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 #include <net/addrconf.h>
  96
  97 /*
  98  *      Build xmit assembly blocks
  99  */
 100
 101 struct icmp_bxm {
 102         struct sk_buff *skb;
 103         int offset;
 104         int data_len;
 105
 106         struct {
 107                 struct icmphdr icmph;
 108                 __be32         times[3];
 109         } data;
 110         int head_len;
 111         struct ip_options_data replyopts;
 112 };
 113
 114 /* An array of errno for error messages from dest unreach. */
 115 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 116
 117 const struct icmp_err icmp_err_convert[] = {
 118         {
 119                 .errno = ENETUNREACH,   /* ICMP_NET_UNREACH */
 120                 .fatal = 0,
 121         },
 122         {
 123                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNREACH */
 124                 .fatal = 0,
 125         },
 126         {
 127                 .errno = ENOPROTOOPT    /* ICMP_PROT_UNREACH */,
 128                 .fatal = 1,
 129         },
 130         {
 131                 .errno = ECONNREFUSED,  /* ICMP_PORT_UNREACH */
 132                 .fatal = 1,
 133         },
 134         {
 135                 .errno = EMSGSIZE,      /* ICMP_FRAG_NEEDED */
 136                 .fatal = 0,
 137         },
 138         {
 139                 .errno = EOPNOTSUPP,    /* ICMP_SR_FAILED */
 140                 .fatal = 0,
 141         },
 142         {
 143                 .errno = ENETUNREACH,   /* ICMP_NET_UNKNOWN */
 144                 .fatal = 1,
 145         },
 146         {
 147                 .errno = EHOSTDOWN,     /* ICMP_HOST_UNKNOWN */
 148                 .fatal = 1,
 149         },
 150         {
 151                 .errno = ENONET,        /* ICMP_HOST_ISOLATED */
 152                 .fatal = 1,
 153         },
 154         {
 155                 .errno = ENETUNREACH,   /* ICMP_NET_ANO */
 156                 .fatal = 1,
 157         },
 158         {
 159                 .errno = EHOSTUNREACH,  /* ICMP_HOST_ANO */
 160                 .fatal = 1,
 161         },
 162         {
 163                 .errno = ENETUNREACH,   /* ICMP_NET_UNR_TOS */
 164                 .fatal = 0,
 165         },
 166         {
 167                 .errno = EHOSTUNREACH,  /* ICMP_HOST_UNR_TOS */
 168                 .fatal = 0,
 169         },
 170         {
 171                 .errno = EHOSTUNREACH,  /* ICMP_PKT_FILTERED */
 172                 .fatal = 1,
 173         },
 174         {
 175                 .errno = EHOSTUNREACH,  /* ICMP_PREC_VIOLATION */
 176                 .fatal = 1,
 177         },
 178         {
 179                 .errno = EHOSTUNREACH,  /* ICMP_PREC_CUTOFF */
 180                 .fatal = 1,
 181         },
 182 };
 183 EXPORT_SYMBOL(icmp_err_convert);
 184
 185 /*
 186  *      ICMP control array. This specifies what to do with each ICMP.
 187  */
 188
 189 struct icmp_control {
 190         enum skb_drop_reason (*handler)(struct sk_buff *skb);
 191         short   error;          /* This ICMP is classed as an error message */
 192 };
 193
 194 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 195
 196 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
 197
 198 /* Called with BH disabled */
 199 static inline struct sock *icmp_xmit_lock(struct net *net)
 200 {
 201         struct sock *sk;
 202
 203         sk = this_cpu_read(ipv4_icmp_sk);
 204
 205         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 206                 /* This can happen if the output path signals a
 207                  * dst_link_failure() for an outgoing ICMP packet.
 208                  */
 209                 return NULL;
 210         }
 211         sock_net_set(sk, net);
 212         return sk;
 213 }
 214
 215 static inline void icmp_xmit_unlock(struct sock *sk)
 216 {
 217         sock_net_set(sk, &init_net);
 218         spin_unlock(&sk->sk_lock.slock);
 219 }
 220
 221 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 222 int sysctl_icmp_msgs_burst __read_mostly = 50;
 223
 224 static struct {
 225         spinlock_t      lock;
 226         u32             credit;
 227         u32             stamp;
 228 } icmp_global = {
 229         .lock           = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 230 };
 231
 232 /**
 233  * icmp_global_allow - Are we allowed to send one more ICMP message ?
 234  *
 235  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 236  * Returns false if we reached the limit and can not send another packet.
 237  * Note: called with BH disabled
 238  */
 239 bool icmp_global_allow(void)
 240 {
 241         u32 credit, delta, incr = 0, now = (u32)jiffies;
 242         bool rc = false;
 243
 244         /* Check if token bucket is empty and cannot be refilled
 245          * without taking the spinlock. The READ_ONCE() are paired
 246          * with the following WRITE_ONCE() in this same function.
 247          */
 248         if (!READ_ONCE(icmp_global.credit)) {
 249                 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 250                 if (delta < HZ / 50)
 251                         return false;
 252         }
 253
 254         spin_lock(&icmp_global.lock);
 255         delta = min_t(u32, now - icmp_global.stamp, HZ);
 256         if (delta >= HZ / 50) {
 257                 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
 258                 if (incr)
 259                         WRITE_ONCE(icmp_global.stamp, now);
 260         }
 261         credit = min_t(u32, icmp_global.credit + incr,
 262                        READ_ONCE(sysctl_icmp_msgs_burst));
 263         if (credit) {
 264                 /* We want to use a credit of one in average, but need to randomize
 265                  * it for security reasons.
 266                  */
 267                 credit = max_t(int, credit - prandom_u32_max(3), 0);
 268                 rc = true;
 269         }
 270         WRITE_ONCE(icmp_global.credit, credit);
 271         spin_unlock(&icmp_global.lock);
 272         return rc;
 273 }
 274 EXPORT_SYMBOL(icmp_global_allow);
 275
 276 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 277 {
 278         if (type > NR_ICMP_TYPES)
 279                 return true;
 280
 281         /* Don't limit PMTU discovery. */
 282         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 283                 return true;
 284
 285         /* Limit if icmp type is enabled in ratemask. */
 286         if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
 287                 return true;
 288
 289         return false;
 290 }
 291
 292 static bool icmpv4_global_allow(struct net *net, int type, int code)
 293 {
 294         if (icmpv4_mask_allow(net, type, code))
 295                 return true;
 296
 297         if (icmp_global_allow())
 298                 return true;
 299
 300         return false;
 301 }
 302
 303 /*
 304  *      Send an ICMP frame.
 305  */
 306
 307 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 308                                struct flowi4 *fl4, int type, int code)
 309 {
 310         struct dst_entry *dst = &rt->dst;
 311         struct inet_peer *peer;
 312         bool rc = true;
 313         int vif;
 314
 315         if (icmpv4_mask_allow(net, type, code))
 316                 goto out;
 317
 318         /* No rate limit on loopback */
 319         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 320                 goto out;
 321
 322         vif = l3mdev_master_ifindex(dst->dev);
 323         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 324         rc = inet_peer_xrlim_allow(peer,
 325                                    READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
 326         if (peer)
 327                 inet_putpeer(peer);
 328 out:
 329         return rc;
 330 }
 331
 332 /*
 333  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 334  */
 335 void icmp_out_count(struct net *net, unsigned char type)
 336 {
 337         ICMPMSGOUT_INC_STATS(net, type);
 338         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 339 }
 340
 341 /*
 342  *      Checksum each fragment, and on the first include the headers and final
 343  *      checksum.
 344  */
 345 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 346                           struct sk_buff *skb)
 347 {
 348         struct icmp_bxm *icmp_param = from;
 349         __wsum csum;
 350
 351         csum = skb_copy_and_csum_bits(icmp_param->skb,
 352                                       icmp_param->offset + offset,
 353                                       to, len);
 354
 355         skb->csum = csum_block_add(skb->csum, csum, odd);
 356         if (icmp_pointers[icmp_param->data.icmph.type].error)
 357                 nf_ct_attach(skb, icmp_param->skb);
 358         return 0;
 359 }
 360
 361 static void icmp_push_reply(struct sock *sk,
 362                             struct icmp_bxm *icmp_param,
 363                             struct flowi4 *fl4,
 364                             struct ipcm_cookie *ipc, struct rtable **rt)
 365 {
 366         struct sk_buff *skb;
 367
 368         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 369                            icmp_param->data_len+icmp_param->head_len,
 370                            icmp_param->head_len,
 371                            ipc, rt, MSG_DONTWAIT) < 0) {
 372                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 373                 ip_flush_pending_frames(sk);
 374         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 375                 struct icmphdr *icmph = icmp_hdr(skb);
 376                 __wsum csum;
 377                 struct sk_buff *skb1;
 378
 379                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 380                                                  (char *)icmph,
 381                                                  icmp_param->head_len);
 382                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 383                         csum = csum_add(csum, skb1->csum);
 384                 }
 385                 icmph->checksum = csum_fold(csum);
 386                 skb->ip_summed = CHECKSUM_NONE;
 387                 ip_push_pending_frames(sk, fl4);
 388         }
 389 }
 390
 391 /*
 392  *      Driving logic for building and sending ICMP messages.
 393  */
 394
 395 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 396 {
 397         struct ipcm_cookie ipc;
 398         struct rtable *rt = skb_rtable(skb);
 399         struct net *net = dev_net(rt->dst.dev);
 400         struct flowi4 fl4;
 401         struct sock *sk;
 402         struct inet_sock *inet;
 403         __be32 daddr, saddr;
 404         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 405         int type = icmp_param->data.icmph.type;
 406         int code = icmp_param->data.icmph.code;
 407
 408         if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 409                 return;
 410
 411         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 412         local_bh_disable();
 413
 414         /* global icmp_msgs_per_sec */
 415         if (!icmpv4_global_allow(net, type, code))
 416                 goto out_bh_enable;
 417
 418         sk = icmp_xmit_lock(net);
 419         if (!sk)
 420                 goto out_bh_enable;
 421         inet = inet_sk(sk);
 422
 423         icmp_param->data.icmph.checksum = 0;
 424
 425         ipcm_init(&ipc);
 426         inet->tos = ip_hdr(skb)->tos;
 427         ipc.sockc.mark = mark;
 428         daddr = ipc.addr = ip_hdr(skb)->saddr;
 429         saddr = fib_compute_spec_dst(skb);
 430
 431         if (icmp_param->replyopts.opt.opt.optlen) {
 432                 ipc.opt = &icmp_param->replyopts.opt;
 433                 if (ipc.opt->opt.srr)
 434                         daddr = icmp_param->replyopts.opt.opt.faddr;
 435         }
 436         memset(&fl4, 0, sizeof(fl4));
 437         fl4.daddr = daddr;
 438         fl4.saddr = saddr;
 439         fl4.flowi4_mark = mark;
 440         fl4.flowi4_uid = sock_net_uid(net, NULL);
 441         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 442         fl4.flowi4_proto = IPPROTO_ICMP;
 443         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 444         security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 445         rt = ip_route_output_key(net, &fl4);
 446         if (IS_ERR(rt))
 447                 goto out_unlock;
 448         if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 449                 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
 450         ip_rt_put(rt);
 451 out_unlock:
 452         icmp_xmit_unlock(sk);
 453 out_bh_enable:
 454         local_bh_enable();
 455 }
 456
 457 /*
 458  * The device used for looking up which routing table to use for sending an ICMP
 459  * error is preferably the source whenever it is set, which should ensure the
 460  * icmp error can be sent to the source host, else lookup using the routing
 461  * table of the destination device, else use the main routing table (index 0).
 462  */
 463 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 464 {
 465         struct net_device *route_lookup_dev = NULL;
 466
 467         if (skb->dev)
 468                 route_lookup_dev = skb->dev;
 469         else if (skb_dst(skb))
 470                 route_lookup_dev = skb_dst(skb)->dev;
 471         return route_lookup_dev;
 472 }
 473
 474 static struct rtable *icmp_route_lookup(struct net *net,
 475                                         struct flowi4 *fl4,
 476                                         struct sk_buff *skb_in,
 477                                         const struct iphdr *iph,
 478                                         __be32 saddr, u8 tos, u32 mark,
 479                                         int type, int code,
 480                                         struct icmp_bxm *param)
 481 {
 482         struct net_device *route_lookup_dev;
 483         struct rtable *rt, *rt2;
 484         struct flowi4 fl4_dec;
 485         int err;
 486
 487         memset(fl4, 0, sizeof(*fl4));
 488         fl4->daddr = (param->replyopts.opt.opt.srr ?
 489                       param->replyopts.opt.opt.faddr : iph->saddr);
 490         fl4->saddr = saddr;
 491         fl4->flowi4_mark = mark;
 492         fl4->flowi4_uid = sock_net_uid(net, NULL);
 493         fl4->flowi4_tos = RT_TOS(tos);
 494         fl4->flowi4_proto = IPPROTO_ICMP;
 495         fl4->fl4_icmp_type = type;
 496         fl4->fl4_icmp_code = code;
 497         route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 498         fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 499
 500         security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
 501         rt = ip_route_output_key_hash(net, fl4, skb_in);
 502         if (IS_ERR(rt))
 503                 return rt;
 504
 505         /* No need to clone since we're just using its address. */
 506         rt2 = rt;
 507
 508         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 509                                            flowi4_to_flowi(fl4), NULL, 0);
 510         if (!IS_ERR(rt)) {
 511                 if (rt != rt2)
 512                         return rt;
 513         } else if (PTR_ERR(rt) == -EPERM) {
 514                 rt = NULL;
 515         } else
 516                 return rt;
 517
 518         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 519         if (err)
 520                 goto relookup_failed;
 521
 522         if (inet_addr_type_dev_table(net, route_lookup_dev,
 523                                      fl4_dec.saddr) == RTN_LOCAL) {
 524                 rt2 = __ip_route_output_key(net, &fl4_dec);
 525                 if (IS_ERR(rt2))
 526                         err = PTR_ERR(rt2);
 527         } else {
 528                 struct flowi4 fl4_2 = {};
 529                 unsigned long orefdst;
 530
 531                 fl4_2.daddr = fl4_dec.saddr;
 532                 rt2 = ip_route_output_key(net, &fl4_2);
 533                 if (IS_ERR(rt2)) {
 534                         err = PTR_ERR(rt2);
 535                         goto relookup_failed;
 536                 }
 537                 /* Ugh! */
 538                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 539                 skb_dst_set(skb_in, NULL);
 540                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 541                                      RT_TOS(tos), rt2->dst.dev);
 542
 543                 dst_release(&rt2->dst);
 544                 rt2 = skb_rtable(skb_in);
 545                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 546         }
 547
 548         if (err)
 549                 goto relookup_failed;
 550
 551         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 552                                             flowi4_to_flowi(&fl4_dec), NULL,
 553                                             XFRM_LOOKUP_ICMP);
 554         if (!IS_ERR(rt2)) {
 555                 dst_release(&rt->dst);
 556                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 557                 rt = rt2;
 558         } else if (PTR_ERR(rt2) == -EPERM) {
 559                 if (rt)
 560                         dst_release(&rt->dst);
 561                 return rt2;
 562         } else {
 563                 err = PTR_ERR(rt2);
 564                 goto relookup_failed;
 565         }
 566         return rt;
 567
 568 relookup_failed:
 569         if (rt)
 570                 return rt;
 571         return ERR_PTR(err);
 572 }
 573
 574 /*
 575  *      Send an ICMP message in response to a situation
 576  *
 577  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 578  *                MAY send more (we do).
 579  *                      MUST NOT change this header information.
 580  *                      MUST NOT reply to a multicast/broadcast IP address.
 581  *                      MUST NOT reply to a multicast/broadcast MAC address.
 582  *                      MUST reply to only the first fragment.
 583  */
 584
 585 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 586                  const struct ip_options *opt)
 587 {
 588         struct iphdr *iph;
 589         int room;
 590         struct icmp_bxm icmp_param;
 591         struct rtable *rt = skb_rtable(skb_in);
 592         struct ipcm_cookie ipc;
 593         struct flowi4 fl4;
 594         __be32 saddr;
 595         u8  tos;
 596         u32 mark;
 597         struct net *net;
 598         struct sock *sk;
 599
 600         if (!rt)
 601                 goto out;
 602
 603         if (rt->dst.dev)
 604                 net = dev_net(rt->dst.dev);
 605         else if (skb_in->dev)
 606                 net = dev_net(skb_in->dev);
 607         else
 608                 goto out;
 609
 610         /*
 611          *      Find the original header. It is expected to be valid, of course.
 612          *      Check this, icmp_send is called from the most obscure devices
 613          *      sometimes.
 614          */
 615         iph = ip_hdr(skb_in);
 616
 617         if ((u8 *)iph < skb_in->head ||
 618             (skb_network_header(skb_in) + sizeof(*iph)) >
 619             skb_tail_pointer(skb_in))
 620                 goto out;
 621
 622         /*
 623          *      No replies to physical multicast/broadcast
 624          */
 625         if (skb_in->pkt_type != PACKET_HOST)
 626                 goto out;
 627
 628         /*
 629          *      Now check at the protocol level
 630          */
 631         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 632                 goto out;
 633
 634         /*
 635          *      Only reply to fragment 0. We byte re-order the constant
 636          *      mask for efficiency.
 637          */
 638         if (iph->frag_off & htons(IP_OFFSET))
 639                 goto out;
 640
 641         /*
 642          *      If we send an ICMP error to an ICMP error a mess would result..
 643          */
 644         if (icmp_pointers[type].error) {
 645                 /*
 646                  *      We are an error, check if we are replying to an
 647                  *      ICMP error
 648                  */
 649                 if (iph->protocol == IPPROTO_ICMP) {
 650                         u8 _inner_type, *itp;
 651
 652                         itp = skb_header_pointer(skb_in,
 653                                                  skb_network_header(skb_in) +
 654                                                  (iph->ihl << 2) +
 655                                                  offsetof(struct icmphdr,
 656                                                           type) -
 657                                                  skb_in->data,
 658                                                  sizeof(_inner_type),
 659                                                  &_inner_type);
 660                         if (!itp)
 661                                 goto out;
 662
 663                         /*
 664                          *      Assume any unknown ICMP type is an error. This
 665                          *      isn't specified by the RFC, but think about it..
 666                          */
 667                         if (*itp > NR_ICMP_TYPES ||
 668                             icmp_pointers[*itp].error)
 669                                 goto out;
 670                 }
 671         }
 672
 673         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 674         local_bh_disable();
 675
 676         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 677          * incoming dev is loopback.  If outgoing dev change to not be
 678          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 679          */
 680         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 681               !icmpv4_global_allow(net, type, code))
 682                 goto out_bh_enable;
 683
 684         sk = icmp_xmit_lock(net);
 685         if (!sk)
 686                 goto out_bh_enable;
 687
 688         /*
 689          *      Construct source address and options.
 690          */
 691
 692         saddr = iph->daddr;
 693         if (!(rt->rt_flags & RTCF_LOCAL)) {
 694                 struct net_device *dev = NULL;
 695
 696                 rcu_read_lock();
 697                 if (rt_is_input_route(rt) &&
 698                     READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
 699                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 700
 701                 if (dev)
 702                         saddr = inet_select_addr(dev, iph->saddr,
 703                                                  RT_SCOPE_LINK);
 704                 else
 705                         saddr = 0;
 706                 rcu_read_unlock();
 707         }
 708
 709         tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 710                                            IPTOS_PREC_INTERNETCONTROL) :
 711                                            iph->tos;
 712         mark = IP4_REPLY_MARK(net, skb_in->mark);
 713
 714         if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 715                 goto out_unlock;
 716
 717
 718         /*
 719          *      Prepare data for ICMP header.
 720          */
 721
 722         icmp_param.data.icmph.type       = type;
 723         icmp_param.data.icmph.code       = code;
 724         icmp_param.data.icmph.un.gateway = info;
 725         icmp_param.data.icmph.checksum   = 0;
 726         icmp_param.skb    = skb_in;
 727         icmp_param.offset = skb_network_offset(skb_in);
 728         inet_sk(sk)->tos = tos;
 729         ipcm_init(&ipc);
 730         ipc.addr = iph->saddr;
 731         ipc.opt = &icmp_param.replyopts.opt;
 732         ipc.sockc.mark = mark;
 733
 734         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 735                                type, code, &icmp_param);
 736         if (IS_ERR(rt))
 737                 goto out_unlock;
 738
 739         /* peer icmp_ratelimit */
 740         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 741                 goto ende;
 742
 743         /* RFC says return as much as we can without exceeding 576 bytes. */
 744
 745         room = dst_mtu(&rt->dst);
 746         if (room > 576)
 747                 room = 576;
 748         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 749         room -= sizeof(struct icmphdr);
 750         /* Guard against tiny mtu. We need to include at least one
 751          * IP network header for this message to make any sense.
 752          */
 753         if (room <= (int)sizeof(struct iphdr))
 754                 goto ende;
 755
 756         icmp_param.data_len = skb_in->len - icmp_param.offset;
 757         if (icmp_param.data_len > room)
 758                 icmp_param.data_len = room;
 759         icmp_param.head_len = sizeof(struct icmphdr);
 760
 761         /* if we don't have a source address at this point, fall back to the
 762          * dummy address instead of sending out a packet with a source address
 763          * of 0.0.0.0
 764          */
 765         if (!fl4.saddr)
 766                 fl4.saddr = htonl(INADDR_DUMMY);
 767
 768         icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
 769 ende:
 770         ip_rt_put(rt);
 771 out_unlock:
 772         icmp_xmit_unlock(sk);
 773 out_bh_enable:
 774         local_bh_enable();
 775 out:;
 776 }
 777 EXPORT_SYMBOL(__icmp_send);
 778
 779 #if IS_ENABLED(CONFIG_NF_NAT)
 780 #include <net/netfilter/nf_conntrack.h>
 781 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 782 {
 783         struct sk_buff *cloned_skb = NULL;
 784         struct ip_options opts = { 0 };
 785         enum ip_conntrack_info ctinfo;
 786         struct nf_conn *ct;
 787         __be32 orig_ip;
 788
 789         ct = nf_ct_get(skb_in, &ctinfo);
 790         if (!ct || !(ct->status & IPS_SRC_NAT)) {
 791                 __icmp_send(skb_in, type, code, info, &opts);
 792                 return;
 793         }
 794
 795         if (skb_shared(skb_in))
 796                 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 797
 798         if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 799             (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 800             skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 801             skb_network_offset(skb_in) + sizeof(struct iphdr))))
 802                 goto out;
 803
 804         orig_ip = ip_hdr(skb_in)->saddr;
 805         ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 806         __icmp_send(skb_in, type, code, info, &opts);
 807         ip_hdr(skb_in)->saddr = orig_ip;
 808 out:
 809         consume_skb(cloned_skb);
 810 }
 811 EXPORT_SYMBOL(icmp_ndo_send);
 812 #endif
 813
 814 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 815 {
 816         const struct iphdr *iph = (const struct iphdr *)skb->data;
 817         const struct net_protocol *ipprot;
 818         int protocol = iph->protocol;
 819
 820         /* Checkin full IP header plus 8 bytes of protocol to
 821          * avoid additional coding at protocol handlers.
 822          */
 823         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 824                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 825                 return;
 826         }
 827
 828         raw_icmp_error(skb, protocol, info);
 829
 830         ipprot = rcu_dereference(inet_protos[protocol]);
 831         if (ipprot && ipprot->err_handler)
 832                 ipprot->err_handler(skb, info);
 833 }
 834
 835 static bool icmp_tag_validation(int proto)
 836 {
 837         bool ok;
 838
 839         rcu_read_lock();
 840         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 841         rcu_read_unlock();
 842         return ok;
 843 }
 844
 845 /*
 846  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 847  *      ICMP_PARAMETERPROB.
 848  */
 849
 850 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
 851 {
 852         enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
 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                 reason = SKB_DROP_REASON_IP_INHDR;
 874                 goto out_err;
 875         }
 876
 877         switch (icmph->type) {
 878         case ICMP_DEST_UNREACH:
 879                 switch (icmph->code & 15) {
 880                 case ICMP_NET_UNREACH:
 881                 case ICMP_HOST_UNREACH:
 882                 case ICMP_PROT_UNREACH:
 883                 case ICMP_PORT_UNREACH:
 884                         break;
 885                 case ICMP_FRAG_NEEDED:
 886                         /* for documentation of the ip_no_pmtu_disc
 887                          * values please see
 888                          * Documentation/networking/ip-sysctl.rst
 889                          */
 890                         switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
 891                         default:
 892                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 893                                                     &iph->daddr);
 894                                 break;
 895                         case 2:
 896                                 goto out;
 897                         case 3:
 898                                 if (!icmp_tag_validation(iph->protocol))
 899                                         goto out;
 900                                 fallthrough;
 901                         case 0:
 902                                 info = ntohs(icmph->un.frag.mtu);
 903                         }
 904                         break;
 905                 case ICMP_SR_FAILED:
 906                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 907                                             &iph->daddr);
 908                         break;
 909                 default:
 910                         break;
 911                 }
 912                 if (icmph->code > NR_ICMP_UNREACH)
 913                         goto out;
 914                 break;
 915         case ICMP_PARAMETERPROB:
 916                 info = ntohl(icmph->un.gateway) >> 24;
 917                 break;
 918         case ICMP_TIME_EXCEEDED:
 919                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 920                 if (icmph->code == ICMP_EXC_FRAGTIME)
 921                         goto out;
 922                 break;
 923         }
 924
 925         /*
 926          *      Throw it at our lower layers
 927          *
 928          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 929          *                header.
 930          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 931          *                transport layer.
 932          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 933          *                transport layer.
 934          */
 935
 936         /*
 937          *      Check the other end isn't violating RFC 1122. Some routers send
 938          *      bogus responses to broadcast frames. If you see this message
 939          *      first check your netmask matches at both ends, if it does then
 940          *      get the other vendor to fix their kit.
 941          */
 942
 943         if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
 944             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 945                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 946                                      &ip_hdr(skb)->saddr,
 947                                      icmph->type, icmph->code,
 948                                      &iph->daddr, skb->dev->name);
 949                 goto out;
 950         }
 951
 952         icmp_socket_deliver(skb, info);
 953
 954 out:
 955         return reason;
 956 out_err:
 957         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 958         return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
 959 }
 960
 961
 962 /*
 963  *      Handle ICMP_REDIRECT.
 964  */
 965
 966 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
 967 {
 968         if (skb->len < sizeof(struct iphdr)) {
 969                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 970                 return SKB_DROP_REASON_PKT_TOO_SMALL;
 971         }
 972
 973         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 974                 /* there aught to be a stat */
 975                 return SKB_DROP_REASON_NOMEM;
 976         }
 977
 978         icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 979         return SKB_NOT_DROPPED_YET;
 980 }
 981
 982 /*
 983  *      Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
 984  *
 985  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 986  *                requests.
 987  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 988  *                included in the reply.
 989  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 990  *                echo requests, MUST have default=NOT.
 991  *      RFC 8335: 8 MUST have a config option to enable/disable ICMP
 992  *                Extended Echo Functionality, MUST be disabled by default
 993  *      See also WRT handling of options once they are done and working.
 994  */
 995
 996 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
 997 {
 998         struct icmp_bxm icmp_param;
 999         struct net *net;
1000
1001         net = dev_net(skb_dst(skb)->dev);
1002         /* should there be an ICMP stat for ignored echos? */
1003         if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1004                 return SKB_NOT_DROPPED_YET;
1005
1006         icmp_param.data.icmph      = *icmp_hdr(skb);
1007         icmp_param.skb             = skb;
1008         icmp_param.offset          = 0;
1009         icmp_param.data_len        = skb->len;
1010         icmp_param.head_len        = sizeof(struct icmphdr);
1011
1012         if (icmp_param.data.icmph.type == ICMP_ECHO)
1013                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1014         else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1015                 return SKB_NOT_DROPPED_YET;
1016
1017         icmp_reply(&icmp_param, skb);
1018         return SKB_NOT_DROPPED_YET;
1019 }
1020
1021 /*      Helper for icmp_echo and icmpv6_echo_reply.
1022  *      Searches for net_device that matches PROBE interface identifier
1023  *              and builds PROBE reply message in icmphdr.
1024  *
1025  *      Returns false if PROBE responses are disabled via sysctl
1026  */
1027
1028 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1029 {
1030         struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1031         struct icmp_ext_echo_iio *iio, _iio;
1032         struct net *net = dev_net(skb->dev);
1033         struct inet6_dev *in6_dev;
1034         struct in_device *in_dev;
1035         struct net_device *dev;
1036         char buff[IFNAMSIZ];
1037         u16 ident_len;
1038         u8 status;
1039
1040         if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1041                 return false;
1042
1043         /* We currently only support probing interfaces on the proxy node
1044          * Check to ensure L-bit is set
1045          */
1046         if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1047                 return false;
1048         /* Clear status bits in reply message */
1049         icmphdr->un.echo.sequence &= htons(0xFF00);
1050         if (icmphdr->type == ICMP_EXT_ECHO)
1051                 icmphdr->type = ICMP_EXT_ECHOREPLY;
1052         else
1053                 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1054         ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1055         /* Size of iio is class_type dependent.
1056          * Only check header here and assign length based on ctype in the switch statement
1057          */
1058         iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1059         if (!ext_hdr || !iio)
1060                 goto send_mal_query;
1061         if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1062             ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1063                 goto send_mal_query;
1064         ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1065         iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1066                                  sizeof(iio->extobj_hdr) + ident_len, &_iio);
1067         if (!iio)
1068                 goto send_mal_query;
1069
1070         status = 0;
1071         dev = NULL;
1072         switch (iio->extobj_hdr.class_type) {
1073         case ICMP_EXT_ECHO_CTYPE_NAME:
1074                 if (ident_len >= IFNAMSIZ)
1075                         goto send_mal_query;
1076                 memset(buff, 0, sizeof(buff));
1077                 memcpy(buff, &iio->ident.name, ident_len);
1078                 dev = dev_get_by_name(net, buff);
1079                 break;
1080         case ICMP_EXT_ECHO_CTYPE_INDEX:
1081                 if (ident_len != sizeof(iio->ident.ifindex))
1082                         goto send_mal_query;
1083                 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1084                 break;
1085         case ICMP_EXT_ECHO_CTYPE_ADDR:
1086                 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1087                     ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1088                                  iio->ident.addr.ctype3_hdr.addrlen)
1089                         goto send_mal_query;
1090                 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1091                 case ICMP_AFI_IP:
1092                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1093                                 goto send_mal_query;
1094                         dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1095                         break;
1096 #if IS_ENABLED(CONFIG_IPV6)
1097                 case ICMP_AFI_IP6:
1098                         if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1099                                 goto send_mal_query;
1100                         dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1101                         dev_hold(dev);
1102                         break;
1103 #endif
1104                 default:
1105                         goto send_mal_query;
1106                 }
1107                 break;
1108         default:
1109                 goto send_mal_query;
1110         }
1111         if (!dev) {
1112                 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1113                 return true;
1114         }
1115         /* Fill bits in reply message */
1116         if (dev->flags & IFF_UP)
1117                 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1118
1119         in_dev = __in_dev_get_rcu(dev);
1120         if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1121                 status |= ICMP_EXT_ECHOREPLY_IPV4;
1122
1123         in6_dev = __in6_dev_get(dev);
1124         if (in6_dev && !list_empty(&in6_dev->addr_list))
1125                 status |= ICMP_EXT_ECHOREPLY_IPV6;
1126
1127         dev_put(dev);
1128         icmphdr->un.echo.sequence |= htons(status);
1129         return true;
1130 send_mal_query:
1131         icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1132         return true;
1133 }
1134 EXPORT_SYMBOL_GPL(icmp_build_probe);
1135
1136 /*
1137  *      Handle ICMP Timestamp requests.
1138  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1139  *                SHOULD be in the kernel for minimum random latency.
1140  *                MUST be accurate to a few minutes.
1141  *                MUST be updated at least at 15Hz.
1142  */
1143 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1144 {
1145         struct icmp_bxm icmp_param;
1146         /*
1147          *      Too short.
1148          */
1149         if (skb->len < 4)
1150                 goto out_err;
1151
1152         /*
1153          *      Fill in the current time as ms since midnight UT:
1154          */
1155         icmp_param.data.times[1] = inet_current_timestamp();
1156         icmp_param.data.times[2] = icmp_param.data.times[1];
1157
1158         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1159
1160         icmp_param.data.icmph      = *icmp_hdr(skb);
1161         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1162         icmp_param.data.icmph.code = 0;
1163         icmp_param.skb             = skb;
1164         icmp_param.offset          = 0;
1165         icmp_param.data_len        = 0;
1166         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
1167         icmp_reply(&icmp_param, skb);
1168         return SKB_NOT_DROPPED_YET;
1169
1170 out_err:
1171         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1172         return SKB_DROP_REASON_PKT_TOO_SMALL;
1173 }
1174
1175 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1176 {
1177         /* pretend it was a success */
1178         return SKB_NOT_DROPPED_YET;
1179 }
1180
1181 /*
1182  *      Deal with incoming ICMP packets.
1183  */
1184 int icmp_rcv(struct sk_buff *skb)
1185 {
1186         enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1187         struct rtable *rt = skb_rtable(skb);
1188         struct net *net = dev_net(rt->dst.dev);
1189         struct icmphdr *icmph;
1190
1191         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1192                 struct sec_path *sp = skb_sec_path(skb);
1193                 int nh;
1194
1195                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1196                                  XFRM_STATE_ICMP)) {
1197                         reason = SKB_DROP_REASON_XFRM_POLICY;
1198                         goto drop;
1199                 }
1200
1201                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1202                         goto drop;
1203
1204                 nh = skb_network_offset(skb);
1205                 skb_set_network_header(skb, sizeof(*icmph));
1206
1207                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1208                                                 skb)) {
1209                         reason = SKB_DROP_REASON_XFRM_POLICY;
1210                         goto drop;
1211                 }
1212
1213                 skb_set_network_header(skb, nh);
1214         }
1215
1216         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1217
1218         if (skb_checksum_simple_validate(skb))
1219                 goto csum_error;
1220
1221         if (!pskb_pull(skb, sizeof(*icmph)))
1222                 goto error;
1223
1224         icmph = icmp_hdr(skb);
1225
1226         ICMPMSGIN_INC_STATS(net, icmph->type);
1227
1228         /* Check for ICMP Extended Echo (PROBE) messages */
1229         if (icmph->type == ICMP_EXT_ECHO) {
1230                 /* We can't use icmp_pointers[].handler() because it is an array of
1231                  * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1232                  */
1233                 reason = icmp_echo(skb);
1234                 goto reason_check;
1235         }
1236
1237         if (icmph->type == ICMP_EXT_ECHOREPLY) {
1238                 reason = ping_rcv(skb);
1239                 goto reason_check;
1240         }
1241
1242         /*
1243          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1244          *
1245          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1246          *                discarded.
1247          */
1248         if (icmph->type > NR_ICMP_TYPES) {
1249                 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1250                 goto error;
1251         }
1252
1253         /*
1254          *      Parse the ICMP message
1255          */
1256
1257         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1258                 /*
1259                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1260                  *        silently ignored (we let user decide with a sysctl).
1261                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1262                  *        discarded if to broadcast/multicast.
1263                  */
1264                 if ((icmph->type == ICMP_ECHO ||
1265                      icmph->type == ICMP_TIMESTAMP) &&
1266                     READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1267                         reason = SKB_DROP_REASON_INVALID_PROTO;
1268                         goto error;
1269                 }
1270                 if (icmph->type != ICMP_ECHO &&
1271                     icmph->type != ICMP_TIMESTAMP &&
1272                     icmph->type != ICMP_ADDRESS &&
1273                     icmph->type != ICMP_ADDRESSREPLY) {
1274                         reason = SKB_DROP_REASON_INVALID_PROTO;
1275                         goto error;
1276                 }
1277         }
1278
1279         reason = icmp_pointers[icmph->type].handler(skb);
1280 reason_check:
1281         if (!reason)  {
1282                 consume_skb(skb);
1283                 return NET_RX_SUCCESS;
1284         }
1285
1286 drop:
1287         kfree_skb_reason(skb, reason);
1288         return NET_RX_DROP;
1289 csum_error:
1290         reason = SKB_DROP_REASON_ICMP_CSUM;
1291         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1292 error:
1293         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1294         goto drop;
1295 }
1296
1297 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1298 {
1299         struct icmp_extobj_hdr *objh, _objh;
1300         struct icmp_ext_hdr *exth, _exth;
1301         u16 olen;
1302
1303         exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1304         if (!exth)
1305                 return false;
1306         if (exth->version != 2)
1307                 return true;
1308
1309         if (exth->checksum &&
1310             csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1311                 return false;
1312
1313         off += sizeof(_exth);
1314         while (off < skb->len) {
1315                 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1316                 if (!objh)
1317                         return false;
1318
1319                 olen = ntohs(objh->length);
1320                 if (olen < sizeof(_objh))
1321                         return false;
1322
1323                 off += olen;
1324                 if (off > skb->len)
1325                         return false;
1326         }
1327
1328         return true;
1329 }
1330
1331 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1332                            struct sock_ee_data_rfc4884 *out,
1333                            int thlen, int off)
1334 {
1335         int hlen;
1336
1337         /* original datagram headers: end of icmph to payload (skb->data) */
1338         hlen = -skb_transport_offset(skb) - thlen;
1339
1340         /* per rfc 4884: minimal datagram length of 128 bytes */
1341         if (off < 128 || off < hlen)
1342                 return;
1343
1344         /* kernel has stripped headers: return payload offset in bytes */
1345         off -= hlen;
1346         if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1347                 return;
1348
1349         out->len = off;
1350
1351         if (!ip_icmp_error_rfc4884_validate(skb, off))
1352                 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1353 }
1354 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1355
1356 int icmp_err(struct sk_buff *skb, u32 info)
1357 {
1358         struct iphdr *iph = (struct iphdr *)skb->data;
1359         int offset = iph->ihl<<2;
1360         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1361         int type = icmp_hdr(skb)->type;
1362         int code = icmp_hdr(skb)->code;
1363         struct net *net = dev_net(skb->dev);
1364
1365         /*
1366          * Use ping_err to handle all icmp errors except those
1367          * triggered by ICMP_ECHOREPLY which sent from kernel.
1368          */
1369         if (icmph->type != ICMP_ECHOREPLY) {
1370                 ping_err(skb, offset, info);
1371                 return 0;
1372         }
1373
1374         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1375                 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1376         else if (type == ICMP_REDIRECT)
1377                 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1378
1379         return 0;
1380 }
1381
1382 /*
1383  *      This table is the definition of how we handle ICMP.
1384  */
1385 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1386         [ICMP_ECHOREPLY] = {
1387                 .handler = ping_rcv,
1388         },
1389         [1] = {
1390                 .handler = icmp_discard,
1391                 .error = 1,
1392         },
1393         [2] = {
1394                 .handler = icmp_discard,
1395                 .error = 1,
1396         },
1397         [ICMP_DEST_UNREACH] = {
1398                 .handler = icmp_unreach,
1399                 .error = 1,
1400         },
1401         [ICMP_SOURCE_QUENCH] = {
1402                 .handler = icmp_unreach,
1403                 .error = 1,
1404         },
1405         [ICMP_REDIRECT] = {
1406                 .handler = icmp_redirect,
1407                 .error = 1,
1408         },
1409         [6] = {
1410                 .handler = icmp_discard,
1411                 .error = 1,
1412         },
1413         [7] = {
1414                 .handler = icmp_discard,
1415                 .error = 1,
1416         },
1417         [ICMP_ECHO] = {
1418                 .handler = icmp_echo,
1419         },
1420         [9] = {
1421                 .handler = icmp_discard,
1422                 .error = 1,
1423         },
1424         [10] = {
1425                 .handler = icmp_discard,
1426                 .error = 1,
1427         },
1428         [ICMP_TIME_EXCEEDED] = {
1429                 .handler = icmp_unreach,
1430                 .error = 1,
1431         },
1432         [ICMP_PARAMETERPROB] = {
1433                 .handler = icmp_unreach,
1434                 .error = 1,
1435         },
1436         [ICMP_TIMESTAMP] = {
1437                 .handler = icmp_timestamp,
1438         },
1439         [ICMP_TIMESTAMPREPLY] = {
1440                 .handler = icmp_discard,
1441         },
1442         [ICMP_INFO_REQUEST] = {
1443                 .handler = icmp_discard,
1444         },
1445         [ICMP_INFO_REPLY] = {
1446                 .handler = icmp_discard,
1447         },
1448         [ICMP_ADDRESS] = {
1449                 .handler = icmp_discard,
1450         },
1451         [ICMP_ADDRESSREPLY] = {
1452                 .handler = icmp_discard,
1453         },
1454 };
1455
1456 static int __net_init icmp_sk_init(struct net *net)
1457 {
1458         /* Control parameters for ECHO replies. */
1459         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1460         net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1461         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1462
1463         /* Control parameter - ignore bogus broadcast responses? */
1464         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1465
1466         /*
1467          *      Configurable global rate limit.
1468          *
1469          *      ratelimit defines tokens/packet consumed for dst->rate_token
1470          *      bucket ratemask defines which icmp types are ratelimited by
1471          *      setting it's bit position.
1472          *
1473          *      default:
1474          *      dest unreachable (3), source quench (4),
1475          *      time exceeded (11), parameter problem (12)
1476          */
1477
1478         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1479         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1480         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1481
1482         return 0;
1483 }
1484
1485 static struct pernet_operations __net_initdata icmp_sk_ops = {
1486        .init = icmp_sk_init,
1487 };
1488
1489 int __init icmp_init(void)
1490 {
1491         int err, i;
1492
1493         for_each_possible_cpu(i) {
1494                 struct sock *sk;
1495
1496                 err = inet_ctl_sock_create(&sk, PF_INET,
1497                                            SOCK_RAW, IPPROTO_ICMP, &init_net);
1498                 if (err < 0)
1499                         return err;
1500
1501                 per_cpu(ipv4_icmp_sk, i) = sk;
1502
1503                 /* Enough space for 2 64K ICMP packets, including
1504                  * sk_buff/skb_shared_info struct overhead.
1505                  */
1506                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1507
1508                 /*
1509                  * Speedup sock_wfree()
1510                  */
1511                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1512                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1513         }
1514         return register_pernet_subsys(&icmp_sk_ops);
1515 }