xref: /freebsd/sys/netinet/ip_fastfwd.c (revision eb9da1ada8b6b2c74378a5c17029ec5a7fb199e6)
1 /*-
2  * Copyright (c) 2003 Andre Oppermann, Internet Business Solutions AG
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. The name of the author may not be used to endorse or promote
14  *    products derived from this software without specific prior written
15  *    permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * ip_fastforward gets its speed from processing the forwarded packet to
32  * completion (if_output on the other side) without any queues or netisr's.
33  * The receiving interface DMAs the packet into memory, the upper half of
34  * driver calls ip_fastforward, we do our routing table lookup and directly
35  * send it off to the outgoing interface, which DMAs the packet to the
36  * network card. The only part of the packet we touch with the CPU is the
37  * IP header (unless there are complex firewall rules touching other parts
38  * of the packet, but that is up to you). We are essentially limited by bus
39  * bandwidth and how fast the network card/driver can set up receives and
40  * transmits.
41  *
42  * We handle basic errors, IP header errors, checksum errors,
43  * destination unreachable, fragmentation and fragmentation needed and
44  * report them via ICMP to the sender.
45  *
46  * Else if something is not pure IPv4 unicast forwarding we fall back to
47  * the normal ip_input processing path. We should only be called from
48  * interfaces connected to the outside world.
49  *
50  * Firewalling is fully supported including divert, ipfw fwd and ipfilter
51  * ipnat and address rewrite.
52  *
53  * IPSEC is not supported if this host is a tunnel broker. IPSEC is
54  * supported for connections to/from local host.
55  *
56  * We try to do the least expensive (in CPU ops) checks and operations
57  * first to catch junk with as little overhead as possible.
58  *
59  * We take full advantage of hardware support for IP checksum and
60  * fragmentation offloading.
61  *
62  * We don't do ICMP redirect in the fast forwarding path. I have had my own
63  * cases where two core routers with Zebra routing suite would send millions
64  * ICMP redirects to connected hosts if the destination router was not the
65  * default gateway. In one case it was filling the routing table of a host
66  * with approximately 300.000 cloned redirect entries until it ran out of
67  * kernel memory. However the networking code proved very robust and it didn't
68  * crash or fail in other ways.
69  */
70 
71 /*
72  * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which
73  * is being followed here.
74  */
75 
76 #include <sys/cdefs.h>
77 __FBSDID("$FreeBSD$");
78 
79 #include "opt_ipstealth.h"
80 
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/kernel.h>
84 #include <sys/malloc.h>
85 #include <sys/mbuf.h>
86 #include <sys/protosw.h>
87 #include <sys/sdt.h>
88 #include <sys/socket.h>
89 #include <sys/sysctl.h>
90 
91 #include <net/pfil.h>
92 #include <net/if.h>
93 #include <net/if_types.h>
94 #include <net/if_var.h>
95 #include <net/if_dl.h>
96 #include <net/route.h>
97 #include <net/vnet.h>
98 
99 #include <netinet/in.h>
100 #include <netinet/in_kdtrace.h>
101 #include <netinet/in_systm.h>
102 #include <netinet/in_var.h>
103 #include <netinet/ip.h>
104 #include <netinet/ip_var.h>
105 #include <netinet/ip_icmp.h>
106 #include <netinet/ip_options.h>
107 
108 #include <machine/in_cksum.h>
109 
110 static struct sockaddr_in *
111 ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
112 {
113 	struct sockaddr_in *dst;
114 	struct rtentry *rt;
115 
116 	/*
117 	 * Find route to destination.
118 	 */
119 	bzero(ro, sizeof(*ro));
120 	dst = (struct sockaddr_in *)&ro->ro_dst;
121 	dst->sin_family = AF_INET;
122 	dst->sin_len = sizeof(*dst);
123 	dst->sin_addr.s_addr = dest.s_addr;
124 	in_rtalloc_ign(ro, 0, M_GETFIB(m));
125 
126 	/*
127 	 * Route there and interface still up?
128 	 */
129 	rt = ro->ro_rt;
130 	if (rt && (rt->rt_flags & RTF_UP) &&
131 	    (rt->rt_ifp->if_flags & IFF_UP) &&
132 	    (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) {
133 		if (rt->rt_flags & RTF_GATEWAY)
134 			dst = (struct sockaddr_in *)rt->rt_gateway;
135 	} else {
136 		IPSTAT_INC(ips_noroute);
137 		IPSTAT_INC(ips_cantforward);
138 		if (rt)
139 			RTFREE(rt);
140 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
141 		return NULL;
142 	}
143 	return dst;
144 }
145 
146 /*
147  * Try to forward a packet based on the destination address.
148  * This is a fast path optimized for the plain forwarding case.
149  * If the packet is handled (and consumed) here then we return NULL;
150  * otherwise mbuf is returned and the packet should be delivered
151  * to ip_input for full processing.
152  */
153 struct mbuf *
154 ip_tryforward(struct mbuf *m)
155 {
156 	struct ip *ip;
157 	struct mbuf *m0 = NULL;
158 	struct route ro;
159 	struct sockaddr_in *dst = NULL;
160 	struct ifnet *ifp;
161 	struct in_addr odest, dest;
162 	uint16_t ip_len, ip_off;
163 	int error = 0;
164 	int mtu;
165 	struct m_tag *fwd_tag = NULL;
166 
167 	/*
168 	 * Are we active and forwarding packets?
169 	 */
170 
171 	M_ASSERTVALID(m);
172 	M_ASSERTPKTHDR(m);
173 
174 	bzero(&ro, sizeof(ro));
175 
176 
177 #ifdef ALTQ
178 	/*
179 	 * Is packet dropped by traffic conditioner?
180 	 */
181 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
182 		goto drop;
183 #endif
184 
185 	/*
186 	 * Only IP packets without options
187 	 */
188 	ip = mtod(m, struct ip *);
189 
190 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
191 		if (V_ip_doopts == 1)
192 			return m;
193 		else if (V_ip_doopts == 2) {
194 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
195 				0, 0);
196 			return NULL;	/* mbuf already free'd */
197 		}
198 		/* else ignore IP options and continue */
199 	}
200 
201 	/*
202 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
203 	 * no multicast, no INADDR_ANY
204 	 *
205 	 * XXX: Probably some of these checks could be direct drop
206 	 * conditions.  However it is not clear whether there are some
207 	 * hacks or obscure behaviours which make it necessary to
208 	 * let ip_input handle it.  We play safe here and let ip_input
209 	 * deal with it until it is proven that we can directly drop it.
210 	 */
211 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
212 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
213 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
214 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
215 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
216 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
217 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
218 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
219 	    ip->ip_src.s_addr == INADDR_ANY ||
220 	    ip->ip_dst.s_addr == INADDR_ANY )
221 		return m;
222 
223 	/*
224 	 * Is it for a local address on this host?
225 	 */
226 	if (in_localip(ip->ip_dst))
227 		return m;
228 
229 	IPSTAT_INC(ips_total);
230 
231 	/*
232 	 * Step 3: incoming packet firewall processing
233 	 */
234 
235 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
236 
237 	/*
238 	 * Run through list of ipfilter hooks for input packets
239 	 */
240 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
241 		goto passin;
242 
243 	if (pfil_run_hooks(
244 	    &V_inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) ||
245 	    m == NULL)
246 		goto drop;
247 
248 	M_ASSERTVALID(m);
249 	M_ASSERTPKTHDR(m);
250 
251 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
252 	dest.s_addr = ip->ip_dst.s_addr;
253 
254 	/*
255 	 * Destination address changed?
256 	 */
257 	if (odest.s_addr != dest.s_addr) {
258 		/*
259 		 * Is it now for a local address on this host?
260 		 */
261 		if (in_localip(dest))
262 			goto forwardlocal;
263 		/*
264 		 * Go on with new destination address
265 		 */
266 	}
267 
268 	if (m->m_flags & M_FASTFWD_OURS) {
269 		/*
270 		 * ipfw changed it for a local address on this host.
271 		 */
272 		goto forwardlocal;
273 	}
274 
275 passin:
276 	/*
277 	 * Step 4: decrement TTL and look up route
278 	 */
279 
280 	/*
281 	 * Check TTL
282 	 */
283 #ifdef IPSTEALTH
284 	if (!V_ipstealth) {
285 #endif
286 	if (ip->ip_ttl <= IPTTLDEC) {
287 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
288 		return NULL;	/* mbuf already free'd */
289 	}
290 
291 	/*
292 	 * Decrement the TTL and incrementally change the IP header checksum.
293 	 * Don't bother doing this with hw checksum offloading, it's faster
294 	 * doing it right here.
295 	 */
296 	ip->ip_ttl -= IPTTLDEC;
297 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
298 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
299 	else
300 		ip->ip_sum += htons(IPTTLDEC << 8);
301 #ifdef IPSTEALTH
302 	}
303 #endif
304 
305 	/*
306 	 * Find route to destination.
307 	 */
308 	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
309 		return NULL;	/* icmp unreach already sent */
310 	ifp = ro.ro_rt->rt_ifp;
311 
312 	/*
313 	 * Immediately drop blackholed traffic, and directed broadcasts
314 	 * for either the all-ones or all-zero subnet addresses on
315 	 * locally attached networks.
316 	 */
317 	if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0)
318 		goto drop;
319 
320 	/*
321 	 * Step 5: outgoing firewall packet processing
322 	 */
323 
324 	/*
325 	 * Run through list of hooks for output packets.
326 	 */
327 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
328 		goto passout;
329 
330 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) {
331 		goto drop;
332 	}
333 
334 	M_ASSERTVALID(m);
335 	M_ASSERTPKTHDR(m);
336 
337 	ip = mtod(m, struct ip *);
338 	dest.s_addr = ip->ip_dst.s_addr;
339 
340 	/*
341 	 * Destination address changed?
342 	 */
343 	if (m->m_flags & M_IP_NEXTHOP)
344 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
345 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
346 		/*
347 		 * Is it now for a local address on this host?
348 		 */
349 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
350 forwardlocal:
351 			/*
352 			 * Return packet for processing by ip_input().
353 			 */
354 			m->m_flags |= M_FASTFWD_OURS;
355 			if (ro.ro_rt)
356 				RTFREE(ro.ro_rt);
357 			return m;
358 		}
359 		/*
360 		 * Redo route lookup with new destination address
361 		 */
362 		if (fwd_tag) {
363 			dest.s_addr = ((struct sockaddr_in *)
364 				    (fwd_tag + 1))->sin_addr.s_addr;
365 			m_tag_delete(m, fwd_tag);
366 			m->m_flags &= ~M_IP_NEXTHOP;
367 		}
368 		RTFREE(ro.ro_rt);
369 		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
370 			return NULL;	/* icmp unreach already sent */
371 		ifp = ro.ro_rt->rt_ifp;
372 	}
373 
374 passout:
375 	/*
376 	 * Step 6: send off the packet
377 	 */
378 	ip_len = ntohs(ip->ip_len);
379 	ip_off = ntohs(ip->ip_off);
380 
381 	/*
382 	 * Check if route is dampned (when ARP is unable to resolve)
383 	 */
384 	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
385 	    (ro.ro_rt->rt_expire == 0 || time_uptime < ro.ro_rt->rt_expire)) {
386 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
387 		goto consumed;
388 	}
389 
390 	/*
391 	 * Check if media link state of interface is not down
392 	 */
393 	if (ifp->if_link_state == LINK_STATE_DOWN) {
394 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
395 		goto consumed;
396 	}
397 
398 	/*
399 	 * Check if packet fits MTU or if hardware will fragment for us
400 	 */
401 	if (ro.ro_rt->rt_mtu)
402 		mtu = min(ro.ro_rt->rt_mtu, ifp->if_mtu);
403 	else
404 		mtu = ifp->if_mtu;
405 
406 	if (ip_len <= mtu) {
407 		/*
408 		 * Avoid confusing lower layers.
409 		 */
410 		m_clrprotoflags(m);
411 		/*
412 		 * Send off the packet via outgoing interface
413 		 */
414 		IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
415 		error = (*ifp->if_output)(ifp, m,
416 				(struct sockaddr *)dst, &ro);
417 	} else {
418 		/*
419 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
420 		 */
421 		if (ip_off & IP_DF) {
422 			IPSTAT_INC(ips_cantfrag);
423 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
424 				0, mtu);
425 			goto consumed;
426 		} else {
427 			/*
428 			 * We have to fragment the packet
429 			 */
430 			m->m_pkthdr.csum_flags |= CSUM_IP;
431 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist))
432 				goto drop;
433 			KASSERT(m != NULL, ("null mbuf and no error"));
434 			/*
435 			 * Send off the fragments via outgoing interface
436 			 */
437 			error = 0;
438 			do {
439 				m0 = m->m_nextpkt;
440 				m->m_nextpkt = NULL;
441 				/*
442 				 * Avoid confusing lower layers.
443 				 */
444 				m_clrprotoflags(m);
445 
446 				IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
447 				error = (*ifp->if_output)(ifp, m,
448 					(struct sockaddr *)dst, &ro);
449 				if (error)
450 					break;
451 			} while ((m = m0) != NULL);
452 			if (error) {
453 				/* Reclaim remaining fragments */
454 				for (m = m0; m; m = m0) {
455 					m0 = m->m_nextpkt;
456 					m_freem(m);
457 				}
458 			} else
459 				IPSTAT_INC(ips_fragmented);
460 		}
461 	}
462 
463 	if (error != 0)
464 		IPSTAT_INC(ips_odropped);
465 	else {
466 		counter_u64_add(ro.ro_rt->rt_pksent, 1);
467 		IPSTAT_INC(ips_forward);
468 		IPSTAT_INC(ips_fastforward);
469 	}
470 consumed:
471 	RTFREE(ro.ro_rt);
472 	return NULL;
473 drop:
474 	if (m)
475 		m_freem(m);
476 	if (ro.ro_rt)
477 		RTFREE(ro.ro_rt);
478 	return NULL;
479 }
480