2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
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.
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
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 1.83 2008/09/13 12:57:07 sephe Exp $
33 * Implement IP packet firewall (new version)
39 #error IPFIREWALL requires INET.
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/malloc.h>
46 #include <sys/kernel.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>
58 #include <net/route.h>
59 #include <net/netmsg2.h>
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>
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>
79 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
82 * Description about per-CPU rule duplication:
84 * Module loading/unloading and all ioctl operations are serialized
85 * by netisr0, so we don't have any ordering or locking problems.
87 * Following graph shows how operation on per-CPU rule list is
88 * performed [2 CPU case]:
92 * netisr0 <------------------------------------+
103 * forwardmsg---------->ifnet1 |
108 * replymsg--------------+
113 * Rules which will not create states (dyn rules) [2 CPU case]
116 * layer3_chain layer3_chain
119 * +-------+ sibling +-------+ sibling
120 * | rule1 |--------->| rule1 |--------->NULL
121 * +-------+ +-------+
125 * +-------+ sibling +-------+ sibling
126 * | rule2 |--------->| rule2 |--------->NULL
127 * +-------+ +-------+
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
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
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
148 * Rules which will create states (dyn rules) [2 CPU case]
149 * (unnecessary parts are omitted; they are same as in the previous figure)
153 * +-------+ +-------+
154 * | rule1 | | rule1 |
155 * +-------+ +-------+
162 * | +--------------------+ |
164 * | | (read-only shared) | |
166 * | | back pointer array | |
167 * | | (indexed by cpuid) | |
169 * +----|---------[0] | |
170 * | [1]--------|----+
172 * +--------------------+
175 * ........|............|............
179 * : +---------+ +---------+ :
180 * : | state1a | | state1b | .... :
181 * : +---------+ +---------+ :
185 * : (protected by dyn_lock) :
186 * ..................................
188 * [state1a and state1b are states created by rule1]
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.
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.
209 #define IPFW_AUTOINC_STEP_MIN 1
210 #define IPFW_AUTOINC_STEP_MAX 1000
211 #define IPFW_AUTOINC_STEP_DEF 100
213 #define IPFW_DEFAULT_RULE 65535 /* rulenum for the default rule */
214 #define IPFW_DEFAULT_SET 31 /* set number for the default rule */
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;
227 struct ip_fw *start_rule;
228 struct ip_fw *prev_rule;
236 struct ip_fw *start_rule;
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) */
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.
252 uint32_t ipfw_set_disable;
253 uint32_t ipfw_gen; /* generation of rule list */
256 static struct ipfw_context *ipfw_ctx[MAXCPU];
260 * Module can not be unloaded, if there are references to
261 * certains rules of ipfw(4), e.g. dummynet(4)
263 static int ipfw_refcnt;
266 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
269 * Following two global variables are accessed and
270 * updated only on CPU0
272 static uint32_t static_count; /* # of static rules */
273 static uint32_t static_ioc_len; /* bytes of static rules */
276 * If 1, then ipfw static rules are being flushed,
277 * ipfw_chk() will skip to the default rule.
279 static int ipfw_flushing;
281 static int fw_verbose;
282 static int verbose_limit;
284 static int fw_debug = 1;
285 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
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);
294 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
295 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
296 &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw");
297 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
298 &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
299 "Rule number autincrement step");
300 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
302 "Only do a single pass through ipfw when using dummynet(4)");
303 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
304 &fw_debug, 0, "Enable printing of debug ip_fw statements");
305 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
306 &fw_verbose, 0, "Log matches to ipfw rules");
307 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
308 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
311 * Description of dynamic rules.
313 * Dynamic rules are stored in lists accessed through a hash table
314 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
315 * be modified through the sysctl variable dyn_buckets which is
316 * updated when the table becomes empty.
318 * XXX currently there is only one list, ipfw_dyn.
320 * When a packet is received, its address fields are first masked
321 * with the mask defined for the rule, then hashed, then matched
322 * against the entries in the corresponding list.
323 * Dynamic rules can be used for different purposes:
325 * + enforcing limits on the number of sessions;
326 * + in-kernel NAT (not implemented yet)
328 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
329 * measured in seconds and depending on the flags.
331 * The total number of dynamic rules is stored in dyn_count.
332 * The max number of dynamic rules is dyn_max. When we reach
333 * the maximum number of rules we do not create anymore. This is
334 * done to avoid consuming too much memory, but also too much
335 * time when searching on each packet (ideally, we should try instead
336 * to put a limit on the length of the list on each bucket...).
338 * Each dynamic rule holds a pointer to the parent ipfw rule so
339 * we know what action to perform. Dynamic rules are removed when
340 * the parent rule is deleted. XXX we should make them survive.
342 * There are some limitations with dynamic rules -- we do not
343 * obey the 'randomized match', and we do not do multiple
344 * passes through the firewall. XXX check the latter!!!
346 * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
347 * Only TCP state transition will change dynamic rule's state and ack
348 * sequences, while all packets of one TCP connection only goes through
349 * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
350 * rule looking up. The keep alive callout uses exclusive lockmgr lock
351 * when it tries to find suitable dynamic rules to send keep alive, so
352 * it will not see half updated state and ack sequences. Though the expire
353 * field updating looks racy for other protocols, the resolution (second)
354 * of expire field makes this kind of race harmless.
355 * XXX statistics' updating is _not_ MPsafe!!!
356 * XXX once UDP output path is fixed, we could use lockless dynamic rule
359 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
360 static uint32_t dyn_buckets = 256; /* must be power of 2 */
361 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
362 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
363 static struct lock dyn_lock; /* dynamic rules' hash table lock */
364 static struct callout ipfw_timeout_h;
367 * Timeouts for various events in handing dynamic rules.
369 static uint32_t dyn_ack_lifetime = 300;
370 static uint32_t dyn_syn_lifetime = 20;
371 static uint32_t dyn_fin_lifetime = 1;
372 static uint32_t dyn_rst_lifetime = 1;
373 static uint32_t dyn_udp_lifetime = 10;
374 static uint32_t dyn_short_lifetime = 5;
377 * Keepalives are sent if dyn_keepalive is set. They are sent every
378 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
379 * seconds of lifetime of a rule.
380 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
381 * than dyn_keepalive_period.
384 static uint32_t dyn_keepalive_interval = 20;
385 static uint32_t dyn_keepalive_period = 5;
386 static uint32_t dyn_keepalive = 1; /* do send keepalives */
388 static uint32_t dyn_count; /* # of dynamic rules */
389 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
391 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
392 &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
393 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
394 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
395 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
396 &dyn_count, 0, "Number of dyn. rules");
397 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
398 &dyn_max, 0, "Max number of dyn. rules");
399 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
400 &static_count, 0, "Number of static rules");
401 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
402 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
403 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
404 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
405 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
406 CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
407 "Lifetime of dyn. rules for fin");
408 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
409 CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
410 "Lifetime of dyn. rules for rst");
411 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
412 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
413 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
414 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
415 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
416 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
418 #endif /* SYSCTL_NODE */
420 static ip_fw_chk_t ipfw_chk;
423 ipfw_free_rule(struct ip_fw *rule)
425 KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d\n", mycpuid));
426 KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
428 if (rule->refcnt == 0) {
436 ipfw_unref_rule(void *priv)
438 ipfw_free_rule(priv);
440 atomic_subtract_int(&ipfw_refcnt, 1);
445 ipfw_ref_rule(struct ip_fw *rule)
447 KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d\n", mycpuid));
449 atomic_add_int(&ipfw_refcnt, 1);
455 * This macro maps an ip pointer into a layer3 header pointer of type T
457 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
460 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
462 int type = L3HDR(struct icmp,ip)->icmp_type;
464 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
467 #define TT ((1 << ICMP_ECHO) | \
468 (1 << ICMP_ROUTERSOLICIT) | \
469 (1 << ICMP_TSTAMP) | \
474 is_icmp_query(struct ip *ip)
476 int type = L3HDR(struct icmp, ip)->icmp_type;
478 return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
484 * The following checks use two arrays of 8 or 16 bits to store the
485 * bits that we want set or clear, respectively. They are in the
486 * low and high half of cmd->arg1 or cmd->d[0].
488 * We scan options and store the bits we find set. We succeed if
490 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
492 * The code is sometimes optimized not to store additional variables.
496 flags_match(ipfw_insn *cmd, uint8_t bits)
501 if (((cmd->arg1 & 0xff) & bits) != 0)
502 return 0; /* some bits we want set were clear */
504 want_clear = (cmd->arg1 >> 8) & 0xff;
505 if ((want_clear & bits) != want_clear)
506 return 0; /* some bits we want clear were set */
511 ipopts_match(struct ip *ip, ipfw_insn *cmd)
513 int optlen, bits = 0;
514 u_char *cp = (u_char *)(ip + 1);
515 int x = (ip->ip_hl << 2) - sizeof(struct ip);
517 for (; x > 0; x -= optlen, cp += optlen) {
518 int opt = cp[IPOPT_OPTVAL];
520 if (opt == IPOPT_EOL)
523 if (opt == IPOPT_NOP) {
526 optlen = cp[IPOPT_OLEN];
527 if (optlen <= 0 || optlen > x)
528 return 0; /* invalid or truncated */
533 bits |= IP_FW_IPOPT_LSRR;
537 bits |= IP_FW_IPOPT_SSRR;
541 bits |= IP_FW_IPOPT_RR;
545 bits |= IP_FW_IPOPT_TS;
552 return (flags_match(cmd, bits));
556 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
558 int optlen, bits = 0;
559 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
560 u_char *cp = (u_char *)(tcp + 1);
561 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
563 for (; x > 0; x -= optlen, cp += optlen) {
566 if (opt == TCPOPT_EOL)
569 if (opt == TCPOPT_NOP) {
579 bits |= IP_FW_TCPOPT_MSS;
583 bits |= IP_FW_TCPOPT_WINDOW;
586 case TCPOPT_SACK_PERMITTED:
588 bits |= IP_FW_TCPOPT_SACK;
591 case TCPOPT_TIMESTAMP:
592 bits |= IP_FW_TCPOPT_TS;
598 bits |= IP_FW_TCPOPT_CC;
605 return (flags_match(cmd, bits));
609 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
611 if (ifp == NULL) /* no iface with this packet, match fails */
614 /* Check by name or by IP address */
615 if (cmd->name[0] != '\0') { /* match by name */
618 if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
621 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
625 struct ifaddr_container *ifac;
627 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
628 struct ifaddr *ia = ifac->ifa;
630 if (ia->ifa_addr == NULL)
632 if (ia->ifa_addr->sa_family != AF_INET)
634 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
635 (ia->ifa_addr))->sin_addr.s_addr)
636 return(1); /* match */
639 return(0); /* no match, fail ... */
642 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
645 * We enter here when we have a rule with O_LOG.
646 * XXX this function alone takes about 2Kbytes of code!
649 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
650 struct mbuf *m, struct ifnet *oif)
653 int limit_reached = 0;
654 char action2[40], proto[48], fragment[28];
659 if (f == NULL) { /* bogus pkt */
660 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
662 if (verbose_limit != 0 &&
663 ctx->ipfw_norule_counter >= verbose_limit)
665 ctx->ipfw_norule_counter++;
666 if (ctx->ipfw_norule_counter == verbose_limit)
667 limit_reached = verbose_limit;
669 } else { /* O_LOG is the first action, find the real one */
670 ipfw_insn *cmd = ACTION_PTR(f);
671 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
673 if (l->max_log != 0 && l->log_left == 0)
676 if (l->log_left == 0)
677 limit_reached = l->max_log;
678 cmd += F_LEN(cmd); /* point to first action */
679 if (cmd->opcode == O_PROB)
683 switch (cmd->opcode) {
689 if (cmd->arg1==ICMP_REJECT_RST) {
691 } else if (cmd->arg1==ICMP_UNREACH_HOST) {
694 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
708 ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
712 ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
716 ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
720 ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
724 ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
729 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
732 len = ksnprintf(SNPARGS(action2, 0),
734 inet_ntoa(sa->sa.sin_addr));
735 if (sa->sa.sin_port) {
736 ksnprintf(SNPARGS(action2, len), ":%d",
748 if (hlen == 0) { /* non-ip */
749 ksnprintf(SNPARGS(proto, 0), "MAC");
751 struct ip *ip = mtod(m, struct ip *);
752 /* these three are all aliases to the same thing */
753 struct icmp *const icmp = L3HDR(struct icmp, ip);
754 struct tcphdr *const tcp = (struct tcphdr *)icmp;
755 struct udphdr *const udp = (struct udphdr *)icmp;
757 int ip_off, offset, ip_len;
760 if (eh != NULL) { /* layer 2 packets are as on the wire */
761 ip_off = ntohs(ip->ip_off);
762 ip_len = ntohs(ip->ip_len);
767 offset = ip_off & IP_OFFMASK;
770 len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
771 inet_ntoa(ip->ip_src));
773 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
774 ntohs(tcp->th_sport),
775 inet_ntoa(ip->ip_dst),
776 ntohs(tcp->th_dport));
778 ksnprintf(SNPARGS(proto, len), " %s",
779 inet_ntoa(ip->ip_dst));
784 len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
785 inet_ntoa(ip->ip_src));
787 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
788 ntohs(udp->uh_sport),
789 inet_ntoa(ip->ip_dst),
790 ntohs(udp->uh_dport));
792 ksnprintf(SNPARGS(proto, len), " %s",
793 inet_ntoa(ip->ip_dst));
799 len = ksnprintf(SNPARGS(proto, 0),
804 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
806 len += ksnprintf(SNPARGS(proto, len), "%s",
807 inet_ntoa(ip->ip_src));
808 ksnprintf(SNPARGS(proto, len), " %s",
809 inet_ntoa(ip->ip_dst));
813 len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
814 inet_ntoa(ip->ip_src));
815 ksnprintf(SNPARGS(proto, len), " %s",
816 inet_ntoa(ip->ip_dst));
820 if (ip_off & (IP_MF | IP_OFFMASK)) {
821 ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
822 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
823 offset << 3, (ip_off & IP_MF) ? "+" : "");
827 if (oif || m->m_pkthdr.rcvif) {
828 log(LOG_SECURITY | LOG_INFO,
829 "ipfw: %d %s %s %s via %s%s\n",
831 action, proto, oif ? "out" : "in",
832 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
835 log(LOG_SECURITY | LOG_INFO,
836 "ipfw: %d %s %s [no if info]%s\n",
838 action, proto, fragment);
842 log(LOG_SECURITY | LOG_NOTICE,
843 "ipfw: limit %d reached on entry %d\n",
844 limit_reached, f ? f->rulenum : -1);
851 * IMPORTANT: the hash function for dynamic rules must be commutative
852 * in source and destination (ip,port), because rules are bidirectional
853 * and we want to find both in the same bucket.
856 hash_packet(struct ipfw_flow_id *id)
860 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
861 i &= (curr_dyn_buckets - 1);
866 * unlink a dynamic rule from a chain. prev is a pointer to
867 * the previous one, q is a pointer to the rule to delete,
868 * head is a pointer to the head of the queue.
869 * Modifies q and potentially also head.
871 #define UNLINK_DYN_RULE(prev, head, q) \
873 ipfw_dyn_rule *old_q = q; \
875 /* remove a refcount to the parent */ \
876 if (q->dyn_type == O_LIMIT) \
877 q->parent->count--; \
878 DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \
879 (q->id.src_ip), (q->id.src_port), \
880 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
882 prev->next = q = q->next; \
884 head = q = q->next; \
885 KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count)); \
887 kfree(old_q, M_IPFW); \
890 #define TIME_LEQ(a, b) ((int)((a) - (b)) <= 0)
893 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
895 * If keep_me == NULL, rules are deleted even if not expired,
896 * otherwise only expired rules are removed.
898 * The value of the second parameter is also used to point to identify
899 * a rule we absolutely do not want to remove (e.g. because we are
900 * holding a reference to it -- this is the case with O_LIMIT_PARENT
901 * rules). The pointer is only used for comparison, so any non-null
905 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
907 static uint32_t last_remove = 0; /* XXX */
909 #define FORCE (keep_me == NULL)
911 ipfw_dyn_rule *prev, *q;
912 int i, pass = 0, max_pass = 0, unlinked = 0;
914 if (ipfw_dyn_v == NULL || dyn_count == 0)
916 /* do not expire more than once per second, it is useless */
917 if (!FORCE && last_remove == time_second)
919 last_remove = time_second;
922 * because O_LIMIT refer to parent rules, during the first pass only
923 * remove child and mark any pending LIMIT_PARENT, and remove
924 * them in a second pass.
927 for (i = 0; i < curr_dyn_buckets; i++) {
928 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
930 * Logic can become complex here, so we split tests.
934 if (rule != NULL && rule->stub != q->stub)
935 goto next; /* not the one we are looking for */
936 if (q->dyn_type == O_LIMIT_PARENT) {
938 * handle parent in the second pass,
939 * record we need one.
944 if (FORCE && q->count != 0) {
945 /* XXX should not happen! */
946 kprintf("OUCH! cannot remove rule, "
947 "count %d\n", q->count);
950 if (!FORCE && !TIME_LEQ(q->expire, time_second))
954 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
961 if (pass++ < max_pass)
971 * lookup a dynamic rule.
973 static ipfw_dyn_rule *
974 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
978 * stateful ipfw extensions.
979 * Lookup into dynamic session queue
981 #define MATCH_REVERSE 0
982 #define MATCH_FORWARD 1
984 #define MATCH_UNKNOWN 3
985 int i, dir = MATCH_NONE;
986 ipfw_dyn_rule *prev, *q=NULL;
988 if (ipfw_dyn_v == NULL)
989 goto done; /* not found */
991 i = hash_packet(pkt);
992 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
993 if (q->dyn_type == O_LIMIT_PARENT)
996 if (TIME_LEQ(q->expire, time_second)) {
998 * Entry expired; skip.
999 * Let ipfw_tick() take care of it
1004 if (pkt->proto == q->id.proto) {
1005 if (pkt->src_ip == q->id.src_ip &&
1006 pkt->dst_ip == q->id.dst_ip &&
1007 pkt->src_port == q->id.src_port &&
1008 pkt->dst_port == q->id.dst_port) {
1009 dir = MATCH_FORWARD;
1012 if (pkt->src_ip == q->id.dst_ip &&
1013 pkt->dst_ip == q->id.src_ip &&
1014 pkt->src_port == q->id.dst_port &&
1015 pkt->dst_port == q->id.src_port) {
1016 dir = MATCH_REVERSE;
1025 goto done; /* q = NULL, not found */
1027 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1028 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1030 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1031 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1033 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1035 case TH_SYN: /* opening */
1036 q->expire = time_second + dyn_syn_lifetime;
1039 case BOTH_SYN: /* move to established */
1040 case BOTH_SYN | TH_FIN : /* one side tries to close */
1041 case BOTH_SYN | (TH_FIN << 8) :
1043 uint32_t ack = ntohl(tcp->th_ack);
1045 #define _SEQ_GE(a, b) ((int)(a) - (int)(b) >= 0)
1047 if (dir == MATCH_FORWARD) {
1048 if (q->ack_fwd == 0 ||
1049 _SEQ_GE(ack, q->ack_fwd))
1051 else /* ignore out-of-sequence */
1054 if (q->ack_rev == 0 ||
1055 _SEQ_GE(ack, q->ack_rev))
1057 else /* ignore out-of-sequence */
1062 q->expire = time_second + dyn_ack_lifetime;
1065 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1066 KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
1067 q->expire = time_second + dyn_fin_lifetime;
1073 * reset or some invalid combination, but can also
1074 * occur if we use keep-state the wrong way.
1076 if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
1077 kprintf("invalid state: 0x%x\n", q->state);
1079 KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
1080 q->expire = time_second + dyn_rst_lifetime;
1083 } else if (pkt->proto == IPPROTO_UDP) {
1084 q->expire = time_second + dyn_udp_lifetime;
1086 /* other protocols */
1087 q->expire = time_second + dyn_short_lifetime;
1090 if (match_direction)
1091 *match_direction = dir;
1095 static struct ip_fw *
1096 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
1097 uint16_t len, int *deny)
1099 struct ip_fw *rule = NULL;
1101 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1105 gen = ctx->ipfw_gen;
1107 lockmgr(&dyn_lock, LK_SHARED);
1109 if (ctx->ipfw_gen != gen) {
1111 * Static rules had been change when we were waiting
1112 * for the dynamic hash table lock; deny this packet,
1113 * since it is _not_ known whether it is safe to keep
1114 * iterating the static rules.
1120 q = lookup_dyn_rule(pkt, match_direction, tcp);
1124 rule = q->stub->rule[mycpuid];
1125 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid);
1132 lockmgr(&dyn_lock, LK_RELEASE);
1137 realloc_dynamic_table(void)
1139 ipfw_dyn_rule **old_dyn_v;
1140 uint32_t old_curr_dyn_buckets;
1142 KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
1143 ("invalid dyn_buckets %d\n", dyn_buckets));
1145 /* Save the current buckets array for later error recovery */
1146 old_dyn_v = ipfw_dyn_v;
1147 old_curr_dyn_buckets = curr_dyn_buckets;
1149 curr_dyn_buckets = dyn_buckets;
1151 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1152 M_IPFW, M_NOWAIT | M_ZERO);
1153 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1156 curr_dyn_buckets /= 2;
1157 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
1158 old_dyn_v != NULL) {
1160 * Don't try allocating smaller buckets array, reuse
1161 * the old one, which alreay contains enough buckets
1167 if (ipfw_dyn_v != NULL) {
1168 if (old_dyn_v != NULL)
1169 kfree(old_dyn_v, M_IPFW);
1171 /* Allocation failed, restore old buckets array */
1172 ipfw_dyn_v = old_dyn_v;
1173 curr_dyn_buckets = old_curr_dyn_buckets;
1176 if (ipfw_dyn_v != NULL)
1181 * Install state of type 'type' for a dynamic session.
1182 * The hash table contains two type of rules:
1183 * - regular rules (O_KEEP_STATE)
1184 * - rules for sessions with limited number of sess per user
1185 * (O_LIMIT). When they are created, the parent is
1186 * increased by 1, and decreased on delete. In this case,
1187 * the third parameter is the parent rule and not the chain.
1188 * - "parent" rules for the above (O_LIMIT_PARENT).
1190 static ipfw_dyn_rule *
1191 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1196 if (ipfw_dyn_v == NULL ||
1197 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1198 realloc_dynamic_table();
1199 if (ipfw_dyn_v == NULL)
1200 return NULL; /* failed ! */
1202 i = hash_packet(id);
1204 r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1206 kprintf ("sorry cannot allocate state\n");
1210 /* increase refcount on parent, and set pointer */
1211 if (dyn_type == O_LIMIT) {
1212 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1214 if (parent->dyn_type != O_LIMIT_PARENT)
1215 panic("invalid parent");
1218 rule = parent->stub->rule[mycpuid];
1219 KKASSERT(rule->stub == parent->stub);
1221 KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1224 r->expire = time_second + dyn_syn_lifetime;
1225 r->stub = rule->stub;
1226 r->dyn_type = dyn_type;
1227 r->pcnt = r->bcnt = 0;
1231 r->next = ipfw_dyn_v[i];
1235 DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1237 (r->id.src_ip), (r->id.src_port),
1238 (r->id.dst_ip), (r->id.dst_port),
1244 * lookup dynamic parent rule using pkt and rule as search keys.
1245 * If the lookup fails, then install one.
1247 static ipfw_dyn_rule *
1248 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1254 i = hash_packet(pkt);
1255 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1256 if (q->dyn_type == O_LIMIT_PARENT &&
1257 rule->stub == q->stub &&
1258 pkt->proto == q->id.proto &&
1259 pkt->src_ip == q->id.src_ip &&
1260 pkt->dst_ip == q->id.dst_ip &&
1261 pkt->src_port == q->id.src_port &&
1262 pkt->dst_port == q->id.dst_port) {
1263 q->expire = time_second + dyn_short_lifetime;
1264 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
1269 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1273 * Install dynamic state for rule type cmd->o.opcode
1275 * Returns 1 (failure) if state is not installed because of errors or because
1276 * session limitations are enforced.
1279 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1280 struct ip_fw_args *args)
1282 static int last_log; /* XXX */
1286 DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1288 (args->f_id.src_ip), (args->f_id.src_port),
1289 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1291 q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1292 if (q != NULL) { /* should never occur */
1293 if (last_log != time_second) {
1294 last_log = time_second;
1295 kprintf(" install_state: entry already present, done\n");
1300 if (dyn_count >= dyn_max) {
1302 * Run out of slots, try to remove any expired rule.
1304 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1305 if (dyn_count >= dyn_max) {
1306 if (last_log != time_second) {
1307 last_log = time_second;
1308 kprintf("install_state: "
1309 "Too many dynamic rules\n");
1311 return 1; /* cannot install, notify caller */
1315 switch (cmd->o.opcode) {
1316 case O_KEEP_STATE: /* bidir rule */
1317 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1321 case O_LIMIT: /* limit number of sessions */
1323 uint16_t limit_mask = cmd->limit_mask;
1324 struct ipfw_flow_id id;
1325 ipfw_dyn_rule *parent;
1327 DEB(kprintf("installing dyn-limit rule %d\n",
1330 id.dst_ip = id.src_ip = 0;
1331 id.dst_port = id.src_port = 0;
1332 id.proto = args->f_id.proto;
1334 if (limit_mask & DYN_SRC_ADDR)
1335 id.src_ip = args->f_id.src_ip;
1336 if (limit_mask & DYN_DST_ADDR)
1337 id.dst_ip = args->f_id.dst_ip;
1338 if (limit_mask & DYN_SRC_PORT)
1339 id.src_port = args->f_id.src_port;
1340 if (limit_mask & DYN_DST_PORT)
1341 id.dst_port = args->f_id.dst_port;
1343 parent = lookup_dyn_parent(&id, rule);
1344 if (parent == NULL) {
1345 kprintf("add parent failed\n");
1349 if (parent->count >= cmd->conn_limit) {
1351 * See if we can remove some expired rule.
1353 remove_dyn_rule_locked(rule, parent);
1354 if (parent->count >= cmd->conn_limit) {
1356 last_log != time_second) {
1357 last_log = time_second;
1358 log(LOG_SECURITY | LOG_DEBUG,
1360 "too many entries\n");
1365 if (add_dyn_rule(&args->f_id, O_LIMIT,
1366 (struct ip_fw *)parent) == NULL)
1371 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1374 lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1379 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1380 struct ip_fw_args *args, int *deny)
1382 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1387 gen = ctx->ipfw_gen;
1389 lockmgr(&dyn_lock, LK_EXCLUSIVE);
1390 if (ctx->ipfw_gen != gen) {
1391 /* See the comment in lookup_rule() */
1394 ret = install_state_locked(rule, cmd, args);
1396 lockmgr(&dyn_lock, LK_RELEASE);
1402 * Transmit a TCP packet, containing either a RST or a keepalive.
1403 * When flags & TH_RST, we are sending a RST packet, because of a
1404 * "reset" action matched the packet.
1405 * Otherwise we are sending a keepalive, and flags & TH_
1408 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1413 struct route sro; /* fake route */
1415 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1418 m->m_pkthdr.rcvif = NULL;
1419 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1420 m->m_data += max_linkhdr;
1422 ip = mtod(m, struct ip *);
1423 bzero(ip, m->m_len);
1424 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1425 ip->ip_p = IPPROTO_TCP;
1429 * Assume we are sending a RST (or a keepalive in the reverse
1430 * direction), swap src and destination addresses and ports.
1432 ip->ip_src.s_addr = htonl(id->dst_ip);
1433 ip->ip_dst.s_addr = htonl(id->src_ip);
1434 tcp->th_sport = htons(id->dst_port);
1435 tcp->th_dport = htons(id->src_port);
1436 if (flags & TH_RST) { /* we are sending a RST */
1437 if (flags & TH_ACK) {
1438 tcp->th_seq = htonl(ack);
1439 tcp->th_ack = htonl(0);
1440 tcp->th_flags = TH_RST;
1444 tcp->th_seq = htonl(0);
1445 tcp->th_ack = htonl(seq);
1446 tcp->th_flags = TH_RST | TH_ACK;
1450 * We are sending a keepalive. flags & TH_SYN determines
1451 * the direction, forward if set, reverse if clear.
1452 * NOTE: seq and ack are always assumed to be correct
1453 * as set by the caller. This may be confusing...
1455 if (flags & TH_SYN) {
1457 * we have to rewrite the correct addresses!
1459 ip->ip_dst.s_addr = htonl(id->dst_ip);
1460 ip->ip_src.s_addr = htonl(id->src_ip);
1461 tcp->th_dport = htons(id->dst_port);
1462 tcp->th_sport = htons(id->src_port);
1464 tcp->th_seq = htonl(seq);
1465 tcp->th_ack = htonl(ack);
1466 tcp->th_flags = TH_ACK;
1470 * set ip_len to the payload size so we can compute
1471 * the tcp checksum on the pseudoheader
1472 * XXX check this, could save a couple of words ?
1474 ip->ip_len = htons(sizeof(struct tcphdr));
1475 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1478 * now fill fields left out earlier
1480 ip->ip_ttl = ip_defttl;
1481 ip->ip_len = m->m_pkthdr.len;
1483 bzero(&sro, sizeof(sro));
1484 ip_rtaddr(ip->ip_dst, &sro);
1486 m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1487 ip_output(m, NULL, &sro, 0, NULL, NULL);
1493 * sends a reject message, consuming the mbuf passed as an argument.
1496 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1498 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1499 /* We need the IP header in host order for icmp_error(). */
1500 if (args->eh != NULL) {
1501 struct ip *ip = mtod(args->m, struct ip *);
1503 ip->ip_len = ntohs(ip->ip_len);
1504 ip->ip_off = ntohs(ip->ip_off);
1506 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1507 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1508 struct tcphdr *const tcp =
1509 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1511 if ((tcp->th_flags & TH_RST) == 0) {
1512 send_pkt(&args->f_id, ntohl(tcp->th_seq),
1513 ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1524 * Given an ip_fw *, lookup_next_rule will return a pointer
1525 * to the next rule, which can be either the jump
1526 * target (for skipto instructions) or the next one in the list (in
1527 * all other cases including a missing jump target).
1528 * The result is also written in the "next_rule" field of the rule.
1529 * Backward jumps are not allowed, so start looking from the next
1532 * This never returns NULL -- in case we do not have an exact match,
1533 * the next rule is returned. When the ruleset is changed,
1534 * pointers are flushed so we are always correct.
1537 static struct ip_fw *
1538 lookup_next_rule(struct ip_fw *me)
1540 struct ip_fw *rule = NULL;
1543 /* look for action, in case it is a skipto */
1544 cmd = ACTION_PTR(me);
1545 if (cmd->opcode == O_LOG)
1547 if (cmd->opcode == O_SKIPTO) {
1548 for (rule = me->next; rule; rule = rule->next) {
1549 if (rule->rulenum >= cmd->arg1)
1553 if (rule == NULL) /* failure or not a skipto */
1555 me->next_rule = rule;
1560 * The main check routine for the firewall.
1562 * All arguments are in args so we can modify them and return them
1563 * back to the caller.
1567 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1568 * Starts with the IP header.
1569 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1570 * args->oif Outgoing interface, or NULL if packet is incoming.
1571 * The incoming interface is in the mbuf. (in)
1573 * args->rule Pointer to the last matching rule (in/out)
1574 * args->f_id Addresses grabbed from the packet (out)
1578 * IP_FW_PORT_DENY_FLAG the packet must be dropped.
1579 * 0 The packet is to be accepted and routed normally OR
1580 * the packet was denied/rejected and has been dropped;
1581 * in the latter case, *m is equal to NULL upon return.
1582 * port Divert the packet to port, with these caveats:
1584 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1585 * of diverting it (ie, 'ipfw tee').
1587 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1588 * 16 bits as a dummynet pipe number instead of diverting
1592 ipfw_chk(struct ip_fw_args *args)
1595 * Local variables hold state during the processing of a packet.
1597 * IMPORTANT NOTE: to speed up the processing of rules, there
1598 * are some assumption on the values of the variables, which
1599 * are documented here. Should you change them, please check
1600 * the implementation of the various instructions to make sure
1601 * that they still work.
1603 * args->eh The MAC header. It is non-null for a layer2
1604 * packet, it is NULL for a layer-3 packet.
1606 * m | args->m Pointer to the mbuf, as received from the caller.
1607 * It may change if ipfw_chk() does an m_pullup, or if it
1608 * consumes the packet because it calls send_reject().
1609 * XXX This has to change, so that ipfw_chk() never modifies
1610 * or consumes the buffer.
1611 * ip is simply an alias of the value of m, and it is kept
1612 * in sync with it (the packet is supposed to start with
1615 struct mbuf *m = args->m;
1616 struct ip *ip = mtod(m, struct ip *);
1619 * oif | args->oif If NULL, ipfw_chk has been called on the
1620 * inbound path (ether_input, ip_input).
1621 * If non-NULL, ipfw_chk has been called on the outbound path
1622 * (ether_output, ip_output).
1624 struct ifnet *oif = args->oif;
1626 struct ip_fw *f = NULL; /* matching rule */
1627 int retval = IP_FW_PASS;
1629 struct divert_info *divinfo;
1632 * hlen The length of the IPv4 header.
1633 * hlen >0 means we have an IPv4 packet.
1635 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
1638 * offset The offset of a fragment. offset != 0 means that
1639 * we have a fragment at this offset of an IPv4 packet.
1640 * offset == 0 means that (if this is an IPv4 packet)
1641 * this is the first or only fragment.
1646 * Local copies of addresses. They are only valid if we have
1649 * proto The protocol. Set to 0 for non-ip packets,
1650 * or to the protocol read from the packet otherwise.
1651 * proto != 0 means that we have an IPv4 packet.
1653 * src_port, dst_port port numbers, in HOST format. Only
1654 * valid for TCP and UDP packets.
1656 * src_ip, dst_ip ip addresses, in NETWORK format.
1657 * Only valid for IPv4 packets.
1660 uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */
1661 struct in_addr src_ip, dst_ip; /* NOTE: network format */
1662 uint16_t ip_len = 0;
1665 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1666 * MATCH_NONE when checked and not matched (dyn_f = NULL),
1667 * MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1669 int dyn_dir = MATCH_UNKNOWN;
1670 struct ip_fw *dyn_f = NULL;
1671 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1673 if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1674 return IP_FW_PASS; /* accept */
1676 if (args->eh == NULL || /* layer 3 packet */
1677 (m->m_pkthdr.len >= sizeof(struct ip) &&
1678 ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1679 hlen = ip->ip_hl << 2;
1682 * Collect parameters into local variables for faster matching.
1684 if (hlen == 0) { /* do not grab addresses for non-ip pkts */
1685 proto = args->f_id.proto = 0; /* mark f_id invalid */
1686 goto after_ip_checks;
1689 proto = args->f_id.proto = ip->ip_p;
1690 src_ip = ip->ip_src;
1691 dst_ip = ip->ip_dst;
1692 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1693 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1694 ip_len = ntohs(ip->ip_len);
1696 offset = ip->ip_off & IP_OFFMASK;
1697 ip_len = ip->ip_len;
1700 #define PULLUP_TO(len) \
1702 if (m->m_len < (len)) { \
1703 args->m = m = m_pullup(m, (len));\
1705 goto pullup_failed; \
1706 ip = mtod(m, struct ip *); \
1716 PULLUP_TO(hlen + sizeof(struct tcphdr));
1717 tcp = L3HDR(struct tcphdr, ip);
1718 dst_port = tcp->th_dport;
1719 src_port = tcp->th_sport;
1720 args->f_id.flags = tcp->th_flags;
1728 PULLUP_TO(hlen + sizeof(struct udphdr));
1729 udp = L3HDR(struct udphdr, ip);
1730 dst_port = udp->uh_dport;
1731 src_port = udp->uh_sport;
1736 PULLUP_TO(hlen + 4); /* type, code and checksum. */
1737 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1747 args->f_id.src_ip = ntohl(src_ip.s_addr);
1748 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1749 args->f_id.src_port = src_port = ntohs(src_port);
1750 args->f_id.dst_port = dst_port = ntohs(dst_port);
1755 * Packet has already been tagged. Look for the next rule
1756 * to restart processing.
1758 * If fw_one_pass != 0 then just accept it.
1759 * XXX should not happen here, but optimized out in
1765 /* This rule is being/has been flushed */
1769 KASSERT(args->rule->cpuid == mycpuid,
1770 ("rule used on cpu%d\n", mycpuid));
1772 /* This rule was deleted */
1773 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1776 f = args->rule->next_rule;
1778 f = lookup_next_rule(args->rule);
1781 * Find the starting rule. It can be either the first
1782 * one, or the one after divert_rule if asked so.
1786 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1788 divinfo = m_tag_data(mtag);
1789 skipto = divinfo->skipto;
1794 f = ctx->ipfw_layer3_chain;
1795 if (args->eh == NULL && skipto != 0) {
1796 /* No skipto during rule flushing */
1800 if (skipto >= IPFW_DEFAULT_RULE)
1801 return IP_FW_DENY; /* invalid */
1803 while (f && f->rulenum <= skipto)
1805 if (f == NULL) /* drop packet */
1807 } else if (ipfw_flushing) {
1808 /* Rules are being flushed; skip to default rule */
1809 f = ctx->ipfw_default_rule;
1812 if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1813 m_tag_delete(m, mtag);
1816 * Now scan the rules, and parse microinstructions for each rule.
1818 for (; f; f = f->next) {
1821 int skip_or; /* skip rest of OR block */
1824 if (ctx->ipfw_set_disable & (1 << f->set))
1828 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1829 l -= cmdlen, cmd += cmdlen) {
1833 * check_body is a jump target used when we find a
1834 * CHECK_STATE, and need to jump to the body of
1839 cmdlen = F_LEN(cmd);
1841 * An OR block (insn_1 || .. || insn_n) has the
1842 * F_OR bit set in all but the last instruction.
1843 * The first match will set "skip_or", and cause
1844 * the following instructions to be skipped until
1845 * past the one with the F_OR bit clear.
1847 if (skip_or) { /* skip this instruction */
1848 if ((cmd->len & F_OR) == 0)
1849 skip_or = 0; /* next one is good */
1852 match = 0; /* set to 1 if we succeed */
1854 switch (cmd->opcode) {
1856 * The first set of opcodes compares the packet's
1857 * fields with some pattern, setting 'match' if a
1858 * match is found. At the end of the loop there is
1859 * logic to deal with F_NOT and F_OR flags associated
1867 kprintf("ipfw: opcode %d unimplemented\n",
1874 * We only check offset == 0 && proto != 0,
1875 * as this ensures that we have an IPv4
1876 * packet with the ports info.
1881 struct inpcbinfo *pi;
1885 if (proto == IPPROTO_TCP) {
1887 pi = &tcbinfo[mycpu->gd_cpuid];
1888 } else if (proto == IPPROTO_UDP) {
1895 in_pcblookup_hash(pi,
1896 dst_ip, htons(dst_port),
1897 src_ip, htons(src_port),
1899 in_pcblookup_hash(pi,
1900 src_ip, htons(src_port),
1901 dst_ip, htons(dst_port),
1904 if (pcb == NULL || pcb->inp_socket == NULL)
1907 if (cmd->opcode == O_UID) {
1908 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1910 !socheckuid(pcb->inp_socket,
1911 (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1914 match = groupmember(
1915 (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1916 pcb->inp_socket->so_cred);
1922 match = iface_match(m->m_pkthdr.rcvif,
1923 (ipfw_insn_if *)cmd);
1927 match = iface_match(oif, (ipfw_insn_if *)cmd);
1931 match = iface_match(oif ? oif :
1932 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1936 if (args->eh != NULL) { /* have MAC header */
1937 uint32_t *want = (uint32_t *)
1938 ((ipfw_insn_mac *)cmd)->addr;
1939 uint32_t *mask = (uint32_t *)
1940 ((ipfw_insn_mac *)cmd)->mask;
1941 uint32_t *hdr = (uint32_t *)args->eh;
1944 (want[0] == (hdr[0] & mask[0]) &&
1945 want[1] == (hdr[1] & mask[1]) &&
1946 want[2] == (hdr[2] & mask[2]));
1951 if (args->eh != NULL) {
1953 ntohs(args->eh->ether_type);
1955 ((ipfw_insn_u16 *)cmd)->ports;
1958 /* Special vlan handling */
1959 if (m->m_flags & M_VLANTAG)
1962 for (i = cmdlen - 1; !match && i > 0;
1965 (t >= p[0] && t <= p[1]);
1971 match = (hlen > 0 && offset != 0);
1974 case O_IN: /* "out" is "not in" */
1975 match = (oif == NULL);
1979 match = (args->eh != NULL);
1984 * We do not allow an arg of 0 so the
1985 * check of "proto" only suffices.
1987 match = (proto == cmd->arg1);
1991 match = (hlen > 0 &&
1992 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1997 match = (hlen > 0 &&
1998 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2000 ((ipfw_insn_ip *)cmd)->mask.s_addr));
2007 tif = INADDR_TO_IFP(&src_ip);
2008 match = (tif != NULL);
2015 uint32_t *d = (uint32_t *)(cmd + 1);
2017 cmd->opcode == O_IP_DST_SET ?
2023 addr -= d[0]; /* subtract base */
2025 (addr < cmd->arg1) &&
2026 (d[1 + (addr >> 5)] &
2027 (1 << (addr & 0x1f)));
2032 match = (hlen > 0 &&
2033 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2038 match = (hlen > 0) &&
2039 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2041 ((ipfw_insn_ip *)cmd)->mask.s_addr));
2048 tif = INADDR_TO_IFP(&dst_ip);
2049 match = (tif != NULL);
2056 * offset == 0 && proto != 0 is enough
2057 * to guarantee that we have an IPv4
2058 * packet with port info.
2060 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2063 (cmd->opcode == O_IP_SRCPORT) ?
2064 src_port : dst_port ;
2066 ((ipfw_insn_u16 *)cmd)->ports;
2069 for (i = cmdlen - 1; !match && i > 0;
2072 (x >= p[0] && x <= p[1]);
2078 match = (offset == 0 && proto==IPPROTO_ICMP &&
2079 icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2083 match = (hlen > 0 && ipopts_match(ip, cmd));
2087 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2091 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2095 match = (hlen > 0 &&
2096 cmd->arg1 == ntohs(ip->ip_id));
2100 match = (hlen > 0 && cmd->arg1 == ip_len);
2103 case O_IPPRECEDENCE:
2104 match = (hlen > 0 &&
2105 (cmd->arg1 == (ip->ip_tos & 0xe0)));
2109 match = (hlen > 0 &&
2110 flags_match(cmd, ip->ip_tos));
2114 match = (proto == IPPROTO_TCP && offset == 0 &&
2116 L3HDR(struct tcphdr,ip)->th_flags));
2120 match = (proto == IPPROTO_TCP && offset == 0 &&
2121 tcpopts_match(ip, cmd));
2125 match = (proto == IPPROTO_TCP && offset == 0 &&
2126 ((ipfw_insn_u32 *)cmd)->d[0] ==
2127 L3HDR(struct tcphdr,ip)->th_seq);
2131 match = (proto == IPPROTO_TCP && offset == 0 &&
2132 ((ipfw_insn_u32 *)cmd)->d[0] ==
2133 L3HDR(struct tcphdr,ip)->th_ack);
2137 match = (proto == IPPROTO_TCP && offset == 0 &&
2139 L3HDR(struct tcphdr,ip)->th_win);
2143 /* reject packets which have SYN only */
2144 /* XXX should i also check for TH_ACK ? */
2145 match = (proto == IPPROTO_TCP && offset == 0 &&
2146 (L3HDR(struct tcphdr,ip)->th_flags &
2147 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2152 ipfw_log(f, hlen, args->eh, m, oif);
2157 match = (krandom() <
2158 ((ipfw_insn_u32 *)cmd)->d[0]);
2162 * The second set of opcodes represents 'actions',
2163 * i.e. the terminal part of a rule once the packet
2164 * matches all previous patterns.
2165 * Typically there is only one action for each rule,
2166 * and the opcode is stored at the end of the rule
2167 * (but there are exceptions -- see below).
2169 * In general, here we set retval and terminate the
2170 * outer loop (would be a 'break 3' in some language,
2171 * but we need to do a 'goto done').
2174 * O_COUNT and O_SKIPTO actions:
2175 * instead of terminating, we jump to the next rule
2176 * ('goto next_rule', equivalent to a 'break 2'),
2177 * or to the SKIPTO target ('goto again' after
2178 * having set f, cmd and l), respectively.
2180 * O_LIMIT and O_KEEP_STATE: these opcodes are
2181 * not real 'actions', and are stored right
2182 * before the 'action' part of the rule.
2183 * These opcodes try to install an entry in the
2184 * state tables; if successful, we continue with
2185 * the next opcode (match=1; break;), otherwise
2186 * the packet must be dropped ('goto done' after
2187 * setting retval). If static rules are changed
2188 * during the state installation, the packet will
2189 * be dropped and rule's stats will not beupdated
2190 * ('return IP_FW_DENY').
2192 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2193 * cause a lookup of the state table, and a jump
2194 * to the 'action' part of the parent rule
2195 * ('goto check_body') if an entry is found, or
2196 * (CHECK_STATE only) a jump to the next rule if
2197 * the entry is not found ('goto next_rule').
2198 * The result of the lookup is cached to make
2199 * further instances of these opcodes are
2200 * effectively NOPs. If static rules are changed
2201 * during the state looking up, the packet will
2202 * be dropped and rule's stats will not be updated
2203 * ('return IP_FW_DENY').
2207 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2208 kprintf("%s rule (%d) is not ready "
2210 cmd->opcode == O_LIMIT ?
2211 "limit" : "keep state",
2212 f->rulenum, f->cpuid);
2215 if (install_state(f,
2216 (ipfw_insn_limit *)cmd, args, &deny)) {
2220 retval = IP_FW_DENY;
2221 goto done; /* error/limit violation */
2231 * dynamic rules are checked at the first
2232 * keep-state or check-state occurrence,
2233 * with the result being stored in dyn_dir.
2234 * The compiler introduces a PROBE_STATE
2235 * instruction for us when we have a
2236 * KEEP_STATE (because PROBE_STATE needs
2239 if (dyn_dir == MATCH_UNKNOWN) {
2240 dyn_f = lookup_rule(&args->f_id,
2242 proto == IPPROTO_TCP ?
2243 L3HDR(struct tcphdr, ip) : NULL,
2247 if (dyn_f != NULL) {
2249 * Found a rule from a dynamic
2250 * entry; jump to the 'action'
2254 cmd = ACTION_PTR(f);
2255 l = f->cmd_len - f->act_ofs;
2260 * Dynamic entry not found. If CHECK_STATE,
2261 * skip to next rule, if PROBE_STATE just
2262 * ignore and continue with next opcode.
2264 if (cmd->opcode == O_CHECK_STATE)
2266 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2267 goto next_rule; /* not ready yet */
2272 retval = IP_FW_PASS; /* accept */
2277 args->rule = f; /* report matching rule */
2278 args->cookie = cmd->arg1;
2279 retval = IP_FW_DUMMYNET;
2284 if (args->eh) /* not on layer 2 */
2287 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2288 sizeof(*divinfo), MB_DONTWAIT);
2290 retval = IP_FW_DENY;
2293 divinfo = m_tag_data(mtag);
2295 divinfo->skipto = f->rulenum;
2296 divinfo->port = cmd->arg1;
2297 divinfo->tee = (cmd->opcode == O_TEE);
2298 m_tag_prepend(m, mtag);
2300 args->cookie = cmd->arg1;
2301 retval = (cmd->opcode == O_DIVERT) ?
2302 IP_FW_DIVERT : IP_FW_TEE;
2307 f->pcnt++; /* update stats */
2309 f->timestamp = time_second;
2310 if (cmd->opcode == O_COUNT)
2313 if (f->next_rule == NULL)
2314 lookup_next_rule(f);
2320 * Drop the packet and send a reject notice
2321 * if the packet is not ICMP (or is an ICMP
2322 * query), and it is not multicast/broadcast.
2325 (proto != IPPROTO_ICMP ||
2326 is_icmp_query(ip)) &&
2327 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2328 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2330 * Update statistics before the possible
2331 * blocking 'send_reject'
2335 f->timestamp = time_second;
2337 send_reject(args, cmd->arg1,
2342 * Return directly here, rule stats
2343 * have been updated above.
2349 retval = IP_FW_DENY;
2353 if (args->eh) /* not valid on layer2 pkts */
2355 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2356 struct sockaddr_in *sin;
2358 mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2359 sizeof(*sin), MB_DONTWAIT);
2361 retval = IP_FW_DENY;
2364 sin = m_tag_data(mtag);
2366 /* Structure copy */
2367 *sin = ((ipfw_insn_sa *)cmd)->sa;
2369 m_tag_prepend(m, mtag);
2370 m->m_pkthdr.fw_flags |=
2371 IPFORWARD_MBUF_TAGGED;
2373 retval = IP_FW_PASS;
2377 panic("-- unknown opcode %d\n", cmd->opcode);
2378 } /* end of switch() on opcodes */
2380 if (cmd->len & F_NOT)
2384 if (cmd->len & F_OR)
2387 if (!(cmd->len & F_OR)) /* not an OR block, */
2388 break; /* try next rule */
2391 } /* end of inner for, scan opcodes */
2393 next_rule:; /* try next rule */
2395 } /* end of outer for, scan rules */
2396 kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2400 /* Update statistics */
2403 f->timestamp = time_second;
2408 kprintf("pullup failed\n");
2413 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2418 const struct ipfw_flow_id *id;
2419 struct dn_flow_id *fid;
2423 mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2428 m_tag_prepend(m, mtag);
2430 pkt = m_tag_data(mtag);
2431 bzero(pkt, sizeof(*pkt));
2433 cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2434 if (cmd->opcode == O_LOG)
2436 KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2437 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2440 pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2441 pkt->ifp = fwa->oif;
2442 pkt->cpuid = mycpu->gd_cpuid;
2443 pkt->pipe_nr = pipe_nr;
2447 fid->fid_dst_ip = id->dst_ip;
2448 fid->fid_src_ip = id->src_ip;
2449 fid->fid_dst_port = id->dst_port;
2450 fid->fid_src_port = id->src_port;
2451 fid->fid_proto = id->proto;
2452 fid->fid_flags = id->flags;
2454 ipfw_ref_rule(fwa->rule);
2455 pkt->dn_priv = fwa->rule;
2456 pkt->dn_unref_priv = ipfw_unref_rule;
2458 if (cmd->opcode == O_PIPE)
2459 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2461 m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2465 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2466 * These will be reconstructed on the fly as packets are matched.
2467 * Must be called at splimp().
2470 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2474 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2475 rule->next_rule = NULL;
2478 static __inline void
2479 ipfw_inc_static_count(struct ip_fw *rule)
2481 KKASSERT(mycpuid == 0);
2484 static_ioc_len += IOC_RULESIZE(rule);
2487 static __inline void
2488 ipfw_dec_static_count(struct ip_fw *rule)
2490 int l = IOC_RULESIZE(rule);
2492 KKASSERT(mycpuid == 0);
2494 KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2497 KASSERT(static_ioc_len >= l,
2498 ("invalid static len %u\n", static_ioc_len));
2499 static_ioc_len -= l;
2503 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2505 if (fwmsg->sibling != NULL) {
2506 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2507 fwmsg->sibling->sibling = rule;
2509 fwmsg->sibling = rule;
2512 static struct ip_fw *
2513 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2517 rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2519 rule->act_ofs = ioc_rule->act_ofs;
2520 rule->cmd_len = ioc_rule->cmd_len;
2521 rule->rulenum = ioc_rule->rulenum;
2522 rule->set = ioc_rule->set;
2523 rule->usr_flags = ioc_rule->usr_flags;
2525 bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2528 rule->cpuid = mycpuid;
2532 stub->rule[mycpuid] = rule;
2538 ipfw_add_rule_dispatch(struct netmsg *nmsg)
2540 struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2541 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2544 rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2547 * Bump generation after ipfw_create_rule(),
2548 * since this function is blocking
2553 * Insert rule into the pre-determined position
2555 if (fwmsg->prev_rule != NULL) {
2556 struct ip_fw *prev, *next;
2558 prev = fwmsg->prev_rule;
2559 KKASSERT(prev->cpuid == mycpuid);
2561 next = fwmsg->next_rule;
2562 KKASSERT(next->cpuid == mycpuid);
2568 * Move to the position on the next CPU
2569 * before the msg is forwarded.
2571 fwmsg->prev_rule = prev->sibling;
2572 fwmsg->next_rule = next->sibling;
2574 KKASSERT(fwmsg->next_rule == NULL);
2575 rule->next = ctx->ipfw_layer3_chain;
2576 ctx->ipfw_layer3_chain = rule;
2579 /* Link rule CPU sibling */
2580 ipfw_link_sibling(fwmsg, rule);
2582 ipfw_flush_rule_ptrs(ctx);
2585 /* Statistics only need to be updated once */
2586 ipfw_inc_static_count(rule);
2588 /* Return the rule on CPU0 */
2589 nmsg->nm_lmsg.u.ms_resultp = rule;
2592 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2596 ipfw_enable_state_dispatch(struct netmsg *nmsg)
2598 struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
2599 struct ip_fw *rule = lmsg->u.ms_resultp;
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;
2607 ifnet_forwardmsg(lmsg, mycpuid + 1);
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
2617 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
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;
2625 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2630 * If rulenum is 0, find highest numbered rule before the
2631 * default rule, and add rule number incremental step.
2633 if (ioc_rule->rulenum == 0) {
2634 int step = autoinc_step;
2636 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2637 step <= IPFW_AUTOINC_STEP_MAX);
2640 * Locate the highest numbered rule before default
2642 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2643 if (f->rulenum == IPFW_DEFAULT_RULE)
2645 ioc_rule->rulenum = f->rulenum;
2647 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2648 ioc_rule->rulenum += step;
2650 KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2651 ioc_rule->rulenum != 0,
2652 ("invalid rule num %d\n", ioc_rule->rulenum));
2655 * Now find the right place for the new rule in the sorted list.
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 */
2664 KASSERT(f != NULL, ("no default rule?!\n"));
2666 if (rule_flags & IPFW_RULE_F_STATE) {
2670 * If the new rule will create states, then allocate
2671 * a rule stub, which will be referenced by states
2674 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2675 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2681 * Duplicate the rule onto each CPU.
2682 * The rule duplicated on CPU0 will be returned.
2684 bzero(&fwmsg, sizeof(fwmsg));
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;
2692 ifnet_domsg(&nmsg->nm_lmsg, 0);
2693 KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2695 rule = nmsg->nm_lmsg.u.ms_resultp;
2696 KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2698 if (rule_flags & IPFW_RULE_F_STATE) {
2700 * Turn on state flag, _after_ everything on all
2701 * CPUs have been setup.
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;
2708 ifnet_domsg(&nmsg->nm_lmsg, 0);
2709 KKASSERT(nmsg->nm_lmsg.u.ms_resultp == NULL);
2714 DEB(kprintf("++ installed rule %d, static count now %d\n",
2715 rule->rulenum, static_count);)
2719 * Free storage associated with a static rule (including derived
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().
2727 static struct ip_fw *
2728 ipfw_delete_rule(struct ipfw_context *ctx,
2729 struct ip_fw *prev, struct ip_fw *rule)
2732 struct ip_fw_stub *stub;
2736 /* STATE flag should have been cleared before we reach here */
2737 KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2742 ctx->ipfw_layer3_chain = n;
2746 /* Mark the rule as invalid */
2747 rule->rule_flags |= IPFW_RULE_F_INVALID;
2748 rule->next_rule = NULL;
2749 rule->sibling = NULL;
2752 /* Don't reset cpuid here; keep various assertion working */
2756 /* Statistics only need to be updated once */
2758 ipfw_dec_static_count(rule);
2760 /* Free 'stub' on the last CPU */
2761 if (stub != NULL && mycpuid == ncpus - 1)
2762 kfree(stub, M_IPFW);
2764 /* Try to free this rule */
2765 ipfw_free_rule(rule);
2767 /* Return the next rule */
2772 ipfw_flush_dispatch(struct netmsg *nmsg)
2774 struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
2775 int kill_default = lmsg->u.ms_result;
2776 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2779 ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2781 while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2782 (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2783 ipfw_delete_rule(ctx, NULL, rule);
2785 ifnet_forwardmsg(lmsg, mycpuid + 1);
2789 ipfw_disable_rule_state_dispatch(struct netmsg *nmsg)
2791 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2794 rule = dmsg->start_rule;
2796 KKASSERT(rule->cpuid == mycpuid);
2799 * Move to the position on the next CPU
2800 * before the msg is forwarded.
2802 dmsg->start_rule = rule->sibling;
2804 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2806 KKASSERT(dmsg->rulenum == 0);
2807 rule = ctx->ipfw_layer3_chain;
2810 while (rule != NULL) {
2811 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2813 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2817 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2821 * Deletes all rules from a chain (including the default rule
2822 * if the second argument is set).
2823 * Must be called at splimp().
2826 ipfw_flush(int kill_default)
2828 struct netmsg_del dmsg;
2830 struct lwkt_msg *lmsg;
2832 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2834 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2837 * If 'kill_default' then caller has done the necessary
2838 * msgport syncing; unnecessary to do it again.
2840 if (!kill_default) {
2842 * Let ipfw_chk() know the rules are going to
2843 * be flushed, so it could jump directly to
2847 netmsg_service_sync();
2851 * Clear STATE flag on rules, so no more states (dyn rules)
2854 bzero(&dmsg, sizeof(dmsg));
2855 netmsg_init(&dmsg.nmsg, &curthread->td_msgport, 0,
2856 ipfw_disable_rule_state_dispatch);
2857 ifnet_domsg(&dmsg.nmsg.nm_lmsg, 0);
2860 * This actually nukes all states (dyn rules)
2862 lockmgr(&dyn_lock, LK_EXCLUSIVE);
2863 for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2865 * Can't check IPFW_RULE_F_STATE here,
2866 * since it has been cleared previously.
2867 * Check 'stub' instead.
2869 if (rule->stub != NULL) {
2871 remove_dyn_rule_locked(rule, NULL);
2874 lockmgr(&dyn_lock, LK_RELEASE);
2877 * Press the 'flush' button
2879 bzero(&nmsg, sizeof(nmsg));
2880 netmsg_init(&nmsg, &curthread->td_msgport, 0, ipfw_flush_dispatch);
2881 lmsg = &nmsg.nm_lmsg;
2882 lmsg->u.ms_result = kill_default;
2883 ifnet_domsg(lmsg, 0);
2885 KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2888 if (ipfw_dyn_v != NULL) {
2890 * Free dynamic rules(state) hash table
2892 kfree(ipfw_dyn_v, M_IPFW);
2896 KASSERT(static_count == 0,
2897 ("%u static rules remains\n", static_count));
2898 KASSERT(static_ioc_len == 0,
2899 ("%u bytes of static rules remains\n", static_ioc_len));
2901 KASSERT(static_count == 1,
2902 ("%u static rules remains\n", static_count));
2903 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2904 ("%u bytes of static rules remains, should be %u\n",
2905 static_ioc_len, IOC_RULESIZE(ctx->ipfw_default_rule)));
2913 ipfw_alt_delete_rule_dispatch(struct netmsg *nmsg)
2915 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2916 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2917 struct ip_fw *rule, *prev;
2919 rule = dmsg->start_rule;
2920 KKASSERT(rule->cpuid == mycpuid);
2921 dmsg->start_rule = rule->sibling;
2923 prev = dmsg->prev_rule;
2925 KKASSERT(prev->cpuid == mycpuid);
2928 * Move to the position on the next CPU
2929 * before the msg is forwarded.
2931 dmsg->prev_rule = prev->sibling;
2935 * flush pointers outside the loop, then delete all matching
2936 * rules. 'prev' remains the same throughout the cycle.
2938 ipfw_flush_rule_ptrs(ctx);
2939 while (rule && rule->rulenum == dmsg->rulenum)
2940 rule = ipfw_delete_rule(ctx, prev, rule);
2942 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2946 ipfw_alt_delete_rule(uint16_t rulenum)
2948 struct ip_fw *prev, *rule, *f;
2949 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2950 struct netmsg_del dmsg;
2951 struct netmsg *nmsg;
2955 * Locate first rule to delete
2957 for (prev = NULL, rule = ctx->ipfw_layer3_chain;
2958 rule && rule->rulenum < rulenum;
2959 prev = rule, rule = rule->next)
2961 if (rule->rulenum != rulenum)
2965 * Check whether any rules with the given number will
2969 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
2970 if (f->rule_flags & IPFW_RULE_F_STATE) {
2978 * Clear the STATE flag, so no more states will be
2979 * created based the rules numbered 'rulenum'.
2981 bzero(&dmsg, sizeof(dmsg));
2983 netmsg_init(nmsg, &curthread->td_msgport, 0,
2984 ipfw_disable_rule_state_dispatch);
2985 dmsg.start_rule = rule;
2986 dmsg.rulenum = rulenum;
2988 ifnet_domsg(&nmsg->nm_lmsg, 0);
2989 KKASSERT(dmsg.start_rule == NULL);
2992 * Nuke all related states
2994 lockmgr(&dyn_lock, LK_EXCLUSIVE);
2995 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
2997 * Can't check IPFW_RULE_F_STATE here,
2998 * since it has been cleared previously.
2999 * Check 'stub' instead.
3001 if (f->stub != NULL) {
3003 remove_dyn_rule_locked(f, NULL);
3006 lockmgr(&dyn_lock, LK_RELEASE);
3010 * Get rid of the rule duplications on all CPUs
3012 bzero(&dmsg, sizeof(dmsg));
3014 netmsg_init(nmsg, &curthread->td_msgport, 0,
3015 ipfw_alt_delete_rule_dispatch);
3016 dmsg.prev_rule = prev;
3017 dmsg.start_rule = rule;
3018 dmsg.rulenum = rulenum;
3020 ifnet_domsg(&nmsg->nm_lmsg, 0);
3021 KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3026 ipfw_alt_delete_ruleset_dispatch(struct netmsg *nmsg)
3028 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3029 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3030 struct ip_fw *prev, *rule;
3035 ipfw_flush_rule_ptrs(ctx);
3038 rule = ctx->ipfw_layer3_chain;
3039 while (rule != NULL) {
3040 if (rule->set == dmsg->from_set) {
3041 rule = ipfw_delete_rule(ctx, prev, rule);
3050 KASSERT(del, ("no match set?!\n"));
3052 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3056 ipfw_disable_ruleset_state_dispatch(struct netmsg *nmsg)
3058 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3059 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3065 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3066 if (rule->set == dmsg->from_set) {
3070 rule->rule_flags &= ~IPFW_RULE_F_STATE;
3073 KASSERT(cleared, ("no match set?!\n"));
3075 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3079 ipfw_alt_delete_ruleset(uint8_t set)
3081 struct netmsg_del dmsg;
3082 struct netmsg *nmsg;
3085 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3088 * Check whether the 'set' exists. If it exists,
3089 * then check whether any rules within the set will
3090 * try to create states.
3094 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3095 if (rule->set == set) {
3097 if (rule->rule_flags & IPFW_RULE_F_STATE) {
3104 return 0; /* XXX EINVAL? */
3108 * Clear the STATE flag, so no more states will be
3109 * created based the rules in this set.
3111 bzero(&dmsg, sizeof(dmsg));
3113 netmsg_init(nmsg, &curthread->td_msgport, 0,
3114 ipfw_disable_ruleset_state_dispatch);
3115 dmsg.from_set = set;
3117 ifnet_domsg(&nmsg->nm_lmsg, 0);
3120 * Nuke all related states
3122 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3123 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3124 if (rule->set != set)
3128 * Can't check IPFW_RULE_F_STATE here,
3129 * since it has been cleared previously.
3130 * Check 'stub' instead.
3132 if (rule->stub != NULL) {
3134 remove_dyn_rule_locked(rule, NULL);
3137 lockmgr(&dyn_lock, LK_RELEASE);
3143 bzero(&dmsg, sizeof(dmsg));
3145 netmsg_init(nmsg, &curthread->td_msgport, 0,
3146 ipfw_alt_delete_ruleset_dispatch);
3147 dmsg.from_set = set;
3149 ifnet_domsg(&nmsg->nm_lmsg, 0);
3154 ipfw_alt_move_rule_dispatch(struct netmsg *nmsg)
3156 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3159 rule = dmsg->start_rule;
3160 KKASSERT(rule->cpuid == mycpuid);
3163 * Move to the position on the next CPU
3164 * before the msg is forwarded.
3166 dmsg->start_rule = rule->sibling;
3168 while (rule && rule->rulenum <= dmsg->rulenum) {
3169 if (rule->rulenum == dmsg->rulenum)
3170 rule->set = dmsg->to_set;
3173 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3177 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3179 struct netmsg_del dmsg;
3180 struct netmsg *nmsg;
3182 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3185 * Locate first rule to move
3187 for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3188 rule = rule->next) {
3189 if (rule->rulenum == rulenum && rule->set != set)
3192 if (rule == NULL || rule->rulenum > rulenum)
3193 return 0; /* XXX error? */
3195 bzero(&dmsg, sizeof(dmsg));
3197 netmsg_init(nmsg, &curthread->td_msgport, 0,
3198 ipfw_alt_move_rule_dispatch);
3199 dmsg.start_rule = rule;
3200 dmsg.rulenum = rulenum;
3203 ifnet_domsg(&nmsg->nm_lmsg, 0);
3204 KKASSERT(dmsg.start_rule == NULL);
3209 ipfw_alt_move_ruleset_dispatch(struct netmsg *nmsg)
3211 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3212 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3215 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3216 if (rule->set == dmsg->from_set)
3217 rule->set = dmsg->to_set;
3219 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3223 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3225 struct netmsg_del dmsg;
3226 struct netmsg *nmsg;
3228 bzero(&dmsg, sizeof(dmsg));
3230 netmsg_init(nmsg, &curthread->td_msgport, 0,
3231 ipfw_alt_move_ruleset_dispatch);
3232 dmsg.from_set = from_set;
3233 dmsg.to_set = to_set;
3235 ifnet_domsg(&nmsg->nm_lmsg, 0);
3240 ipfw_alt_swap_ruleset_dispatch(struct netmsg *nmsg)
3242 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3243 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3246 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3247 if (rule->set == dmsg->from_set)
3248 rule->set = dmsg->to_set;
3249 else if (rule->set == dmsg->to_set)
3250 rule->set = dmsg->from_set;
3252 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3256 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3258 struct netmsg_del dmsg;
3259 struct netmsg *nmsg;
3261 bzero(&dmsg, sizeof(dmsg));
3263 netmsg_init(nmsg, &curthread->td_msgport, 0,
3264 ipfw_alt_swap_ruleset_dispatch);
3265 dmsg.from_set = set1;
3268 ifnet_domsg(&nmsg->nm_lmsg, 0);
3273 * Remove all rules with given number, and also do set manipulation.
3275 * The argument is an uint32_t. The low 16 bit are the rule or set number,
3276 * the next 8 bits are the new set, the top 8 bits are the command:
3278 * 0 delete rules with given number
3279 * 1 delete rules with given set number
3280 * 2 move rules with given number to new set
3281 * 3 move rules with given set number to new set
3282 * 4 swap sets with given numbers
3285 ipfw_ctl_alter(uint32_t arg)
3288 uint8_t cmd, new_set;
3291 rulenum = arg & 0xffff;
3292 cmd = (arg >> 24) & 0xff;
3293 new_set = (arg >> 16) & 0xff;
3297 if (new_set >= IPFW_DEFAULT_SET)
3299 if (cmd == 0 || cmd == 2) {
3300 if (rulenum == IPFW_DEFAULT_RULE)
3303 if (rulenum >= IPFW_DEFAULT_SET)
3308 case 0: /* delete rules with given number */
3309 error = ipfw_alt_delete_rule(rulenum);
3312 case 1: /* delete all rules with given set number */
3313 error = ipfw_alt_delete_ruleset(rulenum);
3316 case 2: /* move rules with given number to new set */
3317 error = ipfw_alt_move_rule(rulenum, new_set);
3320 case 3: /* move rules with given set number to new set */
3321 error = ipfw_alt_move_ruleset(rulenum, new_set);
3324 case 4: /* swap two sets */
3325 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3332 * Clear counters for a specific rule.
3335 clear_counters(struct ip_fw *rule, int log_only)
3337 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3339 if (log_only == 0) {
3340 rule->bcnt = rule->pcnt = 0;
3341 rule->timestamp = 0;
3343 if (l->o.opcode == O_LOG)
3344 l->log_left = l->max_log;
3348 ipfw_zero_entry_dispatch(struct netmsg *nmsg)
3350 struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3351 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3354 if (zmsg->rulenum == 0) {
3355 KKASSERT(zmsg->start_rule == NULL);
3357 ctx->ipfw_norule_counter = 0;
3358 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3359 clear_counters(rule, zmsg->log_only);
3361 struct ip_fw *start = zmsg->start_rule;
3363 KKASSERT(start->cpuid == mycpuid);
3364 KKASSERT(start->rulenum == zmsg->rulenum);
3367 * We can have multiple rules with the same number, so we
3368 * need to clear them all.
3370 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3372 clear_counters(rule, zmsg->log_only);
3375 * Move to the position on the next CPU
3376 * before the msg is forwarded.
3378 zmsg->start_rule = start->sibling;
3380 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3384 * Reset some or all counters on firewall rules.
3385 * @arg frwl is null to clear all entries, or contains a specific
3387 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3390 ipfw_ctl_zero_entry(int rulenum, int log_only)
3392 struct netmsg_zent zmsg;
3393 struct netmsg *nmsg;
3395 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3397 bzero(&zmsg, sizeof(zmsg));
3399 netmsg_init(nmsg, &curthread->td_msgport, 0, ipfw_zero_entry_dispatch);
3400 zmsg.log_only = log_only;
3403 msg = log_only ? "ipfw: All logging counts reset.\n"
3404 : "ipfw: Accounting cleared.\n";
3409 * Locate the first rule with 'rulenum'
3411 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3412 if (rule->rulenum == rulenum)
3415 if (rule == NULL) /* we did not find any matching rules */
3417 zmsg.start_rule = rule;
3418 zmsg.rulenum = rulenum;
3420 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3421 : "ipfw: Entry %d cleared.\n";
3423 ifnet_domsg(&nmsg->nm_lmsg, 0);
3424 KKASSERT(zmsg.start_rule == NULL);
3427 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3432 * Check validity of the structure before insert.
3433 * Fortunately rules are simple, so this mostly need to check rule sizes.
3436 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3439 int have_action = 0;
3444 /* Check for valid size */
3445 if (size < sizeof(*rule)) {
3446 kprintf("ipfw: rule too short\n");
3449 l = IOC_RULESIZE(rule);
3451 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3455 /* Check rule number */
3456 if (rule->rulenum == IPFW_DEFAULT_RULE) {
3457 kprintf("ipfw: invalid rule number\n");
3462 * Now go for the individual checks. Very simple ones, basically only
3463 * instruction sizes.
3465 for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3466 l -= cmdlen, cmd += cmdlen) {
3467 cmdlen = F_LEN(cmd);
3469 kprintf("ipfw: opcode %d size truncated\n",
3474 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
3476 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3477 /* This rule will create states */
3478 *rule_flags |= IPFW_RULE_F_STATE;
3481 switch (cmd->opcode) {
3495 case O_IPPRECEDENCE:
3502 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3514 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3519 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3524 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3527 ((ipfw_insn_log *)cmd)->log_left =
3528 ((ipfw_insn_log *)cmd)->max_log;
3534 if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3536 if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3537 kprintf("ipfw: opcode %d, useless rule\n",
3545 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3546 kprintf("ipfw: invalid set size %d\n",
3550 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3556 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3562 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3563 if (cmdlen < 2 || cmdlen > 31)
3570 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3576 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3581 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3586 fwd_addr = ((ipfw_insn_sa *)cmd)->
3588 if (IN_MULTICAST(ntohl(fwd_addr))) {
3589 kprintf("ipfw: try forwarding to "
3590 "multicast address\n");
3596 case O_FORWARD_MAC: /* XXX not implemented yet */
3605 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3609 kprintf("ipfw: opcode %d, multiple actions"
3616 kprintf("ipfw: opcode %d, action must be"
3623 kprintf("ipfw: opcode %d, unknown opcode\n",
3628 if (have_action == 0) {
3629 kprintf("ipfw: missing action\n");
3635 kprintf("ipfw: opcode %d size %d wrong\n",
3636 cmd->opcode, cmdlen);
3641 ipfw_ctl_add_rule(struct sockopt *sopt)
3643 struct ipfw_ioc_rule *ioc_rule;
3645 uint32_t rule_flags;
3648 size = sopt->sopt_valsize;
3649 if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3650 size < sizeof(*ioc_rule)) {
3653 if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3654 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3655 IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3657 ioc_rule = sopt->sopt_val;
3659 error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3663 ipfw_add_rule(ioc_rule, rule_flags);
3665 if (sopt->sopt_dir == SOPT_GET)
3666 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3671 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3673 const struct ip_fw *sibling;
3678 KKASSERT(rule->cpuid == 0);
3680 ioc_rule->act_ofs = rule->act_ofs;
3681 ioc_rule->cmd_len = rule->cmd_len;
3682 ioc_rule->rulenum = rule->rulenum;
3683 ioc_rule->set = rule->set;
3684 ioc_rule->usr_flags = rule->usr_flags;
3686 ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3687 ioc_rule->static_count = static_count;
3688 ioc_rule->static_len = static_ioc_len;
3691 * Visit (read-only) all of the rule's duplications to get
3692 * the necessary statistics
3699 ioc_rule->timestamp = 0;
3700 for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3701 ioc_rule->pcnt += sibling->pcnt;
3702 ioc_rule->bcnt += sibling->bcnt;
3703 if (sibling->timestamp > ioc_rule->timestamp)
3704 ioc_rule->timestamp = sibling->timestamp;
3709 KASSERT(i == ncpus, ("static rule is not duplicated on every cpu\n"));
3711 bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3713 return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3717 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3718 struct ipfw_ioc_state *ioc_state)
3720 const struct ipfw_flow_id *id;
3721 struct ipfw_ioc_flowid *ioc_id;
3723 ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3724 0 : dyn_rule->expire - time_second;
3725 ioc_state->pcnt = dyn_rule->pcnt;
3726 ioc_state->bcnt = dyn_rule->bcnt;
3728 ioc_state->dyn_type = dyn_rule->dyn_type;
3729 ioc_state->count = dyn_rule->count;
3731 ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3734 ioc_id = &ioc_state->id;
3736 ioc_id->type = ETHERTYPE_IP;
3737 ioc_id->u.ip.dst_ip = id->dst_ip;
3738 ioc_id->u.ip.src_ip = id->src_ip;
3739 ioc_id->u.ip.dst_port = id->dst_port;
3740 ioc_id->u.ip.src_port = id->src_port;
3741 ioc_id->u.ip.proto = id->proto;
3745 ipfw_ctl_get_rules(struct sockopt *sopt)
3747 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3751 uint32_t dcount = 0;
3754 * pass up a copy of the current rules. Static rules
3755 * come first (the last of which has number IPFW_DEFAULT_RULE),
3756 * followed by a possibly empty list of dynamic rule.
3760 size = static_ioc_len; /* size of static rules */
3761 if (ipfw_dyn_v) { /* add size of dyn.rules */
3763 size += dcount * sizeof(struct ipfw_ioc_state);
3766 if (sopt->sopt_valsize < size) {
3767 /* short length, no need to return incomplete rules */
3768 /* XXX: if superuser, no need to zero buffer */
3769 bzero(sopt->sopt_val, sopt->sopt_valsize);
3772 bp = sopt->sopt_val;
3774 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3775 bp = ipfw_copy_rule(rule, bp);
3777 if (ipfw_dyn_v && dcount != 0) {
3778 struct ipfw_ioc_state *ioc_state = bp;
3779 uint32_t dcount2 = 0;
3781 size_t old_size = size;
3785 lockmgr(&dyn_lock, LK_SHARED);
3787 /* Check 'ipfw_dyn_v' again with lock held */
3788 if (ipfw_dyn_v == NULL)
3791 for (i = 0; i < curr_dyn_buckets; i++) {
3795 * The # of dynamic rules may have grown after the
3796 * snapshot of 'dyn_count' was taken, so we will have
3797 * to check 'dcount' (snapshot of dyn_count) here to
3798 * make sure that we don't overflow the pre-allocated
3801 for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
3802 p = p->next, ioc_state++, dcount--, dcount2++)
3803 ipfw_copy_state(p, ioc_state);
3806 lockmgr(&dyn_lock, LK_RELEASE);
3809 * The # of dynamic rules may be shrinked after the
3810 * snapshot of 'dyn_count' was taken. To give user a
3811 * correct dynamic rule count, we use the 'dcount2'
3812 * calculated above (with shared lockmgr lock held).
3814 size = static_ioc_len +
3815 (dcount2 * sizeof(struct ipfw_ioc_state));
3816 KKASSERT(size <= old_size);
3821 sopt->sopt_valsize = size;
3826 ipfw_set_disable_dispatch(struct netmsg *nmsg)
3828 struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
3829 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3832 ctx->ipfw_set_disable = lmsg->u.ms_result32;
3834 ifnet_forwardmsg(lmsg, mycpuid + 1);
3838 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable)
3841 struct lwkt_msg *lmsg;
3842 uint32_t set_disable;
3844 /* IPFW_DEFAULT_SET is always enabled */
3845 enable |= (1 << IPFW_DEFAULT_SET);
3846 set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable;
3848 bzero(&nmsg, sizeof(nmsg));
3849 netmsg_init(&nmsg, &curthread->td_msgport, 0, ipfw_set_disable_dispatch);
3850 lmsg = &nmsg.nm_lmsg;
3851 lmsg->u.ms_result32 = set_disable;
3853 ifnet_domsg(lmsg, 0);
3857 * {set|get}sockopt parser.
3860 ipfw_ctl(struct sockopt *sopt)
3868 switch (sopt->sopt_name) {
3870 error = ipfw_ctl_get_rules(sopt);
3875 * Normally we cannot release the lock on each iteration.
3876 * We could do it here only because we start from the head all
3877 * the times so there is no risk of missing some entries.
3878 * On the other hand, the risk is that we end up with
3879 * a very inconsistent ruleset, so better keep the lock
3880 * around the whole cycle.
3882 * XXX this code can be improved by resetting the head of
3883 * the list to point to the default rule, and then freeing
3884 * the old list without the need for a lock.
3888 ipfw_flush(0 /* keep default rule */);
3893 error = ipfw_ctl_add_rule(sopt);
3898 * IP_FW_DEL is used for deleting single rules or sets,
3899 * and (ab)used to atomically manipulate sets.
3900 * Argument size is used to distinguish between the two:
3902 * delete single rule or set of rules,
3903 * or reassign rules (or sets) to a different set.
3904 * 2 * sizeof(uint32_t)
3905 * atomic disable/enable sets.
3906 * first uint32_t contains sets to be disabled,
3907 * second uint32_t contains sets to be enabled.
3909 masks = sopt->sopt_val;
3910 size = sopt->sopt_valsize;
3911 if (size == sizeof(*masks)) {
3913 * Delete or reassign static rule
3915 error = ipfw_ctl_alter(masks[0]);
3916 } else if (size == (2 * sizeof(*masks))) {
3918 * Set enable/disable
3920 ipfw_ctl_set_disable(masks[0], masks[1]);
3927 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3930 if (sopt->sopt_val != 0) {
3931 error = soopt_to_kbuf(sopt, &rulenum,
3932 sizeof(int), sizeof(int));
3936 error = ipfw_ctl_zero_entry(rulenum,
3937 sopt->sopt_name == IP_FW_RESETLOG);
3941 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3948 * This procedure is only used to handle keepalives. It is invoked
3949 * every dyn_keepalive_period
3952 ipfw_tick(void *dummy __unused)
3958 if (ipfw_dyn_v == NULL || dyn_count == 0)
3961 keep_alive = time_second;
3963 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3965 if (ipfw_dyn_v == NULL || dyn_count == 0) {
3966 lockmgr(&dyn_lock, LK_RELEASE);
3969 gen = dyn_buckets_gen;
3971 for (i = 0; i < curr_dyn_buckets; i++) {
3972 ipfw_dyn_rule *q, *prev;
3974 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
3975 uint32_t ack_rev, ack_fwd;
3976 struct ipfw_flow_id id;
3978 if (q->dyn_type == O_LIMIT_PARENT)
3981 if (TIME_LEQ(q->expire, time_second)) {
3983 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
3988 * Keep alive processing
3993 if (q->id.proto != IPPROTO_TCP)
3995 if ((q->state & BOTH_SYN) != BOTH_SYN)
3997 if (TIME_LEQ(time_second + dyn_keepalive_interval,
3999 goto next; /* too early */
4000 if (q->keep_alive == keep_alive)
4001 goto next; /* alreay done */
4004 * Save necessary information, so that they could
4005 * survive after possible blocking in send_pkt()
4008 ack_rev = q->ack_rev;
4009 ack_fwd = q->ack_fwd;
4011 /* Sending has been started */
4012 q->keep_alive = keep_alive;
4014 /* Release lock to avoid possible dead lock */
4015 lockmgr(&dyn_lock, LK_RELEASE);
4016 send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
4017 send_pkt(&id, ack_fwd - 1, ack_rev, 0);
4018 lockmgr(&dyn_lock, LK_EXCLUSIVE);
4020 if (gen != dyn_buckets_gen) {
4022 * Dyn bucket array has been changed during
4023 * the above two sending; reiterate.
4032 lockmgr(&dyn_lock, LK_RELEASE);
4034 callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
4039 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4041 struct ip_fw_args args;
4042 struct mbuf *m = *m0;
4044 int tee = 0, error = 0, i;
4046 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4047 /* Extract info from dummynet tag */
4048 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4049 KKASSERT(mtag != NULL);
4050 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4051 KKASSERT(args.rule != NULL);
4053 m_tag_delete(m, mtag);
4054 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4062 i = ipfw_chk(&args);
4080 case IP_FW_DUMMYNET:
4081 /* Send packet to the appropriate pipe */
4082 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args);
4090 if (ip_divert_p != NULL) {
4091 m = ip_divert_p(m, tee, 1);
4095 /* not sure this is the right error msg */
4101 panic("unknown ipfw return value: %d\n", i);
4109 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4111 struct ip_fw_args args;
4112 struct mbuf *m = *m0;
4114 int tee = 0, error = 0, off;
4116 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4117 /* Extract info from dummynet tag */
4118 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4119 KKASSERT(mtag != NULL);
4120 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4121 KKASSERT(args.rule != NULL);
4123 m_tag_delete(m, mtag);
4124 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4132 off = ipfw_chk(&args);
4150 case IP_FW_DUMMYNET:
4151 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args);
4159 if (ip_divert_p != NULL) {
4160 m = ip_divert_p(m, tee, 0);
4164 /* not sure this is the right error msg */
4170 panic("unknown ipfw return value: %d\n", off);
4180 struct pfil_head *pfh;
4182 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4184 pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4188 pfil_add_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_WAITOK, pfh);
4189 pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh);
4195 struct pfil_head *pfh;
4197 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4199 pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4203 pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_WAITOK, pfh);
4204 pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh);
4208 ipfw_sysctl_enable_dispatch(struct netmsg *nmsg)
4210 struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
4211 int enable = lmsg->u.ms_result;
4213 if (fw_enable == enable)
4222 lwkt_replymsg(lmsg, 0);
4226 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS)
4229 struct lwkt_msg *lmsg;
4233 error = sysctl_handle_int(oidp, &enable, 0, req);
4234 if (error || req->newptr == NULL)
4237 netmsg_init(&nmsg, &curthread->td_msgport, 0,
4238 ipfw_sysctl_enable_dispatch);
4239 lmsg = &nmsg.nm_lmsg;
4240 lmsg->u.ms_result = enable;
4242 return lwkt_domsg(IPFW_CFGPORT, lmsg, 0);
4246 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
4248 return sysctl_int_range(oidp, arg1, arg2, req,
4249 IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
4253 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
4257 lockmgr(&dyn_lock, LK_EXCLUSIVE);
4259 value = dyn_buckets;
4260 error = sysctl_handle_int(oidp, &value, 0, req);
4261 if (error || !req->newptr)
4265 * Make sure we have a power of 2 and
4266 * do not allow more than 64k entries.
4269 if (value <= 1 || value > 65536)
4271 if ((value & (value - 1)) != 0)
4275 dyn_buckets = value;
4277 lockmgr(&dyn_lock, LK_RELEASE);
4282 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
4284 return sysctl_int_range(oidp, arg1, arg2, req,
4285 1, dyn_keepalive_period - 1);
4289 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
4291 return sysctl_int_range(oidp, arg1, arg2, req,
4292 1, dyn_keepalive_period - 1);
4296 ipfw_ctx_init_dispatch(struct netmsg *nmsg)
4298 struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
4299 struct ipfw_context *ctx;
4300 struct ip_fw *def_rule;
4302 ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO);
4303 ipfw_ctx[mycpuid] = ctx;
4305 def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
4307 def_rule->act_ofs = 0;
4308 def_rule->rulenum = IPFW_DEFAULT_RULE;
4309 def_rule->cmd_len = 1;
4310 def_rule->set = IPFW_DEFAULT_SET;
4312 def_rule->cmd[0].len = 1;
4313 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4314 def_rule->cmd[0].opcode = O_ACCEPT;
4316 def_rule->cmd[0].opcode = O_DENY;
4319 def_rule->refcnt = 1;
4320 def_rule->cpuid = mycpuid;
4322 /* Install the default rule */
4323 ctx->ipfw_default_rule = def_rule;
4324 ctx->ipfw_layer3_chain = def_rule;
4326 /* Link rule CPU sibling */
4327 ipfw_link_sibling(fwmsg, def_rule);
4329 /* Statistics only need to be updated once */
4331 ipfw_inc_static_count(def_rule);
4333 ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
4337 ipfw_init_dispatch(struct netmsg *nmsg)
4339 struct netmsg_ipfw fwmsg;
4345 kprintf("IP firewall already loaded\n");
4350 bzero(&fwmsg, sizeof(fwmsg));
4351 netmsg_init(&fwmsg.nmsg, &curthread->td_msgport, 0,
4352 ipfw_ctx_init_dispatch);
4353 ifnet_domsg(&fwmsg.nmsg.nm_lmsg, 0);
4355 ip_fw_chk_ptr = ipfw_chk;
4356 ip_fw_ctl_ptr = ipfw_ctl;
4357 ip_fw_dn_io_ptr = ipfw_dummynet_io;
4359 kprintf("ipfw2 initialized, default to %s, logging ",
4360 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode ==
4361 O_ACCEPT ? "accept" : "deny");
4363 #ifdef IPFIREWALL_VERBOSE
4366 #ifdef IPFIREWALL_VERBOSE_LIMIT
4367 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4369 if (fw_verbose == 0) {
4370 kprintf("disabled\n");
4371 } else if (verbose_limit == 0) {
4372 kprintf("unlimited\n");
4374 kprintf("limited to %d packets/entry by default\n",
4378 callout_init(&ipfw_timeout_h);
4379 lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
4382 callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
4388 lwkt_replymsg(&nmsg->nm_lmsg, error);
4396 netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_init_dispatch);
4397 return lwkt_domsg(IPFW_CFGPORT, &smsg.nm_lmsg, 0);
4403 ipfw_fini_dispatch(struct netmsg *nmsg)
4409 if (ipfw_refcnt != 0) {
4416 callout_stop(&ipfw_timeout_h);
4419 netmsg_service_sync();
4421 ip_fw_chk_ptr = NULL;
4422 ip_fw_ctl_ptr = NULL;
4423 ip_fw_dn_io_ptr = NULL;
4424 ipfw_flush(1 /* kill default rule */);
4426 /* Free pre-cpu context */
4427 for (cpu = 0; cpu < ncpus; ++cpu)
4428 kfree(ipfw_ctx[cpu], M_IPFW);
4430 kprintf("IP firewall unloaded\n");
4433 lwkt_replymsg(&nmsg->nm_lmsg, error);
4441 netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_fini_dispatch);
4442 return lwkt_domsg(IPFW_CFGPORT, &smsg.nm_lmsg, 0);
4445 #endif /* KLD_MODULE */
4448 ipfw_modevent(module_t mod, int type, void *unused)
4459 kprintf("ipfw statically compiled, cannot unload\n");
4471 static moduledata_t ipfwmod = {
4476 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
4477 MODULE_VERSION(ipfw, 1);