2 * Copyright (c) 2003 Andre Oppermann, Internet Business Solutions AG
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. The name of the author may not be used to endorse or promote
14 * products derived from this software without specific prior written
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * ip_fastforward gets its speed from processing the forwarded packet to
34 * completion (if_output on the other side) without any queues or netisr's.
35 * The receiving interface DMAs the packet into memory, the upper half of
36 * driver calls ip_fastforward, we do our routing table lookup and directly
37 * send it off to the outgoing interface which DMAs the packet to the
38 * network card. The only part of the packet we touch with the CPU is the
39 * IP header (unless there are complex firewall rules touching other parts
40 * of the packet, but that is up to you). We are essentially limited by bus
41 * bandwidth and how fast the network card/driver can set up receives and
44 * We handle basic errors, ip header errors, checksum errors,
45 * destination unreachable, fragmentation and fragmentation needed and
46 * report them via icmp to the sender.
48 * Else if something is not pure IPv4 unicast forwarding we fall back to
49 * the normal ip_input processing path. We should only be called from
50 * interfaces connected to the outside world.
52 * Firewalling is fully supported including divert, ipfw fwd and ipfilter
53 * ipnat and address rewrite.
55 * IPSEC is not supported if this host is a tunnel broker. IPSEC is
56 * supported for connections to/from local host.
58 * We try to do the least expensive (in CPU ops) checks and operations
59 * first to catch junk with as little overhead as possible.
61 * We take full advantage of hardware support for ip checksum and
62 * fragmentation offloading.
64 * We don't do ICMP redirect in the fast forwarding path. I have had my own
65 * cases where two core routers with Zebra routing suite would send millions
66 * ICMP redirects to connected hosts if the router to dest was not the default
67 * gateway. In one case it was filling the routing table of a host with close
68 * 300'000 cloned redirect entries until it ran out of kernel memory. However
69 * the networking code proved very robust and it didn't crash or went ill
74 * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
75 * is being followed here.
80 #include "opt_ipdivert.h"
81 #include "opt_ipfilter.h"
82 #include "opt_ipstealth.h"
84 #include "opt_pfil_hooks.h"
86 #include <sys/param.h>
87 #include <sys/systm.h>
88 #include <sys/kernel.h>
90 #include <sys/malloc.h>
92 #include <sys/protosw.h>
93 #include <sys/socket.h>
94 #include <sys/sysctl.h>
98 #include <net/if_types.h>
99 #include <net/if_var.h>
100 #include <net/if_dl.h>
101 #include <net/route.h>
103 #include <netinet/in.h>
104 #include <netinet/in_systm.h>
105 #include <netinet/in_var.h>
106 #include <netinet/ip.h>
107 #include <netinet/ip_var.h>
108 #include <netinet/ip_icmp.h>
110 #include <machine/in_cksum.h>
112 #include <netinet/ip_fw.h>
113 #include <netinet/ip_dummynet.h>
115 static int ipfastforward_active = 0;
116 SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
117 &ipfastforward_active, 0, "Enable fast IP forwarding");
120 * Try to forward a packet based on the destination address.
121 * This is a fast path optimized for the plain forwarding case.
122 * If the packet is handled (and consumed) here then we return 1;
123 * otherwise 0 is returned and the packet should be delivered
124 * to ip_input for full processing.
127 ip_fastforward(struct mbuf *m)
130 struct mbuf *m0 = NULL;
133 struct mbuf *teem = NULL;
135 struct mbuf *tag = NULL;
137 struct sockaddr_in *dst = NULL;
138 struct in_ifaddr *ia = NULL;
139 struct ifaddr *ifa = NULL;
140 struct ifnet *ifp = NULL;
141 struct ip_fw_args args;
142 in_addr_t odest, dest;
148 * Are we active and forwarding packets?
150 if (!ipfastforward_active || !ipforwarding)
154 * If there is any MT_TAG we fall back to ip_input because we can't
155 * handle TAGs here. Should never happen as we get directly called
156 * from the if_output routines.
158 if (m->m_type == MT_TAG) {
159 KASSERT(0, ("%s: packet with MT_TAG not expected", __func__));
167 * Step 1: check for packet drop conditions (and sanity checks)
171 * Is entire packet big enough?
173 if (m->m_pkthdr.len < sizeof(struct ip)) {
174 ipstat.ips_tooshort++;
179 * Is first mbuf large enough for ip header and is header present?
181 if (m->m_len < sizeof (struct ip) &&
182 (m = m_pullup(m, sizeof (struct ip))) == 0) {
183 ipstat.ips_toosmall++;
187 ip = mtod(m, struct ip *);
192 if (ip->ip_v != IPVERSION) {
193 ipstat.ips_badvers++;
198 * Is IP header length correct and is it in first mbuf?
200 hlen = ip->ip_hl << 2;
201 if (hlen < sizeof(struct ip)) { /* minimum header length */
205 if (hlen > m->m_len) {
206 if ((m = m_pullup(m, hlen)) == 0) {
207 ipstat.ips_badhlen++;
210 ip = mtod(m, struct ip *);
216 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
217 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
219 if (hlen == sizeof(struct ip))
220 sum = in_cksum_hdr(ip);
222 sum = in_cksum(m, hlen);
228 m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
231 * Convert to host representation
233 ip->ip_len = ntohs(ip->ip_len);
234 ip->ip_off = ntohs(ip->ip_off);
237 * Is IP length longer than packet we have got?
239 if (m->m_pkthdr.len < ip->ip_len) {
240 ipstat.ips_tooshort++;
245 * Is packet longer than IP header tells us? If yes, truncate packet.
247 if (m->m_pkthdr.len > ip->ip_len) {
248 if (m->m_len == m->m_pkthdr.len) {
249 m->m_len = ip->ip_len;
250 m->m_pkthdr.len = ip->ip_len;
252 m_adj(m, ip->ip_len - m->m_pkthdr.len);
256 * Is packet from or to 127/8?
258 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
259 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
260 ipstat.ips_badaddr++;
265 * Step 2: fallback conditions to normal ip_input path processing
269 * Only IP packets without options
271 if (ip->ip_hl != (sizeof(struct ip) >> 2))
275 * Only unicast IP, not from loopback, no L2 or IP broadcast,
276 * no multicast, no INADDR_ANY
278 * XXX: Probably some of these checks could be direct drop
279 * conditions. However it is not clear whether there are some
280 * hacks or obscure behaviours which make it neccessary to
281 * let ip_input handle it. We play safe here and let ip_input
282 * deal with it until it is proven that we can directly drop it.
284 if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
285 ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
286 ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
287 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
288 IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
289 ip->ip_dst.s_addr == INADDR_ANY )
293 * Is it for a local address on this host?
295 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
296 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
301 * Or is it for a local IP broadcast address on this host?
303 if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
304 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
305 if (ifa->ifa_addr->sa_family != AF_INET)
308 if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
310 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
315 /* return packet back to netisr for slow processing */
316 ip->ip_len = htons(ip->ip_len);
317 ip->ip_off = htons(ip->ip_off);
324 * Step 3: incoming packet firewall processing
327 odest = dest = ip->ip_dst.s_addr;
330 * Run through list of ipfilter hooks for input packets
332 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN) ||
339 ip = mtod(m, struct ip *); /* m may have changed by pfil hook */
340 dest = ip->ip_dst.s_addr;
344 * Run through ipfw for input packets
346 if (fw_enable && IPFW_LOADED) {
347 bzero(&args, sizeof(args));
350 ipfw = ip_fw_chk_ptr(&args);
357 * Packet denied, drop it
359 if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
362 * Send packet to the appropriate pipe
364 if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
365 ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_IN, &args);
372 if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
374 * See if this is a fragment
376 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
377 MGETHDR(tag, M_DONTWAIT, MT_TAG);
380 tag->m_flags = PACKET_TAG_DIVERT;
381 tag->m_data = (caddr_t)(intptr_t)args.divert_rule;
383 /* XXX: really bloody hack, see ip_input */
384 tag->m_nextpkt = (struct mbuf *)1;
393 if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
394 teem = m_dup(m, M_DONTWAIT);
401 * Delayed checksums are not compatible
403 if (teem->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
404 in_delayed_cksum(teem);
405 teem->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
408 * Restore packet header fields to original values
410 tip = mtod(teem, struct ip *);
411 tip->ip_len = htons(tip->ip_len);
412 tip->ip_off = htons(tip->ip_off);
414 * Deliver packet to divert input routine
416 divert_packet(teem, 0, ipfw & 0xffff, args.divert_rule);
418 * If this was not tee, we are done
420 if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0)
422 /* Continue if it was tee */
426 if (ipfw == 0 && args.next_hop != NULL) {
427 dest = args.next_hop->sin_addr.s_addr;
431 * Let through or not?
437 ip = mtod(m, struct ip *); /* if m changed during fw processing */
440 * Destination address changed?
444 * Is it now for a local address on this host?
446 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
447 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
451 * Go on with new destination address
456 * Step 4: decrement TTL and look up route
465 if (ip->ip_ttl <= IPTTLDEC) {
466 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, NULL);
471 * Decrement the TTL and incrementally change the checksum.
472 * Don't bother doing this with hw checksum offloading.
474 ip->ip_ttl -= IPTTLDEC;
475 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
476 ip->ip_sum -= ~htons(IPTTLDEC << 8);
478 ip->ip_sum += htons(IPTTLDEC << 8);
484 * Find route to destination.
486 bzero(&ro, sizeof(ro));
487 dst = (struct sockaddr_in *)&ro.ro_dst;
488 dst->sin_family = AF_INET;
489 dst->sin_len = sizeof(*dst);
490 dst->sin_addr.s_addr = dest;
491 rtalloc_ign(&ro, RTF_CLONING);
494 * Route there and interface still up?
497 (ro.ro_rt->rt_flags & RTF_UP) &&
498 (ro.ro_rt->rt_ifp->if_flags & IFF_UP)) {
499 ia = ifatoia(ro.ro_rt->rt_ifa);
500 ifp = ro.ro_rt->rt_ifp;
501 if (ro.ro_rt->rt_flags & RTF_GATEWAY)
502 dst = (struct sockaddr_in *)ro.ro_rt->rt_gateway;
504 ipstat.ips_noroute++;
505 ipstat.ips_cantforward++;
506 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
513 * Step 5: outgoing firewall packet processing
518 * Run through list of hooks for output packets.
520 if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT) || m == NULL) {
528 ip = mtod(m, struct ip *);
529 dest = ip->ip_dst.s_addr;
531 if (fw_enable && IPFW_LOADED && !args.next_hop) {
532 bzero(&args, sizeof(args));
536 ipfw = ip_fw_chk_ptr(&args);
542 if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL) {
546 if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
548 * XXX note: if the ifp or rt entry are deleted
549 * while a pkt is in dummynet, we are in trouble!
551 args.ro = &ro; /* dummynet does not save it */
554 ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_OUT, &args);
559 if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
561 * See if this is a fragment
563 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
564 MGETHDR(tag, M_DONTWAIT, MT_TAG);
569 tag->m_flags = PACKET_TAG_DIVERT;
570 tag->m_data = (caddr_t)(intptr_t)args.divert_rule;
572 /* XXX: really bloody hack, see ip_input */
573 tag->m_nextpkt = (struct mbuf *)1;
582 if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
583 teem = m_dup(m, M_DONTWAIT);
590 * Delayed checksums are not compatible with divert
592 if (teem->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
593 in_delayed_cksum(teem);
594 teem->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
597 * Restore packet header fields to original values
599 tip = mtod(teem, struct ip *);
600 tip->ip_len = htons(tip->ip_len);
601 tip->ip_off = htons(tip->ip_off);
603 * Deliver packet to divert input routine
605 divert_packet(teem, 0, ipfw & 0xffff, args.divert_rule);
607 * If this was not tee, we are done
609 if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0) {
613 /* Continue if it was tee */
617 if (ipfw == 0 && args.next_hop != NULL) {
618 dest = args.next_hop->sin_addr.s_addr;
622 * Let through or not?
628 ip = mtod(m, struct ip *);
631 * Destination address changed?
635 * Is it now for a local address on this host?
637 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
638 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
642 MGETHDR(tag, M_DONTWAIT, MT_TAG);
648 tag->m_flags = PACKET_TAG_IPFORWARD;
649 tag->m_data = (caddr_t)args.next_hop;
651 /* XXX: really bloody hack,
653 tag->m_nextpkt = (struct mbuf *)1;
658 droptoours: /* Used for DIVERT */
660 MGETHDR(tag, M_DONTWAIT, MT_TAG);
666 tag->m_flags = PACKET_TAG_IPFASTFWD_OURS;
669 /* XXX: really bloody hack, see ip_input */
670 tag->m_nextpkt = (struct mbuf *)1;
674 /* ip still points to the real packet */
675 ip->ip_len = htons(ip->ip_len);
676 ip->ip_off = htons(ip->ip_off);
679 * Return packet for processing by ip_input
687 * Redo route lookup with new destination address
690 bzero(&ro, sizeof(ro));
691 dst = (struct sockaddr_in *)&ro.ro_dst;
692 dst->sin_family = AF_INET;
693 dst->sin_len = sizeof(*dst);
694 dst->sin_addr.s_addr = dest;
695 rtalloc_ign(&ro, RTF_CLONING);
698 * Route there and interface still up?
701 (ro.ro_rt->rt_flags & RTF_UP) &&
702 (ro.ro_rt->rt_ifp->if_flags & IFF_UP)) {
703 ia = ifatoia(ro.ro_rt->rt_ifa);
704 ifp = ro.ro_rt->rt_ifp;
705 if (ro.ro_rt->rt_flags & RTF_GATEWAY)
706 dst = (struct sockaddr_in *)ro.ro_rt->rt_gateway;
708 ipstat.ips_noroute++;
709 ipstat.ips_cantforward++;
710 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
718 * Step 6: send off the packet
722 * Check if packet fits MTU or if hardware will fragement for us
724 if (ro.ro_rt->rt_rmx.rmx_mtu)
725 mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
729 if (ip->ip_len <= mtu ||
730 (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
732 * Restore packet header fields to original values
734 ip->ip_len = htons(ip->ip_len);
735 ip->ip_off = htons(ip->ip_off);
737 * Send off the packet via outgoing interface
739 error = (*ifp->if_output)(ifp, m,
740 (struct sockaddr *)dst, ro.ro_rt);
742 ia->ia_ifa.if_opackets++;
743 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
747 * Handle EMSGSIZE with icmp reply
748 * needfrag for TCP MTU discovery
750 if (ip->ip_off & IP_DF) {
751 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
753 ipstat.ips_cantfrag++;
758 * We have to fragement the packet
760 m->m_pkthdr.csum_flags |= CSUM_IP;
761 if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
762 (~ifp->if_hwassist & CSUM_DELAY_IP))) {
766 KASSERT(m != NULL, ("null mbuf and no error"));
768 * Send off the fragments via outgoing interface
775 error = (*ifp->if_output)(ifp, m,
776 (struct sockaddr *)dst, ro.ro_rt);
779 } while ((m = m0) != NULL);
781 /* Reclaim remaining fragments */
788 ipstat.ips_fragmented++;
793 ipstat.ips_odropped++;
795 ro.ro_rt->rt_rmx.rmx_pksent++;
796 ipstat.ips_forward++;
797 ipstat.ips_fastforward++;