net/ipv4/icmp.c

   1 /*
   2  *      NET3:   Implementation of the ICMP protocol layer.
   3  *
   4  *              Alan Cox, <alan@redhat.com>
   5  *
   6  *      Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Some of the function names and the icmp unreach table for this
  14  *      module were derived from [icmp.c 1.0.11 06/02/93] by
  15  *      Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  16  *      Other than that this module is a complete rewrite.
  17  *
  18  *      Fixes:
  19  *      Clemens Fruhwirth       :       introduce global icmp rate limiting
  20  *                                      with icmp type masking ability instead
  21  *                                      of broken per type icmp timeouts.
  22  *              Mike Shaver     :       RFC1122 checks.
  23  *              Alan Cox        :       Multicast ping reply as self.
  24  *              Alan Cox        :       Fix atomicity lockup in ip_build_xmit
  25  *                                      call.
  26  *              Alan Cox        :       Added 216,128 byte paths to the MTU
  27  *                                      code.
  28  *              Martin Mares    :       RFC1812 checks.
  29  *              Martin Mares    :       Can be configured to follow redirects
  30  *                                      if acting as a router _without_ a
  31  *                                      routing protocol (RFC 1812).
  32  *              Martin Mares    :       Echo requests may be configured to
  33  *                                      be ignored (RFC 1812).
  34  *              Martin Mares    :       Limitation of ICMP error message
  35  *                                      transmit rate (RFC 1812).
  36  *              Martin Mares    :       TOS and Precedence set correctly
  37  *                                      (RFC 1812).
  38  *              Martin Mares    :       Now copying as much data from the
  39  *                                      original packet as we can without
  40  *                                      exceeding 576 bytes (RFC 1812).
  41  *      Willy Konynenberg       :       Transparent proxying support.
  42  *              Keith Owens     :       RFC1191 correction for 4.2BSD based
  43  *                                      path MTU bug.
  44  *              Thomas Quinot   :       ICMP Dest Unreach codes up to 15 are
  45  *                                      valid (RFC 1812).
  46  *              Andi Kleen      :       Check all packet lengths properly
  47  *                                      and moved all kfree_skb() up to
  48  *                                      icmp_rcv.
  49  *              Andi Kleen      :       Move the rate limit bookkeeping
  50  *                                      into the dest entry and use a token
  51  *                                      bucket filter (thanks to ANK). Make
  52  *                                      the rates sysctl configurable.
  53  *              Yu Tianli       :       Fixed two ugly bugs in icmp_send
  54  *                                      - IP option length was accounted wrongly
  55  *                                      - ICMP header length was not accounted
  56  *                                        at all.
  57  *              Tristan Greaves :       Added sysctl option to ignore bogus
  58  *                                      broadcast responses from broken routers.
  59  *
  60  * To Fix:
  61  *
  62  *      - Should use skb_pull() instead of all the manual checking.
  63  *        This would also greatly simply some upper layer error handlers. --AK
  64  *
  65  */
  66
  67 #include <linux/config.h>
  68 #include <linux/module.h>
  69 #include <linux/types.h>
  70 #include <linux/jiffies.h>
  71 #include <linux/kernel.h>
  72 #include <linux/fcntl.h>
  73 #include <linux/socket.h>
  74 #include <linux/in.h>
  75 #include <linux/inet.h>
  76 #include <linux/netdevice.h>
  77 #include <linux/string.h>
  78 #include <linux/netfilter_ipv4.h>
  79 #include <net/snmp.h>
  80 #include <net/ip.h>
  81 #include <net/route.h>
  82 #include <net/protocol.h>
  83 #include <net/icmp.h>
  84 #include <net/tcp.h>
  85 #include <net/udp.h>
  86 #include <net/raw.h>
  87 #include <linux/skbuff.h>
  88 #include <net/sock.h>
  89 #include <linux/errno.h>
  90 #include <linux/timer.h>
  91 #include <linux/init.h>
  92 #include <asm/system.h>
  93 #include <asm/uaccess.h>
  94 #include <net/checksum.h>
  95
  96 /*
  97  *      Build xmit assembly blocks
  98  */
  99
 100 struct icmp_bxm {
 101         struct sk_buff *skb;
 102         int offset;
 103         int data_len;
 104
 105         struct {
 106                 struct icmphdr icmph;
 107                 __u32          times[3];
 108         } data;
 109         int head_len;
 110         struct ip_options replyopts;
 111         unsigned char  optbuf[40];
 112 };
 113
 114 /*
 115  *      Statistics
 116  */
 117 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics);
 118
 119 /* An array of errno for error messages from dest unreach. */
 120 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 121
 122 struct icmp_err icmp_err_convert[] = {
 123         {
 124                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 125                 .fatal = 0,
 126         },
 127         {
 128                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 129                 .fatal = 0,
 130         },
 131         {
 132                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 133                 .fatal = 1,
 134         },
 135         {
 136                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 137                 .fatal = 1,
 138         },
 139         {
 140                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 141                 .fatal = 0,
 142         },
 143         {
 144                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 145                 .fatal = 0,
 146         },
 147         {
 148                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 149                 .fatal = 1,
 150         },
 151         {
 152                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 153                 .fatal = 1,
 154         },
 155         {
 156                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 157                 .fatal = 1,
 158         },
 159         {
 160                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 161                 .fatal = 1,
 162         },
 163         {
 164                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 165                 .fatal = 1,
 166         },
 167         {
 168                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 169                 .fatal = 0,
 170         },
 171         {
 172                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 173                 .fatal = 0,
 174         },
 175         {
 176                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 177                 .fatal = 1,
 178         },
 179         {
 180                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 181                 .fatal = 1,
 182         },
 183         {
 184                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 185                 .fatal = 1,
 186         },
 187 };
 188
 189 /* Control parameters for ECHO replies. */
 190 int sysctl_icmp_echo_ignore_all;
 191 int sysctl_icmp_echo_ignore_broadcasts;
 192
 193 /* Control parameter - ignore bogus broadcast responses? */
 194 int sysctl_icmp_ignore_bogus_error_responses;
 195
 196 /*
 197  *      Configurable global rate limit.
 198  *
 199  *      ratelimit defines tokens/packet consumed for dst->rate_token bucket
 200  *      ratemask defines which icmp types are ratelimited by setting
 201  *      it's bit position.
 202  *
 203  *      default:
 204  *      dest unreachable (3), source quench (4),
 205  *      time exceeded (11), parameter problem (12)
 206  */
 207
 208 int sysctl_icmp_ratelimit = 1 * HZ;
 209 int sysctl_icmp_ratemask = 0x1818;
 210
 211 /*
 212  *      ICMP control array. This specifies what to do with each ICMP.
 213  */
 214
 215 struct icmp_control {
 216         int output_off;         /* Field offset for increment on output */
 217         int input_off;          /* Field offset for increment on input */
 218         void (*handler)(struct sk_buff *skb);
 219         short   error;          /* This ICMP is classed as an error message */
 220 };
 221
 222 static struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 223
 224 /*
 225  *      The ICMP socket(s). This is the most convenient way to flow control
 226  *      our ICMP output as well as maintain a clean interface throughout
 227  *      all layers. All Socketless IP sends will soon be gone.
 228  *
 229  *      On SMP we have one ICMP socket per-cpu.
 230  */
 231 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
 232 #define icmp_socket     __get_cpu_var(__icmp_socket)
 233
 234 static __inline__ int icmp_xmit_lock(void)
 235 {
 236         local_bh_disable();
 237
 238         if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
 239                 /* This can happen if the output path signals a
 240                  * dst_link_failure() for an outgoing ICMP packet.
 241                  */
 242                 local_bh_enable();
 243                 return 1;
 244         }
 245         return 0;
 246 }
 247
 248 static void icmp_xmit_unlock(void)
 249 {
 250         spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
 251 }
 252
 253 /*
 254  *      Send an ICMP frame.
 255  */
 256
 257 /*
 258  *      Check transmit rate limitation for given message.
 259  *      The rate information is held in the destination cache now.
 260  *      This function is generic and could be used for other purposes
 261  *      too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
 262  *
 263  *      Note that the same dst_entry fields are modified by functions in
 264  *      route.c too, but these work for packet destinations while xrlim_allow
 265  *      works for icmp destinations. This means the rate limiting information
 266  *      for one "ip object" is shared - and these ICMPs are twice limited:
 267  *      by source and by destination.
 268  *
 269  *      RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
 270  *                        SHOULD allow setting of rate limits
 271  *
 272  *      Shared between ICMPv4 and ICMPv6.
 273  */
 274 #define XRLIM_BURST_FACTOR 6
 275 int xrlim_allow(struct dst_entry *dst, int timeout)
 276 {
 277         unsigned long now;
 278         int rc = 0;
 279
 280         now = jiffies;
 281         dst->rate_tokens += now - dst->rate_last;
 282         dst->rate_last = now;
 283         if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
 284                 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
 285         if (dst->rate_tokens >= timeout) {
 286                 dst->rate_tokens -= timeout;
 287                 rc = 1;
 288         }
 289         return rc;
 290 }
 291
 292 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
 293 {
 294         struct dst_entry *dst = &rt->u.dst;
 295         int rc = 1;
 296
 297         if (type > NR_ICMP_TYPES)
 298                 goto out;
 299
 300         /* Don't limit PMTU discovery. */
 301         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 302                 goto out;
 303
 304         /* No rate limit on loopback */
 305         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 306                 goto out;
 307
 308         /* Limit if icmp type is enabled in ratemask. */
 309         if ((1 << type) & sysctl_icmp_ratemask)
 310                 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
 311 out:
 312         return rc;
 313 }
 314
 315 /*
 316  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 317  */
 318 static void icmp_out_count(int type)
 319 {
 320         if (type <= NR_ICMP_TYPES) {
 321                 ICMP_INC_STATS_FIELD(icmp_pointers[type].output_off);
 322                 ICMP_INC_STATS(IcmpOutMsgs);
 323         }
 324 }
 325
 326 /*
 327  *      Checksum each fragment, and on the first include the headers and final
 328  *      checksum.
 329  */
 330 int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 331                    struct sk_buff *skb)
 332 {
 333         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 334         unsigned int csum;
 335
 336         csum = skb_copy_and_csum_bits(icmp_param->skb,
 337                                       icmp_param->offset + offset,
 338                                       to, len, 0);
 339
 340         skb->csum = csum_block_add(skb->csum, csum, odd);
 341         return 0;
 342 }
 343
 344 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 345                             struct ipcm_cookie *ipc, struct rtable *rt)
 346 {
 347         struct sk_buff *skb;
 348
 349         ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
 350                        icmp_param->data_len+icmp_param->head_len,
 351                        icmp_param->head_len,
 352                        ipc, rt, MSG_DONTWAIT);
 353
 354         if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
 355                 struct icmphdr *icmph = skb->h.icmph;
 356                 unsigned int csum = 0;
 357                 struct sk_buff *skb1;
 358
 359                 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
 360                         csum = csum_add(csum, skb1->csum);
 361                 }
 362                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 363                                                  (char *)icmph,
 364                                                  icmp_param->head_len, csum);
 365                 icmph->checksum = csum_fold(csum);
 366                 skb->ip_summed = CHECKSUM_NONE;
 367                 ip_push_pending_frames(icmp_socket->sk);
 368         }
 369 }
 370
 371 /*
 372  *      Driving logic for building and sending ICMP messages.
 373  */
 374
 375 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 376 {
 377         struct sock *sk = icmp_socket->sk;
 378         struct inet_opt *inet = inet_sk(sk);
 379         struct ipcm_cookie ipc;
 380         struct rtable *rt = (struct rtable *)skb->dst;
 381         u32 daddr;
 382
 383         if (ip_options_echo(&icmp_param->replyopts, skb))
 384                 goto out;
 385
 386         if (icmp_xmit_lock())
 387                 return;
 388
 389         icmp_param->data.icmph.checksum = 0;
 390         icmp_out_count(icmp_param->data.icmph.type);
 391
 392         inet->tos = skb->nh.iph->tos;
 393         daddr = ipc.addr = rt->rt_src;
 394         ipc.opt = NULL;
 395         if (icmp_param->replyopts.optlen) {
 396                 ipc.opt = &icmp_param->replyopts;
 397                 if (ipc.opt->srr)
 398                         daddr = icmp_param->replyopts.faddr;
 399         }
 400         {
 401                 struct flowi fl = { .nl_u = { .ip4_u =
 402                                               { .daddr = daddr,
 403                                                 .saddr = rt->rt_spec_dst,
 404                                                 .tos = RT_TOS(skb->nh.iph->tos) } },
 405                                     .proto = IPPROTO_ICMP };
 406                 if (ip_route_output_key(&rt, &fl))
 407                         goto out_unlock;
 408         }
 409         if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
 410                                icmp_param->data.icmph.code))
 411                 icmp_push_reply(icmp_param, &ipc, rt);
 412         ip_rt_put(rt);
 413 out_unlock:
 414         icmp_xmit_unlock();
 415 out:;
 416 }
 417
 418
 419 /*
 420  *      Send an ICMP message in response to a situation
 421  *
 422  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 423  *                MAY send more (we do).
 424  *                      MUST NOT change this header information.
 425  *                      MUST NOT reply to a multicast/broadcast IP address.
 426  *                      MUST NOT reply to a multicast/broadcast MAC address.
 427  *                      MUST reply to only the first fragment.
 428  */
 429
 430 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info)
 431 {
 432         struct iphdr *iph;
 433         int room;
 434         struct icmp_bxm icmp_param;
 435         struct rtable *rt = (struct rtable *)skb_in->dst;
 436         struct ipcm_cookie ipc;
 437         u32 saddr;
 438         u8  tos;
 439
 440         if (!rt)
 441                 goto out;
 442
 443         /*
 444          *      Find the original header. It is expected to be valid, of course.
 445          *      Check this, icmp_send is called from the most obscure devices
 446          *      sometimes.
 447          */
 448         iph = skb_in->nh.iph;
 449
 450         if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > skb_in->tail)
 451                 goto out;
 452
 453         /*
 454          *      No replies to physical multicast/broadcast
 455          */
 456         if (skb_in->pkt_type != PACKET_HOST)
 457                 goto out;
 458
 459         /*
 460          *      Now check at the protocol level
 461          */
 462         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 463                 goto out;
 464
 465         /*
 466          *      Only reply to fragment 0. We byte re-order the constant
 467          *      mask for efficiency.
 468          */
 469         if (iph->frag_off & htons(IP_OFFSET))
 470                 goto out;
 471
 472         /*
 473          *      If we send an ICMP error to an ICMP error a mess would result..
 474          */
 475         if (icmp_pointers[type].error) {
 476                 /*
 477                  *      We are an error, check if we are replying to an
 478                  *      ICMP error
 479                  */
 480                 if (iph->protocol == IPPROTO_ICMP) {
 481                         u8 inner_type;
 482
 483                         if (skb_copy_bits(skb_in,
 484                                           skb_in->nh.raw + (iph->ihl << 2) +
 485                                           offsetof(struct icmphdr, type) -
 486                                           skb_in->data, &inner_type, 1))
 487                                 goto out;
 488
 489                         /*
 490                          *      Assume any unknown ICMP type is an error. This
 491                          *      isn't specified by the RFC, but think about it..
 492                          */
 493                         if (inner_type > NR_ICMP_TYPES ||
 494                             icmp_pointers[inner_type].error)
 495                                 goto out;
 496                 }
 497         }
 498
 499         if (icmp_xmit_lock())
 500                 return;
 501
 502         /*
 503          *      Construct source address and options.
 504          */
 505
 506 #ifdef CONFIG_IP_ROUTE_NAT
 507         /*
 508          *      Restore original addresses if packet has been translated.
 509          */
 510         if (rt->rt_flags & RTCF_NAT && IPCB(skb_in)->flags & IPSKB_TRANSLATED) {
 511                 iph->daddr = rt->fl.fl4_dst;
 512                 iph->saddr = rt->fl.fl4_src;
 513         }
 514 #endif
 515
 516         saddr = iph->daddr;
 517         if (!(rt->rt_flags & RTCF_LOCAL))
 518                 saddr = 0;
 519
 520         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 521                                            IPTOS_PREC_INTERNETCONTROL) :
 522                                           iph->tos;
 523
 524         {
 525                 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = iph->saddr,
 526                                                          .saddr = saddr,
 527                                                          .tos = RT_TOS(tos) } },
 528                                     .proto = IPPROTO_ICMP };
 529
 530                 if (ip_route_output_key(&rt, &fl))
 531                     goto out_unlock;
 532         }
 533         if (ip_options_echo(&icmp_param.replyopts, skb_in))
 534                 goto ende;
 535
 536
 537         /*
 538          *      Prepare data for ICMP header.
 539          */
 540
 541         icmp_param.data.icmph.type       = type;
 542         icmp_param.data.icmph.code       = code;
 543         icmp_param.data.icmph.un.gateway = info;
 544         icmp_param.data.icmph.checksum   = 0;
 545         icmp_param.skb    = skb_in;
 546         icmp_param.offset = skb_in->nh.raw - skb_in->data;
 547         icmp_out_count(icmp_param.data.icmph.type);
 548         inet_sk(icmp_socket->sk)->tos = tos;
 549         ipc.addr = iph->saddr;
 550         ipc.opt = &icmp_param.replyopts;
 551         if (icmp_param.replyopts.srr) {
 552                 struct flowi fl = { .nl_u = { .ip4_u =
 553                                               { .daddr = icmp_param.replyopts.faddr,
 554                                                 .saddr = saddr,
 555                                                 .tos = RT_TOS(tos) } },
 556                                     .proto = IPPROTO_ICMP };
 557                 ip_rt_put(rt);
 558                 if (ip_route_output_key(&rt, &fl))
 559                         goto out_unlock;
 560         }
 561
 562         if (!icmpv4_xrlim_allow(rt, type, code))
 563                 goto ende;
 564
 565         /* RFC says return as much as we can without exceeding 576 bytes. */
 566
 567         room = dst_pmtu(&rt->u.dst);
 568         if (room > 576)
 569                 room = 576;
 570         room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
 571         room -= sizeof(struct icmphdr);
 572
 573         icmp_param.data_len = skb_in->len - icmp_param.offset;
 574         if (icmp_param.data_len > room)
 575                 icmp_param.data_len = room;
 576         icmp_param.head_len = sizeof(struct icmphdr);
 577
 578         icmp_push_reply(&icmp_param, &ipc, rt);
 579 ende:
 580         ip_rt_put(rt);
 581 out_unlock:
 582         icmp_xmit_unlock();
 583 out:;
 584 }
 585
 586
 587 /*
 588  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
 589  */
 590
 591 static void icmp_unreach(struct sk_buff *skb)
 592 {
 593         struct iphdr *iph;
 594         struct icmphdr *icmph;
 595         int hash, protocol;
 596         struct inet_protocol *ipprot;
 597         struct sock *raw_sk;
 598         u32 info = 0;
 599
 600         /*
 601          *      Incomplete header ?
 602          *      Only checks for the IP header, there should be an
 603          *      additional check for longer headers in upper levels.
 604          */
 605
 606         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 607                 goto out_err;
 608
 609         icmph = skb->h.icmph;
 610         iph   = (struct iphdr *)skb->data;
 611
 612         if (iph->ihl < 5) /* Mangled header, drop. */
 613                 goto out_err;
 614
 615         if (icmph->type == ICMP_DEST_UNREACH) {
 616                 switch (icmph->code & 15) {
 617                 case ICMP_NET_UNREACH:
 618                 case ICMP_HOST_UNREACH:
 619                 case ICMP_PROT_UNREACH:
 620                 case ICMP_PORT_UNREACH:
 621                         break;
 622                 case ICMP_FRAG_NEEDED:
 623                         if (ipv4_config.no_pmtu_disc) {
 624                                 LIMIT_NETDEBUG(
 625                                         printk(KERN_INFO "ICMP: %u.%u.%u.%u: "
 626                                                          "fragmentation needed "
 627                                                          "and DF set.\n",
 628                                                NIPQUAD(iph->daddr)));
 629                         } else {
 630                                 info = ip_rt_frag_needed(iph,
 631                                                      ntohs(icmph->un.frag.mtu));
 632                                 if (!info)
 633                                         goto out;
 634                         }
 635                         break;
 636                 case ICMP_SR_FAILED:
 637                         LIMIT_NETDEBUG(
 638                                 printk(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
 639                                                  "Route Failed.\n",
 640                                        NIPQUAD(iph->daddr)));
 641                         break;
 642                 default:
 643                         break;
 644                 }
 645                 if (icmph->code > NR_ICMP_UNREACH)
 646                         goto out;
 647         } else if (icmph->type == ICMP_PARAMETERPROB)
 648                 info = ntohl(icmph->un.gateway) >> 24;
 649
 650         /*
 651          *      Throw it at our lower layers
 652          *
 653          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 654          *                header.
 655          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 656          *                transport layer.
 657          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 658          *                transport layer.
 659          */
 660
 661         /*
 662          *      Check the other end isnt violating RFC 1122. Some routers send
 663          *      bogus responses to broadcast frames. If you see this message
 664          *      first check your netmask matches at both ends, if it does then
 665          *      get the other vendor to fix their kit.
 666          */
 667
 668         if (!sysctl_icmp_ignore_bogus_error_responses &&
 669             inet_addr_type(iph->daddr) == RTN_BROADCAST) {
 670                 if (net_ratelimit())
 671                         printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
 672                                             "type %u, code %u "
 673                                             "error to a broadcast: %u.%u.%u.%u on %s\n",
 674                                NIPQUAD(skb->nh.iph->saddr),
 675                                icmph->type, icmph->code,
 676                                NIPQUAD(iph->daddr),
 677                                skb->dev->name);
 678                 goto out;
 679         }
 680
 681         /* Checkin full IP header plus 8 bytes of protocol to
 682          * avoid additional coding at protocol handlers.
 683          */
 684         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
 685                 goto out;
 686
 687         iph = (struct iphdr *)skb->data;
 688         protocol = iph->protocol;
 689
 690         /*
 691          *      Deliver ICMP message to raw sockets. Pretty useless feature?
 692          */
 693
 694         /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
 695         hash = protocol & (MAX_INET_PROTOS - 1);
 696         read_lock(&raw_v4_lock);
 697         if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
 698                 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
 699                                                  iph->saddr,
 700                                                  skb->dev->ifindex)) != NULL) {
 701                         raw_err(raw_sk, skb, info);
 702                         raw_sk = sk_next(raw_sk);
 703                         iph = (struct iphdr *)skb->data;
 704                 }
 705         }
 706         read_unlock(&raw_v4_lock);
 707
 708         rcu_read_lock();
 709         ipprot = inet_protos[hash];
 710         smp_read_barrier_depends();
 711         if (ipprot && ipprot->err_handler)
 712                 ipprot->err_handler(skb, info);
 713         rcu_read_unlock();
 714
 715 out:
 716         return;
 717 out_err:
 718         ICMP_INC_STATS_BH(IcmpInErrors);
 719         goto out;
 720 }
 721
 722
 723 /*
 724  *      Handle ICMP_REDIRECT.
 725  */
 726
 727 static void icmp_redirect(struct sk_buff *skb)
 728 {
 729         struct iphdr *iph;
 730         unsigned long ip;
 731
 732         if (skb->len < sizeof(struct iphdr))
 733                 goto out_err;
 734
 735         /*
 736          *      Get the copied header of the packet that caused the redirect
 737          */
 738         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 739                 goto out;
 740
 741         iph = (struct iphdr *)skb->data;
 742         ip = iph->daddr;
 743
 744         switch (skb->h.icmph->code & 7) {
 745         case ICMP_REDIR_NET:
 746         case ICMP_REDIR_NETTOS:
 747                 /*
 748                  * As per RFC recommendations now handle it as a host redirect.
 749                  */
 750         case ICMP_REDIR_HOST:
 751         case ICMP_REDIR_HOSTTOS:
 752                 ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway,
 753                                iph->saddr, iph->tos, skb->dev);
 754                 break;
 755         }
 756 out:
 757         return;
 758 out_err:
 759         ICMP_INC_STATS_BH(IcmpInErrors);
 760         goto out;
 761 }
 762
 763 /*
 764  *      Handle ICMP_ECHO ("ping") requests.
 765  *
 766  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 767  *                requests.
 768  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 769  *                included in the reply.
 770  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 771  *                echo requests, MUST have default=NOT.
 772  *      See also WRT handling of options once they are done and working.
 773  */
 774
 775 static void icmp_echo(struct sk_buff *skb)
 776 {
 777         if (!sysctl_icmp_echo_ignore_all) {
 778                 struct icmp_bxm icmp_param;
 779
 780                 icmp_param.data.icmph      = *skb->h.icmph;
 781                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 782                 icmp_param.skb             = skb;
 783                 icmp_param.offset          = 0;
 784                 icmp_param.data_len        = skb->len;
 785                 icmp_param.head_len        = sizeof(struct icmphdr);
 786                 icmp_reply(&icmp_param, skb);
 787         }
 788 }
 789
 790 /*
 791  *      Handle ICMP Timestamp requests.
 792  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 793  *                SHOULD be in the kernel for minimum random latency.
 794  *                MUST be accurate to a few minutes.
 795  *                MUST be updated at least at 15Hz.
 796  */
 797 static void icmp_timestamp(struct sk_buff *skb)
 798 {
 799         struct timeval tv;
 800         struct icmp_bxm icmp_param;
 801         /*
 802          *      Too short.
 803          */
 804         if (skb->len < 4)
 805                 goto out_err;
 806
 807         /*
 808          *      Fill in the current time as ms since midnight UT:
 809          */
 810         do_gettimeofday(&tv);
 811         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
 812                                          tv.tv_usec / 1000);
 813         icmp_param.data.times[2] = icmp_param.data.times[1];
 814         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 815                 BUG();
 816         icmp_param.data.icmph      = *skb->h.icmph;
 817         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 818         icmp_param.data.icmph.code = 0;
 819         icmp_param.skb             = skb;
 820         icmp_param.offset          = 0;
 821         icmp_param.data_len        = 0;
 822         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 823         icmp_reply(&icmp_param, skb);
 824 out:
 825         return;
 826 out_err:
 827         ICMP_INC_STATS_BH(IcmpInErrors);
 828         goto out;
 829 }
 830
 831
 832 /*
 833  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
 834  *
 835  * RFC1122 (3.2.2.9).  A host MUST only send replies to
 836  * ADDRESS_MASK requests if it's been configured as an address mask
 837  * agent.  Receiving a request doesn't constitute implicit permission to
 838  * act as one. Of course, implementing this correctly requires (SHOULD)
 839  * a way to turn the functionality on and off.  Another one for sysctl(),
 840  * I guess. -- MS
 841  *
 842  * RFC1812 (4.3.3.9).   A router MUST implement it.
 843  *                      A router SHOULD have switch turning it on/off.
 844  *                      This switch MUST be ON by default.
 845  *
 846  * Gratuitous replies, zero-source replies are not implemented,
 847  * that complies with RFC. DO NOT implement them!!! All the idea
 848  * of broadcast addrmask replies as specified in RFC950 is broken.
 849  * The problem is that it is not uncommon to have several prefixes
 850  * on one physical interface. Moreover, addrmask agent can even be
 851  * not aware of existing another prefixes.
 852  * If source is zero, addrmask agent cannot choose correct prefix.
 853  * Gratuitous mask announcements suffer from the same problem.
 854  * RFC1812 explains it, but still allows to use ADDRMASK,
 855  * that is pretty silly. --ANK
 856  *
 857  * All these rules are so bizarre, that I removed kernel addrmask
 858  * support at all. It is wrong, it is obsolete, nobody uses it in
 859  * any case. --ANK
 860  *
 861  * Furthermore you can do it with a usermode address agent program
 862  * anyway...
 863  */
 864
 865 static void icmp_address(struct sk_buff *skb)
 866 {
 867 #if 0
 868         if (net_ratelimit())
 869                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
 870 #endif
 871 }
 872
 873 /*
 874  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
 875  *                      loudly if an inconsistency is found.
 876  */
 877
 878 static void icmp_address_reply(struct sk_buff *skb)
 879 {
 880         struct rtable *rt = (struct rtable *)skb->dst;
 881         struct net_device *dev = skb->dev;
 882         struct in_device *in_dev;
 883         struct in_ifaddr *ifa;
 884         u32 mask;
 885
 886         if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
 887                 goto out;
 888
 889         in_dev = in_dev_get(dev);
 890         if (!in_dev)
 891                 goto out;
 892         read_lock(&in_dev->lock);
 893         if (in_dev->ifa_list &&
 894             IN_DEV_LOG_MARTIANS(in_dev) &&
 895             IN_DEV_FORWARD(in_dev)) {
 896                 if (skb_copy_bits(skb, 0, &mask, 4))
 897                         BUG();
 898                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
 899                         if (mask == ifa->ifa_mask &&
 900                             inet_ifa_match(rt->rt_src, ifa))
 901                                 break;
 902                 }
 903                 if (!ifa && net_ratelimit()) {
 904                         printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
 905                                          "%s/%u.%u.%u.%u\n",
 906                                NIPQUAD(mask), dev->name, NIPQUAD(rt->rt_src));
 907                 }
 908         }
 909         read_unlock(&in_dev->lock);
 910         in_dev_put(in_dev);
 911 out:;
 912 }
 913
 914 static void icmp_discard(struct sk_buff *skb)
 915 {
 916 }
 917
 918 /*
 919  *      Deal with incoming ICMP packets.
 920  */
 921 int icmp_rcv(struct sk_buff *skb)
 922 {
 923         struct icmphdr *icmph;
 924         struct rtable *rt = (struct rtable *)skb->dst;
 925
 926         ICMP_INC_STATS_BH(IcmpInMsgs);
 927
 928         switch (skb->ip_summed) {
 929         case CHECKSUM_HW:
 930                 if (!(u16)csum_fold(skb->csum))
 931                         break;
 932                 NETDEBUG(if (net_ratelimit())
 933                                 printk(KERN_DEBUG "icmp v4 hw csum failure\n"));
 934         case CHECKSUM_NONE:
 935                 if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0)))
 936                         goto error;
 937         default:;
 938         }
 939
 940         if (!pskb_pull(skb, sizeof(struct icmphdr)))
 941                 goto error;
 942
 943         icmph = skb->h.icmph;
 944
 945         /*
 946          *      18 is the highest 'known' ICMP type. Anything else is a mystery
 947          *
 948          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
 949          *                discarded.
 950          */
 951         if (icmph->type > NR_ICMP_TYPES)
 952                 goto error;
 953
 954
 955         /*
 956          *      Parse the ICMP message
 957          */
 958
 959         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
 960                 /*
 961                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
 962                  *        silently ignored (we let user decide with a sysctl).
 963                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
 964                  *        discarded if to broadcast/multicast.
 965                  */
 966                 if (icmph->type == ICMP_ECHO &&
 967                     sysctl_icmp_echo_ignore_broadcasts) {
 968                         goto error;
 969                 }
 970                 if (icmph->type != ICMP_ECHO &&
 971                     icmph->type != ICMP_TIMESTAMP &&
 972                     icmph->type != ICMP_ADDRESS &&
 973                     icmph->type != ICMP_ADDRESSREPLY) {
 974                         goto error;
 975                 }
 976         }
 977
 978         ICMP_INC_STATS_BH_FIELD(icmp_pointers[icmph->type].input_off);
 979         icmp_pointers[icmph->type].handler(skb);
 980
 981 drop:
 982         kfree_skb(skb);
 983         return 0;
 984 error:
 985         ICMP_INC_STATS_BH(IcmpInErrors);
 986         goto drop;
 987 }
 988
 989 /*
 990  *      This table is the definition of how we handle ICMP.
 991  */
 992 static struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
 993         [ICMP_ECHOREPLY] = {
 994                 .output_off = offsetof(struct icmp_mib, IcmpOutEchoReps),
 995                 .input_off = offsetof(struct icmp_mib, IcmpInEchoReps),
 996                 .handler = icmp_discard,
 997         },
 998         [1] = {
 999                 .output_off = offsetof(struct icmp_mib, dummy),
1000                 .input_off = offsetof(struct icmp_mib,IcmpInErrors),
1001                 .handler = icmp_discard,
1002                 .error = 1,
1003         },
1004         [2] = {
1005                 .output_off = offsetof(struct icmp_mib, dummy),
1006                 .input_off = offsetof(struct icmp_mib,IcmpInErrors),
1007                 .handler = icmp_discard,
1008                 .error = 1,
1009         },
1010         [ICMP_DEST_UNREACH] = {
1011                 .output_off = offsetof(struct icmp_mib, IcmpOutDestUnreachs),
1012                 .input_off = offsetof(struct icmp_mib, IcmpInDestUnreachs),
1013                 .handler = icmp_unreach,
1014                 .error = 1,
1015         },
1016         [ICMP_SOURCE_QUENCH] = {
1017                 .output_off = offsetof(struct icmp_mib, IcmpOutSrcQuenchs),
1018                 .input_off = offsetof(struct icmp_mib, IcmpInSrcQuenchs),
1019                 .handler = icmp_unreach,
1020                 .error = 1,
1021         },
1022         [ICMP_REDIRECT] = {
1023                 .output_off = offsetof(struct icmp_mib, IcmpOutRedirects),
1024                 .input_off = offsetof(struct icmp_mib, IcmpInRedirects),
1025                 .handler = icmp_redirect,
1026                 .error = 1,
1027         },
1028         [6] = {
1029                 .output_off = offsetof(struct icmp_mib, dummy),
1030                 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1031                 .handler = icmp_discard,
1032                 .error = 1,
1033         },
1034         [7] = {
1035                 .output_off = offsetof(struct icmp_mib, dummy),
1036                 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1037                 .handler = icmp_discard,
1038                 .error = 1,
1039         },
1040         [ICMP_ECHO] = {
1041                 .output_off = offsetof(struct icmp_mib, IcmpOutEchos),
1042                 .input_off = offsetof(struct icmp_mib, IcmpInEchos),
1043                 .handler = icmp_echo,
1044         },
1045         [9] = {
1046                 .output_off = offsetof(struct icmp_mib, dummy),
1047                 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1048                 .handler = icmp_discard,
1049                 .error = 1,
1050         },
1051         [10] = {
1052                 .output_off = offsetof(struct icmp_mib, dummy),
1053                 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1054                 .handler = icmp_discard,
1055                 .error = 1,
1056         },
1057         [ICMP_TIME_EXCEEDED] = {
1058                 .output_off = offsetof(struct icmp_mib, IcmpOutTimeExcds),
1059                 .input_off = offsetof(struct icmp_mib,IcmpInTimeExcds),
1060                 .handler = icmp_unreach,
1061                 .error = 1,
1062         },
1063         [ICMP_PARAMETERPROB] = {
1064                 .output_off = offsetof(struct icmp_mib, IcmpOutParmProbs),
1065                 .input_off = offsetof(struct icmp_mib, IcmpInParmProbs),
1066                 .handler = icmp_unreach,
1067                 .error = 1,
1068         },
1069         [ICMP_TIMESTAMP] = {
1070                 .output_off = offsetof(struct icmp_mib, IcmpOutTimestamps),
1071                 .input_off = offsetof(struct icmp_mib, IcmpInTimestamps),
1072                 .handler = icmp_timestamp,
1073         },
1074         [ICMP_TIMESTAMPREPLY] = {
1075                 .output_off = offsetof(struct icmp_mib, IcmpOutTimestampReps),
1076                 .input_off = offsetof(struct icmp_mib, IcmpInTimestampReps),
1077                 .handler = icmp_discard,
1078         },
1079         [ICMP_INFO_REQUEST] = {
1080                 .output_off = offsetof(struct icmp_mib, dummy),
1081                 .input_off = offsetof(struct icmp_mib, dummy),
1082                 .handler = icmp_discard,
1083         },
1084         [ICMP_INFO_REPLY] = {
1085                 .output_off = offsetof(struct icmp_mib, dummy),
1086                 .input_off = offsetof(struct icmp_mib, dummy),
1087                 .handler = icmp_discard,
1088         },
1089         [ICMP_ADDRESS] = {
1090                 .output_off = offsetof(struct icmp_mib, IcmpOutAddrMasks),
1091                 .input_off = offsetof(struct icmp_mib, IcmpInAddrMasks),
1092                 .handler = icmp_address,
1093         },
1094         [ICMP_ADDRESSREPLY] = {
1095                 .output_off = offsetof(struct icmp_mib, IcmpOutAddrMaskReps),
1096                 .input_off = offsetof(struct icmp_mib, IcmpInAddrMaskReps),
1097                 .handler = icmp_address_reply,
1098         },
1099 };
1100
1101 void __init icmp_init(struct net_proto_family *ops)
1102 {
1103         struct inet_opt *inet;
1104         int i;
1105
1106         for (i = 0; i < NR_CPUS; i++) {
1107                 int err;
1108
1109                 if (!cpu_possible(i))
1110                         continue;
1111
1112                 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1113                                        &per_cpu(__icmp_socket, i));
1114
1115                 if (err < 0)
1116                         panic("Failed to create the ICMP control socket.\n");
1117
1118                 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1119
1120                 /* Enough space for 2 64K ICMP packets, including
1121                  * sk_buff struct overhead.
1122                  */
1123                 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1124                         (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1125
1126                 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1127                 inet->uc_ttl = -1;
1128                 inet->pmtudisc = IP_PMTUDISC_DONT;
1129
1130                 /* Unhash it so that IP input processing does not even
1131                  * see it, we do not wish this socket to see incoming
1132                  * packets.
1133                  */
1134                 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1135         }
1136 }
1137
1138 EXPORT_SYMBOL(icmp_err_convert);
1139 EXPORT_SYMBOL(icmp_send);
1140 EXPORT_SYMBOL(icmp_statistics);
1141 EXPORT_SYMBOL(xrlim_allow);