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 static void ip6_rcv_finish_core(struct net *net, struct sock *sk,
  53                                 struct sk_buff *skb)
  54 {
  55         if (READ_ONCE(net->ipv4.sysctl_ip_early_demux) &&
  56             !skb_dst(skb) && !skb->sk) {
  57                 switch (ipv6_hdr(skb)->nexthdr) {
  58                 case IPPROTO_TCP:
  59                         if (READ_ONCE(net->ipv4.sysctl_tcp_early_demux))
  60                                 tcp_v6_early_demux(skb);
  61                         break;
  62                 case IPPROTO_UDP:
  63                         if (READ_ONCE(net->ipv4.sysctl_udp_early_demux))
  64                                 udp_v6_early_demux(skb);
  65                         break;
  66                 }
  67         }
  68
  69         if (!skb_valid_dst(skb))
  70                 ip6_route_input(skb);
  71 }
  72
  73 int ip6_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  74 {
  75         /* if ingress device is enslaved to an L3 master device pass the
  76          * skb to its handler for processing
  77          */
  78         skb = l3mdev_ip6_rcv(skb);
  79         if (!skb)
  80                 return NET_RX_SUCCESS;
  81         ip6_rcv_finish_core(net, sk, skb);
  82
  83         return dst_input(skb);
  84 }
  85
  86 static void ip6_sublist_rcv_finish(struct list_head *head)
  87 {
  88         struct sk_buff *skb, *next;
  89
  90         list_for_each_entry_safe(skb, next, head, list) {
  91                 skb_list_del_init(skb);
  92                 dst_input(skb);
  93         }
  94 }
  95
  96 static void ip6_list_rcv_finish(struct net *net, struct sock *sk,
  97                                 struct list_head *head)
  98 {
  99         struct dst_entry *curr_dst = NULL;
 100         struct sk_buff *skb, *next;
 101         struct list_head sublist;
 102
 103         INIT_LIST_HEAD(&sublist);
 104         list_for_each_entry_safe(skb, next, head, list) {
 105                 struct dst_entry *dst;
 106
 107                 skb_list_del_init(skb);
 108                 /* if ingress device is enslaved to an L3 master device pass the
 109                  * skb to its handler for processing
 110                  */
 111                 skb = l3mdev_ip6_rcv(skb);
 112                 if (!skb)
 113                         continue;
 114                 ip6_rcv_finish_core(net, sk, skb);
 115                 dst = skb_dst(skb);
 116                 if (curr_dst != dst) {
 117                         /* dispatch old sublist */
 118                         if (!list_empty(&sublist))
 119                                 ip6_sublist_rcv_finish(&sublist);
 120                         /* start new sublist */
 121                         INIT_LIST_HEAD(&sublist);
 122                         curr_dst = dst;
 123                 }
 124                 list_add_tail(&skb->list, &sublist);
 125         }
 126         /* dispatch final sublist */
 127         ip6_sublist_rcv_finish(&sublist);
 128 }
 129
 130 static struct sk_buff *ip6_rcv_core(struct sk_buff *skb, struct net_device *dev,
 131                                     struct net *net)
 132 {
 133         const struct ipv6hdr *hdr;
 134         u32 pkt_len;
 135         struct inet6_dev *idev;
 136
 137         if (skb->pkt_type == PACKET_OTHERHOST) {
 138                 kfree_skb(skb);
 139                 return NULL;
 140         }
 141
 142         rcu_read_lock();
 143
 144         idev = __in6_dev_get(skb->dev);
 145
 146         __IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_IN, skb->len);
 147
 148         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL ||
 149             !idev || unlikely(idev->cnf.disable_ipv6)) {
 150                 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 151                 goto drop;
 152         }
 153
 154         memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 155
 156         /*
 157          * Store incoming device index. When the packet will
 158          * be queued, we cannot refer to skb->dev anymore.
 159          *
 160          * BTW, when we send a packet for our own local address on a
 161          * non-loopback interface (e.g. ethX), it is being delivered
 162          * via the loopback interface (lo) here; skb->dev = loopback_dev.
 163          * It, however, should be considered as if it is being
 164          * arrived via the sending interface (ethX), because of the
 165          * nature of scoping architecture. --yoshfuji
 166          */
 167         IP6CB(skb)->iif = skb_valid_dst(skb) ? ip6_dst_idev(skb_dst(skb))->dev->ifindex : dev->ifindex;
 168
 169         if (unlikely(!pskb_may_pull(skb, sizeof(*hdr))))
 170                 goto err;
 171
 172         hdr = ipv6_hdr(skb);
 173
 174         if (hdr->version != 6)
 175                 goto err;
 176
 177         __IP6_ADD_STATS(net, idev,
 178                         IPSTATS_MIB_NOECTPKTS +
 179                                 (ipv6_get_dsfield(hdr) & INET_ECN_MASK),
 180                         max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
 181         /*
 182          * RFC4291 2.5.3
 183          * The loopback address must not be used as the source address in IPv6
 184          * packets that are sent outside of a single node. [..]
 185          * A packet received on an interface with a destination address
 186          * of loopback must be dropped.
 187          */
 188         if ((ipv6_addr_loopback(&hdr->saddr) ||
 189              ipv6_addr_loopback(&hdr->daddr)) &&
 190              !(dev->flags & IFF_LOOPBACK))
 191                 goto err;
 192
 193         /* RFC4291 Errata ID: 3480
 194          * Interface-Local scope spans only a single interface on a
 195          * node and is useful only for loopback transmission of
 196          * multicast.  Packets with interface-local scope received
 197          * from another node must be discarded.
 198          */
 199         if (!(skb->pkt_type == PACKET_LOOPBACK ||
 200               dev->flags & IFF_LOOPBACK) &&
 201             ipv6_addr_is_multicast(&hdr->daddr) &&
 202             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 1)
 203                 goto err;
 204
 205         /* If enabled, drop unicast packets that were encapsulated in link-layer
 206          * multicast or broadcast to protected against the so-called "hole-196"
 207          * attack in 802.11 wireless.
 208          */
 209         if (!ipv6_addr_is_multicast(&hdr->daddr) &&
 210             (skb->pkt_type == PACKET_BROADCAST ||
 211              skb->pkt_type == PACKET_MULTICAST) &&
 212             idev->cnf.drop_unicast_in_l2_multicast)
 213                 goto err;
 214
 215         /* RFC4291 2.7
 216          * Nodes must not originate a packet to a multicast address whose scope
 217          * field contains the reserved value 0; if such a packet is received, it
 218          * must be silently dropped.
 219          */
 220         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 221             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 0)
 222                 goto err;
 223
 224         /*
 225          * RFC4291 2.7
 226          * Multicast addresses must not be used as source addresses in IPv6
 227          * packets or appear in any Routing header.
 228          */
 229         if (ipv6_addr_is_multicast(&hdr->saddr))
 230                 goto err;
 231
 232         skb->transport_header = skb->network_header + sizeof(*hdr);
 233         IP6CB(skb)->nhoff = offsetof(struct ipv6hdr, nexthdr);
 234
 235         pkt_len = ntohs(hdr->payload_len);
 236
 237         /* pkt_len may be zero if Jumbo payload option is present */
 238         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 239                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len) {
 240                         __IP6_INC_STATS(net,
 241                                         idev, IPSTATS_MIB_INTRUNCATEDPKTS);
 242                         goto drop;
 243                 }
 244                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr))) {
 245                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 246                         goto drop;
 247                 }
 248                 hdr = ipv6_hdr(skb);
 249         }
 250
 251         if (hdr->nexthdr == NEXTHDR_HOP) {
 252                 if (ipv6_parse_hopopts(skb) < 0) {
 253                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 254                         rcu_read_unlock();
 255                         return NULL;
 256                 }
 257         }
 258
 259         rcu_read_unlock();
 260
 261         /* Must drop socket now because of tproxy. */
 262         skb_orphan(skb);
 263
 264         return skb;
 265 err:
 266         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 267 drop:
 268         rcu_read_unlock();
 269         kfree_skb(skb);
 270         return NULL;
 271 }
 272
 273 int ipv6_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 274 {
 275         struct net *net = dev_net(skb->dev);
 276
 277         skb = ip6_rcv_core(skb, dev, net);
 278         if (skb == NULL)
 279                 return NET_RX_DROP;
 280         return NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING,
 281                        net, NULL, skb, dev, NULL,
 282                        ip6_rcv_finish);
 283 }
 284
 285 static void ip6_sublist_rcv(struct list_head *head, struct net_device *dev,
 286                             struct net *net)
 287 {
 288         NF_HOOK_LIST(NFPROTO_IPV6, NF_INET_PRE_ROUTING, net, NULL,
 289                      head, dev, NULL, ip6_rcv_finish);
 290         ip6_list_rcv_finish(net, NULL, head);
 291 }
 292
 293 /* Receive a list of IPv6 packets */
 294 void ipv6_list_rcv(struct list_head *head, struct packet_type *pt,
 295                    struct net_device *orig_dev)
 296 {
 297         struct net_device *curr_dev = NULL;
 298         struct net *curr_net = NULL;
 299         struct sk_buff *skb, *next;
 300         struct list_head sublist;
 301
 302         INIT_LIST_HEAD(&sublist);
 303         list_for_each_entry_safe(skb, next, head, list) {
 304                 struct net_device *dev = skb->dev;
 305                 struct net *net = dev_net(dev);
 306
 307                 skb_list_del_init(skb);
 308                 skb = ip6_rcv_core(skb, dev, net);
 309                 if (skb == NULL)
 310                         continue;
 311
 312                 if (curr_dev != dev || curr_net != net) {
 313                         /* dispatch old sublist */
 314                         if (!list_empty(&sublist))
 315                                 ip6_sublist_rcv(&sublist, curr_dev, curr_net);
 316                         /* start new sublist */
 317                         INIT_LIST_HEAD(&sublist);
 318                         curr_dev = dev;
 319                         curr_net = net;
 320                 }
 321                 list_add_tail(&skb->list, &sublist);
 322         }
 323         /* dispatch final sublist */
 324         ip6_sublist_rcv(&sublist, curr_dev, curr_net);
 325 }
 326
 327 /*
 328  *      Deliver the packet to the host
 329  */
 330
 331
 332 static int ip6_input_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 333 {
 334         const struct inet6_protocol *ipprot;
 335         struct inet6_dev *idev;
 336         unsigned int nhoff;
 337         int nexthdr;
 338         bool raw;
 339         bool have_final = false;
 340
 341         /*
 342          *      Parse extension headers
 343          */
 344
 345         rcu_read_lock();
 346 resubmit:
 347         idev = ip6_dst_idev(skb_dst(skb));
 348         if (!pskb_pull(skb, skb_transport_offset(skb)))
 349                 goto discard;
 350         nhoff = IP6CB(skb)->nhoff;
 351         nexthdr = skb_network_header(skb)[nhoff];
 352
 353 resubmit_final:
 354         raw = raw6_local_deliver(skb, nexthdr);
 355         ipprot = rcu_dereference(inet6_protos[nexthdr]);
 356         if (ipprot) {
 357                 int ret;
 358
 359                 if (have_final) {
 360                         if (!(ipprot->flags & INET6_PROTO_FINAL)) {
 361                                 /* Once we've seen a final protocol don't
 362                                  * allow encapsulation on any non-final
 363                                  * ones. This allows foo in UDP encapsulation
 364                                  * to work.
 365                                  */
 366                                 goto discard;
 367                         }
 368                 } else if (ipprot->flags & INET6_PROTO_FINAL) {
 369                         const struct ipv6hdr *hdr;
 370
 371                         /* Only do this once for first final protocol */
 372                         have_final = true;
 373
 374                         /* Free reference early: we don't need it any more,
 375                            and it may hold ip_conntrack module loaded
 376                            indefinitely. */
 377                         nf_reset(skb);
 378
 379                         skb_postpull_rcsum(skb, skb_network_header(skb),
 380                                            skb_network_header_len(skb));
 381                         hdr = ipv6_hdr(skb);
 382                         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 383                             !ipv6_chk_mcast_addr(skb->dev, &hdr->daddr,
 384                             &hdr->saddr) &&
 385                             !ipv6_is_mld(skb, nexthdr, skb_network_header_len(skb)))
 386                                 goto discard;
 387                 }
 388                 if (!(ipprot->flags & INET6_PROTO_NOPOLICY) &&
 389                     !xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 390                         goto discard;
 391
 392                 ret = ipprot->handler(skb);
 393                 if (ret > 0) {
 394                         if (ipprot->flags & INET6_PROTO_FINAL) {
 395                                 /* Not an extension header, most likely UDP
 396                                  * encapsulation. Use return value as nexthdr
 397                                  * protocol not nhoff (which presumably is
 398                                  * not set by handler).
 399                                  */
 400                                 nexthdr = ret;
 401                                 goto resubmit_final;
 402                         } else {
 403                                 goto resubmit;
 404                         }
 405                 } else if (ret == 0) {
 406                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 407                 }
 408         } else {
 409                 if (!raw) {
 410                         if (xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 411                                 __IP6_INC_STATS(net, idev,
 412                                                 IPSTATS_MIB_INUNKNOWNPROTOS);
 413                                 icmpv6_send(skb, ICMPV6_PARAMPROB,
 414                                             ICMPV6_UNK_NEXTHDR, nhoff);
 415                         }
 416                         kfree_skb(skb);
 417                 } else {
 418                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 419                         consume_skb(skb);
 420                 }
 421         }
 422         rcu_read_unlock();
 423         return 0;
 424
 425 discard:
 426         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 427         rcu_read_unlock();
 428         kfree_skb(skb);
 429         return 0;
 430 }
 431
 432
 433 int ip6_input(struct sk_buff *skb)
 434 {
 435         return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_IN,
 436                        dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 437                        ip6_input_finish);
 438 }
 439 EXPORT_SYMBOL_GPL(ip6_input);
 440
 441 int ip6_mc_input(struct sk_buff *skb)
 442 {
 443         const struct ipv6hdr *hdr;
 444         bool deliver;
 445
 446         __IP6_UPD_PO_STATS(dev_net(skb_dst(skb)->dev),
 447                          __in6_dev_get_safely(skb->dev), IPSTATS_MIB_INMCAST,
 448                          skb->len);
 449
 450         hdr = ipv6_hdr(skb);
 451         deliver = ipv6_chk_mcast_addr(skb->dev, &hdr->daddr, NULL);
 452
 453 #ifdef CONFIG_IPV6_MROUTE
 454         /*
 455          *      IPv6 multicast router mode is now supported ;)
 456          */
 457         if (dev_net(skb->dev)->ipv6.devconf_all->mc_forwarding &&
 458             !(ipv6_addr_type(&hdr->daddr) &
 459               (IPV6_ADDR_LOOPBACK|IPV6_ADDR_LINKLOCAL)) &&
 460             likely(!(IP6CB(skb)->flags & IP6SKB_FORWARDED))) {
 461                 /*
 462                  * Okay, we try to forward - split and duplicate
 463                  * packets.
 464                  */
 465                 struct sk_buff *skb2;
 466                 struct inet6_skb_parm *opt = IP6CB(skb);
 467
 468                 /* Check for MLD */
 469                 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
 470                         /* Check if this is a mld message */
 471                         u8 nexthdr = hdr->nexthdr;
 472                         __be16 frag_off;
 473                         int offset;
 474
 475                         /* Check if the value of Router Alert
 476                          * is for MLD (0x0000).
 477                          */
 478                         if (opt->ra == htons(IPV6_OPT_ROUTERALERT_MLD)) {
 479                                 deliver = false;
 480
 481                                 if (!ipv6_ext_hdr(nexthdr)) {
 482                                         /* BUG */
 483                                         goto out;
 484                                 }
 485                                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr),
 486                                                           &nexthdr, &frag_off);
 487                                 if (offset < 0)
 488                                         goto out;
 489
 490                                 if (ipv6_is_mld(skb, nexthdr, offset))
 491                                         deliver = true;
 492
 493                                 goto out;
 494                         }
 495                         /* unknown RA - process it normally */
 496                 }
 497
 498                 if (deliver)
 499                         skb2 = skb_clone(skb, GFP_ATOMIC);
 500                 else {
 501                         skb2 = skb;
 502                         skb = NULL;
 503                 }
 504
 505                 if (skb2) {
 506                         ip6_mr_input(skb2);
 507                 }
 508         }
 509 out:
 510 #endif
 511         if (likely(deliver))
 512                 ip6_input(skb);
 513         else {
 514                 /* discard */
 515                 kfree_skb(skb);
 516         }
 517
 518         return 0;
 519 }