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 /*
 237  *      Check transmit rate limitation for given message.
 238  *      The rate information is held in the destination cache now.
 239  *      This function is generic and could be used for other purposes
 240  *      too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
 241  *
 242  *      Note that the same dst_entry fields are modified by functions in
 243  *      route.c too, but these work for packet destinations while xrlim_allow
 244  *      works for icmp destinations. This means the rate limiting information
 245  *      for one "ip object" is shared - and these ICMPs are twice limited:
 246  *      by source and by destination.
 247  *
 248  *      RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
 249  *                        SHOULD allow setting of rate limits
 250  *
 251  *      Shared between ICMPv4 and ICMPv6.
 252  */
 253 #define XRLIM_BURST_FACTOR 6
 254 int xrlim_allow(struct dst_entry *dst, int timeout)
 255 {
 256         unsigned long now, token = dst->rate_tokens;
 257         int rc = 0;
 258
 259         now = jiffies;
 260         token += now - dst->rate_last;
 261         dst->rate_last = now;
 262         if (token > XRLIM_BURST_FACTOR * timeout)
 263                 token = XRLIM_BURST_FACTOR * timeout;
 264         if (token >= timeout) {
 265                 token -= timeout;
 266                 rc = 1;
 267         }
 268         dst->rate_tokens = token;
 269         return rc;
 270 }
 271 EXPORT_SYMBOL(xrlim_allow);
 272
 273 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 274                 int type, int code)
 275 {
 276         struct dst_entry *dst = &rt->dst;
 277         int rc = 1;
 278
 279         if (type > NR_ICMP_TYPES)
 280                 goto out;
 281
 282         /* Don't limit PMTU discovery. */
 283         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 284                 goto out;
 285
 286         /* No rate limit on loopback */
 287         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 288                 goto out;
 289
 290         /* Limit if icmp type is enabled in ratemask. */
 291         if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
 292                 rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
 293 out:
 294         return rc;
 295 }
 296
 297 /*
 298  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 299  */
 300 void icmp_out_count(struct net *net, unsigned char type)
 301 {
 302         ICMPMSGOUT_INC_STATS(net, type);
 303         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 304 }
 305
 306 /*
 307  *      Checksum each fragment, and on the first include the headers and final
 308  *      checksum.
 309  */
 310 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 311                           struct sk_buff *skb)
 312 {
 313         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 314         __wsum csum;
 315
 316         csum = skb_copy_and_csum_bits(icmp_param->skb,
 317                                       icmp_param->offset + offset,
 318                                       to, len, 0);
 319
 320         skb->csum = csum_block_add(skb->csum, csum, odd);
 321         if (icmp_pointers[icmp_param->data.icmph.type].error)
 322                 nf_ct_attach(skb, icmp_param->skb);
 323         return 0;
 324 }
 325
 326 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 327                             struct ipcm_cookie *ipc, struct rtable **rt)
 328 {
 329         struct sock *sk;
 330         struct sk_buff *skb;
 331
 332         sk = icmp_sk(dev_net((*rt)->dst.dev));
 333         if (ip_append_data(sk, icmp_glue_bits, icmp_param,
 334                            icmp_param->data_len+icmp_param->head_len,
 335                            icmp_param->head_len,
 336                            ipc, rt, MSG_DONTWAIT) < 0) {
 337                 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
 338                 ip_flush_pending_frames(sk);
 339         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 340                 struct icmphdr *icmph = icmp_hdr(skb);
 341                 __wsum csum = 0;
 342                 struct sk_buff *skb1;
 343
 344                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 345                         csum = csum_add(csum, skb1->csum);
 346                 }
 347                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 348                                                  (char *)icmph,
 349                                                  icmp_param->head_len, csum);
 350                 icmph->checksum = csum_fold(csum);
 351                 skb->ip_summed = CHECKSUM_NONE;
 352                 ip_push_pending_frames(sk);
 353         }
 354 }
 355
 356 /*
 357  *      Driving logic for building and sending ICMP messages.
 358  */
 359
 360 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 361 {
 362         struct ipcm_cookie ipc;
 363         struct rtable *rt = skb_rtable(skb);
 364         struct net *net = dev_net(rt->dst.dev);
 365         struct sock *sk;
 366         struct inet_sock *inet;
 367         __be32 daddr;
 368
 369         if (ip_options_echo(&icmp_param->replyopts, skb))
 370                 return;
 371
 372         sk = icmp_xmit_lock(net);
 373         if (sk == NULL)
 374                 return;
 375         inet = inet_sk(sk);
 376
 377         icmp_param->data.icmph.checksum = 0;
 378
 379         inet->tos = ip_hdr(skb)->tos;
 380         daddr = ipc.addr = rt->rt_src;
 381         ipc.opt = NULL;
 382         ipc.tx_flags = 0;
 383         if (icmp_param->replyopts.optlen) {
 384                 ipc.opt = &icmp_param->replyopts;
 385                 if (ipc.opt->srr)
 386                         daddr = icmp_param->replyopts.faddr;
 387         }
 388         {
 389                 struct flowi fl = { .fl4_dst= daddr,
 390                                     .fl4_src = rt->rt_spec_dst,
 391                                     .fl4_tos = RT_TOS(ip_hdr(skb)->tos),
 392                                     .proto = IPPROTO_ICMP };
 393                 security_skb_classify_flow(skb, &fl);
 394                 if (ip_route_output_key(net, &rt, &fl))
 395                         goto out_unlock;
 396         }
 397         if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
 398                                icmp_param->data.icmph.code))
 399                 icmp_push_reply(icmp_param, &ipc, &rt);
 400         ip_rt_put(rt);
 401 out_unlock:
 402         icmp_xmit_unlock(sk);
 403 }
 404
 405
 406 /*
 407  *      Send an ICMP message in response to a situation
 408  *
 409  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 410  *                MAY send more (we do).
 411  *                      MUST NOT change this header information.
 412  *                      MUST NOT reply to a multicast/broadcast IP address.
 413  *                      MUST NOT reply to a multicast/broadcast MAC address.
 414  *                      MUST reply to only the first fragment.
 415  */
 416
 417 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 418 {
 419         struct iphdr *iph;
 420         int room;
 421         struct icmp_bxm icmp_param;
 422         struct rtable *rt = skb_rtable(skb_in);
 423         struct ipcm_cookie ipc;
 424         __be32 saddr;
 425         u8  tos;
 426         struct net *net;
 427         struct sock *sk;
 428
 429         if (!rt)
 430                 goto out;
 431         net = dev_net(rt->dst.dev);
 432
 433         /*
 434          *      Find the original header. It is expected to be valid, of course.
 435          *      Check this, icmp_send is called from the most obscure devices
 436          *      sometimes.
 437          */
 438         iph = ip_hdr(skb_in);
 439
 440         if ((u8 *)iph < skb_in->head ||
 441             (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
 442                 goto out;
 443
 444         /*
 445          *      No replies to physical multicast/broadcast
 446          */
 447         if (skb_in->pkt_type != PACKET_HOST)
 448                 goto out;
 449
 450         /*
 451          *      Now check at the protocol level
 452          */
 453         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 454                 goto out;
 455
 456         /*
 457          *      Only reply to fragment 0. We byte re-order the constant
 458          *      mask for efficiency.
 459          */
 460         if (iph->frag_off & htons(IP_OFFSET))
 461                 goto out;
 462
 463         /*
 464          *      If we send an ICMP error to an ICMP error a mess would result..
 465          */
 466         if (icmp_pointers[type].error) {
 467                 /*
 468                  *      We are an error, check if we are replying to an
 469                  *      ICMP error
 470                  */
 471                 if (iph->protocol == IPPROTO_ICMP) {
 472                         u8 _inner_type, *itp;
 473
 474                         itp = skb_header_pointer(skb_in,
 475                                                  skb_network_header(skb_in) +
 476                                                  (iph->ihl << 2) +
 477                                                  offsetof(struct icmphdr,
 478                                                           type) -
 479                                                  skb_in->data,
 480                                                  sizeof(_inner_type),
 481                                                  &_inner_type);
 482                         if (itp == NULL)
 483                                 goto out;
 484
 485                         /*
 486                          *      Assume any unknown ICMP type is an error. This
 487                          *      isn't specified by the RFC, but think about it..
 488                          */
 489                         if (*itp > NR_ICMP_TYPES ||
 490                             icmp_pointers[*itp].error)
 491                                 goto out;
 492                 }
 493         }
 494
 495         sk = icmp_xmit_lock(net);
 496         if (sk == NULL)
 497                 return;
 498
 499         /*
 500          *      Construct source address and options.
 501          */
 502
 503         saddr = iph->daddr;
 504         if (!(rt->rt_flags & RTCF_LOCAL)) {
 505                 struct net_device *dev = NULL;
 506
 507                 rcu_read_lock();
 508                 if (rt_is_input_route(rt) &&
 509                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 510                         dev = dev_get_by_index_rcu(net, rt->fl.iif);
 511
 512                 if (dev)
 513                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 514                 else
 515                         saddr = 0;
 516                 rcu_read_unlock();
 517         }
 518
 519         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 520                                            IPTOS_PREC_INTERNETCONTROL) :
 521                                           iph->tos;
 522
 523         if (ip_options_echo(&icmp_param.replyopts, skb_in))
 524                 goto out_unlock;
 525
 526
 527         /*
 528          *      Prepare data for ICMP header.
 529          */
 530
 531         icmp_param.data.icmph.type       = type;
 532         icmp_param.data.icmph.code       = code;
 533         icmp_param.data.icmph.un.gateway = info;
 534         icmp_param.data.icmph.checksum   = 0;
 535         icmp_param.skb    = skb_in;
 536         icmp_param.offset = skb_network_offset(skb_in);
 537         inet_sk(sk)->tos = tos;
 538         ipc.addr = iph->saddr;
 539         ipc.opt = &icmp_param.replyopts;
 540         ipc.tx_flags = 0;
 541
 542         {
 543                 struct flowi fl = {
 544                         .fl4_dst = icmp_param.replyopts.srr ?
 545                                    icmp_param.replyopts.faddr : iph->saddr,
 546                         .fl4_src = saddr,
 547                         .fl4_tos = RT_TOS(tos),
 548                         .proto = IPPROTO_ICMP,
 549                         .fl_icmp_type = type,
 550                         .fl_icmp_code = code,
 551                 };
 552                 int err;
 553                 struct rtable *rt2;
 554
 555                 security_skb_classify_flow(skb_in, &fl);
 556                 if (__ip_route_output_key(net, &rt, &fl))
 557                         goto out_unlock;
 558
 559                 /* No need to clone since we're just using its address. */
 560                 rt2 = rt;
 561
 562                 if (!fl.nl_u.ip4_u.saddr)
 563                         fl.nl_u.ip4_u.saddr = rt->rt_src;
 564
 565                 err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
 566                 switch (err) {
 567                 case 0:
 568                         if (rt != rt2)
 569                                 goto route_done;
 570                         break;
 571                 case -EPERM:
 572                         rt = NULL;
 573                         break;
 574                 default:
 575                         goto out_unlock;
 576                 }
 577
 578                 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
 579                         goto relookup_failed;
 580
 581                 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
 582                         err = __ip_route_output_key(net, &rt2, &fl);
 583                 else {
 584                         struct flowi fl2 = {};
 585                         unsigned long orefdst;
 586
 587                         fl2.fl4_dst = fl.fl4_src;
 588                         if (ip_route_output_key(net, &rt2, &fl2))
 589                                 goto relookup_failed;
 590
 591                         /* Ugh! */
 592                         orefdst = skb_in->_skb_refdst; /* save old refdst */
 593                         err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
 594                                              RT_TOS(tos), rt2->dst.dev);
 595
 596                         dst_release(&rt2->dst);
 597                         rt2 = skb_rtable(skb_in);
 598                         skb_in->_skb_refdst = orefdst; /* restore old refdst */
 599                 }
 600
 601                 if (err)
 602                         goto relookup_failed;
 603
 604                 err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
 605                                   XFRM_LOOKUP_ICMP);
 606                 switch (err) {
 607                 case 0:
 608                         dst_release(&rt->dst);
 609                         rt = rt2;
 610                         break;
 611                 case -EPERM:
 612                         goto ende;
 613                 default:
 614 relookup_failed:
 615                         if (!rt)
 616                                 goto out_unlock;
 617                         break;
 618                 }
 619         }
 620
 621 route_done:
 622         if (!icmpv4_xrlim_allow(net, rt, type, code))
 623                 goto ende;
 624
 625         /* RFC says return as much as we can without exceeding 576 bytes. */
 626
 627         room = dst_mtu(&rt->dst);
 628         if (room > 576)
 629                 room = 576;
 630         room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
 631         room -= sizeof(struct icmphdr);
 632
 633         icmp_param.data_len = skb_in->len - icmp_param.offset;
 634         if (icmp_param.data_len > room)
 635                 icmp_param.data_len = room;
 636         icmp_param.head_len = sizeof(struct icmphdr);
 637
 638         icmp_push_reply(&icmp_param, &ipc, &rt);
 639 ende:
 640         ip_rt_put(rt);
 641 out_unlock:
 642         icmp_xmit_unlock(sk);
 643 out:;
 644 }
 645 EXPORT_SYMBOL(icmp_send);
 646
 647
 648 /*
 649  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
 650  */
 651
 652 static void icmp_unreach(struct sk_buff *skb)
 653 {
 654         struct iphdr *iph;
 655         struct icmphdr *icmph;
 656         int hash, protocol;
 657         const struct net_protocol *ipprot;
 658         u32 info = 0;
 659         struct net *net;
 660
 661         net = dev_net(skb_dst(skb)->dev);
 662
 663         /*
 664          *      Incomplete header ?
 665          *      Only checks for the IP header, there should be an
 666          *      additional check for longer headers in upper levels.
 667          */
 668
 669         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 670                 goto out_err;
 671
 672         icmph = icmp_hdr(skb);
 673         iph   = (struct iphdr *)skb->data;
 674
 675         if (iph->ihl < 5) /* Mangled header, drop. */
 676                 goto out_err;
 677
 678         if (icmph->type == ICMP_DEST_UNREACH) {
 679                 switch (icmph->code & 15) {
 680                 case ICMP_NET_UNREACH:
 681                 case ICMP_HOST_UNREACH:
 682                 case ICMP_PROT_UNREACH:
 683                 case ICMP_PORT_UNREACH:
 684                         break;
 685                 case ICMP_FRAG_NEEDED:
 686                         if (ipv4_config.no_pmtu_disc) {
 687                                 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
 688                                                &iph->daddr);
 689                         } else {
 690                                 info = ip_rt_frag_needed(net, iph,
 691                                                          ntohs(icmph->un.frag.mtu),
 692                                                          skb->dev);
 693                                 if (!info)
 694                                         goto out;
 695                         }
 696                         break;
 697                 case ICMP_SR_FAILED:
 698                         LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
 699                                        &iph->daddr);
 700                         break;
 701                 default:
 702                         break;
 703                 }
 704                 if (icmph->code > NR_ICMP_UNREACH)
 705                         goto out;
 706         } else if (icmph->type == ICMP_PARAMETERPROB)
 707                 info = ntohl(icmph->un.gateway) >> 24;
 708
 709         /*
 710          *      Throw it at our lower layers
 711          *
 712          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 713          *                header.
 714          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 715          *                transport layer.
 716          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 717          *                transport layer.
 718          */
 719
 720         /*
 721          *      Check the other end isnt violating RFC 1122. Some routers send
 722          *      bogus responses to broadcast frames. If you see this message
 723          *      first check your netmask matches at both ends, if it does then
 724          *      get the other vendor to fix their kit.
 725          */
 726
 727         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 728             inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
 729                 if (net_ratelimit())
 730                         printk(KERN_WARNING "%pI4 sent an invalid ICMP "
 731                                             "type %u, code %u "
 732                                             "error to a broadcast: %pI4 on %s\n",
 733                                &ip_hdr(skb)->saddr,
 734                                icmph->type, icmph->code,
 735                                &iph->daddr,
 736                                skb->dev->name);
 737                 goto out;
 738         }
 739
 740         /* Checkin full IP header plus 8 bytes of protocol to
 741          * avoid additional coding at protocol handlers.
 742          */
 743         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
 744                 goto out;
 745
 746         iph = (struct iphdr *)skb->data;
 747         protocol = iph->protocol;
 748
 749         /*
 750          *      Deliver ICMP message to raw sockets. Pretty useless feature?
 751          */
 752         raw_icmp_error(skb, protocol, info);
 753
 754         hash = protocol & (MAX_INET_PROTOS - 1);
 755         rcu_read_lock();
 756         ipprot = rcu_dereference(inet_protos[hash]);
 757         if (ipprot && ipprot->err_handler)
 758                 ipprot->err_handler(skb, info);
 759         rcu_read_unlock();
 760
 761 out:
 762         return;
 763 out_err:
 764         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 765         goto out;
 766 }
 767
 768
 769 /*
 770  *      Handle ICMP_REDIRECT.
 771  */
 772
 773 static void icmp_redirect(struct sk_buff *skb)
 774 {
 775         struct iphdr *iph;
 776
 777         if (skb->len < sizeof(struct iphdr))
 778                 goto out_err;
 779
 780         /*
 781          *      Get the copied header of the packet that caused the redirect
 782          */
 783         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 784                 goto out;
 785
 786         iph = (struct iphdr *)skb->data;
 787
 788         switch (icmp_hdr(skb)->code & 7) {
 789         case ICMP_REDIR_NET:
 790         case ICMP_REDIR_NETTOS:
 791                 /*
 792                  * As per RFC recommendations now handle it as a host redirect.
 793                  */
 794         case ICMP_REDIR_HOST:
 795         case ICMP_REDIR_HOSTTOS:
 796                 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
 797                                icmp_hdr(skb)->un.gateway,
 798                                iph->saddr, skb->dev);
 799                 break;
 800         }
 801 out:
 802         return;
 803 out_err:
 804         ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
 805         goto out;
 806 }
 807
 808 /*
 809  *      Handle ICMP_ECHO ("ping") requests.
 810  *
 811  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 812  *                requests.
 813  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 814  *                included in the reply.
 815  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 816  *                echo requests, MUST have default=NOT.
 817  *      See also WRT handling of options once they are done and working.
 818  */
 819
 820 static void icmp_echo(struct sk_buff *skb)
 821 {
 822         struct net *net;
 823
 824         net = dev_net(skb_dst(skb)->dev);
 825         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 826                 struct icmp_bxm icmp_param;
 827
 828                 icmp_param.data.icmph      = *icmp_hdr(skb);
 829                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 830                 icmp_param.skb             = skb;
 831                 icmp_param.offset          = 0;
 832                 icmp_param.data_len        = skb->len;
 833                 icmp_param.head_len        = sizeof(struct icmphdr);
 834                 icmp_reply(&icmp_param, skb);
 835         }
 836 }
 837
 838 /*
 839  *      Handle ICMP Timestamp requests.
 840  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 841  *                SHOULD be in the kernel for minimum random latency.
 842  *                MUST be accurate to a few minutes.
 843  *                MUST be updated at least at 15Hz.
 844  */
 845 static void icmp_timestamp(struct sk_buff *skb)
 846 {
 847         struct timespec tv;
 848         struct icmp_bxm icmp_param;
 849         /*
 850          *      Too short.
 851          */
 852         if (skb->len < 4)
 853                 goto out_err;
 854
 855         /*
 856          *      Fill in the current time as ms since midnight UT:
 857          */
 858         getnstimeofday(&tv);
 859         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
 860                                          tv.tv_nsec / NSEC_PER_MSEC);
 861         icmp_param.data.times[2] = icmp_param.data.times[1];
 862         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 863                 BUG();
 864         icmp_param.data.icmph      = *icmp_hdr(skb);
 865         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 866         icmp_param.data.icmph.code = 0;
 867         icmp_param.skb             = skb;
 868         icmp_param.offset          = 0;
 869         icmp_param.data_len        = 0;
 870         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 871         icmp_reply(&icmp_param, skb);
 872 out:
 873         return;
 874 out_err:
 875         ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 876         goto out;
 877 }
 878
 879
 880 /*
 881  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
 882  *
 883  * RFC1122 (3.2.2.9).  A host MUST only send replies to
 884  * ADDRESS_MASK requests if it's been configured as an address mask
 885  * agent.  Receiving a request doesn't constitute implicit permission to
 886  * act as one. Of course, implementing this correctly requires (SHOULD)
 887  * a way to turn the functionality on and off.  Another one for sysctl(),
 888  * I guess. -- MS
 889  *
 890  * RFC1812 (4.3.3.9).   A router MUST implement it.
 891  *                      A router SHOULD have switch turning it on/off.
 892  *                      This switch MUST be ON by default.
 893  *
 894  * Gratuitous replies, zero-source replies are not implemented,
 895  * that complies with RFC. DO NOT implement them!!! All the idea
 896  * of broadcast addrmask replies as specified in RFC950 is broken.
 897  * The problem is that it is not uncommon to have several prefixes
 898  * on one physical interface. Moreover, addrmask agent can even be
 899  * not aware of existing another prefixes.
 900  * If source is zero, addrmask agent cannot choose correct prefix.
 901  * Gratuitous mask announcements suffer from the same problem.
 902  * RFC1812 explains it, but still allows to use ADDRMASK,
 903  * that is pretty silly. --ANK
 904  *
 905  * All these rules are so bizarre, that I removed kernel addrmask
 906  * support at all. It is wrong, it is obsolete, nobody uses it in
 907  * any case. --ANK
 908  *
 909  * Furthermore you can do it with a usermode address agent program
 910  * anyway...
 911  */
 912
 913 static void icmp_address(struct sk_buff *skb)
 914 {
 915 #if 0
 916         if (net_ratelimit())
 917                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
 918 #endif
 919 }
 920
 921 /*
 922  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
 923  *                      loudly if an inconsistency is found.
 924  * called with rcu_read_lock()
 925  */
 926
 927 static void icmp_address_reply(struct sk_buff *skb)
 928 {
 929         struct rtable *rt = skb_rtable(skb);
 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                 return;
 936
 937         in_dev = __in_dev_get_rcu(dev);
 938         if (!in_dev)
 939                 return;
 940
 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 }
 959
 960 static void icmp_discard(struct sk_buff *skb)
 961 {
 962 }
 963
 964 /*
 965  *      Deal with incoming ICMP packets.
 966  */
 967 int icmp_rcv(struct sk_buff *skb)
 968 {
 969         struct icmphdr *icmph;
 970         struct rtable *rt = skb_rtable(skb);
 971         struct net *net = dev_net(rt->dst.dev);
 972
 973         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 974                 struct sec_path *sp = skb_sec_path(skb);
 975                 int nh;
 976
 977                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 978                                  XFRM_STATE_ICMP))
 979                         goto drop;
 980
 981                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
 982                         goto drop;
 983
 984                 nh = skb_network_offset(skb);
 985                 skb_set_network_header(skb, sizeof(*icmph));
 986
 987                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 988                         goto drop;
 989
 990                 skb_set_network_header(skb, nh);
 991         }
 992
 993         ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
 994
 995         switch (skb->ip_summed) {
 996         case CHECKSUM_COMPLETE:
 997                 if (!csum_fold(skb->csum))
 998                         break;
 999                 /* fall through */
1000         case CHECKSUM_NONE:
1001                 skb->csum = 0;
1002                 if (__skb_checksum_complete(skb))
1003                         goto error;
1004         }
1005
1006         if (!pskb_pull(skb, sizeof(*icmph)))
1007                 goto error;
1008
1009         icmph = icmp_hdr(skb);
1010
1011         ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1012         /*
1013          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1014          *
1015          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1016          *                discarded.
1017          */
1018         if (icmph->type > NR_ICMP_TYPES)
1019                 goto error;
1020
1021
1022         /*
1023          *      Parse the ICMP message
1024          */
1025
1026         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1027                 /*
1028                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1029                  *        silently ignored (we let user decide with a sysctl).
1030                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1031                  *        discarded if to broadcast/multicast.
1032                  */
1033                 if ((icmph->type == ICMP_ECHO ||
1034                      icmph->type == ICMP_TIMESTAMP) &&
1035                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1036                         goto error;
1037                 }
1038                 if (icmph->type != ICMP_ECHO &&
1039                     icmph->type != ICMP_TIMESTAMP &&
1040                     icmph->type != ICMP_ADDRESS &&
1041                     icmph->type != ICMP_ADDRESSREPLY) {
1042                         goto error;
1043                 }
1044         }
1045
1046         icmp_pointers[icmph->type].handler(skb);
1047
1048 drop:
1049         kfree_skb(skb);
1050         return 0;
1051 error:
1052         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1053         goto drop;
1054 }
1055
1056 /*
1057  *      This table is the definition of how we handle ICMP.
1058  */
1059 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1060         [ICMP_ECHOREPLY] = {
1061                 .handler = icmp_discard,
1062         },
1063         [1] = {
1064                 .handler = icmp_discard,
1065                 .error = 1,
1066         },
1067         [2] = {
1068                 .handler = icmp_discard,
1069                 .error = 1,
1070         },
1071         [ICMP_DEST_UNREACH] = {
1072                 .handler = icmp_unreach,
1073                 .error = 1,
1074         },
1075         [ICMP_SOURCE_QUENCH] = {
1076                 .handler = icmp_unreach,
1077                 .error = 1,
1078         },
1079         [ICMP_REDIRECT] = {
1080                 .handler = icmp_redirect,
1081                 .error = 1,
1082         },
1083         [6] = {
1084                 .handler = icmp_discard,
1085                 .error = 1,
1086         },
1087         [7] = {
1088                 .handler = icmp_discard,
1089                 .error = 1,
1090         },
1091         [ICMP_ECHO] = {
1092                 .handler = icmp_echo,
1093         },
1094         [9] = {
1095                 .handler = icmp_discard,
1096                 .error = 1,
1097         },
1098         [10] = {
1099                 .handler = icmp_discard,
1100                 .error = 1,
1101         },
1102         [ICMP_TIME_EXCEEDED] = {
1103                 .handler = icmp_unreach,
1104                 .error = 1,
1105         },
1106         [ICMP_PARAMETERPROB] = {
1107                 .handler = icmp_unreach,
1108                 .error = 1,
1109         },
1110         [ICMP_TIMESTAMP] = {
1111                 .handler = icmp_timestamp,
1112         },
1113         [ICMP_TIMESTAMPREPLY] = {
1114                 .handler = icmp_discard,
1115         },
1116         [ICMP_INFO_REQUEST] = {
1117                 .handler = icmp_discard,
1118         },
1119         [ICMP_INFO_REPLY] = {
1120                 .handler = icmp_discard,
1121         },
1122         [ICMP_ADDRESS] = {
1123                 .handler = icmp_address,
1124         },
1125         [ICMP_ADDRESSREPLY] = {
1126                 .handler = icmp_address_reply,
1127         },
1128 };
1129
1130 static void __net_exit icmp_sk_exit(struct net *net)
1131 {
1132         int i;
1133
1134         for_each_possible_cpu(i)
1135                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1136         kfree(net->ipv4.icmp_sk);
1137         net->ipv4.icmp_sk = NULL;
1138 }
1139
1140 static int __net_init icmp_sk_init(struct net *net)
1141 {
1142         int i, err;
1143
1144         net->ipv4.icmp_sk =
1145                 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1146         if (net->ipv4.icmp_sk == NULL)
1147                 return -ENOMEM;
1148
1149         for_each_possible_cpu(i) {
1150                 struct sock *sk;
1151
1152                 err = inet_ctl_sock_create(&sk, PF_INET,
1153                                            SOCK_RAW, IPPROTO_ICMP, net);
1154                 if (err < 0)
1155                         goto fail;
1156
1157                 net->ipv4.icmp_sk[i] = sk;
1158
1159                 /* Enough space for 2 64K ICMP packets, including
1160                  * sk_buff struct overhead.
1161                  */
1162                 sk->sk_sndbuf =
1163                         (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1164
1165                 /*
1166                  * Speedup sock_wfree()
1167                  */
1168                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1169                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1170         }
1171
1172         /* Control parameters for ECHO replies. */
1173         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1174         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1175
1176         /* Control parameter - ignore bogus broadcast responses? */
1177         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1178
1179         /*
1180          *      Configurable global rate limit.
1181          *
1182          *      ratelimit defines tokens/packet consumed for dst->rate_token
1183          *      bucket ratemask defines which icmp types are ratelimited by
1184          *      setting it's bit position.
1185          *
1186          *      default:
1187          *      dest unreachable (3), source quench (4),
1188          *      time exceeded (11), parameter problem (12)
1189          */
1190
1191         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1192         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1193         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1194
1195         return 0;
1196
1197 fail:
1198         for_each_possible_cpu(i)
1199                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1200         kfree(net->ipv4.icmp_sk);
1201         return err;
1202 }
1203
1204 static struct pernet_operations __net_initdata icmp_sk_ops = {
1205        .init = icmp_sk_init,
1206        .exit = icmp_sk_exit,
1207 };
1208
1209 int __init icmp_init(void)
1210 {
1211         return register_pernet_subsys(&icmp_sk_ops);
1212 }