net/ipv4/icmp.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      NET3:   Implementation of the ICMP protocol layer.
   4  *
   5  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
   6  *
   7  *      Some of the function names and the icmp unreach table for this
   8  *      module were derived from [icmp.c 1.0.11 06/02/93] by
   9  *      Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  10  *      Other than that this module is a complete rewrite.
  11  *
  12  *      Fixes:
  13  *      Clemens Fruhwirth       :       introduce global icmp rate limiting
  14  *                                      with icmp type masking ability instead
  15  *                                      of broken per type icmp timeouts.
  16  *              Mike Shaver     :       RFC1122 checks.
  17  *              Alan Cox        :       Multicast ping reply as self.
  18  *              Alan Cox        :       Fix atomicity lockup in ip_build_xmit
  19  *                                      call.
  20  *              Alan Cox        :       Added 216,128 byte paths to the MTU
  21  *                                      code.
  22  *              Martin Mares    :       RFC1812 checks.
  23  *              Martin Mares    :       Can be configured to follow redirects
  24  *                                      if acting as a router _without_ a
  25  *                                      routing protocol (RFC 1812).
  26  *              Martin Mares    :       Echo requests may be configured to
  27  *                                      be ignored (RFC 1812).
  28  *              Martin Mares    :       Limitation of ICMP error message
  29  *                                      transmit rate (RFC 1812).
  30  *              Martin Mares    :       TOS and Precedence set correctly
  31  *                                      (RFC 1812).
  32  *              Martin Mares    :       Now copying as much data from the
  33  *                                      original packet as we can without
  34  *                                      exceeding 576 bytes (RFC 1812).
  35  *      Willy Konynenberg       :       Transparent proxying support.
  36  *              Keith Owens     :       RFC1191 correction for 4.2BSD based
  37  *                                      path MTU bug.
  38  *              Thomas Quinot   :       ICMP Dest Unreach codes up to 15 are
  39  *                                      valid (RFC 1812).
  40  *              Andi Kleen      :       Check all packet lengths properly
  41  *                                      and moved all kfree_skb() up to
  42  *                                      icmp_rcv.
  43  *              Andi Kleen      :       Move the rate limit bookkeeping
  44  *                                      into the dest entry and use a token
  45  *                                      bucket filter (thanks to ANK). Make
  46  *                                      the rates sysctl configurable.
  47  *              Yu Tianli       :       Fixed two ugly bugs in icmp_send
  48  *                                      - IP option length was accounted wrongly
  49  *                                      - ICMP header length was not accounted
  50  *                                        at all.
  51  *              Tristan Greaves :       Added sysctl option to ignore bogus
  52  *                                      broadcast responses from broken routers.
  53  *
  54  * To Fix:
  55  *
  56  *      - Should use skb_pull() instead of all the manual checking.
  57  *        This would also greatly simply some upper layer error handlers. --AK
  58  */
  59
  60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  61
  62 #include <linux/module.h>
  63 #include <linux/types.h>
  64 #include <linux/jiffies.h>
  65 #include <linux/kernel.h>
  66 #include <linux/fcntl.h>
  67 #include <linux/socket.h>
  68 #include <linux/in.h>
  69 #include <linux/inet.h>
  70 #include <linux/inetdevice.h>
  71 #include <linux/netdevice.h>
  72 #include <linux/string.h>
  73 #include <linux/netfilter_ipv4.h>
  74 #include <linux/slab.h>
  75 #include <net/snmp.h>
  76 #include <net/ip.h>
  77 #include <net/route.h>
  78 #include <net/protocol.h>
  79 #include <net/icmp.h>
  80 #include <net/tcp.h>
  81 #include <net/udp.h>
  82 #include <net/raw.h>
  83 #include <net/ping.h>
  84 #include <linux/skbuff.h>
  85 #include <net/sock.h>
  86 #include <linux/errno.h>
  87 #include <linux/timer.h>
  88 #include <linux/init.h>
  89 #include <linux/uaccess.h>
  90 #include <net/checksum.h>
  91 #include <net/xfrm.h>
  92 #include <net/inet_common.h>
  93 #include <net/ip_fib.h>
  94 #include <net/l3mdev.h>
  95
  96 /*
  97  *      Build xmit assembly blocks
  98  */
  99
 100 struct icmp_bxm {
 101         struct sk_buff *skb;
 102         int offset;
 103         int data_len;
 104
 105         struct {
 106                 struct icmphdr icmph;
 107                 __be32         times[3];
 108         } data;
 109         int head_len;
 110         struct ip_options_data replyopts;
 111 };
 112
 113 /* An array of errno for error messages from dest unreach. */
 114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 115
 116 const struct icmp_err icmp_err_convert[] = {
 117         {
 118                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 119                 .fatal = 0,
 120         },
 121         {
 122                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 123                 .fatal = 0,
 124         },
 125         {
 126                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 127                 .fatal = 1,
 128         },
 129         {
 130                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 131                 .fatal = 1,
 132         },
 133         {
 134                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 135                 .fatal = 0,
 136         },
 137         {
 138                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 139                 .fatal = 0,
 140         },
 141         {
 142                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 143                 .fatal = 1,
 144         },
 145         {
 146                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 147                 .fatal = 1,
 148         },
 149         {
 150                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 151                 .fatal = 1,
 152         },
 153         {
 154                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 155                 .fatal = 1,
 156         },
 157         {
 158                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 159                 .fatal = 1,
 160         },
 161         {
 162                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 163                 .fatal = 0,
 164         },
 165         {
 166                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 167                 .fatal = 0,
 168         },
 169         {
 170                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 171                 .fatal = 1,
 172         },
 173         {
 174                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 175                 .fatal = 1,
 176         },
 177         {
 178                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 179                 .fatal = 1,
 180         },
 181 };
 182 EXPORT_SYMBOL(icmp_err_convert);
 183
 184 /*
 185  *      ICMP control array. This specifies what to do with each ICMP.
 186  */
 187
 188 struct icmp_control {
 189         bool (*handler)(struct sk_buff *skb);
 190         short   error;          /* This ICMP is classed as an error message */
 191 };
 192
 193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 194
 195 /*
 196  *      The ICMP socket(s). This is the most convenient way to flow control
 197  *      our ICMP output as well as maintain a clean interface throughout
 198  *      all layers. All Socketless IP sends will soon be gone.
 199  *
 200  *      On SMP we have one ICMP socket per-cpu.
 201  */
 202 static struct sock *icmp_sk(struct net *net)
 203 {
 204         return this_cpu_read(*net->ipv4.icmp_sk);
 205 }
 206
 207 /* Called with BH disabled */
 208 static inline struct sock *icmp_xmit_lock(struct net *net)
 209 {
 210         struct sock *sk;
 211
 212         sk = icmp_sk(net);
 213
 214         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 215                 /* This can happen if the output path signals a
 216                  * dst_link_failure() for an outgoing ICMP packet.
 217                  */
 218                 return NULL;
 219         }
 220         return sk;
 221 }
 222
 223 static inline void icmp_xmit_unlock(struct sock *sk)
 224 {
 225         spin_unlock(&sk->sk_lock.slock);
 226 }
 227
 228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 229 int sysctl_icmp_msgs_burst __read_mostly = 50;
 230
 231 static struct {
 232         spinlock_t      lock;
 233         u32             credit;
 234         u32             stamp;
 235 } icmp_global = {
 236         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 237 };
 238
 239 /**
 240  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 241  *
 242  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 243  * Returns false if we reached the limit and can not send another packet.
 244  * Note: called with BH disabled
 245  */
 246 bool icmp_global_allow(void)
 247 {
 248         u32 credit, delta, incr = 0, now = (u32)jiffies;
 249         bool rc = false;
 250
 251         /* Check if token bucket is empty and cannot be refilled
 252          * without taking the spinlock. The READ_ONCE() are paired
 253          * with the following WRITE_ONCE() in this same function.
 254          */
 255         if (!READ_ONCE(icmp_global.credit)) {
 256                 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 257                 if (delta < HZ / 50)
 258                         return false;
 259         }
 260
 261         spin_lock(&icmp_global.lock);
 262         delta = min_t(u32, now - icmp_global.stamp, HZ);
 263         if (delta >= HZ / 50) {
 264                 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
 265                 if (incr)
 266                         WRITE_ONCE(icmp_global.stamp, now);
 267         }
 268         credit = min_t(u32, icmp_global.credit + incr,
 269                        READ_ONCE(sysctl_icmp_msgs_burst));
 270         if (credit) {
 271                 /* We want to use a credit of one in average, but need to randomize
 272                  * it for security reasons.
 273                  */
 274                 credit = max_t(int, credit - prandom_u32_max(3), 0);
 275                 rc = true;
 276         }
 277         WRITE_ONCE(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);
 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;
 384                 struct sk_buff *skb1;
 385
 386                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 387                                                  (char *)icmph,
 388                                                  icmp_param->head_len);
 389                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 390                         csum = csum_add(csum, skb1->csum);
 391                 }
 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         ipc.sockc.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_common(&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 /*
 465  * The device used for looking up which routing table to use for sending an ICMP
 466  * error is preferably the source whenever it is set, which should ensure the
 467  * icmp error can be sent to the source host, else lookup using the routing
 468  * table of the destination device, else use the main routing table (index 0).
 469  */
 470 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 471 {
 472         struct net_device *route_lookup_dev = NULL;
 473
 474         if (skb->dev)
 475                 route_lookup_dev = skb->dev;
 476         else if (skb_dst(skb))
 477                 route_lookup_dev = skb_dst(skb)->dev;
 478         return route_lookup_dev;
 479 }
 480
 481 static struct rtable *icmp_route_lookup(struct net *net,
 482                                         struct flowi4 *fl4,
 483                                         struct sk_buff *skb_in,
 484                                         const struct iphdr *iph,
 485                                         __be32 saddr, u8 tos, u32 mark,
 486                                         int type, int code,
 487                                         struct icmp_bxm *param)
 488 {
 489         struct net_device *route_lookup_dev;
 490         struct rtable *rt, *rt2;
 491         struct flowi4 fl4_dec;
 492         int err;
 493
 494         memset(fl4, 0, sizeof(*fl4));
 495         fl4->daddr = (param->replyopts.opt.opt.srr ?
 496                       param->replyopts.opt.opt.faddr : iph->saddr);
 497         fl4->saddr = saddr;
 498         fl4->flowi4_mark = mark;
 499         fl4->flowi4_uid = sock_net_uid(net, NULL);
 500         fl4->flowi4_tos = RT_TOS(tos);
 501         fl4->flowi4_proto = IPPROTO_ICMP;
 502         fl4->fl4_icmp_type = type;
 503         fl4->fl4_icmp_code = code;
 504         route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 505         fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 506
 507         security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
 508         rt = ip_route_output_key_hash(net, fl4, skb_in);
 509         if (IS_ERR(rt))
 510                 return rt;
 511
 512         /* No need to clone since we're just using its address. */
 513         rt2 = rt;
 514
 515         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 516                                            flowi4_to_flowi(fl4), NULL, 0);
 517         if (!IS_ERR(rt)) {
 518                 if (rt != rt2)
 519                         return rt;
 520         } else if (PTR_ERR(rt) == -EPERM) {
 521                 rt = NULL;
 522         } else
 523                 return rt;
 524
 525         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 526         if (err)
 527                 goto relookup_failed;
 528
 529         if (inet_addr_type_dev_table(net, route_lookup_dev,
 530                                      fl4_dec.saddr) == RTN_LOCAL) {
 531                 rt2 = __ip_route_output_key(net, &fl4_dec);
 532                 if (IS_ERR(rt2))
 533                         err = PTR_ERR(rt2);
 534         } else {
 535                 struct flowi4 fl4_2 = {};
 536                 unsigned long orefdst;
 537
 538                 fl4_2.daddr = fl4_dec.saddr;
 539                 rt2 = ip_route_output_key(net, &fl4_2);
 540                 if (IS_ERR(rt2)) {
 541                         err = PTR_ERR(rt2);
 542                         goto relookup_failed;
 543                 }
 544                 /* Ugh! */
 545                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 546                 skb_dst_set(skb_in, NULL);
 547                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 548                                      RT_TOS(tos), rt2->dst.dev);
 549
 550                 dst_release(&rt2->dst);
 551                 rt2 = skb_rtable(skb_in);
 552                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 553         }
 554
 555         if (err)
 556                 goto relookup_failed;
 557
 558         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 559                                             flowi4_to_flowi(&fl4_dec), NULL,
 560                                             XFRM_LOOKUP_ICMP);
 561         if (!IS_ERR(rt2)) {
 562                 dst_release(&rt->dst);
 563                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 564                 rt = rt2;
 565         } else if (PTR_ERR(rt2) == -EPERM) {
 566                 if (rt)
 567                         dst_release(&rt->dst);
 568                 return rt2;
 569         } else {
 570                 err = PTR_ERR(rt2);
 571                 goto relookup_failed;
 572         }
 573         return rt;
 574
 575 relookup_failed:
 576         if (rt)
 577                 return rt;
 578         return ERR_PTR(err);
 579 }
 580
 581 /*
 582  *      Send an ICMP message in response to a situation
 583  *
 584  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 585  *                MAY send more (we do).
 586  *                      MUST NOT change this header information.
 587  *                      MUST NOT reply to a multicast/broadcast IP address.
 588  *                      MUST NOT reply to a multicast/broadcast MAC address.
 589  *                      MUST reply to only the first fragment.
 590  */
 591
 592 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 593                  const struct ip_options *opt)
 594 {
 595         struct iphdr *iph;
 596         int room;
 597         struct icmp_bxm icmp_param;
 598         struct rtable *rt = skb_rtable(skb_in);
 599         struct ipcm_cookie ipc;
 600         struct flowi4 fl4;
 601         __be32 saddr;
 602         u8  tos;
 603         u32 mark;
 604         struct net *net;
 605         struct sock *sk;
 606
 607         if (!rt)
 608                 goto out;
 609
 610         if (rt->dst.dev)
 611                 net = dev_net(rt->dst.dev);
 612         else if (skb_in->dev)
 613                 net = dev_net(skb_in->dev);
 614         else
 615                 goto out;
 616
 617         /*
 618          *      Find the original header. It is expected to be valid, of course.
 619          *      Check this, icmp_send is called from the most obscure devices
 620          *      sometimes.
 621          */
 622         iph = ip_hdr(skb_in);
 623
 624         if ((u8 *)iph < skb_in->head ||
 625             (skb_network_header(skb_in) + sizeof(*iph)) >
 626             skb_tail_pointer(skb_in))
 627                 goto out;
 628
 629         /*
 630          *      No replies to physical multicast/broadcast
 631          */
 632         if (skb_in->pkt_type != PACKET_HOST)
 633                 goto out;
 634
 635         /*
 636          *      Now check at the protocol level
 637          */
 638         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 639                 goto out;
 640
 641         /*
 642          *      Only reply to fragment 0. We byte re-order the constant
 643          *      mask for efficiency.
 644          */
 645         if (iph->frag_off & htons(IP_OFFSET))
 646                 goto out;
 647
 648         /*
 649          *      If we send an ICMP error to an ICMP error a mess would result..
 650          */
 651         if (icmp_pointers[type].error) {
 652                 /*
 653                  *      We are an error, check if we are replying to an
 654                  *      ICMP error
 655                  */
 656                 if (iph->protocol == IPPROTO_ICMP) {
 657                         u8 _inner_type, *itp;
 658
 659                         itp = skb_header_pointer(skb_in,
 660                                                  skb_network_header(skb_in) +
 661                                                  (iph->ihl << 2) +
 662                                                  offsetof(struct icmphdr,
 663                                                           type) -
 664                                                  skb_in->data,
 665                                                  sizeof(_inner_type),
 666                                                  &_inner_type);
 667                         if (!itp)
 668                                 goto out;
 669
 670                         /*
 671                          *      Assume any unknown ICMP type is an error. This
 672                          *      isn't specified by the RFC, but think about it..
 673                          */
 674                         if (*itp > NR_ICMP_TYPES ||
 675                             icmp_pointers[*itp].error)
 676                                 goto out;
 677                 }
 678         }
 679
 680         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 681         local_bh_disable();
 682
 683         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 684          * incoming dev is loopback.  If outgoing dev change to not be
 685          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 686          */
 687         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 688               !icmpv4_global_allow(net, type, code))
 689                 goto out_bh_enable;
 690
 691         sk = icmp_xmit_lock(net);
 692         if (!sk)
 693                 goto out_bh_enable;
 694
 695         /*
 696          *      Construct source address and options.
 697          */
 698
 699         saddr = iph->daddr;
 700         if (!(rt->rt_flags & RTCF_LOCAL)) {
 701                 struct net_device *dev = NULL;
 702
 703                 rcu_read_lock();
 704                 if (rt_is_input_route(rt) &&
 705                     READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
 706                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 707
 708                 if (dev)
 709                         saddr = inet_select_addr(dev, iph->saddr,
 710                                                  RT_SCOPE_LINK);
 711                 else
 712                         saddr = 0;
 713                 rcu_read_unlock();
 714         }
 715
 716         tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 717                                            IPTOS_PREC_INTERNETCONTROL) :
 718                                            iph->tos;
 719         mark = IP4_REPLY_MARK(net, skb_in->mark);
 720
 721         if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 722                 goto out_unlock;
 723
 724
 725         /*
 726          *      Prepare data for ICMP header.
 727          */
 728
 729         icmp_param.data.icmph.type       = type;
 730         icmp_param.data.icmph.code       = code;
 731         icmp_param.data.icmph.un.gateway = info;
 732         icmp_param.data.icmph.checksum   = 0;
 733         icmp_param.skb    = skb_in;
 734         icmp_param.offset = skb_network_offset(skb_in);
 735         inet_sk(sk)->tos = tos;
 736         ipcm_init(&ipc);
 737         ipc.addr = iph->saddr;
 738         ipc.opt = &icmp_param.replyopts.opt;
 739         ipc.sockc.mark = mark;
 740
 741         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 742                                type, code, &icmp_param);
 743         if (IS_ERR(rt))
 744                 goto out_unlock;
 745
 746         /* peer icmp_ratelimit */
 747         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 748                 goto ende;
 749
 750         /* RFC says return as much as we can without exceeding 576 bytes. */
 751
 752         room = dst_mtu(&rt->dst);
 753         if (room > 576)
 754                 room = 576;
 755         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 756         room -= sizeof(struct icmphdr);
 757
 758         icmp_param.data_len = skb_in->len - icmp_param.offset;
 759         if (icmp_param.data_len > room)
 760                 icmp_param.data_len = room;
 761         icmp_param.head_len = sizeof(struct icmphdr);
 762
 763         /* if we don't have a source address at this point, fall back to the
 764          * dummy address instead of sending out a packet with a source address
 765          * of 0.0.0.0
 766          */
 767         if (!fl4.saddr)
 768                 fl4.saddr = htonl(INADDR_DUMMY);
 769
 770         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 771 ende:
 772         ip_rt_put(rt);
 773 out_unlock:
 774         icmp_xmit_unlock(sk);
 775 out_bh_enable:
 776         local_bh_enable();
 777 out:;
 778 }
 779 EXPORT_SYMBOL(__icmp_send);
 780
 781 #if IS_ENABLED(CONFIG_NF_NAT)
 782 #include <net/netfilter/nf_conntrack.h>
 783 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 784 {
 785         struct sk_buff *cloned_skb = NULL;
 786         struct ip_options opts = { 0 };
 787         enum ip_conntrack_info ctinfo;
 788         struct nf_conn *ct;
 789         __be32 orig_ip;
 790
 791         ct = nf_ct_get(skb_in, &ctinfo);
 792         if (!ct || !(ct->status & IPS_SRC_NAT)) {
 793                 __icmp_send(skb_in, type, code, info, &opts);
 794                 return;
 795         }
 796
 797         if (skb_shared(skb_in))
 798                 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 799
 800         if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 801             (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 802             skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 803             skb_network_offset(skb_in) + sizeof(struct iphdr))))
 804                 goto out;
 805
 806         orig_ip = ip_hdr(skb_in)->saddr;
 807         ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 808         __icmp_send(skb_in, type, code, info, &opts);
 809         ip_hdr(skb_in)->saddr = orig_ip;
 810 out:
 811         consume_skb(cloned_skb);
 812 }
 813 EXPORT_SYMBOL(icmp_ndo_send);
 814 #endif
 815
 816 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 817 {
 818         const struct iphdr *iph = (const struct iphdr *)skb->data;
 819         const struct net_protocol *ipprot;
 820         int protocol = iph->protocol;
 821
 822         /* Checkin full IP header plus 8 bytes of protocol to
 823          * avoid additional coding at protocol handlers.
 824          */
 825         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 826                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 827                 return;
 828         }
 829
 830         raw_icmp_error(skb, protocol, info);
 831
 832         ipprot = rcu_dereference(inet_protos[protocol]);
 833         if (ipprot && ipprot->err_handler)
 834                 ipprot->err_handler(skb, info);
 835 }
 836
 837 static bool icmp_tag_validation(int proto)
 838 {
 839         bool ok;
 840
 841         rcu_read_lock();
 842         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 843         rcu_read_unlock();
 844         return ok;
 845 }
 846
 847 /*
 848  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 849  *      ICMP_PARAMETERPROB.
 850  */
 851
 852 static bool icmp_unreach(struct sk_buff *skb)
 853 {
 854         const struct iphdr *iph;
 855         struct icmphdr *icmph;
 856         struct net *net;
 857         u32 info = 0;
 858
 859         net = dev_net(skb_dst(skb)->dev);
 860
 861         /*
 862          *      Incomplete header ?
 863          *      Only checks for the IP header, there should be an
 864          *      additional check for longer headers in upper levels.
 865          */
 866
 867         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 868                 goto out_err;
 869
 870         icmph = icmp_hdr(skb);
 871         iph   = (const struct iphdr *)skb->data;
 872
 873         if (iph->ihl < 5) /* Mangled header, drop. */
 874                 goto out_err;
 875
 876         switch (icmph->type) {
 877         case ICMP_DEST_UNREACH:
 878                 switch (icmph->code & 15) {
 879                 case ICMP_NET_UNREACH:
 880                 case ICMP_HOST_UNREACH:
 881                 case ICMP_PROT_UNREACH:
 882                 case ICMP_PORT_UNREACH:
 883                         break;
 884                 case ICMP_FRAG_NEEDED:
 885                         /* for documentation of the ip_no_pmtu_disc
 886                          * values please see
 887                          * Documentation/networking/ip-sysctl.rst
 888                          */
 889                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 890                         default:
 891                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 892                                                     &iph->daddr);
 893                                 break;
 894                         case 2:
 895                                 goto out;
 896                         case 3:
 897                                 if (!icmp_tag_validation(iph->protocol))
 898                                         goto out;
 899                                 fallthrough;
 900                         case 0:
 901                                 info = ntohs(icmph->un.frag.mtu);
 902                         }
 903                         break;
 904                 case ICMP_SR_FAILED:
 905                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 906                                             &iph->daddr);
 907                         break;
 908                 default:
 909                         break;
 910                 }
 911                 if (icmph->code > NR_ICMP_UNREACH)
 912                         goto out;
 913                 break;
 914         case ICMP_PARAMETERPROB:
 915                 info = ntohl(icmph->un.gateway) >> 24;
 916                 break;
 917         case ICMP_TIME_EXCEEDED:
 918                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 919                 if (icmph->code == ICMP_EXC_FRAGTIME)
 920                         goto out;
 921                 break;
 922         }
 923
 924         /*
 925          *      Throw it at our lower layers
 926          *
 927          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 928          *                header.
 929          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 930          *                transport layer.
 931          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 932          *                transport layer.
 933          */
 934
 935         /*
 936          *      Check the other end isn't violating RFC 1122. Some routers send
 937          *      bogus responses to broadcast frames. If you see this message
 938          *      first check your netmask matches at both ends, if it does then
 939          *      get the other vendor to fix their kit.
 940          */
 941
 942         if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
 943             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 944                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 945                                      &ip_hdr(skb)->saddr,
 946                                      icmph->type, icmph->code,
 947                                      &iph->daddr, skb->dev->name);
 948                 goto out;
 949         }
 950
 951         icmp_socket_deliver(skb, info);
 952
 953 out:
 954         return true;
 955 out_err:
 956         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 957         return false;
 958 }
 959
 960
 961 /*
 962  *      Handle ICMP_REDIRECT.
 963  */
 964
 965 static bool icmp_redirect(struct sk_buff *skb)
 966 {
 967         if (skb->len < sizeof(struct iphdr)) {
 968                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 969                 return false;
 970         }
 971
 972         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 973                 /* there aught to be a stat */
 974                 return false;
 975         }
 976
 977         icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 978         return true;
 979 }
 980
 981 /*
 982  *      Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
 983  *
 984  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 985  *                requests.
 986  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 987  *                included in the reply.
 988  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 989  *                echo requests, MUST have default=NOT.
 990  *      RFC 8335: 8 MUST have a config option to enable/disable ICMP
 991  *                Extended Echo Functionality, MUST be disabled by default
 992  *      See also WRT handling of options once they are done and working.
 993  */
 994
 995 static bool icmp_echo(struct sk_buff *skb)
 996 {
 997         struct icmp_bxm icmp_param;
 998         struct net *net;
 999
1000         net = dev_net(skb_dst(skb)->dev);
1001         /* should there be an ICMP stat for ignored echos? */
1002         if (net->ipv4.sysctl_icmp_echo_ignore_all)
1003                 return true;
1004
1005         icmp_param.data.icmph      = *icmp_hdr(skb);
1006         icmp_param.skb             = skb;
1007         icmp_param.offset          = 0;
1008         icmp_param.data_len        = skb->len;
1009         icmp_param.head_len        = sizeof(struct icmphdr);
1010
1011         if (icmp_param.data.icmph.type == ICMP_ECHO)
1012                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1013         else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1014                 return true;
1015
1016         icmp_reply(&icmp_param, skb);
1017         return true;
1018 }
1019
1020 /*      Helper for icmp_echo and icmpv6_echo_reply.
1021  *      Searches for net_device that matches PROBE interface identifier
1022  *              and builds PROBE reply message in icmphdr.
1023  *
1024  *      Returns false if PROBE responses are disabled via sysctl
1025  */
1026
1027 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1028 {
1029         struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1030         struct icmp_ext_echo_iio *iio, _iio;
1031         struct net *net = dev_net(skb->dev);
1032         struct net_device *dev;
1033         char buff[IFNAMSIZ];
1034         u16 ident_len;
1035         u8 status;
1036
1037         if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1038                 return false;
1039
1040         /* We currently only support probing interfaces on the proxy node
1041          * Check to ensure L-bit is set
1042          */
1043         if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1044                 return false;
1045         /* Clear status bits in reply message */
1046         icmphdr->un.echo.sequence &= htons(0xFF00);
1047         if (icmphdr->type == ICMP_EXT_ECHO)
1048                 icmphdr->type = ICMP_EXT_ECHOREPLY;
1049         else
1050                 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1051         ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1052         /* Size of iio is class_type dependent.
1053          * Only check header here and assign length based on ctype in the switch statement
1054          */
1055         iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1056         if (!ext_hdr || !iio)
1057                 goto send_mal_query;
1058         if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1059             ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1060                 goto send_mal_query;
1061         ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1062         iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1063                                  sizeof(iio->extobj_hdr) + ident_len, &_iio);
1064         if (!iio)
1065                 goto send_mal_query;
1066
1067         status = 0;
1068         dev = NULL;
1069         switch (iio->extobj_hdr.class_type) {
1070         case ICMP_EXT_ECHO_CTYPE_NAME:
1071                 if (ident_len >= IFNAMSIZ)
1072                         goto send_mal_query;
1073                 memset(buff, 0, sizeof(buff));
1074                 memcpy(buff, &iio->ident.name, ident_len);
1075                 dev = dev_get_by_name(net, buff);
1076                 break;
1077         case ICMP_EXT_ECHO_CTYPE_INDEX:
1078                 if (ident_len != sizeof(iio->ident.ifindex))
1079                         goto send_mal_query;
1080                 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1081                 break;
1082         case ICMP_EXT_ECHO_CTYPE_ADDR:
1083                 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1084                     ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1085                                  iio->ident.addr.ctype3_hdr.addrlen)
1086                         goto send_mal_query;
1087                 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1088                 case ICMP_AFI_IP:
1089                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1090                                 goto send_mal_query;
1091                         dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1092                         break;
1093 #if IS_ENABLED(CONFIG_IPV6)
1094                 case ICMP_AFI_IP6:
1095                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1096                                 goto send_mal_query;
1097                         dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1098                         dev_hold(dev);
1099                         break;
1100 #endif
1101                 default:
1102                         goto send_mal_query;
1103                 }
1104                 break;
1105         default:
1106                 goto send_mal_query;
1107         }
1108         if (!dev) {
1109                 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1110                 return true;
1111         }
1112         /* Fill bits in reply message */
1113         if (dev->flags & IFF_UP)
1114                 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1115         if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1116                 status |= ICMP_EXT_ECHOREPLY_IPV4;
1117         if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1118                 status |= ICMP_EXT_ECHOREPLY_IPV6;
1119         dev_put(dev);
1120         icmphdr->un.echo.sequence |= htons(status);
1121         return true;
1122 send_mal_query:
1123         icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1124         return true;
1125 }
1126 EXPORT_SYMBOL_GPL(icmp_build_probe);
1127
1128 /*
1129  *      Handle ICMP Timestamp requests.
1130  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1131  *                SHOULD be in the kernel for minimum random latency.
1132  *                MUST be accurate to a few minutes.
1133  *                MUST be updated at least at 15Hz.
1134  */
1135 static bool icmp_timestamp(struct sk_buff *skb)
1136 {
1137         struct icmp_bxm icmp_param;
1138         /*
1139          *      Too short.
1140          */
1141         if (skb->len < 4)
1142                 goto out_err;
1143
1144         /*
1145          *      Fill in the current time as ms since midnight UT:
1146          */
1147         icmp_param.data.times[1] = inet_current_timestamp();
1148         icmp_param.data.times[2] = icmp_param.data.times[1];
1149
1150         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1151
1152         icmp_param.data.icmph      = *icmp_hdr(skb);
1153         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1154         icmp_param.data.icmph.code = 0;
1155         icmp_param.skb             = skb;
1156         icmp_param.offset          = 0;
1157         icmp_param.data_len        = 0;
1158         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
1159         icmp_reply(&icmp_param, skb);
1160         return true;
1161
1162 out_err:
1163         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1164         return false;
1165 }
1166
1167 static bool icmp_discard(struct sk_buff *skb)
1168 {
1169         /* pretend it was a success */
1170         return true;
1171 }
1172
1173 /*
1174  *      Deal with incoming ICMP packets.
1175  */
1176 int icmp_rcv(struct sk_buff *skb)
1177 {
1178         struct icmphdr *icmph;
1179         struct rtable *rt = skb_rtable(skb);
1180         struct net *net = dev_net(rt->dst.dev);
1181         bool success;
1182
1183         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1184                 struct sec_path *sp = skb_sec_path(skb);
1185                 int nh;
1186
1187                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1188                                  XFRM_STATE_ICMP))
1189                         goto drop;
1190
1191                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1192                         goto drop;
1193
1194                 nh = skb_network_offset(skb);
1195                 skb_set_network_header(skb, sizeof(*icmph));
1196
1197                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1198                         goto drop;
1199
1200                 skb_set_network_header(skb, nh);
1201         }
1202
1203         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1204
1205         if (skb_checksum_simple_validate(skb))
1206                 goto csum_error;
1207
1208         if (!pskb_pull(skb, sizeof(*icmph)))
1209                 goto error;
1210
1211         icmph = icmp_hdr(skb);
1212
1213         ICMPMSGIN_INC_STATS(net, icmph->type);
1214
1215         /* Check for ICMP Extended Echo (PROBE) messages */
1216         if (icmph->type == ICMP_EXT_ECHO) {
1217                 /* We can't use icmp_pointers[].handler() because it is an array of
1218                  * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1219                  */
1220                 success = icmp_echo(skb);
1221                 goto success_check;
1222         }
1223
1224         if (icmph->type == ICMP_EXT_ECHOREPLY) {
1225                 success = ping_rcv(skb);
1226                 goto success_check;
1227         }
1228
1229         /*
1230          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1231          *
1232          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1233          *                discarded.
1234          */
1235         if (icmph->type > NR_ICMP_TYPES)
1236                 goto error;
1237
1238         /*
1239          *      Parse the ICMP message
1240          */
1241
1242         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1243                 /*
1244                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1245                  *        silently ignored (we let user decide with a sysctl).
1246                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1247                  *        discarded if to broadcast/multicast.
1248                  */
1249                 if ((icmph->type == ICMP_ECHO ||
1250                      icmph->type == ICMP_TIMESTAMP) &&
1251                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1252                         goto error;
1253                 }
1254                 if (icmph->type != ICMP_ECHO &&
1255                     icmph->type != ICMP_TIMESTAMP &&
1256                     icmph->type != ICMP_ADDRESS &&
1257                     icmph->type != ICMP_ADDRESSREPLY) {
1258                         goto error;
1259                 }
1260         }
1261
1262         success = icmp_pointers[icmph->type].handler(skb);
1263 success_check:
1264         if (success)  {
1265                 consume_skb(skb);
1266                 return NET_RX_SUCCESS;
1267         }
1268
1269 drop:
1270         kfree_skb(skb);
1271         return NET_RX_DROP;
1272 csum_error:
1273         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1274 error:
1275         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1276         goto drop;
1277 }
1278
1279 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1280 {
1281         struct icmp_extobj_hdr *objh, _objh;
1282         struct icmp_ext_hdr *exth, _exth;
1283         u16 olen;
1284
1285         exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1286         if (!exth)
1287                 return false;
1288         if (exth->version != 2)
1289                 return true;
1290
1291         if (exth->checksum &&
1292             csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1293                 return false;
1294
1295         off += sizeof(_exth);
1296         while (off < skb->len) {
1297                 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1298                 if (!objh)
1299                         return false;
1300
1301                 olen = ntohs(objh->length);
1302                 if (olen < sizeof(_objh))
1303                         return false;
1304
1305                 off += olen;
1306                 if (off > skb->len)
1307                         return false;
1308         }
1309
1310         return true;
1311 }
1312
1313 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1314                            struct sock_ee_data_rfc4884 *out,
1315                            int thlen, int off)
1316 {
1317         int hlen;
1318
1319         /* original datagram headers: end of icmph to payload (skb->data) */
1320         hlen = -skb_transport_offset(skb) - thlen;
1321
1322         /* per rfc 4884: minimal datagram length of 128 bytes */
1323         if (off < 128 || off < hlen)
1324                 return;
1325
1326         /* kernel has stripped headers: return payload offset in bytes */
1327         off -= hlen;
1328         if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1329                 return;
1330
1331         out->len = off;
1332
1333         if (!ip_icmp_error_rfc4884_validate(skb, off))
1334                 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1335 }
1336 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1337
1338 int icmp_err(struct sk_buff *skb, u32 info)
1339 {
1340         struct iphdr *iph = (struct iphdr *)skb->data;
1341         int offset = iph->ihl<<2;
1342         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1343         int type = icmp_hdr(skb)->type;
1344         int code = icmp_hdr(skb)->code;
1345         struct net *net = dev_net(skb->dev);
1346
1347         /*
1348          * Use ping_err to handle all icmp errors except those
1349          * triggered by ICMP_ECHOREPLY which sent from kernel.
1350          */
1351         if (icmph->type != ICMP_ECHOREPLY) {
1352                 ping_err(skb, offset, info);
1353                 return 0;
1354         }
1355
1356         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1357                 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1358         else if (type == ICMP_REDIRECT)
1359                 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1360
1361         return 0;
1362 }
1363
1364 /*
1365  *      This table is the definition of how we handle ICMP.
1366  */
1367 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1368         [ICMP_ECHOREPLY] = {
1369                 .handler = ping_rcv,
1370         },
1371         [1] = {
1372                 .handler = icmp_discard,
1373                 .error = 1,
1374         },
1375         [2] = {
1376                 .handler = icmp_discard,
1377                 .error = 1,
1378         },
1379         [ICMP_DEST_UNREACH] = {
1380                 .handler = icmp_unreach,
1381                 .error = 1,
1382         },
1383         [ICMP_SOURCE_QUENCH] = {
1384                 .handler = icmp_unreach,
1385                 .error = 1,
1386         },
1387         [ICMP_REDIRECT] = {
1388                 .handler = icmp_redirect,
1389                 .error = 1,
1390         },
1391         [6] = {
1392                 .handler = icmp_discard,
1393                 .error = 1,
1394         },
1395         [7] = {
1396                 .handler = icmp_discard,
1397                 .error = 1,
1398         },
1399         [ICMP_ECHO] = {
1400                 .handler = icmp_echo,
1401         },
1402         [9] = {
1403                 .handler = icmp_discard,
1404                 .error = 1,
1405         },
1406         [10] = {
1407                 .handler = icmp_discard,
1408                 .error = 1,
1409         },
1410         [ICMP_TIME_EXCEEDED] = {
1411                 .handler = icmp_unreach,
1412                 .error = 1,
1413         },
1414         [ICMP_PARAMETERPROB] = {
1415                 .handler = icmp_unreach,
1416                 .error = 1,
1417         },
1418         [ICMP_TIMESTAMP] = {
1419                 .handler = icmp_timestamp,
1420         },
1421         [ICMP_TIMESTAMPREPLY] = {
1422                 .handler = icmp_discard,
1423         },
1424         [ICMP_INFO_REQUEST] = {
1425                 .handler = icmp_discard,
1426         },
1427         [ICMP_INFO_REPLY] = {
1428                 .handler = icmp_discard,
1429         },
1430         [ICMP_ADDRESS] = {
1431                 .handler = icmp_discard,
1432         },
1433         [ICMP_ADDRESSREPLY] = {
1434                 .handler = icmp_discard,
1435         },
1436 };
1437
1438 static void __net_exit icmp_sk_exit(struct net *net)
1439 {
1440         int i;
1441
1442         for_each_possible_cpu(i)
1443                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1444         free_percpu(net->ipv4.icmp_sk);
1445         net->ipv4.icmp_sk = NULL;
1446 }
1447
1448 static int __net_init icmp_sk_init(struct net *net)
1449 {
1450         int i, err;
1451
1452         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1453         if (!net->ipv4.icmp_sk)
1454                 return -ENOMEM;
1455
1456         for_each_possible_cpu(i) {
1457                 struct sock *sk;
1458
1459                 err = inet_ctl_sock_create(&sk, PF_INET,
1460                                            SOCK_RAW, IPPROTO_ICMP, net);
1461                 if (err < 0)
1462                         goto fail;
1463
1464                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1465
1466                 /* Enough space for 2 64K ICMP packets, including
1467                  * sk_buff/skb_shared_info struct overhead.
1468                  */
1469                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1470
1471                 /*
1472                  * Speedup sock_wfree()
1473                  */
1474                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1475                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1476         }
1477
1478         /* Control parameters for ECHO replies. */
1479         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1480         net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1481         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1482
1483         /* Control parameter - ignore bogus broadcast responses? */
1484         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1485
1486         /*
1487          *      Configurable global rate limit.
1488          *
1489          *      ratelimit defines tokens/packet consumed for dst->rate_token
1490          *      bucket ratemask defines which icmp types are ratelimited by
1491          *      setting it's bit position.
1492          *
1493          *      default:
1494          *      dest unreachable (3), source quench (4),
1495          *      time exceeded (11), parameter problem (12)
1496          */
1497
1498         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1499         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1500         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1501
1502         return 0;
1503
1504 fail:
1505         icmp_sk_exit(net);
1506         return err;
1507 }
1508
1509 static struct pernet_operations __net_initdata icmp_sk_ops = {
1510        .init = icmp_sk_init,
1511        .exit = icmp_sk_exit,
1512 };
1513
1514 int __init icmp_init(void)
1515 {
1516         return register_pernet_subsys(&icmp_sk_ops);
1517 }