xref: /freebsd/sys/netinet/ip_input.c (revision 9fd69f37d28cfd7438cac3eeb45fe9dd46b4d7dd)
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 VNET_DEFINE(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 static void	ip_drain_locked(void);
203 
204 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
205     &VNET_NAME(nipq), 0,
206     "Current number of IPv4 fragment reassembly queue entries");
207 
208 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
209     &VNET_NAME(maxfragsperpacket), 0,
210     "Maximum number of IPv4 fragments allowed per packet");
211 
212 struct callout	ipport_tick_callout;
213 
214 #ifdef IPCTL_DEFMTU
215 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
216     &ip_mtu, 0, "Default MTU");
217 #endif
218 
219 #ifdef IPSTEALTH
220 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
221     &VNET_NAME(ipstealth), 0,
222     "IP stealth mode, no TTL decrementation on forwarding");
223 #endif
224 
225 #ifdef FLOWTABLE
226 static VNET_DEFINE(int, ip_output_flowtable_size) = 2048;
227 VNET_DEFINE(struct flowtable *, ip_ft);
228 #define	V_ip_output_flowtable_size	VNET(ip_output_flowtable_size)
229 
230 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
231     &VNET_NAME(ip_output_flowtable_size), 2048,
232     "number of entries in the per-cpu output flow caches");
233 #endif
234 
235 VNET_DEFINE(int, fw_one_pass) = 1;
236 
237 static void	ip_freef(struct ipqhead *, struct ipq *);
238 
239 /*
240  * Kernel module interface for updating ipstat.  The argument is an index
241  * into ipstat treated as an array of u_long.  While this encodes the general
242  * layout of ipstat into the caller, it doesn't encode its location, so that
243  * future changes to add, for example, per-CPU stats support won't cause
244  * binary compatibility problems for kernel modules.
245  */
246 void
247 kmod_ipstat_inc(int statnum)
248 {
249 
250 	(*((u_long *)&V_ipstat + statnum))++;
251 }
252 
253 void
254 kmod_ipstat_dec(int statnum)
255 {
256 
257 	(*((u_long *)&V_ipstat + statnum))--;
258 }
259 
260 static int
261 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
262 {
263 	int error, qlimit;
264 
265 	netisr_getqlimit(&ip_nh, &qlimit);
266 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
267 	if (error || !req->newptr)
268 		return (error);
269 	if (qlimit < 1)
270 		return (EINVAL);
271 	return (netisr_setqlimit(&ip_nh, qlimit));
272 }
273 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
274     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
275     "Maximum size of the IP input queue");
276 
277 static int
278 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
279 {
280 	u_int64_t qdrops_long;
281 	int error, qdrops;
282 
283 	netisr_getqdrops(&ip_nh, &qdrops_long);
284 	qdrops = qdrops_long;
285 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
286 	if (error || !req->newptr)
287 		return (error);
288 	if (qdrops != 0)
289 		return (EINVAL);
290 	netisr_clearqdrops(&ip_nh);
291 	return (0);
292 }
293 
294 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
295     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
296     "Number of packets dropped from the IP input queue");
297 
298 /*
299  * IP initialization: fill in IP protocol switch table.
300  * All protocols not implemented in kernel go to raw IP protocol handler.
301  */
302 void
303 ip_init(void)
304 {
305 	struct protosw *pr;
306 	int i;
307 
308 	V_ip_id = time_second & 0xffff;
309 
310 	TAILQ_INIT(&V_in_ifaddrhead);
311 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
312 
313 	/* Initialize IP reassembly queue. */
314 	for (i = 0; i < IPREASS_NHASH; i++)
315 		TAILQ_INIT(&V_ipq[i]);
316 	V_maxnipq = nmbclusters / 32;
317 	V_maxfragsperpacket = 16;
318 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
319 	    NULL, UMA_ALIGN_PTR, 0);
320 	maxnipq_update();
321 
322 	/* Initialize packet filter hooks. */
323 	V_inet_pfil_hook.ph_type = PFIL_TYPE_AF;
324 	V_inet_pfil_hook.ph_af = AF_INET;
325 	if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0)
326 		printf("%s: WARNING: unable to register pfil hook, "
327 			"error %d\n", __func__, i);
328 
329 #ifdef FLOWTABLE
330 	TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size",
331 	    &V_ip_output_flowtable_size);
332 	V_ip_ft = flowtable_alloc(V_ip_output_flowtable_size, FL_PCPU);
333 #endif
334 
335 	/* Skip initialization of globals for non-default instances. */
336 	if (!IS_DEFAULT_VNET(curvnet))
337 		return;
338 
339 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
340 	if (pr == NULL)
341 		panic("ip_init: PF_INET not found");
342 
343 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
344 	for (i = 0; i < IPPROTO_MAX; i++)
345 		ip_protox[i] = pr - inetsw;
346 	/*
347 	 * Cycle through IP protocols and put them into the appropriate place
348 	 * in ip_protox[].
349 	 */
350 	for (pr = inetdomain.dom_protosw;
351 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
352 		if (pr->pr_domain->dom_family == PF_INET &&
353 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
354 			/* Be careful to only index valid IP protocols. */
355 			if (pr->pr_protocol < IPPROTO_MAX)
356 				ip_protox[pr->pr_protocol] = pr - inetsw;
357 		}
358 
359 	/* Start ipport_tick. */
360 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
361 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
362 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
363 		SHUTDOWN_PRI_DEFAULT);
364 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
365 		NULL, EVENTHANDLER_PRI_ANY);
366 
367 	/* Initialize various other remaining things. */
368 	IPQ_LOCK_INIT();
369 	netisr_register(&ip_nh);
370 }
371 
372 #ifdef VIMAGE
373 void
374 ip_destroy(void)
375 {
376 
377 	/* Cleanup in_ifaddr hash table; should be empty. */
378 	hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask);
379 
380 	IPQ_LOCK();
381 	ip_drain_locked();
382 	IPQ_UNLOCK();
383 
384 	uma_zdestroy(V_ipq_zone);
385 }
386 #endif
387 
388 void
389 ip_fini(void *xtp)
390 {
391 
392 	callout_stop(&ipport_tick_callout);
393 }
394 
395 /*
396  * Ip input routine.  Checksum and byte swap header.  If fragmented
397  * try to reassemble.  Process options.  Pass to next level.
398  */
399 void
400 ip_input(struct mbuf *m)
401 {
402 	struct ip *ip = NULL;
403 	struct in_ifaddr *ia = NULL;
404 	struct ifaddr *ifa;
405 	struct ifnet *ifp;
406 	int    checkif, hlen = 0;
407 	u_short sum;
408 	int dchg = 0;				/* dest changed after fw */
409 	struct in_addr odst;			/* original dst address */
410 
411 	M_ASSERTPKTHDR(m);
412 
413 	if (m->m_flags & M_FASTFWD_OURS) {
414 		/*
415 		 * Firewall or NAT changed destination to local.
416 		 * We expect ip_len and ip_off to be in host byte order.
417 		 */
418 		m->m_flags &= ~M_FASTFWD_OURS;
419 		/* Set up some basics that will be used later. */
420 		ip = mtod(m, struct ip *);
421 		hlen = ip->ip_hl << 2;
422 		goto ours;
423 	}
424 
425 	IPSTAT_INC(ips_total);
426 
427 	if (m->m_pkthdr.len < sizeof(struct ip))
428 		goto tooshort;
429 
430 	if (m->m_len < sizeof (struct ip) &&
431 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
432 		IPSTAT_INC(ips_toosmall);
433 		return;
434 	}
435 	ip = mtod(m, struct ip *);
436 
437 	if (ip->ip_v != IPVERSION) {
438 		IPSTAT_INC(ips_badvers);
439 		goto bad;
440 	}
441 
442 	hlen = ip->ip_hl << 2;
443 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
444 		IPSTAT_INC(ips_badhlen);
445 		goto bad;
446 	}
447 	if (hlen > m->m_len) {
448 		if ((m = m_pullup(m, hlen)) == NULL) {
449 			IPSTAT_INC(ips_badhlen);
450 			return;
451 		}
452 		ip = mtod(m, struct ip *);
453 	}
454 
455 	/* 127/8 must not appear on wire - RFC1122 */
456 	ifp = m->m_pkthdr.rcvif;
457 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
458 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
459 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
460 			IPSTAT_INC(ips_badaddr);
461 			goto bad;
462 		}
463 	}
464 
465 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
466 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
467 	} else {
468 		if (hlen == sizeof(struct ip)) {
469 			sum = in_cksum_hdr(ip);
470 		} else {
471 			sum = in_cksum(m, hlen);
472 		}
473 	}
474 	if (sum) {
475 		IPSTAT_INC(ips_badsum);
476 		goto bad;
477 	}
478 
479 #ifdef ALTQ
480 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
481 		/* packet is dropped by traffic conditioner */
482 		return;
483 #endif
484 
485 	/*
486 	 * Convert fields to host representation.
487 	 */
488 	ip->ip_len = ntohs(ip->ip_len);
489 	if (ip->ip_len < hlen) {
490 		IPSTAT_INC(ips_badlen);
491 		goto bad;
492 	}
493 	ip->ip_off = ntohs(ip->ip_off);
494 
495 	/*
496 	 * Check that the amount of data in the buffers
497 	 * is as at least much as the IP header would have us expect.
498 	 * Trim mbufs if longer than we expect.
499 	 * Drop packet if shorter than we expect.
500 	 */
501 	if (m->m_pkthdr.len < ip->ip_len) {
502 tooshort:
503 		IPSTAT_INC(ips_tooshort);
504 		goto bad;
505 	}
506 	if (m->m_pkthdr.len > ip->ip_len) {
507 		if (m->m_len == m->m_pkthdr.len) {
508 			m->m_len = ip->ip_len;
509 			m->m_pkthdr.len = ip->ip_len;
510 		} else
511 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
512 	}
513 #ifdef IPSEC
514 	/*
515 	 * Bypass packet filtering for packets from a tunnel (gif).
516 	 */
517 	if (ip_ipsec_filtertunnel(m))
518 		goto passin;
519 #endif /* IPSEC */
520 
521 	/*
522 	 * Run through list of hooks for input packets.
523 	 *
524 	 * NB: Beware of the destination address changing (e.g.
525 	 *     by NAT rewriting).  When this happens, tell
526 	 *     ip_forward to do the right thing.
527 	 */
528 
529 	/* Jump over all PFIL processing if hooks are not active. */
530 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
531 		goto passin;
532 
533 	odst = ip->ip_dst;
534 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
535 		return;
536 	if (m == NULL)			/* consumed by filter */
537 		return;
538 
539 	ip = mtod(m, struct ip *);
540 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
541 	ifp = m->m_pkthdr.rcvif;
542 
543 #ifdef IPFIREWALL_FORWARD
544 	if (m->m_flags & M_FASTFWD_OURS) {
545 		m->m_flags &= ~M_FASTFWD_OURS;
546 		goto ours;
547 	}
548 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
549 		/*
550 		 * Directly ship the packet on.  This allows forwarding
551 		 * packets originally destined to us to some other directly
552 		 * connected host.
553 		 */
554 		ip_forward(m, dchg);
555 		return;
556 	}
557 #endif /* IPFIREWALL_FORWARD */
558 
559 passin:
560 	/*
561 	 * Process options and, if not destined for us,
562 	 * ship it on.  ip_dooptions returns 1 when an
563 	 * error was detected (causing an icmp message
564 	 * to be sent and the original packet to be freed).
565 	 */
566 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
567 		return;
568 
569         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
570          * matter if it is destined to another node, or whether it is
571          * a multicast one, RSVP wants it! and prevents it from being forwarded
572          * anywhere else. Also checks if the rsvp daemon is running before
573 	 * grabbing the packet.
574          */
575 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
576 		goto ours;
577 
578 	/*
579 	 * Check our list of addresses, to see if the packet is for us.
580 	 * If we don't have any addresses, assume any unicast packet
581 	 * we receive might be for us (and let the upper layers deal
582 	 * with it).
583 	 */
584 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
585 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
586 		goto ours;
587 
588 	/*
589 	 * Enable a consistency check between the destination address
590 	 * and the arrival interface for a unicast packet (the RFC 1122
591 	 * strong ES model) if IP forwarding is disabled and the packet
592 	 * is not locally generated and the packet is not subject to
593 	 * 'ipfw fwd'.
594 	 *
595 	 * XXX - Checking also should be disabled if the destination
596 	 * address is ipnat'ed to a different interface.
597 	 *
598 	 * XXX - Checking is incompatible with IP aliases added
599 	 * to the loopback interface instead of the interface where
600 	 * the packets are received.
601 	 *
602 	 * XXX - This is the case for carp vhost IPs as well so we
603 	 * insert a workaround. If the packet got here, we already
604 	 * checked with carp_iamatch() and carp_forus().
605 	 */
606 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
607 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
608 #ifdef DEV_CARP
609 	    !ifp->if_carp &&
610 #endif
611 	    (dchg == 0);
612 
613 	/*
614 	 * Check for exact addresses in the hash bucket.
615 	 */
616 	/* IN_IFADDR_RLOCK(); */
617 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
618 		/*
619 		 * If the address matches, verify that the packet
620 		 * arrived via the correct interface if checking is
621 		 * enabled.
622 		 */
623 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
624 		    (!checkif || ia->ia_ifp == ifp)) {
625 			ifa_ref(&ia->ia_ifa);
626 			/* IN_IFADDR_RUNLOCK(); */
627 			goto ours;
628 		}
629 	}
630 	/* IN_IFADDR_RUNLOCK(); */
631 
632 	/*
633 	 * Check for broadcast addresses.
634 	 *
635 	 * Only accept broadcast packets that arrive via the matching
636 	 * interface.  Reception of forwarded directed broadcasts would
637 	 * be handled via ip_forward() and ether_output() with the loopback
638 	 * into the stack for SIMPLEX interfaces handled by ether_output().
639 	 */
640 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
641 		IF_ADDR_LOCK(ifp);
642 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
643 			if (ifa->ifa_addr->sa_family != AF_INET)
644 				continue;
645 			ia = ifatoia(ifa);
646 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
647 			    ip->ip_dst.s_addr) {
648 				ifa_ref(ifa);
649 				IF_ADDR_UNLOCK(ifp);
650 				goto ours;
651 			}
652 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
653 				ifa_ref(ifa);
654 				IF_ADDR_UNLOCK(ifp);
655 				goto ours;
656 			}
657 #ifdef BOOTP_COMPAT
658 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
659 				ifa_ref(ifa);
660 				IF_ADDR_UNLOCK(ifp);
661 				goto ours;
662 			}
663 #endif
664 		}
665 		IF_ADDR_UNLOCK(ifp);
666 		ia = NULL;
667 	}
668 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
669 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
670 		IPSTAT_INC(ips_cantforward);
671 		m_freem(m);
672 		return;
673 	}
674 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
675 		if (V_ip_mrouter) {
676 			/*
677 			 * If we are acting as a multicast router, all
678 			 * incoming multicast packets are passed to the
679 			 * kernel-level multicast forwarding function.
680 			 * The packet is returned (relatively) intact; if
681 			 * ip_mforward() returns a non-zero value, the packet
682 			 * must be discarded, else it may be accepted below.
683 			 */
684 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
685 				IPSTAT_INC(ips_cantforward);
686 				m_freem(m);
687 				return;
688 			}
689 
690 			/*
691 			 * The process-level routing daemon needs to receive
692 			 * all multicast IGMP packets, whether or not this
693 			 * host belongs to their destination groups.
694 			 */
695 			if (ip->ip_p == IPPROTO_IGMP)
696 				goto ours;
697 			IPSTAT_INC(ips_forward);
698 		}
699 		/*
700 		 * Assume the packet is for us, to avoid prematurely taking
701 		 * a lock on the in_multi hash. Protocols must perform
702 		 * their own filtering and update statistics accordingly.
703 		 */
704 		goto ours;
705 	}
706 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
707 		goto ours;
708 	if (ip->ip_dst.s_addr == INADDR_ANY)
709 		goto ours;
710 
711 	/*
712 	 * FAITH(Firewall Aided Internet Translator)
713 	 */
714 	if (ifp && ifp->if_type == IFT_FAITH) {
715 		if (V_ip_keepfaith) {
716 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
717 				goto ours;
718 		}
719 		m_freem(m);
720 		return;
721 	}
722 
723 	/*
724 	 * Not for us; forward if possible and desirable.
725 	 */
726 	if (V_ipforwarding == 0) {
727 		IPSTAT_INC(ips_cantforward);
728 		m_freem(m);
729 	} else {
730 #ifdef IPSEC
731 		if (ip_ipsec_fwd(m))
732 			goto bad;
733 #endif /* IPSEC */
734 		ip_forward(m, dchg);
735 	}
736 	return;
737 
738 ours:
739 #ifdef IPSTEALTH
740 	/*
741 	 * IPSTEALTH: Process non-routing options only
742 	 * if the packet is destined for us.
743 	 */
744 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) {
745 		if (ia != NULL)
746 			ifa_free(&ia->ia_ifa);
747 		return;
748 	}
749 #endif /* IPSTEALTH */
750 
751 	/* Count the packet in the ip address stats */
752 	if (ia != NULL) {
753 		ia->ia_ifa.if_ipackets++;
754 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
755 		ifa_free(&ia->ia_ifa);
756 	}
757 
758 	/*
759 	 * Attempt reassembly; if it succeeds, proceed.
760 	 * ip_reass() will return a different mbuf.
761 	 */
762 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
763 		m = ip_reass(m);
764 		if (m == NULL)
765 			return;
766 		ip = mtod(m, struct ip *);
767 		/* Get the header length of the reassembled packet */
768 		hlen = ip->ip_hl << 2;
769 	}
770 
771 	/*
772 	 * Further protocols expect the packet length to be w/o the
773 	 * IP header.
774 	 */
775 	ip->ip_len -= hlen;
776 
777 #ifdef IPSEC
778 	/*
779 	 * enforce IPsec policy checking if we are seeing last header.
780 	 * note that we do not visit this with protocols with pcb layer
781 	 * code - like udp/tcp/raw ip.
782 	 */
783 	if (ip_ipsec_input(m))
784 		goto bad;
785 #endif /* IPSEC */
786 
787 	/*
788 	 * Switch out to protocol's input routine.
789 	 */
790 	IPSTAT_INC(ips_delivered);
791 
792 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
793 	return;
794 bad:
795 	m_freem(m);
796 }
797 
798 /*
799  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
800  * max has slightly different semantics than the sysctl, for historical
801  * reasons.
802  */
803 static void
804 maxnipq_update(void)
805 {
806 
807 	/*
808 	 * -1 for unlimited allocation.
809 	 */
810 	if (V_maxnipq < 0)
811 		uma_zone_set_max(V_ipq_zone, 0);
812 	/*
813 	 * Positive number for specific bound.
814 	 */
815 	if (V_maxnipq > 0)
816 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
817 	/*
818 	 * Zero specifies no further fragment queue allocation -- set the
819 	 * bound very low, but rely on implementation elsewhere to actually
820 	 * prevent allocation and reclaim current queues.
821 	 */
822 	if (V_maxnipq == 0)
823 		uma_zone_set_max(V_ipq_zone, 1);
824 }
825 
826 static void
827 ipq_zone_change(void *tag)
828 {
829 
830 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
831 		V_maxnipq = nmbclusters / 32;
832 		maxnipq_update();
833 	}
834 }
835 
836 static int
837 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
838 {
839 	int error, i;
840 
841 	i = V_maxnipq;
842 	error = sysctl_handle_int(oidp, &i, 0, req);
843 	if (error || !req->newptr)
844 		return (error);
845 
846 	/*
847 	 * XXXRW: Might be a good idea to sanity check the argument and place
848 	 * an extreme upper bound.
849 	 */
850 	if (i < -1)
851 		return (EINVAL);
852 	V_maxnipq = i;
853 	maxnipq_update();
854 	return (0);
855 }
856 
857 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
858     NULL, 0, sysctl_maxnipq, "I",
859     "Maximum number of IPv4 fragment reassembly queue entries");
860 
861 /*
862  * Take incoming datagram fragment and try to reassemble it into
863  * whole datagram.  If the argument is the first fragment or one
864  * in between the function will return NULL and store the mbuf
865  * in the fragment chain.  If the argument is the last fragment
866  * the packet will be reassembled and the pointer to the new
867  * mbuf returned for further processing.  Only m_tags attached
868  * to the first packet/fragment are preserved.
869  * The IP header is *NOT* adjusted out of iplen.
870  */
871 struct mbuf *
872 ip_reass(struct mbuf *m)
873 {
874 	struct ip *ip;
875 	struct mbuf *p, *q, *nq, *t;
876 	struct ipq *fp = NULL;
877 	struct ipqhead *head;
878 	int i, hlen, next;
879 	u_int8_t ecn, ecn0;
880 	u_short hash;
881 
882 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
883 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
884 		IPSTAT_INC(ips_fragments);
885 		IPSTAT_INC(ips_fragdropped);
886 		m_freem(m);
887 		return (NULL);
888 	}
889 
890 	ip = mtod(m, struct ip *);
891 	hlen = ip->ip_hl << 2;
892 
893 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
894 	head = &V_ipq[hash];
895 	IPQ_LOCK();
896 
897 	/*
898 	 * Look for queue of fragments
899 	 * of this datagram.
900 	 */
901 	TAILQ_FOREACH(fp, head, ipq_list)
902 		if (ip->ip_id == fp->ipq_id &&
903 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
904 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
905 #ifdef MAC
906 		    mac_ipq_match(m, fp) &&
907 #endif
908 		    ip->ip_p == fp->ipq_p)
909 			goto found;
910 
911 	fp = NULL;
912 
913 	/*
914 	 * Attempt to trim the number of allocated fragment queues if it
915 	 * exceeds the administrative limit.
916 	 */
917 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
918 		/*
919 		 * drop something from the tail of the current queue
920 		 * before proceeding further
921 		 */
922 		struct ipq *q = TAILQ_LAST(head, ipqhead);
923 		if (q == NULL) {   /* gak */
924 			for (i = 0; i < IPREASS_NHASH; i++) {
925 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
926 				if (r) {
927 					IPSTAT_ADD(ips_fragtimeout,
928 					    r->ipq_nfrags);
929 					ip_freef(&V_ipq[i], r);
930 					break;
931 				}
932 			}
933 		} else {
934 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
935 			ip_freef(head, q);
936 		}
937 	}
938 
939 found:
940 	/*
941 	 * Adjust ip_len to not reflect header,
942 	 * convert offset of this to bytes.
943 	 */
944 	ip->ip_len -= hlen;
945 	if (ip->ip_off & IP_MF) {
946 		/*
947 		 * Make sure that fragments have a data length
948 		 * that's a non-zero multiple of 8 bytes.
949 		 */
950 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
951 			IPSTAT_INC(ips_toosmall); /* XXX */
952 			goto dropfrag;
953 		}
954 		m->m_flags |= M_FRAG;
955 	} else
956 		m->m_flags &= ~M_FRAG;
957 	ip->ip_off <<= 3;
958 
959 
960 	/*
961 	 * Attempt reassembly; if it succeeds, proceed.
962 	 * ip_reass() will return a different mbuf.
963 	 */
964 	IPSTAT_INC(ips_fragments);
965 	m->m_pkthdr.header = ip;
966 
967 	/* Previous ip_reass() started here. */
968 	/*
969 	 * Presence of header sizes in mbufs
970 	 * would confuse code below.
971 	 */
972 	m->m_data += hlen;
973 	m->m_len -= hlen;
974 
975 	/*
976 	 * If first fragment to arrive, create a reassembly queue.
977 	 */
978 	if (fp == NULL) {
979 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
980 		if (fp == NULL)
981 			goto dropfrag;
982 #ifdef MAC
983 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
984 			uma_zfree(V_ipq_zone, fp);
985 			fp = NULL;
986 			goto dropfrag;
987 		}
988 		mac_ipq_create(m, fp);
989 #endif
990 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
991 		V_nipq++;
992 		fp->ipq_nfrags = 1;
993 		fp->ipq_ttl = IPFRAGTTL;
994 		fp->ipq_p = ip->ip_p;
995 		fp->ipq_id = ip->ip_id;
996 		fp->ipq_src = ip->ip_src;
997 		fp->ipq_dst = ip->ip_dst;
998 		fp->ipq_frags = m;
999 		m->m_nextpkt = NULL;
1000 		goto done;
1001 	} else {
1002 		fp->ipq_nfrags++;
1003 #ifdef MAC
1004 		mac_ipq_update(m, fp);
1005 #endif
1006 	}
1007 
1008 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
1009 
1010 	/*
1011 	 * Handle ECN by comparing this segment with the first one;
1012 	 * if CE is set, do not lose CE.
1013 	 * drop if CE and not-ECT are mixed for the same packet.
1014 	 */
1015 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
1016 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
1017 	if (ecn == IPTOS_ECN_CE) {
1018 		if (ecn0 == IPTOS_ECN_NOTECT)
1019 			goto dropfrag;
1020 		if (ecn0 != IPTOS_ECN_CE)
1021 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
1022 	}
1023 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
1024 		goto dropfrag;
1025 
1026 	/*
1027 	 * Find a segment which begins after this one does.
1028 	 */
1029 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1030 		if (GETIP(q)->ip_off > ip->ip_off)
1031 			break;
1032 
1033 	/*
1034 	 * If there is a preceding segment, it may provide some of
1035 	 * our data already.  If so, drop the data from the incoming
1036 	 * segment.  If it provides all of our data, drop us, otherwise
1037 	 * stick new segment in the proper place.
1038 	 *
1039 	 * If some of the data is dropped from the the preceding
1040 	 * segment, then it's checksum is invalidated.
1041 	 */
1042 	if (p) {
1043 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1044 		if (i > 0) {
1045 			if (i >= ip->ip_len)
1046 				goto dropfrag;
1047 			m_adj(m, i);
1048 			m->m_pkthdr.csum_flags = 0;
1049 			ip->ip_off += i;
1050 			ip->ip_len -= i;
1051 		}
1052 		m->m_nextpkt = p->m_nextpkt;
1053 		p->m_nextpkt = m;
1054 	} else {
1055 		m->m_nextpkt = fp->ipq_frags;
1056 		fp->ipq_frags = m;
1057 	}
1058 
1059 	/*
1060 	 * While we overlap succeeding segments trim them or,
1061 	 * if they are completely covered, dequeue them.
1062 	 */
1063 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1064 	     q = nq) {
1065 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1066 		if (i < GETIP(q)->ip_len) {
1067 			GETIP(q)->ip_len -= i;
1068 			GETIP(q)->ip_off += i;
1069 			m_adj(q, i);
1070 			q->m_pkthdr.csum_flags = 0;
1071 			break;
1072 		}
1073 		nq = q->m_nextpkt;
1074 		m->m_nextpkt = nq;
1075 		IPSTAT_INC(ips_fragdropped);
1076 		fp->ipq_nfrags--;
1077 		m_freem(q);
1078 	}
1079 
1080 	/*
1081 	 * Check for complete reassembly and perform frag per packet
1082 	 * limiting.
1083 	 *
1084 	 * Frag limiting is performed here so that the nth frag has
1085 	 * a chance to complete the packet before we drop the packet.
1086 	 * As a result, n+1 frags are actually allowed per packet, but
1087 	 * only n will ever be stored. (n = maxfragsperpacket.)
1088 	 *
1089 	 */
1090 	next = 0;
1091 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1092 		if (GETIP(q)->ip_off != next) {
1093 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1094 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1095 				ip_freef(head, fp);
1096 			}
1097 			goto done;
1098 		}
1099 		next += GETIP(q)->ip_len;
1100 	}
1101 	/* Make sure the last packet didn't have the IP_MF flag */
1102 	if (p->m_flags & M_FRAG) {
1103 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1104 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1105 			ip_freef(head, fp);
1106 		}
1107 		goto done;
1108 	}
1109 
1110 	/*
1111 	 * Reassembly is complete.  Make sure the packet is a sane size.
1112 	 */
1113 	q = fp->ipq_frags;
1114 	ip = GETIP(q);
1115 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1116 		IPSTAT_INC(ips_toolong);
1117 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1118 		ip_freef(head, fp);
1119 		goto done;
1120 	}
1121 
1122 	/*
1123 	 * Concatenate fragments.
1124 	 */
1125 	m = q;
1126 	t = m->m_next;
1127 	m->m_next = NULL;
1128 	m_cat(m, t);
1129 	nq = q->m_nextpkt;
1130 	q->m_nextpkt = NULL;
1131 	for (q = nq; q != NULL; q = nq) {
1132 		nq = q->m_nextpkt;
1133 		q->m_nextpkt = NULL;
1134 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1135 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1136 		m_cat(m, q);
1137 	}
1138 	/*
1139 	 * In order to do checksumming faster we do 'end-around carry' here
1140 	 * (and not in for{} loop), though it implies we are not going to
1141 	 * reassemble more than 64k fragments.
1142 	 */
1143 	m->m_pkthdr.csum_data =
1144 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1145 #ifdef MAC
1146 	mac_ipq_reassemble(fp, m);
1147 	mac_ipq_destroy(fp);
1148 #endif
1149 
1150 	/*
1151 	 * Create header for new ip packet by modifying header of first
1152 	 * packet;  dequeue and discard fragment reassembly header.
1153 	 * Make header visible.
1154 	 */
1155 	ip->ip_len = (ip->ip_hl << 2) + next;
1156 	ip->ip_src = fp->ipq_src;
1157 	ip->ip_dst = fp->ipq_dst;
1158 	TAILQ_REMOVE(head, fp, ipq_list);
1159 	V_nipq--;
1160 	uma_zfree(V_ipq_zone, fp);
1161 	m->m_len += (ip->ip_hl << 2);
1162 	m->m_data -= (ip->ip_hl << 2);
1163 	/* some debugging cruft by sklower, below, will go away soon */
1164 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1165 		m_fixhdr(m);
1166 	IPSTAT_INC(ips_reassembled);
1167 	IPQ_UNLOCK();
1168 	return (m);
1169 
1170 dropfrag:
1171 	IPSTAT_INC(ips_fragdropped);
1172 	if (fp != NULL)
1173 		fp->ipq_nfrags--;
1174 	m_freem(m);
1175 done:
1176 	IPQ_UNLOCK();
1177 	return (NULL);
1178 
1179 #undef GETIP
1180 }
1181 
1182 /*
1183  * Free a fragment reassembly header and all
1184  * associated datagrams.
1185  */
1186 static void
1187 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1188 {
1189 	struct mbuf *q;
1190 
1191 	IPQ_LOCK_ASSERT();
1192 
1193 	while (fp->ipq_frags) {
1194 		q = fp->ipq_frags;
1195 		fp->ipq_frags = q->m_nextpkt;
1196 		m_freem(q);
1197 	}
1198 	TAILQ_REMOVE(fhp, fp, ipq_list);
1199 	uma_zfree(V_ipq_zone, fp);
1200 	V_nipq--;
1201 }
1202 
1203 /*
1204  * IP timer processing;
1205  * if a timer expires on a reassembly
1206  * queue, discard it.
1207  */
1208 void
1209 ip_slowtimo(void)
1210 {
1211 	VNET_ITERATOR_DECL(vnet_iter);
1212 	struct ipq *fp;
1213 	int i;
1214 
1215 	VNET_LIST_RLOCK_NOSLEEP();
1216 	IPQ_LOCK();
1217 	VNET_FOREACH(vnet_iter) {
1218 		CURVNET_SET(vnet_iter);
1219 		for (i = 0; i < IPREASS_NHASH; i++) {
1220 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1221 				struct ipq *fpp;
1222 
1223 				fpp = fp;
1224 				fp = TAILQ_NEXT(fp, ipq_list);
1225 				if(--fpp->ipq_ttl == 0) {
1226 					IPSTAT_ADD(ips_fragtimeout,
1227 					    fpp->ipq_nfrags);
1228 					ip_freef(&V_ipq[i], fpp);
1229 				}
1230 			}
1231 		}
1232 		/*
1233 		 * If we are over the maximum number of fragments
1234 		 * (due to the limit being lowered), drain off
1235 		 * enough to get down to the new limit.
1236 		 */
1237 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1238 			for (i = 0; i < IPREASS_NHASH; i++) {
1239 				while (V_nipq > V_maxnipq &&
1240 				    !TAILQ_EMPTY(&V_ipq[i])) {
1241 					IPSTAT_ADD(ips_fragdropped,
1242 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1243 					ip_freef(&V_ipq[i],
1244 					    TAILQ_FIRST(&V_ipq[i]));
1245 				}
1246 			}
1247 		}
1248 		CURVNET_RESTORE();
1249 	}
1250 	IPQ_UNLOCK();
1251 	VNET_LIST_RUNLOCK_NOSLEEP();
1252 }
1253 
1254 /*
1255  * Drain off all datagram fragments.
1256  */
1257 static void
1258 ip_drain_locked(void)
1259 {
1260 	int     i;
1261 
1262 	IPQ_LOCK_ASSERT();
1263 
1264 	for (i = 0; i < IPREASS_NHASH; i++) {
1265 		while(!TAILQ_EMPTY(&V_ipq[i])) {
1266 			IPSTAT_ADD(ips_fragdropped,
1267 			    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1268 			ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1269 		}
1270 	}
1271 }
1272 
1273 void
1274 ip_drain(void)
1275 {
1276 	VNET_ITERATOR_DECL(vnet_iter);
1277 
1278 	VNET_LIST_RLOCK_NOSLEEP();
1279 	IPQ_LOCK();
1280 	VNET_FOREACH(vnet_iter) {
1281 		CURVNET_SET(vnet_iter);
1282 		ip_drain_locked();
1283 		CURVNET_RESTORE();
1284 	}
1285 	IPQ_UNLOCK();
1286 	VNET_LIST_RUNLOCK_NOSLEEP();
1287 	in_rtqdrain();
1288 }
1289 
1290 /*
1291  * The protocol to be inserted into ip_protox[] must be already registered
1292  * in inetsw[], either statically or through pf_proto_register().
1293  */
1294 int
1295 ipproto_register(u_char ipproto)
1296 {
1297 	struct protosw *pr;
1298 
1299 	/* Sanity checks. */
1300 	if (ipproto == 0)
1301 		return (EPROTONOSUPPORT);
1302 
1303 	/*
1304 	 * The protocol slot must not be occupied by another protocol
1305 	 * already.  An index pointing to IPPROTO_RAW is unused.
1306 	 */
1307 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1308 	if (pr == NULL)
1309 		return (EPFNOSUPPORT);
1310 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1311 		return (EEXIST);
1312 
1313 	/* Find the protocol position in inetsw[] and set the index. */
1314 	for (pr = inetdomain.dom_protosw;
1315 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1316 		if (pr->pr_domain->dom_family == PF_INET &&
1317 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1318 			/* Be careful to only index valid IP protocols. */
1319 			if (pr->pr_protocol < IPPROTO_MAX) {
1320 				ip_protox[pr->pr_protocol] = pr - inetsw;
1321 				return (0);
1322 			} else
1323 				return (EINVAL);
1324 		}
1325 	}
1326 	return (EPROTONOSUPPORT);
1327 }
1328 
1329 int
1330 ipproto_unregister(u_char ipproto)
1331 {
1332 	struct protosw *pr;
1333 
1334 	/* Sanity checks. */
1335 	if (ipproto == 0)
1336 		return (EPROTONOSUPPORT);
1337 
1338 	/* Check if the protocol was indeed registered. */
1339 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1340 	if (pr == NULL)
1341 		return (EPFNOSUPPORT);
1342 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1343 		return (ENOENT);
1344 
1345 	/* Reset the protocol slot to IPPROTO_RAW. */
1346 	ip_protox[ipproto] = pr - inetsw;
1347 	return (0);
1348 }
1349 
1350 /*
1351  * Given address of next destination (final or next hop), return (referenced)
1352  * internet address info of interface to be used to get there.
1353  */
1354 struct in_ifaddr *
1355 ip_rtaddr(struct in_addr dst, u_int fibnum)
1356 {
1357 	struct route sro;
1358 	struct sockaddr_in *sin;
1359 	struct in_ifaddr *ia;
1360 
1361 	bzero(&sro, sizeof(sro));
1362 	sin = (struct sockaddr_in *)&sro.ro_dst;
1363 	sin->sin_family = AF_INET;
1364 	sin->sin_len = sizeof(*sin);
1365 	sin->sin_addr = dst;
1366 	in_rtalloc_ign(&sro, 0, fibnum);
1367 
1368 	if (sro.ro_rt == NULL)
1369 		return (NULL);
1370 
1371 	ia = ifatoia(sro.ro_rt->rt_ifa);
1372 	ifa_ref(&ia->ia_ifa);
1373 	RTFREE(sro.ro_rt);
1374 	return (ia);
1375 }
1376 
1377 u_char inetctlerrmap[PRC_NCMDS] = {
1378 	0,		0,		0,		0,
1379 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1380 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1381 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1382 	0,		0,		EHOSTUNREACH,	0,
1383 	ENOPROTOOPT,	ECONNREFUSED
1384 };
1385 
1386 /*
1387  * Forward a packet.  If some error occurs return the sender
1388  * an icmp packet.  Note we can't always generate a meaningful
1389  * icmp message because icmp doesn't have a large enough repertoire
1390  * of codes and types.
1391  *
1392  * If not forwarding, just drop the packet.  This could be confusing
1393  * if ipforwarding was zero but some routing protocol was advancing
1394  * us as a gateway to somewhere.  However, we must let the routing
1395  * protocol deal with that.
1396  *
1397  * The srcrt parameter indicates whether the packet is being forwarded
1398  * via a source route.
1399  */
1400 void
1401 ip_forward(struct mbuf *m, int srcrt)
1402 {
1403 	struct ip *ip = mtod(m, struct ip *);
1404 	struct in_ifaddr *ia;
1405 	struct mbuf *mcopy;
1406 	struct in_addr dest;
1407 	struct route ro;
1408 	int error, type = 0, code = 0, mtu = 0;
1409 
1410 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1411 		IPSTAT_INC(ips_cantforward);
1412 		m_freem(m);
1413 		return;
1414 	}
1415 #ifdef IPSTEALTH
1416 	if (!V_ipstealth) {
1417 #endif
1418 		if (ip->ip_ttl <= IPTTLDEC) {
1419 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1420 			    0, 0);
1421 			return;
1422 		}
1423 #ifdef IPSTEALTH
1424 	}
1425 #endif
1426 
1427 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1428 #ifndef IPSEC
1429 	/*
1430 	 * 'ia' may be NULL if there is no route for this destination.
1431 	 * In case of IPsec, Don't discard it just yet, but pass it to
1432 	 * ip_output in case of outgoing IPsec policy.
1433 	 */
1434 	if (!srcrt && ia == NULL) {
1435 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1436 		return;
1437 	}
1438 #endif
1439 
1440 	/*
1441 	 * Save the IP header and at most 8 bytes of the payload,
1442 	 * in case we need to generate an ICMP message to the src.
1443 	 *
1444 	 * XXX this can be optimized a lot by saving the data in a local
1445 	 * buffer on the stack (72 bytes at most), and only allocating the
1446 	 * mbuf if really necessary. The vast majority of the packets
1447 	 * are forwarded without having to send an ICMP back (either
1448 	 * because unnecessary, or because rate limited), so we are
1449 	 * really we are wasting a lot of work here.
1450 	 *
1451 	 * We don't use m_copy() because it might return a reference
1452 	 * to a shared cluster. Both this function and ip_output()
1453 	 * assume exclusive access to the IP header in `m', so any
1454 	 * data in a cluster may change before we reach icmp_error().
1455 	 */
1456 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1457 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1458 		/*
1459 		 * It's probably ok if the pkthdr dup fails (because
1460 		 * the deep copy of the tag chain failed), but for now
1461 		 * be conservative and just discard the copy since
1462 		 * code below may some day want the tags.
1463 		 */
1464 		m_free(mcopy);
1465 		mcopy = NULL;
1466 	}
1467 	if (mcopy != NULL) {
1468 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1469 		mcopy->m_pkthdr.len = mcopy->m_len;
1470 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1471 	}
1472 
1473 #ifdef IPSTEALTH
1474 	if (!V_ipstealth) {
1475 #endif
1476 		ip->ip_ttl -= IPTTLDEC;
1477 #ifdef IPSTEALTH
1478 	}
1479 #endif
1480 
1481 	/*
1482 	 * If forwarding packet using same interface that it came in on,
1483 	 * perhaps should send a redirect to sender to shortcut a hop.
1484 	 * Only send redirect if source is sending directly to us,
1485 	 * and if packet was not source routed (or has any options).
1486 	 * Also, don't send redirect if forwarding using a default route
1487 	 * or a route modified by a redirect.
1488 	 */
1489 	dest.s_addr = 0;
1490 	if (!srcrt && V_ipsendredirects &&
1491 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1492 		struct sockaddr_in *sin;
1493 		struct rtentry *rt;
1494 
1495 		bzero(&ro, sizeof(ro));
1496 		sin = (struct sockaddr_in *)&ro.ro_dst;
1497 		sin->sin_family = AF_INET;
1498 		sin->sin_len = sizeof(*sin);
1499 		sin->sin_addr = ip->ip_dst;
1500 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1501 
1502 		rt = ro.ro_rt;
1503 
1504 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1505 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1506 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1507 			u_long src = ntohl(ip->ip_src.s_addr);
1508 
1509 			if (RTA(rt) &&
1510 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1511 				if (rt->rt_flags & RTF_GATEWAY)
1512 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1513 				else
1514 					dest.s_addr = ip->ip_dst.s_addr;
1515 				/* Router requirements says to only send host redirects */
1516 				type = ICMP_REDIRECT;
1517 				code = ICMP_REDIRECT_HOST;
1518 			}
1519 		}
1520 		if (rt)
1521 			RTFREE(rt);
1522 	}
1523 
1524 	/*
1525 	 * Try to cache the route MTU from ip_output so we can consider it for
1526 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1527 	 */
1528 	bzero(&ro, sizeof(ro));
1529 
1530 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1531 
1532 	if (error == EMSGSIZE && ro.ro_rt)
1533 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1534 	if (ro.ro_rt)
1535 		RTFREE(ro.ro_rt);
1536 
1537 	if (error)
1538 		IPSTAT_INC(ips_cantforward);
1539 	else {
1540 		IPSTAT_INC(ips_forward);
1541 		if (type)
1542 			IPSTAT_INC(ips_redirectsent);
1543 		else {
1544 			if (mcopy)
1545 				m_freem(mcopy);
1546 			if (ia != NULL)
1547 				ifa_free(&ia->ia_ifa);
1548 			return;
1549 		}
1550 	}
1551 	if (mcopy == NULL) {
1552 		if (ia != NULL)
1553 			ifa_free(&ia->ia_ifa);
1554 		return;
1555 	}
1556 
1557 	switch (error) {
1558 
1559 	case 0:				/* forwarded, but need redirect */
1560 		/* type, code set above */
1561 		break;
1562 
1563 	case ENETUNREACH:
1564 	case EHOSTUNREACH:
1565 	case ENETDOWN:
1566 	case EHOSTDOWN:
1567 	default:
1568 		type = ICMP_UNREACH;
1569 		code = ICMP_UNREACH_HOST;
1570 		break;
1571 
1572 	case EMSGSIZE:
1573 		type = ICMP_UNREACH;
1574 		code = ICMP_UNREACH_NEEDFRAG;
1575 
1576 #ifdef IPSEC
1577 		/*
1578 		 * If IPsec is configured for this path,
1579 		 * override any possibly mtu value set by ip_output.
1580 		 */
1581 		mtu = ip_ipsec_mtu(m, mtu);
1582 #endif /* IPSEC */
1583 		/*
1584 		 * If the MTU was set before make sure we are below the
1585 		 * interface MTU.
1586 		 * If the MTU wasn't set before use the interface mtu or
1587 		 * fall back to the next smaller mtu step compared to the
1588 		 * current packet size.
1589 		 */
1590 		if (mtu != 0) {
1591 			if (ia != NULL)
1592 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1593 		} else {
1594 			if (ia != NULL)
1595 				mtu = ia->ia_ifp->if_mtu;
1596 			else
1597 				mtu = ip_next_mtu(ip->ip_len, 0);
1598 		}
1599 		IPSTAT_INC(ips_cantfrag);
1600 		break;
1601 
1602 	case ENOBUFS:
1603 		/*
1604 		 * A router should not generate ICMP_SOURCEQUENCH as
1605 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1606 		 * Source quench could be a big problem under DoS attacks,
1607 		 * or if the underlying interface is rate-limited.
1608 		 * Those who need source quench packets may re-enable them
1609 		 * via the net.inet.ip.sendsourcequench sysctl.
1610 		 */
1611 		if (V_ip_sendsourcequench == 0) {
1612 			m_freem(mcopy);
1613 			if (ia != NULL)
1614 				ifa_free(&ia->ia_ifa);
1615 			return;
1616 		} else {
1617 			type = ICMP_SOURCEQUENCH;
1618 			code = 0;
1619 		}
1620 		break;
1621 
1622 	case EACCES:			/* ipfw denied packet */
1623 		m_freem(mcopy);
1624 		if (ia != NULL)
1625 			ifa_free(&ia->ia_ifa);
1626 		return;
1627 	}
1628 	if (ia != NULL)
1629 		ifa_free(&ia->ia_ifa);
1630 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1631 }
1632 
1633 void
1634 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1635     struct mbuf *m)
1636 {
1637 
1638 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1639 		struct bintime bt;
1640 
1641 		bintime(&bt);
1642 		if (inp->inp_socket->so_options & SO_BINTIME) {
1643 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1644 			SCM_BINTIME, SOL_SOCKET);
1645 			if (*mp)
1646 				mp = &(*mp)->m_next;
1647 		}
1648 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1649 			struct timeval tv;
1650 
1651 			bintime2timeval(&bt, &tv);
1652 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1653 				SCM_TIMESTAMP, SOL_SOCKET);
1654 			if (*mp)
1655 				mp = &(*mp)->m_next;
1656 		}
1657 	}
1658 	if (inp->inp_flags & INP_RECVDSTADDR) {
1659 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1660 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1661 		if (*mp)
1662 			mp = &(*mp)->m_next;
1663 	}
1664 	if (inp->inp_flags & INP_RECVTTL) {
1665 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1666 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1667 		if (*mp)
1668 			mp = &(*mp)->m_next;
1669 	}
1670 #ifdef notyet
1671 	/* XXX
1672 	 * Moving these out of udp_input() made them even more broken
1673 	 * than they already were.
1674 	 */
1675 	/* options were tossed already */
1676 	if (inp->inp_flags & INP_RECVOPTS) {
1677 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1678 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1679 		if (*mp)
1680 			mp = &(*mp)->m_next;
1681 	}
1682 	/* ip_srcroute doesn't do what we want here, need to fix */
1683 	if (inp->inp_flags & INP_RECVRETOPTS) {
1684 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1685 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1686 		if (*mp)
1687 			mp = &(*mp)->m_next;
1688 	}
1689 #endif
1690 	if (inp->inp_flags & INP_RECVIF) {
1691 		struct ifnet *ifp;
1692 		struct sdlbuf {
1693 			struct sockaddr_dl sdl;
1694 			u_char	pad[32];
1695 		} sdlbuf;
1696 		struct sockaddr_dl *sdp;
1697 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1698 
1699 		if (((ifp = m->m_pkthdr.rcvif))
1700 		&& ( ifp->if_index && (ifp->if_index <= V_if_index))) {
1701 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1702 			/*
1703 			 * Change our mind and don't try copy.
1704 			 */
1705 			if ((sdp->sdl_family != AF_LINK)
1706 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1707 				goto makedummy;
1708 			}
1709 			bcopy(sdp, sdl2, sdp->sdl_len);
1710 		} else {
1711 makedummy:
1712 			sdl2->sdl_len
1713 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1714 			sdl2->sdl_family = AF_LINK;
1715 			sdl2->sdl_index = 0;
1716 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1717 		}
1718 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1719 			IP_RECVIF, IPPROTO_IP);
1720 		if (*mp)
1721 			mp = &(*mp)->m_next;
1722 	}
1723 }
1724 
1725 /*
1726  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1727  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1728  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1729  * compiled.
1730  */
1731 int
1732 ip_rsvp_init(struct socket *so)
1733 {
1734 
1735 	if (so->so_type != SOCK_RAW ||
1736 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1737 		return EOPNOTSUPP;
1738 
1739 	if (V_ip_rsvpd != NULL)
1740 		return EADDRINUSE;
1741 
1742 	V_ip_rsvpd = so;
1743 	/*
1744 	 * This may seem silly, but we need to be sure we don't over-increment
1745 	 * the RSVP counter, in case something slips up.
1746 	 */
1747 	if (!V_ip_rsvp_on) {
1748 		V_ip_rsvp_on = 1;
1749 		V_rsvp_on++;
1750 	}
1751 
1752 	return 0;
1753 }
1754 
1755 int
1756 ip_rsvp_done(void)
1757 {
1758 
1759 	V_ip_rsvpd = NULL;
1760 	/*
1761 	 * This may seem silly, but we need to be sure we don't over-decrement
1762 	 * the RSVP counter, in case something slips up.
1763 	 */
1764 	if (V_ip_rsvp_on) {
1765 		V_ip_rsvp_on = 0;
1766 		V_rsvp_on--;
1767 	}
1768 	return 0;
1769 }
1770 
1771 void
1772 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1773 {
1774 
1775 	if (rsvp_input_p) { /* call the real one if loaded */
1776 		rsvp_input_p(m, off);
1777 		return;
1778 	}
1779 
1780 	/* Can still get packets with rsvp_on = 0 if there is a local member
1781 	 * of the group to which the RSVP packet is addressed.  But in this
1782 	 * case we want to throw the packet away.
1783 	 */
1784 
1785 	if (!V_rsvp_on) {
1786 		m_freem(m);
1787 		return;
1788 	}
1789 
1790 	if (V_ip_rsvpd != NULL) {
1791 		rip_input(m, off);
1792 		return;
1793 	}
1794 	/* Drop the packet */
1795 	m_freem(m);
1796 }
1797