xref: /freebsd/sys/netinet/ip_input.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 #include "opt_carp.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/mbuf.h>
46 #include <sys/malloc.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/time.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/rwlock.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
56 
57 #include <net/pfil.h>
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/route.h>
63 #include <net/netisr.h>
64 #include <net/vnet.h>
65 #include <net/flowtable.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/ip_var.h>
73 #include <netinet/ip_fw.h>
74 #include <netinet/ip_icmp.h>
75 #include <netinet/ip_options.h>
76 #include <machine/in_cksum.h>
77 #ifdef DEV_CARP
78 #include <netinet/ip_carp.h>
79 #endif
80 #ifdef IPSEC
81 #include <netinet/ip_ipsec.h>
82 #endif /* IPSEC */
83 
84 #include <sys/socketvar.h>
85 
86 #include <security/mac/mac_framework.h>
87 
88 #ifdef CTASSERT
89 CTASSERT(sizeof(struct ip) == 20);
90 #endif
91 
92 static VNET_DEFINE(int, ipsendredirects) = 1;	/* XXX */
93 static VNET_DEFINE(int, ip_checkinterface);
94 static VNET_DEFINE(int, ip_keepfaith);
95 static VNET_DEFINE(int, ip_sendsourcequench);
96 
97 #define	V_ipsendredirects	VNET(ipsendredirects)
98 #define	V_ip_checkinterface	VNET(ip_checkinterface)
99 #define	V_ip_keepfaith		VNET(ip_keepfaith)
100 #define	V_ip_sendsourcequench	VNET(ip_sendsourcequench)
101 
102 VNET_DEFINE(int, ip_defttl) = IPDEFTTL;
103 VNET_DEFINE(int, ip_do_randomid);
104 VNET_DEFINE(int, ipforwarding);
105 
106 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead);  /* first inet address */
107 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table  */
108 VNET_DEFINE(u_long, in_ifaddrhmask);		/* mask for hash table */
109 VNET_DEFINE(struct ipstat, ipstat);
110 
111 static VNET_DEFINE(int, ip_rsvp_on);
112 VNET_DEFINE(struct socket *, ip_rsvpd);
113 VNET_DEFINE(int, rsvp_on);
114 
115 #define	V_ip_rsvp_on		VNET(ip_rsvp_on)
116 
117 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]);
118 static VNET_DEFINE(int, maxnipq);  /* Administrative limit on # reass queues. */
119 static VNET_DEFINE(int, maxfragsperpacket);
120 static VNET_DEFINE(int, nipq);			/* Total # of reass queues */
121 
122 #define	V_ipq			VNET(ipq)
123 #define	V_maxnipq		VNET(maxnipq)
124 #define	V_maxfragsperpacket	VNET(maxfragsperpacket)
125 #define	V_nipq			VNET(nipq)
126 
127 VNET_DEFINE(int, ipstealth);
128 
129 struct	rwlock in_ifaddr_lock;
130 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock");
131 
132 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
133     &VNET_NAME(ipforwarding), 0,
134     "Enable IP forwarding between interfaces");
135 
136 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
137     &VNET_NAME(ipsendredirects), 0,
138     "Enable sending IP redirects");
139 
140 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
141     &VNET_NAME(ip_defttl), 0,
142     "Maximum TTL on IP packets");
143 
144 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
145     &VNET_NAME(ip_keepfaith), 0,
146     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
147 
148 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
149     &VNET_NAME(ip_sendsourcequench), 0,
150     "Enable the transmission of source quench packets");
151 
152 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
153     &VNET_NAME(ip_do_randomid), 0,
154     "Assign random ip_id values");
155 
156 /*
157  * XXX - Setting ip_checkinterface mostly implements the receive side of
158  * the Strong ES model described in RFC 1122, but since the routing table
159  * and transmit implementation do not implement the Strong ES model,
160  * setting this to 1 results in an odd hybrid.
161  *
162  * XXX - ip_checkinterface currently must be disabled if you use ipnat
163  * to translate the destination address to another local interface.
164  *
165  * XXX - ip_checkinterface must be disabled if you add IP aliases
166  * to the loopback interface instead of the interface where the
167  * packets for those addresses are received.
168  */
169 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
170     &VNET_NAME(ip_checkinterface), 0,
171     "Verify packet arrives on correct interface");
172 
173 struct pfil_head inet_pfil_hook;	/* Packet filter hooks */
174 
175 static struct netisr_handler ip_nh = {
176 	.nh_name = "ip",
177 	.nh_handler = ip_input,
178 	.nh_proto = NETISR_IP,
179 	.nh_policy = NETISR_POLICY_FLOW,
180 };
181 
182 extern	struct domain inetdomain;
183 extern	struct protosw inetsw[];
184 u_char	ip_protox[IPPROTO_MAX];
185 
186 SYSCTL_VNET_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
187     &VNET_NAME(ipstat), ipstat,
188     "IP statistics (struct ipstat, netinet/ip_var.h)");
189 
190 static VNET_DEFINE(uma_zone_t, ipq_zone);
191 #define	V_ipq_zone		VNET(ipq_zone)
192 
193 static struct mtx ipqlock;
194 
195 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
196 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
197 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
198 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
199 
200 static void	maxnipq_update(void);
201 static void	ipq_zone_change(void *);
202 
203 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
204     &VNET_NAME(nipq), 0,
205     "Current number of IPv4 fragment reassembly queue entries");
206 
207 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
208     &VNET_NAME(maxfragsperpacket), 0,
209     "Maximum number of IPv4 fragments allowed per packet");
210 
211 struct callout	ipport_tick_callout;
212 
213 #ifdef IPCTL_DEFMTU
214 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
215     &ip_mtu, 0, "Default MTU");
216 #endif
217 
218 #ifdef IPSTEALTH
219 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
220     &VNET_NAME(ipstealth), 0,
221     "IP stealth mode, no TTL decrementation on forwarding");
222 #endif
223 
224 #ifdef FLOWTABLE
225 static VNET_DEFINE(int, ip_output_flowtable_size) = 2048;
226 VNET_DEFINE(struct flowtable *, ip_ft);
227 #define	V_ip_output_flowtable_size	VNET(ip_output_flowtable_size)
228 
229 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
230     &VNET_NAME(ip_output_flowtable_size), 2048,
231     "number of entries in the per-cpu output flow caches");
232 #endif
233 
234 VNET_DEFINE(int, fw_one_pass) = 1;
235 
236 static void	ip_freef(struct ipqhead *, struct ipq *);
237 
238 /*
239  * Kernel module interface for updating ipstat.  The argument is an index
240  * into ipstat treated as an array of u_long.  While this encodes the general
241  * layout of ipstat into the caller, it doesn't encode its location, so that
242  * future changes to add, for example, per-CPU stats support won't cause
243  * binary compatibility problems for kernel modules.
244  */
245 void
246 kmod_ipstat_inc(int statnum)
247 {
248 
249 	(*((u_long *)&V_ipstat + statnum))++;
250 }
251 
252 void
253 kmod_ipstat_dec(int statnum)
254 {
255 
256 	(*((u_long *)&V_ipstat + statnum))--;
257 }
258 
259 static int
260 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
261 {
262 	int error, qlimit;
263 
264 	netisr_getqlimit(&ip_nh, &qlimit);
265 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
266 	if (error || !req->newptr)
267 		return (error);
268 	if (qlimit < 1)
269 		return (EINVAL);
270 	return (netisr_setqlimit(&ip_nh, qlimit));
271 }
272 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
273     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
274     "Maximum size of the IP input queue");
275 
276 static int
277 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
278 {
279 	u_int64_t qdrops_long;
280 	int error, qdrops;
281 
282 	netisr_getqdrops(&ip_nh, &qdrops_long);
283 	qdrops = qdrops_long;
284 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
285 	if (error || !req->newptr)
286 		return (error);
287 	if (qdrops != 0)
288 		return (EINVAL);
289 	netisr_clearqdrops(&ip_nh);
290 	return (0);
291 }
292 
293 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
294     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
295     "Number of packets dropped from the IP input queue");
296 
297 /*
298  * IP initialization: fill in IP protocol switch table.
299  * All protocols not implemented in kernel go to raw IP protocol handler.
300  */
301 void
302 ip_init(void)
303 {
304 	struct protosw *pr;
305 	int i;
306 
307 	V_ip_id = time_second & 0xffff;
308 
309 	TAILQ_INIT(&V_in_ifaddrhead);
310 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
311 
312 	/* Initialize IP reassembly queue. */
313 	for (i = 0; i < IPREASS_NHASH; i++)
314 		TAILQ_INIT(&V_ipq[i]);
315 	V_maxnipq = nmbclusters / 32;
316 	V_maxfragsperpacket = 16;
317 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
318 	    NULL, UMA_ALIGN_PTR, 0);
319 	maxnipq_update();
320 
321 #ifdef FLOWTABLE
322 	TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size",
323 	    &V_ip_output_flowtable_size);
324 	V_ip_ft = flowtable_alloc(V_ip_output_flowtable_size, FL_PCPU);
325 #endif
326 
327 	/* Skip initialization of globals for non-default instances. */
328 	if (!IS_DEFAULT_VNET(curvnet))
329 		return;
330 
331 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
332 	if (pr == NULL)
333 		panic("ip_init: PF_INET not found");
334 
335 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
336 	for (i = 0; i < IPPROTO_MAX; i++)
337 		ip_protox[i] = pr - inetsw;
338 	/*
339 	 * Cycle through IP protocols and put them into the appropriate place
340 	 * in ip_protox[].
341 	 */
342 	for (pr = inetdomain.dom_protosw;
343 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
344 		if (pr->pr_domain->dom_family == PF_INET &&
345 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
346 			/* Be careful to only index valid IP protocols. */
347 			if (pr->pr_protocol < IPPROTO_MAX)
348 				ip_protox[pr->pr_protocol] = pr - inetsw;
349 		}
350 
351 	/* Initialize packet filter hooks. */
352 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
353 	inet_pfil_hook.ph_af = AF_INET;
354 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
355 		printf("%s: WARNING: unable to register pfil hook, "
356 			"error %d\n", __func__, i);
357 
358 	/* Start ipport_tick. */
359 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
360 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
361 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
362 		SHUTDOWN_PRI_DEFAULT);
363 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
364 		NULL, EVENTHANDLER_PRI_ANY);
365 
366 	/* Initialize various other remaining things. */
367 	IPQ_LOCK_INIT();
368 	netisr_register(&ip_nh);
369 }
370 
371 void
372 ip_fini(void *xtp)
373 {
374 
375 	callout_stop(&ipport_tick_callout);
376 }
377 
378 /*
379  * Ip input routine.  Checksum and byte swap header.  If fragmented
380  * try to reassemble.  Process options.  Pass to next level.
381  */
382 void
383 ip_input(struct mbuf *m)
384 {
385 	struct ip *ip = NULL;
386 	struct in_ifaddr *ia = NULL;
387 	struct ifaddr *ifa;
388 	struct ifnet *ifp;
389 	int    checkif, hlen = 0;
390 	u_short sum;
391 	int dchg = 0;				/* dest changed after fw */
392 	struct in_addr odst;			/* original dst address */
393 
394 	M_ASSERTPKTHDR(m);
395 
396 	if (m->m_flags & M_FASTFWD_OURS) {
397 		/*
398 		 * Firewall or NAT changed destination to local.
399 		 * We expect ip_len and ip_off to be in host byte order.
400 		 */
401 		m->m_flags &= ~M_FASTFWD_OURS;
402 		/* Set up some basics that will be used later. */
403 		ip = mtod(m, struct ip *);
404 		hlen = ip->ip_hl << 2;
405 		goto ours;
406 	}
407 
408 	IPSTAT_INC(ips_total);
409 
410 	if (m->m_pkthdr.len < sizeof(struct ip))
411 		goto tooshort;
412 
413 	if (m->m_len < sizeof (struct ip) &&
414 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
415 		IPSTAT_INC(ips_toosmall);
416 		return;
417 	}
418 	ip = mtod(m, struct ip *);
419 
420 	if (ip->ip_v != IPVERSION) {
421 		IPSTAT_INC(ips_badvers);
422 		goto bad;
423 	}
424 
425 	hlen = ip->ip_hl << 2;
426 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
427 		IPSTAT_INC(ips_badhlen);
428 		goto bad;
429 	}
430 	if (hlen > m->m_len) {
431 		if ((m = m_pullup(m, hlen)) == NULL) {
432 			IPSTAT_INC(ips_badhlen);
433 			return;
434 		}
435 		ip = mtod(m, struct ip *);
436 	}
437 
438 	/* 127/8 must not appear on wire - RFC1122 */
439 	ifp = m->m_pkthdr.rcvif;
440 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
441 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
442 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
443 			IPSTAT_INC(ips_badaddr);
444 			goto bad;
445 		}
446 	}
447 
448 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
449 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
450 	} else {
451 		if (hlen == sizeof(struct ip)) {
452 			sum = in_cksum_hdr(ip);
453 		} else {
454 			sum = in_cksum(m, hlen);
455 		}
456 	}
457 	if (sum) {
458 		IPSTAT_INC(ips_badsum);
459 		goto bad;
460 	}
461 
462 #ifdef ALTQ
463 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
464 		/* packet is dropped by traffic conditioner */
465 		return;
466 #endif
467 
468 	/*
469 	 * Convert fields to host representation.
470 	 */
471 	ip->ip_len = ntohs(ip->ip_len);
472 	if (ip->ip_len < hlen) {
473 		IPSTAT_INC(ips_badlen);
474 		goto bad;
475 	}
476 	ip->ip_off = ntohs(ip->ip_off);
477 
478 	/*
479 	 * Check that the amount of data in the buffers
480 	 * is as at least much as the IP header would have us expect.
481 	 * Trim mbufs if longer than we expect.
482 	 * Drop packet if shorter than we expect.
483 	 */
484 	if (m->m_pkthdr.len < ip->ip_len) {
485 tooshort:
486 		IPSTAT_INC(ips_tooshort);
487 		goto bad;
488 	}
489 	if (m->m_pkthdr.len > ip->ip_len) {
490 		if (m->m_len == m->m_pkthdr.len) {
491 			m->m_len = ip->ip_len;
492 			m->m_pkthdr.len = ip->ip_len;
493 		} else
494 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
495 	}
496 #ifdef IPSEC
497 	/*
498 	 * Bypass packet filtering for packets from a tunnel (gif).
499 	 */
500 	if (ip_ipsec_filtertunnel(m))
501 		goto passin;
502 #endif /* IPSEC */
503 
504 	/*
505 	 * Run through list of hooks for input packets.
506 	 *
507 	 * NB: Beware of the destination address changing (e.g.
508 	 *     by NAT rewriting).  When this happens, tell
509 	 *     ip_forward to do the right thing.
510 	 */
511 
512 	/* Jump over all PFIL processing if hooks are not active. */
513 	if (!PFIL_HOOKED(&inet_pfil_hook))
514 		goto passin;
515 
516 	odst = ip->ip_dst;
517 	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
518 		return;
519 	if (m == NULL)			/* consumed by filter */
520 		return;
521 
522 	ip = mtod(m, struct ip *);
523 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
524 	ifp = m->m_pkthdr.rcvif;
525 
526 #ifdef IPFIREWALL_FORWARD
527 	if (m->m_flags & M_FASTFWD_OURS) {
528 		m->m_flags &= ~M_FASTFWD_OURS;
529 		goto ours;
530 	}
531 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
532 		/*
533 		 * Directly ship on the packet.  This allows to forward packets
534 		 * that were destined for us to some other directly connected
535 		 * host.
536 		 */
537 		ip_forward(m, dchg);
538 		return;
539 	}
540 #endif /* IPFIREWALL_FORWARD */
541 
542 passin:
543 	/*
544 	 * Process options and, if not destined for us,
545 	 * ship it on.  ip_dooptions returns 1 when an
546 	 * error was detected (causing an icmp message
547 	 * to be sent and the original packet to be freed).
548 	 */
549 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
550 		return;
551 
552         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
553          * matter if it is destined to another node, or whether it is
554          * a multicast one, RSVP wants it! and prevents it from being forwarded
555          * anywhere else. Also checks if the rsvp daemon is running before
556 	 * grabbing the packet.
557          */
558 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
559 		goto ours;
560 
561 	/*
562 	 * Check our list of addresses, to see if the packet is for us.
563 	 * If we don't have any addresses, assume any unicast packet
564 	 * we receive might be for us (and let the upper layers deal
565 	 * with it).
566 	 */
567 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
568 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
569 		goto ours;
570 
571 	/*
572 	 * Enable a consistency check between the destination address
573 	 * and the arrival interface for a unicast packet (the RFC 1122
574 	 * strong ES model) if IP forwarding is disabled and the packet
575 	 * is not locally generated and the packet is not subject to
576 	 * 'ipfw fwd'.
577 	 *
578 	 * XXX - Checking also should be disabled if the destination
579 	 * address is ipnat'ed to a different interface.
580 	 *
581 	 * XXX - Checking is incompatible with IP aliases added
582 	 * to the loopback interface instead of the interface where
583 	 * the packets are received.
584 	 *
585 	 * XXX - This is the case for carp vhost IPs as well so we
586 	 * insert a workaround. If the packet got here, we already
587 	 * checked with carp_iamatch() and carp_forus().
588 	 */
589 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
590 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
591 #ifdef DEV_CARP
592 	    !ifp->if_carp &&
593 #endif
594 	    (dchg == 0);
595 
596 	/*
597 	 * Check for exact addresses in the hash bucket.
598 	 */
599 	/* IN_IFADDR_RLOCK(); */
600 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
601 		/*
602 		 * If the address matches, verify that the packet
603 		 * arrived via the correct interface if checking is
604 		 * enabled.
605 		 */
606 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
607 		    (!checkif || ia->ia_ifp == ifp)) {
608 			ifa_ref(&ia->ia_ifa);
609 			/* IN_IFADDR_RUNLOCK(); */
610 			goto ours;
611 		}
612 	}
613 	/* IN_IFADDR_RUNLOCK(); */
614 
615 	/*
616 	 * Check for broadcast addresses.
617 	 *
618 	 * Only accept broadcast packets that arrive via the matching
619 	 * interface.  Reception of forwarded directed broadcasts would
620 	 * be handled via ip_forward() and ether_output() with the loopback
621 	 * into the stack for SIMPLEX interfaces handled by ether_output().
622 	 */
623 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
624 		IF_ADDR_LOCK(ifp);
625 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
626 			if (ifa->ifa_addr->sa_family != AF_INET)
627 				continue;
628 			ia = ifatoia(ifa);
629 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
630 			    ip->ip_dst.s_addr) {
631 				ifa_ref(ifa);
632 				IF_ADDR_UNLOCK(ifp);
633 				goto ours;
634 			}
635 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
636 				ifa_ref(ifa);
637 				IF_ADDR_UNLOCK(ifp);
638 				goto ours;
639 			}
640 #ifdef BOOTP_COMPAT
641 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
642 				ifa_ref(ifa);
643 				IF_ADDR_UNLOCK(ifp);
644 				goto ours;
645 			}
646 #endif
647 		}
648 		IF_ADDR_UNLOCK(ifp);
649 		ia = NULL;
650 	}
651 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
652 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
653 		IPSTAT_INC(ips_cantforward);
654 		m_freem(m);
655 		return;
656 	}
657 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
658 		if (V_ip_mrouter) {
659 			/*
660 			 * If we are acting as a multicast router, all
661 			 * incoming multicast packets are passed to the
662 			 * kernel-level multicast forwarding function.
663 			 * The packet is returned (relatively) intact; if
664 			 * ip_mforward() returns a non-zero value, the packet
665 			 * must be discarded, else it may be accepted below.
666 			 */
667 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
668 				IPSTAT_INC(ips_cantforward);
669 				m_freem(m);
670 				return;
671 			}
672 
673 			/*
674 			 * The process-level routing daemon needs to receive
675 			 * all multicast IGMP packets, whether or not this
676 			 * host belongs to their destination groups.
677 			 */
678 			if (ip->ip_p == IPPROTO_IGMP)
679 				goto ours;
680 			IPSTAT_INC(ips_forward);
681 		}
682 		/*
683 		 * Assume the packet is for us, to avoid prematurely taking
684 		 * a lock on the in_multi hash. Protocols must perform
685 		 * their own filtering and update statistics accordingly.
686 		 */
687 		goto ours;
688 	}
689 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
690 		goto ours;
691 	if (ip->ip_dst.s_addr == INADDR_ANY)
692 		goto ours;
693 
694 	/*
695 	 * FAITH(Firewall Aided Internet Translator)
696 	 */
697 	if (ifp && ifp->if_type == IFT_FAITH) {
698 		if (V_ip_keepfaith) {
699 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
700 				goto ours;
701 		}
702 		m_freem(m);
703 		return;
704 	}
705 
706 	/*
707 	 * Not for us; forward if possible and desirable.
708 	 */
709 	if (V_ipforwarding == 0) {
710 		IPSTAT_INC(ips_cantforward);
711 		m_freem(m);
712 	} else {
713 #ifdef IPSEC
714 		if (ip_ipsec_fwd(m))
715 			goto bad;
716 #endif /* IPSEC */
717 		ip_forward(m, dchg);
718 	}
719 	return;
720 
721 ours:
722 #ifdef IPSTEALTH
723 	/*
724 	 * IPSTEALTH: Process non-routing options only
725 	 * if the packet is destined for us.
726 	 */
727 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) {
728 		if (ia != NULL)
729 			ifa_free(&ia->ia_ifa);
730 		return;
731 	}
732 #endif /* IPSTEALTH */
733 
734 	/* Count the packet in the ip address stats */
735 	if (ia != NULL) {
736 		ia->ia_ifa.if_ipackets++;
737 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
738 		ifa_free(&ia->ia_ifa);
739 	}
740 
741 	/*
742 	 * Attempt reassembly; if it succeeds, proceed.
743 	 * ip_reass() will return a different mbuf.
744 	 */
745 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
746 		m = ip_reass(m);
747 		if (m == NULL)
748 			return;
749 		ip = mtod(m, struct ip *);
750 		/* Get the header length of the reassembled packet */
751 		hlen = ip->ip_hl << 2;
752 	}
753 
754 	/*
755 	 * Further protocols expect the packet length to be w/o the
756 	 * IP header.
757 	 */
758 	ip->ip_len -= hlen;
759 
760 #ifdef IPSEC
761 	/*
762 	 * enforce IPsec policy checking if we are seeing last header.
763 	 * note that we do not visit this with protocols with pcb layer
764 	 * code - like udp/tcp/raw ip.
765 	 */
766 	if (ip_ipsec_input(m))
767 		goto bad;
768 #endif /* IPSEC */
769 
770 	/*
771 	 * Switch out to protocol's input routine.
772 	 */
773 	IPSTAT_INC(ips_delivered);
774 
775 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
776 	return;
777 bad:
778 	m_freem(m);
779 }
780 
781 /*
782  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
783  * max has slightly different semantics than the sysctl, for historical
784  * reasons.
785  */
786 static void
787 maxnipq_update(void)
788 {
789 
790 	/*
791 	 * -1 for unlimited allocation.
792 	 */
793 	if (V_maxnipq < 0)
794 		uma_zone_set_max(V_ipq_zone, 0);
795 	/*
796 	 * Positive number for specific bound.
797 	 */
798 	if (V_maxnipq > 0)
799 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
800 	/*
801 	 * Zero specifies no further fragment queue allocation -- set the
802 	 * bound very low, but rely on implementation elsewhere to actually
803 	 * prevent allocation and reclaim current queues.
804 	 */
805 	if (V_maxnipq == 0)
806 		uma_zone_set_max(V_ipq_zone, 1);
807 }
808 
809 static void
810 ipq_zone_change(void *tag)
811 {
812 
813 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
814 		V_maxnipq = nmbclusters / 32;
815 		maxnipq_update();
816 	}
817 }
818 
819 static int
820 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
821 {
822 	int error, i;
823 
824 	i = V_maxnipq;
825 	error = sysctl_handle_int(oidp, &i, 0, req);
826 	if (error || !req->newptr)
827 		return (error);
828 
829 	/*
830 	 * XXXRW: Might be a good idea to sanity check the argument and place
831 	 * an extreme upper bound.
832 	 */
833 	if (i < -1)
834 		return (EINVAL);
835 	V_maxnipq = i;
836 	maxnipq_update();
837 	return (0);
838 }
839 
840 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
841     NULL, 0, sysctl_maxnipq, "I",
842     "Maximum number of IPv4 fragment reassembly queue entries");
843 
844 /*
845  * Take incoming datagram fragment and try to reassemble it into
846  * whole datagram.  If the argument is the first fragment or one
847  * in between the function will return NULL and store the mbuf
848  * in the fragment chain.  If the argument is the last fragment
849  * the packet will be reassembled and the pointer to the new
850  * mbuf returned for further processing.  Only m_tags attached
851  * to the first packet/fragment are preserved.
852  * The IP header is *NOT* adjusted out of iplen.
853  */
854 struct mbuf *
855 ip_reass(struct mbuf *m)
856 {
857 	struct ip *ip;
858 	struct mbuf *p, *q, *nq, *t;
859 	struct ipq *fp = NULL;
860 	struct ipqhead *head;
861 	int i, hlen, next;
862 	u_int8_t ecn, ecn0;
863 	u_short hash;
864 
865 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
866 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
867 		IPSTAT_INC(ips_fragments);
868 		IPSTAT_INC(ips_fragdropped);
869 		m_freem(m);
870 		return (NULL);
871 	}
872 
873 	ip = mtod(m, struct ip *);
874 	hlen = ip->ip_hl << 2;
875 
876 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
877 	head = &V_ipq[hash];
878 	IPQ_LOCK();
879 
880 	/*
881 	 * Look for queue of fragments
882 	 * of this datagram.
883 	 */
884 	TAILQ_FOREACH(fp, head, ipq_list)
885 		if (ip->ip_id == fp->ipq_id &&
886 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
887 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
888 #ifdef MAC
889 		    mac_ipq_match(m, fp) &&
890 #endif
891 		    ip->ip_p == fp->ipq_p)
892 			goto found;
893 
894 	fp = NULL;
895 
896 	/*
897 	 * Attempt to trim the number of allocated fragment queues if it
898 	 * exceeds the administrative limit.
899 	 */
900 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
901 		/*
902 		 * drop something from the tail of the current queue
903 		 * before proceeding further
904 		 */
905 		struct ipq *q = TAILQ_LAST(head, ipqhead);
906 		if (q == NULL) {   /* gak */
907 			for (i = 0; i < IPREASS_NHASH; i++) {
908 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
909 				if (r) {
910 					IPSTAT_ADD(ips_fragtimeout,
911 					    r->ipq_nfrags);
912 					ip_freef(&V_ipq[i], r);
913 					break;
914 				}
915 			}
916 		} else {
917 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
918 			ip_freef(head, q);
919 		}
920 	}
921 
922 found:
923 	/*
924 	 * Adjust ip_len to not reflect header,
925 	 * convert offset of this to bytes.
926 	 */
927 	ip->ip_len -= hlen;
928 	if (ip->ip_off & IP_MF) {
929 		/*
930 		 * Make sure that fragments have a data length
931 		 * that's a non-zero multiple of 8 bytes.
932 		 */
933 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
934 			IPSTAT_INC(ips_toosmall); /* XXX */
935 			goto dropfrag;
936 		}
937 		m->m_flags |= M_FRAG;
938 	} else
939 		m->m_flags &= ~M_FRAG;
940 	ip->ip_off <<= 3;
941 
942 
943 	/*
944 	 * Attempt reassembly; if it succeeds, proceed.
945 	 * ip_reass() will return a different mbuf.
946 	 */
947 	IPSTAT_INC(ips_fragments);
948 	m->m_pkthdr.header = ip;
949 
950 	/* Previous ip_reass() started here. */
951 	/*
952 	 * Presence of header sizes in mbufs
953 	 * would confuse code below.
954 	 */
955 	m->m_data += hlen;
956 	m->m_len -= hlen;
957 
958 	/*
959 	 * If first fragment to arrive, create a reassembly queue.
960 	 */
961 	if (fp == NULL) {
962 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
963 		if (fp == NULL)
964 			goto dropfrag;
965 #ifdef MAC
966 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
967 			uma_zfree(V_ipq_zone, fp);
968 			fp = NULL;
969 			goto dropfrag;
970 		}
971 		mac_ipq_create(m, fp);
972 #endif
973 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
974 		V_nipq++;
975 		fp->ipq_nfrags = 1;
976 		fp->ipq_ttl = IPFRAGTTL;
977 		fp->ipq_p = ip->ip_p;
978 		fp->ipq_id = ip->ip_id;
979 		fp->ipq_src = ip->ip_src;
980 		fp->ipq_dst = ip->ip_dst;
981 		fp->ipq_frags = m;
982 		m->m_nextpkt = NULL;
983 		goto done;
984 	} else {
985 		fp->ipq_nfrags++;
986 #ifdef MAC
987 		mac_ipq_update(m, fp);
988 #endif
989 	}
990 
991 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
992 
993 	/*
994 	 * Handle ECN by comparing this segment with the first one;
995 	 * if CE is set, do not lose CE.
996 	 * drop if CE and not-ECT are mixed for the same packet.
997 	 */
998 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
999 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
1000 	if (ecn == IPTOS_ECN_CE) {
1001 		if (ecn0 == IPTOS_ECN_NOTECT)
1002 			goto dropfrag;
1003 		if (ecn0 != IPTOS_ECN_CE)
1004 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
1005 	}
1006 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
1007 		goto dropfrag;
1008 
1009 	/*
1010 	 * Find a segment which begins after this one does.
1011 	 */
1012 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1013 		if (GETIP(q)->ip_off > ip->ip_off)
1014 			break;
1015 
1016 	/*
1017 	 * If there is a preceding segment, it may provide some of
1018 	 * our data already.  If so, drop the data from the incoming
1019 	 * segment.  If it provides all of our data, drop us, otherwise
1020 	 * stick new segment in the proper place.
1021 	 *
1022 	 * If some of the data is dropped from the the preceding
1023 	 * segment, then it's checksum is invalidated.
1024 	 */
1025 	if (p) {
1026 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1027 		if (i > 0) {
1028 			if (i >= ip->ip_len)
1029 				goto dropfrag;
1030 			m_adj(m, i);
1031 			m->m_pkthdr.csum_flags = 0;
1032 			ip->ip_off += i;
1033 			ip->ip_len -= i;
1034 		}
1035 		m->m_nextpkt = p->m_nextpkt;
1036 		p->m_nextpkt = m;
1037 	} else {
1038 		m->m_nextpkt = fp->ipq_frags;
1039 		fp->ipq_frags = m;
1040 	}
1041 
1042 	/*
1043 	 * While we overlap succeeding segments trim them or,
1044 	 * if they are completely covered, dequeue them.
1045 	 */
1046 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1047 	     q = nq) {
1048 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1049 		if (i < GETIP(q)->ip_len) {
1050 			GETIP(q)->ip_len -= i;
1051 			GETIP(q)->ip_off += i;
1052 			m_adj(q, i);
1053 			q->m_pkthdr.csum_flags = 0;
1054 			break;
1055 		}
1056 		nq = q->m_nextpkt;
1057 		m->m_nextpkt = nq;
1058 		IPSTAT_INC(ips_fragdropped);
1059 		fp->ipq_nfrags--;
1060 		m_freem(q);
1061 	}
1062 
1063 	/*
1064 	 * Check for complete reassembly and perform frag per packet
1065 	 * limiting.
1066 	 *
1067 	 * Frag limiting is performed here so that the nth frag has
1068 	 * a chance to complete the packet before we drop the packet.
1069 	 * As a result, n+1 frags are actually allowed per packet, but
1070 	 * only n will ever be stored. (n = maxfragsperpacket.)
1071 	 *
1072 	 */
1073 	next = 0;
1074 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1075 		if (GETIP(q)->ip_off != next) {
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 		next += GETIP(q)->ip_len;
1083 	}
1084 	/* Make sure the last packet didn't have the IP_MF flag */
1085 	if (p->m_flags & M_FRAG) {
1086 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1087 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1088 			ip_freef(head, fp);
1089 		}
1090 		goto done;
1091 	}
1092 
1093 	/*
1094 	 * Reassembly is complete.  Make sure the packet is a sane size.
1095 	 */
1096 	q = fp->ipq_frags;
1097 	ip = GETIP(q);
1098 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1099 		IPSTAT_INC(ips_toolong);
1100 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1101 		ip_freef(head, fp);
1102 		goto done;
1103 	}
1104 
1105 	/*
1106 	 * Concatenate fragments.
1107 	 */
1108 	m = q;
1109 	t = m->m_next;
1110 	m->m_next = NULL;
1111 	m_cat(m, t);
1112 	nq = q->m_nextpkt;
1113 	q->m_nextpkt = NULL;
1114 	for (q = nq; q != NULL; q = nq) {
1115 		nq = q->m_nextpkt;
1116 		q->m_nextpkt = NULL;
1117 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1118 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1119 		m_cat(m, q);
1120 	}
1121 	/*
1122 	 * In order to do checksumming faster we do 'end-around carry' here
1123 	 * (and not in for{} loop), though it implies we are not going to
1124 	 * reassemble more than 64k fragments.
1125 	 */
1126 	m->m_pkthdr.csum_data =
1127 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1128 #ifdef MAC
1129 	mac_ipq_reassemble(fp, m);
1130 	mac_ipq_destroy(fp);
1131 #endif
1132 
1133 	/*
1134 	 * Create header for new ip packet by modifying header of first
1135 	 * packet;  dequeue and discard fragment reassembly header.
1136 	 * Make header visible.
1137 	 */
1138 	ip->ip_len = (ip->ip_hl << 2) + next;
1139 	ip->ip_src = fp->ipq_src;
1140 	ip->ip_dst = fp->ipq_dst;
1141 	TAILQ_REMOVE(head, fp, ipq_list);
1142 	V_nipq--;
1143 	uma_zfree(V_ipq_zone, fp);
1144 	m->m_len += (ip->ip_hl << 2);
1145 	m->m_data -= (ip->ip_hl << 2);
1146 	/* some debugging cruft by sklower, below, will go away soon */
1147 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1148 		m_fixhdr(m);
1149 	IPSTAT_INC(ips_reassembled);
1150 	IPQ_UNLOCK();
1151 	return (m);
1152 
1153 dropfrag:
1154 	IPSTAT_INC(ips_fragdropped);
1155 	if (fp != NULL)
1156 		fp->ipq_nfrags--;
1157 	m_freem(m);
1158 done:
1159 	IPQ_UNLOCK();
1160 	return (NULL);
1161 
1162 #undef GETIP
1163 }
1164 
1165 /*
1166  * Free a fragment reassembly header and all
1167  * associated datagrams.
1168  */
1169 static void
1170 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1171 {
1172 	struct mbuf *q;
1173 
1174 	IPQ_LOCK_ASSERT();
1175 
1176 	while (fp->ipq_frags) {
1177 		q = fp->ipq_frags;
1178 		fp->ipq_frags = q->m_nextpkt;
1179 		m_freem(q);
1180 	}
1181 	TAILQ_REMOVE(fhp, fp, ipq_list);
1182 	uma_zfree(V_ipq_zone, fp);
1183 	V_nipq--;
1184 }
1185 
1186 /*
1187  * IP timer processing;
1188  * if a timer expires on a reassembly
1189  * queue, discard it.
1190  */
1191 void
1192 ip_slowtimo(void)
1193 {
1194 	VNET_ITERATOR_DECL(vnet_iter);
1195 	struct ipq *fp;
1196 	int i;
1197 
1198 	VNET_LIST_RLOCK_NOSLEEP();
1199 	IPQ_LOCK();
1200 	VNET_FOREACH(vnet_iter) {
1201 		CURVNET_SET(vnet_iter);
1202 		for (i = 0; i < IPREASS_NHASH; i++) {
1203 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1204 				struct ipq *fpp;
1205 
1206 				fpp = fp;
1207 				fp = TAILQ_NEXT(fp, ipq_list);
1208 				if(--fpp->ipq_ttl == 0) {
1209 					IPSTAT_ADD(ips_fragtimeout,
1210 					    fpp->ipq_nfrags);
1211 					ip_freef(&V_ipq[i], fpp);
1212 				}
1213 			}
1214 		}
1215 		/*
1216 		 * If we are over the maximum number of fragments
1217 		 * (due to the limit being lowered), drain off
1218 		 * enough to get down to the new limit.
1219 		 */
1220 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1221 			for (i = 0; i < IPREASS_NHASH; i++) {
1222 				while (V_nipq > V_maxnipq &&
1223 				    !TAILQ_EMPTY(&V_ipq[i])) {
1224 					IPSTAT_ADD(ips_fragdropped,
1225 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1226 					ip_freef(&V_ipq[i],
1227 					    TAILQ_FIRST(&V_ipq[i]));
1228 				}
1229 			}
1230 		}
1231 		CURVNET_RESTORE();
1232 	}
1233 	IPQ_UNLOCK();
1234 	VNET_LIST_RUNLOCK_NOSLEEP();
1235 }
1236 
1237 /*
1238  * Drain off all datagram fragments.
1239  */
1240 void
1241 ip_drain(void)
1242 {
1243 	VNET_ITERATOR_DECL(vnet_iter);
1244 	int     i;
1245 
1246 	VNET_LIST_RLOCK_NOSLEEP();
1247 	IPQ_LOCK();
1248 	VNET_FOREACH(vnet_iter) {
1249 		CURVNET_SET(vnet_iter);
1250 		for (i = 0; i < IPREASS_NHASH; i++) {
1251 			while(!TAILQ_EMPTY(&V_ipq[i])) {
1252 				IPSTAT_ADD(ips_fragdropped,
1253 				    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1254 				ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1255 			}
1256 		}
1257 		CURVNET_RESTORE();
1258 	}
1259 	IPQ_UNLOCK();
1260 	VNET_LIST_RUNLOCK_NOSLEEP();
1261 	in_rtqdrain();
1262 }
1263 
1264 /*
1265  * The protocol to be inserted into ip_protox[] must be already registered
1266  * in inetsw[], either statically or through pf_proto_register().
1267  */
1268 int
1269 ipproto_register(u_char ipproto)
1270 {
1271 	struct protosw *pr;
1272 
1273 	/* Sanity checks. */
1274 	if (ipproto == 0)
1275 		return (EPROTONOSUPPORT);
1276 
1277 	/*
1278 	 * The protocol slot must not be occupied by another protocol
1279 	 * already.  An index pointing to IPPROTO_RAW is unused.
1280 	 */
1281 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1282 	if (pr == NULL)
1283 		return (EPFNOSUPPORT);
1284 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1285 		return (EEXIST);
1286 
1287 	/* Find the protocol position in inetsw[] and set the index. */
1288 	for (pr = inetdomain.dom_protosw;
1289 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1290 		if (pr->pr_domain->dom_family == PF_INET &&
1291 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1292 			/* Be careful to only index valid IP protocols. */
1293 			if (pr->pr_protocol < IPPROTO_MAX) {
1294 				ip_protox[pr->pr_protocol] = pr - inetsw;
1295 				return (0);
1296 			} else
1297 				return (EINVAL);
1298 		}
1299 	}
1300 	return (EPROTONOSUPPORT);
1301 }
1302 
1303 int
1304 ipproto_unregister(u_char ipproto)
1305 {
1306 	struct protosw *pr;
1307 
1308 	/* Sanity checks. */
1309 	if (ipproto == 0)
1310 		return (EPROTONOSUPPORT);
1311 
1312 	/* Check if the protocol was indeed registered. */
1313 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1314 	if (pr == NULL)
1315 		return (EPFNOSUPPORT);
1316 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1317 		return (ENOENT);
1318 
1319 	/* Reset the protocol slot to IPPROTO_RAW. */
1320 	ip_protox[ipproto] = pr - inetsw;
1321 	return (0);
1322 }
1323 
1324 /*
1325  * Given address of next destination (final or next hop), return (referenced)
1326  * internet address info of interface to be used to get there.
1327  */
1328 struct in_ifaddr *
1329 ip_rtaddr(struct in_addr dst, u_int fibnum)
1330 {
1331 	struct route sro;
1332 	struct sockaddr_in *sin;
1333 	struct in_ifaddr *ia;
1334 
1335 	bzero(&sro, sizeof(sro));
1336 	sin = (struct sockaddr_in *)&sro.ro_dst;
1337 	sin->sin_family = AF_INET;
1338 	sin->sin_len = sizeof(*sin);
1339 	sin->sin_addr = dst;
1340 	in_rtalloc_ign(&sro, 0, fibnum);
1341 
1342 	if (sro.ro_rt == NULL)
1343 		return (NULL);
1344 
1345 	ia = ifatoia(sro.ro_rt->rt_ifa);
1346 	ifa_ref(&ia->ia_ifa);
1347 	RTFREE(sro.ro_rt);
1348 	return (ia);
1349 }
1350 
1351 u_char inetctlerrmap[PRC_NCMDS] = {
1352 	0,		0,		0,		0,
1353 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1354 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1355 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1356 	0,		0,		EHOSTUNREACH,	0,
1357 	ENOPROTOOPT,	ECONNREFUSED
1358 };
1359 
1360 /*
1361  * Forward a packet.  If some error occurs return the sender
1362  * an icmp packet.  Note we can't always generate a meaningful
1363  * icmp message because icmp doesn't have a large enough repertoire
1364  * of codes and types.
1365  *
1366  * If not forwarding, just drop the packet.  This could be confusing
1367  * if ipforwarding was zero but some routing protocol was advancing
1368  * us as a gateway to somewhere.  However, we must let the routing
1369  * protocol deal with that.
1370  *
1371  * The srcrt parameter indicates whether the packet is being forwarded
1372  * via a source route.
1373  */
1374 void
1375 ip_forward(struct mbuf *m, int srcrt)
1376 {
1377 	struct ip *ip = mtod(m, struct ip *);
1378 	struct in_ifaddr *ia;
1379 	struct mbuf *mcopy;
1380 	struct in_addr dest;
1381 	struct route ro;
1382 	int error, type = 0, code = 0, mtu = 0;
1383 
1384 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1385 		IPSTAT_INC(ips_cantforward);
1386 		m_freem(m);
1387 		return;
1388 	}
1389 #ifdef IPSTEALTH
1390 	if (!V_ipstealth) {
1391 #endif
1392 		if (ip->ip_ttl <= IPTTLDEC) {
1393 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1394 			    0, 0);
1395 			return;
1396 		}
1397 #ifdef IPSTEALTH
1398 	}
1399 #endif
1400 
1401 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1402 #ifndef IPSEC
1403 	/*
1404 	 * 'ia' may be NULL if there is no route for this destination.
1405 	 * In case of IPsec, Don't discard it just yet, but pass it to
1406 	 * ip_output in case of outgoing IPsec policy.
1407 	 */
1408 	if (!srcrt && ia == NULL) {
1409 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1410 		return;
1411 	}
1412 #endif
1413 
1414 	/*
1415 	 * Save the IP header and at most 8 bytes of the payload,
1416 	 * in case we need to generate an ICMP message to the src.
1417 	 *
1418 	 * XXX this can be optimized a lot by saving the data in a local
1419 	 * buffer on the stack (72 bytes at most), and only allocating the
1420 	 * mbuf if really necessary. The vast majority of the packets
1421 	 * are forwarded without having to send an ICMP back (either
1422 	 * because unnecessary, or because rate limited), so we are
1423 	 * really we are wasting a lot of work here.
1424 	 *
1425 	 * We don't use m_copy() because it might return a reference
1426 	 * to a shared cluster. Both this function and ip_output()
1427 	 * assume exclusive access to the IP header in `m', so any
1428 	 * data in a cluster may change before we reach icmp_error().
1429 	 */
1430 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1431 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1432 		/*
1433 		 * It's probably ok if the pkthdr dup fails (because
1434 		 * the deep copy of the tag chain failed), but for now
1435 		 * be conservative and just discard the copy since
1436 		 * code below may some day want the tags.
1437 		 */
1438 		m_free(mcopy);
1439 		mcopy = NULL;
1440 	}
1441 	if (mcopy != NULL) {
1442 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1443 		mcopy->m_pkthdr.len = mcopy->m_len;
1444 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1445 	}
1446 
1447 #ifdef IPSTEALTH
1448 	if (!V_ipstealth) {
1449 #endif
1450 		ip->ip_ttl -= IPTTLDEC;
1451 #ifdef IPSTEALTH
1452 	}
1453 #endif
1454 
1455 	/*
1456 	 * If forwarding packet using same interface that it came in on,
1457 	 * perhaps should send a redirect to sender to shortcut a hop.
1458 	 * Only send redirect if source is sending directly to us,
1459 	 * and if packet was not source routed (or has any options).
1460 	 * Also, don't send redirect if forwarding using a default route
1461 	 * or a route modified by a redirect.
1462 	 */
1463 	dest.s_addr = 0;
1464 	if (!srcrt && V_ipsendredirects &&
1465 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1466 		struct sockaddr_in *sin;
1467 		struct rtentry *rt;
1468 
1469 		bzero(&ro, sizeof(ro));
1470 		sin = (struct sockaddr_in *)&ro.ro_dst;
1471 		sin->sin_family = AF_INET;
1472 		sin->sin_len = sizeof(*sin);
1473 		sin->sin_addr = ip->ip_dst;
1474 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1475 
1476 		rt = ro.ro_rt;
1477 
1478 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1479 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1480 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1481 			u_long src = ntohl(ip->ip_src.s_addr);
1482 
1483 			if (RTA(rt) &&
1484 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1485 				if (rt->rt_flags & RTF_GATEWAY)
1486 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1487 				else
1488 					dest.s_addr = ip->ip_dst.s_addr;
1489 				/* Router requirements says to only send host redirects */
1490 				type = ICMP_REDIRECT;
1491 				code = ICMP_REDIRECT_HOST;
1492 			}
1493 		}
1494 		if (rt)
1495 			RTFREE(rt);
1496 	}
1497 
1498 	/*
1499 	 * Try to cache the route MTU from ip_output so we can consider it for
1500 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1501 	 */
1502 	bzero(&ro, sizeof(ro));
1503 
1504 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1505 
1506 	if (error == EMSGSIZE && ro.ro_rt)
1507 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1508 	if (ro.ro_rt)
1509 		RTFREE(ro.ro_rt);
1510 
1511 	if (error)
1512 		IPSTAT_INC(ips_cantforward);
1513 	else {
1514 		IPSTAT_INC(ips_forward);
1515 		if (type)
1516 			IPSTAT_INC(ips_redirectsent);
1517 		else {
1518 			if (mcopy)
1519 				m_freem(mcopy);
1520 			if (ia != NULL)
1521 				ifa_free(&ia->ia_ifa);
1522 			return;
1523 		}
1524 	}
1525 	if (mcopy == NULL) {
1526 		if (ia != NULL)
1527 			ifa_free(&ia->ia_ifa);
1528 		return;
1529 	}
1530 
1531 	switch (error) {
1532 
1533 	case 0:				/* forwarded, but need redirect */
1534 		/* type, code set above */
1535 		break;
1536 
1537 	case ENETUNREACH:
1538 	case EHOSTUNREACH:
1539 	case ENETDOWN:
1540 	case EHOSTDOWN:
1541 	default:
1542 		type = ICMP_UNREACH;
1543 		code = ICMP_UNREACH_HOST;
1544 		break;
1545 
1546 	case EMSGSIZE:
1547 		type = ICMP_UNREACH;
1548 		code = ICMP_UNREACH_NEEDFRAG;
1549 
1550 #ifdef IPSEC
1551 		/*
1552 		 * If IPsec is configured for this path,
1553 		 * override any possibly mtu value set by ip_output.
1554 		 */
1555 		mtu = ip_ipsec_mtu(m, mtu);
1556 #endif /* IPSEC */
1557 		/*
1558 		 * If the MTU was set before make sure we are below the
1559 		 * interface MTU.
1560 		 * If the MTU wasn't set before use the interface mtu or
1561 		 * fall back to the next smaller mtu step compared to the
1562 		 * current packet size.
1563 		 */
1564 		if (mtu != 0) {
1565 			if (ia != NULL)
1566 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1567 		} else {
1568 			if (ia != NULL)
1569 				mtu = ia->ia_ifp->if_mtu;
1570 			else
1571 				mtu = ip_next_mtu(ip->ip_len, 0);
1572 		}
1573 		IPSTAT_INC(ips_cantfrag);
1574 		break;
1575 
1576 	case ENOBUFS:
1577 		/*
1578 		 * A router should not generate ICMP_SOURCEQUENCH as
1579 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1580 		 * Source quench could be a big problem under DoS attacks,
1581 		 * or if the underlying interface is rate-limited.
1582 		 * Those who need source quench packets may re-enable them
1583 		 * via the net.inet.ip.sendsourcequench sysctl.
1584 		 */
1585 		if (V_ip_sendsourcequench == 0) {
1586 			m_freem(mcopy);
1587 			if (ia != NULL)
1588 				ifa_free(&ia->ia_ifa);
1589 			return;
1590 		} else {
1591 			type = ICMP_SOURCEQUENCH;
1592 			code = 0;
1593 		}
1594 		break;
1595 
1596 	case EACCES:			/* ipfw denied packet */
1597 		m_freem(mcopy);
1598 		if (ia != NULL)
1599 			ifa_free(&ia->ia_ifa);
1600 		return;
1601 	}
1602 	if (ia != NULL)
1603 		ifa_free(&ia->ia_ifa);
1604 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1605 }
1606 
1607 void
1608 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1609     struct mbuf *m)
1610 {
1611 
1612 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1613 		struct bintime bt;
1614 
1615 		bintime(&bt);
1616 		if (inp->inp_socket->so_options & SO_BINTIME) {
1617 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1618 			SCM_BINTIME, SOL_SOCKET);
1619 			if (*mp)
1620 				mp = &(*mp)->m_next;
1621 		}
1622 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1623 			struct timeval tv;
1624 
1625 			bintime2timeval(&bt, &tv);
1626 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1627 				SCM_TIMESTAMP, SOL_SOCKET);
1628 			if (*mp)
1629 				mp = &(*mp)->m_next;
1630 		}
1631 	}
1632 	if (inp->inp_flags & INP_RECVDSTADDR) {
1633 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1634 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1635 		if (*mp)
1636 			mp = &(*mp)->m_next;
1637 	}
1638 	if (inp->inp_flags & INP_RECVTTL) {
1639 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1640 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1641 		if (*mp)
1642 			mp = &(*mp)->m_next;
1643 	}
1644 #ifdef notyet
1645 	/* XXX
1646 	 * Moving these out of udp_input() made them even more broken
1647 	 * than they already were.
1648 	 */
1649 	/* options were tossed already */
1650 	if (inp->inp_flags & INP_RECVOPTS) {
1651 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1652 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1653 		if (*mp)
1654 			mp = &(*mp)->m_next;
1655 	}
1656 	/* ip_srcroute doesn't do what we want here, need to fix */
1657 	if (inp->inp_flags & INP_RECVRETOPTS) {
1658 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1659 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1660 		if (*mp)
1661 			mp = &(*mp)->m_next;
1662 	}
1663 #endif
1664 	if (inp->inp_flags & INP_RECVIF) {
1665 		struct ifnet *ifp;
1666 		struct sdlbuf {
1667 			struct sockaddr_dl sdl;
1668 			u_char	pad[32];
1669 		} sdlbuf;
1670 		struct sockaddr_dl *sdp;
1671 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1672 
1673 		if (((ifp = m->m_pkthdr.rcvif))
1674 		&& ( ifp->if_index && (ifp->if_index <= V_if_index))) {
1675 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1676 			/*
1677 			 * Change our mind and don't try copy.
1678 			 */
1679 			if ((sdp->sdl_family != AF_LINK)
1680 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1681 				goto makedummy;
1682 			}
1683 			bcopy(sdp, sdl2, sdp->sdl_len);
1684 		} else {
1685 makedummy:
1686 			sdl2->sdl_len
1687 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1688 			sdl2->sdl_family = AF_LINK;
1689 			sdl2->sdl_index = 0;
1690 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1691 		}
1692 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1693 			IP_RECVIF, IPPROTO_IP);
1694 		if (*mp)
1695 			mp = &(*mp)->m_next;
1696 	}
1697 }
1698 
1699 /*
1700  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1701  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1702  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1703  * compiled.
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