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_ipsec.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/sysctl.h>
  45 #include <sys/syslog.h>
  46
  47 #include <net/if.h>
  48 #include <net/if_types.h>
  49 #include <net/route.h>
  50 #include <net/vnet.h>
  51
  52 #include <netinet/in.h>
  53 #include <netinet/in_pcb.h>
  54 #include <netinet/in_systm.h>
  55 #include <netinet/in_var.h>
  56 #include <netinet/ip.h>
  57 #include <netinet/ip_icmp.h>
  58 #include <netinet/ip_var.h>
  59 #include <netinet/ip_options.h>
  60 #include <netinet/tcp.h>
  61 #include <netinet/tcp_var.h>
  62 #include <netinet/tcpip.h>
  63 #include <netinet/icmp_var.h>
  64
  65 #ifdef IPSEC
  66 #include <netipsec/ipsec.h>
  67 #include <netipsec/key.h>
  68 #endif
  69
  70 #include <machine/in_cksum.h>
  71
  72 #include <security/mac/mac_framework.h>
  73
  74 /*
  75  * ICMP routines: error generation, receive packet processing, and
  76  * routines to turnaround packets back to the originator, and
  77  * host table maintenance routines.
  78  */
  79 VNET_DEFINE(struct icmpstat, icmpstat);
  80 SYSCTL_VNET_STRUCT(_net_inet_icmp, ICMPCTL_STATS, stats, CTLFLAG_RW,
  81         &VNET_NAME(icmpstat), icmpstat, "");
  82
  83 static VNET_DEFINE(int, icmpmaskrepl) = 0;
  84 #define V_icmpmaskrepl                  VNET(icmpmaskrepl)
  85 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_RW,
  86         &VNET_NAME(icmpmaskrepl), 0,
  87         "Reply to ICMP Address Mask Request packets.");
  88
  89 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
  90 #define V_icmpmaskfake                  VNET(icmpmaskfake)
  91 SYSCTL_VNET_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_RW,
  92         &VNET_NAME(icmpmaskfake), 0,
  93         "Fake reply to ICMP Address Mask Request packets.");
  94
  95 static VNET_DEFINE(int, drop_redirect) = 0;
  96 #define V_drop_redirect                 VNET(drop_redirect)
  97 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_RW,
  98         &VNET_NAME(drop_redirect), 0,
  99         "Ignore ICMP redirects");
 100
 101 static VNET_DEFINE(int, log_redirect) = 0;
 102 #define V_log_redirect                  VNET(log_redirect)
 103 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_RW,
 104         &VNET_NAME(log_redirect), 0,
 105         "Log ICMP redirects to the console");
 106
 107 static VNET_DEFINE(int, icmplim) = 200;
 108 #define V_icmplim                       VNET(icmplim)
 109 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_RW,
 110         &VNET_NAME(icmplim), 0,
 111         "Maximum number of ICMP responses per second");
 112
 113 static VNET_DEFINE(int, icmplim_output) = 1;
 114 #define V_icmplim_output                VNET(icmplim_output)
 115 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_RW,
 116         &VNET_NAME(icmplim_output), 0,
 117         "Enable rate limiting of ICMP responses");
 118
 119 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
 120 #define V_reply_src                     VNET(reply_src)
 121 SYSCTL_VNET_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_RW,
 122         &VNET_NAME(reply_src), IFNAMSIZ,
 123         "icmp reply source for non-local packets.");
 124
 125 static VNET_DEFINE(int, icmp_rfi) = 0;
 126 #define V_icmp_rfi                      VNET(icmp_rfi)
 127 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_RW,
 128         &VNET_NAME(icmp_rfi), 0,
 129         "ICMP reply from incoming interface for non-local packets");
 130
 131 static VNET_DEFINE(int, icmp_quotelen) = 8;
 132 #define V_icmp_quotelen                 VNET(icmp_quotelen)
 133 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_RW,
 134         &VNET_NAME(icmp_quotelen), 0,
 135         "Number of bytes from original packet to quote in ICMP reply");
 136
 137 /*
 138  * ICMP broadcast echo sysctl
 139  */
 140 static VNET_DEFINE(int, icmpbmcastecho) = 0;
 141 #define V_icmpbmcastecho                VNET(icmpbmcastecho)
 142 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_RW,
 143         &VNET_NAME(icmpbmcastecho), 0,
 144         "");
 145
 146
 147 #ifdef ICMPPRINTFS
 148 int     icmpprintfs = 0;
 149 #endif
 150
 151 static void     icmp_reflect(struct mbuf *);
 152 static void     icmp_send(struct mbuf *, struct mbuf *);
 153
 154 extern  struct protosw inetsw[];
 155
 156 /*
 157  * Kernel module interface for updating icmpstat.  The argument is an index
 158  * into icmpstat treated as an array of u_long.  While this encodes the
 159  * general layout of icmpstat into the caller, it doesn't encode its
 160  * location, so that future changes to add, for example, per-CPU stats
 161  * support won't cause binary compatibility problems for kernel modules.
 162  */
 163 void
 164 kmod_icmpstat_inc(int statnum)
 165 {
 166
 167         (*((u_long *)&V_icmpstat + statnum))++;
 168 }
 169
 170 /*
 171  * Generate an error packet of type error
 172  * in response to bad packet ip.
 173  */
 174 void
 175 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 176 {
 177         register struct ip *oip = mtod(n, struct ip *), *nip;
 178         register unsigned oiphlen = oip->ip_hl << 2;
 179         register struct icmp *icp;
 180         register struct mbuf *m;
 181         unsigned icmplen, icmpelen, nlen;
 182
 183         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
 184 #ifdef ICMPPRINTFS
 185         if (icmpprintfs)
 186                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 187 #endif
 188         if (type != ICMP_REDIRECT)
 189                 ICMPSTAT_INC(icps_error);
 190         /*
 191          * Don't send error:
 192          *  if the original packet was encrypted.
 193          *  if not the first fragment of message.
 194          *  in response to a multicast or broadcast packet.
 195          *  if the old packet protocol was an ICMP error message.
 196          */
 197         if (n->m_flags & M_DECRYPTED)
 198                 goto freeit;
 199         if (oip->ip_off & ~(IP_MF|IP_DF))
 200                 goto freeit;
 201         if (n->m_flags & (M_BCAST|M_MCAST))
 202                 goto freeit;
 203         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 204           n->m_len >= oiphlen + ICMP_MINLEN &&
 205           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
 206                 ICMPSTAT_INC(icps_oldicmp);
 207                 goto freeit;
 208         }
 209         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
 210         if (oiphlen + 8 > n->m_len)
 211                 goto freeit;
 212         /*
 213          * Calculate length to quote from original packet and
 214          * prevent the ICMP mbuf from overflowing.
 215          * Unfortunatly this is non-trivial since ip_forward()
 216          * sends us truncated packets.
 217          */
 218         nlen = m_length(n, NULL);
 219         if (oip->ip_p == IPPROTO_TCP) {
 220                 struct tcphdr *th;
 221                 int tcphlen;
 222
 223                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 224                     n->m_next == NULL)
 225                         goto stdreply;
 226                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 227                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
 228                         goto freeit;
 229                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
 230                 tcphlen = th->th_off << 2;
 231                 if (tcphlen < sizeof(struct tcphdr))
 232                         goto freeit;
 233                 if (oip->ip_len < oiphlen + tcphlen)
 234                         goto freeit;
 235                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 236                         goto stdreply;
 237                 if (n->m_len < oiphlen + tcphlen &&
 238                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
 239                         goto freeit;
 240                 icmpelen = max(tcphlen, min(V_icmp_quotelen, oip->ip_len - oiphlen));
 241         } else
 242 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, oip->ip_len - oiphlen));
 243
 244         icmplen = min(oiphlen + icmpelen, nlen);
 245         if (icmplen < sizeof(struct ip))
 246                 goto freeit;
 247
 248         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 249                 m = m_gethdr(M_DONTWAIT, MT_DATA);
 250         else
 251                 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
 252         if (m == NULL)
 253                 goto freeit;
 254 #ifdef MAC
 255         mac_netinet_icmp_reply(n, m);
 256 #endif
 257         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
 258         m_align(m, ICMP_MINLEN + icmplen);
 259         m->m_len = ICMP_MINLEN + icmplen;
 260
 261         /* XXX MRT  make the outgoing packet use the same FIB
 262          * that was associated with the incoming packet
 263          */
 264         M_SETFIB(m, M_GETFIB(n));
 265         icp = mtod(m, struct icmp *);
 266         ICMPSTAT_INC(icps_outhist[type]);
 267         icp->icmp_type = type;
 268         if (type == ICMP_REDIRECT)
 269                 icp->icmp_gwaddr.s_addr = dest;
 270         else {
 271                 icp->icmp_void = 0;
 272                 /*
 273                  * The following assignments assume an overlay with the
 274                  * just zeroed icmp_void field.
 275                  */
 276                 if (type == ICMP_PARAMPROB) {
 277                         icp->icmp_pptr = code;
 278                         code = 0;
 279                 } else if (type == ICMP_UNREACH &&
 280                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 281                         icp->icmp_nextmtu = htons(mtu);
 282                 }
 283         }
 284         icp->icmp_code = code;
 285
 286         /*
 287          * Copy the quotation into ICMP message and
 288          * convert quoted IP header back to network representation.
 289          */
 290         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 291         nip = &icp->icmp_ip;
 292         nip->ip_len = htons(nip->ip_len);
 293         nip->ip_off = htons(nip->ip_off);
 294
 295         /*
 296          * Set up ICMP message mbuf and copy old IP header (without options
 297          * in front of ICMP message.
 298          * If the original mbuf was meant to bypass the firewall, the error
 299          * reply should bypass as well.
 300          */
 301         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 302         m->m_data -= sizeof(struct ip);
 303         m->m_len += sizeof(struct ip);
 304         m->m_pkthdr.len = m->m_len;
 305         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 306         nip = mtod(m, struct ip *);
 307         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 308         nip->ip_len = m->m_len;
 309         nip->ip_v = IPVERSION;
 310         nip->ip_hl = 5;
 311         nip->ip_p = IPPROTO_ICMP;
 312         nip->ip_tos = 0;
 313         icmp_reflect(m);
 314
 315 freeit:
 316         m_freem(n);
 317 }
 318
 319 /*
 320  * Process a received ICMP message.
 321  */
 322 void
 323 icmp_input(struct mbuf *m, int off)
 324 {
 325         struct icmp *icp;
 326         struct in_ifaddr *ia;
 327         struct ip *ip = mtod(m, struct ip *);
 328         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 329         int hlen = off;
 330         int icmplen = ip->ip_len;
 331         int i, code;
 332         void (*ctlfunc)(int, struct sockaddr *, void *);
 333         int fibnum;
 334
 335         /*
 336          * Locate icmp structure in mbuf, and check
 337          * that not corrupted and of at least minimum length.
 338          */
 339 #ifdef ICMPPRINTFS
 340         if (icmpprintfs) {
 341                 char buf[4 * sizeof "123"];
 342                 strcpy(buf, inet_ntoa(ip->ip_src));
 343                 printf("icmp_input from %s to %s, len %d\n",
 344                        buf, inet_ntoa(ip->ip_dst), icmplen);
 345         }
 346 #endif
 347         if (icmplen < ICMP_MINLEN) {
 348                 ICMPSTAT_INC(icps_tooshort);
 349                 goto freeit;
 350         }
 351         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 352         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 353                 ICMPSTAT_INC(icps_tooshort);
 354                 return;
 355         }
 356         ip = mtod(m, struct ip *);
 357         m->m_len -= hlen;
 358         m->m_data += hlen;
 359         icp = mtod(m, struct icmp *);
 360         if (in_cksum(m, icmplen)) {
 361                 ICMPSTAT_INC(icps_checksum);
 362                 goto freeit;
 363         }
 364         m->m_len += hlen;
 365         m->m_data -= hlen;
 366
 367         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 368                 /*
 369                  * Deliver very specific ICMP type only.
 370                  */
 371                 switch (icp->icmp_type) {
 372                 case ICMP_UNREACH:
 373                 case ICMP_TIMXCEED:
 374                         break;
 375                 default:
 376                         goto freeit;
 377                 }
 378         }
 379
 380 #ifdef ICMPPRINTFS
 381         if (icmpprintfs)
 382                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 383                     icp->icmp_code);
 384 #endif
 385
 386         /*
 387          * Message type specific processing.
 388          */
 389         if (icp->icmp_type > ICMP_MAXTYPE)
 390                 goto raw;
 391
 392         /* Initialize */
 393         bzero(&icmpsrc, sizeof(icmpsrc));
 394         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 395         icmpsrc.sin_family = AF_INET;
 396         bzero(&icmpdst, sizeof(icmpdst));
 397         icmpdst.sin_len = sizeof(struct sockaddr_in);
 398         icmpdst.sin_family = AF_INET;
 399         bzero(&icmpgw, sizeof(icmpgw));
 400         icmpgw.sin_len = sizeof(struct sockaddr_in);
 401         icmpgw.sin_family = AF_INET;
 402
 403         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 404         code = icp->icmp_code;
 405         switch (icp->icmp_type) {
 406
 407         case ICMP_UNREACH:
 408                 switch (code) {
 409                         case ICMP_UNREACH_NET:
 410                         case ICMP_UNREACH_HOST:
 411                         case ICMP_UNREACH_SRCFAIL:
 412                         case ICMP_UNREACH_NET_UNKNOWN:
 413                         case ICMP_UNREACH_HOST_UNKNOWN:
 414                         case ICMP_UNREACH_ISOLATED:
 415                         case ICMP_UNREACH_TOSNET:
 416                         case ICMP_UNREACH_TOSHOST:
 417                         case ICMP_UNREACH_HOST_PRECEDENCE:
 418                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 419                                 code = PRC_UNREACH_NET;
 420                                 break;
 421
 422                         case ICMP_UNREACH_NEEDFRAG:
 423                                 code = PRC_MSGSIZE;
 424                                 break;
 425
 426                         /*
 427                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 428                          * Treat subcodes 2,3 as immediate RST
 429                          */
 430                         case ICMP_UNREACH_PROTOCOL:
 431                         case ICMP_UNREACH_PORT:
 432                                 code = PRC_UNREACH_PORT;
 433                                 break;
 434
 435                         case ICMP_UNREACH_NET_PROHIB:
 436                         case ICMP_UNREACH_HOST_PROHIB:
 437                         case ICMP_UNREACH_FILTER_PROHIB:
 438                                 code = PRC_UNREACH_ADMIN_PROHIB;
 439                                 break;
 440
 441                         default:
 442                                 goto badcode;
 443                 }
 444                 goto deliver;
 445
 446         case ICMP_TIMXCEED:
 447                 if (code > 1)
 448                         goto badcode;
 449                 code += PRC_TIMXCEED_INTRANS;
 450                 goto deliver;
 451
 452         case ICMP_PARAMPROB:
 453                 if (code > 1)
 454                         goto badcode;
 455                 code = PRC_PARAMPROB;
 456                 goto deliver;
 457
 458         case ICMP_SOURCEQUENCH:
 459                 if (code)
 460                         goto badcode;
 461                 code = PRC_QUENCH;
 462         deliver:
 463                 /*
 464                  * Problem with datagram; advise higher level routines.
 465                  */
 466                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 467                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 468                         ICMPSTAT_INC(icps_badlen);
 469                         goto freeit;
 470                 }
 471                 icp->icmp_ip.ip_len = ntohs(icp->icmp_ip.ip_len);
 472                 /* Discard ICMP's in response to multicast packets */
 473                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 474                         goto badcode;
 475 #ifdef ICMPPRINTFS
 476                 if (icmpprintfs)
 477                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 478 #endif
 479                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 480                 /*
 481                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 482                  * notification to TCP layer.
 483                  */
 484                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 485                 if (ctlfunc)
 486                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 487                                    (void *)&icp->icmp_ip);
 488                 break;
 489
 490         badcode:
 491                 ICMPSTAT_INC(icps_badcode);
 492                 break;
 493
 494         case ICMP_ECHO:
 495                 if (!V_icmpbmcastecho
 496                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 497                         ICMPSTAT_INC(icps_bmcastecho);
 498                         break;
 499                 }
 500                 icp->icmp_type = ICMP_ECHOREPLY;
 501                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 502                         goto freeit;
 503                 else
 504                         goto reflect;
 505
 506         case ICMP_TSTAMP:
 507                 if (!V_icmpbmcastecho
 508                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 509                         ICMPSTAT_INC(icps_bmcasttstamp);
 510                         break;
 511                 }
 512                 if (icmplen < ICMP_TSLEN) {
 513                         ICMPSTAT_INC(icps_badlen);
 514                         break;
 515                 }
 516                 icp->icmp_type = ICMP_TSTAMPREPLY;
 517                 icp->icmp_rtime = iptime();
 518                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 519                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 520                         goto freeit;
 521                 else
 522                         goto reflect;
 523
 524         case ICMP_MASKREQ:
 525                 if (V_icmpmaskrepl == 0)
 526                         break;
 527                 /*
 528                  * We are not able to respond with all ones broadcast
 529                  * unless we receive it over a point-to-point interface.
 530                  */
 531                 if (icmplen < ICMP_MASKLEN)
 532                         break;
 533                 switch (ip->ip_dst.s_addr) {
 534
 535                 case INADDR_BROADCAST:
 536                 case INADDR_ANY:
 537                         icmpdst.sin_addr = ip->ip_src;
 538                         break;
 539
 540                 default:
 541                         icmpdst.sin_addr = ip->ip_dst;
 542                 }
 543                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 544                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 545                 if (ia == NULL)
 546                         break;
 547                 if (ia->ia_ifp == NULL) {
 548                         ifa_free(&ia->ia_ifa);
 549                         break;
 550                 }
 551                 icp->icmp_type = ICMP_MASKREPLY;
 552                 if (V_icmpmaskfake == 0)
 553                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 554                 else
 555                         icp->icmp_mask = V_icmpmaskfake;
 556                 if (ip->ip_src.s_addr == 0) {
 557                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 558                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 559                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 560                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 561                 }
 562                 ifa_free(&ia->ia_ifa);
 563 reflect:
 564                 ip->ip_len += hlen;     /* since ip_input deducts this */
 565                 ICMPSTAT_INC(icps_reflect);
 566                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 567                 icmp_reflect(m);
 568                 return;
 569
 570         case ICMP_REDIRECT:
 571                 if (V_log_redirect) {
 572                         u_long src, dst, gw;
 573
 574                         src = ntohl(ip->ip_src.s_addr);
 575                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 576                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 577                         printf("icmp redirect from %d.%d.%d.%d: "
 578                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 579                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 580                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 581                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 582                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 583                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 584                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 585                 }
 586                 /*
 587                  * RFC1812 says we must ignore ICMP redirects if we
 588                  * are acting as router.
 589                  */
 590                 if (V_drop_redirect || V_ipforwarding)
 591                         break;
 592                 if (code > 3)
 593                         goto badcode;
 594                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 595                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 596                         ICMPSTAT_INC(icps_badlen);
 597                         break;
 598                 }
 599                 /*
 600                  * Short circuit routing redirects to force
 601                  * immediate change in the kernel's routing
 602                  * tables.  The message is also handed to anyone
 603                  * listening on a raw socket (e.g. the routing
 604                  * daemon for use in updating its tables).
 605                  */
 606                 icmpgw.sin_addr = ip->ip_src;
 607                 icmpdst.sin_addr = icp->icmp_gwaddr;
 608 #ifdef  ICMPPRINTFS
 609                 if (icmpprintfs) {
 610                         char buf[4 * sizeof "123"];
 611                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
 612
 613                         printf("redirect dst %s to %s\n",
 614                                buf, inet_ntoa(icp->icmp_gwaddr));
 615                 }
 616 #endif
 617                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 618                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 619                         in_rtredirect((struct sockaddr *)&icmpsrc,
 620                           (struct sockaddr *)&icmpdst,
 621                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 622                           (struct sockaddr *)&icmpgw, fibnum);
 623                 }
 624                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 625 #ifdef IPSEC
 626                 key_sa_routechange((struct sockaddr *)&icmpsrc);
 627 #endif
 628                 break;
 629
 630         /*
 631          * No kernel processing for the following;
 632          * just fall through to send to raw listener.
 633          */
 634         case ICMP_ECHOREPLY:
 635         case ICMP_ROUTERADVERT:
 636         case ICMP_ROUTERSOLICIT:
 637         case ICMP_TSTAMPREPLY:
 638         case ICMP_IREQREPLY:
 639         case ICMP_MASKREPLY:
 640         default:
 641                 break;
 642         }
 643
 644 raw:
 645         rip_input(m, off);
 646         return;
 647
 648 freeit:
 649         m_freem(m);
 650 }
 651
 652 /*
 653  * Reflect the ip packet back to the source
 654  */
 655 static void
 656 icmp_reflect(struct mbuf *m)
 657 {
 658         struct ip *ip = mtod(m, struct ip *);
 659         struct ifaddr *ifa;
 660         struct ifnet *ifp;
 661         struct in_ifaddr *ia;
 662         struct in_addr t;
 663         struct mbuf *opts = 0;
 664         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 665
 666         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 667             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 668             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 669                 m_freem(m);     /* Bad return address */
 670                 ICMPSTAT_INC(icps_badaddr);
 671                 goto done;      /* Ip_output() will check for broadcast */
 672         }
 673
 674         t = ip->ip_dst;
 675         ip->ip_dst = ip->ip_src;
 676
 677         /*
 678          * Source selection for ICMP replies:
 679          *
 680          * If the incoming packet was addressed directly to one of our
 681          * own addresses, use dst as the src for the reply.
 682          */
 683         IN_IFADDR_RLOCK();
 684         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 685                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 686                         t = IA_SIN(ia)->sin_addr;
 687                         IN_IFADDR_RUNLOCK();
 688                         goto match;
 689                 }
 690         }
 691         IN_IFADDR_RUNLOCK();
 692
 693         /*
 694          * If the incoming packet was addressed to one of our broadcast
 695          * addresses, use the first non-broadcast address which corresponds
 696          * to the incoming interface.
 697          */
 698         ifp = m->m_pkthdr.rcvif;
 699         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 700                 IF_ADDR_LOCK(ifp);
 701                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 702                         if (ifa->ifa_addr->sa_family != AF_INET)
 703                                 continue;
 704                         ia = ifatoia(ifa);
 705                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 706                             t.s_addr) {
 707                                 t = IA_SIN(ia)->sin_addr;
 708                                 IF_ADDR_UNLOCK(ifp);
 709                                 goto match;
 710                         }
 711                 }
 712                 IF_ADDR_UNLOCK(ifp);
 713         }
 714         /*
 715          * If the packet was transiting through us, use the address of
 716          * the interface the packet came through in.  If that interface
 717          * doesn't have a suitable IP address, the normal selection
 718          * criteria apply.
 719          */
 720         if (V_icmp_rfi && ifp != NULL) {
 721                 IF_ADDR_LOCK(ifp);
 722                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 723                         if (ifa->ifa_addr->sa_family != AF_INET)
 724                                 continue;
 725                         ia = ifatoia(ifa);
 726                         t = IA_SIN(ia)->sin_addr;
 727                         IF_ADDR_UNLOCK(ifp);
 728                         goto match;
 729                 }
 730                 IF_ADDR_UNLOCK(ifp);
 731         }
 732         /*
 733          * If the incoming packet was not addressed directly to us, use
 734          * designated interface for icmp replies specified by sysctl
 735          * net.inet.icmp.reply_src (default not set). Otherwise continue
 736          * with normal source selection.
 737          */
 738         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 739                 IF_ADDR_LOCK(ifp);
 740                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 741                         if (ifa->ifa_addr->sa_family != AF_INET)
 742                                 continue;
 743                         ia = ifatoia(ifa);
 744                         t = IA_SIN(ia)->sin_addr;
 745                         IF_ADDR_UNLOCK(ifp);
 746                         goto match;
 747                 }
 748                 IF_ADDR_UNLOCK(ifp);
 749         }
 750         /*
 751          * If the packet was transiting through us, use the address of
 752          * the interface that is the closest to the packet source.
 753          * When we don't have a route back to the packet source, stop here
 754          * and drop the packet.
 755          */
 756         ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
 757         if (ia == NULL) {
 758                 m_freem(m);
 759                 ICMPSTAT_INC(icps_noroute);
 760                 goto done;
 761         }
 762         t = IA_SIN(ia)->sin_addr;
 763         ifa_free(&ia->ia_ifa);
 764 match:
 765 #ifdef MAC
 766         mac_netinet_icmp_replyinplace(m);
 767 #endif
 768         ip->ip_src = t;
 769         ip->ip_ttl = V_ip_defttl;
 770
 771         if (optlen > 0) {
 772                 register u_char *cp;
 773                 int opt, cnt;
 774                 u_int len;
 775
 776                 /*
 777                  * Retrieve any source routing from the incoming packet;
 778                  * add on any record-route or timestamp options.
 779                  */
 780                 cp = (u_char *) (ip + 1);
 781                 if ((opts = ip_srcroute(m)) == 0 &&
 782                     (opts = m_gethdr(M_DONTWAIT, MT_DATA))) {
 783                         opts->m_len = sizeof(struct in_addr);
 784                         mtod(opts, struct in_addr *)->s_addr = 0;
 785                 }
 786                 if (opts) {
 787 #ifdef ICMPPRINTFS
 788                     if (icmpprintfs)
 789                             printf("icmp_reflect optlen %d rt %d => ",
 790                                 optlen, opts->m_len);
 791 #endif
 792                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 793                             opt = cp[IPOPT_OPTVAL];
 794                             if (opt == IPOPT_EOL)
 795                                     break;
 796                             if (opt == IPOPT_NOP)
 797                                     len = 1;
 798                             else {
 799                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 800                                             break;
 801                                     len = cp[IPOPT_OLEN];
 802                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 803                                         len > cnt)
 804                                             break;
 805                             }
 806                             /*
 807                              * Should check for overflow, but it "can't happen"
 808                              */
 809                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 810                                 opt == IPOPT_SECURITY) {
 811                                     bcopy((caddr_t)cp,
 812                                         mtod(opts, caddr_t) + opts->m_len, len);
 813                                     opts->m_len += len;
 814                             }
 815                     }
 816                     /* Terminate & pad, if necessary */
 817                     cnt = opts->m_len % 4;
 818                     if (cnt) {
 819                             for (; cnt < 4; cnt++) {
 820                                     *(mtod(opts, caddr_t) + opts->m_len) =
 821                                         IPOPT_EOL;
 822                                     opts->m_len++;
 823                             }
 824                     }
 825 #ifdef ICMPPRINTFS
 826                     if (icmpprintfs)
 827                             printf("%d\n", opts->m_len);
 828 #endif
 829                 }
 830                 /*
 831                  * Now strip out original options by copying rest of first
 832                  * mbuf's data back, and adjust the IP length.
 833                  */
 834                 ip->ip_len -= optlen;
 835                 ip->ip_v = IPVERSION;
 836                 ip->ip_hl = 5;
 837                 m->m_len -= optlen;
 838                 if (m->m_flags & M_PKTHDR)
 839                         m->m_pkthdr.len -= optlen;
 840                 optlen += sizeof(struct ip);
 841                 bcopy((caddr_t)ip + optlen, (caddr_t)(ip + 1),
 842                          (unsigned)(m->m_len - sizeof(struct ip)));
 843         }
 844         m_tag_delete_nonpersistent(m);
 845         m->m_flags &= ~(M_BCAST|M_MCAST);
 846         icmp_send(m, opts);
 847 done:
 848         if (opts)
 849                 (void)m_free(opts);
 850 }
 851
 852 /*
 853  * Send an icmp packet back to the ip level,
 854  * after supplying a checksum.
 855  */
 856 static void
 857 icmp_send(struct mbuf *m, struct mbuf *opts)
 858 {
 859         register struct ip *ip = mtod(m, struct ip *);
 860         register int hlen;
 861         register struct icmp *icp;
 862
 863         hlen = ip->ip_hl << 2;
 864         m->m_data += hlen;
 865         m->m_len -= hlen;
 866         icp = mtod(m, struct icmp *);
 867         icp->icmp_cksum = 0;
 868         icp->icmp_cksum = in_cksum(m, ip->ip_len - hlen);
 869         m->m_data -= hlen;
 870         m->m_len += hlen;
 871         m->m_pkthdr.rcvif = (struct ifnet *)0;
 872 #ifdef ICMPPRINTFS
 873         if (icmpprintfs) {
 874                 char buf[4 * sizeof "123"];
 875                 strcpy(buf, inet_ntoa(ip->ip_dst));
 876                 printf("icmp_send dst %s src %s\n",
 877                        buf, inet_ntoa(ip->ip_src));
 878         }
 879 #endif
 880         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 881 }
 882
 883 /*
 884  * Return milliseconds since 00:00 GMT in network format.
 885  */
 886 uint32_t
 887 iptime(void)
 888 {
 889         struct timeval atv;
 890         u_long t;
 891
 892         getmicrotime(&atv);
 893         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 894         return (htonl(t));
 895 }
 896
 897 /*
 898  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 899  * given current value MTU.  If DIR is less than zero, a larger plateau
 900  * is returned; otherwise, a smaller value is returned.
 901  */
 902 int
 903 ip_next_mtu(int mtu, int dir)
 904 {
 905         static int mtutab[] = {
 906                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 907                 296, 68, 0
 908         };
 909         int i, size;
 910
 911         size = (sizeof mtutab) / (sizeof mtutab[0]);
 912         if (dir >= 0) {
 913                 for (i = 0; i < size; i++)
 914                         if (mtu > mtutab[i])
 915                                 return mtutab[i];
 916         } else {
 917                 for (i = size - 1; i >= 0; i--)
 918                         if (mtu < mtutab[i])
 919                                 return mtutab[i];
 920                 if (mtu == mtutab[0])
 921                         return mtutab[0];
 922         }
 923         return 0;
 924 }
 925
 926
 927 /*
 928  * badport_bandlim() - check for ICMP bandwidth limit
 929  *
 930  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 931  *      hit our bandwidth limit and it is not ok.
 932  *
 933  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 934  *
 935  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 936  *      values.  We may eventually remove this separation (and simplify the
 937  *      code further).
 938  *
 939  *      Note that the printing of the error message is delayed so we can
 940  *      properly print the icmp error rate that the system was trying to do
 941  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 942  *      the 'final' error, but it doesn't make sense to solve the printing
 943  *      delay with more complex code.
 944  */
 945
 946 int
 947 badport_bandlim(int which)
 948 {
 949
 950 #define N(a)    (sizeof (a) / sizeof (a[0]))
 951         static struct rate {
 952                 const char      *type;
 953                 struct timeval  lasttime;
 954                 int             curpps;
 955         } rates[BANDLIM_MAX+1] = {
 956                 { "icmp unreach response" },
 957                 { "icmp ping response" },
 958                 { "icmp tstamp response" },
 959                 { "closed port RST response" },
 960                 { "open port RST response" },
 961                 { "icmp6 unreach response" }
 962         };
 963
 964         /*
 965          * Return ok status if feature disabled or argument out of range.
 966          */
 967         if (V_icmplim > 0 && (u_int) which < N(rates)) {
 968                 struct rate *r = &rates[which];
 969                 int opps = r->curpps;
 970
 971                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
 972                         return -1;      /* discard packet */
 973                 /*
 974                  * If we've dropped below the threshold after having
 975                  * rate-limited traffic print the message.  This preserves
 976                  * the previous behaviour at the expense of added complexity.
 977                  */
 978                 if (V_icmplim_output && opps > V_icmplim)
 979                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
 980                                 r->type, opps, V_icmplim);
 981         }
 982         return 0;                       /* okay to send packet */
 983 #undef N
 984 }