net/ipv4/icmp.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      NET3:   Implementation of the ICMP protocol layer.
   4  *
   5  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
   6  *
   7  *      Some of the function names and the icmp unreach table for this
   8  *      module were derived from [icmp.c 1.0.11 06/02/93] by
   9  *      Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  10  *      Other than that this module is a complete rewrite.
  11  *
  12  *      Fixes:
  13  *      Clemens Fruhwirth       :       introduce global icmp rate limiting
  14  *                                      with icmp type masking ability instead
  15  *                                      of broken per type icmp timeouts.
  16  *              Mike Shaver     :       RFC1122 checks.
  17  *              Alan Cox        :       Multicast ping reply as self.
  18  *              Alan Cox        :       Fix atomicity lockup in ip_build_xmit
  19  *                                      call.
  20  *              Alan Cox        :       Added 216,128 byte paths to the MTU
  21  *                                      code.
  22  *              Martin Mares    :       RFC1812 checks.
  23  *              Martin Mares    :       Can be configured to follow redirects
  24  *                                      if acting as a router _without_ a
  25  *                                      routing protocol (RFC 1812).
  26  *              Martin Mares    :       Echo requests may be configured to
  27  *                                      be ignored (RFC 1812).
  28  *              Martin Mares    :       Limitation of ICMP error message
  29  *                                      transmit rate (RFC 1812).
  30  *              Martin Mares    :       TOS and Precedence set correctly
  31  *                                      (RFC 1812).
  32  *              Martin Mares    :       Now copying as much data from the
  33  *                                      original packet as we can without
  34  *                                      exceeding 576 bytes (RFC 1812).
  35  *      Willy Konynenberg       :       Transparent proxying support.
  36  *              Keith Owens     :       RFC1191 correction for 4.2BSD based
  37  *                                      path MTU bug.
  38  *              Thomas Quinot   :       ICMP Dest Unreach codes up to 15 are
  39  *                                      valid (RFC 1812).
  40  *              Andi Kleen      :       Check all packet lengths properly
  41  *                                      and moved all kfree_skb() up to
  42  *                                      icmp_rcv.
  43  *              Andi Kleen      :       Move the rate limit bookkeeping
  44  *                                      into the dest entry and use a token
  45  *                                      bucket filter (thanks to ANK). Make
  46  *                                      the rates sysctl configurable.
  47  *              Yu Tianli       :       Fixed two ugly bugs in icmp_send
  48  *                                      - IP option length was accounted wrongly
  49  *                                      - ICMP header length was not accounted
  50  *                                        at all.
  51  *              Tristan Greaves :       Added sysctl option to ignore bogus
  52  *                                      broadcast responses from broken routers.
  53  *
  54  * To Fix:
  55  *
  56  *      - Should use skb_pull() instead of all the manual checking.
  57  *        This would also greatly simply some upper layer error handlers. --AK
  58  */
  59
  60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  61
  62 #include <linux/module.h>
  63 #include <linux/types.h>
  64 #include <linux/jiffies.h>
  65 #include <linux/kernel.h>
  66 #include <linux/fcntl.h>
  67 #include <linux/socket.h>
  68 #include <linux/in.h>
  69 #include <linux/inet.h>
  70 #include <linux/inetdevice.h>
  71 #include <linux/netdevice.h>
  72 #include <linux/string.h>
  73 #include <linux/netfilter_ipv4.h>
  74 #include <linux/slab.h>
  75 #include <net/snmp.h>
  76 #include <net/ip.h>
  77 #include <net/route.h>
  78 #include <net/protocol.h>
  79 #include <net/icmp.h>
  80 #include <net/tcp.h>
  81 #include <net/udp.h>
  82 #include <net/raw.h>
  83 #include <net/ping.h>
  84 #include <linux/skbuff.h>
  85 #include <net/sock.h>
  86 #include <linux/errno.h>
  87 #include <linux/timer.h>
  88 #include <linux/init.h>
  89 #include <linux/uaccess.h>
  90 #include <net/checksum.h>
  91 #include <net/xfrm.h>
  92 #include <net/inet_common.h>
  93 #include <net/ip_fib.h>
  94 #include <net/l3mdev.h>
  95
  96 /*
  97  *      Build xmit assembly blocks
  98  */
  99
 100 struct icmp_bxm {
 101         struct sk_buff *skb;
 102         int offset;
 103         int data_len;
 104
 105         struct {
 106                 struct icmphdr icmph;
 107                 __be32         times[3];
 108         } data;
 109         int head_len;
 110         struct ip_options_data replyopts;
 111 };
 112
 113 /* An array of errno for error messages from dest unreach. */
 114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 115
 116 const struct icmp_err icmp_err_convert[] = {
 117         {
 118                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 119                 .fatal = 0,
 120         },
 121         {
 122                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 123                 .fatal = 0,
 124         },
 125         {
 126                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 127                 .fatal = 1,
 128         },
 129         {
 130                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 131                 .fatal = 1,
 132         },
 133         {
 134                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 135                 .fatal = 0,
 136         },
 137         {
 138                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 139                 .fatal = 0,
 140         },
 141         {
 142                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 143                 .fatal = 1,
 144         },
 145         {
 146                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 147                 .fatal = 1,
 148         },
 149         {
 150                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 151                 .fatal = 1,
 152         },
 153         {
 154                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 155                 .fatal = 1,
 156         },
 157         {
 158                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 159                 .fatal = 1,
 160         },
 161         {
 162                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 163                 .fatal = 0,
 164         },
 165         {
 166                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 167                 .fatal = 0,
 168         },
 169         {
 170                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 171                 .fatal = 1,
 172         },
 173         {
 174                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 175                 .fatal = 1,
 176         },
 177         {
 178                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 179                 .fatal = 1,
 180         },
 181 };
 182 EXPORT_SYMBOL(icmp_err_convert);
 183
 184 /*
 185  *      ICMP control array. This specifies what to do with each ICMP.
 186  */
 187
 188 struct icmp_control {
 189         bool (*handler)(struct sk_buff *skb);
 190         short   error;          /* This ICMP is classed as an error message */
 191 };
 192
 193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 194
 195 /*
 196  *      The ICMP socket(s). This is the most convenient way to flow control
 197  *      our ICMP output as well as maintain a clean interface throughout
 198  *      all layers. All Socketless IP sends will soon be gone.
 199  *
 200  *      On SMP we have one ICMP socket per-cpu.
 201  */
 202 static struct sock *icmp_sk(struct net *net)
 203 {
 204         return this_cpu_read(*net->ipv4.icmp_sk);
 205 }
 206
 207 /* Called with BH disabled */
 208 static inline struct sock *icmp_xmit_lock(struct net *net)
 209 {
 210         struct sock *sk;
 211
 212         sk = icmp_sk(net);
 213
 214         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 215                 /* This can happen if the output path signals a
 216                  * dst_link_failure() for an outgoing ICMP packet.
 217                  */
 218                 return NULL;
 219         }
 220         return sk;
 221 }
 222
 223 static inline void icmp_xmit_unlock(struct sock *sk)
 224 {
 225         spin_unlock(&sk->sk_lock.slock);
 226 }
 227
 228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 229 int sysctl_icmp_msgs_burst __read_mostly = 50;
 230
 231 static struct {
 232         spinlock_t      lock;
 233         u32             credit;
 234         u32             stamp;
 235 } icmp_global = {
 236         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 237 };
 238
 239 /**
 240  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 241  *
 242  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 243  * Returns false if we reached the limit and can not send another packet.
 244  * Note: called with BH disabled
 245  */
 246 bool icmp_global_allow(void)
 247 {
 248         u32 credit, delta, incr = 0, now = (u32)jiffies;
 249         bool rc = false;
 250
 251         /* Check if token bucket is empty and cannot be refilled
 252          * without taking the spinlock. The READ_ONCE() are paired
 253          * with the following WRITE_ONCE() in this same function.
 254          */
 255         if (!READ_ONCE(icmp_global.credit)) {
 256                 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 257                 if (delta < HZ / 50)
 258                         return false;
 259         }
 260
 261         spin_lock(&icmp_global.lock);
 262         delta = min_t(u32, now - icmp_global.stamp, HZ);
 263         if (delta >= HZ / 50) {
 264                 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
 265                 if (incr)
 266                         WRITE_ONCE(icmp_global.stamp, now);
 267         }
 268         credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
 269         if (credit) {
 270                 /* We want to use a credit of one in average, but need to randomize
 271                  * it for security reasons.
 272                  */
 273                 credit = max_t(int, credit - prandom_u32_max(3), 0);
 274                 rc = true;
 275         }
 276         WRITE_ONCE(icmp_global.credit, credit);
 277         spin_unlock(&icmp_global.lock);
 278         return rc;
 279 }
 280 EXPORT_SYMBOL(icmp_global_allow);
 281
 282 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 283 {
 284         if (type > NR_ICMP_TYPES)
 285                 return true;
 286
 287         /* Don't limit PMTU discovery. */
 288         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 289                 return true;
 290
 291         /* Limit if icmp type is enabled in ratemask. */
 292         if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 293                 return true;
 294
 295         return false;
 296 }
 297
 298 static bool icmpv4_global_allow(struct net *net, int type, int code)
 299 {
 300         if (icmpv4_mask_allow(net, type, code))
 301                 return true;
 302
 303         if (icmp_global_allow())
 304                 return true;
 305
 306         return false;
 307 }
 308
 309 /*
 310  *      Send an ICMP frame.
 311  */
 312
 313 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 314                                struct flowi4 *fl4, int type, int code)
 315 {
 316         struct dst_entry *dst = &rt->dst;
 317         struct inet_peer *peer;
 318         bool rc = true;
 319         int vif;
 320
 321         if (icmpv4_mask_allow(net, type, code))
 322                 goto out;
 323
 324         /* No rate limit on loopback */
 325         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 326                 goto out;
 327
 328         vif = l3mdev_master_ifindex(dst->dev);
 329         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 330         rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
 331         if (peer)
 332                 inet_putpeer(peer);
 333 out:
 334         return rc;
 335 }
 336
 337 /*
 338  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 339  */
 340 void icmp_out_count(struct net *net, unsigned char type)
 341 {
 342         ICMPMSGOUT_INC_STATS(net, type);
 343         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 344 }
 345
 346 /*
 347  *      Checksum each fragment, and on the first include the headers and final
 348  *      checksum.
 349  */
 350 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 351                           struct sk_buff *skb)
 352 {
 353         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 354         __wsum csum;
 355
 356         csum = skb_copy_and_csum_bits(icmp_param->skb,
 357                                       icmp_param->offset + offset,
 358                                       to, len);
 359
 360         skb->csum = csum_block_add(skb->csum, csum, odd);
 361         if (icmp_pointers[icmp_param->data.icmph.type].error)
 362                 nf_ct_attach(skb, icmp_param->skb);
 363         return 0;
 364 }
 365
 366 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 367                             struct flowi4 *fl4,
 368                             struct ipcm_cookie *ipc, struct rtable **rt)
 369 {
 370         struct sock *sk;
 371         struct sk_buff *skb;
 372
 373         sk = icmp_sk(dev_net((*rt)->dst.dev));
 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         struct flowi4 fl4;
 407         struct sock *sk;
 408         struct inet_sock *inet;
 409         __be32 daddr, saddr;
 410         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 411         int type = icmp_param->data.icmph.type;
 412         int code = icmp_param->data.icmph.code;
 413
 414         if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 415                 return;
 416
 417         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 418         local_bh_disable();
 419
 420         /* global icmp_msgs_per_sec */
 421         if (!icmpv4_global_allow(net, type, code))
 422                 goto out_bh_enable;
 423
 424         sk = icmp_xmit_lock(net);
 425         if (!sk)
 426                 goto out_bh_enable;
 427         inet = inet_sk(sk);
 428
 429         icmp_param->data.icmph.checksum = 0;
 430
 431         ipcm_init(&ipc);
 432         inet->tos = ip_hdr(skb)->tos;
 433         ipc.sockc.mark = mark;
 434         daddr = ipc.addr = ip_hdr(skb)->saddr;
 435         saddr = fib_compute_spec_dst(skb);
 436
 437         if (icmp_param->replyopts.opt.opt.optlen) {
 438                 ipc.opt = &icmp_param->replyopts.opt;
 439                 if (ipc.opt->opt.srr)
 440                         daddr = icmp_param->replyopts.opt.opt.faddr;
 441         }
 442         memset(&fl4, 0, sizeof(fl4));
 443         fl4.daddr = daddr;
 444         fl4.saddr = saddr;
 445         fl4.flowi4_mark = mark;
 446         fl4.flowi4_uid = sock_net_uid(net, NULL);
 447         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 448         fl4.flowi4_proto = IPPROTO_ICMP;
 449         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 450         security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 451         rt = ip_route_output_key(net, &fl4);
 452         if (IS_ERR(rt))
 453                 goto out_unlock;
 454         if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 455                 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 456         ip_rt_put(rt);
 457 out_unlock:
 458         icmp_xmit_unlock(sk);
 459 out_bh_enable:
 460         local_bh_enable();
 461 }
 462
 463 /*
 464  * The device used for looking up which routing table to use for sending an ICMP
 465  * error is preferably the source whenever it is set, which should ensure the
 466  * icmp error can be sent to the source host, else lookup using the routing
 467  * table of the destination device, else use the main routing table (index 0).
 468  */
 469 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 470 {
 471         struct net_device *route_lookup_dev = NULL;
 472
 473         if (skb->dev)
 474                 route_lookup_dev = skb->dev;
 475         else if (skb_dst(skb))
 476                 route_lookup_dev = skb_dst(skb)->dev;
 477         return route_lookup_dev;
 478 }
 479
 480 static struct rtable *icmp_route_lookup(struct net *net,
 481                                         struct flowi4 *fl4,
 482                                         struct sk_buff *skb_in,
 483                                         const struct iphdr *iph,
 484                                         __be32 saddr, u8 tos, u32 mark,
 485                                         int type, int code,
 486                                         struct icmp_bxm *param)
 487 {
 488         struct net_device *route_lookup_dev;
 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 = RT_TOS(tos);
 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         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 515                                            flowi4_to_flowi(fl4), NULL, 0);
 516         if (!IS_ERR(rt)) {
 517                 if (rt != rt2)
 518                         return rt;
 519         } else if (PTR_ERR(rt) == -EPERM) {
 520                 rt = NULL;
 521         } else
 522                 return rt;
 523
 524         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 525         if (err)
 526                 goto relookup_failed;
 527
 528         if (inet_addr_type_dev_table(net, route_lookup_dev,
 529                                      fl4_dec.saddr) == RTN_LOCAL) {
 530                 rt2 = __ip_route_output_key(net, &fl4_dec);
 531                 if (IS_ERR(rt2))
 532                         err = PTR_ERR(rt2);
 533         } else {
 534                 struct flowi4 fl4_2 = {};
 535                 unsigned long orefdst;
 536
 537                 fl4_2.daddr = fl4_dec.saddr;
 538                 rt2 = ip_route_output_key(net, &fl4_2);
 539                 if (IS_ERR(rt2)) {
 540                         err = PTR_ERR(rt2);
 541                         goto relookup_failed;
 542                 }
 543                 /* Ugh! */
 544                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 545                 skb_dst_set(skb_in, NULL);
 546                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 547                                      RT_TOS(tos), rt2->dst.dev);
 548
 549                 dst_release(&rt2->dst);
 550                 rt2 = skb_rtable(skb_in);
 551                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 552         }
 553
 554         if (err)
 555                 goto relookup_failed;
 556
 557         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 558                                             flowi4_to_flowi(&fl4_dec), NULL,
 559                                             XFRM_LOOKUP_ICMP);
 560         if (!IS_ERR(rt2)) {
 561                 dst_release(&rt->dst);
 562                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 563                 rt = rt2;
 564         } else if (PTR_ERR(rt2) == -EPERM) {
 565                 if (rt)
 566                         dst_release(&rt->dst);
 567                 return rt2;
 568         } else {
 569                 err = PTR_ERR(rt2);
 570                 goto relookup_failed;
 571         }
 572         return rt;
 573
 574 relookup_failed:
 575         if (rt)
 576                 return rt;
 577         return ERR_PTR(err);
 578 }
 579
 580 /*
 581  *      Send an ICMP message in response to a situation
 582  *
 583  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 584  *                MAY send more (we do).
 585  *                      MUST NOT change this header information.
 586  *                      MUST NOT reply to a multicast/broadcast IP address.
 587  *                      MUST NOT reply to a multicast/broadcast MAC address.
 588  *                      MUST reply to only the first fragment.
 589  */
 590
 591 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 592                  const struct ip_options *opt)
 593 {
 594         struct iphdr *iph;
 595         int room;
 596         struct icmp_bxm icmp_param;
 597         struct rtable *rt = skb_rtable(skb_in);
 598         struct ipcm_cookie ipc;
 599         struct flowi4 fl4;
 600         __be32 saddr;
 601         u8  tos;
 602         u32 mark;
 603         struct net *net;
 604         struct sock *sk;
 605
 606         if (!rt)
 607                 goto out;
 608
 609         if (rt->dst.dev)
 610                 net = dev_net(rt->dst.dev);
 611         else if (skb_in->dev)
 612                 net = dev_net(skb_in->dev);
 613         else
 614                 goto out;
 615
 616         /*
 617          *      Find the original header. It is expected to be valid, of course.
 618          *      Check this, icmp_send is called from the most obscure devices
 619          *      sometimes.
 620          */
 621         iph = ip_hdr(skb_in);
 622
 623         if ((u8 *)iph < skb_in->head ||
 624             (skb_network_header(skb_in) + sizeof(*iph)) >
 625             skb_tail_pointer(skb_in))
 626                 goto out;
 627
 628         /*
 629          *      No replies to physical multicast/broadcast
 630          */
 631         if (skb_in->pkt_type != PACKET_HOST)
 632                 goto out;
 633
 634         /*
 635          *      Now check at the protocol level
 636          */
 637         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 638                 goto out;
 639
 640         /*
 641          *      Only reply to fragment 0. We byte re-order the constant
 642          *      mask for efficiency.
 643          */
 644         if (iph->frag_off & htons(IP_OFFSET))
 645                 goto out;
 646
 647         /*
 648          *      If we send an ICMP error to an ICMP error a mess would result..
 649          */
 650         if (icmp_pointers[type].error) {
 651                 /*
 652                  *      We are an error, check if we are replying to an
 653                  *      ICMP error
 654                  */
 655                 if (iph->protocol == IPPROTO_ICMP) {
 656                         u8 _inner_type, *itp;
 657
 658                         itp = skb_header_pointer(skb_in,
 659                                                  skb_network_header(skb_in) +
 660                                                  (iph->ihl << 2) +
 661                                                  offsetof(struct icmphdr,
 662                                                           type) -
 663                                                  skb_in->data,
 664                                                  sizeof(_inner_type),
 665                                                  &_inner_type);
 666                         if (!itp)
 667                                 goto out;
 668
 669                         /*
 670                          *      Assume any unknown ICMP type is an error. This
 671                          *      isn't specified by the RFC, but think about it..
 672                          */
 673                         if (*itp > NR_ICMP_TYPES ||
 674                             icmp_pointers[*itp].error)
 675                                 goto out;
 676                 }
 677         }
 678
 679         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 680         local_bh_disable();
 681
 682         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 683          * incoming dev is loopback.  If outgoing dev change to not be
 684          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 685          */
 686         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 687               !icmpv4_global_allow(net, type, code))
 688                 goto out_bh_enable;
 689
 690         sk = icmp_xmit_lock(net);
 691         if (!sk)
 692                 goto out_bh_enable;
 693
 694         /*
 695          *      Construct source address and options.
 696          */
 697
 698         saddr = iph->daddr;
 699         if (!(rt->rt_flags & RTCF_LOCAL)) {
 700                 struct net_device *dev = NULL;
 701
 702                 rcu_read_lock();
 703                 if (rt_is_input_route(rt) &&
 704                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 705                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 706
 707                 if (dev)
 708                         saddr = inet_select_addr(dev, iph->saddr,
 709                                                  RT_SCOPE_LINK);
 710                 else
 711                         saddr = 0;
 712                 rcu_read_unlock();
 713         }
 714
 715         tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 716                                            IPTOS_PREC_INTERNETCONTROL) :
 717                                            iph->tos;
 718         mark = IP4_REPLY_MARK(net, skb_in->mark);
 719
 720         if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 721                 goto out_unlock;
 722
 723
 724         /*
 725          *      Prepare data for ICMP header.
 726          */
 727
 728         icmp_param.data.icmph.type       = type;
 729         icmp_param.data.icmph.code       = code;
 730         icmp_param.data.icmph.un.gateway = info;
 731         icmp_param.data.icmph.checksum   = 0;
 732         icmp_param.skb    = skb_in;
 733         icmp_param.offset = skb_network_offset(skb_in);
 734         inet_sk(sk)->tos = tos;
 735         ipcm_init(&ipc);
 736         ipc.addr = iph->saddr;
 737         ipc.opt = &icmp_param.replyopts.opt;
 738         ipc.sockc.mark = mark;
 739
 740         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 741                                type, code, &icmp_param);
 742         if (IS_ERR(rt))
 743                 goto out_unlock;
 744
 745         /* peer icmp_ratelimit */
 746         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 747                 goto ende;
 748
 749         /* RFC says return as much as we can without exceeding 576 bytes. */
 750
 751         room = dst_mtu(&rt->dst);
 752         if (room > 576)
 753                 room = 576;
 754         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 755         room -= sizeof(struct icmphdr);
 756
 757         icmp_param.data_len = skb_in->len - icmp_param.offset;
 758         if (icmp_param.data_len > room)
 759                 icmp_param.data_len = room;
 760         icmp_param.head_len = sizeof(struct icmphdr);
 761
 762         /* if we don't have a source address at this point, fall back to the
 763          * dummy address instead of sending out a packet with a source address
 764          * of 0.0.0.0
 765          */
 766         if (!fl4.saddr)
 767                 fl4.saddr = htonl(INADDR_DUMMY);
 768
 769         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 770 ende:
 771         ip_rt_put(rt);
 772 out_unlock:
 773         icmp_xmit_unlock(sk);
 774 out_bh_enable:
 775         local_bh_enable();
 776 out:;
 777 }
 778 EXPORT_SYMBOL(__icmp_send);
 779
 780 #if IS_ENABLED(CONFIG_NF_NAT)
 781 #include <net/netfilter/nf_conntrack.h>
 782 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 783 {
 784         struct sk_buff *cloned_skb = NULL;
 785         struct ip_options opts = { 0 };
 786         enum ip_conntrack_info ctinfo;
 787         struct nf_conn *ct;
 788         __be32 orig_ip;
 789
 790         ct = nf_ct_get(skb_in, &ctinfo);
 791         if (!ct || !(ct->status & IPS_SRC_NAT)) {
 792                 __icmp_send(skb_in, type, code, info, &opts);
 793                 return;
 794         }
 795
 796         if (skb_shared(skb_in))
 797                 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 798
 799         if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 800             (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 801             skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 802             skb_network_offset(skb_in) + sizeof(struct iphdr))))
 803                 goto out;
 804
 805         orig_ip = ip_hdr(skb_in)->saddr;
 806         ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 807         __icmp_send(skb_in, type, code, info, &opts);
 808         ip_hdr(skb_in)->saddr = orig_ip;
 809 out:
 810         consume_skb(cloned_skb);
 811 }
 812 EXPORT_SYMBOL(icmp_ndo_send);
 813 #endif
 814
 815 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 816 {
 817         const struct iphdr *iph = (const struct iphdr *)skb->data;
 818         const struct net_protocol *ipprot;
 819         int protocol = iph->protocol;
 820
 821         /* Checkin full IP header plus 8 bytes of protocol to
 822          * avoid additional coding at protocol handlers.
 823          */
 824         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 825                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 826                 return;
 827         }
 828
 829         raw_icmp_error(skb, protocol, info);
 830
 831         ipprot = rcu_dereference(inet_protos[protocol]);
 832         if (ipprot && ipprot->err_handler)
 833                 ipprot->err_handler(skb, info);
 834 }
 835
 836 static bool icmp_tag_validation(int proto)
 837 {
 838         bool ok;
 839
 840         rcu_read_lock();
 841         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 842         rcu_read_unlock();
 843         return ok;
 844 }
 845
 846 /*
 847  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 848  *      ICMP_PARAMETERPROB.
 849  */
 850
 851 static bool icmp_unreach(struct sk_buff *skb)
 852 {
 853         const struct iphdr *iph;
 854         struct icmphdr *icmph;
 855         struct net *net;
 856         u32 info = 0;
 857
 858         net = dev_net(skb_dst(skb)->dev);
 859
 860         /*
 861          *      Incomplete header ?
 862          *      Only checks for the IP header, there should be an
 863          *      additional check for longer headers in upper levels.
 864          */
 865
 866         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 867                 goto out_err;
 868
 869         icmph = icmp_hdr(skb);
 870         iph   = (const struct iphdr *)skb->data;
 871
 872         if (iph->ihl < 5) /* Mangled header, drop. */
 873                 goto out_err;
 874
 875         switch (icmph->type) {
 876         case ICMP_DEST_UNREACH:
 877                 switch (icmph->code & 15) {
 878                 case ICMP_NET_UNREACH:
 879                 case ICMP_HOST_UNREACH:
 880                 case ICMP_PROT_UNREACH:
 881                 case ICMP_PORT_UNREACH:
 882                         break;
 883                 case ICMP_FRAG_NEEDED:
 884                         /* for documentation of the ip_no_pmtu_disc
 885                          * values please see
 886                          * Documentation/networking/ip-sysctl.rst
 887                          */
 888                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 889                         default:
 890                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 891                                                     &iph->daddr);
 892                                 break;
 893                         case 2:
 894                                 goto out;
 895                         case 3:
 896                                 if (!icmp_tag_validation(iph->protocol))
 897                                         goto out;
 898                                 fallthrough;
 899                         case 0:
 900                                 info = ntohs(icmph->un.frag.mtu);
 901                         }
 902                         break;
 903                 case ICMP_SR_FAILED:
 904                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 905                                             &iph->daddr);
 906                         break;
 907                 default:
 908                         break;
 909                 }
 910                 if (icmph->code > NR_ICMP_UNREACH)
 911                         goto out;
 912                 break;
 913         case ICMP_PARAMETERPROB:
 914                 info = ntohl(icmph->un.gateway) >> 24;
 915                 break;
 916         case ICMP_TIME_EXCEEDED:
 917                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 918                 if (icmph->code == ICMP_EXC_FRAGTIME)
 919                         goto out;
 920                 break;
 921         }
 922
 923         /*
 924          *      Throw it at our lower layers
 925          *
 926          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 927          *                header.
 928          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 929          *                transport layer.
 930          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 931          *                transport layer.
 932          */
 933
 934         /*
 935          *      Check the other end isn't violating RFC 1122. Some routers send
 936          *      bogus responses to broadcast frames. If you see this message
 937          *      first check your netmask matches at both ends, if it does then
 938          *      get the other vendor to fix their kit.
 939          */
 940
 941         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 942             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 943                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 944                                      &ip_hdr(skb)->saddr,
 945                                      icmph->type, icmph->code,
 946                                      &iph->daddr, skb->dev->name);
 947                 goto out;
 948         }
 949
 950         icmp_socket_deliver(skb, info);
 951
 952 out:
 953         return true;
 954 out_err:
 955         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 956         return false;
 957 }
 958
 959
 960 /*
 961  *      Handle ICMP_REDIRECT.
 962  */
 963
 964 static bool icmp_redirect(struct sk_buff *skb)
 965 {
 966         if (skb->len < sizeof(struct iphdr)) {
 967                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 968                 return false;
 969         }
 970
 971         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 972                 /* there aught to be a stat */
 973                 return false;
 974         }
 975
 976         icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 977         return true;
 978 }
 979
 980 /*
 981  *      Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
 982  *
 983  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 984  *                requests.
 985  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 986  *                included in the reply.
 987  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 988  *                echo requests, MUST have default=NOT.
 989  *      RFC 8335: 8 MUST have a config option to enable/disable ICMP
 990  *                Extended Echo Functionality, MUST be disabled by default
 991  *      See also WRT handling of options once they are done and working.
 992  */
 993
 994 static bool icmp_echo(struct sk_buff *skb)
 995 {
 996         struct icmp_bxm icmp_param;
 997         struct net *net;
 998
 999         net = dev_net(skb_dst(skb)->dev);
1000         /* should there be an ICMP stat for ignored echos? */
1001         if (net->ipv4.sysctl_icmp_echo_ignore_all)
1002                 return true;
1003
1004         icmp_param.data.icmph      = *icmp_hdr(skb);
1005         icmp_param.skb             = skb;
1006         icmp_param.offset          = 0;
1007         icmp_param.data_len        = skb->len;
1008         icmp_param.head_len        = sizeof(struct icmphdr);
1009
1010         if (icmp_param.data.icmph.type == ICMP_ECHO)
1011                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1012         else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1013                 return true;
1014
1015         icmp_reply(&icmp_param, skb);
1016         return true;
1017 }
1018
1019 /*      Helper for icmp_echo and icmpv6_echo_reply.
1020  *      Searches for net_device that matches PROBE interface identifier
1021  *              and builds PROBE reply message in icmphdr.
1022  *
1023  *      Returns false if PROBE responses are disabled via sysctl
1024  */
1025
1026 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1027 {
1028         struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1029         struct icmp_ext_echo_iio *iio, _iio;
1030         struct net *net = dev_net(skb->dev);
1031         struct net_device *dev;
1032         char buff[IFNAMSIZ];
1033         u16 ident_len;
1034         u8 status;
1035
1036         if (!net->ipv4.sysctl_icmp_echo_enable_probe)
1037                 return false;
1038
1039         /* We currently only support probing interfaces on the proxy node
1040          * Check to ensure L-bit is set
1041          */
1042         if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1043                 return false;
1044         /* Clear status bits in reply message */
1045         icmphdr->un.echo.sequence &= htons(0xFF00);
1046         if (icmphdr->type == ICMP_EXT_ECHO)
1047                 icmphdr->type = ICMP_EXT_ECHOREPLY;
1048         else
1049                 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1050         ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1051         /* Size of iio is class_type dependent.
1052          * Only check header here and assign length based on ctype in the switch statement
1053          */
1054         iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1055         if (!ext_hdr || !iio)
1056                 goto send_mal_query;
1057         if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1058             ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1059                 goto send_mal_query;
1060         ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1061         iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1062                                  sizeof(iio->extobj_hdr) + ident_len, &_iio);
1063         if (!iio)
1064                 goto send_mal_query;
1065
1066         status = 0;
1067         dev = NULL;
1068         switch (iio->extobj_hdr.class_type) {
1069         case ICMP_EXT_ECHO_CTYPE_NAME:
1070                 if (ident_len >= IFNAMSIZ)
1071                         goto send_mal_query;
1072                 memset(buff, 0, sizeof(buff));
1073                 memcpy(buff, &iio->ident.name, ident_len);
1074                 dev = dev_get_by_name(net, buff);
1075                 break;
1076         case ICMP_EXT_ECHO_CTYPE_INDEX:
1077                 if (ident_len != sizeof(iio->ident.ifindex))
1078                         goto send_mal_query;
1079                 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1080                 break;
1081         case ICMP_EXT_ECHO_CTYPE_ADDR:
1082                 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1083                     ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1084                                  iio->ident.addr.ctype3_hdr.addrlen)
1085                         goto send_mal_query;
1086                 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1087                 case ICMP_AFI_IP:
1088                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1089                                 goto send_mal_query;
1090                         dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1091                         break;
1092 #if IS_ENABLED(CONFIG_IPV6)
1093                 case ICMP_AFI_IP6:
1094                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1095                                 goto send_mal_query;
1096                         dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1097                         dev_hold(dev);
1098                         break;
1099 #endif
1100                 default:
1101                         goto send_mal_query;
1102                 }
1103                 break;
1104         default:
1105                 goto send_mal_query;
1106         }
1107         if (!dev) {
1108                 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1109                 return true;
1110         }
1111         /* Fill bits in reply message */
1112         if (dev->flags & IFF_UP)
1113                 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1114         if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1115                 status |= ICMP_EXT_ECHOREPLY_IPV4;
1116         if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1117                 status |= ICMP_EXT_ECHOREPLY_IPV6;
1118         dev_put(dev);
1119         icmphdr->un.echo.sequence |= htons(status);
1120         return true;
1121 send_mal_query:
1122         icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1123         return true;
1124 }
1125 EXPORT_SYMBOL_GPL(icmp_build_probe);
1126
1127 /*
1128  *      Handle ICMP Timestamp requests.
1129  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1130  *                SHOULD be in the kernel for minimum random latency.
1131  *                MUST be accurate to a few minutes.
1132  *                MUST be updated at least at 15Hz.
1133  */
1134 static bool icmp_timestamp(struct sk_buff *skb)
1135 {
1136         struct icmp_bxm icmp_param;
1137         /*
1138          *      Too short.
1139          */
1140         if (skb->len < 4)
1141                 goto out_err;
1142
1143         /*
1144          *      Fill in the current time as ms since midnight UT:
1145          */
1146         icmp_param.data.times[1] = inet_current_timestamp();
1147         icmp_param.data.times[2] = icmp_param.data.times[1];
1148
1149         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1150
1151         icmp_param.data.icmph      = *icmp_hdr(skb);
1152         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1153         icmp_param.data.icmph.code = 0;
1154         icmp_param.skb             = skb;
1155         icmp_param.offset          = 0;
1156         icmp_param.data_len        = 0;
1157         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
1158         icmp_reply(&icmp_param, skb);
1159         return true;
1160
1161 out_err:
1162         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1163         return false;
1164 }
1165
1166 static bool icmp_discard(struct sk_buff *skb)
1167 {
1168         /* pretend it was a success */
1169         return true;
1170 }
1171
1172 /*
1173  *      Deal with incoming ICMP packets.
1174  */
1175 int icmp_rcv(struct sk_buff *skb)
1176 {
1177         struct icmphdr *icmph;
1178         struct rtable *rt = skb_rtable(skb);
1179         struct net *net = dev_net(rt->dst.dev);
1180         bool success;
1181
1182         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1183                 struct sec_path *sp = skb_sec_path(skb);
1184                 int nh;
1185
1186                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1187                                  XFRM_STATE_ICMP))
1188                         goto drop;
1189
1190                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1191                         goto drop;
1192
1193                 nh = skb_network_offset(skb);
1194                 skb_set_network_header(skb, sizeof(*icmph));
1195
1196                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1197                         goto drop;
1198
1199                 skb_set_network_header(skb, nh);
1200         }
1201
1202         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1203
1204         if (skb_checksum_simple_validate(skb))
1205                 goto csum_error;
1206
1207         if (!pskb_pull(skb, sizeof(*icmph)))
1208                 goto error;
1209
1210         icmph = icmp_hdr(skb);
1211
1212         ICMPMSGIN_INC_STATS(net, icmph->type);
1213
1214         /* Check for ICMP Extended Echo (PROBE) messages */
1215         if (icmph->type == ICMP_EXT_ECHO) {
1216                 /* We can't use icmp_pointers[].handler() because it is an array of
1217                  * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1218                  */
1219                 success = icmp_echo(skb);
1220                 goto success_check;
1221         }
1222
1223         if (icmph->type == ICMP_EXT_ECHOREPLY) {
1224                 success = ping_rcv(skb);
1225                 goto success_check;
1226         }
1227
1228         /*
1229          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1230          *
1231          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1232          *                discarded.
1233          */
1234         if (icmph->type > NR_ICMP_TYPES)
1235                 goto error;
1236
1237         /*
1238          *      Parse the ICMP message
1239          */
1240
1241         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1242                 /*
1243                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1244                  *        silently ignored (we let user decide with a sysctl).
1245                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1246                  *        discarded if to broadcast/multicast.
1247                  */
1248                 if ((icmph->type == ICMP_ECHO ||
1249                      icmph->type == ICMP_TIMESTAMP) &&
1250                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1251                         goto error;
1252                 }
1253                 if (icmph->type != ICMP_ECHO &&
1254                     icmph->type != ICMP_TIMESTAMP &&
1255                     icmph->type != ICMP_ADDRESS &&
1256                     icmph->type != ICMP_ADDRESSREPLY) {
1257                         goto error;
1258                 }
1259         }
1260
1261         success = icmp_pointers[icmph->type].handler(skb);
1262 success_check:
1263         if (success)  {
1264                 consume_skb(skb);
1265                 return NET_RX_SUCCESS;
1266         }
1267
1268 drop:
1269         kfree_skb(skb);
1270         return NET_RX_DROP;
1271 csum_error:
1272         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1273 error:
1274         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1275         goto drop;
1276 }
1277
1278 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1279 {
1280         struct icmp_extobj_hdr *objh, _objh;
1281         struct icmp_ext_hdr *exth, _exth;
1282         u16 olen;
1283
1284         exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1285         if (!exth)
1286                 return false;
1287         if (exth->version != 2)
1288                 return true;
1289
1290         if (exth->checksum &&
1291             csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1292                 return false;
1293
1294         off += sizeof(_exth);
1295         while (off < skb->len) {
1296                 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1297                 if (!objh)
1298                         return false;
1299
1300                 olen = ntohs(objh->length);
1301                 if (olen < sizeof(_objh))
1302                         return false;
1303
1304                 off += olen;
1305                 if (off > skb->len)
1306                         return false;
1307         }
1308
1309         return true;
1310 }
1311
1312 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1313                            struct sock_ee_data_rfc4884 *out,
1314                            int thlen, int off)
1315 {
1316         int hlen;
1317
1318         /* original datagram headers: end of icmph to payload (skb->data) */
1319         hlen = -skb_transport_offset(skb) - thlen;
1320
1321         /* per rfc 4884: minimal datagram length of 128 bytes */
1322         if (off < 128 || off < hlen)
1323                 return;
1324
1325         /* kernel has stripped headers: return payload offset in bytes */
1326         off -= hlen;
1327         if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1328                 return;
1329
1330         out->len = off;
1331
1332         if (!ip_icmp_error_rfc4884_validate(skb, off))
1333                 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1334 }
1335 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1336
1337 int icmp_err(struct sk_buff *skb, u32 info)
1338 {
1339         struct iphdr *iph = (struct iphdr *)skb->data;
1340         int offset = iph->ihl<<2;
1341         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1342         int type = icmp_hdr(skb)->type;
1343         int code = icmp_hdr(skb)->code;
1344         struct net *net = dev_net(skb->dev);
1345
1346         /*
1347          * Use ping_err to handle all icmp errors except those
1348          * triggered by ICMP_ECHOREPLY which sent from kernel.
1349          */
1350         if (icmph->type != ICMP_ECHOREPLY) {
1351                 ping_err(skb, offset, info);
1352                 return 0;
1353         }
1354
1355         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1356                 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1357         else if (type == ICMP_REDIRECT)
1358                 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1359
1360         return 0;
1361 }
1362
1363 /*
1364  *      This table is the definition of how we handle ICMP.
1365  */
1366 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1367         [ICMP_ECHOREPLY] = {
1368                 .handler = ping_rcv,
1369         },
1370         [1] = {
1371                 .handler = icmp_discard,
1372                 .error = 1,
1373         },
1374         [2] = {
1375                 .handler = icmp_discard,
1376                 .error = 1,
1377         },
1378         [ICMP_DEST_UNREACH] = {
1379                 .handler = icmp_unreach,
1380                 .error = 1,
1381         },
1382         [ICMP_SOURCE_QUENCH] = {
1383                 .handler = icmp_unreach,
1384                 .error = 1,
1385         },
1386         [ICMP_REDIRECT] = {
1387                 .handler = icmp_redirect,
1388                 .error = 1,
1389         },
1390         [6] = {
1391                 .handler = icmp_discard,
1392                 .error = 1,
1393         },
1394         [7] = {
1395                 .handler = icmp_discard,
1396                 .error = 1,
1397         },
1398         [ICMP_ECHO] = {
1399                 .handler = icmp_echo,
1400         },
1401         [9] = {
1402                 .handler = icmp_discard,
1403                 .error = 1,
1404         },
1405         [10] = {
1406                 .handler = icmp_discard,
1407                 .error = 1,
1408         },
1409         [ICMP_TIME_EXCEEDED] = {
1410                 .handler = icmp_unreach,
1411                 .error = 1,
1412         },
1413         [ICMP_PARAMETERPROB] = {
1414                 .handler = icmp_unreach,
1415                 .error = 1,
1416         },
1417         [ICMP_TIMESTAMP] = {
1418                 .handler = icmp_timestamp,
1419         },
1420         [ICMP_TIMESTAMPREPLY] = {
1421                 .handler = icmp_discard,
1422         },
1423         [ICMP_INFO_REQUEST] = {
1424                 .handler = icmp_discard,
1425         },
1426         [ICMP_INFO_REPLY] = {
1427                 .handler = icmp_discard,
1428         },
1429         [ICMP_ADDRESS] = {
1430                 .handler = icmp_discard,
1431         },
1432         [ICMP_ADDRESSREPLY] = {
1433                 .handler = icmp_discard,
1434         },
1435 };
1436
1437 static void __net_exit icmp_sk_exit(struct net *net)
1438 {
1439         int i;
1440
1441         for_each_possible_cpu(i)
1442                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1443         free_percpu(net->ipv4.icmp_sk);
1444         net->ipv4.icmp_sk = NULL;
1445 }
1446
1447 static int __net_init icmp_sk_init(struct net *net)
1448 {
1449         int i, err;
1450
1451         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1452         if (!net->ipv4.icmp_sk)
1453                 return -ENOMEM;
1454
1455         for_each_possible_cpu(i) {
1456                 struct sock *sk;
1457
1458                 err = inet_ctl_sock_create(&sk, PF_INET,
1459                                            SOCK_RAW, IPPROTO_ICMP, net);
1460                 if (err < 0)
1461                         goto fail;
1462
1463                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1464
1465                 /* Enough space for 2 64K ICMP packets, including
1466                  * sk_buff/skb_shared_info struct overhead.
1467                  */
1468                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1469
1470                 /*
1471                  * Speedup sock_wfree()
1472                  */
1473                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1474                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1475         }
1476
1477         /* Control parameters for ECHO replies. */
1478         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1479         net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1480         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1481
1482         /* Control parameter - ignore bogus broadcast responses? */
1483         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1484
1485         /*
1486          *      Configurable global rate limit.
1487          *
1488          *      ratelimit defines tokens/packet consumed for dst->rate_token
1489          *      bucket ratemask defines which icmp types are ratelimited by
1490          *      setting it's bit position.
1491          *
1492          *      default:
1493          *      dest unreachable (3), source quench (4),
1494          *      time exceeded (11), parameter problem (12)
1495          */
1496
1497         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1498         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1499         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1500
1501         return 0;
1502
1503 fail:
1504         icmp_sk_exit(net);
1505         return err;
1506 }
1507
1508 static struct pernet_operations __net_initdata icmp_sk_ops = {
1509        .init = icmp_sk_init,
1510        .exit = icmp_sk_exit,
1511 };
1512
1513 int __init icmp_init(void)
1514 {
1515         return register_pernet_subsys(&icmp_sk_ops);
1516 }