net/ipv4/icmp.c

   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
 101 struct 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
 118 const 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 EXPORT_SYMBOL(icmp_err_convert);
 185
 186 /*
 187  *      ICMP control array. This specifies what to do with each ICMP.
 188  */
 189
 190 struct icmp_control {
 191         void (*handler)(struct sk_buff *skb);
 192         short   error;          /* This ICMP is classed as an error message */
 193 };
 194
 195 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 196
 197 /*
 198  *      The ICMP socket(s). This is the most convenient way to flow control
 199  *      our ICMP output as well as maintain a clean interface throughout
 200  *      all layers. All Socketless IP sends will soon be gone.
 201  *
 202  *      On SMP we have one ICMP socket per-cpu.
 203  */
 204 static struct sock *icmp_sk(struct net *net)
 205 {
 206         return net->ipv4.icmp_sk[smp_processor_id()];
 207 }
 208
 209 static inline struct sock *icmp_xmit_lock(struct net *net)
 210 {
 211         struct sock *sk;
 212
 213         local_bh_disable();
 214
 215         sk = icmp_sk(net);
 216
 217         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 218                 /* This can happen if the output path signals a
 219                  * dst_link_failure() for an outgoing ICMP packet.
 220                  */
 221                 local_bh_enable();
 222                 return NULL;
 223         }
 224         return sk;
 225 }
 226
 227 static inline void icmp_xmit_unlock(struct sock *sk)
 228 {
 229         spin_unlock_bh(&sk->sk_lock.slock);
 230 }
 231
 232 /*
 233  *      Send an ICMP frame.
 234  */
 235
 236 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 237                 int type, int code)
 238 {
 239         struct dst_entry *dst = &rt->dst;
 240         bool rc = true;
 241
 242         if (type > NR_ICMP_TYPES)
 243                 goto out;
 244
 245         /* Don't limit PMTU discovery. */
 246         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 247                 goto out;
 248
 249         /* No rate limit on loopback */
 250         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 251                 goto out;
 252
 253         /* Limit if icmp type is enabled in ratemask. */
 254         if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
 255                 if (!rt->peer)
 256                         rt_bind_peer(rt, 1);
 257                 rc = inet_peer_xrlim_allow(rt->peer,
 258                                            net->ipv4.sysctl_icmp_ratelimit);
 259         }
 260 out:
 261         return rc;
 262 }
 263
 264 /*
 265  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 266  */
 267 void icmp_out_count(struct net *net, unsigned char type)
 268 {
 269         ICMPMSGOUT_INC_STATS(net, type);
 270         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 271 }
 272
 273 /*
 274  *      Checksum each fragment, and on the first include the headers and final
 275  *      checksum.
 276  */
 277 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 278                           struct sk_buff *skb)
 279 {
 280         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 281         __wsum csum;
 282
 283         csum = skb_copy_and_csum_bits(icmp_param->skb,
 284                                       icmp_param->offset + offset,
 285                                       to, len, 0);
 286
 287         skb->csum = csum_block_add(skb->csum, csum, odd);
 288         if (icmp_pointers[icmp_param->data.icmph.type].error)
 289                 nf_ct_attach(skb, icmp_param->skb);
 290         return 0;
 291 }
 292
 293 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 294                             struct ipcm_cookie *ipc, struct rtable **rt)
 295 {
 296         struct sock *sk;
 297         struct sk_buff *skb;
 298
 299         sk = icmp_sk(dev_net((*rt)->dst.dev));
 300         if (ip_append_data(sk, icmp_glue_bits, icmp_param,
 301                            icmp_param->data_len+icmp_param->head_len,
 302                            icmp_param->head_len,
 303                            ipc, rt, MSG_DONTWAIT) < 0) {
 304                 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
 305                 ip_flush_pending_frames(sk);
 306         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 307                 struct icmphdr *icmph = icmp_hdr(skb);
 308                 __wsum csum = 0;
 309                 struct sk_buff *skb1;
 310
 311                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 312                         csum = csum_add(csum, skb1->csum);
 313                 }
 314                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 315                                                  (char *)icmph,
 316                                                  icmp_param->head_len, csum);
 317                 icmph->checksum = csum_fold(csum);
 318                 skb->ip_summed = CHECKSUM_NONE;
 319                 ip_push_pending_frames(sk);
 320         }
 321 }
 322
 323 /*
 324  *      Driving logic for building and sending ICMP messages.
 325  */
 326
 327 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 328 {
 329         struct ipcm_cookie ipc;
 330         struct rtable *rt = skb_rtable(skb);
 331         struct net *net = dev_net(rt->dst.dev);
 332         struct sock *sk;
 333         struct inet_sock *inet;
 334         __be32 daddr;
 335
 336         if (ip_options_echo(&icmp_param->replyopts, skb))
 337                 return;
 338
 339         sk = icmp_xmit_lock(net);
 340         if (sk == NULL)
 341                 return;
 342         inet = inet_sk(sk);
 343
 344         icmp_param->data.icmph.checksum = 0;
 345
 346         inet->tos = ip_hdr(skb)->tos;
 347         daddr = ipc.addr = rt->rt_src;
 348         ipc.opt = NULL;
 349         ipc.tx_flags = 0;
 350         if (icmp_param->replyopts.optlen) {
 351                 ipc.opt = &icmp_param->replyopts;
 352                 if (ipc.opt->srr)
 353                         daddr = icmp_param->replyopts.faddr;
 354         }
 355         {
 356                 struct flowi fl = {
 357                         .fl4_dst = daddr,
 358                         .fl4_src = rt->rt_spec_dst,
 359                         .fl4_tos = RT_TOS(ip_hdr(skb)->tos),
 360                         .flowi_proto = IPPROTO_ICMP,
 361                 };
 362                 security_skb_classify_flow(skb, &fl);
 363                 rt = ip_route_output_key(net, &fl);
 364                 if (IS_ERR(rt))
 365                         goto out_unlock;
 366         }
 367         if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
 368                                icmp_param->data.icmph.code))
 369                 icmp_push_reply(icmp_param, &ipc, &rt);
 370         ip_rt_put(rt);
 371 out_unlock:
 372         icmp_xmit_unlock(sk);
 373 }
 374
 375 static struct rtable *icmp_route_lookup(struct net *net, struct sk_buff *skb_in,
 376                                         struct iphdr *iph,
 377                                         __be32 saddr, u8 tos,
 378                                         int type, int code,
 379                                         struct icmp_bxm *param)
 380 {
 381         struct flowi fl = {
 382                 .fl4_dst = (param->replyopts.srr ?
 383                             param->replyopts.faddr : iph->saddr),
 384                 .fl4_src = saddr,
 385                 .fl4_tos = RT_TOS(tos),
 386                 .flowi_proto = IPPROTO_ICMP,
 387                 .fl_icmp_type = type,
 388                 .fl_icmp_code = code,
 389         };
 390         struct rtable *rt, *rt2;
 391         int err;
 392
 393         security_skb_classify_flow(skb_in, &fl);
 394         rt = __ip_route_output_key(net, &fl);
 395         if (IS_ERR(rt))
 396                 return rt;
 397
 398         /* No need to clone since we're just using its address. */
 399         rt2 = rt;
 400
 401         if (!fl.fl4_src)
 402                 fl.fl4_src = rt->rt_src;
 403
 404         rt = (struct rtable *) xfrm_lookup(net, &rt->dst, &fl, NULL, 0);
 405         if (!IS_ERR(rt)) {
 406                 if (rt != rt2)
 407                         return rt;
 408         } else if (PTR_ERR(rt) == -EPERM) {
 409                 rt = NULL;
 410         } else
 411                 return rt;
 412
 413         err = xfrm_decode_session_reverse(skb_in, &fl, AF_INET);
 414         if (err)
 415                 goto relookup_failed;
 416
 417         if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL) {
 418                 rt2 = __ip_route_output_key(net, &fl);
 419                 if (IS_ERR(rt2))
 420                         err = PTR_ERR(rt2);
 421         } else {
 422                 struct flowi fl2 = {};
 423                 unsigned long orefdst;
 424
 425                 fl2.fl4_dst = fl.fl4_src;
 426                 rt2 = ip_route_output_key(net, &fl2);
 427                 if (IS_ERR(rt2)) {
 428                         err = PTR_ERR(rt2);
 429                         goto relookup_failed;
 430                 }
 431                 /* Ugh! */
 432                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 433                 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
 434                                      RT_TOS(tos), rt2->dst.dev);
 435
 436                 dst_release(&rt2->dst);
 437                 rt2 = skb_rtable(skb_in);
 438                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 439         }
 440
 441         if (err)
 442                 goto relookup_failed;
 443
 444         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, &fl, NULL, XFRM_LOOKUP_ICMP);
 445         if (!IS_ERR(rt2)) {
 446                 dst_release(&rt->dst);
 447                 rt = rt2;
 448         } else if (PTR_ERR(rt2) == -EPERM) {
 449                 if (rt)
 450                         dst_release(&rt->dst);
 451                 return rt2;
 452         } else {
 453                 err = PTR_ERR(rt2);
 454                 goto relookup_failed;
 455         }
 456         return rt;
 457
 458 relookup_failed:
 459         if (rt)
 460                 return rt;
 461         return ERR_PTR(err);
 462 }
 463
 464 /*
 465  *      Send an ICMP message in response to a situation
 466  *
 467  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 468  *                MAY send more (we do).
 469  *                      MUST NOT change this header information.
 470  *                      MUST NOT reply to a multicast/broadcast IP address.
 471  *                      MUST NOT reply to a multicast/broadcast MAC address.
 472  *                      MUST reply to only the first fragment.
 473  */
 474
 475 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 476 {
 477         struct iphdr *iph;
 478         int room;
 479         struct icmp_bxm icmp_param;
 480         struct rtable *rt = skb_rtable(skb_in);
 481         struct ipcm_cookie ipc;
 482         __be32 saddr;
 483         u8  tos;
 484         struct net *net;
 485         struct sock *sk;
 486
 487         if (!rt)
 488                 goto out;
 489         net = dev_net(rt->dst.dev);
 490
 491         /*
 492          *      Find the original header. It is expected to be valid, of course.
 493          *      Check this, icmp_send is called from the most obscure devices
 494          *      sometimes.
 495          */
 496         iph = ip_hdr(skb_in);
 497
 498         if ((u8 *)iph < skb_in->head ||
 499             (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
 500                 goto out;
 501
 502         /*
 503          *      No replies to physical multicast/broadcast
 504          */
 505         if (skb_in->pkt_type != PACKET_HOST)
 506                 goto out;
 507
 508         /*
 509          *      Now check at the protocol level
 510          */
 511         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 512                 goto out;
 513
 514         /*
 515          *      Only reply to fragment 0. We byte re-order the constant
 516          *      mask for efficiency.
 517          */
 518         if (iph->frag_off & htons(IP_OFFSET))
 519                 goto out;
 520
 521         /*
 522          *      If we send an ICMP error to an ICMP error a mess would result..
 523          */
 524         if (icmp_pointers[type].error) {
 525                 /*
 526                  *      We are an error, check if we are replying to an
 527                  *      ICMP error
 528                  */
 529                 if (iph->protocol == IPPROTO_ICMP) {
 530                         u8 _inner_type, *itp;
 531
 532                         itp = skb_header_pointer(skb_in,
 533                                                  skb_network_header(skb_in) +
 534                                                  (iph->ihl << 2) +
 535                                                  offsetof(struct icmphdr,
 536                                                           type) -
 537                                                  skb_in->data,
 538                                                  sizeof(_inner_type),
 539                                                  &_inner_type);
 540                         if (itp == NULL)
 541                                 goto out;
 542
 543                         /*
 544                          *      Assume any unknown ICMP type is an error. This
 545                          *      isn't specified by the RFC, but think about it..
 546                          */
 547                         if (*itp > NR_ICMP_TYPES ||
 548                             icmp_pointers[*itp].error)
 549                                 goto out;
 550                 }
 551         }
 552
 553         sk = icmp_xmit_lock(net);
 554         if (sk == NULL)
 555                 return;
 556
 557         /*
 558          *      Construct source address and options.
 559          */
 560
 561         saddr = iph->daddr;
 562         if (!(rt->rt_flags & RTCF_LOCAL)) {
 563                 struct net_device *dev = NULL;
 564
 565                 rcu_read_lock();
 566                 if (rt_is_input_route(rt) &&
 567                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 568                         dev = dev_get_by_index_rcu(net, rt->rt_iif);
 569
 570                 if (dev)
 571                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 572                 else
 573                         saddr = 0;
 574                 rcu_read_unlock();
 575         }
 576
 577         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 578                                            IPTOS_PREC_INTERNETCONTROL) :
 579                                           iph->tos;
 580
 581         if (ip_options_echo(&icmp_param.replyopts, skb_in))
 582                 goto out_unlock;
 583
 584
 585         /*
 586          *      Prepare data for ICMP header.
 587          */
 588
 589         icmp_param.data.icmph.type       = type;
 590         icmp_param.data.icmph.code       = code;
 591         icmp_param.data.icmph.un.gateway = info;
 592         icmp_param.data.icmph.checksum   = 0;
 593         icmp_param.skb    = skb_in;
 594         icmp_param.offset = skb_network_offset(skb_in);
 595         inet_sk(sk)->tos = tos;
 596         ipc.addr = iph->saddr;
 597         ipc.opt = &icmp_param.replyopts;
 598         ipc.tx_flags = 0;
 599
 600         rt = icmp_route_lookup(net, skb_in, iph, saddr, tos,
 601                                type, code, &icmp_param);
 602         if (IS_ERR(rt))
 603                 goto out_unlock;
 604
 605         if (!icmpv4_xrlim_allow(net, rt, type, code))
 606                 goto ende;
 607
 608         /* RFC says return as much as we can without exceeding 576 bytes. */
 609
 610         room = dst_mtu(&rt->dst);
 611         if (room > 576)
 612                 room = 576;
 613         room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
 614         room -= sizeof(struct icmphdr);
 615
 616         icmp_param.data_len = skb_in->len - icmp_param.offset;
 617         if (icmp_param.data_len > room)
 618                 icmp_param.data_len = room;
 619         icmp_param.head_len = sizeof(struct icmphdr);
 620
 621         icmp_push_reply(&icmp_param, &ipc, &rt);
 622 ende:
 623         ip_rt_put(rt);
 624 out_unlock:
 625         icmp_xmit_unlock(sk);
 626 out:;
 627 }
 628 EXPORT_SYMBOL(icmp_send);
 629
 630
 631 /*
 632  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
 633  */
 634
 635 static void icmp_unreach(struct sk_buff *skb)
 636 {
 637         struct iphdr *iph;
 638         struct icmphdr *icmph;
 639         int hash, protocol;
 640         const struct net_protocol *ipprot;
 641         u32 info = 0;
 642         struct net *net;
 643
 644         net = dev_net(skb_dst(skb)->dev);
 645
 646         /*
 647          *      Incomplete header ?
 648          *      Only checks for the IP header, there should be an
 649          *      additional check for longer headers in upper levels.
 650          */
 651
 652         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 653                 goto out_err;
 654
 655         icmph = icmp_hdr(skb);
 656         iph   = (struct iphdr *)skb->data;
 657
 658         if (iph->ihl < 5) /* Mangled header, drop. */
 659                 goto out_err;
 660
 661         if (icmph->type == ICMP_DEST_UNREACH) {
 662                 switch (icmph->code & 15) {
 663                 case ICMP_NET_UNREACH:
 664                 case ICMP_HOST_UNREACH:
 665                 case ICMP_PROT_UNREACH:
 666                 case ICMP_PORT_UNREACH:
 667                         break;
 668                 case ICMP_FRAG_NEEDED:
 669                         if (ipv4_config.no_pmtu_disc) {
 670                                 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
 671                                                &iph->daddr);
 672                         } else {
 673                                 info = ip_rt_frag_needed(net, iph,
 674                                                          ntohs(icmph->un.frag.mtu),
 675                                                          skb->dev);
 676                                 if (!info)
 677                                         goto out;
 678                         }
 679                         break;
 680                 case ICMP_SR_FAILED:
 681                         LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
 682                                        &iph->daddr);
 683                         break;
 684                 default:
 685                         break;
 686                 }
 687                 if (icmph->code > NR_ICMP_UNREACH)
 688                         goto out;
 689         } else if (icmph->type == ICMP_PARAMETERPROB)
 690                 info = ntohl(icmph->un.gateway) >> 24;
 691
 692         /*
 693          *      Throw it at our lower layers
 694          *
 695          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 696          *                header.
 697          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 698          *                transport layer.
 699          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 700          *                transport layer.
 701          */
 702
 703         /*
 704          *      Check the other end isnt violating RFC 1122. Some routers send
 705          *      bogus responses to broadcast frames. If you see this message
 706          *      first check your netmask matches at both ends, if it does then
 707          *      get the other vendor to fix their kit.
 708          */
 709
 710         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 711             inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
 712                 if (net_ratelimit())
 713                         printk(KERN_WARNING "%pI4 sent an invalid ICMP "
 714                                             "type %u, code %u "
 715                                             "error to a broadcast: %pI4 on %s\n",
 716                                &ip_hdr(skb)->saddr,
 717                                icmph->type, icmph->code,
 718                                &iph->daddr,
 719                                skb->dev->name);
 720                 goto out;
 721         }
 722
 723         /* Checkin full IP header plus 8 bytes of protocol to
 724          * avoid additional coding at protocol handlers.
 725          */
 726         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
 727                 goto out;
 728
 729         iph = (struct iphdr *)skb->data;
 730         protocol = iph->protocol;
 731
 732         /*
 733          *      Deliver ICMP message to raw sockets. Pretty useless feature?
 734          */
 735         raw_icmp_error(skb, protocol, info);
 736
 737         hash = protocol & (MAX_INET_PROTOS - 1);
 738         rcu_read_lock();
 739         ipprot = rcu_dereference(inet_protos[hash]);
 740         if (ipprot && ipprot->err_handler)
 741                 ipprot->err_handler(skb, info);
 742         rcu_read_unlock();
 743
 744 out:
 745         return;
 746 out_err:
 747         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 748         goto out;
 749 }
 750
 751
 752 /*
 753  *      Handle ICMP_REDIRECT.
 754  */
 755
 756 static void icmp_redirect(struct sk_buff *skb)
 757 {
 758         struct iphdr *iph;
 759
 760         if (skb->len < sizeof(struct iphdr))
 761                 goto out_err;
 762
 763         /*
 764          *      Get the copied header of the packet that caused the redirect
 765          */
 766         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 767                 goto out;
 768
 769         iph = (struct iphdr *)skb->data;
 770
 771         switch (icmp_hdr(skb)->code & 7) {
 772         case ICMP_REDIR_NET:
 773         case ICMP_REDIR_NETTOS:
 774                 /*
 775                  * As per RFC recommendations now handle it as a host redirect.
 776                  */
 777         case ICMP_REDIR_HOST:
 778         case ICMP_REDIR_HOSTTOS:
 779                 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
 780                                icmp_hdr(skb)->un.gateway,
 781                                iph->saddr, skb->dev);
 782                 break;
 783         }
 784 out:
 785         return;
 786 out_err:
 787         ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
 788         goto out;
 789 }
 790
 791 /*
 792  *      Handle ICMP_ECHO ("ping") requests.
 793  *
 794  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 795  *                requests.
 796  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 797  *                included in the reply.
 798  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 799  *                echo requests, MUST have default=NOT.
 800  *      See also WRT handling of options once they are done and working.
 801  */
 802
 803 static void icmp_echo(struct sk_buff *skb)
 804 {
 805         struct net *net;
 806
 807         net = dev_net(skb_dst(skb)->dev);
 808         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 809                 struct icmp_bxm icmp_param;
 810
 811                 icmp_param.data.icmph      = *icmp_hdr(skb);
 812                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 813                 icmp_param.skb             = skb;
 814                 icmp_param.offset          = 0;
 815                 icmp_param.data_len        = skb->len;
 816                 icmp_param.head_len        = sizeof(struct icmphdr);
 817                 icmp_reply(&icmp_param, skb);
 818         }
 819 }
 820
 821 /*
 822  *      Handle ICMP Timestamp requests.
 823  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 824  *                SHOULD be in the kernel for minimum random latency.
 825  *                MUST be accurate to a few minutes.
 826  *                MUST be updated at least at 15Hz.
 827  */
 828 static void icmp_timestamp(struct sk_buff *skb)
 829 {
 830         struct timespec tv;
 831         struct icmp_bxm icmp_param;
 832         /*
 833          *      Too short.
 834          */
 835         if (skb->len < 4)
 836                 goto out_err;
 837
 838         /*
 839          *      Fill in the current time as ms since midnight UT:
 840          */
 841         getnstimeofday(&tv);
 842         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
 843                                          tv.tv_nsec / NSEC_PER_MSEC);
 844         icmp_param.data.times[2] = icmp_param.data.times[1];
 845         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 846                 BUG();
 847         icmp_param.data.icmph      = *icmp_hdr(skb);
 848         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 849         icmp_param.data.icmph.code = 0;
 850         icmp_param.skb             = skb;
 851         icmp_param.offset          = 0;
 852         icmp_param.data_len        = 0;
 853         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 854         icmp_reply(&icmp_param, skb);
 855 out:
 856         return;
 857 out_err:
 858         ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 859         goto out;
 860 }
 861
 862
 863 /*
 864  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
 865  *
 866  * RFC1122 (3.2.2.9).  A host MUST only send replies to
 867  * ADDRESS_MASK requests if it's been configured as an address mask
 868  * agent.  Receiving a request doesn't constitute implicit permission to
 869  * act as one. Of course, implementing this correctly requires (SHOULD)
 870  * a way to turn the functionality on and off.  Another one for sysctl(),
 871  * I guess. -- MS
 872  *
 873  * RFC1812 (4.3.3.9).   A router MUST implement it.
 874  *                      A router SHOULD have switch turning it on/off.
 875  *                      This switch MUST be ON by default.
 876  *
 877  * Gratuitous replies, zero-source replies are not implemented,
 878  * that complies with RFC. DO NOT implement them!!! All the idea
 879  * of broadcast addrmask replies as specified in RFC950 is broken.
 880  * The problem is that it is not uncommon to have several prefixes
 881  * on one physical interface. Moreover, addrmask agent can even be
 882  * not aware of existing another prefixes.
 883  * If source is zero, addrmask agent cannot choose correct prefix.
 884  * Gratuitous mask announcements suffer from the same problem.
 885  * RFC1812 explains it, but still allows to use ADDRMASK,
 886  * that is pretty silly. --ANK
 887  *
 888  * All these rules are so bizarre, that I removed kernel addrmask
 889  * support at all. It is wrong, it is obsolete, nobody uses it in
 890  * any case. --ANK
 891  *
 892  * Furthermore you can do it with a usermode address agent program
 893  * anyway...
 894  */
 895
 896 static void icmp_address(struct sk_buff *skb)
 897 {
 898 #if 0
 899         if (net_ratelimit())
 900                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
 901 #endif
 902 }
 903
 904 /*
 905  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
 906  *                      loudly if an inconsistency is found.
 907  * called with rcu_read_lock()
 908  */
 909
 910 static void icmp_address_reply(struct sk_buff *skb)
 911 {
 912         struct rtable *rt = skb_rtable(skb);
 913         struct net_device *dev = skb->dev;
 914         struct in_device *in_dev;
 915         struct in_ifaddr *ifa;
 916
 917         if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
 918                 return;
 919
 920         in_dev = __in_dev_get_rcu(dev);
 921         if (!in_dev)
 922                 return;
 923
 924         if (in_dev->ifa_list &&
 925             IN_DEV_LOG_MARTIANS(in_dev) &&
 926             IN_DEV_FORWARD(in_dev)) {
 927                 __be32 _mask, *mp;
 928
 929                 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
 930                 BUG_ON(mp == NULL);
 931                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
 932                         if (*mp == ifa->ifa_mask &&
 933                             inet_ifa_match(rt->rt_src, ifa))
 934                                 break;
 935                 }
 936                 if (!ifa && net_ratelimit()) {
 937                         printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
 938                                mp, dev->name, &rt->rt_src);
 939                 }
 940         }
 941 }
 942
 943 static void icmp_discard(struct sk_buff *skb)
 944 {
 945 }
 946
 947 /*
 948  *      Deal with incoming ICMP packets.
 949  */
 950 int icmp_rcv(struct sk_buff *skb)
 951 {
 952         struct icmphdr *icmph;
 953         struct rtable *rt = skb_rtable(skb);
 954         struct net *net = dev_net(rt->dst.dev);
 955
 956         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 957                 struct sec_path *sp = skb_sec_path(skb);
 958                 int nh;
 959
 960                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 961                                  XFRM_STATE_ICMP))
 962                         goto drop;
 963
 964                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
 965                         goto drop;
 966
 967                 nh = skb_network_offset(skb);
 968                 skb_set_network_header(skb, sizeof(*icmph));
 969
 970                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 971                         goto drop;
 972
 973                 skb_set_network_header(skb, nh);
 974         }
 975
 976         ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
 977
 978         switch (skb->ip_summed) {
 979         case CHECKSUM_COMPLETE:
 980                 if (!csum_fold(skb->csum))
 981                         break;
 982                 /* fall through */
 983         case CHECKSUM_NONE:
 984                 skb->csum = 0;
 985                 if (__skb_checksum_complete(skb))
 986                         goto error;
 987         }
 988
 989         if (!pskb_pull(skb, sizeof(*icmph)))
 990                 goto error;
 991
 992         icmph = icmp_hdr(skb);
 993
 994         ICMPMSGIN_INC_STATS_BH(net, icmph->type);
 995         /*
 996          *      18 is the highest 'known' ICMP type. Anything else is a mystery
 997          *
 998          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
 999          *                discarded.
1000          */
1001         if (icmph->type > NR_ICMP_TYPES)
1002                 goto error;
1003
1004
1005         /*
1006          *      Parse the ICMP message
1007          */
1008
1009         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1010                 /*
1011                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1012                  *        silently ignored (we let user decide with a sysctl).
1013                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1014                  *        discarded if to broadcast/multicast.
1015                  */
1016                 if ((icmph->type == ICMP_ECHO ||
1017                      icmph->type == ICMP_TIMESTAMP) &&
1018                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1019                         goto error;
1020                 }
1021                 if (icmph->type != ICMP_ECHO &&
1022                     icmph->type != ICMP_TIMESTAMP &&
1023                     icmph->type != ICMP_ADDRESS &&
1024                     icmph->type != ICMP_ADDRESSREPLY) {
1025                         goto error;
1026                 }
1027         }
1028
1029         icmp_pointers[icmph->type].handler(skb);
1030
1031 drop:
1032         kfree_skb(skb);
1033         return 0;
1034 error:
1035         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1036         goto drop;
1037 }
1038
1039 /*
1040  *      This table is the definition of how we handle ICMP.
1041  */
1042 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1043         [ICMP_ECHOREPLY] = {
1044                 .handler = icmp_discard,
1045         },
1046         [1] = {
1047                 .handler = icmp_discard,
1048                 .error = 1,
1049         },
1050         [2] = {
1051                 .handler = icmp_discard,
1052                 .error = 1,
1053         },
1054         [ICMP_DEST_UNREACH] = {
1055                 .handler = icmp_unreach,
1056                 .error = 1,
1057         },
1058         [ICMP_SOURCE_QUENCH] = {
1059                 .handler = icmp_unreach,
1060                 .error = 1,
1061         },
1062         [ICMP_REDIRECT] = {
1063                 .handler = icmp_redirect,
1064                 .error = 1,
1065         },
1066         [6] = {
1067                 .handler = icmp_discard,
1068                 .error = 1,
1069         },
1070         [7] = {
1071                 .handler = icmp_discard,
1072                 .error = 1,
1073         },
1074         [ICMP_ECHO] = {
1075                 .handler = icmp_echo,
1076         },
1077         [9] = {
1078                 .handler = icmp_discard,
1079                 .error = 1,
1080         },
1081         [10] = {
1082                 .handler = icmp_discard,
1083                 .error = 1,
1084         },
1085         [ICMP_TIME_EXCEEDED] = {
1086                 .handler = icmp_unreach,
1087                 .error = 1,
1088         },
1089         [ICMP_PARAMETERPROB] = {
1090                 .handler = icmp_unreach,
1091                 .error = 1,
1092         },
1093         [ICMP_TIMESTAMP] = {
1094                 .handler = icmp_timestamp,
1095         },
1096         [ICMP_TIMESTAMPREPLY] = {
1097                 .handler = icmp_discard,
1098         },
1099         [ICMP_INFO_REQUEST] = {
1100                 .handler = icmp_discard,
1101         },
1102         [ICMP_INFO_REPLY] = {
1103                 .handler = icmp_discard,
1104         },
1105         [ICMP_ADDRESS] = {
1106                 .handler = icmp_address,
1107         },
1108         [ICMP_ADDRESSREPLY] = {
1109                 .handler = icmp_address_reply,
1110         },
1111 };
1112
1113 static void __net_exit icmp_sk_exit(struct net *net)
1114 {
1115         int i;
1116
1117         for_each_possible_cpu(i)
1118                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1119         kfree(net->ipv4.icmp_sk);
1120         net->ipv4.icmp_sk = NULL;
1121 }
1122
1123 static int __net_init icmp_sk_init(struct net *net)
1124 {
1125         int i, err;
1126
1127         net->ipv4.icmp_sk =
1128                 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1129         if (net->ipv4.icmp_sk == NULL)
1130                 return -ENOMEM;
1131
1132         for_each_possible_cpu(i) {
1133                 struct sock *sk;
1134
1135                 err = inet_ctl_sock_create(&sk, PF_INET,
1136                                            SOCK_RAW, IPPROTO_ICMP, net);
1137                 if (err < 0)
1138                         goto fail;
1139
1140                 net->ipv4.icmp_sk[i] = sk;
1141
1142                 /* Enough space for 2 64K ICMP packets, including
1143                  * sk_buff struct overhead.
1144                  */
1145                 sk->sk_sndbuf =
1146                         (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1147
1148                 /*
1149                  * Speedup sock_wfree()
1150                  */
1151                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1152                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1153         }
1154
1155         /* Control parameters for ECHO replies. */
1156         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1157         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1158
1159         /* Control parameter - ignore bogus broadcast responses? */
1160         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1161
1162         /*
1163          *      Configurable global rate limit.
1164          *
1165          *      ratelimit defines tokens/packet consumed for dst->rate_token
1166          *      bucket ratemask defines which icmp types are ratelimited by
1167          *      setting it's bit position.
1168          *
1169          *      default:
1170          *      dest unreachable (3), source quench (4),
1171          *      time exceeded (11), parameter problem (12)
1172          */
1173
1174         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1175         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1176         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1177
1178         return 0;
1179
1180 fail:
1181         for_each_possible_cpu(i)
1182                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1183         kfree(net->ipv4.icmp_sk);
1184         return err;
1185 }
1186
1187 static struct pernet_operations __net_initdata icmp_sk_ops = {
1188        .init = icmp_sk_init,
1189        .exit = icmp_sk_exit,
1190 };
1191
1192 int __init icmp_init(void)
1193 {
1194         return register_pernet_subsys(&icmp_sk_ops);
1195 }