sys/netinet/ip_icmp.c

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