xref: /freebsd/sys/netinet/ip_input.c (revision f5f7c05209ca2c3748fd8b27c5e80ffad49120eb)
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_route.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.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/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 
55 #include <net/pfil.h>
56 #include <net/if.h>
57 #include <net/if_types.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/netisr.h>
62 #include <net/vnet.h>
63 #include <net/flowtable.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_fw.h>
72 #include <netinet/ip_icmp.h>
73 #include <netinet/ip_options.h>
74 #include <machine/in_cksum.h>
75 #include <netinet/ip_carp.h>
76 #ifdef IPSEC
77 #include <netinet/ip_ipsec.h>
78 #endif /* IPSEC */
79 
80 #include <sys/socketvar.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #ifdef CTASSERT
85 CTASSERT(sizeof(struct ip) == 20);
86 #endif
87 
88 struct	rwlock in_ifaddr_lock;
89 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock");
90 
91 VNET_DEFINE(int, rsvp_on);
92 
93 VNET_DEFINE(int, ipforwarding);
94 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
95     &VNET_NAME(ipforwarding), 0,
96     "Enable IP forwarding between interfaces");
97 
98 static VNET_DEFINE(int, ipsendredirects) = 1;	/* XXX */
99 #define	V_ipsendredirects	VNET(ipsendredirects)
100 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
101     &VNET_NAME(ipsendredirects), 0,
102     "Enable sending IP redirects");
103 
104 static VNET_DEFINE(int, ip_keepfaith);
105 #define	V_ip_keepfaith		VNET(ip_keepfaith)
106 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
107     &VNET_NAME(ip_keepfaith), 0,
108     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
109 
110 static VNET_DEFINE(int, ip_sendsourcequench);
111 #define	V_ip_sendsourcequench	VNET(ip_sendsourcequench)
112 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
113     &VNET_NAME(ip_sendsourcequench), 0,
114     "Enable the transmission of source quench packets");
115 
116 VNET_DEFINE(int, ip_do_randomid);
117 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
118     &VNET_NAME(ip_do_randomid), 0,
119     "Assign random ip_id values");
120 
121 /*
122  * XXX - Setting ip_checkinterface mostly implements the receive side of
123  * the Strong ES model described in RFC 1122, but since the routing table
124  * and transmit implementation do not implement the Strong ES model,
125  * setting this to 1 results in an odd hybrid.
126  *
127  * XXX - ip_checkinterface currently must be disabled if you use ipnat
128  * to translate the destination address to another local interface.
129  *
130  * XXX - ip_checkinterface must be disabled if you add IP aliases
131  * to the loopback interface instead of the interface where the
132  * packets for those addresses are received.
133  */
134 static VNET_DEFINE(int, ip_checkinterface);
135 #define	V_ip_checkinterface	VNET(ip_checkinterface)
136 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
137     &VNET_NAME(ip_checkinterface), 0,
138     "Verify packet arrives on correct interface");
139 
140 VNET_DEFINE(struct pfil_head, inet_pfil_hook);	/* Packet filter hooks */
141 
142 static struct netisr_handler ip_nh = {
143 	.nh_name = "ip",
144 	.nh_handler = ip_input,
145 	.nh_proto = NETISR_IP,
146 	.nh_policy = NETISR_POLICY_FLOW,
147 };
148 
149 extern	struct domain inetdomain;
150 extern	struct protosw inetsw[];
151 u_char	ip_protox[IPPROTO_MAX];
152 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead);  /* first inet address */
153 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table  */
154 VNET_DEFINE(u_long, in_ifaddrhmask);		/* mask for hash table */
155 
156 VNET_DEFINE(struct ipstat, ipstat);
157 SYSCTL_VNET_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
158     &VNET_NAME(ipstat), ipstat,
159     "IP statistics (struct ipstat, netinet/ip_var.h)");
160 
161 static VNET_DEFINE(uma_zone_t, ipq_zone);
162 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]);
163 static struct mtx ipqlock;
164 
165 #define	V_ipq_zone		VNET(ipq_zone)
166 #define	V_ipq			VNET(ipq)
167 
168 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
169 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
170 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
171 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
172 
173 static void	maxnipq_update(void);
174 static void	ipq_zone_change(void *);
175 static void	ip_drain_locked(void);
176 
177 static VNET_DEFINE(int, maxnipq);  /* Administrative limit on # reass queues. */
178 static VNET_DEFINE(int, nipq);			/* Total # of reass queues */
179 #define	V_maxnipq		VNET(maxnipq)
180 #define	V_nipq			VNET(nipq)
181 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
182     &VNET_NAME(nipq), 0,
183     "Current number of IPv4 fragment reassembly queue entries");
184 
185 static VNET_DEFINE(int, maxfragsperpacket);
186 #define	V_maxfragsperpacket	VNET(maxfragsperpacket)
187 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
188     &VNET_NAME(maxfragsperpacket), 0,
189     "Maximum number of IPv4 fragments allowed per packet");
190 
191 #ifdef IPCTL_DEFMTU
192 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
193     &ip_mtu, 0, "Default MTU");
194 #endif
195 
196 #ifdef IPSTEALTH
197 VNET_DEFINE(int, ipstealth);
198 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
199     &VNET_NAME(ipstealth), 0,
200     "IP stealth mode, no TTL decrementation on forwarding");
201 #endif
202 
203 #ifdef FLOWTABLE
204 static VNET_DEFINE(int, ip_output_flowtable_size) = 2048;
205 VNET_DEFINE(struct flowtable *, ip_ft);
206 #define	V_ip_output_flowtable_size	VNET(ip_output_flowtable_size)
207 
208 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
209     &VNET_NAME(ip_output_flowtable_size), 2048,
210     "number of entries in the per-cpu output flow caches");
211 #endif
212 
213 static void	ip_freef(struct ipqhead *, struct ipq *);
214 
215 /*
216  * Kernel module interface for updating ipstat.  The argument is an index
217  * into ipstat treated as an array of u_long.  While this encodes the general
218  * layout of ipstat into the caller, it doesn't encode its location, so that
219  * future changes to add, for example, per-CPU stats support won't cause
220  * binary compatibility problems for kernel modules.
221  */
222 void
223 kmod_ipstat_inc(int statnum)
224 {
225 
226 	(*((u_long *)&V_ipstat + statnum))++;
227 }
228 
229 void
230 kmod_ipstat_dec(int statnum)
231 {
232 
233 	(*((u_long *)&V_ipstat + statnum))--;
234 }
235 
236 static int
237 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
238 {
239 	int error, qlimit;
240 
241 	netisr_getqlimit(&ip_nh, &qlimit);
242 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
243 	if (error || !req->newptr)
244 		return (error);
245 	if (qlimit < 1)
246 		return (EINVAL);
247 	return (netisr_setqlimit(&ip_nh, qlimit));
248 }
249 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
250     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
251     "Maximum size of the IP input queue");
252 
253 static int
254 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
255 {
256 	u_int64_t qdrops_long;
257 	int error, qdrops;
258 
259 	netisr_getqdrops(&ip_nh, &qdrops_long);
260 	qdrops = qdrops_long;
261 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
262 	if (error || !req->newptr)
263 		return (error);
264 	if (qdrops != 0)
265 		return (EINVAL);
266 	netisr_clearqdrops(&ip_nh);
267 	return (0);
268 }
269 
270 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
271     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
272     "Number of packets dropped from the IP input queue");
273 
274 /*
275  * IP initialization: fill in IP protocol switch table.
276  * All protocols not implemented in kernel go to raw IP protocol handler.
277  */
278 void
279 ip_init(void)
280 {
281 	struct protosw *pr;
282 	int i;
283 
284 	V_ip_id = time_second & 0xffff;
285 
286 	TAILQ_INIT(&V_in_ifaddrhead);
287 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
288 
289 	/* Initialize IP reassembly queue. */
290 	for (i = 0; i < IPREASS_NHASH; i++)
291 		TAILQ_INIT(&V_ipq[i]);
292 	V_maxnipq = nmbclusters / 32;
293 	V_maxfragsperpacket = 16;
294 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
295 	    NULL, UMA_ALIGN_PTR, 0);
296 	maxnipq_update();
297 
298 	/* Initialize packet filter hooks. */
299 	V_inet_pfil_hook.ph_type = PFIL_TYPE_AF;
300 	V_inet_pfil_hook.ph_af = AF_INET;
301 	if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0)
302 		printf("%s: WARNING: unable to register pfil hook, "
303 			"error %d\n", __func__, i);
304 
305 #ifdef FLOWTABLE
306 	if (TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size",
307 		&V_ip_output_flowtable_size)) {
308 		if (V_ip_output_flowtable_size < 256)
309 			V_ip_output_flowtable_size = 256;
310 		if (!powerof2(V_ip_output_flowtable_size)) {
311 			printf("flowtable must be power of 2 size\n");
312 			V_ip_output_flowtable_size = 2048;
313 		}
314 	} else {
315 		/*
316 		 * round up to the next power of 2
317 		 */
318 		V_ip_output_flowtable_size = 1 << fls((1024 + maxusers * 64)-1);
319 	}
320 	V_ip_ft = flowtable_alloc("ipv4", V_ip_output_flowtable_size, FL_PCPU);
321 #endif
322 
323 	/* Skip initialization of globals for non-default instances. */
324 	if (!IS_DEFAULT_VNET(curvnet))
325 		return;
326 
327 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
328 	if (pr == NULL)
329 		panic("ip_init: PF_INET not found");
330 
331 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
332 	for (i = 0; i < IPPROTO_MAX; i++)
333 		ip_protox[i] = pr - inetsw;
334 	/*
335 	 * Cycle through IP protocols and put them into the appropriate place
336 	 * in ip_protox[].
337 	 */
338 	for (pr = inetdomain.dom_protosw;
339 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
340 		if (pr->pr_domain->dom_family == PF_INET &&
341 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
342 			/* Be careful to only index valid IP protocols. */
343 			if (pr->pr_protocol < IPPROTO_MAX)
344 				ip_protox[pr->pr_protocol] = pr - inetsw;
345 		}
346 
347 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
348 		NULL, EVENTHANDLER_PRI_ANY);
349 
350 	/* Initialize various other remaining things. */
351 	IPQ_LOCK_INIT();
352 	netisr_register(&ip_nh);
353 }
354 
355 #ifdef VIMAGE
356 void
357 ip_destroy(void)
358 {
359 
360 	/* Cleanup in_ifaddr hash table; should be empty. */
361 	hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask);
362 
363 	IPQ_LOCK();
364 	ip_drain_locked();
365 	IPQ_UNLOCK();
366 
367 	uma_zdestroy(V_ipq_zone);
368 }
369 #endif
370 
371 /*
372  * Ip input routine.  Checksum and byte swap header.  If fragmented
373  * try to reassemble.  Process options.  Pass to next level.
374  */
375 void
376 ip_input(struct mbuf *m)
377 {
378 	struct ip *ip = NULL;
379 	struct in_ifaddr *ia = NULL;
380 	struct ifaddr *ifa;
381 	struct ifnet *ifp;
382 	int    checkif, hlen = 0;
383 	uint16_t sum, ip_len;
384 	int dchg = 0;				/* dest changed after fw */
385 	struct in_addr odst;			/* original dst address */
386 
387 	M_ASSERTPKTHDR(m);
388 
389 	if (m->m_flags & M_FASTFWD_OURS) {
390 		m->m_flags &= ~M_FASTFWD_OURS;
391 		/* Set up some basics that will be used later. */
392 		ip = mtod(m, struct ip *);
393 		hlen = ip->ip_hl << 2;
394 		ip_len = ntohs(ip->ip_len);
395 		goto ours;
396 	}
397 
398 	IPSTAT_INC(ips_total);
399 
400 	if (m->m_pkthdr.len < sizeof(struct ip))
401 		goto tooshort;
402 
403 	if (m->m_len < sizeof (struct ip) &&
404 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
405 		IPSTAT_INC(ips_toosmall);
406 		return;
407 	}
408 	ip = mtod(m, struct ip *);
409 
410 	if (ip->ip_v != IPVERSION) {
411 		IPSTAT_INC(ips_badvers);
412 		goto bad;
413 	}
414 
415 	hlen = ip->ip_hl << 2;
416 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
417 		IPSTAT_INC(ips_badhlen);
418 		goto bad;
419 	}
420 	if (hlen > m->m_len) {
421 		if ((m = m_pullup(m, hlen)) == NULL) {
422 			IPSTAT_INC(ips_badhlen);
423 			return;
424 		}
425 		ip = mtod(m, struct ip *);
426 	}
427 
428 	/* 127/8 must not appear on wire - RFC1122 */
429 	ifp = m->m_pkthdr.rcvif;
430 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
431 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
432 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
433 			IPSTAT_INC(ips_badaddr);
434 			goto bad;
435 		}
436 	}
437 
438 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
439 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
440 	} else {
441 		if (hlen == sizeof(struct ip)) {
442 			sum = in_cksum_hdr(ip);
443 		} else {
444 			sum = in_cksum(m, hlen);
445 		}
446 	}
447 	if (sum) {
448 		IPSTAT_INC(ips_badsum);
449 		goto bad;
450 	}
451 
452 #ifdef ALTQ
453 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
454 		/* packet is dropped by traffic conditioner */
455 		return;
456 #endif
457 
458 	ip_len = ntohs(ip->ip_len);
459 	if (ip_len < hlen) {
460 		IPSTAT_INC(ips_badlen);
461 		goto bad;
462 	}
463 
464 	/*
465 	 * Check that the amount of data in the buffers
466 	 * is as at least much as the IP header would have us expect.
467 	 * Trim mbufs if longer than we expect.
468 	 * Drop packet if shorter than we expect.
469 	 */
470 	if (m->m_pkthdr.len < ip_len) {
471 tooshort:
472 		IPSTAT_INC(ips_tooshort);
473 		goto bad;
474 	}
475 	if (m->m_pkthdr.len > ip_len) {
476 		if (m->m_len == m->m_pkthdr.len) {
477 			m->m_len = ip_len;
478 			m->m_pkthdr.len = ip_len;
479 		} else
480 			m_adj(m, ip_len - m->m_pkthdr.len);
481 	}
482 #ifdef IPSEC
483 	/*
484 	 * Bypass packet filtering for packets previously handled by IPsec.
485 	 */
486 	if (ip_ipsec_filtertunnel(m))
487 		goto passin;
488 #endif /* IPSEC */
489 
490 	/*
491 	 * Run through list of hooks for input packets.
492 	 *
493 	 * NB: Beware of the destination address changing (e.g.
494 	 *     by NAT rewriting).  When this happens, tell
495 	 *     ip_forward to do the right thing.
496 	 */
497 
498 	/* Jump over all PFIL processing if hooks are not active. */
499 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
500 		goto passin;
501 
502 	odst = ip->ip_dst;
503 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
504 		return;
505 	if (m == NULL)			/* consumed by filter */
506 		return;
507 
508 	ip = mtod(m, struct ip *);
509 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
510 	ifp = m->m_pkthdr.rcvif;
511 
512 	if (m->m_flags & M_FASTFWD_OURS) {
513 		m->m_flags &= ~M_FASTFWD_OURS;
514 		goto ours;
515 	}
516 	if (m->m_flags & M_IP_NEXTHOP) {
517 		dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL);
518 		if (dchg != 0) {
519 			/*
520 			 * Directly ship the packet on.  This allows
521 			 * forwarding packets originally destined to us
522 			 * to some other directly connected host.
523 			 */
524 			ip_forward(m, 1);
525 			return;
526 		}
527 	}
528 passin:
529 
530 	/*
531 	 * Process options and, if not destined for us,
532 	 * ship it on.  ip_dooptions returns 1 when an
533 	 * error was detected (causing an icmp message
534 	 * to be sent and the original packet to be freed).
535 	 */
536 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
537 		return;
538 
539         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
540          * matter if it is destined to another node, or whether it is
541          * a multicast one, RSVP wants it! and prevents it from being forwarded
542          * anywhere else. Also checks if the rsvp daemon is running before
543 	 * grabbing the packet.
544          */
545 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
546 		goto ours;
547 
548 	/*
549 	 * Check our list of addresses, to see if the packet is for us.
550 	 * If we don't have any addresses, assume any unicast packet
551 	 * we receive might be for us (and let the upper layers deal
552 	 * with it).
553 	 */
554 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
555 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
556 		goto ours;
557 
558 	/*
559 	 * Enable a consistency check between the destination address
560 	 * and the arrival interface for a unicast packet (the RFC 1122
561 	 * strong ES model) if IP forwarding is disabled and the packet
562 	 * is not locally generated and the packet is not subject to
563 	 * 'ipfw fwd'.
564 	 *
565 	 * XXX - Checking also should be disabled if the destination
566 	 * address is ipnat'ed to a different interface.
567 	 *
568 	 * XXX - Checking is incompatible with IP aliases added
569 	 * to the loopback interface instead of the interface where
570 	 * the packets are received.
571 	 *
572 	 * XXX - This is the case for carp vhost IPs as well so we
573 	 * insert a workaround. If the packet got here, we already
574 	 * checked with carp_iamatch() and carp_forus().
575 	 */
576 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
577 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
578 	    ifp->if_carp == NULL && (dchg == 0);
579 
580 	/*
581 	 * Check for exact addresses in the hash bucket.
582 	 */
583 	/* IN_IFADDR_RLOCK(); */
584 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
585 		/*
586 		 * If the address matches, verify that the packet
587 		 * arrived via the correct interface if checking is
588 		 * enabled.
589 		 */
590 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
591 		    (!checkif || ia->ia_ifp == ifp)) {
592 			ifa_ref(&ia->ia_ifa);
593 			/* IN_IFADDR_RUNLOCK(); */
594 			goto ours;
595 		}
596 	}
597 	/* IN_IFADDR_RUNLOCK(); */
598 
599 	/*
600 	 * Check for broadcast addresses.
601 	 *
602 	 * Only accept broadcast packets that arrive via the matching
603 	 * interface.  Reception of forwarded directed broadcasts would
604 	 * be handled via ip_forward() and ether_output() with the loopback
605 	 * into the stack for SIMPLEX interfaces handled by ether_output().
606 	 */
607 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
608 		IF_ADDR_RLOCK(ifp);
609 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
610 			if (ifa->ifa_addr->sa_family != AF_INET)
611 				continue;
612 			ia = ifatoia(ifa);
613 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
614 			    ip->ip_dst.s_addr) {
615 				ifa_ref(ifa);
616 				IF_ADDR_RUNLOCK(ifp);
617 				goto ours;
618 			}
619 #ifdef BOOTP_COMPAT
620 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
621 				ifa_ref(ifa);
622 				IF_ADDR_RUNLOCK(ifp);
623 				goto ours;
624 			}
625 #endif
626 		}
627 		IF_ADDR_RUNLOCK(ifp);
628 		ia = NULL;
629 	}
630 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
631 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
632 		IPSTAT_INC(ips_cantforward);
633 		m_freem(m);
634 		return;
635 	}
636 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
637 		if (V_ip_mrouter) {
638 			/*
639 			 * If we are acting as a multicast router, all
640 			 * incoming multicast packets are passed to the
641 			 * kernel-level multicast forwarding function.
642 			 * The packet is returned (relatively) intact; if
643 			 * ip_mforward() returns a non-zero value, the packet
644 			 * must be discarded, else it may be accepted below.
645 			 */
646 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
647 				IPSTAT_INC(ips_cantforward);
648 				m_freem(m);
649 				return;
650 			}
651 
652 			/*
653 			 * The process-level routing daemon needs to receive
654 			 * all multicast IGMP packets, whether or not this
655 			 * host belongs to their destination groups.
656 			 */
657 			if (ip->ip_p == IPPROTO_IGMP)
658 				goto ours;
659 			IPSTAT_INC(ips_forward);
660 		}
661 		/*
662 		 * Assume the packet is for us, to avoid prematurely taking
663 		 * a lock on the in_multi hash. Protocols must perform
664 		 * their own filtering and update statistics accordingly.
665 		 */
666 		goto ours;
667 	}
668 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
669 		goto ours;
670 	if (ip->ip_dst.s_addr == INADDR_ANY)
671 		goto ours;
672 
673 	/*
674 	 * FAITH(Firewall Aided Internet Translator)
675 	 */
676 	if (ifp && ifp->if_type == IFT_FAITH) {
677 		if (V_ip_keepfaith) {
678 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
679 				goto ours;
680 		}
681 		m_freem(m);
682 		return;
683 	}
684 
685 	/*
686 	 * Not for us; forward if possible and desirable.
687 	 */
688 	if (V_ipforwarding == 0) {
689 		IPSTAT_INC(ips_cantforward);
690 		m_freem(m);
691 	} else {
692 #ifdef IPSEC
693 		if (ip_ipsec_fwd(m))
694 			goto bad;
695 #endif /* IPSEC */
696 		ip_forward(m, dchg);
697 	}
698 	return;
699 
700 ours:
701 #ifdef IPSTEALTH
702 	/*
703 	 * IPSTEALTH: Process non-routing options only
704 	 * if the packet is destined for us.
705 	 */
706 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) {
707 		if (ia != NULL)
708 			ifa_free(&ia->ia_ifa);
709 		return;
710 	}
711 #endif /* IPSTEALTH */
712 
713 	/* Count the packet in the ip address stats */
714 	if (ia != NULL) {
715 		ia->ia_ifa.if_ipackets++;
716 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
717 		ifa_free(&ia->ia_ifa);
718 	}
719 
720 	/*
721 	 * Attempt reassembly; if it succeeds, proceed.
722 	 * ip_reass() will return a different mbuf.
723 	 */
724 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
725 		m = ip_reass(m);
726 		if (m == NULL)
727 			return;
728 		ip = mtod(m, struct ip *);
729 		/* Get the header length of the reassembled packet */
730 		hlen = ip->ip_hl << 2;
731 	}
732 
733 #ifdef IPSEC
734 	/*
735 	 * enforce IPsec policy checking if we are seeing last header.
736 	 * note that we do not visit this with protocols with pcb layer
737 	 * code - like udp/tcp/raw ip.
738 	 */
739 	if (ip_ipsec_input(m))
740 		goto bad;
741 #endif /* IPSEC */
742 
743 	/*
744 	 * Switch out to protocol's input routine.
745 	 */
746 	IPSTAT_INC(ips_delivered);
747 
748 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
749 	return;
750 bad:
751 	m_freem(m);
752 }
753 
754 /*
755  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
756  * max has slightly different semantics than the sysctl, for historical
757  * reasons.
758  */
759 static void
760 maxnipq_update(void)
761 {
762 
763 	/*
764 	 * -1 for unlimited allocation.
765 	 */
766 	if (V_maxnipq < 0)
767 		uma_zone_set_max(V_ipq_zone, 0);
768 	/*
769 	 * Positive number for specific bound.
770 	 */
771 	if (V_maxnipq > 0)
772 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
773 	/*
774 	 * Zero specifies no further fragment queue allocation -- set the
775 	 * bound very low, but rely on implementation elsewhere to actually
776 	 * prevent allocation and reclaim current queues.
777 	 */
778 	if (V_maxnipq == 0)
779 		uma_zone_set_max(V_ipq_zone, 1);
780 }
781 
782 static void
783 ipq_zone_change(void *tag)
784 {
785 
786 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
787 		V_maxnipq = nmbclusters / 32;
788 		maxnipq_update();
789 	}
790 }
791 
792 static int
793 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
794 {
795 	int error, i;
796 
797 	i = V_maxnipq;
798 	error = sysctl_handle_int(oidp, &i, 0, req);
799 	if (error || !req->newptr)
800 		return (error);
801 
802 	/*
803 	 * XXXRW: Might be a good idea to sanity check the argument and place
804 	 * an extreme upper bound.
805 	 */
806 	if (i < -1)
807 		return (EINVAL);
808 	V_maxnipq = i;
809 	maxnipq_update();
810 	return (0);
811 }
812 
813 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
814     NULL, 0, sysctl_maxnipq, "I",
815     "Maximum number of IPv4 fragment reassembly queue entries");
816 
817 /*
818  * Take incoming datagram fragment and try to reassemble it into
819  * whole datagram.  If the argument is the first fragment or one
820  * in between the function will return NULL and store the mbuf
821  * in the fragment chain.  If the argument is the last fragment
822  * the packet will be reassembled and the pointer to the new
823  * mbuf returned for further processing.  Only m_tags attached
824  * to the first packet/fragment are preserved.
825  * The IP header is *NOT* adjusted out of iplen.
826  */
827 struct mbuf *
828 ip_reass(struct mbuf *m)
829 {
830 	struct ip *ip;
831 	struct mbuf *p, *q, *nq, *t;
832 	struct ipq *fp = NULL;
833 	struct ipqhead *head;
834 	int i, hlen, next;
835 	u_int8_t ecn, ecn0;
836 	u_short hash;
837 
838 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
839 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
840 		IPSTAT_INC(ips_fragments);
841 		IPSTAT_INC(ips_fragdropped);
842 		m_freem(m);
843 		return (NULL);
844 	}
845 
846 	ip = mtod(m, struct ip *);
847 	hlen = ip->ip_hl << 2;
848 
849 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
850 	head = &V_ipq[hash];
851 	IPQ_LOCK();
852 
853 	/*
854 	 * Look for queue of fragments
855 	 * of this datagram.
856 	 */
857 	TAILQ_FOREACH(fp, head, ipq_list)
858 		if (ip->ip_id == fp->ipq_id &&
859 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
860 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
861 #ifdef MAC
862 		    mac_ipq_match(m, fp) &&
863 #endif
864 		    ip->ip_p == fp->ipq_p)
865 			goto found;
866 
867 	fp = NULL;
868 
869 	/*
870 	 * Attempt to trim the number of allocated fragment queues if it
871 	 * exceeds the administrative limit.
872 	 */
873 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
874 		/*
875 		 * drop something from the tail of the current queue
876 		 * before proceeding further
877 		 */
878 		struct ipq *q = TAILQ_LAST(head, ipqhead);
879 		if (q == NULL) {   /* gak */
880 			for (i = 0; i < IPREASS_NHASH; i++) {
881 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
882 				if (r) {
883 					IPSTAT_ADD(ips_fragtimeout,
884 					    r->ipq_nfrags);
885 					ip_freef(&V_ipq[i], r);
886 					break;
887 				}
888 			}
889 		} else {
890 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
891 			ip_freef(head, q);
892 		}
893 	}
894 
895 found:
896 	/*
897 	 * Adjust ip_len to not reflect header,
898 	 * convert offset of this to bytes.
899 	 */
900 	ip->ip_len = htons(ntohs(ip->ip_len) - hlen);
901 	if (ip->ip_off & htons(IP_MF)) {
902 		/*
903 		 * Make sure that fragments have a data length
904 		 * that's a non-zero multiple of 8 bytes.
905 		 */
906 		if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) {
907 			IPSTAT_INC(ips_toosmall); /* XXX */
908 			goto dropfrag;
909 		}
910 		m->m_flags |= M_FRAG;
911 	} else
912 		m->m_flags &= ~M_FRAG;
913 	ip->ip_off = htons(ntohs(ip->ip_off) << 3);
914 
915 	/*
916 	 * Attempt reassembly; if it succeeds, proceed.
917 	 * ip_reass() will return a different mbuf.
918 	 */
919 	IPSTAT_INC(ips_fragments);
920 	m->m_pkthdr.header = ip;
921 
922 	/* Previous ip_reass() started here. */
923 	/*
924 	 * Presence of header sizes in mbufs
925 	 * would confuse code below.
926 	 */
927 	m->m_data += hlen;
928 	m->m_len -= hlen;
929 
930 	/*
931 	 * If first fragment to arrive, create a reassembly queue.
932 	 */
933 	if (fp == NULL) {
934 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
935 		if (fp == NULL)
936 			goto dropfrag;
937 #ifdef MAC
938 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
939 			uma_zfree(V_ipq_zone, fp);
940 			fp = NULL;
941 			goto dropfrag;
942 		}
943 		mac_ipq_create(m, fp);
944 #endif
945 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
946 		V_nipq++;
947 		fp->ipq_nfrags = 1;
948 		fp->ipq_ttl = IPFRAGTTL;
949 		fp->ipq_p = ip->ip_p;
950 		fp->ipq_id = ip->ip_id;
951 		fp->ipq_src = ip->ip_src;
952 		fp->ipq_dst = ip->ip_dst;
953 		fp->ipq_frags = m;
954 		m->m_nextpkt = NULL;
955 		goto done;
956 	} else {
957 		fp->ipq_nfrags++;
958 #ifdef MAC
959 		mac_ipq_update(m, fp);
960 #endif
961 	}
962 
963 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
964 
965 	/*
966 	 * Handle ECN by comparing this segment with the first one;
967 	 * if CE is set, do not lose CE.
968 	 * drop if CE and not-ECT are mixed for the same packet.
969 	 */
970 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
971 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
972 	if (ecn == IPTOS_ECN_CE) {
973 		if (ecn0 == IPTOS_ECN_NOTECT)
974 			goto dropfrag;
975 		if (ecn0 != IPTOS_ECN_CE)
976 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
977 	}
978 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
979 		goto dropfrag;
980 
981 	/*
982 	 * Find a segment which begins after this one does.
983 	 */
984 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
985 		if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off))
986 			break;
987 
988 	/*
989 	 * If there is a preceding segment, it may provide some of
990 	 * our data already.  If so, drop the data from the incoming
991 	 * segment.  If it provides all of our data, drop us, otherwise
992 	 * stick new segment in the proper place.
993 	 *
994 	 * If some of the data is dropped from the preceding
995 	 * segment, then it's checksum is invalidated.
996 	 */
997 	if (p) {
998 		i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) -
999 		    ntohs(ip->ip_off);
1000 		if (i > 0) {
1001 			if (i >= ntohs(ip->ip_len))
1002 				goto dropfrag;
1003 			m_adj(m, i);
1004 			m->m_pkthdr.csum_flags = 0;
1005 			ip->ip_off = htons(ntohs(ip->ip_off) + i);
1006 			ip->ip_len = htons(ntohs(ip->ip_len) - i);
1007 		}
1008 		m->m_nextpkt = p->m_nextpkt;
1009 		p->m_nextpkt = m;
1010 	} else {
1011 		m->m_nextpkt = fp->ipq_frags;
1012 		fp->ipq_frags = m;
1013 	}
1014 
1015 	/*
1016 	 * While we overlap succeeding segments trim them or,
1017 	 * if they are completely covered, dequeue them.
1018 	 */
1019 	for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) >
1020 	    ntohs(GETIP(q)->ip_off); q = nq) {
1021 		i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) -
1022 		    ntohs(GETIP(q)->ip_off);
1023 		if (i < ntohs(GETIP(q)->ip_len)) {
1024 			GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i);
1025 			GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i);
1026 			m_adj(q, i);
1027 			q->m_pkthdr.csum_flags = 0;
1028 			break;
1029 		}
1030 		nq = q->m_nextpkt;
1031 		m->m_nextpkt = nq;
1032 		IPSTAT_INC(ips_fragdropped);
1033 		fp->ipq_nfrags--;
1034 		m_freem(q);
1035 	}
1036 
1037 	/*
1038 	 * Check for complete reassembly and perform frag per packet
1039 	 * limiting.
1040 	 *
1041 	 * Frag limiting is performed here so that the nth frag has
1042 	 * a chance to complete the packet before we drop the packet.
1043 	 * As a result, n+1 frags are actually allowed per packet, but
1044 	 * only n will ever be stored. (n = maxfragsperpacket.)
1045 	 *
1046 	 */
1047 	next = 0;
1048 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1049 		if (ntohs(GETIP(q)->ip_off) != next) {
1050 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1051 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1052 				ip_freef(head, fp);
1053 			}
1054 			goto done;
1055 		}
1056 		next += ntohs(GETIP(q)->ip_len);
1057 	}
1058 	/* Make sure the last packet didn't have the IP_MF flag */
1059 	if (p->m_flags & M_FRAG) {
1060 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1061 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1062 			ip_freef(head, fp);
1063 		}
1064 		goto done;
1065 	}
1066 
1067 	/*
1068 	 * Reassembly is complete.  Make sure the packet is a sane size.
1069 	 */
1070 	q = fp->ipq_frags;
1071 	ip = GETIP(q);
1072 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1073 		IPSTAT_INC(ips_toolong);
1074 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1075 		ip_freef(head, fp);
1076 		goto done;
1077 	}
1078 
1079 	/*
1080 	 * Concatenate fragments.
1081 	 */
1082 	m = q;
1083 	t = m->m_next;
1084 	m->m_next = NULL;
1085 	m_cat(m, t);
1086 	nq = q->m_nextpkt;
1087 	q->m_nextpkt = NULL;
1088 	for (q = nq; q != NULL; q = nq) {
1089 		nq = q->m_nextpkt;
1090 		q->m_nextpkt = NULL;
1091 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1092 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1093 		m_cat(m, q);
1094 	}
1095 	/*
1096 	 * In order to do checksumming faster we do 'end-around carry' here
1097 	 * (and not in for{} loop), though it implies we are not going to
1098 	 * reassemble more than 64k fragments.
1099 	 */
1100 	m->m_pkthdr.csum_data =
1101 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1102 #ifdef MAC
1103 	mac_ipq_reassemble(fp, m);
1104 	mac_ipq_destroy(fp);
1105 #endif
1106 
1107 	/*
1108 	 * Create header for new ip packet by modifying header of first
1109 	 * packet;  dequeue and discard fragment reassembly header.
1110 	 * Make header visible.
1111 	 */
1112 	ip->ip_len = htons((ip->ip_hl << 2) + next);
1113 	ip->ip_src = fp->ipq_src;
1114 	ip->ip_dst = fp->ipq_dst;
1115 	TAILQ_REMOVE(head, fp, ipq_list);
1116 	V_nipq--;
1117 	uma_zfree(V_ipq_zone, fp);
1118 	m->m_len += (ip->ip_hl << 2);
1119 	m->m_data -= (ip->ip_hl << 2);
1120 	/* some debugging cruft by sklower, below, will go away soon */
1121 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1122 		m_fixhdr(m);
1123 	IPSTAT_INC(ips_reassembled);
1124 	IPQ_UNLOCK();
1125 	return (m);
1126 
1127 dropfrag:
1128 	IPSTAT_INC(ips_fragdropped);
1129 	if (fp != NULL)
1130 		fp->ipq_nfrags--;
1131 	m_freem(m);
1132 done:
1133 	IPQ_UNLOCK();
1134 	return (NULL);
1135 
1136 #undef GETIP
1137 }
1138 
1139 /*
1140  * Free a fragment reassembly header and all
1141  * associated datagrams.
1142  */
1143 static void
1144 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1145 {
1146 	struct mbuf *q;
1147 
1148 	IPQ_LOCK_ASSERT();
1149 
1150 	while (fp->ipq_frags) {
1151 		q = fp->ipq_frags;
1152 		fp->ipq_frags = q->m_nextpkt;
1153 		m_freem(q);
1154 	}
1155 	TAILQ_REMOVE(fhp, fp, ipq_list);
1156 	uma_zfree(V_ipq_zone, fp);
1157 	V_nipq--;
1158 }
1159 
1160 /*
1161  * IP timer processing;
1162  * if a timer expires on a reassembly
1163  * queue, discard it.
1164  */
1165 void
1166 ip_slowtimo(void)
1167 {
1168 	VNET_ITERATOR_DECL(vnet_iter);
1169 	struct ipq *fp;
1170 	int i;
1171 
1172 	VNET_LIST_RLOCK_NOSLEEP();
1173 	IPQ_LOCK();
1174 	VNET_FOREACH(vnet_iter) {
1175 		CURVNET_SET(vnet_iter);
1176 		for (i = 0; i < IPREASS_NHASH; i++) {
1177 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1178 				struct ipq *fpp;
1179 
1180 				fpp = fp;
1181 				fp = TAILQ_NEXT(fp, ipq_list);
1182 				if(--fpp->ipq_ttl == 0) {
1183 					IPSTAT_ADD(ips_fragtimeout,
1184 					    fpp->ipq_nfrags);
1185 					ip_freef(&V_ipq[i], fpp);
1186 				}
1187 			}
1188 		}
1189 		/*
1190 		 * If we are over the maximum number of fragments
1191 		 * (due to the limit being lowered), drain off
1192 		 * enough to get down to the new limit.
1193 		 */
1194 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1195 			for (i = 0; i < IPREASS_NHASH; i++) {
1196 				while (V_nipq > V_maxnipq &&
1197 				    !TAILQ_EMPTY(&V_ipq[i])) {
1198 					IPSTAT_ADD(ips_fragdropped,
1199 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1200 					ip_freef(&V_ipq[i],
1201 					    TAILQ_FIRST(&V_ipq[i]));
1202 				}
1203 			}
1204 		}
1205 		CURVNET_RESTORE();
1206 	}
1207 	IPQ_UNLOCK();
1208 	VNET_LIST_RUNLOCK_NOSLEEP();
1209 }
1210 
1211 /*
1212  * Drain off all datagram fragments.
1213  */
1214 static void
1215 ip_drain_locked(void)
1216 {
1217 	int     i;
1218 
1219 	IPQ_LOCK_ASSERT();
1220 
1221 	for (i = 0; i < IPREASS_NHASH; i++) {
1222 		while(!TAILQ_EMPTY(&V_ipq[i])) {
1223 			IPSTAT_ADD(ips_fragdropped,
1224 			    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1225 			ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1226 		}
1227 	}
1228 }
1229 
1230 void
1231 ip_drain(void)
1232 {
1233 	VNET_ITERATOR_DECL(vnet_iter);
1234 
1235 	VNET_LIST_RLOCK_NOSLEEP();
1236 	IPQ_LOCK();
1237 	VNET_FOREACH(vnet_iter) {
1238 		CURVNET_SET(vnet_iter);
1239 		ip_drain_locked();
1240 		CURVNET_RESTORE();
1241 	}
1242 	IPQ_UNLOCK();
1243 	VNET_LIST_RUNLOCK_NOSLEEP();
1244 	in_rtqdrain();
1245 }
1246 
1247 /*
1248  * The protocol to be inserted into ip_protox[] must be already registered
1249  * in inetsw[], either statically or through pf_proto_register().
1250  */
1251 int
1252 ipproto_register(short ipproto)
1253 {
1254 	struct protosw *pr;
1255 
1256 	/* Sanity checks. */
1257 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1258 		return (EPROTONOSUPPORT);
1259 
1260 	/*
1261 	 * The protocol slot must not be occupied by another protocol
1262 	 * already.  An index pointing to IPPROTO_RAW is unused.
1263 	 */
1264 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1265 	if (pr == NULL)
1266 		return (EPFNOSUPPORT);
1267 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1268 		return (EEXIST);
1269 
1270 	/* Find the protocol position in inetsw[] and set the index. */
1271 	for (pr = inetdomain.dom_protosw;
1272 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1273 		if (pr->pr_domain->dom_family == PF_INET &&
1274 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1275 			ip_protox[pr->pr_protocol] = pr - inetsw;
1276 			return (0);
1277 		}
1278 	}
1279 	return (EPROTONOSUPPORT);
1280 }
1281 
1282 int
1283 ipproto_unregister(short ipproto)
1284 {
1285 	struct protosw *pr;
1286 
1287 	/* Sanity checks. */
1288 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1289 		return (EPROTONOSUPPORT);
1290 
1291 	/* Check if the protocol was indeed registered. */
1292 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1293 	if (pr == NULL)
1294 		return (EPFNOSUPPORT);
1295 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1296 		return (ENOENT);
1297 
1298 	/* Reset the protocol slot to IPPROTO_RAW. */
1299 	ip_protox[ipproto] = pr - inetsw;
1300 	return (0);
1301 }
1302 
1303 /*
1304  * Given address of next destination (final or next hop), return (referenced)
1305  * internet address info of interface to be used to get there.
1306  */
1307 struct in_ifaddr *
1308 ip_rtaddr(struct in_addr dst, u_int fibnum)
1309 {
1310 	struct route sro;
1311 	struct sockaddr_in *sin;
1312 	struct in_ifaddr *ia;
1313 
1314 	bzero(&sro, sizeof(sro));
1315 	sin = (struct sockaddr_in *)&sro.ro_dst;
1316 	sin->sin_family = AF_INET;
1317 	sin->sin_len = sizeof(*sin);
1318 	sin->sin_addr = dst;
1319 	in_rtalloc_ign(&sro, 0, fibnum);
1320 
1321 	if (sro.ro_rt == NULL)
1322 		return (NULL);
1323 
1324 	ia = ifatoia(sro.ro_rt->rt_ifa);
1325 	ifa_ref(&ia->ia_ifa);
1326 	RTFREE(sro.ro_rt);
1327 	return (ia);
1328 }
1329 
1330 u_char inetctlerrmap[PRC_NCMDS] = {
1331 	0,		0,		0,		0,
1332 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1333 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1334 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1335 	0,		0,		EHOSTUNREACH,	0,
1336 	ENOPROTOOPT,	ECONNREFUSED
1337 };
1338 
1339 /*
1340  * Forward a packet.  If some error occurs return the sender
1341  * an icmp packet.  Note we can't always generate a meaningful
1342  * icmp message because icmp doesn't have a large enough repertoire
1343  * of codes and types.
1344  *
1345  * If not forwarding, just drop the packet.  This could be confusing
1346  * if ipforwarding was zero but some routing protocol was advancing
1347  * us as a gateway to somewhere.  However, we must let the routing
1348  * protocol deal with that.
1349  *
1350  * The srcrt parameter indicates whether the packet is being forwarded
1351  * via a source route.
1352  */
1353 void
1354 ip_forward(struct mbuf *m, int srcrt)
1355 {
1356 	struct ip *ip = mtod(m, struct ip *);
1357 	struct in_ifaddr *ia;
1358 	struct mbuf *mcopy;
1359 	struct in_addr dest;
1360 	struct route ro;
1361 	int error, type = 0, code = 0, mtu = 0;
1362 
1363 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1364 		IPSTAT_INC(ips_cantforward);
1365 		m_freem(m);
1366 		return;
1367 	}
1368 #ifdef IPSTEALTH
1369 	if (!V_ipstealth) {
1370 #endif
1371 		if (ip->ip_ttl <= IPTTLDEC) {
1372 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1373 			    0, 0);
1374 			return;
1375 		}
1376 #ifdef IPSTEALTH
1377 	}
1378 #endif
1379 
1380 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1381 #ifndef IPSEC
1382 	/*
1383 	 * 'ia' may be NULL if there is no route for this destination.
1384 	 * In case of IPsec, Don't discard it just yet, but pass it to
1385 	 * ip_output in case of outgoing IPsec policy.
1386 	 */
1387 	if (!srcrt && ia == NULL) {
1388 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1389 		return;
1390 	}
1391 #endif
1392 
1393 	/*
1394 	 * Save the IP header and at most 8 bytes of the payload,
1395 	 * in case we need to generate an ICMP message to the src.
1396 	 *
1397 	 * XXX this can be optimized a lot by saving the data in a local
1398 	 * buffer on the stack (72 bytes at most), and only allocating the
1399 	 * mbuf if really necessary. The vast majority of the packets
1400 	 * are forwarded without having to send an ICMP back (either
1401 	 * because unnecessary, or because rate limited), so we are
1402 	 * really we are wasting a lot of work here.
1403 	 *
1404 	 * We don't use m_copy() because it might return a reference
1405 	 * to a shared cluster. Both this function and ip_output()
1406 	 * assume exclusive access to the IP header in `m', so any
1407 	 * data in a cluster may change before we reach icmp_error().
1408 	 */
1409 	MGETHDR(mcopy, M_NOWAIT, m->m_type);
1410 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) {
1411 		/*
1412 		 * It's probably ok if the pkthdr dup fails (because
1413 		 * the deep copy of the tag chain failed), but for now
1414 		 * be conservative and just discard the copy since
1415 		 * code below may some day want the tags.
1416 		 */
1417 		m_free(mcopy);
1418 		mcopy = NULL;
1419 	}
1420 	if (mcopy != NULL) {
1421 		mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy));
1422 		mcopy->m_pkthdr.len = mcopy->m_len;
1423 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1424 	}
1425 
1426 #ifdef IPSTEALTH
1427 	if (!V_ipstealth) {
1428 #endif
1429 		ip->ip_ttl -= IPTTLDEC;
1430 #ifdef IPSTEALTH
1431 	}
1432 #endif
1433 
1434 	/*
1435 	 * If forwarding packet using same interface that it came in on,
1436 	 * perhaps should send a redirect to sender to shortcut a hop.
1437 	 * Only send redirect if source is sending directly to us,
1438 	 * and if packet was not source routed (or has any options).
1439 	 * Also, don't send redirect if forwarding using a default route
1440 	 * or a route modified by a redirect.
1441 	 */
1442 	dest.s_addr = 0;
1443 	if (!srcrt && V_ipsendredirects &&
1444 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1445 		struct sockaddr_in *sin;
1446 		struct rtentry *rt;
1447 
1448 		bzero(&ro, sizeof(ro));
1449 		sin = (struct sockaddr_in *)&ro.ro_dst;
1450 		sin->sin_family = AF_INET;
1451 		sin->sin_len = sizeof(*sin);
1452 		sin->sin_addr = ip->ip_dst;
1453 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1454 
1455 		rt = ro.ro_rt;
1456 
1457 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1458 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1459 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1460 			u_long src = ntohl(ip->ip_src.s_addr);
1461 
1462 			if (RTA(rt) &&
1463 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1464 				if (rt->rt_flags & RTF_GATEWAY)
1465 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1466 				else
1467 					dest.s_addr = ip->ip_dst.s_addr;
1468 				/* Router requirements says to only send host redirects */
1469 				type = ICMP_REDIRECT;
1470 				code = ICMP_REDIRECT_HOST;
1471 			}
1472 		}
1473 		if (rt)
1474 			RTFREE(rt);
1475 	}
1476 
1477 	/*
1478 	 * Try to cache the route MTU from ip_output so we can consider it for
1479 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1480 	 */
1481 	bzero(&ro, sizeof(ro));
1482 
1483 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1484 
1485 	if (error == EMSGSIZE && ro.ro_rt)
1486 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1487 	RO_RTFREE(&ro);
1488 
1489 	if (error)
1490 		IPSTAT_INC(ips_cantforward);
1491 	else {
1492 		IPSTAT_INC(ips_forward);
1493 		if (type)
1494 			IPSTAT_INC(ips_redirectsent);
1495 		else {
1496 			if (mcopy)
1497 				m_freem(mcopy);
1498 			if (ia != NULL)
1499 				ifa_free(&ia->ia_ifa);
1500 			return;
1501 		}
1502 	}
1503 	if (mcopy == NULL) {
1504 		if (ia != NULL)
1505 			ifa_free(&ia->ia_ifa);
1506 		return;
1507 	}
1508 
1509 	switch (error) {
1510 
1511 	case 0:				/* forwarded, but need redirect */
1512 		/* type, code set above */
1513 		break;
1514 
1515 	case ENETUNREACH:
1516 	case EHOSTUNREACH:
1517 	case ENETDOWN:
1518 	case EHOSTDOWN:
1519 	default:
1520 		type = ICMP_UNREACH;
1521 		code = ICMP_UNREACH_HOST;
1522 		break;
1523 
1524 	case EMSGSIZE:
1525 		type = ICMP_UNREACH;
1526 		code = ICMP_UNREACH_NEEDFRAG;
1527 
1528 #ifdef IPSEC
1529 		/*
1530 		 * If IPsec is configured for this path,
1531 		 * override any possibly mtu value set by ip_output.
1532 		 */
1533 		mtu = ip_ipsec_mtu(mcopy, mtu);
1534 #endif /* IPSEC */
1535 		/*
1536 		 * If the MTU was set before make sure we are below the
1537 		 * interface MTU.
1538 		 * If the MTU wasn't set before use the interface mtu or
1539 		 * fall back to the next smaller mtu step compared to the
1540 		 * current packet size.
1541 		 */
1542 		if (mtu != 0) {
1543 			if (ia != NULL)
1544 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1545 		} else {
1546 			if (ia != NULL)
1547 				mtu = ia->ia_ifp->if_mtu;
1548 			else
1549 				mtu = ip_next_mtu(ntohs(ip->ip_len), 0);
1550 		}
1551 		IPSTAT_INC(ips_cantfrag);
1552 		break;
1553 
1554 	case ENOBUFS:
1555 		/*
1556 		 * A router should not generate ICMP_SOURCEQUENCH as
1557 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1558 		 * Source quench could be a big problem under DoS attacks,
1559 		 * or if the underlying interface is rate-limited.
1560 		 * Those who need source quench packets may re-enable them
1561 		 * via the net.inet.ip.sendsourcequench sysctl.
1562 		 */
1563 		if (V_ip_sendsourcequench == 0) {
1564 			m_freem(mcopy);
1565 			if (ia != NULL)
1566 				ifa_free(&ia->ia_ifa);
1567 			return;
1568 		} else {
1569 			type = ICMP_SOURCEQUENCH;
1570 			code = 0;
1571 		}
1572 		break;
1573 
1574 	case EACCES:			/* ipfw denied packet */
1575 		m_freem(mcopy);
1576 		if (ia != NULL)
1577 			ifa_free(&ia->ia_ifa);
1578 		return;
1579 	}
1580 	if (ia != NULL)
1581 		ifa_free(&ia->ia_ifa);
1582 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1583 }
1584 
1585 void
1586 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1587     struct mbuf *m)
1588 {
1589 
1590 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1591 		struct bintime bt;
1592 
1593 		bintime(&bt);
1594 		if (inp->inp_socket->so_options & SO_BINTIME) {
1595 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1596 			SCM_BINTIME, SOL_SOCKET);
1597 			if (*mp)
1598 				mp = &(*mp)->m_next;
1599 		}
1600 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1601 			struct timeval tv;
1602 
1603 			bintime2timeval(&bt, &tv);
1604 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1605 				SCM_TIMESTAMP, SOL_SOCKET);
1606 			if (*mp)
1607 				mp = &(*mp)->m_next;
1608 		}
1609 	}
1610 	if (inp->inp_flags & INP_RECVDSTADDR) {
1611 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1612 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1613 		if (*mp)
1614 			mp = &(*mp)->m_next;
1615 	}
1616 	if (inp->inp_flags & INP_RECVTTL) {
1617 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1618 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1619 		if (*mp)
1620 			mp = &(*mp)->m_next;
1621 	}
1622 #ifdef notyet
1623 	/* XXX
1624 	 * Moving these out of udp_input() made them even more broken
1625 	 * than they already were.
1626 	 */
1627 	/* options were tossed already */
1628 	if (inp->inp_flags & INP_RECVOPTS) {
1629 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1630 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1631 		if (*mp)
1632 			mp = &(*mp)->m_next;
1633 	}
1634 	/* ip_srcroute doesn't do what we want here, need to fix */
1635 	if (inp->inp_flags & INP_RECVRETOPTS) {
1636 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1637 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1638 		if (*mp)
1639 			mp = &(*mp)->m_next;
1640 	}
1641 #endif
1642 	if (inp->inp_flags & INP_RECVIF) {
1643 		struct ifnet *ifp;
1644 		struct sdlbuf {
1645 			struct sockaddr_dl sdl;
1646 			u_char	pad[32];
1647 		} sdlbuf;
1648 		struct sockaddr_dl *sdp;
1649 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1650 
1651 		if (((ifp = m->m_pkthdr.rcvif))
1652 		&& ( ifp->if_index && (ifp->if_index <= V_if_index))) {
1653 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1654 			/*
1655 			 * Change our mind and don't try copy.
1656 			 */
1657 			if ((sdp->sdl_family != AF_LINK)
1658 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1659 				goto makedummy;
1660 			}
1661 			bcopy(sdp, sdl2, sdp->sdl_len);
1662 		} else {
1663 makedummy:
1664 			sdl2->sdl_len
1665 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1666 			sdl2->sdl_family = AF_LINK;
1667 			sdl2->sdl_index = 0;
1668 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1669 		}
1670 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1671 			IP_RECVIF, IPPROTO_IP);
1672 		if (*mp)
1673 			mp = &(*mp)->m_next;
1674 	}
1675 	if (inp->inp_flags & INP_RECVTOS) {
1676 		*mp = sbcreatecontrol((caddr_t) &ip->ip_tos,
1677 		    sizeof(u_char), IP_RECVTOS, IPPROTO_IP);
1678 		if (*mp)
1679 			mp = &(*mp)->m_next;
1680 	}
1681 }
1682 
1683 /*
1684  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1685  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1686  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1687  * compiled.
1688  */
1689 static VNET_DEFINE(int, ip_rsvp_on);
1690 VNET_DEFINE(struct socket *, ip_rsvpd);
1691 
1692 #define	V_ip_rsvp_on		VNET(ip_rsvp_on)
1693 
1694 int
1695 ip_rsvp_init(struct socket *so)
1696 {
1697 
1698 	if (so->so_type != SOCK_RAW ||
1699 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1700 		return EOPNOTSUPP;
1701 
1702 	if (V_ip_rsvpd != NULL)
1703 		return EADDRINUSE;
1704 
1705 	V_ip_rsvpd = so;
1706 	/*
1707 	 * This may seem silly, but we need to be sure we don't over-increment
1708 	 * the RSVP counter, in case something slips up.
1709 	 */
1710 	if (!V_ip_rsvp_on) {
1711 		V_ip_rsvp_on = 1;
1712 		V_rsvp_on++;
1713 	}
1714 
1715 	return 0;
1716 }
1717 
1718 int
1719 ip_rsvp_done(void)
1720 {
1721 
1722 	V_ip_rsvpd = NULL;
1723 	/*
1724 	 * This may seem silly, but we need to be sure we don't over-decrement
1725 	 * the RSVP counter, in case something slips up.
1726 	 */
1727 	if (V_ip_rsvp_on) {
1728 		V_ip_rsvp_on = 0;
1729 		V_rsvp_on--;
1730 	}
1731 	return 0;
1732 }
1733 
1734 void
1735 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1736 {
1737 
1738 	if (rsvp_input_p) { /* call the real one if loaded */
1739 		rsvp_input_p(m, off);
1740 		return;
1741 	}
1742 
1743 	/* Can still get packets with rsvp_on = 0 if there is a local member
1744 	 * of the group to which the RSVP packet is addressed.  But in this
1745 	 * case we want to throw the packet away.
1746 	 */
1747 
1748 	if (!V_rsvp_on) {
1749 		m_freem(m);
1750 		return;
1751 	}
1752 
1753 	if (V_ip_rsvpd != NULL) {
1754 		rip_input(m, off);
1755 		return;
1756 	}
1757 	/* Drop the packet */
1758 	m_freem(m);
1759 }
1760