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 #include "opt_ipsec.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/mbuf.h>
  41 #include <sys/protosw.h>
  42 #include <sys/socket.h>
  43 #include <sys/time.h>
  44 #include <sys/kernel.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/syslog.h>
  47
  48 #include <net/if.h>
  49 #include <net/if_types.h>
  50 #include <net/route.h>
  51 #include <net/vnet.h>
  52
  53 #include <netinet/in.h>
  54 #include <netinet/in_pcb.h>
  55 #include <netinet/in_systm.h>
  56 #include <netinet/in_var.h>
  57 #include <netinet/ip.h>
  58 #include <netinet/ip_icmp.h>
  59 #include <netinet/ip_var.h>
  60 #include <netinet/ip_options.h>
  61 #include <netinet/tcp.h>
  62 #include <netinet/tcp_var.h>
  63 #include <netinet/tcpip.h>
  64 #include <netinet/icmp_var.h>
  65
  66 #ifdef INET
  67 #ifdef IPSEC
  68 #include <netipsec/ipsec.h>
  69 #include <netipsec/key.h>
  70 #endif
  71
  72 #include <machine/in_cksum.h>
  73
  74 #include <security/mac/mac_framework.h>
  75 #endif /* INET */
  76
  77 /*
  78  * ICMP routines: error generation, receive packet processing, and
  79  * routines to turnaround packets back to the originator, and
  80  * host table maintenance routines.
  81  */
  82 static VNET_DEFINE(int, icmplim) = 200;
  83 #define V_icmplim                       VNET(icmplim)
  84 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_RW,
  85         &VNET_NAME(icmplim), 0,
  86         "Maximum number of ICMP responses per second");
  87
  88 static VNET_DEFINE(int, icmplim_output) = 1;
  89 #define V_icmplim_output                VNET(icmplim_output)
  90 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_RW,
  91         &VNET_NAME(icmplim_output), 0,
  92         "Enable logging of ICMP response rate limiting");
  93
  94 #ifdef INET
  95 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
  96 VNET_PCPUSTAT_SYSINIT(icmpstat);
  97 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
  98     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
  99
 100 #ifdef VIMAGE
 101 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
 102 #endif /* VIMAGE */
 103
 104 static VNET_DEFINE(int, icmpmaskrepl) = 0;
 105 #define V_icmpmaskrepl                  VNET(icmpmaskrepl)
 106 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_RW,
 107         &VNET_NAME(icmpmaskrepl), 0,
 108         "Reply to ICMP Address Mask Request packets.");
 109
 110 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
 111 #define V_icmpmaskfake                  VNET(icmpmaskfake)
 112 SYSCTL_VNET_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_RW,
 113         &VNET_NAME(icmpmaskfake), 0,
 114         "Fake reply to ICMP Address Mask Request packets.");
 115
 116 VNET_DEFINE(int, drop_redirect) = 0;
 117
 118 static VNET_DEFINE(int, log_redirect) = 0;
 119 #define V_log_redirect                  VNET(log_redirect)
 120 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_RW,
 121         &VNET_NAME(log_redirect), 0,
 122         "Log ICMP redirects to the console");
 123
 124 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
 125 #define V_reply_src                     VNET(reply_src)
 126 SYSCTL_VNET_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_RW,
 127         &VNET_NAME(reply_src), IFNAMSIZ,
 128         "icmp reply source for non-local packets.");
 129
 130 static VNET_DEFINE(int, icmp_rfi) = 0;
 131 #define V_icmp_rfi                      VNET(icmp_rfi)
 132 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_RW,
 133         &VNET_NAME(icmp_rfi), 0,
 134         "ICMP reply from incoming interface for non-local packets");
 135
 136 static VNET_DEFINE(int, icmp_quotelen) = 8;
 137 #define V_icmp_quotelen                 VNET(icmp_quotelen)
 138 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_RW,
 139         &VNET_NAME(icmp_quotelen), 0,
 140         "Number of bytes from original packet to quote in ICMP reply");
 141
 142 /*
 143  * ICMP broadcast echo sysctl
 144  */
 145 static VNET_DEFINE(int, icmpbmcastecho) = 0;
 146 #define V_icmpbmcastecho                VNET(icmpbmcastecho)
 147 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_RW,
 148         &VNET_NAME(icmpbmcastecho), 0,
 149         "");
 150
 151
 152 #ifdef ICMPPRINTFS
 153 int     icmpprintfs = 0;
 154 #endif
 155
 156 static void     icmp_reflect(struct mbuf *);
 157 static void     icmp_send(struct mbuf *, struct mbuf *);
 158
 159 extern  struct protosw inetsw[];
 160
 161 static int
 162 sysctl_net_icmp_drop_redir(SYSCTL_HANDLER_ARGS)
 163 {
 164         int error, new;
 165         int i;
 166         struct radix_node_head *rnh;
 167
 168         new = V_drop_redirect;
 169         error = sysctl_handle_int(oidp, &new, 0, req);
 170         if (error == 0 && req->newptr) {
 171                 new = (new != 0) ? 1 : 0;
 172
 173                 if (new == V_drop_redirect)
 174                         return (0);
 175
 176                 for (i = 0; i < rt_numfibs; i++) {
 177                         if ((rnh = rt_tables_get_rnh(i, AF_INET)) == NULL)
 178                                 continue;
 179                         RADIX_NODE_HEAD_LOCK(rnh);
 180                         in_setmatchfunc(rnh, new);
 181                         RADIX_NODE_HEAD_UNLOCK(rnh);
 182                 }
 183
 184                 V_drop_redirect = new;
 185         }
 186
 187         return (error);
 188 }
 189
 190 SYSCTL_VNET_PROC(_net_inet_icmp, OID_AUTO, drop_redirect,
 191     CTLTYPE_INT|CTLFLAG_RW, 0, 0,
 192     sysctl_net_icmp_drop_redir, "I", "Ignore ICMP redirects");
 193
 194 /*
 195  * Kernel module interface for updating icmpstat.  The argument is an index
 196  * into icmpstat treated as an array of u_long.  While this encodes the
 197  * general layout of icmpstat into the caller, it doesn't encode its
 198  * location, so that future changes to add, for example, per-CPU stats
 199  * support won't cause binary compatibility problems for kernel modules.
 200  */
 201 void
 202 kmod_icmpstat_inc(int statnum)
 203 {
 204
 205         counter_u64_add(VNET(icmpstat)[statnum], 1);
 206 }
 207
 208 /*
 209  * Generate an error packet of type error
 210  * in response to bad packet ip.
 211  */
 212 void
 213 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 214 {
 215         register struct ip *oip = mtod(n, struct ip *), *nip;
 216         register unsigned oiphlen = oip->ip_hl << 2;
 217         register struct icmp *icp;
 218         register struct mbuf *m;
 219         unsigned icmplen, icmpelen, nlen;
 220
 221         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
 222 #ifdef ICMPPRINTFS
 223         if (icmpprintfs)
 224                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 225 #endif
 226         if (type != ICMP_REDIRECT)
 227                 ICMPSTAT_INC(icps_error);
 228         /*
 229          * Don't send error:
 230          *  if the original packet was encrypted.
 231          *  if not the first fragment of message.
 232          *  in response to a multicast or broadcast packet.
 233          *  if the old packet protocol was an ICMP error message.
 234          */
 235         if (n->m_flags & M_DECRYPTED)
 236                 goto freeit;
 237         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 238                 goto freeit;
 239         if (n->m_flags & (M_BCAST|M_MCAST))
 240                 goto freeit;
 241         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 242           n->m_len >= oiphlen + ICMP_MINLEN &&
 243           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
 244                 ICMPSTAT_INC(icps_oldicmp);
 245                 goto freeit;
 246         }
 247         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
 248         if (oiphlen + 8 > n->m_len)
 249                 goto freeit;
 250         /*
 251          * Calculate length to quote from original packet and
 252          * prevent the ICMP mbuf from overflowing.
 253          * Unfortunatly this is non-trivial since ip_forward()
 254          * sends us truncated packets.
 255          */
 256         nlen = m_length(n, NULL);
 257         if (oip->ip_p == IPPROTO_TCP) {
 258                 struct tcphdr *th;
 259                 int tcphlen;
 260
 261                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 262                     n->m_next == NULL)
 263                         goto stdreply;
 264                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 265                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
 266                         goto freeit;
 267                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
 268                 tcphlen = th->th_off << 2;
 269                 if (tcphlen < sizeof(struct tcphdr))
 270                         goto freeit;
 271                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 272                         goto freeit;
 273                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 274                         goto stdreply;
 275                 if (n->m_len < oiphlen + tcphlen &&
 276                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
 277                         goto freeit;
 278                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 279                     ntohs(oip->ip_len) - oiphlen));
 280         } else
 281 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 282
 283         icmplen = min(oiphlen + icmpelen, nlen);
 284         if (icmplen < sizeof(struct ip))
 285                 goto freeit;
 286
 287         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 288                 m = m_gethdr(M_NOWAIT, MT_DATA);
 289         else
 290                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 291         if (m == NULL)
 292                 goto freeit;
 293 #ifdef MAC
 294         mac_netinet_icmp_reply(n, m);
 295 #endif
 296         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
 297         m_align(m, ICMP_MINLEN + icmplen);
 298         m->m_len = ICMP_MINLEN + icmplen;
 299
 300         /* XXX MRT  make the outgoing packet use the same FIB
 301          * that was associated with the incoming packet
 302          */
 303         M_SETFIB(m, M_GETFIB(n));
 304         icp = mtod(m, struct icmp *);
 305         ICMPSTAT_INC(icps_outhist[type]);
 306         icp->icmp_type = type;
 307         if (type == ICMP_REDIRECT)
 308                 icp->icmp_gwaddr.s_addr = dest;
 309         else {
 310                 icp->icmp_void = 0;
 311                 /*
 312                  * The following assignments assume an overlay with the
 313                  * just zeroed icmp_void field.
 314                  */
 315                 if (type == ICMP_PARAMPROB) {
 316                         icp->icmp_pptr = code;
 317                         code = 0;
 318                 } else if (type == ICMP_UNREACH &&
 319                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 320                         icp->icmp_nextmtu = htons(mtu);
 321                 }
 322         }
 323         icp->icmp_code = code;
 324
 325         /*
 326          * Copy the quotation into ICMP message and
 327          * convert quoted IP header back to network representation.
 328          */
 329         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 330         nip = &icp->icmp_ip;
 331
 332         /*
 333          * Set up ICMP message mbuf and copy old IP header (without options
 334          * in front of ICMP message.
 335          * If the original mbuf was meant to bypass the firewall, the error
 336          * reply should bypass as well.
 337          */
 338         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 339         m->m_data -= sizeof(struct ip);
 340         m->m_len += sizeof(struct ip);
 341         m->m_pkthdr.len = m->m_len;
 342         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 343         nip = mtod(m, struct ip *);
 344         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 345         nip->ip_len = htons(m->m_len);
 346         nip->ip_v = IPVERSION;
 347         nip->ip_hl = 5;
 348         nip->ip_p = IPPROTO_ICMP;
 349         nip->ip_tos = 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_icmpbmcastecho
 544                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 545                         ICMPSTAT_INC(icps_bmcasttstamp);
 546                         break;
 547                 }
 548                 if (icmplen < ICMP_TSLEN) {
 549                         ICMPSTAT_INC(icps_badlen);
 550                         break;
 551                 }
 552                 icp->icmp_type = ICMP_TSTAMPREPLY;
 553                 icp->icmp_rtime = iptime();
 554                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 555                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 556                         goto freeit;
 557                 else
 558                         goto reflect;
 559
 560         case ICMP_MASKREQ:
 561                 if (V_icmpmaskrepl == 0)
 562                         break;
 563                 /*
 564                  * We are not able to respond with all ones broadcast
 565                  * unless we receive it over a point-to-point interface.
 566                  */
 567                 if (icmplen < ICMP_MASKLEN)
 568                         break;
 569                 switch (ip->ip_dst.s_addr) {
 570
 571                 case INADDR_BROADCAST:
 572                 case INADDR_ANY:
 573                         icmpdst.sin_addr = ip->ip_src;
 574                         break;
 575
 576                 default:
 577                         icmpdst.sin_addr = ip->ip_dst;
 578                 }
 579                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 580                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 581                 if (ia == NULL)
 582                         break;
 583                 if (ia->ia_ifp == NULL) {
 584                         ifa_free(&ia->ia_ifa);
 585                         break;
 586                 }
 587                 icp->icmp_type = ICMP_MASKREPLY;
 588                 if (V_icmpmaskfake == 0)
 589                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 590                 else
 591                         icp->icmp_mask = V_icmpmaskfake;
 592                 if (ip->ip_src.s_addr == 0) {
 593                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 594                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 595                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 596                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 597                 }
 598                 ifa_free(&ia->ia_ifa);
 599 reflect:
 600                 ICMPSTAT_INC(icps_reflect);
 601                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 602                 icmp_reflect(m);
 603                 return;
 604
 605         case ICMP_REDIRECT:
 606                 if (V_log_redirect) {
 607                         u_long src, dst, gw;
 608
 609                         src = ntohl(ip->ip_src.s_addr);
 610                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 611                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 612                         printf("icmp redirect from %d.%d.%d.%d: "
 613                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 614                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 615                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 616                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 617                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 618                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 619                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 620                 }
 621                 /*
 622                  * RFC1812 says we must ignore ICMP redirects if we
 623                  * are acting as router.
 624                  */
 625                 if (V_drop_redirect || V_ipforwarding)
 626                         break;
 627                 if (code > 3)
 628                         goto badcode;
 629                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 630                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 631                         ICMPSTAT_INC(icps_badlen);
 632                         break;
 633                 }
 634                 /*
 635                  * Short circuit routing redirects to force
 636                  * immediate change in the kernel's routing
 637                  * tables.  The message is also handed to anyone
 638                  * listening on a raw socket (e.g. the routing
 639                  * daemon for use in updating its tables).
 640                  */
 641                 icmpgw.sin_addr = ip->ip_src;
 642                 icmpdst.sin_addr = icp->icmp_gwaddr;
 643 #ifdef  ICMPPRINTFS
 644                 if (icmpprintfs) {
 645                         char buf[4 * sizeof "123"];
 646                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
 647
 648                         printf("redirect dst %s to %s\n",
 649                                buf, inet_ntoa(icp->icmp_gwaddr));
 650                 }
 651 #endif
 652                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 653                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 654                         in_rtredirect((struct sockaddr *)&icmpsrc,
 655                           (struct sockaddr *)&icmpdst,
 656                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 657                           (struct sockaddr *)&icmpgw, fibnum);
 658                 }
 659                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 660 #ifdef IPSEC
 661                 key_sa_routechange((struct sockaddr *)&icmpsrc);
 662 #endif
 663                 break;
 664
 665         /*
 666          * No kernel processing for the following;
 667          * just fall through to send to raw listener.
 668          */
 669         case ICMP_ECHOREPLY:
 670         case ICMP_ROUTERADVERT:
 671         case ICMP_ROUTERSOLICIT:
 672         case ICMP_TSTAMPREPLY:
 673         case ICMP_IREQREPLY:
 674         case ICMP_MASKREPLY:
 675         default:
 676                 break;
 677         }
 678
 679 raw:
 680         rip_input(m, off);
 681         return;
 682
 683 freeit:
 684         m_freem(m);
 685 }
 686
 687 /*
 688  * Reflect the ip packet back to the source
 689  */
 690 static void
 691 icmp_reflect(struct mbuf *m)
 692 {
 693         struct ip *ip = mtod(m, struct ip *);
 694         struct ifaddr *ifa;
 695         struct ifnet *ifp;
 696         struct in_ifaddr *ia;
 697         struct in_addr t;
 698         struct mbuf *opts = 0;
 699         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 700
 701         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 702             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 703             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 704                 m_freem(m);     /* Bad return address */
 705                 ICMPSTAT_INC(icps_badaddr);
 706                 goto done;      /* Ip_output() will check for broadcast */
 707         }
 708
 709         t = ip->ip_dst;
 710         ip->ip_dst = ip->ip_src;
 711
 712         /*
 713          * Source selection for ICMP replies:
 714          *
 715          * If the incoming packet was addressed directly to one of our
 716          * own addresses, use dst as the src for the reply.
 717          */
 718         IN_IFADDR_RLOCK();
 719         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 720                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 721                         t = IA_SIN(ia)->sin_addr;
 722                         IN_IFADDR_RUNLOCK();
 723                         goto match;
 724                 }
 725         }
 726         IN_IFADDR_RUNLOCK();
 727
 728         /*
 729          * If the incoming packet was addressed to one of our broadcast
 730          * addresses, use the first non-broadcast address which corresponds
 731          * to the incoming interface.
 732          */
 733         ifp = m->m_pkthdr.rcvif;
 734         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 735                 IF_ADDR_RLOCK(ifp);
 736                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 737                         if (ifa->ifa_addr->sa_family != AF_INET)
 738                                 continue;
 739                         ia = ifatoia(ifa);
 740                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 741                             t.s_addr) {
 742                                 t = IA_SIN(ia)->sin_addr;
 743                                 IF_ADDR_RUNLOCK(ifp);
 744                                 goto match;
 745                         }
 746                 }
 747                 IF_ADDR_RUNLOCK(ifp);
 748         }
 749         /*
 750          * If the packet was transiting through us, use the address of
 751          * the interface the packet came through in.  If that interface
 752          * doesn't have a suitable IP address, the normal selection
 753          * criteria apply.
 754          */
 755         if (V_icmp_rfi && ifp != NULL) {
 756                 IF_ADDR_RLOCK(ifp);
 757                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 758                         if (ifa->ifa_addr->sa_family != AF_INET)
 759                                 continue;
 760                         ia = ifatoia(ifa);
 761                         t = IA_SIN(ia)->sin_addr;
 762                         IF_ADDR_RUNLOCK(ifp);
 763                         goto match;
 764                 }
 765                 IF_ADDR_RUNLOCK(ifp);
 766         }
 767         /*
 768          * If the incoming packet was not addressed directly to us, use
 769          * designated interface for icmp replies specified by sysctl
 770          * net.inet.icmp.reply_src (default not set). Otherwise continue
 771          * with normal source selection.
 772          */
 773         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 774                 IF_ADDR_RLOCK(ifp);
 775                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 776                         if (ifa->ifa_addr->sa_family != AF_INET)
 777                                 continue;
 778                         ia = ifatoia(ifa);
 779                         t = IA_SIN(ia)->sin_addr;
 780                         IF_ADDR_RUNLOCK(ifp);
 781                         goto match;
 782                 }
 783                 IF_ADDR_RUNLOCK(ifp);
 784         }
 785         /*
 786          * If the packet was transiting through us, use the address of
 787          * the interface that is the closest to the packet source.
 788          * When we don't have a route back to the packet source, stop here
 789          * and drop the packet.
 790          */
 791         ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
 792         if (ia == NULL) {
 793                 m_freem(m);
 794                 ICMPSTAT_INC(icps_noroute);
 795                 goto done;
 796         }
 797         t = IA_SIN(ia)->sin_addr;
 798         ifa_free(&ia->ia_ifa);
 799 match:
 800 #ifdef MAC
 801         mac_netinet_icmp_replyinplace(m);
 802 #endif
 803         ip->ip_src = t;
 804         ip->ip_ttl = V_ip_defttl;
 805
 806         if (optlen > 0) {
 807                 register u_char *cp;
 808                 int opt, cnt;
 809                 u_int len;
 810
 811                 /*
 812                  * Retrieve any source routing from the incoming packet;
 813                  * add on any record-route or timestamp options.
 814                  */
 815                 cp = (u_char *) (ip + 1);
 816                 if ((opts = ip_srcroute(m)) == 0 &&
 817                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 818                         opts->m_len = sizeof(struct in_addr);
 819                         mtod(opts, struct in_addr *)->s_addr = 0;
 820                 }
 821                 if (opts) {
 822 #ifdef ICMPPRINTFS
 823                     if (icmpprintfs)
 824                             printf("icmp_reflect optlen %d rt %d => ",
 825                                 optlen, opts->m_len);
 826 #endif
 827                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 828                             opt = cp[IPOPT_OPTVAL];
 829                             if (opt == IPOPT_EOL)
 830                                     break;
 831                             if (opt == IPOPT_NOP)
 832                                     len = 1;
 833                             else {
 834                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 835                                             break;
 836                                     len = cp[IPOPT_OLEN];
 837                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 838                                         len > cnt)
 839                                             break;
 840                             }
 841                             /*
 842                              * Should check for overflow, but it "can't happen"
 843                              */
 844                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 845                                 opt == IPOPT_SECURITY) {
 846                                     bcopy((caddr_t)cp,
 847                                         mtod(opts, caddr_t) + opts->m_len, len);
 848                                     opts->m_len += len;
 849                             }
 850                     }
 851                     /* Terminate & pad, if necessary */
 852                     cnt = opts->m_len % 4;
 853                     if (cnt) {
 854                             for (; cnt < 4; cnt++) {
 855                                     *(mtod(opts, caddr_t) + opts->m_len) =
 856                                         IPOPT_EOL;
 857                                     opts->m_len++;
 858                             }
 859                     }
 860 #ifdef ICMPPRINTFS
 861                     if (icmpprintfs)
 862                             printf("%d\n", opts->m_len);
 863 #endif
 864                 }
 865                 ip_stripoptions(m);
 866         }
 867         m_tag_delete_nonpersistent(m);
 868         m->m_flags &= ~(M_BCAST|M_MCAST);
 869         icmp_send(m, opts);
 870 done:
 871         if (opts)
 872                 (void)m_free(opts);
 873 }
 874
 875 /*
 876  * Send an icmp packet back to the ip level,
 877  * after supplying a checksum.
 878  */
 879 static void
 880 icmp_send(struct mbuf *m, struct mbuf *opts)
 881 {
 882         register struct ip *ip = mtod(m, struct ip *);
 883         register int hlen;
 884         register struct icmp *icp;
 885
 886         hlen = ip->ip_hl << 2;
 887         m->m_data += hlen;
 888         m->m_len -= hlen;
 889         icp = mtod(m, struct icmp *);
 890         icp->icmp_cksum = 0;
 891         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 892         m->m_data -= hlen;
 893         m->m_len += hlen;
 894         m->m_pkthdr.rcvif = (struct ifnet *)0;
 895 #ifdef ICMPPRINTFS
 896         if (icmpprintfs) {
 897                 char buf[4 * sizeof "123"];
 898                 strcpy(buf, inet_ntoa(ip->ip_dst));
 899                 printf("icmp_send dst %s src %s\n",
 900                        buf, inet_ntoa(ip->ip_src));
 901         }
 902 #endif
 903         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 904 }
 905
 906 /*
 907  * Return milliseconds since 00:00 GMT in network format.
 908  */
 909 uint32_t
 910 iptime(void)
 911 {
 912         struct timeval atv;
 913         u_long t;
 914
 915         getmicrotime(&atv);
 916         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 917         return (htonl(t));
 918 }
 919
 920 /*
 921  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 922  * given current value MTU.  If DIR is less than zero, a larger plateau
 923  * is returned; otherwise, a smaller value is returned.
 924  */
 925 int
 926 ip_next_mtu(int mtu, int dir)
 927 {
 928         static int mtutab[] = {
 929                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 930                 296, 68, 0
 931         };
 932         int i, size;
 933
 934         size = (sizeof mtutab) / (sizeof mtutab[0]);
 935         if (dir >= 0) {
 936                 for (i = 0; i < size; i++)
 937                         if (mtu > mtutab[i])
 938                                 return mtutab[i];
 939         } else {
 940                 for (i = size - 1; i >= 0; i--)
 941                         if (mtu < mtutab[i])
 942                                 return mtutab[i];
 943                 if (mtu == mtutab[0])
 944                         return mtutab[0];
 945         }
 946         return 0;
 947 }
 948 #endif /* INET */
 949
 950
 951 /*
 952  * badport_bandlim() - check for ICMP bandwidth limit
 953  *
 954  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 955  *      hit our bandwidth limit and it is not ok.
 956  *
 957  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 958  *
 959  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 960  *      values.  We may eventually remove this separation (and simplify the
 961  *      code further).
 962  *
 963  *      Note that the printing of the error message is delayed so we can
 964  *      properly print the icmp error rate that the system was trying to do
 965  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 966  *      the 'final' error, but it doesn't make sense to solve the printing
 967  *      delay with more complex code.
 968  */
 969
 970 int
 971 badport_bandlim(int which)
 972 {
 973
 974 #define N(a)    (sizeof (a) / sizeof (a[0]))
 975         static struct rate {
 976                 const char      *type;
 977                 struct timeval  lasttime;
 978                 int             curpps;
 979         } rates[BANDLIM_MAX+1] = {
 980                 { "icmp unreach response" },
 981                 { "icmp ping response" },
 982                 { "icmp tstamp response" },
 983                 { "closed port RST response" },
 984                 { "open port RST response" },
 985                 { "icmp6 unreach response" },
 986                 { "sctp ootb response" }
 987         };
 988
 989         /*
 990          * Return ok status if feature disabled or argument out of range.
 991          */
 992         if (V_icmplim > 0 && (u_int) which < N(rates)) {
 993                 struct rate *r = &rates[which];
 994                 int opps = r->curpps;
 995
 996                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
 997                         return -1;      /* discard packet */
 998                 /*
 999                  * If we've dropped below the threshold after having
1000                  * rate-limited traffic print the message.  This preserves
1001                  * the previous behaviour at the expense of added complexity.
1002                  */
1003                 if (V_icmplim_output && opps > V_icmplim)
1004                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
1005                                 r->type, opps, V_icmplim);
1006         }
1007         return 0;                       /* okay to send packet */
1008 #undef N
1009 }