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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  66
  67 #include <linux/module.h>
  68 #include <linux/types.h>
  69 #include <linux/jiffies.h>
  70 #include <linux/kernel.h>
  71 #include <linux/fcntl.h>
  72 #include <linux/socket.h>
  73 #include <linux/in.h>
  74 #include <linux/inet.h>
  75 #include <linux/inetdevice.h>
  76 #include <linux/netdevice.h>
  77 #include <linux/string.h>
  78 #include <linux/netfilter_ipv4.h>
  79 #include <linux/slab.h>
  80 #include <net/snmp.h>
  81 #include <net/ip.h>
  82 #include <net/route.h>
  83 #include <net/protocol.h>
  84 #include <net/icmp.h>
  85 #include <net/tcp.h>
  86 #include <net/udp.h>
  87 #include <net/raw.h>
  88 #include <net/ping.h>
  89 #include <linux/skbuff.h>
  90 #include <net/sock.h>
  91 #include <linux/errno.h>
  92 #include <linux/timer.h>
  93 #include <linux/init.h>
  94 #include <asm/uaccess.h>
  95 #include <net/checksum.h>
  96 #include <net/xfrm.h>
  97 #include <net/inet_common.h>
  98 #include <net/ip_fib.h>
  99 #include <net/l3mdev.h>
 100
 101 /*
 102  *      Build xmit assembly blocks
 103  */
 104
 105 struct icmp_bxm {
 106         struct sk_buff *skb;
 107         int offset;
 108         int data_len;
 109
 110         struct {
 111                 struct icmphdr icmph;
 112                 __be32         times[3];
 113         } data;
 114         int head_len;
 115         struct ip_options_data replyopts;
 116 };
 117
 118 /* An array of errno for error messages from dest unreach. */
 119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 120
 121 const struct icmp_err icmp_err_convert[] = {
 122         {
 123                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 124                 .fatal = 0,
 125         },
 126         {
 127                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 128                 .fatal = 0,
 129         },
 130         {
 131                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 132                 .fatal = 1,
 133         },
 134         {
 135                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 136                 .fatal = 1,
 137         },
 138         {
 139                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 140                 .fatal = 0,
 141         },
 142         {
 143                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 144                 .fatal = 0,
 145         },
 146         {
 147                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 148                 .fatal = 1,
 149         },
 150         {
 151                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 152                 .fatal = 1,
 153         },
 154         {
 155                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 156                 .fatal = 1,
 157         },
 158         {
 159                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 160                 .fatal = 1,
 161         },
 162         {
 163                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 164                 .fatal = 1,
 165         },
 166         {
 167                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 168                 .fatal = 0,
 169         },
 170         {
 171                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 172                 .fatal = 0,
 173         },
 174         {
 175                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 176                 .fatal = 1,
 177         },
 178         {
 179                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 180                 .fatal = 1,
 181         },
 182         {
 183                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 184                 .fatal = 1,
 185         },
 186 };
 187 EXPORT_SYMBOL(icmp_err_convert);
 188
 189 /*
 190  *      ICMP control array. This specifies what to do with each ICMP.
 191  */
 192
 193 struct icmp_control {
 194         bool (*handler)(struct sk_buff *skb);
 195         short   error;          /* This ICMP is classed as an error message */
 196 };
 197
 198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 199
 200 /*
 201  *      The ICMP socket(s). This is the most convenient way to flow control
 202  *      our ICMP output as well as maintain a clean interface throughout
 203  *      all layers. All Socketless IP sends will soon be gone.
 204  *
 205  *      On SMP we have one ICMP socket per-cpu.
 206  */
 207 static struct sock *icmp_sk(struct net *net)
 208 {
 209         return *this_cpu_ptr(net->ipv4.icmp_sk);
 210 }
 211
 212 static inline struct sock *icmp_xmit_lock(struct net *net)
 213 {
 214         struct sock *sk;
 215
 216         local_bh_disable();
 217
 218         sk = icmp_sk(net);
 219
 220         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 221                 /* This can happen if the output path signals a
 222                  * dst_link_failure() for an outgoing ICMP packet.
 223                  */
 224                 local_bh_enable();
 225                 return NULL;
 226         }
 227         return sk;
 228 }
 229
 230 static inline void icmp_xmit_unlock(struct sock *sk)
 231 {
 232         spin_unlock_bh(&sk->sk_lock.slock);
 233 }
 234
 235 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 236 int sysctl_icmp_msgs_burst __read_mostly = 50;
 237
 238 static struct {
 239         spinlock_t      lock;
 240         u32             credit;
 241         u32             stamp;
 242 } icmp_global = {
 243         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 244 };
 245
 246 /**
 247  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 248  *
 249  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 250  * Returns false if we reached the limit and can not send another packet.
 251  * Note: called with BH disabled
 252  */
 253 bool icmp_global_allow(void)
 254 {
 255         u32 credit, delta, incr = 0, now = (u32)jiffies;
 256         bool rc = false;
 257
 258         /* Check if token bucket is empty and cannot be refilled
 259          * without taking the spinlock. The READ_ONCE() are paired
 260          * with the following WRITE_ONCE() in this same function.
 261          */
 262         if (!READ_ONCE(icmp_global.credit)) {
 263                 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 264                 if (delta < HZ / 50)
 265                         return false;
 266         }
 267
 268         spin_lock(&icmp_global.lock);
 269         delta = min_t(u32, now - icmp_global.stamp, HZ);
 270         if (delta >= HZ / 50) {
 271                 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
 272                 if (incr)
 273                         WRITE_ONCE(icmp_global.stamp, now);
 274         }
 275         credit = min_t(u32, icmp_global.credit + incr,
 276                        READ_ONCE(sysctl_icmp_msgs_burst));
 277         if (credit) {
 278                 /* We want to use a credit of one in average, but need to randomize
 279                  * it for security reasons.
 280                  */
 281                 credit = max_t(int, credit - prandom_u32_max(3), 0);
 282                 rc = true;
 283         }
 284         WRITE_ONCE(icmp_global.credit, credit);
 285         spin_unlock(&icmp_global.lock);
 286         return rc;
 287 }
 288 EXPORT_SYMBOL(icmp_global_allow);
 289
 290 /*
 291  *      Send an ICMP frame.
 292  */
 293
 294 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 295                                struct flowi4 *fl4, int type, int code)
 296 {
 297         struct dst_entry *dst = &rt->dst;
 298         bool rc = true;
 299
 300         if (type > NR_ICMP_TYPES)
 301                 goto out;
 302
 303         /* Don't limit PMTU discovery. */
 304         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 305                 goto out;
 306
 307         /* No rate limit on loopback */
 308         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 309                 goto out;
 310
 311         /* Limit if icmp type is enabled in ratemask. */
 312         if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 313                 goto out;
 314
 315         rc = false;
 316         if (icmp_global_allow()) {
 317                 int vif = l3mdev_master_ifindex(dst->dev);
 318                 struct inet_peer *peer;
 319
 320                 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 321                 rc = inet_peer_xrlim_allow(peer,
 322                                            net->ipv4.sysctl_icmp_ratelimit);
 323                 if (peer)
 324                         inet_putpeer(peer);
 325         }
 326 out:
 327         return rc;
 328 }
 329
 330 /*
 331  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 332  */
 333 void icmp_out_count(struct net *net, unsigned char type)
 334 {
 335         ICMPMSGOUT_INC_STATS(net, type);
 336         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 337 }
 338
 339 /*
 340  *      Checksum each fragment, and on the first include the headers and final
 341  *      checksum.
 342  */
 343 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 344                           struct sk_buff *skb)
 345 {
 346         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 347         __wsum csum;
 348
 349         csum = skb_copy_and_csum_bits(icmp_param->skb,
 350                                       icmp_param->offset + offset,
 351                                       to, len, 0);
 352
 353         skb->csum = csum_block_add(skb->csum, csum, odd);
 354         if (icmp_pointers[icmp_param->data.icmph.type].error)
 355                 nf_ct_attach(skb, icmp_param->skb);
 356         return 0;
 357 }
 358
 359 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 360                             struct flowi4 *fl4,
 361                             struct ipcm_cookie *ipc, struct rtable **rt)
 362 {
 363         struct sock *sk;
 364         struct sk_buff *skb;
 365
 366         sk = icmp_sk(dev_net((*rt)->dst.dev));
 367         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 368                            icmp_param->data_len+icmp_param->head_len,
 369                            icmp_param->head_len,
 370                            ipc, rt, MSG_DONTWAIT) < 0) {
 371                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 372                 ip_flush_pending_frames(sk);
 373         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 374                 struct icmphdr *icmph = icmp_hdr(skb);
 375                 __wsum csum = 0;
 376                 struct sk_buff *skb1;
 377
 378                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 379                         csum = csum_add(csum, skb1->csum);
 380                 }
 381                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 382                                                  (char *)icmph,
 383                                                  icmp_param->head_len, csum);
 384                 icmph->checksum = csum_fold(csum);
 385                 skb->ip_summed = CHECKSUM_NONE;
 386                 ip_push_pending_frames(sk, fl4);
 387         }
 388 }
 389
 390 /*
 391  *      Driving logic for building and sending ICMP messages.
 392  */
 393
 394 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 395 {
 396         struct ipcm_cookie ipc;
 397         struct rtable *rt = skb_rtable(skb);
 398         struct net *net = dev_net(rt->dst.dev);
 399         struct flowi4 fl4;
 400         struct sock *sk;
 401         struct inet_sock *inet;
 402         __be32 daddr, saddr;
 403         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 404
 405         if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
 406                 return;
 407
 408         sk = icmp_xmit_lock(net);
 409         if (!sk)
 410                 return;
 411         inet = inet_sk(sk);
 412
 413         icmp_param->data.icmph.checksum = 0;
 414
 415         inet->tos = ip_hdr(skb)->tos;
 416         sk->sk_mark = mark;
 417         daddr = ipc.addr = ip_hdr(skb)->saddr;
 418         saddr = fib_compute_spec_dst(skb);
 419         ipc.opt = NULL;
 420         ipc.tx_flags = 0;
 421         ipc.ttl = 0;
 422         ipc.tos = -1;
 423
 424         if (icmp_param->replyopts.opt.opt.optlen) {
 425                 ipc.opt = &icmp_param->replyopts.opt;
 426                 if (ipc.opt->opt.srr)
 427                         daddr = icmp_param->replyopts.opt.opt.faddr;
 428         }
 429         memset(&fl4, 0, sizeof(fl4));
 430         fl4.daddr = daddr;
 431         fl4.saddr = saddr;
 432         fl4.flowi4_mark = mark;
 433         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 434         fl4.flowi4_proto = IPPROTO_ICMP;
 435         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 436         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 437         rt = ip_route_output_key(net, &fl4);
 438         if (IS_ERR(rt))
 439                 goto out_unlock;
 440         if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
 441                                icmp_param->data.icmph.code))
 442                 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 443         ip_rt_put(rt);
 444 out_unlock:
 445         icmp_xmit_unlock(sk);
 446 }
 447
 448 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 449
 450 /* Source and destination is swapped. See ip_multipath_icmp_hash */
 451 static int icmp_multipath_hash_skb(const struct sk_buff *skb)
 452 {
 453         const struct iphdr *iph = ip_hdr(skb);
 454
 455         return fib_multipath_hash(iph->daddr, iph->saddr);
 456 }
 457
 458 #else
 459
 460 #define icmp_multipath_hash_skb(skb) (-1)
 461
 462 #endif
 463
 464 /*
 465  * The device used for looking up which routing table to use for sending an ICMP
 466  * error is preferably the source whenever it is set, which should ensure the
 467  * icmp error can be sent to the source host, else lookup using the routing
 468  * table of the destination device, else use the main routing table (index 0).
 469  */
 470 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 471 {
 472         struct net_device *route_lookup_dev = NULL;
 473
 474         if (skb->dev)
 475                 route_lookup_dev = skb->dev;
 476         else if (skb_dst(skb))
 477                 route_lookup_dev = skb_dst(skb)->dev;
 478         return route_lookup_dev;
 479 }
 480
 481 static struct rtable *icmp_route_lookup(struct net *net,
 482                                         struct flowi4 *fl4,
 483                                         struct sk_buff *skb_in,
 484                                         const struct iphdr *iph,
 485                                         __be32 saddr, u8 tos, u32 mark,
 486                                         int type, int code,
 487                                         struct icmp_bxm *param)
 488 {
 489         struct net_device *route_lookup_dev;
 490         struct rtable *rt, *rt2;
 491         struct flowi4 fl4_dec;
 492         int err;
 493
 494         memset(fl4, 0, sizeof(*fl4));
 495         fl4->daddr = (param->replyopts.opt.opt.srr ?
 496                       param->replyopts.opt.opt.faddr : iph->saddr);
 497         fl4->saddr = saddr;
 498         fl4->flowi4_mark = mark;
 499         fl4->flowi4_tos = RT_TOS(tos);
 500         fl4->flowi4_proto = IPPROTO_ICMP;
 501         fl4->fl4_icmp_type = type;
 502         fl4->fl4_icmp_code = code;
 503         route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 504         fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 505
 506         security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 507         rt = __ip_route_output_key_hash(net, fl4,
 508                                         icmp_multipath_hash_skb(skb_in));
 509         if (IS_ERR(rt))
 510                 return rt;
 511
 512         /* No need to clone since we're just using its address. */
 513         rt2 = rt;
 514
 515         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 516                                            flowi4_to_flowi(fl4), NULL, 0);
 517         if (!IS_ERR(rt)) {
 518                 if (rt != rt2)
 519                         return rt;
 520         } else if (PTR_ERR(rt) == -EPERM) {
 521                 rt = NULL;
 522         } else
 523                 return rt;
 524
 525         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 526         if (err)
 527                 goto relookup_failed;
 528
 529         if (inet_addr_type_dev_table(net, route_lookup_dev,
 530                                      fl4_dec.saddr) == RTN_LOCAL) {
 531                 rt2 = __ip_route_output_key(net, &fl4_dec);
 532                 if (IS_ERR(rt2))
 533                         err = PTR_ERR(rt2);
 534         } else {
 535                 struct flowi4 fl4_2 = {};
 536                 unsigned long orefdst;
 537
 538                 fl4_2.daddr = fl4_dec.saddr;
 539                 rt2 = ip_route_output_key(net, &fl4_2);
 540                 if (IS_ERR(rt2)) {
 541                         err = PTR_ERR(rt2);
 542                         goto relookup_failed;
 543                 }
 544                 /* Ugh! */
 545                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 546                 skb_dst_set(skb_in, NULL);
 547                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 548                                      RT_TOS(tos), rt2->dst.dev);
 549
 550                 dst_release(&rt2->dst);
 551                 rt2 = skb_rtable(skb_in);
 552                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 553         }
 554
 555         if (err)
 556                 goto relookup_failed;
 557
 558         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 559                                             flowi4_to_flowi(&fl4_dec), NULL,
 560                                             XFRM_LOOKUP_ICMP);
 561         if (!IS_ERR(rt2)) {
 562                 dst_release(&rt->dst);
 563                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 564                 rt = rt2;
 565         } else if (PTR_ERR(rt2) == -EPERM) {
 566                 if (rt)
 567                         dst_release(&rt->dst);
 568                 return rt2;
 569         } else {
 570                 err = PTR_ERR(rt2);
 571                 goto relookup_failed;
 572         }
 573         return rt;
 574
 575 relookup_failed:
 576         if (rt)
 577                 return rt;
 578         return ERR_PTR(err);
 579 }
 580
 581 /*
 582  *      Send an ICMP message in response to a situation
 583  *
 584  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 585  *                MAY send more (we do).
 586  *                      MUST NOT change this header information.
 587  *                      MUST NOT reply to a multicast/broadcast IP address.
 588  *                      MUST NOT reply to a multicast/broadcast MAC address.
 589  *                      MUST reply to only the first fragment.
 590  */
 591
 592 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 593                  const struct ip_options *opt)
 594 {
 595         struct iphdr *iph;
 596         int room;
 597         struct icmp_bxm *icmp_param;
 598         struct rtable *rt = skb_rtable(skb_in);
 599         struct ipcm_cookie ipc;
 600         struct flowi4 fl4;
 601         __be32 saddr;
 602         u8  tos;
 603         u32 mark;
 604         struct net *net;
 605         struct sock *sk;
 606
 607         if (!rt)
 608                 goto out;
 609         net = dev_net(rt->dst.dev);
 610
 611         /*
 612          *      Find the original header. It is expected to be valid, of course.
 613          *      Check this, icmp_send is called from the most obscure devices
 614          *      sometimes.
 615          */
 616         iph = ip_hdr(skb_in);
 617
 618         if ((u8 *)iph < skb_in->head ||
 619             (skb_network_header(skb_in) + sizeof(*iph)) >
 620             skb_tail_pointer(skb_in))
 621                 goto out;
 622
 623         /*
 624          *      No replies to physical multicast/broadcast
 625          */
 626         if (skb_in->pkt_type != PACKET_HOST)
 627                 goto out;
 628
 629         /*
 630          *      Now check at the protocol level
 631          */
 632         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 633                 goto out;
 634
 635         /*
 636          *      Only reply to fragment 0. We byte re-order the constant
 637          *      mask for efficiency.
 638          */
 639         if (iph->frag_off & htons(IP_OFFSET))
 640                 goto out;
 641
 642         /*
 643          *      If we send an ICMP error to an ICMP error a mess would result..
 644          */
 645         if (icmp_pointers[type].error) {
 646                 /*
 647                  *      We are an error, check if we are replying to an
 648                  *      ICMP error
 649                  */
 650                 if (iph->protocol == IPPROTO_ICMP) {
 651                         u8 _inner_type, *itp;
 652
 653                         itp = skb_header_pointer(skb_in,
 654                                                  skb_network_header(skb_in) +
 655                                                  (iph->ihl << 2) +
 656                                                  offsetof(struct icmphdr,
 657                                                           type) -
 658                                                  skb_in->data,
 659                                                  sizeof(_inner_type),
 660                                                  &_inner_type);
 661                         if (!itp)
 662                                 goto out;
 663
 664                         /*
 665                          *      Assume any unknown ICMP type is an error. This
 666                          *      isn't specified by the RFC, but think about it..
 667                          */
 668                         if (*itp > NR_ICMP_TYPES ||
 669                             icmp_pointers[*itp].error)
 670                                 goto out;
 671                 }
 672         }
 673
 674         icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
 675         if (!icmp_param)
 676                 return;
 677
 678         sk = icmp_xmit_lock(net);
 679         if (!sk)
 680                 goto out_free;
 681
 682         /*
 683          *      Construct source address and options.
 684          */
 685
 686         saddr = iph->daddr;
 687         if (!(rt->rt_flags & RTCF_LOCAL)) {
 688                 struct net_device *dev = NULL;
 689
 690                 rcu_read_lock();
 691                 if (rt_is_input_route(rt) &&
 692                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 693                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 694
 695                 if (dev)
 696                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 697                 else
 698                         saddr = 0;
 699                 rcu_read_unlock();
 700         }
 701
 702         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 703                                            IPTOS_PREC_INTERNETCONTROL) :
 704                                           iph->tos;
 705         mark = IP4_REPLY_MARK(net, skb_in->mark);
 706
 707         if (__ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in, opt))
 708                 goto out_unlock;
 709
 710
 711         /*
 712          *      Prepare data for ICMP header.
 713          */
 714
 715         icmp_param->data.icmph.type      = type;
 716         icmp_param->data.icmph.code      = code;
 717         icmp_param->data.icmph.un.gateway = info;
 718         icmp_param->data.icmph.checksum  = 0;
 719         icmp_param->skb   = skb_in;
 720         icmp_param->offset = skb_network_offset(skb_in);
 721         inet_sk(sk)->tos = tos;
 722         sk->sk_mark = mark;
 723         ipc.addr = iph->saddr;
 724         ipc.opt = &icmp_param->replyopts.opt;
 725         ipc.tx_flags = 0;
 726         ipc.ttl = 0;
 727         ipc.tos = -1;
 728
 729         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 730                                type, code, icmp_param);
 731         if (IS_ERR(rt))
 732                 goto out_unlock;
 733
 734         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 735                 goto ende;
 736
 737         /* RFC says return as much as we can without exceeding 576 bytes. */
 738
 739         room = dst_mtu(&rt->dst);
 740         if (room > 576)
 741                 room = 576;
 742         room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
 743         room -= sizeof(struct icmphdr);
 744
 745         icmp_param->data_len = skb_in->len - icmp_param->offset;
 746         if (icmp_param->data_len > room)
 747                 icmp_param->data_len = room;
 748         icmp_param->head_len = sizeof(struct icmphdr);
 749
 750         icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 751 ende:
 752         ip_rt_put(rt);
 753 out_unlock:
 754         icmp_xmit_unlock(sk);
 755 out_free:
 756         kfree(icmp_param);
 757 out:;
 758 }
 759 EXPORT_SYMBOL(__icmp_send);
 760
 761 #if IS_ENABLED(CONFIG_NF_NAT)
 762 #include <net/netfilter/nf_conntrack.h>
 763 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 764 {
 765         struct sk_buff *cloned_skb = NULL;
 766         struct ip_options opts = { 0 };
 767         enum ip_conntrack_info ctinfo;
 768         struct nf_conn *ct;
 769         __be32 orig_ip;
 770
 771         ct = nf_ct_get(skb_in, &ctinfo);
 772         if (!ct || !(ct->status & IPS_SRC_NAT)) {
 773                 __icmp_send(skb_in, type, code, info, &opts);
 774                 return;
 775         }
 776
 777         if (skb_shared(skb_in))
 778                 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 779
 780         if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 781             (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 782             skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 783             skb_network_offset(skb_in) + sizeof(struct iphdr))))
 784                 goto out;
 785
 786         orig_ip = ip_hdr(skb_in)->saddr;
 787         ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 788         __icmp_send(skb_in, type, code, info, &opts);
 789         ip_hdr(skb_in)->saddr = orig_ip;
 790 out:
 791         consume_skb(cloned_skb);
 792 }
 793 EXPORT_SYMBOL(icmp_ndo_send);
 794 #endif
 795
 796 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 797 {
 798         const struct iphdr *iph = (const struct iphdr *) skb->data;
 799         const struct net_protocol *ipprot;
 800         int protocol = iph->protocol;
 801
 802         /* Checkin full IP header plus 8 bytes of protocol to
 803          * avoid additional coding at protocol handlers.
 804          */
 805         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 806                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 807                 return;
 808         }
 809
 810         raw_icmp_error(skb, protocol, info);
 811
 812         ipprot = rcu_dereference(inet_protos[protocol]);
 813         if (ipprot && ipprot->err_handler)
 814                 ipprot->err_handler(skb, info);
 815 }
 816
 817 static bool icmp_tag_validation(int proto)
 818 {
 819         bool ok;
 820
 821         rcu_read_lock();
 822         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 823         rcu_read_unlock();
 824         return ok;
 825 }
 826
 827 /*
 828  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 829  *      ICMP_PARAMETERPROB.
 830  */
 831
 832 static bool icmp_unreach(struct sk_buff *skb)
 833 {
 834         const struct iphdr *iph;
 835         struct icmphdr *icmph;
 836         struct net *net;
 837         u32 info = 0;
 838
 839         net = dev_net(skb_dst(skb)->dev);
 840
 841         /*
 842          *      Incomplete header ?
 843          *      Only checks for the IP header, there should be an
 844          *      additional check for longer headers in upper levels.
 845          */
 846
 847         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 848                 goto out_err;
 849
 850         icmph = icmp_hdr(skb);
 851         iph   = (const struct iphdr *)skb->data;
 852
 853         if (iph->ihl < 5) /* Mangled header, drop. */
 854                 goto out_err;
 855
 856         switch (icmph->type) {
 857         case ICMP_DEST_UNREACH:
 858                 switch (icmph->code & 15) {
 859                 case ICMP_NET_UNREACH:
 860                 case ICMP_HOST_UNREACH:
 861                 case ICMP_PROT_UNREACH:
 862                 case ICMP_PORT_UNREACH:
 863                         break;
 864                 case ICMP_FRAG_NEEDED:
 865                         /* for documentation of the ip_no_pmtu_disc
 866                          * values please see
 867                          * Documentation/networking/ip-sysctl.txt
 868                          */
 869                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 870                         default:
 871                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 872                                                     &iph->daddr);
 873                                 break;
 874                         case 2:
 875                                 goto out;
 876                         case 3:
 877                                 if (!icmp_tag_validation(iph->protocol))
 878                                         goto out;
 879                                 /* fall through */
 880                         case 0:
 881                                 info = ntohs(icmph->un.frag.mtu);
 882                         }
 883                         break;
 884                 case ICMP_SR_FAILED:
 885                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 886                                             &iph->daddr);
 887                         break;
 888                 default:
 889                         break;
 890                 }
 891                 if (icmph->code > NR_ICMP_UNREACH)
 892                         goto out;
 893                 break;
 894         case ICMP_PARAMETERPROB:
 895                 info = ntohl(icmph->un.gateway) >> 24;
 896                 break;
 897         case ICMP_TIME_EXCEEDED:
 898                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 899                 if (icmph->code == ICMP_EXC_FRAGTIME)
 900                         goto out;
 901                 break;
 902         }
 903
 904         /*
 905          *      Throw it at our lower layers
 906          *
 907          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 908          *                header.
 909          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 910          *                transport layer.
 911          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 912          *                transport layer.
 913          */
 914
 915         /*
 916          *      Check the other end isn't violating RFC 1122. Some routers send
 917          *      bogus responses to broadcast frames. If you see this message
 918          *      first check your netmask matches at both ends, if it does then
 919          *      get the other vendor to fix their kit.
 920          */
 921
 922         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 923             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 924                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 925                                      &ip_hdr(skb)->saddr,
 926                                      icmph->type, icmph->code,
 927                                      &iph->daddr, skb->dev->name);
 928                 goto out;
 929         }
 930
 931         icmp_socket_deliver(skb, info);
 932
 933 out:
 934         return true;
 935 out_err:
 936         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 937         return false;
 938 }
 939
 940
 941 /*
 942  *      Handle ICMP_REDIRECT.
 943  */
 944
 945 static bool icmp_redirect(struct sk_buff *skb)
 946 {
 947         if (skb->len < sizeof(struct iphdr)) {
 948                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 949                 return false;
 950         }
 951
 952         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 953                 /* there aught to be a stat */
 954                 return false;
 955         }
 956
 957         icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
 958         return true;
 959 }
 960
 961 /*
 962  *      Handle ICMP_ECHO ("ping") requests.
 963  *
 964  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 965  *                requests.
 966  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 967  *                included in the reply.
 968  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 969  *                echo requests, MUST have default=NOT.
 970  *      See also WRT handling of options once they are done and working.
 971  */
 972
 973 static bool icmp_echo(struct sk_buff *skb)
 974 {
 975         struct net *net;
 976
 977         net = dev_net(skb_dst(skb)->dev);
 978         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 979                 struct icmp_bxm icmp_param;
 980
 981                 icmp_param.data.icmph      = *icmp_hdr(skb);
 982                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 983                 icmp_param.skb             = skb;
 984                 icmp_param.offset          = 0;
 985                 icmp_param.data_len        = skb->len;
 986                 icmp_param.head_len        = sizeof(struct icmphdr);
 987                 icmp_reply(&icmp_param, skb);
 988         }
 989         /* should there be an ICMP stat for ignored echos? */
 990         return true;
 991 }
 992
 993 /*
 994  *      Handle ICMP Timestamp requests.
 995  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 996  *                SHOULD be in the kernel for minimum random latency.
 997  *                MUST be accurate to a few minutes.
 998  *                MUST be updated at least at 15Hz.
 999  */
1000 static bool icmp_timestamp(struct sk_buff *skb)
1001 {
1002         struct icmp_bxm icmp_param;
1003         /*
1004          *      Too short.
1005          */
1006         if (skb->len < 4)
1007                 goto out_err;
1008
1009         /*
1010          *      Fill in the current time as ms since midnight UT:
1011          */
1012         icmp_param.data.times[1] = inet_current_timestamp();
1013         icmp_param.data.times[2] = icmp_param.data.times[1];
1014         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
1015                 BUG();
1016         icmp_param.data.icmph      = *icmp_hdr(skb);
1017         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1018         icmp_param.data.icmph.code = 0;
1019         icmp_param.skb             = skb;
1020         icmp_param.offset          = 0;
1021         icmp_param.data_len        = 0;
1022         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
1023         icmp_reply(&icmp_param, skb);
1024         return true;
1025
1026 out_err:
1027         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1028         return false;
1029 }
1030
1031 static bool icmp_discard(struct sk_buff *skb)
1032 {
1033         /* pretend it was a success */
1034         return true;
1035 }
1036
1037 /*
1038  *      Deal with incoming ICMP packets.
1039  */
1040 int icmp_rcv(struct sk_buff *skb)
1041 {
1042         struct icmphdr *icmph;
1043         struct rtable *rt = skb_rtable(skb);
1044         struct net *net = dev_net(rt->dst.dev);
1045         bool success;
1046
1047         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1048                 struct sec_path *sp = skb_sec_path(skb);
1049                 int nh;
1050
1051                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1052                                  XFRM_STATE_ICMP))
1053                         goto drop;
1054
1055                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1056                         goto drop;
1057
1058                 nh = skb_network_offset(skb);
1059                 skb_set_network_header(skb, sizeof(*icmph));
1060
1061                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1062                         goto drop;
1063
1064                 skb_set_network_header(skb, nh);
1065         }
1066
1067         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1068
1069         if (skb_checksum_simple_validate(skb))
1070                 goto csum_error;
1071
1072         if (!pskb_pull(skb, sizeof(*icmph)))
1073                 goto error;
1074
1075         icmph = icmp_hdr(skb);
1076
1077         ICMPMSGIN_INC_STATS(net, icmph->type);
1078         /*
1079          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1080          *
1081          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1082          *                discarded.
1083          */
1084         if (icmph->type > NR_ICMP_TYPES)
1085                 goto error;
1086
1087
1088         /*
1089          *      Parse the ICMP message
1090          */
1091
1092         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1093                 /*
1094                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1095                  *        silently ignored (we let user decide with a sysctl).
1096                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1097                  *        discarded if to broadcast/multicast.
1098                  */
1099                 if ((icmph->type == ICMP_ECHO ||
1100                      icmph->type == ICMP_TIMESTAMP) &&
1101                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1102                         goto error;
1103                 }
1104                 if (icmph->type != ICMP_ECHO &&
1105                     icmph->type != ICMP_TIMESTAMP &&
1106                     icmph->type != ICMP_ADDRESS &&
1107                     icmph->type != ICMP_ADDRESSREPLY) {
1108                         goto error;
1109                 }
1110         }
1111
1112         success = icmp_pointers[icmph->type].handler(skb);
1113
1114         if (success)  {
1115                 consume_skb(skb);
1116                 return 0;
1117         }
1118
1119 drop:
1120         kfree_skb(skb);
1121         return 0;
1122 csum_error:
1123         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1124 error:
1125         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1126         goto drop;
1127 }
1128
1129 void icmp_err(struct sk_buff *skb, u32 info)
1130 {
1131         struct iphdr *iph = (struct iphdr *)skb->data;
1132         int offset = iph->ihl<<2;
1133         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1134         int type = icmp_hdr(skb)->type;
1135         int code = icmp_hdr(skb)->code;
1136         struct net *net = dev_net(skb->dev);
1137
1138         /*
1139          * Use ping_err to handle all icmp errors except those
1140          * triggered by ICMP_ECHOREPLY which sent from kernel.
1141          */
1142         if (icmph->type != ICMP_ECHOREPLY) {
1143                 ping_err(skb, offset, info);
1144                 return;
1145         }
1146
1147         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1148                 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1149         else if (type == ICMP_REDIRECT)
1150                 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1151 }
1152
1153 /*
1154  *      This table is the definition of how we handle ICMP.
1155  */
1156 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1157         [ICMP_ECHOREPLY] = {
1158                 .handler = ping_rcv,
1159         },
1160         [1] = {
1161                 .handler = icmp_discard,
1162                 .error = 1,
1163         },
1164         [2] = {
1165                 .handler = icmp_discard,
1166                 .error = 1,
1167         },
1168         [ICMP_DEST_UNREACH] = {
1169                 .handler = icmp_unreach,
1170                 .error = 1,
1171         },
1172         [ICMP_SOURCE_QUENCH] = {
1173                 .handler = icmp_unreach,
1174                 .error = 1,
1175         },
1176         [ICMP_REDIRECT] = {
1177                 .handler = icmp_redirect,
1178                 .error = 1,
1179         },
1180         [6] = {
1181                 .handler = icmp_discard,
1182                 .error = 1,
1183         },
1184         [7] = {
1185                 .handler = icmp_discard,
1186                 .error = 1,
1187         },
1188         [ICMP_ECHO] = {
1189                 .handler = icmp_echo,
1190         },
1191         [9] = {
1192                 .handler = icmp_discard,
1193                 .error = 1,
1194         },
1195         [10] = {
1196                 .handler = icmp_discard,
1197                 .error = 1,
1198         },
1199         [ICMP_TIME_EXCEEDED] = {
1200                 .handler = icmp_unreach,
1201                 .error = 1,
1202         },
1203         [ICMP_PARAMETERPROB] = {
1204                 .handler = icmp_unreach,
1205                 .error = 1,
1206         },
1207         [ICMP_TIMESTAMP] = {
1208                 .handler = icmp_timestamp,
1209         },
1210         [ICMP_TIMESTAMPREPLY] = {
1211                 .handler = icmp_discard,
1212         },
1213         [ICMP_INFO_REQUEST] = {
1214                 .handler = icmp_discard,
1215         },
1216         [ICMP_INFO_REPLY] = {
1217                 .handler = icmp_discard,
1218         },
1219         [ICMP_ADDRESS] = {
1220                 .handler = icmp_discard,
1221         },
1222         [ICMP_ADDRESSREPLY] = {
1223                 .handler = icmp_discard,
1224         },
1225 };
1226
1227 static void __net_exit icmp_sk_exit(struct net *net)
1228 {
1229         int i;
1230
1231         for_each_possible_cpu(i)
1232                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1233         free_percpu(net->ipv4.icmp_sk);
1234         net->ipv4.icmp_sk = NULL;
1235 }
1236
1237 static int __net_init icmp_sk_init(struct net *net)
1238 {
1239         int i, err;
1240
1241         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1242         if (!net->ipv4.icmp_sk)
1243                 return -ENOMEM;
1244
1245         for_each_possible_cpu(i) {
1246                 struct sock *sk;
1247
1248                 err = inet_ctl_sock_create(&sk, PF_INET,
1249                                            SOCK_RAW, IPPROTO_ICMP, net);
1250                 if (err < 0)
1251                         goto fail;
1252
1253                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1254
1255                 /* Enough space for 2 64K ICMP packets, including
1256                  * sk_buff/skb_shared_info struct overhead.
1257                  */
1258                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1259
1260                 /*
1261                  * Speedup sock_wfree()
1262                  */
1263                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1264                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1265         }
1266
1267         /* Control parameters for ECHO replies. */
1268         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1269         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1270
1271         /* Control parameter - ignore bogus broadcast responses? */
1272         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1273
1274         /*
1275          *      Configurable global rate limit.
1276          *
1277          *      ratelimit defines tokens/packet consumed for dst->rate_token
1278          *      bucket ratemask defines which icmp types are ratelimited by
1279          *      setting it's bit position.
1280          *
1281          *      default:
1282          *      dest unreachable (3), source quench (4),
1283          *      time exceeded (11), parameter problem (12)
1284          */
1285
1286         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1287         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1288         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1289
1290         return 0;
1291
1292 fail:
1293         for_each_possible_cpu(i)
1294                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1295         free_percpu(net->ipv4.icmp_sk);
1296         return err;
1297 }
1298
1299 static struct pernet_operations __net_initdata icmp_sk_ops = {
1300        .init = icmp_sk_init,
1301        .exit = icmp_sk_exit,
1302 };
1303
1304 int __init icmp_init(void)
1305 {
1306         return register_pernet_subsys(&icmp_sk_ops);
1307 }