ip_fastfwd.c revision 134383 
1/*
2 * Copyright (c) 2003 Andre Oppermann, Internet Business Solutions AG
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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
15 *    permission.
16 *
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
27 * SUCH DAMAGE.
28 *
29 * $FreeBSD: head/sys/netinet/ip_fastfwd.c 134383 2004-08-27 15:16:24Z andre $
30 */
31
32/*
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
42 * transmits.
43 *
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.
47 *
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.
51 *
52 * Firewalling is fully supported including divert, ipfw fwd and ipfilter
53 * ipnat and address rewrite.
54 *
55 * IPSEC is not supported if this host is a tunnel broker. IPSEC is
56 * supported for connections to/from local host.
57 *
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.
60 *
61 * We take full advantage of hardware support for ip checksum and
62 * fragmentation offloading.
63 *
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
70 * otherwise.
71 */
72
73/*
74 * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
75 * is being followed here.
76 */
77
78#include "opt_ipfw.h"
79#include "opt_ipstealth.h"
80
81#include <sys/param.h>
82#include <sys/systm.h>
83#include <sys/kernel.h>
84#include <sys/malloc.h>
85#include <sys/mbuf.h>
86#include <sys/protosw.h>
87#include <sys/socket.h>
88#include <sys/sysctl.h>
89
90#include <net/pfil.h>
91#include <net/if.h>
92#include <net/if_types.h>
93#include <net/if_var.h>
94#include <net/if_dl.h>
95#include <net/route.h>
96
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
104#include <machine/in_cksum.h>
105
106static int ipfastforward_active = 0;
107SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
108    &ipfastforward_active, 0, "Enable fast IP forwarding");
109
110static struct sockaddr_in *
111ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
112{
113	struct sockaddr_in *dst;
114	struct rtentry *rt;
115
116	/*
117	 * Find route to destination.
118	 */
119	bzero(ro, sizeof(*ro));
120	dst = (struct sockaddr_in *)&ro->ro_dst;
121	dst->sin_family = AF_INET;
122	dst->sin_len = sizeof(*dst);
123	dst->sin_addr.s_addr = dest.s_addr;
124	rtalloc_ign(ro, RTF_CLONING);
125
126	/*
127	 * Route there and interface still up?
128	 */
129	rt = ro->ro_rt;
130	if (rt && (rt->rt_flags & RTF_UP) &&
131	    (rt->rt_ifp->if_flags & IFF_UP) &&
132	    (rt->rt_ifp->if_flags & IFF_RUNNING)) {
133		if (rt->rt_flags & RTF_GATEWAY)
134			dst = (struct sockaddr_in *)rt->rt_gateway;
135	} else {
136		ipstat.ips_noroute++;
137		ipstat.ips_cantforward++;
138		if (rt)
139			RTFREE(rt);
140		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
141		return NULL;
142	}
143	return dst;
144}
145
146/*
147 * Try to forward a packet based on the destination address.
148 * This is a fast path optimized for the plain forwarding case.
149 * If the packet is handled (and consumed) here then we return 1;
150 * otherwise 0 is returned and the packet should be delivered
151 * to ip_input for full processing.
152 */
153int
154ip_fastforward(struct mbuf *m)
155{
156	struct ip *ip;
157	struct mbuf *m0 = NULL;
158	struct route ro;
159	struct sockaddr_in *dst = NULL;
160	struct in_ifaddr *ia = NULL;
161	struct ifaddr *ifa = NULL;
162	struct ifnet *ifp;
163	struct in_addr odest, dest;
164	u_short sum, ip_len;
165	int error = 0;
166	int hlen, mtu;
167#ifdef IPFIREWALL_FORWARD
168	struct m_tag *fwd_tag;
169#endif
170
171	/*
172	 * Are we active and forwarding packets?
173	 */
174	if (!ipfastforward_active || !ipforwarding)
175		return 0;
176
177	M_ASSERTVALID(m);
178	M_ASSERTPKTHDR(m);
179
180	ro.ro_rt = NULL;
181
182	/*
183	 * Step 1: check for packet drop conditions (and sanity checks)
184	 */
185
186	/*
187	 * Is entire packet big enough?
188	 */
189	if (m->m_pkthdr.len < sizeof(struct ip)) {
190		ipstat.ips_tooshort++;
191		goto drop;
192	}
193
194	/*
195	 * Is first mbuf large enough for ip header and is header present?
196	 */
197	if (m->m_len < sizeof (struct ip) &&
198	   (m = m_pullup(m, sizeof (struct ip))) == 0) {
199		ipstat.ips_toosmall++;
200		goto drop;
201	}
202
203	ip = mtod(m, struct ip *);
204
205	/*
206	 * Is it IPv4?
207	 */
208	if (ip->ip_v != IPVERSION) {
209		ipstat.ips_badvers++;
210		goto drop;
211	}
212
213	/*
214	 * Is IP header length correct and is it in first mbuf?
215	 */
216	hlen = ip->ip_hl << 2;
217	if (hlen < sizeof(struct ip)) {	/* minimum header length */
218		ipstat.ips_badlen++;
219		goto drop;
220	}
221	if (hlen > m->m_len) {
222		if ((m = m_pullup(m, hlen)) == 0) {
223			ipstat.ips_badhlen++;
224			goto drop;
225		}
226		ip = mtod(m, struct ip *);
227	}
228
229	/*
230	 * Checksum correct?
231	 */
232	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
233		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
234	else {
235		if (hlen == sizeof(struct ip))
236			sum = in_cksum_hdr(ip);
237		else
238			sum = in_cksum(m, hlen);
239	}
240	if (sum) {
241		ipstat.ips_badsum++;
242		goto drop;
243	}
244	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
245
246	ip_len = ntohs(ip->ip_len);
247
248	/*
249	 * Is IP length longer than packet we have got?
250	 */
251	if (m->m_pkthdr.len < ip_len) {
252		ipstat.ips_tooshort++;
253		goto drop;
254	}
255
256	/*
257	 * Is packet longer than IP header tells us? If yes, truncate packet.
258	 */
259	if (m->m_pkthdr.len > ip_len) {
260		if (m->m_len == m->m_pkthdr.len) {
261			m->m_len = ip_len;
262			m->m_pkthdr.len = ip_len;
263		} else
264			m_adj(m, ip_len - m->m_pkthdr.len);
265	}
266
267	/*
268	 * Is packet from or to 127/8?
269	 */
270	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
271	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
272		ipstat.ips_badaddr++;
273		goto drop;
274	}
275
276#ifdef ALTQ
277	/*
278	 * Is packet dropped by traffic conditioner?
279	 */
280	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
281		return 1;
282#endif
283
284	/*
285	 * Step 2: fallback conditions to normal ip_input path processing
286	 */
287
288	/*
289	 * Only IP packets without options
290	 */
291	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
292		if (ip_doopts == 1)
293			return 0;
294		else if (ip_doopts == 2) {
295			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
296				0, NULL);
297			return 1;
298		}
299		/* else ignore IP options and continue */
300	}
301
302	/*
303	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
304	 * no multicast, no INADDR_ANY
305	 *
306	 * XXX: Probably some of these checks could be direct drop
307	 * conditions.  However it is not clear whether there are some
308	 * hacks or obscure behaviours which make it neccessary to
309	 * let ip_input handle it.  We play safe here and let ip_input
310	 * deal with it until it is proven that we can directly drop it.
311	 */
312	if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
313	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
314	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
315	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
316	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
317	    ip->ip_dst.s_addr == INADDR_ANY )
318		return 0;
319
320	/*
321	 * Is it for a local address on this host?
322	 */
323	if (in_localip(ip->ip_dst))
324		return 0;
325
326	/*
327	 * Or is it for a local IP broadcast address on this host?
328	 */
329	if ((m->m_flags & M_BCAST) &&
330	    (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST)) {
331	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
332			if (ifa->ifa_addr->sa_family != AF_INET)
333				continue;
334			ia = ifatoia(ifa);
335			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
336				return 0;
337			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
338			    ip->ip_dst.s_addr)
339				return 0;
340		}
341	}
342	ipstat.ips_total++;
343
344	/*
345	 * Step 3: incoming packet firewall processing
346	 */
347
348	/*
349	 * Convert to host representation
350	 */
351	ip->ip_len = ntohs(ip->ip_len);
352	ip->ip_off = ntohs(ip->ip_off);
353
354	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
355
356	/*
357	 * Run through list of ipfilter hooks for input packets
358	 */
359	if (inet_pfil_hook.ph_busy_count == -1)
360		goto passin;
361
362	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN) ||
363	    m == NULL)
364		return 1;
365
366	M_ASSERTVALID(m);
367	M_ASSERTPKTHDR(m);
368
369	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
370	dest.s_addr = ip->ip_dst.s_addr;
371
372	/*
373	 * Destination address changed?
374	 */
375	if (odest.s_addr != dest.s_addr) {
376		/*
377		 * Is it now for a local address on this host?
378		 */
379		if (in_localip(dest))
380			goto forwardlocal;
381		/*
382		 * Go on with new destination address
383		 */
384	}
385#ifdef IPFIREWALL_FORWARD
386	if (m->m_flags & M_FASTFWD_OURS) {
387		/*
388		 * ipfw changed it for a local address on this host.
389		 */
390		goto forwardlocal;
391	}
392#endif /* IPFIREWALL_FORWARD */
393
394passin:
395	/*
396	 * Step 4: decrement TTL and look up route
397	 */
398
399	/*
400	 * Check TTL
401	 */
402#ifdef IPSTEALTH
403	if (!ipstealth) {
404#endif
405	if (ip->ip_ttl <= IPTTLDEC) {
406		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, NULL);
407		return 1;
408	}
409
410	/*
411	 * Decrement the TTL and incrementally change the checksum.
412	 * Don't bother doing this with hw checksum offloading.
413	 */
414	ip->ip_ttl -= IPTTLDEC;
415	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
416		ip->ip_sum -= ~htons(IPTTLDEC << 8);
417	else
418		ip->ip_sum += htons(IPTTLDEC << 8);
419#ifdef IPSTEALTH
420	}
421#endif
422
423	/*
424	 * Find route to destination.
425	 */
426	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
427		return 1;	/* icmp unreach already sent */
428	ifp = ro.ro_rt->rt_ifp;
429
430	/*
431	 * Step 5: outgoing firewall packet processing
432	 */
433
434	/*
435	 * Run through list of hooks for output packets.
436	 */
437	if (inet_pfil_hook.ph_busy_count == -1)
438		goto passout;
439
440	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT) || m == NULL) {
441		goto consumed;
442	}
443
444	M_ASSERTVALID(m);
445	M_ASSERTPKTHDR(m);
446
447	ip = mtod(m, struct ip *);
448	dest.s_addr = ip->ip_dst.s_addr;
449
450	/*
451	 * Destination address changed?
452	 */
453#ifndef IPFIREWALL_FORWARD
454	if (odest.s_addr != dest.s_addr) {
455#else
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 */
459		/*
460		 * Is it now for a local address on this host?
461		 */
462#ifndef IPFIREWALL_FORWARD
463		if (in_localip(dest)) {
464#else
465		if (in_localip(dest) || m->m_flags & M_FASTFWD_OURS) {
466#endif /* IPFIREWALL_FORWARD */
467forwardlocal:
468			/* for ip_input */
469			m->m_flags |= M_FASTFWD_OURS;
470			ip->ip_len = htons(ip->ip_len);
471			ip->ip_off = htons(ip->ip_off);
472
473			/*
474			 * Return packet for processing by ip_input()
475			 */
476			if (ro.ro_rt)
477				RTFREE(ro.ro_rt);
478			return 0;
479		}
480		/*
481		 * Redo route lookup with new destination address
482		 */
483#ifdef IPFIREWALL_FORWARD
484		if (fwd_tag) {
485			if (!in_localip(ip->ip_src) && !in_localaddr(ip->ip_dst))
486				dest.s_addr = ((struct sockaddr_in *)(fwd_tag+1))->sin_addr.s_addr;
487				//bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
488			m_tag_delete(m, fwd_tag);
489		}
490#endif /* IPFIREWALL_FORWARD */
491		RTFREE(ro.ro_rt);
492		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
493			return 1;	/* icmp unreach already sent */
494		ifp = ro.ro_rt->rt_ifp;
495	}
496
497passout:
498	/*
499	 * Step 6: send off the packet
500	 */
501
502	/*
503	 * Check if route is dampned (when ARP is unable to resolve)
504	 */
505	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
506	    ro.ro_rt->rt_rmx.rmx_expire >= time_second) {
507		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
508		goto consumed;
509	}
510
511#ifndef ALTQ
512	/*
513	 * Check if there is enough space in the interface queue
514	 */
515	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
516	    ifp->if_snd.ifq_maxlen) {
517		ipstat.ips_odropped++;
518		/* would send source quench here but that is depreciated */
519		goto drop;
520	}
521#endif
522
523	/*
524	 * Check if media link state of interface is not down
525	 */
526	if (ifp->if_link_state == LINK_STATE_DOWN) {
527		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
528		goto consumed;
529	}
530
531	/*
532	 * Check if packet fits MTU or if hardware will fragement for us
533	 */
534	if (ro.ro_rt->rt_rmx.rmx_mtu)
535		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
536	else
537		mtu = ifp->if_mtu;
538
539	if (ip->ip_len <= mtu ||
540	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
541		/*
542		 * Restore packet header fields to original values
543		 */
544		ip->ip_len = htons(ip->ip_len);
545		ip->ip_off = htons(ip->ip_off);
546		/*
547		 * Send off the packet via outgoing interface
548		 */
549		error = (*ifp->if_output)(ifp, m,
550				(struct sockaddr *)dst, ro.ro_rt);
551	} else {
552		/*
553		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
554		 */
555		if (ip->ip_off & IP_DF) {
556			ipstat.ips_cantfrag++;
557			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
558				0, ifp);
559			goto consumed;
560		} else {
561			/*
562			 * We have to fragement the packet
563			 */
564			m->m_pkthdr.csum_flags |= CSUM_IP;
565			/*
566			 * ip_fragment expects ip_len and ip_off in host byte
567			 * order but returns all packets in network byte order
568			 */
569			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
570					(~ifp->if_hwassist & CSUM_DELAY_IP))) {
571				goto drop;
572			}
573			KASSERT(m != NULL, ("null mbuf and no error"));
574			/*
575			 * Send off the fragments via outgoing interface
576			 */
577			error = 0;
578			do {
579				m0 = m->m_nextpkt;
580				m->m_nextpkt = NULL;
581
582				error = (*ifp->if_output)(ifp, m,
583					(struct sockaddr *)dst, ro.ro_rt);
584				if (error)
585					break;
586			} while ((m = m0) != NULL);
587			if (error) {
588				/* Reclaim remaining fragments */
589				for (; m; m = m0) {
590					m0 = m->m_nextpkt;
591					m->m_nextpkt = NULL;
592					m_freem(m);
593				}
594			} else
595				ipstat.ips_fragmented++;
596		}
597	}
598
599	if (error != 0)
600		ipstat.ips_odropped++;
601	else {
602		ro.ro_rt->rt_rmx.rmx_pksent++;
603		ipstat.ips_forward++;
604		ipstat.ips_fastforward++;
605	}
606consumed:
607	RTFREE(ro.ro_rt);
608	return 1;
609drop:
610	if (m)
611		m_freem(m);
612	if (ro.ro_rt)
613		RTFREE(ro.ro_rt);
614	return 1;
615}
616