]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netinet/ip_icmp.c
Merge OpenSSL 1.0.2g.
[FreeBSD/FreeBSD.git] / sys / netinet / ip_icmp.c
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *      @(#)ip_icmp.c   8.2 (Berkeley) 1/4/94
30  */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_inet.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mbuf.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/time.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/rmlock.h>
46 #include <sys/sysctl.h>
47 #include <sys/syslog.h>
48
49 #include <net/if.h>
50 #include <net/if_var.h>
51 #include <net/if_types.h>
52 #include <net/route.h>
53 #include <net/vnet.h>
54
55 #include <netinet/in.h>
56 #include <netinet/in_fib.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_icmp.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/ip_options.h>
64 #include <netinet/tcp.h>
65 #include <netinet/tcp_var.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 /*
77  * ICMP routines: error generation, receive packet processing, and
78  * routines to turnaround packets back to the originator, and
79  * host table maintenance routines.
80  */
81 static VNET_DEFINE(int, icmplim) = 200;
82 #define V_icmplim                       VNET(icmplim)
83 SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | CTLFLAG_RW,
84         &VNET_NAME(icmplim), 0,
85         "Maximum number of ICMP responses per second");
86
87 static VNET_DEFINE(int, icmplim_output) = 1;
88 #define V_icmplim_output                VNET(icmplim_output)
89 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
90         &VNET_NAME(icmplim_output), 0,
91         "Enable logging of ICMP response rate limiting");
92
93 #ifdef INET
94 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
95 VNET_PCPUSTAT_SYSINIT(icmpstat);
96 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
97     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
98
99 #ifdef VIMAGE
100 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
101 #endif /* VIMAGE */
102
103 static VNET_DEFINE(int, icmpmaskrepl) = 0;
104 #define V_icmpmaskrepl                  VNET(icmpmaskrepl)
105 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
106         &VNET_NAME(icmpmaskrepl), 0,
107         "Reply to ICMP Address Mask Request packets.");
108
109 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
110 #define V_icmpmaskfake                  VNET(icmpmaskfake)
111 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
112         &VNET_NAME(icmpmaskfake), 0,
113         "Fake reply to ICMP Address Mask Request packets.");
114
115 VNET_DEFINE(int, drop_redirect) = 0;
116 #define V_drop_redirect                 VNET(drop_redirect)
117 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
118         &VNET_NAME(drop_redirect), 0, "Ignore ICMP redirects");
119
120 static VNET_DEFINE(int, log_redirect) = 0;
121 #define V_log_redirect                  VNET(log_redirect)
122 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
123         &VNET_NAME(log_redirect), 0,
124         "Log ICMP redirects to the console");
125
126 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
127 #define V_reply_src                     VNET(reply_src)
128 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
129         &VNET_NAME(reply_src), IFNAMSIZ,
130         "icmp reply source for non-local packets.");
131
132 static VNET_DEFINE(int, icmp_rfi) = 0;
133 #define V_icmp_rfi                      VNET(icmp_rfi)
134 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
135         &VNET_NAME(icmp_rfi), 0,
136         "ICMP reply from incoming interface for non-local packets");
137
138 static VNET_DEFINE(int, icmp_quotelen) = 8;
139 #define V_icmp_quotelen                 VNET(icmp_quotelen)
140 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
141         &VNET_NAME(icmp_quotelen), 0,
142         "Number of bytes from original packet to quote in ICMP reply");
143
144 /*
145  * ICMP broadcast echo sysctl
146  */
147 static VNET_DEFINE(int, icmpbmcastecho) = 0;
148 #define V_icmpbmcastecho                VNET(icmpbmcastecho)
149 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
150         &VNET_NAME(icmpbmcastecho), 0,
151         "");
152
153 static VNET_DEFINE(int, icmptstamprepl) = 1;
154 #define V_icmptstamprepl                VNET(icmptstamprepl)
155 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
156         &VNET_NAME(icmptstamprepl), 0, "Respond to ICMP Timestamp packets");
157
158 #ifdef ICMPPRINTFS
159 int     icmpprintfs = 0;
160 #endif
161
162 static void     icmp_reflect(struct mbuf *);
163 static void     icmp_send(struct mbuf *, struct mbuf *);
164
165 extern  struct protosw inetsw[];
166
167 /*
168  * Kernel module interface for updating icmpstat.  The argument is an index
169  * into icmpstat treated as an array of u_long.  While this encodes the
170  * general layout of icmpstat into the caller, it doesn't encode its
171  * location, so that future changes to add, for example, per-CPU stats
172  * support won't cause binary compatibility problems for kernel modules.
173  */
174 void
175 kmod_icmpstat_inc(int statnum)
176 {
177
178         counter_u64_add(VNET(icmpstat)[statnum], 1);
179 }
180
181 /*
182  * Generate an error packet of type error
183  * in response to bad packet ip.
184  */
185 void
186 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
187 {
188         register struct ip *oip = mtod(n, struct ip *), *nip;
189         register unsigned oiphlen = oip->ip_hl << 2;
190         register struct icmp *icp;
191         register struct mbuf *m;
192         unsigned icmplen, icmpelen, nlen;
193
194         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
195 #ifdef ICMPPRINTFS
196         if (icmpprintfs)
197                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
198 #endif
199         if (type != ICMP_REDIRECT)
200                 ICMPSTAT_INC(icps_error);
201         /*
202          * Don't send error:
203          *  if the original packet was encrypted.
204          *  if not the first fragment of message.
205          *  in response to a multicast or broadcast packet.
206          *  if the old packet protocol was an ICMP error message.
207          */
208         if (n->m_flags & M_DECRYPTED)
209                 goto freeit;
210         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
211                 goto freeit;
212         if (n->m_flags & (M_BCAST|M_MCAST))
213                 goto freeit;
214         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
215           n->m_len >= oiphlen + ICMP_MINLEN &&
216           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
217                 ICMPSTAT_INC(icps_oldicmp);
218                 goto freeit;
219         }
220         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
221         if (oiphlen + 8 > n->m_len)
222                 goto freeit;
223         /*
224          * Calculate length to quote from original packet and
225          * prevent the ICMP mbuf from overflowing.
226          * Unfortunatly this is non-trivial since ip_forward()
227          * sends us truncated packets.
228          */
229         nlen = m_length(n, NULL);
230         if (oip->ip_p == IPPROTO_TCP) {
231                 struct tcphdr *th;
232                 int tcphlen;
233
234                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
235                     n->m_next == NULL)
236                         goto stdreply;
237                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
238                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
239                         goto freeit;
240                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
241                 tcphlen = th->th_off << 2;
242                 if (tcphlen < sizeof(struct tcphdr))
243                         goto freeit;
244                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
245                         goto freeit;
246                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
247                         goto stdreply;
248                 if (n->m_len < oiphlen + tcphlen && 
249                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
250                         goto freeit;
251                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
252                     ntohs(oip->ip_len) - oiphlen));
253         } else
254 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
255
256         icmplen = min(oiphlen + icmpelen, nlen);
257         if (icmplen < sizeof(struct ip))
258                 goto freeit;
259
260         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
261                 m = m_gethdr(M_NOWAIT, MT_DATA);
262         else
263                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
264         if (m == NULL)
265                 goto freeit;
266 #ifdef MAC
267         mac_netinet_icmp_reply(n, m);
268 #endif
269         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
270         m_align(m, ICMP_MINLEN + icmplen);
271         m->m_len = ICMP_MINLEN + icmplen;
272
273         /* XXX MRT  make the outgoing packet use the same FIB
274          * that was associated with the incoming packet
275          */
276         M_SETFIB(m, M_GETFIB(n));
277         icp = mtod(m, struct icmp *);
278         ICMPSTAT_INC(icps_outhist[type]);
279         icp->icmp_type = type;
280         if (type == ICMP_REDIRECT)
281                 icp->icmp_gwaddr.s_addr = dest;
282         else {
283                 icp->icmp_void = 0;
284                 /*
285                  * The following assignments assume an overlay with the
286                  * just zeroed icmp_void field.
287                  */
288                 if (type == ICMP_PARAMPROB) {
289                         icp->icmp_pptr = code;
290                         code = 0;
291                 } else if (type == ICMP_UNREACH &&
292                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
293                         icp->icmp_nextmtu = htons(mtu);
294                 }
295         }
296         icp->icmp_code = code;
297
298         /*
299          * Copy the quotation into ICMP message and
300          * convert quoted IP header back to network representation.
301          */
302         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
303         nip = &icp->icmp_ip;
304
305         /*
306          * Set up ICMP message mbuf and copy old IP header (without options
307          * in front of ICMP message.
308          * If the original mbuf was meant to bypass the firewall, the error
309          * reply should bypass as well.
310          */
311         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
312         m->m_data -= sizeof(struct ip);
313         m->m_len += sizeof(struct ip);
314         m->m_pkthdr.len = m->m_len;
315         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
316         nip = mtod(m, struct ip *);
317         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
318         nip->ip_len = htons(m->m_len);
319         nip->ip_v = IPVERSION;
320         nip->ip_hl = 5;
321         nip->ip_p = IPPROTO_ICMP;
322         nip->ip_tos = 0;
323         nip->ip_off = 0;
324         icmp_reflect(m);
325
326 freeit:
327         m_freem(n);
328 }
329
330 /*
331  * Process a received ICMP message.
332  */
333 int
334 icmp_input(struct mbuf **mp, int *offp, int proto)
335 {
336         struct icmp *icp;
337         struct in_ifaddr *ia;
338         struct mbuf *m = *mp;
339         struct ip *ip = mtod(m, struct ip *);
340         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
341         int hlen = *offp;
342         int icmplen = ntohs(ip->ip_len) - *offp;
343         int i, code;
344         void (*ctlfunc)(int, struct sockaddr *, void *);
345         int fibnum;
346
347         *mp = NULL;
348
349         /*
350          * Locate icmp structure in mbuf, and check
351          * that not corrupted and of at least minimum length.
352          */
353 #ifdef ICMPPRINTFS
354         if (icmpprintfs) {
355                 char buf[4 * sizeof "123"];
356                 strcpy(buf, inet_ntoa(ip->ip_src));
357                 printf("icmp_input from %s to %s, len %d\n",
358                        buf, inet_ntoa(ip->ip_dst), icmplen);
359         }
360 #endif
361         if (icmplen < ICMP_MINLEN) {
362                 ICMPSTAT_INC(icps_tooshort);
363                 goto freeit;
364         }
365         i = hlen + min(icmplen, ICMP_ADVLENMIN);
366         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
367                 ICMPSTAT_INC(icps_tooshort);
368                 return (IPPROTO_DONE);
369         }
370         ip = mtod(m, struct ip *);
371         m->m_len -= hlen;
372         m->m_data += hlen;
373         icp = mtod(m, struct icmp *);
374         if (in_cksum(m, icmplen)) {
375                 ICMPSTAT_INC(icps_checksum);
376                 goto freeit;
377         }
378         m->m_len += hlen;
379         m->m_data -= hlen;
380
381 #ifdef ICMPPRINTFS
382         if (icmpprintfs)
383                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
384                     icp->icmp_code);
385 #endif
386
387         /*
388          * Message type specific processing.
389          */
390         if (icp->icmp_type > ICMP_MAXTYPE)
391                 goto raw;
392
393         /* Initialize */
394         bzero(&icmpsrc, sizeof(icmpsrc));
395         icmpsrc.sin_len = sizeof(struct sockaddr_in);
396         icmpsrc.sin_family = AF_INET;
397         bzero(&icmpdst, sizeof(icmpdst));
398         icmpdst.sin_len = sizeof(struct sockaddr_in);
399         icmpdst.sin_family = AF_INET;
400         bzero(&icmpgw, sizeof(icmpgw));
401         icmpgw.sin_len = sizeof(struct sockaddr_in);
402         icmpgw.sin_family = AF_INET;
403
404         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
405         code = icp->icmp_code;
406         switch (icp->icmp_type) {
407
408         case ICMP_UNREACH:
409                 switch (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                                 code = PRC_UNREACH_NET;
421                                 break;
422
423                         case ICMP_UNREACH_NEEDFRAG:
424                                 code = PRC_MSGSIZE;
425                                 break;
426
427                         /*
428                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
429                          * Treat subcodes 2,3 as immediate RST
430                          */
431                         case ICMP_UNREACH_PROTOCOL:
432                         case ICMP_UNREACH_PORT:
433                                 code = PRC_UNREACH_PORT;
434                                 break;
435
436                         case ICMP_UNREACH_NET_PROHIB:
437                         case ICMP_UNREACH_HOST_PROHIB:
438                         case ICMP_UNREACH_FILTER_PROHIB:
439                                 code = PRC_UNREACH_ADMIN_PROHIB;
440                                 break;
441
442                         default:
443                                 goto badcode;
444                 }
445                 goto deliver;
446
447         case ICMP_TIMXCEED:
448                 if (code > 1)
449                         goto badcode;
450                 code += PRC_TIMXCEED_INTRANS;
451                 goto deliver;
452
453         case ICMP_PARAMPROB:
454                 if (code > 1)
455                         goto badcode;
456                 code = PRC_PARAMPROB;
457         deliver:
458                 /*
459                  * Problem with datagram; advise higher level routines.
460                  */
461                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
462                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
463                         ICMPSTAT_INC(icps_badlen);
464                         goto freeit;
465                 }
466                 /* Discard ICMP's in response to multicast packets */
467                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
468                         goto badcode;
469 #ifdef ICMPPRINTFS
470                 if (icmpprintfs)
471                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
472 #endif
473                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
474                 /*
475                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
476                  * notification to TCP layer.
477                  */
478                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
479                 if (ctlfunc)
480                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
481                                    (void *)&icp->icmp_ip);
482                 break;
483
484         badcode:
485                 ICMPSTAT_INC(icps_badcode);
486                 break;
487
488         case ICMP_ECHO:
489                 if (!V_icmpbmcastecho
490                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
491                         ICMPSTAT_INC(icps_bmcastecho);
492                         break;
493                 }
494                 icp->icmp_type = ICMP_ECHOREPLY;
495                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
496                         goto freeit;
497                 else
498                         goto reflect;
499
500         case ICMP_TSTAMP:
501                 if (V_icmptstamprepl == 0)
502                         break;
503                 if (!V_icmpbmcastecho
504                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
505                         ICMPSTAT_INC(icps_bmcasttstamp);
506                         break;
507                 }
508                 if (icmplen < ICMP_TSLEN) {
509                         ICMPSTAT_INC(icps_badlen);
510                         break;
511                 }
512                 icp->icmp_type = ICMP_TSTAMPREPLY;
513                 icp->icmp_rtime = iptime();
514                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
515                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
516                         goto freeit;
517                 else
518                         goto reflect;
519
520         case ICMP_MASKREQ:
521                 if (V_icmpmaskrepl == 0)
522                         break;
523                 /*
524                  * We are not able to respond with all ones broadcast
525                  * unless we receive it over a point-to-point interface.
526                  */
527                 if (icmplen < ICMP_MASKLEN)
528                         break;
529                 switch (ip->ip_dst.s_addr) {
530
531                 case INADDR_BROADCAST:
532                 case INADDR_ANY:
533                         icmpdst.sin_addr = ip->ip_src;
534                         break;
535
536                 default:
537                         icmpdst.sin_addr = ip->ip_dst;
538                 }
539                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
540                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
541                 if (ia == NULL)
542                         break;
543                 if (ia->ia_ifp == NULL) {
544                         ifa_free(&ia->ia_ifa);
545                         break;
546                 }
547                 icp->icmp_type = ICMP_MASKREPLY;
548                 if (V_icmpmaskfake == 0)
549                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
550                 else
551                         icp->icmp_mask = V_icmpmaskfake;
552                 if (ip->ip_src.s_addr == 0) {
553                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
554                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
555                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
556                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
557                 }
558                 ifa_free(&ia->ia_ifa);
559 reflect:
560                 ICMPSTAT_INC(icps_reflect);
561                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
562                 icmp_reflect(m);
563                 return (IPPROTO_DONE);
564
565         case ICMP_REDIRECT:
566                 if (V_log_redirect) {
567                         u_long src, dst, gw;
568
569                         src = ntohl(ip->ip_src.s_addr);
570                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
571                         gw = ntohl(icp->icmp_gwaddr.s_addr);
572                         printf("icmp redirect from %d.%d.%d.%d: "
573                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
574                                (int)(src >> 24), (int)((src >> 16) & 0xff),
575                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
576                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
577                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
578                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
579                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
580                 }
581                 /*
582                  * RFC1812 says we must ignore ICMP redirects if we
583                  * are acting as router.
584                  */
585                 if (V_drop_redirect || V_ipforwarding)
586                         break;
587                 if (code > 3)
588                         goto badcode;
589                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
590                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
591                         ICMPSTAT_INC(icps_badlen);
592                         break;
593                 }
594                 /*
595                  * Short circuit routing redirects to force
596                  * immediate change in the kernel's routing
597                  * tables.  The message is also handed to anyone
598                  * listening on a raw socket (e.g. the routing
599                  * daemon for use in updating its tables).
600                  */
601                 icmpgw.sin_addr = ip->ip_src;
602                 icmpdst.sin_addr = icp->icmp_gwaddr;
603 #ifdef  ICMPPRINTFS
604                 if (icmpprintfs) {
605                         char buf[4 * sizeof "123"];
606                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
607
608                         printf("redirect dst %s to %s\n",
609                                buf, inet_ntoa(icp->icmp_gwaddr));
610                 }
611 #endif
612                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
613                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
614                         in_rtredirect((struct sockaddr *)&icmpsrc,
615                           (struct sockaddr *)&icmpdst,
616                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
617                           (struct sockaddr *)&icmpgw, fibnum);
618                 }
619                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
620                 break;
621
622         /*
623          * No kernel processing for the following;
624          * just fall through to send to raw listener.
625          */
626         case ICMP_ECHOREPLY:
627         case ICMP_ROUTERADVERT:
628         case ICMP_ROUTERSOLICIT:
629         case ICMP_TSTAMPREPLY:
630         case ICMP_IREQREPLY:
631         case ICMP_MASKREPLY:
632         case ICMP_SOURCEQUENCH:
633         default:
634                 break;
635         }
636
637 raw:
638         *mp = m;
639         rip_input(mp, offp, proto);
640         return (IPPROTO_DONE);
641
642 freeit:
643         m_freem(m);
644         return (IPPROTO_DONE);
645 }
646
647 /*
648  * Reflect the ip packet back to the source
649  */
650 static void
651 icmp_reflect(struct mbuf *m)
652 {
653         struct rm_priotracker in_ifa_tracker;
654         struct ip *ip = mtod(m, struct ip *);
655         struct ifaddr *ifa;
656         struct ifnet *ifp;
657         struct in_ifaddr *ia;
658         struct in_addr t;
659         struct nhop4_extended nh_ext;
660         struct mbuf *opts = 0;
661         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
662
663         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
664             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
665             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
666                 m_freem(m);     /* Bad return address */
667                 ICMPSTAT_INC(icps_badaddr);
668                 goto done;      /* Ip_output() will check for broadcast */
669         }
670
671         t = ip->ip_dst;
672         ip->ip_dst = ip->ip_src;
673
674         /*
675          * Source selection for ICMP replies:
676          *
677          * If the incoming packet was addressed directly to one of our
678          * own addresses, use dst as the src for the reply.
679          */
680         IN_IFADDR_RLOCK(&in_ifa_tracker);
681         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
682                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
683                         t = IA_SIN(ia)->sin_addr;
684                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
685                         goto match;
686                 }
687         }
688         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
689
690         /*
691          * If the incoming packet was addressed to one of our broadcast
692          * addresses, use the first non-broadcast address which corresponds
693          * to the incoming interface.
694          */
695         ifp = m->m_pkthdr.rcvif;
696         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
697                 IF_ADDR_RLOCK(ifp);
698                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
699                         if (ifa->ifa_addr->sa_family != AF_INET)
700                                 continue;
701                         ia = ifatoia(ifa);
702                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
703                             t.s_addr) {
704                                 t = IA_SIN(ia)->sin_addr;
705                                 IF_ADDR_RUNLOCK(ifp);
706                                 goto match;
707                         }
708                 }
709                 IF_ADDR_RUNLOCK(ifp);
710         }
711         /*
712          * If the packet was transiting through us, use the address of
713          * the interface the packet came through in.  If that interface
714          * doesn't have a suitable IP address, the normal selection
715          * criteria apply.
716          */
717         if (V_icmp_rfi && ifp != NULL) {
718                 IF_ADDR_RLOCK(ifp);
719                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
720                         if (ifa->ifa_addr->sa_family != AF_INET)
721                                 continue;
722                         ia = ifatoia(ifa);
723                         t = IA_SIN(ia)->sin_addr;
724                         IF_ADDR_RUNLOCK(ifp);
725                         goto match;
726                 }
727                 IF_ADDR_RUNLOCK(ifp);
728         }
729         /*
730          * If the incoming packet was not addressed directly to us, use
731          * designated interface for icmp replies specified by sysctl
732          * net.inet.icmp.reply_src (default not set). Otherwise continue
733          * with normal source selection.
734          */
735         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
736                 IF_ADDR_RLOCK(ifp);
737                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
738                         if (ifa->ifa_addr->sa_family != AF_INET)
739                                 continue;
740                         ia = ifatoia(ifa);
741                         t = IA_SIN(ia)->sin_addr;
742                         IF_ADDR_RUNLOCK(ifp);
743                         goto match;
744                 }
745                 IF_ADDR_RUNLOCK(ifp);
746         }
747         /*
748          * If the packet was transiting through us, use the address of
749          * the interface that is the closest to the packet source.
750          * When we don't have a route back to the packet source, stop here
751          * and drop the packet.
752          */
753         if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
754                 m_freem(m);
755                 ICMPSTAT_INC(icps_noroute);
756                 goto done;
757         }
758         t = nh_ext.nh_src;
759 match:
760 #ifdef MAC
761         mac_netinet_icmp_replyinplace(m);
762 #endif
763         ip->ip_src = t;
764         ip->ip_ttl = V_ip_defttl;
765
766         if (optlen > 0) {
767                 register u_char *cp;
768                 int opt, cnt;
769                 u_int len;
770
771                 /*
772                  * Retrieve any source routing from the incoming packet;
773                  * add on any record-route or timestamp options.
774                  */
775                 cp = (u_char *) (ip + 1);
776                 if ((opts = ip_srcroute(m)) == 0 &&
777                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
778                         opts->m_len = sizeof(struct in_addr);
779                         mtod(opts, struct in_addr *)->s_addr = 0;
780                 }
781                 if (opts) {
782 #ifdef ICMPPRINTFS
783                     if (icmpprintfs)
784                             printf("icmp_reflect optlen %d rt %d => ",
785                                 optlen, opts->m_len);
786 #endif
787                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
788                             opt = cp[IPOPT_OPTVAL];
789                             if (opt == IPOPT_EOL)
790                                     break;
791                             if (opt == IPOPT_NOP)
792                                     len = 1;
793                             else {
794                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
795                                             break;
796                                     len = cp[IPOPT_OLEN];
797                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
798                                         len > cnt)
799                                             break;
800                             }
801                             /*
802                              * Should check for overflow, but it "can't happen"
803                              */
804                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
805                                 opt == IPOPT_SECURITY) {
806                                     bcopy((caddr_t)cp,
807                                         mtod(opts, caddr_t) + opts->m_len, len);
808                                     opts->m_len += len;
809                             }
810                     }
811                     /* Terminate & pad, if necessary */
812                     cnt = opts->m_len % 4;
813                     if (cnt) {
814                             for (; cnt < 4; cnt++) {
815                                     *(mtod(opts, caddr_t) + opts->m_len) =
816                                         IPOPT_EOL;
817                                     opts->m_len++;
818                             }
819                     }
820 #ifdef ICMPPRINTFS
821                     if (icmpprintfs)
822                             printf("%d\n", opts->m_len);
823 #endif
824                 }
825                 ip_stripoptions(m);
826         }
827         m_tag_delete_nonpersistent(m);
828         m->m_flags &= ~(M_BCAST|M_MCAST);
829         icmp_send(m, opts);
830 done:
831         if (opts)
832                 (void)m_free(opts);
833 }
834
835 /*
836  * Send an icmp packet back to the ip level,
837  * after supplying a checksum.
838  */
839 static void
840 icmp_send(struct mbuf *m, struct mbuf *opts)
841 {
842         register struct ip *ip = mtod(m, struct ip *);
843         register int hlen;
844         register struct icmp *icp;
845
846         hlen = ip->ip_hl << 2;
847         m->m_data += hlen;
848         m->m_len -= hlen;
849         icp = mtod(m, struct icmp *);
850         icp->icmp_cksum = 0;
851         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
852         m->m_data -= hlen;
853         m->m_len += hlen;
854         m->m_pkthdr.rcvif = (struct ifnet *)0;
855 #ifdef ICMPPRINTFS
856         if (icmpprintfs) {
857                 char buf[4 * sizeof "123"];
858                 strcpy(buf, inet_ntoa(ip->ip_dst));
859                 printf("icmp_send dst %s src %s\n",
860                        buf, inet_ntoa(ip->ip_src));
861         }
862 #endif
863         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
864 }
865
866 /*
867  * Return milliseconds since 00:00 UTC in network format.
868  */
869 uint32_t
870 iptime(void)
871 {
872         struct timeval atv;
873         u_long t;
874
875         getmicrotime(&atv);
876         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
877         return (htonl(t));
878 }
879
880 /*
881  * Return the next larger or smaller MTU plateau (table from RFC 1191)
882  * given current value MTU.  If DIR is less than zero, a larger plateau
883  * is returned; otherwise, a smaller value is returned.
884  */
885 int
886 ip_next_mtu(int mtu, int dir)
887 {
888         static int mtutab[] = {
889                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
890                 296, 68, 0
891         };
892         int i, size;
893
894         size = (sizeof mtutab) / (sizeof mtutab[0]);
895         if (dir >= 0) {
896                 for (i = 0; i < size; i++)
897                         if (mtu > mtutab[i])
898                                 return mtutab[i];
899         } else {
900                 for (i = size - 1; i >= 0; i--)
901                         if (mtu < mtutab[i])
902                                 return mtutab[i];
903                 if (mtu == mtutab[0])
904                         return mtutab[0];
905         }
906         return 0;
907 }
908 #endif /* INET */
909
910
911 /*
912  * badport_bandlim() - check for ICMP bandwidth limit
913  *
914  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
915  *      hit our bandwidth limit and it is not ok.
916  *
917  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
918  *
919  *      For now we separate the TCP and UDP subsystems w/ different 'which'
920  *      values.  We may eventually remove this separation (and simplify the
921  *      code further).
922  *
923  *      Note that the printing of the error message is delayed so we can
924  *      properly print the icmp error rate that the system was trying to do
925  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
926  *      the 'final' error, but it doesn't make sense to solve the printing
927  *      delay with more complex code.
928  */
929
930 int
931 badport_bandlim(int which)
932 {
933
934 #define N(a)    (sizeof (a) / sizeof (a[0]))
935         static struct rate {
936                 const char      *type;
937                 struct timeval  lasttime;
938                 int             curpps;
939         } rates[BANDLIM_MAX+1] = {
940                 { "icmp unreach response" },
941                 { "icmp ping response" },
942                 { "icmp tstamp response" },
943                 { "closed port RST response" },
944                 { "open port RST response" },
945                 { "icmp6 unreach response" },
946                 { "sctp ootb response" }
947         };
948
949         /*
950          * Return ok status if feature disabled or argument out of range.
951          */
952         if (V_icmplim > 0 && (u_int) which < N(rates)) {
953                 struct rate *r = &rates[which];
954                 int opps = r->curpps;
955
956                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
957                         return -1;      /* discard packet */
958                 /*
959                  * If we've dropped below the threshold after having
960                  * rate-limited traffic print the message.  This preserves
961                  * the previous behaviour at the expense of added complexity.
962                  */
963                 if (V_icmplim_output && opps > V_icmplim)
964                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
965                                 r->type, opps, V_icmplim);
966         }
967         return 0;                       /* okay to send packet */
968 #undef N
969 }