xref: /freebsd/sys/netinet/ip_fastfwd.c (revision 430f7286a566b1407c7b32ce13585caf5aa59b92)
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_ipfw.h"
80 #include "opt_ipstealth.h"
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/kernel.h>
85 #include <sys/malloc.h>
86 #include <sys/mbuf.h>
87 #include <sys/protosw.h>
88 #include <sys/sdt.h>
89 #include <sys/socket.h>
90 #include <sys/sysctl.h>
91 
92 #include <net/pfil.h>
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/route.h>
98 #include <net/vnet.h>
99 
100 #include <netinet/in.h>
101 #include <netinet/in_kdtrace.h>
102 #include <netinet/in_systm.h>
103 #include <netinet/in_var.h>
104 #include <netinet/ip.h>
105 #include <netinet/ip_var.h>
106 #include <netinet/ip_icmp.h>
107 #include <netinet/ip_options.h>
108 
109 #include <machine/in_cksum.h>
110 
111 static struct sockaddr_in *
112 ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m)
113 {
114 	struct sockaddr_in *dst;
115 	struct rtentry *rt;
116 
117 	/*
118 	 * Find route to destination.
119 	 */
120 	bzero(ro, sizeof(*ro));
121 	dst = (struct sockaddr_in *)&ro->ro_dst;
122 	dst->sin_family = AF_INET;
123 	dst->sin_len = sizeof(*dst);
124 	dst->sin_addr.s_addr = dest.s_addr;
125 	in_rtalloc_ign(ro, 0, M_GETFIB(m));
126 
127 	/*
128 	 * Route there and interface still up?
129 	 */
130 	rt = ro->ro_rt;
131 	if (rt && (rt->rt_flags & RTF_UP) &&
132 	    (rt->rt_ifp->if_flags & IFF_UP) &&
133 	    (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) {
134 		if (rt->rt_flags & RTF_GATEWAY)
135 			dst = (struct sockaddr_in *)rt->rt_gateway;
136 	} else {
137 		IPSTAT_INC(ips_noroute);
138 		IPSTAT_INC(ips_cantforward);
139 		if (rt)
140 			RTFREE(rt);
141 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
142 		return NULL;
143 	}
144 	return dst;
145 }
146 
147 /*
148  * Try to forward a packet based on the destination address.
149  * This is a fast path optimized for the plain forwarding case.
150  * If the packet is handled (and consumed) here then we return NULL;
151  * otherwise mbuf is returned and the packet should be delivered
152  * to ip_input for full processing.
153  */
154 struct mbuf *
155 ip_tryforward(struct mbuf *m)
156 {
157 	struct ip *ip;
158 	struct mbuf *m0 = NULL;
159 	struct route ro;
160 	struct sockaddr_in *dst = NULL;
161 	struct ifnet *ifp;
162 	struct in_addr odest, dest;
163 	uint16_t ip_len, ip_off;
164 	int error = 0;
165 	int mtu;
166 	struct m_tag *fwd_tag = NULL;
167 
168 	/*
169 	 * Are we active and forwarding packets?
170 	 */
171 
172 	M_ASSERTVALID(m);
173 	M_ASSERTPKTHDR(m);
174 
175 	bzero(&ro, sizeof(ro));
176 
177 
178 #ifdef ALTQ
179 	/*
180 	 * Is packet dropped by traffic conditioner?
181 	 */
182 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
183 		goto drop;
184 #endif
185 
186 	/*
187 	 * Only IP packets without options
188 	 */
189 	ip = mtod(m, struct ip *);
190 
191 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
192 		if (V_ip_doopts == 1)
193 			return m;
194 		else if (V_ip_doopts == 2) {
195 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
196 				0, 0);
197 			return NULL;	/* mbuf already free'd */
198 		}
199 		/* else ignore IP options and continue */
200 	}
201 
202 	/*
203 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
204 	 * no multicast, no INADDR_ANY
205 	 *
206 	 * XXX: Probably some of these checks could be direct drop
207 	 * conditions.  However it is not clear whether there are some
208 	 * hacks or obscure behaviours which make it necessary to
209 	 * let ip_input handle it.  We play safe here and let ip_input
210 	 * deal with it until it is proven that we can directly drop it.
211 	 */
212 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
213 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
214 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
215 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
216 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
217 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
218 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
219 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
220 	    ip->ip_src.s_addr == INADDR_ANY ||
221 	    ip->ip_dst.s_addr == INADDR_ANY )
222 		return m;
223 
224 	/*
225 	 * Is it for a local address on this host?
226 	 */
227 	if (in_localip(ip->ip_dst))
228 		return m;
229 
230 	IPSTAT_INC(ips_total);
231 
232 	/*
233 	 * Step 3: incoming packet firewall processing
234 	 */
235 
236 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
237 
238 	/*
239 	 * Run through list of ipfilter hooks for input packets
240 	 */
241 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
242 		goto passin;
243 
244 	if (pfil_run_hooks(
245 	    &V_inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) ||
246 	    m == NULL)
247 		goto drop;
248 
249 	M_ASSERTVALID(m);
250 	M_ASSERTPKTHDR(m);
251 
252 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
253 	dest.s_addr = ip->ip_dst.s_addr;
254 
255 	/*
256 	 * Destination address changed?
257 	 */
258 	if (odest.s_addr != dest.s_addr) {
259 		/*
260 		 * Is it now for a local address on this host?
261 		 */
262 		if (in_localip(dest))
263 			goto forwardlocal;
264 		/*
265 		 * Go on with new destination address
266 		 */
267 	}
268 
269 	if (m->m_flags & M_FASTFWD_OURS) {
270 		/*
271 		 * ipfw changed it for a local address on this host.
272 		 */
273 		goto forwardlocal;
274 	}
275 
276 passin:
277 	/*
278 	 * Step 4: decrement TTL and look up route
279 	 */
280 
281 	/*
282 	 * Check TTL
283 	 */
284 #ifdef IPSTEALTH
285 	if (!V_ipstealth) {
286 #endif
287 	if (ip->ip_ttl <= IPTTLDEC) {
288 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
289 		return NULL;	/* mbuf already free'd */
290 	}
291 
292 	/*
293 	 * Decrement the TTL and incrementally change the IP header checksum.
294 	 * Don't bother doing this with hw checksum offloading, it's faster
295 	 * doing it right here.
296 	 */
297 	ip->ip_ttl -= IPTTLDEC;
298 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
299 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
300 	else
301 		ip->ip_sum += htons(IPTTLDEC << 8);
302 #ifdef IPSTEALTH
303 	}
304 #endif
305 
306 	/*
307 	 * Find route to destination.
308 	 */
309 	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
310 		return NULL;	/* icmp unreach already sent */
311 	ifp = ro.ro_rt->rt_ifp;
312 
313 	/*
314 	 * Immediately drop blackholed traffic, and directed broadcasts
315 	 * for either the all-ones or all-zero subnet addresses on
316 	 * locally attached networks.
317 	 */
318 	if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0)
319 		goto drop;
320 
321 	/*
322 	 * Step 5: outgoing firewall packet processing
323 	 */
324 
325 	/*
326 	 * Run through list of hooks for output packets.
327 	 */
328 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
329 		goto passout;
330 
331 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) {
332 		goto drop;
333 	}
334 
335 	M_ASSERTVALID(m);
336 	M_ASSERTPKTHDR(m);
337 
338 	ip = mtod(m, struct ip *);
339 	dest.s_addr = ip->ip_dst.s_addr;
340 
341 	/*
342 	 * Destination address changed?
343 	 */
344 	if (m->m_flags & M_IP_NEXTHOP)
345 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
346 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
347 		/*
348 		 * Is it now for a local address on this host?
349 		 */
350 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
351 forwardlocal:
352 			/*
353 			 * Return packet for processing by ip_input().
354 			 */
355 			m->m_flags |= M_FASTFWD_OURS;
356 			if (ro.ro_rt)
357 				RTFREE(ro.ro_rt);
358 			return m;
359 		}
360 		/*
361 		 * Redo route lookup with new destination address
362 		 */
363 		if (fwd_tag) {
364 			dest.s_addr = ((struct sockaddr_in *)
365 				    (fwd_tag + 1))->sin_addr.s_addr;
366 			m_tag_delete(m, fwd_tag);
367 			m->m_flags &= ~M_IP_NEXTHOP;
368 		}
369 		RTFREE(ro.ro_rt);
370 		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
371 			return NULL;	/* icmp unreach already sent */
372 		ifp = ro.ro_rt->rt_ifp;
373 	}
374 
375 passout:
376 	/*
377 	 * Step 6: send off the packet
378 	 */
379 	ip_len = ntohs(ip->ip_len);
380 	ip_off = ntohs(ip->ip_off);
381 
382 	/*
383 	 * Check if route is dampned (when ARP is unable to resolve)
384 	 */
385 	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
386 	    (ro.ro_rt->rt_expire == 0 || time_uptime < ro.ro_rt->rt_expire)) {
387 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
388 		goto consumed;
389 	}
390 
391 	/*
392 	 * Check if media link state of interface is not down
393 	 */
394 	if (ifp->if_link_state == LINK_STATE_DOWN) {
395 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
396 		goto consumed;
397 	}
398 
399 	/*
400 	 * Check if packet fits MTU or if hardware will fragment for us
401 	 */
402 	if (ro.ro_rt->rt_mtu)
403 		mtu = min(ro.ro_rt->rt_mtu, ifp->if_mtu);
404 	else
405 		mtu = ifp->if_mtu;
406 
407 	if (ip_len <= mtu) {
408 		/*
409 		 * Avoid confusing lower layers.
410 		 */
411 		m_clrprotoflags(m);
412 		/*
413 		 * Send off the packet via outgoing interface
414 		 */
415 		IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
416 		error = (*ifp->if_output)(ifp, m,
417 				(struct sockaddr *)dst, &ro);
418 	} else {
419 		/*
420 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
421 		 */
422 		if (ip_off & IP_DF) {
423 			IPSTAT_INC(ips_cantfrag);
424 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
425 				0, mtu);
426 			goto consumed;
427 		} else {
428 			/*
429 			 * We have to fragment the packet
430 			 */
431 			m->m_pkthdr.csum_flags |= CSUM_IP;
432 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist))
433 				goto drop;
434 			KASSERT(m != NULL, ("null mbuf and no error"));
435 			/*
436 			 * Send off the fragments via outgoing interface
437 			 */
438 			error = 0;
439 			do {
440 				m0 = m->m_nextpkt;
441 				m->m_nextpkt = NULL;
442 				/*
443 				 * Avoid confusing lower layers.
444 				 */
445 				m_clrprotoflags(m);
446 
447 				IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
448 				error = (*ifp->if_output)(ifp, m,
449 					(struct sockaddr *)dst, &ro);
450 				if (error)
451 					break;
452 			} while ((m = m0) != NULL);
453 			if (error) {
454 				/* Reclaim remaining fragments */
455 				for (m = m0; m; m = m0) {
456 					m0 = m->m_nextpkt;
457 					m_freem(m);
458 				}
459 			} else
460 				IPSTAT_INC(ips_fragmented);
461 		}
462 	}
463 
464 	if (error != 0)
465 		IPSTAT_INC(ips_odropped);
466 	else {
467 		counter_u64_add(ro.ro_rt->rt_pksent, 1);
468 		IPSTAT_INC(ips_forward);
469 		IPSTAT_INC(ips_fastforward);
470 	}
471 consumed:
472 	RTFREE(ro.ro_rt);
473 	return NULL;
474 drop:
475 	if (m)
476 		m_freem(m);
477 	if (ro.ro_rt)
478 		RTFREE(ro.ro_rt);
479 	return NULL;
480 }
481