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: head/sys/netinet/ipfw/ip_fw2.c 200601 2009-12-16 10:48:40Z luigi $");
30 * The FreeBSD IP packet firewall, main file
34 #include "opt_ipdivert.h"
37 #error "IPFIREWALL requires INET"
39 #include "opt_inet6.h"
40 #include "opt_ipsec.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/condvar.h>
45 #include <sys/eventhandler.h>
46 #include <sys/malloc.h>
48 #include <sys/kernel.h>
51 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/ucred.h>
60 #include <net/ethernet.h> /* for ETHERTYPE_IP */
62 #include <net/route.h>
63 #include <net/pf_mtag.h>
66 #include <netinet/in.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_icmp.h>
72 #include <netinet/ip_fw.h>
73 #include <netinet/ipfw/ip_fw_private.h>
74 #include <netinet/ip_carp.h>
75 #include <netinet/pim.h>
76 #include <netinet/tcp_var.h>
77 #include <netinet/udp.h>
78 #include <netinet/udp_var.h>
79 #include <netinet/sctp.h>
81 #include <netinet/ip6.h>
82 #include <netinet/icmp6.h>
84 #include <netinet6/in6_pcb.h>
85 #include <netinet6/scope6_var.h>
86 #include <netinet6/ip6_var.h>
89 #include <machine/in_cksum.h> /* XXX for in_cksum */
92 #include <security/mac/mac_framework.h>
96 * static variables followed by global ones.
97 * All ipfw global variables are here.
100 /* ipfw_vnet_ready controls when we are open for business */
101 static VNET_DEFINE(int, ipfw_vnet_ready) = 0;
102 #define V_ipfw_vnet_ready VNET(ipfw_vnet_ready)
104 static VNET_DEFINE(int, fw_deny_unknown_exthdrs);
105 #define V_fw_deny_unknown_exthdrs VNET(fw_deny_unknown_exthdrs)
107 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
108 static int default_to_accept = 1;
110 static int default_to_accept;
113 VNET_DEFINE(int, autoinc_step);
116 * Each rule belongs to one of 32 different sets (0..31).
117 * The variable set_disable contains one bit per set.
118 * If the bit is set, all rules in the corresponding set
119 * are disabled. Set RESVD_SET(31) is reserved for the default rule
120 * and rules that are not deleted by the flush command,
121 * and CANNOT be disabled.
122 * Rules in set RESVD_SET can only be deleted individually.
124 VNET_DEFINE(u_int32_t, set_disable);
125 #define V_set_disable VNET(set_disable)
127 VNET_DEFINE(int, fw_verbose);
128 /* counter for ipfw_log(NULL...) */
129 VNET_DEFINE(u_int64_t, norule_counter);
130 VNET_DEFINE(int, verbose_limit);
132 /* layer3_chain contains the list of rules for layer 3 */
133 VNET_DEFINE(struct ip_fw_chain, layer3_chain);
135 ipfw_nat_t *ipfw_nat_ptr = NULL;
136 struct cfg_nat *(*lookup_nat_ptr)(struct nat_list *, int);
137 ipfw_nat_cfg_t *ipfw_nat_cfg_ptr;
138 ipfw_nat_cfg_t *ipfw_nat_del_ptr;
139 ipfw_nat_cfg_t *ipfw_nat_get_cfg_ptr;
140 ipfw_nat_cfg_t *ipfw_nat_get_log_ptr;
143 uint32_t dummy_def = IPFW_DEFAULT_RULE;
144 uint32_t dummy_tables_max = IPFW_TABLES_MAX;
148 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
149 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
150 CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_one_pass), 0,
151 "Only do a single pass through ipfw when using dummynet(4)");
152 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step,
153 CTLFLAG_RW, &VNET_NAME(autoinc_step), 0,
154 "Rule number auto-increment step");
155 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, verbose,
156 CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_verbose), 0,
157 "Log matches to ipfw rules");
158 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit,
159 CTLFLAG_RW, &VNET_NAME(verbose_limit), 0,
160 "Set upper limit of matches of ipfw rules logged");
161 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, default_rule, CTLFLAG_RD,
163 "The default/max possible rule number.");
164 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, tables_max, CTLFLAG_RD,
165 &dummy_tables_max, 0,
166 "The maximum number of tables.");
167 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, default_to_accept, CTLFLAG_RDTUN,
168 &default_to_accept, 0,
169 "Make the default rule accept all packets.");
170 TUNABLE_INT("net.inet.ip.fw.default_to_accept", &default_to_accept);
171 SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, static_count,
172 CTLFLAG_RD, &VNET_NAME(layer3_chain.n_rules), 0,
173 "Number of static rules");
176 SYSCTL_DECL(_net_inet6_ip6);
177 SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
178 SYSCTL_VNET_INT(_net_inet6_ip6_fw, OID_AUTO, deny_unknown_exthdrs,
179 CTLFLAG_RW | CTLFLAG_SECURE, &VNET_NAME(fw_deny_unknown_exthdrs), 0,
180 "Deny packets with unknown IPv6 Extension Headers");
185 #endif /* SYSCTL_NODE */
189 * Some macros used in the various matching options.
190 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
191 * Other macros just cast void * into the appropriate type
193 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
194 #define TCP(p) ((struct tcphdr *)(p))
195 #define SCTP(p) ((struct sctphdr *)(p))
196 #define UDP(p) ((struct udphdr *)(p))
197 #define ICMP(p) ((struct icmphdr *)(p))
198 #define ICMP6(p) ((struct icmp6_hdr *)(p))
201 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
203 int type = icmp->icmp_type;
205 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
208 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
209 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
212 is_icmp_query(struct icmphdr *icmp)
214 int type = icmp->icmp_type;
216 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
221 * The following checks use two arrays of 8 or 16 bits to store the
222 * bits that we want set or clear, respectively. They are in the
223 * low and high half of cmd->arg1 or cmd->d[0].
225 * We scan options and store the bits we find set. We succeed if
227 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
229 * The code is sometimes optimized not to store additional variables.
233 flags_match(ipfw_insn *cmd, u_int8_t bits)
238 if ( ((cmd->arg1 & 0xff) & bits) != 0)
239 return 0; /* some bits we want set were clear */
240 want_clear = (cmd->arg1 >> 8) & 0xff;
241 if ( (want_clear & bits) != want_clear)
242 return 0; /* some bits we want clear were set */
247 ipopts_match(struct ip *ip, ipfw_insn *cmd)
249 int optlen, bits = 0;
250 u_char *cp = (u_char *)(ip + 1);
251 int x = (ip->ip_hl << 2) - sizeof (struct ip);
253 for (; x > 0; x -= optlen, cp += optlen) {
254 int opt = cp[IPOPT_OPTVAL];
256 if (opt == IPOPT_EOL)
258 if (opt == IPOPT_NOP)
261 optlen = cp[IPOPT_OLEN];
262 if (optlen <= 0 || optlen > x)
263 return 0; /* invalid or truncated */
271 bits |= IP_FW_IPOPT_LSRR;
275 bits |= IP_FW_IPOPT_SSRR;
279 bits |= IP_FW_IPOPT_RR;
283 bits |= IP_FW_IPOPT_TS;
287 return (flags_match(cmd, bits));
291 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
293 int optlen, bits = 0;
294 u_char *cp = (u_char *)(tcp + 1);
295 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
297 for (; x > 0; x -= optlen, cp += optlen) {
299 if (opt == TCPOPT_EOL)
301 if (opt == TCPOPT_NOP)
315 bits |= IP_FW_TCPOPT_MSS;
319 bits |= IP_FW_TCPOPT_WINDOW;
322 case TCPOPT_SACK_PERMITTED:
324 bits |= IP_FW_TCPOPT_SACK;
327 case TCPOPT_TIMESTAMP:
328 bits |= IP_FW_TCPOPT_TS;
333 return (flags_match(cmd, bits));
337 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
339 if (ifp == NULL) /* no iface with this packet, match fails */
341 /* Check by name or by IP address */
342 if (cmd->name[0] != '\0') { /* match by name */
345 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
348 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
352 #ifdef __FreeBSD__ /* and OSX too ? */
356 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
357 if (ia->ifa_addr->sa_family != AF_INET)
359 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
360 (ia->ifa_addr))->sin_addr.s_addr) {
361 if_addr_runlock(ifp);
362 return(1); /* match */
365 if_addr_runlock(ifp);
366 #endif /* __FreeBSD__ */
368 return(0); /* no match, fail ... */
372 * The verify_path function checks if a route to the src exists and
373 * if it is reachable via ifp (when provided).
375 * The 'verrevpath' option checks that the interface that an IP packet
376 * arrives on is the same interface that traffic destined for the
377 * packet's source address would be routed out of.
378 * The 'versrcreach' option just checks that the source address is
379 * reachable via any route (except default) in the routing table.
380 * These two are a measure to block forged packets. This is also
381 * commonly known as "anti-spoofing" or Unicast Reverse Path
382 * Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
383 * is purposely reminiscent of the Cisco IOS command,
385 * ip verify unicast reverse-path
386 * ip verify unicast source reachable-via any
388 * which implements the same functionality. But note that the syntax
389 * is misleading, and the check may be performed on all IP packets
390 * whether unicast, multicast, or broadcast.
393 verify_path(struct in_addr src, struct ifnet *ifp, u_int fib)
399 struct sockaddr_in *dst;
401 bzero(&ro, sizeof(ro));
403 dst = (struct sockaddr_in *)&(ro.ro_dst);
404 dst->sin_family = AF_INET;
405 dst->sin_len = sizeof(*dst);
407 in_rtalloc_ign(&ro, 0, fib);
409 if (ro.ro_rt == NULL)
413 * If ifp is provided, check for equality with rtentry.
414 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
415 * in order to pass packets injected back by if_simloop():
416 * if useloopback == 1 routing entry (via lo0) for our own address
417 * may exist, so we need to handle routing assymetry.
419 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
424 /* if no ifp provided, check if rtentry is not default route */
426 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
431 /* or if this is a blackhole/reject route */
432 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
437 /* found valid route */
440 #endif /* __FreeBSD__ */
445 * ipv6 specific rules here...
448 icmp6type_match (int type, ipfw_insn_u32 *cmd)
450 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
454 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
457 for (i=0; i <= cmd->o.arg1; ++i )
458 if (curr_flow == cmd->d[i] )
463 /* support for IP6_*_ME opcodes */
465 search_ip6_addr_net (struct in6_addr * ip6_addr)
469 struct in6_ifaddr *fdm;
470 struct in6_addr copia;
472 TAILQ_FOREACH(mdc, &V_ifnet, if_link) {
474 TAILQ_FOREACH(mdc2, &mdc->if_addrhead, ifa_link) {
475 if (mdc2->ifa_addr->sa_family == AF_INET6) {
476 fdm = (struct in6_ifaddr *)mdc2;
477 copia = fdm->ia_addr.sin6_addr;
478 /* need for leaving scope_id in the sock_addr */
479 in6_clearscope(&copia);
480 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia)) {
481 if_addr_runlock(mdc);
486 if_addr_runlock(mdc);
492 verify_path6(struct in6_addr *src, struct ifnet *ifp)
495 struct sockaddr_in6 *dst;
497 bzero(&ro, sizeof(ro));
499 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
500 dst->sin6_family = AF_INET6;
501 dst->sin6_len = sizeof(*dst);
502 dst->sin6_addr = *src;
503 /* XXX MRT 0 for ipv6 at this time */
504 rtalloc_ign((struct route *)&ro, 0);
506 if (ro.ro_rt == NULL)
510 * if ifp is provided, check for equality with rtentry
511 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
512 * to support the case of sending packets to an address of our own.
513 * (where the former interface is the first argument of if_simloop()
514 * (=ifp), the latter is lo0)
516 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
521 /* if no ifp provided, check if rtentry is not default route */
523 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
528 /* or if this is a blackhole/reject route */
529 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
534 /* found valid route */
541 is_icmp6_query(int icmp6_type)
543 if ((icmp6_type <= ICMP6_MAXTYPE) &&
544 (icmp6_type == ICMP6_ECHO_REQUEST ||
545 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
546 icmp6_type == ICMP6_WRUREQUEST ||
547 icmp6_type == ICMP6_FQDN_QUERY ||
548 icmp6_type == ICMP6_NI_QUERY))
555 send_reject6(struct ip_fw_args *args, int code, u_int hlen, struct ip6_hdr *ip6)
560 if (code == ICMP6_UNREACH_RST && args->f_id.proto == IPPROTO_TCP) {
562 tcp = (struct tcphdr *)((char *)ip6 + hlen);
564 if ((tcp->th_flags & TH_RST) == 0) {
566 m0 = ipfw_send_pkt(args->m, &(args->f_id),
567 ntohl(tcp->th_seq), ntohl(tcp->th_ack),
568 tcp->th_flags | TH_RST);
570 ip6_output(m0, NULL, NULL, 0, NULL, NULL,
574 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
577 * Unlike above, the mbufs need to line up with the ip6 hdr,
578 * as the contents are read. We need to m_adj() the
580 * The mbuf will however be thrown away so we can adjust it.
581 * Remember we did an m_pullup on it already so we
582 * can make some assumptions about contiguousness.
585 m_adj(m, args->L3offset);
587 icmp6_error(m, ICMP6_DST_UNREACH, code, 0);
598 * sends a reject message, consuming the mbuf passed as an argument.
601 send_reject(struct ip_fw_args *args, int code, int iplen, struct ip *ip)
605 /* XXX When ip is not guaranteed to be at mtod() we will
606 * need to account for this */
607 * The mbuf will however be thrown away so we can adjust it.
608 * Remember we did an m_pullup on it already so we
609 * can make some assumptions about contiguousness.
612 m_adj(m, args->L3offset);
614 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
615 /* We need the IP header in host order for icmp_error(). */
617 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
618 } else if (args->f_id.proto == IPPROTO_TCP) {
619 struct tcphdr *const tcp =
620 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
621 if ( (tcp->th_flags & TH_RST) == 0) {
623 m = ipfw_send_pkt(args->m, &(args->f_id),
624 ntohl(tcp->th_seq), ntohl(tcp->th_ack),
625 tcp->th_flags | TH_RST);
627 ip_output(m, NULL, NULL, 0, NULL, NULL);
636 * Support for uid/gid/jail lookup. These tests are expensive
637 * (because we may need to look into the list of active sockets)
638 * so we cache the results. ugid_lookupp is 0 if we have not
639 * yet done a lookup, 1 if we succeeded, and -1 if we tried
640 * and failed. The function always returns the match value.
641 * We could actually spare the variable and use *uc, setting
642 * it to '(void *)check_uidgid if we have no info, NULL if
643 * we tried and failed, or any other value if successful.
646 check_uidgid(ipfw_insn_u32 *insn, int proto, struct ifnet *oif,
647 struct in_addr dst_ip, u_int16_t dst_port, struct in_addr src_ip,
648 u_int16_t src_port, int *ugid_lookupp,
649 struct ucred **uc, struct inpcb *inp)
652 return cred_check(insn, proto, oif,
653 dst_ip, dst_port, src_ip, src_port,
654 (struct bsd_ucred *)uc, ugid_lookupp, ((struct mbuf *)inp)->m_skb);
656 struct inpcbinfo *pi;
662 * Check to see if the UDP or TCP stack supplied us with
663 * the PCB. If so, rather then holding a lock and looking
664 * up the PCB, we can use the one that was supplied.
666 if (inp && *ugid_lookupp == 0) {
667 INP_LOCK_ASSERT(inp);
668 if (inp->inp_socket != NULL) {
669 *uc = crhold(inp->inp_cred);
675 * If we have already been here and the packet has no
676 * PCB entry associated with it, then we can safely
677 * assume that this is a no match.
679 if (*ugid_lookupp == -1)
681 if (proto == IPPROTO_TCP) {
684 } else if (proto == IPPROTO_UDP) {
685 wildcard = INPLOOKUP_WILDCARD;
690 if (*ugid_lookupp == 0) {
693 in_pcblookup_hash(pi,
694 dst_ip, htons(dst_port),
695 src_ip, htons(src_port),
697 in_pcblookup_hash(pi,
698 src_ip, htons(src_port),
699 dst_ip, htons(dst_port),
702 *uc = crhold(pcb->inp_cred);
705 INP_INFO_RUNLOCK(pi);
706 if (*ugid_lookupp == 0) {
708 * We tried and failed, set the variable to -1
709 * so we will not try again on this packet.
715 if (insn->o.opcode == O_UID)
716 match = ((*uc)->cr_uid == (uid_t)insn->d[0]);
717 else if (insn->o.opcode == O_GID)
718 match = groupmember((gid_t)insn->d[0], *uc);
719 else if (insn->o.opcode == O_JAIL)
720 match = ((*uc)->cr_prison->pr_id == (int)insn->d[0]);
722 #endif /* __FreeBSD__ */
726 * Helper function to set args with info on the rule after the matching
727 * one. slot is precise, whereas we guess rule_id as they are
728 * assigned sequentially.
731 set_match(struct ip_fw_args *args, int slot,
732 struct ip_fw_chain *chain)
734 args->rule.chain_id = chain->id;
735 args->rule.slot = slot + 1; /* we use 0 as a marker */
736 args->rule.rule_id = 1 + chain->map[slot]->id;
737 args->rule.rulenum = chain->map[slot]->rulenum;
741 * The main check routine for the firewall.
743 * All arguments are in args so we can modify them and return them
744 * back to the caller.
748 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
749 * Starts with the IP header.
750 * args->eh (in) Mac header if present, NULL for layer3 packet.
751 * args->L3offset Number of bytes bypassed if we came from L2.
752 * e.g. often sizeof(eh) ** NOTYET **
753 * args->oif Outgoing interface, NULL if packet is incoming.
754 * The incoming interface is in the mbuf. (in)
755 * args->divert_rule (in/out)
756 * Skip up to the first rule past this rule number;
757 * upon return, non-zero port number for divert or tee.
759 * args->rule Pointer to the last matching rule (in/out)
760 * args->next_hop Socket we are forwarding to (out).
761 * args->f_id Addresses grabbed from the packet (out)
762 * args->rule.info a cookie depending on rule action
766 * IP_FW_PASS the packet must be accepted
767 * IP_FW_DENY the packet must be dropped
768 * IP_FW_DIVERT divert packet, port in m_tag
769 * IP_FW_TEE tee packet, port in m_tag
770 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
771 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
772 * args->rule contains the matching rule,
773 * args->rule.info has additional information.
777 ipfw_chk(struct ip_fw_args *args)
781 * Local variables holding state while processing a packet:
783 * IMPORTANT NOTE: to speed up the processing of rules, there
784 * are some assumption on the values of the variables, which
785 * are documented here. Should you change them, please check
786 * the implementation of the various instructions to make sure
787 * that they still work.
789 * args->eh The MAC header. It is non-null for a layer2
790 * packet, it is NULL for a layer-3 packet.
792 * args->L3offset Offset in the packet to the L3 (IP or equiv.) header.
794 * m | args->m Pointer to the mbuf, as received from the caller.
795 * It may change if ipfw_chk() does an m_pullup, or if it
796 * consumes the packet because it calls send_reject().
797 * XXX This has to change, so that ipfw_chk() never modifies
798 * or consumes the buffer.
799 * ip is the beginning of the ip(4 or 6) header.
800 * Calculated by adding the L3offset to the start of data.
801 * (Until we start using L3offset, the packet is
802 * supposed to start with the ip header).
804 struct mbuf *m = args->m;
805 struct ip *ip = mtod(m, struct ip *);
808 * For rules which contain uid/gid or jail constraints, cache
809 * a copy of the users credentials after the pcb lookup has been
810 * executed. This will speed up the processing of rules with
811 * these types of constraints, as well as decrease contention
812 * on pcb related locks.
815 struct bsd_ucred ucred_cache;
817 struct ucred *ucred_cache = NULL;
819 int ucred_lookup = 0;
822 * oif | args->oif If NULL, ipfw_chk has been called on the
823 * inbound path (ether_input, ip_input).
824 * If non-NULL, ipfw_chk has been called on the outbound path
825 * (ether_output, ip_output).
827 struct ifnet *oif = args->oif;
829 int f_pos = 0; /* index of current rule in the array */
833 * hlen The length of the IP header.
835 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
838 * offset The offset of a fragment. offset != 0 means that
839 * we have a fragment at this offset of an IPv4 packet.
840 * offset == 0 means that (if this is an IPv4 packet)
841 * this is the first or only fragment.
842 * For IPv6 offset == 0 means there is no Fragment Header.
843 * If offset != 0 for IPv6 always use correct mask to
844 * get the correct offset because we add IP6F_MORE_FRAG
845 * to be able to dectect the first fragment which would
846 * otherwise have offset = 0.
851 * Local copies of addresses. They are only valid if we have
854 * proto The protocol. Set to 0 for non-ip packets,
855 * or to the protocol read from the packet otherwise.
856 * proto != 0 means that we have an IPv4 packet.
858 * src_port, dst_port port numbers, in HOST format. Only
859 * valid for TCP and UDP packets.
861 * src_ip, dst_ip ip addresses, in NETWORK format.
862 * Only valid for IPv4 packets.
865 uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */
866 struct in_addr src_ip, dst_ip; /* NOTE: network format */
869 uint16_t etype = 0; /* Host order stored ether type */
872 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
873 * MATCH_NONE when checked and not matched (q = NULL),
874 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
876 int dyn_dir = MATCH_UNKNOWN;
877 ipfw_dyn_rule *q = NULL;
878 struct ip_fw_chain *chain = &V_layer3_chain;
881 * We store in ulp a pointer to the upper layer protocol header.
882 * In the ipv4 case this is easy to determine from the header,
883 * but for ipv6 we might have some additional headers in the middle.
884 * ulp is NULL if not found.
886 void *ulp = NULL; /* upper layer protocol pointer. */
888 /* XXX ipv6 variables */
890 uint8_t icmp6_type = 0;
891 uint16_t ext_hd = 0; /* bits vector for extension header filtering */
892 /* end of ipv6 variables */
896 int done = 0; /* flag to exit the outer loop */
898 if (m->m_flags & M_SKIP_FIREWALL || (! V_ipfw_vnet_ready))
899 return (IP_FW_PASS); /* accept */
901 dst_ip.s_addr = 0; /* make sure it is initialized */
902 src_ip.s_addr = 0; /* make sure it is initialized */
903 pktlen = m->m_pkthdr.len;
904 args->f_id.fib = M_GETFIB(m); /* note mbuf not altered) */
905 proto = args->f_id.proto = 0; /* mark f_id invalid */
906 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
909 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
910 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
911 * pointer might become stale after other pullups (but we never use it
914 #define PULLUP_TO(_len, p, T) \
916 int x = (_len) + sizeof(T); \
917 if ((m)->m_len < x) { \
918 args->m = m = m_pullup(m, x); \
920 goto pullup_failed; \
922 p = (mtod(m, char *) + (_len)); \
926 * if we have an ether header,
929 etype = ntohs(args->eh->ether_type);
931 /* Identify IP packets and fill up variables. */
932 if (pktlen >= sizeof(struct ip6_hdr) &&
933 (args->eh == NULL || etype == ETHERTYPE_IPV6) && ip->ip_v == 6) {
934 struct ip6_hdr *ip6 = (struct ip6_hdr *)ip;
936 args->f_id.addr_type = 6;
937 hlen = sizeof(struct ip6_hdr);
938 proto = ip6->ip6_nxt;
940 /* Search extension headers to find upper layer protocols */
941 while (ulp == NULL) {
944 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
945 icmp6_type = ICMP6(ulp)->icmp6_type;
949 PULLUP_TO(hlen, ulp, struct tcphdr);
950 dst_port = TCP(ulp)->th_dport;
951 src_port = TCP(ulp)->th_sport;
952 /* save flags for dynamic rules */
953 args->f_id._flags = TCP(ulp)->th_flags;
957 PULLUP_TO(hlen, ulp, struct sctphdr);
958 src_port = SCTP(ulp)->src_port;
959 dst_port = SCTP(ulp)->dest_port;
963 PULLUP_TO(hlen, ulp, struct udphdr);
964 dst_port = UDP(ulp)->uh_dport;
965 src_port = UDP(ulp)->uh_sport;
968 case IPPROTO_HOPOPTS: /* RFC 2460 */
969 PULLUP_TO(hlen, ulp, struct ip6_hbh);
970 ext_hd |= EXT_HOPOPTS;
971 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
972 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
976 case IPPROTO_ROUTING: /* RFC 2460 */
977 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
978 switch (((struct ip6_rthdr *)ulp)->ip6r_type) {
980 ext_hd |= EXT_RTHDR0;
983 ext_hd |= EXT_RTHDR2;
986 printf("IPFW2: IPV6 - Unknown Routing "
988 ((struct ip6_rthdr *)ulp)->ip6r_type);
989 if (V_fw_deny_unknown_exthdrs)
993 ext_hd |= EXT_ROUTING;
994 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
995 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
999 case IPPROTO_FRAGMENT: /* RFC 2460 */
1000 PULLUP_TO(hlen, ulp, struct ip6_frag);
1001 ext_hd |= EXT_FRAGMENT;
1002 hlen += sizeof (struct ip6_frag);
1003 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
1004 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
1006 /* Add IP6F_MORE_FRAG for offset of first
1007 * fragment to be != 0. */
1008 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
1011 printf("IPFW2: IPV6 - Invalid Fragment "
1013 if (V_fw_deny_unknown_exthdrs)
1014 return (IP_FW_DENY);
1018 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
1022 case IPPROTO_DSTOPTS: /* RFC 2460 */
1023 PULLUP_TO(hlen, ulp, struct ip6_hbh);
1024 ext_hd |= EXT_DSTOPTS;
1025 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
1026 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
1030 case IPPROTO_AH: /* RFC 2402 */
1031 PULLUP_TO(hlen, ulp, struct ip6_ext);
1033 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
1034 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
1038 case IPPROTO_ESP: /* RFC 2406 */
1039 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
1040 /* Anything past Seq# is variable length and
1041 * data past this ext. header is encrypted. */
1045 case IPPROTO_NONE: /* RFC 2460 */
1047 * Packet ends here, and IPv6 header has
1048 * already been pulled up. If ip6e_len!=0
1049 * then octets must be ignored.
1051 ulp = ip; /* non-NULL to get out of loop. */
1054 case IPPROTO_OSPFIGP:
1055 /* XXX OSPF header check? */
1056 PULLUP_TO(hlen, ulp, struct ip6_ext);
1060 /* XXX PIM header check? */
1061 PULLUP_TO(hlen, ulp, struct pim);
1065 PULLUP_TO(hlen, ulp, struct carp_header);
1066 if (((struct carp_header *)ulp)->carp_version !=
1068 return (IP_FW_DENY);
1069 if (((struct carp_header *)ulp)->carp_type !=
1071 return (IP_FW_DENY);
1074 case IPPROTO_IPV6: /* RFC 2893 */
1075 PULLUP_TO(hlen, ulp, struct ip6_hdr);
1078 case IPPROTO_IPV4: /* RFC 2893 */
1079 PULLUP_TO(hlen, ulp, struct ip);
1083 printf("IPFW2: IPV6 - Unknown Extension "
1084 "Header(%d), ext_hd=%x\n", proto, ext_hd);
1085 if (V_fw_deny_unknown_exthdrs)
1086 return (IP_FW_DENY);
1087 PULLUP_TO(hlen, ulp, struct ip6_ext);
1091 ip = mtod(m, struct ip *);
1092 ip6 = (struct ip6_hdr *)ip;
1093 args->f_id.src_ip6 = ip6->ip6_src;
1094 args->f_id.dst_ip6 = ip6->ip6_dst;
1095 args->f_id.src_ip = 0;
1096 args->f_id.dst_ip = 0;
1097 args->f_id.flow_id6 = ntohl(ip6->ip6_flow);
1098 } else if (pktlen >= sizeof(struct ip) &&
1099 (args->eh == NULL || etype == ETHERTYPE_IP) && ip->ip_v == 4) {
1101 hlen = ip->ip_hl << 2;
1102 args->f_id.addr_type = 4;
1105 * Collect parameters into local variables for faster matching.
1108 src_ip = ip->ip_src;
1109 dst_ip = ip->ip_dst;
1110 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1111 iplen = ntohs(ip->ip_len);
1112 pktlen = iplen < pktlen ? iplen : pktlen;
1117 PULLUP_TO(hlen, ulp, struct tcphdr);
1118 dst_port = TCP(ulp)->th_dport;
1119 src_port = TCP(ulp)->th_sport;
1120 /* save flags for dynamic rules */
1121 args->f_id._flags = TCP(ulp)->th_flags;
1125 PULLUP_TO(hlen, ulp, struct udphdr);
1126 dst_port = UDP(ulp)->uh_dport;
1127 src_port = UDP(ulp)->uh_sport;
1131 PULLUP_TO(hlen, ulp, struct icmphdr);
1132 //args->f_id.flags = ICMP(ulp)->icmp_type;
1140 ip = mtod(m, struct ip *);
1141 args->f_id.src_ip = ntohl(src_ip.s_addr);
1142 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1145 if (proto) { /* we may have port numbers, store them */
1146 args->f_id.proto = proto;
1147 args->f_id.src_port = src_port = ntohs(src_port);
1148 args->f_id.dst_port = dst_port = ntohs(dst_port);
1152 if (! V_ipfw_vnet_ready) { /* shutting down, leave NOW. */
1153 IPFW_RUNLOCK(chain);
1154 return (IP_FW_PASS); /* accept */
1156 if (args->rule.slot) {
1158 * Packet has already been tagged as a result of a previous
1159 * match on rule args->rule aka args->rule_id (PIPE, QUEUE,
1160 * REASS, NETGRAPH, DIVERT/TEE...)
1161 * Validate the slot and continue from the next one
1162 * if still present, otherwise do a lookup.
1164 f_pos = (args->rule.chain_id == chain->id) ?
1166 ipfw_find_rule(chain, args->rule.rulenum,
1167 args->rule.rule_id);
1173 * Now scan the rules, and parse microinstructions for each rule.
1174 * We have two nested loops and an inner switch. Sometimes we
1175 * need to break out of one or both loops, or re-enter one of
1176 * the loops with updated variables. Loop variables are:
1178 * f_pos (outer loop) points to the current rule.
1179 * On output it points to the matching rule.
1180 * done (outer loop) is used as a flag to break the loop.
1181 * l (inner loop) residual length of current rule.
1182 * cmd points to the current microinstruction.
1184 * We break the inner loop by setting l=0 and possibly
1185 * cmdlen=0 if we don't want to advance cmd.
1186 * We break the outer loop by setting done=1
1187 * We can restart the inner loop by setting l>0 and f_pos, f, cmd
1190 for (; f_pos < chain->n_rules; f_pos++) {
1192 uint32_t tablearg = 0;
1193 int l, cmdlen, skip_or; /* skip rest of OR block */
1196 f = chain->map[f_pos];
1197 if (V_set_disable & (1 << f->set) )
1201 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1202 l -= cmdlen, cmd += cmdlen) {
1206 * check_body is a jump target used when we find a
1207 * CHECK_STATE, and need to jump to the body of
1212 cmdlen = F_LEN(cmd);
1214 * An OR block (insn_1 || .. || insn_n) has the
1215 * F_OR bit set in all but the last instruction.
1216 * The first match will set "skip_or", and cause
1217 * the following instructions to be skipped until
1218 * past the one with the F_OR bit clear.
1220 if (skip_or) { /* skip this instruction */
1221 if ((cmd->len & F_OR) == 0)
1222 skip_or = 0; /* next one is good */
1225 match = 0; /* set to 1 if we succeed */
1227 switch (cmd->opcode) {
1229 * The first set of opcodes compares the packet's
1230 * fields with some pattern, setting 'match' if a
1231 * match is found. At the end of the loop there is
1232 * logic to deal with F_NOT and F_OR flags associated
1240 printf("ipfw: opcode %d unimplemented\n",
1248 * We only check offset == 0 && proto != 0,
1249 * as this ensures that we have a
1250 * packet with the ports info.
1254 if (is_ipv6) /* XXX to be fixed later */
1256 if (proto == IPPROTO_TCP ||
1257 proto == IPPROTO_UDP)
1258 match = check_uidgid(
1259 (ipfw_insn_u32 *)cmd,
1262 src_ip, src_port, &ucred_lookup,
1264 &ucred_cache, args->inp);
1266 (void *)&ucred_cache,
1267 (struct inpcb *)args->m);
1272 match = iface_match(m->m_pkthdr.rcvif,
1273 (ipfw_insn_if *)cmd);
1277 match = iface_match(oif, (ipfw_insn_if *)cmd);
1281 match = iface_match(oif ? oif :
1282 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1286 if (args->eh != NULL) { /* have MAC header */
1287 u_int32_t *want = (u_int32_t *)
1288 ((ipfw_insn_mac *)cmd)->addr;
1289 u_int32_t *mask = (u_int32_t *)
1290 ((ipfw_insn_mac *)cmd)->mask;
1291 u_int32_t *hdr = (u_int32_t *)args->eh;
1294 ( want[0] == (hdr[0] & mask[0]) &&
1295 want[1] == (hdr[1] & mask[1]) &&
1296 want[2] == (hdr[2] & mask[2]) );
1301 if (args->eh != NULL) {
1303 ((ipfw_insn_u16 *)cmd)->ports;
1306 for (i = cmdlen - 1; !match && i>0;
1308 match = (etype >= p[0] &&
1314 match = (offset != 0);
1317 case O_IN: /* "out" is "not in" */
1318 match = (oif == NULL);
1322 match = (args->eh != NULL);
1327 /* For diverted packets, args->rule.info
1328 * contains the divert port (in host format)
1329 * reason and direction.
1331 uint32_t i = args->rule.info;
1332 match = (i&IPFW_IS_MASK) == IPFW_IS_DIVERT &&
1333 cmd->arg1 & ((i & IPFW_INFO_IN) ? 1 : 2);
1339 * We do not allow an arg of 0 so the
1340 * check of "proto" only suffices.
1342 match = (proto == cmd->arg1);
1347 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1351 case O_IP_SRC_LOOKUP:
1352 case O_IP_DST_LOOKUP:
1355 (cmd->opcode == O_IP_DST_LOOKUP) ?
1356 dst_ip.s_addr : src_ip.s_addr;
1359 if (cmdlen > F_INSN_SIZE(ipfw_insn_u32)) {
1360 /* generic lookup. The key must be
1361 * in 32bit big-endian format.
1363 v = ((ipfw_insn_u32 *)cmd)->d[1];
1365 key = dst_ip.s_addr;
1367 key = src_ip.s_addr;
1368 else if (v == 6) /* dscp */
1369 key = (ip->ip_tos >> 2) & 0x3f;
1370 else if (offset != 0)
1372 else if (proto != IPPROTO_TCP &&
1373 proto != IPPROTO_UDP)
1376 key = htonl(dst_port);
1378 key = htonl(src_port);
1379 else if (v == 4 || v == 5) {
1381 (ipfw_insn_u32 *)cmd,
1384 src_ip, src_port, &ucred_lookup,
1386 &ucred_cache, args->inp);
1387 if (v == 4 /* O_UID */)
1388 key = ucred_cache->cr_uid;
1389 else if (v == 5 /* O_JAIL */)
1390 key = ucred_cache->cr_prison->pr_id;
1391 #else /* !__FreeBSD__ */
1392 (void *)&ucred_cache,
1393 (struct inpcb *)args->m);
1394 if (v ==4 /* O_UID */)
1395 key = ucred_cache.uid;
1396 else if (v == 5 /* O_JAIL */)
1397 key = ucred_cache.xid;
1398 #endif /* !__FreeBSD__ */
1403 match = ipfw_lookup_table(chain,
1404 cmd->arg1, key, &v);
1407 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
1409 ((ipfw_insn_u32 *)cmd)->d[0] == v;
1419 (cmd->opcode == O_IP_DST_MASK) ?
1420 dst_ip.s_addr : src_ip.s_addr;
1421 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
1424 for (; !match && i>0; i-= 2, p+= 2)
1425 match = (p[0] == (a & p[1]));
1433 INADDR_TO_IFP(src_ip, tif);
1434 match = (tif != NULL);
1440 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
1447 u_int32_t *d = (u_int32_t *)(cmd+1);
1449 cmd->opcode == O_IP_DST_SET ?
1455 addr -= d[0]; /* subtract base */
1456 match = (addr < cmd->arg1) &&
1457 ( d[ 1 + (addr>>5)] &
1458 (1<<(addr & 0x1f)) );
1464 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1472 INADDR_TO_IFP(dst_ip, tif);
1473 match = (tif != NULL);
1479 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
1487 * offset == 0 && proto != 0 is enough
1488 * to guarantee that we have a
1489 * packet with port info.
1491 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1494 (cmd->opcode == O_IP_SRCPORT) ?
1495 src_port : dst_port ;
1497 ((ipfw_insn_u16 *)cmd)->ports;
1500 for (i = cmdlen - 1; !match && i>0;
1502 match = (x>=p[0] && x<=p[1]);
1507 match = (offset == 0 && proto==IPPROTO_ICMP &&
1508 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
1513 match = is_ipv6 && offset == 0 &&
1514 proto==IPPROTO_ICMPV6 &&
1516 ICMP6(ulp)->icmp6_type,
1517 (ipfw_insn_u32 *)cmd);
1523 ipopts_match(ip, cmd) );
1528 cmd->arg1 == ip->ip_v);
1534 if (is_ipv4) { /* only for IP packets */
1539 if (cmd->opcode == O_IPLEN)
1541 else if (cmd->opcode == O_IPTTL)
1543 else /* must be IPID */
1544 x = ntohs(ip->ip_id);
1546 match = (cmd->arg1 == x);
1549 /* otherwise we have ranges */
1550 p = ((ipfw_insn_u16 *)cmd)->ports;
1552 for (; !match && i>0; i--, p += 2)
1553 match = (x >= p[0] && x <= p[1]);
1557 case O_IPPRECEDENCE:
1559 (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1564 flags_match(cmd, ip->ip_tos));
1568 if (proto == IPPROTO_TCP && offset == 0) {
1576 ((ip->ip_hl + tcp->th_off) << 2);
1578 match = (cmd->arg1 == x);
1581 /* otherwise we have ranges */
1582 p = ((ipfw_insn_u16 *)cmd)->ports;
1584 for (; !match && i>0; i--, p += 2)
1585 match = (x >= p[0] && x <= p[1]);
1590 match = (proto == IPPROTO_TCP && offset == 0 &&
1591 flags_match(cmd, TCP(ulp)->th_flags));
1595 match = (proto == IPPROTO_TCP && offset == 0 &&
1596 tcpopts_match(TCP(ulp), cmd));
1600 match = (proto == IPPROTO_TCP && offset == 0 &&
1601 ((ipfw_insn_u32 *)cmd)->d[0] ==
1606 match = (proto == IPPROTO_TCP && offset == 0 &&
1607 ((ipfw_insn_u32 *)cmd)->d[0] ==
1612 match = (proto == IPPROTO_TCP && offset == 0 &&
1613 cmd->arg1 == TCP(ulp)->th_win);
1617 /* reject packets which have SYN only */
1618 /* XXX should i also check for TH_ACK ? */
1619 match = (proto == IPPROTO_TCP && offset == 0 &&
1620 (TCP(ulp)->th_flags &
1621 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1626 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
1629 at = pf_find_mtag(m);
1630 if (at != NULL && at->qid != 0)
1632 at = pf_get_mtag(m);
1635 * Let the packet fall back to the
1640 at->qid = altq->qid;
1650 ipfw_log(f, hlen, args, m,
1651 oif, offset, tablearg, ip);
1656 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
1660 /* Outgoing packets automatically pass/match */
1661 match = ((oif != NULL) ||
1662 (m->m_pkthdr.rcvif == NULL) ||
1666 verify_path6(&(args->f_id.src_ip6),
1667 m->m_pkthdr.rcvif) :
1669 verify_path(src_ip, m->m_pkthdr.rcvif,
1674 /* Outgoing packets automatically pass/match */
1675 match = (hlen > 0 && ((oif != NULL) ||
1678 verify_path6(&(args->f_id.src_ip6),
1681 verify_path(src_ip, NULL, args->f_id.fib)));
1685 /* Outgoing packets automatically pass/match */
1686 if (oif == NULL && hlen > 0 &&
1687 ( (is_ipv4 && in_localaddr(src_ip))
1690 in6_localaddr(&(args->f_id.src_ip6)))
1695 is_ipv6 ? verify_path6(
1696 &(args->f_id.src_ip6),
1697 m->m_pkthdr.rcvif) :
1708 match = (m_tag_find(m,
1709 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
1711 /* otherwise no match */
1717 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
1718 &((ipfw_insn_ip6 *)cmd)->addr6);
1723 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
1724 &((ipfw_insn_ip6 *)cmd)->addr6);
1726 case O_IP6_SRC_MASK:
1727 case O_IP6_DST_MASK:
1731 struct in6_addr *d =
1732 &((ipfw_insn_ip6 *)cmd)->addr6;
1734 for (; !match && i > 0; d += 2,
1735 i -= F_INSN_SIZE(struct in6_addr)
1741 APPLY_MASK(&p, &d[1]);
1743 IN6_ARE_ADDR_EQUAL(&d[0],
1751 flow6id_match(args->f_id.flow_id6,
1752 (ipfw_insn_u32 *) cmd);
1757 (ext_hd & ((ipfw_insn *) cmd)->arg1);
1771 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
1772 tablearg : cmd->arg1;
1774 /* Packet is already tagged with this tag? */
1775 mtag = m_tag_locate(m, MTAG_IPFW, tag, NULL);
1777 /* We have `untag' action when F_NOT flag is
1778 * present. And we must remove this mtag from
1779 * mbuf and reset `match' to zero (`match' will
1780 * be inversed later).
1781 * Otherwise we should allocate new mtag and
1782 * push it into mbuf.
1784 if (cmd->len & F_NOT) { /* `untag' action */
1786 m_tag_delete(m, mtag);
1788 } else if (mtag == NULL) {
1789 if ((mtag = m_tag_alloc(MTAG_IPFW,
1790 tag, 0, M_NOWAIT)) != NULL)
1791 m_tag_prepend(m, mtag);
1797 case O_FIB: /* try match the specified fib */
1798 if (args->f_id.fib == cmd->arg1)
1804 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
1805 tablearg : cmd->arg1;
1808 match = m_tag_locate(m, MTAG_IPFW,
1813 /* we have ranges */
1814 for (mtag = m_tag_first(m);
1815 mtag != NULL && !match;
1816 mtag = m_tag_next(m, mtag)) {
1820 if (mtag->m_tag_cookie != MTAG_IPFW)
1823 p = ((ipfw_insn_u16 *)cmd)->ports;
1825 for(; !match && i > 0; i--, p += 2)
1827 mtag->m_tag_id >= p[0] &&
1828 mtag->m_tag_id <= p[1];
1834 * The second set of opcodes represents 'actions',
1835 * i.e. the terminal part of a rule once the packet
1836 * matches all previous patterns.
1837 * Typically there is only one action for each rule,
1838 * and the opcode is stored at the end of the rule
1839 * (but there are exceptions -- see below).
1841 * In general, here we set retval and terminate the
1842 * outer loop (would be a 'break 3' in some language,
1843 * but we need to set l=0, done=1)
1846 * O_COUNT and O_SKIPTO actions:
1847 * instead of terminating, we jump to the next rule
1848 * (setting l=0), or to the SKIPTO target (setting
1849 * f/f_len, cmd and l as needed), respectively.
1851 * O_TAG, O_LOG and O_ALTQ action parameters:
1852 * perform some action and set match = 1;
1854 * O_LIMIT and O_KEEP_STATE: these opcodes are
1855 * not real 'actions', and are stored right
1856 * before the 'action' part of the rule.
1857 * These opcodes try to install an entry in the
1858 * state tables; if successful, we continue with
1859 * the next opcode (match=1; break;), otherwise
1860 * the packet must be dropped (set retval,
1861 * break loops with l=0, done=1)
1863 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1864 * cause a lookup of the state table, and a jump
1865 * to the 'action' part of the parent rule
1866 * if an entry is found, or
1867 * (CHECK_STATE only) a jump to the next rule if
1868 * the entry is not found.
1869 * The result of the lookup is cached so that
1870 * further instances of these opcodes become NOPs.
1871 * The jump to the next rule is done by setting
1876 if (ipfw_install_state(f,
1877 (ipfw_insn_limit *)cmd, args, tablearg)) {
1878 /* error or limit violation */
1879 retval = IP_FW_DENY;
1880 l = 0; /* exit inner loop */
1881 done = 1; /* exit outer loop */
1889 * dynamic rules are checked at the first
1890 * keep-state or check-state occurrence,
1891 * with the result being stored in dyn_dir.
1892 * The compiler introduces a PROBE_STATE
1893 * instruction for us when we have a
1894 * KEEP_STATE (because PROBE_STATE needs
1897 if (dyn_dir == MATCH_UNKNOWN &&
1898 (q = ipfw_lookup_dyn_rule(&args->f_id,
1899 &dyn_dir, proto == IPPROTO_TCP ?
1903 * Found dynamic entry, update stats
1904 * and jump to the 'action' part of
1905 * the parent rule by setting
1906 * f, cmd, l and clearing cmdlen.
1910 /* XXX we would like to have f_pos
1911 * readily accessible in the dynamic
1912 * rule, instead of having to
1916 f_pos = ipfw_find_rule(chain,
1918 cmd = ACTION_PTR(f);
1919 l = f->cmd_len - f->act_ofs;
1926 * Dynamic entry not found. If CHECK_STATE,
1927 * skip to next rule, if PROBE_STATE just
1928 * ignore and continue with next opcode.
1930 if (cmd->opcode == O_CHECK_STATE)
1931 l = 0; /* exit inner loop */
1936 retval = 0; /* accept */
1937 l = 0; /* exit inner loop */
1938 done = 1; /* exit outer loop */
1943 set_match(args, f_pos, chain);
1944 args->rule.info = (cmd->arg1 == IP_FW_TABLEARG) ?
1945 tablearg : cmd->arg1;
1946 if (cmd->opcode == O_PIPE)
1947 args->rule.info |= IPFW_IS_PIPE;
1949 args->rule.info |= IPFW_ONEPASS;
1950 retval = IP_FW_DUMMYNET;
1951 l = 0; /* exit inner loop */
1952 done = 1; /* exit outer loop */
1957 if (args->eh) /* not on layer 2 */
1959 /* otherwise this is terminal */
1960 l = 0; /* exit inner loop */
1961 done = 1; /* exit outer loop */
1962 retval = (cmd->opcode == O_DIVERT) ?
1963 IP_FW_DIVERT : IP_FW_TEE;
1964 set_match(args, f_pos, chain);
1965 args->rule.info = (cmd->arg1 == IP_FW_TABLEARG) ?
1966 tablearg : cmd->arg1;
1970 f->pcnt++; /* update stats */
1972 f->timestamp = time_uptime;
1973 l = 0; /* exit inner loop */
1977 f->pcnt++; /* update stats */
1979 f->timestamp = time_uptime;
1980 /* If possible use cached f_pos (in f->next_rule),
1981 * whose version is written in f->next_rule
1982 * (horrible hacks to avoid changing the ABI).
1984 if (cmd->arg1 != IP_FW_TABLEARG &&
1985 (uintptr_t)f->x_next == chain->id) {
1986 f_pos = (uintptr_t)f->next_rule;
1988 int i = (cmd->arg1 == IP_FW_TABLEARG) ?
1989 tablearg : cmd->arg1;
1990 /* make sure we do not jump backward */
1991 if (i <= f->rulenum)
1993 f_pos = ipfw_find_rule(chain, i, 0);
1994 /* update the cache */
1995 if (cmd->arg1 != IP_FW_TABLEARG) {
1997 (void *)(uintptr_t)f_pos;
1999 (void *)(uintptr_t)chain->id;
2003 * Skip disabled rules, and re-enter
2004 * the inner loop with the correct
2005 * f_pos, f, l and cmd.
2006 * Also clear cmdlen and skip_or
2008 for (; f_pos < chain->n_rules - 1 &&
2010 (1 << chain->map[f_pos]->set));
2013 /* Re-enter the inner loop at the skipto rule. */
2014 f = chain->map[f_pos];
2021 break; /* not reached */
2025 * Drop the packet and send a reject notice
2026 * if the packet is not ICMP (or is an ICMP
2027 * query), and it is not multicast/broadcast.
2029 if (hlen > 0 && is_ipv4 && offset == 0 &&
2030 (proto != IPPROTO_ICMP ||
2031 is_icmp_query(ICMP(ulp))) &&
2032 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2033 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2034 send_reject(args, cmd->arg1, iplen, ip);
2040 if (hlen > 0 && is_ipv6 &&
2041 ((offset & IP6F_OFF_MASK) == 0) &&
2042 (proto != IPPROTO_ICMPV6 ||
2043 (is_icmp6_query(icmp6_type) == 1)) &&
2044 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2045 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
2047 args, cmd->arg1, hlen,
2048 (struct ip6_hdr *)ip);
2054 retval = IP_FW_DENY;
2055 l = 0; /* exit inner loop */
2056 done = 1; /* exit outer loop */
2060 if (args->eh) /* not valid on layer2 pkts */
2062 if (!q || dyn_dir == MATCH_FORWARD) {
2063 struct sockaddr_in *sa;
2064 sa = &(((ipfw_insn_sa *)cmd)->sa);
2065 if (sa->sin_addr.s_addr == INADDR_ANY) {
2066 bcopy(sa, &args->hopstore,
2068 args->hopstore.sin_addr.s_addr =
2070 args->next_hop = &args->hopstore;
2072 args->next_hop = sa;
2075 retval = IP_FW_PASS;
2076 l = 0; /* exit inner loop */
2077 done = 1; /* exit outer loop */
2082 set_match(args, f_pos, chain);
2083 args->rule.info = (cmd->arg1 == IP_FW_TABLEARG) ?
2084 tablearg : cmd->arg1;
2086 args->rule.info |= IPFW_ONEPASS;
2087 retval = (cmd->opcode == O_NETGRAPH) ?
2088 IP_FW_NETGRAPH : IP_FW_NGTEE;
2089 l = 0; /* exit inner loop */
2090 done = 1; /* exit outer loop */
2094 f->pcnt++; /* update stats */
2096 f->timestamp = time_uptime;
2097 M_SETFIB(m, cmd->arg1);
2098 args->f_id.fib = cmd->arg1;
2099 l = 0; /* exit inner loop */
2103 if (!IPFW_NAT_LOADED) {
2104 retval = IP_FW_DENY;
2109 set_match(args, f_pos, chain);
2110 t = ((ipfw_insn_nat *)cmd)->nat;
2112 nat_id = (cmd->arg1 == IP_FW_TABLEARG) ?
2113 tablearg : cmd->arg1;
2114 t = (*lookup_nat_ptr)(&chain->nat, nat_id);
2117 retval = IP_FW_DENY;
2118 l = 0; /* exit inner loop */
2119 done = 1; /* exit outer loop */
2122 if (cmd->arg1 != IP_FW_TABLEARG)
2123 ((ipfw_insn_nat *)cmd)->nat = t;
2125 retval = ipfw_nat_ptr(args, t, m);
2127 l = 0; /* exit inner loop */
2128 done = 1; /* exit outer loop */
2136 l = 0; /* in any case exit inner loop */
2137 ip_off = ntohs(ip->ip_off);
2139 /* if not fragmented, go to next rule */
2140 if ((ip_off & (IP_MF | IP_OFFMASK)) == 0)
2143 * ip_reass() expects len & off in host
2148 args->m = m = ip_reass(m);
2151 * do IP header checksum fixup.
2153 if (m == NULL) { /* fragment got swallowed */
2154 retval = IP_FW_DENY;
2155 } else { /* good, packet complete */
2158 ip = mtod(m, struct ip *);
2159 hlen = ip->ip_hl << 2;
2162 if (hlen == sizeof(struct ip))
2163 ip->ip_sum = in_cksum_hdr(ip);
2165 ip->ip_sum = in_cksum(m, hlen);
2166 retval = IP_FW_REASS;
2167 set_match(args, f_pos, chain);
2169 done = 1; /* exit outer loop */
2174 panic("-- unknown opcode %d\n", cmd->opcode);
2175 } /* end of switch() on opcodes */
2177 * if we get here with l=0, then match is irrelevant.
2180 if (cmd->len & F_NOT)
2184 if (cmd->len & F_OR)
2187 if (!(cmd->len & F_OR)) /* not an OR block, */
2188 break; /* try next rule */
2191 } /* end of inner loop, scan opcodes */
2196 /* next_rule:; */ /* try next rule */
2198 } /* end of outer for, scan rules */
2201 struct ip_fw *rule = chain->map[f_pos];
2202 /* Update statistics */
2204 rule->bcnt += pktlen;
2205 rule->timestamp = time_uptime;
2207 retval = IP_FW_DENY;
2208 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
2210 IPFW_RUNLOCK(chain);
2212 if (ucred_cache != NULL)
2213 crfree(ucred_cache);
2219 printf("ipfw: pullup failed\n");
2220 return (IP_FW_DENY);
2224 * Module and VNET glue
2228 * Stuff that must be initialised only on boot or module load
2237 * Only print out this stuff the first time around,
2238 * when called from the sysinit code.
2244 "initialized, divert %s, nat %s, "
2245 "rule-based forwarding "
2246 #ifdef IPFIREWALL_FORWARD
2251 "default to %s, logging ",
2257 #ifdef IPFIREWALL_NAT
2262 default_to_accept ? "accept" : "deny");
2265 * Note: V_xxx variables can be accessed here but the vnet specific
2266 * initializer may not have been called yet for the VIMAGE case.
2267 * Tuneables will have been processed. We will print out values for
2269 * XXX This should all be rationalized AFTER 8.0
2271 if (V_fw_verbose == 0)
2272 printf("disabled\n");
2273 else if (V_verbose_limit == 0)
2274 printf("unlimited\n");
2276 printf("limited to %d packets/entry by default\n",
2279 ipfw_log_bpf(1); /* init */
2284 * Called for the removal of the last instance only on module unload.
2290 ipfw_log_bpf(0); /* uninit */
2292 printf("IP firewall unloaded\n");
2296 * Stuff that must be initialized for every instance
2297 * (including the first of course).
2300 vnet_ipfw_init(const void *unused)
2303 struct ip_fw *rule = NULL;
2304 struct ip_fw_chain *chain;
2306 chain = &V_layer3_chain;
2308 /* First set up some values that are compile time options */
2309 V_autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
2310 V_fw_deny_unknown_exthdrs = 1;
2311 #ifdef IPFIREWALL_VERBOSE
2314 #ifdef IPFIREWALL_VERBOSE_LIMIT
2315 V_verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2317 #ifdef IPFIREWALL_NAT
2318 LIST_INIT(&chain->nat);
2321 /* insert the default rule and create the initial map */
2323 chain->static_len = sizeof(struct ip_fw);
2324 chain->map = malloc(sizeof(struct ip_fw *), M_IPFW, M_NOWAIT | M_ZERO);
2326 rule = malloc(chain->static_len, M_IPFW, M_NOWAIT | M_ZERO);
2329 free(chain->map, M_IPFW);
2330 printf("ipfw2: ENOSPC initializing default rule "
2331 "(support disabled)\n");
2334 error = ipfw_init_tables(chain);
2336 panic("init_tables"); /* XXX Marko fix this ! */
2339 /* fill and insert the default rule */
2341 rule->rulenum = IPFW_DEFAULT_RULE;
2343 rule->set = RESVD_SET;
2344 rule->cmd[0].len = 1;
2345 rule->cmd[0].opcode = default_to_accept ? O_ACCEPT : O_DENY;
2346 chain->rules = chain->default_rule = chain->map[0] = rule;
2347 chain->id = rule->id = 1;
2349 IPFW_LOCK_INIT(chain);
2352 /* First set up some values that are compile time options */
2353 V_ipfw_vnet_ready = 1; /* Open for business */
2356 * Hook the sockopt handler, and the layer2 (V_ip_fw_chk_ptr)
2357 * and pfil hooks for ipv4 and ipv6. Even if the latter two fail
2358 * we still keep the module alive because the sockopt and
2359 * layer2 paths are still useful.
2360 * ipfw[6]_hook return 0 on success, ENOENT on failure,
2361 * so we can ignore the exact return value and just set a flag.
2363 * Note that V_fw[6]_enable are manipulated by a SYSCTL_PROC so
2364 * changes in the underlying (per-vnet) variables trigger
2365 * immediate hook()/unhook() calls.
2366 * In layer2 we have the same behaviour, except that V_ether_ipfw
2367 * is checked on each packet because there are no pfil hooks.
2369 V_ip_fw_ctl_ptr = ipfw_ctl;
2370 V_ip_fw_chk_ptr = ipfw_chk;
2371 error = ipfw_attach_hooks(1);
2376 * Called for the removal of each instance.
2379 vnet_ipfw_uninit(const void *unused)
2381 struct ip_fw *reap, *rule;
2382 struct ip_fw_chain *chain = &V_layer3_chain;
2385 V_ipfw_vnet_ready = 0; /* tell new callers to go away */
2387 * disconnect from ipv4, ipv6, layer2 and sockopt.
2388 * Then grab, release and grab again the WLOCK so we make
2389 * sure the update is propagated and nobody will be in.
2391 (void)ipfw_attach_hooks(0 /* detach */);
2392 V_ip_fw_chk_ptr = NULL;
2393 V_ip_fw_ctl_ptr = NULL;
2394 IPFW_UH_WLOCK(chain);
2395 IPFW_UH_WUNLOCK(chain);
2396 IPFW_UH_WLOCK(chain);
2399 IPFW_WUNLOCK(chain);
2402 ipfw_dyn_uninit(0); /* run the callout_drain */
2403 ipfw_destroy_tables(chain);
2405 for (i = 0; i < chain->n_rules; i++) {
2406 rule = chain->map[i];
2407 rule->x_next = reap;
2411 free(chain->map, M_IPFW);
2412 IPFW_WUNLOCK(chain);
2413 IPFW_UH_WUNLOCK(chain);
2415 ipfw_reap_rules(reap);
2416 IPFW_LOCK_DESTROY(chain);
2417 ipfw_dyn_uninit(1); /* free the remaining parts */
2422 * Module event handler.
2423 * In general we have the choice of handling most of these events by the
2424 * event handler or by the (VNET_)SYS(UN)INIT handlers. I have chosen to
2425 * use the SYSINIT handlers as they are more capable of expressing the
2426 * flow of control during module and vnet operations, so this is just
2427 * a skeleton. Note there is no SYSINIT equivalent of the module
2428 * SHUTDOWN handler, but we don't have anything to do in that case anyhow.
2431 ipfw_modevent(module_t mod, int type, void *unused)
2437 /* Called once at module load or
2438 * system boot if compiled in. */
2441 /* Called before unload. May veto unloading. */
2444 /* Called during unload. */
2447 /* Called during system shutdown. */
2456 static moduledata_t ipfwmod = {
2462 /* Define startup order. */
2463 #define IPFW_SI_SUB_FIREWALL SI_SUB_PROTO_IFATTACHDOMAIN
2464 #define IPFW_MODEVENT_ORDER (SI_ORDER_ANY - 255) /* On boot slot in here. */
2465 #define IPFW_MODULE_ORDER (IPFW_MODEVENT_ORDER + 1) /* A little later. */
2466 #define IPFW_VNET_ORDER (IPFW_MODEVENT_ORDER + 2) /* Later still. */
2468 DECLARE_MODULE(ipfw, ipfwmod, IPFW_SI_SUB_FIREWALL, IPFW_MODEVENT_ORDER);
2469 MODULE_VERSION(ipfw, 2);
2470 /* should declare some dependencies here */
2473 * Starting up. Done in order after ipfwmod() has been called.
2474 * VNET_SYSINIT is also called for each existing vnet and each new vnet.
2476 SYSINIT(ipfw_init, IPFW_SI_SUB_FIREWALL, IPFW_MODULE_ORDER,
2478 VNET_SYSINIT(vnet_ipfw_init, IPFW_SI_SUB_FIREWALL, IPFW_VNET_ORDER,
2479 vnet_ipfw_init, NULL);
2482 * Closing up shop. These are done in REVERSE ORDER, but still
2483 * after ipfwmod() has been called. Not called on reboot.
2484 * VNET_SYSUNINIT is also called for each exiting vnet as it exits.
2485 * or when the module is unloaded.
2487 SYSUNINIT(ipfw_destroy, IPFW_SI_SUB_FIREWALL, IPFW_MODULE_ORDER,
2488 ipfw_destroy, NULL);
2489 VNET_SYSUNINIT(vnet_ipfw_uninit, IPFW_SI_SUB_FIREWALL, IPFW_VNET_ORDER,
2490 vnet_ipfw_uninit, NULL);