net/ipv4/icmp.c

   1 /*
   2  *      NET3:   Implementation of the ICMP protocol layer.
   3  *
   4  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  *
  11  *      Some of the function names and the icmp unreach table for this
  12  *      module were derived from [icmp.c 1.0.11 06/02/93] by
  13  *      Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  14  *      Other than that this module is a complete rewrite.
  15  *
  16  *      Fixes:
  17  *      Clemens Fruhwirth       :       introduce global icmp rate limiting
  18  *                                      with icmp type masking ability instead
  19  *                                      of broken per type icmp timeouts.
  20  *              Mike Shaver     :       RFC1122 checks.
  21  *              Alan Cox        :       Multicast ping reply as self.
  22  *              Alan Cox        :       Fix atomicity lockup in ip_build_xmit
  23  *                                      call.
  24  *              Alan Cox        :       Added 216,128 byte paths to the MTU
  25  *                                      code.
  26  *              Martin Mares    :       RFC1812 checks.
  27  *              Martin Mares    :       Can be configured to follow redirects
  28  *                                      if acting as a router _without_ a
  29  *                                      routing protocol (RFC 1812).
  30  *              Martin Mares    :       Echo requests may be configured to
  31  *                                      be ignored (RFC 1812).
  32  *              Martin Mares    :       Limitation of ICMP error message
  33  *                                      transmit rate (RFC 1812).
  34  *              Martin Mares    :       TOS and Precedence set correctly
  35  *                                      (RFC 1812).
  36  *              Martin Mares    :       Now copying as much data from the
  37  *                                      original packet as we can without
  38  *                                      exceeding 576 bytes (RFC 1812).
  39  *      Willy Konynenberg       :       Transparent proxying support.
  40  *              Keith Owens     :       RFC1191 correction for 4.2BSD based
  41  *                                      path MTU bug.
  42  *              Thomas Quinot   :       ICMP Dest Unreach codes up to 15 are
  43  *                                      valid (RFC 1812).
  44  *              Andi Kleen      :       Check all packet lengths properly
  45  *                                      and moved all kfree_skb() up to
  46  *                                      icmp_rcv.
  47  *              Andi Kleen      :       Move the rate limit bookkeeping
  48  *                                      into the dest entry and use a token
  49  *                                      bucket filter (thanks to ANK). Make
  50  *                                      the rates sysctl configurable.
  51  *              Yu Tianli       :       Fixed two ugly bugs in icmp_send
  52  *                                      - IP option length was accounted wrongly
  53  *                                      - ICMP header length was not accounted
  54  *                                        at all.
  55  *              Tristan Greaves :       Added sysctl option to ignore bogus
  56  *                                      broadcast responses from broken routers.
  57  *
  58  * To Fix:
  59  *
  60  *      - Should use skb_pull() instead of all the manual checking.
  61  *        This would also greatly simply some upper layer error handlers. --AK
  62  *
  63  */
  64
  65 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  66
  67 #include <linux/module.h>
  68 #include <linux/types.h>
  69 #include <linux/jiffies.h>
  70 #include <linux/kernel.h>
  71 #include <linux/fcntl.h>
  72 #include <linux/socket.h>
  73 #include <linux/in.h>
  74 #include <linux/inet.h>
  75 #include <linux/inetdevice.h>
  76 #include <linux/netdevice.h>
  77 #include <linux/string.h>
  78 #include <linux/netfilter_ipv4.h>
  79 #include <linux/slab.h>
  80 #include <net/snmp.h>
  81 #include <net/ip.h>
  82 #include <net/route.h>
  83 #include <net/protocol.h>
  84 #include <net/icmp.h>
  85 #include <net/tcp.h>
  86 #include <net/udp.h>
  87 #include <net/raw.h>
  88 #include <net/ping.h>
  89 #include <linux/skbuff.h>
  90 #include <net/sock.h>
  91 #include <linux/errno.h>
  92 #include <linux/timer.h>
  93 #include <linux/init.h>
  94 #include <linux/uaccess.h>
  95 #include <net/checksum.h>
  96 #include <net/xfrm.h>
  97 #include <net/inet_common.h>
  98 #include <net/ip_fib.h>
  99 #include <net/l3mdev.h>
 100
 101 /*
 102  *      Build xmit assembly blocks
 103  */
 104
 105 struct icmp_bxm {
 106         struct sk_buff *skb;
 107         int offset;
 108         int data_len;
 109
 110         struct {
 111                 struct icmphdr icmph;
 112                 __be32         times[3];
 113         } data;
 114         int head_len;
 115         struct ip_options_data replyopts;
 116 };
 117
 118 /* An array of errno for error messages from dest unreach. */
 119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 120
 121 const struct icmp_err icmp_err_convert[] = {
 122         {
 123                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 124                 .fatal = 0,
 125         },
 126         {
 127                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 128                 .fatal = 0,
 129         },
 130         {
 131                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 132                 .fatal = 1,
 133         },
 134         {
 135                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 136                 .fatal = 1,
 137         },
 138         {
 139                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 140                 .fatal = 0,
 141         },
 142         {
 143                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 144                 .fatal = 0,
 145         },
 146         {
 147                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 148                 .fatal = 1,
 149         },
 150         {
 151                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 152                 .fatal = 1,
 153         },
 154         {
 155                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 156                 .fatal = 1,
 157         },
 158         {
 159                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 160                 .fatal = 1,
 161         },
 162         {
 163                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 164                 .fatal = 1,
 165         },
 166         {
 167                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 168                 .fatal = 0,
 169         },
 170         {
 171                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 172                 .fatal = 0,
 173         },
 174         {
 175                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 176                 .fatal = 1,
 177         },
 178         {
 179                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 180                 .fatal = 1,
 181         },
 182         {
 183                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 184                 .fatal = 1,
 185         },
 186 };
 187 EXPORT_SYMBOL(icmp_err_convert);
 188
 189 /*
 190  *      ICMP control array. This specifies what to do with each ICMP.
 191  */
 192
 193 struct icmp_control {
 194         bool (*handler)(struct sk_buff *skb);
 195         short   error;          /* This ICMP is classed as an error message */
 196 };
 197
 198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 199
 200 /*
 201  *      The ICMP socket(s). This is the most convenient way to flow control
 202  *      our ICMP output as well as maintain a clean interface throughout
 203  *      all layers. All Socketless IP sends will soon be gone.
 204  *
 205  *      On SMP we have one ICMP socket per-cpu.
 206  */
 207 static struct sock *icmp_sk(struct net *net)
 208 {
 209         return *this_cpu_ptr(net->ipv4.icmp_sk);
 210 }
 211
 212 /* Called with BH disabled */
 213 static inline struct sock *icmp_xmit_lock(struct net *net)
 214 {
 215         struct sock *sk;
 216
 217         sk = icmp_sk(net);
 218
 219         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 220                 /* This can happen if the output path signals a
 221                  * dst_link_failure() for an outgoing ICMP packet.
 222                  */
 223                 return NULL;
 224         }
 225         return sk;
 226 }
 227
 228 static inline void icmp_xmit_unlock(struct sock *sk)
 229 {
 230         spin_unlock(&sk->sk_lock.slock);
 231 }
 232
 233 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 234 int sysctl_icmp_msgs_burst __read_mostly = 50;
 235
 236 static struct {
 237         spinlock_t      lock;
 238         u32             credit;
 239         u32             stamp;
 240 } icmp_global = {
 241         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 242 };
 243
 244 /**
 245  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 246  *
 247  * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
 248  * Returns false if we reached the limit and can not send another packet.
 249  * Note: called with BH disabled
 250  */
 251 bool icmp_global_allow(void)
 252 {
 253         u32 credit, delta, incr = 0, now = (u32)jiffies;
 254         bool rc = false;
 255
 256         /* Check if token bucket is empty and cannot be refilled
 257          * without taking the spinlock.
 258          */
 259         if (!icmp_global.credit) {
 260                 delta = min_t(u32, now - icmp_global.stamp, HZ);
 261                 if (delta < HZ / 50)
 262                         return false;
 263         }
 264
 265         spin_lock(&icmp_global.lock);
 266         delta = min_t(u32, now - icmp_global.stamp, HZ);
 267         if (delta >= HZ / 50) {
 268                 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
 269                 if (incr)
 270                         icmp_global.stamp = now;
 271         }
 272         credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
 273         if (credit) {
 274                 credit--;
 275                 rc = true;
 276         }
 277         icmp_global.credit = credit;
 278         spin_unlock(&icmp_global.lock);
 279         return rc;
 280 }
 281 EXPORT_SYMBOL(icmp_global_allow);
 282
 283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 284 {
 285         if (type > NR_ICMP_TYPES)
 286                 return true;
 287
 288         /* Don't limit PMTU discovery. */
 289         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 290                 return true;
 291
 292         /* Limit if icmp type is enabled in ratemask. */
 293         if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 294                 return true;
 295
 296         return false;
 297 }
 298
 299 static bool icmpv4_global_allow(struct net *net, int type, int code)
 300 {
 301         if (icmpv4_mask_allow(net, type, code))
 302                 return true;
 303
 304         if (icmp_global_allow())
 305                 return true;
 306
 307         return false;
 308 }
 309
 310 /*
 311  *      Send an ICMP frame.
 312  */
 313
 314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 315                                struct flowi4 *fl4, int type, int code)
 316 {
 317         struct dst_entry *dst = &rt->dst;
 318         struct inet_peer *peer;
 319         bool rc = true;
 320         int vif;
 321
 322         if (icmpv4_mask_allow(net, type, code))
 323                 goto out;
 324
 325         /* No rate limit on loopback */
 326         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 327                 goto out;
 328
 329         vif = l3mdev_master_ifindex(dst->dev);
 330         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 331         rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
 332         if (peer)
 333                 inet_putpeer(peer);
 334 out:
 335         return rc;
 336 }
 337
 338 /*
 339  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 340  */
 341 void icmp_out_count(struct net *net, unsigned char type)
 342 {
 343         ICMPMSGOUT_INC_STATS(net, type);
 344         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 345 }
 346
 347 /*
 348  *      Checksum each fragment, and on the first include the headers and final
 349  *      checksum.
 350  */
 351 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 352                           struct sk_buff *skb)
 353 {
 354         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 355         __wsum csum;
 356
 357         csum = skb_copy_and_csum_bits(icmp_param->skb,
 358                                       icmp_param->offset + offset,
 359                                       to, len, 0);
 360
 361         skb->csum = csum_block_add(skb->csum, csum, odd);
 362         if (icmp_pointers[icmp_param->data.icmph.type].error)
 363                 nf_ct_attach(skb, icmp_param->skb);
 364         return 0;
 365 }
 366
 367 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 368                             struct flowi4 *fl4,
 369                             struct ipcm_cookie *ipc, struct rtable **rt)
 370 {
 371         struct sock *sk;
 372         struct sk_buff *skb;
 373
 374         sk = icmp_sk(dev_net((*rt)->dst.dev));
 375         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 376                            icmp_param->data_len+icmp_param->head_len,
 377                            icmp_param->head_len,
 378                            ipc, rt, MSG_DONTWAIT) < 0) {
 379                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 380                 ip_flush_pending_frames(sk);
 381         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 382                 struct icmphdr *icmph = icmp_hdr(skb);
 383                 __wsum csum = 0;
 384                 struct sk_buff *skb1;
 385
 386                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 387                         csum = csum_add(csum, skb1->csum);
 388                 }
 389                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 390                                                  (char *)icmph,
 391                                                  icmp_param->head_len, csum);
 392                 icmph->checksum = csum_fold(csum);
 393                 skb->ip_summed = CHECKSUM_NONE;
 394                 ip_push_pending_frames(sk, fl4);
 395         }
 396 }
 397
 398 /*
 399  *      Driving logic for building and sending ICMP messages.
 400  */
 401
 402 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 403 {
 404         struct ipcm_cookie ipc;
 405         struct rtable *rt = skb_rtable(skb);
 406         struct net *net = dev_net(rt->dst.dev);
 407         struct flowi4 fl4;
 408         struct sock *sk;
 409         struct inet_sock *inet;
 410         __be32 daddr, saddr;
 411         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 412         int type = icmp_param->data.icmph.type;
 413         int code = icmp_param->data.icmph.code;
 414
 415         if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 416                 return;
 417
 418         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 419         local_bh_disable();
 420
 421         /* global icmp_msgs_per_sec */
 422         if (!icmpv4_global_allow(net, type, code))
 423                 goto out_bh_enable;
 424
 425         sk = icmp_xmit_lock(net);
 426         if (!sk)
 427                 goto out_bh_enable;
 428         inet = inet_sk(sk);
 429
 430         icmp_param->data.icmph.checksum = 0;
 431
 432         ipcm_init(&ipc);
 433         inet->tos = ip_hdr(skb)->tos;
 434         sk->sk_mark = mark;
 435         daddr = ipc.addr = ip_hdr(skb)->saddr;
 436         saddr = fib_compute_spec_dst(skb);
 437
 438         if (icmp_param->replyopts.opt.opt.optlen) {
 439                 ipc.opt = &icmp_param->replyopts.opt;
 440                 if (ipc.opt->opt.srr)
 441                         daddr = icmp_param->replyopts.opt.opt.faddr;
 442         }
 443         memset(&fl4, 0, sizeof(fl4));
 444         fl4.daddr = daddr;
 445         fl4.saddr = saddr;
 446         fl4.flowi4_mark = mark;
 447         fl4.flowi4_uid = sock_net_uid(net, NULL);
 448         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 449         fl4.flowi4_proto = IPPROTO_ICMP;
 450         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 451         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 452         rt = ip_route_output_key(net, &fl4);
 453         if (IS_ERR(rt))
 454                 goto out_unlock;
 455         if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 456                 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 457         ip_rt_put(rt);
 458 out_unlock:
 459         icmp_xmit_unlock(sk);
 460 out_bh_enable:
 461         local_bh_enable();
 462 }
 463
 464 static struct rtable *icmp_route_lookup(struct net *net,
 465                                         struct flowi4 *fl4,
 466                                         struct sk_buff *skb_in,
 467                                         const struct iphdr *iph,
 468                                         __be32 saddr, u8 tos, u32 mark,
 469                                         int type, int code,
 470                                         struct icmp_bxm *param)
 471 {
 472         struct rtable *rt, *rt2;
 473         struct flowi4 fl4_dec;
 474         int err;
 475
 476         memset(fl4, 0, sizeof(*fl4));
 477         fl4->daddr = (param->replyopts.opt.opt.srr ?
 478                       param->replyopts.opt.opt.faddr : iph->saddr);
 479         fl4->saddr = saddr;
 480         fl4->flowi4_mark = mark;
 481         fl4->flowi4_uid = sock_net_uid(net, NULL);
 482         fl4->flowi4_tos = RT_TOS(tos);
 483         fl4->flowi4_proto = IPPROTO_ICMP;
 484         fl4->fl4_icmp_type = type;
 485         fl4->fl4_icmp_code = code;
 486         fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
 487
 488         security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 489         rt = ip_route_output_key_hash(net, fl4, skb_in);
 490         if (IS_ERR(rt))
 491                 return rt;
 492
 493         /* No need to clone since we're just using its address. */
 494         rt2 = rt;
 495
 496         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 497                                            flowi4_to_flowi(fl4), NULL, 0);
 498         if (!IS_ERR(rt)) {
 499                 if (rt != rt2)
 500                         return rt;
 501         } else if (PTR_ERR(rt) == -EPERM) {
 502                 rt = NULL;
 503         } else
 504                 return rt;
 505
 506         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 507         if (err)
 508                 goto relookup_failed;
 509
 510         if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
 511                                      fl4_dec.saddr) == RTN_LOCAL) {
 512                 rt2 = __ip_route_output_key(net, &fl4_dec);
 513                 if (IS_ERR(rt2))
 514                         err = PTR_ERR(rt2);
 515         } else {
 516                 struct flowi4 fl4_2 = {};
 517                 unsigned long orefdst;
 518
 519                 fl4_2.daddr = fl4_dec.saddr;
 520                 rt2 = ip_route_output_key(net, &fl4_2);
 521                 if (IS_ERR(rt2)) {
 522                         err = PTR_ERR(rt2);
 523                         goto relookup_failed;
 524                 }
 525                 /* Ugh! */
 526                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 527                 skb_dst_set(skb_in, NULL);
 528                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 529                                      RT_TOS(tos), rt2->dst.dev);
 530
 531                 dst_release(&rt2->dst);
 532                 rt2 = skb_rtable(skb_in);
 533                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 534         }
 535
 536         if (err)
 537                 goto relookup_failed;
 538
 539         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 540                                             flowi4_to_flowi(&fl4_dec), NULL,
 541                                             XFRM_LOOKUP_ICMP);
 542         if (!IS_ERR(rt2)) {
 543                 dst_release(&rt->dst);
 544                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 545                 rt = rt2;
 546         } else if (PTR_ERR(rt2) == -EPERM) {
 547                 if (rt)
 548                         dst_release(&rt->dst);
 549                 return rt2;
 550         } else {
 551                 err = PTR_ERR(rt2);
 552                 goto relookup_failed;
 553         }
 554         return rt;
 555
 556 relookup_failed:
 557         if (rt)
 558                 return rt;
 559         return ERR_PTR(err);
 560 }
 561
 562 /*
 563  *      Send an ICMP message in response to a situation
 564  *
 565  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 566  *                MAY send more (we do).
 567  *                      MUST NOT change this header information.
 568  *                      MUST NOT reply to a multicast/broadcast IP address.
 569  *                      MUST NOT reply to a multicast/broadcast MAC address.
 570  *                      MUST reply to only the first fragment.
 571  */
 572
 573 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 574 {
 575         struct iphdr *iph;
 576         int room;
 577         struct icmp_bxm icmp_param;
 578         struct rtable *rt = skb_rtable(skb_in);
 579         struct ipcm_cookie ipc;
 580         struct flowi4 fl4;
 581         __be32 saddr;
 582         u8  tos;
 583         u32 mark;
 584         struct net *net;
 585         struct sock *sk;
 586
 587         if (!rt)
 588                 goto out;
 589         net = dev_net(rt->dst.dev);
 590
 591         /*
 592          *      Find the original header. It is expected to be valid, of course.
 593          *      Check this, icmp_send is called from the most obscure devices
 594          *      sometimes.
 595          */
 596         iph = ip_hdr(skb_in);
 597
 598         if ((u8 *)iph < skb_in->head ||
 599             (skb_network_header(skb_in) + sizeof(*iph)) >
 600             skb_tail_pointer(skb_in))
 601                 goto out;
 602
 603         /*
 604          *      No replies to physical multicast/broadcast
 605          */
 606         if (skb_in->pkt_type != PACKET_HOST)
 607                 goto out;
 608
 609         /*
 610          *      Now check at the protocol level
 611          */
 612         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 613                 goto out;
 614
 615         /*
 616          *      Only reply to fragment 0. We byte re-order the constant
 617          *      mask for efficiency.
 618          */
 619         if (iph->frag_off & htons(IP_OFFSET))
 620                 goto out;
 621
 622         /*
 623          *      If we send an ICMP error to an ICMP error a mess would result..
 624          */
 625         if (icmp_pointers[type].error) {
 626                 /*
 627                  *      We are an error, check if we are replying to an
 628                  *      ICMP error
 629                  */
 630                 if (iph->protocol == IPPROTO_ICMP) {
 631                         u8 _inner_type, *itp;
 632
 633                         itp = skb_header_pointer(skb_in,
 634                                                  skb_network_header(skb_in) +
 635                                                  (iph->ihl << 2) +
 636                                                  offsetof(struct icmphdr,
 637                                                           type) -
 638                                                  skb_in->data,
 639                                                  sizeof(_inner_type),
 640                                                  &_inner_type);
 641                         if (!itp)
 642                                 goto out;
 643
 644                         /*
 645                          *      Assume any unknown ICMP type is an error. This
 646                          *      isn't specified by the RFC, but think about it..
 647                          */
 648                         if (*itp > NR_ICMP_TYPES ||
 649                             icmp_pointers[*itp].error)
 650                                 goto out;
 651                 }
 652         }
 653
 654         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 655         local_bh_disable();
 656
 657         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 658          * incoming dev is loopback.  If outgoing dev change to not be
 659          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 660          */
 661         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 662               !icmpv4_global_allow(net, type, code))
 663                 goto out_bh_enable;
 664
 665         sk = icmp_xmit_lock(net);
 666         if (!sk)
 667                 goto out_bh_enable;
 668
 669         /*
 670          *      Construct source address and options.
 671          */
 672
 673         saddr = iph->daddr;
 674         if (!(rt->rt_flags & RTCF_LOCAL)) {
 675                 struct net_device *dev = NULL;
 676
 677                 rcu_read_lock();
 678                 if (rt_is_input_route(rt) &&
 679                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 680                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 681
 682                 if (dev)
 683                         saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 684                 else
 685                         saddr = 0;
 686                 rcu_read_unlock();
 687         }
 688
 689         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 690                                            IPTOS_PREC_INTERNETCONTROL) :
 691                                           iph->tos;
 692         mark = IP4_REPLY_MARK(net, skb_in->mark);
 693
 694         if (ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in))
 695                 goto out_unlock;
 696
 697
 698         /*
 699          *      Prepare data for ICMP header.
 700          */
 701
 702         icmp_param.data.icmph.type       = type;
 703         icmp_param.data.icmph.code       = code;
 704         icmp_param.data.icmph.un.gateway = info;
 705         icmp_param.data.icmph.checksum   = 0;
 706         icmp_param.skb    = skb_in;
 707         icmp_param.offset = skb_network_offset(skb_in);
 708         inet_sk(sk)->tos = tos;
 709         sk->sk_mark = mark;
 710         ipcm_init(&ipc);
 711         ipc.addr = iph->saddr;
 712         ipc.opt = &icmp_param.replyopts.opt;
 713
 714         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 715                                type, code, &icmp_param);
 716         if (IS_ERR(rt))
 717                 goto out_unlock;
 718
 719         /* peer icmp_ratelimit */
 720         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 721                 goto ende;
 722
 723         /* RFC says return as much as we can without exceeding 576 bytes. */
 724
 725         room = dst_mtu(&rt->dst);
 726         if (room > 576)
 727                 room = 576;
 728         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 729         room -= sizeof(struct icmphdr);
 730
 731         icmp_param.data_len = skb_in->len - icmp_param.offset;
 732         if (icmp_param.data_len > room)
 733                 icmp_param.data_len = room;
 734         icmp_param.head_len = sizeof(struct icmphdr);
 735
 736         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 737 ende:
 738         ip_rt_put(rt);
 739 out_unlock:
 740         icmp_xmit_unlock(sk);
 741 out_bh_enable:
 742         local_bh_enable();
 743 out:;
 744 }
 745 EXPORT_SYMBOL(icmp_send);
 746
 747
 748 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 749 {
 750         const struct iphdr *iph = (const struct iphdr *) skb->data;
 751         const struct net_protocol *ipprot;
 752         int protocol = iph->protocol;
 753
 754         /* Checkin full IP header plus 8 bytes of protocol to
 755          * avoid additional coding at protocol handlers.
 756          */
 757         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 758                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 759                 return;
 760         }
 761
 762         raw_icmp_error(skb, protocol, info);
 763
 764         ipprot = rcu_dereference(inet_protos[protocol]);
 765         if (ipprot && ipprot->err_handler)
 766                 ipprot->err_handler(skb, info);
 767 }
 768
 769 static bool icmp_tag_validation(int proto)
 770 {
 771         bool ok;
 772
 773         rcu_read_lock();
 774         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 775         rcu_read_unlock();
 776         return ok;
 777 }
 778
 779 /*
 780  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 781  *      ICMP_PARAMETERPROB.
 782  */
 783
 784 static bool icmp_unreach(struct sk_buff *skb)
 785 {
 786         const struct iphdr *iph;
 787         struct icmphdr *icmph;
 788         struct net *net;
 789         u32 info = 0;
 790
 791         net = dev_net(skb_dst(skb)->dev);
 792
 793         /*
 794          *      Incomplete header ?
 795          *      Only checks for the IP header, there should be an
 796          *      additional check for longer headers in upper levels.
 797          */
 798
 799         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 800                 goto out_err;
 801
 802         icmph = icmp_hdr(skb);
 803         iph   = (const struct iphdr *)skb->data;
 804
 805         if (iph->ihl < 5) /* Mangled header, drop. */
 806                 goto out_err;
 807
 808         switch (icmph->type) {
 809         case ICMP_DEST_UNREACH:
 810                 switch (icmph->code & 15) {
 811                 case ICMP_NET_UNREACH:
 812                 case ICMP_HOST_UNREACH:
 813                 case ICMP_PROT_UNREACH:
 814                 case ICMP_PORT_UNREACH:
 815                         break;
 816                 case ICMP_FRAG_NEEDED:
 817                         /* for documentation of the ip_no_pmtu_disc
 818                          * values please see
 819                          * Documentation/networking/ip-sysctl.txt
 820                          */
 821                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 822                         default:
 823                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 824                                                     &iph->daddr);
 825                                 break;
 826                         case 2:
 827                                 goto out;
 828                         case 3:
 829                                 if (!icmp_tag_validation(iph->protocol))
 830                                         goto out;
 831                                 /* fall through */
 832                         case 0:
 833                                 info = ntohs(icmph->un.frag.mtu);
 834                         }
 835                         break;
 836                 case ICMP_SR_FAILED:
 837                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 838                                             &iph->daddr);
 839                         break;
 840                 default:
 841                         break;
 842                 }
 843                 if (icmph->code > NR_ICMP_UNREACH)
 844                         goto out;
 845                 break;
 846         case ICMP_PARAMETERPROB:
 847                 info = ntohl(icmph->un.gateway) >> 24;
 848                 break;
 849         case ICMP_TIME_EXCEEDED:
 850                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 851                 if (icmph->code == ICMP_EXC_FRAGTIME)
 852                         goto out;
 853                 break;
 854         }
 855
 856         /*
 857          *      Throw it at our lower layers
 858          *
 859          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 860          *                header.
 861          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 862          *                transport layer.
 863          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 864          *                transport layer.
 865          */
 866
 867         /*
 868          *      Check the other end isn't violating RFC 1122. Some routers send
 869          *      bogus responses to broadcast frames. If you see this message
 870          *      first check your netmask matches at both ends, if it does then
 871          *      get the other vendor to fix their kit.
 872          */
 873
 874         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 875             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 876                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 877                                      &ip_hdr(skb)->saddr,
 878                                      icmph->type, icmph->code,
 879                                      &iph->daddr, skb->dev->name);
 880                 goto out;
 881         }
 882
 883         icmp_socket_deliver(skb, info);
 884
 885 out:
 886         return true;
 887 out_err:
 888         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 889         return false;
 890 }
 891
 892
 893 /*
 894  *      Handle ICMP_REDIRECT.
 895  */
 896
 897 static bool icmp_redirect(struct sk_buff *skb)
 898 {
 899         if (skb->len < sizeof(struct iphdr)) {
 900                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 901                 return false;
 902         }
 903
 904         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 905                 /* there aught to be a stat */
 906                 return false;
 907         }
 908
 909         icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
 910         return true;
 911 }
 912
 913 /*
 914  *      Handle ICMP_ECHO ("ping") requests.
 915  *
 916  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 917  *                requests.
 918  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 919  *                included in the reply.
 920  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 921  *                echo requests, MUST have default=NOT.
 922  *      See also WRT handling of options once they are done and working.
 923  */
 924
 925 static bool icmp_echo(struct sk_buff *skb)
 926 {
 927         struct net *net;
 928
 929         net = dev_net(skb_dst(skb)->dev);
 930         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 931                 struct icmp_bxm icmp_param;
 932
 933                 icmp_param.data.icmph      = *icmp_hdr(skb);
 934                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 935                 icmp_param.skb             = skb;
 936                 icmp_param.offset          = 0;
 937                 icmp_param.data_len        = skb->len;
 938                 icmp_param.head_len        = sizeof(struct icmphdr);
 939                 icmp_reply(&icmp_param, skb);
 940         }
 941         /* should there be an ICMP stat for ignored echos? */
 942         return true;
 943 }
 944
 945 /*
 946  *      Handle ICMP Timestamp requests.
 947  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 948  *                SHOULD be in the kernel for minimum random latency.
 949  *                MUST be accurate to a few minutes.
 950  *                MUST be updated at least at 15Hz.
 951  */
 952 static bool icmp_timestamp(struct sk_buff *skb)
 953 {
 954         struct icmp_bxm icmp_param;
 955         /*
 956          *      Too short.
 957          */
 958         if (skb->len < 4)
 959                 goto out_err;
 960
 961         /*
 962          *      Fill in the current time as ms since midnight UT:
 963          */
 964         icmp_param.data.times[1] = inet_current_timestamp();
 965         icmp_param.data.times[2] = icmp_param.data.times[1];
 966
 967         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
 968
 969         icmp_param.data.icmph      = *icmp_hdr(skb);
 970         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 971         icmp_param.data.icmph.code = 0;
 972         icmp_param.skb             = skb;
 973         icmp_param.offset          = 0;
 974         icmp_param.data_len        = 0;
 975         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 976         icmp_reply(&icmp_param, skb);
 977         return true;
 978
 979 out_err:
 980         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 981         return false;
 982 }
 983
 984 static bool icmp_discard(struct sk_buff *skb)
 985 {
 986         /* pretend it was a success */
 987         return true;
 988 }
 989
 990 /*
 991  *      Deal with incoming ICMP packets.
 992  */
 993 int icmp_rcv(struct sk_buff *skb)
 994 {
 995         struct icmphdr *icmph;
 996         struct rtable *rt = skb_rtable(skb);
 997         struct net *net = dev_net(rt->dst.dev);
 998         bool success;
 999
1000         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1001                 struct sec_path *sp = skb_sec_path(skb);
1002                 int nh;
1003
1004                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1005                                  XFRM_STATE_ICMP))
1006                         goto drop;
1007
1008                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1009                         goto drop;
1010
1011                 nh = skb_network_offset(skb);
1012                 skb_set_network_header(skb, sizeof(*icmph));
1013
1014                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1015                         goto drop;
1016
1017                 skb_set_network_header(skb, nh);
1018         }
1019
1020         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1021
1022         if (skb_checksum_simple_validate(skb))
1023                 goto csum_error;
1024
1025         if (!pskb_pull(skb, sizeof(*icmph)))
1026                 goto error;
1027
1028         icmph = icmp_hdr(skb);
1029
1030         ICMPMSGIN_INC_STATS(net, icmph->type);
1031         /*
1032          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1033          *
1034          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1035          *                discarded.
1036          */
1037         if (icmph->type > NR_ICMP_TYPES)
1038                 goto error;
1039
1040
1041         /*
1042          *      Parse the ICMP message
1043          */
1044
1045         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1046                 /*
1047                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1048                  *        silently ignored (we let user decide with a sysctl).
1049                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1050                  *        discarded if to broadcast/multicast.
1051                  */
1052                 if ((icmph->type == ICMP_ECHO ||
1053                      icmph->type == ICMP_TIMESTAMP) &&
1054                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1055                         goto error;
1056                 }
1057                 if (icmph->type != ICMP_ECHO &&
1058                     icmph->type != ICMP_TIMESTAMP &&
1059                     icmph->type != ICMP_ADDRESS &&
1060                     icmph->type != ICMP_ADDRESSREPLY) {
1061                         goto error;
1062                 }
1063         }
1064
1065         success = icmp_pointers[icmph->type].handler(skb);
1066
1067         if (success)  {
1068                 consume_skb(skb);
1069                 return NET_RX_SUCCESS;
1070         }
1071
1072 drop:
1073         kfree_skb(skb);
1074         return NET_RX_DROP;
1075 csum_error:
1076         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1077 error:
1078         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1079         goto drop;
1080 }
1081
1082 void icmp_err(struct sk_buff *skb, u32 info)
1083 {
1084         struct iphdr *iph = (struct iphdr *)skb->data;
1085         int offset = iph->ihl<<2;
1086         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1087         int type = icmp_hdr(skb)->type;
1088         int code = icmp_hdr(skb)->code;
1089         struct net *net = dev_net(skb->dev);
1090
1091         /*
1092          * Use ping_err to handle all icmp errors except those
1093          * triggered by ICMP_ECHOREPLY which sent from kernel.
1094          */
1095         if (icmph->type != ICMP_ECHOREPLY) {
1096                 ping_err(skb, offset, info);
1097                 return;
1098         }
1099
1100         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1101                 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1102         else if (type == ICMP_REDIRECT)
1103                 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1104 }
1105
1106 /*
1107  *      This table is the definition of how we handle ICMP.
1108  */
1109 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1110         [ICMP_ECHOREPLY] = {
1111                 .handler = ping_rcv,
1112         },
1113         [1] = {
1114                 .handler = icmp_discard,
1115                 .error = 1,
1116         },
1117         [2] = {
1118                 .handler = icmp_discard,
1119                 .error = 1,
1120         },
1121         [ICMP_DEST_UNREACH] = {
1122                 .handler = icmp_unreach,
1123                 .error = 1,
1124         },
1125         [ICMP_SOURCE_QUENCH] = {
1126                 .handler = icmp_unreach,
1127                 .error = 1,
1128         },
1129         [ICMP_REDIRECT] = {
1130                 .handler = icmp_redirect,
1131                 .error = 1,
1132         },
1133         [6] = {
1134                 .handler = icmp_discard,
1135                 .error = 1,
1136         },
1137         [7] = {
1138                 .handler = icmp_discard,
1139                 .error = 1,
1140         },
1141         [ICMP_ECHO] = {
1142                 .handler = icmp_echo,
1143         },
1144         [9] = {
1145                 .handler = icmp_discard,
1146                 .error = 1,
1147         },
1148         [10] = {
1149                 .handler = icmp_discard,
1150                 .error = 1,
1151         },
1152         [ICMP_TIME_EXCEEDED] = {
1153                 .handler = icmp_unreach,
1154                 .error = 1,
1155         },
1156         [ICMP_PARAMETERPROB] = {
1157                 .handler = icmp_unreach,
1158                 .error = 1,
1159         },
1160         [ICMP_TIMESTAMP] = {
1161                 .handler = icmp_timestamp,
1162         },
1163         [ICMP_TIMESTAMPREPLY] = {
1164                 .handler = icmp_discard,
1165         },
1166         [ICMP_INFO_REQUEST] = {
1167                 .handler = icmp_discard,
1168         },
1169         [ICMP_INFO_REPLY] = {
1170                 .handler = icmp_discard,
1171         },
1172         [ICMP_ADDRESS] = {
1173                 .handler = icmp_discard,
1174         },
1175         [ICMP_ADDRESSREPLY] = {
1176                 .handler = icmp_discard,
1177         },
1178 };
1179
1180 static void __net_exit icmp_sk_exit(struct net *net)
1181 {
1182         int i;
1183
1184         for_each_possible_cpu(i)
1185                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1186         free_percpu(net->ipv4.icmp_sk);
1187         net->ipv4.icmp_sk = NULL;
1188 }
1189
1190 static int __net_init icmp_sk_init(struct net *net)
1191 {
1192         int i, err;
1193
1194         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1195         if (!net->ipv4.icmp_sk)
1196                 return -ENOMEM;
1197
1198         for_each_possible_cpu(i) {
1199                 struct sock *sk;
1200
1201                 err = inet_ctl_sock_create(&sk, PF_INET,
1202                                            SOCK_RAW, IPPROTO_ICMP, net);
1203                 if (err < 0)
1204                         goto fail;
1205
1206                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1207
1208                 /* Enough space for 2 64K ICMP packets, including
1209                  * sk_buff/skb_shared_info struct overhead.
1210                  */
1211                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1212
1213                 /*
1214                  * Speedup sock_wfree()
1215                  */
1216                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1217                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1218         }
1219
1220         /* Control parameters for ECHO replies. */
1221         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1222         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1223
1224         /* Control parameter - ignore bogus broadcast responses? */
1225         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1226
1227         /*
1228          *      Configurable global rate limit.
1229          *
1230          *      ratelimit defines tokens/packet consumed for dst->rate_token
1231          *      bucket ratemask defines which icmp types are ratelimited by
1232          *      setting it's bit position.
1233          *
1234          *      default:
1235          *      dest unreachable (3), source quench (4),
1236          *      time exceeded (11), parameter problem (12)
1237          */
1238
1239         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1240         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1241         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1242
1243         return 0;
1244
1245 fail:
1246         for_each_possible_cpu(i)
1247                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1248         free_percpu(net->ipv4.icmp_sk);
1249         return err;
1250 }
1251
1252 static struct pernet_operations __net_initdata icmp_sk_ops = {
1253        .init = icmp_sk_init,
1254        .exit = icmp_sk_exit,
1255 };
1256
1257 int __init icmp_init(void)
1258 {
1259         return register_pernet_subsys(&icmp_sk_ops);
1260 }