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