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: stable/12/sys/netinet/ip_icmp.c 371883 2022-04-07 13:15:58Z git2svn $");
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_curr_jitter) = 0;
92 #define V_icmplim_curr_jitter		VNET(icmplim_curr_jitter)
93 VNET_DEFINE_STATIC(int, icmplim_jitter) = 16;
94 #define	V_icmplim_jitter		VNET(icmplim_jitter)
95 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_jitter, CTLFLAG_VNET | CTLFLAG_RW,
96 	&VNET_NAME(icmplim_jitter), 0,
97 	"Random icmplim jitter adjustment limit");
98 
99 VNET_DEFINE_STATIC(int, icmplim_output) = 1;
100 #define	V_icmplim_output		VNET(icmplim_output)
101 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
102 	&VNET_NAME(icmplim_output), 0,
103 	"Enable logging of ICMP response rate limiting");
104 
105 #ifdef INET
106 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
107 VNET_PCPUSTAT_SYSINIT(icmpstat);
108 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
109     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
110 
111 #ifdef VIMAGE
112 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
113 #endif /* VIMAGE */
114 
115 VNET_DEFINE_STATIC(int, icmpmaskrepl) = 0;
116 #define	V_icmpmaskrepl			VNET(icmpmaskrepl)
117 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
118 	&VNET_NAME(icmpmaskrepl), 0,
119 	"Reply to ICMP Address Mask Request packets");
120 
121 VNET_DEFINE_STATIC(u_int, icmpmaskfake) = 0;
122 #define	V_icmpmaskfake			VNET(icmpmaskfake)
123 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
124 	&VNET_NAME(icmpmaskfake), 0,
125 	"Fake reply to ICMP Address Mask Request packets");
126 
127 VNET_DEFINE(int, drop_redirect) = 0;
128 #define	V_drop_redirect			VNET(drop_redirect)
129 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
130 	&VNET_NAME(drop_redirect), 0,
131 	"Ignore ICMP redirects");
132 
133 VNET_DEFINE_STATIC(int, log_redirect) = 0;
134 #define	V_log_redirect			VNET(log_redirect)
135 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
136 	&VNET_NAME(log_redirect), 0,
137 	"Log ICMP redirects to the console");
138 
139 VNET_DEFINE_STATIC(char, reply_src[IFNAMSIZ]);
140 #define	V_reply_src			VNET(reply_src)
141 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
142 	&VNET_NAME(reply_src), IFNAMSIZ,
143 	"ICMP reply source for non-local packets");
144 
145 VNET_DEFINE_STATIC(int, icmp_rfi) = 0;
146 #define	V_icmp_rfi			VNET(icmp_rfi)
147 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
148 	&VNET_NAME(icmp_rfi), 0,
149 	"ICMP reply from incoming interface for non-local packets");
150 /* Router requirements RFC 1812 section 4.3.2.3 requires 576 - 28. */
151 VNET_DEFINE_STATIC(int, icmp_quotelen) = 548;
152 #define	V_icmp_quotelen			VNET(icmp_quotelen)
153 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
154 	&VNET_NAME(icmp_quotelen), 0,
155 	"Number of bytes from original packet to quote in ICMP reply");
156 
157 VNET_DEFINE_STATIC(int, icmpbmcastecho) = 0;
158 #define	V_icmpbmcastecho		VNET(icmpbmcastecho)
159 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
160 	&VNET_NAME(icmpbmcastecho), 0,
161 	"Reply to multicast ICMP Echo Request and Timestamp packets");
162 
163 VNET_DEFINE_STATIC(int, icmptstamprepl) = 1;
164 #define	V_icmptstamprepl		VNET(icmptstamprepl)
165 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_VNET | CTLFLAG_RW,
166 	&VNET_NAME(icmptstamprepl), 0,
167 	"Respond to ICMP Timestamp packets");
168 
169 VNET_DEFINE_STATIC(int, error_keeptags) = 0;
170 #define	V_error_keeptags		VNET(error_keeptags)
171 SYSCTL_INT(_net_inet_icmp, OID_AUTO, error_keeptags, CTLFLAG_VNET | CTLFLAG_RW,
172 	&VNET_NAME(error_keeptags), 0,
173 	"ICMP error response keeps copy of mbuf_tags of original packet");
174 
175 #ifdef ICMPPRINTFS
176 int	icmpprintfs = 0;
177 #endif
178 
179 static void	icmp_reflect(struct mbuf *);
180 static void	icmp_send(struct mbuf *, struct mbuf *);
181 
182 extern	struct protosw inetsw[];
183 
184 /*
185  * Kernel module interface for updating icmpstat.  The argument is an index
186  * into icmpstat treated as an array of u_long.  While this encodes the
187  * general layout of icmpstat into the caller, it doesn't encode its
188  * location, so that future changes to add, for example, per-CPU stats
189  * support won't cause binary compatibility problems for kernel modules.
190  */
191 void
kmod_icmpstat_inc(int statnum)192 kmod_icmpstat_inc(int statnum)
193 {
194 
195 	counter_u64_add(VNET(icmpstat)[statnum], 1);
196 }
197 
198 /*
199  * Generate an error packet of type error
200  * in response to bad packet ip.
201  */
202 void
icmp_error(struct mbuf * n,int type,int code,uint32_t dest,int mtu)203 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
204 {
205 	struct ip *oip, *nip;
206 	struct icmp *icp;
207 	struct mbuf *m;
208 	unsigned icmplen, icmpelen, nlen, oiphlen;
209 
210 	KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
211 	    __func__));
212 
213 	if (type != ICMP_REDIRECT)
214 		ICMPSTAT_INC(icps_error);
215 	/*
216 	 * Don't send error:
217 	 *  if the original packet was encrypted.
218 	 *  if not the first fragment of message.
219 	 *  in response to a multicast or broadcast packet.
220 	 *  if the old packet protocol was an ICMP error message.
221 	 */
222 	if (n->m_flags & M_DECRYPTED)
223 		goto freeit;
224 	if (n->m_flags & (M_BCAST|M_MCAST))
225 		goto freeit;
226 
227 	/* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
228 	if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
229 		goto freeit;
230 	oip = mtod(n, struct ip *);
231 	oiphlen = oip->ip_hl << 2;
232 	if (n->m_len < oiphlen + ICMP_MINLEN)
233 		goto freeit;
234 #ifdef ICMPPRINTFS
235 	if (icmpprintfs)
236 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
237 #endif
238 	if (oip->ip_off & htons(~(IP_MF|IP_DF)))
239 		goto freeit;
240 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
241 	    !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
242 		oiphlen))->icmp_type)) {
243 		ICMPSTAT_INC(icps_oldicmp);
244 		goto freeit;
245 	}
246 	/*
247 	 * Calculate length to quote from original packet and
248 	 * prevent the ICMP mbuf from overflowing.
249 	 * Unfortunately this is non-trivial since ip_forward()
250 	 * sends us truncated packets.
251 	 */
252 	nlen = m_length(n, NULL);
253 	if (oip->ip_p == IPPROTO_TCP) {
254 		struct tcphdr *th;
255 		int tcphlen;
256 
257 		if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
258 		    n->m_next == NULL)
259 			goto stdreply;
260 		if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
261 		    (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
262 			goto freeit;
263 		oip = mtod(n, struct ip *);
264 		th = mtodo(n, oiphlen);
265 		tcphlen = th->th_off << 2;
266 		if (tcphlen < sizeof(struct tcphdr))
267 			goto freeit;
268 		if (ntohs(oip->ip_len) < oiphlen + tcphlen)
269 			goto freeit;
270 		if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
271 			goto stdreply;
272 		if (n->m_len < oiphlen + tcphlen &&
273 		    (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
274 			goto freeit;
275 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
276 		    ntohs(oip->ip_len) - oiphlen));
277 	} else if (oip->ip_p == IPPROTO_SCTP) {
278 		struct sctphdr *sh;
279 		struct sctp_chunkhdr *ch;
280 
281 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
282 			goto stdreply;
283 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
284 		    n->m_next == NULL)
285 			goto stdreply;
286 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
287 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
288 			goto freeit;
289 		oip = mtod(n, struct ip *);
290 		icmpelen = max(sizeof(struct sctphdr),
291 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
292 		sh = mtodo(n, oiphlen);
293 		if (ntohl(sh->v_tag) == 0 &&
294 		    ntohs(oip->ip_len) >= oiphlen +
295 		    sizeof(struct sctphdr) + 8 &&
296 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
297 		     n->m_next != NULL)) {
298 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
299 			    (n = m_pullup(n, oiphlen +
300 			    sizeof(struct sctphdr) + 8)) == NULL)
301 				goto freeit;
302 			oip = mtod(n, struct ip *);
303 			sh = mtodo(n, oiphlen);
304 			ch = (struct sctp_chunkhdr *)(sh + 1);
305 			if (ch->chunk_type == SCTP_INITIATION) {
306 				icmpelen = max(sizeof(struct sctphdr) + 8,
307 				    min(V_icmp_quotelen, ntohs(oip->ip_len) -
308 				    oiphlen));
309 			}
310 		}
311 	} else
312 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
313 		    oiphlen));
314 
315 	icmplen = min(oiphlen + icmpelen, nlen);
316 	if (icmplen < sizeof(struct ip))
317 		goto freeit;
318 
319 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
320 		m = m_gethdr(M_NOWAIT, MT_DATA);
321 	else
322 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
323 	if (m == NULL)
324 		goto freeit;
325 #ifdef MAC
326 	mac_netinet_icmp_reply(n, m);
327 #endif
328 	icmplen = min(icmplen, M_TRAILINGSPACE(m) -
329 	    sizeof(struct ip) - ICMP_MINLEN);
330 	m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
331 	m->m_data += sizeof(struct ip);
332 	m->m_len = ICMP_MINLEN + icmplen;
333 
334 	/* XXX MRT  make the outgoing packet use the same FIB
335 	 * that was associated with the incoming packet
336 	 */
337 	M_SETFIB(m, M_GETFIB(n));
338 	icp = mtod(m, struct icmp *);
339 	ICMPSTAT_INC(icps_outhist[type]);
340 	icp->icmp_type = type;
341 	if (type == ICMP_REDIRECT)
342 		icp->icmp_gwaddr.s_addr = dest;
343 	else {
344 		icp->icmp_void = 0;
345 		/*
346 		 * The following assignments assume an overlay with the
347 		 * just zeroed icmp_void field.
348 		 */
349 		if (type == ICMP_PARAMPROB) {
350 			icp->icmp_pptr = code;
351 			code = 0;
352 		} else if (type == ICMP_UNREACH &&
353 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
354 			icp->icmp_nextmtu = htons(mtu);
355 		}
356 	}
357 	icp->icmp_code = code;
358 
359 	/*
360 	 * Copy the quotation into ICMP message and
361 	 * convert quoted IP header back to network representation.
362 	 */
363 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
364 	nip = &icp->icmp_ip;
365 
366 	/*
367 	 * Set up ICMP message mbuf and copy old IP header (without options
368 	 * in front of ICMP message.
369 	 * If the original mbuf was meant to bypass the firewall, the error
370 	 * reply should bypass as well.
371 	 */
372 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
373 	KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
374 	    ("insufficient space for ip header"));
375 	m->m_data -= sizeof(struct ip);
376 	m->m_len += sizeof(struct ip);
377 	m->m_pkthdr.len = m->m_len;
378 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
379 	nip = mtod(m, struct ip *);
380 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
381 	nip->ip_len = htons(m->m_len);
382 	nip->ip_v = IPVERSION;
383 	nip->ip_hl = 5;
384 	nip->ip_p = IPPROTO_ICMP;
385 	nip->ip_tos = 0;
386 	nip->ip_off = 0;
387 
388 	if (V_error_keeptags)
389 		m_tag_copy_chain(m, n, M_NOWAIT);
390 
391 	icmp_reflect(m);
392 
393 freeit:
394 	m_freem(n);
395 }
396 
397 /*
398  * Process a received ICMP message.
399  */
400 int
icmp_input(struct mbuf ** mp,int * offp,int proto)401 icmp_input(struct mbuf **mp, int *offp, int proto)
402 {
403 	struct icmp *icp;
404 	struct in_ifaddr *ia;
405 	struct mbuf *m = *mp;
406 	struct ip *ip = mtod(m, struct ip *);
407 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
408 	int hlen = *offp;
409 	int icmplen = ntohs(ip->ip_len) - *offp;
410 	int i, code;
411 	void (*ctlfunc)(int, struct sockaddr *, void *);
412 	int fibnum;
413 
414 	*mp = NULL;
415 
416 	/*
417 	 * Locate icmp structure in mbuf, and check
418 	 * that not corrupted and of at least minimum length.
419 	 */
420 #ifdef ICMPPRINTFS
421 	if (icmpprintfs) {
422 		char srcbuf[INET_ADDRSTRLEN];
423 		char dstbuf[INET_ADDRSTRLEN];
424 
425 		printf("icmp_input from %s to %s, len %d\n",
426 		    inet_ntoa_r(ip->ip_src, srcbuf),
427 		    inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
428 	}
429 #endif
430 	NET_EPOCH_ENTER();
431 	if (icmplen < ICMP_MINLEN) {
432 		ICMPSTAT_INC(icps_tooshort);
433 		goto freeit;
434 	}
435 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
436 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
437 		ICMPSTAT_INC(icps_tooshort);
438 		NET_EPOCH_EXIT();
439 		return (IPPROTO_DONE);
440 	}
441 	ip = mtod(m, struct ip *);
442 	m->m_len -= hlen;
443 	m->m_data += hlen;
444 	icp = mtod(m, struct icmp *);
445 	if (in_cksum(m, icmplen)) {
446 		ICMPSTAT_INC(icps_checksum);
447 		goto freeit;
448 	}
449 	m->m_len += hlen;
450 	m->m_data -= hlen;
451 
452 #ifdef ICMPPRINTFS
453 	if (icmpprintfs)
454 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
455 		    icp->icmp_code);
456 #endif
457 
458 	/*
459 	 * Message type specific processing.
460 	 */
461 	if (icp->icmp_type > ICMP_MAXTYPE)
462 		goto raw;
463 
464 	/* Initialize */
465 	bzero(&icmpsrc, sizeof(icmpsrc));
466 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
467 	icmpsrc.sin_family = AF_INET;
468 	bzero(&icmpdst, sizeof(icmpdst));
469 	icmpdst.sin_len = sizeof(struct sockaddr_in);
470 	icmpdst.sin_family = AF_INET;
471 	bzero(&icmpgw, sizeof(icmpgw));
472 	icmpgw.sin_len = sizeof(struct sockaddr_in);
473 	icmpgw.sin_family = AF_INET;
474 
475 	ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
476 	code = icp->icmp_code;
477 	switch (icp->icmp_type) {
478 
479 	case ICMP_UNREACH:
480 		switch (code) {
481 			case ICMP_UNREACH_NET:
482 			case ICMP_UNREACH_HOST:
483 			case ICMP_UNREACH_SRCFAIL:
484 			case ICMP_UNREACH_NET_UNKNOWN:
485 			case ICMP_UNREACH_HOST_UNKNOWN:
486 			case ICMP_UNREACH_ISOLATED:
487 			case ICMP_UNREACH_TOSNET:
488 			case ICMP_UNREACH_TOSHOST:
489 			case ICMP_UNREACH_HOST_PRECEDENCE:
490 			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
491 				code = PRC_UNREACH_NET;
492 				break;
493 
494 			case ICMP_UNREACH_NEEDFRAG:
495 				code = PRC_MSGSIZE;
496 				break;
497 
498 			/*
499 			 * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
500 			 * Treat subcodes 2,3 as immediate RST
501 			 */
502 			case ICMP_UNREACH_PROTOCOL:
503 				code = PRC_UNREACH_PROTOCOL;
504 				break;
505 			case ICMP_UNREACH_PORT:
506 				code = PRC_UNREACH_PORT;
507 				break;
508 
509 			case ICMP_UNREACH_NET_PROHIB:
510 			case ICMP_UNREACH_HOST_PROHIB:
511 			case ICMP_UNREACH_FILTER_PROHIB:
512 				code = PRC_UNREACH_ADMIN_PROHIB;
513 				break;
514 
515 			default:
516 				goto badcode;
517 		}
518 		goto deliver;
519 
520 	case ICMP_TIMXCEED:
521 		if (code > 1)
522 			goto badcode;
523 		code += PRC_TIMXCEED_INTRANS;
524 		goto deliver;
525 
526 	case ICMP_PARAMPROB:
527 		if (code > 1)
528 			goto badcode;
529 		code = PRC_PARAMPROB;
530 	deliver:
531 		/*
532 		 * Problem with datagram; advise higher level routines.
533 		 */
534 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
535 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
536 			ICMPSTAT_INC(icps_badlen);
537 			goto freeit;
538 		}
539 		/* Discard ICMP's in response to multicast packets */
540 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
541 			goto badcode;
542 #ifdef ICMPPRINTFS
543 		if (icmpprintfs)
544 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
545 #endif
546 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
547 		/*
548 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
549 		 * notification to TCP layer.
550 		 */
551 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
552 		ip_stripoptions(m);
553 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
554 			/* This should actually not happen */
555 			ICMPSTAT_INC(icps_tooshort);
556 			NET_EPOCH_EXIT();
557 			return (IPPROTO_DONE);
558 		}
559 		ip = mtod(m, struct ip *);
560 		icp = (struct icmp *)(ip + 1);
561 		/*
562 		 * The upper layer handler can rely on:
563 		 * - The outer IP header has no options.
564 		 * - The outer IP header, the ICMP header, the inner IP header,
565 		 *   and the first n bytes of the inner payload are contiguous.
566 		 *   n is at least 8, but might be larger based on
567 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
568 		 */
569 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
570 		if (ctlfunc)
571 			(*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
572 				   (void *)&icp->icmp_ip);
573 		break;
574 
575 	badcode:
576 		ICMPSTAT_INC(icps_badcode);
577 		break;
578 
579 	case ICMP_ECHO:
580 		if (!V_icmpbmcastecho
581 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
582 			ICMPSTAT_INC(icps_bmcastecho);
583 			break;
584 		}
585 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
586 			goto freeit;
587 		icp->icmp_type = ICMP_ECHOREPLY;
588 		goto reflect;
589 
590 	case ICMP_TSTAMP:
591 		if (V_icmptstamprepl == 0)
592 			break;
593 		if (!V_icmpbmcastecho
594 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
595 			ICMPSTAT_INC(icps_bmcasttstamp);
596 			break;
597 		}
598 		if (icmplen < ICMP_TSLEN) {
599 			ICMPSTAT_INC(icps_badlen);
600 			break;
601 		}
602 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
603 			goto freeit;
604 		icp->icmp_type = ICMP_TSTAMPREPLY;
605 		icp->icmp_rtime = iptime();
606 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
607 		goto reflect;
608 
609 	case ICMP_MASKREQ:
610 		if (V_icmpmaskrepl == 0)
611 			break;
612 		/*
613 		 * We are not able to respond with all ones broadcast
614 		 * unless we receive it over a point-to-point interface.
615 		 */
616 		if (icmplen < ICMP_MASKLEN)
617 			break;
618 		switch (ip->ip_dst.s_addr) {
619 
620 		case INADDR_BROADCAST:
621 		case INADDR_ANY:
622 			icmpdst.sin_addr = ip->ip_src;
623 			break;
624 
625 		default:
626 			icmpdst.sin_addr = ip->ip_dst;
627 		}
628 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
629 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
630 		if (ia == NULL)
631 			break;
632 		if (ia->ia_ifp == NULL)
633 			break;
634 		icp->icmp_type = ICMP_MASKREPLY;
635 		if (V_icmpmaskfake == 0)
636 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
637 		else
638 			icp->icmp_mask = V_icmpmaskfake;
639 		if (ip->ip_src.s_addr == 0) {
640 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
641 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
642 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
643 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
644 		}
645 reflect:
646 		ICMPSTAT_INC(icps_reflect);
647 		ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
648 		icmp_reflect(m);
649 		NET_EPOCH_EXIT();
650 		return (IPPROTO_DONE);
651 
652 	case ICMP_REDIRECT:
653 		if (V_log_redirect) {
654 			u_long src, dst, gw;
655 
656 			src = ntohl(ip->ip_src.s_addr);
657 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
658 			gw = ntohl(icp->icmp_gwaddr.s_addr);
659 			printf("icmp redirect from %d.%d.%d.%d: "
660 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
661 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
662 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
663 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
664 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
665 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
666 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
667 		}
668 		/*
669 		 * RFC1812 says we must ignore ICMP redirects if we
670 		 * are acting as router.
671 		 */
672 		if (V_drop_redirect || V_ipforwarding)
673 			break;
674 		if (code > 3)
675 			goto badcode;
676 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
677 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
678 			ICMPSTAT_INC(icps_badlen);
679 			break;
680 		}
681 		/*
682 		 * Short circuit routing redirects to force
683 		 * immediate change in the kernel's routing
684 		 * tables.  The message is also handed to anyone
685 		 * listening on a raw socket (e.g. the routing
686 		 * daemon for use in updating its tables).
687 		 */
688 		icmpgw.sin_addr = ip->ip_src;
689 		icmpdst.sin_addr = icp->icmp_gwaddr;
690 #ifdef	ICMPPRINTFS
691 		if (icmpprintfs) {
692 			char dstbuf[INET_ADDRSTRLEN];
693 			char gwbuf[INET_ADDRSTRLEN];
694 
695 			printf("redirect dst %s to %s\n",
696 			       inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
697 			       inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
698 		}
699 #endif
700 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
701 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
702 			in_rtredirect((struct sockaddr *)&icmpsrc,
703 			  (struct sockaddr *)&icmpdst,
704 			  (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
705 			  (struct sockaddr *)&icmpgw, fibnum);
706 		}
707 		pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
708 		break;
709 
710 	/*
711 	 * No kernel processing for the following;
712 	 * just fall through to send to raw listener.
713 	 */
714 	case ICMP_ECHOREPLY:
715 	case ICMP_ROUTERADVERT:
716 	case ICMP_ROUTERSOLICIT:
717 	case ICMP_TSTAMPREPLY:
718 	case ICMP_IREQREPLY:
719 	case ICMP_MASKREPLY:
720 	case ICMP_SOURCEQUENCH:
721 	default:
722 		break;
723 	}
724 
725 raw:
726 	NET_EPOCH_EXIT();
727 	*mp = m;
728 	rip_input(mp, offp, proto);
729 	return (IPPROTO_DONE);
730 
731 freeit:
732 	NET_EPOCH_EXIT();
733 	m_freem(m);
734 	return (IPPROTO_DONE);
735 }
736 
737 /*
738  * Reflect the ip packet back to the source
739  */
740 static void
icmp_reflect(struct mbuf * m)741 icmp_reflect(struct mbuf *m)
742 {
743 	struct rm_priotracker in_ifa_tracker;
744 	struct ip *ip = mtod(m, struct ip *);
745 	struct ifaddr *ifa;
746 	struct ifnet *ifp;
747 	struct in_ifaddr *ia;
748 	struct in_addr t;
749 	struct nhop4_extended nh_ext;
750 	struct mbuf *opts = NULL;
751 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
752 
753 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
754 	    IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
755 	    IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
756 		m_freem(m);	/* Bad return address */
757 		ICMPSTAT_INC(icps_badaddr);
758 		goto done;	/* Ip_output() will check for broadcast */
759 	}
760 
761 	t = ip->ip_dst;
762 	ip->ip_dst = ip->ip_src;
763 
764 	/*
765 	 * Source selection for ICMP replies:
766 	 *
767 	 * If the incoming packet was addressed directly to one of our
768 	 * own addresses, use dst as the src for the reply.
769 	 */
770 	IN_IFADDR_RLOCK(&in_ifa_tracker);
771 	LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
772 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
773 			t = IA_SIN(ia)->sin_addr;
774 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
775 			goto match;
776 		}
777 	}
778 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
779 
780 	/*
781 	 * If the incoming packet was addressed to one of our broadcast
782 	 * addresses, use the first non-broadcast address which corresponds
783 	 * to the incoming interface.
784 	 */
785 	ifp = m->m_pkthdr.rcvif;
786 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
787 		IF_ADDR_RLOCK(ifp);
788 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
789 			if (ifa->ifa_addr->sa_family != AF_INET)
790 				continue;
791 			ia = ifatoia(ifa);
792 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
793 			    t.s_addr) {
794 				t = IA_SIN(ia)->sin_addr;
795 				IF_ADDR_RUNLOCK(ifp);
796 				goto match;
797 			}
798 		}
799 		IF_ADDR_RUNLOCK(ifp);
800 	}
801 	/*
802 	 * If the packet was transiting through us, use the address of
803 	 * the interface the packet came through in.  If that interface
804 	 * doesn't have a suitable IP address, the normal selection
805 	 * criteria apply.
806 	 */
807 	if (V_icmp_rfi && ifp != NULL) {
808 		IF_ADDR_RLOCK(ifp);
809 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
810 			if (ifa->ifa_addr->sa_family != AF_INET)
811 				continue;
812 			ia = ifatoia(ifa);
813 			t = IA_SIN(ia)->sin_addr;
814 			IF_ADDR_RUNLOCK(ifp);
815 			goto match;
816 		}
817 		IF_ADDR_RUNLOCK(ifp);
818 	}
819 	/*
820 	 * If the incoming packet was not addressed directly to us, use
821 	 * designated interface for icmp replies specified by sysctl
822 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
823 	 * with normal source selection.
824 	 */
825 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
826 		IF_ADDR_RLOCK(ifp);
827 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
828 			if (ifa->ifa_addr->sa_family != AF_INET)
829 				continue;
830 			ia = ifatoia(ifa);
831 			t = IA_SIN(ia)->sin_addr;
832 			IF_ADDR_RUNLOCK(ifp);
833 			goto match;
834 		}
835 		IF_ADDR_RUNLOCK(ifp);
836 	}
837 	/*
838 	 * If the packet was transiting through us, use the address of
839 	 * the interface that is the closest to the packet source.
840 	 * When we don't have a route back to the packet source, stop here
841 	 * and drop the packet.
842 	 */
843 	if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
844 		m_freem(m);
845 		ICMPSTAT_INC(icps_noroute);
846 		goto done;
847 	}
848 	t = nh_ext.nh_src;
849 match:
850 #ifdef MAC
851 	mac_netinet_icmp_replyinplace(m);
852 #endif
853 	ip->ip_src = t;
854 	ip->ip_ttl = V_ip_defttl;
855 
856 	if (optlen > 0) {
857 		u_char *cp;
858 		int opt, cnt;
859 		u_int len;
860 
861 		/*
862 		 * Retrieve any source routing from the incoming packet;
863 		 * add on any record-route or timestamp options.
864 		 */
865 		cp = (u_char *) (ip + 1);
866 		if ((opts = ip_srcroute(m)) == NULL &&
867 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
868 			opts->m_len = sizeof(struct in_addr);
869 			mtod(opts, struct in_addr *)->s_addr = 0;
870 		}
871 		if (opts) {
872 #ifdef ICMPPRINTFS
873 		    if (icmpprintfs)
874 			    printf("icmp_reflect optlen %d rt %d => ",
875 				optlen, opts->m_len);
876 #endif
877 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
878 			    opt = cp[IPOPT_OPTVAL];
879 			    if (opt == IPOPT_EOL)
880 				    break;
881 			    if (opt == IPOPT_NOP)
882 				    len = 1;
883 			    else {
884 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
885 					    break;
886 				    len = cp[IPOPT_OLEN];
887 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
888 				        len > cnt)
889 					    break;
890 			    }
891 			    /*
892 			     * Should check for overflow, but it "can't happen"
893 			     */
894 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
895 				opt == IPOPT_SECURITY) {
896 				    bcopy((caddr_t)cp,
897 					mtod(opts, caddr_t) + opts->m_len, len);
898 				    opts->m_len += len;
899 			    }
900 		    }
901 		    /* Terminate & pad, if necessary */
902 		    cnt = opts->m_len % 4;
903 		    if (cnt) {
904 			    for (; cnt < 4; cnt++) {
905 				    *(mtod(opts, caddr_t) + opts->m_len) =
906 					IPOPT_EOL;
907 				    opts->m_len++;
908 			    }
909 		    }
910 #ifdef ICMPPRINTFS
911 		    if (icmpprintfs)
912 			    printf("%d\n", opts->m_len);
913 #endif
914 		}
915 		ip_stripoptions(m);
916 	}
917 	m_tag_delete_nonpersistent(m);
918 	m->m_flags &= ~(M_BCAST|M_MCAST);
919 	icmp_send(m, opts);
920 done:
921 	if (opts)
922 		(void)m_free(opts);
923 }
924 
925 /*
926  * Send an icmp packet back to the ip level,
927  * after supplying a checksum.
928  */
929 static void
icmp_send(struct mbuf * m,struct mbuf * opts)930 icmp_send(struct mbuf *m, struct mbuf *opts)
931 {
932 	struct ip *ip = mtod(m, struct ip *);
933 	int hlen;
934 	struct icmp *icp;
935 
936 	hlen = ip->ip_hl << 2;
937 	m->m_data += hlen;
938 	m->m_len -= hlen;
939 	icp = mtod(m, struct icmp *);
940 	icp->icmp_cksum = 0;
941 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
942 	m->m_data -= hlen;
943 	m->m_len += hlen;
944 	m->m_pkthdr.rcvif = (struct ifnet *)0;
945 #ifdef ICMPPRINTFS
946 	if (icmpprintfs) {
947 		char dstbuf[INET_ADDRSTRLEN];
948 		char srcbuf[INET_ADDRSTRLEN];
949 
950 		printf("icmp_send dst %s src %s\n",
951 		    inet_ntoa_r(ip->ip_dst, dstbuf),
952 		    inet_ntoa_r(ip->ip_src, srcbuf));
953 	}
954 #endif
955 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
956 }
957 
958 /*
959  * Return milliseconds since 00:00 UTC in network format.
960  */
961 uint32_t
iptime(void)962 iptime(void)
963 {
964 	struct timeval atv;
965 	u_long t;
966 
967 	getmicrotime(&atv);
968 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
969 	return (htonl(t));
970 }
971 
972 /*
973  * Return the next larger or smaller MTU plateau (table from RFC 1191)
974  * given current value MTU.  If DIR is less than zero, a larger plateau
975  * is returned; otherwise, a smaller value is returned.
976  */
977 int
ip_next_mtu(int mtu,int dir)978 ip_next_mtu(int mtu, int dir)
979 {
980 	static int mtutab[] = {
981 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
982 		296, 68, 0
983 	};
984 	int i, size;
985 
986 	size = (sizeof mtutab) / (sizeof mtutab[0]);
987 	if (dir >= 0) {
988 		for (i = 0; i < size; i++)
989 			if (mtu > mtutab[i])
990 				return mtutab[i];
991 	} else {
992 		for (i = size - 1; i >= 0; i--)
993 			if (mtu < mtutab[i])
994 				return mtutab[i];
995 		if (mtu == mtutab[0])
996 			return mtutab[0];
997 	}
998 	return 0;
999 }
1000 #endif /* INET */
1001 
1002 
1003 /*
1004  * badport_bandlim() - check for ICMP bandwidth limit
1005  *
1006  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
1007  *	hit our bandwidth limit and it is not ok.
1008  *
1009  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
1010  *
1011  *	For now we separate the TCP and UDP subsystems w/ different 'which'
1012  *	values.  We may eventually remove this separation (and simplify the
1013  *	code further).
1014  *
1015  *	Note that the printing of the error message is delayed so we can
1016  *	properly print the icmp error rate that the system was trying to do
1017  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1018  *	the 'final' error, but it doesn't make sense to solve the printing
1019  *	delay with more complex code.
1020  */
1021 struct icmp_rate {
1022 	const char *descr;
1023 	struct counter_rate cr;
1024 };
1025 VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1026 	{ "icmp unreach response" },
1027 	{ "icmp ping response" },
1028 	{ "icmp tstamp response" },
1029 	{ "closed port RST response" },
1030 	{ "open port RST response" },
1031 	{ "icmp6 unreach response" },
1032 	{ "sctp ootb response" }
1033 };
1034 #define	V_icmp_rates	VNET(icmp_rates)
1035 
1036 static void
icmp_bandlimit_init(void)1037 icmp_bandlimit_init(void)
1038 {
1039 
1040 	for (int i = 0; i < BANDLIM_MAX; i++) {
1041 		V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1042 		V_icmp_rates[i].cr.cr_ticks = ticks;
1043 	}
1044 }
1045 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1046     icmp_bandlimit_init, NULL);
1047 
1048 static void
icmp_bandlimit_uninit(void)1049 icmp_bandlimit_uninit(void)
1050 {
1051 
1052 	for (int i = 0; i < BANDLIM_MAX; i++)
1053 		counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1054 }
1055 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1056     icmp_bandlimit_uninit, NULL);
1057 
1058 int
badport_bandlim(int which)1059 badport_bandlim(int which)
1060 {
1061 	int64_t pps;
1062 
1063 	if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1064 		return (0);
1065 
1066 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1067 	    ("%s: which %d", __func__, which));
1068 
1069 	if ((V_icmplim + V_icmplim_curr_jitter) <= 0)
1070 		V_icmplim_curr_jitter = -V_icmplim + 1;
1071 
1072 	pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim +
1073 	    V_icmplim_curr_jitter);
1074 	if (pps > 0) {
1075 		/*
1076 		 * Adjust limit +/- to jitter the measurement to deny a
1077 		 * side-channel port scan as in CVE-2020-25705
1078 		 */
1079 		if (V_icmplim_jitter > 0) {
1080 			int32_t inc =
1081 			    arc4random_uniform(V_icmplim_jitter * 2 +1)
1082 			    - V_icmplim_jitter;
1083 
1084 			V_icmplim_curr_jitter = inc;
1085 		}
1086 	}
1087 	if (pps == -1)
1088 		return (-1);
1089 	if (pps > 0 && V_icmplim_output)
1090 		log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1091 		    V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim +
1092 		    V_icmplim_curr_jitter);
1093 	return (0);
1094 }
1095