xref: /freebsd/sys/netinet/ip_input.c (revision 52267f7411adcc76ede961420e08c0e42f42d415)
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  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_bootp.h"
36 #include "opt_ipfw.h"
37 #include "opt_ipstealth.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_carp.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/mbuf.h>
46 #include <sys/malloc.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/time.h>
51 #include <sys/kernel.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 #include <sys/vimage.h>
55 
56 #include <net/pfil.h>
57 #include <net/if.h>
58 #include <net/if_types.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/route.h>
62 #include <net/netisr.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_icmp.h>
72 #include <netinet/ip_options.h>
73 #include <machine/in_cksum.h>
74 #include <netinet/vinet.h>
75 #ifdef DEV_CARP
76 #include <netinet/ip_carp.h>
77 #endif
78 #ifdef IPSEC
79 #include <netinet/ip_ipsec.h>
80 #endif /* IPSEC */
81 
82 #include <sys/socketvar.h>
83 
84 /* XXX: Temporary until ipfw_ether and ipfw_bridge are converted. */
85 #include <netinet/ip_fw.h>
86 #include <netinet/ip_dummynet.h>
87 
88 #include <security/mac/mac_framework.h>
89 
90 #ifdef CTASSERT
91 CTASSERT(sizeof(struct ip) == 20);
92 #endif
93 
94 #ifdef VIMAGE_GLOBALS
95 static int	ipsendredirects;
96 static int	ip_checkinterface;
97 static int	ip_keepfaith;
98 static int	ip_sendsourcequench;
99 int	ip_defttl;
100 int	ip_do_randomid;
101 int	ipforwarding;
102 struct	in_ifaddrhead in_ifaddrhead; 		/* first inet address */
103 struct	in_ifaddrhashhead *in_ifaddrhashtbl;	/* inet addr hash table  */
104 u_long 	in_ifaddrhmask;				/* mask for hash table */
105 struct ipstat ipstat;
106 static int ip_rsvp_on;
107 struct socket *ip_rsvpd;
108 int	rsvp_on;
109 static struct ipqhead ipq[IPREASS_NHASH];
110 static int	maxnipq;	/* Administrative limit on # reass queues. */
111 static int	maxfragsperpacket;
112 int	ipstealth;
113 static int	nipq;	/* Total # of reass queues */
114 #endif
115 
116 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_FORWARDING,
117     forwarding, CTLFLAG_RW, ipforwarding, 0,
118     "Enable IP forwarding between interfaces");
119 
120 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_SENDREDIRECTS,
121     redirect, CTLFLAG_RW, ipsendredirects, 0,
122     "Enable sending IP redirects");
123 
124 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_DEFTTL,
125     ttl, CTLFLAG_RW, ip_defttl, 0, "Maximum TTL on IP packets");
126 
127 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_KEEPFAITH,
128     keepfaith, CTLFLAG_RW, ip_keepfaith,	0,
129     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
130 
131 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO,
132     sendsourcequench, CTLFLAG_RW, ip_sendsourcequench, 0,
133     "Enable the transmission of source quench packets");
134 
135 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, random_id,
136     CTLFLAG_RW, ip_do_randomid, 0, "Assign random ip_id values");
137 
138 /*
139  * XXX - Setting ip_checkinterface mostly implements the receive side of
140  * the Strong ES model described in RFC 1122, but since the routing table
141  * and transmit implementation do not implement the Strong ES model,
142  * setting this to 1 results in an odd hybrid.
143  *
144  * XXX - ip_checkinterface currently must be disabled if you use ipnat
145  * to translate the destination address to another local interface.
146  *
147  * XXX - ip_checkinterface must be disabled if you add IP aliases
148  * to the loopback interface instead of the interface where the
149  * packets for those addresses are received.
150  */
151 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO,
152     check_interface, CTLFLAG_RW, ip_checkinterface, 0,
153     "Verify packet arrives on correct interface");
154 
155 struct pfil_head inet_pfil_hook;	/* Packet filter hooks */
156 
157 static struct	ifqueue ipintrq;
158 static int	ipqmaxlen = IFQ_MAXLEN;
159 
160 extern	struct domain inetdomain;
161 extern	struct protosw inetsw[];
162 u_char	ip_protox[IPPROTO_MAX];
163 
164 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
165     &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
166 SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
167     &ipintrq.ifq_drops, 0,
168     "Number of packets dropped from the IP input queue");
169 
170 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
171     ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
172 
173 static uma_zone_t ipq_zone;
174 static struct mtx ipqlock;
175 
176 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
177 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
178 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
179 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
180 
181 static void	maxnipq_update(void);
182 static void	ipq_zone_change(void *);
183 
184 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, fragpackets,
185     CTLFLAG_RD, nipq, 0,
186     "Current number of IPv4 fragment reassembly queue entries");
187 
188 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, maxfragsperpacket,
189     CTLFLAG_RW, maxfragsperpacket, 0,
190     "Maximum number of IPv4 fragments allowed per packet");
191 
192 struct callout	ipport_tick_callout;
193 
194 #ifdef IPCTL_DEFMTU
195 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
196     &ip_mtu, 0, "Default MTU");
197 #endif
198 
199 #ifdef IPSTEALTH
200 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
201     ipstealth, 0, "IP stealth mode, no TTL decrementation on forwarding");
202 #endif
203 
204 /*
205  * ipfw_ether and ipfw_bridge hooks.
206  * XXX: Temporary until those are converted to pfil_hooks as well.
207  */
208 ip_fw_chk_t *ip_fw_chk_ptr = NULL;
209 ip_dn_io_t *ip_dn_io_ptr = NULL;
210 int fw_one_pass = 1;
211 
212 static void	ip_freef(struct ipqhead *, struct ipq *);
213 
214 /*
215  * IP initialization: fill in IP protocol switch table.
216  * All protocols not implemented in kernel go to raw IP protocol handler.
217  */
218 void
219 ip_init(void)
220 {
221 	INIT_VNET_INET(curvnet);
222 	struct protosw *pr;
223 	int i;
224 
225 	V_ipsendredirects = 1; /* XXX */
226 	V_ip_checkinterface = 0;
227 	V_ip_keepfaith = 0;
228 	V_ip_sendsourcequench = 0;
229 	V_rsvp_on = 0;
230 	V_ip_defttl = IPDEFTTL;
231 	V_ip_do_randomid = 0;
232 	V_ipforwarding = 0;
233 	V_ipstealth = 0;
234 	V_nipq = 0;	/* Total # of reass queues */
235 
236 	V_ipport_lowfirstauto = IPPORT_RESERVED - 1;	/* 1023 */
237 	V_ipport_lowlastauto = IPPORT_RESERVEDSTART;	/* 600 */
238 	V_ipport_firstauto = IPPORT_EPHEMERALFIRST;	/* 10000 */
239 	V_ipport_lastauto = IPPORT_EPHEMERALLAST;	/* 65535 */
240 	V_ipport_hifirstauto = IPPORT_HIFIRSTAUTO;	/* 49152 */
241 	V_ipport_hilastauto = IPPORT_HILASTAUTO;	/* 65535 */
242 	V_ipport_reservedhigh = IPPORT_RESERVED - 1;	/* 1023 */
243 	V_ipport_reservedlow = 0;
244 	V_ipport_randomized = 1;	/* user controlled via sysctl */
245 	V_ipport_randomcps = 10;	/* user controlled via sysctl */
246 	V_ipport_randomtime = 45;	/* user controlled via sysctl */
247 	V_ipport_stoprandom = 0;	/* toggled by ipport_tick */
248 
249 #ifdef NOTYET
250 	/* XXX global static but not instantiated in this file */
251 	V_ipfastforward_active = 0;
252 	V_subnetsarelocal = 0;
253 	V_sameprefixcarponly = 0;
254 #endif
255 
256 	TAILQ_INIT(&V_in_ifaddrhead);
257 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
258 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
259 	if (pr == NULL)
260 		panic("ip_init: PF_INET not found");
261 
262 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
263 	for (i = 0; i < IPPROTO_MAX; i++)
264 		ip_protox[i] = pr - inetsw;
265 	/*
266 	 * Cycle through IP protocols and put them into the appropriate place
267 	 * in ip_protox[].
268 	 */
269 	for (pr = inetdomain.dom_protosw;
270 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
271 		if (pr->pr_domain->dom_family == PF_INET &&
272 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
273 			/* Be careful to only index valid IP protocols. */
274 			if (pr->pr_protocol < IPPROTO_MAX)
275 				ip_protox[pr->pr_protocol] = pr - inetsw;
276 		}
277 
278 	/* Initialize packet filter hooks. */
279 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
280 	inet_pfil_hook.ph_af = AF_INET;
281 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
282 		printf("%s: WARNING: unable to register pfil hook, "
283 			"error %d\n", __func__, i);
284 
285 	/* Initialize IP reassembly queue. */
286 	IPQ_LOCK_INIT();
287 	for (i = 0; i < IPREASS_NHASH; i++)
288 	    TAILQ_INIT(&V_ipq[i]);
289 	V_maxnipq = nmbclusters / 32;
290 	V_maxfragsperpacket = 16;
291 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
292 	    NULL, UMA_ALIGN_PTR, 0);
293 	maxnipq_update();
294 
295 	/* Start ipport_tick. */
296 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
297 	ipport_tick(NULL);
298 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
299 		SHUTDOWN_PRI_DEFAULT);
300 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
301 		NULL, EVENTHANDLER_PRI_ANY);
302 
303 	/* Initialize various other remaining things. */
304 	V_ip_id = time_second & 0xffff;
305 	ipintrq.ifq_maxlen = ipqmaxlen;
306 	mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
307 	netisr_register(NETISR_IP, ip_input, &ipintrq, 0);
308 }
309 
310 void
311 ip_fini(void *xtp)
312 {
313 
314 	callout_stop(&ipport_tick_callout);
315 }
316 
317 /*
318  * Ip input routine.  Checksum and byte swap header.  If fragmented
319  * try to reassemble.  Process options.  Pass to next level.
320  */
321 void
322 ip_input(struct mbuf *m)
323 {
324 	INIT_VNET_INET(curvnet);
325 	struct ip *ip = NULL;
326 	struct in_ifaddr *ia = NULL;
327 	struct ifaddr *ifa;
328 	int    checkif, hlen = 0;
329 	u_short sum;
330 	int dchg = 0;				/* dest changed after fw */
331 	struct in_addr odst;			/* original dst address */
332 
333 	M_ASSERTPKTHDR(m);
334 
335 	if (m->m_flags & M_FASTFWD_OURS) {
336 		/*
337 		 * Firewall or NAT changed destination to local.
338 		 * We expect ip_len and ip_off to be in host byte order.
339 		 */
340 		m->m_flags &= ~M_FASTFWD_OURS;
341 		/* Set up some basics that will be used later. */
342 		ip = mtod(m, struct ip *);
343 		hlen = ip->ip_hl << 2;
344 		goto ours;
345 	}
346 
347 	V_ipstat.ips_total++;
348 
349 	if (m->m_pkthdr.len < sizeof(struct ip))
350 		goto tooshort;
351 
352 	if (m->m_len < sizeof (struct ip) &&
353 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
354 		V_ipstat.ips_toosmall++;
355 		return;
356 	}
357 	ip = mtod(m, struct ip *);
358 
359 	if (ip->ip_v != IPVERSION) {
360 		V_ipstat.ips_badvers++;
361 		goto bad;
362 	}
363 
364 	hlen = ip->ip_hl << 2;
365 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
366 		V_ipstat.ips_badhlen++;
367 		goto bad;
368 	}
369 	if (hlen > m->m_len) {
370 		if ((m = m_pullup(m, hlen)) == NULL) {
371 			V_ipstat.ips_badhlen++;
372 			return;
373 		}
374 		ip = mtod(m, struct ip *);
375 	}
376 
377 	/* 127/8 must not appear on wire - RFC1122 */
378 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
379 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
380 		if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
381 			V_ipstat.ips_badaddr++;
382 			goto bad;
383 		}
384 	}
385 
386 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
387 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
388 	} else {
389 		if (hlen == sizeof(struct ip)) {
390 			sum = in_cksum_hdr(ip);
391 		} else {
392 			sum = in_cksum(m, hlen);
393 		}
394 	}
395 	if (sum) {
396 		V_ipstat.ips_badsum++;
397 		goto bad;
398 	}
399 
400 #ifdef ALTQ
401 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
402 		/* packet is dropped by traffic conditioner */
403 		return;
404 #endif
405 
406 	/*
407 	 * Convert fields to host representation.
408 	 */
409 	ip->ip_len = ntohs(ip->ip_len);
410 	if (ip->ip_len < hlen) {
411 		V_ipstat.ips_badlen++;
412 		goto bad;
413 	}
414 	ip->ip_off = ntohs(ip->ip_off);
415 
416 	/*
417 	 * Check that the amount of data in the buffers
418 	 * is as at least much as the IP header would have us expect.
419 	 * Trim mbufs if longer than we expect.
420 	 * Drop packet if shorter than we expect.
421 	 */
422 	if (m->m_pkthdr.len < ip->ip_len) {
423 tooshort:
424 		V_ipstat.ips_tooshort++;
425 		goto bad;
426 	}
427 	if (m->m_pkthdr.len > ip->ip_len) {
428 		if (m->m_len == m->m_pkthdr.len) {
429 			m->m_len = ip->ip_len;
430 			m->m_pkthdr.len = ip->ip_len;
431 		} else
432 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
433 	}
434 #ifdef IPSEC
435 	/*
436 	 * Bypass packet filtering for packets from a tunnel (gif).
437 	 */
438 	if (ip_ipsec_filtertunnel(m))
439 		goto passin;
440 #endif /* IPSEC */
441 
442 	/*
443 	 * Run through list of hooks for input packets.
444 	 *
445 	 * NB: Beware of the destination address changing (e.g.
446 	 *     by NAT rewriting).  When this happens, tell
447 	 *     ip_forward to do the right thing.
448 	 */
449 
450 	/* Jump over all PFIL processing if hooks are not active. */
451 	if (!PFIL_HOOKED(&inet_pfil_hook))
452 		goto passin;
453 
454 	odst = ip->ip_dst;
455 	if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
456 	    PFIL_IN, NULL) != 0)
457 		return;
458 	if (m == NULL)			/* consumed by filter */
459 		return;
460 
461 	ip = mtod(m, struct ip *);
462 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
463 
464 #ifdef IPFIREWALL_FORWARD
465 	if (m->m_flags & M_FASTFWD_OURS) {
466 		m->m_flags &= ~M_FASTFWD_OURS;
467 		goto ours;
468 	}
469 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
470 		/*
471 		 * Directly ship on the packet.  This allows to forward packets
472 		 * that were destined for us to some other directly connected
473 		 * host.
474 		 */
475 		ip_forward(m, dchg);
476 		return;
477 	}
478 #endif /* IPFIREWALL_FORWARD */
479 
480 passin:
481 	/*
482 	 * Process options and, if not destined for us,
483 	 * ship it on.  ip_dooptions returns 1 when an
484 	 * error was detected (causing an icmp message
485 	 * to be sent and the original packet to be freed).
486 	 */
487 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
488 		return;
489 
490         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
491          * matter if it is destined to another node, or whether it is
492          * a multicast one, RSVP wants it! and prevents it from being forwarded
493          * anywhere else. Also checks if the rsvp daemon is running before
494 	 * grabbing the packet.
495          */
496 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
497 		goto ours;
498 
499 	/*
500 	 * Check our list of addresses, to see if the packet is for us.
501 	 * If we don't have any addresses, assume any unicast packet
502 	 * we receive might be for us (and let the upper layers deal
503 	 * with it).
504 	 */
505 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
506 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
507 		goto ours;
508 
509 	/*
510 	 * Enable a consistency check between the destination address
511 	 * and the arrival interface for a unicast packet (the RFC 1122
512 	 * strong ES model) if IP forwarding is disabled and the packet
513 	 * is not locally generated and the packet is not subject to
514 	 * 'ipfw fwd'.
515 	 *
516 	 * XXX - Checking also should be disabled if the destination
517 	 * address is ipnat'ed to a different interface.
518 	 *
519 	 * XXX - Checking is incompatible with IP aliases added
520 	 * to the loopback interface instead of the interface where
521 	 * the packets are received.
522 	 *
523 	 * XXX - This is the case for carp vhost IPs as well so we
524 	 * insert a workaround. If the packet got here, we already
525 	 * checked with carp_iamatch() and carp_forus().
526 	 */
527 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
528 	    m->m_pkthdr.rcvif != NULL &&
529 	    ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) &&
530 #ifdef DEV_CARP
531 	    !m->m_pkthdr.rcvif->if_carp &&
532 #endif
533 	    (dchg == 0);
534 
535 	/*
536 	 * Check for exact addresses in the hash bucket.
537 	 */
538 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
539 		/*
540 		 * If the address matches, verify that the packet
541 		 * arrived via the correct interface if checking is
542 		 * enabled.
543 		 */
544 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
545 		    (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
546 			goto ours;
547 	}
548 	/*
549 	 * Check for broadcast addresses.
550 	 *
551 	 * Only accept broadcast packets that arrive via the matching
552 	 * interface.  Reception of forwarded directed broadcasts would
553 	 * be handled via ip_forward() and ether_output() with the loopback
554 	 * into the stack for SIMPLEX interfaces handled by ether_output().
555 	 */
556 	if (m->m_pkthdr.rcvif != NULL &&
557 	    m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
558 	        TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
559 			if (ifa->ifa_addr->sa_family != AF_INET)
560 				continue;
561 			ia = ifatoia(ifa);
562 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
563 			    ip->ip_dst.s_addr)
564 				goto ours;
565 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
566 				goto ours;
567 #ifdef BOOTP_COMPAT
568 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
569 				goto ours;
570 #endif
571 		}
572 	}
573 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
574 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
575 		V_ipstat.ips_cantforward++;
576 		m_freem(m);
577 		return;
578 	}
579 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
580 		struct in_multi *inm;
581 		if (V_ip_mrouter) {
582 			/*
583 			 * If we are acting as a multicast router, all
584 			 * incoming multicast packets are passed to the
585 			 * kernel-level multicast forwarding function.
586 			 * The packet is returned (relatively) intact; if
587 			 * ip_mforward() returns a non-zero value, the packet
588 			 * must be discarded, else it may be accepted below.
589 			 */
590 			if (ip_mforward &&
591 			    ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
592 				V_ipstat.ips_cantforward++;
593 				m_freem(m);
594 				return;
595 			}
596 
597 			/*
598 			 * The process-level routing daemon needs to receive
599 			 * all multicast IGMP packets, whether or not this
600 			 * host belongs to their destination groups.
601 			 */
602 			if (ip->ip_p == IPPROTO_IGMP)
603 				goto ours;
604 			V_ipstat.ips_forward++;
605 		}
606 		/*
607 		 * See if we belong to the destination multicast group on the
608 		 * arrival interface.
609 		 */
610 		IN_MULTI_LOCK();
611 		IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
612 		IN_MULTI_UNLOCK();
613 		if (inm == NULL) {
614 			V_ipstat.ips_notmember++;
615 			m_freem(m);
616 			return;
617 		}
618 		goto ours;
619 	}
620 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
621 		goto ours;
622 	if (ip->ip_dst.s_addr == INADDR_ANY)
623 		goto ours;
624 
625 	/*
626 	 * FAITH(Firewall Aided Internet Translator)
627 	 */
628 	if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
629 		if (V_ip_keepfaith) {
630 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
631 				goto ours;
632 		}
633 		m_freem(m);
634 		return;
635 	}
636 
637 	/*
638 	 * Not for us; forward if possible and desirable.
639 	 */
640 	if (V_ipforwarding == 0) {
641 		V_ipstat.ips_cantforward++;
642 		m_freem(m);
643 	} else {
644 #ifdef IPSEC
645 		if (ip_ipsec_fwd(m))
646 			goto bad;
647 #endif /* IPSEC */
648 		ip_forward(m, dchg);
649 	}
650 	return;
651 
652 ours:
653 #ifdef IPSTEALTH
654 	/*
655 	 * IPSTEALTH: Process non-routing options only
656 	 * if the packet is destined for us.
657 	 */
658 	if (V_ipstealth && hlen > sizeof (struct ip) &&
659 	    ip_dooptions(m, 1))
660 		return;
661 #endif /* IPSTEALTH */
662 
663 	/* Count the packet in the ip address stats */
664 	if (ia != NULL) {
665 		ia->ia_ifa.if_ipackets++;
666 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
667 	}
668 
669 	/*
670 	 * Attempt reassembly; if it succeeds, proceed.
671 	 * ip_reass() will return a different mbuf.
672 	 */
673 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
674 		m = ip_reass(m);
675 		if (m == NULL)
676 			return;
677 		ip = mtod(m, struct ip *);
678 		/* Get the header length of the reassembled packet */
679 		hlen = ip->ip_hl << 2;
680 	}
681 
682 	/*
683 	 * Further protocols expect the packet length to be w/o the
684 	 * IP header.
685 	 */
686 	ip->ip_len -= hlen;
687 
688 #ifdef IPSEC
689 	/*
690 	 * enforce IPsec policy checking if we are seeing last header.
691 	 * note that we do not visit this with protocols with pcb layer
692 	 * code - like udp/tcp/raw ip.
693 	 */
694 	if (ip_ipsec_input(m))
695 		goto bad;
696 #endif /* IPSEC */
697 
698 	/*
699 	 * Switch out to protocol's input routine.
700 	 */
701 	V_ipstat.ips_delivered++;
702 
703 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
704 	return;
705 bad:
706 	m_freem(m);
707 }
708 
709 /*
710  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
711  * max has slightly different semantics than the sysctl, for historical
712  * reasons.
713  */
714 static void
715 maxnipq_update(void)
716 {
717 	INIT_VNET_INET(curvnet);
718 
719 	/*
720 	 * -1 for unlimited allocation.
721 	 */
722 	if (V_maxnipq < 0)
723 		uma_zone_set_max(V_ipq_zone, 0);
724 	/*
725 	 * Positive number for specific bound.
726 	 */
727 	if (V_maxnipq > 0)
728 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
729 	/*
730 	 * Zero specifies no further fragment queue allocation -- set the
731 	 * bound very low, but rely on implementation elsewhere to actually
732 	 * prevent allocation and reclaim current queues.
733 	 */
734 	if (V_maxnipq == 0)
735 		uma_zone_set_max(V_ipq_zone, 1);
736 }
737 
738 static void
739 ipq_zone_change(void *tag)
740 {
741 	INIT_VNET_INET(curvnet);
742 
743 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
744 		V_maxnipq = nmbclusters / 32;
745 		maxnipq_update();
746 	}
747 }
748 
749 static int
750 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
751 {
752 	INIT_VNET_INET(curvnet);
753 	int error, i;
754 
755 	i = V_maxnipq;
756 	error = sysctl_handle_int(oidp, &i, 0, req);
757 	if (error || !req->newptr)
758 		return (error);
759 
760 	/*
761 	 * XXXRW: Might be a good idea to sanity check the argument and place
762 	 * an extreme upper bound.
763 	 */
764 	if (i < -1)
765 		return (EINVAL);
766 	V_maxnipq = i;
767 	maxnipq_update();
768 	return (0);
769 }
770 
771 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
772     NULL, 0, sysctl_maxnipq, "I",
773     "Maximum number of IPv4 fragment reassembly queue entries");
774 
775 /*
776  * Take incoming datagram fragment and try to reassemble it into
777  * whole datagram.  If the argument is the first fragment or one
778  * in between the function will return NULL and store the mbuf
779  * in the fragment chain.  If the argument is the last fragment
780  * the packet will be reassembled and the pointer to the new
781  * mbuf returned for further processing.  Only m_tags attached
782  * to the first packet/fragment are preserved.
783  * The IP header is *NOT* adjusted out of iplen.
784  */
785 struct mbuf *
786 ip_reass(struct mbuf *m)
787 {
788 	INIT_VNET_INET(curvnet);
789 	struct ip *ip;
790 	struct mbuf *p, *q, *nq, *t;
791 	struct ipq *fp = NULL;
792 	struct ipqhead *head;
793 	int i, hlen, next;
794 	u_int8_t ecn, ecn0;
795 	u_short hash;
796 
797 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
798 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
799 		V_ipstat.ips_fragments++;
800 		V_ipstat.ips_fragdropped++;
801 		m_freem(m);
802 		return (NULL);
803 	}
804 
805 	ip = mtod(m, struct ip *);
806 	hlen = ip->ip_hl << 2;
807 
808 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
809 	head = &V_ipq[hash];
810 	IPQ_LOCK();
811 
812 	/*
813 	 * Look for queue of fragments
814 	 * of this datagram.
815 	 */
816 	TAILQ_FOREACH(fp, head, ipq_list)
817 		if (ip->ip_id == fp->ipq_id &&
818 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
819 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
820 #ifdef MAC
821 		    mac_ipq_match(m, fp) &&
822 #endif
823 		    ip->ip_p == fp->ipq_p)
824 			goto found;
825 
826 	fp = NULL;
827 
828 	/*
829 	 * Attempt to trim the number of allocated fragment queues if it
830 	 * exceeds the administrative limit.
831 	 */
832 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
833 		/*
834 		 * drop something from the tail of the current queue
835 		 * before proceeding further
836 		 */
837 		struct ipq *q = TAILQ_LAST(head, ipqhead);
838 		if (q == NULL) {   /* gak */
839 			for (i = 0; i < IPREASS_NHASH; i++) {
840 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
841 				if (r) {
842 					V_ipstat.ips_fragtimeout +=
843 					    r->ipq_nfrags;
844 					ip_freef(&V_ipq[i], r);
845 					break;
846 				}
847 			}
848 		} else {
849 			V_ipstat.ips_fragtimeout += q->ipq_nfrags;
850 			ip_freef(head, q);
851 		}
852 	}
853 
854 found:
855 	/*
856 	 * Adjust ip_len to not reflect header,
857 	 * convert offset of this to bytes.
858 	 */
859 	ip->ip_len -= hlen;
860 	if (ip->ip_off & IP_MF) {
861 		/*
862 		 * Make sure that fragments have a data length
863 		 * that's a non-zero multiple of 8 bytes.
864 		 */
865 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
866 			V_ipstat.ips_toosmall++; /* XXX */
867 			goto dropfrag;
868 		}
869 		m->m_flags |= M_FRAG;
870 	} else
871 		m->m_flags &= ~M_FRAG;
872 	ip->ip_off <<= 3;
873 
874 
875 	/*
876 	 * Attempt reassembly; if it succeeds, proceed.
877 	 * ip_reass() will return a different mbuf.
878 	 */
879 	V_ipstat.ips_fragments++;
880 	m->m_pkthdr.header = ip;
881 
882 	/* Previous ip_reass() started here. */
883 	/*
884 	 * Presence of header sizes in mbufs
885 	 * would confuse code below.
886 	 */
887 	m->m_data += hlen;
888 	m->m_len -= hlen;
889 
890 	/*
891 	 * If first fragment to arrive, create a reassembly queue.
892 	 */
893 	if (fp == NULL) {
894 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
895 		if (fp == NULL)
896 			goto dropfrag;
897 #ifdef MAC
898 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
899 			uma_zfree(V_ipq_zone, fp);
900 			fp = NULL;
901 			goto dropfrag;
902 		}
903 		mac_ipq_create(m, fp);
904 #endif
905 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
906 		V_nipq++;
907 		fp->ipq_nfrags = 1;
908 		fp->ipq_ttl = IPFRAGTTL;
909 		fp->ipq_p = ip->ip_p;
910 		fp->ipq_id = ip->ip_id;
911 		fp->ipq_src = ip->ip_src;
912 		fp->ipq_dst = ip->ip_dst;
913 		fp->ipq_frags = m;
914 		m->m_nextpkt = NULL;
915 		goto done;
916 	} else {
917 		fp->ipq_nfrags++;
918 #ifdef MAC
919 		mac_ipq_update(m, fp);
920 #endif
921 	}
922 
923 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
924 
925 	/*
926 	 * Handle ECN by comparing this segment with the first one;
927 	 * if CE is set, do not lose CE.
928 	 * drop if CE and not-ECT are mixed for the same packet.
929 	 */
930 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
931 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
932 	if (ecn == IPTOS_ECN_CE) {
933 		if (ecn0 == IPTOS_ECN_NOTECT)
934 			goto dropfrag;
935 		if (ecn0 != IPTOS_ECN_CE)
936 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
937 	}
938 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
939 		goto dropfrag;
940 
941 	/*
942 	 * Find a segment which begins after this one does.
943 	 */
944 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
945 		if (GETIP(q)->ip_off > ip->ip_off)
946 			break;
947 
948 	/*
949 	 * If there is a preceding segment, it may provide some of
950 	 * our data already.  If so, drop the data from the incoming
951 	 * segment.  If it provides all of our data, drop us, otherwise
952 	 * stick new segment in the proper place.
953 	 *
954 	 * If some of the data is dropped from the the preceding
955 	 * segment, then it's checksum is invalidated.
956 	 */
957 	if (p) {
958 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
959 		if (i > 0) {
960 			if (i >= ip->ip_len)
961 				goto dropfrag;
962 			m_adj(m, i);
963 			m->m_pkthdr.csum_flags = 0;
964 			ip->ip_off += i;
965 			ip->ip_len -= i;
966 		}
967 		m->m_nextpkt = p->m_nextpkt;
968 		p->m_nextpkt = m;
969 	} else {
970 		m->m_nextpkt = fp->ipq_frags;
971 		fp->ipq_frags = m;
972 	}
973 
974 	/*
975 	 * While we overlap succeeding segments trim them or,
976 	 * if they are completely covered, dequeue them.
977 	 */
978 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
979 	     q = nq) {
980 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
981 		if (i < GETIP(q)->ip_len) {
982 			GETIP(q)->ip_len -= i;
983 			GETIP(q)->ip_off += i;
984 			m_adj(q, i);
985 			q->m_pkthdr.csum_flags = 0;
986 			break;
987 		}
988 		nq = q->m_nextpkt;
989 		m->m_nextpkt = nq;
990 		V_ipstat.ips_fragdropped++;
991 		fp->ipq_nfrags--;
992 		m_freem(q);
993 	}
994 
995 	/*
996 	 * Check for complete reassembly and perform frag per packet
997 	 * limiting.
998 	 *
999 	 * Frag limiting is performed here so that the nth frag has
1000 	 * a chance to complete the packet before we drop the packet.
1001 	 * As a result, n+1 frags are actually allowed per packet, but
1002 	 * only n will ever be stored. (n = maxfragsperpacket.)
1003 	 *
1004 	 */
1005 	next = 0;
1006 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1007 		if (GETIP(q)->ip_off != next) {
1008 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1009 				V_ipstat.ips_fragdropped += fp->ipq_nfrags;
1010 				ip_freef(head, fp);
1011 			}
1012 			goto done;
1013 		}
1014 		next += GETIP(q)->ip_len;
1015 	}
1016 	/* Make sure the last packet didn't have the IP_MF flag */
1017 	if (p->m_flags & M_FRAG) {
1018 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1019 			V_ipstat.ips_fragdropped += fp->ipq_nfrags;
1020 			ip_freef(head, fp);
1021 		}
1022 		goto done;
1023 	}
1024 
1025 	/*
1026 	 * Reassembly is complete.  Make sure the packet is a sane size.
1027 	 */
1028 	q = fp->ipq_frags;
1029 	ip = GETIP(q);
1030 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1031 		V_ipstat.ips_toolong++;
1032 		V_ipstat.ips_fragdropped += fp->ipq_nfrags;
1033 		ip_freef(head, fp);
1034 		goto done;
1035 	}
1036 
1037 	/*
1038 	 * Concatenate fragments.
1039 	 */
1040 	m = q;
1041 	t = m->m_next;
1042 	m->m_next = NULL;
1043 	m_cat(m, t);
1044 	nq = q->m_nextpkt;
1045 	q->m_nextpkt = NULL;
1046 	for (q = nq; q != NULL; q = nq) {
1047 		nq = q->m_nextpkt;
1048 		q->m_nextpkt = NULL;
1049 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1050 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1051 		m_cat(m, q);
1052 	}
1053 	/*
1054 	 * In order to do checksumming faster we do 'end-around carry' here
1055 	 * (and not in for{} loop), though it implies we are not going to
1056 	 * reassemble more than 64k fragments.
1057 	 */
1058 	m->m_pkthdr.csum_data =
1059 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1060 #ifdef MAC
1061 	mac_ipq_reassemble(fp, m);
1062 	mac_ipq_destroy(fp);
1063 #endif
1064 
1065 	/*
1066 	 * Create header for new ip packet by modifying header of first
1067 	 * packet;  dequeue and discard fragment reassembly header.
1068 	 * Make header visible.
1069 	 */
1070 	ip->ip_len = (ip->ip_hl << 2) + next;
1071 	ip->ip_src = fp->ipq_src;
1072 	ip->ip_dst = fp->ipq_dst;
1073 	TAILQ_REMOVE(head, fp, ipq_list);
1074 	V_nipq--;
1075 	uma_zfree(V_ipq_zone, fp);
1076 	m->m_len += (ip->ip_hl << 2);
1077 	m->m_data -= (ip->ip_hl << 2);
1078 	/* some debugging cruft by sklower, below, will go away soon */
1079 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1080 		m_fixhdr(m);
1081 	V_ipstat.ips_reassembled++;
1082 	IPQ_UNLOCK();
1083 	return (m);
1084 
1085 dropfrag:
1086 	V_ipstat.ips_fragdropped++;
1087 	if (fp != NULL)
1088 		fp->ipq_nfrags--;
1089 	m_freem(m);
1090 done:
1091 	IPQ_UNLOCK();
1092 	return (NULL);
1093 
1094 #undef GETIP
1095 }
1096 
1097 /*
1098  * Free a fragment reassembly header and all
1099  * associated datagrams.
1100  */
1101 static void
1102 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1103 {
1104 	INIT_VNET_INET(curvnet);
1105 	struct mbuf *q;
1106 
1107 	IPQ_LOCK_ASSERT();
1108 
1109 	while (fp->ipq_frags) {
1110 		q = fp->ipq_frags;
1111 		fp->ipq_frags = q->m_nextpkt;
1112 		m_freem(q);
1113 	}
1114 	TAILQ_REMOVE(fhp, fp, ipq_list);
1115 	uma_zfree(V_ipq_zone, fp);
1116 	V_nipq--;
1117 }
1118 
1119 /*
1120  * IP timer processing;
1121  * if a timer expires on a reassembly
1122  * queue, discard it.
1123  */
1124 void
1125 ip_slowtimo(void)
1126 {
1127 	VNET_ITERATOR_DECL(vnet_iter);
1128 	struct ipq *fp;
1129 	int i;
1130 
1131 	IPQ_LOCK();
1132 	VNET_LIST_RLOCK();
1133 	VNET_FOREACH(vnet_iter) {
1134 		CURVNET_SET(vnet_iter);
1135 		INIT_VNET_INET(vnet_iter);
1136 		for (i = 0; i < IPREASS_NHASH; i++) {
1137 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1138 				struct ipq *fpp;
1139 
1140 				fpp = fp;
1141 				fp = TAILQ_NEXT(fp, ipq_list);
1142 				if(--fpp->ipq_ttl == 0) {
1143 					V_ipstat.ips_fragtimeout +=
1144 					    fpp->ipq_nfrags;
1145 					ip_freef(&V_ipq[i], fpp);
1146 				}
1147 			}
1148 		}
1149 		/*
1150 		 * If we are over the maximum number of fragments
1151 		 * (due to the limit being lowered), drain off
1152 		 * enough to get down to the new limit.
1153 		 */
1154 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1155 			for (i = 0; i < IPREASS_NHASH; i++) {
1156 				while (V_nipq > V_maxnipq &&
1157 				    !TAILQ_EMPTY(&V_ipq[i])) {
1158 					V_ipstat.ips_fragdropped +=
1159 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags;
1160 					ip_freef(&V_ipq[i],
1161 					    TAILQ_FIRST(&V_ipq[i]));
1162 				}
1163 			}
1164 		}
1165 		CURVNET_RESTORE();
1166 	}
1167 	VNET_LIST_RUNLOCK();
1168 	IPQ_UNLOCK();
1169 }
1170 
1171 /*
1172  * Drain off all datagram fragments.
1173  */
1174 void
1175 ip_drain(void)
1176 {
1177 	VNET_ITERATOR_DECL(vnet_iter);
1178 	int     i;
1179 
1180 	IPQ_LOCK();
1181 	VNET_LIST_RLOCK();
1182 	VNET_FOREACH(vnet_iter) {
1183 		CURVNET_SET(vnet_iter);
1184 		INIT_VNET_INET(vnet_iter);
1185 		for (i = 0; i < IPREASS_NHASH; i++) {
1186 			while(!TAILQ_EMPTY(&V_ipq[i])) {
1187 				V_ipstat.ips_fragdropped +=
1188 				    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags;
1189 				ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1190 			}
1191 		}
1192 		CURVNET_RESTORE();
1193 	}
1194 	VNET_LIST_RUNLOCK();
1195 	IPQ_UNLOCK();
1196 	in_rtqdrain();
1197 }
1198 
1199 /*
1200  * The protocol to be inserted into ip_protox[] must be already registered
1201  * in inetsw[], either statically or through pf_proto_register().
1202  */
1203 int
1204 ipproto_register(u_char ipproto)
1205 {
1206 	struct protosw *pr;
1207 
1208 	/* Sanity checks. */
1209 	if (ipproto == 0)
1210 		return (EPROTONOSUPPORT);
1211 
1212 	/*
1213 	 * The protocol slot must not be occupied by another protocol
1214 	 * already.  An index pointing to IPPROTO_RAW is unused.
1215 	 */
1216 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1217 	if (pr == NULL)
1218 		return (EPFNOSUPPORT);
1219 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1220 		return (EEXIST);
1221 
1222 	/* Find the protocol position in inetsw[] and set the index. */
1223 	for (pr = inetdomain.dom_protosw;
1224 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1225 		if (pr->pr_domain->dom_family == PF_INET &&
1226 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1227 			/* Be careful to only index valid IP protocols. */
1228 			if (pr->pr_protocol < IPPROTO_MAX) {
1229 				ip_protox[pr->pr_protocol] = pr - inetsw;
1230 				return (0);
1231 			} else
1232 				return (EINVAL);
1233 		}
1234 	}
1235 	return (EPROTONOSUPPORT);
1236 }
1237 
1238 int
1239 ipproto_unregister(u_char ipproto)
1240 {
1241 	struct protosw *pr;
1242 
1243 	/* Sanity checks. */
1244 	if (ipproto == 0)
1245 		return (EPROTONOSUPPORT);
1246 
1247 	/* Check if the protocol was indeed registered. */
1248 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1249 	if (pr == NULL)
1250 		return (EPFNOSUPPORT);
1251 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1252 		return (ENOENT);
1253 
1254 	/* Reset the protocol slot to IPPROTO_RAW. */
1255 	ip_protox[ipproto] = pr - inetsw;
1256 	return (0);
1257 }
1258 
1259 /*
1260  * Given address of next destination (final or next hop),
1261  * return internet address info of interface to be used to get there.
1262  */
1263 struct in_ifaddr *
1264 ip_rtaddr(struct in_addr dst, u_int fibnum)
1265 {
1266 	struct route sro;
1267 	struct sockaddr_in *sin;
1268 	struct in_ifaddr *ifa;
1269 
1270 	bzero(&sro, sizeof(sro));
1271 	sin = (struct sockaddr_in *)&sro.ro_dst;
1272 	sin->sin_family = AF_INET;
1273 	sin->sin_len = sizeof(*sin);
1274 	sin->sin_addr = dst;
1275 	in_rtalloc_ign(&sro, RTF_CLONING, fibnum);
1276 
1277 	if (sro.ro_rt == NULL)
1278 		return (NULL);
1279 
1280 	ifa = ifatoia(sro.ro_rt->rt_ifa);
1281 	RTFREE(sro.ro_rt);
1282 	return (ifa);
1283 }
1284 
1285 u_char inetctlerrmap[PRC_NCMDS] = {
1286 	0,		0,		0,		0,
1287 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1288 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1289 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1290 	0,		0,		EHOSTUNREACH,	0,
1291 	ENOPROTOOPT,	ECONNREFUSED
1292 };
1293 
1294 /*
1295  * Forward a packet.  If some error occurs return the sender
1296  * an icmp packet.  Note we can't always generate a meaningful
1297  * icmp message because icmp doesn't have a large enough repertoire
1298  * of codes and types.
1299  *
1300  * If not forwarding, just drop the packet.  This could be confusing
1301  * if ipforwarding was zero but some routing protocol was advancing
1302  * us as a gateway to somewhere.  However, we must let the routing
1303  * protocol deal with that.
1304  *
1305  * The srcrt parameter indicates whether the packet is being forwarded
1306  * via a source route.
1307  */
1308 void
1309 ip_forward(struct mbuf *m, int srcrt)
1310 {
1311 	INIT_VNET_INET(curvnet);
1312 	struct ip *ip = mtod(m, struct ip *);
1313 	struct in_ifaddr *ia = NULL;
1314 	struct mbuf *mcopy;
1315 	struct in_addr dest;
1316 	struct route ro;
1317 	int error, type = 0, code = 0, mtu = 0;
1318 
1319 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1320 		V_ipstat.ips_cantforward++;
1321 		m_freem(m);
1322 		return;
1323 	}
1324 #ifdef IPSTEALTH
1325 	if (!V_ipstealth) {
1326 #endif
1327 		if (ip->ip_ttl <= IPTTLDEC) {
1328 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1329 			    0, 0);
1330 			return;
1331 		}
1332 #ifdef IPSTEALTH
1333 	}
1334 #endif
1335 
1336 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1337 	if (!srcrt && ia == NULL) {
1338 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1339 		return;
1340 	}
1341 
1342 	/*
1343 	 * Save the IP header and at most 8 bytes of the payload,
1344 	 * in case we need to generate an ICMP message to the src.
1345 	 *
1346 	 * XXX this can be optimized a lot by saving the data in a local
1347 	 * buffer on the stack (72 bytes at most), and only allocating the
1348 	 * mbuf if really necessary. The vast majority of the packets
1349 	 * are forwarded without having to send an ICMP back (either
1350 	 * because unnecessary, or because rate limited), so we are
1351 	 * really we are wasting a lot of work here.
1352 	 *
1353 	 * We don't use m_copy() because it might return a reference
1354 	 * to a shared cluster. Both this function and ip_output()
1355 	 * assume exclusive access to the IP header in `m', so any
1356 	 * data in a cluster may change before we reach icmp_error().
1357 	 */
1358 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1359 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1360 		/*
1361 		 * It's probably ok if the pkthdr dup fails (because
1362 		 * the deep copy of the tag chain failed), but for now
1363 		 * be conservative and just discard the copy since
1364 		 * code below may some day want the tags.
1365 		 */
1366 		m_free(mcopy);
1367 		mcopy = NULL;
1368 	}
1369 	if (mcopy != NULL) {
1370 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1371 		mcopy->m_pkthdr.len = mcopy->m_len;
1372 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1373 	}
1374 
1375 #ifdef IPSTEALTH
1376 	if (!V_ipstealth) {
1377 #endif
1378 		ip->ip_ttl -= IPTTLDEC;
1379 #ifdef IPSTEALTH
1380 	}
1381 #endif
1382 
1383 	/*
1384 	 * If forwarding packet using same interface that it came in on,
1385 	 * perhaps should send a redirect to sender to shortcut a hop.
1386 	 * Only send redirect if source is sending directly to us,
1387 	 * and if packet was not source routed (or has any options).
1388 	 * Also, don't send redirect if forwarding using a default route
1389 	 * or a route modified by a redirect.
1390 	 */
1391 	dest.s_addr = 0;
1392 	if (!srcrt && V_ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1393 		struct sockaddr_in *sin;
1394 		struct rtentry *rt;
1395 
1396 		bzero(&ro, sizeof(ro));
1397 		sin = (struct sockaddr_in *)&ro.ro_dst;
1398 		sin->sin_family = AF_INET;
1399 		sin->sin_len = sizeof(*sin);
1400 		sin->sin_addr = ip->ip_dst;
1401 		in_rtalloc_ign(&ro, RTF_CLONING, M_GETFIB(m));
1402 
1403 		rt = ro.ro_rt;
1404 
1405 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1406 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1407 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1408 			u_long src = ntohl(ip->ip_src.s_addr);
1409 
1410 			if (RTA(rt) &&
1411 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1412 				if (rt->rt_flags & RTF_GATEWAY)
1413 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1414 				else
1415 					dest.s_addr = ip->ip_dst.s_addr;
1416 				/* Router requirements says to only send host redirects */
1417 				type = ICMP_REDIRECT;
1418 				code = ICMP_REDIRECT_HOST;
1419 			}
1420 		}
1421 		if (rt)
1422 			RTFREE(rt);
1423 	}
1424 
1425 	/*
1426 	 * Try to cache the route MTU from ip_output so we can consider it for
1427 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1428 	 */
1429 	bzero(&ro, sizeof(ro));
1430 
1431 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1432 
1433 	if (error == EMSGSIZE && ro.ro_rt)
1434 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1435 	if (ro.ro_rt)
1436 		RTFREE(ro.ro_rt);
1437 
1438 	if (error)
1439 		V_ipstat.ips_cantforward++;
1440 	else {
1441 		V_ipstat.ips_forward++;
1442 		if (type)
1443 			V_ipstat.ips_redirectsent++;
1444 		else {
1445 			if (mcopy)
1446 				m_freem(mcopy);
1447 			return;
1448 		}
1449 	}
1450 	if (mcopy == NULL)
1451 		return;
1452 
1453 	switch (error) {
1454 
1455 	case 0:				/* forwarded, but need redirect */
1456 		/* type, code set above */
1457 		break;
1458 
1459 	case ENETUNREACH:		/* shouldn't happen, checked above */
1460 	case EHOSTUNREACH:
1461 	case ENETDOWN:
1462 	case EHOSTDOWN:
1463 	default:
1464 		type = ICMP_UNREACH;
1465 		code = ICMP_UNREACH_HOST;
1466 		break;
1467 
1468 	case EMSGSIZE:
1469 		type = ICMP_UNREACH;
1470 		code = ICMP_UNREACH_NEEDFRAG;
1471 
1472 #ifdef IPSEC
1473 		/*
1474 		 * If IPsec is configured for this path,
1475 		 * override any possibly mtu value set by ip_output.
1476 		 */
1477 		mtu = ip_ipsec_mtu(m, mtu);
1478 #endif /* IPSEC */
1479 		/*
1480 		 * If the MTU was set before make sure we are below the
1481 		 * interface MTU.
1482 		 * If the MTU wasn't set before use the interface mtu or
1483 		 * fall back to the next smaller mtu step compared to the
1484 		 * current packet size.
1485 		 */
1486 		if (mtu != 0) {
1487 			if (ia != NULL)
1488 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1489 		} else {
1490 			if (ia != NULL)
1491 				mtu = ia->ia_ifp->if_mtu;
1492 			else
1493 				mtu = ip_next_mtu(ip->ip_len, 0);
1494 		}
1495 		V_ipstat.ips_cantfrag++;
1496 		break;
1497 
1498 	case ENOBUFS:
1499 		/*
1500 		 * A router should not generate ICMP_SOURCEQUENCH as
1501 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1502 		 * Source quench could be a big problem under DoS attacks,
1503 		 * or if the underlying interface is rate-limited.
1504 		 * Those who need source quench packets may re-enable them
1505 		 * via the net.inet.ip.sendsourcequench sysctl.
1506 		 */
1507 		if (V_ip_sendsourcequench == 0) {
1508 			m_freem(mcopy);
1509 			return;
1510 		} else {
1511 			type = ICMP_SOURCEQUENCH;
1512 			code = 0;
1513 		}
1514 		break;
1515 
1516 	case EACCES:			/* ipfw denied packet */
1517 		m_freem(mcopy);
1518 		return;
1519 	}
1520 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1521 }
1522 
1523 void
1524 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1525     struct mbuf *m)
1526 {
1527 	INIT_VNET_NET(inp->inp_vnet);
1528 
1529 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1530 		struct bintime bt;
1531 
1532 		bintime(&bt);
1533 		if (inp->inp_socket->so_options & SO_BINTIME) {
1534 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1535 			SCM_BINTIME, SOL_SOCKET);
1536 			if (*mp)
1537 				mp = &(*mp)->m_next;
1538 		}
1539 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1540 			struct timeval tv;
1541 
1542 			bintime2timeval(&bt, &tv);
1543 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1544 				SCM_TIMESTAMP, SOL_SOCKET);
1545 			if (*mp)
1546 				mp = &(*mp)->m_next;
1547 		}
1548 	}
1549 	if (inp->inp_flags & INP_RECVDSTADDR) {
1550 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1551 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1552 		if (*mp)
1553 			mp = &(*mp)->m_next;
1554 	}
1555 	if (inp->inp_flags & INP_RECVTTL) {
1556 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1557 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1558 		if (*mp)
1559 			mp = &(*mp)->m_next;
1560 	}
1561 #ifdef notyet
1562 	/* XXX
1563 	 * Moving these out of udp_input() made them even more broken
1564 	 * than they already were.
1565 	 */
1566 	/* options were tossed already */
1567 	if (inp->inp_flags & INP_RECVOPTS) {
1568 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1569 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1570 		if (*mp)
1571 			mp = &(*mp)->m_next;
1572 	}
1573 	/* ip_srcroute doesn't do what we want here, need to fix */
1574 	if (inp->inp_flags & INP_RECVRETOPTS) {
1575 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1576 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1577 		if (*mp)
1578 			mp = &(*mp)->m_next;
1579 	}
1580 #endif
1581 	if (inp->inp_flags & INP_RECVIF) {
1582 		struct ifnet *ifp;
1583 		struct sdlbuf {
1584 			struct sockaddr_dl sdl;
1585 			u_char	pad[32];
1586 		} sdlbuf;
1587 		struct sockaddr_dl *sdp;
1588 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1589 
1590 		if (((ifp = m->m_pkthdr.rcvif))
1591 		&& ( ifp->if_index && (ifp->if_index <= V_if_index))) {
1592 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1593 			/*
1594 			 * Change our mind and don't try copy.
1595 			 */
1596 			if ((sdp->sdl_family != AF_LINK)
1597 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1598 				goto makedummy;
1599 			}
1600 			bcopy(sdp, sdl2, sdp->sdl_len);
1601 		} else {
1602 makedummy:
1603 			sdl2->sdl_len
1604 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1605 			sdl2->sdl_family = AF_LINK;
1606 			sdl2->sdl_index = 0;
1607 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1608 		}
1609 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1610 			IP_RECVIF, IPPROTO_IP);
1611 		if (*mp)
1612 			mp = &(*mp)->m_next;
1613 	}
1614 }
1615 
1616 /*
1617  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1618  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1619  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1620  * compiled.
1621  */
1622 int
1623 ip_rsvp_init(struct socket *so)
1624 {
1625 	INIT_VNET_INET(so->so_vnet);
1626 
1627 	if (so->so_type != SOCK_RAW ||
1628 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1629 		return EOPNOTSUPP;
1630 
1631 	if (V_ip_rsvpd != NULL)
1632 		return EADDRINUSE;
1633 
1634 	V_ip_rsvpd = so;
1635 	/*
1636 	 * This may seem silly, but we need to be sure we don't over-increment
1637 	 * the RSVP counter, in case something slips up.
1638 	 */
1639 	if (!V_ip_rsvp_on) {
1640 		V_ip_rsvp_on = 1;
1641 		V_rsvp_on++;
1642 	}
1643 
1644 	return 0;
1645 }
1646 
1647 int
1648 ip_rsvp_done(void)
1649 {
1650 	INIT_VNET_INET(curvnet);
1651 
1652 	V_ip_rsvpd = NULL;
1653 	/*
1654 	 * This may seem silly, but we need to be sure we don't over-decrement
1655 	 * the RSVP counter, in case something slips up.
1656 	 */
1657 	if (V_ip_rsvp_on) {
1658 		V_ip_rsvp_on = 0;
1659 		V_rsvp_on--;
1660 	}
1661 	return 0;
1662 }
1663 
1664 void
1665 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1666 {
1667 	INIT_VNET_INET(curvnet);
1668 
1669 	if (rsvp_input_p) { /* call the real one if loaded */
1670 		rsvp_input_p(m, off);
1671 		return;
1672 	}
1673 
1674 	/* Can still get packets with rsvp_on = 0 if there is a local member
1675 	 * of the group to which the RSVP packet is addressed.  But in this
1676 	 * case we want to throw the packet away.
1677 	 */
1678 
1679 	if (!V_rsvp_on) {
1680 		m_freem(m);
1681 		return;
1682 	}
1683 
1684 	if (V_ip_rsvpd != NULL) {
1685 		rip_input(m, off);
1686 		return;
1687 	}
1688 	/* Drop the packet */
1689 	m_freem(m);
1690 }
1691