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"
83 #include "opt_pfil_hooks.h"
85 #include <sys/param.h>
86 #include <sys/systm.h>
87 #include <sys/kernel.h>
88 #include <sys/malloc.h>
90 #include <sys/protosw.h>
91 #include <sys/socket.h>
92 #include <sys/sysctl.h>
96 #include <net/if_types.h>
97 #include <net/if_var.h>
98 #include <net/if_dl.h>
99 #include <net/route.h>
101 #include <netinet/in.h>
102 #include <netinet/in_systm.h>
103 #include <netinet/in_var.h>
104 #include <netinet/ip.h>
105 #include <netinet/ip_var.h>
106 #include <netinet/ip_icmp.h>
108 #include <machine/in_cksum.h>
110 #include <netinet/ip_fw.h>
111 #include <netinet/ip_divert.h>
112 #include <netinet/ip_dummynet.h>
114 static int ipfastforward_active = 0;
115 SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
116 &ipfastforward_active, 0, "Enable fast IP forwarding");
118 static struct sockaddr_in *
119 ip_findroute(struct route *ro, in_addr_t dest, struct mbuf *m)
121 struct sockaddr_in *dst;
125 * Find route to destination.
127 bzero(ro, sizeof(*ro));
128 dst = (struct sockaddr_in *)&ro->ro_dst;
129 dst->sin_family = AF_INET;
130 dst->sin_len = sizeof(*dst);
131 dst->sin_addr.s_addr = dest;
132 rtalloc_ign(ro, RTF_CLONING);
135 * Route there and interface still up?
138 if (rt && (rt->rt_flags & RTF_UP) &&
139 (rt->rt_ifp->if_flags & IFF_UP) &&
140 (rt->rt_ifp->if_flags & IFF_RUNNING)) {
141 if (rt->rt_flags & RTF_GATEWAY)
142 dst = (struct sockaddr_in *)rt->rt_gateway;
144 ipstat.ips_noroute++;
145 ipstat.ips_cantforward++;
148 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
155 * Try to forward a packet based on the destination address.
156 * This is a fast path optimized for the plain forwarding case.
157 * If the packet is handled (and consumed) here then we return 1;
158 * otherwise 0 is returned and the packet should be delivered
159 * to ip_input for full processing.
162 ip_fastforward(struct mbuf *m)
165 struct mbuf *m0 = NULL;
168 struct mbuf *clone = NULL;
171 struct sockaddr_in *dst = NULL;
172 struct in_ifaddr *ia = NULL;
173 struct ifaddr *ifa = NULL;
175 struct ip_fw_args args;
176 in_addr_t odest, dest;
182 * Are we active and forwarding packets?
184 if (!ipfastforward_active || !ipforwarding)
193 * Step 1: check for packet drop conditions (and sanity checks)
197 * Is entire packet big enough?
199 if (m->m_pkthdr.len < sizeof(struct ip)) {
200 ipstat.ips_tooshort++;
205 * Is first mbuf large enough for ip header and is header present?
207 if (m->m_len < sizeof (struct ip) &&
208 (m = m_pullup(m, sizeof (struct ip))) == 0) {
209 ipstat.ips_toosmall++;
213 ip = mtod(m, struct ip *);
218 if (ip->ip_v != IPVERSION) {
219 ipstat.ips_badvers++;
224 * Is IP header length correct and is it in first mbuf?
226 hlen = ip->ip_hl << 2;
227 if (hlen < sizeof(struct ip)) { /* minimum header length */
231 if (hlen > m->m_len) {
232 if ((m = m_pullup(m, hlen)) == 0) {
233 ipstat.ips_badhlen++;
236 ip = mtod(m, struct ip *);
242 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
243 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
245 if (hlen == sizeof(struct ip))
246 sum = in_cksum_hdr(ip);
248 sum = in_cksum(m, hlen);
254 m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
256 ip_len = ntohs(ip->ip_len);
259 * Is IP length longer than packet we have got?
261 if (m->m_pkthdr.len < ip_len) {
262 ipstat.ips_tooshort++;
267 * Is packet longer than IP header tells us? If yes, truncate packet.
269 if (m->m_pkthdr.len > ip_len) {
270 if (m->m_len == m->m_pkthdr.len) {
272 m->m_pkthdr.len = ip_len;
274 m_adj(m, ip_len - m->m_pkthdr.len);
278 * Is packet from or to 127/8?
280 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
281 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
282 ipstat.ips_badaddr++;
287 * Step 2: fallback conditions to normal ip_input path processing
291 * Only IP packets without options
293 if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
296 else if (ip_doopts == 2) {
297 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
301 /* else ignore IP options and continue */
305 * Only unicast IP, not from loopback, no L2 or IP broadcast,
306 * no multicast, no INADDR_ANY
308 * XXX: Probably some of these checks could be direct drop
309 * conditions. However it is not clear whether there are some
310 * hacks or obscure behaviours which make it neccessary to
311 * let ip_input handle it. We play safe here and let ip_input
312 * deal with it until it is proven that we can directly drop it.
314 if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
315 ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
316 ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
317 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
318 IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
319 ip->ip_dst.s_addr == INADDR_ANY )
323 * Is it for a local address on this host?
325 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
326 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
331 * Or is it for a local IP broadcast address on this host?
333 if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
334 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
335 if (ifa->ifa_addr->sa_family != AF_INET)
338 if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
340 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
348 * Step 3: incoming packet firewall processing
352 * Convert to host representation
354 ip->ip_len = ntohs(ip->ip_len);
355 ip->ip_off = ntohs(ip->ip_off);
357 odest = dest = ip->ip_dst.s_addr;
360 * Run through list of ipfilter hooks for input packets
362 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN) ||
369 ip = mtod(m, struct ip *); /* m may have changed by pfil hook */
370 dest = ip->ip_dst.s_addr;
374 * Run through ipfw for input packets
376 if (fw_enable && IPFW_LOADED) {
377 bzero(&args, sizeof(args));
380 ipfw = ip_fw_chk_ptr(&args);
387 * Packet denied, drop it
389 if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
392 * Send packet to the appropriate pipe
394 if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
395 ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_IN, &args);
402 if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
404 * See if this is a fragment
406 if (ip->ip_off & (IP_MF | IP_OFFMASK))
411 if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
412 clone = divert_clone(m);
419 * Delayed checksums are not compatible
421 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
423 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
426 * Restore packet header fields to original values
428 tip = mtod(m, struct ip *);
429 tip->ip_len = htons(tip->ip_len);
430 tip->ip_off = htons(tip->ip_off);
432 * Deliver packet to divert input routine
436 * If this was not tee, we are done
439 if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0)
441 /* Continue if it was tee */
445 if (ipfw == 0 && args.next_hop != NULL) {
446 dest = args.next_hop->sin_addr.s_addr;
450 * Let through or not?
456 ip = mtod(m, struct ip *); /* if m changed during fw processing */
459 * Destination address changed?
463 * Is it now for a local address on this host?
465 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
466 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
470 * Go on with new destination address
475 * Step 4: decrement TTL and look up route
484 if (ip->ip_ttl <= IPTTLDEC) {
485 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, NULL);
490 * Decrement the TTL and incrementally change the checksum.
491 * Don't bother doing this with hw checksum offloading.
493 ip->ip_ttl -= IPTTLDEC;
494 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
495 ip->ip_sum -= ~htons(IPTTLDEC << 8);
497 ip->ip_sum += htons(IPTTLDEC << 8);
503 * Find route to destination.
505 if ((dst = ip_findroute(&ro, dest, m)) == NULL)
506 return 1; /* icmp unreach already sent */
507 ifp = ro.ro_rt->rt_ifp;
510 * Step 5: outgoing firewall packet processing
515 * Run through list of hooks for output packets.
517 if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT) || m == NULL) {
524 ip = mtod(m, struct ip *);
525 dest = ip->ip_dst.s_addr;
527 if (fw_enable && IPFW_LOADED && !args.next_hop) {
528 bzero(&args, sizeof(args));
532 ipfw = ip_fw_chk_ptr(&args);
538 if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
541 if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
543 * XXX note: if the ifp or rt entry are deleted
544 * while a pkt is in dummynet, we are in trouble!
546 args.ro = &ro; /* dummynet does not save it */
549 ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_OUT, &args);
553 if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
555 * See if this is a fragment
557 if (ip->ip_off & (IP_MF | IP_OFFMASK))
562 if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
563 clone = divert_clone(m);
570 * Delayed checksums are not compatible with divert
572 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
574 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
577 * Restore packet header fields to original values
579 tip = mtod(m, struct ip *);
580 tip->ip_len = htons(tip->ip_len);
581 tip->ip_off = htons(tip->ip_off);
583 * Deliver packet to divert input routine
587 * If this was not tee, we are done
590 if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0) {
593 /* Continue if it was tee */
597 if (ipfw == 0 && args.next_hop != NULL) {
598 dest = args.next_hop->sin_addr.s_addr;
602 * Let through or not?
608 ip = mtod(m, struct ip *);
611 * Destination address changed?
615 * Is it now for a local address on this host?
617 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
618 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
621 struct m_tag *mtag = m_tag_get(
622 PACKET_TAG_IPFORWARD,
623 sizeof(struct sockaddr_in *),
628 *(struct sockaddr_in **)(mtag+1) =
630 m_tag_prepend(m, mtag);
633 droptoours: /* Used for DIVERT */
636 m->m_flags |= M_FASTFWD_OURS;
638 /* ip still points to the real packet */
639 ip->ip_len = htons(ip->ip_len);
640 ip->ip_off = htons(ip->ip_off);
643 * Return packet for processing by ip_input
651 * Redo route lookup with new destination address
654 if ((dst = ip_findroute(&ro, dest, m)) == NULL)
655 return 1; /* icmp unreach already sent */
656 ifp = ro.ro_rt->rt_ifp;
660 * Step 6: send off the packet
664 * Check if route is dampned (when ARP is unable to resolve)
666 if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
667 ro.ro_rt->rt_rmx.rmx_expire >= time_second) {
668 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
673 * Check if there is enough space in the interface queue
675 if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
676 ifp->if_snd.ifq_maxlen) {
677 ipstat.ips_odropped++;
678 /* would send source quench here but that is depreciated */
683 * Check if media link state of interface is not down
685 if (ifp->if_link_state == LINK_STATE_DOWN) {
686 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
691 * Check if packet fits MTU or if hardware will fragement for us
693 if (ro.ro_rt->rt_rmx.rmx_mtu)
694 mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
698 if (ip->ip_len <= mtu ||
699 (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
701 * Restore packet header fields to original values
703 ip->ip_len = htons(ip->ip_len);
704 ip->ip_off = htons(ip->ip_off);
706 * Send off the packet via outgoing interface
708 error = (*ifp->if_output)(ifp, m,
709 (struct sockaddr *)dst, ro.ro_rt);
712 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
714 if (ip->ip_off & IP_DF) {
715 ipstat.ips_cantfrag++;
716 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
721 * We have to fragement the packet
723 m->m_pkthdr.csum_flags |= CSUM_IP;
725 * ip_fragment expects ip_len and ip_off in host byte
726 * order but returns all packets in network byte order
728 if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
729 (~ifp->if_hwassist & CSUM_DELAY_IP))) {
732 KASSERT(m != NULL, ("null mbuf and no error"));
734 * Send off the fragments via outgoing interface
741 error = (*ifp->if_output)(ifp, m,
742 (struct sockaddr *)dst, ro.ro_rt);
745 } while ((m = m0) != NULL);
747 /* Reclaim remaining fragments */
754 ipstat.ips_fragmented++;
759 ipstat.ips_odropped++;
761 ro.ro_rt->rt_rmx.rmx_pksent++;
762 ipstat.ips_forward++;
763 ipstat.ips_fastforward++;