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 #include <linux/indirect_call_wrapper.h>
  33
  34 #include <linux/netfilter.h>
  35 #include <linux/netfilter_ipv6.h>
  36
  37 #include <net/sock.h>
  38 #include <net/snmp.h>
  39
  40 #include <net/ipv6.h>
  41 #include <net/protocol.h>
  42 #include <net/transp_v6.h>
  43 #include <net/rawv6.h>
  44 #include <net/ndisc.h>
  45 #include <net/ip6_route.h>
  46 #include <net/addrconf.h>
  47 #include <net/xfrm.h>
  48 #include <net/inet_ecn.h>
  49 #include <net/dst_metadata.h>
  50
  51 INDIRECT_CALLABLE_DECLARE(void udp_v6_early_demux(struct sk_buff *));
  52 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *));
  53 static void ip6_rcv_finish_core(struct net *net, struct sock *sk,
  54                                 struct sk_buff *skb)
  55 {
  56         void (*edemux)(struct sk_buff *skb);
  57
  58         if (net->ipv4.sysctl_ip_early_demux && !skb_dst(skb) && skb->sk == NULL) {
  59                 const struct inet6_protocol *ipprot;
  60
  61                 ipprot = rcu_dereference(inet6_protos[ipv6_hdr(skb)->nexthdr]);
  62                 if (ipprot && (edemux = READ_ONCE(ipprot->early_demux)))
  63                         INDIRECT_CALL_2(edemux, tcp_v6_early_demux,
  64                                         udp_v6_early_demux, skb);
  65         }
  66         if (!skb_valid_dst(skb))
  67                 ip6_route_input(skb);
  68 }
  69
  70 int ip6_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  71 {
  72         /* if ingress device is enslaved to an L3 master device pass the
  73          * skb to its handler for processing
  74          */
  75         skb = l3mdev_ip6_rcv(skb);
  76         if (!skb)
  77                 return NET_RX_SUCCESS;
  78         ip6_rcv_finish_core(net, sk, skb);
  79
  80         return dst_input(skb);
  81 }
  82
  83 static void ip6_sublist_rcv_finish(struct list_head *head)
  84 {
  85         struct sk_buff *skb, *next;
  86
  87         list_for_each_entry_safe(skb, next, head, list)
  88                 dst_input(skb);
  89 }
  90
  91 static void ip6_list_rcv_finish(struct net *net, struct sock *sk,
  92                                 struct list_head *head)
  93 {
  94         struct dst_entry *curr_dst = NULL;
  95         struct sk_buff *skb, *next;
  96         struct list_head sublist;
  97
  98         INIT_LIST_HEAD(&sublist);
  99         list_for_each_entry_safe(skb, next, head, list) {
 100                 struct dst_entry *dst;
 101
 102                 skb_list_del_init(skb);
 103                 /* if ingress device is enslaved to an L3 master device pass the
 104                  * skb to its handler for processing
 105                  */
 106                 skb = l3mdev_ip6_rcv(skb);
 107                 if (!skb)
 108                         continue;
 109                 ip6_rcv_finish_core(net, sk, skb);
 110                 dst = skb_dst(skb);
 111                 if (curr_dst != dst) {
 112                         /* dispatch old sublist */
 113                         if (!list_empty(&sublist))
 114                                 ip6_sublist_rcv_finish(&sublist);
 115                         /* start new sublist */
 116                         INIT_LIST_HEAD(&sublist);
 117                         curr_dst = dst;
 118                 }
 119                 list_add_tail(&skb->list, &sublist);
 120         }
 121         /* dispatch final sublist */
 122         ip6_sublist_rcv_finish(&sublist);
 123 }
 124
 125 static struct sk_buff *ip6_rcv_core(struct sk_buff *skb, struct net_device *dev,
 126                                     struct net *net)
 127 {
 128         const struct ipv6hdr *hdr;
 129         u32 pkt_len;
 130         struct inet6_dev *idev;
 131
 132         if (skb->pkt_type == PACKET_OTHERHOST) {
 133                 kfree_skb(skb);
 134                 return NULL;
 135         }
 136
 137         rcu_read_lock();
 138
 139         idev = __in6_dev_get(skb->dev);
 140
 141         __IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_IN, skb->len);
 142
 143         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL ||
 144             !idev || unlikely(idev->cnf.disable_ipv6)) {
 145                 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 146                 goto drop;
 147         }
 148
 149         memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
 150
 151         /*
 152          * Store incoming device index. When the packet will
 153          * be queued, we cannot refer to skb->dev anymore.
 154          *
 155          * BTW, when we send a packet for our own local address on a
 156          * non-loopback interface (e.g. ethX), it is being delivered
 157          * via the loopback interface (lo) here; skb->dev = loopback_dev.
 158          * It, however, should be considered as if it is being
 159          * arrived via the sending interface (ethX), because of the
 160          * nature of scoping architecture. --yoshfuji
 161          */
 162         IP6CB(skb)->iif = skb_valid_dst(skb) ? ip6_dst_idev(skb_dst(skb))->dev->ifindex : dev->ifindex;
 163
 164         if (unlikely(!pskb_may_pull(skb, sizeof(*hdr))))
 165                 goto err;
 166
 167         hdr = ipv6_hdr(skb);
 168
 169         if (hdr->version != 6)
 170                 goto err;
 171
 172         __IP6_ADD_STATS(net, idev,
 173                         IPSTATS_MIB_NOECTPKTS +
 174                                 (ipv6_get_dsfield(hdr) & INET_ECN_MASK),
 175                         max_t(unsigned short, 1, skb_shinfo(skb)->gso_segs));
 176         /*
 177          * RFC4291 2.5.3
 178          * The loopback address must not be used as the source address in IPv6
 179          * packets that are sent outside of a single node. [..]
 180          * A packet received on an interface with a destination address
 181          * of loopback must be dropped.
 182          */
 183         if ((ipv6_addr_loopback(&hdr->saddr) ||
 184              ipv6_addr_loopback(&hdr->daddr)) &&
 185             !(dev->flags & IFF_LOOPBACK) &&
 186             !netif_is_l3_master(dev))
 187                 goto err;
 188
 189         /* RFC4291 Errata ID: 3480
 190          * Interface-Local scope spans only a single interface on a
 191          * node and is useful only for loopback transmission of
 192          * multicast.  Packets with interface-local scope received
 193          * from another node must be discarded.
 194          */
 195         if (!(skb->pkt_type == PACKET_LOOPBACK ||
 196               dev->flags & IFF_LOOPBACK) &&
 197             ipv6_addr_is_multicast(&hdr->daddr) &&
 198             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 1)
 199                 goto err;
 200
 201         /* If enabled, drop unicast packets that were encapsulated in link-layer
 202          * multicast or broadcast to protected against the so-called "hole-196"
 203          * attack in 802.11 wireless.
 204          */
 205         if (!ipv6_addr_is_multicast(&hdr->daddr) &&
 206             (skb->pkt_type == PACKET_BROADCAST ||
 207              skb->pkt_type == PACKET_MULTICAST) &&
 208             idev->cnf.drop_unicast_in_l2_multicast)
 209                 goto err;
 210
 211         /* RFC4291 2.7
 212          * Nodes must not originate a packet to a multicast address whose scope
 213          * field contains the reserved value 0; if such a packet is received, it
 214          * must be silently dropped.
 215          */
 216         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 217             IPV6_ADDR_MC_SCOPE(&hdr->daddr) == 0)
 218                 goto err;
 219
 220         /*
 221          * RFC4291 2.7
 222          * Multicast addresses must not be used as source addresses in IPv6
 223          * packets or appear in any Routing header.
 224          */
 225         if (ipv6_addr_is_multicast(&hdr->saddr))
 226                 goto err;
 227
 228         skb->transport_header = skb->network_header + sizeof(*hdr);
 229         IP6CB(skb)->nhoff = offsetof(struct ipv6hdr, nexthdr);
 230
 231         pkt_len = ntohs(hdr->payload_len);
 232
 233         /* pkt_len may be zero if Jumbo payload option is present */
 234         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 235                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len) {
 236                         __IP6_INC_STATS(net,
 237                                         idev, IPSTATS_MIB_INTRUNCATEDPKTS);
 238                         goto drop;
 239                 }
 240                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr))) {
 241                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 242                         goto drop;
 243                 }
 244                 hdr = ipv6_hdr(skb);
 245         }
 246
 247         if (hdr->nexthdr == NEXTHDR_HOP) {
 248                 if (ipv6_parse_hopopts(skb) < 0) {
 249                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 250                         rcu_read_unlock();
 251                         return NULL;
 252                 }
 253         }
 254
 255         rcu_read_unlock();
 256
 257         /* Must drop socket now because of tproxy. */
 258         skb_orphan(skb);
 259
 260         return skb;
 261 err:
 262         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
 263 drop:
 264         rcu_read_unlock();
 265         kfree_skb(skb);
 266         return NULL;
 267 }
 268
 269 int ipv6_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 270 {
 271         struct net *net = dev_net(skb->dev);
 272
 273         skb = ip6_rcv_core(skb, dev, net);
 274         if (skb == NULL)
 275                 return NET_RX_DROP;
 276         return NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING,
 277                        net, NULL, skb, dev, NULL,
 278                        ip6_rcv_finish);
 279 }
 280
 281 static void ip6_sublist_rcv(struct list_head *head, struct net_device *dev,
 282                             struct net *net)
 283 {
 284         NF_HOOK_LIST(NFPROTO_IPV6, NF_INET_PRE_ROUTING, net, NULL,
 285                      head, dev, NULL, ip6_rcv_finish);
 286         ip6_list_rcv_finish(net, NULL, head);
 287 }
 288
 289 /* Receive a list of IPv6 packets */
 290 void ipv6_list_rcv(struct list_head *head, struct packet_type *pt,
 291                    struct net_device *orig_dev)
 292 {
 293         struct net_device *curr_dev = NULL;
 294         struct net *curr_net = NULL;
 295         struct sk_buff *skb, *next;
 296         struct list_head sublist;
 297
 298         INIT_LIST_HEAD(&sublist);
 299         list_for_each_entry_safe(skb, next, head, list) {
 300                 struct net_device *dev = skb->dev;
 301                 struct net *net = dev_net(dev);
 302
 303                 skb_list_del_init(skb);
 304                 skb = ip6_rcv_core(skb, dev, net);
 305                 if (skb == NULL)
 306                         continue;
 307
 308                 if (curr_dev != dev || curr_net != net) {
 309                         /* dispatch old sublist */
 310                         if (!list_empty(&sublist))
 311                                 ip6_sublist_rcv(&sublist, curr_dev, curr_net);
 312                         /* start new sublist */
 313                         INIT_LIST_HEAD(&sublist);
 314                         curr_dev = dev;
 315                         curr_net = net;
 316                 }
 317                 list_add_tail(&skb->list, &sublist);
 318         }
 319         /* dispatch final sublist */
 320         ip6_sublist_rcv(&sublist, curr_dev, curr_net);
 321 }
 322
 323 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
 324 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *));
 325
 326 /*
 327  *      Deliver the packet to the host
 328  */
 329 void ip6_protocol_deliver_rcu(struct net *net, struct sk_buff *skb, int nexthdr,
 330                               bool have_final)
 331 {
 332         const struct inet6_protocol *ipprot;
 333         struct inet6_dev *idev;
 334         unsigned int nhoff;
 335         bool raw;
 336
 337         /*
 338          *      Parse extension headers
 339          */
 340
 341 resubmit:
 342         idev = ip6_dst_idev(skb_dst(skb));
 343         nhoff = IP6CB(skb)->nhoff;
 344         if (!have_final) {
 345                 if (!pskb_pull(skb, skb_transport_offset(skb)))
 346                         goto discard;
 347                 nexthdr = skb_network_header(skb)[nhoff];
 348         }
 349
 350 resubmit_final:
 351         raw = raw6_local_deliver(skb, nexthdr);
 352         ipprot = rcu_dereference(inet6_protos[nexthdr]);
 353         if (ipprot) {
 354                 int ret;
 355
 356                 if (have_final) {
 357                         if (!(ipprot->flags & INET6_PROTO_FINAL)) {
 358                                 /* Once we've seen a final protocol don't
 359                                  * allow encapsulation on any non-final
 360                                  * ones. This allows foo in UDP encapsulation
 361                                  * to work.
 362                                  */
 363                                 goto discard;
 364                         }
 365                 } else if (ipprot->flags & INET6_PROTO_FINAL) {
 366                         const struct ipv6hdr *hdr;
 367                         int sdif = inet6_sdif(skb);
 368                         struct net_device *dev;
 369
 370                         /* Only do this once for first final protocol */
 371                         have_final = true;
 372
 373                         /* Free reference early: we don't need it any more,
 374                            and it may hold ip_conntrack module loaded
 375                            indefinitely. */
 376                         nf_reset(skb);
 377
 378                         skb_postpull_rcsum(skb, skb_network_header(skb),
 379                                            skb_network_header_len(skb));
 380                         hdr = ipv6_hdr(skb);
 381
 382                         /* skb->dev passed may be master dev for vrfs. */
 383                         if (sdif) {
 384                                 dev = dev_get_by_index_rcu(net, sdif);
 385                                 if (!dev)
 386                                         goto discard;
 387                         } else {
 388                                 dev = skb->dev;
 389                         }
 390
 391                         if (ipv6_addr_is_multicast(&hdr->daddr) &&
 392                             !ipv6_chk_mcast_addr(dev, &hdr->daddr,
 393                                                  &hdr->saddr) &&
 394                             !ipv6_is_mld(skb, nexthdr, skb_network_header_len(skb)))
 395                                 goto discard;
 396                 }
 397                 if (!(ipprot->flags & INET6_PROTO_NOPOLICY) &&
 398                     !xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 399                         goto discard;
 400
 401                 ret = INDIRECT_CALL_2(ipprot->handler, tcp_v6_rcv, udpv6_rcv,
 402                                       skb);
 403                 if (ret > 0) {
 404                         if (ipprot->flags & INET6_PROTO_FINAL) {
 405                                 /* Not an extension header, most likely UDP
 406                                  * encapsulation. Use return value as nexthdr
 407                                  * protocol not nhoff (which presumably is
 408                                  * not set by handler).
 409                                  */
 410                                 nexthdr = ret;
 411                                 goto resubmit_final;
 412                         } else {
 413                                 goto resubmit;
 414                         }
 415                 } else if (ret == 0) {
 416                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 417                 }
 418         } else {
 419                 if (!raw) {
 420                         if (xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 421                                 __IP6_INC_STATS(net, idev,
 422                                                 IPSTATS_MIB_INUNKNOWNPROTOS);
 423                                 icmpv6_send(skb, ICMPV6_PARAMPROB,
 424                                             ICMPV6_UNK_NEXTHDR, nhoff);
 425                         }
 426                         kfree_skb(skb);
 427                 } else {
 428                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDELIVERS);
 429                         consume_skb(skb);
 430                 }
 431         }
 432         return;
 433
 434 discard:
 435         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
 436         kfree_skb(skb);
 437 }
 438
 439 static int ip6_input_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 440 {
 441         rcu_read_lock();
 442         ip6_protocol_deliver_rcu(net, skb, 0, false);
 443         rcu_read_unlock();
 444
 445         return 0;
 446 }
 447
 448
 449 int ip6_input(struct sk_buff *skb)
 450 {
 451         return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_IN,
 452                        dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 453                        ip6_input_finish);
 454 }
 455 EXPORT_SYMBOL_GPL(ip6_input);
 456
 457 int ip6_mc_input(struct sk_buff *skb)
 458 {
 459         int sdif = inet6_sdif(skb);
 460         const struct ipv6hdr *hdr;
 461         struct net_device *dev;
 462         bool deliver;
 463
 464         __IP6_UPD_PO_STATS(dev_net(skb_dst(skb)->dev),
 465                          __in6_dev_get_safely(skb->dev), IPSTATS_MIB_INMCAST,
 466                          skb->len);
 467
 468         /* skb->dev passed may be master dev for vrfs. */
 469         if (sdif) {
 470                 rcu_read_lock();
 471                 dev = dev_get_by_index_rcu(dev_net(skb->dev), sdif);
 472                 if (!dev) {
 473                         rcu_read_unlock();
 474                         kfree_skb(skb);
 475                         return -ENODEV;
 476                 }
 477         } else {
 478                 dev = skb->dev;
 479         }
 480
 481         hdr = ipv6_hdr(skb);
 482         deliver = ipv6_chk_mcast_addr(dev, &hdr->daddr, NULL);
 483         if (sdif)
 484                 rcu_read_unlock();
 485
 486 #ifdef CONFIG_IPV6_MROUTE
 487         /*
 488          *      IPv6 multicast router mode is now supported ;)
 489          */
 490         if (dev_net(skb->dev)->ipv6.devconf_all->mc_forwarding &&
 491             !(ipv6_addr_type(&hdr->daddr) &
 492               (IPV6_ADDR_LOOPBACK|IPV6_ADDR_LINKLOCAL)) &&
 493             likely(!(IP6CB(skb)->flags & IP6SKB_FORWARDED))) {
 494                 /*
 495                  * Okay, we try to forward - split and duplicate
 496                  * packets.
 497                  */
 498                 struct sk_buff *skb2;
 499                 struct inet6_skb_parm *opt = IP6CB(skb);
 500
 501                 /* Check for MLD */
 502                 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
 503                         /* Check if this is a mld message */
 504                         u8 nexthdr = hdr->nexthdr;
 505                         __be16 frag_off;
 506                         int offset;
 507
 508                         /* Check if the value of Router Alert
 509                          * is for MLD (0x0000).
 510                          */
 511                         if (opt->ra == htons(IPV6_OPT_ROUTERALERT_MLD)) {
 512                                 deliver = false;
 513
 514                                 if (!ipv6_ext_hdr(nexthdr)) {
 515                                         /* BUG */
 516                                         goto out;
 517                                 }
 518                                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr),
 519                                                           &nexthdr, &frag_off);
 520                                 if (offset < 0)
 521                                         goto out;
 522
 523                                 if (ipv6_is_mld(skb, nexthdr, offset))
 524                                         deliver = true;
 525
 526                                 goto out;
 527                         }
 528                         /* unknown RA - process it normally */
 529                 }
 530
 531                 if (deliver)
 532                         skb2 = skb_clone(skb, GFP_ATOMIC);
 533                 else {
 534                         skb2 = skb;
 535                         skb = NULL;
 536                 }
 537
 538                 if (skb2) {
 539                         ip6_mr_input(skb2);
 540                 }
 541         }
 542 out:
 543 #endif
 544         if (likely(deliver))
 545                 ip6_input(skb);
 546         else {
 547                 /* discard */
 548                 kfree_skb(skb);
 549         }
 550
 551         return 0;
 552 }