2 * Copyright (c) 2002-2009 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
26 #include <sys/cdefs.h>
27 __FBSDID("$FreeBSD$");
30 * The FreeBSD IP packet firewall, main file
33 #if !defined(KLD_MODULE)
35 #include "opt_ipdivert.h"
39 #error IPFIREWALL requires INET.
42 #include "opt_inet6.h"
43 #include "opt_ipsec.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/condvar.h>
48 #include <sys/eventhandler.h>
49 #include <sys/malloc.h>
51 #include <sys/kernel.h>
54 #include <sys/module.h>
57 #include <sys/rwlock.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/sysctl.h>
61 #include <sys/syslog.h>
62 #include <sys/ucred.h>
63 #include <net/ethernet.h> /* for ETHERTYPE_IP */
65 #include <net/route.h>
66 #include <net/pf_mtag.h>
69 #include <netinet/in.h>
70 #include <netinet/in_var.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_icmp.h>
75 #include <netinet/ip_fw.h>
76 #include <netinet/ipfw/ip_fw_private.h>
77 #include <netinet/ip_carp.h>
78 #include <netinet/pim.h>
79 #include <netinet/tcp_var.h>
80 #include <netinet/udp.h>
81 #include <netinet/udp_var.h>
82 #include <netinet/sctp.h>
84 #include <netinet/ip6.h>
85 #include <netinet/icmp6.h>
87 #include <netinet6/in6_pcb.h>
88 #include <netinet6/scope6_var.h>
89 #include <netinet6/ip6_var.h>
92 #include <machine/in_cksum.h> /* XXX for in_cksum */
95 #include <security/mac/mac_framework.h>
99 * static variables followed by global ones.
100 * All ipfw global variables are here.
103 /* ipfw_vnet_ready controls when we are open for business */
104 static VNET_DEFINE(int, ipfw_vnet_ready) = 0;
105 #define V_ipfw_vnet_ready VNET(ipfw_vnet_ready)
107 static VNET_DEFINE(int, fw_deny_unknown_exthdrs);
108 #define V_fw_deny_unknown_exthdrs VNET(fw_deny_unknown_exthdrs)
110 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
111 static int default_to_accept = 1;
113 static int default_to_accept;
116 VNET_DEFINE(int, autoinc_step);
117 VNET_DEFINE(int, fw_one_pass) = 1;
120 * Each rule belongs to one of 32 different sets (0..31).
121 * The variable set_disable contains one bit per set.
122 * If the bit is set, all rules in the corresponding set
123 * are disabled. Set RESVD_SET(31) is reserved for the default rule
124 * and rules that are not deleted by the flush command,
125 * and CANNOT be disabled.
126 * Rules in set RESVD_SET can only be deleted individually.
128 VNET_DEFINE(u_int32_t, set_disable);
129 #define V_set_disable VNET(set_disable)
131 VNET_DEFINE(int, fw_verbose);
132 /* counter for ipfw_log(NULL...) */
133 VNET_DEFINE(u_int64_t, norule_counter);
134 VNET_DEFINE(int, verbose_limit);
136 /* layer3_chain contains the list of rules for layer 3 */
137 VNET_DEFINE(struct ip_fw_chain, layer3_chain);
139 ipfw_nat_t *ipfw_nat_ptr = NULL;
140 struct cfg_nat *(*lookup_nat_ptr)(struct nat_list *, int);
141 ipfw_nat_cfg_t *ipfw_nat_cfg_ptr;
142 ipfw_nat_cfg_t *ipfw_nat_del_ptr;
143 ipfw_nat_cfg_t *ipfw_nat_get_cfg_ptr;
144 ipfw_nat_cfg_t *ipfw_nat_get_log_ptr;
147 uint32_t dummy_def = IPFW_DEFAULT_RULE;
148 uint32_t dummy_tables_max = IPFW_TABLES_MAX;
152 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
153 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
154 CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_one_pass), 0,
155 "Only do a single pass through ipfw when using dummynet(4)");
156 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step,
157 CTLFLAG_RW, &VNET_NAME(autoinc_step), 0,
158 "Rule number auto-increment step");
159 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, verbose,
160 CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_verbose), 0,
161 "Log matches to ipfw rules");
162 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit,
163 CTLFLAG_RW, &VNET_NAME(verbose_limit), 0,
164 "Set upper limit of matches of ipfw rules logged");
165 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, default_rule, CTLFLAG_RD,
167 "The default/max possible rule number.");
168 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, tables_max, CTLFLAG_RD,
169 &dummy_tables_max, 0,
170 "The maximum number of tables.");
171 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, default_to_accept, CTLFLAG_RDTUN,
172 &default_to_accept, 0,
173 "Make the default rule accept all packets.");
174 TUNABLE_INT("net.inet.ip.fw.default_to_accept", &default_to_accept);
175 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, static_count,
176 CTLFLAG_RD, &VNET_NAME(layer3_chain.n_rules), 0,
177 "Number of static rules");
180 SYSCTL_DECL(_net_inet6_ip6);
181 SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
182 SYSCTL_VNET_INT(_net_inet6_ip6_fw, OID_AUTO, deny_unknown_exthdrs,
183 CTLFLAG_RW | CTLFLAG_SECURE, &VNET_NAME(fw_deny_unknown_exthdrs), 0,
184 "Deny packets with unknown IPv6 Extension Headers");
189 #endif /* SYSCTL_NODE */
193 * Some macros used in the various matching options.
194 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
195 * Other macros just cast void * into the appropriate type
197 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
198 #define TCP(p) ((struct tcphdr *)(p))
199 #define SCTP(p) ((struct sctphdr *)(p))
200 #define UDP(p) ((struct udphdr *)(p))
201 #define ICMP(p) ((struct icmphdr *)(p))
202 #define ICMP6(p) ((struct icmp6_hdr *)(p))
205 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
207 int type = icmp->icmp_type;
209 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
212 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
213 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
216 is_icmp_query(struct icmphdr *icmp)
218 int type = icmp->icmp_type;
220 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
225 * The following checks use two arrays of 8 or 16 bits to store the
226 * bits that we want set or clear, respectively. They are in the
227 * low and high half of cmd->arg1 or cmd->d[0].
229 * We scan options and store the bits we find set. We succeed if
231 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
233 * The code is sometimes optimized not to store additional variables.
237 flags_match(ipfw_insn *cmd, u_int8_t bits)
242 if ( ((cmd->arg1 & 0xff) & bits) != 0)
243 return 0; /* some bits we want set were clear */
244 want_clear = (cmd->arg1 >> 8) & 0xff;
245 if ( (want_clear & bits) != want_clear)
246 return 0; /* some bits we want clear were set */
251 ipopts_match(struct ip *ip, ipfw_insn *cmd)
253 int optlen, bits = 0;
254 u_char *cp = (u_char *)(ip + 1);
255 int x = (ip->ip_hl << 2) - sizeof (struct ip);
257 for (; x > 0; x -= optlen, cp += optlen) {
258 int opt = cp[IPOPT_OPTVAL];
260 if (opt == IPOPT_EOL)
262 if (opt == IPOPT_NOP)
265 optlen = cp[IPOPT_OLEN];
266 if (optlen <= 0 || optlen > x)
267 return 0; /* invalid or truncated */
275 bits |= IP_FW_IPOPT_LSRR;
279 bits |= IP_FW_IPOPT_SSRR;
283 bits |= IP_FW_IPOPT_RR;
287 bits |= IP_FW_IPOPT_TS;
291 return (flags_match(cmd, bits));
295 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
297 int optlen, bits = 0;
298 u_char *cp = (u_char *)(tcp + 1);
299 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
301 for (; x > 0; x -= optlen, cp += optlen) {
303 if (opt == TCPOPT_EOL)
305 if (opt == TCPOPT_NOP)
319 bits |= IP_FW_TCPOPT_MSS;
323 bits |= IP_FW_TCPOPT_WINDOW;
326 case TCPOPT_SACK_PERMITTED:
328 bits |= IP_FW_TCPOPT_SACK;
331 case TCPOPT_TIMESTAMP:
332 bits |= IP_FW_TCPOPT_TS;
337 return (flags_match(cmd, bits));
341 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
343 if (ifp == NULL) /* no iface with this packet, match fails */
345 /* Check by name or by IP address */
346 if (cmd->name[0] != '\0') { /* match by name */
349 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
352 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
356 #ifdef __FreeBSD__ /* and OSX too ? */
360 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
361 if (ia->ifa_addr->sa_family != AF_INET)
363 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
364 (ia->ifa_addr))->sin_addr.s_addr) {
365 if_addr_runlock(ifp);
366 return(1); /* match */
369 if_addr_runlock(ifp);
370 #endif /* __FreeBSD__ */
372 return(0); /* no match, fail ... */
376 * The verify_path function checks if a route to the src exists and
377 * if it is reachable via ifp (when provided).
379 * The 'verrevpath' option checks that the interface that an IP packet
380 * arrives on is the same interface that traffic destined for the
381 * packet's source address would be routed out of.
382 * The 'versrcreach' option just checks that the source address is
383 * reachable via any route (except default) in the routing table.
384 * These two are a measure to block forged packets. This is also
385 * commonly known as "anti-spoofing" or Unicast Reverse Path
386 * Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
387 * is purposely reminiscent of the Cisco IOS command,
389 * ip verify unicast reverse-path
390 * ip verify unicast source reachable-via any
392 * which implements the same functionality. But note that the syntax
393 * is misleading, and the check may be performed on all IP packets
394 * whether unicast, multicast, or broadcast.
397 verify_path(struct in_addr src, struct ifnet *ifp, u_int fib)
403 struct sockaddr_in *dst;
405 bzero(&ro, sizeof(ro));
407 dst = (struct sockaddr_in *)&(ro.ro_dst);
408 dst->sin_family = AF_INET;
409 dst->sin_len = sizeof(*dst);
411 in_rtalloc_ign(&ro, 0, fib);
413 if (ro.ro_rt == NULL)
417 * If ifp is provided, check for equality with rtentry.
418 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
419 * in order to pass packets injected back by if_simloop():
420 * if useloopback == 1 routing entry (via lo0) for our own address
421 * may exist, so we need to handle routing assymetry.
423 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
428 /* if no ifp provided, check if rtentry is not default route */
430 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
435 /* or if this is a blackhole/reject route */
436 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
441 /* found valid route */
444 #endif /* __FreeBSD__ */
449 * ipv6 specific rules here...
452 icmp6type_match (int type, ipfw_insn_u32 *cmd)
454 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
458 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
461 for (i=0; i <= cmd->o.arg1; ++i )
462 if (curr_flow == cmd->d[i] )
467 /* support for IP6_*_ME opcodes */
469 search_ip6_addr_net (struct in6_addr * ip6_addr)
473 struct in6_ifaddr *fdm;
474 struct in6_addr copia;
476 TAILQ_FOREACH(mdc, &V_ifnet, if_link) {
478 TAILQ_FOREACH(mdc2, &mdc->if_addrhead, ifa_link) {
479 if (mdc2->ifa_addr->sa_family == AF_INET6) {
480 fdm = (struct in6_ifaddr *)mdc2;
481 copia = fdm->ia_addr.sin6_addr;
482 /* need for leaving scope_id in the sock_addr */
483 in6_clearscope(&copia);
484 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia)) {
485 if_addr_runlock(mdc);
490 if_addr_runlock(mdc);
496 verify_path6(struct in6_addr *src, struct ifnet *ifp)
499 struct sockaddr_in6 *dst;
501 bzero(&ro, sizeof(ro));
503 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
504 dst->sin6_family = AF_INET6;
505 dst->sin6_len = sizeof(*dst);
506 dst->sin6_addr = *src;
507 /* XXX MRT 0 for ipv6 at this time */
508 rtalloc_ign((struct route *)&ro, 0);
510 if (ro.ro_rt == NULL)
514 * if ifp is provided, check for equality with rtentry
515 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
516 * to support the case of sending packets to an address of our own.
517 * (where the former interface is the first argument of if_simloop()
518 * (=ifp), the latter is lo0)
520 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
525 /* if no ifp provided, check if rtentry is not default route */
527 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
532 /* or if this is a blackhole/reject route */
533 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
538 /* found valid route */
545 is_icmp6_query(int icmp6_type)
547 if ((icmp6_type <= ICMP6_MAXTYPE) &&
548 (icmp6_type == ICMP6_ECHO_REQUEST ||
549 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
550 icmp6_type == ICMP6_WRUREQUEST ||
551 icmp6_type == ICMP6_FQDN_QUERY ||
552 icmp6_type == ICMP6_NI_QUERY))
559 send_reject6(struct ip_fw_args *args, int code, u_int hlen, struct ip6_hdr *ip6)
564 if (code == ICMP6_UNREACH_RST && args->f_id.proto == IPPROTO_TCP) {
566 tcp = (struct tcphdr *)((char *)ip6 + hlen);
568 if ((tcp->th_flags & TH_RST) == 0) {
570 m0 = ipfw_send_pkt(args->m, &(args->f_id),
571 ntohl(tcp->th_seq), ntohl(tcp->th_ack),
572 tcp->th_flags | TH_RST);
574 ip6_output(m0, NULL, NULL, 0, NULL, NULL,
578 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
581 * Unlike above, the mbufs need to line up with the ip6 hdr,
582 * as the contents are read. We need to m_adj() the
584 * The mbuf will however be thrown away so we can adjust it.
585 * Remember we did an m_pullup on it already so we
586 * can make some assumptions about contiguousness.
589 m_adj(m, args->L3offset);
591 icmp6_error(m, ICMP6_DST_UNREACH, code, 0);
602 * sends a reject message, consuming the mbuf passed as an argument.
605 send_reject(struct ip_fw_args *args, int code, int iplen, struct ip *ip)
609 /* XXX When ip is not guaranteed to be at mtod() we will
610 * need to account for this */
611 * The mbuf will however be thrown away so we can adjust it.
612 * Remember we did an m_pullup on it already so we
613 * can make some assumptions about contiguousness.
616 m_adj(m, args->L3offset);
618 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
619 /* We need the IP header in host order for icmp_error(). */
621 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
622 } else if (args->f_id.proto == IPPROTO_TCP) {
623 struct tcphdr *const tcp =
624 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
625 if ( (tcp->th_flags & TH_RST) == 0) {
627 m = ipfw_send_pkt(args->m, &(args->f_id),
628 ntohl(tcp->th_seq), ntohl(tcp->th_ack),
629 tcp->th_flags | TH_RST);
631 ip_output(m, NULL, NULL, 0, NULL, NULL);
640 * Support for uid/gid/jail lookup. These tests are expensive
641 * (because we may need to look into the list of active sockets)
642 * so we cache the results. ugid_lookupp is 0 if we have not
643 * yet done a lookup, 1 if we succeeded, and -1 if we tried
644 * and failed. The function always returns the match value.
645 * We could actually spare the variable and use *uc, setting
646 * it to '(void *)check_uidgid if we have no info, NULL if
647 * we tried and failed, or any other value if successful.
650 check_uidgid(ipfw_insn_u32 *insn, struct ip_fw_args *args, int *ugid_lookupp,
655 return cred_check(insn, proto, oif,
656 dst_ip, dst_port, src_ip, src_port,
657 (struct bsd_ucred *)uc, ugid_lookupp, ((struct mbuf *)inp)->m_skb);
659 struct in_addr src_ip, dst_ip;
660 struct inpcbinfo *pi;
661 struct ipfw_flow_id *id;
662 struct inpcb *pcb, *inp;
672 * Check to see if the UDP or TCP stack supplied us with
673 * the PCB. If so, rather then holding a lock and looking
674 * up the PCB, we can use the one that was supplied.
676 if (inp && *ugid_lookupp == 0) {
677 INP_LOCK_ASSERT(inp);
678 if (inp->inp_socket != NULL) {
679 *uc = crhold(inp->inp_cred);
685 * If we have already been here and the packet has no
686 * PCB entry associated with it, then we can safely
687 * assume that this is a no match.
689 if (*ugid_lookupp == -1)
691 if (id->proto == IPPROTO_TCP) {
694 } else if (id->proto == IPPROTO_UDP) {
695 lookupflags = INPLOOKUP_WILDCARD;
699 lookupflags |= INPLOOKUP_RLOCKPCB;
701 if (*ugid_lookupp == 0) {
702 if (id->addr_type == 6) {
705 pcb = in6_pcblookup_mbuf(pi,
706 &id->src_ip6, htons(id->src_port),
707 &id->dst_ip6, htons(id->dst_port),
708 lookupflags, oif, args->m);
710 pcb = in6_pcblookup_mbuf(pi,
711 &id->dst_ip6, htons(id->dst_port),
712 &id->src_ip6, htons(id->src_port),
713 lookupflags, oif, args->m);
719 src_ip.s_addr = htonl(id->src_ip);
720 dst_ip.s_addr = htonl(id->dst_ip);
722 pcb = in_pcblookup_mbuf(pi,
723 src_ip, htons(id->src_port),
724 dst_ip, htons(id->dst_port),
725 lookupflags, oif, args->m);
727 pcb = in_pcblookup_mbuf(pi,
728 dst_ip, htons(id->dst_port),
729 src_ip, htons(id->src_port),
730 lookupflags, oif, args->m);
733 INP_RLOCK_ASSERT(pcb);
734 *uc = crhold(pcb->inp_cred);
738 if (*ugid_lookupp == 0) {
740 * We tried and failed, set the variable to -1
741 * so we will not try again on this packet.
747 if (insn->o.opcode == O_UID)
748 match = ((*uc)->cr_uid == (uid_t)insn->d[0]);
749 else if (insn->o.opcode == O_GID)
750 match = groupmember((gid_t)insn->d[0], *uc);
751 else if (insn->o.opcode == O_JAIL)
752 match = ((*uc)->cr_prison->pr_id == (int)insn->d[0]);
754 #endif /* __FreeBSD__ */
758 * Helper function to set args with info on the rule after the matching
759 * one. slot is precise, whereas we guess rule_id as they are
760 * assigned sequentially.
763 set_match(struct ip_fw_args *args, int slot,
764 struct ip_fw_chain *chain)
766 args->rule.chain_id = chain->id;
767 args->rule.slot = slot + 1; /* we use 0 as a marker */
768 args->rule.rule_id = 1 + chain->map[slot]->id;
769 args->rule.rulenum = chain->map[slot]->rulenum;
773 * The main check routine for the firewall.
775 * All arguments are in args so we can modify them and return them
776 * back to the caller.
780 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
781 * Starts with the IP header.
782 * args->eh (in) Mac header if present, NULL for layer3 packet.
783 * args->L3offset Number of bytes bypassed if we came from L2.
784 * e.g. often sizeof(eh) ** NOTYET **
785 * args->oif Outgoing interface, NULL if packet is incoming.
786 * The incoming interface is in the mbuf. (in)
787 * args->divert_rule (in/out)
788 * Skip up to the first rule past this rule number;
789 * upon return, non-zero port number for divert or tee.
791 * args->rule Pointer to the last matching rule (in/out)
792 * args->next_hop Socket we are forwarding to (out).
793 * args->f_id Addresses grabbed from the packet (out)
794 * args->rule.info a cookie depending on rule action
798 * IP_FW_PASS the packet must be accepted
799 * IP_FW_DENY the packet must be dropped
800 * IP_FW_DIVERT divert packet, port in m_tag
801 * IP_FW_TEE tee packet, port in m_tag
802 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
803 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
804 * args->rule contains the matching rule,
805 * args->rule.info has additional information.
809 ipfw_chk(struct ip_fw_args *args)
813 * Local variables holding state while processing a packet:
815 * IMPORTANT NOTE: to speed up the processing of rules, there
816 * are some assumption on the values of the variables, which
817 * are documented here. Should you change them, please check
818 * the implementation of the various instructions to make sure
819 * that they still work.
821 * args->eh The MAC header. It is non-null for a layer2
822 * packet, it is NULL for a layer-3 packet.
824 * args->L3offset Offset in the packet to the L3 (IP or equiv.) header.
826 * m | args->m Pointer to the mbuf, as received from the caller.
827 * It may change if ipfw_chk() does an m_pullup, or if it
828 * consumes the packet because it calls send_reject().
829 * XXX This has to change, so that ipfw_chk() never modifies
830 * or consumes the buffer.
831 * ip is the beginning of the ip(4 or 6) header.
832 * Calculated by adding the L3offset to the start of data.
833 * (Until we start using L3offset, the packet is
834 * supposed to start with the ip header).
836 struct mbuf *m = args->m;
837 struct ip *ip = mtod(m, struct ip *);
840 * For rules which contain uid/gid or jail constraints, cache
841 * a copy of the users credentials after the pcb lookup has been
842 * executed. This will speed up the processing of rules with
843 * these types of constraints, as well as decrease contention
844 * on pcb related locks.
847 struct bsd_ucred ucred_cache;
849 struct ucred *ucred_cache = NULL;
851 int ucred_lookup = 0;
854 * oif | args->oif If NULL, ipfw_chk has been called on the
855 * inbound path (ether_input, ip_input).
856 * If non-NULL, ipfw_chk has been called on the outbound path
857 * (ether_output, ip_output).
859 struct ifnet *oif = args->oif;
861 int f_pos = 0; /* index of current rule in the array */
865 * hlen The length of the IP header.
867 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
870 * offset The offset of a fragment. offset != 0 means that
871 * we have a fragment at this offset of an IPv4 packet.
872 * offset == 0 means that (if this is an IPv4 packet)
873 * this is the first or only fragment.
874 * For IPv6 offset == 0 means there is no Fragment Header.
875 * If offset != 0 for IPv6 always use correct mask to
876 * get the correct offset because we add IP6F_MORE_FRAG
877 * to be able to dectect the first fragment which would
878 * otherwise have offset = 0.
883 * Local copies of addresses. They are only valid if we have
886 * proto The protocol. Set to 0 for non-ip packets,
887 * or to the protocol read from the packet otherwise.
888 * proto != 0 means that we have an IPv4 packet.
890 * src_port, dst_port port numbers, in HOST format. Only
891 * valid for TCP and UDP packets.
893 * src_ip, dst_ip ip addresses, in NETWORK format.
894 * Only valid for IPv4 packets.
897 uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */
898 struct in_addr src_ip, dst_ip; /* NOTE: network format */
901 uint16_t etype = 0; /* Host order stored ether type */
904 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
905 * MATCH_NONE when checked and not matched (q = NULL),
906 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
908 int dyn_dir = MATCH_UNKNOWN;
909 ipfw_dyn_rule *q = NULL;
910 struct ip_fw_chain *chain = &V_layer3_chain;
913 * We store in ulp a pointer to the upper layer protocol header.
914 * In the ipv4 case this is easy to determine from the header,
915 * but for ipv6 we might have some additional headers in the middle.
916 * ulp is NULL if not found.
918 void *ulp = NULL; /* upper layer protocol pointer. */
920 /* XXX ipv6 variables */
922 uint8_t icmp6_type = 0;
923 uint16_t ext_hd = 0; /* bits vector for extension header filtering */
924 /* end of ipv6 variables */
928 int done = 0; /* flag to exit the outer loop */
930 if (m->m_flags & M_SKIP_FIREWALL || (! V_ipfw_vnet_ready))
931 return (IP_FW_PASS); /* accept */
933 dst_ip.s_addr = 0; /* make sure it is initialized */
934 src_ip.s_addr = 0; /* make sure it is initialized */
935 pktlen = m->m_pkthdr.len;
936 args->f_id.fib = M_GETFIB(m); /* note mbuf not altered) */
937 proto = args->f_id.proto = 0; /* mark f_id invalid */
938 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
941 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
942 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
943 * pointer might become stale after other pullups (but we never use it
946 #define PULLUP_TO(_len, p, T) PULLUP_LEN(_len, p, sizeof(T))
947 #define PULLUP_LEN(_len, p, T) \
949 int x = (_len) + T; \
950 if ((m)->m_len < x) { \
951 args->m = m = m_pullup(m, x); \
953 goto pullup_failed; \
955 p = (mtod(m, char *) + (_len)); \
959 * if we have an ether header,
962 etype = ntohs(args->eh->ether_type);
964 /* Identify IP packets and fill up variables. */
965 if (pktlen >= sizeof(struct ip6_hdr) &&
966 (args->eh == NULL || etype == ETHERTYPE_IPV6) && ip->ip_v == 6) {
967 struct ip6_hdr *ip6 = (struct ip6_hdr *)ip;
969 args->f_id.addr_type = 6;
970 hlen = sizeof(struct ip6_hdr);
971 proto = ip6->ip6_nxt;
973 /* Search extension headers to find upper layer protocols */
974 while (ulp == NULL) {
977 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
978 icmp6_type = ICMP6(ulp)->icmp6_type;
982 PULLUP_TO(hlen, ulp, struct tcphdr);
983 dst_port = TCP(ulp)->th_dport;
984 src_port = TCP(ulp)->th_sport;
985 /* save flags for dynamic rules */
986 args->f_id._flags = TCP(ulp)->th_flags;
990 PULLUP_TO(hlen, ulp, struct sctphdr);
991 src_port = SCTP(ulp)->src_port;
992 dst_port = SCTP(ulp)->dest_port;
996 PULLUP_TO(hlen, ulp, struct udphdr);
997 dst_port = UDP(ulp)->uh_dport;
998 src_port = UDP(ulp)->uh_sport;
1001 case IPPROTO_HOPOPTS: /* RFC 2460 */
1002 PULLUP_TO(hlen, ulp, struct ip6_hbh);
1003 ext_hd |= EXT_HOPOPTS;
1004 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
1005 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
1009 case IPPROTO_ROUTING: /* RFC 2460 */
1010 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
1011 switch (((struct ip6_rthdr *)ulp)->ip6r_type) {
1013 ext_hd |= EXT_RTHDR0;
1016 ext_hd |= EXT_RTHDR2;
1019 printf("IPFW2: IPV6 - Unknown Routing "
1020 "Header type(%d)\n",
1021 ((struct ip6_rthdr *)ulp)->ip6r_type);
1022 if (V_fw_deny_unknown_exthdrs)
1023 return (IP_FW_DENY);
1026 ext_hd |= EXT_ROUTING;
1027 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
1028 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
1032 case IPPROTO_FRAGMENT: /* RFC 2460 */
1033 PULLUP_TO(hlen, ulp, struct ip6_frag);
1034 ext_hd |= EXT_FRAGMENT;
1035 hlen += sizeof (struct ip6_frag);
1036 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
1037 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
1039 /* Add IP6F_MORE_FRAG for offset of first
1040 * fragment to be != 0. */
1041 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
1044 printf("IPFW2: IPV6 - Invalid Fragment "
1046 if (V_fw_deny_unknown_exthdrs)
1047 return (IP_FW_DENY);
1051 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
1055 case IPPROTO_DSTOPTS: /* RFC 2460 */
1056 PULLUP_TO(hlen, ulp, struct ip6_hbh);
1057 ext_hd |= EXT_DSTOPTS;
1058 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
1059 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
1063 case IPPROTO_AH: /* RFC 2402 */
1064 PULLUP_TO(hlen, ulp, struct ip6_ext);
1066 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
1067 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
1071 case IPPROTO_ESP: /* RFC 2406 */
1072 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
1073 /* Anything past Seq# is variable length and
1074 * data past this ext. header is encrypted. */
1078 case IPPROTO_NONE: /* RFC 2460 */
1080 * Packet ends here, and IPv6 header has
1081 * already been pulled up. If ip6e_len!=0
1082 * then octets must be ignored.
1084 ulp = ip; /* non-NULL to get out of loop. */
1087 case IPPROTO_OSPFIGP:
1088 /* XXX OSPF header check? */
1089 PULLUP_TO(hlen, ulp, struct ip6_ext);
1093 /* XXX PIM header check? */
1094 PULLUP_TO(hlen, ulp, struct pim);
1098 PULLUP_TO(hlen, ulp, struct carp_header);
1099 if (((struct carp_header *)ulp)->carp_version !=
1101 return (IP_FW_DENY);
1102 if (((struct carp_header *)ulp)->carp_type !=
1104 return (IP_FW_DENY);
1107 case IPPROTO_IPV6: /* RFC 2893 */
1108 PULLUP_TO(hlen, ulp, struct ip6_hdr);
1111 case IPPROTO_IPV4: /* RFC 2893 */
1112 PULLUP_TO(hlen, ulp, struct ip);
1116 printf("IPFW2: IPV6 - Unknown Extension "
1117 "Header(%d), ext_hd=%x\n", proto, ext_hd);
1118 if (V_fw_deny_unknown_exthdrs)
1119 return (IP_FW_DENY);
1120 PULLUP_TO(hlen, ulp, struct ip6_ext);
1124 ip = mtod(m, struct ip *);
1125 ip6 = (struct ip6_hdr *)ip;
1126 args->f_id.src_ip6 = ip6->ip6_src;
1127 args->f_id.dst_ip6 = ip6->ip6_dst;
1128 args->f_id.src_ip = 0;
1129 args->f_id.dst_ip = 0;
1130 args->f_id.flow_id6 = ntohl(ip6->ip6_flow);
1131 } else if (pktlen >= sizeof(struct ip) &&
1132 (args->eh == NULL || etype == ETHERTYPE_IP) && ip->ip_v == 4) {
1134 hlen = ip->ip_hl << 2;
1135 args->f_id.addr_type = 4;
1138 * Collect parameters into local variables for faster matching.
1141 src_ip = ip->ip_src;
1142 dst_ip = ip->ip_dst;
1143 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1144 iplen = ntohs(ip->ip_len);
1145 pktlen = iplen < pktlen ? iplen : pktlen;
1150 PULLUP_TO(hlen, ulp, struct tcphdr);
1151 dst_port = TCP(ulp)->th_dport;
1152 src_port = TCP(ulp)->th_sport;
1153 /* save flags for dynamic rules */
1154 args->f_id._flags = TCP(ulp)->th_flags;
1158 PULLUP_TO(hlen, ulp, struct sctphdr);
1159 src_port = SCTP(ulp)->src_port;
1160 dst_port = SCTP(ulp)->dest_port;
1164 PULLUP_TO(hlen, ulp, struct udphdr);
1165 dst_port = UDP(ulp)->uh_dport;
1166 src_port = UDP(ulp)->uh_sport;
1170 PULLUP_TO(hlen, ulp, struct icmphdr);
1171 //args->f_id.flags = ICMP(ulp)->icmp_type;
1179 ip = mtod(m, struct ip *);
1180 args->f_id.src_ip = ntohl(src_ip.s_addr);
1181 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1184 if (proto) { /* we may have port numbers, store them */
1185 args->f_id.proto = proto;
1186 args->f_id.src_port = src_port = ntohs(src_port);
1187 args->f_id.dst_port = dst_port = ntohs(dst_port);
1191 if (! V_ipfw_vnet_ready) { /* shutting down, leave NOW. */
1192 IPFW_RUNLOCK(chain);
1193 return (IP_FW_PASS); /* accept */
1195 if (args->rule.slot) {
1197 * Packet has already been tagged as a result of a previous
1198 * match on rule args->rule aka args->rule_id (PIPE, QUEUE,
1199 * REASS, NETGRAPH, DIVERT/TEE...)
1200 * Validate the slot and continue from the next one
1201 * if still present, otherwise do a lookup.
1203 f_pos = (args->rule.chain_id == chain->id) ?
1205 ipfw_find_rule(chain, args->rule.rulenum,
1206 args->rule.rule_id);
1212 * Now scan the rules, and parse microinstructions for each rule.
1213 * We have two nested loops and an inner switch. Sometimes we
1214 * need to break out of one or both loops, or re-enter one of
1215 * the loops with updated variables. Loop variables are:
1217 * f_pos (outer loop) points to the current rule.
1218 * On output it points to the matching rule.
1219 * done (outer loop) is used as a flag to break the loop.
1220 * l (inner loop) residual length of current rule.
1221 * cmd points to the current microinstruction.
1223 * We break the inner loop by setting l=0 and possibly
1224 * cmdlen=0 if we don't want to advance cmd.
1225 * We break the outer loop by setting done=1
1226 * We can restart the inner loop by setting l>0 and f_pos, f, cmd
1229 for (; f_pos < chain->n_rules; f_pos++) {
1231 uint32_t tablearg = 0;
1232 int l, cmdlen, skip_or; /* skip rest of OR block */
1235 f = chain->map[f_pos];
1236 if (V_set_disable & (1 << f->set) )
1240 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1241 l -= cmdlen, cmd += cmdlen) {
1245 * check_body is a jump target used when we find a
1246 * CHECK_STATE, and need to jump to the body of
1251 cmdlen = F_LEN(cmd);
1253 * An OR block (insn_1 || .. || insn_n) has the
1254 * F_OR bit set in all but the last instruction.
1255 * The first match will set "skip_or", and cause
1256 * the following instructions to be skipped until
1257 * past the one with the F_OR bit clear.
1259 if (skip_or) { /* skip this instruction */
1260 if ((cmd->len & F_OR) == 0)
1261 skip_or = 0; /* next one is good */
1264 match = 0; /* set to 1 if we succeed */
1266 switch (cmd->opcode) {
1268 * The first set of opcodes compares the packet's
1269 * fields with some pattern, setting 'match' if a
1270 * match is found. At the end of the loop there is
1271 * logic to deal with F_NOT and F_OR flags associated
1279 printf("ipfw: opcode %d unimplemented\n",
1287 * We only check offset == 0 && proto != 0,
1288 * as this ensures that we have a
1289 * packet with the ports info.
1293 if (proto == IPPROTO_TCP ||
1294 proto == IPPROTO_UDP)
1295 match = check_uidgid(
1296 (ipfw_insn_u32 *)cmd,
1297 args, &ucred_lookup,
1301 (void *)&ucred_cache);
1306 match = iface_match(m->m_pkthdr.rcvif,
1307 (ipfw_insn_if *)cmd);
1311 match = iface_match(oif, (ipfw_insn_if *)cmd);
1315 match = iface_match(oif ? oif :
1316 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1320 if (args->eh != NULL) { /* have MAC header */
1321 u_int32_t *want = (u_int32_t *)
1322 ((ipfw_insn_mac *)cmd)->addr;
1323 u_int32_t *mask = (u_int32_t *)
1324 ((ipfw_insn_mac *)cmd)->mask;
1325 u_int32_t *hdr = (u_int32_t *)args->eh;
1328 ( want[0] == (hdr[0] & mask[0]) &&
1329 want[1] == (hdr[1] & mask[1]) &&
1330 want[2] == (hdr[2] & mask[2]) );
1335 if (args->eh != NULL) {
1337 ((ipfw_insn_u16 *)cmd)->ports;
1340 for (i = cmdlen - 1; !match && i>0;
1342 match = (etype >= p[0] &&
1348 match = (offset != 0);
1351 case O_IN: /* "out" is "not in" */
1352 match = (oif == NULL);
1356 match = (args->eh != NULL);
1361 /* For diverted packets, args->rule.info
1362 * contains the divert port (in host format)
1363 * reason and direction.
1365 uint32_t i = args->rule.info;
1366 match = (i&IPFW_IS_MASK) == IPFW_IS_DIVERT &&
1367 cmd->arg1 & ((i & IPFW_INFO_IN) ? 1 : 2);
1373 * We do not allow an arg of 0 so the
1374 * check of "proto" only suffices.
1376 match = (proto == cmd->arg1);
1381 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1385 case O_IP_SRC_LOOKUP:
1386 case O_IP_DST_LOOKUP:
1389 (cmd->opcode == O_IP_DST_LOOKUP) ?
1390 dst_ip.s_addr : src_ip.s_addr;
1393 if (cmdlen > F_INSN_SIZE(ipfw_insn_u32)) {
1394 /* generic lookup. The key must be
1395 * in 32bit big-endian format.
1397 v = ((ipfw_insn_u32 *)cmd)->d[1];
1399 key = dst_ip.s_addr;
1401 key = src_ip.s_addr;
1402 else if (v == 6) /* dscp */
1403 key = (ip->ip_tos >> 2) & 0x3f;
1404 else if (offset != 0)
1406 else if (proto != IPPROTO_TCP &&
1407 proto != IPPROTO_UDP)
1410 key = htonl(dst_port);
1412 key = htonl(src_port);
1413 else if (v == 4 || v == 5) {
1415 (ipfw_insn_u32 *)cmd,
1416 args, &ucred_lookup,
1419 if (v == 4 /* O_UID */)
1420 key = ucred_cache->cr_uid;
1421 else if (v == 5 /* O_JAIL */)
1422 key = ucred_cache->cr_prison->pr_id;
1423 #else /* !__FreeBSD__ */
1424 (void *)&ucred_cache);
1425 if (v ==4 /* O_UID */)
1426 key = ucred_cache.uid;
1427 else if (v == 5 /* O_JAIL */)
1428 key = ucred_cache.xid;
1429 #endif /* !__FreeBSD__ */
1434 match = ipfw_lookup_table(chain,
1435 cmd->arg1, key, &v);
1438 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
1440 ((ipfw_insn_u32 *)cmd)->d[0] == v;
1450 (cmd->opcode == O_IP_DST_MASK) ?
1451 dst_ip.s_addr : src_ip.s_addr;
1452 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
1455 for (; !match && i>0; i-= 2, p+= 2)
1456 match = (p[0] == (a & p[1]));
1464 INADDR_TO_IFP(src_ip, tif);
1465 match = (tif != NULL);
1471 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
1478 u_int32_t *d = (u_int32_t *)(cmd+1);
1480 cmd->opcode == O_IP_DST_SET ?
1486 addr -= d[0]; /* subtract base */
1487 match = (addr < cmd->arg1) &&
1488 ( d[ 1 + (addr>>5)] &
1489 (1<<(addr & 0x1f)) );
1495 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1503 INADDR_TO_IFP(dst_ip, tif);
1504 match = (tif != NULL);
1510 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
1518 * offset == 0 && proto != 0 is enough
1519 * to guarantee that we have a
1520 * packet with port info.
1522 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1525 (cmd->opcode == O_IP_SRCPORT) ?
1526 src_port : dst_port ;
1528 ((ipfw_insn_u16 *)cmd)->ports;
1531 for (i = cmdlen - 1; !match && i>0;
1533 match = (x>=p[0] && x<=p[1]);
1538 match = (offset == 0 && proto==IPPROTO_ICMP &&
1539 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
1544 match = is_ipv6 && offset == 0 &&
1545 proto==IPPROTO_ICMPV6 &&
1547 ICMP6(ulp)->icmp6_type,
1548 (ipfw_insn_u32 *)cmd);
1554 ipopts_match(ip, cmd) );
1559 cmd->arg1 == ip->ip_v);
1565 if (is_ipv4) { /* only for IP packets */
1570 if (cmd->opcode == O_IPLEN)
1572 else if (cmd->opcode == O_IPTTL)
1574 else /* must be IPID */
1575 x = ntohs(ip->ip_id);
1577 match = (cmd->arg1 == x);
1580 /* otherwise we have ranges */
1581 p = ((ipfw_insn_u16 *)cmd)->ports;
1583 for (; !match && i>0; i--, p += 2)
1584 match = (x >= p[0] && x <= p[1]);
1588 case O_IPPRECEDENCE:
1590 (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1595 flags_match(cmd, ip->ip_tos));
1599 if (proto == IPPROTO_TCP && offset == 0) {
1607 ((ip->ip_hl + tcp->th_off) << 2);
1609 match = (cmd->arg1 == x);
1612 /* otherwise we have ranges */
1613 p = ((ipfw_insn_u16 *)cmd)->ports;
1615 for (; !match && i>0; i--, p += 2)
1616 match = (x >= p[0] && x <= p[1]);
1621 match = (proto == IPPROTO_TCP && offset == 0 &&
1622 flags_match(cmd, TCP(ulp)->th_flags));
1626 PULLUP_LEN(hlen, ulp, (TCP(ulp)->th_off << 2));
1627 match = (proto == IPPROTO_TCP && offset == 0 &&
1628 tcpopts_match(TCP(ulp), cmd));
1632 match = (proto == IPPROTO_TCP && offset == 0 &&
1633 ((ipfw_insn_u32 *)cmd)->d[0] ==
1638 match = (proto == IPPROTO_TCP && offset == 0 &&
1639 ((ipfw_insn_u32 *)cmd)->d[0] ==
1644 match = (proto == IPPROTO_TCP && offset == 0 &&
1645 cmd->arg1 == TCP(ulp)->th_win);
1649 /* reject packets which have SYN only */
1650 /* XXX should i also check for TH_ACK ? */
1651 match = (proto == IPPROTO_TCP && offset == 0 &&
1652 (TCP(ulp)->th_flags &
1653 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1658 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
1661 at = pf_find_mtag(m);
1662 if (at != NULL && at->qid != 0)
1664 at = pf_get_mtag(m);
1667 * Let the packet fall back to the
1672 at->qid = altq->qid;
1678 ipfw_log(f, hlen, args, m,
1679 oif, offset, tablearg, ip);
1684 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
1688 /* Outgoing packets automatically pass/match */
1689 match = ((oif != NULL) ||
1690 (m->m_pkthdr.rcvif == NULL) ||
1694 verify_path6(&(args->f_id.src_ip6),
1695 m->m_pkthdr.rcvif) :
1697 verify_path(src_ip, m->m_pkthdr.rcvif,
1702 /* Outgoing packets automatically pass/match */
1703 match = (hlen > 0 && ((oif != NULL) ||
1706 verify_path6(&(args->f_id.src_ip6),
1709 verify_path(src_ip, NULL, args->f_id.fib)));
1713 /* Outgoing packets automatically pass/match */
1714 if (oif == NULL && hlen > 0 &&
1715 ( (is_ipv4 && in_localaddr(src_ip))
1718 in6_localaddr(&(args->f_id.src_ip6)))
1723 is_ipv6 ? verify_path6(
1724 &(args->f_id.src_ip6),
1725 m->m_pkthdr.rcvif) :
1736 match = (m_tag_find(m,
1737 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
1739 /* otherwise no match */
1745 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
1746 &((ipfw_insn_ip6 *)cmd)->addr6);
1751 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
1752 &((ipfw_insn_ip6 *)cmd)->addr6);
1754 case O_IP6_SRC_MASK:
1755 case O_IP6_DST_MASK:
1759 struct in6_addr *d =
1760 &((ipfw_insn_ip6 *)cmd)->addr6;
1762 for (; !match && i > 0; d += 2,
1763 i -= F_INSN_SIZE(struct in6_addr)
1769 APPLY_MASK(&p, &d[1]);
1771 IN6_ARE_ADDR_EQUAL(&d[0],
1779 flow6id_match(args->f_id.flow_id6,
1780 (ipfw_insn_u32 *) cmd);
1785 (ext_hd & ((ipfw_insn *) cmd)->arg1);
1799 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
1800 tablearg : cmd->arg1;
1802 /* Packet is already tagged with this tag? */
1803 mtag = m_tag_locate(m, MTAG_IPFW, tag, NULL);
1805 /* We have `untag' action when F_NOT flag is
1806 * present. And we must remove this mtag from
1807 * mbuf and reset `match' to zero (`match' will
1808 * be inversed later).
1809 * Otherwise we should allocate new mtag and
1810 * push it into mbuf.
1812 if (cmd->len & F_NOT) { /* `untag' action */
1814 m_tag_delete(m, mtag);
1818 mtag = m_tag_alloc( MTAG_IPFW,
1821 m_tag_prepend(m, mtag);
1828 case O_FIB: /* try match the specified fib */
1829 if (args->f_id.fib == cmd->arg1)
1834 struct inpcb *inp = args->inp;
1835 struct inpcbinfo *pi;
1837 if (is_ipv6) /* XXX can we remove this ? */
1840 if (proto == IPPROTO_TCP)
1842 else if (proto == IPPROTO_UDP)
1848 * XXXRW: so_user_cookie should almost
1849 * certainly be inp_user_cookie?
1852 /* For incomming packet, lookup up the
1853 inpcb using the src/dest ip/port tuple */
1855 inp = in_pcblookup(pi,
1856 src_ip, htons(src_port),
1857 dst_ip, htons(dst_port),
1858 INPLOOKUP_RLOCKPCB, NULL);
1861 inp->inp_socket->so_user_cookie;
1867 if (inp->inp_socket) {
1869 inp->inp_socket->so_user_cookie;
1879 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
1880 tablearg : cmd->arg1;
1883 match = m_tag_locate(m, MTAG_IPFW,
1888 /* we have ranges */
1889 for (mtag = m_tag_first(m);
1890 mtag != NULL && !match;
1891 mtag = m_tag_next(m, mtag)) {
1895 if (mtag->m_tag_cookie != MTAG_IPFW)
1898 p = ((ipfw_insn_u16 *)cmd)->ports;
1900 for(; !match && i > 0; i--, p += 2)
1902 mtag->m_tag_id >= p[0] &&
1903 mtag->m_tag_id <= p[1];
1909 * The second set of opcodes represents 'actions',
1910 * i.e. the terminal part of a rule once the packet
1911 * matches all previous patterns.
1912 * Typically there is only one action for each rule,
1913 * and the opcode is stored at the end of the rule
1914 * (but there are exceptions -- see below).
1916 * In general, here we set retval and terminate the
1917 * outer loop (would be a 'break 3' in some language,
1918 * but we need to set l=0, done=1)
1921 * O_COUNT and O_SKIPTO actions:
1922 * instead of terminating, we jump to the next rule
1923 * (setting l=0), or to the SKIPTO target (setting
1924 * f/f_len, cmd and l as needed), respectively.
1926 * O_TAG, O_LOG and O_ALTQ action parameters:
1927 * perform some action and set match = 1;
1929 * O_LIMIT and O_KEEP_STATE: these opcodes are
1930 * not real 'actions', and are stored right
1931 * before the 'action' part of the rule.
1932 * These opcodes try to install an entry in the
1933 * state tables; if successful, we continue with
1934 * the next opcode (match=1; break;), otherwise
1935 * the packet must be dropped (set retval,
1936 * break loops with l=0, done=1)
1938 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1939 * cause a lookup of the state table, and a jump
1940 * to the 'action' part of the parent rule
1941 * if an entry is found, or
1942 * (CHECK_STATE only) a jump to the next rule if
1943 * the entry is not found.
1944 * The result of the lookup is cached so that
1945 * further instances of these opcodes become NOPs.
1946 * The jump to the next rule is done by setting
1951 if (ipfw_install_state(f,
1952 (ipfw_insn_limit *)cmd, args, tablearg)) {
1953 /* error or limit violation */
1954 retval = IP_FW_DENY;
1955 l = 0; /* exit inner loop */
1956 done = 1; /* exit outer loop */
1964 * dynamic rules are checked at the first
1965 * keep-state or check-state occurrence,
1966 * with the result being stored in dyn_dir.
1967 * The compiler introduces a PROBE_STATE
1968 * instruction for us when we have a
1969 * KEEP_STATE (because PROBE_STATE needs
1972 if (dyn_dir == MATCH_UNKNOWN &&
1973 (q = ipfw_lookup_dyn_rule(&args->f_id,
1974 &dyn_dir, proto == IPPROTO_TCP ?
1978 * Found dynamic entry, update stats
1979 * and jump to the 'action' part of
1980 * the parent rule by setting
1981 * f, cmd, l and clearing cmdlen.
1985 /* XXX we would like to have f_pos
1986 * readily accessible in the dynamic
1987 * rule, instead of having to
1991 f_pos = ipfw_find_rule(chain,
1993 cmd = ACTION_PTR(f);
1994 l = f->cmd_len - f->act_ofs;
2001 * Dynamic entry not found. If CHECK_STATE,
2002 * skip to next rule, if PROBE_STATE just
2003 * ignore and continue with next opcode.
2005 if (cmd->opcode == O_CHECK_STATE)
2006 l = 0; /* exit inner loop */
2011 retval = 0; /* accept */
2012 l = 0; /* exit inner loop */
2013 done = 1; /* exit outer loop */
2018 set_match(args, f_pos, chain);
2019 args->rule.info = (cmd->arg1 == IP_FW_TABLEARG) ?
2020 tablearg : cmd->arg1;
2021 if (cmd->opcode == O_PIPE)
2022 args->rule.info |= IPFW_IS_PIPE;
2024 args->rule.info |= IPFW_ONEPASS;
2025 retval = IP_FW_DUMMYNET;
2026 l = 0; /* exit inner loop */
2027 done = 1; /* exit outer loop */
2032 if (args->eh) /* not on layer 2 */
2034 /* otherwise this is terminal */
2035 l = 0; /* exit inner loop */
2036 done = 1; /* exit outer loop */
2037 retval = (cmd->opcode == O_DIVERT) ?
2038 IP_FW_DIVERT : IP_FW_TEE;
2039 set_match(args, f_pos, chain);
2040 args->rule.info = (cmd->arg1 == IP_FW_TABLEARG) ?
2041 tablearg : cmd->arg1;
2045 f->pcnt++; /* update stats */
2047 f->timestamp = time_uptime;
2048 l = 0; /* exit inner loop */
2052 f->pcnt++; /* update stats */
2054 f->timestamp = time_uptime;
2055 /* If possible use cached f_pos (in f->next_rule),
2056 * whose version is written in f->next_rule
2057 * (horrible hacks to avoid changing the ABI).
2059 if (cmd->arg1 != IP_FW_TABLEARG &&
2060 (uintptr_t)f->x_next == chain->id) {
2061 f_pos = (uintptr_t)f->next_rule;
2063 int i = (cmd->arg1 == IP_FW_TABLEARG) ?
2064 tablearg : cmd->arg1;
2065 /* make sure we do not jump backward */
2066 if (i <= f->rulenum)
2068 f_pos = ipfw_find_rule(chain, i, 0);
2069 /* update the cache */
2070 if (cmd->arg1 != IP_FW_TABLEARG) {
2072 (void *)(uintptr_t)f_pos;
2074 (void *)(uintptr_t)chain->id;
2078 * Skip disabled rules, and re-enter
2079 * the inner loop with the correct
2080 * f_pos, f, l and cmd.
2081 * Also clear cmdlen and skip_or
2083 for (; f_pos < chain->n_rules - 1 &&
2085 (1 << chain->map[f_pos]->set));
2088 /* Re-enter the inner loop at the skipto rule. */
2089 f = chain->map[f_pos];
2096 break; /* not reached */
2100 * Drop the packet and send a reject notice
2101 * if the packet is not ICMP (or is an ICMP
2102 * query), and it is not multicast/broadcast.
2104 if (hlen > 0 && is_ipv4 && offset == 0 &&
2105 (proto != IPPROTO_ICMP ||
2106 is_icmp_query(ICMP(ulp))) &&
2107 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2108 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2109 send_reject(args, cmd->arg1, iplen, ip);
2115 if (hlen > 0 && is_ipv6 &&
2116 ((offset & IP6F_OFF_MASK) == 0) &&
2117 (proto != IPPROTO_ICMPV6 ||
2118 (is_icmp6_query(icmp6_type) == 1)) &&
2119 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2120 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
2122 args, cmd->arg1, hlen,
2123 (struct ip6_hdr *)ip);
2129 retval = IP_FW_DENY;
2130 l = 0; /* exit inner loop */
2131 done = 1; /* exit outer loop */
2135 if (args->eh) /* not valid on layer2 pkts */
2137 if (q == NULL || q->rule != f ||
2138 dyn_dir == MATCH_FORWARD) {
2139 struct sockaddr_in *sa;
2140 sa = &(((ipfw_insn_sa *)cmd)->sa);
2141 if (sa->sin_addr.s_addr == INADDR_ANY) {
2142 bcopy(sa, &args->hopstore,
2144 args->hopstore.sin_addr.s_addr =
2146 args->next_hop = &args->hopstore;
2148 args->next_hop = sa;
2151 retval = IP_FW_PASS;
2152 l = 0; /* exit inner loop */
2153 done = 1; /* exit outer loop */
2158 set_match(args, f_pos, chain);
2159 args->rule.info = (cmd->arg1 == IP_FW_TABLEARG) ?
2160 tablearg : cmd->arg1;
2162 args->rule.info |= IPFW_ONEPASS;
2163 retval = (cmd->opcode == O_NETGRAPH) ?
2164 IP_FW_NETGRAPH : IP_FW_NGTEE;
2165 l = 0; /* exit inner loop */
2166 done = 1; /* exit outer loop */
2172 f->pcnt++; /* update stats */
2174 f->timestamp = time_uptime;
2175 fib = (cmd->arg1 == IP_FW_TABLEARG) ? tablearg:
2177 if (fib >= rt_numfibs)
2180 args->f_id.fib = fib;
2181 l = 0; /* exit inner loop */
2186 if (!IPFW_NAT_LOADED) {
2187 retval = IP_FW_DENY;
2192 set_match(args, f_pos, chain);
2193 /* Check if this is 'global' nat rule */
2194 if (cmd->arg1 == 0) {
2195 retval = ipfw_nat_ptr(args, NULL, m);
2200 t = ((ipfw_insn_nat *)cmd)->nat;
2202 nat_id = (cmd->arg1 == IP_FW_TABLEARG) ?
2203 tablearg : cmd->arg1;
2204 t = (*lookup_nat_ptr)(&chain->nat, nat_id);
2207 retval = IP_FW_DENY;
2208 l = 0; /* exit inner loop */
2209 done = 1; /* exit outer loop */
2212 if (cmd->arg1 != IP_FW_TABLEARG)
2213 ((ipfw_insn_nat *)cmd)->nat = t;
2215 retval = ipfw_nat_ptr(args, t, m);
2217 l = 0; /* exit inner loop */
2218 done = 1; /* exit outer loop */
2226 l = 0; /* in any case exit inner loop */
2227 ip_off = ntohs(ip->ip_off);
2229 /* if not fragmented, go to next rule */
2230 if ((ip_off & (IP_MF | IP_OFFMASK)) == 0)
2233 * ip_reass() expects len & off in host
2238 args->m = m = ip_reass(m);
2241 * do IP header checksum fixup.
2243 if (m == NULL) { /* fragment got swallowed */
2244 retval = IP_FW_DENY;
2245 } else { /* good, packet complete */
2248 ip = mtod(m, struct ip *);
2249 hlen = ip->ip_hl << 2;
2252 if (hlen == sizeof(struct ip))
2253 ip->ip_sum = in_cksum_hdr(ip);
2255 ip->ip_sum = in_cksum(m, hlen);
2256 retval = IP_FW_REASS;
2257 set_match(args, f_pos, chain);
2259 done = 1; /* exit outer loop */
2264 panic("-- unknown opcode %d\n", cmd->opcode);
2265 } /* end of switch() on opcodes */
2267 * if we get here with l=0, then match is irrelevant.
2270 if (cmd->len & F_NOT)
2274 if (cmd->len & F_OR)
2277 if (!(cmd->len & F_OR)) /* not an OR block, */
2278 break; /* try next rule */
2281 } /* end of inner loop, scan opcodes */
2287 /* next_rule:; */ /* try next rule */
2289 } /* end of outer for, scan rules */
2292 struct ip_fw *rule = chain->map[f_pos];
2293 /* Update statistics */
2295 rule->bcnt += pktlen;
2296 rule->timestamp = time_uptime;
2298 retval = IP_FW_DENY;
2299 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
2301 IPFW_RUNLOCK(chain);
2303 if (ucred_cache != NULL)
2304 crfree(ucred_cache);
2310 printf("ipfw: pullup failed\n");
2311 return (IP_FW_DENY);
2315 * Module and VNET glue
2319 * Stuff that must be initialised only on boot or module load
2328 * Only print out this stuff the first time around,
2329 * when called from the sysinit code.
2335 "initialized, divert %s, nat %s, "
2336 "rule-based forwarding "
2337 #ifdef IPFIREWALL_FORWARD
2342 "default to %s, logging ",
2348 #ifdef IPFIREWALL_NAT
2353 default_to_accept ? "accept" : "deny");
2356 * Note: V_xxx variables can be accessed here but the vnet specific
2357 * initializer may not have been called yet for the VIMAGE case.
2358 * Tuneables will have been processed. We will print out values for
2360 * XXX This should all be rationalized AFTER 8.0
2362 if (V_fw_verbose == 0)
2363 printf("disabled\n");
2364 else if (V_verbose_limit == 0)
2365 printf("unlimited\n");
2367 printf("limited to %d packets/entry by default\n",
2370 ipfw_log_bpf(1); /* init */
2375 * Called for the removal of the last instance only on module unload.
2381 ipfw_log_bpf(0); /* uninit */
2383 printf("IP firewall unloaded\n");
2387 * Stuff that must be initialized for every instance
2388 * (including the first of course).
2391 vnet_ipfw_init(const void *unused)
2394 struct ip_fw *rule = NULL;
2395 struct ip_fw_chain *chain;
2397 chain = &V_layer3_chain;
2399 /* First set up some values that are compile time options */
2400 V_autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
2401 V_fw_deny_unknown_exthdrs = 1;
2402 #ifdef IPFIREWALL_VERBOSE
2405 #ifdef IPFIREWALL_VERBOSE_LIMIT
2406 V_verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2408 #ifdef IPFIREWALL_NAT
2409 LIST_INIT(&chain->nat);
2412 /* insert the default rule and create the initial map */
2414 chain->static_len = sizeof(struct ip_fw);
2415 chain->map = malloc(sizeof(struct ip_fw *), M_IPFW, M_NOWAIT | M_ZERO);
2417 rule = malloc(chain->static_len, M_IPFW, M_NOWAIT | M_ZERO);
2420 free(chain->map, M_IPFW);
2421 printf("ipfw2: ENOSPC initializing default rule "
2422 "(support disabled)\n");
2425 error = ipfw_init_tables(chain);
2427 panic("init_tables"); /* XXX Marko fix this ! */
2430 /* fill and insert the default rule */
2432 rule->rulenum = IPFW_DEFAULT_RULE;
2434 rule->set = RESVD_SET;
2435 rule->cmd[0].len = 1;
2436 rule->cmd[0].opcode = default_to_accept ? O_ACCEPT : O_DENY;
2437 chain->rules = chain->default_rule = chain->map[0] = rule;
2438 chain->id = rule->id = 1;
2440 IPFW_LOCK_INIT(chain);
2443 /* First set up some values that are compile time options */
2444 V_ipfw_vnet_ready = 1; /* Open for business */
2447 * Hook the sockopt handler, and the layer2 (V_ip_fw_chk_ptr)
2448 * and pfil hooks for ipv4 and ipv6. Even if the latter two fail
2449 * we still keep the module alive because the sockopt and
2450 * layer2 paths are still useful.
2451 * ipfw[6]_hook return 0 on success, ENOENT on failure,
2452 * so we can ignore the exact return value and just set a flag.
2454 * Note that V_fw[6]_enable are manipulated by a SYSCTL_PROC so
2455 * changes in the underlying (per-vnet) variables trigger
2456 * immediate hook()/unhook() calls.
2457 * In layer2 we have the same behaviour, except that V_ether_ipfw
2458 * is checked on each packet because there are no pfil hooks.
2460 V_ip_fw_ctl_ptr = ipfw_ctl;
2461 V_ip_fw_chk_ptr = ipfw_chk;
2462 error = ipfw_attach_hooks(1);
2467 * Called for the removal of each instance.
2470 vnet_ipfw_uninit(const void *unused)
2472 struct ip_fw *reap, *rule;
2473 struct ip_fw_chain *chain = &V_layer3_chain;
2476 V_ipfw_vnet_ready = 0; /* tell new callers to go away */
2478 * disconnect from ipv4, ipv6, layer2 and sockopt.
2479 * Then grab, release and grab again the WLOCK so we make
2480 * sure the update is propagated and nobody will be in.
2482 (void)ipfw_attach_hooks(0 /* detach */);
2483 V_ip_fw_chk_ptr = NULL;
2484 V_ip_fw_ctl_ptr = NULL;
2485 IPFW_UH_WLOCK(chain);
2486 IPFW_UH_WUNLOCK(chain);
2487 IPFW_UH_WLOCK(chain);
2490 IPFW_WUNLOCK(chain);
2493 ipfw_dyn_uninit(0); /* run the callout_drain */
2494 ipfw_destroy_tables(chain);
2496 for (i = 0; i < chain->n_rules; i++) {
2497 rule = chain->map[i];
2498 rule->x_next = reap;
2502 free(chain->map, M_IPFW);
2503 IPFW_WUNLOCK(chain);
2504 IPFW_UH_WUNLOCK(chain);
2506 ipfw_reap_rules(reap);
2507 IPFW_LOCK_DESTROY(chain);
2508 ipfw_dyn_uninit(1); /* free the remaining parts */
2513 * Module event handler.
2514 * In general we have the choice of handling most of these events by the
2515 * event handler or by the (VNET_)SYS(UN)INIT handlers. I have chosen to
2516 * use the SYSINIT handlers as they are more capable of expressing the
2517 * flow of control during module and vnet operations, so this is just
2518 * a skeleton. Note there is no SYSINIT equivalent of the module
2519 * SHUTDOWN handler, but we don't have anything to do in that case anyhow.
2522 ipfw_modevent(module_t mod, int type, void *unused)
2528 /* Called once at module load or
2529 * system boot if compiled in. */
2532 /* Called before unload. May veto unloading. */
2535 /* Called during unload. */
2538 /* Called during system shutdown. */
2547 static moduledata_t ipfwmod = {
2553 /* Define startup order. */
2554 #define IPFW_SI_SUB_FIREWALL SI_SUB_PROTO_IFATTACHDOMAIN
2555 #define IPFW_MODEVENT_ORDER (SI_ORDER_ANY - 255) /* On boot slot in here. */
2556 #define IPFW_MODULE_ORDER (IPFW_MODEVENT_ORDER + 1) /* A little later. */
2557 #define IPFW_VNET_ORDER (IPFW_MODEVENT_ORDER + 2) /* Later still. */
2559 DECLARE_MODULE(ipfw, ipfwmod, IPFW_SI_SUB_FIREWALL, IPFW_MODEVENT_ORDER);
2560 MODULE_VERSION(ipfw, 2);
2561 /* should declare some dependencies here */
2564 * Starting up. Done in order after ipfwmod() has been called.
2565 * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2567 SYSINIT(ipfw_init, IPFW_SI_SUB_FIREWALL, IPFW_MODULE_ORDER,
2569 VNET_SYSINIT(vnet_ipfw_init, IPFW_SI_SUB_FIREWALL, IPFW_VNET_ORDER,
2570 vnet_ipfw_init, NULL);
2573 * Closing up shop. These are done in REVERSE ORDER, but still
2574 * after ipfwmod() has been called. Not called on reboot.
2575 * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2576 * or when the module is unloaded.
2578 SYSUNINIT(ipfw_destroy, IPFW_SI_SUB_FIREWALL, IPFW_MODULE_ORDER,
2579 ipfw_destroy, NULL);
2580 VNET_SYSUNINIT(vnet_ipfw_uninit, IPFW_SI_SUB_FIREWALL, IPFW_VNET_ORDER,
2581 vnet_ipfw_uninit, NULL);