xref: /freebsd/sys/netinet/ip_icmp.c (revision cd0d51baaa4509a1db83251a601d34404d20c990)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)ip_icmp.c	8.2 (Berkeley) 1/4/94
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/mbuf.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/time.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/rmlock.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
50 
51 #include <net/if.h>
52 #include <net/if_var.h>
53 #include <net/if_types.h>
54 #include <net/route.h>
55 #include <net/vnet.h>
56 
57 #include <netinet/in.h>
58 #include <netinet/in_fib.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/in_var.h>
62 #include <netinet/ip.h>
63 #include <netinet/ip_icmp.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/ip_options.h>
66 #include <netinet/sctp.h>
67 #include <netinet/tcp.h>
68 #include <netinet/tcp_var.h>
69 #include <netinet/tcpip.h>
70 #include <netinet/icmp_var.h>
71 
72 
73 #ifdef INET
74 
75 #include <machine/in_cksum.h>
76 
77 #include <security/mac/mac_framework.h>
78 #endif /* INET */
79 
80 /*
81  * ICMP routines: error generation, receive packet processing, and
82  * routines to turnaround packets back to the originator, and
83  * host table maintenance routines.
84  */
85 VNET_DEFINE_STATIC(int, icmplim) = 200;
86 #define	V_icmplim			VNET(icmplim)
87 SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | CTLFLAG_RW,
88 	&VNET_NAME(icmplim), 0,
89 	"Maximum number of ICMP responses per second");
90 
91 VNET_DEFINE_STATIC(int, icmplim_output) = 1;
92 #define	V_icmplim_output		VNET(icmplim_output)
93 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
94 	&VNET_NAME(icmplim_output), 0,
95 	"Enable logging of ICMP response rate limiting");
96 
97 #ifdef INET
98 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
99 VNET_PCPUSTAT_SYSINIT(icmpstat);
100 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
101     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
102 
103 #ifdef VIMAGE
104 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
105 #endif /* VIMAGE */
106 
107 VNET_DEFINE_STATIC(int, icmpmaskrepl) = 0;
108 #define	V_icmpmaskrepl			VNET(icmpmaskrepl)
109 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
110 	&VNET_NAME(icmpmaskrepl), 0,
111 	"Reply to ICMP Address Mask Request packets");
112 
113 VNET_DEFINE_STATIC(u_int, icmpmaskfake) = 0;
114 #define	V_icmpmaskfake			VNET(icmpmaskfake)
115 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
116 	&VNET_NAME(icmpmaskfake), 0,
117 	"Fake reply to ICMP Address Mask Request packets");
118 
119 VNET_DEFINE(int, drop_redirect) = 0;
120 #define	V_drop_redirect			VNET(drop_redirect)
121 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
122 	&VNET_NAME(drop_redirect), 0,
123 	"Ignore ICMP redirects");
124 
125 VNET_DEFINE_STATIC(int, log_redirect) = 0;
126 #define	V_log_redirect			VNET(log_redirect)
127 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
128 	&VNET_NAME(log_redirect), 0,
129 	"Log ICMP redirects to the console");
130 
131 VNET_DEFINE_STATIC(char, reply_src[IFNAMSIZ]);
132 #define	V_reply_src			VNET(reply_src)
133 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
134 	&VNET_NAME(reply_src), IFNAMSIZ,
135 	"ICMP reply source for non-local packets");
136 
137 VNET_DEFINE_STATIC(int, icmp_rfi) = 0;
138 #define	V_icmp_rfi			VNET(icmp_rfi)
139 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
140 	&VNET_NAME(icmp_rfi), 0,
141 	"ICMP reply from incoming interface for non-local packets");
142 /* Router requirements RFC 1812 section 4.3.2.3 requires 576 - 28. */
143 VNET_DEFINE_STATIC(int, icmp_quotelen) = 548;
144 #define	V_icmp_quotelen			VNET(icmp_quotelen)
145 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
146 	&VNET_NAME(icmp_quotelen), 0,
147 	"Number of bytes from original packet to quote in ICMP reply");
148 
149 VNET_DEFINE_STATIC(int, icmpbmcastecho) = 0;
150 #define	V_icmpbmcastecho		VNET(icmpbmcastecho)
151 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
152 	&VNET_NAME(icmpbmcastecho), 0,
153 	"Reply to multicast ICMP Echo Request and Timestamp packets");
154 
155 VNET_DEFINE_STATIC(int, icmptstamprepl) = 1;
156 #define	V_icmptstamprepl		VNET(icmptstamprepl)
157 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_VNET | CTLFLAG_RW,
158 	&VNET_NAME(icmptstamprepl), 0,
159 	"Respond to ICMP Timestamp packets");
160 
161 VNET_DEFINE_STATIC(int, error_keeptags) = 0;
162 #define	V_error_keeptags		VNET(error_keeptags)
163 SYSCTL_INT(_net_inet_icmp, OID_AUTO, error_keeptags, CTLFLAG_VNET | CTLFLAG_RW,
164 	&VNET_NAME(error_keeptags), 0,
165 	"ICMP error response keeps copy of mbuf_tags of original packet");
166 
167 #ifdef ICMPPRINTFS
168 int	icmpprintfs = 0;
169 #endif
170 
171 static void	icmp_reflect(struct mbuf *);
172 static void	icmp_send(struct mbuf *, struct mbuf *);
173 
174 extern	struct protosw inetsw[];
175 
176 /*
177  * Kernel module interface for updating icmpstat.  The argument is an index
178  * into icmpstat treated as an array of u_long.  While this encodes the
179  * general layout of icmpstat into the caller, it doesn't encode its
180  * location, so that future changes to add, for example, per-CPU stats
181  * support won't cause binary compatibility problems for kernel modules.
182  */
183 void
184 kmod_icmpstat_inc(int statnum)
185 {
186 
187 	counter_u64_add(VNET(icmpstat)[statnum], 1);
188 }
189 
190 /*
191  * Generate an error packet of type error
192  * in response to bad packet ip.
193  */
194 void
195 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
196 {
197 	struct ip *oip, *nip;
198 	struct icmp *icp;
199 	struct mbuf *m;
200 	unsigned icmplen, icmpelen, nlen, oiphlen;
201 
202 	KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
203 	    __func__));
204 
205 	if (type != ICMP_REDIRECT)
206 		ICMPSTAT_INC(icps_error);
207 	/*
208 	 * Don't send error:
209 	 *  if the original packet was encrypted.
210 	 *  if not the first fragment of message.
211 	 *  in response to a multicast or broadcast packet.
212 	 *  if the old packet protocol was an ICMP error message.
213 	 */
214 	if (n->m_flags & M_DECRYPTED)
215 		goto freeit;
216 	if (n->m_flags & (M_BCAST|M_MCAST))
217 		goto freeit;
218 
219 	/* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
220 	if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
221 		goto freeit;
222 	oip = mtod(n, struct ip *);
223 	oiphlen = oip->ip_hl << 2;
224 	if (n->m_len < oiphlen + ICMP_MINLEN)
225 		goto freeit;
226 #ifdef ICMPPRINTFS
227 	if (icmpprintfs)
228 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
229 #endif
230 	if (oip->ip_off & htons(~(IP_MF|IP_DF)))
231 		goto freeit;
232 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
233 	    !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
234 		oiphlen))->icmp_type)) {
235 		ICMPSTAT_INC(icps_oldicmp);
236 		goto freeit;
237 	}
238 	/*
239 	 * Calculate length to quote from original packet and
240 	 * prevent the ICMP mbuf from overflowing.
241 	 * Unfortunately this is non-trivial since ip_forward()
242 	 * sends us truncated packets.
243 	 */
244 	nlen = m_length(n, NULL);
245 	if (oip->ip_p == IPPROTO_TCP) {
246 		struct tcphdr *th;
247 		int tcphlen;
248 
249 		if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
250 		    n->m_next == NULL)
251 			goto stdreply;
252 		if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
253 		    (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
254 			goto freeit;
255 		oip = mtod(n, struct ip *);
256 		th = mtodo(n, oiphlen);
257 		tcphlen = th->th_off << 2;
258 		if (tcphlen < sizeof(struct tcphdr))
259 			goto freeit;
260 		if (ntohs(oip->ip_len) < oiphlen + tcphlen)
261 			goto freeit;
262 		if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
263 			goto stdreply;
264 		if (n->m_len < oiphlen + tcphlen &&
265 		    (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
266 			goto freeit;
267 		oip = mtod(n, struct ip *);
268 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
269 		    ntohs(oip->ip_len) - oiphlen));
270 	} else if (oip->ip_p == IPPROTO_SCTP) {
271 		struct sctphdr *sh;
272 		struct sctp_chunkhdr *ch;
273 
274 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
275 			goto stdreply;
276 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
277 		    n->m_next == NULL)
278 			goto stdreply;
279 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
280 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
281 			goto freeit;
282 		oip = mtod(n, struct ip *);
283 		icmpelen = max(sizeof(struct sctphdr),
284 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
285 		sh = mtodo(n, oiphlen);
286 		if (ntohl(sh->v_tag) == 0 &&
287 		    ntohs(oip->ip_len) >= oiphlen +
288 		    sizeof(struct sctphdr) + 8 &&
289 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
290 		     n->m_next != NULL)) {
291 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
292 			    (n = m_pullup(n, oiphlen +
293 			    sizeof(struct sctphdr) + 8)) == NULL)
294 				goto freeit;
295 			oip = mtod(n, struct ip *);
296 			sh = mtodo(n, oiphlen);
297 			ch = (struct sctp_chunkhdr *)(sh + 1);
298 			if (ch->chunk_type == SCTP_INITIATION) {
299 				icmpelen = max(sizeof(struct sctphdr) + 8,
300 				    min(V_icmp_quotelen, ntohs(oip->ip_len) -
301 				    oiphlen));
302 			}
303 		}
304 	} else
305 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
306 		    oiphlen));
307 
308 	icmplen = min(oiphlen + icmpelen, nlen);
309 	if (icmplen < sizeof(struct ip))
310 		goto freeit;
311 
312 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
313 		m = m_gethdr(M_NOWAIT, MT_DATA);
314 	else
315 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
316 	if (m == NULL)
317 		goto freeit;
318 #ifdef MAC
319 	mac_netinet_icmp_reply(n, m);
320 #endif
321 	icmplen = min(icmplen, M_TRAILINGSPACE(m) -
322 	    sizeof(struct ip) - ICMP_MINLEN);
323 	m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
324 	m->m_data += sizeof(struct ip);
325 	m->m_len = ICMP_MINLEN + icmplen;
326 
327 	/* XXX MRT  make the outgoing packet use the same FIB
328 	 * that was associated with the incoming packet
329 	 */
330 	M_SETFIB(m, M_GETFIB(n));
331 	icp = mtod(m, struct icmp *);
332 	ICMPSTAT_INC(icps_outhist[type]);
333 	icp->icmp_type = type;
334 	if (type == ICMP_REDIRECT)
335 		icp->icmp_gwaddr.s_addr = dest;
336 	else {
337 		icp->icmp_void = 0;
338 		/*
339 		 * The following assignments assume an overlay with the
340 		 * just zeroed icmp_void field.
341 		 */
342 		if (type == ICMP_PARAMPROB) {
343 			icp->icmp_pptr = code;
344 			code = 0;
345 		} else if (type == ICMP_UNREACH &&
346 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
347 			icp->icmp_nextmtu = htons(mtu);
348 		}
349 	}
350 	icp->icmp_code = code;
351 
352 	/*
353 	 * Copy the quotation into ICMP message and
354 	 * convert quoted IP header back to network representation.
355 	 */
356 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
357 	nip = &icp->icmp_ip;
358 
359 	/*
360 	 * Set up ICMP message mbuf and copy old IP header (without options
361 	 * in front of ICMP message.
362 	 * If the original mbuf was meant to bypass the firewall, the error
363 	 * reply should bypass as well.
364 	 */
365 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
366 	KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
367 	    ("insufficient space for ip header"));
368 	m->m_data -= sizeof(struct ip);
369 	m->m_len += sizeof(struct ip);
370 	m->m_pkthdr.len = m->m_len;
371 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
372 	nip = mtod(m, struct ip *);
373 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
374 	nip->ip_len = htons(m->m_len);
375 	nip->ip_v = IPVERSION;
376 	nip->ip_hl = 5;
377 	nip->ip_p = IPPROTO_ICMP;
378 	nip->ip_tos = 0;
379 	nip->ip_off = 0;
380 
381 	if (V_error_keeptags)
382 		m_tag_copy_chain(m, n, M_NOWAIT);
383 
384 	icmp_reflect(m);
385 
386 freeit:
387 	m_freem(n);
388 }
389 
390 /*
391  * Process a received ICMP message.
392  */
393 int
394 icmp_input(struct mbuf **mp, int *offp, int proto)
395 {
396 	struct icmp *icp;
397 	struct in_ifaddr *ia;
398 	struct mbuf *m = *mp;
399 	struct ip *ip = mtod(m, struct ip *);
400 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
401 	int hlen = *offp;
402 	int icmplen = ntohs(ip->ip_len) - *offp;
403 	int i, code;
404 	void (*ctlfunc)(int, struct sockaddr *, void *);
405 	int fibnum;
406 
407 	NET_EPOCH_ASSERT();
408 
409 	*mp = NULL;
410 
411 	/*
412 	 * Locate icmp structure in mbuf, and check
413 	 * that not corrupted and of at least minimum length.
414 	 */
415 #ifdef ICMPPRINTFS
416 	if (icmpprintfs) {
417 		char srcbuf[INET_ADDRSTRLEN];
418 		char dstbuf[INET_ADDRSTRLEN];
419 
420 		printf("icmp_input from %s to %s, len %d\n",
421 		    inet_ntoa_r(ip->ip_src, srcbuf),
422 		    inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
423 	}
424 #endif
425 	if (icmplen < ICMP_MINLEN) {
426 		ICMPSTAT_INC(icps_tooshort);
427 		goto freeit;
428 	}
429 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
430 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
431 		ICMPSTAT_INC(icps_tooshort);
432 		return (IPPROTO_DONE);
433 	}
434 	ip = mtod(m, struct ip *);
435 	m->m_len -= hlen;
436 	m->m_data += hlen;
437 	icp = mtod(m, struct icmp *);
438 	if (in_cksum(m, icmplen)) {
439 		ICMPSTAT_INC(icps_checksum);
440 		goto freeit;
441 	}
442 	m->m_len += hlen;
443 	m->m_data -= hlen;
444 
445 #ifdef ICMPPRINTFS
446 	if (icmpprintfs)
447 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
448 		    icp->icmp_code);
449 #endif
450 
451 	/*
452 	 * Message type specific processing.
453 	 */
454 	if (icp->icmp_type > ICMP_MAXTYPE)
455 		goto raw;
456 
457 	/* Initialize */
458 	bzero(&icmpsrc, sizeof(icmpsrc));
459 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
460 	icmpsrc.sin_family = AF_INET;
461 	bzero(&icmpdst, sizeof(icmpdst));
462 	icmpdst.sin_len = sizeof(struct sockaddr_in);
463 	icmpdst.sin_family = AF_INET;
464 	bzero(&icmpgw, sizeof(icmpgw));
465 	icmpgw.sin_len = sizeof(struct sockaddr_in);
466 	icmpgw.sin_family = AF_INET;
467 
468 	ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
469 	code = icp->icmp_code;
470 	switch (icp->icmp_type) {
471 
472 	case ICMP_UNREACH:
473 		switch (code) {
474 			case ICMP_UNREACH_NET:
475 			case ICMP_UNREACH_HOST:
476 			case ICMP_UNREACH_SRCFAIL:
477 			case ICMP_UNREACH_NET_UNKNOWN:
478 			case ICMP_UNREACH_HOST_UNKNOWN:
479 			case ICMP_UNREACH_ISOLATED:
480 			case ICMP_UNREACH_TOSNET:
481 			case ICMP_UNREACH_TOSHOST:
482 			case ICMP_UNREACH_HOST_PRECEDENCE:
483 			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
484 				code = PRC_UNREACH_NET;
485 				break;
486 
487 			case ICMP_UNREACH_NEEDFRAG:
488 				code = PRC_MSGSIZE;
489 				break;
490 
491 			/*
492 			 * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
493 			 * Treat subcodes 2,3 as immediate RST
494 			 */
495 			case ICMP_UNREACH_PROTOCOL:
496 				code = PRC_UNREACH_PROTOCOL;
497 				break;
498 			case ICMP_UNREACH_PORT:
499 				code = PRC_UNREACH_PORT;
500 				break;
501 
502 			case ICMP_UNREACH_NET_PROHIB:
503 			case ICMP_UNREACH_HOST_PROHIB:
504 			case ICMP_UNREACH_FILTER_PROHIB:
505 				code = PRC_UNREACH_ADMIN_PROHIB;
506 				break;
507 
508 			default:
509 				goto badcode;
510 		}
511 		goto deliver;
512 
513 	case ICMP_TIMXCEED:
514 		if (code > 1)
515 			goto badcode;
516 		code += PRC_TIMXCEED_INTRANS;
517 		goto deliver;
518 
519 	case ICMP_PARAMPROB:
520 		if (code > 1)
521 			goto badcode;
522 		code = PRC_PARAMPROB;
523 	deliver:
524 		/*
525 		 * Problem with datagram; advise higher level routines.
526 		 */
527 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
528 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
529 			ICMPSTAT_INC(icps_badlen);
530 			goto freeit;
531 		}
532 		/* Discard ICMP's in response to multicast packets */
533 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
534 			goto badcode;
535 #ifdef ICMPPRINTFS
536 		if (icmpprintfs)
537 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
538 #endif
539 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
540 		/*
541 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
542 		 * notification to TCP layer.
543 		 */
544 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
545 		ip_stripoptions(m);
546 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
547 			/* This should actually not happen */
548 			ICMPSTAT_INC(icps_tooshort);
549 			return (IPPROTO_DONE);
550 		}
551 		ip = mtod(m, struct ip *);
552 		icp = (struct icmp *)(ip + 1);
553 		/*
554 		 * The upper layer handler can rely on:
555 		 * - The outer IP header has no options.
556 		 * - The outer IP header, the ICMP header, the inner IP header,
557 		 *   and the first n bytes of the inner payload are contiguous.
558 		 *   n is at least 8, but might be larger based on
559 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
560 		 */
561 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
562 		if (ctlfunc)
563 			(*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
564 				   (void *)&icp->icmp_ip);
565 		break;
566 
567 	badcode:
568 		ICMPSTAT_INC(icps_badcode);
569 		break;
570 
571 	case ICMP_ECHO:
572 		if (!V_icmpbmcastecho
573 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
574 			ICMPSTAT_INC(icps_bmcastecho);
575 			break;
576 		}
577 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
578 			goto freeit;
579 		icp->icmp_type = ICMP_ECHOREPLY;
580 		goto reflect;
581 
582 	case ICMP_TSTAMP:
583 		if (V_icmptstamprepl == 0)
584 			break;
585 		if (!V_icmpbmcastecho
586 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
587 			ICMPSTAT_INC(icps_bmcasttstamp);
588 			break;
589 		}
590 		if (icmplen < ICMP_TSLEN) {
591 			ICMPSTAT_INC(icps_badlen);
592 			break;
593 		}
594 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
595 			goto freeit;
596 		icp->icmp_type = ICMP_TSTAMPREPLY;
597 		icp->icmp_rtime = iptime();
598 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
599 		goto reflect;
600 
601 	case ICMP_MASKREQ:
602 		if (V_icmpmaskrepl == 0)
603 			break;
604 		/*
605 		 * We are not able to respond with all ones broadcast
606 		 * unless we receive it over a point-to-point interface.
607 		 */
608 		if (icmplen < ICMP_MASKLEN)
609 			break;
610 		switch (ip->ip_dst.s_addr) {
611 
612 		case INADDR_BROADCAST:
613 		case INADDR_ANY:
614 			icmpdst.sin_addr = ip->ip_src;
615 			break;
616 
617 		default:
618 			icmpdst.sin_addr = ip->ip_dst;
619 		}
620 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
621 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
622 		if (ia == NULL)
623 			break;
624 		if (ia->ia_ifp == NULL)
625 			break;
626 		icp->icmp_type = ICMP_MASKREPLY;
627 		if (V_icmpmaskfake == 0)
628 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
629 		else
630 			icp->icmp_mask = V_icmpmaskfake;
631 		if (ip->ip_src.s_addr == 0) {
632 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
633 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
634 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
635 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
636 		}
637 reflect:
638 		ICMPSTAT_INC(icps_reflect);
639 		ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
640 		icmp_reflect(m);
641 		return (IPPROTO_DONE);
642 
643 	case ICMP_REDIRECT:
644 		if (V_log_redirect) {
645 			u_long src, dst, gw;
646 
647 			src = ntohl(ip->ip_src.s_addr);
648 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
649 			gw = ntohl(icp->icmp_gwaddr.s_addr);
650 			printf("icmp redirect from %d.%d.%d.%d: "
651 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
652 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
653 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
654 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
655 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
656 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
657 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
658 		}
659 		/*
660 		 * RFC1812 says we must ignore ICMP redirects if we
661 		 * are acting as router.
662 		 */
663 		if (V_drop_redirect || V_ipforwarding)
664 			break;
665 		if (code > 3)
666 			goto badcode;
667 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
668 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
669 			ICMPSTAT_INC(icps_badlen);
670 			break;
671 		}
672 		/*
673 		 * Short circuit routing redirects to force
674 		 * immediate change in the kernel's routing
675 		 * tables.  The message is also handed to anyone
676 		 * listening on a raw socket (e.g. the routing
677 		 * daemon for use in updating its tables).
678 		 */
679 		icmpgw.sin_addr = ip->ip_src;
680 		icmpdst.sin_addr = icp->icmp_gwaddr;
681 #ifdef	ICMPPRINTFS
682 		if (icmpprintfs) {
683 			char dstbuf[INET_ADDRSTRLEN];
684 			char gwbuf[INET_ADDRSTRLEN];
685 
686 			printf("redirect dst %s to %s\n",
687 			       inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
688 			       inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
689 		}
690 #endif
691 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
692 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
693 			in_rtredirect((struct sockaddr *)&icmpsrc,
694 			  (struct sockaddr *)&icmpdst,
695 			  (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
696 			  (struct sockaddr *)&icmpgw, fibnum);
697 		}
698 		pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
699 		break;
700 
701 	/*
702 	 * No kernel processing for the following;
703 	 * just fall through to send to raw listener.
704 	 */
705 	case ICMP_ECHOREPLY:
706 	case ICMP_ROUTERADVERT:
707 	case ICMP_ROUTERSOLICIT:
708 	case ICMP_TSTAMPREPLY:
709 	case ICMP_IREQREPLY:
710 	case ICMP_MASKREPLY:
711 	case ICMP_SOURCEQUENCH:
712 	default:
713 		break;
714 	}
715 
716 raw:
717 	*mp = m;
718 	rip_input(mp, offp, proto);
719 	return (IPPROTO_DONE);
720 
721 freeit:
722 	m_freem(m);
723 	return (IPPROTO_DONE);
724 }
725 
726 /*
727  * Reflect the ip packet back to the source
728  */
729 static void
730 icmp_reflect(struct mbuf *m)
731 {
732 	struct rm_priotracker in_ifa_tracker;
733 	struct ip *ip = mtod(m, struct ip *);
734 	struct ifaddr *ifa;
735 	struct ifnet *ifp;
736 	struct in_ifaddr *ia;
737 	struct in_addr t;
738 	struct nhop4_extended nh_ext;
739 	struct mbuf *opts = NULL;
740 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
741 
742 	NET_EPOCH_ASSERT();
743 
744 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
745 	    IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
746 	    IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
747 		m_freem(m);	/* Bad return address */
748 		ICMPSTAT_INC(icps_badaddr);
749 		goto done;	/* Ip_output() will check for broadcast */
750 	}
751 
752 	t = ip->ip_dst;
753 	ip->ip_dst = ip->ip_src;
754 
755 	/*
756 	 * Source selection for ICMP replies:
757 	 *
758 	 * If the incoming packet was addressed directly to one of our
759 	 * own addresses, use dst as the src for the reply.
760 	 */
761 	IN_IFADDR_RLOCK(&in_ifa_tracker);
762 	LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
763 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
764 			t = IA_SIN(ia)->sin_addr;
765 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
766 			goto match;
767 		}
768 	}
769 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
770 
771 	/*
772 	 * If the incoming packet was addressed to one of our broadcast
773 	 * addresses, use the first non-broadcast address which corresponds
774 	 * to the incoming interface.
775 	 */
776 	ifp = m->m_pkthdr.rcvif;
777 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
778 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
779 			if (ifa->ifa_addr->sa_family != AF_INET)
780 				continue;
781 			ia = ifatoia(ifa);
782 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
783 			    t.s_addr) {
784 				t = IA_SIN(ia)->sin_addr;
785 				goto match;
786 			}
787 		}
788 	}
789 	/*
790 	 * If the packet was transiting through us, use the address of
791 	 * the interface the packet came through in.  If that interface
792 	 * doesn't have a suitable IP address, the normal selection
793 	 * criteria apply.
794 	 */
795 	if (V_icmp_rfi && ifp != NULL) {
796 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
797 			if (ifa->ifa_addr->sa_family != AF_INET)
798 				continue;
799 			ia = ifatoia(ifa);
800 			t = IA_SIN(ia)->sin_addr;
801 			goto match;
802 		}
803 	}
804 	/*
805 	 * If the incoming packet was not addressed directly to us, use
806 	 * designated interface for icmp replies specified by sysctl
807 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
808 	 * with normal source selection.
809 	 */
810 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
811 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
812 			if (ifa->ifa_addr->sa_family != AF_INET)
813 				continue;
814 			ia = ifatoia(ifa);
815 			t = IA_SIN(ia)->sin_addr;
816 			goto match;
817 		}
818 	}
819 	/*
820 	 * If the packet was transiting through us, use the address of
821 	 * the interface that is the closest to the packet source.
822 	 * When we don't have a route back to the packet source, stop here
823 	 * and drop the packet.
824 	 */
825 	if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
826 		m_freem(m);
827 		ICMPSTAT_INC(icps_noroute);
828 		goto done;
829 	}
830 	t = nh_ext.nh_src;
831 match:
832 #ifdef MAC
833 	mac_netinet_icmp_replyinplace(m);
834 #endif
835 	ip->ip_src = t;
836 	ip->ip_ttl = V_ip_defttl;
837 
838 	if (optlen > 0) {
839 		u_char *cp;
840 		int opt, cnt;
841 		u_int len;
842 
843 		/*
844 		 * Retrieve any source routing from the incoming packet;
845 		 * add on any record-route or timestamp options.
846 		 */
847 		cp = (u_char *) (ip + 1);
848 		if ((opts = ip_srcroute(m)) == NULL &&
849 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
850 			opts->m_len = sizeof(struct in_addr);
851 			mtod(opts, struct in_addr *)->s_addr = 0;
852 		}
853 		if (opts) {
854 #ifdef ICMPPRINTFS
855 		    if (icmpprintfs)
856 			    printf("icmp_reflect optlen %d rt %d => ",
857 				optlen, opts->m_len);
858 #endif
859 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
860 			    opt = cp[IPOPT_OPTVAL];
861 			    if (opt == IPOPT_EOL)
862 				    break;
863 			    if (opt == IPOPT_NOP)
864 				    len = 1;
865 			    else {
866 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
867 					    break;
868 				    len = cp[IPOPT_OLEN];
869 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
870 				        len > cnt)
871 					    break;
872 			    }
873 			    /*
874 			     * Should check for overflow, but it "can't happen"
875 			     */
876 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
877 				opt == IPOPT_SECURITY) {
878 				    bcopy((caddr_t)cp,
879 					mtod(opts, caddr_t) + opts->m_len, len);
880 				    opts->m_len += len;
881 			    }
882 		    }
883 		    /* Terminate & pad, if necessary */
884 		    cnt = opts->m_len % 4;
885 		    if (cnt) {
886 			    for (; cnt < 4; cnt++) {
887 				    *(mtod(opts, caddr_t) + opts->m_len) =
888 					IPOPT_EOL;
889 				    opts->m_len++;
890 			    }
891 		    }
892 #ifdef ICMPPRINTFS
893 		    if (icmpprintfs)
894 			    printf("%d\n", opts->m_len);
895 #endif
896 		}
897 		ip_stripoptions(m);
898 	}
899 	m_tag_delete_nonpersistent(m);
900 	m->m_flags &= ~(M_BCAST|M_MCAST);
901 	icmp_send(m, opts);
902 done:
903 	if (opts)
904 		(void)m_free(opts);
905 }
906 
907 /*
908  * Send an icmp packet back to the ip level,
909  * after supplying a checksum.
910  */
911 static void
912 icmp_send(struct mbuf *m, struct mbuf *opts)
913 {
914 	struct ip *ip = mtod(m, struct ip *);
915 	int hlen;
916 	struct icmp *icp;
917 
918 	hlen = ip->ip_hl << 2;
919 	m->m_data += hlen;
920 	m->m_len -= hlen;
921 	icp = mtod(m, struct icmp *);
922 	icp->icmp_cksum = 0;
923 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
924 	m->m_data -= hlen;
925 	m->m_len += hlen;
926 	m->m_pkthdr.rcvif = (struct ifnet *)0;
927 #ifdef ICMPPRINTFS
928 	if (icmpprintfs) {
929 		char dstbuf[INET_ADDRSTRLEN];
930 		char srcbuf[INET_ADDRSTRLEN];
931 
932 		printf("icmp_send dst %s src %s\n",
933 		    inet_ntoa_r(ip->ip_dst, dstbuf),
934 		    inet_ntoa_r(ip->ip_src, srcbuf));
935 	}
936 #endif
937 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
938 }
939 
940 /*
941  * Return milliseconds since 00:00 UTC in network format.
942  */
943 uint32_t
944 iptime(void)
945 {
946 	struct timeval atv;
947 	u_long t;
948 
949 	getmicrotime(&atv);
950 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
951 	return (htonl(t));
952 }
953 
954 /*
955  * Return the next larger or smaller MTU plateau (table from RFC 1191)
956  * given current value MTU.  If DIR is less than zero, a larger plateau
957  * is returned; otherwise, a smaller value is returned.
958  */
959 int
960 ip_next_mtu(int mtu, int dir)
961 {
962 	static int mtutab[] = {
963 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
964 		296, 68, 0
965 	};
966 	int i, size;
967 
968 	size = (sizeof mtutab) / (sizeof mtutab[0]);
969 	if (dir >= 0) {
970 		for (i = 0; i < size; i++)
971 			if (mtu > mtutab[i])
972 				return mtutab[i];
973 	} else {
974 		for (i = size - 1; i >= 0; i--)
975 			if (mtu < mtutab[i])
976 				return mtutab[i];
977 		if (mtu == mtutab[0])
978 			return mtutab[0];
979 	}
980 	return 0;
981 }
982 #endif /* INET */
983 
984 
985 /*
986  * badport_bandlim() - check for ICMP bandwidth limit
987  *
988  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
989  *	hit our bandwidth limit and it is not ok.
990  *
991  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
992  *
993  *	For now we separate the TCP and UDP subsystems w/ different 'which'
994  *	values.  We may eventually remove this separation (and simplify the
995  *	code further).
996  *
997  *	Note that the printing of the error message is delayed so we can
998  *	properly print the icmp error rate that the system was trying to do
999  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1000  *	the 'final' error, but it doesn't make sense to solve the printing
1001  *	delay with more complex code.
1002  */
1003 struct icmp_rate {
1004 	const char *descr;
1005 	struct counter_rate cr;
1006 };
1007 VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1008 	{ "icmp unreach response" },
1009 	{ "icmp ping response" },
1010 	{ "icmp tstamp response" },
1011 	{ "closed port RST response" },
1012 	{ "open port RST response" },
1013 	{ "icmp6 unreach response" },
1014 	{ "sctp ootb response" }
1015 };
1016 #define	V_icmp_rates	VNET(icmp_rates)
1017 
1018 static void
1019 icmp_bandlimit_init(void)
1020 {
1021 
1022 	for (int i = 0; i < BANDLIM_MAX; i++) {
1023 		V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1024 		V_icmp_rates[i].cr.cr_ticks = ticks;
1025 	}
1026 }
1027 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1028     icmp_bandlimit_init, NULL);
1029 
1030 static void
1031 icmp_bandlimit_uninit(void)
1032 {
1033 
1034 	for (int i = 0; i < BANDLIM_MAX; i++)
1035 		counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1036 }
1037 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1038     icmp_bandlimit_uninit, NULL);
1039 
1040 int
1041 badport_bandlim(int which)
1042 {
1043 	int64_t pps;
1044 
1045 	if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1046 		return (0);
1047 
1048 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1049 	    ("%s: which %d", __func__, which));
1050 
1051 	pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1052 	if (pps == -1)
1053 		return (-1);
1054 	if (pps > 0 && V_icmplim_output)
1055 		log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1056 			V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1057 	return (0);
1058 }
1059