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 static VNET_DEFINE(int, icmptstamprepl) = 1;
 152 #define V_icmptstamprepl                VNET(icmptstamprepl)
 153 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
 154         &VNET_NAME(icmptstamprepl), 0, "Respond to ICMP Timestamp packets");
 155
 156 #ifdef ICMPPRINTFS
 157 int     icmpprintfs = 0;
 158 #endif
 159
 160 static void     icmp_reflect(struct mbuf *);
 161 static void     icmp_send(struct mbuf *, struct mbuf *);
 162
 163 extern  struct protosw inetsw[];
 164
 165 static int
 166 sysctl_net_icmp_drop_redir(SYSCTL_HANDLER_ARGS)
 167 {
 168         int error, new;
 169         int i;
 170         struct radix_node_head *rnh;
 171
 172         new = V_drop_redirect;
 173         error = sysctl_handle_int(oidp, &new, 0, req);
 174         if (error == 0 && req->newptr) {
 175                 new = (new != 0) ? 1 : 0;
 176
 177                 if (new == V_drop_redirect)
 178                         return (0);
 179
 180                 for (i = 0; i < rt_numfibs; i++) {
 181                         if ((rnh = rt_tables_get_rnh(i, AF_INET)) == NULL)
 182                                 continue;
 183                         RADIX_NODE_HEAD_LOCK(rnh);
 184                         in_setmatchfunc(rnh, new);
 185                         RADIX_NODE_HEAD_UNLOCK(rnh);
 186                 }
 187
 188                 V_drop_redirect = new;
 189         }
 190
 191         return (error);
 192 }
 193
 194 SYSCTL_VNET_PROC(_net_inet_icmp, OID_AUTO, drop_redirect,
 195     CTLTYPE_INT|CTLFLAG_RW, 0, 0,
 196     sysctl_net_icmp_drop_redir, "I", "Ignore ICMP redirects");
 197
 198 /*
 199  * Kernel module interface for updating icmpstat.  The argument is an index
 200  * into icmpstat treated as an array of u_long.  While this encodes the
 201  * general layout of icmpstat into the caller, it doesn't encode its
 202  * location, so that future changes to add, for example, per-CPU stats
 203  * support won't cause binary compatibility problems for kernel modules.
 204  */
 205 void
 206 kmod_icmpstat_inc(int statnum)
 207 {
 208
 209         counter_u64_add(VNET(icmpstat)[statnum], 1);
 210 }
 211
 212 /*
 213  * Generate an error packet of type error
 214  * in response to bad packet ip.
 215  */
 216 void
 217 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 218 {
 219         register struct ip *oip = mtod(n, struct ip *), *nip;
 220         register unsigned oiphlen = oip->ip_hl << 2;
 221         register struct icmp *icp;
 222         register struct mbuf *m;
 223         unsigned icmplen, icmpelen, nlen;
 224
 225         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
 226 #ifdef ICMPPRINTFS
 227         if (icmpprintfs)
 228                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 229 #endif
 230         if (type != ICMP_REDIRECT)
 231                 ICMPSTAT_INC(icps_error);
 232         /*
 233          * Don't send error:
 234          *  if the original packet was encrypted.
 235          *  if not the first fragment of message.
 236          *  in response to a multicast or broadcast packet.
 237          *  if the old packet protocol was an ICMP error message.
 238          */
 239         if (n->m_flags & M_DECRYPTED)
 240                 goto freeit;
 241         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 242                 goto freeit;
 243         if (n->m_flags & (M_BCAST|M_MCAST))
 244                 goto freeit;
 245         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 246           n->m_len >= oiphlen + ICMP_MINLEN &&
 247           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
 248                 ICMPSTAT_INC(icps_oldicmp);
 249                 goto freeit;
 250         }
 251         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
 252         if (oiphlen + 8 > n->m_len)
 253                 goto freeit;
 254         /*
 255          * Calculate length to quote from original packet and
 256          * prevent the ICMP mbuf from overflowing.
 257          * Unfortunatly this is non-trivial since ip_forward()
 258          * sends us truncated packets.
 259          */
 260         nlen = m_length(n, NULL);
 261         if (oip->ip_p == IPPROTO_TCP) {
 262                 struct tcphdr *th;
 263                 int tcphlen;
 264
 265                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 266                     n->m_next == NULL)
 267                         goto stdreply;
 268                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 269                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
 270                         goto freeit;
 271                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
 272                 tcphlen = th->th_off << 2;
 273                 if (tcphlen < sizeof(struct tcphdr))
 274                         goto freeit;
 275                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 276                         goto freeit;
 277                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 278                         goto stdreply;
 279                 if (n->m_len < oiphlen + tcphlen &&
 280                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
 281                         goto freeit;
 282                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 283                     ntohs(oip->ip_len) - oiphlen));
 284         } else
 285 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 286
 287         icmplen = min(oiphlen + icmpelen, nlen);
 288         if (icmplen < sizeof(struct ip))
 289                 goto freeit;
 290
 291         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 292                 m = m_gethdr(M_NOWAIT, MT_DATA);
 293         else
 294                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 295         if (m == NULL)
 296                 goto freeit;
 297 #ifdef MAC
 298         mac_netinet_icmp_reply(n, m);
 299 #endif
 300         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
 301         m_align(m, ICMP_MINLEN + icmplen);
 302         m->m_len = ICMP_MINLEN + icmplen;
 303
 304         /* XXX MRT  make the outgoing packet use the same FIB
 305          * that was associated with the incoming packet
 306          */
 307         M_SETFIB(m, M_GETFIB(n));
 308         icp = mtod(m, struct icmp *);
 309         ICMPSTAT_INC(icps_outhist[type]);
 310         icp->icmp_type = type;
 311         if (type == ICMP_REDIRECT)
 312                 icp->icmp_gwaddr.s_addr = dest;
 313         else {
 314                 icp->icmp_void = 0;
 315                 /*
 316                  * The following assignments assume an overlay with the
 317                  * just zeroed icmp_void field.
 318                  */
 319                 if (type == ICMP_PARAMPROB) {
 320                         icp->icmp_pptr = code;
 321                         code = 0;
 322                 } else if (type == ICMP_UNREACH &&
 323                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 324                         icp->icmp_nextmtu = htons(mtu);
 325                 }
 326         }
 327         icp->icmp_code = code;
 328
 329         /*
 330          * Copy the quotation into ICMP message and
 331          * convert quoted IP header back to network representation.
 332          */
 333         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 334         nip = &icp->icmp_ip;
 335
 336         /*
 337          * Set up ICMP message mbuf and copy old IP header (without options
 338          * in front of ICMP message.
 339          * If the original mbuf was meant to bypass the firewall, the error
 340          * reply should bypass as well.
 341          */
 342         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 343         m->m_data -= sizeof(struct ip);
 344         m->m_len += sizeof(struct ip);
 345         m->m_pkthdr.len = m->m_len;
 346         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 347         nip = mtod(m, struct ip *);
 348         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 349         nip->ip_len = htons(m->m_len);
 350         nip->ip_v = IPVERSION;
 351         nip->ip_hl = 5;
 352         nip->ip_p = IPPROTO_ICMP;
 353         nip->ip_tos = 0;
 354         nip->ip_off = 0;
 355         icmp_reflect(m);
 356
 357 freeit:
 358         m_freem(n);
 359 }
 360
 361 /*
 362  * Process a received ICMP message.
 363  */
 364 void
 365 icmp_input(struct mbuf *m, int off)
 366 {
 367         struct icmp *icp;
 368         struct in_ifaddr *ia;
 369         struct ip *ip = mtod(m, struct ip *);
 370         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 371         int hlen = off;
 372         int icmplen = ntohs(ip->ip_len) - off;
 373         int i, code;
 374         void (*ctlfunc)(int, struct sockaddr *, void *);
 375         int fibnum;
 376
 377         /*
 378          * Locate icmp structure in mbuf, and check
 379          * that not corrupted and of at least minimum length.
 380          */
 381 #ifdef ICMPPRINTFS
 382         if (icmpprintfs) {
 383                 char buf[4 * sizeof "123"];
 384                 strcpy(buf, inet_ntoa(ip->ip_src));
 385                 printf("icmp_input from %s to %s, len %d\n",
 386                        buf, inet_ntoa(ip->ip_dst), icmplen);
 387         }
 388 #endif
 389         if (icmplen < ICMP_MINLEN) {
 390                 ICMPSTAT_INC(icps_tooshort);
 391                 goto freeit;
 392         }
 393         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 394         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 395                 ICMPSTAT_INC(icps_tooshort);
 396                 return;
 397         }
 398         ip = mtod(m, struct ip *);
 399         m->m_len -= hlen;
 400         m->m_data += hlen;
 401         icp = mtod(m, struct icmp *);
 402         if (in_cksum(m, icmplen)) {
 403                 ICMPSTAT_INC(icps_checksum);
 404                 goto freeit;
 405         }
 406         m->m_len += hlen;
 407         m->m_data -= hlen;
 408
 409         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 410                 /*
 411                  * Deliver very specific ICMP type only.
 412                  */
 413                 switch (icp->icmp_type) {
 414                 case ICMP_UNREACH:
 415                 case ICMP_TIMXCEED:
 416                         break;
 417                 default:
 418                         goto freeit;
 419                 }
 420         }
 421
 422 #ifdef ICMPPRINTFS
 423         if (icmpprintfs)
 424                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 425                     icp->icmp_code);
 426 #endif
 427
 428         /*
 429          * Message type specific processing.
 430          */
 431         if (icp->icmp_type > ICMP_MAXTYPE)
 432                 goto raw;
 433
 434         /* Initialize */
 435         bzero(&icmpsrc, sizeof(icmpsrc));
 436         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 437         icmpsrc.sin_family = AF_INET;
 438         bzero(&icmpdst, sizeof(icmpdst));
 439         icmpdst.sin_len = sizeof(struct sockaddr_in);
 440         icmpdst.sin_family = AF_INET;
 441         bzero(&icmpgw, sizeof(icmpgw));
 442         icmpgw.sin_len = sizeof(struct sockaddr_in);
 443         icmpgw.sin_family = AF_INET;
 444
 445         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 446         code = icp->icmp_code;
 447         switch (icp->icmp_type) {
 448
 449         case ICMP_UNREACH:
 450                 switch (code) {
 451                         case ICMP_UNREACH_NET:
 452                         case ICMP_UNREACH_HOST:
 453                         case ICMP_UNREACH_SRCFAIL:
 454                         case ICMP_UNREACH_NET_UNKNOWN:
 455                         case ICMP_UNREACH_HOST_UNKNOWN:
 456                         case ICMP_UNREACH_ISOLATED:
 457                         case ICMP_UNREACH_TOSNET:
 458                         case ICMP_UNREACH_TOSHOST:
 459                         case ICMP_UNREACH_HOST_PRECEDENCE:
 460                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 461                                 code = PRC_UNREACH_NET;
 462                                 break;
 463
 464                         case ICMP_UNREACH_NEEDFRAG:
 465                                 code = PRC_MSGSIZE;
 466                                 break;
 467
 468                         /*
 469                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 470                          * Treat subcodes 2,3 as immediate RST
 471                          */
 472                         case ICMP_UNREACH_PROTOCOL:
 473                         case ICMP_UNREACH_PORT:
 474                                 code = PRC_UNREACH_PORT;
 475                                 break;
 476
 477                         case ICMP_UNREACH_NET_PROHIB:
 478                         case ICMP_UNREACH_HOST_PROHIB:
 479                         case ICMP_UNREACH_FILTER_PROHIB:
 480                                 code = PRC_UNREACH_ADMIN_PROHIB;
 481                                 break;
 482
 483                         default:
 484                                 goto badcode;
 485                 }
 486                 goto deliver;
 487
 488         case ICMP_TIMXCEED:
 489                 if (code > 1)
 490                         goto badcode;
 491                 code += PRC_TIMXCEED_INTRANS;
 492                 goto deliver;
 493
 494         case ICMP_PARAMPROB:
 495                 if (code > 1)
 496                         goto badcode;
 497                 code = PRC_PARAMPROB;
 498                 goto deliver;
 499
 500         case ICMP_SOURCEQUENCH:
 501                 if (code)
 502                         goto badcode;
 503                 code = PRC_QUENCH;
 504         deliver:
 505                 /*
 506                  * Problem with datagram; advise higher level routines.
 507                  */
 508                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 509                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 510                         ICMPSTAT_INC(icps_badlen);
 511                         goto freeit;
 512                 }
 513                 /* Discard ICMP's in response to multicast packets */
 514                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 515                         goto badcode;
 516 #ifdef ICMPPRINTFS
 517                 if (icmpprintfs)
 518                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 519 #endif
 520                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 521                 /*
 522                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 523                  * notification to TCP layer.
 524                  */
 525                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 526                 if (ctlfunc)
 527                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 528                                    (void *)&icp->icmp_ip);
 529                 break;
 530
 531         badcode:
 532                 ICMPSTAT_INC(icps_badcode);
 533                 break;
 534
 535         case ICMP_ECHO:
 536                 if (!V_icmpbmcastecho
 537                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 538                         ICMPSTAT_INC(icps_bmcastecho);
 539                         break;
 540                 }
 541                 icp->icmp_type = ICMP_ECHOREPLY;
 542                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 543                         goto freeit;
 544                 else
 545                         goto reflect;
 546
 547         case ICMP_TSTAMP:
 548                 if (V_icmptstamprepl == 0)
 549                         break;
 550                 if (!V_icmpbmcastecho
 551                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 552                         ICMPSTAT_INC(icps_bmcasttstamp);
 553                         break;
 554                 }
 555                 if (icmplen < ICMP_TSLEN) {
 556                         ICMPSTAT_INC(icps_badlen);
 557                         break;
 558                 }
 559                 icp->icmp_type = ICMP_TSTAMPREPLY;
 560                 icp->icmp_rtime = iptime();
 561                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 562                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 563                         goto freeit;
 564                 else
 565                         goto reflect;
 566
 567         case ICMP_MASKREQ:
 568                 if (V_icmpmaskrepl == 0)
 569                         break;
 570                 /*
 571                  * We are not able to respond with all ones broadcast
 572                  * unless we receive it over a point-to-point interface.
 573                  */
 574                 if (icmplen < ICMP_MASKLEN)
 575                         break;
 576                 switch (ip->ip_dst.s_addr) {
 577
 578                 case INADDR_BROADCAST:
 579                 case INADDR_ANY:
 580                         icmpdst.sin_addr = ip->ip_src;
 581                         break;
 582
 583                 default:
 584                         icmpdst.sin_addr = ip->ip_dst;
 585                 }
 586                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 587                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 588                 if (ia == NULL)
 589                         break;
 590                 if (ia->ia_ifp == NULL) {
 591                         ifa_free(&ia->ia_ifa);
 592                         break;
 593                 }
 594                 icp->icmp_type = ICMP_MASKREPLY;
 595                 if (V_icmpmaskfake == 0)
 596                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 597                 else
 598                         icp->icmp_mask = V_icmpmaskfake;
 599                 if (ip->ip_src.s_addr == 0) {
 600                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 601                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 602                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 603                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 604                 }
 605                 ifa_free(&ia->ia_ifa);
 606 reflect:
 607                 ICMPSTAT_INC(icps_reflect);
 608                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 609                 icmp_reflect(m);
 610                 return;
 611
 612         case ICMP_REDIRECT:
 613                 if (V_log_redirect) {
 614                         u_long src, dst, gw;
 615
 616                         src = ntohl(ip->ip_src.s_addr);
 617                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 618                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 619                         printf("icmp redirect from %d.%d.%d.%d: "
 620                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 621                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 622                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 623                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 624                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 625                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 626                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 627                 }
 628                 /*
 629                  * RFC1812 says we must ignore ICMP redirects if we
 630                  * are acting as router.
 631                  */
 632                 if (V_drop_redirect || V_ipforwarding)
 633                         break;
 634                 if (code > 3)
 635                         goto badcode;
 636                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 637                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 638                         ICMPSTAT_INC(icps_badlen);
 639                         break;
 640                 }
 641                 /*
 642                  * Short circuit routing redirects to force
 643                  * immediate change in the kernel's routing
 644                  * tables.  The message is also handed to anyone
 645                  * listening on a raw socket (e.g. the routing
 646                  * daemon for use in updating its tables).
 647                  */
 648                 icmpgw.sin_addr = ip->ip_src;
 649                 icmpdst.sin_addr = icp->icmp_gwaddr;
 650 #ifdef  ICMPPRINTFS
 651                 if (icmpprintfs) {
 652                         char buf[4 * sizeof "123"];
 653                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
 654
 655                         printf("redirect dst %s to %s\n",
 656                                buf, inet_ntoa(icp->icmp_gwaddr));
 657                 }
 658 #endif
 659                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 660                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 661                         in_rtredirect((struct sockaddr *)&icmpsrc,
 662                           (struct sockaddr *)&icmpdst,
 663                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 664                           (struct sockaddr *)&icmpgw, fibnum);
 665                 }
 666                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 667 #ifdef IPSEC
 668                 key_sa_routechange((struct sockaddr *)&icmpsrc);
 669 #endif
 670                 break;
 671
 672         /*
 673          * No kernel processing for the following;
 674          * just fall through to send to raw listener.
 675          */
 676         case ICMP_ECHOREPLY:
 677         case ICMP_ROUTERADVERT:
 678         case ICMP_ROUTERSOLICIT:
 679         case ICMP_TSTAMPREPLY:
 680         case ICMP_IREQREPLY:
 681         case ICMP_MASKREPLY:
 682         default:
 683                 break;
 684         }
 685
 686 raw:
 687         rip_input(m, off);
 688         return;
 689
 690 freeit:
 691         m_freem(m);
 692 }
 693
 694 /*
 695  * Reflect the ip packet back to the source
 696  */
 697 static void
 698 icmp_reflect(struct mbuf *m)
 699 {
 700         struct ip *ip = mtod(m, struct ip *);
 701         struct ifaddr *ifa;
 702         struct ifnet *ifp;
 703         struct in_ifaddr *ia;
 704         struct in_addr t;
 705         struct mbuf *opts = 0;
 706         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 707
 708         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 709             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 710             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 711                 m_freem(m);     /* Bad return address */
 712                 ICMPSTAT_INC(icps_badaddr);
 713                 goto done;      /* Ip_output() will check for broadcast */
 714         }
 715
 716         t = ip->ip_dst;
 717         ip->ip_dst = ip->ip_src;
 718
 719         /*
 720          * Source selection for ICMP replies:
 721          *
 722          * If the incoming packet was addressed directly to one of our
 723          * own addresses, use dst as the src for the reply.
 724          */
 725         IN_IFADDR_RLOCK();
 726         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 727                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 728                         t = IA_SIN(ia)->sin_addr;
 729                         IN_IFADDR_RUNLOCK();
 730                         goto match;
 731                 }
 732         }
 733         IN_IFADDR_RUNLOCK();
 734
 735         /*
 736          * If the incoming packet was addressed to one of our broadcast
 737          * addresses, use the first non-broadcast address which corresponds
 738          * to the incoming interface.
 739          */
 740         ifp = m->m_pkthdr.rcvif;
 741         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 742                 IF_ADDR_RLOCK(ifp);
 743                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 744                         if (ifa->ifa_addr->sa_family != AF_INET)
 745                                 continue;
 746                         ia = ifatoia(ifa);
 747                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 748                             t.s_addr) {
 749                                 t = IA_SIN(ia)->sin_addr;
 750                                 IF_ADDR_RUNLOCK(ifp);
 751                                 goto match;
 752                         }
 753                 }
 754                 IF_ADDR_RUNLOCK(ifp);
 755         }
 756         /*
 757          * If the packet was transiting through us, use the address of
 758          * the interface the packet came through in.  If that interface
 759          * doesn't have a suitable IP address, the normal selection
 760          * criteria apply.
 761          */
 762         if (V_icmp_rfi && ifp != NULL) {
 763                 IF_ADDR_RLOCK(ifp);
 764                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 765                         if (ifa->ifa_addr->sa_family != AF_INET)
 766                                 continue;
 767                         ia = ifatoia(ifa);
 768                         t = IA_SIN(ia)->sin_addr;
 769                         IF_ADDR_RUNLOCK(ifp);
 770                         goto match;
 771                 }
 772                 IF_ADDR_RUNLOCK(ifp);
 773         }
 774         /*
 775          * If the incoming packet was not addressed directly to us, use
 776          * designated interface for icmp replies specified by sysctl
 777          * net.inet.icmp.reply_src (default not set). Otherwise continue
 778          * with normal source selection.
 779          */
 780         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 781                 IF_ADDR_RLOCK(ifp);
 782                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 783                         if (ifa->ifa_addr->sa_family != AF_INET)
 784                                 continue;
 785                         ia = ifatoia(ifa);
 786                         t = IA_SIN(ia)->sin_addr;
 787                         IF_ADDR_RUNLOCK(ifp);
 788                         goto match;
 789                 }
 790                 IF_ADDR_RUNLOCK(ifp);
 791         }
 792         /*
 793          * If the packet was transiting through us, use the address of
 794          * the interface that is the closest to the packet source.
 795          * When we don't have a route back to the packet source, stop here
 796          * and drop the packet.
 797          */
 798         ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
 799         if (ia == NULL) {
 800                 m_freem(m);
 801                 ICMPSTAT_INC(icps_noroute);
 802                 goto done;
 803         }
 804         t = IA_SIN(ia)->sin_addr;
 805         ifa_free(&ia->ia_ifa);
 806 match:
 807 #ifdef MAC
 808         mac_netinet_icmp_replyinplace(m);
 809 #endif
 810         ip->ip_src = t;
 811         ip->ip_ttl = V_ip_defttl;
 812
 813         if (optlen > 0) {
 814                 register u_char *cp;
 815                 int opt, cnt;
 816                 u_int len;
 817
 818                 /*
 819                  * Retrieve any source routing from the incoming packet;
 820                  * add on any record-route or timestamp options.
 821                  */
 822                 cp = (u_char *) (ip + 1);
 823                 if ((opts = ip_srcroute(m)) == 0 &&
 824                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 825                         opts->m_len = sizeof(struct in_addr);
 826                         mtod(opts, struct in_addr *)->s_addr = 0;
 827                 }
 828                 if (opts) {
 829 #ifdef ICMPPRINTFS
 830                     if (icmpprintfs)
 831                             printf("icmp_reflect optlen %d rt %d => ",
 832                                 optlen, opts->m_len);
 833 #endif
 834                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 835                             opt = cp[IPOPT_OPTVAL];
 836                             if (opt == IPOPT_EOL)
 837                                     break;
 838                             if (opt == IPOPT_NOP)
 839                                     len = 1;
 840                             else {
 841                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 842                                             break;
 843                                     len = cp[IPOPT_OLEN];
 844                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 845                                         len > cnt)
 846                                             break;
 847                             }
 848                             /*
 849                              * Should check for overflow, but it "can't happen"
 850                              */
 851                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 852                                 opt == IPOPT_SECURITY) {
 853                                     bcopy((caddr_t)cp,
 854                                         mtod(opts, caddr_t) + opts->m_len, len);
 855                                     opts->m_len += len;
 856                             }
 857                     }
 858                     /* Terminate & pad, if necessary */
 859                     cnt = opts->m_len % 4;
 860                     if (cnt) {
 861                             for (; cnt < 4; cnt++) {
 862                                     *(mtod(opts, caddr_t) + opts->m_len) =
 863                                         IPOPT_EOL;
 864                                     opts->m_len++;
 865                             }
 866                     }
 867 #ifdef ICMPPRINTFS
 868                     if (icmpprintfs)
 869                             printf("%d\n", opts->m_len);
 870 #endif
 871                 }
 872                 ip_stripoptions(m);
 873         }
 874         m_tag_delete_nonpersistent(m);
 875         m->m_flags &= ~(M_BCAST|M_MCAST);
 876         icmp_send(m, opts);
 877 done:
 878         if (opts)
 879                 (void)m_free(opts);
 880 }
 881
 882 /*
 883  * Send an icmp packet back to the ip level,
 884  * after supplying a checksum.
 885  */
 886 static void
 887 icmp_send(struct mbuf *m, struct mbuf *opts)
 888 {
 889         register struct ip *ip = mtod(m, struct ip *);
 890         register int hlen;
 891         register struct icmp *icp;
 892
 893         hlen = ip->ip_hl << 2;
 894         m->m_data += hlen;
 895         m->m_len -= hlen;
 896         icp = mtod(m, struct icmp *);
 897         icp->icmp_cksum = 0;
 898         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 899         m->m_data -= hlen;
 900         m->m_len += hlen;
 901         m->m_pkthdr.rcvif = (struct ifnet *)0;
 902 #ifdef ICMPPRINTFS
 903         if (icmpprintfs) {
 904                 char buf[4 * sizeof "123"];
 905                 strcpy(buf, inet_ntoa(ip->ip_dst));
 906                 printf("icmp_send dst %s src %s\n",
 907                        buf, inet_ntoa(ip->ip_src));
 908         }
 909 #endif
 910         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 911 }
 912
 913 /*
 914  * Return milliseconds since 00:00 GMT in network format.
 915  */
 916 uint32_t
 917 iptime(void)
 918 {
 919         struct timeval atv;
 920         u_long t;
 921
 922         getmicrotime(&atv);
 923         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 924         return (htonl(t));
 925 }
 926
 927 /*
 928  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 929  * given current value MTU.  If DIR is less than zero, a larger plateau
 930  * is returned; otherwise, a smaller value is returned.
 931  */
 932 int
 933 ip_next_mtu(int mtu, int dir)
 934 {
 935         static int mtutab[] = {
 936                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 937                 296, 68, 0
 938         };
 939         int i, size;
 940
 941         size = (sizeof mtutab) / (sizeof mtutab[0]);
 942         if (dir >= 0) {
 943                 for (i = 0; i < size; i++)
 944                         if (mtu > mtutab[i])
 945                                 return mtutab[i];
 946         } else {
 947                 for (i = size - 1; i >= 0; i--)
 948                         if (mtu < mtutab[i])
 949                                 return mtutab[i];
 950                 if (mtu == mtutab[0])
 951                         return mtutab[0];
 952         }
 953         return 0;
 954 }
 955 #endif /* INET */
 956
 957
 958 /*
 959  * badport_bandlim() - check for ICMP bandwidth limit
 960  *
 961  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 962  *      hit our bandwidth limit and it is not ok.
 963  *
 964  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 965  *
 966  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 967  *      values.  We may eventually remove this separation (and simplify the
 968  *      code further).
 969  *
 970  *      Note that the printing of the error message is delayed so we can
 971  *      properly print the icmp error rate that the system was trying to do
 972  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 973  *      the 'final' error, but it doesn't make sense to solve the printing
 974  *      delay with more complex code.
 975  */
 976
 977 int
 978 badport_bandlim(int which)
 979 {
 980
 981 #define N(a)    (sizeof (a) / sizeof (a[0]))
 982         static struct rate {
 983                 const char      *type;
 984                 struct timeval  lasttime;
 985                 int             curpps;
 986         } rates[BANDLIM_MAX+1] = {
 987                 { "icmp unreach response" },
 988                 { "icmp ping response" },
 989                 { "icmp tstamp response" },
 990                 { "closed port RST response" },
 991                 { "open port RST response" },
 992                 { "icmp6 unreach response" },
 993                 { "sctp ootb response" }
 994         };
 995
 996         /*
 997          * Return ok status if feature disabled or argument out of range.
 998          */
 999         if (V_icmplim > 0 && (u_int) which < N(rates)) {
1000                 struct rate *r = &rates[which];
1001                 int opps = r->curpps;
1002
1003                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
1004                         return -1;      /* discard packet */
1005                 /*
1006                  * If we've dropped below the threshold after having
1007                  * rate-limited traffic print the message.  This preserves
1008                  * the previous behaviour at the expense of added complexity.
1009                  */
1010                 if (V_icmplim_output && opps > V_icmplim)
1011                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
1012                                 r->type, opps, V_icmplim);
1013         }
1014         return 0;                       /* okay to send packet */
1015 #undef N
1016 }