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 destination router was not the
67 * default gateway. In one case it was filling the routing table of a host
68 * with approximately 300.000 cloned redirect entries until it ran out of
69 * kernel memory. However the networking code proved very robust and it didn't
70 * crash or fail in other ways.
74 * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
75 * is being followed here.
79 #include "opt_ipstealth.h"
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/kernel.h>
84 #include <sys/malloc.h>
86 #include <sys/protosw.h>
87 #include <sys/socket.h>
88 #include <sys/sysctl.h>
92 #include <net/if_types.h>
93 #include <net/if_var.h>
94 #include <net/if_dl.h>
95 #include <net/route.h>
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/in_var.h>
100 #include <netinet/ip.h>
101 #include <netinet/ip_var.h>
102 #include <netinet/ip_icmp.h>
103 #include <netinet/ip_options.h>
105 #include <machine/in_cksum.h>
107 static int ipfastforward_active = 0;
108 SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
109 &ipfastforward_active, 0, "Enable fast IP forwarding");
111 static struct sockaddr_in *
112 ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
114 struct sockaddr_in *dst;
118 * Find route to destination.
120 bzero(ro, sizeof(*ro));
121 dst = (struct sockaddr_in *)&ro->ro_dst;
122 dst->sin_family = AF_INET;
123 dst->sin_len = sizeof(*dst);
124 dst->sin_addr.s_addr = dest.s_addr;
125 rtalloc_ign(ro, RTF_CLONING);
128 * Route there and interface still up?
131 if (rt && (rt->rt_flags & RTF_UP) &&
132 (rt->rt_ifp->if_flags & IFF_UP) &&
133 (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) {
134 if (rt->rt_flags & RTF_GATEWAY)
135 dst = (struct sockaddr_in *)rt->rt_gateway;
137 ipstat.ips_noroute++;
138 ipstat.ips_cantforward++;
141 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
148 * Try to forward a packet based on the destination address.
149 * This is a fast path optimized for the plain forwarding case.
150 * If the packet is handled (and consumed) here then we return 1;
151 * otherwise 0 is returned and the packet should be delivered
152 * to ip_input for full processing.
155 ip_fastforward(struct mbuf *m)
158 struct mbuf *m0 = NULL;
160 struct sockaddr_in *dst = NULL;
162 struct in_addr odest, dest;
166 #ifdef IPFIREWALL_FORWARD
167 struct m_tag *fwd_tag;
171 * Are we active and forwarding packets?
173 if (!ipfastforward_active || !ipforwarding)
182 * Step 1: check for packet drop conditions (and sanity checks)
186 * Is entire packet big enough?
188 if (m->m_pkthdr.len < sizeof(struct ip)) {
189 ipstat.ips_tooshort++;
194 * Is first mbuf large enough for ip header and is header present?
196 if (m->m_len < sizeof (struct ip) &&
197 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
198 ipstat.ips_toosmall++;
199 return NULL; /* mbuf already free'd */
202 ip = mtod(m, struct ip *);
207 if (ip->ip_v != IPVERSION) {
208 ipstat.ips_badvers++;
213 * Is IP header length correct and is it in first mbuf?
215 hlen = ip->ip_hl << 2;
216 if (hlen < sizeof(struct ip)) { /* minimum header length */
220 if (hlen > m->m_len) {
221 if ((m = m_pullup(m, hlen)) == NULL) {
222 ipstat.ips_badhlen++;
223 return NULL; /* mbuf already free'd */
225 ip = mtod(m, struct ip *);
231 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
232 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
234 if (hlen == sizeof(struct ip))
235 sum = in_cksum_hdr(ip);
237 sum = in_cksum(m, hlen);
245 * Remember that we have checked the IP header and found it valid.
247 m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
249 ip_len = ntohs(ip->ip_len);
252 * Is IP length longer than packet we have got?
254 if (m->m_pkthdr.len < ip_len) {
255 ipstat.ips_tooshort++;
260 * Is packet longer than IP header tells us? If yes, truncate packet.
262 if (m->m_pkthdr.len > ip_len) {
263 if (m->m_len == m->m_pkthdr.len) {
265 m->m_pkthdr.len = ip_len;
267 m_adj(m, ip_len - m->m_pkthdr.len);
271 * Is packet from or to 127/8?
273 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
274 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
275 ipstat.ips_badaddr++;
281 * Is packet dropped by traffic conditioner?
283 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
288 * Step 2: fallback conditions to normal ip_input path processing
292 * Only IP packets without options
294 if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
297 else if (ip_doopts == 2) {
298 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
300 return NULL; /* mbuf already free'd */
302 /* else ignore IP options and continue */
306 * Only unicast IP, not from loopback, no L2 or IP broadcast,
307 * no multicast, no INADDR_ANY
309 * XXX: Probably some of these checks could be direct drop
310 * conditions. However it is not clear whether there are some
311 * hacks or obscure behaviours which make it neccessary to
312 * let ip_input handle it. We play safe here and let ip_input
313 * deal with it until it is proven that we can directly drop it.
315 if ((m->m_flags & (M_BCAST|M_MCAST)) ||
316 (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
317 ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
318 ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
319 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
320 IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
321 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
322 IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
323 ip->ip_src.s_addr == INADDR_ANY ||
324 ip->ip_dst.s_addr == INADDR_ANY )
328 * Is it for a local address on this host?
330 if (in_localip(ip->ip_dst))
336 * Step 3: incoming packet firewall processing
340 * Convert to host representation
342 ip->ip_len = ntohs(ip->ip_len);
343 ip->ip_off = ntohs(ip->ip_off);
345 odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
348 * Run through list of ipfilter hooks for input packets
350 if (!PFIL_HOOKED(&inet_pfil_hook))
353 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) ||
360 ip = mtod(m, struct ip *); /* m may have changed by pfil hook */
361 dest.s_addr = ip->ip_dst.s_addr;
364 * Destination address changed?
366 if (odest.s_addr != dest.s_addr) {
368 * Is it now for a local address on this host?
370 if (in_localip(dest))
373 * Go on with new destination address
376 #ifdef IPFIREWALL_FORWARD
377 if (m->m_flags & M_FASTFWD_OURS) {
379 * ipfw changed it for a local address on this host.
383 #endif /* IPFIREWALL_FORWARD */
387 * Step 4: decrement TTL and look up route
396 if (ip->ip_ttl <= IPTTLDEC) {
397 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
398 return NULL; /* mbuf already free'd */
402 * Decrement the TTL and incrementally change the IP header checksum.
403 * Don't bother doing this with hw checksum offloading, it's faster
404 * doing it right here.
406 ip->ip_ttl -= IPTTLDEC;
407 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
408 ip->ip_sum -= ~htons(IPTTLDEC << 8);
410 ip->ip_sum += htons(IPTTLDEC << 8);
416 * Find route to destination.
418 if ((dst = ip_findroute(&ro, dest, m)) == NULL)
419 return NULL; /* icmp unreach already sent */
420 ifp = ro.ro_rt->rt_ifp;
423 * Immediately drop blackholed traffic, and directed broadcasts
424 * for either the all-ones or all-zero subnet addresses on
425 * locally attached networks.
427 if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0)
431 * Step 5: outgoing firewall packet processing
435 * Run through list of hooks for output packets.
437 if (!PFIL_HOOKED(&inet_pfil_hook))
440 if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) {
447 ip = mtod(m, struct ip *);
448 dest.s_addr = ip->ip_dst.s_addr;
451 * Destination address changed?
453 #ifndef IPFIREWALL_FORWARD
454 if (odest.s_addr != dest.s_addr) {
456 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
457 if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
458 #endif /* IPFIREWALL_FORWARD */
460 * Is it now for a local address on this host?
462 #ifndef IPFIREWALL_FORWARD
463 if (in_localip(dest)) {
465 if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
466 #endif /* IPFIREWALL_FORWARD */
469 * Return packet for processing by ip_input().
470 * Keep host byte order as expected at ip_input's
473 m->m_flags |= M_FASTFWD_OURS;
479 * Redo route lookup with new destination address
481 #ifdef IPFIREWALL_FORWARD
483 dest.s_addr = ((struct sockaddr_in *)
484 (fwd_tag + 1))->sin_addr.s_addr;
485 m_tag_delete(m, fwd_tag);
487 #endif /* IPFIREWALL_FORWARD */
489 if ((dst = ip_findroute(&ro, dest, m)) == NULL)
490 return NULL; /* icmp unreach already sent */
491 ifp = ro.ro_rt->rt_ifp;
496 * Step 6: send off the packet
500 * Check if route is dampned (when ARP is unable to resolve)
502 if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
503 ro.ro_rt->rt_rmx.rmx_expire >= time_uptime) {
504 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
510 * Check if there is enough space in the interface queue
512 if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
513 ifp->if_snd.ifq_maxlen) {
514 ipstat.ips_odropped++;
515 /* would send source quench here but that is depreciated */
521 * Check if media link state of interface is not down
523 if (ifp->if_link_state == LINK_STATE_DOWN) {
524 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
529 * Check if packet fits MTU or if hardware will fragment for us
531 if (ro.ro_rt->rt_rmx.rmx_mtu)
532 mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
536 if (ip->ip_len <= mtu ||
537 (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
539 * Restore packet header fields to original values
541 ip->ip_len = htons(ip->ip_len);
542 ip->ip_off = htons(ip->ip_off);
544 * Send off the packet via outgoing interface
546 error = (*ifp->if_output)(ifp, m,
547 (struct sockaddr *)dst, ro.ro_rt);
550 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
552 if (ip->ip_off & IP_DF) {
553 ipstat.ips_cantfrag++;
554 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
559 * We have to fragment the packet
561 m->m_pkthdr.csum_flags |= CSUM_IP;
563 * ip_fragment expects ip_len and ip_off in host byte
564 * order but returns all packets in network byte order
566 if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
567 (~ifp->if_hwassist & CSUM_DELAY_IP))) {
570 KASSERT(m != NULL, ("null mbuf and no error"));
572 * Send off the fragments via outgoing interface
579 error = (*ifp->if_output)(ifp, m,
580 (struct sockaddr *)dst, ro.ro_rt);
583 } while ((m = m0) != NULL);
585 /* Reclaim remaining fragments */
586 for (m = m0; m; m = m0) {
591 ipstat.ips_fragmented++;
596 ipstat.ips_odropped++;
598 ro.ro_rt->rt_rmx.rmx_pksent++;
599 ipstat.ips_forward++;
600 ipstat.ips_fastforward++;