linux/net/ipv4/icmp.c
<<
>>
Prefs
   1/*
   2 *      NET3:   Implementation of the ICMP protocol layer.
   3 *
   4 *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5 *
   6 *      This program is free software; you can redistribute it and/or
   7 *      modify it under the terms of the GNU General Public License
   8 *      as published by the Free Software Foundation; either version
   9 *      2 of the License, or (at your option) any later version.
  10 *
  11 *      Some of the function names and the icmp unreach table for this
  12 *      module were derived from [icmp.c 1.0.11 06/02/93] by
  13 *      Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  14 *      Other than that this module is a complete rewrite.
  15 *
  16 *      Fixes:
  17 *      Clemens Fruhwirth       :       introduce global icmp rate limiting
  18 *                                      with icmp type masking ability instead
  19 *                                      of broken per type icmp timeouts.
  20 *              Mike Shaver     :       RFC1122 checks.
  21 *              Alan Cox        :       Multicast ping reply as self.
  22 *              Alan Cox        :       Fix atomicity lockup in ip_build_xmit
  23 *                                      call.
  24 *              Alan Cox        :       Added 216,128 byte paths to the MTU
  25 *                                      code.
  26 *              Martin Mares    :       RFC1812 checks.
  27 *              Martin Mares    :       Can be configured to follow redirects
  28 *                                      if acting as a router _without_ a
  29 *                                      routing protocol (RFC 1812).
  30 *              Martin Mares    :       Echo requests may be configured to
  31 *                                      be ignored (RFC 1812).
  32 *              Martin Mares    :       Limitation of ICMP error message
  33 *                                      transmit rate (RFC 1812).
  34 *              Martin Mares    :       TOS and Precedence set correctly
  35 *                                      (RFC 1812).
  36 *              Martin Mares    :       Now copying as much data from the
  37 *                                      original packet as we can without
  38 *                                      exceeding 576 bytes (RFC 1812).
  39 *      Willy Konynenberg       :       Transparent proxying support.
  40 *              Keith Owens     :       RFC1191 correction for 4.2BSD based
  41 *                                      path MTU bug.
  42 *              Thomas Quinot   :       ICMP Dest Unreach codes up to 15 are
  43 *                                      valid (RFC 1812).
  44 *              Andi Kleen      :       Check all packet lengths properly
  45 *                                      and moved all kfree_skb() up to
  46 *                                      icmp_rcv.
  47 *              Andi Kleen      :       Move the rate limit bookkeeping
  48 *                                      into the dest entry and use a token
  49 *                                      bucket filter (thanks to ANK). Make
  50 *                                      the rates sysctl configurable.
  51 *              Yu Tianli       :       Fixed two ugly bugs in icmp_send
  52 *                                      - IP option length was accounted wrongly
  53 *                                      - ICMP header length was not accounted
  54 *                                        at all.
  55 *              Tristan Greaves :       Added sysctl option to ignore bogus
  56 *                                      broadcast responses from broken routers.
  57 *
  58 * To Fix:
  59 *
  60 *      - Should use skb_pull() instead of all the manual checking.
  61 *        This would also greatly simply some upper layer error handlers. --AK
  62 *
  63 */
  64
  65#include <linux/module.h>
  66#include <linux/types.h>
  67#include <linux/jiffies.h>
  68#include <linux/kernel.h>
  69#include <linux/fcntl.h>
  70#include <linux/socket.h>
  71#include <linux/in.h>
  72#include <linux/inet.h>
  73#include <linux/inetdevice.h>
  74#include <linux/netdevice.h>
  75#include <linux/string.h>
  76#include <linux/netfilter_ipv4.h>
  77#include <linux/slab.h>
  78#include <net/snmp.h>
  79#include <net/ip.h>
  80#include <net/route.h>
  81#include <net/protocol.h>
  82#include <net/icmp.h>
  83#include <net/tcp.h>
  84#include <net/udp.h>
  85#include <net/raw.h>
  86#include <linux/skbuff.h>
  87#include <net/sock.h>
  88#include <linux/errno.h>
  89#include <linux/timer.h>
  90#include <linux/init.h>
  91#include <asm/system.h>
  92#include <asm/uaccess.h>
  93#include <net/checksum.h>
  94#include <net/xfrm.h>
  95#include <net/inet_common.h>
  96
  97/*
  98 *      Build xmit assembly blocks
  99 */
 100
 101struct icmp_bxm {
 102        struct sk_buff *skb;
 103        int offset;
 104        int data_len;
 105
 106        struct {
 107                struct icmphdr icmph;
 108                __be32         times[3];
 109        } data;
 110        int head_len;
 111        struct ip_options replyopts;
 112        unsigned char  optbuf[40];
 113};
 114
 115/* An array of errno for error messages from dest unreach. */
 116/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 117
 118const struct icmp_err icmp_err_convert[] = {
 119        {
 120                .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 121                .fatal = 0,
 122        },
 123        {
 124                .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 125                .fatal = 0,
 126        },
 127        {
 128                .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 129                .fatal = 1,
 130        },
 131        {
 132                .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 133                .fatal = 1,
 134        },
 135        {
 136                .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 137                .fatal = 0,
 138        },
 139        {
 140                .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 141                .fatal = 0,
 142        },
 143        {
 144                .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 145                .fatal = 1,
 146        },
 147        {
 148                .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 149                .fatal = 1,
 150        },
 151        {
 152                .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 153                .fatal = 1,
 154        },
 155        {
 156                .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 157                .fatal = 1,
 158        },
 159        {
 160                .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 161                .fatal = 1,
 162        },
 163        {
 164                .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 165                .fatal = 0,
 166        },
 167        {
 168                .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 169                .fatal = 0,
 170        },
 171        {
 172                .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 173                .fatal = 1,
 174        },
 175        {
 176                .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 177                .fatal = 1,
 178        },
 179        {
 180                .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 181                .fatal = 1,
 182        },
 183};
 184
 185/*
 186 *      ICMP control array. This specifies what to do with each ICMP.
 187 */
 188
 189struct icmp_control {
 190        void (*handler)(struct sk_buff *skb);
 191        short   error;          /* This ICMP is classed as an error message */
 192};
 193
 194static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 195
 196/*
 197 *      The ICMP socket(s). This is the most convenient way to flow control
 198 *      our ICMP output as well as maintain a clean interface throughout
 199 *      all layers. All Socketless IP sends will soon be gone.
 200 *
 201 *      On SMP we have one ICMP socket per-cpu.
 202 */
 203static struct sock *icmp_sk(struct net *net)
 204{
 205        return net->ipv4.icmp_sk[smp_processor_id()];
 206}
 207
 208static inline struct sock *icmp_xmit_lock(struct net *net)
 209{
 210        struct sock *sk;
 211
 212        local_bh_disable();
 213
 214        sk = icmp_sk(net);
 215
 216        if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 217                /* This can happen if the output path signals a
 218                 * dst_link_failure() for an outgoing ICMP packet.
 219                 */
 220                local_bh_enable();
 221                return NULL;
 222        }
 223        return sk;
 224}
 225
 226static inline void icmp_xmit_unlock(struct sock *sk)
 227{
 228        spin_unlock_bh(&sk->sk_lock.slock);
 229}
 230
 231/*
 232 *      Send an ICMP frame.
 233 */
 234
 235/*
 236 *      Check transmit rate limitation for given message.
 237 *      The rate information is held in the destination cache now.
 238 *      This function is generic and could be used for other purposes
 239 *      too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
 240 *
 241 *      Note that the same dst_entry fields are modified by functions in
 242 *      route.c too, but these work for packet destinations while xrlim_allow
 243 *      works for icmp destinations. This means the rate limiting information
 244 *      for one "ip object" is shared - and these ICMPs are twice limited:
 245 *      by source and by destination.
 246 *
 247 *      RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
 248 *                        SHOULD allow setting of rate limits
 249 *
 250 *      Shared between ICMPv4 and ICMPv6.
 251 */
 252#define XRLIM_BURST_FACTOR 6
 253int xrlim_allow(struct dst_entry *dst, int timeout)
 254{
 255        unsigned long now, token = dst->rate_tokens;
 256        int rc = 0;
 257
 258        now = jiffies;
 259        token += now - dst->rate_last;
 260        dst->rate_last = now;
 261        if (token > XRLIM_BURST_FACTOR * timeout)
 262                token = XRLIM_BURST_FACTOR * timeout;
 263        if (token >= timeout) {
 264                token -= timeout;
 265                rc = 1;
 266        }
 267        dst->rate_tokens = token;
 268        return rc;
 269}
 270
 271static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 272                int type, int code)
 273{
 274        struct dst_entry *dst = &rt->u.dst;
 275        int rc = 1;
 276
 277        if (type > NR_ICMP_TYPES)
 278                goto out;
 279
 280        /* Don't limit PMTU discovery. */
 281        if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 282                goto out;
 283
 284        /* No rate limit on loopback */
 285        if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 286                goto out;
 287
 288        /* Limit if icmp type is enabled in ratemask. */
 289        if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
 290                rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
 291out:
 292        return rc;
 293}
 294
 295/*
 296 *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 297 */
 298void icmp_out_count(struct net *net, unsigned char type)
 299{
 300        ICMPMSGOUT_INC_STATS(net, type);
 301        ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 302}
 303
 304/*
 305 *      Checksum each fragment, and on the first include the headers and final
 306 *      checksum.
 307 */
 308static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 309                          struct sk_buff *skb)
 310{
 311        struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 312        __wsum csum;
 313
 314        csum = skb_copy_and_csum_bits(icmp_param->skb,
 315                                      icmp_param->offset + offset,
 316                                      to, len, 0);
 317
 318        skb->csum = csum_block_add(skb->csum, csum, odd);
 319        if (icmp_pointers[icmp_param->data.icmph.type].error)
 320                nf_ct_attach(skb, icmp_param->skb);
 321        return 0;
 322}
 323
 324static void icmp_push_reply(struct icmp_bxm *icmp_param,
 325                            struct ipcm_cookie *ipc, struct rtable **rt)
 326{
 327        struct sock *sk;
 328        struct sk_buff *skb;
 329
 330        sk = icmp_sk(dev_net((*rt)->u.dst.dev));
 331        if (ip_append_data(sk, icmp_glue_bits, icmp_param,
 332                           icmp_param->data_len+icmp_param->head_len,
 333                           icmp_param->head_len,
 334                           ipc, rt, MSG_DONTWAIT) < 0) {
 335                ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
 336                ip_flush_pending_frames(sk);
 337        } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 338                struct icmphdr *icmph = icmp_hdr(skb);
 339                __wsum csum = 0;
 340                struct sk_buff *skb1;
 341
 342                skb_queue_walk(&sk->sk_write_queue, skb1) {
 343                        csum = csum_add(csum, skb1->csum);
 344                }
 345                csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 346                                                 (char *)icmph,
 347                                                 icmp_param->head_len, csum);
 348                icmph->checksum = csum_fold(csum);
 349                skb->ip_summed = CHECKSUM_NONE;
 350                ip_push_pending_frames(sk);
 351        }
 352}
 353
 354/*
 355 *      Driving logic for building and sending ICMP messages.
 356 */
 357
 358static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 359{
 360        struct ipcm_cookie ipc;
 361        struct rtable *rt = skb_rtable(skb);
 362        struct net *net = dev_net(rt->u.dst.dev);
 363        struct sock *sk;
 364        struct inet_sock *inet;
 365        __be32 daddr;
 366
 367        if (ip_options_echo(&icmp_param->replyopts, skb))
 368                return;
 369
 370        sk = icmp_xmit_lock(net);
 371        if (sk == NULL)
 372                return;
 373        inet = inet_sk(sk);
 374
 375        icmp_param->data.icmph.checksum = 0;
 376
 377        inet->tos = ip_hdr(skb)->tos;
 378        daddr = ipc.addr = rt->rt_src;
 379        ipc.opt = NULL;
 380        ipc.shtx.flags = 0;
 381        if (icmp_param->replyopts.optlen) {
 382                ipc.opt = &icmp_param->replyopts;
 383                if (ipc.opt->srr)
 384                        daddr = icmp_param->replyopts.faddr;
 385        }
 386        {
 387                struct flowi fl = { .nl_u = { .ip4_u =
 388                                              { .daddr = daddr,
 389                                                .saddr = rt->rt_spec_dst,
 390                                                .tos = RT_TOS(ip_hdr(skb)->tos) } },
 391                                    .proto = IPPROTO_ICMP };
 392                security_skb_classify_flow(skb, &fl);
 393                if (ip_route_output_key(net, &rt, &fl))
 394                        goto out_unlock;
 395        }
 396        if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
 397                               icmp_param->data.icmph.code))
 398                icmp_push_reply(icmp_param, &ipc, &rt);
 399        ip_rt_put(rt);
 400out_unlock:
 401        icmp_xmit_unlock(sk);
 402}
 403
 404
 405/*
 406 *      Send an ICMP message in response to a situation
 407 *
 408 *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 409 *                MAY send more (we do).
 410 *                      MUST NOT change this header information.
 411 *                      MUST NOT reply to a multicast/broadcast IP address.
 412 *                      MUST NOT reply to a multicast/broadcast MAC address.
 413 *                      MUST reply to only the first fragment.
 414 */
 415
 416void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 417{
 418        struct iphdr *iph;
 419        int room;
 420        struct icmp_bxm icmp_param;
 421        struct rtable *rt = skb_rtable(skb_in);
 422        struct ipcm_cookie ipc;
 423        __be32 saddr;
 424        u8  tos;
 425        struct net *net;
 426        struct sock *sk;
 427
 428        if (!rt)
 429                goto out;
 430        net = dev_net(rt->u.dst.dev);
 431
 432        /*
 433         *      Find the original header. It is expected to be valid, of course.
 434         *      Check this, icmp_send is called from the most obscure devices
 435         *      sometimes.
 436         */
 437        iph = ip_hdr(skb_in);
 438
 439        if ((u8 *)iph < skb_in->head ||
 440            (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
 441                goto out;
 442
 443        /*
 444         *      No replies to physical multicast/broadcast
 445         */
 446        if (skb_in->pkt_type != PACKET_HOST)
 447                goto out;
 448
 449        /*
 450         *      Now check at the protocol level
 451         */
 452        if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 453                goto out;
 454
 455        /*
 456         *      Only reply to fragment 0. We byte re-order the constant
 457         *      mask for efficiency.
 458         */
 459        if (iph->frag_off & htons(IP_OFFSET))
 460                goto out;
 461
 462        /*
 463         *      If we send an ICMP error to an ICMP error a mess would result..
 464         */
 465        if (icmp_pointers[type].error) {
 466                /*
 467                 *      We are an error, check if we are replying to an
 468                 *      ICMP error
 469                 */
 470                if (iph->protocol == IPPROTO_ICMP) {
 471                        u8 _inner_type, *itp;
 472
 473                        itp = skb_header_pointer(skb_in,
 474                                                 skb_network_header(skb_in) +
 475                                                 (iph->ihl << 2) +
 476                                                 offsetof(struct icmphdr,
 477                                                          type) -
 478                                                 skb_in->data,
 479                                                 sizeof(_inner_type),
 480                                                 &_inner_type);
 481                        if (itp == NULL)
 482                                goto out;
 483
 484                        /*
 485                         *      Assume any unknown ICMP type is an error. This
 486                         *      isn't specified by the RFC, but think about it..
 487                         */
 488                        if (*itp > NR_ICMP_TYPES ||
 489                            icmp_pointers[*itp].error)
 490                                goto out;
 491                }
 492        }
 493
 494        sk = icmp_xmit_lock(net);
 495        if (sk == NULL)
 496                return;
 497
 498        /*
 499         *      Construct source address and options.
 500         */
 501
 502        saddr = iph->daddr;
 503        if (!(rt->rt_flags & RTCF_LOCAL)) {
 504                struct net_device *dev = NULL;
 505
 506                rcu_read_lock();
 507                if (rt->fl.iif &&
 508                        net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 509                        dev = dev_get_by_index_rcu(net, rt->fl.iif);
 510
 511                if (dev)
 512                        saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 513                else
 514                        saddr = 0;
 515                rcu_read_unlock();
 516        }
 517
 518        tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 519                                           IPTOS_PREC_INTERNETCONTROL) :
 520                                          iph->tos;
 521
 522        if (ip_options_echo(&icmp_param.replyopts, skb_in))
 523                goto out_unlock;
 524
 525
 526        /*
 527         *      Prepare data for ICMP header.
 528         */
 529
 530        icmp_param.data.icmph.type       = type;
 531        icmp_param.data.icmph.code       = code;
 532        icmp_param.data.icmph.un.gateway = info;
 533        icmp_param.data.icmph.checksum   = 0;
 534        icmp_param.skb    = skb_in;
 535        icmp_param.offset = skb_network_offset(skb_in);
 536        inet_sk(sk)->tos = tos;
 537        ipc.addr = iph->saddr;
 538        ipc.opt = &icmp_param.replyopts;
 539        ipc.shtx.flags = 0;
 540
 541        {
 542                struct flowi fl = {
 543                        .nl_u = {
 544                                .ip4_u = {
 545                                        .daddr = icmp_param.replyopts.srr ?
 546                                                icmp_param.replyopts.faddr :
 547                                                iph->saddr,
 548                                        .saddr = saddr,
 549                                        .tos = RT_TOS(tos)
 550                                }
 551                        },
 552                        .proto = IPPROTO_ICMP,
 553                        .uli_u = {
 554                                .icmpt = {
 555                                        .type = type,
 556                                        .code = code
 557                                }
 558                        }
 559                };
 560                int err;
 561                struct rtable *rt2;
 562
 563                security_skb_classify_flow(skb_in, &fl);
 564                if (__ip_route_output_key(net, &rt, &fl))
 565                        goto out_unlock;
 566
 567                /* No need to clone since we're just using its address. */
 568                rt2 = rt;
 569
 570                err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
 571                switch (err) {
 572                case 0:
 573                        if (rt != rt2)
 574                                goto route_done;
 575                        break;
 576                case -EPERM:
 577                        rt = NULL;
 578                        break;
 579                default:
 580                        goto out_unlock;
 581                }
 582
 583                if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
 584                        goto relookup_failed;
 585
 586                if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
 587                        err = __ip_route_output_key(net, &rt2, &fl);
 588                else {
 589                        struct flowi fl2 = {};
 590                        unsigned long orefdst;
 591
 592                        fl2.fl4_dst = fl.fl4_src;
 593                        if (ip_route_output_key(net, &rt2, &fl2))
 594                                goto relookup_failed;
 595
 596                        /* Ugh! */
 597                        orefdst = skb_in->_skb_refdst; /* save old refdst */
 598                        err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
 599                                             RT_TOS(tos), rt2->u.dst.dev);
 600
 601                        dst_release(&rt2->u.dst);
 602                        rt2 = skb_rtable(skb_in);
 603                        skb_in->_skb_refdst = orefdst; /* restore old refdst */
 604                }
 605
 606                if (err)
 607                        goto relookup_failed;
 608
 609                err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
 610                                  XFRM_LOOKUP_ICMP);
 611                switch (err) {
 612                case 0:
 613                        dst_release(&rt->u.dst);
 614                        rt = rt2;
 615                        break;
 616                case -EPERM:
 617                        goto ende;
 618                default:
 619relookup_failed:
 620                        if (!rt)
 621                                goto out_unlock;
 622                        break;
 623                }
 624        }
 625
 626route_done:
 627        if (!icmpv4_xrlim_allow(net, rt, type, code))
 628                goto ende;
 629
 630        /* RFC says return as much as we can without exceeding 576 bytes. */
 631
 632        room = dst_mtu(&rt->u.dst);
 633        if (room > 576)
 634                room = 576;
 635        room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
 636        room -= sizeof(struct icmphdr);
 637
 638        icmp_param.data_len = skb_in->len - icmp_param.offset;
 639        if (icmp_param.data_len > room)
 640                icmp_param.data_len = room;
 641        icmp_param.head_len = sizeof(struct icmphdr);
 642
 643        icmp_push_reply(&icmp_param, &ipc, &rt);
 644ende:
 645        ip_rt_put(rt);
 646out_unlock:
 647        icmp_xmit_unlock(sk);
 648out:;
 649}
 650
 651
 652/*
 653 *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
 654 */
 655
 656static void icmp_unreach(struct sk_buff *skb)
 657{
 658        struct iphdr *iph;
 659        struct icmphdr *icmph;
 660        int hash, protocol;
 661        const struct net_protocol *ipprot;
 662        u32 info = 0;
 663        struct net *net;
 664
 665        net = dev_net(skb_dst(skb)->dev);
 666
 667        /*
 668         *      Incomplete header ?
 669         *      Only checks for the IP header, there should be an
 670         *      additional check for longer headers in upper levels.
 671         */
 672
 673        if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 674                goto out_err;
 675
 676        icmph = icmp_hdr(skb);
 677        iph   = (struct iphdr *)skb->data;
 678
 679        if (iph->ihl < 5) /* Mangled header, drop. */
 680                goto out_err;
 681
 682        if (icmph->type == ICMP_DEST_UNREACH) {
 683                switch (icmph->code & 15) {
 684                case ICMP_NET_UNREACH:
 685                case ICMP_HOST_UNREACH:
 686                case ICMP_PROT_UNREACH:
 687                case ICMP_PORT_UNREACH:
 688                        break;
 689                case ICMP_FRAG_NEEDED:
 690                        if (ipv4_config.no_pmtu_disc) {
 691                                LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
 692                                               &iph->daddr);
 693                        } else {
 694                                info = ip_rt_frag_needed(net, iph,
 695                                                         ntohs(icmph->un.frag.mtu),
 696                                                         skb->dev);
 697                                if (!info)
 698                                        goto out;
 699                        }
 700                        break;
 701                case ICMP_SR_FAILED:
 702                        LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
 703                                       &iph->daddr);
 704                        break;
 705                default:
 706                        break;
 707                }
 708                if (icmph->code > NR_ICMP_UNREACH)
 709                        goto out;
 710        } else if (icmph->type == ICMP_PARAMETERPROB)
 711                info = ntohl(icmph->un.gateway) >> 24;
 712
 713        /*
 714         *      Throw it at our lower layers
 715         *
 716         *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 717         *                header.
 718         *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 719         *                transport layer.
 720         *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 721         *                transport layer.
 722         */
 723
 724        /*
 725         *      Check the other end isnt violating RFC 1122. Some routers send
 726         *      bogus responses to broadcast frames. If you see this message
 727         *      first check your netmask matches at both ends, if it does then
 728         *      get the other vendor to fix their kit.
 729         */
 730
 731        if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 732            inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
 733                if (net_ratelimit())
 734                        printk(KERN_WARNING "%pI4 sent an invalid ICMP "
 735                                            "type %u, code %u "
 736                                            "error to a broadcast: %pI4 on %s\n",
 737                               &ip_hdr(skb)->saddr,
 738                               icmph->type, icmph->code,
 739                               &iph->daddr,
 740                               skb->dev->name);
 741                goto out;
 742        }
 743
 744        /* Checkin full IP header plus 8 bytes of protocol to
 745         * avoid additional coding at protocol handlers.
 746         */
 747        if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
 748                goto out;
 749
 750        iph = (struct iphdr *)skb->data;
 751        protocol = iph->protocol;
 752
 753        /*
 754         *      Deliver ICMP message to raw sockets. Pretty useless feature?
 755         */
 756        raw_icmp_error(skb, protocol, info);
 757
 758        hash = protocol & (MAX_INET_PROTOS - 1);
 759        rcu_read_lock();
 760        ipprot = rcu_dereference(inet_protos[hash]);
 761        if (ipprot && ipprot->err_handler)
 762                ipprot->err_handler(skb, info);
 763        rcu_read_unlock();
 764
 765out:
 766        return;
 767out_err:
 768        ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 769        goto out;
 770}
 771
 772
 773/*
 774 *      Handle ICMP_REDIRECT.
 775 */
 776
 777static void icmp_redirect(struct sk_buff *skb)
 778{
 779        struct iphdr *iph;
 780
 781        if (skb->len < sizeof(struct iphdr))
 782                goto out_err;
 783
 784        /*
 785         *      Get the copied header of the packet that caused the redirect
 786         */
 787        if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 788                goto out;
 789
 790        iph = (struct iphdr *)skb->data;
 791
 792        switch (icmp_hdr(skb)->code & 7) {
 793        case ICMP_REDIR_NET:
 794        case ICMP_REDIR_NETTOS:
 795                /*
 796                 * As per RFC recommendations now handle it as a host redirect.
 797                 */
 798        case ICMP_REDIR_HOST:
 799        case ICMP_REDIR_HOSTTOS:
 800                ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
 801                               icmp_hdr(skb)->un.gateway,
 802                               iph->saddr, skb->dev);
 803                break;
 804        }
 805out:
 806        return;
 807out_err:
 808        ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
 809        goto out;
 810}
 811
 812/*
 813 *      Handle ICMP_ECHO ("ping") requests.
 814 *
 815 *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 816 *                requests.
 817 *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 818 *                included in the reply.
 819 *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 820 *                echo requests, MUST have default=NOT.
 821 *      See also WRT handling of options once they are done and working.
 822 */
 823
 824static void icmp_echo(struct sk_buff *skb)
 825{
 826        struct net *net;
 827
 828        net = dev_net(skb_dst(skb)->dev);
 829        if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 830                struct icmp_bxm icmp_param;
 831
 832                icmp_param.data.icmph      = *icmp_hdr(skb);
 833                icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 834                icmp_param.skb             = skb;
 835                icmp_param.offset          = 0;
 836                icmp_param.data_len        = skb->len;
 837                icmp_param.head_len        = sizeof(struct icmphdr);
 838                icmp_reply(&icmp_param, skb);
 839        }
 840}
 841
 842/*
 843 *      Handle ICMP Timestamp requests.
 844 *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 845 *                SHOULD be in the kernel for minimum random latency.
 846 *                MUST be accurate to a few minutes.
 847 *                MUST be updated at least at 15Hz.
 848 */
 849static void icmp_timestamp(struct sk_buff *skb)
 850{
 851        struct timespec tv;
 852        struct icmp_bxm icmp_param;
 853        /*
 854         *      Too short.
 855         */
 856        if (skb->len < 4)
 857                goto out_err;
 858
 859        /*
 860         *      Fill in the current time as ms since midnight UT:
 861         */
 862        getnstimeofday(&tv);
 863        icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
 864                                         tv.tv_nsec / NSEC_PER_MSEC);
 865        icmp_param.data.times[2] = icmp_param.data.times[1];
 866        if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 867                BUG();
 868        icmp_param.data.icmph      = *icmp_hdr(skb);
 869        icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 870        icmp_param.data.icmph.code = 0;
 871        icmp_param.skb             = skb;
 872        icmp_param.offset          = 0;
 873        icmp_param.data_len        = 0;
 874        icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 875        icmp_reply(&icmp_param, skb);
 876out:
 877        return;
 878out_err:
 879        ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 880        goto out;
 881}
 882
 883
 884/*
 885 *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
 886 *
 887 * RFC1122 (3.2.2.9).  A host MUST only send replies to
 888 * ADDRESS_MASK requests if it's been configured as an address mask
 889 * agent.  Receiving a request doesn't constitute implicit permission to
 890 * act as one. Of course, implementing this correctly requires (SHOULD)
 891 * a way to turn the functionality on and off.  Another one for sysctl(),
 892 * I guess. -- MS
 893 *
 894 * RFC1812 (4.3.3.9).   A router MUST implement it.
 895 *                      A router SHOULD have switch turning it on/off.
 896 *                      This switch MUST be ON by default.
 897 *
 898 * Gratuitous replies, zero-source replies are not implemented,
 899 * that complies with RFC. DO NOT implement them!!! All the idea
 900 * of broadcast addrmask replies as specified in RFC950 is broken.
 901 * The problem is that it is not uncommon to have several prefixes
 902 * on one physical interface. Moreover, addrmask agent can even be
 903 * not aware of existing another prefixes.
 904 * If source is zero, addrmask agent cannot choose correct prefix.
 905 * Gratuitous mask announcements suffer from the same problem.
 906 * RFC1812 explains it, but still allows to use ADDRMASK,
 907 * that is pretty silly. --ANK
 908 *
 909 * All these rules are so bizarre, that I removed kernel addrmask
 910 * support at all. It is wrong, it is obsolete, nobody uses it in
 911 * any case. --ANK
 912 *
 913 * Furthermore you can do it with a usermode address agent program
 914 * anyway...
 915 */
 916
 917static void icmp_address(struct sk_buff *skb)
 918{
 919#if 0
 920        if (net_ratelimit())
 921                printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
 922#endif
 923}
 924
 925/*
 926 * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
 927 *                      loudly if an inconsistency is found.
 928 */
 929
 930static void icmp_address_reply(struct sk_buff *skb)
 931{
 932        struct rtable *rt = skb_rtable(skb);
 933        struct net_device *dev = skb->dev;
 934        struct in_device *in_dev;
 935        struct in_ifaddr *ifa;
 936
 937        if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
 938                goto out;
 939
 940        in_dev = in_dev_get(dev);
 941        if (!in_dev)
 942                goto out;
 943        rcu_read_lock();
 944        if (in_dev->ifa_list &&
 945            IN_DEV_LOG_MARTIANS(in_dev) &&
 946            IN_DEV_FORWARD(in_dev)) {
 947                __be32 _mask, *mp;
 948
 949                mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
 950                BUG_ON(mp == NULL);
 951                for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
 952                        if (*mp == ifa->ifa_mask &&
 953                            inet_ifa_match(rt->rt_src, ifa))
 954                                break;
 955                }
 956                if (!ifa && net_ratelimit()) {
 957                        printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
 958                               mp, dev->name, &rt->rt_src);
 959                }
 960        }
 961        rcu_read_unlock();
 962        in_dev_put(in_dev);
 963out:;
 964}
 965
 966static void icmp_discard(struct sk_buff *skb)
 967{
 968}
 969
 970/*
 971 *      Deal with incoming ICMP packets.
 972 */
 973int icmp_rcv(struct sk_buff *skb)
 974{
 975        struct icmphdr *icmph;
 976        struct rtable *rt = skb_rtable(skb);
 977        struct net *net = dev_net(rt->u.dst.dev);
 978
 979        if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 980                struct sec_path *sp = skb_sec_path(skb);
 981                int nh;
 982
 983                if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 984                                 XFRM_STATE_ICMP))
 985                        goto drop;
 986
 987                if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
 988                        goto drop;
 989
 990                nh = skb_network_offset(skb);
 991                skb_set_network_header(skb, sizeof(*icmph));
 992
 993                if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 994                        goto drop;
 995
 996                skb_set_network_header(skb, nh);
 997        }
 998
 999        ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
1000
1001        switch (skb->ip_summed) {
1002        case CHECKSUM_COMPLETE:
1003                if (!csum_fold(skb->csum))
1004                        break;
1005                /* fall through */
1006        case CHECKSUM_NONE:
1007                skb->csum = 0;
1008                if (__skb_checksum_complete(skb))
1009                        goto error;
1010        }
1011
1012        if (!pskb_pull(skb, sizeof(*icmph)))
1013                goto error;
1014
1015        icmph = icmp_hdr(skb);
1016
1017        ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1018        /*
1019         *      18 is the highest 'known' ICMP type. Anything else is a mystery
1020         *
1021         *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1022         *                discarded.
1023         */
1024        if (icmph->type > NR_ICMP_TYPES)
1025                goto error;
1026
1027
1028        /*
1029         *      Parse the ICMP message
1030         */
1031
1032        if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1033                /*
1034                 *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1035                 *        silently ignored (we let user decide with a sysctl).
1036                 *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1037                 *        discarded if to broadcast/multicast.
1038                 */
1039                if ((icmph->type == ICMP_ECHO ||
1040                     icmph->type == ICMP_TIMESTAMP) &&
1041                    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1042                        goto error;
1043                }
1044                if (icmph->type != ICMP_ECHO &&
1045                    icmph->type != ICMP_TIMESTAMP &&
1046                    icmph->type != ICMP_ADDRESS &&
1047                    icmph->type != ICMP_ADDRESSREPLY) {
1048                        goto error;
1049                }
1050        }
1051
1052        icmp_pointers[icmph->type].handler(skb);
1053
1054drop:
1055        kfree_skb(skb);
1056        return 0;
1057error:
1058        ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1059        goto drop;
1060}
1061
1062/*
1063 *      This table is the definition of how we handle ICMP.
1064 */
1065static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1066        [ICMP_ECHOREPLY] = {
1067                .handler = icmp_discard,
1068        },
1069        [1] = {
1070                .handler = icmp_discard,
1071                .error = 1,
1072        },
1073        [2] = {
1074                .handler = icmp_discard,
1075                .error = 1,
1076        },
1077        [ICMP_DEST_UNREACH] = {
1078                .handler = icmp_unreach,
1079                .error = 1,
1080        },
1081        [ICMP_SOURCE_QUENCH] = {
1082                .handler = icmp_unreach,
1083                .error = 1,
1084        },
1085        [ICMP_REDIRECT] = {
1086                .handler = icmp_redirect,
1087                .error = 1,
1088        },
1089        [6] = {
1090                .handler = icmp_discard,
1091                .error = 1,
1092        },
1093        [7] = {
1094                .handler = icmp_discard,
1095                .error = 1,
1096        },
1097        [ICMP_ECHO] = {
1098                .handler = icmp_echo,
1099        },
1100        [9] = {
1101                .handler = icmp_discard,
1102                .error = 1,
1103        },
1104        [10] = {
1105                .handler = icmp_discard,
1106                .error = 1,
1107        },
1108        [ICMP_TIME_EXCEEDED] = {
1109                .handler = icmp_unreach,
1110                .error = 1,
1111        },
1112        [ICMP_PARAMETERPROB] = {
1113                .handler = icmp_unreach,
1114                .error = 1,
1115        },
1116        [ICMP_TIMESTAMP] = {
1117                .handler = icmp_timestamp,
1118        },
1119        [ICMP_TIMESTAMPREPLY] = {
1120                .handler = icmp_discard,
1121        },
1122        [ICMP_INFO_REQUEST] = {
1123                .handler = icmp_discard,
1124        },
1125        [ICMP_INFO_REPLY] = {
1126                .handler = icmp_discard,
1127        },
1128        [ICMP_ADDRESS] = {
1129                .handler = icmp_address,
1130        },
1131        [ICMP_ADDRESSREPLY] = {
1132                .handler = icmp_address_reply,
1133        },
1134};
1135
1136static void __net_exit icmp_sk_exit(struct net *net)
1137{
1138        int i;
1139
1140        for_each_possible_cpu(i)
1141                inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1142        kfree(net->ipv4.icmp_sk);
1143        net->ipv4.icmp_sk = NULL;
1144}
1145
1146static int __net_init icmp_sk_init(struct net *net)
1147{
1148        int i, err;
1149
1150        net->ipv4.icmp_sk =
1151                kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1152        if (net->ipv4.icmp_sk == NULL)
1153                return -ENOMEM;
1154
1155        for_each_possible_cpu(i) {
1156                struct sock *sk;
1157
1158                err = inet_ctl_sock_create(&sk, PF_INET,
1159                                           SOCK_RAW, IPPROTO_ICMP, net);
1160                if (err < 0)
1161                        goto fail;
1162
1163                net->ipv4.icmp_sk[i] = sk;
1164
1165                /* Enough space for 2 64K ICMP packets, including
1166                 * sk_buff struct overhead.
1167                 */
1168                sk->sk_sndbuf =
1169                        (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1170
1171                /*
1172                 * Speedup sock_wfree()
1173                 */
1174                sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1175                inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1176        }
1177
1178        /* Control parameters for ECHO replies. */
1179        net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1180        net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1181
1182        /* Control parameter - ignore bogus broadcast responses? */
1183        net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1184
1185        /*
1186         *      Configurable global rate limit.
1187         *
1188         *      ratelimit defines tokens/packet consumed for dst->rate_token
1189         *      bucket ratemask defines which icmp types are ratelimited by
1190         *      setting it's bit position.
1191         *
1192         *      default:
1193         *      dest unreachable (3), source quench (4),
1194         *      time exceeded (11), parameter problem (12)
1195         */
1196
1197        net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1198        net->ipv4.sysctl_icmp_ratemask = 0x1818;
1199        net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1200
1201        return 0;
1202
1203fail:
1204        for_each_possible_cpu(i)
1205                inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1206        kfree(net->ipv4.icmp_sk);
1207        return err;
1208}
1209
1210static struct pernet_operations __net_initdata icmp_sk_ops = {
1211       .init = icmp_sk_init,
1212       .exit = icmp_sk_exit,
1213};
1214
1215int __init icmp_init(void)
1216{
1217        return register_pernet_subsys(&icmp_sk_ops);
1218}
1219
1220EXPORT_SYMBOL(icmp_err_convert);
1221EXPORT_SYMBOL(icmp_send);
1222EXPORT_SYMBOL(xrlim_allow);
1223