xref: /dflybsd-src/sys/netinet/ip_input.c (revision 30e3ae034c9501c319c415ada6d5e23372649c88)
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 1.115 2008/10/28 07:09:26 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 #include <sys/lock.h>
97 
98 #include <sys/mplock2.h>
99 
100 #include <machine/stdarg.h>
101 
102 #include <net/if.h>
103 #include <net/if_types.h>
104 #include <net/if_var.h>
105 #include <net/if_dl.h>
106 #include <net/pfil.h>
107 #include <net/route.h>
108 #include <net/netisr.h>
109 
110 #include <netinet/in.h>
111 #include <netinet/in_systm.h>
112 #include <netinet/in_var.h>
113 #include <netinet/ip.h>
114 #include <netinet/in_pcb.h>
115 #include <netinet/ip_var.h>
116 #include <netinet/ip_icmp.h>
117 #include <netinet/ip_divert.h>
118 #include <netinet/ip_flow.h>
119 
120 #include <sys/thread2.h>
121 #include <sys/msgport2.h>
122 #include <net/netmsg2.h>
123 
124 #include <sys/socketvar.h>
125 
126 #include <net/ipfw/ip_fw.h>
127 #include <net/dummynet/ip_dummynet.h>
128 
129 #ifdef IPSEC
130 #include <netinet6/ipsec.h>
131 #include <netproto/key/key.h>
132 #endif
133 
134 #ifdef FAST_IPSEC
135 #include <netproto/ipsec/ipsec.h>
136 #include <netproto/ipsec/key.h>
137 #endif
138 
139 int rsvp_on = 0;
140 static int ip_rsvp_on;
141 struct socket *ip_rsvpd;
142 
143 int ipforwarding = 0;
144 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
145     &ipforwarding, 0, "Enable IP forwarding between interfaces");
146 
147 static int ipsendredirects = 1; /* XXX */
148 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
149     &ipsendredirects, 0, "Enable sending IP redirects");
150 
151 int ip_defttl = IPDEFTTL;
152 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
153     &ip_defttl, 0, "Maximum TTL on IP packets");
154 
155 static int ip_dosourceroute = 0;
156 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
157     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
158 
159 static int ip_acceptsourceroute = 0;
160 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
161     CTLFLAG_RW, &ip_acceptsourceroute, 0,
162     "Enable accepting source routed IP packets");
163 
164 static int ip_keepfaith = 0;
165 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
166     &ip_keepfaith, 0,
167     "Enable packet capture for FAITH IPv4->IPv6 translator daemon");
168 
169 static int nipq = 0;	/* total # of reass queues */
170 static int maxnipq;
171 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
172     &maxnipq, 0,
173     "Maximum number of IPv4 fragment reassembly queue entries");
174 
175 static int maxfragsperpacket;
176 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
177     &maxfragsperpacket, 0,
178     "Maximum number of IPv4 fragments allowed per packet");
179 
180 static int ip_sendsourcequench = 0;
181 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
182     &ip_sendsourcequench, 0,
183     "Enable the transmission of source quench packets");
184 
185 int ip_do_randomid = 1;
186 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
187     &ip_do_randomid, 0,
188     "Assign random ip_id values");
189 /*
190  * XXX - Setting ip_checkinterface mostly implements the receive side of
191  * the Strong ES model described in RFC 1122, but since the routing table
192  * and transmit implementation do not implement the Strong ES model,
193  * setting this to 1 results in an odd hybrid.
194  *
195  * XXX - ip_checkinterface currently must be disabled if you use ipnat
196  * to translate the destination address to another local interface.
197  *
198  * XXX - ip_checkinterface must be disabled if you add IP aliases
199  * to the loopback interface instead of the interface where the
200  * packets for those addresses are received.
201  */
202 static int ip_checkinterface = 0;
203 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
204     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
205 
206 static int ip_dispatch_fast = 0;
207 static int ip_dispatch_slow = 0;
208 static int ip_dispatch_recheck = 0;
209 static int ip_dispatch_software = 0;
210 SYSCTL_INT(_net_inet_ip, OID_AUTO, dispatch_fast_count, CTLFLAG_RW,
211 	   &ip_dispatch_fast, 0, "");
212 SYSCTL_INT(_net_inet_ip, OID_AUTO, dispatch_slow_count, CTLFLAG_RW,
213 	   &ip_dispatch_slow, 0, "");
214 SYSCTL_INT(_net_inet_ip, OID_AUTO, dispatch_software_count, CTLFLAG_RW,
215 	   &ip_dispatch_software, 0, "");
216 SYSCTL_INT(_net_inet_ip, OID_AUTO, dispatch_recheck_count, CTLFLAG_RW,
217 	   &ip_dispatch_recheck, 0, "");
218 
219 static struct lwkt_token ipq_token = LWKT_TOKEN_MP_INITIALIZER(ipq_token);
220 
221 #ifdef DIAGNOSTIC
222 static int ipprintfs = 0;
223 #endif
224 
225 extern	struct domain inetdomain;
226 extern	struct protosw inetsw[];
227 u_char	ip_protox[IPPROTO_MAX];
228 struct	in_ifaddrhead in_ifaddrheads[MAXCPU];	/* first inet address */
229 struct	in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
230 						/* inet addr hash table */
231 u_long	in_ifaddrhmask;				/* mask for hash table */
232 
233 struct ip_stats ipstats_percpu[MAXCPU];
234 #ifdef SMP
235 static int
236 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
237 {
238 	int cpu, error = 0;
239 
240 	for (cpu = 0; cpu < ncpus; ++cpu) {
241 		if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
242 					sizeof(struct ip_stats))))
243 			break;
244 		if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
245 				       sizeof(struct ip_stats))))
246 			break;
247 	}
248 
249 	return (error);
250 }
251 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
252     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
253 #else
254 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
255     &ipstat, ip_stats, "IP statistics");
256 #endif
257 
258 /* Packet reassembly stuff */
259 #define	IPREASS_NHASH_LOG2	6
260 #define	IPREASS_NHASH		(1 << IPREASS_NHASH_LOG2)
261 #define	IPREASS_HMASK		(IPREASS_NHASH - 1)
262 #define	IPREASS_HASH(x,y)						\
263     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
264 
265 static struct ipq ipq[IPREASS_NHASH];
266 
267 #ifdef IPCTL_DEFMTU
268 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
269     &ip_mtu, 0, "Default MTU");
270 #endif
271 
272 #ifdef IPSTEALTH
273 static int ipstealth = 0;
274 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
275 #else
276 static const int ipstealth = 0;
277 #endif
278 
279 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
280 
281 struct pfil_head inet_pfil_hook;
282 
283 /*
284  * struct ip_srcrt_opt is used to store packet state while it travels
285  * through the stack.
286  *
287  * XXX Note that the code even makes assumptions on the size and
288  * alignment of fields inside struct ip_srcrt so e.g. adding some
289  * fields will break the code.  This needs to be fixed.
290  *
291  * We need to save the IP options in case a protocol wants to respond
292  * to an incoming packet over the same route if the packet got here
293  * using IP source routing.  This allows connection establishment and
294  * maintenance when the remote end is on a network that is not known
295  * to us.
296  */
297 struct ip_srcrt {
298 	struct	in_addr dst;			/* final destination */
299 	char	nop;				/* one NOP to align */
300 	char	srcopt[IPOPT_OFFSET + 1];	/* OPTVAL, OLEN and OFFSET */
301 	struct	in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
302 };
303 
304 struct ip_srcrt_opt {
305 	int		ip_nhops;
306 	struct ip_srcrt	ip_srcrt;
307 };
308 
309 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
310 static struct malloc_pipe ipq_mpipe;
311 
312 static void		save_rte(struct mbuf *, u_char *, struct in_addr);
313 static int		ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
314 static void		ip_freef(struct ipq *);
315 static void		ip_input_handler(struct netmsg *);
316 
317 /*
318  * IP initialization: fill in IP protocol switch table.
319  * All protocols not implemented in kernel go to raw IP protocol handler.
320  */
321 void
322 ip_init(void)
323 {
324 	struct protosw *pr;
325 	int i;
326 #ifdef SMP
327 	int cpu;
328 #endif
329 
330 	/*
331 	 * Make sure we can handle a reasonable number of fragments but
332 	 * cap it at 4000 (XXX).
333 	 */
334 	mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
335 		    IFQ_MAXLEN, 4000, 0, NULL);
336 	for (i = 0; i < ncpus; ++i) {
337 		TAILQ_INIT(&in_ifaddrheads[i]);
338 		in_ifaddrhashtbls[i] =
339 			hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
340 	}
341 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
342 	if (pr == NULL)
343 		panic("ip_init");
344 	for (i = 0; i < IPPROTO_MAX; i++)
345 		ip_protox[i] = pr - inetsw;
346 	for (pr = inetdomain.dom_protosw;
347 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
348 		if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
349 			if (pr->pr_protocol != IPPROTO_RAW)
350 				ip_protox[pr->pr_protocol] = pr - inetsw;
351 		}
352 	}
353 
354 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
355 	inet_pfil_hook.ph_af = AF_INET;
356 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
357 		kprintf("%s: WARNING: unable to register pfil hook, "
358 			"error %d\n", __func__, i);
359 	}
360 
361 	for (i = 0; i < IPREASS_NHASH; i++)
362 		ipq[i].next = ipq[i].prev = &ipq[i];
363 
364 	maxnipq = nmbclusters / 32;
365 	maxfragsperpacket = 16;
366 
367 	ip_id = time_second & 0xffff;
368 
369 	/*
370 	 * Initialize IP statistics counters for each CPU.
371 	 *
372 	 */
373 #ifdef SMP
374 	for (cpu = 0; cpu < ncpus; ++cpu) {
375 		bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
376 	}
377 #else
378 	bzero(&ipstat, sizeof(struct ip_stats));
379 #endif
380 
381 	netisr_register(NETISR_IP, ip_input_handler, ip_cpufn_in);
382 }
383 
384 /* Do transport protocol processing. */
385 static void
386 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
387 {
388 	const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
389 
390 	/*
391 	 * Switch out to protocol's input routine.
392 	 */
393 	PR_GET_MPLOCK(pr);
394 	pr->pr_input(m, hlen, ip->ip_p);
395 	PR_REL_MPLOCK(pr);
396 }
397 
398 static void
399 transport_processing_handler(netmsg_t netmsg)
400 {
401 	struct netmsg_packet *pmsg = (struct netmsg_packet *)netmsg;
402 	struct ip *ip;
403 	int hlen;
404 
405 	ip = mtod(pmsg->nm_packet, struct ip *);
406 	hlen = pmsg->nm_netmsg.nm_lmsg.u.ms_result;
407 
408 	transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
409 	/* netmsg was embedded in the mbuf, do not reply! */
410 }
411 
412 static void
413 ip_input_handler(struct netmsg *msg0)
414 {
415 	struct mbuf *m = ((struct netmsg_packet *)msg0)->nm_packet;
416 
417 	ip_input(m);
418 	/* msg0 was embedded in the mbuf, do not reply! */
419 }
420 
421 /*
422  * IP input routine.  Checksum and byte swap header.  If fragmented
423  * try to reassemble.  Process options.  Pass to next level.
424  */
425 void
426 ip_input(struct mbuf *m)
427 {
428 	struct ip *ip;
429 	struct in_ifaddr *ia = NULL;
430 	struct in_ifaddr_container *iac;
431 	int hlen, checkif;
432 	u_short sum;
433 	struct in_addr pkt_dst;
434 	boolean_t using_srcrt = FALSE;		/* forward (by PFIL_HOOKS) */
435 	struct in_addr odst;			/* original dst address(NAT) */
436 	struct m_tag *mtag;
437 	struct sockaddr_in *next_hop = NULL;
438 	lwkt_port_t port;
439 #ifdef FAST_IPSEC
440 	struct tdb_ident *tdbi;
441 	struct secpolicy *sp;
442 	int error;
443 #endif
444 
445 	M_ASSERTPKTHDR(m);
446 
447 	/*
448 	 * This routine is called from numerous places which may not have
449 	 * characterized the packet.
450 	 */
451 	if ((m->m_flags & M_HASH) == 0) {
452 		++ip_dispatch_software;
453 		ip_cpufn(&m, 0, IP_MPORT_IN);
454 		if (m == NULL)
455 			return;
456 		KKASSERT(m->m_flags & M_HASH);
457 	}
458 	ip = mtod(m, struct ip *);
459 
460 	/*
461 	 * Pull out certain tags
462 	 */
463 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
464 		/* Next hop */
465 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
466 		KKASSERT(mtag != NULL);
467 		next_hop = m_tag_data(mtag);
468 	}
469 
470 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
471 		/* dummynet already filtered us */
472 		ip = mtod(m, struct ip *);
473 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
474 		goto iphack;
475 	}
476 
477 	ipstat.ips_total++;
478 
479 	/* length checks already done in ip_cpufn() */
480 	KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
481 
482 	if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
483 		ipstat.ips_badvers++;
484 		goto bad;
485 	}
486 
487 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
488 	/* length checks already done in ip_cpufn() */
489 	KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
490 	KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
491 
492 	/* 127/8 must not appear on wire - RFC1122 */
493 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
494 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
495 		if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
496 			ipstat.ips_badaddr++;
497 			goto bad;
498 		}
499 	}
500 
501 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
502 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
503 	} else {
504 		if (hlen == sizeof(struct ip))
505 			sum = in_cksum_hdr(ip);
506 		else
507 			sum = in_cksum(m, hlen);
508 	}
509 	if (sum != 0) {
510 		ipstat.ips_badsum++;
511 		goto bad;
512 	}
513 
514 #ifdef ALTQ
515 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
516 		/* packet is dropped by traffic conditioner */
517 		return;
518 	}
519 #endif
520 	/*
521 	 * Convert fields to host representation.
522 	 */
523 	ip->ip_len = ntohs(ip->ip_len);
524 	ip->ip_off = ntohs(ip->ip_off);
525 
526 	/* length checks already done in ip_cpufn() */
527 	KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
528 	KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
529 
530 	/*
531 	 * Trim mbufs if longer than the IP header would have us expect.
532 	 */
533 	if (m->m_pkthdr.len > ip->ip_len) {
534 		if (m->m_len == m->m_pkthdr.len) {
535 			m->m_len = ip->ip_len;
536 			m->m_pkthdr.len = ip->ip_len;
537 		} else {
538 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
539 		}
540 	}
541 #if defined(IPSEC) && !defined(IPSEC_FILTERGIF)
542 	/*
543 	 * Bypass packet filtering for packets from a tunnel (gif).
544 	 */
545 	if (ipsec_gethist(m, NULL))
546 		goto pass;
547 #endif
548 
549 	/*
550 	 * IpHack's section.
551 	 * Right now when no processing on packet has done
552 	 * and it is still fresh out of network we do our black
553 	 * deals with it.
554 	 * - Firewall: deny/allow/divert
555 	 * - Xlate: translate packet's addr/port (NAT).
556 	 * - Pipe: pass pkt through dummynet.
557 	 * - Wrap: fake packet's addr/port <unimpl.>
558 	 * - Encapsulate: put it in another IP and send out. <unimp.>
559 	 */
560 
561 iphack:
562 	/*
563 	 * If we've been forwarded from the output side, then
564 	 * skip the firewall a second time
565 	 */
566 	if (next_hop != NULL)
567 		goto ours;
568 
569 	/* No pfil hooks */
570 	if (!pfil_has_hooks(&inet_pfil_hook)) {
571 		if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
572 			/*
573 			 * Strip dummynet tags from stranded packets
574 			 */
575 			mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
576 			KKASSERT(mtag != NULL);
577 			m_tag_delete(m, mtag);
578 			m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
579 		}
580 		goto pass;
581 	}
582 
583 	/*
584 	 * Run through list of hooks for input packets.
585 	 *
586 	 * NOTE!  If the packet is rewritten pf/ipfw/whoever must
587 	 *	  clear M_HASH.
588 	 */
589 	odst = ip->ip_dst;
590 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
591 		return;
592 	if (m == NULL)	/* consumed by filter */
593 		return;
594 	ip = mtod(m, struct ip *);
595 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
596 	using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
597 
598 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
599 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
600 		KKASSERT(mtag != NULL);
601 		next_hop = m_tag_data(mtag);
602 	}
603 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
604 		ip_dn_queue(m);
605 		return;
606 	}
607 	if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH) {
608 		m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
609 	}
610 pass:
611 	/*
612 	 * Process options and, if not destined for us,
613 	 * ship it on.  ip_dooptions returns 1 when an
614 	 * error was detected (causing an icmp message
615 	 * to be sent and the original packet to be freed).
616 	 */
617 	if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
618 		return;
619 
620 	/* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
621 	 * matter if it is destined to another node, or whether it is
622 	 * a multicast one, RSVP wants it! and prevents it from being forwarded
623 	 * anywhere else. Also checks if the rsvp daemon is running before
624 	 * grabbing the packet.
625 	 */
626 	if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
627 		goto ours;
628 
629 	/*
630 	 * Check our list of addresses, to see if the packet is for us.
631 	 * If we don't have any addresses, assume any unicast packet
632 	 * we receive might be for us (and let the upper layers deal
633 	 * with it).
634 	 */
635 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
636 	    !(m->m_flags & (M_MCAST | M_BCAST)))
637 		goto ours;
638 
639 	/*
640 	 * Cache the destination address of the packet; this may be
641 	 * changed by use of 'ipfw fwd'.
642 	 */
643 	pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
644 
645 	/*
646 	 * Enable a consistency check between the destination address
647 	 * and the arrival interface for a unicast packet (the RFC 1122
648 	 * strong ES model) if IP forwarding is disabled and the packet
649 	 * is not locally generated and the packet is not subject to
650 	 * 'ipfw fwd'.
651 	 *
652 	 * XXX - Checking also should be disabled if the destination
653 	 * address is ipnat'ed to a different interface.
654 	 *
655 	 * XXX - Checking is incompatible with IP aliases added
656 	 * to the loopback interface instead of the interface where
657 	 * the packets are received.
658 	 */
659 	checkif = ip_checkinterface &&
660 		  !ipforwarding &&
661 		  m->m_pkthdr.rcvif != NULL &&
662 		  !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
663 		  next_hop == NULL;
664 
665 	/*
666 	 * Check for exact addresses in the hash bucket.
667 	 */
668 	LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
669 		ia = iac->ia;
670 
671 		/*
672 		 * If the address matches, verify that the packet
673 		 * arrived via the correct interface if checking is
674 		 * enabled.
675 		 */
676 		if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
677 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
678 			goto ours;
679 	}
680 	ia = NULL;
681 
682 	/*
683 	 * Check for broadcast addresses.
684 	 *
685 	 * Only accept broadcast packets that arrive via the matching
686 	 * interface.  Reception of forwarded directed broadcasts would
687 	 * be handled via ip_forward() and ether_output() with the loopback
688 	 * into the stack for SIMPLEX interfaces handled by ether_output().
689 	 */
690 	if (m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
691 		struct ifaddr_container *ifac;
692 
693 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
694 			      ifa_link) {
695 			struct ifaddr *ifa = ifac->ifa;
696 
697 			if (ifa->ifa_addr == NULL) /* shutdown/startup race */
698 				continue;
699 			if (ifa->ifa_addr->sa_family != AF_INET)
700 				continue;
701 			ia = ifatoia(ifa);
702 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
703 								pkt_dst.s_addr)
704 				goto ours;
705 			if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
706 				goto ours;
707 #ifdef BOOTP_COMPAT
708 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
709 				goto ours;
710 #endif
711 		}
712 	}
713 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
714 		struct in_multi *inm;
715 
716 		/* XXX Multicast is not MPSAFE yet */
717 		get_mplock();
718 
719 		if (ip_mrouter != NULL) {
720 			/*
721 			 * If we are acting as a multicast router, all
722 			 * incoming multicast packets are passed to the
723 			 * kernel-level multicast forwarding function.
724 			 * The packet is returned (relatively) intact; if
725 			 * ip_mforward() returns a non-zero value, the packet
726 			 * must be discarded, else it may be accepted below.
727 			 */
728 			if (ip_mforward != NULL &&
729 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
730 				rel_mplock();
731 				ipstat.ips_cantforward++;
732 				m_freem(m);
733 				return;
734 			}
735 
736 			/*
737 			 * The process-level routing daemon needs to receive
738 			 * all multicast IGMP packets, whether or not this
739 			 * host belongs to their destination groups.
740 			 */
741 			if (ip->ip_p == IPPROTO_IGMP) {
742 				rel_mplock();
743 				goto ours;
744 			}
745 			ipstat.ips_forward++;
746 		}
747 		/*
748 		 * See if we belong to the destination multicast group on the
749 		 * arrival interface.
750 		 */
751 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
752 		if (inm == NULL) {
753 			rel_mplock();
754 			ipstat.ips_notmember++;
755 			m_freem(m);
756 			return;
757 		}
758 
759 		rel_mplock();
760 		goto ours;
761 	}
762 	if (ip->ip_dst.s_addr == INADDR_BROADCAST)
763 		goto ours;
764 	if (ip->ip_dst.s_addr == INADDR_ANY)
765 		goto ours;
766 
767 	/*
768 	 * FAITH(Firewall Aided Internet Translator)
769 	 */
770 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
771 		if (ip_keepfaith) {
772 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
773 				goto ours;
774 		}
775 		m_freem(m);
776 		return;
777 	}
778 
779 	/*
780 	 * Not for us; forward if possible and desirable.
781 	 */
782 	if (!ipforwarding) {
783 		ipstat.ips_cantforward++;
784 		m_freem(m);
785 	} else {
786 #ifdef IPSEC
787 		/*
788 		 * Enforce inbound IPsec SPD.
789 		 */
790 		if (ipsec4_in_reject(m, NULL)) {
791 			ipsecstat.in_polvio++;
792 			goto bad;
793 		}
794 #endif
795 #ifdef FAST_IPSEC
796 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
797 		crit_enter();
798 		if (mtag != NULL) {
799 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
800 			sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
801 		} else {
802 			sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
803 						   IP_FORWARDING, &error);
804 		}
805 		if (sp == NULL) {	/* NB: can happen if error */
806 			crit_exit();
807 			/*XXX error stat???*/
808 			DPRINTF(("ip_input: no SP for forwarding\n"));	/*XXX*/
809 			goto bad;
810 		}
811 
812 		/*
813 		 * Check security policy against packet attributes.
814 		 */
815 		error = ipsec_in_reject(sp, m);
816 		KEY_FREESP(&sp);
817 		crit_exit();
818 		if (error) {
819 			ipstat.ips_cantforward++;
820 			goto bad;
821 		}
822 #endif
823 		ip_forward(m, using_srcrt, next_hop);
824 	}
825 	return;
826 
827 ours:
828 
829 	/*
830 	 * IPSTEALTH: Process non-routing options only
831 	 * if the packet is destined for us.
832 	 */
833 	if (ipstealth &&
834 	    hlen > sizeof(struct ip) &&
835 	    ip_dooptions(m, 1, next_hop))
836 		return;
837 
838 	/* Count the packet in the ip address stats */
839 	if (ia != NULL) {
840 		ia->ia_ifa.if_ipackets++;
841 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
842 	}
843 
844 	/*
845 	 * If offset or IP_MF are set, must reassemble.
846 	 * Otherwise, nothing need be done.
847 	 * (We could look in the reassembly queue to see
848 	 * if the packet was previously fragmented,
849 	 * but it's not worth the time; just let them time out.)
850 	 */
851 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
852 		/*
853 		 * Attempt reassembly; if it succeeds, proceed.  ip_reass()
854 		 * will return a different mbuf.
855 		 *
856 		 * NOTE: ip_reass() returns m with M_HASH cleared to force
857 		 *	 us to recharacterize the packet.
858 		 */
859 		m = ip_reass(m);
860 		if (m == NULL)
861 			return;
862 		ip = mtod(m, struct ip *);
863 
864 		/* Get the header length of the reassembled packet */
865 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
866 	} else {
867 		ip->ip_len -= hlen;
868 	}
869 
870 #ifdef IPSEC
871 	/*
872 	 * enforce IPsec policy checking if we are seeing last header.
873 	 * note that we do not visit this with protocols with pcb layer
874 	 * code - like udp/tcp/raw ip.
875 	 */
876 	if ((inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) &&
877 	    ipsec4_in_reject(m, NULL)) {
878 		ipsecstat.in_polvio++;
879 		goto bad;
880 	}
881 #endif
882 #if FAST_IPSEC
883 	/*
884 	 * enforce IPsec policy checking if we are seeing last header.
885 	 * note that we do not visit this with protocols with pcb layer
886 	 * code - like udp/tcp/raw ip.
887 	 */
888 	if (inetsw[ip_protox[ip->ip_p]].pr_flags & PR_LASTHDR) {
889 		/*
890 		 * Check if the packet has already had IPsec processing
891 		 * done.  If so, then just pass it along.  This tag gets
892 		 * set during AH, ESP, etc. input handling, before the
893 		 * packet is returned to the ip input queue for delivery.
894 		 */
895 		mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
896 		crit_enter();
897 		if (mtag != NULL) {
898 			tdbi = (struct tdb_ident *)m_tag_data(mtag);
899 			sp = ipsec_getpolicy(tdbi, IPSEC_DIR_INBOUND);
900 		} else {
901 			sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND,
902 						   IP_FORWARDING, &error);
903 		}
904 		if (sp != NULL) {
905 			/*
906 			 * Check security policy against packet attributes.
907 			 */
908 			error = ipsec_in_reject(sp, m);
909 			KEY_FREESP(&sp);
910 		} else {
911 			/* XXX error stat??? */
912 			error = EINVAL;
913 DPRINTF(("ip_input: no SP, packet discarded\n"));/*XXX*/
914 			goto bad;
915 		}
916 		crit_exit();
917 		if (error)
918 			goto bad;
919 	}
920 #endif /* FAST_IPSEC */
921 
922 	/*
923 	 * We must forward the packet to the correct protocol thread if
924 	 * we are not already in it.
925 	 *
926 	 * NOTE: ip_len is now in host form.  ip_len is not adjusted
927 	 *	 further for protocol processing, instead we pass hlen
928 	 *	 to the protosw and let it deal with it.
929 	 */
930 	ipstat.ips_delivered++;
931 
932 	if ((m->m_flags & M_HASH) == 0) {
933 		++ip_dispatch_recheck;
934 		ip->ip_len = htons(ip->ip_len);
935 		ip->ip_off = htons(ip->ip_off);
936 
937 		ip_cpufn(&m, 0, IP_MPORT_IN);
938 		if (m == NULL)
939 			return;
940 
941 		ip = mtod(m, struct ip *);
942 		ip->ip_len = ntohs(ip->ip_len);
943 		ip->ip_off = ntohs(ip->ip_off);
944 		KKASSERT(m->m_flags & M_HASH);
945 	}
946 	port = cpu_portfn(m->m_pkthdr.hash);
947 
948 	if (port != &curthread->td_msgport) {
949 		struct netmsg_packet *pmsg;
950 
951 		++ip_dispatch_slow;
952 
953 		pmsg = &m->m_hdr.mh_netmsg;
954 		netmsg_init(&pmsg->nm_netmsg, NULL, &netisr_apanic_rport,
955 			    0, transport_processing_handler);
956 		pmsg->nm_packet = m;
957 		pmsg->nm_netmsg.nm_lmsg.u.ms_result = hlen;
958 
959 		lwkt_sendmsg(port, &pmsg->nm_netmsg.nm_lmsg);
960 	} else {
961 		++ip_dispatch_fast;
962 		transport_processing_oncpu(m, hlen, ip);
963 	}
964 	return;
965 
966 bad:
967 	m_freem(m);
968 }
969 
970 /*
971  * Take incoming datagram fragment and try to reassemble it into
972  * whole datagram.  If a chain for reassembly of this datagram already
973  * exists, then it is given as fp; otherwise have to make a chain.
974  */
975 struct mbuf *
976 ip_reass(struct mbuf *m)
977 {
978 	struct ip *ip = mtod(m, struct ip *);
979 	struct mbuf *p = NULL, *q, *nq;
980 	struct mbuf *n;
981 	struct ipq *fp = NULL;
982 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
983 	int i, next;
984 	u_short sum;
985 
986 	/* If maxnipq is 0, never accept fragments. */
987 	if (maxnipq == 0) {
988 		ipstat.ips_fragments++;
989 		ipstat.ips_fragdropped++;
990 		m_freem(m);
991 		return NULL;
992 	}
993 
994 	sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
995 	/*
996 	 * Look for queue of fragments of this datagram.
997 	 */
998 	lwkt_gettoken(&ipq_token);
999 	for (fp = ipq[sum].next; fp != &ipq[sum]; fp = fp->next) {
1000 		if (ip->ip_id == fp->ipq_id &&
1001 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
1002 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
1003 		    ip->ip_p == fp->ipq_p)
1004 			goto found;
1005 	}
1006 
1007 	fp = NULL;
1008 
1009 	/*
1010 	 * Enforce upper bound on number of fragmented packets
1011 	 * for which we attempt reassembly;
1012 	 * If maxnipq is -1, accept all fragments without limitation.
1013 	 */
1014 	if (nipq > maxnipq && maxnipq > 0) {
1015 		/*
1016 		 * drop something from the tail of the current queue
1017 		 * before proceeding further
1018 		 */
1019 		if (ipq[sum].prev == &ipq[sum]) {   /* gak */
1020 			for (i = 0; i < IPREASS_NHASH; i++) {
1021 				if (ipq[i].prev != &ipq[i]) {
1022 					ipstat.ips_fragtimeout +=
1023 					    ipq[i].prev->ipq_nfrags;
1024 					ip_freef(ipq[i].prev);
1025 					break;
1026 				}
1027 			}
1028 		} else {
1029 			ipstat.ips_fragtimeout +=
1030 			    ipq[sum].prev->ipq_nfrags;
1031 			ip_freef(ipq[sum].prev);
1032 		}
1033 	}
1034 found:
1035 	/*
1036 	 * Adjust ip_len to not reflect header,
1037 	 * convert offset of this to bytes.
1038 	 */
1039 	ip->ip_len -= hlen;
1040 	if (ip->ip_off & IP_MF) {
1041 		/*
1042 		 * Make sure that fragments have a data length
1043 		 * that's a non-zero multiple of 8 bytes.
1044 		 */
1045 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
1046 			ipstat.ips_toosmall++; /* XXX */
1047 			m_freem(m);
1048 			goto done;
1049 		}
1050 		m->m_flags |= M_FRAG;
1051 	} else {
1052 		m->m_flags &= ~M_FRAG;
1053 	}
1054 	ip->ip_off <<= 3;
1055 
1056 	ipstat.ips_fragments++;
1057 	m->m_pkthdr.header = ip;
1058 
1059 	/*
1060 	 * If the hardware has not done csum over this fragment
1061 	 * then csum_data is not valid at all.
1062 	 */
1063 	if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1064 	    == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1065 		m->m_pkthdr.csum_data = 0;
1066 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1067 	}
1068 
1069 	/*
1070 	 * Presence of header sizes in mbufs
1071 	 * would confuse code below.
1072 	 */
1073 	m->m_data += hlen;
1074 	m->m_len -= hlen;
1075 
1076 	/*
1077 	 * If first fragment to arrive, create a reassembly queue.
1078 	 */
1079 	if (fp == NULL) {
1080 		if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1081 			goto dropfrag;
1082 		insque(fp, &ipq[sum]);
1083 		nipq++;
1084 		fp->ipq_nfrags = 1;
1085 		fp->ipq_ttl = IPFRAGTTL;
1086 		fp->ipq_p = ip->ip_p;
1087 		fp->ipq_id = ip->ip_id;
1088 		fp->ipq_src = ip->ip_src;
1089 		fp->ipq_dst = ip->ip_dst;
1090 		fp->ipq_frags = m;
1091 		m->m_nextpkt = NULL;
1092 		goto inserted;
1093 	} else {
1094 		fp->ipq_nfrags++;
1095 	}
1096 
1097 #define	GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
1098 
1099 	/*
1100 	 * Find a segment which begins after this one does.
1101 	 */
1102 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1103 		if (GETIP(q)->ip_off > ip->ip_off)
1104 			break;
1105 	}
1106 
1107 	/*
1108 	 * If there is a preceding segment, it may provide some of
1109 	 * our data already.  If so, drop the data from the incoming
1110 	 * segment.  If it provides all of our data, drop us, otherwise
1111 	 * stick new segment in the proper place.
1112 	 *
1113 	 * If some of the data is dropped from the the preceding
1114 	 * segment, then it's checksum is invalidated.
1115 	 */
1116 	if (p) {
1117 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1118 		if (i > 0) {
1119 			if (i >= ip->ip_len)
1120 				goto dropfrag;
1121 			m_adj(m, i);
1122 			m->m_pkthdr.csum_flags = 0;
1123 			ip->ip_off += i;
1124 			ip->ip_len -= i;
1125 		}
1126 		m->m_nextpkt = p->m_nextpkt;
1127 		p->m_nextpkt = m;
1128 	} else {
1129 		m->m_nextpkt = fp->ipq_frags;
1130 		fp->ipq_frags = m;
1131 	}
1132 
1133 	/*
1134 	 * While we overlap succeeding segments trim them or,
1135 	 * if they are completely covered, dequeue them.
1136 	 */
1137 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1138 	     q = nq) {
1139 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1140 		if (i < GETIP(q)->ip_len) {
1141 			GETIP(q)->ip_len -= i;
1142 			GETIP(q)->ip_off += i;
1143 			m_adj(q, i);
1144 			q->m_pkthdr.csum_flags = 0;
1145 			break;
1146 		}
1147 		nq = q->m_nextpkt;
1148 		m->m_nextpkt = nq;
1149 		ipstat.ips_fragdropped++;
1150 		fp->ipq_nfrags--;
1151 		q->m_nextpkt = NULL;
1152 		m_freem(q);
1153 	}
1154 
1155 inserted:
1156 	/*
1157 	 * Check for complete reassembly and perform frag per packet
1158 	 * limiting.
1159 	 *
1160 	 * Frag limiting is performed here so that the nth frag has
1161 	 * a chance to complete the packet before we drop the packet.
1162 	 * As a result, n+1 frags are actually allowed per packet, but
1163 	 * only n will ever be stored. (n = maxfragsperpacket.)
1164 	 *
1165 	 */
1166 	next = 0;
1167 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1168 		if (GETIP(q)->ip_off != next) {
1169 			if (fp->ipq_nfrags > maxfragsperpacket) {
1170 				ipstat.ips_fragdropped += fp->ipq_nfrags;
1171 				ip_freef(fp);
1172 			}
1173 			goto done;
1174 		}
1175 		next += GETIP(q)->ip_len;
1176 	}
1177 	/* Make sure the last packet didn't have the IP_MF flag */
1178 	if (p->m_flags & M_FRAG) {
1179 		if (fp->ipq_nfrags > maxfragsperpacket) {
1180 			ipstat.ips_fragdropped += fp->ipq_nfrags;
1181 			ip_freef(fp);
1182 		}
1183 		goto done;
1184 	}
1185 
1186 	/*
1187 	 * Reassembly is complete.  Make sure the packet is a sane size.
1188 	 */
1189 	q = fp->ipq_frags;
1190 	ip = GETIP(q);
1191 	if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1192 		ipstat.ips_toolong++;
1193 		ipstat.ips_fragdropped += fp->ipq_nfrags;
1194 		ip_freef(fp);
1195 		goto done;
1196 	}
1197 
1198 	/*
1199 	 * Concatenate fragments.
1200 	 */
1201 	m = q;
1202 	n = m->m_next;
1203 	m->m_next = NULL;
1204 	m_cat(m, n);
1205 	nq = q->m_nextpkt;
1206 	q->m_nextpkt = NULL;
1207 	for (q = nq; q != NULL; q = nq) {
1208 		nq = q->m_nextpkt;
1209 		q->m_nextpkt = NULL;
1210 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1211 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1212 		m_cat(m, q);
1213 	}
1214 
1215 	/*
1216 	 * Clean up the 1's complement checksum.  Carry over 16 bits must
1217 	 * be added back.  This assumes no more then 65535 packet fragments
1218 	 * were reassembled.  A second carry can also occur (but not a third).
1219 	 */
1220 	m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1221 				(m->m_pkthdr.csum_data >> 16);
1222 	if (m->m_pkthdr.csum_data > 0xFFFF)
1223 		m->m_pkthdr.csum_data -= 0xFFFF;
1224 
1225 	/*
1226 	 * Create header for new ip packet by
1227 	 * modifying header of first packet;
1228 	 * dequeue and discard fragment reassembly header.
1229 	 * Make header visible.
1230 	 */
1231 	ip->ip_len = next;
1232 	ip->ip_src = fp->ipq_src;
1233 	ip->ip_dst = fp->ipq_dst;
1234 	remque(fp);
1235 	nipq--;
1236 	mpipe_free(&ipq_mpipe, fp);
1237 	m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1238 	m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1239 	/* some debugging cruft by sklower, below, will go away soon */
1240 	if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1241 		int plen = 0;
1242 
1243 		for (n = m; n; n = n->m_next)
1244 			plen += n->m_len;
1245 		m->m_pkthdr.len = plen;
1246 	}
1247 
1248 	/*
1249 	 * Reassembly complete, return the next protocol.
1250 	 *
1251 	 * Be sure to clear M_HASH to force the packet
1252 	 * to be re-characterized.
1253 	 *
1254 	 * Clear M_FRAG, we are no longer a fragment.
1255 	 */
1256 	m->m_flags &= ~(M_HASH | M_FRAG);
1257 
1258 	ipstat.ips_reassembled++;
1259 	lwkt_reltoken(&ipq_token);
1260 	return (m);
1261 
1262 dropfrag:
1263 	ipstat.ips_fragdropped++;
1264 	if (fp != NULL)
1265 		fp->ipq_nfrags--;
1266 	m_freem(m);
1267 done:
1268 	lwkt_reltoken(&ipq_token);
1269 	return (NULL);
1270 
1271 #undef GETIP
1272 }
1273 
1274 /*
1275  * Free a fragment reassembly header and all
1276  * associated datagrams.
1277  *
1278  * Called with ipq_token held.
1279  */
1280 static void
1281 ip_freef(struct ipq *fp)
1282 {
1283 	struct mbuf *q;
1284 
1285 	/*
1286 	 * Remove first to protect against blocking
1287 	 */
1288 	remque(fp);
1289 
1290 	/*
1291 	 * Clean out at our leisure
1292 	 */
1293 	while (fp->ipq_frags) {
1294 		q = fp->ipq_frags;
1295 		fp->ipq_frags = q->m_nextpkt;
1296 		q->m_nextpkt = NULL;
1297 		m_freem(q);
1298 	}
1299 	mpipe_free(&ipq_mpipe, fp);
1300 	nipq--;
1301 }
1302 
1303 /*
1304  * IP timer processing;
1305  * if a timer expires on a reassembly
1306  * queue, discard it.
1307  */
1308 void
1309 ip_slowtimo(void)
1310 {
1311 	struct ipq *fp;
1312 	int i;
1313 
1314 	lwkt_gettoken(&ipq_token);
1315 	for (i = 0; i < IPREASS_NHASH; i++) {
1316 		fp = ipq[i].next;
1317 		if (fp == NULL)
1318 			continue;
1319 		while (fp != &ipq[i]) {
1320 			--fp->ipq_ttl;
1321 			fp = fp->next;
1322 			if (fp->prev->ipq_ttl == 0) {
1323 				ipstat.ips_fragtimeout += fp->prev->ipq_nfrags;
1324 				ip_freef(fp->prev);
1325 			}
1326 		}
1327 	}
1328 	/*
1329 	 * If we are over the maximum number of fragments
1330 	 * (due to the limit being lowered), drain off
1331 	 * enough to get down to the new limit.
1332 	 */
1333 	if (maxnipq >= 0 && nipq > maxnipq) {
1334 		for (i = 0; i < IPREASS_NHASH; i++) {
1335 			while (nipq > maxnipq &&
1336 				(ipq[i].next != &ipq[i])) {
1337 				ipstat.ips_fragdropped +=
1338 				    ipq[i].next->ipq_nfrags;
1339 				ip_freef(ipq[i].next);
1340 			}
1341 		}
1342 	}
1343 	lwkt_reltoken(&ipq_token);
1344 	ipflow_slowtimo();
1345 }
1346 
1347 /*
1348  * Drain off all datagram fragments.
1349  */
1350 void
1351 ip_drain(void)
1352 {
1353 	int i;
1354 
1355 	lwkt_gettoken(&ipq_token);
1356 	for (i = 0; i < IPREASS_NHASH; i++) {
1357 		while (ipq[i].next != &ipq[i]) {
1358 			ipstat.ips_fragdropped += ipq[i].next->ipq_nfrags;
1359 			ip_freef(ipq[i].next);
1360 		}
1361 	}
1362 	lwkt_reltoken(&ipq_token);
1363 	in_rtqdrain();
1364 }
1365 
1366 /*
1367  * Do option processing on a datagram,
1368  * possibly discarding it if bad options are encountered,
1369  * or forwarding it if source-routed.
1370  * The pass argument is used when operating in the IPSTEALTH
1371  * mode to tell what options to process:
1372  * [LS]SRR (pass 0) or the others (pass 1).
1373  * The reason for as many as two passes is that when doing IPSTEALTH,
1374  * non-routing options should be processed only if the packet is for us.
1375  * Returns 1 if packet has been forwarded/freed,
1376  * 0 if the packet should be processed further.
1377  */
1378 static int
1379 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1380 {
1381 	struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1382 	struct ip *ip = mtod(m, struct ip *);
1383 	u_char *cp;
1384 	struct in_ifaddr *ia;
1385 	int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1386 	boolean_t forward = FALSE;
1387 	struct in_addr *sin, dst;
1388 	n_time ntime;
1389 
1390 	dst = ip->ip_dst;
1391 	cp = (u_char *)(ip + 1);
1392 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1393 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1394 		opt = cp[IPOPT_OPTVAL];
1395 		if (opt == IPOPT_EOL)
1396 			break;
1397 		if (opt == IPOPT_NOP)
1398 			optlen = 1;
1399 		else {
1400 			if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1401 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1402 				goto bad;
1403 			}
1404 			optlen = cp[IPOPT_OLEN];
1405 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1406 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1407 				goto bad;
1408 			}
1409 		}
1410 		switch (opt) {
1411 
1412 		default:
1413 			break;
1414 
1415 		/*
1416 		 * Source routing with record.
1417 		 * Find interface with current destination address.
1418 		 * If none on this machine then drop if strictly routed,
1419 		 * or do nothing if loosely routed.
1420 		 * Record interface address and bring up next address
1421 		 * component.  If strictly routed make sure next
1422 		 * address is on directly accessible net.
1423 		 */
1424 		case IPOPT_LSRR:
1425 		case IPOPT_SSRR:
1426 			if (ipstealth && pass > 0)
1427 				break;
1428 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1429 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1430 				goto bad;
1431 			}
1432 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1433 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1434 				goto bad;
1435 			}
1436 			ipaddr.sin_addr = ip->ip_dst;
1437 			ia = (struct in_ifaddr *)
1438 				ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1439 			if (ia == NULL) {
1440 				if (opt == IPOPT_SSRR) {
1441 					type = ICMP_UNREACH;
1442 					code = ICMP_UNREACH_SRCFAIL;
1443 					goto bad;
1444 				}
1445 				if (!ip_dosourceroute)
1446 					goto nosourcerouting;
1447 				/*
1448 				 * Loose routing, and not at next destination
1449 				 * yet; nothing to do except forward.
1450 				 */
1451 				break;
1452 			}
1453 			off--;			/* 0 origin */
1454 			if (off > optlen - (int)sizeof(struct in_addr)) {
1455 				/*
1456 				 * End of source route.  Should be for us.
1457 				 */
1458 				if (!ip_acceptsourceroute)
1459 					goto nosourcerouting;
1460 				save_rte(m, cp, ip->ip_src);
1461 				break;
1462 			}
1463 			if (ipstealth)
1464 				goto dropit;
1465 			if (!ip_dosourceroute) {
1466 				if (ipforwarding) {
1467 					char buf[sizeof "aaa.bbb.ccc.ddd"];
1468 
1469 					/*
1470 					 * Acting as a router, so generate ICMP
1471 					 */
1472 nosourcerouting:
1473 					strcpy(buf, inet_ntoa(ip->ip_dst));
1474 					log(LOG_WARNING,
1475 					    "attempted source route from %s to %s\n",
1476 					    inet_ntoa(ip->ip_src), buf);
1477 					type = ICMP_UNREACH;
1478 					code = ICMP_UNREACH_SRCFAIL;
1479 					goto bad;
1480 				} else {
1481 					/*
1482 					 * Not acting as a router,
1483 					 * so silently drop.
1484 					 */
1485 dropit:
1486 					ipstat.ips_cantforward++;
1487 					m_freem(m);
1488 					return (1);
1489 				}
1490 			}
1491 
1492 			/*
1493 			 * locate outgoing interface
1494 			 */
1495 			memcpy(&ipaddr.sin_addr, cp + off,
1496 			    sizeof ipaddr.sin_addr);
1497 
1498 			if (opt == IPOPT_SSRR) {
1499 #define	INA	struct in_ifaddr *
1500 #define	SA	struct sockaddr *
1501 				if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1502 									== NULL)
1503 					ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1504 			} else {
1505 				ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1506 			}
1507 			if (ia == NULL) {
1508 				type = ICMP_UNREACH;
1509 				code = ICMP_UNREACH_SRCFAIL;
1510 				goto bad;
1511 			}
1512 			ip->ip_dst = ipaddr.sin_addr;
1513 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1514 			    sizeof(struct in_addr));
1515 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1516 			/*
1517 			 * Let ip_intr's mcast routing check handle mcast pkts
1518 			 */
1519 			forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1520 			break;
1521 
1522 		case IPOPT_RR:
1523 			if (ipstealth && pass == 0)
1524 				break;
1525 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1526 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1527 				goto bad;
1528 			}
1529 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1530 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1531 				goto bad;
1532 			}
1533 			/*
1534 			 * If no space remains, ignore.
1535 			 */
1536 			off--;			/* 0 origin */
1537 			if (off > optlen - (int)sizeof(struct in_addr))
1538 				break;
1539 			memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1540 			    sizeof ipaddr.sin_addr);
1541 			/*
1542 			 * locate outgoing interface; if we're the destination,
1543 			 * use the incoming interface (should be same).
1544 			 */
1545 			if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1546 			    (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1547 				type = ICMP_UNREACH;
1548 				code = ICMP_UNREACH_HOST;
1549 				goto bad;
1550 			}
1551 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1552 			    sizeof(struct in_addr));
1553 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1554 			break;
1555 
1556 		case IPOPT_TS:
1557 			if (ipstealth && pass == 0)
1558 				break;
1559 			code = cp - (u_char *)ip;
1560 			if (optlen < 4 || optlen > 40) {
1561 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1562 				goto bad;
1563 			}
1564 			if ((off = cp[IPOPT_OFFSET]) < 5) {
1565 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1566 				goto bad;
1567 			}
1568 			if (off > optlen - (int)sizeof(int32_t)) {
1569 				cp[IPOPT_OFFSET + 1] += (1 << 4);
1570 				if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1571 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1572 					goto bad;
1573 				}
1574 				break;
1575 			}
1576 			off--;				/* 0 origin */
1577 			sin = (struct in_addr *)(cp + off);
1578 			switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1579 
1580 			case IPOPT_TS_TSONLY:
1581 				break;
1582 
1583 			case IPOPT_TS_TSANDADDR:
1584 				if (off + sizeof(n_time) +
1585 				    sizeof(struct in_addr) > optlen) {
1586 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1587 					goto bad;
1588 				}
1589 				ipaddr.sin_addr = dst;
1590 				ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1591 							    m->m_pkthdr.rcvif);
1592 				if (ia == NULL)
1593 					continue;
1594 				memcpy(sin, &IA_SIN(ia)->sin_addr,
1595 				    sizeof(struct in_addr));
1596 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1597 				off += sizeof(struct in_addr);
1598 				break;
1599 
1600 			case IPOPT_TS_PRESPEC:
1601 				if (off + sizeof(n_time) +
1602 				    sizeof(struct in_addr) > optlen) {
1603 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1604 					goto bad;
1605 				}
1606 				memcpy(&ipaddr.sin_addr, sin,
1607 				    sizeof(struct in_addr));
1608 				if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1609 					continue;
1610 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1611 				off += sizeof(struct in_addr);
1612 				break;
1613 
1614 			default:
1615 				code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1616 				goto bad;
1617 			}
1618 			ntime = iptime();
1619 			memcpy(cp + off, &ntime, sizeof(n_time));
1620 			cp[IPOPT_OFFSET] += sizeof(n_time);
1621 		}
1622 	}
1623 	if (forward && ipforwarding) {
1624 		ip_forward(m, TRUE, next_hop);
1625 		return (1);
1626 	}
1627 	return (0);
1628 bad:
1629 	icmp_error(m, type, code, 0, 0);
1630 	ipstat.ips_badoptions++;
1631 	return (1);
1632 }
1633 
1634 /*
1635  * Given address of next destination (final or next hop),
1636  * return internet address info of interface to be used to get there.
1637  */
1638 struct in_ifaddr *
1639 ip_rtaddr(struct in_addr dst, struct route *ro0)
1640 {
1641 	struct route sro, *ro;
1642 	struct sockaddr_in *sin;
1643 	struct in_ifaddr *ia;
1644 
1645 	if (ro0 != NULL) {
1646 		ro = ro0;
1647 	} else {
1648 		bzero(&sro, sizeof(sro));
1649 		ro = &sro;
1650 	}
1651 
1652 	sin = (struct sockaddr_in *)&ro->ro_dst;
1653 
1654 	if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1655 		if (ro->ro_rt != NULL) {
1656 			RTFREE(ro->ro_rt);
1657 			ro->ro_rt = NULL;
1658 		}
1659 		sin->sin_family = AF_INET;
1660 		sin->sin_len = sizeof *sin;
1661 		sin->sin_addr = dst;
1662 		rtalloc_ign(ro, RTF_PRCLONING);
1663 	}
1664 
1665 	if (ro->ro_rt == NULL)
1666 		return (NULL);
1667 
1668 	ia = ifatoia(ro->ro_rt->rt_ifa);
1669 
1670 	if (ro == &sro)
1671 		RTFREE(ro->ro_rt);
1672 	return ia;
1673 }
1674 
1675 /*
1676  * Save incoming source route for use in replies,
1677  * to be picked up later by ip_srcroute if the receiver is interested.
1678  */
1679 static void
1680 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1681 {
1682 	struct m_tag *mtag;
1683 	struct ip_srcrt_opt *opt;
1684 	unsigned olen;
1685 
1686 	mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), MB_DONTWAIT);
1687 	if (mtag == NULL)
1688 		return;
1689 	opt = m_tag_data(mtag);
1690 
1691 	olen = option[IPOPT_OLEN];
1692 #ifdef DIAGNOSTIC
1693 	if (ipprintfs)
1694 		kprintf("save_rte: olen %d\n", olen);
1695 #endif
1696 	if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1697 		m_tag_free(mtag);
1698 		return;
1699 	}
1700 	bcopy(option, opt->ip_srcrt.srcopt, olen);
1701 	opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1702 	opt->ip_srcrt.dst = dst;
1703 	m_tag_prepend(m, mtag);
1704 }
1705 
1706 /*
1707  * Retrieve incoming source route for use in replies,
1708  * in the same form used by setsockopt.
1709  * The first hop is placed before the options, will be removed later.
1710  */
1711 struct mbuf *
1712 ip_srcroute(struct mbuf *m0)
1713 {
1714 	struct in_addr *p, *q;
1715 	struct mbuf *m;
1716 	struct m_tag *mtag;
1717 	struct ip_srcrt_opt *opt;
1718 
1719 	if (m0 == NULL)
1720 		return NULL;
1721 
1722 	mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1723 	if (mtag == NULL)
1724 		return NULL;
1725 	opt = m_tag_data(mtag);
1726 
1727 	if (opt->ip_nhops == 0)
1728 		return (NULL);
1729 	m = m_get(MB_DONTWAIT, MT_HEADER);
1730 	if (m == NULL)
1731 		return (NULL);
1732 
1733 #define	OPTSIZ	(sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1734 
1735 	/* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1736 	m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1737 		   sizeof(struct in_addr) + OPTSIZ;
1738 #ifdef DIAGNOSTIC
1739 	if (ipprintfs) {
1740 		kprintf("ip_srcroute: nhops %d mlen %d",
1741 			opt->ip_nhops, m->m_len);
1742 	}
1743 #endif
1744 
1745 	/*
1746 	 * First save first hop for return route
1747 	 */
1748 	p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1749 	*(mtod(m, struct in_addr *)) = *p--;
1750 #ifdef DIAGNOSTIC
1751 	if (ipprintfs)
1752 		kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1753 #endif
1754 
1755 	/*
1756 	 * Copy option fields and padding (nop) to mbuf.
1757 	 */
1758 	opt->ip_srcrt.nop = IPOPT_NOP;
1759 	opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1760 	memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1761 	    OPTSIZ);
1762 	q = (struct in_addr *)(mtod(m, caddr_t) +
1763 	    sizeof(struct in_addr) + OPTSIZ);
1764 #undef OPTSIZ
1765 	/*
1766 	 * Record return path as an IP source route,
1767 	 * reversing the path (pointers are now aligned).
1768 	 */
1769 	while (p >= opt->ip_srcrt.route) {
1770 #ifdef DIAGNOSTIC
1771 		if (ipprintfs)
1772 			kprintf(" %x", ntohl(q->s_addr));
1773 #endif
1774 		*q++ = *p--;
1775 	}
1776 	/*
1777 	 * Last hop goes to final destination.
1778 	 */
1779 	*q = opt->ip_srcrt.dst;
1780 	m_tag_delete(m0, mtag);
1781 #ifdef DIAGNOSTIC
1782 	if (ipprintfs)
1783 		kprintf(" %x\n", ntohl(q->s_addr));
1784 #endif
1785 	return (m);
1786 }
1787 
1788 /*
1789  * Strip out IP options.
1790  */
1791 void
1792 ip_stripoptions(struct mbuf *m)
1793 {
1794 	int datalen;
1795 	struct ip *ip = mtod(m, struct ip *);
1796 	caddr_t opts;
1797 	int optlen;
1798 
1799 	optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1800 	opts = (caddr_t)(ip + 1);
1801 	datalen = m->m_len - (sizeof(struct ip) + optlen);
1802 	bcopy(opts + optlen, opts, datalen);
1803 	m->m_len -= optlen;
1804 	if (m->m_flags & M_PKTHDR)
1805 		m->m_pkthdr.len -= optlen;
1806 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1807 }
1808 
1809 u_char inetctlerrmap[PRC_NCMDS] = {
1810 	0,		0,		0,		0,
1811 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1812 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1813 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1814 	0,		0,		0,		0,
1815 	ENOPROTOOPT,	ECONNREFUSED
1816 };
1817 
1818 /*
1819  * Forward a packet.  If some error occurs return the sender
1820  * an icmp packet.  Note we can't always generate a meaningful
1821  * icmp message because icmp doesn't have a large enough repertoire
1822  * of codes and types.
1823  *
1824  * If not forwarding, just drop the packet.  This could be confusing
1825  * if ipforwarding was zero but some routing protocol was advancing
1826  * us as a gateway to somewhere.  However, we must let the routing
1827  * protocol deal with that.
1828  *
1829  * The using_srcrt parameter indicates whether the packet is being forwarded
1830  * via a source route.
1831  */
1832 void
1833 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1834 {
1835 	struct ip *ip = mtod(m, struct ip *);
1836 	struct rtentry *rt;
1837 	struct route fwd_ro;
1838 	int error, type = 0, code = 0, destmtu = 0;
1839 	struct mbuf *mcopy;
1840 	n_long dest;
1841 	struct in_addr pkt_dst;
1842 
1843 	dest = INADDR_ANY;
1844 	/*
1845 	 * Cache the destination address of the packet; this may be
1846 	 * changed by use of 'ipfw fwd'.
1847 	 */
1848 	pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1849 
1850 #ifdef DIAGNOSTIC
1851 	if (ipprintfs)
1852 		kprintf("forward: src %x dst %x ttl %x\n",
1853 		       ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1854 #endif
1855 
1856 	if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1857 		ipstat.ips_cantforward++;
1858 		m_freem(m);
1859 		return;
1860 	}
1861 	if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1862 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1863 		return;
1864 	}
1865 
1866 	bzero(&fwd_ro, sizeof(fwd_ro));
1867 	ip_rtaddr(pkt_dst, &fwd_ro);
1868 	if (fwd_ro.ro_rt == NULL) {
1869 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1870 		return;
1871 	}
1872 	rt = fwd_ro.ro_rt;
1873 
1874 	/*
1875 	 * Save the IP header and at most 8 bytes of the payload,
1876 	 * in case we need to generate an ICMP message to the src.
1877 	 *
1878 	 * XXX this can be optimized a lot by saving the data in a local
1879 	 * buffer on the stack (72 bytes at most), and only allocating the
1880 	 * mbuf if really necessary. The vast majority of the packets
1881 	 * are forwarded without having to send an ICMP back (either
1882 	 * because unnecessary, or because rate limited), so we are
1883 	 * really we are wasting a lot of work here.
1884 	 *
1885 	 * We don't use m_copy() because it might return a reference
1886 	 * to a shared cluster. Both this function and ip_output()
1887 	 * assume exclusive access to the IP header in `m', so any
1888 	 * data in a cluster may change before we reach icmp_error().
1889 	 */
1890 	MGETHDR(mcopy, MB_DONTWAIT, m->m_type);
1891 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, MB_DONTWAIT)) {
1892 		/*
1893 		 * It's probably ok if the pkthdr dup fails (because
1894 		 * the deep copy of the tag chain failed), but for now
1895 		 * be conservative and just discard the copy since
1896 		 * code below may some day want the tags.
1897 		 */
1898 		m_free(mcopy);
1899 		mcopy = NULL;
1900 	}
1901 	if (mcopy != NULL) {
1902 		mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1903 		    (int)ip->ip_len);
1904 		mcopy->m_pkthdr.len = mcopy->m_len;
1905 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1906 	}
1907 
1908 	if (!ipstealth)
1909 		ip->ip_ttl -= IPTTLDEC;
1910 
1911 	/*
1912 	 * If forwarding packet using same interface that it came in on,
1913 	 * perhaps should send a redirect to sender to shortcut a hop.
1914 	 * Only send redirect if source is sending directly to us,
1915 	 * and if packet was not source routed (or has any options).
1916 	 * Also, don't send redirect if forwarding using a default route
1917 	 * or a route modified by a redirect.
1918 	 */
1919 	if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1920 	    !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1921 	    satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1922 	    ipsendredirects && !using_srcrt && next_hop == NULL) {
1923 		u_long src = ntohl(ip->ip_src.s_addr);
1924 		struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1925 
1926 		if (rt_ifa != NULL &&
1927 		    (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1928 			if (rt->rt_flags & RTF_GATEWAY)
1929 				dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1930 			else
1931 				dest = pkt_dst.s_addr;
1932 			/*
1933 			 * Router requirements says to only send
1934 			 * host redirects.
1935 			 */
1936 			type = ICMP_REDIRECT;
1937 			code = ICMP_REDIRECT_HOST;
1938 #ifdef DIAGNOSTIC
1939 			if (ipprintfs)
1940 				kprintf("redirect (%d) to %x\n", code, dest);
1941 #endif
1942 		}
1943 	}
1944 
1945 	error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
1946 	if (error == 0) {
1947 		ipstat.ips_forward++;
1948 		if (type == 0) {
1949 			if (mcopy) {
1950 				ipflow_create(&fwd_ro, mcopy);
1951 				m_freem(mcopy);
1952 			}
1953 			goto done;
1954 		} else {
1955 			ipstat.ips_redirectsent++;
1956 		}
1957 	} else {
1958 		ipstat.ips_cantforward++;
1959 	}
1960 
1961 	if (mcopy == NULL)
1962 		goto done;
1963 
1964 	/*
1965 	 * Send ICMP message.
1966 	 */
1967 
1968 	switch (error) {
1969 
1970 	case 0:				/* forwarded, but need redirect */
1971 		/* type, code set above */
1972 		break;
1973 
1974 	case ENETUNREACH:		/* shouldn't happen, checked above */
1975 	case EHOSTUNREACH:
1976 	case ENETDOWN:
1977 	case EHOSTDOWN:
1978 	default:
1979 		type = ICMP_UNREACH;
1980 		code = ICMP_UNREACH_HOST;
1981 		break;
1982 
1983 	case EMSGSIZE:
1984 		type = ICMP_UNREACH;
1985 		code = ICMP_UNREACH_NEEDFRAG;
1986 #ifdef IPSEC
1987 		/*
1988 		 * If the packet is routed over IPsec tunnel, tell the
1989 		 * originator the tunnel MTU.
1990 		 *	tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
1991 		 * XXX quickhack!!!
1992 		 */
1993 		if (fwd_ro.ro_rt != NULL) {
1994 			struct secpolicy *sp = NULL;
1995 			int ipsecerror;
1996 			int ipsechdr;
1997 			struct route *ro;
1998 
1999 			sp = ipsec4_getpolicybyaddr(mcopy,
2000 						    IPSEC_DIR_OUTBOUND,
2001 						    IP_FORWARDING,
2002 						    &ipsecerror);
2003 
2004 			if (sp == NULL)
2005 				destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2006 			else {
2007 				/* count IPsec header size */
2008 				ipsechdr = ipsec4_hdrsiz(mcopy,
2009 							 IPSEC_DIR_OUTBOUND,
2010 							 NULL);
2011 
2012 				/*
2013 				 * find the correct route for outer IPv4
2014 				 * header, compute tunnel MTU.
2015 				 *
2016 				 */
2017 				if (sp->req != NULL && sp->req->sav != NULL &&
2018 				    sp->req->sav->sah != NULL) {
2019 					ro = &sp->req->sav->sah->sa_route;
2020 					if (ro->ro_rt != NULL &&
2021 					    ro->ro_rt->rt_ifp != NULL) {
2022 						destmtu =
2023 						    ro->ro_rt->rt_ifp->if_mtu;
2024 						destmtu -= ipsechdr;
2025 					}
2026 				}
2027 
2028 				key_freesp(sp);
2029 			}
2030 		}
2031 #elif FAST_IPSEC
2032 		/*
2033 		 * If the packet is routed over IPsec tunnel, tell the
2034 		 * originator the tunnel MTU.
2035 		 *	tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
2036 		 * XXX quickhack!!!
2037 		 */
2038 		if (fwd_ro.ro_rt != NULL) {
2039 			struct secpolicy *sp = NULL;
2040 			int ipsecerror;
2041 			int ipsechdr;
2042 			struct route *ro;
2043 
2044 			sp = ipsec_getpolicybyaddr(mcopy,
2045 						   IPSEC_DIR_OUTBOUND,
2046 						   IP_FORWARDING,
2047 						   &ipsecerror);
2048 
2049 			if (sp == NULL)
2050 				destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2051 			else {
2052 				/* count IPsec header size */
2053 				ipsechdr = ipsec4_hdrsiz(mcopy,
2054 							 IPSEC_DIR_OUTBOUND,
2055 							 NULL);
2056 
2057 				/*
2058 				 * find the correct route for outer IPv4
2059 				 * header, compute tunnel MTU.
2060 				 */
2061 
2062 				if (sp->req != NULL &&
2063 				    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 #else /* !IPSEC && !FAST_IPSEC */
2078 		if (fwd_ro.ro_rt != NULL)
2079 			destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2080 #endif /*IPSEC*/
2081 		ipstat.ips_cantfrag++;
2082 		break;
2083 
2084 	case ENOBUFS:
2085 		/*
2086 		 * A router should not generate ICMP_SOURCEQUENCH as
2087 		 * required in RFC1812 Requirements for IP Version 4 Routers.
2088 		 * Source quench could be a big problem under DoS attacks,
2089 		 * or if the underlying interface is rate-limited.
2090 		 * Those who need source quench packets may re-enable them
2091 		 * via the net.inet.ip.sendsourcequench sysctl.
2092 		 */
2093 		if (!ip_sendsourcequench) {
2094 			m_freem(mcopy);
2095 			goto done;
2096 		} else {
2097 			type = ICMP_SOURCEQUENCH;
2098 			code = 0;
2099 		}
2100 		break;
2101 
2102 	case EACCES:			/* ipfw denied packet */
2103 		m_freem(mcopy);
2104 		goto done;
2105 	}
2106 	icmp_error(mcopy, type, code, dest, destmtu);
2107 done:
2108 	if (fwd_ro.ro_rt != NULL)
2109 		RTFREE(fwd_ro.ro_rt);
2110 }
2111 
2112 void
2113 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2114 	       struct mbuf *m)
2115 {
2116 	if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2117 		struct timeval tv;
2118 
2119 		microtime(&tv);
2120 		*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2121 		    SCM_TIMESTAMP, SOL_SOCKET);
2122 		if (*mp)
2123 			mp = &(*mp)->m_next;
2124 	}
2125 	if (inp->inp_flags & INP_RECVDSTADDR) {
2126 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2127 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2128 		if (*mp)
2129 			mp = &(*mp)->m_next;
2130 	}
2131 	if (inp->inp_flags & INP_RECVTTL) {
2132 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2133 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2134 		if (*mp)
2135 			mp = &(*mp)->m_next;
2136 	}
2137 #ifdef notyet
2138 	/* XXX
2139 	 * Moving these out of udp_input() made them even more broken
2140 	 * than they already were.
2141 	 */
2142 	/* options were tossed already */
2143 	if (inp->inp_flags & INP_RECVOPTS) {
2144 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2145 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2146 		if (*mp)
2147 			mp = &(*mp)->m_next;
2148 	}
2149 	/* ip_srcroute doesn't do what we want here, need to fix */
2150 	if (inp->inp_flags & INP_RECVRETOPTS) {
2151 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2152 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2153 		if (*mp)
2154 			mp = &(*mp)->m_next;
2155 	}
2156 #endif
2157 	if (inp->inp_flags & INP_RECVIF) {
2158 		struct ifnet *ifp;
2159 		struct sdlbuf {
2160 			struct sockaddr_dl sdl;
2161 			u_char	pad[32];
2162 		} sdlbuf;
2163 		struct sockaddr_dl *sdp;
2164 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2165 
2166 		if (((ifp = m->m_pkthdr.rcvif)) &&
2167 		    ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2168 			sdp = IF_LLSOCKADDR(ifp);
2169 			/*
2170 			 * Change our mind and don't try copy.
2171 			 */
2172 			if ((sdp->sdl_family != AF_LINK) ||
2173 			    (sdp->sdl_len > sizeof(sdlbuf))) {
2174 				goto makedummy;
2175 			}
2176 			bcopy(sdp, sdl2, sdp->sdl_len);
2177 		} else {
2178 makedummy:
2179 			sdl2->sdl_len =
2180 			    offsetof(struct sockaddr_dl, sdl_data[0]);
2181 			sdl2->sdl_family = AF_LINK;
2182 			sdl2->sdl_index = 0;
2183 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2184 		}
2185 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2186 			IP_RECVIF, IPPROTO_IP);
2187 		if (*mp)
2188 			mp = &(*mp)->m_next;
2189 	}
2190 }
2191 
2192 /*
2193  * XXX these routines are called from the upper part of the kernel.
2194  *
2195  * They could also be moved to ip_mroute.c, since all the RSVP
2196  *  handling is done there already.
2197  */
2198 int
2199 ip_rsvp_init(struct socket *so)
2200 {
2201 	if (so->so_type != SOCK_RAW ||
2202 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2203 		return EOPNOTSUPP;
2204 
2205 	if (ip_rsvpd != NULL)
2206 		return EADDRINUSE;
2207 
2208 	ip_rsvpd = so;
2209 	/*
2210 	 * This may seem silly, but we need to be sure we don't over-increment
2211 	 * the RSVP counter, in case something slips up.
2212 	 */
2213 	if (!ip_rsvp_on) {
2214 		ip_rsvp_on = 1;
2215 		rsvp_on++;
2216 	}
2217 
2218 	return 0;
2219 }
2220 
2221 int
2222 ip_rsvp_done(void)
2223 {
2224 	ip_rsvpd = NULL;
2225 	/*
2226 	 * This may seem silly, but we need to be sure we don't over-decrement
2227 	 * the RSVP counter, in case something slips up.
2228 	 */
2229 	if (ip_rsvp_on) {
2230 		ip_rsvp_on = 0;
2231 		rsvp_on--;
2232 	}
2233 	return 0;
2234 }
2235 
2236 void
2237 rsvp_input(struct mbuf *m, ...)	/* XXX must fixup manually */
2238 {
2239 	int off, proto;
2240 	__va_list ap;
2241 
2242 	__va_start(ap, m);
2243 	off = __va_arg(ap, int);
2244 	proto = __va_arg(ap, int);
2245 	__va_end(ap);
2246 
2247 	if (rsvp_input_p) { /* call the real one if loaded */
2248 		rsvp_input_p(m, off, proto);
2249 		return;
2250 	}
2251 
2252 	/* Can still get packets with rsvp_on = 0 if there is a local member
2253 	 * of the group to which the RSVP packet is addressed.  But in this
2254 	 * case we want to throw the packet away.
2255 	 */
2256 
2257 	if (!rsvp_on) {
2258 		m_freem(m);
2259 		return;
2260 	}
2261 
2262 	if (ip_rsvpd != NULL) {
2263 		rip_input(m, off, proto);
2264 		return;
2265 	}
2266 	/* Drop the packet */
2267 	m_freem(m);
2268 }
2269