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