xref: /freebsd/sys/netinet/ip_input.c (revision a98ff317388a00b992f1bf8404dee596f9383f5e)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_bootp.h"
36 #include "opt_ipfw.h"
37 #include "opt_ipstealth.h"
38 #include "opt_ipsec.h"
39 #include "opt_route.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/mbuf.h>
44 #include <sys/malloc.h>
45 #include <sys/domain.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 
55 #include <net/pfil.h>
56 #include <net/if.h>
57 #include <net/if_types.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/netisr.h>
62 #include <net/vnet.h>
63 #include <net/flowtable.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_fw.h>
72 #include <netinet/ip_icmp.h>
73 #include <netinet/ip_options.h>
74 #include <machine/in_cksum.h>
75 #include <netinet/ip_carp.h>
76 #ifdef IPSEC
77 #include <netinet/ip_ipsec.h>
78 #endif /* IPSEC */
79 
80 #include <sys/socketvar.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #ifdef CTASSERT
85 CTASSERT(sizeof(struct ip) == 20);
86 #endif
87 
88 struct	rwlock in_ifaddr_lock;
89 RW_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock");
90 
91 VNET_DEFINE(int, rsvp_on);
92 
93 VNET_DEFINE(int, ipforwarding);
94 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
95     &VNET_NAME(ipforwarding), 0,
96     "Enable IP forwarding between interfaces");
97 
98 static VNET_DEFINE(int, ipsendredirects) = 1;	/* XXX */
99 #define	V_ipsendredirects	VNET(ipsendredirects)
100 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
101     &VNET_NAME(ipsendredirects), 0,
102     "Enable sending IP redirects");
103 
104 static VNET_DEFINE(int, ip_keepfaith);
105 #define	V_ip_keepfaith		VNET(ip_keepfaith)
106 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
107     &VNET_NAME(ip_keepfaith), 0,
108     "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
109 
110 static VNET_DEFINE(int, ip_sendsourcequench);
111 #define	V_ip_sendsourcequench	VNET(ip_sendsourcequench)
112 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
113     &VNET_NAME(ip_sendsourcequench), 0,
114     "Enable the transmission of source quench packets");
115 
116 VNET_DEFINE(int, ip_do_randomid);
117 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
118     &VNET_NAME(ip_do_randomid), 0,
119     "Assign random ip_id values");
120 
121 /*
122  * XXX - Setting ip_checkinterface mostly implements the receive side of
123  * the Strong ES model described in RFC 1122, but since the routing table
124  * and transmit implementation do not implement the Strong ES model,
125  * setting this to 1 results in an odd hybrid.
126  *
127  * XXX - ip_checkinterface currently must be disabled if you use ipnat
128  * to translate the destination address to another local interface.
129  *
130  * XXX - ip_checkinterface must be disabled if you add IP aliases
131  * to the loopback interface instead of the interface where the
132  * packets for those addresses are received.
133  */
134 static VNET_DEFINE(int, ip_checkinterface);
135 #define	V_ip_checkinterface	VNET(ip_checkinterface)
136 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
137     &VNET_NAME(ip_checkinterface), 0,
138     "Verify packet arrives on correct interface");
139 
140 VNET_DEFINE(struct pfil_head, inet_pfil_hook);	/* Packet filter hooks */
141 
142 static struct netisr_handler ip_nh = {
143 	.nh_name = "ip",
144 	.nh_handler = ip_input,
145 	.nh_proto = NETISR_IP,
146 	.nh_policy = NETISR_POLICY_FLOW,
147 };
148 
149 extern	struct domain inetdomain;
150 extern	struct protosw inetsw[];
151 u_char	ip_protox[IPPROTO_MAX];
152 VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead);  /* first inet address */
153 VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table  */
154 VNET_DEFINE(u_long, in_ifaddrhmask);		/* mask for hash table */
155 
156 static VNET_DEFINE(uma_zone_t, ipq_zone);
157 static VNET_DEFINE(TAILQ_HEAD(ipqhead, ipq), ipq[IPREASS_NHASH]);
158 static struct mtx ipqlock;
159 
160 #define	V_ipq_zone		VNET(ipq_zone)
161 #define	V_ipq			VNET(ipq)
162 
163 #define	IPQ_LOCK()	mtx_lock(&ipqlock)
164 #define	IPQ_UNLOCK()	mtx_unlock(&ipqlock)
165 #define	IPQ_LOCK_INIT()	mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
166 #define	IPQ_LOCK_ASSERT()	mtx_assert(&ipqlock, MA_OWNED)
167 
168 static void	maxnipq_update(void);
169 static void	ipq_zone_change(void *);
170 static void	ip_drain_locked(void);
171 
172 static VNET_DEFINE(int, maxnipq);  /* Administrative limit on # reass queues. */
173 static VNET_DEFINE(int, nipq);			/* Total # of reass queues */
174 #define	V_maxnipq		VNET(maxnipq)
175 #define	V_nipq			VNET(nipq)
176 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
177     &VNET_NAME(nipq), 0,
178     "Current number of IPv4 fragment reassembly queue entries");
179 
180 static VNET_DEFINE(int, maxfragsperpacket);
181 #define	V_maxfragsperpacket	VNET(maxfragsperpacket)
182 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
183     &VNET_NAME(maxfragsperpacket), 0,
184     "Maximum number of IPv4 fragments allowed per packet");
185 
186 #ifdef IPCTL_DEFMTU
187 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
188     &ip_mtu, 0, "Default MTU");
189 #endif
190 
191 #ifdef IPSTEALTH
192 VNET_DEFINE(int, ipstealth);
193 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
194     &VNET_NAME(ipstealth), 0,
195     "IP stealth mode, no TTL decrementation on forwarding");
196 #endif
197 
198 #ifdef FLOWTABLE
199 static VNET_DEFINE(int, ip_output_flowtable_size) = 2048;
200 VNET_DEFINE(struct flowtable *, ip_ft);
201 #define	V_ip_output_flowtable_size	VNET(ip_output_flowtable_size)
202 
203 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
204     &VNET_NAME(ip_output_flowtable_size), 2048,
205     "number of entries in the per-cpu output flow caches");
206 #endif
207 
208 static void	ip_freef(struct ipqhead *, struct ipq *);
209 
210 /*
211  * IP statistics are stored in the "array" of counter(9)s.
212  */
213 VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat);
214 VNET_PCPUSTAT_SYSINIT(ipstat);
215 SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat,
216     "IP statistics (struct ipstat, netinet/ip_var.h)");
217 
218 #ifdef VIMAGE
219 VNET_PCPUSTAT_SYSUNINIT(ipstat);
220 #endif /* VIMAGE */
221 
222 /*
223  * Kernel module interface for updating ipstat.  The argument is an index
224  * into ipstat treated as an array.
225  */
226 void
227 kmod_ipstat_inc(int statnum)
228 {
229 
230 	counter_u64_add(VNET(ipstat)[statnum], 1);
231 }
232 
233 void
234 kmod_ipstat_dec(int statnum)
235 {
236 
237 	counter_u64_add(VNET(ipstat)[statnum], -1);
238 }
239 
240 static int
241 sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS)
242 {
243 	int error, qlimit;
244 
245 	netisr_getqlimit(&ip_nh, &qlimit);
246 	error = sysctl_handle_int(oidp, &qlimit, 0, req);
247 	if (error || !req->newptr)
248 		return (error);
249 	if (qlimit < 1)
250 		return (EINVAL);
251 	return (netisr_setqlimit(&ip_nh, qlimit));
252 }
253 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen,
254     CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I",
255     "Maximum size of the IP input queue");
256 
257 static int
258 sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS)
259 {
260 	u_int64_t qdrops_long;
261 	int error, qdrops;
262 
263 	netisr_getqdrops(&ip_nh, &qdrops_long);
264 	qdrops = qdrops_long;
265 	error = sysctl_handle_int(oidp, &qdrops, 0, req);
266 	if (error || !req->newptr)
267 		return (error);
268 	if (qdrops != 0)
269 		return (EINVAL);
270 	netisr_clearqdrops(&ip_nh);
271 	return (0);
272 }
273 
274 SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops,
275     CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I",
276     "Number of packets dropped from the IP input queue");
277 
278 /*
279  * IP initialization: fill in IP protocol switch table.
280  * All protocols not implemented in kernel go to raw IP protocol handler.
281  */
282 void
283 ip_init(void)
284 {
285 	struct protosw *pr;
286 	int i;
287 
288 	V_ip_id = time_second & 0xffff;
289 
290 	TAILQ_INIT(&V_in_ifaddrhead);
291 	V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask);
292 
293 	/* Initialize IP reassembly queue. */
294 	for (i = 0; i < IPREASS_NHASH; i++)
295 		TAILQ_INIT(&V_ipq[i]);
296 	V_maxnipq = nmbclusters / 32;
297 	V_maxfragsperpacket = 16;
298 	V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
299 	    NULL, UMA_ALIGN_PTR, 0);
300 	maxnipq_update();
301 
302 	/* Initialize packet filter hooks. */
303 	V_inet_pfil_hook.ph_type = PFIL_TYPE_AF;
304 	V_inet_pfil_hook.ph_af = AF_INET;
305 	if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0)
306 		printf("%s: WARNING: unable to register pfil hook, "
307 			"error %d\n", __func__, i);
308 
309 #ifdef FLOWTABLE
310 	if (TUNABLE_INT_FETCH("net.inet.ip.output_flowtable_size",
311 		&V_ip_output_flowtable_size)) {
312 		if (V_ip_output_flowtable_size < 256)
313 			V_ip_output_flowtable_size = 256;
314 		if (!powerof2(V_ip_output_flowtable_size)) {
315 			printf("flowtable must be power of 2 size\n");
316 			V_ip_output_flowtable_size = 2048;
317 		}
318 	} else {
319 		/*
320 		 * round up to the next power of 2
321 		 */
322 		V_ip_output_flowtable_size = 1 << fls((1024 + maxusers * 64)-1);
323 	}
324 	V_ip_ft = flowtable_alloc("ipv4", V_ip_output_flowtable_size, FL_PCPU);
325 #endif
326 
327 	/* Skip initialization of globals for non-default instances. */
328 	if (!IS_DEFAULT_VNET(curvnet))
329 		return;
330 
331 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
332 	if (pr == NULL)
333 		panic("ip_init: PF_INET not found");
334 
335 	/* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
336 	for (i = 0; i < IPPROTO_MAX; i++)
337 		ip_protox[i] = pr - inetsw;
338 	/*
339 	 * Cycle through IP protocols and put them into the appropriate place
340 	 * in ip_protox[].
341 	 */
342 	for (pr = inetdomain.dom_protosw;
343 	    pr < inetdomain.dom_protoswNPROTOSW; pr++)
344 		if (pr->pr_domain->dom_family == PF_INET &&
345 		    pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
346 			/* Be careful to only index valid IP protocols. */
347 			if (pr->pr_protocol < IPPROTO_MAX)
348 				ip_protox[pr->pr_protocol] = pr - inetsw;
349 		}
350 
351 	EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
352 		NULL, EVENTHANDLER_PRI_ANY);
353 
354 	/* Initialize various other remaining things. */
355 	IPQ_LOCK_INIT();
356 	netisr_register(&ip_nh);
357 }
358 
359 #ifdef VIMAGE
360 void
361 ip_destroy(void)
362 {
363 
364 	/* Cleanup in_ifaddr hash table; should be empty. */
365 	hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask);
366 
367 	IPQ_LOCK();
368 	ip_drain_locked();
369 	IPQ_UNLOCK();
370 
371 	uma_zdestroy(V_ipq_zone);
372 }
373 #endif
374 
375 /*
376  * Ip input routine.  Checksum and byte swap header.  If fragmented
377  * try to reassemble.  Process options.  Pass to next level.
378  */
379 void
380 ip_input(struct mbuf *m)
381 {
382 	struct ip *ip = NULL;
383 	struct in_ifaddr *ia = NULL;
384 	struct ifaddr *ifa;
385 	struct ifnet *ifp;
386 	int    checkif, hlen = 0;
387 	uint16_t sum, ip_len;
388 	int dchg = 0;				/* dest changed after fw */
389 	struct in_addr odst;			/* original dst address */
390 
391 	M_ASSERTPKTHDR(m);
392 
393 	if (m->m_flags & M_FASTFWD_OURS) {
394 		m->m_flags &= ~M_FASTFWD_OURS;
395 		/* Set up some basics that will be used later. */
396 		ip = mtod(m, struct ip *);
397 		hlen = ip->ip_hl << 2;
398 		ip_len = ntohs(ip->ip_len);
399 		goto ours;
400 	}
401 
402 	IPSTAT_INC(ips_total);
403 
404 	if (m->m_pkthdr.len < sizeof(struct ip))
405 		goto tooshort;
406 
407 	if (m->m_len < sizeof (struct ip) &&
408 	    (m = m_pullup(m, sizeof (struct ip))) == NULL) {
409 		IPSTAT_INC(ips_toosmall);
410 		return;
411 	}
412 	ip = mtod(m, struct ip *);
413 
414 	if (ip->ip_v != IPVERSION) {
415 		IPSTAT_INC(ips_badvers);
416 		goto bad;
417 	}
418 
419 	hlen = ip->ip_hl << 2;
420 	if (hlen < sizeof(struct ip)) {	/* minimum header length */
421 		IPSTAT_INC(ips_badhlen);
422 		goto bad;
423 	}
424 	if (hlen > m->m_len) {
425 		if ((m = m_pullup(m, hlen)) == NULL) {
426 			IPSTAT_INC(ips_badhlen);
427 			return;
428 		}
429 		ip = mtod(m, struct ip *);
430 	}
431 
432 	/* 127/8 must not appear on wire - RFC1122 */
433 	ifp = m->m_pkthdr.rcvif;
434 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
435 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
436 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
437 			IPSTAT_INC(ips_badaddr);
438 			goto bad;
439 		}
440 	}
441 
442 	if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
443 		sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
444 	} else {
445 		if (hlen == sizeof(struct ip)) {
446 			sum = in_cksum_hdr(ip);
447 		} else {
448 			sum = in_cksum(m, hlen);
449 		}
450 	}
451 	if (sum) {
452 		IPSTAT_INC(ips_badsum);
453 		goto bad;
454 	}
455 
456 #ifdef ALTQ
457 	if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
458 		/* packet is dropped by traffic conditioner */
459 		return;
460 #endif
461 
462 	ip_len = ntohs(ip->ip_len);
463 	if (ip_len < hlen) {
464 		IPSTAT_INC(ips_badlen);
465 		goto bad;
466 	}
467 
468 	/*
469 	 * Check that the amount of data in the buffers
470 	 * is as at least much as the IP header would have us expect.
471 	 * Trim mbufs if longer than we expect.
472 	 * Drop packet if shorter than we expect.
473 	 */
474 	if (m->m_pkthdr.len < ip_len) {
475 tooshort:
476 		IPSTAT_INC(ips_tooshort);
477 		goto bad;
478 	}
479 	if (m->m_pkthdr.len > ip_len) {
480 		if (m->m_len == m->m_pkthdr.len) {
481 			m->m_len = ip_len;
482 			m->m_pkthdr.len = ip_len;
483 		} else
484 			m_adj(m, ip_len - m->m_pkthdr.len);
485 	}
486 #ifdef IPSEC
487 	/*
488 	 * Bypass packet filtering for packets previously handled by IPsec.
489 	 */
490 	if (ip_ipsec_filtertunnel(m))
491 		goto passin;
492 #endif /* IPSEC */
493 
494 	/*
495 	 * Run through list of hooks for input packets.
496 	 *
497 	 * NB: Beware of the destination address changing (e.g.
498 	 *     by NAT rewriting).  When this happens, tell
499 	 *     ip_forward to do the right thing.
500 	 */
501 
502 	/* Jump over all PFIL processing if hooks are not active. */
503 	if (!PFIL_HOOKED(&V_inet_pfil_hook))
504 		goto passin;
505 
506 	odst = ip->ip_dst;
507 	if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, NULL) != 0)
508 		return;
509 	if (m == NULL)			/* consumed by filter */
510 		return;
511 
512 	ip = mtod(m, struct ip *);
513 	dchg = (odst.s_addr != ip->ip_dst.s_addr);
514 	ifp = m->m_pkthdr.rcvif;
515 
516 	if (m->m_flags & M_FASTFWD_OURS) {
517 		m->m_flags &= ~M_FASTFWD_OURS;
518 		goto ours;
519 	}
520 	if (m->m_flags & M_IP_NEXTHOP) {
521 		dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL);
522 		if (dchg != 0) {
523 			/*
524 			 * Directly ship the packet on.  This allows
525 			 * forwarding packets originally destined to us
526 			 * to some other directly connected host.
527 			 */
528 			ip_forward(m, 1);
529 			return;
530 		}
531 	}
532 passin:
533 
534 	/*
535 	 * Process options and, if not destined for us,
536 	 * ship it on.  ip_dooptions returns 1 when an
537 	 * error was detected (causing an icmp message
538 	 * to be sent and the original packet to be freed).
539 	 */
540 	if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
541 		return;
542 
543         /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
544          * matter if it is destined to another node, or whether it is
545          * a multicast one, RSVP wants it! and prevents it from being forwarded
546          * anywhere else. Also checks if the rsvp daemon is running before
547 	 * grabbing the packet.
548          */
549 	if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP)
550 		goto ours;
551 
552 	/*
553 	 * Check our list of addresses, to see if the packet is for us.
554 	 * If we don't have any addresses, assume any unicast packet
555 	 * we receive might be for us (and let the upper layers deal
556 	 * with it).
557 	 */
558 	if (TAILQ_EMPTY(&V_in_ifaddrhead) &&
559 	    (m->m_flags & (M_MCAST|M_BCAST)) == 0)
560 		goto ours;
561 
562 	/*
563 	 * Enable a consistency check between the destination address
564 	 * and the arrival interface for a unicast packet (the RFC 1122
565 	 * strong ES model) if IP forwarding is disabled and the packet
566 	 * is not locally generated and the packet is not subject to
567 	 * 'ipfw fwd'.
568 	 *
569 	 * XXX - Checking also should be disabled if the destination
570 	 * address is ipnat'ed to a different interface.
571 	 *
572 	 * XXX - Checking is incompatible with IP aliases added
573 	 * to the loopback interface instead of the interface where
574 	 * the packets are received.
575 	 *
576 	 * XXX - This is the case for carp vhost IPs as well so we
577 	 * insert a workaround. If the packet got here, we already
578 	 * checked with carp_iamatch() and carp_forus().
579 	 */
580 	checkif = V_ip_checkinterface && (V_ipforwarding == 0) &&
581 	    ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) &&
582 	    ifp->if_carp == NULL && (dchg == 0);
583 
584 	/*
585 	 * Check for exact addresses in the hash bucket.
586 	 */
587 	/* IN_IFADDR_RLOCK(); */
588 	LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
589 		/*
590 		 * If the address matches, verify that the packet
591 		 * arrived via the correct interface if checking is
592 		 * enabled.
593 		 */
594 		if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
595 		    (!checkif || ia->ia_ifp == ifp)) {
596 			ifa_ref(&ia->ia_ifa);
597 			/* IN_IFADDR_RUNLOCK(); */
598 			goto ours;
599 		}
600 	}
601 	/* IN_IFADDR_RUNLOCK(); */
602 
603 	/*
604 	 * Check for broadcast addresses.
605 	 *
606 	 * Only accept broadcast packets that arrive via the matching
607 	 * interface.  Reception of forwarded directed broadcasts would
608 	 * be handled via ip_forward() and ether_output() with the loopback
609 	 * into the stack for SIMPLEX interfaces handled by ether_output().
610 	 */
611 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
612 		IF_ADDR_RLOCK(ifp);
613 	        TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
614 			if (ifa->ifa_addr->sa_family != AF_INET)
615 				continue;
616 			ia = ifatoia(ifa);
617 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
618 			    ip->ip_dst.s_addr) {
619 				ifa_ref(ifa);
620 				IF_ADDR_RUNLOCK(ifp);
621 				goto ours;
622 			}
623 #ifdef BOOTP_COMPAT
624 			if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) {
625 				ifa_ref(ifa);
626 				IF_ADDR_RUNLOCK(ifp);
627 				goto ours;
628 			}
629 #endif
630 		}
631 		IF_ADDR_RUNLOCK(ifp);
632 		ia = NULL;
633 	}
634 	/* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
635 	if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
636 		IPSTAT_INC(ips_cantforward);
637 		m_freem(m);
638 		return;
639 	}
640 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
641 		if (V_ip_mrouter) {
642 			/*
643 			 * If we are acting as a multicast router, all
644 			 * incoming multicast packets are passed to the
645 			 * kernel-level multicast forwarding function.
646 			 * The packet is returned (relatively) intact; if
647 			 * ip_mforward() returns a non-zero value, the packet
648 			 * must be discarded, else it may be accepted below.
649 			 */
650 			if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) {
651 				IPSTAT_INC(ips_cantforward);
652 				m_freem(m);
653 				return;
654 			}
655 
656 			/*
657 			 * The process-level routing daemon needs to receive
658 			 * all multicast IGMP packets, whether or not this
659 			 * host belongs to their destination groups.
660 			 */
661 			if (ip->ip_p == IPPROTO_IGMP)
662 				goto ours;
663 			IPSTAT_INC(ips_forward);
664 		}
665 		/*
666 		 * Assume the packet is for us, to avoid prematurely taking
667 		 * a lock on the in_multi hash. Protocols must perform
668 		 * their own filtering and update statistics accordingly.
669 		 */
670 		goto ours;
671 	}
672 	if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
673 		goto ours;
674 	if (ip->ip_dst.s_addr == INADDR_ANY)
675 		goto ours;
676 
677 	/*
678 	 * FAITH(Firewall Aided Internet Translator)
679 	 */
680 	if (ifp && ifp->if_type == IFT_FAITH) {
681 		if (V_ip_keepfaith) {
682 			if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
683 				goto ours;
684 		}
685 		m_freem(m);
686 		return;
687 	}
688 
689 	/*
690 	 * Not for us; forward if possible and desirable.
691 	 */
692 	if (V_ipforwarding == 0) {
693 		IPSTAT_INC(ips_cantforward);
694 		m_freem(m);
695 	} else {
696 #ifdef IPSEC
697 		if (ip_ipsec_fwd(m))
698 			goto bad;
699 #endif /* IPSEC */
700 		ip_forward(m, dchg);
701 	}
702 	return;
703 
704 ours:
705 #ifdef IPSTEALTH
706 	/*
707 	 * IPSTEALTH: Process non-routing options only
708 	 * if the packet is destined for us.
709 	 */
710 	if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) {
711 		if (ia != NULL)
712 			ifa_free(&ia->ia_ifa);
713 		return;
714 	}
715 #endif /* IPSTEALTH */
716 
717 	/* Count the packet in the ip address stats */
718 	if (ia != NULL) {
719 		ia->ia_ifa.if_ipackets++;
720 		ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
721 		ifa_free(&ia->ia_ifa);
722 	}
723 
724 	/*
725 	 * Attempt reassembly; if it succeeds, proceed.
726 	 * ip_reass() will return a different mbuf.
727 	 */
728 	if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
729 		m = ip_reass(m);
730 		if (m == NULL)
731 			return;
732 		ip = mtod(m, struct ip *);
733 		/* Get the header length of the reassembled packet */
734 		hlen = ip->ip_hl << 2;
735 	}
736 
737 #ifdef IPSEC
738 	/*
739 	 * enforce IPsec policy checking if we are seeing last header.
740 	 * note that we do not visit this with protocols with pcb layer
741 	 * code - like udp/tcp/raw ip.
742 	 */
743 	if (ip_ipsec_input(m))
744 		goto bad;
745 #endif /* IPSEC */
746 
747 	/*
748 	 * Switch out to protocol's input routine.
749 	 */
750 	IPSTAT_INC(ips_delivered);
751 
752 	(*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
753 	return;
754 bad:
755 	m_freem(m);
756 }
757 
758 /*
759  * After maxnipq has been updated, propagate the change to UMA.  The UMA zone
760  * max has slightly different semantics than the sysctl, for historical
761  * reasons.
762  */
763 static void
764 maxnipq_update(void)
765 {
766 
767 	/*
768 	 * -1 for unlimited allocation.
769 	 */
770 	if (V_maxnipq < 0)
771 		uma_zone_set_max(V_ipq_zone, 0);
772 	/*
773 	 * Positive number for specific bound.
774 	 */
775 	if (V_maxnipq > 0)
776 		uma_zone_set_max(V_ipq_zone, V_maxnipq);
777 	/*
778 	 * Zero specifies no further fragment queue allocation -- set the
779 	 * bound very low, but rely on implementation elsewhere to actually
780 	 * prevent allocation and reclaim current queues.
781 	 */
782 	if (V_maxnipq == 0)
783 		uma_zone_set_max(V_ipq_zone, 1);
784 }
785 
786 static void
787 ipq_zone_change(void *tag)
788 {
789 
790 	if (V_maxnipq > 0 && V_maxnipq < (nmbclusters / 32)) {
791 		V_maxnipq = nmbclusters / 32;
792 		maxnipq_update();
793 	}
794 }
795 
796 static int
797 sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
798 {
799 	int error, i;
800 
801 	i = V_maxnipq;
802 	error = sysctl_handle_int(oidp, &i, 0, req);
803 	if (error || !req->newptr)
804 		return (error);
805 
806 	/*
807 	 * XXXRW: Might be a good idea to sanity check the argument and place
808 	 * an extreme upper bound.
809 	 */
810 	if (i < -1)
811 		return (EINVAL);
812 	V_maxnipq = i;
813 	maxnipq_update();
814 	return (0);
815 }
816 
817 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
818     NULL, 0, sysctl_maxnipq, "I",
819     "Maximum number of IPv4 fragment reassembly queue entries");
820 
821 /*
822  * Take incoming datagram fragment and try to reassemble it into
823  * whole datagram.  If the argument is the first fragment or one
824  * in between the function will return NULL and store the mbuf
825  * in the fragment chain.  If the argument is the last fragment
826  * the packet will be reassembled and the pointer to the new
827  * mbuf returned for further processing.  Only m_tags attached
828  * to the first packet/fragment are preserved.
829  * The IP header is *NOT* adjusted out of iplen.
830  */
831 struct mbuf *
832 ip_reass(struct mbuf *m)
833 {
834 	struct ip *ip;
835 	struct mbuf *p, *q, *nq, *t;
836 	struct ipq *fp = NULL;
837 	struct ipqhead *head;
838 	int i, hlen, next;
839 	u_int8_t ecn, ecn0;
840 	u_short hash;
841 
842 	/* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
843 	if (V_maxnipq == 0 || V_maxfragsperpacket == 0) {
844 		IPSTAT_INC(ips_fragments);
845 		IPSTAT_INC(ips_fragdropped);
846 		m_freem(m);
847 		return (NULL);
848 	}
849 
850 	ip = mtod(m, struct ip *);
851 	hlen = ip->ip_hl << 2;
852 
853 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
854 	head = &V_ipq[hash];
855 	IPQ_LOCK();
856 
857 	/*
858 	 * Look for queue of fragments
859 	 * of this datagram.
860 	 */
861 	TAILQ_FOREACH(fp, head, ipq_list)
862 		if (ip->ip_id == fp->ipq_id &&
863 		    ip->ip_src.s_addr == fp->ipq_src.s_addr &&
864 		    ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
865 #ifdef MAC
866 		    mac_ipq_match(m, fp) &&
867 #endif
868 		    ip->ip_p == fp->ipq_p)
869 			goto found;
870 
871 	fp = NULL;
872 
873 	/*
874 	 * Attempt to trim the number of allocated fragment queues if it
875 	 * exceeds the administrative limit.
876 	 */
877 	if ((V_nipq > V_maxnipq) && (V_maxnipq > 0)) {
878 		/*
879 		 * drop something from the tail of the current queue
880 		 * before proceeding further
881 		 */
882 		struct ipq *q = TAILQ_LAST(head, ipqhead);
883 		if (q == NULL) {   /* gak */
884 			for (i = 0; i < IPREASS_NHASH; i++) {
885 				struct ipq *r = TAILQ_LAST(&V_ipq[i], ipqhead);
886 				if (r) {
887 					IPSTAT_ADD(ips_fragtimeout,
888 					    r->ipq_nfrags);
889 					ip_freef(&V_ipq[i], r);
890 					break;
891 				}
892 			}
893 		} else {
894 			IPSTAT_ADD(ips_fragtimeout, q->ipq_nfrags);
895 			ip_freef(head, q);
896 		}
897 	}
898 
899 found:
900 	/*
901 	 * Adjust ip_len to not reflect header,
902 	 * convert offset of this to bytes.
903 	 */
904 	ip->ip_len = htons(ntohs(ip->ip_len) - hlen);
905 	if (ip->ip_off & htons(IP_MF)) {
906 		/*
907 		 * Make sure that fragments have a data length
908 		 * that's a non-zero multiple of 8 bytes.
909 		 */
910 		if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) {
911 			IPSTAT_INC(ips_toosmall); /* XXX */
912 			goto dropfrag;
913 		}
914 		m->m_flags |= M_FRAG;
915 	} else
916 		m->m_flags &= ~M_FRAG;
917 	ip->ip_off = htons(ntohs(ip->ip_off) << 3);
918 
919 	/*
920 	 * Attempt reassembly; if it succeeds, proceed.
921 	 * ip_reass() will return a different mbuf.
922 	 */
923 	IPSTAT_INC(ips_fragments);
924 	m->m_pkthdr.header = ip;
925 
926 	/* Previous ip_reass() started here. */
927 	/*
928 	 * Presence of header sizes in mbufs
929 	 * would confuse code below.
930 	 */
931 	m->m_data += hlen;
932 	m->m_len -= hlen;
933 
934 	/*
935 	 * If first fragment to arrive, create a reassembly queue.
936 	 */
937 	if (fp == NULL) {
938 		fp = uma_zalloc(V_ipq_zone, M_NOWAIT);
939 		if (fp == NULL)
940 			goto dropfrag;
941 #ifdef MAC
942 		if (mac_ipq_init(fp, M_NOWAIT) != 0) {
943 			uma_zfree(V_ipq_zone, fp);
944 			fp = NULL;
945 			goto dropfrag;
946 		}
947 		mac_ipq_create(m, fp);
948 #endif
949 		TAILQ_INSERT_HEAD(head, fp, ipq_list);
950 		V_nipq++;
951 		fp->ipq_nfrags = 1;
952 		fp->ipq_ttl = IPFRAGTTL;
953 		fp->ipq_p = ip->ip_p;
954 		fp->ipq_id = ip->ip_id;
955 		fp->ipq_src = ip->ip_src;
956 		fp->ipq_dst = ip->ip_dst;
957 		fp->ipq_frags = m;
958 		m->m_nextpkt = NULL;
959 		goto done;
960 	} else {
961 		fp->ipq_nfrags++;
962 #ifdef MAC
963 		mac_ipq_update(m, fp);
964 #endif
965 	}
966 
967 #define GETIP(m)	((struct ip*)((m)->m_pkthdr.header))
968 
969 	/*
970 	 * Handle ECN by comparing this segment with the first one;
971 	 * if CE is set, do not lose CE.
972 	 * drop if CE and not-ECT are mixed for the same packet.
973 	 */
974 	ecn = ip->ip_tos & IPTOS_ECN_MASK;
975 	ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
976 	if (ecn == IPTOS_ECN_CE) {
977 		if (ecn0 == IPTOS_ECN_NOTECT)
978 			goto dropfrag;
979 		if (ecn0 != IPTOS_ECN_CE)
980 			GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
981 	}
982 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
983 		goto dropfrag;
984 
985 	/*
986 	 * Find a segment which begins after this one does.
987 	 */
988 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
989 		if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off))
990 			break;
991 
992 	/*
993 	 * If there is a preceding segment, it may provide some of
994 	 * our data already.  If so, drop the data from the incoming
995 	 * segment.  If it provides all of our data, drop us, otherwise
996 	 * stick new segment in the proper place.
997 	 *
998 	 * If some of the data is dropped from the preceding
999 	 * segment, then it's checksum is invalidated.
1000 	 */
1001 	if (p) {
1002 		i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) -
1003 		    ntohs(ip->ip_off);
1004 		if (i > 0) {
1005 			if (i >= ntohs(ip->ip_len))
1006 				goto dropfrag;
1007 			m_adj(m, i);
1008 			m->m_pkthdr.csum_flags = 0;
1009 			ip->ip_off = htons(ntohs(ip->ip_off) + i);
1010 			ip->ip_len = htons(ntohs(ip->ip_len) - i);
1011 		}
1012 		m->m_nextpkt = p->m_nextpkt;
1013 		p->m_nextpkt = m;
1014 	} else {
1015 		m->m_nextpkt = fp->ipq_frags;
1016 		fp->ipq_frags = m;
1017 	}
1018 
1019 	/*
1020 	 * While we overlap succeeding segments trim them or,
1021 	 * if they are completely covered, dequeue them.
1022 	 */
1023 	for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) >
1024 	    ntohs(GETIP(q)->ip_off); q = nq) {
1025 		i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) -
1026 		    ntohs(GETIP(q)->ip_off);
1027 		if (i < ntohs(GETIP(q)->ip_len)) {
1028 			GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i);
1029 			GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i);
1030 			m_adj(q, i);
1031 			q->m_pkthdr.csum_flags = 0;
1032 			break;
1033 		}
1034 		nq = q->m_nextpkt;
1035 		m->m_nextpkt = nq;
1036 		IPSTAT_INC(ips_fragdropped);
1037 		fp->ipq_nfrags--;
1038 		m_freem(q);
1039 	}
1040 
1041 	/*
1042 	 * Check for complete reassembly and perform frag per packet
1043 	 * limiting.
1044 	 *
1045 	 * Frag limiting is performed here so that the nth frag has
1046 	 * a chance to complete the packet before we drop the packet.
1047 	 * As a result, n+1 frags are actually allowed per packet, but
1048 	 * only n will ever be stored. (n = maxfragsperpacket.)
1049 	 *
1050 	 */
1051 	next = 0;
1052 	for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
1053 		if (ntohs(GETIP(q)->ip_off) != next) {
1054 			if (fp->ipq_nfrags > V_maxfragsperpacket) {
1055 				IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1056 				ip_freef(head, fp);
1057 			}
1058 			goto done;
1059 		}
1060 		next += ntohs(GETIP(q)->ip_len);
1061 	}
1062 	/* Make sure the last packet didn't have the IP_MF flag */
1063 	if (p->m_flags & M_FRAG) {
1064 		if (fp->ipq_nfrags > V_maxfragsperpacket) {
1065 			IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1066 			ip_freef(head, fp);
1067 		}
1068 		goto done;
1069 	}
1070 
1071 	/*
1072 	 * Reassembly is complete.  Make sure the packet is a sane size.
1073 	 */
1074 	q = fp->ipq_frags;
1075 	ip = GETIP(q);
1076 	if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
1077 		IPSTAT_INC(ips_toolong);
1078 		IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags);
1079 		ip_freef(head, fp);
1080 		goto done;
1081 	}
1082 
1083 	/*
1084 	 * Concatenate fragments.
1085 	 */
1086 	m = q;
1087 	t = m->m_next;
1088 	m->m_next = NULL;
1089 	m_cat(m, t);
1090 	nq = q->m_nextpkt;
1091 	q->m_nextpkt = NULL;
1092 	for (q = nq; q != NULL; q = nq) {
1093 		nq = q->m_nextpkt;
1094 		q->m_nextpkt = NULL;
1095 		m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1096 		m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1097 		m_cat(m, q);
1098 	}
1099 	/*
1100 	 * In order to do checksumming faster we do 'end-around carry' here
1101 	 * (and not in for{} loop), though it implies we are not going to
1102 	 * reassemble more than 64k fragments.
1103 	 */
1104 	m->m_pkthdr.csum_data =
1105 	    (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1106 #ifdef MAC
1107 	mac_ipq_reassemble(fp, m);
1108 	mac_ipq_destroy(fp);
1109 #endif
1110 
1111 	/*
1112 	 * Create header for new ip packet by modifying header of first
1113 	 * packet;  dequeue and discard fragment reassembly header.
1114 	 * Make header visible.
1115 	 */
1116 	ip->ip_len = htons((ip->ip_hl << 2) + next);
1117 	ip->ip_src = fp->ipq_src;
1118 	ip->ip_dst = fp->ipq_dst;
1119 	TAILQ_REMOVE(head, fp, ipq_list);
1120 	V_nipq--;
1121 	uma_zfree(V_ipq_zone, fp);
1122 	m->m_len += (ip->ip_hl << 2);
1123 	m->m_data -= (ip->ip_hl << 2);
1124 	/* some debugging cruft by sklower, below, will go away soon */
1125 	if (m->m_flags & M_PKTHDR)	/* XXX this should be done elsewhere */
1126 		m_fixhdr(m);
1127 	IPSTAT_INC(ips_reassembled);
1128 	IPQ_UNLOCK();
1129 	return (m);
1130 
1131 dropfrag:
1132 	IPSTAT_INC(ips_fragdropped);
1133 	if (fp != NULL)
1134 		fp->ipq_nfrags--;
1135 	m_freem(m);
1136 done:
1137 	IPQ_UNLOCK();
1138 	return (NULL);
1139 
1140 #undef GETIP
1141 }
1142 
1143 /*
1144  * Free a fragment reassembly header and all
1145  * associated datagrams.
1146  */
1147 static void
1148 ip_freef(struct ipqhead *fhp, struct ipq *fp)
1149 {
1150 	struct mbuf *q;
1151 
1152 	IPQ_LOCK_ASSERT();
1153 
1154 	while (fp->ipq_frags) {
1155 		q = fp->ipq_frags;
1156 		fp->ipq_frags = q->m_nextpkt;
1157 		m_freem(q);
1158 	}
1159 	TAILQ_REMOVE(fhp, fp, ipq_list);
1160 	uma_zfree(V_ipq_zone, fp);
1161 	V_nipq--;
1162 }
1163 
1164 /*
1165  * IP timer processing;
1166  * if a timer expires on a reassembly
1167  * queue, discard it.
1168  */
1169 void
1170 ip_slowtimo(void)
1171 {
1172 	VNET_ITERATOR_DECL(vnet_iter);
1173 	struct ipq *fp;
1174 	int i;
1175 
1176 	VNET_LIST_RLOCK_NOSLEEP();
1177 	IPQ_LOCK();
1178 	VNET_FOREACH(vnet_iter) {
1179 		CURVNET_SET(vnet_iter);
1180 		for (i = 0; i < IPREASS_NHASH; i++) {
1181 			for(fp = TAILQ_FIRST(&V_ipq[i]); fp;) {
1182 				struct ipq *fpp;
1183 
1184 				fpp = fp;
1185 				fp = TAILQ_NEXT(fp, ipq_list);
1186 				if(--fpp->ipq_ttl == 0) {
1187 					IPSTAT_ADD(ips_fragtimeout,
1188 					    fpp->ipq_nfrags);
1189 					ip_freef(&V_ipq[i], fpp);
1190 				}
1191 			}
1192 		}
1193 		/*
1194 		 * If we are over the maximum number of fragments
1195 		 * (due to the limit being lowered), drain off
1196 		 * enough to get down to the new limit.
1197 		 */
1198 		if (V_maxnipq >= 0 && V_nipq > V_maxnipq) {
1199 			for (i = 0; i < IPREASS_NHASH; i++) {
1200 				while (V_nipq > V_maxnipq &&
1201 				    !TAILQ_EMPTY(&V_ipq[i])) {
1202 					IPSTAT_ADD(ips_fragdropped,
1203 					    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1204 					ip_freef(&V_ipq[i],
1205 					    TAILQ_FIRST(&V_ipq[i]));
1206 				}
1207 			}
1208 		}
1209 		CURVNET_RESTORE();
1210 	}
1211 	IPQ_UNLOCK();
1212 	VNET_LIST_RUNLOCK_NOSLEEP();
1213 }
1214 
1215 /*
1216  * Drain off all datagram fragments.
1217  */
1218 static void
1219 ip_drain_locked(void)
1220 {
1221 	int     i;
1222 
1223 	IPQ_LOCK_ASSERT();
1224 
1225 	for (i = 0; i < IPREASS_NHASH; i++) {
1226 		while(!TAILQ_EMPTY(&V_ipq[i])) {
1227 			IPSTAT_ADD(ips_fragdropped,
1228 			    TAILQ_FIRST(&V_ipq[i])->ipq_nfrags);
1229 			ip_freef(&V_ipq[i], TAILQ_FIRST(&V_ipq[i]));
1230 		}
1231 	}
1232 }
1233 
1234 void
1235 ip_drain(void)
1236 {
1237 	VNET_ITERATOR_DECL(vnet_iter);
1238 
1239 	VNET_LIST_RLOCK_NOSLEEP();
1240 	IPQ_LOCK();
1241 	VNET_FOREACH(vnet_iter) {
1242 		CURVNET_SET(vnet_iter);
1243 		ip_drain_locked();
1244 		CURVNET_RESTORE();
1245 	}
1246 	IPQ_UNLOCK();
1247 	VNET_LIST_RUNLOCK_NOSLEEP();
1248 	in_rtqdrain();
1249 }
1250 
1251 /*
1252  * The protocol to be inserted into ip_protox[] must be already registered
1253  * in inetsw[], either statically or through pf_proto_register().
1254  */
1255 int
1256 ipproto_register(short ipproto)
1257 {
1258 	struct protosw *pr;
1259 
1260 	/* Sanity checks. */
1261 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1262 		return (EPROTONOSUPPORT);
1263 
1264 	/*
1265 	 * The protocol slot must not be occupied by another protocol
1266 	 * already.  An index pointing to IPPROTO_RAW is unused.
1267 	 */
1268 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1269 	if (pr == NULL)
1270 		return (EPFNOSUPPORT);
1271 	if (ip_protox[ipproto] != pr - inetsw)	/* IPPROTO_RAW */
1272 		return (EEXIST);
1273 
1274 	/* Find the protocol position in inetsw[] and set the index. */
1275 	for (pr = inetdomain.dom_protosw;
1276 	     pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1277 		if (pr->pr_domain->dom_family == PF_INET &&
1278 		    pr->pr_protocol && pr->pr_protocol == ipproto) {
1279 			ip_protox[pr->pr_protocol] = pr - inetsw;
1280 			return (0);
1281 		}
1282 	}
1283 	return (EPROTONOSUPPORT);
1284 }
1285 
1286 int
1287 ipproto_unregister(short ipproto)
1288 {
1289 	struct protosw *pr;
1290 
1291 	/* Sanity checks. */
1292 	if (ipproto <= 0 || ipproto >= IPPROTO_MAX)
1293 		return (EPROTONOSUPPORT);
1294 
1295 	/* Check if the protocol was indeed registered. */
1296 	pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1297 	if (pr == NULL)
1298 		return (EPFNOSUPPORT);
1299 	if (ip_protox[ipproto] == pr - inetsw)  /* IPPROTO_RAW */
1300 		return (ENOENT);
1301 
1302 	/* Reset the protocol slot to IPPROTO_RAW. */
1303 	ip_protox[ipproto] = pr - inetsw;
1304 	return (0);
1305 }
1306 
1307 /*
1308  * Given address of next destination (final or next hop), return (referenced)
1309  * internet address info of interface to be used to get there.
1310  */
1311 struct in_ifaddr *
1312 ip_rtaddr(struct in_addr dst, u_int fibnum)
1313 {
1314 	struct route sro;
1315 	struct sockaddr_in *sin;
1316 	struct in_ifaddr *ia;
1317 
1318 	bzero(&sro, sizeof(sro));
1319 	sin = (struct sockaddr_in *)&sro.ro_dst;
1320 	sin->sin_family = AF_INET;
1321 	sin->sin_len = sizeof(*sin);
1322 	sin->sin_addr = dst;
1323 	in_rtalloc_ign(&sro, 0, fibnum);
1324 
1325 	if (sro.ro_rt == NULL)
1326 		return (NULL);
1327 
1328 	ia = ifatoia(sro.ro_rt->rt_ifa);
1329 	ifa_ref(&ia->ia_ifa);
1330 	RTFREE(sro.ro_rt);
1331 	return (ia);
1332 }
1333 
1334 u_char inetctlerrmap[PRC_NCMDS] = {
1335 	0,		0,		0,		0,
1336 	0,		EMSGSIZE,	EHOSTDOWN,	EHOSTUNREACH,
1337 	EHOSTUNREACH,	EHOSTUNREACH,	ECONNREFUSED,	ECONNREFUSED,
1338 	EMSGSIZE,	EHOSTUNREACH,	0,		0,
1339 	0,		0,		EHOSTUNREACH,	0,
1340 	ENOPROTOOPT,	ECONNREFUSED
1341 };
1342 
1343 /*
1344  * Forward a packet.  If some error occurs return the sender
1345  * an icmp packet.  Note we can't always generate a meaningful
1346  * icmp message because icmp doesn't have a large enough repertoire
1347  * of codes and types.
1348  *
1349  * If not forwarding, just drop the packet.  This could be confusing
1350  * if ipforwarding was zero but some routing protocol was advancing
1351  * us as a gateway to somewhere.  However, we must let the routing
1352  * protocol deal with that.
1353  *
1354  * The srcrt parameter indicates whether the packet is being forwarded
1355  * via a source route.
1356  */
1357 void
1358 ip_forward(struct mbuf *m, int srcrt)
1359 {
1360 	struct ip *ip = mtod(m, struct ip *);
1361 	struct in_ifaddr *ia;
1362 	struct mbuf *mcopy;
1363 	struct in_addr dest;
1364 	struct route ro;
1365 	int error, type = 0, code = 0, mtu = 0;
1366 
1367 	if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1368 		IPSTAT_INC(ips_cantforward);
1369 		m_freem(m);
1370 		return;
1371 	}
1372 #ifdef IPSTEALTH
1373 	if (!V_ipstealth) {
1374 #endif
1375 		if (ip->ip_ttl <= IPTTLDEC) {
1376 			icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1377 			    0, 0);
1378 			return;
1379 		}
1380 #ifdef IPSTEALTH
1381 	}
1382 #endif
1383 
1384 	ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
1385 #ifndef IPSEC
1386 	/*
1387 	 * 'ia' may be NULL if there is no route for this destination.
1388 	 * In case of IPsec, Don't discard it just yet, but pass it to
1389 	 * ip_output in case of outgoing IPsec policy.
1390 	 */
1391 	if (!srcrt && ia == NULL) {
1392 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1393 		return;
1394 	}
1395 #endif
1396 
1397 	/*
1398 	 * Save the IP header and at most 8 bytes of the payload,
1399 	 * in case we need to generate an ICMP message to the src.
1400 	 *
1401 	 * XXX this can be optimized a lot by saving the data in a local
1402 	 * buffer on the stack (72 bytes at most), and only allocating the
1403 	 * mbuf if really necessary. The vast majority of the packets
1404 	 * are forwarded without having to send an ICMP back (either
1405 	 * because unnecessary, or because rate limited), so we are
1406 	 * really we are wasting a lot of work here.
1407 	 *
1408 	 * We don't use m_copy() because it might return a reference
1409 	 * to a shared cluster. Both this function and ip_output()
1410 	 * assume exclusive access to the IP header in `m', so any
1411 	 * data in a cluster may change before we reach icmp_error().
1412 	 */
1413 	mcopy = m_gethdr(M_NOWAIT, m->m_type);
1414 	if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) {
1415 		/*
1416 		 * It's probably ok if the pkthdr dup fails (because
1417 		 * the deep copy of the tag chain failed), but for now
1418 		 * be conservative and just discard the copy since
1419 		 * code below may some day want the tags.
1420 		 */
1421 		m_free(mcopy);
1422 		mcopy = NULL;
1423 	}
1424 	if (mcopy != NULL) {
1425 		mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy));
1426 		mcopy->m_pkthdr.len = mcopy->m_len;
1427 		m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1428 	}
1429 
1430 #ifdef IPSTEALTH
1431 	if (!V_ipstealth) {
1432 #endif
1433 		ip->ip_ttl -= IPTTLDEC;
1434 #ifdef IPSTEALTH
1435 	}
1436 #endif
1437 
1438 	/*
1439 	 * If forwarding packet using same interface that it came in on,
1440 	 * perhaps should send a redirect to sender to shortcut a hop.
1441 	 * Only send redirect if source is sending directly to us,
1442 	 * and if packet was not source routed (or has any options).
1443 	 * Also, don't send redirect if forwarding using a default route
1444 	 * or a route modified by a redirect.
1445 	 */
1446 	dest.s_addr = 0;
1447 	if (!srcrt && V_ipsendredirects &&
1448 	    ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) {
1449 		struct sockaddr_in *sin;
1450 		struct rtentry *rt;
1451 
1452 		bzero(&ro, sizeof(ro));
1453 		sin = (struct sockaddr_in *)&ro.ro_dst;
1454 		sin->sin_family = AF_INET;
1455 		sin->sin_len = sizeof(*sin);
1456 		sin->sin_addr = ip->ip_dst;
1457 		in_rtalloc_ign(&ro, 0, M_GETFIB(m));
1458 
1459 		rt = ro.ro_rt;
1460 
1461 		if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1462 		    satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1463 #define	RTA(rt)	((struct in_ifaddr *)(rt->rt_ifa))
1464 			u_long src = ntohl(ip->ip_src.s_addr);
1465 
1466 			if (RTA(rt) &&
1467 			    (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1468 				if (rt->rt_flags & RTF_GATEWAY)
1469 					dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1470 				else
1471 					dest.s_addr = ip->ip_dst.s_addr;
1472 				/* Router requirements says to only send host redirects */
1473 				type = ICMP_REDIRECT;
1474 				code = ICMP_REDIRECT_HOST;
1475 			}
1476 		}
1477 		if (rt)
1478 			RTFREE(rt);
1479 	}
1480 
1481 	/*
1482 	 * Try to cache the route MTU from ip_output so we can consider it for
1483 	 * the ICMP_UNREACH_NEEDFRAG "Next-Hop MTU" field described in RFC1191.
1484 	 */
1485 	bzero(&ro, sizeof(ro));
1486 
1487 	error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL);
1488 
1489 	if (error == EMSGSIZE && ro.ro_rt)
1490 		mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1491 	RO_RTFREE(&ro);
1492 
1493 	if (error)
1494 		IPSTAT_INC(ips_cantforward);
1495 	else {
1496 		IPSTAT_INC(ips_forward);
1497 		if (type)
1498 			IPSTAT_INC(ips_redirectsent);
1499 		else {
1500 			if (mcopy)
1501 				m_freem(mcopy);
1502 			if (ia != NULL)
1503 				ifa_free(&ia->ia_ifa);
1504 			return;
1505 		}
1506 	}
1507 	if (mcopy == NULL) {
1508 		if (ia != NULL)
1509 			ifa_free(&ia->ia_ifa);
1510 		return;
1511 	}
1512 
1513 	switch (error) {
1514 
1515 	case 0:				/* forwarded, but need redirect */
1516 		/* type, code set above */
1517 		break;
1518 
1519 	case ENETUNREACH:
1520 	case EHOSTUNREACH:
1521 	case ENETDOWN:
1522 	case EHOSTDOWN:
1523 	default:
1524 		type = ICMP_UNREACH;
1525 		code = ICMP_UNREACH_HOST;
1526 		break;
1527 
1528 	case EMSGSIZE:
1529 		type = ICMP_UNREACH;
1530 		code = ICMP_UNREACH_NEEDFRAG;
1531 
1532 #ifdef IPSEC
1533 		/*
1534 		 * If IPsec is configured for this path,
1535 		 * override any possibly mtu value set by ip_output.
1536 		 */
1537 		mtu = ip_ipsec_mtu(mcopy, mtu);
1538 #endif /* IPSEC */
1539 		/*
1540 		 * If the MTU was set before make sure we are below the
1541 		 * interface MTU.
1542 		 * If the MTU wasn't set before use the interface mtu or
1543 		 * fall back to the next smaller mtu step compared to the
1544 		 * current packet size.
1545 		 */
1546 		if (mtu != 0) {
1547 			if (ia != NULL)
1548 				mtu = min(mtu, ia->ia_ifp->if_mtu);
1549 		} else {
1550 			if (ia != NULL)
1551 				mtu = ia->ia_ifp->if_mtu;
1552 			else
1553 				mtu = ip_next_mtu(ntohs(ip->ip_len), 0);
1554 		}
1555 		IPSTAT_INC(ips_cantfrag);
1556 		break;
1557 
1558 	case ENOBUFS:
1559 		/*
1560 		 * A router should not generate ICMP_SOURCEQUENCH as
1561 		 * required in RFC1812 Requirements for IP Version 4 Routers.
1562 		 * Source quench could be a big problem under DoS attacks,
1563 		 * or if the underlying interface is rate-limited.
1564 		 * Those who need source quench packets may re-enable them
1565 		 * via the net.inet.ip.sendsourcequench sysctl.
1566 		 */
1567 		if (V_ip_sendsourcequench == 0) {
1568 			m_freem(mcopy);
1569 			if (ia != NULL)
1570 				ifa_free(&ia->ia_ifa);
1571 			return;
1572 		} else {
1573 			type = ICMP_SOURCEQUENCH;
1574 			code = 0;
1575 		}
1576 		break;
1577 
1578 	case EACCES:			/* ipfw denied packet */
1579 		m_freem(mcopy);
1580 		if (ia != NULL)
1581 			ifa_free(&ia->ia_ifa);
1582 		return;
1583 	}
1584 	if (ia != NULL)
1585 		ifa_free(&ia->ia_ifa);
1586 	icmp_error(mcopy, type, code, dest.s_addr, mtu);
1587 }
1588 
1589 void
1590 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1591     struct mbuf *m)
1592 {
1593 
1594 	if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1595 		struct bintime bt;
1596 
1597 		bintime(&bt);
1598 		if (inp->inp_socket->so_options & SO_BINTIME) {
1599 			*mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt),
1600 			    SCM_BINTIME, SOL_SOCKET);
1601 			if (*mp)
1602 				mp = &(*mp)->m_next;
1603 		}
1604 		if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1605 			struct timeval tv;
1606 
1607 			bintime2timeval(&bt, &tv);
1608 			*mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv),
1609 			    SCM_TIMESTAMP, SOL_SOCKET);
1610 			if (*mp)
1611 				mp = &(*mp)->m_next;
1612 		}
1613 	}
1614 	if (inp->inp_flags & INP_RECVDSTADDR) {
1615 		*mp = sbcreatecontrol((caddr_t)&ip->ip_dst,
1616 		    sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1617 		if (*mp)
1618 			mp = &(*mp)->m_next;
1619 	}
1620 	if (inp->inp_flags & INP_RECVTTL) {
1621 		*mp = sbcreatecontrol((caddr_t)&ip->ip_ttl,
1622 		    sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1623 		if (*mp)
1624 			mp = &(*mp)->m_next;
1625 	}
1626 #ifdef notyet
1627 	/* XXX
1628 	 * Moving these out of udp_input() made them even more broken
1629 	 * than they already were.
1630 	 */
1631 	/* options were tossed already */
1632 	if (inp->inp_flags & INP_RECVOPTS) {
1633 		*mp = sbcreatecontrol((caddr_t)opts_deleted_above,
1634 		    sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1635 		if (*mp)
1636 			mp = &(*mp)->m_next;
1637 	}
1638 	/* ip_srcroute doesn't do what we want here, need to fix */
1639 	if (inp->inp_flags & INP_RECVRETOPTS) {
1640 		*mp = sbcreatecontrol((caddr_t)ip_srcroute(m),
1641 		    sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1642 		if (*mp)
1643 			mp = &(*mp)->m_next;
1644 	}
1645 #endif
1646 	if (inp->inp_flags & INP_RECVIF) {
1647 		struct ifnet *ifp;
1648 		struct sdlbuf {
1649 			struct sockaddr_dl sdl;
1650 			u_char	pad[32];
1651 		} sdlbuf;
1652 		struct sockaddr_dl *sdp;
1653 		struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1654 
1655 		if ((ifp = m->m_pkthdr.rcvif) &&
1656 		    ifp->if_index && ifp->if_index <= V_if_index) {
1657 			sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1658 			/*
1659 			 * Change our mind and don't try copy.
1660 			 */
1661 			if (sdp->sdl_family != AF_LINK ||
1662 			    sdp->sdl_len > sizeof(sdlbuf)) {
1663 				goto makedummy;
1664 			}
1665 			bcopy(sdp, sdl2, sdp->sdl_len);
1666 		} else {
1667 makedummy:
1668 			sdl2->sdl_len =
1669 			    offsetof(struct sockaddr_dl, sdl_data[0]);
1670 			sdl2->sdl_family = AF_LINK;
1671 			sdl2->sdl_index = 0;
1672 			sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1673 		}
1674 		*mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len,
1675 		    IP_RECVIF, IPPROTO_IP);
1676 		if (*mp)
1677 			mp = &(*mp)->m_next;
1678 	}
1679 	if (inp->inp_flags & INP_RECVTOS) {
1680 		*mp = sbcreatecontrol((caddr_t)&ip->ip_tos,
1681 		    sizeof(u_char), IP_RECVTOS, IPPROTO_IP);
1682 		if (*mp)
1683 			mp = &(*mp)->m_next;
1684 	}
1685 }
1686 
1687 /*
1688  * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1689  * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1690  * locking.  This code remains in ip_input.c as ip_mroute.c is optionally
1691  * compiled.
1692  */
1693 static VNET_DEFINE(int, ip_rsvp_on);
1694 VNET_DEFINE(struct socket *, ip_rsvpd);
1695 
1696 #define	V_ip_rsvp_on		VNET(ip_rsvp_on)
1697 
1698 int
1699 ip_rsvp_init(struct socket *so)
1700 {
1701 
1702 	if (so->so_type != SOCK_RAW ||
1703 	    so->so_proto->pr_protocol != IPPROTO_RSVP)
1704 		return EOPNOTSUPP;
1705 
1706 	if (V_ip_rsvpd != NULL)
1707 		return EADDRINUSE;
1708 
1709 	V_ip_rsvpd = so;
1710 	/*
1711 	 * This may seem silly, but we need to be sure we don't over-increment
1712 	 * the RSVP counter, in case something slips up.
1713 	 */
1714 	if (!V_ip_rsvp_on) {
1715 		V_ip_rsvp_on = 1;
1716 		V_rsvp_on++;
1717 	}
1718 
1719 	return 0;
1720 }
1721 
1722 int
1723 ip_rsvp_done(void)
1724 {
1725 
1726 	V_ip_rsvpd = NULL;
1727 	/*
1728 	 * This may seem silly, but we need to be sure we don't over-decrement
1729 	 * the RSVP counter, in case something slips up.
1730 	 */
1731 	if (V_ip_rsvp_on) {
1732 		V_ip_rsvp_on = 0;
1733 		V_rsvp_on--;
1734 	}
1735 	return 0;
1736 }
1737 
1738 void
1739 rsvp_input(struct mbuf *m, int off)	/* XXX must fixup manually */
1740 {
1741 
1742 	if (rsvp_input_p) { /* call the real one if loaded */
1743 		rsvp_input_p(m, off);
1744 		return;
1745 	}
1746 
1747 	/* Can still get packets with rsvp_on = 0 if there is a local member
1748 	 * of the group to which the RSVP packet is addressed.  But in this
1749 	 * case we want to throw the packet away.
1750 	 */
1751 
1752 	if (!V_rsvp_on) {
1753 		m_freem(m);
1754 		return;
1755 	}
1756 
1757 	if (V_ip_rsvpd != NULL) {
1758 		rip_input(m, off);
1759 		return;
1760 	}
1761 	/* Drop the packet */
1762 	m_freem(m);
1763 }
1764