xref: /freebsd/sys/netinet/ip_icmp.c (revision 7f9dff23d3092aa33ad45b2b63e52469b3c13a6e)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_icmp.c	8.2 (Berkeley) 1/4/94
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mbuf.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/time.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/rmlock.h>
46 #include <sys/sysctl.h>
47 #include <sys/syslog.h>
48 
49 #include <net/if.h>
50 #include <net/if_var.h>
51 #include <net/if_types.h>
52 #include <net/route.h>
53 #include <net/vnet.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/in_fib.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_icmp.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/ip_options.h>
64 #include <netinet/sctp.h>
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_var.h>
67 #include <netinet/tcpip.h>
68 #include <netinet/icmp_var.h>
69 
70 #ifdef INET
71 
72 #include <machine/in_cksum.h>
73 
74 #include <security/mac/mac_framework.h>
75 #endif /* INET */
76 
77 /*
78  * ICMP routines: error generation, receive packet processing, and
79  * routines to turnaround packets back to the originator, and
80  * host table maintenance routines.
81  */
82 static VNET_DEFINE(int, icmplim) = 200;
83 #define	V_icmplim			VNET(icmplim)
84 SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | CTLFLAG_RW,
85 	&VNET_NAME(icmplim), 0,
86 	"Maximum number of ICMP responses per second");
87 
88 static VNET_DEFINE(int, icmplim_output) = 1;
89 #define	V_icmplim_output		VNET(icmplim_output)
90 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
91 	&VNET_NAME(icmplim_output), 0,
92 	"Enable logging of ICMP response rate limiting");
93 
94 #ifdef INET
95 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
96 VNET_PCPUSTAT_SYSINIT(icmpstat);
97 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
98     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
99 
100 #ifdef VIMAGE
101 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
102 #endif /* VIMAGE */
103 
104 static VNET_DEFINE(int, icmpmaskrepl) = 0;
105 #define	V_icmpmaskrepl			VNET(icmpmaskrepl)
106 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
107 	&VNET_NAME(icmpmaskrepl), 0,
108 	"Reply to ICMP Address Mask Request packets");
109 
110 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
111 #define	V_icmpmaskfake			VNET(icmpmaskfake)
112 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
113 	&VNET_NAME(icmpmaskfake), 0,
114 	"Fake reply to ICMP Address Mask Request packets");
115 
116 VNET_DEFINE(int, drop_redirect) = 0;
117 #define	V_drop_redirect			VNET(drop_redirect)
118 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
119 	&VNET_NAME(drop_redirect), 0,
120 	"Ignore ICMP redirects");
121 
122 static VNET_DEFINE(int, log_redirect) = 0;
123 #define	V_log_redirect			VNET(log_redirect)
124 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
125 	&VNET_NAME(log_redirect), 0,
126 	"Log ICMP redirects to the console");
127 
128 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
129 #define	V_reply_src			VNET(reply_src)
130 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
131 	&VNET_NAME(reply_src), IFNAMSIZ,
132 	"ICMP reply source for non-local packets");
133 
134 static VNET_DEFINE(int, icmp_rfi) = 0;
135 #define	V_icmp_rfi			VNET(icmp_rfi)
136 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
137 	&VNET_NAME(icmp_rfi), 0,
138 	"ICMP reply from incoming interface for non-local packets");
139 
140 static VNET_DEFINE(int, icmp_quotelen) = 8;
141 #define	V_icmp_quotelen			VNET(icmp_quotelen)
142 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
143 	&VNET_NAME(icmp_quotelen), 0,
144 	"Number of bytes from original packet to quote in ICMP reply");
145 
146 static VNET_DEFINE(int, icmpbmcastecho) = 0;
147 #define	V_icmpbmcastecho		VNET(icmpbmcastecho)
148 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
149 	&VNET_NAME(icmpbmcastecho), 0,
150 	"Reply to multicast ICMP Echo Request and Timestamp packets");
151 
152 static VNET_DEFINE(int, icmptstamprepl) = 1;
153 #define	V_icmptstamprepl		VNET(icmptstamprepl)
154 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
155 	&VNET_NAME(icmptstamprepl), 0,
156 	"Respond to ICMP Timestamp packets");
157 
158 #ifdef ICMPPRINTFS
159 int	icmpprintfs = 0;
160 #endif
161 
162 static void	icmp_reflect(struct mbuf *);
163 static void	icmp_send(struct mbuf *, struct mbuf *);
164 
165 extern	struct protosw inetsw[];
166 
167 /*
168  * Kernel module interface for updating icmpstat.  The argument is an index
169  * into icmpstat treated as an array of u_long.  While this encodes the
170  * general layout of icmpstat into the caller, it doesn't encode its
171  * location, so that future changes to add, for example, per-CPU stats
172  * support won't cause binary compatibility problems for kernel modules.
173  */
174 void
175 kmod_icmpstat_inc(int statnum)
176 {
177 
178 	counter_u64_add(VNET(icmpstat)[statnum], 1);
179 }
180 
181 /*
182  * Generate an error packet of type error
183  * in response to bad packet ip.
184  */
185 void
186 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
187 {
188 	register struct ip *oip = mtod(n, struct ip *), *nip;
189 	register unsigned oiphlen = oip->ip_hl << 2;
190 	register struct icmp *icp;
191 	register struct mbuf *m;
192 	unsigned icmplen, icmpelen, nlen;
193 
194 	KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
195 #ifdef ICMPPRINTFS
196 	if (icmpprintfs)
197 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
198 #endif
199 	if (type != ICMP_REDIRECT)
200 		ICMPSTAT_INC(icps_error);
201 	/*
202 	 * Don't send error:
203 	 *  if the original packet was encrypted.
204 	 *  if not the first fragment of message.
205 	 *  in response to a multicast or broadcast packet.
206 	 *  if the old packet protocol was an ICMP error message.
207 	 */
208 	if (n->m_flags & M_DECRYPTED)
209 		goto freeit;
210 	if (oip->ip_off & htons(~(IP_MF|IP_DF)))
211 		goto freeit;
212 	if (n->m_flags & (M_BCAST|M_MCAST))
213 		goto freeit;
214 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
215 	  n->m_len >= oiphlen + ICMP_MINLEN &&
216 	  !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
217 		ICMPSTAT_INC(icps_oldicmp);
218 		goto freeit;
219 	}
220 	/* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
221 	if (oiphlen + 8 > n->m_len)
222 		goto freeit;
223 	/*
224 	 * Calculate length to quote from original packet and
225 	 * prevent the ICMP mbuf from overflowing.
226 	 * Unfortunately this is non-trivial since ip_forward()
227 	 * sends us truncated packets.
228 	 */
229 	nlen = m_length(n, NULL);
230 	if (oip->ip_p == IPPROTO_TCP) {
231 		struct tcphdr *th;
232 		int tcphlen;
233 
234 		if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
235 		    n->m_next == NULL)
236 			goto stdreply;
237 		if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
238 		    ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
239 			goto freeit;
240 		th = (struct tcphdr *)((caddr_t)oip + oiphlen);
241 		tcphlen = th->th_off << 2;
242 		if (tcphlen < sizeof(struct tcphdr))
243 			goto freeit;
244 		if (ntohs(oip->ip_len) < oiphlen + tcphlen)
245 			goto freeit;
246 		if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
247 			goto stdreply;
248 		if (n->m_len < oiphlen + tcphlen &&
249 		    ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
250 			goto freeit;
251 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
252 		    ntohs(oip->ip_len) - oiphlen));
253 	} else if (oip->ip_p == IPPROTO_SCTP) {
254 		struct sctphdr *sh;
255 		struct sctp_chunkhdr *ch;
256 
257 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
258 			goto stdreply;
259 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
260 		    n->m_next == NULL)
261 			goto stdreply;
262 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
263 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
264 			goto freeit;
265 		icmpelen = max(sizeof(struct sctphdr),
266 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
267 		sh = (struct sctphdr *)((caddr_t)oip + oiphlen);
268 		if (ntohl(sh->v_tag) == 0 &&
269 		    ntohs(oip->ip_len) >= oiphlen + sizeof(struct sctphdr) + 8 &&
270 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
271 		     n->m_next != NULL)) {
272 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
273 			    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr) + 8)) == NULL)
274 				goto freeit;
275 			ch = (struct sctp_chunkhdr *)(sh + 1);
276 			if (ch->chunk_type == SCTP_INITIATION) {
277 				icmpelen = max(sizeof(struct sctphdr) + 8,
278 				    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
279 			}
280 		}
281 	} else
282 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
283 
284 	icmplen = min(oiphlen + icmpelen, nlen);
285 	if (icmplen < sizeof(struct ip))
286 		goto freeit;
287 
288 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
289 		m = m_gethdr(M_NOWAIT, MT_DATA);
290 	else
291 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
292 	if (m == NULL)
293 		goto freeit;
294 #ifdef MAC
295 	mac_netinet_icmp_reply(n, m);
296 #endif
297 	icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
298 	m_align(m, ICMP_MINLEN + icmplen);
299 	m->m_len = ICMP_MINLEN + icmplen;
300 
301 	/* XXX MRT  make the outgoing packet use the same FIB
302 	 * that was associated with the incoming packet
303 	 */
304 	M_SETFIB(m, M_GETFIB(n));
305 	icp = mtod(m, struct icmp *);
306 	ICMPSTAT_INC(icps_outhist[type]);
307 	icp->icmp_type = type;
308 	if (type == ICMP_REDIRECT)
309 		icp->icmp_gwaddr.s_addr = dest;
310 	else {
311 		icp->icmp_void = 0;
312 		/*
313 		 * The following assignments assume an overlay with the
314 		 * just zeroed icmp_void field.
315 		 */
316 		if (type == ICMP_PARAMPROB) {
317 			icp->icmp_pptr = code;
318 			code = 0;
319 		} else if (type == ICMP_UNREACH &&
320 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
321 			icp->icmp_nextmtu = htons(mtu);
322 		}
323 	}
324 	icp->icmp_code = code;
325 
326 	/*
327 	 * Copy the quotation into ICMP message and
328 	 * convert quoted IP header back to network representation.
329 	 */
330 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
331 	nip = &icp->icmp_ip;
332 
333 	/*
334 	 * Set up ICMP message mbuf and copy old IP header (without options
335 	 * in front of ICMP message.
336 	 * If the original mbuf was meant to bypass the firewall, the error
337 	 * reply should bypass as well.
338 	 */
339 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
340 	m->m_data -= sizeof(struct ip);
341 	m->m_len += sizeof(struct ip);
342 	m->m_pkthdr.len = m->m_len;
343 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
344 	nip = mtod(m, struct ip *);
345 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
346 	nip->ip_len = htons(m->m_len);
347 	nip->ip_v = IPVERSION;
348 	nip->ip_hl = 5;
349 	nip->ip_p = IPPROTO_ICMP;
350 	nip->ip_tos = 0;
351 	nip->ip_off = 0;
352 	icmp_reflect(m);
353 
354 freeit:
355 	m_freem(n);
356 }
357 
358 /*
359  * Process a received ICMP message.
360  */
361 int
362 icmp_input(struct mbuf **mp, int *offp, int proto)
363 {
364 	struct icmp *icp;
365 	struct in_ifaddr *ia;
366 	struct mbuf *m = *mp;
367 	struct ip *ip = mtod(m, struct ip *);
368 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
369 	int hlen = *offp;
370 	int icmplen = ntohs(ip->ip_len) - *offp;
371 	int i, code;
372 	void (*ctlfunc)(int, struct sockaddr *, void *);
373 	int fibnum;
374 
375 	*mp = NULL;
376 
377 	/*
378 	 * Locate icmp structure in mbuf, and check
379 	 * that not corrupted and of at least minimum length.
380 	 */
381 #ifdef ICMPPRINTFS
382 	if (icmpprintfs) {
383 		char buf[4 * sizeof "123"];
384 		strcpy(buf, inet_ntoa(ip->ip_src));
385 		printf("icmp_input from %s to %s, len %d\n",
386 		       buf, inet_ntoa(ip->ip_dst), icmplen);
387 	}
388 #endif
389 	if (icmplen < ICMP_MINLEN) {
390 		ICMPSTAT_INC(icps_tooshort);
391 		goto freeit;
392 	}
393 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
394 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
395 		ICMPSTAT_INC(icps_tooshort);
396 		return (IPPROTO_DONE);
397 	}
398 	ip = mtod(m, struct ip *);
399 	m->m_len -= hlen;
400 	m->m_data += hlen;
401 	icp = mtod(m, struct icmp *);
402 	if (in_cksum(m, icmplen)) {
403 		ICMPSTAT_INC(icps_checksum);
404 		goto freeit;
405 	}
406 	m->m_len += hlen;
407 	m->m_data -= hlen;
408 
409 #ifdef ICMPPRINTFS
410 	if (icmpprintfs)
411 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
412 		    icp->icmp_code);
413 #endif
414 
415 	/*
416 	 * Message type specific processing.
417 	 */
418 	if (icp->icmp_type > ICMP_MAXTYPE)
419 		goto raw;
420 
421 	/* Initialize */
422 	bzero(&icmpsrc, sizeof(icmpsrc));
423 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
424 	icmpsrc.sin_family = AF_INET;
425 	bzero(&icmpdst, sizeof(icmpdst));
426 	icmpdst.sin_len = sizeof(struct sockaddr_in);
427 	icmpdst.sin_family = AF_INET;
428 	bzero(&icmpgw, sizeof(icmpgw));
429 	icmpgw.sin_len = sizeof(struct sockaddr_in);
430 	icmpgw.sin_family = AF_INET;
431 
432 	ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
433 	code = icp->icmp_code;
434 	switch (icp->icmp_type) {
435 
436 	case ICMP_UNREACH:
437 		switch (code) {
438 			case ICMP_UNREACH_NET:
439 			case ICMP_UNREACH_HOST:
440 			case ICMP_UNREACH_SRCFAIL:
441 			case ICMP_UNREACH_NET_UNKNOWN:
442 			case ICMP_UNREACH_HOST_UNKNOWN:
443 			case ICMP_UNREACH_ISOLATED:
444 			case ICMP_UNREACH_TOSNET:
445 			case ICMP_UNREACH_TOSHOST:
446 			case ICMP_UNREACH_HOST_PRECEDENCE:
447 			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
448 				code = PRC_UNREACH_NET;
449 				break;
450 
451 			case ICMP_UNREACH_NEEDFRAG:
452 				code = PRC_MSGSIZE;
453 				break;
454 
455 			/*
456 			 * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
457 			 * Treat subcodes 2,3 as immediate RST
458 			 */
459 			case ICMP_UNREACH_PROTOCOL:
460 				code = PRC_UNREACH_PROTOCOL;
461 				break;
462 			case ICMP_UNREACH_PORT:
463 				code = PRC_UNREACH_PORT;
464 				break;
465 
466 			case ICMP_UNREACH_NET_PROHIB:
467 			case ICMP_UNREACH_HOST_PROHIB:
468 			case ICMP_UNREACH_FILTER_PROHIB:
469 				code = PRC_UNREACH_ADMIN_PROHIB;
470 				break;
471 
472 			default:
473 				goto badcode;
474 		}
475 		goto deliver;
476 
477 	case ICMP_TIMXCEED:
478 		if (code > 1)
479 			goto badcode;
480 		code += PRC_TIMXCEED_INTRANS;
481 		goto deliver;
482 
483 	case ICMP_PARAMPROB:
484 		if (code > 1)
485 			goto badcode;
486 		code = PRC_PARAMPROB;
487 	deliver:
488 		/*
489 		 * Problem with datagram; advise higher level routines.
490 		 */
491 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
492 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
493 			ICMPSTAT_INC(icps_badlen);
494 			goto freeit;
495 		}
496 		/* Discard ICMP's in response to multicast packets */
497 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
498 			goto badcode;
499 #ifdef ICMPPRINTFS
500 		if (icmpprintfs)
501 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
502 #endif
503 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
504 		/*
505 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
506 		 * notification to TCP layer.
507 		 */
508 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
509 		ip_stripoptions(m);
510 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
511 			/* This should actually not happen */
512 			ICMPSTAT_INC(icps_tooshort);
513 			return (IPPROTO_DONE);
514 		}
515 		ip = mtod(m, struct ip *);
516 		icp = (struct icmp *)(ip + 1);
517 		/*
518 		 * The upper layer handler can rely on:
519 		 * - The outer IP header has no options.
520 		 * - The outer IP header, the ICMP header, the inner IP header,
521 		 *   and the first n bytes of the inner payload are contiguous.
522 		 *   n is at least 8, but might be larger based on
523 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
524 		 */
525 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
526 		if (ctlfunc)
527 			(*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
528 				   (void *)&icp->icmp_ip);
529 		break;
530 
531 	badcode:
532 		ICMPSTAT_INC(icps_badcode);
533 		break;
534 
535 	case ICMP_ECHO:
536 		if (!V_icmpbmcastecho
537 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
538 			ICMPSTAT_INC(icps_bmcastecho);
539 			break;
540 		}
541 		icp->icmp_type = ICMP_ECHOREPLY;
542 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
543 			goto freeit;
544 		else
545 			goto reflect;
546 
547 	case ICMP_TSTAMP:
548 		if (V_icmptstamprepl == 0)
549 			break;
550 		if (!V_icmpbmcastecho
551 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
552 			ICMPSTAT_INC(icps_bmcasttstamp);
553 			break;
554 		}
555 		if (icmplen < ICMP_TSLEN) {
556 			ICMPSTAT_INC(icps_badlen);
557 			break;
558 		}
559 		icp->icmp_type = ICMP_TSTAMPREPLY;
560 		icp->icmp_rtime = iptime();
561 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
562 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
563 			goto freeit;
564 		else
565 			goto reflect;
566 
567 	case ICMP_MASKREQ:
568 		if (V_icmpmaskrepl == 0)
569 			break;
570 		/*
571 		 * We are not able to respond with all ones broadcast
572 		 * unless we receive it over a point-to-point interface.
573 		 */
574 		if (icmplen < ICMP_MASKLEN)
575 			break;
576 		switch (ip->ip_dst.s_addr) {
577 
578 		case INADDR_BROADCAST:
579 		case INADDR_ANY:
580 			icmpdst.sin_addr = ip->ip_src;
581 			break;
582 
583 		default:
584 			icmpdst.sin_addr = ip->ip_dst;
585 		}
586 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
587 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
588 		if (ia == NULL)
589 			break;
590 		if (ia->ia_ifp == NULL) {
591 			ifa_free(&ia->ia_ifa);
592 			break;
593 		}
594 		icp->icmp_type = ICMP_MASKREPLY;
595 		if (V_icmpmaskfake == 0)
596 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
597 		else
598 			icp->icmp_mask = V_icmpmaskfake;
599 		if (ip->ip_src.s_addr == 0) {
600 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
601 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
602 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
603 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
604 		}
605 		ifa_free(&ia->ia_ifa);
606 reflect:
607 		ICMPSTAT_INC(icps_reflect);
608 		ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
609 		icmp_reflect(m);
610 		return (IPPROTO_DONE);
611 
612 	case ICMP_REDIRECT:
613 		if (V_log_redirect) {
614 			u_long src, dst, gw;
615 
616 			src = ntohl(ip->ip_src.s_addr);
617 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
618 			gw = ntohl(icp->icmp_gwaddr.s_addr);
619 			printf("icmp redirect from %d.%d.%d.%d: "
620 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
621 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
622 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
623 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
624 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
625 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
626 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
627 		}
628 		/*
629 		 * RFC1812 says we must ignore ICMP redirects if we
630 		 * are acting as router.
631 		 */
632 		if (V_drop_redirect || V_ipforwarding)
633 			break;
634 		if (code > 3)
635 			goto badcode;
636 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
637 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
638 			ICMPSTAT_INC(icps_badlen);
639 			break;
640 		}
641 		/*
642 		 * Short circuit routing redirects to force
643 		 * immediate change in the kernel's routing
644 		 * tables.  The message is also handed to anyone
645 		 * listening on a raw socket (e.g. the routing
646 		 * daemon for use in updating its tables).
647 		 */
648 		icmpgw.sin_addr = ip->ip_src;
649 		icmpdst.sin_addr = icp->icmp_gwaddr;
650 #ifdef	ICMPPRINTFS
651 		if (icmpprintfs) {
652 			char buf[4 * sizeof "123"];
653 			strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
654 
655 			printf("redirect dst %s to %s\n",
656 			       buf, inet_ntoa(icp->icmp_gwaddr));
657 		}
658 #endif
659 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
660 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
661 			in_rtredirect((struct sockaddr *)&icmpsrc,
662 			  (struct sockaddr *)&icmpdst,
663 			  (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
664 			  (struct sockaddr *)&icmpgw, fibnum);
665 		}
666 		pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
667 		break;
668 
669 	/*
670 	 * No kernel processing for the following;
671 	 * just fall through to send to raw listener.
672 	 */
673 	case ICMP_ECHOREPLY:
674 	case ICMP_ROUTERADVERT:
675 	case ICMP_ROUTERSOLICIT:
676 	case ICMP_TSTAMPREPLY:
677 	case ICMP_IREQREPLY:
678 	case ICMP_MASKREPLY:
679 	case ICMP_SOURCEQUENCH:
680 	default:
681 		break;
682 	}
683 
684 raw:
685 	*mp = m;
686 	rip_input(mp, offp, proto);
687 	return (IPPROTO_DONE);
688 
689 freeit:
690 	m_freem(m);
691 	return (IPPROTO_DONE);
692 }
693 
694 /*
695  * Reflect the ip packet back to the source
696  */
697 static void
698 icmp_reflect(struct mbuf *m)
699 {
700 	struct rm_priotracker in_ifa_tracker;
701 	struct ip *ip = mtod(m, struct ip *);
702 	struct ifaddr *ifa;
703 	struct ifnet *ifp;
704 	struct in_ifaddr *ia;
705 	struct in_addr t;
706 	struct nhop4_extended nh_ext;
707 	struct mbuf *opts = NULL;
708 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
709 
710 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
711 	    IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
712 	    IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
713 		m_freem(m);	/* Bad return address */
714 		ICMPSTAT_INC(icps_badaddr);
715 		goto done;	/* Ip_output() will check for broadcast */
716 	}
717 
718 	t = ip->ip_dst;
719 	ip->ip_dst = ip->ip_src;
720 
721 	/*
722 	 * Source selection for ICMP replies:
723 	 *
724 	 * If the incoming packet was addressed directly to one of our
725 	 * own addresses, use dst as the src for the reply.
726 	 */
727 	IN_IFADDR_RLOCK(&in_ifa_tracker);
728 	LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
729 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
730 			t = IA_SIN(ia)->sin_addr;
731 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
732 			goto match;
733 		}
734 	}
735 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
736 
737 	/*
738 	 * If the incoming packet was addressed to one of our broadcast
739 	 * addresses, use the first non-broadcast address which corresponds
740 	 * to the incoming interface.
741 	 */
742 	ifp = m->m_pkthdr.rcvif;
743 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
744 		IF_ADDR_RLOCK(ifp);
745 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
746 			if (ifa->ifa_addr->sa_family != AF_INET)
747 				continue;
748 			ia = ifatoia(ifa);
749 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
750 			    t.s_addr) {
751 				t = IA_SIN(ia)->sin_addr;
752 				IF_ADDR_RUNLOCK(ifp);
753 				goto match;
754 			}
755 		}
756 		IF_ADDR_RUNLOCK(ifp);
757 	}
758 	/*
759 	 * If the packet was transiting through us, use the address of
760 	 * the interface the packet came through in.  If that interface
761 	 * doesn't have a suitable IP address, the normal selection
762 	 * criteria apply.
763 	 */
764 	if (V_icmp_rfi && ifp != NULL) {
765 		IF_ADDR_RLOCK(ifp);
766 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
767 			if (ifa->ifa_addr->sa_family != AF_INET)
768 				continue;
769 			ia = ifatoia(ifa);
770 			t = IA_SIN(ia)->sin_addr;
771 			IF_ADDR_RUNLOCK(ifp);
772 			goto match;
773 		}
774 		IF_ADDR_RUNLOCK(ifp);
775 	}
776 	/*
777 	 * If the incoming packet was not addressed directly to us, use
778 	 * designated interface for icmp replies specified by sysctl
779 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
780 	 * with normal source selection.
781 	 */
782 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
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 packet was transiting through us, use the address of
796 	 * the interface that is the closest to the packet source.
797 	 * When we don't have a route back to the packet source, stop here
798 	 * and drop the packet.
799 	 */
800 	if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
801 		m_freem(m);
802 		ICMPSTAT_INC(icps_noroute);
803 		goto done;
804 	}
805 	t = nh_ext.nh_src;
806 match:
807 #ifdef MAC
808 	mac_netinet_icmp_replyinplace(m);
809 #endif
810 	ip->ip_src = t;
811 	ip->ip_ttl = V_ip_defttl;
812 
813 	if (optlen > 0) {
814 		register u_char *cp;
815 		int opt, cnt;
816 		u_int len;
817 
818 		/*
819 		 * Retrieve any source routing from the incoming packet;
820 		 * add on any record-route or timestamp options.
821 		 */
822 		cp = (u_char *) (ip + 1);
823 		if ((opts = ip_srcroute(m)) == NULL &&
824 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
825 			opts->m_len = sizeof(struct in_addr);
826 			mtod(opts, struct in_addr *)->s_addr = 0;
827 		}
828 		if (opts) {
829 #ifdef ICMPPRINTFS
830 		    if (icmpprintfs)
831 			    printf("icmp_reflect optlen %d rt %d => ",
832 				optlen, opts->m_len);
833 #endif
834 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
835 			    opt = cp[IPOPT_OPTVAL];
836 			    if (opt == IPOPT_EOL)
837 				    break;
838 			    if (opt == IPOPT_NOP)
839 				    len = 1;
840 			    else {
841 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
842 					    break;
843 				    len = cp[IPOPT_OLEN];
844 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
845 				        len > cnt)
846 					    break;
847 			    }
848 			    /*
849 			     * Should check for overflow, but it "can't happen"
850 			     */
851 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
852 				opt == IPOPT_SECURITY) {
853 				    bcopy((caddr_t)cp,
854 					mtod(opts, caddr_t) + opts->m_len, len);
855 				    opts->m_len += len;
856 			    }
857 		    }
858 		    /* Terminate & pad, if necessary */
859 		    cnt = opts->m_len % 4;
860 		    if (cnt) {
861 			    for (; cnt < 4; cnt++) {
862 				    *(mtod(opts, caddr_t) + opts->m_len) =
863 					IPOPT_EOL;
864 				    opts->m_len++;
865 			    }
866 		    }
867 #ifdef ICMPPRINTFS
868 		    if (icmpprintfs)
869 			    printf("%d\n", opts->m_len);
870 #endif
871 		}
872 		ip_stripoptions(m);
873 	}
874 	m_tag_delete_nonpersistent(m);
875 	m->m_flags &= ~(M_BCAST|M_MCAST);
876 	icmp_send(m, opts);
877 done:
878 	if (opts)
879 		(void)m_free(opts);
880 }
881 
882 /*
883  * Send an icmp packet back to the ip level,
884  * after supplying a checksum.
885  */
886 static void
887 icmp_send(struct mbuf *m, struct mbuf *opts)
888 {
889 	register struct ip *ip = mtod(m, struct ip *);
890 	register int hlen;
891 	register struct icmp *icp;
892 
893 	hlen = ip->ip_hl << 2;
894 	m->m_data += hlen;
895 	m->m_len -= hlen;
896 	icp = mtod(m, struct icmp *);
897 	icp->icmp_cksum = 0;
898 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
899 	m->m_data -= hlen;
900 	m->m_len += hlen;
901 	m->m_pkthdr.rcvif = (struct ifnet *)0;
902 #ifdef ICMPPRINTFS
903 	if (icmpprintfs) {
904 		char buf[4 * sizeof "123"];
905 		strcpy(buf, inet_ntoa(ip->ip_dst));
906 		printf("icmp_send dst %s src %s\n",
907 		       buf, inet_ntoa(ip->ip_src));
908 	}
909 #endif
910 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
911 }
912 
913 /*
914  * Return milliseconds since 00:00 UTC in network format.
915  */
916 uint32_t
917 iptime(void)
918 {
919 	struct timeval atv;
920 	u_long t;
921 
922 	getmicrotime(&atv);
923 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
924 	return (htonl(t));
925 }
926 
927 /*
928  * Return the next larger or smaller MTU plateau (table from RFC 1191)
929  * given current value MTU.  If DIR is less than zero, a larger plateau
930  * is returned; otherwise, a smaller value is returned.
931  */
932 int
933 ip_next_mtu(int mtu, int dir)
934 {
935 	static int mtutab[] = {
936 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
937 		296, 68, 0
938 	};
939 	int i, size;
940 
941 	size = (sizeof mtutab) / (sizeof mtutab[0]);
942 	if (dir >= 0) {
943 		for (i = 0; i < size; i++)
944 			if (mtu > mtutab[i])
945 				return mtutab[i];
946 	} else {
947 		for (i = size - 1; i >= 0; i--)
948 			if (mtu < mtutab[i])
949 				return mtutab[i];
950 		if (mtu == mtutab[0])
951 			return mtutab[0];
952 	}
953 	return 0;
954 }
955 #endif /* INET */
956 
957 
958 /*
959  * badport_bandlim() - check for ICMP bandwidth limit
960  *
961  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
962  *	hit our bandwidth limit and it is not ok.
963  *
964  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
965  *
966  *	For now we separate the TCP and UDP subsystems w/ different 'which'
967  *	values.  We may eventually remove this separation (and simplify the
968  *	code further).
969  *
970  *	Note that the printing of the error message is delayed so we can
971  *	properly print the icmp error rate that the system was trying to do
972  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
973  *	the 'final' error, but it doesn't make sense to solve the printing
974  *	delay with more complex code.
975  */
976 struct icmp_rate {
977 	const char *descr;
978 	struct counter_rate cr;
979 };
980 static VNET_DEFINE(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
981 	{ "icmp unreach response" },
982 	{ "icmp ping response" },
983 	{ "icmp tstamp response" },
984 	{ "closed port RST response" },
985 	{ "open port RST response" },
986 	{ "icmp6 unreach response" },
987 	{ "sctp ootb response" }
988 };
989 #define	V_icmp_rates	VNET(icmp_rates)
990 
991 static void
992 icmp_bandlimit_init(void)
993 {
994 
995 	for (int i = 0; i < BANDLIM_MAX; i++) {
996 		V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
997 		V_icmp_rates[i].cr.cr_ticks = ticks;
998 	}
999 }
1000 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1001     icmp_bandlimit_init, NULL);
1002 
1003 static void
1004 icmp_bandlimit_uninit(void)
1005 {
1006 
1007 	for (int i = 0; i < BANDLIM_MAX; i++)
1008 		counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1009 }
1010 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1011     icmp_bandlimit_uninit, NULL);
1012 
1013 int
1014 badport_bandlim(int which)
1015 {
1016 	int64_t pps;
1017 
1018 	if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1019 		return (0);
1020 
1021 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1022 	    ("%s: which %d", __func__, which));
1023 
1024 	pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1025 	if (pps == -1)
1026 		return (-1);
1027 	if (pps > 0 && V_icmplim_output)
1028 		log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1029 			V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1030 	return (0);
1031 }
1032