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
  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         enum skb_drop_reason (*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 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
 196
 197 /* Called with BH disabled */
 198 static inline struct sock *icmp_xmit_lock(struct net *net)
 199 {
 200         struct sock *sk;
 201
 202         sk = this_cpu_read(ipv4_icmp_sk);
 203
 204         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 205                 /* This can happen if the output path signals a
 206                  * dst_link_failure() for an outgoing ICMP packet.
 207                  */
 208                 return NULL;
 209         }
 210         sock_net_set(sk, net);
 211         return sk;
 212 }
 213
 214 static inline void icmp_xmit_unlock(struct sock *sk)
 215 {
 216         sock_net_set(sk, &init_net);
 217         spin_unlock(&sk->sk_lock.slock);
 218 }
 219
 220 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 221 int sysctl_icmp_msgs_burst __read_mostly = 50;
 222
 223 static struct {
 224         spinlock_t      lock;
 225         u32             credit;
 226         u32             stamp;
 227 } icmp_global = {
 228         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 229 };
 230
 231 /**
 232  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 233  *
 234  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 235  * Returns false if we reached the limit and can not send another packet.
 236  * Note: called with BH disabled
 237  */
 238 bool icmp_global_allow(void)
 239 {
 240         u32 credit, delta, incr = 0, now = (u32)jiffies;
 241         bool rc = false;
 242
 243         /* Check if token bucket is empty and cannot be refilled
 244          * without taking the spinlock. The READ_ONCE() are paired
 245          * with the following WRITE_ONCE() in this same function.
 246          */
 247         if (!READ_ONCE(icmp_global.credit)) {
 248                 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 249                 if (delta < HZ / 50)
 250                         return false;
 251         }
 252
 253         spin_lock(&icmp_global.lock);
 254         delta = min_t(u32, now - icmp_global.stamp, HZ);
 255         if (delta >= HZ / 50) {
 256                 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
 257                 if (incr)
 258                         WRITE_ONCE(icmp_global.stamp, now);
 259         }
 260         credit = min_t(u32, icmp_global.credit + incr,
 261                        READ_ONCE(sysctl_icmp_msgs_burst));
 262         if (credit) {
 263                 /* We want to use a credit of one in average, but need to randomize
 264                  * it for security reasons.
 265                  */
 266                 credit = max_t(int, credit - prandom_u32_max(3), 0);
 267                 rc = true;
 268         }
 269         WRITE_ONCE(icmp_global.credit, credit);
 270         spin_unlock(&icmp_global.lock);
 271         return rc;
 272 }
 273 EXPORT_SYMBOL(icmp_global_allow);
 274
 275 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 276 {
 277         if (type > NR_ICMP_TYPES)
 278                 return true;
 279
 280         /* Don't limit PMTU discovery. */
 281         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 282                 return true;
 283
 284         /* Limit if icmp type is enabled in ratemask. */
 285         if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
 286                 return true;
 287
 288         return false;
 289 }
 290
 291 static bool icmpv4_global_allow(struct net *net, int type, int code)
 292 {
 293         if (icmpv4_mask_allow(net, type, code))
 294                 return true;
 295
 296         if (icmp_global_allow())
 297                 return true;
 298
 299         return false;
 300 }
 301
 302 /*
 303  *      Send an ICMP frame.
 304  */
 305
 306 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 307                                struct flowi4 *fl4, int type, int code)
 308 {
 309         struct dst_entry *dst = &rt->dst;
 310         struct inet_peer *peer;
 311         bool rc = true;
 312         int vif;
 313
 314         if (icmpv4_mask_allow(net, type, code))
 315                 goto out;
 316
 317         /* No rate limit on loopback */
 318         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 319                 goto out;
 320
 321         vif = l3mdev_master_ifindex(dst->dev);
 322         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 323         rc = inet_peer_xrlim_allow(peer,
 324                                    READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
 325         if (peer)
 326                 inet_putpeer(peer);
 327 out:
 328         return rc;
 329 }
 330
 331 /*
 332  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 333  */
 334 void icmp_out_count(struct net *net, unsigned char type)
 335 {
 336         ICMPMSGOUT_INC_STATS(net, type);
 337         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 338 }
 339
 340 /*
 341  *      Checksum each fragment, and on the first include the headers and final
 342  *      checksum.
 343  */
 344 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 345                           struct sk_buff *skb)
 346 {
 347         struct icmp_bxm *icmp_param = from;
 348         __wsum csum;
 349
 350         csum = skb_copy_and_csum_bits(icmp_param->skb,
 351                                       icmp_param->offset + offset,
 352                                       to, len);
 353
 354         skb->csum = csum_block_add(skb->csum, csum, odd);
 355         if (icmp_pointers[icmp_param->data.icmph.type].error)
 356                 nf_ct_attach(skb, icmp_param->skb);
 357         return 0;
 358 }
 359
 360 static void icmp_push_reply(struct sock *sk,
 361                             struct icmp_bxm *icmp_param,
 362                             struct flowi4 *fl4,
 363                             struct ipcm_cookie *ipc, struct rtable **rt)
 364 {
 365         struct sk_buff *skb;
 366
 367         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 368                            icmp_param->data_len+icmp_param->head_len,
 369                            icmp_param->head_len,
 370                            ipc, rt, MSG_DONTWAIT) < 0) {
 371                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 372                 ip_flush_pending_frames(sk);
 373         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 374                 struct icmphdr *icmph = icmp_hdr(skb);
 375                 __wsum csum;
 376                 struct sk_buff *skb1;
 377
 378                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 379                                                  (char *)icmph,
 380                                                  icmp_param->head_len);
 381                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 382                         csum = csum_add(csum, skb1->csum);
 383                 }
 384                 icmph->checksum = csum_fold(csum);
 385                 skb->ip_summed = CHECKSUM_NONE;
 386                 ip_push_pending_frames(sk, fl4);
 387         }
 388 }
 389
 390 /*
 391  *      Driving logic for building and sending ICMP messages.
 392  */
 393
 394 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 395 {
 396         struct ipcm_cookie ipc;
 397         struct rtable *rt = skb_rtable(skb);
 398         struct net *net = dev_net(rt->dst.dev);
 399         struct flowi4 fl4;
 400         struct sock *sk;
 401         struct inet_sock *inet;
 402         __be32 daddr, saddr;
 403         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 404         int type = icmp_param->data.icmph.type;
 405         int code = icmp_param->data.icmph.code;
 406
 407         if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 408                 return;
 409
 410         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 411         local_bh_disable();
 412
 413         /* global icmp_msgs_per_sec */
 414         if (!icmpv4_global_allow(net, type, code))
 415                 goto out_bh_enable;
 416
 417         sk = icmp_xmit_lock(net);
 418         if (!sk)
 419                 goto out_bh_enable;
 420         inet = inet_sk(sk);
 421
 422         icmp_param->data.icmph.checksum = 0;
 423
 424         ipcm_init(&ipc);
 425         inet->tos = ip_hdr(skb)->tos;
 426         ipc.sockc.mark = mark;
 427         daddr = ipc.addr = ip_hdr(skb)->saddr;
 428         saddr = fib_compute_spec_dst(skb);
 429
 430         if (icmp_param->replyopts.opt.opt.optlen) {
 431                 ipc.opt = &icmp_param->replyopts.opt;
 432                 if (ipc.opt->opt.srr)
 433                         daddr = icmp_param->replyopts.opt.opt.faddr;
 434         }
 435         memset(&fl4, 0, sizeof(fl4));
 436         fl4.daddr = daddr;
 437         fl4.saddr = saddr;
 438         fl4.flowi4_mark = mark;
 439         fl4.flowi4_uid = sock_net_uid(net, NULL);
 440         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 441         fl4.flowi4_proto = IPPROTO_ICMP;
 442         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 443         security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 444         rt = ip_route_output_key(net, &fl4);
 445         if (IS_ERR(rt))
 446                 goto out_unlock;
 447         if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 448                 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
 449         ip_rt_put(rt);
 450 out_unlock:
 451         icmp_xmit_unlock(sk);
 452 out_bh_enable:
 453         local_bh_enable();
 454 }
 455
 456 /*
 457  * The device used for looking up which routing table to use for sending an ICMP
 458  * error is preferably the source whenever it is set, which should ensure the
 459  * icmp error can be sent to the source host, else lookup using the routing
 460  * table of the destination device, else use the main routing table (index 0).
 461  */
 462 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 463 {
 464         struct net_device *route_lookup_dev = NULL;
 465
 466         if (skb->dev)
 467                 route_lookup_dev = skb->dev;
 468         else if (skb_dst(skb))
 469                 route_lookup_dev = skb_dst(skb)->dev;
 470         return route_lookup_dev;
 471 }
 472
 473 static struct rtable *icmp_route_lookup(struct net *net,
 474                                         struct flowi4 *fl4,
 475                                         struct sk_buff *skb_in,
 476                                         const struct iphdr *iph,
 477                                         __be32 saddr, u8 tos, u32 mark,
 478                                         int type, int code,
 479                                         struct icmp_bxm *param)
 480 {
 481         struct net_device *route_lookup_dev;
 482         struct rtable *rt, *rt2;
 483         struct flowi4 fl4_dec;
 484         int err;
 485
 486         memset(fl4, 0, sizeof(*fl4));
 487         fl4->daddr = (param->replyopts.opt.opt.srr ?
 488                       param->replyopts.opt.opt.faddr : iph->saddr);
 489         fl4->saddr = saddr;
 490         fl4->flowi4_mark = mark;
 491         fl4->flowi4_uid = sock_net_uid(net, NULL);
 492         fl4->flowi4_tos = RT_TOS(tos);
 493         fl4->flowi4_proto = IPPROTO_ICMP;
 494         fl4->fl4_icmp_type = type;
 495         fl4->fl4_icmp_code = code;
 496         route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 497         fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 498
 499         security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
 500         rt = ip_route_output_key_hash(net, fl4, skb_in);
 501         if (IS_ERR(rt))
 502                 return rt;
 503
 504         /* No need to clone since we're just using its address. */
 505         rt2 = rt;
 506
 507         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 508                                            flowi4_to_flowi(fl4), NULL, 0);
 509         if (!IS_ERR(rt)) {
 510                 if (rt != rt2)
 511                         return rt;
 512         } else if (PTR_ERR(rt) == -EPERM) {
 513                 rt = NULL;
 514         } else
 515                 return rt;
 516
 517         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 518         if (err)
 519                 goto relookup_failed;
 520
 521         if (inet_addr_type_dev_table(net, route_lookup_dev,
 522                                      fl4_dec.saddr) == RTN_LOCAL) {
 523                 rt2 = __ip_route_output_key(net, &fl4_dec);
 524                 if (IS_ERR(rt2))
 525                         err = PTR_ERR(rt2);
 526         } else {
 527                 struct flowi4 fl4_2 = {};
 528                 unsigned long orefdst;
 529
 530                 fl4_2.daddr = fl4_dec.saddr;
 531                 rt2 = ip_route_output_key(net, &fl4_2);
 532                 if (IS_ERR(rt2)) {
 533                         err = PTR_ERR(rt2);
 534                         goto relookup_failed;
 535                 }
 536                 /* Ugh! */
 537                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 538                 skb_dst_set(skb_in, NULL);
 539                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 540                                      RT_TOS(tos), rt2->dst.dev);
 541
 542                 dst_release(&rt2->dst);
 543                 rt2 = skb_rtable(skb_in);
 544                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 545         }
 546
 547         if (err)
 548                 goto relookup_failed;
 549
 550         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 551                                             flowi4_to_flowi(&fl4_dec), NULL,
 552                                             XFRM_LOOKUP_ICMP);
 553         if (!IS_ERR(rt2)) {
 554                 dst_release(&rt->dst);
 555                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 556                 rt = rt2;
 557         } else if (PTR_ERR(rt2) == -EPERM) {
 558                 if (rt)
 559                         dst_release(&rt->dst);
 560                 return rt2;
 561         } else {
 562                 err = PTR_ERR(rt2);
 563                 goto relookup_failed;
 564         }
 565         return rt;
 566
 567 relookup_failed:
 568         if (rt)
 569                 return rt;
 570         return ERR_PTR(err);
 571 }
 572
 573 /*
 574  *      Send an ICMP message in response to a situation
 575  *
 576  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 577  *                MAY send more (we do).
 578  *                      MUST NOT change this header information.
 579  *                      MUST NOT reply to a multicast/broadcast IP address.
 580  *                      MUST NOT reply to a multicast/broadcast MAC address.
 581  *                      MUST reply to only the first fragment.
 582  */
 583
 584 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 585                  const struct ip_options *opt)
 586 {
 587         struct iphdr *iph;
 588         int room;
 589         struct icmp_bxm icmp_param;
 590         struct rtable *rt = skb_rtable(skb_in);
 591         struct ipcm_cookie ipc;
 592         struct flowi4 fl4;
 593         __be32 saddr;
 594         u8  tos;
 595         u32 mark;
 596         struct net *net;
 597         struct sock *sk;
 598
 599         if (!rt)
 600                 goto out;
 601
 602         if (rt->dst.dev)
 603                 net = dev_net(rt->dst.dev);
 604         else if (skb_in->dev)
 605                 net = dev_net(skb_in->dev);
 606         else
 607                 goto out;
 608
 609         /*
 610          *      Find the original header. It is expected to be valid, of course.
 611          *      Check this, icmp_send is called from the most obscure devices
 612          *      sometimes.
 613          */
 614         iph = ip_hdr(skb_in);
 615
 616         if ((u8 *)iph < skb_in->head ||
 617             (skb_network_header(skb_in) + sizeof(*iph)) >
 618             skb_tail_pointer(skb_in))
 619                 goto out;
 620
 621         /*
 622          *      No replies to physical multicast/broadcast
 623          */
 624         if (skb_in->pkt_type != PACKET_HOST)
 625                 goto out;
 626
 627         /*
 628          *      Now check at the protocol level
 629          */
 630         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 631                 goto out;
 632
 633         /*
 634          *      Only reply to fragment 0. We byte re-order the constant
 635          *      mask for efficiency.
 636          */
 637         if (iph->frag_off & htons(IP_OFFSET))
 638                 goto out;
 639
 640         /*
 641          *      If we send an ICMP error to an ICMP error a mess would result..
 642          */
 643         if (icmp_pointers[type].error) {
 644                 /*
 645                  *      We are an error, check if we are replying to an
 646                  *      ICMP error
 647                  */
 648                 if (iph->protocol == IPPROTO_ICMP) {
 649                         u8 _inner_type, *itp;
 650
 651                         itp = skb_header_pointer(skb_in,
 652                                                  skb_network_header(skb_in) +
 653                                                  (iph->ihl << 2) +
 654                                                  offsetof(struct icmphdr,
 655                                                           type) -
 656                                                  skb_in->data,
 657                                                  sizeof(_inner_type),
 658                                                  &_inner_type);
 659                         if (!itp)
 660                                 goto out;
 661
 662                         /*
 663                          *      Assume any unknown ICMP type is an error. This
 664                          *      isn't specified by the RFC, but think about it..
 665                          */
 666                         if (*itp > NR_ICMP_TYPES ||
 667                             icmp_pointers[*itp].error)
 668                                 goto out;
 669                 }
 670         }
 671
 672         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 673         local_bh_disable();
 674
 675         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 676          * incoming dev is loopback.  If outgoing dev change to not be
 677          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 678          */
 679         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 680               !icmpv4_global_allow(net, type, code))
 681                 goto out_bh_enable;
 682
 683         sk = icmp_xmit_lock(net);
 684         if (!sk)
 685                 goto out_bh_enable;
 686
 687         /*
 688          *      Construct source address and options.
 689          */
 690
 691         saddr = iph->daddr;
 692         if (!(rt->rt_flags & RTCF_LOCAL)) {
 693                 struct net_device *dev = NULL;
 694
 695                 rcu_read_lock();
 696                 if (rt_is_input_route(rt) &&
 697                     READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
 698                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 699
 700                 if (dev)
 701                         saddr = inet_select_addr(dev, iph->saddr,
 702                                                  RT_SCOPE_LINK);
 703                 else
 704                         saddr = 0;
 705                 rcu_read_unlock();
 706         }
 707
 708         tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 709                                            IPTOS_PREC_INTERNETCONTROL) :
 710                                            iph->tos;
 711         mark = IP4_REPLY_MARK(net, skb_in->mark);
 712
 713         if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 714                 goto out_unlock;
 715
 716
 717         /*
 718          *      Prepare data for ICMP header.
 719          */
 720
 721         icmp_param.data.icmph.type       = type;
 722         icmp_param.data.icmph.code       = code;
 723         icmp_param.data.icmph.un.gateway = info;
 724         icmp_param.data.icmph.checksum   = 0;
 725         icmp_param.skb    = skb_in;
 726         icmp_param.offset = skb_network_offset(skb_in);
 727         inet_sk(sk)->tos = tos;
 728         ipcm_init(&ipc);
 729         ipc.addr = iph->saddr;
 730         ipc.opt = &icmp_param.replyopts.opt;
 731         ipc.sockc.mark = mark;
 732
 733         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 734                                type, code, &icmp_param);
 735         if (IS_ERR(rt))
 736                 goto out_unlock;
 737
 738         /* peer icmp_ratelimit */
 739         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 740                 goto ende;
 741
 742         /* RFC says return as much as we can without exceeding 576 bytes. */
 743
 744         room = dst_mtu(&rt->dst);
 745         if (room > 576)
 746                 room = 576;
 747         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 748         room -= sizeof(struct icmphdr);
 749
 750         icmp_param.data_len = skb_in->len - icmp_param.offset;
 751         if (icmp_param.data_len > room)
 752                 icmp_param.data_len = room;
 753         icmp_param.head_len = sizeof(struct icmphdr);
 754
 755         /* if we don't have a source address at this point, fall back to the
 756          * dummy address instead of sending out a packet with a source address
 757          * of 0.0.0.0
 758          */
 759         if (!fl4.saddr)
 760                 fl4.saddr = htonl(INADDR_DUMMY);
 761
 762         icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
 763 ende:
 764         ip_rt_put(rt);
 765 out_unlock:
 766         icmp_xmit_unlock(sk);
 767 out_bh_enable:
 768         local_bh_enable();
 769 out:;
 770 }
 771 EXPORT_SYMBOL(__icmp_send);
 772
 773 #if IS_ENABLED(CONFIG_NF_NAT)
 774 #include <net/netfilter/nf_conntrack.h>
 775 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 776 {
 777         struct sk_buff *cloned_skb = NULL;
 778         struct ip_options opts = { 0 };
 779         enum ip_conntrack_info ctinfo;
 780         struct nf_conn *ct;
 781         __be32 orig_ip;
 782
 783         ct = nf_ct_get(skb_in, &ctinfo);
 784         if (!ct || !(ct->status & IPS_SRC_NAT)) {
 785                 __icmp_send(skb_in, type, code, info, &opts);
 786                 return;
 787         }
 788
 789         if (skb_shared(skb_in))
 790                 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 791
 792         if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 793             (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 794             skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 795             skb_network_offset(skb_in) + sizeof(struct iphdr))))
 796                 goto out;
 797
 798         orig_ip = ip_hdr(skb_in)->saddr;
 799         ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 800         __icmp_send(skb_in, type, code, info, &opts);
 801         ip_hdr(skb_in)->saddr = orig_ip;
 802 out:
 803         consume_skb(cloned_skb);
 804 }
 805 EXPORT_SYMBOL(icmp_ndo_send);
 806 #endif
 807
 808 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 809 {
 810         const struct iphdr *iph = (const struct iphdr *)skb->data;
 811         const struct net_protocol *ipprot;
 812         int protocol = iph->protocol;
 813
 814         /* Checkin full IP header plus 8 bytes of protocol to
 815          * avoid additional coding at protocol handlers.
 816          */
 817         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 818                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 819                 return;
 820         }
 821
 822         raw_icmp_error(skb, protocol, info);
 823
 824         ipprot = rcu_dereference(inet_protos[protocol]);
 825         if (ipprot && ipprot->err_handler)
 826                 ipprot->err_handler(skb, info);
 827 }
 828
 829 static bool icmp_tag_validation(int proto)
 830 {
 831         bool ok;
 832
 833         rcu_read_lock();
 834         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 835         rcu_read_unlock();
 836         return ok;
 837 }
 838
 839 /*
 840  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 841  *      ICMP_PARAMETERPROB.
 842  */
 843
 844 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
 845 {
 846         enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
 847         const struct iphdr *iph;
 848         struct icmphdr *icmph;
 849         struct net *net;
 850         u32 info = 0;
 851
 852         net = dev_net(skb_dst(skb)->dev);
 853
 854         /*
 855          *      Incomplete header ?
 856          *      Only checks for the IP header, there should be an
 857          *      additional check for longer headers in upper levels.
 858          */
 859
 860         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 861                 goto out_err;
 862
 863         icmph = icmp_hdr(skb);
 864         iph   = (const struct iphdr *)skb->data;
 865
 866         if (iph->ihl < 5)  { /* Mangled header, drop. */
 867                 reason = SKB_DROP_REASON_IP_INHDR;
 868                 goto out_err;
 869         }
 870
 871         switch (icmph->type) {
 872         case ICMP_DEST_UNREACH:
 873                 switch (icmph->code & 15) {
 874                 case ICMP_NET_UNREACH:
 875                 case ICMP_HOST_UNREACH:
 876                 case ICMP_PROT_UNREACH:
 877                 case ICMP_PORT_UNREACH:
 878                         break;
 879                 case ICMP_FRAG_NEEDED:
 880                         /* for documentation of the ip_no_pmtu_disc
 881                          * values please see
 882                          * Documentation/networking/ip-sysctl.rst
 883                          */
 884                         switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
 885                         default:
 886                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 887                                                     &iph->daddr);
 888                                 break;
 889                         case 2:
 890                                 goto out;
 891                         case 3:
 892                                 if (!icmp_tag_validation(iph->protocol))
 893                                         goto out;
 894                                 fallthrough;
 895                         case 0:
 896                                 info = ntohs(icmph->un.frag.mtu);
 897                         }
 898                         break;
 899                 case ICMP_SR_FAILED:
 900                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 901                                             &iph->daddr);
 902                         break;
 903                 default:
 904                         break;
 905                 }
 906                 if (icmph->code > NR_ICMP_UNREACH)
 907                         goto out;
 908                 break;
 909         case ICMP_PARAMETERPROB:
 910                 info = ntohl(icmph->un.gateway) >> 24;
 911                 break;
 912         case ICMP_TIME_EXCEEDED:
 913                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 914                 if (icmph->code == ICMP_EXC_FRAGTIME)
 915                         goto out;
 916                 break;
 917         }
 918
 919         /*
 920          *      Throw it at our lower layers
 921          *
 922          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 923          *                header.
 924          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 925          *                transport layer.
 926          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 927          *                transport layer.
 928          */
 929
 930         /*
 931          *      Check the other end isn't violating RFC 1122. Some routers send
 932          *      bogus responses to broadcast frames. If you see this message
 933          *      first check your netmask matches at both ends, if it does then
 934          *      get the other vendor to fix their kit.
 935          */
 936
 937         if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
 938             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 939                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 940                                      &ip_hdr(skb)->saddr,
 941                                      icmph->type, icmph->code,
 942                                      &iph->daddr, skb->dev->name);
 943                 goto out;
 944         }
 945
 946         icmp_socket_deliver(skb, info);
 947
 948 out:
 949         return reason;
 950 out_err:
 951         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 952         return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
 953 }
 954
 955
 956 /*
 957  *      Handle ICMP_REDIRECT.
 958  */
 959
 960 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
 961 {
 962         if (skb->len < sizeof(struct iphdr)) {
 963                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 964                 return SKB_DROP_REASON_PKT_TOO_SMALL;
 965         }
 966
 967         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 968                 /* there aught to be a stat */
 969                 return SKB_DROP_REASON_NOMEM;
 970         }
 971
 972         icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 973         return SKB_NOT_DROPPED_YET;
 974 }
 975
 976 /*
 977  *      Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
 978  *
 979  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 980  *                requests.
 981  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 982  *                included in the reply.
 983  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 984  *                echo requests, MUST have default=NOT.
 985  *      RFC 8335: 8 MUST have a config option to enable/disable ICMP
 986  *                Extended Echo Functionality, MUST be disabled by default
 987  *      See also WRT handling of options once they are done and working.
 988  */
 989
 990 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
 991 {
 992         struct icmp_bxm icmp_param;
 993         struct net *net;
 994
 995         net = dev_net(skb_dst(skb)->dev);
 996         /* should there be an ICMP stat for ignored echos? */
 997         if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
 998                 return SKB_NOT_DROPPED_YET;
 999
1000         icmp_param.data.icmph      = *icmp_hdr(skb);
1001         icmp_param.skb             = skb;
1002         icmp_param.offset          = 0;
1003         icmp_param.data_len        = skb->len;
1004         icmp_param.head_len        = sizeof(struct icmphdr);
1005
1006         if (icmp_param.data.icmph.type == ICMP_ECHO)
1007                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1008         else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1009                 return SKB_NOT_DROPPED_YET;
1010
1011         icmp_reply(&icmp_param, skb);
1012         return SKB_NOT_DROPPED_YET;
1013 }
1014
1015 /*      Helper for icmp_echo and icmpv6_echo_reply.
1016  *      Searches for net_device that matches PROBE interface identifier
1017  *              and builds PROBE reply message in icmphdr.
1018  *
1019  *      Returns false if PROBE responses are disabled via sysctl
1020  */
1021
1022 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1023 {
1024         struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1025         struct icmp_ext_echo_iio *iio, _iio;
1026         struct net *net = dev_net(skb->dev);
1027         struct net_device *dev;
1028         char buff[IFNAMSIZ];
1029         u16 ident_len;
1030         u8 status;
1031
1032         if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1033                 return false;
1034
1035         /* We currently only support probing interfaces on the proxy node
1036          * Check to ensure L-bit is set
1037          */
1038         if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1039                 return false;
1040         /* Clear status bits in reply message */
1041         icmphdr->un.echo.sequence &= htons(0xFF00);
1042         if (icmphdr->type == ICMP_EXT_ECHO)
1043                 icmphdr->type = ICMP_EXT_ECHOREPLY;
1044         else
1045                 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1046         ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1047         /* Size of iio is class_type dependent.
1048          * Only check header here and assign length based on ctype in the switch statement
1049          */
1050         iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1051         if (!ext_hdr || !iio)
1052                 goto send_mal_query;
1053         if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1054             ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1055                 goto send_mal_query;
1056         ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1057         iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1058                                  sizeof(iio->extobj_hdr) + ident_len, &_iio);
1059         if (!iio)
1060                 goto send_mal_query;
1061
1062         status = 0;
1063         dev = NULL;
1064         switch (iio->extobj_hdr.class_type) {
1065         case ICMP_EXT_ECHO_CTYPE_NAME:
1066                 if (ident_len >= IFNAMSIZ)
1067                         goto send_mal_query;
1068                 memset(buff, 0, sizeof(buff));
1069                 memcpy(buff, &iio->ident.name, ident_len);
1070                 dev = dev_get_by_name(net, buff);
1071                 break;
1072         case ICMP_EXT_ECHO_CTYPE_INDEX:
1073                 if (ident_len != sizeof(iio->ident.ifindex))
1074                         goto send_mal_query;
1075                 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1076                 break;
1077         case ICMP_EXT_ECHO_CTYPE_ADDR:
1078                 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1079                     ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1080                                  iio->ident.addr.ctype3_hdr.addrlen)
1081                         goto send_mal_query;
1082                 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1083                 case ICMP_AFI_IP:
1084                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1085                                 goto send_mal_query;
1086                         dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1087                         break;
1088 #if IS_ENABLED(CONFIG_IPV6)
1089                 case ICMP_AFI_IP6:
1090                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1091                                 goto send_mal_query;
1092                         dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1093                         dev_hold(dev);
1094                         break;
1095 #endif
1096                 default:
1097                         goto send_mal_query;
1098                 }
1099                 break;
1100         default:
1101                 goto send_mal_query;
1102         }
1103         if (!dev) {
1104                 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1105                 return true;
1106         }
1107         /* Fill bits in reply message */
1108         if (dev->flags & IFF_UP)
1109                 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1110         if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1111                 status |= ICMP_EXT_ECHOREPLY_IPV4;
1112         if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1113                 status |= ICMP_EXT_ECHOREPLY_IPV6;
1114         dev_put(dev);
1115         icmphdr->un.echo.sequence |= htons(status);
1116         return true;
1117 send_mal_query:
1118         icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1119         return true;
1120 }
1121 EXPORT_SYMBOL_GPL(icmp_build_probe);
1122
1123 /*
1124  *      Handle ICMP Timestamp requests.
1125  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1126  *                SHOULD be in the kernel for minimum random latency.
1127  *                MUST be accurate to a few minutes.
1128  *                MUST be updated at least at 15Hz.
1129  */
1130 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1131 {
1132         struct icmp_bxm icmp_param;
1133         /*
1134          *      Too short.
1135          */
1136         if (skb->len < 4)
1137                 goto out_err;
1138
1139         /*
1140          *      Fill in the current time as ms since midnight UT:
1141          */
1142         icmp_param.data.times[1] = inet_current_timestamp();
1143         icmp_param.data.times[2] = icmp_param.data.times[1];
1144
1145         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1146
1147         icmp_param.data.icmph      = *icmp_hdr(skb);
1148         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1149         icmp_param.data.icmph.code = 0;
1150         icmp_param.skb             = skb;
1151         icmp_param.offset          = 0;
1152         icmp_param.data_len        = 0;
1153         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
1154         icmp_reply(&icmp_param, skb);
1155         return SKB_NOT_DROPPED_YET;
1156
1157 out_err:
1158         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1159         return SKB_DROP_REASON_PKT_TOO_SMALL;
1160 }
1161
1162 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1163 {
1164         /* pretend it was a success */
1165         return SKB_NOT_DROPPED_YET;
1166 }
1167
1168 /*
1169  *      Deal with incoming ICMP packets.
1170  */
1171 int icmp_rcv(struct sk_buff *skb)
1172 {
1173         enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1174         struct rtable *rt = skb_rtable(skb);
1175         struct net *net = dev_net(rt->dst.dev);
1176         struct icmphdr *icmph;
1177
1178         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1179                 struct sec_path *sp = skb_sec_path(skb);
1180                 int nh;
1181
1182                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1183                                  XFRM_STATE_ICMP)) {
1184                         reason = SKB_DROP_REASON_XFRM_POLICY;
1185                         goto drop;
1186                 }
1187
1188                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1189                         goto drop;
1190
1191                 nh = skb_network_offset(skb);
1192                 skb_set_network_header(skb, sizeof(*icmph));
1193
1194                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1195                                                 skb)) {
1196                         reason = SKB_DROP_REASON_XFRM_POLICY;
1197                         goto drop;
1198                 }
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                 reason = icmp_echo(skb);
1221                 goto reason_check;
1222         }
1223
1224         if (icmph->type == ICMP_EXT_ECHOREPLY) {
1225                 reason = ping_rcv(skb);
1226                 goto reason_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                 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1237                 goto error;
1238         }
1239
1240         /*
1241          *      Parse the ICMP message
1242          */
1243
1244         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1245                 /*
1246                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1247                  *        silently ignored (we let user decide with a sysctl).
1248                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1249                  *        discarded if to broadcast/multicast.
1250                  */
1251                 if ((icmph->type == ICMP_ECHO ||
1252                      icmph->type == ICMP_TIMESTAMP) &&
1253                     READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1254                         reason = SKB_DROP_REASON_INVALID_PROTO;
1255                         goto error;
1256                 }
1257                 if (icmph->type != ICMP_ECHO &&
1258                     icmph->type != ICMP_TIMESTAMP &&
1259                     icmph->type != ICMP_ADDRESS &&
1260                     icmph->type != ICMP_ADDRESSREPLY) {
1261                         reason = SKB_DROP_REASON_INVALID_PROTO;
1262                         goto error;
1263                 }
1264         }
1265
1266         reason = icmp_pointers[icmph->type].handler(skb);
1267 reason_check:
1268         if (!reason)  {
1269                 consume_skb(skb);
1270                 return NET_RX_SUCCESS;
1271         }
1272
1273 drop:
1274         kfree_skb_reason(skb, reason);
1275         return NET_RX_DROP;
1276 csum_error:
1277         reason = SKB_DROP_REASON_ICMP_CSUM;
1278         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1279 error:
1280         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1281         goto drop;
1282 }
1283
1284 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1285 {
1286         struct icmp_extobj_hdr *objh, _objh;
1287         struct icmp_ext_hdr *exth, _exth;
1288         u16 olen;
1289
1290         exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1291         if (!exth)
1292                 return false;
1293         if (exth->version != 2)
1294                 return true;
1295
1296         if (exth->checksum &&
1297             csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1298                 return false;
1299
1300         off += sizeof(_exth);
1301         while (off < skb->len) {
1302                 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1303                 if (!objh)
1304                         return false;
1305
1306                 olen = ntohs(objh->length);
1307                 if (olen < sizeof(_objh))
1308                         return false;
1309
1310                 off += olen;
1311                 if (off > skb->len)
1312                         return false;
1313         }
1314
1315         return true;
1316 }
1317
1318 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1319                            struct sock_ee_data_rfc4884 *out,
1320                            int thlen, int off)
1321 {
1322         int hlen;
1323
1324         /* original datagram headers: end of icmph to payload (skb->data) */
1325         hlen = -skb_transport_offset(skb) - thlen;
1326
1327         /* per rfc 4884: minimal datagram length of 128 bytes */
1328         if (off < 128 || off < hlen)
1329                 return;
1330
1331         /* kernel has stripped headers: return payload offset in bytes */
1332         off -= hlen;
1333         if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1334                 return;
1335
1336         out->len = off;
1337
1338         if (!ip_icmp_error_rfc4884_validate(skb, off))
1339                 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1340 }
1341 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1342
1343 int icmp_err(struct sk_buff *skb, u32 info)
1344 {
1345         struct iphdr *iph = (struct iphdr *)skb->data;
1346         int offset = iph->ihl<<2;
1347         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1348         int type = icmp_hdr(skb)->type;
1349         int code = icmp_hdr(skb)->code;
1350         struct net *net = dev_net(skb->dev);
1351
1352         /*
1353          * Use ping_err to handle all icmp errors except those
1354          * triggered by ICMP_ECHOREPLY which sent from kernel.
1355          */
1356         if (icmph->type != ICMP_ECHOREPLY) {
1357                 ping_err(skb, offset, info);
1358                 return 0;
1359         }
1360
1361         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1362                 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1363         else if (type == ICMP_REDIRECT)
1364                 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1365
1366         return 0;
1367 }
1368
1369 /*
1370  *      This table is the definition of how we handle ICMP.
1371  */
1372 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1373         [ICMP_ECHOREPLY] = {
1374                 .handler = ping_rcv,
1375         },
1376         [1] = {
1377                 .handler = icmp_discard,
1378                 .error = 1,
1379         },
1380         [2] = {
1381                 .handler = icmp_discard,
1382                 .error = 1,
1383         },
1384         [ICMP_DEST_UNREACH] = {
1385                 .handler = icmp_unreach,
1386                 .error = 1,
1387         },
1388         [ICMP_SOURCE_QUENCH] = {
1389                 .handler = icmp_unreach,
1390                 .error = 1,
1391         },
1392         [ICMP_REDIRECT] = {
1393                 .handler = icmp_redirect,
1394                 .error = 1,
1395         },
1396         [6] = {
1397                 .handler = icmp_discard,
1398                 .error = 1,
1399         },
1400         [7] = {
1401                 .handler = icmp_discard,
1402                 .error = 1,
1403         },
1404         [ICMP_ECHO] = {
1405                 .handler = icmp_echo,
1406         },
1407         [9] = {
1408                 .handler = icmp_discard,
1409                 .error = 1,
1410         },
1411         [10] = {
1412                 .handler = icmp_discard,
1413                 .error = 1,
1414         },
1415         [ICMP_TIME_EXCEEDED] = {
1416                 .handler = icmp_unreach,
1417                 .error = 1,
1418         },
1419         [ICMP_PARAMETERPROB] = {
1420                 .handler = icmp_unreach,
1421                 .error = 1,
1422         },
1423         [ICMP_TIMESTAMP] = {
1424                 .handler = icmp_timestamp,
1425         },
1426         [ICMP_TIMESTAMPREPLY] = {
1427                 .handler = icmp_discard,
1428         },
1429         [ICMP_INFO_REQUEST] = {
1430                 .handler = icmp_discard,
1431         },
1432         [ICMP_INFO_REPLY] = {
1433                 .handler = icmp_discard,
1434         },
1435         [ICMP_ADDRESS] = {
1436                 .handler = icmp_discard,
1437         },
1438         [ICMP_ADDRESSREPLY] = {
1439                 .handler = icmp_discard,
1440         },
1441 };
1442
1443 static int __net_init icmp_sk_init(struct net *net)
1444 {
1445         /* Control parameters for ECHO replies. */
1446         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1447         net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1448         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1449
1450         /* Control parameter - ignore bogus broadcast responses? */
1451         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1452
1453         /*
1454          *      Configurable global rate limit.
1455          *
1456          *      ratelimit defines tokens/packet consumed for dst->rate_token
1457          *      bucket ratemask defines which icmp types are ratelimited by
1458          *      setting it's bit position.
1459          *
1460          *      default:
1461          *      dest unreachable (3), source quench (4),
1462          *      time exceeded (11), parameter problem (12)
1463          */
1464
1465         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1466         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1467         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1468
1469         return 0;
1470 }
1471
1472 static struct pernet_operations __net_initdata icmp_sk_ops = {
1473        .init = icmp_sk_init,
1474 };
1475
1476 int __init icmp_init(void)
1477 {
1478         int err, i;
1479
1480         for_each_possible_cpu(i) {
1481                 struct sock *sk;
1482
1483                 err = inet_ctl_sock_create(&sk, PF_INET,
1484                                            SOCK_RAW, IPPROTO_ICMP, &init_net);
1485                 if (err < 0)
1486                         return err;
1487
1488                 per_cpu(ipv4_icmp_sk, i) = sk;
1489
1490                 /* Enough space for 2 64K ICMP packets, including
1491                  * sk_buff/skb_shared_info struct overhead.
1492                  */
1493                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1494
1495                 /*
1496                  * Speedup sock_wfree()
1497                  */
1498                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1499                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1500         }
1501         return register_pernet_subsys(&icmp_sk_ops);
1502 }