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/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 INET
  66
  67 #include <machine/in_cksum.h>
  68
  69 #include <security/mac/mac_framework.h>
  70 #endif /* INET */
  71
  72 /*
  73  * ICMP routines: error generation, receive packet processing, and
  74  * routines to turnaround packets back to the originator, and
  75  * host table maintenance routines.
  76  */
  77 static VNET_DEFINE(int, icmplim) = 200;
  78 #define V_icmplim                       VNET(icmplim)
  79 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_RW,
  80         &VNET_NAME(icmplim), 0,
  81         "Maximum number of ICMP responses per second");
  82
  83 static VNET_DEFINE(int, icmplim_output) = 1;
  84 #define V_icmplim_output                VNET(icmplim_output)
  85 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_RW,
  86         &VNET_NAME(icmplim_output), 0,
  87         "Enable logging of ICMP response rate limiting");
  88
  89 #ifdef INET
  90 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
  91 VNET_PCPUSTAT_SYSINIT(icmpstat);
  92 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
  93     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
  94
  95 #ifdef VIMAGE
  96 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
  97 #endif /* VIMAGE */
  98
  99 static VNET_DEFINE(int, icmpmaskrepl) = 0;
 100 #define V_icmpmaskrepl                  VNET(icmpmaskrepl)
 101 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_RW,
 102         &VNET_NAME(icmpmaskrepl), 0,
 103         "Reply to ICMP Address Mask Request packets.");
 104
 105 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
 106 #define V_icmpmaskfake                  VNET(icmpmaskfake)
 107 SYSCTL_VNET_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_RW,
 108         &VNET_NAME(icmpmaskfake), 0,
 109         "Fake reply to ICMP Address Mask Request packets.");
 110
 111 VNET_DEFINE(int, drop_redirect) = 0;
 112
 113 static VNET_DEFINE(int, log_redirect) = 0;
 114 #define V_log_redirect                  VNET(log_redirect)
 115 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_RW,
 116         &VNET_NAME(log_redirect), 0,
 117         "Log ICMP redirects to the console");
 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 static VNET_DEFINE(int, icmptstamprepl) = 1;
 147 #define V_icmptstamprepl                VNET(icmptstamprepl)
 148 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
 149         &VNET_NAME(icmptstamprepl), 0, "Respond to ICMP Timestamp packets");
 150
 151 #ifdef ICMPPRINTFS
 152 int     icmpprintfs = 0;
 153 #endif
 154
 155 static void     icmp_reflect(struct mbuf *);
 156 static void     icmp_send(struct mbuf *, struct mbuf *);
 157
 158 extern  struct protosw inetsw[];
 159
 160 static int
 161 sysctl_net_icmp_drop_redir(SYSCTL_HANDLER_ARGS)
 162 {
 163         int error, new;
 164         int i;
 165         struct radix_node_head *rnh;
 166
 167         new = V_drop_redirect;
 168         error = sysctl_handle_int(oidp, &new, 0, req);
 169         if (error == 0 && req->newptr) {
 170                 new = (new != 0) ? 1 : 0;
 171
 172                 if (new == V_drop_redirect)
 173                         return (0);
 174
 175                 for (i = 0; i < rt_numfibs; i++) {
 176                         if ((rnh = rt_tables_get_rnh(i, AF_INET)) == NULL)
 177                                 continue;
 178                         RADIX_NODE_HEAD_LOCK(rnh);
 179                         in_setmatchfunc(rnh, new);
 180                         RADIX_NODE_HEAD_UNLOCK(rnh);
 181                 }
 182
 183                 V_drop_redirect = new;
 184         }
 185
 186         return (error);
 187 }
 188
 189 SYSCTL_VNET_PROC(_net_inet_icmp, OID_AUTO, drop_redirect,
 190     CTLTYPE_INT|CTLFLAG_RW, 0, 0,
 191     sysctl_net_icmp_drop_redir, "I", "Ignore ICMP redirects");
 192
 193 /*
 194  * Kernel module interface for updating icmpstat.  The argument is an index
 195  * into icmpstat treated as an array of u_long.  While this encodes the
 196  * general layout of icmpstat into the caller, it doesn't encode its
 197  * location, so that future changes to add, for example, per-CPU stats
 198  * support won't cause binary compatibility problems for kernel modules.
 199  */
 200 void
 201 kmod_icmpstat_inc(int statnum)
 202 {
 203
 204         counter_u64_add(VNET(icmpstat)[statnum], 1);
 205 }
 206
 207 /*
 208  * Generate an error packet of type error
 209  * in response to bad packet ip.
 210  */
 211 void
 212 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 213 {
 214         register struct ip *oip = mtod(n, struct ip *), *nip;
 215         register unsigned oiphlen = oip->ip_hl << 2;
 216         register struct icmp *icp;
 217         register struct mbuf *m;
 218         unsigned icmplen, icmpelen, nlen;
 219
 220         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
 221 #ifdef ICMPPRINTFS
 222         if (icmpprintfs)
 223                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 224 #endif
 225         if (type != ICMP_REDIRECT)
 226                 ICMPSTAT_INC(icps_error);
 227         /*
 228          * Don't send error:
 229          *  if the original packet was encrypted.
 230          *  if not the first fragment of message.
 231          *  in response to a multicast or broadcast packet.
 232          *  if the old packet protocol was an ICMP error message.
 233          */
 234         if (n->m_flags & M_DECRYPTED)
 235                 goto freeit;
 236         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 237                 goto freeit;
 238         if (n->m_flags & (M_BCAST|M_MCAST))
 239                 goto freeit;
 240         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 241           n->m_len >= oiphlen + ICMP_MINLEN &&
 242           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
 243                 ICMPSTAT_INC(icps_oldicmp);
 244                 goto freeit;
 245         }
 246         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
 247         if (oiphlen + 8 > n->m_len)
 248                 goto freeit;
 249         /*
 250          * Calculate length to quote from original packet and
 251          * prevent the ICMP mbuf from overflowing.
 252          * Unfortunatly this is non-trivial since ip_forward()
 253          * sends us truncated packets.
 254          */
 255         nlen = m_length(n, NULL);
 256         if (oip->ip_p == IPPROTO_TCP) {
 257                 struct tcphdr *th;
 258                 int tcphlen;
 259
 260                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 261                     n->m_next == NULL)
 262                         goto stdreply;
 263                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 264                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
 265                         goto freeit;
 266                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
 267                 tcphlen = th->th_off << 2;
 268                 if (tcphlen < sizeof(struct tcphdr))
 269                         goto freeit;
 270                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 271                         goto freeit;
 272                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 273                         goto stdreply;
 274                 if (n->m_len < oiphlen + tcphlen &&
 275                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
 276                         goto freeit;
 277                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 278                     ntohs(oip->ip_len) - oiphlen));
 279         } else
 280 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 281
 282         icmplen = min(oiphlen + icmpelen, nlen);
 283         if (icmplen < sizeof(struct ip))
 284                 goto freeit;
 285
 286         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 287                 m = m_gethdr(M_NOWAIT, MT_DATA);
 288         else
 289                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 290         if (m == NULL)
 291                 goto freeit;
 292 #ifdef MAC
 293         mac_netinet_icmp_reply(n, m);
 294 #endif
 295         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
 296         m_align(m, ICMP_MINLEN + icmplen);
 297         m->m_len = ICMP_MINLEN + icmplen;
 298
 299         /* XXX MRT  make the outgoing packet use the same FIB
 300          * that was associated with the incoming packet
 301          */
 302         M_SETFIB(m, M_GETFIB(n));
 303         icp = mtod(m, struct icmp *);
 304         ICMPSTAT_INC(icps_outhist[type]);
 305         icp->icmp_type = type;
 306         if (type == ICMP_REDIRECT)
 307                 icp->icmp_gwaddr.s_addr = dest;
 308         else {
 309                 icp->icmp_void = 0;
 310                 /*
 311                  * The following assignments assume an overlay with the
 312                  * just zeroed icmp_void field.
 313                  */
 314                 if (type == ICMP_PARAMPROB) {
 315                         icp->icmp_pptr = code;
 316                         code = 0;
 317                 } else if (type == ICMP_UNREACH &&
 318                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 319                         icp->icmp_nextmtu = htons(mtu);
 320                 }
 321         }
 322         icp->icmp_code = code;
 323
 324         /*
 325          * Copy the quotation into ICMP message and
 326          * convert quoted IP header back to network representation.
 327          */
 328         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 329         nip = &icp->icmp_ip;
 330
 331         /*
 332          * Set up ICMP message mbuf and copy old IP header (without options
 333          * in front of ICMP message.
 334          * If the original mbuf was meant to bypass the firewall, the error
 335          * reply should bypass as well.
 336          */
 337         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 338         m->m_data -= sizeof(struct ip);
 339         m->m_len += sizeof(struct ip);
 340         m->m_pkthdr.len = m->m_len;
 341         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 342         nip = mtod(m, struct ip *);
 343         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 344         nip->ip_len = htons(m->m_len);
 345         nip->ip_v = IPVERSION;
 346         nip->ip_hl = 5;
 347         nip->ip_p = IPPROTO_ICMP;
 348         nip->ip_tos = 0;
 349         nip->ip_off = 0;
 350         icmp_reflect(m);
 351
 352 freeit:
 353         m_freem(n);
 354 }
 355
 356 /*
 357  * Process a received ICMP message.
 358  */
 359 void
 360 icmp_input(struct mbuf *m, int off)
 361 {
 362         struct icmp *icp;
 363         struct in_ifaddr *ia;
 364         struct ip *ip = mtod(m, struct ip *);
 365         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 366         int hlen = off;
 367         int icmplen = ntohs(ip->ip_len) - off;
 368         int i, code;
 369         void (*ctlfunc)(int, struct sockaddr *, void *);
 370         int fibnum;
 371
 372         /*
 373          * Locate icmp structure in mbuf, and check
 374          * that not corrupted and of at least minimum length.
 375          */
 376 #ifdef ICMPPRINTFS
 377         if (icmpprintfs) {
 378                 char buf[4 * sizeof "123"];
 379                 strcpy(buf, inet_ntoa(ip->ip_src));
 380                 printf("icmp_input from %s to %s, len %d\n",
 381                        buf, inet_ntoa(ip->ip_dst), icmplen);
 382         }
 383 #endif
 384         if (icmplen < ICMP_MINLEN) {
 385                 ICMPSTAT_INC(icps_tooshort);
 386                 goto freeit;
 387         }
 388         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 389         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 390                 ICMPSTAT_INC(icps_tooshort);
 391                 return;
 392         }
 393         ip = mtod(m, struct ip *);
 394         m->m_len -= hlen;
 395         m->m_data += hlen;
 396         icp = mtod(m, struct icmp *);
 397         if (in_cksum(m, icmplen)) {
 398                 ICMPSTAT_INC(icps_checksum);
 399                 goto freeit;
 400         }
 401         m->m_len += hlen;
 402         m->m_data -= hlen;
 403
 404         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 405                 /*
 406                  * Deliver very specific ICMP type only.
 407                  */
 408                 switch (icp->icmp_type) {
 409                 case ICMP_UNREACH:
 410                 case ICMP_TIMXCEED:
 411                         break;
 412                 default:
 413                         goto freeit;
 414                 }
 415         }
 416
 417 #ifdef ICMPPRINTFS
 418         if (icmpprintfs)
 419                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 420                     icp->icmp_code);
 421 #endif
 422
 423         /*
 424          * Message type specific processing.
 425          */
 426         if (icp->icmp_type > ICMP_MAXTYPE)
 427                 goto raw;
 428
 429         /* Initialize */
 430         bzero(&icmpsrc, sizeof(icmpsrc));
 431         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 432         icmpsrc.sin_family = AF_INET;
 433         bzero(&icmpdst, sizeof(icmpdst));
 434         icmpdst.sin_len = sizeof(struct sockaddr_in);
 435         icmpdst.sin_family = AF_INET;
 436         bzero(&icmpgw, sizeof(icmpgw));
 437         icmpgw.sin_len = sizeof(struct sockaddr_in);
 438         icmpgw.sin_family = AF_INET;
 439
 440         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 441         code = icp->icmp_code;
 442         switch (icp->icmp_type) {
 443
 444         case ICMP_UNREACH:
 445                 switch (code) {
 446                         case ICMP_UNREACH_NET:
 447                         case ICMP_UNREACH_HOST:
 448                         case ICMP_UNREACH_SRCFAIL:
 449                         case ICMP_UNREACH_NET_UNKNOWN:
 450                         case ICMP_UNREACH_HOST_UNKNOWN:
 451                         case ICMP_UNREACH_ISOLATED:
 452                         case ICMP_UNREACH_TOSNET:
 453                         case ICMP_UNREACH_TOSHOST:
 454                         case ICMP_UNREACH_HOST_PRECEDENCE:
 455                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 456                                 code = PRC_UNREACH_NET;
 457                                 break;
 458
 459                         case ICMP_UNREACH_NEEDFRAG:
 460                                 code = PRC_MSGSIZE;
 461                                 break;
 462
 463                         /*
 464                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 465                          * Treat subcodes 2,3 as immediate RST
 466                          */
 467                         case ICMP_UNREACH_PROTOCOL:
 468                         case ICMP_UNREACH_PORT:
 469                                 code = PRC_UNREACH_PORT;
 470                                 break;
 471
 472                         case ICMP_UNREACH_NET_PROHIB:
 473                         case ICMP_UNREACH_HOST_PROHIB:
 474                         case ICMP_UNREACH_FILTER_PROHIB:
 475                                 code = PRC_UNREACH_ADMIN_PROHIB;
 476                                 break;
 477
 478                         default:
 479                                 goto badcode;
 480                 }
 481                 goto deliver;
 482
 483         case ICMP_TIMXCEED:
 484                 if (code > 1)
 485                         goto badcode;
 486                 code += PRC_TIMXCEED_INTRANS;
 487                 goto deliver;
 488
 489         case ICMP_PARAMPROB:
 490                 if (code > 1)
 491                         goto badcode;
 492                 code = PRC_PARAMPROB;
 493                 goto deliver;
 494
 495         case ICMP_SOURCEQUENCH:
 496                 if (code)
 497                         goto badcode;
 498                 code = PRC_QUENCH;
 499         deliver:
 500                 /*
 501                  * Problem with datagram; advise higher level routines.
 502                  */
 503                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 504                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 505                         ICMPSTAT_INC(icps_badlen);
 506                         goto freeit;
 507                 }
 508                 /* Discard ICMP's in response to multicast packets */
 509                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 510                         goto badcode;
 511 #ifdef ICMPPRINTFS
 512                 if (icmpprintfs)
 513                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 514 #endif
 515                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 516                 /*
 517                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 518                  * notification to TCP layer.
 519                  */
 520                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 521                 if (ctlfunc)
 522                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 523                                    (void *)&icp->icmp_ip);
 524                 break;
 525
 526         badcode:
 527                 ICMPSTAT_INC(icps_badcode);
 528                 break;
 529
 530         case ICMP_ECHO:
 531                 if (!V_icmpbmcastecho
 532                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 533                         ICMPSTAT_INC(icps_bmcastecho);
 534                         break;
 535                 }
 536                 icp->icmp_type = ICMP_ECHOREPLY;
 537                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 538                         goto freeit;
 539                 else
 540                         goto reflect;
 541
 542         case ICMP_TSTAMP:
 543                 if (V_icmptstamprepl == 0)
 544                         break;
 545                 if (!V_icmpbmcastecho
 546                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 547                         ICMPSTAT_INC(icps_bmcasttstamp);
 548                         break;
 549                 }
 550                 if (icmplen < ICMP_TSLEN) {
 551                         ICMPSTAT_INC(icps_badlen);
 552                         break;
 553                 }
 554                 icp->icmp_type = ICMP_TSTAMPREPLY;
 555                 icp->icmp_rtime = iptime();
 556                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 557                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 558                         goto freeit;
 559                 else
 560                         goto reflect;
 561
 562         case ICMP_MASKREQ:
 563                 if (V_icmpmaskrepl == 0)
 564                         break;
 565                 /*
 566                  * We are not able to respond with all ones broadcast
 567                  * unless we receive it over a point-to-point interface.
 568                  */
 569                 if (icmplen < ICMP_MASKLEN)
 570                         break;
 571                 switch (ip->ip_dst.s_addr) {
 572
 573                 case INADDR_BROADCAST:
 574                 case INADDR_ANY:
 575                         icmpdst.sin_addr = ip->ip_src;
 576                         break;
 577
 578                 default:
 579                         icmpdst.sin_addr = ip->ip_dst;
 580                 }
 581                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 582                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 583                 if (ia == NULL)
 584                         break;
 585                 if (ia->ia_ifp == NULL) {
 586                         ifa_free(&ia->ia_ifa);
 587                         break;
 588                 }
 589                 icp->icmp_type = ICMP_MASKREPLY;
 590                 if (V_icmpmaskfake == 0)
 591                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 592                 else
 593                         icp->icmp_mask = V_icmpmaskfake;
 594                 if (ip->ip_src.s_addr == 0) {
 595                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 596                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 597                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 598                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 599                 }
 600                 ifa_free(&ia->ia_ifa);
 601 reflect:
 602                 ICMPSTAT_INC(icps_reflect);
 603                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 604                 icmp_reflect(m);
 605                 return;
 606
 607         case ICMP_REDIRECT:
 608                 if (V_log_redirect) {
 609                         u_long src, dst, gw;
 610
 611                         src = ntohl(ip->ip_src.s_addr);
 612                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 613                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 614                         printf("icmp redirect from %d.%d.%d.%d: "
 615                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 616                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 617                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 618                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 619                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 620                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 621                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 622                 }
 623                 /*
 624                  * RFC1812 says we must ignore ICMP redirects if we
 625                  * are acting as router.
 626                  */
 627                 if (V_drop_redirect || V_ipforwarding)
 628                         break;
 629                 if (code > 3)
 630                         goto badcode;
 631                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 632                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 633                         ICMPSTAT_INC(icps_badlen);
 634                         break;
 635                 }
 636                 /*
 637                  * Short circuit routing redirects to force
 638                  * immediate change in the kernel's routing
 639                  * tables.  The message is also handed to anyone
 640                  * listening on a raw socket (e.g. the routing
 641                  * daemon for use in updating its tables).
 642                  */
 643                 icmpgw.sin_addr = ip->ip_src;
 644                 icmpdst.sin_addr = icp->icmp_gwaddr;
 645 #ifdef  ICMPPRINTFS
 646                 if (icmpprintfs) {
 647                         char buf[4 * sizeof "123"];
 648                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
 649
 650                         printf("redirect dst %s to %s\n",
 651                                buf, inet_ntoa(icp->icmp_gwaddr));
 652                 }
 653 #endif
 654                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 655                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 656                         in_rtredirect((struct sockaddr *)&icmpsrc,
 657                           (struct sockaddr *)&icmpdst,
 658                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 659                           (struct sockaddr *)&icmpgw, fibnum);
 660                 }
 661                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 662                 break;
 663
 664         /*
 665          * No kernel processing for the following;
 666          * just fall through to send to raw listener.
 667          */
 668         case ICMP_ECHOREPLY:
 669         case ICMP_ROUTERADVERT:
 670         case ICMP_ROUTERSOLICIT:
 671         case ICMP_TSTAMPREPLY:
 672         case ICMP_IREQREPLY:
 673         case ICMP_MASKREPLY:
 674         default:
 675                 break;
 676         }
 677
 678 raw:
 679         rip_input(m, off);
 680         return;
 681
 682 freeit:
 683         m_freem(m);
 684 }
 685
 686 /*
 687  * Reflect the ip packet back to the source
 688  */
 689 static void
 690 icmp_reflect(struct mbuf *m)
 691 {
 692         struct ip *ip = mtod(m, struct ip *);
 693         struct ifaddr *ifa;
 694         struct ifnet *ifp;
 695         struct in_ifaddr *ia;
 696         struct in_addr t;
 697         struct mbuf *opts = 0;
 698         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 699
 700         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 701             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 702             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 703                 m_freem(m);     /* Bad return address */
 704                 ICMPSTAT_INC(icps_badaddr);
 705                 goto done;      /* Ip_output() will check for broadcast */
 706         }
 707
 708         t = ip->ip_dst;
 709         ip->ip_dst = ip->ip_src;
 710
 711         /*
 712          * Source selection for ICMP replies:
 713          *
 714          * If the incoming packet was addressed directly to one of our
 715          * own addresses, use dst as the src for the reply.
 716          */
 717         IN_IFADDR_RLOCK();
 718         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 719                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 720                         t = IA_SIN(ia)->sin_addr;
 721                         IN_IFADDR_RUNLOCK();
 722                         goto match;
 723                 }
 724         }
 725         IN_IFADDR_RUNLOCK();
 726
 727         /*
 728          * If the incoming packet was addressed to one of our broadcast
 729          * addresses, use the first non-broadcast address which corresponds
 730          * to the incoming interface.
 731          */
 732         ifp = m->m_pkthdr.rcvif;
 733         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 734                 IF_ADDR_RLOCK(ifp);
 735                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 736                         if (ifa->ifa_addr->sa_family != AF_INET)
 737                                 continue;
 738                         ia = ifatoia(ifa);
 739                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 740                             t.s_addr) {
 741                                 t = IA_SIN(ia)->sin_addr;
 742                                 IF_ADDR_RUNLOCK(ifp);
 743                                 goto match;
 744                         }
 745                 }
 746                 IF_ADDR_RUNLOCK(ifp);
 747         }
 748         /*
 749          * If the packet was transiting through us, use the address of
 750          * the interface the packet came through in.  If that interface
 751          * doesn't have a suitable IP address, the normal selection
 752          * criteria apply.
 753          */
 754         if (V_icmp_rfi && ifp != NULL) {
 755                 IF_ADDR_RLOCK(ifp);
 756                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 757                         if (ifa->ifa_addr->sa_family != AF_INET)
 758                                 continue;
 759                         ia = ifatoia(ifa);
 760                         t = IA_SIN(ia)->sin_addr;
 761                         IF_ADDR_RUNLOCK(ifp);
 762                         goto match;
 763                 }
 764                 IF_ADDR_RUNLOCK(ifp);
 765         }
 766         /*
 767          * If the incoming packet was not addressed directly to us, use
 768          * designated interface for icmp replies specified by sysctl
 769          * net.inet.icmp.reply_src (default not set). Otherwise continue
 770          * with normal source selection.
 771          */
 772         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 773                 IF_ADDR_RLOCK(ifp);
 774                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 775                         if (ifa->ifa_addr->sa_family != AF_INET)
 776                                 continue;
 777                         ia = ifatoia(ifa);
 778                         t = IA_SIN(ia)->sin_addr;
 779                         IF_ADDR_RUNLOCK(ifp);
 780                         goto match;
 781                 }
 782                 IF_ADDR_RUNLOCK(ifp);
 783         }
 784         /*
 785          * If the packet was transiting through us, use the address of
 786          * the interface that is the closest to the packet source.
 787          * When we don't have a route back to the packet source, stop here
 788          * and drop the packet.
 789          */
 790         ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
 791         if (ia == NULL) {
 792                 m_freem(m);
 793                 ICMPSTAT_INC(icps_noroute);
 794                 goto done;
 795         }
 796         t = IA_SIN(ia)->sin_addr;
 797         ifa_free(&ia->ia_ifa);
 798 match:
 799 #ifdef MAC
 800         mac_netinet_icmp_replyinplace(m);
 801 #endif
 802         ip->ip_src = t;
 803         ip->ip_ttl = V_ip_defttl;
 804
 805         if (optlen > 0) {
 806                 register u_char *cp;
 807                 int opt, cnt;
 808                 u_int len;
 809
 810                 /*
 811                  * Retrieve any source routing from the incoming packet;
 812                  * add on any record-route or timestamp options.
 813                  */
 814                 cp = (u_char *) (ip + 1);
 815                 if ((opts = ip_srcroute(m)) == 0 &&
 816                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 817                         opts->m_len = sizeof(struct in_addr);
 818                         mtod(opts, struct in_addr *)->s_addr = 0;
 819                 }
 820                 if (opts) {
 821 #ifdef ICMPPRINTFS
 822                     if (icmpprintfs)
 823                             printf("icmp_reflect optlen %d rt %d => ",
 824                                 optlen, opts->m_len);
 825 #endif
 826                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 827                             opt = cp[IPOPT_OPTVAL];
 828                             if (opt == IPOPT_EOL)
 829                                     break;
 830                             if (opt == IPOPT_NOP)
 831                                     len = 1;
 832                             else {
 833                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 834                                             break;
 835                                     len = cp[IPOPT_OLEN];
 836                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 837                                         len > cnt)
 838                                             break;
 839                             }
 840                             /*
 841                              * Should check for overflow, but it "can't happen"
 842                              */
 843                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 844                                 opt == IPOPT_SECURITY) {
 845                                     bcopy((caddr_t)cp,
 846                                         mtod(opts, caddr_t) + opts->m_len, len);
 847                                     opts->m_len += len;
 848                             }
 849                     }
 850                     /* Terminate & pad, if necessary */
 851                     cnt = opts->m_len % 4;
 852                     if (cnt) {
 853                             for (; cnt < 4; cnt++) {
 854                                     *(mtod(opts, caddr_t) + opts->m_len) =
 855                                         IPOPT_EOL;
 856                                     opts->m_len++;
 857                             }
 858                     }
 859 #ifdef ICMPPRINTFS
 860                     if (icmpprintfs)
 861                             printf("%d\n", opts->m_len);
 862 #endif
 863                 }
 864                 ip_stripoptions(m);
 865         }
 866         m_tag_delete_nonpersistent(m);
 867         m->m_flags &= ~(M_BCAST|M_MCAST);
 868         icmp_send(m, opts);
 869 done:
 870         if (opts)
 871                 (void)m_free(opts);
 872 }
 873
 874 /*
 875  * Send an icmp packet back to the ip level,
 876  * after supplying a checksum.
 877  */
 878 static void
 879 icmp_send(struct mbuf *m, struct mbuf *opts)
 880 {
 881         register struct ip *ip = mtod(m, struct ip *);
 882         register int hlen;
 883         register struct icmp *icp;
 884
 885         hlen = ip->ip_hl << 2;
 886         m->m_data += hlen;
 887         m->m_len -= hlen;
 888         icp = mtod(m, struct icmp *);
 889         icp->icmp_cksum = 0;
 890         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 891         m->m_data -= hlen;
 892         m->m_len += hlen;
 893         m->m_pkthdr.rcvif = (struct ifnet *)0;
 894 #ifdef ICMPPRINTFS
 895         if (icmpprintfs) {
 896                 char buf[4 * sizeof "123"];
 897                 strcpy(buf, inet_ntoa(ip->ip_dst));
 898                 printf("icmp_send dst %s src %s\n",
 899                        buf, inet_ntoa(ip->ip_src));
 900         }
 901 #endif
 902         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 903 }
 904
 905 /*
 906  * Return milliseconds since 00:00 GMT in network format.
 907  */
 908 uint32_t
 909 iptime(void)
 910 {
 911         struct timeval atv;
 912         u_long t;
 913
 914         getmicrotime(&atv);
 915         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 916         return (htonl(t));
 917 }
 918
 919 /*
 920  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 921  * given current value MTU.  If DIR is less than zero, a larger plateau
 922  * is returned; otherwise, a smaller value is returned.
 923  */
 924 int
 925 ip_next_mtu(int mtu, int dir)
 926 {
 927         static int mtutab[] = {
 928                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 929                 296, 68, 0
 930         };
 931         int i, size;
 932
 933         size = (sizeof mtutab) / (sizeof mtutab[0]);
 934         if (dir >= 0) {
 935                 for (i = 0; i < size; i++)
 936                         if (mtu > mtutab[i])
 937                                 return mtutab[i];
 938         } else {
 939                 for (i = size - 1; i >= 0; i--)
 940                         if (mtu < mtutab[i])
 941                                 return mtutab[i];
 942                 if (mtu == mtutab[0])
 943                         return mtutab[0];
 944         }
 945         return 0;
 946 }
 947 #endif /* INET */
 948
 949
 950 /*
 951  * badport_bandlim() - check for ICMP bandwidth limit
 952  *
 953  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 954  *      hit our bandwidth limit and it is not ok.
 955  *
 956  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 957  *
 958  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 959  *      values.  We may eventually remove this separation (and simplify the
 960  *      code further).
 961  *
 962  *      Note that the printing of the error message is delayed so we can
 963  *      properly print the icmp error rate that the system was trying to do
 964  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 965  *      the 'final' error, but it doesn't make sense to solve the printing
 966  *      delay with more complex code.
 967  */
 968
 969 int
 970 badport_bandlim(int which)
 971 {
 972
 973 #define N(a)    (sizeof (a) / sizeof (a[0]))
 974         static struct rate {
 975                 const char      *type;
 976                 struct timeval  lasttime;
 977                 int             curpps;
 978         } rates[BANDLIM_MAX+1] = {
 979                 { "icmp unreach response" },
 980                 { "icmp ping response" },
 981                 { "icmp tstamp response" },
 982                 { "closed port RST response" },
 983                 { "open port RST response" },
 984                 { "icmp6 unreach response" },
 985                 { "sctp ootb response" }
 986         };
 987
 988         /*
 989          * Return ok status if feature disabled or argument out of range.
 990          */
 991         if (V_icmplim > 0 && (u_int) which < N(rates)) {
 992                 struct rate *r = &rates[which];
 993                 int opps = r->curpps;
 994
 995                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
 996                         return -1;      /* discard packet */
 997                 /*
 998                  * If we've dropped below the threshold after having
 999                  * rate-limited traffic print the message.  This preserves
1000                  * the previous behaviour at the expense of added complexity.
1001                  */
1002                 if (V_icmplim_output && opps > V_icmplim)
1003                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
1004                                 r->type, opps, V_icmplim);
1005         }
1006         return 0;                       /* okay to send packet */
1007 #undef N
1008 }