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"
41 #error IPFIREWALL requires INET.
44 #include "opt_inet6.h"
45 #include "opt_ipsec.h"
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];
129 #define IPFW_TABLES_MAX 128
131 struct ip_fw *rules; /* list of rules */
132 struct ip_fw *reap; /* list of rules to reap */
133 struct radix_node_head *tables[IPFW_TABLES_MAX];
134 struct mtx mtx; /* lock guarding rule list */
135 int busy_count; /* busy count for rw locks */
139 #define IPFW_LOCK_INIT(_chain) \
140 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
141 MTX_DEF | MTX_RECURSE)
142 #define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
143 #define IPFW_WLOCK_ASSERT(_chain) do { \
144 mtx_assert(&(_chain)->mtx, MA_OWNED); \
145 NET_ASSERT_GIANT(); \
149 IPFW_RLOCK(struct ip_fw_chain *chain)
151 mtx_lock(&chain->mtx);
153 mtx_unlock(&chain->mtx);
157 IPFW_RUNLOCK(struct ip_fw_chain *chain)
159 mtx_lock(&chain->mtx);
161 if (chain->busy_count == 0 && chain->want_write)
162 cv_signal(&chain->cv);
163 mtx_unlock(&chain->mtx);
167 IPFW_WLOCK(struct ip_fw_chain *chain)
169 mtx_lock(&chain->mtx);
171 while (chain->busy_count > 0)
172 cv_wait(&chain->cv, &chain->mtx);
176 IPFW_WUNLOCK(struct ip_fw_chain *chain)
179 cv_signal(&chain->cv);
180 mtx_unlock(&chain->mtx);
184 * list of rules for layer 3
186 static struct ip_fw_chain layer3_chain;
188 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
189 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
192 struct radix_node rn[2];
193 struct sockaddr_in addr, mask;
197 static int fw_debug = 1;
198 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
201 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
202 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
203 CTLFLAG_RW | CTLFLAG_SECURE3,
204 &fw_enable, 0, "Enable ipfw");
205 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
206 &autoinc_step, 0, "Rule number autincrement step");
207 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
208 CTLFLAG_RW | CTLFLAG_SECURE3,
210 "Only do a single pass through ipfw when using dummynet(4)");
211 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
212 &fw_debug, 0, "Enable printing of debug ip_fw statements");
213 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
214 CTLFLAG_RW | CTLFLAG_SECURE3,
215 &fw_verbose, 0, "Log matches to ipfw rules");
216 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
217 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
220 * Description of dynamic rules.
222 * Dynamic rules are stored in lists accessed through a hash table
223 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
224 * be modified through the sysctl variable dyn_buckets which is
225 * updated when the table becomes empty.
227 * XXX currently there is only one list, ipfw_dyn.
229 * When a packet is received, its address fields are first masked
230 * with the mask defined for the rule, then hashed, then matched
231 * against the entries in the corresponding list.
232 * Dynamic rules can be used for different purposes:
234 * + enforcing limits on the number of sessions;
235 * + in-kernel NAT (not implemented yet)
237 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
238 * measured in seconds and depending on the flags.
240 * The total number of dynamic rules is stored in dyn_count.
241 * The max number of dynamic rules is dyn_max. When we reach
242 * the maximum number of rules we do not create anymore. This is
243 * done to avoid consuming too much memory, but also too much
244 * time when searching on each packet (ideally, we should try instead
245 * to put a limit on the length of the list on each bucket...).
247 * Each dynamic rule holds a pointer to the parent ipfw rule so
248 * we know what action to perform. Dynamic rules are removed when
249 * the parent rule is deleted. XXX we should make them survive.
251 * There are some limitations with dynamic rules -- we do not
252 * obey the 'randomized match', and we do not do multiple
253 * passes through the firewall. XXX check the latter!!!
255 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
256 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
257 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
259 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
260 #define IPFW_DYN_LOCK_INIT() \
261 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
262 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
263 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
264 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
265 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
268 * Timeouts for various events in handing dynamic rules.
270 static u_int32_t dyn_ack_lifetime = 300;
271 static u_int32_t dyn_syn_lifetime = 20;
272 static u_int32_t dyn_fin_lifetime = 1;
273 static u_int32_t dyn_rst_lifetime = 1;
274 static u_int32_t dyn_udp_lifetime = 10;
275 static u_int32_t dyn_short_lifetime = 5;
278 * Keepalives are sent if dyn_keepalive is set. They are sent every
279 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
280 * seconds of lifetime of a rule.
281 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
282 * than dyn_keepalive_period.
285 static u_int32_t dyn_keepalive_interval = 20;
286 static u_int32_t dyn_keepalive_period = 5;
287 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
289 static u_int32_t static_count; /* # of static rules */
290 static u_int32_t static_len; /* size in bytes of static rules */
291 static u_int32_t dyn_count; /* # of dynamic rules */
292 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
294 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
295 &dyn_buckets, 0, "Number of dyn. buckets");
296 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
297 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
298 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
299 &dyn_count, 0, "Number of dyn. rules");
300 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
301 &dyn_max, 0, "Max number of dyn. rules");
302 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
303 &static_count, 0, "Number of static rules");
304 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
305 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
306 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
307 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
308 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
309 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
310 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
311 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
312 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
313 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
314 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
315 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
316 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
317 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
321 * IPv6 specific variables
323 SYSCTL_DECL(_net_inet6_ip6);
325 static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
326 static struct sysctl_oid *ip6_fw_sysctl_tree;
328 #endif /* SYSCTL_NODE */
330 static int fw_deny_unknown_exthdrs = 1;
334 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
335 * Other macros just cast void * into the appropriate type
337 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
338 #define TCP(p) ((struct tcphdr *)(p))
339 #define UDP(p) ((struct udphdr *)(p))
340 #define ICMP(p) ((struct icmphdr *)(p))
341 #define ICMP6(p) ((struct icmp6_hdr *)(p))
344 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
346 int type = icmp->icmp_type;
348 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
351 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
352 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
355 is_icmp_query(struct icmphdr *icmp)
357 int type = icmp->icmp_type;
359 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
364 * The following checks use two arrays of 8 or 16 bits to store the
365 * bits that we want set or clear, respectively. They are in the
366 * low and high half of cmd->arg1 or cmd->d[0].
368 * We scan options and store the bits we find set. We succeed if
370 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
372 * The code is sometimes optimized not to store additional variables.
376 flags_match(ipfw_insn *cmd, u_int8_t bits)
381 if ( ((cmd->arg1 & 0xff) & bits) != 0)
382 return 0; /* some bits we want set were clear */
383 want_clear = (cmd->arg1 >> 8) & 0xff;
384 if ( (want_clear & bits) != want_clear)
385 return 0; /* some bits we want clear were set */
390 ipopts_match(struct ip *ip, ipfw_insn *cmd)
392 int optlen, bits = 0;
393 u_char *cp = (u_char *)(ip + 1);
394 int x = (ip->ip_hl << 2) - sizeof (struct ip);
396 for (; x > 0; x -= optlen, cp += optlen) {
397 int opt = cp[IPOPT_OPTVAL];
399 if (opt == IPOPT_EOL)
401 if (opt == IPOPT_NOP)
404 optlen = cp[IPOPT_OLEN];
405 if (optlen <= 0 || optlen > x)
406 return 0; /* invalid or truncated */
414 bits |= IP_FW_IPOPT_LSRR;
418 bits |= IP_FW_IPOPT_SSRR;
422 bits |= IP_FW_IPOPT_RR;
426 bits |= IP_FW_IPOPT_TS;
430 return (flags_match(cmd, bits));
434 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
436 int optlen, bits = 0;
437 u_char *cp = (u_char *)(tcp + 1);
438 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
440 for (; x > 0; x -= optlen, cp += optlen) {
442 if (opt == TCPOPT_EOL)
444 if (opt == TCPOPT_NOP)
458 bits |= IP_FW_TCPOPT_MSS;
462 bits |= IP_FW_TCPOPT_WINDOW;
465 case TCPOPT_SACK_PERMITTED:
467 bits |= IP_FW_TCPOPT_SACK;
470 case TCPOPT_TIMESTAMP:
471 bits |= IP_FW_TCPOPT_TS;
476 return (flags_match(cmd, bits));
480 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
482 if (ifp == NULL) /* no iface with this packet, match fails */
484 /* Check by name or by IP address */
485 if (cmd->name[0] != '\0') { /* match by name */
488 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
491 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
498 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
499 if (ia->ifa_addr == NULL)
501 if (ia->ifa_addr->sa_family != AF_INET)
503 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
504 (ia->ifa_addr))->sin_addr.s_addr)
505 return(1); /* match */
508 return(0); /* no match, fail ... */
512 * The verify_path function checks if a route to the src exists and
513 * if it is reachable via ifp (when provided).
515 * The 'verrevpath' option checks that the interface that an IP packet
516 * arrives on is the same interface that traffic destined for the
517 * packet's source address would be routed out of. The 'versrcreach'
518 * option just checks that the source address is reachable via any route
519 * (except default) in the routing table. These two are a measure to block
520 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
521 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
522 * is purposely reminiscent of the Cisco IOS command,
524 * ip verify unicast reverse-path
525 * ip verify unicast source reachable-via any
527 * which implements the same functionality. But note that syntax is
528 * misleading. The check may be performed on all IP packets whether unicast,
529 * multicast, or broadcast.
532 verify_path(struct in_addr src, struct ifnet *ifp)
535 struct sockaddr_in *dst;
537 bzero(&ro, sizeof(ro));
539 dst = (struct sockaddr_in *)&(ro.ro_dst);
540 dst->sin_family = AF_INET;
541 dst->sin_len = sizeof(*dst);
543 rtalloc_ign(&ro, RTF_CLONING);
545 if (ro.ro_rt == NULL)
548 /* if ifp is provided, check for equality with rtentry */
549 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) {
554 /* if no ifp provided, check if rtentry is not default route */
556 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
561 /* or if this is a blackhole/reject route */
562 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
567 /* found valid route */
574 * ipv6 specific rules here...
577 icmp6type_match (int type, ipfw_insn_u32 *cmd)
579 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
583 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
586 for (i=0; i <= cmd->o.arg1; ++i )
587 if (curr_flow == cmd->d[i] )
592 /* support for IP6_*_ME opcodes */
594 search_ip6_addr_net (struct in6_addr * ip6_addr)
598 struct in6_ifaddr *fdm;
599 struct in6_addr copia;
601 TAILQ_FOREACH(mdc, &ifnet, if_link)
602 for (mdc2 = mdc->if_addrlist.tqh_first; mdc2;
603 mdc2 = mdc2->ifa_list.tqe_next) {
606 if (mdc2->ifa_addr->sa_family == AF_INET6) {
607 fdm = (struct in6_ifaddr *)mdc2;
608 copia = fdm->ia_addr.sin6_addr;
609 /* need for leaving scope_id in the sock_addr */
610 in6_clearscope(&copia);
611 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
619 verify_path6(struct in6_addr *src, struct ifnet *ifp)
622 struct sockaddr_in6 *dst;
624 bzero(&ro, sizeof(ro));
626 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
627 dst->sin6_family = AF_INET6;
628 dst->sin6_len = sizeof(*dst);
629 dst->sin6_addr = *src;
630 rtalloc_ign((struct route *)&ro, RTF_CLONING);
632 if (ro.ro_rt == NULL)
636 * if ifp is provided, check for equality with rtentry
637 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
638 * to support the case of sending packets to an address of our own.
639 * (where the former interface is the first argument of if_simloop()
640 * (=ifp), the latter is lo0)
642 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
647 /* if no ifp provided, check if rtentry is not default route */
649 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
654 /* or if this is a blackhole/reject route */
655 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
660 /* found valid route */
666 hash_packet6(struct ipfw_flow_id *id)
669 i = (id->dst_ip6.__u6_addr.__u6_addr32[0]) ^
670 (id->dst_ip6.__u6_addr.__u6_addr32[1]) ^
671 (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
672 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
673 (id->dst_port) ^ (id->src_port) ^ (id->flow_id6);
678 is_icmp6_query(int icmp6_type)
680 if ((icmp6_type <= ICMP6_MAXTYPE) &&
681 (icmp6_type == ICMP6_ECHO_REQUEST ||
682 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
683 icmp6_type == ICMP6_WRUREQUEST ||
684 icmp6_type == ICMP6_FQDN_QUERY ||
685 icmp6_type == ICMP6_NI_QUERY))
692 send_reject6(struct ip_fw_args *args, int code, u_short offset, u_int hlen)
694 if (code == ICMP6_UNREACH_RST && offset == 0 &&
695 args->f_id.proto == IPPROTO_TCP) {
705 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
706 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
711 ip6 = mtod(args->m, struct ip6_hdr *);
712 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
714 if ((tcp->th_flags & TH_RST) != 0) {
721 ti.th.th_seq = ntohl(ti.th.th_seq);
722 ti.th.th_ack = ntohl(ti.th.th_ack);
723 ti.ip6.ip6_nxt = IPPROTO_TCP;
725 if (ti.th.th_flags & TH_ACK) {
731 if (((args->m)->m_flags & M_PKTHDR) != 0) {
732 ack += (args->m)->m_pkthdr.len - hlen
733 - (ti.th.th_off << 2);
734 } else if (ip6->ip6_plen) {
735 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
736 - hlen - (ti.th.th_off << 2);
741 if (tcp->th_flags & TH_SYN)
744 flags = TH_RST|TH_ACK;
746 bcopy(&ti, ip6, sizeof(ti));
747 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
748 args->m, ack, seq, flags);
750 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
751 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
761 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
763 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
764 #define SNP(buf) buf, sizeof(buf)
767 * We enter here when we have a rule with O_LOG.
768 * XXX this function alone takes about 2Kbytes of code!
771 ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
772 struct mbuf *m, struct ifnet *oif, u_short offset)
774 struct ether_header *eh = args->eh;
776 int limit_reached = 0;
777 char action2[40], proto[128], fragment[32];
782 if (f == NULL) { /* bogus pkt */
783 if (verbose_limit != 0 && norule_counter >= verbose_limit)
786 if (norule_counter == verbose_limit)
787 limit_reached = verbose_limit;
789 } else { /* O_LOG is the first action, find the real one */
790 ipfw_insn *cmd = ACTION_PTR(f);
791 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
793 if (l->max_log != 0 && l->log_left == 0)
796 if (l->log_left == 0)
797 limit_reached = l->max_log;
798 cmd += F_LEN(cmd); /* point to first action */
799 if (cmd->opcode == O_ALTQ) {
800 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
802 snprintf(SNPARGS(action2, 0), "Altq %d",
806 if (cmd->opcode == O_PROB)
810 switch (cmd->opcode) {
816 if (cmd->arg1==ICMP_REJECT_RST)
818 else if (cmd->arg1==ICMP_UNREACH_HOST)
821 snprintf(SNPARGS(action2, 0), "Unreach %d",
826 if (cmd->arg1==ICMP6_UNREACH_RST)
829 snprintf(SNPARGS(action2, 0), "Unreach %d",
840 snprintf(SNPARGS(action2, 0), "Divert %d",
844 snprintf(SNPARGS(action2, 0), "Tee %d",
848 snprintf(SNPARGS(action2, 0), "SkipTo %d",
852 snprintf(SNPARGS(action2, 0), "Pipe %d",
856 snprintf(SNPARGS(action2, 0), "Queue %d",
860 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
863 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
864 inet_ntoa(sa->sa.sin_addr));
866 snprintf(SNPARGS(action2, len), ":%d",
871 snprintf(SNPARGS(action2, 0), "Netgraph %d",
875 snprintf(SNPARGS(action2, 0), "Ngtee %d",
884 if (hlen == 0) { /* non-ip */
885 snprintf(SNPARGS(proto, 0), "MAC");
889 char src[48], dst[48];
890 struct icmphdr *icmp;
893 /* Initialize to make compiler happy. */
894 struct ip *ip = NULL;
896 struct ip6_hdr *ip6 = NULL;
897 struct icmp6_hdr *icmp6;
902 if (args->f_id.addr_type == 6) {
903 snprintf(src, sizeof(src), "[%s]",
904 ip6_sprintf(&args->f_id.src_ip6));
905 snprintf(dst, sizeof(dst), "[%s]",
906 ip6_sprintf(&args->f_id.dst_ip6));
908 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
909 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
910 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
914 ip = mtod(m, struct ip *);
915 tcp = L3HDR(struct tcphdr, ip);
916 udp = L3HDR(struct udphdr, ip);
918 inet_ntoa_r(ip->ip_src, src);
919 inet_ntoa_r(ip->ip_dst, dst);
922 switch (args->f_id.proto) {
924 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
926 snprintf(SNPARGS(proto, len), ":%d %s:%d",
927 ntohs(tcp->th_sport),
929 ntohs(tcp->th_dport));
931 snprintf(SNPARGS(proto, len), " %s", dst);
935 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
937 snprintf(SNPARGS(proto, len), ":%d %s:%d",
938 ntohs(udp->uh_sport),
940 ntohs(udp->uh_dport));
942 snprintf(SNPARGS(proto, len), " %s", dst);
946 icmp = L3HDR(struct icmphdr, ip);
948 len = snprintf(SNPARGS(proto, 0),
950 icmp->icmp_type, icmp->icmp_code);
952 len = snprintf(SNPARGS(proto, 0), "ICMP ");
953 len += snprintf(SNPARGS(proto, len), "%s", src);
954 snprintf(SNPARGS(proto, len), " %s", dst);
958 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
960 len = snprintf(SNPARGS(proto, 0),
962 icmp6->icmp6_type, icmp6->icmp6_code);
964 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
965 len += snprintf(SNPARGS(proto, len), "%s", src);
966 snprintf(SNPARGS(proto, len), " %s", dst);
970 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
971 args->f_id.proto, src);
972 snprintf(SNPARGS(proto, len), " %s", dst);
977 if (args->f_id.addr_type == 6) {
978 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
979 snprintf(SNPARGS(fragment, 0),
980 " (frag %08x:%d@%d%s)",
982 ntohs(ip6->ip6_plen) - hlen,
983 ntohs(offset & IP6F_OFF_MASK) << 3,
984 (offset & IP6F_MORE_FRAG) ? "+" : "");
989 if (eh != NULL) { /* layer 2 packets are as on the wire */
990 ip_off = ntohs(ip->ip_off);
991 ip_len = ntohs(ip->ip_len);
996 if (ip_off & (IP_MF | IP_OFFMASK))
997 snprintf(SNPARGS(fragment, 0),
998 " (frag %d:%d@%d%s)",
999 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
1001 (ip_off & IP_MF) ? "+" : "");
1004 if (oif || m->m_pkthdr.rcvif)
1005 log(LOG_SECURITY | LOG_INFO,
1006 "ipfw: %d %s %s %s via %s%s\n",
1007 f ? f->rulenum : -1,
1008 action, proto, oif ? "out" : "in",
1009 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
1012 log(LOG_SECURITY | LOG_INFO,
1013 "ipfw: %d %s %s [no if info]%s\n",
1014 f ? f->rulenum : -1,
1015 action, proto, fragment);
1017 log(LOG_SECURITY | LOG_NOTICE,
1018 "ipfw: limit %d reached on entry %d\n",
1019 limit_reached, f ? f->rulenum : -1);
1023 * IMPORTANT: the hash function for dynamic rules must be commutative
1024 * in source and destination (ip,port), because rules are bidirectional
1025 * and we want to find both in the same bucket.
1028 hash_packet(struct ipfw_flow_id *id)
1033 if (IS_IP6_FLOW_ID(id))
1034 i = hash_packet6(id);
1037 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1038 i &= (curr_dyn_buckets - 1);
1043 * unlink a dynamic rule from a chain. prev is a pointer to
1044 * the previous one, q is a pointer to the rule to delete,
1045 * head is a pointer to the head of the queue.
1046 * Modifies q and potentially also head.
1048 #define UNLINK_DYN_RULE(prev, head, q) { \
1049 ipfw_dyn_rule *old_q = q; \
1051 /* remove a refcount to the parent */ \
1052 if (q->dyn_type == O_LIMIT) \
1053 q->parent->count--; \
1054 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1055 (q->id.src_ip), (q->id.src_port), \
1056 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1058 prev->next = q = q->next; \
1060 head = q = q->next; \
1062 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1064 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1067 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1069 * If keep_me == NULL, rules are deleted even if not expired,
1070 * otherwise only expired rules are removed.
1072 * The value of the second parameter is also used to point to identify
1073 * a rule we absolutely do not want to remove (e.g. because we are
1074 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1075 * rules). The pointer is only used for comparison, so any non-null
1079 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1081 static u_int32_t last_remove = 0;
1083 #define FORCE (keep_me == NULL)
1085 ipfw_dyn_rule *prev, *q;
1086 int i, pass = 0, max_pass = 0;
1088 IPFW_DYN_LOCK_ASSERT();
1090 if (ipfw_dyn_v == NULL || dyn_count == 0)
1092 /* do not expire more than once per second, it is useless */
1093 if (!FORCE && last_remove == time_uptime)
1095 last_remove = time_uptime;
1098 * because O_LIMIT refer to parent rules, during the first pass only
1099 * remove child and mark any pending LIMIT_PARENT, and remove
1100 * them in a second pass.
1103 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1104 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1106 * Logic can become complex here, so we split tests.
1110 if (rule != NULL && rule != q->rule)
1111 goto next; /* not the one we are looking for */
1112 if (q->dyn_type == O_LIMIT_PARENT) {
1114 * handle parent in the second pass,
1115 * record we need one.
1120 if (FORCE && q->count != 0 ) {
1121 /* XXX should not happen! */
1122 printf("ipfw: OUCH! cannot remove rule,"
1123 " count %d\n", q->count);
1127 !TIME_LEQ( q->expire, time_uptime ))
1130 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1131 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1139 if (pass++ < max_pass)
1145 * lookup a dynamic rule.
1147 static ipfw_dyn_rule *
1148 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1152 * stateful ipfw extensions.
1153 * Lookup into dynamic session queue
1155 #define MATCH_REVERSE 0
1156 #define MATCH_FORWARD 1
1157 #define MATCH_NONE 2
1158 #define MATCH_UNKNOWN 3
1159 int i, dir = MATCH_NONE;
1160 ipfw_dyn_rule *prev, *q=NULL;
1162 IPFW_DYN_LOCK_ASSERT();
1164 if (ipfw_dyn_v == NULL)
1165 goto done; /* not found */
1166 i = hash_packet( pkt );
1167 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1168 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1170 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1171 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1174 if (pkt->proto == q->id.proto &&
1175 q->dyn_type != O_LIMIT_PARENT) {
1176 if (IS_IP6_FLOW_ID(pkt)) {
1177 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1178 &(q->id.src_ip6)) &&
1179 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1180 &(q->id.dst_ip6)) &&
1181 pkt->src_port == q->id.src_port &&
1182 pkt->dst_port == q->id.dst_port ) {
1183 dir = MATCH_FORWARD;
1186 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1187 &(q->id.dst_ip6)) &&
1188 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1189 &(q->id.src_ip6)) &&
1190 pkt->src_port == q->id.dst_port &&
1191 pkt->dst_port == q->id.src_port ) {
1192 dir = MATCH_REVERSE;
1196 if (pkt->src_ip == q->id.src_ip &&
1197 pkt->dst_ip == q->id.dst_ip &&
1198 pkt->src_port == q->id.src_port &&
1199 pkt->dst_port == q->id.dst_port ) {
1200 dir = MATCH_FORWARD;
1203 if (pkt->src_ip == q->id.dst_ip &&
1204 pkt->dst_ip == q->id.src_ip &&
1205 pkt->src_port == q->id.dst_port &&
1206 pkt->dst_port == q->id.src_port ) {
1207 dir = MATCH_REVERSE;
1217 goto done; /* q = NULL, not found */
1219 if ( prev != NULL) { /* found and not in front */
1220 prev->next = q->next;
1221 q->next = ipfw_dyn_v[i];
1224 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1225 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1227 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1228 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1229 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1231 case TH_SYN: /* opening */
1232 q->expire = time_uptime + dyn_syn_lifetime;
1235 case BOTH_SYN: /* move to established */
1236 case BOTH_SYN | TH_FIN : /* one side tries to close */
1237 case BOTH_SYN | (TH_FIN << 8) :
1239 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1240 u_int32_t ack = ntohl(tcp->th_ack);
1241 if (dir == MATCH_FORWARD) {
1242 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1244 else { /* ignore out-of-sequence */
1248 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1250 else { /* ignore out-of-sequence */
1255 q->expire = time_uptime + dyn_ack_lifetime;
1258 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1259 if (dyn_fin_lifetime >= dyn_keepalive_period)
1260 dyn_fin_lifetime = dyn_keepalive_period - 1;
1261 q->expire = time_uptime + dyn_fin_lifetime;
1267 * reset or some invalid combination, but can also
1268 * occur if we use keep-state the wrong way.
1270 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1271 printf("invalid state: 0x%x\n", q->state);
1273 if (dyn_rst_lifetime >= dyn_keepalive_period)
1274 dyn_rst_lifetime = dyn_keepalive_period - 1;
1275 q->expire = time_uptime + dyn_rst_lifetime;
1278 } else if (pkt->proto == IPPROTO_UDP) {
1279 q->expire = time_uptime + dyn_udp_lifetime;
1281 /* other protocols */
1282 q->expire = time_uptime + dyn_short_lifetime;
1285 if (match_direction)
1286 *match_direction = dir;
1290 static ipfw_dyn_rule *
1291 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1297 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1300 /* NB: return table locked when q is not NULL */
1305 realloc_dynamic_table(void)
1307 IPFW_DYN_LOCK_ASSERT();
1310 * Try reallocation, make sure we have a power of 2 and do
1311 * not allow more than 64k entries. In case of overflow,
1315 if (dyn_buckets > 65536)
1317 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1318 dyn_buckets = curr_dyn_buckets; /* reset */
1321 curr_dyn_buckets = dyn_buckets;
1322 if (ipfw_dyn_v != NULL)
1323 free(ipfw_dyn_v, M_IPFW);
1325 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1326 M_IPFW, M_NOWAIT | M_ZERO);
1327 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1329 curr_dyn_buckets /= 2;
1334 * Install state of type 'type' for a dynamic session.
1335 * The hash table contains two type of rules:
1336 * - regular rules (O_KEEP_STATE)
1337 * - rules for sessions with limited number of sess per user
1338 * (O_LIMIT). When they are created, the parent is
1339 * increased by 1, and decreased on delete. In this case,
1340 * the third parameter is the parent rule and not the chain.
1341 * - "parent" rules for the above (O_LIMIT_PARENT).
1343 static ipfw_dyn_rule *
1344 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1349 IPFW_DYN_LOCK_ASSERT();
1351 if (ipfw_dyn_v == NULL ||
1352 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1353 realloc_dynamic_table();
1354 if (ipfw_dyn_v == NULL)
1355 return NULL; /* failed ! */
1357 i = hash_packet(id);
1359 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1361 printf ("ipfw: sorry cannot allocate state\n");
1365 /* increase refcount on parent, and set pointer */
1366 if (dyn_type == O_LIMIT) {
1367 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1368 if ( parent->dyn_type != O_LIMIT_PARENT)
1369 panic("invalid parent");
1372 rule = parent->rule;
1376 r->expire = time_uptime + dyn_syn_lifetime;
1378 r->dyn_type = dyn_type;
1379 r->pcnt = r->bcnt = 0;
1383 r->next = ipfw_dyn_v[i];
1386 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1388 (r->id.src_ip), (r->id.src_port),
1389 (r->id.dst_ip), (r->id.dst_port),
1395 * lookup dynamic parent rule using pkt and rule as search keys.
1396 * If the lookup fails, then install one.
1398 static ipfw_dyn_rule *
1399 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1404 IPFW_DYN_LOCK_ASSERT();
1407 int is_v6 = IS_IP6_FLOW_ID(pkt);
1408 i = hash_packet( pkt );
1409 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1410 if (q->dyn_type == O_LIMIT_PARENT &&
1412 pkt->proto == q->id.proto &&
1413 pkt->src_port == q->id.src_port &&
1414 pkt->dst_port == q->id.dst_port &&
1417 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1418 &(q->id.src_ip6)) &&
1419 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1420 &(q->id.dst_ip6))) ||
1422 pkt->src_ip == q->id.src_ip &&
1423 pkt->dst_ip == q->id.dst_ip)
1426 q->expire = time_uptime + dyn_short_lifetime;
1427 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1431 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1435 * Install dynamic state for rule type cmd->o.opcode
1437 * Returns 1 (failure) if state is not installed because of errors or because
1438 * session limitations are enforced.
1441 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1442 struct ip_fw_args *args)
1444 static int last_log;
1448 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1450 (args->f_id.src_ip), (args->f_id.src_port),
1451 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1455 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1457 if (q != NULL) { /* should never occur */
1458 if (last_log != time_uptime) {
1459 last_log = time_uptime;
1460 printf("ipfw: install_state: entry already present, done\n");
1466 if (dyn_count >= dyn_max)
1468 * Run out of slots, try to remove any expired rule.
1470 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1472 if (dyn_count >= dyn_max) {
1473 if (last_log != time_uptime) {
1474 last_log = time_uptime;
1475 printf("ipfw: install_state: Too many dynamic rules\n");
1478 return 1; /* cannot install, notify caller */
1481 switch (cmd->o.opcode) {
1482 case O_KEEP_STATE: /* bidir rule */
1483 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1486 case O_LIMIT: /* limit number of sessions */
1488 u_int16_t limit_mask = cmd->limit_mask;
1489 struct ipfw_flow_id id;
1490 ipfw_dyn_rule *parent;
1492 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1495 id.dst_ip = id.src_ip = 0;
1496 id.dst_port = id.src_port = 0;
1497 id.proto = args->f_id.proto;
1499 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1500 if (limit_mask & DYN_SRC_ADDR)
1501 id.src_ip6 = args->f_id.src_ip6;
1502 if (limit_mask & DYN_DST_ADDR)
1503 id.dst_ip6 = args->f_id.dst_ip6;
1505 if (limit_mask & DYN_SRC_ADDR)
1506 id.src_ip = args->f_id.src_ip;
1507 if (limit_mask & DYN_DST_ADDR)
1508 id.dst_ip = args->f_id.dst_ip;
1510 if (limit_mask & DYN_SRC_PORT)
1511 id.src_port = args->f_id.src_port;
1512 if (limit_mask & DYN_DST_PORT)
1513 id.dst_port = args->f_id.dst_port;
1514 parent = lookup_dyn_parent(&id, rule);
1515 if (parent == NULL) {
1516 printf("ipfw: add parent failed\n");
1520 if (parent->count >= cmd->conn_limit) {
1522 * See if we can remove some expired rule.
1524 remove_dyn_rule(rule, parent);
1525 if (parent->count >= cmd->conn_limit) {
1526 if (fw_verbose && last_log != time_uptime) {
1527 last_log = time_uptime;
1528 log(LOG_SECURITY | LOG_DEBUG,
1529 "drop session, too many entries\n");
1535 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1539 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1543 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1549 * Generate a TCP packet, containing either a RST or a keepalive.
1550 * When flags & TH_RST, we are sending a RST packet, because of a
1551 * "reset" action matched the packet.
1552 * Otherwise we are sending a keepalive, and flags & TH_
1554 static struct mbuf *
1555 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1561 MGETHDR(m, M_DONTWAIT, MT_DATA);
1564 m->m_pkthdr.rcvif = (struct ifnet *)0;
1565 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1566 m->m_data += max_linkhdr;
1568 ip = mtod(m, struct ip *);
1569 bzero(ip, m->m_len);
1570 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1571 ip->ip_p = IPPROTO_TCP;
1574 * Assume we are sending a RST (or a keepalive in the reverse
1575 * direction), swap src and destination addresses and ports.
1577 ip->ip_src.s_addr = htonl(id->dst_ip);
1578 ip->ip_dst.s_addr = htonl(id->src_ip);
1579 tcp->th_sport = htons(id->dst_port);
1580 tcp->th_dport = htons(id->src_port);
1581 if (flags & TH_RST) { /* we are sending a RST */
1582 if (flags & TH_ACK) {
1583 tcp->th_seq = htonl(ack);
1584 tcp->th_ack = htonl(0);
1585 tcp->th_flags = TH_RST;
1589 tcp->th_seq = htonl(0);
1590 tcp->th_ack = htonl(seq);
1591 tcp->th_flags = TH_RST | TH_ACK;
1595 * We are sending a keepalive. flags & TH_SYN determines
1596 * the direction, forward if set, reverse if clear.
1597 * NOTE: seq and ack are always assumed to be correct
1598 * as set by the caller. This may be confusing...
1600 if (flags & TH_SYN) {
1602 * we have to rewrite the correct addresses!
1604 ip->ip_dst.s_addr = htonl(id->dst_ip);
1605 ip->ip_src.s_addr = htonl(id->src_ip);
1606 tcp->th_dport = htons(id->dst_port);
1607 tcp->th_sport = htons(id->src_port);
1609 tcp->th_seq = htonl(seq);
1610 tcp->th_ack = htonl(ack);
1611 tcp->th_flags = TH_ACK;
1614 * set ip_len to the payload size so we can compute
1615 * the tcp checksum on the pseudoheader
1616 * XXX check this, could save a couple of words ?
1618 ip->ip_len = htons(sizeof(struct tcphdr));
1619 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1621 * now fill fields left out earlier
1623 ip->ip_ttl = ip_defttl;
1624 ip->ip_len = m->m_pkthdr.len;
1625 m->m_flags |= M_SKIP_FIREWALL;
1630 * sends a reject message, consuming the mbuf passed as an argument.
1633 send_reject(struct ip_fw_args *args, int code, u_short offset, int ip_len)
1636 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1637 /* We need the IP header in host order for icmp_error(). */
1638 if (args->eh != NULL) {
1639 struct ip *ip = mtod(args->m, struct ip *);
1640 ip->ip_len = ntohs(ip->ip_len);
1641 ip->ip_off = ntohs(ip->ip_off);
1643 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1644 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1645 struct tcphdr *const tcp =
1646 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1647 if ( (tcp->th_flags & TH_RST) == 0) {
1649 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1651 tcp->th_flags | TH_RST);
1653 ip_output(m, NULL, NULL, 0, NULL, NULL);
1663 * Given an ip_fw *, lookup_next_rule will return a pointer
1664 * to the next rule, which can be either the jump
1665 * target (for skipto instructions) or the next one in the list (in
1666 * all other cases including a missing jump target).
1667 * The result is also written in the "next_rule" field of the rule.
1668 * Backward jumps are not allowed, so start looking from the next
1671 * This never returns NULL -- in case we do not have an exact match,
1672 * the next rule is returned. When the ruleset is changed,
1673 * pointers are flushed so we are always correct.
1676 static struct ip_fw *
1677 lookup_next_rule(struct ip_fw *me)
1679 struct ip_fw *rule = NULL;
1682 /* look for action, in case it is a skipto */
1683 cmd = ACTION_PTR(me);
1684 if (cmd->opcode == O_LOG)
1686 if (cmd->opcode == O_ALTQ)
1688 if ( cmd->opcode == O_SKIPTO )
1689 for (rule = me->next; rule ; rule = rule->next)
1690 if (rule->rulenum >= cmd->arg1)
1692 if (rule == NULL) /* failure or not a skipto */
1694 me->next_rule = rule;
1699 init_tables(struct ip_fw_chain *ch)
1703 for (i = 0; i < IPFW_TABLES_MAX; i++)
1704 rn_inithead((void **)&ch->tables[i], 32);
1708 add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1709 uint8_t mlen, uint32_t value)
1711 struct radix_node_head *rnh;
1712 struct table_entry *ent;
1714 if (tbl >= IPFW_TABLES_MAX)
1716 rnh = ch->tables[tbl];
1717 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1721 ent->addr.sin_len = ent->mask.sin_len = 8;
1722 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1723 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1724 IPFW_WLOCK(&layer3_chain);
1725 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1727 IPFW_WUNLOCK(&layer3_chain);
1728 free(ent, M_IPFW_TBL);
1731 IPFW_WUNLOCK(&layer3_chain);
1736 del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1739 struct radix_node_head *rnh;
1740 struct table_entry *ent;
1741 struct sockaddr_in sa, mask;
1743 if (tbl >= IPFW_TABLES_MAX)
1745 rnh = ch->tables[tbl];
1746 sa.sin_len = mask.sin_len = 8;
1747 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1748 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1750 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1756 free(ent, M_IPFW_TBL);
1761 flush_table_entry(struct radix_node *rn, void *arg)
1763 struct radix_node_head * const rnh = arg;
1764 struct table_entry *ent;
1766 ent = (struct table_entry *)
1767 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1769 free(ent, M_IPFW_TBL);
1774 flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1776 struct radix_node_head *rnh;
1778 IPFW_WLOCK_ASSERT(ch);
1780 if (tbl >= IPFW_TABLES_MAX)
1782 rnh = ch->tables[tbl];
1783 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1788 flush_tables(struct ip_fw_chain *ch)
1792 IPFW_WLOCK_ASSERT(ch);
1794 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1795 flush_table(ch, tbl);
1799 lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1802 struct radix_node_head *rnh;
1803 struct table_entry *ent;
1804 struct sockaddr_in sa;
1806 if (tbl >= IPFW_TABLES_MAX)
1808 rnh = ch->tables[tbl];
1810 sa.sin_addr.s_addr = addr;
1811 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1820 count_table_entry(struct radix_node *rn, void *arg)
1822 u_int32_t * const cnt = arg;
1829 count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1831 struct radix_node_head *rnh;
1833 if (tbl >= IPFW_TABLES_MAX)
1835 rnh = ch->tables[tbl];
1837 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1842 dump_table_entry(struct radix_node *rn, void *arg)
1844 struct table_entry * const n = (struct table_entry *)rn;
1845 ipfw_table * const tbl = arg;
1846 ipfw_table_entry *ent;
1848 if (tbl->cnt == tbl->size)
1850 ent = &tbl->ent[tbl->cnt];
1851 ent->tbl = tbl->tbl;
1852 if (in_nullhost(n->mask.sin_addr))
1855 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1856 ent->addr = n->addr.sin_addr.s_addr;
1857 ent->value = n->value;
1863 dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1865 struct radix_node_head *rnh;
1867 IPFW_WLOCK_ASSERT(ch);
1869 if (tbl->tbl >= IPFW_TABLES_MAX)
1871 rnh = ch->tables[tbl->tbl];
1873 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1878 fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1882 if (inp->inp_socket != NULL) {
1883 cr = inp->inp_socket->so_cred;
1884 ugp->fw_prid = jailed(cr) ?
1885 cr->cr_prison->pr_id : -1;
1886 ugp->fw_uid = cr->cr_uid;
1887 ugp->fw_ngroups = cr->cr_ngroups;
1888 bcopy(cr->cr_groups, ugp->fw_groups,
1889 sizeof(ugp->fw_groups));
1894 check_uidgid(ipfw_insn_u32 *insn,
1895 int proto, struct ifnet *oif,
1896 struct in_addr dst_ip, u_int16_t dst_port,
1897 struct in_addr src_ip, u_int16_t src_port,
1898 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1900 struct inpcbinfo *pi;
1907 * Check to see if the UDP or TCP stack supplied us with
1908 * the PCB. If so, rather then holding a lock and looking
1909 * up the PCB, we can use the one that was supplied.
1911 if (inp && *lookup == 0) {
1912 INP_LOCK_ASSERT(inp);
1913 if (inp->inp_socket != NULL) {
1914 fill_ugid_cache(inp, ugp);
1919 * If we have already been here and the packet has no
1920 * PCB entry associated with it, then we can safely
1921 * assume that this is a no match.
1925 if (proto == IPPROTO_TCP) {
1928 } else if (proto == IPPROTO_UDP) {
1937 in_pcblookup_hash(pi,
1938 dst_ip, htons(dst_port),
1939 src_ip, htons(src_port),
1941 in_pcblookup_hash(pi,
1942 src_ip, htons(src_port),
1943 dst_ip, htons(dst_port),
1947 if (pcb->inp_socket != NULL) {
1948 fill_ugid_cache(pcb, ugp);
1953 INP_INFO_RUNLOCK(pi);
1956 * If the lookup did not yield any results, there
1957 * is no sense in coming back and trying again. So
1958 * we can set lookup to -1 and ensure that we wont
1959 * bother the pcb system again.
1965 if (insn->o.opcode == O_UID)
1966 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1967 else if (insn->o.opcode == O_GID) {
1968 for (gp = ugp->fw_groups;
1969 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1970 if (*gp == (gid_t)insn->d[0]) {
1974 } else if (insn->o.opcode == O_JAIL)
1975 match = (ugp->fw_prid == (int)insn->d[0]);
1980 * The main check routine for the firewall.
1982 * All arguments are in args so we can modify them and return them
1983 * back to the caller.
1987 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1988 * Starts with the IP header.
1989 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1990 * args->oif Outgoing interface, or NULL if packet is incoming.
1991 * The incoming interface is in the mbuf. (in)
1992 * args->divert_rule (in/out)
1993 * Skip up to the first rule past this rule number;
1994 * upon return, non-zero port number for divert or tee.
1996 * args->rule Pointer to the last matching rule (in/out)
1997 * args->next_hop Socket we are forwarding to (out).
1998 * args->f_id Addresses grabbed from the packet (out)
1999 * args->cookie a cookie depending on rule action
2003 * IP_FW_PASS the packet must be accepted
2004 * IP_FW_DENY the packet must be dropped
2005 * IP_FW_DIVERT divert packet, port in m_tag
2006 * IP_FW_TEE tee packet, port in m_tag
2007 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2008 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2013 ipfw_chk(struct ip_fw_args *args)
2016 * Local variables hold state during the processing of a packet.
2018 * IMPORTANT NOTE: to speed up the processing of rules, there
2019 * are some assumption on the values of the variables, which
2020 * are documented here. Should you change them, please check
2021 * the implementation of the various instructions to make sure
2022 * that they still work.
2024 * args->eh The MAC header. It is non-null for a layer2
2025 * packet, it is NULL for a layer-3 packet.
2027 * m | args->m Pointer to the mbuf, as received from the caller.
2028 * It may change if ipfw_chk() does an m_pullup, or if it
2029 * consumes the packet because it calls send_reject().
2030 * XXX This has to change, so that ipfw_chk() never modifies
2031 * or consumes the buffer.
2032 * ip is simply an alias of the value of m, and it is kept
2033 * in sync with it (the packet is supposed to start with
2036 struct mbuf *m = args->m;
2037 struct ip *ip = mtod(m, struct ip *);
2040 * For rules which contain uid/gid or jail constraints, cache
2041 * a copy of the users credentials after the pcb lookup has been
2042 * executed. This will speed up the processing of rules with
2043 * these types of constraints, as well as decrease contention
2044 * on pcb related locks.
2046 struct ip_fw_ugid fw_ugid_cache;
2047 int ugid_lookup = 0;
2050 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2051 * associated with a packet input on a divert socket. This
2052 * will allow to distinguish traffic and its direction when
2053 * it originates from a divert socket.
2055 u_int divinput_flags = 0;
2058 * oif | args->oif If NULL, ipfw_chk has been called on the
2059 * inbound path (ether_input, ip_input).
2060 * If non-NULL, ipfw_chk has been called on the outbound path
2061 * (ether_output, ip_output).
2063 struct ifnet *oif = args->oif;
2065 struct ip_fw *f = NULL; /* matching rule */
2069 * hlen The length of the IP header.
2071 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2074 * offset The offset of a fragment. offset != 0 means that
2075 * we have a fragment at this offset of an IPv4 packet.
2076 * offset == 0 means that (if this is an IPv4 packet)
2077 * this is the first or only fragment.
2078 * For IPv6 offset == 0 means there is no Fragment Header.
2079 * If offset != 0 for IPv6 always use correct mask to
2080 * get the correct offset because we add IP6F_MORE_FRAG
2081 * to be able to dectect the first fragment which would
2082 * otherwise have offset = 0.
2087 * Local copies of addresses. They are only valid if we have
2090 * proto The protocol. Set to 0 for non-ip packets,
2091 * or to the protocol read from the packet otherwise.
2092 * proto != 0 means that we have an IPv4 packet.
2094 * src_port, dst_port port numbers, in HOST format. Only
2095 * valid for TCP and UDP packets.
2097 * src_ip, dst_ip ip addresses, in NETWORK format.
2098 * Only valid for IPv4 packets.
2101 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2102 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2107 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2108 * MATCH_NONE when checked and not matched (q = NULL),
2109 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2111 int dyn_dir = MATCH_UNKNOWN;
2112 ipfw_dyn_rule *q = NULL;
2113 struct ip_fw_chain *chain = &layer3_chain;
2117 * We store in ulp a pointer to the upper layer protocol header.
2118 * In the ipv4 case this is easy to determine from the header,
2119 * but for ipv6 we might have some additional headers in the middle.
2120 * ulp is NULL if not found.
2122 void *ulp = NULL; /* upper layer protocol pointer. */
2123 /* XXX ipv6 variables */
2125 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2126 /* end of ipv6 variables */
2129 if (m->m_flags & M_SKIP_FIREWALL)
2130 return (IP_FW_PASS); /* accept */
2132 pktlen = m->m_pkthdr.len;
2133 proto = args->f_id.proto = 0; /* mark f_id invalid */
2134 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2137 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2138 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2139 * pointer might become stale after other pullups (but we never use it
2142 #define PULLUP_TO(len, p, T) \
2144 int x = (len) + sizeof(T); \
2145 if ((m)->m_len < x) { \
2146 args->m = m = m_pullup(m, x); \
2148 goto pullup_failed; \
2150 p = (mtod(m, char *) + (len)); \
2153 /* Identify IP packets and fill up variables. */
2154 if (pktlen >= sizeof(struct ip6_hdr) &&
2155 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2156 mtod(m, struct ip *)->ip_v == 6) {
2158 args->f_id.addr_type = 6;
2159 hlen = sizeof(struct ip6_hdr);
2160 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2162 /* Search extension headers to find upper layer protocols */
2163 while (ulp == NULL) {
2165 case IPPROTO_ICMPV6:
2166 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2167 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2171 PULLUP_TO(hlen, ulp, struct tcphdr);
2172 dst_port = TCP(ulp)->th_dport;
2173 src_port = TCP(ulp)->th_sport;
2174 args->f_id.flags = TCP(ulp)->th_flags;
2178 PULLUP_TO(hlen, ulp, struct udphdr);
2179 dst_port = UDP(ulp)->uh_dport;
2180 src_port = UDP(ulp)->uh_sport;
2183 case IPPROTO_HOPOPTS: /* RFC 2460 */
2184 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2185 ext_hd |= EXT_HOPOPTS;
2186 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2187 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2191 case IPPROTO_ROUTING: /* RFC 2460 */
2192 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2193 if (((struct ip6_rthdr *)ulp)->ip6r_type != 0) {
2194 printf("IPFW2: IPV6 - Unknown Routing "
2195 "Header type(%d)\n",
2196 ((struct ip6_rthdr *)ulp)->ip6r_type);
2197 if (fw_deny_unknown_exthdrs)
2198 return (IP_FW_DENY);
2201 ext_hd |= EXT_ROUTING;
2202 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2203 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2207 case IPPROTO_FRAGMENT: /* RFC 2460 */
2208 PULLUP_TO(hlen, ulp, struct ip6_frag);
2209 ext_hd |= EXT_FRAGMENT;
2210 hlen += sizeof (struct ip6_frag);
2211 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2212 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2214 /* Add IP6F_MORE_FRAG for offset of first
2215 * fragment to be != 0. */
2216 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2219 printf("IPFW2: IPV6 - Invalid Fragment "
2221 if (fw_deny_unknown_exthdrs)
2222 return (IP_FW_DENY);
2225 args->f_id.frag_id6 =
2226 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2230 case IPPROTO_DSTOPTS: /* RFC 2460 */
2231 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2232 ext_hd |= EXT_DSTOPTS;
2233 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2234 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2238 case IPPROTO_AH: /* RFC 2402 */
2239 PULLUP_TO(hlen, ulp, struct ip6_ext);
2241 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2242 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2246 case IPPROTO_ESP: /* RFC 2406 */
2247 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2248 /* Anything past Seq# is variable length and
2249 * data past this ext. header is encrypted. */
2253 case IPPROTO_NONE: /* RFC 2460 */
2254 PULLUP_TO(hlen, ulp, struct ip6_ext);
2255 /* Packet ends here. if ip6e_len!=0 octets
2256 * must be ignored. */
2259 case IPPROTO_OSPFIGP:
2260 /* XXX OSPF header check? */
2261 PULLUP_TO(hlen, ulp, struct ip6_ext);
2265 printf("IPFW2: IPV6 - Unknown Extension "
2266 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2267 if (fw_deny_unknown_exthdrs)
2268 return (IP_FW_DENY);
2272 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2273 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2274 args->f_id.src_ip = 0;
2275 args->f_id.dst_ip = 0;
2276 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2277 } else if (pktlen >= sizeof(struct ip) &&
2278 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2279 mtod(m, struct ip *)->ip_v == 4) {
2281 ip = mtod(m, struct ip *);
2282 hlen = ip->ip_hl << 2;
2283 args->f_id.addr_type = 4;
2286 * Collect parameters into local variables for faster matching.
2289 src_ip = ip->ip_src;
2290 dst_ip = ip->ip_dst;
2291 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2292 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2293 ip_len = ntohs(ip->ip_len);
2295 offset = ip->ip_off & IP_OFFMASK;
2296 ip_len = ip->ip_len;
2298 pktlen = ip_len < pktlen ? ip_len : pktlen;
2303 PULLUP_TO(hlen, ulp, struct tcphdr);
2304 dst_port = TCP(ulp)->th_dport;
2305 src_port = TCP(ulp)->th_sport;
2306 args->f_id.flags = TCP(ulp)->th_flags;
2310 PULLUP_TO(hlen, ulp, struct udphdr);
2311 dst_port = UDP(ulp)->uh_dport;
2312 src_port = UDP(ulp)->uh_sport;
2316 PULLUP_TO(hlen, ulp, struct icmphdr);
2317 args->f_id.flags = ICMP(ulp)->icmp_type;
2325 args->f_id.src_ip = ntohl(src_ip.s_addr);
2326 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2329 if (proto) { /* we may have port numbers, store them */
2330 args->f_id.proto = proto;
2331 args->f_id.src_port = src_port = ntohs(src_port);
2332 args->f_id.dst_port = dst_port = ntohs(dst_port);
2336 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2339 * Packet has already been tagged. Look for the next rule
2340 * to restart processing.
2342 * If fw_one_pass != 0 then just accept it.
2343 * XXX should not happen here, but optimized out in
2347 IPFW_RUNLOCK(chain);
2348 return (IP_FW_PASS);
2351 f = args->rule->next_rule;
2353 f = lookup_next_rule(args->rule);
2356 * Find the starting rule. It can be either the first
2357 * one, or the one after divert_rule if asked so.
2359 int skipto = mtag ? divert_cookie(mtag) : 0;
2362 if (args->eh == NULL && skipto != 0) {
2363 if (skipto >= IPFW_DEFAULT_RULE) {
2364 IPFW_RUNLOCK(chain);
2365 return (IP_FW_DENY); /* invalid */
2367 while (f && f->rulenum <= skipto)
2369 if (f == NULL) { /* drop packet */
2370 IPFW_RUNLOCK(chain);
2371 return (IP_FW_DENY);
2375 /* reset divert rule to avoid confusion later */
2377 divinput_flags = divert_info(mtag) &
2378 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2379 m_tag_delete(m, mtag);
2383 * Now scan the rules, and parse microinstructions for each rule.
2385 for (; f; f = f->next) {
2388 int skip_or; /* skip rest of OR block */
2391 if (set_disable & (1 << f->set) )
2395 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
2396 l -= cmdlen, cmd += cmdlen) {
2400 * check_body is a jump target used when we find a
2401 * CHECK_STATE, and need to jump to the body of
2406 cmdlen = F_LEN(cmd);
2408 * An OR block (insn_1 || .. || insn_n) has the
2409 * F_OR bit set in all but the last instruction.
2410 * The first match will set "skip_or", and cause
2411 * the following instructions to be skipped until
2412 * past the one with the F_OR bit clear.
2414 if (skip_or) { /* skip this instruction */
2415 if ((cmd->len & F_OR) == 0)
2416 skip_or = 0; /* next one is good */
2419 match = 0; /* set to 1 if we succeed */
2421 switch (cmd->opcode) {
2423 * The first set of opcodes compares the packet's
2424 * fields with some pattern, setting 'match' if a
2425 * match is found. At the end of the loop there is
2426 * logic to deal with F_NOT and F_OR flags associated
2434 printf("ipfw: opcode %d unimplemented\n",
2442 * We only check offset == 0 && proto != 0,
2443 * as this ensures that we have a
2444 * packet with the ports info.
2448 if (is_ipv6) /* XXX to be fixed later */
2450 if (proto == IPPROTO_TCP ||
2451 proto == IPPROTO_UDP)
2452 match = check_uidgid(
2453 (ipfw_insn_u32 *)cmd,
2456 src_ip, src_port, &fw_ugid_cache,
2457 &ugid_lookup, args->inp);
2461 match = iface_match(m->m_pkthdr.rcvif,
2462 (ipfw_insn_if *)cmd);
2466 match = iface_match(oif, (ipfw_insn_if *)cmd);
2470 match = iface_match(oif ? oif :
2471 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
2475 if (args->eh != NULL) { /* have MAC header */
2476 u_int32_t *want = (u_int32_t *)
2477 ((ipfw_insn_mac *)cmd)->addr;
2478 u_int32_t *mask = (u_int32_t *)
2479 ((ipfw_insn_mac *)cmd)->mask;
2480 u_int32_t *hdr = (u_int32_t *)args->eh;
2483 ( want[0] == (hdr[0] & mask[0]) &&
2484 want[1] == (hdr[1] & mask[1]) &&
2485 want[2] == (hdr[2] & mask[2]) );
2490 if (args->eh != NULL) {
2492 ntohs(args->eh->ether_type);
2494 ((ipfw_insn_u16 *)cmd)->ports;
2497 for (i = cmdlen - 1; !match && i>0;
2499 match = (t>=p[0] && t<=p[1]);
2504 match = (offset != 0);
2507 case O_IN: /* "out" is "not in" */
2508 match = (oif == NULL);
2512 match = (args->eh != NULL);
2516 match = (cmd->arg1 & 1 && divinput_flags &
2517 IP_FW_DIVERT_LOOPBACK_FLAG) ||
2518 (cmd->arg1 & 2 && divinput_flags &
2519 IP_FW_DIVERT_OUTPUT_FLAG);
2524 * We do not allow an arg of 0 so the
2525 * check of "proto" only suffices.
2527 match = (proto == cmd->arg1);
2532 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2536 case O_IP_SRC_LOOKUP:
2537 case O_IP_DST_LOOKUP:
2540 (cmd->opcode == O_IP_DST_LOOKUP) ?
2541 dst_ip.s_addr : src_ip.s_addr;
2544 match = lookup_table(chain, cmd->arg1, a,
2548 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2550 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2558 (cmd->opcode == O_IP_DST_MASK) ?
2559 dst_ip.s_addr : src_ip.s_addr;
2560 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2563 for (; !match && i>0; i-= 2, p+= 2)
2564 match = (p[0] == (a & p[1]));
2572 INADDR_TO_IFP(src_ip, tif);
2573 match = (tif != NULL);
2580 u_int32_t *d = (u_int32_t *)(cmd+1);
2582 cmd->opcode == O_IP_DST_SET ?
2588 addr -= d[0]; /* subtract base */
2589 match = (addr < cmd->arg1) &&
2590 ( d[ 1 + (addr>>5)] &
2591 (1<<(addr & 0x1f)) );
2597 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2605 INADDR_TO_IFP(dst_ip, tif);
2606 match = (tif != NULL);
2613 * offset == 0 && proto != 0 is enough
2614 * to guarantee that we have a
2615 * packet with port info.
2617 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2620 (cmd->opcode == O_IP_SRCPORT) ?
2621 src_port : dst_port ;
2623 ((ipfw_insn_u16 *)cmd)->ports;
2626 for (i = cmdlen - 1; !match && i>0;
2628 match = (x>=p[0] && x<=p[1]);
2633 match = (offset == 0 && proto==IPPROTO_ICMP &&
2634 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
2639 match = is_ipv6 && offset == 0 &&
2640 proto==IPPROTO_ICMPV6 &&
2642 ICMP6(ulp)->icmp6_type,
2643 (ipfw_insn_u32 *)cmd);
2649 ipopts_match(mtod(m, struct ip *), cmd) );
2654 cmd->arg1 == mtod(m, struct ip *)->ip_v);
2660 if (is_ipv4) { /* only for IP packets */
2665 if (cmd->opcode == O_IPLEN)
2667 else if (cmd->opcode == O_IPTTL)
2668 x = mtod(m, struct ip *)->ip_ttl;
2669 else /* must be IPID */
2670 x = ntohs(mtod(m, struct ip *)->ip_id);
2672 match = (cmd->arg1 == x);
2675 /* otherwise we have ranges */
2676 p = ((ipfw_insn_u16 *)cmd)->ports;
2678 for (; !match && i>0; i--, p += 2)
2679 match = (x >= p[0] && x <= p[1]);
2683 case O_IPPRECEDENCE:
2685 (cmd->arg1 == (mtod(m, struct ip *)->ip_tos & 0xe0)) );
2690 flags_match(cmd, mtod(m, struct ip *)->ip_tos));
2694 if (proto == IPPROTO_TCP && offset == 0) {
2702 ((ip->ip_hl + tcp->th_off) << 2);
2704 match = (cmd->arg1 == x);
2707 /* otherwise we have ranges */
2708 p = ((ipfw_insn_u16 *)cmd)->ports;
2710 for (; !match && i>0; i--, p += 2)
2711 match = (x >= p[0] && x <= p[1]);
2716 match = (proto == IPPROTO_TCP && offset == 0 &&
2717 flags_match(cmd, TCP(ulp)->th_flags));
2721 match = (proto == IPPROTO_TCP && offset == 0 &&
2722 tcpopts_match(TCP(ulp), cmd));
2726 match = (proto == IPPROTO_TCP && offset == 0 &&
2727 ((ipfw_insn_u32 *)cmd)->d[0] ==
2732 match = (proto == IPPROTO_TCP && offset == 0 &&
2733 ((ipfw_insn_u32 *)cmd)->d[0] ==
2738 match = (proto == IPPROTO_TCP && offset == 0 &&
2739 cmd->arg1 == TCP(ulp)->th_win);
2743 /* reject packets which have SYN only */
2744 /* XXX should i also check for TH_ACK ? */
2745 match = (proto == IPPROTO_TCP && offset == 0 &&
2746 (TCP(ulp)->th_flags &
2747 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2751 struct altq_tag *at;
2752 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
2755 mtag = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
2758 mtag = m_tag_get(PACKET_TAG_PF_QID,
2759 sizeof(struct altq_tag),
2763 * Let the packet fall back to the
2768 at = (struct altq_tag *)(mtag+1);
2769 at->qid = altq->qid;
2775 m_tag_prepend(m, mtag);
2781 ipfw_log(f, hlen, args, m, oif, offset);
2786 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2790 /* Outgoing packets automatically pass/match */
2791 match = ((oif != NULL) ||
2792 (m->m_pkthdr.rcvif == NULL) ||
2796 verify_path6(&(args->f_id.src_ip6),
2797 m->m_pkthdr.rcvif) :
2799 verify_path(src_ip, m->m_pkthdr.rcvif)));
2803 /* Outgoing packets automatically pass/match */
2804 match = (hlen > 0 && ((oif != NULL) ||
2807 verify_path6(&(args->f_id.src_ip6),
2810 verify_path(src_ip, NULL)));
2814 /* Outgoing packets automatically pass/match */
2815 if (oif == NULL && hlen > 0 &&
2816 ( (is_ipv4 && in_localaddr(src_ip))
2819 in6_localaddr(&(args->f_id.src_ip6)))
2824 is_ipv6 ? verify_path6(
2825 &(args->f_id.src_ip6),
2826 m->m_pkthdr.rcvif) :
2836 match = (m_tag_find(m,
2837 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2840 match = (ipsec_getnhist(m) != 0);
2842 /* otherwise no match */
2848 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
2849 &((ipfw_insn_ip6 *)cmd)->addr6);
2854 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
2855 &((ipfw_insn_ip6 *)cmd)->addr6);
2857 case O_IP6_SRC_MASK:
2859 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2860 struct in6_addr p = args->f_id.src_ip6;
2862 APPLY_MASK(&p, &te->mask6);
2863 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2867 case O_IP6_DST_MASK:
2869 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2870 struct in6_addr p = args->f_id.dst_ip6;
2872 APPLY_MASK(&p, &te->mask6);
2873 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2878 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
2882 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
2887 flow6id_match(args->f_id.flow_id6,
2888 (ipfw_insn_u32 *) cmd);
2893 (ext_hd & ((ipfw_insn *) cmd)->arg1);
2906 * The second set of opcodes represents 'actions',
2907 * i.e. the terminal part of a rule once the packet
2908 * matches all previous patterns.
2909 * Typically there is only one action for each rule,
2910 * and the opcode is stored at the end of the rule
2911 * (but there are exceptions -- see below).
2913 * In general, here we set retval and terminate the
2914 * outer loop (would be a 'break 3' in some language,
2915 * but we need to do a 'goto done').
2918 * O_COUNT and O_SKIPTO actions:
2919 * instead of terminating, we jump to the next rule
2920 * ('goto next_rule', equivalent to a 'break 2'),
2921 * or to the SKIPTO target ('goto again' after
2922 * having set f, cmd and l), respectively.
2924 * O_LOG and O_ALTQ action parameters:
2925 * perform some action and set match = 1;
2927 * O_LIMIT and O_KEEP_STATE: these opcodes are
2928 * not real 'actions', and are stored right
2929 * before the 'action' part of the rule.
2930 * These opcodes try to install an entry in the
2931 * state tables; if successful, we continue with
2932 * the next opcode (match=1; break;), otherwise
2933 * the packet * must be dropped
2934 * ('goto done' after setting retval);
2936 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2937 * cause a lookup of the state table, and a jump
2938 * to the 'action' part of the parent rule
2939 * ('goto check_body') if an entry is found, or
2940 * (CHECK_STATE only) a jump to the next rule if
2941 * the entry is not found ('goto next_rule').
2942 * The result of the lookup is cached to make
2943 * further instances of these opcodes are
2948 if (install_state(f,
2949 (ipfw_insn_limit *)cmd, args)) {
2950 retval = IP_FW_DENY;
2951 goto done; /* error/limit violation */
2959 * dynamic rules are checked at the first
2960 * keep-state or check-state occurrence,
2961 * with the result being stored in dyn_dir.
2962 * The compiler introduces a PROBE_STATE
2963 * instruction for us when we have a
2964 * KEEP_STATE (because PROBE_STATE needs
2967 if (dyn_dir == MATCH_UNKNOWN &&
2968 (q = lookup_dyn_rule(&args->f_id,
2969 &dyn_dir, proto == IPPROTO_TCP ?
2973 * Found dynamic entry, update stats
2974 * and jump to the 'action' part of
2980 cmd = ACTION_PTR(f);
2981 l = f->cmd_len - f->act_ofs;
2986 * Dynamic entry not found. If CHECK_STATE,
2987 * skip to next rule, if PROBE_STATE just
2988 * ignore and continue with next opcode.
2990 if (cmd->opcode == O_CHECK_STATE)
2996 retval = 0; /* accept */
3001 args->rule = f; /* report matching rule */
3002 args->cookie = cmd->arg1;
3003 retval = IP_FW_DUMMYNET;
3008 struct divert_tag *dt;
3010 if (args->eh) /* not on layer 2 */
3012 mtag = m_tag_get(PACKET_TAG_DIVERT,
3013 sizeof(struct divert_tag),
3018 IPFW_RUNLOCK(chain);
3019 return (IP_FW_DENY);
3021 dt = (struct divert_tag *)(mtag+1);
3022 dt->cookie = f->rulenum;
3023 dt->info = cmd->arg1;
3024 m_tag_prepend(m, mtag);
3025 retval = (cmd->opcode == O_DIVERT) ?
3026 IP_FW_DIVERT : IP_FW_TEE;
3032 f->pcnt++; /* update stats */
3034 f->timestamp = time_uptime;
3035 if (cmd->opcode == O_COUNT)
3038 if (f->next_rule == NULL)
3039 lookup_next_rule(f);
3045 * Drop the packet and send a reject notice
3046 * if the packet is not ICMP (or is an ICMP
3047 * query), and it is not multicast/broadcast.
3049 if (hlen > 0 && is_ipv4 &&
3050 (proto != IPPROTO_ICMP ||
3051 is_icmp_query(ICMP(ulp))) &&
3052 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3053 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
3054 send_reject(args, cmd->arg1,
3061 if (hlen > 0 && is_ipv6 &&
3062 (proto != IPPROTO_ICMPV6 ||
3063 (is_icmp6_query(args->f_id.flags) == 1)) &&
3064 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3065 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
3066 send_reject6(args, cmd->arg1,
3073 retval = IP_FW_DENY;
3077 if (args->eh) /* not valid on layer2 pkts */
3079 if (!q || dyn_dir == MATCH_FORWARD)
3081 &((ipfw_insn_sa *)cmd)->sa;
3082 retval = IP_FW_PASS;
3087 args->rule = f; /* report matching rule */
3088 args->cookie = cmd->arg1;
3089 retval = (cmd->opcode == O_NETGRAPH) ?
3090 IP_FW_NETGRAPH : IP_FW_NGTEE;
3094 panic("-- unknown opcode %d\n", cmd->opcode);
3095 } /* end of switch() on opcodes */
3097 if (cmd->len & F_NOT)
3101 if (cmd->len & F_OR)
3104 if (!(cmd->len & F_OR)) /* not an OR block, */
3105 break; /* try next rule */
3108 } /* end of inner for, scan opcodes */
3110 next_rule:; /* try next rule */
3112 } /* end of outer for, scan rules */
3113 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3114 IPFW_RUNLOCK(chain);
3115 return (IP_FW_DENY);
3118 /* Update statistics */
3121 f->timestamp = time_uptime;
3122 IPFW_RUNLOCK(chain);
3127 printf("ipfw: pullup failed\n");
3128 return (IP_FW_DENY);
3132 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3133 * These will be reconstructed on the fly as packets are matched.
3136 flush_rule_ptrs(struct ip_fw_chain *chain)
3140 IPFW_WLOCK_ASSERT(chain);
3142 for (rule = chain->rules; rule; rule = rule->next)
3143 rule->next_rule = NULL;
3147 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3148 * possibly create a rule number and add the rule to the list.
3149 * Update the rule_number in the input struct so the caller knows it as well.
3152 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3154 struct ip_fw *rule, *f, *prev;
3155 int l = RULESIZE(input_rule);
3157 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3160 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3164 bcopy(input_rule, rule, l);
3167 rule->next_rule = NULL;
3171 rule->timestamp = 0;
3175 if (chain->rules == NULL) { /* default rule */
3176 chain->rules = rule;
3181 * If rulenum is 0, find highest numbered rule before the
3182 * default rule, and add autoinc_step
3184 if (autoinc_step < 1)
3186 else if (autoinc_step > 1000)
3187 autoinc_step = 1000;
3188 if (rule->rulenum == 0) {
3190 * locate the highest numbered rule before default
3192 for (f = chain->rules; f; f = f->next) {
3193 if (f->rulenum == IPFW_DEFAULT_RULE)
3195 rule->rulenum = f->rulenum;
3197 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3198 rule->rulenum += autoinc_step;
3199 input_rule->rulenum = rule->rulenum;
3203 * Now insert the new rule in the right place in the sorted list.
3205 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3206 if (f->rulenum > rule->rulenum) { /* found the location */
3210 } else { /* head insert */
3211 rule->next = chain->rules;
3212 chain->rules = rule;
3217 flush_rule_ptrs(chain);
3221 IPFW_WUNLOCK(chain);
3222 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3223 rule->rulenum, static_count);)
3228 * Remove a static rule (including derived * dynamic rules)
3229 * and place it on the ``reap list'' for later reclamation.
3230 * The caller is in charge of clearing rule pointers to avoid
3231 * dangling pointers.
3232 * @return a pointer to the next entry.
3233 * Arguments are not checked, so they better be correct.
3235 static struct ip_fw *
3236 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3239 int l = RULESIZE(rule);
3241 IPFW_WLOCK_ASSERT(chain);
3245 remove_dyn_rule(rule, NULL /* force removal */);
3254 rule->next = chain->reap;
3261 * Reclaim storage associated with a list of rules. This is
3262 * typically the list created using remove_rule.
3265 reap_rules(struct ip_fw *head)
3269 while ((rule = head) != NULL) {
3271 if (DUMMYNET_LOADED)
3272 ip_dn_ruledel_ptr(rule);
3278 * Remove all rules from a chain (except rules in set RESVD_SET
3279 * unless kill_default = 1). The caller is responsible for
3280 * reclaiming storage for the rules left in chain->reap.
3283 free_chain(struct ip_fw_chain *chain, int kill_default)
3285 struct ip_fw *prev, *rule;
3287 IPFW_WLOCK_ASSERT(chain);
3289 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3290 for (prev = NULL, rule = chain->rules; rule ; )
3291 if (kill_default || rule->set != RESVD_SET)
3292 rule = remove_rule(chain, rule, prev);
3300 * Remove all rules with given number, and also do set manipulation.
3301 * Assumes chain != NULL && *chain != NULL.
3303 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3304 * the next 8 bits are the new set, the top 8 bits are the command:
3306 * 0 delete rules with given number
3307 * 1 delete rules with given set number
3308 * 2 move rules with given number to new set
3309 * 3 move rules with given set number to new set
3310 * 4 swap sets with given numbers
3313 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3315 struct ip_fw *prev = NULL, *rule;
3316 u_int16_t rulenum; /* rule or old_set */
3317 u_int8_t cmd, new_set;
3319 rulenum = arg & 0xffff;
3320 cmd = (arg >> 24) & 0xff;
3321 new_set = (arg >> 16) & 0xff;
3325 if (new_set > RESVD_SET)
3327 if (cmd == 0 || cmd == 2) {
3328 if (rulenum >= IPFW_DEFAULT_RULE)
3331 if (rulenum > RESVD_SET) /* old_set */
3336 rule = chain->rules;
3339 case 0: /* delete rules with given number */
3341 * locate first rule to delete
3343 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3345 if (rule->rulenum != rulenum) {
3346 IPFW_WUNLOCK(chain);
3351 * flush pointers outside the loop, then delete all matching
3352 * rules. prev remains the same throughout the cycle.
3354 flush_rule_ptrs(chain);
3355 while (rule->rulenum == rulenum)
3356 rule = remove_rule(chain, rule, prev);
3359 case 1: /* delete all rules with given set number */
3360 flush_rule_ptrs(chain);
3361 rule = chain->rules;
3362 while (rule->rulenum < IPFW_DEFAULT_RULE)
3363 if (rule->set == rulenum)
3364 rule = remove_rule(chain, rule, prev);
3371 case 2: /* move rules with given number to new set */
3372 rule = chain->rules;
3373 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3374 if (rule->rulenum == rulenum)
3375 rule->set = new_set;
3378 case 3: /* move rules with given set number to new set */
3379 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3380 if (rule->set == rulenum)
3381 rule->set = new_set;
3384 case 4: /* swap two sets */
3385 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3386 if (rule->set == rulenum)
3387 rule->set = new_set;
3388 else if (rule->set == new_set)
3389 rule->set = rulenum;
3393 * Look for rules to reclaim. We grab the list before
3394 * releasing the lock then reclaim them w/o the lock to
3395 * avoid a LOR with dummynet.
3399 IPFW_WUNLOCK(chain);
3406 * Clear counters for a specific rule.
3407 * The enclosing "table" is assumed locked.
3410 clear_counters(struct ip_fw *rule, int log_only)
3412 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3414 if (log_only == 0) {
3415 rule->bcnt = rule->pcnt = 0;
3416 rule->timestamp = 0;
3418 if (l->o.opcode == O_LOG)
3419 l->log_left = l->max_log;
3423 * Reset some or all counters on firewall rules.
3424 * @arg frwl is null to clear all entries, or contains a specific
3426 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3429 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3437 for (rule = chain->rules; rule; rule = rule->next)
3438 clear_counters(rule, log_only);
3439 msg = log_only ? "ipfw: All logging counts reset.\n" :
3440 "ipfw: Accounting cleared.\n";
3444 * We can have multiple rules with the same number, so we
3445 * need to clear them all.
3447 for (rule = chain->rules; rule; rule = rule->next)
3448 if (rule->rulenum == rulenum) {
3449 while (rule && rule->rulenum == rulenum) {
3450 clear_counters(rule, log_only);
3456 if (!cleared) { /* we did not find any matching rules */
3457 IPFW_WUNLOCK(chain);
3460 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3461 "ipfw: Entry %d cleared.\n";
3463 IPFW_WUNLOCK(chain);
3466 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3471 * Check validity of the structure before insert.
3472 * Fortunately rules are simple, so this mostly need to check rule sizes.
3475 check_ipfw_struct(struct ip_fw *rule, int size)
3481 if (size < sizeof(*rule)) {
3482 printf("ipfw: rule too short\n");
3485 /* first, check for valid size */
3488 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3491 if (rule->act_ofs >= rule->cmd_len) {
3492 printf("ipfw: bogus action offset (%u > %u)\n",
3493 rule->act_ofs, rule->cmd_len - 1);
3497 * Now go for the individual checks. Very simple ones, basically only
3498 * instruction sizes.
3500 for (l = rule->cmd_len, cmd = rule->cmd ;
3501 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3502 cmdlen = F_LEN(cmd);
3504 printf("ipfw: opcode %d size truncated\n",
3508 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3509 switch (cmd->opcode) {
3521 case O_IPPRECEDENCE:
3538 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3551 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3556 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3561 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3564 ((ipfw_insn_log *)cmd)->log_left =
3565 ((ipfw_insn_log *)cmd)->max_log;
3571 /* only odd command lengths */
3572 if ( !(cmdlen & 1) || cmdlen > 31)
3578 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3579 printf("ipfw: invalid set size %d\n",
3583 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3588 case O_IP_SRC_LOOKUP:
3589 case O_IP_DST_LOOKUP:
3590 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3591 printf("ipfw: invalid table number %d\n",
3595 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3596 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3601 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3610 if (cmdlen < 1 || cmdlen > 31)
3616 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3617 if (cmdlen < 2 || cmdlen > 31)
3624 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3629 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3635 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3640 #ifdef IPFIREWALL_FORWARD
3641 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3650 if (ip_divert_ptr == NULL)
3656 if (!NG_IPFW_LOADED)
3660 case O_FORWARD_MAC: /* XXX not implemented yet */
3671 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3675 printf("ipfw: opcode %d, multiple actions"
3682 printf("ipfw: opcode %d, action must be"
3691 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3692 F_INSN_SIZE(ipfw_insn))
3697 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3698 ((ipfw_insn_u32 *)cmd)->o.arg1)
3702 case O_IP6_SRC_MASK:
3703 case O_IP6_DST_MASK:
3704 if ( !(cmdlen & 1) || cmdlen > 127)
3708 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3714 switch (cmd->opcode) {
3724 case O_IP6_SRC_MASK:
3725 case O_IP6_DST_MASK:
3727 printf("ipfw: no IPv6 support in kernel\n");
3728 return EPROTONOSUPPORT;
3731 printf("ipfw: opcode %d, unknown opcode\n",
3737 if (have_action == 0) {
3738 printf("ipfw: missing action\n");
3744 printf("ipfw: opcode %d size %d wrong\n",
3745 cmd->opcode, cmdlen);
3750 * Copy the static and dynamic rules to the supplied buffer
3751 * and return the amount of space actually used.
3754 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3757 char *ep = bp + space;
3761 /* XXX this can take a long time and locking will block packet flow */
3763 for (rule = chain->rules; rule ; rule = rule->next) {
3765 * Verify the entry fits in the buffer in case the
3766 * rules changed between calculating buffer space and
3767 * now. This would be better done using a generation
3768 * number but should suffice for now.
3773 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3774 sizeof(set_disable));
3778 IPFW_RUNLOCK(chain);
3780 ipfw_dyn_rule *p, *last = NULL;
3783 for (i = 0 ; i < curr_dyn_buckets; i++)
3784 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3785 if (bp + sizeof *p <= ep) {
3786 ipfw_dyn_rule *dst =
3787 (ipfw_dyn_rule *)bp;
3788 bcopy(p, dst, sizeof *p);
3789 bcopy(&(p->rule->rulenum), &(dst->rule),
3790 sizeof(p->rule->rulenum));
3792 * store a non-null value in "next".
3793 * The userland code will interpret a
3794 * NULL here as a marker
3795 * for the last dynamic rule.
3797 bcopy(&dst, &dst->next, sizeof(dst));
3800 TIME_LEQ(dst->expire, time_uptime) ?
3801 0 : dst->expire - time_uptime ;
3802 bp += sizeof(ipfw_dyn_rule);
3806 if (last != NULL) /* mark last dynamic rule */
3807 bzero(&last->next, sizeof(last));
3809 return (bp - (char *)buf);
3814 * {set|get}sockopt parser.
3817 ipfw_ctl(struct sockopt *sopt)
3819 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
3820 int error, rule_num;
3822 struct ip_fw *buf, *rule;
3823 u_int32_t rulenum[2];
3825 error = suser(sopt->sopt_td);
3830 * Disallow modifications in really-really secure mode, but still allow
3831 * the logging counters to be reset.
3833 if (sopt->sopt_name == IP_FW_ADD ||
3834 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3835 #if __FreeBSD_version >= 500034
3836 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3839 #else /* FreeBSD 4.x */
3840 if (securelevel >= 3)
3847 switch (sopt->sopt_name) {
3850 * pass up a copy of the current rules. Static rules
3851 * come first (the last of which has number IPFW_DEFAULT_RULE),
3852 * followed by a possibly empty list of dynamic rule.
3853 * The last dynamic rule has NULL in the "next" field.
3855 * Note that the calculated size is used to bound the
3856 * amount of data returned to the user. The rule set may
3857 * change between calculating the size and returning the
3858 * data in which case we'll just return what fits.
3860 size = static_len; /* size of static rules */
3861 if (ipfw_dyn_v) /* add size of dyn.rules */
3862 size += (dyn_count * sizeof(ipfw_dyn_rule));
3865 * XXX todo: if the user passes a short length just to know
3866 * how much room is needed, do not bother filling up the
3867 * buffer, just jump to the sooptcopyout.
3869 buf = malloc(size, M_TEMP, M_WAITOK);
3870 error = sooptcopyout(sopt, buf,
3871 ipfw_getrules(&layer3_chain, buf, size));
3877 * Normally we cannot release the lock on each iteration.
3878 * We could do it here only because we start from the head all
3879 * the times so there is no risk of missing some entries.
3880 * On the other hand, the risk is that we end up with
3881 * a very inconsistent ruleset, so better keep the lock
3882 * around the whole cycle.
3884 * XXX this code can be improved by resetting the head of
3885 * the list to point to the default rule, and then freeing
3886 * the old list without the need for a lock.
3889 IPFW_WLOCK(&layer3_chain);
3890 layer3_chain.reap = NULL;
3891 free_chain(&layer3_chain, 0 /* keep default rule */);
3892 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3893 IPFW_WUNLOCK(&layer3_chain);
3894 if (layer3_chain.reap != NULL)
3899 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3900 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3901 sizeof(struct ip_fw) );
3903 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3905 error = add_rule(&layer3_chain, rule);
3906 size = RULESIZE(rule);
3907 if (!error && sopt->sopt_dir == SOPT_GET)
3908 error = sooptcopyout(sopt, rule, size);
3915 * IP_FW_DEL is used for deleting single rules or sets,
3916 * and (ab)used to atomically manipulate sets. Argument size
3917 * is used to distinguish between the two:
3919 * delete single rule or set of rules,
3920 * or reassign rules (or sets) to a different set.
3921 * 2*sizeof(u_int32_t)
3922 * atomic disable/enable sets.
3923 * first u_int32_t contains sets to be disabled,
3924 * second u_int32_t contains sets to be enabled.
3926 error = sooptcopyin(sopt, rulenum,
3927 2*sizeof(u_int32_t), sizeof(u_int32_t));
3930 size = sopt->sopt_valsize;
3931 if (size == sizeof(u_int32_t)) /* delete or reassign */
3932 error = del_entry(&layer3_chain, rulenum[0]);
3933 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3935 (set_disable | rulenum[0]) & ~rulenum[1] &
3936 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3942 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3944 if (sopt->sopt_val != 0) {
3945 error = sooptcopyin(sopt, &rule_num,
3946 sizeof(int), sizeof(int));
3950 error = zero_entry(&layer3_chain, rule_num,
3951 sopt->sopt_name == IP_FW_RESETLOG);
3954 case IP_FW_TABLE_ADD:
3956 ipfw_table_entry ent;
3958 error = sooptcopyin(sopt, &ent,
3959 sizeof(ent), sizeof(ent));
3962 error = add_table_entry(&layer3_chain, ent.tbl,
3963 ent.addr, ent.masklen, ent.value);
3967 case IP_FW_TABLE_DEL:
3969 ipfw_table_entry ent;
3971 error = sooptcopyin(sopt, &ent,
3972 sizeof(ent), sizeof(ent));
3975 error = del_table_entry(&layer3_chain, ent.tbl,
3976 ent.addr, ent.masklen);
3980 case IP_FW_TABLE_FLUSH:
3984 error = sooptcopyin(sopt, &tbl,
3985 sizeof(tbl), sizeof(tbl));
3988 IPFW_WLOCK(&layer3_chain);
3989 error = flush_table(&layer3_chain, tbl);
3990 IPFW_WUNLOCK(&layer3_chain);
3994 case IP_FW_TABLE_GETSIZE:
3998 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
4001 IPFW_RLOCK(&layer3_chain);
4002 if ((error = count_table(&layer3_chain, tbl, &cnt)))
4004 IPFW_RUNLOCK(&layer3_chain);
4005 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4009 case IP_FW_TABLE_LIST:
4013 if (sopt->sopt_valsize < sizeof(*tbl)) {
4017 size = sopt->sopt_valsize;
4018 tbl = malloc(size, M_TEMP, M_WAITOK);
4023 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4028 tbl->size = (size - sizeof(*tbl)) /
4029 sizeof(ipfw_table_entry);
4030 IPFW_WLOCK(&layer3_chain);
4031 error = dump_table(&layer3_chain, tbl);
4033 IPFW_WUNLOCK(&layer3_chain);
4037 IPFW_WUNLOCK(&layer3_chain);
4038 error = sooptcopyout(sopt, tbl, size);
4044 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4053 * dummynet needs a reference to the default rule, because rules can be
4054 * deleted while packets hold a reference to them. When this happens,
4055 * dummynet changes the reference to the default rule (it could well be a
4056 * NULL pointer, but this way we do not need to check for the special
4057 * case, plus here he have info on the default behaviour).
4059 struct ip_fw *ip_fw_default_rule;
4062 * This procedure is only used to handle keepalives. It is invoked
4063 * every dyn_keepalive_period
4066 ipfw_tick(void * __unused unused)
4068 struct mbuf *m0, *m, *mnext, **mtailp;
4072 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4076 * We make a chain of packets to go out here -- not deferring
4077 * until after we drop the IPFW dynamic rule lock would result
4078 * in a lock order reversal with the normal packet input -> ipfw
4084 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4085 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4086 if (q->dyn_type == O_LIMIT_PARENT)
4088 if (q->id.proto != IPPROTO_TCP)
4090 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4092 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4094 continue; /* too early */
4095 if (TIME_LEQ(q->expire, time_uptime))
4096 continue; /* too late, rule expired */
4098 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4099 q->ack_fwd, TH_SYN);
4100 if (*mtailp != NULL)
4101 mtailp = &(*mtailp)->m_nextpkt;
4102 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4104 if (*mtailp != NULL)
4105 mtailp = &(*mtailp)->m_nextpkt;
4109 for (m = mnext = m0; m != NULL; m = mnext) {
4110 mnext = m->m_nextpkt;
4111 m->m_nextpkt = NULL;
4112 ip_output(m, NULL, NULL, 0, NULL, NULL);
4115 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4121 struct ip_fw default_rule;
4125 /* Setup IPv6 fw sysctl tree. */
4126 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4127 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4128 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4129 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4130 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4131 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4132 &fw_deny_unknown_exthdrs, 0,
4133 "Deny packets with unknown IPv6 Extension Headers");
4136 layer3_chain.rules = NULL;
4137 layer3_chain.want_write = 0;
4138 layer3_chain.busy_count = 0;
4139 cv_init(&layer3_chain.cv, "Condition variable for IPFW rw locks");
4140 IPFW_LOCK_INIT(&layer3_chain);
4141 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4142 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4144 IPFW_DYN_LOCK_INIT();
4145 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4147 bzero(&default_rule, sizeof default_rule);
4149 default_rule.act_ofs = 0;
4150 default_rule.rulenum = IPFW_DEFAULT_RULE;
4151 default_rule.cmd_len = 1;
4152 default_rule.set = RESVD_SET;
4154 default_rule.cmd[0].len = 1;
4155 default_rule.cmd[0].opcode =
4156 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4161 error = add_rule(&layer3_chain, &default_rule);
4163 printf("ipfw2: error %u initializing default rule "
4164 "(support disabled)\n", error);
4165 IPFW_DYN_LOCK_DESTROY();
4166 IPFW_LOCK_DESTROY(&layer3_chain);
4170 ip_fw_default_rule = layer3_chain.rules;
4171 printf("ipfw2 (+ipv6) initialized, divert %s, "
4172 "rule-based forwarding "
4173 #ifdef IPFIREWALL_FORWARD
4178 "default to %s, logging ",
4184 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4186 #ifdef IPFIREWALL_VERBOSE
4189 #ifdef IPFIREWALL_VERBOSE_LIMIT
4190 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4192 if (fw_verbose == 0)
4193 printf("disabled\n");
4194 else if (verbose_limit == 0)
4195 printf("unlimited\n");
4197 printf("limited to %d packets/entry by default\n",
4200 init_tables(&layer3_chain);
4201 ip_fw_ctl_ptr = ipfw_ctl;
4202 ip_fw_chk_ptr = ipfw_chk;
4203 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4213 ip_fw_chk_ptr = NULL;
4214 ip_fw_ctl_ptr = NULL;
4215 callout_drain(&ipfw_timeout);
4216 IPFW_WLOCK(&layer3_chain);
4217 flush_tables(&layer3_chain);
4218 layer3_chain.reap = NULL;
4219 free_chain(&layer3_chain, 1 /* kill default rule */);
4220 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4221 IPFW_WUNLOCK(&layer3_chain);
4224 IPFW_DYN_LOCK_DESTROY();
4225 uma_zdestroy(ipfw_dyn_rule_zone);
4226 IPFW_LOCK_DESTROY(&layer3_chain);
4229 /* Free IPv6 fw sysctl tree. */
4230 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4233 printf("IP firewall unloaded\n");