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, <[email protected]>
   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 #include <net/addrconf.h>
  96 #include <net/inet_dscp.h>
  97 #define CREATE_TRACE_POINTS
  98 #include <trace/events/icmp.h>
  99
 100 /*
 101  *      Build xmit assembly blocks
 102  */
 103
 104 struct icmp_bxm {
 105         struct sk_buff *skb;
 106         int offset;
 107         int data_len;
 108
 109         struct {
 110                 struct icmphdr icmph;
 111                 __be32         times[3];
 112         } data;
 113         int head_len;
 114         struct ip_options_data replyopts;
 115 };
 116
 117 /* An array of errno for error messages from dest unreach. */
 118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 119
 120 const struct icmp_err icmp_err_convert[] = {
 121         {
 122                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 123                 .fatal = 0,
 124         },
 125         {
 126                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 127                 .fatal = 0,
 128         },
 129         {
 130                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 131                 .fatal = 1,
 132         },
 133         {
 134                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 135                 .fatal = 1,
 136         },
 137         {
 138                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 139                 .fatal = 0,
 140         },
 141         {
 142                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 143                 .fatal = 0,
 144         },
 145         {
 146                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 147                 .fatal = 1,
 148         },
 149         {
 150                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 151                 .fatal = 1,
 152         },
 153         {
 154                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 155                 .fatal = 1,
 156         },
 157         {
 158                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 159                 .fatal = 1,
 160         },
 161         {
 162                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 163                 .fatal = 1,
 164         },
 165         {
 166                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 167                 .fatal = 0,
 168         },
 169         {
 170                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 171                 .fatal = 0,
 172         },
 173         {
 174                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 175                 .fatal = 1,
 176         },
 177         {
 178                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 179                 .fatal = 1,
 180         },
 181         {
 182                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 183                 .fatal = 1,
 184         },
 185 };
 186 EXPORT_SYMBOL(icmp_err_convert);
 187
 188 /*
 189  *      ICMP control array. This specifies what to do with each ICMP.
 190  */
 191
 192 struct icmp_control {
 193         enum skb_drop_reason (*handler)(struct sk_buff *skb);
 194         short   error;          /* This ICMP is classed as an error message */
 195 };
 196
 197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 198
 199 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
 200
 201 /* Called with BH disabled */
 202 static inline struct sock *icmp_xmit_lock(struct net *net)
 203 {
 204         struct sock *sk;
 205
 206         sk = this_cpu_read(ipv4_icmp_sk);
 207
 208         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 209                 /* This can happen if the output path signals a
 210                  * dst_link_failure() for an outgoing ICMP packet.
 211                  */
 212                 return NULL;
 213         }
 214         sock_net_set(sk, net);
 215         return sk;
 216 }
 217
 218 static inline void icmp_xmit_unlock(struct sock *sk)
 219 {
 220         sock_net_set(sk, &init_net);
 221         spin_unlock(&sk->sk_lock.slock);
 222 }
 223
 224 /**
 225  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 226  * @net: network namespace
 227  *
 228  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 229  * Returns false if we reached the limit and can not send another packet.
 230  * Works in tandem with icmp_global_consume().
 231  */
 232 bool icmp_global_allow(struct net *net)
 233 {
 234         u32 delta, now, oldstamp;
 235         int incr, new, old;
 236
 237         /* Note: many cpus could find this condition true.
 238          * Then later icmp_global_consume() could consume more credits,
 239          * this is an acceptable race.
 240          */
 241         if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
 242                 return true;
 243
 244         now = jiffies;
 245         oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
 246         delta = min_t(u32, now - oldstamp, HZ);
 247         if (delta < HZ / 50)
 248                 return false;
 249
 250         incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ;
 251         if (!incr)
 252                 return false;
 253
 254         if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
 255                 old = atomic_read(&net->ipv4.icmp_global_credit);
 256                 do {
 257                         new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
 258                 } while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
 259         }
 260         return true;
 261 }
 262 EXPORT_SYMBOL(icmp_global_allow);
 263
 264 void icmp_global_consume(struct net *net)
 265 {
 266         int credits = get_random_u32_below(3);
 267
 268         /* Note: this might make icmp_global.credit negative. */
 269         if (credits)
 270                 atomic_sub(credits, &net->ipv4.icmp_global_credit);
 271 }
 272 EXPORT_SYMBOL(icmp_global_consume);
 273
 274 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 275 {
 276         if (type > NR_ICMP_TYPES)
 277                 return true;
 278
 279         /* Don't limit PMTU discovery. */
 280         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 281                 return true;
 282
 283         /* Limit if icmp type is enabled in ratemask. */
 284         if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
 285                 return true;
 286
 287         return false;
 288 }
 289
 290 static bool icmpv4_global_allow(struct net *net, int type, int code,
 291                                 bool *apply_ratelimit)
 292 {
 293         if (icmpv4_mask_allow(net, type, code))
 294                 return true;
 295
 296         if (icmp_global_allow(net)) {
 297                 *apply_ratelimit = true;
 298                 return true;
 299         }
 300         __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
 301         return false;
 302 }
 303
 304 /*
 305  *      Send an ICMP frame.
 306  */
 307
 308 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 309                                struct flowi4 *fl4, int type, int code,
 310                                bool apply_ratelimit)
 311 {
 312         struct dst_entry *dst = &rt->dst;
 313         struct inet_peer *peer;
 314         bool rc = true;
 315         int vif;
 316
 317         if (!apply_ratelimit)
 318                 return true;
 319
 320         /* No rate limit on loopback */
 321         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 322                 goto out;
 323
 324         vif = l3mdev_master_ifindex(dst->dev);
 325         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 326         rc = inet_peer_xrlim_allow(peer,
 327                                    READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
 328         if (peer)
 329                 inet_putpeer(peer);
 330 out:
 331         if (!rc)
 332                 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
 333         else
 334                 icmp_global_consume(net);
 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 = 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 sock *sk,
 368                             struct icmp_bxm *icmp_param,
 369                             struct flowi4 *fl4,
 370                             struct ipcm_cookie *ipc, struct rtable **rt)
 371 {
 372         struct sk_buff *skb;
 373
 374         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 375                            icmp_param->data_len+icmp_param->head_len,
 376                            icmp_param->head_len,
 377                            ipc, rt, MSG_DONTWAIT) < 0) {
 378                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 379                 ip_flush_pending_frames(sk);
 380         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 381                 struct icmphdr *icmph = icmp_hdr(skb);
 382                 __wsum csum;
 383                 struct sk_buff *skb1;
 384
 385                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 386                                                  (char *)icmph,
 387                                                  icmp_param->head_len);
 388                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 389                         csum = csum_add(csum, skb1->csum);
 390                 }
 391                 icmph->checksum = csum_fold(csum);
 392                 skb->ip_summed = CHECKSUM_NONE;
 393                 ip_push_pending_frames(sk, fl4);
 394         }
 395 }
 396
 397 /*
 398  *      Driving logic for building and sending ICMP messages.
 399  */
 400
 401 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 402 {
 403         struct ipcm_cookie ipc;
 404         struct rtable *rt = skb_rtable(skb);
 405         struct net *net = dev_net(rt->dst.dev);
 406         bool apply_ratelimit = false;
 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 icmpv4_global_allow and icmp_xmit_lock */
 419         local_bh_disable();
 420
 421         /* is global icmp_msgs_per_sec exhausted ? */
 422         if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
 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 = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb)));
 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, apply_ratelimit))
 456                 icmp_push_reply(sk, 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, struct flowi4 *fl4,
 482                                         struct sk_buff *skb_in,
 483                                         const struct iphdr *iph, __be32 saddr,
 484                                         dscp_t dscp, u32 mark, int type,
 485                                         int code, struct icmp_bxm *param)
 486 {
 487         struct net_device *route_lookup_dev;
 488         struct dst_entry *dst, *dst2;
 489         struct rtable *rt, *rt2;
 490         struct flowi4 fl4_dec;
 491         int err;
 492
 493         memset(fl4, 0, sizeof(*fl4));
 494         fl4->daddr = (param->replyopts.opt.opt.srr ?
 495                       param->replyopts.opt.opt.faddr : iph->saddr);
 496         fl4->saddr = saddr;
 497         fl4->flowi4_mark = mark;
 498         fl4->flowi4_uid = sock_net_uid(net, NULL);
 499         fl4->flowi4_tos = inet_dscp_to_dsfield(dscp);
 500         fl4->flowi4_proto = IPPROTO_ICMP;
 501         fl4->fl4_icmp_type = type;
 502         fl4->fl4_icmp_code = code;
 503         route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 504         fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 505
 506         security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
 507         rt = ip_route_output_key_hash(net, fl4, skb_in);
 508         if (IS_ERR(rt))
 509                 return rt;
 510
 511         /* No need to clone since we're just using its address. */
 512         rt2 = rt;
 513
 514         dst = xfrm_lookup(net, &rt->dst,
 515                           flowi4_to_flowi(fl4), NULL, 0);
 516         rt = dst_rtable(dst);
 517         if (!IS_ERR(dst)) {
 518                 if (rt != rt2)
 519                         return rt;
 520                 if (inet_addr_type_dev_table(net, route_lookup_dev,
 521                                              fl4->daddr) == RTN_LOCAL)
 522                         return rt;
 523         } else if (PTR_ERR(dst) == -EPERM) {
 524                 rt = NULL;
 525         } else {
 526                 return rt;
 527         }
 528         err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 529         if (err)
 530                 goto relookup_failed;
 531
 532         if (inet_addr_type_dev_table(net, route_lookup_dev,
 533                                      fl4_dec.saddr) == RTN_LOCAL) {
 534                 rt2 = __ip_route_output_key(net, &fl4_dec);
 535                 if (IS_ERR(rt2))
 536                         err = PTR_ERR(rt2);
 537         } else {
 538                 struct flowi4 fl4_2 = {};
 539                 unsigned long orefdst;
 540
 541                 fl4_2.daddr = fl4_dec.saddr;
 542                 rt2 = ip_route_output_key(net, &fl4_2);
 543                 if (IS_ERR(rt2)) {
 544                         err = PTR_ERR(rt2);
 545                         goto relookup_failed;
 546                 }
 547                 /* Ugh! */
 548                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 549                 skb_dst_set(skb_in, NULL);
 550                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 551                                      dscp, rt2->dst.dev) ? -EINVAL : 0;
 552
 553                 dst_release(&rt2->dst);
 554                 rt2 = skb_rtable(skb_in);
 555                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 556         }
 557
 558         if (err)
 559                 goto relookup_failed;
 560
 561         dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
 562                            XFRM_LOOKUP_ICMP);
 563         rt2 = dst_rtable(dst2);
 564         if (!IS_ERR(dst2)) {
 565                 dst_release(&rt->dst);
 566                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 567                 rt = rt2;
 568         } else if (PTR_ERR(dst2) == -EPERM) {
 569                 if (rt)
 570                         dst_release(&rt->dst);
 571                 return rt2;
 572         } else {
 573                 err = PTR_ERR(dst2);
 574                 goto relookup_failed;
 575         }
 576         return rt;
 577
 578 relookup_failed:
 579         if (rt)
 580                 return rt;
 581         return ERR_PTR(err);
 582 }
 583
 584 /*
 585  *      Send an ICMP message in response to a situation
 586  *
 587  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 588  *                MAY send more (we do).
 589  *                      MUST NOT change this header information.
 590  *                      MUST NOT reply to a multicast/broadcast IP address.
 591  *                      MUST NOT reply to a multicast/broadcast MAC address.
 592  *                      MUST reply to only the first fragment.
 593  */
 594
 595 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 596                  const struct ip_options *opt)
 597 {
 598         struct iphdr *iph;
 599         int room;
 600         struct icmp_bxm icmp_param;
 601         struct rtable *rt = skb_rtable(skb_in);
 602         bool apply_ratelimit = false;
 603         struct ipcm_cookie ipc;
 604         struct flowi4 fl4;
 605         __be32 saddr;
 606         u8  tos;
 607         u32 mark;
 608         struct net *net;
 609         struct sock *sk;
 610
 611         if (!rt)
 612                 goto out;
 613
 614         if (rt->dst.dev)
 615                 net = dev_net(rt->dst.dev);
 616         else if (skb_in->dev)
 617                 net = dev_net(skb_in->dev);
 618         else
 619                 goto out;
 620
 621         /*
 622          *      Find the original header. It is expected to be valid, of course.
 623          *      Check this, icmp_send is called from the most obscure devices
 624          *      sometimes.
 625          */
 626         iph = ip_hdr(skb_in);
 627
 628         if ((u8 *)iph < skb_in->head ||
 629             (skb_network_header(skb_in) + sizeof(*iph)) >
 630             skb_tail_pointer(skb_in))
 631                 goto out;
 632
 633         /*
 634          *      No replies to physical multicast/broadcast
 635          */
 636         if (skb_in->pkt_type != PACKET_HOST)
 637                 goto out;
 638
 639         /*
 640          *      Now check at the protocol level
 641          */
 642         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 643                 goto out;
 644
 645         /*
 646          *      Only reply to fragment 0. We byte re-order the constant
 647          *      mask for efficiency.
 648          */
 649         if (iph->frag_off & htons(IP_OFFSET))
 650                 goto out;
 651
 652         /*
 653          *      If we send an ICMP error to an ICMP error a mess would result..
 654          */
 655         if (icmp_pointers[type].error) {
 656                 /*
 657                  *      We are an error, check if we are replying to an
 658                  *      ICMP error
 659                  */
 660                 if (iph->protocol == IPPROTO_ICMP) {
 661                         u8 _inner_type, *itp;
 662
 663                         itp = skb_header_pointer(skb_in,
 664                                                  skb_network_header(skb_in) +
 665                                                  (iph->ihl << 2) +
 666                                                  offsetof(struct icmphdr,
 667                                                           type) -
 668                                                  skb_in->data,
 669                                                  sizeof(_inner_type),
 670                                                  &_inner_type);
 671                         if (!itp)
 672                                 goto out;
 673
 674                         /*
 675                          *      Assume any unknown ICMP type is an error. This
 676                          *      isn't specified by the RFC, but think about it..
 677                          */
 678                         if (*itp > NR_ICMP_TYPES ||
 679                             icmp_pointers[*itp].error)
 680                                 goto out;
 681                 }
 682         }
 683
 684         /* Needed by both icmpv4_global_allow and icmp_xmit_lock */
 685         local_bh_disable();
 686
 687         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 688          * incoming dev is loopback.  If outgoing dev change to not be
 689          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 690          */
 691         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 692               !icmpv4_global_allow(net, type, code, &apply_ratelimit))
 693                 goto out_bh_enable;
 694
 695         sk = icmp_xmit_lock(net);
 696         if (!sk)
 697                 goto out_bh_enable;
 698
 699         /*
 700          *      Construct source address and options.
 701          */
 702
 703         saddr = iph->daddr;
 704         if (!(rt->rt_flags & RTCF_LOCAL)) {
 705                 struct net_device *dev = NULL;
 706
 707                 rcu_read_lock();
 708                 if (rt_is_input_route(rt) &&
 709                     READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
 710                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 711
 712                 if (dev)
 713                         saddr = inet_select_addr(dev, iph->saddr,
 714                                                  RT_SCOPE_LINK);
 715                 else
 716                         saddr = 0;
 717                 rcu_read_unlock();
 718         }
 719
 720         tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 721                                            IPTOS_PREC_INTERNETCONTROL) :
 722                                            iph->tos;
 723         mark = IP4_REPLY_MARK(net, skb_in->mark);
 724
 725         if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 726                 goto out_unlock;
 727
 728
 729         /*
 730          *      Prepare data for ICMP header.
 731          */
 732
 733         icmp_param.data.icmph.type       = type;
 734         icmp_param.data.icmph.code       = code;
 735         icmp_param.data.icmph.un.gateway = info;
 736         icmp_param.data.icmph.checksum   = 0;
 737         icmp_param.skb    = skb_in;
 738         icmp_param.offset = skb_network_offset(skb_in);
 739         inet_sk(sk)->tos = tos;
 740         ipcm_init(&ipc);
 741         ipc.addr = iph->saddr;
 742         ipc.opt = &icmp_param.replyopts.opt;
 743         ipc.sockc.mark = mark;
 744
 745         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
 746                                inet_dsfield_to_dscp(tos), mark, type, code,
 747                                &icmp_param);
 748         if (IS_ERR(rt))
 749                 goto out_unlock;
 750
 751         /* peer icmp_ratelimit */
 752         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
 753                 goto ende;
 754
 755         /* RFC says return as much as we can without exceeding 576 bytes. */
 756
 757         room = dst_mtu(&rt->dst);
 758         if (room > 576)
 759                 room = 576;
 760         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 761         room -= sizeof(struct icmphdr);
 762         /* Guard against tiny mtu. We need to include at least one
 763          * IP network header for this message to make any sense.
 764          */
 765         if (room <= (int)sizeof(struct iphdr))
 766                 goto ende;
 767
 768         icmp_param.data_len = skb_in->len - icmp_param.offset;
 769         if (icmp_param.data_len > room)
 770                 icmp_param.data_len = room;
 771         icmp_param.head_len = sizeof(struct icmphdr);
 772
 773         /* if we don't have a source address at this point, fall back to the
 774          * dummy address instead of sending out a packet with a source address
 775          * of 0.0.0.0
 776          */
 777         if (!fl4.saddr)
 778                 fl4.saddr = htonl(INADDR_DUMMY);
 779
 780         trace_icmp_send(skb_in, type, code);
 781
 782         icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
 783 ende:
 784         ip_rt_put(rt);
 785 out_unlock:
 786         icmp_xmit_unlock(sk);
 787 out_bh_enable:
 788         local_bh_enable();
 789 out:;
 790 }
 791 EXPORT_SYMBOL(__icmp_send);
 792
 793 #if IS_ENABLED(CONFIG_NF_NAT)
 794 #include <net/netfilter/nf_conntrack.h>
 795 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 796 {
 797         struct sk_buff *cloned_skb = NULL;
 798         struct ip_options opts = { 0 };
 799         enum ip_conntrack_info ctinfo;
 800         struct nf_conn *ct;
 801         __be32 orig_ip;
 802
 803         ct = nf_ct_get(skb_in, &ctinfo);
 804         if (!ct || !(ct->status & IPS_SRC_NAT)) {
 805                 __icmp_send(skb_in, type, code, info, &opts);
 806                 return;
 807         }
 808
 809         if (skb_shared(skb_in))
 810                 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 811
 812         if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 813             (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 814             skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 815             skb_network_offset(skb_in) + sizeof(struct iphdr))))
 816                 goto out;
 817
 818         orig_ip = ip_hdr(skb_in)->saddr;
 819         ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 820         __icmp_send(skb_in, type, code, info, &opts);
 821         ip_hdr(skb_in)->saddr = orig_ip;
 822 out:
 823         consume_skb(cloned_skb);
 824 }
 825 EXPORT_SYMBOL(icmp_ndo_send);
 826 #endif
 827
 828 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 829 {
 830         const struct iphdr *iph = (const struct iphdr *)skb->data;
 831         const struct net_protocol *ipprot;
 832         int protocol = iph->protocol;
 833
 834         /* Checkin full IP header plus 8 bytes of protocol to
 835          * avoid additional coding at protocol handlers.
 836          */
 837         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 838                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 839                 return;
 840         }
 841
 842         raw_icmp_error(skb, protocol, info);
 843
 844         ipprot = rcu_dereference(inet_protos[protocol]);
 845         if (ipprot && ipprot->err_handler)
 846                 ipprot->err_handler(skb, info);
 847 }
 848
 849 static bool icmp_tag_validation(int proto)
 850 {
 851         bool ok;
 852
 853         rcu_read_lock();
 854         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 855         rcu_read_unlock();
 856         return ok;
 857 }
 858
 859 /*
 860  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 861  *      ICMP_PARAMETERPROB.
 862  */
 863
 864 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
 865 {
 866         enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
 867         const struct iphdr *iph;
 868         struct icmphdr *icmph;
 869         struct net *net;
 870         u32 info = 0;
 871
 872         net = dev_net(skb_dst(skb)->dev);
 873
 874         /*
 875          *      Incomplete header ?
 876          *      Only checks for the IP header, there should be an
 877          *      additional check for longer headers in upper levels.
 878          */
 879
 880         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 881                 goto out_err;
 882
 883         icmph = icmp_hdr(skb);
 884         iph   = (const struct iphdr *)skb->data;
 885
 886         if (iph->ihl < 5)  { /* Mangled header, drop. */
 887                 reason = SKB_DROP_REASON_IP_INHDR;
 888                 goto out_err;
 889         }
 890
 891         switch (icmph->type) {
 892         case ICMP_DEST_UNREACH:
 893                 switch (icmph->code & 15) {
 894                 case ICMP_NET_UNREACH:
 895                 case ICMP_HOST_UNREACH:
 896                 case ICMP_PROT_UNREACH:
 897                 case ICMP_PORT_UNREACH:
 898                         break;
 899                 case ICMP_FRAG_NEEDED:
 900                         /* for documentation of the ip_no_pmtu_disc
 901                          * values please see
 902                          * Documentation/networking/ip-sysctl.rst
 903                          */
 904                         switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
 905                         default:
 906                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 907                                                     &iph->daddr);
 908                                 break;
 909                         case 2:
 910                                 goto out;
 911                         case 3:
 912                                 if (!icmp_tag_validation(iph->protocol))
 913                                         goto out;
 914                                 fallthrough;
 915                         case 0:
 916                                 info = ntohs(icmph->un.frag.mtu);
 917                         }
 918                         break;
 919                 case ICMP_SR_FAILED:
 920                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 921                                             &iph->daddr);
 922                         break;
 923                 default:
 924                         break;
 925                 }
 926                 if (icmph->code > NR_ICMP_UNREACH)
 927                         goto out;
 928                 break;
 929         case ICMP_PARAMETERPROB:
 930                 info = ntohl(icmph->un.gateway) >> 24;
 931                 break;
 932         case ICMP_TIME_EXCEEDED:
 933                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 934                 if (icmph->code == ICMP_EXC_FRAGTIME)
 935                         goto out;
 936                 break;
 937         }
 938
 939         /*
 940          *      Throw it at our lower layers
 941          *
 942          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 943          *                header.
 944          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 945          *                transport layer.
 946          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 947          *                transport layer.
 948          */
 949
 950         /*
 951          *      Check the other end isn't violating RFC 1122. Some routers send
 952          *      bogus responses to broadcast frames. If you see this message
 953          *      first check your netmask matches at both ends, if it does then
 954          *      get the other vendor to fix their kit.
 955          */
 956
 957         if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
 958             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 959                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 960                                      &ip_hdr(skb)->saddr,
 961                                      icmph->type, icmph->code,
 962                                      &iph->daddr, skb->dev->name);
 963                 goto out;
 964         }
 965
 966         icmp_socket_deliver(skb, info);
 967
 968 out:
 969         return reason;
 970 out_err:
 971         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 972         return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
 973 }
 974
 975
 976 /*
 977  *      Handle ICMP_REDIRECT.
 978  */
 979
 980 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
 981 {
 982         if (skb->len < sizeof(struct iphdr)) {
 983                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 984                 return SKB_DROP_REASON_PKT_TOO_SMALL;
 985         }
 986
 987         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 988                 /* there aught to be a stat */
 989                 return SKB_DROP_REASON_NOMEM;
 990         }
 991
 992         icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 993         return SKB_NOT_DROPPED_YET;
 994 }
 995
 996 /*
 997  *      Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
 998  *
 999  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1000  *                requests.
1001  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1002  *                included in the reply.
1003  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1004  *                echo requests, MUST have default=NOT.
1005  *      RFC 8335: 8 MUST have a config option to enable/disable ICMP
1006  *                Extended Echo Functionality, MUST be disabled by default
1007  *      See also WRT handling of options once they are done and working.
1008  */
1009
1010 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1011 {
1012         struct icmp_bxm icmp_param;
1013         struct net *net;
1014
1015         net = dev_net(skb_dst(skb)->dev);
1016         /* should there be an ICMP stat for ignored echos? */
1017         if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1018                 return SKB_NOT_DROPPED_YET;
1019
1020         icmp_param.data.icmph      = *icmp_hdr(skb);
1021         icmp_param.skb             = skb;
1022         icmp_param.offset          = 0;
1023         icmp_param.data_len        = skb->len;
1024         icmp_param.head_len        = sizeof(struct icmphdr);
1025
1026         if (icmp_param.data.icmph.type == ICMP_ECHO)
1027                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1028         else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1029                 return SKB_NOT_DROPPED_YET;
1030
1031         icmp_reply(&icmp_param, skb);
1032         return SKB_NOT_DROPPED_YET;
1033 }
1034
1035 /*      Helper for icmp_echo and icmpv6_echo_reply.
1036  *      Searches for net_device that matches PROBE interface identifier
1037  *              and builds PROBE reply message in icmphdr.
1038  *
1039  *      Returns false if PROBE responses are disabled via sysctl
1040  */
1041
1042 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1043 {
1044         struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1045         struct icmp_ext_echo_iio *iio, _iio;
1046         struct net *net = dev_net(skb->dev);
1047         struct inet6_dev *in6_dev;
1048         struct in_device *in_dev;
1049         struct net_device *dev;
1050         char buff[IFNAMSIZ];
1051         u16 ident_len;
1052         u8 status;
1053
1054         if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1055                 return false;
1056
1057         /* We currently only support probing interfaces on the proxy node
1058          * Check to ensure L-bit is set
1059          */
1060         if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1061                 return false;
1062         /* Clear status bits in reply message */
1063         icmphdr->un.echo.sequence &= htons(0xFF00);
1064         if (icmphdr->type == ICMP_EXT_ECHO)
1065                 icmphdr->type = ICMP_EXT_ECHOREPLY;
1066         else
1067                 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1068         ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1069         /* Size of iio is class_type dependent.
1070          * Only check header here and assign length based on ctype in the switch statement
1071          */
1072         iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1073         if (!ext_hdr || !iio)
1074                 goto send_mal_query;
1075         if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1076             ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1077                 goto send_mal_query;
1078         ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1079         iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1080                                  sizeof(iio->extobj_hdr) + ident_len, &_iio);
1081         if (!iio)
1082                 goto send_mal_query;
1083
1084         status = 0;
1085         dev = NULL;
1086         switch (iio->extobj_hdr.class_type) {
1087         case ICMP_EXT_ECHO_CTYPE_NAME:
1088                 if (ident_len >= IFNAMSIZ)
1089                         goto send_mal_query;
1090                 memset(buff, 0, sizeof(buff));
1091                 memcpy(buff, &iio->ident.name, ident_len);
1092                 dev = dev_get_by_name(net, buff);
1093                 break;
1094         case ICMP_EXT_ECHO_CTYPE_INDEX:
1095                 if (ident_len != sizeof(iio->ident.ifindex))
1096                         goto send_mal_query;
1097                 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1098                 break;
1099         case ICMP_EXT_ECHO_CTYPE_ADDR:
1100                 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1101                     ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1102                                  iio->ident.addr.ctype3_hdr.addrlen)
1103                         goto send_mal_query;
1104                 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1105                 case ICMP_AFI_IP:
1106                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1107                                 goto send_mal_query;
1108                         dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1109                         break;
1110 #if IS_ENABLED(CONFIG_IPV6)
1111                 case ICMP_AFI_IP6:
1112                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1113                                 goto send_mal_query;
1114                         dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1115                         dev_hold(dev);
1116                         break;
1117 #endif
1118                 default:
1119                         goto send_mal_query;
1120                 }
1121                 break;
1122         default:
1123                 goto send_mal_query;
1124         }
1125         if (!dev) {
1126                 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1127                 return true;
1128         }
1129         /* Fill bits in reply message */
1130         if (dev->flags & IFF_UP)
1131                 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1132
1133         in_dev = __in_dev_get_rcu(dev);
1134         if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1135                 status |= ICMP_EXT_ECHOREPLY_IPV4;
1136
1137         in6_dev = __in6_dev_get(dev);
1138         if (in6_dev && !list_empty(&in6_dev->addr_list))
1139                 status |= ICMP_EXT_ECHOREPLY_IPV6;
1140
1141         dev_put(dev);
1142         icmphdr->un.echo.sequence |= htons(status);
1143         return true;
1144 send_mal_query:
1145         icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1146         return true;
1147 }
1148 EXPORT_SYMBOL_GPL(icmp_build_probe);
1149
1150 /*
1151  *      Handle ICMP Timestamp requests.
1152  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1153  *                SHOULD be in the kernel for minimum random latency.
1154  *                MUST be accurate to a few minutes.
1155  *                MUST be updated at least at 15Hz.
1156  */
1157 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1158 {
1159         struct icmp_bxm icmp_param;
1160         /*
1161          *      Too short.
1162          */
1163         if (skb->len < 4)
1164                 goto out_err;
1165
1166         /*
1167          *      Fill in the current time as ms since midnight UT:
1168          */
1169         icmp_param.data.times[1] = inet_current_timestamp();
1170         icmp_param.data.times[2] = icmp_param.data.times[1];
1171
1172         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1173
1174         icmp_param.data.icmph      = *icmp_hdr(skb);
1175         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1176         icmp_param.data.icmph.code = 0;
1177         icmp_param.skb             = skb;
1178         icmp_param.offset          = 0;
1179         icmp_param.data_len        = 0;
1180         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
1181         icmp_reply(&icmp_param, skb);
1182         return SKB_NOT_DROPPED_YET;
1183
1184 out_err:
1185         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1186         return SKB_DROP_REASON_PKT_TOO_SMALL;
1187 }
1188
1189 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1190 {
1191         /* pretend it was a success */
1192         return SKB_NOT_DROPPED_YET;
1193 }
1194
1195 /*
1196  *      Deal with incoming ICMP packets.
1197  */
1198 int icmp_rcv(struct sk_buff *skb)
1199 {
1200         enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1201         struct rtable *rt = skb_rtable(skb);
1202         struct net *net = dev_net(rt->dst.dev);
1203         struct icmphdr *icmph;
1204
1205         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1206                 struct sec_path *sp = skb_sec_path(skb);
1207                 int nh;
1208
1209                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1210                                  XFRM_STATE_ICMP)) {
1211                         reason = SKB_DROP_REASON_XFRM_POLICY;
1212                         goto drop;
1213                 }
1214
1215                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1216                         goto drop;
1217
1218                 nh = skb_network_offset(skb);
1219                 skb_set_network_header(skb, sizeof(*icmph));
1220
1221                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1222                                                 skb)) {
1223                         reason = SKB_DROP_REASON_XFRM_POLICY;
1224                         goto drop;
1225                 }
1226
1227                 skb_set_network_header(skb, nh);
1228         }
1229
1230         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1231
1232         if (skb_checksum_simple_validate(skb))
1233                 goto csum_error;
1234
1235         if (!pskb_pull(skb, sizeof(*icmph)))
1236                 goto error;
1237
1238         icmph = icmp_hdr(skb);
1239
1240         ICMPMSGIN_INC_STATS(net, icmph->type);
1241
1242         /* Check for ICMP Extended Echo (PROBE) messages */
1243         if (icmph->type == ICMP_EXT_ECHO) {
1244                 /* We can't use icmp_pointers[].handler() because it is an array of
1245                  * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1246                  */
1247                 reason = icmp_echo(skb);
1248                 goto reason_check;
1249         }
1250
1251         if (icmph->type == ICMP_EXT_ECHOREPLY) {
1252                 reason = ping_rcv(skb);
1253                 goto reason_check;
1254         }
1255
1256         /*
1257          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1258          *
1259          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1260          *                discarded.
1261          */
1262         if (icmph->type > NR_ICMP_TYPES) {
1263                 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1264                 goto error;
1265         }
1266
1267         /*
1268          *      Parse the ICMP message
1269          */
1270
1271         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1272                 /*
1273                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1274                  *        silently ignored (we let user decide with a sysctl).
1275                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1276                  *        discarded if to broadcast/multicast.
1277                  */
1278                 if ((icmph->type == ICMP_ECHO ||
1279                      icmph->type == ICMP_TIMESTAMP) &&
1280                     READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1281                         reason = SKB_DROP_REASON_INVALID_PROTO;
1282                         goto error;
1283                 }
1284                 if (icmph->type != ICMP_ECHO &&
1285                     icmph->type != ICMP_TIMESTAMP &&
1286                     icmph->type != ICMP_ADDRESS &&
1287                     icmph->type != ICMP_ADDRESSREPLY) {
1288                         reason = SKB_DROP_REASON_INVALID_PROTO;
1289                         goto error;
1290                 }
1291         }
1292
1293         reason = icmp_pointers[icmph->type].handler(skb);
1294 reason_check:
1295         if (!reason)  {
1296                 consume_skb(skb);
1297                 return NET_RX_SUCCESS;
1298         }
1299
1300 drop:
1301         kfree_skb_reason(skb, reason);
1302         return NET_RX_DROP;
1303 csum_error:
1304         reason = SKB_DROP_REASON_ICMP_CSUM;
1305         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1306 error:
1307         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1308         goto drop;
1309 }
1310
1311 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1312 {
1313         struct icmp_extobj_hdr *objh, _objh;
1314         struct icmp_ext_hdr *exth, _exth;
1315         u16 olen;
1316
1317         exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1318         if (!exth)
1319                 return false;
1320         if (exth->version != 2)
1321                 return true;
1322
1323         if (exth->checksum &&
1324             csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1325                 return false;
1326
1327         off += sizeof(_exth);
1328         while (off < skb->len) {
1329                 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1330                 if (!objh)
1331                         return false;
1332
1333                 olen = ntohs(objh->length);
1334                 if (olen < sizeof(_objh))
1335                         return false;
1336
1337                 off += olen;
1338                 if (off > skb->len)
1339                         return false;
1340         }
1341
1342         return true;
1343 }
1344
1345 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1346                            struct sock_ee_data_rfc4884 *out,
1347                            int thlen, int off)
1348 {
1349         int hlen;
1350
1351         /* original datagram headers: end of icmph to payload (skb->data) */
1352         hlen = -skb_transport_offset(skb) - thlen;
1353
1354         /* per rfc 4884: minimal datagram length of 128 bytes */
1355         if (off < 128 || off < hlen)
1356                 return;
1357
1358         /* kernel has stripped headers: return payload offset in bytes */
1359         off -= hlen;
1360         if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1361                 return;
1362
1363         out->len = off;
1364
1365         if (!ip_icmp_error_rfc4884_validate(skb, off))
1366                 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1367 }
1368 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1369
1370 int icmp_err(struct sk_buff *skb, u32 info)
1371 {
1372         struct iphdr *iph = (struct iphdr *)skb->data;
1373         int offset = iph->ihl<<2;
1374         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1375         int type = icmp_hdr(skb)->type;
1376         int code = icmp_hdr(skb)->code;
1377         struct net *net = dev_net(skb->dev);
1378
1379         /*
1380          * Use ping_err to handle all icmp errors except those
1381          * triggered by ICMP_ECHOREPLY which sent from kernel.
1382          */
1383         if (icmph->type != ICMP_ECHOREPLY) {
1384                 ping_err(skb, offset, info);
1385                 return 0;
1386         }
1387
1388         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1389                 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1390         else if (type == ICMP_REDIRECT)
1391                 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1392
1393         return 0;
1394 }
1395
1396 /*
1397  *      This table is the definition of how we handle ICMP.
1398  */
1399 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1400         [ICMP_ECHOREPLY] = {
1401                 .handler = ping_rcv,
1402         },
1403         [1] = {
1404                 .handler = icmp_discard,
1405                 .error = 1,
1406         },
1407         [2] = {
1408                 .handler = icmp_discard,
1409                 .error = 1,
1410         },
1411         [ICMP_DEST_UNREACH] = {
1412                 .handler = icmp_unreach,
1413                 .error = 1,
1414         },
1415         [ICMP_SOURCE_QUENCH] = {
1416                 .handler = icmp_unreach,
1417                 .error = 1,
1418         },
1419         [ICMP_REDIRECT] = {
1420                 .handler = icmp_redirect,
1421                 .error = 1,
1422         },
1423         [6] = {
1424                 .handler = icmp_discard,
1425                 .error = 1,
1426         },
1427         [7] = {
1428                 .handler = icmp_discard,
1429                 .error = 1,
1430         },
1431         [ICMP_ECHO] = {
1432                 .handler = icmp_echo,
1433         },
1434         [9] = {
1435                 .handler = icmp_discard,
1436                 .error = 1,
1437         },
1438         [10] = {
1439                 .handler = icmp_discard,
1440                 .error = 1,
1441         },
1442         [ICMP_TIME_EXCEEDED] = {
1443                 .handler = icmp_unreach,
1444                 .error = 1,
1445         },
1446         [ICMP_PARAMETERPROB] = {
1447                 .handler = icmp_unreach,
1448                 .error = 1,
1449         },
1450         [ICMP_TIMESTAMP] = {
1451                 .handler = icmp_timestamp,
1452         },
1453         [ICMP_TIMESTAMPREPLY] = {
1454                 .handler = icmp_discard,
1455         },
1456         [ICMP_INFO_REQUEST] = {
1457                 .handler = icmp_discard,
1458         },
1459         [ICMP_INFO_REPLY] = {
1460                 .handler = icmp_discard,
1461         },
1462         [ICMP_ADDRESS] = {
1463                 .handler = icmp_discard,
1464         },
1465         [ICMP_ADDRESSREPLY] = {
1466                 .handler = icmp_discard,
1467         },
1468 };
1469
1470 static int __net_init icmp_sk_init(struct net *net)
1471 {
1472         /* Control parameters for ECHO replies. */
1473         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1474         net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1475         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1476
1477         /* Control parameter - ignore bogus broadcast responses? */
1478         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1479
1480         /*
1481          *      Configurable global rate limit.
1482          *
1483          *      ratelimit defines tokens/packet consumed for dst->rate_token
1484          *      bucket ratemask defines which icmp types are ratelimited by
1485          *      setting it's bit position.
1486          *
1487          *      default:
1488          *      dest unreachable (3), source quench (4),
1489          *      time exceeded (11), parameter problem (12)
1490          */
1491
1492         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1493         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1494         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1495         net->ipv4.sysctl_icmp_msgs_per_sec = 1000;
1496         net->ipv4.sysctl_icmp_msgs_burst = 50;
1497
1498         return 0;
1499 }
1500
1501 static struct pernet_operations __net_initdata icmp_sk_ops = {
1502        .init = icmp_sk_init,
1503 };
1504
1505 int __init icmp_init(void)
1506 {
1507         int err, i;
1508
1509         for_each_possible_cpu(i) {
1510                 struct sock *sk;
1511
1512                 err = inet_ctl_sock_create(&sk, PF_INET,
1513                                            SOCK_RAW, IPPROTO_ICMP, &init_net);
1514                 if (err < 0)
1515                         return err;
1516
1517                 per_cpu(ipv4_icmp_sk, i) = sk;
1518
1519                 /* Enough space for 2 64K ICMP packets, including
1520                  * sk_buff/skb_shared_info struct overhead.
1521                  */
1522                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1523
1524                 /*
1525                  * Speedup sock_wfree()
1526                  */
1527                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1528                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1529         }
1530         return register_pernet_subsys(&icmp_sk_ops);
1531 }