xref: /freebsd/sys/netinet/ip_fastfwd.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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 #include <net/vnet.h>
98 
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/in_var.h>
102 #include <netinet/ip.h>
103 #include <netinet/ip_var.h>
104 #include <netinet/ip_icmp.h>
105 #include <netinet/ip_options.h>
106 
107 #include <machine/in_cksum.h>
108 
109 static VNET_DEFINE(int, ipfastforward_active);
110 #define	V_ipfastforward_active		VNET(ipfastforward_active)
111 
112 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
113     &VNET_NAME(ipfastforward_active), 0, "Enable fast IP forwarding");
114 
115 static struct sockaddr_in *
116 ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
117 {
118 	struct sockaddr_in *dst;
119 	struct rtentry *rt;
120 
121 	/*
122 	 * Find route to destination.
123 	 */
124 	bzero(ro, sizeof(*ro));
125 	dst = (struct sockaddr_in *)&ro->ro_dst;
126 	dst->sin_family = AF_INET;
127 	dst->sin_len = sizeof(*dst);
128 	dst->sin_addr.s_addr = dest.s_addr;
129 	in_rtalloc_ign(ro, 0, M_GETFIB(m));
130 
131 	/*
132 	 * Route there and interface still up?
133 	 */
134 	rt = ro->ro_rt;
135 	if (rt && (rt->rt_flags & RTF_UP) &&
136 	    (rt->rt_ifp->if_flags & IFF_UP) &&
137 	    (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) {
138 		if (rt->rt_flags & RTF_GATEWAY)
139 			dst = (struct sockaddr_in *)rt->rt_gateway;
140 	} else {
141 		IPSTAT_INC(ips_noroute);
142 		IPSTAT_INC(ips_cantforward);
143 		if (rt)
144 			RTFREE(rt);
145 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
146 		return NULL;
147 	}
148 	return dst;
149 }
150 
151 /*
152  * Try to forward a packet based on the destination address.
153  * This is a fast path optimized for the plain forwarding case.
154  * If the packet is handled (and consumed) here then we return NULL;
155  * otherwise mbuf is returned and the packet should be delivered
156  * to ip_input for full processing.
157  */
158 struct mbuf *
159 ip_fastforward(struct mbuf *m)
160 {
161 	struct ip *ip;
162 	struct mbuf *m0 = NULL;
163 	struct route ro;
164 	struct sockaddr_in *dst = NULL;
165 	struct ifnet *ifp;
166 	struct in_addr odest, dest;
167 	uint16_t sum, ip_len, ip_off;
168 	int error = 0;
169 	int hlen, mtu;
170 	struct m_tag *fwd_tag = NULL;
171 
172 	/*
173 	 * Are we active and forwarding packets?
174 	 */
175 	if (!V_ipfastforward_active || !V_ipforwarding)
176 		return m;
177 
178 	M_ASSERTVALID(m);
179 	M_ASSERTPKTHDR(m);
180 
181 	bzero(&ro, sizeof(ro));
182 
183 	/*
184 	 * Step 1: check for packet drop conditions (and sanity checks)
185 	 */
186 
187 	/*
188 	 * Is entire packet big enough?
189 	 */
190 	if (m->m_pkthdr.len < sizeof(struct ip)) {
191 		IPSTAT_INC(ips_tooshort);
192 		goto drop;
193 	}
194 
195 	/*
196 	 * Is first mbuf large enough for ip header and is header present?
197 	 */
198 	if (m->m_len < sizeof (struct ip) &&
199 	   (m = m_pullup(m, sizeof (struct ip))) == NULL) {
200 		IPSTAT_INC(ips_toosmall);
201 		return NULL;	/* mbuf already free'd */
202 	}
203 
204 	ip = mtod(m, struct ip *);
205 
206 	/*
207 	 * Is it IPv4?
208 	 */
209 	if (ip->ip_v != IPVERSION) {
210 		IPSTAT_INC(ips_badvers);
211 		goto drop;
212 	}
213 
214 	/*
215 	 * Is IP header length correct and is it in first mbuf?
216 	 */
217 	hlen = ip->ip_hl << 2;
218 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
219 		IPSTAT_INC(ips_badhlen);
220 		goto drop;
221 	}
222 	if (hlen > m->m_len) {
223 		if ((m = m_pullup(m, hlen)) == NULL) {
224 			IPSTAT_INC(ips_badhlen);
225 			return NULL;	/* mbuf already free'd */
226 		}
227 		ip = mtod(m, struct ip *);
228 	}
229 
230 	/*
231 	 * Checksum correct?
232 	 */
233 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
234 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
235 	else {
236 		if (hlen == sizeof(struct ip))
237 			sum = in_cksum_hdr(ip);
238 		else
239 			sum = in_cksum(m, hlen);
240 	}
241 	if (sum) {
242 		IPSTAT_INC(ips_badsum);
243 		goto drop;
244 	}
245 
246 	/*
247 	 * Remember that we have checked the IP header and found it valid.
248 	 */
249 	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
250 
251 	ip_len = ntohs(ip->ip_len);
252 
253 	/*
254 	 * Is IP length longer than packet we have got?
255 	 */
256 	if (m->m_pkthdr.len < ip_len) {
257 		IPSTAT_INC(ips_tooshort);
258 		goto drop;
259 	}
260 
261 	/*
262 	 * Is packet longer than IP header tells us? If yes, truncate packet.
263 	 */
264 	if (m->m_pkthdr.len > ip_len) {
265 		if (m->m_len == m->m_pkthdr.len) {
266 			m->m_len = ip_len;
267 			m->m_pkthdr.len = ip_len;
268 		} else
269 			m_adj(m, ip_len - m->m_pkthdr.len);
270 	}
271 
272 	/*
273 	 * Is packet from or to 127/8?
274 	 */
275 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
276 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
277 		IPSTAT_INC(ips_badaddr);
278 		goto drop;
279 	}
280 
281 #ifdef ALTQ
282 	/*
283 	 * Is packet dropped by traffic conditioner?
284 	 */
285 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
286 		goto drop;
287 #endif
288 
289 	/*
290 	 * Step 2: fallback conditions to normal ip_input path processing
291 	 */
292 
293 	/*
294 	 * Only IP packets without options
295 	 */
296 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
297 		if (ip_doopts == 1)
298 			return m;
299 		else if (ip_doopts == 2) {
300 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
301 				0, 0);
302 			return NULL;	/* mbuf already free'd */
303 		}
304 		/* else ignore IP options and continue */
305 	}
306 
307 	/*
308 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
309 	 * no multicast, no INADDR_ANY
310 	 *
311 	 * XXX: Probably some of these checks could be direct drop
312 	 * conditions.  However it is not clear whether there are some
313 	 * hacks or obscure behaviours which make it neccessary to
314 	 * let ip_input handle it.  We play safe here and let ip_input
315 	 * deal with it until it is proven that we can directly drop it.
316 	 */
317 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
318 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
319 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
320 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
321 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
322 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
323 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
324 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
325 	    ip->ip_src.s_addr == INADDR_ANY ||
326 	    ip->ip_dst.s_addr == INADDR_ANY )
327 		return m;
328 
329 	/*
330 	 * Is it for a local address on this host?
331 	 */
332 	if (in_localip(ip->ip_dst))
333 		return m;
334 
335 	IPSTAT_INC(ips_total);
336 
337 	/*
338 	 * Step 3: incoming packet firewall processing
339 	 */
340 
341 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
342 
343 	/*
344 	 * Run through list of ipfilter hooks for input packets
345 	 */
346 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
347 		goto passin;
348 
349 	if (pfil_run_hooks(
350 	    &V_inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) ||
351 	    m == NULL)
352 		goto drop;
353 
354 	M_ASSERTVALID(m);
355 	M_ASSERTPKTHDR(m);
356 
357 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
358 	dest.s_addr = ip->ip_dst.s_addr;
359 
360 	/*
361 	 * Destination address changed?
362 	 */
363 	if (odest.s_addr != dest.s_addr) {
364 		/*
365 		 * Is it now for a local address on this host?
366 		 */
367 		if (in_localip(dest))
368 			goto forwardlocal;
369 		/*
370 		 * Go on with new destination address
371 		 */
372 	}
373 
374 	if (m->m_flags & M_FASTFWD_OURS) {
375 		/*
376 		 * ipfw changed it for a local address on this host.
377 		 */
378 		goto forwardlocal;
379 	}
380 
381 passin:
382 	/*
383 	 * Step 4: decrement TTL and look up route
384 	 */
385 
386 	/*
387 	 * Check TTL
388 	 */
389 #ifdef IPSTEALTH
390 	if (!V_ipstealth) {
391 #endif
392 	if (ip->ip_ttl <= IPTTLDEC) {
393 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
394 		return NULL;	/* mbuf already free'd */
395 	}
396 
397 	/*
398 	 * Decrement the TTL and incrementally change the IP header checksum.
399 	 * Don't bother doing this with hw checksum offloading, it's faster
400 	 * doing it right here.
401 	 */
402 	ip->ip_ttl -= IPTTLDEC;
403 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
404 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
405 	else
406 		ip->ip_sum += htons(IPTTLDEC << 8);
407 #ifdef IPSTEALTH
408 	}
409 #endif
410 
411 	/*
412 	 * Find route to destination.
413 	 */
414 	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
415 		return NULL;	/* icmp unreach already sent */
416 	ifp = ro.ro_rt->rt_ifp;
417 
418 	/*
419 	 * Immediately drop blackholed traffic, and directed broadcasts
420 	 * for either the all-ones or all-zero subnet addresses on
421 	 * locally attached networks.
422 	 */
423 	if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0)
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(&V_inet_pfil_hook))
434 		goto passout;
435 
436 	if (pfil_run_hooks(&V_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 	if (m->m_flags & M_IP_NEXTHOP)
450 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
451 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
452 		/*
453 		 * Is it now for a local address on this host?
454 		 */
455 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
456 forwardlocal:
457 			/*
458 			 * Return packet for processing by ip_input().
459 			 */
460 			m->m_flags |= M_FASTFWD_OURS;
461 			if (ro.ro_rt)
462 				RTFREE(ro.ro_rt);
463 			return m;
464 		}
465 		/*
466 		 * Redo route lookup with new destination address
467 		 */
468 		if (fwd_tag) {
469 			dest.s_addr = ((struct sockaddr_in *)
470 				    (fwd_tag + 1))->sin_addr.s_addr;
471 			m_tag_delete(m, fwd_tag);
472 			m->m_flags &= ~M_IP_NEXTHOP;
473 		}
474 		RTFREE(ro.ro_rt);
475 		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
476 			return NULL;	/* icmp unreach already sent */
477 		ifp = ro.ro_rt->rt_ifp;
478 	}
479 
480 passout:
481 	/*
482 	 * Step 6: send off the packet
483 	 */
484 	ip_len = ntohs(ip->ip_len);
485 	ip_off = ntohs(ip->ip_off);
486 
487 	/*
488 	 * Check if route is dampned (when ARP is unable to resolve)
489 	 */
490 	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
491 	    (ro.ro_rt->rt_rmx.rmx_expire == 0 ||
492 	    time_uptime < ro.ro_rt->rt_rmx.rmx_expire)) {
493 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
494 		goto consumed;
495 	}
496 
497 #ifndef ALTQ
498 	/*
499 	 * Check if there is enough space in the interface queue
500 	 */
501 	if ((ifp->if_snd.ifq_len + ip_len / ifp->if_mtu + 1) >=
502 	    ifp->if_snd.ifq_maxlen) {
503 		IPSTAT_INC(ips_odropped);
504 		/* would send source quench here but that is depreciated */
505 		goto drop;
506 	}
507 #endif
508 
509 	/*
510 	 * Check if media link state of interface is not down
511 	 */
512 	if (ifp->if_link_state == LINK_STATE_DOWN) {
513 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
514 		goto consumed;
515 	}
516 
517 	/*
518 	 * Check if packet fits MTU or if hardware will fragment for us
519 	 */
520 	if (ro.ro_rt->rt_rmx.rmx_mtu)
521 		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
522 	else
523 		mtu = ifp->if_mtu;
524 
525 	if (ip_len <= mtu ||
526 	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip_off & IP_DF) == 0)) {
527 		/*
528 		 * Send off the packet via outgoing interface
529 		 */
530 		error = (*ifp->if_output)(ifp, m,
531 				(struct sockaddr *)dst, &ro);
532 	} else {
533 		/*
534 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
535 		 */
536 		if (ip_off & IP_DF) {
537 			IPSTAT_INC(ips_cantfrag);
538 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
539 				0, mtu);
540 			goto consumed;
541 		} else {
542 			/*
543 			 * We have to fragment the packet
544 			 */
545 			m->m_pkthdr.csum_flags |= CSUM_IP;
546 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist))
547 				goto drop;
548 			KASSERT(m != NULL, ("null mbuf and no error"));
549 			/*
550 			 * Send off the fragments via outgoing interface
551 			 */
552 			error = 0;
553 			do {
554 				m0 = m->m_nextpkt;
555 				m->m_nextpkt = NULL;
556 
557 				error = (*ifp->if_output)(ifp, m,
558 					(struct sockaddr *)dst, &ro);
559 				if (error)
560 					break;
561 			} while ((m = m0) != NULL);
562 			if (error) {
563 				/* Reclaim remaining fragments */
564 				for (m = m0; m; m = m0) {
565 					m0 = m->m_nextpkt;
566 					m_freem(m);
567 				}
568 			} else
569 				IPSTAT_INC(ips_fragmented);
570 		}
571 	}
572 
573 	if (error != 0)
574 		IPSTAT_INC(ips_odropped);
575 	else {
576 		ro.ro_rt->rt_rmx.rmx_pksent++;
577 		IPSTAT_INC(ips_forward);
578 		IPSTAT_INC(ips_fastforward);
579 	}
580 consumed:
581 	RTFREE(ro.ro_rt);
582 	return NULL;
583 drop:
584 	if (m)
585 		m_freem(m);
586 	if (ro.ro_rt)
587 		RTFREE(ro.ro_rt);
588 	return NULL;
589 }
590