xref: /freebsd/sys/netinet/ip_fastfwd.c (revision acd3428b7d3e94cef0e1881c868cb4b131d4ff41)
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  * $FreeBSD$
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 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.
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 #include <netinet/ip_options.h>
104 
105 #include <machine/in_cksum.h>
106 
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");
110 
111 static struct sockaddr_in *
112 ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
113 {
114 	struct sockaddr_in *dst;
115 	struct rtentry *rt;
116 
117 	/*
118 	 * Find route to destination.
119 	 */
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);
126 
127 	/*
128 	 * Route there and interface still up?
129 	 */
130 	rt = ro->ro_rt;
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;
136 	} else {
137 		ipstat.ips_noroute++;
138 		ipstat.ips_cantforward++;
139 		if (rt)
140 			RTFREE(rt);
141 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
142 		return NULL;
143 	}
144 	return dst;
145 }
146 
147 /*
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.
153  */
154 struct mbuf *
155 ip_fastforward(struct mbuf *m)
156 {
157 	struct ip *ip;
158 	struct mbuf *m0 = NULL;
159 	struct route ro;
160 	struct sockaddr_in *dst = NULL;
161 	struct ifnet *ifp;
162 	struct in_addr odest, dest;
163 	u_short sum, ip_len;
164 	int error = 0;
165 	int hlen, mtu;
166 #ifdef IPFIREWALL_FORWARD
167 	struct m_tag *fwd_tag;
168 #endif
169 
170 	/*
171 	 * Are we active and forwarding packets?
172 	 */
173 	if (!ipfastforward_active || !ipforwarding)
174 		return m;
175 
176 	M_ASSERTVALID(m);
177 	M_ASSERTPKTHDR(m);
178 
179 	ro.ro_rt = NULL;
180 
181 	/*
182 	 * Step 1: check for packet drop conditions (and sanity checks)
183 	 */
184 
185 	/*
186 	 * Is entire packet big enough?
187 	 */
188 	if (m->m_pkthdr.len < sizeof(struct ip)) {
189 		ipstat.ips_tooshort++;
190 		goto drop;
191 	}
192 
193 	/*
194 	 * Is first mbuf large enough for ip header and is header present?
195 	 */
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 */
200 	}
201 
202 	ip = mtod(m, struct ip *);
203 
204 	/*
205 	 * Is it IPv4?
206 	 */
207 	if (ip->ip_v != IPVERSION) {
208 		ipstat.ips_badvers++;
209 		goto drop;
210 	}
211 
212 	/*
213 	 * Is IP header length correct and is it in first mbuf?
214 	 */
215 	hlen = ip->ip_hl << 2;
216 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
217 		ipstat.ips_badlen++;
218 		goto drop;
219 	}
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 */
224 		}
225 		ip = mtod(m, struct ip *);
226 	}
227 
228 	/*
229 	 * Checksum correct?
230 	 */
231 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
232 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
233 	else {
234 		if (hlen == sizeof(struct ip))
235 			sum = in_cksum_hdr(ip);
236 		else
237 			sum = in_cksum(m, hlen);
238 	}
239 	if (sum) {
240 		ipstat.ips_badsum++;
241 		goto drop;
242 	}
243 
244 	/*
245 	 * Remember that we have checked the IP header and found it valid.
246 	 */
247 	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
248 
249 	ip_len = ntohs(ip->ip_len);
250 
251 	/*
252 	 * Is IP length longer than packet we have got?
253 	 */
254 	if (m->m_pkthdr.len < ip_len) {
255 		ipstat.ips_tooshort++;
256 		goto drop;
257 	}
258 
259 	/*
260 	 * Is packet longer than IP header tells us? If yes, truncate packet.
261 	 */
262 	if (m->m_pkthdr.len > ip_len) {
263 		if (m->m_len == m->m_pkthdr.len) {
264 			m->m_len = ip_len;
265 			m->m_pkthdr.len = ip_len;
266 		} else
267 			m_adj(m, ip_len - m->m_pkthdr.len);
268 	}
269 
270 	/*
271 	 * Is packet from or to 127/8?
272 	 */
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++;
276 		goto drop;
277 	}
278 
279 #ifdef ALTQ
280 	/*
281 	 * Is packet dropped by traffic conditioner?
282 	 */
283 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
284 		goto drop;
285 #endif
286 
287 	/*
288 	 * Step 2: fallback conditions to normal ip_input path processing
289 	 */
290 
291 	/*
292 	 * Only IP packets without options
293 	 */
294 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
295 		if (ip_doopts == 1)
296 			return m;
297 		else if (ip_doopts == 2) {
298 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
299 				0, 0);
300 			return NULL;	/* mbuf already free'd */
301 		}
302 		/* else ignore IP options and continue */
303 	}
304 
305 	/*
306 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
307 	 * no multicast, no INADDR_ANY
308 	 *
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.
314 	 */
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 	    ip->ip_src.s_addr == INADDR_ANY ||
322 	    ip->ip_dst.s_addr == INADDR_ANY )
323 		return m;
324 
325 	/*
326 	 * Is it for a local address on this host?
327 	 */
328 	if (in_localip(ip->ip_dst))
329 		return m;
330 
331 	ipstat.ips_total++;
332 
333 	/*
334 	 * Step 3: incoming packet firewall processing
335 	 */
336 
337 	/*
338 	 * Convert to host representation
339 	 */
340 	ip->ip_len = ntohs(ip->ip_len);
341 	ip->ip_off = ntohs(ip->ip_off);
342 
343 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
344 
345 	/*
346 	 * Run through list of ipfilter hooks for input packets
347 	 */
348 	if (!PFIL_HOOKED(&inet_pfil_hook))
349 		goto passin;
350 
351 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) ||
352 	    m == NULL)
353 		goto drop;
354 
355 	M_ASSERTVALID(m);
356 	M_ASSERTPKTHDR(m);
357 
358 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
359 	dest.s_addr = ip->ip_dst.s_addr;
360 
361 	/*
362 	 * Destination address changed?
363 	 */
364 	if (odest.s_addr != dest.s_addr) {
365 		/*
366 		 * Is it now for a local address on this host?
367 		 */
368 		if (in_localip(dest))
369 			goto forwardlocal;
370 		/*
371 		 * Go on with new destination address
372 		 */
373 	}
374 #ifdef IPFIREWALL_FORWARD
375 	if (m->m_flags & M_FASTFWD_OURS) {
376 		/*
377 		 * ipfw changed it for a local address on this host.
378 		 */
379 		goto forwardlocal;
380 	}
381 #endif /* IPFIREWALL_FORWARD */
382 
383 passin:
384 	/*
385 	 * Step 4: decrement TTL and look up route
386 	 */
387 
388 	/*
389 	 * Check TTL
390 	 */
391 #ifdef IPSTEALTH
392 	if (!ipstealth) {
393 #endif
394 	if (ip->ip_ttl <= IPTTLDEC) {
395 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
396 		return NULL;	/* mbuf already free'd */
397 	}
398 
399 	/*
400 	 * Decrement the TTL and incrementally change the IP header checksum.
401 	 * Don't bother doing this with hw checksum offloading, it's faster
402 	 * doing it right here.
403 	 */
404 	ip->ip_ttl -= IPTTLDEC;
405 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
406 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
407 	else
408 		ip->ip_sum += htons(IPTTLDEC << 8);
409 #ifdef IPSTEALTH
410 	}
411 #endif
412 
413 	/*
414 	 * Find route to destination.
415 	 */
416 	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
417 		return NULL;	/* icmp unreach already sent */
418 	ifp = ro.ro_rt->rt_ifp;
419 
420 	/*
421 	 * Immediately drop blackholed traffic.
422 	 */
423 	if (ro.ro_rt->rt_flags & RTF_BLACKHOLE)
424 		goto drop;
425 
426 	/*
427 	 * Step 5: outgoing firewall packet processing
428 	 */
429 
430 	/*
431 	 * Run through list of hooks for output packets.
432 	 */
433 	if (!PFIL_HOOKED(&inet_pfil_hook))
434 		goto passout;
435 
436 	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) {
437 		goto drop;
438 	}
439 
440 	M_ASSERTVALID(m);
441 	M_ASSERTPKTHDR(m);
442 
443 	ip = mtod(m, struct ip *);
444 	dest.s_addr = ip->ip_dst.s_addr;
445 
446 	/*
447 	 * Destination address changed?
448 	 */
449 #ifndef IPFIREWALL_FORWARD
450 	if (odest.s_addr != dest.s_addr) {
451 #else
452 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
453 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
454 #endif /* IPFIREWALL_FORWARD */
455 		/*
456 		 * Is it now for a local address on this host?
457 		 */
458 #ifndef IPFIREWALL_FORWARD
459 		if (in_localip(dest)) {
460 #else
461 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
462 #endif /* IPFIREWALL_FORWARD */
463 forwardlocal:
464 			/*
465 			 * Return packet for processing by ip_input().
466 			 * Keep host byte order as expected at ip_input's
467 			 * "ours"-label.
468 			 */
469 			m->m_flags |= M_FASTFWD_OURS;
470 			if (ro.ro_rt)
471 				RTFREE(ro.ro_rt);
472 			return m;
473 		}
474 		/*
475 		 * Redo route lookup with new destination address
476 		 */
477 #ifdef IPFIREWALL_FORWARD
478 		if (fwd_tag) {
479 			dest.s_addr = ((struct sockaddr_in *)
480 				    (fwd_tag + 1))->sin_addr.s_addr;
481 			m_tag_delete(m, fwd_tag);
482 		}
483 #endif /* IPFIREWALL_FORWARD */
484 		RTFREE(ro.ro_rt);
485 		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
486 			return NULL;	/* icmp unreach already sent */
487 		ifp = ro.ro_rt->rt_ifp;
488 	}
489 
490 passout:
491 	/*
492 	 * Step 6: send off the packet
493 	 */
494 
495 	/*
496 	 * Check if route is dampned (when ARP is unable to resolve)
497 	 */
498 	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
499 	    ro.ro_rt->rt_rmx.rmx_expire >= time_uptime) {
500 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
501 		goto consumed;
502 	}
503 
504 #ifndef ALTQ
505 	/*
506 	 * Check if there is enough space in the interface queue
507 	 */
508 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
509 	    ifp->if_snd.ifq_maxlen) {
510 		ipstat.ips_odropped++;
511 		/* would send source quench here but that is depreciated */
512 		goto drop;
513 	}
514 #endif
515 
516 	/*
517 	 * Check if media link state of interface is not down
518 	 */
519 	if (ifp->if_link_state == LINK_STATE_DOWN) {
520 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
521 		goto consumed;
522 	}
523 
524 	/*
525 	 * Check if packet fits MTU or if hardware will fragment for us
526 	 */
527 	if (ro.ro_rt->rt_rmx.rmx_mtu)
528 		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
529 	else
530 		mtu = ifp->if_mtu;
531 
532 	if (ip->ip_len <= mtu ||
533 	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
534 		/*
535 		 * Restore packet header fields to original values
536 		 */
537 		ip->ip_len = htons(ip->ip_len);
538 		ip->ip_off = htons(ip->ip_off);
539 		/*
540 		 * Send off the packet via outgoing interface
541 		 */
542 		error = (*ifp->if_output)(ifp, m,
543 				(struct sockaddr *)dst, ro.ro_rt);
544 	} else {
545 		/*
546 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
547 		 */
548 		if (ip->ip_off & IP_DF) {
549 			ipstat.ips_cantfrag++;
550 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
551 				0, mtu);
552 			goto consumed;
553 		} else {
554 			/*
555 			 * We have to fragment the packet
556 			 */
557 			m->m_pkthdr.csum_flags |= CSUM_IP;
558 			/*
559 			 * ip_fragment expects ip_len and ip_off in host byte
560 			 * order but returns all packets in network byte order
561 			 */
562 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
563 					(~ifp->if_hwassist & CSUM_DELAY_IP))) {
564 				goto drop;
565 			}
566 			KASSERT(m != NULL, ("null mbuf and no error"));
567 			/*
568 			 * Send off the fragments via outgoing interface
569 			 */
570 			error = 0;
571 			do {
572 				m0 = m->m_nextpkt;
573 				m->m_nextpkt = NULL;
574 
575 				error = (*ifp->if_output)(ifp, m,
576 					(struct sockaddr *)dst, ro.ro_rt);
577 				if (error)
578 					break;
579 			} while ((m = m0) != NULL);
580 			if (error) {
581 				/* Reclaim remaining fragments */
582 				for (m = m0; m; m = m0) {
583 					m0 = m->m_nextpkt;
584 					m_freem(m);
585 				}
586 			} else
587 				ipstat.ips_fragmented++;
588 		}
589 	}
590 
591 	if (error != 0)
592 		ipstat.ips_odropped++;
593 	else {
594 		ro.ro_rt->rt_rmx.rmx_pksent++;
595 		ipstat.ips_forward++;
596 		ipstat.ips_fastforward++;
597 	}
598 consumed:
599 	RTFREE(ro.ro_rt);
600 	return NULL;
601 drop:
602 	if (m)
603 		m_freem(m);
604 	if (ro.ro_rt)
605 		RTFREE(ro.ro_rt);
606 	return NULL;
607 }
608