xref: /freebsd/sys/netinet/ip_fastfwd.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
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 
30 /*
31  * ip_fastforward gets its speed from processing the forwarded packet to
32  * completion (if_output on the other side) without any queues or netisr's.
33  * The receiving interface DMAs the packet into memory, the upper half of
34  * driver calls ip_fastforward, we do our routing table lookup and directly
35  * send it off to the outgoing interface, which DMAs the packet to the
36  * network card. The only part of the packet we touch with the CPU is the
37  * IP header (unless there are complex firewall rules touching other parts
38  * of the packet, but that is up to you). We are essentially limited by bus
39  * bandwidth and how fast the network card/driver can set up receives and
40  * transmits.
41  *
42  * We handle basic errors, IP header errors, checksum errors,
43  * destination unreachable, fragmentation and fragmentation needed and
44  * report them via ICMP to the sender.
45  *
46  * Else if something is not pure IPv4 unicast forwarding we fall back to
47  * the normal ip_input processing path. We should only be called from
48  * interfaces connected to the outside world.
49  *
50  * Firewalling is fully supported including divert, ipfw fwd and ipfilter
51  * ipnat and address rewrite.
52  *
53  * IPSEC is not supported if this host is a tunnel broker. IPSEC is
54  * supported for connections to/from local host.
55  *
56  * We try to do the least expensive (in CPU ops) checks and operations
57  * first to catch junk with as little overhead as possible.
58  *
59  * We take full advantage of hardware support for IP checksum and
60  * fragmentation offloading.
61  *
62  * We don't do ICMP redirect in the fast forwarding path. I have had my own
63  * cases where two core routers with Zebra routing suite would send millions
64  * ICMP redirects to connected hosts if the destination router was not the
65  * default gateway. In one case it was filling the routing table of a host
66  * with approximately 300.000 cloned redirect entries until it ran out of
67  * kernel memory. However the networking code proved very robust and it didn't
68  * crash or fail in other ways.
69  */
70 
71 /*
72  * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
73  * is being followed here.
74  */
75 
76 #include <sys/cdefs.h>
77 __FBSDID("$FreeBSD$");
78 
79 #include "opt_ipfw.h"
80 #include "opt_ipstealth.h"
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/kernel.h>
85 #include <sys/malloc.h>
86 #include <sys/mbuf.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/sysctl.h>
90 
91 #include <net/pfil.h>
92 #include <net/if.h>
93 #include <net/if_types.h>
94 #include <net/if_var.h>
95 #include <net/if_dl.h>
96 #include <net/route.h>
97 
98 #include <netinet/in.h>
99 #include <netinet/in_systm.h>
100 #include <netinet/in_var.h>
101 #include <netinet/ip.h>
102 #include <netinet/ip_var.h>
103 #include <netinet/ip_icmp.h>
104 #include <netinet/ip_options.h>
105 
106 #include <machine/in_cksum.h>
107 
108 static int ipfastforward_active = 0;
109 SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
110     &ipfastforward_active, 0, "Enable fast IP forwarding");
111 
112 static struct sockaddr_in *
113 ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
114 {
115 	struct sockaddr_in *dst;
116 	struct rtentry *rt;
117 
118 	/*
119 	 * Find route to destination.
120 	 */
121 	bzero(ro, sizeof(*ro));
122 	dst = (struct sockaddr_in *)&ro->ro_dst;
123 	dst->sin_family = AF_INET;
124 	dst->sin_len = sizeof(*dst);
125 	dst->sin_addr.s_addr = dest.s_addr;
126 	rtalloc_ign(ro, RTF_CLONING);
127 
128 	/*
129 	 * Route there and interface still up?
130 	 */
131 	rt = ro->ro_rt;
132 	if (rt && (rt->rt_flags & RTF_UP) &&
133 	    (rt->rt_ifp->if_flags & IFF_UP) &&
134 	    (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) {
135 		if (rt->rt_flags & RTF_GATEWAY)
136 			dst = (struct sockaddr_in *)rt->rt_gateway;
137 	} else {
138 		ipstat.ips_noroute++;
139 		ipstat.ips_cantforward++;
140 		if (rt)
141 			RTFREE(rt);
142 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
143 		return NULL;
144 	}
145 	return dst;
146 }
147 
148 /*
149  * Try to forward a packet based on the destination address.
150  * This is a fast path optimized for the plain forwarding case.
151  * If the packet is handled (and consumed) here then we return 1;
152  * otherwise 0 is returned and the packet should be delivered
153  * to ip_input for full processing.
154  */
155 struct mbuf *
156 ip_fastforward(struct mbuf *m)
157 {
158 	struct ip *ip;
159 	struct mbuf *m0 = NULL;
160 	struct route ro;
161 	struct sockaddr_in *dst = 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 m;
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))) == NULL) {
199 		ipstat.ips_toosmall++;
200 		return NULL;	/* mbuf already free'd */
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)) == NULL) {
223 			ipstat.ips_badhlen++;
224 			return NULL;	/* mbuf already free'd */
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 
245 	/*
246 	 * Remember that we have checked the IP header and found it valid.
247 	 */
248 	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
249 
250 	ip_len = ntohs(ip->ip_len);
251 
252 	/*
253 	 * Is IP length longer than packet we have got?
254 	 */
255 	if (m->m_pkthdr.len < ip_len) {
256 		ipstat.ips_tooshort++;
257 		goto drop;
258 	}
259 
260 	/*
261 	 * Is packet longer than IP header tells us? If yes, truncate packet.
262 	 */
263 	if (m->m_pkthdr.len > ip_len) {
264 		if (m->m_len == m->m_pkthdr.len) {
265 			m->m_len = ip_len;
266 			m->m_pkthdr.len = ip_len;
267 		} else
268 			m_adj(m, ip_len - m->m_pkthdr.len);
269 	}
270 
271 	/*
272 	 * Is packet from or to 127/8?
273 	 */
274 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
275 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
276 		ipstat.ips_badaddr++;
277 		goto drop;
278 	}
279 
280 #ifdef ALTQ
281 	/*
282 	 * Is packet dropped by traffic conditioner?
283 	 */
284 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
285 		goto drop;
286 #endif
287 
288 	/*
289 	 * Step 2: fallback conditions to normal ip_input path processing
290 	 */
291 
292 	/*
293 	 * Only IP packets without options
294 	 */
295 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
296 		if (ip_doopts == 1)
297 			return m;
298 		else if (ip_doopts == 2) {
299 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
300 				0, 0);
301 			return NULL;	/* mbuf already free'd */
302 		}
303 		/* else ignore IP options and continue */
304 	}
305 
306 	/*
307 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
308 	 * no multicast, no INADDR_ANY
309 	 *
310 	 * XXX: Probably some of these checks could be direct drop
311 	 * conditions.  However it is not clear whether there are some
312 	 * hacks or obscure behaviours which make it neccessary to
313 	 * let ip_input handle it.  We play safe here and let ip_input
314 	 * deal with it until it is proven that we can directly drop it.
315 	 */
316 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
317 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
318 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
319 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
320 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
321 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
322 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
323 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
324 	    ip->ip_src.s_addr == INADDR_ANY ||
325 	    ip->ip_dst.s_addr == INADDR_ANY )
326 		return m;
327 
328 	/*
329 	 * Is it for a local address on this host?
330 	 */
331 	if (in_localip(ip->ip_dst))
332 		return m;
333 
334 	ipstat.ips_total++;
335 
336 	/*
337 	 * Step 3: incoming packet firewall processing
338 	 */
339 
340 	/*
341 	 * Convert to host representation
342 	 */
343 	ip->ip_len = ntohs(ip->ip_len);
344 	ip->ip_off = ntohs(ip->ip_off);
345 
346 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
347 
348 	/*
349 	 * Run through list of ipfilter hooks for input packets
350 	 */
351 	if (!PFIL_HOOKED(&inet_pfil_hook))
352 		goto passin;
353 
354 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) ||
355 	    m == NULL)
356 		goto drop;
357 
358 	M_ASSERTVALID(m);
359 	M_ASSERTPKTHDR(m);
360 
361 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
362 	dest.s_addr = ip->ip_dst.s_addr;
363 
364 	/*
365 	 * Destination address changed?
366 	 */
367 	if (odest.s_addr != dest.s_addr) {
368 		/*
369 		 * Is it now for a local address on this host?
370 		 */
371 		if (in_localip(dest))
372 			goto forwardlocal;
373 		/*
374 		 * Go on with new destination address
375 		 */
376 	}
377 #ifdef IPFIREWALL_FORWARD
378 	if (m->m_flags & M_FASTFWD_OURS) {
379 		/*
380 		 * ipfw changed it for a local address on this host.
381 		 */
382 		goto forwardlocal;
383 	}
384 #endif /* IPFIREWALL_FORWARD */
385 
386 passin:
387 	/*
388 	 * Step 4: decrement TTL and look up route
389 	 */
390 
391 	/*
392 	 * Check TTL
393 	 */
394 #ifdef IPSTEALTH
395 	if (!ipstealth) {
396 #endif
397 	if (ip->ip_ttl <= IPTTLDEC) {
398 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
399 		return NULL;	/* mbuf already free'd */
400 	}
401 
402 	/*
403 	 * Decrement the TTL and incrementally change the IP header checksum.
404 	 * Don't bother doing this with hw checksum offloading, it's faster
405 	 * doing it right here.
406 	 */
407 	ip->ip_ttl -= IPTTLDEC;
408 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
409 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
410 	else
411 		ip->ip_sum += htons(IPTTLDEC << 8);
412 #ifdef IPSTEALTH
413 	}
414 #endif
415 
416 	/*
417 	 * Find route to destination.
418 	 */
419 	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
420 		return NULL;	/* icmp unreach already sent */
421 	ifp = ro.ro_rt->rt_ifp;
422 
423 	/*
424 	 * Immediately drop blackholed traffic, and directed broadcasts
425 	 * for either the all-ones or all-zero subnet addresses on
426 	 * locally attached networks.
427 	 */
428 	if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0)
429 		goto drop;
430 
431 	/*
432 	 * Step 5: outgoing firewall packet processing
433 	 */
434 
435 	/*
436 	 * Run through list of hooks for output packets.
437 	 */
438 	if (!PFIL_HOOKED(&inet_pfil_hook))
439 		goto passout;
440 
441 	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) {
442 		goto drop;
443 	}
444 
445 	M_ASSERTVALID(m);
446 	M_ASSERTPKTHDR(m);
447 
448 	ip = mtod(m, struct ip *);
449 	dest.s_addr = ip->ip_dst.s_addr;
450 
451 	/*
452 	 * Destination address changed?
453 	 */
454 #ifndef IPFIREWALL_FORWARD
455 	if (odest.s_addr != dest.s_addr) {
456 #else
457 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
458 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
459 #endif /* IPFIREWALL_FORWARD */
460 		/*
461 		 * Is it now for a local address on this host?
462 		 */
463 #ifndef IPFIREWALL_FORWARD
464 		if (in_localip(dest)) {
465 #else
466 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
467 #endif /* IPFIREWALL_FORWARD */
468 forwardlocal:
469 			/*
470 			 * Return packet for processing by ip_input().
471 			 * Keep host byte order as expected at ip_input's
472 			 * "ours"-label.
473 			 */
474 			m->m_flags |= M_FASTFWD_OURS;
475 			if (ro.ro_rt)
476 				RTFREE(ro.ro_rt);
477 			return m;
478 		}
479 		/*
480 		 * Redo route lookup with new destination address
481 		 */
482 #ifdef IPFIREWALL_FORWARD
483 		if (fwd_tag) {
484 			dest.s_addr = ((struct sockaddr_in *)
485 				    (fwd_tag + 1))->sin_addr.s_addr;
486 			m_tag_delete(m, fwd_tag);
487 		}
488 #endif /* IPFIREWALL_FORWARD */
489 		RTFREE(ro.ro_rt);
490 		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
491 			return NULL;	/* icmp unreach already sent */
492 		ifp = ro.ro_rt->rt_ifp;
493 	}
494 
495 passout:
496 	/*
497 	 * Step 6: send off the packet
498 	 */
499 
500 	/*
501 	 * Check if route is dampned (when ARP is unable to resolve)
502 	 */
503 	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
504 	    (ro.ro_rt->rt_rmx.rmx_expire == 0 ||
505 	    time_uptime < ro.ro_rt->rt_rmx.rmx_expire)) {
506 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
507 		goto consumed;
508 	}
509 
510 #ifndef ALTQ
511 	/*
512 	 * Check if there is enough space in the interface queue
513 	 */
514 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
515 	    ifp->if_snd.ifq_maxlen) {
516 		ipstat.ips_odropped++;
517 		/* would send source quench here but that is depreciated */
518 		goto drop;
519 	}
520 #endif
521 
522 	/*
523 	 * Check if media link state of interface is not down
524 	 */
525 	if (ifp->if_link_state == LINK_STATE_DOWN) {
526 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
527 		goto consumed;
528 	}
529 
530 	/*
531 	 * Check if packet fits MTU or if hardware will fragment for us
532 	 */
533 	if (ro.ro_rt->rt_rmx.rmx_mtu)
534 		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
535 	else
536 		mtu = ifp->if_mtu;
537 
538 	if (ip->ip_len <= mtu ||
539 	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
540 		/*
541 		 * Restore packet header fields to original values
542 		 */
543 		ip->ip_len = htons(ip->ip_len);
544 		ip->ip_off = htons(ip->ip_off);
545 		/*
546 		 * Send off the packet via outgoing interface
547 		 */
548 		error = (*ifp->if_output)(ifp, m,
549 				(struct sockaddr *)dst, ro.ro_rt);
550 	} else {
551 		/*
552 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
553 		 */
554 		if (ip->ip_off & IP_DF) {
555 			ipstat.ips_cantfrag++;
556 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
557 				0, mtu);
558 			goto consumed;
559 		} else {
560 			/*
561 			 * We have to fragment the packet
562 			 */
563 			m->m_pkthdr.csum_flags |= CSUM_IP;
564 			/*
565 			 * ip_fragment expects ip_len and ip_off in host byte
566 			 * order but returns all packets in network byte order
567 			 */
568 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
569 					(~ifp->if_hwassist & CSUM_DELAY_IP))) {
570 				goto drop;
571 			}
572 			KASSERT(m != NULL, ("null mbuf and no error"));
573 			/*
574 			 * Send off the fragments via outgoing interface
575 			 */
576 			error = 0;
577 			do {
578 				m0 = m->m_nextpkt;
579 				m->m_nextpkt = NULL;
580 
581 				error = (*ifp->if_output)(ifp, m,
582 					(struct sockaddr *)dst, ro.ro_rt);
583 				if (error)
584 					break;
585 			} while ((m = m0) != NULL);
586 			if (error) {
587 				/* Reclaim remaining fragments */
588 				for (m = m0; m; m = m0) {
589 					m0 = m->m_nextpkt;
590 					m_freem(m);
591 				}
592 			} else
593 				ipstat.ips_fragmented++;
594 		}
595 	}
596 
597 	if (error != 0)
598 		ipstat.ips_odropped++;
599 	else {
600 		ro.ro_rt->rt_rmx.rmx_pksent++;
601 		ipstat.ips_forward++;
602 		ipstat.ips_fastforward++;
603 	}
604 consumed:
605 	RTFREE(ro.ro_rt);
606 	return NULL;
607 drop:
608 	if (m)
609 		m_freem(m);
610 	if (ro.ro_rt)
611 		RTFREE(ro.ro_rt);
612 	return NULL;
613 }
614