xref: /freebsd-13-stable/sys/netinet/ip_fastfwd.c (revision 3bc80996974a61a4223eae4c1ccd47b6ee32a48a)
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 #include "opt_ipstealth.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/pfil.h>
88 #include <net/route.h>
89 #include <net/route/nhop.h>
90 #include <net/vnet.h>
91 
92 #include <netinet/in.h>
93 #include <netinet/in_fib.h>
94 #include <netinet/in_kdtrace.h>
95 #include <netinet/in_systm.h>
96 #include <netinet/in_var.h>
97 #include <netinet/ip.h>
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_icmp.h>
100 #include <netinet/ip_options.h>
101 
102 #include <machine/in_cksum.h>
103 
104 #define	V_ipsendredirects	VNET(ipsendredirects)
105 
106 static struct mbuf *
ip_redir_alloc(struct mbuf * m,struct nhop_object * nh,u_short ip_len,struct in_addr * osrc,struct in_addr * newgw)107 ip_redir_alloc(struct mbuf *m, struct nhop_object *nh, u_short ip_len,
108     struct in_addr *osrc, struct in_addr *newgw)
109 {
110 	struct in_ifaddr *nh_ia;
111 	struct mbuf *mcopy;
112 
113 	KASSERT(nh != NULL, ("%s: m %p nh is NULL\n", __func__, m));
114 
115 	/*
116 	 * Only send a redirect if:
117 	 * - Redirects are not disabled (must be checked by caller),
118 	 * - We have not applied NAT (must be checked by caller as possible),
119 	 * - Neither a MCAST or BCAST packet (must be checked by caller)
120 	 *   [RFC1009 Appendix A.2].
121 	 * - The packet does not do IP source routing or having any other
122 	 *   IP options (this case was handled already by ip_input() calling
123 	 *   ip_dooptions() [RFC792, p13],
124 	 * - The packet is being forwarded out the same physical interface
125 	 *   that it was received from [RFC1812, 5.2.7.2].
126 	 */
127 
128 	/*
129 	 * - The forwarding route was not created by a redirect
130 	 *   [RFC1812, 5.2.7.2], or
131 	 *   if it was to follow a default route (see below).
132 	 * - The next-hop is reachable by us [RFC1009 Appendix A.2].
133 	 */
134 	if ((nh->nh_flags & (NHF_DEFAULT | NHF_REDIRECT |
135 	    NHF_BLACKHOLE | NHF_REJECT)) != 0)
136 		return (NULL);
137 
138 	/* Get the new gateway. */
139 	if ((nh->nh_flags & NHF_GATEWAY) == 0 || nh->gw_sa.sa_family != AF_INET)
140 		return (NULL);
141 	newgw->s_addr = nh->gw4_sa.sin_addr.s_addr;
142 
143 	/*
144 	 * - The resulting forwarding destination is not "This host on this
145 	 *   network" [RFC1122, Section 3.2.1.3] (default route check above).
146 	 */
147 	if (newgw->s_addr == 0)
148 		return (NULL);
149 
150 	/*
151 	 * - We know how to reach the sender and the source address is
152 	 *   directly connected to us [RFC792, p13].
153 	 * + The new gateway address and the source address are on the same
154 	 *   subnet [RFC1009 Appendix A.2, RFC1122 3.2.2.2, RFC1812, 5.2.7.2].
155 	 * NB: if you think multiple logical subnets on the same wire should
156 	 *     receive redirects read [RFC1812, APPENDIX C (14->15)].
157 	 */
158 	nh_ia = (struct in_ifaddr *)nh->nh_ifa;
159 	if ((ntohl(osrc->s_addr) & nh_ia->ia_subnetmask) != nh_ia->ia_subnet)
160 		return (NULL);
161 
162 	/* Prepare for sending the redirect. */
163 
164 	/*
165 	 * Make a copy of as much as we need of the packet as the original
166 	 * one will be forwarded but we need (a portion) for icmp_error().
167 	 */
168 	mcopy = m_gethdr(M_NOWAIT, m->m_type);
169 	if (mcopy == NULL)
170 		return (NULL);
171 
172 	if (m_dup_pkthdr(mcopy, m, M_NOWAIT) == 0) {
173 		/*
174 		 * It's probably ok if the pkthdr dup fails (because
175 		 * the deep copy of the tag chain failed), but for now
176 		 * be conservative and just discard the copy since
177 		 * code below may some day want the tags.
178 		 */
179 		m_free(mcopy);
180 		return (NULL);
181 	}
182 	mcopy->m_len = min(ip_len, M_TRAILINGSPACE(mcopy));
183 	mcopy->m_pkthdr.len = mcopy->m_len;
184 	m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
185 
186 	return (mcopy);
187 }
188 
189 
190 static int
ip_findroute(struct nhop_object ** pnh,struct in_addr dest,struct mbuf * m)191 ip_findroute(struct nhop_object **pnh, struct in_addr dest, struct mbuf *m)
192 {
193 	struct nhop_object *nh;
194 
195 	nh = fib4_lookup(M_GETFIB(m), dest, 0, NHR_NONE,
196 	    m->m_pkthdr.flowid);
197 	if (nh == NULL) {
198 		IPSTAT_INC(ips_noroute);
199 		IPSTAT_INC(ips_cantforward);
200 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
201 		return (EHOSTUNREACH);
202 	}
203 	/*
204 	 * Drop blackholed traffic and directed broadcasts.
205 	 */
206 	if ((nh->nh_flags & (NHF_BLACKHOLE | NHF_BROADCAST)) != 0) {
207 		IPSTAT_INC(ips_cantforward);
208 		m_freem(m);
209 		return (EHOSTUNREACH);
210 	}
211 
212 	if (nh->nh_flags & NHF_REJECT) {
213 		IPSTAT_INC(ips_cantforward);
214 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
215 		return (EHOSTUNREACH);
216 	}
217 
218 	*pnh = nh;
219 
220 	return (0);
221 }
222 
223 /*
224  * Try to forward a packet based on the destination address.
225  * This is a fast path optimized for the plain forwarding case.
226  * If the packet is handled (and consumed) here then we return NULL;
227  * otherwise mbuf is returned and the packet should be delivered
228  * to ip_input for full processing.
229  */
230 struct mbuf *
ip_tryforward(struct mbuf * m)231 ip_tryforward(struct mbuf *m)
232 {
233 	struct ip *ip;
234 	struct mbuf *m0 = NULL;
235 	struct nhop_object *nh = NULL;
236 	struct route ro;
237 	struct sockaddr_in *dst;
238 	const struct sockaddr *gw;
239 	struct in_addr dest, odest, rtdest, osrc;
240 	uint16_t ip_len, ip_off;
241 	int error = 0;
242 	struct m_tag *fwd_tag = NULL;
243 	struct mbuf *mcopy = NULL;
244 	struct in_addr redest;
245 	/*
246 	 * Are we active and forwarding packets?
247 	 */
248 
249 	M_ASSERTVALID(m);
250 	M_ASSERTPKTHDR(m);
251 
252 #ifdef ALTQ
253 	/*
254 	 * Is packet dropped by traffic conditioner?
255 	 */
256 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
257 		goto drop;
258 #endif
259 
260 	/*
261 	 * Only IP packets without options
262 	 */
263 	ip = mtod(m, struct ip *);
264 
265 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
266 		if (V_ip_doopts == 1)
267 			return m;
268 		else if (V_ip_doopts == 2) {
269 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
270 				0, 0);
271 			return NULL;	/* mbuf already free'd */
272 		}
273 		/* else ignore IP options and continue */
274 	}
275 
276 	/*
277 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
278 	 * no multicast, no INADDR_ANY
279 	 *
280 	 * XXX: Probably some of these checks could be direct drop
281 	 * conditions.  However it is not clear whether there are some
282 	 * hacks or obscure behaviours which make it necessary to
283 	 * let ip_input handle it.  We play safe here and let ip_input
284 	 * deal with it until it is proven that we can directly drop it.
285 	 */
286 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
287 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
288 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
289 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
290 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
291 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
292 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
293 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
294 	    ip->ip_src.s_addr == INADDR_ANY ||
295 	    ip->ip_dst.s_addr == INADDR_ANY )
296 		return m;
297 
298 	/*
299 	 * Is it for a local address on this host?
300 	 */
301 	if (in_localip(ip->ip_dst))
302 		return m;
303 
304 	IPSTAT_INC(ips_total);
305 
306 	/*
307 	 * Step 3: incoming packet firewall processing
308 	 */
309 
310 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
311 	osrc.s_addr = ip->ip_src.s_addr;
312 
313 	/*
314 	 * Run through list of ipfilter hooks for input packets
315 	 */
316 	if (!PFIL_HOOKED_IN(V_inet_pfil_head))
317 		goto passin;
318 
319 	if (pfil_run_hooks(V_inet_pfil_head, &m, m->m_pkthdr.rcvif, PFIL_IN,
320 	    NULL) != PFIL_PASS)
321 		goto drop;
322 
323 	M_ASSERTVALID(m);
324 	M_ASSERTPKTHDR(m);
325 
326 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
327 	dest.s_addr = ip->ip_dst.s_addr;
328 
329 	/*
330 	 * Destination address changed?
331 	 */
332 	if (odest.s_addr != dest.s_addr) {
333 		/*
334 		 * Is it now for a local address on this host?
335 		 */
336 		if (in_localip(dest))
337 			goto forwardlocal;
338 		/*
339 		 * Go on with new destination address
340 		 */
341 	}
342 
343 	if (m->m_flags & M_FASTFWD_OURS) {
344 		/*
345 		 * ipfw changed it for a local address on this host.
346 		 */
347 		goto forwardlocal;
348 	}
349 
350 passin:
351 	/*
352 	 * Step 4: decrement TTL and look up route
353 	 */
354 
355 	/*
356 	 * Check TTL
357 	 */
358 #ifdef IPSTEALTH
359 	if (!V_ipstealth) {
360 #endif
361 	if (ip->ip_ttl <= IPTTLDEC) {
362 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
363 		return NULL;	/* mbuf already free'd */
364 	}
365 
366 	/*
367 	 * Decrement the TTL and incrementally change the IP header checksum.
368 	 * Don't bother doing this with hw checksum offloading, it's faster
369 	 * doing it right here.
370 	 */
371 	ip->ip_ttl -= IPTTLDEC;
372 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
373 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
374 	else
375 		ip->ip_sum += htons(IPTTLDEC << 8);
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_run_hooks(V_inet_pfil_head, &m, nh->nh_ifp,
412 	    PFIL_OUT | PFIL_FWD, 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 	/* Handle redirect case. */
473 	redest.s_addr = 0;
474 	if (V_ipsendredirects && osrc.s_addr == ip->ip_src.s_addr &&
475 	    nh->nh_ifp == m->m_pkthdr.rcvif)
476 		mcopy = ip_redir_alloc(m, nh, ip_len, &osrc, &redest);
477 
478 	/*
479 	 * Check if packet fits MTU or if hardware will fragment for us
480 	 */
481 	if (ip_len <= nh->nh_mtu) {
482 		/*
483 		 * Avoid confusing lower layers.
484 		 */
485 		m_clrprotoflags(m);
486 		/*
487 		 * Send off the packet via outgoing interface
488 		 */
489 		IP_PROBE(send, NULL, NULL, ip, nh->nh_ifp, ip, NULL);
490 		error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m, gw, &ro);
491 	} else {
492 		/*
493 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
494 		 */
495 		if (ip_off & IP_DF) {
496 			IPSTAT_INC(ips_cantfrag);
497 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
498 				0, nh->nh_mtu);
499 			goto consumed;
500 		} else {
501 			/*
502 			 * We have to fragment the packet
503 			 */
504 			m->m_pkthdr.csum_flags |= CSUM_IP;
505 			if (ip_fragment(ip, &m, nh->nh_mtu,
506 			    nh->nh_ifp->if_hwassist) != 0)
507 				goto drop;
508 			KASSERT(m != NULL, ("null mbuf and no error"));
509 			/*
510 			 * Send off the fragments via outgoing interface
511 			 */
512 			error = 0;
513 			do {
514 				m0 = m->m_nextpkt;
515 				m->m_nextpkt = NULL;
516 				/*
517 				 * Avoid confusing lower layers.
518 				 */
519 				m_clrprotoflags(m);
520 
521 				IP_PROBE(send, NULL, NULL,
522 				    mtod(m, struct ip *), nh->nh_ifp,
523 				    mtod(m, struct ip *), NULL);
524 				error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m,
525 				    gw, &ro);
526 				if (error)
527 					break;
528 			} while ((m = m0) != NULL);
529 			if (error) {
530 				/* Reclaim remaining fragments */
531 				for (m = m0; m; m = m0) {
532 					m0 = m->m_nextpkt;
533 					m_freem(m);
534 				}
535 			} else
536 				IPSTAT_INC(ips_fragmented);
537 		}
538 	}
539 
540 	if (error != 0)
541 		IPSTAT_INC(ips_odropped);
542 	else {
543 		IPSTAT_INC(ips_forward);
544 		IPSTAT_INC(ips_fastforward);
545 	}
546 
547 	/* Send required redirect */
548 	if (mcopy != NULL) {
549 		icmp_error(mcopy, ICMP_REDIRECT, ICMP_REDIRECT_HOST, redest.s_addr, 0);
550 		mcopy = NULL; /* Was consumed by callee. */
551 	}
552 
553 consumed:
554 	if (mcopy != NULL)
555 		m_freem(mcopy);
556 	return NULL;
557 drop:
558 	if (m)
559 		m_freem(m);
560 	return NULL;
561 }
562