xref: /freebsd/sys/netinet/ip_input.c (revision 164e1f20ecf89ccacec38e82360e96dd2e2060b9)
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 #include <sys/vimage.h>
57 
58 #include <net/pfil.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_var.h>
62 #include <net/if_dl.h>
63 #include <net/route.h>
64 #include <net/netisr.h>
65 #include <net/vnet.h>
66 #include <net/flowtable.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_icmp.h>
75 #include <netinet/ip_options.h>
76 #include <machine/in_cksum.h>
77 #include <netinet/vinet.h>
78 #ifdef DEV_CARP
79 #include <netinet/ip_carp.h>
80 #endif
81 #ifdef IPSEC
82 #include <netinet/ip_ipsec.h>
83 #endif /* IPSEC */
84 
85 #include <sys/socketvar.h>
86 
87 #include <security/mac/mac_framework.h>
88 
89 #ifdef CTASSERT
90 CTASSERT(sizeof(struct ip) == 20);
91 #endif
92 
93 #ifndef VIMAGE
94 #ifndef VIMAGE_GLOBALS
95 struct vnet_inet vnet_inet_0;
96 #endif
97 #endif
98 
99 #ifdef VIMAGE_GLOBALS
100 static int	ipsendredirects;
101 static int	ip_checkinterface;
102 static int	ip_keepfaith;
103 static int	ip_sendsourcequench;
104 int	ip_defttl;
105 int	ip_do_randomid;
106 int	ipforwarding;
107 struct	in_ifaddrhead in_ifaddrhead; 		/* first inet address */
108 struct	in_ifaddrhashhead *in_ifaddrhashtbl;	/* inet addr hash table  */
109 u_long 	in_ifaddrhmask;				/* mask for hash table */
110 struct ipstat ipstat;
111 static int ip_rsvp_on;
112 struct socket *ip_rsvpd;
113 int	rsvp_on;
114 static struct ipqhead ipq[IPREASS_NHASH];
115 static int	maxnipq;	/* Administrative limit on # reass queues. */
116 static int	maxfragsperpacket;
117 int	ipstealth;
118 static int	nipq;	/* Total # of reass queues */
119 #endif
120 
121 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_FORWARDING,
122     forwarding, CTLFLAG_RW, ipforwarding, 0,
123     "Enable IP forwarding between interfaces");
124 
125 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_SENDREDIRECTS,
126     redirect, CTLFLAG_RW, ipsendredirects, 0,
127     "Enable sending IP redirects");
128 
129 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_DEFTTL,
130     ttl, CTLFLAG_RW, ip_defttl, 0, "Maximum TTL on IP packets");
131 
132 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_KEEPFAITH,
133     keepfaith, CTLFLAG_RW, ip_keepfaith,	0,
134     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
135 
136 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO,
137     sendsourcequench, CTLFLAG_RW, ip_sendsourcequench, 0,
138     "Enable the transmission of source quench packets");
139 
140 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, random_id,
141     CTLFLAG_RW, ip_do_randomid, 0, "Assign random ip_id values");
142 
143 /*
144  * XXX - Setting ip_checkinterface mostly implements the receive side of
145  * the Strong ES model described in RFC 1122, but since the routing table
146  * and transmit implementation do not implement the Strong ES model,
147  * setting this to 1 results in an odd hybrid.
148  *
149  * XXX - ip_checkinterface currently must be disabled if you use ipnat
150  * to translate the destination address to another local interface.
151  *
152  * XXX - ip_checkinterface must be disabled if you add IP aliases
153  * to the loopback interface instead of the interface where the
154  * packets for those addresses are received.
155  */
156 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO,
157     check_interface, CTLFLAG_RW, ip_checkinterface, 0,
158     "Verify packet arrives on correct interface");
159 
160 struct pfil_head inet_pfil_hook;	/* Packet filter hooks */
161 
162 static struct netisr_handler ip_nh = {
163 	.nh_name = "ip",
164 	.nh_handler = ip_input,
165 	.nh_proto = NETISR_IP,
166 	.nh_policy = NETISR_POLICY_FLOW,
167 };
168 
169 extern	struct domain inetdomain;
170 extern	struct protosw inetsw[];
171 u_char	ip_protox[IPPROTO_MAX];
172 
173 
174 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
175     ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
176 
177 #ifdef VIMAGE_GLOBALS
178 static uma_zone_t ipq_zone;
179 #endif
180 static struct mtx ipqlock;
181 
182 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
183 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
184 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
185 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
186 
187 static void	maxnipq_update(void);
188 static void	ipq_zone_change(void *);
189 
190 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, fragpackets,
191     CTLFLAG_RD, nipq, 0,
192     "Current number of IPv4 fragment reassembly queue entries");
193 
194 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, maxfragsperpacket,
195     CTLFLAG_RW, maxfragsperpacket, 0,
196     "Maximum number of IPv4 fragments allowed per packet");
197 
198 struct callout	ipport_tick_callout;
199 
200 #ifdef IPCTL_DEFMTU
201 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
202     &ip_mtu, 0, "Default MTU");
203 #endif
204 
205 #ifdef IPSTEALTH
206 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
207     ipstealth, 0, "IP stealth mode, no TTL decrementation on forwarding");
208 #endif
209 #ifdef FLOWTABLE
210 static int ip_output_flowtable_size = 2048;
211 TUNABLE_INT("net.inet.ip.output_flowtable_size", &ip_output_flowtable_size);
212 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, output_flowtable_size,
213     CTLFLAG_RDTUN, ip_output_flowtable_size, 2048,
214     "number of entries in the per-cpu output flow caches");
215 
216 struct flowtable *ip_ft;
217 #endif
218 
219 #ifdef VIMAGE_GLOBALS
220 int fw_one_pass;
221 #endif
222 
223 static void	ip_freef(struct ipqhead *, struct ipq *);
224 
225 #ifndef VIMAGE_GLOBALS
226 static void vnet_inet_register(void);
227 
228 static const vnet_modinfo_t vnet_inet_modinfo = {
229 	.vmi_id		= VNET_MOD_INET,
230 	.vmi_name	= "inet",
231 	.vmi_size	= sizeof(struct vnet_inet)
232 };
233 
234 static void vnet_inet_register()
235 {
236 
237 	vnet_mod_register(&vnet_inet_modinfo);
238 }
239 
240 SYSINIT(inet, SI_SUB_PROTO_BEGIN, SI_ORDER_FIRST, vnet_inet_register, 0);
241 #endif
242 
243 static int
244 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
245 {
246 	int error, qlimit;
247 
248 	netisr_getqlimit(&ip_nh, &qlimit);
249 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
250 	if (error || !req->newptr)
251 		return (error);
252 	if (qlimit < 1)
253 		return (EINVAL);
254 	return (netisr_setqlimit(&ip_nh, qlimit));
255 }
256 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
257     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
258     "Maximum size of the IP input queue");
259 
260 static int
261 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
262 {
263 	u_int64_t qdrops_long;
264 	int error, qdrops;
265 
266 	netisr_getqdrops(&ip_nh, &qdrops_long);
267 	qdrops = qdrops_long;
268 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
269 	if (error || !req->newptr)
270 		return (error);
271 	if (qdrops != 0)
272 		return (EINVAL);
273 	netisr_clearqdrops(&ip_nh);
274 	return (0);
275 }
276 
277 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
278     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
279     "Number of packets dropped from the IP input queue");
280 
281 /*
282  * IP initialization: fill in IP protocol switch table.
283  * All protocols not implemented in kernel go to raw IP protocol handler.
284  */
285 void
286 ip_init(void)
287 {
288 	INIT_VNET_INET(curvnet);
289 	struct protosw *pr;
290 	int i;
291 
292 	V_ipsendredirects = 1; /* XXX */
293 	V_ip_checkinterface = 0;
294 	V_ip_keepfaith = 0;
295 	V_ip_sendsourcequench = 0;
296 	V_rsvp_on = 0;
297 	V_ip_defttl = IPDEFTTL;
298 	V_ip_do_randomid = 0;
299 	V_ip_id = time_second & 0xffff;
300 	V_ipforwarding = 0;
301 	V_ipstealth = 0;
302 	V_nipq = 0;	/* Total # of reass queues */
303 
304 	V_ipport_lowfirstauto = IPPORT_RESERVED - 1;	/* 1023 */
305 	V_ipport_lowlastauto = IPPORT_RESERVEDSTART;	/* 600 */
306 	V_ipport_firstauto = IPPORT_EPHEMERALFIRST;	/* 10000 */
307 	V_ipport_lastauto = IPPORT_EPHEMERALLAST;	/* 65535 */
308 	V_ipport_hifirstauto = IPPORT_HIFIRSTAUTO;	/* 49152 */
309 	V_ipport_hilastauto = IPPORT_HILASTAUTO;	/* 65535 */
310 	V_ipport_reservedhigh = IPPORT_RESERVED - 1;	/* 1023 */
311 	V_ipport_reservedlow = 0;
312 	V_ipport_randomized = 1;	/* user controlled via sysctl */
313 	V_ipport_randomcps = 10;	/* user controlled via sysctl */
314 	V_ipport_randomtime = 45;	/* user controlled via sysctl */
315 	V_ipport_stoprandom = 0;	/* toggled by ipport_tick */
316 
317 	V_fw_one_pass = 1;
318 
319 #ifdef NOTYET
320 	/* XXX global static but not instantiated in this file */
321 	V_ipfastforward_active = 0;
322 	V_subnetsarelocal = 0;
323 	V_sameprefixcarponly = 0;
324 #endif
325 
326 	TAILQ_INIT(&V_in_ifaddrhead);
327 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
328 
329 	/* Initialize IP reassembly queue. */
330 	for (i = 0; i < IPREASS_NHASH; i++)
331 		TAILQ_INIT(&V_ipq[i]);
332 	V_maxnipq = nmbclusters / 32;
333 	V_maxfragsperpacket = 16;
334 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
335 	    NULL, UMA_ALIGN_PTR, 0);
336 	maxnipq_update();
337 
338 	/* Skip initialization of globals for non-default instances. */
339 	if (!IS_DEFAULT_VNET(curvnet))
340 		return;
341 
342 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
343 	if (pr == NULL)
344 		panic("ip_init: PF_INET not found");
345 
346 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
347 	for (i = 0; i < IPPROTO_MAX; i++)
348 		ip_protox[i] = pr - inetsw;
349 	/*
350 	 * Cycle through IP protocols and put them into the appropriate place
351 	 * in ip_protox[].
352 	 */
353 	for (pr = inetdomain.dom_protosw;
354 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
355 		if (pr->pr_domain->dom_family == PF_INET &&
356 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
357 			/* Be careful to only index valid IP protocols. */
358 			if (pr->pr_protocol < IPPROTO_MAX)
359 				ip_protox[pr->pr_protocol] = pr - inetsw;
360 		}
361 
362 	/* Initialize packet filter hooks. */
363 	inet_pfil_hook.ph_type = PFIL_TYPE_AF;
364 	inet_pfil_hook.ph_af = AF_INET;
365 	if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
366 		printf("%s: WARNING: unable to register pfil hook, "
367 			"error %d\n", __func__, i);
368 
369 	/* Start ipport_tick. */
370 	callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
371 	callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
372 	EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
373 		SHUTDOWN_PRI_DEFAULT);
374 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
375 		NULL, EVENTHANDLER_PRI_ANY);
376 
377 	/* Initialize various other remaining things. */
378 	IPQ_LOCK_INIT();
379 	netisr_register(&ip_nh);
380 #ifdef FLOWTABLE
381 	ip_ft = flowtable_alloc(ip_output_flowtable_size, FL_PCPU);
382 #endif
383 }
384 
385 void
386 ip_fini(void *xtp)
387 {
388 
389 	callout_stop(&ipport_tick_callout);
390 }
391 
392 /*
393  * Ip input routine.  Checksum and byte swap header.  If fragmented
394  * try to reassemble.  Process options.  Pass to next level.
395  */
396 void
397 ip_input(struct mbuf *m)
398 {
399 	INIT_VNET_INET(curvnet);
400 	struct ip *ip = NULL;
401 	struct in_ifaddr *ia = NULL;
402 	struct ifaddr *ifa;
403 	struct ifnet *ifp;
404 	int    checkif, hlen = 0;
405 	u_short sum;
406 	int dchg = 0;				/* dest changed after fw */
407 	struct in_addr odst;			/* original dst address */
408 
409 	M_ASSERTPKTHDR(m);
410 
411 	if (m->m_flags & M_FASTFWD_OURS) {
412 		/*
413 		 * Firewall or NAT changed destination to local.
414 		 * We expect ip_len and ip_off to be in host byte order.
415 		 */
416 		m->m_flags &= ~M_FASTFWD_OURS;
417 		/* Set up some basics that will be used later. */
418 		ip = mtod(m, struct ip *);
419 		hlen = ip->ip_hl << 2;
420 		goto ours;
421 	}
422 
423 	IPSTAT_INC(ips_total);
424 
425 	if (m->m_pkthdr.len < sizeof(struct ip))
426 		goto tooshort;
427 
428 	if (m->m_len < sizeof (struct ip) &&
429 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
430 		IPSTAT_INC(ips_toosmall);
431 		return;
432 	}
433 	ip = mtod(m, struct ip *);
434 
435 	if (ip->ip_v != IPVERSION) {
436 		IPSTAT_INC(ips_badvers);
437 		goto bad;
438 	}
439 
440 	hlen = ip->ip_hl << 2;
441 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
442 		IPSTAT_INC(ips_badhlen);
443 		goto bad;
444 	}
445 	if (hlen > m->m_len) {
446 		if ((m = m_pullup(m, hlen)) == NULL) {
447 			IPSTAT_INC(ips_badhlen);
448 			return;
449 		}
450 		ip = mtod(m, struct ip *);
451 	}
452 
453 	/* 127/8 must not appear on wire - RFC1122 */
454 	ifp = m->m_pkthdr.rcvif;
455 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
456 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
457 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
458 			IPSTAT_INC(ips_badaddr);
459 			goto bad;
460 		}
461 	}
462 
463 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
464 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
465 	} else {
466 		if (hlen == sizeof(struct ip)) {
467 			sum = in_cksum_hdr(ip);
468 		} else {
469 			sum = in_cksum(m, hlen);
470 		}
471 	}
472 	if (sum) {
473 		IPSTAT_INC(ips_badsum);
474 		goto bad;
475 	}
476 
477 #ifdef ALTQ
478 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
479 		/* packet is dropped by traffic conditioner */
480 		return;
481 #endif
482 
483 	/*
484 	 * Convert fields to host representation.
485 	 */
486 	ip->ip_len = ntohs(ip->ip_len);
487 	if (ip->ip_len < hlen) {
488 		IPSTAT_INC(ips_badlen);
489 		goto bad;
490 	}
491 	ip->ip_off = ntohs(ip->ip_off);
492 
493 	/*
494 	 * Check that the amount of data in the buffers
495 	 * is as at least much as the IP header would have us expect.
496 	 * Trim mbufs if longer than we expect.
497 	 * Drop packet if shorter than we expect.
498 	 */
499 	if (m->m_pkthdr.len < ip->ip_len) {
500 tooshort:
501 		IPSTAT_INC(ips_tooshort);
502 		goto bad;
503 	}
504 	if (m->m_pkthdr.len > ip->ip_len) {
505 		if (m->m_len == m->m_pkthdr.len) {
506 			m->m_len = ip->ip_len;
507 			m->m_pkthdr.len = ip->ip_len;
508 		} else
509 			m_adj(m, ip->ip_len - m->m_pkthdr.len);
510 	}
511 #ifdef IPSEC
512 	/*
513 	 * Bypass packet filtering for packets from a tunnel (gif).
514 	 */
515 	if (ip_ipsec_filtertunnel(m))
516 		goto passin;
517 #endif /* IPSEC */
518 
519 	/*
520 	 * Run through list of hooks for input packets.
521 	 *
522 	 * NB: Beware of the destination address changing (e.g.
523 	 *     by NAT rewriting).  When this happens, tell
524 	 *     ip_forward to do the right thing.
525 	 */
526 
527 	/* Jump over all PFIL processing if hooks are not active. */
528 	if (!PFIL_HOOKED(&inet_pfil_hook))
529 		goto passin;
530 
531 	odst = ip->ip_dst;
532 	if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
533 		return;
534 	if (m == NULL)			/* consumed by filter */
535 		return;
536 
537 	ip = mtod(m, struct ip *);
538 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
539 	ifp = m->m_pkthdr.rcvif;
540 
541 #ifdef IPFIREWALL_FORWARD
542 	if (m->m_flags & M_FASTFWD_OURS) {
543 		m->m_flags &= ~M_FASTFWD_OURS;
544 		goto ours;
545 	}
546 	if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
547 		/*
548 		 * Directly ship on the packet.  This allows to forward packets
549 		 * that were destined for us to some other directly connected
550 		 * host.
551 		 */
552 		ip_forward(m, dchg);
553 		return;
554 	}
555 #endif /* IPFIREWALL_FORWARD */
556 
557 passin:
558 	/*
559 	 * Process options and, if not destined for us,
560 	 * ship it on.  ip_dooptions returns 1 when an
561 	 * error was detected (causing an icmp message
562 	 * to be sent and the original packet to be freed).
563 	 */
564 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
565 		return;
566 
567         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
568          * matter if it is destined to another node, or whether it is
569          * a multicast one, RSVP wants it! and prevents it from being forwarded
570          * anywhere else. Also checks if the rsvp daemon is running before
571 	 * grabbing the packet.
572          */
573 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
574 		goto ours;
575 
576 	/*
577 	 * Check our list of addresses, to see if the packet is for us.
578 	 * If we don't have any addresses, assume any unicast packet
579 	 * we receive might be for us (and let the upper layers deal
580 	 * with it).
581 	 */
582 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
583 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
584 		goto ours;
585 
586 	/*
587 	 * Enable a consistency check between the destination address
588 	 * and the arrival interface for a unicast packet (the RFC 1122
589 	 * strong ES model) if IP forwarding is disabled and the packet
590 	 * is not locally generated and the packet is not subject to
591 	 * 'ipfw fwd'.
592 	 *
593 	 * XXX - Checking also should be disabled if the destination
594 	 * address is ipnat'ed to a different interface.
595 	 *
596 	 * XXX - Checking is incompatible with IP aliases added
597 	 * to the loopback interface instead of the interface where
598 	 * the packets are received.
599 	 *
600 	 * XXX - This is the case for carp vhost IPs as well so we
601 	 * insert a workaround. If the packet got here, we already
602 	 * checked with carp_iamatch() and carp_forus().
603 	 */
604 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
605 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
606 #ifdef DEV_CARP
607 	    !ifp->if_carp &&
608 #endif
609 	    (dchg == 0);
610 
611 	/*
612 	 * Check for exact addresses in the hash bucket.
613 	 */
614 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
615 		/*
616 		 * If the address matches, verify that the packet
617 		 * arrived via the correct interface if checking is
618 		 * enabled.
619 		 */
620 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
621 		    (!checkif || ia->ia_ifp == ifp))
622 			goto ours;
623 	}
624 	/*
625 	 * Check for broadcast addresses.
626 	 *
627 	 * Only accept broadcast packets that arrive via the matching
628 	 * interface.  Reception of forwarded directed broadcasts would
629 	 * be handled via ip_forward() and ether_output() with the loopback
630 	 * into the stack for SIMPLEX interfaces handled by ether_output().
631 	 */
632 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
633 		IF_ADDR_LOCK(ifp);
634 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
635 			if (ifa->ifa_addr->sa_family != AF_INET)
636 				continue;
637 			ia = ifatoia(ifa);
638 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
639 			    ip->ip_dst.s_addr) {
640 				IF_ADDR_UNLOCK(ifp);
641 				goto ours;
642 			}
643 			if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
644 				IF_ADDR_UNLOCK(ifp);
645 				goto ours;
646 			}
647 #ifdef BOOTP_COMPAT
648 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
649 				IF_ADDR_UNLOCK(ifp);
650 				goto ours;
651 			}
652 #endif
653 		}
654 		IF_ADDR_UNLOCK(ifp);
655 	}
656 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
657 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
658 		IPSTAT_INC(ips_cantforward);
659 		m_freem(m);
660 		return;
661 	}
662 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
663 		if (V_ip_mrouter) {
664 			/*
665 			 * If we are acting as a multicast router, all
666 			 * incoming multicast packets are passed to the
667 			 * kernel-level multicast forwarding function.
668 			 * The packet is returned (relatively) intact; if
669 			 * ip_mforward() returns a non-zero value, the packet
670 			 * must be discarded, else it may be accepted below.
671 			 */
672 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
673 				IPSTAT_INC(ips_cantforward);
674 				m_freem(m);
675 				return;
676 			}
677 
678 			/*
679 			 * The process-level routing daemon needs to receive
680 			 * all multicast IGMP packets, whether or not this
681 			 * host belongs to their destination groups.
682 			 */
683 			if (ip->ip_p == IPPROTO_IGMP)
684 				goto ours;
685 			IPSTAT_INC(ips_forward);
686 		}
687 		/*
688 		 * Assume the packet is for us, to avoid prematurely taking
689 		 * a lock on the in_multi hash. Protocols must perform
690 		 * their own filtering and update statistics accordingly.
691 		 */
692 		goto ours;
693 	}
694 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
695 		goto ours;
696 	if (ip->ip_dst.s_addr == INADDR_ANY)
697 		goto ours;
698 
699 	/*
700 	 * FAITH(Firewall Aided Internet Translator)
701 	 */
702 	if (ifp && ifp->if_type == IFT_FAITH) {
703 		if (V_ip_keepfaith) {
704 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
705 				goto ours;
706 		}
707 		m_freem(m);
708 		return;
709 	}
710 
711 	/*
712 	 * Not for us; forward if possible and desirable.
713 	 */
714 	if (V_ipforwarding == 0) {
715 		IPSTAT_INC(ips_cantforward);
716 		m_freem(m);
717 	} else {
718 #ifdef IPSEC
719 		if (ip_ipsec_fwd(m))
720 			goto bad;
721 #endif /* IPSEC */
722 		ip_forward(m, dchg);
723 	}
724 	return;
725 
726 ours:
727 #ifdef IPSTEALTH
728 	/*
729 	 * IPSTEALTH: Process non-routing options only
730 	 * if the packet is destined for us.
731 	 */
732 	if (V_ipstealth && hlen > sizeof (struct ip) &&
733 	    ip_dooptions(m, 1))
734 		return;
735 #endif /* IPSTEALTH */
736 
737 	/* Count the packet in the ip address stats */
738 	if (ia != NULL) {
739 		ia->ia_ifa.if_ipackets++;
740 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
741 	}
742 
743 	/*
744 	 * Attempt reassembly; if it succeeds, proceed.
745 	 * ip_reass() will return a different mbuf.
746 	 */
747 	if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
748 		m = ip_reass(m);
749 		if (m == NULL)
750 			return;
751 		ip = mtod(m, struct ip *);
752 		/* Get the header length of the reassembled packet */
753 		hlen = ip->ip_hl << 2;
754 	}
755 
756 	/*
757 	 * Further protocols expect the packet length to be w/o the
758 	 * IP header.
759 	 */
760 	ip->ip_len -= hlen;
761 
762 #ifdef IPSEC
763 	/*
764 	 * enforce IPsec policy checking if we are seeing last header.
765 	 * note that we do not visit this with protocols with pcb layer
766 	 * code - like udp/tcp/raw ip.
767 	 */
768 	if (ip_ipsec_input(m))
769 		goto bad;
770 #endif /* IPSEC */
771 
772 	/*
773 	 * Switch out to protocol's input routine.
774 	 */
775 	IPSTAT_INC(ips_delivered);
776 
777 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
778 	return;
779 bad:
780 	m_freem(m);
781 }
782 
783 /*
784  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
785  * max has slightly different semantics than the sysctl, for historical
786  * reasons.
787  */
788 static void
789 maxnipq_update(void)
790 {
791 	INIT_VNET_INET(curvnet);
792 
793 	/*
794 	 * -1 for unlimited allocation.
795 	 */
796 	if (V_maxnipq < 0)
797 		uma_zone_set_max(V_ipq_zone, 0);
798 	/*
799 	 * Positive number for specific bound.
800 	 */
801 	if (V_maxnipq > 0)
802 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
803 	/*
804 	 * Zero specifies no further fragment queue allocation -- set the
805 	 * bound very low, but rely on implementation elsewhere to actually
806 	 * prevent allocation and reclaim current queues.
807 	 */
808 	if (V_maxnipq == 0)
809 		uma_zone_set_max(V_ipq_zone, 1);
810 }
811 
812 static void
813 ipq_zone_change(void *tag)
814 {
815 	INIT_VNET_INET(curvnet);
816 
817 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
818 		V_maxnipq = nmbclusters / 32;
819 		maxnipq_update();
820 	}
821 }
822 
823 static int
824 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
825 {
826 	INIT_VNET_INET(curvnet);
827 	int error, i;
828 
829 	i = V_maxnipq;
830 	error = sysctl_handle_int(oidp, &i, 0, req);
831 	if (error || !req->newptr)
832 		return (error);
833 
834 	/*
835 	 * XXXRW: Might be a good idea to sanity check the argument and place
836 	 * an extreme upper bound.
837 	 */
838 	if (i < -1)
839 		return (EINVAL);
840 	V_maxnipq = i;
841 	maxnipq_update();
842 	return (0);
843 }
844 
845 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
846     NULL, 0, sysctl_maxnipq, "I",
847     "Maximum number of IPv4 fragment reassembly queue entries");
848 
849 /*
850  * Take incoming datagram fragment and try to reassemble it into
851  * whole datagram.  If the argument is the first fragment or one
852  * in between the function will return NULL and store the mbuf
853  * in the fragment chain.  If the argument is the last fragment
854  * the packet will be reassembled and the pointer to the new
855  * mbuf returned for further processing.  Only m_tags attached
856  * to the first packet/fragment are preserved.
857  * The IP header is *NOT* adjusted out of iplen.
858  */
859 struct mbuf *
860 ip_reass(struct mbuf *m)
861 {
862 	INIT_VNET_INET(curvnet);
863 	struct ip *ip;
864 	struct mbuf *p, *q, *nq, *t;
865 	struct ipq *fp = NULL;
866 	struct ipqhead *head;
867 	int i, hlen, next;
868 	u_int8_t ecn, ecn0;
869 	u_short hash;
870 
871 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
872 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
873 		IPSTAT_INC(ips_fragments);
874 		IPSTAT_INC(ips_fragdropped);
875 		m_freem(m);
876 		return (NULL);
877 	}
878 
879 	ip = mtod(m, struct ip *);
880 	hlen = ip->ip_hl << 2;
881 
882 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
883 	head = &V_ipq[hash];
884 	IPQ_LOCK();
885 
886 	/*
887 	 * Look for queue of fragments
888 	 * of this datagram.
889 	 */
890 	TAILQ_FOREACH(fp, head, ipq_list)
891 		if (ip->ip_id == fp->ipq_id &&
892 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
893 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
894 #ifdef MAC
895 		    mac_ipq_match(m, fp) &&
896 #endif
897 		    ip->ip_p == fp->ipq_p)
898 			goto found;
899 
900 	fp = NULL;
901 
902 	/*
903 	 * Attempt to trim the number of allocated fragment queues if it
904 	 * exceeds the administrative limit.
905 	 */
906 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
907 		/*
908 		 * drop something from the tail of the current queue
909 		 * before proceeding further
910 		 */
911 		struct ipq *q = TAILQ_LAST(head, ipqhead);
912 		if (q == NULL) {   /* gak */
913 			for (i = 0; i < IPREASS_NHASH; i++) {
914 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
915 				if (r) {
916 					IPSTAT_ADD(ips_fragtimeout,
917 					    r->ipq_nfrags);
918 					ip_freef(&V_ipq[i], r);
919 					break;
920 				}
921 			}
922 		} else {
923 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
924 			ip_freef(head, q);
925 		}
926 	}
927 
928 found:
929 	/*
930 	 * Adjust ip_len to not reflect header,
931 	 * convert offset of this to bytes.
932 	 */
933 	ip->ip_len -= hlen;
934 	if (ip->ip_off & IP_MF) {
935 		/*
936 		 * Make sure that fragments have a data length
937 		 * that's a non-zero multiple of 8 bytes.
938 		 */
939 		if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
940 			IPSTAT_INC(ips_toosmall); /* XXX */
941 			goto dropfrag;
942 		}
943 		m->m_flags |= M_FRAG;
944 	} else
945 		m->m_flags &= ~M_FRAG;
946 	ip->ip_off <<= 3;
947 
948 
949 	/*
950 	 * Attempt reassembly; if it succeeds, proceed.
951 	 * ip_reass() will return a different mbuf.
952 	 */
953 	IPSTAT_INC(ips_fragments);
954 	m->m_pkthdr.header = ip;
955 
956 	/* Previous ip_reass() started here. */
957 	/*
958 	 * Presence of header sizes in mbufs
959 	 * would confuse code below.
960 	 */
961 	m->m_data += hlen;
962 	m->m_len -= hlen;
963 
964 	/*
965 	 * If first fragment to arrive, create a reassembly queue.
966 	 */
967 	if (fp == NULL) {
968 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
969 		if (fp == NULL)
970 			goto dropfrag;
971 #ifdef MAC
972 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
973 			uma_zfree(V_ipq_zone, fp);
974 			fp = NULL;
975 			goto dropfrag;
976 		}
977 		mac_ipq_create(m, fp);
978 #endif
979 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
980 		V_nipq++;
981 		fp->ipq_nfrags = 1;
982 		fp->ipq_ttl = IPFRAGTTL;
983 		fp->ipq_p = ip->ip_p;
984 		fp->ipq_id = ip->ip_id;
985 		fp->ipq_src = ip->ip_src;
986 		fp->ipq_dst = ip->ip_dst;
987 		fp->ipq_frags = m;
988 		m->m_nextpkt = NULL;
989 		goto done;
990 	} else {
991 		fp->ipq_nfrags++;
992 #ifdef MAC
993 		mac_ipq_update(m, fp);
994 #endif
995 	}
996 
997 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
998 
999 	/*
1000 	 * Handle ECN by comparing this segment with the first one;
1001 	 * if CE is set, do not lose CE.
1002 	 * drop if CE and not-ECT are mixed for the same packet.
1003 	 */
1004 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
1005 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
1006 	if (ecn == IPTOS_ECN_CE) {
1007 		if (ecn0 == IPTOS_ECN_NOTECT)
1008 			goto dropfrag;
1009 		if (ecn0 != IPTOS_ECN_CE)
1010 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
1011 	}
1012 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
1013 		goto dropfrag;
1014 
1015 	/*
1016 	 * Find a segment which begins after this one does.
1017 	 */
1018 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
1019 		if (GETIP(q)->ip_off > ip->ip_off)
1020 			break;
1021 
1022 	/*
1023 	 * If there is a preceding segment, it may provide some of
1024 	 * our data already.  If so, drop the data from the incoming
1025 	 * segment.  If it provides all of our data, drop us, otherwise
1026 	 * stick new segment in the proper place.
1027 	 *
1028 	 * If some of the data is dropped from the the preceding
1029 	 * segment, then it's checksum is invalidated.
1030 	 */
1031 	if (p) {
1032 		i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
1033 		if (i > 0) {
1034 			if (i >= ip->ip_len)
1035 				goto dropfrag;
1036 			m_adj(m, i);
1037 			m->m_pkthdr.csum_flags = 0;
1038 			ip->ip_off += i;
1039 			ip->ip_len -= i;
1040 		}
1041 		m->m_nextpkt = p->m_nextpkt;
1042 		p->m_nextpkt = m;
1043 	} else {
1044 		m->m_nextpkt = fp->ipq_frags;
1045 		fp->ipq_frags = m;
1046 	}
1047 
1048 	/*
1049 	 * While we overlap succeeding segments trim them or,
1050 	 * if they are completely covered, dequeue them.
1051 	 */
1052 	for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
1053 	     q = nq) {
1054 		i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
1055 		if (i < GETIP(q)->ip_len) {
1056 			GETIP(q)->ip_len -= i;
1057 			GETIP(q)->ip_off += i;
1058 			m_adj(q, i);
1059 			q->m_pkthdr.csum_flags = 0;
1060 			break;
1061 		}
1062 		nq = q->m_nextpkt;
1063 		m->m_nextpkt = nq;
1064 		IPSTAT_INC(ips_fragdropped);
1065 		fp->ipq_nfrags--;
1066 		m_freem(q);
1067 	}
1068 
1069 	/*
1070 	 * Check for complete reassembly and perform frag per packet
1071 	 * limiting.
1072 	 *
1073 	 * Frag limiting is performed here so that the nth frag has
1074 	 * a chance to complete the packet before we drop the packet.
1075 	 * As a result, n+1 frags are actually allowed per packet, but
1076 	 * only n will ever be stored. (n = maxfragsperpacket.)
1077 	 *
1078 	 */
1079 	next = 0;
1080 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1081 		if (GETIP(q)->ip_off != next) {
1082 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1083 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1084 				ip_freef(head, fp);
1085 			}
1086 			goto done;
1087 		}
1088 		next += GETIP(q)->ip_len;
1089 	}
1090 	/* Make sure the last packet didn't have the IP_MF flag */
1091 	if (p->m_flags & M_FRAG) {
1092 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1093 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1094 			ip_freef(head, fp);
1095 		}
1096 		goto done;
1097 	}
1098 
1099 	/*
1100 	 * Reassembly is complete.  Make sure the packet is a sane size.
1101 	 */
1102 	q = fp->ipq_frags;
1103 	ip = GETIP(q);
1104 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1105 		IPSTAT_INC(ips_toolong);
1106 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1107 		ip_freef(head, fp);
1108 		goto done;
1109 	}
1110 
1111 	/*
1112 	 * Concatenate fragments.
1113 	 */
1114 	m = q;
1115 	t = m->m_next;
1116 	m->m_next = NULL;
1117 	m_cat(m, t);
1118 	nq = q->m_nextpkt;
1119 	q->m_nextpkt = NULL;
1120 	for (q = nq; q != NULL; q = nq) {
1121 		nq = q->m_nextpkt;
1122 		q->m_nextpkt = NULL;
1123 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1124 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1125 		m_cat(m, q);
1126 	}
1127 	/*
1128 	 * In order to do checksumming faster we do 'end-around carry' here
1129 	 * (and not in for{} loop), though it implies we are not going to
1130 	 * reassemble more than 64k fragments.
1131 	 */
1132 	m->m_pkthdr.csum_data =
1133 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1134 #ifdef MAC
1135 	mac_ipq_reassemble(fp, m);
1136 	mac_ipq_destroy(fp);
1137 #endif
1138 
1139 	/*
1140 	 * Create header for new ip packet by modifying header of first
1141 	 * packet;  dequeue and discard fragment reassembly header.
1142 	 * Make header visible.
1143 	 */
1144 	ip->ip_len = (ip->ip_hl << 2) + next;
1145 	ip->ip_src = fp->ipq_src;
1146 	ip->ip_dst = fp->ipq_dst;
1147 	TAILQ_REMOVE(head, fp, ipq_list);
1148 	V_nipq--;
1149 	uma_zfree(V_ipq_zone, fp);
1150 	m->m_len += (ip->ip_hl << 2);
1151 	m->m_data -= (ip->ip_hl << 2);
1152 	/* some debugging cruft by sklower, below, will go away soon */
1153 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1154 		m_fixhdr(m);
1155 	IPSTAT_INC(ips_reassembled);
1156 	IPQ_UNLOCK();
1157 	return (m);
1158 
1159 dropfrag:
1160 	IPSTAT_INC(ips_fragdropped);
1161 	if (fp != NULL)
1162 		fp->ipq_nfrags--;
1163 	m_freem(m);
1164 done:
1165 	IPQ_UNLOCK();
1166 	return (NULL);
1167 
1168 #undef GETIP
1169 }
1170 
1171 /*
1172  * Free a fragment reassembly header and all
1173  * associated datagrams.
1174  */
1175 static void
1176 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1177 {
1178 	INIT_VNET_INET(curvnet);
1179 	struct mbuf *q;
1180 
1181 	IPQ_LOCK_ASSERT();
1182 
1183 	while (fp->ipq_frags) {
1184 		q = fp->ipq_frags;
1185 		fp->ipq_frags = q->m_nextpkt;
1186 		m_freem(q);
1187 	}
1188 	TAILQ_REMOVE(fhp, fp, ipq_list);
1189 	uma_zfree(V_ipq_zone, fp);
1190 	V_nipq--;
1191 }
1192 
1193 /*
1194  * IP timer processing;
1195  * if a timer expires on a reassembly
1196  * queue, discard it.
1197  */
1198 void
1199 ip_slowtimo(void)
1200 {
1201 	VNET_ITERATOR_DECL(vnet_iter);
1202 	struct ipq *fp;
1203 	int i;
1204 
1205 	IPQ_LOCK();
1206 	VNET_LIST_RLOCK();
1207 	VNET_FOREACH(vnet_iter) {
1208 		CURVNET_SET(vnet_iter);
1209 		INIT_VNET_INET(vnet_iter);
1210 		for (i = 0; i < IPREASS_NHASH; i++) {
1211 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1212 				struct ipq *fpp;
1213 
1214 				fpp = fp;
1215 				fp = TAILQ_NEXT(fp, ipq_list);
1216 				if(--fpp->ipq_ttl == 0) {
1217 					IPSTAT_ADD(ips_fragtimeout,
1218 					    fpp->ipq_nfrags);
1219 					ip_freef(&V_ipq[i], fpp);
1220 				}
1221 			}
1222 		}
1223 		/*
1224 		 * If we are over the maximum number of fragments
1225 		 * (due to the limit being lowered), drain off
1226 		 * enough to get down to the new limit.
1227 		 */
1228 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1229 			for (i = 0; i < IPREASS_NHASH; i++) {
1230 				while (V_nipq > V_maxnipq &&
1231 				    !TAILQ_EMPTY(&V_ipq[i])) {
1232 					IPSTAT_ADD(ips_fragdropped,
1233 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1234 					ip_freef(&V_ipq[i],
1235 					    TAILQ_FIRST(&V_ipq[i]));
1236 				}
1237 			}
1238 		}
1239 		CURVNET_RESTORE();
1240 	}
1241 	VNET_LIST_RUNLOCK();
1242 	IPQ_UNLOCK();
1243 }
1244 
1245 /*
1246  * Drain off all datagram fragments.
1247  */
1248 void
1249 ip_drain(void)
1250 {
1251 	VNET_ITERATOR_DECL(vnet_iter);
1252 	int     i;
1253 
1254 	IPQ_LOCK();
1255 	VNET_LIST_RLOCK();
1256 	VNET_FOREACH(vnet_iter) {
1257 		CURVNET_SET(vnet_iter);
1258 		INIT_VNET_INET(vnet_iter);
1259 		for (i = 0; i < IPREASS_NHASH; i++) {
1260 			while(!TAILQ_EMPTY(&V_ipq[i])) {
1261 				IPSTAT_ADD(ips_fragdropped,
1262 				    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1263 				ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1264 			}
1265 		}
1266 		CURVNET_RESTORE();
1267 	}
1268 	VNET_LIST_RUNLOCK();
1269 	IPQ_UNLOCK();
1270 	in_rtqdrain();
1271 }
1272 
1273 /*
1274  * The protocol to be inserted into ip_protox[] must be already registered
1275  * in inetsw[], either statically or through pf_proto_register().
1276  */
1277 int
1278 ipproto_register(u_char ipproto)
1279 {
1280 	struct protosw *pr;
1281 
1282 	/* Sanity checks. */
1283 	if (ipproto == 0)
1284 		return (EPROTONOSUPPORT);
1285 
1286 	/*
1287 	 * The protocol slot must not be occupied by another protocol
1288 	 * already.  An index pointing to IPPROTO_RAW is unused.
1289 	 */
1290 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1291 	if (pr == NULL)
1292 		return (EPFNOSUPPORT);
1293 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1294 		return (EEXIST);
1295 
1296 	/* Find the protocol position in inetsw[] and set the index. */
1297 	for (pr = inetdomain.dom_protosw;
1298 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1299 		if (pr->pr_domain->dom_family == PF_INET &&
1300 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1301 			/* Be careful to only index valid IP protocols. */
1302 			if (pr->pr_protocol < IPPROTO_MAX) {
1303 				ip_protox[pr->pr_protocol] = pr - inetsw;
1304 				return (0);
1305 			} else
1306 				return (EINVAL);
1307 		}
1308 	}
1309 	return (EPROTONOSUPPORT);
1310 }
1311 
1312 int
1313 ipproto_unregister(u_char ipproto)
1314 {
1315 	struct protosw *pr;
1316 
1317 	/* Sanity checks. */
1318 	if (ipproto == 0)
1319 		return (EPROTONOSUPPORT);
1320 
1321 	/* Check if the protocol was indeed registered. */
1322 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1323 	if (pr == NULL)
1324 		return (EPFNOSUPPORT);
1325 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1326 		return (ENOENT);
1327 
1328 	/* Reset the protocol slot to IPPROTO_RAW. */
1329 	ip_protox[ipproto] = pr - inetsw;
1330 	return (0);
1331 }
1332 
1333 /*
1334  * Given address of next destination (final or next hop),
1335  * return internet address info of interface to be used to get there.
1336  */
1337 struct in_ifaddr *
1338 ip_rtaddr(struct in_addr dst, u_int fibnum)
1339 {
1340 	struct route sro;
1341 	struct sockaddr_in *sin;
1342 	struct in_ifaddr *ifa;
1343 
1344 	bzero(&sro, sizeof(sro));
1345 	sin = (struct sockaddr_in *)&sro.ro_dst;
1346 	sin->sin_family = AF_INET;
1347 	sin->sin_len = sizeof(*sin);
1348 	sin->sin_addr = dst;
1349 	in_rtalloc_ign(&sro, 0, fibnum);
1350 
1351 	if (sro.ro_rt == NULL)
1352 		return (NULL);
1353 
1354 	ifa = ifatoia(sro.ro_rt->rt_ifa);
1355 	RTFREE(sro.ro_rt);
1356 	return (ifa);
1357 }
1358 
1359 u_char inetctlerrmap[PRC_NCMDS] = {
1360 	0,		0,		0,		0,
1361 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1362 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1363 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1364 	0,		0,		EHOSTUNREACH,	0,
1365 	ENOPROTOOPT,	ECONNREFUSED
1366 };
1367 
1368 /*
1369  * Forward a packet.  If some error occurs return the sender
1370  * an icmp packet.  Note we can't always generate a meaningful
1371  * icmp message because icmp doesn't have a large enough repertoire
1372  * of codes and types.
1373  *
1374  * If not forwarding, just drop the packet.  This could be confusing
1375  * if ipforwarding was zero but some routing protocol was advancing
1376  * us as a gateway to somewhere.  However, we must let the routing
1377  * protocol deal with that.
1378  *
1379  * The srcrt parameter indicates whether the packet is being forwarded
1380  * via a source route.
1381  */
1382 void
1383 ip_forward(struct mbuf *m, int srcrt)
1384 {
1385 	INIT_VNET_INET(curvnet);
1386 	struct ip *ip = mtod(m, struct ip *);
1387 	struct in_ifaddr *ia;
1388 	struct mbuf *mcopy;
1389 	struct in_addr dest;
1390 	struct route ro;
1391 	int error, type = 0, code = 0, mtu = 0;
1392 
1393 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1394 		IPSTAT_INC(ips_cantforward);
1395 		m_freem(m);
1396 		return;
1397 	}
1398 #ifdef IPSTEALTH
1399 	if (!V_ipstealth) {
1400 #endif
1401 		if (ip->ip_ttl <= IPTTLDEC) {
1402 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1403 			    0, 0);
1404 			return;
1405 		}
1406 #ifdef IPSTEALTH
1407 	}
1408 #endif
1409 
1410 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1411 #ifndef IPSEC
1412 	/*
1413 	 * 'ia' may be NULL if there is no route for this destination.
1414 	 * In case of IPsec, Don't discard it just yet, but pass it to
1415 	 * ip_output in case of outgoing IPsec policy.
1416 	 */
1417 	if (!srcrt && ia == NULL) {
1418 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1419 		return;
1420 	}
1421 #endif
1422 
1423 	/*
1424 	 * Save the IP header and at most 8 bytes of the payload,
1425 	 * in case we need to generate an ICMP message to the src.
1426 	 *
1427 	 * XXX this can be optimized a lot by saving the data in a local
1428 	 * buffer on the stack (72 bytes at most), and only allocating the
1429 	 * mbuf if really necessary. The vast majority of the packets
1430 	 * are forwarded without having to send an ICMP back (either
1431 	 * because unnecessary, or because rate limited), so we are
1432 	 * really we are wasting a lot of work here.
1433 	 *
1434 	 * We don't use m_copy() because it might return a reference
1435 	 * to a shared cluster. Both this function and ip_output()
1436 	 * assume exclusive access to the IP header in `m', so any
1437 	 * data in a cluster may change before we reach icmp_error().
1438 	 */
1439 	MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1440 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1441 		/*
1442 		 * It's probably ok if the pkthdr dup fails (because
1443 		 * the deep copy of the tag chain failed), but for now
1444 		 * be conservative and just discard the copy since
1445 		 * code below may some day want the tags.
1446 		 */
1447 		m_free(mcopy);
1448 		mcopy = NULL;
1449 	}
1450 	if (mcopy != NULL) {
1451 		mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1452 		mcopy->m_pkthdr.len = mcopy->m_len;
1453 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1454 	}
1455 
1456 #ifdef IPSTEALTH
1457 	if (!V_ipstealth) {
1458 #endif
1459 		ip->ip_ttl -= IPTTLDEC;
1460 #ifdef IPSTEALTH
1461 	}
1462 #endif
1463 
1464 	/*
1465 	 * If forwarding packet using same interface that it came in on,
1466 	 * perhaps should send a redirect to sender to shortcut a hop.
1467 	 * Only send redirect if source is sending directly to us,
1468 	 * and if packet was not source routed (or has any options).
1469 	 * Also, don't send redirect if forwarding using a default route
1470 	 * or a route modified by a redirect.
1471 	 */
1472 	dest.s_addr = 0;
1473 	if (!srcrt && V_ipsendredirects &&
1474 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1475 		struct sockaddr_in *sin;
1476 		struct rtentry *rt;
1477 
1478 		bzero(&ro, sizeof(ro));
1479 		sin = (struct sockaddr_in *)&ro.ro_dst;
1480 		sin->sin_family = AF_INET;
1481 		sin->sin_len = sizeof(*sin);
1482 		sin->sin_addr = ip->ip_dst;
1483 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1484 
1485 		rt = ro.ro_rt;
1486 
1487 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1488 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1489 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1490 			u_long src = ntohl(ip->ip_src.s_addr);
1491 
1492 			if (RTA(rt) &&
1493 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1494 				if (rt->rt_flags & RTF_GATEWAY)
1495 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1496 				else
1497 					dest.s_addr = ip->ip_dst.s_addr;
1498 				/* Router requirements says to only send host redirects */
1499 				type = ICMP_REDIRECT;
1500 				code = ICMP_REDIRECT_HOST;
1501 			}
1502 		}
1503 		if (rt)
1504 			RTFREE(rt);
1505 	}
1506 
1507 	/*
1508 	 * Try to cache the route MTU from ip_output so we can consider it for
1509 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1510 	 */
1511 	bzero(&ro, sizeof(ro));
1512 
1513 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1514 
1515 	if (error == EMSGSIZE && ro.ro_rt)
1516 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1517 	if (ro.ro_rt)
1518 		RTFREE(ro.ro_rt);
1519 
1520 	if (error)
1521 		IPSTAT_INC(ips_cantforward);
1522 	else {
1523 		IPSTAT_INC(ips_forward);
1524 		if (type)
1525 			IPSTAT_INC(ips_redirectsent);
1526 		else {
1527 			if (mcopy)
1528 				m_freem(mcopy);
1529 			return;
1530 		}
1531 	}
1532 	if (mcopy == NULL)
1533 		return;
1534 
1535 	switch (error) {
1536 
1537 	case 0:				/* forwarded, but need redirect */
1538 		/* type, code set above */
1539 		break;
1540 
1541 	case ENETUNREACH:
1542 	case EHOSTUNREACH:
1543 	case ENETDOWN:
1544 	case EHOSTDOWN:
1545 	default:
1546 		type = ICMP_UNREACH;
1547 		code = ICMP_UNREACH_HOST;
1548 		break;
1549 
1550 	case EMSGSIZE:
1551 		type = ICMP_UNREACH;
1552 		code = ICMP_UNREACH_NEEDFRAG;
1553 
1554 #ifdef IPSEC
1555 		/*
1556 		 * If IPsec is configured for this path,
1557 		 * override any possibly mtu value set by ip_output.
1558 		 */
1559 		mtu = ip_ipsec_mtu(m, mtu);
1560 #endif /* IPSEC */
1561 		/*
1562 		 * If the MTU was set before make sure we are below the
1563 		 * interface MTU.
1564 		 * If the MTU wasn't set before use the interface mtu or
1565 		 * fall back to the next smaller mtu step compared to the
1566 		 * current packet size.
1567 		 */
1568 		if (mtu != 0) {
1569 			if (ia != NULL)
1570 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1571 		} else {
1572 			if (ia != NULL)
1573 				mtu = ia->ia_ifp->if_mtu;
1574 			else
1575 				mtu = ip_next_mtu(ip->ip_len, 0);
1576 		}
1577 		IPSTAT_INC(ips_cantfrag);
1578 		break;
1579 
1580 	case ENOBUFS:
1581 		/*
1582 		 * A router should not generate ICMP_SOURCEQUENCH as
1583 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1584 		 * Source quench could be a big problem under DoS attacks,
1585 		 * or if the underlying interface is rate-limited.
1586 		 * Those who need source quench packets may re-enable them
1587 		 * via the net.inet.ip.sendsourcequench sysctl.
1588 		 */
1589 		if (V_ip_sendsourcequench == 0) {
1590 			m_freem(mcopy);
1591 			return;
1592 		} else {
1593 			type = ICMP_SOURCEQUENCH;
1594 			code = 0;
1595 		}
1596 		break;
1597 
1598 	case EACCES:			/* ipfw denied packet */
1599 		m_freem(mcopy);
1600 		return;
1601 	}
1602 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1603 }
1604 
1605 void
1606 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1607     struct mbuf *m)
1608 {
1609 	INIT_VNET_NET(inp->inp_vnet);
1610 
1611 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1612 		struct bintime bt;
1613 
1614 		bintime(&bt);
1615 		if (inp->inp_socket->so_options & SO_BINTIME) {
1616 			*mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1617 			SCM_BINTIME, SOL_SOCKET);
1618 			if (*mp)
1619 				mp = &(*mp)->m_next;
1620 		}
1621 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1622 			struct timeval tv;
1623 
1624 			bintime2timeval(&bt, &tv);
1625 			*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1626 				SCM_TIMESTAMP, SOL_SOCKET);
1627 			if (*mp)
1628 				mp = &(*mp)->m_next;
1629 		}
1630 	}
1631 	if (inp->inp_flags & INP_RECVDSTADDR) {
1632 		*mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1633 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1634 		if (*mp)
1635 			mp = &(*mp)->m_next;
1636 	}
1637 	if (inp->inp_flags & INP_RECVTTL) {
1638 		*mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1639 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1640 		if (*mp)
1641 			mp = &(*mp)->m_next;
1642 	}
1643 #ifdef notyet
1644 	/* XXX
1645 	 * Moving these out of udp_input() made them even more broken
1646 	 * than they already were.
1647 	 */
1648 	/* options were tossed already */
1649 	if (inp->inp_flags & INP_RECVOPTS) {
1650 		*mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1651 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1652 		if (*mp)
1653 			mp = &(*mp)->m_next;
1654 	}
1655 	/* ip_srcroute doesn't do what we want here, need to fix */
1656 	if (inp->inp_flags & INP_RECVRETOPTS) {
1657 		*mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1658 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1659 		if (*mp)
1660 			mp = &(*mp)->m_next;
1661 	}
1662 #endif
1663 	if (inp->inp_flags & INP_RECVIF) {
1664 		struct ifnet *ifp;
1665 		struct sdlbuf {
1666 			struct sockaddr_dl sdl;
1667 			u_char	pad[32];
1668 		} sdlbuf;
1669 		struct sockaddr_dl *sdp;
1670 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1671 
1672 		if (((ifp = m->m_pkthdr.rcvif))
1673 		&& ( ifp->if_index && (ifp->if_index <= V_if_index))) {
1674 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1675 			/*
1676 			 * Change our mind and don't try copy.
1677 			 */
1678 			if ((sdp->sdl_family != AF_LINK)
1679 			|| (sdp->sdl_len > sizeof(sdlbuf))) {
1680 				goto makedummy;
1681 			}
1682 			bcopy(sdp, sdl2, sdp->sdl_len);
1683 		} else {
1684 makedummy:
1685 			sdl2->sdl_len
1686 				= offsetof(struct sockaddr_dl, sdl_data[0]);
1687 			sdl2->sdl_family = AF_LINK;
1688 			sdl2->sdl_index = 0;
1689 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1690 		}
1691 		*mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1692 			IP_RECVIF, IPPROTO_IP);
1693 		if (*mp)
1694 			mp = &(*mp)->m_next;
1695 	}
1696 }
1697 
1698 /*
1699  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1700  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1701  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1702  * compiled.
1703  */
1704 int
1705 ip_rsvp_init(struct socket *so)
1706 {
1707 	INIT_VNET_INET(so->so_vnet);
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 	INIT_VNET_INET(curvnet);
1733 
1734 	V_ip_rsvpd = NULL;
1735 	/*
1736 	 * This may seem silly, but we need to be sure we don't over-decrement
1737 	 * the RSVP counter, in case something slips up.
1738 	 */
1739 	if (V_ip_rsvp_on) {
1740 		V_ip_rsvp_on = 0;
1741 		V_rsvp_on--;
1742 	}
1743 	return 0;
1744 }
1745 
1746 void
1747 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1748 {
1749 	INIT_VNET_INET(curvnet);
1750 
1751 	if (rsvp_input_p) { /* call the real one if loaded */
1752 		rsvp_input_p(m, off);
1753 		return;
1754 	}
1755 
1756 	/* Can still get packets with rsvp_on = 0 if there is a local member
1757 	 * of the group to which the RSVP packet is addressed.  But in this
1758 	 * case we want to throw the packet away.
1759 	 */
1760 
1761 	if (!V_rsvp_on) {
1762 		m_freem(m);
1763 		return;
1764 	}
1765 
1766 	if (V_ip_rsvpd != NULL) {
1767 		rip_input(m, off);
1768 		return;
1769 	}
1770 	/* Drop the packet */
1771 	m_freem(m);
1772 }
1773