xref: /freebsd/sys/netinet/ip_input.c (revision 71099ec5097cff9b4a566e5474b7f214bd539e8a)
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/sdt.h>
53 #include <sys/syslog.h>
54 #include <sys/sysctl.h>
55 
56 #include <net/pfil.h>
57 #include <net/if.h>
58 #include <net/if_types.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/route.h>
62 #include <net/netisr.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_kdtrace.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip.h>
70 #include <netinet/in_pcb.h>
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_fw.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/ip_options.h>
75 #include <machine/in_cksum.h>
76 #include <netinet/ip_carp.h>
77 #ifdef IPSEC
78 #include <netinet/ip_ipsec.h>
79 #endif /* IPSEC */
80 
81 #include <sys/socketvar.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 #ifdef CTASSERT
86 CTASSERT(sizeof(struct ip) == 20);
87 #endif
88 
89 struct	rwlock in_ifaddr_lock;
90 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock");
91 
92 VNET_DEFINE(int, rsvp_on);
93 
94 VNET_DEFINE(int, ipforwarding);
95 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
96     &VNET_NAME(ipforwarding), 0,
97     "Enable IP forwarding between interfaces");
98 
99 static VNET_DEFINE(int, ipsendredirects) = 1;	/* XXX */
100 #define	V_ipsendredirects	VNET(ipsendredirects)
101 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
102     &VNET_NAME(ipsendredirects), 0,
103     "Enable sending IP redirects");
104 
105 static VNET_DEFINE(int, ip_keepfaith);
106 #define	V_ip_keepfaith		VNET(ip_keepfaith)
107 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
108     &VNET_NAME(ip_keepfaith), 0,
109     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
110 
111 static VNET_DEFINE(int, ip_sendsourcequench);
112 #define	V_ip_sendsourcequench	VNET(ip_sendsourcequench)
113 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
114     &VNET_NAME(ip_sendsourcequench), 0,
115     "Enable the transmission of source quench packets");
116 
117 VNET_DEFINE(int, ip_do_randomid);
118 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
119     &VNET_NAME(ip_do_randomid), 0,
120     "Assign random ip_id values");
121 
122 /*
123  * XXX - Setting ip_checkinterface mostly implements the receive side of
124  * the Strong ES model described in RFC 1122, but since the routing table
125  * and transmit implementation do not implement the Strong ES model,
126  * setting this to 1 results in an odd hybrid.
127  *
128  * XXX - ip_checkinterface currently must be disabled if you use ipnat
129  * to translate the destination address to another local interface.
130  *
131  * XXX - ip_checkinterface must be disabled if you add IP aliases
132  * to the loopback interface instead of the interface where the
133  * packets for those addresses are received.
134  */
135 static VNET_DEFINE(int, ip_checkinterface);
136 #define	V_ip_checkinterface	VNET(ip_checkinterface)
137 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
138     &VNET_NAME(ip_checkinterface), 0,
139     "Verify packet arrives on correct interface");
140 
141 VNET_DEFINE(struct pfil_head, inet_pfil_hook);	/* Packet filter hooks */
142 
143 static struct netisr_handler ip_nh = {
144 	.nh_name = "ip",
145 	.nh_handler = ip_input,
146 	.nh_proto = NETISR_IP,
147 	.nh_policy = NETISR_POLICY_FLOW,
148 };
149 
150 extern	struct domain inetdomain;
151 extern	struct protosw inetsw[];
152 u_char	ip_protox[IPPROTO_MAX];
153 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead);  /* first inet address */
154 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table  */
155 VNET_DEFINE(u_long, in_ifaddrhmask);		/* mask for hash table */
156 
157 static VNET_DEFINE(uma_zone_t, ipq_zone);
158 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]);
159 static struct mtx ipqlock;
160 
161 #define	V_ipq_zone		VNET(ipq_zone)
162 #define	V_ipq			VNET(ipq)
163 
164 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
165 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
166 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
167 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
168 
169 static void	maxnipq_update(void);
170 static void	ipq_zone_change(void *);
171 static void	ip_drain_locked(void);
172 
173 static VNET_DEFINE(int, maxnipq);  /* Administrative limit on # reass queues. */
174 static VNET_DEFINE(int, nipq);			/* Total # of reass queues */
175 #define	V_maxnipq		VNET(maxnipq)
176 #define	V_nipq			VNET(nipq)
177 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
178     &VNET_NAME(nipq), 0,
179     "Current number of IPv4 fragment reassembly queue entries");
180 
181 static VNET_DEFINE(int, maxfragsperpacket);
182 #define	V_maxfragsperpacket	VNET(maxfragsperpacket)
183 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
184     &VNET_NAME(maxfragsperpacket), 0,
185     "Maximum number of IPv4 fragments allowed per packet");
186 
187 #ifdef IPCTL_DEFMTU
188 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
189     &ip_mtu, 0, "Default MTU");
190 #endif
191 
192 #ifdef IPSTEALTH
193 VNET_DEFINE(int, ipstealth);
194 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
195     &VNET_NAME(ipstealth), 0,
196     "IP stealth mode, no TTL decrementation on forwarding");
197 #endif
198 
199 static void	ip_freef(struct ipqhead *, struct ipq *);
200 
201 /*
202  * IP statistics are stored in the "array" of counter(9)s.
203  */
204 VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat);
205 VNET_PCPUSTAT_SYSINIT(ipstat);
206 SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat,
207     "IP statistics (struct ipstat, netinet/ip_var.h)");
208 
209 #ifdef VIMAGE
210 VNET_PCPUSTAT_SYSUNINIT(ipstat);
211 #endif /* VIMAGE */
212 
213 /*
214  * Kernel module interface for updating ipstat.  The argument is an index
215  * into ipstat treated as an array.
216  */
217 void
218 kmod_ipstat_inc(int statnum)
219 {
220 
221 	counter_u64_add(VNET(ipstat)[statnum], 1);
222 }
223 
224 void
225 kmod_ipstat_dec(int statnum)
226 {
227 
228 	counter_u64_add(VNET(ipstat)[statnum], -1);
229 }
230 
231 static int
232 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
233 {
234 	int error, qlimit;
235 
236 	netisr_getqlimit(&ip_nh, &qlimit);
237 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
238 	if (error || !req->newptr)
239 		return (error);
240 	if (qlimit < 1)
241 		return (EINVAL);
242 	return (netisr_setqlimit(&ip_nh, qlimit));
243 }
244 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
245     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
246     "Maximum size of the IP input queue");
247 
248 static int
249 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
250 {
251 	u_int64_t qdrops_long;
252 	int error, qdrops;
253 
254 	netisr_getqdrops(&ip_nh, &qdrops_long);
255 	qdrops = qdrops_long;
256 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
257 	if (error || !req->newptr)
258 		return (error);
259 	if (qdrops != 0)
260 		return (EINVAL);
261 	netisr_clearqdrops(&ip_nh);
262 	return (0);
263 }
264 
265 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
266     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
267     "Number of packets dropped from the IP input queue");
268 
269 /*
270  * IP initialization: fill in IP protocol switch table.
271  * All protocols not implemented in kernel go to raw IP protocol handler.
272  */
273 void
274 ip_init(void)
275 {
276 	struct protosw *pr;
277 	int i;
278 
279 	V_ip_id = time_second & 0xffff;
280 
281 	TAILQ_INIT(&V_in_ifaddrhead);
282 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
283 
284 	/* Initialize IP reassembly queue. */
285 	for (i = 0; i < IPREASS_NHASH; i++)
286 		TAILQ_INIT(&V_ipq[i]);
287 	V_maxnipq = nmbclusters / 32;
288 	V_maxfragsperpacket = 16;
289 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
290 	    NULL, UMA_ALIGN_PTR, 0);
291 	maxnipq_update();
292 
293 	/* Initialize packet filter hooks. */
294 	V_inet_pfil_hook.ph_type = PFIL_TYPE_AF;
295 	V_inet_pfil_hook.ph_af = AF_INET;
296 	if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0)
297 		printf("%s: WARNING: unable to register pfil hook, "
298 			"error %d\n", __func__, i);
299 
300 	/* Skip initialization of globals for non-default instances. */
301 	if (!IS_DEFAULT_VNET(curvnet))
302 		return;
303 
304 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
305 	if (pr == NULL)
306 		panic("ip_init: PF_INET not found");
307 
308 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
309 	for (i = 0; i < IPPROTO_MAX; i++)
310 		ip_protox[i] = pr - inetsw;
311 	/*
312 	 * Cycle through IP protocols and put them into the appropriate place
313 	 * in ip_protox[].
314 	 */
315 	for (pr = inetdomain.dom_protosw;
316 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
317 		if (pr->pr_domain->dom_family == PF_INET &&
318 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
319 			/* Be careful to only index valid IP protocols. */
320 			if (pr->pr_protocol < IPPROTO_MAX)
321 				ip_protox[pr->pr_protocol] = pr - inetsw;
322 		}
323 
324 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
325 		NULL, EVENTHANDLER_PRI_ANY);
326 
327 	/* Initialize various other remaining things. */
328 	IPQ_LOCK_INIT();
329 	netisr_register(&ip_nh);
330 }
331 
332 #ifdef VIMAGE
333 void
334 ip_destroy(void)
335 {
336 	int i;
337 
338 	if ((i = pfil_head_unregister(&V_inet_pfil_hook)) != 0)
339 		printf("%s: WARNING: unable to unregister pfil hook, "
340 		    "error %d\n", __func__, i);
341 
342 	/* Cleanup in_ifaddr hash table; should be empty. */
343 	hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask);
344 
345 	IPQ_LOCK();
346 	ip_drain_locked();
347 	IPQ_UNLOCK();
348 
349 	uma_zdestroy(V_ipq_zone);
350 }
351 #endif
352 
353 /*
354  * Ip input routine.  Checksum and byte swap header.  If fragmented
355  * try to reassemble.  Process options.  Pass to next level.
356  */
357 void
358 ip_input(struct mbuf *m)
359 {
360 	struct ip *ip = NULL;
361 	struct in_ifaddr *ia = NULL;
362 	struct ifaddr *ifa;
363 	struct ifnet *ifp;
364 	int    checkif, hlen = 0;
365 	uint16_t sum, ip_len;
366 	int dchg = 0;				/* dest changed after fw */
367 	struct in_addr odst;			/* original dst address */
368 
369 	M_ASSERTPKTHDR(m);
370 
371 	if (m->m_flags & M_FASTFWD_OURS) {
372 		m->m_flags &= ~M_FASTFWD_OURS;
373 		/* Set up some basics that will be used later. */
374 		ip = mtod(m, struct ip *);
375 		hlen = ip->ip_hl << 2;
376 		ip_len = ntohs(ip->ip_len);
377 		goto ours;
378 	}
379 
380 	IPSTAT_INC(ips_total);
381 
382 	if (m->m_pkthdr.len < sizeof(struct ip))
383 		goto tooshort;
384 
385 	if (m->m_len < sizeof (struct ip) &&
386 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
387 		IPSTAT_INC(ips_toosmall);
388 		return;
389 	}
390 	ip = mtod(m, struct ip *);
391 
392 	if (ip->ip_v != IPVERSION) {
393 		IPSTAT_INC(ips_badvers);
394 		goto bad;
395 	}
396 
397 	hlen = ip->ip_hl << 2;
398 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
399 		IPSTAT_INC(ips_badhlen);
400 		goto bad;
401 	}
402 	if (hlen > m->m_len) {
403 		if ((m = m_pullup(m, hlen)) == NULL) {
404 			IPSTAT_INC(ips_badhlen);
405 			return;
406 		}
407 		ip = mtod(m, struct ip *);
408 	}
409 
410 	IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL);
411 
412 	/* 127/8 must not appear on wire - RFC1122 */
413 	ifp = m->m_pkthdr.rcvif;
414 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
415 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
416 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
417 			IPSTAT_INC(ips_badaddr);
418 			goto bad;
419 		}
420 	}
421 
422 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
423 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
424 	} else {
425 		if (hlen == sizeof(struct ip)) {
426 			sum = in_cksum_hdr(ip);
427 		} else {
428 			sum = in_cksum(m, hlen);
429 		}
430 	}
431 	if (sum) {
432 		IPSTAT_INC(ips_badsum);
433 		goto bad;
434 	}
435 
436 #ifdef ALTQ
437 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
438 		/* packet is dropped by traffic conditioner */
439 		return;
440 #endif
441 
442 	ip_len = ntohs(ip->ip_len);
443 	if (ip_len < hlen) {
444 		IPSTAT_INC(ips_badlen);
445 		goto bad;
446 	}
447 
448 	/*
449 	 * Check that the amount of data in the buffers
450 	 * is as at least much as the IP header would have us expect.
451 	 * Trim mbufs if longer than we expect.
452 	 * Drop packet if shorter than we expect.
453 	 */
454 	if (m->m_pkthdr.len < ip_len) {
455 tooshort:
456 		IPSTAT_INC(ips_tooshort);
457 		goto bad;
458 	}
459 	if (m->m_pkthdr.len > ip_len) {
460 		if (m->m_len == m->m_pkthdr.len) {
461 			m->m_len = ip_len;
462 			m->m_pkthdr.len = ip_len;
463 		} else
464 			m_adj(m, ip_len - m->m_pkthdr.len);
465 	}
466 #ifdef IPSEC
467 	/*
468 	 * Bypass packet filtering for packets previously handled by IPsec.
469 	 */
470 	if (ip_ipsec_filtertunnel(m))
471 		goto passin;
472 #endif /* IPSEC */
473 
474 	/*
475 	 * Run through list of hooks for input packets.
476 	 *
477 	 * NB: Beware of the destination address changing (e.g.
478 	 *     by NAT rewriting).  When this happens, tell
479 	 *     ip_forward to do the right thing.
480 	 */
481 
482 	/* Jump over all PFIL processing if hooks are not active. */
483 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
484 		goto passin;
485 
486 	odst = ip->ip_dst;
487 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
488 		return;
489 	if (m == NULL)			/* consumed by filter */
490 		return;
491 
492 	ip = mtod(m, struct ip *);
493 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
494 	ifp = m->m_pkthdr.rcvif;
495 
496 	if (m->m_flags & M_FASTFWD_OURS) {
497 		m->m_flags &= ~M_FASTFWD_OURS;
498 		goto ours;
499 	}
500 	if (m->m_flags & M_IP_NEXTHOP) {
501 		dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL);
502 		if (dchg != 0) {
503 			/*
504 			 * Directly ship the packet on.  This allows
505 			 * forwarding packets originally destined to us
506 			 * to some other directly connected host.
507 			 */
508 			ip_forward(m, 1);
509 			return;
510 		}
511 	}
512 passin:
513 
514 	/*
515 	 * Process options and, if not destined for us,
516 	 * ship it on.  ip_dooptions returns 1 when an
517 	 * error was detected (causing an icmp message
518 	 * to be sent and the original packet to be freed).
519 	 */
520 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
521 		return;
522 
523         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
524          * matter if it is destined to another node, or whether it is
525          * a multicast one, RSVP wants it! and prevents it from being forwarded
526          * anywhere else. Also checks if the rsvp daemon is running before
527 	 * grabbing the packet.
528          */
529 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
530 		goto ours;
531 
532 	/*
533 	 * Check our list of addresses, to see if the packet is for us.
534 	 * If we don't have any addresses, assume any unicast packet
535 	 * we receive might be for us (and let the upper layers deal
536 	 * with it).
537 	 */
538 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
539 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
540 		goto ours;
541 
542 	/*
543 	 * Enable a consistency check between the destination address
544 	 * and the arrival interface for a unicast packet (the RFC 1122
545 	 * strong ES model) if IP forwarding is disabled and the packet
546 	 * is not locally generated and the packet is not subject to
547 	 * 'ipfw fwd'.
548 	 *
549 	 * XXX - Checking also should be disabled if the destination
550 	 * address is ipnat'ed to a different interface.
551 	 *
552 	 * XXX - Checking is incompatible with IP aliases added
553 	 * to the loopback interface instead of the interface where
554 	 * the packets are received.
555 	 *
556 	 * XXX - This is the case for carp vhost IPs as well so we
557 	 * insert a workaround. If the packet got here, we already
558 	 * checked with carp_iamatch() and carp_forus().
559 	 */
560 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
561 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
562 	    ifp->if_carp == NULL && (dchg == 0);
563 
564 	/*
565 	 * Check for exact addresses in the hash bucket.
566 	 */
567 	/* IN_IFADDR_RLOCK(); */
568 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
569 		/*
570 		 * If the address matches, verify that the packet
571 		 * arrived via the correct interface if checking is
572 		 * enabled.
573 		 */
574 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
575 		    (!checkif || ia->ia_ifp == ifp)) {
576 			counter_u64_add(ia->ia_ifa.ifa_ipackets, 1);
577 			counter_u64_add(ia->ia_ifa.ifa_ibytes,
578 			    m->m_pkthdr.len);
579 			/* IN_IFADDR_RUNLOCK(); */
580 			goto ours;
581 		}
582 	}
583 	/* IN_IFADDR_RUNLOCK(); */
584 
585 	/*
586 	 * Check for broadcast addresses.
587 	 *
588 	 * Only accept broadcast packets that arrive via the matching
589 	 * interface.  Reception of forwarded directed broadcasts would
590 	 * be handled via ip_forward() and ether_output() with the loopback
591 	 * into the stack for SIMPLEX interfaces handled by ether_output().
592 	 */
593 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
594 		IF_ADDR_RLOCK(ifp);
595 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
596 			if (ifa->ifa_addr->sa_family != AF_INET)
597 				continue;
598 			ia = ifatoia(ifa);
599 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
600 			    ip->ip_dst.s_addr) {
601 				counter_u64_add(ia->ia_ifa.ifa_ipackets, 1);
602 				counter_u64_add(ia->ia_ifa.ifa_ibytes,
603 				    m->m_pkthdr.len);
604 				IF_ADDR_RUNLOCK(ifp);
605 				goto ours;
606 			}
607 #ifdef BOOTP_COMPAT
608 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
609 				counter_u64_add(ia->ia_ifa.ifa_ipackets, 1);
610 				counter_u64_add(ia->ia_ifa.ifa_ibytes,
611 				    m->m_pkthdr.len);
612 				IF_ADDR_RUNLOCK(ifp);
613 				goto ours;
614 			}
615 #endif
616 		}
617 		IF_ADDR_RUNLOCK(ifp);
618 		ia = NULL;
619 	}
620 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
621 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
622 		IPSTAT_INC(ips_cantforward);
623 		m_freem(m);
624 		return;
625 	}
626 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
627 		if (V_ip_mrouter) {
628 			/*
629 			 * If we are acting as a multicast router, all
630 			 * incoming multicast packets are passed to the
631 			 * kernel-level multicast forwarding function.
632 			 * The packet is returned (relatively) intact; if
633 			 * ip_mforward() returns a non-zero value, the packet
634 			 * must be discarded, else it may be accepted below.
635 			 */
636 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
637 				IPSTAT_INC(ips_cantforward);
638 				m_freem(m);
639 				return;
640 			}
641 
642 			/*
643 			 * The process-level routing daemon needs to receive
644 			 * all multicast IGMP packets, whether or not this
645 			 * host belongs to their destination groups.
646 			 */
647 			if (ip->ip_p == IPPROTO_IGMP)
648 				goto ours;
649 			IPSTAT_INC(ips_forward);
650 		}
651 		/*
652 		 * Assume the packet is for us, to avoid prematurely taking
653 		 * a lock on the in_multi hash. Protocols must perform
654 		 * their own filtering and update statistics accordingly.
655 		 */
656 		goto ours;
657 	}
658 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
659 		goto ours;
660 	if (ip->ip_dst.s_addr == INADDR_ANY)
661 		goto ours;
662 
663 	/*
664 	 * FAITH(Firewall Aided Internet Translator)
665 	 */
666 	if (ifp && ifp->if_type == IFT_FAITH) {
667 		if (V_ip_keepfaith) {
668 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
669 				goto ours;
670 		}
671 		m_freem(m);
672 		return;
673 	}
674 
675 	/*
676 	 * Not for us; forward if possible and desirable.
677 	 */
678 	if (V_ipforwarding == 0) {
679 		IPSTAT_INC(ips_cantforward);
680 		m_freem(m);
681 	} else {
682 #ifdef IPSEC
683 		if (ip_ipsec_fwd(m))
684 			goto bad;
685 #endif /* IPSEC */
686 		ip_forward(m, dchg);
687 	}
688 	return;
689 
690 ours:
691 #ifdef IPSTEALTH
692 	/*
693 	 * IPSTEALTH: Process non-routing options only
694 	 * if the packet is destined for us.
695 	 */
696 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1))
697 		return;
698 #endif /* IPSTEALTH */
699 
700 	/*
701 	 * Attempt reassembly; if it succeeds, proceed.
702 	 * ip_reass() will return a different mbuf.
703 	 */
704 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
705 		m = ip_reass(m);
706 		if (m == NULL)
707 			return;
708 		ip = mtod(m, struct ip *);
709 		/* Get the header length of the reassembled packet */
710 		hlen = ip->ip_hl << 2;
711 	}
712 
713 #ifdef IPSEC
714 	/*
715 	 * enforce IPsec policy checking if we are seeing last header.
716 	 * note that we do not visit this with protocols with pcb layer
717 	 * code - like udp/tcp/raw ip.
718 	 */
719 	if (ip_ipsec_input(m))
720 		goto bad;
721 #endif /* IPSEC */
722 
723 	/*
724 	 * Switch out to protocol's input routine.
725 	 */
726 	IPSTAT_INC(ips_delivered);
727 
728 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
729 	return;
730 bad:
731 	m_freem(m);
732 }
733 
734 /*
735  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
736  * max has slightly different semantics than the sysctl, for historical
737  * reasons.
738  */
739 static void
740 maxnipq_update(void)
741 {
742 
743 	/*
744 	 * -1 for unlimited allocation.
745 	 */
746 	if (V_maxnipq < 0)
747 		uma_zone_set_max(V_ipq_zone, 0);
748 	/*
749 	 * Positive number for specific bound.
750 	 */
751 	if (V_maxnipq > 0)
752 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
753 	/*
754 	 * Zero specifies no further fragment queue allocation -- set the
755 	 * bound very low, but rely on implementation elsewhere to actually
756 	 * prevent allocation and reclaim current queues.
757 	 */
758 	if (V_maxnipq == 0)
759 		uma_zone_set_max(V_ipq_zone, 1);
760 }
761 
762 static void
763 ipq_zone_change(void *tag)
764 {
765 
766 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
767 		V_maxnipq = nmbclusters / 32;
768 		maxnipq_update();
769 	}
770 }
771 
772 static int
773 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
774 {
775 	int error, i;
776 
777 	i = V_maxnipq;
778 	error = sysctl_handle_int(oidp, &i, 0, req);
779 	if (error || !req->newptr)
780 		return (error);
781 
782 	/*
783 	 * XXXRW: Might be a good idea to sanity check the argument and place
784 	 * an extreme upper bound.
785 	 */
786 	if (i < -1)
787 		return (EINVAL);
788 	V_maxnipq = i;
789 	maxnipq_update();
790 	return (0);
791 }
792 
793 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
794     NULL, 0, sysctl_maxnipq, "I",
795     "Maximum number of IPv4 fragment reassembly queue entries");
796 
797 /*
798  * Take incoming datagram fragment and try to reassemble it into
799  * whole datagram.  If the argument is the first fragment or one
800  * in between the function will return NULL and store the mbuf
801  * in the fragment chain.  If the argument is the last fragment
802  * the packet will be reassembled and the pointer to the new
803  * mbuf returned for further processing.  Only m_tags attached
804  * to the first packet/fragment are preserved.
805  * The IP header is *NOT* adjusted out of iplen.
806  */
807 struct mbuf *
808 ip_reass(struct mbuf *m)
809 {
810 	struct ip *ip;
811 	struct mbuf *p, *q, *nq, *t;
812 	struct ipq *fp = NULL;
813 	struct ipqhead *head;
814 	int i, hlen, next;
815 	u_int8_t ecn, ecn0;
816 	u_short hash;
817 
818 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
819 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
820 		IPSTAT_INC(ips_fragments);
821 		IPSTAT_INC(ips_fragdropped);
822 		m_freem(m);
823 		return (NULL);
824 	}
825 
826 	ip = mtod(m, struct ip *);
827 	hlen = ip->ip_hl << 2;
828 
829 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
830 	head = &V_ipq[hash];
831 	IPQ_LOCK();
832 
833 	/*
834 	 * Look for queue of fragments
835 	 * of this datagram.
836 	 */
837 	TAILQ_FOREACH(fp, head, ipq_list)
838 		if (ip->ip_id == fp->ipq_id &&
839 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
840 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
841 #ifdef MAC
842 		    mac_ipq_match(m, fp) &&
843 #endif
844 		    ip->ip_p == fp->ipq_p)
845 			goto found;
846 
847 	fp = NULL;
848 
849 	/*
850 	 * Attempt to trim the number of allocated fragment queues if it
851 	 * exceeds the administrative limit.
852 	 */
853 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
854 		/*
855 		 * drop something from the tail of the current queue
856 		 * before proceeding further
857 		 */
858 		struct ipq *q = TAILQ_LAST(head, ipqhead);
859 		if (q == NULL) {   /* gak */
860 			for (i = 0; i < IPREASS_NHASH; i++) {
861 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
862 				if (r) {
863 					IPSTAT_ADD(ips_fragtimeout,
864 					    r->ipq_nfrags);
865 					ip_freef(&V_ipq[i], r);
866 					break;
867 				}
868 			}
869 		} else {
870 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
871 			ip_freef(head, q);
872 		}
873 	}
874 
875 found:
876 	/*
877 	 * Adjust ip_len to not reflect header,
878 	 * convert offset of this to bytes.
879 	 */
880 	ip->ip_len = htons(ntohs(ip->ip_len) - hlen);
881 	if (ip->ip_off & htons(IP_MF)) {
882 		/*
883 		 * Make sure that fragments have a data length
884 		 * that's a non-zero multiple of 8 bytes.
885 		 */
886 		if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) {
887 			IPSTAT_INC(ips_toosmall); /* XXX */
888 			goto dropfrag;
889 		}
890 		m->m_flags |= M_IP_FRAG;
891 	} else
892 		m->m_flags &= ~M_IP_FRAG;
893 	ip->ip_off = htons(ntohs(ip->ip_off) << 3);
894 
895 	/*
896 	 * Attempt reassembly; if it succeeds, proceed.
897 	 * ip_reass() will return a different mbuf.
898 	 */
899 	IPSTAT_INC(ips_fragments);
900 	m->m_pkthdr.PH_loc.ptr = ip;
901 
902 	/* Previous ip_reass() started here. */
903 	/*
904 	 * Presence of header sizes in mbufs
905 	 * would confuse code below.
906 	 */
907 	m->m_data += hlen;
908 	m->m_len -= hlen;
909 
910 	/*
911 	 * If first fragment to arrive, create a reassembly queue.
912 	 */
913 	if (fp == NULL) {
914 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
915 		if (fp == NULL)
916 			goto dropfrag;
917 #ifdef MAC
918 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
919 			uma_zfree(V_ipq_zone, fp);
920 			fp = NULL;
921 			goto dropfrag;
922 		}
923 		mac_ipq_create(m, fp);
924 #endif
925 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
926 		V_nipq++;
927 		fp->ipq_nfrags = 1;
928 		fp->ipq_ttl = IPFRAGTTL;
929 		fp->ipq_p = ip->ip_p;
930 		fp->ipq_id = ip->ip_id;
931 		fp->ipq_src = ip->ip_src;
932 		fp->ipq_dst = ip->ip_dst;
933 		fp->ipq_frags = m;
934 		m->m_nextpkt = NULL;
935 		goto done;
936 	} else {
937 		fp->ipq_nfrags++;
938 #ifdef MAC
939 		mac_ipq_update(m, fp);
940 #endif
941 	}
942 
943 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.PH_loc.ptr))
944 
945 	/*
946 	 * Handle ECN by comparing this segment with the first one;
947 	 * if CE is set, do not lose CE.
948 	 * drop if CE and not-ECT are mixed for the same packet.
949 	 */
950 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
951 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
952 	if (ecn == IPTOS_ECN_CE) {
953 		if (ecn0 == IPTOS_ECN_NOTECT)
954 			goto dropfrag;
955 		if (ecn0 != IPTOS_ECN_CE)
956 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
957 	}
958 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
959 		goto dropfrag;
960 
961 	/*
962 	 * Find a segment which begins after this one does.
963 	 */
964 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
965 		if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off))
966 			break;
967 
968 	/*
969 	 * If there is a preceding segment, it may provide some of
970 	 * our data already.  If so, drop the data from the incoming
971 	 * segment.  If it provides all of our data, drop us, otherwise
972 	 * stick new segment in the proper place.
973 	 *
974 	 * If some of the data is dropped from the preceding
975 	 * segment, then it's checksum is invalidated.
976 	 */
977 	if (p) {
978 		i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) -
979 		    ntohs(ip->ip_off);
980 		if (i > 0) {
981 			if (i >= ntohs(ip->ip_len))
982 				goto dropfrag;
983 			m_adj(m, i);
984 			m->m_pkthdr.csum_flags = 0;
985 			ip->ip_off = htons(ntohs(ip->ip_off) + i);
986 			ip->ip_len = htons(ntohs(ip->ip_len) - i);
987 		}
988 		m->m_nextpkt = p->m_nextpkt;
989 		p->m_nextpkt = m;
990 	} else {
991 		m->m_nextpkt = fp->ipq_frags;
992 		fp->ipq_frags = m;
993 	}
994 
995 	/*
996 	 * While we overlap succeeding segments trim them or,
997 	 * if they are completely covered, dequeue them.
998 	 */
999 	for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) >
1000 	    ntohs(GETIP(q)->ip_off); q = nq) {
1001 		i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) -
1002 		    ntohs(GETIP(q)->ip_off);
1003 		if (i < ntohs(GETIP(q)->ip_len)) {
1004 			GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i);
1005 			GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i);
1006 			m_adj(q, i);
1007 			q->m_pkthdr.csum_flags = 0;
1008 			break;
1009 		}
1010 		nq = q->m_nextpkt;
1011 		m->m_nextpkt = nq;
1012 		IPSTAT_INC(ips_fragdropped);
1013 		fp->ipq_nfrags--;
1014 		m_freem(q);
1015 	}
1016 
1017 	/*
1018 	 * Check for complete reassembly and perform frag per packet
1019 	 * limiting.
1020 	 *
1021 	 * Frag limiting is performed here so that the nth frag has
1022 	 * a chance to complete the packet before we drop the packet.
1023 	 * As a result, n+1 frags are actually allowed per packet, but
1024 	 * only n will ever be stored. (n = maxfragsperpacket.)
1025 	 *
1026 	 */
1027 	next = 0;
1028 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1029 		if (ntohs(GETIP(q)->ip_off) != next) {
1030 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1031 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1032 				ip_freef(head, fp);
1033 			}
1034 			goto done;
1035 		}
1036 		next += ntohs(GETIP(q)->ip_len);
1037 	}
1038 	/* Make sure the last packet didn't have the IP_MF flag */
1039 	if (p->m_flags & M_IP_FRAG) {
1040 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1041 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1042 			ip_freef(head, fp);
1043 		}
1044 		goto done;
1045 	}
1046 
1047 	/*
1048 	 * Reassembly is complete.  Make sure the packet is a sane size.
1049 	 */
1050 	q = fp->ipq_frags;
1051 	ip = GETIP(q);
1052 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1053 		IPSTAT_INC(ips_toolong);
1054 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1055 		ip_freef(head, fp);
1056 		goto done;
1057 	}
1058 
1059 	/*
1060 	 * Concatenate fragments.
1061 	 */
1062 	m = q;
1063 	t = m->m_next;
1064 	m->m_next = NULL;
1065 	m_cat(m, t);
1066 	nq = q->m_nextpkt;
1067 	q->m_nextpkt = NULL;
1068 	for (q = nq; q != NULL; q = nq) {
1069 		nq = q->m_nextpkt;
1070 		q->m_nextpkt = NULL;
1071 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1072 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1073 		m_cat(m, q);
1074 	}
1075 	/*
1076 	 * In order to do checksumming faster we do 'end-around carry' here
1077 	 * (and not in for{} loop), though it implies we are not going to
1078 	 * reassemble more than 64k fragments.
1079 	 */
1080 	m->m_pkthdr.csum_data =
1081 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1082 #ifdef MAC
1083 	mac_ipq_reassemble(fp, m);
1084 	mac_ipq_destroy(fp);
1085 #endif
1086 
1087 	/*
1088 	 * Create header for new ip packet by modifying header of first
1089 	 * packet;  dequeue and discard fragment reassembly header.
1090 	 * Make header visible.
1091 	 */
1092 	ip->ip_len = htons((ip->ip_hl << 2) + next);
1093 	ip->ip_src = fp->ipq_src;
1094 	ip->ip_dst = fp->ipq_dst;
1095 	TAILQ_REMOVE(head, fp, ipq_list);
1096 	V_nipq--;
1097 	uma_zfree(V_ipq_zone, fp);
1098 	m->m_len += (ip->ip_hl << 2);
1099 	m->m_data -= (ip->ip_hl << 2);
1100 	/* some debugging cruft by sklower, below, will go away soon */
1101 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1102 		m_fixhdr(m);
1103 	IPSTAT_INC(ips_reassembled);
1104 	IPQ_UNLOCK();
1105 	return (m);
1106 
1107 dropfrag:
1108 	IPSTAT_INC(ips_fragdropped);
1109 	if (fp != NULL)
1110 		fp->ipq_nfrags--;
1111 	m_freem(m);
1112 done:
1113 	IPQ_UNLOCK();
1114 	return (NULL);
1115 
1116 #undef GETIP
1117 }
1118 
1119 /*
1120  * Free a fragment reassembly header and all
1121  * associated datagrams.
1122  */
1123 static void
1124 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1125 {
1126 	struct mbuf *q;
1127 
1128 	IPQ_LOCK_ASSERT();
1129 
1130 	while (fp->ipq_frags) {
1131 		q = fp->ipq_frags;
1132 		fp->ipq_frags = q->m_nextpkt;
1133 		m_freem(q);
1134 	}
1135 	TAILQ_REMOVE(fhp, fp, ipq_list);
1136 	uma_zfree(V_ipq_zone, fp);
1137 	V_nipq--;
1138 }
1139 
1140 /*
1141  * IP timer processing;
1142  * if a timer expires on a reassembly
1143  * queue, discard it.
1144  */
1145 void
1146 ip_slowtimo(void)
1147 {
1148 	VNET_ITERATOR_DECL(vnet_iter);
1149 	struct ipq *fp;
1150 	int i;
1151 
1152 	VNET_LIST_RLOCK_NOSLEEP();
1153 	IPQ_LOCK();
1154 	VNET_FOREACH(vnet_iter) {
1155 		CURVNET_SET(vnet_iter);
1156 		for (i = 0; i < IPREASS_NHASH; i++) {
1157 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1158 				struct ipq *fpp;
1159 
1160 				fpp = fp;
1161 				fp = TAILQ_NEXT(fp, ipq_list);
1162 				if(--fpp->ipq_ttl == 0) {
1163 					IPSTAT_ADD(ips_fragtimeout,
1164 					    fpp->ipq_nfrags);
1165 					ip_freef(&V_ipq[i], fpp);
1166 				}
1167 			}
1168 		}
1169 		/*
1170 		 * If we are over the maximum number of fragments
1171 		 * (due to the limit being lowered), drain off
1172 		 * enough to get down to the new limit.
1173 		 */
1174 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1175 			for (i = 0; i < IPREASS_NHASH; i++) {
1176 				while (V_nipq > V_maxnipq &&
1177 				    !TAILQ_EMPTY(&V_ipq[i])) {
1178 					IPSTAT_ADD(ips_fragdropped,
1179 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1180 					ip_freef(&V_ipq[i],
1181 					    TAILQ_FIRST(&V_ipq[i]));
1182 				}
1183 			}
1184 		}
1185 		CURVNET_RESTORE();
1186 	}
1187 	IPQ_UNLOCK();
1188 	VNET_LIST_RUNLOCK_NOSLEEP();
1189 }
1190 
1191 /*
1192  * Drain off all datagram fragments.
1193  */
1194 static void
1195 ip_drain_locked(void)
1196 {
1197 	int     i;
1198 
1199 	IPQ_LOCK_ASSERT();
1200 
1201 	for (i = 0; i < IPREASS_NHASH; i++) {
1202 		while(!TAILQ_EMPTY(&V_ipq[i])) {
1203 			IPSTAT_ADD(ips_fragdropped,
1204 			    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1205 			ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1206 		}
1207 	}
1208 }
1209 
1210 void
1211 ip_drain(void)
1212 {
1213 	VNET_ITERATOR_DECL(vnet_iter);
1214 
1215 	VNET_LIST_RLOCK_NOSLEEP();
1216 	IPQ_LOCK();
1217 	VNET_FOREACH(vnet_iter) {
1218 		CURVNET_SET(vnet_iter);
1219 		ip_drain_locked();
1220 		CURVNET_RESTORE();
1221 	}
1222 	IPQ_UNLOCK();
1223 	VNET_LIST_RUNLOCK_NOSLEEP();
1224 	in_rtqdrain();
1225 }
1226 
1227 /*
1228  * The protocol to be inserted into ip_protox[] must be already registered
1229  * in inetsw[], either statically or through pf_proto_register().
1230  */
1231 int
1232 ipproto_register(short ipproto)
1233 {
1234 	struct protosw *pr;
1235 
1236 	/* Sanity checks. */
1237 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1238 		return (EPROTONOSUPPORT);
1239 
1240 	/*
1241 	 * The protocol slot must not be occupied by another protocol
1242 	 * already.  An index pointing to IPPROTO_RAW is unused.
1243 	 */
1244 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1245 	if (pr == NULL)
1246 		return (EPFNOSUPPORT);
1247 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1248 		return (EEXIST);
1249 
1250 	/* Find the protocol position in inetsw[] and set the index. */
1251 	for (pr = inetdomain.dom_protosw;
1252 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1253 		if (pr->pr_domain->dom_family == PF_INET &&
1254 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1255 			ip_protox[pr->pr_protocol] = pr - inetsw;
1256 			return (0);
1257 		}
1258 	}
1259 	return (EPROTONOSUPPORT);
1260 }
1261 
1262 int
1263 ipproto_unregister(short ipproto)
1264 {
1265 	struct protosw *pr;
1266 
1267 	/* Sanity checks. */
1268 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1269 		return (EPROTONOSUPPORT);
1270 
1271 	/* Check if the protocol was indeed registered. */
1272 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1273 	if (pr == NULL)
1274 		return (EPFNOSUPPORT);
1275 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1276 		return (ENOENT);
1277 
1278 	/* Reset the protocol slot to IPPROTO_RAW. */
1279 	ip_protox[ipproto] = pr - inetsw;
1280 	return (0);
1281 }
1282 
1283 /*
1284  * Given address of next destination (final or next hop), return (referenced)
1285  * internet address info of interface to be used to get there.
1286  */
1287 struct in_ifaddr *
1288 ip_rtaddr(struct in_addr dst, u_int fibnum)
1289 {
1290 	struct route sro;
1291 	struct sockaddr_in *sin;
1292 	struct in_ifaddr *ia;
1293 
1294 	bzero(&sro, sizeof(sro));
1295 	sin = (struct sockaddr_in *)&sro.ro_dst;
1296 	sin->sin_family = AF_INET;
1297 	sin->sin_len = sizeof(*sin);
1298 	sin->sin_addr = dst;
1299 	in_rtalloc_ign(&sro, 0, fibnum);
1300 
1301 	if (sro.ro_rt == NULL)
1302 		return (NULL);
1303 
1304 	ia = ifatoia(sro.ro_rt->rt_ifa);
1305 	ifa_ref(&ia->ia_ifa);
1306 	RTFREE(sro.ro_rt);
1307 	return (ia);
1308 }
1309 
1310 u_char inetctlerrmap[PRC_NCMDS] = {
1311 	0,		0,		0,		0,
1312 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1313 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1314 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1315 	0,		0,		EHOSTUNREACH,	0,
1316 	ENOPROTOOPT,	ECONNREFUSED
1317 };
1318 
1319 /*
1320  * Forward a packet.  If some error occurs return the sender
1321  * an icmp packet.  Note we can't always generate a meaningful
1322  * icmp message because icmp doesn't have a large enough repertoire
1323  * of codes and types.
1324  *
1325  * If not forwarding, just drop the packet.  This could be confusing
1326  * if ipforwarding was zero but some routing protocol was advancing
1327  * us as a gateway to somewhere.  However, we must let the routing
1328  * protocol deal with that.
1329  *
1330  * The srcrt parameter indicates whether the packet is being forwarded
1331  * via a source route.
1332  */
1333 void
1334 ip_forward(struct mbuf *m, int srcrt)
1335 {
1336 	struct ip *ip = mtod(m, struct ip *);
1337 	struct in_ifaddr *ia;
1338 	struct mbuf *mcopy;
1339 	struct in_addr dest;
1340 	struct route ro;
1341 	int error, type = 0, code = 0, mtu = 0;
1342 
1343 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1344 		IPSTAT_INC(ips_cantforward);
1345 		m_freem(m);
1346 		return;
1347 	}
1348 #ifdef IPSTEALTH
1349 	if (!V_ipstealth) {
1350 #endif
1351 		if (ip->ip_ttl <= IPTTLDEC) {
1352 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1353 			    0, 0);
1354 			return;
1355 		}
1356 #ifdef IPSTEALTH
1357 	}
1358 #endif
1359 
1360 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1361 #ifndef IPSEC
1362 	/*
1363 	 * 'ia' may be NULL if there is no route for this destination.
1364 	 * In case of IPsec, Don't discard it just yet, but pass it to
1365 	 * ip_output in case of outgoing IPsec policy.
1366 	 */
1367 	if (!srcrt && ia == NULL) {
1368 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1369 		return;
1370 	}
1371 #endif
1372 
1373 	/*
1374 	 * Save the IP header and at most 8 bytes of the payload,
1375 	 * in case we need to generate an ICMP message to the src.
1376 	 *
1377 	 * XXX this can be optimized a lot by saving the data in a local
1378 	 * buffer on the stack (72 bytes at most), and only allocating the
1379 	 * mbuf if really necessary. The vast majority of the packets
1380 	 * are forwarded without having to send an ICMP back (either
1381 	 * because unnecessary, or because rate limited), so we are
1382 	 * really we are wasting a lot of work here.
1383 	 *
1384 	 * We don't use m_copy() because it might return a reference
1385 	 * to a shared cluster. Both this function and ip_output()
1386 	 * assume exclusive access to the IP header in `m', so any
1387 	 * data in a cluster may change before we reach icmp_error().
1388 	 */
1389 	mcopy = m_gethdr(M_NOWAIT, m->m_type);
1390 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) {
1391 		/*
1392 		 * It's probably ok if the pkthdr dup fails (because
1393 		 * the deep copy of the tag chain failed), but for now
1394 		 * be conservative and just discard the copy since
1395 		 * code below may some day want the tags.
1396 		 */
1397 		m_free(mcopy);
1398 		mcopy = NULL;
1399 	}
1400 	if (mcopy != NULL) {
1401 		mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy));
1402 		mcopy->m_pkthdr.len = mcopy->m_len;
1403 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1404 	}
1405 
1406 #ifdef IPSTEALTH
1407 	if (!V_ipstealth) {
1408 #endif
1409 		ip->ip_ttl -= IPTTLDEC;
1410 #ifdef IPSTEALTH
1411 	}
1412 #endif
1413 
1414 	/*
1415 	 * If forwarding packet using same interface that it came in on,
1416 	 * perhaps should send a redirect to sender to shortcut a hop.
1417 	 * Only send redirect if source is sending directly to us,
1418 	 * and if packet was not source routed (or has any options).
1419 	 * Also, don't send redirect if forwarding using a default route
1420 	 * or a route modified by a redirect.
1421 	 */
1422 	dest.s_addr = 0;
1423 	if (!srcrt && V_ipsendredirects &&
1424 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1425 		struct sockaddr_in *sin;
1426 		struct rtentry *rt;
1427 
1428 		bzero(&ro, sizeof(ro));
1429 		sin = (struct sockaddr_in *)&ro.ro_dst;
1430 		sin->sin_family = AF_INET;
1431 		sin->sin_len = sizeof(*sin);
1432 		sin->sin_addr = ip->ip_dst;
1433 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1434 
1435 		rt = ro.ro_rt;
1436 
1437 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1438 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1439 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1440 			u_long src = ntohl(ip->ip_src.s_addr);
1441 
1442 			if (RTA(rt) &&
1443 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1444 				if (rt->rt_flags & RTF_GATEWAY)
1445 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1446 				else
1447 					dest.s_addr = ip->ip_dst.s_addr;
1448 				/* Router requirements says to only send host redirects */
1449 				type = ICMP_REDIRECT;
1450 				code = ICMP_REDIRECT_HOST;
1451 			}
1452 		}
1453 		if (rt)
1454 			RTFREE(rt);
1455 	}
1456 
1457 	/*
1458 	 * Try to cache the route MTU from ip_output so we can consider it for
1459 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1460 	 */
1461 	bzero(&ro, sizeof(ro));
1462 
1463 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1464 
1465 	if (error == EMSGSIZE && ro.ro_rt)
1466 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1467 	RO_RTFREE(&ro);
1468 
1469 	if (error)
1470 		IPSTAT_INC(ips_cantforward);
1471 	else {
1472 		IPSTAT_INC(ips_forward);
1473 		if (type)
1474 			IPSTAT_INC(ips_redirectsent);
1475 		else {
1476 			if (mcopy)
1477 				m_freem(mcopy);
1478 			if (ia != NULL)
1479 				ifa_free(&ia->ia_ifa);
1480 			return;
1481 		}
1482 	}
1483 	if (mcopy == NULL) {
1484 		if (ia != NULL)
1485 			ifa_free(&ia->ia_ifa);
1486 		return;
1487 	}
1488 
1489 	switch (error) {
1490 
1491 	case 0:				/* forwarded, but need redirect */
1492 		/* type, code set above */
1493 		break;
1494 
1495 	case ENETUNREACH:
1496 	case EHOSTUNREACH:
1497 	case ENETDOWN:
1498 	case EHOSTDOWN:
1499 	default:
1500 		type = ICMP_UNREACH;
1501 		code = ICMP_UNREACH_HOST;
1502 		break;
1503 
1504 	case EMSGSIZE:
1505 		type = ICMP_UNREACH;
1506 		code = ICMP_UNREACH_NEEDFRAG;
1507 
1508 #ifdef IPSEC
1509 		/*
1510 		 * If IPsec is configured for this path,
1511 		 * override any possibly mtu value set by ip_output.
1512 		 */
1513 		mtu = ip_ipsec_mtu(mcopy, mtu);
1514 #endif /* IPSEC */
1515 		/*
1516 		 * If the MTU was set before make sure we are below the
1517 		 * interface MTU.
1518 		 * If the MTU wasn't set before use the interface mtu or
1519 		 * fall back to the next smaller mtu step compared to the
1520 		 * current packet size.
1521 		 */
1522 		if (mtu != 0) {
1523 			if (ia != NULL)
1524 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1525 		} else {
1526 			if (ia != NULL)
1527 				mtu = ia->ia_ifp->if_mtu;
1528 			else
1529 				mtu = ip_next_mtu(ntohs(ip->ip_len), 0);
1530 		}
1531 		IPSTAT_INC(ips_cantfrag);
1532 		break;
1533 
1534 	case ENOBUFS:
1535 		/*
1536 		 * A router should not generate ICMP_SOURCEQUENCH as
1537 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1538 		 * Source quench could be a big problem under DoS attacks,
1539 		 * or if the underlying interface is rate-limited.
1540 		 * Those who need source quench packets may re-enable them
1541 		 * via the net.inet.ip.sendsourcequench sysctl.
1542 		 */
1543 		if (V_ip_sendsourcequench == 0) {
1544 			m_freem(mcopy);
1545 			if (ia != NULL)
1546 				ifa_free(&ia->ia_ifa);
1547 			return;
1548 		} else {
1549 			type = ICMP_SOURCEQUENCH;
1550 			code = 0;
1551 		}
1552 		break;
1553 
1554 	case EACCES:			/* ipfw denied packet */
1555 		m_freem(mcopy);
1556 		if (ia != NULL)
1557 			ifa_free(&ia->ia_ifa);
1558 		return;
1559 	}
1560 	if (ia != NULL)
1561 		ifa_free(&ia->ia_ifa);
1562 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1563 }
1564 
1565 void
1566 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1567     struct mbuf *m)
1568 {
1569 
1570 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1571 		struct bintime bt;
1572 
1573 		bintime(&bt);
1574 		if (inp->inp_socket->so_options & SO_BINTIME) {
1575 			*mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt),
1576 			    SCM_BINTIME, SOL_SOCKET);
1577 			if (*mp)
1578 				mp = &(*mp)->m_next;
1579 		}
1580 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1581 			struct timeval tv;
1582 
1583 			bintime2timeval(&bt, &tv);
1584 			*mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv),
1585 			    SCM_TIMESTAMP, SOL_SOCKET);
1586 			if (*mp)
1587 				mp = &(*mp)->m_next;
1588 		}
1589 	}
1590 	if (inp->inp_flags & INP_RECVDSTADDR) {
1591 		*mp = sbcreatecontrol((caddr_t)&ip->ip_dst,
1592 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1593 		if (*mp)
1594 			mp = &(*mp)->m_next;
1595 	}
1596 	if (inp->inp_flags & INP_RECVTTL) {
1597 		*mp = sbcreatecontrol((caddr_t)&ip->ip_ttl,
1598 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1599 		if (*mp)
1600 			mp = &(*mp)->m_next;
1601 	}
1602 #ifdef notyet
1603 	/* XXX
1604 	 * Moving these out of udp_input() made them even more broken
1605 	 * than they already were.
1606 	 */
1607 	/* options were tossed already */
1608 	if (inp->inp_flags & INP_RECVOPTS) {
1609 		*mp = sbcreatecontrol((caddr_t)opts_deleted_above,
1610 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1611 		if (*mp)
1612 			mp = &(*mp)->m_next;
1613 	}
1614 	/* ip_srcroute doesn't do what we want here, need to fix */
1615 	if (inp->inp_flags & INP_RECVRETOPTS) {
1616 		*mp = sbcreatecontrol((caddr_t)ip_srcroute(m),
1617 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1618 		if (*mp)
1619 			mp = &(*mp)->m_next;
1620 	}
1621 #endif
1622 	if (inp->inp_flags & INP_RECVIF) {
1623 		struct ifnet *ifp;
1624 		struct sdlbuf {
1625 			struct sockaddr_dl sdl;
1626 			u_char	pad[32];
1627 		} sdlbuf;
1628 		struct sockaddr_dl *sdp;
1629 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1630 
1631 		if ((ifp = m->m_pkthdr.rcvif) &&
1632 		    ifp->if_index && ifp->if_index <= V_if_index) {
1633 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1634 			/*
1635 			 * Change our mind and don't try copy.
1636 			 */
1637 			if (sdp->sdl_family != AF_LINK ||
1638 			    sdp->sdl_len > sizeof(sdlbuf)) {
1639 				goto makedummy;
1640 			}
1641 			bcopy(sdp, sdl2, sdp->sdl_len);
1642 		} else {
1643 makedummy:
1644 			sdl2->sdl_len =
1645 			    offsetof(struct sockaddr_dl, sdl_data[0]);
1646 			sdl2->sdl_family = AF_LINK;
1647 			sdl2->sdl_index = 0;
1648 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1649 		}
1650 		*mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len,
1651 		    IP_RECVIF, IPPROTO_IP);
1652 		if (*mp)
1653 			mp = &(*mp)->m_next;
1654 	}
1655 	if (inp->inp_flags & INP_RECVTOS) {
1656 		*mp = sbcreatecontrol((caddr_t)&ip->ip_tos,
1657 		    sizeof(u_char), IP_RECVTOS, IPPROTO_IP);
1658 		if (*mp)
1659 			mp = &(*mp)->m_next;
1660 	}
1661 }
1662 
1663 /*
1664  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1665  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1666  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1667  * compiled.
1668  */
1669 static VNET_DEFINE(int, ip_rsvp_on);
1670 VNET_DEFINE(struct socket *, ip_rsvpd);
1671 
1672 #define	V_ip_rsvp_on		VNET(ip_rsvp_on)
1673 
1674 int
1675 ip_rsvp_init(struct socket *so)
1676 {
1677 
1678 	if (so->so_type != SOCK_RAW ||
1679 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1680 		return EOPNOTSUPP;
1681 
1682 	if (V_ip_rsvpd != NULL)
1683 		return EADDRINUSE;
1684 
1685 	V_ip_rsvpd = so;
1686 	/*
1687 	 * This may seem silly, but we need to be sure we don't over-increment
1688 	 * the RSVP counter, in case something slips up.
1689 	 */
1690 	if (!V_ip_rsvp_on) {
1691 		V_ip_rsvp_on = 1;
1692 		V_rsvp_on++;
1693 	}
1694 
1695 	return 0;
1696 }
1697 
1698 int
1699 ip_rsvp_done(void)
1700 {
1701 
1702 	V_ip_rsvpd = NULL;
1703 	/*
1704 	 * This may seem silly, but we need to be sure we don't over-decrement
1705 	 * the RSVP counter, in case something slips up.
1706 	 */
1707 	if (V_ip_rsvp_on) {
1708 		V_ip_rsvp_on = 0;
1709 		V_rsvp_on--;
1710 	}
1711 	return 0;
1712 }
1713 
1714 void
1715 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1716 {
1717 
1718 	if (rsvp_input_p) { /* call the real one if loaded */
1719 		rsvp_input_p(m, off);
1720 		return;
1721 	}
1722 
1723 	/* Can still get packets with rsvp_on = 0 if there is a local member
1724 	 * of the group to which the RSVP packet is addressed.  But in this
1725 	 * case we want to throw the packet away.
1726 	 */
1727 
1728 	if (!V_rsvp_on) {
1729 		m_freem(m);
1730 		return;
1731 	}
1732 
1733 	if (V_ip_rsvpd != NULL) {
1734 		rip_input(m, off);
1735 		return;
1736 	}
1737 	/* Drop the packet */
1738 	m_freem(m);
1739 }
1740