xref: /freebsd/sys/netinet/ip_fastfwd.c (revision 0f8f86b71f022b803e99151c19db81b280f245dc)
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  * $FreeBSD$
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 router to dest was not the default
67  * gateway. In one case it was filling the routing table of a host with close
68  * 300'000 cloned redirect entries until it ran out of kernel memory. However
69  * the networking code proved very robust and it didn't crash or went ill
70  * otherwise.
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 "opt_ipfw.h"
79 #include "opt_ipdn.h"
80 #include "opt_ipdivert.h"
81 #include "opt_ipfilter.h"
82 #include "opt_ipstealth.h"
83 #include "opt_mac.h"
84 #include "opt_pfil_hooks.h"
85 
86 #include <sys/param.h>
87 #include <sys/systm.h>
88 #include <sys/kernel.h>
89 #include <sys/mac.h>
90 #include <sys/malloc.h>
91 #include <sys/mbuf.h>
92 #include <sys/protosw.h>
93 #include <sys/socket.h>
94 #include <sys/sysctl.h>
95 
96 #include <net/pfil.h>
97 #include <net/if.h>
98 #include <net/if_types.h>
99 #include <net/if_var.h>
100 #include <net/if_dl.h>
101 #include <net/route.h>
102 
103 #include <netinet/in.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 
110 #include <machine/in_cksum.h>
111 
112 #include <netinet/ip_fw.h>
113 #include <netinet/ip_divert.h>
114 #include <netinet/ip_dummynet.h>
115 
116 static int ipfastforward_active = 0;
117 SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
118     &ipfastforward_active, 0, "Enable fast IP forwarding");
119 
120 /*
121  * Try to forward a packet based on the destination address.
122  * This is a fast path optimized for the plain forwarding case.
123  * If the packet is handled (and consumed) here then we return 1;
124  * otherwise 0 is returned and the packet should be delivered
125  * to ip_input for full processing.
126  */
127 int
128 ip_fastforward(struct mbuf *m)
129 {
130 	struct ip *ip;
131 	struct mbuf *m0 = NULL;
132 #ifdef IPDIVERT
133 	struct ip *tip;
134 	struct mbuf *clone = NULL;
135 #endif
136 	struct route ro;
137 	struct sockaddr_in *dst = NULL;
138 	struct in_ifaddr *ia = NULL;
139 	struct ifaddr *ifa = NULL;
140 	struct ifnet *ifp = NULL;
141 	struct ip_fw_args args;
142 	in_addr_t odest, dest;
143 	u_short sum;
144 	int error = 0;
145 	int hlen, ipfw, mtu;
146 
147 	/*
148 	 * Are we active and forwarding packets?
149 	 */
150 	if (!ipfastforward_active || !ipforwarding)
151 		return 0;
152 
153 	M_ASSERTVALID(m);
154 	M_ASSERTPKTHDR(m);
155 
156 	/*
157 	 * Step 1: check for packet drop conditions (and sanity checks)
158 	 */
159 
160 	/*
161 	 * Is entire packet big enough?
162 	 */
163 	if (m->m_pkthdr.len < sizeof(struct ip)) {
164 		ipstat.ips_tooshort++;
165 		goto drop;
166 	}
167 
168 	/*
169 	 * Is first mbuf large enough for ip header and is header present?
170 	 */
171 	if (m->m_len < sizeof (struct ip) &&
172 	   (m = m_pullup(m, sizeof (struct ip))) == 0) {
173 		ipstat.ips_toosmall++;
174 		goto drop;
175 	}
176 
177 	ip = mtod(m, struct ip *);
178 
179 	/*
180 	 * Is it IPv4?
181 	 */
182 	if (ip->ip_v != IPVERSION) {
183 		ipstat.ips_badvers++;
184 		goto drop;
185 	}
186 
187 	/*
188 	 * Is IP header length correct and is it in first mbuf?
189 	 */
190 	hlen = ip->ip_hl << 2;
191 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
192 		ipstat.ips_badlen++;
193 		goto drop;
194 	}
195 	if (hlen > m->m_len) {
196 		if ((m = m_pullup(m, hlen)) == 0) {
197 			ipstat.ips_badhlen++;
198 			goto drop;
199 		}
200 		ip = mtod(m, struct ip *);
201 	}
202 
203 	/*
204 	 * Checksum correct?
205 	 */
206 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
207 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
208 	else {
209 		if (hlen == sizeof(struct ip))
210 			sum = in_cksum_hdr(ip);
211 		else
212 			sum = in_cksum(m, hlen);
213 	}
214 	if (sum) {
215 		ipstat.ips_badsum++;
216 		goto drop;
217 	}
218 	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
219 
220 	/*
221 	 * Convert to host representation
222 	 */
223 	ip->ip_len = ntohs(ip->ip_len);
224 	ip->ip_off = ntohs(ip->ip_off);
225 
226 	/*
227 	 * Is IP length longer than packet we have got?
228 	 */
229 	if (m->m_pkthdr.len < ip->ip_len) {
230 		ipstat.ips_tooshort++;
231 		goto drop;
232 	}
233 
234 	/*
235 	 * Is packet longer than IP header tells us? If yes, truncate packet.
236 	 */
237 	if (m->m_pkthdr.len > ip->ip_len) {
238 		if (m->m_len == m->m_pkthdr.len) {
239 			m->m_len = ip->ip_len;
240 			m->m_pkthdr.len = ip->ip_len;
241 		} else
242 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
243 	}
244 
245 	/*
246 	 * Is packet from or to 127/8?
247 	 */
248 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
249 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
250 		ipstat.ips_badaddr++;
251 		goto drop;
252 	}
253 
254 	/*
255 	 * Step 2: fallback conditions to normal ip_input path processing
256 	 */
257 
258 	/*
259 	 * Only IP packets without options
260 	 */
261 	if (ip->ip_hl != (sizeof(struct ip) >> 2))
262 		goto fallback;
263 
264 	/*
265 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
266 	 * no multicast, no INADDR_ANY
267 	 *
268 	 * XXX: Probably some of these checks could be direct drop
269 	 * conditions.  However it is not clear whether there are some
270 	 * hacks or obscure behaviours which make it neccessary to
271 	 * let ip_input handle it.  We play safe here and let ip_input
272 	 * deal with it until it is proven that we can directly drop it.
273 	 */
274 	if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
275 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
276 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
277 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
278 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
279 	    ip->ip_dst.s_addr == INADDR_ANY )
280 		goto fallback;
281 
282 	/*
283 	 * Is it for a local address on this host?
284 	 */
285 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
286 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
287 			goto fallback;
288 	}
289 
290 	/*
291 	 * Or is it for a local IP broadcast address on this host?
292 	 */
293 	if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
294 	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
295 			if (ifa->ifa_addr->sa_family != AF_INET)
296 				continue;
297 			ia = ifatoia(ifa);
298 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
299 				goto fallback;
300 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
301 			    ip->ip_dst.s_addr)
302 				goto fallback;
303 			continue;
304 fallback:
305 			/* return packet back to netisr for slow processing */
306 			ip->ip_len = htons(ip->ip_len);
307 			ip->ip_off = htons(ip->ip_off);
308 			return 0;
309 		}
310 	}
311 	ipstat.ips_total++;
312 
313 	/*
314 	 * Step 3: incoming packet firewall processing
315 	 */
316 
317 	odest = dest = ip->ip_dst.s_addr;
318 #ifdef PFIL_HOOKS
319 	/*
320 	 * Run through list of ipfilter hooks for input packets
321 	 */
322 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN) ||
323 	    m == NULL)
324 		return 1;
325 
326 	M_ASSERTVALID(m);
327 	M_ASSERTPKTHDR(m);
328 
329 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
330 	dest = ip->ip_dst.s_addr;
331 #endif
332 
333 	/*
334 	 * Run through ipfw for input packets
335 	 */
336 	if (fw_enable && IPFW_LOADED) {
337 		bzero(&args, sizeof(args));
338 		args.m = m;
339 
340 		ipfw = ip_fw_chk_ptr(&args);
341 		m = args.m;
342 
343 		M_ASSERTVALID(m);
344 		M_ASSERTPKTHDR(m);
345 
346 		/*
347 		 * Packet denied, drop it
348 		 */
349 		if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
350 			goto drop;
351 		/*
352 		 * Send packet to the appropriate pipe
353 		 */
354 		if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
355 			ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_IN, &args);
356 			return 1;
357 		}
358 #ifdef IPDIVERT
359 		/*
360 		 * Divert packet
361 		 */
362 		if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
363 			/*
364 			 * See if this is a fragment
365 			 */
366 			if (ip->ip_off & (IP_MF | IP_OFFMASK))
367 				goto droptoours;
368 			/*
369 			 * Tee packet
370 			 */
371 			if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
372 				clone = divert_clone(m);
373 			else
374 				clone = m;
375 			if (clone == NULL)
376 				goto passin;
377 
378 			/*
379 			 * Delayed checksums are not compatible
380 			 */
381 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
382 				in_delayed_cksum(m);
383 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
384 			}
385 			/*
386 			 * Restore packet header fields to original values
387 			 */
388 			tip = mtod(m, struct ip *);
389 			tip->ip_len = htons(tip->ip_len);
390 			tip->ip_off = htons(tip->ip_off);
391 			/*
392 			 * Deliver packet to divert input routine
393 			 */
394 			divert_packet(m, 0);
395 			/*
396 			 * If this was not tee, we are done
397 			 */
398 			m = clone;
399 			if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0)
400 				return 1;
401 			/* Continue if it was tee */
402 			goto passin;
403 		}
404 #endif
405 		if (ipfw == 0 && args.next_hop != NULL) {
406 			dest = args.next_hop->sin_addr.s_addr;
407 			goto passin;
408 		}
409 		/*
410 		 * Let through or not?
411 		 */
412 		if (ipfw != 0)
413 			goto drop;
414 	}
415 passin:
416 	ip = mtod(m, struct ip *);	/* if m changed during fw processing */
417 
418 	/*
419 	 * Destination address changed?
420 	 */
421 	if (odest != dest) {
422 		/*
423 		 * Is it now for a local address on this host?
424 		 */
425 		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
426 			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
427 				goto forwardlocal;
428 		}
429 		/*
430 		 * Go on with new destination address
431 		 */
432 	}
433 
434 	/*
435 	 * Step 4: decrement TTL and look up route
436 	 */
437 
438 	/*
439 	 * Check TTL
440 	 */
441 #ifdef IPSTEALTH
442 	if (!ipstealth) {
443 #endif
444 	if (ip->ip_ttl <= IPTTLDEC) {
445 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, NULL);
446 		return 1;
447 	}
448 
449 	/*
450 	 * Decrement the TTL and incrementally change the checksum.
451 	 * Don't bother doing this with hw checksum offloading.
452 	 */
453 	ip->ip_ttl -= IPTTLDEC;
454 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
455 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
456 	else
457 		ip->ip_sum += htons(IPTTLDEC << 8);
458 #ifdef IPSTEALTH
459 	}
460 #endif
461 
462 	/*
463 	 * Find route to destination.
464 	 */
465 	bzero(&ro, sizeof(ro));
466 	dst = (struct sockaddr_in *)&ro.ro_dst;
467 	dst->sin_family = AF_INET;
468 	dst->sin_len = sizeof(*dst);
469 	dst->sin_addr.s_addr = dest;
470 	rtalloc_ign(&ro, RTF_CLONING);
471 
472 	/*
473 	 * Route there and interface still up?
474 	 */
475 	if (ro.ro_rt &&
476 	    (ro.ro_rt->rt_flags & RTF_UP) &&
477 	    (ro.ro_rt->rt_ifp->if_flags & IFF_UP)) {
478 		ia = ifatoia(ro.ro_rt->rt_ifa);
479 		ifp = ro.ro_rt->rt_ifp;
480 		if (ro.ro_rt->rt_flags & RTF_GATEWAY)
481 			dst = (struct sockaddr_in *)ro.ro_rt->rt_gateway;
482 	} else {
483 		ipstat.ips_noroute++;
484 		ipstat.ips_cantforward++;
485 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
486 		if (ro.ro_rt)
487 			RTFREE(ro.ro_rt);
488 		return 1;
489 	}
490 
491 	/*
492 	 * Step 5: outgoing firewall packet processing
493 	 */
494 
495 #ifdef PFIL_HOOKS
496 	/*
497 	 * Run through list of hooks for output packets.
498 	 */
499 	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT) || m == NULL) {
500 		RTFREE(ro.ro_rt);
501 		return 1;
502 	}
503 
504 	M_ASSERTVALID(m);
505 	M_ASSERTPKTHDR(m);
506 
507 	ip = mtod(m, struct ip *);
508 	dest = ip->ip_dst.s_addr;
509 #endif
510 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
511 		bzero(&args, sizeof(args));
512 		args.m = m;
513 		args.oif = ifp;
514 
515 		ipfw = ip_fw_chk_ptr(&args);
516 		m = args.m;
517 
518 		M_ASSERTVALID(m);
519 		M_ASSERTPKTHDR(m);
520 
521 		if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL) {
522 			RTFREE(ro.ro_rt);
523 			goto drop;
524 		}
525 		if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
526 			/*
527 			 * XXX note: if the ifp or rt entry are deleted
528 			 * while a pkt is in dummynet, we are in trouble!
529 			 */
530 			args.ro = &ro;		/* dummynet does not save it */
531 			args.dst = dst;
532 
533 			ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_OUT, &args);
534 			RTFREE(ro.ro_rt);
535 			return 1;
536 		}
537 #ifdef IPDIVERT
538 		if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
539 			/*
540 			 * See if this is a fragment
541 			 */
542 			if (ip->ip_off & (IP_MF | IP_OFFMASK))
543 				goto droptoours;
544 			/*
545 			 * Tee packet
546 			 */
547 			if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
548 				clone = divert_clone(m);
549 			else
550 				clone = m;
551 			if (clone == NULL)
552 				goto passout;
553 
554 			/*
555 			 * Delayed checksums are not compatible with divert
556 			 */
557 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
558 				in_delayed_cksum(m);
559 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
560 			}
561 			/*
562 			 * Restore packet header fields to original values
563 			 */
564 			tip = mtod(m, struct ip *);
565 			tip->ip_len = htons(tip->ip_len);
566 			tip->ip_off = htons(tip->ip_off);
567 			/*
568 			 * Deliver packet to divert input routine
569 			 */
570 			divert_packet(m, 0);
571 			/*
572 			 * If this was not tee, we are done
573 			 */
574 			m = clone;
575 			if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0) {
576 				RTFREE(ro.ro_rt);
577 				return 1;
578 			}
579 			/* Continue if it was tee */
580 			goto passout;
581 		}
582 #endif
583 		if (ipfw == 0 && args.next_hop != NULL) {
584 			dest = args.next_hop->sin_addr.s_addr;
585 			goto passout;
586 		}
587 		/*
588 		 * Let through or not?
589 		 */
590 		if (ipfw != 0)
591 			goto drop;
592 	}
593 passout:
594 	ip = mtod(m, struct ip *);
595 
596 	/*
597 	 * Destination address changed?
598 	 */
599 	if (odest != dest) {
600 		/*
601 		 * Is it now for a local address on this host?
602 		 */
603 		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
604 			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
605 forwardlocal:
606 				if (args.next_hop) {
607 					struct m_tag *mtag = m_tag_get(
608 					    PACKET_TAG_IPFORWARD,
609 					    sizeof(struct sockaddr_in *),
610 					    M_NOWAIT);
611 					if (mtag == NULL) {
612 						if (ro.ro_rt)
613 							RTFREE(ro.ro_rt);
614 						goto drop;
615 					}
616 					*(struct sockaddr_in **)(mtag+1) =
617 					    args.next_hop;
618 					m_tag_prepend(m, mtag);
619 				}
620 #ifdef IPDIVERT
621 droptoours:	/* Used for DIVERT */
622 #endif
623 				/* for ip_input */
624 				m->m_flags |= M_FASTFWD_OURS;
625 
626 				/* ip still points to the real packet */
627 				ip->ip_len = htons(ip->ip_len);
628 				ip->ip_off = htons(ip->ip_off);
629 
630 				/*
631 				 * Return packet for processing by ip_input
632 				 */
633 				if (ro.ro_rt)
634 					RTFREE(ro.ro_rt);
635 				return 0;
636 			}
637 		}
638 		/*
639 		 * Redo route lookup with new destination address
640 		 */
641 		RTFREE(ro.ro_rt);
642 		bzero(&ro, sizeof(ro));
643 		dst = (struct sockaddr_in *)&ro.ro_dst;
644 		dst->sin_family = AF_INET;
645 		dst->sin_len = sizeof(*dst);
646 		dst->sin_addr.s_addr = dest;
647 		rtalloc_ign(&ro, RTF_CLONING);
648 
649 		/*
650 		 * Route there and interface still up?
651 		 */
652 		if (ro.ro_rt &&
653 		    (ro.ro_rt->rt_flags & RTF_UP) &&
654 		    (ro.ro_rt->rt_ifp->if_flags & IFF_UP)) {
655 			ia = ifatoia(ro.ro_rt->rt_ifa);
656 			ifp = ro.ro_rt->rt_ifp;
657 			if (ro.ro_rt->rt_flags & RTF_GATEWAY)
658 				dst = (struct sockaddr_in *)ro.ro_rt->rt_gateway;
659 		} else {
660 			ipstat.ips_noroute++;
661 			ipstat.ips_cantforward++;
662 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
663 			if (ro.ro_rt)
664 				RTFREE(ro.ro_rt);
665 			return 1;
666 		}
667 	}
668 
669 	/*
670 	 * Step 6: send off the packet
671 	 */
672 
673 	/*
674 	 * Check if packet fits MTU or if hardware will fragement for us
675 	 */
676 	if (ro.ro_rt->rt_rmx.rmx_mtu)
677 		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
678 	else
679 		mtu = ifp->if_mtu;
680 
681 	if (ip->ip_len <= mtu ||
682 	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
683 		/*
684 		 * Restore packet header fields to original values
685 		 */
686 		ip->ip_len = htons(ip->ip_len);
687 		ip->ip_off = htons(ip->ip_off);
688 		/*
689 		 * Send off the packet via outgoing interface
690 		 */
691 		error = (*ifp->if_output)(ifp, m,
692 				(struct sockaddr *)dst, ro.ro_rt);
693 		if (ia) {
694 			ia->ia_ifa.if_opackets++;
695 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
696 		}
697 	} else {
698 		/*
699 		 * Handle EMSGSIZE with icmp reply
700 		 * needfrag for TCP MTU discovery
701 		 */
702 		if (ip->ip_off & IP_DF) {
703 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
704 				0, ifp);
705 			ipstat.ips_cantfrag++;
706 			RTFREE(ro.ro_rt);
707 			return 1;
708 		} else {
709 			/*
710 			 * We have to fragement the packet
711 			 */
712 			m->m_pkthdr.csum_flags |= CSUM_IP;
713 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
714 					(~ifp->if_hwassist & CSUM_DELAY_IP))) {
715 				RTFREE(ro.ro_rt);
716 				goto drop;
717 			}
718 			KASSERT(m != NULL, ("null mbuf and no error"));
719 			/*
720 			 * Send off the fragments via outgoing interface
721 			 */
722 			error = 0;
723 			do {
724 				m0 = m->m_nextpkt;
725 				m->m_nextpkt = NULL;
726 
727 				error = (*ifp->if_output)(ifp, m,
728 					(struct sockaddr *)dst, ro.ro_rt);
729 				if (error)
730 					break;
731 			} while ((m = m0) != NULL);
732 			if (error) {
733 				/* Reclaim remaining fragments */
734 				for (; m; m = m0) {
735 					m0 = m->m_nextpkt;
736 					m->m_nextpkt = NULL;
737 					m_freem(m);
738 				}
739 			} else
740 				ipstat.ips_fragmented++;
741 		}
742 	}
743 
744 	if (error != 0)
745 		ipstat.ips_odropped++;
746 	else {
747 		ro.ro_rt->rt_rmx.rmx_pksent++;
748 		ipstat.ips_forward++;
749 		ipstat.ips_fastforward++;
750 	}
751 	RTFREE(ro.ro_rt);
752 	return 1;
753 drop:
754 	if (m)
755 		m_freem(m);
756 	return 1;
757 }
758