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                 credit--;
 271                 rc = true;
 272         }
 273         WRITE_ONCE(icmp_global.credit, credit);
 274         spin_unlock(&icmp_global.lock);
 275         return rc;
 276 }
 277 EXPORT_SYMBOL(icmp_global_allow);
 278
 279 static bool icmpv4_mask_allow(struct net *net, int type, int code)
 280 {
 281         if (type > NR_ICMP_TYPES)
 282                 return true;
 283
 284         /* Don't limit PMTU discovery. */
 285         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 286                 return true;
 287
 288         /* Limit if icmp type is enabled in ratemask. */
 289         if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 290                 return true;
 291
 292         return false;
 293 }
 294
 295 static bool icmpv4_global_allow(struct net *net, int type, int code)
 296 {
 297         if (icmpv4_mask_allow(net, type, code))
 298                 return true;
 299
 300         if (icmp_global_allow())
 301                 return true;
 302
 303         return false;
 304 }
 305
 306 /*
 307  *      Send an ICMP frame.
 308  */
 309
 310 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 311                                struct flowi4 *fl4, int type, int code)
 312 {
 313         struct dst_entry *dst = &rt->dst;
 314         struct inet_peer *peer;
 315         bool rc = true;
 316         int vif;
 317
 318         if (icmpv4_mask_allow(net, type, code))
 319                 goto out;
 320
 321         /* No rate limit on loopback */
 322         if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 323                 goto out;
 324
 325         vif = l3mdev_master_ifindex(dst->dev);
 326         peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 327         rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
 328         if (peer)
 329                 inet_putpeer(peer);
 330 out:
 331         return rc;
 332 }
 333
 334 /*
 335  *      Maintain the counters used in the SNMP statistics for outgoing ICMP
 336  */
 337 void icmp_out_count(struct net *net, unsigned char type)
 338 {
 339         ICMPMSGOUT_INC_STATS(net, type);
 340         ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 341 }
 342
 343 /*
 344  *      Checksum each fragment, and on the first include the headers and final
 345  *      checksum.
 346  */
 347 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 348                           struct sk_buff *skb)
 349 {
 350         struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 351         __wsum csum;
 352
 353         csum = skb_copy_and_csum_bits(icmp_param->skb,
 354                                       icmp_param->offset + offset,
 355                                       to, len, 0);
 356
 357         skb->csum = csum_block_add(skb->csum, csum, odd);
 358         if (icmp_pointers[icmp_param->data.icmph.type].error)
 359                 nf_ct_attach(skb, icmp_param->skb);
 360         return 0;
 361 }
 362
 363 static void icmp_push_reply(struct icmp_bxm *icmp_param,
 364                             struct flowi4 *fl4,
 365                             struct ipcm_cookie *ipc, struct rtable **rt)
 366 {
 367         struct sock *sk;
 368         struct sk_buff *skb;
 369
 370         sk = icmp_sk(dev_net((*rt)->dst.dev));
 371         if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 372                            icmp_param->data_len+icmp_param->head_len,
 373                            icmp_param->head_len,
 374                            ipc, rt, MSG_DONTWAIT) < 0) {
 375                 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 376                 ip_flush_pending_frames(sk);
 377         } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 378                 struct icmphdr *icmph = icmp_hdr(skb);
 379                 __wsum csum = 0;
 380                 struct sk_buff *skb1;
 381
 382                 skb_queue_walk(&sk->sk_write_queue, skb1) {
 383                         csum = csum_add(csum, skb1->csum);
 384                 }
 385                 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 386                                                  (char *)icmph,
 387                                                  icmp_param->head_len, csum);
 388                 icmph->checksum = csum_fold(csum);
 389                 skb->ip_summed = CHECKSUM_NONE;
 390                 ip_push_pending_frames(sk, fl4);
 391         }
 392 }
 393
 394 /*
 395  *      Driving logic for building and sending ICMP messages.
 396  */
 397
 398 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 399 {
 400         struct ipcm_cookie ipc;
 401         struct rtable *rt = skb_rtable(skb);
 402         struct net *net = dev_net(rt->dst.dev);
 403         struct flowi4 fl4;
 404         struct sock *sk;
 405         struct inet_sock *inet;
 406         __be32 daddr, saddr;
 407         u32 mark = IP4_REPLY_MARK(net, skb->mark);
 408         int type = icmp_param->data.icmph.type;
 409         int code = icmp_param->data.icmph.code;
 410
 411         if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 412                 return;
 413
 414         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 415         local_bh_disable();
 416
 417         /* global icmp_msgs_per_sec */
 418         if (!icmpv4_global_allow(net, type, code))
 419                 goto out_bh_enable;
 420
 421         sk = icmp_xmit_lock(net);
 422         if (!sk)
 423                 goto out_bh_enable;
 424         inet = inet_sk(sk);
 425
 426         icmp_param->data.icmph.checksum = 0;
 427
 428         ipcm_init(&ipc);
 429         inet->tos = ip_hdr(skb)->tos;
 430         sk->sk_mark = mark;
 431         daddr = ipc.addr = ip_hdr(skb)->saddr;
 432         saddr = fib_compute_spec_dst(skb);
 433
 434         if (icmp_param->replyopts.opt.opt.optlen) {
 435                 ipc.opt = &icmp_param->replyopts.opt;
 436                 if (ipc.opt->opt.srr)
 437                         daddr = icmp_param->replyopts.opt.opt.faddr;
 438         }
 439         memset(&fl4, 0, sizeof(fl4));
 440         fl4.daddr = daddr;
 441         fl4.saddr = saddr;
 442         fl4.flowi4_mark = mark;
 443         fl4.flowi4_uid = sock_net_uid(net, NULL);
 444         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 445         fl4.flowi4_proto = IPPROTO_ICMP;
 446         fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 447         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 448         rt = ip_route_output_key(net, &fl4);
 449         if (IS_ERR(rt))
 450                 goto out_unlock;
 451         if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 452                 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 453         ip_rt_put(rt);
 454 out_unlock:
 455         icmp_xmit_unlock(sk);
 456 out_bh_enable:
 457         local_bh_enable();
 458 }
 459
 460 static struct rtable *icmp_route_lookup(struct net *net,
 461                                         struct flowi4 *fl4,
 462                                         struct sk_buff *skb_in,
 463                                         const struct iphdr *iph,
 464                                         __be32 saddr, u8 tos, u32 mark,
 465                                         int type, int code,
 466                                         struct icmp_bxm *param)
 467 {
 468         struct rtable *rt, *rt2;
 469         struct flowi4 fl4_dec;
 470         int err;
 471
 472         memset(fl4, 0, sizeof(*fl4));
 473         fl4->daddr = (param->replyopts.opt.opt.srr ?
 474                       param->replyopts.opt.opt.faddr : iph->saddr);
 475         fl4->saddr = saddr;
 476         fl4->flowi4_mark = mark;
 477         fl4->flowi4_uid = sock_net_uid(net, NULL);
 478         fl4->flowi4_tos = RT_TOS(tos);
 479         fl4->flowi4_proto = IPPROTO_ICMP;
 480         fl4->fl4_icmp_type = type;
 481         fl4->fl4_icmp_code = code;
 482         fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
 483
 484         security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 485         rt = ip_route_output_key_hash(net, fl4, skb_in);
 486         if (IS_ERR(rt))
 487                 return rt;
 488
 489         /* No need to clone since we're just using its address. */
 490         rt2 = rt;
 491
 492         rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 493                                            flowi4_to_flowi(fl4), NULL, 0);
 494         if (!IS_ERR(rt)) {
 495                 if (rt != rt2)
 496                         return rt;
 497         } else if (PTR_ERR(rt) == -EPERM) {
 498                 rt = NULL;
 499         } else
 500                 return rt;
 501
 502         err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 503         if (err)
 504                 goto relookup_failed;
 505
 506         if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
 507                                      fl4_dec.saddr) == RTN_LOCAL) {
 508                 rt2 = __ip_route_output_key(net, &fl4_dec);
 509                 if (IS_ERR(rt2))
 510                         err = PTR_ERR(rt2);
 511         } else {
 512                 struct flowi4 fl4_2 = {};
 513                 unsigned long orefdst;
 514
 515                 fl4_2.daddr = fl4_dec.saddr;
 516                 rt2 = ip_route_output_key(net, &fl4_2);
 517                 if (IS_ERR(rt2)) {
 518                         err = PTR_ERR(rt2);
 519                         goto relookup_failed;
 520                 }
 521                 /* Ugh! */
 522                 orefdst = skb_in->_skb_refdst; /* save old refdst */
 523                 skb_dst_set(skb_in, NULL);
 524                 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 525                                      RT_TOS(tos), rt2->dst.dev);
 526
 527                 dst_release(&rt2->dst);
 528                 rt2 = skb_rtable(skb_in);
 529                 skb_in->_skb_refdst = orefdst; /* restore old refdst */
 530         }
 531
 532         if (err)
 533                 goto relookup_failed;
 534
 535         rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 536                                             flowi4_to_flowi(&fl4_dec), NULL,
 537                                             XFRM_LOOKUP_ICMP);
 538         if (!IS_ERR(rt2)) {
 539                 dst_release(&rt->dst);
 540                 memcpy(fl4, &fl4_dec, sizeof(*fl4));
 541                 rt = rt2;
 542         } else if (PTR_ERR(rt2) == -EPERM) {
 543                 if (rt)
 544                         dst_release(&rt->dst);
 545                 return rt2;
 546         } else {
 547                 err = PTR_ERR(rt2);
 548                 goto relookup_failed;
 549         }
 550         return rt;
 551
 552 relookup_failed:
 553         if (rt)
 554                 return rt;
 555         return ERR_PTR(err);
 556 }
 557
 558 /*
 559  *      Send an ICMP message in response to a situation
 560  *
 561  *      RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
 562  *                MAY send more (we do).
 563  *                      MUST NOT change this header information.
 564  *                      MUST NOT reply to a multicast/broadcast IP address.
 565  *                      MUST NOT reply to a multicast/broadcast MAC address.
 566  *                      MUST reply to only the first fragment.
 567  */
 568
 569 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 570                  const struct ip_options *opt)
 571 {
 572         struct iphdr *iph;
 573         int room;
 574         struct icmp_bxm icmp_param;
 575         struct rtable *rt = skb_rtable(skb_in);
 576         struct ipcm_cookie ipc;
 577         struct flowi4 fl4;
 578         __be32 saddr;
 579         u8  tos;
 580         u32 mark;
 581         struct net *net;
 582         struct sock *sk;
 583
 584         if (!rt)
 585                 goto out;
 586
 587         if (rt->dst.dev)
 588                 net = dev_net(rt->dst.dev);
 589         else if (skb_in->dev)
 590                 net = dev_net(skb_in->dev);
 591         else
 592                 goto out;
 593
 594         /*
 595          *      Find the original header. It is expected to be valid, of course.
 596          *      Check this, icmp_send is called from the most obscure devices
 597          *      sometimes.
 598          */
 599         iph = ip_hdr(skb_in);
 600
 601         if ((u8 *)iph < skb_in->head ||
 602             (skb_network_header(skb_in) + sizeof(*iph)) >
 603             skb_tail_pointer(skb_in))
 604                 goto out;
 605
 606         /*
 607          *      No replies to physical multicast/broadcast
 608          */
 609         if (skb_in->pkt_type != PACKET_HOST)
 610                 goto out;
 611
 612         /*
 613          *      Now check at the protocol level
 614          */
 615         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 616                 goto out;
 617
 618         /*
 619          *      Only reply to fragment 0. We byte re-order the constant
 620          *      mask for efficiency.
 621          */
 622         if (iph->frag_off & htons(IP_OFFSET))
 623                 goto out;
 624
 625         /*
 626          *      If we send an ICMP error to an ICMP error a mess would result..
 627          */
 628         if (icmp_pointers[type].error) {
 629                 /*
 630                  *      We are an error, check if we are replying to an
 631                  *      ICMP error
 632                  */
 633                 if (iph->protocol == IPPROTO_ICMP) {
 634                         u8 _inner_type, *itp;
 635
 636                         itp = skb_header_pointer(skb_in,
 637                                                  skb_network_header(skb_in) +
 638                                                  (iph->ihl << 2) +
 639                                                  offsetof(struct icmphdr,
 640                                                           type) -
 641                                                  skb_in->data,
 642                                                  sizeof(_inner_type),
 643                                                  &_inner_type);
 644                         if (!itp)
 645                                 goto out;
 646
 647                         /*
 648                          *      Assume any unknown ICMP type is an error. This
 649                          *      isn't specified by the RFC, but think about it..
 650                          */
 651                         if (*itp > NR_ICMP_TYPES ||
 652                             icmp_pointers[*itp].error)
 653                                 goto out;
 654                 }
 655         }
 656
 657         /* Needed by both icmp_global_allow and icmp_xmit_lock */
 658         local_bh_disable();
 659
 660         /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 661          * incoming dev is loopback.  If outgoing dev change to not be
 662          * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 663          */
 664         if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 665               !icmpv4_global_allow(net, type, code))
 666                 goto out_bh_enable;
 667
 668         sk = icmp_xmit_lock(net);
 669         if (!sk)
 670                 goto out_bh_enable;
 671
 672         /*
 673          *      Construct source address and options.
 674          */
 675
 676         saddr = iph->daddr;
 677         if (!(rt->rt_flags & RTCF_LOCAL)) {
 678                 struct net_device *dev = NULL;
 679
 680                 rcu_read_lock();
 681                 if (rt_is_input_route(rt) &&
 682                     net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 683                         dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 684
 685                 if (dev)
 686                         saddr = inet_select_addr(dev, iph->saddr,
 687                                                  RT_SCOPE_LINK);
 688                 else
 689                         saddr = 0;
 690                 rcu_read_unlock();
 691         }
 692
 693         tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 694                                            IPTOS_PREC_INTERNETCONTROL) :
 695                                           iph->tos;
 696         mark = IP4_REPLY_MARK(net, skb_in->mark);
 697
 698         if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 699                 goto out_unlock;
 700
 701
 702         /*
 703          *      Prepare data for ICMP header.
 704          */
 705
 706         icmp_param.data.icmph.type       = type;
 707         icmp_param.data.icmph.code       = code;
 708         icmp_param.data.icmph.un.gateway = info;
 709         icmp_param.data.icmph.checksum   = 0;
 710         icmp_param.skb    = skb_in;
 711         icmp_param.offset = skb_network_offset(skb_in);
 712         inet_sk(sk)->tos = tos;
 713         sk->sk_mark = mark;
 714         ipcm_init(&ipc);
 715         ipc.addr = iph->saddr;
 716         ipc.opt = &icmp_param.replyopts.opt;
 717
 718         rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 719                                type, code, &icmp_param);
 720         if (IS_ERR(rt))
 721                 goto out_unlock;
 722
 723         /* peer icmp_ratelimit */
 724         if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 725                 goto ende;
 726
 727         /* RFC says return as much as we can without exceeding 576 bytes. */
 728
 729         room = dst_mtu(&rt->dst);
 730         if (room > 576)
 731                 room = 576;
 732         room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 733         room -= sizeof(struct icmphdr);
 734
 735         icmp_param.data_len = skb_in->len - icmp_param.offset;
 736         if (icmp_param.data_len > room)
 737                 icmp_param.data_len = room;
 738         icmp_param.head_len = sizeof(struct icmphdr);
 739
 740         icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 741 ende:
 742         ip_rt_put(rt);
 743 out_unlock:
 744         icmp_xmit_unlock(sk);
 745 out_bh_enable:
 746         local_bh_enable();
 747 out:;
 748 }
 749 EXPORT_SYMBOL(__icmp_send);
 750
 751
 752 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 753 {
 754         const struct iphdr *iph = (const struct iphdr *) skb->data;
 755         const struct net_protocol *ipprot;
 756         int protocol = iph->protocol;
 757
 758         /* Checkin full IP header plus 8 bytes of protocol to
 759          * avoid additional coding at protocol handlers.
 760          */
 761         if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 762                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 763                 return;
 764         }
 765
 766         raw_icmp_error(skb, protocol, info);
 767
 768         ipprot = rcu_dereference(inet_protos[protocol]);
 769         if (ipprot && ipprot->err_handler)
 770                 ipprot->err_handler(skb, info);
 771 }
 772
 773 static bool icmp_tag_validation(int proto)
 774 {
 775         bool ok;
 776
 777         rcu_read_lock();
 778         ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 779         rcu_read_unlock();
 780         return ok;
 781 }
 782
 783 /*
 784  *      Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 785  *      ICMP_PARAMETERPROB.
 786  */
 787
 788 static bool icmp_unreach(struct sk_buff *skb)
 789 {
 790         const struct iphdr *iph;
 791         struct icmphdr *icmph;
 792         struct net *net;
 793         u32 info = 0;
 794
 795         net = dev_net(skb_dst(skb)->dev);
 796
 797         /*
 798          *      Incomplete header ?
 799          *      Only checks for the IP header, there should be an
 800          *      additional check for longer headers in upper levels.
 801          */
 802
 803         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 804                 goto out_err;
 805
 806         icmph = icmp_hdr(skb);
 807         iph   = (const struct iphdr *)skb->data;
 808
 809         if (iph->ihl < 5) /* Mangled header, drop. */
 810                 goto out_err;
 811
 812         switch (icmph->type) {
 813         case ICMP_DEST_UNREACH:
 814                 switch (icmph->code & 15) {
 815                 case ICMP_NET_UNREACH:
 816                 case ICMP_HOST_UNREACH:
 817                 case ICMP_PROT_UNREACH:
 818                 case ICMP_PORT_UNREACH:
 819                         break;
 820                 case ICMP_FRAG_NEEDED:
 821                         /* for documentation of the ip_no_pmtu_disc
 822                          * values please see
 823                          * Documentation/networking/ip-sysctl.txt
 824                          */
 825                         switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 826                         default:
 827                                 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 828                                                     &iph->daddr);
 829                                 break;
 830                         case 2:
 831                                 goto out;
 832                         case 3:
 833                                 if (!icmp_tag_validation(iph->protocol))
 834                                         goto out;
 835                                 /* fall through */
 836                         case 0:
 837                                 info = ntohs(icmph->un.frag.mtu);
 838                         }
 839                         break;
 840                 case ICMP_SR_FAILED:
 841                         net_dbg_ratelimited("%pI4: Source Route Failed\n",
 842                                             &iph->daddr);
 843                         break;
 844                 default:
 845                         break;
 846                 }
 847                 if (icmph->code > NR_ICMP_UNREACH)
 848                         goto out;
 849                 break;
 850         case ICMP_PARAMETERPROB:
 851                 info = ntohl(icmph->un.gateway) >> 24;
 852                 break;
 853         case ICMP_TIME_EXCEEDED:
 854                 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 855                 if (icmph->code == ICMP_EXC_FRAGTIME)
 856                         goto out;
 857                 break;
 858         }
 859
 860         /*
 861          *      Throw it at our lower layers
 862          *
 863          *      RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 864          *                header.
 865          *      RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 866          *                transport layer.
 867          *      RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 868          *                transport layer.
 869          */
 870
 871         /*
 872          *      Check the other end isn't violating RFC 1122. Some routers send
 873          *      bogus responses to broadcast frames. If you see this message
 874          *      first check your netmask matches at both ends, if it does then
 875          *      get the other vendor to fix their kit.
 876          */
 877
 878         if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 879             inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 880                 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 881                                      &ip_hdr(skb)->saddr,
 882                                      icmph->type, icmph->code,
 883                                      &iph->daddr, skb->dev->name);
 884                 goto out;
 885         }
 886
 887         icmp_socket_deliver(skb, info);
 888
 889 out:
 890         return true;
 891 out_err:
 892         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 893         return false;
 894 }
 895
 896
 897 /*
 898  *      Handle ICMP_REDIRECT.
 899  */
 900
 901 static bool icmp_redirect(struct sk_buff *skb)
 902 {
 903         if (skb->len < sizeof(struct iphdr)) {
 904                 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 905                 return false;
 906         }
 907
 908         if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 909                 /* there aught to be a stat */
 910                 return false;
 911         }
 912
 913         icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 914         return true;
 915 }
 916
 917 /*
 918  *      Handle ICMP_ECHO ("ping") requests.
 919  *
 920  *      RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 921  *                requests.
 922  *      RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 923  *                included in the reply.
 924  *      RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 925  *                echo requests, MUST have default=NOT.
 926  *      See also WRT handling of options once they are done and working.
 927  */
 928
 929 static bool icmp_echo(struct sk_buff *skb)
 930 {
 931         struct net *net;
 932
 933         net = dev_net(skb_dst(skb)->dev);
 934         if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 935                 struct icmp_bxm icmp_param;
 936
 937                 icmp_param.data.icmph      = *icmp_hdr(skb);
 938                 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 939                 icmp_param.skb             = skb;
 940                 icmp_param.offset          = 0;
 941                 icmp_param.data_len        = skb->len;
 942                 icmp_param.head_len        = sizeof(struct icmphdr);
 943                 icmp_reply(&icmp_param, skb);
 944         }
 945         /* should there be an ICMP stat for ignored echos? */
 946         return true;
 947 }
 948
 949 /*
 950  *      Handle ICMP Timestamp requests.
 951  *      RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 952  *                SHOULD be in the kernel for minimum random latency.
 953  *                MUST be accurate to a few minutes.
 954  *                MUST be updated at least at 15Hz.
 955  */
 956 static bool icmp_timestamp(struct sk_buff *skb)
 957 {
 958         struct icmp_bxm icmp_param;
 959         /*
 960          *      Too short.
 961          */
 962         if (skb->len < 4)
 963                 goto out_err;
 964
 965         /*
 966          *      Fill in the current time as ms since midnight UT:
 967          */
 968         icmp_param.data.times[1] = inet_current_timestamp();
 969         icmp_param.data.times[2] = icmp_param.data.times[1];
 970
 971         BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
 972
 973         icmp_param.data.icmph      = *icmp_hdr(skb);
 974         icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 975         icmp_param.data.icmph.code = 0;
 976         icmp_param.skb             = skb;
 977         icmp_param.offset          = 0;
 978         icmp_param.data_len        = 0;
 979         icmp_param.head_len        = sizeof(struct icmphdr) + 12;
 980         icmp_reply(&icmp_param, skb);
 981         return true;
 982
 983 out_err:
 984         __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 985         return false;
 986 }
 987
 988 static bool icmp_discard(struct sk_buff *skb)
 989 {
 990         /* pretend it was a success */
 991         return true;
 992 }
 993
 994 /*
 995  *      Deal with incoming ICMP packets.
 996  */
 997 int icmp_rcv(struct sk_buff *skb)
 998 {
 999         struct icmphdr *icmph;
1000         struct rtable *rt = skb_rtable(skb);
1001         struct net *net = dev_net(rt->dst.dev);
1002         bool success;
1003
1004         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1005                 struct sec_path *sp = skb_sec_path(skb);
1006                 int nh;
1007
1008                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1009                                  XFRM_STATE_ICMP))
1010                         goto drop;
1011
1012                 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1013                         goto drop;
1014
1015                 nh = skb_network_offset(skb);
1016                 skb_set_network_header(skb, sizeof(*icmph));
1017
1018                 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1019                         goto drop;
1020
1021                 skb_set_network_header(skb, nh);
1022         }
1023
1024         __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1025
1026         if (skb_checksum_simple_validate(skb))
1027                 goto csum_error;
1028
1029         if (!pskb_pull(skb, sizeof(*icmph)))
1030                 goto error;
1031
1032         icmph = icmp_hdr(skb);
1033
1034         ICMPMSGIN_INC_STATS(net, icmph->type);
1035         /*
1036          *      18 is the highest 'known' ICMP type. Anything else is a mystery
1037          *
1038          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1039          *                discarded.
1040          */
1041         if (icmph->type > NR_ICMP_TYPES)
1042                 goto error;
1043
1044
1045         /*
1046          *      Parse the ICMP message
1047          */
1048
1049         if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1050                 /*
1051                  *      RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1052                  *        silently ignored (we let user decide with a sysctl).
1053                  *      RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1054                  *        discarded if to broadcast/multicast.
1055                  */
1056                 if ((icmph->type == ICMP_ECHO ||
1057                      icmph->type == ICMP_TIMESTAMP) &&
1058                     net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1059                         goto error;
1060                 }
1061                 if (icmph->type != ICMP_ECHO &&
1062                     icmph->type != ICMP_TIMESTAMP &&
1063                     icmph->type != ICMP_ADDRESS &&
1064                     icmph->type != ICMP_ADDRESSREPLY) {
1065                         goto error;
1066                 }
1067         }
1068
1069         success = icmp_pointers[icmph->type].handler(skb);
1070
1071         if (success)  {
1072                 consume_skb(skb);
1073                 return NET_RX_SUCCESS;
1074         }
1075
1076 drop:
1077         kfree_skb(skb);
1078         return NET_RX_DROP;
1079 csum_error:
1080         __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1081 error:
1082         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1083         goto drop;
1084 }
1085
1086 int icmp_err(struct sk_buff *skb, u32 info)
1087 {
1088         struct iphdr *iph = (struct iphdr *)skb->data;
1089         int offset = iph->ihl<<2;
1090         struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1091         int type = icmp_hdr(skb)->type;
1092         int code = icmp_hdr(skb)->code;
1093         struct net *net = dev_net(skb->dev);
1094
1095         /*
1096          * Use ping_err to handle all icmp errors except those
1097          * triggered by ICMP_ECHOREPLY which sent from kernel.
1098          */
1099         if (icmph->type != ICMP_ECHOREPLY) {
1100                 ping_err(skb, offset, info);
1101                 return 0;
1102         }
1103
1104         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1105                 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1106         else if (type == ICMP_REDIRECT)
1107                 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1108
1109         return 0;
1110 }
1111
1112 /*
1113  *      This table is the definition of how we handle ICMP.
1114  */
1115 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1116         [ICMP_ECHOREPLY] = {
1117                 .handler = ping_rcv,
1118         },
1119         [1] = {
1120                 .handler = icmp_discard,
1121                 .error = 1,
1122         },
1123         [2] = {
1124                 .handler = icmp_discard,
1125                 .error = 1,
1126         },
1127         [ICMP_DEST_UNREACH] = {
1128                 .handler = icmp_unreach,
1129                 .error = 1,
1130         },
1131         [ICMP_SOURCE_QUENCH] = {
1132                 .handler = icmp_unreach,
1133                 .error = 1,
1134         },
1135         [ICMP_REDIRECT] = {
1136                 .handler = icmp_redirect,
1137                 .error = 1,
1138         },
1139         [6] = {
1140                 .handler = icmp_discard,
1141                 .error = 1,
1142         },
1143         [7] = {
1144                 .handler = icmp_discard,
1145                 .error = 1,
1146         },
1147         [ICMP_ECHO] = {
1148                 .handler = icmp_echo,
1149         },
1150         [9] = {
1151                 .handler = icmp_discard,
1152                 .error = 1,
1153         },
1154         [10] = {
1155                 .handler = icmp_discard,
1156                 .error = 1,
1157         },
1158         [ICMP_TIME_EXCEEDED] = {
1159                 .handler = icmp_unreach,
1160                 .error = 1,
1161         },
1162         [ICMP_PARAMETERPROB] = {
1163                 .handler = icmp_unreach,
1164                 .error = 1,
1165         },
1166         [ICMP_TIMESTAMP] = {
1167                 .handler = icmp_timestamp,
1168         },
1169         [ICMP_TIMESTAMPREPLY] = {
1170                 .handler = icmp_discard,
1171         },
1172         [ICMP_INFO_REQUEST] = {
1173                 .handler = icmp_discard,
1174         },
1175         [ICMP_INFO_REPLY] = {
1176                 .handler = icmp_discard,
1177         },
1178         [ICMP_ADDRESS] = {
1179                 .handler = icmp_discard,
1180         },
1181         [ICMP_ADDRESSREPLY] = {
1182                 .handler = icmp_discard,
1183         },
1184 };
1185
1186 static void __net_exit icmp_sk_exit(struct net *net)
1187 {
1188         int i;
1189
1190         for_each_possible_cpu(i)
1191                 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1192         free_percpu(net->ipv4.icmp_sk);
1193         net->ipv4.icmp_sk = NULL;
1194 }
1195
1196 static int __net_init icmp_sk_init(struct net *net)
1197 {
1198         int i, err;
1199
1200         net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1201         if (!net->ipv4.icmp_sk)
1202                 return -ENOMEM;
1203
1204         for_each_possible_cpu(i) {
1205                 struct sock *sk;
1206
1207                 err = inet_ctl_sock_create(&sk, PF_INET,
1208                                            SOCK_RAW, IPPROTO_ICMP, net);
1209                 if (err < 0)
1210                         goto fail;
1211
1212                 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1213
1214                 /* Enough space for 2 64K ICMP packets, including
1215                  * sk_buff/skb_shared_info struct overhead.
1216                  */
1217                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1218
1219                 /*
1220                  * Speedup sock_wfree()
1221                  */
1222                 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1223                 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1224         }
1225
1226         /* Control parameters for ECHO replies. */
1227         net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1228         net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1229
1230         /* Control parameter - ignore bogus broadcast responses? */
1231         net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1232
1233         /*
1234          *      Configurable global rate limit.
1235          *
1236          *      ratelimit defines tokens/packet consumed for dst->rate_token
1237          *      bucket ratemask defines which icmp types are ratelimited by
1238          *      setting it's bit position.
1239          *
1240          *      default:
1241          *      dest unreachable (3), source quench (4),
1242          *      time exceeded (11), parameter problem (12)
1243          */
1244
1245         net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1246         net->ipv4.sysctl_icmp_ratemask = 0x1818;
1247         net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1248
1249         return 0;
1250
1251 fail:
1252         icmp_sk_exit(net);
1253         return err;
1254 }
1255
1256 static struct pernet_operations __net_initdata icmp_sk_ops = {
1257        .init = icmp_sk_init,
1258        .exit = icmp_sk_exit,
1259 };
1260
1261 int __init icmp_init(void)
1262 {
1263         return register_pernet_subsys(&icmp_sk_ops);
1264 }