xref: /freebsd/sys/netinet/ip_icmp.c (revision 52f72944b8f5abb2386eae924357dee8aea17d5b)
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 #ifdef INET
73 
74 #include <machine/in_cksum.h>
75 
76 #include <security/mac/mac_framework.h>
77 #endif /* INET */
78 
79 /*
80  * ICMP routines: error generation, receive packet processing, and
81  * routines to turnaround packets back to the originator, and
82  * host table maintenance routines.
83  */
84 static VNET_DEFINE(int, icmplim) = 200;
85 #define	V_icmplim			VNET(icmplim)
86 SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | CTLFLAG_RW,
87 	&VNET_NAME(icmplim), 0,
88 	"Maximum number of ICMP responses per second");
89 
90 static VNET_DEFINE(int, icmplim_output) = 1;
91 #define	V_icmplim_output		VNET(icmplim_output)
92 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
93 	&VNET_NAME(icmplim_output), 0,
94 	"Enable logging of ICMP response rate limiting");
95 
96 #ifdef INET
97 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
98 VNET_PCPUSTAT_SYSINIT(icmpstat);
99 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
100     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
101 
102 #ifdef VIMAGE
103 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
104 #endif /* VIMAGE */
105 
106 static VNET_DEFINE(int, icmpmaskrepl) = 0;
107 #define	V_icmpmaskrepl			VNET(icmpmaskrepl)
108 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
109 	&VNET_NAME(icmpmaskrepl), 0,
110 	"Reply to ICMP Address Mask Request packets");
111 
112 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
113 #define	V_icmpmaskfake			VNET(icmpmaskfake)
114 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
115 	&VNET_NAME(icmpmaskfake), 0,
116 	"Fake reply to ICMP Address Mask Request packets");
117 
118 VNET_DEFINE(int, drop_redirect) = 0;
119 #define	V_drop_redirect			VNET(drop_redirect)
120 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
121 	&VNET_NAME(drop_redirect), 0,
122 	"Ignore ICMP redirects");
123 
124 static VNET_DEFINE(int, log_redirect) = 0;
125 #define	V_log_redirect			VNET(log_redirect)
126 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
127 	&VNET_NAME(log_redirect), 0,
128 	"Log ICMP redirects to the console");
129 
130 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
131 #define	V_reply_src			VNET(reply_src)
132 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
133 	&VNET_NAME(reply_src), IFNAMSIZ,
134 	"ICMP reply source for non-local packets");
135 
136 static VNET_DEFINE(int, icmp_rfi) = 0;
137 #define	V_icmp_rfi			VNET(icmp_rfi)
138 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
139 	&VNET_NAME(icmp_rfi), 0,
140 	"ICMP reply from incoming interface for non-local packets");
141 
142 static VNET_DEFINE(int, icmp_quotelen) = 8;
143 #define	V_icmp_quotelen			VNET(icmp_quotelen)
144 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
145 	&VNET_NAME(icmp_quotelen), 0,
146 	"Number of bytes from original packet to quote in ICMP reply");
147 
148 static VNET_DEFINE(int, icmpbmcastecho) = 0;
149 #define	V_icmpbmcastecho		VNET(icmpbmcastecho)
150 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
151 	&VNET_NAME(icmpbmcastecho), 0,
152 	"Reply to multicast ICMP Echo Request and Timestamp packets");
153 
154 static VNET_DEFINE(int, icmptstamprepl) = 1;
155 #define	V_icmptstamprepl		VNET(icmptstamprepl)
156 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
157 	&VNET_NAME(icmptstamprepl), 0,
158 	"Respond to ICMP Timestamp packets");
159 
160 #ifdef ICMPPRINTFS
161 int	icmpprintfs = 0;
162 #endif
163 
164 static void	icmp_reflect(struct mbuf *);
165 static void	icmp_send(struct mbuf *, struct mbuf *);
166 
167 extern	struct protosw inetsw[];
168 
169 /*
170  * Kernel module interface for updating icmpstat.  The argument is an index
171  * into icmpstat treated as an array of u_long.  While this encodes the
172  * general layout of icmpstat into the caller, it doesn't encode its
173  * location, so that future changes to add, for example, per-CPU stats
174  * support won't cause binary compatibility problems for kernel modules.
175  */
176 void
177 kmod_icmpstat_inc(int statnum)
178 {
179 
180 	counter_u64_add(VNET(icmpstat)[statnum], 1);
181 }
182 
183 /*
184  * Generate an error packet of type error
185  * in response to bad packet ip.
186  */
187 void
188 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
189 {
190 	struct ip *oip, *nip;
191 	struct icmp *icp;
192 	struct mbuf *m;
193 	unsigned icmplen, icmpelen, nlen, oiphlen;
194 
195 	KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
196 	    __func__));
197 
198 	if (type != ICMP_REDIRECT)
199 		ICMPSTAT_INC(icps_error);
200 	/*
201 	 * Don't send error:
202 	 *  if the original packet was encrypted.
203 	 *  if not the first fragment of message.
204 	 *  in response to a multicast or broadcast packet.
205 	 *  if the old packet protocol was an ICMP error message.
206 	 */
207 	if (n->m_flags & M_DECRYPTED)
208 		goto freeit;
209 	if (n->m_flags & (M_BCAST|M_MCAST))
210 		goto freeit;
211 
212 	/* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
213 	if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
214 		goto freeit;
215 	oip = mtod(n, struct ip *);
216 	oiphlen = oip->ip_hl << 2;
217 	if (n->m_len < oiphlen + ICMP_MINLEN)
218 		goto freeit;
219 #ifdef ICMPPRINTFS
220 	if (icmpprintfs)
221 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
222 #endif
223 	if (oip->ip_off & htons(~(IP_MF|IP_DF)))
224 		goto freeit;
225 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
226 	    !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
227 		oiphlen))->icmp_type)) {
228 		ICMPSTAT_INC(icps_oldicmp);
229 		goto freeit;
230 	}
231 	/*
232 	 * Calculate length to quote from original packet and
233 	 * prevent the ICMP mbuf from overflowing.
234 	 * Unfortunately this is non-trivial since ip_forward()
235 	 * sends us truncated packets.
236 	 */
237 	nlen = m_length(n, NULL);
238 	if (oip->ip_p == IPPROTO_TCP) {
239 		struct tcphdr *th;
240 		int tcphlen;
241 
242 		if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
243 		    n->m_next == NULL)
244 			goto stdreply;
245 		if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
246 		    (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
247 			goto freeit;
248 		oip = mtod(n, struct ip *);
249 		th = mtodo(n, oiphlen);
250 		tcphlen = th->th_off << 2;
251 		if (tcphlen < sizeof(struct tcphdr))
252 			goto freeit;
253 		if (ntohs(oip->ip_len) < oiphlen + tcphlen)
254 			goto freeit;
255 		if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
256 			goto stdreply;
257 		if (n->m_len < oiphlen + tcphlen &&
258 		    (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
259 			goto freeit;
260 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
261 		    ntohs(oip->ip_len) - oiphlen));
262 	} else if (oip->ip_p == IPPROTO_SCTP) {
263 		struct sctphdr *sh;
264 		struct sctp_chunkhdr *ch;
265 
266 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
267 			goto stdreply;
268 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
269 		    n->m_next == NULL)
270 			goto stdreply;
271 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
272 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
273 			goto freeit;
274 		oip = mtod(n, struct ip *);
275 		icmpelen = max(sizeof(struct sctphdr),
276 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
277 		sh = mtodo(n, oiphlen);
278 		if (ntohl(sh->v_tag) == 0 &&
279 		    ntohs(oip->ip_len) >= oiphlen +
280 		    sizeof(struct sctphdr) + 8 &&
281 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
282 		     n->m_next != NULL)) {
283 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
284 			    (n = m_pullup(n, oiphlen +
285 			    sizeof(struct sctphdr) + 8)) == NULL)
286 				goto freeit;
287 			oip = mtod(n, struct ip *);
288 			sh = mtodo(n, oiphlen);
289 			ch = (struct sctp_chunkhdr *)(sh + 1);
290 			if (ch->chunk_type == SCTP_INITIATION) {
291 				icmpelen = max(sizeof(struct sctphdr) + 8,
292 				    min(V_icmp_quotelen, ntohs(oip->ip_len) -
293 				    oiphlen));
294 			}
295 		}
296 	} else
297 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
298 		    oiphlen));
299 
300 	icmplen = min(oiphlen + icmpelen, nlen);
301 	if (icmplen < sizeof(struct ip))
302 		goto freeit;
303 
304 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
305 		m = m_gethdr(M_NOWAIT, MT_DATA);
306 	else
307 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
308 	if (m == NULL)
309 		goto freeit;
310 #ifdef MAC
311 	mac_netinet_icmp_reply(n, m);
312 #endif
313 	icmplen = min(icmplen, M_TRAILINGSPACE(m) -
314 	    sizeof(struct ip) - ICMP_MINLEN);
315 	m_align(m, ICMP_MINLEN + icmplen);
316 	m->m_len = ICMP_MINLEN + icmplen;
317 
318 	/* XXX MRT  make the outgoing packet use the same FIB
319 	 * that was associated with the incoming packet
320 	 */
321 	M_SETFIB(m, M_GETFIB(n));
322 	icp = mtod(m, struct icmp *);
323 	ICMPSTAT_INC(icps_outhist[type]);
324 	icp->icmp_type = type;
325 	if (type == ICMP_REDIRECT)
326 		icp->icmp_gwaddr.s_addr = dest;
327 	else {
328 		icp->icmp_void = 0;
329 		/*
330 		 * The following assignments assume an overlay with the
331 		 * just zeroed icmp_void field.
332 		 */
333 		if (type == ICMP_PARAMPROB) {
334 			icp->icmp_pptr = code;
335 			code = 0;
336 		} else if (type == ICMP_UNREACH &&
337 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
338 			icp->icmp_nextmtu = htons(mtu);
339 		}
340 	}
341 	icp->icmp_code = code;
342 
343 	/*
344 	 * Copy the quotation into ICMP message and
345 	 * convert quoted IP header back to network representation.
346 	 */
347 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
348 	nip = &icp->icmp_ip;
349 
350 	/*
351 	 * Set up ICMP message mbuf and copy old IP header (without options
352 	 * in front of ICMP message.
353 	 * If the original mbuf was meant to bypass the firewall, the error
354 	 * reply should bypass as well.
355 	 */
356 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
357 	m->m_data -= sizeof(struct ip);
358 	m->m_len += sizeof(struct ip);
359 	m->m_pkthdr.len = m->m_len;
360 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
361 	nip = mtod(m, struct ip *);
362 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
363 	nip->ip_len = htons(m->m_len);
364 	nip->ip_v = IPVERSION;
365 	nip->ip_hl = 5;
366 	nip->ip_p = IPPROTO_ICMP;
367 	nip->ip_tos = 0;
368 	nip->ip_off = 0;
369 	icmp_reflect(m);
370 
371 freeit:
372 	m_freem(n);
373 }
374 
375 /*
376  * Process a received ICMP message.
377  */
378 int
379 icmp_input(struct mbuf **mp, int *offp, int proto)
380 {
381 	struct icmp *icp;
382 	struct in_ifaddr *ia;
383 	struct mbuf *m = *mp;
384 	struct ip *ip = mtod(m, struct ip *);
385 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
386 	int hlen = *offp;
387 	int icmplen = ntohs(ip->ip_len) - *offp;
388 	int i, code;
389 	void (*ctlfunc)(int, struct sockaddr *, void *);
390 	int fibnum;
391 
392 	*mp = NULL;
393 
394 	/*
395 	 * Locate icmp structure in mbuf, and check
396 	 * that not corrupted and of at least minimum length.
397 	 */
398 #ifdef ICMPPRINTFS
399 	if (icmpprintfs) {
400 		char srcbuf[INET_ADDRSTRLEN];
401 		char dstbuf[INET_ADDRSTRLEN];
402 
403 		printf("icmp_input from %s to %s, len %d\n",
404 		    inet_ntoa_r(ip->ip_src, srcbuf),
405 		    inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
406 	}
407 #endif
408 	if (icmplen < ICMP_MINLEN) {
409 		ICMPSTAT_INC(icps_tooshort);
410 		goto freeit;
411 	}
412 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
413 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
414 		ICMPSTAT_INC(icps_tooshort);
415 		return (IPPROTO_DONE);
416 	}
417 	ip = mtod(m, struct ip *);
418 	m->m_len -= hlen;
419 	m->m_data += hlen;
420 	icp = mtod(m, struct icmp *);
421 	if (in_cksum(m, icmplen)) {
422 		ICMPSTAT_INC(icps_checksum);
423 		goto freeit;
424 	}
425 	m->m_len += hlen;
426 	m->m_data -= hlen;
427 
428 #ifdef ICMPPRINTFS
429 	if (icmpprintfs)
430 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
431 		    icp->icmp_code);
432 #endif
433 
434 	/*
435 	 * Message type specific processing.
436 	 */
437 	if (icp->icmp_type > ICMP_MAXTYPE)
438 		goto raw;
439 
440 	/* Initialize */
441 	bzero(&icmpsrc, sizeof(icmpsrc));
442 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
443 	icmpsrc.sin_family = AF_INET;
444 	bzero(&icmpdst, sizeof(icmpdst));
445 	icmpdst.sin_len = sizeof(struct sockaddr_in);
446 	icmpdst.sin_family = AF_INET;
447 	bzero(&icmpgw, sizeof(icmpgw));
448 	icmpgw.sin_len = sizeof(struct sockaddr_in);
449 	icmpgw.sin_family = AF_INET;
450 
451 	ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
452 	code = icp->icmp_code;
453 	switch (icp->icmp_type) {
454 
455 	case ICMP_UNREACH:
456 		switch (code) {
457 			case ICMP_UNREACH_NET:
458 			case ICMP_UNREACH_HOST:
459 			case ICMP_UNREACH_SRCFAIL:
460 			case ICMP_UNREACH_NET_UNKNOWN:
461 			case ICMP_UNREACH_HOST_UNKNOWN:
462 			case ICMP_UNREACH_ISOLATED:
463 			case ICMP_UNREACH_TOSNET:
464 			case ICMP_UNREACH_TOSHOST:
465 			case ICMP_UNREACH_HOST_PRECEDENCE:
466 			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
467 				code = PRC_UNREACH_NET;
468 				break;
469 
470 			case ICMP_UNREACH_NEEDFRAG:
471 				code = PRC_MSGSIZE;
472 				break;
473 
474 			/*
475 			 * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
476 			 * Treat subcodes 2,3 as immediate RST
477 			 */
478 			case ICMP_UNREACH_PROTOCOL:
479 				code = PRC_UNREACH_PROTOCOL;
480 				break;
481 			case ICMP_UNREACH_PORT:
482 				code = PRC_UNREACH_PORT;
483 				break;
484 
485 			case ICMP_UNREACH_NET_PROHIB:
486 			case ICMP_UNREACH_HOST_PROHIB:
487 			case ICMP_UNREACH_FILTER_PROHIB:
488 				code = PRC_UNREACH_ADMIN_PROHIB;
489 				break;
490 
491 			default:
492 				goto badcode;
493 		}
494 		goto deliver;
495 
496 	case ICMP_TIMXCEED:
497 		if (code > 1)
498 			goto badcode;
499 		code += PRC_TIMXCEED_INTRANS;
500 		goto deliver;
501 
502 	case ICMP_PARAMPROB:
503 		if (code > 1)
504 			goto badcode;
505 		code = PRC_PARAMPROB;
506 	deliver:
507 		/*
508 		 * Problem with datagram; advise higher level routines.
509 		 */
510 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
511 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
512 			ICMPSTAT_INC(icps_badlen);
513 			goto freeit;
514 		}
515 		/* Discard ICMP's in response to multicast packets */
516 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
517 			goto badcode;
518 #ifdef ICMPPRINTFS
519 		if (icmpprintfs)
520 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
521 #endif
522 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
523 		/*
524 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
525 		 * notification to TCP layer.
526 		 */
527 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
528 		ip_stripoptions(m);
529 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
530 			/* This should actually not happen */
531 			ICMPSTAT_INC(icps_tooshort);
532 			return (IPPROTO_DONE);
533 		}
534 		ip = mtod(m, struct ip *);
535 		icp = (struct icmp *)(ip + 1);
536 		/*
537 		 * The upper layer handler can rely on:
538 		 * - The outer IP header has no options.
539 		 * - The outer IP header, the ICMP header, the inner IP header,
540 		 *   and the first n bytes of the inner payload are contiguous.
541 		 *   n is at least 8, but might be larger based on
542 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
543 		 */
544 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
545 		if (ctlfunc)
546 			(*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
547 				   (void *)&icp->icmp_ip);
548 		break;
549 
550 	badcode:
551 		ICMPSTAT_INC(icps_badcode);
552 		break;
553 
554 	case ICMP_ECHO:
555 		if (!V_icmpbmcastecho
556 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
557 			ICMPSTAT_INC(icps_bmcastecho);
558 			break;
559 		}
560 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
561 			goto freeit;
562 		icp->icmp_type = ICMP_ECHOREPLY;
563 		goto reflect;
564 
565 	case ICMP_TSTAMP:
566 		if (V_icmptstamprepl == 0)
567 			break;
568 		if (!V_icmpbmcastecho
569 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
570 			ICMPSTAT_INC(icps_bmcasttstamp);
571 			break;
572 		}
573 		if (icmplen < ICMP_TSLEN) {
574 			ICMPSTAT_INC(icps_badlen);
575 			break;
576 		}
577 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
578 			goto freeit;
579 		icp->icmp_type = ICMP_TSTAMPREPLY;
580 		icp->icmp_rtime = iptime();
581 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
582 		goto reflect;
583 
584 	case ICMP_MASKREQ:
585 		if (V_icmpmaskrepl == 0)
586 			break;
587 		/*
588 		 * We are not able to respond with all ones broadcast
589 		 * unless we receive it over a point-to-point interface.
590 		 */
591 		if (icmplen < ICMP_MASKLEN)
592 			break;
593 		switch (ip->ip_dst.s_addr) {
594 
595 		case INADDR_BROADCAST:
596 		case INADDR_ANY:
597 			icmpdst.sin_addr = ip->ip_src;
598 			break;
599 
600 		default:
601 			icmpdst.sin_addr = ip->ip_dst;
602 		}
603 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
604 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
605 		if (ia == NULL)
606 			break;
607 		if (ia->ia_ifp == NULL) {
608 			ifa_free(&ia->ia_ifa);
609 			break;
610 		}
611 		icp->icmp_type = ICMP_MASKREPLY;
612 		if (V_icmpmaskfake == 0)
613 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
614 		else
615 			icp->icmp_mask = V_icmpmaskfake;
616 		if (ip->ip_src.s_addr == 0) {
617 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
618 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
619 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
620 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
621 		}
622 		ifa_free(&ia->ia_ifa);
623 reflect:
624 		ICMPSTAT_INC(icps_reflect);
625 		ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
626 		icmp_reflect(m);
627 		return (IPPROTO_DONE);
628 
629 	case ICMP_REDIRECT:
630 		if (V_log_redirect) {
631 			u_long src, dst, gw;
632 
633 			src = ntohl(ip->ip_src.s_addr);
634 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
635 			gw = ntohl(icp->icmp_gwaddr.s_addr);
636 			printf("icmp redirect from %d.%d.%d.%d: "
637 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
638 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
639 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
640 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
641 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
642 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
643 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
644 		}
645 		/*
646 		 * RFC1812 says we must ignore ICMP redirects if we
647 		 * are acting as router.
648 		 */
649 		if (V_drop_redirect || V_ipforwarding)
650 			break;
651 		if (code > 3)
652 			goto badcode;
653 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
654 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
655 			ICMPSTAT_INC(icps_badlen);
656 			break;
657 		}
658 		/*
659 		 * Short circuit routing redirects to force
660 		 * immediate change in the kernel's routing
661 		 * tables.  The message is also handed to anyone
662 		 * listening on a raw socket (e.g. the routing
663 		 * daemon for use in updating its tables).
664 		 */
665 		icmpgw.sin_addr = ip->ip_src;
666 		icmpdst.sin_addr = icp->icmp_gwaddr;
667 #ifdef	ICMPPRINTFS
668 		if (icmpprintfs) {
669 			char dstbuf[INET_ADDRSTRLEN];
670 			char gwbuf[INET_ADDRSTRLEN];
671 
672 			printf("redirect dst %s to %s\n",
673 			       inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
674 			       inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
675 		}
676 #endif
677 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
678 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
679 			in_rtredirect((struct sockaddr *)&icmpsrc,
680 			  (struct sockaddr *)&icmpdst,
681 			  (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
682 			  (struct sockaddr *)&icmpgw, fibnum);
683 		}
684 		pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
685 		break;
686 
687 	/*
688 	 * No kernel processing for the following;
689 	 * just fall through to send to raw listener.
690 	 */
691 	case ICMP_ECHOREPLY:
692 	case ICMP_ROUTERADVERT:
693 	case ICMP_ROUTERSOLICIT:
694 	case ICMP_TSTAMPREPLY:
695 	case ICMP_IREQREPLY:
696 	case ICMP_MASKREPLY:
697 	case ICMP_SOURCEQUENCH:
698 	default:
699 		break;
700 	}
701 
702 raw:
703 	*mp = m;
704 	rip_input(mp, offp, proto);
705 	return (IPPROTO_DONE);
706 
707 freeit:
708 	m_freem(m);
709 	return (IPPROTO_DONE);
710 }
711 
712 /*
713  * Reflect the ip packet back to the source
714  */
715 static void
716 icmp_reflect(struct mbuf *m)
717 {
718 	struct rm_priotracker in_ifa_tracker;
719 	struct ip *ip = mtod(m, struct ip *);
720 	struct ifaddr *ifa;
721 	struct ifnet *ifp;
722 	struct in_ifaddr *ia;
723 	struct in_addr t;
724 	struct nhop4_extended nh_ext;
725 	struct mbuf *opts = NULL;
726 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
727 
728 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
729 	    IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
730 	    IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
731 		m_freem(m);	/* Bad return address */
732 		ICMPSTAT_INC(icps_badaddr);
733 		goto done;	/* Ip_output() will check for broadcast */
734 	}
735 
736 	t = ip->ip_dst;
737 	ip->ip_dst = ip->ip_src;
738 
739 	/*
740 	 * Source selection for ICMP replies:
741 	 *
742 	 * If the incoming packet was addressed directly to one of our
743 	 * own addresses, use dst as the src for the reply.
744 	 */
745 	IN_IFADDR_RLOCK(&in_ifa_tracker);
746 	LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
747 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
748 			t = IA_SIN(ia)->sin_addr;
749 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
750 			goto match;
751 		}
752 	}
753 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
754 
755 	/*
756 	 * If the incoming packet was addressed to one of our broadcast
757 	 * addresses, use the first non-broadcast address which corresponds
758 	 * to the incoming interface.
759 	 */
760 	ifp = m->m_pkthdr.rcvif;
761 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
762 		IF_ADDR_RLOCK(ifp);
763 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
764 			if (ifa->ifa_addr->sa_family != AF_INET)
765 				continue;
766 			ia = ifatoia(ifa);
767 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
768 			    t.s_addr) {
769 				t = IA_SIN(ia)->sin_addr;
770 				IF_ADDR_RUNLOCK(ifp);
771 				goto match;
772 			}
773 		}
774 		IF_ADDR_RUNLOCK(ifp);
775 	}
776 	/*
777 	 * If the packet was transiting through us, use the address of
778 	 * the interface the packet came through in.  If that interface
779 	 * doesn't have a suitable IP address, the normal selection
780 	 * criteria apply.
781 	 */
782 	if (V_icmp_rfi && ifp != NULL) {
783 		IF_ADDR_RLOCK(ifp);
784 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
785 			if (ifa->ifa_addr->sa_family != AF_INET)
786 				continue;
787 			ia = ifatoia(ifa);
788 			t = IA_SIN(ia)->sin_addr;
789 			IF_ADDR_RUNLOCK(ifp);
790 			goto match;
791 		}
792 		IF_ADDR_RUNLOCK(ifp);
793 	}
794 	/*
795 	 * If the incoming packet was not addressed directly to us, use
796 	 * designated interface for icmp replies specified by sysctl
797 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
798 	 * with normal source selection.
799 	 */
800 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
801 		IF_ADDR_RLOCK(ifp);
802 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
803 			if (ifa->ifa_addr->sa_family != AF_INET)
804 				continue;
805 			ia = ifatoia(ifa);
806 			t = IA_SIN(ia)->sin_addr;
807 			IF_ADDR_RUNLOCK(ifp);
808 			goto match;
809 		}
810 		IF_ADDR_RUNLOCK(ifp);
811 	}
812 	/*
813 	 * If the packet was transiting through us, use the address of
814 	 * the interface that is the closest to the packet source.
815 	 * When we don't have a route back to the packet source, stop here
816 	 * and drop the packet.
817 	 */
818 	if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
819 		m_freem(m);
820 		ICMPSTAT_INC(icps_noroute);
821 		goto done;
822 	}
823 	t = nh_ext.nh_src;
824 match:
825 #ifdef MAC
826 	mac_netinet_icmp_replyinplace(m);
827 #endif
828 	ip->ip_src = t;
829 	ip->ip_ttl = V_ip_defttl;
830 
831 	if (optlen > 0) {
832 		u_char *cp;
833 		int opt, cnt;
834 		u_int len;
835 
836 		/*
837 		 * Retrieve any source routing from the incoming packet;
838 		 * add on any record-route or timestamp options.
839 		 */
840 		cp = (u_char *) (ip + 1);
841 		if ((opts = ip_srcroute(m)) == NULL &&
842 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
843 			opts->m_len = sizeof(struct in_addr);
844 			mtod(opts, struct in_addr *)->s_addr = 0;
845 		}
846 		if (opts) {
847 #ifdef ICMPPRINTFS
848 		    if (icmpprintfs)
849 			    printf("icmp_reflect optlen %d rt %d => ",
850 				optlen, opts->m_len);
851 #endif
852 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
853 			    opt = cp[IPOPT_OPTVAL];
854 			    if (opt == IPOPT_EOL)
855 				    break;
856 			    if (opt == IPOPT_NOP)
857 				    len = 1;
858 			    else {
859 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
860 					    break;
861 				    len = cp[IPOPT_OLEN];
862 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
863 				        len > cnt)
864 					    break;
865 			    }
866 			    /*
867 			     * Should check for overflow, but it "can't happen"
868 			     */
869 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
870 				opt == IPOPT_SECURITY) {
871 				    bcopy((caddr_t)cp,
872 					mtod(opts, caddr_t) + opts->m_len, len);
873 				    opts->m_len += len;
874 			    }
875 		    }
876 		    /* Terminate & pad, if necessary */
877 		    cnt = opts->m_len % 4;
878 		    if (cnt) {
879 			    for (; cnt < 4; cnt++) {
880 				    *(mtod(opts, caddr_t) + opts->m_len) =
881 					IPOPT_EOL;
882 				    opts->m_len++;
883 			    }
884 		    }
885 #ifdef ICMPPRINTFS
886 		    if (icmpprintfs)
887 			    printf("%d\n", opts->m_len);
888 #endif
889 		}
890 		ip_stripoptions(m);
891 	}
892 	m_tag_delete_nonpersistent(m);
893 	m->m_flags &= ~(M_BCAST|M_MCAST);
894 	icmp_send(m, opts);
895 done:
896 	if (opts)
897 		(void)m_free(opts);
898 }
899 
900 /*
901  * Send an icmp packet back to the ip level,
902  * after supplying a checksum.
903  */
904 static void
905 icmp_send(struct mbuf *m, struct mbuf *opts)
906 {
907 	struct ip *ip = mtod(m, struct ip *);
908 	int hlen;
909 	struct icmp *icp;
910 
911 	hlen = ip->ip_hl << 2;
912 	m->m_data += hlen;
913 	m->m_len -= hlen;
914 	icp = mtod(m, struct icmp *);
915 	icp->icmp_cksum = 0;
916 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
917 	m->m_data -= hlen;
918 	m->m_len += hlen;
919 	m->m_pkthdr.rcvif = (struct ifnet *)0;
920 #ifdef ICMPPRINTFS
921 	if (icmpprintfs) {
922 		char dstbuf[INET_ADDRSTRLEN];
923 		char srcbuf[INET_ADDRSTRLEN];
924 
925 		printf("icmp_send dst %s src %s\n",
926 		    inet_ntoa_r(ip->ip_dst, dstbuf),
927 		    inet_ntoa_r(ip->ip_src, srcbuf));
928 	}
929 #endif
930 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
931 }
932 
933 /*
934  * Return milliseconds since 00:00 UTC in network format.
935  */
936 uint32_t
937 iptime(void)
938 {
939 	struct timeval atv;
940 	u_long t;
941 
942 	getmicrotime(&atv);
943 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
944 	return (htonl(t));
945 }
946 
947 /*
948  * Return the next larger or smaller MTU plateau (table from RFC 1191)
949  * given current value MTU.  If DIR is less than zero, a larger plateau
950  * is returned; otherwise, a smaller value is returned.
951  */
952 int
953 ip_next_mtu(int mtu, int dir)
954 {
955 	static int mtutab[] = {
956 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
957 		296, 68, 0
958 	};
959 	int i, size;
960 
961 	size = (sizeof mtutab) / (sizeof mtutab[0]);
962 	if (dir >= 0) {
963 		for (i = 0; i < size; i++)
964 			if (mtu > mtutab[i])
965 				return mtutab[i];
966 	} else {
967 		for (i = size - 1; i >= 0; i--)
968 			if (mtu < mtutab[i])
969 				return mtutab[i];
970 		if (mtu == mtutab[0])
971 			return mtutab[0];
972 	}
973 	return 0;
974 }
975 #endif /* INET */
976 
977 
978 /*
979  * badport_bandlim() - check for ICMP bandwidth limit
980  *
981  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
982  *	hit our bandwidth limit and it is not ok.
983  *
984  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
985  *
986  *	For now we separate the TCP and UDP subsystems w/ different 'which'
987  *	values.  We may eventually remove this separation (and simplify the
988  *	code further).
989  *
990  *	Note that the printing of the error message is delayed so we can
991  *	properly print the icmp error rate that the system was trying to do
992  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
993  *	the 'final' error, but it doesn't make sense to solve the printing
994  *	delay with more complex code.
995  */
996 struct icmp_rate {
997 	const char *descr;
998 	struct counter_rate cr;
999 };
1000 static VNET_DEFINE(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1001 	{ "icmp unreach response" },
1002 	{ "icmp ping response" },
1003 	{ "icmp tstamp response" },
1004 	{ "closed port RST response" },
1005 	{ "open port RST response" },
1006 	{ "icmp6 unreach response" },
1007 	{ "sctp ootb response" }
1008 };
1009 #define	V_icmp_rates	VNET(icmp_rates)
1010 
1011 static void
1012 icmp_bandlimit_init(void)
1013 {
1014 
1015 	for (int i = 0; i < BANDLIM_MAX; i++) {
1016 		V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1017 		V_icmp_rates[i].cr.cr_ticks = ticks;
1018 	}
1019 }
1020 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1021     icmp_bandlimit_init, NULL);
1022 
1023 static void
1024 icmp_bandlimit_uninit(void)
1025 {
1026 
1027 	for (int i = 0; i < BANDLIM_MAX; i++)
1028 		counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1029 }
1030 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1031     icmp_bandlimit_uninit, NULL);
1032 
1033 int
1034 badport_bandlim(int which)
1035 {
1036 	int64_t pps;
1037 
1038 	if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1039 		return (0);
1040 
1041 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1042 	    ("%s: which %d", __func__, which));
1043 
1044 	pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1045 	if (pps == -1)
1046 		return (-1);
1047 	if (pps > 0 && V_icmplim_output)
1048 		log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1049 			V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1050 	return (0);
1051 }
1052