_TCPACK:
2128                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2129                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2130                                         L3HDR(struct tcphdr,ip)->th_ack);
2131                                 break;
2132
2133                         case O_TCPWIN:
2134                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2135                                     cmd->arg1 ==
2136                                         L3HDR(struct tcphdr,ip)->th_win);
2137                                 break;
2138
2139                         case O_ESTAB:
2140                                 /* reject packets which have SYN only */
2141                                 /* XXX should i also check for TH_ACK ? */
2142                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2143                                     (L3HDR(struct tcphdr,ip)->th_flags &
2144                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2145                                 break;
2146
2147                         case O_LOG:
2148                                 if (fw_verbose)
2149                                         ipfw_log(f, hlen, args->eh, m, oif);
2150                                 match = 1;
2151                                 break;
2152
2153                         case O_PROB:
2154                                 match = (krandom() <
2155                                         ((ipfw_insn_u32 *)cmd)->d[0]);
2156                                 break;
2157
2158                         /*
2159                          * The second set of opcodes represents 'actions',
2160                          * i.e. the terminal part of a rule once the packet
2161                          * matches all previous patterns.
2162                          * Typically there is only one action for each rule,
2163                          * and the opcode is stored at the end of the rule
2164                          * (but there are exceptions -- see below).
2165                          *
2166                          * In general, here we set retval and terminate the
2167                          * outer loop (would be a 'break 3' in some language,
2168                          * but we need to do a 'goto done').
2169                          *
2170                          * Exceptions:
2171                          * O_COUNT and O_SKIPTO actions:
2172                          *   instead of terminating, we jump to the next rule
2173                          *   ('goto next_rule', equivalent to a 'break 2'),
2174                          *   or to the SKIPTO target ('goto again' after
2175                          *   having set f, cmd and l), respectively.
2176                          *
2177                          * O_LIMIT and O_KEEP_STATE: these opcodes are
2178                          *   not real 'actions', and are stored right
2179                          *   before the 'action' part of the rule.
2180                          *   These opcodes try to install an entry in the
2181                          *   state tables; if successful, we continue with
2182                          *   the next opcode (match=1; break;), otherwise
2183                          *   the packet must be dropped ('goto done' after
2184                          *   setting retval).  If static rules are changed
2185                          *   during the state installation, the packet will
2186                          *   be dropped and rule's stats will not beupdated
2187                          *   ('return IP_FW_DENY').
2188                          *
2189                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2190                          *   cause a lookup of the state table, and a jump
2191                          *   to the 'action' part of the parent rule
2192                          *   ('goto check_body') if an entry is found, or
2193                          *   (CHECK_STATE only) a jump to the next rule if
2194                          *   the entry is not found ('goto next_rule').
2195                          *   The result of the lookup is cached to make
2196                          *   further instances of these opcodes are
2197                          *   effectively NOPs.  If static rules are changed
2198                          *   during the state looking up, the packet will
2199                          *   be dropped and rule's stats will not be updated
2200                          *   ('return IP_FW_DENY').
2201                          */
2202                         case O_LIMIT:
2203                         case O_KEEP_STATE:
2204                                 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2205                                         kprintf("%s rule (%d) is not ready "
2206                                                 "on cpu%d\n",
2207                                                 cmd->opcode == O_LIMIT ?
2208                                                 "limit" : "keep state",
2209                                                 f->rulenum, f->cpuid);
2210                                         goto next_rule;
2211                                 }
2212                                 if (install_state(f,
2213                                     (ipfw_insn_limit *)cmd, args, &deny)) {
2214                                         if (deny)
2215                                                 return IP_FW_DENY;
2216
2217                                         retval = IP_FW_DENY;
2218                                         goto done; /* error/limit violation */
2219                                 }
2220                                 if (deny)
2221                                         return IP_FW_DENY;
2222                                 match = 1;
2223                                 break;
2224
2225                         case O_PROBE_STATE:
2226                         case O_CHECK_STATE:
2227                                 /*
2228                                  * dynamic rules are checked at the first
2229                                  * keep-state or check-state occurrence,
2230                                  * with the result being stored in dyn_dir.
2231                                  * The compiler introduces a PROBE_STATE
2232                                  * instruction for us when we have a
2233                                  * KEEP_STATE (because PROBE_STATE needs
2234                                  * to be run first).
2235                                  */
2236                                 if (dyn_dir == MATCH_UNKNOWN) {
2237                                         dyn_f = lookup_rule(&args->f_id,
2238                                                 &dyn_dir,
2239                                                 proto == IPPROTO_TCP ?
2240                                                 L3HDR(struct tcphdr, ip) : NULL,
2241                                                 ip_len, &deny);
2242                                         if (deny)
2243                                                 return IP_FW_DENY;
2244                                         if (dyn_f != NULL) {
2245                                                 /*
2246                                                  * Found a rule from a dynamic
2247                                                  * entry; jump to the 'action'
2248                                                  * part of the rule.
2249                                                  */
2250                                                 f = dyn_f;
2251                                                 cmd = ACTION_PTR(f);
2252                                                 l = f->cmd_len - f->act_ofs;
2253                                                 goto check_body;
2254                                         }
2255                                 }
2256                                 /*
2257                                  * Dynamic entry not found. If CHECK_STATE,
2258                                  * skip to next rule, if PROBE_STATE just
2259                                  * ignore and continue with next opcode.
2260                                  */
2261                                 if (cmd->opcode == O_CHECK_STATE)
2262                                         goto next_rule;
2263                                 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2264                                         goto next_rule; /* not ready yet */
2265                                 match = 1;
2266                                 break;
2267
2268                         case O_ACCEPT:
2269                                 retval = IP_FW_PASS;    /* accept */
2270                                 goto done;
2271
2272                         case O_PIPE:
2273                         case O_QUEUE:
2274                                 args->rule = f; /* report matching rule */
2275                                 args->cookie = cmd->arg1;
2276                                 retval = IP_FW_DUMMYNET;
2277                                 goto done;
2278
2279                         case O_DIVERT:
2280                         case O_TEE:
2281                                 if (args->eh) /* not on layer 2 */
2282                                         break;
2283
2284                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2285                                                  sizeof(*divinfo), MB_DONTWAIT);
2286                                 if (mtag == NULL) {
2287                                         retval = IP_FW_DENY;
2288                                         goto done;
2289                                 }
2290                                 divinfo = m_tag_data(mtag);
2291
2292                                 divinfo->skipto = f->rulenum;
2293                                 divinfo->port = cmd->arg1;
2294                                 divinfo->tee = (cmd->opcode == O_TEE);
2295                                 m_tag_prepend(m, mtag);
2296
2297                                 args->cookie = cmd->arg1;
2298                                 retval = (cmd->opcode == O_DIVERT) ?
2299                                          IP_FW_DIVERT : IP_FW_TEE;
2300                                 goto done;
2301
2302                         case O_COUNT:
2303                         case O_SKIPTO:
2304                                 f->pcnt++;      /* update stats */
2305                                 f->bcnt += ip_len;
2306                                 f->timestamp = time_second;
2307                                 if (cmd->opcode == O_COUNT)
2308                                         goto next_rule;
2309                                 /* handle skipto */
2310                                 if (f->next_rule == NULL)
2311                                         lookup_next_rule(f);
2312                                 f = f->next_rule;
2313                                 goto again;
2314
2315                         case O_REJECT:
2316                                 /*
2317                                  * Drop the packet and send a reject notice
2318                                  * if the packet is not ICMP (or is an ICMP
2319                                  * query), and it is not multicast/broadcast.
2320                                  */
2321                                 if (hlen > 0 &&
2322                                     (proto != IPPROTO_ICMP ||
2323                                      is_icmp_query(ip)) &&
2324                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
2325                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2326                                         /*
2327                                          * Update statistics before the possible
2328                                          * blocking 'send_reject'
2329                                          */
2330                                         f->pcnt++;
2331                                         f->bcnt += ip_len;
2332                                         f->timestamp = time_second;
2333
2334                                         send_reject(args, cmd->arg1,
2335                                             offset,ip_len);
2336                                         m = args->m;
2337
2338                                         /*
2339                                          * Return directly here, rule stats
2340                                          * have been updated above.
2341                                          */
2342                                         return IP_FW_DENY;
2343                                 }
2344                                 /* FALLTHROUGH */
2345                         case O_DENY:
2346                                 retval = IP_FW_DENY;
2347                                 goto done;
2348
2349                         case O_FORWARD_IP:
2350                                 if (args->eh)   /* not valid on layer2 pkts */
2351                                         break;
2352                                 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2353                                         struct sockaddr_in *sin;
2354
2355                                         mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2356                                                sizeof(*sin), MB_DONTWAIT);
2357                                         if (mtag == NULL) {
2358                                                 retval = IP_FW_DENY;
2359                                                 goto done;
2360                                         }
2361                                         sin = m_tag_data(mtag);
2362
2363                                         /* Structure copy */
2364                                         *sin = ((ipfw_insn_sa *)cmd)->sa;
2365
2366                                         m_tag_prepend(m, mtag);
2367                                         m->m_pkthdr.fw_flags |=
2368                                                 IPFORWARD_MBUF_TAGGED;
2369                                 }
2370                                 retval = IP_FW_PASS;
2371                                 goto done;
2372
2373                         default:
2374                                 panic("-- unknown opcode %d\n", cmd->opcode);
2375                         } /* end of switch() on opcodes */
2376
2377                         if (cmd->len & F_NOT)
2378                                 match = !match;
2379
2380                         if (match) {
2381                                 if (cmd->len & F_OR)
2382                                         skip_or = 1;
2383                         } else {
2384                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2385                                         break;          /* try next rule    */
2386                         }
2387
2388                 }       /* end of inner for, scan opcodes */
2389
2390 next_rule:;             /* try next rule                */
2391
2392         }               /* end of outer for, scan rules */
2393         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2394         return IP_FW_DENY;
2395
2396 done:
2397         /* Update statistics */
2398         f->pcnt++;
2399         f->bcnt += ip_len;
2400         f->timestamp = time_second;
2401         return retval;
2402
2403 pullup_failed:
2404         if (fw_verbose)
2405                 kprintf("pullup failed\n");
2406         return IP_FW_DENY;
2407 }
2408
2409 static void
2410 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2411 {
2412         struct m_tag *mtag;
2413         struct dn_pkt *pkt;
2414         ipfw_insn *cmd;
2415         const struct ipfw_flow_id *id;
2416         struct dn_flow_id *fid;
2417
2418         M_ASSERTPKTHDR(m);
2419
2420         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2421         if (mtag == NULL) {
2422                 m_freem(m);
2423                 return;
2424         }
2425         m_tag_prepend(m, mtag);
2426
2427         pkt = m_tag_data(mtag);
2428         bzero(pkt, sizeof(*pkt));
2429
2430         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2431         if (cmd->opcode == O_LOG)
2432                 cmd += F_LEN(cmd);
2433         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2434                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2435
2436         pkt->dn_m = m;
2437         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2438         pkt->ifp = fwa->oif;
2439         pkt->cpuid = mycpu->gd_cpuid;
2440         pkt->pipe_nr = pipe_nr;
2441
2442         id = &fwa->f_id;
2443         fid = &pkt->id;
2444         fid->fid_dst_ip = id->dst_ip;
2445         fid->fid_src_ip = id->src_ip;
2446         fid->fid_dst_port = id->dst_port;
2447         fid->fid_src_port = id->src_port;
2448         fid->fid_proto = id->proto;
2449         fid->fid_flags = id->flags;
2450
2451         ipfw_ref_rule(fwa->rule);
2452         pkt->dn_priv = fwa->rule;
2453         pkt->dn_unref_priv = ipfw_unref_rule;
2454
2455         if (cmd->opcode == O_PIPE)
2456                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2457
2458         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2459 }
2460
2461 /*
2462  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2463  * These will be reconstructed on the fly as packets are matched.
2464  * Must be called at splimp().
2465  */
2466 static void
2467 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2468 {
2469         struct ip_fw *rule;
2470
2471         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2472                 rule->next_rule = NULL;
2473 }
2474
2475 static __inline void
2476 ipfw_inc_static_count(struct ip_fw *rule)
2477 {
2478         KKASSERT(mycpuid == 0);
2479
2480         static_count++;
2481         static_ioc_len += IOC_RULESIZE(rule);
2482 }
2483
2484 static __inline void
2485 ipfw_dec_static_count(struct ip_fw *rule)
2486 {
2487         int l = IOC_RULESIZE(rule);
2488
2489         KKASSERT(mycpuid == 0);
2490
2491         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2492         static_count--;
2493
2494         KASSERT(static_ioc_len >= l,
2495                 ("invalid static len %u\n", static_ioc_len));
2496         static_ioc_len -= l;
2497 }
2498
2499 static void
2500 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2501 {
2502         if (fwmsg->sibling != NULL) {
2503                 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2504                 fwmsg->sibling->sibling = rule;
2505         }
2506         fwmsg->sibling = rule;
2507 }
2508
2509 static struct ip_fw *
2510 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2511 {
2512         struct ip_fw *rule;
2513
2514         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2515
2516         rule->act_ofs = ioc_rule->act_ofs;
2517         rule->cmd_len = ioc_rule->cmd_len;
2518         rule->rulenum = ioc_rule->rulenum;
2519         rule->set = ioc_rule->set;
2520         rule->usr_flags = ioc_rule->usr_flags;
2521
2522         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2523
2524         rule->refcnt = 1;
2525         rule->cpuid = mycpuid;
2526
2527         rule->stub = stub;
2528         if (stub != NULL)
2529                 stub->rule[mycpuid] = rule;
2530
2531         return rule;
2532 }
2533
2534 static void
2535 ipfw_add_rule_dispatch(struct netmsg *nmsg)
2536 {
2537         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2538         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2539         struct ip_fw *rule;
2540
2541         rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2542
2543         /*
2544          * Bump generation after ipfw_create_rule(),
2545          * since this function is blocking
2546          */
2547         ctx->ipfw_gen++;
2548
2549         /*
2550          * Insert rule into the pre-determined position
2551          */
2552         if (fwmsg->prev_rule != NULL) {
2553                 struct ip_fw *prev, *next;
2554
2555                 prev = fwmsg->prev_rule;
2556                 KKASSERT(prev->cpuid == mycpuid);
2557
2558                 next = fwmsg->next_rule;
2559                 KKASSERT(next->cpuid == mycpuid);
2560
2561                 rule->next = next;
2562                 prev->next = rule;
2563
2564                 /*
2565                  * Move to the position on the next CPU
2566                  * before the msg is forwarded.
2567                  */
2568                 fwmsg->prev_rule = prev->sibling;
2569                 fwmsg->next_rule = next->sibling;
2570         } else {
2571                 KKASSERT(fwmsg->next_rule == NULL);
2572                 rule->next = ctx->ipfw_layer3_chain;
2573                 ctx->ipfw_layer3_chain = rule;
2574         }
2575
2576         /* Link rule CPU sibling */
2577         ipfw_link_sibling(fwmsg, rule);
2578
2579         ipfw_flush_rule_ptrs(ctx);
2580
2581         if (mycpuid == 0) {
2582                 /* Statistics only need to be updated once */
2583                 ipfw_inc_static_count(rule);
2584
2585                 /* Return the rule on CPU0 */
2586                 nmsg->nm_lmsg.u.ms_resultp = rule;
2587         }
2588
2589         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2590 }
2591
2592 static void
2593 ipfw_enable_state_dispatch(struct netmsg *nmsg)
2594 {
2595         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
2596         struct ip_fw *rule = lmsg->u.ms_resultp;
2597         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2598
2599         ctx->ipfw_gen++;
2600
2601         KKASSERT(rule->cpuid == mycpuid);
2602         KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule);
2603         KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE));
2604         rule->rule_flags |= IPFW_RULE_F_STATE;
2605         lmsg->u.ms_resultp = rule->sibling;
2606
2607         ifnet_forwardmsg(lmsg, mycpuid + 1);
2608 }
2609
2610 /*
2611  * Add a new rule to the list.  Copy the rule into a malloc'ed area,
2612  * then possibly create a rule number and add the rule to the list.
2613  * Update the rule_number in the input struct so the caller knows
2614  * it as well.
2615  */
2616 static void
2617 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
2618 {
2619         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2620         struct netmsg_ipfw fwmsg;
2621         struct netmsg *nmsg;
2622         struct ip_fw *f, *prev, *rule;
2623         struct ip_fw_stub *stub;
2624
2625         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2626
2627         crit_enter();
2628
2629         /*
2630          * If rulenum is 0, find highest numbered rule before the
2631          * default rule, and add rule number incremental step.
2632          */
2633         if (ioc_rule->rulenum == 0) {
2634                 int step = autoinc_step;
2635
2636                 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2637                          step <= IPFW_AUTOINC_STEP_MAX);
2638
2639                 /*
2640                  * Locate the highest numbered rule before default
2641                  */
2642                 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2643                         if (f->rulenum == IPFW_DEFAULT_RULE)
2644                                 break;
2645                         ioc_rule->rulenum = f->rulenum;
2646                 }
2647                 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2648                         ioc_rule->rulenum += step;
2649         }
2650         KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2651                 ioc_rule->rulenum != 0,
2652                 ("invalid rule num %d\n", ioc_rule->rulenum));
2653
2654         /*
2655          * Now find the right place for the new rule in the sorted list.
2656          */
2657         for (prev = NULL, f = ctx->ipfw_layer3_chain; f;
2658              prev = f, f = f->next) {
2659                 if (f->rulenum > ioc_rule->rulenum) {
2660                         /* Found the location */
2661                         break;
2662                 }
2663         }
2664         KASSERT(f != NULL, ("no default rule?!\n"));
2665
2666         if (rule_flags & IPFW_RULE_F_STATE) {
2667                 int size;
2668
2669                 /*
2670                  * If the new rule will create states, then allocate
2671                  * a rule stub, which will be referenced by states
2672                  * (dyn rules)
2673                  */
2674                 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2675                 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2676         } else {
2677                 stub = NULL;
2678         }
2679
2680         /*
2681          * Duplicate the rule onto each CPU.
2682          * The rule duplicated on CPU0 will be returned.
2683          */
2684         bzero(&fwmsg, sizeof(fwmsg));
2685         nmsg = &fwmsg.nmsg;
2686         netmsg_init(nmsg, &curthread->td_msgport, 0, ipfw_add_rule_dispatch);
2687         fwmsg.ioc_rule = ioc_rule;
2688         fwmsg.prev_rule = prev;
2689         fwmsg.next_rule = prev == NULL ? NULL : f;
2690         fwmsg.stub = stub;
2691
2692         ifnet_domsg(&nmsg->nm_lmsg, 0);
2693         KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2694
2695         rule = nmsg->nm_lmsg.u.ms_resultp;
2696         KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2697
2698         if (rule_flags & IPFW_RULE_F_STATE) {
2699                 /*
2700                  * Turn on state flag, _after_ everything on all
2701                  * CPUs have been setup.
2702                  */
2703                 bzero(nmsg, sizeof(*nmsg));
2704                 netmsg_init(nmsg, &curthread->td_msgport, 0,
2705                             ipfw_enable_state_dispatch);
2706                 nmsg->nm_lmsg.u.ms_resultp = rule;
2707
2708                 ifnet_domsg(&nmsg->nm_lmsg, 0);
2709                 KKASSERT(nmsg->nm_lmsg.u.ms_resultp == NULL);
2710         }
2711
2712         crit_exit();
2713
2714         DEB(kprintf("++ installed rule %d, static count now %d\n",
2715                 rule->rulenum, static_count);)
2716 }
2717
2718 /**
2719  * Free storage associated with a static rule (including derived
2720  * dynamic rules).
2721  * The caller is in charge of clearing rule pointers to avoid
2722  * dangling pointers.
2723  * @return a pointer to the next entry.
2724  * Arguments are not checked, so they better be correct.
2725  * Must be called at splimp().
2726  */
2727 static struct ip_fw *
2728 ipfw_delete_rule(struct ipfw_context *ctx,
2729                  struct ip_fw *prev, struct ip_fw *rule)
2730 {
2731         struct ip_fw *n;
2732         struct ip_fw_stub *stub;
2733
2734         ctx->ipfw_gen++;
2735
2736         /* STATE flag should have been cleared before we reach here */
2737         KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2738
2739         stub = rule->stub;
2740         n = rule->next;
2741         if (prev == NULL)
2742                 ctx->ipfw_layer3_chain = n;
2743         else
2744                 prev->next = n;
2745
2746         /* Mark the rule as invalid */
2747         rule->rule_flags |= IPFW_RULE_F_INVALID;
2748         rule->next_rule = NULL;
2749         rule->sibling = NULL;
2750         rule->stub = NULL;
2751 #ifdef foo
2752         /* Don't reset cpuid here; keep various assertion working */
2753         rule->cpuid = -1;
2754 #endif
2755
2756         /* Statistics only need to be updated once */
2757         if (mycpuid == 0)
2758                 ipfw_dec_static_count(rule);
2759
2760         /* Free 'stub' on the last CPU */
2761         if (stub != NULL && mycpuid == ncpus - 1)
2762                 kfree(stub, M_IPFW);
2763
2764         /* Try to free this rule */
2765         ipfw_free_rule(rule);
2766
2767         /* Return the next rule */
2768         return n;
2769 }
2770
2771 static void
2772 ipfw_flush_dispatch(struct netmsg *nmsg)
2773 {
2774         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
2775         int kill_default = lmsg->u.ms_result;
2776         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2777         struct ip_fw *rule;
2778
2779         ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2780
2781         while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2782                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2783                 ipfw_delete_rule(ctx, NULL, rule);
2784
2785         ifnet_forwardmsg(lmsg, mycpuid + 1);
2786 }
2787
2788 static void
2789 ipfw_disable_rule_state_dispatch(struct netmsg *nmsg)
2790 {
2791         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2792         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2793         struct ip_fw *rule;
2794
2795         ctx->ipfw_gen++;
2796
2797         rule = dmsg->start_rule;
2798         if (rule != NULL) {
2799                 KKASSERT(rule->cpuid == mycpuid);
2800
2801                 /*
2802                  * Move to the position on the next CPU
2803                  * before the msg is forwarded.
2804                  */
2805                 dmsg->start_rule = rule->sibling;
2806         } else {
2807                 KKASSERT(dmsg->rulenum == 0);
2808                 rule = ctx->ipfw_layer3_chain;
2809         }
2810
2811         while (rule != NULL) {
2812                 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2813                         break;
2814                 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2815                 rule = rule->next;
2816         }
2817
2818         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2819 }
2820
2821 /*
2822  * Deletes all rules from a chain (including the default rule
2823  * if the second argument is set).
2824  * Must be called at splimp().
2825  */
2826 static void
2827 ipfw_flush(int kill_default)
2828 {
2829         struct netmsg_del dmsg;
2830         struct netmsg nmsg;
2831         struct lwkt_msg *lmsg;
2832         struct ip_fw *rule;
2833         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2834
2835         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2836
2837         /*
2838          * If 'kill_default' then caller has done the necessary
2839          * msgport syncing; unnecessary to do it again.
2840          */
2841         if (!kill_default) {
2842                 /*
2843                  * Let ipfw_chk() know the rules are going to
2844                  * be flushed, so it could jump directly to
2845                  * the default rule.
2846                  */
2847                 ipfw_flushing = 1;
2848                 netmsg_service_sync();
2849         }
2850
2851         /*
2852          * Clear STATE flag on rules, so no more states (dyn rules)
2853          * will be created.
2854          */
2855         bzero(&dmsg, sizeof(dmsg));
2856         netmsg_init(&dmsg.nmsg, &curthread->td_msgport, 0,
2857                     ipfw_disable_rule_state_dispatch);
2858         ifnet_domsg(&dmsg.nmsg.nm_lmsg, 0);
2859
2860         /*
2861          * This actually nukes all states (dyn rules)
2862          */
2863         lockmgr(&dyn_lock, LK_EXCLUSIVE);
2864         for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2865                 /*
2866                  * Can't check IPFW_RULE_F_STATE here,
2867                  * since it has been cleared previously.
2868                  * Check 'stub' instead.
2869                  */
2870                 if (rule->stub != NULL) {
2871                         /* Force removal */
2872                         remove_dyn_rule_locked(rule, NULL);
2873                 }
2874         }
2875         lockmgr(&dyn_lock, LK_RELEASE);
2876
2877         /*
2878          * Press the 'flush' button
2879          */
2880         bzero(&nmsg, sizeof(nmsg));
2881         netmsg_init(&nmsg, &curthread->td_msgport, 0, ipfw_flush_dispatch);
2882         lmsg = &nmsg.nm_lmsg;
2883         lmsg->u.ms_result = kill_default;
2884         ifnet_domsg(lmsg, 0);
2885
2886         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2887
2888         if (kill_default) {
2889                 if (ipfw_dyn_v != NULL) {
2890                         /*
2891                          * Free dynamic rules(state) hash table
2892                          */
2893                         kfree(ipfw_dyn_v, M_IPFW);
2894                         ipfw_dyn_v = NULL;
2895                 }
2896
2897                 KASSERT(static_count == 0,
2898                         ("%u static rules remains\n", static_count));
2899                 KASSERT(static_ioc_len == 0,
2900                         ("%u bytes of static rules remains\n", static_ioc_len));
2901         } else {
2902                 KASSERT(static_count == 1,
2903                         ("%u static rules remains\n", static_count));
2904                 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2905                         ("%u bytes of static rules remains, should be %u\n",
2906                          static_ioc_len, IOC_RULESIZE(ctx->ipfw_default_rule)));
2907         }
2908
2909         /* Flush is done */
2910         ipfw_flushing = 0;
2911 }
2912
2913 static void
2914 ipfw_alt_delete_rule_dispatch(struct netmsg *nmsg)
2915 {
2916         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2917         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2918         struct ip_fw *rule, *prev;
2919
2920         rule = dmsg->start_rule;
2921         KKASSERT(rule->cpuid == mycpuid);
2922         dmsg->start_rule = rule->sibling;
2923
2924         prev = dmsg->prev_rule;
2925         if (prev != NULL) {
2926                 KKASSERT(prev->cpuid == mycpuid);
2927
2928                 /*
2929                  * Move to the position on the next CPU
2930                  * before the msg is forwarded.
2931                  */
2932                 dmsg->prev_rule = prev->sibling;
2933         }
2934
2935         /*
2936          * flush pointers outside the loop, then delete all matching
2937          * rules.  'prev' remains the same throughout the cycle.
2938          */
2939         ipfw_flush_rule_ptrs(ctx);
2940         while (rule && rule->rulenum == dmsg->rulenum)
2941                 rule = ipfw_delete_rule(ctx, prev, rule);
2942
2943         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2944 }
2945
2946 static int
2947 ipfw_alt_delete_rule(uint16_t rulenum)
2948 {
2949         struct ip_fw *prev, *rule, *f;
2950         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2951         struct netmsg_del dmsg;
2952         struct netmsg *nmsg;
2953         int state;
2954
2955         /*
2956          * Locate first rule to delete
2957          */
2958         for (prev = NULL, rule = ctx->ipfw_layer3_chain;
2959              rule && rule->rulenum < rulenum;
2960              prev = rule, rule = rule->next)
2961                 ; /* EMPTY */
2962         if (rule->rulenum != rulenum)
2963                 return EINVAL;
2964
2965         /*
2966          * Check whether any rules with the given number will
2967          * create states.
2968          */
2969         state = 0;
2970         for (f = rule; f && f->rulenum == rulenum; f = f->next) {
2971                 if (f->rule_flags & IPFW_RULE_F_STATE) {
2972                         state = 1;
2973                         break;
2974                 }
2975         }
2976
2977         if (state) {
2978                 /*
2979                  * Clear the STATE flag, so no more states will be
2980                  * created based the rules numbered 'rulenum'.
2981                  */
2982                 bzero(&dmsg, sizeof(dmsg));
2983                 nmsg = &dmsg.nmsg;
2984                 netmsg_init(nmsg, &curthread->td_msgport, 0,
2985                             ipfw_disable_rule_state_dispatch);
2986                 dmsg.start_rule = rule;
2987                 dmsg.rulenum = rulenum;
2988
2989                 ifnet_domsg(&nmsg->nm_lmsg, 0);
2990                 KKASSERT(dmsg.start_rule == NULL);
2991
2992                 /*
2993                  * Nuke all related states
2994                  */
2995                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
2996                 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
2997                         /*
2998                          * Can't check IPFW_RULE_F_STATE here,
2999                          * since it has been cleared previously.
3000                          * Check 'stub' instead.
3001                          */
3002                         if (f->stub != NULL) {
3003                                 /* Force removal */
3004                                 remove_dyn_rule_locked(f, NULL);
3005                         }
3006                 }
3007                 lockmgr(&dyn_lock, LK_RELEASE);
3008         }
3009
3010         /*
3011          * Get rid of the rule duplications on all CPUs
3012          */
3013         bzero(&dmsg, sizeof(dmsg));
3014         nmsg = &dmsg.nmsg;
3015         netmsg_init(nmsg, &curthread->td_msgport, 0,
3016                     ipfw_alt_delete_rule_dispatch);
3017         dmsg.prev_rule = prev;
3018         dmsg.start_rule = rule;
3019         dmsg.rulenum = rulenum;
3020
3021         ifnet_domsg(&nmsg->nm_lmsg, 0);
3022         KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3023         return 0;
3024 }
3025
3026 static void
3027 ipfw_alt_delete_ruleset_dispatch(struct netmsg *nmsg)
3028 {
3029         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3030         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3031         struct ip_fw *prev, *rule;
3032 #ifdef INVARIANTS
3033         int del = 0;
3034 #endif
3035
3036         ipfw_flush_rule_ptrs(ctx);
3037
3038         prev = NULL;
3039         rule = ctx->ipfw_layer3_chain;
3040         while (rule != NULL) {
3041                 if (rule->set == dmsg->from_set) {
3042                         rule = ipfw_delete_rule(ctx, prev, rule);
3043 #ifdef INVARIANTS
3044                         del = 1;
3045 #endif
3046                 } else {
3047                         prev = rule;
3048                         rule = rule->next;
3049                 }
3050         }
3051         KASSERT(del, ("no match set?!\n"));
3052
3053         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3054 }
3055
3056 static void
3057 ipfw_disable_ruleset_state_dispatch(struct netmsg *nmsg)
3058 {
3059         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3060         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3061         struct ip_fw *rule;
3062 #ifdef INVARIANTS
3063         int cleared = 0;
3064 #endif
3065
3066         ctx->ipfw_gen++;
3067
3068         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3069                 if (rule->set == dmsg->from_set) {
3070 #ifdef INVARIANTS
3071                         cleared = 1;
3072 #endif
3073                         rule->rule_flags &= ~IPFW_RULE_F_STATE;
3074                 }
3075         }
3076         KASSERT(cleared, ("no match set?!\n"));
3077
3078         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3079 }
3080
3081 static int
3082 ipfw_alt_delete_ruleset(uint8_t set)
3083 {
3084         struct netmsg_del dmsg;
3085         struct netmsg *nmsg;
3086         int state, del;
3087         struct ip_fw *rule;
3088         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3089
3090         /*
3091          * Check whether the 'set' exists.  If it exists,
3092          * then check whether any rules within the set will
3093          * try to create states.
3094          */
3095         state = 0;
3096         del = 0;
3097         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3098                 if (rule->set == set) {
3099                         del = 1;
3100                         if (rule->rule_flags & IPFW_RULE_F_STATE) {
3101                                 state = 1;
3102                                 break;
3103                         }
3104                 }
3105         }
3106         if (!del)
3107                 return 0; /* XXX EINVAL? */
3108
3109         if (state) {
3110                 /*
3111                  * Clear the STATE flag, so no more states will be
3112                  * created based the rules in this set.
3113                  */
3114                 bzero(&dmsg, sizeof(dmsg));
3115                 nmsg = &dmsg.nmsg;
3116                 netmsg_init(nmsg, &curthread->td_msgport, 0,
3117                             ipfw_disable_ruleset_state_dispatch);
3118                 dmsg.from_set = set;
3119
3120                 ifnet_domsg(&nmsg->nm_lmsg, 0);
3121
3122                 /*
3123                  * Nuke all related states
3124                  */
3125                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3126                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3127                         if (rule->set != set)
3128                                 continue;
3129
3130                         /*
3131                          * Can't check IPFW_RULE_F_STATE here,
3132                          * since it has been cleared previously.
3133                          * Check 'stub' instead.
3134                          */
3135                         if (rule->stub != NULL) {
3136                                 /* Force removal */
3137                                 remove_dyn_rule_locked(rule, NULL);
3138                         }
3139                 }
3140                 lockmgr(&dyn_lock, LK_RELEASE);
3141         }
3142
3143         /*
3144          * Delete this set
3145          */
3146         bzero(&dmsg, sizeof(dmsg));
3147         nmsg = &dmsg.nmsg;
3148         netmsg_init(nmsg, &curthread->td_msgport, 0,
3149                     ipfw_alt_delete_ruleset_dispatch);
3150         dmsg.from_set = set;
3151
3152         ifnet_domsg(&nmsg->nm_lmsg, 0);
3153         return 0;
3154 }
3155
3156 static void
3157 ipfw_alt_move_rule_dispatch(struct netmsg *nmsg)
3158 {
3159         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3160         struct ip_fw *rule;
3161
3162         rule = dmsg->start_rule;
3163         KKASSERT(rule->cpuid == mycpuid);
3164
3165         /*
3166          * Move to the position on the next CPU
3167          * before the msg is forwarded.
3168          */
3169         dmsg->start_rule = rule->sibling;
3170
3171         while (rule && rule->rulenum <= dmsg->rulenum) {
3172                 if (rule->rulenum == dmsg->rulenum)
3173                         rule->set = dmsg->to_set;
3174                 rule = rule->next;
3175         }
3176         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3177 }
3178
3179 static int
3180 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3181 {
3182         struct netmsg_del dmsg;
3183         struct netmsg *nmsg;
3184         struct ip_fw *rule;
3185         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3186
3187         /*
3188          * Locate first rule to move
3189          */
3190         for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3191              rule = rule->next) {
3192                 if (rule->rulenum == rulenum && rule->set != set)
3193                         break;
3194         }
3195         if (rule == NULL || rule->rulenum > rulenum)
3196                 return 0; /* XXX error? */
3197
3198         bzero(&dmsg, sizeof(dmsg));
3199         nmsg = &dmsg.nmsg;
3200         netmsg_init(nmsg, &curthread->td_msgport, 0,
3201                     ipfw_alt_move_rule_dispatch);
3202         dmsg.start_rule = rule;
3203         dmsg.rulenum = rulenum;
3204         dmsg.to_set = set;
3205
3206         ifnet_domsg(&nmsg->nm_lmsg, 0);
3207         KKASSERT(dmsg.start_rule == NULL);
3208         return 0;
3209 }
3210
3211 static void
3212 ipfw_alt_move_ruleset_dispatch(struct netmsg *nmsg)
3213 {
3214         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3215         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3216         struct ip_fw *rule;
3217
3218         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3219                 if (rule->set == dmsg->from_set)
3220                         rule->set = dmsg->to_set;
3221         }
3222         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3223 }
3224
3225 static int
3226 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3227 {
3228         struct netmsg_del dmsg;
3229         struct netmsg *nmsg;
3230
3231         bzero(&dmsg, sizeof(dmsg));
3232         nmsg = &dmsg.nmsg;
3233         netmsg_init(nmsg, &curthread->td_msgport, 0,
3234                     ipfw_alt_move_ruleset_dispatch);
3235         dmsg.from_set = from_set;
3236         dmsg.to_set = to_set;
3237
3238         ifnet_domsg(&nmsg->nm_lmsg, 0);
3239         return 0;
3240 }
3241
3242 static void
3243 ipfw_alt_swap_ruleset_dispatch(struct netmsg *nmsg)
3244 {
3245         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3246         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3247         struct ip_fw *rule;
3248
3249         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3250                 if (rule->set == dmsg->from_set)
3251                         rule->set = dmsg->to_set;
3252                 else if (rule->set == dmsg->to_set)
3253                         rule->set = dmsg->from_set;
3254         }
3255         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3256 }
3257
3258 static int
3259 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3260 {
3261         struct netmsg_del dmsg;
3262         struct netmsg *nmsg;
3263
3264         bzero(&dmsg, sizeof(dmsg));
3265         nmsg = &dmsg.nmsg;
3266         netmsg_init(nmsg, &curthread->td_msgport, 0,
3267                     ipfw_alt_swap_ruleset_dispatch);
3268         dmsg.from_set = set1;
3269         dmsg.to_set = set2;
3270
3271         ifnet_domsg(&nmsg->nm_lmsg, 0);
3272         return 0;
3273 }
3274
3275 /**
3276  * Remove all rules with given number, and also do set manipulation.
3277  *
3278  * The argument is an uint32_t. The low 16 bit are the rule or set number,
3279  * the next 8 bits are the new set, the top 8 bits are the command:
3280  *
3281  *      0       delete rules with given number
3282  *      1       delete rules with given set number
3283  *      2       move rules with given number to new set
3284  *      3       move rules with given set number to new set
3285  *      4       swap sets with given numbers
3286  */
3287 static int
3288 ipfw_ctl_alter(uint32_t arg)
3289 {
3290         uint16_t rulenum;
3291         uint8_t cmd, new_set;
3292         int error = 0;
3293
3294         rulenum = arg & 0xffff;
3295         cmd = (arg >> 24) & 0xff;
3296         new_set = (arg >> 16) & 0xff;
3297
3298         if (cmd > 4)
3299                 return EINVAL;
3300         if (new_set >= IPFW_DEFAULT_SET)
3301                 return EINVAL;
3302         if (cmd == 0 || cmd == 2) {
3303                 if (rulenum == IPFW_DEFAULT_RULE)
3304                         return EINVAL;
3305         } else {
3306                 if (rulenum >= IPFW_DEFAULT_SET)
3307                         return EINVAL;
3308         }
3309
3310         switch (cmd) {
3311         case 0: /* delete rules with given number */
3312                 error = ipfw_alt_delete_rule(rulenum);
3313                 break;
3314
3315         case 1: /* delete all rules with given set number */
3316                 error = ipfw_alt_delete_ruleset(rulenum);
3317                 break;
3318
3319         case 2: /* move rules with given number to new set */
3320                 error = ipfw_alt_move_rule(rulenum, new_set);
3321                 break;
3322
3323         case 3: /* move rules with given set number to new set */
3324                 error = ipfw_alt_move_ruleset(rulenum, new_set);
3325                 break;
3326
3327         case 4: /* swap two sets */
3328                 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3329                 break;
3330         }
3331         return error;
3332 }
3333
3334 /*
3335  * Clear counters for a specific rule.
3336  */
3337 static void
3338 clear_counters(struct ip_fw *rule, int log_only)
3339 {
3340         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3341
3342         if (log_only == 0) {
3343                 rule->bcnt = rule->pcnt = 0;
3344                 rule->timestamp = 0;
3345         }
3346         if (l->o.opcode == O_LOG)
3347                 l->log_left = l->max_log;
3348 }
3349
3350 static void
3351 ipfw_zero_entry_dispatch(struct netmsg *nmsg)
3352 {
3353         struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3354         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3355         struct ip_fw *rule;
3356
3357         if (zmsg->rulenum == 0) {
3358                 KKASSERT(zmsg->start_rule == NULL);
3359
3360                 ctx->ipfw_norule_counter = 0;
3361                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3362                         clear_counters(rule, zmsg->log_only);
3363         } else {
3364                 struct ip_fw *start = zmsg->start_rule;
3365
3366                 KKASSERT(start->cpuid == mycpuid);
3367                 KKASSERT(start->rulenum == zmsg->rulenum);
3368
3369                 /*
3370                  * We can have multiple rules with the same number, so we
3371                  * need to clear them all.
3372                  */
3373                 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3374                      rule = rule->next)
3375                         clear_counters(rule, zmsg->log_only);
3376
3377                 /*
3378                  * Move to the position on the next CPU
3379                  * before the msg is forwarded.
3380                  */
3381                 zmsg->start_rule = start->sibling;
3382         }
3383         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3384 }
3385
3386 /**
3387  * Reset some or all counters on firewall rules.
3388  * @arg frwl is null to clear all entries, or contains a specific
3389  * rule number.
3390  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3391  */
3392 static int
3393 ipfw_ctl_zero_entry(int rulenum, int log_only)
3394 {
3395         struct netmsg_zent zmsg;
3396         struct netmsg *nmsg;
3397         const char *msg;
3398         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3399
3400         bzero(&zmsg, sizeof(zmsg));
3401         nmsg = &zmsg.nmsg;
3402         netmsg_init(nmsg, &curthread->td_msgport, 0, ipfw_zero_entry_dispatch);
3403         zmsg.log_only = log_only;
3404
3405         if (rulenum == 0) {
3406                 msg = log_only ? "ipfw: All logging counts reset.\n"
3407                                : "ipfw: Accounting cleared.\n";
3408         } else {
3409                 struct ip_fw *rule;
3410
3411                 /*
3412                  * Locate the first rule with 'rulenum'
3413                  */
3414                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3415                         if (rule->rulenum == rulenum)
3416                                 break;
3417                 }
3418                 if (rule == NULL) /* we did not find any matching rules */
3419                         return (EINVAL);
3420                 zmsg.start_rule = rule;
3421                 zmsg.rulenum = rulenum;
3422
3423                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3424                                : "ipfw: Entry %d cleared.\n";
3425         }
3426         ifnet_domsg(&nmsg->nm_lmsg, 0);
3427         KKASSERT(zmsg.start_rule == NULL);
3428
3429         if (fw_verbose)
3430                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3431         return (0);
3432 }
3433
3434 /*
3435  * Check validity of the structure before insert.
3436  * Fortunately rules are simple, so this mostly need to check rule sizes.
3437  */
3438 static int
3439 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3440 {
3441         int l, cmdlen = 0;
3442         int have_action = 0;
3443         ipfw_insn *cmd;
3444
3445         *rule_flags = 0;
3446
3447         /* Check for valid size */
3448         if (size < sizeof(*rule)) {
3449                 kprintf("ipfw: rule too short\n");
3450                 return EINVAL;
3451         }
3452         l = IOC_RULESIZE(rule);
3453         if (l != size) {
3454                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3455                 return EINVAL;
3456         }
3457
3458         /* Check rule number */
3459         if (rule->rulenum == IPFW_DEFAULT_RULE) {
3460                 kprintf("ipfw: invalid rule number\n");
3461                 return EINVAL;
3462         }
3463
3464         /*
3465          * Now go for the individual checks. Very simple ones, basically only
3466          * instruction sizes.
3467          */
3468         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3469              l -= cmdlen, cmd += cmdlen) {
3470                 cmdlen = F_LEN(cmd);
3471                 if (cmdlen > l) {
3472                         kprintf("ipfw: opcode %d size truncated\n",
3473                                 cmd->opcode);
3474                         return EINVAL;
3475                 }
3476
3477                 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
3478
3479                 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3480                         /* This rule will create states */
3481                         *rule_flags |= IPFW_RULE_F_STATE;
3482                 }
3483
3484                 switch (cmd->opcode) {
3485                 case O_NOP:
3486                 case O_PROBE_STATE:
3487                 case O_KEEP_STATE:
3488                 case O_PROTO:
3489                 case O_IP_SRC_ME:
3490                 case O_IP_DST_ME:
3491                 case O_LAYER2:
3492                 case O_IN:
3493                 case O_FRAG:
3494                 case O_IPOPT:
3495                 case O_IPLEN:
3496                 case O_IPID:
3497                 case O_IPTOS:
3498                 case O_IPPRECEDENCE:
3499                 case O_IPTTL:
3500                 case O_IPVER:
3501                 case O_TCPWIN:
3502                 case O_TCPFLAGS:
3503                 case O_TCPOPTS:
3504                 case O_ESTAB:
3505                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3506                                 goto bad_size;
3507                         break;
3508
3509                 case O_UID:
3510                 case O_GID:
3511                 case O_IP_SRC:
3512                 case O_IP_DST:
3513                 case O_TCPSEQ:
3514                 case O_TCPACK:
3515                 case O_PROB:
3516                 case O_ICMPTYPE:
3517                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3518                                 goto bad_size;
3519                         break;
3520
3521                 case O_LIMIT:
3522                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3523                                 goto bad_size;
3524                         break;
3525
3526                 case O_LOG:
3527                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3528                                 goto bad_size;
3529
3530                         ((ipfw_insn_log *)cmd)->log_left =
3531                             ((ipfw_insn_log *)cmd)->max_log;
3532
3533                         break;
3534
3535                 case O_IP_SRC_MASK:
3536                 case O_IP_DST_MASK:
3537                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3538                                 goto bad_size;
3539                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3540                                 kprintf("ipfw: opcode %d, useless rule\n",
3541                                         cmd->opcode);
3542                                 return EINVAL;
3543                         }
3544                         break;
3545
3546                 case O_IP_SRC_SET:
3547                 case O_IP_DST_SET:
3548                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3549                                 kprintf("ipfw: invalid set size %d\n",
3550                                         cmd->arg1);
3551                                 return EINVAL;
3552                         }
3553                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3554                             (cmd->arg1+31)/32 )
3555                                 goto bad_size;
3556                         break;
3557
3558                 case O_MACADDR2:
3559                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3560                                 goto bad_size;
3561                         break;
3562
3563                 case O_MAC_TYPE:
3564                 case O_IP_SRCPORT:
3565                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3566                         if (cmdlen < 2 || cmdlen > 31)
3567                                 goto bad_size;
3568                         break;
3569
3570                 case O_RECV:
3571                 case O_XMIT:
3572                 case O_VIA:
3573                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3574                                 goto bad_size;
3575                         break;
3576
3577                 case O_PIPE:
3578                 case O_QUEUE:
3579                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3580                                 goto bad_size;
3581                         goto check_action;
3582
3583                 case O_FORWARD_IP:
3584                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3585                                 goto bad_size;
3586                         } else {
3587                                 in_addr_t fwd_addr;
3588
3589                                 fwd_addr = ((ipfw_insn_sa *)cmd)->
3590                                            sa.sin_addr.s_addr;
3591                                 if (IN_MULTICAST(ntohl(fwd_addr))) {
3592                                         kprintf("ipfw: try forwarding to "
3593                                                 "multicast address\n");
3594                                         return EINVAL;
3595                                 }
3596                         }
3597                         goto check_action;
3598
3599                 case O_FORWARD_MAC: /* XXX not implemented yet */
3600                 case O_CHECK_STATE:
3601                 case O_COUNT:
3602                 case O_ACCEPT:
3603                 case O_DENY:
3604                 case O_REJECT:
3605                 case O_SKIPTO:
3606                 case O_DIVERT:
3607                 case O_TEE:
3608                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3609                                 goto bad_size;
3610 check_action:
3611                         if (have_action) {
3612                                 kprintf("ipfw: opcode %d, multiple actions"
3613                                         " not allowed\n",
3614                                         cmd->opcode);
3615                                 return EINVAL;
3616                         }
3617                         have_action = 1;
3618                         if (l != cmdlen) {
3619                                 kprintf("ipfw: opcode %d, action must be"
3620                                         " last opcode\n",
3621                                         cmd->opcode);
3622                                 return EINVAL;
3623                         }
3624                         break;
3625                 default:
3626                         kprintf("ipfw: opcode %d, unknown opcode\n",
3627                                 cmd->opcode);
3628                         return EINVAL;
3629                 }
3630         }
3631         if (have_action == 0) {
3632                 kprintf("ipfw: missing action\n");
3633                 return EINVAL;
3634         }
3635         return 0;
3636
3637 bad_size:
3638         kprintf("ipfw: opcode %d size %d wrong\n",
3639                 cmd->opcode, cmdlen);
3640         return EINVAL;
3641 }
3642
3643 static int
3644 ipfw_ctl_add_rule(struct sockopt *sopt)
3645 {
3646         struct ipfw_ioc_rule *ioc_rule;
3647         size_t size;
3648         uint32_t rule_flags;
3649         int error;
3650
3651         size = sopt->sopt_valsize;
3652         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3653             size < sizeof(*ioc_rule)) {
3654                 return EINVAL;
3655         }
3656         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3657                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3658                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3659         }
3660         ioc_rule = sopt->sopt_val;
3661
3662         error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3663         if (error)
3664                 return error;
3665
3666         ipfw_add_rule(ioc_rule, rule_flags);
3667
3668         if (sopt->sopt_dir == SOPT_GET)
3669                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3670         return 0;
3671 }
3672
3673 static void *
3674 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3675 {
3676         const struct ip_fw *sibling;
3677 #ifdef INVARIANTS
3678         int i;
3679 #endif
3680
3681         KKASSERT(rule->cpuid == 0);
3682
3683         ioc_rule->act_ofs = rule->act_ofs;
3684         ioc_rule->cmd_len = rule->cmd_len;
3685         ioc_rule->rulenum = rule->rulenum;
3686         ioc_rule->set = rule->set;
3687         ioc_rule->usr_flags = rule->usr_flags;
3688
3689         ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3690         ioc_rule->static_count = static_count;
3691         ioc_rule->static_len = static_ioc_len;
3692
3693         /*
3694          * Visit (read-only) all of the rule's duplications to get
3695          * the necessary statistics
3696          */
3697 #ifdef INVARIANTS
3698         i = 0;
3699 #endif
3700         ioc_rule->pcnt = 0;
3701         ioc_rule->bcnt = 0;
3702         ioc_rule->timestamp = 0;
3703         for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3704                 ioc_rule->pcnt += sibling->pcnt;
3705                 ioc_rule->bcnt += sibling->bcnt;
3706                 if (sibling->timestamp > ioc_rule->timestamp)
3707                         ioc_rule->timestamp = sibling->timestamp;
3708 #ifdef INVARIANTS
3709                 ++i;
3710 #endif
3711         }
3712         KASSERT(i == ncpus, ("static rule is not duplicated on every cpu\n"));
3713
3714         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3715
3716         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3717 }
3718
3719 static void
3720 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3721                 struct ipfw_ioc_state *ioc_state)
3722 {
3723         const struct ipfw_flow_id *id;
3724         struct ipfw_ioc_flowid *ioc_id;
3725
3726         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3727                             0 : dyn_rule->expire - time_second;
3728         ioc_state->pcnt = dyn_rule->pcnt;
3729         ioc_state->bcnt = dyn_rule->bcnt;
3730
3731         ioc_state->dyn_type = dyn_rule->dyn_type;
3732         ioc_state->count = dyn_rule->count;
3733
3734         ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3735
3736         id = &dyn_rule->id;
3737         ioc_id = &ioc_state->id;
3738
3739         ioc_id->type = ETHERTYPE_IP;
3740         ioc_id->u.ip.dst_ip = id->dst_ip;
3741         ioc_id->u.ip.src_ip = id->src_ip;
3742         ioc_id->u.ip.dst_port = id->dst_port;
3743         ioc_id->u.ip.src_port = id->src_port;
3744         ioc_id->u.ip.proto = id->proto;
3745 }
3746
3747 static int
3748 ipfw_ctl_get_rules(struct sockopt *sopt)
3749 {
3750         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3751         struct ip_fw *rule;
3752         void *bp;
3753         size_t size;
3754         uint32_t dcount = 0;
3755
3756         /*
3757          * pass up a copy of the current rules. Static rules
3758          * come first (the last of which has number IPFW_DEFAULT_RULE),
3759          * followed by a possibly empty list of dynamic rule.
3760          */
3761         crit_enter();
3762
3763         size = static_ioc_len;  /* size of static rules */
3764         if (ipfw_dyn_v) {       /* add size of dyn.rules */
3765                 dcount = dyn_count;
3766                 size += dcount * sizeof(struct ipfw_ioc_state);
3767         }
3768
3769         if (sopt->sopt_valsize < size) {
3770                 /* short length, no need to return incomplete rules */
3771                 /* XXX: if superuser, no need to zero buffer */
3772                 bzero(sopt->sopt_val, sopt->sopt_valsize);
3773                 return 0;
3774         }
3775         bp = sopt->sopt_val;
3776
3777         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3778                 bp = ipfw_copy_rule(rule, bp);
3779
3780         if (ipfw_dyn_v && dcount != 0) {
3781                 struct ipfw_ioc_state *ioc_state = bp;
3782                 uint32_t dcount2 = 0;
3783 #ifdef INVARIANTS
3784                 size_t old_size = size;
3785 #endif
3786                 int i;
3787
3788                 lockmgr(&dyn_lock, LK_SHARED);
3789
3790                 /* Check 'ipfw_dyn_v' again with lock held */
3791                 if (ipfw_dyn_v == NULL)
3792                         goto skip;
3793
3794                 for (i = 0; i < curr_dyn_buckets; i++) {
3795                         ipfw_dyn_rule *p;
3796
3797                         /*
3798                          * The # of dynamic rules may have grown after the
3799                          * snapshot of 'dyn_count' was taken, so we will have
3800                          * to check 'dcount' (snapshot of dyn_count) here to
3801                          * make sure that we don't overflow the pre-allocated
3802                          * buffer.
3803                          */
3804                         for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
3805                              p = p->next, ioc_state++, dcount--, dcount2++)
3806                                 ipfw_copy_state(p, ioc_state);
3807                 }
3808 skip:
3809                 lockmgr(&dyn_lock, LK_RELEASE);
3810
3811                 /*
3812                  * The # of dynamic rules may be shrinked after the
3813                  * snapshot of 'dyn_count' was taken.  To give user a
3814                  * correct dynamic rule count, we use the 'dcount2'
3815                  * calculated above (with shared lockmgr lock held).
3816                  */
3817                 size = static_ioc_len +
3818                        (dcount2 * sizeof(struct ipfw_ioc_state));
3819                 KKASSERT(size <= old_size);
3820         }
3821
3822         crit_exit();
3823
3824         sopt->sopt_valsize = size;
3825         return 0;
3826 }
3827
3828 static void
3829 ipfw_set_disable_dispatch(struct netmsg *nmsg)
3830 {
3831         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
3832         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3833
3834         ctx->ipfw_gen++;
3835         ctx->ipfw_set_disable = lmsg->u.ms_result32;
3836
3837         ifnet_forwardmsg(lmsg, mycpuid + 1);
3838 }
3839
3840 static void
3841 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable)
3842 {
3843         struct netmsg nmsg;
3844         struct lwkt_msg *lmsg;
3845         uint32_t set_disable;
3846
3847         /* IPFW_DEFAULT_SET is always enabled */
3848         enable |= (1 << IPFW_DEFAULT_SET);
3849         set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable;
3850
3851         bzero(&nmsg, sizeof(nmsg));
3852         netmsg_init(&nmsg, &curthread->td_msgport, 0, ipfw_set_disable_dispatch);
3853         lmsg = &nmsg.nm_lmsg;
3854         lmsg->u.ms_result32 = set_disable;
3855
3856         ifnet_domsg(lmsg, 0);
3857 }
3858
3859 /**
3860  * {set|get}sockopt parser.
3861  */
3862 static int
3863 ipfw_ctl(struct sockopt *sopt)
3864 {
3865         int error, rulenum;
3866         uint32_t *masks;
3867         size_t size;
3868
3869         error = 0;
3870
3871         switch (sopt->sopt_name) {
3872         case IP_FW_GET:
3873                 error = ipfw_ctl_get_rules(sopt);
3874                 break;
3875
3876         case IP_FW_FLUSH:
3877                 /*
3878                  * Normally we cannot release the lock on each iteration.
3879                  * We could do it here only because we start from the head all
3880                  * the times so there is no risk of missing some entries.
3881                  * On the other hand, the risk is that we end up with
3882                  * a very inconsistent ruleset, so better keep the lock
3883                  * around the whole cycle.
3884                  *
3885                  * XXX this code can be improved by resetting the head of
3886                  * the list to point to the default rule, and then freeing
3887                  * the old list without the need for a lock.
3888                  */
3889
3890                 crit_enter();
3891                 ipfw_flush(0 /* keep default rule */);
3892                 crit_exit();
3893                 break;
3894
3895         case IP_FW_ADD:
3896                 error = ipfw_ctl_add_rule(sopt);
3897                 break;
3898
3899         case IP_FW_DEL:
3900                 /*
3901                  * IP_FW_DEL is used for deleting single rules or sets,
3902                  * and (ab)used to atomically manipulate sets.
3903                  * Argument size is used to distinguish between the two:
3904                  *    sizeof(uint32_t)
3905                  *      delete single rule or set of rules,
3906                  *      or reassign rules (or sets) to a different set.
3907                  *    2 * sizeof(uint32_t)
3908                  *      atomic disable/enable sets.
3909                  *      first uint32_t contains sets to be disabled,
3910                  *      second uint32_t contains sets to be enabled.
3911                  */
3912                 masks = sopt->sopt_val;
3913                 size = sopt->sopt_valsize;
3914                 if (size == sizeof(*masks)) {
3915                         /*
3916                          * Delete or reassign static rule
3917                          */
3918                         error = ipfw_ctl_alter(masks[0]);
3919                 } else if (size == (2 * sizeof(*masks))) {
3920                         /*
3921                          * Set enable/disable
3922                          */
3923                         ipfw_ctl_set_disable(masks[0], masks[1]);
3924                 } else {
3925                         error = EINVAL;
3926                 }
3927                 break;
3928
3929         case IP_FW_ZERO:
3930         case IP_FW_RESETLOG: /* argument is an int, the rule number */
3931                 rulenum = 0;
3932
3933                 if (sopt->sopt_val != 0) {
3934                     error = soopt_to_kbuf(sopt, &rulenum,
3935                             sizeof(int), sizeof(int));
3936                     if (error)
3937                         break;
3938                 }
3939                 error = ipfw_ctl_zero_entry(rulenum,
3940                         sopt->sopt_name == IP_FW_RESETLOG);
3941                 break;
3942
3943         default:
3944                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3945                 error = EINVAL;
3946         }
3947         return error;
3948 }
3949
3950 /*
3951  * This procedure is only used to handle keepalives. It is invoked
3952  * every dyn_keepalive_period
3953  */
3954 static void
3955 ipfw_tick(void *dummy __unused)
3956 {
3957         time_t keep_alive;
3958         uint32_t gen;
3959         int i;
3960
3961         if (ipfw_dyn_v == NULL || dyn_count == 0)
3962                 goto done;
3963
3964         keep_alive = time_second;
3965
3966         lockmgr(&dyn_lock, LK_EXCLUSIVE);
3967 again:
3968         if (ipfw_dyn_v == NULL || dyn_count == 0) {
3969                 lockmgr(&dyn_lock, LK_RELEASE);
3970                 goto done;
3971         }
3972         gen = dyn_buckets_gen;
3973
3974         for (i = 0; i < curr_dyn_buckets; i++) {
3975                 ipfw_dyn_rule *q, *prev;
3976
3977                 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
3978                         uint32_t ack_rev, ack_fwd;
3979                         struct ipfw_flow_id id;
3980
3981                         if (q->dyn_type == O_LIMIT_PARENT)
3982                                 goto next;
3983
3984                         if (TIME_LEQ(q->expire, time_second)) {
3985                                 /* State expired */
3986                                 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
3987                                 continue;
3988                         }
3989
3990                         /*
3991                          * Keep alive processing
3992                          */
3993
3994                         if (!dyn_keepalive)
3995                                 goto next;
3996                         if (q->id.proto != IPPROTO_TCP)
3997                                 goto next;
3998                         if ((q->state & BOTH_SYN) != BOTH_SYN)
3999                                 goto next;
4000                         if (TIME_LEQ(time_second + dyn_keepalive_interval,
4001                             q->expire))
4002                                 goto next;      /* too early */
4003                         if (q->keep_alive == keep_alive)
4004                                 goto next;      /* alreay done */
4005
4006                         /*
4007                          * Save necessary information, so that they could
4008                          * survive after possible blocking in send_pkt()
4009                          */
4010                         id = q->id;
4011                         ack_rev = q->ack_rev;
4012                         ack_fwd = q->ack_fwd;
4013
4014                         /* Sending has been started */
4015                         q->keep_alive = keep_alive;
4016
4017                         /* Release lock to avoid possible dead lock */
4018                         lockmgr(&dyn_lock, LK_RELEASE);
4019                         send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
4020                         send_pkt(&id, ack_fwd - 1, ack_rev, 0);
4021                         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4022
4023                         if (gen != dyn_buckets_gen) {
4024                                 /*
4025                                  * Dyn bucket array has been changed during
4026                                  * the above two sending; reiterate.
4027                                  */
4028                                 goto again;
4029                         }
4030 next:
4031                         prev = q;
4032                         q = q->next;
4033                 }
4034         }
4035         lockmgr(&dyn_lock, LK_RELEASE);
4036 done:
4037         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
4038                       ipfw_tick, NULL);
4039 }
4040
4041 static int
4042 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4043 {
4044         struct ip_fw_args args;
4045         struct mbuf *m = *m0;
4046         struct m_tag *mtag;
4047         int tee = 0, error = 0, i;
4048
4049         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4050                 /* Extract info from dummynet tag */
4051                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4052                 KKASSERT(mtag != NULL);
4053                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4054                 KKASSERT(args.rule != NULL);
4055
4056                 m_tag_delete(m, mtag);
4057                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4058         } else {
4059                 args.rule = NULL;
4060         }
4061
4062         args.eh = NULL;
4063         args.oif = NULL;
4064         args.m = m;
4065         i = ipfw_chk(&args);
4066         m = args.m;
4067
4068         if (m == NULL) {
4069                 error = EACCES;
4070                 goto back;
4071         }
4072
4073         switch (i) {
4074         case IP_FW_PASS:
4075                 break;
4076
4077         case IP_FW_DENY:
4078                 m_freem(m);
4079                 m = NULL;
4080                 error = EACCES;
4081                 break;
4082
4083         case IP_FW_DUMMYNET:
4084                 /* Send packet to the appropriate pipe */
4085                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args);
4086                 break;
4087
4088         case IP_FW_TEE:
4089                 tee = 1;
4090                 /* FALL THROUGH */
4091
4092         case IP_FW_DIVERT:
4093                 if (ip_divert_p != NULL) {
4094                         m = ip_divert_p(m, tee, 1);
4095                 } else {
4096                         m_freem(m);
4097                         m = NULL;
4098                         /* not sure this is the right error msg */
4099                         error = EACCES;
4100                 }
4101                 break;
4102
4103         default:
4104                 panic("unknown ipfw return value: %d\n", i);
4105         }
4106 back:
4107         *m0 = m;
4108         return error;
4109 }
4110
4111 static int
4112 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4113 {
4114         struct ip_fw_args args;
4115         struct mbuf *m = *m0;
4116         struct m_tag *mtag;
4117         int tee = 0, error = 0, off;
4118
4119         if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4120                 /* Extract info from dummynet tag */
4121                 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4122                 KKASSERT(mtag != NULL);
4123                 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4124                 KKASSERT(args.rule != NULL);
4125
4126                 m_tag_delete(m, mtag);
4127                 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4128         } else {
4129                 args.rule = NULL;
4130         }
4131
4132         args.eh = NULL;
4133         args.m = m;
4134         args.oif = ifp;
4135         off = ipfw_chk(&args);
4136         m = args.m;
4137
4138         if (m == NULL) {
4139                 error = EACCES;
4140                 goto back;
4141         }
4142
4143         switch (off) {
4144         case IP_FW_PASS:
4145                 break;
4146
4147         case IP_FW_DENY:
4148                 m_freem(m);
4149                 m = NULL;
4150                 error = EACCES;
4151                 break;
4152
4153         case IP_FW_DUMMYNET:
4154                 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args);
4155                 break;
4156
4157         case IP_FW_TEE:
4158                 tee = 1;
4159                 /* FALL THROUGH */
4160
4161         case IP_FW_DIVERT:
4162                 if (ip_divert_p != NULL) {
4163                         m = ip_divert_p(m, tee, 0);
4164                 } else {
4165                         m_freem(m);
4166                         m = NULL;
4167                         /* not sure this is the right error msg */
4168                         error = EACCES;
4169                 }
4170                 break;
4171
4172         default:
4173                 panic("unknown ipfw return value: %d\n", off);
4174         }
4175 back:
4176         *m0 = m;
4177         return error;
4178 }
4179
4180 static void
4181 ipfw_hook(void)
4182 {
4183         struct pfil_head *pfh;
4184
4185         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4186
4187         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4188         if (pfh == NULL)
4189                 return;
4190
4191         pfil_add_hook(ipfw_check_in, NULL, PFIL_IN, pfh);
4192         pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT, pfh);
4193 }
4194
4195 static void
4196 ipfw_dehook(void)
4197 {
4198         struct pfil_head *pfh;
4199
4200         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4201
4202         pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4203         if (pfh == NULL)
4204                 return;
4205
4206         pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN, pfh);
4207         pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT, pfh);
4208 }
4209
4210 static void
4211 ipfw_sysctl_enable_dispatch(struct netmsg *nmsg)
4212 {
4213         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
4214         int enable = lmsg->u.ms_result;
4215
4216         if (fw_enable == enable)
4217                 goto reply;
4218
4219         fw_enable = enable;
4220         if (fw_enable)
4221                 ipfw_hook();
4222         else
4223                 ipfw_dehook();
4224 reply:
4225         lwkt_replymsg(lmsg, 0);
4226 }
4227
4228 static int
4229 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS)
4230 {
4231         struct netmsg nmsg;
4232         struct lwkt_msg *lmsg;
4233         int enable, error;
4234
4235         enable = fw_enable;
4236         error = sysctl_handle_int(oidp, &enable, 0, req);
4237         if (error || req->newptr == NULL)
4238                 return error;
4239
4240         netmsg_init(&nmsg, &curthread->td_msgport, 0,
4241                     ipfw_sysctl_enable_dispatch);
4242         lmsg = &nmsg.nm_lmsg;
4243         lmsg->u.ms_result = enable;
4244
4245         return lwkt_domsg(IPFW_CFGPORT, lmsg, 0);
4246 }
4247
4248 static int
4249 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
4250 {
4251         return sysctl_int_range(oidp, arg1, arg2, req,
4252                IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
4253 }
4254
4255 static int
4256 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
4257 {
4258         int error, value;
4259
4260         lockmgr(&dyn_lock, LK_EXCLUSIVE);
4261
4262         value = dyn_buckets;
4263         error = sysctl_handle_int(oidp, &value, 0, req);
4264         if (error || !req->newptr)
4265                 goto back;
4266
4267         /*
4268          * Make sure we have a power of 2 and
4269          * do not allow more than 64k entries.
4270          */
4271         error = EINVAL;
4272         if (value <= 1 || value > 65536)
4273                 goto back;
4274         if ((value & (value - 1)) != 0)
4275                 goto back;
4276
4277         error = 0;
4278         dyn_buckets = value;
4279 back:
4280         lockmgr(&dyn_lock, LK_RELEASE);
4281         return error;
4282 }
4283
4284 static int
4285 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
4286 {
4287         return sysctl_int_range(oidp, arg1, arg2, req,
4288                                 1, dyn_keepalive_period - 1);
4289 }
4290
4291 static int
4292 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
4293 {
4294         return sysctl_int_range(oidp, arg1, arg2, req,
4295                                 1, dyn_keepalive_period - 1);
4296 }
4297
4298 static void
4299 ipfw_ctx_init_dispatch(struct netmsg *nmsg)
4300 {
4301         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
4302         struct ipfw_context *ctx;
4303         struct ip_fw *def_rule;
4304
4305         ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO);
4306         ipfw_ctx[mycpuid] = ctx;
4307
4308         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
4309
4310         def_rule->act_ofs = 0;
4311         def_rule->rulenum = IPFW_DEFAULT_RULE;
4312         def_rule->cmd_len = 1;
4313         def_rule->set = IPFW_DEFAULT_SET;
4314
4315         def_rule->cmd[0].len = 1;
4316 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4317         def_rule->cmd[0].opcode = O_ACCEPT;
4318 #else
4319         def_rule->cmd[0].opcode = O_DENY;
4320 #endif
4321
4322         def_rule->refcnt = 1;
4323         def_rule->cpuid = mycpuid;
4324
4325         /* Install the default rule */
4326         ctx->ipfw_default_rule = def_rule;
4327         ctx->ipfw_layer3_chain = def_rule;
4328
4329         /* Link rule CPU sibling */
4330         ipfw_link_sibling(fwmsg, def_rule);
4331
4332         /* Statistics only need to be updated once */
4333         if (mycpuid == 0)
4334                 ipfw_inc_static_count(def_rule);
4335
4336         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
4337 }
4338
4339 static void
4340 ipfw_init_dispatch(struct netmsg *nmsg)
4341 {
4342         struct netmsg_ipfw fwmsg;
4343         int error = 0;
4344
4345         crit_enter();
4346
4347         if (IPFW_LOADED) {
4348                 kprintf("IP firewall already loaded\n");
4349                 error = EEXIST;
4350                 goto reply;
4351         }
4352
4353         bzero(&fwmsg, sizeof(fwmsg));
4354         netmsg_init(&fwmsg.nmsg, &curthread->td_msgport, 0,
4355                     ipfw_ctx_init_dispatch);
4356         ifnet_domsg(&fwmsg.nmsg.nm_lmsg, 0);
4357
4358         ip_fw_chk_ptr = ipfw_chk;
4359         ip_fw_ctl_ptr = ipfw_ctl;
4360         ip_fw_dn_io_ptr = ipfw_dummynet_io;
4361
4362         kprintf("ipfw2 initialized, default to %s, logging ",
4363                 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode ==
4364                 O_ACCEPT ? "accept" : "deny");
4365
4366 #ifdef IPFIREWALL_VERBOSE
4367         fw_verbose = 1;
4368 #endif
4369 #ifdef IPFIREWALL_VERBOSE_LIMIT
4370         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4371 #endif
4372         if (fw_verbose == 0) {
4373                 kprintf("disabled\n");
4374         } else if (verbose_limit == 0) {
4375                 kprintf("unlimited\n");
4376         } else {
4377                 kprintf("limited to %d packets/entry by default\n",
4378                         verbose_limit);
4379         }
4380
4381         callout_init(&ipfw_timeout_h);
4382         lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
4383
4384         ip_fw_loaded = 1;
4385         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
4386
4387         if (fw_enable)
4388                 ipfw_hook();
4389 reply:
4390         crit_exit();
4391         lwkt_replymsg(&nmsg->nm_lmsg, error);
4392 }
4393
4394 static int
4395 ipfw_init(void)
4396 {
4397         struct netmsg smsg;
4398
4399         netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_init_dispatch);
4400         return lwkt_domsg(IPFW_CFGPORT, &smsg.nm_lmsg, 0);
4401 }
4402
4403 #ifdef KLD_MODULE
4404
4405 static void
4406 ipfw_fini_dispatch(struct netmsg *nmsg)
4407 {
4408         int error = 0, cpu;
4409
4410         crit_enter();
4411
4412         if (ipfw_refcnt != 0) {
4413                 error = EBUSY;
4414                 goto reply;
4415         }
4416
4417         ipfw_dehook();
4418
4419         callout_stop(&ipfw_timeout_h);
4420
4421         ip_fw_loaded = 0;
4422         netmsg_service_sync();
4423
4424         ip_fw_chk_ptr = NULL;
4425         ip_fw_ctl_ptr = NULL;
4426         ip_fw_dn_io_ptr = NULL;
4427         ipfw_flush(1 /* kill default rule */);
4428
4429         /* Free pre-cpu context */
4430         for (cpu = 0; cpu < ncpus; ++cpu)
4431                 kfree(ipfw_ctx[cpu], M_IPFW);
4432
4433         kprintf("IP firewall unloaded\n");
4434 reply:
4435         crit_exit();
4436         lwkt_replymsg(&nmsg->nm_lmsg, error);
4437 }
4438
4439 static int
4440 ipfw_fini(void)
4441 {
4442         struct netmsg smsg;
4443
4444         netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_fini_dispatch);
4445         return lwkt_domsg(IPFW_CFGPORT, &smsg.nm_lmsg, 0);
4446 }
4447
4448 #endif  /* KLD_MODULE */
4449
4450 static int
4451 ipfw_modevent(module_t mod, int type, void *unused)
4452 {
4453         int err = 0;
4454
4455         switch (type) {
4456         case MOD_LOAD:
4457                 err = ipfw_init();
4458                 break;
4459
4460         case MOD_UNLOAD:
4461 #ifndef KLD_MODULE
4462                 kprintf("ipfw statically compiled, cannot unload\n");
4463                 err = EBUSY;
4464 #else
4465                 err = ipfw_fini();
4466 #endif
4467                 break;
4468         default:
4469                 break;
4470         }
4471         return err;
4472 }
4473
4474 static moduledata_t ipfwmod = {
4475         "ipfw",
4476         ipfw_modevent,
4477         0
4478 };
4479 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
4480 MODULE_VERSION(ipfw, 1);