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
32 * Implement IP packet firewall (new version)
35 #if !defined(KLD_MODULE)
37 #include "opt_ip6fw.h"
40 #include "opt_inet6.h"
41 #include "opt_ipsec.h"
43 #error IPFIREWALL requires INET.
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/condvar.h>
50 #include <sys/malloc.h>
52 #include <sys/kernel.h>
54 #include <sys/module.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/sysctl.h>
59 #include <sys/syslog.h>
60 #include <sys/ucred.h>
62 #include <net/radix.h>
63 #include <net/route.h>
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_icmp.h>
71 #include <netinet/ip_fw.h>
72 #include <netinet/ip_divert.h>
73 #include <netinet/ip_dummynet.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_timer.h>
76 #include <netinet/tcp_var.h>
77 #include <netinet/tcpip.h>
78 #include <netinet/udp.h>
79 #include <netinet/udp_var.h>
81 #include <netgraph/ng_ipfw.h>
83 #include <altq/if_altq.h>
86 #include <netinet6/ipsec.h>
89 #include <netinet/ip6.h>
90 #include <netinet/icmp6.h>
92 #include <netinet6/scope6_var.h>
95 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
97 #include <machine/in_cksum.h> /* XXX for in_cksum */
100 * set_disable contains one bit per set value (0..31).
101 * If the bit is set, all rules with the corresponding set
102 * are disabled. Set RESVD_SET(31) is reserved for the default rule
103 * and rules that are not deleted by the flush command,
104 * and CANNOT be disabled.
105 * Rules in set RESVD_SET can only be deleted explicitly.
107 static u_int32_t set_disable;
109 static int fw_verbose;
110 static int verbose_limit;
112 static struct callout ipfw_timeout;
113 static uma_zone_t ipfw_dyn_rule_zone;
114 #define IPFW_DEFAULT_RULE 65535
117 * Data structure to cache our ucred related
118 * information. This structure only gets used if
119 * the user specified UID/GID based constraints in
123 gid_t fw_groups[NGROUPS];
130 struct ip_fw *rules; /* list of rules */
131 struct ip_fw *reap; /* list of rules to reap */
132 struct mtx mtx; /* lock guarding rule list */
133 int busy_count; /* busy count for rw locks */
137 #define IPFW_LOCK_INIT(_chain) \
138 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
139 MTX_DEF | MTX_RECURSE)
140 #define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
141 #define IPFW_WLOCK_ASSERT(_chain) do { \
142 mtx_assert(&(_chain)->mtx, MA_OWNED); \
143 NET_ASSERT_GIANT(); \
147 IPFW_RLOCK(struct ip_fw_chain *chain)
149 mtx_lock(&chain->mtx);
151 mtx_unlock(&chain->mtx);
155 IPFW_RUNLOCK(struct ip_fw_chain *chain)
157 mtx_lock(&chain->mtx);
159 if (chain->busy_count == 0 && chain->want_write)
160 cv_signal(&chain->cv);
161 mtx_unlock(&chain->mtx);
165 IPFW_WLOCK(struct ip_fw_chain *chain)
167 mtx_lock(&chain->mtx);
169 while (chain->busy_count > 0)
170 cv_wait(&chain->cv, &chain->mtx);
174 IPFW_WUNLOCK(struct ip_fw_chain *chain)
177 cv_signal(&chain->cv);
178 mtx_unlock(&chain->mtx);
182 * list of rules for layer 3
184 static struct ip_fw_chain layer3_chain;
186 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
187 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
190 struct radix_node rn[2];
191 struct sockaddr_in addr, mask;
195 #define IPFW_TABLES_MAX 128
196 static struct ip_fw_table {
197 struct radix_node_head *rnh;
201 u_int32_t last_value;
202 } ipfw_tables[IPFW_TABLES_MAX];
204 static int fw_debug = 1;
205 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
208 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
209 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
210 CTLFLAG_RW | CTLFLAG_SECURE3,
211 &fw_enable, 0, "Enable ipfw");
212 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
213 &autoinc_step, 0, "Rule number autincrement step");
214 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
215 CTLFLAG_RW | CTLFLAG_SECURE3,
217 "Only do a single pass through ipfw when using dummynet(4)");
218 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
219 &fw_debug, 0, "Enable printing of debug ip_fw statements");
220 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
221 CTLFLAG_RW | CTLFLAG_SECURE3,
222 &fw_verbose, 0, "Log matches to ipfw rules");
223 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
224 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
227 * Description of dynamic rules.
229 * Dynamic rules are stored in lists accessed through a hash table
230 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
231 * be modified through the sysctl variable dyn_buckets which is
232 * updated when the table becomes empty.
234 * XXX currently there is only one list, ipfw_dyn.
236 * When a packet is received, its address fields are first masked
237 * with the mask defined for the rule, then hashed, then matched
238 * against the entries in the corresponding list.
239 * Dynamic rules can be used for different purposes:
241 * + enforcing limits on the number of sessions;
242 * + in-kernel NAT (not implemented yet)
244 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
245 * measured in seconds and depending on the flags.
247 * The total number of dynamic rules is stored in dyn_count.
248 * The max number of dynamic rules is dyn_max. When we reach
249 * the maximum number of rules we do not create anymore. This is
250 * done to avoid consuming too much memory, but also too much
251 * time when searching on each packet (ideally, we should try instead
252 * to put a limit on the length of the list on each bucket...).
254 * Each dynamic rule holds a pointer to the parent ipfw rule so
255 * we know what action to perform. Dynamic rules are removed when
256 * the parent rule is deleted. XXX we should make them survive.
258 * There are some limitations with dynamic rules -- we do not
259 * obey the 'randomized match', and we do not do multiple
260 * passes through the firewall. XXX check the latter!!!
262 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
263 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
264 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
266 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
267 #define IPFW_DYN_LOCK_INIT() \
268 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
269 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
270 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
271 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
272 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
275 * Timeouts for various events in handing dynamic rules.
277 static u_int32_t dyn_ack_lifetime = 300;
278 static u_int32_t dyn_syn_lifetime = 20;
279 static u_int32_t dyn_fin_lifetime = 1;
280 static u_int32_t dyn_rst_lifetime = 1;
281 static u_int32_t dyn_udp_lifetime = 10;
282 static u_int32_t dyn_short_lifetime = 5;
285 * Keepalives are sent if dyn_keepalive is set. They are sent every
286 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
287 * seconds of lifetime of a rule.
288 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
289 * than dyn_keepalive_period.
292 static u_int32_t dyn_keepalive_interval = 20;
293 static u_int32_t dyn_keepalive_period = 5;
294 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
296 static u_int32_t static_count; /* # of static rules */
297 static u_int32_t static_len; /* size in bytes of static rules */
298 static u_int32_t dyn_count; /* # of dynamic rules */
299 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
301 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
302 &dyn_buckets, 0, "Number of dyn. buckets");
303 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
304 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
305 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
306 &dyn_count, 0, "Number of dyn. rules");
307 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
308 &dyn_max, 0, "Max number of dyn. rules");
309 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
310 &static_count, 0, "Number of static rules");
311 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
312 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
313 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
314 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
315 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
316 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
317 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
318 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
319 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
320 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
321 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
322 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
323 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
324 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
328 * IPv6 specific variables
330 SYSCTL_DECL(_net_inet6_ip6);
332 static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
333 static struct sysctl_oid *ip6_fw_sysctl_tree;
335 #endif /* SYSCTL_NODE */
337 static int fw_deny_unknown_exthdrs = 1;
341 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
342 * Other macros just cast void * into the appropriate type
344 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
345 #define TCP(p) ((struct tcphdr *)(p))
346 #define UDP(p) ((struct udphdr *)(p))
347 #define ICMP(p) ((struct icmphdr *)(p))
348 #define ICMP6(p) ((struct icmp6_hdr *)(p))
351 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
353 int type = icmp->icmp_type;
355 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
358 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
359 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
362 is_icmp_query(struct icmphdr *icmp)
364 int type = icmp->icmp_type;
366 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
371 * The following checks use two arrays of 8 or 16 bits to store the
372 * bits that we want set or clear, respectively. They are in the
373 * low and high half of cmd->arg1 or cmd->d[0].
375 * We scan options and store the bits we find set. We succeed if
377 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
379 * The code is sometimes optimized not to store additional variables.
383 flags_match(ipfw_insn *cmd, u_int8_t bits)
388 if ( ((cmd->arg1 & 0xff) & bits) != 0)
389 return 0; /* some bits we want set were clear */
390 want_clear = (cmd->arg1 >> 8) & 0xff;
391 if ( (want_clear & bits) != want_clear)
392 return 0; /* some bits we want clear were set */
397 ipopts_match(struct ip *ip, ipfw_insn *cmd)
399 int optlen, bits = 0;
400 u_char *cp = (u_char *)(ip + 1);
401 int x = (ip->ip_hl << 2) - sizeof (struct ip);
403 for (; x > 0; x -= optlen, cp += optlen) {
404 int opt = cp[IPOPT_OPTVAL];
406 if (opt == IPOPT_EOL)
408 if (opt == IPOPT_NOP)
411 optlen = cp[IPOPT_OLEN];
412 if (optlen <= 0 || optlen > x)
413 return 0; /* invalid or truncated */
421 bits |= IP_FW_IPOPT_LSRR;
425 bits |= IP_FW_IPOPT_SSRR;
429 bits |= IP_FW_IPOPT_RR;
433 bits |= IP_FW_IPOPT_TS;
437 return (flags_match(cmd, bits));
441 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
443 int optlen, bits = 0;
444 u_char *cp = (u_char *)(tcp + 1);
445 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
447 for (; x > 0; x -= optlen, cp += optlen) {
449 if (opt == TCPOPT_EOL)
451 if (opt == TCPOPT_NOP)
465 bits |= IP_FW_TCPOPT_MSS;
469 bits |= IP_FW_TCPOPT_WINDOW;
472 case TCPOPT_SACK_PERMITTED:
474 bits |= IP_FW_TCPOPT_SACK;
477 case TCPOPT_TIMESTAMP:
478 bits |= IP_FW_TCPOPT_TS;
483 return (flags_match(cmd, bits));
487 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
489 if (ifp == NULL) /* no iface with this packet, match fails */
491 /* Check by name or by IP address */
492 if (cmd->name[0] != '\0') { /* match by name */
495 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
498 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
505 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
506 if (ia->ifa_addr == NULL)
508 if (ia->ifa_addr->sa_family != AF_INET)
510 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
511 (ia->ifa_addr))->sin_addr.s_addr)
512 return(1); /* match */
515 return(0); /* no match, fail ... */
519 * The verify_path function checks if a route to the src exists and
520 * if it is reachable via ifp (when provided).
522 * The 'verrevpath' option checks that the interface that an IP packet
523 * arrives on is the same interface that traffic destined for the
524 * packet's source address would be routed out of. The 'versrcreach'
525 * option just checks that the source address is reachable via any route
526 * (except default) in the routing table. These two are a measure to block
527 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
528 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
529 * is purposely reminiscent of the Cisco IOS command,
531 * ip verify unicast reverse-path
532 * ip verify unicast source reachable-via any
534 * which implements the same functionality. But note that syntax is
535 * misleading. The check may be performed on all IP packets whether unicast,
536 * multicast, or broadcast.
539 verify_path(struct in_addr src, struct ifnet *ifp)
542 struct sockaddr_in *dst;
544 bzero(&ro, sizeof(ro));
546 dst = (struct sockaddr_in *)&(ro.ro_dst);
547 dst->sin_family = AF_INET;
548 dst->sin_len = sizeof(*dst);
550 rtalloc_ign(&ro, RTF_CLONING);
552 if (ro.ro_rt == NULL)
555 /* if ifp is provided, check for equality with rtentry */
556 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) {
561 /* if no ifp provided, check if rtentry is not default route */
563 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
568 /* or if this is a blackhole/reject route */
569 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
574 /* found valid route */
581 * ipv6 specific rules here...
584 icmp6type_match (int type, ipfw_insn_u32 *cmd)
586 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
590 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
593 for (i=0; i <= cmd->o.arg1; ++i )
594 if (curr_flow == cmd->d[i] )
599 /* support for IP6_*_ME opcodes */
601 search_ip6_addr_net (struct in6_addr * ip6_addr)
605 struct in6_ifaddr *fdm;
606 struct in6_addr copia;
608 TAILQ_FOREACH(mdc, &ifnet, if_link)
609 for (mdc2 = mdc->if_addrlist.tqh_first; mdc2;
610 mdc2 = mdc2->ifa_list.tqe_next) {
613 if (mdc2->ifa_addr->sa_family == AF_INET6) {
614 fdm = (struct in6_ifaddr *)mdc2;
615 copia = fdm->ia_addr.sin6_addr;
616 /* need for leaving scope_id in the sock_addr */
617 in6_clearscope(&copia);
618 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
626 verify_path6(struct in6_addr *src, struct ifnet *ifp)
629 struct sockaddr_in6 *dst;
631 bzero(&ro, sizeof(ro));
633 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
634 dst->sin6_family = AF_INET6;
635 dst->sin6_len = sizeof(*dst);
636 dst->sin6_addr = *src;
637 rtalloc_ign((struct route *)&ro, RTF_CLONING);
639 if (ro.ro_rt == NULL)
643 * if ifp is provided, check for equality with rtentry
644 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
645 * to support the case of sending packets to an address of our own.
646 * (where the former interface is the first argument of if_simloop()
647 * (=ifp), the latter is lo0)
649 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
654 /* if no ifp provided, check if rtentry is not default route */
656 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
661 /* or if this is a blackhole/reject route */
662 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
667 /* found valid route */
673 hash_packet6(struct ipfw_flow_id *id)
676 i = (id->dst_ip6.__u6_addr.__u6_addr32[0]) ^
677 (id->dst_ip6.__u6_addr.__u6_addr32[1]) ^
678 (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
679 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
680 (id->dst_port) ^ (id->src_port) ^ (id->flow_id6);
685 is_icmp6_query(int icmp6_type)
687 if ((icmp6_type <= ICMP6_MAXTYPE) &&
688 (icmp6_type == ICMP6_ECHO_REQUEST ||
689 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
690 icmp6_type == ICMP6_WRUREQUEST ||
691 icmp6_type == ICMP6_FQDN_QUERY ||
692 icmp6_type == ICMP6_NI_QUERY))
699 send_reject6(struct ip_fw_args *args, int code, u_short offset, u_int hlen)
701 if (code == ICMP6_UNREACH_RST && offset == 0 &&
702 args->f_id.proto == IPPROTO_TCP) {
712 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
713 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
718 ip6 = mtod(args->m, struct ip6_hdr *);
719 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
721 if ((tcp->th_flags & TH_RST) != 0) {
728 ti.th.th_seq = ntohl(ti.th.th_seq);
729 ti.th.th_ack = ntohl(ti.th.th_ack);
730 ti.ip6.ip6_nxt = IPPROTO_TCP;
732 if (ti.th.th_flags & TH_ACK) {
738 if (((args->m)->m_flags & M_PKTHDR) != 0) {
739 ack += (args->m)->m_pkthdr.len - hlen
740 - (ti.th.th_off << 2);
741 } else if (ip6->ip6_plen) {
742 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
743 - hlen - (ti.th.th_off << 2);
748 if (tcp->th_flags & TH_SYN)
751 flags = TH_RST|TH_ACK;
753 bcopy(&ti, ip6, sizeof(ti));
754 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
755 args->m, ack, seq, flags);
757 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
758 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
768 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
770 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
771 #define SNP(buf) buf, sizeof(buf)
774 * We enter here when we have a rule with O_LOG.
775 * XXX this function alone takes about 2Kbytes of code!
778 ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
779 struct mbuf *m, struct ifnet *oif, u_short offset)
781 struct ether_header *eh = args->eh;
783 int limit_reached = 0;
784 char action2[40], proto[128], fragment[32];
789 if (f == NULL) { /* bogus pkt */
790 if (verbose_limit != 0 && norule_counter >= verbose_limit)
793 if (norule_counter == verbose_limit)
794 limit_reached = verbose_limit;
796 } else { /* O_LOG is the first action, find the real one */
797 ipfw_insn *cmd = ACTION_PTR(f);
798 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
800 if (l->max_log != 0 && l->log_left == 0)
803 if (l->log_left == 0)
804 limit_reached = l->max_log;
805 cmd += F_LEN(cmd); /* point to first action */
806 if (cmd->opcode == O_ALTQ) {
807 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
809 snprintf(SNPARGS(action2, 0), "Altq %d",
813 if (cmd->opcode == O_PROB)
817 switch (cmd->opcode) {
823 if (cmd->arg1==ICMP_REJECT_RST)
825 else if (cmd->arg1==ICMP_UNREACH_HOST)
828 snprintf(SNPARGS(action2, 0), "Unreach %d",
833 if (cmd->arg1==ICMP6_UNREACH_RST)
836 snprintf(SNPARGS(action2, 0), "Unreach %d",
847 snprintf(SNPARGS(action2, 0), "Divert %d",
851 snprintf(SNPARGS(action2, 0), "Tee %d",
855 snprintf(SNPARGS(action2, 0), "SkipTo %d",
859 snprintf(SNPARGS(action2, 0), "Pipe %d",
863 snprintf(SNPARGS(action2, 0), "Queue %d",
867 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
870 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
871 inet_ntoa(sa->sa.sin_addr));
873 snprintf(SNPARGS(action2, len), ":%d",
878 snprintf(SNPARGS(action2, 0), "Netgraph %d",
882 snprintf(SNPARGS(action2, 0), "Ngtee %d",
891 if (hlen == 0) { /* non-ip */
892 snprintf(SNPARGS(proto, 0), "MAC");
896 char src[48], dst[48];
897 struct icmphdr *icmp;
900 /* Initialize to make compiler happy. */
901 struct ip *ip = NULL;
903 struct ip6_hdr *ip6 = NULL;
904 struct icmp6_hdr *icmp6;
909 if (args->f_id.addr_type == 6) {
910 snprintf(src, sizeof(src), "[%s]",
911 ip6_sprintf(&args->f_id.src_ip6));
912 snprintf(dst, sizeof(dst), "[%s]",
913 ip6_sprintf(&args->f_id.dst_ip6));
915 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
916 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
917 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
921 ip = mtod(m, struct ip *);
922 tcp = L3HDR(struct tcphdr, ip);
923 udp = L3HDR(struct udphdr, ip);
925 inet_ntoa_r(ip->ip_src, src);
926 inet_ntoa_r(ip->ip_dst, dst);
929 switch (args->f_id.proto) {
931 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
933 snprintf(SNPARGS(proto, len), ":%d %s:%d",
934 ntohs(tcp->th_sport),
936 ntohs(tcp->th_dport));
938 snprintf(SNPARGS(proto, len), " %s", dst);
942 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
944 snprintf(SNPARGS(proto, len), ":%d %s:%d",
945 ntohs(udp->uh_sport),
947 ntohs(udp->uh_dport));
949 snprintf(SNPARGS(proto, len), " %s", dst);
953 icmp = L3HDR(struct icmphdr, ip);
955 len = snprintf(SNPARGS(proto, 0),
957 icmp->icmp_type, icmp->icmp_code);
959 len = snprintf(SNPARGS(proto, 0), "ICMP ");
960 len += snprintf(SNPARGS(proto, len), "%s", src);
961 snprintf(SNPARGS(proto, len), " %s", dst);
965 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
967 len = snprintf(SNPARGS(proto, 0),
969 icmp6->icmp6_type, icmp6->icmp6_code);
971 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
972 len += snprintf(SNPARGS(proto, len), "%s", src);
973 snprintf(SNPARGS(proto, len), " %s", dst);
977 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
978 args->f_id.proto, src);
979 snprintf(SNPARGS(proto, len), " %s", dst);
984 if (args->f_id.addr_type == 6) {
985 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
986 snprintf(SNPARGS(fragment, 0),
987 " (frag %08x:%d@%d%s)",
989 ntohs(ip6->ip6_plen) - hlen,
990 ntohs(offset & IP6F_OFF_MASK) << 3,
991 (offset & IP6F_MORE_FRAG) ? "+" : "");
996 if (eh != NULL) { /* layer 2 packets are as on the wire */
997 ip_off = ntohs(ip->ip_off);
998 ip_len = ntohs(ip->ip_len);
1000 ip_off = ip->ip_off;
1001 ip_len = ip->ip_len;
1003 if (ip_off & (IP_MF | IP_OFFMASK))
1004 snprintf(SNPARGS(fragment, 0),
1005 " (frag %d:%d@%d%s)",
1006 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
1008 (ip_off & IP_MF) ? "+" : "");
1011 if (oif || m->m_pkthdr.rcvif)
1012 log(LOG_SECURITY | LOG_INFO,
1013 "ipfw: %d %s %s %s via %s%s\n",
1014 f ? f->rulenum : -1,
1015 action, proto, oif ? "out" : "in",
1016 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
1019 log(LOG_SECURITY | LOG_INFO,
1020 "ipfw: %d %s %s [no if info]%s\n",
1021 f ? f->rulenum : -1,
1022 action, proto, fragment);
1024 log(LOG_SECURITY | LOG_NOTICE,
1025 "ipfw: limit %d reached on entry %d\n",
1026 limit_reached, f ? f->rulenum : -1);
1030 * IMPORTANT: the hash function for dynamic rules must be commutative
1031 * in source and destination (ip,port), because rules are bidirectional
1032 * and we want to find both in the same bucket.
1035 hash_packet(struct ipfw_flow_id *id)
1040 if (IS_IP6_FLOW_ID(id))
1041 i = hash_packet6(id);
1044 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1045 i &= (curr_dyn_buckets - 1);
1050 * unlink a dynamic rule from a chain. prev is a pointer to
1051 * the previous one, q is a pointer to the rule to delete,
1052 * head is a pointer to the head of the queue.
1053 * Modifies q and potentially also head.
1055 #define UNLINK_DYN_RULE(prev, head, q) { \
1056 ipfw_dyn_rule *old_q = q; \
1058 /* remove a refcount to the parent */ \
1059 if (q->dyn_type == O_LIMIT) \
1060 q->parent->count--; \
1061 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1062 (q->id.src_ip), (q->id.src_port), \
1063 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1065 prev->next = q = q->next; \
1067 head = q = q->next; \
1069 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1071 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1074 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1076 * If keep_me == NULL, rules are deleted even if not expired,
1077 * otherwise only expired rules are removed.
1079 * The value of the second parameter is also used to point to identify
1080 * a rule we absolutely do not want to remove (e.g. because we are
1081 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1082 * rules). The pointer is only used for comparison, so any non-null
1086 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1088 static u_int32_t last_remove = 0;
1090 #define FORCE (keep_me == NULL)
1092 ipfw_dyn_rule *prev, *q;
1093 int i, pass = 0, max_pass = 0;
1095 IPFW_DYN_LOCK_ASSERT();
1097 if (ipfw_dyn_v == NULL || dyn_count == 0)
1099 /* do not expire more than once per second, it is useless */
1100 if (!FORCE && last_remove == time_uptime)
1102 last_remove = time_uptime;
1105 * because O_LIMIT refer to parent rules, during the first pass only
1106 * remove child and mark any pending LIMIT_PARENT, and remove
1107 * them in a second pass.
1110 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1111 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1113 * Logic can become complex here, so we split tests.
1117 if (rule != NULL && rule != q->rule)
1118 goto next; /* not the one we are looking for */
1119 if (q->dyn_type == O_LIMIT_PARENT) {
1121 * handle parent in the second pass,
1122 * record we need one.
1127 if (FORCE && q->count != 0 ) {
1128 /* XXX should not happen! */
1129 printf("ipfw: OUCH! cannot remove rule,"
1130 " count %d\n", q->count);
1134 !TIME_LEQ( q->expire, time_uptime ))
1137 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1138 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1146 if (pass++ < max_pass)
1152 * lookup a dynamic rule.
1154 static ipfw_dyn_rule *
1155 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1159 * stateful ipfw extensions.
1160 * Lookup into dynamic session queue
1162 #define MATCH_REVERSE 0
1163 #define MATCH_FORWARD 1
1164 #define MATCH_NONE 2
1165 #define MATCH_UNKNOWN 3
1166 int i, dir = MATCH_NONE;
1167 ipfw_dyn_rule *prev, *q=NULL;
1169 IPFW_DYN_LOCK_ASSERT();
1171 if (ipfw_dyn_v == NULL)
1172 goto done; /* not found */
1173 i = hash_packet( pkt );
1174 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1175 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1177 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1178 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1181 if (pkt->proto == q->id.proto &&
1182 q->dyn_type != O_LIMIT_PARENT) {
1183 if (IS_IP6_FLOW_ID(pkt)) {
1184 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1185 &(q->id.src_ip6)) &&
1186 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1187 &(q->id.dst_ip6)) &&
1188 pkt->src_port == q->id.src_port &&
1189 pkt->dst_port == q->id.dst_port ) {
1190 dir = MATCH_FORWARD;
1193 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1194 &(q->id.dst_ip6)) &&
1195 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1196 &(q->id.src_ip6)) &&
1197 pkt->src_port == q->id.dst_port &&
1198 pkt->dst_port == q->id.src_port ) {
1199 dir = MATCH_REVERSE;
1203 if (pkt->src_ip == q->id.src_ip &&
1204 pkt->dst_ip == q->id.dst_ip &&
1205 pkt->src_port == q->id.src_port &&
1206 pkt->dst_port == q->id.dst_port ) {
1207 dir = MATCH_FORWARD;
1210 if (pkt->src_ip == q->id.dst_ip &&
1211 pkt->dst_ip == q->id.src_ip &&
1212 pkt->src_port == q->id.dst_port &&
1213 pkt->dst_port == q->id.src_port ) {
1214 dir = MATCH_REVERSE;
1224 goto done; /* q = NULL, not found */
1226 if ( prev != NULL) { /* found and not in front */
1227 prev->next = q->next;
1228 q->next = ipfw_dyn_v[i];
1231 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1232 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1234 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1235 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1236 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1238 case TH_SYN: /* opening */
1239 q->expire = time_uptime + dyn_syn_lifetime;
1242 case BOTH_SYN: /* move to established */
1243 case BOTH_SYN | TH_FIN : /* one side tries to close */
1244 case BOTH_SYN | (TH_FIN << 8) :
1246 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1247 u_int32_t ack = ntohl(tcp->th_ack);
1248 if (dir == MATCH_FORWARD) {
1249 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1251 else { /* ignore out-of-sequence */
1255 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1257 else { /* ignore out-of-sequence */
1262 q->expire = time_uptime + dyn_ack_lifetime;
1265 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1266 if (dyn_fin_lifetime >= dyn_keepalive_period)
1267 dyn_fin_lifetime = dyn_keepalive_period - 1;
1268 q->expire = time_uptime + dyn_fin_lifetime;
1274 * reset or some invalid combination, but can also
1275 * occur if we use keep-state the wrong way.
1277 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1278 printf("invalid state: 0x%x\n", q->state);
1280 if (dyn_rst_lifetime >= dyn_keepalive_period)
1281 dyn_rst_lifetime = dyn_keepalive_period - 1;
1282 q->expire = time_uptime + dyn_rst_lifetime;
1285 } else if (pkt->proto == IPPROTO_UDP) {
1286 q->expire = time_uptime + dyn_udp_lifetime;
1288 /* other protocols */
1289 q->expire = time_uptime + dyn_short_lifetime;
1292 if (match_direction)
1293 *match_direction = dir;
1297 static ipfw_dyn_rule *
1298 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1304 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1307 /* NB: return table locked when q is not NULL */
1312 realloc_dynamic_table(void)
1314 IPFW_DYN_LOCK_ASSERT();
1317 * Try reallocation, make sure we have a power of 2 and do
1318 * not allow more than 64k entries. In case of overflow,
1322 if (dyn_buckets > 65536)
1324 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1325 dyn_buckets = curr_dyn_buckets; /* reset */
1328 curr_dyn_buckets = dyn_buckets;
1329 if (ipfw_dyn_v != NULL)
1330 free(ipfw_dyn_v, M_IPFW);
1332 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1333 M_IPFW, M_NOWAIT | M_ZERO);
1334 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1336 curr_dyn_buckets /= 2;
1341 * Install state of type 'type' for a dynamic session.
1342 * The hash table contains two type of rules:
1343 * - regular rules (O_KEEP_STATE)
1344 * - rules for sessions with limited number of sess per user
1345 * (O_LIMIT). When they are created, the parent is
1346 * increased by 1, and decreased on delete. In this case,
1347 * the third parameter is the parent rule and not the chain.
1348 * - "parent" rules for the above (O_LIMIT_PARENT).
1350 static ipfw_dyn_rule *
1351 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1356 IPFW_DYN_LOCK_ASSERT();
1358 if (ipfw_dyn_v == NULL ||
1359 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1360 realloc_dynamic_table();
1361 if (ipfw_dyn_v == NULL)
1362 return NULL; /* failed ! */
1364 i = hash_packet(id);
1366 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1368 printf ("ipfw: sorry cannot allocate state\n");
1372 /* increase refcount on parent, and set pointer */
1373 if (dyn_type == O_LIMIT) {
1374 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1375 if ( parent->dyn_type != O_LIMIT_PARENT)
1376 panic("invalid parent");
1379 rule = parent->rule;
1383 r->expire = time_uptime + dyn_syn_lifetime;
1385 r->dyn_type = dyn_type;
1386 r->pcnt = r->bcnt = 0;
1390 r->next = ipfw_dyn_v[i];
1393 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1395 (r->id.src_ip), (r->id.src_port),
1396 (r->id.dst_ip), (r->id.dst_port),
1402 * lookup dynamic parent rule using pkt and rule as search keys.
1403 * If the lookup fails, then install one.
1405 static ipfw_dyn_rule *
1406 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1411 IPFW_DYN_LOCK_ASSERT();
1414 int is_v6 = IS_IP6_FLOW_ID(pkt);
1415 i = hash_packet( pkt );
1416 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1417 if (q->dyn_type == O_LIMIT_PARENT &&
1419 pkt->proto == q->id.proto &&
1420 pkt->src_port == q->id.src_port &&
1421 pkt->dst_port == q->id.dst_port &&
1424 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1425 &(q->id.src_ip6)) &&
1426 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1427 &(q->id.dst_ip6))) ||
1429 pkt->src_ip == q->id.src_ip &&
1430 pkt->dst_ip == q->id.dst_ip)
1433 q->expire = time_uptime + dyn_short_lifetime;
1434 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1438 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1442 * Install dynamic state for rule type cmd->o.opcode
1444 * Returns 1 (failure) if state is not installed because of errors or because
1445 * session limitations are enforced.
1448 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1449 struct ip_fw_args *args)
1451 static int last_log;
1455 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1457 (args->f_id.src_ip), (args->f_id.src_port),
1458 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1462 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1464 if (q != NULL) { /* should never occur */
1465 if (last_log != time_uptime) {
1466 last_log = time_uptime;
1467 printf("ipfw: install_state: entry already present, done\n");
1473 if (dyn_count >= dyn_max)
1475 * Run out of slots, try to remove any expired rule.
1477 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1479 if (dyn_count >= dyn_max) {
1480 if (last_log != time_uptime) {
1481 last_log = time_uptime;
1482 printf("ipfw: install_state: Too many dynamic rules\n");
1485 return 1; /* cannot install, notify caller */
1488 switch (cmd->o.opcode) {
1489 case O_KEEP_STATE: /* bidir rule */
1490 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1493 case O_LIMIT: /* limit number of sessions */
1495 u_int16_t limit_mask = cmd->limit_mask;
1496 struct ipfw_flow_id id;
1497 ipfw_dyn_rule *parent;
1499 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1502 id.dst_ip = id.src_ip = 0;
1503 id.dst_port = id.src_port = 0;
1504 id.proto = args->f_id.proto;
1506 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1507 if (limit_mask & DYN_SRC_ADDR)
1508 id.src_ip6 = args->f_id.src_ip6;
1509 if (limit_mask & DYN_DST_ADDR)
1510 id.dst_ip6 = args->f_id.dst_ip6;
1512 if (limit_mask & DYN_SRC_ADDR)
1513 id.src_ip = args->f_id.src_ip;
1514 if (limit_mask & DYN_DST_ADDR)
1515 id.dst_ip = args->f_id.dst_ip;
1517 if (limit_mask & DYN_SRC_PORT)
1518 id.src_port = args->f_id.src_port;
1519 if (limit_mask & DYN_DST_PORT)
1520 id.dst_port = args->f_id.dst_port;
1521 parent = lookup_dyn_parent(&id, rule);
1522 if (parent == NULL) {
1523 printf("ipfw: add parent failed\n");
1527 if (parent->count >= cmd->conn_limit) {
1529 * See if we can remove some expired rule.
1531 remove_dyn_rule(rule, parent);
1532 if (parent->count >= cmd->conn_limit) {
1533 if (fw_verbose && last_log != time_uptime) {
1534 last_log = time_uptime;
1535 log(LOG_SECURITY | LOG_DEBUG,
1536 "drop session, too many entries\n");
1542 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1546 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1550 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1556 * Generate a TCP packet, containing either a RST or a keepalive.
1557 * When flags & TH_RST, we are sending a RST packet, because of a
1558 * "reset" action matched the packet.
1559 * Otherwise we are sending a keepalive, and flags & TH_
1561 static struct mbuf *
1562 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1568 MGETHDR(m, M_DONTWAIT, MT_DATA);
1571 m->m_pkthdr.rcvif = (struct ifnet *)0;
1572 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1573 m->m_data += max_linkhdr;
1575 ip = mtod(m, struct ip *);
1576 bzero(ip, m->m_len);
1577 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1578 ip->ip_p = IPPROTO_TCP;
1581 * Assume we are sending a RST (or a keepalive in the reverse
1582 * direction), swap src and destination addresses and ports.
1584 ip->ip_src.s_addr = htonl(id->dst_ip);
1585 ip->ip_dst.s_addr = htonl(id->src_ip);
1586 tcp->th_sport = htons(id->dst_port);
1587 tcp->th_dport = htons(id->src_port);
1588 if (flags & TH_RST) { /* we are sending a RST */
1589 if (flags & TH_ACK) {
1590 tcp->th_seq = htonl(ack);
1591 tcp->th_ack = htonl(0);
1592 tcp->th_flags = TH_RST;
1596 tcp->th_seq = htonl(0);
1597 tcp->th_ack = htonl(seq);
1598 tcp->th_flags = TH_RST | TH_ACK;
1602 * We are sending a keepalive. flags & TH_SYN determines
1603 * the direction, forward if set, reverse if clear.
1604 * NOTE: seq and ack are always assumed to be correct
1605 * as set by the caller. This may be confusing...
1607 if (flags & TH_SYN) {
1609 * we have to rewrite the correct addresses!
1611 ip->ip_dst.s_addr = htonl(id->dst_ip);
1612 ip->ip_src.s_addr = htonl(id->src_ip);
1613 tcp->th_dport = htons(id->dst_port);
1614 tcp->th_sport = htons(id->src_port);
1616 tcp->th_seq = htonl(seq);
1617 tcp->th_ack = htonl(ack);
1618 tcp->th_flags = TH_ACK;
1621 * set ip_len to the payload size so we can compute
1622 * the tcp checksum on the pseudoheader
1623 * XXX check this, could save a couple of words ?
1625 ip->ip_len = htons(sizeof(struct tcphdr));
1626 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1628 * now fill fields left out earlier
1630 ip->ip_ttl = ip_defttl;
1631 ip->ip_len = m->m_pkthdr.len;
1632 m->m_flags |= M_SKIP_FIREWALL;
1637 * sends a reject message, consuming the mbuf passed as an argument.
1640 send_reject(struct ip_fw_args *args, int code, u_short offset, int ip_len)
1643 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1644 /* We need the IP header in host order for icmp_error(). */
1645 if (args->eh != NULL) {
1646 struct ip *ip = mtod(args->m, struct ip *);
1647 ip->ip_len = ntohs(ip->ip_len);
1648 ip->ip_off = ntohs(ip->ip_off);
1650 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1651 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1652 struct tcphdr *const tcp =
1653 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1654 if ( (tcp->th_flags & TH_RST) == 0) {
1656 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1658 tcp->th_flags | TH_RST);
1660 ip_output(m, NULL, NULL, 0, NULL, NULL);
1670 * Given an ip_fw *, lookup_next_rule will return a pointer
1671 * to the next rule, which can be either the jump
1672 * target (for skipto instructions) or the next one in the list (in
1673 * all other cases including a missing jump target).
1674 * The result is also written in the "next_rule" field of the rule.
1675 * Backward jumps are not allowed, so start looking from the next
1678 * This never returns NULL -- in case we do not have an exact match,
1679 * the next rule is returned. When the ruleset is changed,
1680 * pointers are flushed so we are always correct.
1683 static struct ip_fw *
1684 lookup_next_rule(struct ip_fw *me)
1686 struct ip_fw *rule = NULL;
1689 /* look for action, in case it is a skipto */
1690 cmd = ACTION_PTR(me);
1691 if (cmd->opcode == O_LOG)
1693 if (cmd->opcode == O_ALTQ)
1695 if ( cmd->opcode == O_SKIPTO )
1696 for (rule = me->next; rule ; rule = rule->next)
1697 if (rule->rulenum >= cmd->arg1)
1699 if (rule == NULL) /* failure or not a skipto */
1701 me->next_rule = rule;
1710 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1711 rn_inithead((void **)&ipfw_tables[i].rnh, 32);
1712 ipfw_tables[i].modified = 1;
1717 add_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen, u_int32_t value)
1719 struct radix_node_head *rnh;
1720 struct table_entry *ent;
1722 if (tbl >= IPFW_TABLES_MAX)
1724 rnh = ipfw_tables[tbl].rnh;
1725 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1729 ent->addr.sin_len = ent->mask.sin_len = 8;
1730 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1731 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1732 RADIX_NODE_HEAD_LOCK(rnh);
1733 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1735 RADIX_NODE_HEAD_UNLOCK(rnh);
1736 free(ent, M_IPFW_TBL);
1739 ipfw_tables[tbl].modified = 1;
1740 RADIX_NODE_HEAD_UNLOCK(rnh);
1745 del_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen)
1747 struct radix_node_head *rnh;
1748 struct table_entry *ent;
1749 struct sockaddr_in sa, mask;
1751 if (tbl >= IPFW_TABLES_MAX)
1753 rnh = ipfw_tables[tbl].rnh;
1754 sa.sin_len = mask.sin_len = 8;
1755 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1756 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1757 RADIX_NODE_HEAD_LOCK(rnh);
1758 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1760 RADIX_NODE_HEAD_UNLOCK(rnh);
1763 ipfw_tables[tbl].modified = 1;
1764 RADIX_NODE_HEAD_UNLOCK(rnh);
1765 free(ent, M_IPFW_TBL);
1770 flush_table_entry(struct radix_node *rn, void *arg)
1772 struct radix_node_head * const rnh = arg;
1773 struct table_entry *ent;
1775 ent = (struct table_entry *)
1776 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1778 free(ent, M_IPFW_TBL);
1783 flush_table(u_int16_t tbl)
1785 struct radix_node_head *rnh;
1787 if (tbl >= IPFW_TABLES_MAX)
1789 rnh = ipfw_tables[tbl].rnh;
1790 RADIX_NODE_HEAD_LOCK(rnh);
1791 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1792 ipfw_tables[tbl].modified = 1;
1793 RADIX_NODE_HEAD_UNLOCK(rnh);
1802 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1807 lookup_table(u_int16_t tbl, in_addr_t addr, u_int32_t *val)
1809 struct radix_node_head *rnh;
1810 struct ip_fw_table *table;
1811 struct table_entry *ent;
1812 struct sockaddr_in sa;
1815 if (tbl >= IPFW_TABLES_MAX)
1817 table = &ipfw_tables[tbl];
1819 RADIX_NODE_HEAD_LOCK(rnh);
1820 if (addr == table->last_addr && !table->modified) {
1821 last_match = table->last_match;
1823 *val = table->last_value;
1824 RADIX_NODE_HEAD_UNLOCK(rnh);
1825 return (last_match);
1827 table->modified = 0;
1829 sa.sin_addr.s_addr = addr;
1830 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1831 table->last_addr = addr;
1833 table->last_value = *val = ent->value;
1834 table->last_match = 1;
1835 RADIX_NODE_HEAD_UNLOCK(rnh);
1838 table->last_match = 0;
1839 RADIX_NODE_HEAD_UNLOCK(rnh);
1844 count_table_entry(struct radix_node *rn, void *arg)
1846 u_int32_t * const cnt = arg;
1853 count_table(u_int32_t tbl, u_int32_t *cnt)
1855 struct radix_node_head *rnh;
1857 if (tbl >= IPFW_TABLES_MAX)
1859 rnh = ipfw_tables[tbl].rnh;
1861 RADIX_NODE_HEAD_LOCK(rnh);
1862 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1863 RADIX_NODE_HEAD_UNLOCK(rnh);
1868 dump_table_entry(struct radix_node *rn, void *arg)
1870 struct table_entry * const n = (struct table_entry *)rn;
1871 ipfw_table * const tbl = arg;
1872 ipfw_table_entry *ent;
1874 if (tbl->cnt == tbl->size)
1876 ent = &tbl->ent[tbl->cnt];
1877 ent->tbl = tbl->tbl;
1878 if (in_nullhost(n->mask.sin_addr))
1881 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1882 ent->addr = n->addr.sin_addr.s_addr;
1883 ent->value = n->value;
1889 dump_table(ipfw_table *tbl)
1891 struct radix_node_head *rnh;
1893 if (tbl->tbl >= IPFW_TABLES_MAX)
1895 rnh = ipfw_tables[tbl->tbl].rnh;
1897 RADIX_NODE_HEAD_LOCK(rnh);
1898 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1899 RADIX_NODE_HEAD_UNLOCK(rnh);
1904 fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1908 if (inp->inp_socket != NULL) {
1909 cr = inp->inp_socket->so_cred;
1910 ugp->fw_prid = jailed(cr) ?
1911 cr->cr_prison->pr_id : -1;
1912 ugp->fw_uid = cr->cr_uid;
1913 ugp->fw_ngroups = cr->cr_ngroups;
1914 bcopy(cr->cr_groups, ugp->fw_groups,
1915 sizeof(ugp->fw_groups));
1920 check_uidgid(ipfw_insn_u32 *insn,
1921 int proto, struct ifnet *oif,
1922 struct in_addr dst_ip, u_int16_t dst_port,
1923 struct in_addr src_ip, u_int16_t src_port,
1924 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1926 struct inpcbinfo *pi;
1933 * Check to see if the UDP or TCP stack supplied us with
1934 * the PCB. If so, rather then holding a lock and looking
1935 * up the PCB, we can use the one that was supplied.
1937 if (inp && *lookup == 0) {
1938 INP_LOCK_ASSERT(inp);
1939 if (inp->inp_socket != NULL) {
1940 fill_ugid_cache(inp, ugp);
1945 * If we have already been here and the packet has no
1946 * PCB entry associated with it, then we can safely
1947 * assume that this is a no match.
1951 if (proto == IPPROTO_TCP) {
1954 } else if (proto == IPPROTO_UDP) {
1963 in_pcblookup_hash(pi,
1964 dst_ip, htons(dst_port),
1965 src_ip, htons(src_port),
1967 in_pcblookup_hash(pi,
1968 src_ip, htons(src_port),
1969 dst_ip, htons(dst_port),
1973 if (pcb->inp_socket != NULL) {
1974 fill_ugid_cache(pcb, ugp);
1979 INP_INFO_RUNLOCK(pi);
1982 * If the lookup did not yield any results, there
1983 * is no sense in coming back and trying again. So
1984 * we can set lookup to -1 and ensure that we wont
1985 * bother the pcb system again.
1991 if (insn->o.opcode == O_UID)
1992 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1993 else if (insn->o.opcode == O_GID) {
1994 for (gp = ugp->fw_groups;
1995 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1996 if (*gp == (gid_t)insn->d[0]) {
2000 } else if (insn->o.opcode == O_JAIL)
2001 match = (ugp->fw_prid == (int)insn->d[0]);
2006 * The main check routine for the firewall.
2008 * All arguments are in args so we can modify them and return them
2009 * back to the caller.
2013 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
2014 * Starts with the IP header.
2015 * args->eh (in) Mac header if present, or NULL for layer3 packet.
2016 * args->oif Outgoing interface, or NULL if packet is incoming.
2017 * The incoming interface is in the mbuf. (in)
2018 * args->divert_rule (in/out)
2019 * Skip up to the first rule past this rule number;
2020 * upon return, non-zero port number for divert or tee.
2022 * args->rule Pointer to the last matching rule (in/out)
2023 * args->next_hop Socket we are forwarding to (out).
2024 * args->f_id Addresses grabbed from the packet (out)
2025 * args->cookie a cookie depending on rule action
2029 * IP_FW_PASS the packet must be accepted
2030 * IP_FW_DENY the packet must be dropped
2031 * IP_FW_DIVERT divert packet, port in m_tag
2032 * IP_FW_TEE tee packet, port in m_tag
2033 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2034 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2039 ipfw_chk(struct ip_fw_args *args)
2042 * Local variables hold state during the processing of a packet.
2044 * IMPORTANT NOTE: to speed up the processing of rules, there
2045 * are some assumption on the values of the variables, which
2046 * are documented here. Should you change them, please check
2047 * the implementation of the various instructions to make sure
2048 * that they still work.
2050 * args->eh The MAC header. It is non-null for a layer2
2051 * packet, it is NULL for a layer-3 packet.
2053 * m | args->m Pointer to the mbuf, as received from the caller.
2054 * It may change if ipfw_chk() does an m_pullup, or if it
2055 * consumes the packet because it calls send_reject().
2056 * XXX This has to change, so that ipfw_chk() never modifies
2057 * or consumes the buffer.
2058 * ip is simply an alias of the value of m, and it is kept
2059 * in sync with it (the packet is supposed to start with
2062 struct mbuf *m = args->m;
2063 struct ip *ip = mtod(m, struct ip *);
2066 * For rules which contain uid/gid or jail constraints, cache
2067 * a copy of the users credentials after the pcb lookup has been
2068 * executed. This will speed up the processing of rules with
2069 * these types of constraints, as well as decrease contention
2070 * on pcb related locks.
2072 struct ip_fw_ugid fw_ugid_cache;
2073 int ugid_lookup = 0;
2076 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2077 * associated with a packet input on a divert socket. This
2078 * will allow to distinguish traffic and its direction when
2079 * it originates from a divert socket.
2081 u_int divinput_flags = 0;
2084 * oif | args->oif If NULL, ipfw_chk has been called on the
2085 * inbound path (ether_input, ip_input).
2086 * If non-NULL, ipfw_chk has been called on the outbound path
2087 * (ether_output, ip_output).
2089 struct ifnet *oif = args->oif;
2091 struct ip_fw *f = NULL; /* matching rule */
2095 * hlen The length of the IP header.
2097 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2100 * offset The offset of a fragment. offset != 0 means that
2101 * we have a fragment at this offset of an IPv4 packet.
2102 * offset == 0 means that (if this is an IPv4 packet)
2103 * this is the first or only fragment.
2104 * For IPv6 offset == 0 means there is no Fragment Header.
2105 * If offset != 0 for IPv6 always use correct mask to
2106 * get the correct offset because we add IP6F_MORE_FRAG
2107 * to be able to dectect the first fragment which would
2108 * otherwise have offset = 0.
2113 * Local copies of addresses. They are only valid if we have
2116 * proto The protocol. Set to 0 for non-ip packets,
2117 * or to the protocol read from the packet otherwise.
2118 * proto != 0 means that we have an IPv4 packet.
2120 * src_port, dst_port port numbers, in HOST format. Only
2121 * valid for TCP and UDP packets.
2123 * src_ip, dst_ip ip addresses, in NETWORK format.
2124 * Only valid for IPv4 packets.
2127 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2128 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2133 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2134 * MATCH_NONE when checked and not matched (q = NULL),
2135 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2137 int dyn_dir = MATCH_UNKNOWN;
2138 ipfw_dyn_rule *q = NULL;
2139 struct ip_fw_chain *chain = &layer3_chain;
2143 * We store in ulp a pointer to the upper layer protocol header.
2144 * In the ipv4 case this is easy to determine from the header,
2145 * but for ipv6 we might have some additional headers in the middle.
2146 * ulp is NULL if not found.
2148 void *ulp = NULL; /* upper layer protocol pointer. */
2149 /* XXX ipv6 variables */
2151 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2152 /* end of ipv6 variables */
2155 if (m->m_flags & M_SKIP_FIREWALL)
2156 return (IP_FW_PASS); /* accept */
2158 pktlen = m->m_pkthdr.len;
2159 proto = args->f_id.proto = 0; /* mark f_id invalid */
2160 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2163 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2164 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2165 * pointer might become stale after other pullups (but we never use it
2168 #define PULLUP_TO(len, p, T) \
2170 int x = (len) + sizeof(T); \
2171 if ((m)->m_len < x) { \
2172 args->m = m = m_pullup(m, x); \
2174 goto pullup_failed; \
2176 p = (mtod(m, char *) + (len)); \
2179 /* Identify IP packets and fill up variables. */
2180 if (pktlen >= sizeof(struct ip6_hdr) &&
2181 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2182 mtod(m, struct ip *)->ip_v == 6) {
2184 args->f_id.addr_type = 6;
2185 hlen = sizeof(struct ip6_hdr);
2186 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2188 /* Search extension headers to find upper layer protocols */
2189 while (ulp == NULL) {
2191 case IPPROTO_ICMPV6:
2192 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2193 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2197 PULLUP_TO(hlen, ulp, struct tcphdr);
2198 dst_port = TCP(ulp)->th_dport;
2199 src_port = TCP(ulp)->th_sport;
2200 args->f_id.flags = TCP(ulp)->th_flags;
2204 PULLUP_TO(hlen, ulp, struct udphdr);
2205 dst_port = UDP(ulp)->uh_dport;
2206 src_port = UDP(ulp)->uh_sport;
2209 case IPPROTO_HOPOPTS: /* RFC 2460 */
2210 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2211 ext_hd |= EXT_HOPOPTS;
2212 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2213 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2217 case IPPROTO_ROUTING: /* RFC 2460 */
2218 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2219 if (((struct ip6_rthdr *)ulp)->ip6r_type != 0) {
2220 printf("IPFW2: IPV6 - Unknown Routing "
2221 "Header type(%d)\n",
2222 ((struct ip6_rthdr *)ulp)->ip6r_type);
2223 if (fw_deny_unknown_exthdrs)
2224 return (IP_FW_DENY);
2227 ext_hd |= EXT_ROUTING;
2228 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2229 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2233 case IPPROTO_FRAGMENT: /* RFC 2460 */
2234 PULLUP_TO(hlen, ulp, struct ip6_frag);
2235 ext_hd |= EXT_FRAGMENT;
2236 hlen += sizeof (struct ip6_frag);
2237 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2238 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2240 /* Add IP6F_MORE_FRAG for offset of first
2241 * fragment to be != 0. */
2242 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2245 printf("IPFW2: IPV6 - Invalid Fragment "
2247 if (fw_deny_unknown_exthdrs)
2248 return (IP_FW_DENY);
2251 args->f_id.frag_id6 =
2252 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2256 case IPPROTO_DSTOPTS: /* RFC 2460 */
2257 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2258 ext_hd |= EXT_DSTOPTS;
2259 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2260 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2264 case IPPROTO_AH: /* RFC 2402 */
2265 PULLUP_TO(hlen, ulp, struct ip6_ext);
2267 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2268 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2272 case IPPROTO_ESP: /* RFC 2406 */
2273 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2274 /* Anything past Seq# is variable length and
2275 * data past this ext. header is encrypted. */
2279 case IPPROTO_NONE: /* RFC 2460 */
2280 PULLUP_TO(hlen, ulp, struct ip6_ext);
2281 /* Packet ends here. if ip6e_len!=0 octets
2282 * must be ignored. */
2285 case IPPROTO_OSPFIGP:
2286 /* XXX OSPF header check? */
2287 PULLUP_TO(hlen, ulp, struct ip6_ext);
2291 printf("IPFW2: IPV6 - Unknown Extension "
2292 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2293 if (fw_deny_unknown_exthdrs)
2294 return (IP_FW_DENY);
2298 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2299 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2300 args->f_id.src_ip = 0;
2301 args->f_id.dst_ip = 0;
2302 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2303 } else if (pktlen >= sizeof(struct ip) &&
2304 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2305 mtod(m, struct ip *)->ip_v == 4) {
2307 ip = mtod(m, struct ip *);
2308 hlen = ip->ip_hl << 2;
2309 args->f_id.addr_type = 4;
2312 * Collect parameters into local variables for faster matching.
2315 src_ip = ip->ip_src;
2316 dst_ip = ip->ip_dst;
2317 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2318 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2319 ip_len = ntohs(ip->ip_len);
2321 offset = ip->ip_off & IP_OFFMASK;
2322 ip_len = ip->ip_len;
2324 pktlen = ip_len < pktlen ? ip_len : pktlen;
2329 PULLUP_TO(hlen, ulp, struct tcphdr);
2330 dst_port = TCP(ulp)->th_dport;
2331 src_port = TCP(ulp)->th_sport;
2332 args->f_id.flags = TCP(ulp)->th_flags;
2336 PULLUP_TO(hlen, ulp, struct udphdr);
2337 dst_port = UDP(ulp)->uh_dport;
2338 src_port = UDP(ulp)->uh_sport;
2342 PULLUP_TO(hlen, ulp, struct icmphdr);
2343 args->f_id.flags = ICMP(ulp)->icmp_type;
2351 args->f_id.src_ip = ntohl(src_ip.s_addr);
2352 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2355 if (proto) { /* we may have port numbers, store them */
2356 args->f_id.proto = proto;
2357 args->f_id.src_port = src_port = ntohs(src_port);
2358 args->f_id.dst_port = dst_port = ntohs(dst_port);
2362 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2365 * Packet has already been tagged. Look for the next rule
2366 * to restart processing.
2368 * If fw_one_pass != 0 then just accept it.
2369 * XXX should not happen here, but optimized out in
2373 IPFW_RUNLOCK(chain);
2374 return (IP_FW_PASS);
2377 f = args->rule->next_rule;
2379 f = lookup_next_rule(args->rule);
2382 * Find the starting rule. It can be either the first
2383 * one, or the one after divert_rule if asked so.
2385 int skipto = mtag ? divert_cookie(mtag) : 0;
2388 if (args->eh == NULL && skipto != 0) {
2389 if (skipto >= IPFW_DEFAULT_RULE) {
2390 IPFW_RUNLOCK(chain);
2391 return (IP_FW_DENY); /* invalid */
2393 while (f && f->rulenum <= skipto)
2395 if (f == NULL) { /* drop packet */
2396 IPFW_RUNLOCK(chain);
2397 return (IP_FW_DENY);
2401 /* reset divert rule to avoid confusion later */
2403 divinput_flags = divert_info(mtag) &
2404 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2405 m_tag_delete(m, mtag);
2409 * Now scan the rules, and parse microinstructions for each rule.
2411 for (; f; f = f->next) {
2414 int skip_or; /* skip rest of OR block */
2417 if (set_disable & (1 << f->set) )
2421 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
2422 l -= cmdlen, cmd += cmdlen) {
2426 * check_body is a jump target used when we find a
2427 * CHECK_STATE, and need to jump to the body of
2432 cmdlen = F_LEN(cmd);
2434 * An OR block (insn_1 || .. || insn_n) has the
2435 * F_OR bit set in all but the last instruction.
2436 * The first match will set "skip_or", and cause
2437 * the following instructions to be skipped until
2438 * past the one with the F_OR bit clear.
2440 if (skip_or) { /* skip this instruction */
2441 if ((cmd->len & F_OR) == 0)
2442 skip_or = 0; /* next one is good */
2445 match = 0; /* set to 1 if we succeed */
2447 switch (cmd->opcode) {
2449 * The first set of opcodes compares the packet's
2450 * fields with some pattern, setting 'match' if a
2451 * match is found. At the end of the loop there is
2452 * logic to deal with F_NOT and F_OR flags associated
2460 printf("ipfw: opcode %d unimplemented\n",
2468 * We only check offset == 0 && proto != 0,
2469 * as this ensures that we have a
2470 * packet with the ports info.
2474 if (is_ipv6) /* XXX to be fixed later */
2476 if (proto == IPPROTO_TCP ||
2477 proto == IPPROTO_UDP)
2478 match = check_uidgid(
2479 (ipfw_insn_u32 *)cmd,
2482 src_ip, src_port, &fw_ugid_cache,
2483 &ugid_lookup, args->inp);
2487 match = iface_match(m->m_pkthdr.rcvif,
2488 (ipfw_insn_if *)cmd);
2492 match = iface_match(oif, (ipfw_insn_if *)cmd);
2496 match = iface_match(oif ? oif :
2497 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
2501 if (args->eh != NULL) { /* have MAC header */
2502 u_int32_t *want = (u_int32_t *)
2503 ((ipfw_insn_mac *)cmd)->addr;
2504 u_int32_t *mask = (u_int32_t *)
2505 ((ipfw_insn_mac *)cmd)->mask;
2506 u_int32_t *hdr = (u_int32_t *)args->eh;
2509 ( want[0] == (hdr[0] & mask[0]) &&
2510 want[1] == (hdr[1] & mask[1]) &&
2511 want[2] == (hdr[2] & mask[2]) );
2516 if (args->eh != NULL) {
2518 ntohs(args->eh->ether_type);
2520 ((ipfw_insn_u16 *)cmd)->ports;
2523 for (i = cmdlen - 1; !match && i>0;
2525 match = (t>=p[0] && t<=p[1]);
2530 match = (offset != 0);
2533 case O_IN: /* "out" is "not in" */
2534 match = (oif == NULL);
2538 match = (args->eh != NULL);
2542 match = (cmd->arg1 & 1 && divinput_flags &
2543 IP_FW_DIVERT_LOOPBACK_FLAG) ||
2544 (cmd->arg1 & 2 && divinput_flags &
2545 IP_FW_DIVERT_OUTPUT_FLAG);
2550 * We do not allow an arg of 0 so the
2551 * check of "proto" only suffices.
2553 match = (proto == cmd->arg1);
2558 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2562 case O_IP_SRC_LOOKUP:
2563 case O_IP_DST_LOOKUP:
2566 (cmd->opcode == O_IP_DST_LOOKUP) ?
2567 dst_ip.s_addr : src_ip.s_addr;
2570 match = lookup_table(cmd->arg1, a, &v);
2573 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2575 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2583 (cmd->opcode == O_IP_DST_MASK) ?
2584 dst_ip.s_addr : src_ip.s_addr;
2585 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2588 for (; !match && i>0; i-= 2, p+= 2)
2589 match = (p[0] == (a & p[1]));
2597 INADDR_TO_IFP(src_ip, tif);
2598 match = (tif != NULL);
2605 u_int32_t *d = (u_int32_t *)(cmd+1);
2607 cmd->opcode == O_IP_DST_SET ?
2613 addr -= d[0]; /* subtract base */
2614 match = (addr < cmd->arg1) &&
2615 ( d[ 1 + (addr>>5)] &
2616 (1<<(addr & 0x1f)) );
2622 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2630 INADDR_TO_IFP(dst_ip, tif);
2631 match = (tif != NULL);
2638 * offset == 0 && proto != 0 is enough
2639 * to guarantee that we have a
2640 * packet with port info.
2642 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2645 (cmd->opcode == O_IP_SRCPORT) ?
2646 src_port : dst_port ;
2648 ((ipfw_insn_u16 *)cmd)->ports;
2651 for (i = cmdlen - 1; !match && i>0;
2653 match = (x>=p[0] && x<=p[1]);
2658 match = (offset == 0 && proto==IPPROTO_ICMP &&
2659 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
2664 match = is_ipv6 && offset == 0 &&
2665 proto==IPPROTO_ICMPV6 &&
2667 ICMP6(ulp)->icmp6_type,
2668 (ipfw_insn_u32 *)cmd);
2674 ipopts_match(mtod(m, struct ip *), cmd) );
2679 cmd->arg1 == mtod(m, struct ip *)->ip_v);
2685 if (is_ipv4) { /* only for IP packets */
2690 if (cmd->opcode == O_IPLEN)
2692 else if (cmd->opcode == O_IPTTL)
2693 x = mtod(m, struct ip *)->ip_ttl;
2694 else /* must be IPID */
2695 x = ntohs(mtod(m, struct ip *)->ip_id);
2697 match = (cmd->arg1 == x);
2700 /* otherwise we have ranges */
2701 p = ((ipfw_insn_u16 *)cmd)->ports;
2703 for (; !match && i>0; i--, p += 2)
2704 match = (x >= p[0] && x <= p[1]);
2708 case O_IPPRECEDENCE:
2710 (cmd->arg1 == (mtod(m, struct ip *)->ip_tos & 0xe0)) );
2715 flags_match(cmd, mtod(m, struct ip *)->ip_tos));
2719 if (proto == IPPROTO_TCP && offset == 0) {
2727 ((ip->ip_hl + tcp->th_off) << 2);
2729 match = (cmd->arg1 == x);
2732 /* otherwise we have ranges */
2733 p = ((ipfw_insn_u16 *)cmd)->ports;
2735 for (; !match && i>0; i--, p += 2)
2736 match = (x >= p[0] && x <= p[1]);
2741 match = (proto == IPPROTO_TCP && offset == 0 &&
2742 flags_match(cmd, TCP(ulp)->th_flags));
2746 match = (proto == IPPROTO_TCP && offset == 0 &&
2747 tcpopts_match(TCP(ulp), cmd));
2751 match = (proto == IPPROTO_TCP && offset == 0 &&
2752 ((ipfw_insn_u32 *)cmd)->d[0] ==
2757 match = (proto == IPPROTO_TCP && offset == 0 &&
2758 ((ipfw_insn_u32 *)cmd)->d[0] ==
2763 match = (proto == IPPROTO_TCP && offset == 0 &&
2764 cmd->arg1 == TCP(ulp)->th_win);
2768 /* reject packets which have SYN only */
2769 /* XXX should i also check for TH_ACK ? */
2770 match = (proto == IPPROTO_TCP && offset == 0 &&
2771 (TCP(ulp)->th_flags &
2772 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2776 struct altq_tag *at;
2777 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
2780 mtag = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
2783 mtag = m_tag_get(PACKET_TAG_PF_QID,
2784 sizeof(struct altq_tag),
2788 * Let the packet fall back to the
2793 at = (struct altq_tag *)(mtag+1);
2794 at->qid = altq->qid;
2800 m_tag_prepend(m, mtag);
2806 ipfw_log(f, hlen, args, m, oif, offset);
2811 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2815 /* Outgoing packets automatically pass/match */
2816 match = ((oif != NULL) ||
2817 (m->m_pkthdr.rcvif == NULL) ||
2821 verify_path6(&(args->f_id.src_ip6),
2822 m->m_pkthdr.rcvif) :
2824 verify_path(src_ip, m->m_pkthdr.rcvif)));
2828 /* Outgoing packets automatically pass/match */
2829 match = (hlen > 0 && ((oif != NULL) ||
2832 verify_path6(&(args->f_id.src_ip6),
2835 verify_path(src_ip, NULL)));
2839 /* Outgoing packets automatically pass/match */
2840 if (oif == NULL && hlen > 0 &&
2841 ( (is_ipv4 && in_localaddr(src_ip))
2844 in6_localaddr(&(args->f_id.src_ip6)))
2849 is_ipv6 ? verify_path6(
2850 &(args->f_id.src_ip6),
2851 m->m_pkthdr.rcvif) :
2861 match = (m_tag_find(m,
2862 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2865 match = (ipsec_getnhist(m) != 0);
2867 /* otherwise no match */
2873 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
2874 &((ipfw_insn_ip6 *)cmd)->addr6);
2879 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
2880 &((ipfw_insn_ip6 *)cmd)->addr6);
2882 case O_IP6_SRC_MASK:
2884 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2885 struct in6_addr p = args->f_id.src_ip6;
2887 APPLY_MASK(&p, &te->mask6);
2888 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2892 case O_IP6_DST_MASK:
2894 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2895 struct in6_addr p = args->f_id.dst_ip6;
2897 APPLY_MASK(&p, &te->mask6);
2898 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2903 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
2907 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
2912 flow6id_match(args->f_id.flow_id6,
2913 (ipfw_insn_u32 *) cmd);
2918 (ext_hd & ((ipfw_insn *) cmd)->arg1);
2931 * The second set of opcodes represents 'actions',
2932 * i.e. the terminal part of a rule once the packet
2933 * matches all previous patterns.
2934 * Typically there is only one action for each rule,
2935 * and the opcode is stored at the end of the rule
2936 * (but there are exceptions -- see below).
2938 * In general, here we set retval and terminate the
2939 * outer loop (would be a 'break 3' in some language,
2940 * but we need to do a 'goto done').
2943 * O_COUNT and O_SKIPTO actions:
2944 * instead of terminating, we jump to the next rule
2945 * ('goto next_rule', equivalent to a 'break 2'),
2946 * or to the SKIPTO target ('goto again' after
2947 * having set f, cmd and l), respectively.
2949 * O_LOG and O_ALTQ action parameters:
2950 * perform some action and set match = 1;
2952 * O_LIMIT and O_KEEP_STATE: these opcodes are
2953 * not real 'actions', and are stored right
2954 * before the 'action' part of the rule.
2955 * These opcodes try to install an entry in the
2956 * state tables; if successful, we continue with
2957 * the next opcode (match=1; break;), otherwise
2958 * the packet * must be dropped
2959 * ('goto done' after setting retval);
2961 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2962 * cause a lookup of the state table, and a jump
2963 * to the 'action' part of the parent rule
2964 * ('goto check_body') if an entry is found, or
2965 * (CHECK_STATE only) a jump to the next rule if
2966 * the entry is not found ('goto next_rule').
2967 * The result of the lookup is cached to make
2968 * further instances of these opcodes are
2973 if (install_state(f,
2974 (ipfw_insn_limit *)cmd, args)) {
2975 retval = IP_FW_DENY;
2976 goto done; /* error/limit violation */
2984 * dynamic rules are checked at the first
2985 * keep-state or check-state occurrence,
2986 * with the result being stored in dyn_dir.
2987 * The compiler introduces a PROBE_STATE
2988 * instruction for us when we have a
2989 * KEEP_STATE (because PROBE_STATE needs
2992 if (dyn_dir == MATCH_UNKNOWN &&
2993 (q = lookup_dyn_rule(&args->f_id,
2994 &dyn_dir, proto == IPPROTO_TCP ?
2998 * Found dynamic entry, update stats
2999 * and jump to the 'action' part of
3005 cmd = ACTION_PTR(f);
3006 l = f->cmd_len - f->act_ofs;
3011 * Dynamic entry not found. If CHECK_STATE,
3012 * skip to next rule, if PROBE_STATE just
3013 * ignore and continue with next opcode.
3015 if (cmd->opcode == O_CHECK_STATE)
3021 retval = 0; /* accept */
3026 args->rule = f; /* report matching rule */
3027 args->cookie = cmd->arg1;
3028 retval = IP_FW_DUMMYNET;
3033 struct divert_tag *dt;
3035 if (args->eh) /* not on layer 2 */
3037 mtag = m_tag_get(PACKET_TAG_DIVERT,
3038 sizeof(struct divert_tag),
3043 IPFW_RUNLOCK(chain);
3044 return (IP_FW_DENY);
3046 dt = (struct divert_tag *)(mtag+1);
3047 dt->cookie = f->rulenum;
3048 dt->info = cmd->arg1;
3049 m_tag_prepend(m, mtag);
3050 retval = (cmd->opcode == O_DIVERT) ?
3051 IP_FW_DIVERT : IP_FW_TEE;
3057 f->pcnt++; /* update stats */
3059 f->timestamp = time_uptime;
3060 if (cmd->opcode == O_COUNT)
3063 if (f->next_rule == NULL)
3064 lookup_next_rule(f);
3070 * Drop the packet and send a reject notice
3071 * if the packet is not ICMP (or is an ICMP
3072 * query), and it is not multicast/broadcast.
3074 if (hlen > 0 && is_ipv4 &&
3075 (proto != IPPROTO_ICMP ||
3076 is_icmp_query(ICMP(ulp))) &&
3077 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3078 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
3079 send_reject(args, cmd->arg1,
3086 if (hlen > 0 && is_ipv6 &&
3087 (proto != IPPROTO_ICMPV6 ||
3088 (is_icmp6_query(args->f_id.flags) == 1)) &&
3089 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3090 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
3091 send_reject6(args, cmd->arg1,
3098 retval = IP_FW_DENY;
3102 if (args->eh) /* not valid on layer2 pkts */
3104 if (!q || dyn_dir == MATCH_FORWARD)
3106 &((ipfw_insn_sa *)cmd)->sa;
3107 retval = IP_FW_PASS;
3112 args->rule = f; /* report matching rule */
3113 args->cookie = cmd->arg1;
3114 retval = (cmd->opcode == O_NETGRAPH) ?
3115 IP_FW_NETGRAPH : IP_FW_NGTEE;
3119 panic("-- unknown opcode %d\n", cmd->opcode);
3120 } /* end of switch() on opcodes */
3122 if (cmd->len & F_NOT)
3126 if (cmd->len & F_OR)
3129 if (!(cmd->len & F_OR)) /* not an OR block, */
3130 break; /* try next rule */
3133 } /* end of inner for, scan opcodes */
3135 next_rule:; /* try next rule */
3137 } /* end of outer for, scan rules */
3138 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3139 IPFW_RUNLOCK(chain);
3140 return (IP_FW_DENY);
3143 /* Update statistics */
3146 f->timestamp = time_uptime;
3147 IPFW_RUNLOCK(chain);
3152 printf("ipfw: pullup failed\n");
3153 return (IP_FW_DENY);
3157 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3158 * These will be reconstructed on the fly as packets are matched.
3161 flush_rule_ptrs(struct ip_fw_chain *chain)
3165 IPFW_WLOCK_ASSERT(chain);
3167 for (rule = chain->rules; rule; rule = rule->next)
3168 rule->next_rule = NULL;
3172 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
3173 * pipe/queue, or to all of them (match == NULL).
3176 flush_pipe_ptrs(struct dn_flow_set *match)
3180 IPFW_WLOCK(&layer3_chain);
3181 for (rule = layer3_chain.rules; rule; rule = rule->next) {
3182 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
3184 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
3187 * XXX Use bcmp/bzero to handle pipe_ptr to overcome
3188 * possible alignment problems on 64-bit architectures.
3189 * This code is seldom used so we do not worry too
3190 * much about efficiency.
3192 if (match == NULL ||
3193 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) )
3194 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr));
3196 IPFW_WUNLOCK(&layer3_chain);
3200 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3201 * possibly create a rule number and add the rule to the list.
3202 * Update the rule_number in the input struct so the caller knows it as well.
3205 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3207 struct ip_fw *rule, *f, *prev;
3208 int l = RULESIZE(input_rule);
3210 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3213 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3217 bcopy(input_rule, rule, l);
3220 rule->next_rule = NULL;
3224 rule->timestamp = 0;
3228 if (chain->rules == NULL) { /* default rule */
3229 chain->rules = rule;
3234 * If rulenum is 0, find highest numbered rule before the
3235 * default rule, and add autoinc_step
3237 if (autoinc_step < 1)
3239 else if (autoinc_step > 1000)
3240 autoinc_step = 1000;
3241 if (rule->rulenum == 0) {
3243 * locate the highest numbered rule before default
3245 for (f = chain->rules; f; f = f->next) {
3246 if (f->rulenum == IPFW_DEFAULT_RULE)
3248 rule->rulenum = f->rulenum;
3250 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3251 rule->rulenum += autoinc_step;
3252 input_rule->rulenum = rule->rulenum;
3256 * Now insert the new rule in the right place in the sorted list.
3258 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3259 if (f->rulenum > rule->rulenum) { /* found the location */
3263 } else { /* head insert */
3264 rule->next = chain->rules;
3265 chain->rules = rule;
3270 flush_rule_ptrs(chain);
3274 IPFW_WUNLOCK(chain);
3275 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3276 rule->rulenum, static_count);)
3281 * Remove a static rule (including derived * dynamic rules)
3282 * and place it on the ``reap list'' for later reclamation.
3283 * The caller is in charge of clearing rule pointers to avoid
3284 * dangling pointers.
3285 * @return a pointer to the next entry.
3286 * Arguments are not checked, so they better be correct.
3288 static struct ip_fw *
3289 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3292 int l = RULESIZE(rule);
3294 IPFW_WLOCK_ASSERT(chain);
3298 remove_dyn_rule(rule, NULL /* force removal */);
3307 rule->next = chain->reap;
3314 * Reclaim storage associated with a list of rules. This is
3315 * typically the list created using remove_rule.
3318 reap_rules(struct ip_fw *head)
3322 while ((rule = head) != NULL) {
3324 if (DUMMYNET_LOADED)
3325 ip_dn_ruledel_ptr(rule);
3331 * Remove all rules from a chain (except rules in set RESVD_SET
3332 * unless kill_default = 1). The caller is responsible for
3333 * reclaiming storage for the rules left in chain->reap.
3336 free_chain(struct ip_fw_chain *chain, int kill_default)
3338 struct ip_fw *prev, *rule;
3340 IPFW_WLOCK_ASSERT(chain);
3342 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3343 for (prev = NULL, rule = chain->rules; rule ; )
3344 if (kill_default || rule->set != RESVD_SET)
3345 rule = remove_rule(chain, rule, prev);
3353 * Remove all rules with given number, and also do set manipulation.
3354 * Assumes chain != NULL && *chain != NULL.
3356 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3357 * the next 8 bits are the new set, the top 8 bits are the command:
3359 * 0 delete rules with given number
3360 * 1 delete rules with given set number
3361 * 2 move rules with given number to new set
3362 * 3 move rules with given set number to new set
3363 * 4 swap sets with given numbers
3366 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3368 struct ip_fw *prev = NULL, *rule;
3369 u_int16_t rulenum; /* rule or old_set */
3370 u_int8_t cmd, new_set;
3372 rulenum = arg & 0xffff;
3373 cmd = (arg >> 24) & 0xff;
3374 new_set = (arg >> 16) & 0xff;
3378 if (new_set > RESVD_SET)
3380 if (cmd == 0 || cmd == 2) {
3381 if (rulenum >= IPFW_DEFAULT_RULE)
3384 if (rulenum > RESVD_SET) /* old_set */
3389 rule = chain->rules;
3392 case 0: /* delete rules with given number */
3394 * locate first rule to delete
3396 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3398 if (rule->rulenum != rulenum) {
3399 IPFW_WUNLOCK(chain);
3404 * flush pointers outside the loop, then delete all matching
3405 * rules. prev remains the same throughout the cycle.
3407 flush_rule_ptrs(chain);
3408 while (rule->rulenum == rulenum)
3409 rule = remove_rule(chain, rule, prev);
3412 case 1: /* delete all rules with given set number */
3413 flush_rule_ptrs(chain);
3414 rule = chain->rules;
3415 while (rule->rulenum < IPFW_DEFAULT_RULE)
3416 if (rule->set == rulenum)
3417 rule = remove_rule(chain, rule, prev);
3424 case 2: /* move rules with given number to new set */
3425 rule = chain->rules;
3426 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3427 if (rule->rulenum == rulenum)
3428 rule->set = new_set;
3431 case 3: /* move rules with given set number to new set */
3432 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3433 if (rule->set == rulenum)
3434 rule->set = new_set;
3437 case 4: /* swap two sets */
3438 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3439 if (rule->set == rulenum)
3440 rule->set = new_set;
3441 else if (rule->set == new_set)
3442 rule->set = rulenum;
3446 * Look for rules to reclaim. We grab the list before
3447 * releasing the lock then reclaim them w/o the lock to
3448 * avoid a LOR with dummynet.
3452 IPFW_WUNLOCK(chain);
3459 * Clear counters for a specific rule.
3460 * The enclosing "table" is assumed locked.
3463 clear_counters(struct ip_fw *rule, int log_only)
3465 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3467 if (log_only == 0) {
3468 rule->bcnt = rule->pcnt = 0;
3469 rule->timestamp = 0;
3471 if (l->o.opcode == O_LOG)
3472 l->log_left = l->max_log;
3476 * Reset some or all counters on firewall rules.
3477 * @arg frwl is null to clear all entries, or contains a specific
3479 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3482 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3490 for (rule = chain->rules; rule; rule = rule->next)
3491 clear_counters(rule, log_only);
3492 msg = log_only ? "ipfw: All logging counts reset.\n" :
3493 "ipfw: Accounting cleared.\n";
3497 * We can have multiple rules with the same number, so we
3498 * need to clear them all.
3500 for (rule = chain->rules; rule; rule = rule->next)
3501 if (rule->rulenum == rulenum) {
3502 while (rule && rule->rulenum == rulenum) {
3503 clear_counters(rule, log_only);
3509 if (!cleared) { /* we did not find any matching rules */
3510 IPFW_WUNLOCK(chain);
3513 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3514 "ipfw: Entry %d cleared.\n";
3516 IPFW_WUNLOCK(chain);
3519 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3524 * Check validity of the structure before insert.
3525 * Fortunately rules are simple, so this mostly need to check rule sizes.
3528 check_ipfw_struct(struct ip_fw *rule, int size)
3534 if (size < sizeof(*rule)) {
3535 printf("ipfw: rule too short\n");
3538 /* first, check for valid size */
3541 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3544 if (rule->act_ofs >= rule->cmd_len) {
3545 printf("ipfw: bogus action offset (%u > %u)\n",
3546 rule->act_ofs, rule->cmd_len - 1);
3550 * Now go for the individual checks. Very simple ones, basically only
3551 * instruction sizes.
3553 for (l = rule->cmd_len, cmd = rule->cmd ;
3554 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3555 cmdlen = F_LEN(cmd);
3557 printf("ipfw: opcode %d size truncated\n",
3561 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3562 switch (cmd->opcode) {
3574 case O_IPPRECEDENCE:
3591 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3604 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3609 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3614 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3617 ((ipfw_insn_log *)cmd)->log_left =
3618 ((ipfw_insn_log *)cmd)->max_log;
3624 /* only odd command lengths */
3625 if ( !(cmdlen & 1) || cmdlen > 31)
3631 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3632 printf("ipfw: invalid set size %d\n",
3636 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3641 case O_IP_SRC_LOOKUP:
3642 case O_IP_DST_LOOKUP:
3643 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3644 printf("ipfw: invalid table number %d\n",
3648 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3649 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3654 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3663 if (cmdlen < 1 || cmdlen > 31)
3669 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3670 if (cmdlen < 2 || cmdlen > 31)
3677 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3682 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3688 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3693 #ifdef IPFIREWALL_FORWARD
3694 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3703 if (ip_divert_ptr == NULL)
3709 if (!NG_IPFW_LOADED)
3713 case O_FORWARD_MAC: /* XXX not implemented yet */
3724 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3728 printf("ipfw: opcode %d, multiple actions"
3735 printf("ipfw: opcode %d, action must be"
3744 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3745 F_INSN_SIZE(ipfw_insn))
3750 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3751 ((ipfw_insn_u32 *)cmd)->o.arg1)
3755 case O_IP6_SRC_MASK:
3756 case O_IP6_DST_MASK:
3757 if ( !(cmdlen & 1) || cmdlen > 127)
3761 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3767 switch (cmd->opcode) {
3777 case O_IP6_SRC_MASK:
3778 case O_IP6_DST_MASK:
3780 printf("ipfw: no IPv6 support in kernel\n");
3781 return EPROTONOSUPPORT;
3784 printf("ipfw: opcode %d, unknown opcode\n",
3790 if (have_action == 0) {
3791 printf("ipfw: missing action\n");
3797 printf("ipfw: opcode %d size %d wrong\n",
3798 cmd->opcode, cmdlen);
3803 * Copy the static and dynamic rules to the supplied buffer
3804 * and return the amount of space actually used.
3807 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3810 char *ep = bp + space;
3814 /* XXX this can take a long time and locking will block packet flow */
3816 for (rule = chain->rules; rule ; rule = rule->next) {
3818 * Verify the entry fits in the buffer in case the
3819 * rules changed between calculating buffer space and
3820 * now. This would be better done using a generation
3821 * number but should suffice for now.
3826 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3827 sizeof(set_disable));
3831 IPFW_RUNLOCK(chain);
3833 ipfw_dyn_rule *p, *last = NULL;
3836 for (i = 0 ; i < curr_dyn_buckets; i++)
3837 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3838 if (bp + sizeof *p <= ep) {
3839 ipfw_dyn_rule *dst =
3840 (ipfw_dyn_rule *)bp;
3841 bcopy(p, dst, sizeof *p);
3842 bcopy(&(p->rule->rulenum), &(dst->rule),
3843 sizeof(p->rule->rulenum));
3845 * store a non-null value in "next".
3846 * The userland code will interpret a
3847 * NULL here as a marker
3848 * for the last dynamic rule.
3850 bcopy(&dst, &dst->next, sizeof(dst));
3853 TIME_LEQ(dst->expire, time_uptime) ?
3854 0 : dst->expire - time_uptime ;
3855 bp += sizeof(ipfw_dyn_rule);
3859 if (last != NULL) /* mark last dynamic rule */
3860 bzero(&last->next, sizeof(last));
3862 return (bp - (char *)buf);
3867 * {set|get}sockopt parser.
3870 ipfw_ctl(struct sockopt *sopt)
3872 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
3873 int error, rule_num;
3875 struct ip_fw *buf, *rule;
3876 u_int32_t rulenum[2];
3878 error = suser(sopt->sopt_td);
3883 * Disallow modifications in really-really secure mode, but still allow
3884 * the logging counters to be reset.
3886 if (sopt->sopt_name == IP_FW_ADD ||
3887 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3888 #if __FreeBSD_version >= 500034
3889 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3892 #else /* FreeBSD 4.x */
3893 if (securelevel >= 3)
3900 switch (sopt->sopt_name) {
3903 * pass up a copy of the current rules. Static rules
3904 * come first (the last of which has number IPFW_DEFAULT_RULE),
3905 * followed by a possibly empty list of dynamic rule.
3906 * The last dynamic rule has NULL in the "next" field.
3908 * Note that the calculated size is used to bound the
3909 * amount of data returned to the user. The rule set may
3910 * change between calculating the size and returning the
3911 * data in which case we'll just return what fits.
3913 size = static_len; /* size of static rules */
3914 if (ipfw_dyn_v) /* add size of dyn.rules */
3915 size += (dyn_count * sizeof(ipfw_dyn_rule));
3918 * XXX todo: if the user passes a short length just to know
3919 * how much room is needed, do not bother filling up the
3920 * buffer, just jump to the sooptcopyout.
3922 buf = malloc(size, M_TEMP, M_WAITOK);
3923 error = sooptcopyout(sopt, buf,
3924 ipfw_getrules(&layer3_chain, buf, size));
3930 * Normally we cannot release the lock on each iteration.
3931 * We could do it here only because we start from the head all
3932 * the times so there is no risk of missing some entries.
3933 * On the other hand, the risk is that we end up with
3934 * a very inconsistent ruleset, so better keep the lock
3935 * around the whole cycle.
3937 * XXX this code can be improved by resetting the head of
3938 * the list to point to the default rule, and then freeing
3939 * the old list without the need for a lock.
3942 IPFW_WLOCK(&layer3_chain);
3943 layer3_chain.reap = NULL;
3944 free_chain(&layer3_chain, 0 /* keep default rule */);
3945 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3946 IPFW_WUNLOCK(&layer3_chain);
3947 if (layer3_chain.reap != NULL)
3952 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3953 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3954 sizeof(struct ip_fw) );
3956 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3958 error = add_rule(&layer3_chain, rule);
3959 size = RULESIZE(rule);
3960 if (!error && sopt->sopt_dir == SOPT_GET)
3961 error = sooptcopyout(sopt, rule, size);
3968 * IP_FW_DEL is used for deleting single rules or sets,
3969 * and (ab)used to atomically manipulate sets. Argument size
3970 * is used to distinguish between the two:
3972 * delete single rule or set of rules,
3973 * or reassign rules (or sets) to a different set.
3974 * 2*sizeof(u_int32_t)
3975 * atomic disable/enable sets.
3976 * first u_int32_t contains sets to be disabled,
3977 * second u_int32_t contains sets to be enabled.
3979 error = sooptcopyin(sopt, rulenum,
3980 2*sizeof(u_int32_t), sizeof(u_int32_t));
3983 size = sopt->sopt_valsize;
3984 if (size == sizeof(u_int32_t)) /* delete or reassign */
3985 error = del_entry(&layer3_chain, rulenum[0]);
3986 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3988 (set_disable | rulenum[0]) & ~rulenum[1] &
3989 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3995 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3997 if (sopt->sopt_val != 0) {
3998 error = sooptcopyin(sopt, &rule_num,
3999 sizeof(int), sizeof(int));
4003 error = zero_entry(&layer3_chain, rule_num,
4004 sopt->sopt_name == IP_FW_RESETLOG);
4007 case IP_FW_TABLE_ADD:
4009 ipfw_table_entry ent;
4011 error = sooptcopyin(sopt, &ent,
4012 sizeof(ent), sizeof(ent));
4015 error = add_table_entry(ent.tbl, ent.addr,
4016 ent.masklen, ent.value);
4020 case IP_FW_TABLE_DEL:
4022 ipfw_table_entry ent;
4024 error = sooptcopyin(sopt, &ent,
4025 sizeof(ent), sizeof(ent));
4028 error = del_table_entry(ent.tbl, ent.addr, ent.masklen);
4032 case IP_FW_TABLE_FLUSH:
4036 error = sooptcopyin(sopt, &tbl,
4037 sizeof(tbl), sizeof(tbl));
4040 error = flush_table(tbl);
4044 case IP_FW_TABLE_GETSIZE:
4048 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
4051 if ((error = count_table(tbl, &cnt)))
4053 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4057 case IP_FW_TABLE_LIST:
4061 if (sopt->sopt_valsize < sizeof(*tbl)) {
4065 size = sopt->sopt_valsize;
4066 tbl = malloc(size, M_TEMP, M_WAITOK);
4071 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4076 tbl->size = (size - sizeof(*tbl)) /
4077 sizeof(ipfw_table_entry);
4078 error = dump_table(tbl);
4083 error = sooptcopyout(sopt, tbl, size);
4089 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4098 * dummynet needs a reference to the default rule, because rules can be
4099 * deleted while packets hold a reference to them. When this happens,
4100 * dummynet changes the reference to the default rule (it could well be a
4101 * NULL pointer, but this way we do not need to check for the special
4102 * case, plus here he have info on the default behaviour).
4104 struct ip_fw *ip_fw_default_rule;
4107 * This procedure is only used to handle keepalives. It is invoked
4108 * every dyn_keepalive_period
4111 ipfw_tick(void * __unused unused)
4113 struct mbuf *m0, *m, *mnext, **mtailp;
4117 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4121 * We make a chain of packets to go out here -- not deferring
4122 * until after we drop the IPFW dynamic rule lock would result
4123 * in a lock order reversal with the normal packet input -> ipfw
4129 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4130 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4131 if (q->dyn_type == O_LIMIT_PARENT)
4133 if (q->id.proto != IPPROTO_TCP)
4135 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4137 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4139 continue; /* too early */
4140 if (TIME_LEQ(q->expire, time_uptime))
4141 continue; /* too late, rule expired */
4143 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4144 q->ack_fwd, TH_SYN);
4145 if (*mtailp != NULL)
4146 mtailp = &(*mtailp)->m_nextpkt;
4147 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4149 if (*mtailp != NULL)
4150 mtailp = &(*mtailp)->m_nextpkt;
4154 for (m = mnext = m0; m != NULL; m = mnext) {
4155 mnext = m->m_nextpkt;
4156 m->m_nextpkt = NULL;
4157 ip_output(m, NULL, NULL, 0, NULL, NULL);
4160 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4166 struct ip_fw default_rule;
4170 /* Setup IPv6 fw sysctl tree. */
4171 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4172 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4173 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4174 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4175 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4176 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4177 &fw_deny_unknown_exthdrs, 0,
4178 "Deny packets with unknown IPv6 Extension Headers");
4181 layer3_chain.rules = NULL;
4182 layer3_chain.want_write = 0;
4183 layer3_chain.busy_count = 0;
4184 cv_init(&layer3_chain.cv, "Condition variable for IPFW rw locks");
4185 IPFW_LOCK_INIT(&layer3_chain);
4186 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4187 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4189 IPFW_DYN_LOCK_INIT();
4190 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4192 bzero(&default_rule, sizeof default_rule);
4194 default_rule.act_ofs = 0;
4195 default_rule.rulenum = IPFW_DEFAULT_RULE;
4196 default_rule.cmd_len = 1;
4197 default_rule.set = RESVD_SET;
4199 default_rule.cmd[0].len = 1;
4200 default_rule.cmd[0].opcode =
4201 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4206 error = add_rule(&layer3_chain, &default_rule);
4208 printf("ipfw2: error %u initializing default rule "
4209 "(support disabled)\n", error);
4210 IPFW_DYN_LOCK_DESTROY();
4211 IPFW_LOCK_DESTROY(&layer3_chain);
4215 ip_fw_default_rule = layer3_chain.rules;
4216 printf("ipfw2 (+ipv6) initialized, divert %s, "
4217 "rule-based forwarding "
4218 #ifdef IPFIREWALL_FORWARD
4223 "default to %s, logging ",
4229 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4231 #ifdef IPFIREWALL_VERBOSE
4234 #ifdef IPFIREWALL_VERBOSE_LIMIT
4235 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4237 if (fw_verbose == 0)
4238 printf("disabled\n");
4239 else if (verbose_limit == 0)
4240 printf("unlimited\n");
4242 printf("limited to %d packets/entry by default\n",
4246 ip_fw_ctl_ptr = ipfw_ctl;
4247 ip_fw_chk_ptr = ipfw_chk;
4248 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4258 ip_fw_chk_ptr = NULL;
4259 ip_fw_ctl_ptr = NULL;
4260 callout_drain(&ipfw_timeout);
4261 IPFW_WLOCK(&layer3_chain);
4262 layer3_chain.reap = NULL;
4263 free_chain(&layer3_chain, 1 /* kill default rule */);
4264 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4265 IPFW_WUNLOCK(&layer3_chain);
4269 IPFW_DYN_LOCK_DESTROY();
4270 uma_zdestroy(ipfw_dyn_rule_zone);
4271 IPFW_LOCK_DESTROY(&layer3_chain);
4274 /* Free IPv6 fw sysctl tree. */
4275 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4278 printf("IP firewall unloaded\n");