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(&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 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 468
 469 /* Source and destination is swapped. See ip_multipath_icmp_hash */
 470 static int icmp_multipath_hash_skb(const struct sk_buff *skb)
 471 {
 472         const struct iphdr *iph = ip_hdr(skb);
 473
 474         return fib_multipath_hash(iph->daddr, iph->saddr);
 475 }
 476
 477 #else
 478
 479 #define icmp_multipath_hash_skb(skb) (-1)
 480
 481 #endif
 482
 483 static struct rtable *icmp_route_lookup(struct net *net,
 484                                         struct flowi4 *fl4,
 485                                         struct sk_buff *skb_in,
 486                                         const struct iphdr *iph,
 487                                         __be32 saddr, u8 tos, u32 mark,
 488                                         int type, int code,
 489                                         struct icmp_bxm *param)
 490 {
 491         struct rtable *rt, *rt2;
 492         struct flowi4 fl4_dec;
 493         int err;
 494
 495         memset(fl4, 0, sizeof(*fl4));
 496         fl4->daddr = (param->replyopts.opt.opt.srr ?
 497                       param->replyopts.opt.opt.faddr : iph->saddr);
 498         fl4->saddr = saddr;
 499         fl4->flowi4_mark = mark;
 500         fl4->flowi4_uid = sock_net_uid(net, NULL);
 501         fl4->flowi4_tos = RT_TOS(tos);
 502         fl4->flowi4_proto = IPPROTO_ICMP;
 503         fl4->fl4_icmp_type = type;
 504         fl4->fl4_icmp_code = code;
 505         fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
 506
 507         security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 508         rt = __ip_route_output_key_hash(net, fl4,
 509                                         icmp_multipath_hash_skb(skb_in));
 510         if (IS_ERR(rt))
 511                 return rt;
 512
 513         /* No need to clone since we're just using its address. */
 514         rt2 = rt;
 515
 516         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 517                                            flowi4_to_flowi(fl4), NULL, 0);
 518         if (!IS_ERR(rt)) {
 519                 if (rt != rt2)
 520                         return rt;
 521         } else if (PTR_ERR(rt) == -EPERM) {
 522                 rt = NULL;
 523         } else
 524                 return rt;
 525
 526         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 527         if (err)
 528                 goto relookup_failed;
 529
 530         if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
 531                                      fl4_dec.saddr) == RTN_LOCAL) {
 532                 rt2 = __ip_route_output_key(net, &fl4_dec);
 533                 if (IS_ERR(rt2))
 534                         err = PTR_ERR(rt2);
 535         } else {
 536                 struct flowi4 fl4_2 = {};
 537                 unsigned long orefdst;
 538
 539                 fl4_2.daddr = fl4_dec.saddr;
 540                 rt2 = ip_route_output_key(net, &fl4_2);
 541                 if (IS_ERR(rt2)) {
 542                         err = PTR_ERR(rt2);
 543                         goto relookup_failed;
 544                 }
 545                 /* Ugh! */
 546                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 547                 skb_dst_set(skb_in, NULL);
 548                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 549                                      RT_TOS(tos), rt2->dst.dev);
 550
 551                 dst_release(&rt2->dst);
 552                 rt2 = skb_rtable(skb_in);
 553                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 554         }
 555
 556         if (err)
 557                 goto relookup_failed;
 558
 559         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 560                                             flowi4_to_flowi(&fl4_dec), NULL,
 561                                             XFRM_LOOKUP_ICMP);
 562         if (!IS_ERR(rt2)) {
 563                 dst_release(&rt->dst);
 564                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 565                 rt = rt2;
 566         } else if (PTR_ERR(rt2) == -EPERM) {
 567                 if (rt)
 568                         dst_release(&rt->dst);
 569                 return rt2;
 570         } else {
 571                 err = PTR_ERR(rt2);
 572                 goto relookup_failed;
 573         }
 574         return rt;
 575
 576 relookup_failed:
 577         if (rt)
 578                 return rt;
 579         return ERR_PTR(err);
 580 }
 581
 582 /*
 583  *      Send an ICMP message in response to a situation
 584  *
 585  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 586  *                MAY send more (we do).
 587  *                      MUST NOT change this header information.
 588  *                      MUST NOT reply to a multicast/broadcast IP address.
 589  *                      MUST NOT reply to a multicast/broadcast MAC address.
 590  *                      MUST reply to only the first fragment.
 591  */
 592
 593 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 594 {
 595         struct iphdr *iph;
 596         int room;
 597         struct icmp_bxm icmp_param;
 598         struct rtable *rt = skb_rtable(skb_in);
 599         struct ipcm_cookie ipc;
 600         struct flowi4 fl4;
 601         __be32 saddr;
 602         u8  tos;
 603         u32 mark;
 604         struct net *net;
 605         struct sock *sk;
 606
 607         if (!rt)
 608                 goto out;
 609         net = dev_net(rt->dst.dev);
 610
 611         /*
 612          *      Find the original header. It is expected to be valid, of course.
 613          *      Check this, icmp_send is called from the most obscure devices
 614          *      sometimes.
 615          */
 616         iph = ip_hdr(skb_in);
 617
 618         if ((u8 *)iph < skb_in->head ||
 619             (skb_network_header(skb_in) + sizeof(*iph)) >
 620             skb_tail_pointer(skb_in))
 621                 goto out;
 622
 623         /*
 624          *      No replies to physical multicast/broadcast
 625          */
 626         if (skb_in->pkt_type != PACKET_HOST)
 627                 goto out;
 628
 629         /*
 630          *      Now check at the protocol level
 631          */
 632         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 633                 goto out;
 634
 635         /*
 636          *      Only reply to fragment 0. We byte re-order the constant
 637          *      mask for efficiency.
 638          */
 639         if (iph->frag_off & htons(IP_OFFSET))
 640                 goto out;
 641
 642         /*
 643          *      If we send an ICMP error to an ICMP error a mess would result..
 644          */
 645         if (icmp_pointers[type].error) {
 646                 /*
 647                  *      We are an error, check if we are replying to an
 648                  *      ICMP error
 649                  */
 650                 if (iph->protocol == IPPROTO_ICMP) {
 651                         u8 _inner_type, *itp;
 652
 653                         itp = skb_header_pointer(skb_in,
 654                                                  skb_network_header(skb_in) +
 655                                                  (iph->ihl << 2) +
 656                                                  offsetof(struct icmphdr,
 657                                                           type) -
 658                                                  skb_in->data,
 659                                                  sizeof(_inner_type),
 660                                                  &_inner_type);
 661                         if (!itp)
 662                                 goto out;
 663
 664                         /*
 665                          *      Assume any unknown ICMP type is an error. This
 666                          *      isn't specified by the RFC, but think about it..
 667                          */
 668                         if (*itp > NR_ICMP_TYPES ||
 669                             icmp_pointers[*itp].error)
 670                                 goto out;
 671                 }
 672         }
 673
 674         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 675         local_bh_disable();
 676
 677         /* Check global sysctl_icmp_msgs_per_sec ratelimit */
 678         if (!icmpv4_global_allow(net, type, code))
 679                 goto out_bh_enable;
 680
 681         sk = icmp_xmit_lock(net);
 682         if (!sk)
 683                 goto out_bh_enable;
 684
 685         /*
 686          *      Construct source address and options.
 687          */
 688
 689         saddr = iph->daddr;
 690         if (!(rt->rt_flags & RTCF_LOCAL)) {
 691                 struct net_device *dev = NULL;
 692
 693                 rcu_read_lock();
 694                 if (rt_is_input_route(rt) &&
 695                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 696                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 697
 698                 if (dev)
 699                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 700                 else
 701                         saddr = 0;
 702                 rcu_read_unlock();
 703         }
 704
 705         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 706                                            IPTOS_PREC_INTERNETCONTROL) :
 707                                           iph->tos;
 708         mark = IP4_REPLY_MARK(net, skb_in->mark);
 709
 710         if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
 711                 goto out_unlock;
 712
 713
 714         /*
 715          *      Prepare data for ICMP header.
 716          */
 717
 718         icmp_param.data.icmph.type       = type;
 719         icmp_param.data.icmph.code       = code;
 720         icmp_param.data.icmph.un.gateway = info;
 721         icmp_param.data.icmph.checksum   = 0;
 722         icmp_param.skb    = skb_in;
 723         icmp_param.offset = skb_network_offset(skb_in);
 724         inet_sk(sk)->tos = tos;
 725         sk->sk_mark = mark;
 726         ipc.addr = iph->saddr;
 727         ipc.opt = &icmp_param.replyopts.opt;
 728         ipc.tx_flags = 0;
 729         ipc.ttl = 0;
 730         ipc.tos = -1;
 731
 732         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 733                                type, code, &icmp_param);
 734         if (IS_ERR(rt))
 735                 goto out_unlock;
 736
 737         /* peer icmp_ratelimit */
 738         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 739                 goto ende;
 740
 741         /* RFC says return as much as we can without exceeding 576 bytes. */
 742
 743         room = dst_mtu(&rt->dst);
 744         if (room > 576)
 745                 room = 576;
 746         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 747         room -= sizeof(struct icmphdr);
 748
 749         icmp_param.data_len = skb_in->len - icmp_param.offset;
 750         if (icmp_param.data_len > room)
 751                 icmp_param.data_len = room;
 752         icmp_param.head_len = sizeof(struct icmphdr);
 753
 754         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 755 ende:
 756         ip_rt_put(rt);
 757 out_unlock:
 758         icmp_xmit_unlock(sk);
 759 out_bh_enable:
 760         local_bh_enable();
 761 out:;
 762 }
 763 EXPORT_SYMBOL(icmp_send);
 764
 765
 766 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 767 {
 768         const struct iphdr *iph = (const struct iphdr *) skb->data;
 769         const struct net_protocol *ipprot;
 770         int protocol = iph->protocol;
 771
 772         /* Checkin full IP header plus 8 bytes of protocol to
 773          * avoid additional coding at protocol handlers.
 774          */
 775         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 776                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 777                 return;
 778         }
 779
 780         raw_icmp_error(skb, protocol, info);
 781
 782         ipprot = rcu_dereference(inet_protos[protocol]);
 783         if (ipprot && ipprot->err_handler)
 784                 ipprot->err_handler(skb, info);
 785 }
 786
 787 static bool icmp_tag_validation(int proto)
 788 {
 789         bool ok;
 790
 791         rcu_read_lock();
 792         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 793         rcu_read_unlock();
 794         return ok;
 795 }
 796
 797 /*
 798  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
 799  *      ICMP_PARAMETERPROB.
 800  */
 801
 802 static bool icmp_unreach(struct sk_buff *skb)
 803 {
 804         const struct iphdr *iph;
 805         struct icmphdr *icmph;
 806         struct net *net;
 807         u32 info = 0;
 808
 809         net = dev_net(skb_dst(skb)->dev);
 810
 811         /*
 812          *      Incomplete header ?
 813          *      Only checks for the IP header, there should be an
 814          *      additional check for longer headers in upper levels.
 815          */
 816
 817         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 818                 goto out_err;
 819
 820         icmph = icmp_hdr(skb);
 821         iph   = (const struct iphdr *)skb->data;
 822
 823         if (iph->ihl < 5) /* Mangled header, drop. */
 824                 goto out_err;
 825
 826         if (icmph->type == ICMP_DEST_UNREACH) {
 827                 switch (icmph->code & 15) {
 828                 case ICMP_NET_UNREACH:
 829                 case ICMP_HOST_UNREACH:
 830                 case ICMP_PROT_UNREACH:
 831                 case ICMP_PORT_UNREACH:
 832                         break;
 833                 case ICMP_FRAG_NEEDED:
 834                         /* for documentation of the ip_no_pmtu_disc
 835                          * values please see
 836                          * Documentation/networking/ip-sysctl.txt
 837                          */
 838                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 839                         default:
 840                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 841                                                     &iph->daddr);
 842                                 break;
 843                         case 2:
 844                                 goto out;
 845                         case 3:
 846                                 if (!icmp_tag_validation(iph->protocol))
 847                                         goto out;
 848                                 /* fall through */
 849                         case 0:
 850                                 info = ntohs(icmph->un.frag.mtu);
 851                         }
 852                         break;
 853                 case ICMP_SR_FAILED:
 854                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 855                                             &iph->daddr);
 856                         break;
 857                 default:
 858                         break;
 859                 }
 860                 if (icmph->code > NR_ICMP_UNREACH)
 861                         goto out;
 862         } else if (icmph->type == ICMP_PARAMETERPROB)
 863                 info = ntohl(icmph->un.gateway) >> 24;
 864
 865         /*
 866          *      Throw it at our lower layers
 867          *
 868          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 869          *                header.
 870          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 871          *                transport layer.
 872          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 873          *                transport layer.
 874          */
 875
 876         /*
 877          *      Check the other end isn't violating RFC 1122. Some routers send
 878          *      bogus responses to broadcast frames. If you see this message
 879          *      first check your netmask matches at both ends, if it does then
 880          *      get the other vendor to fix their kit.
 881          */
 882
 883         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 884             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 885                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 886                                      &ip_hdr(skb)->saddr,
 887                                      icmph->type, icmph->code,
 888                                      &iph->daddr, skb->dev->name);
 889                 goto out;
 890         }
 891
 892         icmp_socket_deliver(skb, info);
 893
 894 out:
 895         return true;
 896 out_err:
 897         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 898         return false;
 899 }
 900
 901
 902 /*
 903  *      Handle ICMP_REDIRECT.
 904  */
 905
 906 static bool icmp_redirect(struct sk_buff *skb)
 907 {
 908         if (skb->len < sizeof(struct iphdr)) {
 909                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 910                 return false;
 911         }
 912
 913         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 914                 /* there aught to be a stat */
 915                 return false;
 916         }
 917
 918         icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
 919         return true;
 920 }
 921
 922 /*
 923  *      Handle ICMP_ECHO ("ping") requests.
 924  *
 925  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 926  *                requests.
 927  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 928  *                included in the reply.
 929  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 930  *                echo requests, MUST have default=NOT.
 931  *      See also WRT handling of options once they are done and working.
 932  */
 933
 934 static bool icmp_echo(struct sk_buff *skb)
 935 {
 936         struct net *net;
 937
 938         net = dev_net(skb_dst(skb)->dev);
 939         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 940                 struct icmp_bxm icmp_param;
 941
 942                 icmp_param.data.icmph      = *icmp_hdr(skb);
 943                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 944                 icmp_param.skb             = skb;
 945                 icmp_param.offset          = 0;
 946                 icmp_param.data_len        = skb->len;
 947                 icmp_param.head_len        = sizeof(struct icmphdr);
 948                 icmp_reply(&icmp_param, skb);
 949         }
 950         /* should there be an ICMP stat for ignored echos? */
 951         return true;
 952 }
 953
 954 /*
 955  *      Handle ICMP Timestamp requests.
 956  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 957  *                SHOULD be in the kernel for minimum random latency.
 958  *                MUST be accurate to a few minutes.
 959  *                MUST be updated at least at 15Hz.
 960  */
 961 static bool icmp_timestamp(struct sk_buff *skb)
 962 {
 963         struct icmp_bxm icmp_param;
 964         /*
 965          *      Too short.
 966          */
 967         if (skb->len < 4)
 968                 goto out_err;
 969
 970         /*
 971          *      Fill in the current time as ms since midnight UT:
 972          */
 973         icmp_param.data.times[1] = inet_current_timestamp();
 974         icmp_param.data.times[2] = icmp_param.data.times[1];
 975         if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
 976                 BUG();
 977         icmp_param.data.icmph      = *icmp_hdr(skb);
 978         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 979         icmp_param.data.icmph.code = 0;
 980         icmp_param.skb             = skb;
 981         icmp_param.offset          = 0;
 982         icmp_param.data_len        = 0;
 983         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 984         icmp_reply(&icmp_param, skb);
 985         return true;
 986
 987 out_err:
 988         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 989         return false;
 990 }
 991
 992 static bool icmp_discard(struct sk_buff *skb)
 993 {
 994         /* pretend it was a success */
 995         return true;
 996 }
 997
 998 /*
 999  *      Deal with incoming ICMP packets.
1000  */
1001 int icmp_rcv(struct sk_buff *skb)
1002 {
1003         struct icmphdr *icmph;
1004         struct rtable *rt = skb_rtable(skb);
1005         struct net *net = dev_net(rt->dst.dev);
1006         bool success;
1007
1008         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1009                 struct sec_path *sp = skb_sec_path(skb);
1010                 int nh;
1011
1012                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1013                                  XFRM_STATE_ICMP))
1014                         goto drop;
1015
1016                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1017                         goto drop;
1018
1019                 nh = skb_network_offset(skb);
1020                 skb_set_network_header(skb, sizeof(*icmph));
1021
1022                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1023                         goto drop;
1024
1025                 skb_set_network_header(skb, nh);
1026         }
1027
1028         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1029
1030         if (skb_checksum_simple_validate(skb))
1031                 goto csum_error;
1032
1033         if (!pskb_pull(skb, sizeof(*icmph)))
1034                 goto error;
1035
1036         icmph = icmp_hdr(skb);
1037
1038         ICMPMSGIN_INC_STATS(net, icmph->type);
1039         /*
1040          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1041          *
1042          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1043          *                discarded.
1044          */
1045         if (icmph->type > NR_ICMP_TYPES)
1046                 goto error;
1047
1048
1049         /*
1050          *      Parse the ICMP message
1051          */
1052
1053         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1054                 /*
1055                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1056                  *        silently ignored (we let user decide with a sysctl).
1057                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1058                  *        discarded if to broadcast/multicast.
1059                  */
1060                 if ((icmph->type == ICMP_ECHO ||
1061                      icmph->type == ICMP_TIMESTAMP) &&
1062                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1063                         goto error;
1064                 }
1065                 if (icmph->type != ICMP_ECHO &&
1066                     icmph->type != ICMP_TIMESTAMP &&
1067                     icmph->type != ICMP_ADDRESS &&
1068                     icmph->type != ICMP_ADDRESSREPLY) {
1069                         goto error;
1070                 }
1071         }
1072
1073         success = icmp_pointers[icmph->type].handler(skb);
1074
1075         if (success)  {
1076                 consume_skb(skb);
1077                 return NET_RX_SUCCESS;
1078         }
1079
1080 drop:
1081         kfree_skb(skb);
1082         return NET_RX_DROP;
1083 csum_error:
1084         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1085 error:
1086         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1087         goto drop;
1088 }
1089
1090 void icmp_err(struct sk_buff *skb, u32 info)
1091 {
1092         struct iphdr *iph = (struct iphdr *)skb->data;
1093         int offset = iph->ihl<<2;
1094         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1095         int type = icmp_hdr(skb)->type;
1096         int code = icmp_hdr(skb)->code;
1097         struct net *net = dev_net(skb->dev);
1098
1099         /*
1100          * Use ping_err to handle all icmp errors except those
1101          * triggered by ICMP_ECHOREPLY which sent from kernel.
1102          */
1103         if (icmph->type != ICMP_ECHOREPLY) {
1104                 ping_err(skb, offset, info);
1105                 return;
1106         }
1107
1108         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1109                 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1110         else if (type == ICMP_REDIRECT)
1111                 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1112 }
1113
1114 /*
1115  *      This table is the definition of how we handle ICMP.
1116  */
1117 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1118         [ICMP_ECHOREPLY] = {
1119                 .handler = ping_rcv,
1120         },
1121         [1] = {
1122                 .handler = icmp_discard,
1123                 .error = 1,
1124         },
1125         [2] = {
1126                 .handler = icmp_discard,
1127                 .error = 1,
1128         },
1129         [ICMP_DEST_UNREACH] = {
1130                 .handler = icmp_unreach,
1131                 .error = 1,
1132         },
1133         [ICMP_SOURCE_QUENCH] = {
1134                 .handler = icmp_unreach,
1135                 .error = 1,
1136         },
1137         [ICMP_REDIRECT] = {
1138                 .handler = icmp_redirect,
1139                 .error = 1,
1140         },
1141         [6] = {
1142                 .handler = icmp_discard,
1143                 .error = 1,
1144         },
1145         [7] = {
1146                 .handler = icmp_discard,
1147                 .error = 1,
1148         },
1149         [ICMP_ECHO] = {
1150                 .handler = icmp_echo,
1151         },
1152         [9] = {
1153                 .handler = icmp_discard,
1154                 .error = 1,
1155         },
1156         [10] = {
1157                 .handler = icmp_discard,
1158                 .error = 1,
1159         },
1160         [ICMP_TIME_EXCEEDED] = {
1161                 .handler = icmp_unreach,
1162                 .error = 1,
1163         },
1164         [ICMP_PARAMETERPROB] = {
1165                 .handler = icmp_unreach,
1166                 .error = 1,
1167         },
1168         [ICMP_TIMESTAMP] = {
1169                 .handler = icmp_timestamp,
1170         },
1171         [ICMP_TIMESTAMPREPLY] = {
1172                 .handler = icmp_discard,
1173         },
1174         [ICMP_INFO_REQUEST] = {
1175                 .handler = icmp_discard,
1176         },
1177         [ICMP_INFO_REPLY] = {
1178                 .handler = icmp_discard,
1179         },
1180         [ICMP_ADDRESS] = {
1181                 .handler = icmp_discard,
1182         },
1183         [ICMP_ADDRESSREPLY] = {
1184                 .handler = icmp_discard,
1185         },
1186 };
1187
1188 static void __net_exit icmp_sk_exit(struct net *net)
1189 {
1190         int i;
1191
1192         for_each_possible_cpu(i)
1193                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1194         free_percpu(net->ipv4.icmp_sk);
1195         net->ipv4.icmp_sk = NULL;
1196 }
1197
1198 static int __net_init icmp_sk_init(struct net *net)
1199 {
1200         int i, err;
1201
1202         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1203         if (!net->ipv4.icmp_sk)
1204                 return -ENOMEM;
1205
1206         for_each_possible_cpu(i) {
1207                 struct sock *sk;
1208
1209                 err = inet_ctl_sock_create(&sk, PF_INET,
1210                                            SOCK_RAW, IPPROTO_ICMP, net);
1211                 if (err < 0)
1212                         goto fail;
1213
1214                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1215
1216                 /* Enough space for 2 64K ICMP packets, including
1217                  * sk_buff/skb_shared_info struct overhead.
1218                  */
1219                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1220
1221                 /*
1222                  * Speedup sock_wfree()
1223                  */
1224                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1225                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1226         }
1227
1228         /* Control parameters for ECHO replies. */
1229         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1230         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1231
1232         /* Control parameter - ignore bogus broadcast responses? */
1233         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1234
1235         /*
1236          *      Configurable global rate limit.
1237          *
1238          *      ratelimit defines tokens/packet consumed for dst->rate_token
1239          *      bucket ratemask defines which icmp types are ratelimited by
1240          *      setting it's bit position.
1241          *
1242          *      default:
1243          *      dest unreachable (3), source quench (4),
1244          *      time exceeded (11), parameter problem (12)
1245          */
1246
1247         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1248         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1249         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1250
1251         return 0;
1252
1253 fail:
1254         for_each_possible_cpu(i)
1255                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1256         free_percpu(net->ipv4.icmp_sk);
1257         return err;
1258 }
1259
1260 static struct pernet_operations __net_initdata icmp_sk_ops = {
1261        .init = icmp_sk_init,
1262        .exit = icmp_sk_exit,
1263 };
1264
1265 int __init icmp_init(void)
1266 {
1267         return register_pernet_subsys(&icmp_sk_ops);
1268 }