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_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                 oip = mtod(n, struct ip *);
 268                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 269                     ntohs(oip->ip_len) - oiphlen));
 270         } else if (oip->ip_p == IPPROTO_SCTP) {
 271                 struct sctphdr *sh;
 272                 struct sctp_chunkhdr *ch;
 273
 274                 if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
 275                         goto stdreply;
 276                 if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
 277                     n->m_next == NULL)
 278                         goto stdreply;
 279                 if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
 280                     (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
 281                         goto freeit;
 282                 oip = mtod(n, struct ip *);
 283                 icmpelen = max(sizeof(struct sctphdr),
 284                     min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 285                 sh = mtodo(n, oiphlen);
 286                 if (ntohl(sh->v_tag) == 0 &&
 287                     ntohs(oip->ip_len) >= oiphlen +
 288                     sizeof(struct sctphdr) + 8 &&
 289                     (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
 290                      n->m_next != NULL)) {
 291                         if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
 292                             (n = m_pullup(n, oiphlen +
 293                             sizeof(struct sctphdr) + 8)) == NULL)
 294                                 goto freeit;
 295                         oip = mtod(n, struct ip *);
 296                         sh = mtodo(n, oiphlen);
 297                         ch = (struct sctp_chunkhdr *)(sh + 1);
 298                         if (ch->chunk_type == SCTP_INITIATION) {
 299                                 icmpelen = max(sizeof(struct sctphdr) + 8,
 300                                     min(V_icmp_quotelen, ntohs(oip->ip_len) -
 301                                     oiphlen));
 302                         }
 303                 }
 304         } else
 305 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
 306                     oiphlen));
 307
 308         icmplen = min(oiphlen + icmpelen, nlen);
 309         if (icmplen < sizeof(struct ip))
 310                 goto freeit;
 311
 312         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 313                 m = m_gethdr(M_NOWAIT, MT_DATA);
 314         else
 315                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 316         if (m == NULL)
 317                 goto freeit;
 318 #ifdef MAC
 319         mac_netinet_icmp_reply(n, m);
 320 #endif
 321         icmplen = min(icmplen, M_TRAILINGSPACE(m) -
 322             sizeof(struct ip) - ICMP_MINLEN);
 323         m_align(m, sizeof(struct ip) + ICMP_MINLEN + icmplen);
 324         m->m_data += sizeof(struct ip);
 325         m->m_len = ICMP_MINLEN + icmplen;
 326
 327         /* XXX MRT  make the outgoing packet use the same FIB
 328          * that was associated with the incoming packet
 329          */
 330         M_SETFIB(m, M_GETFIB(n));
 331         icp = mtod(m, struct icmp *);
 332         ICMPSTAT_INC(icps_outhist[type]);
 333         icp->icmp_type = type;
 334         if (type == ICMP_REDIRECT)
 335                 icp->icmp_gwaddr.s_addr = dest;
 336         else {
 337                 icp->icmp_void = 0;
 338                 /*
 339                  * The following assignments assume an overlay with the
 340                  * just zeroed icmp_void field.
 341                  */
 342                 if (type == ICMP_PARAMPROB) {
 343                         icp->icmp_pptr = code;
 344                         code = 0;
 345                 } else if (type == ICMP_UNREACH &&
 346                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 347                         icp->icmp_nextmtu = htons(mtu);
 348                 }
 349         }
 350         icp->icmp_code = code;
 351
 352         /*
 353          * Copy the quotation into ICMP message and
 354          * convert quoted IP header back to network representation.
 355          */
 356         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 357         nip = &icp->icmp_ip;
 358
 359         /*
 360          * Set up ICMP message mbuf and copy old IP header (without options
 361          * in front of ICMP message.
 362          * If the original mbuf was meant to bypass the firewall, the error
 363          * reply should bypass as well.
 364          */
 365         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 366         KASSERT(M_LEADINGSPACE(m) >= sizeof(struct ip),
 367             ("insufficient space for ip header"));
 368         m->m_data -= sizeof(struct ip);
 369         m->m_len += sizeof(struct ip);
 370         m->m_pkthdr.len = m->m_len;
 371         m->m_pkthdr.rcvif = n->m_pkthdr.rcvif;
 372         nip = mtod(m, struct ip *);
 373         bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip));
 374         nip->ip_len = htons(m->m_len);
 375         nip->ip_v = IPVERSION;
 376         nip->ip_hl = 5;
 377         nip->ip_p = IPPROTO_ICMP;
 378         nip->ip_tos = 0;
 379         nip->ip_off = 0;
 380
 381         if (V_error_keeptags)
 382                 m_tag_copy_chain(m, n, M_NOWAIT);
 383
 384         icmp_reflect(m);
 385
 386 freeit:
 387         m_freem(n);
 388 }
 389
 390 /*
 391  * Process a received ICMP message.
 392  */
 393 int
 394 icmp_input(struct mbuf **mp, int *offp, int proto)
 395 {
 396         struct icmp *icp;
 397         struct in_ifaddr *ia;
 398         struct mbuf *m = *mp;
 399         struct ip *ip = mtod(m, struct ip *);
 400         struct sockaddr_in icmpsrc, icmpdst, icmpgw;
 401         int hlen = *offp;
 402         int icmplen = ntohs(ip->ip_len) - *offp;
 403         int i, code;
 404         void (*ctlfunc)(int, struct sockaddr *, void *);
 405         int fibnum;
 406
 407         *mp = NULL;
 408
 409         /*
 410          * Locate icmp structure in mbuf, and check
 411          * that not corrupted and of at least minimum length.
 412          */
 413 #ifdef ICMPPRINTFS
 414         if (icmpprintfs) {
 415                 char srcbuf[INET_ADDRSTRLEN];
 416                 char dstbuf[INET_ADDRSTRLEN];
 417
 418                 printf("icmp_input from %s to %s, len %d\n",
 419                     inet_ntoa_r(ip->ip_src, srcbuf),
 420                     inet_ntoa_r(ip->ip_dst, dstbuf), icmplen);
 421         }
 422 #endif
 423         NET_EPOCH_ENTER();
 424         if (icmplen < ICMP_MINLEN) {
 425                 ICMPSTAT_INC(icps_tooshort);
 426                 goto freeit;
 427         }
 428         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 429         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 430                 ICMPSTAT_INC(icps_tooshort);
 431                 NET_EPOCH_EXIT();
 432                 return (IPPROTO_DONE);
 433         }
 434         ip = mtod(m, struct ip *);
 435         m->m_len -= hlen;
 436         m->m_data += hlen;
 437         icp = mtod(m, struct icmp *);
 438         if (in_cksum(m, icmplen)) {
 439                 ICMPSTAT_INC(icps_checksum);
 440                 goto freeit;
 441         }
 442         m->m_len += hlen;
 443         m->m_data -= hlen;
 444
 445 #ifdef ICMPPRINTFS
 446         if (icmpprintfs)
 447                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 448                     icp->icmp_code);
 449 #endif
 450
 451         /*
 452          * Message type specific processing.
 453          */
 454         if (icp->icmp_type > ICMP_MAXTYPE)
 455                 goto raw;
 456
 457         /* Initialize */
 458         bzero(&icmpsrc, sizeof(icmpsrc));
 459         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 460         icmpsrc.sin_family = AF_INET;
 461         bzero(&icmpdst, sizeof(icmpdst));
 462         icmpdst.sin_len = sizeof(struct sockaddr_in);
 463         icmpdst.sin_family = AF_INET;
 464         bzero(&icmpgw, sizeof(icmpgw));
 465         icmpgw.sin_len = sizeof(struct sockaddr_in);
 466         icmpgw.sin_family = AF_INET;
 467
 468         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 469         code = icp->icmp_code;
 470         switch (icp->icmp_type) {
 471
 472         case ICMP_UNREACH:
 473                 switch (code) {
 474                         case ICMP_UNREACH_NET:
 475                         case ICMP_UNREACH_HOST:
 476                         case ICMP_UNREACH_SRCFAIL:
 477                         case ICMP_UNREACH_NET_UNKNOWN:
 478                         case ICMP_UNREACH_HOST_UNKNOWN:
 479                         case ICMP_UNREACH_ISOLATED:
 480                         case ICMP_UNREACH_TOSNET:
 481                         case ICMP_UNREACH_TOSHOST:
 482                         case ICMP_UNREACH_HOST_PRECEDENCE:
 483                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 484                                 code = PRC_UNREACH_NET;
 485                                 break;
 486
 487                         case ICMP_UNREACH_NEEDFRAG:
 488                                 code = PRC_MSGSIZE;
 489                                 break;
 490
 491                         /*
 492                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 493                          * Treat subcodes 2,3 as immediate RST
 494                          */
 495                         case ICMP_UNREACH_PROTOCOL:
 496                                 code = PRC_UNREACH_PROTOCOL;
 497                                 break;
 498                         case ICMP_UNREACH_PORT:
 499                                 code = PRC_UNREACH_PORT;
 500                                 break;
 501
 502                         case ICMP_UNREACH_NET_PROHIB:
 503                         case ICMP_UNREACH_HOST_PROHIB:
 504                         case ICMP_UNREACH_FILTER_PROHIB:
 505                                 code = PRC_UNREACH_ADMIN_PROHIB;
 506                                 break;
 507
 508                         default:
 509                                 goto badcode;
 510                 }
 511                 goto deliver;
 512
 513         case ICMP_TIMXCEED:
 514                 if (code > 1)
 515                         goto badcode;
 516                 code += PRC_TIMXCEED_INTRANS;
 517                 goto deliver;
 518
 519         case ICMP_PARAMPROB:
 520                 if (code > 1)
 521                         goto badcode;
 522                 code = PRC_PARAMPROB;
 523         deliver:
 524                 /*
 525                  * Problem with datagram; advise higher level routines.
 526                  */
 527                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 528                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 529                         ICMPSTAT_INC(icps_badlen);
 530                         goto freeit;
 531                 }
 532                 /* Discard ICMP's in response to multicast packets */
 533                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 534                         goto badcode;
 535 #ifdef ICMPPRINTFS
 536                 if (icmpprintfs)
 537                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 538 #endif
 539                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 540                 /*
 541                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 542                  * notification to TCP layer.
 543                  */
 544                 i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
 545                 ip_stripoptions(m);
 546                 if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
 547                         /* This should actually not happen */
 548                         ICMPSTAT_INC(icps_tooshort);
 549                         NET_EPOCH_EXIT();
 550                         return (IPPROTO_DONE);
 551                 }
 552                 ip = mtod(m, struct ip *);
 553                 icp = (struct icmp *)(ip + 1);
 554                 /*
 555                  * The upper layer handler can rely on:
 556                  * - The outer IP header has no options.
 557                  * - The outer IP header, the ICMP header, the inner IP header,
 558                  *   and the first n bytes of the inner payload are contiguous.
 559                  *   n is at least 8, but might be larger based on
 560                  *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
 561                  */
 562                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 563                 if (ctlfunc)
 564                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 565                                    (void *)&icp->icmp_ip);
 566                 break;
 567
 568         badcode:
 569                 ICMPSTAT_INC(icps_badcode);
 570                 break;
 571
 572         case ICMP_ECHO:
 573                 if (!V_icmpbmcastecho
 574                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 575                         ICMPSTAT_INC(icps_bmcastecho);
 576                         break;
 577                 }
 578                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 579                         goto freeit;
 580                 icp->icmp_type = ICMP_ECHOREPLY;
 581                 goto reflect;
 582
 583         case ICMP_TSTAMP:
 584                 if (V_icmptstamprepl == 0)
 585                         break;
 586                 if (!V_icmpbmcastecho
 587                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 588                         ICMPSTAT_INC(icps_bmcasttstamp);
 589                         break;
 590                 }
 591                 if (icmplen < ICMP_TSLEN) {
 592                         ICMPSTAT_INC(icps_badlen);
 593                         break;
 594                 }
 595                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 596                         goto freeit;
 597                 icp->icmp_type = ICMP_TSTAMPREPLY;
 598                 icp->icmp_rtime = iptime();
 599                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 600                 goto reflect;
 601
 602         case ICMP_MASKREQ:
 603                 if (V_icmpmaskrepl == 0)
 604                         break;
 605                 /*
 606                  * We are not able to respond with all ones broadcast
 607                  * unless we receive it over a point-to-point interface.
 608                  */
 609                 if (icmplen < ICMP_MASKLEN)
 610                         break;
 611                 switch (ip->ip_dst.s_addr) {
 612
 613                 case INADDR_BROADCAST:
 614                 case INADDR_ANY:
 615                         icmpdst.sin_addr = ip->ip_src;
 616                         break;
 617
 618                 default:
 619                         icmpdst.sin_addr = ip->ip_dst;
 620                 }
 621                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 622                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 623                 if (ia == NULL)
 624                         break;
 625                 if (ia->ia_ifp == NULL)
 626                         break;
 627                 icp->icmp_type = ICMP_MASKREPLY;
 628                 if (V_icmpmaskfake == 0)
 629                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 630                 else
 631                         icp->icmp_mask = V_icmpmaskfake;
 632                 if (ip->ip_src.s_addr == 0) {
 633                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 634                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 635                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 636                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 637                 }
 638 reflect:
 639                 ICMPSTAT_INC(icps_reflect);
 640                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 641                 icmp_reflect(m);
 642                 NET_EPOCH_EXIT();
 643                 return (IPPROTO_DONE);
 644
 645         case ICMP_REDIRECT:
 646                 if (V_log_redirect) {
 647                         u_long src, dst, gw;
 648
 649                         src = ntohl(ip->ip_src.s_addr);
 650                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 651                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 652                         printf("icmp redirect from %d.%d.%d.%d: "
 653                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 654                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 655                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 656                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 657                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 658                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 659                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 660                 }
 661                 /*
 662                  * RFC1812 says we must ignore ICMP redirects if we
 663                  * are acting as router.
 664                  */
 665                 if (V_drop_redirect || V_ipforwarding)
 666                         break;
 667                 if (code > 3)
 668                         goto badcode;
 669                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 670                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 671                         ICMPSTAT_INC(icps_badlen);
 672                         break;
 673                 }
 674                 /*
 675                  * Short circuit routing redirects to force
 676                  * immediate change in the kernel's routing
 677                  * tables.  The message is also handed to anyone
 678                  * listening on a raw socket (e.g. the routing
 679                  * daemon for use in updating its tables).
 680                  */
 681                 icmpgw.sin_addr = ip->ip_src;
 682                 icmpdst.sin_addr = icp->icmp_gwaddr;
 683 #ifdef  ICMPPRINTFS
 684                 if (icmpprintfs) {
 685                         char dstbuf[INET_ADDRSTRLEN];
 686                         char gwbuf[INET_ADDRSTRLEN];
 687
 688                         printf("redirect dst %s to %s\n",
 689                                inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
 690                                inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
 691                 }
 692 #endif
 693                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 694                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 695                         in_rtredirect((struct sockaddr *)&icmpsrc,
 696                           (struct sockaddr *)&icmpdst,
 697                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 698                           (struct sockaddr *)&icmpgw, fibnum);
 699                 }
 700                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 701                 break;
 702
 703         /*
 704          * No kernel processing for the following;
 705          * just fall through to send to raw listener.
 706          */
 707         case ICMP_ECHOREPLY:
 708         case ICMP_ROUTERADVERT:
 709         case ICMP_ROUTERSOLICIT:
 710         case ICMP_TSTAMPREPLY:
 711         case ICMP_IREQREPLY:
 712         case ICMP_MASKREPLY:
 713         case ICMP_SOURCEQUENCH:
 714         default:
 715                 break;
 716         }
 717
 718 raw:
 719         NET_EPOCH_EXIT();
 720         *mp = m;
 721         rip_input(mp, offp, proto);
 722         return (IPPROTO_DONE);
 723
 724 freeit:
 725         NET_EPOCH_EXIT();
 726         m_freem(m);
 727         return (IPPROTO_DONE);
 728 }
 729
 730 /*
 731  * Reflect the ip packet back to the source
 732  */
 733 static void
 734 icmp_reflect(struct mbuf *m)
 735 {
 736         struct rm_priotracker in_ifa_tracker;
 737         struct ip *ip = mtod(m, struct ip *);
 738         struct ifaddr *ifa;
 739         struct ifnet *ifp;
 740         struct in_ifaddr *ia;
 741         struct in_addr t;
 742         struct nhop4_extended nh_ext;
 743         struct mbuf *opts = NULL;
 744         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 745
 746         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 747             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 748             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 749                 m_freem(m);     /* Bad return address */
 750                 ICMPSTAT_INC(icps_badaddr);
 751                 goto done;      /* Ip_output() will check for broadcast */
 752         }
 753
 754         t = ip->ip_dst;
 755         ip->ip_dst = ip->ip_src;
 756
 757         /*
 758          * Source selection for ICMP replies:
 759          *
 760          * If the incoming packet was addressed directly to one of our
 761          * own addresses, use dst as the src for the reply.
 762          */
 763         IN_IFADDR_RLOCK(&in_ifa_tracker);
 764         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 765                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 766                         t = IA_SIN(ia)->sin_addr;
 767                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 768                         goto match;
 769                 }
 770         }
 771         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 772
 773         /*
 774          * If the incoming packet was addressed to one of our broadcast
 775          * addresses, use the first non-broadcast address which corresponds
 776          * to the incoming interface.
 777          */
 778         ifp = m->m_pkthdr.rcvif;
 779         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 780                 IF_ADDR_RLOCK(ifp);
 781                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 782                         if (ifa->ifa_addr->sa_family != AF_INET)
 783                                 continue;
 784                         ia = ifatoia(ifa);
 785                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 786                             t.s_addr) {
 787                                 t = IA_SIN(ia)->sin_addr;
 788                                 IF_ADDR_RUNLOCK(ifp);
 789                                 goto match;
 790                         }
 791                 }
 792                 IF_ADDR_RUNLOCK(ifp);
 793         }
 794         /*
 795          * If the packet was transiting through us, use the address of
 796          * the interface the packet came through in.  If that interface
 797          * doesn't have a suitable IP address, the normal selection
 798          * criteria apply.
 799          */
 800         if (V_icmp_rfi && ifp != NULL) {
 801                 IF_ADDR_RLOCK(ifp);
 802                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 803                         if (ifa->ifa_addr->sa_family != AF_INET)
 804                                 continue;
 805                         ia = ifatoia(ifa);
 806                         t = IA_SIN(ia)->sin_addr;
 807                         IF_ADDR_RUNLOCK(ifp);
 808                         goto match;
 809                 }
 810                 IF_ADDR_RUNLOCK(ifp);
 811         }
 812         /*
 813          * If the incoming packet was not addressed directly to us, use
 814          * designated interface for icmp replies specified by sysctl
 815          * net.inet.icmp.reply_src (default not set). Otherwise continue
 816          * with normal source selection.
 817          */
 818         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 819                 IF_ADDR_RLOCK(ifp);
 820                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 821                         if (ifa->ifa_addr->sa_family != AF_INET)
 822                                 continue;
 823                         ia = ifatoia(ifa);
 824                         t = IA_SIN(ia)->sin_addr;
 825                         IF_ADDR_RUNLOCK(ifp);
 826                         goto match;
 827                 }
 828                 IF_ADDR_RUNLOCK(ifp);
 829         }
 830         /*
 831          * If the packet was transiting through us, use the address of
 832          * the interface that is the closest to the packet source.
 833          * When we don't have a route back to the packet source, stop here
 834          * and drop the packet.
 835          */
 836         if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
 837                 m_freem(m);
 838                 ICMPSTAT_INC(icps_noroute);
 839                 goto done;
 840         }
 841         t = nh_ext.nh_src;
 842 match:
 843 #ifdef MAC
 844         mac_netinet_icmp_replyinplace(m);
 845 #endif
 846         ip->ip_src = t;
 847         ip->ip_ttl = V_ip_defttl;
 848
 849         if (optlen > 0) {
 850                 u_char *cp;
 851                 int opt, cnt;
 852                 u_int len;
 853
 854                 /*
 855                  * Retrieve any source routing from the incoming packet;
 856                  * add on any record-route or timestamp options.
 857                  */
 858                 cp = (u_char *) (ip + 1);
 859                 if ((opts = ip_srcroute(m)) == NULL &&
 860                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 861                         opts->m_len = sizeof(struct in_addr);
 862                         mtod(opts, struct in_addr *)->s_addr = 0;
 863                 }
 864                 if (opts) {
 865 #ifdef ICMPPRINTFS
 866                     if (icmpprintfs)
 867                             printf("icmp_reflect optlen %d rt %d => ",
 868                                 optlen, opts->m_len);
 869 #endif
 870                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 871                             opt = cp[IPOPT_OPTVAL];
 872                             if (opt == IPOPT_EOL)
 873                                     break;
 874                             if (opt == IPOPT_NOP)
 875                                     len = 1;
 876                             else {
 877                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 878                                             break;
 879                                     len = cp[IPOPT_OLEN];
 880                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 881                                         len > cnt)
 882                                             break;
 883                             }
 884                             /*
 885                              * Should check for overflow, but it "can't happen"
 886                              */
 887                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 888                                 opt == IPOPT_SECURITY) {
 889                                     bcopy((caddr_t)cp,
 890                                         mtod(opts, caddr_t) + opts->m_len, len);
 891                                     opts->m_len += len;
 892                             }
 893                     }
 894                     /* Terminate & pad, if necessary */
 895                     cnt = opts->m_len % 4;
 896                     if (cnt) {
 897                             for (; cnt < 4; cnt++) {
 898                                     *(mtod(opts, caddr_t) + opts->m_len) =
 899                                         IPOPT_EOL;
 900                                     opts->m_len++;
 901                             }
 902                     }
 903 #ifdef ICMPPRINTFS
 904                     if (icmpprintfs)
 905                             printf("%d\n", opts->m_len);
 906 #endif
 907                 }
 908                 ip_stripoptions(m);
 909         }
 910         m_tag_delete_nonpersistent(m);
 911         m->m_flags &= ~(M_BCAST|M_MCAST);
 912         icmp_send(m, opts);
 913 done:
 914         if (opts)
 915                 (void)m_free(opts);
 916 }
 917
 918 /*
 919  * Send an icmp packet back to the ip level,
 920  * after supplying a checksum.
 921  */
 922 static void
 923 icmp_send(struct mbuf *m, struct mbuf *opts)
 924 {
 925         struct ip *ip = mtod(m, struct ip *);
 926         int hlen;
 927         struct icmp *icp;
 928
 929         hlen = ip->ip_hl << 2;
 930         m->m_data += hlen;
 931         m->m_len -= hlen;
 932         icp = mtod(m, struct icmp *);
 933         icp->icmp_cksum = 0;
 934         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 935         m->m_data -= hlen;
 936         m->m_len += hlen;
 937         m->m_pkthdr.rcvif = (struct ifnet *)0;
 938 #ifdef ICMPPRINTFS
 939         if (icmpprintfs) {
 940                 char dstbuf[INET_ADDRSTRLEN];
 941                 char srcbuf[INET_ADDRSTRLEN];
 942
 943                 printf("icmp_send dst %s src %s\n",
 944                     inet_ntoa_r(ip->ip_dst, dstbuf),
 945                     inet_ntoa_r(ip->ip_src, srcbuf));
 946         }
 947 #endif
 948         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 949 }
 950
 951 /*
 952  * Return milliseconds since 00:00 UTC in network format.
 953  */
 954 uint32_t
 955 iptime(void)
 956 {
 957         struct timeval atv;
 958         u_long t;
 959
 960         getmicrotime(&atv);
 961         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 962         return (htonl(t));
 963 }
 964
 965 /*
 966  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 967  * given current value MTU.  If DIR is less than zero, a larger plateau
 968  * is returned; otherwise, a smaller value is returned.
 969  */
 970 int
 971 ip_next_mtu(int mtu, int dir)
 972 {
 973         static int mtutab[] = {
 974                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 975                 296, 68, 0
 976         };
 977         int i, size;
 978
 979         size = (sizeof mtutab) / (sizeof mtutab[0]);
 980         if (dir >= 0) {
 981                 for (i = 0; i < size; i++)
 982                         if (mtu > mtutab[i])
 983                                 return mtutab[i];
 984         } else {
 985                 for (i = size - 1; i >= 0; i--)
 986                         if (mtu < mtutab[i])
 987                                 return mtutab[i];
 988                 if (mtu == mtutab[0])
 989                         return mtutab[0];
 990         }
 991         return 0;
 992 }
 993 #endif /* INET */
 994
 995
 996 /*
 997  * badport_bandlim() - check for ICMP bandwidth limit
 998  *
 999  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
1000  *      hit our bandwidth limit and it is not ok.
1001  *
1002  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
1003  *
1004  *      For now we separate the TCP and UDP subsystems w/ different 'which'
1005  *      values.  We may eventually remove this separation (and simplify the
1006  *      code further).
1007  *
1008  *      Note that the printing of the error message is delayed so we can
1009  *      properly print the icmp error rate that the system was trying to do
1010  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
1011  *      the 'final' error, but it doesn't make sense to solve the printing
1012  *      delay with more complex code.
1013  */
1014 struct icmp_rate {
1015         const char *descr;
1016         struct counter_rate cr;
1017 };
1018 VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1019         { "icmp unreach response" },
1020         { "icmp ping response" },
1021         { "icmp tstamp response" },
1022         { "closed port RST response" },
1023         { "open port RST response" },
1024         { "icmp6 unreach response" },
1025         { "sctp ootb response" }
1026 };
1027 #define V_icmp_rates    VNET(icmp_rates)
1028
1029 static void
1030 icmp_bandlimit_init(void)
1031 {
1032
1033         for (int i = 0; i < BANDLIM_MAX; i++) {
1034                 V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1035                 V_icmp_rates[i].cr.cr_ticks = ticks;
1036         }
1037 }
1038 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1039     icmp_bandlimit_init, NULL);
1040
1041 static void
1042 icmp_bandlimit_uninit(void)
1043 {
1044
1045         for (int i = 0; i < BANDLIM_MAX; i++)
1046                 counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1047 }
1048 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1049     icmp_bandlimit_uninit, NULL);
1050
1051 int
1052 badport_bandlim(int which)
1053 {
1054         int64_t pps;
1055
1056         if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1057                 return (0);
1058
1059         KASSERT(which >= 0 && which < BANDLIM_MAX,
1060             ("%s: which %d", __func__, which));
1061
1062         pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1063         if (pps == -1)
1064                 return (-1);
1065         if (pps > 0 && V_icmplim_output)
1066                 log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1067                         V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1068         return (0);
1069 }