xref: /dflybsd-src/sys/netinet/ip_input.c (revision 1d53fdce6974a00c68a126d57b230993ffa9a938)
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
63  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
64  */
65 
66 #define	_IP_VHL
67 
68 #include "opt_bootp.h"
69 #include "opt_ipdn.h"
70 #include "opt_ipdivert.h"
71 #include "opt_ipstealth.h"
72 #include "opt_ipsec.h"
73 #include "opt_rss.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/mbuf.h>
78 #include <sys/malloc.h>
79 #include <sys/mpipe.h>
80 #include <sys/domain.h>
81 #include <sys/protosw.h>
82 #include <sys/socket.h>
83 #include <sys/time.h>
84 #include <sys/globaldata.h>
85 #include <sys/thread.h>
86 #include <sys/kernel.h>
87 #include <sys/syslog.h>
88 #include <sys/sysctl.h>
89 #include <sys/in_cksum.h>
90 #include <sys/lock.h>
91 
92 #include <sys/mplock2.h>
93 
94 #include <machine/stdarg.h>
95 
96 #include <net/if.h>
97 #include <net/if_types.h>
98 #include <net/if_var.h>
99 #include <net/if_dl.h>
100 #include <net/pfil.h>
101 #include <net/route.h>
102 #include <net/netisr2.h>
103 
104 #include <netinet/in.h>
105 #include <netinet/in_systm.h>
106 #include <netinet/in_var.h>
107 #include <netinet/ip.h>
108 #include <netinet/in_pcb.h>
109 #include <netinet/ip_var.h>
110 #include <netinet/ip_icmp.h>
111 #include <netinet/ip_divert.h>
112 #include <netinet/ip_flow.h>
113 
114 #include <sys/thread2.h>
115 #include <sys/msgport2.h>
116 #include <net/netmsg2.h>
117 
118 #include <sys/socketvar.h>
119 
120 #include <net/ipfw/ip_fw.h>
121 #include <net/dummynet/ip_dummynet.h>
122 
123 #ifdef IPSEC
124 #include <netinet6/ipsec.h>
125 #include <netproto/key/key.h>
126 #endif
127 
128 #ifdef FAST_IPSEC
129 #include <netproto/ipsec/ipsec.h>
130 #include <netproto/ipsec/key.h>
131 #endif
132 
133 int rsvp_on = 0;
134 static int ip_rsvp_on;
135 struct socket *ip_rsvpd;
136 
137 int ipforwarding = 0;
138 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
139     &ipforwarding, 0, "Enable IP forwarding between interfaces");
140 
141 static int ipsendredirects = 1; /* XXX */
142 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
143     &ipsendredirects, 0, "Enable sending IP redirects");
144 
145 int ip_defttl = IPDEFTTL;
146 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
147     &ip_defttl, 0, "Maximum TTL on IP packets");
148 
149 static int ip_dosourceroute = 0;
150 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
151     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
152 
153 static int ip_acceptsourceroute = 0;
154 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
155     CTLFLAG_RW, &ip_acceptsourceroute, 0,
156     "Enable accepting source routed IP packets");
157 
158 static int ip_keepfaith = 0;
159 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
160     &ip_keepfaith, 0,
161     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
162 
163 static int nipq = 0;	/* total # of reass queues */
164 static int maxnipq;
165 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
166     &maxnipq, 0,
167     "Maximum number of IPv4 fragment reassembly queue entries");
168 
169 static int maxfragsperpacket;
170 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
171     &maxfragsperpacket, 0,
172     "Maximum number of IPv4 fragments allowed per packet");
173 
174 static int ip_sendsourcequench = 0;
175 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
176     &ip_sendsourcequench, 0,
177     "Enable the transmission of source quench packets");
178 
179 int ip_do_randomid = 1;
180 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
181     &ip_do_randomid, 0,
182     "Assign random ip_id values");
183 /*
184  * XXX - Setting ip_checkinterface mostly implements the receive side of
185  * the Strong ES model described in RFC 1122, but since the routing table
186  * and transmit implementation do not implement the Strong ES model,
187  * setting this to 1 results in an odd hybrid.
188  *
189  * XXX - ip_checkinterface currently must be disabled if you use ipnat
190  * to translate the destination address to another local interface.
191  *
192  * XXX - ip_checkinterface must be disabled if you add IP aliases
193  * to the loopback interface instead of the interface where the
194  * packets for those addresses are received.
195  */
196 static int ip_checkinterface = 0;
197 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
198     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
199 
200 static u_long ip_hash_count = 0;
201 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, hash_count, CTLFLAG_RD,
202     &ip_hash_count, 0, "Number of packets hashed by IP");
203 
204 #ifdef RSS_DEBUG
205 static u_long ip_rehash_count = 0;
206 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, rehash_count, CTLFLAG_RD,
207     &ip_rehash_count, 0, "Number of packets rehashed by IP");
208 
209 static u_long ip_dispatch_fast = 0;
210 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_fast_count, CTLFLAG_RD,
211     &ip_dispatch_fast, 0, "Number of packets handled on current CPU");
212 
213 static u_long ip_dispatch_slow = 0;
214 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_slow_count, CTLFLAG_RD,
215     &ip_dispatch_slow, 0, "Number of packets messaged to another CPU");
216 #endif
217 
218 static struct lwkt_token ipq_token = LWKT_TOKEN_INITIALIZER(ipq_token);
219 
220 #ifdef DIAGNOSTIC
221 static int ipprintfs = 0;
222 #endif
223 
224 extern	struct domain inetdomain;
225 extern	struct protosw inetsw[];
226 u_char	ip_protox[IPPROTO_MAX];
227 struct	in_ifaddrhead in_ifaddrheads[MAXCPU];	/* first inet address */
228 struct	in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
229 						/* inet addr hash table */
230 u_long	in_ifaddrhmask;				/* mask for hash table */
231 
232 static struct mbuf *ipforward_mtemp[MAXCPU];
233 
234 struct ip_stats ipstats_percpu[MAXCPU] __cachealign;
235 
236 static int
237 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
238 {
239 	int cpu, error = 0;
240 
241 	for (cpu = 0; cpu < ncpus; ++cpu) {
242 		if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
243 					sizeof(struct ip_stats))))
244 			break;
245 		if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
246 				       sizeof(struct ip_stats))))
247 			break;
248 	}
249 
250 	return (error);
251 }
252 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
253     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
254 
255 /* Packet reassembly stuff */
256 #define	IPREASS_NHASH_LOG2	6
257 #define	IPREASS_NHASH		(1 << IPREASS_NHASH_LOG2)
258 #define	IPREASS_HMASK		(IPREASS_NHASH - 1)
259 #define	IPREASS_HASH(x,y)						\
260     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
261 
262 static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
263 
264 #ifdef IPCTL_DEFMTU
265 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
266     &ip_mtu, 0, "Default MTU");
267 #endif
268 
269 #ifdef IPSTEALTH
270 static int ipstealth = 0;
271 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
272 #else
273 static const int ipstealth = 0;
274 #endif
275 
276 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
277 
278 struct pfil_head inet_pfil_hook;
279 
280 /*
281  * struct ip_srcrt_opt is used to store packet state while it travels
282  * through the stack.
283  *
284  * XXX Note that the code even makes assumptions on the size and
285  * alignment of fields inside struct ip_srcrt so e.g. adding some
286  * fields will break the code.  This needs to be fixed.
287  *
288  * We need to save the IP options in case a protocol wants to respond
289  * to an incoming packet over the same route if the packet got here
290  * using IP source routing.  This allows connection establishment and
291  * maintenance when the remote end is on a network that is not known
292  * to us.
293  */
294 struct ip_srcrt {
295 	struct	in_addr dst;			/* final destination */
296 	char	nop;				/* one NOP to align */
297 	char	srcopt[IPOPT_OFFSET + 1];	/* OPTVAL, OLEN and OFFSET */
298 	struct	in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
299 };
300 
301 struct ip_srcrt_opt {
302 	int		ip_nhops;
303 	struct ip_srcrt	ip_srcrt;
304 };
305 
306 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
307 static struct malloc_pipe ipq_mpipe;
308 
309 static void		save_rte(struct mbuf *, u_char *, struct in_addr);
310 static int		ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
311 static void		ip_freef(struct ipqhead *, struct ipq *);
312 static void		ip_input_handler(netmsg_t);
313 
314 /*
315  * IP initialization: fill in IP protocol switch table.
316  * All protocols not implemented in kernel go to raw IP protocol handler.
317  */
318 void
319 ip_init(void)
320 {
321 	struct protosw *pr;
322 	int i;
323 	int cpu;
324 
325 	/*
326 	 * Make sure we can handle a reasonable number of fragments but
327 	 * cap it at 4000 (XXX).
328 	 */
329 	mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
330 		    IFQ_MAXLEN, 4000, 0, NULL, NULL, NULL);
331 	for (i = 0; i < ncpus; ++i) {
332 		TAILQ_INIT(&in_ifaddrheads[i]);
333 		in_ifaddrhashtbls[i] =
334 			hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
335 	}
336 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
337 	if (pr == NULL)
338 		panic("ip_init");
339 	for (i = 0; i < IPPROTO_MAX; i++)
340 		ip_protox[i] = pr - inetsw;
341 	for (pr = inetdomain.dom_protosw;
342 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
343 		if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
344 			if (pr->pr_protocol != IPPROTO_RAW)
345 				ip_protox[pr->pr_protocol] = pr - inetsw;
346 		}
347 	}
348 
349 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
350 	inet_pfil_hook.ph_af = AF_INET;
351 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
352 		kprintf("%s: WARNING: unable to register pfil hook, "
353 			"error %d\n", __func__, i);
354 	}
355 
356 	for (i = 0; i < IPREASS_NHASH; i++)
357 		TAILQ_INIT(&ipq[i]);
358 
359 	maxnipq = nmbclusters / 32;
360 	maxfragsperpacket = 16;
361 
362 	ip_id = time_second & 0xffff;	/* time_second survives reboots */
363 
364 	for (cpu = 0; cpu < ncpus; ++cpu) {
365 		/*
366 		 * Initialize IP statistics counters for each CPU.
367 		 */
368 		bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
369 
370 		/*
371 		 * Preallocate mbuf template for forwarding
372 		 */
373 		MGETHDR(ipforward_mtemp[cpu], MB_WAIT, MT_DATA);
374 	}
375 
376 	netisr_register(NETISR_IP, ip_input_handler, ip_hashfn_in);
377 	netisr_register_hashcheck(NETISR_IP, ip_hashcheck);
378 }
379 
380 /* Do transport protocol processing. */
381 static void
382 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
383 {
384 	const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
385 
386 	/*
387 	 * Switch out to protocol's input routine.
388 	 */
389 	PR_GET_MPLOCK(pr);
390 	pr->pr_input(&m, &hlen, ip->ip_p);
391 	PR_REL_MPLOCK(pr);
392 }
393 
394 static void
395 transport_processing_handler(netmsg_t msg)
396 {
397 	struct netmsg_packet *pmsg = &msg->packet;
398 	struct ip *ip;
399 	int hlen;
400 
401 	ip = mtod(pmsg->nm_packet, struct ip *);
402 	hlen = pmsg->base.lmsg.u.ms_result;
403 
404 	transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
405 	/* msg was embedded in the mbuf, do not reply! */
406 }
407 
408 static void
409 ip_input_handler(netmsg_t msg)
410 {
411 	ip_input(msg->packet.nm_packet);
412 	/* msg was embedded in the mbuf, do not reply! */
413 }
414 
415 /*
416  * IP input routine.  Checksum and byte swap header.  If fragmented
417  * try to reassemble.  Process options.  Pass to next level.
418  */
419 void
420 ip_input(struct mbuf *m)
421 {
422 	struct ip *ip;
423 	struct in_ifaddr *ia = NULL;
424 	struct in_ifaddr_container *iac;
425 	int hlen, checkif;
426 	u_short sum;
427 	struct in_addr pkt_dst;
428 	boolean_t check_msgport = FALSE;
429 	boolean_t using_srcrt = FALSE;		/* forward (by PFIL_HOOKS) */
430 	struct in_addr odst;			/* original dst address(NAT) */
431 	struct m_tag *mtag;
432 	struct sockaddr_in *next_hop = NULL;
433 	lwkt_port_t port;
434 #ifdef FAST_IPSEC
435 	struct tdb_ident *tdbi;
436 	struct secpolicy *sp;
437 	int error;
438 #endif
439 
440 	M_ASSERTPKTHDR(m);
441 
442 	/*
443 	 * This routine is called from numerous places which may not have
444 	 * characterized the packet.
445 	 */
446 	if ((m->m_flags & M_HASH) == 0) {
447 		atomic_add_long(&ip_hash_count, 1);
448 		ip_hashfn(&m, 0, IP_MPORT_IN);
449 		if (m == NULL)
450 			return;
451 		KKASSERT(m->m_flags & M_HASH);
452 		check_msgport = TRUE;
453 	}
454 	ip = mtod(m, struct ip *);
455 
456 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
457 	    ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK)) {
458 		/*
459 		 * XXX handle multicast and fragment on CPU0 for now.
460 		 *
461 		 * This could happen for IP packets hashed by hardwares
462 		 * using RSS:
463 		 * - Hardware may not differentiate multicast IP packets
464 		 *   from unicast IP packets.
465 		 * - Hardware may not differentiate IP fragments from
466 		 *   unfragmented IP packets.
467 		 */
468 		m->m_pkthdr.hash = 0;
469 		check_msgport = TRUE;
470 	}
471 
472 	if (check_msgport &&
473 	    &curthread->td_msgport != netisr_hashport(m->m_pkthdr.hash)) {
474 		netisr_queue(NETISR_IP, m);
475 		/* Requeued to other netisr msgport; done */
476 		return;
477 	}
478 
479 	/*
480 	 * Pull out certain tags
481 	 */
482 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
483 		/* Next hop */
484 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
485 		KKASSERT(mtag != NULL);
486 		next_hop = m_tag_data(mtag);
487 	}
488 
489 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
490 		/* dummynet already filtered us */
491 		ip = mtod(m, struct ip *);
492 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
493 		goto iphack;
494 	}
495 
496 	ipstat.ips_total++;
497 
498 	/* length checks already done in ip_hashfn() */
499 	KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
500 
501 	if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
502 		ipstat.ips_badvers++;
503 		goto bad;
504 	}
505 
506 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
507 	/* length checks already done in ip_hashfn() */
508 	KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
509 	KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
510 
511 	/* 127/8 must not appear on wire - RFC1122 */
512 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
513 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
514 		if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
515 			ipstat.ips_badaddr++;
516 			goto bad;
517 		}
518 	}
519 
520 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
521 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
522 	} else {
523 		if (hlen == sizeof(struct ip))
524 			sum = in_cksum_hdr(ip);
525 		else
526 			sum = in_cksum(m, hlen);
527 	}
528 	if (sum != 0) {
529 		ipstat.ips_badsum++;
530 		goto bad;
531 	}
532 
533 #ifdef ALTQ
534 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
535 		/* packet is dropped by traffic conditioner */
536 		return;
537 	}
538 #endif
539 	/*
540 	 * Convert fields to host representation.
541 	 */
542 	ip->ip_len = ntohs(ip->ip_len);
543 	ip->ip_off = ntohs(ip->ip_off);
544 
545 	/* length checks already done in ip_hashfn() */
546 	KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
547 	KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
548 
549 	/*
550 	 * Trim mbufs if longer than the IP header would have us expect.
551 	 */
552 	if (m->m_pkthdr.len > ip->ip_len) {
553 		if (m->m_len == m->m_pkthdr.len) {
554 			m->m_len = ip->ip_len;
555 			m->m_pkthdr.len = ip->ip_len;
556 		} else {
557 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
558 		}
559 	}
560 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
561 	/*
562 	 * Bypass packet filtering for packets from a tunnel (gif).
563 	 */
564 	if (ipsec_gethist(m, NULL))
565 		goto pass;
566 #endif
567 
568 	/*
569 	 * IpHack's section.
570 	 * Right now when no processing on packet has done
571 	 * and it is still fresh out of network we do our black
572 	 * deals with it.
573 	 * - Firewall: deny/allow/divert
574 	 * - Xlate: translate packet's addr/port (NAT).
575 	 * - Pipe: pass pkt through dummynet.
576 	 * - Wrap: fake packet's addr/port <unimpl.>
577 	 * - Encapsulate: put it in another IP and send out. <unimp.>
578 	 */
579 
580 iphack:
581 	/*
582 	 * If we've been forwarded from the output side, then
583 	 * skip the firewall a second time
584 	 */
585 	if (next_hop != NULL)
586 		goto ours;
587 
588 	/* No pfil hooks */
589 	if (!pfil_has_hooks(&inet_pfil_hook)) {
590 		if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
591 			/*
592 			 * Strip dummynet tags from stranded packets
593 			 */
594 			mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
595 			KKASSERT(mtag != NULL);
596 			m_tag_delete(m, mtag);
597 			m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
598 		}
599 		goto pass;
600 	}
601 
602 	/*
603 	 * Run through list of hooks for input packets.
604 	 *
605 	 * NOTE!  If the packet is rewritten pf/ipfw/whoever must
606 	 *	  clear M_HASH.
607 	 */
608 	odst = ip->ip_dst;
609 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
610 		return;
611 	if (m == NULL)	/* consumed by filter */
612 		return;
613 	ip = mtod(m, struct ip *);
614 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
615 	using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
616 
617 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
618 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
619 		KKASSERT(mtag != NULL);
620 		next_hop = m_tag_data(mtag);
621 	}
622 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
623 		ip_dn_queue(m);
624 		return;
625 	}
626 	if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
627 		m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
628 	}
629 pass:
630 	/*
631 	 * Process options and, if not destined for us,
632 	 * ship it on.  ip_dooptions returns 1 when an
633 	 * error was detected (causing an icmp message
634 	 * to be sent and the original packet to be freed).
635 	 */
636 	if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
637 		return;
638 
639 	/* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
640 	 * matter if it is destined to another node, or whether it is
641 	 * a multicast one, RSVP wants it! and prevents it from being forwarded
642 	 * anywhere else. Also checks if the rsvp daemon is running before
643 	 * grabbing the packet.
644 	 */
645 	if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
646 		goto ours;
647 
648 	/*
649 	 * Check our list of addresses, to see if the packet is for us.
650 	 * If we don't have any addresses, assume any unicast packet
651 	 * we receive might be for us (and let the upper layers deal
652 	 * with it).
653 	 */
654 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
655 	    !(m->m_flags & (M_MCAST | M_BCAST)))
656 		goto ours;
657 
658 	/*
659 	 * Cache the destination address of the packet; this may be
660 	 * changed by use of 'ipfw fwd'.
661 	 */
662 	pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
663 
664 	/*
665 	 * Enable a consistency check between the destination address
666 	 * and the arrival interface for a unicast packet (the RFC 1122
667 	 * strong ES model) if IP forwarding is disabled and the packet
668 	 * is not locally generated and the packet is not subject to
669 	 * 'ipfw fwd'.
670 	 *
671 	 * XXX - Checking also should be disabled if the destination
672 	 * address is ipnat'ed to a different interface.
673 	 *
674 	 * XXX - Checking is incompatible with IP aliases added
675 	 * to the loopback interface instead of the interface where
676 	 * the packets are received.
677 	 */
678 	checkif = ip_checkinterface &&
679 		  !ipforwarding &&
680 		  m->m_pkthdr.rcvif != NULL &&
681 		  !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
682 		  next_hop == NULL;
683 
684 	/*
685 	 * Check for exact addresses in the hash bucket.
686 	 */
687 	LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
688 		ia = iac->ia;
689 
690 		/*
691 		 * If the address matches, verify that the packet
692 		 * arrived via the correct interface if checking is
693 		 * enabled.
694 		 */
695 		if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
696 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
697 			goto ours;
698 	}
699 	ia = NULL;
700 
701 	/*
702 	 * Check for broadcast addresses.
703 	 *
704 	 * Only accept broadcast packets that arrive via the matching
705 	 * interface.  Reception of forwarded directed broadcasts would
706 	 * be handled via ip_forward() and ether_output() with the loopback
707 	 * into the stack for SIMPLEX interfaces handled by ether_output().
708 	 */
709 	if (m->m_pkthdr.rcvif != NULL &&
710 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
711 		struct ifaddr_container *ifac;
712 
713 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
714 			      ifa_link) {
715 			struct ifaddr *ifa = ifac->ifa;
716 
717 			if (ifa->ifa_addr == NULL) /* shutdown/startup race */
718 				continue;
719 			if (ifa->ifa_addr->sa_family != AF_INET)
720 				continue;
721 			ia = ifatoia(ifa);
722 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
723 								pkt_dst.s_addr)
724 				goto ours;
725 			if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
726 				goto ours;
727 #ifdef BOOTP_COMPAT
728 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
729 				goto ours;
730 #endif
731 		}
732 	}
733 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
734 		struct in_multi *inm;
735 
736 		if (ip_mrouter != NULL) {
737 			/* XXX Multicast routing is not MPSAFE yet */
738 			get_mplock();
739 
740 			/*
741 			 * If we are acting as a multicast router, all
742 			 * incoming multicast packets are passed to the
743 			 * kernel-level multicast forwarding function.
744 			 * The packet is returned (relatively) intact; if
745 			 * ip_mforward() returns a non-zero value, the packet
746 			 * must be discarded, else it may be accepted below.
747 			 */
748 			if (ip_mforward != NULL &&
749 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
750 				rel_mplock();
751 				ipstat.ips_cantforward++;
752 				m_freem(m);
753 				return;
754 			}
755 
756 			rel_mplock();
757 
758 			/*
759 			 * The process-level routing daemon needs to receive
760 			 * all multicast IGMP packets, whether or not this
761 			 * host belongs to their destination groups.
762 			 */
763 			if (ip->ip_p == IPPROTO_IGMP)
764 				goto ours;
765 			ipstat.ips_forward++;
766 		}
767 		/*
768 		 * See if we belong to the destination multicast group on the
769 		 * arrival interface.
770 		 */
771 		inm = IN_LOOKUP_MULTI(&ip->ip_dst, m->m_pkthdr.rcvif);
772 		if (inm == NULL) {
773 			ipstat.ips_notmember++;
774 			m_freem(m);
775 			return;
776 		}
777 		goto ours;
778 	}
779 	if (ip->ip_dst.s_addr == INADDR_BROADCAST)
780 		goto ours;
781 	if (ip->ip_dst.s_addr == INADDR_ANY)
782 		goto ours;
783 
784 	/*
785 	 * FAITH(Firewall Aided Internet Translator)
786 	 */
787 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
788 		if (ip_keepfaith) {
789 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
790 				goto ours;
791 		}
792 		m_freem(m);
793 		return;
794 	}
795 
796 	/*
797 	 * Not for us; forward if possible and desirable.
798 	 */
799 	if (!ipforwarding) {
800 		ipstat.ips_cantforward++;
801 		m_freem(m);
802 	} else {
803 #ifdef IPSEC
804 		/*
805 		 * Enforce inbound IPsec SPD.
806 		 */
807 		if (ipsec4_in_reject(m, NULL)) {
808 			ipsecstat.in_polvio++;
809 			goto bad;
810 		}
811 #endif
812 #ifdef FAST_IPSEC
813 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
814 		crit_enter();
815 		if (mtag != NULL) {
816 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
817 			sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
818 		} else {
819 			sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
820 						   IP_FORWARDING, &error);
821 		}
822 		if (sp == NULL) {	/* NB: can happen if error */
823 			crit_exit();
824 			/*XXX error stat???*/
825 			DPRINTF(("ip_input: no SP for forwarding\n"));	/*XXX*/
826 			goto bad;
827 		}
828 
829 		/*
830 		 * Check security policy against packet attributes.
831 		 */
832 		error = ipsec_in_reject(sp, m);
833 		KEY_FREESP(&sp);
834 		crit_exit();
835 		if (error) {
836 			ipstat.ips_cantforward++;
837 			goto bad;
838 		}
839 #endif
840 		ip_forward(m, using_srcrt, next_hop);
841 	}
842 	return;
843 
844 ours:
845 
846 	/*
847 	 * IPSTEALTH: Process non-routing options only
848 	 * if the packet is destined for us.
849 	 */
850 	if (ipstealth &&
851 	    hlen > sizeof(struct ip) &&
852 	    ip_dooptions(m, 1, next_hop))
853 		return;
854 
855 	/* Count the packet in the ip address stats */
856 	if (ia != NULL) {
857 		IFA_STAT_INC(&ia->ia_ifa, ipackets, 1);
858 		IFA_STAT_INC(&ia->ia_ifa, ibytes, m->m_pkthdr.len);
859 	}
860 
861 	/*
862 	 * If offset or IP_MF are set, must reassemble.
863 	 * Otherwise, nothing need be done.
864 	 * (We could look in the reassembly queue to see
865 	 * if the packet was previously fragmented,
866 	 * but it's not worth the time; just let them time out.)
867 	 */
868 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
869 		/*
870 		 * Attempt reassembly; if it succeeds, proceed.  ip_reass()
871 		 * will return a different mbuf.
872 		 *
873 		 * NOTE: ip_reass() returns m with M_HASH cleared to force
874 		 *	 us to recharacterize the packet.
875 		 */
876 		m = ip_reass(m);
877 		if (m == NULL)
878 			return;
879 		ip = mtod(m, struct ip *);
880 
881 		/* Get the header length of the reassembled packet */
882 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
883 	} else {
884 		ip->ip_len -= hlen;
885 	}
886 
887 #ifdef IPSEC
888 	/*
889 	 * enforce IPsec policy checking if we are seeing last header.
890 	 * note that we do not visit this with protocols with pcb layer
891 	 * code - like udp/tcp/raw ip.
892 	 */
893 	if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
894 	    ipsec4_in_reject(m, NULL)) {
895 		ipsecstat.in_polvio++;
896 		goto bad;
897 	}
898 #endif
899 #if FAST_IPSEC
900 	/*
901 	 * enforce IPsec policy checking if we are seeing last header.
902 	 * note that we do not visit this with protocols with pcb layer
903 	 * code - like udp/tcp/raw ip.
904 	 */
905 	if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
906 		/*
907 		 * Check if the packet has already had IPsec processing
908 		 * done.  If so, then just pass it along.  This tag gets
909 		 * set during AH, ESP, etc. input handling, before the
910 		 * packet is returned to the ip input queue for delivery.
911 		 */
912 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
913 		crit_enter();
914 		if (mtag != NULL) {
915 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
916 			sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
917 		} else {
918 			sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
919 						   IP_FORWARDING, &error);
920 		}
921 		if (sp != NULL) {
922 			/*
923 			 * Check security policy against packet attributes.
924 			 */
925 			error = ipsec_in_reject(sp, m);
926 			KEY_FREESP(&sp);
927 		} else {
928 			/* XXX error stat??? */
929 			error = EINVAL;
930 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
931 			crit_exit();
932 			goto bad;
933 		}
934 		crit_exit();
935 		if (error)
936 			goto bad;
937 	}
938 #endif /* FAST_IPSEC */
939 
940 	/*
941 	 * We must forward the packet to the correct protocol thread if
942 	 * we are not already in it.
943 	 *
944 	 * NOTE: ip_len is now in host form.  ip_len is not adjusted
945 	 *	 further for protocol processing, instead we pass hlen
946 	 *	 to the protosw and let it deal with it.
947 	 */
948 	ipstat.ips_delivered++;
949 
950 	if ((m->m_flags & M_HASH) == 0) {
951 #ifdef RSS_DEBUG
952 		atomic_add_long(&ip_rehash_count, 1);
953 #endif
954 		ip->ip_len = htons(ip->ip_len + hlen);
955 		ip->ip_off = htons(ip->ip_off);
956 
957 		ip_hashfn(&m, 0, IP_MPORT_IN);
958 		if (m == NULL)
959 			return;
960 
961 		ip = mtod(m, struct ip *);
962 		ip->ip_len = ntohs(ip->ip_len) - hlen;
963 		ip->ip_off = ntohs(ip->ip_off);
964 		KKASSERT(m->m_flags & M_HASH);
965 	}
966 	port = netisr_hashport(m->m_pkthdr.hash);
967 
968 	if (port != &curthread->td_msgport) {
969 		struct netmsg_packet *pmsg;
970 
971 #ifdef RSS_DEBUG
972 		atomic_add_long(&ip_dispatch_slow, 1);
973 #endif
974 
975 		pmsg = &m->m_hdr.mh_netmsg;
976 		netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
977 			    0, transport_processing_handler);
978 		pmsg->nm_packet = m;
979 		pmsg->base.lmsg.u.ms_result = hlen;
980 		lwkt_sendmsg(port, &pmsg->base.lmsg);
981 	} else {
982 #ifdef RSS_DEBUG
983 		atomic_add_long(&ip_dispatch_fast, 1);
984 #endif
985 		transport_processing_oncpu(m, hlen, ip);
986 	}
987 	return;
988 
989 bad:
990 	m_freem(m);
991 }
992 
993 /*
994  * Take incoming datagram fragment and try to reassemble it into
995  * whole datagram.  If a chain for reassembly of this datagram already
996  * exists, then it is given as fp; otherwise have to make a chain.
997  */
998 struct mbuf *
999 ip_reass(struct mbuf *m)
1000 {
1001 	struct ip *ip = mtod(m, struct ip *);
1002 	struct mbuf *p = NULL, *q, *nq;
1003 	struct mbuf *n;
1004 	struct ipq *fp = NULL;
1005 	struct ipqhead *head;
1006 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1007 	int i, next;
1008 	u_short sum;
1009 
1010 	/* If maxnipq is 0, never accept fragments. */
1011 	if (maxnipq == 0) {
1012 		ipstat.ips_fragments++;
1013 		ipstat.ips_fragdropped++;
1014 		m_freem(m);
1015 		return NULL;
1016 	}
1017 
1018 	sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1019 	/*
1020 	 * Look for queue of fragments of this datagram.
1021 	 */
1022 	lwkt_gettoken(&ipq_token);
1023 	head = &ipq[sum];
1024 	TAILQ_FOREACH(fp, head, ipq_list) {
1025 		if (ip->ip_id == fp->ipq_id &&
1026 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1027 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1028 		    ip->ip_p == fp->ipq_p)
1029 			goto found;
1030 	}
1031 
1032 	fp = NULL;
1033 
1034 	/*
1035 	 * Enforce upper bound on number of fragmented packets
1036 	 * for which we attempt reassembly;
1037 	 * If maxnipq is -1, accept all fragments without limitation.
1038 	 */
1039 	if (nipq > maxnipq && maxnipq > 0) {
1040 		/*
1041 		 * drop something from the tail of the current queue
1042 		 * before proceeding further
1043 		 */
1044 		struct ipq *q = TAILQ_LAST(head, ipqhead);
1045 		if (q == NULL) {
1046 			/*
1047 			 * The current queue is empty,
1048 			 * so drop from one of the others.
1049 			 */
1050 			for (i = 0; i < IPREASS_NHASH; i++) {
1051 				struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
1052 				if (r) {
1053 					ipstat.ips_fragtimeout += r->ipq_nfrags;
1054 					ip_freef(&ipq[i], r);
1055 					break;
1056 				}
1057 			}
1058 		} else {
1059 			ipstat.ips_fragtimeout += q->ipq_nfrags;
1060 			ip_freef(head, q);
1061 		}
1062 	}
1063 found:
1064 	/*
1065 	 * Adjust ip_len to not reflect header,
1066 	 * convert offset of this to bytes.
1067 	 */
1068 	ip->ip_len -= hlen;
1069 	if (ip->ip_off & IP_MF) {
1070 		/*
1071 		 * Make sure that fragments have a data length
1072 		 * that's a non-zero multiple of 8 bytes.
1073 		 */
1074 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1075 			ipstat.ips_toosmall++; /* XXX */
1076 			m_freem(m);
1077 			goto done;
1078 		}
1079 		m->m_flags |= M_FRAG;
1080 	} else {
1081 		m->m_flags &= ~M_FRAG;
1082 	}
1083 	ip->ip_off <<= 3;
1084 
1085 	ipstat.ips_fragments++;
1086 	m->m_pkthdr.header = ip;
1087 
1088 	/*
1089 	 * If the hardware has not done csum over this fragment
1090 	 * then csum_data is not valid at all.
1091 	 */
1092 	if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1093 	    == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1094 		m->m_pkthdr.csum_data = 0;
1095 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1096 	}
1097 
1098 	/*
1099 	 * Presence of header sizes in mbufs
1100 	 * would confuse code below.
1101 	 */
1102 	m->m_data += hlen;
1103 	m->m_len -= hlen;
1104 
1105 	/*
1106 	 * If first fragment to arrive, create a reassembly queue.
1107 	 */
1108 	if (fp == NULL) {
1109 		if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1110 			goto dropfrag;
1111 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
1112 		nipq++;
1113 		fp->ipq_nfrags = 1;
1114 		fp->ipq_ttl = IPFRAGTTL;
1115 		fp->ipq_p = ip->ip_p;
1116 		fp->ipq_id = ip->ip_id;
1117 		fp->ipq_src = ip->ip_src;
1118 		fp->ipq_dst = ip->ip_dst;
1119 		fp->ipq_frags = m;
1120 		m->m_nextpkt = NULL;
1121 		goto inserted;
1122 	} else {
1123 		fp->ipq_nfrags++;
1124 	}
1125 
1126 #define	GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
1127 
1128 	/*
1129 	 * Find a segment which begins after this one does.
1130 	 */
1131 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1132 		if (GETIP(q)->ip_off > ip->ip_off)
1133 			break;
1134 	}
1135 
1136 	/*
1137 	 * If there is a preceding segment, it may provide some of
1138 	 * our data already.  If so, drop the data from the incoming
1139 	 * segment.  If it provides all of our data, drop us, otherwise
1140 	 * stick new segment in the proper place.
1141 	 *
1142 	 * If some of the data is dropped from the the preceding
1143 	 * segment, then it's checksum is invalidated.
1144 	 */
1145 	if (p) {
1146 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1147 		if (i > 0) {
1148 			if (i >= ip->ip_len)
1149 				goto dropfrag;
1150 			m_adj(m, i);
1151 			m->m_pkthdr.csum_flags = 0;
1152 			ip->ip_off += i;
1153 			ip->ip_len -= i;
1154 		}
1155 		m->m_nextpkt = p->m_nextpkt;
1156 		p->m_nextpkt = m;
1157 	} else {
1158 		m->m_nextpkt = fp->ipq_frags;
1159 		fp->ipq_frags = m;
1160 	}
1161 
1162 	/*
1163 	 * While we overlap succeeding segments trim them or,
1164 	 * if they are completely covered, dequeue them.
1165 	 */
1166 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1167 	     q = nq) {
1168 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1169 		if (i < GETIP(q)->ip_len) {
1170 			GETIP(q)->ip_len -= i;
1171 			GETIP(q)->ip_off += i;
1172 			m_adj(q, i);
1173 			q->m_pkthdr.csum_flags = 0;
1174 			break;
1175 		}
1176 		nq = q->m_nextpkt;
1177 		m->m_nextpkt = nq;
1178 		ipstat.ips_fragdropped++;
1179 		fp->ipq_nfrags--;
1180 		q->m_nextpkt = NULL;
1181 		m_freem(q);
1182 	}
1183 
1184 inserted:
1185 	/*
1186 	 * Check for complete reassembly and perform frag per packet
1187 	 * limiting.
1188 	 *
1189 	 * Frag limiting is performed here so that the nth frag has
1190 	 * a chance to complete the packet before we drop the packet.
1191 	 * As a result, n+1 frags are actually allowed per packet, but
1192 	 * only n will ever be stored. (n = maxfragsperpacket.)
1193 	 *
1194 	 */
1195 	next = 0;
1196 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1197 		if (GETIP(q)->ip_off != next) {
1198 			if (fp->ipq_nfrags > maxfragsperpacket) {
1199 				ipstat.ips_fragdropped += fp->ipq_nfrags;
1200 				ip_freef(head, fp);
1201 			}
1202 			goto done;
1203 		}
1204 		next += GETIP(q)->ip_len;
1205 	}
1206 	/* Make sure the last packet didn't have the IP_MF flag */
1207 	if (p->m_flags & M_FRAG) {
1208 		if (fp->ipq_nfrags > maxfragsperpacket) {
1209 			ipstat.ips_fragdropped += fp->ipq_nfrags;
1210 			ip_freef(head, fp);
1211 		}
1212 		goto done;
1213 	}
1214 
1215 	/*
1216 	 * Reassembly is complete.  Make sure the packet is a sane size.
1217 	 */
1218 	q = fp->ipq_frags;
1219 	ip = GETIP(q);
1220 	if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1221 		ipstat.ips_toolong++;
1222 		ipstat.ips_fragdropped += fp->ipq_nfrags;
1223 		ip_freef(head, fp);
1224 		goto done;
1225 	}
1226 
1227 	/*
1228 	 * Concatenate fragments.
1229 	 */
1230 	m = q;
1231 	n = m->m_next;
1232 	m->m_next = NULL;
1233 	m_cat(m, n);
1234 	nq = q->m_nextpkt;
1235 	q->m_nextpkt = NULL;
1236 	for (q = nq; q != NULL; q = nq) {
1237 		nq = q->m_nextpkt;
1238 		q->m_nextpkt = NULL;
1239 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1240 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1241 		m_cat(m, q);
1242 	}
1243 
1244 	/*
1245 	 * Clean up the 1's complement checksum.  Carry over 16 bits must
1246 	 * be added back.  This assumes no more then 65535 packet fragments
1247 	 * were reassembled.  A second carry can also occur (but not a third).
1248 	 */
1249 	m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1250 				(m->m_pkthdr.csum_data >> 16);
1251 	if (m->m_pkthdr.csum_data > 0xFFFF)
1252 		m->m_pkthdr.csum_data -= 0xFFFF;
1253 
1254 	/*
1255 	 * Create header for new ip packet by
1256 	 * modifying header of first packet;
1257 	 * dequeue and discard fragment reassembly header.
1258 	 * Make header visible.
1259 	 */
1260 	ip->ip_len = next;
1261 	ip->ip_src = fp->ipq_src;
1262 	ip->ip_dst = fp->ipq_dst;
1263 	TAILQ_REMOVE(head, fp, ipq_list);
1264 	nipq--;
1265 	mpipe_free(&ipq_mpipe, fp);
1266 	m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1267 	m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1268 	/* some debugging cruft by sklower, below, will go away soon */
1269 	if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1270 		int plen = 0;
1271 
1272 		for (n = m; n; n = n->m_next)
1273 			plen += n->m_len;
1274 		m->m_pkthdr.len = plen;
1275 	}
1276 
1277 	/*
1278 	 * Reassembly complete, return the next protocol.
1279 	 *
1280 	 * Be sure to clear M_HASH to force the packet
1281 	 * to be re-characterized.
1282 	 *
1283 	 * Clear M_FRAG, we are no longer a fragment.
1284 	 */
1285 	m->m_flags &= ~(M_HASH | M_FRAG);
1286 
1287 	ipstat.ips_reassembled++;
1288 	lwkt_reltoken(&ipq_token);
1289 	return (m);
1290 
1291 dropfrag:
1292 	ipstat.ips_fragdropped++;
1293 	if (fp != NULL)
1294 		fp->ipq_nfrags--;
1295 	m_freem(m);
1296 done:
1297 	lwkt_reltoken(&ipq_token);
1298 	return (NULL);
1299 
1300 #undef GETIP
1301 }
1302 
1303 /*
1304  * Free a fragment reassembly header and all
1305  * associated datagrams.
1306  *
1307  * Called with ipq_token held.
1308  */
1309 static void
1310 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1311 {
1312 	struct mbuf *q;
1313 
1314 	/*
1315 	 * Remove first to protect against blocking
1316 	 */
1317 	TAILQ_REMOVE(fhp, fp, ipq_list);
1318 
1319 	/*
1320 	 * Clean out at our leisure
1321 	 */
1322 	while (fp->ipq_frags) {
1323 		q = fp->ipq_frags;
1324 		fp->ipq_frags = q->m_nextpkt;
1325 		q->m_nextpkt = NULL;
1326 		m_freem(q);
1327 	}
1328 	mpipe_free(&ipq_mpipe, fp);
1329 	nipq--;
1330 }
1331 
1332 /*
1333  * IP timer processing;
1334  * if a timer expires on a reassembly
1335  * queue, discard it.
1336  */
1337 void
1338 ip_slowtimo(void)
1339 {
1340 	struct ipq *fp, *fp_temp;
1341 	struct ipqhead *head;
1342 	int i;
1343 
1344 	lwkt_gettoken(&ipq_token);
1345 	for (i = 0; i < IPREASS_NHASH; i++) {
1346 		head = &ipq[i];
1347 		TAILQ_FOREACH_MUTABLE(fp, head, ipq_list, fp_temp) {
1348 			if (--fp->ipq_ttl == 0) {
1349 				ipstat.ips_fragtimeout += fp->ipq_nfrags;
1350 				ip_freef(head, fp);
1351 			}
1352 		}
1353 	}
1354 	/*
1355 	 * If we are over the maximum number of fragments
1356 	 * (due to the limit being lowered), drain off
1357 	 * enough to get down to the new limit.
1358 	 */
1359 	if (maxnipq >= 0 && nipq > maxnipq) {
1360 		for (i = 0; i < IPREASS_NHASH; i++) {
1361 			head = &ipq[i];
1362 			while (nipq > maxnipq && !TAILQ_EMPTY(head)) {
1363 				ipstat.ips_fragdropped +=
1364 				    TAILQ_FIRST(head)->ipq_nfrags;
1365 				ip_freef(head, TAILQ_FIRST(head));
1366 			}
1367 		}
1368 	}
1369 	lwkt_reltoken(&ipq_token);
1370 	ipflow_slowtimo();
1371 }
1372 
1373 /*
1374  * Drain off all datagram fragments.
1375  */
1376 void
1377 ip_drain(void)
1378 {
1379 	struct ipqhead *head;
1380 	int i;
1381 
1382 	lwkt_gettoken(&ipq_token);
1383 	for (i = 0; i < IPREASS_NHASH; i++) {
1384 		head = &ipq[i];
1385 		while (!TAILQ_EMPTY(head)) {
1386 			ipstat.ips_fragdropped += TAILQ_FIRST(head)->ipq_nfrags;
1387 			ip_freef(head, TAILQ_FIRST(head));
1388 		}
1389 	}
1390 	lwkt_reltoken(&ipq_token);
1391 	in_rtqdrain();
1392 }
1393 
1394 /*
1395  * Do option processing on a datagram,
1396  * possibly discarding it if bad options are encountered,
1397  * or forwarding it if source-routed.
1398  * The pass argument is used when operating in the IPSTEALTH
1399  * mode to tell what options to process:
1400  * [LS]SRR (pass 0) or the others (pass 1).
1401  * The reason for as many as two passes is that when doing IPSTEALTH,
1402  * non-routing options should be processed only if the packet is for us.
1403  * Returns 1 if packet has been forwarded/freed,
1404  * 0 if the packet should be processed further.
1405  */
1406 static int
1407 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1408 {
1409 	struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1410 	struct ip *ip = mtod(m, struct ip *);
1411 	u_char *cp;
1412 	struct in_ifaddr *ia;
1413 	int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1414 	boolean_t forward = FALSE;
1415 	struct in_addr *sin, dst;
1416 	n_time ntime;
1417 
1418 	dst = ip->ip_dst;
1419 	cp = (u_char *)(ip + 1);
1420 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1421 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1422 		opt = cp[IPOPT_OPTVAL];
1423 		if (opt == IPOPT_EOL)
1424 			break;
1425 		if (opt == IPOPT_NOP)
1426 			optlen = 1;
1427 		else {
1428 			if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1429 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1430 				goto bad;
1431 			}
1432 			optlen = cp[IPOPT_OLEN];
1433 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1434 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1435 				goto bad;
1436 			}
1437 		}
1438 		switch (opt) {
1439 
1440 		default:
1441 			break;
1442 
1443 		/*
1444 		 * Source routing with record.
1445 		 * Find interface with current destination address.
1446 		 * If none on this machine then drop if strictly routed,
1447 		 * or do nothing if loosely routed.
1448 		 * Record interface address and bring up next address
1449 		 * component.  If strictly routed make sure next
1450 		 * address is on directly accessible net.
1451 		 */
1452 		case IPOPT_LSRR:
1453 		case IPOPT_SSRR:
1454 			if (ipstealth && pass > 0)
1455 				break;
1456 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1457 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1458 				goto bad;
1459 			}
1460 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1461 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1462 				goto bad;
1463 			}
1464 			ipaddr.sin_addr = ip->ip_dst;
1465 			ia = (struct in_ifaddr *)
1466 				ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1467 			if (ia == NULL) {
1468 				if (opt == IPOPT_SSRR) {
1469 					type = ICMP_UNREACH;
1470 					code = ICMP_UNREACH_SRCFAIL;
1471 					goto bad;
1472 				}
1473 				if (!ip_dosourceroute)
1474 					goto nosourcerouting;
1475 				/*
1476 				 * Loose routing, and not at next destination
1477 				 * yet; nothing to do except forward.
1478 				 */
1479 				break;
1480 			}
1481 			off--;			/* 0 origin */
1482 			if (off > optlen - (int)sizeof(struct in_addr)) {
1483 				/*
1484 				 * End of source route.  Should be for us.
1485 				 */
1486 				if (!ip_acceptsourceroute)
1487 					goto nosourcerouting;
1488 				save_rte(m, cp, ip->ip_src);
1489 				break;
1490 			}
1491 			if (ipstealth)
1492 				goto dropit;
1493 			if (!ip_dosourceroute) {
1494 				if (ipforwarding) {
1495 					char buf[sizeof "aaa.bbb.ccc.ddd"];
1496 
1497 					/*
1498 					 * Acting as a router, so generate ICMP
1499 					 */
1500 nosourcerouting:
1501 					strcpy(buf, inet_ntoa(ip->ip_dst));
1502 					log(LOG_WARNING,
1503 					    "attempted source route from %s to %s\n",
1504 					    inet_ntoa(ip->ip_src), buf);
1505 					type = ICMP_UNREACH;
1506 					code = ICMP_UNREACH_SRCFAIL;
1507 					goto bad;
1508 				} else {
1509 					/*
1510 					 * Not acting as a router,
1511 					 * so silently drop.
1512 					 */
1513 dropit:
1514 					ipstat.ips_cantforward++;
1515 					m_freem(m);
1516 					return (1);
1517 				}
1518 			}
1519 
1520 			/*
1521 			 * locate outgoing interface
1522 			 */
1523 			memcpy(&ipaddr.sin_addr, cp + off,
1524 			    sizeof ipaddr.sin_addr);
1525 
1526 			if (opt == IPOPT_SSRR) {
1527 #define	INA	struct in_ifaddr *
1528 #define	SA	struct sockaddr *
1529 				if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1530 									== NULL)
1531 					ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1532 			} else {
1533 				ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1534 			}
1535 			if (ia == NULL) {
1536 				type = ICMP_UNREACH;
1537 				code = ICMP_UNREACH_SRCFAIL;
1538 				goto bad;
1539 			}
1540 			ip->ip_dst = ipaddr.sin_addr;
1541 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1542 			    sizeof(struct in_addr));
1543 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1544 			/*
1545 			 * Let ip_intr's mcast routing check handle mcast pkts
1546 			 */
1547 			forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1548 			break;
1549 
1550 		case IPOPT_RR:
1551 			if (ipstealth && pass == 0)
1552 				break;
1553 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1554 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1555 				goto bad;
1556 			}
1557 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1558 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1559 				goto bad;
1560 			}
1561 			/*
1562 			 * If no space remains, ignore.
1563 			 */
1564 			off--;			/* 0 origin */
1565 			if (off > optlen - (int)sizeof(struct in_addr))
1566 				break;
1567 			memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1568 			    sizeof ipaddr.sin_addr);
1569 			/*
1570 			 * locate outgoing interface; if we're the destination,
1571 			 * use the incoming interface (should be same).
1572 			 */
1573 			if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1574 			    (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1575 				type = ICMP_UNREACH;
1576 				code = ICMP_UNREACH_HOST;
1577 				goto bad;
1578 			}
1579 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1580 			    sizeof(struct in_addr));
1581 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1582 			break;
1583 
1584 		case IPOPT_TS:
1585 			if (ipstealth && pass == 0)
1586 				break;
1587 			code = cp - (u_char *)ip;
1588 			if (optlen < 4 || optlen > 40) {
1589 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1590 				goto bad;
1591 			}
1592 			if ((off = cp[IPOPT_OFFSET]) < 5) {
1593 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1594 				goto bad;
1595 			}
1596 			if (off > optlen - (int)sizeof(int32_t)) {
1597 				cp[IPOPT_OFFSET + 1] += (1 << 4);
1598 				if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1599 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1600 					goto bad;
1601 				}
1602 				break;
1603 			}
1604 			off--;				/* 0 origin */
1605 			sin = (struct in_addr *)(cp + off);
1606 			switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1607 
1608 			case IPOPT_TS_TSONLY:
1609 				break;
1610 
1611 			case IPOPT_TS_TSANDADDR:
1612 				if (off + sizeof(n_time) +
1613 				    sizeof(struct in_addr) > optlen) {
1614 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1615 					goto bad;
1616 				}
1617 				ipaddr.sin_addr = dst;
1618 				ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1619 							    m->m_pkthdr.rcvif);
1620 				if (ia == NULL)
1621 					continue;
1622 				memcpy(sin, &IA_SIN(ia)->sin_addr,
1623 				    sizeof(struct in_addr));
1624 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1625 				off += sizeof(struct in_addr);
1626 				break;
1627 
1628 			case IPOPT_TS_PRESPEC:
1629 				if (off + sizeof(n_time) +
1630 				    sizeof(struct in_addr) > optlen) {
1631 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1632 					goto bad;
1633 				}
1634 				memcpy(&ipaddr.sin_addr, sin,
1635 				    sizeof(struct in_addr));
1636 				if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1637 					continue;
1638 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1639 				off += sizeof(struct in_addr);
1640 				break;
1641 
1642 			default:
1643 				code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1644 				goto bad;
1645 			}
1646 			ntime = iptime();
1647 			memcpy(cp + off, &ntime, sizeof(n_time));
1648 			cp[IPOPT_OFFSET] += sizeof(n_time);
1649 		}
1650 	}
1651 	if (forward && ipforwarding) {
1652 		ip_forward(m, TRUE, next_hop);
1653 		return (1);
1654 	}
1655 	return (0);
1656 bad:
1657 	icmp_error(m, type, code, 0, 0);
1658 	ipstat.ips_badoptions++;
1659 	return (1);
1660 }
1661 
1662 /*
1663  * Given address of next destination (final or next hop),
1664  * return internet address info of interface to be used to get there.
1665  */
1666 struct in_ifaddr *
1667 ip_rtaddr(struct in_addr dst, struct route *ro0)
1668 {
1669 	struct route sro, *ro;
1670 	struct sockaddr_in *sin;
1671 	struct in_ifaddr *ia;
1672 
1673 	if (ro0 != NULL) {
1674 		ro = ro0;
1675 	} else {
1676 		bzero(&sro, sizeof(sro));
1677 		ro = &sro;
1678 	}
1679 
1680 	sin = (struct sockaddr_in *)&ro->ro_dst;
1681 
1682 	if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1683 		if (ro->ro_rt != NULL) {
1684 			RTFREE(ro->ro_rt);
1685 			ro->ro_rt = NULL;
1686 		}
1687 		sin->sin_family = AF_INET;
1688 		sin->sin_len = sizeof *sin;
1689 		sin->sin_addr = dst;
1690 		rtalloc_ign(ro, RTF_PRCLONING);
1691 	}
1692 
1693 	if (ro->ro_rt == NULL)
1694 		return (NULL);
1695 
1696 	ia = ifatoia(ro->ro_rt->rt_ifa);
1697 
1698 	if (ro == &sro)
1699 		RTFREE(ro->ro_rt);
1700 	return ia;
1701 }
1702 
1703 /*
1704  * Save incoming source route for use in replies,
1705  * to be picked up later by ip_srcroute if the receiver is interested.
1706  */
1707 static void
1708 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1709 {
1710 	struct m_tag *mtag;
1711 	struct ip_srcrt_opt *opt;
1712 	unsigned olen;
1713 
1714 	mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1715 	if (mtag == NULL)
1716 		return;
1717 	opt = m_tag_data(mtag);
1718 
1719 	olen = option[IPOPT_OLEN];
1720 #ifdef DIAGNOSTIC
1721 	if (ipprintfs)
1722 		kprintf("save_rte: olen %d\n", olen);
1723 #endif
1724 	if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1725 		m_tag_free(mtag);
1726 		return;
1727 	}
1728 	bcopy(option, opt->ip_srcrt.srcopt, olen);
1729 	opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1730 	opt->ip_srcrt.dst = dst;
1731 	m_tag_prepend(m, mtag);
1732 }
1733 
1734 /*
1735  * Retrieve incoming source route for use in replies,
1736  * in the same form used by setsockopt.
1737  * The first hop is placed before the options, will be removed later.
1738  */
1739 struct mbuf *
1740 ip_srcroute(struct mbuf *m0)
1741 {
1742 	struct in_addr *p, *q;
1743 	struct mbuf *m;
1744 	struct m_tag *mtag;
1745 	struct ip_srcrt_opt *opt;
1746 
1747 	if (m0 == NULL)
1748 		return NULL;
1749 
1750 	mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1751 	if (mtag == NULL)
1752 		return NULL;
1753 	opt = m_tag_data(mtag);
1754 
1755 	if (opt->ip_nhops == 0)
1756 		return (NULL);
1757 	m = m_get(MB_DONTWAIT, MT_HEADER);
1758 	if (m == NULL)
1759 		return (NULL);
1760 
1761 #define	OPTSIZ	(sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1762 
1763 	/* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1764 	m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1765 		   sizeof(struct in_addr) + OPTSIZ;
1766 #ifdef DIAGNOSTIC
1767 	if (ipprintfs) {
1768 		kprintf("ip_srcroute: nhops %d mlen %d",
1769 			opt->ip_nhops, m->m_len);
1770 	}
1771 #endif
1772 
1773 	/*
1774 	 * First save first hop for return route
1775 	 */
1776 	p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1777 	*(mtod(m, struct in_addr *)) = *p--;
1778 #ifdef DIAGNOSTIC
1779 	if (ipprintfs)
1780 		kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1781 #endif
1782 
1783 	/*
1784 	 * Copy option fields and padding (nop) to mbuf.
1785 	 */
1786 	opt->ip_srcrt.nop = IPOPT_NOP;
1787 	opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1788 	memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1789 	    OPTSIZ);
1790 	q = (struct in_addr *)(mtod(m, caddr_t) +
1791 	    sizeof(struct in_addr) + OPTSIZ);
1792 #undef OPTSIZ
1793 	/*
1794 	 * Record return path as an IP source route,
1795 	 * reversing the path (pointers are now aligned).
1796 	 */
1797 	while (p >= opt->ip_srcrt.route) {
1798 #ifdef DIAGNOSTIC
1799 		if (ipprintfs)
1800 			kprintf(" %x", ntohl(q->s_addr));
1801 #endif
1802 		*q++ = *p--;
1803 	}
1804 	/*
1805 	 * Last hop goes to final destination.
1806 	 */
1807 	*q = opt->ip_srcrt.dst;
1808 	m_tag_delete(m0, mtag);
1809 #ifdef DIAGNOSTIC
1810 	if (ipprintfs)
1811 		kprintf(" %x\n", ntohl(q->s_addr));
1812 #endif
1813 	return (m);
1814 }
1815 
1816 /*
1817  * Strip out IP options.
1818  */
1819 void
1820 ip_stripoptions(struct mbuf *m)
1821 {
1822 	int datalen;
1823 	struct ip *ip = mtod(m, struct ip *);
1824 	caddr_t opts;
1825 	int optlen;
1826 
1827 	optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1828 	opts = (caddr_t)(ip + 1);
1829 	datalen = m->m_len - (sizeof(struct ip) + optlen);
1830 	bcopy(opts + optlen, opts, datalen);
1831 	m->m_len -= optlen;
1832 	if (m->m_flags & M_PKTHDR)
1833 		m->m_pkthdr.len -= optlen;
1834 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1835 }
1836 
1837 u_char inetctlerrmap[PRC_NCMDS] = {
1838 	0,		0,		0,		0,
1839 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1840 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1841 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1842 	0,		0,		0,		0,
1843 	ENOPROTOOPT,	ECONNREFUSED
1844 };
1845 
1846 /*
1847  * Forward a packet.  If some error occurs return the sender
1848  * an icmp packet.  Note we can't always generate a meaningful
1849  * icmp message because icmp doesn't have a large enough repertoire
1850  * of codes and types.
1851  *
1852  * If not forwarding, just drop the packet.  This could be confusing
1853  * if ipforwarding was zero but some routing protocol was advancing
1854  * us as a gateway to somewhere.  However, we must let the routing
1855  * protocol deal with that.
1856  *
1857  * The using_srcrt parameter indicates whether the packet is being forwarded
1858  * via a source route.
1859  */
1860 void
1861 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1862 {
1863 	struct ip *ip = mtod(m, struct ip *);
1864 	struct rtentry *rt;
1865 	struct route fwd_ro;
1866 	int error, type = 0, code = 0, destmtu = 0;
1867 	struct mbuf *mcopy, *mtemp = NULL;
1868 	n_long dest;
1869 	struct in_addr pkt_dst;
1870 
1871 	dest = INADDR_ANY;
1872 	/*
1873 	 * Cache the destination address of the packet; this may be
1874 	 * changed by use of 'ipfw fwd'.
1875 	 */
1876 	pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1877 
1878 #ifdef DIAGNOSTIC
1879 	if (ipprintfs)
1880 		kprintf("forward: src %x dst %x ttl %x\n",
1881 		       ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1882 #endif
1883 
1884 	if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1885 		ipstat.ips_cantforward++;
1886 		m_freem(m);
1887 		return;
1888 	}
1889 	if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1890 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1891 		return;
1892 	}
1893 
1894 	bzero(&fwd_ro, sizeof(fwd_ro));
1895 	ip_rtaddr(pkt_dst, &fwd_ro);
1896 	if (fwd_ro.ro_rt == NULL) {
1897 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1898 		return;
1899 	}
1900 	rt = fwd_ro.ro_rt;
1901 
1902 	if (curthread->td_type == TD_TYPE_NETISR) {
1903 		/*
1904 		 * Save the IP header and at most 8 bytes of the payload,
1905 		 * in case we need to generate an ICMP message to the src.
1906 		 */
1907 		mtemp = ipforward_mtemp[mycpuid];
1908 		KASSERT((mtemp->m_flags & M_EXT) == 0 &&
1909 		    mtemp->m_data == mtemp->m_pktdat &&
1910 		    m_tag_first(mtemp) == NULL,
1911 		    ("ip_forward invalid mtemp1"));
1912 
1913 		if (!m_dup_pkthdr(mtemp, m, MB_DONTWAIT)) {
1914 			/*
1915 			 * It's probably ok if the pkthdr dup fails (because
1916 			 * the deep copy of the tag chain failed), but for now
1917 			 * be conservative and just discard the copy since
1918 			 * code below may some day want the tags.
1919 			 */
1920 			mtemp = NULL;
1921 		} else {
1922 			mtemp->m_type = m->m_type;
1923 			mtemp->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1924 			    (int)ip->ip_len);
1925 			mtemp->m_pkthdr.len = mtemp->m_len;
1926 			m_copydata(m, 0, mtemp->m_len, mtod(mtemp, caddr_t));
1927 		}
1928 	}
1929 
1930 	if (!ipstealth)
1931 		ip->ip_ttl -= IPTTLDEC;
1932 
1933 	/*
1934 	 * If forwarding packet using same interface that it came in on,
1935 	 * perhaps should send a redirect to sender to shortcut a hop.
1936 	 * Only send redirect if source is sending directly to us,
1937 	 * and if packet was not source routed (or has any options).
1938 	 * Also, don't send redirect if forwarding using a default route
1939 	 * or a route modified by a redirect.
1940 	 */
1941 	if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1942 	    !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1943 	    satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1944 	    ipsendredirects && !using_srcrt && next_hop == NULL) {
1945 		u_long src = ntohl(ip->ip_src.s_addr);
1946 		struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1947 
1948 		if (rt_ifa != NULL &&
1949 		    (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1950 			if (rt->rt_flags & RTF_GATEWAY)
1951 				dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1952 			else
1953 				dest = pkt_dst.s_addr;
1954 			/*
1955 			 * Router requirements says to only send
1956 			 * host redirects.
1957 			 */
1958 			type = ICMP_REDIRECT;
1959 			code = ICMP_REDIRECT_HOST;
1960 #ifdef DIAGNOSTIC
1961 			if (ipprintfs)
1962 				kprintf("redirect (%d) to %x\n", code, dest);
1963 #endif
1964 		}
1965 	}
1966 
1967 	error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
1968 	if (error == 0) {
1969 		ipstat.ips_forward++;
1970 		if (type == 0) {
1971 			if (mtemp)
1972 				ipflow_create(&fwd_ro, mtemp);
1973 			goto done;
1974 		} else {
1975 			ipstat.ips_redirectsent++;
1976 		}
1977 	} else {
1978 		ipstat.ips_cantforward++;
1979 	}
1980 
1981 	if (mtemp == NULL)
1982 		goto done;
1983 
1984 	/*
1985 	 * Errors that do not require generating ICMP message
1986 	 */
1987 	switch (error) {
1988 	case ENOBUFS:
1989 		/*
1990 		 * A router should not generate ICMP_SOURCEQUENCH as
1991 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1992 		 * Source quench could be a big problem under DoS attacks,
1993 		 * or if the underlying interface is rate-limited.
1994 		 * Those who need source quench packets may re-enable them
1995 		 * via the net.inet.ip.sendsourcequench sysctl.
1996 		 */
1997 		if (!ip_sendsourcequench)
1998 			goto done;
1999 		break;
2000 
2001 	case EACCES:			/* ipfw denied packet */
2002 		goto done;
2003 	}
2004 
2005 	KASSERT((mtemp->m_flags & M_EXT) == 0 &&
2006 	    mtemp->m_data == mtemp->m_pktdat,
2007 	    ("ip_forward invalid mtemp2"));
2008 	mcopy = m_copym(mtemp, 0, mtemp->m_len, MB_DONTWAIT);
2009 	if (mcopy == NULL)
2010 		goto done;
2011 
2012 	/*
2013 	 * Send ICMP message.
2014 	 */
2015 	switch (error) {
2016 	case 0:				/* forwarded, but need redirect */
2017 		/* type, code set above */
2018 		break;
2019 
2020 	case ENETUNREACH:		/* shouldn't happen, checked above */
2021 	case EHOSTUNREACH:
2022 	case ENETDOWN:
2023 	case EHOSTDOWN:
2024 	default:
2025 		type = ICMP_UNREACH;
2026 		code = ICMP_UNREACH_HOST;
2027 		break;
2028 
2029 	case EMSGSIZE:
2030 		type = ICMP_UNREACH;
2031 		code = ICMP_UNREACH_NEEDFRAG;
2032 #ifdef IPSEC
2033 		/*
2034 		 * If the packet is routed over IPsec tunnel, tell the
2035 		 * originator the tunnel MTU.
2036 		 *	tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2037 		 * XXX quickhack!!!
2038 		 */
2039 		if (fwd_ro.ro_rt != NULL) {
2040 			struct secpolicy *sp = NULL;
2041 			int ipsecerror;
2042 			int ipsechdr;
2043 			struct route *ro;
2044 
2045 			sp = ipsec4_getpolicybyaddr(mcopy,
2046 						    IPSEC_DIR_OUTBOUND,
2047 						    IP_FORWARDING,
2048 						    &ipsecerror);
2049 
2050 			if (sp == NULL)
2051 				destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2052 			else {
2053 				/* count IPsec header size */
2054 				ipsechdr = ipsec4_hdrsiz(mcopy,
2055 							 IPSEC_DIR_OUTBOUND,
2056 							 NULL);
2057 
2058 				/*
2059 				 * find the correct route for outer IPv4
2060 				 * header, compute tunnel MTU.
2061 				 *
2062 				 */
2063 				if (sp->req != NULL && sp->req->sav != NULL &&
2064 				    sp->req->sav->sah != NULL) {
2065 					ro = &sp->req->sav->sah->sa_route;
2066 					if (ro->ro_rt != NULL &&
2067 					    ro->ro_rt->rt_ifp != NULL) {
2068 						destmtu =
2069 						    ro->ro_rt->rt_ifp->if_mtu;
2070 						destmtu -= ipsechdr;
2071 					}
2072 				}
2073 
2074 				key_freesp(sp);
2075 			}
2076 		}
2077 #elif FAST_IPSEC
2078 		/*
2079 		 * If the packet is routed over IPsec tunnel, tell the
2080 		 * originator the tunnel MTU.
2081 		 *	tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2082 		 * XXX quickhack!!!
2083 		 */
2084 		if (fwd_ro.ro_rt != NULL) {
2085 			struct secpolicy *sp = NULL;
2086 			int ipsecerror;
2087 			int ipsechdr;
2088 			struct route *ro;
2089 
2090 			sp = ipsec_getpolicybyaddr(mcopy,
2091 						   IPSEC_DIR_OUTBOUND,
2092 						   IP_FORWARDING,
2093 						   &ipsecerror);
2094 
2095 			if (sp == NULL)
2096 				destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2097 			else {
2098 				/* count IPsec header size */
2099 				ipsechdr = ipsec4_hdrsiz(mcopy,
2100 							 IPSEC_DIR_OUTBOUND,
2101 							 NULL);
2102 
2103 				/*
2104 				 * find the correct route for outer IPv4
2105 				 * header, compute tunnel MTU.
2106 				 */
2107 
2108 				if (sp->req != NULL &&
2109 				    sp->req->sav != NULL &&
2110 				    sp->req->sav->sah != NULL) {
2111 					ro = &sp->req->sav->sah->sa_route;
2112 					if (ro->ro_rt != NULL &&
2113 					    ro->ro_rt->rt_ifp != NULL) {
2114 						destmtu =
2115 						    ro->ro_rt->rt_ifp->if_mtu;
2116 						destmtu -= ipsechdr;
2117 					}
2118 				}
2119 
2120 				KEY_FREESP(&sp);
2121 			}
2122 		}
2123 #else /* !IPSEC && !FAST_IPSEC */
2124 		if (fwd_ro.ro_rt != NULL)
2125 			destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2126 #endif /*IPSEC*/
2127 		ipstat.ips_cantfrag++;
2128 		break;
2129 
2130 	case ENOBUFS:
2131 		type = ICMP_SOURCEQUENCH;
2132 		code = 0;
2133 		break;
2134 
2135 	case EACCES:			/* ipfw denied packet */
2136 		panic("ip_forward EACCES should not reach");
2137 	}
2138 	icmp_error(mcopy, type, code, dest, destmtu);
2139 done:
2140 	if (mtemp != NULL)
2141 		m_tag_delete_chain(mtemp);
2142 	if (fwd_ro.ro_rt != NULL)
2143 		RTFREE(fwd_ro.ro_rt);
2144 }
2145 
2146 void
2147 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2148 	       struct mbuf *m)
2149 {
2150 	if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2151 		struct timeval tv;
2152 
2153 		microtime(&tv);
2154 		*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2155 		    SCM_TIMESTAMP, SOL_SOCKET);
2156 		if (*mp)
2157 			mp = &(*mp)->m_next;
2158 	}
2159 	if (inp->inp_flags & INP_RECVDSTADDR) {
2160 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2161 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2162 		if (*mp)
2163 			mp = &(*mp)->m_next;
2164 	}
2165 	if (inp->inp_flags & INP_RECVTTL) {
2166 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2167 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2168 		if (*mp)
2169 			mp = &(*mp)->m_next;
2170 	}
2171 #ifdef notyet
2172 	/* XXX
2173 	 * Moving these out of udp_input() made them even more broken
2174 	 * than they already were.
2175 	 */
2176 	/* options were tossed already */
2177 	if (inp->inp_flags & INP_RECVOPTS) {
2178 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2179 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2180 		if (*mp)
2181 			mp = &(*mp)->m_next;
2182 	}
2183 	/* ip_srcroute doesn't do what we want here, need to fix */
2184 	if (inp->inp_flags & INP_RECVRETOPTS) {
2185 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2186 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2187 		if (*mp)
2188 			mp = &(*mp)->m_next;
2189 	}
2190 #endif
2191 	if (inp->inp_flags & INP_RECVIF) {
2192 		struct ifnet *ifp;
2193 		struct sdlbuf {
2194 			struct sockaddr_dl sdl;
2195 			u_char	pad[32];
2196 		} sdlbuf;
2197 		struct sockaddr_dl *sdp;
2198 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2199 
2200 		if (((ifp = m->m_pkthdr.rcvif)) &&
2201 		    ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2202 			sdp = IF_LLSOCKADDR(ifp);
2203 			/*
2204 			 * Change our mind and don't try copy.
2205 			 */
2206 			if ((sdp->sdl_family != AF_LINK) ||
2207 			    (sdp->sdl_len > sizeof(sdlbuf))) {
2208 				goto makedummy;
2209 			}
2210 			bcopy(sdp, sdl2, sdp->sdl_len);
2211 		} else {
2212 makedummy:
2213 			sdl2->sdl_len =
2214 			    offsetof(struct sockaddr_dl, sdl_data[0]);
2215 			sdl2->sdl_family = AF_LINK;
2216 			sdl2->sdl_index = 0;
2217 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2218 		}
2219 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2220 			IP_RECVIF, IPPROTO_IP);
2221 		if (*mp)
2222 			mp = &(*mp)->m_next;
2223 	}
2224 }
2225 
2226 /*
2227  * XXX these routines are called from the upper part of the kernel.
2228  *
2229  * They could also be moved to ip_mroute.c, since all the RSVP
2230  *  handling is done there already.
2231  */
2232 int
2233 ip_rsvp_init(struct socket *so)
2234 {
2235 	if (so->so_type != SOCK_RAW ||
2236 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2237 		return EOPNOTSUPP;
2238 
2239 	if (ip_rsvpd != NULL)
2240 		return EADDRINUSE;
2241 
2242 	ip_rsvpd = so;
2243 	/*
2244 	 * This may seem silly, but we need to be sure we don't over-increment
2245 	 * the RSVP counter, in case something slips up.
2246 	 */
2247 	if (!ip_rsvp_on) {
2248 		ip_rsvp_on = 1;
2249 		rsvp_on++;
2250 	}
2251 
2252 	return 0;
2253 }
2254 
2255 int
2256 ip_rsvp_done(void)
2257 {
2258 	ip_rsvpd = NULL;
2259 	/*
2260 	 * This may seem silly, but we need to be sure we don't over-decrement
2261 	 * the RSVP counter, in case something slips up.
2262 	 */
2263 	if (ip_rsvp_on) {
2264 		ip_rsvp_on = 0;
2265 		rsvp_on--;
2266 	}
2267 	return 0;
2268 }
2269 
2270 int
2271 rsvp_input(struct mbuf **mp, int *offp, int proto)
2272 {
2273 	struct mbuf *m = *mp;
2274 
2275 	*mp = NULL;
2276 
2277 	if (rsvp_input_p) { /* call the real one if loaded */
2278 		*mp = m;
2279 		rsvp_input_p(mp, offp, proto);
2280 		return(IPPROTO_DONE);
2281 	}
2282 
2283 	/* Can still get packets with rsvp_on = 0 if there is a local member
2284 	 * of the group to which the RSVP packet is addressed.  But in this
2285 	 * case we want to throw the packet away.
2286 	 */
2287 
2288 	if (!rsvp_on) {
2289 		m_freem(m);
2290 		return(IPPROTO_DONE);
2291 	}
2292 
2293 	if (ip_rsvpd != NULL) {
2294 		*mp = m;
2295 		rip_input(mp, offp, proto);
2296 		return(IPPROTO_DONE);
2297 	}
2298 	/* Drop the packet */
2299 	m_freem(m);
2300 	return(IPPROTO_DONE);
2301 }
2302