xref: /freebsd/sys/netinet/ip_input.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_bootp.h"
36 #include "opt_ipfw.h"
37 #include "opt_ipstealth.h"
38 #include "opt_ipsec.h"
39 #include "opt_route.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/mbuf.h>
44 #include <sys/malloc.h>
45 #include <sys/domain.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 
55 #include <net/pfil.h>
56 #include <net/if.h>
57 #include <net/if_types.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/netisr.h>
62 #include <net/vnet.h>
63 #include <net/flowtable.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_fw.h>
72 #include <netinet/ip_icmp.h>
73 #include <netinet/ip_options.h>
74 #include <machine/in_cksum.h>
75 #include <netinet/ip_carp.h>
76 #ifdef IPSEC
77 #include <netinet/ip_ipsec.h>
78 #endif /* IPSEC */
79 
80 #include <sys/socketvar.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #ifdef CTASSERT
85 CTASSERT(sizeof(struct ip) == 20);
86 #endif
87 
88 struct	rwlock in_ifaddr_lock;
89 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock");
90 
91 VNET_DEFINE(int, rsvp_on);
92 
93 VNET_DEFINE(int, ipforwarding);
94 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
95     &VNET_NAME(ipforwarding), 0,
96     "Enable IP forwarding between interfaces");
97 
98 static VNET_DEFINE(int, ipsendredirects) = 1;	/* XXX */
99 #define	V_ipsendredirects	VNET(ipsendredirects)
100 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
101     &VNET_NAME(ipsendredirects), 0,
102     "Enable sending IP redirects");
103 
104 static VNET_DEFINE(int, ip_keepfaith);
105 #define	V_ip_keepfaith		VNET(ip_keepfaith)
106 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
107     &VNET_NAME(ip_keepfaith), 0,
108     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
109 
110 static VNET_DEFINE(int, ip_sendsourcequench);
111 #define	V_ip_sendsourcequench	VNET(ip_sendsourcequench)
112 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
113     &VNET_NAME(ip_sendsourcequench), 0,
114     "Enable the transmission of source quench packets");
115 
116 VNET_DEFINE(int, ip_do_randomid);
117 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
118     &VNET_NAME(ip_do_randomid), 0,
119     "Assign random ip_id values");
120 
121 /*
122  * XXX - Setting ip_checkinterface mostly implements the receive side of
123  * the Strong ES model described in RFC 1122, but since the routing table
124  * and transmit implementation do not implement the Strong ES model,
125  * setting this to 1 results in an odd hybrid.
126  *
127  * XXX - ip_checkinterface currently must be disabled if you use ipnat
128  * to translate the destination address to another local interface.
129  *
130  * XXX - ip_checkinterface must be disabled if you add IP aliases
131  * to the loopback interface instead of the interface where the
132  * packets for those addresses are received.
133  */
134 static VNET_DEFINE(int, ip_checkinterface);
135 #define	V_ip_checkinterface	VNET(ip_checkinterface)
136 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
137     &VNET_NAME(ip_checkinterface), 0,
138     "Verify packet arrives on correct interface");
139 
140 VNET_DEFINE(struct pfil_head, inet_pfil_hook);	/* Packet filter hooks */
141 
142 static struct netisr_handler ip_nh = {
143 	.nh_name = "ip",
144 	.nh_handler = ip_input,
145 	.nh_proto = NETISR_IP,
146 	.nh_policy = NETISR_POLICY_FLOW,
147 };
148 
149 extern	struct domain inetdomain;
150 extern	struct protosw inetsw[];
151 u_char	ip_protox[IPPROTO_MAX];
152 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead);  /* first inet address */
153 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table  */
154 VNET_DEFINE(u_long, in_ifaddrhmask);		/* mask for hash table */
155 
156 VNET_DEFINE(struct ipstat, ipstat);
157 SYSCTL_VNET_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
158     &VNET_NAME(ipstat), ipstat,
159     "IP statistics (struct ipstat, netinet/ip_var.h)");
160 
161 static VNET_DEFINE(uma_zone_t, ipq_zone);
162 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]);
163 static struct mtx ipqlock;
164 
165 #define	V_ipq_zone		VNET(ipq_zone)
166 #define	V_ipq			VNET(ipq)
167 
168 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
169 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
170 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
171 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
172 
173 static void	maxnipq_update(void);
174 static void	ipq_zone_change(void *);
175 static void	ip_drain_locked(void);
176 
177 static VNET_DEFINE(int, maxnipq);  /* Administrative limit on # reass queues. */
178 static VNET_DEFINE(int, nipq);			/* Total # of reass queues */
179 #define	V_maxnipq		VNET(maxnipq)
180 #define	V_nipq			VNET(nipq)
181 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
182     &VNET_NAME(nipq), 0,
183     "Current number of IPv4 fragment reassembly queue entries");
184 
185 static VNET_DEFINE(int, maxfragsperpacket);
186 #define	V_maxfragsperpacket	VNET(maxfragsperpacket)
187 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
188     &VNET_NAME(maxfragsperpacket), 0,
189     "Maximum number of IPv4 fragments allowed per packet");
190 
191 #ifdef IPCTL_DEFMTU
192 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
193     &ip_mtu, 0, "Default MTU");
194 #endif
195 
196 #ifdef IPSTEALTH
197 VNET_DEFINE(int, ipstealth);
198 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
199     &VNET_NAME(ipstealth), 0,
200     "IP stealth mode, no TTL decrementation on forwarding");
201 #endif
202 
203 #ifdef FLOWTABLE
204 static VNET_DEFINE(int, ip_output_flowtable_size) = 2048;
205 VNET_DEFINE(struct flowtable *, ip_ft);
206 #define	V_ip_output_flowtable_size	VNET(ip_output_flowtable_size)
207 
208 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
209     &VNET_NAME(ip_output_flowtable_size), 2048,
210     "number of entries in the per-cpu output flow caches");
211 #endif
212 
213 static void	ip_freef(struct ipqhead *, struct ipq *);
214 
215 /*
216  * Kernel module interface for updating ipstat.  The argument is an index
217  * into ipstat treated as an array of u_long.  While this encodes the general
218  * layout of ipstat into the caller, it doesn't encode its location, so that
219  * future changes to add, for example, per-CPU stats support won't cause
220  * binary compatibility problems for kernel modules.
221  */
222 void
223 kmod_ipstat_inc(int statnum)
224 {
225 
226 	(*((u_long *)&V_ipstat + statnum))++;
227 }
228 
229 void
230 kmod_ipstat_dec(int statnum)
231 {
232 
233 	(*((u_long *)&V_ipstat + statnum))--;
234 }
235 
236 static int
237 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
238 {
239 	int error, qlimit;
240 
241 	netisr_getqlimit(&ip_nh, &qlimit);
242 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
243 	if (error || !req->newptr)
244 		return (error);
245 	if (qlimit < 1)
246 		return (EINVAL);
247 	return (netisr_setqlimit(&ip_nh, qlimit));
248 }
249 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
250     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
251     "Maximum size of the IP input queue");
252 
253 static int
254 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
255 {
256 	u_int64_t qdrops_long;
257 	int error, qdrops;
258 
259 	netisr_getqdrops(&ip_nh, &qdrops_long);
260 	qdrops = qdrops_long;
261 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
262 	if (error || !req->newptr)
263 		return (error);
264 	if (qdrops != 0)
265 		return (EINVAL);
266 	netisr_clearqdrops(&ip_nh);
267 	return (0);
268 }
269 
270 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
271     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
272     "Number of packets dropped from the IP input queue");
273 
274 /*
275  * IP initialization: fill in IP protocol switch table.
276  * All protocols not implemented in kernel go to raw IP protocol handler.
277  */
278 void
279 ip_init(void)
280 {
281 	struct protosw *pr;
282 	int i;
283 
284 	V_ip_id = time_second & 0xffff;
285 
286 	TAILQ_INIT(&V_in_ifaddrhead);
287 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
288 
289 	/* Initialize IP reassembly queue. */
290 	for (i = 0; i < IPREASS_NHASH; i++)
291 		TAILQ_INIT(&V_ipq[i]);
292 	V_maxnipq = nmbclusters / 32;
293 	V_maxfragsperpacket = 16;
294 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
295 	    NULL, UMA_ALIGN_PTR, 0);
296 	maxnipq_update();
297 
298 	/* Initialize packet filter hooks. */
299 	V_inet_pfil_hook.ph_type = PFIL_TYPE_AF;
300 	V_inet_pfil_hook.ph_af = AF_INET;
301 	if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0)
302 		printf("%s: WARNING: unable to register pfil hook, "
303 			"error %d\n", __func__, i);
304 
305 #ifdef FLOWTABLE
306 	if (TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size",
307 		&V_ip_output_flowtable_size)) {
308 		if (V_ip_output_flowtable_size < 256)
309 			V_ip_output_flowtable_size = 256;
310 		if (!powerof2(V_ip_output_flowtable_size)) {
311 			printf("flowtable must be power of 2 size\n");
312 			V_ip_output_flowtable_size = 2048;
313 		}
314 	} else {
315 		/*
316 		 * round up to the next power of 2
317 		 */
318 		V_ip_output_flowtable_size = 1 << fls((1024 + maxusers * 64)-1);
319 	}
320 	V_ip_ft = flowtable_alloc("ipv4", V_ip_output_flowtable_size, FL_PCPU);
321 #endif
322 
323 	/* Skip initialization of globals for non-default instances. */
324 	if (!IS_DEFAULT_VNET(curvnet))
325 		return;
326 
327 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
328 	if (pr == NULL)
329 		panic("ip_init: PF_INET not found");
330 
331 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
332 	for (i = 0; i < IPPROTO_MAX; i++)
333 		ip_protox[i] = pr - inetsw;
334 	/*
335 	 * Cycle through IP protocols and put them into the appropriate place
336 	 * in ip_protox[].
337 	 */
338 	for (pr = inetdomain.dom_protosw;
339 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
340 		if (pr->pr_domain->dom_family == PF_INET &&
341 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
342 			/* Be careful to only index valid IP protocols. */
343 			if (pr->pr_protocol < IPPROTO_MAX)
344 				ip_protox[pr->pr_protocol] = pr - inetsw;
345 		}
346 
347 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
348 		NULL, EVENTHANDLER_PRI_ANY);
349 
350 	/* Initialize various other remaining things. */
351 	IPQ_LOCK_INIT();
352 	netisr_register(&ip_nh);
353 }
354 
355 #ifdef VIMAGE
356 void
357 ip_destroy(void)
358 {
359 
360 	/* Cleanup in_ifaddr hash table; should be empty. */
361 	hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask);
362 
363 	IPQ_LOCK();
364 	ip_drain_locked();
365 	IPQ_UNLOCK();
366 
367 	uma_zdestroy(V_ipq_zone);
368 }
369 #endif
370 
371 /*
372  * Ip input routine.  Checksum and byte swap header.  If fragmented
373  * try to reassemble.  Process options.  Pass to next level.
374  */
375 void
376 ip_input(struct mbuf *m)
377 {
378 	struct ip *ip = NULL;
379 	struct in_ifaddr *ia = NULL;
380 	struct ifaddr *ifa;
381 	struct ifnet *ifp;
382 	int    checkif, hlen = 0;
383 	u_short sum;
384 	int dchg = 0;				/* dest changed after fw */
385 	struct in_addr odst;			/* original dst address */
386 
387 	M_ASSERTPKTHDR(m);
388 
389 	if (m->m_flags & M_FASTFWD_OURS) {
390 		/*
391 		 * Firewall or NAT changed destination to local.
392 		 * We expect ip_len and ip_off to be in host byte order.
393 		 */
394 		m->m_flags &= ~M_FASTFWD_OURS;
395 		/* Set up some basics that will be used later. */
396 		ip = mtod(m, struct ip *);
397 		hlen = ip->ip_hl << 2;
398 		goto ours;
399 	}
400 
401 	IPSTAT_INC(ips_total);
402 
403 	if (m->m_pkthdr.len < sizeof(struct ip))
404 		goto tooshort;
405 
406 	if (m->m_len < sizeof (struct ip) &&
407 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
408 		IPSTAT_INC(ips_toosmall);
409 		return;
410 	}
411 	ip = mtod(m, struct ip *);
412 
413 	if (ip->ip_v != IPVERSION) {
414 		IPSTAT_INC(ips_badvers);
415 		goto bad;
416 	}
417 
418 	hlen = ip->ip_hl << 2;
419 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
420 		IPSTAT_INC(ips_badhlen);
421 		goto bad;
422 	}
423 	if (hlen > m->m_len) {
424 		if ((m = m_pullup(m, hlen)) == NULL) {
425 			IPSTAT_INC(ips_badhlen);
426 			return;
427 		}
428 		ip = mtod(m, struct ip *);
429 	}
430 
431 	/* 127/8 must not appear on wire - RFC1122 */
432 	ifp = m->m_pkthdr.rcvif;
433 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
434 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
435 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
436 			IPSTAT_INC(ips_badaddr);
437 			goto bad;
438 		}
439 	}
440 
441 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
442 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
443 	} else {
444 		if (hlen == sizeof(struct ip)) {
445 			sum = in_cksum_hdr(ip);
446 		} else {
447 			sum = in_cksum(m, hlen);
448 		}
449 	}
450 	if (sum) {
451 		IPSTAT_INC(ips_badsum);
452 		goto bad;
453 	}
454 
455 #ifdef ALTQ
456 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
457 		/* packet is dropped by traffic conditioner */
458 		return;
459 #endif
460 
461 	/*
462 	 * Convert fields to host representation.
463 	 */
464 	ip->ip_len = ntohs(ip->ip_len);
465 	if (ip->ip_len < hlen) {
466 		IPSTAT_INC(ips_badlen);
467 		goto bad;
468 	}
469 	ip->ip_off = ntohs(ip->ip_off);
470 
471 	/*
472 	 * Check that the amount of data in the buffers
473 	 * is as at least much as the IP header would have us expect.
474 	 * Trim mbufs if longer than we expect.
475 	 * Drop packet if shorter than we expect.
476 	 */
477 	if (m->m_pkthdr.len < ip->ip_len) {
478 tooshort:
479 		IPSTAT_INC(ips_tooshort);
480 		goto bad;
481 	}
482 	if (m->m_pkthdr.len > ip->ip_len) {
483 		if (m->m_len == m->m_pkthdr.len) {
484 			m->m_len = ip->ip_len;
485 			m->m_pkthdr.len = ip->ip_len;
486 		} else
487 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
488 	}
489 #ifdef IPSEC
490 	/*
491 	 * Bypass packet filtering for packets from a tunnel (gif).
492 	 */
493 	if (ip_ipsec_filtertunnel(m))
494 		goto passin;
495 #endif /* IPSEC */
496 
497 	/*
498 	 * Run through list of hooks for input packets.
499 	 *
500 	 * NB: Beware of the destination address changing (e.g.
501 	 *     by NAT rewriting).  When this happens, tell
502 	 *     ip_forward to do the right thing.
503 	 */
504 
505 	/* Jump over all PFIL processing if hooks are not active. */
506 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
507 		goto passin;
508 
509 	odst = ip->ip_dst;
510 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
511 		return;
512 	if (m == NULL)			/* consumed by filter */
513 		return;
514 
515 	ip = mtod(m, struct ip *);
516 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
517 	ifp = m->m_pkthdr.rcvif;
518 
519 #ifdef IPFIREWALL_FORWARD
520 	if (m->m_flags & M_FASTFWD_OURS) {
521 		m->m_flags &= ~M_FASTFWD_OURS;
522 		goto ours;
523 	}
524 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
525 		/*
526 		 * Directly ship the packet on.  This allows forwarding
527 		 * packets originally destined to us to some other directly
528 		 * connected host.
529 		 */
530 		ip_forward(m, dchg);
531 		return;
532 	}
533 #endif /* IPFIREWALL_FORWARD */
534 
535 passin:
536 	/*
537 	 * Process options and, if not destined for us,
538 	 * ship it on.  ip_dooptions returns 1 when an
539 	 * error was detected (causing an icmp message
540 	 * to be sent and the original packet to be freed).
541 	 */
542 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
543 		return;
544 
545         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
546          * matter if it is destined to another node, or whether it is
547          * a multicast one, RSVP wants it! and prevents it from being forwarded
548          * anywhere else. Also checks if the rsvp daemon is running before
549 	 * grabbing the packet.
550          */
551 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
552 		goto ours;
553 
554 	/*
555 	 * Check our list of addresses, to see if the packet is for us.
556 	 * If we don't have any addresses, assume any unicast packet
557 	 * we receive might be for us (and let the upper layers deal
558 	 * with it).
559 	 */
560 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
561 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
562 		goto ours;
563 
564 	/*
565 	 * Enable a consistency check between the destination address
566 	 * and the arrival interface for a unicast packet (the RFC 1122
567 	 * strong ES model) if IP forwarding is disabled and the packet
568 	 * is not locally generated and the packet is not subject to
569 	 * 'ipfw fwd'.
570 	 *
571 	 * XXX - Checking also should be disabled if the destination
572 	 * address is ipnat'ed to a different interface.
573 	 *
574 	 * XXX - Checking is incompatible with IP aliases added
575 	 * to the loopback interface instead of the interface where
576 	 * the packets are received.
577 	 *
578 	 * XXX - This is the case for carp vhost IPs as well so we
579 	 * insert a workaround. If the packet got here, we already
580 	 * checked with carp_iamatch() and carp_forus().
581 	 */
582 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
583 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
584 	    ifp->if_carp == NULL && (dchg == 0);
585 
586 	/*
587 	 * Check for exact addresses in the hash bucket.
588 	 */
589 	/* IN_IFADDR_RLOCK(); */
590 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
591 		/*
592 		 * If the address matches, verify that the packet
593 		 * arrived via the correct interface if checking is
594 		 * enabled.
595 		 */
596 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
597 		    (!checkif || ia->ia_ifp == ifp)) {
598 			ifa_ref(&ia->ia_ifa);
599 			/* IN_IFADDR_RUNLOCK(); */
600 			goto ours;
601 		}
602 	}
603 	/* IN_IFADDR_RUNLOCK(); */
604 
605 	/*
606 	 * Check for broadcast addresses.
607 	 *
608 	 * Only accept broadcast packets that arrive via the matching
609 	 * interface.  Reception of forwarded directed broadcasts would
610 	 * be handled via ip_forward() and ether_output() with the loopback
611 	 * into the stack for SIMPLEX interfaces handled by ether_output().
612 	 */
613 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
614 		IF_ADDR_LOCK(ifp);
615 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
616 			if (ifa->ifa_addr->sa_family != AF_INET)
617 				continue;
618 			ia = ifatoia(ifa);
619 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
620 			    ip->ip_dst.s_addr) {
621 				ifa_ref(ifa);
622 				IF_ADDR_UNLOCK(ifp);
623 				goto ours;
624 			}
625 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
626 				ifa_ref(ifa);
627 				IF_ADDR_UNLOCK(ifp);
628 				goto ours;
629 			}
630 #ifdef BOOTP_COMPAT
631 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
632 				ifa_ref(ifa);
633 				IF_ADDR_UNLOCK(ifp);
634 				goto ours;
635 			}
636 #endif
637 		}
638 		IF_ADDR_UNLOCK(ifp);
639 		ia = NULL;
640 	}
641 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
642 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
643 		IPSTAT_INC(ips_cantforward);
644 		m_freem(m);
645 		return;
646 	}
647 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
648 		if (V_ip_mrouter) {
649 			/*
650 			 * If we are acting as a multicast router, all
651 			 * incoming multicast packets are passed to the
652 			 * kernel-level multicast forwarding function.
653 			 * The packet is returned (relatively) intact; if
654 			 * ip_mforward() returns a non-zero value, the packet
655 			 * must be discarded, else it may be accepted below.
656 			 */
657 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
658 				IPSTAT_INC(ips_cantforward);
659 				m_freem(m);
660 				return;
661 			}
662 
663 			/*
664 			 * The process-level routing daemon needs to receive
665 			 * all multicast IGMP packets, whether or not this
666 			 * host belongs to their destination groups.
667 			 */
668 			if (ip->ip_p == IPPROTO_IGMP)
669 				goto ours;
670 			IPSTAT_INC(ips_forward);
671 		}
672 		/*
673 		 * Assume the packet is for us, to avoid prematurely taking
674 		 * a lock on the in_multi hash. Protocols must perform
675 		 * their own filtering and update statistics accordingly.
676 		 */
677 		goto ours;
678 	}
679 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
680 		goto ours;
681 	if (ip->ip_dst.s_addr == INADDR_ANY)
682 		goto ours;
683 
684 	/*
685 	 * FAITH(Firewall Aided Internet Translator)
686 	 */
687 	if (ifp && ifp->if_type == IFT_FAITH) {
688 		if (V_ip_keepfaith) {
689 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
690 				goto ours;
691 		}
692 		m_freem(m);
693 		return;
694 	}
695 
696 	/*
697 	 * Not for us; forward if possible and desirable.
698 	 */
699 	if (V_ipforwarding == 0) {
700 		IPSTAT_INC(ips_cantforward);
701 		m_freem(m);
702 	} else {
703 #ifdef IPSEC
704 		if (ip_ipsec_fwd(m))
705 			goto bad;
706 #endif /* IPSEC */
707 		ip_forward(m, dchg);
708 	}
709 	return;
710 
711 ours:
712 #ifdef IPSTEALTH
713 	/*
714 	 * IPSTEALTH: Process non-routing options only
715 	 * if the packet is destined for us.
716 	 */
717 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) {
718 		if (ia != NULL)
719 			ifa_free(&ia->ia_ifa);
720 		return;
721 	}
722 #endif /* IPSTEALTH */
723 
724 	/* Count the packet in the ip address stats */
725 	if (ia != NULL) {
726 		ia->ia_ifa.if_ipackets++;
727 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
728 		ifa_free(&ia->ia_ifa);
729 	}
730 
731 	/*
732 	 * Attempt reassembly; if it succeeds, proceed.
733 	 * ip_reass() will return a different mbuf.
734 	 */
735 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
736 		m = ip_reass(m);
737 		if (m == NULL)
738 			return;
739 		ip = mtod(m, struct ip *);
740 		/* Get the header length of the reassembled packet */
741 		hlen = ip->ip_hl << 2;
742 	}
743 
744 	/*
745 	 * Further protocols expect the packet length to be w/o the
746 	 * IP header.
747 	 */
748 	ip->ip_len -= hlen;
749 
750 #ifdef IPSEC
751 	/*
752 	 * enforce IPsec policy checking if we are seeing last header.
753 	 * note that we do not visit this with protocols with pcb layer
754 	 * code - like udp/tcp/raw ip.
755 	 */
756 	if (ip_ipsec_input(m))
757 		goto bad;
758 #endif /* IPSEC */
759 
760 	/*
761 	 * Switch out to protocol's input routine.
762 	 */
763 	IPSTAT_INC(ips_delivered);
764 
765 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
766 	return;
767 bad:
768 	m_freem(m);
769 }
770 
771 /*
772  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
773  * max has slightly different semantics than the sysctl, for historical
774  * reasons.
775  */
776 static void
777 maxnipq_update(void)
778 {
779 
780 	/*
781 	 * -1 for unlimited allocation.
782 	 */
783 	if (V_maxnipq < 0)
784 		uma_zone_set_max(V_ipq_zone, 0);
785 	/*
786 	 * Positive number for specific bound.
787 	 */
788 	if (V_maxnipq > 0)
789 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
790 	/*
791 	 * Zero specifies no further fragment queue allocation -- set the
792 	 * bound very low, but rely on implementation elsewhere to actually
793 	 * prevent allocation and reclaim current queues.
794 	 */
795 	if (V_maxnipq == 0)
796 		uma_zone_set_max(V_ipq_zone, 1);
797 }
798 
799 static void
800 ipq_zone_change(void *tag)
801 {
802 
803 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
804 		V_maxnipq = nmbclusters / 32;
805 		maxnipq_update();
806 	}
807 }
808 
809 static int
810 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
811 {
812 	int error, i;
813 
814 	i = V_maxnipq;
815 	error = sysctl_handle_int(oidp, &i, 0, req);
816 	if (error || !req->newptr)
817 		return (error);
818 
819 	/*
820 	 * XXXRW: Might be a good idea to sanity check the argument and place
821 	 * an extreme upper bound.
822 	 */
823 	if (i < -1)
824 		return (EINVAL);
825 	V_maxnipq = i;
826 	maxnipq_update();
827 	return (0);
828 }
829 
830 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
831     NULL, 0, sysctl_maxnipq, "I",
832     "Maximum number of IPv4 fragment reassembly queue entries");
833 
834 /*
835  * Take incoming datagram fragment and try to reassemble it into
836  * whole datagram.  If the argument is the first fragment or one
837  * in between the function will return NULL and store the mbuf
838  * in the fragment chain.  If the argument is the last fragment
839  * the packet will be reassembled and the pointer to the new
840  * mbuf returned for further processing.  Only m_tags attached
841  * to the first packet/fragment are preserved.
842  * The IP header is *NOT* adjusted out of iplen.
843  */
844 struct mbuf *
845 ip_reass(struct mbuf *m)
846 {
847 	struct ip *ip;
848 	struct mbuf *p, *q, *nq, *t;
849 	struct ipq *fp = NULL;
850 	struct ipqhead *head;
851 	int i, hlen, next;
852 	u_int8_t ecn, ecn0;
853 	u_short hash;
854 
855 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
856 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
857 		IPSTAT_INC(ips_fragments);
858 		IPSTAT_INC(ips_fragdropped);
859 		m_freem(m);
860 		return (NULL);
861 	}
862 
863 	ip = mtod(m, struct ip *);
864 	hlen = ip->ip_hl << 2;
865 
866 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
867 	head = &V_ipq[hash];
868 	IPQ_LOCK();
869 
870 	/*
871 	 * Look for queue of fragments
872 	 * of this datagram.
873 	 */
874 	TAILQ_FOREACH(fp, head, ipq_list)
875 		if (ip->ip_id == fp->ipq_id &&
876 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
877 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
878 #ifdef MAC
879 		    mac_ipq_match(m, fp) &&
880 #endif
881 		    ip->ip_p == fp->ipq_p)
882 			goto found;
883 
884 	fp = NULL;
885 
886 	/*
887 	 * Attempt to trim the number of allocated fragment queues if it
888 	 * exceeds the administrative limit.
889 	 */
890 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
891 		/*
892 		 * drop something from the tail of the current queue
893 		 * before proceeding further
894 		 */
895 		struct ipq *q = TAILQ_LAST(head, ipqhead);
896 		if (q == NULL) {   /* gak */
897 			for (i = 0; i < IPREASS_NHASH; i++) {
898 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
899 				if (r) {
900 					IPSTAT_ADD(ips_fragtimeout,
901 					    r->ipq_nfrags);
902 					ip_freef(&V_ipq[i], r);
903 					break;
904 				}
905 			}
906 		} else {
907 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
908 			ip_freef(head, q);
909 		}
910 	}
911 
912 found:
913 	/*
914 	 * Adjust ip_len to not reflect header,
915 	 * convert offset of this to bytes.
916 	 */
917 	ip->ip_len -= hlen;
918 	if (ip->ip_off & IP_MF) {
919 		/*
920 		 * Make sure that fragments have a data length
921 		 * that's a non-zero multiple of 8 bytes.
922 		 */
923 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
924 			IPSTAT_INC(ips_toosmall); /* XXX */
925 			goto dropfrag;
926 		}
927 		m->m_flags |= M_FRAG;
928 	} else
929 		m->m_flags &= ~M_FRAG;
930 	ip->ip_off <<= 3;
931 
932 
933 	/*
934 	 * Attempt reassembly; if it succeeds, proceed.
935 	 * ip_reass() will return a different mbuf.
936 	 */
937 	IPSTAT_INC(ips_fragments);
938 	m->m_pkthdr.header = ip;
939 
940 	/* Previous ip_reass() started here. */
941 	/*
942 	 * Presence of header sizes in mbufs
943 	 * would confuse code below.
944 	 */
945 	m->m_data += hlen;
946 	m->m_len -= hlen;
947 
948 	/*
949 	 * If first fragment to arrive, create a reassembly queue.
950 	 */
951 	if (fp == NULL) {
952 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
953 		if (fp == NULL)
954 			goto dropfrag;
955 #ifdef MAC
956 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
957 			uma_zfree(V_ipq_zone, fp);
958 			fp = NULL;
959 			goto dropfrag;
960 		}
961 		mac_ipq_create(m, fp);
962 #endif
963 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
964 		V_nipq++;
965 		fp->ipq_nfrags = 1;
966 		fp->ipq_ttl = IPFRAGTTL;
967 		fp->ipq_p = ip->ip_p;
968 		fp->ipq_id = ip->ip_id;
969 		fp->ipq_src = ip->ip_src;
970 		fp->ipq_dst = ip->ip_dst;
971 		fp->ipq_frags = m;
972 		m->m_nextpkt = NULL;
973 		goto done;
974 	} else {
975 		fp->ipq_nfrags++;
976 #ifdef MAC
977 		mac_ipq_update(m, fp);
978 #endif
979 	}
980 
981 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
982 
983 	/*
984 	 * Handle ECN by comparing this segment with the first one;
985 	 * if CE is set, do not lose CE.
986 	 * drop if CE and not-ECT are mixed for the same packet.
987 	 */
988 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
989 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
990 	if (ecn == IPTOS_ECN_CE) {
991 		if (ecn0 == IPTOS_ECN_NOTECT)
992 			goto dropfrag;
993 		if (ecn0 != IPTOS_ECN_CE)
994 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
995 	}
996 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
997 		goto dropfrag;
998 
999 	/*
1000 	 * Find a segment which begins after this one does.
1001 	 */
1002 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1003 		if (GETIP(q)->ip_off > ip->ip_off)
1004 			break;
1005 
1006 	/*
1007 	 * If there is a preceding segment, it may provide some of
1008 	 * our data already.  If so, drop the data from the incoming
1009 	 * segment.  If it provides all of our data, drop us, otherwise
1010 	 * stick new segment in the proper place.
1011 	 *
1012 	 * If some of the data is dropped from the preceding
1013 	 * segment, then it's checksum is invalidated.
1014 	 */
1015 	if (p) {
1016 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1017 		if (i > 0) {
1018 			if (i >= ip->ip_len)
1019 				goto dropfrag;
1020 			m_adj(m, i);
1021 			m->m_pkthdr.csum_flags = 0;
1022 			ip->ip_off += i;
1023 			ip->ip_len -= i;
1024 		}
1025 		m->m_nextpkt = p->m_nextpkt;
1026 		p->m_nextpkt = m;
1027 	} else {
1028 		m->m_nextpkt = fp->ipq_frags;
1029 		fp->ipq_frags = m;
1030 	}
1031 
1032 	/*
1033 	 * While we overlap succeeding segments trim them or,
1034 	 * if they are completely covered, dequeue them.
1035 	 */
1036 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1037 	     q = nq) {
1038 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1039 		if (i < GETIP(q)->ip_len) {
1040 			GETIP(q)->ip_len -= i;
1041 			GETIP(q)->ip_off += i;
1042 			m_adj(q, i);
1043 			q->m_pkthdr.csum_flags = 0;
1044 			break;
1045 		}
1046 		nq = q->m_nextpkt;
1047 		m->m_nextpkt = nq;
1048 		IPSTAT_INC(ips_fragdropped);
1049 		fp->ipq_nfrags--;
1050 		m_freem(q);
1051 	}
1052 
1053 	/*
1054 	 * Check for complete reassembly and perform frag per packet
1055 	 * limiting.
1056 	 *
1057 	 * Frag limiting is performed here so that the nth frag has
1058 	 * a chance to complete the packet before we drop the packet.
1059 	 * As a result, n+1 frags are actually allowed per packet, but
1060 	 * only n will ever be stored. (n = maxfragsperpacket.)
1061 	 *
1062 	 */
1063 	next = 0;
1064 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1065 		if (GETIP(q)->ip_off != next) {
1066 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1067 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1068 				ip_freef(head, fp);
1069 			}
1070 			goto done;
1071 		}
1072 		next += GETIP(q)->ip_len;
1073 	}
1074 	/* Make sure the last packet didn't have the IP_MF flag */
1075 	if (p->m_flags & M_FRAG) {
1076 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1077 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1078 			ip_freef(head, fp);
1079 		}
1080 		goto done;
1081 	}
1082 
1083 	/*
1084 	 * Reassembly is complete.  Make sure the packet is a sane size.
1085 	 */
1086 	q = fp->ipq_frags;
1087 	ip = GETIP(q);
1088 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1089 		IPSTAT_INC(ips_toolong);
1090 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1091 		ip_freef(head, fp);
1092 		goto done;
1093 	}
1094 
1095 	/*
1096 	 * Concatenate fragments.
1097 	 */
1098 	m = q;
1099 	t = m->m_next;
1100 	m->m_next = NULL;
1101 	m_cat(m, t);
1102 	nq = q->m_nextpkt;
1103 	q->m_nextpkt = NULL;
1104 	for (q = nq; q != NULL; q = nq) {
1105 		nq = q->m_nextpkt;
1106 		q->m_nextpkt = NULL;
1107 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1108 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1109 		m_cat(m, q);
1110 	}
1111 	/*
1112 	 * In order to do checksumming faster we do 'end-around carry' here
1113 	 * (and not in for{} loop), though it implies we are not going to
1114 	 * reassemble more than 64k fragments.
1115 	 */
1116 	m->m_pkthdr.csum_data =
1117 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1118 #ifdef MAC
1119 	mac_ipq_reassemble(fp, m);
1120 	mac_ipq_destroy(fp);
1121 #endif
1122 
1123 	/*
1124 	 * Create header for new ip packet by modifying header of first
1125 	 * packet;  dequeue and discard fragment reassembly header.
1126 	 * Make header visible.
1127 	 */
1128 	ip->ip_len = (ip->ip_hl << 2) + next;
1129 	ip->ip_src = fp->ipq_src;
1130 	ip->ip_dst = fp->ipq_dst;
1131 	TAILQ_REMOVE(head, fp, ipq_list);
1132 	V_nipq--;
1133 	uma_zfree(V_ipq_zone, fp);
1134 	m->m_len += (ip->ip_hl << 2);
1135 	m->m_data -= (ip->ip_hl << 2);
1136 	/* some debugging cruft by sklower, below, will go away soon */
1137 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1138 		m_fixhdr(m);
1139 	IPSTAT_INC(ips_reassembled);
1140 	IPQ_UNLOCK();
1141 	return (m);
1142 
1143 dropfrag:
1144 	IPSTAT_INC(ips_fragdropped);
1145 	if (fp != NULL)
1146 		fp->ipq_nfrags--;
1147 	m_freem(m);
1148 done:
1149 	IPQ_UNLOCK();
1150 	return (NULL);
1151 
1152 #undef GETIP
1153 }
1154 
1155 /*
1156  * Free a fragment reassembly header and all
1157  * associated datagrams.
1158  */
1159 static void
1160 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1161 {
1162 	struct mbuf *q;
1163 
1164 	IPQ_LOCK_ASSERT();
1165 
1166 	while (fp->ipq_frags) {
1167 		q = fp->ipq_frags;
1168 		fp->ipq_frags = q->m_nextpkt;
1169 		m_freem(q);
1170 	}
1171 	TAILQ_REMOVE(fhp, fp, ipq_list);
1172 	uma_zfree(V_ipq_zone, fp);
1173 	V_nipq--;
1174 }
1175 
1176 /*
1177  * IP timer processing;
1178  * if a timer expires on a reassembly
1179  * queue, discard it.
1180  */
1181 void
1182 ip_slowtimo(void)
1183 {
1184 	VNET_ITERATOR_DECL(vnet_iter);
1185 	struct ipq *fp;
1186 	int i;
1187 
1188 	VNET_LIST_RLOCK_NOSLEEP();
1189 	IPQ_LOCK();
1190 	VNET_FOREACH(vnet_iter) {
1191 		CURVNET_SET(vnet_iter);
1192 		for (i = 0; i < IPREASS_NHASH; i++) {
1193 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1194 				struct ipq *fpp;
1195 
1196 				fpp = fp;
1197 				fp = TAILQ_NEXT(fp, ipq_list);
1198 				if(--fpp->ipq_ttl == 0) {
1199 					IPSTAT_ADD(ips_fragtimeout,
1200 					    fpp->ipq_nfrags);
1201 					ip_freef(&V_ipq[i], fpp);
1202 				}
1203 			}
1204 		}
1205 		/*
1206 		 * If we are over the maximum number of fragments
1207 		 * (due to the limit being lowered), drain off
1208 		 * enough to get down to the new limit.
1209 		 */
1210 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1211 			for (i = 0; i < IPREASS_NHASH; i++) {
1212 				while (V_nipq > V_maxnipq &&
1213 				    !TAILQ_EMPTY(&V_ipq[i])) {
1214 					IPSTAT_ADD(ips_fragdropped,
1215 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1216 					ip_freef(&V_ipq[i],
1217 					    TAILQ_FIRST(&V_ipq[i]));
1218 				}
1219 			}
1220 		}
1221 		CURVNET_RESTORE();
1222 	}
1223 	IPQ_UNLOCK();
1224 	VNET_LIST_RUNLOCK_NOSLEEP();
1225 }
1226 
1227 /*
1228  * Drain off all datagram fragments.
1229  */
1230 static void
1231 ip_drain_locked(void)
1232 {
1233 	int     i;
1234 
1235 	IPQ_LOCK_ASSERT();
1236 
1237 	for (i = 0; i < IPREASS_NHASH; i++) {
1238 		while(!TAILQ_EMPTY(&V_ipq[i])) {
1239 			IPSTAT_ADD(ips_fragdropped,
1240 			    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1241 			ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1242 		}
1243 	}
1244 }
1245 
1246 void
1247 ip_drain(void)
1248 {
1249 	VNET_ITERATOR_DECL(vnet_iter);
1250 
1251 	VNET_LIST_RLOCK_NOSLEEP();
1252 	IPQ_LOCK();
1253 	VNET_FOREACH(vnet_iter) {
1254 		CURVNET_SET(vnet_iter);
1255 		ip_drain_locked();
1256 		CURVNET_RESTORE();
1257 	}
1258 	IPQ_UNLOCK();
1259 	VNET_LIST_RUNLOCK_NOSLEEP();
1260 	in_rtqdrain();
1261 }
1262 
1263 /*
1264  * The protocol to be inserted into ip_protox[] must be already registered
1265  * in inetsw[], either statically or through pf_proto_register().
1266  */
1267 int
1268 ipproto_register(short ipproto)
1269 {
1270 	struct protosw *pr;
1271 
1272 	/* Sanity checks. */
1273 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1274 		return (EPROTONOSUPPORT);
1275 
1276 	/*
1277 	 * The protocol slot must not be occupied by another protocol
1278 	 * already.  An index pointing to IPPROTO_RAW is unused.
1279 	 */
1280 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1281 	if (pr == NULL)
1282 		return (EPFNOSUPPORT);
1283 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1284 		return (EEXIST);
1285 
1286 	/* Find the protocol position in inetsw[] and set the index. */
1287 	for (pr = inetdomain.dom_protosw;
1288 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1289 		if (pr->pr_domain->dom_family == PF_INET &&
1290 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1291 			ip_protox[pr->pr_protocol] = pr - inetsw;
1292 			return (0);
1293 		}
1294 	}
1295 	return (EPROTONOSUPPORT);
1296 }
1297 
1298 int
1299 ipproto_unregister(short ipproto)
1300 {
1301 	struct protosw *pr;
1302 
1303 	/* Sanity checks. */
1304 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1305 		return (EPROTONOSUPPORT);
1306 
1307 	/* Check if the protocol was indeed registered. */
1308 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1309 	if (pr == NULL)
1310 		return (EPFNOSUPPORT);
1311 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1312 		return (ENOENT);
1313 
1314 	/* Reset the protocol slot to IPPROTO_RAW. */
1315 	ip_protox[ipproto] = pr - inetsw;
1316 	return (0);
1317 }
1318 
1319 /*
1320  * Given address of next destination (final or next hop), return (referenced)
1321  * internet address info of interface to be used to get there.
1322  */
1323 struct in_ifaddr *
1324 ip_rtaddr(struct in_addr dst, u_int fibnum)
1325 {
1326 	struct route sro;
1327 	struct sockaddr_in *sin;
1328 	struct in_ifaddr *ia;
1329 
1330 	bzero(&sro, sizeof(sro));
1331 	sin = (struct sockaddr_in *)&sro.ro_dst;
1332 	sin->sin_family = AF_INET;
1333 	sin->sin_len = sizeof(*sin);
1334 	sin->sin_addr = dst;
1335 	in_rtalloc_ign(&sro, 0, fibnum);
1336 
1337 	if (sro.ro_rt == NULL)
1338 		return (NULL);
1339 
1340 	ia = ifatoia(sro.ro_rt->rt_ifa);
1341 	ifa_ref(&ia->ia_ifa);
1342 	RTFREE(sro.ro_rt);
1343 	return (ia);
1344 }
1345 
1346 u_char inetctlerrmap[PRC_NCMDS] = {
1347 	0,		0,		0,		0,
1348 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1349 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1350 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1351 	0,		0,		EHOSTUNREACH,	0,
1352 	ENOPROTOOPT,	ECONNREFUSED
1353 };
1354 
1355 /*
1356  * Forward a packet.  If some error occurs return the sender
1357  * an icmp packet.  Note we can't always generate a meaningful
1358  * icmp message because icmp doesn't have a large enough repertoire
1359  * of codes and types.
1360  *
1361  * If not forwarding, just drop the packet.  This could be confusing
1362  * if ipforwarding was zero but some routing protocol was advancing
1363  * us as a gateway to somewhere.  However, we must let the routing
1364  * protocol deal with that.
1365  *
1366  * The srcrt parameter indicates whether the packet is being forwarded
1367  * via a source route.
1368  */
1369 void
1370 ip_forward(struct mbuf *m, int srcrt)
1371 {
1372 	struct ip *ip = mtod(m, struct ip *);
1373 	struct in_ifaddr *ia;
1374 	struct mbuf *mcopy;
1375 	struct in_addr dest;
1376 	struct route ro;
1377 	int error, type = 0, code = 0, mtu = 0;
1378 
1379 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1380 		IPSTAT_INC(ips_cantforward);
1381 		m_freem(m);
1382 		return;
1383 	}
1384 #ifdef IPSTEALTH
1385 	if (!V_ipstealth) {
1386 #endif
1387 		if (ip->ip_ttl <= IPTTLDEC) {
1388 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1389 			    0, 0);
1390 			return;
1391 		}
1392 #ifdef IPSTEALTH
1393 	}
1394 #endif
1395 
1396 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1397 #ifndef IPSEC
1398 	/*
1399 	 * 'ia' may be NULL if there is no route for this destination.
1400 	 * In case of IPsec, Don't discard it just yet, but pass it to
1401 	 * ip_output in case of outgoing IPsec policy.
1402 	 */
1403 	if (!srcrt && ia == NULL) {
1404 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1405 		return;
1406 	}
1407 #endif
1408 
1409 	/*
1410 	 * Save the IP header and at most 8 bytes of the payload,
1411 	 * in case we need to generate an ICMP message to the src.
1412 	 *
1413 	 * XXX this can be optimized a lot by saving the data in a local
1414 	 * buffer on the stack (72 bytes at most), and only allocating the
1415 	 * mbuf if really necessary. The vast majority of the packets
1416 	 * are forwarded without having to send an ICMP back (either
1417 	 * because unnecessary, or because rate limited), so we are
1418 	 * really we are wasting a lot of work here.
1419 	 *
1420 	 * We don't use m_copy() because it might return a reference
1421 	 * to a shared cluster. Both this function and ip_output()
1422 	 * assume exclusive access to the IP header in `m', so any
1423 	 * data in a cluster may change before we reach icmp_error().
1424 	 */
1425 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1426 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1427 		/*
1428 		 * It's probably ok if the pkthdr dup fails (because
1429 		 * the deep copy of the tag chain failed), but for now
1430 		 * be conservative and just discard the copy since
1431 		 * code below may some day want the tags.
1432 		 */
1433 		m_free(mcopy);
1434 		mcopy = NULL;
1435 	}
1436 	if (mcopy != NULL) {
1437 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1438 		mcopy->m_pkthdr.len = mcopy->m_len;
1439 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1440 	}
1441 
1442 #ifdef IPSTEALTH
1443 	if (!V_ipstealth) {
1444 #endif
1445 		ip->ip_ttl -= IPTTLDEC;
1446 #ifdef IPSTEALTH
1447 	}
1448 #endif
1449 
1450 	/*
1451 	 * If forwarding packet using same interface that it came in on,
1452 	 * perhaps should send a redirect to sender to shortcut a hop.
1453 	 * Only send redirect if source is sending directly to us,
1454 	 * and if packet was not source routed (or has any options).
1455 	 * Also, don't send redirect if forwarding using a default route
1456 	 * or a route modified by a redirect.
1457 	 */
1458 	dest.s_addr = 0;
1459 	if (!srcrt && V_ipsendredirects &&
1460 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1461 		struct sockaddr_in *sin;
1462 		struct rtentry *rt;
1463 
1464 		bzero(&ro, sizeof(ro));
1465 		sin = (struct sockaddr_in *)&ro.ro_dst;
1466 		sin->sin_family = AF_INET;
1467 		sin->sin_len = sizeof(*sin);
1468 		sin->sin_addr = ip->ip_dst;
1469 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1470 
1471 		rt = ro.ro_rt;
1472 
1473 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1474 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1475 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1476 			u_long src = ntohl(ip->ip_src.s_addr);
1477 
1478 			if (RTA(rt) &&
1479 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1480 				if (rt->rt_flags & RTF_GATEWAY)
1481 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1482 				else
1483 					dest.s_addr = ip->ip_dst.s_addr;
1484 				/* Router requirements says to only send host redirects */
1485 				type = ICMP_REDIRECT;
1486 				code = ICMP_REDIRECT_HOST;
1487 			}
1488 		}
1489 		if (rt)
1490 			RTFREE(rt);
1491 	}
1492 
1493 	/*
1494 	 * Try to cache the route MTU from ip_output so we can consider it for
1495 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1496 	 */
1497 	bzero(&ro, sizeof(ro));
1498 
1499 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1500 
1501 	if (error == EMSGSIZE && ro.ro_rt)
1502 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1503 	if (ro.ro_rt)
1504 		RTFREE(ro.ro_rt);
1505 
1506 	if (error)
1507 		IPSTAT_INC(ips_cantforward);
1508 	else {
1509 		IPSTAT_INC(ips_forward);
1510 		if (type)
1511 			IPSTAT_INC(ips_redirectsent);
1512 		else {
1513 			if (mcopy)
1514 				m_freem(mcopy);
1515 			if (ia != NULL)
1516 				ifa_free(&ia->ia_ifa);
1517 			return;
1518 		}
1519 	}
1520 	if (mcopy == NULL) {
1521 		if (ia != NULL)
1522 			ifa_free(&ia->ia_ifa);
1523 		return;
1524 	}
1525 
1526 	switch (error) {
1527 
1528 	case 0:				/* forwarded, but need redirect */
1529 		/* type, code set above */
1530 		break;
1531 
1532 	case ENETUNREACH:
1533 	case EHOSTUNREACH:
1534 	case ENETDOWN:
1535 	case EHOSTDOWN:
1536 	default:
1537 		type = ICMP_UNREACH;
1538 		code = ICMP_UNREACH_HOST;
1539 		break;
1540 
1541 	case EMSGSIZE:
1542 		type = ICMP_UNREACH;
1543 		code = ICMP_UNREACH_NEEDFRAG;
1544 
1545 #ifdef IPSEC
1546 		/*
1547 		 * If IPsec is configured for this path,
1548 		 * override any possibly mtu value set by ip_output.
1549 		 */
1550 		mtu = ip_ipsec_mtu(mcopy, mtu);
1551 #endif /* IPSEC */
1552 		/*
1553 		 * If the MTU was set before make sure we are below the
1554 		 * interface MTU.
1555 		 * If the MTU wasn't set before use the interface mtu or
1556 		 * fall back to the next smaller mtu step compared to the
1557 		 * current packet size.
1558 		 */
1559 		if (mtu != 0) {
1560 			if (ia != NULL)
1561 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1562 		} else {
1563 			if (ia != NULL)
1564 				mtu = ia->ia_ifp->if_mtu;
1565 			else
1566 				mtu = ip_next_mtu(ip->ip_len, 0);
1567 		}
1568 		IPSTAT_INC(ips_cantfrag);
1569 		break;
1570 
1571 	case ENOBUFS:
1572 		/*
1573 		 * A router should not generate ICMP_SOURCEQUENCH as
1574 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1575 		 * Source quench could be a big problem under DoS attacks,
1576 		 * or if the underlying interface is rate-limited.
1577 		 * Those who need source quench packets may re-enable them
1578 		 * via the net.inet.ip.sendsourcequench sysctl.
1579 		 */
1580 		if (V_ip_sendsourcequench == 0) {
1581 			m_freem(mcopy);
1582 			if (ia != NULL)
1583 				ifa_free(&ia->ia_ifa);
1584 			return;
1585 		} else {
1586 			type = ICMP_SOURCEQUENCH;
1587 			code = 0;
1588 		}
1589 		break;
1590 
1591 	case EACCES:			/* ipfw denied packet */
1592 		m_freem(mcopy);
1593 		if (ia != NULL)
1594 			ifa_free(&ia->ia_ifa);
1595 		return;
1596 	}
1597 	if (ia != NULL)
1598 		ifa_free(&ia->ia_ifa);
1599 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1600 }
1601 
1602 void
1603 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1604     struct mbuf *m)
1605 {
1606 
1607 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1608 		struct bintime bt;
1609 
1610 		bintime(&bt);
1611 		if (inp->inp_socket->so_options & SO_BINTIME) {
1612 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1613 			SCM_BINTIME, SOL_SOCKET);
1614 			if (*mp)
1615 				mp = &(*mp)->m_next;
1616 		}
1617 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1618 			struct timeval tv;
1619 
1620 			bintime2timeval(&bt, &tv);
1621 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1622 				SCM_TIMESTAMP, SOL_SOCKET);
1623 			if (*mp)
1624 				mp = &(*mp)->m_next;
1625 		}
1626 	}
1627 	if (inp->inp_flags & INP_RECVDSTADDR) {
1628 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1629 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1630 		if (*mp)
1631 			mp = &(*mp)->m_next;
1632 	}
1633 	if (inp->inp_flags & INP_RECVTTL) {
1634 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1635 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1636 		if (*mp)
1637 			mp = &(*mp)->m_next;
1638 	}
1639 #ifdef notyet
1640 	/* XXX
1641 	 * Moving these out of udp_input() made them even more broken
1642 	 * than they already were.
1643 	 */
1644 	/* options were tossed already */
1645 	if (inp->inp_flags & INP_RECVOPTS) {
1646 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1647 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1648 		if (*mp)
1649 			mp = &(*mp)->m_next;
1650 	}
1651 	/* ip_srcroute doesn't do what we want here, need to fix */
1652 	if (inp->inp_flags & INP_RECVRETOPTS) {
1653 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1654 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1655 		if (*mp)
1656 			mp = &(*mp)->m_next;
1657 	}
1658 #endif
1659 	if (inp->inp_flags & INP_RECVIF) {
1660 		struct ifnet *ifp;
1661 		struct sdlbuf {
1662 			struct sockaddr_dl sdl;
1663 			u_char	pad[32];
1664 		} sdlbuf;
1665 		struct sockaddr_dl *sdp;
1666 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1667 
1668 		if (((ifp = m->m_pkthdr.rcvif))
1669 		&& ( ifp->if_index && (ifp->if_index <= V_if_index))) {
1670 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1671 			/*
1672 			 * Change our mind and don't try copy.
1673 			 */
1674 			if ((sdp->sdl_family != AF_LINK)
1675 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1676 				goto makedummy;
1677 			}
1678 			bcopy(sdp, sdl2, sdp->sdl_len);
1679 		} else {
1680 makedummy:
1681 			sdl2->sdl_len
1682 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1683 			sdl2->sdl_family = AF_LINK;
1684 			sdl2->sdl_index = 0;
1685 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1686 		}
1687 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1688 			IP_RECVIF, IPPROTO_IP);
1689 		if (*mp)
1690 			mp = &(*mp)->m_next;
1691 	}
1692 }
1693 
1694 /*
1695  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1696  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1697  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1698  * compiled.
1699  */
1700 static VNET_DEFINE(int, ip_rsvp_on);
1701 VNET_DEFINE(struct socket *, ip_rsvpd);
1702 
1703 #define	V_ip_rsvp_on		VNET(ip_rsvp_on)
1704 
1705 int
1706 ip_rsvp_init(struct socket *so)
1707 {
1708 
1709 	if (so->so_type != SOCK_RAW ||
1710 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1711 		return EOPNOTSUPP;
1712 
1713 	if (V_ip_rsvpd != NULL)
1714 		return EADDRINUSE;
1715 
1716 	V_ip_rsvpd = so;
1717 	/*
1718 	 * This may seem silly, but we need to be sure we don't over-increment
1719 	 * the RSVP counter, in case something slips up.
1720 	 */
1721 	if (!V_ip_rsvp_on) {
1722 		V_ip_rsvp_on = 1;
1723 		V_rsvp_on++;
1724 	}
1725 
1726 	return 0;
1727 }
1728 
1729 int
1730 ip_rsvp_done(void)
1731 {
1732 
1733 	V_ip_rsvpd = NULL;
1734 	/*
1735 	 * This may seem silly, but we need to be sure we don't over-decrement
1736 	 * the RSVP counter, in case something slips up.
1737 	 */
1738 	if (V_ip_rsvp_on) {
1739 		V_ip_rsvp_on = 0;
1740 		V_rsvp_on--;
1741 	}
1742 	return 0;
1743 }
1744 
1745 void
1746 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1747 {
1748 
1749 	if (rsvp_input_p) { /* call the real one if loaded */
1750 		rsvp_input_p(m, off);
1751 		return;
1752 	}
1753 
1754 	/* Can still get packets with rsvp_on = 0 if there is a local member
1755 	 * of the group to which the RSVP packet is addressed.  But in this
1756 	 * case we want to throw the packet away.
1757 	 */
1758 
1759 	if (!V_rsvp_on) {
1760 		m_freem(m);
1761 		return;
1762 	}
1763 
1764 	if (V_ip_rsvpd != NULL) {
1765 		rip_input(m, off);
1766 		return;
1767 	}
1768 	/* Drop the packet */
1769 	m_freem(m);
1770 }
1771