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 <linux/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 /* Called with BH disabled */
 213 static inline struct sock *icmp_xmit_lock(struct net *net)
 214 {
 215         struct sock *sk;
 216
 217         sk = icmp_sk(net);
 218
 219         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 220                 /* This can happen if the output path signals a
 221                  * dst_link_failure() for an outgoing ICMP packet.
 222                  */
 223                 return NULL;
 224         }
 225         return sk;
 226 }
 227
 228 static inline void icmp_xmit_unlock(struct sock *sk)
 229 {
 230         spin_unlock(&sk->sk_lock.slock);
 231 }
 232
 233 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 234 int sysctl_icmp_msgs_burst __read_mostly = 50;
 235
 236 static struct {
 237         spinlock_t      lock;
 238         u32             credit;
 239         u32             stamp;
 240 } icmp_global = {
 241         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 242 };
 243
 244 /**
 245  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 246  *
 247  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
 248  * Returns false if we reached the limit and can not send another packet.
 249  * Note: called with BH disabled
 250  */
 251 bool icmp_global_allow(void)
 252 {
 253         u32 credit, delta, incr = 0, now = (u32)jiffies;
 254         bool rc = false;
 255
 256         /* Check if token bucket is empty and cannot be refilled
 257          * without taking the spinlock.
 258          */
 259         if (!icmp_global.credit) {
 260                 delta = min_t(u32, now - icmp_global.stamp, HZ);
 261                 if (delta < HZ / 50)
 262                         return false;
 263         }
 264
 265         spin_lock(&icmp_global.lock);
 266         delta = min_t(u32, now - icmp_global.stamp, HZ);
 267         if (delta >= HZ / 50) {
 268                 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
 269                 if (incr)
 270                         icmp_global.stamp = now;
 271         }
 272         credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
 273         if (credit) {
 274                 credit--;
 275                 rc = true;
 276         }
 277         icmp_global.credit = credit;
 278         spin_unlock(&icmp_global.lock);
 279         return rc;
 280 }
 281 EXPORT_SYMBOL(icmp_global_allow);
 282
 283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 284 {
 285         if (type > NR_ICMP_TYPES)
 286                 return true;
 287
 288         /* Don't limit PMTU discovery. */
 289         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 290                 return true;
 291
 292         /* Limit if icmp type is enabled in ratemask. */
 293         if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 294                 return true;
 295
 296         return false;
 297 }
 298
 299 static bool icmpv4_global_allow(struct net *net, int type, int code)
 300 {
 301         if (icmpv4_mask_allow(net, type, code))
 302                 return true;
 303
 304         if (icmp_global_allow())
 305                 return true;
 306
 307         return false;
 308 }
 309
 310 /*
 311  *      Send an ICMP frame.
 312  */
 313
 314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 315                                struct flowi4 *fl4, int type, int code)
 316 {
 317         struct dst_entry *dst = &rt->dst;
 318         struct inet_peer *peer;
 319         bool rc = true;
 320         int vif;
 321
 322         if (icmpv4_mask_allow(net, type, code))
 323                 goto out;
 324
 325         /* No rate limit on loopback */
 326         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 327                 goto out;
 328
 329         vif = l3mdev_master_ifindex(dst->dev);
 330         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 331         rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
 332         if (peer)
 333                 inet_putpeer(peer);
 334 out:
 335         return rc;
 336 }
 337
 338 /*
 339  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 340  */
 341 void icmp_out_count(struct net *net, unsigned char type)
 342 {
 343         ICMPMSGOUT_INC_STATS(net, type);
 344         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 345 }
 346
 347 /*
 348  *      Checksum each fragment, and on the first include the headers and final
 349  *      checksum.
 350  */
 351 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 352                           struct sk_buff *skb)
 353 {
 354         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 355         __wsum csum;
 356
 357         csum = skb_copy_and_csum_bits(icmp_param->skb,
 358                                       icmp_param->offset + offset,
 359                                       to, len, 0);
 360
 361         skb->csum = csum_block_add(skb->csum, csum, odd);
 362         if (icmp_pointers[icmp_param->data.icmph.type].error)
 363                 nf_ct_attach(skb, icmp_param->skb);
 364         return 0;
 365 }
 366
 367 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 368                             struct flowi4 *fl4,
 369                             struct ipcm_cookie *ipc, struct rtable **rt)
 370 {
 371         struct sock *sk;
 372         struct sk_buff *skb;
 373
 374         sk = icmp_sk(dev_net((*rt)->dst.dev));
 375         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 376                            icmp_param->data_len+icmp_param->head_len,
 377                            icmp_param->head_len,
 378                            ipc, rt, MSG_DONTWAIT) < 0) {
 379                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 380                 ip_flush_pending_frames(sk);
 381         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 382                 struct icmphdr *icmph = icmp_hdr(skb);
 383                 __wsum csum = 0;
 384                 struct sk_buff *skb1;
 385
 386                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 387                         csum = csum_add(csum, skb1->csum);
 388                 }
 389                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 390                                                  (char *)icmph,
 391                                                  icmp_param->head_len, csum);
 392                 icmph->checksum = csum_fold(csum);
 393                 skb->ip_summed = CHECKSUM_NONE;
 394                 ip_push_pending_frames(sk, fl4);
 395         }
 396 }
 397
 398 /*
 399  *      Driving logic for building and sending ICMP messages.
 400  */
 401
 402 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 403 {
 404         struct ipcm_cookie ipc;
 405         struct rtable *rt = skb_rtable(skb);
 406         struct net *net = dev_net(rt->dst.dev);
 407         struct flowi4 fl4;
 408         struct sock *sk;
 409         struct inet_sock *inet;
 410         __be32 daddr, saddr;
 411         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 412         int type = icmp_param->data.icmph.type;
 413         int code = icmp_param->data.icmph.code;
 414
 415         if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 416                 return;
 417
 418         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 419         local_bh_disable();
 420
 421         /* global icmp_msgs_per_sec */
 422         if (!icmpv4_global_allow(net, type, code))
 423                 goto out_bh_enable;
 424
 425         sk = icmp_xmit_lock(net);
 426         if (!sk)
 427                 goto out_bh_enable;
 428         inet = inet_sk(sk);
 429
 430         icmp_param->data.icmph.checksum = 0;
 431
 432         inet->tos = ip_hdr(skb)->tos;
 433         sk->sk_mark = mark;
 434         daddr = ipc.addr = ip_hdr(skb)->saddr;
 435         saddr = fib_compute_spec_dst(skb);
 436         ipc.opt = NULL;
 437         ipc.tx_flags = 0;
 438         ipc.ttl = 0;
 439         ipc.tos = -1;
 440
 441         if (icmp_param->replyopts.opt.opt.optlen) {
 442                 ipc.opt = &icmp_param->replyopts.opt;
 443                 if (ipc.opt->opt.srr)
 444                         daddr = icmp_param->replyopts.opt.opt.faddr;
 445         }
 446         memset(&fl4, 0, sizeof(fl4));
 447         fl4.daddr = daddr;
 448         fl4.saddr = saddr;
 449         fl4.flowi4_mark = mark;
 450         fl4.flowi4_uid = sock_net_uid(net, NULL);
 451         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 452         fl4.flowi4_proto = IPPROTO_ICMP;
 453         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 454         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 455         rt = ip_route_output_key(net, &fl4);
 456         if (IS_ERR(rt))
 457                 goto out_unlock;
 458         if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 459                 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 460         ip_rt_put(rt);
 461 out_unlock:
 462         icmp_xmit_unlock(sk);
 463 out_bh_enable:
 464         local_bh_enable();
 465 }
 466
 467 static struct rtable *icmp_route_lookup(struct net *net,
 468                                         struct flowi4 *fl4,
 469                                         struct sk_buff *skb_in,
 470                                         const struct iphdr *iph,
 471                                         __be32 saddr, u8 tos, u32 mark,
 472                                         int type, int code,
 473                                         struct icmp_bxm *param)
 474 {
 475         struct rtable *rt, *rt2;
 476         struct flowi4 fl4_dec;
 477         int err;
 478
 479         memset(fl4, 0, sizeof(*fl4));
 480         fl4->daddr = (param->replyopts.opt.opt.srr ?
 481                       param->replyopts.opt.opt.faddr : iph->saddr);
 482         fl4->saddr = saddr;
 483         fl4->flowi4_mark = mark;
 484         fl4->flowi4_uid = sock_net_uid(net, NULL);
 485         fl4->flowi4_tos = RT_TOS(tos);
 486         fl4->flowi4_proto = IPPROTO_ICMP;
 487         fl4->fl4_icmp_type = type;
 488         fl4->fl4_icmp_code = code;
 489         fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
 490
 491         security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 492         rt = ip_route_output_key_hash(net, fl4, skb_in);
 493         if (IS_ERR(rt))
 494                 return rt;
 495
 496         /* No need to clone since we're just using its address. */
 497         rt2 = rt;
 498
 499         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 500                                            flowi4_to_flowi(fl4), NULL, 0);
 501         if (!IS_ERR(rt)) {
 502                 if (rt != rt2)
 503                         return rt;
 504         } else if (PTR_ERR(rt) == -EPERM) {
 505                 rt = NULL;
 506         } else
 507                 return rt;
 508
 509         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 510         if (err)
 511                 goto relookup_failed;
 512
 513         if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
 514                                      fl4_dec.saddr) == RTN_LOCAL) {
 515                 rt2 = __ip_route_output_key(net, &fl4_dec);
 516                 if (IS_ERR(rt2))
 517                         err = PTR_ERR(rt2);
 518         } else {
 519                 struct flowi4 fl4_2 = {};
 520                 unsigned long orefdst;
 521
 522                 fl4_2.daddr = fl4_dec.saddr;
 523                 rt2 = ip_route_output_key(net, &fl4_2);
 524                 if (IS_ERR(rt2)) {
 525                         err = PTR_ERR(rt2);
 526                         goto relookup_failed;
 527                 }
 528                 /* Ugh! */
 529                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 530                 skb_dst_set(skb_in, NULL);
 531                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 532                                      RT_TOS(tos), rt2->dst.dev);
 533
 534                 dst_release(&rt2->dst);
 535                 rt2 = skb_rtable(skb_in);
 536                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 537         }
 538
 539         if (err)
 540                 goto relookup_failed;
 541
 542         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 543                                             flowi4_to_flowi(&fl4_dec), NULL,
 544                                             XFRM_LOOKUP_ICMP);
 545         if (!IS_ERR(rt2)) {
 546                 dst_release(&rt->dst);
 547                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 548                 rt = rt2;
 549         } else if (PTR_ERR(rt2) == -EPERM) {
 550                 if (rt)
 551                         dst_release(&rt->dst);
 552                 return rt2;
 553         } else {
 554                 err = PTR_ERR(rt2);
 555                 goto relookup_failed;
 556         }
 557         return rt;
 558
 559 relookup_failed:
 560         if (rt)
 561                 return rt;
 562         return ERR_PTR(err);
 563 }
 564
 565 /*
 566  *      Send an ICMP message in response to a situation
 567  *
 568  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 569  *                MAY send more (we do).
 570  *                      MUST NOT change this header information.
 571  *                      MUST NOT reply to a multicast/broadcast IP address.
 572  *                      MUST NOT reply to a multicast/broadcast MAC address.
 573  *                      MUST reply to only the first fragment.
 574  */
 575
 576 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 577 {
 578         struct iphdr *iph;
 579         int room;
 580         struct icmp_bxm icmp_param;
 581         struct rtable *rt = skb_rtable(skb_in);
 582         struct ipcm_cookie ipc;
 583         struct flowi4 fl4;
 584         __be32 saddr;
 585         u8  tos;
 586         u32 mark;
 587         struct net *net;
 588         struct sock *sk;
 589
 590         if (!rt)
 591                 goto out;
 592         net = dev_net(rt->dst.dev);
 593
 594         /*
 595          *      Find the original header. It is expected to be valid, of course.
 596          *      Check this, icmp_send is called from the most obscure devices
 597          *      sometimes.
 598          */
 599         iph = ip_hdr(skb_in);
 600
 601         if ((u8 *)iph < skb_in->head ||
 602             (skb_network_header(skb_in) + sizeof(*iph)) >
 603             skb_tail_pointer(skb_in))
 604                 goto out;
 605
 606         /*
 607          *      No replies to physical multicast/broadcast
 608          */
 609         if (skb_in->pkt_type != PACKET_HOST)
 610                 goto out;
 611
 612         /*
 613          *      Now check at the protocol level
 614          */
 615         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 616                 goto out;
 617
 618         /*
 619          *      Only reply to fragment 0. We byte re-order the constant
 620          *      mask for efficiency.
 621          */
 622         if (iph->frag_off & htons(IP_OFFSET))
 623                 goto out;
 624
 625         /*
 626          *      If we send an ICMP error to an ICMP error a mess would result..
 627          */
 628         if (icmp_pointers[type].error) {
 629                 /*
 630                  *      We are an error, check if we are replying to an
 631                  *      ICMP error
 632                  */
 633                 if (iph->protocol == IPPROTO_ICMP) {
 634                         u8 _inner_type, *itp;
 635
 636                         itp = skb_header_pointer(skb_in,
 637                                                  skb_network_header(skb_in) +
 638                                                  (iph->ihl << 2) +
 639                                                  offsetof(struct icmphdr,
 640                                                           type) -
 641                                                  skb_in->data,
 642                                                  sizeof(_inner_type),
 643                                                  &_inner_type);
 644                         if (!itp)
 645                                 goto out;
 646
 647                         /*
 648                          *      Assume any unknown ICMP type is an error. This
 649                          *      isn't specified by the RFC, but think about it..
 650                          */
 651                         if (*itp > NR_ICMP_TYPES ||
 652                             icmp_pointers[*itp].error)
 653                                 goto out;
 654                 }
 655         }
 656
 657         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 658         local_bh_disable();
 659
 660         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 661          * incoming dev is loopback.  If outgoing dev change to not be
 662          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 663          */
 664         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 665               !icmpv4_global_allow(net, type, code))
 666                 goto out_bh_enable;
 667
 668         sk = icmp_xmit_lock(net);
 669         if (!sk)
 670                 goto out_bh_enable;
 671
 672         /*
 673          *      Construct source address and options.
 674          */
 675
 676         saddr = iph->daddr;
 677         if (!(rt->rt_flags & RTCF_LOCAL)) {
 678                 struct net_device *dev = NULL;
 679
 680                 rcu_read_lock();
 681                 if (rt_is_input_route(rt) &&
 682                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 683                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 684
 685                 if (dev)
 686                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 687                 else
 688                         saddr = 0;
 689                 rcu_read_unlock();
 690         }
 691
 692         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 693                                            IPTOS_PREC_INTERNETCONTROL) :
 694                                           iph->tos;
 695         mark = IP4_REPLY_MARK(net, skb_in->mark);
 696
 697         if (ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in))
 698                 goto out_unlock;
 699
 700
 701         /*
 702          *      Prepare data for ICMP header.
 703          */
 704
 705         icmp_param.data.icmph.type       = type;
 706         icmp_param.data.icmph.code       = code;
 707         icmp_param.data.icmph.un.gateway = info;
 708         icmp_param.data.icmph.checksum   = 0;
 709         icmp_param.skb    = skb_in;
 710         icmp_param.offset = skb_network_offset(skb_in);
 711         inet_sk(sk)->tos = tos;
 712         sk->sk_mark = mark;
 713         ipc.addr = iph->saddr;
 714         ipc.opt = &icmp_param.replyopts.opt;
 715         ipc.tx_flags = 0;
 716         ipc.ttl = 0;
 717         ipc.tos = -1;
 718
 719         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 720                                type, code, &icmp_param);
 721         if (IS_ERR(rt))
 722                 goto out_unlock;
 723
 724         /* peer icmp_ratelimit */
 725         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 726                 goto ende;
 727
 728         /* RFC says return as much as we can without exceeding 576 bytes. */
 729
 730         room = dst_mtu(&rt->dst);
 731         if (room > 576)
 732                 room = 576;
 733         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 734         room -= sizeof(struct icmphdr);
 735
 736         icmp_param.data_len = skb_in->len - icmp_param.offset;
 737         if (icmp_param.data_len > room)
 738                 icmp_param.data_len = room;
 739         icmp_param.head_len = sizeof(struct icmphdr);
 740
 741         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 742 ende:
 743         ip_rt_put(rt);
 744 out_unlock:
 745         icmp_xmit_unlock(sk);
 746 out_bh_enable:
 747         local_bh_enable();
 748 out:;
 749 }
 750 EXPORT_SYMBOL(icmp_send);
 751
 752
 753 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 754 {
 755         const struct iphdr *iph = (const struct iphdr *) skb->data;
 756         const struct net_protocol *ipprot;
 757         int protocol = iph->protocol;
 758
 759         /* Checkin full IP header plus 8 bytes of protocol to
 760          * avoid additional coding at protocol handlers.
 761          */
 762         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 763                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 764                 return;
 765         }
 766
 767         raw_icmp_error(skb, protocol, info);
 768
 769         ipprot = rcu_dereference(inet_protos[protocol]);
 770         if (ipprot && ipprot->err_handler)
 771                 ipprot->err_handler(skb, info);
 772 }
 773
 774 static bool icmp_tag_validation(int proto)
 775 {
 776         bool ok;
 777
 778         rcu_read_lock();
 779         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 780         rcu_read_unlock();
 781         return ok;
 782 }
 783
 784 /*
 785  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 786  *      ICMP_PARAMETERPROB.
 787  */
 788
 789 static bool icmp_unreach(struct sk_buff *skb)
 790 {
 791         const struct iphdr *iph;
 792         struct icmphdr *icmph;
 793         struct net *net;
 794         u32 info = 0;
 795
 796         net = dev_net(skb_dst(skb)->dev);
 797
 798         /*
 799          *      Incomplete header ?
 800          *      Only checks for the IP header, there should be an
 801          *      additional check for longer headers in upper levels.
 802          */
 803
 804         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 805                 goto out_err;
 806
 807         icmph = icmp_hdr(skb);
 808         iph   = (const struct iphdr *)skb->data;
 809
 810         if (iph->ihl < 5) /* Mangled header, drop. */
 811                 goto out_err;
 812
 813         switch (icmph->type) {
 814         case ICMP_DEST_UNREACH:
 815                 switch (icmph->code & 15) {
 816                 case ICMP_NET_UNREACH:
 817                 case ICMP_HOST_UNREACH:
 818                 case ICMP_PROT_UNREACH:
 819                 case ICMP_PORT_UNREACH:
 820                         break;
 821                 case ICMP_FRAG_NEEDED:
 822                         /* for documentation of the ip_no_pmtu_disc
 823                          * values please see
 824                          * Documentation/networking/ip-sysctl.txt
 825                          */
 826                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 827                         default:
 828                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 829                                                     &iph->daddr);
 830                                 break;
 831                         case 2:
 832                                 goto out;
 833                         case 3:
 834                                 if (!icmp_tag_validation(iph->protocol))
 835                                         goto out;
 836                                 /* fall through */
 837                         case 0:
 838                                 info = ntohs(icmph->un.frag.mtu);
 839                         }
 840                         break;
 841                 case ICMP_SR_FAILED:
 842                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 843                                             &iph->daddr);
 844                         break;
 845                 default:
 846                         break;
 847                 }
 848                 if (icmph->code > NR_ICMP_UNREACH)
 849                         goto out;
 850                 break;
 851         case ICMP_PARAMETERPROB:
 852                 info = ntohl(icmph->un.gateway) >> 24;
 853                 break;
 854         case ICMP_TIME_EXCEEDED:
 855                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 856                 if (icmph->code == ICMP_EXC_FRAGTIME)
 857                         goto out;
 858                 break;
 859         }
 860
 861         /*
 862          *      Throw it at our lower layers
 863          *
 864          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 865          *                header.
 866          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 867          *                transport layer.
 868          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 869          *                transport layer.
 870          */
 871
 872         /*
 873          *      Check the other end isn't violating RFC 1122. Some routers send
 874          *      bogus responses to broadcast frames. If you see this message
 875          *      first check your netmask matches at both ends, if it does then
 876          *      get the other vendor to fix their kit.
 877          */
 878
 879         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 880             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 881                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 882                                      &ip_hdr(skb)->saddr,
 883                                      icmph->type, icmph->code,
 884                                      &iph->daddr, skb->dev->name);
 885                 goto out;
 886         }
 887
 888         icmp_socket_deliver(skb, info);
 889
 890 out:
 891         return true;
 892 out_err:
 893         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 894         return false;
 895 }
 896
 897
 898 /*
 899  *      Handle ICMP_REDIRECT.
 900  */
 901
 902 static bool icmp_redirect(struct sk_buff *skb)
 903 {
 904         if (skb->len < sizeof(struct iphdr)) {
 905                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 906                 return false;
 907         }
 908
 909         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 910                 /* there aught to be a stat */
 911                 return false;
 912         }
 913
 914         icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
 915         return true;
 916 }
 917
 918 /*
 919  *      Handle ICMP_ECHO ("ping") requests.
 920  *
 921  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 922  *                requests.
 923  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 924  *                included in the reply.
 925  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 926  *                echo requests, MUST have default=NOT.
 927  *      See also WRT handling of options once they are done and working.
 928  */
 929
 930 static bool icmp_echo(struct sk_buff *skb)
 931 {
 932         struct net *net;
 933
 934         net = dev_net(skb_dst(skb)->dev);
 935         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 936                 struct icmp_bxm icmp_param;
 937
 938                 icmp_param.data.icmph      = *icmp_hdr(skb);
 939                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 940                 icmp_param.skb             = skb;
 941                 icmp_param.offset          = 0;
 942                 icmp_param.data_len        = skb->len;
 943                 icmp_param.head_len        = sizeof(struct icmphdr);
 944                 icmp_reply(&icmp_param, skb);
 945         }
 946         /* should there be an ICMP stat for ignored echos? */
 947         return true;
 948 }
 949
 950 /*
 951  *      Handle ICMP Timestamp requests.
 952  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 953  *                SHOULD be in the kernel for minimum random latency.
 954  *                MUST be accurate to a few minutes.
 955  *                MUST be updated at least at 15Hz.
 956  */
 957 static bool icmp_timestamp(struct sk_buff *skb)
 958 {
 959         struct icmp_bxm icmp_param;
 960         /*
 961          *      Too short.
 962          */
 963         if (skb->len < 4)
 964                 goto out_err;
 965
 966         /*
 967          *      Fill in the current time as ms since midnight UT:
 968          */
 969         icmp_param.data.times[1] = inet_current_timestamp();
 970         icmp_param.data.times[2] = icmp_param.data.times[1];
 971
 972         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
 973
 974         icmp_param.data.icmph      = *icmp_hdr(skb);
 975         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 976         icmp_param.data.icmph.code = 0;
 977         icmp_param.skb             = skb;
 978         icmp_param.offset          = 0;
 979         icmp_param.data_len        = 0;
 980         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 981         icmp_reply(&icmp_param, skb);
 982         return true;
 983
 984 out_err:
 985         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 986         return false;
 987 }
 988
 989 static bool icmp_discard(struct sk_buff *skb)
 990 {
 991         /* pretend it was a success */
 992         return true;
 993 }
 994
 995 /*
 996  *      Deal with incoming ICMP packets.
 997  */
 998 int icmp_rcv(struct sk_buff *skb)
 999 {
1000         struct icmphdr *icmph;
1001         struct rtable *rt = skb_rtable(skb);
1002         struct net *net = dev_net(rt->dst.dev);
1003         bool success;
1004
1005         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1006                 struct sec_path *sp = skb_sec_path(skb);
1007                 int nh;
1008
1009                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1010                                  XFRM_STATE_ICMP))
1011                         goto drop;
1012
1013                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1014                         goto drop;
1015
1016                 nh = skb_network_offset(skb);
1017                 skb_set_network_header(skb, sizeof(*icmph));
1018
1019                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1020                         goto drop;
1021
1022                 skb_set_network_header(skb, nh);
1023         }
1024
1025         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1026
1027         if (skb_checksum_simple_validate(skb))
1028                 goto csum_error;
1029
1030         if (!pskb_pull(skb, sizeof(*icmph)))
1031                 goto error;
1032
1033         icmph = icmp_hdr(skb);
1034
1035         ICMPMSGIN_INC_STATS(net, icmph->type);
1036         /*
1037          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1038          *
1039          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1040          *                discarded.
1041          */
1042         if (icmph->type > NR_ICMP_TYPES)
1043                 goto error;
1044
1045
1046         /*
1047          *      Parse the ICMP message
1048          */
1049
1050         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1051                 /*
1052                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1053                  *        silently ignored (we let user decide with a sysctl).
1054                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1055                  *        discarded if to broadcast/multicast.
1056                  */
1057                 if ((icmph->type == ICMP_ECHO ||
1058                      icmph->type == ICMP_TIMESTAMP) &&
1059                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1060                         goto error;
1061                 }
1062                 if (icmph->type != ICMP_ECHO &&
1063                     icmph->type != ICMP_TIMESTAMP &&
1064                     icmph->type != ICMP_ADDRESS &&
1065                     icmph->type != ICMP_ADDRESSREPLY) {
1066                         goto error;
1067                 }
1068         }
1069
1070         success = icmp_pointers[icmph->type].handler(skb);
1071
1072         if (success)  {
1073                 consume_skb(skb);
1074                 return NET_RX_SUCCESS;
1075         }
1076
1077 drop:
1078         kfree_skb(skb);
1079         return NET_RX_DROP;
1080 csum_error:
1081         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1082 error:
1083         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1084         goto drop;
1085 }
1086
1087 void icmp_err(struct sk_buff *skb, u32 info)
1088 {
1089         struct iphdr *iph = (struct iphdr *)skb->data;
1090         int offset = iph->ihl<<2;
1091         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1092         int type = icmp_hdr(skb)->type;
1093         int code = icmp_hdr(skb)->code;
1094         struct net *net = dev_net(skb->dev);
1095
1096         /*
1097          * Use ping_err to handle all icmp errors except those
1098          * triggered by ICMP_ECHOREPLY which sent from kernel.
1099          */
1100         if (icmph->type != ICMP_ECHOREPLY) {
1101                 ping_err(skb, offset, info);
1102                 return;
1103         }
1104
1105         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1106                 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1107         else if (type == ICMP_REDIRECT)
1108                 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1109 }
1110
1111 /*
1112  *      This table is the definition of how we handle ICMP.
1113  */
1114 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1115         [ICMP_ECHOREPLY] = {
1116                 .handler = ping_rcv,
1117         },
1118         [1] = {
1119                 .handler = icmp_discard,
1120                 .error = 1,
1121         },
1122         [2] = {
1123                 .handler = icmp_discard,
1124                 .error = 1,
1125         },
1126         [ICMP_DEST_UNREACH] = {
1127                 .handler = icmp_unreach,
1128                 .error = 1,
1129         },
1130         [ICMP_SOURCE_QUENCH] = {
1131                 .handler = icmp_unreach,
1132                 .error = 1,
1133         },
1134         [ICMP_REDIRECT] = {
1135                 .handler = icmp_redirect,
1136                 .error = 1,
1137         },
1138         [6] = {
1139                 .handler = icmp_discard,
1140                 .error = 1,
1141         },
1142         [7] = {
1143                 .handler = icmp_discard,
1144                 .error = 1,
1145         },
1146         [ICMP_ECHO] = {
1147                 .handler = icmp_echo,
1148         },
1149         [9] = {
1150                 .handler = icmp_discard,
1151                 .error = 1,
1152         },
1153         [10] = {
1154                 .handler = icmp_discard,
1155                 .error = 1,
1156         },
1157         [ICMP_TIME_EXCEEDED] = {
1158                 .handler = icmp_unreach,
1159                 .error = 1,
1160         },
1161         [ICMP_PARAMETERPROB] = {
1162                 .handler = icmp_unreach,
1163                 .error = 1,
1164         },
1165         [ICMP_TIMESTAMP] = {
1166                 .handler = icmp_timestamp,
1167         },
1168         [ICMP_TIMESTAMPREPLY] = {
1169                 .handler = icmp_discard,
1170         },
1171         [ICMP_INFO_REQUEST] = {
1172                 .handler = icmp_discard,
1173         },
1174         [ICMP_INFO_REPLY] = {
1175                 .handler = icmp_discard,
1176         },
1177         [ICMP_ADDRESS] = {
1178                 .handler = icmp_discard,
1179         },
1180         [ICMP_ADDRESSREPLY] = {
1181                 .handler = icmp_discard,
1182         },
1183 };
1184
1185 static void __net_exit icmp_sk_exit(struct net *net)
1186 {
1187         int i;
1188
1189         for_each_possible_cpu(i)
1190                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1191         free_percpu(net->ipv4.icmp_sk);
1192         net->ipv4.icmp_sk = NULL;
1193 }
1194
1195 static int __net_init icmp_sk_init(struct net *net)
1196 {
1197         int i, err;
1198
1199         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1200         if (!net->ipv4.icmp_sk)
1201                 return -ENOMEM;
1202
1203         for_each_possible_cpu(i) {
1204                 struct sock *sk;
1205
1206                 err = inet_ctl_sock_create(&sk, PF_INET,
1207                                            SOCK_RAW, IPPROTO_ICMP, net);
1208                 if (err < 0)
1209                         goto fail;
1210
1211                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1212
1213                 /* Enough space for 2 64K ICMP packets, including
1214                  * sk_buff/skb_shared_info struct overhead.
1215                  */
1216                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1217
1218                 /*
1219                  * Speedup sock_wfree()
1220                  */
1221                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1222                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1223         }
1224
1225         /* Control parameters for ECHO replies. */
1226         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1227         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1228
1229         /* Control parameter - ignore bogus broadcast responses? */
1230         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1231
1232         /*
1233          *      Configurable global rate limit.
1234          *
1235          *      ratelimit defines tokens/packet consumed for dst->rate_token
1236          *      bucket ratemask defines which icmp types are ratelimited by
1237          *      setting it's bit position.
1238          *
1239          *      default:
1240          *      dest unreachable (3), source quench (4),
1241          *      time exceeded (11), parameter problem (12)
1242          */
1243
1244         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1245         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1246         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1247
1248         return 0;
1249
1250 fail:
1251         for_each_possible_cpu(i)
1252                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1253         free_percpu(net->ipv4.icmp_sk);
1254         return err;
1255 }
1256
1257 static struct pernet_operations __net_initdata icmp_sk_ops = {
1258        .init = icmp_sk_init,
1259        .exit = icmp_sk_exit,
1260 };
1261
1262 int __init icmp_init(void)
1263 {
1264         return register_pernet_subsys(&icmp_sk_ops);
1265 }