xref: /dflybsd-src/sys/netinet/ip_input.c (revision 6b7bde44184a7a4502ef3ec6f061b4268b50ea82)
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. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
67  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
68  * $DragonFly: src/sys/netinet/ip_input.c,v 1.98 2008/09/06 14:12:50 sephe Exp $
69  */
70 
71 #define	_IP_VHL
72 
73 #include "opt_bootp.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipdn.h"
76 #include "opt_ipdivert.h"
77 #include "opt_ipfilter.h"
78 #include "opt_ipstealth.h"
79 #include "opt_ipsec.h"
80 
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/mpipe.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/socket.h>
89 #include <sys/time.h>
90 #include <sys/globaldata.h>
91 #include <sys/thread.h>
92 #include <sys/kernel.h>
93 #include <sys/syslog.h>
94 #include <sys/sysctl.h>
95 #include <sys/in_cksum.h>
96 
97 #include <machine/stdarg.h>
98 
99 #include <net/if.h>
100 #include <net/if_types.h>
101 #include <net/if_var.h>
102 #include <net/if_dl.h>
103 #include <net/pfil.h>
104 #include <net/route.h>
105 #include <net/netisr.h>
106 
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/in_var.h>
110 #include <netinet/ip.h>
111 #include <netinet/in_pcb.h>
112 #include <netinet/ip_var.h>
113 #include <netinet/ip_icmp.h>
114 #ifdef IPDIVERT
115 #include <netinet/ip_divert.h>
116 #endif
117 
118 #include <sys/thread2.h>
119 #include <sys/msgport2.h>
120 #include <net/netmsg2.h>
121 
122 #include <sys/socketvar.h>
123 
124 #include <net/ipfw/ip_fw.h>
125 #include <net/dummynet/ip_dummynet.h>
126 
127 #ifdef IPSEC
128 #include <netinet6/ipsec.h>
129 #include <netproto/key/key.h>
130 #endif
131 
132 #ifdef FAST_IPSEC
133 #include <netproto/ipsec/ipsec.h>
134 #include <netproto/ipsec/key.h>
135 #endif
136 
137 int rsvp_on = 0;
138 static int ip_rsvp_on;
139 struct socket *ip_rsvpd;
140 
141 int ipforwarding = 0;
142 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
143     &ipforwarding, 0, "Enable IP forwarding between interfaces");
144 
145 static int ipsendredirects = 1; /* XXX */
146 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
147     &ipsendredirects, 0, "Enable sending IP redirects");
148 
149 int ip_defttl = IPDEFTTL;
150 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
151     &ip_defttl, 0, "Maximum TTL on IP packets");
152 
153 static int ip_dosourceroute = 0;
154 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
155     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
156 
157 static int ip_acceptsourceroute = 0;
158 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
159     CTLFLAG_RW, &ip_acceptsourceroute, 0,
160     "Enable accepting source routed IP packets");
161 
162 static int ip_keepfaith = 0;
163 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
164     &ip_keepfaith, 0,
165     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
166 
167 static int nipq = 0;	/* total # of reass queues */
168 static int maxnipq;
169 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
170     &maxnipq, 0,
171     "Maximum number of IPv4 fragment reassembly queue entries");
172 
173 static int maxfragsperpacket;
174 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
175     &maxfragsperpacket, 0,
176     "Maximum number of IPv4 fragments allowed per packet");
177 
178 static int ip_sendsourcequench = 0;
179 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
180     &ip_sendsourcequench, 0,
181     "Enable the transmission of source quench packets");
182 
183 int ip_do_randomid = 1;
184 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
185     &ip_do_randomid, 0,
186     "Assign random ip_id values");
187 /*
188  * XXX - Setting ip_checkinterface mostly implements the receive side of
189  * the Strong ES model described in RFC 1122, but since the routing table
190  * and transmit implementation do not implement the Strong ES model,
191  * setting this to 1 results in an odd hybrid.
192  *
193  * XXX - ip_checkinterface currently must be disabled if you use ipnat
194  * to translate the destination address to another local interface.
195  *
196  * XXX - ip_checkinterface must be disabled if you add IP aliases
197  * to the loopback interface instead of the interface where the
198  * packets for those addresses are received.
199  */
200 static int ip_checkinterface = 0;
201 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
202     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
203 
204 #ifdef DIAGNOSTIC
205 static int ipprintfs = 0;
206 #endif
207 
208 extern	struct domain inetdomain;
209 extern	struct protosw inetsw[];
210 u_char	ip_protox[IPPROTO_MAX];
211 struct	in_ifaddrhead in_ifaddrheads[MAXCPU];	/* first inet address */
212 struct	in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
213 						/* inet addr hash table */
214 u_long	in_ifaddrhmask;				/* mask for hash table */
215 
216 struct ip_stats ipstats_percpu[MAXCPU];
217 #ifdef SMP
218 static int
219 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
220 {
221 	int cpu, error = 0;
222 
223 	for (cpu = 0; cpu < ncpus; ++cpu) {
224 		if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
225 					sizeof(struct ip_stats))))
226 			break;
227 		if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
228 				       sizeof(struct ip_stats))))
229 			break;
230 	}
231 
232 	return (error);
233 }
234 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
235     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
236 #else
237 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
238     &ipstat, ip_stats, "IP statistics");
239 #endif
240 
241 /* Packet reassembly stuff */
242 #define	IPREASS_NHASH_LOG2	6
243 #define	IPREASS_NHASH		(1 << IPREASS_NHASH_LOG2)
244 #define	IPREASS_HMASK		(IPREASS_NHASH - 1)
245 #define	IPREASS_HASH(x,y)						\
246     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
247 
248 static struct ipq ipq[IPREASS_NHASH];
249 
250 #ifdef IPCTL_DEFMTU
251 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
252     &ip_mtu, 0, "Default MTU");
253 #endif
254 
255 #ifdef IPSTEALTH
256 static int ipstealth = 0;
257 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
258 #else
259 static const int ipstealth = 0;
260 #endif
261 
262 
263 /* Firewall hooks */
264 ip_fw_chk_t *ip_fw_chk_ptr;
265 ip_fw_dn_io_t *ip_fw_dn_io_ptr;
266 int ip_fw_loaded;
267 int fw_enable = 1;
268 int fw_one_pass = 1;
269 
270 struct pfil_head inet_pfil_hook;
271 
272 /*
273  * struct ip_srcrt_opt is used to store packet state while it travels
274  * through the stack.
275  *
276  * XXX Note that the code even makes assumptions on the size and
277  * alignment of fields inside struct ip_srcrt so e.g. adding some
278  * fields will break the code.  This needs to be fixed.
279  *
280  * We need to save the IP options in case a protocol wants to respond
281  * to an incoming packet over the same route if the packet got here
282  * using IP source routing.  This allows connection establishment and
283  * maintenance when the remote end is on a network that is not known
284  * to us.
285  */
286 struct ip_srcrt {
287 	struct	in_addr dst;			/* final destination */
288 	char	nop;				/* one NOP to align */
289 	char	srcopt[IPOPT_OFFSET + 1];	/* OPTVAL, OLEN and OFFSET */
290 	struct	in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
291 };
292 
293 struct ip_srcrt_opt {
294 	int		ip_nhops;
295 	struct ip_srcrt	ip_srcrt;
296 };
297 
298 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
299 static struct malloc_pipe ipq_mpipe;
300 
301 static void		save_rte(struct mbuf *, u_char *, struct in_addr);
302 static int		ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
303 static void		ip_freef(struct ipq *);
304 static void		ip_input_handler(struct netmsg *);
305 static struct mbuf	*ip_reass(struct mbuf *);
306 
307 /*
308  * IP initialization: fill in IP protocol switch table.
309  * All protocols not implemented in kernel go to raw IP protocol handler.
310  */
311 void
312 ip_init(void)
313 {
314 	struct protosw *pr;
315 	int i;
316 #ifdef SMP
317 	int cpu;
318 #endif
319 
320 	/*
321 	 * Make sure we can handle a reasonable number of fragments but
322 	 * cap it at 4000 (XXX).
323 	 */
324 	mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
325 		    IFQ_MAXLEN, 4000, 0, NULL);
326 	for (i = 0; i < ncpus; ++i) {
327 		TAILQ_INIT(&in_ifaddrheads[i]);
328 		in_ifaddrhashtbls[i] =
329 			hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
330 	}
331 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
332 	if (pr == NULL)
333 		panic("ip_init");
334 	for (i = 0; i < IPPROTO_MAX; i++)
335 		ip_protox[i] = pr - inetsw;
336 	for (pr = inetdomain.dom_protosw;
337 	     pr < inetdomain.dom_protoswNPROTOSW; pr++)
338 		if (pr->pr_domain->dom_family == PF_INET &&
339 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW)
340 			ip_protox[pr->pr_protocol] = pr - inetsw;
341 
342 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
343 	inet_pfil_hook.ph_af = AF_INET;
344 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
345 		kprintf("%s: WARNING: unable to register pfil hook, "
346 			"error %d\n", __func__, i);
347 	}
348 
349 	for (i = 0; i < IPREASS_NHASH; i++)
350 		ipq[i].next = ipq[i].prev = &ipq[i];
351 
352 	maxnipq = nmbclusters / 32;
353 	maxfragsperpacket = 16;
354 
355 	ip_id = time_second & 0xffff;
356 
357 	/*
358 	 * Initialize IP statistics counters for each CPU.
359 	 *
360 	 */
361 #ifdef SMP
362 	for (cpu = 0; cpu < ncpus; ++cpu) {
363 		bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
364 	}
365 #else
366 	bzero(&ipstat, sizeof(struct ip_stats));
367 #endif
368 
369 	netisr_register(NETISR_IP, ip_mport_in, ip_input_handler);
370 }
371 
372 /*
373  * XXX watch out this one. It is perhaps used as a cache for
374  * the most recently used route ? it is cleared in in_addroute()
375  * when a new route is successfully created.
376  */
377 struct route ipforward_rt[MAXCPU];
378 
379 /* Do transport protocol processing. */
380 static void
381 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
382 {
383 	/*
384 	 * Switch out to protocol's input routine.
385 	 */
386 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen, ip->ip_p);
387 }
388 
389 static void
390 transport_processing_handler(netmsg_t netmsg)
391 {
392 	struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
393 	struct ip *ip;
394 	int hlen;
395 
396 	ip = mtod(pmsg->nm_packet, struct ip *);
397 	hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
398 
399 	transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
400 	/* netmsg was embedded in the mbuf, do not reply! */
401 }
402 
403 static void
404 ip_input_handler(struct netmsg *msg0)
405 {
406 	struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
407 
408 	ip_input(m);
409 	/* msg0 was embedded in the mbuf, do not reply! */
410 }
411 
412 /*
413  * IP input routine.  Checksum and byte swap header.  If fragmented
414  * try to reassemble.  Process options.  Pass to next level.
415  */
416 void
417 ip_input(struct mbuf *m)
418 {
419 	struct ip *ip;
420 	struct in_ifaddr *ia = NULL;
421 	struct in_ifaddr_container *iac;
422 	int i, hlen, checkif;
423 	u_short sum;
424 	struct in_addr pkt_dst;
425 	boolean_t using_srcrt = FALSE;		/* forward (by PFIL_HOOKS) */
426 	boolean_t needredispatch = FALSE;
427 	struct in_addr odst;			/* original dst address(NAT) */
428 	struct m_tag *mtag;
429 	struct sockaddr_in *next_hop = NULL;
430 #ifdef FAST_IPSEC
431 	struct tdb_ident *tdbi;
432 	struct secpolicy *sp;
433 	int error;
434 #endif
435 
436 	M_ASSERTPKTHDR(m);
437 
438 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
439 		/* Next hop */
440 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
441 		KKASSERT(mtag != NULL);
442 		next_hop = m_tag_data(mtag);
443 	}
444 
445 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
446 		/* dummynet already filtered us */
447 		ip = mtod(m, struct ip *);
448 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
449 		goto iphack;
450 	}
451 
452 	ipstat.ips_total++;
453 
454 	/* length checks already done in ip_demux() */
455 	KASSERT(m->m_len >= sizeof(ip), ("IP header not in one mbuf"));
456 
457 	ip = mtod(m, struct ip *);
458 
459 	if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
460 		ipstat.ips_badvers++;
461 		goto bad;
462 	}
463 
464 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
465 	/* length checks already done in ip_demux() */
466 	KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
467 	KASSERT(m->m_len >= hlen, ("packet shorter than IP header length"));
468 
469 	/* 127/8 must not appear on wire - RFC1122 */
470 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
471 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
472 		if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
473 			ipstat.ips_badaddr++;
474 			goto bad;
475 		}
476 	}
477 
478 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
479 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
480 	} else {
481 		if (hlen == sizeof(struct ip)) {
482 			sum = in_cksum_hdr(ip);
483 		} else {
484 			sum = in_cksum(m, hlen);
485 		}
486 	}
487 	if (sum != 0) {
488 		ipstat.ips_badsum++;
489 		goto bad;
490 	}
491 
492 #ifdef ALTQ
493 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
494 		/* packet is dropped by traffic conditioner */
495 		return;
496 	}
497 #endif
498 	/*
499 	 * Convert fields to host representation.
500 	 */
501 	ip->ip_len = ntohs(ip->ip_len);
502 	if (ip->ip_len < hlen) {
503 		ipstat.ips_badlen++;
504 		goto bad;
505 	}
506 	ip->ip_off = ntohs(ip->ip_off);
507 
508 	/*
509 	 * Check that the amount of data in the buffers
510 	 * is as at least much as the IP header would have us expect.
511 	 * Trim mbufs if longer than we expect.
512 	 * Drop packet if shorter than we expect.
513 	 */
514 	if (m->m_pkthdr.len < ip->ip_len) {
515 		ipstat.ips_tooshort++;
516 		goto bad;
517 	}
518 	if (m->m_pkthdr.len > ip->ip_len) {
519 		if (m->m_len == m->m_pkthdr.len) {
520 			m->m_len = ip->ip_len;
521 			m->m_pkthdr.len = ip->ip_len;
522 		} else
523 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
524 	}
525 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
526 	/*
527 	 * Bypass packet filtering for packets from a tunnel (gif).
528 	 */
529 	if (ipsec_gethist(m, NULL))
530 		goto pass;
531 #endif
532 
533 	/*
534 	 * IpHack's section.
535 	 * Right now when no processing on packet has done
536 	 * and it is still fresh out of network we do our black
537 	 * deals with it.
538 	 * - Firewall: deny/allow/divert
539 	 * - Xlate: translate packet's addr/port (NAT).
540 	 * - Pipe: pass pkt through dummynet.
541 	 * - Wrap: fake packet's addr/port <unimpl.>
542 	 * - Encapsulate: put it in another IP and send out. <unimp.>
543 	 */
544 
545 iphack:
546 
547 	/*
548 	 * Run through list of hooks for input packets.
549 	 *
550 	 * NB: Beware of the destination address changing (e.g.
551 	 *     by NAT rewriting). When this happens, tell
552 	 *     ip_forward to do the right thing.
553 	 */
554 	if (pfil_has_hooks(&inet_pfil_hook)) {
555 		odst = ip->ip_dst;
556 		if (pfil_run_hooks(&inet_pfil_hook, &m,
557 		    m->m_pkthdr.rcvif, PFIL_IN)) {
558 			return;
559 		}
560 		if (m == NULL)			/* consumed by filter */
561 			return;
562 		ip = mtod(m, struct ip *);
563 		using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
564 	}
565 
566 	if (fw_enable && IPFW_LOADED) {
567 		struct ip_fw_args args;
568 
569 		/*
570 		 * If we've been forwarded from the output side, then
571 		 * skip the firewall a second time
572 		 */
573 		if (next_hop != NULL)
574 			goto ours;
575 
576 		if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
577 			/* Extract info from dummynet tag */
578 			mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
579 			KKASSERT(mtag != NULL);
580 			args.rule =
581 			((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
582 			KKASSERT(args.rule != NULL);
583 
584 			m_tag_delete(m, mtag);
585 			m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
586 		} else {
587 			args.rule = NULL;
588 		}
589 
590 		args.eh = NULL;
591 		args.oif = NULL;
592 		args.m = m;
593 		i = ip_fw_chk_ptr(&args);
594 		m = args.m;
595 
596 		if ((i & IP_FW_PORT_DENY_FLAG) || m == NULL) {	/* drop */
597 			if (m != NULL)
598 				m_freem(m);
599 			return;
600 		}
601 		ip = mtod(m, struct ip *);	/* just in case m changed */
602 
603 		if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
604 			mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
605 			KKASSERT(mtag != NULL);
606 			next_hop = m_tag_data(mtag);
607 		}
608 
609 		if (i == 0 && next_hop == NULL)	/* common case */
610 			goto pass;
611 		if (i & IP_FW_PORT_DYNT_FLAG) {
612 			/* Send packet to the appropriate pipe */
613 			ip_fw_dn_io_ptr(m, i&0xffff, DN_TO_IP_IN, &args);
614 			return;
615 		}
616 #ifdef IPDIVERT
617 		if (i != 0 && !(i & IP_FW_PORT_DYNT_FLAG)) {
618 			struct mbuf *clone = NULL;
619 			int tee = 0;
620 
621 			/* Divert or 'tee'? */
622 			if (i & IP_FW_PORT_TEE_FLAG)
623 				tee = 1;
624 
625 			if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
626 				const struct divert_info *divinfo;
627 				u_short frag_off;
628 
629 				/*
630 				 * Only trust divert info in the fragment
631 				 * at offset 0.
632 				 */
633 				frag_off = ip->ip_off << 3;
634 				if (frag_off != 0) {
635 					mtag = m_tag_find(m,
636 					PACKET_TAG_IPFW_DIVERT, NULL);
637 					m_tag_delete(m, mtag);
638 				}
639 
640 				/*
641 				 * Attempt reassembly; if it succeeds, proceed.
642 				 * ip_reass() will return a different mbuf.
643 				 */
644 				m = ip_reass(m);
645 				if (m == NULL)
646 					return;
647 				ip = mtod(m, struct ip *);
648 
649 				needredispatch = TRUE;
650 
651 				/*
652 				 * Get the header length of the reassembled
653 				 * packet
654 				 */
655 				hlen = IP_VHL_HL(ip->ip_vhl) << 2;
656 
657 				/*
658 				 * Restore original checksum before diverting
659 				 * packet
660 				 */
661 				ip->ip_len += hlen;
662 				ip->ip_len = htons(ip->ip_len);
663 				ip->ip_off = htons(ip->ip_off);
664 				ip->ip_sum = 0;
665 				if (hlen == sizeof(struct ip))
666 					ip->ip_sum = in_cksum_hdr(ip);
667 				else
668 					ip->ip_sum = in_cksum(m, hlen);
669 				ip->ip_off = ntohs(ip->ip_off);
670 				ip->ip_len = ntohs(ip->ip_len);
671 
672 				/*
673 				 * Only use the saved divert info
674 				 */
675 				mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT,
676 						  NULL);
677 				if (mtag == NULL) {
678 					/* Wrongly configured ipfw */
679 					kprintf("ip_input no divert info\n");
680 					m_freem(m);
681 					return;
682 				}
683 				divinfo = m_tag_data(mtag);
684 				tee = divinfo->tee;
685 			}
686 
687 			/*
688 			 * Divert or tee packet to the divert protocol if
689 			 * required.
690 			 */
691 
692 			/* Clone packet if we're doing a 'tee' */
693 			if (tee)
694 				clone = m_dup(m, MB_DONTWAIT);
695 
696 			/*
697 			 * Restore packet header fields to original
698 			 * values
699 			 */
700 			ip->ip_len = htons(ip->ip_len);
701 			ip->ip_off = htons(ip->ip_off);
702 
703 			/* Deliver packet to divert input routine */
704 			divert_packet(m, 1);
705 			ipstat.ips_delivered++;
706 
707 			/* If 'tee', continue with original packet */
708 			if (clone == NULL)
709 				return;
710 			m = clone;
711 			ip = mtod(m, struct ip *);
712 
713 			/*
714 			 * Complete processing of the packet.
715 			 *
716 			 * XXX
717 			 * Better safe than sorry, remove the DIVERT
718 			 * tag.
719 			 */
720 			mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT,
721 					  NULL);
722 			KKASSERT(mtag != NULL);
723 			m_tag_delete(m, mtag);
724 
725 			goto pass;
726 		}
727 #endif
728 		if (i == 0 && next_hop != NULL)
729 			goto pass;
730 		/*
731 		 * if we get here, the packet must be dropped
732 		 */
733 		m_freem(m);
734 		return;
735 	}
736 pass:
737 
738 	/*
739 	 * Process options and, if not destined for us,
740 	 * ship it on.  ip_dooptions returns 1 when an
741 	 * error was detected (causing an icmp message
742 	 * to be sent and the original packet to be freed).
743 	 */
744 	if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
745 		return;
746 
747 	/* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
748 	 * matter if it is destined to another node, or whether it is
749 	 * a multicast one, RSVP wants it! and prevents it from being forwarded
750 	 * anywhere else. Also checks if the rsvp daemon is running before
751 	 * grabbing the packet.
752 	 */
753 	if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
754 		goto ours;
755 
756 	/*
757 	 * Check our list of addresses, to see if the packet is for us.
758 	 * If we don't have any addresses, assume any unicast packet
759 	 * we receive might be for us (and let the upper layers deal
760 	 * with it).
761 	 */
762 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
763 	    !(m->m_flags & (M_MCAST | M_BCAST)))
764 		goto ours;
765 
766 	/*
767 	 * Cache the destination address of the packet; this may be
768 	 * changed by use of 'ipfw fwd'.
769 	 */
770 	pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
771 
772 	/*
773 	 * Enable a consistency check between the destination address
774 	 * and the arrival interface for a unicast packet (the RFC 1122
775 	 * strong ES model) if IP forwarding is disabled and the packet
776 	 * is not locally generated and the packet is not subject to
777 	 * 'ipfw fwd'.
778 	 *
779 	 * XXX - Checking also should be disabled if the destination
780 	 * address is ipnat'ed to a different interface.
781 	 *
782 	 * XXX - Checking is incompatible with IP aliases added
783 	 * to the loopback interface instead of the interface where
784 	 * the packets are received.
785 	 */
786 	checkif = ip_checkinterface &&
787 		  !ipforwarding &&
788 		  m->m_pkthdr.rcvif != NULL &&
789 		  !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
790 		  next_hop == NULL;
791 
792 	/*
793 	 * Check for exact addresses in the hash bucket.
794 	 */
795 	LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
796 		ia = iac->ia;
797 
798 		/*
799 		 * If the address matches, verify that the packet
800 		 * arrived via the correct interface if checking is
801 		 * enabled.
802 		 */
803 		if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
804 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
805 			goto ours;
806 	}
807 	ia = NULL;
808 
809 	/*
810 	 * Check for broadcast addresses.
811 	 *
812 	 * Only accept broadcast packets that arrive via the matching
813 	 * interface.  Reception of forwarded directed broadcasts would
814 	 * be handled via ip_forward() and ether_output() with the loopback
815 	 * into the stack for SIMPLEX interfaces handled by ether_output().
816 	 */
817 	if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
818 		struct ifaddr_container *ifac;
819 
820 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
821 			      ifa_link) {
822 			struct ifaddr *ifa = ifac->ifa;
823 
824 			if (ifa->ifa_addr == NULL) /* shutdown/startup race */
825 				continue;
826 			if (ifa->ifa_addr->sa_family != AF_INET)
827 				continue;
828 			ia = ifatoia(ifa);
829 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
830 								pkt_dst.s_addr)
831 				goto ours;
832 			if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
833 				goto ours;
834 #ifdef BOOTP_COMPAT
835 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
836 				goto ours;
837 #endif
838 		}
839 	}
840 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
841 		struct in_multi *inm;
842 
843 		if (ip_mrouter != NULL) {
844 			/*
845 			 * If we are acting as a multicast router, all
846 			 * incoming multicast packets are passed to the
847 			 * kernel-level multicast forwarding function.
848 			 * The packet is returned (relatively) intact; if
849 			 * ip_mforward() returns a non-zero value, the packet
850 			 * must be discarded, else it may be accepted below.
851 			 */
852 			if (ip_mforward != NULL &&
853 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
854 				ipstat.ips_cantforward++;
855 				m_freem(m);
856 				return;
857 			}
858 
859 			/*
860 			 * The process-level routing daemon needs to receive
861 			 * all multicast IGMP packets, whether or not this
862 			 * host belongs to their destination groups.
863 			 */
864 			if (ip->ip_p == IPPROTO_IGMP)
865 				goto ours;
866 			ipstat.ips_forward++;
867 		}
868 		/*
869 		 * See if we belong to the destination multicast group on the
870 		 * arrival interface.
871 		 */
872 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
873 		if (inm == NULL) {
874 			ipstat.ips_notmember++;
875 			m_freem(m);
876 			return;
877 		}
878 		goto ours;
879 	}
880 	if (ip->ip_dst.s_addr == INADDR_BROADCAST)
881 		goto ours;
882 	if (ip->ip_dst.s_addr == INADDR_ANY)
883 		goto ours;
884 
885 	/*
886 	 * FAITH(Firewall Aided Internet Translator)
887 	 */
888 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
889 		if (ip_keepfaith) {
890 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
891 				goto ours;
892 		}
893 		m_freem(m);
894 		return;
895 	}
896 
897 	/*
898 	 * Not for us; forward if possible and desirable.
899 	 */
900 	if (!ipforwarding) {
901 		ipstat.ips_cantforward++;
902 		m_freem(m);
903 	} else {
904 #ifdef IPSEC
905 		/*
906 		 * Enforce inbound IPsec SPD.
907 		 */
908 		if (ipsec4_in_reject(m, NULL)) {
909 			ipsecstat.in_polvio++;
910 			goto bad;
911 		}
912 #endif
913 #ifdef FAST_IPSEC
914 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
915 		crit_enter();
916 		if (mtag != NULL) {
917 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
918 			sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
919 		} else {
920 			sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
921 						   IP_FORWARDING, &error);
922 		}
923 		if (sp == NULL) {	/* NB: can happen if error */
924 			crit_exit();
925 			/*XXX error stat???*/
926 			DPRINTF(("ip_input: no SP for forwarding\n"));	/*XXX*/
927 			goto bad;
928 		}
929 
930 		/*
931 		 * Check security policy against packet attributes.
932 		 */
933 		error = ipsec_in_reject(sp, m);
934 		KEY_FREESP(&sp);
935 		crit_exit();
936 		if (error) {
937 			ipstat.ips_cantforward++;
938 			goto bad;
939 		}
940 #endif
941 		ip_forward(m, using_srcrt, next_hop);
942 	}
943 	return;
944 
945 ours:
946 
947 	/*
948 	 * IPSTEALTH: Process non-routing options only
949 	 * if the packet is destined for us.
950 	 */
951 	if (ipstealth &&
952 	    hlen > sizeof(struct ip) &&
953 	    ip_dooptions(m, 1, next_hop))
954 		return;
955 
956 	/* Count the packet in the ip address stats */
957 	if (ia != NULL) {
958 		ia->ia_ifa.if_ipackets++;
959 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
960 	}
961 
962 	/*
963 	 * If offset or IP_MF are set, must reassemble.
964 	 * Otherwise, nothing need be done.
965 	 * (We could look in the reassembly queue to see
966 	 * if the packet was previously fragmented,
967 	 * but it's not worth the time; just let them time out.)
968 	 */
969 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
970 		/*
971 		 * Attempt reassembly; if it succeeds, proceed.
972 		 * ip_reass() will return a different mbuf.
973 		 */
974 		m = ip_reass(m);
975 		if (m == NULL)
976 			return;
977 		ip = mtod(m, struct ip *);
978 
979 		/* Get the header length of the reassembled packet */
980 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
981 
982 		needredispatch = TRUE;
983 	} else {
984 		ip->ip_len -= hlen;
985 	}
986 
987 #ifdef IPSEC
988 	/*
989 	 * enforce IPsec policy checking if we are seeing last header.
990 	 * note that we do not visit this with protocols with pcb layer
991 	 * code - like udp/tcp/raw ip.
992 	 */
993 	if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
994 	    ipsec4_in_reject(m, NULL)) {
995 		ipsecstat.in_polvio++;
996 		goto bad;
997 	}
998 #endif
999 #if FAST_IPSEC
1000 	/*
1001 	 * enforce IPsec policy checking if we are seeing last header.
1002 	 * note that we do not visit this with protocols with pcb layer
1003 	 * code - like udp/tcp/raw ip.
1004 	 */
1005 	if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
1006 		/*
1007 		 * Check if the packet has already had IPsec processing
1008 		 * done.  If so, then just pass it along.  This tag gets
1009 		 * set during AH, ESP, etc. input handling, before the
1010 		 * packet is returned to the ip input queue for delivery.
1011 		 */
1012 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
1013 		crit_enter();
1014 		if (mtag != NULL) {
1015 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
1016 			sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
1017 		} else {
1018 			sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
1019 						   IP_FORWARDING, &error);
1020 		}
1021 		if (sp != NULL) {
1022 			/*
1023 			 * Check security policy against packet attributes.
1024 			 */
1025 			error = ipsec_in_reject(sp, m);
1026 			KEY_FREESP(&sp);
1027 		} else {
1028 			/* XXX error stat??? */
1029 			error = EINVAL;
1030 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
1031 			goto bad;
1032 		}
1033 		crit_exit();
1034 		if (error)
1035 			goto bad;
1036 	}
1037 #endif /* FAST_IPSEC */
1038 
1039 	ipstat.ips_delivered++;
1040 	if (needredispatch) {
1041 		struct netmsg_packet *pmsg;
1042 		lwkt_port_t port;
1043 
1044 		ip->ip_off = htons(ip->ip_off);
1045 		ip->ip_len = htons(ip->ip_len);
1046 		port = ip_mport_in(&m);
1047 		if (port == NULL)
1048 			return;
1049 
1050 		pmsg = &m->m_hdr.mh_netmsg;
1051 		netmsg_init(&pmsg->nm_netmsg, &netisr_apanic_rport, 0,
1052 			    transport_processing_handler);
1053 		pmsg->nm_packet = m;
1054 		pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
1055 
1056 		ip = mtod(m, struct ip *);
1057 		ip->ip_len = ntohs(ip->ip_len);
1058 		ip->ip_off = ntohs(ip->ip_off);
1059 		lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
1060 	} else {
1061 		transport_processing_oncpu(m, hlen, ip);
1062 	}
1063 	return;
1064 
1065 bad:
1066 	m_freem(m);
1067 }
1068 
1069 /*
1070  * Take incoming datagram fragment and try to reassemble it into
1071  * whole datagram.  If a chain for reassembly of this datagram already
1072  * exists, then it is given as fp; otherwise have to make a chain.
1073  */
1074 static struct mbuf *
1075 ip_reass(struct mbuf *m)
1076 {
1077 	struct ip *ip = mtod(m, struct ip *);
1078 	struct mbuf *p = NULL, *q, *nq;
1079 	struct mbuf *n;
1080 	struct ipq *fp = NULL;
1081 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1082 	int i, next;
1083 	u_short sum;
1084 
1085 	/* If maxnipq is 0, never accept fragments. */
1086 	if (maxnipq == 0) {
1087 		ipstat.ips_fragments++;
1088 		ipstat.ips_fragdropped++;
1089 		m_freem(m);
1090 		return NULL;
1091 	}
1092 
1093 	sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
1094 	/*
1095 	 * Look for queue of fragments of this datagram.
1096 	 */
1097 	for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next)
1098 		if (ip->ip_id == fp->ipq_id &&
1099 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1100 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1101 		    ip->ip_p == fp->ipq_p)
1102 			goto found;
1103 
1104 	fp = NULL;
1105 
1106 	/*
1107 	 * Enforce upper bound on number of fragmented packets
1108 	 * for which we attempt reassembly;
1109 	 * If maxnipq is -1, accept all fragments without limitation.
1110 	 */
1111 	if (nipq > maxnipq && maxnipq > 0) {
1112 		/*
1113 		 * drop something from the tail of the current queue
1114 		 * before proceeding further
1115 		 */
1116 		if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1117 			for (i = 0; i < IPREASS_NHASH; i++) {
1118 				if (ipq[i].prev != &ipq[i]) {
1119 					ipstat.ips_fragtimeout +=
1120 					    ipq[i].prev->ipq_nfrags;
1121 					ip_freef(ipq[i].prev);
1122 					break;
1123 				}
1124 			}
1125 		} else {
1126 			ipstat.ips_fragtimeout +=
1127 			    ipq[sum].prev->ipq_nfrags;
1128 			ip_freef(ipq[sum].prev);
1129 		}
1130 	}
1131 found:
1132 	/*
1133 	 * Adjust ip_len to not reflect header,
1134 	 * convert offset of this to bytes.
1135 	 */
1136 	ip->ip_len -= hlen;
1137 	if (ip->ip_off & IP_MF) {
1138 		/*
1139 		 * Make sure that fragments have a data length
1140 		 * that's a non-zero multiple of 8 bytes.
1141 		 */
1142 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1143 			ipstat.ips_toosmall++; /* XXX */
1144 			m_freem(m);
1145 			return NULL;
1146 		}
1147 		m->m_flags |= M_FRAG;
1148 	} else
1149 		m->m_flags &= ~M_FRAG;
1150 	ip->ip_off <<= 3;
1151 
1152 	ipstat.ips_fragments++;
1153 	m->m_pkthdr.header = ip;
1154 
1155 	/*
1156 	 * If the hardware has not done csum over this fragment
1157 	 * then csum_data is not valid at all.
1158 	 */
1159 	if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1160 	    == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1161 		m->m_pkthdr.csum_data = 0;
1162 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1163 	}
1164 
1165 	/*
1166 	 * Presence of header sizes in mbufs
1167 	 * would confuse code below.
1168 	 */
1169 	m->m_data += hlen;
1170 	m->m_len -= hlen;
1171 
1172 	/*
1173 	 * If first fragment to arrive, create a reassembly queue.
1174 	 */
1175 	if (fp == NULL) {
1176 		if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1177 			goto dropfrag;
1178 		insque(fp, &ipq[sum]);
1179 		nipq++;
1180 		fp->ipq_nfrags = 1;
1181 		fp->ipq_ttl = IPFRAGTTL;
1182 		fp->ipq_p = ip->ip_p;
1183 		fp->ipq_id = ip->ip_id;
1184 		fp->ipq_src = ip->ip_src;
1185 		fp->ipq_dst = ip->ip_dst;
1186 		fp->ipq_frags = m;
1187 		m->m_nextpkt = NULL;
1188 		goto inserted;
1189 	} else {
1190 		fp->ipq_nfrags++;
1191 	}
1192 
1193 #define	GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
1194 
1195 	/*
1196 	 * Find a segment which begins after this one does.
1197 	 */
1198 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1199 		if (GETIP(q)->ip_off > ip->ip_off)
1200 			break;
1201 
1202 	/*
1203 	 * If there is a preceding segment, it may provide some of
1204 	 * our data already.  If so, drop the data from the incoming
1205 	 * segment.  If it provides all of our data, drop us, otherwise
1206 	 * stick new segment in the proper place.
1207 	 *
1208 	 * If some of the data is dropped from the the preceding
1209 	 * segment, then it's checksum is invalidated.
1210 	 */
1211 	if (p) {
1212 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1213 		if (i > 0) {
1214 			if (i >= ip->ip_len)
1215 				goto dropfrag;
1216 			m_adj(m, i);
1217 			m->m_pkthdr.csum_flags = 0;
1218 			ip->ip_off += i;
1219 			ip->ip_len -= i;
1220 		}
1221 		m->m_nextpkt = p->m_nextpkt;
1222 		p->m_nextpkt = m;
1223 	} else {
1224 		m->m_nextpkt = fp->ipq_frags;
1225 		fp->ipq_frags = m;
1226 	}
1227 
1228 	/*
1229 	 * While we overlap succeeding segments trim them or,
1230 	 * if they are completely covered, dequeue them.
1231 	 */
1232 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1233 	     q = nq) {
1234 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1235 		if (i < GETIP(q)->ip_len) {
1236 			GETIP(q)->ip_len -= i;
1237 			GETIP(q)->ip_off += i;
1238 			m_adj(q, i);
1239 			q->m_pkthdr.csum_flags = 0;
1240 			break;
1241 		}
1242 		nq = q->m_nextpkt;
1243 		m->m_nextpkt = nq;
1244 		ipstat.ips_fragdropped++;
1245 		fp->ipq_nfrags--;
1246 		q->m_nextpkt = NULL;
1247 		m_freem(q);
1248 	}
1249 
1250 inserted:
1251 	/*
1252 	 * Check for complete reassembly and perform frag per packet
1253 	 * limiting.
1254 	 *
1255 	 * Frag limiting is performed here so that the nth frag has
1256 	 * a chance to complete the packet before we drop the packet.
1257 	 * As a result, n+1 frags are actually allowed per packet, but
1258 	 * only n will ever be stored. (n = maxfragsperpacket.)
1259 	 *
1260 	 */
1261 	next = 0;
1262 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1263 		if (GETIP(q)->ip_off != next) {
1264 			if (fp->ipq_nfrags > maxfragsperpacket) {
1265 				ipstat.ips_fragdropped += fp->ipq_nfrags;
1266 				ip_freef(fp);
1267 			}
1268 			return (NULL);
1269 		}
1270 		next += GETIP(q)->ip_len;
1271 	}
1272 	/* Make sure the last packet didn't have the IP_MF flag */
1273 	if (p->m_flags & M_FRAG) {
1274 		if (fp->ipq_nfrags > maxfragsperpacket) {
1275 			ipstat.ips_fragdropped += fp->ipq_nfrags;
1276 			ip_freef(fp);
1277 		}
1278 		return (NULL);
1279 	}
1280 
1281 	/*
1282 	 * Reassembly is complete.  Make sure the packet is a sane size.
1283 	 */
1284 	q = fp->ipq_frags;
1285 	ip = GETIP(q);
1286 	if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1287 		ipstat.ips_toolong++;
1288 		ipstat.ips_fragdropped += fp->ipq_nfrags;
1289 		ip_freef(fp);
1290 		return (NULL);
1291 	}
1292 
1293 	/*
1294 	 * Concatenate fragments.
1295 	 */
1296 	m = q;
1297 	n = m->m_next;
1298 	m->m_next = NULL;
1299 	m_cat(m, n);
1300 	nq = q->m_nextpkt;
1301 	q->m_nextpkt = NULL;
1302 	for (q = nq; q != NULL; q = nq) {
1303 		nq = q->m_nextpkt;
1304 		q->m_nextpkt = NULL;
1305 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1306 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1307 		m_cat(m, q);
1308 	}
1309 
1310 	/*
1311 	 * Clean up the 1's complement checksum.  Carry over 16 bits must
1312 	 * be added back.  This assumes no more then 65535 packet fragments
1313 	 * were reassembled.  A second carry can also occur (but not a third).
1314 	 */
1315 	m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1316 				(m->m_pkthdr.csum_data >> 16);
1317 	if (m->m_pkthdr.csum_data > 0xFFFF)
1318 		m->m_pkthdr.csum_data -= 0xFFFF;
1319 
1320 	/*
1321 	 * Create header for new ip packet by
1322 	 * modifying header of first packet;
1323 	 * dequeue and discard fragment reassembly header.
1324 	 * Make header visible.
1325 	 */
1326 	ip->ip_len = next;
1327 	ip->ip_src = fp->ipq_src;
1328 	ip->ip_dst = fp->ipq_dst;
1329 	remque(fp);
1330 	nipq--;
1331 	mpipe_free(&ipq_mpipe, fp);
1332 	m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1333 	m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1334 	/* some debugging cruft by sklower, below, will go away soon */
1335 	if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1336 		int plen = 0;
1337 
1338 		for (n = m; n; n = n->m_next)
1339 			plen += n->m_len;
1340 		m->m_pkthdr.len = plen;
1341 	}
1342 
1343 	ipstat.ips_reassembled++;
1344 	return (m);
1345 
1346 dropfrag:
1347 	ipstat.ips_fragdropped++;
1348 	if (fp != NULL)
1349 		fp->ipq_nfrags--;
1350 	m_freem(m);
1351 	return (NULL);
1352 
1353 #undef GETIP
1354 }
1355 
1356 /*
1357  * Free a fragment reassembly header and all
1358  * associated datagrams.
1359  */
1360 static void
1361 ip_freef(struct ipq *fp)
1362 {
1363 	struct mbuf *q;
1364 
1365 	while (fp->ipq_frags) {
1366 		q = fp->ipq_frags;
1367 		fp->ipq_frags = q->m_nextpkt;
1368 		q->m_nextpkt = NULL;
1369 		m_freem(q);
1370 	}
1371 	remque(fp);
1372 	mpipe_free(&ipq_mpipe, fp);
1373 	nipq--;
1374 }
1375 
1376 /*
1377  * IP timer processing;
1378  * if a timer expires on a reassembly
1379  * queue, discard it.
1380  */
1381 void
1382 ip_slowtimo(void)
1383 {
1384 	struct ipq *fp;
1385 	int i;
1386 
1387 	crit_enter();
1388 	for (i = 0; i < IPREASS_NHASH; i++) {
1389 		fp = ipq[i].next;
1390 		if (fp == NULL)
1391 			continue;
1392 		while (fp != &ipq[i]) {
1393 			--fp->ipq_ttl;
1394 			fp = fp->next;
1395 			if (fp->prev->ipq_ttl == 0) {
1396 				ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1397 				ip_freef(fp->prev);
1398 			}
1399 		}
1400 	}
1401 	/*
1402 	 * If we are over the maximum number of fragments
1403 	 * (due to the limit being lowered), drain off
1404 	 * enough to get down to the new limit.
1405 	 */
1406 	if (maxnipq >= 0 && nipq > maxnipq) {
1407 		for (i = 0; i < IPREASS_NHASH; i++) {
1408 			while (nipq > maxnipq &&
1409 				(ipq[i].next != &ipq[i])) {
1410 				ipstat.ips_fragdropped +=
1411 				    ipq[i].next->ipq_nfrags;
1412 				ip_freef(ipq[i].next);
1413 			}
1414 		}
1415 	}
1416 	ipflow_slowtimo();
1417 	crit_exit();
1418 }
1419 
1420 /*
1421  * Drain off all datagram fragments.
1422  */
1423 void
1424 ip_drain(void)
1425 {
1426 	int i;
1427 
1428 	for (i = 0; i < IPREASS_NHASH; i++) {
1429 		while (ipq[i].next != &ipq[i]) {
1430 			ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1431 			ip_freef(ipq[i].next);
1432 		}
1433 	}
1434 	in_rtqdrain();
1435 }
1436 
1437 /*
1438  * Do option processing on a datagram,
1439  * possibly discarding it if bad options are encountered,
1440  * or forwarding it if source-routed.
1441  * The pass argument is used when operating in the IPSTEALTH
1442  * mode to tell what options to process:
1443  * [LS]SRR (pass 0) or the others (pass 1).
1444  * The reason for as many as two passes is that when doing IPSTEALTH,
1445  * non-routing options should be processed only if the packet is for us.
1446  * Returns 1 if packet has been forwarded/freed,
1447  * 0 if the packet should be processed further.
1448  */
1449 static int
1450 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1451 {
1452 	struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1453 	struct ip *ip = mtod(m, struct ip *);
1454 	u_char *cp;
1455 	struct in_ifaddr *ia;
1456 	int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1457 	boolean_t forward = FALSE;
1458 	struct in_addr *sin, dst;
1459 	n_time ntime;
1460 
1461 	dst = ip->ip_dst;
1462 	cp = (u_char *)(ip + 1);
1463 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1464 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1465 		opt = cp[IPOPT_OPTVAL];
1466 		if (opt == IPOPT_EOL)
1467 			break;
1468 		if (opt == IPOPT_NOP)
1469 			optlen = 1;
1470 		else {
1471 			if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1472 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1473 				goto bad;
1474 			}
1475 			optlen = cp[IPOPT_OLEN];
1476 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1477 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1478 				goto bad;
1479 			}
1480 		}
1481 		switch (opt) {
1482 
1483 		default:
1484 			break;
1485 
1486 		/*
1487 		 * Source routing with record.
1488 		 * Find interface with current destination address.
1489 		 * If none on this machine then drop if strictly routed,
1490 		 * or do nothing if loosely routed.
1491 		 * Record interface address and bring up next address
1492 		 * component.  If strictly routed make sure next
1493 		 * address is on directly accessible net.
1494 		 */
1495 		case IPOPT_LSRR:
1496 		case IPOPT_SSRR:
1497 			if (ipstealth && pass > 0)
1498 				break;
1499 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1500 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1501 				goto bad;
1502 			}
1503 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1504 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1505 				goto bad;
1506 			}
1507 			ipaddr.sin_addr = ip->ip_dst;
1508 			ia = (struct in_ifaddr *)
1509 				ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1510 			if (ia == NULL) {
1511 				if (opt == IPOPT_SSRR) {
1512 					type = ICMP_UNREACH;
1513 					code = ICMP_UNREACH_SRCFAIL;
1514 					goto bad;
1515 				}
1516 				if (!ip_dosourceroute)
1517 					goto nosourcerouting;
1518 				/*
1519 				 * Loose routing, and not at next destination
1520 				 * yet; nothing to do except forward.
1521 				 */
1522 				break;
1523 			}
1524 			off--;			/* 0 origin */
1525 			if (off > optlen - (int)sizeof(struct in_addr)) {
1526 				/*
1527 				 * End of source route.  Should be for us.
1528 				 */
1529 				if (!ip_acceptsourceroute)
1530 					goto nosourcerouting;
1531 				save_rte(m, cp, ip->ip_src);
1532 				break;
1533 			}
1534 			if (ipstealth)
1535 				goto dropit;
1536 			if (!ip_dosourceroute) {
1537 				if (ipforwarding) {
1538 					char buf[sizeof "aaa.bbb.ccc.ddd"];
1539 
1540 					/*
1541 					 * Acting as a router, so generate ICMP
1542 					 */
1543 nosourcerouting:
1544 					strcpy(buf, inet_ntoa(ip->ip_dst));
1545 					log(LOG_WARNING,
1546 					    "attempted source route from %s to %s\n",
1547 					    inet_ntoa(ip->ip_src), buf);
1548 					type = ICMP_UNREACH;
1549 					code = ICMP_UNREACH_SRCFAIL;
1550 					goto bad;
1551 				} else {
1552 					/*
1553 					 * Not acting as a router,
1554 					 * so silently drop.
1555 					 */
1556 dropit:
1557 					ipstat.ips_cantforward++;
1558 					m_freem(m);
1559 					return (1);
1560 				}
1561 			}
1562 
1563 			/*
1564 			 * locate outgoing interface
1565 			 */
1566 			memcpy(&ipaddr.sin_addr, cp + off,
1567 			    sizeof ipaddr.sin_addr);
1568 
1569 			if (opt == IPOPT_SSRR) {
1570 #define	INA	struct in_ifaddr *
1571 #define	SA	struct sockaddr *
1572 				if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1573 									== NULL)
1574 					ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1575 			} else
1576 				ia = ip_rtaddr(ipaddr.sin_addr,
1577 					       &ipforward_rt[mycpuid]);
1578 			if (ia == NULL) {
1579 				type = ICMP_UNREACH;
1580 				code = ICMP_UNREACH_SRCFAIL;
1581 				goto bad;
1582 			}
1583 			ip->ip_dst = ipaddr.sin_addr;
1584 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1585 			    sizeof(struct in_addr));
1586 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1587 			/*
1588 			 * Let ip_intr's mcast routing check handle mcast pkts
1589 			 */
1590 			forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1591 			break;
1592 
1593 		case IPOPT_RR:
1594 			if (ipstealth && pass == 0)
1595 				break;
1596 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1597 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1598 				goto bad;
1599 			}
1600 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1601 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1602 				goto bad;
1603 			}
1604 			/*
1605 			 * If no space remains, ignore.
1606 			 */
1607 			off--;			/* 0 origin */
1608 			if (off > optlen - (int)sizeof(struct in_addr))
1609 				break;
1610 			memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1611 			    sizeof ipaddr.sin_addr);
1612 			/*
1613 			 * locate outgoing interface; if we're the destination,
1614 			 * use the incoming interface (should be same).
1615 			 */
1616 			if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1617 			    (ia = ip_rtaddr(ipaddr.sin_addr,
1618 			    		    &ipforward_rt[mycpuid]))
1619 								     == NULL) {
1620 				type = ICMP_UNREACH;
1621 				code = ICMP_UNREACH_HOST;
1622 				goto bad;
1623 			}
1624 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1625 			    sizeof(struct in_addr));
1626 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1627 			break;
1628 
1629 		case IPOPT_TS:
1630 			if (ipstealth && pass == 0)
1631 				break;
1632 			code = cp - (u_char *)ip;
1633 			if (optlen < 4 || optlen > 40) {
1634 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1635 				goto bad;
1636 			}
1637 			if ((off = cp[IPOPT_OFFSET]) < 5) {
1638 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1639 				goto bad;
1640 			}
1641 			if (off > optlen - (int)sizeof(int32_t)) {
1642 				cp[IPOPT_OFFSET + 1] += (1 << 4);
1643 				if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1644 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1645 					goto bad;
1646 				}
1647 				break;
1648 			}
1649 			off--;				/* 0 origin */
1650 			sin = (struct in_addr *)(cp + off);
1651 			switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1652 
1653 			case IPOPT_TS_TSONLY:
1654 				break;
1655 
1656 			case IPOPT_TS_TSANDADDR:
1657 				if (off + sizeof(n_time) +
1658 				    sizeof(struct in_addr) > optlen) {
1659 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1660 					goto bad;
1661 				}
1662 				ipaddr.sin_addr = dst;
1663 				ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1664 							    m->m_pkthdr.rcvif);
1665 				if (ia == NULL)
1666 					continue;
1667 				memcpy(sin, &IA_SIN(ia)->sin_addr,
1668 				    sizeof(struct in_addr));
1669 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1670 				off += sizeof(struct in_addr);
1671 				break;
1672 
1673 			case IPOPT_TS_PRESPEC:
1674 				if (off + sizeof(n_time) +
1675 				    sizeof(struct in_addr) > optlen) {
1676 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1677 					goto bad;
1678 				}
1679 				memcpy(&ipaddr.sin_addr, sin,
1680 				    sizeof(struct in_addr));
1681 				if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1682 					continue;
1683 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1684 				off += sizeof(struct in_addr);
1685 				break;
1686 
1687 			default:
1688 				code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1689 				goto bad;
1690 			}
1691 			ntime = iptime();
1692 			memcpy(cp + off, &ntime, sizeof(n_time));
1693 			cp[IPOPT_OFFSET] += sizeof(n_time);
1694 		}
1695 	}
1696 	if (forward && ipforwarding) {
1697 		ip_forward(m, TRUE, next_hop);
1698 		return (1);
1699 	}
1700 	return (0);
1701 bad:
1702 	icmp_error(m, type, code, 0, 0);
1703 	ipstat.ips_badoptions++;
1704 	return (1);
1705 }
1706 
1707 /*
1708  * Given address of next destination (final or next hop),
1709  * return internet address info of interface to be used to get there.
1710  */
1711 struct in_ifaddr *
1712 ip_rtaddr(struct in_addr dst, struct route *ro)
1713 {
1714 	struct sockaddr_in *sin;
1715 
1716 	sin = (struct sockaddr_in *)&ro->ro_dst;
1717 
1718 	if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1719 		if (ro->ro_rt != NULL) {
1720 			RTFREE(ro->ro_rt);
1721 			ro->ro_rt = NULL;
1722 		}
1723 		sin->sin_family = AF_INET;
1724 		sin->sin_len = sizeof *sin;
1725 		sin->sin_addr = dst;
1726 		rtalloc_ign(ro, RTF_PRCLONING);
1727 	}
1728 
1729 	if (ro->ro_rt == NULL)
1730 		return (NULL);
1731 
1732 	return (ifatoia(ro->ro_rt->rt_ifa));
1733 }
1734 
1735 /*
1736  * Save incoming source route for use in replies,
1737  * to be picked up later by ip_srcroute if the receiver is interested.
1738  */
1739 static void
1740 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1741 {
1742 	struct m_tag *mtag;
1743 	struct ip_srcrt_opt *opt;
1744 	unsigned olen;
1745 
1746 	mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1747 	if (mtag == NULL)
1748 		return;
1749 	opt = m_tag_data(mtag);
1750 
1751 	olen = option[IPOPT_OLEN];
1752 #ifdef DIAGNOSTIC
1753 	if (ipprintfs)
1754 		kprintf("save_rte: olen %d\n", olen);
1755 #endif
1756 	if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1757 		m_tag_free(mtag);
1758 		return;
1759 	}
1760 	bcopy(option, opt->ip_srcrt.srcopt, olen);
1761 	opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1762 	opt->ip_srcrt.dst = dst;
1763 	m_tag_prepend(m, mtag);
1764 }
1765 
1766 /*
1767  * Retrieve incoming source route for use in replies,
1768  * in the same form used by setsockopt.
1769  * The first hop is placed before the options, will be removed later.
1770  */
1771 struct mbuf *
1772 ip_srcroute(struct mbuf *m0)
1773 {
1774 	struct in_addr *p, *q;
1775 	struct mbuf *m;
1776 	struct m_tag *mtag;
1777 	struct ip_srcrt_opt *opt;
1778 
1779 	if (m0 == NULL)
1780 		return NULL;
1781 
1782 	mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1783 	if (mtag == NULL)
1784 		return NULL;
1785 	opt = m_tag_data(mtag);
1786 
1787 	if (opt->ip_nhops == 0)
1788 		return (NULL);
1789 	m = m_get(MB_DONTWAIT, MT_HEADER);
1790 	if (m == NULL)
1791 		return (NULL);
1792 
1793 #define	OPTSIZ	(sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1794 
1795 	/* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1796 	m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1797 		   sizeof(struct in_addr) + OPTSIZ;
1798 #ifdef DIAGNOSTIC
1799 	if (ipprintfs) {
1800 		kprintf("ip_srcroute: nhops %d mlen %d",
1801 			opt->ip_nhops, m->m_len);
1802 	}
1803 #endif
1804 
1805 	/*
1806 	 * First save first hop for return route
1807 	 */
1808 	p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1809 	*(mtod(m, struct in_addr *)) = *p--;
1810 #ifdef DIAGNOSTIC
1811 	if (ipprintfs)
1812 		kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1813 #endif
1814 
1815 	/*
1816 	 * Copy option fields and padding (nop) to mbuf.
1817 	 */
1818 	opt->ip_srcrt.nop = IPOPT_NOP;
1819 	opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1820 	memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1821 	    OPTSIZ);
1822 	q = (struct in_addr *)(mtod(m, caddr_t) +
1823 	    sizeof(struct in_addr) + OPTSIZ);
1824 #undef OPTSIZ
1825 	/*
1826 	 * Record return path as an IP source route,
1827 	 * reversing the path (pointers are now aligned).
1828 	 */
1829 	while (p >= opt->ip_srcrt.route) {
1830 #ifdef DIAGNOSTIC
1831 		if (ipprintfs)
1832 			kprintf(" %x", ntohl(q->s_addr));
1833 #endif
1834 		*q++ = *p--;
1835 	}
1836 	/*
1837 	 * Last hop goes to final destination.
1838 	 */
1839 	*q = opt->ip_srcrt.dst;
1840 	m_tag_delete(m0, mtag);
1841 #ifdef DIAGNOSTIC
1842 	if (ipprintfs)
1843 		kprintf(" %x\n", ntohl(q->s_addr));
1844 #endif
1845 	return (m);
1846 }
1847 
1848 /*
1849  * Strip out IP options.
1850  */
1851 void
1852 ip_stripoptions(struct mbuf *m)
1853 {
1854 	int datalen;
1855 	struct ip *ip = mtod(m, struct ip *);
1856 	caddr_t opts;
1857 	int optlen;
1858 
1859 	optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1860 	opts = (caddr_t)(ip + 1);
1861 	datalen = m->m_len - (sizeof(struct ip) + optlen);
1862 	bcopy(opts + optlen, opts, datalen);
1863 	m->m_len -= optlen;
1864 	if (m->m_flags & M_PKTHDR)
1865 		m->m_pkthdr.len -= optlen;
1866 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1867 }
1868 
1869 u_char inetctlerrmap[PRC_NCMDS] = {
1870 	0,		0,		0,		0,
1871 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1872 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1873 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1874 	0,		0,		0,		0,
1875 	ENOPROTOOPT,	ECONNREFUSED
1876 };
1877 
1878 /*
1879  * Forward a packet.  If some error occurs return the sender
1880  * an icmp packet.  Note we can't always generate a meaningful
1881  * icmp message because icmp doesn't have a large enough repertoire
1882  * of codes and types.
1883  *
1884  * If not forwarding, just drop the packet.  This could be confusing
1885  * if ipforwarding was zero but some routing protocol was advancing
1886  * us as a gateway to somewhere.  However, we must let the routing
1887  * protocol deal with that.
1888  *
1889  * The using_srcrt parameter indicates whether the packet is being forwarded
1890  * via a source route.
1891  */
1892 void
1893 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1894 {
1895 	struct ip *ip = mtod(m, struct ip *);
1896 	struct sockaddr_in *ipforward_rtaddr;
1897 	struct rtentry *rt;
1898 	int error, type = 0, code = 0, destmtu = 0;
1899 	struct mbuf *mcopy;
1900 	n_long dest;
1901 	struct in_addr pkt_dst;
1902 	struct route *cache_rt = &ipforward_rt[mycpuid];
1903 
1904 	dest = INADDR_ANY;
1905 	/*
1906 	 * Cache the destination address of the packet; this may be
1907 	 * changed by use of 'ipfw fwd'.
1908 	 */
1909 	pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1910 
1911 #ifdef DIAGNOSTIC
1912 	if (ipprintfs)
1913 		kprintf("forward: src %x dst %x ttl %x\n",
1914 		       ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1915 #endif
1916 
1917 	if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1918 		ipstat.ips_cantforward++;
1919 		m_freem(m);
1920 		return;
1921 	}
1922 	if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1923 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1924 		return;
1925 	}
1926 
1927 	ipforward_rtaddr = (struct sockaddr_in *) &cache_rt->ro_dst;
1928 	if (cache_rt->ro_rt == NULL ||
1929 	    ipforward_rtaddr->sin_addr.s_addr != pkt_dst.s_addr) {
1930 		if (cache_rt->ro_rt != NULL) {
1931 			RTFREE(cache_rt->ro_rt);
1932 			cache_rt->ro_rt = NULL;
1933 		}
1934 		ipforward_rtaddr->sin_family = AF_INET;
1935 		ipforward_rtaddr->sin_len = sizeof(struct sockaddr_in);
1936 		ipforward_rtaddr->sin_addr = pkt_dst;
1937 		rtalloc_ign(cache_rt, RTF_PRCLONING);
1938 		if (cache_rt->ro_rt == NULL) {
1939 			icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1940 			return;
1941 		}
1942 	}
1943 	rt = cache_rt->ro_rt;
1944 
1945 	/*
1946 	 * Save the IP header and at most 8 bytes of the payload,
1947 	 * in case we need to generate an ICMP message to the src.
1948 	 *
1949 	 * XXX this can be optimized a lot by saving the data in a local
1950 	 * buffer on the stack (72 bytes at most), and only allocating the
1951 	 * mbuf if really necessary. The vast majority of the packets
1952 	 * are forwarded without having to send an ICMP back (either
1953 	 * because unnecessary, or because rate limited), so we are
1954 	 * really we are wasting a lot of work here.
1955 	 *
1956 	 * We don't use m_copy() because it might return a reference
1957 	 * to a shared cluster. Both this function and ip_output()
1958 	 * assume exclusive access to the IP header in `m', so any
1959 	 * data in a cluster may change before we reach icmp_error().
1960 	 */
1961 	MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1962 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1963 		/*
1964 		 * It's probably ok if the pkthdr dup fails (because
1965 		 * the deep copy of the tag chain failed), but for now
1966 		 * be conservative and just discard the copy since
1967 		 * code below may some day want the tags.
1968 		 */
1969 		m_free(mcopy);
1970 		mcopy = NULL;
1971 	}
1972 	if (mcopy != NULL) {
1973 		mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1974 		    (int)ip->ip_len);
1975 		mcopy->m_pkthdr.len = mcopy->m_len;
1976 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1977 	}
1978 
1979 	if (!ipstealth)
1980 		ip->ip_ttl -= IPTTLDEC;
1981 
1982 	/*
1983 	 * If forwarding packet using same interface that it came in on,
1984 	 * perhaps should send a redirect to sender to shortcut a hop.
1985 	 * Only send redirect if source is sending directly to us,
1986 	 * and if packet was not source routed (or has any options).
1987 	 * Also, don't send redirect if forwarding using a default route
1988 	 * or a route modified by a redirect.
1989 	 */
1990 	if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1991 	    !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1992 	    satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1993 	    ipsendredirects && !using_srcrt && next_hop == NULL) {
1994 		u_long src = ntohl(ip->ip_src.s_addr);
1995 		struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1996 
1997 		if (rt_ifa != NULL &&
1998 		    (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1999 			if (rt->rt_flags & RTF_GATEWAY)
2000 				dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
2001 			else
2002 				dest = pkt_dst.s_addr;
2003 			/*
2004 			 * Router requirements says to only send
2005 			 * host redirects.
2006 			 */
2007 			type = ICMP_REDIRECT;
2008 			code = ICMP_REDIRECT_HOST;
2009 #ifdef DIAGNOSTIC
2010 			if (ipprintfs)
2011 				kprintf("redirect (%d) to %x\n", code, dest);
2012 #endif
2013 		}
2014 	}
2015 
2016 	error = ip_output(m, NULL, cache_rt, IP_FORWARDING, NULL, NULL);
2017 	if (error == 0) {
2018 		ipstat.ips_forward++;
2019 		if (type == 0) {
2020 			if (mcopy) {
2021 				ipflow_create(cache_rt, mcopy);
2022 				m_freem(mcopy);
2023 			}
2024 			return;		/* most common case */
2025 		} else {
2026 			ipstat.ips_redirectsent++;
2027 		}
2028 	} else {
2029 		ipstat.ips_cantforward++;
2030 	}
2031 
2032 	if (mcopy == NULL)
2033 		return;
2034 
2035 	/*
2036 	 * Send ICMP message.
2037 	 */
2038 
2039 	switch (error) {
2040 
2041 	case 0:				/* forwarded, but need redirect */
2042 		/* type, code set above */
2043 		break;
2044 
2045 	case ENETUNREACH:		/* shouldn't happen, checked above */
2046 	case EHOSTUNREACH:
2047 	case ENETDOWN:
2048 	case EHOSTDOWN:
2049 	default:
2050 		type = ICMP_UNREACH;
2051 		code = ICMP_UNREACH_HOST;
2052 		break;
2053 
2054 	case EMSGSIZE:
2055 		type = ICMP_UNREACH;
2056 		code = ICMP_UNREACH_NEEDFRAG;
2057 #ifdef IPSEC
2058 		/*
2059 		 * If the packet is routed over IPsec tunnel, tell the
2060 		 * originator the tunnel MTU.
2061 		 *	tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2062 		 * XXX quickhack!!!
2063 		 */
2064 		if (cache_rt->ro_rt != NULL) {
2065 			struct secpolicy *sp = NULL;
2066 			int ipsecerror;
2067 			int ipsechdr;
2068 			struct route *ro;
2069 
2070 			sp = ipsec4_getpolicybyaddr(mcopy,
2071 						    IPSEC_DIR_OUTBOUND,
2072 						    IP_FORWARDING,
2073 						    &ipsecerror);
2074 
2075 			if (sp == NULL)
2076 				destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2077 			else {
2078 				/* count IPsec header size */
2079 				ipsechdr = ipsec4_hdrsiz(mcopy,
2080 							 IPSEC_DIR_OUTBOUND,
2081 							 NULL);
2082 
2083 				/*
2084 				 * find the correct route for outer IPv4
2085 				 * header, compute tunnel MTU.
2086 				 *
2087 				 */
2088 				if (sp->req != NULL && sp->req->sav != NULL &&
2089 				    sp->req->sav->sah != NULL) {
2090 					ro = &sp->req->sav->sah->sa_route;
2091 					if (ro->ro_rt != NULL &&
2092 					    ro->ro_rt->rt_ifp != NULL) {
2093 						destmtu =
2094 						    ro->ro_rt->rt_ifp->if_mtu;
2095 						destmtu -= ipsechdr;
2096 					}
2097 				}
2098 
2099 				key_freesp(sp);
2100 			}
2101 		}
2102 #elif FAST_IPSEC
2103 		/*
2104 		 * If the packet is routed over IPsec tunnel, tell the
2105 		 * originator the tunnel MTU.
2106 		 *	tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2107 		 * XXX quickhack!!!
2108 		 */
2109 		if (cache_rt->ro_rt != NULL) {
2110 			struct secpolicy *sp = NULL;
2111 			int ipsecerror;
2112 			int ipsechdr;
2113 			struct route *ro;
2114 
2115 			sp = ipsec_getpolicybyaddr(mcopy,
2116 						   IPSEC_DIR_OUTBOUND,
2117 						   IP_FORWARDING,
2118 						   &ipsecerror);
2119 
2120 			if (sp == NULL)
2121 				destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2122 			else {
2123 				/* count IPsec header size */
2124 				ipsechdr = ipsec4_hdrsiz(mcopy,
2125 							 IPSEC_DIR_OUTBOUND,
2126 							 NULL);
2127 
2128 				/*
2129 				 * find the correct route for outer IPv4
2130 				 * header, compute tunnel MTU.
2131 				 */
2132 
2133 				if (sp->req != NULL &&
2134 				    sp->req->sav != NULL &&
2135 				    sp->req->sav->sah != NULL) {
2136 					ro = &sp->req->sav->sah->sa_route;
2137 					if (ro->ro_rt != NULL &&
2138 					    ro->ro_rt->rt_ifp != NULL) {
2139 						destmtu =
2140 						    ro->ro_rt->rt_ifp->if_mtu;
2141 						destmtu -= ipsechdr;
2142 					}
2143 				}
2144 
2145 				KEY_FREESP(&sp);
2146 			}
2147 		}
2148 #else /* !IPSEC && !FAST_IPSEC */
2149 		if (cache_rt->ro_rt != NULL)
2150 			destmtu = cache_rt->ro_rt->rt_ifp->if_mtu;
2151 #endif /*IPSEC*/
2152 		ipstat.ips_cantfrag++;
2153 		break;
2154 
2155 	case ENOBUFS:
2156 		/*
2157 		 * A router should not generate ICMP_SOURCEQUENCH as
2158 		 * required in RFC1812 Requirements for IP Version 4 Routers.
2159 		 * Source quench could be a big problem under DoS attacks,
2160 		 * or if the underlying interface is rate-limited.
2161 		 * Those who need source quench packets may re-enable them
2162 		 * via the net.inet.ip.sendsourcequench sysctl.
2163 		 */
2164 		if (!ip_sendsourcequench) {
2165 			m_freem(mcopy);
2166 			return;
2167 		} else {
2168 			type = ICMP_SOURCEQUENCH;
2169 			code = 0;
2170 		}
2171 		break;
2172 
2173 	case EACCES:			/* ipfw denied packet */
2174 		m_freem(mcopy);
2175 		return;
2176 	}
2177 	icmp_error(mcopy, type, code, dest, destmtu);
2178 }
2179 
2180 void
2181 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2182 	       struct mbuf *m)
2183 {
2184 	if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2185 		struct timeval tv;
2186 
2187 		microtime(&tv);
2188 		*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2189 		    SCM_TIMESTAMP, SOL_SOCKET);
2190 		if (*mp)
2191 			mp = &(*mp)->m_next;
2192 	}
2193 	if (inp->inp_flags & INP_RECVDSTADDR) {
2194 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2195 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2196 		if (*mp)
2197 			mp = &(*mp)->m_next;
2198 	}
2199 	if (inp->inp_flags & INP_RECVTTL) {
2200 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2201 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2202 		if (*mp)
2203 			mp = &(*mp)->m_next;
2204 	}
2205 #ifdef notyet
2206 	/* XXX
2207 	 * Moving these out of udp_input() made them even more broken
2208 	 * than they already were.
2209 	 */
2210 	/* options were tossed already */
2211 	if (inp->inp_flags & INP_RECVOPTS) {
2212 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2213 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2214 		if (*mp)
2215 			mp = &(*mp)->m_next;
2216 	}
2217 	/* ip_srcroute doesn't do what we want here, need to fix */
2218 	if (inp->inp_flags & INP_RECVRETOPTS) {
2219 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2220 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2221 		if (*mp)
2222 			mp = &(*mp)->m_next;
2223 	}
2224 #endif
2225 	if (inp->inp_flags & INP_RECVIF) {
2226 		struct ifnet *ifp;
2227 		struct sdlbuf {
2228 			struct sockaddr_dl sdl;
2229 			u_char	pad[32];
2230 		} sdlbuf;
2231 		struct sockaddr_dl *sdp;
2232 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2233 
2234 		if (((ifp = m->m_pkthdr.rcvif)) &&
2235 		    ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2236 			sdp = IF_LLSOCKADDR(ifp);
2237 			/*
2238 			 * Change our mind and don't try copy.
2239 			 */
2240 			if ((sdp->sdl_family != AF_LINK) ||
2241 			    (sdp->sdl_len > sizeof(sdlbuf))) {
2242 				goto makedummy;
2243 			}
2244 			bcopy(sdp, sdl2, sdp->sdl_len);
2245 		} else {
2246 makedummy:
2247 			sdl2->sdl_len =
2248 			    offsetof(struct sockaddr_dl, sdl_data[0]);
2249 			sdl2->sdl_family = AF_LINK;
2250 			sdl2->sdl_index = 0;
2251 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2252 		}
2253 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2254 			IP_RECVIF, IPPROTO_IP);
2255 		if (*mp)
2256 			mp = &(*mp)->m_next;
2257 	}
2258 }
2259 
2260 /*
2261  * XXX these routines are called from the upper part of the kernel.
2262  *
2263  * They could also be moved to ip_mroute.c, since all the RSVP
2264  *  handling is done there already.
2265  */
2266 int
2267 ip_rsvp_init(struct socket *so)
2268 {
2269 	if (so->so_type != SOCK_RAW ||
2270 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2271 		return EOPNOTSUPP;
2272 
2273 	if (ip_rsvpd != NULL)
2274 		return EADDRINUSE;
2275 
2276 	ip_rsvpd = so;
2277 	/*
2278 	 * This may seem silly, but we need to be sure we don't over-increment
2279 	 * the RSVP counter, in case something slips up.
2280 	 */
2281 	if (!ip_rsvp_on) {
2282 		ip_rsvp_on = 1;
2283 		rsvp_on++;
2284 	}
2285 
2286 	return 0;
2287 }
2288 
2289 int
2290 ip_rsvp_done(void)
2291 {
2292 	ip_rsvpd = NULL;
2293 	/*
2294 	 * This may seem silly, but we need to be sure we don't over-decrement
2295 	 * the RSVP counter, in case something slips up.
2296 	 */
2297 	if (ip_rsvp_on) {
2298 		ip_rsvp_on = 0;
2299 		rsvp_on--;
2300 	}
2301 	return 0;
2302 }
2303 
2304 void
2305 rsvp_input(struct mbuf *m, ...)	/* XXX must fixup manually */
2306 {
2307 	int off, proto;
2308 	__va_list ap;
2309 
2310 	__va_start(ap, m);
2311 	off = __va_arg(ap, int);
2312 	proto = __va_arg(ap, int);
2313 	__va_end(ap);
2314 
2315 	if (rsvp_input_p) { /* call the real one if loaded */
2316 		rsvp_input_p(m, off, proto);
2317 		return;
2318 	}
2319 
2320 	/* Can still get packets with rsvp_on = 0 if there is a local member
2321 	 * of the group to which the RSVP packet is addressed.  But in this
2322 	 * case we want to throw the packet away.
2323 	 */
2324 
2325 	if (!rsvp_on) {
2326 		m_freem(m);
2327 		return;
2328 	}
2329 
2330 	if (ip_rsvpd != NULL) {
2331 		rip_input(m, off, proto);
2332 		return;
2333 	}
2334 	/* Drop the packet */
2335 	m_freem(m);
2336 }
2337