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 <net/ping.h>
  87 #include <linux/skbuff.h>
  88 #include <net/sock.h>
  89 #include <linux/errno.h>
  90 #include <linux/timer.h>
  91 #include <linux/init.h>
  92 #include <asm/system.h>
  93 #include <asm/uaccess.h>
  94 #include <net/checksum.h>
  95 #include <net/xfrm.h>
  96 #include <net/inet_common.h>
  97
  98 /*
  99  *      Build xmit assembly blocks
 100  */
 101
 102 struct icmp_bxm {
 103         struct sk_buff *skb;
 104         int offset;
 105         int data_len;
 106
 107         struct {
 108                 struct icmphdr icmph;
 109                 __be32         times[3];
 110         } data;
 111         int head_len;
 112         struct ip_options_data replyopts;
 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                                       struct flowi4 *fl4, 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, fl4->daddr, 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 flowi4 *fl4,
 295                             struct ipcm_cookie *ipc, struct rtable **rt)
 296 {
 297         struct sock *sk;
 298         struct sk_buff *skb;
 299
 300         sk = icmp_sk(dev_net((*rt)->dst.dev));
 301         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 302                            icmp_param->data_len+icmp_param->head_len,
 303                            icmp_param->head_len,
 304                            ipc, rt, MSG_DONTWAIT) < 0) {
 305                 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
 306                 ip_flush_pending_frames(sk);
 307         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 308                 struct icmphdr *icmph = icmp_hdr(skb);
 309                 __wsum csum = 0;
 310                 struct sk_buff *skb1;
 311
 312                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 313                         csum = csum_add(csum, skb1->csum);
 314                 }
 315                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 316                                                  (char *)icmph,
 317                                                  icmp_param->head_len, csum);
 318                 icmph->checksum = csum_fold(csum);
 319                 skb->ip_summed = CHECKSUM_NONE;
 320                 ip_push_pending_frames(sk, fl4);
 321         }
 322 }
 323
 324 /*
 325  *      Driving logic for building and sending ICMP messages.
 326  */
 327
 328 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 329 {
 330         struct ipcm_cookie ipc;
 331         struct rtable *rt = skb_rtable(skb);
 332         struct net *net = dev_net(rt->dst.dev);
 333         struct flowi4 fl4;
 334         struct sock *sk;
 335         struct inet_sock *inet;
 336         __be32 daddr;
 337
 338         if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
 339                 return;
 340
 341         sk = icmp_xmit_lock(net);
 342         if (sk == NULL)
 343                 return;
 344         inet = inet_sk(sk);
 345
 346         icmp_param->data.icmph.checksum = 0;
 347
 348         inet->tos = ip_hdr(skb)->tos;
 349         daddr = ipc.addr = ip_hdr(skb)->saddr;
 350         ipc.opt = NULL;
 351         ipc.tx_flags = 0;
 352         if (icmp_param->replyopts.opt.opt.optlen) {
 353                 ipc.opt = &icmp_param->replyopts.opt;
 354                 if (ipc.opt->opt.srr)
 355                         daddr = icmp_param->replyopts.opt.opt.faddr;
 356         }
 357         memset(&fl4, 0, sizeof(fl4));
 358         fl4.daddr = daddr;
 359         fl4.saddr = rt->rt_spec_dst;
 360         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 361         fl4.flowi4_proto = IPPROTO_ICMP;
 362         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 363         rt = ip_route_output_key(net, &fl4);
 364         if (IS_ERR(rt))
 365                 goto out_unlock;
 366         if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
 367                                icmp_param->data.icmph.code))
 368                 icmp_push_reply(icmp_param, &fl4, &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,
 375                                         struct flowi4 *fl4,
 376                                         struct sk_buff *skb_in,
 377                                         const struct iphdr *iph,
 378                                         __be32 saddr, u8 tos,
 379                                         int type, int code,
 380                                         struct icmp_bxm *param)
 381 {
 382         struct rtable *rt, *rt2;
 383         struct flowi4 fl4_dec;
 384         int err;
 385
 386         memset(fl4, 0, sizeof(*fl4));
 387         fl4->daddr = (param->replyopts.opt.opt.srr ?
 388                       param->replyopts.opt.opt.faddr : iph->saddr);
 389         fl4->saddr = saddr;
 390         fl4->flowi4_tos = RT_TOS(tos);
 391         fl4->flowi4_proto = IPPROTO_ICMP;
 392         fl4->fl4_icmp_type = type;
 393         fl4->fl4_icmp_code = code;
 394         security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 395         rt = __ip_route_output_key(net, fl4);
 396         if (IS_ERR(rt))
 397                 return rt;
 398
 399         /* No need to clone since we're just using its address. */
 400         rt2 = rt;
 401
 402         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 403                                            flowi4_to_flowi(fl4), NULL, 0);
 404         if (!IS_ERR(rt)) {
 405                 if (rt != rt2)
 406                         return rt;
 407         } else if (PTR_ERR(rt) == -EPERM) {
 408                 rt = NULL;
 409         } else
 410                 return rt;
 411
 412         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 413         if (err)
 414                 goto relookup_failed;
 415
 416         if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
 417                 rt2 = __ip_route_output_key(net, &fl4_dec);
 418                 if (IS_ERR(rt2))
 419                         err = PTR_ERR(rt2);
 420         } else {
 421                 struct flowi4 fl4_2 = {};
 422                 unsigned long orefdst;
 423
 424                 fl4_2.daddr = fl4_dec.saddr;
 425                 rt2 = ip_route_output_key(net, &fl4_2);
 426                 if (IS_ERR(rt2)) {
 427                         err = PTR_ERR(rt2);
 428                         goto relookup_failed;
 429                 }
 430                 /* Ugh! */
 431                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 432                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 433                                      RT_TOS(tos), rt2->dst.dev);
 434
 435                 dst_release(&rt2->dst);
 436                 rt2 = skb_rtable(skb_in);
 437                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 438         }
 439
 440         if (err)
 441                 goto relookup_failed;
 442
 443         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 444                                             flowi4_to_flowi(&fl4_dec), NULL,
 445                                             XFRM_LOOKUP_ICMP);
 446         if (!IS_ERR(rt2)) {
 447                 dst_release(&rt->dst);
 448                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 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         struct flowi4 fl4;
 485         __be32 saddr;
 486         u8  tos;
 487         struct net *net;
 488         struct sock *sk;
 489
 490         if (!rt)
 491                 goto out;
 492         net = dev_net(rt->dst.dev);
 493
 494         /*
 495          *      Find the original header. It is expected to be valid, of course.
 496          *      Check this, icmp_send is called from the most obscure devices
 497          *      sometimes.
 498          */
 499         iph = ip_hdr(skb_in);
 500
 501         if ((u8 *)iph < skb_in->head ||
 502             (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
 503                 goto out;
 504
 505         /*
 506          *      No replies to physical multicast/broadcast
 507          */
 508         if (skb_in->pkt_type != PACKET_HOST)
 509                 goto out;
 510
 511         /*
 512          *      Now check at the protocol level
 513          */
 514         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 515                 goto out;
 516
 517         /*
 518          *      Only reply to fragment 0. We byte re-order the constant
 519          *      mask for efficiency.
 520          */
 521         if (iph->frag_off & htons(IP_OFFSET))
 522                 goto out;
 523
 524         /*
 525          *      If we send an ICMP error to an ICMP error a mess would result..
 526          */
 527         if (icmp_pointers[type].error) {
 528                 /*
 529                  *      We are an error, check if we are replying to an
 530                  *      ICMP error
 531                  */
 532                 if (iph->protocol == IPPROTO_ICMP) {
 533                         u8 _inner_type, *itp;
 534
 535                         itp = skb_header_pointer(skb_in,
 536                                                  skb_network_header(skb_in) +
 537                                                  (iph->ihl << 2) +
 538                                                  offsetof(struct icmphdr,
 539                                                           type) -
 540                                                  skb_in->data,
 541                                                  sizeof(_inner_type),
 542                                                  &_inner_type);
 543                         if (itp == NULL)
 544                                 goto out;
 545
 546                         /*
 547                          *      Assume any unknown ICMP type is an error. This
 548                          *      isn't specified by the RFC, but think about it..
 549                          */
 550                         if (*itp > NR_ICMP_TYPES ||
 551                             icmp_pointers[*itp].error)
 552                                 goto out;
 553                 }
 554         }
 555
 556         sk = icmp_xmit_lock(net);
 557         if (sk == NULL)
 558                 return;
 559
 560         /*
 561          *      Construct source address and options.
 562          */
 563
 564         saddr = iph->daddr;
 565         if (!(rt->rt_flags & RTCF_LOCAL)) {
 566                 struct net_device *dev = NULL;
 567
 568                 rcu_read_lock();
 569                 if (rt_is_input_route(rt) &&
 570                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 571                         dev = dev_get_by_index_rcu(net, rt->rt_iif);
 572
 573                 if (dev)
 574                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 575                 else
 576                         saddr = 0;
 577                 rcu_read_unlock();
 578         }
 579
 580         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 581                                            IPTOS_PREC_INTERNETCONTROL) :
 582                                           iph->tos;
 583
 584         if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
 585                 goto out_unlock;
 586
 587
 588         /*
 589          *      Prepare data for ICMP header.
 590          */
 591
 592         icmp_param.data.icmph.type       = type;
 593         icmp_param.data.icmph.code       = code;
 594         icmp_param.data.icmph.un.gateway = info;
 595         icmp_param.data.icmph.checksum   = 0;
 596         icmp_param.skb    = skb_in;
 597         icmp_param.offset = skb_network_offset(skb_in);
 598         inet_sk(sk)->tos = tos;
 599         ipc.addr = iph->saddr;
 600         ipc.opt = &icmp_param.replyopts.opt;
 601         ipc.tx_flags = 0;
 602
 603         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
 604                                type, code, &icmp_param);
 605         if (IS_ERR(rt))
 606                 goto out_unlock;
 607
 608         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 609                 goto ende;
 610
 611         /* RFC says return as much as we can without exceeding 576 bytes. */
 612
 613         room = dst_mtu(&rt->dst);
 614         if (room > 576)
 615                 room = 576;
 616         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 617         room -= sizeof(struct icmphdr);
 618
 619         icmp_param.data_len = skb_in->len - icmp_param.offset;
 620         if (icmp_param.data_len > room)
 621                 icmp_param.data_len = room;
 622         icmp_param.head_len = sizeof(struct icmphdr);
 623
 624         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 625 ende:
 626         ip_rt_put(rt);
 627 out_unlock:
 628         icmp_xmit_unlock(sk);
 629 out:;
 630 }
 631 EXPORT_SYMBOL(icmp_send);
 632
 633
 634 /*
 635  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
 636  */
 637
 638 static void icmp_unreach(struct sk_buff *skb)
 639 {
 640         const struct iphdr *iph;
 641         struct icmphdr *icmph;
 642         int hash, protocol;
 643         const struct net_protocol *ipprot;
 644         u32 info = 0;
 645         struct net *net;
 646
 647         net = dev_net(skb_dst(skb)->dev);
 648
 649         /*
 650          *      Incomplete header ?
 651          *      Only checks for the IP header, there should be an
 652          *      additional check for longer headers in upper levels.
 653          */
 654
 655         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 656                 goto out_err;
 657
 658         icmph = icmp_hdr(skb);
 659         iph   = (const struct iphdr *)skb->data;
 660
 661         if (iph->ihl < 5) /* Mangled header, drop. */
 662                 goto out_err;
 663
 664         if (icmph->type == ICMP_DEST_UNREACH) {
 665                 switch (icmph->code & 15) {
 666                 case ICMP_NET_UNREACH:
 667                 case ICMP_HOST_UNREACH:
 668                 case ICMP_PROT_UNREACH:
 669                 case ICMP_PORT_UNREACH:
 670                         break;
 671                 case ICMP_FRAG_NEEDED:
 672                         if (ipv4_config.no_pmtu_disc) {
 673                                 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
 674                                                &iph->daddr);
 675                         } else {
 676                                 info = ip_rt_frag_needed(net, iph,
 677                                                          ntohs(icmph->un.frag.mtu),
 678                                                          skb->dev);
 679                                 if (!info)
 680                                         goto out;
 681                         }
 682                         break;
 683                 case ICMP_SR_FAILED:
 684                         LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
 685                                        &iph->daddr);
 686                         break;
 687                 default:
 688                         break;
 689                 }
 690                 if (icmph->code > NR_ICMP_UNREACH)
 691                         goto out;
 692         } else if (icmph->type == ICMP_PARAMETERPROB)
 693                 info = ntohl(icmph->un.gateway) >> 24;
 694
 695         /*
 696          *      Throw it at our lower layers
 697          *
 698          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 699          *                header.
 700          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 701          *                transport layer.
 702          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 703          *                transport layer.
 704          */
 705
 706         /*
 707          *      Check the other end isn't violating RFC 1122. Some routers send
 708          *      bogus responses to broadcast frames. If you see this message
 709          *      first check your netmask matches at both ends, if it does then
 710          *      get the other vendor to fix their kit.
 711          */
 712
 713         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 714             inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
 715                 if (net_ratelimit())
 716                         printk(KERN_WARNING "%pI4 sent an invalid ICMP "
 717                                             "type %u, code %u "
 718                                             "error to a broadcast: %pI4 on %s\n",
 719                                &ip_hdr(skb)->saddr,
 720                                icmph->type, icmph->code,
 721                                &iph->daddr,
 722                                skb->dev->name);
 723                 goto out;
 724         }
 725
 726         /* Checkin full IP header plus 8 bytes of protocol to
 727          * avoid additional coding at protocol handlers.
 728          */
 729         if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
 730                 goto out;
 731
 732         iph = (const struct iphdr *)skb->data;
 733         protocol = iph->protocol;
 734
 735         /*
 736          *      Deliver ICMP message to raw sockets. Pretty useless feature?
 737          */
 738         raw_icmp_error(skb, protocol, info);
 739
 740         hash = protocol & (MAX_INET_PROTOS - 1);
 741         rcu_read_lock();
 742         ipprot = rcu_dereference(inet_protos[hash]);
 743         if (ipprot && ipprot->err_handler)
 744                 ipprot->err_handler(skb, info);
 745         rcu_read_unlock();
 746
 747 out:
 748         return;
 749 out_err:
 750         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 751         goto out;
 752 }
 753
 754
 755 /*
 756  *      Handle ICMP_REDIRECT.
 757  */
 758
 759 static void icmp_redirect(struct sk_buff *skb)
 760 {
 761         const struct iphdr *iph;
 762
 763         if (skb->len < sizeof(struct iphdr))
 764                 goto out_err;
 765
 766         /*
 767          *      Get the copied header of the packet that caused the redirect
 768          */
 769         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 770                 goto out;
 771
 772         iph = (const struct iphdr *)skb->data;
 773
 774         switch (icmp_hdr(skb)->code & 7) {
 775         case ICMP_REDIR_NET:
 776         case ICMP_REDIR_NETTOS:
 777                 /*
 778                  * As per RFC recommendations now handle it as a host redirect.
 779                  */
 780         case ICMP_REDIR_HOST:
 781         case ICMP_REDIR_HOSTTOS:
 782                 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
 783                                icmp_hdr(skb)->un.gateway,
 784                                iph->saddr, skb->dev);
 785                 break;
 786         }
 787
 788         /* Ping wants to see redirects.
 789          * Let's pretend they are errors of sorts... */
 790         if (iph->protocol == IPPROTO_ICMP &&
 791             iph->ihl >= 5 &&
 792             pskb_may_pull(skb, (iph->ihl<<2)+8)) {
 793                 ping_err(skb, icmp_hdr(skb)->un.gateway);
 794         }
 795
 796 out:
 797         return;
 798 out_err:
 799         ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
 800         goto out;
 801 }
 802
 803 /*
 804  *      Handle ICMP_ECHO ("ping") requests.
 805  *
 806  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 807  *                requests.
 808  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 809  *                included in the reply.
 810  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 811  *                echo requests, MUST have default=NOT.
 812  *      See also WRT handling of options once they are done and working.
 813  */
 814
 815 static void icmp_echo(struct sk_buff *skb)
 816 {
 817         struct net *net;
 818
 819         net = dev_net(skb_dst(skb)->dev);
 820         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 821                 struct icmp_bxm icmp_param;
 822
 823                 icmp_param.data.icmph      = *icmp_hdr(skb);
 824                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 825                 icmp_param.skb             = skb;
 826                 icmp_param.offset          = 0;
 827                 icmp_param.data_len        = skb->len;
 828                 icmp_param.head_len        = sizeof(struct icmphdr);
 829                 icmp_reply(&icmp_param, skb);
 830         }
 831 }
 832
 833 /*
 834  *      Handle ICMP Timestamp requests.
 835  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 836  *                SHOULD be in the kernel for minimum random latency.
 837  *                MUST be accurate to a few minutes.
 838  *                MUST be updated at least at 15Hz.
 839  */
 840 static void icmp_timestamp(struct sk_buff *skb)
 841 {
 842         struct timespec tv;
 843         struct icmp_bxm icmp_param;
 844         /*
 845          *      Too short.
 846          */
 847         if (skb->len < 4)
 848                 goto out_err;
 849
 850         /*
 851          *      Fill in the current time as ms since midnight UT:
 852          */
 853         getnstimeofday(&tv);
 854         icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
 855                                          tv.tv_nsec / NSEC_PER_MSEC);
 856         icmp_param.data.times[2] = icmp_param.data.times[1];
 857         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 858                 BUG();
 859         icmp_param.data.icmph      = *icmp_hdr(skb);
 860         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 861         icmp_param.data.icmph.code = 0;
 862         icmp_param.skb             = skb;
 863         icmp_param.offset          = 0;
 864         icmp_param.data_len        = 0;
 865         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 866         icmp_reply(&icmp_param, skb);
 867 out:
 868         return;
 869 out_err:
 870         ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 871         goto out;
 872 }
 873
 874
 875 /*
 876  *      Handle ICMP_ADDRESS_MASK requests.  (RFC950)
 877  *
 878  * RFC1122 (3.2.2.9).  A host MUST only send replies to
 879  * ADDRESS_MASK requests if it's been configured as an address mask
 880  * agent.  Receiving a request doesn't constitute implicit permission to
 881  * act as one. Of course, implementing this correctly requires (SHOULD)
 882  * a way to turn the functionality on and off.  Another one for sysctl(),
 883  * I guess. -- MS
 884  *
 885  * RFC1812 (4.3.3.9).   A router MUST implement it.
 886  *                      A router SHOULD have switch turning it on/off.
 887  *                      This switch MUST be ON by default.
 888  *
 889  * Gratuitous replies, zero-source replies are not implemented,
 890  * that complies with RFC. DO NOT implement them!!! All the idea
 891  * of broadcast addrmask replies as specified in RFC950 is broken.
 892  * The problem is that it is not uncommon to have several prefixes
 893  * on one physical interface. Moreover, addrmask agent can even be
 894  * not aware of existing another prefixes.
 895  * If source is zero, addrmask agent cannot choose correct prefix.
 896  * Gratuitous mask announcements suffer from the same problem.
 897  * RFC1812 explains it, but still allows to use ADDRMASK,
 898  * that is pretty silly. --ANK
 899  *
 900  * All these rules are so bizarre, that I removed kernel addrmask
 901  * support at all. It is wrong, it is obsolete, nobody uses it in
 902  * any case. --ANK
 903  *
 904  * Furthermore you can do it with a usermode address agent program
 905  * anyway...
 906  */
 907
 908 static void icmp_address(struct sk_buff *skb)
 909 {
 910 #if 0
 911         if (net_ratelimit())
 912                 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
 913 #endif
 914 }
 915
 916 /*
 917  * RFC1812 (4.3.3.9).   A router SHOULD listen all replies, and complain
 918  *                      loudly if an inconsistency is found.
 919  * called with rcu_read_lock()
 920  */
 921
 922 static void icmp_address_reply(struct sk_buff *skb)
 923 {
 924         struct rtable *rt = skb_rtable(skb);
 925         struct net_device *dev = skb->dev;
 926         struct in_device *in_dev;
 927         struct in_ifaddr *ifa;
 928
 929         if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
 930                 return;
 931
 932         in_dev = __in_dev_get_rcu(dev);
 933         if (!in_dev)
 934                 return;
 935
 936         if (in_dev->ifa_list &&
 937             IN_DEV_LOG_MARTIANS(in_dev) &&
 938             IN_DEV_FORWARD(in_dev)) {
 939                 __be32 _mask, *mp;
 940
 941                 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
 942                 BUG_ON(mp == NULL);
 943                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
 944                         if (*mp == ifa->ifa_mask &&
 945                             inet_ifa_match(ip_hdr(skb)->saddr, ifa))
 946                                 break;
 947                 }
 948                 if (!ifa && net_ratelimit()) {
 949                         printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
 950                                mp, dev->name, &ip_hdr(skb)->saddr);
 951                 }
 952         }
 953 }
 954
 955 static void icmp_discard(struct sk_buff *skb)
 956 {
 957 }
 958
 959 /*
 960  *      Deal with incoming ICMP packets.
 961  */
 962 int icmp_rcv(struct sk_buff *skb)
 963 {
 964         struct icmphdr *icmph;
 965         struct rtable *rt = skb_rtable(skb);
 966         struct net *net = dev_net(rt->dst.dev);
 967
 968         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 969                 struct sec_path *sp = skb_sec_path(skb);
 970                 int nh;
 971
 972                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 973                                  XFRM_STATE_ICMP))
 974                         goto drop;
 975
 976                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
 977                         goto drop;
 978
 979                 nh = skb_network_offset(skb);
 980                 skb_set_network_header(skb, sizeof(*icmph));
 981
 982                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 983                         goto drop;
 984
 985                 skb_set_network_header(skb, nh);
 986         }
 987
 988         ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
 989
 990         switch (skb->ip_summed) {
 991         case CHECKSUM_COMPLETE:
 992                 if (!csum_fold(skb->csum))
 993                         break;
 994                 /* fall through */
 995         case CHECKSUM_NONE:
 996                 skb->csum = 0;
 997                 if (__skb_checksum_complete(skb))
 998                         goto error;
 999         }
1000
1001         if (!pskb_pull(skb, sizeof(*icmph)))
1002                 goto error;
1003
1004         icmph = icmp_hdr(skb);
1005
1006         ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1007         /*
1008          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1009          *
1010          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1011          *                discarded.
1012          */
1013         if (icmph->type > NR_ICMP_TYPES)
1014                 goto error;
1015
1016
1017         /*
1018          *      Parse the ICMP message
1019          */
1020
1021         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1022                 /*
1023                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1024                  *        silently ignored (we let user decide with a sysctl).
1025                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1026                  *        discarded if to broadcast/multicast.
1027                  */
1028                 if ((icmph->type == ICMP_ECHO ||
1029                      icmph->type == ICMP_TIMESTAMP) &&
1030                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1031                         goto error;
1032                 }
1033                 if (icmph->type != ICMP_ECHO &&
1034                     icmph->type != ICMP_TIMESTAMP &&
1035                     icmph->type != ICMP_ADDRESS &&
1036                     icmph->type != ICMP_ADDRESSREPLY) {
1037                         goto error;
1038                 }
1039         }
1040
1041         icmp_pointers[icmph->type].handler(skb);
1042
1043 drop:
1044         kfree_skb(skb);
1045         return 0;
1046 error:
1047         ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1048         goto drop;
1049 }
1050
1051 /*
1052  *      This table is the definition of how we handle ICMP.
1053  */
1054 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1055         [ICMP_ECHOREPLY] = {
1056                 .handler = ping_rcv,
1057         },
1058         [1] = {
1059                 .handler = icmp_discard,
1060                 .error = 1,
1061         },
1062         [2] = {
1063                 .handler = icmp_discard,
1064                 .error = 1,
1065         },
1066         [ICMP_DEST_UNREACH] = {
1067                 .handler = icmp_unreach,
1068                 .error = 1,
1069         },
1070         [ICMP_SOURCE_QUENCH] = {
1071                 .handler = icmp_unreach,
1072                 .error = 1,
1073         },
1074         [ICMP_REDIRECT] = {
1075                 .handler = icmp_redirect,
1076                 .error = 1,
1077         },
1078         [6] = {
1079                 .handler = icmp_discard,
1080                 .error = 1,
1081         },
1082         [7] = {
1083                 .handler = icmp_discard,
1084                 .error = 1,
1085         },
1086         [ICMP_ECHO] = {
1087                 .handler = icmp_echo,
1088         },
1089         [9] = {
1090                 .handler = icmp_discard,
1091                 .error = 1,
1092         },
1093         [10] = {
1094                 .handler = icmp_discard,
1095                 .error = 1,
1096         },
1097         [ICMP_TIME_EXCEEDED] = {
1098                 .handler = icmp_unreach,
1099                 .error = 1,
1100         },
1101         [ICMP_PARAMETERPROB] = {
1102                 .handler = icmp_unreach,
1103                 .error = 1,
1104         },
1105         [ICMP_TIMESTAMP] = {
1106                 .handler = icmp_timestamp,
1107         },
1108         [ICMP_TIMESTAMPREPLY] = {
1109                 .handler = icmp_discard,
1110         },
1111         [ICMP_INFO_REQUEST] = {
1112                 .handler = icmp_discard,
1113         },
1114         [ICMP_INFO_REPLY] = {
1115                 .handler = icmp_discard,
1116         },
1117         [ICMP_ADDRESS] = {
1118                 .handler = icmp_address,
1119         },
1120         [ICMP_ADDRESSREPLY] = {
1121                 .handler = icmp_address_reply,
1122         },
1123 };
1124
1125 static void __net_exit icmp_sk_exit(struct net *net)
1126 {
1127         int i;
1128
1129         for_each_possible_cpu(i)
1130                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1131         kfree(net->ipv4.icmp_sk);
1132         net->ipv4.icmp_sk = NULL;
1133 }
1134
1135 static int __net_init icmp_sk_init(struct net *net)
1136 {
1137         int i, err;
1138
1139         net->ipv4.icmp_sk =
1140                 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1141         if (net->ipv4.icmp_sk == NULL)
1142                 return -ENOMEM;
1143
1144         for_each_possible_cpu(i) {
1145                 struct sock *sk;
1146
1147                 err = inet_ctl_sock_create(&sk, PF_INET,
1148                                            SOCK_RAW, IPPROTO_ICMP, net);
1149                 if (err < 0)
1150                         goto fail;
1151
1152                 net->ipv4.icmp_sk[i] = sk;
1153
1154                 /* Enough space for 2 64K ICMP packets, including
1155                  * sk_buff/skb_shared_info struct overhead.
1156                  */
1157                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1158
1159                 /*
1160                  * Speedup sock_wfree()
1161                  */
1162                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1163                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1164         }
1165
1166         /* Control parameters for ECHO replies. */
1167         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1168         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1169
1170         /* Control parameter - ignore bogus broadcast responses? */
1171         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1172
1173         /*
1174          *      Configurable global rate limit.
1175          *
1176          *      ratelimit defines tokens/packet consumed for dst->rate_token
1177          *      bucket ratemask defines which icmp types are ratelimited by
1178          *      setting it's bit position.
1179          *
1180          *      default:
1181          *      dest unreachable (3), source quench (4),
1182          *      time exceeded (11), parameter problem (12)
1183          */
1184
1185         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1186         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1187         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1188
1189         return 0;
1190
1191 fail:
1192         for_each_possible_cpu(i)
1193                 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1194         kfree(net->ipv4.icmp_sk);
1195         return err;
1196 }
1197
1198 static struct pernet_operations __net_initdata icmp_sk_ops = {
1199        .init = icmp_sk_init,
1200        .exit = icmp_sk_exit,
1201 };
1202
1203 int __init icmp_init(void)
1204 {
1205         return register_pernet_subsys(&icmp_sk_ops);
1206 }