sys/netinet/ip_icmp.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1993
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      @(#)ip_icmp.c   8.2 (Berkeley) 1/4/94
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include "opt_inet.h"
  38
  39 #include <sys/param.h>
  40 #include <sys/systm.h>
  41 #include <sys/mbuf.h>
  42 #include <sys/protosw.h>
  43 #include <sys/socket.h>
  44 #include <sys/time.h>
  45 #include <sys/kernel.h>
  46 #include <sys/lock.h>
  47 #include <sys/rmlock.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/syslog.h>
  50
  51 #include <net/if.h>
  52 #include <net/if_var.h>
  53 #include <net/if_types.h>
  54 #include <net/route.h>
  55 #include <net/vnet.h>
  56
  57 #include <netinet/in.h>
  58 #include <netinet/in_fib.h>
  59 #include <netinet/in_pcb.h>
  60 #include <netinet/in_systm.h>
  61 #include <netinet/in_var.h>
  62 #include <netinet/ip.h>
  63 #include <netinet/ip_icmp.h>
  64 #include <netinet/ip_var.h>
  65 #include <netinet/ip_options.h>
  66 #include <netinet/sctp.h>
  67 #include <netinet/tcp.h>
  68 #include <netinet/tcp_var.h>
  69 #include <netinet/tcpip.h>
  70 #include <netinet/icmp_var.h>
  71
  72
  73 #ifdef INET
  74
  75 #include <machine/in_cksum.h>
  76
  77 #include <security/mac/mac_framework.h>
  78 #endif /* INET */
  79
  80 /*
  81  * ICMP routines: error generation, receive packet processing, and
  82  * routines to turnaround packets back to the originator, and
  83  * host table maintenance routines.
  84  */
  85 VNET_DEFINE_STATIC(int, icmplim) = 200;
  86 #define V_icmplim                       VNET(icmplim)
  87 SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | CTLFLAG_RW,
  88         &VNET_NAME(icmplim), 0,
  89         "Maximum number of ICMP responses per second");
  90
  91 VNET_DEFINE_STATIC(int, icmplim_output) = 1;
  92 #define V_icmplim_output                VNET(icmplim_output)
  93 SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW,
  94         &VNET_NAME(icmplim_output), 0,
  95         "Enable logging of ICMP response rate limiting");
  96
  97 #ifdef INET
  98 VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat);
  99 VNET_PCPUSTAT_SYSINIT(icmpstat);
 100 SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat,
 101     icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)");
 102
 103 #ifdef VIMAGE
 104 VNET_PCPUSTAT_SYSUNINIT(icmpstat);
 105 #endif /* VIMAGE */
 106
 107 VNET_DEFINE_STATIC(int, icmpmaskrepl) = 0;
 108 #define V_icmpmaskrepl                  VNET(icmpmaskrepl)
 109 SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW,
 110         &VNET_NAME(icmpmaskrepl), 0,
 111         "Reply to ICMP Address Mask Request packets");
 112
 113 VNET_DEFINE_STATIC(u_int, icmpmaskfake) = 0;
 114 #define V_icmpmaskfake                  VNET(icmpmaskfake)
 115 SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW,
 116         &VNET_NAME(icmpmaskfake), 0,
 117         "Fake reply to ICMP Address Mask Request packets");
 118
 119 VNET_DEFINE(int, drop_redirect) = 0;
 120 #define V_drop_redirect                 VNET(drop_redirect)
 121 SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW,
 122         &VNET_NAME(drop_redirect), 0,
 123         "Ignore ICMP redirects");
 124
 125 VNET_DEFINE_STATIC(int, log_redirect) = 0;
 126 #define V_log_redirect                  VNET(log_redirect)
 127 SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW,
 128         &VNET_NAME(log_redirect), 0,
 129         "Log ICMP redirects to the console");
 130
 131 VNET_DEFINE_STATIC(char, reply_src[IFNAMSIZ]);
 132 #define V_reply_src                     VNET(reply_src)
 133 SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW,
 134         &VNET_NAME(reply_src), IFNAMSIZ,
 135         "ICMP reply source for non-local packets");
 136
 137 VNET_DEFINE_STATIC(int, icmp_rfi) = 0;
 138 #define V_icmp_rfi                      VNET(icmp_rfi)
 139 SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW,
 140         &VNET_NAME(icmp_rfi), 0,
 141         "ICMP reply from incoming interface for non-local packets");
 142 /* Router requirements RFC 1812 section 4.3.2.3 requires 576 - 28. */
 143 VNET_DEFINE_STATIC(int, icmp_quotelen) = 548;
 144 #define V_icmp_quotelen                 VNET(icmp_quotelen)
 145 SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW,
 146         &VNET_NAME(icmp_quotelen), 0,
 147         "Number of bytes from original packet to quote in ICMP reply");
 148
 149 VNET_DEFINE_STATIC(int, icmpbmcastecho) = 0;
 150 #define V_icmpbmcastecho                VNET(icmpbmcastecho)
 151 SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW,
 152         &VNET_NAME(icmpbmcastecho), 0,
 153         "Reply to multicast ICMP Echo Request and Timestamp packets");
 154
 155 VNET_DEFINE_STATIC(int, icmptstamprepl) = 1;
 156 #define V_icmptstamprepl                VNET(icmptstamprepl)
 157 SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_VNET | CTLFLAG_RW,
 158         &VNET_NAME(icmptstamprepl), 0,
 159         "Respond to ICMP Timestamp packets");
 160
 161 VNET_DEFINE_STATIC(int, error_keeptags) = 0;
 162 #define V_error_keeptags                VNET(error_keeptags)
 163 SYSCTL_INT(_net_inet_icmp, OID_AUTO, error_keeptags, CTLFLAG_VNET | CTLFLAG_RW,
 164         &VNET_NAME(error_keeptags), 0,
 165         "ICMP error response keeps copy of mbuf_tags of original packet");
 166
 167 #ifdef ICMPPRINTFS
 168 int     icmpprintfs = 0;
 169 #endif
 170
 171 static void     icmp_reflect(struct mbuf *);
 172 static void     icmp_send(struct mbuf *, struct mbuf *);
 173
 174 extern  struct protosw inetsw[];
 175
 176 /*
 177  * Kernel module interface for updating icmpstat.  The argument is an index
 178  * into icmpstat treated as an array of u_long.  While this encodes the
 179  * general layout of icmpstat into the caller, it doesn't encode its
 180  * location, so that future changes to add, for example, per-CPU stats
 181  * support won't cause binary compatibility problems for kernel modules.
 182  */
 183 void
 184 kmod_icmpstat_inc(int statnum)
 185 {
 186
 187         counter_u64_add(VNET(icmpstat)[statnum], 1);
 188 }
 189
 190 /*
 191  * Generate an error packet of type error
 192  * in response to bad packet ip.
 193  */
 194 void
 195 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 196 {
 197         struct ip *oip, *nip;
 198         struct icmp *icp;
 199         struct mbuf *m;
 200         unsigned icmplen, icmpelen, nlen, oiphlen;
 201
 202         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
 203             __func__));
 204
 205         if (type != ICMP_REDIRECT)
 206                 ICMPSTAT_INC(icps_error);
 207         /*
 208          * Don't send error:
 209          *  if the original packet was encrypted.
 210          *  if not the first fragment of message.
 211          *  in response to a multicast or broadcast packet.
 212          *  if the old packet protocol was an ICMP error message.
 213          */
 214         if (n->m_flags & M_DECRYPTED)
 215                 goto freeit;
 216         if (n->m_flags & (M_BCAST|M_MCAST))
 217                 goto freeit;
 218
 219         /* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
 220         if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
 221                 goto freeit;
 222         oip = mtod(n, struct ip *);
 223         oiphlen = oip->ip_hl << 2;
 224         if (n->m_len < oiphlen + ICMP_MINLEN)
 225                 goto freeit;
 226 #ifdef ICMPPRINTFS
 227         if (icmpprintfs)
 228                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 229 #endif
 230         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 231                 goto freeit;
 232         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 233             !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
 234                 oiphlen))->icmp_type)) {
 235                 ICMPSTAT_INC(icps_oldicmp);
 236                 goto freeit;
 237         }
 238         /*
 239          * Calculate length to quote from original packet and
 240          * prevent the ICMP mbuf from overflowing.
 241          * Unfortunately this is non-trivial since ip_forward()
 242          * sends us truncated packets.
 243          */
 244         nlen = m_length(n, NULL);
 245         if (oip->ip_p == IPPROTO_TCP) {
 246                 struct tcphdr *th;
 247                 int tcphlen;
 248
 249                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 250                     n->m_next == NULL)
 251                         goto stdreply;
 252                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 253                     (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
 254                         goto freeit;
 255                 oip = mtod(n, struct ip *);
 256                 th = mtodo(n, oiphlen);
 257                 tcphlen = th->th_off << 2;
 258                 if (tcphlen < sizeof(struct tcphdr))
 259                         goto freeit;
 260                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 261                         goto freeit;
 262                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 263                         goto stdreply;
 264                 if (n->m_len < oiphlen + tcphlen &&
 265                     (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
 266                         goto freeit;
 267                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 268                     ntohs(oip->ip_len) - oiphlen));
 269         } else if (oip->ip_p == IPPROTO_SCTP) {
 270                 struct sctphdr *sh;
 271                 struct sctp_chunkhdr *ch;
 272
 273                 if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
 274                         goto stdreply;
 275                 if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
 276                     n->m_next == NULL)
 277                         goto stdreply;
 278                 if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
 279                     (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
 280                         goto freeit;
 281                 oip = mtod(n, struct ip *);
 282                 icmpelen = max(sizeof(struct sctphdr),
 283                     min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 284                 sh = mtodo(n, oiphlen);
 285                 if (ntohl(sh->v_tag) == 0 &&
 286                     ntohs(oip->ip_len) >= oiphlen +
 287                     sizeof(struct sctphdr) + 8 &&
 288                     (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
 289                      n->m_next != NULL)) {
 290                         if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
 291                             (n = m_pullup(n, oiphlen +
 292                             sizeof(struct sctphdr) + 8)) == NULL)
 293                                 goto freeit;
 294                         oip = mtod(n, struct ip *);
 295                         sh = mtodo(n, oiphlen);
 296                         ch = (struct sctp_chunkhdr *)(sh + 1);
 297                         if (ch->chunk_type == SCTP_INITIATION) {
 298                                 icmpelen = max(sizeof(struct sctphdr) + 8,
 299                                     min(V_icmp_quotelen, ntohs(oip->ip_len) -
 300                                     oiphlen));
 301                         }
 302                 }
 303         } else
 304 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
 305                     oiphlen));
 306
 307         icmplen = min(oiphlen + icmpelen, nlen);
 308         if (icmplen < sizeof(struct ip))
 309                 goto freeit;
 310
 311         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 312                 m = m_gethdr(M_NOWAIT, MT_DATA);
 313         else
 314                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 315         if (m == NULL)
 316                 goto freeit;
 317 #ifdef MAC
 318         mac_netinet_icmp_reply(n, m);
 319 #endif
 320         icmplen = min(icmplen, M_TRAILINGSPACE(m) -
 321             sizeof(struct ip) - ICMP_MINLEN);
 322         m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
 323         m->m_data += sizeof(struct ip);
 324         m->m_len = ICMP_MINLEN + icmplen;
 325
 326         /* XXX MRT  make the outgoing packet use the same FIB
 327          * that was associated with the incoming packet
 328          */
 329         M_SETFIB(m, M_GETFIB(n));
 330         icp = mtod(m, struct icmp *);
 331         ICMPSTAT_INC(icps_outhist[type]);
 332         icp->icmp_type = type;
 333         if (type == ICMP_REDIRECT)
 334                 icp->icmp_gwaddr.s_addr = dest;
 335         else {
 336                 icp->icmp_void = 0;
 337                 /*
 338                  * The following assignments assume an overlay with the
 339                  * just zeroed icmp_void field.
 340                  */
 341                 if (type == ICMP_PARAMPROB) {
 342                         icp->icmp_pptr = code;
 343                         code = 0;
 344                 } else if (type == ICMP_UNREACH &&
 345                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 346                         icp->icmp_nextmtu = htons(mtu);
 347                 }
 348         }
 349         icp->icmp_code = code;
 350
 351         /*
 352          * Copy the quotation into ICMP message and
 353          * convert quoted IP header back to network representation.
 354          */
 355         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 356         nip = &icp->icmp_ip;
 357
 358         /*
 359          * Set up ICMP message mbuf and copy old IP header (without options
 360          * in front of ICMP message.
 361          * If the original mbuf was meant to bypass the firewall, the error
 362          * reply should bypass as well.
 363          */
 364         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 365         KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
 366             ("insufficient space for ip header"));
 367         m->m_data -= sizeof(struct ip);
 368         m->m_len += sizeof(struct ip);
 369         m->m_pkthdr.len = m->m_len;
 370         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 371         nip = mtod(m, struct ip *);
 372         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 373         nip->ip_len = htons(m->m_len);
 374         nip->ip_v = IPVERSION;
 375         nip->ip_hl = 5;
 376         nip->ip_p = IPPROTO_ICMP;
 377         nip->ip_tos = 0;
 378         nip->ip_off = 0;
 379
 380         if (V_error_keeptags)
 381                 m_tag_copy_chain(m, n, M_NOWAIT);
 382
 383         icmp_reflect(m);
 384
 385 freeit:
 386         m_freem(n);
 387 }
 388
 389 /*
 390  * Process a received ICMP message.
 391  */
 392 int
 393 icmp_input(struct mbuf **mp, int *offp, int proto)
 394 {
 395         struct icmp *icp;
 396         struct in_ifaddr *ia;
 397         struct mbuf *m = *mp;
 398         struct ip *ip = mtod(m, struct ip *);
 399         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 400         int hlen = *offp;
 401         int icmplen = ntohs(ip->ip_len) - *offp;
 402         int i, code;
 403         void (*ctlfunc)(int, struct sockaddr *, void *);
 404         int fibnum;
 405
 406         *mp = NULL;
 407
 408         /*
 409          * Locate icmp structure in mbuf, and check
 410          * that not corrupted and of at least minimum length.
 411          */
 412 #ifdef ICMPPRINTFS
 413         if (icmpprintfs) {
 414                 char srcbuf[INET_ADDRSTRLEN];
 415                 char dstbuf[INET_ADDRSTRLEN];
 416
 417                 printf("icmp_input from %s to %s, len %d\n",
 418                     inet_ntoa_r(ip->ip_src, srcbuf),
 419                     inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
 420         }
 421 #endif
 422         NET_EPOCH_ENTER();
 423         if (icmplen < ICMP_MINLEN) {
 424                 ICMPSTAT_INC(icps_tooshort);
 425                 goto freeit;
 426         }
 427         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 428         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 429                 ICMPSTAT_INC(icps_tooshort);
 430                 NET_EPOCH_EXIT();
 431                 return (IPPROTO_DONE);
 432         }
 433         ip = mtod(m, struct ip *);
 434         m->m_len -= hlen;
 435         m->m_data += hlen;
 436         icp = mtod(m, struct icmp *);
 437         if (in_cksum(m, icmplen)) {
 438                 ICMPSTAT_INC(icps_checksum);
 439                 goto freeit;
 440         }
 441         m->m_len += hlen;
 442         m->m_data -= hlen;
 443
 444 #ifdef ICMPPRINTFS
 445         if (icmpprintfs)
 446                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 447                     icp->icmp_code);
 448 #endif
 449
 450         /*
 451          * Message type specific processing.
 452          */
 453         if (icp->icmp_type > ICMP_MAXTYPE)
 454                 goto raw;
 455
 456         /* Initialize */
 457         bzero(&icmpsrc, sizeof(icmpsrc));
 458         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 459         icmpsrc.sin_family = AF_INET;
 460         bzero(&icmpdst, sizeof(icmpdst));
 461         icmpdst.sin_len = sizeof(struct sockaddr_in);
 462         icmpdst.sin_family = AF_INET;
 463         bzero(&icmpgw, sizeof(icmpgw));
 464         icmpgw.sin_len = sizeof(struct sockaddr_in);
 465         icmpgw.sin_family = AF_INET;
 466
 467         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 468         code = icp->icmp_code;
 469         switch (icp->icmp_type) {
 470
 471         case ICMP_UNREACH:
 472                 switch (code) {
 473                         case ICMP_UNREACH_NET:
 474                         case ICMP_UNREACH_HOST:
 475                         case ICMP_UNREACH_SRCFAIL:
 476                         case ICMP_UNREACH_NET_UNKNOWN:
 477                         case ICMP_UNREACH_HOST_UNKNOWN:
 478                         case ICMP_UNREACH_ISOLATED:
 479                         case ICMP_UNREACH_TOSNET:
 480                         case ICMP_UNREACH_TOSHOST:
 481                         case ICMP_UNREACH_HOST_PRECEDENCE:
 482                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 483                                 code = PRC_UNREACH_NET;
 484                                 break;
 485
 486                         case ICMP_UNREACH_NEEDFRAG:
 487                                 code = PRC_MSGSIZE;
 488                                 break;
 489
 490                         /*
 491                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 492                          * Treat subcodes 2,3 as immediate RST
 493                          */
 494                         case ICMP_UNREACH_PROTOCOL:
 495                                 code = PRC_UNREACH_PROTOCOL;
 496                                 break;
 497                         case ICMP_UNREACH_PORT:
 498                                 code = PRC_UNREACH_PORT;
 499                                 break;
 500
 501                         case ICMP_UNREACH_NET_PROHIB:
 502                         case ICMP_UNREACH_HOST_PROHIB:
 503                         case ICMP_UNREACH_FILTER_PROHIB:
 504                                 code = PRC_UNREACH_ADMIN_PROHIB;
 505                                 break;
 506
 507                         default:
 508                                 goto badcode;
 509                 }
 510                 goto deliver;
 511
 512         case ICMP_TIMXCEED:
 513                 if (code > 1)
 514                         goto badcode;
 515                 code += PRC_TIMXCEED_INTRANS;
 516                 goto deliver;
 517
 518         case ICMP_PARAMPROB:
 519                 if (code > 1)
 520                         goto badcode;
 521                 code = PRC_PARAMPROB;
 522         deliver:
 523                 /*
 524                  * Problem with datagram; advise higher level routines.
 525                  */
 526                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 527                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 528                         ICMPSTAT_INC(icps_badlen);
 529                         goto freeit;
 530                 }
 531                 /* Discard ICMP's in response to multicast packets */
 532                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 533                         goto badcode;
 534 #ifdef ICMPPRINTFS
 535                 if (icmpprintfs)
 536                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 537 #endif
 538                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 539                 /*
 540                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 541                  * notification to TCP layer.
 542                  */
 543                 i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
 544                 ip_stripoptions(m);
 545                 if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
 546                         /* This should actually not happen */
 547                         ICMPSTAT_INC(icps_tooshort);
 548                         NET_EPOCH_EXIT();
 549                         return (IPPROTO_DONE);
 550                 }
 551                 ip = mtod(m, struct ip *);
 552                 icp = (struct icmp *)(ip + 1);
 553                 /*
 554                  * The upper layer handler can rely on:
 555                  * - The outer IP header has no options.
 556                  * - The outer IP header, the ICMP header, the inner IP header,
 557                  *   and the first n bytes of the inner payload are contiguous.
 558                  *   n is at least 8, but might be larger based on
 559                  *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
 560                  */
 561                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 562                 if (ctlfunc)
 563                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 564                                    (void *)&icp->icmp_ip);
 565                 break;
 566
 567         badcode:
 568                 ICMPSTAT_INC(icps_badcode);
 569                 break;
 570
 571         case ICMP_ECHO:
 572                 if (!V_icmpbmcastecho
 573                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 574                         ICMPSTAT_INC(icps_bmcastecho);
 575                         break;
 576                 }
 577                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 578                         goto freeit;
 579                 icp->icmp_type = ICMP_ECHOREPLY;
 580                 goto reflect;
 581
 582         case ICMP_TSTAMP:
 583                 if (V_icmptstamprepl == 0)
 584                         break;
 585                 if (!V_icmpbmcastecho
 586                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 587                         ICMPSTAT_INC(icps_bmcasttstamp);
 588                         break;
 589                 }
 590                 if (icmplen < ICMP_TSLEN) {
 591                         ICMPSTAT_INC(icps_badlen);
 592                         break;
 593                 }
 594                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 595                         goto freeit;
 596                 icp->icmp_type = ICMP_TSTAMPREPLY;
 597                 icp->icmp_rtime = iptime();
 598                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 599                 goto reflect;
 600
 601         case ICMP_MASKREQ:
 602                 if (V_icmpmaskrepl == 0)
 603                         break;
 604                 /*
 605                  * We are not able to respond with all ones broadcast
 606                  * unless we receive it over a point-to-point interface.
 607                  */
 608                 if (icmplen < ICMP_MASKLEN)
 609                         break;
 610                 switch (ip->ip_dst.s_addr) {
 611
 612                 case INADDR_BROADCAST:
 613                 case INADDR_ANY:
 614                         icmpdst.sin_addr = ip->ip_src;
 615                         break;
 616
 617                 default:
 618                         icmpdst.sin_addr = ip->ip_dst;
 619                 }
 620                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 621                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 622                 if (ia == NULL)
 623                         break;
 624                 if (ia->ia_ifp == NULL)
 625                         break;
 626                 icp->icmp_type = ICMP_MASKREPLY;
 627                 if (V_icmpmaskfake == 0)
 628                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 629                 else
 630                         icp->icmp_mask = V_icmpmaskfake;
 631                 if (ip->ip_src.s_addr == 0) {
 632                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 633                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 634                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 635                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 636                 }
 637 reflect:
 638                 ICMPSTAT_INC(icps_reflect);
 639                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 640                 icmp_reflect(m);
 641                 NET_EPOCH_EXIT();
 642                 return (IPPROTO_DONE);
 643
 644         case ICMP_REDIRECT:
 645                 if (V_log_redirect) {
 646                         u_long src, dst, gw;
 647
 648                         src = ntohl(ip->ip_src.s_addr);
 649                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 650                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 651                         printf("icmp redirect from %d.%d.%d.%d: "
 652                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 653                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 654                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 655                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 656                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 657                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 658                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 659                 }
 660                 /*
 661                  * RFC1812 says we must ignore ICMP redirects if we
 662                  * are acting as router.
 663                  */
 664                 if (V_drop_redirect || V_ipforwarding)
 665                         break;
 666                 if (code > 3)
 667                         goto badcode;
 668                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 669                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 670                         ICMPSTAT_INC(icps_badlen);
 671                         break;
 672                 }
 673                 /*
 674                  * Short circuit routing redirects to force
 675                  * immediate change in the kernel's routing
 676                  * tables.  The message is also handed to anyone
 677                  * listening on a raw socket (e.g. the routing
 678                  * daemon for use in updating its tables).
 679                  */
 680                 icmpgw.sin_addr = ip->ip_src;
 681                 icmpdst.sin_addr = icp->icmp_gwaddr;
 682 #ifdef  ICMPPRINTFS
 683                 if (icmpprintfs) {
 684                         char dstbuf[INET_ADDRSTRLEN];
 685                         char gwbuf[INET_ADDRSTRLEN];
 686
 687                         printf("redirect dst %s to %s\n",
 688                                inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
 689                                inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
 690                 }
 691 #endif
 692                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 693                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 694                         in_rtredirect((struct sockaddr *)&icmpsrc,
 695                           (struct sockaddr *)&icmpdst,
 696                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 697                           (struct sockaddr *)&icmpgw, fibnum);
 698                 }
 699                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 700                 break;
 701
 702         /*
 703          * No kernel processing for the following;
 704          * just fall through to send to raw listener.
 705          */
 706         case ICMP_ECHOREPLY:
 707         case ICMP_ROUTERADVERT:
 708         case ICMP_ROUTERSOLICIT:
 709         case ICMP_TSTAMPREPLY:
 710         case ICMP_IREQREPLY:
 711         case ICMP_MASKREPLY:
 712         case ICMP_SOURCEQUENCH:
 713         default:
 714                 break;
 715         }
 716
 717 raw:
 718         NET_EPOCH_EXIT();
 719         *mp = m;
 720         rip_input(mp, offp, proto);
 721         return (IPPROTO_DONE);
 722
 723 freeit:
 724         NET_EPOCH_EXIT();
 725         m_freem(m);
 726         return (IPPROTO_DONE);
 727 }
 728
 729 /*
 730  * Reflect the ip packet back to the source
 731  */
 732 static void
 733 icmp_reflect(struct mbuf *m)
 734 {
 735         struct rm_priotracker in_ifa_tracker;
 736         struct ip *ip = mtod(m, struct ip *);
 737         struct ifaddr *ifa;
 738         struct ifnet *ifp;
 739         struct in_ifaddr *ia;
 740         struct in_addr t;
 741         struct nhop4_extended nh_ext;
 742         struct mbuf *opts = NULL;
 743         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 744
 745         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 746             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 747             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 748                 m_freem(m);     /* Bad return address */
 749                 ICMPSTAT_INC(icps_badaddr);
 750                 goto done;      /* Ip_output() will check for broadcast */
 751         }
 752
 753         t = ip->ip_dst;
 754         ip->ip_dst = ip->ip_src;
 755
 756         /*
 757          * Source selection for ICMP replies:
 758          *
 759          * If the incoming packet was addressed directly to one of our
 760          * own addresses, use dst as the src for the reply.
 761          */
 762         IN_IFADDR_RLOCK(&in_ifa_tracker);
 763         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 764                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 765                         t = IA_SIN(ia)->sin_addr;
 766                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 767                         goto match;
 768                 }
 769         }
 770         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 771
 772         /*
 773          * If the incoming packet was addressed to one of our broadcast
 774          * addresses, use the first non-broadcast address which corresponds
 775          * to the incoming interface.
 776          */
 777         ifp = m->m_pkthdr.rcvif;
 778         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 779                 IF_ADDR_RLOCK(ifp);
 780                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 781                         if (ifa->ifa_addr->sa_family != AF_INET)
 782                                 continue;
 783                         ia = ifatoia(ifa);
 784                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 785                             t.s_addr) {
 786                                 t = IA_SIN(ia)->sin_addr;
 787                                 IF_ADDR_RUNLOCK(ifp);
 788                                 goto match;
 789                         }
 790                 }
 791                 IF_ADDR_RUNLOCK(ifp);
 792         }
 793         /*
 794          * If the packet was transiting through us, use the address of
 795          * the interface the packet came through in.  If that interface
 796          * doesn't have a suitable IP address, the normal selection
 797          * criteria apply.
 798          */
 799         if (V_icmp_rfi && ifp != NULL) {
 800                 IF_ADDR_RLOCK(ifp);
 801                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 802                         if (ifa->ifa_addr->sa_family != AF_INET)
 803                                 continue;
 804                         ia = ifatoia(ifa);
 805                         t = IA_SIN(ia)->sin_addr;
 806                         IF_ADDR_RUNLOCK(ifp);
 807                         goto match;
 808                 }
 809                 IF_ADDR_RUNLOCK(ifp);
 810         }
 811         /*
 812          * If the incoming packet was not addressed directly to us, use
 813          * designated interface for icmp replies specified by sysctl
 814          * net.inet.icmp.reply_src (default not set). Otherwise continue
 815          * with normal source selection.
 816          */
 817         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 818                 IF_ADDR_RLOCK(ifp);
 819                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 820                         if (ifa->ifa_addr->sa_family != AF_INET)
 821                                 continue;
 822                         ia = ifatoia(ifa);
 823                         t = IA_SIN(ia)->sin_addr;
 824                         IF_ADDR_RUNLOCK(ifp);
 825                         goto match;
 826                 }
 827                 IF_ADDR_RUNLOCK(ifp);
 828         }
 829         /*
 830          * If the packet was transiting through us, use the address of
 831          * the interface that is the closest to the packet source.
 832          * When we don't have a route back to the packet source, stop here
 833          * and drop the packet.
 834          */
 835         if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
 836                 m_freem(m);
 837                 ICMPSTAT_INC(icps_noroute);
 838                 goto done;
 839         }
 840         t = nh_ext.nh_src;
 841 match:
 842 #ifdef MAC
 843         mac_netinet_icmp_replyinplace(m);
 844 #endif
 845         ip->ip_src = t;
 846         ip->ip_ttl = V_ip_defttl;
 847
 848         if (optlen > 0) {
 849                 u_char *cp;
 850                 int opt, cnt;
 851                 u_int len;
 852
 853                 /*
 854                  * Retrieve any source routing from the incoming packet;
 855                  * add on any record-route or timestamp options.
 856                  */
 857                 cp = (u_char *) (ip + 1);
 858                 if ((opts = ip_srcroute(m)) == NULL &&
 859                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 860                         opts->m_len = sizeof(struct in_addr);
 861                         mtod(opts, struct in_addr *)->s_addr = 0;
 862                 }
 863                 if (opts) {
 864 #ifdef ICMPPRINTFS
 865                     if (icmpprintfs)
 866                             printf("icmp_reflect optlen %d rt %d => ",
 867                                 optlen, opts->m_len);
 868 #endif
 869                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 870                             opt = cp[IPOPT_OPTVAL];
 871                             if (opt == IPOPT_EOL)
 872                                     break;
 873                             if (opt == IPOPT_NOP)
 874                                     len = 1;
 875                             else {
 876                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 877                                             break;
 878                                     len = cp[IPOPT_OLEN];
 879                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 880                                         len > cnt)
 881                                             break;
 882                             }
 883                             /*
 884                              * Should check for overflow, but it "can't happen"
 885                              */
 886                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 887                                 opt == IPOPT_SECURITY) {
 888                                     bcopy((caddr_t)cp,
 889                                         mtod(opts, caddr_t) + opts->m_len, len);
 890                                     opts->m_len += len;
 891                             }
 892                     }
 893                     /* Terminate & pad, if necessary */
 894                     cnt = opts->m_len % 4;
 895                     if (cnt) {
 896                             for (; cnt < 4; cnt++) {
 897                                     *(mtod(opts, caddr_t) + opts->m_len) =
 898                                         IPOPT_EOL;
 899                                     opts->m_len++;
 900                             }
 901                     }
 902 #ifdef ICMPPRINTFS
 903                     if (icmpprintfs)
 904                             printf("%d\n", opts->m_len);
 905 #endif
 906                 }
 907                 ip_stripoptions(m);
 908         }
 909         m_tag_delete_nonpersistent(m);
 910         m->m_flags &= ~(M_BCAST|M_MCAST);
 911         icmp_send(m, opts);
 912 done:
 913         if (opts)
 914                 (void)m_free(opts);
 915 }
 916
 917 /*
 918  * Send an icmp packet back to the ip level,
 919  * after supplying a checksum.
 920  */
 921 static void
 922 icmp_send(struct mbuf *m, struct mbuf *opts)
 923 {
 924         struct ip *ip = mtod(m, struct ip *);
 925         int hlen;
 926         struct icmp *icp;
 927
 928         hlen = ip->ip_hl << 2;
 929         m->m_data += hlen;
 930         m->m_len -= hlen;
 931         icp = mtod(m, struct icmp *);
 932         icp->icmp_cksum = 0;
 933         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 934         m->m_data -= hlen;
 935         m->m_len += hlen;
 936         m->m_pkthdr.rcvif = (struct ifnet *)0;
 937 #ifdef ICMPPRINTFS
 938         if (icmpprintfs) {
 939                 char dstbuf[INET_ADDRSTRLEN];
 940                 char srcbuf[INET_ADDRSTRLEN];
 941
 942                 printf("icmp_send dst %s src %s\n",
 943                     inet_ntoa_r(ip->ip_dst, dstbuf),
 944                     inet_ntoa_r(ip->ip_src, srcbuf));
 945         }
 946 #endif
 947         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 948 }
 949
 950 /*
 951  * Return milliseconds since 00:00 UTC in network format.
 952  */
 953 uint32_t
 954 iptime(void)
 955 {
 956         struct timeval atv;
 957         u_long t;
 958
 959         getmicrotime(&atv);
 960         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 961         return (htonl(t));
 962 }
 963
 964 /*
 965  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 966  * given current value MTU.  If DIR is less than zero, a larger plateau
 967  * is returned; otherwise, a smaller value is returned.
 968  */
 969 int
 970 ip_next_mtu(int mtu, int dir)
 971 {
 972         static int mtutab[] = {
 973                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 974                 296, 68, 0
 975         };
 976         int i, size;
 977
 978         size = (sizeof mtutab) / (sizeof mtutab[0]);
 979         if (dir >= 0) {
 980                 for (i = 0; i < size; i++)
 981                         if (mtu > mtutab[i])
 982                                 return mtutab[i];
 983         } else {
 984                 for (i = size - 1; i >= 0; i--)
 985                         if (mtu < mtutab[i])
 986                                 return mtutab[i];
 987                 if (mtu == mtutab[0])
 988                         return mtutab[0];
 989         }
 990         return 0;
 991 }
 992 #endif /* INET */
 993
 994
 995 /*
 996  * badport_bandlim() - check for ICMP bandwidth limit
 997  *
 998  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 999  *      hit our bandwidth limit and it is not ok.
1000  *
1001  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
1002  *
1003  *      For now we separate the TCP and UDP subsystems w/ different 'which'
1004  *      values.  We may eventually remove this separation (and simplify the
1005  *      code further).
1006  *
1007  *      Note that the printing of the error message is delayed so we can
1008  *      properly print the icmp error rate that the system was trying to do
1009  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1010  *      the 'final' error, but it doesn't make sense to solve the printing
1011  *      delay with more complex code.
1012  */
1013 struct icmp_rate {
1014         const char *descr;
1015         struct counter_rate cr;
1016 };
1017 VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1018         { "icmp unreach response" },
1019         { "icmp ping response" },
1020         { "icmp tstamp response" },
1021         { "closed port RST response" },
1022         { "open port RST response" },
1023         { "icmp6 unreach response" },
1024         { "sctp ootb response" }
1025 };
1026 #define V_icmp_rates    VNET(icmp_rates)
1027
1028 static void
1029 icmp_bandlimit_init(void)
1030 {
1031
1032         for (int i = 0; i < BANDLIM_MAX; i++) {
1033                 V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1034                 V_icmp_rates[i].cr.cr_ticks = ticks;
1035         }
1036 }
1037 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1038     icmp_bandlimit_init, NULL);
1039
1040 static void
1041 icmp_bandlimit_uninit(void)
1042 {
1043
1044         for (int i = 0; i < BANDLIM_MAX; i++)
1045                 counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1046 }
1047 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1048     icmp_bandlimit_uninit, NULL);
1049
1050 int
1051 badport_bandlim(int which)
1052 {
1053         int64_t pps;
1054
1055         if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1056                 return (0);
1057
1058         KASSERT(which >= 0 && which < BANDLIM_MAX,
1059             ("%s: which %d", __func__, which));
1060
1061         pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1062         if (pps == -1)
1063                 return (-1);
1064         if (pps > 0 && V_icmplim_output)
1065                 log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1066                         V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1067         return (0);
1068 }