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         struct sockcm_cookie sockc_unused = {0};
 404         int iif = 0;
 405         int addr_type = 0;
 406         int len;
 407         int hlimit;
 408         int err = 0;
 409         u32 mark = IP6_REPLY_MARK(net, skb->mark);
 410
 411         if ((u8 *)hdr < skb->head ||
 412             (skb_network_header(skb) + sizeof(*hdr)) > skb_tail_pointer(skb))
 413                 return;
 414
 415         /*
 416          *      Make sure we respect the rules
 417          *      i.e. RFC 1885 2.4(e)
 418          *      Rule (e.1) is enforced by not using icmp6_send
 419          *      in any code that processes icmp errors.
 420          */
 421         addr_type = ipv6_addr_type(&hdr->daddr);
 422
 423         if (ipv6_chk_addr(net, &hdr->daddr, skb->dev, 0) ||
 424             ipv6_chk_acast_addr_src(net, skb->dev, &hdr->daddr))
 425                 saddr = &hdr->daddr;
 426
 427         /*
 428          *      Dest addr check
 429          */
 430
 431         if (addr_type & IPV6_ADDR_MULTICAST || skb->pkt_type != PACKET_HOST) {
 432                 if (type != ICMPV6_PKT_TOOBIG &&
 433                     !(type == ICMPV6_PARAMPROB &&
 434                       code == ICMPV6_UNK_OPTION &&
 435                       (opt_unrec(skb, info))))
 436                         return;
 437
 438                 saddr = NULL;
 439         }
 440
 441         addr_type = ipv6_addr_type(&hdr->saddr);
 442
 443         /*
 444          *      Source addr check
 445          */
 446
 447         if (__ipv6_addr_needs_scope_id(addr_type))
 448                 iif = skb->dev->ifindex;
 449
 450         /*
 451          *      Must not send error if the source does not uniquely
 452          *      identify a single node (RFC2463 Section 2.4).
 453          *      We check unspecified / multicast addresses here,
 454          *      and anycast addresses will be checked later.
 455          */
 456         if ((addr_type == IPV6_ADDR_ANY) || (addr_type & IPV6_ADDR_MULTICAST)) {
 457                 net_dbg_ratelimited("icmp6_send: addr_any/mcast source [%pI6c > %pI6c]\n",
 458                                     &hdr->saddr, &hdr->daddr);
 459                 return;
 460         }
 461
 462         /*
 463          *      Never answer to a ICMP packet.
 464          */
 465         if (is_ineligible(skb)) {
 466                 net_dbg_ratelimited("icmp6_send: no reply to icmp error [%pI6c > %pI6c]\n",
 467                                     &hdr->saddr, &hdr->daddr);
 468                 return;
 469         }
 470
 471         mip6_addr_swap(skb);
 472
 473         memset(&fl6, 0, sizeof(fl6));
 474         fl6.flowi6_proto = IPPROTO_ICMPV6;
 475         fl6.daddr = hdr->saddr;
 476         if (saddr)
 477                 fl6.saddr = *saddr;
 478         fl6.flowi6_mark = mark;
 479         fl6.flowi6_oif = iif;
 480         fl6.fl6_icmp_type = type;
 481         fl6.fl6_icmp_code = code;
 482         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 483
 484         sk = icmpv6_xmit_lock(net);
 485         if (!sk)
 486                 return;
 487         sk->sk_mark = mark;
 488         np = inet6_sk(sk);
 489
 490         if (!icmpv6_xrlim_allow(sk, type, &fl6))
 491                 goto out;
 492
 493         tmp_hdr.icmp6_type = type;
 494         tmp_hdr.icmp6_code = code;
 495         tmp_hdr.icmp6_cksum = 0;
 496         tmp_hdr.icmp6_pointer = htonl(info);
 497
 498         if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr))
 499                 fl6.flowi6_oif = np->mcast_oif;
 500         else if (!fl6.flowi6_oif)
 501                 fl6.flowi6_oif = np->ucast_oif;
 502
 503         if (!fl6.flowi6_oif)
 504                 fl6.flowi6_oif = l3mdev_master_ifindex(skb->dev);
 505
 506         dst = icmpv6_route_lookup(net, skb, sk, &fl6);
 507         if (IS_ERR(dst))
 508                 goto out;
 509
 510         hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
 511
 512         msg.skb = skb;
 513         msg.offset = skb_network_offset(skb);
 514         msg.type = type;
 515
 516         len = skb->len - msg.offset;
 517         len = min_t(unsigned int, len, IPV6_MIN_MTU - sizeof(struct ipv6hdr) - sizeof(struct icmp6hdr));
 518         if (len < 0) {
 519                 net_dbg_ratelimited("icmp: len problem [%pI6c > %pI6c]\n",
 520                                     &hdr->saddr, &hdr->daddr);
 521                 goto out_dst_release;
 522         }
 523
 524         rcu_read_lock();
 525         idev = __in6_dev_get(skb->dev);
 526
 527         err = ip6_append_data(sk, icmpv6_getfrag, &msg,
 528                               len + sizeof(struct icmp6hdr),
 529                               sizeof(struct icmp6hdr), hlimit,
 530                               np->tclass, NULL, &fl6, (struct rt6_info *)dst,
 531                               MSG_DONTWAIT, np->dontfrag, &sockc_unused);
 532         if (err) {
 533                 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTERRORS);
 534                 ip6_flush_pending_frames(sk);
 535         } else {
 536                 err = icmpv6_push_pending_frames(sk, &fl6, &tmp_hdr,
 537                                                  len + sizeof(struct icmp6hdr));
 538         }
 539         rcu_read_unlock();
 540 out_dst_release:
 541         dst_release(dst);
 542 out:
 543         icmpv6_xmit_unlock(sk);
 544 }
 545
 546 /* Slightly more convenient version of icmp6_send.
 547  */
 548 void icmpv6_param_prob(struct sk_buff *skb, u8 code, int pos)
 549 {
 550         icmp6_send(skb, ICMPV6_PARAMPROB, code, pos);
 551         kfree_skb(skb);
 552 }
 553
 554 static void icmpv6_echo_reply(struct sk_buff *skb)
 555 {
 556         struct net *net = dev_net(skb->dev);
 557         struct sock *sk;
 558         struct inet6_dev *idev;
 559         struct ipv6_pinfo *np;
 560         const struct in6_addr *saddr = NULL;
 561         struct icmp6hdr *icmph = icmp6_hdr(skb);
 562         struct icmp6hdr tmp_hdr;
 563         struct flowi6 fl6;
 564         struct icmpv6_msg msg;
 565         struct dst_entry *dst;
 566         int err = 0;
 567         int hlimit;
 568         u8 tclass;
 569         u32 mark = IP6_REPLY_MARK(net, skb->mark);
 570         struct sockcm_cookie sockc_unused = {0};
 571
 572         saddr = &ipv6_hdr(skb)->daddr;
 573
 574         if (!ipv6_unicast_destination(skb) &&
 575             !(net->ipv6.sysctl.anycast_src_echo_reply &&
 576               ipv6_anycast_destination(skb_dst(skb), saddr)))
 577                 saddr = NULL;
 578
 579         memcpy(&tmp_hdr, icmph, sizeof(tmp_hdr));
 580         tmp_hdr.icmp6_type = ICMPV6_ECHO_REPLY;
 581
 582         memset(&fl6, 0, sizeof(fl6));
 583         fl6.flowi6_proto = IPPROTO_ICMPV6;
 584         fl6.daddr = ipv6_hdr(skb)->saddr;
 585         if (saddr)
 586                 fl6.saddr = *saddr;
 587         fl6.flowi6_oif = l3mdev_fib_oif(skb->dev);
 588         fl6.fl6_icmp_type = ICMPV6_ECHO_REPLY;
 589         fl6.flowi6_mark = mark;
 590         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 591
 592         sk = icmpv6_xmit_lock(net);
 593         if (!sk)
 594                 return;
 595         sk->sk_mark = mark;
 596         np = inet6_sk(sk);
 597
 598         if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr))
 599                 fl6.flowi6_oif = np->mcast_oif;
 600         else if (!fl6.flowi6_oif)
 601                 fl6.flowi6_oif = np->ucast_oif;
 602
 603         err = ip6_dst_lookup(net, sk, &dst, &fl6);
 604         if (err)
 605                 goto out;
 606         dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), sk, 0);
 607         if (IS_ERR(dst))
 608                 goto out;
 609
 610         hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
 611
 612         idev = __in6_dev_get(skb->dev);
 613
 614         msg.skb = skb;
 615         msg.offset = 0;
 616         msg.type = ICMPV6_ECHO_REPLY;
 617
 618         tclass = ipv6_get_dsfield(ipv6_hdr(skb));
 619         err = ip6_append_data(sk, icmpv6_getfrag, &msg, skb->len + sizeof(struct icmp6hdr),
 620                                 sizeof(struct icmp6hdr), hlimit, tclass, NULL, &fl6,
 621                                 (struct rt6_info *)dst, MSG_DONTWAIT,
 622                                 np->dontfrag, &sockc_unused);
 623
 624         if (err) {
 625                 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTERRORS);
 626                 ip6_flush_pending_frames(sk);
 627         } else {
 628                 err = icmpv6_push_pending_frames(sk, &fl6, &tmp_hdr,
 629                                                  skb->len + sizeof(struct icmp6hdr));
 630         }
 631         dst_release(dst);
 632 out:
 633         icmpv6_xmit_unlock(sk);
 634 }
 635
 636 void icmpv6_notify(struct sk_buff *skb, u8 type, u8 code, __be32 info)
 637 {
 638         const struct inet6_protocol *ipprot;
 639         int inner_offset;
 640         __be16 frag_off;
 641         u8 nexthdr;
 642         struct net *net = dev_net(skb->dev);
 643
 644         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 645                 goto out;
 646
 647         nexthdr = ((struct ipv6hdr *)skb->data)->nexthdr;
 648         if (ipv6_ext_hdr(nexthdr)) {
 649                 /* now skip over extension headers */
 650                 inner_offset = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr),
 651                                                 &nexthdr, &frag_off);
 652                 if (inner_offset < 0)
 653                         goto out;
 654         } else {
 655                 inner_offset = sizeof(struct ipv6hdr);
 656         }
 657
 658         /* Checkin header including 8 bytes of inner protocol header. */
 659         if (!pskb_may_pull(skb, inner_offset+8))
 660                 goto out;
 661
 662         /* BUGGG_FUTURE: we should try to parse exthdrs in this packet.
 663            Without this we will not able f.e. to make source routed
 664            pmtu discovery.
 665            Corresponding argument (opt) to notifiers is already added.
 666            --ANK (980726)
 667          */
 668
 669         ipprot = rcu_dereference(inet6_protos[nexthdr]);
 670         if (ipprot && ipprot->err_handler)
 671                 ipprot->err_handler(skb, NULL, type, code, inner_offset, info);
 672
 673         raw6_icmp_error(skb, nexthdr, type, code, inner_offset, info);
 674         return;
 675
 676 out:
 677         ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
 678 }
 679
 680 /*
 681  *      Handle icmp messages
 682  */
 683
 684 static int icmpv6_rcv(struct sk_buff *skb)
 685 {
 686         struct net_device *dev = skb->dev;
 687         struct inet6_dev *idev = __in6_dev_get(dev);
 688         const struct in6_addr *saddr, *daddr;
 689         struct icmp6hdr *hdr;
 690         u8 type;
 691         bool success = false;
 692
 693         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 694                 struct sec_path *sp = skb_sec_path(skb);
 695                 int nh;
 696
 697                 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
 698                                  XFRM_STATE_ICMP))
 699                         goto drop_no_count;
 700
 701                 if (!pskb_may_pull(skb, sizeof(*hdr) + sizeof(struct ipv6hdr)))
 702                         goto drop_no_count;
 703
 704                 nh = skb_network_offset(skb);
 705                 skb_set_network_header(skb, sizeof(*hdr));
 706
 707                 if (!xfrm6_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
 708                         goto drop_no_count;
 709
 710                 skb_set_network_header(skb, nh);
 711         }
 712
 713         ICMP6_INC_STATS_BH(dev_net(dev), idev, ICMP6_MIB_INMSGS);
 714
 715         saddr = &ipv6_hdr(skb)->saddr;
 716         daddr = &ipv6_hdr(skb)->daddr;
 717
 718         if (skb_checksum_validate(skb, IPPROTO_ICMPV6, ip6_compute_pseudo)) {
 719                 net_dbg_ratelimited("ICMPv6 checksum failed [%pI6c > %pI6c]\n",
 720                                     saddr, daddr);
 721                 goto csum_error;
 722         }
 723
 724         if (!pskb_pull(skb, sizeof(*hdr)))
 725                 goto discard_it;
 726
 727         hdr = icmp6_hdr(skb);
 728
 729         type = hdr->icmp6_type;
 730
 731         ICMP6MSGIN_INC_STATS_BH(dev_net(dev), idev, type);
 732
 733         switch (type) {
 734         case ICMPV6_ECHO_REQUEST:
 735                 icmpv6_echo_reply(skb);
 736                 break;
 737
 738         case ICMPV6_ECHO_REPLY:
 739                 success = ping_rcv(skb);
 740                 break;
 741
 742         case ICMPV6_PKT_TOOBIG:
 743                 /* BUGGG_FUTURE: if packet contains rthdr, we cannot update
 744                    standard destination cache. Seems, only "advanced"
 745                    destination cache will allow to solve this problem
 746                    --ANK (980726)
 747                  */
 748                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 749                         goto discard_it;
 750                 hdr = icmp6_hdr(skb);
 751
 752                 /*
 753                  *      Drop through to notify
 754                  */
 755
 756         case ICMPV6_DEST_UNREACH:
 757         case ICMPV6_TIME_EXCEED:
 758         case ICMPV6_PARAMPROB:
 759                 icmpv6_notify(skb, type, hdr->icmp6_code, hdr->icmp6_mtu);
 760                 break;
 761
 762         case NDISC_ROUTER_SOLICITATION:
 763         case NDISC_ROUTER_ADVERTISEMENT:
 764         case NDISC_NEIGHBOUR_SOLICITATION:
 765         case NDISC_NEIGHBOUR_ADVERTISEMENT:
 766         case NDISC_REDIRECT:
 767                 ndisc_rcv(skb);
 768                 break;
 769
 770         case ICMPV6_MGM_QUERY:
 771                 igmp6_event_query(skb);
 772                 break;
 773
 774         case ICMPV6_MGM_REPORT:
 775                 igmp6_event_report(skb);
 776                 break;
 777
 778         case ICMPV6_MGM_REDUCTION:
 779         case ICMPV6_NI_QUERY:
 780         case ICMPV6_NI_REPLY:
 781         case ICMPV6_MLD2_REPORT:
 782         case ICMPV6_DHAAD_REQUEST:
 783         case ICMPV6_DHAAD_REPLY:
 784         case ICMPV6_MOBILE_PREFIX_SOL:
 785         case ICMPV6_MOBILE_PREFIX_ADV:
 786                 break;
 787
 788         default:
 789                 /* informational */
 790                 if (type & ICMPV6_INFOMSG_MASK)
 791                         break;
 792
 793                 net_dbg_ratelimited("icmpv6: msg of unknown type [%pI6c > %pI6c]\n",
 794                                     saddr, daddr);
 795
 796                 /*
 797                  * error of unknown type.
 798                  * must pass to upper level
 799                  */
 800
 801                 icmpv6_notify(skb, type, hdr->icmp6_code, hdr->icmp6_mtu);
 802         }
 803
 804         /* until the v6 path can be better sorted assume failure and
 805          * preserve the status quo behaviour for the rest of the paths to here
 806          */
 807         if (success)
 808                 consume_skb(skb);
 809         else
 810                 kfree_skb(skb);
 811
 812         return 0;
 813
 814 csum_error:
 815         ICMP6_INC_STATS_BH(dev_net(dev), idev, ICMP6_MIB_CSUMERRORS);
 816 discard_it:
 817         ICMP6_INC_STATS_BH(dev_net(dev), idev, ICMP6_MIB_INERRORS);
 818 drop_no_count:
 819         kfree_skb(skb);
 820         return 0;
 821 }
 822
 823 void icmpv6_flow_init(struct sock *sk, struct flowi6 *fl6,
 824                       u8 type,
 825                       const struct in6_addr *saddr,
 826                       const struct in6_addr *daddr,
 827                       int oif)
 828 {
 829         memset(fl6, 0, sizeof(*fl6));
 830         fl6->saddr = *saddr;
 831         fl6->daddr = *daddr;
 832         fl6->flowi6_proto       = IPPROTO_ICMPV6;
 833         fl6->fl6_icmp_type      = type;
 834         fl6->fl6_icmp_code      = 0;
 835         fl6->flowi6_oif         = oif;
 836         security_sk_classify_flow(sk, flowi6_to_flowi(fl6));
 837 }
 838
 839 static int __net_init icmpv6_sk_init(struct net *net)
 840 {
 841         struct sock *sk;
 842         int err, i, j;
 843
 844         net->ipv6.icmp_sk =
 845                 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
 846         if (!net->ipv6.icmp_sk)
 847                 return -ENOMEM;
 848
 849         for_each_possible_cpu(i) {
 850                 err = inet_ctl_sock_create(&sk, PF_INET6,
 851                                            SOCK_RAW, IPPROTO_ICMPV6, net);
 852                 if (err < 0) {
 853                         pr_err("Failed to initialize the ICMP6 control socket (err %d)\n",
 854                                err);
 855                         goto fail;
 856                 }
 857
 858                 net->ipv6.icmp_sk[i] = sk;
 859
 860                 /* Enough space for 2 64K ICMP packets, including
 861                  * sk_buff struct overhead.
 862                  */
 863                 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
 864         }
 865         return 0;
 866
 867  fail:
 868         for (j = 0; j < i; j++)
 869                 inet_ctl_sock_destroy(net->ipv6.icmp_sk[j]);
 870         kfree(net->ipv6.icmp_sk);
 871         return err;
 872 }
 873
 874 static void __net_exit icmpv6_sk_exit(struct net *net)
 875 {
 876         int i;
 877
 878         for_each_possible_cpu(i) {
 879                 inet_ctl_sock_destroy(net->ipv6.icmp_sk[i]);
 880         }
 881         kfree(net->ipv6.icmp_sk);
 882 }
 883
 884 static struct pernet_operations icmpv6_sk_ops = {
 885         .init = icmpv6_sk_init,
 886         .exit = icmpv6_sk_exit,
 887 };
 888
 889 int __init icmpv6_init(void)
 890 {
 891         int err;
 892
 893         err = register_pernet_subsys(&icmpv6_sk_ops);
 894         if (err < 0)
 895                 return err;
 896
 897         err = -EAGAIN;
 898         if (inet6_add_protocol(&icmpv6_protocol, IPPROTO_ICMPV6) < 0)
 899                 goto fail;
 900
 901         err = inet6_register_icmp_sender(icmp6_send);
 902         if (err)
 903                 goto sender_reg_err;
 904         return 0;
 905
 906 sender_reg_err:
 907         inet6_del_protocol(&icmpv6_protocol, IPPROTO_ICMPV6);
 908 fail:
 909         pr_err("Failed to register ICMP6 protocol\n");
 910         unregister_pernet_subsys(&icmpv6_sk_ops);
 911         return err;
 912 }
 913
 914 void icmpv6_cleanup(void)
 915 {
 916         inet6_unregister_icmp_sender(icmp6_send);
 917         unregister_pernet_subsys(&icmpv6_sk_ops);
 918         inet6_del_protocol(&icmpv6_protocol, IPPROTO_ICMPV6);
 919 }
 920
 921
 922 static const struct icmp6_err {
 923         int err;
 924         int fatal;
 925 } tab_unreach[] = {
 926         {       /* NOROUTE */
 927                 .err    = ENETUNREACH,
 928                 .fatal  = 0,
 929         },
 930         {       /* ADM_PROHIBITED */
 931                 .err    = EACCES,
 932                 .fatal  = 1,
 933         },
 934         {       /* Was NOT_NEIGHBOUR, now reserved */
 935                 .err    = EHOSTUNREACH,
 936                 .fatal  = 0,
 937         },
 938         {       /* ADDR_UNREACH */
 939                 .err    = EHOSTUNREACH,
 940                 .fatal  = 0,
 941         },
 942         {       /* PORT_UNREACH */
 943                 .err    = ECONNREFUSED,
 944                 .fatal  = 1,
 945         },
 946         {       /* POLICY_FAIL */
 947                 .err    = EACCES,
 948                 .fatal  = 1,
 949         },
 950         {       /* REJECT_ROUTE */
 951                 .err    = EACCES,
 952                 .fatal  = 1,
 953         },
 954 };
 955
 956 int icmpv6_err_convert(u8 type, u8 code, int *err)
 957 {
 958         int fatal = 0;
 959
 960         *err = EPROTO;
 961
 962         switch (type) {
 963         case ICMPV6_DEST_UNREACH:
 964                 fatal = 1;
 965                 if (code < ARRAY_SIZE(tab_unreach)) {
 966                         *err  = tab_unreach[code].err;
 967                         fatal = tab_unreach[code].fatal;
 968                 }
 969                 break;
 970
 971         case ICMPV6_PKT_TOOBIG:
 972                 *err = EMSGSIZE;
 973                 break;
 974
 975         case ICMPV6_PARAMPROB:
 976                 *err = EPROTO;
 977                 fatal = 1;
 978                 break;
 979
 980         case ICMPV6_TIME_EXCEED:
 981                 *err = EHOSTUNREACH;
 982                 break;
 983         }
 984
 985         return fatal;
 986 }
 987 EXPORT_SYMBOL(icmpv6_err_convert);
 988
 989 #ifdef CONFIG_SYSCTL
 990 static struct ctl_table ipv6_icmp_table_template[] = {
 991         {
 992                 .procname       = "ratelimit",
 993                 .data           = &init_net.ipv6.sysctl.icmpv6_time,
 994                 .maxlen         = sizeof(int),
 995                 .mode           = 0644,
 996                 .proc_handler   = proc_dointvec_ms_jiffies,
 997         },
 998         { },
 999 };
1000
1001 struct ctl_table * __net_init ipv6_icmp_sysctl_init(struct net *net)
1002 {
1003         struct ctl_table *table;
1004
1005         table = kmemdup(ipv6_icmp_table_template,
1006                         sizeof(ipv6_icmp_table_template),
1007                         GFP_KERNEL);
1008
1009         if (table)
1010                 table[0].data = &net->ipv6.sysctl.icmpv6_time;
1011
1012         return table;
1013 }
1014 #endif