xref: /freebsd/sys/netinet/ip_fastfwd.c (revision fe6060f10f634930ff71b7c50291ddc610da2475)
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  * We don't do ICMP redirect in the fast forwarding path. I have had my own
65  * cases where two core routers with Zebra routing suite would send millions
66  * ICMP redirects to connected hosts if the destination router was not the
67  * default gateway. In one case it was filling the routing table of a host
68  * with approximately 300.000 cloned redirect entries until it ran out of
69  * kernel memory. However the networking code proved very robust and it didn't
70  * crash or fail in other ways.
71  */
72 
73 /*
74  * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
75  * is being followed here.
76  */
77 
78 #include <sys/cdefs.h>
79 __FBSDID("$FreeBSD$");
80 
81 #include "opt_ipstealth.h"
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/kernel.h>
86 #include <sys/malloc.h>
87 #include <sys/mbuf.h>
88 #include <sys/protosw.h>
89 #include <sys/sdt.h>
90 #include <sys/socket.h>
91 #include <sys/sysctl.h>
92 
93 #include <net/if.h>
94 #include <net/if_types.h>
95 #include <net/if_var.h>
96 #include <net/if_dl.h>
97 #include <net/pfil.h>
98 #include <net/route.h>
99 #include <net/route/nhop.h>
100 #include <net/vnet.h>
101 
102 #include <netinet/in.h>
103 #include <netinet/in_fib.h>
104 #include <netinet/in_kdtrace.h>
105 #include <netinet/in_systm.h>
106 #include <netinet/in_var.h>
107 #include <netinet/ip.h>
108 #include <netinet/ip_var.h>
109 #include <netinet/ip_icmp.h>
110 #include <netinet/ip_options.h>
111 
112 #include <machine/in_cksum.h>
113 
114 #define	V_ipsendredirects	VNET(ipsendredirects)
115 
116 static struct mbuf *
117 ip_redir_alloc(struct mbuf *m, struct nhop_object *nh,
118     struct ip *ip, in_addr_t *addr)
119 {
120 	struct mbuf *mcopy = m_gethdr(M_NOWAIT, m->m_type);
121 
122 	if (mcopy == NULL)
123 		return (NULL);
124 
125 	if (m_dup_pkthdr(mcopy, m, M_NOWAIT) == 0) {
126 		/*
127 		 * It's probably ok if the pkthdr dup fails (because
128 		 * the deep copy of the tag chain failed), but for now
129 		 * be conservative and just discard the copy since
130 		 * code below may some day want the tags.
131 		 */
132 		m_free(mcopy);
133 		return (NULL);
134 	}
135 	mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy));
136 	mcopy->m_pkthdr.len = mcopy->m_len;
137 	m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
138 
139 	if (nh != NULL &&
140 	    ((nh->nh_flags & (NHF_REDIRECT|NHF_DEFAULT)) == 0)) {
141 		struct in_ifaddr *nh_ia = (struct in_ifaddr *)(nh->nh_ifa);
142 		u_long src = ntohl(ip->ip_src.s_addr);
143 
144 		if (nh_ia != NULL &&
145 		    (src & nh_ia->ia_subnetmask) == nh_ia->ia_subnet) {
146 			if (nh->nh_flags & NHF_GATEWAY)
147 				*addr = nh->gw4_sa.sin_addr.s_addr;
148 			else
149 				*addr = ip->ip_dst.s_addr;
150 		}
151 	}
152 	return (mcopy);
153 }
154 
155 
156 static int
157 ip_findroute(struct nhop_object **pnh, struct in_addr dest, struct mbuf *m)
158 {
159 	struct nhop_object *nh;
160 
161 	nh = fib4_lookup(M_GETFIB(m), dest, 0, NHR_NONE,
162 	    m->m_pkthdr.flowid);
163 	if (nh == NULL) {
164 		IPSTAT_INC(ips_noroute);
165 		IPSTAT_INC(ips_cantforward);
166 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
167 		return (EHOSTUNREACH);
168 	}
169 	/*
170 	 * Drop blackholed traffic and directed broadcasts.
171 	 */
172 	if ((nh->nh_flags & (NHF_BLACKHOLE | NHF_BROADCAST)) != 0) {
173 		IPSTAT_INC(ips_cantforward);
174 		m_freem(m);
175 		return (EHOSTUNREACH);
176 	}
177 
178 	if (nh->nh_flags & NHF_REJECT) {
179 		IPSTAT_INC(ips_cantforward);
180 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
181 		return (EHOSTUNREACH);
182 	}
183 
184 	*pnh = nh;
185 
186 	return (0);
187 }
188 
189 /*
190  * Try to forward a packet based on the destination address.
191  * This is a fast path optimized for the plain forwarding case.
192  * If the packet is handled (and consumed) here then we return NULL;
193  * otherwise mbuf is returned and the packet should be delivered
194  * to ip_input for full processing.
195  */
196 struct mbuf *
197 ip_tryforward(struct mbuf *m)
198 {
199 	struct ip *ip;
200 	struct mbuf *m0 = NULL;
201 	struct nhop_object *nh = NULL;
202 	struct route ro;
203 	struct sockaddr_in *dst;
204 	const struct sockaddr *gw;
205 	struct in_addr dest, odest, rtdest;
206 	uint16_t ip_len, ip_off;
207 	int error = 0;
208 	struct m_tag *fwd_tag = NULL;
209 	struct mbuf *mcopy = NULL;
210 	struct in_addr redest;
211 	/*
212 	 * Are we active and forwarding packets?
213 	 */
214 
215 	M_ASSERTVALID(m);
216 	M_ASSERTPKTHDR(m);
217 
218 #ifdef ALTQ
219 	/*
220 	 * Is packet dropped by traffic conditioner?
221 	 */
222 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
223 		goto drop;
224 #endif
225 
226 	/*
227 	 * Only IP packets without options
228 	 */
229 	ip = mtod(m, struct ip *);
230 
231 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
232 		if (V_ip_doopts == 1)
233 			return m;
234 		else if (V_ip_doopts == 2) {
235 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
236 				0, 0);
237 			return NULL;	/* mbuf already free'd */
238 		}
239 		/* else ignore IP options and continue */
240 	}
241 
242 	/*
243 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
244 	 * no multicast, no INADDR_ANY
245 	 *
246 	 * XXX: Probably some of these checks could be direct drop
247 	 * conditions.  However it is not clear whether there are some
248 	 * hacks or obscure behaviours which make it necessary to
249 	 * let ip_input handle it.  We play safe here and let ip_input
250 	 * deal with it until it is proven that we can directly drop it.
251 	 */
252 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
253 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
254 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
255 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
256 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
257 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
258 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
259 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
260 	    ip->ip_src.s_addr == INADDR_ANY ||
261 	    ip->ip_dst.s_addr == INADDR_ANY )
262 		return m;
263 
264 	/*
265 	 * Is it for a local address on this host?
266 	 */
267 	if (in_localip(ip->ip_dst))
268 		return m;
269 
270 	IPSTAT_INC(ips_total);
271 
272 	/*
273 	 * Step 3: incoming packet firewall processing
274 	 */
275 
276 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
277 
278 	/*
279 	 * Run through list of ipfilter hooks for input packets
280 	 */
281 	if (!PFIL_HOOKED_IN(V_inet_pfil_head))
282 		goto passin;
283 
284 	if (pfil_run_hooks(V_inet_pfil_head, &m, m->m_pkthdr.rcvif, PFIL_IN,
285 	    NULL) != PFIL_PASS)
286 		goto drop;
287 
288 	M_ASSERTVALID(m);
289 	M_ASSERTPKTHDR(m);
290 
291 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
292 	dest.s_addr = ip->ip_dst.s_addr;
293 
294 	/*
295 	 * Destination address changed?
296 	 */
297 	if (odest.s_addr != dest.s_addr) {
298 		/*
299 		 * Is it now for a local address on this host?
300 		 */
301 		if (in_localip(dest))
302 			goto forwardlocal;
303 		/*
304 		 * Go on with new destination address
305 		 */
306 	}
307 
308 	if (m->m_flags & M_FASTFWD_OURS) {
309 		/*
310 		 * ipfw changed it for a local address on this host.
311 		 */
312 		goto forwardlocal;
313 	}
314 
315 passin:
316 	/*
317 	 * Step 4: decrement TTL and look up route
318 	 */
319 
320 	/*
321 	 * Check TTL
322 	 */
323 #ifdef IPSTEALTH
324 	if (!V_ipstealth) {
325 #endif
326 	if (ip->ip_ttl <= IPTTLDEC) {
327 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
328 		return NULL;	/* mbuf already free'd */
329 	}
330 
331 	/*
332 	 * Decrement the TTL and incrementally change the IP header checksum.
333 	 * Don't bother doing this with hw checksum offloading, it's faster
334 	 * doing it right here.
335 	 */
336 	ip->ip_ttl -= IPTTLDEC;
337 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
338 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
339 	else
340 		ip->ip_sum += htons(IPTTLDEC << 8);
341 #ifdef IPSTEALTH
342 	}
343 #endif
344 
345 	/*
346 	 * Next hop forced by pfil(9) hook?
347 	 */
348 	if ((m->m_flags & M_IP_NEXTHOP) &&
349 	    ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
350 		/*
351 		 * Now we will find route to forced destination.
352 		 */
353 		dest.s_addr = ((struct sockaddr_in *)
354 			    (fwd_tag + 1))->sin_addr.s_addr;
355 		m_tag_delete(m, fwd_tag);
356 		m->m_flags &= ~M_IP_NEXTHOP;
357 	}
358 
359 	/*
360 	 * Find route to destination.
361 	 */
362 	if (ip_findroute(&nh, dest, m) != 0)
363 		return (NULL);	/* icmp unreach already sent */
364 
365 	/*
366 	 * Avoid second route lookup by caching destination.
367 	 */
368 	rtdest.s_addr = dest.s_addr;
369 
370 	/*
371 	 * Step 5: outgoing firewall packet processing
372 	 */
373 	if (!PFIL_HOOKED_OUT(V_inet_pfil_head))
374 		goto passout;
375 
376 	if (pfil_run_hooks(V_inet_pfil_head, &m, nh->nh_ifp,
377 	    PFIL_OUT | PFIL_FWD, NULL) != PFIL_PASS)
378 		goto drop;
379 
380 	M_ASSERTVALID(m);
381 	M_ASSERTPKTHDR(m);
382 
383 	ip = mtod(m, struct ip *);
384 	dest.s_addr = ip->ip_dst.s_addr;
385 
386 	/*
387 	 * Destination address changed?
388 	 */
389 	if (m->m_flags & M_IP_NEXTHOP)
390 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
391 	else
392 		fwd_tag = NULL;
393 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
394 		/*
395 		 * Is it now for a local address on this host?
396 		 */
397 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
398 forwardlocal:
399 			/*
400 			 * Return packet for processing by ip_input().
401 			 */
402 			m->m_flags |= M_FASTFWD_OURS;
403 			return (m);
404 		}
405 		/*
406 		 * Redo route lookup with new destination address
407 		 */
408 		if (fwd_tag) {
409 			dest.s_addr = ((struct sockaddr_in *)
410 				    (fwd_tag + 1))->sin_addr.s_addr;
411 			m_tag_delete(m, fwd_tag);
412 			m->m_flags &= ~M_IP_NEXTHOP;
413 		}
414 		if (dest.s_addr != rtdest.s_addr &&
415 		    ip_findroute(&nh, dest, m) != 0)
416 			return (NULL);	/* icmp unreach already sent */
417 	}
418 
419 passout:
420 	/*
421 	 * Step 6: send off the packet
422 	 */
423 	ip_len = ntohs(ip->ip_len);
424 	ip_off = ntohs(ip->ip_off);
425 
426 	bzero(&ro, sizeof(ro));
427 	dst = (struct sockaddr_in *)&ro.ro_dst;
428 	dst->sin_family = AF_INET;
429 	dst->sin_len = sizeof(*dst);
430 	dst->sin_addr = dest;
431 	if (nh->nh_flags & NHF_GATEWAY) {
432 		gw = &nh->gw_sa;
433 		ro.ro_flags |= RT_HAS_GW;
434 	} else
435 		gw = (const struct sockaddr *)dst;
436 
437 	/*
438 	 * Handle redirect case.
439 	 */
440 	redest.s_addr = 0;
441 	if (V_ipsendredirects && (nh->nh_ifp == m->m_pkthdr.rcvif) &&
442 	    gw->sa_family == AF_INET)
443 		mcopy = ip_redir_alloc(m, nh, ip, &redest.s_addr);
444 
445 	/*
446 	 * Check if packet fits MTU or if hardware will fragment for us
447 	 */
448 	if (ip_len <= nh->nh_mtu) {
449 		/*
450 		 * Avoid confusing lower layers.
451 		 */
452 		m_clrprotoflags(m);
453 		/*
454 		 * Send off the packet via outgoing interface
455 		 */
456 		IP_PROBE(send, NULL, NULL, ip, nh->nh_ifp, ip, NULL);
457 		error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m, gw, &ro);
458 	} else {
459 		/*
460 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
461 		 */
462 		if (ip_off & IP_DF) {
463 			IPSTAT_INC(ips_cantfrag);
464 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
465 				0, nh->nh_mtu);
466 			goto consumed;
467 		} else {
468 			/*
469 			 * We have to fragment the packet
470 			 */
471 			m->m_pkthdr.csum_flags |= CSUM_IP;
472 			if (ip_fragment(ip, &m, nh->nh_mtu,
473 			    nh->nh_ifp->if_hwassist) != 0)
474 				goto drop;
475 			KASSERT(m != NULL, ("null mbuf and no error"));
476 			/*
477 			 * Send off the fragments via outgoing interface
478 			 */
479 			error = 0;
480 			do {
481 				m0 = m->m_nextpkt;
482 				m->m_nextpkt = NULL;
483 				/*
484 				 * Avoid confusing lower layers.
485 				 */
486 				m_clrprotoflags(m);
487 
488 				IP_PROBE(send, NULL, NULL,
489 				    mtod(m, struct ip *), nh->nh_ifp,
490 				    mtod(m, struct ip *), NULL);
491 				error = (*nh->nh_ifp->if_output)(nh->nh_ifp, m,
492 				    gw, &ro);
493 				if (error)
494 					break;
495 			} while ((m = m0) != NULL);
496 			if (error) {
497 				/* Reclaim remaining fragments */
498 				for (m = m0; m; m = m0) {
499 					m0 = m->m_nextpkt;
500 					m_freem(m);
501 				}
502 			} else
503 				IPSTAT_INC(ips_fragmented);
504 		}
505 	}
506 
507 	if (error != 0)
508 		IPSTAT_INC(ips_odropped);
509 	else {
510 		IPSTAT_INC(ips_forward);
511 		IPSTAT_INC(ips_fastforward);
512 	}
513 
514 	/* Send required redirect */
515 	if (mcopy != NULL) {
516 		icmp_error(mcopy, ICMP_REDIRECT, ICMP_REDIRECT_HOST, redest.s_addr, 0);
517 		mcopy = NULL; /* Freed by caller */
518 	}
519 
520 consumed:
521 	if (mcopy != NULL)
522 		m_freem(mcopy);
523 	return NULL;
524 drop:
525 	if (m)
526 		m_freem(m);
527 	return NULL;
528 }
529