xref: /freebsd/sys/netinet/ip_icmp.c (revision 7b71f57f4e514a2ab7308ce4147e14d90e099ad0)
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 
32 #include "opt_inet.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/mbuf.h>
37 #include <sys/protosw.h>
38 #include <sys/socket.h>
39 #include <sys/time.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/sysctl.h>
43 #include <sys/syslog.h>
44 
45 #include <net/if.h>
46 #include <net/if_var.h>
47 #include <net/if_private.h>
48 #include <net/if_types.h>
49 #include <net/route.h>
50 #include <net/route/route_ctl.h>
51 #include <net/route/nhop.h>
52 #include <net/vnet.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_fib.h>
56 #include <netinet/in_pcb.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip_icmp.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/ip_options.h>
63 #include <netinet/sctp.h>
64 #include <netinet/tcp.h>
65 #include <netinet/tcpip.h>
66 #include <netinet/icmp_var.h>
67 
68 #ifdef INET
69 
70 #include <machine/in_cksum.h>
71 
72 #include <security/mac/mac_framework.h>
73 #endif /* INET */
74 
75 extern ipproto_ctlinput_t	*ip_ctlprotox[];
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 int sysctl_icmplim_and_jitter(SYSCTL_HANDLER_ARGS);
83 VNET_DEFINE_STATIC(u_int, icmplim) = 200;
84 #define	V_icmplim			VNET(icmplim)
85 SYSCTL_PROC(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLTYPE_UINT |
86     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmplim), 0,
87     &sysctl_icmplim_and_jitter, "IU",
88     "Maximum number of ICMP responses per second");
89 
90 VNET_DEFINE_STATIC(int, icmplim_curr_jitter[BANDLIM_MAX]) = {0};
91 #define V_icmplim_curr_jitter		VNET(icmplim_curr_jitter)
92 VNET_DEFINE_STATIC(u_int, icmplim_jitter) = 16;
93 #define	V_icmplim_jitter		VNET(icmplim_jitter)
94 SYSCTL_PROC(_net_inet_icmp, OID_AUTO, icmplim_jitter, CTLTYPE_UINT |
95     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmplim_jitter), 0,
96     &sysctl_icmplim_and_jitter, "IU",
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(int, redirtimeout) = 60 * 10; /* 10 minutes */
140 #define	V_redirtimeout			VNET(redirtimeout)
141 SYSCTL_INT(_net_inet_icmp, OID_AUTO, redirtimeout, CTLFLAG_VNET | CTLFLAG_RW,
142 	&VNET_NAME(redirtimeout), 0,
143 	"Delay in seconds before expiring redirect route");
144 
145 VNET_DEFINE_STATIC(char, reply_src[IFNAMSIZ]);
146 #define	V_reply_src			VNET(reply_src)
147 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
148 	&VNET_NAME(reply_src), IFNAMSIZ,
149 	"ICMP reply source for non-local packets");
150 
151 VNET_DEFINE_STATIC(int, icmp_rfi) = 0;
152 #define	V_icmp_rfi			VNET(icmp_rfi)
153 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
154 	&VNET_NAME(icmp_rfi), 0,
155 	"ICMP reply from incoming interface for non-local packets");
156 /* Router requirements RFC 1812 section 4.3.2.3 requires 576 - 28. */
157 VNET_DEFINE_STATIC(int, icmp_quotelen) = 548;
158 #define	V_icmp_quotelen			VNET(icmp_quotelen)
159 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
160 	&VNET_NAME(icmp_quotelen), 0,
161 	"Number of bytes from original packet to quote in ICMP reply");
162 
163 VNET_DEFINE_STATIC(int, icmpbmcastecho) = 0;
164 #define	V_icmpbmcastecho		VNET(icmpbmcastecho)
165 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
166 	&VNET_NAME(icmpbmcastecho), 0,
167 	"Reply to multicast ICMP Echo Request and Timestamp packets");
168 
169 VNET_DEFINE_STATIC(int, icmptstamprepl) = 1;
170 #define	V_icmptstamprepl		VNET(icmptstamprepl)
171 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_VNET | CTLFLAG_RW,
172 	&VNET_NAME(icmptstamprepl), 0,
173 	"Respond to ICMP Timestamp packets");
174 
175 VNET_DEFINE_STATIC(int, error_keeptags) = 0;
176 #define	V_error_keeptags		VNET(error_keeptags)
177 SYSCTL_INT(_net_inet_icmp, OID_AUTO, error_keeptags, CTLFLAG_VNET | CTLFLAG_RW,
178 	&VNET_NAME(error_keeptags), 0,
179 	"ICMP error response keeps copy of mbuf_tags of original packet");
180 
181 #ifdef ICMPPRINTFS
182 int	icmpprintfs = 0;
183 #endif
184 
185 static void	icmp_reflect(struct mbuf *);
186 static void	icmp_send(struct mbuf *, struct mbuf *);
187 static int	icmp_verify_redirect_gateway(struct sockaddr_in *,
188     struct sockaddr_in *, struct sockaddr_in *, u_int);
189 
190 /*
191  * Kernel module interface for updating icmpstat.  The argument is an index
192  * into icmpstat treated as an array of u_long.  While this encodes the
193  * general layout of icmpstat into the caller, it doesn't encode its
194  * location, so that future changes to add, for example, per-CPU stats
195  * support won't cause binary compatibility problems for kernel modules.
196  */
197 void
kmod_icmpstat_inc(int statnum)198 kmod_icmpstat_inc(int statnum)
199 {
200 
201 	counter_u64_add(VNET(icmpstat)[statnum], 1);
202 }
203 
204 /*
205  * Generate an error packet of type error
206  * in response to bad packet ip.
207  */
208 void
icmp_error(struct mbuf * n,int type,int code,uint32_t dest,int mtu)209 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
210 {
211 	struct ip *oip, *nip;
212 	struct icmp *icp;
213 	struct mbuf *m;
214 	unsigned icmplen, icmpelen, nlen, oiphlen;
215 
216 	KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
217 	    __func__));
218 
219 	if (type != ICMP_REDIRECT)
220 		ICMPSTAT_INC(icps_error);
221 	/*
222 	 * Don't send error:
223 	 *  if the original packet was encrypted.
224 	 *  if not the first fragment of message.
225 	 *  in response to a multicast or broadcast packet.
226 	 *  if the old packet protocol was an ICMP error message.
227 	 */
228 	if (n->m_flags & M_DECRYPTED)
229 		goto freeit;
230 	if (n->m_flags & (M_BCAST|M_MCAST))
231 		goto freeit;
232 
233 	/* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
234 	if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
235 		goto freeit;
236 	oip = mtod(n, struct ip *);
237 	oiphlen = oip->ip_hl << 2;
238 	if (n->m_len < oiphlen + ICMP_MINLEN)
239 		goto freeit;
240 #ifdef ICMPPRINTFS
241 	if (icmpprintfs)
242 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
243 #endif
244 	if (oip->ip_off & htons(~(IP_MF|IP_DF)))
245 		goto freeit;
246 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
247 	    !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
248 		oiphlen))->icmp_type)) {
249 		ICMPSTAT_INC(icps_oldicmp);
250 		goto freeit;
251 	}
252 	/*
253 	 * Calculate length to quote from original packet and
254 	 * prevent the ICMP mbuf from overflowing.
255 	 * Unfortunately this is non-trivial since ip_forward()
256 	 * sends us truncated packets.
257 	 */
258 	nlen = m_length(n, NULL);
259 	if (oip->ip_p == IPPROTO_TCP) {
260 		struct tcphdr *th;
261 		int tcphlen;
262 
263 		if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
264 		    n->m_next == NULL)
265 			goto stdreply;
266 		if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
267 		    (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
268 			goto freeit;
269 		oip = mtod(n, struct ip *);
270 		th = mtodo(n, oiphlen);
271 		tcphlen = th->th_off << 2;
272 		if (tcphlen < sizeof(struct tcphdr))
273 			goto freeit;
274 		if (ntohs(oip->ip_len) < oiphlen + tcphlen)
275 			goto freeit;
276 		if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
277 			goto stdreply;
278 		if (n->m_len < oiphlen + tcphlen &&
279 		    (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
280 			goto freeit;
281 		oip = mtod(n, struct ip *);
282 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
283 		    ntohs(oip->ip_len) - oiphlen));
284 	} else if (oip->ip_p == IPPROTO_SCTP) {
285 		struct sctphdr *sh;
286 		struct sctp_chunkhdr *ch;
287 
288 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
289 			goto stdreply;
290 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
291 		    n->m_next == NULL)
292 			goto stdreply;
293 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
294 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
295 			goto freeit;
296 		oip = mtod(n, struct ip *);
297 		icmpelen = max(sizeof(struct sctphdr),
298 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
299 		sh = mtodo(n, oiphlen);
300 		if (ntohl(sh->v_tag) == 0 &&
301 		    ntohs(oip->ip_len) >= oiphlen +
302 		    sizeof(struct sctphdr) + 8 &&
303 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
304 		     n->m_next != NULL)) {
305 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
306 			    (n = m_pullup(n, oiphlen +
307 			    sizeof(struct sctphdr) + 8)) == NULL)
308 				goto freeit;
309 			oip = mtod(n, struct ip *);
310 			sh = mtodo(n, oiphlen);
311 			ch = (struct sctp_chunkhdr *)(sh + 1);
312 			if (ch->chunk_type == SCTP_INITIATION) {
313 				icmpelen = max(sizeof(struct sctphdr) + 8,
314 				    min(V_icmp_quotelen, ntohs(oip->ip_len) -
315 				    oiphlen));
316 			}
317 		}
318 	} else
319 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
320 		    oiphlen));
321 
322 	icmplen = min(oiphlen + icmpelen, nlen);
323 	if (icmplen < sizeof(struct ip))
324 		goto freeit;
325 
326 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
327 		m = m_gethdr(M_NOWAIT, MT_DATA);
328 	else
329 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
330 	if (m == NULL)
331 		goto freeit;
332 #ifdef MAC
333 	mac_netinet_icmp_reply(n, m);
334 #endif
335 	icmplen = min(icmplen, M_TRAILINGSPACE(m) -
336 	    sizeof(struct ip) - ICMP_MINLEN);
337 	m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
338 	m->m_data += sizeof(struct ip);
339 	m->m_len = ICMP_MINLEN + icmplen;
340 
341 	/* XXX MRT  make the outgoing packet use the same FIB
342 	 * that was associated with the incoming packet
343 	 */
344 	M_SETFIB(m, M_GETFIB(n));
345 	icp = mtod(m, struct icmp *);
346 	ICMPSTAT_INC2(icps_outhist, type);
347 	icp->icmp_type = type;
348 	if (type == ICMP_REDIRECT)
349 		icp->icmp_gwaddr.s_addr = dest;
350 	else {
351 		icp->icmp_void = 0;
352 		/*
353 		 * The following assignments assume an overlay with the
354 		 * just zeroed icmp_void field.
355 		 */
356 		if (type == ICMP_PARAMPROB) {
357 			icp->icmp_pptr = code;
358 			code = 0;
359 		} else if (type == ICMP_UNREACH &&
360 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
361 			icp->icmp_nextmtu = htons(mtu);
362 		}
363 	}
364 	icp->icmp_code = code;
365 
366 	/*
367 	 * Copy the quotation into ICMP message and
368 	 * convert quoted IP header back to network representation.
369 	 */
370 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
371 	nip = &icp->icmp_ip;
372 
373 	/*
374 	 * Set up ICMP message mbuf and copy old IP header (without options
375 	 * in front of ICMP message.
376 	 * If the original mbuf was meant to bypass the firewall, the error
377 	 * reply should bypass as well.
378 	 */
379 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
380 	KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
381 	    ("insufficient space for ip header"));
382 	m->m_data -= sizeof(struct ip);
383 	m->m_len += sizeof(struct ip);
384 	m->m_pkthdr.len = m->m_len;
385 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
386 	nip = mtod(m, struct ip *);
387 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
388 	nip->ip_len = htons(m->m_len);
389 	nip->ip_v = IPVERSION;
390 	nip->ip_hl = 5;
391 	nip->ip_p = IPPROTO_ICMP;
392 	nip->ip_tos = 0;
393 
394 	if (V_error_keeptags)
395 		m_tag_copy_chain(m, n, M_NOWAIT);
396 
397 	icmp_reflect(m);
398 
399 freeit:
400 	m_freem(n);
401 }
402 
403 int
icmp_errmap(const struct icmp * icp)404 icmp_errmap(const struct icmp *icp)
405 {
406 
407 	switch (icp->icmp_type) {
408 	case ICMP_UNREACH:
409 		switch (icp->icmp_code) {
410 		case ICMP_UNREACH_NET:
411 		case ICMP_UNREACH_HOST:
412 		case ICMP_UNREACH_SRCFAIL:
413 		case ICMP_UNREACH_NET_UNKNOWN:
414 		case ICMP_UNREACH_HOST_UNKNOWN:
415 		case ICMP_UNREACH_ISOLATED:
416 		case ICMP_UNREACH_TOSNET:
417 		case ICMP_UNREACH_TOSHOST:
418 		case ICMP_UNREACH_HOST_PRECEDENCE:
419 		case ICMP_UNREACH_PRECEDENCE_CUTOFF:
420 			return (EHOSTUNREACH);
421 		case ICMP_UNREACH_NEEDFRAG:
422 			return (EMSGSIZE);
423 		case ICMP_UNREACH_PROTOCOL:
424 		case ICMP_UNREACH_PORT:
425 		case ICMP_UNREACH_NET_PROHIB:
426 		case ICMP_UNREACH_HOST_PROHIB:
427 		case ICMP_UNREACH_FILTER_PROHIB:
428 			return (ECONNREFUSED);
429 		default:
430 			return (0);
431 		}
432 	case ICMP_TIMXCEED:
433 		switch (icp->icmp_code) {
434 		case ICMP_TIMXCEED_INTRANS:
435 			return (EHOSTUNREACH);
436 		default:
437 			return (0);
438 		}
439 	case ICMP_PARAMPROB:
440 		switch (icp->icmp_code) {
441 		case ICMP_PARAMPROB_ERRATPTR:
442 		case ICMP_PARAMPROB_OPTABSENT:
443 			return (ENOPROTOOPT);
444 		default:
445 			return (0);
446 		}
447 	default:
448 		return (0);
449 	}
450 }
451 
452 /*
453  * Process a received ICMP message.
454  */
455 int
icmp_input(struct mbuf ** mp,int * offp,int proto)456 icmp_input(struct mbuf **mp, int *offp, int proto)
457 {
458 	struct icmp *icp;
459 	struct in_ifaddr *ia;
460 	struct mbuf *m = *mp;
461 	struct ip *ip = mtod(m, struct ip *);
462 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
463 	int hlen = *offp;
464 	int icmplen = ntohs(ip->ip_len) - *offp;
465 	int i, code;
466 	int fibnum;
467 
468 	NET_EPOCH_ASSERT();
469 
470 	*mp = NULL;
471 
472 	/*
473 	 * Locate icmp structure in mbuf, and check
474 	 * that not corrupted and of at least minimum length.
475 	 */
476 #ifdef ICMPPRINTFS
477 	if (icmpprintfs) {
478 		char srcbuf[INET_ADDRSTRLEN];
479 		char dstbuf[INET_ADDRSTRLEN];
480 
481 		printf("icmp_input from %s to %s, len %d\n",
482 		    inet_ntoa_r(ip->ip_src, srcbuf),
483 		    inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
484 	}
485 #endif
486 	if (icmplen < ICMP_MINLEN) {
487 		ICMPSTAT_INC(icps_tooshort);
488 		goto freeit;
489 	}
490 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
491 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
492 		ICMPSTAT_INC(icps_tooshort);
493 		return (IPPROTO_DONE);
494 	}
495 	ip = mtod(m, struct ip *);
496 	m->m_len -= hlen;
497 	m->m_data += hlen;
498 	icp = mtod(m, struct icmp *);
499 	if (in_cksum(m, icmplen)) {
500 		ICMPSTAT_INC(icps_checksum);
501 		goto freeit;
502 	}
503 	m->m_len += hlen;
504 	m->m_data -= hlen;
505 
506 #ifdef ICMPPRINTFS
507 	if (icmpprintfs)
508 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
509 		    icp->icmp_code);
510 #endif
511 
512 	/*
513 	 * Message type specific processing.
514 	 */
515 	if (icp->icmp_type > ICMP_MAXTYPE)
516 		goto raw;
517 
518 	/* Initialize */
519 	bzero(&icmpsrc, sizeof(icmpsrc));
520 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
521 	icmpsrc.sin_family = AF_INET;
522 	bzero(&icmpdst, sizeof(icmpdst));
523 	icmpdst.sin_len = sizeof(struct sockaddr_in);
524 	icmpdst.sin_family = AF_INET;
525 	bzero(&icmpgw, sizeof(icmpgw));
526 	icmpgw.sin_len = sizeof(struct sockaddr_in);
527 	icmpgw.sin_family = AF_INET;
528 
529 	ICMPSTAT_INC2(icps_inhist, icp->icmp_type);
530 	code = icp->icmp_code;
531 	switch (icp->icmp_type) {
532 	case ICMP_UNREACH:
533 		if (code > ICMP_UNREACH_PRECEDENCE_CUTOFF)
534 			goto badcode;
535 		else
536 			goto deliver;
537 
538 	case ICMP_TIMXCEED:
539 		if (code > ICMP_TIMXCEED_REASS)
540 			goto badcode;
541 		else
542 			goto deliver;
543 
544 	case ICMP_PARAMPROB:
545 		if (code > ICMP_PARAMPROB_LENGTH)
546 			goto badcode;
547 
548 	deliver:
549 		/*
550 		 * Problem with datagram; advise higher level routines.
551 		 */
552 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
553 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
554 			ICMPSTAT_INC(icps_badlen);
555 			goto freeit;
556 		}
557 		/* Discard ICMP's in response to multicast packets */
558 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
559 			goto badcode;
560 		/* Filter out responses to INADDR_ANY, protocols ignore it. */
561 		if (icp->icmp_ip.ip_dst.s_addr == INADDR_ANY ||
562 		    icp->icmp_ip.ip_src.s_addr == INADDR_ANY)
563 			goto freeit;
564 #ifdef ICMPPRINTFS
565 		if (icmpprintfs)
566 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
567 #endif
568 		/*
569 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
570 		 * notification to TCP layer.
571 		 */
572 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
573 		ip_stripoptions(m);
574 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
575 			/* This should actually not happen */
576 			ICMPSTAT_INC(icps_tooshort);
577 			return (IPPROTO_DONE);
578 		}
579 		ip = mtod(m, struct ip *);
580 		icp = (struct icmp *)(ip + 1);
581 		/*
582 		 * The upper layer handler can rely on:
583 		 * - The outer IP header has no options.
584 		 * - The outer IP header, the ICMP header, the inner IP header,
585 		 *   and the first n bytes of the inner payload are contiguous.
586 		 *   n is at least 8, but might be larger based on
587 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
588 		 */
589 		if (ip_ctlprotox[icp->icmp_ip.ip_p] != NULL)
590 			ip_ctlprotox[icp->icmp_ip.ip_p](icp);
591 		break;
592 
593 	badcode:
594 		ICMPSTAT_INC(icps_badcode);
595 		break;
596 
597 	case ICMP_ECHO:
598 		if (!V_icmpbmcastecho
599 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
600 			ICMPSTAT_INC(icps_bmcastecho);
601 			break;
602 		}
603 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
604 			goto freeit;
605 		icp->icmp_type = ICMP_ECHOREPLY;
606 		goto reflect;
607 
608 	case ICMP_TSTAMP:
609 		if (V_icmptstamprepl == 0)
610 			break;
611 		if (!V_icmpbmcastecho
612 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
613 			ICMPSTAT_INC(icps_bmcasttstamp);
614 			break;
615 		}
616 		if (icmplen < ICMP_TSLEN) {
617 			ICMPSTAT_INC(icps_badlen);
618 			break;
619 		}
620 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
621 			goto freeit;
622 		icp->icmp_type = ICMP_TSTAMPREPLY;
623 		icp->icmp_rtime = iptime();
624 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
625 		goto reflect;
626 
627 	case ICMP_MASKREQ:
628 		if (V_icmpmaskrepl == 0)
629 			break;
630 		/*
631 		 * We are not able to respond with all ones broadcast
632 		 * unless we receive it over a point-to-point interface.
633 		 */
634 		if (icmplen < ICMP_MASKLEN)
635 			break;
636 		if (in_broadcast(ip->ip_dst))
637 			icmpdst.sin_addr = ip->ip_src;
638 		else
639 			icmpdst.sin_addr = ip->ip_dst;
640 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
641 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
642 		if (ia == NULL)
643 			break;
644 		if (ia->ia_ifp == NULL)
645 			break;
646 		icp->icmp_type = ICMP_MASKREPLY;
647 		if (V_icmpmaskfake == 0)
648 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
649 		else
650 			icp->icmp_mask = V_icmpmaskfake;
651 		if (ip->ip_src.s_addr == 0) {
652 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
653 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
654 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
655 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
656 		}
657 reflect:
658 		ICMPSTAT_INC(icps_reflect);
659 		ICMPSTAT_INC2(icps_outhist, icp->icmp_type);
660 		icmp_reflect(m);
661 		return (IPPROTO_DONE);
662 
663 	case ICMP_REDIRECT:
664 		if (V_log_redirect) {
665 			u_long src, dst, gw;
666 
667 			src = ntohl(ip->ip_src.s_addr);
668 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
669 			gw = ntohl(icp->icmp_gwaddr.s_addr);
670 			printf("icmp redirect from %d.%d.%d.%d: "
671 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
672 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
673 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
674 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
675 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
676 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
677 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
678 		}
679 		/*
680 		 * RFC1812 says we must ignore ICMP redirects if we
681 		 * are acting as router.
682 		 */
683 		if (V_drop_redirect || V_ipforwarding)
684 			break;
685 		if (code > 3)
686 			goto badcode;
687 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
688 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
689 			ICMPSTAT_INC(icps_badlen);
690 			break;
691 		}
692 		/*
693 		 * Short circuit routing redirects to force
694 		 * immediate change in the kernel's routing
695 		 * tables.  The message is also handed to anyone
696 		 * listening on a raw socket (e.g. the routing
697 		 * daemon for use in updating its tables).
698 		 */
699 		icmpgw.sin_addr = ip->ip_src;
700 		icmpdst.sin_addr = icp->icmp_gwaddr;
701 #ifdef	ICMPPRINTFS
702 		if (icmpprintfs) {
703 			char dstbuf[INET_ADDRSTRLEN];
704 			char gwbuf[INET_ADDRSTRLEN];
705 
706 			printf("redirect dst %s to %s\n",
707 			       inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
708 			       inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
709 		}
710 #endif
711 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
712 
713 		/*
714 		 * RFC 1122 says network (code 0,2) redirects SHOULD
715 		 * be treated identically to the host redirects.
716 		 * Given that, ignore network masks.
717 		 */
718 
719 		/*
720 		 * Variable values:
721 		 * icmpsrc: route destination
722 		 * icmpdst: route gateway
723 		 * icmpgw: message source
724 		 */
725 
726 		if (icmp_verify_redirect_gateway(&icmpgw, &icmpsrc, &icmpdst,
727 		    M_GETFIB(m)) != 0) {
728 			/* TODO: increment bad redirects here */
729 			break;
730 		}
731 
732 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
733 			rib_add_redirect(fibnum, (struct sockaddr *)&icmpsrc,
734 			    (struct sockaddr *)&icmpdst,
735 			    (struct sockaddr *)&icmpgw, m->m_pkthdr.rcvif,
736 			    RTF_GATEWAY, V_redirtimeout);
737 		}
738 		break;
739 
740 	/*
741 	 * No kernel processing for the following;
742 	 * just fall through to send to raw listener.
743 	 */
744 	case ICMP_ECHOREPLY:
745 	case ICMP_ROUTERADVERT:
746 	case ICMP_ROUTERSOLICIT:
747 	case ICMP_TSTAMPREPLY:
748 	case ICMP_IREQREPLY:
749 	case ICMP_MASKREPLY:
750 	case ICMP_SOURCEQUENCH:
751 	default:
752 		break;
753 	}
754 
755 raw:
756 	*mp = m;
757 	rip_input(mp, offp, proto);
758 	return (IPPROTO_DONE);
759 
760 freeit:
761 	m_freem(m);
762 	return (IPPROTO_DONE);
763 }
764 
765 /*
766  * Reflect the ip packet back to the source
767  */
768 static void
icmp_reflect(struct mbuf * m)769 icmp_reflect(struct mbuf *m)
770 {
771 	struct ip *ip = mtod(m, struct ip *);
772 	struct ifaddr *ifa;
773 	struct ifnet *ifp;
774 	struct in_ifaddr *ia;
775 	struct in_addr t;
776 	struct nhop_object *nh;
777 	struct mbuf *opts = NULL;
778 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
779 
780 	NET_EPOCH_ASSERT();
781 
782 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
783 	    (IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) && !V_ip_allow_net240) ||
784 	    (IN_ZERONET(ntohl(ip->ip_src.s_addr)) && !V_ip_allow_net0) ||
785 	    in_nullhost(ip->ip_src) ) {
786 		m_freem(m);	/* Bad return address */
787 		ICMPSTAT_INC(icps_badaddr);
788 		goto done;	/* ip_output() will check for broadcast */
789 	}
790 
791 	t = ip->ip_dst;
792 	ip->ip_dst = ip->ip_src;
793 
794 	/*
795 	 * Source selection for ICMP replies:
796 	 *
797 	 * If the incoming packet was addressed directly to one of our
798 	 * own addresses, use dst as the src for the reply.
799 	 */
800 	CK_LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
801 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
802 			t = IA_SIN(ia)->sin_addr;
803 			goto match;
804 		}
805 	}
806 
807 	/*
808 	 * If the incoming packet was addressed to one of our broadcast
809 	 * addresses, use the first non-broadcast address which corresponds
810 	 * to the incoming interface.
811 	 */
812 	ifp = m->m_pkthdr.rcvif;
813 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
814 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
815 			if (ifa->ifa_addr->sa_family != AF_INET)
816 				continue;
817 			ia = ifatoia(ifa);
818 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
819 			    t.s_addr) {
820 				t = IA_SIN(ia)->sin_addr;
821 				goto match;
822 			}
823 		}
824 	}
825 	/*
826 	 * If the packet was transiting through us, use the address of
827 	 * the interface the packet came through in.  If that interface
828 	 * doesn't have a suitable IP address, the normal selection
829 	 * criteria apply.
830 	 */
831 	if (V_icmp_rfi && ifp != NULL) {
832 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
833 			if (ifa->ifa_addr->sa_family != AF_INET)
834 				continue;
835 			ia = ifatoia(ifa);
836 			t = IA_SIN(ia)->sin_addr;
837 			goto match;
838 		}
839 	}
840 	/*
841 	 * If the incoming packet was not addressed directly to us, use
842 	 * designated interface for icmp replies specified by sysctl
843 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
844 	 * with normal source selection.
845 	 */
846 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
847 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
848 			if (ifa->ifa_addr->sa_family != AF_INET)
849 				continue;
850 			ia = ifatoia(ifa);
851 			t = IA_SIN(ia)->sin_addr;
852 			goto match;
853 		}
854 	}
855 	/*
856 	 * If the packet was transiting through us, use the address of
857 	 * the interface that is the closest to the packet source.
858 	 * When we don't have a route back to the packet source, stop here
859 	 * and drop the packet.
860 	 */
861 	nh = fib4_lookup(M_GETFIB(m), ip->ip_dst, 0, NHR_NONE, 0);
862 	if (nh == NULL) {
863 		m_freem(m);
864 		ICMPSTAT_INC(icps_noroute);
865 		goto done;
866 	}
867 	t = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr;
868 match:
869 #ifdef MAC
870 	mac_netinet_icmp_replyinplace(m);
871 #endif
872 	ip->ip_src = t;
873 	/* ip->ip_tos will be reflected. */
874 	ip->ip_off = htons(0);
875 	ip->ip_ttl = V_ip_defttl;
876 
877 	if (optlen > 0) {
878 		u_char *cp;
879 		int opt, cnt;
880 		u_int len;
881 
882 		/*
883 		 * Retrieve any source routing from the incoming packet;
884 		 * add on any record-route or timestamp options.
885 		 */
886 		cp = (u_char *) (ip + 1);
887 		if ((opts = ip_srcroute(m)) == NULL &&
888 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
889 			opts->m_len = sizeof(struct in_addr);
890 			mtod(opts, struct in_addr *)->s_addr = 0;
891 		}
892 		if (opts) {
893 #ifdef ICMPPRINTFS
894 		    if (icmpprintfs)
895 			    printf("icmp_reflect optlen %d rt %d => ",
896 				optlen, opts->m_len);
897 #endif
898 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
899 			    opt = cp[IPOPT_OPTVAL];
900 			    if (opt == IPOPT_EOL)
901 				    break;
902 			    if (opt == IPOPT_NOP)
903 				    len = 1;
904 			    else {
905 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
906 					    break;
907 				    len = cp[IPOPT_OLEN];
908 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
909 				        len > cnt)
910 					    break;
911 			    }
912 			    /*
913 			     * Should check for overflow, but it "can't happen"
914 			     */
915 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
916 				opt == IPOPT_SECURITY) {
917 				    bcopy((caddr_t)cp,
918 					mtod(opts, caddr_t) + opts->m_len, len);
919 				    opts->m_len += len;
920 			    }
921 		    }
922 		    /* Terminate & pad, if necessary */
923 		    cnt = opts->m_len % 4;
924 		    if (cnt) {
925 			    for (; cnt < 4; cnt++) {
926 				    *(mtod(opts, caddr_t) + opts->m_len) =
927 					IPOPT_EOL;
928 				    opts->m_len++;
929 			    }
930 		    }
931 #ifdef ICMPPRINTFS
932 		    if (icmpprintfs)
933 			    printf("%d\n", opts->m_len);
934 #endif
935 		}
936 		ip_stripoptions(m);
937 	}
938 	m_tag_delete_nonpersistent(m);
939 	m->m_flags &= ~(M_BCAST|M_MCAST);
940 	icmp_send(m, opts);
941 done:
942 	if (opts)
943 		(void)m_free(opts);
944 }
945 
946 /*
947  * Verifies if redirect message is valid, according to RFC 1122
948  *
949  * @src: sockaddr with address of redirect originator
950  * @dst: sockaddr with destination in question
951  * @gateway: new proposed gateway
952  *
953  * Returns 0 on success.
954  */
955 static int
icmp_verify_redirect_gateway(struct sockaddr_in * src,struct sockaddr_in * dst,struct sockaddr_in * gateway,u_int fibnum)956 icmp_verify_redirect_gateway(struct sockaddr_in *src, struct sockaddr_in *dst,
957     struct sockaddr_in *gateway, u_int fibnum)
958 {
959 	struct nhop_object *nh;
960 	struct ifaddr *ifa;
961 
962 	NET_EPOCH_ASSERT();
963 
964 	/* Verify the gateway is directly reachable. */
965 	if ((ifa = ifa_ifwithnet((struct sockaddr *)gateway, 0, fibnum))==NULL)
966 		return (ENETUNREACH);
967 
968 	/* TODO: fib-aware. */
969 	if (ifa_ifwithaddr_check((struct sockaddr *)gateway))
970 		return (EHOSTUNREACH);
971 
972 	nh = fib4_lookup(fibnum, dst->sin_addr, 0, NHR_NONE, 0);
973 	if (nh == NULL)
974 		return (EINVAL);
975 
976 	/*
977 	 * If the redirect isn't from our current router for this dst,
978 	 * it's either old or wrong.  If it redirects us to ourselves,
979 	 * we have a routing loop, perhaps as a result of an interface
980 	 * going down recently.
981 	 */
982 	if (!sa_equal((struct sockaddr *)src, &nh->gw_sa))
983 		return (EINVAL);
984 	if (nh->nh_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK)
985 		return (EINVAL);
986 
987 	/* If host route already exists, ignore redirect. */
988 	if (nh->nh_flags & NHF_HOST)
989 		return (EEXIST);
990 
991 	/* If the prefix is directly reachable, ignore redirect. */
992 	if (!(nh->nh_flags & NHF_GATEWAY))
993 		return (EEXIST);
994 
995 	return (0);
996 }
997 
998 /*
999  * Send an icmp packet back to the ip level,
1000  * after supplying a checksum.
1001  */
1002 static void
icmp_send(struct mbuf * m,struct mbuf * opts)1003 icmp_send(struct mbuf *m, struct mbuf *opts)
1004 {
1005 	struct ip *ip = mtod(m, struct ip *);
1006 	int hlen;
1007 	struct icmp *icp;
1008 
1009 	hlen = ip->ip_hl << 2;
1010 	m->m_data += hlen;
1011 	m->m_len -= hlen;
1012 	icp = mtod(m, struct icmp *);
1013 	icp->icmp_cksum = 0;
1014 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
1015 	m->m_data -= hlen;
1016 	m->m_len += hlen;
1017 	m->m_pkthdr.rcvif = (struct ifnet *)0;
1018 #ifdef ICMPPRINTFS
1019 	if (icmpprintfs) {
1020 		char dstbuf[INET_ADDRSTRLEN];
1021 		char srcbuf[INET_ADDRSTRLEN];
1022 
1023 		printf("icmp_send dst %s src %s\n",
1024 		    inet_ntoa_r(ip->ip_dst, dstbuf),
1025 		    inet_ntoa_r(ip->ip_src, srcbuf));
1026 	}
1027 #endif
1028 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
1029 }
1030 
1031 /*
1032  * Return milliseconds since 00:00 UTC in network format.
1033  */
1034 uint32_t
iptime(void)1035 iptime(void)
1036 {
1037 	struct timeval atv;
1038 	u_long t;
1039 
1040 	getmicrotime(&atv);
1041 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
1042 	return (htonl(t));
1043 }
1044 
1045 /*
1046  * Return the next larger or smaller MTU plateau (table from RFC 1191)
1047  * given current value MTU.  If DIR is less than zero, a larger plateau
1048  * is returned; otherwise, a smaller value is returned.
1049  */
1050 int
ip_next_mtu(int mtu,int dir)1051 ip_next_mtu(int mtu, int dir)
1052 {
1053 	static int mtutab[] = {
1054 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
1055 		296, 68, 0
1056 	};
1057 	int i, size;
1058 
1059 	size = (sizeof mtutab) / (sizeof mtutab[0]);
1060 	if (dir >= 0) {
1061 		for (i = 0; i < size; i++)
1062 			if (mtu > mtutab[i])
1063 				return mtutab[i];
1064 	} else {
1065 		for (i = size - 1; i >= 0; i--)
1066 			if (mtu < mtutab[i])
1067 				return mtutab[i];
1068 		if (mtu == mtutab[0])
1069 			return mtutab[0];
1070 	}
1071 	return 0;
1072 }
1073 #endif /* INET */
1074 
1075 /*
1076  * badport_bandlim() - check for ICMP bandwidth limit
1077  *
1078  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
1079  *	hit our bandwidth limit and it is not ok.
1080  *
1081  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
1082  *
1083  *	For now we separate the TCP and UDP subsystems w/ different 'which'
1084  *	values.  We may eventually remove this separation (and simplify the
1085  *	code further).
1086  *
1087  *	Note that the printing of the error message is delayed so we can
1088  *	properly print the icmp error rate that the system was trying to do
1089  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1090  *	the 'final' error, but it doesn't make sense to solve the printing
1091  *	delay with more complex code.
1092  */
1093 VNET_DEFINE_STATIC(struct counter_rate *, icmp_rates[BANDLIM_MAX]);
1094 #define	V_icmp_rates	VNET(icmp_rates)
1095 
1096 static const char *icmp_rate_descrs[BANDLIM_MAX] = {
1097 	[BANDLIM_ICMP_UNREACH] = "icmp unreach",
1098 	[BANDLIM_ICMP_ECHO] = "icmp ping",
1099 	[BANDLIM_ICMP_TSTAMP] = "icmp tstamp",
1100 	[BANDLIM_TCP_RST] = "tcp reset",
1101 	[BANDLIM_ICMP6_UNREACH] = "icmp6 unreach",
1102 	[BANDLIM_SCTP_OOTB] = "sctp ootb",
1103 };
1104 
1105 static void
icmplim_new_jitter(int which)1106 icmplim_new_jitter(int which)
1107 {
1108 	/*
1109 	 * Adjust limit +/- to jitter the measurement to deny a side-channel
1110 	 * port scan as in https://dl.acm.org/doi/10.1145/3372297.3417280
1111 	 */
1112 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1113 	    ("%s: which %d", __func__, which));
1114 	if (V_icmplim_jitter > 0)
1115 		V_icmplim_curr_jitter[which] =
1116 		    arc4random_uniform(V_icmplim_jitter * 2 + 1) -
1117 		    V_icmplim_jitter;
1118 }
1119 
1120 static int
sysctl_icmplim_and_jitter(SYSCTL_HANDLER_ARGS)1121 sysctl_icmplim_and_jitter(SYSCTL_HANDLER_ARGS)
1122 {
1123 	uint32_t new;
1124 	int error;
1125 	bool lim;
1126 
1127 	MPASS(oidp->oid_arg1 == &VNET_NAME(icmplim) ||
1128 	    oidp->oid_arg1 == &VNET_NAME(icmplim_jitter));
1129 
1130 	lim = (oidp->oid_arg1 == &VNET_NAME(icmplim));
1131 	new = lim ? V_icmplim : V_icmplim_jitter;
1132 	error = sysctl_handle_int(oidp, &new, 0, req);
1133 	if (error == 0 && req->newptr) {
1134 		if (lim) {
1135 			if (new != 0 && new <= V_icmplim_jitter)
1136 				error = EINVAL;
1137 			else
1138 				V_icmplim = new;
1139 		} else {
1140 			if (new >= V_icmplim)
1141 				error = EINVAL;
1142 			else {
1143 				V_icmplim_jitter = new;
1144 				for (int i = 0; i < BANDLIM_MAX; i++) {
1145 					icmplim_new_jitter(i);
1146 				}
1147 			}
1148 		}
1149 	}
1150 	MPASS(V_icmplim == 0 || V_icmplim > V_icmplim_jitter);
1151 
1152 	return (error);
1153 }
1154 
1155 static void
icmp_bandlimit_init(void)1156 icmp_bandlimit_init(void)
1157 {
1158 
1159 	for (int i = 0; i < BANDLIM_MAX; i++) {
1160 		V_icmp_rates[i] = counter_rate_alloc(M_WAITOK, 1);
1161 		icmplim_new_jitter(i);
1162 	}
1163 }
1164 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1165     icmp_bandlimit_init, NULL);
1166 
1167 #ifdef VIMAGE
1168 static void
icmp_bandlimit_uninit(void)1169 icmp_bandlimit_uninit(void)
1170 {
1171 
1172 	for (int i = 0; i < BANDLIM_MAX; i++)
1173 		counter_rate_free(V_icmp_rates[i]);
1174 }
1175 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1176     icmp_bandlimit_uninit, NULL);
1177 #endif
1178 
1179 int
badport_bandlim(int which)1180 badport_bandlim(int which)
1181 {
1182 	int64_t pps;
1183 
1184 	if (V_icmplim == 0)
1185 		return (0);
1186 
1187 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1188 	    ("%s: which %d", __func__, which));
1189 
1190 	pps = counter_ratecheck(V_icmp_rates[which], V_icmplim +
1191 	    V_icmplim_curr_jitter[which]);
1192 	if (pps > 0) {
1193 		if (V_icmplim_output)
1194 			log(LOG_NOTICE,
1195 			    "Limiting %s response from %jd to %d packets/sec\n",
1196 			    icmp_rate_descrs[which], (intmax_t )pps,
1197 			    V_icmplim + V_icmplim_curr_jitter[which]);
1198 		icmplim_new_jitter(which);
1199 	}
1200 	if (pps == -1)
1201 		return (-1);
1202 	return (0);
1203 }
1204