sys/netinet/ip_icmp.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 4. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_icmp.c   8.2 (Berkeley) 1/4/94
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __FBSDID("$FreeBSD$");
  34
  35 #include "opt_inet.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/mbuf.h>
  40 #include <sys/protosw.h>
  41 #include <sys/socket.h>
  42 #include <sys/time.h>
  43 #include <sys/kernel.h>
  44 #include <sys/sysctl.h>
  45 #include <sys/syslog.h>
  46
  47 #include <net/if.h>
  48 #include <net/if_types.h>
  49 #include <net/route.h>
  50 #include <net/vnet.h>
  51
  52 #include <netinet/in.h>
  53 #include <netinet/in_pcb.h>
  54 #include <netinet/in_systm.h>
  55 #include <netinet/in_var.h>
  56 #include <netinet/ip.h>
  57 #include <netinet/ip_icmp.h>
  58 #include <netinet/ip_var.h>
  59 #include <netinet/ip_options.h>
  60 #include <netinet/tcp.h>
  61 #include <netinet/tcp_var.h>
  62 #include <netinet/tcpip.h>
  63 #include <netinet/icmp_var.h>
  64
  65 #ifdef INET
  66
  67 #include <machine/in_cksum.h>
  68
  69 #include <security/mac/mac_framework.h>
  70 #endif /* INET */
  71
  72 /*
  73  * ICMP routines: error generation, receive packet processing, and
  74  * routines to turnaround packets back to the originator, and
  75  * host table maintenance routines.
  76  */
  77 static VNET_DEFINE(int, icmplim) = 200;
  78 #define V_icmplim                       VNET(icmplim)
  79 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_RW,
  80         &VNET_NAME(icmplim), 0,
  81         "Maximum number of ICMP responses per second");
  82
  83 static VNET_DEFINE(int, icmplim_output) = 1;
  84 #define V_icmplim_output                VNET(icmplim_output)
  85 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_RW,
  86         &VNET_NAME(icmplim_output), 0,
  87         "Enable logging of ICMP response rate limiting");
  88
  89 #ifdef INET
  90 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
  91 VNET_PCPUSTAT_SYSINIT(icmpstat);
  92 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
  93     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
  94
  95 #ifdef VIMAGE
  96 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
  97 #endif /* VIMAGE */
  98
  99 static VNET_DEFINE(int, icmpmaskrepl) = 0;
 100 #define V_icmpmaskrepl                  VNET(icmpmaskrepl)
 101 SYSCTL_VNET_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_RW,
 102         &VNET_NAME(icmpmaskrepl), 0,
 103         "Reply to ICMP Address Mask Request packets.");
 104
 105 static VNET_DEFINE(u_int, icmpmaskfake) = 0;
 106 #define V_icmpmaskfake                  VNET(icmpmaskfake)
 107 SYSCTL_VNET_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_RW,
 108         &VNET_NAME(icmpmaskfake), 0,
 109         "Fake reply to ICMP Address Mask Request packets.");
 110
 111 VNET_DEFINE(int, drop_redirect) = 0;
 112
 113 static VNET_DEFINE(int, log_redirect) = 0;
 114 #define V_log_redirect                  VNET(log_redirect)
 115 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_RW,
 116         &VNET_NAME(log_redirect), 0,
 117         "Log ICMP redirects to the console");
 118
 119 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
 120 #define V_reply_src                     VNET(reply_src)
 121 SYSCTL_VNET_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_RW,
 122         &VNET_NAME(reply_src), IFNAMSIZ,
 123         "icmp reply source for non-local packets.");
 124
 125 static VNET_DEFINE(int, icmp_rfi) = 0;
 126 #define V_icmp_rfi                      VNET(icmp_rfi)
 127 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_RW,
 128         &VNET_NAME(icmp_rfi), 0,
 129         "ICMP reply from incoming interface for non-local packets");
 130
 131 static VNET_DEFINE(int, icmp_quotelen) = 8;
 132 #define V_icmp_quotelen                 VNET(icmp_quotelen)
 133 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_RW,
 134         &VNET_NAME(icmp_quotelen), 0,
 135         "Number of bytes from original packet to quote in ICMP reply");
 136
 137 /*
 138  * ICMP broadcast echo sysctl
 139  */
 140 static VNET_DEFINE(int, icmpbmcastecho) = 0;
 141 #define V_icmpbmcastecho                VNET(icmpbmcastecho)
 142 SYSCTL_VNET_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_RW,
 143         &VNET_NAME(icmpbmcastecho), 0,
 144         "");
 145
 146 static VNET_DEFINE(int, icmptstamprepl) = 1;
 147 #define V_icmptstamprepl                VNET(icmptstamprepl)
 148 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
 149         &VNET_NAME(icmptstamprepl), 0, "Respond to ICMP Timestamp packets");
 150
 151 static VNET_DEFINE(int, error_keeptags) = 0;
 152 #define V_error_keeptags                VNET(error_keeptags)
 153 SYSCTL_INT(_net_inet_icmp, OID_AUTO, error_keeptags, CTLFLAG_VNET | CTLFLAG_RW,
 154         &VNET_NAME(error_keeptags), 0,
 155         "ICMP error response keeps copy of mbuf_tags of original packet");
 156
 157 #ifdef ICMPPRINTFS
 158 int     icmpprintfs = 0;
 159 #endif
 160
 161 static void     icmp_reflect(struct mbuf *);
 162 static void     icmp_send(struct mbuf *, struct mbuf *);
 163
 164 extern  struct protosw inetsw[];
 165
 166 static int
 167 sysctl_net_icmp_drop_redir(SYSCTL_HANDLER_ARGS)
 168 {
 169         int error, new;
 170         int i;
 171         struct radix_node_head *rnh;
 172
 173         new = V_drop_redirect;
 174         error = sysctl_handle_int(oidp, &new, 0, req);
 175         if (error == 0 && req->newptr) {
 176                 new = (new != 0) ? 1 : 0;
 177
 178                 if (new == V_drop_redirect)
 179                         return (0);
 180
 181                 for (i = 0; i < rt_numfibs; i++) {
 182                         if ((rnh = rt_tables_get_rnh(i, AF_INET)) == NULL)
 183                                 continue;
 184                         RADIX_NODE_HEAD_LOCK(rnh);
 185                         in_setmatchfunc(rnh, new);
 186                         RADIX_NODE_HEAD_UNLOCK(rnh);
 187                 }
 188
 189                 V_drop_redirect = new;
 190         }
 191
 192         return (error);
 193 }
 194
 195 SYSCTL_VNET_PROC(_net_inet_icmp, OID_AUTO, drop_redirect,
 196     CTLTYPE_INT|CTLFLAG_RW, 0, 0,
 197     sysctl_net_icmp_drop_redir, "I", "Ignore ICMP redirects");
 198
 199 /*
 200  * Kernel module interface for updating icmpstat.  The argument is an index
 201  * into icmpstat treated as an array of u_long.  While this encodes the
 202  * general layout of icmpstat into the caller, it doesn't encode its
 203  * location, so that future changes to add, for example, per-CPU stats
 204  * support won't cause binary compatibility problems for kernel modules.
 205  */
 206 void
 207 kmod_icmpstat_inc(int statnum)
 208 {
 209
 210         counter_u64_add(VNET(icmpstat)[statnum], 1);
 211 }
 212
 213 /*
 214  * Generate an error packet of type error
 215  * in response to bad packet ip.
 216  */
 217 void
 218 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 219 {
 220         register struct ip *oip = mtod(n, struct ip *), *nip;
 221         register unsigned oiphlen = oip->ip_hl << 2;
 222         register struct icmp *icp;
 223         register struct mbuf *m;
 224         unsigned icmplen, icmpelen, nlen;
 225
 226         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__));
 227 #ifdef ICMPPRINTFS
 228         if (icmpprintfs)
 229                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 230 #endif
 231         if (type != ICMP_REDIRECT)
 232                 ICMPSTAT_INC(icps_error);
 233         /*
 234          * Don't send error:
 235          *  if the original packet was encrypted.
 236          *  if not the first fragment of message.
 237          *  in response to a multicast or broadcast packet.
 238          *  if the old packet protocol was an ICMP error message.
 239          */
 240         if (n->m_flags & M_DECRYPTED)
 241                 goto freeit;
 242         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 243                 goto freeit;
 244         if (n->m_flags & (M_BCAST|M_MCAST))
 245                 goto freeit;
 246         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 247           n->m_len >= oiphlen + ICMP_MINLEN &&
 248           !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) {
 249                 ICMPSTAT_INC(icps_oldicmp);
 250                 goto freeit;
 251         }
 252         /* Drop if IP header plus 8 bytes is not contignous in first mbuf. */
 253         if (oiphlen + 8 > n->m_len)
 254                 goto freeit;
 255         /*
 256          * Calculate length to quote from original packet and
 257          * prevent the ICMP mbuf from overflowing.
 258          * Unfortunatly this is non-trivial since ip_forward()
 259          * sends us truncated packets.
 260          */
 261         nlen = m_length(n, NULL);
 262         if (oip->ip_p == IPPROTO_TCP) {
 263                 struct tcphdr *th;
 264                 int tcphlen;
 265
 266                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 267                     n->m_next == NULL)
 268                         goto stdreply;
 269                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 270                     ((n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL))
 271                         goto freeit;
 272                 th = (struct tcphdr *)((caddr_t)oip + oiphlen);
 273                 tcphlen = th->th_off << 2;
 274                 if (tcphlen < sizeof(struct tcphdr))
 275                         goto freeit;
 276                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 277                         goto freeit;
 278                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 279                         goto stdreply;
 280                 if (n->m_len < oiphlen + tcphlen &&
 281                     ((n = m_pullup(n, oiphlen + tcphlen)) == NULL))
 282                         goto freeit;
 283                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 284                     ntohs(oip->ip_len) - oiphlen));
 285         } else
 286 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 287
 288         icmplen = min(oiphlen + icmpelen, nlen);
 289         if (icmplen < sizeof(struct ip))
 290                 goto freeit;
 291
 292         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 293                 m = m_gethdr(M_NOWAIT, MT_DATA);
 294         else
 295                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 296         if (m == NULL)
 297                 goto freeit;
 298 #ifdef MAC
 299         mac_netinet_icmp_reply(n, m);
 300 #endif
 301         icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN);
 302         m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
 303         m->m_data += sizeof(struct ip);
 304         m->m_len = ICMP_MINLEN + icmplen;
 305
 306         /* XXX MRT  make the outgoing packet use the same FIB
 307          * that was associated with the incoming packet
 308          */
 309         M_SETFIB(m, M_GETFIB(n));
 310         icp = mtod(m, struct icmp *);
 311         ICMPSTAT_INC(icps_outhist[type]);
 312         icp->icmp_type = type;
 313         if (type == ICMP_REDIRECT)
 314                 icp->icmp_gwaddr.s_addr = dest;
 315         else {
 316                 icp->icmp_void = 0;
 317                 /*
 318                  * The following assignments assume an overlay with the
 319                  * just zeroed icmp_void field.
 320                  */
 321                 if (type == ICMP_PARAMPROB) {
 322                         icp->icmp_pptr = code;
 323                         code = 0;
 324                 } else if (type == ICMP_UNREACH &&
 325                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 326                         icp->icmp_nextmtu = htons(mtu);
 327                 }
 328         }
 329         icp->icmp_code = code;
 330
 331         /*
 332          * Copy the quotation into ICMP message and
 333          * convert quoted IP header back to network representation.
 334          */
 335         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 336         nip = &icp->icmp_ip;
 337
 338         /*
 339          * Set up ICMP message mbuf and copy old IP header (without options
 340          * in front of ICMP message.
 341          * If the original mbuf was meant to bypass the firewall, the error
 342          * reply should bypass as well.
 343          */
 344         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 345         KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
 346             ("insufficient space for ip header"));
 347         m->m_data -= sizeof(struct ip);
 348         m->m_len += sizeof(struct ip);
 349         m->m_pkthdr.len = m->m_len;
 350         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 351         nip = mtod(m, struct ip *);
 352         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 353         nip->ip_len = htons(m->m_len);
 354         nip->ip_v = IPVERSION;
 355         nip->ip_hl = 5;
 356         nip->ip_p = IPPROTO_ICMP;
 357         nip->ip_tos = 0;
 358         nip->ip_off = 0;
 359
 360         if (V_error_keeptags)
 361                 m_tag_copy_chain(m, n, M_NOWAIT);
 362
 363         icmp_reflect(m);
 364
 365 freeit:
 366         m_freem(n);
 367 }
 368
 369 /*
 370  * Process a received ICMP message.
 371  */
 372 void
 373 icmp_input(struct mbuf *m, int off)
 374 {
 375         struct icmp *icp;
 376         struct in_ifaddr *ia;
 377         struct ip *ip = mtod(m, struct ip *);
 378         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 379         int hlen = off;
 380         int icmplen = ntohs(ip->ip_len) - off;
 381         int i, code;
 382         void (*ctlfunc)(int, struct sockaddr *, void *);
 383         int fibnum;
 384
 385         /*
 386          * Locate icmp structure in mbuf, and check
 387          * that not corrupted and of at least minimum length.
 388          */
 389 #ifdef ICMPPRINTFS
 390         if (icmpprintfs) {
 391                 char buf[4 * sizeof "123"];
 392                 strcpy(buf, inet_ntoa(ip->ip_src));
 393                 printf("icmp_input from %s to %s, len %d\n",
 394                        buf, inet_ntoa(ip->ip_dst), icmplen);
 395         }
 396 #endif
 397         if (icmplen < ICMP_MINLEN) {
 398                 ICMPSTAT_INC(icps_tooshort);
 399                 goto freeit;
 400         }
 401         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 402         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 403                 ICMPSTAT_INC(icps_tooshort);
 404                 return;
 405         }
 406         ip = mtod(m, struct ip *);
 407         m->m_len -= hlen;
 408         m->m_data += hlen;
 409         icp = mtod(m, struct icmp *);
 410         if (in_cksum(m, icmplen)) {
 411                 ICMPSTAT_INC(icps_checksum);
 412                 goto freeit;
 413         }
 414         m->m_len += hlen;
 415         m->m_data -= hlen;
 416
 417         if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
 418                 /*
 419                  * Deliver very specific ICMP type only.
 420                  */
 421                 switch (icp->icmp_type) {
 422                 case ICMP_UNREACH:
 423                 case ICMP_TIMXCEED:
 424                         break;
 425                 default:
 426                         goto freeit;
 427                 }
 428         }
 429
 430 #ifdef ICMPPRINTFS
 431         if (icmpprintfs)
 432                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 433                     icp->icmp_code);
 434 #endif
 435
 436         /*
 437          * Message type specific processing.
 438          */
 439         if (icp->icmp_type > ICMP_MAXTYPE)
 440                 goto raw;
 441
 442         /* Initialize */
 443         bzero(&icmpsrc, sizeof(icmpsrc));
 444         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 445         icmpsrc.sin_family = AF_INET;
 446         bzero(&icmpdst, sizeof(icmpdst));
 447         icmpdst.sin_len = sizeof(struct sockaddr_in);
 448         icmpdst.sin_family = AF_INET;
 449         bzero(&icmpgw, sizeof(icmpgw));
 450         icmpgw.sin_len = sizeof(struct sockaddr_in);
 451         icmpgw.sin_family = AF_INET;
 452
 453         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 454         code = icp->icmp_code;
 455         switch (icp->icmp_type) {
 456
 457         case ICMP_UNREACH:
 458                 switch (code) {
 459                         case ICMP_UNREACH_NET:
 460                         case ICMP_UNREACH_HOST:
 461                         case ICMP_UNREACH_SRCFAIL:
 462                         case ICMP_UNREACH_NET_UNKNOWN:
 463                         case ICMP_UNREACH_HOST_UNKNOWN:
 464                         case ICMP_UNREACH_ISOLATED:
 465                         case ICMP_UNREACH_TOSNET:
 466                         case ICMP_UNREACH_TOSHOST:
 467                         case ICMP_UNREACH_HOST_PRECEDENCE:
 468                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 469                                 code = PRC_UNREACH_NET;
 470                                 break;
 471
 472                         case ICMP_UNREACH_NEEDFRAG:
 473                                 code = PRC_MSGSIZE;
 474                                 break;
 475
 476                         /*
 477                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 478                          * Treat subcodes 2,3 as immediate RST
 479                          */
 480                         case ICMP_UNREACH_PROTOCOL:
 481                         case ICMP_UNREACH_PORT:
 482                                 code = PRC_UNREACH_PORT;
 483                                 break;
 484
 485                         case ICMP_UNREACH_NET_PROHIB:
 486                         case ICMP_UNREACH_HOST_PROHIB:
 487                         case ICMP_UNREACH_FILTER_PROHIB:
 488                                 code = PRC_UNREACH_ADMIN_PROHIB;
 489                                 break;
 490
 491                         default:
 492                                 goto badcode;
 493                 }
 494                 goto deliver;
 495
 496         case ICMP_TIMXCEED:
 497                 if (code > 1)
 498                         goto badcode;
 499                 code += PRC_TIMXCEED_INTRANS;
 500                 goto deliver;
 501
 502         case ICMP_PARAMPROB:
 503                 if (code > 1)
 504                         goto badcode;
 505                 code = PRC_PARAMPROB;
 506                 goto deliver;
 507
 508         case ICMP_SOURCEQUENCH:
 509                 if (code)
 510                         goto badcode;
 511                 code = PRC_QUENCH;
 512         deliver:
 513                 /*
 514                  * Problem with datagram; advise higher level routines.
 515                  */
 516                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 517                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 518                         ICMPSTAT_INC(icps_badlen);
 519                         goto freeit;
 520                 }
 521                 /* Discard ICMP's in response to multicast packets */
 522                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 523                         goto badcode;
 524 #ifdef ICMPPRINTFS
 525                 if (icmpprintfs)
 526                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 527 #endif
 528                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 529                 /*
 530                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 531                  * notification to TCP layer.
 532                  */
 533                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 534                 if (ctlfunc)
 535                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 536                                    (void *)&icp->icmp_ip);
 537                 break;
 538
 539         badcode:
 540                 ICMPSTAT_INC(icps_badcode);
 541                 break;
 542
 543         case ICMP_ECHO:
 544                 if (!V_icmpbmcastecho
 545                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 546                         ICMPSTAT_INC(icps_bmcastecho);
 547                         break;
 548                 }
 549                 icp->icmp_type = ICMP_ECHOREPLY;
 550                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 551                         goto freeit;
 552                 else
 553                         goto reflect;
 554
 555         case ICMP_TSTAMP:
 556                 if (V_icmptstamprepl == 0)
 557                         break;
 558                 if (!V_icmpbmcastecho
 559                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 560                         ICMPSTAT_INC(icps_bmcasttstamp);
 561                         break;
 562                 }
 563                 if (icmplen < ICMP_TSLEN) {
 564                         ICMPSTAT_INC(icps_badlen);
 565                         break;
 566                 }
 567                 icp->icmp_type = ICMP_TSTAMPREPLY;
 568                 icp->icmp_rtime = iptime();
 569                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 570                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 571                         goto freeit;
 572                 else
 573                         goto reflect;
 574
 575         case ICMP_MASKREQ:
 576                 if (V_icmpmaskrepl == 0)
 577                         break;
 578                 /*
 579                  * We are not able to respond with all ones broadcast
 580                  * unless we receive it over a point-to-point interface.
 581                  */
 582                 if (icmplen < ICMP_MASKLEN)
 583                         break;
 584                 switch (ip->ip_dst.s_addr) {
 585
 586                 case INADDR_BROADCAST:
 587                 case INADDR_ANY:
 588                         icmpdst.sin_addr = ip->ip_src;
 589                         break;
 590
 591                 default:
 592                         icmpdst.sin_addr = ip->ip_dst;
 593                 }
 594                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 595                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 596                 if (ia == NULL)
 597                         break;
 598                 if (ia->ia_ifp == NULL) {
 599                         ifa_free(&ia->ia_ifa);
 600                         break;
 601                 }
 602                 icp->icmp_type = ICMP_MASKREPLY;
 603                 if (V_icmpmaskfake == 0)
 604                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 605                 else
 606                         icp->icmp_mask = V_icmpmaskfake;
 607                 if (ip->ip_src.s_addr == 0) {
 608                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 609                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 610                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 611                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 612                 }
 613                 ifa_free(&ia->ia_ifa);
 614 reflect:
 615                 ICMPSTAT_INC(icps_reflect);
 616                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 617                 icmp_reflect(m);
 618                 return;
 619
 620         case ICMP_REDIRECT:
 621                 if (V_log_redirect) {
 622                         u_long src, dst, gw;
 623
 624                         src = ntohl(ip->ip_src.s_addr);
 625                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 626                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 627                         printf("icmp redirect from %d.%d.%d.%d: "
 628                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 629                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 630                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 631                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 632                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 633                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 634                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 635                 }
 636                 /*
 637                  * RFC1812 says we must ignore ICMP redirects if we
 638                  * are acting as router.
 639                  */
 640                 if (V_drop_redirect || V_ipforwarding)
 641                         break;
 642                 if (code > 3)
 643                         goto badcode;
 644                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 645                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 646                         ICMPSTAT_INC(icps_badlen);
 647                         break;
 648                 }
 649                 /*
 650                  * Short circuit routing redirects to force
 651                  * immediate change in the kernel's routing
 652                  * tables.  The message is also handed to anyone
 653                  * listening on a raw socket (e.g. the routing
 654                  * daemon for use in updating its tables).
 655                  */
 656                 icmpgw.sin_addr = ip->ip_src;
 657                 icmpdst.sin_addr = icp->icmp_gwaddr;
 658 #ifdef  ICMPPRINTFS
 659                 if (icmpprintfs) {
 660                         char buf[4 * sizeof "123"];
 661                         strcpy(buf, inet_ntoa(icp->icmp_ip.ip_dst));
 662
 663                         printf("redirect dst %s to %s\n",
 664                                buf, inet_ntoa(icp->icmp_gwaddr));
 665                 }
 666 #endif
 667                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 668                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 669                         in_rtredirect((struct sockaddr *)&icmpsrc,
 670                           (struct sockaddr *)&icmpdst,
 671                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 672                           (struct sockaddr *)&icmpgw, fibnum);
 673                 }
 674                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 675                 break;
 676
 677         /*
 678          * No kernel processing for the following;
 679          * just fall through to send to raw listener.
 680          */
 681         case ICMP_ECHOREPLY:
 682         case ICMP_ROUTERADVERT:
 683         case ICMP_ROUTERSOLICIT:
 684         case ICMP_TSTAMPREPLY:
 685         case ICMP_IREQREPLY:
 686         case ICMP_MASKREPLY:
 687         default:
 688                 break;
 689         }
 690
 691 raw:
 692         rip_input(m, off);
 693         return;
 694
 695 freeit:
 696         m_freem(m);
 697 }
 698
 699 /*
 700  * Reflect the ip packet back to the source
 701  */
 702 static void
 703 icmp_reflect(struct mbuf *m)
 704 {
 705         struct ip *ip = mtod(m, struct ip *);
 706         struct ifaddr *ifa;
 707         struct ifnet *ifp;
 708         struct in_ifaddr *ia;
 709         struct in_addr t;
 710         struct mbuf *opts = 0;
 711         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 712
 713         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 714             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 715             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 716                 m_freem(m);     /* Bad return address */
 717                 ICMPSTAT_INC(icps_badaddr);
 718                 goto done;      /* Ip_output() will check for broadcast */
 719         }
 720
 721         t = ip->ip_dst;
 722         ip->ip_dst = ip->ip_src;
 723
 724         /*
 725          * Source selection for ICMP replies:
 726          *
 727          * If the incoming packet was addressed directly to one of our
 728          * own addresses, use dst as the src for the reply.
 729          */
 730         IN_IFADDR_RLOCK();
 731         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 732                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 733                         t = IA_SIN(ia)->sin_addr;
 734                         IN_IFADDR_RUNLOCK();
 735                         goto match;
 736                 }
 737         }
 738         IN_IFADDR_RUNLOCK();
 739
 740         /*
 741          * If the incoming packet was addressed to one of our broadcast
 742          * addresses, use the first non-broadcast address which corresponds
 743          * to the incoming interface.
 744          */
 745         ifp = m->m_pkthdr.rcvif;
 746         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 747                 IF_ADDR_RLOCK(ifp);
 748                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 749                         if (ifa->ifa_addr->sa_family != AF_INET)
 750                                 continue;
 751                         ia = ifatoia(ifa);
 752                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 753                             t.s_addr) {
 754                                 t = IA_SIN(ia)->sin_addr;
 755                                 IF_ADDR_RUNLOCK(ifp);
 756                                 goto match;
 757                         }
 758                 }
 759                 IF_ADDR_RUNLOCK(ifp);
 760         }
 761         /*
 762          * If the packet was transiting through us, use the address of
 763          * the interface the packet came through in.  If that interface
 764          * doesn't have a suitable IP address, the normal selection
 765          * criteria apply.
 766          */
 767         if (V_icmp_rfi && ifp != NULL) {
 768                 IF_ADDR_RLOCK(ifp);
 769                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 770                         if (ifa->ifa_addr->sa_family != AF_INET)
 771                                 continue;
 772                         ia = ifatoia(ifa);
 773                         t = IA_SIN(ia)->sin_addr;
 774                         IF_ADDR_RUNLOCK(ifp);
 775                         goto match;
 776                 }
 777                 IF_ADDR_RUNLOCK(ifp);
 778         }
 779         /*
 780          * If the incoming packet was not addressed directly to us, use
 781          * designated interface for icmp replies specified by sysctl
 782          * net.inet.icmp.reply_src (default not set). Otherwise continue
 783          * with normal source selection.
 784          */
 785         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 786                 IF_ADDR_RLOCK(ifp);
 787                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 788                         if (ifa->ifa_addr->sa_family != AF_INET)
 789                                 continue;
 790                         ia = ifatoia(ifa);
 791                         t = IA_SIN(ia)->sin_addr;
 792                         IF_ADDR_RUNLOCK(ifp);
 793                         goto match;
 794                 }
 795                 IF_ADDR_RUNLOCK(ifp);
 796         }
 797         /*
 798          * If the packet was transiting through us, use the address of
 799          * the interface that is the closest to the packet source.
 800          * When we don't have a route back to the packet source, stop here
 801          * and drop the packet.
 802          */
 803         ia = ip_rtaddr(ip->ip_dst, M_GETFIB(m));
 804         if (ia == NULL) {
 805                 m_freem(m);
 806                 ICMPSTAT_INC(icps_noroute);
 807                 goto done;
 808         }
 809         t = IA_SIN(ia)->sin_addr;
 810         ifa_free(&ia->ia_ifa);
 811 match:
 812 #ifdef MAC
 813         mac_netinet_icmp_replyinplace(m);
 814 #endif
 815         ip->ip_src = t;
 816         ip->ip_ttl = V_ip_defttl;
 817
 818         if (optlen > 0) {
 819                 register u_char *cp;
 820                 int opt, cnt;
 821                 u_int len;
 822
 823                 /*
 824                  * Retrieve any source routing from the incoming packet;
 825                  * add on any record-route or timestamp options.
 826                  */
 827                 cp = (u_char *) (ip + 1);
 828                 if ((opts = ip_srcroute(m)) == 0 &&
 829                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 830                         opts->m_len = sizeof(struct in_addr);
 831                         mtod(opts, struct in_addr *)->s_addr = 0;
 832                 }
 833                 if (opts) {
 834 #ifdef ICMPPRINTFS
 835                     if (icmpprintfs)
 836                             printf("icmp_reflect optlen %d rt %d => ",
 837                                 optlen, opts->m_len);
 838 #endif
 839                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 840                             opt = cp[IPOPT_OPTVAL];
 841                             if (opt == IPOPT_EOL)
 842                                     break;
 843                             if (opt == IPOPT_NOP)
 844                                     len = 1;
 845                             else {
 846                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 847                                             break;
 848                                     len = cp[IPOPT_OLEN];
 849                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 850                                         len > cnt)
 851                                             break;
 852                             }
 853                             /*
 854                              * Should check for overflow, but it "can't happen"
 855                              */
 856                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 857                                 opt == IPOPT_SECURITY) {
 858                                     bcopy((caddr_t)cp,
 859                                         mtod(opts, caddr_t) + opts->m_len, len);
 860                                     opts->m_len += len;
 861                             }
 862                     }
 863                     /* Terminate & pad, if necessary */
 864                     cnt = opts->m_len % 4;
 865                     if (cnt) {
 866                             for (; cnt < 4; cnt++) {
 867                                     *(mtod(opts, caddr_t) + opts->m_len) =
 868                                         IPOPT_EOL;
 869                                     opts->m_len++;
 870                             }
 871                     }
 872 #ifdef ICMPPRINTFS
 873                     if (icmpprintfs)
 874                             printf("%d\n", opts->m_len);
 875 #endif
 876                 }
 877                 ip_stripoptions(m);
 878         }
 879         m_tag_delete_nonpersistent(m);
 880         m->m_flags &= ~(M_BCAST|M_MCAST);
 881         icmp_send(m, opts);
 882 done:
 883         if (opts)
 884                 (void)m_free(opts);
 885 }
 886
 887 /*
 888  * Send an icmp packet back to the ip level,
 889  * after supplying a checksum.
 890  */
 891 static void
 892 icmp_send(struct mbuf *m, struct mbuf *opts)
 893 {
 894         register struct ip *ip = mtod(m, struct ip *);
 895         register int hlen;
 896         register struct icmp *icp;
 897
 898         hlen = ip->ip_hl << 2;
 899         m->m_data += hlen;
 900         m->m_len -= hlen;
 901         icp = mtod(m, struct icmp *);
 902         icp->icmp_cksum = 0;
 903         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 904         m->m_data -= hlen;
 905         m->m_len += hlen;
 906         m->m_pkthdr.rcvif = (struct ifnet *)0;
 907 #ifdef ICMPPRINTFS
 908         if (icmpprintfs) {
 909                 char buf[4 * sizeof "123"];
 910                 strcpy(buf, inet_ntoa(ip->ip_dst));
 911                 printf("icmp_send dst %s src %s\n",
 912                        buf, inet_ntoa(ip->ip_src));
 913         }
 914 #endif
 915         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 916 }
 917
 918 /*
 919  * Return milliseconds since 00:00 GMT in network format.
 920  */
 921 uint32_t
 922 iptime(void)
 923 {
 924         struct timeval atv;
 925         u_long t;
 926
 927         getmicrotime(&atv);
 928         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 929         return (htonl(t));
 930 }
 931
 932 /*
 933  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 934  * given current value MTU.  If DIR is less than zero, a larger plateau
 935  * is returned; otherwise, a smaller value is returned.
 936  */
 937 int
 938 ip_next_mtu(int mtu, int dir)
 939 {
 940         static int mtutab[] = {
 941                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 942                 296, 68, 0
 943         };
 944         int i, size;
 945
 946         size = (sizeof mtutab) / (sizeof mtutab[0]);
 947         if (dir >= 0) {
 948                 for (i = 0; i < size; i++)
 949                         if (mtu > mtutab[i])
 950                                 return mtutab[i];
 951         } else {
 952                 for (i = size - 1; i >= 0; i--)
 953                         if (mtu < mtutab[i])
 954                                 return mtutab[i];
 955                 if (mtu == mtutab[0])
 956                         return mtutab[0];
 957         }
 958         return 0;
 959 }
 960 #endif /* INET */
 961
 962
 963 /*
 964  * badport_bandlim() - check for ICMP bandwidth limit
 965  *
 966  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 967  *      hit our bandwidth limit and it is not ok.
 968  *
 969  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 970  *
 971  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 972  *      values.  We may eventually remove this separation (and simplify the
 973  *      code further).
 974  *
 975  *      Note that the printing of the error message is delayed so we can
 976  *      properly print the icmp error rate that the system was trying to do
 977  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 978  *      the 'final' error, but it doesn't make sense to solve the printing
 979  *      delay with more complex code.
 980  */
 981
 982 int
 983 badport_bandlim(int which)
 984 {
 985
 986 #define N(a)    (sizeof (a) / sizeof (a[0]))
 987         static struct rate {
 988                 const char      *type;
 989                 struct timeval  lasttime;
 990                 int             curpps;
 991         } rates[BANDLIM_MAX+1] = {
 992                 { "icmp unreach response" },
 993                 { "icmp ping response" },
 994                 { "icmp tstamp response" },
 995                 { "closed port RST response" },
 996                 { "open port RST response" },
 997                 { "icmp6 unreach response" },
 998                 { "sctp ootb response" }
 999         };
1000
1001         /*
1002          * Return ok status if feature disabled or argument out of range.
1003          */
1004         if (V_icmplim > 0 && (u_int) which < N(rates)) {
1005                 struct rate *r = &rates[which];
1006                 int opps = r->curpps;
1007
1008                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
1009                         return -1;      /* discard packet */
1010                 /*
1011                  * If we've dropped below the threshold after having
1012                  * rate-limited traffic print the message.  This preserves
1013                  * the previous behaviour at the expense of added complexity.
1014                  */
1015                 if (V_icmplim_output && opps > V_icmplim)
1016                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
1017                                 r->type, opps, V_icmplim);
1018         }
1019         return 0;                       /* okay to send packet */
1020 #undef N
1021 }