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