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/lock.h>
  45 #include <sys/rmlock.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/syslog.h>
  48
  49 #include <net/if.h>
  50 #include <net/if_var.h>
  51 #include <net/if_types.h>
  52 #include <net/route.h>
  53 #include <net/vnet.h>
  54
  55 #include <netinet/in.h>
  56 #include <netinet/in_fib.h>
  57 #include <netinet/in_pcb.h>
  58 #include <netinet/in_systm.h>
  59 #include <netinet/in_var.h>
  60 #include <netinet/ip.h>
  61 #include <netinet/ip_icmp.h>
  62 #include <netinet/ip_var.h>
  63 #include <netinet/ip_options.h>
  64 #include <netinet/sctp.h>
  65 #include <netinet/tcp.h>
  66 #include <netinet/tcp_var.h>
  67 #include <netinet/tcpip.h>
  68 #include <netinet/icmp_var.h>
  69
  70 #ifdef INET
  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_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | 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_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | 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_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | 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_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | 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 #define V_drop_redirect                 VNET(drop_redirect)
 118 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
 119         &VNET_NAME(drop_redirect), 0,
 120         "Ignore ICMP redirects");
 121
 122 static VNET_DEFINE(int, log_redirect) = 0;
 123 #define V_log_redirect                  VNET(log_redirect)
 124 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
 125         &VNET_NAME(log_redirect), 0,
 126         "Log ICMP redirects to the console");
 127
 128 static VNET_DEFINE(char, reply_src[IFNAMSIZ]);
 129 #define V_reply_src                     VNET(reply_src)
 130 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
 131         &VNET_NAME(reply_src), IFNAMSIZ,
 132         "ICMP reply source for non-local packets");
 133
 134 static VNET_DEFINE(int, icmp_rfi) = 0;
 135 #define V_icmp_rfi                      VNET(icmp_rfi)
 136 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
 137         &VNET_NAME(icmp_rfi), 0,
 138         "ICMP reply from incoming interface for non-local packets");
 139
 140 static VNET_DEFINE(int, icmp_quotelen) = 8;
 141 #define V_icmp_quotelen                 VNET(icmp_quotelen)
 142 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
 143         &VNET_NAME(icmp_quotelen), 0,
 144         "Number of bytes from original packet to quote in ICMP reply");
 145
 146 static VNET_DEFINE(int, icmpbmcastecho) = 0;
 147 #define V_icmpbmcastecho                VNET(icmpbmcastecho)
 148 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
 149         &VNET_NAME(icmpbmcastecho), 0,
 150         "Reply to multicast ICMP Echo Request and Timestamp packets");
 151
 152 static VNET_DEFINE(int, icmptstamprepl) = 1;
 153 #define V_icmptstamprepl                VNET(icmptstamprepl)
 154 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW,
 155         &VNET_NAME(icmptstamprepl), 0,
 156         "Respond to ICMP Timestamp packets");
 157
 158 #ifdef ICMPPRINTFS
 159 int     icmpprintfs = 0;
 160 #endif
 161
 162 static void     icmp_reflect(struct mbuf *);
 163 static void     icmp_send(struct mbuf *, struct mbuf *);
 164
 165 extern  struct protosw inetsw[];
 166
 167 /*
 168  * Kernel module interface for updating icmpstat.  The argument is an index
 169  * into icmpstat treated as an array of u_long.  While this encodes the
 170  * general layout of icmpstat into the caller, it doesn't encode its
 171  * location, so that future changes to add, for example, per-CPU stats
 172  * support won't cause binary compatibility problems for kernel modules.
 173  */
 174 void
 175 kmod_icmpstat_inc(int statnum)
 176 {
 177
 178         counter_u64_add(VNET(icmpstat)[statnum], 1);
 179 }
 180
 181 /*
 182  * Generate an error packet of type error
 183  * in response to bad packet ip.
 184  */
 185 void
 186 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 187 {
 188         struct ip *oip, *nip;
 189         struct icmp *icp;
 190         struct mbuf *m;
 191         unsigned icmplen, icmpelen, nlen, oiphlen;
 192
 193         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
 194             __func__));
 195
 196         if (type != ICMP_REDIRECT)
 197                 ICMPSTAT_INC(icps_error);
 198         /*
 199          * Don't send error:
 200          *  if the original packet was encrypted.
 201          *  if not the first fragment of message.
 202          *  in response to a multicast or broadcast packet.
 203          *  if the old packet protocol was an ICMP error message.
 204          */
 205         if (n->m_flags & M_DECRYPTED)
 206                 goto freeit;
 207         if (n->m_flags & (M_BCAST|M_MCAST))
 208                 goto freeit;
 209
 210         /* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
 211         if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
 212                 goto freeit;
 213         oip = mtod(n, struct ip *);
 214         oiphlen = oip->ip_hl << 2;
 215         if (n->m_len < oiphlen + ICMP_MINLEN)
 216                 goto freeit;
 217 #ifdef ICMPPRINTFS
 218         if (icmpprintfs)
 219                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 220 #endif
 221         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 222                 goto freeit;
 223         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 224             !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
 225                 oiphlen))->icmp_type)) {
 226                 ICMPSTAT_INC(icps_oldicmp);
 227                 goto freeit;
 228         }
 229         /*
 230          * Calculate length to quote from original packet and
 231          * prevent the ICMP mbuf from overflowing.
 232          * Unfortunately this is non-trivial since ip_forward()
 233          * sends us truncated packets.
 234          */
 235         nlen = m_length(n, NULL);
 236         if (oip->ip_p == IPPROTO_TCP) {
 237                 struct tcphdr *th;
 238                 int tcphlen;
 239
 240                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 241                     n->m_next == NULL)
 242                         goto stdreply;
 243                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 244                     (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
 245                         goto freeit;
 246                 oip = mtod(n, struct ip *);
 247                 th = mtodo(n, oiphlen);
 248                 tcphlen = th->th_off << 2;
 249                 if (tcphlen < sizeof(struct tcphdr))
 250                         goto freeit;
 251                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 252                         goto freeit;
 253                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 254                         goto stdreply;
 255                 if (n->m_len < oiphlen + tcphlen &&
 256                     (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
 257                         goto freeit;
 258                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 259                     ntohs(oip->ip_len) - oiphlen));
 260         } else if (oip->ip_p == IPPROTO_SCTP) {
 261                 struct sctphdr *sh;
 262                 struct sctp_chunkhdr *ch;
 263
 264                 if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
 265                         goto stdreply;
 266                 if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
 267                     n->m_next == NULL)
 268                         goto stdreply;
 269                 if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
 270                     (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
 271                         goto freeit;
 272                 oip = mtod(n, struct ip *);
 273                 icmpelen = max(sizeof(struct sctphdr),
 274                     min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 275                 sh = mtodo(n, oiphlen);
 276                 if (ntohl(sh->v_tag) == 0 &&
 277                     ntohs(oip->ip_len) >= oiphlen +
 278                     sizeof(struct sctphdr) + 8 &&
 279                     (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
 280                      n->m_next != NULL)) {
 281                         if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
 282                             (n = m_pullup(n, oiphlen +
 283                             sizeof(struct sctphdr) + 8)) == NULL)
 284                                 goto freeit;
 285                         oip = mtod(n, struct ip *);
 286                         sh = mtodo(n, oiphlen);
 287                         ch = (struct sctp_chunkhdr *)(sh + 1);
 288                         if (ch->chunk_type == SCTP_INITIATION) {
 289                                 icmpelen = max(sizeof(struct sctphdr) + 8,
 290                                     min(V_icmp_quotelen, ntohs(oip->ip_len) -
 291                                     oiphlen));
 292                         }
 293                 }
 294         } else
 295 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
 296                     oiphlen));
 297
 298         icmplen = min(oiphlen + icmpelen, nlen);
 299         if (icmplen < sizeof(struct ip))
 300                 goto freeit;
 301
 302         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 303                 m = m_gethdr(M_NOWAIT, MT_DATA);
 304         else
 305                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 306         if (m == NULL)
 307                 goto freeit;
 308 #ifdef MAC
 309         mac_netinet_icmp_reply(n, m);
 310 #endif
 311         icmplen = min(icmplen, M_TRAILINGSPACE(m) -
 312             sizeof(struct ip) - ICMP_MINLEN);
 313         m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
 314         m->m_data += sizeof(struct ip);
 315         m->m_len = ICMP_MINLEN + icmplen;
 316
 317         /* XXX MRT  make the outgoing packet use the same FIB
 318          * that was associated with the incoming packet
 319          */
 320         M_SETFIB(m, M_GETFIB(n));
 321         icp = mtod(m, struct icmp *);
 322         ICMPSTAT_INC(icps_outhist[type]);
 323         icp->icmp_type = type;
 324         if (type == ICMP_REDIRECT)
 325                 icp->icmp_gwaddr.s_addr = dest;
 326         else {
 327                 icp->icmp_void = 0;
 328                 /*
 329                  * The following assignments assume an overlay with the
 330                  * just zeroed icmp_void field.
 331                  */
 332                 if (type == ICMP_PARAMPROB) {
 333                         icp->icmp_pptr = code;
 334                         code = 0;
 335                 } else if (type == ICMP_UNREACH &&
 336                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 337                         icp->icmp_nextmtu = htons(mtu);
 338                 }
 339         }
 340         icp->icmp_code = code;
 341
 342         /*
 343          * Copy the quotation into ICMP message and
 344          * convert quoted IP header back to network representation.
 345          */
 346         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 347         nip = &icp->icmp_ip;
 348
 349         /*
 350          * Set up ICMP message mbuf and copy old IP header (without options
 351          * in front of ICMP message.
 352          * If the original mbuf was meant to bypass the firewall, the error
 353          * reply should bypass as well.
 354          */
 355         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 356         KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
 357             ("insufficient space for ip header"));
 358         m->m_data -= sizeof(struct ip);
 359         m->m_len += sizeof(struct ip);
 360         m->m_pkthdr.len = m->m_len;
 361         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 362         nip = mtod(m, struct ip *);
 363         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 364         nip->ip_len = htons(m->m_len);
 365         nip->ip_v = IPVERSION;
 366         nip->ip_hl = 5;
 367         nip->ip_p = IPPROTO_ICMP;
 368         nip->ip_tos = 0;
 369         nip->ip_off = 0;
 370         icmp_reflect(m);
 371
 372 freeit:
 373         m_freem(n);
 374 }
 375
 376 /*
 377  * Process a received ICMP message.
 378  */
 379 int
 380 icmp_input(struct mbuf **mp, int *offp, int proto)
 381 {
 382         struct icmp *icp;
 383         struct in_ifaddr *ia;
 384         struct mbuf *m = *mp;
 385         struct ip *ip = mtod(m, struct ip *);
 386         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 387         int hlen = *offp;
 388         int icmplen = ntohs(ip->ip_len) - *offp;
 389         int i, code;
 390         void (*ctlfunc)(int, struct sockaddr *, void *);
 391         int fibnum;
 392
 393         *mp = NULL;
 394
 395         /*
 396          * Locate icmp structure in mbuf, and check
 397          * that not corrupted and of at least minimum length.
 398          */
 399 #ifdef ICMPPRINTFS
 400         if (icmpprintfs) {
 401                 char srcbuf[INET_ADDRSTRLEN];
 402                 char dstbuf[INET_ADDRSTRLEN];
 403
 404                 printf("icmp_input from %s to %s, len %d\n",
 405                     inet_ntoa_r(ip->ip_src, srcbuf),
 406                     inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
 407         }
 408 #endif
 409         if (icmplen < ICMP_MINLEN) {
 410                 ICMPSTAT_INC(icps_tooshort);
 411                 goto freeit;
 412         }
 413         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 414         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 415                 ICMPSTAT_INC(icps_tooshort);
 416                 return (IPPROTO_DONE);
 417         }
 418         ip = mtod(m, struct ip *);
 419         m->m_len -= hlen;
 420         m->m_data += hlen;
 421         icp = mtod(m, struct icmp *);
 422         if (in_cksum(m, icmplen)) {
 423                 ICMPSTAT_INC(icps_checksum);
 424                 goto freeit;
 425         }
 426         m->m_len += hlen;
 427         m->m_data -= hlen;
 428
 429 #ifdef ICMPPRINTFS
 430         if (icmpprintfs)
 431                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 432                     icp->icmp_code);
 433 #endif
 434
 435         /*
 436          * Message type specific processing.
 437          */
 438         if (icp->icmp_type > ICMP_MAXTYPE)
 439                 goto raw;
 440
 441         /* Initialize */
 442         bzero(&icmpsrc, sizeof(icmpsrc));
 443         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 444         icmpsrc.sin_family = AF_INET;
 445         bzero(&icmpdst, sizeof(icmpdst));
 446         icmpdst.sin_len = sizeof(struct sockaddr_in);
 447         icmpdst.sin_family = AF_INET;
 448         bzero(&icmpgw, sizeof(icmpgw));
 449         icmpgw.sin_len = sizeof(struct sockaddr_in);
 450         icmpgw.sin_family = AF_INET;
 451
 452         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 453         code = icp->icmp_code;
 454         switch (icp->icmp_type) {
 455
 456         case ICMP_UNREACH:
 457                 switch (code) {
 458                         case ICMP_UNREACH_NET:
 459                         case ICMP_UNREACH_HOST:
 460                         case ICMP_UNREACH_SRCFAIL:
 461                         case ICMP_UNREACH_NET_UNKNOWN:
 462                         case ICMP_UNREACH_HOST_UNKNOWN:
 463                         case ICMP_UNREACH_ISOLATED:
 464                         case ICMP_UNREACH_TOSNET:
 465                         case ICMP_UNREACH_TOSHOST:
 466                         case ICMP_UNREACH_HOST_PRECEDENCE:
 467                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 468                                 code = PRC_UNREACH_NET;
 469                                 break;
 470
 471                         case ICMP_UNREACH_NEEDFRAG:
 472                                 code = PRC_MSGSIZE;
 473                                 break;
 474
 475                         /*
 476                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 477                          * Treat subcodes 2,3 as immediate RST
 478                          */
 479                         case ICMP_UNREACH_PROTOCOL:
 480                                 code = PRC_UNREACH_PROTOCOL;
 481                                 break;
 482                         case ICMP_UNREACH_PORT:
 483                                 code = PRC_UNREACH_PORT;
 484                                 break;
 485
 486                         case ICMP_UNREACH_NET_PROHIB:
 487                         case ICMP_UNREACH_HOST_PROHIB:
 488                         case ICMP_UNREACH_FILTER_PROHIB:
 489                                 code = PRC_UNREACH_ADMIN_PROHIB;
 490                                 break;
 491
 492                         default:
 493                                 goto badcode;
 494                 }
 495                 goto deliver;
 496
 497         case ICMP_TIMXCEED:
 498                 if (code > 1)
 499                         goto badcode;
 500                 code += PRC_TIMXCEED_INTRANS;
 501                 goto deliver;
 502
 503         case ICMP_PARAMPROB:
 504                 if (code > 1)
 505                         goto badcode;
 506                 code = PRC_PARAMPROB;
 507         deliver:
 508                 /*
 509                  * Problem with datagram; advise higher level routines.
 510                  */
 511                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 512                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 513                         ICMPSTAT_INC(icps_badlen);
 514                         goto freeit;
 515                 }
 516                 /* Discard ICMP's in response to multicast packets */
 517                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 518                         goto badcode;
 519 #ifdef ICMPPRINTFS
 520                 if (icmpprintfs)
 521                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 522 #endif
 523                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 524                 /*
 525                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 526                  * notification to TCP layer.
 527                  */
 528                 i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
 529                 ip_stripoptions(m);
 530                 if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
 531                         /* This should actually not happen */
 532                         ICMPSTAT_INC(icps_tooshort);
 533                         return (IPPROTO_DONE);
 534                 }
 535                 ip = mtod(m, struct ip *);
 536                 icp = (struct icmp *)(ip + 1);
 537                 /*
 538                  * The upper layer handler can rely on:
 539                  * - The outer IP header has no options.
 540                  * - The outer IP header, the ICMP header, the inner IP header,
 541                  *   and the first n bytes of the inner payload are contiguous.
 542                  *   n is at least 8, but might be larger based on
 543                  *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
 544                  */
 545                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 546                 if (ctlfunc)
 547                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 548                                    (void *)&icp->icmp_ip);
 549                 break;
 550
 551         badcode:
 552                 ICMPSTAT_INC(icps_badcode);
 553                 break;
 554
 555         case ICMP_ECHO:
 556                 if (!V_icmpbmcastecho
 557                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 558                         ICMPSTAT_INC(icps_bmcastecho);
 559                         break;
 560                 }
 561                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 562                         goto freeit;
 563                 icp->icmp_type = ICMP_ECHOREPLY;
 564                 goto reflect;
 565
 566         case ICMP_TSTAMP:
 567                 if (V_icmptstamprepl == 0)
 568                         break;
 569                 if (!V_icmpbmcastecho
 570                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 571                         ICMPSTAT_INC(icps_bmcasttstamp);
 572                         break;
 573                 }
 574                 if (icmplen < ICMP_TSLEN) {
 575                         ICMPSTAT_INC(icps_badlen);
 576                         break;
 577                 }
 578                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 579                         goto freeit;
 580                 icp->icmp_type = ICMP_TSTAMPREPLY;
 581                 icp->icmp_rtime = iptime();
 582                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 583                 goto reflect;
 584
 585         case ICMP_MASKREQ:
 586                 if (V_icmpmaskrepl == 0)
 587                         break;
 588                 /*
 589                  * We are not able to respond with all ones broadcast
 590                  * unless we receive it over a point-to-point interface.
 591                  */
 592                 if (icmplen < ICMP_MASKLEN)
 593                         break;
 594                 switch (ip->ip_dst.s_addr) {
 595
 596                 case INADDR_BROADCAST:
 597                 case INADDR_ANY:
 598                         icmpdst.sin_addr = ip->ip_src;
 599                         break;
 600
 601                 default:
 602                         icmpdst.sin_addr = ip->ip_dst;
 603                 }
 604                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 605                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 606                 if (ia == NULL)
 607                         break;
 608                 if (ia->ia_ifp == NULL) {
 609                         ifa_free(&ia->ia_ifa);
 610                         break;
 611                 }
 612                 icp->icmp_type = ICMP_MASKREPLY;
 613                 if (V_icmpmaskfake == 0)
 614                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 615                 else
 616                         icp->icmp_mask = V_icmpmaskfake;
 617                 if (ip->ip_src.s_addr == 0) {
 618                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 619                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 620                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 621                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 622                 }
 623                 ifa_free(&ia->ia_ifa);
 624 reflect:
 625                 ICMPSTAT_INC(icps_reflect);
 626                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 627                 icmp_reflect(m);
 628                 return (IPPROTO_DONE);
 629
 630         case ICMP_REDIRECT:
 631                 if (V_log_redirect) {
 632                         u_long src, dst, gw;
 633
 634                         src = ntohl(ip->ip_src.s_addr);
 635                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 636                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 637                         printf("icmp redirect from %d.%d.%d.%d: "
 638                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 639                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 640                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 641                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 642                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 643                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 644                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 645                 }
 646                 /*
 647                  * RFC1812 says we must ignore ICMP redirects if we
 648                  * are acting as router.
 649                  */
 650                 if (V_drop_redirect || V_ipforwarding)
 651                         break;
 652                 if (code > 3)
 653                         goto badcode;
 654                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 655                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 656                         ICMPSTAT_INC(icps_badlen);
 657                         break;
 658                 }
 659                 /*
 660                  * Short circuit routing redirects to force
 661                  * immediate change in the kernel's routing
 662                  * tables.  The message is also handed to anyone
 663                  * listening on a raw socket (e.g. the routing
 664                  * daemon for use in updating its tables).
 665                  */
 666                 icmpgw.sin_addr = ip->ip_src;
 667                 icmpdst.sin_addr = icp->icmp_gwaddr;
 668 #ifdef  ICMPPRINTFS
 669                 if (icmpprintfs) {
 670                         char dstbuf[INET_ADDRSTRLEN];
 671                         char gwbuf[INET_ADDRSTRLEN];
 672
 673                         printf("redirect dst %s to %s\n",
 674                                inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
 675                                inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
 676                 }
 677 #endif
 678                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 679                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 680                         in_rtredirect((struct sockaddr *)&icmpsrc,
 681                           (struct sockaddr *)&icmpdst,
 682                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 683                           (struct sockaddr *)&icmpgw, fibnum);
 684                 }
 685                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 686                 break;
 687
 688         /*
 689          * No kernel processing for the following;
 690          * just fall through to send to raw listener.
 691          */
 692         case ICMP_ECHOREPLY:
 693         case ICMP_ROUTERADVERT:
 694         case ICMP_ROUTERSOLICIT:
 695         case ICMP_TSTAMPREPLY:
 696         case ICMP_IREQREPLY:
 697         case ICMP_MASKREPLY:
 698         case ICMP_SOURCEQUENCH:
 699         default:
 700                 break;
 701         }
 702
 703 raw:
 704         *mp = m;
 705         rip_input(mp, offp, proto);
 706         return (IPPROTO_DONE);
 707
 708 freeit:
 709         m_freem(m);
 710         return (IPPROTO_DONE);
 711 }
 712
 713 /*
 714  * Reflect the ip packet back to the source
 715  */
 716 static void
 717 icmp_reflect(struct mbuf *m)
 718 {
 719         struct rm_priotracker in_ifa_tracker;
 720         struct ip *ip = mtod(m, struct ip *);
 721         struct ifaddr *ifa;
 722         struct ifnet *ifp;
 723         struct in_ifaddr *ia;
 724         struct in_addr t;
 725         struct nhop4_extended nh_ext;
 726         struct mbuf *opts = NULL;
 727         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 728
 729         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 730             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 731             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 732                 m_freem(m);     /* Bad return address */
 733                 ICMPSTAT_INC(icps_badaddr);
 734                 goto done;      /* Ip_output() will check for broadcast */
 735         }
 736
 737         t = ip->ip_dst;
 738         ip->ip_dst = ip->ip_src;
 739
 740         /*
 741          * Source selection for ICMP replies:
 742          *
 743          * If the incoming packet was addressed directly to one of our
 744          * own addresses, use dst as the src for the reply.
 745          */
 746         IN_IFADDR_RLOCK(&in_ifa_tracker);
 747         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 748                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 749                         t = IA_SIN(ia)->sin_addr;
 750                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 751                         goto match;
 752                 }
 753         }
 754         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 755
 756         /*
 757          * If the incoming packet was addressed to one of our broadcast
 758          * addresses, use the first non-broadcast address which corresponds
 759          * to the incoming interface.
 760          */
 761         ifp = m->m_pkthdr.rcvif;
 762         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 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                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 769                             t.s_addr) {
 770                                 t = IA_SIN(ia)->sin_addr;
 771                                 IF_ADDR_RUNLOCK(ifp);
 772                                 goto match;
 773                         }
 774                 }
 775                 IF_ADDR_RUNLOCK(ifp);
 776         }
 777         /*
 778          * If the packet was transiting through us, use the address of
 779          * the interface the packet came through in.  If that interface
 780          * doesn't have a suitable IP address, the normal selection
 781          * criteria apply.
 782          */
 783         if (V_icmp_rfi && ifp != NULL) {
 784                 IF_ADDR_RLOCK(ifp);
 785                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 786                         if (ifa->ifa_addr->sa_family != AF_INET)
 787                                 continue;
 788                         ia = ifatoia(ifa);
 789                         t = IA_SIN(ia)->sin_addr;
 790                         IF_ADDR_RUNLOCK(ifp);
 791                         goto match;
 792                 }
 793                 IF_ADDR_RUNLOCK(ifp);
 794         }
 795         /*
 796          * If the incoming packet was not addressed directly to us, use
 797          * designated interface for icmp replies specified by sysctl
 798          * net.inet.icmp.reply_src (default not set). Otherwise continue
 799          * with normal source selection.
 800          */
 801         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 802                 IF_ADDR_RLOCK(ifp);
 803                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 804                         if (ifa->ifa_addr->sa_family != AF_INET)
 805                                 continue;
 806                         ia = ifatoia(ifa);
 807                         t = IA_SIN(ia)->sin_addr;
 808                         IF_ADDR_RUNLOCK(ifp);
 809                         goto match;
 810                 }
 811                 IF_ADDR_RUNLOCK(ifp);
 812         }
 813         /*
 814          * If the packet was transiting through us, use the address of
 815          * the interface that is the closest to the packet source.
 816          * When we don't have a route back to the packet source, stop here
 817          * and drop the packet.
 818          */
 819         if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
 820                 m_freem(m);
 821                 ICMPSTAT_INC(icps_noroute);
 822                 goto done;
 823         }
 824         t = nh_ext.nh_src;
 825 match:
 826 #ifdef MAC
 827         mac_netinet_icmp_replyinplace(m);
 828 #endif
 829         ip->ip_src = t;
 830         ip->ip_ttl = V_ip_defttl;
 831
 832         if (optlen > 0) {
 833                 u_char *cp;
 834                 int opt, cnt;
 835                 u_int len;
 836
 837                 /*
 838                  * Retrieve any source routing from the incoming packet;
 839                  * add on any record-route or timestamp options.
 840                  */
 841                 cp = (u_char *) (ip + 1);
 842                 if ((opts = ip_srcroute(m)) == NULL &&
 843                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 844                         opts->m_len = sizeof(struct in_addr);
 845                         mtod(opts, struct in_addr *)->s_addr = 0;
 846                 }
 847                 if (opts) {
 848 #ifdef ICMPPRINTFS
 849                     if (icmpprintfs)
 850                             printf("icmp_reflect optlen %d rt %d => ",
 851                                 optlen, opts->m_len);
 852 #endif
 853                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 854                             opt = cp[IPOPT_OPTVAL];
 855                             if (opt == IPOPT_EOL)
 856                                     break;
 857                             if (opt == IPOPT_NOP)
 858                                     len = 1;
 859                             else {
 860                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 861                                             break;
 862                                     len = cp[IPOPT_OLEN];
 863                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 864                                         len > cnt)
 865                                             break;
 866                             }
 867                             /*
 868                              * Should check for overflow, but it "can't happen"
 869                              */
 870                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 871                                 opt == IPOPT_SECURITY) {
 872                                     bcopy((caddr_t)cp,
 873                                         mtod(opts, caddr_t) + opts->m_len, len);
 874                                     opts->m_len += len;
 875                             }
 876                     }
 877                     /* Terminate & pad, if necessary */
 878                     cnt = opts->m_len % 4;
 879                     if (cnt) {
 880                             for (; cnt < 4; cnt++) {
 881                                     *(mtod(opts, caddr_t) + opts->m_len) =
 882                                         IPOPT_EOL;
 883                                     opts->m_len++;
 884                             }
 885                     }
 886 #ifdef ICMPPRINTFS
 887                     if (icmpprintfs)
 888                             printf("%d\n", opts->m_len);
 889 #endif
 890                 }
 891                 ip_stripoptions(m);
 892         }
 893         m_tag_delete_nonpersistent(m);
 894         m->m_flags &= ~(M_BCAST|M_MCAST);
 895         icmp_send(m, opts);
 896 done:
 897         if (opts)
 898                 (void)m_free(opts);
 899 }
 900
 901 /*
 902  * Send an icmp packet back to the ip level,
 903  * after supplying a checksum.
 904  */
 905 static void
 906 icmp_send(struct mbuf *m, struct mbuf *opts)
 907 {
 908         struct ip *ip = mtod(m, struct ip *);
 909         int hlen;
 910         struct icmp *icp;
 911
 912         hlen = ip->ip_hl << 2;
 913         m->m_data += hlen;
 914         m->m_len -= hlen;
 915         icp = mtod(m, struct icmp *);
 916         icp->icmp_cksum = 0;
 917         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 918         m->m_data -= hlen;
 919         m->m_len += hlen;
 920         m->m_pkthdr.rcvif = (struct ifnet *)0;
 921 #ifdef ICMPPRINTFS
 922         if (icmpprintfs) {
 923                 char dstbuf[INET_ADDRSTRLEN];
 924                 char srcbuf[INET_ADDRSTRLEN];
 925
 926                 printf("icmp_send dst %s src %s\n",
 927                     inet_ntoa_r(ip->ip_dst, dstbuf),
 928                     inet_ntoa_r(ip->ip_src, srcbuf));
 929         }
 930 #endif
 931         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 932 }
 933
 934 /*
 935  * Return milliseconds since 00:00 UTC in network format.
 936  */
 937 uint32_t
 938 iptime(void)
 939 {
 940         struct timeval atv;
 941         u_long t;
 942
 943         getmicrotime(&atv);
 944         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 945         return (htonl(t));
 946 }
 947
 948 /*
 949  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 950  * given current value MTU.  If DIR is less than zero, a larger plateau
 951  * is returned; otherwise, a smaller value is returned.
 952  */
 953 int
 954 ip_next_mtu(int mtu, int dir)
 955 {
 956         static int mtutab[] = {
 957                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 958                 296, 68, 0
 959         };
 960         int i, size;
 961
 962         size = (sizeof mtutab) / (sizeof mtutab[0]);
 963         if (dir >= 0) {
 964                 for (i = 0; i < size; i++)
 965                         if (mtu > mtutab[i])
 966                                 return mtutab[i];
 967         } else {
 968                 for (i = size - 1; i >= 0; i--)
 969                         if (mtu < mtutab[i])
 970                                 return mtutab[i];
 971                 if (mtu == mtutab[0])
 972                         return mtutab[0];
 973         }
 974         return 0;
 975 }
 976 #endif /* INET */
 977
 978
 979 /*
 980  * badport_bandlim() - check for ICMP bandwidth limit
 981  *
 982  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 983  *      hit our bandwidth limit and it is not ok.
 984  *
 985  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 986  *
 987  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 988  *      values.  We may eventually remove this separation (and simplify the
 989  *      code further).
 990  *
 991  *      Note that the printing of the error message is delayed so we can
 992  *      properly print the icmp error rate that the system was trying to do
 993  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 994  *      the 'final' error, but it doesn't make sense to solve the printing
 995  *      delay with more complex code.
 996  */
 997
 998 int
 999 badport_bandlim(int which)
1000 {
1001
1002 #define N(a)    (sizeof (a) / sizeof (a[0]))
1003         static struct rate {
1004                 const char      *type;
1005                 struct timeval  lasttime;
1006                 int             curpps;
1007         } rates[BANDLIM_MAX+1] = {
1008                 { "icmp unreach response" },
1009                 { "icmp ping response" },
1010                 { "icmp tstamp response" },
1011                 { "closed port RST response" },
1012                 { "open port RST response" },
1013                 { "icmp6 unreach response" },
1014                 { "sctp ootb response" }
1015         };
1016
1017         /*
1018          * Return ok status if feature disabled or argument out of range.
1019          */
1020         if (V_icmplim > 0 && (u_int) which < N(rates)) {
1021                 struct rate *r = &rates[which];
1022                 int opps = r->curpps;
1023
1024                 if (!ppsratecheck(&r->lasttime, &r->curpps, V_icmplim))
1025                         return -1;      /* discard packet */
1026                 /*
1027                  * If we've dropped below the threshold after having
1028                  * rate-limited traffic print the message.  This preserves
1029                  * the previous behaviour at the expense of added complexity.
1030                  */
1031                 if (V_icmplim_output && opps > V_icmplim)
1032                         log(LOG_NOTICE, "Limiting %s from %d to %d packets/sec\n",
1033                                 r->type, opps, V_icmplim);
1034         }
1035         return 0;                       /* okay to send packet */
1036 #undef N
1037 }