net/ipv6/icmp.c

   1 /*
   2  *      Internet Control Message Protocol (ICMPv6)
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      Based on net/ipv4/icmp.c
   9  *
  10  *      RFC 1885
  11  *
  12  *      This program is free software; you can redistribute it and/or
  13  *      modify it under the terms of the GNU General Public License
  14  *      as published by the Free Software Foundation; either version
  15  *      2 of the License, or (at your option) any later version.
  16  */
  17
  18 /*
  19  *      Changes:
  20  *
  21  *      Andi Kleen              :       exception handling
  22  *      Andi Kleen                      add rate limits. never reply to a icmp.
  23  *                                      add more length checks and other fixes.
  24  *      yoshfuji                :       ensure to sent parameter problem for
  25  *                                      fragments.
  26  *      YOSHIFUJI Hideaki @USAGI:       added sysctl for icmp rate limit.
  27  *      Randy Dunlap and
  28  *      YOSHIFUJI Hideaki @USAGI:       Per-interface statistics support
  29  *      Kazunori MIYAZAWA @USAGI:       change output process to use ip6_append_data
  30  */
  31
  32 #define pr_fmt(fmt) "IPv6: " fmt
  33
  34 #include <linux/module.h>
  35 #include <linux/errno.h>
  36 #include <linux/types.h>
  37 #include <linux/socket.h>
  38 #include <linux/in.h>
  39 #include <linux/kernel.h>
  40 #include <linux/sockios.h>
  41 #include <linux/net.h>
  42 #include <linux/skbuff.h>
  43 #include <linux/init.h>
  44 #include <linux/netfilter.h>
  45 #include <linux/slab.h>
  46
  47 #ifdef CONFIG_SYSCTL
  48 #include <linux/sysctl.h>
  49 #endif
  50
  51 #include <linux/inet.h>
  52 #include <linux/netdevice.h>
  53 #include <linux/icmpv6.h>
  54
  55 #include <net/ip.h>
  56 #include <net/sock.h>
  57
  58 #include <net/ipv6.h>
  59 #include <net/ip6_checksum.h>
  60 #include <net/ping.h>
  61 #include <net/protocol.h>
  62 #include <net/raw.h>
  63 #include <net/rawv6.h>
  64 #include <net/transp_v6.h>
  65 #include <net/ip6_route.h>
  66 #include <net/addrconf.h>
  67 #include <net/icmp.h>
  68 #include <net/xfrm.h>
  69 #include <net/inet_common.h>
  70 #include <net/dsfield.h>
  71 #include <net/l3mdev.h>
  72
  73 #include <asm/uaccess.h>
  74
  75 /*
  76  *      The ICMP socket(s). This is the most convenient way to flow control
  77  *      our ICMP output as well as maintain a clean interface throughout
  78  *      all layers. All Socketless IP sends will soon be gone.
  79  *
  80  *      On SMP we have one ICMP socket per-cpu.
  81  */
  82 static inline struct sock *icmpv6_sk(struct net *net)
  83 {
  84         return net->ipv6.icmp_sk[smp_processor_id()];
  85 }
  86
  87 static void icmpv6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
  88                        u8 type, u8 code, int offset, __be32 info)
  89 {
  90         /* icmpv6_notify checks 8 bytes can be pulled, icmp6hdr is 8 bytes */
  91         struct icmp6hdr *icmp6 = (struct icmp6hdr *) (skb->data + offset);
  92         struct net *net = dev_net(skb->dev);
  93
  94         if (type == ICMPV6_PKT_TOOBIG)
  95                 ip6_update_pmtu(skb, net, info, 0, 0);
  96         else if (type == NDISC_REDIRECT)
  97                 ip6_redirect(skb, net, skb->dev->ifindex, 0);
  98
  99         if (!(type & ICMPV6_INFOMSG_MASK))
 100                 if (icmp6->icmp6_type == ICMPV6_ECHO_REQUEST)
 101                         ping_err(skb, offset, info);
 102 }
 103
 104 static int icmpv6_rcv(struct sk_buff *skb);
 105
 106 static const struct inet6_protocol icmpv6_protocol = {
 107         .handler        =       icmpv6_rcv,
 108         .err_handler    =       icmpv6_err,
 109         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
 110 };
 111
 112 static __inline__ struct sock *icmpv6_xmit_lock(struct net *net)
 113 {
 114         struct sock *sk;
 115
 116         local_bh_disable();
 117
 118         sk = icmpv6_sk(net);
 119         if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 120                 /* This can happen if the output path (f.e. SIT or
 121                  * ip6ip6 tunnel) signals dst_link_failure() for an
 122                  * outgoing ICMP6 packet.
 123                  */
 124                 local_bh_enable();
 125                 return NULL;
 126         }
 127         return sk;
 128 }
 129
 130 static __inline__ void icmpv6_xmit_unlock(struct sock *sk)
 131 {
 132         spin_unlock_bh(&sk->sk_lock.slock);
 133 }
 134
 135 /*
 136  * Figure out, may we reply to this packet with icmp error.
 137  *
 138  * We do not reply, if:
 139  *      - it was icmp error message.
 140  *      - it is truncated, so that it is known, that protocol is ICMPV6
 141  *        (i.e. in the middle of some exthdr)
 142  *
 143  *      --ANK (980726)
 144  */
 145
 146 static bool is_ineligible(const struct sk_buff *skb)
 147 {
 148         int ptr = (u8 *)(ipv6_hdr(skb) + 1) - skb->data;
 149         int len = skb->len - ptr;
 150         __u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 151         __be16 frag_off;
 152
 153         if (len < 0)
 154                 return true;
 155
 156         ptr = ipv6_skip_exthdr(skb, ptr, &nexthdr, &frag_off);
 157         if (ptr < 0)
 158                 return false;
 159         if (nexthdr == IPPROTO_ICMPV6) {
 160                 u8 _type, *tp;
 161                 tp = skb_header_pointer(skb,
 162                         ptr+offsetof(struct icmp6hdr, icmp6_type),
 163                         sizeof(_type), &_type);
 164                 if (!tp || !(*tp & ICMPV6_INFOMSG_MASK))
 165                         return true;
 166         }
 167         return false;
 168 }
 169
 170 /*
 171  * Check the ICMP output rate limit
 172  */
 173 static bool icmpv6_xrlim_allow(struct sock *sk, u8 type,
 174                                struct flowi6 *fl6)
 175 {
 176         struct net *net = sock_net(sk);
 177         struct dst_entry *dst;
 178         bool res = false;
 179
 180         /* Informational messages are not limited. */
 181         if (type & ICMPV6_INFOMSG_MASK)
 182                 return true;
 183
 184         /* Do not limit pmtu discovery, it would break it. */
 185         if (type == ICMPV6_PKT_TOOBIG)
 186                 return true;
 187
 188         /*
 189          * Look up the output route.
 190          * XXX: perhaps the expire for routing entries cloned by
 191          * this lookup should be more aggressive (not longer than timeout).
 192          */
 193         dst = ip6_route_output(net, sk, fl6);
 194         if (dst->error) {
 195                 IP6_INC_STATS(net, ip6_dst_idev(dst),
 196                               IPSTATS_MIB_OUTNOROUTES);
 197         } else if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) {
 198                 res = true;
 199         } else {
 200                 struct rt6_info *rt = (struct rt6_info *)dst;
 201                 int tmo = net->ipv6.sysctl.icmpv6_time;
 202
 203                 /* Give more bandwidth to wider prefixes. */
 204                 if (rt->rt6i_dst.plen < 128)
 205                         tmo >>= ((128 - rt->rt6i_dst.plen)>>5);
 206
 207                 if (icmp_global_allow()) {
 208                         struct inet_peer *peer;
 209
 210                         peer = inet_getpeer_v6(net->ipv6.peers,
 211                                                &fl6->daddr, 1);
 212                         res = inet_peer_xrlim_allow(peer, tmo);
 213                         if (peer)
 214                                 inet_putpeer(peer);
 215                 }
 216         }
 217         dst_release(dst);
 218         return res;
 219 }
 220
 221 /*
 222  *      an inline helper for the "simple" if statement below
 223  *      checks if parameter problem report is caused by an
 224  *      unrecognized IPv6 option that has the Option Type
 225  *      highest-order two bits set to 10
 226  */
 227
 228 static bool opt_unrec(struct sk_buff *skb, __u32 offset)
 229 {
 230         u8 _optval, *op;
 231
 232         offset += skb_network_offset(skb);
 233         op = skb_header_pointer(skb, offset, sizeof(_optval), &_optval);
 234         if (!op)
 235                 return true;
 236         return (*op & 0xC0) == 0x80;
 237 }
 238
 239 int icmpv6_push_pending_frames(struct sock *sk, struct flowi6 *fl6,
 240                                struct icmp6hdr *thdr, int len)
 241 {
 242         struct sk_buff *skb;
 243         struct icmp6hdr *icmp6h;
 244         int err = 0;
 245
 246         skb = skb_peek(&sk->sk_write_queue);
 247         if (!skb)
 248                 goto out;
 249
 250         icmp6h = icmp6_hdr(skb);
 251         memcpy(icmp6h, thdr, sizeof(struct icmp6hdr));
 252         icmp6h->icmp6_cksum = 0;
 253
 254         if (skb_queue_len(&sk->sk_write_queue) == 1) {
 255                 skb->csum = csum_partial(icmp6h,
 256                                         sizeof(struct icmp6hdr), skb->csum);
 257                 icmp6h->icmp6_cksum = csum_ipv6_magic(&fl6->saddr,
 258                                                       &fl6->daddr,
 259                                                       len, fl6->flowi6_proto,
 260                                                       skb->csum);
 261         } else {
 262                 __wsum tmp_csum = 0;
 263
 264                 skb_queue_walk(&sk->sk_write_queue, skb) {
 265                         tmp_csum = csum_add(tmp_csum, skb->csum);
 266                 }
 267
 268                 tmp_csum = csum_partial(icmp6h,
 269                                         sizeof(struct icmp6hdr), tmp_csum);
 270                 icmp6h->icmp6_cksum = csum_ipv6_magic(&fl6->saddr,
 271                                                       &fl6->daddr,
 272                                                       len, fl6->flowi6_proto,
 273                                                       tmp_csum);
 274         }
 275         ip6_push_pending_frames(sk);
 276 out:
 277         return err;
 278 }
 279
 280 struct icmpv6_msg {
 281         struct sk_buff  *skb;
 282         int             offset;
 283         uint8_t         type;
 284 };
 285
 286 static int icmpv6_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 287 {
 288         struct icmpv6_msg *msg = (struct icmpv6_msg *) from;
 289         struct sk_buff *org_skb = msg->skb;
 290         __wsum csum = 0;
 291
 292         csum = skb_copy_and_csum_bits(org_skb, msg->offset + offset,
 293                                       to, len, csum);
 294         skb->csum = csum_block_add(skb->csum, csum, odd);
 295         if (!(msg->type & ICMPV6_INFOMSG_MASK))
 296                 nf_ct_attach(skb, org_skb);
 297         return 0;
 298 }
 299
 300 #if IS_ENABLED(CONFIG_IPV6_MIP6)
 301 static void mip6_addr_swap(struct sk_buff *skb)
 302 {
 303         struct ipv6hdr *iph = ipv6_hdr(skb);
 304         struct inet6_skb_parm *opt = IP6CB(skb);
 305         struct ipv6_destopt_hao *hao;
 306         struct in6_addr tmp;
 307         int off;
 308
 309         if (opt->dsthao) {
 310                 off = ipv6_find_tlv(skb, opt->dsthao, IPV6_TLV_HAO);
 311                 if (likely(off >= 0)) {
 312                         hao = (struct ipv6_destopt_hao *)
 313                                         (skb_network_header(skb) + off);
 314                         tmp = iph->saddr;
 315                         iph->saddr = hao->addr;
 316                         hao->addr = tmp;
 317                 }
 318         }
 319 }
 320 #else
 321 static inline void mip6_addr_swap(struct sk_buff *skb) {}
 322 #endif
 323
 324 static struct dst_entry *icmpv6_route_lookup(struct net *net,
 325                                              struct sk_buff *skb,
 326                                              struct sock *sk,
 327                                              struct flowi6 *fl6)
 328 {
 329         struct dst_entry *dst, *dst2;
 330         struct flowi6 fl2;
 331         int err;
 332
 333         err = ip6_dst_lookup(net, sk, &dst, fl6);
 334         if (err)
 335                 return ERR_PTR(err);
 336
 337         /*
 338          * We won't send icmp if the destination is known
 339          * anycast.
 340          */
 341         if (ipv6_anycast_destination(dst, &fl6->daddr)) {
 342                 net_dbg_ratelimited("icmp6_send: acast source\n");
 343                 dst_release(dst);
 344                 return ERR_PTR(-EINVAL);
 345         }
 346
 347         /* No need to clone since we're just using its address. */
 348         dst2 = dst;
 349
 350         dst = xfrm_lookup(net, dst, flowi6_to_flowi(fl6), sk, 0);
 351         if (!IS_ERR(dst)) {
 352                 if (dst != dst2)
 353                         return dst;
 354         } else {
 355                 if (PTR_ERR(dst) == -EPERM)
 356                         dst = NULL;
 357                 else
 358                         return dst;
 359         }
 360
 361         err = xfrm_decode_session_reverse(skb, flowi6_to_flowi(&fl2), AF_INET6);
 362         if (err)
 363                 goto relookup_failed;
 364
 365         err = ip6_dst_lookup(net, sk, &dst2, &fl2);
 366         if (err)
 367                 goto relookup_failed;
 368
 369         dst2 = xfrm_lookup(net, dst2, flowi6_to_flowi(&fl2), sk, XFRM_LOOKUP_ICMP);
 370         if (!IS_ERR(dst2)) {
 371                 dst_release(dst);
 372                 dst = dst2;
 373         } else {
 374                 err = PTR_ERR(dst2);
 375                 if (err == -EPERM) {
 376                         dst_release(dst);
 377                         return dst2;
 378                 } else
 379                         goto relookup_failed;
 380         }
 381
 382 relookup_failed:
 383         if (dst)
 384                 return dst;
 385         return ERR_PTR(err);
 386 }
 387
 388 /*
 389  *      Send an ICMP message in response to a packet in error
 390  */
 391 static void icmp6_send(struct sk_buff *skb, u8 type, u8 code, __u32 info)
 392 {
 393         struct net *net = dev_net(skb->dev);
 394         struct inet6_dev *idev = NULL;
 395         struct ipv6hdr *hdr = ipv6_hdr(skb);
 396         struct sock *sk;
 397         struct ipv6_pinfo *np;
 398         const struct in6_addr *saddr = NULL;
 399         struct dst_entry *dst;
 400         struct icmp6hdr tmp_hdr;
 401         struct flowi6 fl6;
 402         struct icmpv6_msg msg;
 403         int iif = 0;
 404         int addr_type = 0;
 405         int len;
 406         int hlimit;
 407         int err = 0;
 408         u32 mark = IP6_REPLY_MARK(net, skb->mark);
 409
 410         if ((u8 *)hdr < skb->head ||
 411             (skb_network_header(skb) + sizeof(*hdr)) > skb_tail_pointer(skb))
 412                 return;
 413
 414         /*
 415          *      Make sure we respect the rules
 416          *      i.e. RFC 1885 2.4(e)
 417          *      Rule (e.1) is enforced by not using icmp6_send
 418          *      in any code that processes icmp errors.
 419          */
 420         addr_type = ipv6_addr_type(&hdr->daddr);
 421
 422         if (ipv6_chk_addr(net, &hdr->daddr, skb->dev, 0) ||
 423             ipv6_chk_acast_addr_src(net, skb->dev, &hdr->daddr))
 424                 saddr = &hdr->daddr;
 425
 426         /*
 427          *      Dest addr check
 428          */
 429
 430         if (addr_type & IPV6_ADDR_MULTICAST || skb->pkt_type != PACKET_HOST) {
 431                 if (type != ICMPV6_PKT_TOOBIG &&
 432                     !(type == ICMPV6_PARAMPROB &&
 433                       code == ICMPV6_UNK_OPTION &&
 434                       (opt_unrec(skb, info))))
 435                         return;
 436
 437                 saddr = NULL;
 438         }
 439
 440         addr_type = ipv6_addr_type(&hdr->saddr);
 441
 442         /*
 443          *      Source addr check
 444          */
 445
 446         if (__ipv6_addr_needs_scope_id(addr_type))
 447                 iif = skb->dev->ifindex;
 448
 449         /*
 450          *      Must not send error if the source does not uniquely
 451          *      identify a single node (RFC2463 Section 2.4).
 452          *      We check unspecified / multicast addresses here,
 453          *      and anycast addresses will be checked later.
 454          */
 455         if ((addr_type == IPV6_ADDR_ANY) || (addr_type & IPV6_ADDR_MULTICAST)) {
 456                 net_dbg_ratelimited("icmp6_send: addr_any/mcast source\n");
 457                 return;
 458         }
 459
 460         /*
 461          *      Never answer to a ICMP packet.
 462          */
 463         if (is_ineligible(skb)) {
 464                 net_dbg_ratelimited("icmp6_send: no reply to icmp error\n");
 465                 return;
 466         }
 467
 468         mip6_addr_swap(skb);
 469
 470         memset(&fl6, 0, sizeof(fl6));
 471         fl6.flowi6_proto = IPPROTO_ICMPV6;
 472         fl6.daddr = hdr->saddr;
 473         if (saddr)
 474                 fl6.saddr = *saddr;
 475         fl6.flowi6_mark = mark;
 476         fl6.flowi6_oif = iif;
 477         fl6.fl6_icmp_type = type;
 478         fl6.fl6_icmp_code = code;
 479         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 480
 481         sk = icmpv6_xmit_lock(net);
 482         if (!sk)
 483                 return;
 484         sk->sk_mark = mark;
 485         np = inet6_sk(sk);
 486
 487         if (!icmpv6_xrlim_allow(sk, type, &fl6))
 488                 goto out;
 489
 490         tmp_hdr.icmp6_type = type;
 491         tmp_hdr.icmp6_code = code;
 492         tmp_hdr.icmp6_cksum = 0;
 493         tmp_hdr.icmp6_pointer = htonl(info);
 494
 495         if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr))
 496                 fl6.flowi6_oif = np->mcast_oif;
 497         else if (!fl6.flowi6_oif)
 498                 fl6.flowi6_oif = np->ucast_oif;
 499
 500         if (!fl6.flowi6_oif)
 501                 fl6.flowi6_oif = l3mdev_master_ifindex(skb->dev);
 502
 503         dst = icmpv6_route_lookup(net, skb, sk, &fl6);
 504         if (IS_ERR(dst))
 505                 goto out;
 506
 507         hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
 508
 509         msg.skb = skb;
 510         msg.offset = skb_network_offset(skb);
 511         msg.type = type;
 512
 513         len = skb->len - msg.offset;
 514         len = min_t(unsigned int, len, IPV6_MIN_MTU - sizeof(struct ipv6hdr) - sizeof(struct icmp6hdr));
 515         if (len < 0) {
 516                 net_dbg_ratelimited("icmp: len problem\n");
 517                 goto out_dst_release;
 518         }
 519
 520         rcu_read_lock();
 521         idev = __in6_dev_get(skb->dev);
 522
 523         err = ip6_append_data(sk, icmpv6_getfrag, &msg,
 524                               len + sizeof(struct icmp6hdr),
 525                               sizeof(struct icmp6hdr), hlimit,
 526                               np->tclass, NULL, &fl6, (struct rt6_info *)dst,
 527                               MSG_DONTWAIT, np->dontfrag);
 528         if (err) {
 529                 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTERRORS);
 530                 ip6_flush_pending_frames(sk);
 531         } else {
 532                 err = icmpv6_push_pending_frames(sk, &fl6, &tmp_hdr,
 533                                                  len + sizeof(struct icmp6hdr));
 534         }
 535         rcu_read_unlock();
 536 out_dst_release:
 537         dst_release(dst);
 538 out:
 539         icmpv6_xmit_unlock(sk);
 540 }
 541
 542 /* Slightly more convenient version of icmp6_send.
 543  */
 544 void icmpv6_param_prob(struct sk_buff *skb, u8 code, int pos)
 545 {
 546         icmp6_send(skb, ICMPV6_PARAMPROB, code, pos);
 547         kfree_skb(skb);
 548 }
 549
 550 static void icmpv6_echo_reply(struct sk_buff *skb)
 551 {
 552         struct net *net = dev_net(skb->dev);
 553         struct sock *sk;
 554         struct inet6_dev *idev;
 555         struct ipv6_pinfo *np;
 556         const struct in6_addr *saddr = NULL;
 557         struct icmp6hdr *icmph = icmp6_hdr(skb);
 558         struct icmp6hdr tmp_hdr;
 559         struct flowi6 fl6;
 560         struct icmpv6_msg msg;
 561         struct dst_entry *dst;
 562         int err = 0;
 563         int hlimit;
 564         u8 tclass;
 565         u32 mark = IP6_REPLY_MARK(net, skb->mark);
 566
 567         saddr = &ipv6_hdr(skb)->daddr;
 568
 569         if (!ipv6_unicast_destination(skb) &&
 570             !(net->ipv6.sysctl.anycast_src_echo_reply &&
 571               ipv6_anycast_destination(skb_dst(skb), saddr)))
 572                 saddr = NULL;
 573
 574         memcpy(&tmp_hdr, icmph, sizeof(tmp_hdr));
 575         tmp_hdr.icmp6_type = ICMPV6_ECHO_REPLY;
 576
 577         memset(&fl6, 0, sizeof(fl6));
 578         fl6.flowi6_proto = IPPROTO_ICMPV6;
 579         fl6.daddr = ipv6_hdr(skb)->saddr;
 580         if (saddr)
 581                 fl6.saddr = *saddr;
 582         fl6.flowi6_oif = l3mdev_fib_oif(skb->dev);
 583         fl6.fl6_icmp_type = ICMPV6_ECHO_REPLY;
 584         fl6.flowi6_mark = mark;
 585         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 586
 587         sk = icmpv6_xmit_lock(net);
 588         if (!sk)
 589                 return;
 590         sk->sk_mark = mark;
 591         np = inet6_sk(sk);
 592
 593         if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr))
 594                 fl6.flowi6_oif = np->mcast_oif;
 595         else if (!fl6.flowi6_oif)
 596                 fl6.flowi6_oif = np->ucast_oif;
 597
 598         err = ip6_dst_lookup(net, sk, &dst, &fl6);
 599         if (err)
 600                 goto out;
 601         dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), sk, 0);
 602         if (IS_ERR(dst))
 603                 goto out;
 604
 605         hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
 606
 607         idev = __in6_dev_get(skb->dev);
 608
 609         msg.skb = skb;
 610         msg.offset = 0;
 611         msg.type = ICMPV6_ECHO_REPLY;
 612
 613         tclass = ipv6_get_dsfield(ipv6_hdr(skb));
 614         err = ip6_append_data(sk, icmpv6_getfrag, &msg, skb->len + sizeof(struct icmp6hdr),
 615                                 sizeof(struct icmp6hdr), hlimit, tclass, NULL, &fl6,
 616                                 (struct rt6_info *)dst, MSG_DONTWAIT,
 617                                 np->dontfrag);
 618
 619         if (err) {
 620                 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTERRORS);
 621                 ip6_flush_pending_frames(sk);
 622         } else {
 623                 err = icmpv6_push_pending_frames(sk, &fl6, &tmp_hdr,
 624                                                  skb->len + sizeof(struct icmp6hdr));
 625         }
 626         dst_release(dst);
 627 out:
 628         icmpv6_xmit_unlock(sk);
 629 }
 630
 631 void icmpv6_notify(struct sk_buff *skb, u8 type, u8 code, __be32 info)
 632 {
 633         const struct inet6_protocol *ipprot;
 634         int inner_offset;
 635         __be16 frag_off;
 636         u8 nexthdr;
 637         struct net *net = dev_net(skb->dev);
 638
 639         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 640                 goto out;
 641
 642         nexthdr = ((struct ipv6hdr *)skb->data)->nexthdr;
 643         if (ipv6_ext_hdr(nexthdr)) {
 644                 /* now skip over extension headers */
 645                 inner_offset = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr),
 646                                                 &nexthdr, &frag_off);
 647                 if (inner_offset < 0)
 648                         goto out;
 649         } else {
 650                 inner_offset = sizeof(struct ipv6hdr);
 651         }
 652
 653         /* Checkin header including 8 bytes of inner protocol header. */
 654         if (!pskb_may_pull(skb, inner_offset+8))
 655                 goto out;
 656
 657         /* BUGGG_FUTURE: we should try to parse exthdrs in this packet.
 658            Without this we will not able f.e. to make source routed
 659            pmtu discovery.
 660            Corresponding argument (opt) to notifiers is already added.
 661            --ANK (980726)
 662          */
 663
 664         ipprot = rcu_dereference(inet6_protos[nexthdr]);
 665         if (ipprot && ipprot->err_handler)
 666                 ipprot->err_handler(skb, NULL, type, code, inner_offset, info);
 667
 668         raw6_icmp_error(skb, nexthdr, type, code, inner_offset, info);
 669         return;
 670
 671 out:
 672         ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
 673 }
 674
 675 /*
 676  *      Handle icmp messages
 677  */
 678
 679 static int icmpv6_rcv(struct sk_buff *skb)
 680 {
 681         struct net_device *dev = skb->dev;
 682         struct inet6_dev *idev = __in6_dev_get(dev);
 683         const struct in6_addr *saddr, *daddr;
 684         struct icmp6hdr *hdr;
 685         u8 type;
 686         bool success = false;
 687
 688         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 689                 struct sec_path *sp = skb_sec_path(skb);
 690                 int nh;
 691
 692                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 693                                  XFRM_STATE_ICMP))
 694                         goto drop_no_count;
 695
 696                 if (!pskb_may_pull(skb, sizeof(*hdr) + sizeof(struct ipv6hdr)))
 697                         goto drop_no_count;
 698
 699                 nh = skb_network_offset(skb);
 700                 skb_set_network_header(skb, sizeof(*hdr));
 701
 702                 if (!xfrm6_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 703                         goto drop_no_count;
 704
 705                 skb_set_network_header(skb, nh);
 706         }
 707
 708         ICMP6_INC_STATS_BH(dev_net(dev), idev, ICMP6_MIB_INMSGS);
 709
 710         saddr = &ipv6_hdr(skb)->saddr;
 711         daddr = &ipv6_hdr(skb)->daddr;
 712
 713         if (skb_checksum_validate(skb, IPPROTO_ICMPV6, ip6_compute_pseudo)) {
 714                 net_dbg_ratelimited("ICMPv6 checksum failed [%pI6c > %pI6c]\n",
 715                                     saddr, daddr);
 716                 goto csum_error;
 717         }
 718
 719         if (!pskb_pull(skb, sizeof(*hdr)))
 720                 goto discard_it;
 721
 722         hdr = icmp6_hdr(skb);
 723
 724         type = hdr->icmp6_type;
 725
 726         ICMP6MSGIN_INC_STATS_BH(dev_net(dev), idev, type);
 727
 728         switch (type) {
 729         case ICMPV6_ECHO_REQUEST:
 730                 icmpv6_echo_reply(skb);
 731                 break;
 732
 733         case ICMPV6_ECHO_REPLY:
 734                 success = ping_rcv(skb);
 735                 break;
 736
 737         case ICMPV6_PKT_TOOBIG:
 738                 /* BUGGG_FUTURE: if packet contains rthdr, we cannot update
 739                    standard destination cache. Seems, only "advanced"
 740                    destination cache will allow to solve this problem
 741                    --ANK (980726)
 742                  */
 743                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 744                         goto discard_it;
 745                 hdr = icmp6_hdr(skb);
 746
 747                 /*
 748                  *      Drop through to notify
 749                  */
 750
 751         case ICMPV6_DEST_UNREACH:
 752         case ICMPV6_TIME_EXCEED:
 753         case ICMPV6_PARAMPROB:
 754                 icmpv6_notify(skb, type, hdr->icmp6_code, hdr->icmp6_mtu);
 755                 break;
 756
 757         case NDISC_ROUTER_SOLICITATION:
 758         case NDISC_ROUTER_ADVERTISEMENT:
 759         case NDISC_NEIGHBOUR_SOLICITATION:
 760         case NDISC_NEIGHBOUR_ADVERTISEMENT:
 761         case NDISC_REDIRECT:
 762                 ndisc_rcv(skb);
 763                 break;
 764
 765         case ICMPV6_MGM_QUERY:
 766                 igmp6_event_query(skb);
 767                 break;
 768
 769         case ICMPV6_MGM_REPORT:
 770                 igmp6_event_report(skb);
 771                 break;
 772
 773         case ICMPV6_MGM_REDUCTION:
 774         case ICMPV6_NI_QUERY:
 775         case ICMPV6_NI_REPLY:
 776         case ICMPV6_MLD2_REPORT:
 777         case ICMPV6_DHAAD_REQUEST:
 778         case ICMPV6_DHAAD_REPLY:
 779         case ICMPV6_MOBILE_PREFIX_SOL:
 780         case ICMPV6_MOBILE_PREFIX_ADV:
 781                 break;
 782
 783         default:
 784                 /* informational */
 785                 if (type & ICMPV6_INFOMSG_MASK)
 786                         break;
 787
 788                 net_dbg_ratelimited("icmpv6: msg of unknown type\n");
 789
 790                 /*
 791                  * error of unknown type.
 792                  * must pass to upper level
 793                  */
 794
 795                 icmpv6_notify(skb, type, hdr->icmp6_code, hdr->icmp6_mtu);
 796         }
 797
 798         /* until the v6 path can be better sorted assume failure and
 799          * preserve the status quo behaviour for the rest of the paths to here
 800          */
 801         if (success)
 802                 consume_skb(skb);
 803         else
 804                 kfree_skb(skb);
 805
 806         return 0;
 807
 808 csum_error:
 809         ICMP6_INC_STATS_BH(dev_net(dev), idev, ICMP6_MIB_CSUMERRORS);
 810 discard_it:
 811         ICMP6_INC_STATS_BH(dev_net(dev), idev, ICMP6_MIB_INERRORS);
 812 drop_no_count:
 813         kfree_skb(skb);
 814         return 0;
 815 }
 816
 817 void icmpv6_flow_init(struct sock *sk, struct flowi6 *fl6,
 818                       u8 type,
 819                       const struct in6_addr *saddr,
 820                       const struct in6_addr *daddr,
 821                       int oif)
 822 {
 823         memset(fl6, 0, sizeof(*fl6));
 824         fl6->saddr = *saddr;
 825         fl6->daddr = *daddr;
 826         fl6->flowi6_proto       = IPPROTO_ICMPV6;
 827         fl6->fl6_icmp_type      = type;
 828         fl6->fl6_icmp_code      = 0;
 829         fl6->flowi6_oif         = oif;
 830         security_sk_classify_flow(sk, flowi6_to_flowi(fl6));
 831 }
 832
 833 /*
 834  * Special lock-class for __icmpv6_sk:
 835  */
 836 static struct lock_class_key icmpv6_socket_sk_dst_lock_key;
 837
 838 static int __net_init icmpv6_sk_init(struct net *net)
 839 {
 840         struct sock *sk;
 841         int err, i, j;
 842
 843         net->ipv6.icmp_sk =
 844                 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
 845         if (!net->ipv6.icmp_sk)
 846                 return -ENOMEM;
 847
 848         for_each_possible_cpu(i) {
 849                 err = inet_ctl_sock_create(&sk, PF_INET6,
 850                                            SOCK_RAW, IPPROTO_ICMPV6, net);
 851                 if (err < 0) {
 852                         pr_err("Failed to initialize the ICMP6 control socket (err %d)\n",
 853                                err);
 854                         goto fail;
 855                 }
 856
 857                 net->ipv6.icmp_sk[i] = sk;
 858
 859                 /*
 860                  * Split off their lock-class, because sk->sk_dst_lock
 861                  * gets used from softirqs, which is safe for
 862                  * __icmpv6_sk (because those never get directly used
 863                  * via userspace syscalls), but unsafe for normal sockets.
 864                  */
 865                 lockdep_set_class(&sk->sk_dst_lock,
 866                                   &icmpv6_socket_sk_dst_lock_key);
 867
 868                 /* Enough space for 2 64K ICMP packets, including
 869                  * sk_buff struct overhead.
 870                  */
 871                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
 872         }
 873         return 0;
 874
 875  fail:
 876         for (j = 0; j < i; j++)
 877                 inet_ctl_sock_destroy(net->ipv6.icmp_sk[j]);
 878         kfree(net->ipv6.icmp_sk);
 879         return err;
 880 }
 881
 882 static void __net_exit icmpv6_sk_exit(struct net *net)
 883 {
 884         int i;
 885
 886         for_each_possible_cpu(i) {
 887                 inet_ctl_sock_destroy(net->ipv6.icmp_sk[i]);
 888         }
 889         kfree(net->ipv6.icmp_sk);
 890 }
 891
 892 static struct pernet_operations icmpv6_sk_ops = {
 893         .init = icmpv6_sk_init,
 894         .exit = icmpv6_sk_exit,
 895 };
 896
 897 int __init icmpv6_init(void)
 898 {
 899         int err;
 900
 901         err = register_pernet_subsys(&icmpv6_sk_ops);
 902         if (err < 0)
 903                 return err;
 904
 905         err = -EAGAIN;
 906         if (inet6_add_protocol(&icmpv6_protocol, IPPROTO_ICMPV6) < 0)
 907                 goto fail;
 908
 909         err = inet6_register_icmp_sender(icmp6_send);
 910         if (err)
 911                 goto sender_reg_err;
 912         return 0;
 913
 914 sender_reg_err:
 915         inet6_del_protocol(&icmpv6_protocol, IPPROTO_ICMPV6);
 916 fail:
 917         pr_err("Failed to register ICMP6 protocol\n");
 918         unregister_pernet_subsys(&icmpv6_sk_ops);
 919         return err;
 920 }
 921
 922 void icmpv6_cleanup(void)
 923 {
 924         inet6_unregister_icmp_sender(icmp6_send);
 925         unregister_pernet_subsys(&icmpv6_sk_ops);
 926         inet6_del_protocol(&icmpv6_protocol, IPPROTO_ICMPV6);
 927 }
 928
 929
 930 static const struct icmp6_err {
 931         int err;
 932         int fatal;
 933 } tab_unreach[] = {
 934         {       /* NOROUTE */
 935                 .err    = ENETUNREACH,
 936                 .fatal  = 0,
 937         },
 938         {       /* ADM_PROHIBITED */
 939                 .err    = EACCES,
 940                 .fatal  = 1,
 941         },
 942         {       /* Was NOT_NEIGHBOUR, now reserved */
 943                 .err    = EHOSTUNREACH,
 944                 .fatal  = 0,
 945         },
 946         {       /* ADDR_UNREACH */
 947                 .err    = EHOSTUNREACH,
 948                 .fatal  = 0,
 949         },
 950         {       /* PORT_UNREACH */
 951                 .err    = ECONNREFUSED,
 952                 .fatal  = 1,
 953         },
 954         {       /* POLICY_FAIL */
 955                 .err    = EACCES,
 956                 .fatal  = 1,
 957         },
 958         {       /* REJECT_ROUTE */
 959                 .err    = EACCES,
 960                 .fatal  = 1,
 961         },
 962 };
 963
 964 int icmpv6_err_convert(u8 type, u8 code, int *err)
 965 {
 966         int fatal = 0;
 967
 968         *err = EPROTO;
 969
 970         switch (type) {
 971         case ICMPV6_DEST_UNREACH:
 972                 fatal = 1;
 973                 if (code < ARRAY_SIZE(tab_unreach)) {
 974                         *err  = tab_unreach[code].err;
 975                         fatal = tab_unreach[code].fatal;
 976                 }
 977                 break;
 978
 979         case ICMPV6_PKT_TOOBIG:
 980                 *err = EMSGSIZE;
 981                 break;
 982
 983         case ICMPV6_PARAMPROB:
 984                 *err = EPROTO;
 985                 fatal = 1;
 986                 break;
 987
 988         case ICMPV6_TIME_EXCEED:
 989                 *err = EHOSTUNREACH;
 990                 break;
 991         }
 992
 993         return fatal;
 994 }
 995 EXPORT_SYMBOL(icmpv6_err_convert);
 996
 997 #ifdef CONFIG_SYSCTL
 998 static struct ctl_table ipv6_icmp_table_template[] = {
 999         {
1000                 .procname       = "ratelimit",
1001                 .data           = &init_net.ipv6.sysctl.icmpv6_time,
1002                 .maxlen         = sizeof(int),
1003                 .mode           = 0644,
1004                 .proc_handler   = proc_dointvec_ms_jiffies,
1005         },
1006         { },
1007 };
1008
1009 struct ctl_table * __net_init ipv6_icmp_sysctl_init(struct net *net)
1010 {
1011         struct ctl_table *table;
1012
1013         table = kmemdup(ipv6_icmp_table_template,
1014                         sizeof(ipv6_icmp_table_template),
1015                         GFP_KERNEL);
1016
1017         if (table)
1018                 table[0].data = &net->ipv6.sysctl.icmpv6_time;
1019
1020         return table;
1021 }
1022 #endif