xref: /freebsd/sys/netinet/ip_input.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
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 
171 static int	maxnipq;	/* Administrative limit on # reass queues. */
172 static int	nipq = 0;	/* Total # of reass queues */
173 SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD, &nipq, 0,
174 	"Current number of IPv4 fragment reassembly queue entries");
175 
176 static int	maxfragsperpacket;
177 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
178 	&maxfragsperpacket, 0,
179 	"Maximum number of IPv4 fragments allowed per packet");
180 
181 struct callout	ipport_tick_callout;
182 
183 #ifdef IPCTL_DEFMTU
184 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
185     &ip_mtu, 0, "Default MTU");
186 #endif
187 
188 #ifdef IPSTEALTH
189 int	ipstealth = 0;
190 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
191     &ipstealth, 0, "");
192 #endif
193 
194 /*
195  * ipfw_ether and ipfw_bridge hooks.
196  * XXX: Temporary until those are converted to pfil_hooks as well.
197  */
198 ip_fw_chk_t *ip_fw_chk_ptr = NULL;
199 ip_dn_io_t *ip_dn_io_ptr = NULL;
200 int fw_enable = 1;
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 
260 	/* Initialize various other remaining things. */
261 	ip_id = time_second & 0xffff;
262 	ipintrq.ifq_maxlen = ipqmaxlen;
263 	mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
264 	netisr_register(NETISR_IP, ip_input, &ipintrq, NETISR_MPSAFE);
265 }
266 
267 void ip_fini(xtp)
268 	void *xtp;
269 {
270 	callout_stop(&ipport_tick_callout);
271 }
272 
273 /*
274  * Ip input routine.  Checksum and byte swap header.  If fragmented
275  * try to reassemble.  Process options.  Pass to next level.
276  */
277 void
278 ip_input(struct mbuf *m)
279 {
280 	struct ip *ip = NULL;
281 	struct in_ifaddr *ia = NULL;
282 	struct ifaddr *ifa;
283 	int    checkif, hlen = 0;
284 	u_short sum;
285 	int dchg = 0;				/* dest changed after fw */
286 	struct in_addr odst;			/* original dst address */
287 
288   	M_ASSERTPKTHDR(m);
289 
290 	if (m->m_flags & M_FASTFWD_OURS) {
291 		/*
292 		 * Firewall or NAT changed destination to local.
293 		 * We expect ip_len and ip_off to be in host byte order.
294 		 */
295 		m->m_flags &= ~M_FASTFWD_OURS;
296 		/* Set up some basics that will be used later. */
297 		ip = mtod(m, struct ip *);
298 		hlen = ip->ip_hl << 2;
299   		goto ours;
300   	}
301 
302 	ipstat.ips_total++;
303 
304 	if (m->m_pkthdr.len < sizeof(struct ip))
305 		goto tooshort;
306 
307 	if (m->m_len < sizeof (struct ip) &&
308 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
309 		ipstat.ips_toosmall++;
310 		return;
311 	}
312 	ip = mtod(m, struct ip *);
313 
314 	if (ip->ip_v != IPVERSION) {
315 		ipstat.ips_badvers++;
316 		goto bad;
317 	}
318 
319 	hlen = ip->ip_hl << 2;
320 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
321 		ipstat.ips_badhlen++;
322 		goto bad;
323 	}
324 	if (hlen > m->m_len) {
325 		if ((m = m_pullup(m, hlen)) == NULL) {
326 			ipstat.ips_badhlen++;
327 			return;
328 		}
329 		ip = mtod(m, struct ip *);
330 	}
331 
332 	/* 127/8 must not appear on wire - RFC1122 */
333 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
334 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
335 		if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
336 			ipstat.ips_badaddr++;
337 			goto bad;
338 		}
339 	}
340 
341 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
342 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
343 	} else {
344 		if (hlen == sizeof(struct ip)) {
345 			sum = in_cksum_hdr(ip);
346 		} else {
347 			sum = in_cksum(m, hlen);
348 		}
349 	}
350 	if (sum) {
351 		ipstat.ips_badsum++;
352 		goto bad;
353 	}
354 
355 #ifdef ALTQ
356 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
357 		/* packet is dropped by traffic conditioner */
358 		return;
359 #endif
360 
361 	/*
362 	 * Convert fields to host representation.
363 	 */
364 	ip->ip_len = ntohs(ip->ip_len);
365 	if (ip->ip_len < hlen) {
366 		ipstat.ips_badlen++;
367 		goto bad;
368 	}
369 	ip->ip_off = ntohs(ip->ip_off);
370 
371 	/*
372 	 * Check that the amount of data in the buffers
373 	 * is as at least much as the IP header would have us expect.
374 	 * Trim mbufs if longer than we expect.
375 	 * Drop packet if shorter than we expect.
376 	 */
377 	if (m->m_pkthdr.len < ip->ip_len) {
378 tooshort:
379 		ipstat.ips_tooshort++;
380 		goto bad;
381 	}
382 	if (m->m_pkthdr.len > ip->ip_len) {
383 		if (m->m_len == m->m_pkthdr.len) {
384 			m->m_len = ip->ip_len;
385 			m->m_pkthdr.len = ip->ip_len;
386 		} else
387 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
388 	}
389 #if defined(IPSEC) || defined(FAST_IPSEC)
390 	/*
391 	 * Bypass packet filtering for packets from a tunnel (gif).
392 	 */
393 	if (ip_ipsec_filtergif(m))
394 		goto passin;
395 #endif /* IPSEC */
396 
397 	/*
398 	 * Run through list of hooks for input packets.
399 	 *
400 	 * NB: Beware of the destination address changing (e.g.
401 	 *     by NAT rewriting).  When this happens, tell
402 	 *     ip_forward to do the right thing.
403 	 */
404 
405 	/* Jump over all PFIL processing if hooks are not active. */
406 	if (!PFIL_HOOKED(&inet_pfil_hook))
407 		goto passin;
408 
409 	odst = ip->ip_dst;
410 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
411 	    PFIL_IN, NULL) != 0)
412 		return;
413 	if (m == NULL)			/* consumed by filter */
414 		return;
415 
416 	ip = mtod(m, struct ip *);
417 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
418 
419 #ifdef IPFIREWALL_FORWARD
420 	if (m->m_flags & M_FASTFWD_OURS) {
421 		m->m_flags &= ~M_FASTFWD_OURS;
422 		goto ours;
423 	}
424 #ifndef IPFIREWALL_FORWARD_EXTENDED
425 	dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL);
426 #else
427 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
428 		/*
429 		 * Directly ship on the packet.  This allows to forward packets
430 		 * that were destined for us to some other directly connected
431 		 * host.
432 		 */
433 		ip_forward(m, dchg);
434 		return;
435 	}
436 #endif /* IPFIREWALL_FORWARD_EXTENDED */
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 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
533 		struct in_multi *inm;
534 		if (ip_mrouter) {
535 			/*
536 			 * If we are acting as a multicast router, all
537 			 * incoming multicast packets are passed to the
538 			 * kernel-level multicast forwarding function.
539 			 * The packet is returned (relatively) intact; if
540 			 * ip_mforward() returns a non-zero value, the packet
541 			 * must be discarded, else it may be accepted below.
542 			 */
543 			if (ip_mforward &&
544 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
545 				ipstat.ips_cantforward++;
546 				m_freem(m);
547 				return;
548 			}
549 
550 			/*
551 			 * The process-level routing daemon needs to receive
552 			 * all multicast IGMP packets, whether or not this
553 			 * host belongs to their destination groups.
554 			 */
555 			if (ip->ip_p == IPPROTO_IGMP)
556 				goto ours;
557 			ipstat.ips_forward++;
558 		}
559 		/*
560 		 * See if we belong to the destination multicast group on the
561 		 * arrival interface.
562 		 */
563 		IN_MULTI_LOCK();
564 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
565 		IN_MULTI_UNLOCK();
566 		if (inm == NULL) {
567 			ipstat.ips_notmember++;
568 			m_freem(m);
569 			return;
570 		}
571 		goto ours;
572 	}
573 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
574 		goto ours;
575 	if (ip->ip_dst.s_addr == INADDR_ANY)
576 		goto ours;
577 
578 	/*
579 	 * FAITH(Firewall Aided Internet Translator)
580 	 */
581 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
582 		if (ip_keepfaith) {
583 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
584 				goto ours;
585 		}
586 		m_freem(m);
587 		return;
588 	}
589 
590 	/*
591 	 * Not for us; forward if possible and desirable.
592 	 */
593 	if (ipforwarding == 0) {
594 		ipstat.ips_cantforward++;
595 		m_freem(m);
596 	} else {
597 #if defined(IPSEC) || defined(FAST_IPSEC)
598 		if (ip_ipsec_fwd(m))
599 			goto bad;
600 #endif /* IPSEC */
601 		ip_forward(m, dchg);
602 	}
603 	return;
604 
605 ours:
606 #ifdef IPSTEALTH
607 	/*
608 	 * IPSTEALTH: Process non-routing options only
609 	 * if the packet is destined for us.
610 	 */
611 	if (ipstealth && hlen > sizeof (struct ip) &&
612 	    ip_dooptions(m, 1))
613 		return;
614 #endif /* IPSTEALTH */
615 
616 	/* Count the packet in the ip address stats */
617 	if (ia != NULL) {
618 		ia->ia_ifa.if_ipackets++;
619 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
620 	}
621 
622 	/*
623 	 * Attempt reassembly; if it succeeds, proceed.
624 	 * ip_reass() will return a different mbuf.
625 	 */
626 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
627 		m = ip_reass(m);
628 		if (m == NULL)
629 			return;
630 		ip = mtod(m, struct ip *);
631 		/* Get the header length of the reassembled packet */
632 		hlen = ip->ip_hl << 2;
633 	}
634 
635 	/*
636 	 * Further protocols expect the packet length to be w/o the
637 	 * IP header.
638 	 */
639 	ip->ip_len -= hlen;
640 
641 #if defined(IPSEC) || defined(FAST_IPSEC)
642 	/*
643 	 * enforce IPsec policy checking if we are seeing last header.
644 	 * note that we do not visit this with protocols with pcb layer
645 	 * code - like udp/tcp/raw ip.
646 	 */
647 	if (ip_ipsec_input(m))
648 		goto bad;
649 #endif /* IPSEC */
650 
651 	/*
652 	 * Switch out to protocol's input routine.
653 	 */
654 	ipstat.ips_delivered++;
655 
656 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
657 	return;
658 bad:
659 	m_freem(m);
660 }
661 
662 /*
663  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
664  * max has slightly different semantics than the sysctl, for historical
665  * reasons.
666  */
667 static void
668 maxnipq_update(void)
669 {
670 
671 	/*
672 	 * -1 for unlimited allocation.
673 	 */
674 	if (maxnipq < 0)
675 		uma_zone_set_max(ipq_zone, 0);
676 	/*
677 	 * Positive number for specific bound.
678 	 */
679 	if (maxnipq > 0)
680 		uma_zone_set_max(ipq_zone, maxnipq);
681 	/*
682 	 * Zero specifies no further fragment queue allocation -- set the
683 	 * bound very low, but rely on implementation elsewhere to actually
684 	 * prevent allocation and reclaim current queues.
685 	 */
686 	if (maxnipq == 0)
687 		uma_zone_set_max(ipq_zone, 1);
688 }
689 
690 static int
691 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
692 {
693 	int error, i;
694 
695 	i = maxnipq;
696 	error = sysctl_handle_int(oidp, &i, 0, req);
697 	if (error || !req->newptr)
698 		return (error);
699 
700 	/*
701 	 * XXXRW: Might be a good idea to sanity check the argument and place
702 	 * an extreme upper bound.
703 	 */
704 	if (i < -1)
705 		return (EINVAL);
706 	maxnipq = i;
707 	maxnipq_update();
708 	return (0);
709 }
710 
711 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
712     NULL, 0, sysctl_maxnipq, "I",
713     "Maximum number of IPv4 fragment reassembly queue entries");
714 
715 /*
716  * Take incoming datagram fragment and try to reassemble it into
717  * whole datagram.  If the argument is the first fragment or one
718  * in between the function will return NULL and store the mbuf
719  * in the fragment chain.  If the argument is the last fragment
720  * the packet will be reassembled and the pointer to the new
721  * mbuf returned for further processing.  Only m_tags attached
722  * to the first packet/fragment are preserved.
723  * The IP header is *NOT* adjusted out of iplen.
724  */
725 
726 struct mbuf *
727 ip_reass(struct mbuf *m)
728 {
729 	struct ip *ip;
730 	struct mbuf *p, *q, *nq, *t;
731 	struct ipq *fp = NULL;
732 	struct ipqhead *head;
733 	int i, hlen, next;
734 	u_int8_t ecn, ecn0;
735 	u_short hash;
736 
737 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
738 	if (maxnipq == 0 || maxfragsperpacket == 0) {
739 		ipstat.ips_fragments++;
740 		ipstat.ips_fragdropped++;
741 		m_freem(m);
742 		return (NULL);
743 	}
744 
745 	ip = mtod(m, struct ip *);
746 	hlen = ip->ip_hl << 2;
747 
748 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
749 	head = &ipq[hash];
750 	IPQ_LOCK();
751 
752 	/*
753 	 * Look for queue of fragments
754 	 * of this datagram.
755 	 */
756 	TAILQ_FOREACH(fp, head, ipq_list)
757 		if (ip->ip_id == fp->ipq_id &&
758 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
759 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
760 #ifdef MAC
761 		    mac_fragment_match(m, fp) &&
762 #endif
763 		    ip->ip_p == fp->ipq_p)
764 			goto found;
765 
766 	fp = NULL;
767 
768 	/*
769 	 * Attempt to trim the number of allocated fragment queues if it
770 	 * exceeds the administrative limit.
771 	 */
772 	if ((nipq > maxnipq) && (maxnipq > 0)) {
773 		/*
774 		 * drop something from the tail of the current queue
775 		 * before proceeding further
776 		 */
777 		struct ipq *q = TAILQ_LAST(head, ipqhead);
778 		if (q == NULL) {   /* gak */
779 			for (i = 0; i < IPREASS_NHASH; i++) {
780 				struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
781 				if (r) {
782 					ipstat.ips_fragtimeout += r->ipq_nfrags;
783 					ip_freef(&ipq[i], r);
784 					break;
785 				}
786 			}
787 		} else {
788 			ipstat.ips_fragtimeout += q->ipq_nfrags;
789 			ip_freef(head, q);
790 		}
791 	}
792 
793 found:
794 	/*
795 	 * Adjust ip_len to not reflect header,
796 	 * convert offset of this to bytes.
797 	 */
798 	ip->ip_len -= hlen;
799 	if (ip->ip_off & IP_MF) {
800 		/*
801 		 * Make sure that fragments have a data length
802 		 * that's a non-zero multiple of 8 bytes.
803 		 */
804 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
805 			ipstat.ips_toosmall++; /* XXX */
806 			goto dropfrag;
807 		}
808 		m->m_flags |= M_FRAG;
809 	} else
810 		m->m_flags &= ~M_FRAG;
811 	ip->ip_off <<= 3;
812 
813 
814 	/*
815 	 * Attempt reassembly; if it succeeds, proceed.
816 	 * ip_reass() will return a different mbuf.
817 	 */
818 	ipstat.ips_fragments++;
819 	m->m_pkthdr.header = ip;
820 
821 	/* Previous ip_reass() started here. */
822 	/*
823 	 * Presence of header sizes in mbufs
824 	 * would confuse code below.
825 	 */
826 	m->m_data += hlen;
827 	m->m_len -= hlen;
828 
829 	/*
830 	 * If first fragment to arrive, create a reassembly queue.
831 	 */
832 	if (fp == NULL) {
833 		fp = uma_zalloc(ipq_zone, M_NOWAIT);
834 		if (fp == NULL)
835 			goto dropfrag;
836 #ifdef MAC
837 		if (mac_init_ipq(fp, M_NOWAIT) != 0) {
838 			uma_zfree(ipq_zone, fp);
839 			goto dropfrag;
840 		}
841 		mac_create_ipq(m, fp);
842 #endif
843 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
844 		nipq++;
845 		fp->ipq_nfrags = 1;
846 		fp->ipq_ttl = IPFRAGTTL;
847 		fp->ipq_p = ip->ip_p;
848 		fp->ipq_id = ip->ip_id;
849 		fp->ipq_src = ip->ip_src;
850 		fp->ipq_dst = ip->ip_dst;
851 		fp->ipq_frags = m;
852 		m->m_nextpkt = NULL;
853 		goto done;
854 	} else {
855 		fp->ipq_nfrags++;
856 #ifdef MAC
857 		mac_update_ipq(m, fp);
858 #endif
859 	}
860 
861 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
862 
863 	/*
864 	 * Handle ECN by comparing this segment with the first one;
865 	 * if CE is set, do not lose CE.
866 	 * drop if CE and not-ECT are mixed for the same packet.
867 	 */
868 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
869 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
870 	if (ecn == IPTOS_ECN_CE) {
871 		if (ecn0 == IPTOS_ECN_NOTECT)
872 			goto dropfrag;
873 		if (ecn0 != IPTOS_ECN_CE)
874 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
875 	}
876 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
877 		goto dropfrag;
878 
879 	/*
880 	 * Find a segment which begins after this one does.
881 	 */
882 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
883 		if (GETIP(q)->ip_off > ip->ip_off)
884 			break;
885 
886 	/*
887 	 * If there is a preceding segment, it may provide some of
888 	 * our data already.  If so, drop the data from the incoming
889 	 * segment.  If it provides all of our data, drop us, otherwise
890 	 * stick new segment in the proper place.
891 	 *
892 	 * If some of the data is dropped from the the preceding
893 	 * segment, then it's checksum is invalidated.
894 	 */
895 	if (p) {
896 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
897 		if (i > 0) {
898 			if (i >= ip->ip_len)
899 				goto dropfrag;
900 			m_adj(m, i);
901 			m->m_pkthdr.csum_flags = 0;
902 			ip->ip_off += i;
903 			ip->ip_len -= i;
904 		}
905 		m->m_nextpkt = p->m_nextpkt;
906 		p->m_nextpkt = m;
907 	} else {
908 		m->m_nextpkt = fp->ipq_frags;
909 		fp->ipq_frags = m;
910 	}
911 
912 	/*
913 	 * While we overlap succeeding segments trim them or,
914 	 * if they are completely covered, dequeue them.
915 	 */
916 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
917 	     q = nq) {
918 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
919 		if (i < GETIP(q)->ip_len) {
920 			GETIP(q)->ip_len -= i;
921 			GETIP(q)->ip_off += i;
922 			m_adj(q, i);
923 			q->m_pkthdr.csum_flags = 0;
924 			break;
925 		}
926 		nq = q->m_nextpkt;
927 		m->m_nextpkt = nq;
928 		ipstat.ips_fragdropped++;
929 		fp->ipq_nfrags--;
930 		m_freem(q);
931 	}
932 
933 	/*
934 	 * Check for complete reassembly and perform frag per packet
935 	 * limiting.
936 	 *
937 	 * Frag limiting is performed here so that the nth frag has
938 	 * a chance to complete the packet before we drop the packet.
939 	 * As a result, n+1 frags are actually allowed per packet, but
940 	 * only n will ever be stored. (n = maxfragsperpacket.)
941 	 *
942 	 */
943 	next = 0;
944 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
945 		if (GETIP(q)->ip_off != next) {
946 			if (fp->ipq_nfrags > maxfragsperpacket) {
947 				ipstat.ips_fragdropped += fp->ipq_nfrags;
948 				ip_freef(head, fp);
949 			}
950 			goto done;
951 		}
952 		next += GETIP(q)->ip_len;
953 	}
954 	/* Make sure the last packet didn't have the IP_MF flag */
955 	if (p->m_flags & M_FRAG) {
956 		if (fp->ipq_nfrags > maxfragsperpacket) {
957 			ipstat.ips_fragdropped += fp->ipq_nfrags;
958 			ip_freef(head, fp);
959 		}
960 		goto done;
961 	}
962 
963 	/*
964 	 * Reassembly is complete.  Make sure the packet is a sane size.
965 	 */
966 	q = fp->ipq_frags;
967 	ip = GETIP(q);
968 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
969 		ipstat.ips_toolong++;
970 		ipstat.ips_fragdropped += fp->ipq_nfrags;
971 		ip_freef(head, fp);
972 		goto done;
973 	}
974 
975 	/*
976 	 * Concatenate fragments.
977 	 */
978 	m = q;
979 	t = m->m_next;
980 	m->m_next = NULL;
981 	m_cat(m, t);
982 	nq = q->m_nextpkt;
983 	q->m_nextpkt = NULL;
984 	for (q = nq; q != NULL; q = nq) {
985 		nq = q->m_nextpkt;
986 		q->m_nextpkt = NULL;
987 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
988 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
989 		m_cat(m, q);
990 	}
991 	/*
992 	 * In order to do checksumming faster we do 'end-around carry' here
993 	 * (and not in for{} loop), though it implies we are not going to
994 	 * reassemble more than 64k fragments.
995 	 */
996 	m->m_pkthdr.csum_data =
997 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
998 #ifdef MAC
999 	mac_create_datagram_from_ipq(fp, m);
1000 	mac_destroy_ipq(fp);
1001 #endif
1002 
1003 	/*
1004 	 * Create header for new ip packet by modifying header of first
1005 	 * packet;  dequeue and discard fragment reassembly header.
1006 	 * Make header visible.
1007 	 */
1008 	ip->ip_len = (ip->ip_hl << 2) + next;
1009 	ip->ip_src = fp->ipq_src;
1010 	ip->ip_dst = fp->ipq_dst;
1011 	TAILQ_REMOVE(head, fp, ipq_list);
1012 	nipq--;
1013 	uma_zfree(ipq_zone, fp);
1014 	m->m_len += (ip->ip_hl << 2);
1015 	m->m_data -= (ip->ip_hl << 2);
1016 	/* some debugging cruft by sklower, below, will go away soon */
1017 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1018 		m_fixhdr(m);
1019 	ipstat.ips_reassembled++;
1020 	IPQ_UNLOCK();
1021 	return (m);
1022 
1023 dropfrag:
1024 	ipstat.ips_fragdropped++;
1025 	if (fp != NULL)
1026 		fp->ipq_nfrags--;
1027 	m_freem(m);
1028 done:
1029 	IPQ_UNLOCK();
1030 	return (NULL);
1031 
1032 #undef GETIP
1033 }
1034 
1035 /*
1036  * Free a fragment reassembly header and all
1037  * associated datagrams.
1038  */
1039 static void
1040 ip_freef(fhp, fp)
1041 	struct ipqhead *fhp;
1042 	struct ipq *fp;
1043 {
1044 	register struct mbuf *q;
1045 
1046 	IPQ_LOCK_ASSERT();
1047 
1048 	while (fp->ipq_frags) {
1049 		q = fp->ipq_frags;
1050 		fp->ipq_frags = q->m_nextpkt;
1051 		m_freem(q);
1052 	}
1053 	TAILQ_REMOVE(fhp, fp, ipq_list);
1054 	uma_zfree(ipq_zone, fp);
1055 	nipq--;
1056 }
1057 
1058 /*
1059  * IP timer processing;
1060  * if a timer expires on a reassembly
1061  * queue, discard it.
1062  */
1063 void
1064 ip_slowtimo()
1065 {
1066 	register struct ipq *fp;
1067 	int i;
1068 
1069 	IPQ_LOCK();
1070 	for (i = 0; i < IPREASS_NHASH; i++) {
1071 		for(fp = TAILQ_FIRST(&ipq[i]); fp;) {
1072 			struct ipq *fpp;
1073 
1074 			fpp = fp;
1075 			fp = TAILQ_NEXT(fp, ipq_list);
1076 			if(--fpp->ipq_ttl == 0) {
1077 				ipstat.ips_fragtimeout += fpp->ipq_nfrags;
1078 				ip_freef(&ipq[i], fpp);
1079 			}
1080 		}
1081 	}
1082 	/*
1083 	 * If we are over the maximum number of fragments
1084 	 * (due to the limit being lowered), drain off
1085 	 * enough to get down to the new limit.
1086 	 */
1087 	if (maxnipq >= 0 && nipq > maxnipq) {
1088 		for (i = 0; i < IPREASS_NHASH; i++) {
1089 			while (nipq > maxnipq && !TAILQ_EMPTY(&ipq[i])) {
1090 				ipstat.ips_fragdropped +=
1091 				    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1092 				ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1093 			}
1094 		}
1095 	}
1096 	IPQ_UNLOCK();
1097 }
1098 
1099 /*
1100  * Drain off all datagram fragments.
1101  */
1102 void
1103 ip_drain()
1104 {
1105 	int     i;
1106 
1107 	IPQ_LOCK();
1108 	for (i = 0; i < IPREASS_NHASH; i++) {
1109 		while(!TAILQ_EMPTY(&ipq[i])) {
1110 			ipstat.ips_fragdropped +=
1111 			    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1112 			ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1113 		}
1114 	}
1115 	IPQ_UNLOCK();
1116 	in_rtqdrain();
1117 }
1118 
1119 /*
1120  * The protocol to be inserted into ip_protox[] must be already registered
1121  * in inetsw[], either statically or through pf_proto_register().
1122  */
1123 int
1124 ipproto_register(u_char ipproto)
1125 {
1126 	struct protosw *pr;
1127 
1128 	/* Sanity checks. */
1129 	if (ipproto == 0)
1130 		return (EPROTONOSUPPORT);
1131 
1132 	/*
1133 	 * The protocol slot must not be occupied by another protocol
1134 	 * already.  An index pointing to IPPROTO_RAW is unused.
1135 	 */
1136 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1137 	if (pr == NULL)
1138 		return (EPFNOSUPPORT);
1139 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1140 		return (EEXIST);
1141 
1142 	/* Find the protocol position in inetsw[] and set the index. */
1143 	for (pr = inetdomain.dom_protosw;
1144 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1145 		if (pr->pr_domain->dom_family == PF_INET &&
1146 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1147 			/* Be careful to only index valid IP protocols. */
1148 			if (pr->pr_protocol < IPPROTO_MAX) {
1149 				ip_protox[pr->pr_protocol] = pr - inetsw;
1150 				return (0);
1151 			} else
1152 				return (EINVAL);
1153 		}
1154 	}
1155 	return (EPROTONOSUPPORT);
1156 }
1157 
1158 int
1159 ipproto_unregister(u_char ipproto)
1160 {
1161 	struct protosw *pr;
1162 
1163 	/* Sanity checks. */
1164 	if (ipproto == 0)
1165 		return (EPROTONOSUPPORT);
1166 
1167 	/* Check if the protocol was indeed registered. */
1168 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1169 	if (pr == NULL)
1170 		return (EPFNOSUPPORT);
1171 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1172 		return (ENOENT);
1173 
1174 	/* Reset the protocol slot to IPPROTO_RAW. */
1175 	ip_protox[ipproto] = pr - inetsw;
1176 	return (0);
1177 }
1178 
1179 /*
1180  * Given address of next destination (final or next hop),
1181  * return internet address info of interface to be used to get there.
1182  */
1183 struct in_ifaddr *
1184 ip_rtaddr(dst)
1185 	struct in_addr dst;
1186 {
1187 	struct route sro;
1188 	struct sockaddr_in *sin;
1189 	struct in_ifaddr *ifa;
1190 
1191 	bzero(&sro, sizeof(sro));
1192 	sin = (struct sockaddr_in *)&sro.ro_dst;
1193 	sin->sin_family = AF_INET;
1194 	sin->sin_len = sizeof(*sin);
1195 	sin->sin_addr = dst;
1196 	rtalloc_ign(&sro, RTF_CLONING);
1197 
1198 	if (sro.ro_rt == NULL)
1199 		return (NULL);
1200 
1201 	ifa = ifatoia(sro.ro_rt->rt_ifa);
1202 	RTFREE(sro.ro_rt);
1203 	return (ifa);
1204 }
1205 
1206 u_char inetctlerrmap[PRC_NCMDS] = {
1207 	0,		0,		0,		0,
1208 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1209 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1210 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1211 	0,		0,		EHOSTUNREACH,	0,
1212 	ENOPROTOOPT,	ECONNREFUSED
1213 };
1214 
1215 /*
1216  * Forward a packet.  If some error occurs return the sender
1217  * an icmp packet.  Note we can't always generate a meaningful
1218  * icmp message because icmp doesn't have a large enough repertoire
1219  * of codes and types.
1220  *
1221  * If not forwarding, just drop the packet.  This could be confusing
1222  * if ipforwarding was zero but some routing protocol was advancing
1223  * us as a gateway to somewhere.  However, we must let the routing
1224  * protocol deal with that.
1225  *
1226  * The srcrt parameter indicates whether the packet is being forwarded
1227  * via a source route.
1228  */
1229 void
1230 ip_forward(struct mbuf *m, int srcrt)
1231 {
1232 	struct ip *ip = mtod(m, struct ip *);
1233 	struct in_ifaddr *ia = NULL;
1234 	struct mbuf *mcopy;
1235 	struct in_addr dest;
1236 	int error, type = 0, code = 0, mtu = 0;
1237 
1238 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1239 		ipstat.ips_cantforward++;
1240 		m_freem(m);
1241 		return;
1242 	}
1243 #ifdef IPSTEALTH
1244 	if (!ipstealth) {
1245 #endif
1246 		if (ip->ip_ttl <= IPTTLDEC) {
1247 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1248 			    0, 0);
1249 			return;
1250 		}
1251 #ifdef IPSTEALTH
1252 	}
1253 #endif
1254 
1255 	if (!srcrt && (ia = ip_rtaddr(ip->ip_dst)) == NULL) {
1256 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1257 		return;
1258 	}
1259 
1260 	/*
1261 	 * Save the IP header and at most 8 bytes of the payload,
1262 	 * in case we need to generate an ICMP message to the src.
1263 	 *
1264 	 * XXX this can be optimized a lot by saving the data in a local
1265 	 * buffer on the stack (72 bytes at most), and only allocating the
1266 	 * mbuf if really necessary. The vast majority of the packets
1267 	 * are forwarded without having to send an ICMP back (either
1268 	 * because unnecessary, or because rate limited), so we are
1269 	 * really we are wasting a lot of work here.
1270 	 *
1271 	 * We don't use m_copy() because it might return a reference
1272 	 * to a shared cluster. Both this function and ip_output()
1273 	 * assume exclusive access to the IP header in `m', so any
1274 	 * data in a cluster may change before we reach icmp_error().
1275 	 */
1276 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1277 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1278 		/*
1279 		 * It's probably ok if the pkthdr dup fails (because
1280 		 * the deep copy of the tag chain failed), but for now
1281 		 * be conservative and just discard the copy since
1282 		 * code below may some day want the tags.
1283 		 */
1284 		m_free(mcopy);
1285 		mcopy = NULL;
1286 	}
1287 	if (mcopy != NULL) {
1288 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1289 		mcopy->m_pkthdr.len = mcopy->m_len;
1290 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1291 	}
1292 
1293 #ifdef IPSTEALTH
1294 	if (!ipstealth) {
1295 #endif
1296 		ip->ip_ttl -= IPTTLDEC;
1297 #ifdef IPSTEALTH
1298 	}
1299 #endif
1300 
1301 	/*
1302 	 * If forwarding packet using same interface that it came in on,
1303 	 * perhaps should send a redirect to sender to shortcut a hop.
1304 	 * Only send redirect if source is sending directly to us,
1305 	 * and if packet was not source routed (or has any options).
1306 	 * Also, don't send redirect if forwarding using a default route
1307 	 * or a route modified by a redirect.
1308 	 */
1309 	dest.s_addr = 0;
1310 	if (!srcrt && ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1311 		struct sockaddr_in *sin;
1312 		struct route ro;
1313 		struct rtentry *rt;
1314 
1315 		bzero(&ro, sizeof(ro));
1316 		sin = (struct sockaddr_in *)&ro.ro_dst;
1317 		sin->sin_family = AF_INET;
1318 		sin->sin_len = sizeof(*sin);
1319 		sin->sin_addr = ip->ip_dst;
1320 		rtalloc_ign(&ro, RTF_CLONING);
1321 
1322 		rt = ro.ro_rt;
1323 
1324 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1325 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1326 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1327 			u_long src = ntohl(ip->ip_src.s_addr);
1328 
1329 			if (RTA(rt) &&
1330 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1331 				if (rt->rt_flags & RTF_GATEWAY)
1332 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1333 				else
1334 					dest.s_addr = ip->ip_dst.s_addr;
1335 				/* Router requirements says to only send host redirects */
1336 				type = ICMP_REDIRECT;
1337 				code = ICMP_REDIRECT_HOST;
1338 			}
1339 		}
1340 		if (rt)
1341 			RTFREE(rt);
1342 	}
1343 
1344 	error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1345 	if (error)
1346 		ipstat.ips_cantforward++;
1347 	else {
1348 		ipstat.ips_forward++;
1349 		if (type)
1350 			ipstat.ips_redirectsent++;
1351 		else {
1352 			if (mcopy)
1353 				m_freem(mcopy);
1354 			return;
1355 		}
1356 	}
1357 	if (mcopy == NULL)
1358 		return;
1359 
1360 	switch (error) {
1361 
1362 	case 0:				/* forwarded, but need redirect */
1363 		/* type, code set above */
1364 		break;
1365 
1366 	case ENETUNREACH:		/* shouldn't happen, checked above */
1367 	case EHOSTUNREACH:
1368 	case ENETDOWN:
1369 	case EHOSTDOWN:
1370 	default:
1371 		type = ICMP_UNREACH;
1372 		code = ICMP_UNREACH_HOST;
1373 		break;
1374 
1375 	case EMSGSIZE:
1376 		type = ICMP_UNREACH;
1377 		code = ICMP_UNREACH_NEEDFRAG;
1378 
1379 #if defined(IPSEC) || defined(FAST_IPSEC)
1380 		mtu = ip_ipsec_mtu(m);
1381 #endif /* IPSEC */
1382 		/*
1383 		 * If the MTU wasn't set before use the interface mtu or
1384 		 * fall back to the next smaller mtu step compared to the
1385 		 * current packet size.
1386 		 */
1387 		if (mtu == 0) {
1388 			if (ia != NULL)
1389 				mtu = ia->ia_ifp->if_mtu;
1390 			else
1391 				mtu = ip_next_mtu(ip->ip_len, 0);
1392 		}
1393 		ipstat.ips_cantfrag++;
1394 		break;
1395 
1396 	case ENOBUFS:
1397 		/*
1398 		 * A router should not generate ICMP_SOURCEQUENCH as
1399 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1400 		 * Source quench could be a big problem under DoS attacks,
1401 		 * or if the underlying interface is rate-limited.
1402 		 * Those who need source quench packets may re-enable them
1403 		 * via the net.inet.ip.sendsourcequench sysctl.
1404 		 */
1405 		if (ip_sendsourcequench == 0) {
1406 			m_freem(mcopy);
1407 			return;
1408 		} else {
1409 			type = ICMP_SOURCEQUENCH;
1410 			code = 0;
1411 		}
1412 		break;
1413 
1414 	case EACCES:			/* ipfw denied packet */
1415 		m_freem(mcopy);
1416 		return;
1417 	}
1418 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1419 }
1420 
1421 void
1422 ip_savecontrol(inp, mp, ip, m)
1423 	register struct inpcb *inp;
1424 	register struct mbuf **mp;
1425 	register struct ip *ip;
1426 	register struct mbuf *m;
1427 {
1428 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1429 		struct bintime bt;
1430 
1431 		bintime(&bt);
1432 		if (inp->inp_socket->so_options & SO_BINTIME) {
1433 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1434 			SCM_BINTIME, SOL_SOCKET);
1435 			if (*mp)
1436 				mp = &(*mp)->m_next;
1437 		}
1438 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1439 			struct timeval tv;
1440 
1441 			bintime2timeval(&bt, &tv);
1442 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1443 				SCM_TIMESTAMP, SOL_SOCKET);
1444 			if (*mp)
1445 				mp = &(*mp)->m_next;
1446 		}
1447 	}
1448 	if (inp->inp_flags & INP_RECVDSTADDR) {
1449 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1450 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1451 		if (*mp)
1452 			mp = &(*mp)->m_next;
1453 	}
1454 	if (inp->inp_flags & INP_RECVTTL) {
1455 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1456 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1457 		if (*mp)
1458 			mp = &(*mp)->m_next;
1459 	}
1460 #ifdef notyet
1461 	/* XXX
1462 	 * Moving these out of udp_input() made them even more broken
1463 	 * than they already were.
1464 	 */
1465 	/* options were tossed already */
1466 	if (inp->inp_flags & INP_RECVOPTS) {
1467 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1468 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1469 		if (*mp)
1470 			mp = &(*mp)->m_next;
1471 	}
1472 	/* ip_srcroute doesn't do what we want here, need to fix */
1473 	if (inp->inp_flags & INP_RECVRETOPTS) {
1474 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1475 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1476 		if (*mp)
1477 			mp = &(*mp)->m_next;
1478 	}
1479 #endif
1480 	if (inp->inp_flags & INP_RECVIF) {
1481 		struct ifnet *ifp;
1482 		struct sdlbuf {
1483 			struct sockaddr_dl sdl;
1484 			u_char	pad[32];
1485 		} sdlbuf;
1486 		struct sockaddr_dl *sdp;
1487 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1488 
1489 		if (((ifp = m->m_pkthdr.rcvif))
1490 		&& ( ifp->if_index && (ifp->if_index <= if_index))) {
1491 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1492 			/*
1493 			 * Change our mind and don't try copy.
1494 			 */
1495 			if ((sdp->sdl_family != AF_LINK)
1496 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1497 				goto makedummy;
1498 			}
1499 			bcopy(sdp, sdl2, sdp->sdl_len);
1500 		} else {
1501 makedummy:
1502 			sdl2->sdl_len
1503 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1504 			sdl2->sdl_family = AF_LINK;
1505 			sdl2->sdl_index = 0;
1506 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1507 		}
1508 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1509 			IP_RECVIF, IPPROTO_IP);
1510 		if (*mp)
1511 			mp = &(*mp)->m_next;
1512 	}
1513 }
1514 
1515 /*
1516  * XXX these routines are called from the upper part of the kernel.
1517  * They need to be locked when we remove Giant.
1518  *
1519  * They could also be moved to ip_mroute.c, since all the RSVP
1520  *  handling is done there already.
1521  */
1522 static int ip_rsvp_on;
1523 struct socket *ip_rsvpd;
1524 int
1525 ip_rsvp_init(struct socket *so)
1526 {
1527 	if (so->so_type != SOCK_RAW ||
1528 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1529 		return EOPNOTSUPP;
1530 
1531 	if (ip_rsvpd != NULL)
1532 		return EADDRINUSE;
1533 
1534 	ip_rsvpd = so;
1535 	/*
1536 	 * This may seem silly, but we need to be sure we don't over-increment
1537 	 * the RSVP counter, in case something slips up.
1538 	 */
1539 	if (!ip_rsvp_on) {
1540 		ip_rsvp_on = 1;
1541 		rsvp_on++;
1542 	}
1543 
1544 	return 0;
1545 }
1546 
1547 int
1548 ip_rsvp_done(void)
1549 {
1550 	ip_rsvpd = NULL;
1551 	/*
1552 	 * This may seem silly, but we need to be sure we don't over-decrement
1553 	 * the RSVP counter, in case something slips up.
1554 	 */
1555 	if (ip_rsvp_on) {
1556 		ip_rsvp_on = 0;
1557 		rsvp_on--;
1558 	}
1559 	return 0;
1560 }
1561 
1562 void
1563 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1564 {
1565 	if (rsvp_input_p) { /* call the real one if loaded */
1566 		rsvp_input_p(m, off);
1567 		return;
1568 	}
1569 
1570 	/* Can still get packets with rsvp_on = 0 if there is a local member
1571 	 * of the group to which the RSVP packet is addressed.  But in this
1572 	 * case we want to throw the packet away.
1573 	 */
1574 
1575 	if (!rsvp_on) {
1576 		m_freem(m);
1577 		return;
1578 	}
1579 
1580 	if (ip_rsvpd != NULL) {
1581 		rip_input(m, off);
1582 		return;
1583 	}
1584 	/* Drop the packet */
1585 	m_freem(m);
1586 }
1587