xref: /freebsd/sys/netinet/ip_input.c (revision 91c878a6935c5c2e99866eb267e5bc3028bf6d2f)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
30  * $FreeBSD$
31  */
32 
33 #include "opt_bootp.h"
34 #include "opt_ipfw.h"
35 #include "opt_ipstealth.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_carp.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/mbuf.h>
44 #include <sys/malloc.h>
45 #include <sys/domain.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/syslog.h>
51 #include <sys/sysctl.h>
52 
53 #include <net/pfil.h>
54 #include <net/if.h>
55 #include <net/if_types.h>
56 #include <net/if_var.h>
57 #include <net/if_dl.h>
58 #include <net/route.h>
59 #include <net/netisr.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_icmp.h>
68 #include <netinet/ip_options.h>
69 #include <machine/in_cksum.h>
70 #ifdef DEV_CARP
71 #include <netinet/ip_carp.h>
72 #endif
73 #if defined(IPSEC) || defined(FAST_IPSEC)
74 #include <netinet/ip_ipsec.h>
75 #endif /* IPSEC */
76 
77 #include <sys/socketvar.h>
78 
79 /* XXX: Temporary until ipfw_ether and ipfw_bridge are converted. */
80 #include <netinet/ip_fw.h>
81 #include <netinet/ip_dummynet.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 int rsvp_on = 0;
86 
87 int	ipforwarding = 0;
88 SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
89     &ipforwarding, 0, "Enable IP forwarding between interfaces");
90 
91 static int	ipsendredirects = 1; /* XXX */
92 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
93     &ipsendredirects, 0, "Enable sending IP redirects");
94 
95 int	ip_defttl = IPDEFTTL;
96 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
97     &ip_defttl, 0, "Maximum TTL on IP packets");
98 
99 static int	ip_keepfaith = 0;
100 SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
101 	&ip_keepfaith,	0,
102 	"Enable packet capture for FAITH IPv4->IPv6 translater daemon");
103 
104 static int	ip_sendsourcequench = 0;
105 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
106 	&ip_sendsourcequench, 0,
107 	"Enable the transmission of source quench packets");
108 
109 int	ip_do_randomid = 0;
110 SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
111 	&ip_do_randomid, 0,
112 	"Assign random ip_id values");
113 
114 /*
115  * XXX - Setting ip_checkinterface mostly implements the receive side of
116  * the Strong ES model described in RFC 1122, but since the routing table
117  * and transmit implementation do not implement the Strong ES model,
118  * setting this to 1 results in an odd hybrid.
119  *
120  * XXX - ip_checkinterface currently must be disabled if you use ipnat
121  * to translate the destination address to another local interface.
122  *
123  * XXX - ip_checkinterface must be disabled if you add IP aliases
124  * to the loopback interface instead of the interface where the
125  * packets for those addresses are received.
126  */
127 static int	ip_checkinterface = 0;
128 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
129     &ip_checkinterface, 0, "Verify packet arrives on correct interface");
130 
131 struct pfil_head inet_pfil_hook;	/* Packet filter hooks */
132 
133 static struct	ifqueue ipintrq;
134 static int	ipqmaxlen = IFQ_MAXLEN;
135 
136 extern	struct domain inetdomain;
137 extern	struct protosw inetsw[];
138 u_char	ip_protox[IPPROTO_MAX];
139 struct	in_ifaddrhead in_ifaddrhead; 		/* first inet address */
140 struct	in_ifaddrhashhead *in_ifaddrhashtbl;	/* inet addr hash table  */
141 u_long 	in_ifaddrhmask;				/* mask for hash table */
142 
143 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
144     &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
145 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
146     &ipintrq.ifq_drops, 0, "Number of packets dropped from the IP input queue");
147 
148 struct ipstat ipstat;
149 SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
150     &ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
151 
152 /*
153  * IP datagram reassembly.
154  */
155 #define IPREASS_NHASH_LOG2      6
156 #define IPREASS_NHASH           (1 << IPREASS_NHASH_LOG2)
157 #define IPREASS_HMASK           (IPREASS_NHASH - 1)
158 #define IPREASS_HASH(x,y) \
159 	(((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
160 
161 static uma_zone_t ipq_zone;
162 static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
163 static struct mtx ipqlock;
164 
165 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
166 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
167 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
168 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
169 
170 static void	maxnipq_update(void);
171 static void	ipq_zone_change(void *);
172 
173 static int	maxnipq;	/* Administrative limit on # reass queues. */
174 static int	nipq = 0;	/* Total # of reass queues */
175 SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD, &nipq, 0,
176 	"Current number of IPv4 fragment reassembly queue entries");
177 
178 static int	maxfragsperpacket;
179 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
180 	&maxfragsperpacket, 0,
181 	"Maximum number of IPv4 fragments allowed per packet");
182 
183 struct callout	ipport_tick_callout;
184 
185 #ifdef IPCTL_DEFMTU
186 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
187     &ip_mtu, 0, "Default MTU");
188 #endif
189 
190 #ifdef IPSTEALTH
191 int	ipstealth = 0;
192 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
193     &ipstealth, 0, "");
194 #endif
195 
196 /*
197  * ipfw_ether and ipfw_bridge hooks.
198  * XXX: Temporary until those are converted to pfil_hooks as well.
199  */
200 ip_fw_chk_t *ip_fw_chk_ptr = NULL;
201 ip_dn_io_t *ip_dn_io_ptr = NULL;
202 int fw_one_pass = 1;
203 
204 static void	ip_freef(struct ipqhead *, struct ipq *);
205 
206 /*
207  * IP initialization: fill in IP protocol switch table.
208  * All protocols not implemented in kernel go to raw IP protocol handler.
209  */
210 void
211 ip_init()
212 {
213 	register struct protosw *pr;
214 	register int i;
215 
216 	TAILQ_INIT(&in_ifaddrhead);
217 	in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
218 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
219 	if (pr == NULL)
220 		panic("ip_init: PF_INET not found");
221 
222 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
223 	for (i = 0; i < IPPROTO_MAX; i++)
224 		ip_protox[i] = pr - inetsw;
225 	/*
226 	 * Cycle through IP protocols and put them into the appropriate place
227 	 * in ip_protox[].
228 	 */
229 	for (pr = inetdomain.dom_protosw;
230 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
231 		if (pr->pr_domain->dom_family == PF_INET &&
232 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
233 			/* Be careful to only index valid IP protocols. */
234 			if (pr->pr_protocol < IPPROTO_MAX)
235 				ip_protox[pr->pr_protocol] = pr - inetsw;
236 		}
237 
238 	/* Initialize packet filter hooks. */
239 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
240 	inet_pfil_hook.ph_af = AF_INET;
241 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
242 		printf("%s: WARNING: unable to register pfil hook, "
243 			"error %d\n", __func__, i);
244 
245 	/* Initialize IP reassembly queue. */
246 	IPQ_LOCK_INIT();
247 	for (i = 0; i < IPREASS_NHASH; i++)
248 	    TAILQ_INIT(&ipq[i]);
249 	maxnipq = nmbclusters / 32;
250 	maxfragsperpacket = 16;
251 	ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
252 	    NULL, UMA_ALIGN_PTR, 0);
253 	maxnipq_update();
254 
255 	/* Start ipport_tick. */
256 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
257 	ipport_tick(NULL);
258 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
259 		SHUTDOWN_PRI_DEFAULT);
260 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
261 		NULL, EVENTHANDLER_PRI_ANY);
262 
263 	/* Initialize various other remaining things. */
264 	ip_id = time_second & 0xffff;
265 	ipintrq.ifq_maxlen = ipqmaxlen;
266 	mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
267 	netisr_register(NETISR_IP, ip_input, &ipintrq, NETISR_MPSAFE);
268 }
269 
270 void ip_fini(xtp)
271 	void *xtp;
272 {
273 	callout_stop(&ipport_tick_callout);
274 }
275 
276 /*
277  * Ip input routine.  Checksum and byte swap header.  If fragmented
278  * try to reassemble.  Process options.  Pass to next level.
279  */
280 void
281 ip_input(struct mbuf *m)
282 {
283 	struct ip *ip = NULL;
284 	struct in_ifaddr *ia = NULL;
285 	struct ifaddr *ifa;
286 	int    checkif, hlen = 0;
287 	u_short sum;
288 	int dchg = 0;				/* dest changed after fw */
289 	struct in_addr odst;			/* original dst address */
290 
291   	M_ASSERTPKTHDR(m);
292 
293 	if (m->m_flags & M_FASTFWD_OURS) {
294 		/*
295 		 * Firewall or NAT changed destination to local.
296 		 * We expect ip_len and ip_off to be in host byte order.
297 		 */
298 		m->m_flags &= ~M_FASTFWD_OURS;
299 		/* Set up some basics that will be used later. */
300 		ip = mtod(m, struct ip *);
301 		hlen = ip->ip_hl << 2;
302   		goto ours;
303   	}
304 
305 	ipstat.ips_total++;
306 
307 	if (m->m_pkthdr.len < sizeof(struct ip))
308 		goto tooshort;
309 
310 	if (m->m_len < sizeof (struct ip) &&
311 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
312 		ipstat.ips_toosmall++;
313 		return;
314 	}
315 	ip = mtod(m, struct ip *);
316 
317 	if (ip->ip_v != IPVERSION) {
318 		ipstat.ips_badvers++;
319 		goto bad;
320 	}
321 
322 	hlen = ip->ip_hl << 2;
323 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
324 		ipstat.ips_badhlen++;
325 		goto bad;
326 	}
327 	if (hlen > m->m_len) {
328 		if ((m = m_pullup(m, hlen)) == NULL) {
329 			ipstat.ips_badhlen++;
330 			return;
331 		}
332 		ip = mtod(m, struct ip *);
333 	}
334 
335 	/* 127/8 must not appear on wire - RFC1122 */
336 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
337 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
338 		if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
339 			ipstat.ips_badaddr++;
340 			goto bad;
341 		}
342 	}
343 
344 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
345 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
346 	} else {
347 		if (hlen == sizeof(struct ip)) {
348 			sum = in_cksum_hdr(ip);
349 		} else {
350 			sum = in_cksum(m, hlen);
351 		}
352 	}
353 	if (sum) {
354 		ipstat.ips_badsum++;
355 		goto bad;
356 	}
357 
358 #ifdef ALTQ
359 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
360 		/* packet is dropped by traffic conditioner */
361 		return;
362 #endif
363 
364 	/*
365 	 * Convert fields to host representation.
366 	 */
367 	ip->ip_len = ntohs(ip->ip_len);
368 	if (ip->ip_len < hlen) {
369 		ipstat.ips_badlen++;
370 		goto bad;
371 	}
372 	ip->ip_off = ntohs(ip->ip_off);
373 
374 	/*
375 	 * Check that the amount of data in the buffers
376 	 * is as at least much as the IP header would have us expect.
377 	 * Trim mbufs if longer than we expect.
378 	 * Drop packet if shorter than we expect.
379 	 */
380 	if (m->m_pkthdr.len < ip->ip_len) {
381 tooshort:
382 		ipstat.ips_tooshort++;
383 		goto bad;
384 	}
385 	if (m->m_pkthdr.len > ip->ip_len) {
386 		if (m->m_len == m->m_pkthdr.len) {
387 			m->m_len = ip->ip_len;
388 			m->m_pkthdr.len = ip->ip_len;
389 		} else
390 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
391 	}
392 #if defined(IPSEC) || defined(FAST_IPSEC)
393 	/*
394 	 * Bypass packet filtering for packets from a tunnel (gif).
395 	 */
396 	if (ip_ipsec_filtergif(m))
397 		goto passin;
398 #endif /* IPSEC */
399 
400 	/*
401 	 * Run through list of hooks for input packets.
402 	 *
403 	 * NB: Beware of the destination address changing (e.g.
404 	 *     by NAT rewriting).  When this happens, tell
405 	 *     ip_forward to do the right thing.
406 	 */
407 
408 	/* Jump over all PFIL processing if hooks are not active. */
409 	if (!PFIL_HOOKED(&inet_pfil_hook))
410 		goto passin;
411 
412 	odst = ip->ip_dst;
413 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
414 	    PFIL_IN, NULL) != 0)
415 		return;
416 	if (m == NULL)			/* consumed by filter */
417 		return;
418 
419 	ip = mtod(m, struct ip *);
420 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
421 
422 #ifdef IPFIREWALL_FORWARD
423 	if (m->m_flags & M_FASTFWD_OURS) {
424 		m->m_flags &= ~M_FASTFWD_OURS;
425 		goto ours;
426 	}
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 */
437 
438 passin:
439 	/*
440 	 * Process options and, if not destined for us,
441 	 * ship it on.  ip_dooptions returns 1 when an
442 	 * error was detected (causing an icmp message
443 	 * to be sent and the original packet to be freed).
444 	 */
445 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
446 		return;
447 
448         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
449          * matter if it is destined to another node, or whether it is
450          * a multicast one, RSVP wants it! and prevents it from being forwarded
451          * anywhere else. Also checks if the rsvp daemon is running before
452 	 * grabbing the packet.
453          */
454 	if (rsvp_on && ip->ip_p==IPPROTO_RSVP)
455 		goto ours;
456 
457 	/*
458 	 * Check our list of addresses, to see if the packet is for us.
459 	 * If we don't have any addresses, assume any unicast packet
460 	 * we receive might be for us (and let the upper layers deal
461 	 * with it).
462 	 */
463 	if (TAILQ_EMPTY(&in_ifaddrhead) &&
464 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
465 		goto ours;
466 
467 	/*
468 	 * Enable a consistency check between the destination address
469 	 * and the arrival interface for a unicast packet (the RFC 1122
470 	 * strong ES model) if IP forwarding is disabled and the packet
471 	 * is not locally generated and the packet is not subject to
472 	 * 'ipfw fwd'.
473 	 *
474 	 * XXX - Checking also should be disabled if the destination
475 	 * address is ipnat'ed to a different interface.
476 	 *
477 	 * XXX - Checking is incompatible with IP aliases added
478 	 * to the loopback interface instead of the interface where
479 	 * the packets are received.
480 	 *
481 	 * XXX - This is the case for carp vhost IPs as well so we
482 	 * insert a workaround. If the packet got here, we already
483 	 * checked with carp_iamatch() and carp_forus().
484 	 */
485 	checkif = ip_checkinterface && (ipforwarding == 0) &&
486 	    m->m_pkthdr.rcvif != NULL &&
487 	    ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) &&
488 #ifdef DEV_CARP
489 	    !m->m_pkthdr.rcvif->if_carp &&
490 #endif
491 	    (dchg == 0);
492 
493 	/*
494 	 * Check for exact addresses in the hash bucket.
495 	 */
496 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
497 		/*
498 		 * If the address matches, verify that the packet
499 		 * arrived via the correct interface if checking is
500 		 * enabled.
501 		 */
502 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
503 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
504 			goto ours;
505 	}
506 	/*
507 	 * Check for broadcast addresses.
508 	 *
509 	 * Only accept broadcast packets that arrive via the matching
510 	 * interface.  Reception of forwarded directed broadcasts would
511 	 * be handled via ip_forward() and ether_output() with the loopback
512 	 * into the stack for SIMPLEX interfaces handled by ether_output().
513 	 */
514 	if (m->m_pkthdr.rcvif != NULL &&
515 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
516 	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
517 			if (ifa->ifa_addr->sa_family != AF_INET)
518 				continue;
519 			ia = ifatoia(ifa);
520 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
521 			    ip->ip_dst.s_addr)
522 				goto ours;
523 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
524 				goto ours;
525 #ifdef BOOTP_COMPAT
526 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
527 				goto ours;
528 #endif
529 		}
530 	}
531 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
532 		struct in_multi *inm;
533 		if (ip_mrouter) {
534 			/*
535 			 * If we are acting as a multicast router, all
536 			 * incoming multicast packets are passed to the
537 			 * kernel-level multicast forwarding function.
538 			 * The packet is returned (relatively) intact; if
539 			 * ip_mforward() returns a non-zero value, the packet
540 			 * must be discarded, else it may be accepted below.
541 			 */
542 			if (ip_mforward &&
543 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
544 				ipstat.ips_cantforward++;
545 				m_freem(m);
546 				return;
547 			}
548 
549 			/*
550 			 * The process-level routing daemon needs to receive
551 			 * all multicast IGMP packets, whether or not this
552 			 * host belongs to their destination groups.
553 			 */
554 			if (ip->ip_p == IPPROTO_IGMP)
555 				goto ours;
556 			ipstat.ips_forward++;
557 		}
558 		/*
559 		 * See if we belong to the destination multicast group on the
560 		 * arrival interface.
561 		 */
562 		IN_MULTI_LOCK();
563 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
564 		IN_MULTI_UNLOCK();
565 		if (inm == NULL) {
566 			ipstat.ips_notmember++;
567 			m_freem(m);
568 			return;
569 		}
570 		goto ours;
571 	}
572 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
573 		goto ours;
574 	if (ip->ip_dst.s_addr == INADDR_ANY)
575 		goto ours;
576 
577 	/*
578 	 * FAITH(Firewall Aided Internet Translator)
579 	 */
580 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
581 		if (ip_keepfaith) {
582 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
583 				goto ours;
584 		}
585 		m_freem(m);
586 		return;
587 	}
588 
589 	/*
590 	 * Not for us; forward if possible and desirable.
591 	 */
592 	if (ipforwarding == 0) {
593 		ipstat.ips_cantforward++;
594 		m_freem(m);
595 	} else {
596 #if defined(IPSEC) || defined(FAST_IPSEC)
597 		if (ip_ipsec_fwd(m))
598 			goto bad;
599 #endif /* IPSEC */
600 		ip_forward(m, dchg);
601 	}
602 	return;
603 
604 ours:
605 #ifdef IPSTEALTH
606 	/*
607 	 * IPSTEALTH: Process non-routing options only
608 	 * if the packet is destined for us.
609 	 */
610 	if (ipstealth && hlen > sizeof (struct ip) &&
611 	    ip_dooptions(m, 1))
612 		return;
613 #endif /* IPSTEALTH */
614 
615 	/* Count the packet in the ip address stats */
616 	if (ia != NULL) {
617 		ia->ia_ifa.if_ipackets++;
618 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
619 	}
620 
621 	/*
622 	 * Attempt reassembly; if it succeeds, proceed.
623 	 * ip_reass() will return a different mbuf.
624 	 */
625 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
626 		m = ip_reass(m);
627 		if (m == NULL)
628 			return;
629 		ip = mtod(m, struct ip *);
630 		/* Get the header length of the reassembled packet */
631 		hlen = ip->ip_hl << 2;
632 	}
633 
634 	/*
635 	 * Further protocols expect the packet length to be w/o the
636 	 * IP header.
637 	 */
638 	ip->ip_len -= hlen;
639 
640 #if defined(IPSEC) || defined(FAST_IPSEC)
641 	/*
642 	 * enforce IPsec policy checking if we are seeing last header.
643 	 * note that we do not visit this with protocols with pcb layer
644 	 * code - like udp/tcp/raw ip.
645 	 */
646 	if (ip_ipsec_input(m))
647 		goto bad;
648 #endif /* IPSEC */
649 
650 	/*
651 	 * Switch out to protocol's input routine.
652 	 */
653 	ipstat.ips_delivered++;
654 
655 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
656 	return;
657 bad:
658 	m_freem(m);
659 }
660 
661 /*
662  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
663  * max has slightly different semantics than the sysctl, for historical
664  * reasons.
665  */
666 static void
667 maxnipq_update(void)
668 {
669 
670 	/*
671 	 * -1 for unlimited allocation.
672 	 */
673 	if (maxnipq < 0)
674 		uma_zone_set_max(ipq_zone, 0);
675 	/*
676 	 * Positive number for specific bound.
677 	 */
678 	if (maxnipq > 0)
679 		uma_zone_set_max(ipq_zone, maxnipq);
680 	/*
681 	 * Zero specifies no further fragment queue allocation -- set the
682 	 * bound very low, but rely on implementation elsewhere to actually
683 	 * prevent allocation and reclaim current queues.
684 	 */
685 	if (maxnipq == 0)
686 		uma_zone_set_max(ipq_zone, 1);
687 }
688 
689 static void
690 ipq_zone_change(void *tag)
691 {
692 
693 	if (maxnipq > 0 && maxnipq < (nmbclusters / 32)) {
694 		maxnipq = nmbclusters / 32;
695 		maxnipq_update();
696 	}
697 }
698 
699 static int
700 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
701 {
702 	int error, i;
703 
704 	i = maxnipq;
705 	error = sysctl_handle_int(oidp, &i, 0, req);
706 	if (error || !req->newptr)
707 		return (error);
708 
709 	/*
710 	 * XXXRW: Might be a good idea to sanity check the argument and place
711 	 * an extreme upper bound.
712 	 */
713 	if (i < -1)
714 		return (EINVAL);
715 	maxnipq = i;
716 	maxnipq_update();
717 	return (0);
718 }
719 
720 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
721     NULL, 0, sysctl_maxnipq, "I",
722     "Maximum number of IPv4 fragment reassembly queue entries");
723 
724 /*
725  * Take incoming datagram fragment and try to reassemble it into
726  * whole datagram.  If the argument is the first fragment or one
727  * in between the function will return NULL and store the mbuf
728  * in the fragment chain.  If the argument is the last fragment
729  * the packet will be reassembled and the pointer to the new
730  * mbuf returned for further processing.  Only m_tags attached
731  * to the first packet/fragment are preserved.
732  * The IP header is *NOT* adjusted out of iplen.
733  */
734 
735 struct mbuf *
736 ip_reass(struct mbuf *m)
737 {
738 	struct ip *ip;
739 	struct mbuf *p, *q, *nq, *t;
740 	struct ipq *fp = NULL;
741 	struct ipqhead *head;
742 	int i, hlen, next;
743 	u_int8_t ecn, ecn0;
744 	u_short hash;
745 
746 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
747 	if (maxnipq == 0 || maxfragsperpacket == 0) {
748 		ipstat.ips_fragments++;
749 		ipstat.ips_fragdropped++;
750 		m_freem(m);
751 		return (NULL);
752 	}
753 
754 	ip = mtod(m, struct ip *);
755 	hlen = ip->ip_hl << 2;
756 
757 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
758 	head = &ipq[hash];
759 	IPQ_LOCK();
760 
761 	/*
762 	 * Look for queue of fragments
763 	 * of this datagram.
764 	 */
765 	TAILQ_FOREACH(fp, head, ipq_list)
766 		if (ip->ip_id == fp->ipq_id &&
767 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
768 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
769 #ifdef MAC
770 		    mac_fragment_match(m, fp) &&
771 #endif
772 		    ip->ip_p == fp->ipq_p)
773 			goto found;
774 
775 	fp = NULL;
776 
777 	/*
778 	 * Attempt to trim the number of allocated fragment queues if it
779 	 * exceeds the administrative limit.
780 	 */
781 	if ((nipq > maxnipq) && (maxnipq > 0)) {
782 		/*
783 		 * drop something from the tail of the current queue
784 		 * before proceeding further
785 		 */
786 		struct ipq *q = TAILQ_LAST(head, ipqhead);
787 		if (q == NULL) {   /* gak */
788 			for (i = 0; i < IPREASS_NHASH; i++) {
789 				struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
790 				if (r) {
791 					ipstat.ips_fragtimeout += r->ipq_nfrags;
792 					ip_freef(&ipq[i], r);
793 					break;
794 				}
795 			}
796 		} else {
797 			ipstat.ips_fragtimeout += q->ipq_nfrags;
798 			ip_freef(head, q);
799 		}
800 	}
801 
802 found:
803 	/*
804 	 * Adjust ip_len to not reflect header,
805 	 * convert offset of this to bytes.
806 	 */
807 	ip->ip_len -= hlen;
808 	if (ip->ip_off & IP_MF) {
809 		/*
810 		 * Make sure that fragments have a data length
811 		 * that's a non-zero multiple of 8 bytes.
812 		 */
813 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
814 			ipstat.ips_toosmall++; /* XXX */
815 			goto dropfrag;
816 		}
817 		m->m_flags |= M_FRAG;
818 	} else
819 		m->m_flags &= ~M_FRAG;
820 	ip->ip_off <<= 3;
821 
822 
823 	/*
824 	 * Attempt reassembly; if it succeeds, proceed.
825 	 * ip_reass() will return a different mbuf.
826 	 */
827 	ipstat.ips_fragments++;
828 	m->m_pkthdr.header = ip;
829 
830 	/* Previous ip_reass() started here. */
831 	/*
832 	 * Presence of header sizes in mbufs
833 	 * would confuse code below.
834 	 */
835 	m->m_data += hlen;
836 	m->m_len -= hlen;
837 
838 	/*
839 	 * If first fragment to arrive, create a reassembly queue.
840 	 */
841 	if (fp == NULL) {
842 		fp = uma_zalloc(ipq_zone, M_NOWAIT);
843 		if (fp == NULL)
844 			goto dropfrag;
845 #ifdef MAC
846 		if (mac_init_ipq(fp, M_NOWAIT) != 0) {
847 			uma_zfree(ipq_zone, fp);
848 			fp = NULL;
849 			goto dropfrag;
850 		}
851 		mac_create_ipq(m, fp);
852 #endif
853 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
854 		nipq++;
855 		fp->ipq_nfrags = 1;
856 		fp->ipq_ttl = IPFRAGTTL;
857 		fp->ipq_p = ip->ip_p;
858 		fp->ipq_id = ip->ip_id;
859 		fp->ipq_src = ip->ip_src;
860 		fp->ipq_dst = ip->ip_dst;
861 		fp->ipq_frags = m;
862 		m->m_nextpkt = NULL;
863 		goto done;
864 	} else {
865 		fp->ipq_nfrags++;
866 #ifdef MAC
867 		mac_update_ipq(m, fp);
868 #endif
869 	}
870 
871 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
872 
873 	/*
874 	 * Handle ECN by comparing this segment with the first one;
875 	 * if CE is set, do not lose CE.
876 	 * drop if CE and not-ECT are mixed for the same packet.
877 	 */
878 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
879 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
880 	if (ecn == IPTOS_ECN_CE) {
881 		if (ecn0 == IPTOS_ECN_NOTECT)
882 			goto dropfrag;
883 		if (ecn0 != IPTOS_ECN_CE)
884 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
885 	}
886 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
887 		goto dropfrag;
888 
889 	/*
890 	 * Find a segment which begins after this one does.
891 	 */
892 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
893 		if (GETIP(q)->ip_off > ip->ip_off)
894 			break;
895 
896 	/*
897 	 * If there is a preceding segment, it may provide some of
898 	 * our data already.  If so, drop the data from the incoming
899 	 * segment.  If it provides all of our data, drop us, otherwise
900 	 * stick new segment in the proper place.
901 	 *
902 	 * If some of the data is dropped from the the preceding
903 	 * segment, then it's checksum is invalidated.
904 	 */
905 	if (p) {
906 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
907 		if (i > 0) {
908 			if (i >= ip->ip_len)
909 				goto dropfrag;
910 			m_adj(m, i);
911 			m->m_pkthdr.csum_flags = 0;
912 			ip->ip_off += i;
913 			ip->ip_len -= i;
914 		}
915 		m->m_nextpkt = p->m_nextpkt;
916 		p->m_nextpkt = m;
917 	} else {
918 		m->m_nextpkt = fp->ipq_frags;
919 		fp->ipq_frags = m;
920 	}
921 
922 	/*
923 	 * While we overlap succeeding segments trim them or,
924 	 * if they are completely covered, dequeue them.
925 	 */
926 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
927 	     q = nq) {
928 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
929 		if (i < GETIP(q)->ip_len) {
930 			GETIP(q)->ip_len -= i;
931 			GETIP(q)->ip_off += i;
932 			m_adj(q, i);
933 			q->m_pkthdr.csum_flags = 0;
934 			break;
935 		}
936 		nq = q->m_nextpkt;
937 		m->m_nextpkt = nq;
938 		ipstat.ips_fragdropped++;
939 		fp->ipq_nfrags--;
940 		m_freem(q);
941 	}
942 
943 	/*
944 	 * Check for complete reassembly and perform frag per packet
945 	 * limiting.
946 	 *
947 	 * Frag limiting is performed here so that the nth frag has
948 	 * a chance to complete the packet before we drop the packet.
949 	 * As a result, n+1 frags are actually allowed per packet, but
950 	 * only n will ever be stored. (n = maxfragsperpacket.)
951 	 *
952 	 */
953 	next = 0;
954 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
955 		if (GETIP(q)->ip_off != next) {
956 			if (fp->ipq_nfrags > maxfragsperpacket) {
957 				ipstat.ips_fragdropped += fp->ipq_nfrags;
958 				ip_freef(head, fp);
959 			}
960 			goto done;
961 		}
962 		next += GETIP(q)->ip_len;
963 	}
964 	/* Make sure the last packet didn't have the IP_MF flag */
965 	if (p->m_flags & M_FRAG) {
966 		if (fp->ipq_nfrags > maxfragsperpacket) {
967 			ipstat.ips_fragdropped += fp->ipq_nfrags;
968 			ip_freef(head, fp);
969 		}
970 		goto done;
971 	}
972 
973 	/*
974 	 * Reassembly is complete.  Make sure the packet is a sane size.
975 	 */
976 	q = fp->ipq_frags;
977 	ip = GETIP(q);
978 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
979 		ipstat.ips_toolong++;
980 		ipstat.ips_fragdropped += fp->ipq_nfrags;
981 		ip_freef(head, fp);
982 		goto done;
983 	}
984 
985 	/*
986 	 * Concatenate fragments.
987 	 */
988 	m = q;
989 	t = m->m_next;
990 	m->m_next = NULL;
991 	m_cat(m, t);
992 	nq = q->m_nextpkt;
993 	q->m_nextpkt = NULL;
994 	for (q = nq; q != NULL; q = nq) {
995 		nq = q->m_nextpkt;
996 		q->m_nextpkt = NULL;
997 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
998 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
999 		m_cat(m, q);
1000 	}
1001 	/*
1002 	 * In order to do checksumming faster we do 'end-around carry' here
1003 	 * (and not in for{} loop), though it implies we are not going to
1004 	 * reassemble more than 64k fragments.
1005 	 */
1006 	m->m_pkthdr.csum_data =
1007 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1008 #ifdef MAC
1009 	mac_create_datagram_from_ipq(fp, m);
1010 	mac_destroy_ipq(fp);
1011 #endif
1012 
1013 	/*
1014 	 * Create header for new ip packet by modifying header of first
1015 	 * packet;  dequeue and discard fragment reassembly header.
1016 	 * Make header visible.
1017 	 */
1018 	ip->ip_len = (ip->ip_hl << 2) + next;
1019 	ip->ip_src = fp->ipq_src;
1020 	ip->ip_dst = fp->ipq_dst;
1021 	TAILQ_REMOVE(head, fp, ipq_list);
1022 	nipq--;
1023 	uma_zfree(ipq_zone, fp);
1024 	m->m_len += (ip->ip_hl << 2);
1025 	m->m_data -= (ip->ip_hl << 2);
1026 	/* some debugging cruft by sklower, below, will go away soon */
1027 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1028 		m_fixhdr(m);
1029 	ipstat.ips_reassembled++;
1030 	IPQ_UNLOCK();
1031 	return (m);
1032 
1033 dropfrag:
1034 	ipstat.ips_fragdropped++;
1035 	if (fp != NULL)
1036 		fp->ipq_nfrags--;
1037 	m_freem(m);
1038 done:
1039 	IPQ_UNLOCK();
1040 	return (NULL);
1041 
1042 #undef GETIP
1043 }
1044 
1045 /*
1046  * Free a fragment reassembly header and all
1047  * associated datagrams.
1048  */
1049 static void
1050 ip_freef(fhp, fp)
1051 	struct ipqhead *fhp;
1052 	struct ipq *fp;
1053 {
1054 	register struct mbuf *q;
1055 
1056 	IPQ_LOCK_ASSERT();
1057 
1058 	while (fp->ipq_frags) {
1059 		q = fp->ipq_frags;
1060 		fp->ipq_frags = q->m_nextpkt;
1061 		m_freem(q);
1062 	}
1063 	TAILQ_REMOVE(fhp, fp, ipq_list);
1064 	uma_zfree(ipq_zone, fp);
1065 	nipq--;
1066 }
1067 
1068 /*
1069  * IP timer processing;
1070  * if a timer expires on a reassembly
1071  * queue, discard it.
1072  */
1073 void
1074 ip_slowtimo()
1075 {
1076 	register struct ipq *fp;
1077 	int i;
1078 
1079 	IPQ_LOCK();
1080 	for (i = 0; i < IPREASS_NHASH; i++) {
1081 		for(fp = TAILQ_FIRST(&ipq[i]); fp;) {
1082 			struct ipq *fpp;
1083 
1084 			fpp = fp;
1085 			fp = TAILQ_NEXT(fp, ipq_list);
1086 			if(--fpp->ipq_ttl == 0) {
1087 				ipstat.ips_fragtimeout += fpp->ipq_nfrags;
1088 				ip_freef(&ipq[i], fpp);
1089 			}
1090 		}
1091 	}
1092 	/*
1093 	 * If we are over the maximum number of fragments
1094 	 * (due to the limit being lowered), drain off
1095 	 * enough to get down to the new limit.
1096 	 */
1097 	if (maxnipq >= 0 && nipq > maxnipq) {
1098 		for (i = 0; i < IPREASS_NHASH; i++) {
1099 			while (nipq > maxnipq && !TAILQ_EMPTY(&ipq[i])) {
1100 				ipstat.ips_fragdropped +=
1101 				    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1102 				ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1103 			}
1104 		}
1105 	}
1106 	IPQ_UNLOCK();
1107 }
1108 
1109 /*
1110  * Drain off all datagram fragments.
1111  */
1112 void
1113 ip_drain()
1114 {
1115 	int     i;
1116 
1117 	IPQ_LOCK();
1118 	for (i = 0; i < IPREASS_NHASH; i++) {
1119 		while(!TAILQ_EMPTY(&ipq[i])) {
1120 			ipstat.ips_fragdropped +=
1121 			    TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1122 			ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1123 		}
1124 	}
1125 	IPQ_UNLOCK();
1126 	in_rtqdrain();
1127 }
1128 
1129 /*
1130  * The protocol to be inserted into ip_protox[] must be already registered
1131  * in inetsw[], either statically or through pf_proto_register().
1132  */
1133 int
1134 ipproto_register(u_char ipproto)
1135 {
1136 	struct protosw *pr;
1137 
1138 	/* Sanity checks. */
1139 	if (ipproto == 0)
1140 		return (EPROTONOSUPPORT);
1141 
1142 	/*
1143 	 * The protocol slot must not be occupied by another protocol
1144 	 * already.  An index pointing to IPPROTO_RAW is unused.
1145 	 */
1146 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1147 	if (pr == NULL)
1148 		return (EPFNOSUPPORT);
1149 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1150 		return (EEXIST);
1151 
1152 	/* Find the protocol position in inetsw[] and set the index. */
1153 	for (pr = inetdomain.dom_protosw;
1154 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1155 		if (pr->pr_domain->dom_family == PF_INET &&
1156 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1157 			/* Be careful to only index valid IP protocols. */
1158 			if (pr->pr_protocol < IPPROTO_MAX) {
1159 				ip_protox[pr->pr_protocol] = pr - inetsw;
1160 				return (0);
1161 			} else
1162 				return (EINVAL);
1163 		}
1164 	}
1165 	return (EPROTONOSUPPORT);
1166 }
1167 
1168 int
1169 ipproto_unregister(u_char ipproto)
1170 {
1171 	struct protosw *pr;
1172 
1173 	/* Sanity checks. */
1174 	if (ipproto == 0)
1175 		return (EPROTONOSUPPORT);
1176 
1177 	/* Check if the protocol was indeed registered. */
1178 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1179 	if (pr == NULL)
1180 		return (EPFNOSUPPORT);
1181 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1182 		return (ENOENT);
1183 
1184 	/* Reset the protocol slot to IPPROTO_RAW. */
1185 	ip_protox[ipproto] = pr - inetsw;
1186 	return (0);
1187 }
1188 
1189 /*
1190  * Given address of next destination (final or next hop),
1191  * return internet address info of interface to be used to get there.
1192  */
1193 struct in_ifaddr *
1194 ip_rtaddr(dst)
1195 	struct in_addr dst;
1196 {
1197 	struct route sro;
1198 	struct sockaddr_in *sin;
1199 	struct in_ifaddr *ifa;
1200 
1201 	bzero(&sro, sizeof(sro));
1202 	sin = (struct sockaddr_in *)&sro.ro_dst;
1203 	sin->sin_family = AF_INET;
1204 	sin->sin_len = sizeof(*sin);
1205 	sin->sin_addr = dst;
1206 	rtalloc_ign(&sro, RTF_CLONING);
1207 
1208 	if (sro.ro_rt == NULL)
1209 		return (NULL);
1210 
1211 	ifa = ifatoia(sro.ro_rt->rt_ifa);
1212 	RTFREE(sro.ro_rt);
1213 	return (ifa);
1214 }
1215 
1216 u_char inetctlerrmap[PRC_NCMDS] = {
1217 	0,		0,		0,		0,
1218 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1219 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1220 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1221 	0,		0,		EHOSTUNREACH,	0,
1222 	ENOPROTOOPT,	ECONNREFUSED
1223 };
1224 
1225 /*
1226  * Forward a packet.  If some error occurs return the sender
1227  * an icmp packet.  Note we can't always generate a meaningful
1228  * icmp message because icmp doesn't have a large enough repertoire
1229  * of codes and types.
1230  *
1231  * If not forwarding, just drop the packet.  This could be confusing
1232  * if ipforwarding was zero but some routing protocol was advancing
1233  * us as a gateway to somewhere.  However, we must let the routing
1234  * protocol deal with that.
1235  *
1236  * The srcrt parameter indicates whether the packet is being forwarded
1237  * via a source route.
1238  */
1239 void
1240 ip_forward(struct mbuf *m, int srcrt)
1241 {
1242 	struct ip *ip = mtod(m, struct ip *);
1243 	struct in_ifaddr *ia = NULL;
1244 	struct mbuf *mcopy;
1245 	struct in_addr dest;
1246 	int error, type = 0, code = 0, mtu = 0;
1247 
1248 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1249 		ipstat.ips_cantforward++;
1250 		m_freem(m);
1251 		return;
1252 	}
1253 #ifdef IPSTEALTH
1254 	if (!ipstealth) {
1255 #endif
1256 		if (ip->ip_ttl <= IPTTLDEC) {
1257 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1258 			    0, 0);
1259 			return;
1260 		}
1261 #ifdef IPSTEALTH
1262 	}
1263 #endif
1264 
1265 	if (!srcrt && (ia = ip_rtaddr(ip->ip_dst)) == NULL) {
1266 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1267 		return;
1268 	}
1269 
1270 	/*
1271 	 * Save the IP header and at most 8 bytes of the payload,
1272 	 * in case we need to generate an ICMP message to the src.
1273 	 *
1274 	 * XXX this can be optimized a lot by saving the data in a local
1275 	 * buffer on the stack (72 bytes at most), and only allocating the
1276 	 * mbuf if really necessary. The vast majority of the packets
1277 	 * are forwarded without having to send an ICMP back (either
1278 	 * because unnecessary, or because rate limited), so we are
1279 	 * really we are wasting a lot of work here.
1280 	 *
1281 	 * We don't use m_copy() because it might return a reference
1282 	 * to a shared cluster. Both this function and ip_output()
1283 	 * assume exclusive access to the IP header in `m', so any
1284 	 * data in a cluster may change before we reach icmp_error().
1285 	 */
1286 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1287 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1288 		/*
1289 		 * It's probably ok if the pkthdr dup fails (because
1290 		 * the deep copy of the tag chain failed), but for now
1291 		 * be conservative and just discard the copy since
1292 		 * code below may some day want the tags.
1293 		 */
1294 		m_free(mcopy);
1295 		mcopy = NULL;
1296 	}
1297 	if (mcopy != NULL) {
1298 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1299 		mcopy->m_pkthdr.len = mcopy->m_len;
1300 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1301 	}
1302 
1303 #ifdef IPSTEALTH
1304 	if (!ipstealth) {
1305 #endif
1306 		ip->ip_ttl -= IPTTLDEC;
1307 #ifdef IPSTEALTH
1308 	}
1309 #endif
1310 
1311 	/*
1312 	 * If forwarding packet using same interface that it came in on,
1313 	 * perhaps should send a redirect to sender to shortcut a hop.
1314 	 * Only send redirect if source is sending directly to us,
1315 	 * and if packet was not source routed (or has any options).
1316 	 * Also, don't send redirect if forwarding using a default route
1317 	 * or a route modified by a redirect.
1318 	 */
1319 	dest.s_addr = 0;
1320 	if (!srcrt && ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1321 		struct sockaddr_in *sin;
1322 		struct route ro;
1323 		struct rtentry *rt;
1324 
1325 		bzero(&ro, sizeof(ro));
1326 		sin = (struct sockaddr_in *)&ro.ro_dst;
1327 		sin->sin_family = AF_INET;
1328 		sin->sin_len = sizeof(*sin);
1329 		sin->sin_addr = ip->ip_dst;
1330 		rtalloc_ign(&ro, RTF_CLONING);
1331 
1332 		rt = ro.ro_rt;
1333 
1334 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1335 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1336 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1337 			u_long src = ntohl(ip->ip_src.s_addr);
1338 
1339 			if (RTA(rt) &&
1340 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1341 				if (rt->rt_flags & RTF_GATEWAY)
1342 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1343 				else
1344 					dest.s_addr = ip->ip_dst.s_addr;
1345 				/* Router requirements says to only send host redirects */
1346 				type = ICMP_REDIRECT;
1347 				code = ICMP_REDIRECT_HOST;
1348 			}
1349 		}
1350 		if (rt)
1351 			RTFREE(rt);
1352 	}
1353 
1354 	error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1355 	if (error)
1356 		ipstat.ips_cantforward++;
1357 	else {
1358 		ipstat.ips_forward++;
1359 		if (type)
1360 			ipstat.ips_redirectsent++;
1361 		else {
1362 			if (mcopy)
1363 				m_freem(mcopy);
1364 			return;
1365 		}
1366 	}
1367 	if (mcopy == NULL)
1368 		return;
1369 
1370 	switch (error) {
1371 
1372 	case 0:				/* forwarded, but need redirect */
1373 		/* type, code set above */
1374 		break;
1375 
1376 	case ENETUNREACH:		/* shouldn't happen, checked above */
1377 	case EHOSTUNREACH:
1378 	case ENETDOWN:
1379 	case EHOSTDOWN:
1380 	default:
1381 		type = ICMP_UNREACH;
1382 		code = ICMP_UNREACH_HOST;
1383 		break;
1384 
1385 	case EMSGSIZE:
1386 		type = ICMP_UNREACH;
1387 		code = ICMP_UNREACH_NEEDFRAG;
1388 
1389 #if defined(IPSEC) || defined(FAST_IPSEC)
1390 		mtu = ip_ipsec_mtu(m);
1391 #endif /* IPSEC */
1392 		/*
1393 		 * If the MTU wasn't set before use the interface mtu or
1394 		 * fall back to the next smaller mtu step compared to the
1395 		 * current packet size.
1396 		 */
1397 		if (mtu == 0) {
1398 			if (ia != NULL)
1399 				mtu = ia->ia_ifp->if_mtu;
1400 			else
1401 				mtu = ip_next_mtu(ip->ip_len, 0);
1402 		}
1403 		ipstat.ips_cantfrag++;
1404 		break;
1405 
1406 	case ENOBUFS:
1407 		/*
1408 		 * A router should not generate ICMP_SOURCEQUENCH as
1409 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1410 		 * Source quench could be a big problem under DoS attacks,
1411 		 * or if the underlying interface is rate-limited.
1412 		 * Those who need source quench packets may re-enable them
1413 		 * via the net.inet.ip.sendsourcequench sysctl.
1414 		 */
1415 		if (ip_sendsourcequench == 0) {
1416 			m_freem(mcopy);
1417 			return;
1418 		} else {
1419 			type = ICMP_SOURCEQUENCH;
1420 			code = 0;
1421 		}
1422 		break;
1423 
1424 	case EACCES:			/* ipfw denied packet */
1425 		m_freem(mcopy);
1426 		return;
1427 	}
1428 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1429 }
1430 
1431 void
1432 ip_savecontrol(inp, mp, ip, m)
1433 	register struct inpcb *inp;
1434 	register struct mbuf **mp;
1435 	register struct ip *ip;
1436 	register struct mbuf *m;
1437 {
1438 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1439 		struct bintime bt;
1440 
1441 		bintime(&bt);
1442 		if (inp->inp_socket->so_options & SO_BINTIME) {
1443 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1444 			SCM_BINTIME, SOL_SOCKET);
1445 			if (*mp)
1446 				mp = &(*mp)->m_next;
1447 		}
1448 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1449 			struct timeval tv;
1450 
1451 			bintime2timeval(&bt, &tv);
1452 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1453 				SCM_TIMESTAMP, SOL_SOCKET);
1454 			if (*mp)
1455 				mp = &(*mp)->m_next;
1456 		}
1457 	}
1458 	if (inp->inp_flags & INP_RECVDSTADDR) {
1459 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1460 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1461 		if (*mp)
1462 			mp = &(*mp)->m_next;
1463 	}
1464 	if (inp->inp_flags & INP_RECVTTL) {
1465 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1466 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1467 		if (*mp)
1468 			mp = &(*mp)->m_next;
1469 	}
1470 #ifdef notyet
1471 	/* XXX
1472 	 * Moving these out of udp_input() made them even more broken
1473 	 * than they already were.
1474 	 */
1475 	/* options were tossed already */
1476 	if (inp->inp_flags & INP_RECVOPTS) {
1477 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1478 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1479 		if (*mp)
1480 			mp = &(*mp)->m_next;
1481 	}
1482 	/* ip_srcroute doesn't do what we want here, need to fix */
1483 	if (inp->inp_flags & INP_RECVRETOPTS) {
1484 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1485 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1486 		if (*mp)
1487 			mp = &(*mp)->m_next;
1488 	}
1489 #endif
1490 	if (inp->inp_flags & INP_RECVIF) {
1491 		struct ifnet *ifp;
1492 		struct sdlbuf {
1493 			struct sockaddr_dl sdl;
1494 			u_char	pad[32];
1495 		} sdlbuf;
1496 		struct sockaddr_dl *sdp;
1497 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1498 
1499 		if (((ifp = m->m_pkthdr.rcvif))
1500 		&& ( ifp->if_index && (ifp->if_index <= if_index))) {
1501 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1502 			/*
1503 			 * Change our mind and don't try copy.
1504 			 */
1505 			if ((sdp->sdl_family != AF_LINK)
1506 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1507 				goto makedummy;
1508 			}
1509 			bcopy(sdp, sdl2, sdp->sdl_len);
1510 		} else {
1511 makedummy:
1512 			sdl2->sdl_len
1513 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1514 			sdl2->sdl_family = AF_LINK;
1515 			sdl2->sdl_index = 0;
1516 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1517 		}
1518 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1519 			IP_RECVIF, IPPROTO_IP);
1520 		if (*mp)
1521 			mp = &(*mp)->m_next;
1522 	}
1523 }
1524 
1525 /*
1526  * XXX these routines are called from the upper part of the kernel.
1527  * They need to be locked when we remove Giant.
1528  *
1529  * They could also be moved to ip_mroute.c, since all the RSVP
1530  *  handling is done there already.
1531  */
1532 static int ip_rsvp_on;
1533 struct socket *ip_rsvpd;
1534 int
1535 ip_rsvp_init(struct socket *so)
1536 {
1537 	if (so->so_type != SOCK_RAW ||
1538 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1539 		return EOPNOTSUPP;
1540 
1541 	if (ip_rsvpd != NULL)
1542 		return EADDRINUSE;
1543 
1544 	ip_rsvpd = so;
1545 	/*
1546 	 * This may seem silly, but we need to be sure we don't over-increment
1547 	 * the RSVP counter, in case something slips up.
1548 	 */
1549 	if (!ip_rsvp_on) {
1550 		ip_rsvp_on = 1;
1551 		rsvp_on++;
1552 	}
1553 
1554 	return 0;
1555 }
1556 
1557 int
1558 ip_rsvp_done(void)
1559 {
1560 	ip_rsvpd = NULL;
1561 	/*
1562 	 * This may seem silly, but we need to be sure we don't over-decrement
1563 	 * the RSVP counter, in case something slips up.
1564 	 */
1565 	if (ip_rsvp_on) {
1566 		ip_rsvp_on = 0;
1567 		rsvp_on--;
1568 	}
1569 	return 0;
1570 }
1571 
1572 void
1573 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1574 {
1575 	if (rsvp_input_p) { /* call the real one if loaded */
1576 		rsvp_input_p(m, off);
1577 		return;
1578 	}
1579 
1580 	/* Can still get packets with rsvp_on = 0 if there is a local member
1581 	 * of the group to which the RSVP packet is addressed.  But in this
1582 	 * case we want to throw the packet away.
1583 	 */
1584 
1585 	if (!rsvp_on) {
1586 		m_freem(m);
1587 		return;
1588 	}
1589 
1590 	if (ip_rsvpd != NULL) {
1591 		rip_input(m, off);
1592 		return;
1593 	}
1594 	/* Drop the packet */
1595 	m_freem(m);
1596 }
1597