xref: /freebsd/sys/netinet/ip_fastfwd.c (revision ec965063070e5753c166cf592c9336444b74720a)
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 	/*
255 	 * Only IP packets without options
256 	 */
257 	ip = mtod(m, struct ip *);
258 
259 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
260 		if (V_ip_doopts == 1)
261 			return m;
262 		else if (V_ip_doopts == 2) {
263 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
264 				0, 0);
265 			return NULL;	/* mbuf already free'd */
266 		}
267 		/* else ignore IP options and continue */
268 	}
269 
270 	/*
271 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
272 	 * no multicast, no INADDR_ANY
273 	 *
274 	 * XXX: Probably some of these checks could be direct drop
275 	 * conditions.  However it is not clear whether there are some
276 	 * hacks or obscure behaviours which make it necessary to
277 	 * let ip_input handle it.  We play safe here and let ip_input
278 	 * deal with it until it is proven that we can directly drop it.
279 	 */
280 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
281 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
282 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
283 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
284 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
285 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
286 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
287 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
288 	    ip->ip_src.s_addr == INADDR_ANY ||
289 	    ip->ip_dst.s_addr == INADDR_ANY )
290 		return m;
291 
292 	/*
293 	 * Is it for a local address on this host?
294 	 */
295 	if (in_localip(ip->ip_dst))
296 		return m;
297 
298 	IPSTAT_INC(ips_total);
299 
300 	/*
301 	 * Step 3: incoming packet firewall processing
302 	 */
303 
304 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
305 	osrc.s_addr = ip->ip_src.s_addr;
306 
307 	/*
308 	 * Run through list of ipfilter hooks for input packets
309 	 */
310 	if (!PFIL_HOOKED_IN(V_inet_pfil_head))
311 		goto passin;
312 
313 	if (pfil_mbuf_in(V_inet_pfil_head, &m, m->m_pkthdr.rcvif,
314 	    NULL) != PFIL_PASS)
315 		goto drop;
316 
317 	M_ASSERTVALID(m);
318 	M_ASSERTPKTHDR(m);
319 
320 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
321 	dest.s_addr = ip->ip_dst.s_addr;
322 
323 	/*
324 	 * Destination address changed?
325 	 */
326 	if (odest.s_addr != dest.s_addr) {
327 		/*
328 		 * Is it now for a local address on this host?
329 		 */
330 		if (in_localip(dest))
331 			goto forwardlocal;
332 		/*
333 		 * Go on with new destination address
334 		 */
335 	}
336 
337 	if (m->m_flags & M_FASTFWD_OURS) {
338 		/*
339 		 * ipfw changed it for a local address on this host.
340 		 */
341 		goto forwardlocal;
342 	}
343 
344 passin:
345 	/*
346 	 * Step 4: decrement TTL and look up route
347 	 */
348 
349 	/*
350 	 * Check TTL
351 	 */
352 #ifdef IPSTEALTH
353 	if (!V_ipstealth) {
354 #endif
355 	if (ip->ip_ttl <= IPTTLDEC) {
356 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
357 		return NULL;	/* mbuf already free'd */
358 	}
359 
360 	/*
361 	 * Decrement the TTL and incrementally change the IP header checksum.
362 	 * Don't bother doing this with hw checksum offloading, it's faster
363 	 * doing it right here.
364 	 */
365 	ip->ip_ttl -= IPTTLDEC;
366 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
367 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
368 	else
369 		ip->ip_sum += htons(IPTTLDEC << 8);
370 #ifdef IPSTEALTH
371 	}
372 #endif
373 
374 	/*
375 	 * Next hop forced by pfil(9) hook?
376 	 */
377 	if ((m->m_flags & M_IP_NEXTHOP) &&
378 	    ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
379 		/*
380 		 * Now we will find route to forced destination.
381 		 */
382 		dest.s_addr = ((struct sockaddr_in *)
383 			    (fwd_tag + 1))->sin_addr.s_addr;
384 		m_tag_delete(m, fwd_tag);
385 		m->m_flags &= ~M_IP_NEXTHOP;
386 	}
387 
388 	/*
389 	 * Find route to destination.
390 	 */
391 	if (ip_findroute(&nh, dest, m) != 0)
392 		return (NULL);	/* icmp unreach already sent */
393 
394 	/*
395 	 * Avoid second route lookup by caching destination.
396 	 */
397 	rtdest.s_addr = dest.s_addr;
398 
399 	/*
400 	 * Step 5: outgoing firewall packet processing
401 	 */
402 	if (!PFIL_HOOKED_OUT(V_inet_pfil_head))
403 		goto passout;
404 
405 	if (pfil_mbuf_out(V_inet_pfil_head, &m, nh->nh_ifp,
406 	    NULL) != PFIL_PASS)
407 		goto drop;
408 
409 	M_ASSERTVALID(m);
410 	M_ASSERTPKTHDR(m);
411 
412 	ip = mtod(m, struct ip *);
413 	dest.s_addr = ip->ip_dst.s_addr;
414 
415 	/*
416 	 * Destination address changed?
417 	 */
418 	if (m->m_flags & M_IP_NEXTHOP)
419 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
420 	else
421 		fwd_tag = NULL;
422 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
423 		/*
424 		 * Is it now for a local address on this host?
425 		 */
426 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
427 forwardlocal:
428 			/*
429 			 * Return packet for processing by ip_input().
430 			 */
431 			m->m_flags |= M_FASTFWD_OURS;
432 			return (m);
433 		}
434 		/*
435 		 * Redo route lookup with new destination address
436 		 */
437 		if (fwd_tag) {
438 			dest.s_addr = ((struct sockaddr_in *)
439 				    (fwd_tag + 1))->sin_addr.s_addr;
440 			m_tag_delete(m, fwd_tag);
441 			m->m_flags &= ~M_IP_NEXTHOP;
442 		}
443 		if (dest.s_addr != rtdest.s_addr &&
444 		    ip_findroute(&nh, dest, m) != 0)
445 			return (NULL);	/* icmp unreach already sent */
446 	}
447 
448 passout:
449 	/*
450 	 * Step 6: send off the packet
451 	 */
452 	ip_len = ntohs(ip->ip_len);
453 	ip_off = ntohs(ip->ip_off);
454 
455 	bzero(&ro, sizeof(ro));
456 	dst = (struct sockaddr_in *)&ro.ro_dst;
457 	dst->sin_family = AF_INET;
458 	dst->sin_len = sizeof(*dst);
459 	dst->sin_addr = dest;
460 	if (nh->nh_flags & NHF_GATEWAY) {
461 		gw = &nh->gw_sa;
462 		ro.ro_flags |= RT_HAS_GW;
463 	} else
464 		gw = (const struct sockaddr *)dst;
465 
466 	/* Handle redirect case. */
467 	redest.s_addr = 0;
468 	if (V_ipsendredirects && osrc.s_addr == ip->ip_src.s_addr &&
469 	    nh->nh_ifp == m->m_pkthdr.rcvif)
470 		mcopy = ip_redir_alloc(m, nh, ip_len, &osrc, &redest);
471 
472 	/*
473 	 * Check if packet fits MTU or if hardware will fragment for us
474 	 */
475 	if (ip_len <= nh->nh_mtu) {
476 		/*
477 		 * Avoid confusing lower layers.
478 		 */
479 		m_clrprotoflags(m);
480 		/*
481 		 * Send off the packet via outgoing interface
482 		 */
483 		IP_PROBE(send, NULL, NULL, ip, nh->nh_ifp, ip, NULL);
484 		error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m, gw, &ro);
485 	} else {
486 		/*
487 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
488 		 */
489 		if (ip_off & IP_DF) {
490 			IPSTAT_INC(ips_cantfrag);
491 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
492 				0, nh->nh_mtu);
493 			goto consumed;
494 		} else {
495 			/*
496 			 * We have to fragment the packet
497 			 */
498 			m->m_pkthdr.csum_flags |= CSUM_IP;
499 			if (ip_fragment(ip, &m, nh->nh_mtu,
500 			    nh->nh_ifp->if_hwassist) != 0)
501 				goto drop;
502 			KASSERT(m != NULL, ("null mbuf and no error"));
503 			/*
504 			 * Send off the fragments via outgoing interface
505 			 */
506 			error = 0;
507 			do {
508 				m0 = m->m_nextpkt;
509 				m->m_nextpkt = NULL;
510 				/*
511 				 * Avoid confusing lower layers.
512 				 */
513 				m_clrprotoflags(m);
514 
515 				IP_PROBE(send, NULL, NULL,
516 				    mtod(m, struct ip *), nh->nh_ifp,
517 				    mtod(m, struct ip *), NULL);
518 				error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m,
519 				    gw, &ro);
520 				if (error)
521 					break;
522 			} while ((m = m0) != NULL);
523 			if (error) {
524 				/* Reclaim remaining fragments */
525 				for (m = m0; m; m = m0) {
526 					m0 = m->m_nextpkt;
527 					m_freem(m);
528 				}
529 			} else
530 				IPSTAT_INC(ips_fragmented);
531 		}
532 	}
533 
534 	if (error != 0)
535 		IPSTAT_INC(ips_odropped);
536 	else {
537 		IPSTAT_INC(ips_forward);
538 		IPSTAT_INC(ips_fastforward);
539 	}
540 
541 	/* Send required redirect */
542 	if (mcopy != NULL) {
543 		icmp_error(mcopy, ICMP_REDIRECT, ICMP_REDIRECT_HOST, redest.s_addr, 0);
544 		mcopy = NULL; /* Was consumed by callee. */
545 	}
546 
547 consumed:
548 	if (mcopy != NULL)
549 		m_freem(mcopy);
550 	return NULL;
551 drop:
552 	if (m)
553 		m_freem(m);
554 	return NULL;
555 }
556