xref: /freebsd/sys/netinet/ip_fastfwd.c (revision dafba19e42e78cd3d7c9264ece49ddd3d7d70da5)
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 "opt_ipstealth.h"
71 #include "opt_sctp.h"
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/kernel.h>
76 #include <sys/malloc.h>
77 #include <sys/mbuf.h>
78 #include <sys/protosw.h>
79 #include <sys/sdt.h>
80 #include <sys/socket.h>
81 #include <sys/sysctl.h>
82 
83 #include <net/if.h>
84 #include <net/if_types.h>
85 #include <net/if_var.h>
86 #include <net/if_dl.h>
87 #include <net/if_private.h>
88 #include <net/pfil.h>
89 #include <net/route.h>
90 #include <net/route/nhop.h>
91 #include <net/vnet.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/in_fib.h>
95 #include <netinet/in_kdtrace.h>
96 #include <netinet/in_systm.h>
97 #include <netinet/in_var.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_var.h>
100 #include <netinet/ip_icmp.h>
101 #include <netinet/ip_options.h>
102 
103 #include <machine/in_cksum.h>
104 
105 #if defined(SCTP) || defined(SCTP_SUPPORT)
106 #include <netinet/sctp_crc32.h>
107 #endif
108 
109 #define	V_ipsendredirects	VNET(ipsendredirects)
110 
111 static struct mbuf *
112 ip_redir_alloc(struct mbuf *m, struct nhop_object *nh, u_short ip_len,
113     struct in_addr *osrc, struct in_addr *newgw)
114 {
115 	struct in_ifaddr *nh_ia;
116 	struct mbuf *mcopy;
117 
118 	KASSERT(nh != NULL, ("%s: m %p nh is NULL\n", __func__, m));
119 
120 	/*
121 	 * Only send a redirect if:
122 	 * - Redirects are not disabled (must be checked by caller),
123 	 * - We have not applied NAT (must be checked by caller as possible),
124 	 * - Neither a MCAST or BCAST packet (must be checked by caller)
125 	 *   [RFC1009 Appendix A.2].
126 	 * - The packet does not do IP source routing or having any other
127 	 *   IP options (this case was handled already by ip_input() calling
128 	 *   ip_dooptions() [RFC792, p13],
129 	 * - The packet is being forwarded out the same physical interface
130 	 *   that it was received from [RFC1812, 5.2.7.2].
131 	 */
132 
133 	/*
134 	 * - The forwarding route was not created by a redirect
135 	 *   [RFC1812, 5.2.7.2], or
136 	 *   if it was to follow a default route (see below).
137 	 * - The next-hop is reachable by us [RFC1009 Appendix A.2].
138 	 */
139 	if ((nh->nh_flags & (NHF_DEFAULT | NHF_REDIRECT |
140 	    NHF_BLACKHOLE | NHF_REJECT)) != 0)
141 		return (NULL);
142 
143 	/* Get the new gateway. */
144 	if ((nh->nh_flags & NHF_GATEWAY) == 0 || nh->gw_sa.sa_family != AF_INET)
145 		return (NULL);
146 	newgw->s_addr = nh->gw4_sa.sin_addr.s_addr;
147 
148 	/*
149 	 * - The resulting forwarding destination is not "This host on this
150 	 *   network" [RFC1122, Section 3.2.1.3] (default route check above).
151 	 */
152 	if (newgw->s_addr == 0)
153 		return (NULL);
154 
155 	/*
156 	 * - We know how to reach the sender and the source address is
157 	 *   directly connected to us [RFC792, p13].
158 	 * + The new gateway address and the source address are on the same
159 	 *   subnet [RFC1009 Appendix A.2, RFC1122 3.2.2.2, RFC1812, 5.2.7.2].
160 	 * NB: if you think multiple logical subnets on the same wire should
161 	 *     receive redirects read [RFC1812, APPENDIX C (14->15)].
162 	 */
163 	nh_ia = (struct in_ifaddr *)nh->nh_ifa;
164 	if ((ntohl(osrc->s_addr) & nh_ia->ia_subnetmask) != nh_ia->ia_subnet)
165 		return (NULL);
166 
167 	/* Prepare for sending the redirect. */
168 
169 	/*
170 	 * Make a copy of as much as we need of the packet as the original
171 	 * one will be forwarded but we need (a portion) for icmp_error().
172 	 */
173 	mcopy = m_gethdr(M_NOWAIT, m->m_type);
174 	if (mcopy == NULL)
175 		return (NULL);
176 
177 	if (m_dup_pkthdr(mcopy, m, M_NOWAIT) == 0) {
178 		/*
179 		 * It's probably ok if the pkthdr dup fails (because
180 		 * the deep copy of the tag chain failed), but for now
181 		 * be conservative and just discard the copy since
182 		 * code below may some day want the tags.
183 		 */
184 		m_free(mcopy);
185 		return (NULL);
186 	}
187 	mcopy->m_len = min(ip_len, M_TRAILINGSPACE(mcopy));
188 	mcopy->m_pkthdr.len = mcopy->m_len;
189 	m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
190 
191 	return (mcopy);
192 }
193 
194 
195 static int
196 ip_findroute(struct nhop_object **pnh, struct in_addr dest, struct mbuf *m)
197 {
198 	struct nhop_object *nh;
199 
200 	nh = fib4_lookup(M_GETFIB(m), dest, 0, NHR_NONE,
201 	    m->m_pkthdr.flowid);
202 	if (nh == NULL) {
203 		IPSTAT_INC(ips_noroute);
204 		IPSTAT_INC(ips_cantforward);
205 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
206 		return (EHOSTUNREACH);
207 	}
208 	/*
209 	 * Drop blackholed traffic and directed broadcasts.
210 	 */
211 	if ((nh->nh_flags & (NHF_BLACKHOLE | NHF_BROADCAST)) != 0) {
212 		IPSTAT_INC(ips_cantforward);
213 		m_freem(m);
214 		return (EHOSTUNREACH);
215 	}
216 
217 	if (nh->nh_flags & NHF_REJECT) {
218 		IPSTAT_INC(ips_cantforward);
219 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
220 		return (EHOSTUNREACH);
221 	}
222 
223 	*pnh = nh;
224 
225 	return (0);
226 }
227 
228 /*
229  * Try to forward a packet based on the destination address.
230  * This is a fast path optimized for the plain forwarding case.
231  * If the packet is handled (and consumed) here then we return NULL;
232  * otherwise mbuf is returned and the packet should be delivered
233  * to ip_input for full processing.
234  */
235 struct mbuf *
236 ip_tryforward(struct mbuf *m)
237 {
238 	struct ip *ip;
239 	struct mbuf *m0 = NULL;
240 	struct nhop_object *nh = NULL;
241 	struct route ro;
242 	struct sockaddr_in *dst;
243 	const struct sockaddr *gw;
244 	struct in_addr dest, odest, rtdest, osrc;
245 	uint16_t ip_len, ip_off;
246 	int error = 0;
247 	struct m_tag *fwd_tag = NULL;
248 	struct mbuf *mcopy = NULL;
249 	struct in_addr redest;
250 	/*
251 	 * Are we active and forwarding packets?
252 	 */
253 
254 	M_ASSERTVALID(m);
255 	M_ASSERTPKTHDR(m);
256 
257 	/*
258 	 * Only IP packets without options
259 	 */
260 	ip = mtod(m, struct ip *);
261 
262 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
263 		if (V_ip_doopts == 1)
264 			return m;
265 		else if (V_ip_doopts == 2) {
266 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
267 				0, 0);
268 			return NULL;	/* mbuf already free'd */
269 		}
270 		/* else ignore IP options and continue */
271 	}
272 
273 	/*
274 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
275 	 * no multicast, no INADDR_ANY
276 	 *
277 	 * XXX: Probably some of these checks could be direct drop
278 	 * conditions.  However it is not clear whether there are some
279 	 * hacks or obscure behaviours which make it necessary to
280 	 * let ip_input handle it.  We play safe here and let ip_input
281 	 * deal with it until it is proven that we can directly drop it.
282 	 */
283 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
284 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
285 	    in_broadcast(ip->ip_src) ||
286 	    in_broadcast(ip->ip_dst) ||
287 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
288 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
289 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
290 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) )
291 		return m;
292 
293 	/*
294 	 * Is it for a local address on this host?
295 	 */
296 	if (in_localip(ip->ip_dst))
297 		return m;
298 
299 	IPSTAT_INC(ips_total);
300 
301 	/*
302 	 * Step 3: incoming packet firewall processing
303 	 */
304 
305 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
306 	osrc.s_addr = ip->ip_src.s_addr;
307 
308 	/*
309 	 * Run through list of ipfilter hooks for input packets
310 	 */
311 	if (!PFIL_HOOKED_IN(V_inet_pfil_head))
312 		goto passin;
313 
314 	if (pfil_mbuf_in(V_inet_pfil_head, &m, m->m_pkthdr.rcvif,
315 	    NULL) != PFIL_PASS)
316 		goto drop;
317 
318 	M_ASSERTVALID(m);
319 	M_ASSERTPKTHDR(m);
320 
321 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
322 	dest.s_addr = ip->ip_dst.s_addr;
323 
324 	/*
325 	 * Destination address changed?
326 	 */
327 	if (odest.s_addr != dest.s_addr) {
328 		/*
329 		 * Is it now for a local address on this host?
330 		 */
331 		if (in_localip(dest))
332 			goto forwardlocal;
333 		/*
334 		 * Go on with new destination address
335 		 */
336 	}
337 
338 	if (m->m_flags & M_FASTFWD_OURS) {
339 		/*
340 		 * ipfw changed it for a local address on this host.
341 		 */
342 		goto forwardlocal;
343 	}
344 
345 passin:
346 	/*
347 	 * Step 4: decrement TTL and look up route
348 	 */
349 
350 	/*
351 	 * Check TTL
352 	 */
353 #ifdef IPSTEALTH
354 	if (!V_ipstealth) {
355 #endif
356 	if (ip->ip_ttl <= IPTTLDEC) {
357 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
358 		return NULL;	/* mbuf already free'd */
359 	}
360 
361 	/*
362 	 * Decrement the TTL.
363 	 * If the IP header checksum field contains a valid value, incrementally
364 	 * change this value. Don't use hw checksum offloading, which would
365 	 * recompute the checksum. It's faster to just change it here
366 	 * according to the decremented TTL.
367 	 * If the checksum still needs to be computed, don't touch it.
368 	 */
369 	ip->ip_ttl -= IPTTLDEC;
370 	if (__predict_true((m->m_pkthdr.csum_flags & CSUM_IP) == 0)) {
371 		if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
372 			ip->ip_sum -= ~htons(IPTTLDEC << 8);
373 		else
374 			ip->ip_sum += htons(IPTTLDEC << 8);
375 	}
376 #ifdef IPSTEALTH
377 	}
378 #endif
379 
380 	/*
381 	 * Next hop forced by pfil(9) hook?
382 	 */
383 	if ((m->m_flags & M_IP_NEXTHOP) &&
384 	    ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
385 		/*
386 		 * Now we will find route to forced destination.
387 		 */
388 		dest.s_addr = ((struct sockaddr_in *)
389 			    (fwd_tag + 1))->sin_addr.s_addr;
390 		m_tag_delete(m, fwd_tag);
391 		m->m_flags &= ~M_IP_NEXTHOP;
392 	}
393 
394 	/*
395 	 * Find route to destination.
396 	 */
397 	if (ip_findroute(&nh, dest, m) != 0)
398 		return (NULL);	/* icmp unreach already sent */
399 
400 	/*
401 	 * Avoid second route lookup by caching destination.
402 	 */
403 	rtdest.s_addr = dest.s_addr;
404 
405 	/*
406 	 * Step 5: outgoing firewall packet processing
407 	 */
408 	if (!PFIL_HOOKED_OUT(V_inet_pfil_head))
409 		goto passout;
410 
411 	if (pfil_mbuf_fwd(V_inet_pfil_head, &m, nh->nh_ifp,
412 	    NULL) != PFIL_PASS)
413 		goto drop;
414 
415 	M_ASSERTVALID(m);
416 	M_ASSERTPKTHDR(m);
417 
418 	ip = mtod(m, struct ip *);
419 	dest.s_addr = ip->ip_dst.s_addr;
420 
421 	/*
422 	 * Destination address changed?
423 	 */
424 	if (m->m_flags & M_IP_NEXTHOP)
425 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
426 	else
427 		fwd_tag = NULL;
428 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
429 		/*
430 		 * Is it now for a local address on this host?
431 		 */
432 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
433 forwardlocal:
434 			/*
435 			 * Return packet for processing by ip_input().
436 			 */
437 			m->m_flags |= M_FASTFWD_OURS;
438 			return (m);
439 		}
440 		/*
441 		 * Redo route lookup with new destination address
442 		 */
443 		if (fwd_tag) {
444 			dest.s_addr = ((struct sockaddr_in *)
445 				    (fwd_tag + 1))->sin_addr.s_addr;
446 			m_tag_delete(m, fwd_tag);
447 			m->m_flags &= ~M_IP_NEXTHOP;
448 		}
449 		if (dest.s_addr != rtdest.s_addr &&
450 		    ip_findroute(&nh, dest, m) != 0)
451 			return (NULL);	/* icmp unreach already sent */
452 	}
453 
454 passout:
455 	/*
456 	 * Step 6: send off the packet
457 	 */
458 	ip_len = ntohs(ip->ip_len);
459 	ip_off = ntohs(ip->ip_off);
460 
461 	bzero(&ro, sizeof(ro));
462 	dst = (struct sockaddr_in *)&ro.ro_dst;
463 	dst->sin_family = AF_INET;
464 	dst->sin_len = sizeof(*dst);
465 	dst->sin_addr = dest;
466 	if (nh->nh_flags & NHF_GATEWAY) {
467 		gw = &nh->gw_sa;
468 		ro.ro_flags |= RT_HAS_GW;
469 	} else
470 		gw = (const struct sockaddr *)dst;
471 
472 	/*
473 	 * If the IP/SCTP/TCP/UDP header still needs a valid checksum and the
474 	 * interface will not calculate it for us, do it here.
475 	 * Note that if we defer checksum calculation, we might send an ICMP
476 	 * message later that reflects this packet, which still has an
477 	 * invalid checksum.
478 	 */
479 	if (__predict_false(m->m_pkthdr.csum_flags & CSUM_IP &
480 	    ~nh->nh_ifp->if_hwassist)) {
481 		ip->ip_sum = 0;
482 		ip->ip_sum = in_cksum(m, (ip->ip_hl << 2));
483 		m->m_pkthdr.csum_flags &= ~CSUM_IP;
484 	}
485 	if (__predict_false(m->m_pkthdr.csum_flags & CSUM_DELAY_DATA &
486 	    ~nh->nh_ifp->if_hwassist)) {
487 		in_delayed_cksum(m);
488 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
489 	}
490 #if defined(SCTP) || defined(SCTP_SUPPORT)
491 	if (__predict_false(m->m_pkthdr.csum_flags & CSUM_IP_SCTP &
492 	    ~nh->nh_ifp->if_hwassist)) {
493 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
494 		m->m_pkthdr.csum_flags &= ~CSUM_IP_SCTP;
495 	}
496 #endif
497 
498 	/* Handle redirect case. */
499 	redest.s_addr = 0;
500 	if (V_ipsendredirects && osrc.s_addr == ip->ip_src.s_addr &&
501 	    nh->nh_ifp == m->m_pkthdr.rcvif)
502 		mcopy = ip_redir_alloc(m, nh, ip_len, &osrc, &redest);
503 
504 	/*
505 	 * Check if packet fits MTU or if hardware will fragment for us
506 	 */
507 	if (ip_len <= nh->nh_mtu) {
508 		/*
509 		 * Avoid confusing lower layers.
510 		 */
511 		m_clrprotoflags(m);
512 		/*
513 		 * Send off the packet via outgoing interface
514 		 */
515 		IP_PROBE(send, NULL, NULL, ip, nh->nh_ifp, ip, NULL);
516 		error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m, gw, &ro);
517 	} else {
518 		/*
519 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
520 		 */
521 		if (ip_off & IP_DF) {
522 			IPSTAT_INC(ips_cantfrag);
523 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
524 				0, nh->nh_mtu);
525 			goto consumed;
526 		} else {
527 			/*
528 			 * We have to fragment the packet
529 			 */
530 			m->m_pkthdr.csum_flags |= CSUM_IP;
531 			if (ip_fragment(ip, &m, nh->nh_mtu,
532 			    nh->nh_ifp->if_hwassist) != 0)
533 				goto drop;
534 			KASSERT(m != NULL, ("null mbuf and no error"));
535 			/*
536 			 * Send off the fragments via outgoing interface
537 			 */
538 			error = 0;
539 			do {
540 				m0 = m->m_nextpkt;
541 				m->m_nextpkt = NULL;
542 				/*
543 				 * Avoid confusing lower layers.
544 				 */
545 				m_clrprotoflags(m);
546 
547 				IP_PROBE(send, NULL, NULL,
548 				    mtod(m, struct ip *), nh->nh_ifp,
549 				    mtod(m, struct ip *), NULL);
550 				error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m,
551 				    gw, &ro);
552 				if (error)
553 					break;
554 			} while ((m = m0) != NULL);
555 			if (error) {
556 				/* Reclaim remaining fragments */
557 				for (m = m0; m; m = m0) {
558 					m0 = m->m_nextpkt;
559 					m_freem(m);
560 				}
561 			} else
562 				IPSTAT_INC(ips_fragmented);
563 		}
564 	}
565 
566 	if (error != 0)
567 		IPSTAT_INC(ips_odropped);
568 	else {
569 		IPSTAT_INC(ips_forward);
570 		IPSTAT_INC(ips_fastforward);
571 	}
572 
573 	/* Send required redirect */
574 	if (mcopy != NULL) {
575 		icmp_error(mcopy, ICMP_REDIRECT, ICMP_REDIRECT_HOST, redest.s_addr, 0);
576 		mcopy = NULL; /* Was consumed by callee. */
577 	}
578 
579 consumed:
580 	if (mcopy != NULL)
581 		m_freem(mcopy);
582 	return NULL;
583 drop:
584 	if (m)
585 		m_freem(m);
586 	return NULL;
587 }
588