xref: /freebsd/sys/netinet/ip_fastfwd.c (revision b88cc53d4ddda4486683ee2121f131b10ed21c30)
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/vnet.h>
100 
101 #include <netinet/in.h>
102 #include <netinet/in_fib.h>
103 #include <netinet/in_kdtrace.h>
104 #include <netinet/in_systm.h>
105 #include <netinet/in_var.h>
106 #include <netinet/ip.h>
107 #include <netinet/ip_var.h>
108 #include <netinet/ip_icmp.h>
109 #include <netinet/ip_options.h>
110 
111 #include <machine/in_cksum.h>
112 
113 static int
114 ip_findroute(struct nhop4_basic *pnh, struct in_addr dest, struct mbuf *m)
115 {
116 
117 	bzero(pnh, sizeof(*pnh));
118 	if (fib4_lookup_nh_basic(M_GETFIB(m), dest, 0, 0, pnh) != 0) {
119 		IPSTAT_INC(ips_noroute);
120 		IPSTAT_INC(ips_cantforward);
121 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
122 		return (EHOSTUNREACH);
123 	}
124 	/*
125 	 * Drop blackholed traffic and directed broadcasts.
126 	 */
127 	if ((pnh->nh_flags & (NHF_BLACKHOLE | NHF_BROADCAST)) != 0) {
128 		IPSTAT_INC(ips_cantforward);
129 		m_freem(m);
130 		return (EHOSTUNREACH);
131 	}
132 
133 	if (pnh->nh_flags & NHF_REJECT) {
134 		IPSTAT_INC(ips_cantforward);
135 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
136 		return (EHOSTUNREACH);
137 	}
138 
139 	return (0);
140 }
141 
142 /*
143  * Try to forward a packet based on the destination address.
144  * This is a fast path optimized for the plain forwarding case.
145  * If the packet is handled (and consumed) here then we return NULL;
146  * otherwise mbuf is returned and the packet should be delivered
147  * to ip_input for full processing.
148  */
149 struct mbuf *
150 ip_tryforward(struct mbuf *m)
151 {
152 	struct ip *ip;
153 	struct mbuf *m0 = NULL;
154 	struct nhop4_basic nh;
155 	struct sockaddr_in dst;
156 	struct in_addr dest, odest, rtdest;
157 	uint16_t ip_len, ip_off;
158 	int error = 0;
159 	struct m_tag *fwd_tag = NULL;
160 
161 	/*
162 	 * Are we active and forwarding packets?
163 	 */
164 
165 	M_ASSERTVALID(m);
166 	M_ASSERTPKTHDR(m);
167 
168 #ifdef ALTQ
169 	/*
170 	 * Is packet dropped by traffic conditioner?
171 	 */
172 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
173 		goto drop;
174 #endif
175 
176 	/*
177 	 * Only IP packets without options
178 	 */
179 	ip = mtod(m, struct ip *);
180 
181 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
182 		if (V_ip_doopts == 1)
183 			return m;
184 		else if (V_ip_doopts == 2) {
185 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
186 				0, 0);
187 			return NULL;	/* mbuf already free'd */
188 		}
189 		/* else ignore IP options and continue */
190 	}
191 
192 	/*
193 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
194 	 * no multicast, no INADDR_ANY
195 	 *
196 	 * XXX: Probably some of these checks could be direct drop
197 	 * conditions.  However it is not clear whether there are some
198 	 * hacks or obscure behaviours which make it necessary to
199 	 * let ip_input handle it.  We play safe here and let ip_input
200 	 * deal with it until it is proven that we can directly drop it.
201 	 */
202 	if ((m->m_flags & (M_BCAST|M_MCAST)) ||
203 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
204 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
205 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
206 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
207 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
208 	    IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
209 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
210 	    ip->ip_src.s_addr == INADDR_ANY ||
211 	    ip->ip_dst.s_addr == INADDR_ANY )
212 		return m;
213 
214 	/*
215 	 * Is it for a local address on this host?
216 	 */
217 	if (in_localip(ip->ip_dst))
218 		return m;
219 
220 	IPSTAT_INC(ips_total);
221 
222 	/*
223 	 * Step 3: incoming packet firewall processing
224 	 */
225 
226 	odest.s_addr = dest.s_addr = ip->ip_dst.s_addr;
227 
228 	/*
229 	 * Run through list of ipfilter hooks for input packets
230 	 */
231 	if (!PFIL_HOOKED_IN(V_inet_pfil_head))
232 		goto passin;
233 
234 	if (pfil_run_hooks(V_inet_pfil_head, &m, m->m_pkthdr.rcvif, PFIL_IN,
235 	    NULL) != PFIL_PASS)
236 		goto drop;
237 
238 	M_ASSERTVALID(m);
239 	M_ASSERTPKTHDR(m);
240 
241 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
242 	dest.s_addr = ip->ip_dst.s_addr;
243 
244 	/*
245 	 * Destination address changed?
246 	 */
247 	if (odest.s_addr != dest.s_addr) {
248 		/*
249 		 * Is it now for a local address on this host?
250 		 */
251 		if (in_localip(dest))
252 			goto forwardlocal;
253 		/*
254 		 * Go on with new destination address
255 		 */
256 	}
257 
258 	if (m->m_flags & M_FASTFWD_OURS) {
259 		/*
260 		 * ipfw changed it for a local address on this host.
261 		 */
262 		goto forwardlocal;
263 	}
264 
265 passin:
266 	/*
267 	 * Step 4: decrement TTL and look up route
268 	 */
269 
270 	/*
271 	 * Check TTL
272 	 */
273 #ifdef IPSTEALTH
274 	if (!V_ipstealth) {
275 #endif
276 	if (ip->ip_ttl <= IPTTLDEC) {
277 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0);
278 		return NULL;	/* mbuf already free'd */
279 	}
280 
281 	/*
282 	 * Decrement the TTL and incrementally change the IP header checksum.
283 	 * Don't bother doing this with hw checksum offloading, it's faster
284 	 * doing it right here.
285 	 */
286 	ip->ip_ttl -= IPTTLDEC;
287 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
288 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
289 	else
290 		ip->ip_sum += htons(IPTTLDEC << 8);
291 #ifdef IPSTEALTH
292 	}
293 #endif
294 
295 	/*
296 	 * Next hop forced by pfil(9) hook?
297 	 */
298 	if ((m->m_flags & M_IP_NEXTHOP) &&
299 	    ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
300 		/*
301 		 * Now we will find route to forced destination.
302 		 */
303 		dest.s_addr = ((struct sockaddr_in *)
304 			    (fwd_tag + 1))->sin_addr.s_addr;
305 		m_tag_delete(m, fwd_tag);
306 		m->m_flags &= ~M_IP_NEXTHOP;
307 	}
308 
309 	/*
310 	 * Find route to destination.
311 	 */
312 	if (ip_findroute(&nh, dest, m) != 0)
313 		return (NULL);	/* icmp unreach already sent */
314 
315 	/*
316 	 * Avoid second route lookup by caching destination.
317 	 */
318 	rtdest.s_addr = dest.s_addr;
319 
320 	/*
321 	 * Step 5: outgoing firewall packet processing
322 	 */
323 	if (!PFIL_HOOKED_OUT(V_inet_pfil_head))
324 		goto passout;
325 
326 	if (pfil_run_hooks(V_inet_pfil_head, &m, nh.nh_ifp,
327 	    PFIL_OUT | PFIL_FWD, NULL) != PFIL_PASS)
328 		goto drop;
329 
330 	M_ASSERTVALID(m);
331 	M_ASSERTPKTHDR(m);
332 
333 	ip = mtod(m, struct ip *);
334 	dest.s_addr = ip->ip_dst.s_addr;
335 
336 	/*
337 	 * Destination address changed?
338 	 */
339 	if (m->m_flags & M_IP_NEXTHOP)
340 		fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
341 	else
342 		fwd_tag = NULL;
343 	if (odest.s_addr != dest.s_addr || fwd_tag != NULL) {
344 		/*
345 		 * Is it now for a local address on this host?
346 		 */
347 		if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) {
348 forwardlocal:
349 			/*
350 			 * Return packet for processing by ip_input().
351 			 */
352 			m->m_flags |= M_FASTFWD_OURS;
353 			return (m);
354 		}
355 		/*
356 		 * Redo route lookup with new destination address
357 		 */
358 		if (fwd_tag) {
359 			dest.s_addr = ((struct sockaddr_in *)
360 				    (fwd_tag + 1))->sin_addr.s_addr;
361 			m_tag_delete(m, fwd_tag);
362 			m->m_flags &= ~M_IP_NEXTHOP;
363 		}
364 		if (dest.s_addr != rtdest.s_addr &&
365 		    ip_findroute(&nh, dest, m) != 0)
366 			return (NULL);	/* icmp unreach already sent */
367 	}
368 
369 passout:
370 	/*
371 	 * Step 6: send off the packet
372 	 */
373 	ip_len = ntohs(ip->ip_len);
374 	ip_off = ntohs(ip->ip_off);
375 
376 	bzero(&dst, sizeof(dst));
377 	dst.sin_family = AF_INET;
378 	dst.sin_len = sizeof(dst);
379 	dst.sin_addr = nh.nh_addr;
380 
381 	/*
382 	 * Check if packet fits MTU or if hardware will fragment for us
383 	 */
384 	if (ip_len <= nh.nh_mtu) {
385 		/*
386 		 * Avoid confusing lower layers.
387 		 */
388 		m_clrprotoflags(m);
389 		/*
390 		 * Send off the packet via outgoing interface
391 		 */
392 		IP_PROBE(send, NULL, NULL, ip, nh.nh_ifp, ip, NULL);
393 		error = (*nh.nh_ifp->if_output)(nh.nh_ifp, m,
394 		    (struct sockaddr *)&dst, NULL);
395 	} else {
396 		/*
397 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
398 		 */
399 		if (ip_off & IP_DF) {
400 			IPSTAT_INC(ips_cantfrag);
401 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
402 				0, nh.nh_mtu);
403 			goto consumed;
404 		} else {
405 			/*
406 			 * We have to fragment the packet
407 			 */
408 			m->m_pkthdr.csum_flags |= CSUM_IP;
409 			if (ip_fragment(ip, &m, nh.nh_mtu,
410 			    nh.nh_ifp->if_hwassist) != 0)
411 				goto drop;
412 			KASSERT(m != NULL, ("null mbuf and no error"));
413 			/*
414 			 * Send off the fragments via outgoing interface
415 			 */
416 			error = 0;
417 			do {
418 				m0 = m->m_nextpkt;
419 				m->m_nextpkt = NULL;
420 				/*
421 				 * Avoid confusing lower layers.
422 				 */
423 				m_clrprotoflags(m);
424 
425 				IP_PROBE(send, NULL, NULL,
426 				    mtod(m, struct ip *), nh.nh_ifp,
427 				    mtod(m, struct ip *), NULL);
428 				/* XXX: we can use cached route here */
429 				error = (*nh.nh_ifp->if_output)(nh.nh_ifp, m,
430 				    (struct sockaddr *)&dst, NULL);
431 				if (error)
432 					break;
433 			} while ((m = m0) != NULL);
434 			if (error) {
435 				/* Reclaim remaining fragments */
436 				for (m = m0; m; m = m0) {
437 					m0 = m->m_nextpkt;
438 					m_freem(m);
439 				}
440 			} else
441 				IPSTAT_INC(ips_fragmented);
442 		}
443 	}
444 
445 	if (error != 0)
446 		IPSTAT_INC(ips_odropped);
447 	else {
448 		IPSTAT_INC(ips_forward);
449 		IPSTAT_INC(ips_fastforward);
450 	}
451 consumed:
452 	return NULL;
453 drop:
454 	if (m)
455 		m_freem(m);
456 	return NULL;
457 }
458