xref: /freebsd/sys/netinet/ip_fastfwd.c (revision 6b3455a7665208c366849f0b2b3bc916fb97516e)
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_pfil_hooks.h"
84 
85 #include <sys/param.h>
86 #include <sys/systm.h>
87 #include <sys/kernel.h>
88 #include <sys/malloc.h>
89 #include <sys/mbuf.h>
90 #include <sys/protosw.h>
91 #include <sys/socket.h>
92 #include <sys/sysctl.h>
93 
94 #include <net/pfil.h>
95 #include <net/if.h>
96 #include <net/if_types.h>
97 #include <net/if_var.h>
98 #include <net/if_dl.h>
99 #include <net/route.h>
100 
101 #include <netinet/in.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 
108 #include <machine/in_cksum.h>
109 
110 #include <netinet/ip_fw.h>
111 #include <netinet/ip_divert.h>
112 #include <netinet/ip_dummynet.h>
113 
114 static int ipfastforward_active = 0;
115 SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
116     &ipfastforward_active, 0, "Enable fast IP forwarding");
117 
118 static struct sockaddr_in *
119 ip_findroute(struct route *ro, in_addr_t dest, struct mbuf *m)
120 {
121 	struct sockaddr_in *dst;
122 	struct rtentry *rt;
123 
124 	/*
125 	 * Find route to destination.
126 	 */
127 	bzero(ro, sizeof(*ro));
128 	dst = (struct sockaddr_in *)&ro->ro_dst;
129 	dst->sin_family = AF_INET;
130 	dst->sin_len = sizeof(*dst);
131 	dst->sin_addr.s_addr = dest;
132 	rtalloc_ign(ro, RTF_CLONING);
133 
134 	/*
135 	 * Route there and interface still up?
136 	 */
137 	rt = ro->ro_rt;
138 	if (rt && (rt->rt_flags & RTF_UP) &&
139 	    (rt->rt_ifp->if_flags & IFF_UP) &&
140 	    (rt->rt_ifp->if_flags & IFF_RUNNING)) {
141 		if (rt->rt_flags & RTF_GATEWAY)
142 			dst = (struct sockaddr_in *)rt->rt_gateway;
143 	} else {
144 		ipstat.ips_noroute++;
145 		ipstat.ips_cantforward++;
146 		if (rt)
147 			RTFREE(rt);
148 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
149 		return NULL;
150 	}
151 	return dst;
152 }
153 
154 /*
155  * Try to forward a packet based on the destination address.
156  * This is a fast path optimized for the plain forwarding case.
157  * If the packet is handled (and consumed) here then we return 1;
158  * otherwise 0 is returned and the packet should be delivered
159  * to ip_input for full processing.
160  */
161 int
162 ip_fastforward(struct mbuf *m)
163 {
164 	struct ip *ip;
165 	struct mbuf *m0 = NULL;
166 #ifdef IPDIVERT
167 	struct ip *tip;
168 	struct mbuf *clone = NULL;
169 #endif
170 	struct route ro;
171 	struct sockaddr_in *dst = NULL;
172 	struct in_ifaddr *ia = NULL;
173 	struct ifaddr *ifa = NULL;
174 	struct ifnet *ifp;
175 	struct ip_fw_args args;
176 	in_addr_t odest, dest;
177 	u_short sum, ip_len;
178 	int error = 0;
179 	int hlen, ipfw, mtu;
180 
181 	/*
182 	 * Are we active and forwarding packets?
183 	 */
184 	if (!ipfastforward_active || !ipforwarding)
185 		return 0;
186 
187 	M_ASSERTVALID(m);
188 	M_ASSERTPKTHDR(m);
189 
190 	ro.ro_rt = NULL;
191 
192 	/*
193 	 * Step 1: check for packet drop conditions (and sanity checks)
194 	 */
195 
196 	/*
197 	 * Is entire packet big enough?
198 	 */
199 	if (m->m_pkthdr.len < sizeof(struct ip)) {
200 		ipstat.ips_tooshort++;
201 		goto drop;
202 	}
203 
204 	/*
205 	 * Is first mbuf large enough for ip header and is header present?
206 	 */
207 	if (m->m_len < sizeof (struct ip) &&
208 	   (m = m_pullup(m, sizeof (struct ip))) == 0) {
209 		ipstat.ips_toosmall++;
210 		goto drop;
211 	}
212 
213 	ip = mtod(m, struct ip *);
214 
215 	/*
216 	 * Is it IPv4?
217 	 */
218 	if (ip->ip_v != IPVERSION) {
219 		ipstat.ips_badvers++;
220 		goto drop;
221 	}
222 
223 	/*
224 	 * Is IP header length correct and is it in first mbuf?
225 	 */
226 	hlen = ip->ip_hl << 2;
227 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
228 		ipstat.ips_badlen++;
229 		goto drop;
230 	}
231 	if (hlen > m->m_len) {
232 		if ((m = m_pullup(m, hlen)) == 0) {
233 			ipstat.ips_badhlen++;
234 			goto drop;
235 		}
236 		ip = mtod(m, struct ip *);
237 	}
238 
239 	/*
240 	 * Checksum correct?
241 	 */
242 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
243 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
244 	else {
245 		if (hlen == sizeof(struct ip))
246 			sum = in_cksum_hdr(ip);
247 		else
248 			sum = in_cksum(m, hlen);
249 	}
250 	if (sum) {
251 		ipstat.ips_badsum++;
252 		goto drop;
253 	}
254 	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
255 
256 	ip_len = ntohs(ip->ip_len);
257 
258 	/*
259 	 * Is IP length longer than packet we have got?
260 	 */
261 	if (m->m_pkthdr.len < ip_len) {
262 		ipstat.ips_tooshort++;
263 		goto drop;
264 	}
265 
266 	/*
267 	 * Is packet longer than IP header tells us? If yes, truncate packet.
268 	 */
269 	if (m->m_pkthdr.len > ip_len) {
270 		if (m->m_len == m->m_pkthdr.len) {
271 			m->m_len = ip_len;
272 			m->m_pkthdr.len = ip_len;
273 		} else
274 			m_adj(m, ip_len - m->m_pkthdr.len);
275 	}
276 
277 	/*
278 	 * Is packet from or to 127/8?
279 	 */
280 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
281 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
282 		ipstat.ips_badaddr++;
283 		goto drop;
284 	}
285 
286 	/*
287 	 * Step 2: fallback conditions to normal ip_input path processing
288 	 */
289 
290 	/*
291 	 * Only IP packets without options
292 	 */
293 	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
294 		if (ip_doopts == 1)
295 			return 0;
296 		else if (ip_doopts == 2) {
297 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
298 				0, NULL);
299 			return 1;
300 		}
301 		/* else ignore IP options and continue */
302 	}
303 
304 	/*
305 	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
306 	 * no multicast, no INADDR_ANY
307 	 *
308 	 * XXX: Probably some of these checks could be direct drop
309 	 * conditions.  However it is not clear whether there are some
310 	 * hacks or obscure behaviours which make it neccessary to
311 	 * let ip_input handle it.  We play safe here and let ip_input
312 	 * deal with it until it is proven that we can directly drop it.
313 	 */
314 	if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
315 	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
316 	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
317 	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
318 	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
319 	    ip->ip_dst.s_addr == INADDR_ANY )
320 		return 0;
321 
322 	/*
323 	 * Is it for a local address on this host?
324 	 */
325 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
326 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
327 			return 0;
328 	}
329 
330 	/*
331 	 * Or is it for a local IP broadcast address on this host?
332 	 */
333 	if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
334 	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
335 			if (ifa->ifa_addr->sa_family != AF_INET)
336 				continue;
337 			ia = ifatoia(ifa);
338 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
339 				return 0;
340 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
341 			    ip->ip_dst.s_addr)
342 				return 0;
343 		}
344 	}
345 	ipstat.ips_total++;
346 
347 	/*
348 	 * Step 3: incoming packet firewall processing
349 	 */
350 
351 	/*
352 	 * Convert to host representation
353 	 */
354 	ip->ip_len = ntohs(ip->ip_len);
355 	ip->ip_off = ntohs(ip->ip_off);
356 
357 	odest = dest = ip->ip_dst.s_addr;
358 #ifdef PFIL_HOOKS
359 	/*
360 	 * Run through list of ipfilter hooks for input packets
361 	 */
362 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN) ||
363 	    m == NULL)
364 		return 1;
365 
366 	M_ASSERTVALID(m);
367 	M_ASSERTPKTHDR(m);
368 
369 	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
370 	dest = ip->ip_dst.s_addr;
371 #endif
372 
373 	/*
374 	 * Run through ipfw for input packets
375 	 */
376 	if (fw_enable && IPFW_LOADED) {
377 		bzero(&args, sizeof(args));
378 		args.m = m;
379 
380 		ipfw = ip_fw_chk_ptr(&args);
381 		m = args.m;
382 
383 		M_ASSERTVALID(m);
384 		M_ASSERTPKTHDR(m);
385 
386 		/*
387 		 * Packet denied, drop it
388 		 */
389 		if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
390 			goto drop;
391 		/*
392 		 * Send packet to the appropriate pipe
393 		 */
394 		if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
395 			ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_IN, &args);
396 			return 1;
397 		}
398 #ifdef IPDIVERT
399 		/*
400 		 * Divert packet
401 		 */
402 		if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
403 			/*
404 			 * See if this is a fragment
405 			 */
406 			if (ip->ip_off & (IP_MF | IP_OFFMASK))
407 				goto droptoours;
408 			/*
409 			 * Tee packet
410 			 */
411 			if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
412 				clone = divert_clone(m);
413 			else
414 				clone = m;
415 			if (clone == NULL)
416 				goto passin;
417 
418 			/*
419 			 * Delayed checksums are not compatible
420 			 */
421 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
422 				in_delayed_cksum(m);
423 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
424 			}
425 			/*
426 			 * Restore packet header fields to original values
427 			 */
428 			tip = mtod(m, struct ip *);
429 			tip->ip_len = htons(tip->ip_len);
430 			tip->ip_off = htons(tip->ip_off);
431 			/*
432 			 * Deliver packet to divert input routine
433 			 */
434 			divert_packet(m, 0);
435 			/*
436 			 * If this was not tee, we are done
437 			 */
438 			m = clone;
439 			if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0)
440 				return 1;
441 			/* Continue if it was tee */
442 			goto passin;
443 		}
444 #endif
445 		if (ipfw == 0 && args.next_hop != NULL) {
446 			dest = args.next_hop->sin_addr.s_addr;
447 			goto passin;
448 		}
449 		/*
450 		 * Let through or not?
451 		 */
452 		if (ipfw != 0)
453 			goto drop;
454 	}
455 passin:
456 	ip = mtod(m, struct ip *);	/* if m changed during fw processing */
457 
458 	/*
459 	 * Destination address changed?
460 	 */
461 	if (odest != dest) {
462 		/*
463 		 * Is it now for a local address on this host?
464 		 */
465 		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
466 			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
467 				goto forwardlocal;
468 		}
469 		/*
470 		 * Go on with new destination address
471 		 */
472 	}
473 
474 	/*
475 	 * Step 4: decrement TTL and look up route
476 	 */
477 
478 	/*
479 	 * Check TTL
480 	 */
481 #ifdef IPSTEALTH
482 	if (!ipstealth) {
483 #endif
484 	if (ip->ip_ttl <= IPTTLDEC) {
485 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, NULL);
486 		return 1;
487 	}
488 
489 	/*
490 	 * Decrement the TTL and incrementally change the checksum.
491 	 * Don't bother doing this with hw checksum offloading.
492 	 */
493 	ip->ip_ttl -= IPTTLDEC;
494 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
495 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
496 	else
497 		ip->ip_sum += htons(IPTTLDEC << 8);
498 #ifdef IPSTEALTH
499 	}
500 #endif
501 
502 	/*
503 	 * Find route to destination.
504 	 */
505 	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
506 		return 1;	/* icmp unreach already sent */
507 	ifp = ro.ro_rt->rt_ifp;
508 
509 	/*
510 	 * Step 5: outgoing firewall packet processing
511 	 */
512 
513 #ifdef PFIL_HOOKS
514 	/*
515 	 * Run through list of hooks for output packets.
516 	 */
517 	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT) || m == NULL) {
518 		goto consumed;
519 	}
520 
521 	M_ASSERTVALID(m);
522 	M_ASSERTPKTHDR(m);
523 
524 	ip = mtod(m, struct ip *);
525 	dest = ip->ip_dst.s_addr;
526 #endif
527 	if (fw_enable && IPFW_LOADED && !args.next_hop) {
528 		bzero(&args, sizeof(args));
529 		args.m = m;
530 		args.oif = ifp;
531 
532 		ipfw = ip_fw_chk_ptr(&args);
533 		m = args.m;
534 
535 		M_ASSERTVALID(m);
536 		M_ASSERTPKTHDR(m);
537 
538 		if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
539 			goto drop;
540 
541 		if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
542 			/*
543 			 * XXX note: if the ifp or rt entry are deleted
544 			 * while a pkt is in dummynet, we are in trouble!
545 			 */
546 			args.ro = &ro;		/* dummynet does not save it */
547 			args.dst = dst;
548 
549 			ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_OUT, &args);
550 			goto consumed;
551 		}
552 #ifdef IPDIVERT
553 		if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
554 			/*
555 			 * See if this is a fragment
556 			 */
557 			if (ip->ip_off & (IP_MF | IP_OFFMASK))
558 				goto droptoours;
559 			/*
560 			 * Tee packet
561 			 */
562 			if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
563 				clone = divert_clone(m);
564 			else
565 				clone = m;
566 			if (clone == NULL)
567 				goto passout;
568 
569 			/*
570 			 * Delayed checksums are not compatible with divert
571 			 */
572 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
573 				in_delayed_cksum(m);
574 				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
575 			}
576 			/*
577 			 * Restore packet header fields to original values
578 			 */
579 			tip = mtod(m, struct ip *);
580 			tip->ip_len = htons(tip->ip_len);
581 			tip->ip_off = htons(tip->ip_off);
582 			/*
583 			 * Deliver packet to divert input routine
584 			 */
585 			divert_packet(m, 0);
586 			/*
587 			 * If this was not tee, we are done
588 			 */
589 			m = clone;
590 			if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0) {
591 				goto consumed;
592 			}
593 			/* Continue if it was tee */
594 			goto passout;
595 		}
596 #endif
597 		if (ipfw == 0 && args.next_hop != NULL) {
598 			dest = args.next_hop->sin_addr.s_addr;
599 			goto passout;
600 		}
601 		/*
602 		 * Let through or not?
603 		 */
604 		if (ipfw != 0)
605 			goto drop;
606 	}
607 passout:
608 	ip = mtod(m, struct ip *);
609 
610 	/*
611 	 * Destination address changed?
612 	 */
613 	if (odest != dest) {
614 		/*
615 		 * Is it now for a local address on this host?
616 		 */
617 		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
618 			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
619 forwardlocal:
620 				if (args.next_hop) {
621 					struct m_tag *mtag = m_tag_get(
622 					    PACKET_TAG_IPFORWARD,
623 					    sizeof(struct sockaddr_in *),
624 					    M_NOWAIT);
625 					if (mtag == NULL) {
626 						goto drop;
627 					}
628 					*(struct sockaddr_in **)(mtag+1) =
629 					    args.next_hop;
630 					m_tag_prepend(m, mtag);
631 				}
632 #ifdef IPDIVERT
633 droptoours:	/* Used for DIVERT */
634 #endif
635 				/* for ip_input */
636 				m->m_flags |= M_FASTFWD_OURS;
637 
638 				/* ip still points to the real packet */
639 				ip->ip_len = htons(ip->ip_len);
640 				ip->ip_off = htons(ip->ip_off);
641 
642 				/*
643 				 * Return packet for processing by ip_input
644 				 */
645 				if (ro.ro_rt)
646 					RTFREE(ro.ro_rt);
647 				return 0;
648 			}
649 		}
650 		/*
651 		 * Redo route lookup with new destination address
652 		 */
653 		RTFREE(ro.ro_rt);
654 		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
655 			return 1;	/* icmp unreach already sent */
656 		ifp = ro.ro_rt->rt_ifp;
657 	}
658 
659 	/*
660 	 * Step 6: send off the packet
661 	 */
662 
663 	/*
664 	 * Check if route is dampned (when ARP is unable to resolve)
665 	 */
666 	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
667 	    ro.ro_rt->rt_rmx.rmx_expire >= time_second) {
668 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
669 		goto consumed;
670 	}
671 
672 	/*
673 	 * Check if there is enough space in the interface queue
674 	 */
675 	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
676 	    ifp->if_snd.ifq_maxlen) {
677 		ipstat.ips_odropped++;
678 		/* would send source quench here but that is depreciated */
679 		goto drop;
680 	}
681 
682 	/*
683 	 * Check if media link state of interface is not down
684 	 */
685 	if (ifp->if_link_state == LINK_STATE_DOWN) {
686 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
687 		goto consumed;
688 	}
689 
690 	/*
691 	 * Check if packet fits MTU or if hardware will fragement for us
692 	 */
693 	if (ro.ro_rt->rt_rmx.rmx_mtu)
694 		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
695 	else
696 		mtu = ifp->if_mtu;
697 
698 	if (ip->ip_len <= mtu ||
699 	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
700 		/*
701 		 * Restore packet header fields to original values
702 		 */
703 		ip->ip_len = htons(ip->ip_len);
704 		ip->ip_off = htons(ip->ip_off);
705 		/*
706 		 * Send off the packet via outgoing interface
707 		 */
708 		error = (*ifp->if_output)(ifp, m,
709 				(struct sockaddr *)dst, ro.ro_rt);
710 	} else {
711 		/*
712 		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
713 		 */
714 		if (ip->ip_off & IP_DF) {
715 			ipstat.ips_cantfrag++;
716 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
717 				0, ifp);
718 			goto consumed;
719 		} else {
720 			/*
721 			 * We have to fragement the packet
722 			 */
723 			m->m_pkthdr.csum_flags |= CSUM_IP;
724 			/*
725 			 * ip_fragment expects ip_len and ip_off in host byte
726 			 * order but returns all packets in network byte order
727 			 */
728 			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
729 					(~ifp->if_hwassist & CSUM_DELAY_IP))) {
730 				goto drop;
731 			}
732 			KASSERT(m != NULL, ("null mbuf and no error"));
733 			/*
734 			 * Send off the fragments via outgoing interface
735 			 */
736 			error = 0;
737 			do {
738 				m0 = m->m_nextpkt;
739 				m->m_nextpkt = NULL;
740 
741 				error = (*ifp->if_output)(ifp, m,
742 					(struct sockaddr *)dst, ro.ro_rt);
743 				if (error)
744 					break;
745 			} while ((m = m0) != NULL);
746 			if (error) {
747 				/* Reclaim remaining fragments */
748 				for (; m; m = m0) {
749 					m0 = m->m_nextpkt;
750 					m->m_nextpkt = NULL;
751 					m_freem(m);
752 				}
753 			} else
754 				ipstat.ips_fragmented++;
755 		}
756 	}
757 
758 	if (error != 0)
759 		ipstat.ips_odropped++;
760 	else {
761 		ro.ro_rt->rt_rmx.rmx_pksent++;
762 		ipstat.ips_forward++;
763 		ipstat.ips_fastforward++;
764 	}
765 consumed:
766 	RTFREE(ro.ro_rt);
767 	return 1;
768 drop:
769 	if (m)
770 		m_freem(m);
771 	if (ro.ro_rt)
772 		RTFREE(ro.ro_rt);
773 	return 1;
774 }
775