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_data replyopts;
 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
 117 const struct 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 EXPORT_SYMBOL(icmp_err_convert);
 184
 185 /*
 186  *      ICMP control array. This specifies what to do with each ICMP.
 187  */
 188
 189 struct icmp_control {
 190         void (*handler)(struct sk_buff *skb);
 191         short   error;          /* This ICMP is classed as an error message */
 192 };
 193
 194 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 195
 196 /*
 197  *      The ICMP socket(s). This is the most convenient way to flow control
 198  *      our ICMP output as well as maintain a clean interface throughout
 199  *      all layers. All Socketless IP sends will soon be gone.
 200  *
 201  *      On SMP we have one ICMP socket per-cpu.
 202  */
 203 static struct sock *icmp_sk(struct net *net)
 204 {
 205         return net->ipv4.icmp_sk[smp_processor_id()];
 206 }
 207
 208 static inline struct sock *icmp_xmit_lock(struct net *net)
 209 {
 210         struct sock *sk;
 211
 212         local_bh_disable();
 213
 214         sk = icmp_sk(net);
 215
 216         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 217                 /* This can happen if the output path signals a
 218                  * dst_link_failure() for an outgoing ICMP packet.
 219                  */
 220                 local_bh_enable();
 221                 return NULL;
 222         }
 223         return sk;
 224 }
 225
 226 static inline void icmp_xmit_unlock(struct sock *sk)
 227 {
 228         spin_unlock_bh(&sk->sk_lock.slock);
 229 }
 230
 231 /*
 232  *      Send an ICMP frame.
 233  */
 234
 235 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 236                 int type, int code)
 237 {
 238         struct dst_entry *dst = &rt->dst;
 239         bool rc = true;
 240
 241         if (type > NR_ICMP_TYPES)
 242                 goto out;
 243
 244         /* Don't limit PMTU discovery. */
 245         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 246                 goto out;
 247
 248         /* No rate limit on loopback */
 249         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 250                 goto out;
 251
 252         /* Limit if icmp type is enabled in ratemask. */
 253         if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
 254                 if (!rt->peer)
 255                         rt_bind_peer(rt, 1);
 256                 rc = inet_peer_xrlim_allow(rt->peer,
 257                                            net->ipv4.sysctl_icmp_ratelimit);
 258         }
 259 out:
 260         return rc;
 261 }
 262
 263 /*
 264  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 265  */
 266 void icmp_out_count(struct net *net, unsigned char type)
 267 {
 268         ICMPMSGOUT_INC_STATS(net, type);
 269         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 270 }
 271
 272 /*
 273  *      Checksum each fragment, and on the first include the headers and final
 274  *      checksum.
 275  */
 276 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 277                           struct sk_buff *skb)
 278 {
 279         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 280         __wsum csum;
 281
 282         csum = skb_copy_and_csum_bits(icmp_param->skb,
 283                                       icmp_param->offset + offset,
 284                                       to, len, 0);
 285
 286         skb->csum = csum_block_add(skb->csum, csum, odd);
 287         if (icmp_pointers[icmp_param->data.icmph.type].error)
 288                 nf_ct_attach(skb, icmp_param->skb);
 289         return 0;
 290 }
 291
 292 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 293                             struct ipcm_cookie *ipc, struct rtable **rt)
 294 {
 295         struct sock *sk;
 296         struct sk_buff *skb;
 297
 298         sk = icmp_sk(dev_net((*rt)->dst.dev));
 299         if (ip_append_data(sk, icmp_glue_bits, icmp_param,
 300                            icmp_param->data_len+icmp_param->head_len,
 301                            icmp_param->head_len,
 302                            ipc, rt, MSG_DONTWAIT) < 0) {
 303                 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
 304                 ip_flush_pending_frames(sk);
 305         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 306                 struct icmphdr *icmph = icmp_hdr(skb);
 307                 __wsum csum = 0;
 308                 struct sk_buff *skb1;
 309
 310                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 311                         csum = csum_add(csum, skb1->csum);
 312                 }
 313                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 314                                                  (char *)icmph,
 315                                                  icmp_param->head_len, csum);
 316                 icmph->checksum = csum_fold(csum);
 317                 skb->ip_summed = CHECKSUM_NONE;
 318                 ip_push_pending_frames(sk);
 319         }
 320 }
 321
 322 /*
 323  *      Driving logic for building and sending ICMP messages.
 324  */
 325
 326 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 327 {
 328         struct ipcm_cookie ipc;
 329         struct rtable *rt = skb_rtable(skb);
 330         struct net *net = dev_net(rt->dst.dev);
 331         struct sock *sk;
 332         struct inet_sock *inet;
 333         __be32 daddr;
 334
 335         if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
 336                 return;
 337
 338         sk = icmp_xmit_lock(net);
 339         if (sk == NULL)
 340                 return;
 341         inet = inet_sk(sk);
 342
 343         icmp_param->data.icmph.checksum = 0;
 344
 345         inet->tos = ip_hdr(skb)->tos;
 346         daddr = ipc.addr = rt->rt_src;
 347         ipc.opt = NULL;
 348         ipc.tx_flags = 0;
 349         if (icmp_param->replyopts.opt.opt.optlen) {
 350                 ipc.opt = &icmp_param->replyopts.opt;
 351                 if (ipc.opt->opt.srr)
 352                         daddr = icmp_param->replyopts.opt.opt.faddr;
 353         }
 354         {
 355                 struct flowi4 fl4 = {
 356                         .daddr = daddr,
 357                         .saddr = rt->rt_spec_dst,
 358                         .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
 359                         .flowi4_proto = IPPROTO_ICMP,
 360                 };
 361                 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 362                 rt = ip_route_output_key(net, &fl4);
 363                 if (IS_ERR(rt))
 364                         goto out_unlock;
 365         }
 366         if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
 367                                icmp_param->data.icmph.code))
 368                 icmp_push_reply(icmp_param, &ipc, &rt);
 369         ip_rt_put(rt);
 370 out_unlock:
 371         icmp_xmit_unlock(sk);
 372 }
 373
 374 static struct rtable *icmp_route_lookup(struct net *net, struct sk_buff *skb_in,
 375                                         const struct iphdr *iph,
 376                                         __be32 saddr, u8 tos,
 377                                         int type, int code,
 378                                         struct icmp_bxm *param)
 379 {
 380         struct flowi4 fl4 = {
 381                 .daddr = (param->replyopts.opt.opt.srr ?
 382                           param->replyopts.opt.opt.faddr : iph->saddr),
 383                 .saddr = saddr,
 384                 .flowi4_tos = RT_TOS(tos),
 385                 .flowi4_proto = IPPROTO_ICMP,
 386                 .fl4_icmp_type = type,
 387                 .fl4_icmp_code = code,
 388         };
 389         struct rtable *rt, *rt2;
 390         int err;
 391
 392         security_skb_classify_flow(skb_in, flowi4_to_flowi(&fl4));
 393         rt = __ip_route_output_key(net, &fl4);
 394         if (IS_ERR(rt))
 395                 return rt;
 396
 397         /* No need to clone since we're just using its address. */
 398         rt2 = rt;
 399
 400         if (!fl4.saddr)
 401                 fl4.saddr = rt->rt_src;
 402
 403         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 404                                            flowi4_to_flowi(&fl4), 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, flowi4_to_flowi(&fl4), AF_INET);
 414         if (err)
 415                 goto relookup_failed;
 416
 417         if (inet_addr_type(net, fl4.saddr) == RTN_LOCAL) {
 418                 rt2 = __ip_route_output_key(net, &fl4);
 419                 if (IS_ERR(rt2))
 420                         err = PTR_ERR(rt2);
 421         } else {
 422                 struct flowi4 fl4_2 = {};
 423                 unsigned long orefdst;
 424
 425                 fl4_2.daddr = fl4.saddr;
 426                 rt2 = ip_route_output_key(net, &fl4_2);
 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, fl4.daddr, fl4.saddr,
 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,
 445                                             flowi4_to_flowi(&fl4), NULL,
 446                                             XFRM_LOOKUP_ICMP);
 447         if (!IS_ERR(rt2)) {
 448                 dst_release(&rt->dst);
 449                 rt = rt2;
 450         } else if (PTR_ERR(rt2) == -EPERM) {
 451                 if (rt)
 452                         dst_release(&rt->dst);
 453                 return rt2;
 454         } else {
 455                 err = PTR_ERR(rt2);
 456                 goto relookup_failed;
 457         }
 458         return rt;
 459
 460 relookup_failed:
 461         if (rt)
 462                 return rt;
 463         return ERR_PTR(err);
 464 }
 465
 466 /*
 467  *      Send an ICMP message in response to a situation
 468  *
 469  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 470  *                MAY send more (we do).
 471  *                      MUST NOT change this header information.
 472  *                      MUST NOT reply to a multicast/broadcast IP address.
 473  *                      MUST NOT reply to a multicast/broadcast MAC address.
 474  *                      MUST reply to only the first fragment.
 475  */
 476
 477 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 478 {
 479         struct iphdr *iph;
 480         int room;
 481         struct icmp_bxm icmp_param;
 482         struct rtable *rt = skb_rtable(skb_in);
 483         struct ipcm_cookie ipc;
 484         __be32 saddr;
 485         u8  tos;
 486         struct net *net;
 487         struct sock *sk;
 488
 489         if (!rt)
 490                 goto out;
 491         net = dev_net(rt->dst.dev);
 492
 493         /*
 494          *      Find the original header. It is expected to be valid, of course.
 495          *      Check this, icmp_send is called from the most obscure devices
 496          *      sometimes.
 497          */
 498         iph = ip_hdr(skb_in);
 499
 500         if ((u8 *)iph < skb_in->head ||
 501             (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
 502                 goto out;
 503
 504         /*
 505          *      No replies to physical multicast/broadcast
 506          */
 507         if (skb_in->pkt_type != PACKET_HOST)
 508                 goto out;
 509
 510         /*
 511          *      Now check at the protocol level
 512          */
 513         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 514                 goto out;
 515
 516         /*
 517          *      Only reply to fragment 0. We byte re-order the constant
 518          *      mask for efficiency.
 519          */
 520         if (iph->frag_off & htons(IP_OFFSET))
 521                 goto out;
 522
 523         /*
 524          *      If we send an ICMP error to an ICMP error a mess would result..
 525          */
 526         if (icmp_pointers[type].error) {
 527                 /*
 528                  *      We are an error, check if we are replying to an
 529                  *      ICMP error
 530                  */
 531                 if (iph->protocol == IPPROTO_ICMP) {
 532                         u8 _inner_type, *itp;
 533
 534                         itp = skb_header_pointer(skb_in,
 535                                                  skb_network_header(skb_in) +
 536                                                  (iph->ihl << 2) +
 537                                                  offsetof(struct icmphdr,
 538                                                           type) -
 539                                                  skb_in->data,
 540                                                  sizeof(_inner_type),
 541                                                  &_inner_type);
 542                         if (itp == NULL)
 543                                 goto out;
 544
 545                         /*
 546                          *      Assume any unknown ICMP type is an error. This
 547                          *      isn't specified by the RFC, but think about it..
 548                          */
 549                         if (*itp > NR_ICMP_TYPES ||
 550                             icmp_pointers[*itp].error)
 551                                 goto out;
 552                 }
 553         }
 554
 555         sk = icmp_xmit_lock(net);
 556         if (sk == NULL)
 557                 return;
 558
 559         /*
 560          *      Construct source address and options.
 561          */
 562
 563         saddr = iph->daddr;
 564         if (!(rt->rt_flags & RTCF_LOCAL)) {
 565                 struct net_device *dev = NULL;
 566
 567                 rcu_read_lock();
 568                 if (rt_is_input_route(rt) &&
 569                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 570                         dev = dev_get_by_index_rcu(net, rt->rt_iif);
 571
 572                 if (dev)
 573                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 574                 else
 575                         saddr = 0;
 576                 rcu_read_unlock();
 577         }
 578
 579         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 580                                            IPTOS_PREC_INTERNETCONTROL) :
 581                                           iph->tos;
 582
 583         if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
 584                 goto out_unlock;
 585
 586
 587         /*
 588          *      Prepare data for ICMP header.
 589          */
 590
 591         icmp_param.data.icmph.type       = type;
 592         icmp_param.data.icmph.code       = code;
 593         icmp_param.data.icmph.un.gateway = info;
 594         icmp_param.data.icmph.checksum   = 0;
 595         icmp_param.skb    = skb_in;
 596         icmp_param.offset = skb_network_offset(skb_in);
 597         inet_sk(sk)->tos = tos;
 598         ipc.addr = iph->saddr;
 599         ipc.opt = &icmp_param.replyopts.opt;
 600         ipc.tx_flags = 0;
 601
 602         rt = icmp_route_lookup(net, skb_in, iph, saddr, tos,
 603                                type, code, &icmp_param);
 604         if (IS_ERR(rt))
 605                 goto out_unlock;
 606
 607         if (!icmpv4_xrlim_allow(net, rt, type, code))
 608                 goto ende;
 609
 610         /* RFC says return as much as we can without exceeding 576 bytes. */
 611
 612         room = dst_mtu(&rt->dst);
 613         if (room > 576)
 614                 room = 576;
 615         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 616         room -= sizeof(struct icmphdr);
 617
 618         icmp_param.data_len = skb_in->len - icmp_param.offset;
 619         if (icmp_param.data_len > room)
 620                 icmp_param.data_len = room;
 621         icmp_param.head_len = sizeof(struct icmphdr);
 622
 623         icmp_push_reply(&icmp_param, &ipc, &rt);
 624 ende:
 625         ip_rt_put(rt);
 626 out_unlock:
 627         icmp_xmit_unlock(sk);
 628 out:;
 629 }
 630 EXPORT_SYMBOL(icmp_send);
 631
 632
 633 /*
 634  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
 635  */
 636
 637 static void icmp_unreach(struct sk_buff *skb)
 638 {
 639         const struct iphdr *iph;
 640         struct icmphdr *icmph;
 641         int hash, protocol;
 642         const struct net_protocol *ipprot;
 643         u32 info = 0;
 644         struct net *net;
 645
 646         net = dev_net(skb_dst(skb)->dev);
 647
 648         /*
 649          *      Incomplete header ?
 650          *      Only checks for the IP header, there should be an
 651          *      additional check for longer headers in upper levels.
 652          */
 653
 654         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 655                 goto out_err;
 656
 657         icmph = icmp_hdr(skb);
 658         iph   = (const struct iphdr *)skb->data;
 659
 660         if (iph->ihl < 5) /* Mangled header, drop. */
 661                 goto out_err;
 662
 663         if (icmph->type == ICMP_DEST_UNREACH) {
 664                 switch (icmph->code & 15) {
 665                 case ICMP_NET_UNREACH:
 666                 case ICMP_HOST_UNREACH:
 667                 case ICMP_PROT_UNREACH:
 668                 case ICMP_PORT_UNREACH:
 669                         break;
 670                 case ICMP_FRAG_NEEDED:
 671                         if (ipv4_config.no_pmtu_disc) {
 672                                 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
 673                                                &iph->daddr);
 674                         } else {
 675                                 info = ip_rt_frag_needed(net, iph,
 676                                                          ntohs(icmph->un.frag.mtu),
 677                                                          skb->dev);
 678                                 if (!info)
 679                                         goto out;
 680                         }
 681                         break;
 682                 case ICMP_SR_FAILED:
 683                         LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
 684                                        &iph->daddr);
 685                         break;
 686                 default:
 687                         break;
 688                 }
 689                 if (icmph->code > NR_ICMP_UNREACH)
 690                         goto out;
 691         } else if (icmph->type == ICMP_PARAMETERPROB)
 692                 info = ntohl(icmph->un.gateway) >> 24;
 693
 694         /*
 695          *      Throw it at our lower layers
 696          *
 697          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 698          *                header.
 699          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 700          *                transport layer.
 701          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 702          *                transport layer.
 703          */
 704
 705         /*
 706          *      Check the other end isn't violating RFC 1122. Some routers send
 707          *      bogus responses to broadcast frames. If you see this message
 708          *      first check your netmask matches at both ends, if it does then
 709          *      get the other vendor to fix their kit.
 710          */
 711
 712         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 713             inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
 714                 if (net_ratelimit())
 715                         printk(KERN_WARNING "%pI4 sent an invalid ICMP "
 716                                             "type %u, code %u "
 717                                             "error to a broadcast: %pI4 on %s\n",
 718                                &ip_hdr(skb)->saddr,
 719                                icmph->type, icmph->code,
 720                                &iph->daddr,
 721                                skb->dev->name);
 722                 goto out;
 723         }
 724
 725         /* Checkin full IP header plus 8 bytes of protocol to
 726          * avoid additional coding at protocol handlers.
 727          */
 728         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
 729                 goto out;
 730
 731         iph = (const struct iphdr *)skb->data;
 732         protocol = iph->protocol;
 733
 734         /*
 735          *      Deliver ICMP message to raw sockets. Pretty useless feature?
 736          */
 737         raw_icmp_error(skb, protocol, info);
 738
 739         hash = protocol & (MAX_INET_PROTOS - 1);
 740         rcu_read_lock();
 741         ipprot = rcu_dereference(inet_protos[hash]);
 742         if (ipprot && ipprot->err_handler)
 743                 ipprot->err_handler(skb, info);
 744         rcu_read_unlock();
 745
 746 out:
 747         return;
 748 out_err:
 749         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 750         goto out;
 751 }
 752
 753
 754 /*
 755  *      Handle ICMP_REDIRECT.
 756  */
 757
 758 static void icmp_redirect(struct sk_buff *skb)
 759 {
 760         const struct iphdr *iph;
 761
 762         if (skb->len < sizeof(struct iphdr))
 763                 goto out_err;
 764
 765         /*
 766          *      Get the copied header of the packet that caused the redirect
 767          */
 768         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 769                 goto out;
 770
 771         iph = (const struct iphdr *)skb->data;
 772
 773         switch (icmp_hdr(skb)->code & 7) {
 774         case ICMP_REDIR_NET:
 775         case ICMP_REDIR_NETTOS:
 776                 /*
 777                  * As per RFC recommendations now handle it as a host redirect.
 778                  */
 779         case ICMP_REDIR_HOST:
 780         case ICMP_REDIR_HOSTTOS:
 781                 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
 782                                icmp_hdr(skb)->un.gateway,
 783                                iph->saddr, skb->dev);
 784                 break;
 785         }
 786 out:
 787         return;
 788 out_err:
 789         ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
 790         goto out;
 791 }
 792
 793 /*
 794  *      Handle ICMP_ECHO ("ping") requests.
 795  *
 796  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 797  *                requests.
 798  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 799  *                included in the reply.
 800  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 801  *                echo requests, MUST have default=NOT.
 802  *      See also WRT handling of options once they are done and working.
 803  */
 804
 805 static void icmp_echo(struct sk_buff *skb)
 806 {
 807         struct net *net;
 808
 809         net = dev_net(skb_dst(skb)->dev);
 810         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 811                 struct icmp_bxm icmp_param;
 812
 813                 icmp_param.data.icmph      = *icmp_hdr(skb);
 814                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 815                 icmp_param.skb             = skb;
 816                 icmp_param.offset          = 0;
 817                 icmp_param.data_len        = skb->len;
 818                 icmp_param.head_len        = sizeof(struct icmphdr);
 819                 icmp_reply(&icmp_param, skb);
 820         }
 821 }
 822
 823 /*
 824  *      Handle ICMP Timestamp requests.
 825  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 826  *                SHOULD be in the kernel for minimum random latency.
 827  *                MUST be accurate to a few minutes.
 828  *                MUST be updated at least at 15Hz.
 829  */
 830 static void icmp_timestamp(struct sk_buff *skb)
 831 {
 832         struct timespec tv;
 833         struct icmp_bxm icmp_param;
 834         /*
 835          *      Too short.
 836          */
 837         if (skb->len < 4)
 838                 goto out_err;
 839
 840         /*
 841          *      Fill in the current time as ms since midnight UT:
 842          */
 843         getnstimeofday(&tv);
 844         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
 845                                          tv.tv_nsec / NSEC_PER_MSEC);
 846         icmp_param.data.times[2] = icmp_param.data.times[1];
 847         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 848                 BUG();
 849         icmp_param.data.icmph      = *icmp_hdr(skb);
 850         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 851         icmp_param.data.icmph.code = 0;
 852         icmp_param.skb             = skb;
 853         icmp_param.offset          = 0;
 854         icmp_param.data_len        = 0;
 855         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 856         icmp_reply(&icmp_param, skb);
 857 out:
 858         return;
 859 out_err:
 860         ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 861         goto out;
 862 }
 863
 864
 865 /*
 866  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
 867  *
 868  * RFC1122 (3.2.2.9).  A host MUST only send replies to
 869  * ADDRESS_MASK requests if it's been configured as an address mask
 870  * agent.  Receiving a request doesn't constitute implicit permission to
 871  * act as one. Of course, implementing this correctly requires (SHOULD)
 872  * a way to turn the functionality on and off.  Another one for sysctl(),
 873  * I guess. -- MS
 874  *
 875  * RFC1812 (4.3.3.9).   A router MUST implement it.
 876  *                      A router SHOULD have switch turning it on/off.
 877  *                      This switch MUST be ON by default.
 878  *
 879  * Gratuitous replies, zero-source replies are not implemented,
 880  * that complies with RFC. DO NOT implement them!!! All the idea
 881  * of broadcast addrmask replies as specified in RFC950 is broken.
 882  * The problem is that it is not uncommon to have several prefixes
 883  * on one physical interface. Moreover, addrmask agent can even be
 884  * not aware of existing another prefixes.
 885  * If source is zero, addrmask agent cannot choose correct prefix.
 886  * Gratuitous mask announcements suffer from the same problem.
 887  * RFC1812 explains it, but still allows to use ADDRMASK,
 888  * that is pretty silly. --ANK
 889  *
 890  * All these rules are so bizarre, that I removed kernel addrmask
 891  * support at all. It is wrong, it is obsolete, nobody uses it in
 892  * any case. --ANK
 893  *
 894  * Furthermore you can do it with a usermode address agent program
 895  * anyway...
 896  */
 897
 898 static void icmp_address(struct sk_buff *skb)
 899 {
 900 #if 0
 901         if (net_ratelimit())
 902                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
 903 #endif
 904 }
 905
 906 /*
 907  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
 908  *                      loudly if an inconsistency is found.
 909  * called with rcu_read_lock()
 910  */
 911
 912 static void icmp_address_reply(struct sk_buff *skb)
 913 {
 914         struct rtable *rt = skb_rtable(skb);
 915         struct net_device *dev = skb->dev;
 916         struct in_device *in_dev;
 917         struct in_ifaddr *ifa;
 918
 919         if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
 920                 return;
 921
 922         in_dev = __in_dev_get_rcu(dev);
 923         if (!in_dev)
 924                 return;
 925
 926         if (in_dev->ifa_list &&
 927             IN_DEV_LOG_MARTIANS(in_dev) &&
 928             IN_DEV_FORWARD(in_dev)) {
 929                 __be32 _mask, *mp;
 930
 931                 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
 932                 BUG_ON(mp == NULL);
 933                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
 934                         if (*mp == ifa->ifa_mask &&
 935                             inet_ifa_match(rt->rt_src, ifa))
 936                                 break;
 937                 }
 938                 if (!ifa && net_ratelimit()) {
 939                         printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
 940                                mp, dev->name, &rt->rt_src);
 941                 }
 942         }
 943 }
 944
 945 static void icmp_discard(struct sk_buff *skb)
 946 {
 947 }
 948
 949 /*
 950  *      Deal with incoming ICMP packets.
 951  */
 952 int icmp_rcv(struct sk_buff *skb)
 953 {
 954         struct icmphdr *icmph;
 955         struct rtable *rt = skb_rtable(skb);
 956         struct net *net = dev_net(rt->dst.dev);
 957
 958         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 959                 struct sec_path *sp = skb_sec_path(skb);
 960                 int nh;
 961
 962                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 963                                  XFRM_STATE_ICMP))
 964                         goto drop;
 965
 966                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
 967                         goto drop;
 968
 969                 nh = skb_network_offset(skb);
 970                 skb_set_network_header(skb, sizeof(*icmph));
 971
 972                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 973                         goto drop;
 974
 975                 skb_set_network_header(skb, nh);
 976         }
 977
 978         ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
 979
 980         switch (skb->ip_summed) {
 981         case CHECKSUM_COMPLETE:
 982                 if (!csum_fold(skb->csum))
 983                         break;
 984                 /* fall through */
 985         case CHECKSUM_NONE:
 986                 skb->csum = 0;
 987                 if (__skb_checksum_complete(skb))
 988                         goto error;
 989         }
 990
 991         if (!pskb_pull(skb, sizeof(*icmph)))
 992                 goto error;
 993
 994         icmph = icmp_hdr(skb);
 995
 996         ICMPMSGIN_INC_STATS_BH(net, icmph->type);
 997         /*
 998          *      18 is the highest 'known' ICMP type. Anything else is a mystery
 999          *
1000          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1001          *                discarded.
1002          */
1003         if (icmph->type > NR_ICMP_TYPES)
1004                 goto error;
1005
1006
1007         /*
1008          *      Parse the ICMP message
1009          */
1010
1011         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1012                 /*
1013                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1014                  *        silently ignored (we let user decide with a sysctl).
1015                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1016                  *        discarded if to broadcast/multicast.
1017                  */
1018                 if ((icmph->type == ICMP_ECHO ||
1019                      icmph->type == ICMP_TIMESTAMP) &&
1020                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1021                         goto error;
1022                 }
1023                 if (icmph->type != ICMP_ECHO &&
1024                     icmph->type != ICMP_TIMESTAMP &&
1025                     icmph->type != ICMP_ADDRESS &&
1026                     icmph->type != ICMP_ADDRESSREPLY) {
1027                         goto error;
1028                 }
1029         }
1030
1031         icmp_pointers[icmph->type].handler(skb);
1032
1033 drop:
1034         kfree_skb(skb);
1035         return 0;
1036 error:
1037         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1038         goto drop;
1039 }
1040
1041 /*
1042  *      This table is the definition of how we handle ICMP.
1043  */
1044 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1045         [ICMP_ECHOREPLY] = {
1046                 .handler = icmp_discard,
1047         },
1048         [1] = {
1049                 .handler = icmp_discard,
1050                 .error = 1,
1051         },
1052         [2] = {
1053                 .handler = icmp_discard,
1054                 .error = 1,
1055         },
1056         [ICMP_DEST_UNREACH] = {
1057                 .handler = icmp_unreach,
1058                 .error = 1,
1059         },
1060         [ICMP_SOURCE_QUENCH] = {
1061                 .handler = icmp_unreach,
1062                 .error = 1,
1063         },
1064         [ICMP_REDIRECT] = {
1065                 .handler = icmp_redirect,
1066                 .error = 1,
1067         },
1068         [6] = {
1069                 .handler = icmp_discard,
1070                 .error = 1,
1071         },
1072         [7] = {
1073                 .handler = icmp_discard,
1074                 .error = 1,
1075         },
1076         [ICMP_ECHO] = {
1077                 .handler = icmp_echo,
1078         },
1079         [9] = {
1080                 .handler = icmp_discard,
1081                 .error = 1,
1082         },
1083         [10] = {
1084                 .handler = icmp_discard,
1085                 .error = 1,
1086         },
1087         [ICMP_TIME_EXCEEDED] = {
1088                 .handler = icmp_unreach,
1089                 .error = 1,
1090         },
1091         [ICMP_PARAMETERPROB] = {
1092                 .handler = icmp_unreach,
1093                 .error = 1,
1094         },
1095         [ICMP_TIMESTAMP] = {
1096                 .handler = icmp_timestamp,
1097         },
1098         [ICMP_TIMESTAMPREPLY] = {
1099                 .handler = icmp_discard,
1100         },
1101         [ICMP_INFO_REQUEST] = {
1102                 .handler = icmp_discard,
1103         },
1104         [ICMP_INFO_REPLY] = {
1105                 .handler = icmp_discard,
1106         },
1107         [ICMP_ADDRESS] = {
1108                 .handler = icmp_address,
1109         },
1110         [ICMP_ADDRESSREPLY] = {
1111                 .handler = icmp_address_reply,
1112         },
1113 };
1114
1115 static void __net_exit icmp_sk_exit(struct net *net)
1116 {
1117         int i;
1118
1119         for_each_possible_cpu(i)
1120                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1121         kfree(net->ipv4.icmp_sk);
1122         net->ipv4.icmp_sk = NULL;
1123 }
1124
1125 static int __net_init icmp_sk_init(struct net *net)
1126 {
1127         int i, err;
1128
1129         net->ipv4.icmp_sk =
1130                 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1131         if (net->ipv4.icmp_sk == NULL)
1132                 return -ENOMEM;
1133
1134         for_each_possible_cpu(i) {
1135                 struct sock *sk;
1136
1137                 err = inet_ctl_sock_create(&sk, PF_INET,
1138                                            SOCK_RAW, IPPROTO_ICMP, net);
1139                 if (err < 0)
1140                         goto fail;
1141
1142                 net->ipv4.icmp_sk[i] = sk;
1143
1144                 /* Enough space for 2 64K ICMP packets, including
1145                  * sk_buff struct overhead.
1146                  */
1147                 sk->sk_sndbuf =
1148                         (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1149
1150                 /*
1151                  * Speedup sock_wfree()
1152                  */
1153                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1154                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1155         }
1156
1157         /* Control parameters for ECHO replies. */
1158         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1159         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1160
1161         /* Control parameter - ignore bogus broadcast responses? */
1162         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1163
1164         /*
1165          *      Configurable global rate limit.
1166          *
1167          *      ratelimit defines tokens/packet consumed for dst->rate_token
1168          *      bucket ratemask defines which icmp types are ratelimited by
1169          *      setting it's bit position.
1170          *
1171          *      default:
1172          *      dest unreachable (3), source quench (4),
1173          *      time exceeded (11), parameter problem (12)
1174          */
1175
1176         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1177         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1178         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1179
1180         return 0;
1181
1182 fail:
1183         for_each_possible_cpu(i)
1184                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1185         kfree(net->ipv4.icmp_sk);
1186         return err;
1187 }
1188
1189 static struct pernet_operations __net_initdata icmp_sk_ops = {
1190        .init = icmp_sk_init,
1191        .exit = icmp_sk_exit,
1192 };
1193
1194 int __init icmp_init(void)
1195 {
1196         return register_pernet_subsys(&icmp_sk_ops);
1197 }