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 rate limiting of ICMP responses");
  93
  94 #ifdef INET
  95 VNET_DEFINE(struct icmpstat, icmpstat);
  96 SYSCTL_VNET_STRUCT(_net_inet_icmp, ICMPCTL_STATS, stats, CTLFLAG_RW,
  97         &VNET_NAME(icmpstat), icmpstat, "");
  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
 147 #ifdef ICMPPRINTFS
 148 int     icmpprintfs = 0;
 149 #endif
 150
 151 static void     icmp_reflect(struct mbuf *);
 152 static void     icmp_send(struct mbuf *, struct mbuf *);
 153
 154 extern  struct protosw inetsw[];
 155
 156 static int
 157 sysctl_net_icmp_drop_redir(SYSCTL_HANDLER_ARGS)
 158 {
 159         int error, new;
 160         int i;
 161         struct radix_node_head *rnh;
 162
 163         new = V_drop_redirect;
 164         error = sysctl_handle_int(oidp, &new, 0, req);
 165         if (error == 0 && req->newptr) {
 166                 new = (new != 0) ? 1 : 0;
 167
 168                 if (new == V_drop_redirect)
 169                         return (0);
 170
 171                 for (i = 0; i < rt_numfibs; i++) {
 172                         if ((rnh = rt_tables_get_rnh(i, AF_INET)) == NULL)
 173                                 continue;
 174                         RADIX_NODE_HEAD_LOCK(rnh);
 175                         in_setmatchfunc(rnh, new);
 176                         RADIX_NODE_HEAD_UNLOCK(rnh);
 177                 }
 178
 179                 V_drop_redirect = new;
 180         }
 181
 182         return (error);
 183 }
 184
 185 SYSCTL_VNET_PROC(_net_inet_icmp, OID_AUTO, drop_redirect,
 186     CTLTYPE_INT|CTLFLAG_RW, 0, 0,
 187     sysctl_net_icmp_drop_redir, "I", "Ignore ICMP redirects");
 188
 189 /*
 190  * Kernel module interface for updating icmpstat.  The argument is an index
 191  * into icmpstat treated as an array of u_long.  While this encodes the
 192  * general layout of icmpstat into the caller, it doesn't encode its
 193  * location, so that future changes to add, for example, per-CPU stats
 194  * support won't cause binary compatibility problems for kernel modules.
 195  */
 196 void
 197 kmod_icmpstat_inc(int statnum)
 198 {
 199
 200         (*((u_long *)&V_icmpstat + statnum))++;
 201 }
 202
 203 /*
 204  * Generate an error packet of type error
 205  * in response to bad packet ip.
 206  */
 207 void
 208 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 209 {
 210         register struct ip *oip = mtod(n, struct ip *), *nip;
 211         register unsigned oiphlen = oip->ip_hl << 2;
 212         register struct icmp *icp;
 213         register struct mbuf *m;
 214         unsigned icmplen, icmpelen, nlen;
 215
 216         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
 217 #ifdef ICMPPRINTFS
 218         if (icmpprintfs)
 219                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 220 #endif
 221         if (type != ICMP_REDIRECT)
 222                 ICMPSTAT_INC(icps_error);
 223         /*
 224          * Don't send error:
 225          *  if the original packet was encrypted.
 226          *  if not the first fragment of message.
 227          *  in response to a multicast or broadcast packet.
 228          *  if the old packet protocol was an ICMP error message.
 229          */
 230         if (n->m_flags & M_DECRYPTED)
 231                 goto freeit;
 232         if (oip->ip_off & ~(IP_MF|IP_DF))
 233                 goto freeit;
 234         if (n->m_flags & (M_BCAST|M_MCAST))
 235                 goto freeit;
 236         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 237           n->m_len >= oiphlen + ICMP_MINLEN &&
 238           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
 239                 ICMPSTAT_INC(icps_oldicmp);
 240                 goto freeit;
 241         }
 242         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
 243         if (oiphlen + 8 > n->m_len)
 244                 goto freeit;
 245         /*
 246          * Calculate length to quote from original packet and
 247          * prevent the ICMP mbuf from overflowing.
 248          * Unfortunatly this is non-trivial since ip_forward()
 249          * sends us truncated packets.
 250          */
 251         nlen = m_length(n, NULL);
 252         if (oip->ip_p == IPPROTO_TCP) {
 253                 struct tcphdr *th;
 254                 int tcphlen;
 255
 256                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 257                     n->m_next == NULL)
 258                         goto stdreply;
 259                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 260                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
 261                         goto freeit;
 262                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
 263                 tcphlen = th->th_off << 2;
 264                 if (tcphlen < sizeof(struct tcphdr))
 265                         goto freeit;
 266                 if (oip->ip_len < oiphlen + tcphlen)
 267                         goto freeit;
 268                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 269                         goto stdreply;
 270                 if (n->m_len < oiphlen + tcphlen &&
 271                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
 272                         goto freeit;
 273                 icmpelen = max(tcphlen, min(V_icmp_quotelen, oip->ip_len - oiphlen));
 274         } else
 275 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, oip->ip_len - oiphlen));
 276
 277         icmplen = min(oiphlen + icmpelen, nlen);
 278         if (icmplen < sizeof(struct ip))
 279                 goto freeit;
 280
 281         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 282                 m = m_gethdr(M_DONTWAIT, MT_DATA);
 283         else
 284                 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
 285         if (m == NULL)
 286                 goto freeit;
 287 #ifdef MAC
 288         mac_netinet_icmp_reply(n, m);
 289 #endif
 290         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
 291         m_align(m, ICMP_MINLEN + icmplen);
 292         m->m_len = ICMP_MINLEN + icmplen;
 293
 294         /* XXX MRT  make the outgoing packet use the same FIB
 295          * that was associated with the incoming packet
 296          */
 297         M_SETFIB(m, M_GETFIB(n));
 298         icp = mtod(m, struct icmp *);
 299         ICMPSTAT_INC(icps_outhist[type]);
 300         icp->icmp_type = type;
 301         if (type == ICMP_REDIRECT)
 302                 icp->icmp_gwaddr.s_addr = dest;
 303         else {
 304                 icp->icmp_void = 0;
 305                 /*
 306                  * The following assignments assume an overlay with the
 307                  * just zeroed icmp_void field.
 308                  */
 309                 if (type == ICMP_PARAMPROB) {
 310                         icp->icmp_pptr = code;
 311                         code = 0;
 312                 } else if (type == ICMP_UNREACH &&
 313                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 314                         icp->icmp_nextmtu = htons(mtu);
 315                 }
 316         }
 317         icp->icmp_code = code;
 318
 319         /*
 320          * Copy the quotation into ICMP message and
 321          * convert quoted IP header back to network representation.
 322          */
 323         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 324         nip = &icp->icmp_ip;
 325         nip->ip_len = htons(nip->ip_len);
 326         nip->ip_off = htons(nip->ip_off);
 327
 328         /*
 329          * Set up ICMP message mbuf and copy old IP header (without options
 330          * in front of ICMP message.
 331          * If the original mbuf was meant to bypass the firewall, the error
 332          * reply should bypass as well.
 333          */
 334         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 335         m->m_data -= sizeof(struct ip);
 336         m->m_len += sizeof(struct ip);
 337         m->m_pkthdr.len = m->m_len;
 338         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 339         nip = mtod(m, struct ip *);
 340         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 341         nip->ip_len = m->m_len;
 342         nip->ip_v = IPVERSION;
 343         nip->ip_hl = 5;
 344         nip->ip_p = IPPROTO_ICMP;
 345         nip->ip_tos = 0;
 346         nip->ip_off = 0;
 347         icmp_reflect(m);
 348
 349 freeit:
 350         m_freem(n);
 351 }
 352
 353 /*
 354  * Process a received ICMP message.
 355  */
 356 void
 357 icmp_input(struct mbuf *m, int off)
 358 {
 359         struct icmp *icp;
 360         struct in_ifaddr *ia;
 361         struct ip *ip = mtod(m, struct ip *);
 362         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 363         int hlen = off;
 364         int icmplen = ip->ip_len;
 365         int i, code;
 366         void (*ctlfunc)(int, struct sockaddr *, void *);
 367         int fibnum;
 368
 369         /*
 370          * Locate icmp structure in mbuf, and check
 371          * that not corrupted and of at least minimum length.
 372          */
 373 #ifdef ICMPPRINTFS
 374         if (icmpprintfs) {
 375                 char buf[4 * sizeof "123"];
 376                 strcpy(buf, inet_ntoa(ip->ip_src));
 377                 printf("icmp_input from %s to %s, len %d\n",
 378                        buf, inet_ntoa(ip->ip_dst), icmplen);
 379         }
 380 #endif
 381         if (icmplen < ICMP_MINLEN) {
 382                 ICMPSTAT_INC(icps_tooshort);
 383                 goto freeit;
 384         }
 385         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 386         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 387                 ICMPSTAT_INC(icps_tooshort);
 388                 return;
 389         }
 390         ip = mtod(m, struct ip *);
 391         m->m_len -= hlen;
 392         m->m_data += hlen;
 393         icp = mtod(m, struct icmp *);
 394         if (in_cksum(m, icmplen)) {
 395                 ICMPSTAT_INC(icps_checksum);
 396                 goto freeit;
 397         }
 398         m->m_len += hlen;
 399         m->m_data -= hlen;
 400
 401         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 402                 /*
 403                  * Deliver very specific ICMP type only.
 404                  */
 405                 switch (icp->icmp_type) {
 406                 case ICMP_UNREACH:
 407                 case ICMP_TIMXCEED:
 408                         break;
 409                 default:
 410                         goto freeit;
 411                 }
 412         }
 413
 414 #ifdef ICMPPRINTFS
 415         if (icmpprintfs)
 416                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 417                     icp->icmp_code);
 418 #endif
 419
 420         /*
 421          * Message type specific processing.
 422          */
 423         if (icp->icmp_type > ICMP_MAXTYPE)
 424                 goto raw;
 425
 426         /* Initialize */
 427         bzero(&icmpsrc, sizeof(icmpsrc));
 428         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 429         icmpsrc.sin_family = AF_INET;
 430         bzero(&icmpdst, sizeof(icmpdst));
 431         icmpdst.sin_len = sizeof(struct sockaddr_in);
 432         icmpdst.sin_family = AF_INET;
 433         bzero(&icmpgw, sizeof(icmpgw));
 434         icmpgw.sin_len = sizeof(struct sockaddr_in);
 435         icmpgw.sin_family = AF_INET;
 436
 437         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 438         code = icp->icmp_code;
 439         switch (icp->icmp_type) {
 440
 441         case ICMP_UNREACH:
 442                 switch (code) {
 443                         case ICMP_UNREACH_NET:
 444                         case ICMP_UNREACH_HOST:
 445                         case ICMP_UNREACH_SRCFAIL:
 446                         case ICMP_UNREACH_NET_UNKNOWN:
 447                         case ICMP_UNREACH_HOST_UNKNOWN:
 448                         case ICMP_UNREACH_ISOLATED:
 449                         case ICMP_UNREACH_TOSNET:
 450                         case ICMP_UNREACH_TOSHOST:
 451                         case ICMP_UNREACH_HOST_PRECEDENCE:
 452                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 453                                 code = PRC_UNREACH_NET;
 454                                 break;
 455
 456                         case ICMP_UNREACH_NEEDFRAG:
 457                                 code = PRC_MSGSIZE;
 458                                 break;
 459
 460                         /*
 461                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 462                          * Treat subcodes 2,3 as immediate RST
 463                          */
 464                         case ICMP_UNREACH_PROTOCOL:
 465                         case ICMP_UNREACH_PORT:
 466                                 code = PRC_UNREACH_PORT;
 467                                 break;
 468
 469                         case ICMP_UNREACH_NET_PROHIB:
 470                         case ICMP_UNREACH_HOST_PROHIB:
 471                         case ICMP_UNREACH_FILTER_PROHIB:
 472                                 code = PRC_UNREACH_ADMIN_PROHIB;
 473                                 break;
 474
 475                         default:
 476                                 goto badcode;
 477                 }
 478                 goto deliver;
 479
 480         case ICMP_TIMXCEED:
 481                 if (code > 1)
 482                         goto badcode;
 483                 code += PRC_TIMXCEED_INTRANS;
 484                 goto deliver;
 485
 486         case ICMP_PARAMPROB:
 487                 if (code > 1)
 488                         goto badcode;
 489                 code = PRC_PARAMPROB;
 490                 goto deliver;
 491
 492         case ICMP_SOURCEQUENCH:
 493                 if (code)
 494                         goto badcode;
 495                 code = PRC_QUENCH;
 496         deliver:
 497                 /*
 498                  * Problem with datagram; advise higher level routines.
 499                  */
 500                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 501                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 502                         ICMPSTAT_INC(icps_badlen);
 503                         goto freeit;
 504                 }
 505                 icp->icmp_ip.ip_len = ntohs(icp->icmp_ip.ip_len);
 506                 /* Discard ICMP's in response to multicast packets */
 507                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 508                         goto badcode;
 509 #ifdef ICMPPRINTFS
 510                 if (icmpprintfs)
 511                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 512 #endif
 513                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 514                 /*
 515                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 516                  * notification to TCP layer.
 517                  */
 518                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 519                 if (ctlfunc)
 520                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 521                                    (void *)&icp->icmp_ip);
 522                 break;
 523
 524         badcode:
 525                 ICMPSTAT_INC(icps_badcode);
 526                 break;
 527
 528         case ICMP_ECHO:
 529                 if (!V_icmpbmcastecho
 530                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 531                         ICMPSTAT_INC(icps_bmcastecho);
 532                         break;
 533                 }
 534                 icp->icmp_type = ICMP_ECHOREPLY;
 535                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 536                         goto freeit;
 537                 else
 538                         goto reflect;
 539
 540         case ICMP_TSTAMP:
 541                 if (!V_icmpbmcastecho
 542                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 543                         ICMPSTAT_INC(icps_bmcasttstamp);
 544                         break;
 545                 }
 546                 if (icmplen < ICMP_TSLEN) {
 547                         ICMPSTAT_INC(icps_badlen);
 548                         break;
 549                 }
 550                 icp->icmp_type = ICMP_TSTAMPREPLY;
 551                 icp->icmp_rtime = iptime();
 552                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 553                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 554                         goto freeit;
 555                 else
 556                         goto reflect;
 557
 558         case ICMP_MASKREQ:
 559                 if (V_icmpmaskrepl == 0)
 560                         break;
 561                 /*
 562                  * We are not able to respond with all ones broadcast
 563                  * unless we receive it over a point-to-point interface.
 564                  */
 565                 if (icmplen < ICMP_MASKLEN)
 566                         break;
 567                 switch (ip->ip_dst.s_addr) {
 568
 569                 case INADDR_BROADCAST:
 570                 case INADDR_ANY:
 571                         icmpdst.sin_addr = ip->ip_src;
 572                         break;
 573
 574                 default:
 575                         icmpdst.sin_addr = ip->ip_dst;
 576                 }
 577                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 578                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 579                 if (ia == NULL)
 580                         break;
 581                 if (ia->ia_ifp == NULL) {
 582                         ifa_free(&ia->ia_ifa);
 583                         break;
 584                 }
 585                 icp->icmp_type = ICMP_MASKREPLY;
 586                 if (V_icmpmaskfake == 0)
 587                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 588                 else
 589                         icp->icmp_mask = V_icmpmaskfake;
 590                 if (ip->ip_src.s_addr == 0) {
 591                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 592                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 593                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 594                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 595                 }
 596                 ifa_free(&ia->ia_ifa);
 597 reflect:
 598                 ip->ip_len += hlen;     /* since ip_input deducts this */
 599                 ICMPSTAT_INC(icps_reflect);
 600                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 601                 icmp_reflect(m);
 602                 return;
 603
 604         case ICMP_REDIRECT:
 605                 if (V_log_redirect) {
 606                         u_long src, dst, gw;
 607
 608                         src = ntohl(ip->ip_src.s_addr);
 609                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 610                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 611                         printf("icmp redirect from %d.%d.%d.%d: "
 612                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 613                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 614                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 615                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 616                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 617                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 618                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 619                 }
 620                 /*
 621                  * RFC1812 says we must ignore ICMP redirects if we
 622                  * are acting as router.
 623                  */
 624                 if (V_drop_redirect || V_ipforwarding)
 625                         break;
 626                 if (code > 3)
 627                         goto badcode;
 628                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 629                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 630                         ICMPSTAT_INC(icps_badlen);
 631                         break;
 632                 }
 633                 /*
 634                  * Short circuit routing redirects to force
 635                  * immediate change in the kernel's routing
 636                  * tables.  The message is also handed to anyone
 637                  * listening on a raw socket (e.g. the routing
 638                  * daemon for use in updating its tables).
 639                  */
 640                 icmpgw.sin_addr = ip->ip_src;
 641                 icmpdst.sin_addr = icp->icmp_gwaddr;
 642 #ifdef  ICMPPRINTFS
 643                 if (icmpprintfs) {
 644                         char buf[4 * sizeof "123"];
 645                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
 646
 647                         printf("redirect dst %s to %s\n",
 648                                buf, inet_ntoa(icp->icmp_gwaddr));
 649                 }
 650 #endif
 651                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 652                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 653                         in_rtredirect((struct sockaddr *)&icmpsrc,
 654                           (struct sockaddr *)&icmpdst,
 655                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 656                           (struct sockaddr *)&icmpgw, fibnum);
 657                 }
 658                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 659 #ifdef IPSEC
 660                 key_sa_routechange((struct sockaddr *)&icmpsrc);
 661 #endif
 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         m_addr_changed(m);
 709
 710         t = ip->ip_dst;
 711         ip->ip_dst = ip->ip_src;
 712
 713         /*
 714          * Source selection for ICMP replies:
 715          *
 716          * If the incoming packet was addressed directly to one of our
 717          * own addresses, use dst as the src for the reply.
 718          */
 719         IN_IFADDR_RLOCK();
 720         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 721                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 722                         t = IA_SIN(ia)->sin_addr;
 723                         IN_IFADDR_RUNLOCK();
 724                         goto match;
 725                 }
 726         }
 727         IN_IFADDR_RUNLOCK();
 728
 729         /*
 730          * If the incoming packet was addressed to one of our broadcast
 731          * addresses, use the first non-broadcast address which corresponds
 732          * to the incoming interface.
 733          */
 734         ifp = m->m_pkthdr.rcvif;
 735         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 736                 IF_ADDR_RLOCK(ifp);
 737                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 738                         if (ifa->ifa_addr->sa_family != AF_INET)
 739                                 continue;
 740                         ia = ifatoia(ifa);
 741                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 742                             t.s_addr) {
 743                                 t = IA_SIN(ia)->sin_addr;
 744                                 IF_ADDR_RUNLOCK(ifp);
 745                                 goto match;
 746                         }
 747                 }
 748                 IF_ADDR_RUNLOCK(ifp);
 749         }
 750         /*
 751          * If the packet was transiting through us, use the address of
 752          * the interface the packet came through in.  If that interface
 753          * doesn't have a suitable IP address, the normal selection
 754          * criteria apply.
 755          */
 756         if (V_icmp_rfi && ifp != NULL) {
 757                 IF_ADDR_RLOCK(ifp);
 758                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 759                         if (ifa->ifa_addr->sa_family != AF_INET)
 760                                 continue;
 761                         ia = ifatoia(ifa);
 762                         t = IA_SIN(ia)->sin_addr;
 763                         IF_ADDR_RUNLOCK(ifp);
 764                         goto match;
 765                 }
 766                 IF_ADDR_RUNLOCK(ifp);
 767         }
 768         /*
 769          * If the incoming packet was not addressed directly to us, use
 770          * designated interface for icmp replies specified by sysctl
 771          * net.inet.icmp.reply_src (default not set). Otherwise continue
 772          * with normal source selection.
 773          */
 774         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 775                 IF_ADDR_RLOCK(ifp);
 776                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 777                         if (ifa->ifa_addr->sa_family != AF_INET)
 778                                 continue;
 779                         ia = ifatoia(ifa);
 780                         t = IA_SIN(ia)->sin_addr;
 781                         IF_ADDR_RUNLOCK(ifp);
 782                         goto match;
 783                 }
 784                 IF_ADDR_RUNLOCK(ifp);
 785         }
 786         /*
 787          * If the packet was transiting through us, use the address of
 788          * the interface that is the closest to the packet source.
 789          * When we don't have a route back to the packet source, stop here
 790          * and drop the packet.
 791          */
 792         ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
 793         if (ia == NULL) {
 794                 m_freem(m);
 795                 ICMPSTAT_INC(icps_noroute);
 796                 goto done;
 797         }
 798         t = IA_SIN(ia)->sin_addr;
 799         ifa_free(&ia->ia_ifa);
 800 match:
 801 #ifdef MAC
 802         mac_netinet_icmp_replyinplace(m);
 803 #endif
 804         ip->ip_src = t;
 805         ip->ip_ttl = V_ip_defttl;
 806
 807         if (optlen > 0) {
 808                 register u_char *cp;
 809                 int opt, cnt;
 810                 u_int len;
 811
 812                 /*
 813                  * Retrieve any source routing from the incoming packet;
 814                  * add on any record-route or timestamp options.
 815                  */
 816                 cp = (u_char *) (ip + 1);
 817                 if ((opts = ip_srcroute(m)) == 0 &&
 818                     (opts = m_gethdr(M_DONTWAIT, MT_DATA))) {
 819                         opts->m_len = sizeof(struct in_addr);
 820                         mtod(opts, struct in_addr *)->s_addr = 0;
 821                 }
 822                 if (opts) {
 823 #ifdef ICMPPRINTFS
 824                     if (icmpprintfs)
 825                             printf("icmp_reflect optlen %d rt %d => ",
 826                                 optlen, opts->m_len);
 827 #endif
 828                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 829                             opt = cp[IPOPT_OPTVAL];
 830                             if (opt == IPOPT_EOL)
 831                                     break;
 832                             if (opt == IPOPT_NOP)
 833                                     len = 1;
 834                             else {
 835                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 836                                             break;
 837                                     len = cp[IPOPT_OLEN];
 838                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 839                                         len > cnt)
 840                                             break;
 841                             }
 842                             /*
 843                              * Should check for overflow, but it "can't happen"
 844                              */
 845                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 846                                 opt == IPOPT_SECURITY) {
 847                                     bcopy((caddr_t)cp,
 848                                         mtod(opts, caddr_t) + opts->m_len, len);
 849                                     opts->m_len += len;
 850                             }
 851                     }
 852                     /* Terminate & pad, if necessary */
 853                     cnt = opts->m_len % 4;
 854                     if (cnt) {
 855                             for (; cnt < 4; cnt++) {
 856                                     *(mtod(opts, caddr_t) + opts->m_len) =
 857                                         IPOPT_EOL;
 858                                     opts->m_len++;
 859                             }
 860                     }
 861 #ifdef ICMPPRINTFS
 862                     if (icmpprintfs)
 863                             printf("%d\n", opts->m_len);
 864 #endif
 865                 }
 866                 /*
 867                  * Now strip out original options by copying rest of first
 868                  * mbuf's data back, and adjust the IP length.
 869                  */
 870                 ip->ip_len -= optlen;
 871                 ip->ip_v = IPVERSION;
 872                 ip->ip_hl = 5;
 873                 m->m_len -= optlen;
 874                 if (m->m_flags & M_PKTHDR)
 875                         m->m_pkthdr.len -= optlen;
 876                 optlen += sizeof(struct ip);
 877                 bcopy((caddr_t)ip + optlen, (caddr_t)(ip + 1),
 878                          (unsigned)(m->m_len - sizeof(struct ip)));
 879         }
 880         m_tag_delete_nonpersistent(m);
 881         m->m_flags &= ~(M_BCAST|M_MCAST);
 882         icmp_send(m, opts);
 883 done:
 884         if (opts)
 885                 (void)m_free(opts);
 886 }
 887
 888 /*
 889  * Send an icmp packet back to the ip level,
 890  * after supplying a checksum.
 891  */
 892 static void
 893 icmp_send(struct mbuf *m, struct mbuf *opts)
 894 {
 895         register struct ip *ip = mtod(m, struct ip *);
 896         register int hlen;
 897         register struct icmp *icp;
 898
 899         hlen = ip->ip_hl << 2;
 900         m->m_data += hlen;
 901         m->m_len -= hlen;
 902         icp = mtod(m, struct icmp *);
 903         icp->icmp_cksum = 0;
 904         icp->icmp_cksum = in_cksum(m, ip->ip_len - hlen);
 905         m->m_data -= hlen;
 906         m->m_len += hlen;
 907         m->m_pkthdr.rcvif = (struct ifnet *)0;
 908 #ifdef ICMPPRINTFS
 909         if (icmpprintfs) {
 910                 char buf[4 * sizeof "123"];
 911                 strcpy(buf, inet_ntoa(ip->ip_dst));
 912                 printf("icmp_send dst %s src %s\n",
 913                        buf, inet_ntoa(ip->ip_src));
 914         }
 915 #endif
 916         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 917 }
 918
 919 /*
 920  * Return milliseconds since 00:00 GMT in network format.
 921  */
 922 uint32_t
 923 iptime(void)
 924 {
 925         struct timeval atv;
 926         u_long t;
 927
 928         getmicrotime(&atv);
 929         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 930         return (htonl(t));
 931 }
 932
 933 /*
 934  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 935  * given current value MTU.  If DIR is less than zero, a larger plateau
 936  * is returned; otherwise, a smaller value is returned.
 937  */
 938 int
 939 ip_next_mtu(int mtu, int dir)
 940 {
 941         static int mtutab[] = {
 942                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 943                 296, 68, 0
 944         };
 945         int i, size;
 946
 947         size = (sizeof mtutab) / (sizeof mtutab[0]);
 948         if (dir >= 0) {
 949                 for (i = 0; i < size; i++)
 950                         if (mtu > mtutab[i])
 951                                 return mtutab[i];
 952         } else {
 953                 for (i = size - 1; i >= 0; i--)
 954                         if (mtu < mtutab[i])
 955                                 return mtutab[i];
 956                 if (mtu == mtutab[0])
 957                         return mtutab[0];
 958         }
 959         return 0;
 960 }
 961 #endif /* INET */
 962
 963
 964 /*
 965  * badport_bandlim() - check for ICMP bandwidth limit
 966  *
 967  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 968  *      hit our bandwidth limit and it is not ok.
 969  *
 970  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 971  *
 972  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 973  *      values.  We may eventually remove this separation (and simplify the
 974  *      code further).
 975  *
 976  *      Note that the printing of the error message is delayed so we can
 977  *      properly print the icmp error rate that the system was trying to do
 978  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 979  *      the 'final' error, but it doesn't make sense to solve the printing
 980  *      delay with more complex code.
 981  */
 982
 983 int
 984 badport_bandlim(int which)
 985 {
 986
 987 #define N(a)    (sizeof (a) / sizeof (a[0]))
 988         static struct rate {
 989                 const char      *type;
 990                 struct timeval  lasttime;
 991                 int             curpps;
 992         } rates[BANDLIM_MAX+1] = {
 993                 { "icmp unreach response" },
 994                 { "icmp ping response" },
 995                 { "icmp tstamp response" },
 996                 { "closed port RST response" },
 997                 { "open port RST response" },
 998                 { "icmp6 unreach response" },
 999                 { "sctp ootb response" }
1000         };
1001
1002         /*
1003          * Return ok status if feature disabled or argument out of range.
1004          */
1005         if (V_icmplim > 0 && (u_int) which < N(rates)) {
1006                 struct rate *r = &rates[which];
1007                 int opps = r->curpps;
1008
1009                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
1010                         return -1;      /* discard packet */
1011                 /*
1012                  * If we've dropped below the threshold after having
1013                  * rate-limited traffic print the message.  This preserves
1014                  * the previous behaviour at the expense of added complexity.
1015                  */
1016                 if (V_icmplim_output && opps > V_icmplim)
1017                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
1018                                 r->type, opps, V_icmplim);
1019         }
1020         return 0;                       /* okay to send packet */
1021 #undef N
1022 }