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 #ifdef ICMPPRINTFS
 162 int     icmpprintfs = 0;
 163 #endif
 164
 165 static void     icmp_reflect(struct mbuf *);
 166 static void     icmp_send(struct mbuf *, struct mbuf *);
 167
 168 extern  struct protosw inetsw[];
 169
 170 /*
 171  * Kernel module interface for updating icmpstat.  The argument is an index
 172  * into icmpstat treated as an array of u_long.  While this encodes the
 173  * general layout of icmpstat into the caller, it doesn't encode its
 174  * location, so that future changes to add, for example, per-CPU stats
 175  * support won't cause binary compatibility problems for kernel modules.
 176  */
 177 void
 178 kmod_icmpstat_inc(int statnum)
 179 {
 180
 181         counter_u64_add(VNET(icmpstat)[statnum], 1);
 182 }
 183
 184 /*
 185  * Generate an error packet of type error
 186  * in response to bad packet ip.
 187  */
 188 void
 189 icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu)
 190 {
 191         struct ip *oip, *nip;
 192         struct icmp *icp;
 193         struct mbuf *m;
 194         unsigned icmplen, icmpelen, nlen, oiphlen;
 195
 196         KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type",
 197             __func__));
 198
 199         if (type != ICMP_REDIRECT)
 200                 ICMPSTAT_INC(icps_error);
 201         /*
 202          * Don't send error:
 203          *  if the original packet was encrypted.
 204          *  if not the first fragment of message.
 205          *  in response to a multicast or broadcast packet.
 206          *  if the old packet protocol was an ICMP error message.
 207          */
 208         if (n->m_flags & M_DECRYPTED)
 209                 goto freeit;
 210         if (n->m_flags & (M_BCAST|M_MCAST))
 211                 goto freeit;
 212
 213         /* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */
 214         if (n->m_len < sizeof(struct ip) + ICMP_MINLEN)
 215                 goto freeit;
 216         oip = mtod(n, struct ip *);
 217         oiphlen = oip->ip_hl << 2;
 218         if (n->m_len < oiphlen + ICMP_MINLEN)
 219                 goto freeit;
 220 #ifdef ICMPPRINTFS
 221         if (icmpprintfs)
 222                 printf("icmp_error(%p, %x, %d)\n", oip, type, code);
 223 #endif
 224         if (oip->ip_off & htons(~(IP_MF|IP_DF)))
 225                 goto freeit;
 226         if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT &&
 227             !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip +
 228                 oiphlen))->icmp_type)) {
 229                 ICMPSTAT_INC(icps_oldicmp);
 230                 goto freeit;
 231         }
 232         /*
 233          * Calculate length to quote from original packet and
 234          * prevent the ICMP mbuf from overflowing.
 235          * Unfortunately this is non-trivial since ip_forward()
 236          * sends us truncated packets.
 237          */
 238         nlen = m_length(n, NULL);
 239         if (oip->ip_p == IPPROTO_TCP) {
 240                 struct tcphdr *th;
 241                 int tcphlen;
 242
 243                 if (oiphlen + sizeof(struct tcphdr) > n->m_len &&
 244                     n->m_next == NULL)
 245                         goto stdreply;
 246                 if (n->m_len < oiphlen + sizeof(struct tcphdr) &&
 247                     (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL)
 248                         goto freeit;
 249                 oip = mtod(n, struct ip *);
 250                 th = mtodo(n, oiphlen);
 251                 tcphlen = th->th_off << 2;
 252                 if (tcphlen < sizeof(struct tcphdr))
 253                         goto freeit;
 254                 if (ntohs(oip->ip_len) < oiphlen + tcphlen)
 255                         goto freeit;
 256                 if (oiphlen + tcphlen > n->m_len && n->m_next == NULL)
 257                         goto stdreply;
 258                 if (n->m_len < oiphlen + tcphlen &&
 259                     (n = m_pullup(n, oiphlen + tcphlen)) == NULL)
 260                         goto freeit;
 261                 icmpelen = max(tcphlen, min(V_icmp_quotelen,
 262                     ntohs(oip->ip_len) - oiphlen));
 263         } else if (oip->ip_p == IPPROTO_SCTP) {
 264                 struct sctphdr *sh;
 265                 struct sctp_chunkhdr *ch;
 266
 267                 if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr))
 268                         goto stdreply;
 269                 if (oiphlen + sizeof(struct sctphdr) > n->m_len &&
 270                     n->m_next == NULL)
 271                         goto stdreply;
 272                 if (n->m_len < oiphlen + sizeof(struct sctphdr) &&
 273                     (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL)
 274                         goto freeit;
 275                 oip = mtod(n, struct ip *);
 276                 icmpelen = max(sizeof(struct sctphdr),
 277                     min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen));
 278                 sh = mtodo(n, oiphlen);
 279                 if (ntohl(sh->v_tag) == 0 &&
 280                     ntohs(oip->ip_len) >= oiphlen +
 281                     sizeof(struct sctphdr) + 8 &&
 282                     (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 ||
 283                      n->m_next != NULL)) {
 284                         if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 &&
 285                             (n = m_pullup(n, oiphlen +
 286                             sizeof(struct sctphdr) + 8)) == NULL)
 287                                 goto freeit;
 288                         oip = mtod(n, struct ip *);
 289                         sh = mtodo(n, oiphlen);
 290                         ch = (struct sctp_chunkhdr *)(sh + 1);
 291                         if (ch->chunk_type == SCTP_INITIATION) {
 292                                 icmpelen = max(sizeof(struct sctphdr) + 8,
 293                                     min(V_icmp_quotelen, ntohs(oip->ip_len) -
 294                                     oiphlen));
 295                         }
 296                 }
 297         } else
 298 stdreply:       icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) -
 299                     oiphlen));
 300
 301         icmplen = min(oiphlen + icmpelen, nlen);
 302         if (icmplen < sizeof(struct ip))
 303                 goto freeit;
 304
 305         if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen)
 306                 m = m_gethdr(M_NOWAIT, MT_DATA);
 307         else
 308                 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 309         if (m == NULL)
 310                 goto freeit;
 311 #ifdef MAC
 312         mac_netinet_icmp_reply(n, m);
 313 #endif
 314         icmplen = min(icmplen, M_TRAILINGSPACE(m) -
 315             sizeof(struct ip) - ICMP_MINLEN);
 316         m_align(m, ICMP_MINLEN + icmplen);
 317         m->m_len = ICMP_MINLEN + icmplen;
 318
 319         /* XXX MRT  make the outgoing packet use the same FIB
 320          * that was associated with the incoming packet
 321          */
 322         M_SETFIB(m, M_GETFIB(n));
 323         icp = mtod(m, struct icmp *);
 324         ICMPSTAT_INC(icps_outhist[type]);
 325         icp->icmp_type = type;
 326         if (type == ICMP_REDIRECT)
 327                 icp->icmp_gwaddr.s_addr = dest;
 328         else {
 329                 icp->icmp_void = 0;
 330                 /*
 331                  * The following assignments assume an overlay with the
 332                  * just zeroed icmp_void field.
 333                  */
 334                 if (type == ICMP_PARAMPROB) {
 335                         icp->icmp_pptr = code;
 336                         code = 0;
 337                 } else if (type == ICMP_UNREACH &&
 338                         code == ICMP_UNREACH_NEEDFRAG && mtu) {
 339                         icp->icmp_nextmtu = htons(mtu);
 340                 }
 341         }
 342         icp->icmp_code = code;
 343
 344         /*
 345          * Copy the quotation into ICMP message and
 346          * convert quoted IP header back to network representation.
 347          */
 348         m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip);
 349         nip = &icp->icmp_ip;
 350
 351         /*
 352          * Set up ICMP message mbuf and copy old IP header (without options
 353          * in front of ICMP message.
 354          * If the original mbuf was meant to bypass the firewall, the error
 355          * reply should bypass as well.
 356          */
 357         m->m_flags |= n->m_flags & M_SKIP_FIREWALL;
 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         NET_EPOCH_ENTER();
 410         if (icmplen < ICMP_MINLEN) {
 411                 ICMPSTAT_INC(icps_tooshort);
 412                 goto freeit;
 413         }
 414         i = hlen + min(icmplen, ICMP_ADVLENMIN);
 415         if (m->m_len < i && (m = m_pullup(m, i)) == NULL)  {
 416                 ICMPSTAT_INC(icps_tooshort);
 417                 NET_EPOCH_EXIT();
 418                 return (IPPROTO_DONE);
 419         }
 420         ip = mtod(m, struct ip *);
 421         m->m_len -= hlen;
 422         m->m_data += hlen;
 423         icp = mtod(m, struct icmp *);
 424         if (in_cksum(m, icmplen)) {
 425                 ICMPSTAT_INC(icps_checksum);
 426                 goto freeit;
 427         }
 428         m->m_len += hlen;
 429         m->m_data -= hlen;
 430
 431 #ifdef ICMPPRINTFS
 432         if (icmpprintfs)
 433                 printf("icmp_input, type %d code %d\n", icp->icmp_type,
 434                     icp->icmp_code);
 435 #endif
 436
 437         /*
 438          * Message type specific processing.
 439          */
 440         if (icp->icmp_type > ICMP_MAXTYPE)
 441                 goto raw;
 442
 443         /* Initialize */
 444         bzero(&icmpsrc, sizeof(icmpsrc));
 445         icmpsrc.sin_len = sizeof(struct sockaddr_in);
 446         icmpsrc.sin_family = AF_INET;
 447         bzero(&icmpdst, sizeof(icmpdst));
 448         icmpdst.sin_len = sizeof(struct sockaddr_in);
 449         icmpdst.sin_family = AF_INET;
 450         bzero(&icmpgw, sizeof(icmpgw));
 451         icmpgw.sin_len = sizeof(struct sockaddr_in);
 452         icmpgw.sin_family = AF_INET;
 453
 454         ICMPSTAT_INC(icps_inhist[icp->icmp_type]);
 455         code = icp->icmp_code;
 456         switch (icp->icmp_type) {
 457
 458         case ICMP_UNREACH:
 459                 switch (code) {
 460                         case ICMP_UNREACH_NET:
 461                         case ICMP_UNREACH_HOST:
 462                         case ICMP_UNREACH_SRCFAIL:
 463                         case ICMP_UNREACH_NET_UNKNOWN:
 464                         case ICMP_UNREACH_HOST_UNKNOWN:
 465                         case ICMP_UNREACH_ISOLATED:
 466                         case ICMP_UNREACH_TOSNET:
 467                         case ICMP_UNREACH_TOSHOST:
 468                         case ICMP_UNREACH_HOST_PRECEDENCE:
 469                         case ICMP_UNREACH_PRECEDENCE_CUTOFF:
 470                                 code = PRC_UNREACH_NET;
 471                                 break;
 472
 473                         case ICMP_UNREACH_NEEDFRAG:
 474                                 code = PRC_MSGSIZE;
 475                                 break;
 476
 477                         /*
 478                          * RFC 1122, Sections 3.2.2.1 and 4.2.3.9.
 479                          * Treat subcodes 2,3 as immediate RST
 480                          */
 481                         case ICMP_UNREACH_PROTOCOL:
 482                                 code = PRC_UNREACH_PROTOCOL;
 483                                 break;
 484                         case ICMP_UNREACH_PORT:
 485                                 code = PRC_UNREACH_PORT;
 486                                 break;
 487
 488                         case ICMP_UNREACH_NET_PROHIB:
 489                         case ICMP_UNREACH_HOST_PROHIB:
 490                         case ICMP_UNREACH_FILTER_PROHIB:
 491                                 code = PRC_UNREACH_ADMIN_PROHIB;
 492                                 break;
 493
 494                         default:
 495                                 goto badcode;
 496                 }
 497                 goto deliver;
 498
 499         case ICMP_TIMXCEED:
 500                 if (code > 1)
 501                         goto badcode;
 502                 code += PRC_TIMXCEED_INTRANS;
 503                 goto deliver;
 504
 505         case ICMP_PARAMPROB:
 506                 if (code > 1)
 507                         goto badcode;
 508                 code = PRC_PARAMPROB;
 509         deliver:
 510                 /*
 511                  * Problem with datagram; advise higher level routines.
 512                  */
 513                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 514                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 515                         ICMPSTAT_INC(icps_badlen);
 516                         goto freeit;
 517                 }
 518                 /* Discard ICMP's in response to multicast packets */
 519                 if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr)))
 520                         goto badcode;
 521 #ifdef ICMPPRINTFS
 522                 if (icmpprintfs)
 523                         printf("deliver to protocol %d\n", icp->icmp_ip.ip_p);
 524 #endif
 525                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 526                 /*
 527                  * XXX if the packet contains [IPv4 AH TCP], we can't make a
 528                  * notification to TCP layer.
 529                  */
 530                 i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp));
 531                 ip_stripoptions(m);
 532                 if (m->m_len < i && (m = m_pullup(m, i)) == NULL) {
 533                         /* This should actually not happen */
 534                         ICMPSTAT_INC(icps_tooshort);
 535                         NET_EPOCH_EXIT();
 536                         return (IPPROTO_DONE);
 537                 }
 538                 ip = mtod(m, struct ip *);
 539                 icp = (struct icmp *)(ip + 1);
 540                 /*
 541                  * The upper layer handler can rely on:
 542                  * - The outer IP header has no options.
 543                  * - The outer IP header, the ICMP header, the inner IP header,
 544                  *   and the first n bytes of the inner payload are contiguous.
 545                  *   n is at least 8, but might be larger based on
 546                  *   ICMP_ADVLENPREF. See its definition in ip_icmp.h.
 547                  */
 548                 ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 549                 if (ctlfunc)
 550                         (*ctlfunc)(code, (struct sockaddr *)&icmpsrc,
 551                                    (void *)&icp->icmp_ip);
 552                 break;
 553
 554         badcode:
 555                 ICMPSTAT_INC(icps_badcode);
 556                 break;
 557
 558         case ICMP_ECHO:
 559                 if (!V_icmpbmcastecho
 560                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 561                         ICMPSTAT_INC(icps_bmcastecho);
 562                         break;
 563                 }
 564                 if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0)
 565                         goto freeit;
 566                 icp->icmp_type = ICMP_ECHOREPLY;
 567                 goto reflect;
 568
 569         case ICMP_TSTAMP:
 570                 if (V_icmptstamprepl == 0)
 571                         break;
 572                 if (!V_icmpbmcastecho
 573                     && (m->m_flags & (M_MCAST | M_BCAST)) != 0) {
 574                         ICMPSTAT_INC(icps_bmcasttstamp);
 575                         break;
 576                 }
 577                 if (icmplen < ICMP_TSLEN) {
 578                         ICMPSTAT_INC(icps_badlen);
 579                         break;
 580                 }
 581                 if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0)
 582                         goto freeit;
 583                 icp->icmp_type = ICMP_TSTAMPREPLY;
 584                 icp->icmp_rtime = iptime();
 585                 icp->icmp_ttime = icp->icmp_rtime;      /* bogus, do later! */
 586                 goto reflect;
 587
 588         case ICMP_MASKREQ:
 589                 if (V_icmpmaskrepl == 0)
 590                         break;
 591                 /*
 592                  * We are not able to respond with all ones broadcast
 593                  * unless we receive it over a point-to-point interface.
 594                  */
 595                 if (icmplen < ICMP_MASKLEN)
 596                         break;
 597                 switch (ip->ip_dst.s_addr) {
 598
 599                 case INADDR_BROADCAST:
 600                 case INADDR_ANY:
 601                         icmpdst.sin_addr = ip->ip_src;
 602                         break;
 603
 604                 default:
 605                         icmpdst.sin_addr = ip->ip_dst;
 606                 }
 607                 ia = (struct in_ifaddr *)ifaof_ifpforaddr(
 608                             (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif);
 609                 if (ia == NULL)
 610                         break;
 611                 if (ia->ia_ifp == NULL)
 612                         break;
 613                 icp->icmp_type = ICMP_MASKREPLY;
 614                 if (V_icmpmaskfake == 0)
 615                         icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr;
 616                 else
 617                         icp->icmp_mask = V_icmpmaskfake;
 618                 if (ip->ip_src.s_addr == 0) {
 619                         if (ia->ia_ifp->if_flags & IFF_BROADCAST)
 620                             ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr;
 621                         else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT)
 622                             ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr;
 623                 }
 624 reflect:
 625                 ICMPSTAT_INC(icps_reflect);
 626                 ICMPSTAT_INC(icps_outhist[icp->icmp_type]);
 627                 icmp_reflect(m);
 628                 NET_EPOCH_EXIT();
 629                 return (IPPROTO_DONE);
 630
 631         case ICMP_REDIRECT:
 632                 if (V_log_redirect) {
 633                         u_long src, dst, gw;
 634
 635                         src = ntohl(ip->ip_src.s_addr);
 636                         dst = ntohl(icp->icmp_ip.ip_dst.s_addr);
 637                         gw = ntohl(icp->icmp_gwaddr.s_addr);
 638                         printf("icmp redirect from %d.%d.%d.%d: "
 639                                "%d.%d.%d.%d => %d.%d.%d.%d\n",
 640                                (int)(src >> 24), (int)((src >> 16) & 0xff),
 641                                (int)((src >> 8) & 0xff), (int)(src & 0xff),
 642                                (int)(dst >> 24), (int)((dst >> 16) & 0xff),
 643                                (int)((dst >> 8) & 0xff), (int)(dst & 0xff),
 644                                (int)(gw >> 24), (int)((gw >> 16) & 0xff),
 645                                (int)((gw >> 8) & 0xff), (int)(gw & 0xff));
 646                 }
 647                 /*
 648                  * RFC1812 says we must ignore ICMP redirects if we
 649                  * are acting as router.
 650                  */
 651                 if (V_drop_redirect || V_ipforwarding)
 652                         break;
 653                 if (code > 3)
 654                         goto badcode;
 655                 if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) ||
 656                     icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) {
 657                         ICMPSTAT_INC(icps_badlen);
 658                         break;
 659                 }
 660                 /*
 661                  * Short circuit routing redirects to force
 662                  * immediate change in the kernel's routing
 663                  * tables.  The message is also handed to anyone
 664                  * listening on a raw socket (e.g. the routing
 665                  * daemon for use in updating its tables).
 666                  */
 667                 icmpgw.sin_addr = ip->ip_src;
 668                 icmpdst.sin_addr = icp->icmp_gwaddr;
 669 #ifdef  ICMPPRINTFS
 670                 if (icmpprintfs) {
 671                         char dstbuf[INET_ADDRSTRLEN];
 672                         char gwbuf[INET_ADDRSTRLEN];
 673
 674                         printf("redirect dst %s to %s\n",
 675                                inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf),
 676                                inet_ntoa_r(icp->icmp_gwaddr, gwbuf));
 677                 }
 678 #endif
 679                 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
 680                 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
 681                         in_rtredirect((struct sockaddr *)&icmpsrc,
 682                           (struct sockaddr *)&icmpdst,
 683                           (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST,
 684                           (struct sockaddr *)&icmpgw, fibnum);
 685                 }
 686                 pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc);
 687                 break;
 688
 689         /*
 690          * No kernel processing for the following;
 691          * just fall through to send to raw listener.
 692          */
 693         case ICMP_ECHOREPLY:
 694         case ICMP_ROUTERADVERT:
 695         case ICMP_ROUTERSOLICIT:
 696         case ICMP_TSTAMPREPLY:
 697         case ICMP_IREQREPLY:
 698         case ICMP_MASKREPLY:
 699         case ICMP_SOURCEQUENCH:
 700         default:
 701                 break;
 702         }
 703
 704 raw:
 705         NET_EPOCH_EXIT();
 706         *mp = m;
 707         rip_input(mp, offp, proto);
 708         return (IPPROTO_DONE);
 709
 710 freeit:
 711         NET_EPOCH_EXIT();
 712         m_freem(m);
 713         return (IPPROTO_DONE);
 714 }
 715
 716 /*
 717  * Reflect the ip packet back to the source
 718  */
 719 static void
 720 icmp_reflect(struct mbuf *m)
 721 {
 722         struct rm_priotracker in_ifa_tracker;
 723         struct ip *ip = mtod(m, struct ip *);
 724         struct ifaddr *ifa;
 725         struct ifnet *ifp;
 726         struct in_ifaddr *ia;
 727         struct in_addr t;
 728         struct nhop4_extended nh_ext;
 729         struct mbuf *opts = NULL;
 730         int optlen = (ip->ip_hl << 2) - sizeof(struct ip);
 731
 732         if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
 733             IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) ||
 734             IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) {
 735                 m_freem(m);     /* Bad return address */
 736                 ICMPSTAT_INC(icps_badaddr);
 737                 goto done;      /* Ip_output() will check for broadcast */
 738         }
 739
 740         t = ip->ip_dst;
 741         ip->ip_dst = ip->ip_src;
 742
 743         /*
 744          * Source selection for ICMP replies:
 745          *
 746          * If the incoming packet was addressed directly to one of our
 747          * own addresses, use dst as the src for the reply.
 748          */
 749         IN_IFADDR_RLOCK(&in_ifa_tracker);
 750         LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) {
 751                 if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) {
 752                         t = IA_SIN(ia)->sin_addr;
 753                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 754                         goto match;
 755                 }
 756         }
 757         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
 758
 759         /*
 760          * If the incoming packet was addressed to one of our broadcast
 761          * addresses, use the first non-broadcast address which corresponds
 762          * to the incoming interface.
 763          */
 764         ifp = m->m_pkthdr.rcvif;
 765         if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) {
 766                 IF_ADDR_RLOCK(ifp);
 767                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 768                         if (ifa->ifa_addr->sa_family != AF_INET)
 769                                 continue;
 770                         ia = ifatoia(ifa);
 771                         if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
 772                             t.s_addr) {
 773                                 t = IA_SIN(ia)->sin_addr;
 774                                 IF_ADDR_RUNLOCK(ifp);
 775                                 goto match;
 776                         }
 777                 }
 778                 IF_ADDR_RUNLOCK(ifp);
 779         }
 780         /*
 781          * If the packet was transiting through us, use the address of
 782          * the interface the packet came through in.  If that interface
 783          * doesn't have a suitable IP address, the normal selection
 784          * criteria apply.
 785          */
 786         if (V_icmp_rfi && ifp != NULL) {
 787                 IF_ADDR_RLOCK(ifp);
 788                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 789                         if (ifa->ifa_addr->sa_family != AF_INET)
 790                                 continue;
 791                         ia = ifatoia(ifa);
 792                         t = IA_SIN(ia)->sin_addr;
 793                         IF_ADDR_RUNLOCK(ifp);
 794                         goto match;
 795                 }
 796                 IF_ADDR_RUNLOCK(ifp);
 797         }
 798         /*
 799          * If the incoming packet was not addressed directly to us, use
 800          * designated interface for icmp replies specified by sysctl
 801          * net.inet.icmp.reply_src (default not set). Otherwise continue
 802          * with normal source selection.
 803          */
 804         if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) {
 805                 IF_ADDR_RLOCK(ifp);
 806                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 807                         if (ifa->ifa_addr->sa_family != AF_INET)
 808                                 continue;
 809                         ia = ifatoia(ifa);
 810                         t = IA_SIN(ia)->sin_addr;
 811                         IF_ADDR_RUNLOCK(ifp);
 812                         goto match;
 813                 }
 814                 IF_ADDR_RUNLOCK(ifp);
 815         }
 816         /*
 817          * If the packet was transiting through us, use the address of
 818          * the interface that is the closest to the packet source.
 819          * When we don't have a route back to the packet source, stop here
 820          * and drop the packet.
 821          */
 822         if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) {
 823                 m_freem(m);
 824                 ICMPSTAT_INC(icps_noroute);
 825                 goto done;
 826         }
 827         t = nh_ext.nh_src;
 828 match:
 829 #ifdef MAC
 830         mac_netinet_icmp_replyinplace(m);
 831 #endif
 832         ip->ip_src = t;
 833         ip->ip_ttl = V_ip_defttl;
 834
 835         if (optlen > 0) {
 836                 u_char *cp;
 837                 int opt, cnt;
 838                 u_int len;
 839
 840                 /*
 841                  * Retrieve any source routing from the incoming packet;
 842                  * add on any record-route or timestamp options.
 843                  */
 844                 cp = (u_char *) (ip + 1);
 845                 if ((opts = ip_srcroute(m)) == NULL &&
 846                     (opts = m_gethdr(M_NOWAIT, MT_DATA))) {
 847                         opts->m_len = sizeof(struct in_addr);
 848                         mtod(opts, struct in_addr *)->s_addr = 0;
 849                 }
 850                 if (opts) {
 851 #ifdef ICMPPRINTFS
 852                     if (icmpprintfs)
 853                             printf("icmp_reflect optlen %d rt %d => ",
 854                                 optlen, opts->m_len);
 855 #endif
 856                     for (cnt = optlen; cnt > 0; cnt -= len, cp += len) {
 857                             opt = cp[IPOPT_OPTVAL];
 858                             if (opt == IPOPT_EOL)
 859                                     break;
 860                             if (opt == IPOPT_NOP)
 861                                     len = 1;
 862                             else {
 863                                     if (cnt < IPOPT_OLEN + sizeof(*cp))
 864                                             break;
 865                                     len = cp[IPOPT_OLEN];
 866                                     if (len < IPOPT_OLEN + sizeof(*cp) ||
 867                                         len > cnt)
 868                                             break;
 869                             }
 870                             /*
 871                              * Should check for overflow, but it "can't happen"
 872                              */
 873                             if (opt == IPOPT_RR || opt == IPOPT_TS ||
 874                                 opt == IPOPT_SECURITY) {
 875                                     bcopy((caddr_t)cp,
 876                                         mtod(opts, caddr_t) + opts->m_len, len);
 877                                     opts->m_len += len;
 878                             }
 879                     }
 880                     /* Terminate & pad, if necessary */
 881                     cnt = opts->m_len % 4;
 882                     if (cnt) {
 883                             for (; cnt < 4; cnt++) {
 884                                     *(mtod(opts, caddr_t) + opts->m_len) =
 885                                         IPOPT_EOL;
 886                                     opts->m_len++;
 887                             }
 888                     }
 889 #ifdef ICMPPRINTFS
 890                     if (icmpprintfs)
 891                             printf("%d\n", opts->m_len);
 892 #endif
 893                 }
 894                 ip_stripoptions(m);
 895         }
 896         m_tag_delete_nonpersistent(m);
 897         m->m_flags &= ~(M_BCAST|M_MCAST);
 898         icmp_send(m, opts);
 899 done:
 900         if (opts)
 901                 (void)m_free(opts);
 902 }
 903
 904 /*
 905  * Send an icmp packet back to the ip level,
 906  * after supplying a checksum.
 907  */
 908 static void
 909 icmp_send(struct mbuf *m, struct mbuf *opts)
 910 {
 911         struct ip *ip = mtod(m, struct ip *);
 912         int hlen;
 913         struct icmp *icp;
 914
 915         hlen = ip->ip_hl << 2;
 916         m->m_data += hlen;
 917         m->m_len -= hlen;
 918         icp = mtod(m, struct icmp *);
 919         icp->icmp_cksum = 0;
 920         icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen);
 921         m->m_data -= hlen;
 922         m->m_len += hlen;
 923         m->m_pkthdr.rcvif = (struct ifnet *)0;
 924 #ifdef ICMPPRINTFS
 925         if (icmpprintfs) {
 926                 char dstbuf[INET_ADDRSTRLEN];
 927                 char srcbuf[INET_ADDRSTRLEN];
 928
 929                 printf("icmp_send dst %s src %s\n",
 930                     inet_ntoa_r(ip->ip_dst, dstbuf),
 931                     inet_ntoa_r(ip->ip_src, srcbuf));
 932         }
 933 #endif
 934         (void) ip_output(m, opts, NULL, 0, NULL, NULL);
 935 }
 936
 937 /*
 938  * Return milliseconds since 00:00 UTC in network format.
 939  */
 940 uint32_t
 941 iptime(void)
 942 {
 943         struct timeval atv;
 944         u_long t;
 945
 946         getmicrotime(&atv);
 947         t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000;
 948         return (htonl(t));
 949 }
 950
 951 /*
 952  * Return the next larger or smaller MTU plateau (table from RFC 1191)
 953  * given current value MTU.  If DIR is less than zero, a larger plateau
 954  * is returned; otherwise, a smaller value is returned.
 955  */
 956 int
 957 ip_next_mtu(int mtu, int dir)
 958 {
 959         static int mtutab[] = {
 960                 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508,
 961                 296, 68, 0
 962         };
 963         int i, size;
 964
 965         size = (sizeof mtutab) / (sizeof mtutab[0]);
 966         if (dir >= 0) {
 967                 for (i = 0; i < size; i++)
 968                         if (mtu > mtutab[i])
 969                                 return mtutab[i];
 970         } else {
 971                 for (i = size - 1; i >= 0; i--)
 972                         if (mtu < mtutab[i])
 973                                 return mtutab[i];
 974                 if (mtu == mtutab[0])
 975                         return mtutab[0];
 976         }
 977         return 0;
 978 }
 979 #endif /* INET */
 980
 981
 982 /*
 983  * badport_bandlim() - check for ICMP bandwidth limit
 984  *
 985  *      Return 0 if it is ok to send an ICMP error response, -1 if we have
 986  *      hit our bandwidth limit and it is not ok.
 987  *
 988  *      If icmplim is <= 0, the feature is disabled and 0 is returned.
 989  *
 990  *      For now we separate the TCP and UDP subsystems w/ different 'which'
 991  *      values.  We may eventually remove this separation (and simplify the
 992  *      code further).
 993  *
 994  *      Note that the printing of the error message is delayed so we can
 995  *      properly print the icmp error rate that the system was trying to do
 996  *      (i.e. 22000/100 pps, etc...).  This can cause long delays in printing
 997  *      the 'final' error, but it doesn't make sense to solve the printing
 998  *      delay with more complex code.
 999  */
1000 struct icmp_rate {
1001         const char *descr;
1002         struct counter_rate cr;
1003 };
1004 VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = {
1005         { "icmp unreach response" },
1006         { "icmp ping response" },
1007         { "icmp tstamp response" },
1008         { "closed port RST response" },
1009         { "open port RST response" },
1010         { "icmp6 unreach response" },
1011         { "sctp ootb response" }
1012 };
1013 #define V_icmp_rates    VNET(icmp_rates)
1014
1015 static void
1016 icmp_bandlimit_init(void)
1017 {
1018
1019         for (int i = 0; i < BANDLIM_MAX; i++) {
1020                 V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK);
1021                 V_icmp_rates[i].cr.cr_ticks = ticks;
1022         }
1023 }
1024 VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY,
1025     icmp_bandlimit_init, NULL);
1026
1027 static void
1028 icmp_bandlimit_uninit(void)
1029 {
1030
1031         for (int i = 0; i < BANDLIM_MAX; i++)
1032                 counter_u64_free(V_icmp_rates[i].cr.cr_rate);
1033 }
1034 VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1035     icmp_bandlimit_uninit, NULL);
1036
1037 int
1038 badport_bandlim(int which)
1039 {
1040         int64_t pps;
1041
1042         if (V_icmplim == 0 || which == BANDLIM_UNLIMITED)
1043                 return (0);
1044
1045         KASSERT(which >= 0 && which < BANDLIM_MAX,
1046             ("%s: which %d", __func__, which));
1047
1048         pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim);
1049         if (pps == -1)
1050                 return (-1);
1051         if (pps > 0 && V_icmplim_output)
1052                 log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n",
1053                         V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim);
1054         return (0);
1055 }