xref: /freebsd/sys/netinet/ip_input.c (revision 2d0d168606f477d58d1a9aa72f94915da2f1733a)
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_kdtrace.h"
40 #include "opt_route.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/mbuf.h>
45 #include <sys/malloc.h>
46 #include <sys/domain.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/rwlock.h>
53 #include <sys/sdt.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
56 
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/route.h>
63 #include <net/netisr.h>
64 #include <net/vnet.h>
65 #include <net/flowtable.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_kdtrace.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_fw.h>
75 #include <netinet/ip_icmp.h>
76 #include <netinet/ip_options.h>
77 #include <machine/in_cksum.h>
78 #include <netinet/ip_carp.h>
79 #ifdef IPSEC
80 #include <netinet/ip_ipsec.h>
81 #endif /* IPSEC */
82 
83 #include <sys/socketvar.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #ifdef CTASSERT
88 CTASSERT(sizeof(struct ip) == 20);
89 #endif
90 
91 struct	rwlock in_ifaddr_lock;
92 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock");
93 
94 VNET_DEFINE(int, rsvp_on);
95 
96 VNET_DEFINE(int, ipforwarding);
97 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
98     &VNET_NAME(ipforwarding), 0,
99     "Enable IP forwarding between interfaces");
100 
101 static VNET_DEFINE(int, ipsendredirects) = 1;	/* XXX */
102 #define	V_ipsendredirects	VNET(ipsendredirects)
103 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
104     &VNET_NAME(ipsendredirects), 0,
105     "Enable sending IP redirects");
106 
107 static VNET_DEFINE(int, ip_keepfaith);
108 #define	V_ip_keepfaith		VNET(ip_keepfaith)
109 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
110     &VNET_NAME(ip_keepfaith), 0,
111     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
112 
113 static VNET_DEFINE(int, ip_sendsourcequench);
114 #define	V_ip_sendsourcequench	VNET(ip_sendsourcequench)
115 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
116     &VNET_NAME(ip_sendsourcequench), 0,
117     "Enable the transmission of source quench packets");
118 
119 VNET_DEFINE(int, ip_do_randomid);
120 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
121     &VNET_NAME(ip_do_randomid), 0,
122     "Assign random ip_id values");
123 
124 /*
125  * XXX - Setting ip_checkinterface mostly implements the receive side of
126  * the Strong ES model described in RFC 1122, but since the routing table
127  * and transmit implementation do not implement the Strong ES model,
128  * setting this to 1 results in an odd hybrid.
129  *
130  * XXX - ip_checkinterface currently must be disabled if you use ipnat
131  * to translate the destination address to another local interface.
132  *
133  * XXX - ip_checkinterface must be disabled if you add IP aliases
134  * to the loopback interface instead of the interface where the
135  * packets for those addresses are received.
136  */
137 static VNET_DEFINE(int, ip_checkinterface);
138 #define	V_ip_checkinterface	VNET(ip_checkinterface)
139 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
140     &VNET_NAME(ip_checkinterface), 0,
141     "Verify packet arrives on correct interface");
142 
143 VNET_DEFINE(struct pfil_head, inet_pfil_hook);	/* Packet filter hooks */
144 
145 static struct netisr_handler ip_nh = {
146 	.nh_name = "ip",
147 	.nh_handler = ip_input,
148 	.nh_proto = NETISR_IP,
149 	.nh_policy = NETISR_POLICY_FLOW,
150 };
151 
152 extern	struct domain inetdomain;
153 extern	struct protosw inetsw[];
154 u_char	ip_protox[IPPROTO_MAX];
155 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead);  /* first inet address */
156 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table  */
157 VNET_DEFINE(u_long, in_ifaddrhmask);		/* mask for hash table */
158 
159 static VNET_DEFINE(uma_zone_t, ipq_zone);
160 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]);
161 static struct mtx ipqlock;
162 
163 #define	V_ipq_zone		VNET(ipq_zone)
164 #define	V_ipq			VNET(ipq)
165 
166 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
167 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
168 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
169 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
170 
171 static void	maxnipq_update(void);
172 static void	ipq_zone_change(void *);
173 static void	ip_drain_locked(void);
174 
175 static VNET_DEFINE(int, maxnipq);  /* Administrative limit on # reass queues. */
176 static VNET_DEFINE(int, nipq);			/* Total # of reass queues */
177 #define	V_maxnipq		VNET(maxnipq)
178 #define	V_nipq			VNET(nipq)
179 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
180     &VNET_NAME(nipq), 0,
181     "Current number of IPv4 fragment reassembly queue entries");
182 
183 static VNET_DEFINE(int, maxfragsperpacket);
184 #define	V_maxfragsperpacket	VNET(maxfragsperpacket)
185 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
186     &VNET_NAME(maxfragsperpacket), 0,
187     "Maximum number of IPv4 fragments allowed per packet");
188 
189 #ifdef IPCTL_DEFMTU
190 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
191     &ip_mtu, 0, "Default MTU");
192 #endif
193 
194 #ifdef IPSTEALTH
195 VNET_DEFINE(int, ipstealth);
196 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
197     &VNET_NAME(ipstealth), 0,
198     "IP stealth mode, no TTL decrementation on forwarding");
199 #endif
200 
201 #ifdef FLOWTABLE
202 static VNET_DEFINE(int, ip_output_flowtable_size) = 2048;
203 VNET_DEFINE(struct flowtable *, ip_ft);
204 #define	V_ip_output_flowtable_size	VNET(ip_output_flowtable_size)
205 
206 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
207     &VNET_NAME(ip_output_flowtable_size), 2048,
208     "number of entries in the per-cpu output flow caches");
209 #endif
210 
211 static void	ip_freef(struct ipqhead *, struct ipq *);
212 
213 /*
214  * IP statistics are stored in the "array" of counter(9)s.
215  */
216 VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat);
217 VNET_PCPUSTAT_SYSINIT(ipstat);
218 SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat,
219     "IP statistics (struct ipstat, netinet/ip_var.h)");
220 
221 #ifdef VIMAGE
222 VNET_PCPUSTAT_SYSUNINIT(ipstat);
223 #endif /* VIMAGE */
224 
225 /*
226  * Kernel module interface for updating ipstat.  The argument is an index
227  * into ipstat treated as an array.
228  */
229 void
230 kmod_ipstat_inc(int statnum)
231 {
232 
233 	counter_u64_add(VNET(ipstat)[statnum], 1);
234 }
235 
236 void
237 kmod_ipstat_dec(int statnum)
238 {
239 
240 	counter_u64_add(VNET(ipstat)[statnum], -1);
241 }
242 
243 static int
244 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
245 {
246 	int error, qlimit;
247 
248 	netisr_getqlimit(&ip_nh, &qlimit);
249 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
250 	if (error || !req->newptr)
251 		return (error);
252 	if (qlimit < 1)
253 		return (EINVAL);
254 	return (netisr_setqlimit(&ip_nh, qlimit));
255 }
256 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
257     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
258     "Maximum size of the IP input queue");
259 
260 static int
261 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
262 {
263 	u_int64_t qdrops_long;
264 	int error, qdrops;
265 
266 	netisr_getqdrops(&ip_nh, &qdrops_long);
267 	qdrops = qdrops_long;
268 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
269 	if (error || !req->newptr)
270 		return (error);
271 	if (qdrops != 0)
272 		return (EINVAL);
273 	netisr_clearqdrops(&ip_nh);
274 	return (0);
275 }
276 
277 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
278     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
279     "Number of packets dropped from the IP input queue");
280 
281 /*
282  * IP initialization: fill in IP protocol switch table.
283  * All protocols not implemented in kernel go to raw IP protocol handler.
284  */
285 void
286 ip_init(void)
287 {
288 	struct protosw *pr;
289 	int i;
290 
291 	V_ip_id = time_second & 0xffff;
292 
293 	TAILQ_INIT(&V_in_ifaddrhead);
294 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
295 
296 	/* Initialize IP reassembly queue. */
297 	for (i = 0; i < IPREASS_NHASH; i++)
298 		TAILQ_INIT(&V_ipq[i]);
299 	V_maxnipq = nmbclusters / 32;
300 	V_maxfragsperpacket = 16;
301 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
302 	    NULL, UMA_ALIGN_PTR, 0);
303 	maxnipq_update();
304 
305 	/* Initialize packet filter hooks. */
306 	V_inet_pfil_hook.ph_type = PFIL_TYPE_AF;
307 	V_inet_pfil_hook.ph_af = AF_INET;
308 	if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0)
309 		printf("%s: WARNING: unable to register pfil hook, "
310 			"error %d\n", __func__, i);
311 
312 #ifdef FLOWTABLE
313 	if (TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size",
314 		&V_ip_output_flowtable_size)) {
315 		if (V_ip_output_flowtable_size < 256)
316 			V_ip_output_flowtable_size = 256;
317 		if (!powerof2(V_ip_output_flowtable_size)) {
318 			printf("flowtable must be power of 2 size\n");
319 			V_ip_output_flowtable_size = 2048;
320 		}
321 	} else {
322 		/*
323 		 * round up to the next power of 2
324 		 */
325 		V_ip_output_flowtable_size = 1 << fls((1024 + maxusers * 64)-1);
326 	}
327 	V_ip_ft = flowtable_alloc("ipv4", V_ip_output_flowtable_size, FL_PCPU);
328 #endif
329 
330 	/* Skip initialization of globals for non-default instances. */
331 	if (!IS_DEFAULT_VNET(curvnet))
332 		return;
333 
334 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
335 	if (pr == NULL)
336 		panic("ip_init: PF_INET not found");
337 
338 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
339 	for (i = 0; i < IPPROTO_MAX; i++)
340 		ip_protox[i] = pr - inetsw;
341 	/*
342 	 * Cycle through IP protocols and put them into the appropriate place
343 	 * in ip_protox[].
344 	 */
345 	for (pr = inetdomain.dom_protosw;
346 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
347 		if (pr->pr_domain->dom_family == PF_INET &&
348 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
349 			/* Be careful to only index valid IP protocols. */
350 			if (pr->pr_protocol < IPPROTO_MAX)
351 				ip_protox[pr->pr_protocol] = pr - inetsw;
352 		}
353 
354 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
355 		NULL, EVENTHANDLER_PRI_ANY);
356 
357 	/* Initialize various other remaining things. */
358 	IPQ_LOCK_INIT();
359 	netisr_register(&ip_nh);
360 }
361 
362 #ifdef VIMAGE
363 void
364 ip_destroy(void)
365 {
366 	int i;
367 
368 	if ((i = pfil_head_unregister(&V_inet_pfil_hook)) != 0)
369 		printf("%s: WARNING: unable to unregister pfil hook, "
370 		    "error %d\n", __func__, i);
371 
372 	/* Cleanup in_ifaddr hash table; should be empty. */
373 	hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask);
374 
375 	IPQ_LOCK();
376 	ip_drain_locked();
377 	IPQ_UNLOCK();
378 
379 	uma_zdestroy(V_ipq_zone);
380 }
381 #endif
382 
383 /*
384  * Ip input routine.  Checksum and byte swap header.  If fragmented
385  * try to reassemble.  Process options.  Pass to next level.
386  */
387 void
388 ip_input(struct mbuf *m)
389 {
390 	struct ip *ip = NULL;
391 	struct in_ifaddr *ia = NULL;
392 	struct ifaddr *ifa;
393 	struct ifnet *ifp;
394 	int    checkif, hlen = 0;
395 	uint16_t sum, ip_len;
396 	int dchg = 0;				/* dest changed after fw */
397 	struct in_addr odst;			/* original dst address */
398 
399 	M_ASSERTPKTHDR(m);
400 
401 	if (m->m_flags & M_FASTFWD_OURS) {
402 		m->m_flags &= ~M_FASTFWD_OURS;
403 		/* Set up some basics that will be used later. */
404 		ip = mtod(m, struct ip *);
405 		hlen = ip->ip_hl << 2;
406 		ip_len = ntohs(ip->ip_len);
407 		goto ours;
408 	}
409 
410 	IPSTAT_INC(ips_total);
411 
412 	if (m->m_pkthdr.len < sizeof(struct ip))
413 		goto tooshort;
414 
415 	if (m->m_len < sizeof (struct ip) &&
416 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
417 		IPSTAT_INC(ips_toosmall);
418 		return;
419 	}
420 	ip = mtod(m, struct ip *);
421 
422 	if (ip->ip_v != IPVERSION) {
423 		IPSTAT_INC(ips_badvers);
424 		goto bad;
425 	}
426 
427 	hlen = ip->ip_hl << 2;
428 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
429 		IPSTAT_INC(ips_badhlen);
430 		goto bad;
431 	}
432 	if (hlen > m->m_len) {
433 		if ((m = m_pullup(m, hlen)) == NULL) {
434 			IPSTAT_INC(ips_badhlen);
435 			return;
436 		}
437 		ip = mtod(m, struct ip *);
438 	}
439 
440 	IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL);
441 
442 	/* 127/8 must not appear on wire - RFC1122 */
443 	ifp = m->m_pkthdr.rcvif;
444 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
445 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
446 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
447 			IPSTAT_INC(ips_badaddr);
448 			goto bad;
449 		}
450 	}
451 
452 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
453 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
454 	} else {
455 		if (hlen == sizeof(struct ip)) {
456 			sum = in_cksum_hdr(ip);
457 		} else {
458 			sum = in_cksum(m, hlen);
459 		}
460 	}
461 	if (sum) {
462 		IPSTAT_INC(ips_badsum);
463 		goto bad;
464 	}
465 
466 #ifdef ALTQ
467 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
468 		/* packet is dropped by traffic conditioner */
469 		return;
470 #endif
471 
472 	ip_len = ntohs(ip->ip_len);
473 	if (ip_len < hlen) {
474 		IPSTAT_INC(ips_badlen);
475 		goto bad;
476 	}
477 
478 	/*
479 	 * Check that the amount of data in the buffers
480 	 * is as at least much as the IP header would have us expect.
481 	 * Trim mbufs if longer than we expect.
482 	 * Drop packet if shorter than we expect.
483 	 */
484 	if (m->m_pkthdr.len < ip_len) {
485 tooshort:
486 		IPSTAT_INC(ips_tooshort);
487 		goto bad;
488 	}
489 	if (m->m_pkthdr.len > ip_len) {
490 		if (m->m_len == m->m_pkthdr.len) {
491 			m->m_len = ip_len;
492 			m->m_pkthdr.len = ip_len;
493 		} else
494 			m_adj(m, ip_len - m->m_pkthdr.len);
495 	}
496 #ifdef IPSEC
497 	/*
498 	 * Bypass packet filtering for packets previously handled by IPsec.
499 	 */
500 	if (ip_ipsec_filtertunnel(m))
501 		goto passin;
502 #endif /* IPSEC */
503 
504 	/*
505 	 * Run through list of hooks for input packets.
506 	 *
507 	 * NB: Beware of the destination address changing (e.g.
508 	 *     by NAT rewriting).  When this happens, tell
509 	 *     ip_forward to do the right thing.
510 	 */
511 
512 	/* Jump over all PFIL processing if hooks are not active. */
513 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
514 		goto passin;
515 
516 	odst = ip->ip_dst;
517 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
518 		return;
519 	if (m == NULL)			/* consumed by filter */
520 		return;
521 
522 	ip = mtod(m, struct ip *);
523 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
524 	ifp = m->m_pkthdr.rcvif;
525 
526 	if (m->m_flags & M_FASTFWD_OURS) {
527 		m->m_flags &= ~M_FASTFWD_OURS;
528 		goto ours;
529 	}
530 	if (m->m_flags & M_IP_NEXTHOP) {
531 		dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL);
532 		if (dchg != 0) {
533 			/*
534 			 * Directly ship the packet on.  This allows
535 			 * forwarding packets originally destined to us
536 			 * to some other directly connected host.
537 			 */
538 			ip_forward(m, 1);
539 			return;
540 		}
541 	}
542 passin:
543 
544 	/*
545 	 * Process options and, if not destined for us,
546 	 * ship it on.  ip_dooptions returns 1 when an
547 	 * error was detected (causing an icmp message
548 	 * to be sent and the original packet to be freed).
549 	 */
550 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
551 		return;
552 
553         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
554          * matter if it is destined to another node, or whether it is
555          * a multicast one, RSVP wants it! and prevents it from being forwarded
556          * anywhere else. Also checks if the rsvp daemon is running before
557 	 * grabbing the packet.
558          */
559 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
560 		goto ours;
561 
562 	/*
563 	 * Check our list of addresses, to see if the packet is for us.
564 	 * If we don't have any addresses, assume any unicast packet
565 	 * we receive might be for us (and let the upper layers deal
566 	 * with it).
567 	 */
568 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
569 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
570 		goto ours;
571 
572 	/*
573 	 * Enable a consistency check between the destination address
574 	 * and the arrival interface for a unicast packet (the RFC 1122
575 	 * strong ES model) if IP forwarding is disabled and the packet
576 	 * is not locally generated and the packet is not subject to
577 	 * 'ipfw fwd'.
578 	 *
579 	 * XXX - Checking also should be disabled if the destination
580 	 * address is ipnat'ed to a different interface.
581 	 *
582 	 * XXX - Checking is incompatible with IP aliases added
583 	 * to the loopback interface instead of the interface where
584 	 * the packets are received.
585 	 *
586 	 * XXX - This is the case for carp vhost IPs as well so we
587 	 * insert a workaround. If the packet got here, we already
588 	 * checked with carp_iamatch() and carp_forus().
589 	 */
590 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
591 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
592 	    ifp->if_carp == NULL && (dchg == 0);
593 
594 	/*
595 	 * Check for exact addresses in the hash bucket.
596 	 */
597 	/* IN_IFADDR_RLOCK(); */
598 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
599 		/*
600 		 * If the address matches, verify that the packet
601 		 * arrived via the correct interface if checking is
602 		 * enabled.
603 		 */
604 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
605 		    (!checkif || ia->ia_ifp == ifp)) {
606 			ifa_ref(&ia->ia_ifa);
607 			/* IN_IFADDR_RUNLOCK(); */
608 			goto ours;
609 		}
610 	}
611 	/* IN_IFADDR_RUNLOCK(); */
612 
613 	/*
614 	 * Check for broadcast addresses.
615 	 *
616 	 * Only accept broadcast packets that arrive via the matching
617 	 * interface.  Reception of forwarded directed broadcasts would
618 	 * be handled via ip_forward() and ether_output() with the loopback
619 	 * into the stack for SIMPLEX interfaces handled by ether_output().
620 	 */
621 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
622 		IF_ADDR_RLOCK(ifp);
623 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
624 			if (ifa->ifa_addr->sa_family != AF_INET)
625 				continue;
626 			ia = ifatoia(ifa);
627 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
628 			    ip->ip_dst.s_addr) {
629 				ifa_ref(ifa);
630 				IF_ADDR_RUNLOCK(ifp);
631 				goto ours;
632 			}
633 #ifdef BOOTP_COMPAT
634 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
635 				ifa_ref(ifa);
636 				IF_ADDR_RUNLOCK(ifp);
637 				goto ours;
638 			}
639 #endif
640 		}
641 		IF_ADDR_RUNLOCK(ifp);
642 		ia = NULL;
643 	}
644 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
645 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
646 		IPSTAT_INC(ips_cantforward);
647 		m_freem(m);
648 		return;
649 	}
650 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
651 		if (V_ip_mrouter) {
652 			/*
653 			 * If we are acting as a multicast router, all
654 			 * incoming multicast packets are passed to the
655 			 * kernel-level multicast forwarding function.
656 			 * The packet is returned (relatively) intact; if
657 			 * ip_mforward() returns a non-zero value, the packet
658 			 * must be discarded, else it may be accepted below.
659 			 */
660 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
661 				IPSTAT_INC(ips_cantforward);
662 				m_freem(m);
663 				return;
664 			}
665 
666 			/*
667 			 * The process-level routing daemon needs to receive
668 			 * all multicast IGMP packets, whether or not this
669 			 * host belongs to their destination groups.
670 			 */
671 			if (ip->ip_p == IPPROTO_IGMP)
672 				goto ours;
673 			IPSTAT_INC(ips_forward);
674 		}
675 		/*
676 		 * Assume the packet is for us, to avoid prematurely taking
677 		 * a lock on the in_multi hash. Protocols must perform
678 		 * their own filtering and update statistics accordingly.
679 		 */
680 		goto ours;
681 	}
682 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
683 		goto ours;
684 	if (ip->ip_dst.s_addr == INADDR_ANY)
685 		goto ours;
686 
687 	/*
688 	 * FAITH(Firewall Aided Internet Translator)
689 	 */
690 	if (ifp && ifp->if_type == IFT_FAITH) {
691 		if (V_ip_keepfaith) {
692 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
693 				goto ours;
694 		}
695 		m_freem(m);
696 		return;
697 	}
698 
699 	/*
700 	 * Not for us; forward if possible and desirable.
701 	 */
702 	if (V_ipforwarding == 0) {
703 		IPSTAT_INC(ips_cantforward);
704 		m_freem(m);
705 	} else {
706 #ifdef IPSEC
707 		if (ip_ipsec_fwd(m))
708 			goto bad;
709 #endif /* IPSEC */
710 		ip_forward(m, dchg);
711 	}
712 	return;
713 
714 ours:
715 #ifdef IPSTEALTH
716 	/*
717 	 * IPSTEALTH: Process non-routing options only
718 	 * if the packet is destined for us.
719 	 */
720 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) {
721 		if (ia != NULL)
722 			ifa_free(&ia->ia_ifa);
723 		return;
724 	}
725 #endif /* IPSTEALTH */
726 
727 	/* Count the packet in the ip address stats */
728 	if (ia != NULL) {
729 		ia->ia_ifa.if_ipackets++;
730 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
731 		ifa_free(&ia->ia_ifa);
732 	}
733 
734 	/*
735 	 * Attempt reassembly; if it succeeds, proceed.
736 	 * ip_reass() will return a different mbuf.
737 	 */
738 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
739 		m = ip_reass(m);
740 		if (m == NULL)
741 			return;
742 		ip = mtod(m, struct ip *);
743 		/* Get the header length of the reassembled packet */
744 		hlen = ip->ip_hl << 2;
745 	}
746 
747 #ifdef IPSEC
748 	/*
749 	 * enforce IPsec policy checking if we are seeing last header.
750 	 * note that we do not visit this with protocols with pcb layer
751 	 * code - like udp/tcp/raw ip.
752 	 */
753 	if (ip_ipsec_input(m))
754 		goto bad;
755 #endif /* IPSEC */
756 
757 	/*
758 	 * Switch out to protocol's input routine.
759 	 */
760 	IPSTAT_INC(ips_delivered);
761 
762 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
763 	return;
764 bad:
765 	m_freem(m);
766 }
767 
768 /*
769  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
770  * max has slightly different semantics than the sysctl, for historical
771  * reasons.
772  */
773 static void
774 maxnipq_update(void)
775 {
776 
777 	/*
778 	 * -1 for unlimited allocation.
779 	 */
780 	if (V_maxnipq < 0)
781 		uma_zone_set_max(V_ipq_zone, 0);
782 	/*
783 	 * Positive number for specific bound.
784 	 */
785 	if (V_maxnipq > 0)
786 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
787 	/*
788 	 * Zero specifies no further fragment queue allocation -- set the
789 	 * bound very low, but rely on implementation elsewhere to actually
790 	 * prevent allocation and reclaim current queues.
791 	 */
792 	if (V_maxnipq == 0)
793 		uma_zone_set_max(V_ipq_zone, 1);
794 }
795 
796 static void
797 ipq_zone_change(void *tag)
798 {
799 
800 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
801 		V_maxnipq = nmbclusters / 32;
802 		maxnipq_update();
803 	}
804 }
805 
806 static int
807 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
808 {
809 	int error, i;
810 
811 	i = V_maxnipq;
812 	error = sysctl_handle_int(oidp, &i, 0, req);
813 	if (error || !req->newptr)
814 		return (error);
815 
816 	/*
817 	 * XXXRW: Might be a good idea to sanity check the argument and place
818 	 * an extreme upper bound.
819 	 */
820 	if (i < -1)
821 		return (EINVAL);
822 	V_maxnipq = i;
823 	maxnipq_update();
824 	return (0);
825 }
826 
827 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
828     NULL, 0, sysctl_maxnipq, "I",
829     "Maximum number of IPv4 fragment reassembly queue entries");
830 
831 /*
832  * Take incoming datagram fragment and try to reassemble it into
833  * whole datagram.  If the argument is the first fragment or one
834  * in between the function will return NULL and store the mbuf
835  * in the fragment chain.  If the argument is the last fragment
836  * the packet will be reassembled and the pointer to the new
837  * mbuf returned for further processing.  Only m_tags attached
838  * to the first packet/fragment are preserved.
839  * The IP header is *NOT* adjusted out of iplen.
840  */
841 struct mbuf *
842 ip_reass(struct mbuf *m)
843 {
844 	struct ip *ip;
845 	struct mbuf *p, *q, *nq, *t;
846 	struct ipq *fp = NULL;
847 	struct ipqhead *head;
848 	int i, hlen, next;
849 	u_int8_t ecn, ecn0;
850 	u_short hash;
851 
852 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
853 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
854 		IPSTAT_INC(ips_fragments);
855 		IPSTAT_INC(ips_fragdropped);
856 		m_freem(m);
857 		return (NULL);
858 	}
859 
860 	ip = mtod(m, struct ip *);
861 	hlen = ip->ip_hl << 2;
862 
863 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
864 	head = &V_ipq[hash];
865 	IPQ_LOCK();
866 
867 	/*
868 	 * Look for queue of fragments
869 	 * of this datagram.
870 	 */
871 	TAILQ_FOREACH(fp, head, ipq_list)
872 		if (ip->ip_id == fp->ipq_id &&
873 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
874 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
875 #ifdef MAC
876 		    mac_ipq_match(m, fp) &&
877 #endif
878 		    ip->ip_p == fp->ipq_p)
879 			goto found;
880 
881 	fp = NULL;
882 
883 	/*
884 	 * Attempt to trim the number of allocated fragment queues if it
885 	 * exceeds the administrative limit.
886 	 */
887 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
888 		/*
889 		 * drop something from the tail of the current queue
890 		 * before proceeding further
891 		 */
892 		struct ipq *q = TAILQ_LAST(head, ipqhead);
893 		if (q == NULL) {   /* gak */
894 			for (i = 0; i < IPREASS_NHASH; i++) {
895 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
896 				if (r) {
897 					IPSTAT_ADD(ips_fragtimeout,
898 					    r->ipq_nfrags);
899 					ip_freef(&V_ipq[i], r);
900 					break;
901 				}
902 			}
903 		} else {
904 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
905 			ip_freef(head, q);
906 		}
907 	}
908 
909 found:
910 	/*
911 	 * Adjust ip_len to not reflect header,
912 	 * convert offset of this to bytes.
913 	 */
914 	ip->ip_len = htons(ntohs(ip->ip_len) - hlen);
915 	if (ip->ip_off & htons(IP_MF)) {
916 		/*
917 		 * Make sure that fragments have a data length
918 		 * that's a non-zero multiple of 8 bytes.
919 		 */
920 		if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) {
921 			IPSTAT_INC(ips_toosmall); /* XXX */
922 			goto dropfrag;
923 		}
924 		m->m_flags |= M_IP_FRAG;
925 	} else
926 		m->m_flags &= ~M_IP_FRAG;
927 	ip->ip_off = htons(ntohs(ip->ip_off) << 3);
928 
929 	/*
930 	 * Attempt reassembly; if it succeeds, proceed.
931 	 * ip_reass() will return a different mbuf.
932 	 */
933 	IPSTAT_INC(ips_fragments);
934 	m->m_pkthdr.PH_loc.ptr = ip;
935 
936 	/* Previous ip_reass() started here. */
937 	/*
938 	 * Presence of header sizes in mbufs
939 	 * would confuse code below.
940 	 */
941 	m->m_data += hlen;
942 	m->m_len -= hlen;
943 
944 	/*
945 	 * If first fragment to arrive, create a reassembly queue.
946 	 */
947 	if (fp == NULL) {
948 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
949 		if (fp == NULL)
950 			goto dropfrag;
951 #ifdef MAC
952 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
953 			uma_zfree(V_ipq_zone, fp);
954 			fp = NULL;
955 			goto dropfrag;
956 		}
957 		mac_ipq_create(m, fp);
958 #endif
959 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
960 		V_nipq++;
961 		fp->ipq_nfrags = 1;
962 		fp->ipq_ttl = IPFRAGTTL;
963 		fp->ipq_p = ip->ip_p;
964 		fp->ipq_id = ip->ip_id;
965 		fp->ipq_src = ip->ip_src;
966 		fp->ipq_dst = ip->ip_dst;
967 		fp->ipq_frags = m;
968 		m->m_nextpkt = NULL;
969 		goto done;
970 	} else {
971 		fp->ipq_nfrags++;
972 #ifdef MAC
973 		mac_ipq_update(m, fp);
974 #endif
975 	}
976 
977 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.PH_loc.ptr))
978 
979 	/*
980 	 * Handle ECN by comparing this segment with the first one;
981 	 * if CE is set, do not lose CE.
982 	 * drop if CE and not-ECT are mixed for the same packet.
983 	 */
984 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
985 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
986 	if (ecn == IPTOS_ECN_CE) {
987 		if (ecn0 == IPTOS_ECN_NOTECT)
988 			goto dropfrag;
989 		if (ecn0 != IPTOS_ECN_CE)
990 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
991 	}
992 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
993 		goto dropfrag;
994 
995 	/*
996 	 * Find a segment which begins after this one does.
997 	 */
998 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
999 		if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off))
1000 			break;
1001 
1002 	/*
1003 	 * If there is a preceding segment, it may provide some of
1004 	 * our data already.  If so, drop the data from the incoming
1005 	 * segment.  If it provides all of our data, drop us, otherwise
1006 	 * stick new segment in the proper place.
1007 	 *
1008 	 * If some of the data is dropped from the preceding
1009 	 * segment, then it's checksum is invalidated.
1010 	 */
1011 	if (p) {
1012 		i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) -
1013 		    ntohs(ip->ip_off);
1014 		if (i > 0) {
1015 			if (i >= ntohs(ip->ip_len))
1016 				goto dropfrag;
1017 			m_adj(m, i);
1018 			m->m_pkthdr.csum_flags = 0;
1019 			ip->ip_off = htons(ntohs(ip->ip_off) + i);
1020 			ip->ip_len = htons(ntohs(ip->ip_len) - i);
1021 		}
1022 		m->m_nextpkt = p->m_nextpkt;
1023 		p->m_nextpkt = m;
1024 	} else {
1025 		m->m_nextpkt = fp->ipq_frags;
1026 		fp->ipq_frags = m;
1027 	}
1028 
1029 	/*
1030 	 * While we overlap succeeding segments trim them or,
1031 	 * if they are completely covered, dequeue them.
1032 	 */
1033 	for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) >
1034 	    ntohs(GETIP(q)->ip_off); q = nq) {
1035 		i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) -
1036 		    ntohs(GETIP(q)->ip_off);
1037 		if (i < ntohs(GETIP(q)->ip_len)) {
1038 			GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i);
1039 			GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i);
1040 			m_adj(q, i);
1041 			q->m_pkthdr.csum_flags = 0;
1042 			break;
1043 		}
1044 		nq = q->m_nextpkt;
1045 		m->m_nextpkt = nq;
1046 		IPSTAT_INC(ips_fragdropped);
1047 		fp->ipq_nfrags--;
1048 		m_freem(q);
1049 	}
1050 
1051 	/*
1052 	 * Check for complete reassembly and perform frag per packet
1053 	 * limiting.
1054 	 *
1055 	 * Frag limiting is performed here so that the nth frag has
1056 	 * a chance to complete the packet before we drop the packet.
1057 	 * As a result, n+1 frags are actually allowed per packet, but
1058 	 * only n will ever be stored. (n = maxfragsperpacket.)
1059 	 *
1060 	 */
1061 	next = 0;
1062 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1063 		if (ntohs(GETIP(q)->ip_off) != next) {
1064 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1065 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1066 				ip_freef(head, fp);
1067 			}
1068 			goto done;
1069 		}
1070 		next += ntohs(GETIP(q)->ip_len);
1071 	}
1072 	/* Make sure the last packet didn't have the IP_MF flag */
1073 	if (p->m_flags & M_IP_FRAG) {
1074 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1075 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1076 			ip_freef(head, fp);
1077 		}
1078 		goto done;
1079 	}
1080 
1081 	/*
1082 	 * Reassembly is complete.  Make sure the packet is a sane size.
1083 	 */
1084 	q = fp->ipq_frags;
1085 	ip = GETIP(q);
1086 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1087 		IPSTAT_INC(ips_toolong);
1088 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1089 		ip_freef(head, fp);
1090 		goto done;
1091 	}
1092 
1093 	/*
1094 	 * Concatenate fragments.
1095 	 */
1096 	m = q;
1097 	t = m->m_next;
1098 	m->m_next = NULL;
1099 	m_cat(m, t);
1100 	nq = q->m_nextpkt;
1101 	q->m_nextpkt = NULL;
1102 	for (q = nq; q != NULL; q = nq) {
1103 		nq = q->m_nextpkt;
1104 		q->m_nextpkt = NULL;
1105 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1106 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1107 		m_cat(m, q);
1108 	}
1109 	/*
1110 	 * In order to do checksumming faster we do 'end-around carry' here
1111 	 * (and not in for{} loop), though it implies we are not going to
1112 	 * reassemble more than 64k fragments.
1113 	 */
1114 	m->m_pkthdr.csum_data =
1115 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1116 #ifdef MAC
1117 	mac_ipq_reassemble(fp, m);
1118 	mac_ipq_destroy(fp);
1119 #endif
1120 
1121 	/*
1122 	 * Create header for new ip packet by modifying header of first
1123 	 * packet;  dequeue and discard fragment reassembly header.
1124 	 * Make header visible.
1125 	 */
1126 	ip->ip_len = htons((ip->ip_hl << 2) + next);
1127 	ip->ip_src = fp->ipq_src;
1128 	ip->ip_dst = fp->ipq_dst;
1129 	TAILQ_REMOVE(head, fp, ipq_list);
1130 	V_nipq--;
1131 	uma_zfree(V_ipq_zone, fp);
1132 	m->m_len += (ip->ip_hl << 2);
1133 	m->m_data -= (ip->ip_hl << 2);
1134 	/* some debugging cruft by sklower, below, will go away soon */
1135 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1136 		m_fixhdr(m);
1137 	IPSTAT_INC(ips_reassembled);
1138 	IPQ_UNLOCK();
1139 	return (m);
1140 
1141 dropfrag:
1142 	IPSTAT_INC(ips_fragdropped);
1143 	if (fp != NULL)
1144 		fp->ipq_nfrags--;
1145 	m_freem(m);
1146 done:
1147 	IPQ_UNLOCK();
1148 	return (NULL);
1149 
1150 #undef GETIP
1151 }
1152 
1153 /*
1154  * Free a fragment reassembly header and all
1155  * associated datagrams.
1156  */
1157 static void
1158 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1159 {
1160 	struct mbuf *q;
1161 
1162 	IPQ_LOCK_ASSERT();
1163 
1164 	while (fp->ipq_frags) {
1165 		q = fp->ipq_frags;
1166 		fp->ipq_frags = q->m_nextpkt;
1167 		m_freem(q);
1168 	}
1169 	TAILQ_REMOVE(fhp, fp, ipq_list);
1170 	uma_zfree(V_ipq_zone, fp);
1171 	V_nipq--;
1172 }
1173 
1174 /*
1175  * IP timer processing;
1176  * if a timer expires on a reassembly
1177  * queue, discard it.
1178  */
1179 void
1180 ip_slowtimo(void)
1181 {
1182 	VNET_ITERATOR_DECL(vnet_iter);
1183 	struct ipq *fp;
1184 	int i;
1185 
1186 	VNET_LIST_RLOCK_NOSLEEP();
1187 	IPQ_LOCK();
1188 	VNET_FOREACH(vnet_iter) {
1189 		CURVNET_SET(vnet_iter);
1190 		for (i = 0; i < IPREASS_NHASH; i++) {
1191 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1192 				struct ipq *fpp;
1193 
1194 				fpp = fp;
1195 				fp = TAILQ_NEXT(fp, ipq_list);
1196 				if(--fpp->ipq_ttl == 0) {
1197 					IPSTAT_ADD(ips_fragtimeout,
1198 					    fpp->ipq_nfrags);
1199 					ip_freef(&V_ipq[i], fpp);
1200 				}
1201 			}
1202 		}
1203 		/*
1204 		 * If we are over the maximum number of fragments
1205 		 * (due to the limit being lowered), drain off
1206 		 * enough to get down to the new limit.
1207 		 */
1208 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1209 			for (i = 0; i < IPREASS_NHASH; i++) {
1210 				while (V_nipq > V_maxnipq &&
1211 				    !TAILQ_EMPTY(&V_ipq[i])) {
1212 					IPSTAT_ADD(ips_fragdropped,
1213 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1214 					ip_freef(&V_ipq[i],
1215 					    TAILQ_FIRST(&V_ipq[i]));
1216 				}
1217 			}
1218 		}
1219 		CURVNET_RESTORE();
1220 	}
1221 	IPQ_UNLOCK();
1222 	VNET_LIST_RUNLOCK_NOSLEEP();
1223 }
1224 
1225 /*
1226  * Drain off all datagram fragments.
1227  */
1228 static void
1229 ip_drain_locked(void)
1230 {
1231 	int     i;
1232 
1233 	IPQ_LOCK_ASSERT();
1234 
1235 	for (i = 0; i < IPREASS_NHASH; i++) {
1236 		while(!TAILQ_EMPTY(&V_ipq[i])) {
1237 			IPSTAT_ADD(ips_fragdropped,
1238 			    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1239 			ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1240 		}
1241 	}
1242 }
1243 
1244 void
1245 ip_drain(void)
1246 {
1247 	VNET_ITERATOR_DECL(vnet_iter);
1248 
1249 	VNET_LIST_RLOCK_NOSLEEP();
1250 	IPQ_LOCK();
1251 	VNET_FOREACH(vnet_iter) {
1252 		CURVNET_SET(vnet_iter);
1253 		ip_drain_locked();
1254 		CURVNET_RESTORE();
1255 	}
1256 	IPQ_UNLOCK();
1257 	VNET_LIST_RUNLOCK_NOSLEEP();
1258 	in_rtqdrain();
1259 }
1260 
1261 /*
1262  * The protocol to be inserted into ip_protox[] must be already registered
1263  * in inetsw[], either statically or through pf_proto_register().
1264  */
1265 int
1266 ipproto_register(short ipproto)
1267 {
1268 	struct protosw *pr;
1269 
1270 	/* Sanity checks. */
1271 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1272 		return (EPROTONOSUPPORT);
1273 
1274 	/*
1275 	 * The protocol slot must not be occupied by another protocol
1276 	 * already.  An index pointing to IPPROTO_RAW is unused.
1277 	 */
1278 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1279 	if (pr == NULL)
1280 		return (EPFNOSUPPORT);
1281 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1282 		return (EEXIST);
1283 
1284 	/* Find the protocol position in inetsw[] and set the index. */
1285 	for (pr = inetdomain.dom_protosw;
1286 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1287 		if (pr->pr_domain->dom_family == PF_INET &&
1288 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1289 			ip_protox[pr->pr_protocol] = pr - inetsw;
1290 			return (0);
1291 		}
1292 	}
1293 	return (EPROTONOSUPPORT);
1294 }
1295 
1296 int
1297 ipproto_unregister(short ipproto)
1298 {
1299 	struct protosw *pr;
1300 
1301 	/* Sanity checks. */
1302 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1303 		return (EPROTONOSUPPORT);
1304 
1305 	/* Check if the protocol was indeed registered. */
1306 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1307 	if (pr == NULL)
1308 		return (EPFNOSUPPORT);
1309 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1310 		return (ENOENT);
1311 
1312 	/* Reset the protocol slot to IPPROTO_RAW. */
1313 	ip_protox[ipproto] = pr - inetsw;
1314 	return (0);
1315 }
1316 
1317 /*
1318  * Given address of next destination (final or next hop), return (referenced)
1319  * internet address info of interface to be used to get there.
1320  */
1321 struct in_ifaddr *
1322 ip_rtaddr(struct in_addr dst, u_int fibnum)
1323 {
1324 	struct route sro;
1325 	struct sockaddr_in *sin;
1326 	struct in_ifaddr *ia;
1327 
1328 	bzero(&sro, sizeof(sro));
1329 	sin = (struct sockaddr_in *)&sro.ro_dst;
1330 	sin->sin_family = AF_INET;
1331 	sin->sin_len = sizeof(*sin);
1332 	sin->sin_addr = dst;
1333 	in_rtalloc_ign(&sro, 0, fibnum);
1334 
1335 	if (sro.ro_rt == NULL)
1336 		return (NULL);
1337 
1338 	ia = ifatoia(sro.ro_rt->rt_ifa);
1339 	ifa_ref(&ia->ia_ifa);
1340 	RTFREE(sro.ro_rt);
1341 	return (ia);
1342 }
1343 
1344 u_char inetctlerrmap[PRC_NCMDS] = {
1345 	0,		0,		0,		0,
1346 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1347 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1348 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1349 	0,		0,		EHOSTUNREACH,	0,
1350 	ENOPROTOOPT,	ECONNREFUSED
1351 };
1352 
1353 /*
1354  * Forward a packet.  If some error occurs return the sender
1355  * an icmp packet.  Note we can't always generate a meaningful
1356  * icmp message because icmp doesn't have a large enough repertoire
1357  * of codes and types.
1358  *
1359  * If not forwarding, just drop the packet.  This could be confusing
1360  * if ipforwarding was zero but some routing protocol was advancing
1361  * us as a gateway to somewhere.  However, we must let the routing
1362  * protocol deal with that.
1363  *
1364  * The srcrt parameter indicates whether the packet is being forwarded
1365  * via a source route.
1366  */
1367 void
1368 ip_forward(struct mbuf *m, int srcrt)
1369 {
1370 	struct ip *ip = mtod(m, struct ip *);
1371 	struct in_ifaddr *ia;
1372 	struct mbuf *mcopy;
1373 	struct in_addr dest;
1374 	struct route ro;
1375 	int error, type = 0, code = 0, mtu = 0;
1376 
1377 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1378 		IPSTAT_INC(ips_cantforward);
1379 		m_freem(m);
1380 		return;
1381 	}
1382 #ifdef IPSTEALTH
1383 	if (!V_ipstealth) {
1384 #endif
1385 		if (ip->ip_ttl <= IPTTLDEC) {
1386 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1387 			    0, 0);
1388 			return;
1389 		}
1390 #ifdef IPSTEALTH
1391 	}
1392 #endif
1393 
1394 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1395 #ifndef IPSEC
1396 	/*
1397 	 * 'ia' may be NULL if there is no route for this destination.
1398 	 * In case of IPsec, Don't discard it just yet, but pass it to
1399 	 * ip_output in case of outgoing IPsec policy.
1400 	 */
1401 	if (!srcrt && ia == NULL) {
1402 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1403 		return;
1404 	}
1405 #endif
1406 
1407 	/*
1408 	 * Save the IP header and at most 8 bytes of the payload,
1409 	 * in case we need to generate an ICMP message to the src.
1410 	 *
1411 	 * XXX this can be optimized a lot by saving the data in a local
1412 	 * buffer on the stack (72 bytes at most), and only allocating the
1413 	 * mbuf if really necessary. The vast majority of the packets
1414 	 * are forwarded without having to send an ICMP back (either
1415 	 * because unnecessary, or because rate limited), so we are
1416 	 * really we are wasting a lot of work here.
1417 	 *
1418 	 * We don't use m_copy() because it might return a reference
1419 	 * to a shared cluster. Both this function and ip_output()
1420 	 * assume exclusive access to the IP header in `m', so any
1421 	 * data in a cluster may change before we reach icmp_error().
1422 	 */
1423 	mcopy = m_gethdr(M_NOWAIT, m->m_type);
1424 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) {
1425 		/*
1426 		 * It's probably ok if the pkthdr dup fails (because
1427 		 * the deep copy of the tag chain failed), but for now
1428 		 * be conservative and just discard the copy since
1429 		 * code below may some day want the tags.
1430 		 */
1431 		m_free(mcopy);
1432 		mcopy = NULL;
1433 	}
1434 	if (mcopy != NULL) {
1435 		mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy));
1436 		mcopy->m_pkthdr.len = mcopy->m_len;
1437 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1438 	}
1439 
1440 #ifdef IPSTEALTH
1441 	if (!V_ipstealth) {
1442 #endif
1443 		ip->ip_ttl -= IPTTLDEC;
1444 #ifdef IPSTEALTH
1445 	}
1446 #endif
1447 
1448 	/*
1449 	 * If forwarding packet using same interface that it came in on,
1450 	 * perhaps should send a redirect to sender to shortcut a hop.
1451 	 * Only send redirect if source is sending directly to us,
1452 	 * and if packet was not source routed (or has any options).
1453 	 * Also, don't send redirect if forwarding using a default route
1454 	 * or a route modified by a redirect.
1455 	 */
1456 	dest.s_addr = 0;
1457 	if (!srcrt && V_ipsendredirects &&
1458 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1459 		struct sockaddr_in *sin;
1460 		struct rtentry *rt;
1461 
1462 		bzero(&ro, sizeof(ro));
1463 		sin = (struct sockaddr_in *)&ro.ro_dst;
1464 		sin->sin_family = AF_INET;
1465 		sin->sin_len = sizeof(*sin);
1466 		sin->sin_addr = ip->ip_dst;
1467 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1468 
1469 		rt = ro.ro_rt;
1470 
1471 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1472 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1473 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1474 			u_long src = ntohl(ip->ip_src.s_addr);
1475 
1476 			if (RTA(rt) &&
1477 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1478 				if (rt->rt_flags & RTF_GATEWAY)
1479 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1480 				else
1481 					dest.s_addr = ip->ip_dst.s_addr;
1482 				/* Router requirements says to only send host redirects */
1483 				type = ICMP_REDIRECT;
1484 				code = ICMP_REDIRECT_HOST;
1485 			}
1486 		}
1487 		if (rt)
1488 			RTFREE(rt);
1489 	}
1490 
1491 	/*
1492 	 * Try to cache the route MTU from ip_output so we can consider it for
1493 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1494 	 */
1495 	bzero(&ro, sizeof(ro));
1496 
1497 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1498 
1499 	if (error == EMSGSIZE && ro.ro_rt)
1500 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1501 	RO_RTFREE(&ro);
1502 
1503 	if (error)
1504 		IPSTAT_INC(ips_cantforward);
1505 	else {
1506 		IPSTAT_INC(ips_forward);
1507 		if (type)
1508 			IPSTAT_INC(ips_redirectsent);
1509 		else {
1510 			if (mcopy)
1511 				m_freem(mcopy);
1512 			if (ia != NULL)
1513 				ifa_free(&ia->ia_ifa);
1514 			return;
1515 		}
1516 	}
1517 	if (mcopy == NULL) {
1518 		if (ia != NULL)
1519 			ifa_free(&ia->ia_ifa);
1520 		return;
1521 	}
1522 
1523 	switch (error) {
1524 
1525 	case 0:				/* forwarded, but need redirect */
1526 		/* type, code set above */
1527 		break;
1528 
1529 	case ENETUNREACH:
1530 	case EHOSTUNREACH:
1531 	case ENETDOWN:
1532 	case EHOSTDOWN:
1533 	default:
1534 		type = ICMP_UNREACH;
1535 		code = ICMP_UNREACH_HOST;
1536 		break;
1537 
1538 	case EMSGSIZE:
1539 		type = ICMP_UNREACH;
1540 		code = ICMP_UNREACH_NEEDFRAG;
1541 
1542 #ifdef IPSEC
1543 		/*
1544 		 * If IPsec is configured for this path,
1545 		 * override any possibly mtu value set by ip_output.
1546 		 */
1547 		mtu = ip_ipsec_mtu(mcopy, mtu);
1548 #endif /* IPSEC */
1549 		/*
1550 		 * If the MTU was set before make sure we are below the
1551 		 * interface MTU.
1552 		 * If the MTU wasn't set before use the interface mtu or
1553 		 * fall back to the next smaller mtu step compared to the
1554 		 * current packet size.
1555 		 */
1556 		if (mtu != 0) {
1557 			if (ia != NULL)
1558 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1559 		} else {
1560 			if (ia != NULL)
1561 				mtu = ia->ia_ifp->if_mtu;
1562 			else
1563 				mtu = ip_next_mtu(ntohs(ip->ip_len), 0);
1564 		}
1565 		IPSTAT_INC(ips_cantfrag);
1566 		break;
1567 
1568 	case ENOBUFS:
1569 		/*
1570 		 * A router should not generate ICMP_SOURCEQUENCH as
1571 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1572 		 * Source quench could be a big problem under DoS attacks,
1573 		 * or if the underlying interface is rate-limited.
1574 		 * Those who need source quench packets may re-enable them
1575 		 * via the net.inet.ip.sendsourcequench sysctl.
1576 		 */
1577 		if (V_ip_sendsourcequench == 0) {
1578 			m_freem(mcopy);
1579 			if (ia != NULL)
1580 				ifa_free(&ia->ia_ifa);
1581 			return;
1582 		} else {
1583 			type = ICMP_SOURCEQUENCH;
1584 			code = 0;
1585 		}
1586 		break;
1587 
1588 	case EACCES:			/* ipfw denied packet */
1589 		m_freem(mcopy);
1590 		if (ia != NULL)
1591 			ifa_free(&ia->ia_ifa);
1592 		return;
1593 	}
1594 	if (ia != NULL)
1595 		ifa_free(&ia->ia_ifa);
1596 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1597 }
1598 
1599 void
1600 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1601     struct mbuf *m)
1602 {
1603 
1604 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1605 		struct bintime bt;
1606 
1607 		bintime(&bt);
1608 		if (inp->inp_socket->so_options & SO_BINTIME) {
1609 			*mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt),
1610 			    SCM_BINTIME, SOL_SOCKET);
1611 			if (*mp)
1612 				mp = &(*mp)->m_next;
1613 		}
1614 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1615 			struct timeval tv;
1616 
1617 			bintime2timeval(&bt, &tv);
1618 			*mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv),
1619 			    SCM_TIMESTAMP, SOL_SOCKET);
1620 			if (*mp)
1621 				mp = &(*mp)->m_next;
1622 		}
1623 	}
1624 	if (inp->inp_flags & INP_RECVDSTADDR) {
1625 		*mp = sbcreatecontrol((caddr_t)&ip->ip_dst,
1626 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1627 		if (*mp)
1628 			mp = &(*mp)->m_next;
1629 	}
1630 	if (inp->inp_flags & INP_RECVTTL) {
1631 		*mp = sbcreatecontrol((caddr_t)&ip->ip_ttl,
1632 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1633 		if (*mp)
1634 			mp = &(*mp)->m_next;
1635 	}
1636 #ifdef notyet
1637 	/* XXX
1638 	 * Moving these out of udp_input() made them even more broken
1639 	 * than they already were.
1640 	 */
1641 	/* options were tossed already */
1642 	if (inp->inp_flags & INP_RECVOPTS) {
1643 		*mp = sbcreatecontrol((caddr_t)opts_deleted_above,
1644 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1645 		if (*mp)
1646 			mp = &(*mp)->m_next;
1647 	}
1648 	/* ip_srcroute doesn't do what we want here, need to fix */
1649 	if (inp->inp_flags & INP_RECVRETOPTS) {
1650 		*mp = sbcreatecontrol((caddr_t)ip_srcroute(m),
1651 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1652 		if (*mp)
1653 			mp = &(*mp)->m_next;
1654 	}
1655 #endif
1656 	if (inp->inp_flags & INP_RECVIF) {
1657 		struct ifnet *ifp;
1658 		struct sdlbuf {
1659 			struct sockaddr_dl sdl;
1660 			u_char	pad[32];
1661 		} sdlbuf;
1662 		struct sockaddr_dl *sdp;
1663 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1664 
1665 		if ((ifp = m->m_pkthdr.rcvif) &&
1666 		    ifp->if_index && ifp->if_index <= V_if_index) {
1667 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1668 			/*
1669 			 * Change our mind and don't try copy.
1670 			 */
1671 			if (sdp->sdl_family != AF_LINK ||
1672 			    sdp->sdl_len > sizeof(sdlbuf)) {
1673 				goto makedummy;
1674 			}
1675 			bcopy(sdp, sdl2, sdp->sdl_len);
1676 		} else {
1677 makedummy:
1678 			sdl2->sdl_len =
1679 			    offsetof(struct sockaddr_dl, sdl_data[0]);
1680 			sdl2->sdl_family = AF_LINK;
1681 			sdl2->sdl_index = 0;
1682 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1683 		}
1684 		*mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len,
1685 		    IP_RECVIF, IPPROTO_IP);
1686 		if (*mp)
1687 			mp = &(*mp)->m_next;
1688 	}
1689 	if (inp->inp_flags & INP_RECVTOS) {
1690 		*mp = sbcreatecontrol((caddr_t)&ip->ip_tos,
1691 		    sizeof(u_char), IP_RECVTOS, IPPROTO_IP);
1692 		if (*mp)
1693 			mp = &(*mp)->m_next;
1694 	}
1695 }
1696 
1697 /*
1698  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1699  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1700  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1701  * compiled.
1702  */
1703 static VNET_DEFINE(int, ip_rsvp_on);
1704 VNET_DEFINE(struct socket *, ip_rsvpd);
1705 
1706 #define	V_ip_rsvp_on		VNET(ip_rsvp_on)
1707 
1708 int
1709 ip_rsvp_init(struct socket *so)
1710 {
1711 
1712 	if (so->so_type != SOCK_RAW ||
1713 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1714 		return EOPNOTSUPP;
1715 
1716 	if (V_ip_rsvpd != NULL)
1717 		return EADDRINUSE;
1718 
1719 	V_ip_rsvpd = so;
1720 	/*
1721 	 * This may seem silly, but we need to be sure we don't over-increment
1722 	 * the RSVP counter, in case something slips up.
1723 	 */
1724 	if (!V_ip_rsvp_on) {
1725 		V_ip_rsvp_on = 1;
1726 		V_rsvp_on++;
1727 	}
1728 
1729 	return 0;
1730 }
1731 
1732 int
1733 ip_rsvp_done(void)
1734 {
1735 
1736 	V_ip_rsvpd = NULL;
1737 	/*
1738 	 * This may seem silly, but we need to be sure we don't over-decrement
1739 	 * the RSVP counter, in case something slips up.
1740 	 */
1741 	if (V_ip_rsvp_on) {
1742 		V_ip_rsvp_on = 0;
1743 		V_rsvp_on--;
1744 	}
1745 	return 0;
1746 }
1747 
1748 void
1749 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1750 {
1751 
1752 	if (rsvp_input_p) { /* call the real one if loaded */
1753 		rsvp_input_p(m, off);
1754 		return;
1755 	}
1756 
1757 	/* Can still get packets with rsvp_on = 0 if there is a local member
1758 	 * of the group to which the RSVP packet is addressed.  But in this
1759 	 * case we want to throw the packet away.
1760 	 */
1761 
1762 	if (!V_rsvp_on) {
1763 		m_freem(m);
1764 		return;
1765 	}
1766 
1767 	if (V_ip_rsvpd != NULL) {
1768 		rip_input(m, off);
1769 		return;
1770 	}
1771 	/* Drop the packet */
1772 	m_freem(m);
1773 }
1774