xref: /freebsd/sys/netinet/ip_input.c (revision f0a75d274af375d15b97b830966b99a02b7db911)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
30  * $FreeBSD$
31  */
32 
33 #include "opt_bootp.h"
34 #include "opt_ipfw.h"
35 #include "opt_ipstealth.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_carp.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/mbuf.h>
44 #include <sys/malloc.h>
45 #include <sys/domain.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/syslog.h>
51 #include <sys/sysctl.h>
52 
53 #include <net/pfil.h>
54 #include <net/if.h>
55 #include <net/if_types.h>
56 #include <net/if_var.h>
57 #include <net/if_dl.h>
58 #include <net/route.h>
59 #include <net/netisr.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_icmp.h>
68 #include <netinet/ip_options.h>
69 #include <machine/in_cksum.h>
70 #ifdef DEV_CARP
71 #include <netinet/ip_carp.h>
72 #endif
73 #if defined(IPSEC) || defined(FAST_IPSEC)
74 #include <netinet/ip_ipsec.h>
75 #endif /* IPSEC */
76 
77 #include <sys/socketvar.h>
78 
79 /* XXX: Temporary until ipfw_ether and ipfw_bridge are converted. */
80 #include <netinet/ip_fw.h>
81 #include <netinet/ip_dummynet.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 int rsvp_on = 0;
86 
87 int	ipforwarding = 0;
88 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
89     &ipforwarding, 0, "Enable IP forwarding between interfaces");
90 
91 static int	ipsendredirects = 1; /* XXX */
92 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
93     &ipsendredirects, 0, "Enable sending IP redirects");
94 
95 int	ip_defttl = IPDEFTTL;
96 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
97     &ip_defttl, 0, "Maximum TTL on IP packets");
98 
99 static int	ip_keepfaith = 0;
100 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
101     &ip_keepfaith,	0,
102     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
103 
104 static int	ip_sendsourcequench = 0;
105 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
106     &ip_sendsourcequench, 0,
107     "Enable the transmission of source quench packets");
108 
109 int	ip_do_randomid = 0;
110 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
111     &ip_do_randomid, 0,
112     "Assign random ip_id values");
113 
114 /*
115  * XXX - Setting ip_checkinterface mostly implements the receive side of
116  * the Strong ES model described in RFC 1122, but since the routing table
117  * and transmit implementation do not implement the Strong ES model,
118  * setting this to 1 results in an odd hybrid.
119  *
120  * XXX - ip_checkinterface currently must be disabled if you use ipnat
121  * to translate the destination address to another local interface.
122  *
123  * XXX - ip_checkinterface must be disabled if you add IP aliases
124  * to the loopback interface instead of the interface where the
125  * packets for those addresses are received.
126  */
127 static int	ip_checkinterface = 0;
128 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
129     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
130 
131 struct pfil_head inet_pfil_hook;	/* Packet filter hooks */
132 
133 static struct	ifqueue ipintrq;
134 static int	ipqmaxlen = IFQ_MAXLEN;
135 
136 extern	struct domain inetdomain;
137 extern	struct protosw inetsw[];
138 u_char	ip_protox[IPPROTO_MAX];
139 struct	in_ifaddrhead in_ifaddrhead; 		/* first inet address */
140 struct	in_ifaddrhashhead *in_ifaddrhashtbl;	/* inet addr hash table  */
141 u_long 	in_ifaddrhmask;				/* mask for hash table */
142 
143 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
144     &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
145 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
146     &ipintrq.ifq_drops, 0,
147     "Number of packets dropped from the IP input queue");
148 
149 struct ipstat ipstat;
150 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
151     &ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
152 
153 /*
154  * IP datagram reassembly.
155  */
156 #define IPREASS_NHASH_LOG2      6
157 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
158 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
159 #define IPREASS_HASH(x,y) \
160 	(((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
161 
162 static uma_zone_t ipq_zone;
163 static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
164 static struct mtx ipqlock;
165 
166 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
167 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
168 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
169 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
170 
171 static void	maxnipq_update(void);
172 static void	ipq_zone_change(void *);
173 
174 static int	maxnipq;	/* Administrative limit on # reass queues. */
175 static int	nipq = 0;	/* Total # of reass queues */
176 SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
177     &nipq, 0, "Current number of IPv4 fragment reassembly queue entries");
178 
179 static int	maxfragsperpacket;
180 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
181     &maxfragsperpacket, 0,
182     "Maximum number of IPv4 fragments allowed per packet");
183 
184 struct callout	ipport_tick_callout;
185 
186 #ifdef IPCTL_DEFMTU
187 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
188     &ip_mtu, 0, "Default MTU");
189 #endif
190 
191 #ifdef IPSTEALTH
192 int	ipstealth = 0;
193 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
194     &ipstealth, 0, "IP stealth mode, no TTL decrementation on forwarding");
195 #endif
196 
197 /*
198  * ipfw_ether and ipfw_bridge hooks.
199  * XXX: Temporary until those are converted to pfil_hooks as well.
200  */
201 ip_fw_chk_t *ip_fw_chk_ptr = NULL;
202 ip_dn_io_t *ip_dn_io_ptr = NULL;
203 int fw_one_pass = 1;
204 
205 static void	ip_freef(struct ipqhead *, struct ipq *);
206 
207 /*
208  * IP initialization: fill in IP protocol switch table.
209  * All protocols not implemented in kernel go to raw IP protocol handler.
210  */
211 void
212 ip_init()
213 {
214 	register struct protosw *pr;
215 	register int i;
216 
217 	TAILQ_INIT(&in_ifaddrhead);
218 	in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
219 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
220 	if (pr == NULL)
221 		panic("ip_init: PF_INET not found");
222 
223 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
224 	for (i = 0; i < IPPROTO_MAX; i++)
225 		ip_protox[i] = pr - inetsw;
226 	/*
227 	 * Cycle through IP protocols and put them into the appropriate place
228 	 * in ip_protox[].
229 	 */
230 	for (pr = inetdomain.dom_protosw;
231 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
232 		if (pr->pr_domain->dom_family == PF_INET &&
233 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
234 			/* Be careful to only index valid IP protocols. */
235 			if (pr->pr_protocol < IPPROTO_MAX)
236 				ip_protox[pr->pr_protocol] = pr - inetsw;
237 		}
238 
239 	/* Initialize packet filter hooks. */
240 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
241 	inet_pfil_hook.ph_af = AF_INET;
242 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
243 		printf("%s: WARNING: unable to register pfil hook, "
244 			"error %d\n", __func__, i);
245 
246 	/* Initialize IP reassembly queue. */
247 	IPQ_LOCK_INIT();
248 	for (i = 0; i < IPREASS_NHASH; i++)
249 	    TAILQ_INIT(&ipq[i]);
250 	maxnipq = nmbclusters / 32;
251 	maxfragsperpacket = 16;
252 	ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
253 	    NULL, UMA_ALIGN_PTR, 0);
254 	maxnipq_update();
255 
256 	/* Start ipport_tick. */
257 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
258 	ipport_tick(NULL);
259 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
260 		SHUTDOWN_PRI_DEFAULT);
261 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
262 		NULL, EVENTHANDLER_PRI_ANY);
263 
264 	/* Initialize various other remaining things. */
265 	ip_id = time_second & 0xffff;
266 	ipintrq.ifq_maxlen = ipqmaxlen;
267 	mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
268 	netisr_register(NETISR_IP, ip_input, &ipintrq, NETISR_MPSAFE);
269 }
270 
271 void ip_fini(xtp)
272 	void *xtp;
273 {
274 	callout_stop(&ipport_tick_callout);
275 }
276 
277 /*
278  * Ip input routine.  Checksum and byte swap header.  If fragmented
279  * try to reassemble.  Process options.  Pass to next level.
280  */
281 void
282 ip_input(struct mbuf *m)
283 {
284 	struct ip *ip = NULL;
285 	struct in_ifaddr *ia = NULL;
286 	struct ifaddr *ifa;
287 	int    checkif, hlen = 0;
288 	u_short sum;
289 	int dchg = 0;				/* dest changed after fw */
290 	struct in_addr odst;			/* original dst address */
291 
292   	M_ASSERTPKTHDR(m);
293 
294 	if (m->m_flags & M_FASTFWD_OURS) {
295 		/*
296 		 * Firewall or NAT changed destination to local.
297 		 * We expect ip_len and ip_off to be in host byte order.
298 		 */
299 		m->m_flags &= ~M_FASTFWD_OURS;
300 		/* Set up some basics that will be used later. */
301 		ip = mtod(m, struct ip *);
302 		hlen = ip->ip_hl << 2;
303   		goto ours;
304   	}
305 
306 	ipstat.ips_total++;
307 
308 	if (m->m_pkthdr.len < sizeof(struct ip))
309 		goto tooshort;
310 
311 	if (m->m_len < sizeof (struct ip) &&
312 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
313 		ipstat.ips_toosmall++;
314 		return;
315 	}
316 	ip = mtod(m, struct ip *);
317 
318 	if (ip->ip_v != IPVERSION) {
319 		ipstat.ips_badvers++;
320 		goto bad;
321 	}
322 
323 	hlen = ip->ip_hl << 2;
324 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
325 		ipstat.ips_badhlen++;
326 		goto bad;
327 	}
328 	if (hlen > m->m_len) {
329 		if ((m = m_pullup(m, hlen)) == NULL) {
330 			ipstat.ips_badhlen++;
331 			return;
332 		}
333 		ip = mtod(m, struct ip *);
334 	}
335 
336 	/* 127/8 must not appear on wire - RFC1122 */
337 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
338 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
339 		if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
340 			ipstat.ips_badaddr++;
341 			goto bad;
342 		}
343 	}
344 
345 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
346 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
347 	} else {
348 		if (hlen == sizeof(struct ip)) {
349 			sum = in_cksum_hdr(ip);
350 		} else {
351 			sum = in_cksum(m, hlen);
352 		}
353 	}
354 	if (sum) {
355 		ipstat.ips_badsum++;
356 		goto bad;
357 	}
358 
359 #ifdef ALTQ
360 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
361 		/* packet is dropped by traffic conditioner */
362 		return;
363 #endif
364 
365 	/*
366 	 * Convert fields to host representation.
367 	 */
368 	ip->ip_len = ntohs(ip->ip_len);
369 	if (ip->ip_len < hlen) {
370 		ipstat.ips_badlen++;
371 		goto bad;
372 	}
373 	ip->ip_off = ntohs(ip->ip_off);
374 
375 	/*
376 	 * Check that the amount of data in the buffers
377 	 * is as at least much as the IP header would have us expect.
378 	 * Trim mbufs if longer than we expect.
379 	 * Drop packet if shorter than we expect.
380 	 */
381 	if (m->m_pkthdr.len < ip->ip_len) {
382 tooshort:
383 		ipstat.ips_tooshort++;
384 		goto bad;
385 	}
386 	if (m->m_pkthdr.len > ip->ip_len) {
387 		if (m->m_len == m->m_pkthdr.len) {
388 			m->m_len = ip->ip_len;
389 			m->m_pkthdr.len = ip->ip_len;
390 		} else
391 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
392 	}
393 #if defined(IPSEC) || defined(FAST_IPSEC)
394 	/*
395 	 * Bypass packet filtering for packets from a tunnel (gif).
396 	 */
397 	if (ip_ipsec_filtergif(m))
398 		goto passin;
399 #endif /* IPSEC */
400 
401 	/*
402 	 * Run through list of hooks for input packets.
403 	 *
404 	 * NB: Beware of the destination address changing (e.g.
405 	 *     by NAT rewriting).  When this happens, tell
406 	 *     ip_forward to do the right thing.
407 	 */
408 
409 	/* Jump over all PFIL processing if hooks are not active. */
410 	if (!PFIL_HOOKED(&inet_pfil_hook))
411 		goto passin;
412 
413 	odst = ip->ip_dst;
414 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
415 	    PFIL_IN, NULL) != 0)
416 		return;
417 	if (m == NULL)			/* consumed by filter */
418 		return;
419 
420 	ip = mtod(m, struct ip *);
421 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
422 
423 #ifdef IPFIREWALL_FORWARD
424 	if (m->m_flags & M_FASTFWD_OURS) {
425 		m->m_flags &= ~M_FASTFWD_OURS;
426 		goto ours;
427 	}
428 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
429 		/*
430 		 * Directly ship on the packet.  This allows to forward packets
431 		 * that were destined for us to some other directly connected
432 		 * host.
433 		 */
434 		ip_forward(m, dchg);
435 		return;
436 	}
437 #endif /* IPFIREWALL_FORWARD */
438 
439 passin:
440 	/*
441 	 * Process options and, if not destined for us,
442 	 * ship it on.  ip_dooptions returns 1 when an
443 	 * error was detected (causing an icmp message
444 	 * to be sent and the original packet to be freed).
445 	 */
446 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
447 		return;
448 
449         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
450          * matter if it is destined to another node, or whether it is
451          * a multicast one, RSVP wants it! and prevents it from being forwarded
452          * anywhere else. Also checks if the rsvp daemon is running before
453 	 * grabbing the packet.
454          */
455 	if (rsvp_on && ip->ip_p==IPPROTO_RSVP)
456 		goto ours;
457 
458 	/*
459 	 * Check our list of addresses, to see if the packet is for us.
460 	 * If we don't have any addresses, assume any unicast packet
461 	 * we receive might be for us (and let the upper layers deal
462 	 * with it).
463 	 */
464 	if (TAILQ_EMPTY(&in_ifaddrhead) &&
465 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
466 		goto ours;
467 
468 	/*
469 	 * Enable a consistency check between the destination address
470 	 * and the arrival interface for a unicast packet (the RFC 1122
471 	 * strong ES model) if IP forwarding is disabled and the packet
472 	 * is not locally generated and the packet is not subject to
473 	 * 'ipfw fwd'.
474 	 *
475 	 * XXX - Checking also should be disabled if the destination
476 	 * address is ipnat'ed to a different interface.
477 	 *
478 	 * XXX - Checking is incompatible with IP aliases added
479 	 * to the loopback interface instead of the interface where
480 	 * the packets are received.
481 	 *
482 	 * XXX - This is the case for carp vhost IPs as well so we
483 	 * insert a workaround. If the packet got here, we already
484 	 * checked with carp_iamatch() and carp_forus().
485 	 */
486 	checkif = ip_checkinterface && (ipforwarding == 0) &&
487 	    m->m_pkthdr.rcvif != NULL &&
488 	    ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) &&
489 #ifdef DEV_CARP
490 	    !m->m_pkthdr.rcvif->if_carp &&
491 #endif
492 	    (dchg == 0);
493 
494 	/*
495 	 * Check for exact addresses in the hash bucket.
496 	 */
497 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
498 		/*
499 		 * If the address matches, verify that the packet
500 		 * arrived via the correct interface if checking is
501 		 * enabled.
502 		 */
503 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
504 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
505 			goto ours;
506 	}
507 	/*
508 	 * Check for broadcast addresses.
509 	 *
510 	 * Only accept broadcast packets that arrive via the matching
511 	 * interface.  Reception of forwarded directed broadcasts would
512 	 * be handled via ip_forward() and ether_output() with the loopback
513 	 * into the stack for SIMPLEX interfaces handled by ether_output().
514 	 */
515 	if (m->m_pkthdr.rcvif != NULL &&
516 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
517 	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
518 			if (ifa->ifa_addr->sa_family != AF_INET)
519 				continue;
520 			ia = ifatoia(ifa);
521 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
522 			    ip->ip_dst.s_addr)
523 				goto ours;
524 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
525 				goto ours;
526 #ifdef BOOTP_COMPAT
527 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
528 				goto ours;
529 #endif
530 		}
531 	}
532 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
533 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
534 		ipstat.ips_cantforward++;
535 		m_freem(m);
536 		return;
537 	}
538 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
539 		struct in_multi *inm;
540 		if (ip_mrouter) {
541 			/*
542 			 * If we are acting as a multicast router, all
543 			 * incoming multicast packets are passed to the
544 			 * kernel-level multicast forwarding function.
545 			 * The packet is returned (relatively) intact; if
546 			 * ip_mforward() returns a non-zero value, the packet
547 			 * must be discarded, else it may be accepted below.
548 			 */
549 			if (ip_mforward &&
550 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
551 				ipstat.ips_cantforward++;
552 				m_freem(m);
553 				return;
554 			}
555 
556 			/*
557 			 * The process-level routing daemon needs to receive
558 			 * all multicast IGMP packets, whether or not this
559 			 * host belongs to their destination groups.
560 			 */
561 			if (ip->ip_p == IPPROTO_IGMP)
562 				goto ours;
563 			ipstat.ips_forward++;
564 		}
565 		/*
566 		 * See if we belong to the destination multicast group on the
567 		 * arrival interface.
568 		 */
569 		IN_MULTI_LOCK();
570 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
571 		IN_MULTI_UNLOCK();
572 		if (inm == NULL) {
573 			ipstat.ips_notmember++;
574 			m_freem(m);
575 			return;
576 		}
577 		goto ours;
578 	}
579 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
580 		goto ours;
581 	if (ip->ip_dst.s_addr == INADDR_ANY)
582 		goto ours;
583 
584 	/*
585 	 * FAITH(Firewall Aided Internet Translator)
586 	 */
587 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
588 		if (ip_keepfaith) {
589 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
590 				goto ours;
591 		}
592 		m_freem(m);
593 		return;
594 	}
595 
596 	/*
597 	 * Not for us; forward if possible and desirable.
598 	 */
599 	if (ipforwarding == 0) {
600 		ipstat.ips_cantforward++;
601 		m_freem(m);
602 	} else {
603 #if defined(IPSEC) || defined(FAST_IPSEC)
604 		if (ip_ipsec_fwd(m))
605 			goto bad;
606 #endif /* IPSEC */
607 		ip_forward(m, dchg);
608 	}
609 	return;
610 
611 ours:
612 #ifdef IPSTEALTH
613 	/*
614 	 * IPSTEALTH: Process non-routing options only
615 	 * if the packet is destined for us.
616 	 */
617 	if (ipstealth && hlen > sizeof (struct ip) &&
618 	    ip_dooptions(m, 1))
619 		return;
620 #endif /* IPSTEALTH */
621 
622 	/* Count the packet in the ip address stats */
623 	if (ia != NULL) {
624 		ia->ia_ifa.if_ipackets++;
625 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
626 	}
627 
628 	/*
629 	 * Attempt reassembly; if it succeeds, proceed.
630 	 * ip_reass() will return a different mbuf.
631 	 */
632 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
633 		m = ip_reass(m);
634 		if (m == NULL)
635 			return;
636 		ip = mtod(m, struct ip *);
637 		/* Get the header length of the reassembled packet */
638 		hlen = ip->ip_hl << 2;
639 	}
640 
641 	/*
642 	 * Further protocols expect the packet length to be w/o the
643 	 * IP header.
644 	 */
645 	ip->ip_len -= hlen;
646 
647 #if defined(IPSEC) || defined(FAST_IPSEC)
648 	/*
649 	 * enforce IPsec policy checking if we are seeing last header.
650 	 * note that we do not visit this with protocols with pcb layer
651 	 * code - like udp/tcp/raw ip.
652 	 */
653 	if (ip_ipsec_input(m))
654 		goto bad;
655 #endif /* IPSEC */
656 
657 	/*
658 	 * Switch out to protocol's input routine.
659 	 */
660 	ipstat.ips_delivered++;
661 
662 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
663 	return;
664 bad:
665 	m_freem(m);
666 }
667 
668 /*
669  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
670  * max has slightly different semantics than the sysctl, for historical
671  * reasons.
672  */
673 static void
674 maxnipq_update(void)
675 {
676 
677 	/*
678 	 * -1 for unlimited allocation.
679 	 */
680 	if (maxnipq < 0)
681 		uma_zone_set_max(ipq_zone, 0);
682 	/*
683 	 * Positive number for specific bound.
684 	 */
685 	if (maxnipq > 0)
686 		uma_zone_set_max(ipq_zone, maxnipq);
687 	/*
688 	 * Zero specifies no further fragment queue allocation -- set the
689 	 * bound very low, but rely on implementation elsewhere to actually
690 	 * prevent allocation and reclaim current queues.
691 	 */
692 	if (maxnipq == 0)
693 		uma_zone_set_max(ipq_zone, 1);
694 }
695 
696 static void
697 ipq_zone_change(void *tag)
698 {
699 
700 	if (maxnipq > 0 && maxnipq < (nmbclusters / 32)) {
701 		maxnipq = nmbclusters / 32;
702 		maxnipq_update();
703 	}
704 }
705 
706 static int
707 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
708 {
709 	int error, i;
710 
711 	i = maxnipq;
712 	error = sysctl_handle_int(oidp, &i, 0, req);
713 	if (error || !req->newptr)
714 		return (error);
715 
716 	/*
717 	 * XXXRW: Might be a good idea to sanity check the argument and place
718 	 * an extreme upper bound.
719 	 */
720 	if (i < -1)
721 		return (EINVAL);
722 	maxnipq = i;
723 	maxnipq_update();
724 	return (0);
725 }
726 
727 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
728     NULL, 0, sysctl_maxnipq, "I",
729     "Maximum number of IPv4 fragment reassembly queue entries");
730 
731 /*
732  * Take incoming datagram fragment and try to reassemble it into
733  * whole datagram.  If the argument is the first fragment or one
734  * in between the function will return NULL and store the mbuf
735  * in the fragment chain.  If the argument is the last fragment
736  * the packet will be reassembled and the pointer to the new
737  * mbuf returned for further processing.  Only m_tags attached
738  * to the first packet/fragment are preserved.
739  * The IP header is *NOT* adjusted out of iplen.
740  */
741 
742 struct mbuf *
743 ip_reass(struct mbuf *m)
744 {
745 	struct ip *ip;
746 	struct mbuf *p, *q, *nq, *t;
747 	struct ipq *fp = NULL;
748 	struct ipqhead *head;
749 	int i, hlen, next;
750 	u_int8_t ecn, ecn0;
751 	u_short hash;
752 
753 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
754 	if (maxnipq == 0 || maxfragsperpacket == 0) {
755 		ipstat.ips_fragments++;
756 		ipstat.ips_fragdropped++;
757 		m_freem(m);
758 		return (NULL);
759 	}
760 
761 	ip = mtod(m, struct ip *);
762 	hlen = ip->ip_hl << 2;
763 
764 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
765 	head = &ipq[hash];
766 	IPQ_LOCK();
767 
768 	/*
769 	 * Look for queue of fragments
770 	 * of this datagram.
771 	 */
772 	TAILQ_FOREACH(fp, head, ipq_list)
773 		if (ip->ip_id == fp->ipq_id &&
774 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
775 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
776 #ifdef MAC
777 		    mac_fragment_match(m, fp) &&
778 #endif
779 		    ip->ip_p == fp->ipq_p)
780 			goto found;
781 
782 	fp = NULL;
783 
784 	/*
785 	 * Attempt to trim the number of allocated fragment queues if it
786 	 * exceeds the administrative limit.
787 	 */
788 	if ((nipq > maxnipq) && (maxnipq > 0)) {
789 		/*
790 		 * drop something from the tail of the current queue
791 		 * before proceeding further
792 		 */
793 		struct ipq *q = TAILQ_LAST(head, ipqhead);
794 		if (q == NULL) {   /* gak */
795 			for (i = 0; i < IPREASS_NHASH; i++) {
796 				struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
797 				if (r) {
798 					ipstat.ips_fragtimeout += r->ipq_nfrags;
799 					ip_freef(&ipq[i], r);
800 					break;
801 				}
802 			}
803 		} else {
804 			ipstat.ips_fragtimeout += q->ipq_nfrags;
805 			ip_freef(head, q);
806 		}
807 	}
808 
809 found:
810 	/*
811 	 * Adjust ip_len to not reflect header,
812 	 * convert offset of this to bytes.
813 	 */
814 	ip->ip_len -= hlen;
815 	if (ip->ip_off & IP_MF) {
816 		/*
817 		 * Make sure that fragments have a data length
818 		 * that's a non-zero multiple of 8 bytes.
819 		 */
820 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
821 			ipstat.ips_toosmall++; /* XXX */
822 			goto dropfrag;
823 		}
824 		m->m_flags |= M_FRAG;
825 	} else
826 		m->m_flags &= ~M_FRAG;
827 	ip->ip_off <<= 3;
828 
829 
830 	/*
831 	 * Attempt reassembly; if it succeeds, proceed.
832 	 * ip_reass() will return a different mbuf.
833 	 */
834 	ipstat.ips_fragments++;
835 	m->m_pkthdr.header = ip;
836 
837 	/* Previous ip_reass() started here. */
838 	/*
839 	 * Presence of header sizes in mbufs
840 	 * would confuse code below.
841 	 */
842 	m->m_data += hlen;
843 	m->m_len -= hlen;
844 
845 	/*
846 	 * If first fragment to arrive, create a reassembly queue.
847 	 */
848 	if (fp == NULL) {
849 		fp = uma_zalloc(ipq_zone, M_NOWAIT);
850 		if (fp == NULL)
851 			goto dropfrag;
852 #ifdef MAC
853 		if (mac_init_ipq(fp, M_NOWAIT) != 0) {
854 			uma_zfree(ipq_zone, fp);
855 			fp = NULL;
856 			goto dropfrag;
857 		}
858 		mac_create_ipq(m, fp);
859 #endif
860 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
861 		nipq++;
862 		fp->ipq_nfrags = 1;
863 		fp->ipq_ttl = IPFRAGTTL;
864 		fp->ipq_p = ip->ip_p;
865 		fp->ipq_id = ip->ip_id;
866 		fp->ipq_src = ip->ip_src;
867 		fp->ipq_dst = ip->ip_dst;
868 		fp->ipq_frags = m;
869 		m->m_nextpkt = NULL;
870 		goto done;
871 	} else {
872 		fp->ipq_nfrags++;
873 #ifdef MAC
874 		mac_update_ipq(m, fp);
875 #endif
876 	}
877 
878 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
879 
880 	/*
881 	 * Handle ECN by comparing this segment with the first one;
882 	 * if CE is set, do not lose CE.
883 	 * drop if CE and not-ECT are mixed for the same packet.
884 	 */
885 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
886 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
887 	if (ecn == IPTOS_ECN_CE) {
888 		if (ecn0 == IPTOS_ECN_NOTECT)
889 			goto dropfrag;
890 		if (ecn0 != IPTOS_ECN_CE)
891 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
892 	}
893 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
894 		goto dropfrag;
895 
896 	/*
897 	 * Find a segment which begins after this one does.
898 	 */
899 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
900 		if (GETIP(q)->ip_off > ip->ip_off)
901 			break;
902 
903 	/*
904 	 * If there is a preceding segment, it may provide some of
905 	 * our data already.  If so, drop the data from the incoming
906 	 * segment.  If it provides all of our data, drop us, otherwise
907 	 * stick new segment in the proper place.
908 	 *
909 	 * If some of the data is dropped from the the preceding
910 	 * segment, then it's checksum is invalidated.
911 	 */
912 	if (p) {
913 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
914 		if (i > 0) {
915 			if (i >= ip->ip_len)
916 				goto dropfrag;
917 			m_adj(m, i);
918 			m->m_pkthdr.csum_flags = 0;
919 			ip->ip_off += i;
920 			ip->ip_len -= i;
921 		}
922 		m->m_nextpkt = p->m_nextpkt;
923 		p->m_nextpkt = m;
924 	} else {
925 		m->m_nextpkt = fp->ipq_frags;
926 		fp->ipq_frags = m;
927 	}
928 
929 	/*
930 	 * While we overlap succeeding segments trim them or,
931 	 * if they are completely covered, dequeue them.
932 	 */
933 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
934 	     q = nq) {
935 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
936 		if (i < GETIP(q)->ip_len) {
937 			GETIP(q)->ip_len -= i;
938 			GETIP(q)->ip_off += i;
939 			m_adj(q, i);
940 			q->m_pkthdr.csum_flags = 0;
941 			break;
942 		}
943 		nq = q->m_nextpkt;
944 		m->m_nextpkt = nq;
945 		ipstat.ips_fragdropped++;
946 		fp->ipq_nfrags--;
947 		m_freem(q);
948 	}
949 
950 	/*
951 	 * Check for complete reassembly and perform frag per packet
952 	 * limiting.
953 	 *
954 	 * Frag limiting is performed here so that the nth frag has
955 	 * a chance to complete the packet before we drop the packet.
956 	 * As a result, n+1 frags are actually allowed per packet, but
957 	 * only n will ever be stored. (n = maxfragsperpacket.)
958 	 *
959 	 */
960 	next = 0;
961 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
962 		if (GETIP(q)->ip_off != next) {
963 			if (fp->ipq_nfrags > maxfragsperpacket) {
964 				ipstat.ips_fragdropped += fp->ipq_nfrags;
965 				ip_freef(head, fp);
966 			}
967 			goto done;
968 		}
969 		next += GETIP(q)->ip_len;
970 	}
971 	/* Make sure the last packet didn't have the IP_MF flag */
972 	if (p->m_flags & M_FRAG) {
973 		if (fp->ipq_nfrags > maxfragsperpacket) {
974 			ipstat.ips_fragdropped += fp->ipq_nfrags;
975 			ip_freef(head, fp);
976 		}
977 		goto done;
978 	}
979 
980 	/*
981 	 * Reassembly is complete.  Make sure the packet is a sane size.
982 	 */
983 	q = fp->ipq_frags;
984 	ip = GETIP(q);
985 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
986 		ipstat.ips_toolong++;
987 		ipstat.ips_fragdropped += fp->ipq_nfrags;
988 		ip_freef(head, fp);
989 		goto done;
990 	}
991 
992 	/*
993 	 * Concatenate fragments.
994 	 */
995 	m = q;
996 	t = m->m_next;
997 	m->m_next = NULL;
998 	m_cat(m, t);
999 	nq = q->m_nextpkt;
1000 	q->m_nextpkt = NULL;
1001 	for (q = nq; q != NULL; q = nq) {
1002 		nq = q->m_nextpkt;
1003 		q->m_nextpkt = NULL;
1004 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1005 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1006 		m_cat(m, q);
1007 	}
1008 	/*
1009 	 * In order to do checksumming faster we do 'end-around carry' here
1010 	 * (and not in for{} loop), though it implies we are not going to
1011 	 * reassemble more than 64k fragments.
1012 	 */
1013 	m->m_pkthdr.csum_data =
1014 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1015 #ifdef MAC
1016 	mac_create_datagram_from_ipq(fp, m);
1017 	mac_destroy_ipq(fp);
1018 #endif
1019 
1020 	/*
1021 	 * Create header for new ip packet by modifying header of first
1022 	 * packet;  dequeue and discard fragment reassembly header.
1023 	 * Make header visible.
1024 	 */
1025 	ip->ip_len = (ip->ip_hl << 2) + next;
1026 	ip->ip_src = fp->ipq_src;
1027 	ip->ip_dst = fp->ipq_dst;
1028 	TAILQ_REMOVE(head, fp, ipq_list);
1029 	nipq--;
1030 	uma_zfree(ipq_zone, fp);
1031 	m->m_len += (ip->ip_hl << 2);
1032 	m->m_data -= (ip->ip_hl << 2);
1033 	/* some debugging cruft by sklower, below, will go away soon */
1034 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1035 		m_fixhdr(m);
1036 	ipstat.ips_reassembled++;
1037 	IPQ_UNLOCK();
1038 	return (m);
1039 
1040 dropfrag:
1041 	ipstat.ips_fragdropped++;
1042 	if (fp != NULL)
1043 		fp->ipq_nfrags--;
1044 	m_freem(m);
1045 done:
1046 	IPQ_UNLOCK();
1047 	return (NULL);
1048 
1049 #undef GETIP
1050 }
1051 
1052 /*
1053  * Free a fragment reassembly header and all
1054  * associated datagrams.
1055  */
1056 static void
1057 ip_freef(fhp, fp)
1058 	struct ipqhead *fhp;
1059 	struct ipq *fp;
1060 {
1061 	register struct mbuf *q;
1062 
1063 	IPQ_LOCK_ASSERT();
1064 
1065 	while (fp->ipq_frags) {
1066 		q = fp->ipq_frags;
1067 		fp->ipq_frags = q->m_nextpkt;
1068 		m_freem(q);
1069 	}
1070 	TAILQ_REMOVE(fhp, fp, ipq_list);
1071 	uma_zfree(ipq_zone, fp);
1072 	nipq--;
1073 }
1074 
1075 /*
1076  * IP timer processing;
1077  * if a timer expires on a reassembly
1078  * queue, discard it.
1079  */
1080 void
1081 ip_slowtimo()
1082 {
1083 	register struct ipq *fp;
1084 	int i;
1085 
1086 	IPQ_LOCK();
1087 	for (i = 0; i < IPREASS_NHASH; i++) {
1088 		for(fp = TAILQ_FIRST(&ipq[i]); fp;) {
1089 			struct ipq *fpp;
1090 
1091 			fpp = fp;
1092 			fp = TAILQ_NEXT(fp, ipq_list);
1093 			if(--fpp->ipq_ttl == 0) {
1094 				ipstat.ips_fragtimeout += fpp->ipq_nfrags;
1095 				ip_freef(&ipq[i], fpp);
1096 			}
1097 		}
1098 	}
1099 	/*
1100 	 * If we are over the maximum number of fragments
1101 	 * (due to the limit being lowered), drain off
1102 	 * enough to get down to the new limit.
1103 	 */
1104 	if (maxnipq >= 0 && nipq > maxnipq) {
1105 		for (i = 0; i < IPREASS_NHASH; i++) {
1106 			while (nipq > maxnipq && !TAILQ_EMPTY(&ipq[i])) {
1107 				ipstat.ips_fragdropped +=
1108 				    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1109 				ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1110 			}
1111 		}
1112 	}
1113 	IPQ_UNLOCK();
1114 }
1115 
1116 /*
1117  * Drain off all datagram fragments.
1118  */
1119 void
1120 ip_drain()
1121 {
1122 	int     i;
1123 
1124 	IPQ_LOCK();
1125 	for (i = 0; i < IPREASS_NHASH; i++) {
1126 		while(!TAILQ_EMPTY(&ipq[i])) {
1127 			ipstat.ips_fragdropped +=
1128 			    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1129 			ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1130 		}
1131 	}
1132 	IPQ_UNLOCK();
1133 	in_rtqdrain();
1134 }
1135 
1136 /*
1137  * The protocol to be inserted into ip_protox[] must be already registered
1138  * in inetsw[], either statically or through pf_proto_register().
1139  */
1140 int
1141 ipproto_register(u_char ipproto)
1142 {
1143 	struct protosw *pr;
1144 
1145 	/* Sanity checks. */
1146 	if (ipproto == 0)
1147 		return (EPROTONOSUPPORT);
1148 
1149 	/*
1150 	 * The protocol slot must not be occupied by another protocol
1151 	 * already.  An index pointing to IPPROTO_RAW is unused.
1152 	 */
1153 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1154 	if (pr == NULL)
1155 		return (EPFNOSUPPORT);
1156 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1157 		return (EEXIST);
1158 
1159 	/* Find the protocol position in inetsw[] and set the index. */
1160 	for (pr = inetdomain.dom_protosw;
1161 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1162 		if (pr->pr_domain->dom_family == PF_INET &&
1163 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1164 			/* Be careful to only index valid IP protocols. */
1165 			if (pr->pr_protocol < IPPROTO_MAX) {
1166 				ip_protox[pr->pr_protocol] = pr - inetsw;
1167 				return (0);
1168 			} else
1169 				return (EINVAL);
1170 		}
1171 	}
1172 	return (EPROTONOSUPPORT);
1173 }
1174 
1175 int
1176 ipproto_unregister(u_char ipproto)
1177 {
1178 	struct protosw *pr;
1179 
1180 	/* Sanity checks. */
1181 	if (ipproto == 0)
1182 		return (EPROTONOSUPPORT);
1183 
1184 	/* Check if the protocol was indeed registered. */
1185 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1186 	if (pr == NULL)
1187 		return (EPFNOSUPPORT);
1188 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1189 		return (ENOENT);
1190 
1191 	/* Reset the protocol slot to IPPROTO_RAW. */
1192 	ip_protox[ipproto] = pr - inetsw;
1193 	return (0);
1194 }
1195 
1196 /*
1197  * Given address of next destination (final or next hop),
1198  * return internet address info of interface to be used to get there.
1199  */
1200 struct in_ifaddr *
1201 ip_rtaddr(dst)
1202 	struct in_addr dst;
1203 {
1204 	struct route sro;
1205 	struct sockaddr_in *sin;
1206 	struct in_ifaddr *ifa;
1207 
1208 	bzero(&sro, sizeof(sro));
1209 	sin = (struct sockaddr_in *)&sro.ro_dst;
1210 	sin->sin_family = AF_INET;
1211 	sin->sin_len = sizeof(*sin);
1212 	sin->sin_addr = dst;
1213 	rtalloc_ign(&sro, RTF_CLONING);
1214 
1215 	if (sro.ro_rt == NULL)
1216 		return (NULL);
1217 
1218 	ifa = ifatoia(sro.ro_rt->rt_ifa);
1219 	RTFREE(sro.ro_rt);
1220 	return (ifa);
1221 }
1222 
1223 u_char inetctlerrmap[PRC_NCMDS] = {
1224 	0,		0,		0,		0,
1225 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1226 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1227 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1228 	0,		0,		EHOSTUNREACH,	0,
1229 	ENOPROTOOPT,	ECONNREFUSED
1230 };
1231 
1232 /*
1233  * Forward a packet.  If some error occurs return the sender
1234  * an icmp packet.  Note we can't always generate a meaningful
1235  * icmp message because icmp doesn't have a large enough repertoire
1236  * of codes and types.
1237  *
1238  * If not forwarding, just drop the packet.  This could be confusing
1239  * if ipforwarding was zero but some routing protocol was advancing
1240  * us as a gateway to somewhere.  However, we must let the routing
1241  * protocol deal with that.
1242  *
1243  * The srcrt parameter indicates whether the packet is being forwarded
1244  * via a source route.
1245  */
1246 void
1247 ip_forward(struct mbuf *m, int srcrt)
1248 {
1249 	struct ip *ip = mtod(m, struct ip *);
1250 	struct in_ifaddr *ia = NULL;
1251 	struct mbuf *mcopy;
1252 	struct in_addr dest;
1253 	int error, type = 0, code = 0, mtu = 0;
1254 
1255 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1256 		ipstat.ips_cantforward++;
1257 		m_freem(m);
1258 		return;
1259 	}
1260 #ifdef IPSTEALTH
1261 	if (!ipstealth) {
1262 #endif
1263 		if (ip->ip_ttl <= IPTTLDEC) {
1264 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1265 			    0, 0);
1266 			return;
1267 		}
1268 #ifdef IPSTEALTH
1269 	}
1270 #endif
1271 
1272 	if (!srcrt && (ia = ip_rtaddr(ip->ip_dst)) == NULL) {
1273 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1274 		return;
1275 	}
1276 
1277 	/*
1278 	 * Save the IP header and at most 8 bytes of the payload,
1279 	 * in case we need to generate an ICMP message to the src.
1280 	 *
1281 	 * XXX this can be optimized a lot by saving the data in a local
1282 	 * buffer on the stack (72 bytes at most), and only allocating the
1283 	 * mbuf if really necessary. The vast majority of the packets
1284 	 * are forwarded without having to send an ICMP back (either
1285 	 * because unnecessary, or because rate limited), so we are
1286 	 * really we are wasting a lot of work here.
1287 	 *
1288 	 * We don't use m_copy() because it might return a reference
1289 	 * to a shared cluster. Both this function and ip_output()
1290 	 * assume exclusive access to the IP header in `m', so any
1291 	 * data in a cluster may change before we reach icmp_error().
1292 	 */
1293 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1294 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1295 		/*
1296 		 * It's probably ok if the pkthdr dup fails (because
1297 		 * the deep copy of the tag chain failed), but for now
1298 		 * be conservative and just discard the copy since
1299 		 * code below may some day want the tags.
1300 		 */
1301 		m_free(mcopy);
1302 		mcopy = NULL;
1303 	}
1304 	if (mcopy != NULL) {
1305 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1306 		mcopy->m_pkthdr.len = mcopy->m_len;
1307 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1308 	}
1309 
1310 #ifdef IPSTEALTH
1311 	if (!ipstealth) {
1312 #endif
1313 		ip->ip_ttl -= IPTTLDEC;
1314 #ifdef IPSTEALTH
1315 	}
1316 #endif
1317 
1318 	/*
1319 	 * If forwarding packet using same interface that it came in on,
1320 	 * perhaps should send a redirect to sender to shortcut a hop.
1321 	 * Only send redirect if source is sending directly to us,
1322 	 * and if packet was not source routed (or has any options).
1323 	 * Also, don't send redirect if forwarding using a default route
1324 	 * or a route modified by a redirect.
1325 	 */
1326 	dest.s_addr = 0;
1327 	if (!srcrt && ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1328 		struct sockaddr_in *sin;
1329 		struct route ro;
1330 		struct rtentry *rt;
1331 
1332 		bzero(&ro, sizeof(ro));
1333 		sin = (struct sockaddr_in *)&ro.ro_dst;
1334 		sin->sin_family = AF_INET;
1335 		sin->sin_len = sizeof(*sin);
1336 		sin->sin_addr = ip->ip_dst;
1337 		rtalloc_ign(&ro, RTF_CLONING);
1338 
1339 		rt = ro.ro_rt;
1340 
1341 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1342 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1343 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1344 			u_long src = ntohl(ip->ip_src.s_addr);
1345 
1346 			if (RTA(rt) &&
1347 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1348 				if (rt->rt_flags & RTF_GATEWAY)
1349 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1350 				else
1351 					dest.s_addr = ip->ip_dst.s_addr;
1352 				/* Router requirements says to only send host redirects */
1353 				type = ICMP_REDIRECT;
1354 				code = ICMP_REDIRECT_HOST;
1355 			}
1356 		}
1357 		if (rt)
1358 			RTFREE(rt);
1359 	}
1360 
1361 	error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1362 	if (error)
1363 		ipstat.ips_cantforward++;
1364 	else {
1365 		ipstat.ips_forward++;
1366 		if (type)
1367 			ipstat.ips_redirectsent++;
1368 		else {
1369 			if (mcopy)
1370 				m_freem(mcopy);
1371 			return;
1372 		}
1373 	}
1374 	if (mcopy == NULL)
1375 		return;
1376 
1377 	switch (error) {
1378 
1379 	case 0:				/* forwarded, but need redirect */
1380 		/* type, code set above */
1381 		break;
1382 
1383 	case ENETUNREACH:		/* shouldn't happen, checked above */
1384 	case EHOSTUNREACH:
1385 	case ENETDOWN:
1386 	case EHOSTDOWN:
1387 	default:
1388 		type = ICMP_UNREACH;
1389 		code = ICMP_UNREACH_HOST;
1390 		break;
1391 
1392 	case EMSGSIZE:
1393 		type = ICMP_UNREACH;
1394 		code = ICMP_UNREACH_NEEDFRAG;
1395 
1396 #if defined(IPSEC) || defined(FAST_IPSEC)
1397 		mtu = ip_ipsec_mtu(m);
1398 #endif /* IPSEC */
1399 		/*
1400 		 * If the MTU wasn't set before use the interface mtu or
1401 		 * fall back to the next smaller mtu step compared to the
1402 		 * current packet size.
1403 		 */
1404 		if (mtu == 0) {
1405 			if (ia != NULL)
1406 				mtu = ia->ia_ifp->if_mtu;
1407 			else
1408 				mtu = ip_next_mtu(ip->ip_len, 0);
1409 		}
1410 		ipstat.ips_cantfrag++;
1411 		break;
1412 
1413 	case ENOBUFS:
1414 		/*
1415 		 * A router should not generate ICMP_SOURCEQUENCH as
1416 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1417 		 * Source quench could be a big problem under DoS attacks,
1418 		 * or if the underlying interface is rate-limited.
1419 		 * Those who need source quench packets may re-enable them
1420 		 * via the net.inet.ip.sendsourcequench sysctl.
1421 		 */
1422 		if (ip_sendsourcequench == 0) {
1423 			m_freem(mcopy);
1424 			return;
1425 		} else {
1426 			type = ICMP_SOURCEQUENCH;
1427 			code = 0;
1428 		}
1429 		break;
1430 
1431 	case EACCES:			/* ipfw denied packet */
1432 		m_freem(mcopy);
1433 		return;
1434 	}
1435 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1436 }
1437 
1438 void
1439 ip_savecontrol(inp, mp, ip, m)
1440 	register struct inpcb *inp;
1441 	register struct mbuf **mp;
1442 	register struct ip *ip;
1443 	register struct mbuf *m;
1444 {
1445 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1446 		struct bintime bt;
1447 
1448 		bintime(&bt);
1449 		if (inp->inp_socket->so_options & SO_BINTIME) {
1450 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1451 			SCM_BINTIME, SOL_SOCKET);
1452 			if (*mp)
1453 				mp = &(*mp)->m_next;
1454 		}
1455 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1456 			struct timeval tv;
1457 
1458 			bintime2timeval(&bt, &tv);
1459 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1460 				SCM_TIMESTAMP, SOL_SOCKET);
1461 			if (*mp)
1462 				mp = &(*mp)->m_next;
1463 		}
1464 	}
1465 	if (inp->inp_flags & INP_RECVDSTADDR) {
1466 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1467 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1468 		if (*mp)
1469 			mp = &(*mp)->m_next;
1470 	}
1471 	if (inp->inp_flags & INP_RECVTTL) {
1472 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1473 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1474 		if (*mp)
1475 			mp = &(*mp)->m_next;
1476 	}
1477 #ifdef notyet
1478 	/* XXX
1479 	 * Moving these out of udp_input() made them even more broken
1480 	 * than they already were.
1481 	 */
1482 	/* options were tossed already */
1483 	if (inp->inp_flags & INP_RECVOPTS) {
1484 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1485 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1486 		if (*mp)
1487 			mp = &(*mp)->m_next;
1488 	}
1489 	/* ip_srcroute doesn't do what we want here, need to fix */
1490 	if (inp->inp_flags & INP_RECVRETOPTS) {
1491 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1492 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1493 		if (*mp)
1494 			mp = &(*mp)->m_next;
1495 	}
1496 #endif
1497 	if (inp->inp_flags & INP_RECVIF) {
1498 		struct ifnet *ifp;
1499 		struct sdlbuf {
1500 			struct sockaddr_dl sdl;
1501 			u_char	pad[32];
1502 		} sdlbuf;
1503 		struct sockaddr_dl *sdp;
1504 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1505 
1506 		if (((ifp = m->m_pkthdr.rcvif))
1507 		&& ( ifp->if_index && (ifp->if_index <= if_index))) {
1508 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1509 			/*
1510 			 * Change our mind and don't try copy.
1511 			 */
1512 			if ((sdp->sdl_family != AF_LINK)
1513 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1514 				goto makedummy;
1515 			}
1516 			bcopy(sdp, sdl2, sdp->sdl_len);
1517 		} else {
1518 makedummy:
1519 			sdl2->sdl_len
1520 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1521 			sdl2->sdl_family = AF_LINK;
1522 			sdl2->sdl_index = 0;
1523 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1524 		}
1525 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1526 			IP_RECVIF, IPPROTO_IP);
1527 		if (*mp)
1528 			mp = &(*mp)->m_next;
1529 	}
1530 }
1531 
1532 /*
1533  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1534  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1535  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1536  * compiled.
1537  */
1538 static int ip_rsvp_on;
1539 struct socket *ip_rsvpd;
1540 int
1541 ip_rsvp_init(struct socket *so)
1542 {
1543 	if (so->so_type != SOCK_RAW ||
1544 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1545 		return EOPNOTSUPP;
1546 
1547 	if (ip_rsvpd != NULL)
1548 		return EADDRINUSE;
1549 
1550 	ip_rsvpd = so;
1551 	/*
1552 	 * This may seem silly, but we need to be sure we don't over-increment
1553 	 * the RSVP counter, in case something slips up.
1554 	 */
1555 	if (!ip_rsvp_on) {
1556 		ip_rsvp_on = 1;
1557 		rsvp_on++;
1558 	}
1559 
1560 	return 0;
1561 }
1562 
1563 int
1564 ip_rsvp_done(void)
1565 {
1566 	ip_rsvpd = NULL;
1567 	/*
1568 	 * This may seem silly, but we need to be sure we don't over-decrement
1569 	 * the RSVP counter, in case something slips up.
1570 	 */
1571 	if (ip_rsvp_on) {
1572 		ip_rsvp_on = 0;
1573 		rsvp_on--;
1574 	}
1575 	return 0;
1576 }
1577 
1578 void
1579 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1580 {
1581 	if (rsvp_input_p) { /* call the real one if loaded */
1582 		rsvp_input_p(m, off);
1583 		return;
1584 	}
1585 
1586 	/* Can still get packets with rsvp_on = 0 if there is a local member
1587 	 * of the group to which the RSVP packet is addressed.  But in this
1588 	 * case we want to throw the packet away.
1589 	 */
1590 
1591 	if (!rsvp_on) {
1592 		m_freem(m);
1593 		return;
1594 	}
1595 
1596 	if (ip_rsvpd != NULL) {
1597 		rip_input(m, off);
1598 		return;
1599 	}
1600 	/* Drop the packet */
1601 	m_freem(m);
1602 }
1603