ip_fastfwd.c revision 129017
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: head/sys/netinet/ip_fastfwd.c 129017 2004-05-06 18:46:03Z andre $
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
116static int ipfastforward_active = 0;
117SYSCTL_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW,
118    &ipfastforward_active, 0, "Enable fast IP forwarding");
119
120static struct sockaddr_in *
121ip_findroute(struct route *ro, in_addr_t dest, struct mbuf *m)
122{
123	struct sockaddr_in *dst;
124	struct rtentry *rt;
125
126	/*
127	 * Find route to destination.
128	 */
129	bzero(ro, sizeof(*ro));
130	dst = (struct sockaddr_in *)&ro->ro_dst;
131	dst->sin_family = AF_INET;
132	dst->sin_len = sizeof(*dst);
133	dst->sin_addr.s_addr = dest;
134	rtalloc_ign(ro, RTF_CLONING);
135
136	/*
137	 * Route there and interface still up?
138	 */
139	rt = ro->ro_rt;
140	if (rt && (rt->rt_flags & RTF_UP) &&
141	    (rt->rt_ifp->if_flags & IFF_UP) &&
142	    (rt->rt_ifp->if_flags & IFF_RUNNING)) {
143		if (rt->rt_flags & RTF_GATEWAY)
144			dst = (struct sockaddr_in *)rt->rt_gateway;
145	} else {
146		ipstat.ips_noroute++;
147		ipstat.ips_cantforward++;
148		if (rt)
149			RTFREE(rt);
150		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
151		return NULL;
152	}
153	return dst;
154}
155
156/*
157 * Try to forward a packet based on the destination address.
158 * This is a fast path optimized for the plain forwarding case.
159 * If the packet is handled (and consumed) here then we return 1;
160 * otherwise 0 is returned and the packet should be delivered
161 * to ip_input for full processing.
162 */
163int
164ip_fastforward(struct mbuf *m)
165{
166	struct ip *ip;
167	struct mbuf *m0 = NULL;
168#ifdef IPDIVERT
169	struct ip *tip;
170	struct mbuf *clone = NULL;
171#endif
172	struct route ro;
173	struct sockaddr_in *dst = NULL;
174	struct in_ifaddr *ia = NULL;
175	struct ifaddr *ifa = NULL;
176	struct ifnet *ifp;
177	struct ip_fw_args args;
178	in_addr_t odest, dest;
179	u_short sum, ip_len;
180	int error = 0;
181	int hlen, ipfw, mtu;
182
183	/*
184	 * Are we active and forwarding packets?
185	 */
186	if (!ipfastforward_active || !ipforwarding)
187		return 0;
188
189	M_ASSERTVALID(m);
190	M_ASSERTPKTHDR(m);
191
192	ro.ro_rt = NULL;
193
194	/*
195	 * Step 1: check for packet drop conditions (and sanity checks)
196	 */
197
198	/*
199	 * Is entire packet big enough?
200	 */
201	if (m->m_pkthdr.len < sizeof(struct ip)) {
202		ipstat.ips_tooshort++;
203		goto drop;
204	}
205
206	/*
207	 * Is first mbuf large enough for ip header and is header present?
208	 */
209	if (m->m_len < sizeof (struct ip) &&
210	   (m = m_pullup(m, sizeof (struct ip))) == 0) {
211		ipstat.ips_toosmall++;
212		goto drop;
213	}
214
215	ip = mtod(m, struct ip *);
216
217	/*
218	 * Is it IPv4?
219	 */
220	if (ip->ip_v != IPVERSION) {
221		ipstat.ips_badvers++;
222		goto drop;
223	}
224
225	/*
226	 * Is IP header length correct and is it in first mbuf?
227	 */
228	hlen = ip->ip_hl << 2;
229	if (hlen < sizeof(struct ip)) {	/* minimum header length */
230		ipstat.ips_badlen++;
231		goto drop;
232	}
233	if (hlen > m->m_len) {
234		if ((m = m_pullup(m, hlen)) == 0) {
235			ipstat.ips_badhlen++;
236			goto drop;
237		}
238		ip = mtod(m, struct ip *);
239	}
240
241	/*
242	 * Checksum correct?
243	 */
244	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED)
245		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
246	else {
247		if (hlen == sizeof(struct ip))
248			sum = in_cksum_hdr(ip);
249		else
250			sum = in_cksum(m, hlen);
251	}
252	if (sum) {
253		ipstat.ips_badsum++;
254		goto drop;
255	}
256	m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID);
257
258	ip_len = ntohs(ip->ip_len);
259
260	/*
261	 * Is IP length longer than packet we have got?
262	 */
263	if (m->m_pkthdr.len < ip_len) {
264		ipstat.ips_tooshort++;
265		goto drop;
266	}
267
268	/*
269	 * Is packet longer than IP header tells us? If yes, truncate packet.
270	 */
271	if (m->m_pkthdr.len > ip_len) {
272		if (m->m_len == m->m_pkthdr.len) {
273			m->m_len = ip_len;
274			m->m_pkthdr.len = ip_len;
275		} else
276			m_adj(m, ip_len - m->m_pkthdr.len);
277	}
278
279	/*
280	 * Is packet from or to 127/8?
281	 */
282	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
283	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
284		ipstat.ips_badaddr++;
285		goto drop;
286	}
287
288	/*
289	 * Step 2: fallback conditions to normal ip_input path processing
290	 */
291
292	/*
293	 * Only IP packets without options
294	 */
295	if (ip->ip_hl != (sizeof(struct ip) >> 2)) {
296		if (ip_doopts == 1)
297			return 0;
298		else if (ip_doopts == 2) {
299			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB,
300				0, NULL);
301			return 1;
302		}
303		/* else ignore IP options and continue */
304	}
305
306	/*
307	 * Only unicast IP, not from loopback, no L2 or IP broadcast,
308	 * no multicast, no INADDR_ANY
309	 *
310	 * XXX: Probably some of these checks could be direct drop
311	 * conditions.  However it is not clear whether there are some
312	 * hacks or obscure behaviours which make it neccessary to
313	 * let ip_input handle it.  We play safe here and let ip_input
314	 * deal with it until it is proven that we can directly drop it.
315	 */
316	if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
317	    ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST ||
318	    ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST ||
319	    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
320	    IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
321	    ip->ip_dst.s_addr == INADDR_ANY )
322		return 0;
323
324	/*
325	 * Is it for a local address on this host?
326	 */
327	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
328		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
329			return 0;
330	}
331
332	/*
333	 * Or is it for a local IP broadcast address on this host?
334	 */
335	if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
336	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
337			if (ifa->ifa_addr->sa_family != AF_INET)
338				continue;
339			ia = ifatoia(ifa);
340			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
341				return 0;
342			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
343			    ip->ip_dst.s_addr)
344				return 0;
345		}
346	}
347	ipstat.ips_total++;
348
349	/*
350	 * Step 3: incoming packet firewall processing
351	 */
352
353	/*
354	 * Convert to host representation
355	 */
356	ip->ip_len = ntohs(ip->ip_len);
357	ip->ip_off = ntohs(ip->ip_off);
358
359	odest = dest = ip->ip_dst.s_addr;
360#ifdef PFIL_HOOKS
361	/*
362	 * Run through list of ipfilter hooks for input packets
363	 */
364	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN) ||
365	    m == NULL)
366		return 1;
367
368	M_ASSERTVALID(m);
369	M_ASSERTPKTHDR(m);
370
371	ip = mtod(m, struct ip *);	/* m may have changed by pfil hook */
372	dest = ip->ip_dst.s_addr;
373#endif
374
375	/*
376	 * Run through ipfw for input packets
377	 */
378	if (fw_enable && IPFW_LOADED) {
379		bzero(&args, sizeof(args));
380		args.m = m;
381
382		ipfw = ip_fw_chk_ptr(&args);
383		m = args.m;
384
385		M_ASSERTVALID(m);
386		M_ASSERTPKTHDR(m);
387
388		/*
389		 * Packet denied, drop it
390		 */
391		if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
392			goto drop;
393		/*
394		 * Send packet to the appropriate pipe
395		 */
396		if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
397			ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_IN, &args);
398			return 1;
399		}
400#ifdef IPDIVERT
401		/*
402		 * Divert packet
403		 */
404		if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
405			/*
406			 * See if this is a fragment
407			 */
408			if (ip->ip_off & (IP_MF | IP_OFFMASK))
409				goto droptoours;
410			/*
411			 * Tee packet
412			 */
413			if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
414				clone = divert_clone(m);
415			else
416				clone = m;
417			if (clone == NULL)
418				goto passin;
419
420			/*
421			 * Delayed checksums are not compatible
422			 */
423			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
424				in_delayed_cksum(m);
425				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
426			}
427			/*
428			 * Restore packet header fields to original values
429			 */
430			tip = mtod(m, struct ip *);
431			tip->ip_len = htons(tip->ip_len);
432			tip->ip_off = htons(tip->ip_off);
433			/*
434			 * Deliver packet to divert input routine
435			 */
436			divert_packet(m, 0);
437			/*
438			 * If this was not tee, we are done
439			 */
440			m = clone;
441			if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0)
442				return 1;
443			/* Continue if it was tee */
444			goto passin;
445		}
446#endif
447		if (ipfw == 0 && args.next_hop != NULL) {
448			dest = args.next_hop->sin_addr.s_addr;
449			goto passin;
450		}
451		/*
452		 * Let through or not?
453		 */
454		if (ipfw != 0)
455			goto drop;
456	}
457passin:
458	ip = mtod(m, struct ip *);	/* if m changed during fw processing */
459
460	/*
461	 * Destination address changed?
462	 */
463	if (odest != dest) {
464		/*
465		 * Is it now for a local address on this host?
466		 */
467		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
468			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr)
469				goto forwardlocal;
470		}
471		/*
472		 * Go on with new destination address
473		 */
474	}
475
476	/*
477	 * Step 4: decrement TTL and look up route
478	 */
479
480	/*
481	 * Check TTL
482	 */
483#ifdef IPSTEALTH
484	if (!ipstealth) {
485#endif
486	if (ip->ip_ttl <= IPTTLDEC) {
487		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, NULL);
488		return 1;
489	}
490
491	/*
492	 * Decrement the TTL and incrementally change the checksum.
493	 * Don't bother doing this with hw checksum offloading.
494	 */
495	ip->ip_ttl -= IPTTLDEC;
496	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
497		ip->ip_sum -= ~htons(IPTTLDEC << 8);
498	else
499		ip->ip_sum += htons(IPTTLDEC << 8);
500#ifdef IPSTEALTH
501	}
502#endif
503
504	/*
505	 * Find route to destination.
506	 */
507	if ((dst = ip_findroute(&ro, dest, m)) == NULL)
508		return 1;	/* icmp unreach already sent */
509	ifp = ro.ro_rt->rt_ifp;
510
511	/*
512	 * Step 5: outgoing firewall packet processing
513	 */
514
515#ifdef PFIL_HOOKS
516	/*
517	 * Run through list of hooks for output packets.
518	 */
519	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT) || m == NULL) {
520		goto consumed;
521	}
522
523	M_ASSERTVALID(m);
524	M_ASSERTPKTHDR(m);
525
526	ip = mtod(m, struct ip *);
527	dest = ip->ip_dst.s_addr;
528#endif
529	if (fw_enable && IPFW_LOADED && !args.next_hop) {
530		bzero(&args, sizeof(args));
531		args.m = m;
532		args.oif = ifp;
533
534		ipfw = ip_fw_chk_ptr(&args);
535		m = args.m;
536
537		M_ASSERTVALID(m);
538		M_ASSERTPKTHDR(m);
539
540		if ((ipfw & IP_FW_PORT_DENY_FLAG) || m == NULL)
541			goto drop;
542
543		if (DUMMYNET_LOADED && (ipfw & IP_FW_PORT_DYNT_FLAG) != 0) {
544			/*
545			 * XXX note: if the ifp or rt entry are deleted
546			 * while a pkt is in dummynet, we are in trouble!
547			 */
548			args.ro = &ro;		/* dummynet does not save it */
549			args.dst = dst;
550
551			ip_dn_io_ptr(m, ipfw & 0xffff, DN_TO_IP_OUT, &args);
552			goto consumed;
553		}
554#ifdef IPDIVERT
555		if (ipfw != 0 && (ipfw & IP_FW_PORT_DYNT_FLAG) == 0) {
556			/*
557			 * See if this is a fragment
558			 */
559			if (ip->ip_off & (IP_MF | IP_OFFMASK))
560				goto droptoours;
561			/*
562			 * Tee packet
563			 */
564			if ((ipfw & IP_FW_PORT_TEE_FLAG) != 0)
565				clone = divert_clone(m);
566			else
567				clone = m;
568			if (clone == NULL)
569				goto passout;
570
571			/*
572			 * Delayed checksums are not compatible with divert
573			 */
574			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
575				in_delayed_cksum(m);
576				m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
577			}
578			/*
579			 * Restore packet header fields to original values
580			 */
581			tip = mtod(m, struct ip *);
582			tip->ip_len = htons(tip->ip_len);
583			tip->ip_off = htons(tip->ip_off);
584			/*
585			 * Deliver packet to divert input routine
586			 */
587			divert_packet(m, 0);
588			/*
589			 * If this was not tee, we are done
590			 */
591			m = clone;
592			if ((ipfw & IP_FW_PORT_TEE_FLAG) == 0) {
593				goto consumed;
594			}
595			/* Continue if it was tee */
596			goto passout;
597		}
598#endif
599		if (ipfw == 0 && args.next_hop != NULL) {
600			dest = args.next_hop->sin_addr.s_addr;
601			goto passout;
602		}
603		/*
604		 * Let through or not?
605		 */
606		if (ipfw != 0)
607			goto drop;
608	}
609passout:
610	ip = mtod(m, struct ip *);
611
612	/*
613	 * Destination address changed?
614	 */
615	if (odest != dest) {
616		/*
617		 * Is it now for a local address on this host?
618		 */
619		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
620			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
621forwardlocal:
622				if (args.next_hop) {
623					struct m_tag *mtag = m_tag_get(
624					    PACKET_TAG_IPFORWARD,
625					    sizeof(struct sockaddr_in *),
626					    M_NOWAIT);
627					if (mtag == NULL) {
628						goto drop;
629					}
630					*(struct sockaddr_in **)(mtag+1) =
631					    args.next_hop;
632					m_tag_prepend(m, mtag);
633				}
634#ifdef IPDIVERT
635droptoours:	/* Used for DIVERT */
636#endif
637				/* for ip_input */
638				m->m_flags |= M_FASTFWD_OURS;
639
640				/* ip still points to the real packet */
641				ip->ip_len = htons(ip->ip_len);
642				ip->ip_off = htons(ip->ip_off);
643
644				/*
645				 * Return packet for processing by ip_input
646				 */
647				if (ro.ro_rt)
648					RTFREE(ro.ro_rt);
649				return 0;
650			}
651		}
652		/*
653		 * Redo route lookup with new destination address
654		 */
655		RTFREE(ro.ro_rt);
656		if ((dst = ip_findroute(&ro, dest, m)) == NULL)
657			return 1;	/* icmp unreach already sent */
658		ifp = ro.ro_rt->rt_ifp;
659	}
660
661	/*
662	 * Step 6: send off the packet
663	 */
664
665	/*
666	 * Check if route is dampned (when ARP is unable to resolve)
667	 */
668	if ((ro.ro_rt->rt_flags & RTF_REJECT) &&
669	    ro.ro_rt->rt_rmx.rmx_expire >= time_second) {
670		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
671		goto consumed;
672	}
673
674	/*
675	 * Check if there is enough space in the interface queue
676	 */
677	if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
678	    ifp->if_snd.ifq_maxlen) {
679		ipstat.ips_odropped++;
680		/* would send source quench here but that is depreciated */
681		goto drop;
682	}
683
684	/*
685	 * Check if media link state of interface is not down
686	 */
687	if (ifp->if_link_state == LINK_STATE_DOWN) {
688		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, NULL);
689		goto consumed;
690	}
691
692	/*
693	 * Check if packet fits MTU or if hardware will fragement for us
694	 */
695	if (ro.ro_rt->rt_rmx.rmx_mtu)
696		mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
697	else
698		mtu = ifp->if_mtu;
699
700	if (ip->ip_len <= mtu ||
701	    (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) {
702		/*
703		 * Restore packet header fields to original values
704		 */
705		ip->ip_len = htons(ip->ip_len);
706		ip->ip_off = htons(ip->ip_off);
707		/*
708		 * Send off the packet via outgoing interface
709		 */
710		error = (*ifp->if_output)(ifp, m,
711				(struct sockaddr *)dst, ro.ro_rt);
712	} else {
713		/*
714		 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery
715		 */
716		if (ip->ip_off & IP_DF) {
717			ipstat.ips_cantfrag++;
718			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
719				0, ifp);
720			goto consumed;
721		} else {
722			/*
723			 * We have to fragement the packet
724			 */
725			m->m_pkthdr.csum_flags |= CSUM_IP;
726			/*
727			 * ip_fragment expects ip_len and ip_off in host byte
728			 * order but returns all packets in network byte order
729			 */
730			if (ip_fragment(ip, &m, mtu, ifp->if_hwassist,
731					(~ifp->if_hwassist & CSUM_DELAY_IP))) {
732				goto drop;
733			}
734			KASSERT(m != NULL, ("null mbuf and no error"));
735			/*
736			 * Send off the fragments via outgoing interface
737			 */
738			error = 0;
739			do {
740				m0 = m->m_nextpkt;
741				m->m_nextpkt = NULL;
742
743				error = (*ifp->if_output)(ifp, m,
744					(struct sockaddr *)dst, ro.ro_rt);
745				if (error)
746					break;
747			} while ((m = m0) != NULL);
748			if (error) {
749				/* Reclaim remaining fragments */
750				for (; m; m = m0) {
751					m0 = m->m_nextpkt;
752					m->m_nextpkt = NULL;
753					m_freem(m);
754				}
755			} else
756				ipstat.ips_fragmented++;
757		}
758	}
759
760	if (error != 0)
761		ipstat.ips_odropped++;
762	else {
763		ro.ro_rt->rt_rmx.rmx_pksent++;
764		ipstat.ips_forward++;
765		ipstat.ips_fastforward++;
766	}
767consumed:
768	RTFREE(ro.ro_rt);
769	return 1;
770drop:
771	if (m)
772		m_freem(m);
773	if (ro.ro_rt)
774		RTFREE(ro.ro_rt);
775	return 1;
776}
777