xref: /dragonfly/sys/netinet/ip_input.c (revision 8bf5b238)
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
63  * $FreeBSD: src/sys/netinet/ip_input.c,v 1.130.2.52 2003/03/07 07:01:28 silby Exp $
64  */
65 
66 #define	_IP_VHL
67 
68 #include "opt_bootp.h"
69 #include "opt_ipdn.h"
70 #include "opt_ipdivert.h"
71 #include "opt_ipstealth.h"
72 #include "opt_rss.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mbuf.h>
77 #include <sys/malloc.h>
78 #include <sys/mpipe.h>
79 #include <sys/domain.h>
80 #include <sys/protosw.h>
81 #include <sys/socket.h>
82 #include <sys/time.h>
83 #include <sys/globaldata.h>
84 #include <sys/thread.h>
85 #include <sys/kernel.h>
86 #include <sys/syslog.h>
87 #include <sys/sysctl.h>
88 #include <sys/in_cksum.h>
89 #include <sys/lock.h>
90 
91 #include <sys/mplock2.h>
92 
93 #include <machine/stdarg.h>
94 
95 #include <net/if.h>
96 #include <net/if_types.h>
97 #include <net/if_var.h>
98 #include <net/if_dl.h>
99 #include <net/pfil.h>
100 #include <net/route.h>
101 #include <net/netisr2.h>
102 
103 #include <netinet/in.h>
104 #include <netinet/in_systm.h>
105 #include <netinet/in_var.h>
106 #include <netinet/ip.h>
107 #include <netinet/in_pcb.h>
108 #include <netinet/ip_var.h>
109 #include <netinet/ip_icmp.h>
110 #include <netinet/ip_divert.h>
111 #include <netinet/ip_flow.h>
112 
113 #include <sys/thread2.h>
114 #include <sys/msgport2.h>
115 #include <net/netmsg2.h>
116 
117 #include <sys/socketvar.h>
118 
119 #include <net/ipfw/ip_fw.h>
120 #include <net/dummynet/ip_dummynet.h>
121 
122 int rsvp_on = 0;
123 static int ip_rsvp_on;
124 struct socket *ip_rsvpd;
125 
126 int ipforwarding = 0;
127 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
128     &ipforwarding, 0, "Enable IP forwarding between interfaces");
129 
130 static int ipsendredirects = 1; /* XXX */
131 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
132     &ipsendredirects, 0, "Enable sending IP redirects");
133 
134 int ip_defttl = IPDEFTTL;
135 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
136     &ip_defttl, 0, "Maximum TTL on IP packets");
137 
138 static int ip_dosourceroute = 0;
139 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_RW,
140     &ip_dosourceroute, 0, "Enable forwarding source routed IP packets");
141 
142 static int ip_acceptsourceroute = 0;
143 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
144     CTLFLAG_RW, &ip_acceptsourceroute, 0,
145     "Enable accepting source routed IP packets");
146 
147 static int maxnipq;
148 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_RW,
149     &maxnipq, 0,
150     "Maximum number of IPv4 fragment reassembly queue entries");
151 
152 static int maxfragsperpacket;
153 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
154     &maxfragsperpacket, 0,
155     "Maximum number of IPv4 fragments allowed per packet");
156 
157 static int ip_sendsourcequench = 0;
158 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
159     &ip_sendsourcequench, 0,
160     "Enable the transmission of source quench packets");
161 
162 int ip_do_randomid = 1;
163 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
164     &ip_do_randomid, 0,
165     "Assign random ip_id values");
166 /*
167  * XXX - Setting ip_checkinterface mostly implements the receive side of
168  * the Strong ES model described in RFC 1122, but since the routing table
169  * and transmit implementation do not implement the Strong ES model,
170  * setting this to 1 results in an odd hybrid.
171  *
172  * XXX - ip_checkinterface currently must be disabled if you use ipnat
173  * to translate the destination address to another local interface.
174  *
175  * XXX - ip_checkinterface must be disabled if you add IP aliases
176  * to the loopback interface instead of the interface where the
177  * packets for those addresses are received.
178  */
179 static int ip_checkinterface = 0;
180 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
181     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
182 
183 static u_long ip_hash_count = 0;
184 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, hash_count, CTLFLAG_RD,
185     &ip_hash_count, 0, "Number of packets hashed by IP");
186 
187 #ifdef RSS_DEBUG
188 static u_long ip_rehash_count = 0;
189 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, rehash_count, CTLFLAG_RD,
190     &ip_rehash_count, 0, "Number of packets rehashed by IP");
191 
192 static u_long ip_dispatch_fast = 0;
193 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_fast_count, CTLFLAG_RD,
194     &ip_dispatch_fast, 0, "Number of packets handled on current CPU");
195 
196 static u_long ip_dispatch_slow = 0;
197 SYSCTL_ULONG(_net_inet_ip, OID_AUTO, dispatch_slow_count, CTLFLAG_RD,
198     &ip_dispatch_slow, 0, "Number of packets messaged to another CPU");
199 #endif
200 
201 #ifdef DIAGNOSTIC
202 static int ipprintfs = 0;
203 #endif
204 
205 extern	struct domain inetdomain;
206 extern	struct protosw inetsw[];
207 u_char	ip_protox[IPPROTO_MAX];
208 struct	in_ifaddrhead in_ifaddrheads[MAXCPU];	/* first inet address */
209 struct	in_ifaddrhashhead *in_ifaddrhashtbls[MAXCPU];
210 						/* inet addr hash table */
211 u_long	in_ifaddrhmask;				/* mask for hash table */
212 
213 static struct mbuf *ipforward_mtemp[MAXCPU];
214 
215 struct ip_stats ipstats_percpu[MAXCPU] __cachealign;
216 
217 static int
218 sysctl_ipstats(SYSCTL_HANDLER_ARGS)
219 {
220 	int cpu, error = 0;
221 
222 	for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
223 		if ((error = SYSCTL_OUT(req, &ipstats_percpu[cpu],
224 					sizeof(struct ip_stats))))
225 			break;
226 		if ((error = SYSCTL_IN(req, &ipstats_percpu[cpu],
227 				       sizeof(struct ip_stats))))
228 			break;
229 	}
230 
231 	return (error);
232 }
233 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
234     0, 0, sysctl_ipstats, "S,ip_stats", "IP statistics");
235 
236 /* Packet reassembly stuff */
237 #define	IPREASS_NHASH_LOG2	6
238 #define	IPREASS_NHASH		(1 << IPREASS_NHASH_LOG2)
239 #define	IPREASS_HMASK		(IPREASS_NHASH - 1)
240 #define	IPREASS_HASH(x,y)						\
241     (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
242 
243 TAILQ_HEAD(ipqhead, ipq);
244 struct ipfrag_queue {
245 	int			nipq;
246 	volatile int		draining;
247 	struct netmsg_base	timeo_netmsg;
248 	struct callout		timeo_ch;
249 	struct netmsg_base	drain_netmsg;
250 	struct ipqhead		ipq[IPREASS_NHASH];
251 } __cachealign;
252 
253 static struct ipfrag_queue	ipfrag_queue_pcpu[MAXCPU];
254 
255 #ifdef IPCTL_DEFMTU
256 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
257     &ip_mtu, 0, "Default MTU");
258 #endif
259 
260 #ifdef IPSTEALTH
261 static int ipstealth = 0;
262 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW, &ipstealth, 0, "");
263 #else
264 static const int ipstealth = 0;
265 #endif
266 
267 struct mbuf *(*ip_divert_p)(struct mbuf *, int, int);
268 
269 struct pfil_head inet_pfil_hook;
270 
271 /*
272  * struct ip_srcrt_opt is used to store packet state while it travels
273  * through the stack.
274  *
275  * XXX Note that the code even makes assumptions on the size and
276  * alignment of fields inside struct ip_srcrt so e.g. adding some
277  * fields will break the code.  This needs to be fixed.
278  *
279  * We need to save the IP options in case a protocol wants to respond
280  * to an incoming packet over the same route if the packet got here
281  * using IP source routing.  This allows connection establishment and
282  * maintenance when the remote end is on a network that is not known
283  * to us.
284  */
285 struct ip_srcrt {
286 	struct	in_addr dst;			/* final destination */
287 	char	nop;				/* one NOP to align */
288 	char	srcopt[IPOPT_OFFSET + 1];	/* OPTVAL, OLEN and OFFSET */
289 	struct	in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)];
290 };
291 
292 struct ip_srcrt_opt {
293 	int		ip_nhops;
294 	struct ip_srcrt	ip_srcrt;
295 };
296 
297 #define IPFRAG_MPIPE_MAX	4096
298 #define MAXIPFRAG_MIN		((IPFRAG_MPIPE_MAX * 2) / 256)
299 
300 #define IPFRAG_TIMEO		(hz / PR_SLOWHZ)
301 
302 static MALLOC_DEFINE(M_IPQ, "ipq", "IP Fragment Management");
303 static struct malloc_pipe ipq_mpipe;
304 
305 static void		save_rte(struct mbuf *, u_char *, struct in_addr);
306 static int		ip_dooptions(struct mbuf *m, int, struct sockaddr_in *);
307 static void		ip_freef(struct ipfrag_queue *, struct ipqhead *,
308 			    struct ipq *);
309 static void		ip_input_handler(netmsg_t);
310 
311 static void		ipfrag_timeo_dispatch(netmsg_t);
312 static void		ipfrag_timeo(void *);
313 static void		ipfrag_drain_dispatch(netmsg_t);
314 
315 /*
316  * IP initialization: fill in IP protocol switch table.
317  * All protocols not implemented in kernel go to raw IP protocol handler.
318  */
319 void
320 ip_init(void)
321 {
322 	struct ipfrag_queue *fragq;
323 	struct protosw *pr;
324 	int cpu, i;
325 
326 	/*
327 	 * Make sure we can handle a reasonable number of fragments but
328 	 * cap it at IPFRAG_MPIPE_MAX.
329 	 */
330 	mpipe_init(&ipq_mpipe, M_IPQ, sizeof(struct ipq),
331 	    IFQ_MAXLEN, IPFRAG_MPIPE_MAX, 0, NULL, NULL, NULL);
332 
333 	/*
334 	 * Make in_ifaddrhead and in_ifaddrhashtbl available on all CPUs,
335 	 * since they could be accessed by any threads.
336 	 */
337 	for (cpu = 0; cpu < ncpus; ++cpu) {
338 		TAILQ_INIT(&in_ifaddrheads[cpu]);
339 		in_ifaddrhashtbls[cpu] =
340 		    hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
341 	}
342 
343 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
344 	if (pr == NULL)
345 		panic("ip_init");
346 	for (i = 0; i < IPPROTO_MAX; i++)
347 		ip_protox[i] = pr - inetsw;
348 	for (pr = inetdomain.dom_protosw;
349 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
350 		if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol) {
351 			if (pr->pr_protocol != IPPROTO_RAW)
352 				ip_protox[pr->pr_protocol] = pr - inetsw;
353 		}
354 	}
355 
356 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
357 	inet_pfil_hook.ph_af = AF_INET;
358 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0) {
359 		kprintf("%s: WARNING: unable to register pfil hook, "
360 			"error %d\n", __func__, i);
361 	}
362 
363 	maxnipq = (nmbclusters / 32) / netisr_ncpus;
364 	if (maxnipq < MAXIPFRAG_MIN)
365 		maxnipq = MAXIPFRAG_MIN;
366 	maxfragsperpacket = 16;
367 
368 	ip_id = time_second & 0xffff;	/* time_second survives reboots */
369 
370 	for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
371 		/*
372 		 * Initialize IP statistics counters for each CPU.
373 		 */
374 		bzero(&ipstats_percpu[cpu], sizeof(struct ip_stats));
375 
376 		/*
377 		 * Preallocate mbuf template for forwarding
378 		 */
379 		MGETHDR(ipforward_mtemp[cpu], M_WAITOK, MT_DATA);
380 
381 		/*
382 		 * Initialize per-cpu ip fragments queues
383 		 */
384 		fragq = &ipfrag_queue_pcpu[cpu];
385 		for (i = 0; i < IPREASS_NHASH; i++)
386 			TAILQ_INIT(&fragq->ipq[i]);
387 
388 		callout_init_mp(&fragq->timeo_ch);
389 		netmsg_init(&fragq->timeo_netmsg, NULL, &netisr_adone_rport,
390 		    MSGF_PRIORITY, ipfrag_timeo_dispatch);
391 		netmsg_init(&fragq->drain_netmsg, NULL, &netisr_adone_rport,
392 		    MSGF_PRIORITY, ipfrag_drain_dispatch);
393 	}
394 
395 	netisr_register(NETISR_IP, ip_input_handler, ip_hashfn);
396 	netisr_register_hashcheck(NETISR_IP, ip_hashcheck);
397 
398 	for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
399 		fragq = &ipfrag_queue_pcpu[cpu];
400 		callout_reset_bycpu(&fragq->timeo_ch, IPFRAG_TIMEO,
401 		    ipfrag_timeo, NULL, cpu);
402 	}
403 
404 	ip_porthash_trycount = 2 * netisr_ncpus;
405 }
406 
407 /* Do transport protocol processing. */
408 static void
409 transport_processing_oncpu(struct mbuf *m, int hlen, struct ip *ip)
410 {
411 	const struct protosw *pr = &inetsw[ip_protox[ip->ip_p]];
412 
413 	/*
414 	 * Switch out to protocol's input routine.
415 	 */
416 	PR_GET_MPLOCK(pr);
417 	pr->pr_input(&m, &hlen, ip->ip_p);
418 	PR_REL_MPLOCK(pr);
419 }
420 
421 static void
422 transport_processing_handler(netmsg_t msg)
423 {
424 	struct netmsg_packet *pmsg = &msg->packet;
425 	struct ip *ip;
426 	int hlen;
427 
428 	ip = mtod(pmsg->nm_packet, struct ip *);
429 	hlen = pmsg->base.lmsg.u.ms_result;
430 
431 	transport_processing_oncpu(pmsg->nm_packet, hlen, ip);
432 	/* msg was embedded in the mbuf, do not reply! */
433 }
434 
435 static void
436 ip_input_handler(netmsg_t msg)
437 {
438 	ip_input(msg->packet.nm_packet);
439 	/* msg was embedded in the mbuf, do not reply! */
440 }
441 
442 /*
443  * IP input routine.  Checksum and byte swap header.  If fragmented
444  * try to reassemble.  Process options.  Pass to next level.
445  */
446 void
447 ip_input(struct mbuf *m)
448 {
449 	struct ip *ip;
450 	struct in_ifaddr *ia = NULL;
451 	struct in_ifaddr_container *iac;
452 	int hlen, checkif;
453 	u_short sum;
454 	struct in_addr pkt_dst;
455 	boolean_t using_srcrt = FALSE;		/* forward (by PFIL_HOOKS) */
456 	struct in_addr odst;			/* original dst address(NAT) */
457 	struct m_tag *mtag;
458 	struct sockaddr_in *next_hop = NULL;
459 	lwkt_port_t port;
460 
461 	ASSERT_NETISR_NCPUS(mycpuid);
462 	M_ASSERTPKTHDR(m);
463 
464 	/* length checks already done in ip_hashfn() */
465 	KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf"));
466 
467 	/*
468 	 * This routine is called from numerous places which may not have
469 	 * characterized the packet.
470 	 */
471 	ip = mtod(m, struct ip *);
472 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
473 	    (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))) {
474 		/*
475 		 * Force hash recalculation for fragments and multicast
476 		 * packets; hardware may not do it correctly.
477 		 * XXX add flag to indicate the hash is from hardware
478 		 */
479 		m->m_flags &= ~M_HASH;
480 	}
481 	if ((m->m_flags & M_HASH) == 0) {
482 		ip_hashfn(&m, 0);
483 		if (m == NULL)
484 			return;
485 		KKASSERT(m->m_flags & M_HASH);
486 
487 		if (&curthread->td_msgport !=
488 		    netisr_hashport(m->m_pkthdr.hash)) {
489 			netisr_queue(NETISR_IP, m);
490 			/* Requeued to other netisr msgport; done */
491 			return;
492 		}
493 
494 		/* mbuf could have been changed */
495 		ip = mtod(m, struct ip *);
496 	}
497 
498 	/*
499 	 * Pull out certain tags
500 	 */
501 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
502 		/* Next hop */
503 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
504 		KKASSERT(mtag != NULL);
505 		next_hop = m_tag_data(mtag);
506 	}
507 
508 	if (m->m_pkthdr.fw_flags &
509 	    (DUMMYNET_MBUF_TAGGED | IPFW_MBUF_CONTINUE)) {
510 		/*
511 		 * - Dummynet already filtered this packet.
512 		 * - This packet was processed by ipfw on another
513 		 *   cpu, and the rest of the ipfw processing should
514 		 *   be carried out on this cpu.
515 		 */
516 		ip = mtod(m, struct ip *);
517 		ip->ip_len = ntohs(ip->ip_len);
518 		ip->ip_off = ntohs(ip->ip_off);
519 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
520 		goto iphack;
521 	}
522 
523 	ipstat.ips_total++;
524 
525 	if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
526 		ipstat.ips_badvers++;
527 		goto bad;
528 	}
529 
530 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
531 	/* length checks already done in ip_hashfn() */
532 	KASSERT(hlen >= sizeof(struct ip), ("IP header len too small"));
533 	KASSERT(m->m_len >= hlen, ("complete IP header not in one mbuf"));
534 
535 	/* 127/8 must not appear on wire - RFC1122 */
536 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
537 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
538 		if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK)) {
539 			ipstat.ips_badaddr++;
540 			goto bad;
541 		}
542 	}
543 
544 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
545 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
546 	} else {
547 		if (hlen == sizeof(struct ip))
548 			sum = in_cksum_hdr(ip);
549 		else
550 			sum = in_cksum(m, hlen);
551 	}
552 	if (sum != 0) {
553 		ipstat.ips_badsum++;
554 		goto bad;
555 	}
556 
557 #ifdef ALTQ
558 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) {
559 		/* packet is dropped by traffic conditioner */
560 		return;
561 	}
562 #endif
563 	/*
564 	 * Convert fields to host representation.
565 	 */
566 	ip->ip_len = ntohs(ip->ip_len);
567 	ip->ip_off = ntohs(ip->ip_off);
568 
569 	/* length checks already done in ip_hashfn() */
570 	KASSERT(ip->ip_len >= hlen, ("total length less then header length"));
571 	KASSERT(m->m_pkthdr.len >= ip->ip_len, ("mbuf too short"));
572 
573 	/*
574 	 * Trim mbufs if longer than the IP header would have us expect.
575 	 */
576 	if (m->m_pkthdr.len > ip->ip_len) {
577 		if (m->m_len == m->m_pkthdr.len) {
578 			m->m_len = ip->ip_len;
579 			m->m_pkthdr.len = ip->ip_len;
580 		} else {
581 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
582 		}
583 	}
584 
585 	/*
586 	 * IpHack's section.
587 	 * Right now when no processing on packet has done
588 	 * and it is still fresh out of network we do our black
589 	 * deals with it.
590 	 * - Firewall: deny/allow/divert
591 	 * - Xlate: translate packet's addr/port (NAT).
592 	 * - Pipe: pass pkt through dummynet.
593 	 * - Wrap: fake packet's addr/port <unimpl.>
594 	 * - Encapsulate: put it in another IP and send out. <unimp.>
595 	 */
596 
597 iphack:
598 	/*
599 	 * If we've been forwarded from the output side, then
600 	 * skip the firewall a second time
601 	 */
602 	if (next_hop != NULL)
603 		goto ours;
604 
605 	/* No pfil hooks */
606 	if (!pfil_has_hooks(&inet_pfil_hook)) {
607 		if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
608 			/*
609 			 * Strip dummynet tags from stranded packets
610 			 */
611 			mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
612 			KKASSERT(mtag != NULL);
613 			m_tag_delete(m, mtag);
614 			m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
615 		}
616 		goto pass;
617 	}
618 
619 	/*
620 	 * Run through list of hooks for input packets.
621 	 *
622 	 * NOTE!  If the packet is rewritten pf/ipfw/whoever must
623 	 *	  clear M_HASH.
624 	 */
625 	odst = ip->ip_dst;
626 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN))
627 		return;
628 	if (m == NULL)	/* consumed by filter */
629 		return;
630 	ip = mtod(m, struct ip *);
631 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
632 	using_srcrt = (odst.s_addr != ip->ip_dst.s_addr);
633 
634 	if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
635 		mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
636 		KKASSERT(mtag != NULL);
637 		next_hop = m_tag_data(mtag);
638 	}
639 	if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
640 		ip_dn_queue(m);
641 		return;
642 	}
643 	if (m->m_pkthdr.fw_flags & FW_MBUF_REDISPATCH)
644 		m->m_pkthdr.fw_flags &= ~FW_MBUF_REDISPATCH;
645 	if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) {
646 		/* ipfw was disabled/unloaded. */
647 		goto bad;
648 	}
649 pass:
650 	/*
651 	 * Process options and, if not destined for us,
652 	 * ship it on.  ip_dooptions returns 1 when an
653 	 * error was detected (causing an icmp message
654 	 * to be sent and the original packet to be freed).
655 	 */
656 	if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, next_hop))
657 		return;
658 
659 	/* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
660 	 * matter if it is destined to another node, or whether it is
661 	 * a multicast one, RSVP wants it! and prevents it from being forwarded
662 	 * anywhere else. Also checks if the rsvp daemon is running before
663 	 * grabbing the packet.
664 	 */
665 	if (rsvp_on && ip->ip_p == IPPROTO_RSVP)
666 		goto ours;
667 
668 	/*
669 	 * Check our list of addresses, to see if the packet is for us.
670 	 * If we don't have any addresses, assume any unicast packet
671 	 * we receive might be for us (and let the upper layers deal
672 	 * with it).
673 	 */
674 	if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid]) &&
675 	    !(m->m_flags & (M_MCAST | M_BCAST)))
676 		goto ours;
677 
678 	/*
679 	 * Cache the destination address of the packet; this may be
680 	 * changed by use of 'ipfw fwd'.
681 	 */
682 	pkt_dst = next_hop ? next_hop->sin_addr : ip->ip_dst;
683 
684 	/*
685 	 * Enable a consistency check between the destination address
686 	 * and the arrival interface for a unicast packet (the RFC 1122
687 	 * strong ES model) if IP forwarding is disabled and the packet
688 	 * is not locally generated and the packet is not subject to
689 	 * 'ipfw fwd'.
690 	 *
691 	 * XXX - Checking also should be disabled if the destination
692 	 * address is ipnat'ed to a different interface.
693 	 *
694 	 * XXX - Checking is incompatible with IP aliases added
695 	 * to the loopback interface instead of the interface where
696 	 * the packets are received.
697 	 */
698 	checkif = ip_checkinterface &&
699 		  !ipforwarding &&
700 		  m->m_pkthdr.rcvif != NULL &&
701 		  !(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
702 		  next_hop == NULL;
703 
704 	/*
705 	 * Check for exact addresses in the hash bucket.
706 	 */
707 	LIST_FOREACH(iac, INADDR_HASH(pkt_dst.s_addr), ia_hash) {
708 		ia = iac->ia;
709 
710 		/*
711 		 * If the address matches, verify that the packet
712 		 * arrived via the correct interface if checking is
713 		 * enabled.
714 		 */
715 		if (IA_SIN(ia)->sin_addr.s_addr == pkt_dst.s_addr &&
716 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
717 			goto ours;
718 	}
719 	ia = NULL;
720 
721 	/*
722 	 * Check for broadcast addresses.
723 	 *
724 	 * Only accept broadcast packets that arrive via the matching
725 	 * interface.  Reception of forwarded directed broadcasts would
726 	 * be handled via ip_forward() and ether_output() with the loopback
727 	 * into the stack for SIMPLEX interfaces handled by ether_output().
728 	 */
729 	if (m->m_pkthdr.rcvif != NULL &&
730 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
731 		struct ifaddr_container *ifac;
732 
733 		TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid],
734 			      ifa_link) {
735 			struct ifaddr *ifa = ifac->ifa;
736 
737 			if (ifa->ifa_addr == NULL) /* shutdown/startup race */
738 				continue;
739 			if (ifa->ifa_addr->sa_family != AF_INET)
740 				continue;
741 			ia = ifatoia(ifa);
742 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
743 								pkt_dst.s_addr)
744 				goto ours;
745 			if (ia->ia_netbroadcast.s_addr == pkt_dst.s_addr)
746 				goto ours;
747 #ifdef BOOTP_COMPAT
748 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
749 				goto ours;
750 #endif
751 		}
752 	}
753 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
754 		struct in_multi *inm;
755 
756 		if (ip_mrouter != NULL) {
757 			/* XXX Multicast routing is not MPSAFE yet */
758 			get_mplock();
759 
760 			/*
761 			 * If we are acting as a multicast router, all
762 			 * incoming multicast packets are passed to the
763 			 * kernel-level multicast forwarding function.
764 			 * The packet is returned (relatively) intact; if
765 			 * ip_mforward() returns a non-zero value, the packet
766 			 * must be discarded, else it may be accepted below.
767 			 */
768 			if (ip_mforward != NULL &&
769 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, NULL) != 0) {
770 				rel_mplock();
771 				ipstat.ips_cantforward++;
772 				m_freem(m);
773 				return;
774 			}
775 
776 			rel_mplock();
777 
778 			/*
779 			 * The process-level routing daemon needs to receive
780 			 * all multicast IGMP packets, whether or not this
781 			 * host belongs to their destination groups.
782 			 */
783 			if (ip->ip_p == IPPROTO_IGMP)
784 				goto ours;
785 			ipstat.ips_forward++;
786 		}
787 		/*
788 		 * See if we belong to the destination multicast group on the
789 		 * arrival interface.
790 		 */
791 		inm = IN_LOOKUP_MULTI(&ip->ip_dst, m->m_pkthdr.rcvif);
792 		if (inm == NULL) {
793 			ipstat.ips_notmember++;
794 			m_freem(m);
795 			return;
796 		}
797 		goto ours;
798 	}
799 	if (ip->ip_dst.s_addr == INADDR_BROADCAST)
800 		goto ours;
801 	if (ip->ip_dst.s_addr == INADDR_ANY)
802 		goto ours;
803 
804 	/*
805 	 * Not for us; forward if possible and desirable.
806 	 */
807 	if (!ipforwarding) {
808 		ipstat.ips_cantforward++;
809 		m_freem(m);
810 	} else {
811 		ip_forward(m, using_srcrt, next_hop);
812 	}
813 	return;
814 
815 ours:
816 
817 	/*
818 	 * IPSTEALTH: Process non-routing options only
819 	 * if the packet is destined for us.
820 	 */
821 	if (ipstealth &&
822 	    hlen > sizeof(struct ip) &&
823 	    ip_dooptions(m, 1, next_hop))
824 		return;
825 
826 	/* Count the packet in the ip address stats */
827 	if (ia != NULL) {
828 		IFA_STAT_INC(&ia->ia_ifa, ipackets, 1);
829 		IFA_STAT_INC(&ia->ia_ifa, ibytes, m->m_pkthdr.len);
830 	}
831 
832 	/*
833 	 * If offset or IP_MF are set, must reassemble.
834 	 * Otherwise, nothing need be done.
835 	 * (We could look in the reassembly queue to see
836 	 * if the packet was previously fragmented,
837 	 * but it's not worth the time; just let them time out.)
838 	 */
839 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
840 		/*
841 		 * Attempt reassembly; if it succeeds, proceed.  ip_reass()
842 		 * will return a different mbuf.
843 		 *
844 		 * NOTE: ip_reass() returns m with M_HASH cleared to force
845 		 *	 us to recharacterize the packet.
846 		 */
847 		m = ip_reass(m);
848 		if (m == NULL)
849 			return;
850 		ip = mtod(m, struct ip *);
851 
852 		/* Get the header length of the reassembled packet */
853 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
854 	} else {
855 		ip->ip_len -= hlen;
856 	}
857 
858 
859 	/*
860 	 * We must forward the packet to the correct protocol thread if
861 	 * we are not already in it.
862 	 *
863 	 * NOTE: ip_len is now in host form.  ip_len is not adjusted
864 	 *	 further for protocol processing, instead we pass hlen
865 	 *	 to the protosw and let it deal with it.
866 	 */
867 	ipstat.ips_delivered++;
868 
869 	if ((m->m_flags & M_HASH) == 0) {
870 #ifdef RSS_DEBUG
871 		atomic_add_long(&ip_rehash_count, 1);
872 #endif
873 		ip->ip_len = htons(ip->ip_len + hlen);
874 		ip->ip_off = htons(ip->ip_off);
875 
876 		ip_hashfn(&m, 0);
877 		if (m == NULL)
878 			return;
879 
880 		ip = mtod(m, struct ip *);
881 		ip->ip_len = ntohs(ip->ip_len) - hlen;
882 		ip->ip_off = ntohs(ip->ip_off);
883 		KKASSERT(m->m_flags & M_HASH);
884 	}
885 	port = netisr_hashport(m->m_pkthdr.hash);
886 
887 	if (port != &curthread->td_msgport) {
888 		struct netmsg_packet *pmsg;
889 
890 #ifdef RSS_DEBUG
891 		atomic_add_long(&ip_dispatch_slow, 1);
892 #endif
893 
894 		pmsg = &m->m_hdr.mh_netmsg;
895 		netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
896 			    0, transport_processing_handler);
897 		pmsg->nm_packet = m;
898 		pmsg->base.lmsg.u.ms_result = hlen;
899 		lwkt_sendmsg(port, &pmsg->base.lmsg);
900 	} else {
901 #ifdef RSS_DEBUG
902 		atomic_add_long(&ip_dispatch_fast, 1);
903 #endif
904 		transport_processing_oncpu(m, hlen, ip);
905 	}
906 	return;
907 
908 bad:
909 	m_freem(m);
910 }
911 
912 /*
913  * Take incoming datagram fragment and try to reassemble it into
914  * whole datagram.  If a chain for reassembly of this datagram already
915  * exists, then it is given as fp; otherwise have to make a chain.
916  */
917 struct mbuf *
918 ip_reass(struct mbuf *m)
919 {
920 	struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[mycpuid];
921 	struct ip *ip = mtod(m, struct ip *);
922 	struct mbuf *p = NULL, *q, *nq;
923 	struct mbuf *n;
924 	struct ipq *fp = NULL;
925 	struct ipqhead *head;
926 	int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
927 	int i, next;
928 	u_short sum;
929 
930 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
931 	if (maxnipq == 0 || maxfragsperpacket == 0) {
932 		ipstat.ips_fragments++;
933 		ipstat.ips_fragdropped++;
934 		m_freem(m);
935 		return NULL;
936 	}
937 
938 	sum = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
939 	/*
940 	 * Look for queue of fragments of this datagram.
941 	 */
942 	head = &fragq->ipq[sum];
943 	TAILQ_FOREACH(fp, head, ipq_list) {
944 		if (ip->ip_id == fp->ipq_id &&
945 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
946 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
947 		    ip->ip_p == fp->ipq_p)
948 			goto found;
949 	}
950 
951 	fp = NULL;
952 
953 	/*
954 	 * Enforce upper bound on number of fragmented packets
955 	 * for which we attempt reassembly;
956 	 * If maxnipq is -1, accept all fragments without limitation.
957 	 */
958 	if (fragq->nipq > maxnipq && maxnipq > 0) {
959 		/*
960 		 * drop something from the tail of the current queue
961 		 * before proceeding further
962 		 */
963 		struct ipq *q = TAILQ_LAST(head, ipqhead);
964 		if (q == NULL) {
965 			/*
966 			 * The current queue is empty,
967 			 * so drop from one of the others.
968 			 */
969 			for (i = 0; i < IPREASS_NHASH; i++) {
970 				struct ipq *r = TAILQ_LAST(&fragq->ipq[i],
971 				    ipqhead);
972 				if (r) {
973 					ipstat.ips_fragtimeout += r->ipq_nfrags;
974 					ip_freef(fragq, &fragq->ipq[i], r);
975 					break;
976 				}
977 			}
978 		} else {
979 			ipstat.ips_fragtimeout += q->ipq_nfrags;
980 			ip_freef(fragq, head, q);
981 		}
982 	}
983 found:
984 	/*
985 	 * Adjust ip_len to not reflect header,
986 	 * convert offset of this to bytes.
987 	 */
988 	ip->ip_len -= hlen;
989 	if (ip->ip_off & IP_MF) {
990 		/*
991 		 * Make sure that fragments have a data length
992 		 * that's a non-zero multiple of 8 bytes.
993 		 */
994 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
995 			ipstat.ips_toosmall++; /* XXX */
996 			m_freem(m);
997 			goto done;
998 		}
999 		m->m_flags |= M_FRAG;
1000 	} else {
1001 		m->m_flags &= ~M_FRAG;
1002 	}
1003 	ip->ip_off <<= 3;
1004 
1005 	ipstat.ips_fragments++;
1006 	m->m_pkthdr.header = ip;
1007 
1008 	/*
1009 	 * If the hardware has not done csum over this fragment
1010 	 * then csum_data is not valid at all.
1011 	 */
1012 	if ((m->m_pkthdr.csum_flags & (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID))
1013 	    == (CSUM_FRAG_NOT_CHECKED | CSUM_DATA_VALID)) {
1014 		m->m_pkthdr.csum_data = 0;
1015 		m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1016 	}
1017 
1018 	/*
1019 	 * Presence of header sizes in mbufs
1020 	 * would confuse code below.
1021 	 */
1022 	m->m_data += hlen;
1023 	m->m_len -= hlen;
1024 
1025 	/*
1026 	 * If first fragment to arrive, create a reassembly queue.
1027 	 */
1028 	if (fp == NULL) {
1029 		if ((fp = mpipe_alloc_nowait(&ipq_mpipe)) == NULL)
1030 			goto dropfrag;
1031 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
1032 		fragq->nipq++;
1033 		fp->ipq_nfrags = 1;
1034 		fp->ipq_ttl = IPFRAGTTL;
1035 		fp->ipq_p = ip->ip_p;
1036 		fp->ipq_id = ip->ip_id;
1037 		fp->ipq_src = ip->ip_src;
1038 		fp->ipq_dst = ip->ip_dst;
1039 		fp->ipq_frags = m;
1040 		m->m_nextpkt = NULL;
1041 		goto inserted;
1042 	}
1043 	fp->ipq_nfrags++;
1044 
1045 #define	GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
1046 
1047 	/*
1048 	 * Find a segment which begins after this one does.
1049 	 */
1050 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1051 		if (GETIP(q)->ip_off > ip->ip_off)
1052 			break;
1053 	}
1054 
1055 	/*
1056 	 * If there is a preceding segment, it may provide some of
1057 	 * our data already.  If so, drop the data from the incoming
1058 	 * segment.  If it provides all of our data, drop us, otherwise
1059 	 * stick new segment in the proper place.
1060 	 *
1061 	 * If some of the data is dropped from the the preceding
1062 	 * segment, then it's checksum is invalidated.
1063 	 */
1064 	if (p) {
1065 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1066 		if (i > 0) {
1067 			if (i >= ip->ip_len)
1068 				goto dropfrag;
1069 			m_adj(m, i);
1070 			m->m_pkthdr.csum_flags = 0;
1071 			ip->ip_off += i;
1072 			ip->ip_len -= i;
1073 		}
1074 		m->m_nextpkt = p->m_nextpkt;
1075 		p->m_nextpkt = m;
1076 	} else {
1077 		m->m_nextpkt = fp->ipq_frags;
1078 		fp->ipq_frags = m;
1079 	}
1080 
1081 	/*
1082 	 * While we overlap succeeding segments trim them or,
1083 	 * if they are completely covered, dequeue them.
1084 	 */
1085 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1086 	     q = nq) {
1087 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1088 		if (i < GETIP(q)->ip_len) {
1089 			GETIP(q)->ip_len -= i;
1090 			GETIP(q)->ip_off += i;
1091 			m_adj(q, i);
1092 			q->m_pkthdr.csum_flags = 0;
1093 			break;
1094 		}
1095 		nq = q->m_nextpkt;
1096 		m->m_nextpkt = nq;
1097 		ipstat.ips_fragdropped++;
1098 		fp->ipq_nfrags--;
1099 		q->m_nextpkt = NULL;
1100 		m_freem(q);
1101 	}
1102 
1103 inserted:
1104 	/*
1105 	 * Check for complete reassembly and perform frag per packet
1106 	 * limiting.
1107 	 *
1108 	 * Frag limiting is performed here so that the nth frag has
1109 	 * a chance to complete the packet before we drop the packet.
1110 	 * As a result, n+1 frags are actually allowed per packet, but
1111 	 * only n will ever be stored. (n = maxfragsperpacket.)
1112 	 *
1113 	 */
1114 	next = 0;
1115 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1116 		if (GETIP(q)->ip_off != next) {
1117 			if (fp->ipq_nfrags > maxfragsperpacket) {
1118 				ipstat.ips_fragdropped += fp->ipq_nfrags;
1119 				ip_freef(fragq, head, fp);
1120 			}
1121 			goto done;
1122 		}
1123 		next += GETIP(q)->ip_len;
1124 	}
1125 	/* Make sure the last packet didn't have the IP_MF flag */
1126 	if (p->m_flags & M_FRAG) {
1127 		if (fp->ipq_nfrags > maxfragsperpacket) {
1128 			ipstat.ips_fragdropped += fp->ipq_nfrags;
1129 			ip_freef(fragq, head, fp);
1130 		}
1131 		goto done;
1132 	}
1133 
1134 	/*
1135 	 * Reassembly is complete.  Make sure the packet is a sane size.
1136 	 */
1137 	q = fp->ipq_frags;
1138 	ip = GETIP(q);
1139 	if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
1140 		ipstat.ips_toolong++;
1141 		ipstat.ips_fragdropped += fp->ipq_nfrags;
1142 		ip_freef(fragq, head, fp);
1143 		goto done;
1144 	}
1145 
1146 	/*
1147 	 * Concatenate fragments.
1148 	 */
1149 	m = q;
1150 	n = m->m_next;
1151 	m->m_next = NULL;
1152 	m_cat(m, n);
1153 	nq = q->m_nextpkt;
1154 	q->m_nextpkt = NULL;
1155 	for (q = nq; q != NULL; q = nq) {
1156 		nq = q->m_nextpkt;
1157 		q->m_nextpkt = NULL;
1158 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1159 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1160 		m_cat(m, q);
1161 	}
1162 
1163 	/*
1164 	 * Clean up the 1's complement checksum.  Carry over 16 bits must
1165 	 * be added back.  This assumes no more then 65535 packet fragments
1166 	 * were reassembled.  A second carry can also occur (but not a third).
1167 	 */
1168 	m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
1169 				(m->m_pkthdr.csum_data >> 16);
1170 	if (m->m_pkthdr.csum_data > 0xFFFF)
1171 		m->m_pkthdr.csum_data -= 0xFFFF;
1172 
1173 	/*
1174 	 * Create header for new ip packet by
1175 	 * modifying header of first packet;
1176 	 * dequeue and discard fragment reassembly header.
1177 	 * Make header visible.
1178 	 */
1179 	ip->ip_len = next;
1180 	ip->ip_src = fp->ipq_src;
1181 	ip->ip_dst = fp->ipq_dst;
1182 	TAILQ_REMOVE(head, fp, ipq_list);
1183 	fragq->nipq--;
1184 	mpipe_free(&ipq_mpipe, fp);
1185 	m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
1186 	m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
1187 	/* some debugging cruft by sklower, below, will go away soon */
1188 	if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
1189 		int plen = 0;
1190 
1191 		for (n = m; n; n = n->m_next)
1192 			plen += n->m_len;
1193 		m->m_pkthdr.len = plen;
1194 	}
1195 
1196 	/*
1197 	 * Reassembly complete, return the next protocol.
1198 	 *
1199 	 * Be sure to clear M_HASH to force the packet
1200 	 * to be re-characterized.
1201 	 *
1202 	 * Clear M_FRAG, we are no longer a fragment.
1203 	 */
1204 	m->m_flags &= ~(M_HASH | M_FRAG);
1205 
1206 	ipstat.ips_reassembled++;
1207 	return (m);
1208 
1209 dropfrag:
1210 	ipstat.ips_fragdropped++;
1211 	if (fp != NULL)
1212 		fp->ipq_nfrags--;
1213 	m_freem(m);
1214 done:
1215 	return (NULL);
1216 
1217 #undef GETIP
1218 }
1219 
1220 /*
1221  * Free a fragment reassembly header and all
1222  * associated datagrams.
1223  */
1224 static void
1225 ip_freef(struct ipfrag_queue *fragq, struct ipqhead *fhp, struct ipq *fp)
1226 {
1227 	struct mbuf *q;
1228 
1229 	/*
1230 	 * Remove first to protect against blocking
1231 	 */
1232 	TAILQ_REMOVE(fhp, fp, ipq_list);
1233 
1234 	/*
1235 	 * Clean out at our leisure
1236 	 */
1237 	while (fp->ipq_frags) {
1238 		q = fp->ipq_frags;
1239 		fp->ipq_frags = q->m_nextpkt;
1240 		q->m_nextpkt = NULL;
1241 		m_freem(q);
1242 	}
1243 	mpipe_free(&ipq_mpipe, fp);
1244 	fragq->nipq--;
1245 }
1246 
1247 /*
1248  * If a timer expires on a reassembly queue, discard it.
1249  */
1250 static void
1251 ipfrag_timeo_dispatch(netmsg_t nmsg)
1252 {
1253 	struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[mycpuid];
1254 	struct ipq *fp, *fp_temp;
1255 	struct ipqhead *head;
1256 	int i;
1257 
1258 	crit_enter();
1259 	netisr_replymsg(&nmsg->base, 0);  /* reply ASAP */
1260 	crit_exit();
1261 
1262 	if (fragq->nipq == 0)
1263 		goto done;
1264 
1265 	for (i = 0; i < IPREASS_NHASH; i++) {
1266 		head = &fragq->ipq[i];
1267 		TAILQ_FOREACH_MUTABLE(fp, head, ipq_list, fp_temp) {
1268 			if (--fp->ipq_ttl == 0) {
1269 				ipstat.ips_fragtimeout += fp->ipq_nfrags;
1270 				ip_freef(fragq, head, fp);
1271 			}
1272 		}
1273 	}
1274 	/*
1275 	 * If we are over the maximum number of fragments
1276 	 * (due to the limit being lowered), drain off
1277 	 * enough to get down to the new limit.
1278 	 */
1279 	if (maxnipq >= 0 && fragq->nipq > maxnipq) {
1280 		for (i = 0; i < IPREASS_NHASH; i++) {
1281 			head = &fragq->ipq[i];
1282 			while (fragq->nipq > maxnipq && !TAILQ_EMPTY(head)) {
1283 				ipstat.ips_fragdropped +=
1284 				    TAILQ_FIRST(head)->ipq_nfrags;
1285 				ip_freef(fragq, head, TAILQ_FIRST(head));
1286 			}
1287 		}
1288 	}
1289 done:
1290 	callout_reset(&fragq->timeo_ch, IPFRAG_TIMEO, ipfrag_timeo, NULL);
1291 }
1292 
1293 static void
1294 ipfrag_timeo(void *dummy __unused)
1295 {
1296 	struct netmsg_base *msg = &ipfrag_queue_pcpu[mycpuid].timeo_netmsg;
1297 
1298 	crit_enter();
1299 	if (msg->lmsg.ms_flags & MSGF_DONE)
1300 		netisr_sendmsg_oncpu(msg);
1301 	crit_exit();
1302 }
1303 
1304 /*
1305  * Drain off all datagram fragments.
1306  */
1307 static void
1308 ipfrag_drain_oncpu(struct ipfrag_queue *fragq)
1309 {
1310 	struct ipqhead *head;
1311 	int i;
1312 
1313 	for (i = 0; i < IPREASS_NHASH; i++) {
1314 		head = &fragq->ipq[i];
1315 		while (!TAILQ_EMPTY(head)) {
1316 			ipstat.ips_fragdropped += TAILQ_FIRST(head)->ipq_nfrags;
1317 			ip_freef(fragq, head, TAILQ_FIRST(head));
1318 		}
1319 	}
1320 }
1321 
1322 static void
1323 ipfrag_drain_dispatch(netmsg_t nmsg)
1324 {
1325 	struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[mycpuid];
1326 
1327 	crit_enter();
1328 	lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1329 	crit_exit();
1330 
1331 	ipfrag_drain_oncpu(fragq);
1332 	fragq->draining = 0;
1333 }
1334 
1335 static void
1336 ipfrag_drain_ipi(void *arg __unused)
1337 {
1338 	int cpu = mycpuid;
1339 	struct lwkt_msg *msg = &ipfrag_queue_pcpu[cpu].drain_netmsg.lmsg;
1340 
1341 	crit_enter();
1342 	if (msg->ms_flags & MSGF_DONE)
1343 		lwkt_sendmsg_oncpu(netisr_cpuport(cpu), msg);
1344 	crit_exit();
1345 }
1346 
1347 static void
1348 ipfrag_drain(void)
1349 {
1350 	cpumask_t mask;
1351 	int cpu;
1352 
1353 	CPUMASK_ASSBMASK(mask, netisr_ncpus);
1354 	CPUMASK_ANDMASK(mask, smp_active_mask);
1355 
1356 	if (IN_NETISR_NCPUS(mycpuid)) {
1357 		ipfrag_drain_oncpu(&ipfrag_queue_pcpu[mycpuid]);
1358 		CPUMASK_NANDBIT(mask, mycpuid);
1359 	}
1360 
1361 	for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
1362 		struct ipfrag_queue *fragq = &ipfrag_queue_pcpu[cpu];
1363 
1364 		if (!CPUMASK_TESTBIT(mask, cpu))
1365 			continue;
1366 
1367 		if (fragq->nipq == 0 || fragq->draining) {
1368 			/* No fragments or is draining; skip this cpu. */
1369 			CPUMASK_NANDBIT(mask, cpu);
1370 			continue;
1371 		}
1372 		fragq->draining = 1;
1373 	}
1374 
1375 	if (CPUMASK_TESTNZERO(mask))
1376 		lwkt_send_ipiq_mask(mask, ipfrag_drain_ipi, NULL);
1377 }
1378 
1379 void
1380 ip_drain(void)
1381 {
1382 	ipfrag_drain();
1383 	in_rtqdrain();
1384 }
1385 
1386 /*
1387  * Do option processing on a datagram,
1388  * possibly discarding it if bad options are encountered,
1389  * or forwarding it if source-routed.
1390  * The pass argument is used when operating in the IPSTEALTH
1391  * mode to tell what options to process:
1392  * [LS]SRR (pass 0) or the others (pass 1).
1393  * The reason for as many as two passes is that when doing IPSTEALTH,
1394  * non-routing options should be processed only if the packet is for us.
1395  * Returns 1 if packet has been forwarded/freed,
1396  * 0 if the packet should be processed further.
1397  */
1398 static int
1399 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
1400 {
1401 	struct sockaddr_in ipaddr = { sizeof ipaddr, AF_INET };
1402 	struct ip *ip = mtod(m, struct ip *);
1403 	u_char *cp;
1404 	struct in_ifaddr *ia;
1405 	int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB;
1406 	boolean_t forward = FALSE;
1407 	struct in_addr *sin, dst;
1408 	n_time ntime;
1409 
1410 	dst = ip->ip_dst;
1411 	cp = (u_char *)(ip + 1);
1412 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1413 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1414 		opt = cp[IPOPT_OPTVAL];
1415 		if (opt == IPOPT_EOL)
1416 			break;
1417 		if (opt == IPOPT_NOP)
1418 			optlen = 1;
1419 		else {
1420 			if (cnt < IPOPT_OLEN + sizeof(*cp)) {
1421 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1422 				goto bad;
1423 			}
1424 			optlen = cp[IPOPT_OLEN];
1425 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
1426 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1427 				goto bad;
1428 			}
1429 		}
1430 		switch (opt) {
1431 
1432 		default:
1433 			break;
1434 
1435 		/*
1436 		 * Source routing with record.
1437 		 * Find interface with current destination address.
1438 		 * If none on this machine then drop if strictly routed,
1439 		 * or do nothing if loosely routed.
1440 		 * Record interface address and bring up next address
1441 		 * component.  If strictly routed make sure next
1442 		 * address is on directly accessible net.
1443 		 */
1444 		case IPOPT_LSRR:
1445 		case IPOPT_SSRR:
1446 			if (ipstealth && pass > 0)
1447 				break;
1448 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1449 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1450 				goto bad;
1451 			}
1452 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1453 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1454 				goto bad;
1455 			}
1456 			ipaddr.sin_addr = ip->ip_dst;
1457 			ia = (struct in_ifaddr *)
1458 				ifa_ifwithaddr((struct sockaddr *)&ipaddr);
1459 			if (ia == NULL) {
1460 				if (opt == IPOPT_SSRR) {
1461 					type = ICMP_UNREACH;
1462 					code = ICMP_UNREACH_SRCFAIL;
1463 					goto bad;
1464 				}
1465 				if (!ip_dosourceroute)
1466 					goto nosourcerouting;
1467 				/*
1468 				 * Loose routing, and not at next destination
1469 				 * yet; nothing to do except forward.
1470 				 */
1471 				break;
1472 			}
1473 			off--;			/* 0 origin */
1474 			if (off > optlen - (int)sizeof(struct in_addr)) {
1475 				/*
1476 				 * End of source route.  Should be for us.
1477 				 */
1478 				if (!ip_acceptsourceroute)
1479 					goto nosourcerouting;
1480 				save_rte(m, cp, ip->ip_src);
1481 				break;
1482 			}
1483 			if (ipstealth)
1484 				goto dropit;
1485 			if (!ip_dosourceroute) {
1486 				if (ipforwarding) {
1487 					char sbuf[INET_ADDRSTRLEN];
1488 					char dbuf[INET_ADDRSTRLEN];
1489 
1490 					/*
1491 					 * Acting as a router, so generate ICMP
1492 					 */
1493 nosourcerouting:
1494 					log(LOG_WARNING,
1495 					    "attempted source route from %s to %s\n",
1496 					    kinet_ntoa(ip->ip_src, sbuf),
1497 					    kinet_ntoa(ip->ip_dst, dbuf));
1498 					type = ICMP_UNREACH;
1499 					code = ICMP_UNREACH_SRCFAIL;
1500 					goto bad;
1501 				} else {
1502 					/*
1503 					 * Not acting as a router,
1504 					 * so silently drop.
1505 					 */
1506 dropit:
1507 					ipstat.ips_cantforward++;
1508 					m_freem(m);
1509 					return (1);
1510 				}
1511 			}
1512 
1513 			/*
1514 			 * locate outgoing interface
1515 			 */
1516 			memcpy(&ipaddr.sin_addr, cp + off,
1517 			    sizeof ipaddr.sin_addr);
1518 
1519 			if (opt == IPOPT_SSRR) {
1520 #define	INA	struct in_ifaddr *
1521 #define	SA	struct sockaddr *
1522 				if ((ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr))
1523 									== NULL)
1524 					ia = (INA)ifa_ifwithnet((SA)&ipaddr);
1525 			} else {
1526 				ia = ip_rtaddr(ipaddr.sin_addr, NULL);
1527 			}
1528 			if (ia == NULL) {
1529 				type = ICMP_UNREACH;
1530 				code = ICMP_UNREACH_SRCFAIL;
1531 				goto bad;
1532 			}
1533 			ip->ip_dst = ipaddr.sin_addr;
1534 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1535 			    sizeof(struct in_addr));
1536 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1537 			/*
1538 			 * Let ip_intr's mcast routing check handle mcast pkts
1539 			 */
1540 			forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
1541 			break;
1542 
1543 		case IPOPT_RR:
1544 			if (ipstealth && pass == 0)
1545 				break;
1546 			if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
1547 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1548 				goto bad;
1549 			}
1550 			if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
1551 				code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1552 				goto bad;
1553 			}
1554 			/*
1555 			 * If no space remains, ignore.
1556 			 */
1557 			off--;			/* 0 origin */
1558 			if (off > optlen - (int)sizeof(struct in_addr))
1559 				break;
1560 			memcpy(&ipaddr.sin_addr, &ip->ip_dst,
1561 			    sizeof ipaddr.sin_addr);
1562 			/*
1563 			 * locate outgoing interface; if we're the destination,
1564 			 * use the incoming interface (should be same).
1565 			 */
1566 			if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) == NULL &&
1567 			    (ia = ip_rtaddr(ipaddr.sin_addr, NULL)) == NULL) {
1568 				type = ICMP_UNREACH;
1569 				code = ICMP_UNREACH_HOST;
1570 				goto bad;
1571 			}
1572 			memcpy(cp + off, &IA_SIN(ia)->sin_addr,
1573 			    sizeof(struct in_addr));
1574 			cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1575 			break;
1576 
1577 		case IPOPT_TS:
1578 			if (ipstealth && pass == 0)
1579 				break;
1580 			code = cp - (u_char *)ip;
1581 			if (optlen < 4 || optlen > 40) {
1582 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1583 				goto bad;
1584 			}
1585 			if ((off = cp[IPOPT_OFFSET]) < 5) {
1586 				code = &cp[IPOPT_OLEN] - (u_char *)ip;
1587 				goto bad;
1588 			}
1589 			if (off > optlen - (int)sizeof(int32_t)) {
1590 				cp[IPOPT_OFFSET + 1] += (1 << 4);
1591 				if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) {
1592 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1593 					goto bad;
1594 				}
1595 				break;
1596 			}
1597 			off--;				/* 0 origin */
1598 			sin = (struct in_addr *)(cp + off);
1599 			switch (cp[IPOPT_OFFSET + 1] & 0x0f) {
1600 
1601 			case IPOPT_TS_TSONLY:
1602 				break;
1603 
1604 			case IPOPT_TS_TSANDADDR:
1605 				if (off + sizeof(n_time) +
1606 				    sizeof(struct in_addr) > optlen) {
1607 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1608 					goto bad;
1609 				}
1610 				ipaddr.sin_addr = dst;
1611 				ia = (INA)ifaof_ifpforaddr((SA)&ipaddr,
1612 							    m->m_pkthdr.rcvif);
1613 				if (ia == NULL)
1614 					continue;
1615 				memcpy(sin, &IA_SIN(ia)->sin_addr,
1616 				    sizeof(struct in_addr));
1617 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1618 				off += sizeof(struct in_addr);
1619 				break;
1620 
1621 			case IPOPT_TS_PRESPEC:
1622 				if (off + sizeof(n_time) +
1623 				    sizeof(struct in_addr) > optlen) {
1624 					code = &cp[IPOPT_OFFSET] - (u_char *)ip;
1625 					goto bad;
1626 				}
1627 				memcpy(&ipaddr.sin_addr, sin,
1628 				    sizeof(struct in_addr));
1629 				if (ifa_ifwithaddr((SA)&ipaddr) == NULL)
1630 					continue;
1631 				cp[IPOPT_OFFSET] += sizeof(struct in_addr);
1632 				off += sizeof(struct in_addr);
1633 				break;
1634 
1635 			default:
1636 				code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip;
1637 				goto bad;
1638 			}
1639 			ntime = iptime();
1640 			memcpy(cp + off, &ntime, sizeof(n_time));
1641 			cp[IPOPT_OFFSET] += sizeof(n_time);
1642 		}
1643 	}
1644 	if (forward && ipforwarding) {
1645 		ip_forward(m, TRUE, next_hop);
1646 		return (1);
1647 	}
1648 	return (0);
1649 bad:
1650 	icmp_error(m, type, code, 0, 0);
1651 	ipstat.ips_badoptions++;
1652 	return (1);
1653 }
1654 
1655 /*
1656  * Given address of next destination (final or next hop),
1657  * return internet address info of interface to be used to get there.
1658  */
1659 struct in_ifaddr *
1660 ip_rtaddr(struct in_addr dst, struct route *ro0)
1661 {
1662 	struct route sro, *ro;
1663 	struct sockaddr_in *sin;
1664 	struct in_ifaddr *ia;
1665 
1666 	if (ro0 != NULL) {
1667 		ro = ro0;
1668 	} else {
1669 		bzero(&sro, sizeof(sro));
1670 		ro = &sro;
1671 	}
1672 
1673 	sin = (struct sockaddr_in *)&ro->ro_dst;
1674 
1675 	if (ro->ro_rt == NULL || dst.s_addr != sin->sin_addr.s_addr) {
1676 		if (ro->ro_rt != NULL) {
1677 			RTFREE(ro->ro_rt);
1678 			ro->ro_rt = NULL;
1679 		}
1680 		sin->sin_family = AF_INET;
1681 		sin->sin_len = sizeof *sin;
1682 		sin->sin_addr = dst;
1683 		rtalloc_ign(ro, RTF_PRCLONING);
1684 	}
1685 
1686 	if (ro->ro_rt == NULL)
1687 		return (NULL);
1688 
1689 	ia = ifatoia(ro->ro_rt->rt_ifa);
1690 
1691 	if (ro == &sro)
1692 		RTFREE(ro->ro_rt);
1693 	return ia;
1694 }
1695 
1696 /*
1697  * Save incoming source route for use in replies,
1698  * to be picked up later by ip_srcroute if the receiver is interested.
1699  */
1700 static void
1701 save_rte(struct mbuf *m, u_char *option, struct in_addr dst)
1702 {
1703 	struct m_tag *mtag;
1704 	struct ip_srcrt_opt *opt;
1705 	unsigned olen;
1706 
1707 	mtag = m_tag_get(PACKET_TAG_IPSRCRT, sizeof(*opt), M_NOWAIT);
1708 	if (mtag == NULL)
1709 		return;
1710 	opt = m_tag_data(mtag);
1711 
1712 	olen = option[IPOPT_OLEN];
1713 #ifdef DIAGNOSTIC
1714 	if (ipprintfs)
1715 		kprintf("save_rte: olen %d\n", olen);
1716 #endif
1717 	if (olen > sizeof(opt->ip_srcrt) - (1 + sizeof(dst))) {
1718 		m_tag_free(mtag);
1719 		return;
1720 	}
1721 	bcopy(option, opt->ip_srcrt.srcopt, olen);
1722 	opt->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
1723 	opt->ip_srcrt.dst = dst;
1724 	m_tag_prepend(m, mtag);
1725 }
1726 
1727 /*
1728  * Retrieve incoming source route for use in replies,
1729  * in the same form used by setsockopt.
1730  * The first hop is placed before the options, will be removed later.
1731  */
1732 struct mbuf *
1733 ip_srcroute(struct mbuf *m0)
1734 {
1735 	struct in_addr *p, *q;
1736 	struct mbuf *m;
1737 	struct m_tag *mtag;
1738 	struct ip_srcrt_opt *opt;
1739 
1740 	if (m0 == NULL)
1741 		return NULL;
1742 
1743 	mtag = m_tag_find(m0, PACKET_TAG_IPSRCRT, NULL);
1744 	if (mtag == NULL)
1745 		return NULL;
1746 	opt = m_tag_data(mtag);
1747 
1748 	if (opt->ip_nhops == 0)
1749 		return (NULL);
1750 	m = m_get(M_NOWAIT, MT_HEADER);
1751 	if (m == NULL)
1752 		return (NULL);
1753 
1754 #define	OPTSIZ	(sizeof(opt->ip_srcrt.nop) + sizeof(opt->ip_srcrt.srcopt))
1755 
1756 	/* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
1757 	m->m_len = opt->ip_nhops * sizeof(struct in_addr) +
1758 		   sizeof(struct in_addr) + OPTSIZ;
1759 #ifdef DIAGNOSTIC
1760 	if (ipprintfs) {
1761 		kprintf("ip_srcroute: nhops %d mlen %d",
1762 			opt->ip_nhops, m->m_len);
1763 	}
1764 #endif
1765 
1766 	/*
1767 	 * First save first hop for return route
1768 	 */
1769 	p = &opt->ip_srcrt.route[opt->ip_nhops - 1];
1770 	*(mtod(m, struct in_addr *)) = *p--;
1771 #ifdef DIAGNOSTIC
1772 	if (ipprintfs)
1773 		kprintf(" hops %x", ntohl(mtod(m, struct in_addr *)->s_addr));
1774 #endif
1775 
1776 	/*
1777 	 * Copy option fields and padding (nop) to mbuf.
1778 	 */
1779 	opt->ip_srcrt.nop = IPOPT_NOP;
1780 	opt->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
1781 	memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &opt->ip_srcrt.nop,
1782 	    OPTSIZ);
1783 	q = (struct in_addr *)(mtod(m, caddr_t) +
1784 	    sizeof(struct in_addr) + OPTSIZ);
1785 #undef OPTSIZ
1786 	/*
1787 	 * Record return path as an IP source route,
1788 	 * reversing the path (pointers are now aligned).
1789 	 */
1790 	while (p >= opt->ip_srcrt.route) {
1791 #ifdef DIAGNOSTIC
1792 		if (ipprintfs)
1793 			kprintf(" %x", ntohl(q->s_addr));
1794 #endif
1795 		*q++ = *p--;
1796 	}
1797 	/*
1798 	 * Last hop goes to final destination.
1799 	 */
1800 	*q = opt->ip_srcrt.dst;
1801 	m_tag_delete(m0, mtag);
1802 #ifdef DIAGNOSTIC
1803 	if (ipprintfs)
1804 		kprintf(" %x\n", ntohl(q->s_addr));
1805 #endif
1806 	return (m);
1807 }
1808 
1809 /*
1810  * Strip out IP options.
1811  */
1812 void
1813 ip_stripoptions(struct mbuf *m)
1814 {
1815 	int datalen;
1816 	struct ip *ip = mtod(m, struct ip *);
1817 	caddr_t opts;
1818 	int optlen;
1819 
1820 	optlen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1821 	opts = (caddr_t)(ip + 1);
1822 	datalen = m->m_len - (sizeof(struct ip) + optlen);
1823 	bcopy(opts + optlen, opts, datalen);
1824 	m->m_len -= optlen;
1825 	if (m->m_flags & M_PKTHDR)
1826 		m->m_pkthdr.len -= optlen;
1827 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
1828 }
1829 
1830 u_char inetctlerrmap[PRC_NCMDS] = {
1831 	0,		0,		0,		0,
1832 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1833 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1834 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1835 	0,		0,		0,		0,
1836 	ENOPROTOOPT,	ECONNREFUSED
1837 };
1838 
1839 /*
1840  * Forward a packet.  If some error occurs return the sender
1841  * an icmp packet.  Note we can't always generate a meaningful
1842  * icmp message because icmp doesn't have a large enough repertoire
1843  * of codes and types.
1844  *
1845  * If not forwarding, just drop the packet.  This could be confusing
1846  * if ipforwarding was zero but some routing protocol was advancing
1847  * us as a gateway to somewhere.  However, we must let the routing
1848  * protocol deal with that.
1849  *
1850  * The using_srcrt parameter indicates whether the packet is being forwarded
1851  * via a source route.
1852  */
1853 void
1854 ip_forward(struct mbuf *m, boolean_t using_srcrt, struct sockaddr_in *next_hop)
1855 {
1856 	struct ip *ip = mtod(m, struct ip *);
1857 	struct rtentry *rt;
1858 	struct route fwd_ro;
1859 	int error, type = 0, code = 0, destmtu = 0;
1860 	struct mbuf *mcopy, *mtemp = NULL;
1861 	n_long dest;
1862 	struct in_addr pkt_dst;
1863 
1864 	dest = INADDR_ANY;
1865 	/*
1866 	 * Cache the destination address of the packet; this may be
1867 	 * changed by use of 'ipfw fwd'.
1868 	 */
1869 	pkt_dst = (next_hop != NULL) ? next_hop->sin_addr : ip->ip_dst;
1870 
1871 #ifdef DIAGNOSTIC
1872 	if (ipprintfs)
1873 		kprintf("forward: src %x dst %x ttl %x\n",
1874 		       ip->ip_src.s_addr, pkt_dst.s_addr, ip->ip_ttl);
1875 #endif
1876 
1877 	if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
1878 		ipstat.ips_cantforward++;
1879 		m_freem(m);
1880 		return;
1881 	}
1882 	if (!ipstealth && ip->ip_ttl <= IPTTLDEC) {
1883 		icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0);
1884 		return;
1885 	}
1886 
1887 	bzero(&fwd_ro, sizeof(fwd_ro));
1888 	ip_rtaddr(pkt_dst, &fwd_ro);
1889 	if (fwd_ro.ro_rt == NULL) {
1890 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
1891 		return;
1892 	}
1893 	rt = fwd_ro.ro_rt;
1894 
1895 	if (curthread->td_type == TD_TYPE_NETISR) {
1896 		/*
1897 		 * Save the IP header and at most 8 bytes of the payload,
1898 		 * in case we need to generate an ICMP message to the src.
1899 		 */
1900 		mtemp = ipforward_mtemp[mycpuid];
1901 		KASSERT((mtemp->m_flags & M_EXT) == 0 &&
1902 		    mtemp->m_data == mtemp->m_pktdat &&
1903 		    m_tag_first(mtemp) == NULL,
1904 		    ("ip_forward invalid mtemp1"));
1905 
1906 		if (!m_dup_pkthdr(mtemp, m, M_NOWAIT)) {
1907 			/*
1908 			 * It's probably ok if the pkthdr dup fails (because
1909 			 * the deep copy of the tag chain failed), but for now
1910 			 * be conservative and just discard the copy since
1911 			 * code below may some day want the tags.
1912 			 */
1913 			mtemp = NULL;
1914 		} else {
1915 			mtemp->m_type = m->m_type;
1916 			mtemp->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
1917 			    (int)ip->ip_len);
1918 			mtemp->m_pkthdr.len = mtemp->m_len;
1919 			m_copydata(m, 0, mtemp->m_len, mtod(mtemp, caddr_t));
1920 		}
1921 	}
1922 
1923 	if (!ipstealth)
1924 		ip->ip_ttl -= IPTTLDEC;
1925 
1926 	/*
1927 	 * If forwarding packet using same interface that it came in on,
1928 	 * perhaps should send a redirect to sender to shortcut a hop.
1929 	 * Only send redirect if source is sending directly to us,
1930 	 * and if packet was not source routed (or has any options).
1931 	 * Also, don't send redirect if forwarding using a default route
1932 	 * or a route modified by a redirect.
1933 	 */
1934 	if (rt->rt_ifp == m->m_pkthdr.rcvif &&
1935 	    !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
1936 	    satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
1937 	    ipsendredirects && !using_srcrt && next_hop == NULL) {
1938 		u_long src = ntohl(ip->ip_src.s_addr);
1939 		struct in_ifaddr *rt_ifa = (struct in_ifaddr *)rt->rt_ifa;
1940 
1941 		if (rt_ifa != NULL &&
1942 		    (src & rt_ifa->ia_subnetmask) == rt_ifa->ia_subnet) {
1943 			if (rt->rt_flags & RTF_GATEWAY)
1944 				dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
1945 			else
1946 				dest = pkt_dst.s_addr;
1947 			/*
1948 			 * Router requirements says to only send
1949 			 * host redirects.
1950 			 */
1951 			type = ICMP_REDIRECT;
1952 			code = ICMP_REDIRECT_HOST;
1953 #ifdef DIAGNOSTIC
1954 			if (ipprintfs)
1955 				kprintf("redirect (%d) to %x\n", code, dest);
1956 #endif
1957 		}
1958 	}
1959 
1960 	error = ip_output(m, NULL, &fwd_ro, IP_FORWARDING, NULL, NULL);
1961 	if (error == 0) {
1962 		ipstat.ips_forward++;
1963 		if (type == 0) {
1964 			if (mtemp)
1965 				ipflow_create(&fwd_ro, mtemp);
1966 			goto done;
1967 		}
1968 		ipstat.ips_redirectsent++;
1969 	} else {
1970 		ipstat.ips_cantforward++;
1971 	}
1972 
1973 	if (mtemp == NULL)
1974 		goto done;
1975 
1976 	/*
1977 	 * Errors that do not require generating ICMP message
1978 	 */
1979 	switch (error) {
1980 	case ENOBUFS:
1981 		/*
1982 		 * A router should not generate ICMP_SOURCEQUENCH as
1983 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1984 		 * Source quench could be a big problem under DoS attacks,
1985 		 * or if the underlying interface is rate-limited.
1986 		 * Those who need source quench packets may re-enable them
1987 		 * via the net.inet.ip.sendsourcequench sysctl.
1988 		 */
1989 		if (!ip_sendsourcequench)
1990 			goto done;
1991 		break;
1992 
1993 	case EACCES:			/* ipfw denied packet */
1994 		goto done;
1995 	}
1996 
1997 	KASSERT((mtemp->m_flags & M_EXT) == 0 &&
1998 	    mtemp->m_data == mtemp->m_pktdat,
1999 	    ("ip_forward invalid mtemp2"));
2000 	mcopy = m_copym(mtemp, 0, mtemp->m_len, M_NOWAIT);
2001 	if (mcopy == NULL)
2002 		goto done;
2003 
2004 	/*
2005 	 * Send ICMP message.
2006 	 */
2007 	switch (error) {
2008 	case 0:				/* forwarded, but need redirect */
2009 		/* type, code set above */
2010 		break;
2011 
2012 	case ENETUNREACH:		/* shouldn't happen, checked above */
2013 	case EHOSTUNREACH:
2014 	case ENETDOWN:
2015 	case EHOSTDOWN:
2016 	default:
2017 		type = ICMP_UNREACH;
2018 		code = ICMP_UNREACH_HOST;
2019 		break;
2020 
2021 	case EMSGSIZE:
2022 		type = ICMP_UNREACH;
2023 		code = ICMP_UNREACH_NEEDFRAG;
2024 		if (fwd_ro.ro_rt != NULL)
2025 			destmtu = fwd_ro.ro_rt->rt_ifp->if_mtu;
2026 		ipstat.ips_cantfrag++;
2027 		break;
2028 
2029 	case ENOBUFS:
2030 		type = ICMP_SOURCEQUENCH;
2031 		code = 0;
2032 		break;
2033 
2034 	case EACCES:			/* ipfw denied packet */
2035 		panic("ip_forward EACCES should not reach");
2036 	}
2037 	icmp_error(mcopy, type, code, dest, destmtu);
2038 done:
2039 	if (mtemp != NULL)
2040 		m_tag_delete_chain(mtemp);
2041 	if (fwd_ro.ro_rt != NULL)
2042 		RTFREE(fwd_ro.ro_rt);
2043 }
2044 
2045 void
2046 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
2047 	       struct mbuf *m)
2048 {
2049 	if (inp->inp_socket->so_options & SO_TIMESTAMP) {
2050 		struct timeval tv;
2051 
2052 		microtime(&tv);
2053 		*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
2054 		    SCM_TIMESTAMP, SOL_SOCKET);
2055 		if (*mp)
2056 			mp = &(*mp)->m_next;
2057 	}
2058 	if (inp->inp_flags & INP_RECVDSTADDR) {
2059 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
2060 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
2061 		if (*mp)
2062 			mp = &(*mp)->m_next;
2063 	}
2064 	if (inp->inp_flags & INP_RECVTTL) {
2065 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
2066 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
2067 		if (*mp)
2068 			mp = &(*mp)->m_next;
2069 	}
2070 #ifdef notyet
2071 	/* XXX
2072 	 * Moving these out of udp_input() made them even more broken
2073 	 * than they already were.
2074 	 */
2075 	/* options were tossed already */
2076 	if (inp->inp_flags & INP_RECVOPTS) {
2077 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
2078 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
2079 		if (*mp)
2080 			mp = &(*mp)->m_next;
2081 	}
2082 	/* ip_srcroute doesn't do what we want here, need to fix */
2083 	if (inp->inp_flags & INP_RECVRETOPTS) {
2084 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
2085 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
2086 		if (*mp)
2087 			mp = &(*mp)->m_next;
2088 	}
2089 #endif
2090 	if (inp->inp_flags & INP_RECVIF) {
2091 		struct ifnet *ifp;
2092 		struct sdlbuf {
2093 			struct sockaddr_dl sdl;
2094 			u_char	pad[32];
2095 		} sdlbuf;
2096 		struct sockaddr_dl *sdp;
2097 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
2098 
2099 		if (((ifp = m->m_pkthdr.rcvif)) &&
2100 		    ((ifp->if_index != 0) && (ifp->if_index <= if_index))) {
2101 			sdp = IF_LLSOCKADDR(ifp);
2102 			/*
2103 			 * Change our mind and don't try copy.
2104 			 */
2105 			if ((sdp->sdl_family != AF_LINK) ||
2106 			    (sdp->sdl_len > sizeof(sdlbuf))) {
2107 				goto makedummy;
2108 			}
2109 			bcopy(sdp, sdl2, sdp->sdl_len);
2110 		} else {
2111 makedummy:
2112 			sdl2->sdl_len =
2113 			    offsetof(struct sockaddr_dl, sdl_data[0]);
2114 			sdl2->sdl_family = AF_LINK;
2115 			sdl2->sdl_index = 0;
2116 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
2117 		}
2118 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
2119 			IP_RECVIF, IPPROTO_IP);
2120 		if (*mp)
2121 			mp = &(*mp)->m_next;
2122 	}
2123 }
2124 
2125 /*
2126  * XXX these routines are called from the upper part of the kernel.
2127  *
2128  * They could also be moved to ip_mroute.c, since all the RSVP
2129  *  handling is done there already.
2130  */
2131 int
2132 ip_rsvp_init(struct socket *so)
2133 {
2134 	if (so->so_type != SOCK_RAW ||
2135 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
2136 		return EOPNOTSUPP;
2137 
2138 	if (ip_rsvpd != NULL)
2139 		return EADDRINUSE;
2140 
2141 	ip_rsvpd = so;
2142 	/*
2143 	 * This may seem silly, but we need to be sure we don't over-increment
2144 	 * the RSVP counter, in case something slips up.
2145 	 */
2146 	if (!ip_rsvp_on) {
2147 		ip_rsvp_on = 1;
2148 		rsvp_on++;
2149 	}
2150 
2151 	return 0;
2152 }
2153 
2154 int
2155 ip_rsvp_done(void)
2156 {
2157 	ip_rsvpd = NULL;
2158 	/*
2159 	 * This may seem silly, but we need to be sure we don't over-decrement
2160 	 * the RSVP counter, in case something slips up.
2161 	 */
2162 	if (ip_rsvp_on) {
2163 		ip_rsvp_on = 0;
2164 		rsvp_on--;
2165 	}
2166 	return 0;
2167 }
2168 
2169 int
2170 rsvp_input(struct mbuf **mp, int *offp, int proto)
2171 {
2172 	struct mbuf *m = *mp;
2173 
2174 	*mp = NULL;
2175 
2176 	if (rsvp_input_p) { /* call the real one if loaded */
2177 		*mp = m;
2178 		rsvp_input_p(mp, offp, proto);
2179 		return(IPPROTO_DONE);
2180 	}
2181 
2182 	/* Can still get packets with rsvp_on = 0 if there is a local member
2183 	 * of the group to which the RSVP packet is addressed.  But in this
2184 	 * case we want to throw the packet away.
2185 	 */
2186 
2187 	if (!rsvp_on) {
2188 		m_freem(m);
2189 		return(IPPROTO_DONE);
2190 	}
2191 
2192 	if (ip_rsvpd != NULL) {
2193 		*mp = m;
2194 		rip_input(mp, offp, proto);
2195 		return(IPPROTO_DONE);
2196 	}
2197 	/* Drop the packet */
2198 	m_freem(m);
2199 	return(IPPROTO_DONE);
2200 }
2201