xref: /freebsd/sys/netinet/ip_icmp.c (revision 783d3ff6d7fae619db8a7990b8a6387de0c677b5)
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 <sys/cdefs.h>
33 #include "opt_inet.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/mbuf.h>
38 #include <sys/protosw.h>
39 #include <sys/socket.h>
40 #include <sys/time.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/sysctl.h>
44 #include <sys/syslog.h>
45 
46 #include <net/if.h>
47 #include <net/if_var.h>
48 #include <net/if_private.h>
49 #include <net/if_types.h>
50 #include <net/route.h>
51 #include <net/route/route_ctl.h>
52 #include <net/route/nhop.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/tcpip.h>
67 #include <netinet/icmp_var.h>
68 
69 #ifdef INET
70 
71 #include <machine/in_cksum.h>
72 
73 #include <security/mac/mac_framework.h>
74 #endif /* INET */
75 
76 extern ipproto_ctlinput_t	*ip_ctlprotox[];
77 
78 /*
79  * ICMP routines: error generation, receive packet processing, and
80  * routines to turnaround packets back to the originator, and
81  * host table maintenance routines.
82  */
83 static int sysctl_icmplim_and_jitter(SYSCTL_HANDLER_ARGS);
84 VNET_DEFINE_STATIC(u_int, icmplim) = 200;
85 #define	V_icmplim			VNET(icmplim)
86 SYSCTL_PROC(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLTYPE_UINT |
87     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmplim), 0,
88     &sysctl_icmplim_and_jitter, "IU",
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(u_int, icmplim_jitter) = 16;
94 #define	V_icmplim_jitter		VNET(icmplim_jitter)
95 SYSCTL_PROC(_net_inet_icmp, OID_AUTO, icmplim_jitter, CTLTYPE_UINT |
96     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmplim_jitter), 0,
97     &sysctl_icmplim_and_jitter, "IU",
98     "Random icmplim jitter adjustment limit");
99 
100 VNET_DEFINE_STATIC(int, icmplim_output) = 1;
101 #define	V_icmplim_output		VNET(icmplim_output)
102 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
103 	&VNET_NAME(icmplim_output), 0,
104 	"Enable logging of ICMP response rate limiting");
105 
106 #ifdef INET
107 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
108 VNET_PCPUSTAT_SYSINIT(icmpstat);
109 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
110     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
111 
112 #ifdef VIMAGE
113 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
114 #endif /* VIMAGE */
115 
116 VNET_DEFINE_STATIC(int, icmpmaskrepl) = 0;
117 #define	V_icmpmaskrepl			VNET(icmpmaskrepl)
118 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
119 	&VNET_NAME(icmpmaskrepl), 0,
120 	"Reply to ICMP Address Mask Request packets");
121 
122 VNET_DEFINE_STATIC(u_int, icmpmaskfake) = 0;
123 #define	V_icmpmaskfake			VNET(icmpmaskfake)
124 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
125 	&VNET_NAME(icmpmaskfake), 0,
126 	"Fake reply to ICMP Address Mask Request packets");
127 
128 VNET_DEFINE(int, drop_redirect) = 0;
129 #define	V_drop_redirect			VNET(drop_redirect)
130 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
131 	&VNET_NAME(drop_redirect), 0,
132 	"Ignore ICMP redirects");
133 
134 VNET_DEFINE_STATIC(int, log_redirect) = 0;
135 #define	V_log_redirect			VNET(log_redirect)
136 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
137 	&VNET_NAME(log_redirect), 0,
138 	"Log ICMP redirects to the console");
139 
140 VNET_DEFINE_STATIC(int, redirtimeout) = 60 * 10; /* 10 minutes */
141 #define	V_redirtimeout			VNET(redirtimeout)
142 SYSCTL_INT(_net_inet_icmp, OID_AUTO, redirtimeout, CTLFLAG_VNET | CTLFLAG_RW,
143 	&VNET_NAME(redirtimeout), 0,
144 	"Delay in seconds before expiring redirect route");
145 
146 VNET_DEFINE_STATIC(char, reply_src[IFNAMSIZ]);
147 #define	V_reply_src			VNET(reply_src)
148 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
149 	&VNET_NAME(reply_src), IFNAMSIZ,
150 	"ICMP reply source for non-local packets");
151 
152 VNET_DEFINE_STATIC(int, icmp_rfi) = 0;
153 #define	V_icmp_rfi			VNET(icmp_rfi)
154 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
155 	&VNET_NAME(icmp_rfi), 0,
156 	"ICMP reply from incoming interface for non-local packets");
157 /* Router requirements RFC 1812 section 4.3.2.3 requires 576 - 28. */
158 VNET_DEFINE_STATIC(int, icmp_quotelen) = 548;
159 #define	V_icmp_quotelen			VNET(icmp_quotelen)
160 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
161 	&VNET_NAME(icmp_quotelen), 0,
162 	"Number of bytes from original packet to quote in ICMP reply");
163 
164 VNET_DEFINE_STATIC(int, icmpbmcastecho) = 0;
165 #define	V_icmpbmcastecho		VNET(icmpbmcastecho)
166 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
167 	&VNET_NAME(icmpbmcastecho), 0,
168 	"Reply to multicast ICMP Echo Request and Timestamp packets");
169 
170 VNET_DEFINE_STATIC(int, icmptstamprepl) = 1;
171 #define	V_icmptstamprepl		VNET(icmptstamprepl)
172 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_VNET | CTLFLAG_RW,
173 	&VNET_NAME(icmptstamprepl), 0,
174 	"Respond to ICMP Timestamp packets");
175 
176 VNET_DEFINE_STATIC(int, error_keeptags) = 0;
177 #define	V_error_keeptags		VNET(error_keeptags)
178 SYSCTL_INT(_net_inet_icmp, OID_AUTO, error_keeptags, CTLFLAG_VNET | CTLFLAG_RW,
179 	&VNET_NAME(error_keeptags), 0,
180 	"ICMP error response keeps copy of mbuf_tags of original packet");
181 
182 #ifdef ICMPPRINTFS
183 int	icmpprintfs = 0;
184 #endif
185 
186 static void	icmp_reflect(struct mbuf *);
187 static void	icmp_send(struct mbuf *, struct mbuf *);
188 static int	icmp_verify_redirect_gateway(struct sockaddr_in *,
189     struct sockaddr_in *, struct sockaddr_in *, u_int);
190 
191 /*
192  * Kernel module interface for updating icmpstat.  The argument is an index
193  * into icmpstat treated as an array of u_long.  While this encodes the
194  * general layout of icmpstat into the caller, it doesn't encode its
195  * location, so that future changes to add, for example, per-CPU stats
196  * support won't cause binary compatibility problems for kernel modules.
197  */
198 void
199 kmod_icmpstat_inc(int statnum)
200 {
201 
202 	counter_u64_add(VNET(icmpstat)[statnum], 1);
203 }
204 
205 /*
206  * Generate an error packet of type error
207  * in response to bad packet ip.
208  */
209 void
210 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
211 {
212 	struct ip *oip, *nip;
213 	struct icmp *icp;
214 	struct mbuf *m;
215 	unsigned icmplen, icmpelen, nlen, oiphlen;
216 
217 	KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
218 	    __func__));
219 
220 	if (type != ICMP_REDIRECT)
221 		ICMPSTAT_INC(icps_error);
222 	/*
223 	 * Don't send error:
224 	 *  if the original packet was encrypted.
225 	 *  if not the first fragment of message.
226 	 *  in response to a multicast or broadcast packet.
227 	 *  if the old packet protocol was an ICMP error message.
228 	 */
229 	if (n->m_flags & M_DECRYPTED)
230 		goto freeit;
231 	if (n->m_flags & (M_BCAST|M_MCAST))
232 		goto freeit;
233 
234 	/* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
235 	if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
236 		goto freeit;
237 	oip = mtod(n, struct ip *);
238 	oiphlen = oip->ip_hl << 2;
239 	if (n->m_len < oiphlen + ICMP_MINLEN)
240 		goto freeit;
241 #ifdef ICMPPRINTFS
242 	if (icmpprintfs)
243 		printf("icmp_error(%p, %x, %d)\n", oip, type, code);
244 #endif
245 	if (oip->ip_off & htons(~(IP_MF|IP_DF)))
246 		goto freeit;
247 	if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
248 	    !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
249 		oiphlen))->icmp_type)) {
250 		ICMPSTAT_INC(icps_oldicmp);
251 		goto freeit;
252 	}
253 	/*
254 	 * Calculate length to quote from original packet and
255 	 * prevent the ICMP mbuf from overflowing.
256 	 * Unfortunately this is non-trivial since ip_forward()
257 	 * sends us truncated packets.
258 	 */
259 	nlen = m_length(n, NULL);
260 	if (oip->ip_p == IPPROTO_TCP) {
261 		struct tcphdr *th;
262 		int tcphlen;
263 
264 		if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
265 		    n->m_next == NULL)
266 			goto stdreply;
267 		if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
268 		    (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
269 			goto freeit;
270 		oip = mtod(n, struct ip *);
271 		th = mtodo(n, oiphlen);
272 		tcphlen = th->th_off << 2;
273 		if (tcphlen < sizeof(struct tcphdr))
274 			goto freeit;
275 		if (ntohs(oip->ip_len) < oiphlen + tcphlen)
276 			goto freeit;
277 		if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
278 			goto stdreply;
279 		if (n->m_len < oiphlen + tcphlen &&
280 		    (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
281 			goto freeit;
282 		oip = mtod(n, struct ip *);
283 		icmpelen = max(tcphlen, min(V_icmp_quotelen,
284 		    ntohs(oip->ip_len) - oiphlen));
285 	} else if (oip->ip_p == IPPROTO_SCTP) {
286 		struct sctphdr *sh;
287 		struct sctp_chunkhdr *ch;
288 
289 		if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
290 			goto stdreply;
291 		if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
292 		    n->m_next == NULL)
293 			goto stdreply;
294 		if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
295 		    (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
296 			goto freeit;
297 		oip = mtod(n, struct ip *);
298 		icmpelen = max(sizeof(struct sctphdr),
299 		    min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
300 		sh = mtodo(n, oiphlen);
301 		if (ntohl(sh->v_tag) == 0 &&
302 		    ntohs(oip->ip_len) >= oiphlen +
303 		    sizeof(struct sctphdr) + 8 &&
304 		    (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
305 		     n->m_next != NULL)) {
306 			if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
307 			    (n = m_pullup(n, oiphlen +
308 			    sizeof(struct sctphdr) + 8)) == NULL)
309 				goto freeit;
310 			oip = mtod(n, struct ip *);
311 			sh = mtodo(n, oiphlen);
312 			ch = (struct sctp_chunkhdr *)(sh + 1);
313 			if (ch->chunk_type == SCTP_INITIATION) {
314 				icmpelen = max(sizeof(struct sctphdr) + 8,
315 				    min(V_icmp_quotelen, ntohs(oip->ip_len) -
316 				    oiphlen));
317 			}
318 		}
319 	} else
320 stdreply:	icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
321 		    oiphlen));
322 
323 	icmplen = min(oiphlen + icmpelen, nlen);
324 	if (icmplen < sizeof(struct ip))
325 		goto freeit;
326 
327 	if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
328 		m = m_gethdr(M_NOWAIT, MT_DATA);
329 	else
330 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
331 	if (m == NULL)
332 		goto freeit;
333 #ifdef MAC
334 	mac_netinet_icmp_reply(n, m);
335 #endif
336 	icmplen = min(icmplen, M_TRAILINGSPACE(m) -
337 	    sizeof(struct ip) - ICMP_MINLEN);
338 	m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
339 	m->m_data += sizeof(struct ip);
340 	m->m_len = ICMP_MINLEN + icmplen;
341 
342 	/* XXX MRT  make the outgoing packet use the same FIB
343 	 * that was associated with the incoming packet
344 	 */
345 	M_SETFIB(m, M_GETFIB(n));
346 	icp = mtod(m, struct icmp *);
347 	ICMPSTAT_INC(icps_outhist[type]);
348 	icp->icmp_type = type;
349 	if (type == ICMP_REDIRECT)
350 		icp->icmp_gwaddr.s_addr = dest;
351 	else {
352 		icp->icmp_void = 0;
353 		/*
354 		 * The following assignments assume an overlay with the
355 		 * just zeroed icmp_void field.
356 		 */
357 		if (type == ICMP_PARAMPROB) {
358 			icp->icmp_pptr = code;
359 			code = 0;
360 		} else if (type == ICMP_UNREACH &&
361 			code == ICMP_UNREACH_NEEDFRAG && mtu) {
362 			icp->icmp_nextmtu = htons(mtu);
363 		}
364 	}
365 	icp->icmp_code = code;
366 
367 	/*
368 	 * Copy the quotation into ICMP message and
369 	 * convert quoted IP header back to network representation.
370 	 */
371 	m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
372 	nip = &icp->icmp_ip;
373 
374 	/*
375 	 * Set up ICMP message mbuf and copy old IP header (without options
376 	 * in front of ICMP message.
377 	 * If the original mbuf was meant to bypass the firewall, the error
378 	 * reply should bypass as well.
379 	 */
380 	m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
381 	KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
382 	    ("insufficient space for ip header"));
383 	m->m_data -= sizeof(struct ip);
384 	m->m_len += sizeof(struct ip);
385 	m->m_pkthdr.len = m->m_len;
386 	m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
387 	nip = mtod(m, struct ip *);
388 	bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
389 	nip->ip_len = htons(m->m_len);
390 	nip->ip_v = IPVERSION;
391 	nip->ip_hl = 5;
392 	nip->ip_p = IPPROTO_ICMP;
393 	nip->ip_tos = 0;
394 	nip->ip_off = 0;
395 
396 	if (V_error_keeptags)
397 		m_tag_copy_chain(m, n, M_NOWAIT);
398 
399 	icmp_reflect(m);
400 
401 freeit:
402 	m_freem(n);
403 }
404 
405 int
406 icmp_errmap(const struct icmp *icp)
407 {
408 
409 	switch (icp->icmp_type) {
410 	case ICMP_UNREACH:
411 		switch (icp->icmp_code) {
412 		case ICMP_UNREACH_NET:
413 		case ICMP_UNREACH_HOST:
414 		case ICMP_UNREACH_SRCFAIL:
415 		case ICMP_UNREACH_NET_UNKNOWN:
416 		case ICMP_UNREACH_HOST_UNKNOWN:
417 		case ICMP_UNREACH_ISOLATED:
418 		case ICMP_UNREACH_TOSNET:
419 		case ICMP_UNREACH_TOSHOST:
420 		case ICMP_UNREACH_HOST_PRECEDENCE:
421 		case ICMP_UNREACH_PRECEDENCE_CUTOFF:
422 			return (EHOSTUNREACH);
423 		case ICMP_UNREACH_NEEDFRAG:
424 			return (EMSGSIZE);
425 		case ICMP_UNREACH_PROTOCOL:
426 		case ICMP_UNREACH_PORT:
427 		case ICMP_UNREACH_NET_PROHIB:
428 		case ICMP_UNREACH_HOST_PROHIB:
429 		case ICMP_UNREACH_FILTER_PROHIB:
430 			return (ECONNREFUSED);
431 		default:
432 			return (0);
433 		}
434 	case ICMP_TIMXCEED:
435 		switch (icp->icmp_code) {
436 		case ICMP_TIMXCEED_INTRANS:
437 			return (EHOSTUNREACH);
438 		default:
439 			return (0);
440 		}
441 	case ICMP_PARAMPROB:
442 		switch (icp->icmp_code) {
443 		case ICMP_PARAMPROB_ERRATPTR:
444 		case ICMP_PARAMPROB_OPTABSENT:
445 			return (ENOPROTOOPT);
446 		default:
447 			return (0);
448 		}
449 	default:
450 		return (0);
451 	}
452 }
453 
454 /*
455  * Process a received ICMP message.
456  */
457 int
458 icmp_input(struct mbuf **mp, int *offp, int proto)
459 {
460 	struct icmp *icp;
461 	struct in_ifaddr *ia;
462 	struct mbuf *m = *mp;
463 	struct ip *ip = mtod(m, struct ip *);
464 	struct sockaddr_in icmpsrc, icmpdst, icmpgw;
465 	int hlen = *offp;
466 	int icmplen = ntohs(ip->ip_len) - *offp;
467 	int i, code;
468 	int fibnum;
469 
470 	NET_EPOCH_ASSERT();
471 
472 	*mp = NULL;
473 
474 	/*
475 	 * Locate icmp structure in mbuf, and check
476 	 * that not corrupted and of at least minimum length.
477 	 */
478 #ifdef ICMPPRINTFS
479 	if (icmpprintfs) {
480 		char srcbuf[INET_ADDRSTRLEN];
481 		char dstbuf[INET_ADDRSTRLEN];
482 
483 		printf("icmp_input from %s to %s, len %d\n",
484 		    inet_ntoa_r(ip->ip_src, srcbuf),
485 		    inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
486 	}
487 #endif
488 	if (icmplen < ICMP_MINLEN) {
489 		ICMPSTAT_INC(icps_tooshort);
490 		goto freeit;
491 	}
492 	i = hlen + min(icmplen, ICMP_ADVLENMIN);
493 	if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
494 		ICMPSTAT_INC(icps_tooshort);
495 		return (IPPROTO_DONE);
496 	}
497 	ip = mtod(m, struct ip *);
498 	m->m_len -= hlen;
499 	m->m_data += hlen;
500 	icp = mtod(m, struct icmp *);
501 	if (in_cksum(m, icmplen)) {
502 		ICMPSTAT_INC(icps_checksum);
503 		goto freeit;
504 	}
505 	m->m_len += hlen;
506 	m->m_data -= hlen;
507 
508 #ifdef ICMPPRINTFS
509 	if (icmpprintfs)
510 		printf("icmp_input, type %d code %d\n", icp->icmp_type,
511 		    icp->icmp_code);
512 #endif
513 
514 	/*
515 	 * Message type specific processing.
516 	 */
517 	if (icp->icmp_type > ICMP_MAXTYPE)
518 		goto raw;
519 
520 	/* Initialize */
521 	bzero(&icmpsrc, sizeof(icmpsrc));
522 	icmpsrc.sin_len = sizeof(struct sockaddr_in);
523 	icmpsrc.sin_family = AF_INET;
524 	bzero(&icmpdst, sizeof(icmpdst));
525 	icmpdst.sin_len = sizeof(struct sockaddr_in);
526 	icmpdst.sin_family = AF_INET;
527 	bzero(&icmpgw, sizeof(icmpgw));
528 	icmpgw.sin_len = sizeof(struct sockaddr_in);
529 	icmpgw.sin_family = AF_INET;
530 
531 	ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
532 	code = icp->icmp_code;
533 	switch (icp->icmp_type) {
534 	case ICMP_UNREACH:
535 		if (code > ICMP_UNREACH_PRECEDENCE_CUTOFF)
536 			goto badcode;
537 		else
538 			goto deliver;
539 
540 	case ICMP_TIMXCEED:
541 		if (code > ICMP_TIMXCEED_REASS)
542 			goto badcode;
543 		else
544 			goto deliver;
545 
546 	case ICMP_PARAMPROB:
547 		if (code > ICMP_PARAMPROB_LENGTH)
548 			goto badcode;
549 
550 	deliver:
551 		/*
552 		 * Problem with datagram; advise higher level routines.
553 		 */
554 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
555 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
556 			ICMPSTAT_INC(icps_badlen);
557 			goto freeit;
558 		}
559 		/* Discard ICMP's in response to multicast packets */
560 		if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
561 			goto badcode;
562 		/* Filter out responses to INADDR_ANY, protocols ignore it. */
563 		if (icp->icmp_ip.ip_dst.s_addr == INADDR_ANY ||
564 		    icp->icmp_ip.ip_src.s_addr == INADDR_ANY)
565 			goto freeit;
566 #ifdef ICMPPRINTFS
567 		if (icmpprintfs)
568 			printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
569 #endif
570 		/*
571 		 * XXX if the packet contains [IPv4 AH TCP], we can't make a
572 		 * notification to TCP layer.
573 		 */
574 		i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
575 		ip_stripoptions(m);
576 		if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
577 			/* This should actually not happen */
578 			ICMPSTAT_INC(icps_tooshort);
579 			return (IPPROTO_DONE);
580 		}
581 		ip = mtod(m, struct ip *);
582 		icp = (struct icmp *)(ip + 1);
583 		/*
584 		 * The upper layer handler can rely on:
585 		 * - The outer IP header has no options.
586 		 * - The outer IP header, the ICMP header, the inner IP header,
587 		 *   and the first n bytes of the inner payload are contiguous.
588 		 *   n is at least 8, but might be larger based on
589 		 *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
590 		 */
591 		if (ip_ctlprotox[icp->icmp_ip.ip_p] != NULL)
592 			ip_ctlprotox[icp->icmp_ip.ip_p](icp);
593 		break;
594 
595 	badcode:
596 		ICMPSTAT_INC(icps_badcode);
597 		break;
598 
599 	case ICMP_ECHO:
600 		if (!V_icmpbmcastecho
601 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
602 			ICMPSTAT_INC(icps_bmcastecho);
603 			break;
604 		}
605 		if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
606 			goto freeit;
607 		icp->icmp_type = ICMP_ECHOREPLY;
608 		goto reflect;
609 
610 	case ICMP_TSTAMP:
611 		if (V_icmptstamprepl == 0)
612 			break;
613 		if (!V_icmpbmcastecho
614 		    && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
615 			ICMPSTAT_INC(icps_bmcasttstamp);
616 			break;
617 		}
618 		if (icmplen < ICMP_TSLEN) {
619 			ICMPSTAT_INC(icps_badlen);
620 			break;
621 		}
622 		if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
623 			goto freeit;
624 		icp->icmp_type = ICMP_TSTAMPREPLY;
625 		icp->icmp_rtime = iptime();
626 		icp->icmp_ttime = icp->icmp_rtime;	/* bogus, do later! */
627 		goto reflect;
628 
629 	case ICMP_MASKREQ:
630 		if (V_icmpmaskrepl == 0)
631 			break;
632 		/*
633 		 * We are not able to respond with all ones broadcast
634 		 * unless we receive it over a point-to-point interface.
635 		 */
636 		if (icmplen < ICMP_MASKLEN)
637 			break;
638 		switch (ip->ip_dst.s_addr) {
639 		case INADDR_BROADCAST:
640 		case INADDR_ANY:
641 			icmpdst.sin_addr = ip->ip_src;
642 			break;
643 
644 		default:
645 			icmpdst.sin_addr = ip->ip_dst;
646 		}
647 		ia = (struct in_ifaddr *)ifaof_ifpforaddr(
648 			    (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
649 		if (ia == NULL)
650 			break;
651 		if (ia->ia_ifp == NULL)
652 			break;
653 		icp->icmp_type = ICMP_MASKREPLY;
654 		if (V_icmpmaskfake == 0)
655 			icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
656 		else
657 			icp->icmp_mask = V_icmpmaskfake;
658 		if (ip->ip_src.s_addr == 0) {
659 			if (ia->ia_ifp->if_flags & IFF_BROADCAST)
660 			    ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
661 			else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
662 			    ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
663 		}
664 reflect:
665 		ICMPSTAT_INC(icps_reflect);
666 		ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
667 		icmp_reflect(m);
668 		return (IPPROTO_DONE);
669 
670 	case ICMP_REDIRECT:
671 		if (V_log_redirect) {
672 			u_long src, dst, gw;
673 
674 			src = ntohl(ip->ip_src.s_addr);
675 			dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
676 			gw = ntohl(icp->icmp_gwaddr.s_addr);
677 			printf("icmp redirect from %d.%d.%d.%d: "
678 			       "%d.%d.%d.%d => %d.%d.%d.%d\n",
679 			       (int)(src >> 24), (int)((src >> 16) & 0xff),
680 			       (int)((src >> 8) & 0xff), (int)(src & 0xff),
681 			       (int)(dst >> 24), (int)((dst >> 16) & 0xff),
682 			       (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
683 			       (int)(gw >> 24), (int)((gw >> 16) & 0xff),
684 			       (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
685 		}
686 		/*
687 		 * RFC1812 says we must ignore ICMP redirects if we
688 		 * are acting as router.
689 		 */
690 		if (V_drop_redirect || V_ipforwarding)
691 			break;
692 		if (code > 3)
693 			goto badcode;
694 		if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
695 		    icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
696 			ICMPSTAT_INC(icps_badlen);
697 			break;
698 		}
699 		/*
700 		 * Short circuit routing redirects to force
701 		 * immediate change in the kernel's routing
702 		 * tables.  The message is also handed to anyone
703 		 * listening on a raw socket (e.g. the routing
704 		 * daemon for use in updating its tables).
705 		 */
706 		icmpgw.sin_addr = ip->ip_src;
707 		icmpdst.sin_addr = icp->icmp_gwaddr;
708 #ifdef	ICMPPRINTFS
709 		if (icmpprintfs) {
710 			char dstbuf[INET_ADDRSTRLEN];
711 			char gwbuf[INET_ADDRSTRLEN];
712 
713 			printf("redirect dst %s to %s\n",
714 			       inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
715 			       inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
716 		}
717 #endif
718 		icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
719 
720 		/*
721 		 * RFC 1122 says network (code 0,2) redirects SHOULD
722 		 * be treated identically to the host redirects.
723 		 * Given that, ignore network masks.
724 		 */
725 
726 		/*
727 		 * Variable values:
728 		 * icmpsrc: route destination
729 		 * icmpdst: route gateway
730 		 * icmpgw: message source
731 		 */
732 
733 		if (icmp_verify_redirect_gateway(&icmpgw, &icmpsrc, &icmpdst,
734 		    M_GETFIB(m)) != 0) {
735 			/* TODO: increment bad redirects here */
736 			break;
737 		}
738 
739 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
740 			rib_add_redirect(fibnum, (struct sockaddr *)&icmpsrc,
741 			    (struct sockaddr *)&icmpdst,
742 			    (struct sockaddr *)&icmpgw, m->m_pkthdr.rcvif,
743 			    RTF_GATEWAY, V_redirtimeout);
744 		}
745 		break;
746 
747 	/*
748 	 * No kernel processing for the following;
749 	 * just fall through to send to raw listener.
750 	 */
751 	case ICMP_ECHOREPLY:
752 	case ICMP_ROUTERADVERT:
753 	case ICMP_ROUTERSOLICIT:
754 	case ICMP_TSTAMPREPLY:
755 	case ICMP_IREQREPLY:
756 	case ICMP_MASKREPLY:
757 	case ICMP_SOURCEQUENCH:
758 	default:
759 		break;
760 	}
761 
762 raw:
763 	*mp = m;
764 	rip_input(mp, offp, proto);
765 	return (IPPROTO_DONE);
766 
767 freeit:
768 	m_freem(m);
769 	return (IPPROTO_DONE);
770 }
771 
772 /*
773  * Reflect the ip packet back to the source
774  */
775 static void
776 icmp_reflect(struct mbuf *m)
777 {
778 	struct ip *ip = mtod(m, struct ip *);
779 	struct ifaddr *ifa;
780 	struct ifnet *ifp;
781 	struct in_ifaddr *ia;
782 	struct in_addr t;
783 	struct nhop_object *nh;
784 	struct mbuf *opts = NULL;
785 	int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
786 
787 	NET_EPOCH_ASSERT();
788 
789 	if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
790 	    (IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) && !V_ip_allow_net240) ||
791 	    (IN_ZERONET(ntohl(ip->ip_src.s_addr)) && !V_ip_allow_net0) ) {
792 		m_freem(m);	/* Bad return address */
793 		ICMPSTAT_INC(icps_badaddr);
794 		goto done;	/* Ip_output() will check for broadcast */
795 	}
796 
797 	t = ip->ip_dst;
798 	ip->ip_dst = ip->ip_src;
799 
800 	/*
801 	 * Source selection for ICMP replies:
802 	 *
803 	 * If the incoming packet was addressed directly to one of our
804 	 * own addresses, use dst as the src for the reply.
805 	 */
806 	CK_LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
807 		if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
808 			t = IA_SIN(ia)->sin_addr;
809 			goto match;
810 		}
811 	}
812 
813 	/*
814 	 * If the incoming packet was addressed to one of our broadcast
815 	 * addresses, use the first non-broadcast address which corresponds
816 	 * to the incoming interface.
817 	 */
818 	ifp = m->m_pkthdr.rcvif;
819 	if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
820 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
821 			if (ifa->ifa_addr->sa_family != AF_INET)
822 				continue;
823 			ia = ifatoia(ifa);
824 			if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
825 			    t.s_addr) {
826 				t = IA_SIN(ia)->sin_addr;
827 				goto match;
828 			}
829 		}
830 	}
831 	/*
832 	 * If the packet was transiting through us, use the address of
833 	 * the interface the packet came through in.  If that interface
834 	 * doesn't have a suitable IP address, the normal selection
835 	 * criteria apply.
836 	 */
837 	if (V_icmp_rfi && ifp != NULL) {
838 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
839 			if (ifa->ifa_addr->sa_family != AF_INET)
840 				continue;
841 			ia = ifatoia(ifa);
842 			t = IA_SIN(ia)->sin_addr;
843 			goto match;
844 		}
845 	}
846 	/*
847 	 * If the incoming packet was not addressed directly to us, use
848 	 * designated interface for icmp replies specified by sysctl
849 	 * net.inet.icmp.reply_src (default not set). Otherwise continue
850 	 * with normal source selection.
851 	 */
852 	if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
853 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
854 			if (ifa->ifa_addr->sa_family != AF_INET)
855 				continue;
856 			ia = ifatoia(ifa);
857 			t = IA_SIN(ia)->sin_addr;
858 			goto match;
859 		}
860 	}
861 	/*
862 	 * If the packet was transiting through us, use the address of
863 	 * the interface that is the closest to the packet source.
864 	 * When we don't have a route back to the packet source, stop here
865 	 * and drop the packet.
866 	 */
867 	nh = fib4_lookup(M_GETFIB(m), ip->ip_dst, 0, NHR_NONE, 0);
868 	if (nh == NULL) {
869 		m_freem(m);
870 		ICMPSTAT_INC(icps_noroute);
871 		goto done;
872 	}
873 	t = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr;
874 match:
875 #ifdef MAC
876 	mac_netinet_icmp_replyinplace(m);
877 #endif
878 	ip->ip_src = t;
879 	ip->ip_ttl = V_ip_defttl;
880 
881 	if (optlen > 0) {
882 		u_char *cp;
883 		int opt, cnt;
884 		u_int len;
885 
886 		/*
887 		 * Retrieve any source routing from the incoming packet;
888 		 * add on any record-route or timestamp options.
889 		 */
890 		cp = (u_char *) (ip + 1);
891 		if ((opts = ip_srcroute(m)) == NULL &&
892 		    (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
893 			opts->m_len = sizeof(struct in_addr);
894 			mtod(opts, struct in_addr *)->s_addr = 0;
895 		}
896 		if (opts) {
897 #ifdef ICMPPRINTFS
898 		    if (icmpprintfs)
899 			    printf("icmp_reflect optlen %d rt %d => ",
900 				optlen, opts->m_len);
901 #endif
902 		    for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
903 			    opt = cp[IPOPT_OPTVAL];
904 			    if (opt == IPOPT_EOL)
905 				    break;
906 			    if (opt == IPOPT_NOP)
907 				    len = 1;
908 			    else {
909 				    if (cnt < IPOPT_OLEN + sizeof(*cp))
910 					    break;
911 				    len = cp[IPOPT_OLEN];
912 				    if (len < IPOPT_OLEN + sizeof(*cp) ||
913 				        len > cnt)
914 					    break;
915 			    }
916 			    /*
917 			     * Should check for overflow, but it "can't happen"
918 			     */
919 			    if (opt == IPOPT_RR || opt == IPOPT_TS ||
920 				opt == IPOPT_SECURITY) {
921 				    bcopy((caddr_t)cp,
922 					mtod(opts, caddr_t) + opts->m_len, len);
923 				    opts->m_len += len;
924 			    }
925 		    }
926 		    /* Terminate & pad, if necessary */
927 		    cnt = opts->m_len % 4;
928 		    if (cnt) {
929 			    for (; cnt < 4; cnt++) {
930 				    *(mtod(opts, caddr_t) + opts->m_len) =
931 					IPOPT_EOL;
932 				    opts->m_len++;
933 			    }
934 		    }
935 #ifdef ICMPPRINTFS
936 		    if (icmpprintfs)
937 			    printf("%d\n", opts->m_len);
938 #endif
939 		}
940 		ip_stripoptions(m);
941 	}
942 	m_tag_delete_nonpersistent(m);
943 	m->m_flags &= ~(M_BCAST|M_MCAST);
944 	icmp_send(m, opts);
945 done:
946 	if (opts)
947 		(void)m_free(opts);
948 }
949 
950 /*
951  * Verifies if redirect message is valid, according to RFC 1122
952  *
953  * @src: sockaddr with address of redirect originator
954  * @dst: sockaddr with destination in question
955  * @gateway: new proposed gateway
956  *
957  * Returns 0 on success.
958  */
959 static int
960 icmp_verify_redirect_gateway(struct sockaddr_in *src, struct sockaddr_in *dst,
961     struct sockaddr_in *gateway, u_int fibnum)
962 {
963 	struct nhop_object *nh;
964 	struct ifaddr *ifa;
965 
966 	NET_EPOCH_ASSERT();
967 
968 	/* Verify the gateway is directly reachable. */
969 	if ((ifa = ifa_ifwithnet((struct sockaddr *)gateway, 0, fibnum))==NULL)
970 		return (ENETUNREACH);
971 
972 	/* TODO: fib-aware. */
973 	if (ifa_ifwithaddr_check((struct sockaddr *)gateway))
974 		return (EHOSTUNREACH);
975 
976 	nh = fib4_lookup(fibnum, dst->sin_addr, 0, NHR_NONE, 0);
977 	if (nh == NULL)
978 		return (EINVAL);
979 
980 	/*
981 	 * If the redirect isn't from our current router for this dst,
982 	 * it's either old or wrong.  If it redirects us to ourselves,
983 	 * we have a routing loop, perhaps as a result of an interface
984 	 * going down recently.
985 	 */
986 	if (!sa_equal((struct sockaddr *)src, &nh->gw_sa))
987 		return (EINVAL);
988 	if (nh->nh_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK)
989 		return (EINVAL);
990 
991 	/* If host route already exists, ignore redirect. */
992 	if (nh->nh_flags & NHF_HOST)
993 		return (EEXIST);
994 
995 	/* If the prefix is directly reachable, ignore redirect. */
996 	if (!(nh->nh_flags & NHF_GATEWAY))
997 		return (EEXIST);
998 
999 	return (0);
1000 }
1001 
1002 /*
1003  * Send an icmp packet back to the ip level,
1004  * after supplying a checksum.
1005  */
1006 static void
1007 icmp_send(struct mbuf *m, struct mbuf *opts)
1008 {
1009 	struct ip *ip = mtod(m, struct ip *);
1010 	int hlen;
1011 	struct icmp *icp;
1012 
1013 	hlen = ip->ip_hl << 2;
1014 	m->m_data += hlen;
1015 	m->m_len -= hlen;
1016 	icp = mtod(m, struct icmp *);
1017 	icp->icmp_cksum = 0;
1018 	icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
1019 	m->m_data -= hlen;
1020 	m->m_len += hlen;
1021 	m->m_pkthdr.rcvif = (struct ifnet *)0;
1022 #ifdef ICMPPRINTFS
1023 	if (icmpprintfs) {
1024 		char dstbuf[INET_ADDRSTRLEN];
1025 		char srcbuf[INET_ADDRSTRLEN];
1026 
1027 		printf("icmp_send dst %s src %s\n",
1028 		    inet_ntoa_r(ip->ip_dst, dstbuf),
1029 		    inet_ntoa_r(ip->ip_src, srcbuf));
1030 	}
1031 #endif
1032 	(void) ip_output(m, opts, NULL, 0, NULL, NULL);
1033 }
1034 
1035 /*
1036  * Return milliseconds since 00:00 UTC in network format.
1037  */
1038 uint32_t
1039 iptime(void)
1040 {
1041 	struct timeval atv;
1042 	u_long t;
1043 
1044 	getmicrotime(&atv);
1045 	t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
1046 	return (htonl(t));
1047 }
1048 
1049 /*
1050  * Return the next larger or smaller MTU plateau (table from RFC 1191)
1051  * given current value MTU.  If DIR is less than zero, a larger plateau
1052  * is returned; otherwise, a smaller value is returned.
1053  */
1054 int
1055 ip_next_mtu(int mtu, int dir)
1056 {
1057 	static int mtutab[] = {
1058 		65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
1059 		296, 68, 0
1060 	};
1061 	int i, size;
1062 
1063 	size = (sizeof mtutab) / (sizeof mtutab[0]);
1064 	if (dir >= 0) {
1065 		for (i = 0; i < size; i++)
1066 			if (mtu > mtutab[i])
1067 				return mtutab[i];
1068 	} else {
1069 		for (i = size - 1; i >= 0; i--)
1070 			if (mtu < mtutab[i])
1071 				return mtutab[i];
1072 		if (mtu == mtutab[0])
1073 			return mtutab[0];
1074 	}
1075 	return 0;
1076 }
1077 #endif /* INET */
1078 
1079 /*
1080  * badport_bandlim() - check for ICMP bandwidth limit
1081  *
1082  *	Return 0 if it is ok to send an ICMP error response, -1 if we have
1083  *	hit our bandwidth limit and it is not ok.
1084  *
1085  *	If icmplim is <= 0, the feature is disabled and 0 is returned.
1086  *
1087  *	For now we separate the TCP and UDP subsystems w/ different 'which'
1088  *	values.  We may eventually remove this separation (and simplify the
1089  *	code further).
1090  *
1091  *	Note that the printing of the error message is delayed so we can
1092  *	properly print the icmp error rate that the system was trying to do
1093  *	(i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1094  *	the 'final' error, but it doesn't make sense to solve the printing
1095  *	delay with more complex code.
1096  */
1097 VNET_DEFINE_STATIC(struct counter_rate, icmp_rates[BANDLIM_MAX]);
1098 #define	V_icmp_rates	VNET(icmp_rates)
1099 
1100 static const char *icmp_rate_descrs[BANDLIM_MAX] = {
1101 	[BANDLIM_ICMP_UNREACH] = "icmp unreach",
1102 	[BANDLIM_ICMP_ECHO] = "icmp ping",
1103 	[BANDLIM_ICMP_TSTAMP] = "icmp tstamp",
1104 	[BANDLIM_RST_CLOSEDPORT] = "closed port RST",
1105 	[BANDLIM_RST_OPENPORT] = "open port RST",
1106 	[BANDLIM_ICMP6_UNREACH] = "icmp6 unreach",
1107 	[BANDLIM_SCTP_OOTB] = "sctp ootb",
1108 };
1109 
1110 static void
1111 icmplim_new_jitter(void)
1112 {
1113 	/*
1114 	 * Adjust limit +/- to jitter the measurement to deny a side-channel
1115 	 * port scan as in https://dl.acm.org/doi/10.1145/3372297.3417280
1116 	 */
1117 	if (V_icmplim_jitter > 0)
1118 		V_icmplim_curr_jitter =
1119 		    arc4random_uniform(V_icmplim_jitter * 2 + 1) -
1120 		    V_icmplim_jitter;
1121 }
1122 
1123 static int
1124 sysctl_icmplim_and_jitter(SYSCTL_HANDLER_ARGS)
1125 {
1126 	uint32_t new;
1127 	int error;
1128 	bool lim;
1129 
1130 	MPASS(oidp->oid_arg1 == &VNET_NAME(icmplim) ||
1131 	    oidp->oid_arg1 == &VNET_NAME(icmplim_jitter));
1132 
1133 	lim = (oidp->oid_arg1 == &VNET_NAME(icmplim));
1134 	new = lim ? V_icmplim : V_icmplim_jitter;
1135 	error = sysctl_handle_int(oidp, &new, 0, req);
1136 	if (error == 0 && req->newptr) {
1137 		if (lim) {
1138 			if (new != 0 && new <= V_icmplim_jitter)
1139 				error = EINVAL;
1140 			else
1141 				V_icmplim = new;
1142 		} else {
1143 			if (new >= V_icmplim)
1144 				error = EINVAL;
1145 			else {
1146 				V_icmplim_jitter = new;
1147 				icmplim_new_jitter();
1148 			}
1149 		}
1150 	}
1151 	MPASS(V_icmplim + V_icmplim_curr_jitter > 0);
1152 
1153 	return (error);
1154 }
1155 
1156 static void
1157 icmp_bandlimit_init(void)
1158 {
1159 
1160 	for (int i = 0; i < BANDLIM_MAX; i++) {
1161 		V_icmp_rates[i].cr_rate = counter_u64_alloc(M_WAITOK);
1162 		V_icmp_rates[i].cr_ticks = ticks;
1163 	}
1164 	icmplim_new_jitter();
1165 }
1166 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1167     icmp_bandlimit_init, NULL);
1168 
1169 #ifdef VIMAGE
1170 static void
1171 icmp_bandlimit_uninit(void)
1172 {
1173 
1174 	for (int i = 0; i < BANDLIM_MAX; i++)
1175 		counter_u64_free(V_icmp_rates[i].cr_rate);
1176 }
1177 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1178     icmp_bandlimit_uninit, NULL);
1179 #endif
1180 
1181 int
1182 badport_bandlim(int which)
1183 {
1184 	int64_t pps;
1185 
1186 	if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1187 		return (0);
1188 
1189 	KASSERT(which >= 0 && which < BANDLIM_MAX,
1190 	    ("%s: which %d", __func__, which));
1191 
1192 	pps = counter_ratecheck(&V_icmp_rates[which], V_icmplim +
1193 	    V_icmplim_curr_jitter);
1194 	if (pps > 0) {
1195 		if (V_icmplim_output)
1196 			log(LOG_NOTICE,
1197 			    "Limiting %s response from %jd to %d packets/sec\n",
1198 			    icmp_rate_descrs[which], (intmax_t )pps,
1199 			    V_icmplim + V_icmplim_curr_jitter);
1200 		icmplim_new_jitter();
1201 	}
1202 	if (pps == -1)
1203 		return (-1);
1204 	return (0);
1205 }
1206