net/ipv6/ip6_input.c

   1 /*
   2  *      IPv6 input
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *      Ian P. Morris           <I.P.Morris@soton.ac.uk>
   8  *
   9  *      Based in linux/net/ipv4/ip_input.c
  10  *
  11  *      This program is free software; you can redistribute it and/or
  12  *      modify it under the terms of the GNU General Public License
  13  *      as published by the Free Software Foundation; either version
  14  *      2 of the License, or (at your option) any later version.
  15  */
  16 /* Changes
  17  *
  18  *      Mitsuru KANDA @USAGI and
  19  *      YOSHIFUJI Hideaki @USAGI: Remove ipv6_parse_exthdrs().
  20  */
  21
  22 #include <linux/errno.h>
  23 #include <linux/types.h>
  24 #include <linux/socket.h>
  25 #include <linux/sockios.h>
  26 #include <linux/net.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/in6.h>
  29 #include <linux/icmpv6.h>
  30 #include <linux/mroute6.h>
  31 #include <linux/slab.h>
  32
  33 #include <linux/netfilter.h>
  34 #include <linux/netfilter_ipv6.h>
  35
  36 #include <net/sock.h>
  37 #include <net/snmp.h>
  38
  39 #include <net/ipv6.h>
  40 #include <net/protocol.h>
  41 #include <net/transp_v6.h>
  42 #include <net/rawv6.h>
  43 #include <net/ndisc.h>
  44 #include <net/ip6_route.h>
  45 #include <net/addrconf.h>
  46 #include <net/xfrm.h>
  47 #include <net/inet_ecn.h>
  48 #include <net/dst_metadata.h>
  49
  50 int ip6_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  51 {
  52         /* if ingress device is enslaved to an L3 master device pass the
  53          * skb to its handler for processing
  54          */
  55         skb = l3mdev_ip6_rcv(skb);
  56         if (!skb)
  57                 return NET_RX_SUCCESS;
  58
  59         if (net->ipv4.sysctl_ip_early_demux && !skb_dst(skb) && skb->sk == NULL) {
  60                 const struct inet6_protocol *ipprot;
  61
  62                 ipprot = rcu_dereference(inet6_protos[ipv6_hdr(skb)->nexthdr]);
  63                 if (ipprot && ipprot->early_demux)
  64                         ipprot->early_demux(skb);
  65         }
  66         if (!skb_valid_dst(skb))
  67                 ip6_route_input(skb);
  68
  69         return dst_input(skb);
  70 }
  71
  72 int ipv6_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
  73 {
  74         const struct ipv6hdr *hdr;
  75         u32 pkt_len;
  76         struct inet6_dev *idev;
  77         struct net *net = dev_net(skb->dev);
  78
  79         if (skb->pkt_type == PACKET_OTHERHOST) {
  80                 kfree_skb(skb);
  81                 return NET_RX_DROP;
  82         }
  83
  84         rcu_read_lock();
  85
  86         idev = __in6_dev_get(skb->dev);
  87
  88         __IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_IN, skb->len);
  89
  90         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL ||
  91             !idev || unlikely(idev->cnf.disable_ipv6)) {
  92                 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
  93                 goto drop;
  94         }
  95
  96         memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
  97
  98         /*
  99          * Store incoming device index. When the packet will
 100          * be queued, we cannot refer to skb->dev anymore.
 101          *
 102          * BTW, when we send a packet for our own local address on a
 103          * non-loopback interface (e.g. ethX), it is being delivered
 104          * via the loopback interface (lo) here; skb->dev = loopback_dev.
 105          * It, however, should be considered as if it is being
 106          * arrived via the sending interface (ethX), because of the
 107          * nature of scoping architecture. --yoshfuji
 108          */
 109         IP6CB(skb)->iif = skb_valid_dst(skb) ? ip6_dst_idev(skb_dst(skb))->dev->ifindex : dev->ifindex;
 110
 111         if (unlikely(!pskb_may_pull(skb, sizeof(*hdr))))
 112                 goto err;
 113
 114         hdr = ipv6_hdr(skb);
 115
 116         if (hdr->version != 6)
 117                 goto err;
 118
 119         __IP6_ADD_STATS(net, idev,
 120                         IPSTATS_MIB_NOECTPKTS +
 121                                 (ipv6_get_dsfield(hdr) & INET_ECN_MASK),
 122                         max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
 123         /*
 124          * RFC4291 2.5.3
 125          * The loopback address must not be used as the source address in IPv6
 126          * packets that are sent outside of a single node. [..]
 127          * A packet received on an interface with a destination address
 128          * of loopback must be dropped.
 129          */
 130         if ((ipv6_addr_loopback(&hdr->saddr) ||
 131              ipv6_addr_loopback(&hdr->daddr)) &&
 132              !(dev->flags & IFF_LOOPBACK))
 133                 goto err;
 134
 135         /* RFC4291 Errata ID: 3480
 136          * Interface-Local scope spans only a single interface on a
 137          * node and is useful only for loopback transmission of
 138          * multicast.  Packets with interface-local scope received
 139          * from another node must be discarded.
 140          */
 141         if (!(skb->pkt_type == PACKET_LOOPBACK ||
 142               dev->flags & IFF_LOOPBACK) &&
 143             ipv6_addr_is_multicast(&hdr->daddr) &&
 144             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 1)
 145                 goto err;
 146
 147         /* If enabled, drop unicast packets that were encapsulated in link-layer
 148          * multicast or broadcast to protected against the so-called "hole-196"
 149          * attack in 802.11 wireless.
 150          */
 151         if (!ipv6_addr_is_multicast(&hdr->daddr) &&
 152             (skb->pkt_type == PACKET_BROADCAST ||
 153              skb->pkt_type == PACKET_MULTICAST) &&
 154             idev->cnf.drop_unicast_in_l2_multicast)
 155                 goto err;
 156
 157         /* RFC4291 2.7
 158          * Nodes must not originate a packet to a multicast address whose scope
 159          * field contains the reserved value 0; if such a packet is received, it
 160          * must be silently dropped.
 161          */
 162         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 163             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 0)
 164                 goto err;
 165
 166         /*
 167          * RFC4291 2.7
 168          * Multicast addresses must not be used as source addresses in IPv6
 169          * packets or appear in any Routing header.
 170          */
 171         if (ipv6_addr_is_multicast(&hdr->saddr))
 172                 goto err;
 173
 174         skb->transport_header = skb->network_header + sizeof(*hdr);
 175         IP6CB(skb)->nhoff = offsetof(struct ipv6hdr, nexthdr);
 176
 177         pkt_len = ntohs(hdr->payload_len);
 178
 179         /* pkt_len may be zero if Jumbo payload option is present */
 180         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 181                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len) {
 182                         __IP6_INC_STATS(net,
 183                                         idev, IPSTATS_MIB_INTRUNCATEDPKTS);
 184                         goto drop;
 185                 }
 186                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr))) {
 187                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 188                         goto drop;
 189                 }
 190                 hdr = ipv6_hdr(skb);
 191         }
 192
 193         if (hdr->nexthdr == NEXTHDR_HOP) {
 194                 if (ipv6_parse_hopopts(skb) < 0) {
 195                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 196                         rcu_read_unlock();
 197                         return NET_RX_DROP;
 198                 }
 199         }
 200
 201         rcu_read_unlock();
 202
 203         /* Must drop socket now because of tproxy. */
 204         skb_orphan(skb);
 205
 206         return NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING,
 207                        net, NULL, skb, dev, NULL,
 208                        ip6_rcv_finish);
 209 err:
 210         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 211 drop:
 212         rcu_read_unlock();
 213         kfree_skb(skb);
 214         return NET_RX_DROP;
 215 }
 216
 217 /*
 218  *      Deliver the packet to the host
 219  */
 220
 221
 222 static int ip6_input_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 223 {
 224         const struct inet6_protocol *ipprot;
 225         struct inet6_dev *idev;
 226         unsigned int nhoff;
 227         int nexthdr;
 228         bool raw;
 229         bool have_final = false;
 230
 231         /*
 232          *      Parse extension headers
 233          */
 234
 235         rcu_read_lock();
 236 resubmit:
 237         idev = ip6_dst_idev(skb_dst(skb));
 238         if (!pskb_pull(skb, skb_transport_offset(skb)))
 239                 goto discard;
 240         nhoff = IP6CB(skb)->nhoff;
 241         nexthdr = skb_network_header(skb)[nhoff];
 242
 243 resubmit_final:
 244         raw = raw6_local_deliver(skb, nexthdr);
 245         ipprot = rcu_dereference(inet6_protos[nexthdr]);
 246         if (ipprot) {
 247                 int ret;
 248
 249                 if (have_final) {
 250                         if (!(ipprot->flags & INET6_PROTO_FINAL)) {
 251                                 /* Once we've seen a final protocol don't
 252                                  * allow encapsulation on any non-final
 253                                  * ones. This allows foo in UDP encapsulation
 254                                  * to work.
 255                                  */
 256                                 goto discard;
 257                         }
 258                 } else if (ipprot->flags & INET6_PROTO_FINAL) {
 259                         const struct ipv6hdr *hdr;
 260
 261                         /* Only do this once for first final protocol */
 262                         have_final = true;
 263
 264                         /* Free reference early: we don't need it any more,
 265                            and it may hold ip_conntrack module loaded
 266                            indefinitely. */
 267                         nf_reset(skb);
 268
 269                         skb_postpull_rcsum(skb, skb_network_header(skb),
 270                                            skb_network_header_len(skb));
 271                         hdr = ipv6_hdr(skb);
 272                         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 273                             !ipv6_chk_mcast_addr(skb->dev, &hdr->daddr,
 274                             &hdr->saddr) &&
 275                             !ipv6_is_mld(skb, nexthdr, skb_network_header_len(skb)))
 276                                 goto discard;
 277                 }
 278                 if (!(ipprot->flags & INET6_PROTO_NOPOLICY) &&
 279                     !xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 280                         goto discard;
 281
 282                 ret = ipprot->handler(skb);
 283                 if (ret > 0) {
 284                         if (ipprot->flags & INET6_PROTO_FINAL) {
 285                                 /* Not an extension header, most likely UDP
 286                                  * encapsulation. Use return value as nexthdr
 287                                  * protocol not nhoff (which presumably is
 288                                  * not set by handler).
 289                                  */
 290                                 nexthdr = ret;
 291                                 goto resubmit_final;
 292                         } else {
 293                                 goto resubmit;
 294                         }
 295                 } else if (ret == 0) {
 296                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 297                 }
 298         } else {
 299                 if (!raw) {
 300                         if (xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 301                                 __IP6_INC_STATS(net, idev,
 302                                                 IPSTATS_MIB_INUNKNOWNPROTOS);
 303                                 icmpv6_send(skb, ICMPV6_PARAMPROB,
 304                                             ICMPV6_UNK_NEXTHDR, nhoff);
 305                         }
 306                         kfree_skb(skb);
 307                 } else {
 308                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 309                         consume_skb(skb);
 310                 }
 311         }
 312         rcu_read_unlock();
 313         return 0;
 314
 315 discard:
 316         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 317         rcu_read_unlock();
 318         kfree_skb(skb);
 319         return 0;
 320 }
 321
 322
 323 int ip6_input(struct sk_buff *skb)
 324 {
 325         return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_IN,
 326                        dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 327                        ip6_input_finish);
 328 }
 329 EXPORT_SYMBOL_GPL(ip6_input);
 330
 331 int ip6_mc_input(struct sk_buff *skb)
 332 {
 333         const struct ipv6hdr *hdr;
 334         bool deliver;
 335
 336         __IP6_UPD_PO_STATS(dev_net(skb_dst(skb)->dev),
 337                          ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INMCAST,
 338                          skb->len);
 339
 340         hdr = ipv6_hdr(skb);
 341         deliver = ipv6_chk_mcast_addr(skb->dev, &hdr->daddr, NULL);
 342
 343 #ifdef CONFIG_IPV6_MROUTE
 344         /*
 345          *      IPv6 multicast router mode is now supported ;)
 346          */
 347         if (dev_net(skb->dev)->ipv6.devconf_all->mc_forwarding &&
 348             !(ipv6_addr_type(&hdr->daddr) &
 349               (IPV6_ADDR_LOOPBACK|IPV6_ADDR_LINKLOCAL)) &&
 350             likely(!(IP6CB(skb)->flags & IP6SKB_FORWARDED))) {
 351                 /*
 352                  * Okay, we try to forward - split and duplicate
 353                  * packets.
 354                  */
 355                 struct sk_buff *skb2;
 356                 struct inet6_skb_parm *opt = IP6CB(skb);
 357
 358                 /* Check for MLD */
 359                 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
 360                         /* Check if this is a mld message */
 361                         u8 nexthdr = hdr->nexthdr;
 362                         __be16 frag_off;
 363                         int offset;
 364
 365                         /* Check if the value of Router Alert
 366                          * is for MLD (0x0000).
 367                          */
 368                         if (opt->ra == htons(IPV6_OPT_ROUTERALERT_MLD)) {
 369                                 deliver = false;
 370
 371                                 if (!ipv6_ext_hdr(nexthdr)) {
 372                                         /* BUG */
 373                                         goto out;
 374                                 }
 375                                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr),
 376                                                           &nexthdr, &frag_off);
 377                                 if (offset < 0)
 378                                         goto out;
 379
 380                                 if (ipv6_is_mld(skb, nexthdr, offset))
 381                                         deliver = true;
 382
 383                                 goto out;
 384                         }
 385                         /* unknown RA - process it normally */
 386                 }
 387
 388                 if (deliver)
 389                         skb2 = skb_clone(skb, GFP_ATOMIC);
 390                 else {
 391                         skb2 = skb;
 392                         skb = NULL;
 393                 }
 394
 395                 if (skb2) {
 396                         ip6_mr_input(skb2);
 397                 }
 398         }
 399 out:
 400 #endif
 401         if (likely(deliver))
 402                 ip6_input(skb);
 403         else {
 404                 /* discard */
 405                 kfree_skb(skb);
 406         }
 407
 408         return 0;
 409 }