xref: /freebsd/sys/netinet/ip_fastfwd.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2003 Andre Oppermann, Internet Business Solutions AG
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote
16  *    products derived from this software without specific prior written
17  *    permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
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 
65 /*
66  * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
67  * is being followed here.
68  */
69 
70 #include <sys/cdefs.h>
71 __FBSDID("$FreeBSD$");
72 
73 #include "opt_ipstealth.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/mbuf.h>
80 #include <sys/protosw.h>
81 #include <sys/sdt.h>
82 #include <sys/socket.h>
83 #include <sys/sysctl.h>
84 
85 #include <net/if.h>
86 #include <net/if_types.h>
87 #include <net/if_var.h>
88 #include <net/if_dl.h>
89 #include <net/pfil.h>
90 #include <net/route.h>
91 #include <net/route/nhop.h>
92 #include <net/vnet.h>
93 
94 #include <netinet/in.h>
95 #include <netinet/in_fib.h>
96 #include <netinet/in_kdtrace.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/in_var.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip_var.h>
101 #include <netinet/ip_icmp.h>
102 #include <netinet/ip_options.h>
103 
104 #include <machine/in_cksum.h>
105 
106 #define	V_ipsendredirects	VNET(ipsendredirects)
107 
108 static struct mbuf *
109 ip_redir_alloc(struct mbuf *m, struct nhop_object *nh, u_short ip_len,
110     struct in_addr *osrc, struct in_addr *newgw)
111 {
112 	struct in_ifaddr *nh_ia;
113 	struct mbuf *mcopy;
114 
115 	KASSERT(nh != NULL, ("%s: m %p nh is NULL\n", __func__, m));
116 
117 	/*
118 	 * Only send a redirect if:
119 	 * - Redirects are not disabled (must be checked by caller),
120 	 * - We have not applied NAT (must be checked by caller as possible),
121 	 * - Neither a MCAST or BCAST packet (must be checked by caller)
122 	 *   [RFC1009 Appendix A.2].
123 	 * - The packet does not do IP source routing or having any other
124 	 *   IP options (this case was handled already by ip_input() calling
125 	 *   ip_dooptions() [RFC792, p13],
126 	 * - The packet is being forwarded out the same physical interface
127 	 *   that it was received from [RFC1812, 5.2.7.2].
128 	 */
129 
130 	/*
131 	 * - The forwarding route was not created by a redirect
132 	 *   [RFC1812, 5.2.7.2], or
133 	 *   if it was to follow a default route (see below).
134 	 * - The next-hop is reachable by us [RFC1009 Appendix A.2].
135 	 */
136 	if ((nh->nh_flags & (NHF_DEFAULT | NHF_REDIRECT |
137 	    NHF_BLACKHOLE | NHF_REJECT)) != 0)
138 		return (NULL);
139 
140 	/* Get the new gateway. */
141 	if ((nh->nh_flags & NHF_GATEWAY) == 0 || nh->gw_sa.sa_family != AF_INET)
142 		return (NULL);
143 	newgw->s_addr = nh->gw4_sa.sin_addr.s_addr;
144 
145 	/*
146 	 * - The resulting forwarding destination is not "This host on this
147 	 *   network" [RFC1122, Section 3.2.1.3] (default route check above).
148 	 */
149 	if (newgw->s_addr == 0)
150 		return (NULL);
151 
152 	/*
153 	 * - We know how to reach the sender and the source address is
154 	 *   directly connected to us [RFC792, p13].
155 	 * + The new gateway address and the source address are on the same
156 	 *   subnet [RFC1009 Appendix A.2, RFC1122 3.2.2.2, RFC1812, 5.2.7.2].
157 	 * NB: if you think multiple logical subnets on the same wire should
158 	 *     receive redirects read [RFC1812, APPENDIX C (14->15)].
159 	 */
160 	nh_ia = (struct in_ifaddr *)nh->nh_ifa;
161 	if ((ntohl(osrc->s_addr) & nh_ia->ia_subnetmask) != nh_ia->ia_subnet)
162 		return (NULL);
163 
164 	/* Prepare for sending the redirect. */
165 
166 	/*
167 	 * Make a copy of as much as we need of the packet as the original
168 	 * one will be forwarded but we need (a portion) for icmp_error().
169 	 */
170 	mcopy = m_gethdr(M_NOWAIT, m->m_type);
171 	if (mcopy == NULL)
172 		return (NULL);
173 
174 	if (m_dup_pkthdr(mcopy, m, M_NOWAIT) == 0) {
175 		/*
176 		 * It's probably ok if the pkthdr dup fails (because
177 		 * the deep copy of the tag chain failed), but for now
178 		 * be conservative and just discard the copy since
179 		 * code below may some day want the tags.
180 		 */
181 		m_free(mcopy);
182 		return (NULL);
183 	}
184 	mcopy->m_len = min(ip_len, M_TRAILINGSPACE(mcopy));
185 	mcopy->m_pkthdr.len = mcopy->m_len;
186 	m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
187 
188 	return (mcopy);
189 }
190 
191 
192 static int
193 ip_findroute(struct nhop_object **pnh, struct in_addr dest, struct mbuf *m)
194 {
195 	struct nhop_object *nh;
196 
197 	nh = fib4_lookup(M_GETFIB(m), dest, 0, NHR_NONE,
198 	    m->m_pkthdr.flowid);
199 	if (nh == NULL) {
200 		IPSTAT_INC(ips_noroute);
201 		IPSTAT_INC(ips_cantforward);
202 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
203 		return (EHOSTUNREACH);
204 	}
205 	/*
206 	 * Drop blackholed traffic and directed broadcasts.
207 	 */
208 	if ((nh->nh_flags & (NHF_BLACKHOLE | NHF_BROADCAST)) != 0) {
209 		IPSTAT_INC(ips_cantforward);
210 		m_freem(m);
211 		return (EHOSTUNREACH);
212 	}
213 
214 	if (nh->nh_flags & NHF_REJECT) {
215 		IPSTAT_INC(ips_cantforward);
216 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
217 		return (EHOSTUNREACH);
218 	}
219 
220 	*pnh = nh;
221 
222 	return (0);
223 }
224 
225 /*
226  * Try to forward a packet based on the destination address.
227  * This is a fast path optimized for the plain forwarding case.
228  * If the packet is handled (and consumed) here then we return NULL;
229  * otherwise mbuf is returned and the packet should be delivered
230  * to ip_input for full processing.
231  */
232 struct mbuf *
233 ip_tryforward(struct mbuf *m)
234 {
235 	struct ip *ip;
236 	struct mbuf *m0 = NULL;
237 	struct nhop_object *nh = NULL;
238 	struct route ro;
239 	struct sockaddr_in *dst;
240 	const struct sockaddr *gw;
241 	struct in_addr dest, odest, rtdest, osrc;
242 	uint16_t ip_len, ip_off;
243 	int error = 0;
244 	struct m_tag *fwd_tag = NULL;
245 	struct mbuf *mcopy = NULL;
246 	struct in_addr redest;
247 	/*
248 	 * Are we active and forwarding packets?
249 	 */
250 
251 	M_ASSERTVALID(m);
252 	M_ASSERTPKTHDR(m);
253 
254 #ifdef ALTQ
255 	/*
256 	 * Is packet dropped by traffic conditioner?
257 	 */
258 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
259 		goto drop;
260 #endif
261 
262 	/*
263 	 * Only IP packets without options
264 	 */
265 	ip = mtod(m, struct ip *);
266 
267 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
268 		if (V_ip_doopts == 1)
269 			return m;
270 		else if (V_ip_doopts == 2) {
271 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
272 				0, 0);
273 			return NULL;	/* mbuf already free'd */
274 		}
275 		/* else ignore IP options and continue */
276 	}
277 
278 	/*
279 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
280 	 * no multicast, no INADDR_ANY
281 	 *
282 	 * XXX: Probably some of these checks could be direct drop
283 	 * conditions.  However it is not clear whether there are some
284 	 * hacks or obscure behaviours which make it necessary to
285 	 * let ip_input handle it.  We play safe here and let ip_input
286 	 * deal with it until it is proven that we can directly drop it.
287 	 */
288 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
289 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
290 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
291 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
292 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
293 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
294 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
295 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
296 	    ip->ip_src.s_addr == INADDR_ANY ||
297 	    ip->ip_dst.s_addr == INADDR_ANY )
298 		return m;
299 
300 	/*
301 	 * Is it for a local address on this host?
302 	 */
303 	if (in_localip(ip->ip_dst))
304 		return m;
305 
306 	IPSTAT_INC(ips_total);
307 
308 	/*
309 	 * Step 3: incoming packet firewall processing
310 	 */
311 
312 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
313 	osrc.s_addr = ip->ip_src.s_addr;
314 
315 	/*
316 	 * Run through list of ipfilter hooks for input packets
317 	 */
318 	if (!PFIL_HOOKED_IN(V_inet_pfil_head))
319 		goto passin;
320 
321 	if (pfil_run_hooks(V_inet_pfil_head, &m, m->m_pkthdr.rcvif, PFIL_IN,
322 	    NULL) != PFIL_PASS)
323 		goto drop;
324 
325 	M_ASSERTVALID(m);
326 	M_ASSERTPKTHDR(m);
327 
328 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
329 	dest.s_addr = ip->ip_dst.s_addr;
330 
331 	/*
332 	 * Destination address changed?
333 	 */
334 	if (odest.s_addr != dest.s_addr) {
335 		/*
336 		 * Is it now for a local address on this host?
337 		 */
338 		if (in_localip(dest))
339 			goto forwardlocal;
340 		/*
341 		 * Go on with new destination address
342 		 */
343 	}
344 
345 	if (m->m_flags & M_FASTFWD_OURS) {
346 		/*
347 		 * ipfw changed it for a local address on this host.
348 		 */
349 		goto forwardlocal;
350 	}
351 
352 passin:
353 	/*
354 	 * Step 4: decrement TTL and look up route
355 	 */
356 
357 	/*
358 	 * Check TTL
359 	 */
360 #ifdef IPSTEALTH
361 	if (!V_ipstealth) {
362 #endif
363 	if (ip->ip_ttl <= IPTTLDEC) {
364 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
365 		return NULL;	/* mbuf already free'd */
366 	}
367 
368 	/*
369 	 * Decrement the TTL and incrementally change the IP header checksum.
370 	 * Don't bother doing this with hw checksum offloading, it's faster
371 	 * doing it right here.
372 	 */
373 	ip->ip_ttl -= IPTTLDEC;
374 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
375 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
376 	else
377 		ip->ip_sum += htons(IPTTLDEC << 8);
378 #ifdef IPSTEALTH
379 	}
380 #endif
381 
382 	/*
383 	 * Next hop forced by pfil(9) hook?
384 	 */
385 	if ((m->m_flags & M_IP_NEXTHOP) &&
386 	    ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
387 		/*
388 		 * Now we will find route to forced destination.
389 		 */
390 		dest.s_addr = ((struct sockaddr_in *)
391 			    (fwd_tag + 1))->sin_addr.s_addr;
392 		m_tag_delete(m, fwd_tag);
393 		m->m_flags &= ~M_IP_NEXTHOP;
394 	}
395 
396 	/*
397 	 * Find route to destination.
398 	 */
399 	if (ip_findroute(&nh, dest, m) != 0)
400 		return (NULL);	/* icmp unreach already sent */
401 
402 	/*
403 	 * Avoid second route lookup by caching destination.
404 	 */
405 	rtdest.s_addr = dest.s_addr;
406 
407 	/*
408 	 * Step 5: outgoing firewall packet processing
409 	 */
410 	if (!PFIL_HOOKED_OUT(V_inet_pfil_head))
411 		goto passout;
412 
413 	if (pfil_run_hooks(V_inet_pfil_head, &m, nh->nh_ifp,
414 	    PFIL_OUT | PFIL_FWD, NULL) != PFIL_PASS)
415 		goto drop;
416 
417 	M_ASSERTVALID(m);
418 	M_ASSERTPKTHDR(m);
419 
420 	ip = mtod(m, struct ip *);
421 	dest.s_addr = ip->ip_dst.s_addr;
422 
423 	/*
424 	 * Destination address changed?
425 	 */
426 	if (m->m_flags & M_IP_NEXTHOP)
427 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
428 	else
429 		fwd_tag = NULL;
430 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
431 		/*
432 		 * Is it now for a local address on this host?
433 		 */
434 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
435 forwardlocal:
436 			/*
437 			 * Return packet for processing by ip_input().
438 			 */
439 			m->m_flags |= M_FASTFWD_OURS;
440 			return (m);
441 		}
442 		/*
443 		 * Redo route lookup with new destination address
444 		 */
445 		if (fwd_tag) {
446 			dest.s_addr = ((struct sockaddr_in *)
447 				    (fwd_tag + 1))->sin_addr.s_addr;
448 			m_tag_delete(m, fwd_tag);
449 			m->m_flags &= ~M_IP_NEXTHOP;
450 		}
451 		if (dest.s_addr != rtdest.s_addr &&
452 		    ip_findroute(&nh, dest, m) != 0)
453 			return (NULL);	/* icmp unreach already sent */
454 	}
455 
456 passout:
457 	/*
458 	 * Step 6: send off the packet
459 	 */
460 	ip_len = ntohs(ip->ip_len);
461 	ip_off = ntohs(ip->ip_off);
462 
463 	bzero(&ro, sizeof(ro));
464 	dst = (struct sockaddr_in *)&ro.ro_dst;
465 	dst->sin_family = AF_INET;
466 	dst->sin_len = sizeof(*dst);
467 	dst->sin_addr = dest;
468 	if (nh->nh_flags & NHF_GATEWAY) {
469 		gw = &nh->gw_sa;
470 		ro.ro_flags |= RT_HAS_GW;
471 	} else
472 		gw = (const struct sockaddr *)dst;
473 
474 	/* Handle redirect case. */
475 	redest.s_addr = 0;
476 	if (V_ipsendredirects && osrc.s_addr == ip->ip_src.s_addr &&
477 	    nh->nh_ifp == m->m_pkthdr.rcvif)
478 		mcopy = ip_redir_alloc(m, nh, ip_len, &osrc, &redest);
479 
480 	/*
481 	 * Check if packet fits MTU or if hardware will fragment for us
482 	 */
483 	if (ip_len <= nh->nh_mtu) {
484 		/*
485 		 * Avoid confusing lower layers.
486 		 */
487 		m_clrprotoflags(m);
488 		/*
489 		 * Send off the packet via outgoing interface
490 		 */
491 		IP_PROBE(send, NULL, NULL, ip, nh->nh_ifp, ip, NULL);
492 		error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m, gw, &ro);
493 	} else {
494 		/*
495 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
496 		 */
497 		if (ip_off & IP_DF) {
498 			IPSTAT_INC(ips_cantfrag);
499 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
500 				0, nh->nh_mtu);
501 			goto consumed;
502 		} else {
503 			/*
504 			 * We have to fragment the packet
505 			 */
506 			m->m_pkthdr.csum_flags |= CSUM_IP;
507 			if (ip_fragment(ip, &m, nh->nh_mtu,
508 			    nh->nh_ifp->if_hwassist) != 0)
509 				goto drop;
510 			KASSERT(m != NULL, ("null mbuf and no error"));
511 			/*
512 			 * Send off the fragments via outgoing interface
513 			 */
514 			error = 0;
515 			do {
516 				m0 = m->m_nextpkt;
517 				m->m_nextpkt = NULL;
518 				/*
519 				 * Avoid confusing lower layers.
520 				 */
521 				m_clrprotoflags(m);
522 
523 				IP_PROBE(send, NULL, NULL,
524 				    mtod(m, struct ip *), nh->nh_ifp,
525 				    mtod(m, struct ip *), NULL);
526 				error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m,
527 				    gw, &ro);
528 				if (error)
529 					break;
530 			} while ((m = m0) != NULL);
531 			if (error) {
532 				/* Reclaim remaining fragments */
533 				for (m = m0; m; m = m0) {
534 					m0 = m->m_nextpkt;
535 					m_freem(m);
536 				}
537 			} else
538 				IPSTAT_INC(ips_fragmented);
539 		}
540 	}
541 
542 	if (error != 0)
543 		IPSTAT_INC(ips_odropped);
544 	else {
545 		IPSTAT_INC(ips_forward);
546 		IPSTAT_INC(ips_fastforward);
547 	}
548 
549 	/* Send required redirect */
550 	if (mcopy != NULL) {
551 		icmp_error(mcopy, ICMP_REDIRECT, ICMP_REDIRECT_HOST, redest.s_addr, 0);
552 		mcopy = NULL; /* Was consumed by callee. */
553 	}
554 
555 consumed:
556 	if (mcopy != NULL)
557 		m_freem(mcopy);
558 	return NULL;
559 drop:
560 	if (m)
561 		m_freem(m);
562 	return NULL;
563 }
564