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 void udp_v6_early_demux(struct sk_buff *);
  51 void tcp_v6_early_demux(struct sk_buff *);
  52 int ip6_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  53 {
  54         /* if ingress device is enslaved to an L3 master device pass the
  55          * skb to its handler for processing
  56          */
  57         skb = l3mdev_ip6_rcv(skb);
  58         if (!skb)
  59                 return NET_RX_SUCCESS;
  60
  61         if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
  62             !skb_dst(skb) && !skb->sk) {
  63                 switch (ipv6_hdr(skb)->nexthdr) {
  64                 case IPPROTO_TCP:
  65                         if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux))
  66                                 tcp_v6_early_demux(skb);
  67                         break;
  68                 case IPPROTO_UDP:
  69                         if (READ_ONCE(net->ipv4.sysctl_udp_early_demux))
  70                                 udp_v6_early_demux(skb);
  71                         break;
  72                 }
  73         }
  74
  75         if (!skb_valid_dst(skb))
  76                 ip6_route_input(skb);
  77
  78         return dst_input(skb);
  79 }
  80
  81 int ipv6_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
  82 {
  83         const struct ipv6hdr *hdr;
  84         u32 pkt_len;
  85         struct inet6_dev *idev;
  86         struct net *net = dev_net(skb->dev);
  87
  88         if (skb->pkt_type == PACKET_OTHERHOST) {
  89                 kfree_skb(skb);
  90                 return NET_RX_DROP;
  91         }
  92
  93         rcu_read_lock();
  94
  95         idev = __in6_dev_get(skb->dev);
  96
  97         __IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_IN, skb->len);
  98
  99         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL ||
 100             !idev || unlikely(idev->cnf.disable_ipv6)) {
 101                 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 102                 goto drop;
 103         }
 104
 105         memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 106
 107         /*
 108          * Store incoming device index. When the packet will
 109          * be queued, we cannot refer to skb->dev anymore.
 110          *
 111          * BTW, when we send a packet for our own local address on a
 112          * non-loopback interface (e.g. ethX), it is being delivered
 113          * via the loopback interface (lo) here; skb->dev = loopback_dev.
 114          * It, however, should be considered as if it is being
 115          * arrived via the sending interface (ethX), because of the
 116          * nature of scoping architecture. --yoshfuji
 117          */
 118         IP6CB(skb)->iif = skb_valid_dst(skb) ? ip6_dst_idev(skb_dst(skb))->dev->ifindex : dev->ifindex;
 119
 120         if (unlikely(!pskb_may_pull(skb, sizeof(*hdr))))
 121                 goto err;
 122
 123         hdr = ipv6_hdr(skb);
 124
 125         if (hdr->version != 6)
 126                 goto err;
 127
 128         __IP6_ADD_STATS(net, idev,
 129                         IPSTATS_MIB_NOECTPKTS +
 130                                 (ipv6_get_dsfield(hdr) & INET_ECN_MASK),
 131                         max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
 132         /*
 133          * RFC4291 2.5.3
 134          * The loopback address must not be used as the source address in IPv6
 135          * packets that are sent outside of a single node. [..]
 136          * A packet received on an interface with a destination address
 137          * of loopback must be dropped.
 138          */
 139         if ((ipv6_addr_loopback(&hdr->saddr) ||
 140              ipv6_addr_loopback(&hdr->daddr)) &&
 141              !(dev->flags & IFF_LOOPBACK))
 142                 goto err;
 143
 144         /* RFC4291 Errata ID: 3480
 145          * Interface-Local scope spans only a single interface on a
 146          * node and is useful only for loopback transmission of
 147          * multicast.  Packets with interface-local scope received
 148          * from another node must be discarded.
 149          */
 150         if (!(skb->pkt_type == PACKET_LOOPBACK ||
 151               dev->flags & IFF_LOOPBACK) &&
 152             ipv6_addr_is_multicast(&hdr->daddr) &&
 153             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 1)
 154                 goto err;
 155
 156         /* If enabled, drop unicast packets that were encapsulated in link-layer
 157          * multicast or broadcast to protected against the so-called "hole-196"
 158          * attack in 802.11 wireless.
 159          */
 160         if (!ipv6_addr_is_multicast(&hdr->daddr) &&
 161             (skb->pkt_type == PACKET_BROADCAST ||
 162              skb->pkt_type == PACKET_MULTICAST) &&
 163             idev->cnf.drop_unicast_in_l2_multicast)
 164                 goto err;
 165
 166         /* RFC4291 2.7
 167          * Nodes must not originate a packet to a multicast address whose scope
 168          * field contains the reserved value 0; if such a packet is received, it
 169          * must be silently dropped.
 170          */
 171         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 172             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 0)
 173                 goto err;
 174
 175         /*
 176          * RFC4291 2.7
 177          * Multicast addresses must not be used as source addresses in IPv6
 178          * packets or appear in any Routing header.
 179          */
 180         if (ipv6_addr_is_multicast(&hdr->saddr))
 181                 goto err;
 182
 183         skb->transport_header = skb->network_header + sizeof(*hdr);
 184         IP6CB(skb)->nhoff = offsetof(struct ipv6hdr, nexthdr);
 185
 186         pkt_len = ntohs(hdr->payload_len);
 187
 188         /* pkt_len may be zero if Jumbo payload option is present */
 189         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 190                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len) {
 191                         __IP6_INC_STATS(net,
 192                                         idev, IPSTATS_MIB_INTRUNCATEDPKTS);
 193                         goto drop;
 194                 }
 195                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr))) {
 196                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 197                         goto drop;
 198                 }
 199                 hdr = ipv6_hdr(skb);
 200         }
 201
 202         if (hdr->nexthdr == NEXTHDR_HOP) {
 203                 if (ipv6_parse_hopopts(skb) < 0) {
 204                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 205                         rcu_read_unlock();
 206                         return NET_RX_DROP;
 207                 }
 208         }
 209
 210         rcu_read_unlock();
 211
 212         /* Must drop socket now because of tproxy. */
 213         skb_orphan(skb);
 214
 215         return NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING,
 216                        net, NULL, skb, dev, NULL,
 217                        ip6_rcv_finish);
 218 err:
 219         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 220 drop:
 221         rcu_read_unlock();
 222         kfree_skb(skb);
 223         return NET_RX_DROP;
 224 }
 225
 226 /*
 227  *      Deliver the packet to the host
 228  */
 229
 230
 231 static int ip6_input_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 232 {
 233         const struct inet6_protocol *ipprot;
 234         struct inet6_dev *idev;
 235         unsigned int nhoff;
 236         int nexthdr;
 237         bool raw;
 238         bool have_final = false;
 239
 240         /*
 241          *      Parse extension headers
 242          */
 243
 244         rcu_read_lock();
 245 resubmit:
 246         idev = ip6_dst_idev(skb_dst(skb));
 247         if (!pskb_pull(skb, skb_transport_offset(skb)))
 248                 goto discard;
 249         nhoff = IP6CB(skb)->nhoff;
 250         nexthdr = skb_network_header(skb)[nhoff];
 251
 252 resubmit_final:
 253         raw = raw6_local_deliver(skb, nexthdr);
 254         ipprot = rcu_dereference(inet6_protos[nexthdr]);
 255         if (ipprot) {
 256                 int ret;
 257
 258                 if (have_final) {
 259                         if (!(ipprot->flags & INET6_PROTO_FINAL)) {
 260                                 /* Once we've seen a final protocol don't
 261                                  * allow encapsulation on any non-final
 262                                  * ones. This allows foo in UDP encapsulation
 263                                  * to work.
 264                                  */
 265                                 goto discard;
 266                         }
 267                 } else if (ipprot->flags & INET6_PROTO_FINAL) {
 268                         const struct ipv6hdr *hdr;
 269
 270                         /* Only do this once for first final protocol */
 271                         have_final = true;
 272
 273                         /* Free reference early: we don't need it any more,
 274                            and it may hold ip_conntrack module loaded
 275                            indefinitely. */
 276                         nf_reset(skb);
 277
 278                         skb_postpull_rcsum(skb, skb_network_header(skb),
 279                                            skb_network_header_len(skb));
 280                         hdr = ipv6_hdr(skb);
 281                         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 282                             !ipv6_chk_mcast_addr(skb->dev, &hdr->daddr,
 283                             &hdr->saddr) &&
 284                             !ipv6_is_mld(skb, nexthdr, skb_network_header_len(skb)))
 285                                 goto discard;
 286                 }
 287                 if (!(ipprot->flags & INET6_PROTO_NOPOLICY) &&
 288                     !xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 289                         goto discard;
 290
 291                 ret = ipprot->handler(skb);
 292                 if (ret > 0) {
 293                         if (ipprot->flags & INET6_PROTO_FINAL) {
 294                                 /* Not an extension header, most likely UDP
 295                                  * encapsulation. Use return value as nexthdr
 296                                  * protocol not nhoff (which presumably is
 297                                  * not set by handler).
 298                                  */
 299                                 nexthdr = ret;
 300                                 goto resubmit_final;
 301                         } else {
 302                                 goto resubmit;
 303                         }
 304                 } else if (ret == 0) {
 305                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 306                 }
 307         } else {
 308                 if (!raw) {
 309                         if (xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 310                                 __IP6_INC_STATS(net, idev,
 311                                                 IPSTATS_MIB_INUNKNOWNPROTOS);
 312                                 icmpv6_send(skb, ICMPV6_PARAMPROB,
 313                                             ICMPV6_UNK_NEXTHDR, nhoff);
 314                         }
 315                         kfree_skb(skb);
 316                 } else {
 317                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 318                         consume_skb(skb);
 319                 }
 320         }
 321         rcu_read_unlock();
 322         return 0;
 323
 324 discard:
 325         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 326         rcu_read_unlock();
 327         kfree_skb(skb);
 328         return 0;
 329 }
 330
 331
 332 int ip6_input(struct sk_buff *skb)
 333 {
 334         return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_IN,
 335                        dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 336                        ip6_input_finish);
 337 }
 338 EXPORT_SYMBOL_GPL(ip6_input);
 339
 340 int ip6_mc_input(struct sk_buff *skb)
 341 {
 342         const struct ipv6hdr *hdr;
 343         bool deliver;
 344
 345         __IP6_UPD_PO_STATS(dev_net(skb_dst(skb)->dev),
 346                          ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INMCAST,
 347                          skb->len);
 348
 349         hdr = ipv6_hdr(skb);
 350         deliver = ipv6_chk_mcast_addr(skb->dev, &hdr->daddr, NULL);
 351
 352 #ifdef CONFIG_IPV6_MROUTE
 353         /*
 354          *      IPv6 multicast router mode is now supported ;)
 355          */
 356         if (dev_net(skb->dev)->ipv6.devconf_all->mc_forwarding &&
 357             !(ipv6_addr_type(&hdr->daddr) &
 358               (IPV6_ADDR_LOOPBACK|IPV6_ADDR_LINKLOCAL)) &&
 359             likely(!(IP6CB(skb)->flags & IP6SKB_FORWARDED))) {
 360                 /*
 361                  * Okay, we try to forward - split and duplicate
 362                  * packets.
 363                  */
 364                 struct sk_buff *skb2;
 365                 struct inet6_skb_parm *opt = IP6CB(skb);
 366
 367                 /* Check for MLD */
 368                 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
 369                         /* Check if this is a mld message */
 370                         u8 nexthdr = hdr->nexthdr;
 371                         __be16 frag_off;
 372                         int offset;
 373
 374                         /* Check if the value of Router Alert
 375                          * is for MLD (0x0000).
 376                          */
 377                         if (opt->ra == htons(IPV6_OPT_ROUTERALERT_MLD)) {
 378                                 deliver = false;
 379
 380                                 if (!ipv6_ext_hdr(nexthdr)) {
 381                                         /* BUG */
 382                                         goto out;
 383                                 }
 384                                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr),
 385                                                           &nexthdr, &frag_off);
 386                                 if (offset < 0)
 387                                         goto out;
 388
 389                                 if (ipv6_is_mld(skb, nexthdr, offset))
 390                                         deliver = true;
 391
 392                                 goto out;
 393                         }
 394                         /* unknown RA - process it normally */
 395                 }
 396
 397                 if (deliver)
 398                         skb2 = skb_clone(skb, GFP_ATOMIC);
 399                 else {
 400                         skb2 = skb;
 401                         skb = NULL;
 402                 }
 403
 404                 if (skb2) {
 405                         ip6_mr_input(skb2);
 406                 }
 407         }
 408 out:
 409 #endif
 410         if (likely(deliver))
 411                 ip6_input(skb);
 412         else {
 413                 /* discard */
 414                 kfree_skb(skb);
 415         }
 416
 417         return 0;
 418 }