xref: /freebsd/sys/netinet/ip_icmp.c (revision c1cdf6a42f0d951ba720688dfc6ce07608b02f6e)
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_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 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
268 		    ntohs(oip->ip_len) - oiphlen));
269 	} else if (oip->ip_p == IPPROTO_SCTP) {
270 		struct sctphdr *sh;
271 		struct sctp_chunkhdr *ch;
272 
273 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
274 			goto stdreply;
275 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
276 		    n->m_next == NULL)
277 			goto stdreply;
278 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
279 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
280 			goto freeit;
281 		oip = mtod(n, struct ip *);
282 		icmpelen = max(sizeof(struct sctphdr),
283 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
284 		sh = mtodo(n, oiphlen);
285 		if (ntohl(sh->v_tag) == 0 &&
286 		    ntohs(oip->ip_len) >= oiphlen +
287 		    sizeof(struct sctphdr) + 8 &&
288 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
289 		     n->m_next != NULL)) {
290 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
291 			    (n = m_pullup(n, oiphlen +
292 			    sizeof(struct sctphdr) + 8)) == NULL)
293 				goto freeit;
294 			oip = mtod(n, struct ip *);
295 			sh = mtodo(n, oiphlen);
296 			ch = (struct sctp_chunkhdr *)(sh + 1);
297 			if (ch->chunk_type == SCTP_INITIATION) {
298 				icmpelen = max(sizeof(struct sctphdr) + 8,
299 				    min(V_icmp_quotelen, ntohs(oip->ip_len) -
300 				    oiphlen));
301 			}
302 		}
303 	} else
304 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
305 		    oiphlen));
306 
307 	icmplen = min(oiphlen + icmpelen, nlen);
308 	if (icmplen < sizeof(struct ip))
309 		goto freeit;
310 
311 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
312 		m = m_gethdr(M_NOWAIT, MT_DATA);
313 	else
314 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
315 	if (m == NULL)
316 		goto freeit;
317 #ifdef MAC
318 	mac_netinet_icmp_reply(n, m);
319 #endif
320 	icmplen = min(icmplen, M_TRAILINGSPACE(m) -
321 	    sizeof(struct ip) - ICMP_MINLEN);
322 	m_align(m, ICMP_MINLEN + icmplen);
323 	m->m_len = ICMP_MINLEN + icmplen;
324 
325 	/* XXX MRT  make the outgoing packet use the same FIB
326 	 * that was associated with the incoming packet
327 	 */
328 	M_SETFIB(m, M_GETFIB(n));
329 	icp = mtod(m, struct icmp *);
330 	ICMPSTAT_INC(icps_outhist[type]);
331 	icp->icmp_type = type;
332 	if (type == ICMP_REDIRECT)
333 		icp->icmp_gwaddr.s_addr = dest;
334 	else {
335 		icp->icmp_void = 0;
336 		/*
337 		 * The following assignments assume an overlay with the
338 		 * just zeroed icmp_void field.
339 		 */
340 		if (type == ICMP_PARAMPROB) {
341 			icp->icmp_pptr = code;
342 			code = 0;
343 		} else if (type == ICMP_UNREACH &&
344 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
345 			icp->icmp_nextmtu = htons(mtu);
346 		}
347 	}
348 	icp->icmp_code = code;
349 
350 	/*
351 	 * Copy the quotation into ICMP message and
352 	 * convert quoted IP header back to network representation.
353 	 */
354 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
355 	nip = &icp->icmp_ip;
356 
357 	/*
358 	 * Set up ICMP message mbuf and copy old IP header (without options
359 	 * in front of ICMP message.
360 	 * If the original mbuf was meant to bypass the firewall, the error
361 	 * reply should bypass as well.
362 	 */
363 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
364 	m->m_data -= sizeof(struct ip);
365 	m->m_len += sizeof(struct ip);
366 	m->m_pkthdr.len = m->m_len;
367 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
368 	nip = mtod(m, struct ip *);
369 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
370 	nip->ip_len = htons(m->m_len);
371 	nip->ip_v = IPVERSION;
372 	nip->ip_hl = 5;
373 	nip->ip_p = IPPROTO_ICMP;
374 	nip->ip_tos = 0;
375 	nip->ip_off = 0;
376 
377 	if (V_error_keeptags)
378 		m_tag_copy_chain(m, n, M_NOWAIT);
379 
380 	icmp_reflect(m);
381 
382 freeit:
383 	m_freem(n);
384 }
385 
386 /*
387  * Process a received ICMP message.
388  */
389 int
390 icmp_input(struct mbuf **mp, int *offp, int proto)
391 {
392 	struct icmp *icp;
393 	struct in_ifaddr *ia;
394 	struct mbuf *m = *mp;
395 	struct ip *ip = mtod(m, struct ip *);
396 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
397 	int hlen = *offp;
398 	int icmplen = ntohs(ip->ip_len) - *offp;
399 	int i, code;
400 	void (*ctlfunc)(int, struct sockaddr *, void *);
401 	int fibnum;
402 
403 	*mp = NULL;
404 
405 	/*
406 	 * Locate icmp structure in mbuf, and check
407 	 * that not corrupted and of at least minimum length.
408 	 */
409 #ifdef ICMPPRINTFS
410 	if (icmpprintfs) {
411 		char srcbuf[INET_ADDRSTRLEN];
412 		char dstbuf[INET_ADDRSTRLEN];
413 
414 		printf("icmp_input from %s to %s, len %d\n",
415 		    inet_ntoa_r(ip->ip_src, srcbuf),
416 		    inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
417 	}
418 #endif
419 	NET_EPOCH_ENTER();
420 	if (icmplen < ICMP_MINLEN) {
421 		ICMPSTAT_INC(icps_tooshort);
422 		goto freeit;
423 	}
424 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
425 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
426 		ICMPSTAT_INC(icps_tooshort);
427 		NET_EPOCH_EXIT();
428 		return (IPPROTO_DONE);
429 	}
430 	ip = mtod(m, struct ip *);
431 	m->m_len -= hlen;
432 	m->m_data += hlen;
433 	icp = mtod(m, struct icmp *);
434 	if (in_cksum(m, icmplen)) {
435 		ICMPSTAT_INC(icps_checksum);
436 		goto freeit;
437 	}
438 	m->m_len += hlen;
439 	m->m_data -= hlen;
440 
441 #ifdef ICMPPRINTFS
442 	if (icmpprintfs)
443 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
444 		    icp->icmp_code);
445 #endif
446 
447 	/*
448 	 * Message type specific processing.
449 	 */
450 	if (icp->icmp_type > ICMP_MAXTYPE)
451 		goto raw;
452 
453 	/* Initialize */
454 	bzero(&icmpsrc, sizeof(icmpsrc));
455 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
456 	icmpsrc.sin_family = AF_INET;
457 	bzero(&icmpdst, sizeof(icmpdst));
458 	icmpdst.sin_len = sizeof(struct sockaddr_in);
459 	icmpdst.sin_family = AF_INET;
460 	bzero(&icmpgw, sizeof(icmpgw));
461 	icmpgw.sin_len = sizeof(struct sockaddr_in);
462 	icmpgw.sin_family = AF_INET;
463 
464 	ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
465 	code = icp->icmp_code;
466 	switch (icp->icmp_type) {
467 
468 	case ICMP_UNREACH:
469 		switch (code) {
470 			case ICMP_UNREACH_NET:
471 			case ICMP_UNREACH_HOST:
472 			case ICMP_UNREACH_SRCFAIL:
473 			case ICMP_UNREACH_NET_UNKNOWN:
474 			case ICMP_UNREACH_HOST_UNKNOWN:
475 			case ICMP_UNREACH_ISOLATED:
476 			case ICMP_UNREACH_TOSNET:
477 			case ICMP_UNREACH_TOSHOST:
478 			case ICMP_UNREACH_HOST_PRECEDENCE:
479 			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
480 				code = PRC_UNREACH_NET;
481 				break;
482 
483 			case ICMP_UNREACH_NEEDFRAG:
484 				code = PRC_MSGSIZE;
485 				break;
486 
487 			/*
488 			 * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
489 			 * Treat subcodes 2,3 as immediate RST
490 			 */
491 			case ICMP_UNREACH_PROTOCOL:
492 				code = PRC_UNREACH_PROTOCOL;
493 				break;
494 			case ICMP_UNREACH_PORT:
495 				code = PRC_UNREACH_PORT;
496 				break;
497 
498 			case ICMP_UNREACH_NET_PROHIB:
499 			case ICMP_UNREACH_HOST_PROHIB:
500 			case ICMP_UNREACH_FILTER_PROHIB:
501 				code = PRC_UNREACH_ADMIN_PROHIB;
502 				break;
503 
504 			default:
505 				goto badcode;
506 		}
507 		goto deliver;
508 
509 	case ICMP_TIMXCEED:
510 		if (code > 1)
511 			goto badcode;
512 		code += PRC_TIMXCEED_INTRANS;
513 		goto deliver;
514 
515 	case ICMP_PARAMPROB:
516 		if (code > 1)
517 			goto badcode;
518 		code = PRC_PARAMPROB;
519 	deliver:
520 		/*
521 		 * Problem with datagram; advise higher level routines.
522 		 */
523 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
524 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
525 			ICMPSTAT_INC(icps_badlen);
526 			goto freeit;
527 		}
528 		/* Discard ICMP's in response to multicast packets */
529 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
530 			goto badcode;
531 #ifdef ICMPPRINTFS
532 		if (icmpprintfs)
533 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
534 #endif
535 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
536 		/*
537 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
538 		 * notification to TCP layer.
539 		 */
540 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
541 		ip_stripoptions(m);
542 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
543 			/* This should actually not happen */
544 			ICMPSTAT_INC(icps_tooshort);
545 			NET_EPOCH_EXIT();
546 			return (IPPROTO_DONE);
547 		}
548 		ip = mtod(m, struct ip *);
549 		icp = (struct icmp *)(ip + 1);
550 		/*
551 		 * The upper layer handler can rely on:
552 		 * - The outer IP header has no options.
553 		 * - The outer IP header, the ICMP header, the inner IP header,
554 		 *   and the first n bytes of the inner payload are contiguous.
555 		 *   n is at least 8, but might be larger based on
556 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
557 		 */
558 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
559 		if (ctlfunc)
560 			(*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
561 				   (void *)&icp->icmp_ip);
562 		break;
563 
564 	badcode:
565 		ICMPSTAT_INC(icps_badcode);
566 		break;
567 
568 	case ICMP_ECHO:
569 		if (!V_icmpbmcastecho
570 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
571 			ICMPSTAT_INC(icps_bmcastecho);
572 			break;
573 		}
574 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
575 			goto freeit;
576 		icp->icmp_type = ICMP_ECHOREPLY;
577 		goto reflect;
578 
579 	case ICMP_TSTAMP:
580 		if (V_icmptstamprepl == 0)
581 			break;
582 		if (!V_icmpbmcastecho
583 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
584 			ICMPSTAT_INC(icps_bmcasttstamp);
585 			break;
586 		}
587 		if (icmplen < ICMP_TSLEN) {
588 			ICMPSTAT_INC(icps_badlen);
589 			break;
590 		}
591 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
592 			goto freeit;
593 		icp->icmp_type = ICMP_TSTAMPREPLY;
594 		icp->icmp_rtime = iptime();
595 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
596 		goto reflect;
597 
598 	case ICMP_MASKREQ:
599 		if (V_icmpmaskrepl == 0)
600 			break;
601 		/*
602 		 * We are not able to respond with all ones broadcast
603 		 * unless we receive it over a point-to-point interface.
604 		 */
605 		if (icmplen < ICMP_MASKLEN)
606 			break;
607 		switch (ip->ip_dst.s_addr) {
608 
609 		case INADDR_BROADCAST:
610 		case INADDR_ANY:
611 			icmpdst.sin_addr = ip->ip_src;
612 			break;
613 
614 		default:
615 			icmpdst.sin_addr = ip->ip_dst;
616 		}
617 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
618 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
619 		if (ia == NULL)
620 			break;
621 		if (ia->ia_ifp == NULL)
622 			break;
623 		icp->icmp_type = ICMP_MASKREPLY;
624 		if (V_icmpmaskfake == 0)
625 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
626 		else
627 			icp->icmp_mask = V_icmpmaskfake;
628 		if (ip->ip_src.s_addr == 0) {
629 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
630 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
631 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
632 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
633 		}
634 reflect:
635 		ICMPSTAT_INC(icps_reflect);
636 		ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
637 		icmp_reflect(m);
638 		NET_EPOCH_EXIT();
639 		return (IPPROTO_DONE);
640 
641 	case ICMP_REDIRECT:
642 		if (V_log_redirect) {
643 			u_long src, dst, gw;
644 
645 			src = ntohl(ip->ip_src.s_addr);
646 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
647 			gw = ntohl(icp->icmp_gwaddr.s_addr);
648 			printf("icmp redirect from %d.%d.%d.%d: "
649 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
650 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
651 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
652 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
653 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
654 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
655 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
656 		}
657 		/*
658 		 * RFC1812 says we must ignore ICMP redirects if we
659 		 * are acting as router.
660 		 */
661 		if (V_drop_redirect || V_ipforwarding)
662 			break;
663 		if (code > 3)
664 			goto badcode;
665 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
666 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
667 			ICMPSTAT_INC(icps_badlen);
668 			break;
669 		}
670 		/*
671 		 * Short circuit routing redirects to force
672 		 * immediate change in the kernel's routing
673 		 * tables.  The message is also handed to anyone
674 		 * listening on a raw socket (e.g. the routing
675 		 * daemon for use in updating its tables).
676 		 */
677 		icmpgw.sin_addr = ip->ip_src;
678 		icmpdst.sin_addr = icp->icmp_gwaddr;
679 #ifdef	ICMPPRINTFS
680 		if (icmpprintfs) {
681 			char dstbuf[INET_ADDRSTRLEN];
682 			char gwbuf[INET_ADDRSTRLEN];
683 
684 			printf("redirect dst %s to %s\n",
685 			       inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
686 			       inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
687 		}
688 #endif
689 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
690 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
691 			in_rtredirect((struct sockaddr *)&icmpsrc,
692 			  (struct sockaddr *)&icmpdst,
693 			  (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
694 			  (struct sockaddr *)&icmpgw, fibnum);
695 		}
696 		pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
697 		break;
698 
699 	/*
700 	 * No kernel processing for the following;
701 	 * just fall through to send to raw listener.
702 	 */
703 	case ICMP_ECHOREPLY:
704 	case ICMP_ROUTERADVERT:
705 	case ICMP_ROUTERSOLICIT:
706 	case ICMP_TSTAMPREPLY:
707 	case ICMP_IREQREPLY:
708 	case ICMP_MASKREPLY:
709 	case ICMP_SOURCEQUENCH:
710 	default:
711 		break;
712 	}
713 
714 raw:
715 	NET_EPOCH_EXIT();
716 	*mp = m;
717 	rip_input(mp, offp, proto);
718 	return (IPPROTO_DONE);
719 
720 freeit:
721 	NET_EPOCH_EXIT();
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 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
743 	    IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
744 	    IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
745 		m_freem(m);	/* Bad return address */
746 		ICMPSTAT_INC(icps_badaddr);
747 		goto done;	/* Ip_output() will check for broadcast */
748 	}
749 
750 	t = ip->ip_dst;
751 	ip->ip_dst = ip->ip_src;
752 
753 	/*
754 	 * Source selection for ICMP replies:
755 	 *
756 	 * If the incoming packet was addressed directly to one of our
757 	 * own addresses, use dst as the src for the reply.
758 	 */
759 	IN_IFADDR_RLOCK(&in_ifa_tracker);
760 	LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
761 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
762 			t = IA_SIN(ia)->sin_addr;
763 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
764 			goto match;
765 		}
766 	}
767 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
768 
769 	/*
770 	 * If the incoming packet was addressed to one of our broadcast
771 	 * addresses, use the first non-broadcast address which corresponds
772 	 * to the incoming interface.
773 	 */
774 	ifp = m->m_pkthdr.rcvif;
775 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
776 		IF_ADDR_RLOCK(ifp);
777 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
778 			if (ifa->ifa_addr->sa_family != AF_INET)
779 				continue;
780 			ia = ifatoia(ifa);
781 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
782 			    t.s_addr) {
783 				t = IA_SIN(ia)->sin_addr;
784 				IF_ADDR_RUNLOCK(ifp);
785 				goto match;
786 			}
787 		}
788 		IF_ADDR_RUNLOCK(ifp);
789 	}
790 	/*
791 	 * If the packet was transiting through us, use the address of
792 	 * the interface the packet came through in.  If that interface
793 	 * doesn't have a suitable IP address, the normal selection
794 	 * criteria apply.
795 	 */
796 	if (V_icmp_rfi && ifp != NULL) {
797 		IF_ADDR_RLOCK(ifp);
798 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
799 			if (ifa->ifa_addr->sa_family != AF_INET)
800 				continue;
801 			ia = ifatoia(ifa);
802 			t = IA_SIN(ia)->sin_addr;
803 			IF_ADDR_RUNLOCK(ifp);
804 			goto match;
805 		}
806 		IF_ADDR_RUNLOCK(ifp);
807 	}
808 	/*
809 	 * If the incoming packet was not addressed directly to us, use
810 	 * designated interface for icmp replies specified by sysctl
811 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
812 	 * with normal source selection.
813 	 */
814 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
815 		IF_ADDR_RLOCK(ifp);
816 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
817 			if (ifa->ifa_addr->sa_family != AF_INET)
818 				continue;
819 			ia = ifatoia(ifa);
820 			t = IA_SIN(ia)->sin_addr;
821 			IF_ADDR_RUNLOCK(ifp);
822 			goto match;
823 		}
824 		IF_ADDR_RUNLOCK(ifp);
825 	}
826 	/*
827 	 * If the packet was transiting through us, use the address of
828 	 * the interface that is the closest to the packet source.
829 	 * When we don't have a route back to the packet source, stop here
830 	 * and drop the packet.
831 	 */
832 	if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
833 		m_freem(m);
834 		ICMPSTAT_INC(icps_noroute);
835 		goto done;
836 	}
837 	t = nh_ext.nh_src;
838 match:
839 #ifdef MAC
840 	mac_netinet_icmp_replyinplace(m);
841 #endif
842 	ip->ip_src = t;
843 	ip->ip_ttl = V_ip_defttl;
844 
845 	if (optlen > 0) {
846 		u_char *cp;
847 		int opt, cnt;
848 		u_int len;
849 
850 		/*
851 		 * Retrieve any source routing from the incoming packet;
852 		 * add on any record-route or timestamp options.
853 		 */
854 		cp = (u_char *) (ip + 1);
855 		if ((opts = ip_srcroute(m)) == NULL &&
856 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
857 			opts->m_len = sizeof(struct in_addr);
858 			mtod(opts, struct in_addr *)->s_addr = 0;
859 		}
860 		if (opts) {
861 #ifdef ICMPPRINTFS
862 		    if (icmpprintfs)
863 			    printf("icmp_reflect optlen %d rt %d => ",
864 				optlen, opts->m_len);
865 #endif
866 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
867 			    opt = cp[IPOPT_OPTVAL];
868 			    if (opt == IPOPT_EOL)
869 				    break;
870 			    if (opt == IPOPT_NOP)
871 				    len = 1;
872 			    else {
873 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
874 					    break;
875 				    len = cp[IPOPT_OLEN];
876 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
877 				        len > cnt)
878 					    break;
879 			    }
880 			    /*
881 			     * Should check for overflow, but it "can't happen"
882 			     */
883 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
884 				opt == IPOPT_SECURITY) {
885 				    bcopy((caddr_t)cp,
886 					mtod(opts, caddr_t) + opts->m_len, len);
887 				    opts->m_len += len;
888 			    }
889 		    }
890 		    /* Terminate & pad, if necessary */
891 		    cnt = opts->m_len % 4;
892 		    if (cnt) {
893 			    for (; cnt < 4; cnt++) {
894 				    *(mtod(opts, caddr_t) + opts->m_len) =
895 					IPOPT_EOL;
896 				    opts->m_len++;
897 			    }
898 		    }
899 #ifdef ICMPPRINTFS
900 		    if (icmpprintfs)
901 			    printf("%d\n", opts->m_len);
902 #endif
903 		}
904 		ip_stripoptions(m);
905 	}
906 	m_tag_delete_nonpersistent(m);
907 	m->m_flags &= ~(M_BCAST|M_MCAST);
908 	icmp_send(m, opts);
909 done:
910 	if (opts)
911 		(void)m_free(opts);
912 }
913 
914 /*
915  * Send an icmp packet back to the ip level,
916  * after supplying a checksum.
917  */
918 static void
919 icmp_send(struct mbuf *m, struct mbuf *opts)
920 {
921 	struct ip *ip = mtod(m, struct ip *);
922 	int hlen;
923 	struct icmp *icp;
924 
925 	hlen = ip->ip_hl << 2;
926 	m->m_data += hlen;
927 	m->m_len -= hlen;
928 	icp = mtod(m, struct icmp *);
929 	icp->icmp_cksum = 0;
930 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
931 	m->m_data -= hlen;
932 	m->m_len += hlen;
933 	m->m_pkthdr.rcvif = (struct ifnet *)0;
934 #ifdef ICMPPRINTFS
935 	if (icmpprintfs) {
936 		char dstbuf[INET_ADDRSTRLEN];
937 		char srcbuf[INET_ADDRSTRLEN];
938 
939 		printf("icmp_send dst %s src %s\n",
940 		    inet_ntoa_r(ip->ip_dst, dstbuf),
941 		    inet_ntoa_r(ip->ip_src, srcbuf));
942 	}
943 #endif
944 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
945 }
946 
947 /*
948  * Return milliseconds since 00:00 UTC in network format.
949  */
950 uint32_t
951 iptime(void)
952 {
953 	struct timeval atv;
954 	u_long t;
955 
956 	getmicrotime(&atv);
957 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
958 	return (htonl(t));
959 }
960 
961 /*
962  * Return the next larger or smaller MTU plateau (table from RFC 1191)
963  * given current value MTU.  If DIR is less than zero, a larger plateau
964  * is returned; otherwise, a smaller value is returned.
965  */
966 int
967 ip_next_mtu(int mtu, int dir)
968 {
969 	static int mtutab[] = {
970 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
971 		296, 68, 0
972 	};
973 	int i, size;
974 
975 	size = (sizeof mtutab) / (sizeof mtutab[0]);
976 	if (dir >= 0) {
977 		for (i = 0; i < size; i++)
978 			if (mtu > mtutab[i])
979 				return mtutab[i];
980 	} else {
981 		for (i = size - 1; i >= 0; i--)
982 			if (mtu < mtutab[i])
983 				return mtutab[i];
984 		if (mtu == mtutab[0])
985 			return mtutab[0];
986 	}
987 	return 0;
988 }
989 #endif /* INET */
990 
991 
992 /*
993  * badport_bandlim() - check for ICMP bandwidth limit
994  *
995  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
996  *	hit our bandwidth limit and it is not ok.
997  *
998  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
999  *
1000  *	For now we separate the TCP and UDP subsystems w/ different 'which'
1001  *	values.  We may eventually remove this separation (and simplify the
1002  *	code further).
1003  *
1004  *	Note that the printing of the error message is delayed so we can
1005  *	properly print the icmp error rate that the system was trying to do
1006  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1007  *	the 'final' error, but it doesn't make sense to solve the printing
1008  *	delay with more complex code.
1009  */
1010 struct icmp_rate {
1011 	const char *descr;
1012 	struct counter_rate cr;
1013 };
1014 VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1015 	{ "icmp unreach response" },
1016 	{ "icmp ping response" },
1017 	{ "icmp tstamp response" },
1018 	{ "closed port RST response" },
1019 	{ "open port RST response" },
1020 	{ "icmp6 unreach response" },
1021 	{ "sctp ootb response" }
1022 };
1023 #define	V_icmp_rates	VNET(icmp_rates)
1024 
1025 static void
1026 icmp_bandlimit_init(void)
1027 {
1028 
1029 	for (int i = 0; i < BANDLIM_MAX; i++) {
1030 		V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1031 		V_icmp_rates[i].cr.cr_ticks = ticks;
1032 	}
1033 }
1034 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1035     icmp_bandlimit_init, NULL);
1036 
1037 static void
1038 icmp_bandlimit_uninit(void)
1039 {
1040 
1041 	for (int i = 0; i < BANDLIM_MAX; i++)
1042 		counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1043 }
1044 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1045     icmp_bandlimit_uninit, NULL);
1046 
1047 int
1048 badport_bandlim(int which)
1049 {
1050 	int64_t pps;
1051 
1052 	if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1053 		return (0);
1054 
1055 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1056 	    ("%s: which %d", __func__, which));
1057 
1058 	pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1059 	if (pps == -1)
1060 		return (-1);
1061 	if (pps > 0 && V_icmplim_output)
1062 		log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1063 			V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1064 	return (0);
1065 }
1066