xref: /freebsd/sys/netinet/ip_input.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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/mac.h>
44 #include <sys/mbuf.h>
45 #include <sys/malloc.h>
46 #include <sys/domain.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/syslog.h>
52 #include <sys/sysctl.h>
53 
54 #include <net/pfil.h>
55 #include <net/if.h>
56 #include <net/if_types.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/netisr.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/ip_var.h>
68 #include <netinet/ip_icmp.h>
69 #include <netinet/ip_options.h>
70 #include <machine/in_cksum.h>
71 #ifdef DEV_CARP
72 #include <netinet/ip_carp.h>
73 #endif
74 #if defined(IPSEC) || defined(FAST_IPSEC)
75 #include <netinet/ip_ipsec.h>
76 #endif /* IPSEC */
77 
78 #include <sys/socketvar.h>
79 
80 /* XXX: Temporary until ipfw_ether and ipfw_bridge are converted. */
81 #include <netinet/ip_fw.h>
82 #include <netinet/ip_dummynet.h>
83 
84 int rsvp_on = 0;
85 
86 int	ipforwarding = 0;
87 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
88     &ipforwarding, 0, "Enable IP forwarding between interfaces");
89 
90 static int	ipsendredirects = 1; /* XXX */
91 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
92     &ipsendredirects, 0, "Enable sending IP redirects");
93 
94 int	ip_defttl = IPDEFTTL;
95 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
96     &ip_defttl, 0, "Maximum TTL on IP packets");
97 
98 static int	ip_keepfaith = 0;
99 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
100 	&ip_keepfaith,	0,
101 	"Enable packet capture for FAITH IPv4->IPv6 translater daemon");
102 
103 static int	ip_sendsourcequench = 0;
104 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
105 	&ip_sendsourcequench, 0,
106 	"Enable the transmission of source quench packets");
107 
108 int	ip_do_randomid = 0;
109 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
110 	&ip_do_randomid, 0,
111 	"Assign random ip_id values");
112 
113 /*
114  * XXX - Setting ip_checkinterface mostly implements the receive side of
115  * the Strong ES model described in RFC 1122, but since the routing table
116  * and transmit implementation do not implement the Strong ES model,
117  * setting this to 1 results in an odd hybrid.
118  *
119  * XXX - ip_checkinterface currently must be disabled if you use ipnat
120  * to translate the destination address to another local interface.
121  *
122  * XXX - ip_checkinterface must be disabled if you add IP aliases
123  * to the loopback interface instead of the interface where the
124  * packets for those addresses are received.
125  */
126 static int	ip_checkinterface = 0;
127 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
128     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
129 
130 struct pfil_head inet_pfil_hook;	/* Packet filter hooks */
131 
132 static struct	ifqueue ipintrq;
133 static int	ipqmaxlen = IFQ_MAXLEN;
134 
135 extern	struct domain inetdomain;
136 extern	struct protosw inetsw[];
137 u_char	ip_protox[IPPROTO_MAX];
138 struct	in_ifaddrhead in_ifaddrhead; 		/* first inet address */
139 struct	in_ifaddrhashhead *in_ifaddrhashtbl;	/* inet addr hash table  */
140 u_long 	in_ifaddrhmask;				/* mask for hash table */
141 
142 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
143     &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
144 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
145     &ipintrq.ifq_drops, 0, "Number of packets dropped from the IP input queue");
146 
147 struct ipstat ipstat;
148 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
149     &ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
150 
151 /*
152  * IP datagram reassembly.
153  */
154 #define IPREASS_NHASH_LOG2      6
155 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
156 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
157 #define IPREASS_HASH(x,y) \
158 	(((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
159 
160 static uma_zone_t ipq_zone;
161 static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
162 static struct mtx ipqlock;
163 
164 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
165 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
166 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
167 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
168 
169 static void	maxnipq_update(void);
170 static void	ipq_zone_change(void *);
171 
172 static int	maxnipq;	/* Administrative limit on # reass queues. */
173 static int	nipq = 0;	/* Total # of reass queues */
174 SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD, &nipq, 0,
175 	"Current number of IPv4 fragment reassembly queue entries");
176 
177 static int	maxfragsperpacket;
178 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
179 	&maxfragsperpacket, 0,
180 	"Maximum number of IPv4 fragments allowed per packet");
181 
182 struct callout	ipport_tick_callout;
183 
184 #ifdef IPCTL_DEFMTU
185 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
186     &ip_mtu, 0, "Default MTU");
187 #endif
188 
189 #ifdef IPSTEALTH
190 int	ipstealth = 0;
191 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
192     &ipstealth, 0, "");
193 #endif
194 
195 /*
196  * ipfw_ether and ipfw_bridge hooks.
197  * XXX: Temporary until those are converted to pfil_hooks as well.
198  */
199 ip_fw_chk_t *ip_fw_chk_ptr = NULL;
200 ip_dn_io_t *ip_dn_io_ptr = NULL;
201 int fw_one_pass = 1;
202 
203 static void	ip_freef(struct ipqhead *, struct ipq *);
204 
205 /*
206  * IP initialization: fill in IP protocol switch table.
207  * All protocols not implemented in kernel go to raw IP protocol handler.
208  */
209 void
210 ip_init()
211 {
212 	register struct protosw *pr;
213 	register int i;
214 
215 	TAILQ_INIT(&in_ifaddrhead);
216 	in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
217 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
218 	if (pr == NULL)
219 		panic("ip_init: PF_INET not found");
220 
221 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
222 	for (i = 0; i < IPPROTO_MAX; i++)
223 		ip_protox[i] = pr - inetsw;
224 	/*
225 	 * Cycle through IP protocols and put them into the appropriate place
226 	 * in ip_protox[].
227 	 */
228 	for (pr = inetdomain.dom_protosw;
229 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
230 		if (pr->pr_domain->dom_family == PF_INET &&
231 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
232 			/* Be careful to only index valid IP protocols. */
233 			if (pr->pr_protocol < IPPROTO_MAX)
234 				ip_protox[pr->pr_protocol] = pr - inetsw;
235 		}
236 
237 	/* Initialize packet filter hooks. */
238 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
239 	inet_pfil_hook.ph_af = AF_INET;
240 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
241 		printf("%s: WARNING: unable to register pfil hook, "
242 			"error %d\n", __func__, i);
243 
244 	/* Initialize IP reassembly queue. */
245 	IPQ_LOCK_INIT();
246 	for (i = 0; i < IPREASS_NHASH; i++)
247 	    TAILQ_INIT(&ipq[i]);
248 	maxnipq = nmbclusters / 32;
249 	maxfragsperpacket = 16;
250 	ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
251 	    NULL, UMA_ALIGN_PTR, 0);
252 	maxnipq_update();
253 
254 	/* Start ipport_tick. */
255 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
256 	ipport_tick(NULL);
257 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
258 		SHUTDOWN_PRI_DEFAULT);
259 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
260 		NULL, EVENTHANDLER_PRI_ANY);
261 
262 	/* Initialize various other remaining things. */
263 	ip_id = time_second & 0xffff;
264 	ipintrq.ifq_maxlen = ipqmaxlen;
265 	mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
266 	netisr_register(NETISR_IP, ip_input, &ipintrq, NETISR_MPSAFE);
267 }
268 
269 void ip_fini(xtp)
270 	void *xtp;
271 {
272 	callout_stop(&ipport_tick_callout);
273 }
274 
275 /*
276  * Ip input routine.  Checksum and byte swap header.  If fragmented
277  * try to reassemble.  Process options.  Pass to next level.
278  */
279 void
280 ip_input(struct mbuf *m)
281 {
282 	struct ip *ip = NULL;
283 	struct in_ifaddr *ia = NULL;
284 	struct ifaddr *ifa;
285 	int    checkif, hlen = 0;
286 	u_short sum;
287 	int dchg = 0;				/* dest changed after fw */
288 	struct in_addr odst;			/* original dst address */
289 
290   	M_ASSERTPKTHDR(m);
291 
292 	if (m->m_flags & M_FASTFWD_OURS) {
293 		/*
294 		 * Firewall or NAT changed destination to local.
295 		 * We expect ip_len and ip_off to be in host byte order.
296 		 */
297 		m->m_flags &= ~M_FASTFWD_OURS;
298 		/* Set up some basics that will be used later. */
299 		ip = mtod(m, struct ip *);
300 		hlen = ip->ip_hl << 2;
301   		goto ours;
302   	}
303 
304 	ipstat.ips_total++;
305 
306 	if (m->m_pkthdr.len < sizeof(struct ip))
307 		goto tooshort;
308 
309 	if (m->m_len < sizeof (struct ip) &&
310 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
311 		ipstat.ips_toosmall++;
312 		return;
313 	}
314 	ip = mtod(m, struct ip *);
315 
316 	if (ip->ip_v != IPVERSION) {
317 		ipstat.ips_badvers++;
318 		goto bad;
319 	}
320 
321 	hlen = ip->ip_hl << 2;
322 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
323 		ipstat.ips_badhlen++;
324 		goto bad;
325 	}
326 	if (hlen > m->m_len) {
327 		if ((m = m_pullup(m, hlen)) == NULL) {
328 			ipstat.ips_badhlen++;
329 			return;
330 		}
331 		ip = mtod(m, struct ip *);
332 	}
333 
334 	/* 127/8 must not appear on wire - RFC1122 */
335 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
336 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
337 		if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
338 			ipstat.ips_badaddr++;
339 			goto bad;
340 		}
341 	}
342 
343 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
344 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
345 	} else {
346 		if (hlen == sizeof(struct ip)) {
347 			sum = in_cksum_hdr(ip);
348 		} else {
349 			sum = in_cksum(m, hlen);
350 		}
351 	}
352 	if (sum) {
353 		ipstat.ips_badsum++;
354 		goto bad;
355 	}
356 
357 #ifdef ALTQ
358 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
359 		/* packet is dropped by traffic conditioner */
360 		return;
361 #endif
362 
363 	/*
364 	 * Convert fields to host representation.
365 	 */
366 	ip->ip_len = ntohs(ip->ip_len);
367 	if (ip->ip_len < hlen) {
368 		ipstat.ips_badlen++;
369 		goto bad;
370 	}
371 	ip->ip_off = ntohs(ip->ip_off);
372 
373 	/*
374 	 * Check that the amount of data in the buffers
375 	 * is as at least much as the IP header would have us expect.
376 	 * Trim mbufs if longer than we expect.
377 	 * Drop packet if shorter than we expect.
378 	 */
379 	if (m->m_pkthdr.len < ip->ip_len) {
380 tooshort:
381 		ipstat.ips_tooshort++;
382 		goto bad;
383 	}
384 	if (m->m_pkthdr.len > ip->ip_len) {
385 		if (m->m_len == m->m_pkthdr.len) {
386 			m->m_len = ip->ip_len;
387 			m->m_pkthdr.len = ip->ip_len;
388 		} else
389 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
390 	}
391 #if defined(IPSEC) || defined(FAST_IPSEC)
392 	/*
393 	 * Bypass packet filtering for packets from a tunnel (gif).
394 	 */
395 	if (ip_ipsec_filtergif(m))
396 		goto passin;
397 #endif /* IPSEC */
398 
399 	/*
400 	 * Run through list of hooks for input packets.
401 	 *
402 	 * NB: Beware of the destination address changing (e.g.
403 	 *     by NAT rewriting).  When this happens, tell
404 	 *     ip_forward to do the right thing.
405 	 */
406 
407 	/* Jump over all PFIL processing if hooks are not active. */
408 	if (!PFIL_HOOKED(&inet_pfil_hook))
409 		goto passin;
410 
411 	odst = ip->ip_dst;
412 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
413 	    PFIL_IN, NULL) != 0)
414 		return;
415 	if (m == NULL)			/* consumed by filter */
416 		return;
417 
418 	ip = mtod(m, struct ip *);
419 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
420 
421 #ifdef IPFIREWALL_FORWARD
422 	if (m->m_flags & M_FASTFWD_OURS) {
423 		m->m_flags &= ~M_FASTFWD_OURS;
424 		goto ours;
425 	}
426 #ifndef IPFIREWALL_FORWARD_EXTENDED
427 	dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL);
428 #else
429 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
430 		/*
431 		 * Directly ship on the packet.  This allows to forward packets
432 		 * that were destined for us to some other directly connected
433 		 * host.
434 		 */
435 		ip_forward(m, dchg);
436 		return;
437 	}
438 #endif /* IPFIREWALL_FORWARD_EXTENDED */
439 #endif /* IPFIREWALL_FORWARD */
440 
441 passin:
442 	/*
443 	 * Process options and, if not destined for us,
444 	 * ship it on.  ip_dooptions returns 1 when an
445 	 * error was detected (causing an icmp message
446 	 * to be sent and the original packet to be freed).
447 	 */
448 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
449 		return;
450 
451         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
452          * matter if it is destined to another node, or whether it is
453          * a multicast one, RSVP wants it! and prevents it from being forwarded
454          * anywhere else. Also checks if the rsvp daemon is running before
455 	 * grabbing the packet.
456          */
457 	if (rsvp_on && ip->ip_p==IPPROTO_RSVP)
458 		goto ours;
459 
460 	/*
461 	 * Check our list of addresses, to see if the packet is for us.
462 	 * If we don't have any addresses, assume any unicast packet
463 	 * we receive might be for us (and let the upper layers deal
464 	 * with it).
465 	 */
466 	if (TAILQ_EMPTY(&in_ifaddrhead) &&
467 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
468 		goto ours;
469 
470 	/*
471 	 * Enable a consistency check between the destination address
472 	 * and the arrival interface for a unicast packet (the RFC 1122
473 	 * strong ES model) if IP forwarding is disabled and the packet
474 	 * is not locally generated and the packet is not subject to
475 	 * 'ipfw fwd'.
476 	 *
477 	 * XXX - Checking also should be disabled if the destination
478 	 * address is ipnat'ed to a different interface.
479 	 *
480 	 * XXX - Checking is incompatible with IP aliases added
481 	 * to the loopback interface instead of the interface where
482 	 * the packets are received.
483 	 *
484 	 * XXX - This is the case for carp vhost IPs as well so we
485 	 * insert a workaround. If the packet got here, we already
486 	 * checked with carp_iamatch() and carp_forus().
487 	 */
488 	checkif = ip_checkinterface && (ipforwarding == 0) &&
489 	    m->m_pkthdr.rcvif != NULL &&
490 	    ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) &&
491 #ifdef DEV_CARP
492 	    !m->m_pkthdr.rcvif->if_carp &&
493 #endif
494 	    (dchg == 0);
495 
496 	/*
497 	 * Check for exact addresses in the hash bucket.
498 	 */
499 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
500 		/*
501 		 * If the address matches, verify that the packet
502 		 * arrived via the correct interface if checking is
503 		 * enabled.
504 		 */
505 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
506 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
507 			goto ours;
508 	}
509 	/*
510 	 * Check for broadcast addresses.
511 	 *
512 	 * Only accept broadcast packets that arrive via the matching
513 	 * interface.  Reception of forwarded directed broadcasts would
514 	 * be handled via ip_forward() and ether_output() with the loopback
515 	 * into the stack for SIMPLEX interfaces handled by ether_output().
516 	 */
517 	if (m->m_pkthdr.rcvif != NULL &&
518 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
519 	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
520 			if (ifa->ifa_addr->sa_family != AF_INET)
521 				continue;
522 			ia = ifatoia(ifa);
523 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
524 			    ip->ip_dst.s_addr)
525 				goto ours;
526 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
527 				goto ours;
528 #ifdef BOOTP_COMPAT
529 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
530 				goto ours;
531 #endif
532 		}
533 	}
534 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
535 		struct in_multi *inm;
536 		if (ip_mrouter) {
537 			/*
538 			 * If we are acting as a multicast router, all
539 			 * incoming multicast packets are passed to the
540 			 * kernel-level multicast forwarding function.
541 			 * The packet is returned (relatively) intact; if
542 			 * ip_mforward() returns a non-zero value, the packet
543 			 * must be discarded, else it may be accepted below.
544 			 */
545 			if (ip_mforward &&
546 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
547 				ipstat.ips_cantforward++;
548 				m_freem(m);
549 				return;
550 			}
551 
552 			/*
553 			 * The process-level routing daemon needs to receive
554 			 * all multicast IGMP packets, whether or not this
555 			 * host belongs to their destination groups.
556 			 */
557 			if (ip->ip_p == IPPROTO_IGMP)
558 				goto ours;
559 			ipstat.ips_forward++;
560 		}
561 		/*
562 		 * See if we belong to the destination multicast group on the
563 		 * arrival interface.
564 		 */
565 		IN_MULTI_LOCK();
566 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
567 		IN_MULTI_UNLOCK();
568 		if (inm == NULL) {
569 			ipstat.ips_notmember++;
570 			m_freem(m);
571 			return;
572 		}
573 		goto ours;
574 	}
575 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
576 		goto ours;
577 	if (ip->ip_dst.s_addr == INADDR_ANY)
578 		goto ours;
579 
580 	/*
581 	 * FAITH(Firewall Aided Internet Translator)
582 	 */
583 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
584 		if (ip_keepfaith) {
585 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
586 				goto ours;
587 		}
588 		m_freem(m);
589 		return;
590 	}
591 
592 	/*
593 	 * Not for us; forward if possible and desirable.
594 	 */
595 	if (ipforwarding == 0) {
596 		ipstat.ips_cantforward++;
597 		m_freem(m);
598 	} else {
599 #if defined(IPSEC) || defined(FAST_IPSEC)
600 		if (ip_ipsec_fwd(m))
601 			goto bad;
602 #endif /* IPSEC */
603 		ip_forward(m, dchg);
604 	}
605 	return;
606 
607 ours:
608 #ifdef IPSTEALTH
609 	/*
610 	 * IPSTEALTH: Process non-routing options only
611 	 * if the packet is destined for us.
612 	 */
613 	if (ipstealth && hlen > sizeof (struct ip) &&
614 	    ip_dooptions(m, 1))
615 		return;
616 #endif /* IPSTEALTH */
617 
618 	/* Count the packet in the ip address stats */
619 	if (ia != NULL) {
620 		ia->ia_ifa.if_ipackets++;
621 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
622 	}
623 
624 	/*
625 	 * Attempt reassembly; if it succeeds, proceed.
626 	 * ip_reass() will return a different mbuf.
627 	 */
628 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
629 		m = ip_reass(m);
630 		if (m == NULL)
631 			return;
632 		ip = mtod(m, struct ip *);
633 		/* Get the header length of the reassembled packet */
634 		hlen = ip->ip_hl << 2;
635 	}
636 
637 	/*
638 	 * Further protocols expect the packet length to be w/o the
639 	 * IP header.
640 	 */
641 	ip->ip_len -= hlen;
642 
643 #if defined(IPSEC) || defined(FAST_IPSEC)
644 	/*
645 	 * enforce IPsec policy checking if we are seeing last header.
646 	 * note that we do not visit this with protocols with pcb layer
647 	 * code - like udp/tcp/raw ip.
648 	 */
649 	if (ip_ipsec_input(m))
650 		goto bad;
651 #endif /* IPSEC */
652 
653 	/*
654 	 * Switch out to protocol's input routine.
655 	 */
656 	ipstat.ips_delivered++;
657 
658 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
659 	return;
660 bad:
661 	m_freem(m);
662 }
663 
664 /*
665  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
666  * max has slightly different semantics than the sysctl, for historical
667  * reasons.
668  */
669 static void
670 maxnipq_update(void)
671 {
672 
673 	/*
674 	 * -1 for unlimited allocation.
675 	 */
676 	if (maxnipq < 0)
677 		uma_zone_set_max(ipq_zone, 0);
678 	/*
679 	 * Positive number for specific bound.
680 	 */
681 	if (maxnipq > 0)
682 		uma_zone_set_max(ipq_zone, maxnipq);
683 	/*
684 	 * Zero specifies no further fragment queue allocation -- set the
685 	 * bound very low, but rely on implementation elsewhere to actually
686 	 * prevent allocation and reclaim current queues.
687 	 */
688 	if (maxnipq == 0)
689 		uma_zone_set_max(ipq_zone, 1);
690 }
691 
692 static void
693 ipq_zone_change(void *tag)
694 {
695 
696 	if (maxnipq > 0 && maxnipq < (nmbclusters / 32)) {
697 		maxnipq = nmbclusters / 32;
698 		maxnipq_update();
699 	}
700 }
701 
702 static int
703 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
704 {
705 	int error, i;
706 
707 	i = maxnipq;
708 	error = sysctl_handle_int(oidp, &i, 0, req);
709 	if (error || !req->newptr)
710 		return (error);
711 
712 	/*
713 	 * XXXRW: Might be a good idea to sanity check the argument and place
714 	 * an extreme upper bound.
715 	 */
716 	if (i < -1)
717 		return (EINVAL);
718 	maxnipq = i;
719 	maxnipq_update();
720 	return (0);
721 }
722 
723 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
724     NULL, 0, sysctl_maxnipq, "I",
725     "Maximum number of IPv4 fragment reassembly queue entries");
726 
727 /*
728  * Take incoming datagram fragment and try to reassemble it into
729  * whole datagram.  If the argument is the first fragment or one
730  * in between the function will return NULL and store the mbuf
731  * in the fragment chain.  If the argument is the last fragment
732  * the packet will be reassembled and the pointer to the new
733  * mbuf returned for further processing.  Only m_tags attached
734  * to the first packet/fragment are preserved.
735  * The IP header is *NOT* adjusted out of iplen.
736  */
737 
738 struct mbuf *
739 ip_reass(struct mbuf *m)
740 {
741 	struct ip *ip;
742 	struct mbuf *p, *q, *nq, *t;
743 	struct ipq *fp = NULL;
744 	struct ipqhead *head;
745 	int i, hlen, next;
746 	u_int8_t ecn, ecn0;
747 	u_short hash;
748 
749 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
750 	if (maxnipq == 0 || maxfragsperpacket == 0) {
751 		ipstat.ips_fragments++;
752 		ipstat.ips_fragdropped++;
753 		m_freem(m);
754 		return (NULL);
755 	}
756 
757 	ip = mtod(m, struct ip *);
758 	hlen = ip->ip_hl << 2;
759 
760 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
761 	head = &ipq[hash];
762 	IPQ_LOCK();
763 
764 	/*
765 	 * Look for queue of fragments
766 	 * of this datagram.
767 	 */
768 	TAILQ_FOREACH(fp, head, ipq_list)
769 		if (ip->ip_id == fp->ipq_id &&
770 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
771 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
772 #ifdef MAC
773 		    mac_fragment_match(m, fp) &&
774 #endif
775 		    ip->ip_p == fp->ipq_p)
776 			goto found;
777 
778 	fp = NULL;
779 
780 	/*
781 	 * Attempt to trim the number of allocated fragment queues if it
782 	 * exceeds the administrative limit.
783 	 */
784 	if ((nipq > maxnipq) && (maxnipq > 0)) {
785 		/*
786 		 * drop something from the tail of the current queue
787 		 * before proceeding further
788 		 */
789 		struct ipq *q = TAILQ_LAST(head, ipqhead);
790 		if (q == NULL) {   /* gak */
791 			for (i = 0; i < IPREASS_NHASH; i++) {
792 				struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
793 				if (r) {
794 					ipstat.ips_fragtimeout += r->ipq_nfrags;
795 					ip_freef(&ipq[i], r);
796 					break;
797 				}
798 			}
799 		} else {
800 			ipstat.ips_fragtimeout += q->ipq_nfrags;
801 			ip_freef(head, q);
802 		}
803 	}
804 
805 found:
806 	/*
807 	 * Adjust ip_len to not reflect header,
808 	 * convert offset of this to bytes.
809 	 */
810 	ip->ip_len -= hlen;
811 	if (ip->ip_off & IP_MF) {
812 		/*
813 		 * Make sure that fragments have a data length
814 		 * that's a non-zero multiple of 8 bytes.
815 		 */
816 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
817 			ipstat.ips_toosmall++; /* XXX */
818 			goto dropfrag;
819 		}
820 		m->m_flags |= M_FRAG;
821 	} else
822 		m->m_flags &= ~M_FRAG;
823 	ip->ip_off <<= 3;
824 
825 
826 	/*
827 	 * Attempt reassembly; if it succeeds, proceed.
828 	 * ip_reass() will return a different mbuf.
829 	 */
830 	ipstat.ips_fragments++;
831 	m->m_pkthdr.header = ip;
832 
833 	/* Previous ip_reass() started here. */
834 	/*
835 	 * Presence of header sizes in mbufs
836 	 * would confuse code below.
837 	 */
838 	m->m_data += hlen;
839 	m->m_len -= hlen;
840 
841 	/*
842 	 * If first fragment to arrive, create a reassembly queue.
843 	 */
844 	if (fp == NULL) {
845 		fp = uma_zalloc(ipq_zone, M_NOWAIT);
846 		if (fp == NULL)
847 			goto dropfrag;
848 #ifdef MAC
849 		if (mac_init_ipq(fp, M_NOWAIT) != 0) {
850 			uma_zfree(ipq_zone, fp);
851 			fp = NULL;
852 			goto dropfrag;
853 		}
854 		mac_create_ipq(m, fp);
855 #endif
856 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
857 		nipq++;
858 		fp->ipq_nfrags = 1;
859 		fp->ipq_ttl = IPFRAGTTL;
860 		fp->ipq_p = ip->ip_p;
861 		fp->ipq_id = ip->ip_id;
862 		fp->ipq_src = ip->ip_src;
863 		fp->ipq_dst = ip->ip_dst;
864 		fp->ipq_frags = m;
865 		m->m_nextpkt = NULL;
866 		goto done;
867 	} else {
868 		fp->ipq_nfrags++;
869 #ifdef MAC
870 		mac_update_ipq(m, fp);
871 #endif
872 	}
873 
874 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
875 
876 	/*
877 	 * Handle ECN by comparing this segment with the first one;
878 	 * if CE is set, do not lose CE.
879 	 * drop if CE and not-ECT are mixed for the same packet.
880 	 */
881 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
882 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
883 	if (ecn == IPTOS_ECN_CE) {
884 		if (ecn0 == IPTOS_ECN_NOTECT)
885 			goto dropfrag;
886 		if (ecn0 != IPTOS_ECN_CE)
887 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
888 	}
889 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
890 		goto dropfrag;
891 
892 	/*
893 	 * Find a segment which begins after this one does.
894 	 */
895 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
896 		if (GETIP(q)->ip_off > ip->ip_off)
897 			break;
898 
899 	/*
900 	 * If there is a preceding segment, it may provide some of
901 	 * our data already.  If so, drop the data from the incoming
902 	 * segment.  If it provides all of our data, drop us, otherwise
903 	 * stick new segment in the proper place.
904 	 *
905 	 * If some of the data is dropped from the the preceding
906 	 * segment, then it's checksum is invalidated.
907 	 */
908 	if (p) {
909 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
910 		if (i > 0) {
911 			if (i >= ip->ip_len)
912 				goto dropfrag;
913 			m_adj(m, i);
914 			m->m_pkthdr.csum_flags = 0;
915 			ip->ip_off += i;
916 			ip->ip_len -= i;
917 		}
918 		m->m_nextpkt = p->m_nextpkt;
919 		p->m_nextpkt = m;
920 	} else {
921 		m->m_nextpkt = fp->ipq_frags;
922 		fp->ipq_frags = m;
923 	}
924 
925 	/*
926 	 * While we overlap succeeding segments trim them or,
927 	 * if they are completely covered, dequeue them.
928 	 */
929 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
930 	     q = nq) {
931 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
932 		if (i < GETIP(q)->ip_len) {
933 			GETIP(q)->ip_len -= i;
934 			GETIP(q)->ip_off += i;
935 			m_adj(q, i);
936 			q->m_pkthdr.csum_flags = 0;
937 			break;
938 		}
939 		nq = q->m_nextpkt;
940 		m->m_nextpkt = nq;
941 		ipstat.ips_fragdropped++;
942 		fp->ipq_nfrags--;
943 		m_freem(q);
944 	}
945 
946 	/*
947 	 * Check for complete reassembly and perform frag per packet
948 	 * limiting.
949 	 *
950 	 * Frag limiting is performed here so that the nth frag has
951 	 * a chance to complete the packet before we drop the packet.
952 	 * As a result, n+1 frags are actually allowed per packet, but
953 	 * only n will ever be stored. (n = maxfragsperpacket.)
954 	 *
955 	 */
956 	next = 0;
957 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
958 		if (GETIP(q)->ip_off != next) {
959 			if (fp->ipq_nfrags > maxfragsperpacket) {
960 				ipstat.ips_fragdropped += fp->ipq_nfrags;
961 				ip_freef(head, fp);
962 			}
963 			goto done;
964 		}
965 		next += GETIP(q)->ip_len;
966 	}
967 	/* Make sure the last packet didn't have the IP_MF flag */
968 	if (p->m_flags & M_FRAG) {
969 		if (fp->ipq_nfrags > maxfragsperpacket) {
970 			ipstat.ips_fragdropped += fp->ipq_nfrags;
971 			ip_freef(head, fp);
972 		}
973 		goto done;
974 	}
975 
976 	/*
977 	 * Reassembly is complete.  Make sure the packet is a sane size.
978 	 */
979 	q = fp->ipq_frags;
980 	ip = GETIP(q);
981 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
982 		ipstat.ips_toolong++;
983 		ipstat.ips_fragdropped += fp->ipq_nfrags;
984 		ip_freef(head, fp);
985 		goto done;
986 	}
987 
988 	/*
989 	 * Concatenate fragments.
990 	 */
991 	m = q;
992 	t = m->m_next;
993 	m->m_next = NULL;
994 	m_cat(m, t);
995 	nq = q->m_nextpkt;
996 	q->m_nextpkt = NULL;
997 	for (q = nq; q != NULL; q = nq) {
998 		nq = q->m_nextpkt;
999 		q->m_nextpkt = NULL;
1000 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1001 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1002 		m_cat(m, q);
1003 	}
1004 	/*
1005 	 * In order to do checksumming faster we do 'end-around carry' here
1006 	 * (and not in for{} loop), though it implies we are not going to
1007 	 * reassemble more than 64k fragments.
1008 	 */
1009 	m->m_pkthdr.csum_data =
1010 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1011 #ifdef MAC
1012 	mac_create_datagram_from_ipq(fp, m);
1013 	mac_destroy_ipq(fp);
1014 #endif
1015 
1016 	/*
1017 	 * Create header for new ip packet by modifying header of first
1018 	 * packet;  dequeue and discard fragment reassembly header.
1019 	 * Make header visible.
1020 	 */
1021 	ip->ip_len = (ip->ip_hl << 2) + next;
1022 	ip->ip_src = fp->ipq_src;
1023 	ip->ip_dst = fp->ipq_dst;
1024 	TAILQ_REMOVE(head, fp, ipq_list);
1025 	nipq--;
1026 	uma_zfree(ipq_zone, fp);
1027 	m->m_len += (ip->ip_hl << 2);
1028 	m->m_data -= (ip->ip_hl << 2);
1029 	/* some debugging cruft by sklower, below, will go away soon */
1030 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1031 		m_fixhdr(m);
1032 	ipstat.ips_reassembled++;
1033 	IPQ_UNLOCK();
1034 	return (m);
1035 
1036 dropfrag:
1037 	ipstat.ips_fragdropped++;
1038 	if (fp != NULL)
1039 		fp->ipq_nfrags--;
1040 	m_freem(m);
1041 done:
1042 	IPQ_UNLOCK();
1043 	return (NULL);
1044 
1045 #undef GETIP
1046 }
1047 
1048 /*
1049  * Free a fragment reassembly header and all
1050  * associated datagrams.
1051  */
1052 static void
1053 ip_freef(fhp, fp)
1054 	struct ipqhead *fhp;
1055 	struct ipq *fp;
1056 {
1057 	register struct mbuf *q;
1058 
1059 	IPQ_LOCK_ASSERT();
1060 
1061 	while (fp->ipq_frags) {
1062 		q = fp->ipq_frags;
1063 		fp->ipq_frags = q->m_nextpkt;
1064 		m_freem(q);
1065 	}
1066 	TAILQ_REMOVE(fhp, fp, ipq_list);
1067 	uma_zfree(ipq_zone, fp);
1068 	nipq--;
1069 }
1070 
1071 /*
1072  * IP timer processing;
1073  * if a timer expires on a reassembly
1074  * queue, discard it.
1075  */
1076 void
1077 ip_slowtimo()
1078 {
1079 	register struct ipq *fp;
1080 	int i;
1081 
1082 	IPQ_LOCK();
1083 	for (i = 0; i < IPREASS_NHASH; i++) {
1084 		for(fp = TAILQ_FIRST(&ipq[i]); fp;) {
1085 			struct ipq *fpp;
1086 
1087 			fpp = fp;
1088 			fp = TAILQ_NEXT(fp, ipq_list);
1089 			if(--fpp->ipq_ttl == 0) {
1090 				ipstat.ips_fragtimeout += fpp->ipq_nfrags;
1091 				ip_freef(&ipq[i], fpp);
1092 			}
1093 		}
1094 	}
1095 	/*
1096 	 * If we are over the maximum number of fragments
1097 	 * (due to the limit being lowered), drain off
1098 	 * enough to get down to the new limit.
1099 	 */
1100 	if (maxnipq >= 0 && nipq > maxnipq) {
1101 		for (i = 0; i < IPREASS_NHASH; i++) {
1102 			while (nipq > maxnipq && !TAILQ_EMPTY(&ipq[i])) {
1103 				ipstat.ips_fragdropped +=
1104 				    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1105 				ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1106 			}
1107 		}
1108 	}
1109 	IPQ_UNLOCK();
1110 }
1111 
1112 /*
1113  * Drain off all datagram fragments.
1114  */
1115 void
1116 ip_drain()
1117 {
1118 	int     i;
1119 
1120 	IPQ_LOCK();
1121 	for (i = 0; i < IPREASS_NHASH; i++) {
1122 		while(!TAILQ_EMPTY(&ipq[i])) {
1123 			ipstat.ips_fragdropped +=
1124 			    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1125 			ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1126 		}
1127 	}
1128 	IPQ_UNLOCK();
1129 	in_rtqdrain();
1130 }
1131 
1132 /*
1133  * The protocol to be inserted into ip_protox[] must be already registered
1134  * in inetsw[], either statically or through pf_proto_register().
1135  */
1136 int
1137 ipproto_register(u_char ipproto)
1138 {
1139 	struct protosw *pr;
1140 
1141 	/* Sanity checks. */
1142 	if (ipproto == 0)
1143 		return (EPROTONOSUPPORT);
1144 
1145 	/*
1146 	 * The protocol slot must not be occupied by another protocol
1147 	 * already.  An index pointing to IPPROTO_RAW is unused.
1148 	 */
1149 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1150 	if (pr == NULL)
1151 		return (EPFNOSUPPORT);
1152 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1153 		return (EEXIST);
1154 
1155 	/* Find the protocol position in inetsw[] and set the index. */
1156 	for (pr = inetdomain.dom_protosw;
1157 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1158 		if (pr->pr_domain->dom_family == PF_INET &&
1159 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1160 			/* Be careful to only index valid IP protocols. */
1161 			if (pr->pr_protocol < IPPROTO_MAX) {
1162 				ip_protox[pr->pr_protocol] = pr - inetsw;
1163 				return (0);
1164 			} else
1165 				return (EINVAL);
1166 		}
1167 	}
1168 	return (EPROTONOSUPPORT);
1169 }
1170 
1171 int
1172 ipproto_unregister(u_char ipproto)
1173 {
1174 	struct protosw *pr;
1175 
1176 	/* Sanity checks. */
1177 	if (ipproto == 0)
1178 		return (EPROTONOSUPPORT);
1179 
1180 	/* Check if the protocol was indeed registered. */
1181 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1182 	if (pr == NULL)
1183 		return (EPFNOSUPPORT);
1184 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1185 		return (ENOENT);
1186 
1187 	/* Reset the protocol slot to IPPROTO_RAW. */
1188 	ip_protox[ipproto] = pr - inetsw;
1189 	return (0);
1190 }
1191 
1192 /*
1193  * Given address of next destination (final or next hop),
1194  * return internet address info of interface to be used to get there.
1195  */
1196 struct in_ifaddr *
1197 ip_rtaddr(dst)
1198 	struct in_addr dst;
1199 {
1200 	struct route sro;
1201 	struct sockaddr_in *sin;
1202 	struct in_ifaddr *ifa;
1203 
1204 	bzero(&sro, sizeof(sro));
1205 	sin = (struct sockaddr_in *)&sro.ro_dst;
1206 	sin->sin_family = AF_INET;
1207 	sin->sin_len = sizeof(*sin);
1208 	sin->sin_addr = dst;
1209 	rtalloc_ign(&sro, RTF_CLONING);
1210 
1211 	if (sro.ro_rt == NULL)
1212 		return (NULL);
1213 
1214 	ifa = ifatoia(sro.ro_rt->rt_ifa);
1215 	RTFREE(sro.ro_rt);
1216 	return (ifa);
1217 }
1218 
1219 u_char inetctlerrmap[PRC_NCMDS] = {
1220 	0,		0,		0,		0,
1221 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1222 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1223 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1224 	0,		0,		EHOSTUNREACH,	0,
1225 	ENOPROTOOPT,	ECONNREFUSED
1226 };
1227 
1228 /*
1229  * Forward a packet.  If some error occurs return the sender
1230  * an icmp packet.  Note we can't always generate a meaningful
1231  * icmp message because icmp doesn't have a large enough repertoire
1232  * of codes and types.
1233  *
1234  * If not forwarding, just drop the packet.  This could be confusing
1235  * if ipforwarding was zero but some routing protocol was advancing
1236  * us as a gateway to somewhere.  However, we must let the routing
1237  * protocol deal with that.
1238  *
1239  * The srcrt parameter indicates whether the packet is being forwarded
1240  * via a source route.
1241  */
1242 void
1243 ip_forward(struct mbuf *m, int srcrt)
1244 {
1245 	struct ip *ip = mtod(m, struct ip *);
1246 	struct in_ifaddr *ia = NULL;
1247 	struct mbuf *mcopy;
1248 	struct in_addr dest;
1249 	int error, type = 0, code = 0, mtu = 0;
1250 
1251 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1252 		ipstat.ips_cantforward++;
1253 		m_freem(m);
1254 		return;
1255 	}
1256 #ifdef IPSTEALTH
1257 	if (!ipstealth) {
1258 #endif
1259 		if (ip->ip_ttl <= IPTTLDEC) {
1260 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1261 			    0, 0);
1262 			return;
1263 		}
1264 #ifdef IPSTEALTH
1265 	}
1266 #endif
1267 
1268 	if (!srcrt && (ia = ip_rtaddr(ip->ip_dst)) == NULL) {
1269 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1270 		return;
1271 	}
1272 
1273 	/*
1274 	 * Save the IP header and at most 8 bytes of the payload,
1275 	 * in case we need to generate an ICMP message to the src.
1276 	 *
1277 	 * XXX this can be optimized a lot by saving the data in a local
1278 	 * buffer on the stack (72 bytes at most), and only allocating the
1279 	 * mbuf if really necessary. The vast majority of the packets
1280 	 * are forwarded without having to send an ICMP back (either
1281 	 * because unnecessary, or because rate limited), so we are
1282 	 * really we are wasting a lot of work here.
1283 	 *
1284 	 * We don't use m_copy() because it might return a reference
1285 	 * to a shared cluster. Both this function and ip_output()
1286 	 * assume exclusive access to the IP header in `m', so any
1287 	 * data in a cluster may change before we reach icmp_error().
1288 	 */
1289 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1290 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1291 		/*
1292 		 * It's probably ok if the pkthdr dup fails (because
1293 		 * the deep copy of the tag chain failed), but for now
1294 		 * be conservative and just discard the copy since
1295 		 * code below may some day want the tags.
1296 		 */
1297 		m_free(mcopy);
1298 		mcopy = NULL;
1299 	}
1300 	if (mcopy != NULL) {
1301 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1302 		mcopy->m_pkthdr.len = mcopy->m_len;
1303 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1304 	}
1305 
1306 #ifdef IPSTEALTH
1307 	if (!ipstealth) {
1308 #endif
1309 		ip->ip_ttl -= IPTTLDEC;
1310 #ifdef IPSTEALTH
1311 	}
1312 #endif
1313 
1314 	/*
1315 	 * If forwarding packet using same interface that it came in on,
1316 	 * perhaps should send a redirect to sender to shortcut a hop.
1317 	 * Only send redirect if source is sending directly to us,
1318 	 * and if packet was not source routed (or has any options).
1319 	 * Also, don't send redirect if forwarding using a default route
1320 	 * or a route modified by a redirect.
1321 	 */
1322 	dest.s_addr = 0;
1323 	if (!srcrt && ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1324 		struct sockaddr_in *sin;
1325 		struct route ro;
1326 		struct rtentry *rt;
1327 
1328 		bzero(&ro, sizeof(ro));
1329 		sin = (struct sockaddr_in *)&ro.ro_dst;
1330 		sin->sin_family = AF_INET;
1331 		sin->sin_len = sizeof(*sin);
1332 		sin->sin_addr = ip->ip_dst;
1333 		rtalloc_ign(&ro, RTF_CLONING);
1334 
1335 		rt = ro.ro_rt;
1336 
1337 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1338 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1339 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1340 			u_long src = ntohl(ip->ip_src.s_addr);
1341 
1342 			if (RTA(rt) &&
1343 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1344 				if (rt->rt_flags & RTF_GATEWAY)
1345 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1346 				else
1347 					dest.s_addr = ip->ip_dst.s_addr;
1348 				/* Router requirements says to only send host redirects */
1349 				type = ICMP_REDIRECT;
1350 				code = ICMP_REDIRECT_HOST;
1351 			}
1352 		}
1353 		if (rt)
1354 			RTFREE(rt);
1355 	}
1356 
1357 	error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1358 	if (error)
1359 		ipstat.ips_cantforward++;
1360 	else {
1361 		ipstat.ips_forward++;
1362 		if (type)
1363 			ipstat.ips_redirectsent++;
1364 		else {
1365 			if (mcopy)
1366 				m_freem(mcopy);
1367 			return;
1368 		}
1369 	}
1370 	if (mcopy == NULL)
1371 		return;
1372 
1373 	switch (error) {
1374 
1375 	case 0:				/* forwarded, but need redirect */
1376 		/* type, code set above */
1377 		break;
1378 
1379 	case ENETUNREACH:		/* shouldn't happen, checked above */
1380 	case EHOSTUNREACH:
1381 	case ENETDOWN:
1382 	case EHOSTDOWN:
1383 	default:
1384 		type = ICMP_UNREACH;
1385 		code = ICMP_UNREACH_HOST;
1386 		break;
1387 
1388 	case EMSGSIZE:
1389 		type = ICMP_UNREACH;
1390 		code = ICMP_UNREACH_NEEDFRAG;
1391 
1392 #if defined(IPSEC) || defined(FAST_IPSEC)
1393 		mtu = ip_ipsec_mtu(m);
1394 #endif /* IPSEC */
1395 		/*
1396 		 * If the MTU wasn't set before use the interface mtu or
1397 		 * fall back to the next smaller mtu step compared to the
1398 		 * current packet size.
1399 		 */
1400 		if (mtu == 0) {
1401 			if (ia != NULL)
1402 				mtu = ia->ia_ifp->if_mtu;
1403 			else
1404 				mtu = ip_next_mtu(ip->ip_len, 0);
1405 		}
1406 		ipstat.ips_cantfrag++;
1407 		break;
1408 
1409 	case ENOBUFS:
1410 		/*
1411 		 * A router should not generate ICMP_SOURCEQUENCH as
1412 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1413 		 * Source quench could be a big problem under DoS attacks,
1414 		 * or if the underlying interface is rate-limited.
1415 		 * Those who need source quench packets may re-enable them
1416 		 * via the net.inet.ip.sendsourcequench sysctl.
1417 		 */
1418 		if (ip_sendsourcequench == 0) {
1419 			m_freem(mcopy);
1420 			return;
1421 		} else {
1422 			type = ICMP_SOURCEQUENCH;
1423 			code = 0;
1424 		}
1425 		break;
1426 
1427 	case EACCES:			/* ipfw denied packet */
1428 		m_freem(mcopy);
1429 		return;
1430 	}
1431 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1432 }
1433 
1434 void
1435 ip_savecontrol(inp, mp, ip, m)
1436 	register struct inpcb *inp;
1437 	register struct mbuf **mp;
1438 	register struct ip *ip;
1439 	register struct mbuf *m;
1440 {
1441 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1442 		struct bintime bt;
1443 
1444 		bintime(&bt);
1445 		if (inp->inp_socket->so_options & SO_BINTIME) {
1446 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1447 			SCM_BINTIME, SOL_SOCKET);
1448 			if (*mp)
1449 				mp = &(*mp)->m_next;
1450 		}
1451 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1452 			struct timeval tv;
1453 
1454 			bintime2timeval(&bt, &tv);
1455 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1456 				SCM_TIMESTAMP, SOL_SOCKET);
1457 			if (*mp)
1458 				mp = &(*mp)->m_next;
1459 		}
1460 	}
1461 	if (inp->inp_flags & INP_RECVDSTADDR) {
1462 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1463 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1464 		if (*mp)
1465 			mp = &(*mp)->m_next;
1466 	}
1467 	if (inp->inp_flags & INP_RECVTTL) {
1468 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1469 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1470 		if (*mp)
1471 			mp = &(*mp)->m_next;
1472 	}
1473 #ifdef notyet
1474 	/* XXX
1475 	 * Moving these out of udp_input() made them even more broken
1476 	 * than they already were.
1477 	 */
1478 	/* options were tossed already */
1479 	if (inp->inp_flags & INP_RECVOPTS) {
1480 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1481 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1482 		if (*mp)
1483 			mp = &(*mp)->m_next;
1484 	}
1485 	/* ip_srcroute doesn't do what we want here, need to fix */
1486 	if (inp->inp_flags & INP_RECVRETOPTS) {
1487 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1488 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1489 		if (*mp)
1490 			mp = &(*mp)->m_next;
1491 	}
1492 #endif
1493 	if (inp->inp_flags & INP_RECVIF) {
1494 		struct ifnet *ifp;
1495 		struct sdlbuf {
1496 			struct sockaddr_dl sdl;
1497 			u_char	pad[32];
1498 		} sdlbuf;
1499 		struct sockaddr_dl *sdp;
1500 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1501 
1502 		if (((ifp = m->m_pkthdr.rcvif))
1503 		&& ( ifp->if_index && (ifp->if_index <= if_index))) {
1504 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1505 			/*
1506 			 * Change our mind and don't try copy.
1507 			 */
1508 			if ((sdp->sdl_family != AF_LINK)
1509 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1510 				goto makedummy;
1511 			}
1512 			bcopy(sdp, sdl2, sdp->sdl_len);
1513 		} else {
1514 makedummy:
1515 			sdl2->sdl_len
1516 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1517 			sdl2->sdl_family = AF_LINK;
1518 			sdl2->sdl_index = 0;
1519 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1520 		}
1521 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1522 			IP_RECVIF, IPPROTO_IP);
1523 		if (*mp)
1524 			mp = &(*mp)->m_next;
1525 	}
1526 }
1527 
1528 /*
1529  * XXX these routines are called from the upper part of the kernel.
1530  * They need to be locked when we remove Giant.
1531  *
1532  * They could also be moved to ip_mroute.c, since all the RSVP
1533  *  handling is done there already.
1534  */
1535 static int ip_rsvp_on;
1536 struct socket *ip_rsvpd;
1537 int
1538 ip_rsvp_init(struct socket *so)
1539 {
1540 	if (so->so_type != SOCK_RAW ||
1541 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1542 		return EOPNOTSUPP;
1543 
1544 	if (ip_rsvpd != NULL)
1545 		return EADDRINUSE;
1546 
1547 	ip_rsvpd = so;
1548 	/*
1549 	 * This may seem silly, but we need to be sure we don't over-increment
1550 	 * the RSVP counter, in case something slips up.
1551 	 */
1552 	if (!ip_rsvp_on) {
1553 		ip_rsvp_on = 1;
1554 		rsvp_on++;
1555 	}
1556 
1557 	return 0;
1558 }
1559 
1560 int
1561 ip_rsvp_done(void)
1562 {
1563 	ip_rsvpd = NULL;
1564 	/*
1565 	 * This may seem silly, but we need to be sure we don't over-decrement
1566 	 * the RSVP counter, in case something slips up.
1567 	 */
1568 	if (ip_rsvp_on) {
1569 		ip_rsvp_on = 0;
1570 		rsvp_on--;
1571 	}
1572 	return 0;
1573 }
1574 
1575 void
1576 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1577 {
1578 	if (rsvp_input_p) { /* call the real one if loaded */
1579 		rsvp_input_p(m, off);
1580 		return;
1581 	}
1582 
1583 	/* Can still get packets with rsvp_on = 0 if there is a local member
1584 	 * of the group to which the RSVP packet is addressed.  But in this
1585 	 * case we want to throw the packet away.
1586 	 */
1587 
1588 	if (!rsvp_on) {
1589 		m_freem(m);
1590 		return;
1591 	}
1592 
1593 	if (ip_rsvpd != NULL) {
1594 		rip_input(m, off);
1595 		return;
1596 	}
1597 	/* Drop the packet */
1598 	m_freem(m);
1599 }
1600